2014 Book TopologyOptimizationInStructur PDF

CISM International Centre for Mechanical Sciences 549
Courses and Lectures
George I. N. Rozvany
Tomasz Lewiński
Editors
Topology
Optimization
in Structural
and Continuum
Mechanics
International Centre
for Mechanical Sciences
CISM Courses and Lectures
Series Editors:
The Rectors
Friedrich Pfeiffer - Munich
Franz G. Rammerstorfer - Wien
Elisabeth Guazzelli - Marseille
The Secretary General

Bernhard Schrefler - Padua
Executive Editor
Paolo Serafini - Udine
The series presents lecture notes, monographs, edited works and

proceedings in the field of Mechanics, Engineering, Computer Science
and Applied Mathematics.
Purpose of the series is to make known in the international scientific
and technical community results obtained in some of the activities
organized by CISM, the International Centre for Mechanical Sciences.
International Centre for Mechanical Sciences
Courses and Lectures Vol. 549
For further volumes:

www.springer.com/series/76
George I. N. Rozvany · Tomasz Lewiński
Editors
Topology Optimization in
Structural and
Continuum Mechanics
Editors
George I. N. Rozvany
Budapest University of Technology and Economics, Hungary
Tomasz Lewiński
Warsaw University of Technology, Poland
ISSN 0254-1971
ISBN 978-3-7091-1642-5 ISBN 978-3-7091-1643-2 (eBook)
DOI 10.1007/ 978-3-7091-1643-2
Springer Wien Heidelberg New York Dordrecht London
© CISM, Udine 2014
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole
or part of the material is concerned, specifically the rights of translation, reprinting, reuse of
illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical
way, and transmission or information storage and retrieval, electronic adaptation, computer
software, or by similar or dissimilar methodology now known or hereafter developed. Ex-
empted from this legal reservation are brief excerpts in connection with reviews or scholarly
analysis or material supplied specifically for the purpose of being entered and executed on
a computer system, for exclusive use by the purchaser of the work. Duplication of this pub-
lication or parts thereof is permitted only under the provisions of the Copyright Law of the
Publisher’s location, in its current version, and permission for use must always be obtained
from Springer. Permissions for use may be obtained through RightsLink at the Copyright
Clearance Center. Violations are liable to prosecution under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are
exempt from the relevant protective laws and regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at
the date of publication, neither the authors nor the editors nor the publisher can ac-
cept any legal responsibility for any errors or omissions that may be made. The publish-
er makes no warranty, express or implied, with respect to the material contained herein.
All contributions have been typeset by the authors

Printed in Italy
Printed on acid-free paper
Springer is part of Springer Science+Business Media (www.springer.com)

PREFACE
Structural Topology Optimization (STO) is a relatively new, but ra-

pidly expanding and extremely popular field of structural mechanics.
Various theoretical aspects, as well as a great variety of numerical
methods and applications are discussed extensively in international
journals and at conferences. The high level of interest in this field
is due to the substantial savings that can be achieved by topology
optimization in industrial applications. Moreover, STO has interest-
ing theoretical implications in mathematics, mechanics, multi-physics
and computer science.
This is the third CISM Advanced Course on Structural Topology
Optimization. The two previous ones were organized by the first au-
thor of this Preface, the current one – by both authors.
The aim of the present course is to cover new developments in this
field since the previous CISM meeting on STO in 1997. The topics
reviewed by various lecturers of this course are summarized briefly
below.
In his first lecture, George I. N. Rozvany reviews the basic features
and limitations of Michell’s (1904) truss theory, and its extension to
a broader class of support conditions.
In the second lecture, George Rozvany and Erika Pinter give an
overview of generalizations of truss topology optimization, via the
Prager-Rozvany (1977) optimal layout theory, to multiple load con-
ditions, probabilistic design and optimization with pre-existing mem-
bers, also briefly reviewing optimal grillage theory and cognitive pro-
cesses in deriving exact optimal topologies.
George Rozvany’s third lecture discusses fundamental properties of
exact optimal structural topologies, including (non)uniqueness, sym-
metry, skew-symmetry, domain augmentation and reduction, and the
effect of non-zero support cost.
In a joint lecture with Tomasz Sokól, the verification of various
numerical methods by exact analytical benchmarks is explained, and
conversely, the confirmation of exact analytical solutions by Sokól’s
numerical method is discussed. The latter can currently handle ground
structures with several billion potential members. In his final lecture,
George Rozvany gives a concise historical overview of structural topol-
ogy optimization, and critically reviews various numerical methods in
this field.
The lecture by Tomasz Lewiński and Tomasz Sokól is focused on
one aspect of the lectures by George Rozvany, namely on the Michell
continua. This theory is constructed for volume minimization of
trusses which finally reduces to a locking material problem.
The Michell problem belongs to the class of optimization of stat-
ically determinate structures whose behavior is governed only by the
equilibrium conditions and constraints bounding the stress level. More
complex problems arise if one optimizes the shape of elastic bodies,
even those being homogeneous and isotropic. In general, the layout
problems in linear elasticity are ill-posed, which is the central ques-
tion of the lecture by François Jouve. This author discusses the above
problem and clears up the remedies: either to extend the design space
and to relax the problem, or to reduce the design space by introducing
new regularity constraints. The relaxation by homogenization method
is outlined in Sec. 2 of this lecture, along with numerical techniques.
The method is efficient due to fundamental results concerning opti-
mal bounds on the energy. Although this exact and explicit result is
restricted to the compliance minimization for a single load condition,
it has served as the basis for various researchers to develop other
homogenization-based methods, such as the one by Grégoire Allaire,
Eric Bonnetier, Gilles Francfort and François Jouve in 1997. In his
lecture François Jouve discusses also the methods of partial relax-
ation of selected problems for which the exact relaxations are not at
our disposal, or they assume a non-explicit form. The last chapter of
the lecture concerns the level set method proposed in the early 2000’s,
which gives very promising results, even in an industrial context, with
complex state equations, objective functions and constraints. This au-
thor shows how this method can be combined with shape derivatives
and by the topology derivatives of selected functionals.
The lecture by Grzegorz Dzierżanowski and Tomasz Lewiński de-
livers a complete derivation of the crucial result mentioned: the opti-
mal bounds on the energy. The derivation is based on the translation
method for the case of two isotropic constituents and then reduced to
the case if one constituent is a void.
Structural topology optimization comprises also the design of ma-
terial characteristics without linking them with the density of mass.
This optimization field is called the Free Material Design (FMD).
The classical FMD problem is aimed at finding the optimum values of
all components of the Hooke tensor from the criterion of compliance
minimization, under the isoperimetric condition of boundedness of
the integral of the trace of the Hooke tensor. The lecture by Slawomir
Czarnecki and Tomasz Lewiński shows that the FMD problem can be
reduced to a locking material problem, even in the multi-load case.
The six lectures by Niels Olhoff, Jianbin Du and Bin Niu concern
the optimization of structures subjected to dynamic loads. These au-
thors explain how to design a structure such that the structural eigen-
frequencies of vibration are as far away as possible from a prescribed
external excitation frequency - or band of excitation frequencies - to
avoid resonance phenomena with high vibration and noise levels. This
objective may be achieved by
- maximizing the fundamental eigenfrequency of the structure,
- maximizing the distance (gap) between two consecutive eigen-
frequencies,
- maximizing the dynamic stiffness of the structure subject to
forced vibration,
- minimizing the sound power flow radiated from the structural
surface into an acoustic medium.
A special lecture by Niels Olhoff and Bin Niu discusses how max-
imization of the gap between two consecutive eigenfrequencies gen-
erates significant design periodicity, and the final (sixth) lecture pre-
sents the application of a novel topology based method of simultaneous
optimization of fiber angles, stacking sequence, and selection of ma-
terials, for vibrating laminate composite plates with minimum sound
radiation.
In the first three of his five lectures, Kurt Maute discusses ap-
plications of the density method to diffusive and convective transport
processes, as well as to multi-physics problems. The complexity of
selecting appropriate objectives and constraints are emphasized in the
chapter on diffusive transport optimization problems. The extension
of the porosity model to fluid problems is presented for flow topology
optimization problems, characterized by the Darcy-Stokes and Navier-
Stokes equations at steady state conditions. The fundamental differ-
ences in solving multi-physics problems that are either coupled via
constitutive laws or via surface interactions are discussed and illus-
trated with applications to piezo-electric coupling and fluid-structure
interaction problems. The fourth lecture introduces an alternative
to topology optimization approaches that employ density or Ersatz
material approaches to represent the material layout in the mechan-
ical model. The integration of the eXtended Finite Element Method
(XFEM) into a level-set topology optimization method is discussed
and illustrated with applications to flow topology optimization. The
last lecture by Kurt Maute is devoted to topology optimization methods
that account for uncertainty in material parameters, geometry, and
operating conditions. Here, the aim is to arrive at reliable and robust
designs. This lecture introduces basic techniques in reliability based
design optimization (RBDO) and robust design optimization (RDO),
and discusses their application to topology optimization.
G.I.N. Rozvany and T. Lewiński

CONTENTS
Preface
Structural Topology Optimization (STO) – Exact Analyt-

ical Solutions: Part I
by G.I.N. Rozvany . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Structural Topology Optimization (STO) – Exact Analyt-
ical Solutions: Part II
by G.I.N. Rozvany and E. Pinter . . . . . . . . . . . . . . . . . . . . . . . . 15
Some Fundamental Properties of Exact Optimal Structural
Topologies
by G.I.N. Rozvany . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Validation of Numerical Methods by Analytical Bench-
marks, and Verification of Exact Solutions by Numerical
Methods
by G.I.N. Rozvany and T. Sokól . . . . . . . . . . . . . . . . . . . . . . . . . 53
A Brief Review of Numerical Methods of Structural Topol-
ogy Optimization
by G.I.N. Rozvany . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
On Basic Properties of Michell’s Structures
by T. Lewiński and T. Sokól . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
Structural Shape and Topology Optimization
by F. Jouve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
Compliance Minimization of Two-Material Elastic Struc-
tures
by G. Dzierżanowski and T. Lewiński . . . . . . . . . . . . . . . . . . . . 175
The Free Material Design in Linear Elasticity
by S. Czarnecki and T. Lewiński . . . . . . . . . . . . . . . . . . . . . . . . . 213
Introductory Notes on Topological Design Optimization of
Vibrating Continuum Structures
by N. Olhoff and J. Du . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
Structural Topology Optimization with Respect to Eigen-
frequencies of Vibration
by N. Olhoff and J. Du . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
On Optimum Design and Periodicity of Band-gap Struc-
tures
by N. Olhoff and B. Niu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
Topological Design for Minimum Dynamic Compliance of
Structures under Forced Vibration
by N. Olhoff and J. Du . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
Topological Design for Minimum Sound Emission from
Structures under Forced Vibration
by N. Olhoff and J. Du . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
Discrete Material Optimization of Vibrating Laminated
Composite Plates for Minimum Sound Emission
by N. Olhoff and B. Niu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
Topology Optimization of Diffusive Transport Problems
by K. Maute . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
Topology Optimization of Flows: Stokes and Navier-Stokes
Models
by K. Maute . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 409
Topology Optimization of Coupled Multi-Physics Problems
by K. Maute . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421
The Extended Finite Element Method

by K. Maute . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 439
Topology Optimization under Uncertainty
by K. Maute . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 457

6WUXFWXUDO7RSRORJ\2SWLPL]DWLRQ672±([DFW

$QDO\WLFDO6ROXWLRQV3DUW,
*HRUJH,15R]YDQ\

'HSDUWPHQWRI6WUXFWXUDO0HFKDQLFV%XGDSHVW8QLYHUVLW\RI7HFKQRORJ\DQG(FRQRPLFV
%XGDSHVW+XQJDU\
,QWURGXFWLRQ6RPH3UHYLRXV0HHWLQJV5HYLHZ/HFWXUHVDQG
5HYLHZ$UWLFOHVRQ6WUXFWXUDO7RSRORJ\2SWLPL]DWLRQ672
$VPHQWLRQHGLQWKH3UHIDFHRQWKHVXEMHFWRI672WKHDXWKRUKDVRUJDQL]HGWZR
SUHYLRXV&,60 $GYDQFHG &RXUVHV LQ 8GLQH 5R]YDQ\ DQG DQG ZLWK
1LHOV2OKRIID1$72$5:LQ%XGDSHVW5R]YDQ\DQG2OKRII
,QUHFHQW\HDUVWKHDXWKRUJDYHSULQFLSDONH\QRWHRUUHYLHZOHFWXUHVRQ672DW
LPSRUWDQWPHHWLQJVVXFKDVWK(XURSHDQ6ROLG0HFKDQLFV&RQIHUHQFH%XGDSHVW
:RUOG&RQJUHVVRI&RPSXWDWLRQDO0HFKDQLFV9HQLFH861DWLRQDO
&RQJUHVV RQ &RPSXWDWLRQDO 0HFKDQLFV &ROXPEXV 2KLR &KLQHVH 6ROLG
0HFKDQLFV&RQJUHVV'DOLDQ+XQJDULDQ$FDGHP\RI6FLHQFHPHHWLQJ%X
GDSHVW WK $602,6602 &RQIHUHQFH /RQGRQ &RPSXWDWLRQDO
6WUXFWXUHV7HFKQRORJ\&RQIHUHQFH9DOHQFLD6SDLQ&RPSXWDWLRQDO0HWKRGV
LQ0HFKDQLFV&RQIHUHQFH:DUVDZDQG(QJLQHHULQJ&RPSXWDWLRQDO7HFK
QRORJ\&RQIHUHQFH'XEURYQLN
7KHDXWKRUDQGFRDXWKRUVKDYHDOVRZULWWHQVRPHPXFKFLWHGUHYLHZDUWLFOHVRQ
6725R]YDQ\%HQGVRHDQG.LUVFK5R]YDQ\DQG$PRVWFRP
SUHKHQVLYHDQGDXWKRULWDWLYHERRNRQ672LVE\%HQGVRHDQG6LJPXQG
3UHOLPLQDULHV
6LQFHWKHDERYHOHFWXUHVGHDOWPRVWO\ZLWKQXPHULFDOWRSRORJ\RSWLPL]DWLRQWKHILUVW
WKUHH OHFWXUHV RI WKH DXWKRU ZLOO GLVFXVV H[DFW DQDO\WLFDO VROXWLRQV 7KHVH DUH H[
WUHPHO\LPSRUWDQWDVEHQFKPDUNVVLQFHQXPHULFDOPHWKRGVRIWRSRORJ\RSWLPL]DWLRQ
KDYHPDQ\VRXUFHVRIHUURU
$EDVLFIHDWXUHRIWKHDXWKRU¶VOHFWXUHVLVWKDWWKH\HPSOR\YHU\VLPSOHH[DPSOHV
VRWKDWIXQGDPHQWDOSULQFLSOHVRIPHFKDQLFVDUHQRWREVFXUHGE\FRPSXWDWLRQDORU
PDWKHPDWLFDO FRPSOH[LWLHV 0RUHRYHU IROORZLQJ WKH DGYLFH RI WKH ODWH 3URIHVVRU
:LOOLDP3UDJHU%URZQ8QLYHUVLW\YDULRXVJHQHUDOIHDWXUHVRIRSWLPDOWRSRORJLHV
ZLOOEHH[SODLQHGDQGSURYHGDWDQµHQJLQHHULQJOHYHO¶XVLQJSULQFLSOHVRIPHFKDQLFV
LQSUHIHUHQFHWRKLJKHUPDWKHPDWLFDOFRQFHSWV
G. Rozvany, T. Lewiński (Eds.), Topology Optimization in Structural

and Continuum Mechanics, CISM International Centre for Mechanical Sciences,
DOI 10.1007/978-3-7091-1643-2_1, © CISM, Udine 2014
2 G.I.N. Rozvany
6XEILHOGVRI672
$VH[SODLQHGSUHYLRXVO\VHHHJWKHUHYLHZDUWLFOH5R]YDQ\%HQGVRHDQG.LUVFK
VWUXFWXUDOWRSRORJ\RSWLPL]DWLRQFDQEHGLYLGHGLQWRWZRVXEILHOGV
/D\RXW RSWLPL]DWLRQ /2 GHDOV ZLWK JULGOLNH ORZ YROXPH IUDFWLRQ VWUXFWXUHV

WUXVVHVJULOODJHVVKHOOJULGVDQGGHQVHV\VWHPVRILQWHUVHFWLQJVKHOOVVHH)LJ
RSWLPL]LQJ VLPXOWDQHRXVO\ WKH WRSRORJ\ FRQQHFWLYLW\ RI WKH PHPEHUV JHRPHWU\
ORFDWLRQRIWKHMRLQWVDQGVL]LQJFURVVVHFWLRQDOGLPHQVLRQV

)LJXUH6RPHH[DPSOHVRIJULGOLNHVWUXFWXUHVDWUXVVEJULOODJHFVKHOOJULGG
LQWHUVHFWLQJVKHOOVKRQH\FRPE
*HQHUDOL]HG VKDSH RSWLPL]DWLRQ *62 ILQGV WKH EHVW WRSRORJ\ RI WKH LQWHUQDO
ERXQGDULHVDQGWKHVKDSHRIERWKLQWHUQDODQGH[WHUQDOERXQGDULHVRISHUIRUDWHGRU
FRPSRVLWHFRQWLQXD
/D\RXW2SWLPL]DWLRQ/2
,QOD\RXWRSWLPL]DWLRQ/2ZHVWDUWRIIZLWKDµJURXQGVWUXFWXUH¶ZKLFKFRQWDLQVDOO
SRWHQWLDO PHPEHUV µJURXQG HOHPHQWV¶ 'XULQJ WKH RSWLPL]DWLRQ SURFHGXUH ZH
UHPRYHQRQRSWLPDOJURXQGHOHPHQWVDQGGHWHUPLQHWKHRSWLPDOVL]HVRIWKHRSWLPDO
RQHV
Structural Topology Optimization (STO)… Part I 3
,QGLVFUHWL]HG/2WKHQXPEHURIJURXQGHOHPHQWVLVILQLWH,QH[DFW/2ZHKDYH
DQLQILQLWHQXPEHURIJURXQGHOHPHQWV$JURXQGVWUXFWXUHZLWKDQLQILQLWHQXPEHURI
HOHPHQWV LV VRPHWLPHV FDOOHG µVWUXFWXUDO XQLYHUVH¶ HJ 5R]YDQ\ DQG DO
$OWHUQDWLYHO\ D VWUXFWXUDO XQLYHUVH PD\ EH UHSODFHG E\ D µGHVLJQ GRPDLQ¶ DW DOO
SRLQWVRIZKLFKPHPEHUVPD\UXQLQDQ\GLUHFWLRQ
*HQHUDOL]HG6KDSH2SWLPL]DWLRQ*62
7KH VLPSOHVW IRUP RI VWUXFWXUDO WRSRORJ\ RSWLPL]DWLRQ FRQFHUQV VRFDOOHG ,6( WR
SRORJLHVZLWK,VRWURSLF6ROLGRU(PSW\HOHPHQWVVHHHJ5R]YDQ\
,Q GLVFUHWL]HG *62 WKH VWUXFWXUDO GRPDLQ LV GLYLGHG LQWR D ILQLWH QXPEHU RI
JURXQGHOHPHQWV)RU,6(WRSRORJLHVWKHWKLFNQHVVRUGHQVLW\RIHDFKJURXQGHOHPHQW
PD\RQO\WDNHRQD]HURYDOXHµZKLWHHOHPHQW¶RUDSUHDVVLJQHGRWKHUYDOXHµEODFN
HOHPHQW¶,QDQRUPDOL]HGIRUPXODWLRQWKHSUHDVVLJQHGYDOXHLVXQLW\DQGWKHQZH
KDYHDSUREOHP
,QH[DFW*62WKHQXPEHURIJURXQGHOHPHQWVWHQGVWRLQILQLW\DQGWKHUHIRUHWKH\
VKULQNWRSRLQWV,QRWKHUZRUGVH[DFW*62FRQVLVWVRIDVVLJQLQJRSWLPDOO\HLWKHU
µPDWHULDO¶RUµQRPDWHULDO¶WRHDFKSRLQWRIWKHGHVLJQGRPDLQ
*62FDQEHJHQHUDOL]HGWRPXOWLPDWHULDORSWLPL]DWLRQLQZKLFKFDVHIRUHDFK
JURXQGHOHPHQWRUDWHDFKSRLQWRIWKHGHVLJQGRPDLQZHPD\FKRVHRQHRXWRI
VHYHUDOJLYHQPDWHULDOVRUQRPDWHULDO
6WUXFWXUDOWRSRORJ\RSWLPL]DWLRQPD\FRQVLVWRIPLQLPL]LQJDQREMHFWLYHIXQFWLRQ
W\SLFDOO\WKHWRWDOVWUXFWXUDOYROXPHZHLJKWRUFRVWVXEMHFWWRFRQVWUDLQWVRQWKH
VWUXFWXUDO UHVSRQVH HJ VWUHVVHV GLVSODFHPHQWV EXFNOLQJ ORDG QDWXUDO IUHTXHQF\
HWF$GGLWLRQDOFRQVWUDLQWVPD\SXWOLPLWVRQWKHPHPEHUVL]HVVLGHFRQVWUDLQWV
HQVXUHPDQXIDFWXUDELOLW\RUSUHVFULEHGLVFUHWHFURVVVHFWLRQDOGLPHQVLRQV$OWHUQD
WLYHO\ ZH PD\ PD[LPL]H RU PLQLPL]H VRPH VWUXFWXUDO UHVSRQVH RU D ZHLJKWHG
FRPELQDWLRQRIVHYHUDOUHVSRQVHVIRUJLYHQYROXPHZHLJKWRUFRVW
0LFKHOO¶V7UXVV7KHRU\
7KHILUVWSDSHURQH[DFWVWUXFWXUDOWRSRORJ\RSWLPL]DWLRQZDVSXEOLVKHGPRUHWKDQD
FHQWXU\ DJR E\ WKH LQJHQLRXV $XVWUDOLDQ LQYHQWRU $ * 0 µ*HRUJH¶ 0LFKHOO
,WGHDOWZLWKWKHOHDVWYROXPHWRSRORJ\RIWUXVVHVZLWKDVLQJOHORDGFRQGLWLRQ
DQGDVWUHVVFRQVWUDLQW
6RPH%LRGDWDDERXW0LFKHOO
$*00LFKHOOZHQWWRVFKRROLQ&DPEULGJH8.DQGOLYHGODWHULQ
0HOERXUQH 9LFWRULD $XVWUDOLD +H ZDV VWLOO DOLYH ZKHQ WKH DXWKRU PRYHG IURP
)UDQFHWRWKDWVWDWHRI$XVWUDOLD0LFKHOOKDGDQXPEHURILQJHQLRXVLQYHQWLRQVRI
4 G.I.N. Rozvany
JUHDWLPSRUWDQFHWRWKHLQGXVWU\VXFKDVWKUXVWEHDULQJIRUVKLSVFUDQNOHVVHQJLQH
OXEULFDWLRQ V\VWHPV WLPEHU SUHVHUYDWLYHV K\GUDXOLF SRZHU WUDQVPLVVLRQV HWF
+RZHYHUWKHVHZHUHQRWIROORZHGE\ILQDQFLDOVXFFHVVIRU0LFKHOO+HOLNHGVROLWXGH
ZDVQRWLQWHUHVWHGLQWHDPVSRUWVQHYHUJRWPDUULHGDQGOLNHGWRZDONKXQGUHGVRI
PLOHVLQWKH$XVWUDOLDQEXVK
7KH6LJQLILFDQFHRI0LFKHOO¶V3DSHURQ7UXVVHV
0LFKHOO¶V PLOHVWRQH FRQWULEXWLRQ WR WUXVV WRSRORJ\ RSWLPL]DWLRQ ZDV XQ
GRXEWHGO\ ERWK UHYROXWLRQDU\ DQG LQJHQLRXV +H LQWURGXFHG HVVHQWLDO HOHPHQWV RI
ZKDW ZH QRZ FDOO OD\RXW RSWLPL]DWLRQ FRQWLQXXPW\SH RSWLPDOLW\ FULWHULD DGMRLQW
VWUDLQILHOG DQG JURXQG VWUXFWXUH VWUXFWXUDO XQLYHUVH +H DFKLHYHG DOO WKLV RYHU D
FHQWXU\DJRZKHQDOPRVWQRWKLQJZDVNQRZQDERXWHVVHQWLDOWHFKQLTXHVRIVWUXFWXUDO
RSWLPL]DWLRQ)RUWKLVUHDVRQ0LFKHOOKDGWRUHO\RQQHZDQGLPDJLQDWLYHLGHDVLQ
GHULYLQJKLVRSWLPDOLW\FULWHULD
0LFKHOO¶V2SWLPDOLW\&ULWHULD
,Q FXUUHQW QRWDWLRQ 0LFKHOO¶V SUREOHP FDQ EH VWDWHG DV IROORZV &RQVLGHULQJ D
VWUXFWXUDOGRPDLQ'ZLWKJLYHQH[WHUQDOIRUFHVPLQLPL]HWKHYROXPH9RIDWUXVV
VXEMHFWWRWKHVWUHVVFRQVWUDLQW
V & d V d V 7
ZKHUHVLVWKHORQJLWXGLQDOVWUHVVLQDQ\WUXVVPHPEHU V & LVWKHSHUPLVVLEOHVWUHVV

LQFRPSUHVVLRQDQG V 7 LVWKHSHUPLVVLEOHVWUHVVLQWHQVLRQ$VVXPLQJWKDWDOOWUXVV
PHPEHUVGHYHORSRQHRIWKHSHUPLVVLEOHVWUHVVHVVHHHQGRI6HFWLRQLQ/HFWXUH
WKHWRWDOWUXVVYROXPH9FDQEHH[SUHVVHGDV
9 ¦ /L )L V & ¦ /L )L V 7
& 7
ZKHUH/LDQG)LDUHWKHOHQJWKRIDQGWKHIRUFHLQWKHPHPEHULDQGWKHVXPPDWLRQ
RYHU & DQG 7 UHIHU WR DOO WKH FRPSUHVVLRQ DQG WHQVLRQ PHPEHUV UHVSHFWLYHO\ $
0LFKHOOWUXVVPD\KDYHDQLQILQLWHQXPEHURIPHPEHUV
1RWH ,Q 0LFKHOO¶V WKHRU\ QHLWKHU XSSHU QRU ORZHU OLPLWV DUH LPSRVHG RQ WKH
FURVVVHFWLRQDODUHDVZKLFKYDU\FRQWLQXRXVO\
,QFXUUHQWQRWDWLRQ0LFKHOO¶VRSWLPDOLW\FULWHULRQFDQEHVXPPDUL]HGDVIROORZV
İ N VJQ ) IRU ) z İ dN IRU )
ZKHUH İ LVDVPDOONLQHPDWLFDOO\DGPLVVLEOHVWUDLQDQGNLVDSRVLWLYHFRQVWDQW

7KHPHPEHUIRUFHV)PXVWEHVWDWLFDOO\DGPLVVLEOH
7KHWHUPµNLQHPDWLFDOO\DGPLVVLEOH¶LPSOLHVWKDWWKHFRQVLGHUHGVWUDLQILHOGRU
WKHFRUUHVSRQGLQJGLVSODFHPHQWILHOGVDWLVILHVNLQHPDWLFFRQWLQXLW\FRPSDWLELOLW\
DQGERXQGDU\VXSSRUWFRQGLWLRQVµ6WDWLFDOO\DGPLVVLEOH¶PHDQVWKDWWKHFRQVLGHUHG
IRUFHVVWUHVVHVVDWLVI\HTXLOLEULXPDQGVWDWLFERXQGDU\FRQGLWLRQV
0LFKHOOXVHGWKHV\PERO H IRUWKHOLPLWRQUHIHUHQFHVWUDLQLQVWHDGRINLQ
WKHUHODWLRQDERYH
/LPLWDWLRQVRI0LFKHOO¶V2SWLPDOLW\&ULWHULD
$IWHU UHH[DPLQLQJ 0LFKHOO¶V SURRI WKH DXWKRU FRQFOXGHG 5R]YDQ\ D WKDW
0LFKHOO¶VRULJLQDORSWLPDOLW\FULWHULDDUHYDOLGLIRQHRUERWKRIWKHIROORZLQJWZR
VXIILFLHQWFRQGLWLRQVDUHVDWLVILHG
D WKH VWUXFWXUDO GRPDLQ LV VXEMHFW WR SUHVFULEHG H[WHUQDO IRUFHV RQO\ RU
HTXLYDOHQWO\WRµVWDWLFDOO\GHWHUPLQDWHVXSSRUWV¶VXSSRUWVLQZKLFKWKHUHDF
WLRQVFDQEHXQLTXHO\GHWHUPLQHGIURPHTXLOLEULXPLHWKH\GRQRWGHSHQGRQ
WKHFKRLFHRIDVWDEOHWRSRORJ\QRURQWKHFURVVVHFWLRQDODUHDVZLWKLQDJLYHQ
WRSRORJ\RU
EWKHSHUPLVVLEOHVWUHVVHVLQWHQVLRQDQGFRPSUHVVLRQDUHWKHVDPH
7KHVHOLPLWDWLRQVZLOOEHLOOXVWUDWHGZLWKH[DPSOHVLQ6HFWLRQ
5HVHDUFKHUV¶5HVSRQVHWR0LFKHOO¶V3DSHU

0LFKHOO¶V SLRQHHULQJ HIIRUW ZDV FRPSOHWHO\ LJQRUHG IRU RYHU KDOI D FHQWXU\ DIWHU
ZKLFKLWZDVUHGLVFRYHUHGE\&R[LQ&R[ZKRDSSOLHG0LFKHOO¶V
FULWHULDWRDIHZQHZEXWVLPSOHORDGFRQGLWLRQV)ROORZLQJWKLV2ZHQDOVR
UHYLHZHGEULHIO\0LFKHOO¶VZRUNLQKLVERRN%XWDGHWDLOHGLQYHVWLJDWLRQRIWKLVWRSLF
LVGXHWR+HPSZKRZLWKLPSRUWDQWFRQWULEXWLRQVE\KLVFRZRUNHUV$6/&KDQ
+6<&KDQDQG0F&RQHOOVWXGLHG0LFKHOOVWUXFWXUHVPRVWRIKLVSURIHVVLRQDOOLIH
9DOXDEOHUHVXOWVE\KLPDQGKLVUHVHDUFKDVVRFLDWHVZHUHVXPPDUL]HGLQKLVERRN
+HPS
([WHQVLRQ RI 0LFKHOO¶V 7KHRU\ WR 6WDWLFDOO\ ,QGHWHUPLQDWH

6XSSRUW&RQGLWLRQVDQG8QHTXDO3HUPLVVLEOH6WUHVVHV
,IQHLWKHURIWKHFRQGLWLRQVDRUEXQGHU6HFWLRQDUHVDWLVILHGWKHQ0LFKHOO¶V
RSWLPDOLW\FULWHULDDUHQRWYDOLGDQGPXVWEHPRGLILHG6XFKPRUHJHQHUDORSWLPDOLW\
FULWHULDZHUHVWDWHGE\+HPSDQGDOVRIROORZUHDGLO\IURPWKH3UDJHU6KLHOG
FRQGLWLRQV ZKLFK ZDV XVHG IRU WKH RSWLPDO OD\RXW WKHRU\ E\ 3UDJHU DQG
5R]YDQ\DGLVFXVVHGLQ/HFWXUH6HFWLRQ
6 G.I.N. Rozvany
)RUGLIIHUHQWSHUPLVVLEOHVWUHVVHVLQWHQVLRQDQGFRPSUHVVLRQDQGVWDWLFDOO\LQ
GHWHUPLQDWHVXSSRUWFRQGLWLRQVRQHPXVWXVHWKHIROORZLQJRSWLPDOLW\FULWHULD
İ V & IRU ) İ V 7 IRU ) !

V & d İ d V 7 IRU )
LQ ZKLFK H UHSUHVHQWV D ILFWLWLRXV VWUDLQILHOG WHUPHG µDGMRLQW¶ VWUDLQ ILHOG :H
VRPHWLPHVXVHWKHVKRUWHUQRWDWLRQ
V & N& V 7 N7
)RU V 7 V& V S ZHKDYH

N7 N& N V S
DQGWKHQUHGXFHVWR
1RWH5HODWLRQVRUDUHQHFHVVDU\DQGVXIILFLHQWFRQGLWLRQVRIRSWLPDOLW\IRUWKH
UHOHYDQW FODVV RI SUREOHPV LI D IHDVLEOH VROXWLRQ H[LVWV WKH DGMRLQW VWUDLQV H DUH
NLQHPDWLFDOO\DGPLVVLEOHDQGWKHPHPEHUIRUFHV)VWDWLFDOO\DGPLVVLEOHLHVDWLV
I\LQJHTXLOLEULXP7KLVLVEHFDXVH0LFKHOO¶VRULJLQDOSUREOHPDQGLWVH[WHQVLRQLQ
DUHFRQYH[SUREOHPV
0RUHRYHUIRUHTXDOSHUPLVVLEOHVWUHVVHVLQWHQVLRQDQGFRPSUHVVLRQ0LFKHOO¶V
SUREOHPLVVHOIDGMRLQWLHWKHµUHDO¶HODVWLFVWUDLQVDQGWKHDGMRLQWVWUDLQVDUHOLQHDUO\
SURSRUWLRQDO
,W LV WR EH UHPDUNHG WKDW GXULQJ KLV SRVWGRFWRUDO ZRUN ZLWK +HPS LQ 2[IRUG
DURXQG WKH ILUVW DXWKRU XVHG WKH 3UDJHU6KLHOG FRQGLWLRQV IRU DOO WR
SRORJ\SUREOHPVZKLFKRIFRXUVHFOHDUO\LPSO\WKHH[WHQGHGRSWLPDOLW\FRQGLWLRQVLQ

7KHVWUXFWXUDOYROXPHLVJLYHQE\WKHVXPRISURGXFWVRIFURVVVHFWLRQDODUHDV$L
DQGPHPEHUOHQJWK/LGLYLGHGE\WKHDSSURSULDWHSHUPLVVLEOHVWUHVVSULPDOIRUPX
ODWLRQDVLQRUIURPWKHµGXDOIRUPXOD¶
9 37 ǻ
ZKHUH WKH YHFWRU 3 GHQRWHV WKH H[WHUQDO IRUFHV DQG ǻ WKH DGMRLQW GLVSODFHPHQWV
JLYHQE\WKHDGMRLQWVWDLQILHOG H DWSRLQWVRIDSSOLFDWLRQRIWKHVHIRUFHV
5HYLHZ RI WKH $XWKRU¶V 3DSHU RQ WKH µ6KRUWFRPLQJV¶ RI
0LFKHOO¶V7KHRU\5R]YDQ\
,QWKHDERYHSDSHUWKHDXWKRU5R]YDQ\
LSRLQWHGRXWWKHOLPLWDWLRQVRI0LFKHOO¶VSURRI
LLVWDWHGWKHUDQJHRIYDOLGLW\RI0LFKHOO¶VRULJLQDOFULWHULDIRUXQHTXDOSHU
PLVVLEOHVWUHVVHV
LLLJDYHDVLPSOHH[DPSOHWRVKRZWKDWWKHFRUUHFWHGRSWLPDOLW\FULWHULDLQ
UHVXOWLQDORZHUYROXPHWKDQWKHRULJLQDOFULWHULDE\0LFKHOOVHH)LJVE
DQGEODWHUDQG
LYSRLQWHGRXWWKHH[DPSOHVLQ0LFKHOO¶VSDSHUZKLFKLQKLVRSLQLRQKDYHD
QRQRSWLPDOWRSRORJ\VHH6HFWLRQKHUHLQ
,QKLVSDSHU5R]YDQ\WKHDXWKRUSUHVHQWHGWKUHHSURRIVRIWKHH[WHQGHG
RSWLPDOLW\FULWHULDLQ7KHVHSURRIVZHUHEDVHGRQ
DWKH3UDJHU6KLHOGFRQGLWLRQVHH/HFWXUH6HFWLRQ
EWKH.DUXVK.XKQ7XFNHU..7FRQGLWLRQ
FWKHSULQFLSOHRIYLUWXDOZRUN
,QKLVERRN+HPSVWDWHGWKHPRUHJHQHUDORSWLPDOLW\FRQGLWLRQVVHH
DERYHEXWGLGQRWSUHVHQWDQ\H[DPSOHVIRUZKLFKWKHVHJLYHDQRSWLPDOWRSRORJ\
WKDWLVGLIIHUHQWIURPWKRVHJLYHQE\0LFKHOO¶VPRUHUHVWULFWHGFULWHULD+HGLGQRW
SRLQWRXWHLWKHUZKLFKRQHVRI0LFKHOO¶VH[DPSOHVUHSUHVHQWDQRQRSWLPDOVROXWLRQ
IRU XQHTXDO SHUPLVVLEOH VWUHVVHV QRU WKH SDUW RI 0LFKHOO¶V SURRI WKDW UHQGHUV KLV
FULWHULDRIPRUHUHVWULFWHGYDOLGLW\3RVVLEO\+HPSFDUHIXOO\DYRLGHGDOOVWDWHPHQWV
ZKLFK FRXOG WDUQLVK 0LFKHOO¶V ZHOOGHVHUYHG KLVWRULFDO LPDJH +LV JHQWOHPDQO\
DWWLWXGHZDVDOVRREYLRXVZKHQKHUHYLHZHGWKHDXWKRU¶VSDSHUDQGVXJJHVWHG
WDFWIXOIRUPXODWLRQVHJµVKRUWFRPLQJV¶LQVWHDGRIµHUURUV¶
([DPSOHV ,OOXVWUDWLQJ WKH /LPLWHG 5DQJH RI 9DOLGLW\ RI

0LFKHOO¶V2SWLPDOLW\&ULWHULD
6WDWLFDOO\'HWHUPLQDWH6XSSRUW&RQGLWLRQV(TXDO3HUPLVVLEOH6WUHVVHV
,Q)LJDZHVKRZDQH[DPSOHLQZKLFKWKHVXSSRUWFRQGLWLRQVDSLQDQGDUROOHU
DUHVWDWLFDOO\GHWHUPLQDWHLHHTXLOLEULXPXQLTXHO\GHWHUPLQHVWKHUHDFWLRQVLQWKHP
LUUHVSHFWLYHRIDQ\VWDWLFDOO\DGPLVVLEOHWRSRORJ\RIWKHWUXVV7KLVPHDQVWKDWIRU
WKHORDG3ZHDOZD\VKDYHWKHUHDFWLRQV 4 3 DQG 5 3
)URPWKHYLHZSRLQWRIWUXVVWRSRORJ\RSWLPL]DWLRQWKLVLVFRPSOHWHO\HTXLYDOHQW
WRKDYLQJWKHJLYHQH[WHUQDOORDGV45DQG3DVLQ0LFKHOO¶VRULJLQDOIRUPXODWLRQ
7DNLQJ ILUVW WKH VSHFLDO FDVH V 7 V & V S LPSO\LQJ N7 N& N LQ )LJ D ZH
KDYH H T H U N DQG WKHQ WKH TXDUWHU FLUFXODU IDQ ZLWK HTXDO DGMRLQW VWUDLQV LQ
ERWK UDGLDO DQG FLUFXPIHUHQWLDO GLUHFWLRQV FOHDUO\ VDWLVILHV NLQHPDWLF DGPLVVLELOLW\
ZLWK
X Y DW SRLQW % X DW SRLQW $
8 G.I.N. Rozvany
D E
G
D (
4 $
G
*
9 N3G
D
3 G
%
ș ș
5 3 )
U
[
X Y N[ [ V S
HT V 7 H U V & \
H N H N D q

)LJXUH ([DPSOHV LOOXVWUDWLQJ WKH OLPLWHG UDQJH RI YDOLGLW\ RI 0LFKHOO¶V RSWLPDOLW\
FULWHULD
ZKHUH X DQG Y DUHWKHDGMRLQWGLVSODFHPHQWVLQGLUHFWLRQV[DQG\

8VLQJWKHSULPDOIRUPXODWLRQLQLWLVNQRZQHJ5R]YDQ\HWDOWKDWIRU
DQ\FLUFXODUIDQVHH)LJDWKHVXPRIWKHSURGXFWRIIRUFHVDQGPHPEHUOHQJWKVIRU
ERWKLWKHFLUFXODUFKRUGDQGLLWKHUDGLDOPHPEHUVLVWKHVDPHDQGHTXDOV
¦L )L /L )UT
ZKHUH)LVWKHFRQVWDQWIRUFHLQWKHFLUFXODUFKRUGULVWKHUDGLXVRIWKHIDQDQGșLVWKH
FHQWUDODQJOHRIWKHIDQ,QRXUFDVH) 3U DDQGș ʌVRWKHH[SUHVVLRQLQ
UHGXFHVWRʌ3/7KHQE\ZHJHWDWUXVVYROXPHRI
9 NS3D S3D V S
8VLQJWKHGXDOIRUPXODWLRQLQLWFDQEHVKRZQHJ5R]YDQ\HWDOWKDWWKH

UHODWLYHFLUFXPIHUHQWLDOGLVSODFHPHQWDWWKHHQGRIWKHFLUFXODUEDURIDIDQLV)LJD
' UT H T H U
6XEVWLWXWLQJ H T N H U N U D DQG T S LQWR ZH ILQG WKDW 3'

JLYHVWKHVDPHWUXVVYROXPHDVWKHSULPDOIRUPXODWLRQ
1RWH,QWKHDXWKRU¶VOHFWXUHVWKLFNFRQWLQXRXVOLQHVGHQRWHWHQVLRQPHPEHUVDQG
WKLFNEURNHQOLQHVFRPSUHVVLRQPHPEHUV
6WDWLFDOO\'HWHUPLQDWH6XSSRUW&RQGLWLRQV8QHTXDO3HUPLVVLEOH6WUHVVHV
&RQVLGHULQJ DJDLQ WKH SUREOHP LQ )LJ D DVVLJQ WKH YDOXHV V & V 7 WR WKH
SHUPLVVLEOHVWUHVVHV:LWKWKHDGMRLQWVWUDLQVJLYHQE\ZHKDYH
HT V7 HU V& V 7
7KH DERYH DGMRLQW VWUDLQV DUH NLQHPDWLFDOO\ DGPLVVLEOH VHH )LJ D 7KHQ IURP
SULPDOIRUPXODWLRQE\ZLWK
9 S3D V 7 V & S3D V 7 S3D V 7
8VLQJGXDOIRUPXODWLRQZHKDYHE\DQG)LJVDDQGD
HT V7 HU V 7 U D DQG T S
DQGWKHQDQGFRQILUPWKHYROXPHYDOXHLQ

1RWH)RUWKHVWDWLFDOO\GHWHUPLQDWHVXSSRUWFRQGLWLRQVRIWKLVH[DPSOH 0LFKHOO¶V
RSWLPDOLW\FULWHULDLQZRXOGDOVRJLYHWKHFRUUHFWWUXVVWRSRORJ\+RZHYHUIRU
FDOFXODWLQJWKHYROXPHIURPWKHGXDOIRUPXODLQRQHZRXOGKDYHWRWDNHN LQ
DQGWKHQPXOWLSO\WKHUHVXOWE\ V 7 V & 6LPLODUIRUPXODHDUHXVHGLQ
WKHH[DPSOHVRI0LFKHOO+RZHYHUWKLVµGXDO¶PHWKRGLVUDWKHUFXPEHUVRPH
DQGLVRQO\YDOLGLIWKHVXPRISURGXFWVRIEDUIRUFHVDQGEDUOHQJWKVLVHTXDOIRU
FRPSUHVVLRQDQGWHQVLRQEDUV
¦ /L )L ¦ /L )L
& 7
ZKLFKZDVDFWXDOO\SRLQWHGRXWWRWKHDXWKRUE\0DULDQR9i]TXH](VSL&OHDUO\D
JUHDWDGYDQWDJHRIWKHH[WHQGHGRSWLPDOFULWHULDLQLVWKDWWKHFRUUHVSRQGLQJGXDO
IRUPXODLQLVYDOLGHYHQLIWKHUHVWULFWLRQLVQRWVDWLVILHG
6WDWLFDOO\,QGHWHUPLQDWH6XSSRUW&RQGLWLRQV(TXDO3HUPLVVLEOH6WUHVVHV
&RQVLGHULQJWKHSUREOHPLQ)LJEWKHH[WHUQDOORDG3FDQEHWUDQVPLWWHGE\DWUXVV
WRDQ\DUELWUDU\SRLQWVRIWKHULJLGOLQHVXSSRUW()7KLVVXSSRUWFRQGLWLRQLVVWDWLFDOO\
LQGHWHUPLQDWHEHFDXVHWKHORFDWLRQGLUHFWLRQDQGPDJQLWXGHRIWKHUHDFWLRQIRUFHV
GHSHQGRQWKHOD\RXWRIWKHWUXVV$VLPLODUSUREOHPZDVFRQVLGHUHGE\3UDJHUDQG
5R]YDQ\D)LJLQWKDWSDSHU

10 G.I.N. Rozvany
7KHDGMRLQWGLVSODFHPHQWILHOGLQ)LJEJLYHVSULQFLSDODGMRLQWVWUDLQVRI
H H N
DW rq WRWKHD[LV[7KLVVWUDLQILHOGLVNLQHPDWLFDOO\DGPLVVLEOHEHFDXVHLWVDWLVILHV

X Y
DORQJWKHULJLGOLQHVXSSRUW(),WDOVRVDWLVILHVWKHLQHTXDOLW\LQEHFDXVHVWUDLQVLQ
QRQSULQFLSDOGLUHFWLRQVFDQQRWKDYHJUHDWHUDEVROXWHYDOXHWKDQWKHSULQFLSDOVWUDLQV
LQ
)RUWKHUHVXOWLQ)LJEWKHSULPDOYROXPHIRUPXODLQZLWK V 7 V & V S
JLYHV
9 G 3 V S 3G V S
)RUWKHGXDOIRUPXODLQWKHDGMRLQWGLVSODFHPHQWDWWKHSRLQW*LQ)LJELVE\WKH

IUDPHGUHODWLRQVLQ)LJE
' Y* G V S
DQGWKHQFRQILUPVWKHUHVXOWLQ
6WDWLFDOO\,QGHWHUPLQDWH6XSSRUW&RQGLWLRQV8QHTXDO3HUPLVVLEOH6WUHVVHV
:HFRQVLGHUWKHSUREOHPLQ)LJEEXWZLWKXQHTXDOSHUPLVVLEOHVWUHVVHVLQWHQVLRQ
DQGFRPSUHVVLRQ V & V 7
)RUWKDWSUREOHP0LFKHOO¶VRULJLQDORSWLPDOLW\FULWHULDZRXOGJLYHWKHOD\RXWLQ
)LJEEXWZLWKWKHDERYHSHUPLVVLEOHVWUHVVHVWKHYROXPHZRXOGWKHQEHFRPH
9 G V 7 V 7 3 G 3 V 7
2Q WKH RWKHU KDQG WKH VROXWLRQ EDVHG RQ WKH H[WHQGHG RSWLPDOLW\ FULWHULD LQ
VKRZQLQ)LJEJLYHVWKHIROORZLQJYROXPHE\SULPDOIRUPXODWLRQ
9 G 3 V 7 G 3 V 7 V 7 | G 3V 7
ZKLFK LV DERXW SHU FHQW VPDOOHU WKDQ WKH YROXPH RI WKH VROXWLRQ EDVHG RQ
0LFKHOO¶VFULWHULD7KHGXDOIRUPXODLQZLWK ǻ Y* G V 7 FRQILUPVWKH
UHVXOWLQ
7KLVVKRZVWKHVLJQLILFDQFHRIXVLQJWKHPRGLILHGFULWHULDIRUVWDWLFDOO\LQGHWHU
PLQDWH VXSSRUW FRQGLWLRQV DQG XQHTXDO SHUPLVVLEOH VWUHVVHV 7KH DERYH FRXQWHU
H[DPSOHZDVRULJLQDOO\SUHVHQWHGE\5R]YDQ\
V& V7
X [ V 7 Y [ V 7
T H V 7 H V 7

/ G ) 3
'
*

3
U )
/ G ) 3
D G E

)LJXUH D $GMRLQW GLVSODFHPHQW IRU D FLUFXODU IDQ E 7KH FRUUHFW VROXWLRQ RI WKH
SUREOHPLQ)LJEIRUXQHTXDOSHUPLVVLEOHVWUHVVHV
7KH 2SWLPDO 2ULHQWDWLRQ RI 7UXVV 0HPEHUV DW D 3RLQW RI D
/LQH6XSSRUW
&RQVLGHULQJDSRLQW$RIDOLQHVXSSRUW)LJDE\WKHDGMRLQWVWUDLQVDORQJWZR
WUXVVPHPEHUVDUH H V 7 DQG H V & ZKLFKDUHDOVRSULQFLSDOVWUDLQVGXH
WRWKHLQHTXDOLW\LQ
,QD0RKUFLUFOHUHSUHVHQWLQJWKHDGMRLQWVWUDLQVDWSRLQW$ZHKDYH)LJE
U V 7 V & D V & V 7
FRVN DU
7KHQHOHPHQWDU\WUDQVIRUPDWLRQVLPSO\
V V&
N DUFFRV 7
V7 V&

12 G.I.N. Rozvany
)RU V 7 V & V S JLYHV N q DQG IRU V & V 7 ZH JHW N q SUR
YLGLQJDQLQGHSHQGHQWFRQILUPDWLRQRIWKHUHVXOWVLQ)LJVEDQGE
7KHRSWLPDORULHQWDWLRQRIPHPEHUVDORQJOLQHVXSSRUWVZDVGHULYHGSUHYLRXVO\
5R]YDQ\IURP+HPS¶VHTXDWLRQVEXWWKHDERYHGHULYDWLRQLVPXFK
VLPSOHUDQGHDVLHUWRXQGHUVWDQGIRUQRQVSHFLDOLVWV
&ULWLFDO5HH[DPLQDWLRQRIWKH)LUVW([DPSOHE\0LFKHOO
)RUDSRLQWORDGDQGµIRUFHVHTXLYDOHQWWRDIRUFH«DQGDFRXSOH«RYHUDVPDOOFLUFOH¶
0LFKHOOSUHVHQWVWKHVROXWLRQLQ)LJFKHUHUHGUDZQDIWHU5R]YDQ\
)RUWKHRSWLPDOYROXPH0LFKHOOJLYHVLQRXUQRWDWLRQ
D
9 )D OQ
U V 7 V &
,QRXU)LJFZHVKRZDFLUFXODUVXSSRUWEXW0LFKHOOUHIHUVWRGLVWULEXWHGIRUFHV
DORQJDFLUFOH
,WLVXQIRUWXQDWHWKDW0LFKHOOLVQRWZLWKXVWRFODULI\KLVUHVXOWVEXWWKLVDXWKRUKDV
WKHIROORZLQJGLIILFXOWLHVLQDFFHSWLQJWKHVROXWLRQLQ)LJFDQG
D7KHUHODWLRQFOHDUO\LPSOLHVWKDWWKHDERYHVROXWLRQLVPHDQWWREHDOVR
IRU GLIIHUHQW SHUPLVVLEOH VWUHVVHV LQ WHQVLRQ DQG FRPSUHVVLRQ )RU WKDW FDVH
0LFKHOO¶VRSWLPDOLW\FULWHULD LQ DUH UHVWULFWHG WR JLYHQ H[WHUQDO IRUFHV RU
HTXLYDOHQWO\VWDWLFDOO\GHWHUPLQDWHVXSSRUWFRQGLWLRQV)RUWKHSUREOHPLQ)LJ
F WKH RSWLPDO GLVWULEXWLRQ RI WKH IRUFHV DORQJ WKH FLUFOH ZLWK WKH UDGLXV U
GHSHQGVRQWKHUDWLR D U VRWKHVHIRUFHVDUHQRWJLYHQ7KLVPHDQVWKDWZHDUH
FKRRVLQJWKHIRUFHVDORQJWKHFLUFOHRSWLPDOO\ZKLFKLVHTXLYDOHQWWRRSWLPL]LQJ
IRUDFLUFXODUOLQHVXSSRUW,QWKDWFDVHKRZHYHUZHKDYHWRXVHWKHH[WHQGHG
RSWLPDOLW\FULWHULDLQ
E7KHVROXWLRQLQ)LJFIDLOVWRVDWLVI\WKHFRQGLWLRQIRURSWLPDOPHPEHU
GLUHFWLRQVDORQJOLQHVXSSRUWVVRLWFDQQRWEHRSWLPDO
F7KHSUREOHPLQ)LJVEDQGELVWKHVSHFLDOOLPLWLQJFDVHRIWKHSUREOHPLQ
)LJFZLWK U o f )RUWKLVVSHFLDOFDVHLWZDVVKRZQWKDWWKHH[WHQGHGRS
WLPDOLW\FULWHULDLQJLYHDORZHUYROXPHWKDQ0LFKHOO¶VRULJLQDOFULWHULD
G)RUWKHSUREOHPLQ)LJF'HZKXUVWDQG6ULWKRQJFKDLREWDLQHGD
ORZHU YROXPH WKDQ 0LFKHOO¶V VROXWLRQ ZKHQ WKH\ XVHG WKH RSWLPDO PHPEHU
DQJOHVRIDQGJLYHQE\WKHH[WHQGHGRSWLPDOLW\FULWHULD

D E
q N
q N
U
$ H V & N
D
N
H V & H V 7
H V 7
3
U
%
D

)LJXUHDDQGE2SWLPDOGLUHFWLRQVRIWUXVVPHPEHUVDWOLQHVXSSRUWVF0LFKHOO¶VILUVW
H[DPSOH
1RWH 0LFKHOO¶V VROXWLRQ LV FRPSOHWHO\ FRUUHFW IRU HTXDO SHUPLVVLEOH VWUHVVHV
V 7 V & V S ,Q KLV ILUVW H[DPSOH )LJ F KHUHLQ KH VWDUWV RII ZLWK KDYLQJ D
FRQFHQWUDWHGFRXSOHDQGIRUFHDWSRLQW%LQ)LJFZKLFKLVHTXLYDOHQWWR U o
)RU WKDW FDVH JLYHV DQ LQILQLWH YROXPH $ PLQLPL]DWLRQ SUREOHP JLYLQJ DQ
LQILQLWHREMHFWLYHIXQFWLRQYDOXHGRHVQRWPDNHPXFKVHQVHEHFDXVHWKLVZRXOGLPSO\
WKDWDQ\RWKHUIHDVLEOHVROXWLRQLVHTXDOO\RSWLPDO
0LFKHOO¶V VHFRQG WKLUG DQG IRXUWK H[DPSOHV DUH FRUUHFW EHFDXVH WKH\ DUH IRU
JLYHQH[WHUQDOIRUFHVHTXLYDOHQWWRVWDWLFDOO\GHWHUPLQDWHVXSSRUWFRQGLWLRQV+LV
ILIWKH[DPSOHKDVWKHVDPHSUREOHPDVWKHRQHGLVFXVVHGDERYHIRUWKHILUVWH[DPSOH
14 G.I.N. Rozvany
&RQFOXGLQJ5HPDUNV
L([DFWDQDO\WLFDOEHQFKPDUNVDUHH[WUHPHO\LPSRUWDQWLQVWUXFWXUDOWRSRORJ\
RSWLPL]DWLRQ IRU FKHFNLQJ RQ WKH YDOLGLW\ HIILFLHQF\ DQG FRQYHUJHQFH RI QX
PHULFDOPHWKRGVZKLFKKDYHPDQ\VRXUFHVRIHUURU
LL 0LFKHOO¶V ODQGPDUN SDSHU LV RI SLYRWDO LPSRUWDQFH WR WKLV ILHOG
+RZHYHULWLVEHQHILFLDOWRDQ\WKHRU\WRGHWHUPLQHLWVUDQJHRIYDOLGLW\DQGWR
GHULYH LWV H[WHQVLRQ WR D PXFK EURDGHU FODVV RI SUREOHPV 7KLV KDV EHHQ DW
WHPSWHG LQ WKLV OHFWXUH )XUWKHU JHQHUDOL]DWLRQV RI 0LFKHOO¶V WKHRU\ ZLOO EH
SUHVHQWHGLQWKHQH[WOHFWXUH
5HIHUHQFHVDUHOLVWHGDIWHUWKHODVWOHFWXUHRIWKHDXWKRU
6WUXFWXUDO7RSRORJ\2SWLPL]DWLRQ672±([DFW
$QDO\WLFDO6ROXWLRQV3DUW,,
*HRUJH,15R]YDQ\ DQG(ULND3LQWHU

%XGDSHVW+XQJDU\
,QWURGXFWLRQ
,QWKLVOHFWXUHZHUHYLHZWKHRULJLQDQGEDVLFIHDWXUHVRIWKHRSWLPDOOD\RXWWKHRU\
3UDJHU DQG 5R]YDQ\ D LQFOXGLQJ RSWLPDO UHJLRQV DQG GLIILFXOWLHV ZLWK
VRFDOOHG 2UHJLRQV 7KHQ ZH GLVFXVV YDULRXV H[WHQVLRQV RI WKLV WKHRU\ LQFOXGLQJ
PXOWLSOH ORDG FRQGLWLRQV DQG PXOWLSOH GHVLJQ FRQVWUDLQWV SUREDELOLVWLF GHVLJQ DQG
SUHH[LVWLQJPHPEHUV
2SWLPDO/D\RXW7KHRU\3UDJHUDQG5R]YDQ\D
3UHOLPLQDULHV
,WKDVEHHQVKRZQHJ'UXFNHU*UHHQEHUJDQG3UDJHUWKDWIRUHODVWLFLGHDO
SODVWLFVWUXFWXUHVDORZHUERXQGRQWKHFROODSVHORDGLVJLYHQE\DQ\VWDWLFDOO\DG
PLVVLEOH VWUHVV ILHOG ZKLFK QRZKHUH YLRODWHV WKH \LHOG FRQGLWLRQ ¶6WDWLFDOO\
DGPLVVLEOH¶PHDQVWKDWDVWUHVVILHOGVDWLVILHVVWDWLFERXQGDU\FRQGLWLRQVDQGHTXL
OLEULXP$GHVLJQEDVHGRQWKHDERYH¶ORZHUERXQGWKHRUHP¶LVWHUPHG¶SODVWLFORZHU
ERXQGGHVLJQ¶,WQHHGQRWWDNHNLQHPDWLFFRPSDWLELOLW\FRQGLWLRQVLQWRFRQVLGHUD
WLRQ
,QWKHSUREOHPLQ)LJZHKDYHDEHDPZLWKHTXDOSRVLWLYHDQGQHJDWLYH\LHOG
PRPHQWV ZKLFK DUH FRQVWDQW RYHU WKH HQWLUH EHDP OHQJWK 7KH PRPHQW GLDJUDP
VKRZQ LQ WKLFN OLQH LV VWDWLFDOO\ DGPLVVLEOH EXW NLQHPDWLFDOO\ LQDGPLVVLEOH IRU D
OLQHDUO\HODVWLFEHDPYLRODWLQJNLQHPDWLFERXQGDU\FRQGLWLRQV<HWLQµSODVWLFORZHU
ERXQGGHVLJQ¶LWFDQEHXVHGEHFDXVHWKHEHQGLQJPRPHQWVQRZKHUHH[FHHGWKH\LHOG
PRPHQWV
6YHGKDVVKRZQWKDWRSWLPDOOHDVWZHLJKWHODVWLFWUXVVHVXQGHUDVWUHVV
FRQVWUDLQWDQGDVLQJOHORDGFRQGLWLRQDUHDOZD\V¶VWDWLFDOO\GHWHUPLQDWH¶RUDFRQYH[
FRPELQDWLRQRIVWDWLFDOO\GHWHUPLQDWHVROXWLRQV7KLVWHUPPHDQVWKDWLWKHUHDUHQR
¶UHGXQGDQW¶PHPEHUVLQWKHVWUXFWXUHDµUHGXQGDQWPHPEHU¶FDQEHUHPRYHGZLWKRXW
PDNLQJWKHVWUXFWXUHXQVWDEOHLLLWVLQWHUQDOIRUFHVFDQEHFRPSXWHGRQWKHEDVLVRI

DOI 10.1007/978-3-7091-1643-2_2, © CISM, Udine 2014
16 G.I.N. Rozvany and E. Pinter
VWDWLFHTXDWLRQVRQO\LJQRULQJNLQHPDWLFFRPSDWLELOLW\UHTXLUHPHQWVDQGWKHUHIRUH

LLLWKHLQWHUQDOIRUFHVGRQRWGHSHQGRQPHPEHUVL]HV
/
3 3
/
0\ 3/
0\ 3/

)LJXUH([DPSOHRIXVLQJWKHORZHUERXQGWKHRUHPRISODVWLFGHVLJQ
%DUWD H[WHQGHG 6YHG¶V WKHRUHP RQ VWDWLFDO GHWHUPLQDF\ WR WUXVVHV ZLWK
ORFDOEXFNOLQJDQG0F.HRZQ WR DQ\ FRPELQDWLRQ RI VWUHVV DQG GLV
SODFHPHQW FRQVWUDLQWV XQGHU D VLQJOH ORDG 3HGHUVHQ REWDLQHG PRUH
ULJRURXV SURRIV RI %DUWD¶V WKHRUHP DQG H[WHQGHG WKH VWDWLFDO GHWHUPLQDF\
SURSHUW\ WR WUXVVHV ZLWK YDULDEOH VXSSRUW FRQGLWLRQV 3HGHUVHQ 7KH
HTXLYDOHQWRI6YHG¶VWKHRUHPZDVDOVRSURYHGULJRURXVO\E\$FKW]LJHULQD
FRPSUHKHQVLYHUHYLHZRQRSWLPL]DWLRQRIGLVFUHWHVWUXFWXUHV
:HPD\DGGWKDWDFFRUGLQJWR:DVLXW\QVNLDQG%UDQGWWKHILUVWSURRIRI
VWDWLFDO GHWHUPLQDF\ RI VLQJOHORDG WUXVVHV H[SUHVVHG DV WKH ¶WKHRUHP RQ WKH
QRQH[LVWHQFHRIVWDWLFDOO\LQGHWHUPLQDWHODWWLFHVRIXQLIRUPVWUHQJWK¶PHDQLQJIXOO\
VWUHVVHGWUXVVHVLVGXHWR/pY\
6LQFHWKHRSWLPDOVROXWLRQIRUFHUWDLQFODVVHVRIUHGXQGDQWVWUXFWXUHVHJWUXVVHV
JULOODJHV ULJLG IUDPHV ZLWK RQH ORDG FRQGLWLRQ LV VWDWLFDOO\ GHWHUPLQDWH ZH FDQ
HQODUJHRXUIHDVLEOHVHWIRURSWLPDOHODVWLFGHVLJQZLWKVWDWLFDQGNLQHPDWLFDGPLV
VLELOLW\WRLQFOXGHDOOVWDWLFDOO\DGPLVVLEOHVROXWLRQVDOVRWKRVHYLRODWLQJNLQHPDWLF
DGPLVVLELOLW\7KLVLVEHFDXVHE\WKHDERYHWKHRUHPVWKHILQDOVROXWLRQZLOOEHVWDWL
FDOO\ GHWHUPLQDWH DQG WKDW DXWRPDWLFDOO\ VDWLVILHV HODVWLF FRPSDWLELOLW\ ,Q RWKHU
Structural Topology Optimization (STO)… Part II 17
ZRUGVWKHRSWLPDOVROXWLRQIRUWKHHQODUJHGIHDVLEOHVHWZLOODOZD\VEHFRQWDLQHGLQ
WKHRULJLQDOVPDOOHUIHDVLEOHVHW7KLVPHDQVWKDWRSWLPDOHODVWLFGHVLJQRIFHUWDLQ
VWUXFWXUHV ZLWK RQH ORDG FRQGLWLRQ UHGXFHV WR WKH RSWLPDO SODVWLF GHVLJQ RI WKHVH
VWUXFWXUHV0RUHRYHUDVWUHVVEDVHGGHVLJQRIDVWDWLFDOO\GHWHUPLQDWHWUXVVFDQRQO\
EHRSWLPDOLIDOOPHPEHUVGHYHORSWKHSHUPLVVLEOHVWUHVV
%DVLF)HDWXUHVRIWKH2SWLPDO/D\RXW7KHRU\
7KLV WKHRU\ ZDV DOUHDG\ LQWURGXFHG LQ HIIHFW LQ WKH DXWKRU¶V ILUVW ERRN 5R]YDQ\
IRUIOH[XUDOV\VWHPVHJJULOODJHVVKHOOVDQGIRUPXODWHGLQDPRUHFRQFLVH
IRUP ODWHU 3UDJHU DQG 5R]YDQ\ D ,W LV EDVHG RQ WKH 3UDJHU6KLHOG
RSWLPDOLW\ FRQGLWLRQ IRU SODVWLF GHVLJQ +RZHYHU WKH QHZ HOHPHQW LQ WKH OD\RXW
WKHRU\LVWKDWLWDOVRJLYHVRSWLPDOLW\FRQGLWLRQVIRU¶YDQLVKLQJ¶PHPEHUVRI]HUR
FURVVVHFWLRQDO DUHD LQ WHUPV RI ¶DGMRLQW¶ VWUDLQV DORQJ WKHVH PHPEHUV ,Q RWKHU
ZRUGVRSWLPDOOD\RXWWKHRU\VWDUWVRIIZLWKDJURXQGVWUXFWXUHRUVWUXFWXUDOXQLYHUVH
RI DOO SRWHQWLDO PHPEHUV DQG VHOHFWV WKH RSWLPDO PHPEHUV RI QRQ]HUR
FURVVVHFWLRQDODUHDRXWRIWKRVH
8VLQJHLWKHURQHRIWKHDERYHWKHRULHVZHQHHGWRILQG
L D NLQHPDWLFDOO\ DGPLVVLEOH ¶DGMRLQW¶ GLVSODFHPHQWVWUDLQ ILHOG VDWLVI\LQJ
NLQHPDWLFFRQWLQXLW\DQGERXQGDU\FRQGLWLRQV
LL D VWDWLFDOO\ DGPLVVLEOH VWUHVV ILHOG VDWLVI\LQJ HTXLOLEULXP DQG VWDWLFDO
ERXQGDU\FRQGLWLRQV
LLLVXFKWKDWFHUWDLQµVWDWLFNLQHPDWLF¶RSWLPDOLW\FULWHULDDUHDOVRIXOILOOHG
,QWKHVHRSWLPDOLW\FULWHULDWKHDGMRLQWVWUDLQVDUHJLYHQE\WKH VXEJUDGLHQW RI WKH
¶VSHFLILF FRVW IXQFWLRQ¶ ZLWK UHVSHFW WR VWUHVVHV RU VWUHVV UHVXOWDQWV HJ PHPEHU
IRUFHV)RUEHDPPRPHQWV0
7KHVXEJUDGLHQWRIDIXQFWLRQLVWKHXVXDOJUDGLHQWEXWDWGLVFRQWLQXLWLHVRIWKH
JUDGLHQWDQ\FRQYH[FRPELQDWLRQRIWKHDGMDFHQWJUDGLHQWVFDQEHWDNHQ
)RUVLJQLQGHSHQGHQWVWUHVVEDVHGGHVLJQRIWUXVVHVDQGJULOODJHVRIJLYHQGHSWKIRU
H[DPSOHWKH¶VSHFLILFFRVWIXQFWLRQV¶UHSUHVHQWLQJLQWKLVWH[WFURVVVHFWLRQDODUHDV
$DUH
$ N ) DQG $ N0
ZKHUHNLVDFRQVWDQW)LVDPHPEHUIRUFHDQG0LVDEHQGLQJPRPHQW
7KHQHJIRUWUXVVHVWKHRSWLPDOLW\FRQGLWLRQVUHGXFHWRWKRVHRI0LFKHOO
H N VJQ ) IRU ) z H d N IRU )
ZKHUH İ LVWKHDGMRLQWVWUDLQ
$ $ $
$ N)
NF ) NW )
$ N) $ N)
% %
) % ) % % )
N NF NW
H H H
N N NW
N ) N ) NF )
D E F

)LJXUH([DPSOHVRIVSHFLILFFRVWIXQFWLRQVDQGWKHFRUUHVSRQGLQJDGMRLQWVWUDLQV
)RUJULOODJHVZHKDYH
N N VJQ 0 IRU 0 z N d N IRU 0
ZKHUH N X

LVWKHDGMRLQWEHDPFXUYDWXUHDQG X LVWKHDGMRLQWEHDPGHIOHFWLRQ
7KHVSHFLILFFRVWIXQFWLRQDQGDGMRLQWVWUDLQYDOXHVIRU0LFKHOOWUXVVHVDUHVKRZQ
LQ)LJD7KHVHUHODWLRQVDUHH[WHQGHGWRWUXVVHVZLWKDSUHVFULEHGPLQLPXPFURVV
VHFWLRQ DQG UHVSHFWLYHO\ HTXDO DQG XQHTXDO SHUPLVVLEOH VWUHVVHV LQ WHQVLRQ DQG
FRPSUHVVLRQLQ)LJVEDQGF
$6LPSOH,OOXVWUDWLYH([DPSOH8VLQJWKH3UDJHU6KLHOG&RQGLWLRQ
&RQVLGHUDFODPSHGEHDPRIFRQVWDQWGHSWKDQGYDULDEOHZLGWKZLWKDFHQWUDOSRLQW
ORDG )LJ D 7KHQ E\ WKH DGMRLQW FXUYDWXUHV LH VHFRQG GHULYDWLYHV RI WKH
DGMRLQW EHDP GHIOHFWLRQV IRU SRVLWLYH DQG QHJDWLYH PRPHQWV UHVSHFWLYHO\ DUH
N N DQG N N
)RUDQ\V\VWHPRIGRZQZDUGIRUFHVRQWKHEHDPWKHVLJQRIWKHPRPHQWGLDJUDP
0PD\RQO\FKDQJHDWWZRSODFHV7KHVLPSOHVWRIVXFKORDGVDVLQJOHSRLQWORDGLV
FRQVLGHUHG LQ )LJ )RU WZR QHJDWLYH DQG RQH SRVLWLYH VHJPHQWV RI WKH PRPHQW
GLDJUDPRQHJHWVWKHDGMRLQWFXUYDWXUHVLQ)LJE)RURXUVLPSOHORDGWKLVJLYHVWKH
PRPHQWGLDJUDPLQ)LJFEXWWKH]HURPRPHQWSRLQWVZRXOGEHWKHVDPHIRUPRVW
GRZQZDUGORDGV

3
D

/
N N N N
E
X

N N

/ /
/ F

3/
0
3/

)LJXUH([DPSOHRIDSSO\LQJWKH3UDJHU6KLHOGRSWLPDOLW\FRQGLWLRQSUREOHPRI+H\
PDQ
7KHJHQHUDOIHDWXUHRIHDUOLHUDSSOLFDWLRQVRIWKH3UDJHU6KLHOGFRQGLWLRQZDVWKDW
QRPHPEHUVRUFRPSRQHQWVGLVDSSHDUHGIURPWKHVWUXFWXUH7KLVFRQGLWLRQZDVRIWHQ
DSSOLHGWRSODWHVRUVKHOOVZLWKVHYHUDOVWUHVVFRPSRQHQWVVHHWKHDXWKRU¶VILUVWERRN
5R]YDQ\ ,Q OD\RXW RSWLPL]DWLRQ KRZHYHU PRVW RI WKH RULJLQDO PHPEHUV
YDQLVKIURPWKHJURXQGVWUXFWXUHEXWRSWLPDOLW\FULWHULDXVXDOO\LQHTXDOLWLHVPXVWEH
VDWLVILHGDORQJYDQLVKLQJPHPEHUVDOVR
2SWLPDO¶5HJLRQV¶RIWKH$GMRLQW6WUDLQ)LHOGIRU0LFKHOO6WUXFWXUHV
,WIROORZVIURPWKDWLQ0LFKHOOWUXVVHVWKHPHPEHUVDUHLQWKHGLUHFWLRQRIWKH
DGMRLQW SULQFLSDO VWUDLQV RI D FRQVWDQW PDJQLWXGH N DQG WKH DGMRLQW VWUDLQV PD\
QRZKHUH H[FHHG WKLV YDOXH 7KH RSWLPDO WRSRORJ\ XVXDOO\ FRQVLVWV RI VHYHUDO ¶UH
JLRQV¶'HQRWLQJWKHSULQFLSDODGMRLQWVWUDLQVE\ İ İ DWDQ\SRLQWRID'WUXVV
IRUHTXDOSHUPLVVLEOHVWUHVVHVLQWHQVLRQDQGFRPSUHVVLRQZHPD\KDYHD
7UHJLRQZLWKDWHQVLOHDQGDFRPSUHVVLRQPHPEHUDWULJKWDQJOHV İ İ N
6UHJLRQ ZLWK PHPEHUV KDYLQJ IRUFHV RI WKH VDPH VLJQ LQ DQ\ GLUHFWLRQ
İ İ H L N L
5UHJLRQVZLWKRQO\RQHPHPEHUDWDQ\SRLQW İ N İ d N RU
2UHJLRQZLWKQRPHPEHUV İ d N İ d N ZLWK N V S ZKHUH V S LVWKH
SHUPLVVLEOHVWUHVVLQERWKWHQVLRQDQGFRPSUHVVLRQ
7KHV\PEROVXVHGIRURSWLPDOUHJLRQVDUHVKRZQLQ)LJD
)RUXQHTXDOSHUPLVVLEOHVWUHVVHVLQWHQVLRQDQGFRPSUHVVLRQZHKDYH
IRU7UHJLRQV H N7 H N &
IRU 5UHJLRQV H N7 RU H N & DQG N7 t H t N& ZLWK V & N&
V 7 N7
'HSHQGLQJRQWKHVLJQRIWKHIRUFHV6DQG5UHJLRQVPD\EHIXUWKHUVXEGLYLGHG
LQWR 6 6 5 RU 5 UHJLRQV
1RWH0RVWRIWKH0LFKHOOOLWHUDWXUHGHDOVZLWK7UHJLRQVEHFDXVHRIWKHOLQNZLWK
VOLSOLQHVLQSODVWLFLW\+HQFN\3UDQGWOQHWVZKLFKZDVGHYHORSHGHDUOLHU
6RPHQHZW\SHVRIUHJLRQVIRU0LFKHOOVWUXFWXUHVZLOOEHLQWURGXFHGLQ6HFWLRQ
2SWLPDOUHJLRQVZHUHDOVRGHULYHGIRU'SHUIRUDWHGDQGFRPSRVLWHFRQWLQXDE\
5R]YDQ\2OKRII%HQGVRH2QJDQG6]HWR2QJ5R]YDQ\DQG6]HWR
DQG6]HWRIRUDUHYLHZVHH5R]YDQ\%HQGVRHDQG.LUVFKRUDERRNE\
5R]YDQ\7KHRSWLPDOUHJLRQVIRU'SHUIRUDWHGFRQWLQXDZHUHSUHVHQWHGLQ
DQRXWVWDQGLQJSDSHUE\2OKRII5¡QKROWDQG6FKHHO
$6LPSOH([DPSOHRI$SSO\LQJWKH3UDJHU5R]YDQ\D/D\RXW7KH

RU\WRD0LFKHOO7\SH3UREOHP
,Q)LJEZHVKRZVRPHRSWLPDOWRSRORJLHVDQGLQ)LJFWKHFRUUHVSRQGLQJRSWLPDO
DGMRLQWGLVSODFHPHQWILHOGV7KHUHDUHWZROLQHVXSSRUWVDWULJKWDQJOHV:HXVHD
QRUPDOL]HGIRUPXODWLRQZLWKN
(OHPHQWDU\FDOFXODWLRQVVKRZWKDWIRUWKHDGMRLQWGLVSODFHPHQWV X DQG Y LQWKH
XSSHUUHJLRQZHKDYHWKHSULQFLSDODGMRLQWVWUDLQV
H H
DWWRWKHKRUL]RQWDO7KLVFRUUHVSRQGVWRD7UHJLRQGHQRWHGE\DQDUURZFURVV
LQ)LJF,QWKHORZHUUHJLRQZHKDYH
H H
ZKLFKVLJQLILHVDQ 5 UHJLRQGHQRWHGE\DGRXEOHDUURZ%RWKVDWLVI\WKHRSWLPDOLW\
FRQGLWLRQVLQ
7 5 5 9 9
D
E 7
E
X
Y [

E
E X 5
Y \
R
[
R d E L d R
\ F
E

)LJXUHD6\PEROVXVHGIRURSWLPDOUHJLRQVEF([DPSOHRIDSSOLFDWLRQRIRSWLPDO
OD\RXWWKHRU\WR0LFKHOOWUXVVHV
,WFDQEHUHDGLO\VHHQWKDWDORQJWKHOLQHVXSSRUWVZLWK[ DQG\ ZHKDYH

X Y VDWLVI\LQJ WKH NLQHPDWLF ERXQGDU\ FRQGLWLRQV 0RUHRYHU DORQJ WKH
UHJLRQERXQGDU\ZLWK\ [ERWKDGMRLQWGLVSODFHPHQWILHOGVLQ)LJFJLYH
X Y [
VDWLVI\LQJNLQHPDWLFFRQWLQXLW\
,WFDQEHVHHQIURP)LJEWKDWLQWKHXSSHUUHJLRQZHKDYHWZREDUWUXVVHVZLWK
RQHWHQVLRQDQGRQHFRPSUHVVLRQEDULQWKHORZHUUHJLRQZHKDYHVLQJOHEDUWUXVVHV
LQFRPSUHVVLRQ,IWKHORDGLVDFWLQJRQWKHUHJLRQERXQGDU\ZHPD\KDYHDQ\FRQYH[
FRPELQDWLRQRIDWZREDUWUXVVDQGDVLQJOHEDUWUXVV
2SWLPDO7RSRORJLHVIRU/HDVW:HLJKW*ULOODJHV
7KHWKHRU\RIWKHRSWLPDOWRSRORJ\RIJULOODJHVEHDPV\VWHPVDFWXDOO\SUHFHGHGWKH
GHYHORSPHQWRIWKHRSWLPDOOD\RXWWKHRU\3UDJHUDQG5R]YDQ\D7KHJULOODJH
WKHRU\HJ5R]YDQ\DDQGEZDVGHYHORSHGE\WKHDXWKRU¶VUHVHDUFKJURXSEXW
LWKDVVRPHHOHPHQWVRI0RUOH\¶VWKHRU\IRURSWLPDOUHLQIRUFHPHQWLQFRQFUHWH
VODEV5HYLHZVRIWKHJULOODJHWKHRU\PD\EHIRXQGLQVHYHUDOWH[WVHJ5R]YDQ\DQG
+LOO3UDJHUDQG5R]YDQ\EDQG5R]YDQ\%HQGVRHDQG.LUVFK
([DFWRSWLPDOJULOODJHWRSRORJLHVVKRXOGDFWXDOO\EHXVHGPRUHRIWHQDVEHQFK
PDUNVEHFDXVHWKH\KDYHWKHIROORZLQJDGYDQWDJHV
L7KHRSWLPDOJULOODJHWKHRU\KDVDGYDQFHGPXFKIXUWKHUWKDQWKHWUXVVWKHRU\
EHFDXVHH[DFWRSWLPDOJULOODJHWRSRORJLHVDUHDYDLODEOHIRUDOPRVWDOOSRVVLEOHVXSSRUW
DQGORDGFRQGLWLRQVDQGWKHVHDQDO\WLFDOVROXWLRQVFDQHYHQEHJHQHUDWHGE\FRP
SXWHU VHH HJ 5R]YDQ\ %HQGVRH DQG .LUVFK )LJ $ FRPSOH[ RSWLPDO
JULOODJHWRSRORJ\LVVKRZQRQWKHFRYHURIWKHDXWKRU¶VILUVWERRN5R]YDQ\
VHH)LJKHUHLQ
LL0LFKHOOWUXVVHVLJQRUHEXFNOLQJZKLFKSOD\VDPXFKPRUHLPSRUWDQWUROHIRU
WUXVVHVWKDQIRUJULOODJHV
LLL*ULOODJHVPD\KDYHWKUHHW\SHVRIERXQGDU\FRQGLWLRQVFODPSHGDQGVLPSO\
VXSSRUWHGERXQGDULHVDQGIUHHHGJHVZKLOVWWUXVVHVPD\RQO\KDYHWZRVXSSRUWRU
QRVXSSRUW
LY $ ODUJH QXPEHU RI H[WHQVLRQV RI WKH JULOODJH WKHRU\ H[LVW VHH 5R]YDQ\
ESS
Y:KLOVWPRVW0LFKHOOWUXVVHVDUHPHFKDQLVPVPRVWRSWLPDOJULOODJHVDUHVWDEOH
VWUXFWXUHV
(OHPHQWDU\ FDOFXODWLRQV VKRZ WKDW IRU WKH DGMRLQW GLVSODFHPHQWV X DQG Y LQ WKH
XSSHUUHJLRQZHKDYHWKHSULQFLSDODGMRLQWVWUDLQV
H H
DWWRWKHKRUL]RQWDO7KLVFRUUHVSRQGVWRD7UHJLRQGHQRWHGE\DQDUURZFURVV
LQ)LJF,QWKHORZHUUHJLRQZHKDYH
H H

)LJXUH&RYHURIWKHDXWKRU¶VILUVWERRNVKRZLQJDFRPSOH[RSWLPDOJULOODJHWRSRORJ\
,QVSLWHRIWKHVHDGYDQWDJHVRQO\6LJPXQG6LJPXQGHWDOKDVFRPSDUHG

QXPHULFDODQGDQDO\WLFDOVROXWLRQVIRUJULOODJHVDQGJRWDYHU\JRRGDJUHHPHQW
:H DUH QRW JRLQJ WR UHYLHZ KHUH WKH JULOODJH WKHRU\ LQ GHWDLO EHFDXVH LW ZDV
GLVFXVVHGDWFRQVLGHUDEOHOHQJWKDWDQRWKHU&,60$GYDQFHG&RXUVHVHH5R]YDQ\
ESS
7KHRSWLPDOOD\RXWWKHRU\ZDVDOVRXVHGIRUOHDVWZHLJKWVKHOOJULGVVWDUWLQJZLWK
5R]YDQ\DQG3UDJHULQFOXGLQJVHOIZHLJKWIRUDUHYLHZVHHWKHDXWKRU¶VERRN
5R]YDQ\SS
&RJQLWLYH 3URFHVVHV LQ 'HULYLQJ ([DFW $QDO\WLFDO 6WUXFWXUDO

7RSRORJLHV
7KHRSWLPDOWRSRORJ\IRUDFRPSOHWHO\QHZSUREOHPFDQQRWEHGHULYHGE\DGHGXF
WLYHSURFHVVXVLQJORJLFDOUHDVRQLQJ7KHURXJKDUUDQJHPHQWRIRSWLPDOUHJLRQVKDV
WREHILUVWµGUHDPWXS¶0HOFKHUV LH DVVXPHG E\ µLQWHOOLJHQW¶ JXHVVLQJ RU
LQYHQWLRQ DQG WKHQ FKHFNHG LI VXFK D VROXWLRQ VDWLVILHV DOO WKH RSWLPDOLW\ FULWHULD
'XULQJWKLVVHFRQGVWDJHWKHH[DFWVKDSHRIWKHUHJLRQERXQGDULHVLVDOVRDVVHVVHG
7KH ILUVW VWDJH UHTXLUHV FRQVLGHUDEOH LQVLJKW WKH VHFRQG VWDJH D ORW RI KLJK OHYHO
PDWKHPDWLFDOZRUN
D

E

)LJXUH2SWLPDOWUXVVWRSRORJ\IRUDVTXDUHVXSSRUWDVROXWLRQJXHVVHGE\5R]YDQ\
EH[DFWVROXWLRQE\/HZLQVNL/HZLQVNLDQG5R]YDQ\
$V DQ H[DPSOH IRU D VTXDUH VXSSRUW WKH DXWKRU JXHVVHG WKH RSWLPDO WRSRORJ\
IDLUO\DFFXUDWHO\LQDWH[WE\WKHDXWKRU5R]YDQ\SUHVHQWHGKHUH)LJD
H[DFWUHSURGXFWLRQDQGWKLVZDVFRQILUPHGE\/HZLQVNLILIWHHQ\HDUVODWHU/HZLQVNL
DQG5R]YDQ\E\DQH[WUHPHO\ORQJGHULYDWLRQ)LJE
0RUH UHFHQWO\ KRZHYHU LW KDV EHHQ SRVVLEOH WR FRPSXWH KLJKO\ DFFXUDWH QX
PHULFDOVROXWLRQVHJIRU0LFKHOOWUXVVHVXVLQJRYHUDELOOLRQSRWHQWLDOPHPEHUVLQ
WKHJURXQGVWUXFWXUH*LOEHUWDQG7\DV6RNRODDQGE3LFKXJLQ7\DVDQG
*LOEHUWVHHDOVR/HFWXUHLQWKLVFRXUVH7KHVHQXPHULFDOVROXWLRQVJLYHDYHU\
JRRGJHQHUDOLGHDDERXWWKHDGMRLQWGLVSODFHPHQWILHOGIRUWKHH[DFWRSWLPDOWRSRORJ\
H[FHSW IRU 2UHJLRQV UHJLRQV ZLWKRXW PHPEHUV ZKLFK DUH GLVFXVVHG LQ WKH QH[W
VHFWLRQ
'LIILFXOWLHVLQ'HULYLQJWKH$GMRLQW'LVSODFHPHQW6WUDLQ)LHOG
IRU25HJLRQV±5HFHQW'HYHORSPHQWV
25HJLRQVZHUHXVHGE\WKHDXWKRULQVRPHFRQWULEXWLRQVWR SDSHUV E\ /HZLQVNL
=KRX DQG 5R]YDQ\ E 5R]YDQ\ *ROOXE DQG =KRX DQG 6RNRO DQG
5R]YDQ\
,W LV H[SODLQHG LQ WKH ODVW RI WKHVH SDSHUV WKDW 25HJLRQV PD\ FRQWDLQ WKH XVXDO
55HJLRQVDQG75HJLRQVEXWDOVRWKHIROORZLQJQHZW\SHVRIUHJLRQV=5HJLRQVLQ
ZKLFKERWKSULQFLSDOVWUDLQVDUH]HURULJLGUHJLRQZLWK H H DQG9UHJLRQV
ZLWK H d N H
([DPSOHVRI2UHJLRQVFRQVLVWLQJRI75DQG=5HJLRQVDUHJLYHQLQ)LJEF
DQGGDIWHU6RNRODQG5R]YDQ\$QH[DPSOHRIDQ25HJLRQFRQVLVWLQJRI7
DQG95HJLRQVLVJLYHQLQ)LJ7KHDERYHGLDJUDPVVKRZRQO\RQHTXDUWHURIWKH
GRPDLQD[HVRIV\PPHWU\DUHLQGLFDWHGLQGDVKGRWOLQHV7KH\DUHJHQHUDOL]DWLRQVRI
0LFKHOO¶VVROXWLRQVKRZQLQ)LJD
7KHVWDWHRIDGMRLQWVWUDLQVLQWKHVH7DQG=5HJLRQVLQ)LJDUHVSHFWLYHO\LV
UHSUHVHQWHGE\WKH0RKUFLUFOHVLQ)LJVEDQGF,WFDQEHVHHQIURP)LJEWKDW
WKHVWUDLQDORQJWKHERXQGDU\RIWKH7DQG9UHJLRQVLV
H% N FRVD
0RUHRYHURQHFDQLQIHUIURP)LJFWKDWIRUWKH9UHJLRQZHKDYH
N FRVD H FRVD H H\ N WDQ D
,WFDQEHVHHQWKDWIRU D DQG D q JLYHVWKHFRUUHFW H YDOXHVIRUWKH

OLPLWLQJFDVHVLQ)LJVDDQGE,WLVLPSRUWDQWWRQRWHWKDWLQ2UHJLRQVWKHDGMRLQW
VWUDLQGLVSODFHPHQWILHOGVPD\EHQRQXQLTXH
/ G
D
/±G
0LFKHOO

7
7
K G
3
E
/±G K

7 K G
=
3
/±G
F

K/±GK
5
7
D K G
=
3
G

/±G K

7 = 7
K G
3
/G

)LJXUH6RPH2UHJLRQVFRQVLVWLQJRI75DQG=UHJLRQV
D

7
9
Į
[
\
E F
J J
H N D H % H N H H[ D H % H H\

)LJXUH25HJLRQFRQVLVWLQJRI7DQG95HJLRQV
$ VLPSOH H[DPSOH RI QRQXQLTXHQHVV RI WKH DGMRLQW VWUDLQ ILHOG LV VKRZQ LQ
)LJ,QWKLVSUREOHPZHKDYHWZRYHUWLFDOOLQHVXSSRUWVDWDGLVWDQFHRI/IURP
HDFKRWKHUDQGDKRUL]RQWDOSRLQWORDGDWDGLVWDQFH/IURPWKHULJKWKDQGVXSSRUW7R
WKHSHUPLVVLEOHVWUHVVHVWKHVDPHYDOXHLVDVVLJQHGLQWHQVLRQDQGFRPSUHVVLRQ7KH
RSWLPDOOD\RXWREYLRXVO\FRQVLVWVRIDVLQJOHKRUL]RQWDOEDUEHWZHHQWKHORDGDQGWKH
QHDUHUVXSSRUW7KHSULQFLSDODGMRLQWVWUDLQLQWKHYHUWLFDOGLUHFWLRQLV HYHU\ZKHUH
]HUR H \ ,Q )LJD ZH KDYH DQ 5UHJLRQ RQ WKH ULJKW KDQG VLGH DQG DQ
2UHJLRQRQWKHOHIWKDQGVLGHZKHUHWKHLQHTXDOLW\LQDGPLWVDKRUL]RQWDOVWUDLQ
RI N $Q DOWHUQDWLYH GLVFRQWLQXRXV DGMRLQW VWUDLQ ILHOG LV JLYHQ LQ )LJE LQ
ZKLFKZHKDYHD=UHJLRQDQGDQ5UHJLRQRQWKHOHIWERWKDGPLVVLEOHE\
$QRWKHUQHZW\SHRIUHJLRQZLWKLQDQ2UHJLRQLVDVFDOHG7UHJLRQWHUPHG7¶
UHJLRQKDYLQJWKHSURSHUW\ İ İ ON ZLWK O IRUHTXDOSHUPLVVLEOHVWUHVVHV
DQG H ON7 H ON & IRUXQHTXDOSHUPLVVLEOHVWUHVVHV)LJVKRZVDQRSWLPDO
WRSRORJ\ZLWKD7¶UHJLRQIRUZKLFKZHKDYH
O S S
7KHVROXWLRQLQ)LJKDVEHHQYHULILHGQXPHULFDOO\WRDKLJKGHJUHHRIDFFXUDF\E\
6RNROVHHWKHSDSHUE\5R]YDQ\DQG6RNRO
6LPLODUO\ZHKDYHVFDOHG5UHJLRQVWHUPHG5¶UHJLRQV

9 5 D
H N H N
[ [
/ /
[
\
= 55 E
H H N H N
[ [ [

)LJXUH $ WULYLDOO\ VLPSOH H[DPSOH RI QRQXQLTXHQHVV RI DGMRLQW VWUDLQGLVSODFHPHQW
ILHOGVLQ25HJLRQV

SULQFLSDODGMRLQWVWUDLQV
LQWKH2UHJLRQ
UHJLRQERXQGDULHV
2
µFRQFHQWUDWHG¶PHPEHUV
Į
7 7 µGLVWULEXWHG¶PHPEHUV
7

)LJXUH2SWLPDOWRSRORJ\ZLWKWKUHH7¶UHJLRQVLQWKH2UHJLRQ
([WHQVLRQRIWKH/D\RXW7KHRU\WR(ODVWLF'HVLJQZLWK0XOWL
SOH/RDG&RQGLWLRQV
7KHRULJLQDORSWLPDOOD\RXWWKHRU\3UDJHUDQG5R]YDQ\DKDVEHHQH[WHQGHGWR
PDQ\DGGLWLRQDOFODVVHVRISUREOHPVLQFOXGLQJFRPELQHG VWUHVV DQG GLVSODFHPHQW
FRQVWUDLQWVHJ5R]YDQ\%LUNHUDQG*HUGHV
*HQHUDORSWLPDOLW\FRQGLWLRQVIRUWUXVVHVZLWKVHYHUDOORDGFRQGLWLRQVDQGVHYHUDO
GLVSODFHPHQWFRQVWUDLQWVZHUHGHULYHGE\WKHDXWKRU5R]YDQ\D$FRPSOLDQFH
FRQVWUDLQWSXWVDOLPLWRQWKHVXPRIWKHVFDODUSURGXFWRIH[WHUQDOIRUFHVDQGWKH
FRUUHVSRQGLQJGLVSODFHPHQWVIRUDQ\RQHRIWKHORDGFRQGLWLRQVVRPHWLPHVFDOOHG
WKHµZRUVWVFHQDULR¶FRQVWUDLQW7KLVLVDVSHFLDOFDVHRIGLVSODFHPHQWFRQVWUDLQWLQ
ZKLFK D ZHLJKWHG FRPELQDWLRQ RI WKH GLVSODFHPHQWV DW WKH H[WHUQDO ORDGV LV FRQ
VWUDLQHGDQGWKHZHLJKWLQJIDFWRUVDUHWKHH[WHUQDOIRUFHV
)RUDFRPSOLDQFHFRQVWUDLQWWKHDXWKRU¶V5R]YDQ\DRSWLPDOLW\FRQGLWLRQV
UHGXFHWRWKHIROORZLQJ5R]YDQ\HWDO
)LN Q N )LN
H LN
(L $L
H LN
(L $L
$L ¦ Q N )LN (L U L
N
(L U L ¦Q H
N LN IRU $L !
N
(L U L ¦Q N H LN d IRU $L
N
ZKHUH H LN DQG H LN DUHNLQHPDWLFDOO\DGPLVVLEOHUHDOHODVWLFDQGDGMRLQWVWUDLQVLQ

WKH PHPEHU L XQGHU WKH ORDG N Q N /DJUDQJH PXOWLSOLHUV )LN )LN WKH UHDO DQG
DGMRLQWPHPEHUIRUFHVZKLOVW $L (L U L GHQRWHWKHFURVVVHFWLRQDODUHD<RXQJ¶V
PRGXOXVDQGGHQVLW\UHVSHFWLYHO\IRUWKHPHPEHUL
)RUWKHSDUWLFXODUFDVHRIDYHUWLFDOOLQHVXSSRUWDQGRIWZRDOWHUQDWLYHORDGLQJ
FDVHV FRQVLVWLQJ RI FRQFHQWUDWHG IRUFHV DW r E WKH DERYH RSWLPDOLW\ FULWHULD OHDG
5R]YDQ\HWDOWRDQRSWLPDOWRSRORJ\RIWZREDUVVKRZQLQ)LJZLWKD
YHU\QHDWFORVHGIRUPUHVXOWIRUWKHEDUDQJOHVWRSRI)LJ
7KHRSWLPDOYROXPHRIWKHWUXVVLVJLYHQE\5R]YDQ\HWDO
/ 3 § FRV E VLQ E ·
9RSW ¨¨ ¸¸
&( FRV D © FRV D VLQ D ¹

ZKHUH / LV WKH KRUL]RQWDO GLVWDQFH RI WKH ORDG IURP WKH YHUWLFDO VXSSRUW 3 LV WKH
PDJQLWXGHRIWKHORDGV(LV<RXQJ¶VPRGXOXVDQG&LVWKHJLYHQFRPSOLDQFHYDOXH
,Q DGGLWLRQ WR JOREDO SURRI YLD OD\RXW WKHRU\ LQ WKLV UDWKHU HODERUDWH SDSHU
5R]YDQ\=KRXDQG%LUNHURSWLPDOLW\RIWKHVROXWLRQVZDVDOVRFKHFNHGE\
WKHFRDXWKRUVQXPHULFDOO\E\
DXVLQJDGHQVHJULGRISRWHQWLDOWUXVVPHPEHUV
ERSWLPL]LQJDSHUIRUDWHGSODWHE\PHDQVRI6,03%HQGVRH=KRXDQG
5R]YDQ\5R]YDQ\=KRXDQG%LUNHU
FGHULYLQJIRUWKHJLYHQWRSRORJ\WZREDUWUXVVWKHRSWLPDODQJOHIURPWKH
.XKQ7XFNHUFRQGLWLRQ
DRSW
R

DRSW
R
DUFWDQ WDQ E WDQ E WDQ E

3
D $ E
D E
3
ER

)LJXUH2SWLPDOWRSRORJ\IRUWZRDOWHUQDWLYHORDGV
$OOVROXWLRQVGHULYHGE\WKUHHGLIIHUHQWDXWKRUVKDYHVKRZQDFRPSOHWHDJUHHPHQW
)RUWKLVUHDVRQ³UHOLDELOLW\´RIWKHDERYHDQDO\WLFDOVROXWLRQVLVIDLUO\KLJK
([WHQVLRQRIWKH/D\RXW7KHRU\WR3UREDELOLVWLF'HVLJQ
7KHDLPRIUHYLHZLQJWKHSUREOHPLQWKHODVWVHFWLRQZDVWRH[WHQGWKHVDPHRSWLPDO
WRSRORJ\WRSUREDELOLVWLFORDGV7KLVWRSLFZDVFRQVLGHUHGE\5R]YDQ\DQG0DXWH

7KHJHQHUDOIRUPRIWKHFRQVLGHUHGSUREOHPFODVVLVDVIROORZV
PLQ 9 ¦$/ L L
L
VXEMHFWWR
3U >& d . @ t 5
)L /L
& ¦ $(
L

L L
ZKHUH 9 WUXVV YROXPH $L FURVVVHFWLRQDO DUHD RI PHPEHU L /L OHQJWK RI
PHPEHUL3U SUREDELOLW\& WRWDOFRPSOLDQFH. OLPLWLQJYDOXHRIWRWDOFRP
SOLDQFH5 OLPLWLQJYDOXHRISUREDELOLW\ )L IRUFHLQPHPEHULDQG (L <RXQJ¶V
PRGXOXVRIPHPEHUL
7KHSDUWLFXODUH[DPSOHFRQVLGHUHGLVVKRZQLQ)LJD7KHWRSRORJ\RIDWUXVVLV
WREHRSWLPL]HGZLWKLQWKHGHVLJQGRPDLQ$%'*ZLWKVXSSRUWVDORQJ$%VXEMHFWWR
DQRQUDQGRPYHUWLFDOORDGRI9 DQGDUDQGRPKRUL]RQWDOORDG+ZLWKDPHDQRI
]HUR DQG D QRUPDO GLVWULEXWLRQ KDYLQJ D VWDQGDUG GHYLDWLRQ RI V 7KH WUXVV
YROXPHLVWREHPLQLPL]HGIRUWKHFRQGLWLRQVLQ,WZLOOEHVKRZQWKDWWKH
VROXWLRQIRUWKHFRQVLGHUHGSUREOHPLVDV\PPHWULFWZREDUWUXVVDVVKRZQLQ)LJ
E)RUVLPSOLFLW\ZHDVVLJQDXQLWYDOXHWR<RXQJ¶VPRGXOXV
$ %
7\SLFDO
QXPHULFDO
/ VROXWLRQ
+
D ([DFW
' *
DQDO\WLFDO
9
E VROXWLRQ
9
D E
+

)LJXUH(OHPHQWDU\EHQFKPDUNH[DPSOH
6LQFHWKLVSUREOHPLVV\PPHWULFLQWKHRQO\UDQGRPYDULDEOH+FRQGLWLRQLV

IXOILOOHGLIZHFRQVLGHUWKHUDQJHRIYDOXHVIRUWKHUDQGRPYDULDEOH
+ d + d +
ZKHUHWKHYDOXH + FDQEHFDOFXODWHGIURPWKHLQYHUVHQRUPDOGLVWULEXWLRQFXPXOD
WLYHSUREDELOLW\IXQFWLRQDOVRFDOOHG³TXDQWLOH´RU³SURELWÍXQFWLRQ ) DQG5LQ

5
+ V )

,Q5LVXVHGVLQFHWKHIDLOXUHPD\RFFXUDWERWKHQGVRIWKHLQWHUYDORI

7KLVLPSOLHV
5 ) P + V ) P + V

) P + V ) P + V ) P + V
IURPZKLFKZLWK P IROORZV
)RUH[DPSOHLIZHUHTXLUHDSUREDELOLW\RIIDLOXUHSUREDELOLW\RI
WKHQZHKDYH ) DQGKHQFH
+ V
LQ
,WFDQEHVKRZQ5R]YDQ\DQG0DXWHWKDWIRUWKHFRQVLGHUHGSUREOHPRQO\
+ r + FDQEHFULWLFDOIRU7KHUHIRUHRQFHZHNQRZWKHYDOXHRI + ZHFDQ
FDOFXODWHWKHRSWLPDOWRSRORJ\DQGWKHRSWLPDOYROXPHIURPWKHUHODWLRQLQ)LJ
DQG
7KHDERYHUHVXOWVZHUHFRQILUPHGE\0DXWHERWKDQDO\WLFDOO\DQGE\DILUVWRUGHU
UHOLDELOLW\DSSURDFK)250FRPELQHGZLWKDPDWHULDOGLVWULEXWLRQPHWKRG6,03
%HQGVRH=KRXDQG5R]YDQ\5R]YDQ\=KRXDQG%LUNHU
([WHQVLRQRIWKH/D\RXW7KHRU\WR3UH([LVWLQJ0HPEHUV

)LJXUH2SWLPDOWRSRORJ\DZLWKRXWDQGEDQGFZLWKSUHH[LVWLQJPHPEHUV
7KLVSUREOHPZDVGLVFXVVHGLQDSDSHUE\5R]YDQ\4XHULQ/RJRDQG3RPH]DQVNL
,IZHKDYHVRPHDOUHDG\ H[LVWLQJ PHPEHUV RI FURVVVHFWLRQDO DUHD % DQG
ZDQWWRDGGQHZPHPEHUVWRVDWLVI\ VRPH VWUHVV FRQGLWLRQ WKHQ ZH KDYH WKH IRO
ORZLQJRSWLPDOLW\FULWHULDIRUHTXDOSHUPLVVLEOHVWUHVVHVVHH)LJE
IRU ) % N H
IRU ) ! % N H N VJQ)
IRU ) % N d H N VJQH VJQ)
)RUXQHTXDOSHUPLVVLEOHVWUHVVHVZHKDYHVLPLODUFULWHULDEDVHGRQ)LJF)LJD
VKRZVWKHRSWLPDOWRSRORJ\ZLWKRXWSUHH[LVWLQJPHPEHUV)LJVEDQGFDUHIRU
SUHH[LVWLQJ PHPEHUV DORQJ 45 ZLWK HTXDO DQG XQHTXDO SHUPLVVLEOH VWUHVVHV UH
VSHFWLYHO\
7KHVH DQDO\WLFDO VROXWLRQV ZHUH REWDLQHG E\ WKH DXWKRU ZLWK LQGHSHQGHQW QX
PHULFDOFRQILUPDWLRQVE\4XHULQ/RJRDQG3RPH]DQVNL
D

7 2
=
E

)LJXUHD0LFKHOO¶VILUVWH[DPSOHELWVH[WHQVLRQWRWZRSRLQWORDGV5R]YDQ\
D6RNRODQG5R]YDQ\
+LVWRULF$VSHFWVRI6RPH5HFHQW'HYHORSPHQWV
,WLVUDWKHUUHPDUNDEOHWKDWWKHILUVWH[DPSOHRI0LFKHOOVKRZQLQ)LJWRS
KDVQRWEHH[WHQGHGIURPRQHSRLQWORDGWRWZRSRLQWORDGVIRURYHUDFHQWXU\7KLV
H[WHQVLRQZDVILUVWVWDWHGUHFHQWO\5R]YDQ\DDQGGLVFXVVHGLQJUHDWHUGHWDLO
DOVRIRURWKHUDVSHFWUDWLRVD\HDUODWHU6RNRODQG5R]YDQ\
,W LV HYHQ PRUH VXUSULVLQJ WKDW 0LFKHOO¶V WUXVV WKHRU\ KDV QRW EHHQ H[
WHQGHGXQWLOWKLV\HDUWRVHYHUDOORDGFRQGLWLRQVVWUHVVFRQVWUDLQWVHODVWLFGHVLJQ
7KLV ZLOO EH GLVFXVVHG LQ D IRUWKFRPLQJ SDSHU 5R]YDQ\ 6RNRO DQG 3RPH]DQVNL

&RQFOXGLQJ5HPDUNV
,QWKLVOHFWXUHZHKDYHUHYLHZHGWKHRSWLPDOOD\RXWWKHRU\E\3UDJHUDQG5R]YDQ\
D DQG VRPH RI LWV H[WHQVLRQV WR YDULRXV GHVLJQ FRQVWUDLQWV 'LIILFXOWLHV LQ
REWDLQLQJ VROXWLRQV ZHUH DOVR H[SODLQHG DQG QHZ W\SHV RI RSWLPDO UHJLRQV ZHUH
LQWURGXFHG IRU RYHUFRPLQJ WKHP 6RPH KLVWRULF DVSHFWV RI FXUUHQW WUXVV WRSRORJ\
UHVHDUFKZHUHDOVRQRWHG

6RPH)XQGDPHQWDO3URSHUWLHVRI([DFW2SWLPDO
6WUXFWXUDO7RSRORJLHV
*HRUJH,15R]YDQ\

%XGDSHVW+XQJDU\
,QWURGXFWLRQ
,Q/HFWXUHRIWKLVDXWKRUZHGLVFXVVHG0LFKHOO¶VWKHRU\RIRSWLPDO WUXVV
GHVLJQH[DPLQHGLWVUDQJHRIYDOLGLW\DQGORRNHGDWH[WHQGHGRSWLPDOLW\FULWHULDIRUD
EURDGHUFODVVRIERXQGDU\FRQGLWLRQV
,Q /HFWXUH WKH RSWLPDO OD\RXW WKHRU\ 3UDJHU DQG 5R]YDQ\ DV ZHOO DV
RSWLPDOUHJLRQVLQH[DFWWUXVVWRSRORJLHVZHUHUHYLHZHGDQGVHYHUDOH[WHQVLRQVRIWKH
OD\RXWWKHRU\SUHVHQWHG
:KLOVWWKHVHWZROHFWXUHVWRXFKHGRQVRPHEDVLFIHDWXUHVRIRSWLPDOVWUXFWXUDO
WRSRORJLHVIXQGDPHQWDOSURSHUWLHVRIWKHVHZLOOEHH[DPLQHGLQGHWDLOLQWKLVOHFWXUH
3DUWLDO 5HOD[DWLRQ RI WKH 2UWKRJRQDOLW\ 5HTXLUHPHQW IRU

0LFKHOO7UXVVHV
+HPS VWDWHV DERXW 0LFKHOO WUXVVHV µ,I D SDLU RI WHQVLRQ DQG FRPSUHVVLRQ
PHPEHUVPHHWDWDSRLQWWKH\PXVWEHRUWKRJRQDO«QRRWKHUPHPEHUFDQEHFR
SODQDU ZLWK WKHP¶ ,Q WKH IROORZLQJ ZH PHQWLRQ FDVHV LQ ZKLFK WKH DERYH
RUWKRJRQDOLW\FDQEHUHOD[HG
1RQ2UWKRJRQDO7HQVLRQDQG&RPSUHVVLRQ0HPEHUVDW%RXQGDU\3RLQWV
RI7UHJLRQV
$VDVLPSOHH[DPSOHZHFRQVLGHU0LFKHOO¶VVROXWLRQVHH)LJDLQ/HFWXUH
RU)LJLQWKLVOHFWXUH
7KLVVROXWLRQFRQVLVWVRIIRXU7UHJLRQVWZRRIWKHPKDYHFRQVWDQWGLUHFWLRQVRI
SULQFLSDO DGMRLQW VWUDLQV DQG WKH RWKHU WZR DUH µFLUFXODU IDQV¶ $W WKH SRLQW $ DQ
LQILQLWHQXPEHURIPHPEHUVPHHW EXW RQO\ WKH RXWVLGH RQHV VDWLVI\ WKH DERYH RU
WKRJRQDOLW\UXOH

DOI 10.1007/978-3-7091-1643-2_3, © CISM, Udine 2014
36 G.I.N. Rozvany
1RQ2UWKRJRQDO7HQVLRQDQG&RPSUHVVLRQ0HPEHUV$ORQJWKH%RXQGDU\
RIDQ5DQGDQ55HJLRQ
$VDQH[DPSOHZHTXRWHWKHUHVXOWVRI5R]YDQ\DQG*ROOXEZKRVROYHGWKH
0LFKHOOWUXVVSUREOHPIRUDQ\FRQYH[SRO\JRQDOGRPDLQZLWKOLQHVXSSRUWVDORQJWKH
ERXQGDULHV7KHRSWLPDOWUXVVWRSRORJ\IRUWKHVHVXSSRUWFRQGLWLRQVZDVHYHQFDO
FXODWHGE\DQRQQXPHULFFRPSXWHUSURJUDP2QHRIWKHVHWRSRORJLHVLVVKRZQLQ
)LJLQZKLFKGRXEOHDUURZVLQGLFDWHSULQFLSOHGLUHFWLRQVRI5UHJLRQV$W WKH
SRLQWORDGWKHFRPSUHVVLRQDQGWHQVLRQPHPEHUVDUHFOHDUO\QRWRUWKRJRQDO

)LJXUH1RQRUWKRJRQDOFRPSUHVVLRQDQGWHQVLRQPHPEHUVIRUORDGVDWWKHERXQGDU\
EHWZHHQDQ 5 DQGDQ 5 UHJLRQDIWHU5R]YDQ\DQG*ROOXE
$QRWKHUH[DPSOHRIQRQRUWKRJRQDOLW\ZDVGHULYHGE\WKHDXWKRU5R]YDQ\
&RQVLGHULQJDYHUWLFDOSRLQWORDGDQGWZRSLQVXSSRUWVWKHRSWLPDOWUXVVWRSRORJLHV
IRUYDULRXVKRUL]RQWDOGLVWDQFHVRIWKHSRLQWORDGDUHVKRZQLQ)LJVDGHDQGI
7KHNLQHPDWLFERXQGDU\FRQGLWLRQVIRUWKHDGMRLQWVWUDLQILHOGVDUHVKRZQLQ)LJE
DQG WKH FDOFXODWLRQ RI WKH SULQFLSDO DGMRLQW VWUDLQV H , DQG H ,, E\ PHDQV RI D
0RKUFLUFOHLQ)LJF,Q)LJZHXVHDQRUPDOL]HGIRUPXODWLRQZLWKN
,WFDQEHVHHQWKDWWKHRSWLPDODGMRLQWVWUDLQILHOGLQ)LJVDEDQGFFRQVLVWRIDQ
5 DQGDQ 5 UHJLRQ:HOONQRZQVROXWLRQVLQ)LJVGDQGHDUHVSHFLDOFDVHVRI
WKH WRSRORJ\ LQ )LJ D 7KH VROXWLRQ LQ )LJ I LV D VKRUW µ0LFKHOO FDQWLOHYHU¶
+HPS/HZLQVNL=KRXDQG5R]YDQ\D
Some Fundamental Properties of Exact Optimal Structural Topologies 37
7KHWRSRORJ\LQ)LJDKDVEHHQH[WHQGHGWRQRQV\PPHWULFORFDWLRQVRIWKH
SRLQW ORDG LQ D SDSHU WR EH VXEPLWWHG 6RNRO DQG 5R]YDQ\ $OO QHZ
QRQRUWKRJRQDORSWLPDOWRSRORJLHVKDYHEHHQFRQILUPHGQXPHULFDOO\WRDKLJKGHJUHH
RIDFFXUDF\E\6RNRO¶VDDQGEWUXVVRSWLPL]DWLRQSURJUDP

)LJXUH DGHDQGI2SWLPDOWUXVVWRSRORJLHVIRUDYHUWLFDOSRLQWORDGDQGWZRSLQ
VXSSRUWV 7KH DGMRLQW VWUDLQ ILHOG IRU WKH VROXWLRQ LQ VXEILJXUH D LV H[SODLQHG LQ
VXEILJXUHVEDQGF
'RPDLQ$XJPHQWDWLRQDQG5HGXFWLRQ
7KHGRPDLQDXJPHQWDWLRQWKHRUHPVWDWHVWKDWIRUDGRPDLQZLWKOLQHVXSSRUWVDORQJ
WKHERXQGDULHVWKHRSWLPDOWRSRORJ\GRHVQRWFKDQJHLIZHPRGLI\WKHERXQGDU\VXFK
WKDW
DDFWLYHVXSSRUWLQJSRLQWVDORQJWKHROGERXQGDU\DUHFRQWDLQHGLQWKHQHZ
ERXQGDU\DQG
EQRSRLQWRIWKHQHZERXQGDU\LVFRQWDLQHGLQWKHLQWHULRURIWKHROGGRPDLQ
$SRLQWRIDOLQHVXSSRUWLVµDFWLYH¶LIPHPEHUVZLWKQRQ]HURFURVVVHFWLRQDODUHD
FRQQHFWWRLW
38 G.I.N. Rozvany
3URRIRIWKHDERYHWKHRUHPLVEDVHGRQDQDGMRLQWVWUDLQGLVSODFHPHQWILHOGLQ
ZKLFKWKHUHDUH]HURGLVSODFHPHQWVVWUDLQVLQEHWZHHQWKHROGDQGWKHQHZGRPDLQ
ERXQGDULHV
7KHH[DPSOHLQ)LJLOOXVWUDWHVWKHGRPDLQDXJPHQWDWLRQSULQFLSOH)RUWKHROG
ERXQGDU\WULSOHOLQHVZHGHULYHGWKHRSWLPDOWRSRORJ\LQ)LJRI/HFWXUH%DVHG
RQ WKH GRPDLQ DXJPHQWDWLRQ WKHRUHP WKH VDPH WRSRORJ\ LV YDOLG IRU WKH QHZ
ERXQGDU\GRWWHGOLQHV
$PRUHXVHIXOH[DPSOHLVVKRZQLQ)LJLQZKLFKDZHOONQRZQVROXWLRQIRU
WZRYHUWLFDOVXSSRUWVWULSOHOLQHVLVH[WHQGHGWRWZRFLUFXODUVXSSRUWVGRWWHGOLQHV

)LJXUH([DPSOHGHPRQVWUDWLQJWKHWKHRUHPRIGRPDLQDXJPHQWDWLRQ

)LJXUH$QRWKHUH[DPSOHRIGRPDLQDXJPHQWDWLRQ
7KH WKHRUHP RI GRPDLQ UHGXFWLRQ VWDWHV WKDW DQ RSWLPDO 0LFKHOO WRSRORJ\ UH
PDLQVWKHVDPHLIZHFKDQJHWKHGHVLJQGRPDLQVXFKWKDW
DDOOSRLQWVLQWKHQHZERXQGDU\DUHFRQWDLQHGLQWKHROGGRPDLQDQG
EWKHRULJLQDO0LFKHOOWUXVVLQFOXGLQJLWVDFWLYHVXSSRUWVLVFRQWDLQHGLQWKH

QHZGRPDLQ
7KLVWKHRUHPLVQRWUHVWULFWHGWRGRPDLQVZLWKOLQHVXSSRUWVDORQJWKHLUERXQGDU\)RU
DSURRIRIWKLVWKHRUHPVHHWKHDXWKRU¶VSDSHU5R]YDQ\E

)LJXUH([DPSOHGHPRQVWUDWLQJWKHGRPDLQUHGXFWLRQWKHRUHP
$QH[DPSOHRIGRPDLQUHGXFWLRQLVVKRZQLQ)LJ7KHRSWLPDOWRSRORJ\IRUWZR
OLQHVXSSRUWVDWDQDQJOHZDVGHWHUPLQHGSUHYLRXVO\5R]YDQ\DQG*ROOXE)LJ
DLQWKDWSDSHU7KHGLUHFWLRQRIWKHSRLQWORDGPD\YDU\EHWZHHQ7KHQHZ
ERXQGDU\ IRU WKH GRPDLQ UHGXFWLRQ WKHRUHP LV VKRZQ LQ EURNHQ OLQH JLYLQJ WKH
RSWLPDOWRSRORJ\IRUWZRKLQJHVXSSRUWV
,QIDFWWKHGRPDLQUHGXFWLRQWKHRUHPZDVXVHGIRUREWDLQLQJRSWLPDO0LFKHOO
WRSRORJLHVIRUWZRKLQJHVDQGDSRLQWORDGLQYDULRXVORFDWLRQV6RNRODQG5R]YDQ\

6\PPHWU\DQG1RQXQLTXHQHVV3ULQFLSOHV
7\SHRI3UREOHPV&RQVLGHUHG
,Q6HFWLRQZHFRQVLGHUSUREOHPVZLWKWKHIROORZLQJSURSHUWLHV
L7KHVSHFLILFFRVWIXQFWLRQLVDPRGXOXVRUDEVROXWHYDOXHIXQFWLRQDVLQ
7KLVLVWKHFDVHIRUH[DPSOHIRUJULOODJHVRIJLYHQGHSWKRUWUXVVHVZLWKVLJQ
DQGVL]HLQGHSHQGHQWSHUPLVVLEOHVWUHVVHV
40 G.I.N. Rozvany
LL7KHFURVVVHFWLRQDODUHDVDUHFRQWLQXRXVYDULDEOHVDQGXQFRQVWUDLQHG
LLL7KHWRWDOYROXPHZHLJKWRUFRVWLVXQFRQVWUDLQHG
LY7KHJURXQGVWUXFWXUHLHVWUXFWXUDOXQLYHUVHFRQVLVWVRIDQLQILQLWHQXPEHURI
SRWHQWLDOPHPEHUVLQDOOGLUHFWLRQVDWDOOSRLQWVRIWKHGHVLJQGRPDLQ+RZHYHU
DQH[WHQVLRQWRGLVFUHWHVWUXFWXUHVLVJLYHQLQ6HFWLRQ
Y7KHUHH[LVWVRPHIHDVLEOHVROXWLRQVIRUWKHFRQVLGHUHGSUREOHPV
'HILQLWLRQV
D:KHQHYHUWKHWHUPµV\PPHWU\¶LVXVHGLQWKLVWH[WZHPHDQOLQHV\PPHWU\
ZKLFK LPSOLHV WKDW DQ\WKLQJ FRQWDLQHG LQ WKH WZR KDOYHV RI RXU GRPDLQ DUH HDFK
RWKHUV¶PLUURULPDJHVDERXWDOLQHWHUPHGµD[LVRIV\PPHWU\¶
E6XSSRUWFRQGLWLRQVDUHV\PPHWULFLIDOOVXSSRUWVDUHV\PPHWULFDOO\ORFDWHG
ZLWKUHVSHFWWRWKHD[LVRIV\PPHWU\DVGHILQHGDERYHDQGDOOUROOHUVXSSRUWVPRYH
DORQJOLQHVWKDWDUHDOVRV\PPHWULF
F$SDLURIORDGVRUGLVSODFHPHQWVLVV\PPHWULFLILWKH\DUHORFDWHGV\PPHW
ULFDOO\ZLWKUHVSHFWWRWKHD[LVLLWKHLUFRPSRQHQWVSDUDOOHOWRWKHD[LVDUHLQWKH
VDPHGLUHFWLRQLLLWKHLUFRPSRQHQWVQRUPDOWRWKHD[LVDUHLQRSSRVLWHGLUHFWLRQDQG
LY WKHLU FRPSRQHQWV KDYH WKH VDPH PDJQLWXGH RQ ERWK VLGHV RI WKH D[LV /RDGV
DFWLQJRQDQGSDUDOOHOWRWKHV\PPHWU\D[LVDUHV\PPHWULF
G$SDLURIORDGVRUGLVSODFHPHQWV LV VNHZV\PPHWULF LI L WKH\ DUH ORFDWHG
V\PPHWULFDOO\ZLWKUHVSHFWWRWKHD[LVLLWKHLUFRPSRQHQWVSDUDOOHOWRWKHD[LVDUHLQ
WKH RSSRVLWH GLUHFWLRQ LLL WKHLU FRPSRQHQWV QRUPDO WR WKH D[LV DUH LQ WKH VDPH
GLUHFWLRQDQGLYWKHLUFRPSRQHQWVKDYHWKHVDPHPDJQLWXGHRQERWKVLGHVRIWKH
D[LV/RDGVDFWLQJRQDQGQRUPDOWRWKHV\PPHWU\D[LVDUHVNHZV\PPHWULF
H$VWDWHRIVWUHVVRUVWUDLQLVV\PPHWULFLIDWDQ\SDLURISRLQWVORFDWHGV\P
PHWULFDOO\WRWKHD[LVLWKHGLUHFWLRQRIWKHSULQFLSDOVWUHVVHVRUVWUDLQVIRUPDPLUURU
LPDJHRIHDFKRWKHUDQGLLWKHLUVLJQVDQGPDJQLWXGHVDUHWKHVDPH
I$VWDWHRIVWUHVVRUVWUDLQLVVNHZV\PPHWULFLIDWDQ\SDLURISRLQWVORFDWHG
V\PPHWULFDOO\WRWKHD[LVLWKHGLUHFWLRQRIWKHSULQFLSDOVWUHVVHVRUVWUDLQVIRUPD
PLUURULPDJHRIHDFKRWKHUDQGLLWKHLUVLJQVDUHRSSRVLWHEXWWKHLUPDJQLWXGHVWKH
VDPH
J$VSHFLILFFRVWIXQFWLRQLVV\PPHWULFRUVLJQLQGHSHQGHQWLIFKDQJLQJWKH
VLJQRIWKHVWUHVVHVRUVWUHVVUHVXOWDQWVGRHVQRWDOWHUWKHYDOXHRIWKHFRVWIXQFWLRQ
K $ WRSRORJ\ RSWLPL]DWLRQ SUREOHP LV V\PPHWULF LI WKH GRPDLQ ERXQGDULHV
ORDGLQJVXSSRUWVDQGVSHFLILFFRVWIXQFWLRQDUHV\PPHWULF
L$WRSRORJ\RSWLPL]DWLRQSUREOHPLVVNHZV\PPHWULFLIWKHGRPDLQERXQGD
ULHV VXSSRUWV DQG VSHFLILF FRVW IXQFWLRQ DUH V\PPHWULF DQG WKH ORDGLQJ
VNHZV\PPHWULF
)RUIXUWKHUGHILQLWLRQVWKHUHDGHUPD\UHIHUWRWKHUHYLHZDUWLFOHRIWKHDXWKRU5R]
YDQ\D
1RQXQLTXHQHVV7KHRUHP
7KHRUHP$WRSRORJ\RSWLPL]DWLRQSUREOHPRIWKHFRQVLGHUHGFODVVWKHUHH[LVWV
HLWKHURQHRUDQLQILQLWHQXPEHURIRSWLPDOGHVLJQVZLWKWKHVDPHOHDVWYROXPH
3URRIRXWOLQH,WFDQEHVKRZQWKDWWKHOHDVWYROXPHSUREOHPXQGHUFRQVLGHUDWLRQ
FDQEHKDQGOHGE\WKHOLQHDUSURJUDPPLQJ/3PHWKRG,WLVZHOONQRZQHJ6WUDQJ
WKDWLQ/3WKHRSWLPDOVROXWLRQLVJLYHQE\WKHEDVLFIHDVLEOHVROXWLRQWKDWKDV
WKHORZHVWFRVWLQRXUFDVHORZHVWYROXPH:HVKDOOWHUPWKLVRSWLPDOEDVLFIHDVLEOH
VROXWLRQRUEULHIO\2%)6,IWKHUHDUHPRUHWKDQRQHEDVLFIHDVLEOHVROXWLRQVZLWKWKH
VDPHOHDVWYROXPHWKHQWKH\DUHDOOWHUPHG2%)6¶VDQGWKHLUFRQYH[FRPELQDWLRQV
DUHDOORSWLPDOVROXWLRQV7KLVPHDQVWKDWWKHRSWLPDOWRSRORJ\IRURXUSUREOHPVPD\
EHQRQXQLTXHDQGWKHQDQLQILQLWHQXPEHURIVROXWLRQVLVRSWLPDO

5
7 7
D E
F

)LJXUHD7RSRORJ\RSWLPL]DWLRQSUREOHPEDQGF
([DPSOHQRQXQLTXHRSWLPDOVROXWLRQ)RUWKHVXSSRUWDQGORDGFRQGLWLRQVLQ)LJ
DWKHWZR2%)6¶VRIWKHVDPHRSWLPDOZHLJKWDUHVKRZQLQ)LJVEDQGFIRUD
SURRIVHH5R]YDQ\DQG*ROOXE$Q\FRQYH[FRPELQDWLRQRIWKHVHWZR2%)6¶V
LVDOVRDQRSWLPDOVROXWLRQ7KLVPHDQVWKDWZHPD\PXOWLSO\DOOFURVVVHFWLRQDODUHDV
RIRQH2%)6E\ Ȟ d Ȟ d DQGWKHFURVVVHFWLRQDODUHDVRIWKHRWKHU2%)6E\
Ȟ DQGWKHUHVXOWLQJEDUWUXVVHVZLOOEHIXOO\VWUHVVHGDQGRIWKHVDPHYROXPH
DVWKHWZR2%)6¶V
1RWHWKDWLQ)LJWKHSRLQWORDGVDFWRQRSWLPDOUHJLRQERXQGDULHVDQGWKLV
FDXVHVPXOWLSOH2%)6¶V7KHVDPHZLOOEHREVHUYDEOHRQDOORWKHUH[DPSOHVZLWK
42 G.I.N. Rozvany
PXOWLSOH2%)6¶VZLWKWKHH[FHSWLRQRI6UHJLRQV7KLVLVEHFDXVHDORDGRQDUHJLRQ

ERXQGDU\FDQVRPHWLPHVEHWUDQVPLWWHGDORQJWZRGLIIHUHQWORDGSDWKVRYHUHLWKHU
UHJLRQDQGERWKOD\RXWVKDYHWKHVDPHYROXPH+RZHYHUORDGVRQUHJLRQERXQGDULHV
GRQRWDOZD\VFDXVHPXOWLSOH2%)6¶VVHHHJ)LJ
([DPSOHRQO\RQH2%)6 )RUWKHRSWLPDOVROXWLRQLQ)LJZHKDYHRQO\RQH
2%)6ZKLFKUHSUHVHQWVWKHXQLTXHRSWLPDOVROXWLRQ5R]YDQ\DQG*ROOXE
OLQHVXSSRUW
5
D
D
5

)LJXUH6LPSOHH[DPSOHLOOXVWUDWLQJ7KHRUHPXQLTXHRSWLPDOVROXWLRQDIWHU5R]YDQ\
DQG*ROOXE
6\PPHWU\7KHRUHP
7KHRUHP)RUDV\PPHWULFWRSRORJ\RSWLPL]DWLRQSUREOHPRIWKHFRQVLGHUHGFODVV
DWOHDVWRQHRSWLPDOGHVLJQDQGWKHLQWHUQDOIRUFHVLQLWDUHV\PPHWULF
3URRI RXWOLQH ,I WKHUH H[LVWV DW OHDVW RQH V\PPHWULF 2%)6 IRU D SUREOHP WKHQ
7KHRUHPLVREYLRXVO\VDWLVILHGWKLVLVWKHFDVHZLWKDOONQRZQH[DFWVROXWLRQVLQWKH
OLWHUDWXUH+RZHYHUIRUWKHVDNHRIDFRPSOHWHSURRIRI7KHRUHPDVVXPHWKDWIRUD
V\PPHWULFSUREOHPZLWKVRPHIHDVLEOHVROXWLRQVWKHUHH[LVWVQRV\PPHWULF2%)6
7KLVLPSOLHVWKDWWKHUHH[LVWVRPHQRQV\PPHWULF2%)6¶V,IZHWKHQWDNHDUHIOHFWLRQ
PLUURULPDJHRIDQ\RQHRIWKHVH2%)6¶VDERXWWKHD[LVRIV\PPHWU\RIRXUSURE
OHPWKHQWKHUHIOHFWLRQPXVWDOVREHDQ2%)6EHFDXVHDQ\UHIOHFWLRQLVDFRQJUXHQW
PDSSLQJDQGKHQFHWKHUHIOHFWLRQPXVWKDYHWKHVDPHRSWLPDOYROXPHDVWKHRULJLQDO
2%)6 ,I ZH QRZ IRUP D FRQYH[ FRPELQDWLRQ RI WKH DERYH WZR QRQV\PPHWULF
2%)6¶VZLWKPXOWLSOLHUVDQGZHJHWDQRSWLPDOVROXWLRQWKDWLVV\PPHWULF
7KLVSURYHV7KHRUHP
([DPSOHV,Q)LJWKHWZRRSWLPDOWRSRORJLHVLQEDQGFDUHV\PPHWULF,Q)LJ

WKHXQLTXHRSWLPDOWRSRORJ\LVV\PPHWULF%RWKH[DPSOHVGHPRQVWUDWHWKHYD
OLGLW\RI7KHRUHP
6NHZ6\PPHWU\7KHRUHP
7KHRUHP&RQVLGHULQJDVNHZV\PPHWULFWRSRORJ\RSWLPL]DWLRQSUREOHPRIWKH
FRQVLGHUHGFODVVDWOHDVWRQHRSWLPDOGHVLJQLVV\PPHWULFDQGWKHLQWHUQDOIRUFHVLQLW
DUHVNHZV\PPHWULF
3URRIRXWOLQH7KHSURRILVVLPLODUWRWKDWRI7KHRUHPH[FHSWWKDWWKHORDGVDQG
LQWHUQDOIRUFHVDUHVNHZV\PPHWULFLQVWHDGRIEHLQJV\PPHWULF
([DPSOH VNHZV\PPHWULF ORDG RQO\ RQH VNHZV\PPHWULF 2%)6 0LFKHOO¶V

ZHOONQRZQH[DPSOHKDVDXQLTXH DQG VNHZV\PPHWULF VROXWLRQ WR IRU WKH
VXSSRUWDQGORDGFRQGLWLRQVVKRZQLQ)LJ

)LJXUH0LFKHOO¶VH[DPSOHZLWKDXQLTXHVROXWLRQKDYLQJDV\PPHWULFRSWLPDOGHVLJQ
ZLWKVNHZV\PPHWULFLQWHUQDOIRUFHV
44 G.I.N. Rozvany
([DPSOH VNHZV\PPHWULF ORDG DQ LQILQLWH QXPEHU RI VNHZV\PPHWULF DQG

QRQVNHZV\PPHWULFVROXWLRQV)RUWKHVXSSRUWDQGORDGFRQGLWLRQLQ)LJZH
KDYHWKUHH2%)6¶VRQHLVDV\PPHWULFGHVLJQZLWKVNHZV\PPHWULFIRUFHVLQWKH
RWKHUWZRZHKDYHWZRGHVLJQVZKLFKDUHPLUURULPDJHVRIHDFKRWKHUDERXWWKHD[LV
RIV\PPHWU\$Q\FRQYH[FRPELQDWLRQRIWKHVHWKUHH2%)6¶VZLWKHTXDOPXOWLSOLHUV
IRUWKRVHLQ)LJVEDQGFEHFRPHV\PPHWULFGHVLJQVZLWKVNHZV\PPHWULFLQWHUQDO
IRUFHV7KLVLVWKHFDVHIRUH[DPSOHLIZHXVHWKHPXOWLSOLHUVIRUWKH
2%)6¶V LQ )LJV D E DQG F UHVSHFWLYHO\ 2WKHU FRQYH[ FRPELQDWLRQV DUH
QRQVNHZV\PPHWULF7KLVH[DPSOHLVDOVRFRQVLVWHQWZLWK7KHRUHP
7 5

5
DEF

)LJXUH([DPSOHZLWKDQLQILQLWHQXPEHURIVNHZV\PPHWULFDQGQRQVNHZV\PPHWULF
VROXWLRQV
&RUROODULHVRI7KHRUHPVDQG
&RUROODU\ )RUDV\PPHWULFWRSRORJ\RSWLPL]DWLRQSUREOHPRIWKHFRQVLGHUHG
FODVVHLWKHURQHRUDQLQILQLWHQXPEHURIRSWLPDOGHVLJQVWRJHWKHUZLWKWKHLULQWHUQDO
IRUFHVDUHV\PPHWULF7KHQXPEHURIDGGLWLRQDOQRQV\PPHWULFRSWLPDOGHVLJQVLV
HLWKHU]HURRULQILQLWH
3URRIRXWOLQH%\7KHRUHPWKHUHH[LVWDWOHDVWRQHV\PPHWULF2%)6,IWKLVLVWKH
RQO\2%)6WKHQZHKDYHDXQLTXHRSWLPDOGHVLJQWKDWLVV\PPHWULFVHHHJ)LJ
,I WKHUH H[LVW DGGLWLRQDO V\PPHWULF 2%)6¶V WKHQ WKH FRQYH[ FRPELQDWLRQV RI DOO
V\PPHWULF2%)6¶VLVDOVRV\PPHWULFDQGZHKDYHDQLQILQLWHQXPEHURIV\PPHWULF
2%)6¶VVHH)LJ7KHQXPEHURIQRQV\PPHWULFRSWLPDOVROXWLRQVLV]HUR LI
WKHUHH[LVWQRQRQV\PPHWULF2%)6IRUDJLYHQSUREOHP%XWLIWKHUHLVDWOHDVWRQH
QRQV\PPHWULF2%)6WKHQWKHLQILQLWHQXPEHURIFRQYH[FRPELQDWLRQVRIWKLVZLWK
RWKHU2%)6¶VDUHDOOQRQV\PPHWULFRSWLPDOGHVLJQVZLWKWKHH[FHSWLRQRIWKHFDVH
VKRZQLQ)LJZLWKHTXDOPXOWLSOLHUVIRUWKH2%)6¶VLQ)LJVEDQGF
&RUROODU\ )RUD VNHZV\PPHWULF WRSRORJ\ RSWLPL]DWLRQ SUREOHP RI WKH FRQ
VLGHUHG FODVV WKHUH H[LVWV HLWKHU RQH RU DQ LQILQLWH QXPEHU RI V\PPHWULF RSWLPDO
GHVLJQV ZLWK VNHZV\PPHWULF LQWHUQDO IRUFHV LQ WKHP 7KH QXPEHU RI DGGLWLRQDO
QRQV\PPHWULFRSWLPDOGHVLJQVLVHLWKHU]HURRULQILQLWH
3URRIRXWOLQH7KLVLVVLPLODUWRWKDWIRU&RUROODU\
&RUROODU\,QWKHRSWLPDOVROXWLRQRIDVNHZV\PPHWULFWRSRORJ\RSWLPL]DWLRQ
SUREOHPLWKHDGMRLQWVWUDLQLV]HURDORQJWKHD[LVRIVNHZV\PPHWU\LLDWDQ\SLQ
VXSSRUWDORQJWKHD[LVRIV\PPHWU\WKHUHDFWLRQIRUFHLVQRUPDOWRWKHD[LVRIV\P
PHWU\ DQG LLL VXFK SLQ VXSSRUW FDQ EH UHSODFHG ZLWK D UROOHU VXSSRUW ZLWKRXW
FKDQJLQJWKHRSWLPDOGHVLJQ
2XWOLQHRISURRILDQGLLIROORZIURPWKHVNHZV\PPHWU\RIWKHRSWLPDOVWUHVV

DQGIRUFHILHOGVLLLIROORZVIURPLDQGLLZKHQWKHUHDFWLRQVDUHQRUPDOWRWKH
D[LVRIV\PPHWU\ WKHQ LW PDNHV QR GLIIHUHQFH WR WKH IRUFH DQG VWUHVV ILHOGV LI ZH
UHSODFHDSLQZLWKDUROOHURUYLYHYHUVD0RUHRYHUWKHDGMRLQWVWUDLQILHOGGRHVQRW
FKDQJHHLWKHUEHFDXVHE\LZHKDYH]HURGLVSODFHPHQWVDORQJWKHD[LVRIV\PPHWU\
([DPSOH LQ )LJ UHYLVLWHG DOVR LOOXVWUDWLQJ &RUROODU\ 7KH ORDGV DQG
DGMRLQW VWUDLQV LQ 0LFKHOO¶V ZHOONQRZQ VROXWLRQ LQ )LJ DUH
VNHZV\PPHWULFDERXWWKHKRUL]RQWDOD[LV(LWKHURUERWKVXSSRUWVFRXOGEHUHSODFHG
E\ D UROOHU VXSSRUW ZLWKRXW FKDQJLQJ VWUHVVHV DQG VWUDLQV LQ WKH VWUXFWXUH 7KLV LV
EHFDXVHWKHUHDFWLRQVLQERWKVXSSRUWVDUHYHUWLFDO0RUHRYHUWKHDGMRLQWVWUDLQVDUH
]HURDORQJWKHKRUL]RQWDOD[LVRIVNHZV\PPHWU\
6\PPHWU\7KHRUHPIRU0XOWLSOH/RDGV
7KHRUHP,IWKHVXSSRUWFRQGLWLRQVDQGGHVLJQGRPDLQERXQGDU\DUHV\PPHWULF
DQGWZRDOWHUQDWHORDGFRQGLWLRQVDUHPLUURULPDJHVUHIOHFWLRQVRIHDFKRWKHUDERXW
WKHV\PPHWU\D[LVWKHQDWOHDVWRQHRSWLPDOWRSRORJ\LVV\PPHWULF
([DPSOH LOOXVWUDWLQJ 7KHRUHP &RQVLGHU D YHUWLFDO VXSSRUW DQG WZR DOWHUQDWH
SRLQWORDGVDWGHJUHHVWRWKHV\PPHWU\D[LV)LJVHHDOVR)LJLQ/HFWXUH
46 G.I.N. Rozvany
7KHQWKHRSWLPDOWRSRORJ\IRUDZRUVWFDVHFRPSOLDQFHFRQVWUDLQWFRQVLVWVRIWZR
EDUV)LJZKLFKDUHDWDQDQJOHRI
D DUFWDQ
WR WKH V\PPHWU\ D[LV )RU D FRPSOHWH SURRI VHH 5R]YDQ\ HW DO ZLWK WKH
GHULYDWLRQRIRSWLPDOLW\FULWHULDLQ5R]YDQ\D
D
DE

)LJXUH([DPSOHLOOXVWUDWLQJ7KHRUHP
0RGLILFDWLRQRIWKH$ERYH7KHRUHPVDQG&RUROODULHVIRU'LVFUHWH*URXQG
6WUXFWXUHV
,IWKHJURXQGVWUXFWXUHFRQVLVWVRIDILQLWHQXPEHURIPHPEHUVDQGWKHSUREOHPLVRI
WKHW\SHVHWRXWLQ6HFWLRQSDUDJUDSKVLWRLLLDQGYWKHQWKHDERYH7KHRUHPV
DQG&RUROODULHVDUHVWLOOYDOLGLIZHUHSODFHLQ'HILQLWLRQVKDQGLLQ6HFWLRQ
´GHVLJQGRPDLQERXQGDU\´ZLWK´JURXQGVWUXFWXUH´
7KLVLVEHFDXVHIRUV\PPHWU\RIDGLVFUHWHSUREOHPQRWWKHGHVLJQGRPDLQERXQGDU\
EXWWKHJURXQGVWUXFWXUHQHHGVWREHV\PPHWULF
([DPSOH $UHPDUNDEOHH[DPSOHRIDV\PPHWULF GLVFUHWL]HG RSWLPDO WRSRORJ\ E\

2OKRIIHWDOLVVKRZQLQ)LJ
1RWH)RUIXUWKHUGHWDLOVRIV\PPHWU\DQGQRQXQLTXHQHVVVHHWKHUHYLHZDUWLFOH
E\5R]YDQ\D
D
E

)LJXUH2QHRIWKHHDUOLHVWH[DPSOHVRIDV\PPHWULFKLJKUHVROXWLRQGLVFUHWL]HGRSWLPDO
' WRSRORJ\ IRU D V\PPHWULF LQSXW E\ 2OKRII HW DO D SUREOHP VWDWHPHQW E
GLVFUHWL]HGRSWLPDOWRSRORJ\
48 G.I.N. Rozvany
1RQ6\PPHWULF([DFW2SWLPDO6ROXWLRQV IRU 6\PPHWULF 7RSRORJ\ 3URE

OHPVZLWK&RQWLQXRXVO\9DU\LQJ&URVVVHFWLRQV
1RQFRQYH[ FRVW IXQFWLRQV )RUQRQFRQYH[FRVWIXQFWLRQVWKHRSWLPDOWRSRORJ\

PD\EHQRQV\PPHWULFHYHQIRUV\PPHWULFJURXQGVWUXFWXUHVDQGV\PPHWULFORDGV
7KLVLVGHPRQVWUDWHGLQWZRH[DPSOHVLQ)LJ
7KHILUVWSUREOHPZHFRQVLGHUKHUHLVDQHOHPHQWDU\WUXVVFRQVLVWLQJRIWZRSDU
DOOHOEDUV)LJDWDNLQJ(XOHUEXFNOLQJLQWRFRQVLGHUDWLRQ,Q)LJDWKHIRXU
UROOHUVSUHYHQWWKHULJLGGLVNIURPURWDWLQJ7KHIRUFHVLQWKHWZREDUVPXVWWKHUHIRUH
VDWLVI\WKHHTXLOLEULXPFRQGLWLRQ3 457KHFURVVVHFWLRQVDUHHLWKHUVROLG
VTXDUHRIVROLGFLUFXODU7KHQRXUFRVWIXQFWLRQEHFRPHV
IRU ) ! $ N) IRU ) $ N )
ZKHUH)LVWKHPHPEHUIRUFHDQGNDQGNDUHNQRZQFRQVWDQWV
3
ULJLG
GLVN
4 5
E
D
4 5 4 5
FGH

)LJXUH&DVHVZLWKSLHFHZLVHFRQFDYHFRVWIXQFWLRQVZKHUHWKHRSWLPDOWRSRORJ\LV
QRQV\PPHWULF HYHQ IRU V\PPHWULF JURXQG VWUXFWXUHV DQG V\PPHWULF ORDGV D DQG E
SUREOHPVWDWHPHQWVFDQGHQRQV\PPHWULFRSWLPDOVROXWLRQVE DQG G V\PPHWULF
QRQRSWLPDOVROXWLRQV
,QWKLVH[DPSOHZHDUHXVLQJRQO\WKHVHFRQGSDUWRIEHFDXVHDWHQVLOHIRUFH
LQHLWKHUEDUZRXOGEHFOHDUO\QRQRSWLPDO$VVLJQLQJWKHYDOXH N DQGXQLWEDU
OHQJWKVWKHYROXPHRIWKHQRQV\PPHWULFVROXWLRQLQ)LJFLV9 DQGWKDWRIWKH
V\PPHWULFVROXWLRQLQ)LJGLV 9 ZKLFKLVDOPRVWRQHDQGDKDOIWLPHVWKH
YROXPHRIWKHRSWLPDOVROXWLRQ
)RUDEHDPZLWKDUHFWDQJXODUFURVVVHFWLRQRIYDULDEOHGHSWKEXWFRQVWDQWZLGWKWKH
FURVVVHFWLRQDODUHDLVJLYHQE\
$ N 0
ZKHUHNLVDJLYHQFRQVWDQWDQG0LVWKHEHDPEHQGLQJPRPHQW
,WFDQEHHDVLO\FKHFNHGWKDWLQWKLVFDVHWKHEHVWV\PPHWULFVROXWLRQ)LJE
KDVDJDLQ WLPHVWKHYROXPHRIWKHQRQV\PPHWULFRSWLPDOVROXWLRQV)LJH
DOWKRXJKWKHVSHFLILFFRVWIXQFWLRQLVV\PPHWULF$PRUHGHWDLOHGGLVFXVVLRQRIWKLV
SUREOHPFDQEHIRXQGLQWKHDXWKRU¶VILUVWERRN5R]YDQ\SS
$[LV\PPHWULF SHUIRUDWHG SODWHV .RODQHN DQG /HZLQVNL VKRZHG WKDW WKH
RSWLPDOULEOD\RXWIRUSHUIRUDWHGSODWHVZLWKD[LV\PPHWULFORDGDQGVXSSRUWVFDQEH
QRQD[LV\PPHWULFLIZHGRDZD\ZLWKWKHDVVXPSWLRQRID[LDOV\PPHWU\XVHGLQWKH
PLOHVWRQHSDSHUVE\&KHQJDQG2OKRIIDQGLQWKHSDSHUVE\5R]YDQ\
2OKRII%HQGVRH2QJDQG6]HWRDQG2QJ5R]YDQ\DQG6]HWR
1RQV\PPHWULFVSHFLILFFRVWIXQFWLRQV,WZDVVWDWHGLQ6HFWLRQWKDWRXUV\P
PHWU\ VWDWHPHQWV DUH YDOLG RQO\ IRU D FHUWDLQ W\SH RI V\PPHWULF VSHFLILF FRVW
IXQFWLRQVH[DPSOHVRIZKLFKDUHJLYHQLQRI/HFWXUH7KLVLPSOLHVWKDWIRU
0LFKHOOWUXVVHVIRUH[DPSOHZHDVVXPHWKHVDPHSHUPLVVLEOHVWUHVVHVLQWHQVLRQDQG
FRPSUHVVLRQ7KHDERYHDVVXPSWLRQLVRIFRXUVHXQUHDOLVWLFEHFDXVHGXHWREXFN
OLQJWKHFRPSUHVVLYHSHUPLVVLEOHVWUHVVHVIRUVOHQGHUWUXVVPHPEHUVDUHXVXDOO\PXFK
ORZHUWKDQWKHWHQVLOHRQHV
)RUGLIIHUHQWSHUPLVVLEOHVWUHVVHVLQWHQVLRQDQGFRPSUHVVLRQZHKDYHWRXVHLQ
JHQHUDOWKHH[WHQGHGRSWLPDOLW\FRQGLWLRQVLQRI/HFWXUH)LJXUHDVKRZVWKH
FRUUHFWRSWLPDOWRSRORJ\IRUDµ0LFKHOOFDQWLOHYHU¶IRUDVWUHVVUDWLRRIGHULYHGE\
*UDF]\NRZVNLDQG/HZLQVNLDDQGEDDQGE7KHVROXWLRQLQ)LJ
ELVDFWXDOO\YDOLGIRUDVWUHVVUDWLRRI/HZLQVNLHWDOD
)LJXUH D GHPRQVWUDWHV WKH IDFW WKDW HYHQ IRU V\PPHWULF GRPDLQ ERXQGDULHV
VXSSRUWV DQG ORDGV WKH RSWLPDO GHVLJQ FDQ EH QRQV\PPHWULF LI RXU VSHFLILF FRVW
IXQFWLRQLVQRWV\PPHWULFLHWKHVSHFLILFFRVWYROXPHGHSHQGVRQWKHVLJQRIWKH
IRUFHV
50 G.I.N. Rozvany
D
E

)LJXUH2SWLPDOWUXVVWRSRORJ\IRUXQHTXDOSHUPLVVLEOHVWUHVVHVXVLQJDWKHFRUUHFWHG
RSWLPDOLW\FULWHULD+HPS5R]YDQ\E0LFKHOO¶VRULJLQDORSWLPDOLW\FULWHULD
7KHDERYHVROXWLRQVZHUHSORWWHGRQWKHEDVLVRIWKHLUDQDO\WLFDOO\GHULYHGRSWLPDOVROXWLRQV
E\*UDF]\NRZVNLDQG/HZLQVNLDDQGEDDQGE
7KH(IIHFWRI1RQ]HUR6XSSRUW&RVWV
,WZDVSRLQWHGRXWE\0DULDQR9i]TXH](VSL(VSLDQG%UDYRWKDWLQWKHFDVH
RIYDULDEOHUHDFWLRQVDWVXSSRUWVWKHHIIHFWRIUHDFWLRQIRUFHVRQVXSSRUWVVKRXOGEH
LQFOXGHGLQWKHRSWLPL]DWLRQSURFHGXUH7KHHIIHFWRIUHDFWLRQFRVWVRQWKHRSWLPDO
WUXVVWRSRORJ\ZDVLQYHVWLJDWHGE\5R]YDQ\DQG6RNRO
,WZDVVKRZQLQWKHDERYHSDSHUWKDWDOORZDQFHIRUVXSSRUWFRVWVFDQEHPDGHDV
IROORZV,IWKHWKUHHUHDFWLRQFRPSRQHQWVDWDVXSSRUWRIDVSDFHWUXVVDUH;<DQG=
DQGWKHLUFRVWVDUH
[ ; K< ] =
UHVSHFWLYHO\WKHQZHFDQXVHWKHOD\RXWWKHRU\RXWOLQHGLQ/HFWXUHZLWKWKHGLI
IHUHQFHWKDWLQVWHDGRI]HURDGMRLQW GLVSODFHPHQW DW WKH VXSSRUWV ZH KDYH DGMRLQW
GLVSODFHPHQWVJLYHQE\WKHVXEJUDGLHQWVRI WKH UHDFWLRQ FRVW IXQFWLRQV LQ )RU
H[DPSOHLQWKH[GLUHFWLRQZHKDYH
X [ VJQ ; IRU ; z X d [ IRU ;
ZKHUH X LVWKHDGMRLQWGLVSODFHPHQWDWWKHVXSSRUWLQGLUHFWLRQ[
$Q,OOXVWUDWLYH([DPSOH±0RGLILHG0LFKHOO3UREOHP

&RQVLGHUDYHUWLFDOSRLQWORDGLQEHWZHHQWZRSLQVXSSRUWVDVLQ)LJDZKLFKDOVR
VKRZVWKHRSWLPDOWRSRORJ\IRUQRQ]HURUHDFWLRQFRVWV7KHFRVWRIYHUWLFDOUHDFWLRQV
GRHV QRW LQIOXHQFH WKH VROXWLRQ EHFDXVH WKHVH UHDFWLRQV DUH LQYDULDEOH $OVR WKH
FRUUHVSRQGLQJ DGMRLQW GLVSODFHPHQWV DW WKH YHUWLFDO VXSSRUWV FDXVH RQO\ D YHUWLFDO
ULJLGWUDQVODWLRQRIWKHHQWLUHDGMRLQWGLVSODFHPHQWILHOG
/FRVD
%
$
D &
[ 7
3
// D

E

F

)LJXUH,OOXVWUDWLYHH[DPSOHRSWLPDOWRSRORJLHVIRUUHDFWLRQFRVWIDFWRUVD [Ö /
E [Ö DQGF [Ö t /
52 G.I.N. Rozvany
7KHRSWLPDOWRSRORJ\)LJDFRQVLVWVRIDFLUFXODUIDQDQGWZRVWUDLJKWEDUV

HJ$%%\LQ/HFWXUHZHKDYHDGMRLQWSULQFLSDOVWUDLQVRINDQGNDORQJEDUV
%&DQG$%UHVSHFWLYHO\WKHWULDQJOH$%&EHLQJD7UHJLRQZLWKSULQFLSDOVWUDLQVRI
FRQVWDQW PDJQLWXGH DQG GLUHFWLRQ 7KHQ HOHPHQWDU\ FDOFXODWLRQV JLYH WKH DGMRLQW
VWUDLQLQWKHKRUL]RQWDO[GLUHFWLRQDV H [ N FRVD 7KLVLPSOLHVWKDWWKHDQJOHD
LVRSWLPDOLIWKHIROORZLQJHTXDOLW\LVVDWLVILHG
[ N[Ö N /FRVD
IRUWKLVRSWLPDOWRSRORJ\
)RU]HURUHDFWLRQFRVW [ ZHJHWE\ D q DQGWKHVROXWLRQUHGXFHVWR
RQHKDOIRIWKHZHOONQRZQ0LFKHOOWRSRORJ\LQ)LJE
)RU [ N/ JLYHV D ZLWK WKH RSWLPDO WRSRORJ\ LQ )LJ F %\ WKH
LQHTXDOLW\LQWKLVLVDOVRYDOLGIRU [ ! N/ 7KHWRSRORJ\LQ)LJFLVDFWXDOO\GXH
WR0LFKHOOEXWLWZDVIRUWKUHHJLYHQYHUWLFDOIRUFHVVWDWLFDOO\HTXLYDOHQWWRD
SLQDQGDUROOHU VOLGLQJVXSSRUW
&RQFOXGLQJ5HPDUNV
,QWKLVOHFWXUHZHGLVFXVVHGVRPHIXQGDPHQWDOSURSHUWLHVRIH[DFWRSWLPDOVWUXFWXUDO
WRSRORJLHVVXFKDVRUWKRJRQDOLW\QRQXQLTXHQHVVV\PPHWU\DQGVNHZV\PPHWU\
DQGPHWKRGVIRUWUDQVIRUPLQJVROXWLRQVIRUPRGLILHGERXQGDU\FRQGLWLRQV7KHHIIHFW
RIVXSSRUWFRVWRQWKHRSWLPDOWRSRORJ\ZDVDOVRUHYLHZHG

9DOLGDWLRQRI1XPHULFDO0HWKRGVE\$QDO\WLFDO
%HQFKPDUNVDQG9HULILFDWLRQRI([DFW6ROXWLRQV
E\1XPHULFDO0HWKRGV
*HRUJH,15R]YDQ\ DQG7RPDV]6RNyá

%XGDSHVW+XQJDU\

)DFXOW\RI&LYLO(QJLQHHULQJ:DUVDZ8QLYHUVLW\RI7HFKQRORJ\:DUVDZ3RODQG
,QWURGXFWLRQ
,QWKLVOHFWXUHZHGLVFXVVWKHYDOLGDWLRQRIYDULRXVQXPHULFDOPHWKRGVLQVWUXFWXUDO
WRSRORJ\RSWLPL]DWLRQ7KLVLVGRQHE\FRPSXWLQJQXPHULFDOO\RSWLPDOWRSRORJLHV
HJIRUSHUIRUDWHGSODWHVLQSODQHVWUHVVXVLQJYDULRXVQXPEHUVRIJURXQGHOHPHQWV
DQGYDULRXVYROXPHIUDFWLRQV7KHQWKHVWUXFWXUDOYROXPHYDOXHLVH[WUDSRODWHGIRU
WKHRUHWLFDOO\]HURYROXPHIUDFWLRQDQGLQILQLWHQXPEHURIJURXQGHOHPHQWVDQGWKLV
H[WUDSRODWHG YDOXH LV FRPSDUHG ZLWK WKDW RI WKH DQDO\WLFDO VROXWLRQ $V D UHODWHG
VXEMHFWYROXPHLQFUHDVLQJHIIHFWRIWRSRORJ\VLPSOLILFDWLRQLVDOVRGLVFXVVHG
,QWKHVHFRQGSDUWRIWKHOHFWXUHQHZDQGKLJKO\HIILFLHQWQXPHULFDOPHWKRGVDUH
GLVFXVVHG ZKLFK DUH FDSDEOH RI RSWLPL]LQJ WUXVV WRSRORJLHV ZLWK XS WR RYHU RQH
ELOOLRQ SRWHQWLDO WUXVV HOHPHQWV DQG WKHUHE\ SURYLGLQJ WKH YHULILFDWLRQ RI RSWLPDO
YROXPHVZLWKDQDFFXUDF\RIXSWRIRXUVLJQLILFDQWGLJLWV
7KHILUVWDXWKRUZLOOSUHVHQWWKHILUVWSDUWRIWKLVOHFWXUHDQGWKHVHFRQGDXWKRUWKH
VHFRQGSDUW
9DOLGDWLRQ RI 1XPHULFDO 0HWKRGV LQ 6WUXFWXUDO 7RSRORJ\

2SWLPL]DWLRQ
7KHDXWKRUKDVUHSHDWHGO\H[SUHVVHGFRQFHUQDERXWWKHODFNRI³TXDOLW\FRQWURO´LQ
WRSRORJ\RSWLPL]DWLRQ5R]YDQ\3RPH]DQVNL*DVSDUDQG4XHULQ5R]YDQ\
/HZLQVNL4XHULQDQG/RJRDDQG5R]YDQ\3RPH]DQVNL4XHULQ*DVSDUDQG
/RJRE+LVPHWKRGRIYHULI\LQJQXPHULFDOVROXWLRQVLQWRSRORJ\RSWLPL]DWLRQ
LVEDVHGRQWKHIDFWWKDWWKHRSWLPDOWRSRORJ\RISHUIRUDWHGSODWHVLQSODQHVWUHVVXQGHU
DFRPSOLDQFHFRQVWUDLQWWHQGVWRWKDWIRUSODQHWUXVVHVLIWKHYROXPHIUDFWLRQRIWKH
SODWHDSSURDFKHV]HUR5R]YDQ\2OKRII%HQGVRHHWDO%HQGVRHDQG
+DEHU$OODLUHDQG.RKQ5R]YDQ\D0RVWRIWKHDXWKRUVLQQX
PHULFDO WRSRORJ\ RSWLPL]DWLRQ VLPSO\ FRPSDUH WKHLU VROXWLRQV RSWLFDOO\ ZLWK WKH
H[DFWRSWLPDOWUXVVWRSRORJ\DQGDUHVDWLVILHGZLWKDYDJXHUHVHPEOHQFH7KLVLVD

DOI 10.1007/978-3-7091-1643-2_4, © CISM, Udine 2014
54 G.I.N. Rozvany and T. Sokół
YHU\VXEMHFWLYHPHWKRGIRUYHULI\LQJWRSRORJ\RSWLPL]DWLRQPHWKRGVDQGVROXWLRQV
,QRUGHUWRHQDEOHDUHOLDEOHFKHFNRQQXPHULFDOVROXWLRQVLQWRSRORJ\RSWLPL]D
WLRQ/HZLQVNL6RNRODQGWKHDXWKRUDUHGHULYLQJDQDO\WLFDOO\DVHULHVRIH[DFWWUXVV
WRSRORJLHVIRUSRSXODUEHQFKPDUNSUREOHPV5R]YDQ\/HZLQVNLDQG5R]YDQ\
DDQGE6RNRODQG/HZLQVNL5R]YDQ\DQG6RNRO6RNRODQG
5R]YDQ\7KHDFWXDOYHULILFDWLRQRIDVROXWLRQFRQVLVWVRIWKHIROORZLQJSUR
FHGXUH5R]YDQ\HWDODDQGE
D)RUJLYHQVHWRIUHVSRQVHRUEHKDYLRXUFRQVWUDLQWVGHULYHQXPHULFDOO\WKH
RSWLPDOWRSRORJ\IRUYDULRXVYROXPHIUDFWLRQVDQGYDULRXVQXPEHUVRIHOHPHQWV
E&DOFXODWHWKHVWUXFWXUDOYROXPHZHLJKWIRUHDFKVROXWLRQ
F([WUDSRODWHWKHYROXPHZHLJKWYDOXHIRU]HURYROXPHIUDFWLRQDQGLQILQLWH
QXPEHURIHOHPHQWV
G &RPSDUH WKLV H[WUDSRODWHG YDOXH ZLWK WKDW FDOFXODWHG DQDO\WLFDOO\ IRU WKH
H[DFWEHQFKPDUNV
LJKHU
1K
HU
RZ
(IILFLHQF\(
1O
1 QXPEHURIJURXQG
HOHPHQWV

9ROXPH)UDFWLRQ)
)LJXUH4XDQWLWDWLYHYHULILFDWLRQRIDQXPHULFDOWRSRORJ\VROXWLRQE\XVLQJH[DFWWUXVV
WRSRORJLHVDVEHQFKPDUNV
)RU FRPSOLDQFH FRQVWUDLQWV WKH DERYH PHWKRG UHGXFHV WR WKH RQH UHSUHVHQWHG
JUDSKLFDOO\LQ)LJ,QWKHODWWHUZHLQGLFDWHWKDWWKH³HIILFLHQF\´RIDWRSRORJ\
VKRXOGWHQGWRXQLW\DVZHGHFUHDVHWKHYROXPHIUDFWLRQDQGLQFUHDVHWKHQXPEHURI
Validation of Numerical Method by Analytical Benchmarks… 55
HOHPHQWV ,Q )LJ ( &090&9 HIILFLHQF\ RI D QXPHULFDO VROXWLRQ

& FRPSOLDQFH RI D QXPHULFDO VROXWLRQ &0 FRPSOLDQFH RI 0LFKHOO WUXVV
) YROXPH IUDFWLRQ 1 HOHPHQW QXPEHU 9 VWUXFWXUDO YROXPH RI D QXPHULFDO
VROXWLRQDQG90 VWUXFWXUDOYROXPHRI0LFKHOOWUXVV
$V FDQ EH VHHQ IURP WKH QH[W VHFWLRQ FHUWDLQ QXPHULFDO GLIILFXOWLHV KDG WR EH
RYHUFRPH 5R]YDQ\ HW DO D EXW WKH DERYH SURFHGXUH JLYHV D TXDQWLWDWLYH
FRQILUPDWLRQRI6,03VROXWLRQV
4XDOLW\ &RQWURO LQ 7RSRORJ\ 2SWLPL]DWLRQ 'LIILFXOWLHV DQG +RZ WR
2YHUFRPHWKHP
7KHSURFHGXUHRXWOLQHGLQ6HFWLRQ)LJPD\HQFRXQWHUFRPSXWDWLRQDOGLIIL
FXOWLHV,QWKLVVHFWLRQZHH[SODLQKRZWRKDQGOHWKHVHSLWIDOOV
D0HVKLQGHSHQGHQWPHWKRGV0HVKLQGHSHQGHQFHIRUPHWKRGVRIFKHFNHUERDUG
FRQWUROLVDPL[HGEOHVVLQJEHFDXVHLWSUHYHQWVJHWWLQJKLJKHUUHVROXWLRQVROXWLRQV
ZLWKPHVKUHILQHPHQW7KLVPHDQVWKDWLQ)LJZHJHWVWXFNRQWKHORZHVWFXUYH
PDUNHG³1ORZHU´DQGFDQQRWJHWRQWRWKHKLJKHUFXUYHV,WIROORZVWKDWZHFDQQRW
H[WUDSRODWHWRJHWDQHVWLPDWHIRUDQLQILQLWHO\GHQVHV\VWHP7KHVROXWLRQLVVLPSOH
QRW WR XVH PHVK LQGHSHQGHQW PHWKRGV LQ WKH SURSRVHG SURFHGXUH )RU JHWWLQJ LQ
FUHDVLQJUHVROXWLRQDQGFKHFNHUERDUGFRQWUROZHPD\XVHHJKLJKHURUGHUHOHPHQWV
RUVHYHUDO)(¶VSHUJURXQGHOHPHQW,QTXDOLW\FRQWUROFRPSXWHUWLPHLVRIVHFRQGDU\
LPSRUWDQFH EHFDXVH ZH FKHFN RQ WKH HIILFLHQF\ RI D SDUWLFXODU PHWKRG UHODWLYHO\
LQIUHTXHQWO\
7KLVLVLQFRQWUDVWWRLQGXVWULDODSSOLFDWLRQVZKHUHFRPSXWHUWLPHLVRISULPDU\
FRQVLGHUDWLRQ0DQXIDFWXUDELOLW\LVRIFRXUVHLPSRUWDQWLQSUDFWLFHDQGPD\MXVWLI\
ORZUHVROXWLRQVROXWLRQVEXWKLJKUHVROXWLRQVROXWLRQVZLOOEHLQFUHDVLQJO\LPSRUWDQW
ZLWKKLJKWHFKPDQXIDFWXULQJPHWKRGVPDVVSURGXFWLRQDQGPLFURQDQRWHFKQRORJ\
2I FRXUVH WKH DXWKRU DJUHHV ZLWK 6LJPXQG WKDW RQH FRXOG JHW D VHULHV RI
VROXWLRQV RI GLIIHUHQW UHVROXWLRQV ZLWK PHVKLQGHSHQGHQW ILOWHUV E\ GHFUHDVLQJ WKH
ILOWHUUDGLXVSURJUHVVLYHO\
E(UURQHRXVO\KLJKHUVWLIIQHVVIRUFRDUVHUPHVKHV,WZDVIRXQGLQHDUOLHUVWXGLHV
HJ5R]YDQ\3RPH]DQVNL*DVSDUDQG4XHULQWKDWFRQWUDU\WRWKHWKHRUHWLFDO
WUHQGVLQ)LJWKH³HIILFLHQF\´GHFUHDVHGZLWKHOHPHQWUHILQHPHQWLQQXPHULFDO
H[SHULPHQWVXVLQJ6,03$VLWZDVUHDOL]HGODWHUWKLVZDVEHFDXVHWKHQXPEHURI
)(¶VDQGWKHQXPEHURIJURXQGHOHPHQWVZHUHLQFUHDVHGVLPXOWDQHRXVO\NHHSLQJWKH
QXPEHURI)(¶VSHUJURXQGHOHPHQWFRQVWDQWHJ7KHFRDUVHUQHWRIODUJHUVL]H
HOHPHQWVLQFUHDVHGWKHGLVFUHWL]DWLRQHUURUIRUVPDOOHUQXPEHURIHOHPHQWVHUURQH
RXVO\LQFUHDVLQJWKHVWLIIQHVVDQGGHFUHDVLQJWKHFRPSOLDQFHHYHQZLWKRXWFRUQHU
FRQWDFWVLQWKHVROXWLRQ7KLVSKHQRPHQRQLVZHOONQRZQIURPWKHILQLWHHOHPHQW
OLWHUDWXUH
7KHDERYHSUREOHPZDVRYHUFRPHLQLPSURYHGH[SHULPHQWVHJ5R]YDQ\HWDO
DDQGELQZKLFKWKHWRWDOQXPEHURI)(¶VZDVNHSWFRQVWDQWLQDOOFDOFXODWLRQV
EXWWKHQXPEHURIJURXQGHOHPHQWVZDVSURJUHVVLYHO\LQFUHDVHGLHWKHQXPEHURI
)(¶V SHU JURXQG HOHPHQW GHFUHDVHG $IWHU WKLV HVVHQWLDO FRUUHFWLRQ WKH H[SHFWHG
WUHQGZDVREVHUYHGLQDOOUHVXOWV
F6XGGHQGURSRIWKHHIILFLHQF\FXUYHDWORZYROXPHIUDFWLRQV,WZDVDOVRIRXQG
WKDWWKHHIILFLHQF\FXUYHVLQ)LJVXGGHQO\GURSWDNHRQYHU\ORZYDOXHVDWORZ
YROXPHIUDFWLRQV7KLVLVEHFDXVHIRUDJLYHQQXPEHURIJURXQGHOHPHQWVDYHU\ORZ
YROXPHIUDFWLRQDOORZVWRRIHZ³EODFN´HOHPHQWVIRUDQHIILFLHQWORDGWUDQVPLVVLRQ,Q
0LFKHOO¶V³ELF\FOHZKHHOSUREOHPÍRUH[DPSOHDORZYROXPHIUDFWLRQZRXOGDOORZ
RQO\ WRR IHZ ³VSRNHV´ LQ WKH VROXWLRQ UHVXOWLQJ LQ KLJKHU FRPSOLDQFH YDOXH VHH
5R]YDQ\4XHULQ*DVSDU3RPH]DQVNL$WHYHQORZHUYROXPH IUDFWLRQ WKH
DOORZHGQXPEHURIHOHPHQWVLVLQVXIILFLHQWIRUWUDQVPLWWLQJWKHSUHVFULEHGORDGVDWDOO
DQGWKHFRPSOLDQFHYDOXHJRHVWRLQILQLW\RQO\WKHRUHWLFDOO\LQSUDFWLFHZHXVHDYHU\
ORZILQLWHWKLFNQHVVIRU³ZKLWH´HOHPHQWV
7KHVROXWLRQWRWKHDERYHGLIILFXOW\LVWRLJQRUHV\VWHPDWLFDOO\WKHVROXWLRQVZLWK
VXFK³IRUFHG´LQFRUUHFWWRSRORJLHVZKHQZHH[WUDSRODWHIRUQHDU]HURYROXPHIUDF
WLRQDQGQHDULQILQLWH)(QXPEHUVHH5R]YDQ\HWDOE

)LJXUH 2SWLPDO WRSRORJ\ IRU WZR GLVSODFHPHQW FRQVWUDLQWV DQG QRQ]HUR YROXPH
IUDFWLRQV.iURO\LDQG5R]YDQ\
G3RLVVRQ¶VUDWLR,Q0LFKHOOVWUXFWXUHVWKHPHPEHUVDUHLQXQLD[LDOVWUHVVDQGWKH
YROXPHRIPHPEHULQWHUVHFWLRQVLVQHJOLJLEOH)RUWKLVUHDVRQZHKDYHIRXQGWKDWZH
JHWWKHEHVWFRQYHUJHQFHWRWKHFRUUHVSRQGLQJ0LFKHOOWUXVVLIZHXVH]HUR3RLVVRQ¶V
UDWLRIRUWKHSHUIRUDWHGSODWHLQWKHSURSRVHGTXDOLW\WHVW
H7KHFKRLFHRIWKHWHVWSUREOHP,WKDVEHHQVKRZQ5R]YDQ\4XHULQDQG*DVSDU
5R]YDQ\EWKDWLQVRPHFDVHVWKH0LFKHOOWUXVVZLWKDQHDU]HURYROXPH
IUDFWLRQJLYHVDPXFKKLJKHUYDOXHIRUWKHµH[WHQGHG¶REMHFWLYHIXQFWLRQ&9WKDQ
SHUIRUDWHGSODWHVROXWLRQVIRURWKHUYROXPHIUDFWLRQV,QVXFKEHQFKPDUNSUREOHPV
WKHHIILFLHQF\FXUYHVLQ)LJDSSURDFKWKHOLQHZLWK( IURPDERYH+RZHYHULWLV
SUHIHUDEOHWRVHOHFWEHQFKPDUNSUREOHPVIRUZKLFKD0LFKHOOWUXVVFRUUHVSRQGVWR
WKHH[WHQGHGRSWLPXP
(YLGHQFH RI &RQYHUJHQFH RI 3HUIRUDWHG 3ODWH 6ROXWLRQV WR WKH 2SWLPDO
7UXVV 6ROXWLRQ IRU 1RQVHOIDGMRLQW 3UREOHPV 7ZR 'LVSODFHPHQW &RQ
VWUDLQWV
.iURO\L REWDLQHG DQDO\WLFDO VROXWLRQV IRU L WZR GLVSODFHPHQW FRQVWUDLQWV D
QRQVHOIDGMRLQWSUREOHPDQGLLQRQ]HURYROXPHIUDFWLRQV.iURO\LDQG5R]YDQ\
D UDWKHU UHPDUNDEOH DFKLHYHPHQW 7KH SUREOHP LV VKRZQ LQ )LJ WKH
QRQRUWKRJRQDO UDQN PLFURVWUXFWXUH LV EDVHG RQ /XULH¶V D DQG E
ILQGLQJV
7KHFRQYHUJHQFHRI.iURO\L¶VDQDO\WLFDOVROXWLRQVWRWKHRSWLPDOWUXVVVROXWLRQIRUWKH
VDPHSUREOHP%LUNHU/HZLQVNLDQG5R]YDQ\LVVKRZQLQ)LJIURPWKH
UHYLHZ E\ 5R]YDQ\ ,QVWHDG RI GHFUHDVLQJ WKH YROXPH IUDFWLRQ .iURO\L LQ
FUHDVHGWKHYDOXHRI<RXQJ¶VPRGXOXVIRUJLYHQGLVSODFHPHQWVZKLFKKDVWKHVDPH
HIIHFW
9ROXPH,QFUHDVLQJ(IIHFWRI7RSRORJ\6LPSOLILFDWLRQ
)RUFLQJDVLPSOHUWRSRORJ\WKDQWKHWKHRUHWLFDORSWLPXPE\DGGLWLRQDOFRQVWUDLQWVRU
RWKHUSUREOHPPRGLILFDWLRQPD\KDYHVHYHUDODGYDQWDJHVVXFKDVHDVHRIPDQXIDF
WXULQJ PHVK LQGHSHQGHQFH DQG FKHFNHUERDUG FRQWURO +RZHYHU WRSRORJ\
VLPSOLILFDWLRQPD\UHVXOWLQDFRQVLGHUDEOHLQFUHDVHRIWKHVWUXFWXUDOYROXPH
7KHDERYHSUREOHPLVGLVFXVVHGE\5R]YDQ\4XHULQ*DVSDUDQG3RPH]DQVNL
2QHRIWKHH[DPSOHVWKH\FRQVLGHUHGLVWKHµ0LFKHOOELF\FOHZKHHOSUREOHP¶
LQ)LJDLQZKLFKZHVKRZRQO\WKHFHQWHUOLQHRIµPHPEHUV¶7KHODWWHUDFWXDOO\
FRQVLVW RI JURXQG HOHPHQW RI GLPHQVLRQ D u D WKH PHPEHU L KDYLQJ D ZLGWK RI
EL ZLWK D EL / VHH)LJEWRSULJKW,IDOOPHPEHUVKDYHWKHVDPHZLGWK
ZHPD\UHSODFH EL ZLWK E ,WZDVVKRZQE\WKHDXWKRUHJ5R]YDQ\EWKDWIRU
]HUR3RLVVRQ¶VUDWLRWKHZHLJKWFDQEHH[SUHVVHGDV

§ S · ) /J
: ¨ Q WDQ ¸
© Q ¹ (&
ZKHUHQLVWKHQXPEHURIFRPSUHVVLRQPHPEHUVLQWKHWUXVVVHH)LJD)LVWKH
SRLQWORDG/LVWKHVSDQ J LVWKHVSHFLILFZHLJKW(LV<RXQJ¶VPRGXOXVDQG&LV
WKHVSHFLILHGOLPLWRQWKHFRPSOLDQFH

)LJXUH&RQYHUJHQFHRIDQDO\WLFDOVROXWLRQVIRUSHUIRUDWHGSODWHVZLWKQRQ]HURYROXPH
IUDFWLRQV.iURO\LDQG5R]YDQ\WRWKHRSWLPDOWUXVVVROXWLRQ%LUNHUHWDOIRU
WKHVDPHGLVSODFHPHQWFRQVWUDLQWV
)RU Q f ZH JHW E\ WDNLQJ WKH ILUVW WHUP RI WKH 7D\ORUVHULHV WKH UHVXOW E\
0LFKHOO
) /J ) /J
: S
(& (&

)LJXUH([DPSOHVKRZLQJWKHZHLJKWLQFUHDVLQJHIIHFWRIWRSRORJ\VLPSOLILFDWLRQ
7DNLQJDPRUHSUDFWLFDOVROXWLRQRIQ ZHJHWIURP
) / J
:
(&
ZKLFKLVRQO\SHUFHQWKHDYLHUWKDQWKHWKHRUHWLFDORSWLPXP
)RUQ ZHJHWWKHVROXWLRQLQ)LJEZLWKWKHZHLJKWRI
) / J
:
(&
ZKLFKLVSHUFHQWKHDYLHUWKDQWKHVLGHGVROXWLRQ7KHVROXWLRQLQ)LJE
FDQEHIRUFHGE\RQHRIWKHIROORZLQJFRQVWUDLQWV
) /
Et RU S d /
&(W
ZKHUHELVWKHPHPEHUZLGWKWLVWKHSODWHWKLFNQHVVDQGSLVWKHSHULPHWHU
,W ZDV SRLQWHG RXW E\ 0LQJ =KRX WKDW LI WKH SHULPHWHU LV FRQVWUDLQHG WR VD\
S d / WKHQWKHVROXWLRQGHJHQHUDWHVLQWRDVLQJOHIOH[XUDOEHDPFRQQHFWLQJWKH
VXSSRUWVDWWKHERWWRPRIWKHGRPDLQ7KHQZHKDYHDZHLJKWRI
) / J
:
(&
ZKLFKLVRYHUSHUFHQWRIWKHZHLJKWRIWKHWZHQW\VLGHGWUXVV1DWXUDOO\WKLVLVD
SXUHO\ DFDGHPLF VROXWLRQ EXW LW VKRZV WKH ZHLJKW LQFUHDVLQJ HIIHFW RI WRR VWURQJ
WRSRORJ\VLPSOLILFDWLRQV
7RSRORJ\2SWLPL]DWLRQRI/DUJH6FDOH7UXVVHV8VLQJ*URXQG
6WUXFWXUH$SSURDFKZLWK6HOHFWLYH6XEVHWVRI$FWLYH%DUV
&ODVVLFDO3ODVWLF/D\RXW2SWLPL]DWLRQRI7UXVVHV
7KH WUXVV WRSRORJ\ RSWLPL]DWLRQ SUREOHP FDQ EH GHILQHG DQG VROYHG XVLQJ PDQ\
GLIIHUHQW DSSURDFKHV ,W FDQ DOVR EH ZULWWHQ LQ PDQ\ IRUPV RIWHQ HTXLYDOHQW IURP
PDWKHPDWLFDO SRLQW RI YLHZ EXW GLIIHUHQW UHJDUGLQJ WR FRPSXWDWLRQDO HIILFLHQF\
,QWKHSUHVHQWOHFWXUHZHSD\WKHDWWHQWLRQWRWKHZHOONQRZQJURXQGVWUXFWXUHDS
SURDFKDQGVRFDOOHGSODVWLFGHVLJQRSWLPL]DWLRQSUREOHP7KHPDLQGLIILFXOW\RIWKLV
DSSURDFKFRQVLVWVLQDKXJHRIWHQLQWUDFWDEOHQXPEHURISRWHQWLDOEDUVFRQQHFWLQJ
DQ\ SDLU RI QRGHV )RU H[DPSOH WKH JURXQG VWUXFWXUH GHILQHG LQ D VTXDUH GRPDLQ
GLYLGHG E\ QuQ FHOOV Q QRGHV LQ [ DQG \ GLUHFWLRQV KDV Q QRGHV DQG
0 Q>Q@SRWHQWLDOEDUV,WPHDQVWKDWWKHQXPEHURIEDUVLQFUHDVHV
DVUDSLGO\DVQIRUQ ZHJHW0 EDUVDQGMXVWIRU Q WKH
QXPEHURISRWHQWLDOEDUVH[FHHGVRQHELOOLRQ7KLVGLIILFXOW\FDQEHRYHUFRPH
XVLQJWKHDGDSWLYHJURXQGVWUXFWXUHDSSURDFKZLWKWKHVHOHFWLYHVXEVHWVRIDFWLYHEDUV
7KLVQHZVROXWLRQPHWKRGZDVRULJLQDOO\SURSRVHGE\*LOEHUWDQG7\DVDQG
PRGLILHG E\ 6RNyá E 7KH ODWWHU PHWKRG ZLOO EH EULHIO\ SUHVHQWHG EHOORZ
'HVSLWH WKH QHFHVVLW\ RI VROYLQJ ODUJHVFDOH RSWLPL]DWLRQ SUREOHPV WKH JURXQG
VWUXFWXUHDSSURDFKKDVDYHU\LPSRUWDQWIXWXUH'XHWRWKH IL[HG QRGHV ZLWK SUH

VFULEHGFRQVWDQWFRRUGLQDWHVWKHRSWLPL]DWLRQSUREOHPFDQEHIRUPXODWHGDQGVROYHG
XVLQJWKHOLQHDUSURJUDPPLQJ/3PHWKRGV&RQWUDU\WRRWKHUIRUPXODWLRQVHVSH
FLDOO\QRQOLQHDUDQGQRQFRQYH[LWDVVXUHVILQGLQJWKHJOREDORSWLPXPZKLFKLVWKH
OHDVWYROXPHWUXVVWUDQVPLWWLQJJLYHQORDGLQJWRJLYHQVXSSRUWV
7KHSULPDOORZHUERXQGIRUPXODWLRQRIWKHSODVWLFGHVLJQRSWLPL]DWLRQSUREOHP
RI WUXVV ZLWK GLIIHUHQW VWUHVV OLPLWV IRU WHQVLRQ V7 DQG FRPSUHVVLRQ V& DQG IRU
RQHORDGFDVHFDQEHGHILQHGDVIROORZV
/7 7 /7 &
PLQ0 9
7 &5 V7 V&

VW % 7 & 3
7
7 t & t
ZKHUHWKHREMHFWLYHIXQFWLRQ9LVWKHYROXPHRIWKHIXOO\VWUHVVHGWUXVV/LVWKHYHFWRU
RIEDUOHQJWKV7DQG&DUHWKHYHFWRUVRIPHPEHUWHQVLRQDQGFRPSUHVVLRQIRUFHV
UHVSHFWLYHO\3LVWKHYHFWRURIQRGDOORDGVH[WHUQDOORDGLQJDQG%LVWKHJHRPHWULF
PDWUL[LQFOXGLQJGLUHFWLRQDOFRVLQHVRIEDUV
,QWKLVIRUPXODWLRQHYHU\EDULVUHSUHVHQWHGE\WZRLQGHSHQGHQWGHVLJQYDULDEOHV
FRUUHVSRQGLQJWRWHQVLRQ7LDQGFRPSUHVVLRQ&LIRUFHVL «0+HQFHIRUWUXVV
FRPSRVHGRI0PHPEHUVWKH/3SUREOHPLQYROYHV0GHVLJQYDULDEOHVZLWK1
HTXDOLW\FRQVWUDLQVZKHUH1GHQRWHVWKHQXPEHURIGHJUHHVRIIUHHGRP3ULPDOIRUP
LVDOVRFDOOHGWKHORZHUERXQGIRUPEHFDXVHHYHU\IHDVLEOHVROXWLRQ±DQ\WUXVV
FKRVHQIURPWKHJURXQGVWUXFWXUHDQGVDWLVI\LQJWKHHTXLOLEULXPHTXDWLRQV±KDV
WKHYROXPHJUHDWHUWKDQRUHTXDOWRWKHRSWLPXPRQH7KHGLUHFWVROXWLRQRILV
SRVVLEOHIRUPRGHUDWHO\ODUJHSUREOHPVXSWRIHZPLOOLRQRIEDUVXVLQJWKHLQWHULRU
SRLQWPHWKRGWRJHWKHUZLWKVSDUVHPDWUL[UHSUHVHQWDWLRQVHH6RNyáD/DUJHU
SUREOHPV ZLWK GHQVHU JURXQG VWUXFWXUHV SURYLGLQJ PRUH DFFXUDWH UHVXOWV UHTXLUH
PXFKPRUHVXEWOHPHWKRGV
7KHGXDOXSSHUERXQGIRUPXODWLRQIRUWKHVDPHSUREOHPLVJLYHQE\
PD[
1
: 37 X
X5
VW % X d / V 7
% X t / V &
ZKHUH X UHSUHVHQWV WKH YHFWRU RI DGMRLQW QRGDO GLVSODFHPHQWV FRUUHVSRQGLQJ WR
/DJUDQJHPXOWLSOLHUVDQG:LVWKHZRUNRIWKHH[WHUQDOIRUFHV3RQ X )RUWKHWUXVV
PDGHRIRQHPDWHULDOZLWKFRQVWDQW<RXQJPRGXOXV(WKHUHDOGLVSODFHPHQWVDUH
SURSRUWLRQDOWRDGMRLQWGLVSODFHPHQWV X ( X
7KHODWWHUSUREOHPLQYROYHVRQO\1GHVLJQYDULDEOHVIRUGHQVHJURXQGVWUXFWXUH1
LVPDQ\WLPHVVPDOOHUWKDQ0EXWVXEMHFWHGWR0LQHTXDOLW\FRQVWUDLQV$IWHU
WUDQVIRUPDWLRQWRDVWDQGDUG/3IRUPWKHGXDOSUREOHPLVVOLJKWO\JUHDWHUWKDQWKH
SULPDO RQH 1 EDVLF GHVLJQ YDULDEOHV SOXV 0 VODFN YDULDEOHV ZKLFK FRQYHUW LQH
TXDOLW\WRHTXDOLW\FRQVWUDLQV+HQFHLWLVPRUHHIILFLHQWWRVROYHWKDQ7KH
GXDOIRUPXODWLRQFDQKRZHYHUEHXWLOL]HGLQWKHLWHUDWLYHPHWKRGGHVFULEHGEHOORZ
DQGFDQVHUYHDVWKHILOWHUIRUXSGDWLQJWKHVHWRIDFWLYHEDUVLQVXFFHHGLQJLWHUDWLRQV
,W VKRXOG EH QRWHG WKDW WKH LQHTXDOLW\ FRQVWUDLQWV JLYHQ LQ FRUUHVSRQG WR
0LFKHOO¶VRSWLPDOLW\FULWHULD0LFKHOOZKLFKVWDWHVWKDWLQDQRSWLPDOWUXVVWKH
PDJQLWXGHRIWKHDGMRLQWPHPEHUVWUDLQVDUHUHVWULFWHGE\
V & d H L d V 7 IRUL «0
$IWHUVLPSOHDOJHEUDLFRSHUDWLRQVRQHFDQGHGXFHWKDWWKHFRQVWUDLQWVRISUREOHP
DUHHTXLYDOHQWWR0LFKHOO¶VFRQVWUDLQWVDQGKHQFHWKH\DUHVXIILFLHQWWRREWDLQDQ
RSWLPDOWUXVVLIRQHFDQDGMXVWWKHGLVSODFHPHQWILHOGKHUHUHSUHVHQWHGE\GLVFUHWH
QRGDOGLVSODFHPHQWV X LQVXFKDZD\WKDWKROGVIRUHYHU\SRWHQWLDOEDURIWKH
JURXQGVWUXFWXUH)RUFRQYHQLHQFHLWLVZRUWKLQWURGXFLQJDQRUPDOL]HGVWUDLQPHDV
XUHGHILQHGE\
V7HL IRU H L t
HL ® RU H L PD[V &H L V 7 H L
¯ V & H L IRU H L
KHQFHWKH0LFKHOO¶VFRQVWUDLQWVFDQQRZEHZULWWHQDV
H L d IRUL «0
7KH$FWLYH6HW0HWKRG:LWK$GDSWLYH*URXQG6WUXFWXUHV
7KHPDLQLGHDRIWKHSURSRVHGPHWKRGLVWRUHSODFHRQHODUJH/3SUREOHPE\DQ
DSSURSULDWHVHULHVRIVLJQLILFDQWO\VPDOOHUVXESUREOHPVZKLFKLPSURYHWKHVROXWLRQL
LWHUDWLYHO\7KHLQLWLDOJURXQGVWUXFWXUHLQFOXGHVRQO\WKHVKRUWHVWEDUVFRQQHFWLQJWKH
QHLJKERXULQJQRGHVVHH)LJD,QWKHQH[WLWHUDWLRQVORQJHUDQGORQJHUEDUVDUH
FRQVLGHUHGDVWKHFDQGLGDWHVIRUDGGLWLRQVHH)LJEGDQGDWWKHVDPHWLPHWKHROG
XQQHFHVVDU\ EDUV DUH HOLPLQDWHG IURP WKH VWUXFWXUH 7KH 0LFKHOO FRQVWUDLQWV
VHUYHDVWKHFULWHULRQIRUDGGLQJRUUHPRYLQJEDUV
,QWKHFXUUHQWYHUVLRQRIWKHSURJUDPGHYHORSHGWKHZKROHRSWLPL]DWLRQSURFHVV
LVSHUIRUPHGRYHUDIL[HGJULGRIQRGHVUHJXODUO\VSDFHGLQ[DQG\GLUHFWLRQVVHH
)LJ7KLVUHJXODULW\KHOSVLQSUHVHUYLQJDORWRI5$0DQG&38E\XVLQJ³SDW
WHUQV´RIHOHPHQWVGHILQLQJIDPLOLHVRILGHQWLFDOEDUVE\PHDQVRIWUDQVODWLRQV6RNyá
D(DFKLWHUDWLRQEHJLQVZLWKDSURSHUO\XSGDWHGVXEVHWRIPDFWLYHEDUVZKHUH
P0)RUWKHUHPDLQLQJKXJHVXEVHWRILQDFWLYHEDUVZHDVVXPHWKDWWKHLUFURVV
VHFWLRQDODUHDVDUH]HURVR7L &L IRUDSSURSULDWHLQGLFHVL7KXVPRVWGHVLJQ

YDULDEOHVDUHHOLPLQDWHGDSULRULDQGWKHQWKHUHGXFHGSUREOHPLVVROYHGXVLQJ
WKHSULPDOGXDOLQWHULRUSRLQWPHWKRG1RWHWKDWXVLQJWKLVPHWKRGZHJHWIURPWKH
VROXWLRQERWKSULPDO7&DQGGXDO X GHVLJQYDULDEOHVDQGWKHODWWHURQHVGRQRW
UHTXLUH DGGLWLRQDO FRPSXWDWLRQV ZH JHW WKHP IRU IUHH 2I FRXUVH DFFRUGLQJ WR
SULPDOGXDOWKHRUHPIRUDJLYHQVXEVHWRIDFWLYHEDUVZHKDYH9PLQ :PD[EXWEHIRUH
JRLQJWRWKHQH[WLWHUDWLRQZHFKHFNZKDWKDSSHQVLIZHDGGVRPHQHZEDUVLQWRWKH
FXUUHQWVHW2EYLRXVO\DGGLQJDQ\QHZEDUWKDWVDWLVILHVGRHVQRWFKDQJHWKH
FXUUHQWVROXWLRQVRLWLVQDWXUDOWKDWZHDUHORRNLQJIRUQHZFDQGLGDWHEDUVIRUZKLFK
WKH QRUPDOL]HG VWUDLQ LV JUHDWHU WKHQ RQH H ! +RZHYHU WDNLQJ LQWR DFFRXQW
L
LQHYLWDEOHQXPHULFDOHUURUVLWLVEHWWHUWRDFWLYDWHDQHZEDULI H L ! H ZKHUHHLVD
VPDOOSRVLWLYHQXPEHUHJ H LVDJRRGFKRLFH$WWKHVDPHWLPHZHFDQ
UHPRYHXQQHFHVVDU\EDUVLIWKHLUQRUPDOL]HGVWUDLQVDUHVPDOOHQRXJK,WVKRXOGEH
QRWHGWKDWUHPRYLQJWRRPDQ\EDUVHYHQWKRVHZLWK]HURFURVVVHFWLRQDODUHDFDQ
OHDG WR D ZRUVH GLVWULEXWLRQ RI QRGDO GLVSODFHPHQWV X DQG WKH SURILW REWDLQHG LQ
SUHYLRXVLWHUDWLRQVFDQEHORVWLQODWHURQHV2QWKHEDVLVRIPDQ\QXPHULFDOWHVWVZH
SURSRVHWRUHPRYHEDUVIURPWKHDFWLYHVHWLI H L

D[[EDUV E[[EDUV

F[[EDUV G[[EDUV
)LJXUH7KHVXFFHVVLYHJURXQGVWUXFWXUHVZLWKGLIIHUHQWGLVWDQFHVRIFRQQHFWLRQV
7KH PD[LPXP YDOXH RI QRUPDOL]HG VWUDLQV H PD[ PD[ H L H PD[ t LV WKH JRRG
L
HVWLPDWRURISRVVLEOHLPSURYHPHQWRIWKHVROXWLRQLQWKHQH[WLWHUDWLRQ,IWKHFXUUHQW
GLVSODFHPHQWV X DUH SURSRUWLRQDOO\ UHGXFHG E\ H PD[ WKHQ WKH\ ZLOO VDWLVI\ DOO
FRQVWUDLQWVRIDQGKHQFHGHILQHDIHDVLEOHVROXWLRQIRUWKHQH[WLWHUDWLRQ,WPHDQV
ROG
WKDW :PD[ t :PD[
QHZ
t :PD[
ROG
H PD[ $W WKH VDPH WLPH WKH YROXPH 9PLQ FDQ DOVR EH
UHGXFHG E\ WKH VDPH IDFWRU 0RUHRYHU LI H PD[ IRU D IXOO\ FRQQHFWHG JURXQG
VWUXFWXUHWKHQQRPRUHLPSURYHPHQWVFDQEHDFKLHYHGDQGWKHFXUUHQWVROXWLRQKDVWR
EHRSWLPDO7KXVWKHDERYHFRQGLWLRQLVDVLPSOHVWRSSLQJFULWHULRQIRUWKHDOJRULWKP
SUHVHQWHG
7KHVWHSE\VWHSSURFHGXUHIRUWKHDOJRULWKPGLVFXVVHGFDQEHGHVFULEHGDVIRO
ORZV
)LUVWLWHUDWLRQ
6HWLWHU DQGGLVW
6HW WKH LQLWLDO JURXQG VWUXFWXUH PuQu ZLWK EDUV FRQQHFWLQJ WKH
QHLJKERXULQJQRGHVDVVKRZQLQ)LJD7KHVHEDUVZLOODOZD\VEH
DFWLYH HYHQ IRU ]HUR FURVV VHFWLRQDO DUHD WKH\ SOD\ D UROH RI
JURXQGZRUN WR REWDLQ QRGDO GLVSODFHPHQWV RI µHPSW\¶ QRGHV QRW
FRQQHFWHGZLWKRWKHUEDUV
6ROYHSULPDOSUREOHPDQGJHWGXDOYDULDEOHV X
1H[WLWHUDWLRQV
,QFUHPHQWLWHUDQGGLVWLWHUDQGGLVW PD[GLVWPD[GLVW
6HOHFWWKHVHWRIDFWLYHEDUVIURPJURXQGVWUXFWXUHPuQGLVWuGLVW
± FRPSXWHQRUPDOL]HGVWUDLQIRUHYHU\QHZEDU X LWHU H LWHU
± LI H L t WRO WKHQDFWLYDWHDGGEDU
± RWKHUZLVHLI H L DQGGLVWGLVWPD[WKHQGHDFWLYDWHUHPRYH
EDU
± LIGLVWGLVWPD[DQGµWKHQXPEHURIDGGHGEDUVLVWRRVPDOO¶WKHQ
GLVWDQGUHSHDWIURPVWHS
&KHFNVWRSSLQJFULWHULRQ
± LIGLVW GLVWPD[DQGµWKHUHDUHQRQHZEDUVDGGHG¶WKHQILQLVKZH
DSSURDFKWKHRSWLPXPVROXWLRQEHFDXVHIRUDOOSRWHQWLDOEDUVWKH
FRQVWUDLQWVDUHVDWLVILHGRULQRWKHUZRUGV H PD[ DQGWKH
VROXWLRQFDQQRWEHIXUWKHULPSURYHG
6ROYHSULPDOSUREOHPDQGJHWGXDOYDULDEOHV X LWHU
5HSHDWIURPVWHS
([DPSOH±0RGLILHG0LFKHOO:KHHO3UREOHP
&RQVLGHUWKHSUREOHPVKRZQLQ)LJ,WLVVLPLODUWRZHOONQRZQ0LFKHOO
SUREOHP FRQVLVWLQJ LQ ILQGLQJ WKH OHDVWYROXPH VWUXFWXUH WUDQVPLWWLQJ WKH FHQWUDO
SRLQWORDGWRWZRSLQVXSSRUWVVHH)LJEXWQRZLWLVFRQVLGHUHGIRUGLIIHUHQW
SHUPLVVLEOHVWUHVVOLPLWVIRUWHQVLRQ V7DQGFRPSUHVVLRQ V&7KLVSUREOHPKDVUH
FHQWO\EHHQVROYHGE\DXWKRUV5R]YDQ\DQG6RNyáERWKIRUKDOIDQGIXOOSODQH
GRPDLQDQGFRQILUPHGZLWKWKHQXPHULFDOUHVXOWVREWDLQHGE\WKHSURJUDPGHVFULEHG
LQSUHYLRXVVHFWLRQ)LJVKRZVDVZHOOWKHSUREOHPVHWWLQJDVWKHRSWLPDOWRSRORJ\
GHGXFHGIURPQXPHULFDOSUHGLFWLRQV

SULQFLSDODGMRLQWVWUDLQV
LQWKH2UHJLRQ
UHJLRQERXQGDULHV
2
µFRQFHQWUDWHG¶PHPEHUV
Į
µGLVWULEXWHG¶PHPEHUV
7 7
7

)LJXUH2SWLPDOWRSRORJ\RIPRGLILHG0LFKHOOSUREOHP
7KHH[DFWRSWLPDOYROXPHIRUV7 V&DQGWRSKDOISODLQGRPDLQLVHTXDOWRVHH

5R]YDQ\DQG6RNyá
3/ § S · 3/
9+3 ¨ ¸ #
V7 © ¹ V 7
ZKLOHWKHVDPHYROXPHIRUIXOOSODQHRUERWWRPKDOISODQHGRPDLQLVHTXDOWR

3/ § S · 3/
9)3 ¨ ¸ #
V7 © ¹ V7
7KHKLVWRU\RIVXFFHVVLYHLWHUDWLRQVRIWKHPHWKRGSUHVHQWHGLQ6HFIRUWKH
DERYHSUREOHPVLVJLYHQLQ)LJVDQGUHVSHFWLYHO\'HVSLWHVRPHLQHYLWDEOH
QXPHULFDOQRLVHVLQPLGGOHLWHUDWLRQVWKHSURSRVHGPHWKRGVHHPVWREHYHU\VWDEOH
DQG WKH RSWLPDO VROXWLRQ LV REWDLQHG DIWHU D VPDOO QXPEHU RI LWHUDWLRQV 7KXV WKH
FRQYHUJHQFHRIWKHPHWKRGLVYHU\JRRG

)LJXUH 1XPHULFDO OD\RXWV IRU VXFFHVVLYH LWHUDWLRQV IRU WRSKDOI SODQH GRPDLQ DQG
JURXQG VWUXFWXUH [[ ZLWK SRWHQWLDO EDUV 1XPHULFDO YROXPH
9QXP 3/V7ZLWKUHODWLYHHUURURI«
([DPSOH±IURP0F&RQQHO
VWR%ULGJH3UREOHP
,QWKHIROORZLQJZHJLYHQXPHULFDODQGVRPHDQDO\WLFDOVROXWLRQVIRUV\PPHWULFDOO\
ORFDWHG WZR WKUHH VHYHQ DQG SRLQW ORDGV EHWZHHQ VXSSRUWV 7KH SUREOHP
VWDWHPHQWLVVKRZQLQ)LJDQGWKHVROXWLRQVLQ)LJVWR$QDO\WLFDOOD\RXWV
RI)LJVDDQGDZHUHREWDLQHGE\6RNyáDQG/HZLĔVNL6LPLODUVROXWLRQV
EXWIRUGLIIHUHQWVXSSRUWVDQGGLIIHUHQWGRPDLQVZHUHVWXGLHGE\6RNyáDQG/HZLĔVNL
DQG6RNyáDQG5R]YDQ\

)LJXUH1XPHULFDOOD\RXWVIRUVXFFHVVLYHLWHUDWLRQVIRUIXOOSODQHGRPDLQDQGJURXQG
VWUXFWXUH [[ ZLWK SRWHQWLDO EDUV 1XPHULFDO YROXPH
9QXP 3/V7ZLWKUHODWLYHHUURU
IHDVLEOHGRPDLQWKHKDOISODQH
3 3
G / G / G G
/

)LJXUH7KHIRXUIRUFHVSUREOHPVHWWLQJ

)LJXUH([DFWDDQGQXPHULFDOEOD\RXWVIRUSUREOHPRI)LJDQGG/

E

)LJXUH([DFWDDQGQXPHULFDOEOD\RXWVIRUSUREOHPRI)LJDQGG/

)LJXUH7KH0F&RQQHOSUREOHPWUDQVPLWWKHJLYHQYHUWLFDODQGXQLIRUPO\GLVWULEXWHG
IRUFHVWRWZRIL[HGKLQJHVWKHIHDVLEOHGRPDLQLVWKHXSSHUKDOISODQH1XPHULFDOVROXWLRQV
IRUWKHFDVHRIWKUHHIRUFHVDVHYHQIRUFHVEDQGIRUFHVF
5HIHUHQFHVDUHOLVWHGDIWHUWKHODVWOHFWXUHRIWKHILUVWDXWKRU

$%ULHI5HYLHZRI1XPHULFDO0HWKRGVRI6WUXFWXUDO
7RSRORJ\2SWLPL]DWLRQ
*HRUJH,15R]YDQ\

%XGDSHVW+XQJDU\
,QWURGXFWLRQ
,QKLVILUVWIRXUOHFWXUHVWKHDXWKRUGLVFXVVHGH[DFWDQDO\WLFDOVROXWLRQVRIVWUXFWXUDO
WRSRORJ\RSWLPL]DWLRQ0RVWUHDOZRUOGSUREOHPVFDQRQO\EHVROYHGE\QXPHULFDO
GLVFUHWL]HGPHWKRGVZKLFKDUHWKHUHIRUHRIFRQVLGHUDEOHSUDFWLFDOLPSRUWDQFH
7KH%HJLQQLQJVRI1XPHULFDO7RSRORJ\2SWLPL]DWLRQ
7KH &HQWUDO 5ROH RI 7KUHH 'DQLVK 6FLHQWLVWV LQ 3LRQHHULQJ 'LVFUHWL]HG
6WUXFWXUDO7RSRORJ\2SWLPL]DWLRQ
6WUXFWXUDO WRSRORJ\ RSWLPL]DWLRQ ZRXOG QRW KDYH UHDFKHG LWV HPLQHQW VWDWXV LQ
VWUXFWXUDO PHFKDQLFV ZLWKRXW VRPH PLOHVWRQH FRQWULEXWLRQV RI WKUHH RXWVWDQGLQJ
'DQLVKVFLHQWLVWV
D+LVWRULFDFKLHYHPHQWVRI1LHOV2OKRII3URJUHVVLQWRSRORJ\RSWLPL]DWLRQZDV
EURXJKW DERXW E\ DQ LPSRUWDQW GHYHORSPHQW LQ WKH HDUO\ HLJKWLHV $ QHZ FODVV RI
RSWLPDOVWUXFWXUHVZDVGLVFRYHUHGE\&KHQJDQG2OKRII,WZDVIRXQG
E\WKHVHDXWKRUVWKDWDQ)(EDVHGRSWLPDOSODWHGHVLJQGHYHORSVDV\VWHPRIULEV
ZKRVHGHQVLW\LQFUHDVHVZLWKPHVKUHILQHPHQW3UDJHUSRLQWHGRXWLQDOHWWHUVKRUWO\
EHIRUHKLVWUDJLFGHDWKLQWKDWWKHOD\RXWRIWKHVHULEVLVDOPRVWLGHQWLFDOWRWKDW
RIRSWLPDOJULOODJHVHJ5R]YDQ\DQG$GLGDP3UDJHUDFWXDOO\WROGWKHDXWKRU
WKDWWZRLQJHQLRXVILQGLQJVE\2OKRIIPXOWLSOHHLJHQYDOXHVLQFROXPQRSWLPL]DWLRQ
DQGQRQVPRRWKVROXWLRQVLQSODWHRSWLPL]DWLRQZHUHSUREDEO\WKHPRVWLPSRUWDQW
GLVFRYHULHVLQWKH WK FHQWXU\ LQ VWUXFWXUDO RSWLPL]DWLRQ 0XFK RI WKH VXEVHTXHQW
GHYHORSPHQW RI H[DFW WRSRORJ\ RSWLPL]DWLRQ E\ PDWKHPDWLFLDQV ZDV SURPSWHG E\
2OKRII¶VDQG&KHQJ¶VSLRQHHULQJGLVFRYHU\
7KH GHQVHO\ ULEEHG QDWXUH RI SODVWLFDOO\ GHVLJQHG RSWLPDO VROLG SODWHV ZDV LQ
YHVWLJDWHGLQJUHDWHUGHWDLOODWHUE\5R]YDQ\2OKRII&KHQJDQG7D\ORU

DOI 10.1007/978-3-7091-1643-2_5, © CISM, Udine 2014
72 G.I.N. Rozvany
2OKRIIHWDODOVRGHWHUPLQHGWKHRSWLPDOPLFURVWUXFWXUHVDQGRSWLPDOUH

JLRQV IRU ' WRSRORJ\ RSWLPL]DWLRQ SUREOHPV DQG GHULYHG QXPHULFDOO\ KLJK
UHVROXWLRQ'RSWLPDOWRSRORJLHVIRUYDULRXVSUREOHPVVHHHJ)LJLQ/HFWXUH
+HKDVDOVREHHQH[WUHPHO\DFWLYHLQDSSO\LQJWKHPRVWSRSXODU QXPHULFDOWR
SRORJ\ RSWLPL]DWLRQ PHWKRG 6,03 WR D YDULHW\ RI SUREOHPV LQFOXGLQJ
GHVLJQGHSHQGHQW ORDGV HLJHQYDOXH VWDELOLW\ DQG YLEUDWLRQ SUREOHPV DQG SURE
DELOLVWLFWRSRORJ\RSWLPL]DWLRQ
E +LVWRULF DFKLHYHPHQWV RI 0DUWLQ %HQGVRH 7KH ILHOG RI FRPSXWHUDLGHG WR
SRORJ\RSWLPL]DWLRQKDVEHHQFRQFHLYHGDQGLQLWLDWHGE\WKLVRXWVWDQGLQJUHVHDUFKHU
2XWRIPDQ\RIKLVYLWDOFRQWULEXWLRQVHDFKRIWKHIROORZLQJIRXUUHSUHVHQWDUHYR
OXWLRQDU\EUHDNWKURXJKLQWRSRORJ\RSWLPL]DWLRQ
LGHYHORSPHQWRIWKHILUVW)(EDVHGPHWKRGVIRUJHQHUDWLQJRSWLPDOWRSRORJLHV
HJ%HQGVRHDQG.LNXFKL
LLWKHFRQFHSWRI6,03W\SHSRZHUODZPHWKRGV%HQGVRH
LLL GHULYDWLRQ RI GLVFUHWL]HG RSWLPDO WRSRORJLHV ZLWK H[DFW RSWLPDO PLFUR
VWUXFWXUHVHJ-RJ+DEHUDQG%HQGVRH7KHVHFDQEHXVHGDVDOLPLWRI
PDWHULDOHFRQRP\DQGDEDVLVIRUDVVHVVLQJWKHUHODWLYHHFRQRP\RISUDFWLFDO
GHVLJQV
LYDXWKRULQJWKHILUVWFRPSUHKHQVLYHDQGDXWKRULWDWLYHERRNRQVWUXFWXUDOWR
SRORJ\RSWLPL]DWLRQ%HQGVRH
F+LVWRULFDFKLHYHPHQWVRI2OH6LJPXQG $OUHDG\GXULQJKLVH[WHQGHGYLVLWWR
(VVHQ8QLYHUVLW\6LJPXQGFRQWULEXWHGVLJQLILFDQWO\WRWKHGHYHORSPHQWRIWKH6,03
DOJRULWKP+HZDVWKHRQO\UHVHDUFKHUZKRDSSOLHG6,03QRWRQO\WRWUXVVHVEXWDOVR
WRJULOODJHVFRPSDULQJKLVUHVXOWVWRDQDO\WLFDOVROXWLRQV6LJPXQGHWDO+H
DOVRGHWHUPLQHGLPSRUWDQWJHQHUDOSURSHUWLHVRIRSWLPDOWRSRORJLHVIRUVHYHUDOGLV
SODFHPHQWFRQVWUDLQWV
$IWHUKLVUHWXUQWR'HQPDUNLWZDVPRVWO\2OH6LJPXQG¶VDFKLHYHPHQWWKDWWKH
6,03DOJRULWKPLVQRZDFFHSWHGE\WKHVWUXFWXUDOWRSRORJ\FRPPXQLW\WKURXJKRXW
WKHZRUOG7KLVUHPDUNDEOHHIIRUWZDVEURXJKWDERXWE\VXSHUEOHFWXUHVDWFRQJUHVVHV
DQG E\ PLOHVWRQH SXEOLFDWLRQV UDQJLQJ IURP HGXFDWLRQDO DUWLFOHV WKURXJK UHVHDUFK
SDSHUVDERXWFUXFLDOGLVFRYHULHVWRDFRPSUHKHQVLYHWH[WERRNZLWK0DUWLQ%HQGVRH
%HQGVRHDQG6LJPXQGRQWRSRORJ\RSWLPL]DWLRQZKLFKLVJHQHUDOO\UHJDUGHG
DVWKHµELEOH¶RIWKLVILHOG
$ FUXFLDO GLVFRYHU\ E\ 6LJPXQG 'LD] DQG 6LJPXQG 6LJPXQG DQG 3H
WHUVVRQ ZDV WKH IDFW WKDW WKDW WKH IDWDO FKHFNHUERDUG HUURU LV FDXVHG E\
GLVFUHWL]DWLRQHUURUVGXHWRVLPSOHHJIRXUQRGHHOHPHQWVZKLFKFDQEHDYRLGHG
E\XVLQJKLJKHURUGHU)(¶V$QRWKHUYLWDOFRQWULEXWLRQZDVWKHLQWURGXFWLRQRIILO
A Brief Review of Numerical Methods of STO 73
WHULQJLQVWUXFWXUDOWRSRORJ\RSWLPL]DWLRQZKLFKQRWRQO\VXSSUHVVHVWKH FKHFNHU
ERDUGHUURUEXWDOVRDFKLHYHVOHQJWKFRQWUROPLQLPXPPHPEHUVL]H:LWKRXWWKH
ODWWHUWKHRXWSXWWXUQVRXWWREHUDWKHUµQRLV\¶ZLWKVRPHWKLQPHPEHUVEHLQJGLV
FRQWLQXRXVRUORFNHGLQWRFHUWDLQGLUHFWLRQVHJDWDQGGHJUHHV
0RUHUHFHQWUHVHDUFKDFWLYLW\RI2OH6LJPXQGFRQFHUQVWKHDSSOLFDWLRQVRIWKH
6,03 DOJRULWKP WR D JUHDW YDULHW\ RI SUREOHPV LQFOXGLQJ FRPSOLDQW PHFKDQLVPV
PXOWLSK\VLFVVROLGIOXLGLQWHUDFWLRQHWF
%HQGVRH DQG 6LJPXQG DOVR IRXQG PLFURVWUXFWXUHV ZLWK WKH PHFKDQLFDO
SURSHUWLHVXVHGLQWKH6,03DOJRULWKPIRUSHQDOL]LQJLQWHUPHGLDWHGHQVLWLHV)RUWKLV
WKH\XVHG6LJPXQG¶VµLQYHUVHKRPRJHQL]DWLRQ¶FRQFHSW
,PSRUWDQW &RQWULEXWLRQV E\ &KLQHVH 6FLHQWLVWV WR 1XPHULFDO 7RSRORJ\

2SWLPL]DWLRQ
D&RQWULEXWLRQVE\*HQJGRQJ&KHQJ$VPHQWLRQHGHDUOLHUWKHHQWLUHILHOGRI
GLVFUHWL]HG VWUXFWXUDO WRSRORJ\ RSWLPL]DWLRQ ZDV VSDUNHG RII E\ WKH GLVFRYHU\ RI
&KHQJDDQGEDQG&KHQJDQG2OKRIIWKDWVWUXFWXUHVOLNHVROLG
SODWHVDQGVKHOOVGHYHORSDGHQVHV\VWHPRIULEVLQWKHLURSWLPDOVROXWLRQV
*HQJRQJ&KHQJKDVPDGHPDQ\RWKHULPSRUWDQWFRQWULEXWLRQVWRWRSRORJ\RS
WLPL]DWLRQ 7KH PRVW VLJQLILFDQW RI WKHVH ZDV WKH GLVFRYHU\ RI WKH UHDO FDXVHV RI
VLQJXODUWRSRORJLHV&KHQJDQG-LDQJ&KHQJDQGWKHHSVLORQUHOD[DWLRQ
PHWKRGIRUKDQGOLQJWKHVHSUREOHPV&KHQJDQG*XR
E&RQWULEXWLRQVE\0LQJ=KRX2QHRIWKHPDQ\FRQWULEXWLRQVRI0LQJ=KRXWR
VWUXFWXUDO WRSRORJ\ RSWLPL]DWLRQ ZDV WKH LQGHSHQGHQW GHYHORSPHQW RI WKH 6,03
DOJRULWKPV DW (VVHQ XQLYHUVLW\ LQ FROODERUDWLRQ ZLWK WKH DXWKRU VROYLQJ VRPH DG
YDQFHG SUREOHPV DOUHDG\ LQ HJ =KRX DQG 5R]YDQ\ SUHVHQWHG DW D
PHHWLQJLQ$IWHUMRLQLQJ$OWDLULQ=KRXLPSOHPHQWHGWKH6,03PHWKRG
IRUJHQHUDOPXOWLSO\FRQVWUDLQHGRSWLPL]DWLRQSUREOHPVLQWKHZHOONQRZQVRIWZDUH
2SWL6WUXFW )XUWKHUPRUH KH H[WHQGHG WKH PHWKRG WR LQFOXGH D YDULHW\ RI GHVLJQ
FRQGLWLRQVLQFOXGLQJPDQXIDFWXULQJFRQVWUDLQWVDQGVWUHVVFRQVWUDLQWV
6RPH,PSRUWDQW(DUO\5HVHDUFK5HVXOWVE\2WKHUV
,WLVLPSRUWDQWWRPHQWLRQWKDWWKHILUVWDWWHPSWWRREWDLQRSWLPDOWRSRORJLHVRIFRQ
WLQXDE\ILQLWHHOHPHQWFDOFXODWLRQVZDVPDGHE\-RKQ7D\ORU5RVVRZDQG7D\ORU
6LQFH KH GLG QRW XVH SHQDOL]DWLRQ KLV RXWSXWV FRQVLVWHG RI SODWHV KDYLQJ
YDU\LQJWKLFNQHVVZLWKPRVWO\µJUH\¶HOHPHQWV
$6,03OLNHPHWKRGZDVDOVRSURSRVHGLQGHSHQGHQWO\E\0OHMQHNHJ
74 G.I.N. Rozvany
0HWKRGVRI'LVFUHWL]HG6WUXFWXUDO7RSRORJ\2SWLPL]DWLRQ
7KH PRVW SRSXODU WHFKQLTXH LV FXUUHQWO\ WKH 6,03 6ROLG ,VRWURSLF 0DWHULDO RU
0LFURVWUXFWXUHZLWK3HQDOL]DWLRQPHWKRG,Q WKLV DSSURDFK LVRWURSLF JURXQG HOH
PHQWVRIFRQWLQXRXVO\YDU\LQJWKLFNQHVVRUGHQVLW\DUHXVHGEXWLQWHUPHGLDWHYDOXHV
EHWZHHQ]HURDQGWKHSUHDVVLJQHGPD[LPXPYDOXHDUHSHQDOL]HG,QWKHµFRQWLQXD
WLRQ¶YHUVLRQRI6,03WKHSHQDOW\IDFWRULVLQFUHDVHGSURJUHVVLYHO\IURPXQLW\WRLWV
KLJKHVWYDOXHGXULQJWKHLWHUDWLRQSURFHGXUH0RVWFRPPHUFLDOVRIWZDUHLVHPSOR\LQJ
WKH6,03PHWKRGDWSUHVHQWDQGLWFDQEHXVHGIRUFRPELQDWLRQVRIDJUHDWYDULHW\RI
FRQVWUDLQWV
$QRWKHU HDUO\ WRSRORJ\ RSWLPL]DWLRQ WHFKQLTXH ZDV WKH KRPRJHQL]DWLRQ
PHWKRGLQZKLFKLQHDFKHOHPHQWWKHGLPHQVLRQVDQGRULHQWDWLRQRIVTXDUHRUUHF
WDQJXODU KROHV ZHUH RSWLPL]HG DQG VXFK HOHPHQWV ZHUH KRPRJHQL]HG IRU WKH
VXEVHTXHQW)(DQDO\VLVLQHDFKLWHUDWLRQ
2WKHULQYHVWLJDWRUVXVHGHDUOLHURSWLPDOPLFURVWUXFWXUHVLQHDFKHOHPHQWDQG
VXEVHTXHQWO\SHQDOL]HGLQWHUPHGLDWHGHQVLWLHV
+RZHYHUFXUUHQWO\6,03LVXVHGLQSUHIHUHQFHWRWKHODWWHUWZRPHWKRGVZKLFK
UHTXLUHDPXFKKLJKHUFRPSXWDWLRQDOHIIRUW
0RUHRYHU )OHXU\ DQG %HFNHUV HJ %HFNHUV XVHG GLVFUHWH GXDO SUR
JUDPPLQJVXFFHVVIXOO\IRUWRSRORJ\RSWLPL]DWLRQ
0RUH UHFHQW PHWKRGV DUH WKH WRSRORJLFDO GHULYDWLYH DQG OHYHO VHW PHWKRGV
ZKLFKKDYHYHU\PXFKSURPLVHEXWKDYHQRWTXLWHUHDFKHGDV\HWWKHVWDJHRILQ
GXVWULDODSSOLFDWLRQV7KHVHWHFKQLTXHVZLOOEHGLVFXVVHGE\RWKHUOHFWXUHUVLQWKLV
FRXUVH
5DWKHUFRQWURYHUVLDODUHWKHFRPSOHWHO\KHXULVWLF¶VXGGHQGHDWK¶RU¶KDUGNLOO¶
PHWKRGVLQZKLFKHOHPHQWVZLWKDORZYDOXHRIVRPHUHVSRQVHIXQFWLRQDUHWDNHQRXW
LQHDFKLWHUDWLRQ7KHPRVWSXEOLFL]HGRIWKHVHPHWKRGVLVWKHµ(62¶(YROXWLRQDU\
6WUXFWXUDO 2SWLPL]DWLRQ PHWKRG ZKLFK LV QRW XVHG FXUUHQWO\ LQ FRPPHUFLDO VRIW
ZDUH,WZDVVKRZQE\0LQJ=KRXDQGWKHDXWKRU=KRXDQG5R]YDQ\WKDW(62
PD\ \LHOG KLJKO\ QRQRSWLPDO VROXWLRQV $ FRPSUHKHQVLYH UHYLHZ RI HVWDEOLVKHG
PHWKRGVLQVWUXFWXUDOWRSRORJ\RSWLPL]DWLRQZDVSUHVHQWHGE\WKHDXWKRU5R]YDQ\

7KHDERYHPHWKRGVDUHDOOEDVLFDOO\¶JUDGLHQWPHWKRGV¶EHFDXVHWKHUHGHVLJQRI
HDFKHOHPHQWLVEDVHGRQVRPHJUDGLHQWYDOXHRUDWOHDVWWKHYDOXHRIWKHUHVSRQVH
SDUDPHWHUIRUWKHFRQVLGHUHGHOHPHQWDVLQ(62
2QWKHRWKHUKDQG¶QRQJUDGLHQWPHWKRGV¶XVHVRPHRWKHULQIRUPDWLRQIRUWKH
UHGHVLJQ7KH\DUHRIWHQFODLPHGWREH¶JOREDOPHWKRGV¶VXSSRVHGO\JXDUDQWHHLQJ
JOREDOFRQYHUJHQFHEXWWKLVFODLPLVXVXDOO\XQIRXQGHG2IFRXUVHWKHUHH[LVWWUXO\
JOREDO PHWKRGV VXFK DV HQXPHUDWLRQ PHWKRGV ZLWK HIILFLHQW UHGHVLJQ WHFKQLTXHV
EUDQFK DQG ERXQG RU EUDQFK DQG FXW PHWKRGV VHH HJ WKH SDSHU E\ 6WROSH DQG
%HQGVRHEXWPRVWQRQJUDGLHQWWHFKQLTXHVDUHEDVHGRQKHXULVWLFVDQGWKH\
RIWHQ XVH IDQF\ PHWDSKRUV HJ 'ROSKLQ (FKRORFDWLRQ $UWLILFLDO %HH &RORQ\
0DJQHWLF &KDUJHG 6\VWHP 6HDUFK $QW &RORQ\ 3DUWLFOH 6ZDUP 7DEX 6HDUFK
6KXIIOHG)URJ/HDSLQJ&XFNRR6HDUFK6LPXODWHG$QQHDOLQJ,PSHULDOLVW&RPSHWL
WLRQ $OJRULWKP +DUPRQ\ 6HDUFK 'LIIHUHQWLDO (YROXWLRQ HWF 6RPH RI WKHVH
PHWKRGVDUHDOVRWHUPHGµQDWXUHLQVSLUHG¶µELRPLPLFU\¶µPHWDKHXULVWLF¶RUµSRSX
ODWLRQW\SH¶DOJRULWKPVDQGXVXDOO\LQFOXGHVRPHUDQGRPSURFHVVHV7KH\DUHRIWHQ
YHULILHGRQO\E\FRPSDULQJWKHPZLWKRWKHUQRQJUDGLHQWPHWKRGVXVLQJSDUWLFXODU
EHQFKPDUN H[DPSOHV 7KH EHVW HVWDEOLVKHG QDWXUHLQVSLUHG PHWKRG LV WKH *HQHWLF
$OJRULWKP*$PHWKRG
6LJPXQG¶V 0RVW 6LJQLILFDQW )RUXP $UWLFOH RQ 1RQJUDGLHQW

0HWKRGVLQ7RSRORJ\2SWLPL]DWLRQ
6LJPXQGSRLQWHGRXWWKDWJUDGLHQWW\SHWRSRORJ\DOJRULWKPVFDQVROYHSURE
OHPVZLWKXSWRPLOOLRQVRIYDULDEOHVXVLQJDIHZKXQGUHGDQG VRPHFRPPHUFLDO
FRGHV HYHQ OHVV WKDQ IXQFWLRQ HYDOXDWLRQV 2Q WKH RWKHU KDQG QRQJUDGLHQW
PHWKRGV QHHG W\SLFDOO\ RYHU IXQFWLRQ HYDOXDWLRQV HYHQ IRU YHU\ FRDUVHO\
GLVFUHWL]HGSUREOHPV
6LJPXQG FRQYLQFLQJO\ GLVPLVVHV DUJXPHQWV IRU QRQJUDGLHQW PHWKRGV
WKDWLWKH\JHWEHWWHURSWLPDEHLQJJOREDOVHDUFKPHWKRGVLLWKH\SURYLGHGLVFUHWH
GHVLJQV ZKLFK DUH EHWWHU WKDQ JUH\VFDOH GHVLJQV LLL WKH\ DUH HDV\ WR LPSOHPHQW
EHFDXVHWKH\QHHGQRJUDGLHQWVDQGLYWKHLUDGYDQWDJHLVWKDWWKH\UXQHDVLO\RQ
SDUDOOHO FRPSXWHUV 6LJPXQG DOVR VXPPDUL]HV WKH GLVDGYDQWDJHV RI
QRQJUDGLHQWPHWKRGVLQFOXGLQJWKHIDFWWKDWWKH\FDQQRWEHXVHGIRUPHWKRGYHUL
ILFDWLRQ E\ JUDGXDOO\ UHILQLQJ WKH ILQLWH HOHPHQW PHVK IRU FRPSDULVRQ ZLWK WKH
DQDO\WLFDO VROXWLRQ VHH /HFWXUH +RZHYHU 6LJPXQG UHPDUNV WKDW QRQJUDGLHQW
PHWKRGVPD\EHDSSURSULDWHIRUVRPHYHU\VSHFLDOSUREOHPVZLWKPDQ\ORFDOPLQLPD
RUGLVMRLQWGHVLJQGRPDLQV
%DVHG RQ WKH DERYH FRQFOXVLRQV 6LJPXQG UHFRPPHQGV WKDW SDSHUV RQ
QRQJUDGLHQWPHWKRGVVKRXOGRQO\EHSXEOLVKHGLIWKH\LDUHVKRZQWREHDEOHWR
KDQGOHDWOHDVWGHVLJQYDULDEOHVLLVKRZDSSOLFDWLRQVZKLFKWKH\FDQKDQGOH
EHWWHU WKDQ VWDQGDUG JUDGLHQW PHWKRGV LLL GR QRW LQFOXGH REYLRXVO\ QRQRSWLPDO
VROXWLRQV DQG LY DUH VKRZQ WR EH DEOH WR KDQGOH PRUH GLIILFXOW QRQVHOIDGMRLQW
SUREOHPVQRWRQO\FRPSOLDQFHH[DPSOHV
,WLVWREHUHPDUNHGWKDWWZRHGLWRUVRI6020LQJ=KRXDQGWKHDXWKRUIXOO\
VXSSRUWHG6LJPXQG¶VLGHDRISXEOLVKLQJDIRUXPDUWLFOHRQQRQJUDGLHQWPHWKRGV/H
5LFKHDQG+DIWNDKDYHSXEOLVKHGDVRUWRIDµ5HSO\¶WR6LJPXQG¶V
DUWLFOH LQ ZKLFK WKH\ XVH IRU QRQJUDGLHQW PHWKRGV WKH WHUP µJOREDO RSWLPL]DWLRQ
²²²²²²²²²²

7KHDXWKRUPD\DGGWKDWIRUVRPHSUREOHPVVHH/HFWXUHJUDGLHQWPHWKRGVPD\KDQGOH
PRUHWKDQDELOOLRQYDULDEOHV
76 G.I.N. Rozvany
PHWKRGV¶DOWKRXJKWKH\FRQFHGHWKDWWKHRUHWLFDOJOREDOFRQYHUJHQFHSURRIVIRUWKHVH
PHWKRGVKDYHOLWWOHSUDFWLFDOVLJQLILFDQFH$VHGLWRULDOJXLGHOLQHVWKH\SURSRVHWKDW
VXFKSDSHUVVKRXOGFRQWDLQDSVHXGRFRGHRUPDWKHPDWLFDOIRUPXODWLRQGHYRLGRID
PHWDSKRU DQG WKH GLIIHUHQFH IURP H[LVWLQJ PHWDSKRULFDO RSWLPL]DWLRQ PHWKRGV
VKRXOGEHFOHDUO\H[SODLQHG7HVWSUREOHPVVKRXOGLQFOXGHRQHVWKDWFDQQRWEHVROYHG
PRUHHIILFLHQWO\E\UHVWDUWHGJUDGLHQWPHWKRGV)LQDOO\WKH\OLVWµJOREDO¶RSWLPL]DWLRQ
SDSHUV ZLWK RYHU *RRJOH FLWDWLRQV FODLPLQJ WKDW LW LV µWKH EHVW PHDVXUH ZH
SUHVHQWO\KDYH¶SUHVXPDEO\IRUDVVHVVLQJWKHLPSRUWDQFHRIDSDSHU
&RPSXWDWLRQDO'LIILFXOWLHVLQ1XPHULFDO7RSRORJ\2SWLPL]D
WLRQ
&KHFNHUERDUG(UURU
7KLVLVDVHYHUHGLVFUHWL]DWLRQHUURULQWRSRORJ\RSWLPL]DWLRQZKLFKZDVILUVWPHQ
WLRQHGE\%HQGVRH'LD]DQG.LNXFKL,WZDVH[DPLQHGLQJUHDWHUGHWDLOE\
'LD]DQG6LJPXQGDQG6LJPXQGDQG3HWHUVVRQDQGDOVRUHYLHZHGLQ
%HQGVRH¶VILUVWERRN%HQGVRH7KLVHUURULVFDXVHGE\WKHRYHUHVWLPDWLRQRI
WKHVWLIIQHVVRIVLPSOHILQLWHHOHPHQWVLIWZRVROLGJURXQGHOHPHQWVRQO\KDYHFRUQHU
FRQWDFW&KHFNHUERDUGHUURUFDQEHDYRLGHGE\
DXVLQJDKLJKHURUGHUILQLWHHOHPHQWLQHDFKJURXQGHOHPHQW'LD]DQG6LJ
PXQG6LJPXQGDQG3HWHUVVRQRUPRUHWKDQRQHILQLWHHOHPHQWSHU
JURXQGHOHPHQW=KRXDQG5R]YDQ\
EILOWHULQJ6LJPXQG6LJPXQGDQG3HWHUVVRQ
FSHULPHWHUFRQWURO+DEHUDQG%HQGVRH
GOHQJWKPHPEHUVL]HFRQWURO3RXOVHQ
HFRQVWUDLQWVRQFKHFNHUERDUGLQJ3RXOVHQ
ISHQDOL]DWLRQRIFKHFNHUERDUGLQJ3RPH]DQVNLHWDO
JFRQGLWLRQDOVLGHFRQVWUDLQWV=KRXHWDO
KVSDWLDOVORSHFRQVWUDLQWV3HWHUVVRQDQG6LJPXQGRU
LKH[DJRQDOHOHPHQWV7DOLVFKLHWDO
6LQJXODU2SWLPD
6LQJXODURSWLPDDUHRXWVLGHWKHPDLQSDUWRIWKHIHDVLEOHGHVLJQGRPDLQDQGWKHUHIRUH
FDQQRWEHUHDFKHGE\LWHUDWLYHFRPSXWDWLRQDOPHWKRGV7KH\ZHUHILUVWGLVFRYHUHGE\
6YHGDQGWKHWHUPµVLQJXODU¶FRPHVIURP.LUVFK7KHUHDOQDWXUHRI
VLQJXODURSWLPDZDVH[SODLQHGSDUWO\E\&KHQJDQG&KHQJDQG-LDQJ
DQGDWDGHHSHUDQDO\WLFDOOHYHOE\WKHDXWKRUDQG%LUNHU5R]YDQ\DQG%LUNHU
%RWK &KHQJ &KHQJ DQG *XR DQG WKH DXWKRU KDYH VXJJHVWHG PHWKRGV IRU
KDQGOLQJVLQJXODURSWLPDIRUDUHYLHZRIWKHODWWHUVHHWH[WVE\5R]YDQ\%HQGVRH
DQG.LUVFKRU5R]YDQ\
/RFDO2SWLPD
/RFDORSWLPDPD\EHGXHWRQRQFRQYH[LW\RIWKHXQGHUO\LQJSUREOHPRUWKH\PD\
DULVH DV WKH UHVXOW RI SHQDOL]DWLRQ LQ WKH 6,03 PHWKRG 7KH ODWWHU FDQ EH ODUJHO\
DYRLGHG E\ XVLQJ WKH VRFDOOHG µFRQWLQXDWLRQ PHWKRG¶ LQ ZKLFK RQH VWDUWV RII WKH
LWHUDWLRQVZLWKQRSHQDOW\DQGWKHQWKHSHQDOW\IDFWRULVLQFUHDVHGSURJUHVVLYHO\LQ
VXEVHTXHQWLWHUDWLRQV7KHLGHDRIWKLVDSSURDFKZDVILUVWPHQWLRQHGE\ 5R]YDQ\
=KRXDQG6LJPXQGSS
/RFDODQG*OREDO%XFNOLQJ&RQVWUDLQWV
'LIILFXOWLHV ZLWK ORFDO DQG JOREDO EXFNOLQJ FRQVWUDLQWV ZHUH SRLQWHG RXW E\ =KRX
DQGWKHDXWKRUUHVSHFWLYHO\
0RUH 5HFHQW 5HVHDUFK E\ 6LJPXQG DQG &RDXWKRUV LQ 'LV
FUHWL]HG7RSRORJ\2SWLPL]DWLRQ
6LQFHUHVHDUFKLQWKLVILHOGE\RWKHUUHVHDUFKJURXSVLVFRYHUHGE\VRPHRIWKHRWKHU
OHFWXUHUVLQWKLVVHFWLRQZHVXPPDUL]HYHU\EULHIO\WKHPRVWVLJQLILFDQWUHFHQWUH
VHDUFKGHYHORSPHQWVE\6LJPXQGDQGKLVFRDXWKRUV7KHVHKDYHLQFOXGHGDPRQJ
RWKHUVWKHIROORZLQJWRSLFV
0XOWLSK\VLFVDFWXDWRUV6LJPXQGODUJHGLVSODFHPHQWFRPSOLDQWPHFKD
QLVPV3HGHUVHQ%XKODQG6LJPXQGSKRWRQLFFU\VWDOVWUXFWXUHV%RUHOHWDO
-HQVHQDQG6LJPXQG6LJPXQGDQG+RXJDUG3KRQRQLFEDQGJDS
PDWHULDO 6LJPXQG DQG -HQVHQ DFRXVWLFVWUXFWXUDO LQWHUDFWLRQ <RXQ HW DO
SUHVVXUHORDGSUREOHPV6LJPXQGDQG&ODXVHQDQGPRUSKRORJ\EDVHG
EODFNDQGZKLWHILOWHULQJ6LJPXQG
7KHODWHVWUHVHDUFKE\6LJPXQGDQGFRDXWKRUVLQFOXGHUREXVWGHVLJQZLWKUHVSHFW
WRVSDWLDOO\YDU\LQJPDQXIDFWXULQJYDULDWLRQV6FKHYHQHOVHWDOH[WHQVLRQVRI
7RS2SWWRPHWDOOLF0LFURZDYHDSSOLFDWLRQVLQ'$DJHHWDOUHYLHZDQG
UREXVW GHVLJQ ZLWK UHVSHFW WR XQGHURYHU HWFKLQJ :DQJ HW DO VHH )LJ
UHYLHZ RQ DFWLYLWLHV LQ 3KRWRQLFV -HQVHQ DQG 6LJPXQG GHVLJQ RI RSWLFDO
FORDNV$QGNMDHUDQG6LJPXQGDQOLQHLPSURYHG0DWODEFRGHIRUWRSRORJ\
RSWLPL]DWLRQ$QGUHDVVHQHWDOFRQFUHWHUHLQIRUFHPHQWGHVLJQZLWK7RS2SW
DQGH[SODQDWLRQRIWKHVHQVLWLYLW\ILOWHU6LJPXQGDQG0DXWH
78 G.I.N. Rozvany

)LJXUH7RSRORJ\RSWLPL]DWLRQRIKHDWFRQGXFWLRQXVLQJWKUHVKROGSURMHFWLRQDIWHU
:DQJHWDO
$KLJKO\LQWHUHVWLQJGHYHORSPHQWE\WKH6LJPXQGJURXSLVDQLQWHUDFWLYHWRSRO
RJ\ RSWLPL]DWLRQ SURJUDP ZKLFK FDQ DOVR EH XVHG RQ L26 DQG $QGURLG GHYLFHV
$DJHHWDO$QRSWLPDOWRSRORJ\JHQHUDWHGE\WKLVSURJUDPFDQEHVHHQLQ
)LJ

)LJXUH2SWLPDOWRSRORJ\JHQHUDWHGE\DQLQWHUDFWLYHRSWLPL]DWLRQSURJUDPDIWHU$DJH
HWDOWKLV7RS2SW$SSFDQEHGRZQORDGHGRQL26RU$QGURLGGHYLFHV
$FNQRZOHGJHPHQWV 7KH ILUVW DXWKRU RI WKHVH ILYH OHFWXUHV ZRXOG OLNH WR WKDQN
7RPDV] 6RNyá IRU KLV H[WHQVLYH KHOS LQ WKH SUHSDUDWLRQ RI WKH PDQXVFULRSWV 7KH
SUHVHQWVWXG\ZDVVXSSRUWHGE\27.$*UDQWQXPEHU
5HIHUHQFHVIRU/HFWXUHVE\*,15R]YDQ\DQG&RDXWKRUV
$DJH1-RUJHQVHQ01$QGUHDVHQ&66LJPXQG2,QWHUDFWLYHWRSRORJ\RSWLPL]D
WLRQRQKDQGKHOGGHYLFHV6WUXFW0XOWLGLVF2SWLP
$DJH10RUWHQVHQ1$6LJPXQG27RSRORJ\RSWLPL]DWLRQRIPHWDOOLFGHYLFHVIRU
PLFURZDYHDSSOLFDWLRQV,QW-1XPHU0HWK(QJ
$FKW]LJHU:7RSRORJ\RSWLPL]DWLRQRIGLVFUHWHVWUXFWXUHVDQLQWURGXFWLRQLQYLHZ
RIFRPSXWDWLRQDODQGQRQVPRRWKDVSHFWV,Q5R]YDQ\*,1(G7RSRORJ\RSWLPL]D
WLRQ LQ VWUXFWXUDO PHFKDQLFV &,60 &RXUVH 1R KHOG LQ 8GLQH SS
6SULQJHU9LHQQD
$OODLUH*.RKQ592SWLPDOGHVLJQIRUPLQLPXPZHLJKWDQGFRPSOLDQFHLQSODQH
VWUHVVXVLQJH[WUHPDOPLFURVWUXFWXUHV(XU-0HFK±$6ROLGV
$PLU26LJPXQG25HLQIRUFHPHQWOD\RXWGHVLJQIRUFRQFUHWHVWUXFWXUHVEDVHGRQ
FRQWLQXXP GDPDJH DQG WUXVV WRSRORJ\ RSWLPL]DWLRQ 6WUXFW 0XOWLGLVF 2SWLP

$QGUHDVVHQ ( &ODXVHQ $ 6FKHYHQHOV 0 /D]DURY %6 6LJPXQG 2 (IILFLHQW WR
SRORJ\ RSWLPL]DWLRQ LQ 0$7/$% XVLQJ OLQHV RI FRGH 6WUXFW 0XOWLGLVF 2SWLP

$QGNMDHU - 6LJPXQG 2 7RSRORJ\ RSWLPL]HG ORZFRQWUDVW DOOGLHOHFWULF RSWLFDO
FORDN$SSO3K\VLFVOHWWHUV
%DUWD-2QWKHPLQLPXPZHLJKWRIFHUWDLQUHGXQGDQWVWUXFWXUHV$FWD7HFKQ$FDG
6FL+XQJ
%HFNHUV 0 7RSRORJ\ RSWLPL]DWLRQ XVLQJ D GXDO PHWKRG ZLWK GLVFUHWH YDULDEOHV
6WUXFW2SWLP
%HQGVRH032SWLPDOVKDSHGHVLJQDVDPDWHULDOGLVWULEXWLRQSUREOHP6WUXFW2SWLP

%HQGVRH 03 2SWLPL]DWLRQ RI VWUXFWXUDO WRSRORJ\ VKDSH DQG PDWHULDO 6SULQJHU
%HUOLQ
%HQGVRH03'LD]$.LNXFKL17RSRORJ\DQGJHQHUDOL]HGOD\RXWRSWLPL]DWLRQRI
HODVWLFVWUXFWXUHV,Q%HQGVRH030RWD6RDUHV&$(GV7RSRORJ\'HVLJQRI6WUXF
WXUHV1$72$5:6HVLPEUD3RUWXJDOSS.OXZHU'RUGUHFKW
%HQGVRH03+DEHU5%7KH0LFKHOOOD\RXWSUREOHPDVDORZYROXPHIUDFWLRQOLPLW
RI WKH SHUIRUDWHG SODWH WRSRORJ\ RSWLPL]DWLRQ SUREOHP DQ DV\PSWRWLF VWXG\ 6WUXFW
2SWLP
%HQGVRH03.LNXFKL1*HQHUDWLQJRSWLPDOWRSRORJLHVLQVWUXFWXUDOGHVLJQXVLQJD
KRPRJHQL]DWLRQPHWKRG&RPS0HWK$SSO0HFK(QJUJ
%HQGVRH036LJPXQG20DWHULDOLQWHUSRODWLRQVFKHPHVLQWRSRORJ\RSWLPL]DWLRQ
$UFK$SSO0HFK
%HQGVRH 03 6LJPXQG 2 7RSRORJ\ RSWLPL]DWLRQ WKHRU\ PHWKRGV DQG DSSOLFD
WLRQV6SULQJHU%HUOLQ
80 G.I.N. Rozvany
%LUNHU7/HZLQVNL75R]YDQ\*,1$ZHOOSRVHGQRQVHOIDGMRLQWOD\RXWSUREOHP

OHDVWZHLJKWSODQHWUXVVIRURQHORDGFRQGLWLRQDQGWZRGLVSODFHPHQWFRQVWUDLQWV6WUXFW
0XOWLGLVF2SW
%RUHO 3, +DUSRWK )UDGVHQ /+ .ULVWLHQVHQ 0 6KL 3 -HQVHQ -6 6LJPXQG 2
7RSRORJ\RSWLPL]DWLRQDQGIDEULFDWLRQRISKRWRQLFFU\VWDOVWUXFWXUHV2SWLFV([SUHVV

&KHQJ.72QQRQVPRRWKQHVVLQRSWLPDOGHVLJQRIVROLGHODVWLFSODWHV,QW-6ROLGV
6WUXFW
&KHQJ.72QVRPHQHZRSWLPDOGHVLJQIRUPXODWLRQVIRUSODWHV,Q+DXJ(-&HD-
(GV2SWLPL]DWLRQRI'LVWULEXWHG3DUDPHWHU3UREOHPV3URFHHGLQJVRI1$72$6,
,RZD&LW\6LMWKRIIDQG1RRUGKRII$OSKHQDDQGHU5LMQ7KH1HWKHUODQGV
&KHQJ.76RPHDVSHFWVRIWUXVVWRSRORJ\RSWLPL]DWLRQ6WUXFW2SWLP
&KHQJ.7*XR; HUHOD[HGDSSURDFKLQVWUXFWXUDOWRSRORJ\RSWLPL]DWLRQ6WUXFW
2SWLP
&KHQJ .7 -LDQJ = 6WXG\ RI WRSRORJ\ RSWLPL]DWLRQ ZLWK VWUHVV FRQVWUDLQWV (QJ
2SWLP
&KHQJ.72OKRII1$QLQYHVWLJDWLRQFRQFHUQLQJRSWLPDOGHVLJQRIVROLGHODVWLF
SODWHV,QW-6ROLGV6WUXFW
&KHQJ.72OKRII15HJXODUL]HGIRUPXODWLRQIRURSWLPDOGHVLJQRID[LV\PPHWULF
SODWHV,QW-6ROLGV6WUXFW
&R[+/7KHWKHRU\RIGHVLJQ5HS$HURQDXW5HV&RXQFLO1R
&R[+/7KHGHVLJQRIVWUXFWXUHVIRUOHDVWZHLJKW3HUJDPRQ3UHVV2[IRUG
'HZKXUVW 3 6ULWKRQJFKDL 6 $Q LQYHVWLJDWLRQ RI PLQLPXPZHLJKW GXDO PDWHULDO
V\PPHWULFDOO\ORDGHGZKHHOVDQGWRUVLRQDUPV-$SSO0HFK$60(
'LD]$56LJPXQG2&KHFNHUERDUGSDWWHUQVLQOD\RXWRSWLPL]DWLRQ6WUXFW2SWLP

'UXFNHU'&*UHHQEHUJ-+3UDJHU:7KHVDIHW\IDFWRURIDQHODVWLFSODVWLFERG\LQ
SODQHVWUHVV-$SSO0HFK
*LOEHUW 0 7\DV $ /D\RXW RSWLPL]DWLRQ RI ODUJHVFDOH SLQMRLQWHG IUDPHV (QJ
&RPSXW±
*UDF]\NRZVNL&/HZLQVNL7D0LFKHOOFDQWLOHYHUVFRQVWUXFWHGZLWKLQWUDSH]LRGDO
GRPDLQV3DUW,*HRPHWU\RI+HQFN\QHWV6WUXFW0XOWLGLVF2SW
*UDF]\NRZVNL&/HZLQVNL7E0LFKHOOFDQWLOHYHUVFRQVWUXFWHGZLWKLQWUDSH]LRGDO
GRPDLQV3DUW,,9LUWXDOGLVSODFHPHQWILHOGV6WUXFW0XOWLGLVF2SW
*UDF]\NRZVNL&/HZLQVNL7D0LFKHOOFDQWLOHYHUVFRQVWUXFWHGZLWKLQWUDSH]LRGDO
GRPDLQV3DUW,,,)RUFHILHOGV6WUXFW0XOWLGLVF2SW
*UDF]\NRZVNL&/HZLQVNL7E0LFKHOOFDQWLOHYHUVFRQVWUXFWHGZLWKLQWUDSH]LRGDO
GRPDLQV3DUW,9&RPSOHWHH[DFWVROXWLRQVRIVHOHFWHGRSWLPDOGHVLJQVDQGWKHLUDS
SUR[LPDWLRQVE\WUXVVHVRIILQLWHQXPEHURIMRLQWV6WUXFW0XOWLGLVF2SW
*UDF]\NRZVNL & /HZLQVNL 7 0LFKHOO FDQWLOHYHUV FRQVWUXFWHG ZLWKLQ D KDOIVWULS
7DEXODWLRQRIVHOHFWHGEHQFKPDUNUHVXOWV6WUXFW0XOWLGLVF2SW
+DEHU5%%HQGVRH03$QHZDSSURDFKWRYDULDEOHWRSRORJ\VKDSHGHVLJQXVLQJ
FRQVWUDLQWRQWKHSHULPHWHU6WUXFW2SW
+HPS:62SWLPXPVWUXFWXUHV&ODUHQGRQ2[IRUG
-HQVHQ -6 6LJPXQG 2 6\VWHPDWLF GHVLJQ RI SKRQRQLF FU\VWDO VWUXFWXUHV XVLQJ
WRSRORJ\ RSWLPL]DWLRQ /RZORVV ZDYHJXLGH EHQGV $SSO 3K\VLFV /HWWHUV

-HQVHQ-66LJPXQG27RSRORJ\RSWLPL]DWLRQIRUQDQRSKRWRQLFV/DVHU3KRWRQLFV
5HY
-RJ&+DEHU5%HQGVRH03$GLVSODFHPHQWEDVHGWRSRORJ\RSWLPL]DWLRQPHWKRG
ZLWKVHOIDGDSWLYHOD\HUHGPDWHULDOV,Q%HQGVRH030RWD6RDUHV&$HGV7RSRORJ\
'HVLJQRIVWUXFWXUHVSS.OXZHU'RUGUHFKW
.iURO\L * 5R]YDQ\ *,1 ([DFW DQDO\WLFDO VROXWLRQV IRU QRQVHOIDGMRLQW YDULDEOH
WRSRORJ\ VKDSH RSWLPL]DWLRQ SHUIRUDWHG FDQWLOHYHU SODWHV LQ SODQH VWUHVV VXEMHFW WR
GLVSODFHPHQWFRQVWUDLQWV6WUXFW0XOWLGLVF2SWLP
.LUVFK82QVLQJXODUWRSRORJLHVLQVWUXFWXUDOGHVLJQ6WUXFW2SWLP
.RODQHN./HZLQVNL7&LUFXODUDQGDQQXDOWZRSKDVHSODWHVRIPLQLPDOFRPSOL
DQFH&RPS$VVLVWHG0HFK(QJ6FL
OH5LFKH5+DIWND572QJOREDORSWLPL]DWLRQDUWLFOHVLQ6026WUXFW 0XOWLGLVF
2SWLP
/pY\0/DVWDWLTXHJUDSKLTXHHWVHVDSSOLFDWLRQVDX[FRQVWUXFWLRQV
/HZLQVNL75R]YDQ\*,1([DFWDQDO\WLFDOVROXWLRQVIRUVRPHSRSXODUEHQFKPDUN
SUREOHPVLQWRSRORJ\RSWLPL]DWLRQ,,WKUHHVLGHGSRO\JRQDOVXSSRUWV6WUXFW0XOWLGLVF
2SWLP
/HZLQVNL75R]YDQ\*,1D([DFWDQDO\WLFDOVROXWLRQVIRUVRPHSRSXODUEHQFKPDUN
SUREOHPVLQWRSRORJ\RSWLPL]DWLRQ,,,/VKDSHG GRPDLQV6WUXFW 0XOWLGLVF 2SWLP

/HZLQVNL75R]YDQ\*,1E$QDO\WLFDOEHQFKPDUNVIRUWRSRORJ\RSWLPL]DWLRQ,9
VTXDUHVKDSHGOLQHVXSSRUW6WUXFW0XOWLGLVF2SWLP
/HZLĔVNL75R]YDQ\*,16RNyá7%RáERWRZVNL.([DFWDQDO\WLFDOVROXWLRQVIRU
VRPH SRSXODU EHQFKPDUN SUREOHPV LQ WRSRORJ\ RSWLPL]DWLRQ ,,, /VKDSHG GRPDLQV
UHYLVLWHG6WUXFW0XOWLGLVF2SWLPSXEOLVKHGRQOLQH-DQXDU\
/HZLQVNL7=KRX05R]YDQ\*,1D([WHQGHGH[DFWVROXWLRQVIRUOHDVWZHLJKWWUXVV
OD\RXWV 3DUW , FDQWLOHYHUV ZLWK D KRUL]RQWDO D[LV RI V\PPHWU\ ,QW - 0HFK 6FL

/HZLQVNL7=KRX05R]YDQ\*,1E([WHQGHGH[DFWVROXWLRQVIRUOHDVWZHLJKWWUXVV
OD\RXWV3DUW,,XQV\PPHWULFFDQWLOHYHUV,QW-0HFK6FL
/XULH.$'LUHFWUHOD[DWLRQRIRSWLPDOOD\RXWSUREOHPVIRUSODWHV-27$
/XULH .$ D 2Q GLUHFW UHOD[DWLRQ RI RSWLPDO PDWHULDO GHVLJQ SUREOHPV IRU SODWHV
:66,$$
/XULH.$E$QRSWLPDOSODWH,Q2OKRII15R]YDQ\*,1(GV3URF:&602±
SS3HUJDPRQ2[IRUG
0DXWH . )UDQJRSRO ' 5HOLDELOLW\EDVHG 'HVLJQ RI 0(06 0HFKDQLVPV E\ 7R
SRORJ\2SWLPL]DWLRQ&RPS6WUXFW
0F.HRZQ -- $ QRWH RQ WKH PD[LPXP QXPEHU DQG GHQVLW\ RI GLVWULEXWLRQ RI
PHPEHUVLQHODVWLFVWUXFWXUHVRIPLQLPXPZHLJKWXQGHUPXOWLSOHORDGLQJFRQGLWLRQV,QW
-6ROLGV6WUXFW
82 G.I.N. Rozvany
0F.HRZQ--8SSHUOLPLWVRQWKHQXPEHURIHOHPHQWVLQHODVWLFVWUXFWXUHVRIPLQ

LPXPZHLJKW6WUXFW2SWLP
0HOFKHUV 5 2Q H[WHQGLQJ WKH UDQJH RI 0LFKHOOOLNH RSWLPDO WRSRORJ\ VWUXFWXUHV
6WUXFW0XOWLGLVF2SWLP
0LFKHOO$*07KHOLPLWVRIHFRQRP\RIPDWHULDOLQIUDPHVWUXFWXUHV3KLO0DJ

0OHMQHN+36FKLUUPDFKHU5$QHQJLQHHULQJDSSURDFKWRRSWLPDOPDWHULDOGLVWUL
EXWLRQDQGVKDSHILQGLQJ&RPS0HWK$SSO0HFK(QJUJ
0RUOH\&77KHPLQLPXPUHLQIRUFHPHQWRIFRQFUHWHVODEV,QW-0HFK6FL
2OKRII15¡QKROW(6FKHHO-7RSRORJ\RSWLPL]DWLRQRIWKUHHGLPHQVLRQDOVWUXF
WXUHVXVLQJRSWLPDOPLFURVWUXFWXUHV6WUXFW2SW
2QJ7*5R]YDQ\*,16]HWR:7/HDVWZHLJKWGHVLJQRISHUIRUDWHGHODVWLFSODWHV
IRU JLYHQ FRPSOLDQFH QRQ]HUR 3RLVVRQ¶V UDWLR &RPS 0HWK $SSO 0HFK (QJ

2ZHQ-%%7KHDQDO\VLVDQGGHVLJQRIOLJKWVWUXFWXUHV(GZDUG$UQROG/WG/RQGRQ
3HGHUVHQ &%: %XKO 7 6LJPXQG 2 7RSRORJ\ V\QWKHVLV RI ODUJH GLVSODFHPHQW
FRPSOLDQWPHFKDQLVPV,QW-1XPHU0HWK(QJUJ
3HGHUVHQ 3 2Q WKH PLQLPXP PDVV OD\RXW RI WUXVVHV LQ µ$*$5' 6\PSRVLXP RQ
6WUXFWXUDO2SWLPL]DWLRQ9RO1$72,VWDQEXO7XUNH\SS±±
3HGHUVHQ 3 2SWLPDO /D\RXW RI 7UXVVHV '&$00 6 6ROLG 0HFKDQLFV '78 ±
WKHVLVIRUWKH3K'
3HGHUVHQ3D7RSRORJ\RSWLPL]DWLRQRIWKUHHGLPHQVLRQDOWUXVVHV,Q0%HQGVRH
DQG &$ 0RWD6RDUHV (GV 7RSRORJ\ 'HVLJQ RI 6WUXFWXUHV 3URF 1$72 $6, 6H
VLPEUD3RUWXJDOSS±.OXZHU'RUGUHFKW
3HGHUVHQ 3 E 7RSRORJ\ RSWLPL]DWLRQ RI G WUXVVHV ZLWK FRVW RI VXSSRUWV ,Q -
+HUVNRYLWVHG6WUXFWXUDO2SWLPL]DWLRQSS±
3LFKXJLQ$97\DV$*LOEHUW02QWKHRSWLPDOLW\RI+HPS¶VDUFKZLWKYHUWLFDO
KDQJHUV6WUXFW0XOWLGLVF2SWLP
3RPH]DQVNL94XHULQ20 5R]YDQ\ *,1 &26,03 H[WHQGHG 6,03 DOJRULWKP
ZLWKGLUHFW&2UQHU&2QWDFW&2QWURO6WUXFW0XOWLGLVF2SWLP
3RXOVHQ 7$ $ VLPSOH VFKHPH WR SUHYHQW FKHFNHUERDUG SDWWHUQV DQG RQH QRGH
FRQQHFWHGKLQJHVLQWRSRORJ\RSWLPL]DWLRQ6WUXFW0XOWLGLVF2SWLP
3RXOVHQ 7$ $ QHZ VFKHPH IRU LPSRVLQJ D PLQLPXP OHQJWK VFDOH LQ WRSRORJ\
RSWLPL]DWLRQ,QW-1XPHU0HWK(QJUJ
3UDJHU: 5R]YDQ\ *,1 D 2SWLPL]DWLRQ RI WKH VWUXFWXUDO JHRPHWU\ ,Q %HGQDUHN
$5 &HVDUL / (GV '\QDPLFDO 6\VWHPV 3URF ,QW &RQI *DLQVYLOOH )ORULGD SS
$FDGHPLF3UHVV1HZ<RUN
3UDJHU:5R]YDQ\*,1E2SWLPDOOD\RXWRIJULOODJHV-6WUXFW0HFK
3UDJHU : 6KLHOG 57 $ JHQHUDO WKHRU\ RI RSWLPDO SODVWLF GHVLJQ - $SSO 0HFK

5RVVRZ037D\ORU-($ILQLWHHOHPHQWPHWKRGIRUWKHRSWLPDOGHVLJQRIYDULDEOH
WKLFNQHVVVKHHWV$,$$-
5R]YDQ\*,1D*ULOODJHVRIPD[LPXPVWUHQJWKDQGPD[LPXPVWLIIQHVV,QW-0HFK
6FL
5R]YDQ\ *,1 E 2SWLPDO ORDG WUDQVPLVVLRQ E\ IOH[XUH &RPS 0HWK $SSO 0HFK
(QJUJ
5R]YDQ\*,12SWLPDOGHVLJQRIIOH[XUDOV\VWHPV3HUJDPRQ2[IRUG
5R]YDQ\*,16WUXFWXUDOGHVLJQYLDRSWLPDOLW\FULWHULD.OXZHU'RUGUHFKW
5R]YDQ\*,1D2SWLPDOOD\RXWWKHRU\DQDO\WLFDOVROXWLRQVIRUHODVWLFVWUXFWXUHVZLWK
VHYHUDOGHIOHFWLRQFRQVWUDLQWVDQGORDGFRQGLWLRQV6WUXFW2SWLP
5R]YDQ\*,1E6KDSHDQGOD\RXWRSWLPL]DWLRQRIVWUXFWXUDOV\VWHPVDQGRSWLPDOLW\
FULWHULDPHWKRGV&,60&RXUVHVDQG/HFWXUHV1R6SULQJHU9HUODJ9LHQQD1HZ
<RUN
5R]YDQ\*,1:KDWLVPHDQLQJIXOLQWRSRORJ\GHVLJQ$QHQJLQHHU¶VYLHZSRLQW,Q
+HUVNRYLWV-(G$GYDQFHVLQ6WUXFWXUDO2SWLPL]DWLRQ3URF6WUXFWXUDO2SWLPL]DWLRQ
5LRGH-DQHLUR.OXZHU'RUGUHFKWSS
5R]YDQ\ *,1 D 6RPH VKRUWFRPLQJV RI 0LFKHOO¶V WUXVV WKHRU\ 6WUXFW 2SWLP

5R]YDQ\*,1E'LIILFXOWLHVLQWUXVVWRSRORJ\RSWLPL]DWLRQZLWKVWUHVVORFDOEXFNOLQJ
DQGV\VWHPVWDELOLW\FRQVWUDLQWV6WUXFW2SWLP
5R]YDQ\ *,1 D 3DUWLDO UHOD[DWLRQ RI WKH RUWKRJRQDOLW\ UHTXLUHPHQW IRU FODVVLFDO
0LFKHOOWUXVVHV6WUXFW2SWLP
5R]YDQ\*,1E([DFWRSWLPDOOD\RXWRIJULOODJHVIRUSDUWLDOO\XSZDUGDQGSDUWLDOO\
GRZQZDUGORDG6WUXFW2SWLP
5R]YDQ\*,1F6WUXFWXUDORSWLPL]DWLRQLQVWUXFWXUDOPHFKDQLFV&,60&RXUVHVDQG
/HFWXUHV1R6SULQJHU9HUODJ9LHQQD1HZ<RUN
5R]YDQ\ *,1 7RSRORJ\ RSWLPL]DWLRQ RI PXOWLSXUSRVH VWUXFWXUHV 0DWK 0HWK
2SHUDWLRQDO5HV
5R]YDQ\ *,1 D $LPV VFRSH PHWKRGV KLVWRU\ DQG XQLILHG WHUPLQRORJ\ RI FRP
SXWHUDLGHG WRSRORJ\ RSWLPL]DWLRQ LQ VWUXFWXUDO PHFKDQLFV 6WUXFW 0XOWLGLVF 2SWLP

5R]YDQ\*,1E6WUHVVUDWLRDQGFRPSOLDQFHEDVHGPHWKRGVLQWRSRORJ\RSWLPL]D
WLRQ±DFULWLFDOUHYLHZ6WUXFW0XOWLGLVF2SWLP
5R]YDQ\*,1D([DFWDQDO\WLFDOVROXWLRQVIRUEHQFKPDUNSUREOHPVLQSUREDELOLVWLF
WRSRORJ\RSWLPL]DWLRQ,Q+HUVNRYLWV- (G 3URF (QJ2SW ,QW &RQI 5LR GH
-DQHLUR
5R]YDQ\ *,1 E $QDO\WLFDO EHQFKPDUNV LQ WRSRORJ\ RSWLPL]DWLRQ ± LQFOXGLQJ
SUREDELOLVWLFGHVLJQ3URFWK$,$$,66020XOWLGLVFLSOLQDU\$QDO\VLVDQG2SWLPL
]DWLRQ&RQIHUHQFH9LFWRULD%&&DQDGD
5R]YDQ\*,1F7UDGLWLRQDOYVH[WHQGHGRSWLPDOLW\LQWRSRORJ\RSWLPL]DWLRQ6WUXFW
0XOWLGLVF2SWLP
5R]YDQ\ *,1 D $ FULWLFDO UHYLHZ RI HVWDEOLVKHG PHWKRGV LQ VWUXFWXUDO WRSRORJ\
RSWLPL]DWLRQ6WUXFW0XOWLGLVF2SWLP
5R]YDQ\*,1E7UDGLWLRQDOYVH[WHQGHGRSWLPDOLW\LQWRSRORJ\RSWLPL]DWLRQ6WUXFW
0XOWLGLVF2SWLP
5R]YDQ\*,1D2QV\PPHWU\DQGQRQXQLTXHQHVVLQH[DFWWRSRORJ\RSWLPL]DWLRQ
84 G.I.N. Rozvany
5R]YDQ\ *,1 E 'RPDLQ DXJPHQWDWLRQ DQG UHGXFWLRQ LQ VWUXFWXUDO WRSRORJ\ RSWL
PL]DWLRQ6WUXFW0XOWLGLVF2SWLP
5R]YDQ\*,1$GLGDP6565HFWDQJXODUJULOODJHVRIOHDVWZHJKW-(QJ0HFK$6&(

5R]YDQ\*,1%HQGVRH03.LUVFK8/D\RXWRSWLPL]DWLRQRIVWUXFWXUHV$SSO0HFK
5HY$60(
5R]YDQ\*,1%LUNHU72QVLQJXODUWRSRORJLHVLQH[DFWOD\RXWRSWLPL]DWLRQ6WUXFW
2SWLP
5R]YDQ\*,1*ROOXE:0LFKHOOOD\RXWVIRUYDULRXVFRPELQDWLRQVRIOLQHVXSSRUWV
±,,QW-0HFK6FL
5R]YDQ\*,1*ROOXE:=KRX0([DFW0LFKHOOOD\RXWVIRUYDULRXVFRPELQDWLRQV
RIOLQHVXSSRUWV±3DUW,,6WUXFW2SWLP
5R]YDQ\ *,1 +LOO 5+ *HQHUDO WKHRU\ RI RSWLPDO ORDG WUDQVPLVVLRQ E\ IOH[XUH
$GYDQFHVLQ$SSO0HFK
5R]YDQ\*,1/HZLQVNL74XHULQ20/RJR-4XDOLW\FRQWUROLQWRSRORJ\RSWL
PL]DWLRQ XVLQJ DQDO\WLFDOO\ GHULYHG EHQFKPDUNV 3URF WK $,$$,6602 0XOWLGLVF
$QDO2SW&RQI3RUWVRXWK9LUJLQLD86$$,$$SDSHU1R$,$$&'
SDSHU S &ROOHFWLRQ RI 7HFKQLFDO 3DSHUV WK $,$$,6602 0XOWLGLVFLSOLQDU\
$QDO\VLVDQG2SWLPL]DWLRQ&RQIHUHQFHSS
5R]YDQ\*,1/LHEHUPDQQ6([DFWRSWLPDOJULOODJHOD\RXWV±3DUW,&RPELQDWLRQV
RIIUHHDQGVLPSO\VXSSRUWHGHGJHV6WUXFW2SWLP
5R]YDQ\*,10DXWH.$QDO\WLFDODQGQXPHULFDOVROXWLRQVIRUDUHOLDELOLW\EDVHG
EHQFKPDUNH[DPSOH6WUXFW0XOWLGLVF2SWLP
5R]YDQ\*,12OKRII17RSRORJ\RSWLPL]DWLRQRIVWUXFWXUHVDQGFRPSRVLWHFRQ
WLQXD3URF1$72$5:%XGDSHVW+XQJDU\0D\.OXZHU'RUGUHFKW
5R]YDQ\*,12OKRII1%HQGVRH032QJ7*6]HWR:7/HDVWZHLJKWGHVLJQRI
SHUIRUDWHGHODVWLFSODWHV'&$005HSRUW1R7HFKQ8QLY'HQPDUN/\QJE\
5R]YDQ\*,12OKRII1%HQGVRH032QJ7*6]HWR:7/HDVWZHLJKWGHVLJQRI
SHUIRUDWHGHODVWLFSODWHV,,,,QW-6ROLGV6WUXFW
5R]YDQ\ *,1 2OKRII 1 &KHQJ .7 7D\ORU -( 2Q WKH VROLG SODWH SDUDGR[ LQ
VWUXFWXUDORSWLPL]DWLRQ-6WUXFW0HFK
5R]YDQ\ *,1 3RPH]DQVNL 9 *DVSDU = 4XHULQ 20 6RPH SLYRWDO LVVXHV LQ
VWUXFWXUDOWRSRORJ\RSWLPL]DWLRQ,Q+HUVNRYLWV-0D]RUFKH6&DQHODVHGV3URF
:&602 &'520 KHOG LQ 5LR GH -DQHLUR %UDVLO &' SDSHU ,6%1
±±
5R]YDQ\*,13RPH]DQVNL94XHULQ20*DVSDU=/RJR-6RPHEDVLFLVVXHVRI
WRSRORJ\ RSWLPL]DWLRQ ,Q %HQGVRH 03 2OKRII 1 6LJPXQG 2 HGV 3URF ,87$0
6\PSRQ7RSRORJ\'HVLJQ2SWLPL]DWLRQRI6WUXFWXUHV0DFKLQHVDQG0DWHULDOVKHOGLQ
5XQJVWHGJDDUG2FWSS6SULQJHU+HLGHOEHUJ%HUOLQ1HZ<RUN
5R]YDQ\*,13UDJHU:$QHZFODVVRIVWUXFWXUDORSWLPL]DWLRQSUREOHPVRSWLPDO
DUFKJULGV&RPS0HWK$SSO0HFK(QJUJ
5R]YDQ\*,1 4XHULQ 20 *DVSDU = 3RPH]DQVNL 9 ([WHQGHG RSWLPDOLW\ LQ WR
SRORJ\GHVLJQ6WUXFW0XOWLGLVF2SWLP
5R]YDQ\*,14XHULQ20*DVSDU=3RPH]DQVNL9:HLJKWLQFUHDVLQJHIIHFWRI

WRSRORJ\VLPSOLILFDWLRQ6WUXFW0XOWLGLVF2SWLP
5R]YDQ\ *,1 4XHULQ 20 /yJy - 3RPH]DQVNL 9 ([DFW DQDO\WLFDO WKHRU\ RI
WRSRORJ\RSWLPL]DWLRQZLWKVRPHSUHH[LVWLQJPHPEHUVRUHOHPHQWV6WUXFW0XOWLGLVF
2SWLP±
5R]YDQ\*,16RNyá7([DFWWUXVVWRSRORJ\RSWLPL]DWLRQ DOORZDQFHIRUVXSSRUW
FRVWVDQGGLIIHUHQWSHUPLVVLEOHVWUHVVHVLQWHQVLRQDQGFRPSUHVVLRQ±H[WHQVLRQVRID
FODVVLFDOVROXWLRQE\0LFKHOO6WUXFW0XOWLGLVF2SWLP
5R]YDQ\*,1=KRX0%LUNHU7*HQHUDOL]HGVKDSHRSWLPL]DWLRQZLWKRXWKRPRJ
HQL]DWLRQ6WUXFW2SWLP
5R]YDQ\ *,1 =KRX 0 %LUNHU 7 :K\ PXOWLORDG WRSRORJ\ GHVLJQV EDVHG RQ
RUWKRJRQDOPLFURVWUXFWXUHVDUHLQJHQHUDOQRQRSWLPDO6WUXFW2SWLP
5R]YDQ\*,1=KRX06LJPXQG2 2SWLPL]DWLRQ RI WRSRORJ\ ,Q $GHOL + (G
$GYDQFHVLQGHVLJQRSWLPL]DWLRQ&KDSPDQDQG+DOO/RQGRQ
6FKHYHQHOV0/D]DURY%66LJPXQG25REXVWWRSRORJ\RSWLPL]DWLRQDFFRXQWLQJ
IRU VSDWLDOO\ YDU\LQJ PDQXIDFWXULQJ HUURUV &RPS 0HWK $SSO PHFK (QJUJ

6LJPXQG 2 'HVLJQ RI PDWHULDO VWUXFWXUHV XVLQJ WRSRORJ\ RSWLPL]DWLRQ 3K '
7KHVLV'HSDUWPHQWRI6ROLG0HFKDQLFV7HFKQLFDO8QLYHUVLW\RI'HQPDUN
6LJPXQG2'HVLJQRIPXOWLSK\VLFVDFWXDWRUVXVLQJWRSRORJ\RSWLPL]DWLRQ±3DUW,
2QHPDWHULDOVWUXFWXUHV&RPS0HWK$SSO0HFK(QJUJ
6LJPXQG20RUSKRORJ\EDVHGEODFNDQGZKLWHILOWHUVIRUWRSRORJ\RSWLPL]DWLRQ
6LJPXQG22QWKHXVHIXOQHVVRIQRQJUDGLHQWDSSURDFKHVLQWRSRORJ\SWLPL]DWLRQ
6LJPXQG 2 &ODXVHQ 30 7RSRORJ\ RSWLPL]DWLRQ XVLQJ D PL[HG IRUPXODWLRQ $Q
DOWHUQDWLYH ZD\ WR VROYH SUHVVXUH ORDG SUREOHPV &RPS 0HWK $SSO 0HFK (QJ

6LJPXQG2+RXJDUG.*HRPHWULFSURSHUWLHV RI RSWLPDO SKRWRQLF FU\VWDOV 3K\V
5HY/HWWHUV$UWLFOH1XPEHU
6LJPXQG 2 -HQVHQ -6 6\VWHPDWLF GHVLJQ RI SKRQRQLF EDQGJDS PDWHULDOV DQG
VWUXFWXUHVE\WRSRORJ\RSWLPL]DWLRQ3KLO7UDQV5R\6RFLQ/RQGRQ6HULHV$±0DWK
3K\V(QJ6FL
6LJPXQG23HWHUVVRQ-1XPHULFDOLQVWDELOLWLHVLQWRSRORJ\RSWLPL]DWLRQ$VXUYH\
RQ SURFHGXUHV GHDOLQJ ZLWK FKHFNHUERDUGV PHVKGHSHQGHQFLHV DQG ORFDO PLQLPD
6WUXFW2SWLP
6LJPXQG2=KRX05R]YDQ\*,1/D\RXWRSWLPL]DWLRQRIODUJH)(V\VWHPVE\
QHZRSWLPDOLW\FULWHULDPHWKRGVDSSOLFDWLRQVWREHDPV\VWHPV,Q+DXJ(-HG3URF
1$72 $6, &RQFXUUHQW (QJLQHHULQJ WRROV DQG WHFKQRORJLHV IRU PHFKDQLFDO V\VWHP
GHVLJQ,RZD6SULQJHU9HUODJ%HUOLQ
6RNyá 7 D $ OLQH FRGH IRU GLVFUHWL]HG 0LFKHOO WUXVV RSWLPL]DWLRQ ZULWWHQ LQ
0DWKHPDWLFD6WUXFW0XOWLGLVF2SWLP
86 G.I.N. Rozvany
6RNyá 7 E 7RSRORJ\ RSWLPL]DWLRQ RI ODUJHVFDOH WUXVVHV XVLQJ JURXQG VWUXFWXUH
DSSURDFKZLWKVHOHFWLYHVXEVHWVRIDFWLYHEDUV,Q%RUNRZVNL$/HZLĔVNL7']LHU
ĪDQRZVNL*(GVWK,QW&RQIRQµ&RPSXWHU0HWKRGVLQ0HFKDQLFV¶&00
±0D\:DUVDZ%RRNRI$EVWUDFWSS±
6RNyá7/HZLĔVNL72QWKHVROXWLRQRIWKHWKUHHIRUFHVSUREOHPDQGLWVDSSOLFDWLRQ
LQRSWLPDOGHVLJQLQJRIDFODVVRIV\PPHWULFSODQHIUDPHZRUNVRIOHDVWZHLJKW6WUXFW
0XOWLGLVF2SWLP
6RNyá7/HZLĔVNL72SWLPDOGHVLJQRIDFODVVRIV\PPHWULFSODQHIUDPHZRUNVRI
OHDVWZHLJKW6WUXFW0XOWLGLVF2SWLP±
6RNyá75R]YDQ\*,11HZDQDO\WLFDOEHQFKPDUNVIRUWRSRORJ\RSWLPL]DWLRQDQG
WKHLULPSOLFDWLRQV3DUW,ELV\PPHWULFWUXVVHVZLWKWZRSRLQWORDGVEHWZHHQVXSSRUWV
6RNyá75R]YDQ\*,1D([DFWOHDVWYROXPHWUXVVHVIRUWZRV\PPHWULFSRLQWORDGV
DQGXQHTXDOSHUPLVVLEOHVWUHVVHVLQWHQVLRQDQGFRPSUHVVLRQ6WUXFW0XOWLGLVF2SWLP

6RNyá75R]YDQ\*,1E1HZDQDO\WLFDOEHQFKPDUNVIRUWRSRORJ\RSWLPL]DWLRQDQG
WKHLULPSOLFDWLRQV3DUW,,2SWLPDOWUXVVWRSRORJLHVIRUWZRVXSSRUWVDQGDSRLQWORDG
6WUXFW0XOWLGLVF2SWLPWREHVXEPLWWHG
6WROSH 0 %HQGVRH 03 *OREDO RSWLPD IRU WKH =KRX5R]YDQ\ SUREOHP VWUXFW
0XOWLGLVF2SWLP
6YHG*7KHPLQLPXPZHLJKWRIFHUWDLQUHGXQGDQWVWUXFWXUHV$XVWUDO-$SSO6FL

6]HWR :7 0LFURVWUXFWXUH VWXGLHV DQG VWUXFWXUDO RSWLPL]DWLRQ 3K ' 7KHVLV
0RQDVK8QLYHUVLW\0HOERXUQH
7DOLVFKL & 3DXOLQR *+ /H &+ +RQH\FRPE :DFKVSUHVV ILQLWH HOHPHQWV IRU WR
SRORJ\RSWLPL]DWLRQ6WUXFW0XOWLGLVF2SWLP
9i]TXH](VSL0&HUYHUD%UDYR-2QWKHVROXWLRQRIWKHWKUHHIRUFHVSUREOHPDQGLWV
DSSOLFDWLRQ LQ RSWLPDO GHVLJQLQJ RI D FODVV RI V\PPHWULF SODQH IUDPHZRUNV RI OHDVW
ZHLJKW'LVFXVVLRQRQWKH0LFKHOOFODVVFRQFHSW6WUXFW0XOWLGLVF2SWLP
:DQJ)/D]DURY%66LJPXQG22QSURMHFWLRQPHWKRGVFRQYHUJHQFHDQGUREXVW
IRUPXODWLRQVLQWRSRORJ\RSWLPL]DWLRQ6WUXFW0XOWLGLVF2SWLP
:DVLXW\QVNL=%UDQGW$7KHSUHVHQWVWDWHRINQRZOHGJHLQWKHILHOGRIRSWLPXP
GHVLJQRIVWUXFWXUHV$SSO0HFK5HY
<RRQ *+ -HQVHQ -6 6LJPXQG 2 7RSRORJ\ RSWLPL]DWLRQ RI DFFRXVWLFVWUXFWXUH
LQWHUDFWLRQSUREOHPVXVLQJDPL[HGILQLWHHOHPHQWIRUPXODWLRQ,QW-1XPHU0HWK(Q
JUJ
=KRX0'LIILFXOWLHVLQWUXVVWRSRORJ\RSWLPL]DWLRQZLWKVWUHVVDQGORFDOEXFNOLQJ
FRQVWUDLQWV6WUXFW2SWLP
=KRX05R]YDQ\*,17KH&2&DOJRULWKP3DUW,,WRSRORJLFDOJHRPHWULFDODQG
JHQHUDOL]HGVKDSHRSWLPL]DWLRQ&RPS0HWK$SSO0HFK(QJUJ
=KRX 0 5R]YDQ\ *,1 '&2& $Q RSWLPDOLW\ FULWHULRQ PHWKRG IRU ODUJH
V\VWHPV3DUW,7KHRU\3DUW,,DOJRULWKP6WUXFW2SWLP
=KRX06K\\=.7KRPDV+/&KHFNHUERDUGDQGPLQLPXPPHPEHUVL]HFRQWUROLQ
WRSRORJ\RSWLPL]DWLRQ6WUXFW0XOWLGLVF2SWLP

2Q%DVLF3URSHUWLHVRI0LFKHOO
V6WUXFWXUHV
7RPDV]/HZLĔVNL DQG7RPDV]6RNyá

)DFXOW\RI&LYLO(QJLQHHULQJ:DUVDZ8QLYHUVLW\RI7HFKQRORJ\:DUVDZ3RODQG
$EVWUDFW7KHSDSHULVDQLQWURGXFWLRQWRZDUGV0LFKHOO¶VRSWLPXPGHVLJQ
SUREOHPV RI SLQMRLQWHG IUDPHZRUNV 6WDUWLQJ IURP WKH RSWLPXP GHVLJQ
SUREOHP LQ LWV GLVFUHWH VHWWLQJ ZH VKRZ WKH SDVVDJH WR WKH GLV
FUHWHFRQWLQXRXV VHWWLQJ LQ WKH NLQHPDWLF IRUP DQG WKHQ WKH SULPDO
VWUHVVEDVHGVHWWLQJ7KHFODVVLFSURSHUWLHVRI+HQFN\QHWVDUHFRQVLVWHQWO\
GHULYHGIURPWKHNLQHPDWLFRSWLPDOLW\FRQGLWLRQV7KH5LHPDQQPHWKRGRI
WKH QHW FRQVWUXFWLRQ LV EULHIO\ UHFDOOHG 7KLV LV WKH EDVLV IRU ILQGLQJ WKH
DQDO\WLFDO VROXWLRQV WR WKH RSWLPXP GHVLJQ SUREOHPV 7KH SDSHU UHIHUV
PDLQO\WRWKHZHOONQRZQVROXWLRQVHJWKHFDQWLOHYHUV\HWGLVFXVVHVDOVR
WKHRSHQSUREOHPVFRQFHUQLQJWKRVHFODVVHVRIVROXWLRQVLQZKLFKWKHNL
QHPDWLFDSSURDFKFDQQRWSUHFHGHWKHVWDWLFDQDO\VLV
,QWURGXFWLRQ
7KHVXEMHFWRIFRQVLGHUDWLRQLVDQRSWLPXPGHVLJQSUREOHPRIVWUXFWXUDOPHFKDQLFV
WRIRUPDVWUXFWXUHZLWKLQDJLYHQIHDVLEOHGRPDLQ:ZKRVHERXQGDU\VHJPHQW*LVD
VXSSRUWIRUWKHJLYHQLQSODQHORDGDSSOLHGDWRWKHUSDUWRIWKHERXQGDU\,WLVQRW
LPSRVHGWKDWDOOSRLQWVRIWKHVHJPHQW*ZLOOSOD\WKHUROHRIWKHVXSSRUWLQJQRGHV
7KHVWUHVVLVVDLGWREHDGPLVVLEOHLIERWKWKHSULQFLSDOVWUHVVHVDUHERXQGHGE\V&
DQGV7WKHVHYDOXHVEHLQJFDOOHGWKHOLPLWVWUHVVHVIRUFRPSUHVVLRQDQGWHQVLRQ7KH
SUREOHPGLVFXVVHGFDOOHG0LFKHOO¶VSUREOHPUHDGV ILQGWKHVWUXFWXUHRIPLQLPDO
YROXPHDGPLVVLEO\VWUHVVHGFDSDEOHRIWUDQVPLWWLQJDJLYHQORDGWRWKHJLYHQVXS
SRUWFI)LJ
'XH WR WKH SUHVHQFH RI WKH SRLQW ORDGV WKH VROXWLRQ LV H[SHFWHG WR EH GLV
FUHWHFRQWLQXRXV,QWKHSODQHFDVHWKHRSWLPXPVWUXFWXUHVDUHFRPSRVHGRIDWKH
EDUVRIILQLWHFURVVVHFWLRQVWKH\UHLQIRUFHWKHVWUXFWXUHDQGPDNHSRVVLEOHWKHHTXL
OLEULXPRIWKHQRGHVZKHUHWKHSRLQWORDGVDUHDSSOLHGDQGZKLFKPDNHLWSRVVLEOHWR
VXSSRUWWKHVWUXFWXUHDWVRPHSRLQWVEWKHVROLGERG\SODFHGEHWZHHQWKHEDUVRI
ILQLWHFURVVVHFWLRQV
7KHOLPLWVWUHVVHVV&V7FRUUHVSRQGWRERWKWKHD[LDOVWUHVVVWDWHLQEDUVDQGWR
WKHELD[LDOVWUHVVVWDWHLQWKHVROLGSDUWRIWKHVWUXFWXUH7KLVVWDWHPHQWQHHGVH[
SODQDWLRQ,QWKHPHFKDQLFVRIFRPSRVLWHVWKHVWUHVVFRPSRQHQWVKDYHWKHPHDQLQJRI
WKHDYHUDJHVWUHVVHV7KHTXDQWLWLHVV&V7GRQRWUHIHUWRWKLVVWUHVVVWDWH7KH\UHIHU

DOI 10.1007/978-3-7091-1643-2_6, © CISM, Udine 2014
88 T. Lewiński and T. Sokół
WRWKHVWDWHRIVWUHVVLQWKHPLFURVWUXFWXUHRUXVLQJWKHWHUPVRIWKHWKHRU\RIKR
PRJHQL]DWLRQWKH\DUHWKHVWUHVVOLPLWVIRUWKHVWUHVVVWDWHZLWKLQWKHSHULRGLFLW\FHOOV
RQHOHYHOORZHUWKDQWKHOHYHODWZKLFKWKHSUREOHPLVVHW

)LJXUH7KHSUREOHPVHWWLQJLQWKHSODQHFDVHIRURQHSRLQWORDG$QH[DPSOHEHORQJLQJ
WRWKHFODVVRI0LFKHOO¶VFDQWLOHYHUV
$OUHDG\LQWKHRULJLQDOSDSHUE\0LFKHOOLWZDVGLVFORVHGWKDWWKHRSWLPXP

VWUXFWXUH VKRXOG KDYH D ILEURXV PLFURVWUXFWXUH UHLQIRUFHG E\ EDUV RI ILQLWH FURVV
VHFWLRQV7KHERG\ZKLFKILOOVXSWKHVWUXFWXUHLVDFRPSRVLWHRIWKHPLFURVWUXFWXUH
EXLOW IURP RQH RU WZR IDPLOLHV RI ILEUHV LQ WKH ODWWHU FDVH WKH ILEUHV IRUP DQ RU
WKRJRQDOFXUYLOLQHDUQHW6RPHGRPDLQVFDQEHHPSW\
7KH H[DFW VROXWLRQV WR WKH RSWLPXP GHVLJQ SUREOHP FRQVLGHUHG NQRZQ XS WR
HDUO\VHYHQWLHVDUHSXWIRUZDUGLQWKHERRNVE\&R[DQG+HPSEDVHG
SDUWO\ RQ WKH UHSRUWV DQG SDSHUV E\ $6/&KDQ DQG +6<&KDQ
7KHDQDO\WLFDOVROXWLRQVXSWRPLGGOH
VDUHUHYLHZHGE\
5R]YDQ\ H[FHSW IRU OLQH VXSSRUWV ZKLFK ZHUH GLVFXVVHG E\ 5R]YDQ\ DQG
*ROOXE DQG 5R]YDQ\ HW DO 7KH UHFHQW DQDO\WLFDO VROXWLRQV FDQ EH
IRXQGLQWKHSDSHUVE\*UDF]\NRZVNLDQG/HZLĔVNL/HZLĔVNL
DQG5R]YDQ\'DUZLFKHWDO3LFKXJLQHWDO3ULWFKDUGHW
DO 7\DV HW DO 0HOFKHUV 'HZKXUVW 'HZKXUVW DQG
&ROOLQV 6RNyá DQG /HZLĔVNL 5R]YDQ\ DQG 6RNyá
3LFKXJLQHWDO
7KHDLPRIWKHSUHVHQWZRUNLVWRSXWIRUZDUGDQLQWURGXFWLRQWRWKHVXEMHFWE\
UHFDOOLQJWKHJHRPHWULFDODVSHFWVRIWKHSUREOHPRIWKHILEURXVQHWVFRQVWUXFWLRQVDQG
SXWWLQJIRUZDUGWKH5LHPDQQ
VPHWKRGRIVROYLQJWKHJRYHUQLQJK\SHUEROLFSUREOHP
On Basic Properties of Michell’s Structures 89
0LFKHOO¶VSUREOHPDSSHDUVDVDOLPLWFDVHRIWKHJHQHUDOL]HGVKDSHRSWLPL]DWLRQ
DVSURYHGLQ$OODLUH%HQGV¡H2QWKHRWKHUKDQGWKH0LFKHOO
VROXWLRQVDUHORZHUOLPLWVIRUWKHWUXVVRSWLPL]DWLRQVHH3UDJHUDE0F&RQQHO
3ULWFKDUG HW DO 0D]XUHN HW DO 0D]XUHN 6RNyá
D7KHQXPHULFDOPHWKRGVWREHXVHGLQWKHSUHVHQWDUWLFOHDUHEDVHGRQWKH
JURXQGVWUXFWXUHFRQFHSWVHH'RUQHWDO
0LFKHOO¶VSUREOHPKDVPXFKLQFRPPRQZLWKHTXLOLEULXPRIDORFNLQJPDWHULDO
DVQRWHGE\:LOOLDP3UDJHU5R]YDQ\6HF5R]YDQ\DQG$GLGDP
DDQG6WUDQJDQG.RKQ)RUWKHFRQWHPSRUDU\WKHRU\RIWKHORFNLQJ
PDWHULDOVWKHUHDGHULVUHIHUUHGWR'HPHQJHODQG6XTXHW
7KLVSDSHUUHIHUVWRWKHFDVHRIWKHORDGVDWIL[HGSRLQWVWUDQVPLVVLEOHORDGVKDYH
EHHQGLVFXVVHGLQ5R]YDQ\DQG:DQJ:DQJDQG5R]YDQ\DQG'DU
ZLFKHWDO
0LFKHOOVWUXFWXUHVZHUHLQVSLUDWLRQIRUFLYLOHQJLQHHUVDQGDUFKLWHFWVLQWKHLUYL
VLRQVRIQHZVKDSHVRIEXLOGLQJVVHH=DOHZVNLDQG$OOHQDQG0D]XUHNHWDO
6WURPEHUJHWDO
7KHXVXDOVXPPDWLRQFRQYHQWLRQIRUUHSHDWHGLQGLFHVLVDGRSWHG7KHIROORZLQJ
UDQJHV IRU WKH LQGLFHV DUH DVVXPHG D V D Q LM G N H
. H/ H«HZKHUHVLVWKHQXPEHURIORDGVGLVWKHGLPHQVLRQRIWKH
SUREOHPG LQWKHSODQHFDVHDQGG LQWKHVSDWLDOFDVHQLVWKHQXPEHURI
XQORDGHGQRGHVHLVWKHQXPEHURIEDUVHLVWKHQXPEHURIEDUVLQWHQVLRQLQGH[HG
E\.WKHEDUVLQFRPSUHVVLRQDUHLQGH[HGE\/
(TXLOLEULXPRI7UXVVHV
$VVXPHWKDWWKHYHFWRUVUDD «QGHWHUPLQHSRVLWLRQVRIWKHQRGHVRIDWUXVVLQ
WKH'FDVHG RULQ'FDVHG 7KHORDGVRIPDJQLWXGHV3DD VDUH
FRQFHQWUDWHGDWWKHQRGHV7KHQRGDOGLVSODFHPHQWVYDGLUHFWHGDORQJWKHORDGV3D
GHWHUPLQHWKHGHIRUPHGFRQILJXUDWLRQRIWKHWUXVV7KHVXSSRUWVDUHQRWVXEMHFWWR
VHWWOHPHQWV7KHYHFWRUVUDGHWHUPLQHWKHSRVLWLRQVRIEDUVWKHEDURILQGH[NLVRI
OHQJWKONWKHIRUFHLQWKLVEDULVGHQRWHGE\1N7KHVHLQWHUQDOIRUFHVDUHOLQNHGZLWK
WKHORDGV3DE\WKHHTXLOLEULXPHTXDWLRQVRIWKHQRGHV7KHHORQJDWLRQ'NRIWKHEDU
RILQGH[NLVH[SUHVVHGE\WKHQRGDOGLVSODFHPHQWVDVIROORZV
V
'N ¦ % D YD N
N H
D
,I YD DUH GLUHFWHG DORQJ WKH D[HV [L L G WKH FRHIILFLHQWV %ND DUH GLUHFWLRQDO
FRVLQHVRIEDUV7KHVHFRHIILFLHQWVIRUPDQHE\VJHRPHWULFPDWUL[7KHVHFRHIIL
FLHQWVDUHGHWHUPLQHGE\WKHYHFWRUVUDEXWWKHVHHTXDWLRQVZLOOQRWEHGLVSOD\HG
&RQVLGHUWKHYDULDWLRQDOHTXDWLRQFDOOHGWKHYLUWXDOZRUNHTXDWLRQ
H V
¦1 N 'N ¦ 3D YD Y 5 V
N D
ZKHUH'NLVOLQNHGZLWKYDE\7KHDERYHHTXDWLRQVLPSO\

H
¦% D1 N N 3D D V
N
7KH HTXDWLRQV RU H[SUHVV DOO HTXDWLRQV RI HTXLOLEULXP RI QRGHV
7KHVHHTXDWLRQVGRQRWLQFOXGHHTXLOLEULXPHTXDWLRQVRIWKHVXSSRUWLQJQRGHVVLQFH
DOOGLVSODFHPHQWVYDDUHNLQHPDWLFDOO\DGPLVVLEOH7KHUHODWLYHHORQJDWLRQRIWKHNWK
EDULVH[SUHVVHGE\
V
HN ¦ %ND YD
ON D

7KH EDUV LQ WHQVLRQ ZLOO EH LQGH[HG E\ . «H RU 1.t DQG WKH EDUV LQ
FRPSUHVVLRQZLOOEHLQGH[HGE\/ H«H1/d/HWXVZULWHDVIROORZV
H H V
¦1
.
O H
. . . ¦1 O H
/ H
/ / / ¦ 3D YD
D
Y 5 V
ZKHUHH.H/DUHOLQNHGZLWKYDE\
7KH3UREOHPRI0LQLPL]DWLRQRIWKH9ROXPHRID7UXVVRID
%RXQGHG6WUHVV/HYHO
$VVXPHWKHVWUHVVVLQEDUVWREHERXQGHG
V & d V d V 7
ZKHUHV&V7DUHDGPLVVLEOHVWUHVVHVLQFRPSUHVVLRQDQGWHQVLRQUHVSHFWLYHO\/HW$.
$/EHDUHDVRIWKHFURVVVHFWLRQVRIEDUVLQWHQVLRQDQGFRPSUHVVLRQUHVSHFWLYHO\
7KXVZHUHTXLUHWKDW
1 / t V & $/ 1 . d V 7 $.
/HW V V 7 V & 7KHYROXPHRIWKHWUXVVLVHVWLPDWHGIURPEHORZE\

H
Ö V
H
V
9t
V
9 9Ö ¦
. V7
1 . O. ¦ 1 /O/
/ H V&

DQGWKHHTXDOLW\WDNHVSODFHLIWKHFURVVVHFWLRQVDUHVWUHVVHGXSWRWKHOLPLWV
$VVXPHQRZWKDWWKHIRUFHVDQGWKHSRLQWVRIWKHLUDSSOLFDWLRQDUHIL[HG$VVXPHWKDW
WKHQRGHVRILQGLFHVDIURPWKHVHW,DUHQRWORDGHG7KHLUSRVLWLRQGRQRWXQGHUJR
RSWLPL]DWLRQ:HFRQVLGHUWKHSUREOHP

9PLQ PLQ PLQ 9Ö
V UD D, 1 . t 1 / d
VDWLVI\LQJ
,WZLOOEHVKRZQODWHUWKDWWKHRSWLPDOWUXVVHVDUHVWDWLFDOO\GHWHUPLQDWHKHQFHIXOO\
VWUHVVHG7KLVPHDQVWKDWZLOOSURYLGHWKHYDOXHRIWKHYROXPHDQGQRWLWVORZHU
ERXQG
1RZ ZH VKDOO LQWURGXFH WKH FRQGLWLRQV LQWR XVLQJ WKH WHFKQLTXH RI
/DJUDQJHPXOWLSOLHUV:HUHZULWHDV

9PLQ PLQ /ÛD `
V UD D,
ZKHUH
ª V H H º
/ÛD ` PLQ PD[V «9Ö ¦ 3D YD ¦ 1 . O . H . ¦ 1 / O /H / »
1 . t Y5
1 / d «¬ D . / H »¼
ZKHUHWKHUHODWLYHHORQJDWLRQVDUHOLQNHGZLWKYE\7KHRUGHURIPLQDQGPD[

RSHUDWLRQVFDQEHLQWHUFKDQJHGOHDGLQJWR
ª V º
/ÛD ` PD[ «¦ 3D YD . Y YV »
¬D ¼
V
Y 5
ZLWK
ª H § V · H
§ V · º
. Y Y V PLQ « ¦ ¨¨ H . Y ¸¸1 . O. ¦ ¨¨ H / Y ¸¸1 /O/ »
¬ . © V 7 V&
1 t
1 d « ¹ / H © ¹ »¼
.
/
7KHODVWTXDQWLW\FDQEHFRPSXWHGDQDO\WLFDOO\
LI H . V V7 DQG H / V V & D

°
. Y YV ® LI H . V V7 DQG H / ! V V & E
° f LI H. ! V V7 RU H / V V & F
¯
7KHPLQLPXPLVDWWDLQHGIRU
1. t DQG 1/ d D
1. DQG 1/ E
1 . f RU 1 / f F
7KHPD[RSHUDWLRQLQH[FOXGHVWKHFDVHF+HQFH

V
/ÛD ` PD[V ¦ 3D YD
Y5 D
H N > V V & V V 7 @
ZKHUH HNGHSHQGVRQYE\/HWXVFRPHEDFNWR7KHPLQLPDOYROXPHLV

JLYHQE\
V

9PLQ PLQ PD[V
V UD D, Y5
¦ 3D YD
D

H N >V V & V V 7 @
5HFDOOWKDW%NDLQYROYHGLQDUHGHWHUPLQHGE\UD

0LFKHOO7UXVVHVDVD6XEFODVVRI*OREDO0LQLPL]HUVRI3URE
OHP
,IWKHVHWRIQRGHVUDD,LVPDGHELJJHUDQGELJJHUE\DGPLWWLQJQHZSRLQWVDV
QRGHVDQGDGPLWWLQJQHZEDUVWKHQWKHTXDQWLW\9PLQZLOOGHFUHDVH6XFKDVHTXHQFH
RI VXERSWLPDO WUXVVHV LV VKRZQ LQ *UDF]\NRZVNL DQG /HZLĔVNL 7KXV WKH
YROXPH GHFUHDVHV LI WKH QXPEHU RI WKH PHPEHUV LQFUHDVHV 7KLV SURFHVV GRHV QRW
VWRSZKLFKPHDQVWKDWWKHOLJKWHVWIXOO\VWUHVVHGVWUXFWXUHWUDQVPLWWLQJDJLYHQORDG
WRJLYHQVXSSRUWVLVDILEURXVVWUXFWXUHSRVVLEO\UHLQIRUFHGE\EDUVRIILQLWHFURVV
VHFWLRQVLIWKHSRLQWORDGVDUHDSSOLHG<HWWKHILEURXVGRPDLQVPD\EHDEVHQWLIWKH
NLQHPDWLF FRQGLWLRQV SUHYHQW IURP IRUPLQJ WKHP 7KHQ WKH RSWLPDO VWUXFWXUH LV
FRPSRVHGRIILQLWHQXPEHURIPHPEHUV$QH[DPSOHRIVXFKDWUXVVLVWKHWZREDU
WUXVVEHLQJWKHVROXWLRQWRWKHSUREOHPRI)LJFGLQZKLFKOdDOEHLQJWKHGLVWDQFH
RIWKHORDGWRWKHYHUWLFDOOLQH$%_$%_ D7KHVHWZRVROXWLRQVGHULYHGE\5R]
YDQ\KDYHEHHQFRQILUPHGKHUHE\XVLQJPLOOLRQRISRVVLEOHPHPEHUVE\
SHUIRUPLQJWKHFRPSXWDWLRQVEDVHGRQWKHPHWKRGRIWKHJURXQGVWUXFWXUHXVLQJWKH
SURJUDPE\76RNyáE
/HWXVFRPHEDFNWR:HVKDOOFRQVLGHUDVXEFODVVRIWKHIDPLO\RIVROX
WLRQVRIWKLVSUREOHPLQZKLFKDOOGLVSODFHPHQWVYDFDQEHWUHDWHGDVLPSOLHGE\D
GLVSODFHPHQWILHOGYJLYHQLQWKHGHVLJQGRPDLQ:/HWXVVWUHVVWKDWYDLVQRWWKHDWK
FRPSRQHQWRIWKHYHFWRUY7KHILHOGYGHWHUPLQHVDWHDFKSRLQWWKHGHIRUPDWLRQV
H LM Y Y L M Y M L
ZKHUH L w w[L DQGYLLVWKHLWKFRPSRQHQWRIY7KHFRPSRQHQWV HLMIRUPWKH

V\PPHWULFWHQVRUHHRIRUGHUZKLFKLVGHQRWHGDV İ (V 7KHGHIRUPDWLRQV
GHWHUPLQHWKHVWUDLQDORQJDQ\YHFWRUQRIFRPSRQHQWVQL
H LM Y QL Q M
H Q Y

Q
ZKHUH__Q__ QLQL7KHTXDQWLWLHVHLMYDQGQDUHUHIHUUHGWRWKHVDPHSRLQW[RI:,I
WKHHQGSRLQWVRIYHFWRUVUDD,UXQRYHUDOOSRLQWV[RI:WKHQWKHFRQGLWLRQV
V V & d H N d V V 7 ZLOOEHUHSODFHGE\
V V & d H Q Y [ d V V 7 IRUDH [ :
ZKHUHQLVDUELWUDU\7KXVWKHPLQRSHUDWLRQRYHUUDD,LVQRZUHGXQGDQWVLQFHWKH
HQGSRLQWVRIYHFWRUVUDUXQRYHUWKHZKROHGRPDLQ:7KXVOHWXVVWUHVVRQFHDJDLQ
WKDWZHFRQVLGHUVXFKDFODVVRIVROXWLRQVWRWKHSUREOHPIRUZKLFK
± WKHGLVSODFHPHQWVYDDUHLPSOLHGE\WKHILHOGY7KH\DUHGLUHFWHGDORQJWKH
IRUFHV3D7KHILHOGYLVNLQHPDWLFDOO\DGPLVVLEOHKHQFHLVFRQWLQXRXV7KH
ILHOGYZLOOEHYLHZHGDVDQHOHPHQWRI+:LHWKHVSDFHRINLQHPDWLFDOO\
DGPLVVLEOHGLVSODFHPHQWVRIDPHGLXPLQZKLFKWKHXQNQRZQIUDPHZRUNLV
HPEHGGHG7KHYLUWXDOZRUN63DYDZLOOEHYLHZHGDVWKHYDOXHRIDOLQHDU
IRUPIRQYRUIY
± WKHVWUDLQVHNLQFDQEHUHSODFHGE\HHQY[ZLWKDUELWUDU\YHFWRUVQ
O
% D E F G
D
D
$

)LJXUH ([DPSOHV RI RUWKRJRQDO VROXWLRQV D O D E O D F O D DQG
QRQRUWKRJRQDOVROXWLRQGO D
6XFKVROXWLRQVZLOOEHFDOOHG0LFKHOO¶VIUDPHVRU0LFKHOO¶VWUXVVHV7KXVWKHYROXPH
RID0LFKHOOIUDPHLVJLYHQE\

90 PD[ I Y
V HY+ :
Y [ >V V V V @
Q & 7 IRUDH [:
Q5 G
,Q WKH FDVH RI RQH SRLQW ORDG 3 >3 3G@ DW [ WKH YLUWXDO ZRUN HTXDOV
IY 3LYL[7KHOLQHDUIRUPIWKXVGHILQHGLVQRWFRQWLQXRXV
/HWG DQG H,tH,,tH,,,EHSULQFLSDOVWUDLQVWKHQWKHH[WUHPDOHLJHQYDOXHVRI
WKHWHQVRUHHDUHH[SUHVVHGE\
H, PD[ H Q H ,,, PLQ H Q
Q5 Q5
/HWQ Q,EHWKHPD[LPL]HUDQGQ Q,,,WKHPLQLPL]HURIWKHDERYHIRUPXODH7KHVH

YHFWRUVDUHPXWXDOO\RUWKRJRQDO7KH\GHWHUPLQHWKHGLUHFWLRQVFDOOHG,DQG,,,7KXV
IRUWKHFDVHRIG WKHFRQGLWLRQ V V & d İ Q d V V 7 PD\EHUHSODFHGE\WKH
FRQGLWLRQV
H , d V V 7 H ,,, t V V &
,QWKHFDVHRIG WKHVHFRQGLWLRQVDVVXPHWKHIRUP
H , d V V 7 H ,, t V V &
7KHFRQGLWLRQVRUZLOOEHZULWWHQHH%V&V7ZKHUH%V&V7LVWKHVHW

RIVXFKWHQVRUV İ (V WKDWWKHVPDOOHVWHLJHQYDOXHRIHHLVQRWVPDOOHUWKDQVV&
DQGWKHELJJHVWHLJHQYDOXHRIHHLVQRWJUHDWHUWKDQVV7/HWXVUHZULWHLQWKH
IRUP

90 PD[ I Y
V Y+ :
İ Y [ % V & V 7 IRUDH [:
1RWHWKDWWKHVSDWLDO0LFKHOOIUDPHVDUHFRPSRVHGRI
± WKHVXUIDFHQHWVFRPSRVHGRIWZRIDPLOLHVRIILEUHVVWUHVVHGDVDPHPEUDQH
± WKHXQLGLUHFWLRQDOIDQVRIYDU\LQJVWUHDPRIILEUHV
7KHDERYHSURSHUWLHVDUHFRQVHTXHQFHVRIDQGLIWKHVWUDLQOLHVEH
WZHHQWKHERXQGVWKHILEUHVGRQRWZRUNWKH\EHJDQWRZRUNLIWKHVWUDLQVDWWDLQWKH
ERXQGV,Q'LIERWKERXQGVDUHDWWDLQHGWKHQWKHLQWHUPHGLDWHSULQFLSDOVWUDLQ
OLHVEHWZHHQWKHERXQGVDQGKHQFHWKHFRUUHVSRQGLQJVWUHVVYDQLVKHV7KLVPHDQVWKDW
WKHUHDUHQRILEUHVLQWKHGLUHFWLRQRIWKHLQWHUPHGLDWHSULQFLSDOGLUHFWLRQDQGVLQFHWKH
VWUDLQVLQRWKHUGLUHFWLRQVWKDQ,DQG,,,YDQLVKWKHRQO\ILEHUVOLHDORQJWKH,DQG,,,
GLUHFWLRQV7KHRSWLPDOVWUXFWXUHEHFRPHVDILEURXVVKHOOLQWKHPHPEUDQHVWDWHRI
VWUHVV WKH QRUPDO GLUHFWLRQ WR WKLV VKHOO LV WKH GLUHFWLRQ ,, 7KH ILEUHV IRUP WZR
IDPLOLHVRIRUWKRJRQDOQHWVVLQFHGLUHFWLRQV,DQG,,,DUHRUWKRJRQDO7KHRUWKRJR
QDOLW\ FRQGLWLRQV FDQ EH YLRODWHG DW WKH ERXQGDULHV RU LQ WKH FDVH LI WKH RSWLPDO
VWUXFWXUH LV FRPSRVHG RI ILQLWH QXPEHU RI EDUV 7KH RQO\ NQRZQ 0LFKHOO IUDPH
ZRUNLQJLQVSDFHLV0LFKHOO¶VVSKHUHVHH/HZLĔVNLZKHUHWKHSURRILVJLYHQ
WKDWWKHVSKHULFDOVKHOOLV OLJKWHU WKDQ RWKHU VKHOOV RI UHYROXWLRQ VXEMHFWHG WR D[LDO
WRUVLRQ
D E
F
T FRQVW

)LJXUH6ROXWLRQVRIVHOHFWHG0LFKHOO¶VSUREOHPVDV\PPHWULFDOFDQWLOHYHUIRUWZRIL[HG
VXSSRUWVDQGKG EGFDQWLOHYHUVIRUIL[HGDQGUROOHUVXSSRUWVDQGKG DQG
UHVSHFWLYHO\HµEULGJH¶SUREOHPIRUXQLIRUPO\GLVWULEXWHGORDGLQJVDORQJVSDQDQGIRU
SLQUROOHUVXSSRUWVIH[DPSOHRIWKHRSWLPDOOD\RXWZLWKLQDFRQFDYHGRPDLQ
/HWXVGLVFXVVVHOHFWHGVROXWLRQVWRWKH0LFKHOOSUREOHPJLYHQLQ)LJ,Q

DOOH[DPSOHVWKHORDGLVRQHFRQFHQWUDWHGIRUFH&RQVHTXHQWO\WKHRSWLPDOVWUXFWXUHV
EHFRPHUHLQIRUFHGE\EDUVRIILQLWHFURVVVHFWLRQVWRHTXLOLEUDWHWKHQRGHZKHUHWKH
SRLQWORDGLVDSSOLHG7KHDSSHDUDQFHRIWKHEDUVRIILQLWHFURVVVHFWLRQVSUHYHQWV
IURP VWUHVV VLQJXODULWLHV $OO WKH 0LFKHOO WUXVVHV VKRZQ LQ )LJ DUH GLV
FUHWHFRQWLQXRXV VWUXFWXUHV KHQFH WKH\ DUH QHLWKHU SODWHV QRU IUDPHV QRU WUXVVHV
7KHLUWKHRU\H[WHQGVWKHFODVVLFDOIUDPHZRUNRIERWKOLQHDUHODVWLFLW\DQGVWUXFWXUDO
PHFKDQLFV7KHLQWHUQDOPLFURVWUXFWXUHDVZHOODVWKHZKROHJHRPHWU\RIWKHVROXWLRQ
GHSHQGVRQWKHGDWDDQGLWLVZRUWKQRWLQJWKDWWKHOD\RXWVFDQMXPSVXGGHQO\WKH
GHSHQGHQFHRIWKHOD\RXWRQWKHGDWDLVQRWFRQWLQXRXV)RULQVWDQFHWKHSUREOHPVLQ
)LJFDQGGDUHVLPLODUEXWWKHUHVXOWLQJOD\RXWVGLIIHUFRQVLGHUDEO\
7KH VROXWLRQV RI )LJ KDYH EHHQ REWDLQHG QXPHULFDOO\ E\ XVLQJ WKH DGDSWLYH
JURXQGVWUXFWXUHDSSURDFKGHYHORSHGLQ6RNyáE7KHOD\RXWVSUHVHQWHGLQ)LJV
DHUHIHUWRWKHIHDVLEOHGRPDLQEHLQJWKHKDOISODQHRYHUWKHOLQHOLQNLQJWKHVXS
SRUWV +HUH G UHSUHVHQWV WKH GLVWDQFH EHWZHHQ WKH VXSSRUWV DQG K LV WKH YHUWLFDO
RUGLQDWHRIWKHSRLQWRIWKHDSSOLHGORDGPHDVXUHGZLWKUHVSHFWWRWKHOLQHOLQNLQJWKH
VXSSRUWV7KHIHDVLEOHGRPDLQRIWKHH[DPSOHRI)LJILVPRUHFRPSOH[±LWLVD
FRQFDYHSRO\JRQPDUNHGE\WKHGDVKHGOLQHIRUPRUHGHWDLOVRQWKHODWWHUH[DPSOH
WKHUHDGHULVUHIHUUHGWR/HZLĔVNLHWDO
D E
9Q 3G V 9D 3G V

)LJXUH&RPSDULVRQRIQXPHULFDODQGH[DFWVROXWLRQVRISUREOHPRI)LJG
7KHH[DPSOHVRI)LJGHPRQVWUDWHDKLJKVHQVLWLYLW\RIWKHRSWLPDOWRSRORJ\WR
WKHFKRLFHRIERXQGDU\ FRQGLWLRQV WKXV LQGLFDWLQJ GLIILFXOWLHV LQ FRQVWUXFWLQJ ERWK
KLJKDFFXUDF\ QXPHULFDO VROXWLRQV DQG H[DFW DQDO\WLFDO VROXWLRQV WR WKH FODVV RI
0LFKHOO SUREOHPV GLVFXVVHG 6RPH RI WKHVH GLIILFXOWLHV KDYH EHHQ RYHUFRPH RQO\
UHFHQWO\,QSDUWLFXODUWKHKLJKDFFXUDF\OD\RXWVLQ)LJVEGHIDUHSXEOLVKHG
KHUHIRUWKHILUVWWLPH2QWKHRWKHUKDQGWKHH[DFWVROXWLRQVRI)LJVDFKDYHEHHQ
FRQVWUXFWHG PXFK HDUOLHU FI +HPS /HZLĔVNL HW DO 7KH SUREOHP RI
)LJGFRQFHUQLQJWKHFDVHRIWKHSRLQWORDGO\LQJDWWKHPLGGOHSRLQWRIWKHVTXDUH
RIWKHVLGHGRUIRUKG KDVEHHQUHFHQWO\VROYHGDQDO\WLFDOO\E\WKHSUHVHQW
DXWKRUVWKHGHULYDWLRQRIWKHVROXWLRQKDVQRW\HWEHHQSXEOLVKHG%HORZWKHILQDO
UHVXOWV RQO\ DUH SUHVHQWHG VHH )LJ VKRZLQJ JRRG DJUHHPHQW EHWZHHQ WKH QX
PHULFDODQGDQDO\WLFDOOD\RXWV7KHQXPHULFDOHVWLPDWLRQRIWKHRSWLPDOYROXPHGRHV
QRWH[FHHGRIWKHUHODWLYHHUURU
.QRZLQJWKHH[DPSOHVLQ)LJZHVKDOOEHEHWWHUSUHSDUHGWRVWXG\WKHH[DFW
VROXWLRQV WR WKH SUREOHP DQG WR WKH SUREOHP GXDO WR LW RXWOLQHG LQ WKH QH[W
VHFWLRQ
7KH3UREOHPDVD3UREOHP'XDOWRD6WUHVVEDVHG)RU

PXODWLRQ'&DVH
%\XVLQJWKHPHWKRGRI6WUDQJDQG.RKQZHVKDOOILQGWKHSUREOHPIRUZKLFK
WKHSUREOHPLVGXDO,QIRUWKH'FDVHWKHVHW%V&V7KDVWKHIRUP
%V & V 7 ^ İ(
V H ,, t V V & H , d V V 7 `
/HW1EHWKHILHOGRIWKHLQSODQHVWUHVVUHVXOWDQWVVXFKWKDW

1 / : ( V GLY 1 / : 5
DQGWKLVILHOGLVOLQNHGZLWKWKHORDGE\WKHYDULDWLRQDOHTXDWLRQRIHTXLOLEULXP
³ 1 İ Y G[ I Y Y + :
:
6XFKILHOGVIRUPDVHW6:RIVWDWLFDOO\DGPLVVLEOHVWUHVVUHVXOWDQWV7KHQRWDWLRQ1LV
VWRLQGLFDWHWKDWZHGHDOZLWKDSODWH:HVKDOOSURYHWKDWWKHRSWLPDO
XVHGQRWV
YROXPHJLYHQE\FDQEHDOWHUQDWLYHO\H[SUHVVHGE\
° § V V · ½°
90 PLQ ® ³ ¨¨ 1 , 1 ,, ¸¸G[ 1 6: ¾
V °̄ : © V 7 V& ¹ °¿
:HFRQVLGHUVXFKSUREOHPVLQZKLFKWKHSULQFLSDOYDOXHVRI1VDWLVI\1,tDQG
1,,d /HW XV UHZULWH WKH DERYH H[SUHVVLRQ E\ LQWURGXFLQJ WKH HTXLOLEULXP FRQ
VWUDLQWVLQWRWKHPDLQWHUP
° § V V · ½°
90 PLQ PD[ ® ³ ¨¨ 1 , 1 ,, ¸¸G[ I Y ³ 1 İY G[ ¾
V 1/ Y+ : °̄ : © V 7

: ( V V& ¹ °¿
GLY1/ : 5
:
:HLQWHUFKDQJHWKHVHTXHQFHRIPLQDQGPD[RSHUDWLRQVWRDUULYHDW

90 PD[ ^ I Y 5 İY `
V Y+ :
ZLWK
§ V V ·
5 İ PLQ ³ ¨¨© V 7 1, 1 1 İ Y ¸¸G[

1/ : ( V V & ,, ¹

GLY1/ : 5 :
1RZZHPDNHXVHRIWKHUHSUHVHQWDWLRQRIWKHVFDODUSURGXFWRI1DQGHHGHULYHGLQ
/HZLĔVNL
1İ 1 , İ, 1 ,, İ ,, VLQ M H , H ,, 1 , 1 ,,
ZKHUH MLVDQDQJOHEHWZHHQILUVWSULQFLSDOGLUHFWLRQVRIWHQVRUV1DQGHH7KXVWKH
RSHUDWLRQPLQ5HHLPSOLHVMEHLQJRUS7KHQWKHSULQFLSDOGLUHFWLRQVRI1DQGHH
FRLQFLGH0RUHRYHU
§§ V · § V · ·
5 İ PLQ ³ ¨¨ ¨¨© V H , ¸¸ 1 , ¨¨ H ,, ¸¸ 1 ,, ¸G[
¸
© V&

1/ : ( V

GLY1/ : 5 : © 7 ¹ ¹ ¹
RU
V V
° LI H, DQG H ,, D
° V7 V&
° V V
5 İ ® LI H, DQG H ,, ! E
° V7 V&
° f LI V V
° H, ! DQG H ,, F
¯ V7 V&
0LQLPXPRYHU1LVDWWDLQHGLI
D 1, z DQG 1 ,, z
E 1, DQG 1 ,,
F 1, f RU 1 ,, f
7KXV WKH RSHUDWLRQ PD[ LQ UHDUUDQJHV WKLV SUREOHP WR WKH IRUP 7KLV
SURYHVWKDWWKHSUREOHPLVGXDOWR
7KHFRQGLWLRQHH%V&V7DORQJZLWKWKHFROLQHDULW\RIWKHSULQFLSDOGLUHFWLRQV
RI 1 DQG H DQG DORQJ ZLWK DQG GHWHUPLQH WKH PLFURVWUXFWXUHV RI WKH
RSWLPDOSUREOHP7KHVXEGRPDLQVRIWKHVROXWLRQFRUUHVSRQGLQJWRWKHLQWHULRURIWKH
ORFNLQJORFXVVHW%V&V7ZLOOEHQDPHG2GRPDLQV7KHUHQREDUVRFFXUVHHWKH
QRWDWLRQLQ)LJ

)LJXUH 7KH ORFNLQJ ORFXV %V&V7 GHVFULSWLRQ 'LYLVLRQ RI WKH GRPDLQ _H,_dVV7
_H,,_dVV&H,tH,,LQWRWKHUHJLRQV7652
7KH ILQDO GHVLJQ RI HDFK 0LFKHOO SUREOHP LV GLYLGHG LQWR VXEGRPDLQV FRUUH
VSRQGLQJWRVRPHVHJPHQWVRIWKHERXQGDU\RIWKHORFNLQJORFXV7KXVLWLVKHOSIXOWR
LQWURGXFHVSHFLDOV\PEROVWRFRUUHODWHWKHVROXWLRQDWHDFKSRLQWZLWKDSDUWRIWKH
ERXQGDU\RIWKHORFNLQJORFXV7RWKLVHQGZHVKDOOPDNHXVHRI5R]YDQ\
VQRWDWLRQ
LQWURGXFHGLQWKHHDUO\
VVHH5R]YDQ\DQG$GLGDPEDQG5R]YDQ\
7KHGRPDLQVQDPHG5RUWKHGRPDLQVZKHUH H, VV7DQG H,,!VV&RUWKHVH
GRPDLQVZKHUH H,VV7DQG H,, VV&DUHILOOHGXSZLWKEDUVJRLQJLQRQHGL
UHFWLRQ7KHGRPDLQVQDPHG7ZKHUHH, VV7DQGH,, VV&DUHILOOHGXSZLWK
WZR IDPLOLHV RI PXWXDOO\ RUWKRJRQDO ILEUHV VHH )LJ 7KH PDMRULW\ RI 0LFKHOO¶V
VROXWLRQVFRUUHVSRQGWRWKHFDVHRIV7 V& V7KHPHWKRGVRIFRQVWUXFWLRQRIVXFK
VROXWLRQVDUHRXWOLQHGLQWKHQH[WVHFWLRQ
7KHGLVFXVVLRQDERYHVKRZVWKDWWKH0LFKHOOSUREOHPLVVLPXOWDQHRXVO\JRYHUQHG
E\WZRPXWXDOO\GXDOIRUPXODWLRQVVWDWLFDQGNLQHPDWLF%RWKWKHVHWWLQJV
FDQ EH YLHZHG DV D ORFNLQJ PDWHULDO HTXLOLEULXP SUREOHP LQ :LOOLDP 3UDJHU
V
PHDQLQJ7KHORFNLQJPDWHULDODSSURDFKWRWKHOD\RXWRSWLPL]DWLRQSUREOHPV
LVWKHOHLWPRWLYRIWKHERRNE\5R]YDQ\RQIOH[XUDOV\VWHPVDQGLVDOVRSUHVHQW
LQKLVQXPHURXVSDSHUVRQRSWLPDOWUXVVHV$QLPSRUWDQWFRQWULEXWLRQLVGXHWR6WUDQJ
DQG.RKQZKRVKRZHGH[SOLFLWO\KRZWRGHULYHWKHNLQHPDWLFVHWWLQJRIWKH
0LFKHOOSUREOHPIURPWKHVWDWLFRQH2WKHUOLQNVEHWZHHQWKHORFNLQJPDWHULDOWKHRU\
DQGOD\RXWRSWLPL]DWLRQDUHGLVFXVVHGLQ/HZLĔVNLDQG7HOHJD
3ODQH 0LFKHOO )UDPHV LQ WKH &DVH RI (TXDO 3HUPLVVLEOH
6WUHVVHVLQ7HQVLRQDQG&RPSUHVVLRQ*HRPHWU\RI7UDMHFWR
ULHVRI3ULQFLSDO6WUDLQV
,QWKHFDVHFRQVLGHUHGV7 V& VWKHSUREOHPUHGXFHVWR
½
PLQ ® ³ 1 , 1 ,, G[ 1 6: ¾

90
V ¯: ¿
DQGWKHSUREOHPGXDOWRLWUHDGV

90 PD[ I Y
V Y+ :
İ Y [ %IRUDH [:

ZLWK
% ^İ (
V H , d H ,, d `
7KXVWKHORFNLQJORFXVLVDVTXDUHVHH)LJ

)LJXUH'LYLVLRQRIWKHGRPDLQ_H,_d_H,,_dH,tH,,LQWRWKHUHJLRQV7652
$SDUWIURPWKHGRPDLQV257ZHLQWURGXFHWKHGRPDLQV6ZKHUHH, H, RU

H, H, 7KHUHWKHOD\RXWRIEDUVLVDUELWUDU\7KHPRVWLPSRUWDQWUROHDUHSOD\HG
E\WKHGRPDLQV7ZKHUHZHKDYHWKHVWURQJHVWLQIRUPDWLRQRQWKHJHRPHWU\RIWKH
ILEUHV
H, H ,,
DWHDFKSRLQW[RIDVXEGRPDLQ:7RIWKHGRPDLQ:
7KHUHDUHVROXWLRQVLQZKLFK:7 :7KHFRQGLWLRQVGHWHUPLQHWKH+HQFN\

QHWVRUVOLSOLQHVRIULJLGSODVWLFLW\VHH+LOO7KHOLWHUDWXUHRQ+HQFN\QHWVLV
ULFK7KHPRVWLPSRUWDQWJHRPHWULFSURSHUWLHVRIWKH+HQFN\QHWVKDYHEHHQRULJL
QDOO\IRXQGE\&DUDWKpRGRU\DQG6FKPLGW7KHDQDO\WLFDOPHWKRGVXVHGE\
+HPSGRQRWJREH\RQGWKHWRROVRIWKLVROGSDSHU+RZHYHUWKHUHLVDORQJ
ZD\IURPWKHPDWKHPDWLFDOUHVXOWVIRXQGLQWKLVSDSHURIWRWKHVROXWLRQVRIWKH
0LFKHOO SUREOHPV UHSRUWHG E\ +HPS EDVHG SDUWO\ RQ WKH SDSHUV E\
$6/&KDQ +6<&KDQ DQG ODWHU IRXQG E\
+6<&KDQ,QIDFWHDFKDQDO\WLFDOVROXWLRQVKRXOGEHJXHVVHGDQGWKHQRQH
VKRXOGFKHFNLWVFRUUHFWQHVVWKHUHLVQRGLUHFWZD\IURPWKHGDWDWRWKHH[DFWUHVXOW
7KHHDVLHVWJXHVVVWDUWVIURPWKHWKHRU\RI+HQFN\QHWFKDUDFWHUL]HGE\

)LJXUH$QRUWKRJRQDOFXUYLOLQHDUV\VWHPRISDUDPHWULFOLQHVWKHEDVLVYHFWRUVDWDUEL
WUDU\SRLQWGHILQLWLRQRIWKHDQJOHI
,PDJLQHWZRIDPLOLHVRIPXWXDOO\RUWKRJRQDOOLQHVJLYHQE\WKHPDSSLQJ
[ [ o U >[ [ [ [ [ [ @

ZKHUH[ [[ \7KHQZHLQWURGXFHWKHEDVLVYHFWRUV J L wU w[ L WDQJHQWWRWKH

OLQHV [ L WKHPHWULFWHQVRUEHLQJ J L J M J LM :HDVVXPHRUWKRJRQDOLW\J /HW
XVLQWURGXFHWKH/DPpFRHIILFLHQWV
$[ [ J [ [ % [ [ J [ [
DQG WKH FRQWUDYDULDQW FRPSRQHQWV JLM RI WKH PHWULF WHQVRU 7KLV VWDUWLQJ SRLQW LV
VLPLODUWRWKDWZHUHPHPEHUIURPOHFWXUHVRQGLIIHUHQWLDOJHRPHWU\DQGRQWKHWKHRU\
RIVKHOOVZKHUHXVXDOO\WKHPDLQWHUPVRIWKHVXUIDFHJHRPHWU\DUHUHFDOOHGVHHHJ
1DJKGLDQG1RYR]KLORY7KH/DPpFRHIILFLHQWVDUHQRWLQGHSHQGHQW
DVWKH\GHWHUPLQHWKH*DXVVFXUYDWXUH
ª w § w% · w § w$ · º
. « »
$% ¬ w[ ¨© $ w[ ¹¸ w[ ¨© % w[ ¹¸ ¼
ZKLFKPXVWYDQLVKIRUWKHSODQHDWHDFKSRLQWRIWKHSODQHGRPDLQFRQVLGHUHG
/HW Y YL J L EHWKHGLVSODFHPHQWYHFWRU7KHVWUDLQFRPSRQHQWVUHIHUUHGWRWKH
FXUYLOLQHDUV\VWHPFRQVLGHUHGDUH
HLM

YL_ M Y M_L

ZKHUH YL M LVWKHFRYDULDQWGHULYDWLYHGHILQHGE\
wYL
YL _ M *LMN YN
w[ M
ZKHUHWKH&KULVWRIIHOV\PEROVFDQEHH[SUHVVHGE\WKH/DPpFRHIILFLHQWVDVIROORZV
w$ $ w$ w$
* *

*

*
$ w[ % w[ $ w[

w% % w% w%
*

*

* *
% w[ $ w[ % w[
:HLQWURGXFHWKHSK\VLFDOFRPSRQHQWVRIWKHYHFWRUYE\UHIHUULQJWKHPWRWKHEDVLV
YHFWRUVRIXQLWOHQJWK+HQFHZHGHILQHWKHPE\

X Y Y Y
$ %
DQGVLPLODUO\LQWURGXFHWKHSK\VLFDOFRPSRQHQWVRIWHQVRUHHE\XVLQJWKHUXOHV

H H H H H H
$ % $%
7KHILHOGVXYGHQRWHGLVSODFHPHQWVDORQJWKH[[OLQHVHHGHQRWHWKHUHODWLYH
HORQJDWLRQVRIWKHILEUHVDORQJWKHVDPHOLQHV:HGHILQHWKHGHIRUPDWLRQPHDVXUHVRI
VKHDUZZDVWKHSK\VLFDOFRPSRQHQWVRIWKHJUDGLHQWVY_Y_UHVSHFWLYHO\

Z Y_ Z Y_
$% $%
$VLQ+HPSZHXVHWKHQRWDWLRQ[ D[ E:HH[SUHVVQRZWKHSK\VLFDO
FRPSRQHQWV H H Z ZE\PHDQVRIWKHSK\VLFDOFRPSRQHQWVXYRIWKHGLV

SODFHPHQWYHFWRU
§ wX w$ · § wY w% ·
H ¨ Y ¸ H ¨ X ¸
$ ¨© wD % wE ¸¹ % ¨© wE $ wD ¸¹
§ wY w$ · § wX w% ·
Z ¨ X ¸ Z ¨ Y¸
$ ¨© wD % wE ¸¹ % ¨© wE $ wD ¸¹
:HH[SUHVVWKHSK\VLFDOFRPSRQHQWVRIWKHVKHDUPHDVXUHHE\ZZ

H Z Z

DQGUHFDOOWKHH[SUHVVLRQIRUWKHULJLGURWDWLRQ

Z Z Z

DURXQGWKHQRUPDOWRWKHSODQH
7KHLQWURGXFWLRQRIWKHSK\VLFDOFRPSRQHQWVLVLQGLVSHQVLEOHWRPDNHXVHFRU
UHFWO\ RI WKH FRQGLWLRQV UHIHUUHG WR WKH SK\VLFDO EDVLV /HW XV GLVFXVV WKH
FRQVHTXHQFHV RI RUWKRJRQDOLW\ RI WKH SDUDPHWULF OLQHV [ [ /HW XV ZULWH
U >[ D E \ D E @ 5HFDOOWKDWWKHYHFWRUVRIWKHEDVLV J J DUHJLYHQE\
wU wU
J J
wD wE
$FFRUGLQJWRZHKDYH

§ w[ · § w\ · § w[ · § w\ ·
¨ ¸ ¨ ¸ $D E ¨¨

¸¸ ¨¨ ¸¸ % D E
© wD ¹ © wD ¹ © wE ¹ © wE ¹
ZKLOHWKHRUWKRJRQDOLW\FRQGLWLRQ JJ JLYHV
w[ w[ w\ w\

wD wE wD wE
:HLQWURGXFHWKHXQNQRZQV
w[ w\ w[ w\
] ] ] ]
$ wD $ wD % wE % wE
7KHHTXDWLRQVDUHZULWWHQDV
] ] ] ]

] ] ] ]
7KXVWKHUHH[LVWWKHDQJOHV\MVXFKWKDW
] FRV\ ] VLQ\ ] FRV M ] VLQ M FRV M \
/HW\ I M I S 7KHQ
] FRV I ] VLQ I ] VLQ I ] FRV I
/HW XV FRPSXWH J H J H FRVJ H $ FRVJ H DQG QRWH WKDW
J H w[ wD 7KXV FRVI FRVJ H ZKLFKFOHDUVXSWKDW I J H :H
ZULWHDJDLQ
w[ w[ w[ w\
FRV I VLQ I VLQ I FRV I
$ wD % wE $ wD % wE
7KHIXQFWLRQV$DQG%DUHQRWLQGHSHQGHQWVLQFHWKH*DXVVFXUYDWXUHYDQLVKHV
w § w% · w § w$ ·
¨ ¸ ¨ ¸
wD © $ wD ¹ wE © % wE ¹
/HWXVZULWHWKHIRUPXODIRUWKHGLIIHUHQWLDOV
w[ w[ w\ w\
G[ GD GE G\ GD GE
wD wE wD wE
/HWXVLQVHUWWRILQG
G[ $ FRV I GD % VLQ I G E

G\ $ VLQ I G D % FRV I G E
ZKHUH I I D E +DYLQJ I I D E $D E % D E RQH FDQ UHFRYHU WKH

VKDSHVRIWKHSDUDPHWULFOLQHVE\DGLUHFWLQWHJUDWLRQRI7KHHTXDWLRQV
LPSO\WKHLGHQWLWLHV
w [ w w
$ FRV I % VLQ I
wDwE wE wD

w \ w w
$ VLQ I % FRV I
wDwE wE wD
)URPWKHVHHTXDWLRQVZHLQIHU
ª w$ wI º
« %
ªFRV I VLQ I º wE wD » ªº
« VLQ I » « » «»
¬ FRV I ¼ « w% wI » ¬ ¼
« wD $ wE »
¬ ¼
ZKLFKOHDGVWRWKH]HURVROXWLRQ
w$ wI w% wI
% $
wE wD wD wE
/HWXVUHZULWHWKHIRUPXODHIRUWKHFRPSRQHQWVRIVWUDLQZLWKXVLQJ
§ wX wI · § wY wI ·
H ¨ X ¸ H X¸
$ © wD wD ¹ % ¨© wE wE ¹

§ wY wI · § wX wI ·
Z ¨ X ¸ Z Y
$ © wD wD ¹ % ¨© wE wE ¹¸
7KHDERYHHTXDWLRQVLPSO\
G X LX L X LX G I $ H LZ G D % Z LH G E
%\XVLQJWKHLGHQWLW\
GI LIGI H LI G HLI I
ZHUHDUUDQJHWRWKHIRUP
G ¬ª HLI X LY ¼º HLI ª¬ $ H LZ GD % Z LH G E º¼
7KXVLIZHNQRZWKHPHDVXUHVHHZZZHFDQUHFRYHUWKHGLVSODFHPHQWVXY
E\DSSURSULDWHLQWHJUDWLRQDORQJWKHSDUDPHWULFOLQHVDE7KHUHODWLRQV
UHIHUWRDOOSDUDPHWHUL]DWLRQVZKLFKDUHRUWKRJRQDORQWKHSODQH7KHVH
HTXDWLRQVFDQEHIRXQGLQW\SLFDOERRNVRQWKHVKHOOWKHRU\VLQFHWKHUHDFXUYLOLQHDU
SDUDPHWHUL]DWLRQFDQQRWEHDYRLGHG,WLVPRUHGLIILFXOWWRILQGWKHVHHTXDWLRQVLQWKH
ERRNVRQVROLGPHFKDQLFVZKHUHHLWKHUDQDEVROXWHQRWDWLRQLVXVHGRUWKHHTXDWLRQV
DUHH[SUHVVHGE\XVLQJWKHFRYDULDQWGHULYDWLYHVDQGRWKHUW\SLFDOWRROVRIWKHWHQVRU
FDOFXOXV7KHSDVVDJHIURPWKHWHQVRULDOWRSK\VLFDOFRPSRQHQWVLVVKRZQRQO\LQWKH
VSHFLDOFDVHVRISRSXODUSDUDPHWHUL]DWLRQVOLNHSRODUFRRUGLQDWHVLQSODQHVSKHULFDO
FRRUGLQDWHVLQVSDFHHWF7KHH[FHSWLRQDUHROGHUWH[WERRNVOLNHWKHIDPRXVERRN
E\ /RYH RU WKH OHVV NQRZQ ERRN E\ 9=9ODVRY RQ WKH WKHRU\ RI
VKHOOV ZKHUH D JHQHUDO GHVFULSWLRQ LQ WKH FXUYLOLQHDU FRRUGLQDWHV LV XVHG ZLWK DQ
H[SOLFLWXVDJHRI/DPpILHOGV-XVWWKLVOHYHORIGHVFULSWLRQLVQHFHVVDU\WRSXWIRUZDUG
WKHWKHRU\RI0LFKHOOVWUXFWXUHV,QSDUWLFXODUWKH/DPpFRHIILFLHQWV$DE%DE

DUHILHOGVZKLFKVKRXOGH[SOLFLWO\DSSHDULQWKHOHFWXUHVLQFHWKH\EHORQJWRWKHVHWRI
WKHXQNQRZQVRIWKHRSWLPXPGHVLJQSUREOHP
:H LPSRVH WKH RSWLPDOLW\ FRQGLWLRQV ZLWKLQ WKH IUDPH RI DE SDUDPH
WHUL]DWLRQRU
H H Z Z Z Z
DW HDFK SRLQW DE RI WKH GRPDLQ 7KH ODVW WZR HTXDWLRQV DVVXUH WKDW WKH VKHDU
YDQLVKHV
7KHDERYHFRQGLWLRQVUHIHUQRWRQO\WRWKHILHOGVXDEYDEVLQFHWKHQXPEHU
RIWKHFRQGLWLRQVLVWKUHHZKLOHWKHQXPEHURIGLVSODFHPHQWVLVWZR7KXVWKH
HTXDWLRQVLQWURGXFHRQHHTXDWLRQFKDUDFWHUL]LQJWKHIDPLO\RIOLQHVDE/HW
XVLQVHUWLQWR
G ª¬ HLI X LX º¼ HLI ¬ª $G D L%G E LZ $G D L%G E ¼º
RU
G ª¬ HLI X LX º¼ HLI ª¬ $ LZ G D % L Z G E º¼
7KHULJKWKDQGVLGHRILVDFRPSOHWHGLIIHUHQWLDOLIWKHIROORZLQJHTXDOLW\KROGV

w LI w LI
ª H $ LZ º¼ ª H % L Z º¼
wE ¬ wD ¬
+HQFHZHILQGXVLQJ
wZ wI wZ wI

wD wD wE wE
ZKLFKLPSOLHV
w I

wD wE
7KLVLVMXVWWKHIDPRXVHTXDWLRQFKDUDFWHUL]LQJWKHVKDSHVRIWKHSDUDPHWULFOLQHV
VXFKOLQHVDUHFDOOHG+HQFN\QHWVVHH+LOO2QHFDQSURYHWKDWWKHIROORZLQJ
IRUPRIWKHVROXWLRQRI
I D E E D
LVQRWDUHVWULFWLRQVHH&DUDWKpRGRU\DQG6FKPLGW(T/HWXVLQVHUW

LQWR
w$ w%
% $
wE wD
DQG±LQWR
wZ wZ

wD wE
KHQFH
Z D E Z ZKHUH Z Z
6XEVWLWXWLRQRILQWRJLYHVWKHIRUPXODHIRUGLVSODFHPHQWV

[
X [ K XT [ K ³ ª¬ FRV [ D Z D K VLQ [ D º¼ $ D K G D
T
[

Y [ K YT [ K ³ ª¬VLQ [ D Z D K FRV [ D º¼ $ D K G D
T
ZKHUH
XT [ K FRV [ T X T K VLQ [ T Y T K

YT [ K VLQ [ T X T K FRV [ T Y T K
,QWKHLQWHJUDWLRQLVSHUIRUPHGDORQJWKHOLQHVD6LPLODUIRUPXODHRILQWHJUD

WLRQDORQJWKHOLQHVEFDQEHDOVRSXWIRUZDUG
7KHHTXDWLRQVOHDGWRWKHJRYHUQLQJHTXDWLRQV
/$ /%
ZKHUH
w
/
wD wE
+DYLQJ$%RQHFDQILQGXYXSWRDFRQVWDQWZ/HWXVWDNHLQWRDFFRXQWLQ

wX wX
X $ X %
wD wD
§ wY · § wX ·
Z ¨ X ¸ Z ¨ X ¸
$ © wD ¹ % © wE ¹
/HWXVGHILQHWKHDX[LOLDU\ILHOGV
X X D $ X X E %
7KHHTXDWLRQVLPSO\

/X % /X $
ZKLOHWKHILHOGVXYVDWLVI\
/ X / Y
2QHQRWHVWKDWWKHHTXDWLRQVFDQEHVROYHGILUVWDQGWKHQWKHGLVSODFHPHQW

ILHOGVXYFDQEHFRQVWUXFWHGE\.QRZLQJWKH/DPpILHOGVZHFDQFRPSXWHWKH
FRRUGLQDWHV[\E\LQWHJUDWLQJDORQJWKHDOLQHVWRDUULYHDW
O
[ O P [ T P ³ FRV P D $ D P GD
T
O

\ O P \ T P ³ VLQ P D $ D P GD
T
,QVRPHSUREOHPVRQHVKRXOGFRQVWUXFWWKHQHWQRWNQRZLQJWKHILHOGV$%7KHQ
ZHFDQXVHWKHIROORZLQJWRROZHLQWURGXFHWKHDX[LOLDU\YDULDEOHV
[ [ FRV I \ VLQ I

\ [ VLQ I \ FRV I
ZKHUH [ [ D E \ \ D E I E D 2QHFDQSURYHWKDW
w[ w\
\ [ /[ /\
wE wD
&RQFOXGLQJWKHJRYHUQLQJHTXDWLRQRIK\SHUEROLFW\SH
w I
I
wD wE
LVVDWLVILHGE\WKHIROORZLQJIXQFWLRQV I I D E
[ D E \ D E $ D E % D E X D E X D E
&RQFOXGLQJ ZH QRWH WKH +HQFN\ QHWV GHWHUPLQH WKH OD\RXW RI ILEUHV RI WKH
0LFKHOOVWUXFWXUHVWKH\DUHWKHSDUDPHWULFOLQHVRIDFXUYLOLQHDURUWKRJRQDOV\VWHP
JLYHQE\[ [DE\ \DE7KHWDQJHQWOLQHVWRWKHDOLQHVPDNHDQJOHV IZLWK
WKH D[LV [ DQG WKLV DQJOH LV GHWHUPLQHG E\ WKH YDOXHV RI WKH SDUDPHWHUV ED E\
I ED7KHRSWLPDOLW\FRQGLWLRQVUHIHUULQJWRWKH7UHJLRQV)LJLPSRVH

UHVWULFWLRQVRQWKHVKDSHRIWKHOLQHVEDRURQWKHOD\RXWVRI/DPHILHOGV$DQG%
DV ZHOO DV RQ WKH OD\RXWV RI YLUWXDO GLVSODFHPHQWV XY 7KH IXQFWLRQV DUH
JRYHUQHGE\WKHVDPHK\SHUEROLFGLIIHUHQWLDOHTXDWLRQ$OOWKH NQRZQ DQD
O\WLFDOVROXWLRQVRIWKH0LFKHOOSUREOHPUHIHUWRWKHHTXDWLRQ/HWXVGHILQH
DIWHU&DUDWKpRGRU\DQG6FKPLGWDIXQFWLRQEHORQJLQJWRWKHFODVVRI%HVVHO
IXQFWLRQVDVIROORZV
f
]Q
- ] ¦

Q Q
/HWXVQRWHWKDWIDE -DEVDWLVILHV7KHSURRILVVWUDLJKWIRUZDUGKHQFHLV

RPLWWHG7KHIXQFWLRQ-LVOLQNHGZLWKWKH%HVVHOIXQFWLRQ,E\
, ] - ]
2QH FDQ VKRZ PDQ\ IXQFWLRQV VROYLQJ WKH HTXDWLRQ +6<&KDQ KDV FRQ
VWUXFWHGWKHIROORZLQJIDPLO\RIVXFKIXQFWLRQV
Q
*Q D E
§ D ·
¨¨ ¸¸ , Q DE DQ¦
f
DE N
©E ¹ N N N Q
f
)Q D E ¦ P *P Q D E
P
ZKHUH Q r r VHH&KDQ/HZLĔVNLHWDODDQG
DE ¦ DE
Q f N
, Q DE
N N Q N
LVDPRGLILHG%HVVHOIXQFWLRQ7KHFRQYHUJHQFHRIWKHVHULHVIRUVPDOODUJXPHQWVLV
YHU\IDVW7KHVHIXQFWLRQVDUHOLQNHGZLWK/RPPHOIXQFWLRQVWKHWKHRU\RIZKLFKLV
DYDLODEOH LQ WKH WH[WERRNV RQ %HVVHO IXQFWLRQV 2QH FDQ SURYH WKDW WKH IXQFWLRQV
VDWLVI\
w+ Q E D w+ Q E D
+ Q E D + Q E D
wD wE
ZKHUH+ *RU+ )7KXVWKHIXQFWLRQV+QDEVDWLVI\
7KH 5LHPDQQ 3UREOHP DQG LWV 6ROXWLRQ E\ WKH 0HWKRG RI
&DUDWKpRGRU\DQG6FKPLGW
:HFRQVLGHUWKHSUREOHPRI5LHPDQQW\SHJLYHQWKHYDOXHVZTEZDTRID
IXQFWLRQZDEDORQJWKHOLQHVD TDQGE TDQGKDYLQJWKHYDOXHZTTILQG
ZOPNQRZLQJWKDWZVDWLVILHVLQWKHGRPDLQD!TE!T
7KHVROXWLRQKDVDQH[SOLFLWIRUPVHH&DUDWKpRGRU\DQG6FKPLGW
Z O P Z T T * O T P T
O
wZ D T P
wZ T E
³ * O D P T
T
wD
G D ³ * O T P E
T
wE
GE
ZKHUHDFFRUGLQJWR
* D E
, DE
RUDFFRUGLQJWR
* D E -DE
ZKHUH-LVJLYHQE\:HQRWHWKDWWKHIXQFWLRQSOD\VWKHUROHRID*UHHQ

IXQFWLRQ7KHSURRIRILVGHOLYHUHGLQ/HZLĔVNLHWDODKHQFHFDQEH
RPLWWHG
:HFRQFOXGHWKDWWKHYDOXHVZDORQJWKHOLQHVTEDTGHWHUPLQHWKHYDOXHV
ZOPIRUO!TP!T

)LJXUH7KH5LHPDQQSUREOHP
([DPSOH)LQGDIXQFWLRQZVDWLVI\LQJZKLFKIXOILOVWKHERXQGDU\FRQGLWLRQV

ZD UDQGZE UEU+HUHUDQGUDUHNQRZQSRVLWLYHQXPEHUV

)LJXUH7KHERXQGDU\FRQGLWLRQVIRUWKHLOOXVWUDWLYH5LHPDQQSUREOHP
7KXVZHKDYH
wZ E
Z D U Z E U E U U
wE
:HQRWHWKDW Z U DQG
wZ D
D !
wD
:HPDNHXVHRIIRU T T
O
wZ D
Z O P Z * O P ³ * O D P GD

wD

P
wZ E
³ * O P E wE G E
ZKLFKUHGXFHVWR
P
Z O P U* O P U ³ * O P E G E

6LQFH * O P E * P E O DQGPDNLQJXVHRIZHILQG

w* P E O
* P E O
wE
KHQFHZHLQWHJUDWH
P P
w* P E O E P

³ * O P E G E

³

wE
GE * P E O E

ª¬* O * P O º¼ * P O
DQGREWDLQ
ZO P U* O P U* P O
ZKLFKLVWKHILQDOUHVXOW2QHFDQHDV\FKHFNWKDWWKLVIXQFWLRQVDWLVILHVDOOWKHFRQGL
WLRQVUHTXLUHG
7KHUHVXOWLVRIYLWDOLPSRUWDQFHLQWKHWKHRU\RI0LFKHOOVWUXFWXUHVVLQFH
WKLVUHVXOWGHVFULEHVWKHOD\RXWRI/DPHIXQFWLRQ$OPLQWKHGRPDLQJHQHUDWHGE\
WZRFLUFOHVRIGLIIHUHQWUDGLLVHH)LJ

)LJXUH5LHPDQQSUREOHPSRVHGRQWZRFLUFOHVRIGLIIHUHQWUDGLL
,QGHHGQRWHWKDW$D UVLQFHWKHOHQJWKRIHOHPHQWDU\DUFRQ$&LVHTXDOUGD

0RUHRYHUE\
w$ D E
% D E
wE
ZKLOH $ E U EU 7KXV % E U ,QGHHGGVDORQJ$%LVHTXDOWR U GE
:HQRWHWKDWWKHVROXWLRQIRXQGDERYHFDQEHUHZULWWHQLQWHUPVRI/DPHIXQFWLRQVDV
IROORZVILQGWKHILHOG $O P RIWKH+HQFN\QHWJHQHUDWHGE\WZRFLUFOHVRIUDGLLU
DQGU
+DYLQJ $O P RQHFDQILQGWKHIXQFWLRQ %O P E\:HXVHWKHUXOHV

WRILQG
% O P U*R O P U* O P
(TXLOLEULXPRIWKH0LFKHOO6WUXFWXUHV
7KHRSWLPDO0LFKHOOVWUXFWXUHFDOOHGVRPHWLPHVWUXVVZKLFKZLOOEHH[SODLQHGODWHU
LVLPEHGGHGLQWRDYLUWXDOFDOOHGVRPHWLPHVDGMRLQWILHOGY YYVXFKWKDWWKH
ILEUHVDQGEDUVIROORZWKHWUDMHFWRULHVRIHHYVDWLVI\LQJWKHFRQGLWLRQVRIWKHUHJLRQV
7562DVGLVFXVVHGLQ6HF7KHPRVWSRSXODUSDWWHUQVRI0LFKHOOVWUXFWXUHV
FRUUHVSRQGWRWKH7UHJLRQ%XWHYHQLIWKLVLPEHGGLQJLVIXOILOOHGLWLVRXUGXW\WR
FKHFNZKHWKHUWKHVWUXFWXUHFDQEHHTXLOLEUDWHG
/HWXVFRQVLGHUWKH0LFKHOOVWUXFWXUHDVD SODWH RI XQLW WKLFNQHVV LQ ZKLFK WKH
VWUHVVUHVXOWDQWV1LMDSSHDUFRUUHVSRQGLQJWRWKHSDUDPHWULFOLQHV[ D[ E7KH
HTXLOLEULXPHTXDWLRQVUHDG
1 LM_ M L M
ZLWK L_M EHLQJ WKH FRYDULDQW GHULYDWLYH DV QRWHG LQ 6HF /HW XV LQWURGXFH WKH
SK\VLFDOFRPSRQHQWVRI1LME\WKHPHWKRGVXVXDOO\XVHGLQWKHWKHRU\RIVKHOOV:H
GHILQH

1
1 1 1 1 1
$ % $%
7KHSK\VLFDOFRPSRQHQWVDERYHUHIHUWRWKHFRYDULDQWEDVLV\HWLIUHIHUUHGWRWKH
FRQWUDYDULDQWEDVLVWKHVHVGHILQLWLRQVGRQRWFKDQJHEHFDXVHWKHFRRUGLQDWHV\VWHP
LVRUWKRJRQDO'XHWRRUWKRJRQDOLW\RIWKHSDUDPHWHUL]DWLRQWKHSK\VLFDOFRPSRQHQWV
RIDOOWHQVRUVDUHXQLTXHO\GHILQHG7KHHTXLOLEULXPHTXDWLRQVH[SUHVVHGLQWHUPVRI
WKHSK\VLFDOFRPSRQHQWVDUHGHULYHGIURP:HJHWWKHHTXDWLRQV
ZHOONQRZQIURPWKHWKHRU\RIPHPEUDQHVKHOOV
w %1 w% w

wD

wD
1
$ wE
$ 1
DQGWKHVHFRQGHTXDWLRQLVIRUPHGE\UHSODFLQJ%ZLWK$ZLWKDQGDZLWKE
,WLVUHTXLUHGKHUHWKDW1 VLQFHWKHPLFURVWUXFWXUHFDQQRWEHVKHDUHG7KLV
PHDQVWKDW[L FRQVWEHFRPHWUDMHFWRULHVRISULQFLSDOVWUHVVUHVXOWDQWV1,1,,DQGWKH
SK\VLFDOFRPSRQHQWVEHFRPH
1
1 , 1 1 ,,
7KHHTXDWLRQVH[SUHVVHGLQWKHOLQHVFRLQFLGLQJZLWKWUDMHFWRULHVRISULQFLSDO

VWUHVVHVUHGXFHWR
w w%
%1 , 1 ,,
wD wD

w w$
$1 ,, 1 ,
wE wE
1RWHWKDWWKHRSWLPDOLW\FRQGLWLRQVGRQRWDIIHFW:HUHPHPEHUWKDW1,
1,,DUHPHDVXUHGLQ1PVLQFHWKHWKLFNQHVV OLNHLQWKHWKHRU\RISODWHVXQGHUWKH
LQSODQHORDGLQJ

)LJXUH7KHDYHUDJHVWUHVVUHVXOWDQWVYHUVXVIRUFHVLQWKHLQILQLWHVLPDOVWULSV
7KHIRUFHVDFWLQJRQWKHVLGHVRIWKHLQILQLWHVLPDOVHJPHQWRIWKHSODWHDUH
G1D 1 , %G E G1 E 1 ,, $G D
/HWXVUHIHUWKHVHIRUFHVWRWKHZLGWKVRIWKHVWULSVVHH)LJ:HQRWHWKDWWKHVWUHVV
OLPLWVLVUHDFKHGKHQFH
G1D V R GVE& G1 E V R GVD&

&RPELQLQJWKHVHHTXDWLRQVZLWKZHILQGWKHZLGWKVRIWKHIXOO\VWUHVVHGVWULSV

GVD& 1 ,, $G D GVE& 1 , %G E
VR VR
7KHYROXPHRIWKHPDWHULDOZLWKLQWKHGRPDLQRIVLGHV $GD u %GE LVHTXDOWRVHH
)LJ
G9 GVD& %G E GVE& $G D
+HQFH

G9 ª 1 , 1 ,, º¼ $%GD G E
VR ¬
:HH[SUHVVWKHYROXPHLQWHUPVRIWKHIRUFHV1,1,,E\

9R
VR ³ 1
:
, 1 ,, G :
WKXVDUULYLQJDWWKHIRUPXODGHULYHGSUHYLRXVO\LQDGLIIHUHQWPDQQHU:HGHILQH

DIWHU+HPS
7 %1 , 7 $1 ,,
DQGDVVXPHWKDW1,!1,,:HVHHWKDW 77KDYHXQLWVRIDIRUFH>1@7KH
HTXDWLRQVFDQEHVLPSOLILHGHVVHQWLDOO\ZLWKXVLQJWRWKHIRUP
w w
7 7 7 7
wD wE
/HWXVVWUHVVWKDWWKHVHHTXDWLRQVDUHYDOLGRQO\LQ+HQFN\QHWVVDWLVI\LQJ1RWH
WKDW
V R GVD& 7 G D V R GVE& 7 G E
/HWXVQRWHDOVRWKDWWKHILHOGV77VDWLVI\RU
w 7 w 7
7 7
wDwE wDwE
/HWXVILQGQRZDYDULDWLRQDOIRUPRIWKH HTXLOLEULXP HTXDWLRQV %\ PXOWL
SO\LQJ ERWK HTXDWLRQV E\ WKH WULDO ILHOGV Y X Y Y DQG LQWHJUDWLQJ RYHU WKH
GRPDLQ:RIYDULDWLRQRIDQGDEZHJHW
§ wX wY ·
³ ¨© 7
:R

wD
7
wE
7X 7 X ¸G D G E
¹
/
ZKHUHWKHUKVUHSUHVHQWVDYLUWXDOZRUNRIWKHORDGLQJ7KHVDPHHTXDWLRQFDQEH
LQIHUUHGIURPWKHYLUWXDOZRUNHTXDWLRQRIWKHSODWH7KLVSDUWLFXODUIRUPRIWKH
YLUWXDOZRUNHTXDWLRQLVYDOLGRQO\LQ+HQFN\QHWV
5HPDUNRQWKHWUXVVDQDORJ\
7KHHTXDWLRQWXUQVRXWWREHDFRQWLQXXPOLPLWRIWKHVHTXHQFHRIWKHYLUWXDO
ZRUNHTXDWLRQVFRUUHVSRQGLQJWRWKHVHTXHQFHRIWUXVVHVDSSUR[LPDWLQJWKH0LFKHOO
VROXWLRQV 7R VKRZ LW ZH GLVFUHWL]H WKH GRPDLQ LQWR VWULSV RI ZLGWKV GD 'D
GE 'E7KXVZHUHSODFHE\
¦¦ ª¬7 'E 'X X'D 7 'D 'X X 'E º¼

/
1RWHWKDWE\ZHKDYH

H $GD GX X G D H %G E GX XG E
/HWXVGLVFUHWL]HWKHVHHTXDWLRQV
' 'X Y 'D ' 'Y X 'E
ZKHUH ' ' GHQRWH HORQJDWLRQV RI WKH VWULSV LQ WHQVLRQ DQG FRPSUHVVLRQ :H
GHILQHWKHIRUFHVLQWKHVWULSVE\
7'E 7 7 'D 7
7KXVWKHHTXDWLRQLVUHDUUDQJHGWRWKHIRUP
¦ 7
' 7 ' /
ZKLFKLVNQRZQIURPWKHWKHRU\RIWUXVVHV7KHOHIWKDQGVLGHUHSUHVHQWVWKHYLUWXDO
ZRUNRIWKHD[LDOIRUFHVLQEDUVRQEDUVHORQJDWLRQVZKLOHWKHULJKWKDQGVLGHLVHTXDO
WRWKHYLUWXDOZRUNRIWKHORDGLQJ7KHYLUWXDOHORQJDWLRQV ǻ DUHOLQNHGZLWKYLUWXDO
QRGDO GLVSODFHPHQWV T E\ WKH HTXDWLRQ ǻ %T ZLWK % FDOOHG WKH JHRPHWULF
PDWUL[7KHHTXDWLRQFDQEHUHZULWWHQLQWKHPDWUL[IRUP ǻ 7 7 T 7 3 ZKHUH
3LVWKHQRGDOORDGYHFWRU7KHVHWZRHTXDWLRQVLPSO\WKHHTXDWLRQVOLQNLQJWKHD[LDO
IRUFHVZLWKWKHQRGDOORDGLQJ%77 37KHDERYHHTXDWLRQVUHSUHVHQWWKHFRQGL
WLRQVRIHTXLOLEULXPRIWKHQRGHV-XVWWKLVPDWUL[HTXDWLRQLVZHOONQRZQIURPWKH
WKHRU\RIWUXVVHVLWLVWKHEDVLVIRUWKHWUXVVRSWLPL]DWLRQEHLQJWKHVXEMHFWRIWKH
DFFRPSDQ\LQJDUWLFOHE\*5R]YDQ\DQG76RNyáRQWUXVVOLNHDSSUR[LPDWLRQVRI
0LFKHOOOD\RXWV6LQFHWKHWUXVVHVDSSUR[LPDWLQJWKH0LFKHOOVWUXFWXUHDUHVWDWLFDOO\
GHWHUPLQDWHWKHHTXDWLRQVDERYHDUHVROYDEOHWKXVGHWHUPLQLQJWKHYROXPHVRIWKH
IXOO\ VWUHVVHG WUXVVHV 7KH VHTXHQFH RI VXFK YROXPHV LV D GHFUHDVLQJ VHTXHQFH
WHQGLQJWRWKHH[DFW0LFKHOOUHVXOWIURPWKHXSSHU
/HWXVFRPHEDFNWR(TV7KLVV\VWHPFDQEHVROYHGE\XVLQJWKH
VDPH5LHPDQQPHWKRGVDVGHVFULEHGDERYHWRILQGWKHQHWRIILEUHV$QLOOXVWUDWLYH
VWDWLFDODQDO\VLVFRQFHUQLQJWKH0LFKHOOFDQWLOHYHUVXSSRUWHGRQDFLUFOHDQGGHVLJQHG
LQLWVH[WHULRUFDQEHIRXQGLQ*UDF]\NRZVNLDQG/HZLĔVNL
$Q$OJRULWKPRI&RQVWUXFWLQJDQ([DFW6ROXWLRQWRD0LFKHOO
3UREOHPRI77\SH
6HYHUDO LPSRUWDQW VROXWLRQV DUH FKDUDFWHUL]HG H[FOXVLYHO\ E\ WKH 7 UHJLRQV 7KH
FRQVWUXFWLRQRIVXFKVROXWLRQVFRQVLVWVRIWKHIROORZLQJVWHSV
)RUPDWLRQRIWKHQHWRIILEUHVRISURSHUWLHVRXWOLQHGLQ6HF2QHVKRXOGILQG
WKHILHOGV $D E %D E [ D E \ D E
&RQVWUXFWLRQ RI YLUWXDO ILHOGV X D E Y D E RI SURSHUWLHV DV LQ 7UHJLRQV
VXFKWKDWWKHNLQHPDWLFERXQGDU\FRQGLWLRQVDUHVDWLVILHGKDYLQJXDQGYZH
KDYHWUDMHFWRULHVRIWKHSULQFLSDOVWUDLQVMXVWWKHOLQHV:HFRPSXWHWKHYLUWXDO
ZRUNRIWKHORDGDQGWKHRSWLPDOYROXPHE\
)LQGWKHIRUFHILHOGV 7 D E 7 D E WKHILUVWDVVXPLQJSRVLWLYHYDOXHVWKH
VHFRQGQHJDWLYHVDWLVI\LQJWKHHTXLOLEULXPHTXDWLRQVZLWKLQWKHERG\
DQGWKHERXQGDU\FRQGLWLRQVDORQJLWVORDGHGFRQWRXU,ISRLQWORDGVDUHDS
SOLHGWKH\FDQRQO\EHHTXLOLEUDWHGE\WKHEDUVRIILQLWHFURVVVHFWLRQVZKLFK
UHLQIRUFHWKHVWUXFWXUHXVXDOO\DORQJWKHFRQWRXU7KHVHEDUV RIILQLWHFURVV
VHFWLRQVFDQEHFXUYHGZKLFKLVSRVVLEOHGXHWRWKHSUHVHQFHRIWKHQRUPDO
WUDFWLRQVZKLFKDSSHDUXSRQGLVFRQQHFWLQJWKHEDUVDQGWKHILEURXVGRPDLQVRI
WKH 0LFKHOO VWUXFWXUH +DYLQJ WKH IRUFH ILHOGV 7 D E 7 D E RQH FDQ
FRPSXWHWKHUHDFWLRQVDWWKHVXSSRUWV)LQDOO\WKHWKUHHJOREDOFRQGLWLRQVRI
HTXLOLEULXP VKRXOG EH FKHFNHG +DYLQJ WKH IRUFH ILHOGV 7 D E 7 D E
RQHFDQFRPSXWHWKHVWUHVVUHVXOWDQWV 1 , D E 1 ,, D E DFFRUGLQJWR
DQGWKHQFRPSXWHWKHYROXPHE\7KLVYDOXHVKRXOGFRLQFLGHZLWKWKH
YDOXHE\ZKLFKLVWKHILQDOFKHFNRIFRUUHFWQHVV
0LFKHOOGLGQRWSURYLGHWKHIRUPXODLWFDQEHIRXQGLQ+HPS
(T$PRGLILHGGHULYDWLRQRIFDQEHIRXQGLQ6HFRIWKHDUWLFOHE\
']LHUĪDQRZVNLDQG/HZLĔVNLLQWKLVYROXPH,IWKHSRLQWORDGVDUHSUHVHQWWKHIRUPXOD
VKRXOGEHDXJPHQWHGZLWKDSSURSULDWHWHUPVFKDUDFWHUL]LQJWKHEDUVRIILQLWH
FURVV VHFWLRQV VHH *UDF]\NRZVNL DQG /HZLĔVNL ,Q SDUWLFXODU WKH
YROXPHVRIWKHEDUVH[SUHVVHGE\OLQHLQWHJUDOVVKRXOGEHDGGHGWRWKHVXUIDFHLQWH
JUDO
0LFKHOO¶V&DQWLOHYHUV
7KHSUREOHPVHWKHUHDVLQ)LJLVDFDQWLOHYHUSUREOHPEHFDXVHWKHUHDFWLRQVDUHD
SULRULXQNQRZQ7KHUHDFWLRQVFDQEHFRPSXWHGLQWKHODVWVWDJHRIWKHFRQVWUXFWLRQ
7KH 0LFKHOO FDQWLOHYHU SUREOHPV DUH FKDUDFWHUL]HG E\ D UHPDUNDEOH SURSHUW\ WKH
+HQFN\QHWLVGHWHUPLQHGE\WKHVKDSHRIWKHFRQWRXU*RISRVVLEOHVXSSRUWDQGE\
WKHVKDSHRI:DQGLVWKHVDPHIRUDEURDGFODVVRIORDGV7KHRSWLPDOVWUXFWXUHLV
FRQVWUXFWHGE\FXWWLQJLWIURPWKHJLYHQQHW7KHVROXWLRQVRIWKLVFODVVKDYHEHHQ
FRQVWUXFWHG LQ SDSHUV $6/&KDQ +6<&KDQ
'HZKXUVW*UDF]\NRZVNLDQG/HZLĔVNLDEDE
+HPS/HZLĔVNLHWDODEDE5R]YDQ\
)RUWKLVFODVVRIVROXWLRQVWKHRSWLPDOYROXPHFDQEHFRPSXWHGE\WKHVWDWLF
DQDO\VLV FDQ EH HYHQ GLVUHJDUGHG LW LV RXU GXW\ RQO\ WR FKHFN ZKHWKHU WKH ILHOGV
7 ! DQG 7 H[LVWWRVDWLVI\DOOHTXLOLEULXPFRQGLWLRQV

)LJXUH0LFKHOO¶VFDQWLOHYHUVXSSRUWHGRQ D FLUFOH DIWHU *UDF]\NRZVNL DQG /HZLĔVNL
7KHVWUXFWXUHFRUUHVSRQGVWRDSRLQWORDGDWWKHWRSGLUHFWHGVXFKWKDWRQHIDPLO\RI
EDUVLVLQWHQVLRQDQGWKHVHFRQGIDPLO\LVLQFRPSUHVVLRQ
3UREOHPV ZKLFK $ULVH GXH WR %RXQGDU\ /LQHV $IIHFWLQJ WKH

)LQDO/D\RXW
7KH SUREOHP RI 6HF LQ +HPS FDOOHG ³F\FORLGV´ LV WKH RQO\ H[DPSOH
UHSRUWHGWKHUHRIWKLVFODVV7KHVROXWLRQLVFRQVWUXFWHGZLWKLQDVWULSZKLFKUHVXOWVLQ
WKH+HQFN\QHWRQHIDPLO\RIILEUHVRIZKLFKEHFRPHWDQJHQWWRWKHERXQGDU\
7KH ILUVW SURIRXQG SDSHU RQ WKH VROXWLRQV DGMDFHQW WR WKH VWUDLJKW OLQHV ZKLFK
EHFRPHHQYHORSHVIRUWKH+HQFN\QHWVLVGXHWR+HQU\&KDQ-XVWWKHUHDQ
LQJHQLRXVPDWKHPDWLFDOFRQVWUXFWLRQKDVEHHQ SXW IRUZDUG OHDGLQJ WR H[SOLFLW IRU
PXODHIRUWKH[\$%XYILHOGV&KDQ¶VFRQVWUXFWLRQKDVEHHQDSSOLHGLQVRPH
VXEGRPDLQVRIWKHFDQWLOHYHUGLVFXVVHGLQ/HZLĔVNLHWDODE*UDF]\NRZVNL
DQG/HZLĔVNLDEDEDQG/HZLĔVNLDQG5R]YDQ\DE
DQG /HZLĔVNL HW DO 7KHUH WKH FRQVWUXFWLRQ LV QRW EDVHG RQ WKH 5LHPDQQ
IRUPXODEXWLVPXFKPRUHFRPSOLFDWHGUHTXLULQJVROYLQJDFRQYROXWLRQW\SH
LQWHJUDOHTXDWLRQ7KHDQDO\WLFDOPHWKRGVRIVROYLQJWKHVHSUREOHPVDUHQRWUHFDOOHG
KHUH $Q LOOXVWUDWLYH 0LFKHOO¶V FDQWLOHYHU FRQVWUXFWHG ZLWKLQ D ERXQGHG GRPDLQ LV
VKRZQLQ)LJ7KHVWUDLJKWSDUWVRIWKHHGJHVRIWKHFDQWLOHYHUFRLQFLGHZLWKWKH
VWUDLJKWOLQHVEHLQJWKHERXQGDULHVRIWKHIHDVLEOHGRPDLQ)LJXUH UHIHUV WR WKH
SRLQWORDGDSSOLHGDWWKHWRSGLUHFWHGVXFKWKDWWKHVWUHVVHVLQERWKIDPLOLHVRIILEUHV
DUH RI GLIIHUHQW VLJQ /HW XV VWUHVV WKDW WKH QHW LQ )LJ LV ORDGLQGHSHQGHQW DQG
)LJXUH 0LFKHOO FDQWLOHYHUV GHVLJQHG LQ D WUDSH]RLGDO GRPDLQ D V7V& E
V7V& DIWHU*UDF]\NRZVNLDQG/HZLĔVNL
GHWHUPLQHV D IDPLO\ RI RSWLPDO FDQWLOHYHUV FRUUHVSRQGLQJ WR WKH ORDGV DSSOLHG DW
DUELWUDU\LQWHULRUSRLQWRIWKHQHWVKRZQ7KHRSWLPDOVWUXFWXUHVDUHFXWRXWIURPWKH
QHWDORQJWKHOLQHVFRQQHFWHGDWWKHSRLQWRIDSSOLFDWLRQRIWKHSRLQWORDG&RQVH
TXHQWO\ WKH VROXWLRQ FRUUHVSRQGLQJ WR Q IRUFHV LV D GLUHFW FRPSRVLWLRQ RI WKH Q
FRUUHVSRQGLQJFDQWLOHYHUV
8QHTXDO/LPLW6WUHVVHVLQ7HQVLRQDQG&RPSUHVVLRQ
&RQVLGHUWKHFDVHRIV&dVdV77KHZKROHOHFWXUHE\+HPS&KDSWHULV
FRQFHUQHGZLWKWKLVPRUHJHQHUDOFDVH:KLOHWKHFUXFLDOHTXDWLRQKROGVKHUH
VRPHRWKHUIRUPXODHVKRXOGEHFKDQJHGDSSURSULDWHO\
7KHUHH[LVWSUREOHPVLQZKLFKWKHUDWLRV7V&GRHVQRWLQIOXHQFHWKHJHRPHWU\RI
WKH+HQFN\QHWDVQRWHGE\*,15R]YDQ\VHHKLVDUWLFOHLQWKLVYROXPH)RUPRVW
SUREOHPVOLNHIRUWKHFDQWLOHYHUVWKHUDWLRV7V&GRHVDIIHFWWKHVROXWLRQHVVHQWLDOO\
VHH HJ WKH EHQFKPDUNV LQ *UDF]\NRZVNL DQG /HZLĔVNL 7KH
0LFKHOOFDQWLOHYHUIRUWKHFDVHRIV7V& LVVKRZQLQ)LJE7KHUHPDUNVDWWKH
HQGRI6HFFRQFHUQLQJ)LJDSSO\WRWKLVUHVXOW/HWXVVWUHVVWKLVQHWLVVWDEOH
ZLWKUHVSHFWWRWKHORDGVDSSOLHGXQGHUWKHNQRZQOLPLWDWLRQV
&RXSOLQJRI.LQHPDWLFVDQG6WDWLFV
,QWKHW\SLFDO0LFKHOOSUREOHPVWKHVROXWLRQFDQEHFRQVWUXFWHGVHTXHQWLDOO\$WWKH
ILUVWVWDJHWKHQHWLVFRQVWUXFWHGVXFKWKDWHDFKUHJLRQFRUUHVSRQGVWRWKHUHJLPHV7
65DQG2,QWKLVILUVWVWDJHWKHXQNQRZQVDUH[\$%XYZEHLQJWKHIXQFWLRQV
RIDE,QWKHQH[WVWHSZHILQGWKHILHOGV77HTXLOLEUDWLQJWKHORDGLQJDQGWKHQ
ILQDOO\ ILQG WKH UHDFWLRQV $OO WKH SUREOHPV VROYHG LQ WKH ERRN E\ +HPS
&KDSWHUDVZHOODVWKHSUREOHPVVROYHGLQSDSHUVE\/HZLĔVNLHWDODE

)LJXUH7KHWKUHHIRUFHVSUREOHPVHWWLQJ
EHORQJ WR WKLV FODVV ,Q PDQ\ RI WKHVH VROXWLRQV WKH QHW LV VWDEOH ZLWK UHVSHFW WR
FKDQJLQJWKHSRVLWLRQDQGGLUHFWLRQRIWKHORDGSURYLGHGWKDW77DWHDFKSRLQW
DQGDOOHTXLOLEULXPFRQGLWLRQVDUHDVVXUHG<HWWKHUHH[LVWVDEURDGIDPLO\RISURE
OHPVH[FHHGLQJWKLVFODVV7KHPRVWLPSRUWDQWLVVWLOORQO\SDUWO\VROYHGWKHWKUHH
IRUFHVSUREOHPFRQVWUXFWWKHOLJKWHVWVWUXFWXUHLQZKLFKV&dVdV7ZKHUH VLV
WKHD[LDOVWUHVVHTXLOLEUDWLQJDJLYHQV\VWHPRIWKUHHIRUFHVDWWKUHHJLYHQSRLQWVVHH
)LJ

)LJXUH 2SWLPDO IUDPHZRUN IRU JLYHQ [' DQG \' G ZKHUH G _51_ +HUH WKH
RSWLPDOYROXPH9 8 J qDQG T T qWKXVWKHDQJOHVRIERWK
FLUFXODUIDQVDUHHTXDO

)LJXUH2SWLPDOIUDPHZRUNIRUJLYHQ[' G\' G)[ )\ +HUH9
8J qT qT q
7KHIRXUIRUFHVSUREOHPEHORQJVWRWKHVDPHFODVVDPRQJZKLFKLWLVZRUWKZKLOH
WRVROYHWKHSUREOHPRIHTXLOLEUDWLQJIRXUSDUDOOHOIRUFHVZKRVHSRLQWVRIDSSOLFDWLRQV
OLHRQRQHOLQH
7KHILUVWSDSHURQWKHWKUHHIRUFHVSUREOHPLVWKHUHSRUWE\&KDQWKHPDLQ
UHVXOWVRIZKLFKKDYHEHHQLQFOXGHGLQ6RNyODQG/HZLĔVNL
,QWKHWKUHHIRUFHVSUREOHPWKHQHWRIEDUVLVIXOO\GHSHQGHQWRQWKHIRUFHSRVLWLRQ
2QHFDQVD\WKDWWKHQHWIROORZVWKHIRUFHV7KHH[SHULHQFHVIURPWKHFRQVWUXFWLRQRI
0LFKHOO¶V FDQWLOHYHUV DUH KHUH ERWK XVHOHVV DQG PLVOHDGLQJ 1RWH WKDW WKH QHWV RI
VROXWLRQV RI )LJ UHSHDWHG DIWHU 6RNyá DQG /HZLĔVNL IRUP D IDPLO\
LQGH[HGE\WKHIRUFHVSRVLWLRQ1RWHWKDWWKHPDLQILEURXVGRPDLQVDUHEXLOWRQWZR
FLUFOHVRIGLIIHUHQWUDGLLZLWKXVLQJWKHWRROVRXWOLQHGLQ6HF7ZRUHSUHVHQWDWLYH
OD\RXWVRIWKUHHIRUFHVSUREOHPIRUGLIIHUHQWIRUFHGLUHFWLRQDQGSRVLWLRQVDUHVKRZQ
LQ)LJV7KHUHIHUHQFHYROXPHLVGHILQHGE\8 )[GVSZKHUHGGHQRWHVWKH
GLVWDQFHEHWZHHQSRLQWV5DQG1
7KHUHVXOWRI)LJLOOXVWUDWHVWKHWKHRUHPE\&KDQZKLFKVD\VWKDWLIWKH
ORDGDW'LVSDUDOOHOWR51DQGLIWKHWKUHHIRUFHVWZRUHDFWLRQVDQGWKHORDGLQWHUVHFW
DW'WKHQWKHDQJOHVRIFLUFXODUIDQVDUHHTXDO7KHUHVXOWRI)LJFRQILUPVWKLV
WKHRUHP&RQWUDU\LQ)LJWKHIDQDQJOHVDUHGLIIHUHQWDQGZHVHHWKDWWKHLQWHU
VHFWLRQSRLQWOLHVIDUIURPSRLQWZKHUHWKHORDGLVDSSOLHG
2QWKH5ROHRIWKH&KRLFHRIWKH)HDVLEOH'RPDLQ
7KLVWRSLFLVGLVFXVVHGLQ5R]YDQ\¶VOHFWXUHLQWKHVDPHYROXPH<HWLWLVZRUWKZKLOH
WRUHFDOOKHUH WKH UHVXOW RI SDSHU E\ 6RNyá DQG /HZLĔVNL ZKLFK VKRZV WKH
VROXWLRQV WR WKH IRXU IRUFHV SUREOHPV LQ LWV PRVW VLPSOH VHWWLQJ LI WKH IRUFHV DUH
SDUDOOHO DQG V\PPHWULFDOO\ DSSOLHG DORQJ D OLQH /HW XV ORRN DW WKH OD\RXWV LQ
)LJVIRUWKHVDPHV\VWHPRIORDGLQJDQGWZRFDVHVRIVXSSRUWVWZRKLQJHVRU
VOLGLQJVXSSRUWV1RWHWKDWWKHVROXWLRQVIRUWKHKDOISODQH)LJVGLIIHUIURPWKH
VROXWLRQVIRUWKHIXOOSODQH)LJ,QWKLVODVWFDVHWKHHPSW\UHJLRQV2DUHVTXDUHV
,WLVUHPDUNDEOHWKDWWKHUHVXOWVLQ)LJFRQFHUQLQJWKHGHVLJQGRPDLQEHLQJWKH
IXOOSODQHDUHYDOLGIRUWZRFDVHVIRUVOLGLQJVXSSRUWVDQGIRUKLQJHVDWWKHVXSSRUWV
VHHWKHFRPPHQWVLQWKHDUWLFOHE\*,15R]YDQ\LQWKHVDPHYROXPH&RPSDUHQRZ
WKHUHVXOWVRI)LJVDQG:HVHHWKHLQIOXHQFHRIWKHERXQGDU\FRQGLWLRQV7KLV
FRPSDULVRQLQGLFDWHVWKDW
± WKHVROXWLRQIRUWKHKDOISODQH)LJLVGLIIHUHQWWKDQWKHXSSHUKDOIRIWKH
VROXWLRQIRUWKHIXOOSODQH
± WKHVROXWLRQRI)LJIRUWKHKDOISODQHLVQRWWKHXSSHUKDOIRIWKHVROXWLRQ
IRUWKHIXOOSODQHDVLQ)LJ
D
L
0.5L
P P P P
E
L
0.75L
P P P P
F
L
2P 2P
)LJXUH([DFWVROXWLRQVRIWKH0F&RQQHODEDQG0LFKHOOSUREOHPF
0.5L
P P P P
0.75L
P P P P
2P 2P
)LJXUH([DFWVROXWLRQVIRUSUREOHPVZLWKRQHVOLGLQJVXSSRUW
D
L
0.5L
P P P P
E
L
0.75L
P P P P
F
L
2P 2P
)LJXUH([DFWOD\RXWVIRUWKHFDVHRIWKHIHDVLEOHGRPDLQEHLQJWKHZKROHSODQH7KH
VXSSRUWVPD\EHVOLGLQJLQWKHKRUL]RQWDOGLUHFWLRQZLWKQRHIIHFWRQWKHIRUPRIWKHVROX
WLRQVLQDOOFDVHVVKRZQ
:HGRQRWGLVFXVVKHUHWKHVRFDOOHGEULGJHSUREOHPFRQFHUQLQJDXQLIRUPORDG
EHWZHHQWZRIL[HGKLQJHVWKHPDLQUHVXOWVDUHGXHWR+HPS'DUZLFKHWDO
7\DVHWDO3LFKXJLQHWDO
)LQDO5HPDUNV
7KHDQDO\WLFDOVROXWLRQVWRWKH0LFKHOOSUREOHPVFDQQRWEHIRXQGIRUFRPSOLFDWHG
GRPDLQVDQGORDGFDVHV)RUWXQDWHO\WKHFRQWHPSRUDU\QXPHULFDOPHWKRGVPDNHLW
SRVVLEOHWRSUHGLFWWKHRSWLPDOOD\RXWVZLWKKLJKDFFXUDF\DVVKRZQHJLQ'DUZLFK
HWDO7\DVHWDO6RNyáDQG/HZLĔVNL6RNyá6RNyáDQG
5R]YDQ\/HZLĔVNLHWDO7KHVHVROXWLRQVSDYHWKHZD\WRUDWLRQDOGHVLJQ
RIVWUXFWXUHVLQFOXGLQJWKHUHLQIRUFHGFRQFUHWHVWUXFWXUHV7KHPHWKRGVVKRZQKHUH
UHSODFHWRVRPHH[WHQWDSRSXODUPHWKRGRIVWUXWVDQGWLHRUDWOHDVWSURYLGHWKH
LGHDVRISXWWLQJWKHUHLQIRUFHPHQWLQWRWKH5&VWUXFWXUHV
5HIHUHQFHV
:$FKW]LJHU 7RSRORJ\ 2SWLPL]DWLRQ RI 'LVFUHWH 6WUXFWXUHV $Q ,QWURGXFWRU\ &RXUVH
8VLQJ&RQYH[DQG1RQVPRRWK$QDO\VLV$GYDQFHG6FKRRO,QWHUQDWLRQDO&HQWUH IRU
0HFKDQLFDO6FLHQFHV&,608GLQH-XQH
*$OODLUH6KDSHRSWLPL]DWLRQE\WKHKRPRJHQL]DWLRQPHWKRG6SULQJHU1HZ<RUN
*$OODLUH(%RQQHWLHU*)UDQFIRUWDQG)-RXYH6KDSHRSWLPLVDWLRQE\WKHKRPRJHQL
VDWLRQPHWKRG1XPHU0DWK
03%HQGV¡H2SWLPL]DWLRQRIVWUXFWXUDOWRSRORJ\VKDSHDQGPDWHULDO6SULQJHU%HUOLQ

&&DUDWKpRGRU\ DQG (6FKPLGW ĥEHU GLH +HQFN\3UDQGWOVFKHQ .XUYHQ =HLWVFKULIW IU
$QJHZDQGWH0DWKHPDWLNXQG0HFKDQLN
$6/&KDQ7KHGHVLJQRI0LFKHOORSWLPXPVWUXFWXUHV5HS&ROO$HURQDXW&UDQILHOG
1R
+6<&KDQ 2SWLPXP 0LFKHOO IUDPHZRUNV IRU WKUHH SDUDOOHO IRUFHV 7KH &ROOHJH RI
$HURQDXWLFV&UDQILHOG5HSRUW$(521R
+6<&KDQ7DEXODWLRQRIVRPHOD\RXWVDQGYLUWXDOGLVSODFHPHQWILHOGVLQWKHWKHRU\RI
0LFKHOORSWLPXPVWUXFWXUHV7KH&ROOHJHRI$HURQDXWLFV&UDQILHOG&R$1RWH$HUR1R

+6<&KDQ 0LQLPXP ZHLJKW FDQWLOHYHU IUDPHV ZLWK VSHFLILHG UHDFWLRQV 8QLYHUVLW\ RI
2[IRUG'HSDUWRI(QJ6FL(QJ/DERUDWRU\3DUNV5RDG2[IRUG1R-XQH

+6<&KDQ+DOISODQHVOLSOLQHILHOGVDQG0LFKHOOVWUXFWXUHV4-0HFK$SSO0DWK

+6<&KDQ 6\PPHWULF SODQH IUDPHZRUNV RI OHDVW ZHLJKW ,Q $6DZF]XN =0Uy]
HGLWRUV2SWLPL]DWLRQLQ6WUXFWXUDO'HVLJQ6SULQJHU%HUOLQSS
+/&R[7KHGHVLJQRIVWUXFWXUHVRIOHDVWZHLJKW3HUJDPRQ3UHVV2[IRUG
:'DUZLFK0*LOEHUWDQG$7\DV2SWLPXPVWUXFWXUHWRFDUU\DXQLIRUPORDGEHWZHHQ
SLQQHGVXSSRUWV6WUXFW0XOWLGLVF2SWLP
)'HPHQJHODQG36XTXHW2QORFNLQJPDWHULDOV$FWD$SSO0DWK
3'HZKXUVW$QDO\WLFDOVROXWLRQVDQGQXPHULFDOSURFHGXUHVIRUPLQLPXPZHLJKW0LFKHOO
VWUXFWXUHV-0HFK3K\V6ROLGV
3'HZKXUVWDQG,)&ROOLQV$PDWUL[WHFKQLTXHIRUFRQVWUXFWLQJVOLSOLQHILHOGVROXWLRQV
WR D FODVV RI SODQH VWUDLQ SODVWLFLW\ SUREOHPV ,QW - 1XPHU 0HWK (QJ

:'RUQ 5*RPRU\ DQG 0*UHHQEHUJ $XWRPDWLF GHVLJQ RI RSWLPDO VWUXFWXUHV - GH
0HFDQLTXH
0*LOEHUW DQG $7\DV /D\RXW RSWLPL]DWLRQ RI ODUJHVFDOH SLQMRLQWHG IUDPHV (QJ
&RPSXW
&*UDF]\NRZVNLDQG7/HZLĔVNL7KHOLJKWHVWSODQHVWUXFWXUHVRIDERXQGHGVWUHVVOHYHO
WUDQVPLWWLQJDSRLQWORDGWRDFLUFXODUVXSSRUW&RQWURODQG&\EHUQHWLFV

&*UDF]\NRZVNL DQG 7/HZLĔVNL 0LFKHOO FDQWLOHYHUV FRQVWUXFWHG ZLWKLQ WUDSH]RLGDO
GRPDLQV3DUW,*HRPHWU\RI+HQFN\QHWV6WUXFW0XOWLGLVF2SWLP
D
GRPDLQV 3DUW ,, 9LUWXDO GLVSODFHPHQW ILHOGV 6WUXFW 0XOWLGLVF 2SWLP
E
GRPDLQV3DUW,,,)RUFHILHOGV6WUXFW0XOWLGLVF2SWLPD
GRPDLQV3DUW,9&RPSOHWHH[DFWVROXWLRQVRIVHOHFWHGRSWLPDOGHVLJQVDQGWKHLUDS
SUR[LPDWLRQV E\ WUXVVHV RI ILQLWH QXPEHU RI MRLQWV 6WUXFW 0XOWLGLVF 2SWLP
E
&*UDF]\NRZVNL DQG 7/HZLĔVNL 0LFKHOO FDQWLOHYHUV FRQVWUXFWHG ZLWKLQ D KDOIVWULS
7DEXODWLRQ RI VHOHFWHG EHQFKPDUN UHVXOWV 6WUXFW 0XOWLGLVF 2SWLP

:6+HPS2SWLPXP6WUXFWXUHV&ODUHQGRQ3UHVV2[IRUG
:6+HPS0LFKHOOIUDPHZRUNIRUXQLIRUPORDGEHWZHHQIL[HGVXSSRUWV(QJ2SWLPL]

5+LOO7KH0DWKHPDWLFDO7KHRU\RI3ODVWLFLW\2[IRUG&ODUHQGRQ3UHVV
7/HZLĔVNL0LFKHOOVWUXFWXUHVIRUPHGRQVXUIDFHVRIUHYROXWLRQ6WUXFW0XOWLGLVF2SWLP

7/HZLĔVNLDQG*,15R]YDQ\([DFWDQDO\WLFDOVROXWLRQVIRUVRPHSRSXODUEHQFKPDUN
SUREOHPV LQ WRSRORJ\ RSWLPL]DWLRQ ,, WKUHHVLGHG SRO\JRQDO VXSSRUWV 6WUXFW 0XOWL
GLVF2SWLP
7/HZLĔVNLDQG*,15R]YDQ\([DFWDQDO\WLFDOVROXWLRQVIRUVRPHSRSXODUEHQFKPDUN
SUREOHPVLQWRSRORJ\RSWLPL]DWLRQ,,,/VKDSHG GRPDLQV6WUXFW 0XOWLGLVF 2SWLP
D
7/HZLĔVNLDQG*,15R]YDQ\$QDO\WLFDOEHQFKPDUNVIRUWRSRORJLFDORSWLPL]DWLRQ,9
VTXDUHVKDSHGOLQHVXSSRUW6WUXFW0XOWLGLVF2SWLPE
7/HZLĔVNL*,15R]YDQ\76RNyáDQG.%RáERWRZVNL([DFWDQDO\WLFDOVROXWLRQVIRU
VRPH SRSXODU EHQFKPDUN SUREOHPV LQ WRSRORJ\ RSWLPL]DWLRQ ,,, /VKDSHG GRPDLQV
UHYLVLWHG6WUXFW0XOWLGLVF2SWLP'2,V
7/HZLĔVNLDQG--7HOHJD3ODWHV/DPLQDWHVDQG6KHOOV$V\PSWRWLF$QDO\VLVDQG+R
PRJHQL]DWLRQ :RUOG 6FLHQWLILF 3XEOLVKLQJ 6LQJDSRUH 1HZ -HUVH\ /RQGRQ +RQJ
.RQJ
7/HZLĔVNL DQG --7HOHJD 0LFKHOOOLNH JULOODJHV DQG VWUXFWXUHV ZLWK ORFNLQJ $UFK
0HFK
7/HZLĔVNL0=KRXDQG*,15R]YDQ\([WHQGHGH[DFWVROXWLRQVIRUOHDVWZHLJKWWUXVV
OD\RXWV 3DUW , &DQWLOHYHU ZLWK D KRUL]RQWDO D[LV RI V\PPHWU\ ,QW - 0HFK 6FL
D
7/HZLĔVNL0=KRXDQG*,15R]YDQ\([WHQGHGH[DFWVROXWLRQVIRUOHDVWZHLJKWWUXVV
OD\RXWV3DUW,,8QV\PPHWULFFDQWLOHYHUV,QW-0HFK6FLE
$(+/RYH$WUHDWLVHRQWKHPDWKHPDWLFDOWKHRU\RIHODVWLFLW\&DPEULGJH
&0D[ZHOO 2Q UHFLSURFDO ILJXUHV IUDPHV DQG GLDJUDPV RI IRUFHV 7UDQV 5R\ 6RF RI
(GLQEXUJK6FLHQWLILF3DSHUV,,
$0D]XUHN*HRPHWULFDODVSHFWVRIRSWLPXPWUXVVOLNHVWUXFWXUHVIRUWKUHHIRUFHSUREOHP
$0D]XUHN:%DNHUDQG&7RUW*HRPHWULFDODVSHFWVRIRSWLPXPWUXVVOLNHVWUXFWXUHV
5(0F&RQQHO /HDVWZHLJKW IUDPHZRUNV IRU ORDGV DFURVV VSDQ - (QJ 0HFK 'LY
±
5(0HOFKHUV2QH[WHQGLQJWKHUDQJHRI0LFKHOOOLNHRSWLPDOWRSRORJ\VWUXFWXUHV6WUXFW
0XOWLGLVF2SWLP
$*00LFKHOO 7KH OLPLWV RI HFRQRP\ RI PDWHULDO LQ IUDPH VWUXFWXUHV 3KLO 0DJ

301DJKGL)RXQGDWLRQRIHODVWLFVKHOOWKHRU\,Q3URJUHVVLQ6ROLG0HFKDQLFVSS
(GV,16QHGGRQDQG5+LOOYRO1RUWK+ROODQG$PVWHUGDP
$93LFKXJLQ$7\DVDQG0*LOEHUW0LFKHOOVWUXFWXUHIRUDXQLIRUPORDGRYHUPXOWLSOH
VSDQV WK :RUOG &RQJUHVV RQ 6WUXFWXUDO DQG 0XOWLGLVFLSOLQDU\ 2SWLPL]DWLRQ
:&602-XQH6KL]XRND-DSDQ&'520
$93LFKXJLQ $7\DV DQG 0*LOEHUW 2Q WKH RSWLPDOLW\ RI +HPS
V DUFK ZLWK YHUWLFDO
KDQJHUV6WUXFW0XOWLGLVF2SWLP
:3UDJHU2QLGHDOORFNLQJPDWHULDOV7UDQV6RF5KHRORJ\
:3UDJHU 2SWLPDO OD\RXW RI WUXVVHV RI ILQLWH QXPEHU RI MRLQWV - 0HFK 3K\V 6ROLGV
D
:3UDJHU1HDUO\RSWLPDOGHVLJQRIWUXVVHV&RPS6WUXFWE
7-3ULWFKDUG 0*LOEHUW DQG $7\DV 3ODVWLF OD\RXW RSWLPL]DWLRQ RI ODUJHVFDOH IUDPH
ZRUNV VXEMHFW WR PXOWLSOH ORDG FDVHV PHPEHU VHOIZHLJKW DQG ZLWK MRLQW OHQJWK
SHQDOWLHVWK:RUOG&RQJUHVVRI6WUXFWXUDODQG0XOWLGLVFLSOLQDU\2SWLPL]DWLRQ5LRGH
-DQHLUR0D\-XQH%UD]LO
*,15R]YDQ\2SWLPDOGHVLJQRIIOH[XUDOV\VWHPV3HUJDPRQ3UHVV/RQGRQ
*,15R]YDQ\6WUXFWXUDO'HVLJQYLD2SWLPDOLW\&ULWHULD.OXZHU$FDGHPLF3XEOLVKHUV
'RUGUHFKW7KH1HWKHUODQGV
*,15R]YDQ\6RPHVKRUWFRPLQJVLQ0LFKHOO¶VWUXVVWKHRU\6WUXFW2SWLP

*,15R]YDQ\3DUWLDOUHOD[DWLRQRIWKHRUWKRJRQDOLW\UHTXLUHPHQWIRUFODVVLFDO0LFKHOO
WUXVVHV6WUXFW2SWLP
*,15R]YDQ\ ([DFW DQDO\WLFDO VROXWLRQV IRU VRPH SRSXODU EHQFKPDUN SUREOHPV LQ
WRSRORJ\RSWLPL]DWLRQ6WUXFW2SWLP
*,15R]YDQ\ DQG 65$GLGDP 'XDO IRUPXODWLRQ RI YDULDWLRQDO SUREOHPV LQ RSWLPDO
GHVLJQ7UDQV$60((QJUJIRU,QGD
*,15R]YDQ\DQG65$GLGDP$EVROXWHPLQLPXPYROXPHRIUHLQIRUFHPHQWLQVODEV-
6WUXFW'LY3URF$P6RF&LY(QJE
*,15R]YDQ\ DQG :*ROOXE 0LFKHOO OD\RXWV IRU YDULRXV FRPELQDWLRQV RI OLQH VXS
SRUWV,,QW-0HFK6FL
*,15R]YDQ\:*ROOXE0=KRX([DFW0LFKHOOOD\RXWVIRUYDULRXVFRPELQDWLRQVRI
OLQHVXSSRUWV3DUW,,6WUXFW2SW
*,1 5R]YDQ\ DQG 7 6RNyá ([DFW WUXVV WRSRORJ\ RSWLPL]DWLRQ DOORZDQFH IRU VXSSRUW
FRVWVDQGGLIIHUHQWSHUPLVVLEOHVWUHVVHVLQWHQVLRQDQGFRPSUHVVLRQH[WHQVLRQVRID
FODVVLFDOVROXWLRQE\0LFKHOO6WUXFW0XOWLGLVF2SWLP
*,15R]YDQ\ DQG &0:DQJ 2Q SODQH 3UDJHUVWUXFWXUHV, ,QW - 0HFK 6FL

76RNyá$OLQHFRGHIRUGLVFUHWL]HG0LFKHOOWUXVVRSWLPL]DWLRQZULWWHQLQ0DWKHPDWLFD
6WUXFW0XOWLGLVF2SWLP±D
76RNyá 7RSRORJ\ RSWLPL]DWLRQ RI ODUJHVFDOH WUXVVHV XVLQJ JURXQG VWUXFWXUH DSSURDFK
ZLWK VHOHFWLYH VXEVHWV RI DFWLYH EDUV (GV $ %RUNRZVNL 7 /HZLĔVNL * ']LHU
ĪDQRZVNLWKLQWFRQIRQµFRPSXWHUPHWKRGVLQPHFKDQLFV¶&00±0D\
:DUVDZ%RRNRI$EVWUDFWSS±E
76RNyáDQG7/HZLĔVNL2QWKHVROXWLRQRIWKHWKUHHIRUFHVSUREOHPDQGLWVDSSOLFDWLRQWR
RSWLPDO GHVLJQLQJ RI D FODVV RI V\PPHWULF SODQH IUDPHZRUNV RI OHDVW ZHLJKW 6WUXFW
0XOWLGLVF2SWLP
76RNyODQG7/HZLĔVNL2SWLPDOGHVLJQRIDFODVVRIV\PPHWULFSODQHIUDPHZRUNVRIOHDVW
ZHLJKW6WUXFW0XOWLGLVF2SWLP
76RNyá DQG 7/HZLĔVNL 2Q WKH WKUHH IRUFHV SUREOHP LQ WUXVV WRSRORJ\ RSWLPL]DWLRQ
$QDO\WLFDODQGQXPHULFDOVROXWLRQVWK:RUOG&RQJUHVVRQ6WUXFWXUDODQG0XOWLGLV
FLSOLQDU\ 2SWLPL]DWLRQ :&602 -XQH 6KL]XRND -DSDQ %RRN RI
$EVWUDFWS)XOOSDSHURQ&'
76RNyáDQG*,15R]YDQ\1HZDQDO\WLFDOEHQFKPDUNVIRUWRSRORJ\RSWLPL]DWLRQDQG
WKHLULPSOLFDWLRQV3DUW,ELV\PPHWULFWUXVVHVZLWKWZRSRLQWORDGVEHWZHHQVXSSRUWV
6WUXFW0XOWLGLVF2SWLP±
*6WUDQJ DQG 59.RKQ +HQFN\3UDQGWO QHWV DQG FRQVWUDLQHG 0LFKHOO WUXVVHV &RPS
0HWK$SSO0HFK(QJ
//6WURPEHUJ $%HJKLQL :)%DNHU DQG *+3DXOLQR 7RSRORJ\ RSWLPL]DWLRQ IRU
EUDFHGIUDPHVFRPELQLQJFRQWLQXXPDQGGLVFUHWHHOHPHQWV(QJLQHHULQJ6WUXFWXUHV

$7\DV$93LFKXJLQDQG0*LOEHUW2SWLPXPVWUXFWXUHWRFDUU\DXQLIRUPORDGEHWZHHQ
SLQQHGVXSSRUWVH[DFWDQDO\WLFDOVROXWLRQ3URF5R\6RF$
9=9ODVRY*HQHUDO7KHRU\IRU6KHOOVDQGLWV$SSOLFDWLRQLQ(QJLQHHULQJ*RVWHNKL]GDW
3XEO0RVFRZWUDQVODWHGLQWR(QJOLVKLQ1$&$7HFKQLFDO7UDQVODWLRQ
&0:DQJDQG*,15R]YDQ\2QSODQH3UDJHUVWUXFWXUHV,,1RQSDUDOOHOH[WHUQDOORDGV
DQGDOORZDQFHVIRUVHOIZHLJKW,QW-0HFK6FL
:=DOHZVNLDQG$$OOHQ6KDSLQJVWUXFWXUHV6WDWLFV-:LOH\1HZ<RUN

Structural Shape and Topology Optimization
François Jouve
University Paris Diderot (Paris 7) and Laboratoire J.L.Lions
Abstract This chapter is the printed matter of a series of lectures

on shape optimization given in Udine (Italy) in June 2012. It is
mainly focused on the optimal design problem in elasticity.
It is well known, since the pioneer work of Tartar and Murat in
the late 70’s, that this problem – as well as the scalar shape opti-
mization problem in conductivity – is ill posed and does not always
admit solutions. If the goal is to construct numerical methods to
address this optimal design problem, this illposedness will lead to
algorithms showing instabilities and erratic behaviours.
Two ways are possible to slightly change the initial problem so
that it may be solved in an efficient and reliable way by numerical
methods. The first one consists in enlarging the set of admissible
shapes and allow, as solutions, “generalized shapes” that may con-
tain fine mixtures of the initial material and void. This process is
called “relaxation” and the main mathematical tool involved is the
homogenization theory. It leads to the so called “homogenization
method for shape optimization”. It is described in Section 2, both
from theoretical and numerical point of views, including practical
issues necessary to write a numerical code.
The second approach consists in restricting the set of admissible
solutions. A few theorems prove the existence of solutions under fur-
ther geometrical, topological or regularity constraints. The classical
domain variation method is revisited using the level set represen-
tation, that may be seen as a constraint on the set of admissible
solutions. The main tool is here the shape derivative. The level set
algorithm is presented in Section 3.
Of course this chapter is too short to develop new results and
methods and present extensive proofs. Its goal is rather to give an
overview on different parts of our work on this topics. Interested
readers should check the references cited to have further details.

DOI 10.1007/978-3-7091-1643-2_7, © CISM, Udine 2014
130 F. Jouve
1 Basic principles of shape optimization

1.1 Model problem
Consider a given bounded domain Ω ⊂ Rd (with d = 2 or 3). The model
problem consists in finding a subdomain ω ⊂ Ω that it minimizes a cost
function (or objective function) J. It can be written, for example, in the
generic form:
inf J(ω)
ω∈Uad
where Uad is the set of admissible shapes among all the subdomains included
into Ω. It may include some constraints, like, for example, an upper bound
over the volume of ω. J depends on ω through the solution of a partial
differential equation (or a system of PDE’s) posed on ω.
In the sequel we will focus on structural shape optimization. The govern-
ing equations will be, most of the time, the linearized elasticity system. Of
course the choice of the objective function, together with the set of admis-
sible solutions (including constraints) is very important. Popular objective
functions will be described later with examples of applications.
Mathematicians usually choose classical objective functions (sometimes
because of theirs good properties rather than their relevance for applica-
tions), while real applications have sometimes “fuzzy” needs. An example
of a popular objective function is the compliance for elastic solids, that leads
to the following problem: “find the most rigid structure for a given set of
external forces and a given volume of constitutive material”.
But real problems coming from industry are usually more complex, in-
volving multiple constraints and loading cases. Examples of such industrial
constraints are: maximum local admissible stress, molding constraints, min-
imal size of small parts, minimal radii of curvature everywhere, cooling of
the molded piece in less than a given time, etc... Some of them are even
difficult to express mathematically in an explicit way (for example the con-
straints related to the molding process), and may involve the solution of
additional complex problems.
1.2 Shape representation

Without getting further into the detailed description of the setting, it
is clear that the shape representation chosen for the computation of the
discrete problem introduces an additional constraint to the set of admissible
solutions. There are essentially two classes of approaches:
• Rather than searching for ω in its full generality, its boundary can
be parametrized by a (small) number of points (e.g. using splines). The
optimization of the control points can be done by classical optimization
Structural Shape and Topology Optimization 131
algorithms. This method is called parametric optimization. It amounts to

a discrete optimization problem.
• Search for ω ⊂ Ω as an arbitrary layout of constitutive material and
do topology optimization.
1.3 Ill posed problem

Generally, the optimal design problems are known to be ill-posed. It
means that there is not always existence of solutions in the set of admissi-
ble shapes chosen (and of course, if there is existence, there is usually no
uniqueness of the solution). The objective function may also be discontinu-
ous with respect to variations of the shape. It means that small changes in
the shape can sometimes have a huge influence on the value of the objective
function.
The practical consequence is that numerical methods may exhibit numer-
ical instabilities and may not have good convergence properties, for example
under mesh refinements. The functional to optimize could also have many
local minima where the algorithms may be trapped.
The following example, for a more simple scalar problem, illustrates some
of these phenomenons.
Model problem: optimal radiator (Tartar)

We consider the classic conductivity equations with two conducting ma-
terials characterized by their conductivities 0 < α < β. The optimization
domain Ω is the unit disc in R2 . The domain is uniformly heated and a
prescribed temperature u = 0 is applied on the boundary.
We introduce a characteristic function χ(x) ∈ {0, 1} that characterizes
the best conductor. Thus, for each x ∈ Ω, the local conductivity can be
written
A(x) = χ(x)β + (1 − χ(x))α
For a given layout χ, the temperature u is solution of

−div (A(x)∇u(x)) = 1 in Ω
u(x) = 0 on ∂Ω
We can then consider two interesting problems: find the best configura-
tion maximizing the mean temperature over Ω or minimizing it. Without
further constraints, these two problems are trivial and admit respectively
χ ≡ 0 and χ ≡ 1 as solutions (domain full of respectively α and β material).
If the most conducting material β is supposed to be more expensive
than the other one, we don’t want to use too much of it. The control of
132 F. Jouve
the balance between the performance we want to optimize and the cost (i.e.
the total volume of material β used) is done introducing a given positive
parameter . The two problems above can now be written:
1. Find the cheapest configuration maximizing the mean temperature
over Ω:

max u(x)dx − χ(x)dx , for a given > 0
χ Ω Ω
2. Find the cheapest configuration minimizing the mean temperature

over Ω:

min u(x)dx + χ(x)dx , for a given > 0
χ Ω Ω
Problem 1 can be easily solved. A simple computation in polar coordi-

nates leads to an explicit classical solution (see Figure below):
0 < r < R0 : material β,

l
R0 < r < 1: material α, with R0 = 2α
β−α
For the Problem 2, it is possible to exhibit a minimizing sequence con-

verging to the global minimum of the functional that can be described as:
0 < r < R0 : material α,

l
R1 < r < 1: material β, with R1 = 2β
β−α
R0 < r < R1 : fine mixture of α and β
Figure 1 shows the solutions of both problems with pure materials α

and β respectively coloured in red and blue. Grey zones on the solution of
problem 2 are composite materials, fine mixtures of α and β. The grey level
representing the local density of β material, from white = pure α, to black
= pure β. The solution of problem 1 is a “classical” solution, involving only
pure materials.
Figure 2 shows the numerical solution obtained by the homogenization
method that will be described in Section 2, after penalization of the com-
posites. This penalization is a numerical procedure allowing to eliminate
grey zones from the solution. The resulting structure is strongly mesh de-
pendent, the size of the peaks depending on the spatial resolution allowed
by the mesh size.
Figure 1. Solution of problem 2 (left) and 1 (right).
Figure 2. Optimal radiator for problem 2: numerical solution using ho-

mogenization after penalization of intermediates densities.
There are three possible ways to change the initial ill-posed problem into a
well posed one:
• change the problem and enlarge the set of admissible solutions to allow
microstructures;
• add constraints (e.g. smoothness of the boundaries, topology con-
straints) or regularizing terms (e.g. a perimeter term) to narrow the set of
admissible solutions in such a way that there is existence of solutions;
• work on finite dimensional sets and do numerical computations.
134 F. Jouve
Although the above classes are not very strict, since a given method could
belong to more than one category, it is possible to give popular examples
belonging to each of then: the first approach leads to topology methods
based on homogenization (homogenization method, SIMP); the second one
includes classical boundary methods and the level set method; parametric
methods, topological gradient and genetic algorithm, for example, belong
to the third one
The remaining of this chapter will be restricted to the first two classes,
and more precisely the homogenization method and the level set method.
2 Homogenization method for shape optimization

Methods based on homogenization have been developed after the pioneer-
ing theoretical works of Murat-Tartar (59), Lurie-Cherkaev (54) and Kohn-
Strang (52). In the late 80’s, Bendsoe and Kikuchi (22) proposed a first
efficient algorithm using particular microstructures (although the paper of
Lurie and Cherkaev, in Russian, contained some basic principles that could
lead to the development of numerical algorithms). It has been then devel-
oped in the 90’s, leading to the SIMP (21) method and the homogenization
method (7)(5).
2.1 Notations and basic equations of linearized elasticity
We consider a linear elastic and isotropic material. The Hooke’s tensor

is denoted A and characterized by the Lamé coefficients λ, μ.
1
If u is the displacement field and e(u) = (∇u + ∇uT ) the strain tensor,
2
the stress tensor σ writes
σ = Ae(u) = λ tr(e(u))I + 2μe(u).

On a given domain ω, the elasticity system writes:

⎧
⎨ −div σ = 0 in ω
σn = f on ∂ω ∩ ΓN
⎩
u = u0 on ∂ω ∩ ΓD
It admits a unique solution if meas(ωD ) > 0 or if the external forces are

well balanced, i.e.

f ds = 0
∂ωN
ω ⊂ Ω, can be characterized by a characteristic function χ : Ω −→ {0, 1}

such that
ω = {x ∈ Ω ; χ(x) = 1}
Extending A by zero on the whole domain Ω, the system of elasticity can

be written: ⎧
⎪
⎪ −div χ(x)Ae(u) = 0 in Ω
⎨
σn = f on ∂Ω
⎪
⎪ σn = 0 on ∂ω \ ∂Ω
⎩
σ = 0 in Ω \ ω
with a natural condition (zero traction) on the boundaries of the holes of ω.

Remark: of course this system is singular if A = 0 in Ω\ω, but in numerical
computations of topology optimization, “void” is often replaced by a “soft”
material in order to get a well posed problem over Ω. This trick is known
under the name of “fictitious material” approach.
2.2 Objective functions

In this section we describe a series of classical objective functions that
have been implemented in either our code based on the homogenization
method or the one that uses the level set method (or both of them).
Compliance:

J(ω) = χ(x)Ae(u) : e(u)dx = A−1 σ : σdx = f · u ds = c(ω) (1)
Ω Ω ∂Ω
It is a global measure of the rigidity, and the most popular objective func-
tion in topology optimization of structures (for practical and mathematical
reasons).
136 F. Jouve
Example of cost function depending on the displacement:
1/α
J(ω) = k(x)|u(x) − u0 (x)| dx
α
(2)
ω
where u0 is a given target displacement, α ≥ 2 a given constant and k is

a bounded positive function. Its purpose is to assign different weights to
different zones of the domain. Typically, if the displacement is only to be
controlled over a small subset of Ω, k will be 1 on this subset and 0 outside.
According to the value of u0 chosen, this objective function can be used
either to minimize u (if u0 = 0), maximize u along a given direction (if |u0 |
is large), or reach a target displacement (if u0 is moderate and reachable).
This objective function has been used to design compliant micromechanisms
(MEMS) like the force inverter pictured below (in reference and deformed
configurations).
Cost function depending on the stresses: as an example of cost func-

tion depending
on the stresses, we can use the Lp norm of the stress tensor,
J(ω) = |σ|p or functions of the Von Mises stresses. If p is large enough,
Ω
this expression could give a reasonable approximation of the L∞ norm,
which is often needed in real applications, but is much more difficult to
handle because it is not differentiable.
Weighted sum of compliances: find the best compromise for a struc-

ture successively submitted to several loadings fi :

J(ω) = Ci χ(x)Ae(ui ) : e(ui )dx

i Ω
= Ci A−1 σi : σi dx (3)

i Ω
= Ci fi · ui ds
i ∂Ω
Each displacement field ui is solution of the elasticity system for the

loading case fi .
Eigenfrequencies: one relevant problem is to maximize the smallest eigen-

values of the structure. Using Rayleigh quotient, the problem is similar to
the compliance case, since, for example, the first eigenvalue ω1 writes:

A e(v) : e(v) dx
ω
ω12 = min .
v∈H 1 (ω)N , v=0 2
v=0 on ΓD ρ|v| dx
ω
Robust optimization (worst case compliance):

J(ω) = max (f + α) · u ds
α admissible perturbation ∂ω
Optimization of the buckling criterion:

1
J(ω) =
λ
where λ is the critical buckling load (solution of a generalized eigenvalue
problem).
2.3 Compliance case

Volume constraint
To get rid of the volume constraint, we introduce a positive Lagrange
multiplier and, for a given value of , look for a minimizer of
inf {c(ω) + |ω|}.

ω⊂Ω
138 F. Jouve
Since the compliance is a measure of the rigidity, it is obvious that, starting

from a given structure and adding some material to a given layout, will
lead to an heavier and more rigid structure. Thus we optimize the sum
of two contradictory terms and try to find the best compromise between
rigidity and weight, being a balance parameter between the two terms. In
the compliance case this problem becomes trivial as → 0 since a solution
is ω = Ω. If a strict volume constraint is needed, the parameter can be
iteratively adjusted (for example using an augmented Lagrangian algorithm)
to reach a target volume at convergence.
For other objective functions like (2) or eigenfrequencies maximization,
the relationship between the performance and the volume of the structure
is not straightforward (and monotone). The use of the Lagrange multiplier
is not relevant any more. A real volume constraint is preferred instead, if
it makes sense for the problem at hand. No volume constraint is another
possible option, depending on the target application.
Relaxation of the problem

We write the complementary energy principle:

c(ω) = f · u ds = 8 min A−1 τ : τ,
>
< −div τ =0
∂Ω > in ω ω
τ n=f on ∂ω∩∂Ω
>
>
:
τ n=0 on ∂ω\∂Ω
where τ is the statically admissible stress tensor. If τ is extended by 0 inside

holes, the above expression can be rewritten on the whole domain Ω:

c(ω) = f · uds = 8 min (χ(x)A)−1 τ : τ,
∂Ω < −div τ =0 in Ω Ω
: τ n=f on ∂Ω
and the minimization of c(ω) over all ω ⊂ Ω leads to a double minimiza-

tion formulation. The two minimizations can be switched, and the new
expression writes

inf {c(ω) + |ω|} = 8 inf F (τ )
ω⊂Ω < −div τ =0 in Ω Ω
: τ n=f on ∂Ω

A−1 τ : τ + if τ = 0
where F (τ ) =
0 if τ = 0
This problem is still ill posed, since we have only made some algebraic ma-
nipulations up to now: the minimizing sequences of characteristic functions
can converge to something else than a characteristic function. Mechanically

it means that many small holes may be more effective than a big one having
the same volume: composite material with micro-holes may be solutions (cf.
Figure 3 where finely layered material have a better performance).
Two kinds of remedies:
• Fight: reduce the space of admissible shape to be sure that the com-
puted solution belongs to it: add constraints and look for classical
solutions (cf. Section 3: level set method).
• Give up: admit there is a problem and allow weird things: enlarge
the space of admissible solution to allow composite materials with
micro-holes. It is the key idea of the homogenization method.
≥ ≥ ≥ ?
Figure 3. Uniaxial traction: fine layers are more effective than large ones
for the same volume of material.
2.4 Homogenization
Homogenization is the rigorous way of computing the effective (macroscopic)
properties of heterogeneous and composite media.
For an heterogeneous medium where ε is the typical size of small details,
the direct computation is too expensive. We look for an equivalent medium,
at the macroscopic scale, that will be obtained as the limit material when
ε → 0.
1d scalar example
Consider a > 0, a 1-periodic function and the following problem, with ε > 0
140 F. Jouve
a small parameter:
⎧
⎨ d x du
− a = f in ]α, β[
dx ε dx
⎩
u(α) = u(β) = 0
The limit problem as ε → 0 writes:

⎧
⎨ d ∗ du
− a = f in ]α, β[
dx dx
⎩
u(α) = u(β) = 0
where
1
a∗ = β
1 dy
β−α α a(y)
The homogenized conductivity is the harmonic mean value and not the
arithmetic one!
Homogenization - Periodic case

If the microstructure is periodic, the homogenized coefficients can be
computed solving some partial differential equations posed on the period-
icity cell. In practice this computation is not explicit, except in some very
particular cases.
There are two steps in the computation of the homogenized properties:
• Formal asymptotic expansion of the problem and identification of the
factors of the successive powers of ε. It leads to the homogenized equation
and the cell problems that are used to compute the homogenized coefficients.
• Proof of the convergence of the ε−system to the limit homogenized
equation (e.g. using the Tartar energy method) as ε → 0, and estimation of
the error between the exact solution and the homogenized one for a given
value of ε.
Let Y = (0, 1)d and consider a Y −periodic Hooke’s tensor A(y) for the
following elasticity problem with oscillating coefficients:
x
−div A e(uε ) = f in Ω
ε
uε = 0 on ∂Ω,
We assume (formally) the following ansazt for the two-scales asymptotic
expansion:

+∞ x
uε (x) = ui x, ,
i=0
ε
where ui (x, y) is a function of both variables x and y, periodic in y with
period Y . The following derivation rule is used
x x
∇ ui x, = ε−1 ∇y ui + ∇x ui x, ,
ε ε
where ∇y and ∇x denote respectively the partial derivative with respect to
the first and second variable of ui (x, y). This series is then plugged into the
equation
x
f (x) = −ε−2 [divy (Aey (u0 ))] x,
ε
x
−ε−1 [divy (A(ex (u0 ) + ey (u1 ))) + divx (Aey (u0 ))] x,
ε

+∞ x
− εi [divx (A(ex (ui ) + ey (ui+1 ))) + divy (A(ex (ui+1 ) + ey (ui+2 )))] x,
i=0
ε
Identifying each coefficient as an individual equation yields a cascade of
equations that leads to the homogenized equation
−divx (A∗ ex (u(x))) = f (x) in Ω,
and the homogenized Hooke’s tensor A∗ is given by

A∗ijkl = (A(y)ey (wij )kl + Aijkl (y)) dy.
Y
For each (i, j) ∈ {1, . . . , d}2 , the fields wij are solutions of the cell problem:

−divy (A(y) (eij + ey (wij (y)))) = 0 in Y
y → wij (y) Y -periodic,
where eij are the elements of the canonical basis of the space of symmet-
ric matrices in dimension d. Thus, d(d + 1)/2 different cell problems are
necessary to solve, in order to compute A∗ .
142 F. Jouve
Homogenization - Non periodic case

In non periodic settings, there is no explicit characterization of the ho-
mogenized Hooke’s tensor. But the limit problem can be established as well,
using the H-convergence (or G-convergence) theory.
Moreover, if θ ∈ [0, 1]

all Hooke’s tensors that can be built by homogenization
Gθ = ,
mixing 2 materials A1 and A2 in proportions θ and (1 − θ)
some bounds of the energy over Gθ can be computed and are achieved,
for example, by a particular and explicit class of periodic materials: the
sequential laminates.
Sequential laminates
It is a particular class of composites, built sequentially. Consider two
given materials. In a first step they are stacked in successive layers along
one given direction, with a given proportion of each of them. It leads to the
rank 1 laminate, whose mechanical properties can be computed explicitly,
function of the two given materials and the lamination parameters. Then
this rank 1 laminate can be stacked again with one of the initial pure ma-
terials, with a different stacking direction and different proportions, to get
a rank 2 laminate. Mechanical properties of the rank 2 material can be ex-
plicitly computed as well, and the same occurs for any rank of lamination.
Sequential laminates are useful in two ways:
• They are described by a small number of parameters: the rank of
lamination, the global proportion of each constitutive material, the
lamination directions and proportions along each direction.
• There are explicit formulas for the homogenized properties, functions
of these characterizing parameters.
Optimal bounds of the energy

In 2d, consider a given 2 × 2 symmetric matrix τ whose eigenvalues
are τ1 and τ2 (the principal stresses). Let A be a given isotropic Hooke’s
tensor, characterized by its bulk modulus κ and its shear modulus μ, and
let θ ∈ [0, 1] be a given constant. If Gθ is the set of all materials that can
be built by mixing A and void in proportions θ and (1 − θ), then
(κ + μ)(1 − θ)
min A∗−1 τ : τ = A−1 τ : τ + (|τ1 | + |τ2 |)2
A∗ ∈Gθ 4κμθ
This minimum is attained for a rank 2 laminate whose lamination directions

|τ1 |
are the eigenvectors of τ . The optimal lamination proportions are
|τ1 | + |τ2 |
|τ2 |
and .
|τ1 | + |τ2 |
Then the optimization over θ ∈ [0, 1] is easy and explicit:

(κ + μ)
θoptimal = min 1, (|τ1 | + |τ2 |)
4κμ
In 3d there is an analogous result, although involving more complicated

formulas (4). The energy bounds are attained for a rank 3 laminate (or
rank 1 or 2 in some particular cases) and the lamination directions are
aligned with the principal directions of τ .
Remark 1: there is no uniqueness of the optimal microstructures: there
exist others, less usable in practice, optimal microstructures: for example
confocal ellipsoids (2d, 3d) or Vigdergauz inclusions in 2d (cf. Figure 4).
Remark 2: there is no uniqueness of the optimal laminates: for example

if τ = pI (stress tensor corresponding to an hydrostatic pressure p), the
lamination directions can be any directions. For a given τ in 2d, it is also
possible to find rank 2n optimal laminates, for any positive integer n.
2.5 Relaxation by homogenization

Definition: enlarge, just as much as necessary but not more, the set of
admissible shapes and change the less possible the cost function so that the
modified problem has a solution in the new set of admissible solutions, and
144 F. Jouve
Figure 4. Examples of Vigdergauz inclusions in a square cell.
the minimizing sequences of the initial problem converge to a solution of

the relaxed problem.
We recall the initial formulation of the shape optimization problem in the
compliance case:

−1
∞
inf J(χ) = (χ(x)A) τ : τ dx + χ(x)dx,
χ∈L (Ω;{0,1}) Ω Ω
where τ is solution of

−div τ = 0 in Ω
τn = f on ∂Ω,
Then the relaxed formulation can be written:

∗ ∗ ∗ −1
min
∗
J (θ, A ) = (A ) τ : τ dx + θ(x)dx,
(θ,A )∈CD Ω Ω
where τ is solution of

−div τ = 0 in Ω
τn = f on ∂Ω,
and CD denotes the set of the generalized shapes that includes composite
materials made by homogenization. The new set of admissible shapes is

CD = θ ∈ L∞ (Ω; [0, 1]) , A∗ (x) ∈ Gθ(x) , ∀x ∈ Ω ,
where for each given θ(x) ∈ [0, 1], Gθ is the set of all homogenized Hooke’s
tensors obtained by mixing A and void in respective proportions θ and
(1 − θ).
Unfortunately Gθ is unknown for elasticity (although it can be fully
characterized by laminates for conductivity problems). But it can be proved
that Gθ is the set of all the Hooke’s tensors obtained by mixing A and void
in a periodic way with proportions θ and (1 − θ).
Relaxation does not change (too much) the problem:
inf J(χ) = min J ∗ (θ, A∗ )
χ∈L∞ (Ω;{0,1}) (θ,A∗ )∈CD
All the homogenized solutions can be attained as limits of minimizing se-

quences of classical designs. If the initial problem admits a classical (com-
posite free) solution, it is solution of the relaxed problem as well.
The relaxed formulation is expressed as a minimization problem over a
set that is partially unknown. But the two minimizations, over the design
variables (θ, A∗ ) and the admissible stresses τ , can be interchanged. More-
over, thanks to the local character of the G−closure, the minimization over
the admissible designs can be put under the integral:

8 min min (A∗−1 τ : τ + θ)
< −div τ =0 in Ω Ω A∗ ∈Gθ
: τ n=f on ∂Ω 0≤θ≤1
Then, for a given stress tensor τ , minimizing A∗−1 τ · τ over Gθ amounts to

compute the bound of the energy. It can be done explicitly using sequential
laminates.
2.6 Numerical algorithm

The above developments lead to a straightforward algorithm for the ho-
mogenization method in the compliance case. As the relaxed formulation
involves two minimizations, they are performed alternatively in the algo-
rithm:
8 min min (A∗−1 τ : τ + θ)
< −div τ =0 in Ω Ω A∗ ∈Gθ
: τ n=f on ∂Ω 0≤θ≤1
• First minimization: solve an elasticity problem for a given A∗ (x). It

is done, for example, by a classical Finite Elements Method (or any other
efficient method).
• Second minimization: minimization over the shape parameters (θ, A∗ )
for given τ (x): it is done explicitly using optimality conditions and optimal
146 F. Jouve
composites (e.g. laminates).

The algorithm is then quite simple. It belongs to the class of alternate
directions algorithms:
1. Initialization of the design parameters (θ0 , A∗0 ) (e.g. θ0 ≡ 1 and A∗0 ≡
A).
2. Iterations until convergence (n being the iteration number):
a) Compute τn solving the elasticity system for the design variables
(θn−1 , A∗n−1 ).
b) Update the design variable (θn , A∗n ) using the explicit optimal
composites for the stress tensor → explicit expression as a func-
tion of τn .
Remarks:
• By construction, the energy is decreasing at each iteration.
• The solution is not sensitive to initial guess and to mesh refinement.
• Convergence is considered to be reached when the energy and the shape
parameters are stationary enough.
• The Lagrange parameter , piloting the global volume of the struc-
ture, can be adjusted numerically to satisfy an explicit volume constraint
(augmented Lagrangian algorithm).
• “Void” is replaced by a “weak” material in order to have a valid system
on all Ω.
• In 2d, orthogonal rank 2 laminates of material and void are degen-
erate and do not support shear stresses. Their Hooke’s tensor have to be
regularized in order to perform numerical computations.
• In 2d, with square elements, some “checkerboard instabilities” may
appear. It is a numeric artifact related to the discretization of the problem.
They can be easily filtered.
• After convergence to a generalized solution (involving laminates) it is
possible to penalize the composites to find a classical solution, free of homog-
enized material, made of pure material and void. It is a numerical procedure
and gives mesh dependent results, related to the spatial discretization and
the amount of details that can be resolved by the mesh, while the homoge-
nized design is mesh independent (cf. Figure 6). But the performance of the
penalized design is usually very close to the one of the generalized design
(cf. Figure 5).
• Displacement u is discretized by the FEM, and the shape parameters
(θ, A∗ ) are taken as constant in each cell.
2.7 Other methods related to homogenization

Rectangular holes + rotation in each cell. Homogenization computed
numerically and tabulated. It is the first effective practical attempt to use
Figure 5. Cantilever: setting of the problem and convergence history (left);

composite and penalized designs (right).
Figure 6. Stability by mesh refinement: composite designs (above) and

penalized designs (below) for different mesh sizes.
homogenization for structural optimization (Bendsoe-Kikuchi 88 (22)).

Drawbacks: No explicit properties of the composites and suboptimal com-
posites (less efficient).
148 F. Jouve
Isotropic composites: Consider only the isotropic composite in Gθ . The

energy bound over this subset is attained respectively by rank 3 and 6 lami-
nates in 2d and 3d, with given directions and fixed proportions (respectively
1/3 and 1/6).
Advantage: only one remaining shape parameter (θ) → simpler method.
Drawback: suboptimal composites leading to poorer solutions.
Convexification: instead of using real composites with macroscopic prop-
erties computed by homogenization, use a single density parameter θ(x) ∈
[0, 1] at each point x ∈ Ω. Mathematically, it amounts to convexify the
problem (while the relaxed formulation presented above is mathematically
the quasi-convexification of the initial problem).

min min ((θA)−1 τ : τ + θ)
τ Ω 0≤θ≤1
Advantage: explicit and simple optimization of θ (the only shape parame-
ter).
Drawback: Some admissible shapes lie outside of Gθ and some composites
(e.g. all the anisotropic ones) cannot be described by an isotropic law.
In fact, the set of Hooke’s tensors {θA, θ ∈ [0, 1]} is neither included into
{Gθ , θ ∈ [0, 1]}, nor containing it. The local information on the lamina-
tion directions is lost and the resulting algorithm gives poorer solutions (cf.
Figure 7).
Variant: the SIMP method
min ((θp A)−1 τ : τ + θ),

0≤θ≤1
where p is a (quite) secret parameter. Adjusting p during the algorithm
allows to penalize the “grey zones” progressively and get classical solutions
at convergence. It is a very popular method, and quite efficient, although
involving suboptimal composites (21)(23).
2.8 Multiple loads

Consider the following objective function

J(ω) = A−1 σi : σi dx = fi · ui ds
i Ω i ∂Ω
where fi are different loadings applied successively to the structure. The

relaxed problem can be established in the same way than in the single load
compliance case:

∗ ∗
∗ −1
min
∗
J (θ, A ) = (A ) τi : τi dx + θ(x)dx,
(θ,A )∈CD Ω Ω
i
1.5
Relaxation
1.4 Convexification
Penalizing from the start
Cost function 1.3
1.2
1.1
1
0 50 100 150
Iterations
Figure 7. Arch problem: homogenization solutions are presented on Fig-

ure 6. Above: convexified problem, composite solution; middle: convexified
problem, penalized solution; bottom: homogenization + penalization from
the start of the algorithm.
where ∀i, τi is solution of

−div τi = 0 in Ω
τi n = fi on ∂Ω,
It can be proved (Francfort-Murat-Tartar (45)) that there exists a rank 3

laminate in 2d and a rank 6 laminate in 3d that attains the bound of a
sum of energies. But these laminates cannot be computed explicitly like in
the single load case. Thus the explicit phase of optimization over the shape
parameters is replaced by a numerical procedure:
• We look for a quasi-optimal composite in the class of the sequential
laminates.
• The number of lamination directions q is fixed as well as the directions:
typically q = 12 directions in 2d and q = 100 in 3d are enough to cover
reasonably uniformly the whole half space.
• The remaining parameters θ (global proportion) and (mi )1≤i≤q are
optimized in each cell of the mesh.
• The optimization of mi is done by a gradient method. It is efficient
since the problem is convex in mi for fixed θ and lamination directions.
• The optimization of θ remains explicit.
150 F. Jouve
1 2 3 4 5
Solution for loading 1
Solution for the 5 loadings together Solution for loading 2
Multiple loads solution Solution for loading 3

Figure 8. Multiple loads example.
2.9 Another cost function depending on the displacement
1/α
J(ω) = k(x)|u(x) − u0 (x)|α dx
ω
where u0 is a given target displacement, α ≥ 2 and k ∈ L∞ (Ω) is a given

bounded function used to localize the computation of the objective function.
This cost function can be used for two objectives:
• minimize the displacement over a given area (u0 = 0) → more local
criterion than the compliance to optimize rigidity;
• minimize the difference between the displacement field and the target
displacement u0 . It can be used as well to maximize the displacement along
a given direction. It has been applied, for example, to the optimization of
compliant mechanisms and MEMS.
Relaxed formulation (total relaxation)

The relaxed problem can be obtained in the same way than in the compli-
ance case and writes:

min J ∗ (θ, A∗ ),
(θ,A∗ )∈CD
α1
∗ ∗
J (θ, A ) = θ(x)k(x)|u(x) − u0 (x)| dx α
+ θ(x) dx,
Ω Ω
where u is solution of

−div (A∗ e(u)) = 0 in Ω
A∗ e(u) · n = f on ∂Ω,
and CD is the set of admissible generalized shapes

CD = θ ∈ L∞ (Ω; [0, 1]) , A∗ (x) ∈ Gθ(x) , ∀x ∈ Ω ,
But now – and it is the case for any objective function different from the
compliance – the optimal composites cannot be characterized explicitly.
We do not even know if optimal homogenized materials can be reached
by sequential laminates. Thus we have to use a numerical procedure to
optimize, in an approximate way, the local composites.
Partial relaxation
min J ∗ (θ, A∗ ),
(θ,A∗ )∈LD
with
LD = θ ∈ L∞ (Ω; [0, 1]) , A∗ (x) ∈ Lθ(x) , ∀x ∈ Ω ,
where

set of all the Hooke’s tensors obtained by sequential
Lθ(x) = .
lamination of A and void in proportions θ and (1 − θ)
Remarks:
• In the compliance case (or multiple loads case optimizing a weighted
sum of compliances), the partial relaxation is equivalent to the genuine
relaxation.
• For all other objective functions, it is not true (it may be true but not
proved yet...)
• An ill posed problem is replaced by another, a priori, ill posed problem.
We expect that it is at least “less ill posed”...
• In practical computations, the set of admissible composites is further
restricted to rank q laminates (q is fixed) with fixed lamination directions,
and the rest of the algorithm is similar to the multiple loads case.
152 F. Jouve
Adjoint state
Since we derive, with respect to shape parameters, a function depending
on the solution of a partial differential equation posed on this domain, the
computation of the adjoint state is needed.
Remark: the compliance case is self-adjoint. That is why no adjoint state
is needed for the compliance problem, and why it is so popular.
For u solution of the elasticity problem, the adjoint state p is solution of

−div (A∗ e(p)) = cα θk|u − u0 |α−2 (u − u0 ) on Ω
+boundary conditions
1−α
α
where cα = θ(x)k(x)|u(x) − u0 (x)| dx
α
.
Ω
∗
p allows to compute the derivatives of J with respect to the shape pa-
rameters θ and (mi ), and obtain numerically, through a classical descent
algorithm, the optimal parameters for the class of laminates chosen.
Numerical algorithm (projected gradient like)

• Initialization of the shape parameters θ0 and (mi )0 in each cell.
• Iteration until convergence (n ≥ 0 is the iteration number):
* Computation of un (displacement) and pn (adjoint state) for fixed

θn and (mi )n .
* Update of the shape parameters:
θn+1 = max(0, min(1, θn − tn ∇θn Jn∗ ))

mi,n+1 = max(0, mi,n − t n ∇mi,n Jn∗ )

* The constraint i mi,n = 1 is adjusted by a Lagrange multiplier.
* Descent steps tn and t n , adjusted using a line search method, are chosen
such that J ∗ (θn+1 , mi,n+1 ) < J ∗ (θn , mi,n )
3 Level set method

3.1 Shape sensitivity optimization versus Topology optimization
Roughly speaking, the shape optimization techniques can be separated
into two classes:
• methods based on the shape sensitivity, where a given initial shape is
deformed into another one, hopefully “better”;
• topology optimization methods that are looking for a shape without
restrictions on its topology.
These two classes of methods have advantages and drawbacks.
Shape sensitivity optimization

Based on Hadamard’s ideas, at the beginning of the 20th century, the
shape sensitivity methods have been developed by many authors (Murat-
Simon, Pironneau, Nice’s school, etc.) in the 70’s and 80’s.
They face the intrinsic ill-posedness of the structural optimization prob-
lems and the algorithms are expected to fall into local minima and show
no convergence under mesh refinement. In addition, at numerical level, the
necessary remeshing of the shape at each step of the iterative algorithm is
usually costly in 2d, and very difficult and costly in 3d. The remeshing pro-
cess may also induce additional instabilities of the algorithms. Moreover,
topology changes are very difficult to handle.
But this class of method is very general and can quite easily address a
wide range of direct problems, objective functions and constraints.
Topology optimization
The topology optimization methods can be classified in three distinct
families:
• Methods based on homogenization, like the one described in the Section
2. These techniques involve the relaxation of the problem (or its partial
relaxation for the SIMP method) to enlarge the set of admissible solutions,
and look for a density of material at each point of the admissible domain,
discretized on a fixed grid. The main advantages of this approach is to
avoid local minima, mesh and initialization dependency, and topological
restrictions.
Their main weaknesses lie in the difficulty to deal with complex direct
problems, objective functions and geometrical constraints (e.g. thickness
constraints).
• Evolutionary algorithms: purely numeric approach, treating the dis-
cretized problem.
154 F. Jouve
• Topological asymptotics, topological gradient that are strongly related

to shape sensitivity but allow naturally topological changes.
All these methods capture shapes on a fixed mesh, leading to cheap and
quite stable algorithms.
Level set method

The purpose of the level set method is to try to combine some advan-
tages of the shape sensitivity method and the topology approach. Its main
characteristics are:
• Based on the shape derivative:
* easy handling of various objective functions;
* straightforward to apply to any direct problem (e.g. nonlinear).
• Shape representation by the zero level set of a scalar field on a fixed
mesh (Osher-Sethian (65)):
* moderate cost;
* no numerical instabilities due to remeshing;
* easy topology changes.
• Remaining drawbacks:
* reduction of topology (rather than variation) in 2d. It means that
cancelling holes is an easy and natural process, while creating new ones is
more difficult;
* may be trapped in local minima.
Coupling the level set method with the topological gradient is a way to
circumvent these two remaining drawbacks.
3.2 Existence theory for shape optimization problems in the

compliance case
This section summarizes the main existence theorems for shape optimization
problems. Without further assumptions, the minimal conditions to impose
is to look for shapes in the following set

Uad = ω ⊂ Ω, vol(ω) = V0 , ΓN ⊂ ∂ω, meas(ΓD ∩ ∂ω) > 0
where Ω is open and bounded in RN .

Usually this minimization problem has no solutions in Uad . Under additional
conditions some existence results can be proved. We recall four of them:
Existence theorem 1: Uniform cone condition

Definition:
for θ ∈ (0, π/2), h > 0, ξ ∈ RN , |ξ| = 1,
the cone of direction ξ, angle θ and height h is defined by
C(θ, h, ξ) = {x ∈ RN , x.ξ > |x| cos θ, |x| < h}
Definition: ω ⊂ RN satisfies the uniform cone condition if and only if

∃θ ∈ (0, π/2), h > 0, r > 0 such that ∀x ∈ ∂ω, ∃ξ(x) such that
∀y ∈ B(x, r) ∩ ∂ω, y + C(θ, h, ξ(x)) ⊂ ω
Remark: the cone condition does not imply regularity (corners allowed),
but forbids cups and rapidly oscillating boundaries.
Theorem [Chenais 75 (35)]:
If
Uad1 = {ω ∈ Uad , ω verifies the uniform cone condition}
then
inf J(ω)
ω∈Uad1
admits at least one solution.

Remark: not useful for applications
Existence theorem 2: Regularity condition

Definition: diffeomorphisms of RN :
T = {T : RN → RN , (T − Id) ∈ W 1,∞ and (T −1 − Id) ∈ W 1,∞ }
If ω0 ⊂ RN is a given open reference domain, bounded and smooth (e.g.

C 1 ) and
D(ω0 ) = {ω ⊂ RN , ∃T ∈ T such that ω = T (ω0 )}
156 F. Jouve
then for ω1 , ω2 ∈ D(ω0 ) we can define the pseudo-distance

d(ω1 , ω2 ) = inf ||T − Id||1,∞ + ||T −1 − Id||1,∞
T ∈T ,T (ω1 )=ω2
Theorem [Murat-Simon 76 (58)]:

For a given C > 0, small enough, if
Uad2 = {ω ∈ Uad ∩ D(ω0 ), d(ω, ω0 ) ≤ C}
then
inf J(Ω)
ω∈Uad2

Remark: not useful for applications
Existence theorem 3: Additional regularizing term

Theorem [Ambrosio-Buttazzo 93 (17)]:
˜
If J(ω) = J(ω) + P (∂ω), where P (∂ω) is an ad-hoc regularizing term in-
volving the perimeter of the shape, then
˜
inf J(ω)
ω∈Uad

Remark 1: avoids rapidly oscillating boundaries (like the cone condition).
Remark 2: useful in practice. The perimeter is easy to compute (e.g. using
the level set representation).
Existence theorem 4: Topological condition in 2d

Theorem [Chambolle-Larsen 03 (34)]: For a given k ∈ N,
Uad4 = {ω ∈ Uad , Ω \ ω has less than k connected components}
then
inf J(ω)
ω∈Uad4

Remark 1: strongly restricted to the 2d case.
Remark 2: not (very) useful for applications.
3.3 Shape derivative (Murat-Simon (58), Céa (32))
Let ω0 be a given domain. We are inter-

ested in variations of the form

ω = {x + θ(x) | x ∈ ω0 } = Id + θ ω0
with θ ∈ W 1,∞ (Rd ; Rd ).

θ(x) is a vector field that maps the refer-
ence domain ω0 into ω.
Lemma: For any θ ∈ W 1,∞ (Rd ; Rd ) such that θ W 1,∞ (Rd ;Rd ) < 1, ( Id + θ)
is a diffeomorphism in Rd .
Definition: The shape derivative
of ω → J(ω) at ω0 is the Fréchet deriva-
tive of θ → J ( Id + θ)ω0 at 0.
The following lemma ensures that the

shape derivative J (ω0 )(θ) depends only on
θ·n on the boundary ∂ω0 . Thus, admissible
deformations can be restricted to normal
ones.
Lemma: Let ω0 be a smooth bounded
open set and J(ω) a differentiable function
at ω0 . Its derivative satisfies
J (ω0 )(θ1 ) = J (ω0 )(θ2 )
if θ1 , θ2 ∈ W 1,∞ (RN ; RN ) are such that

θ2 − θ1 ∈ C 1 (RN ; RN )
θ1 · n = θ2 · n on ∂ω0 .
Examples of shape derivatives

Objective-function defined in the domain:
let ω0 be a smooth bounded open set of class C 1 of RN and f (x) ∈ W 1,1 (RN ).
If J is defined by

J(ω) = f (x) dx,
ω
158 F. Jouve
then J is differentiable at ω0 and

J (ω0 )(θ) = div θ(x) f (x) dx = θ(x) · n(x) f (x) ds
ω0 ∂ω0
for all θ ∈ W 1,∞ (RN ; RN ).

Objective-function defined on the boundary:
let ω0 be a smooth bounded open set of class C 1 of RN and f (x) ∈ W 2,1 (RN ).
If J is defined by
J(ω) = f (x) ds,
∂ω
then J is differentiable at ω0 and

J (ω0 )(θ) = ∇f · θ + f div θ − ∇θn · n ds
∂ω0
for all θ ∈ W 1,∞ (RN ; RN ). Moreover if ω0 is of class C 2 , then

∂f
J (ω0 )(θ) = θ·n + Hf ds,
∂ω0 ∂n
where H is the mean curvature of ∂ω0 defined by H = div n.
Shape derivative of the compliance If

J(ω) = f · u ds = A e(u) : e(u) dx,
∂ω∪ΓN ω
then

∂(f · u)
J (ω0 )(θ) = 2 + Hf · u − Ae(u) : e(u) θ · n ds, (4)
∂ω0 ∂n
where u is the displacement field in ω0 , and H is the mean curvature of ∂ω0 .

No adjoint state involved since the compliance problem is self-adjoint.
Shape derivative of the least-square criterion

To give an example of shape derivative involving the adjoint state, let
us consider the objective function depending on the displacement presented
in Section 2.2: 1/α
J(ω) = k(x)|u − u0 |α dx ,
ω
then

∂(g · p) C0
J (ω0 )(θ) = + Hg · p − Ae(p) : e(u) + k|u − u0 |α θ·n ds,
∂ω0 ∂n α
where the state u is solution of the elasticity system and the adjoint state
p is solution of the following adjoint problem:
⎧
⎨ −div (A e(p)) = C0 k(x)|u − u0 |α−2 (u − u0 ) in ω0
p = 0 on ΓD (5)
⎩
Ae(p)n = 0 on ΓN ∪ ∂ω0 ,
1/α−1
with C0 = k(x)|u(x) − u0 (x)|α dx .
ω0
Formal computation of the shape derivative (Céa 86)

The expression of the shape derivative is not always straightforward.
The following tool, although formal, can be used to do it. It gives a general
method to compute explicitly the shape derivative.
Consider a general objective function

J(ω) = j x, u(x) dx + l x, u(x) ds,
ω ∂ω
2
Introduce the Lagrangian defined for (v, q) ∈ H 1 (Rd ; Rd ) by

L(ω, v, q) = j(v) dx + l(v) ds + Ae(v) : e(q) dx − q · f dx
ω ∂ω ω ω

− q · g ds − q · Ae(v)n + v · Ae(q)n ds.
ΓN ΓD
v and q belong to a functional space that does not depend on ω, so we can

apply the usual differentiation rule to L.
The stationarity of the Lagrangian gives the optimality conditions of the
minimization problem.
First, derivating with respect to its third variable q gives the state equa-
tion: and its boundary conditions. For any test function φ we have:

∂L
(ω, u, p), φ = 0 = − φ · div (Ae(u)) + f dx
∂q ω

+ φ · Ae(u) n − g ds
Γ
N
− u · Ae(φ)n ds.
ΓD
160 F. Jouve
Then by varying φ:
• φ with compact support in ω leads to the state equation.
• Vary the trace of φ on ΓN leads to the Neumann boundary conditions
on u.
• Vary Ae(φ) n on ΓD leads to the Dirichlet boundary conditions on
u.
Then the Lagrangian is derivated with respect to its second variable v:

∂L
(ω, u, p), φ = 0 = j (u) · φ dx + l (u) · φ ds
∂v ω ∂ω

+ Ae(φ) · e(p) dx
ω

− p · Ae(φ)n + φ · Ae(p)n ds.
ΓD
An integration by parts leads to

∂L

(ω, u, p), φ = j (u) − div (Ae(p)) · φ dx + φ · Ae(p)n + l (u) ds
∂v ω ΓN

+ φ · l (u) − p · Ae(φ)n ds.
ΓD
Again, varying φ gives the adjoint problem and its boundary conditions:
• Taking φ with compact support in ω leads to the adjoint equation:
−div (Ae(p)) = −j (u) in ω.
• Vary the trace of φ on ΓN leads to the Neumann boundary conditions for

p:
Ae(p)n = −l (u) on ΓN .

• Vary Ae(φ) n on ΓD leads to the Dirichlet boundary conditions for p:
p=0 on ΓD .
We have found a well-posed boundary value problem for the adjoint state
p.
The shape derivative of the objective function is then obtained by dif-

ferentiating L with respect to ω in the direction θ:

∂L
(ω, u, p)(θ) = θ · n j(u) + Ae(u) : e(p) − p · f ds
∂ω ∂ω

∂l(u)
+ θ·n + H l(u) ds
∂ω ∂n

∂(g · p)
− θ·n + H g · p ds
ΓN ∂n

∂h
− θ·n + Hh ds,
ΓD ∂n
where h = u · Ae(p)n + p · Ae(u)n and H = div n is the mean curvature of
the boundary.
Taking into account the boundary conditions u = p = 0 on ΓD gives (after
computation)

∂L ∂(g · p)
(ω, u, p)(θ) = θ · n j(u) + Ae(u) : e(p) − p · f − − H g · p ds
∂ω ΓN ∂n

+ θ · n j(u) − Ae(u) : e(p) ds
ΓD

∂l(u)
+ θ·n + H l(u) ds.
∂ω ∂n
Remark: this computation is only valid for a domain (one phase + void).
In the two phases case, the spirit is the same but the result is much more
complicated (see (14)).
3.4 Front propagation by level set

The key point of the level set method is the coupling of the shape derivative
and the level set representation. Shapes are never meshed, but captured
on a fixed mesh of a fixed given domain Ω enclosing all the admissible
shapes. All over the algorithm, the geometry of the shape ω ⊂ Ω can be
kept implicit through the level set function ψ defined as a scalar function
over Ω. By convention we set:
⎧
⎨ ψ(x) = 0 ⇔ x ∈ ∂ω ∩ Ω
ψ(x) < 0 ⇔ x ∈ ω
⎩
ψ(x) > 0 ⇔ x ∈ (Ω \ ω)
With this convention, the exterior normal to ω writes n = ∇ψ/|∇ψ|,
and the mean curvature is obtained by H = div n.
162 F. Jouve
These formula make sense everywhere in Ω, not only on the boundary

∂ω. The level set representation provides a natural extension of quantities
a priori defined only on the boundary. As shown on Figure 9, the level set
representation allows easy topological changes due to the implicit definition
of the boundary.
Figure 9. Topology changes are easily handled by the level set representa-
tion.
3.5 Hamilton-Jacobi equation

We can now describe a gradient method for the minimization of an ob-
jective function J(ω). Assume that the shape derivative can be expressed
as
J (ω)(θ) = v θ · n ds,
∂ω
where v is given by a formula of the type established in the examples above.

If regularity problems are neglected, a valid descent direction is for example
θ = −v n. (6)
If we use this vector field to evolve the shape ω as

ωt = ( Id + tθ)ω,
where t > 0 is the descent parameter, we obtain formally

J(ωt ) = J(ω) − t v 2 ds + O(t2 )
∂ω
that ensures that the objective function is decreasing as soon as the descent
parameter is small enough.
Suppose that the shape ωt evolves with a pseudo-time t and a normal
velocity V (t, x). Then, as the boundary of the shape is characterized by the
level set 0 of ψ, we have

ψ t, x(t) = 0 for all x(t) ∈ ∂ωt .
Differentiating with respect to t yields

∂ψ ∂ψ
+ ẋ(t) · ∇ψ = + V n · ∇ψ = 0.
∂t ∂t
As n = ∇ψ/|∇ψ| we obtain
∂ψ
+ V |∇ψ| = 0.
∂t
This Hamilton-Jacobi equation is valid on the whole domain Ω, not only
on the boundary ∂ω, assuming that the velocity V is known everywhere.
Thus the description of the boundary of ω can remain implicit during the
algorithm. As pointed out above, the choice of V = −v is a possible one,
but it may be useful to use a proper regularization of v instead.
3.6 Numerical algorithm

1. Initialization of the level set function ψ0 .
2. Iterations until convergence for n ≥ 1:
a) Computation of un , and pn if necessary, by solving a linearized
elasticity problem on the shape defined by ψn . Computation
of the shape derivative for un and pn that provides a descent
direction (normal velocity Vn )
b) Transport of the shape with the speed Vn (Hamilton-Jacobi equa-
tion) to obtain a new shape ψn+1 . (Several successive time steps
can be applied for a same velocity field). The descent step is
controlled by the CFL condition on the transport equation and
by the decreasing of the objective function.
164 F. Jouve
c) Possible reinitialization of the level set function such that ψn+1

is the signed distance to the interface.
3. Optionally: computation of the topological gradient to guess where
holes may be dig, and return to step 2.
3.7 Topological Gradient (75)(33)

Although the level set representation allows naturally topological chan-
ges, the method could not always find the good topology if there is no
minimizing path (starting from the current state) that reaches it. It is
particularly the case in 2d, where digging a hole inside a plain zone may be
difficult for the algorithm, that can be trapped in local minima. There are
more possible paths in 3d and the problem of unreachable topology changes
is very rare.
The topological gradient is a scalar criterion to guess where it may be
useful to dig additional holes in a converged solution that could be a local
minimum. It has usually an explicit expression and is quite easy to compute.
Examples of topological gradients in bidimensional elasticity, plane strains,
with Neumann boundary conditions for the holes). u is the displacement
field, e(u) and σ(u) are respectively the strain and stress tensors. λ and μ
are the Lamé coefficients of the given isotropic material.
• Example 1 (Compliance optimization):
π(λ + 2μ)
TG = {4μσ(u) : e(u) + (λ − μ) tr(σ(u))tre(u)}
2μ(λ + μ)

• Example 2 (Minimization of k(x)|u − u0 |α ). p is the adjoint state,
Ω
solution of the problem (5)
π
TG = C0 k(x)|u(x) − u0 (x)|α
α
π(λ + 2μ)
+ {4μσ(u) : e(p) + (λ − μ) tr(σ(u)) tr(e(p))}
2μ(λ + μ)
A new hole may be dig at point x if T G is smaller than a given value.
3.8 Some technical points

• Q1 quadrangular structured meshes (2d-3d) −→ classical upwind trans-
port schemes.
• Possibilities of unstructured meshes −→ more complicated schemes for
Hamilton-Jacobi equation (Abgrall (1)).
• ψ is discretized at mesh nodes.
• Approximate solution of elasticity systems by finite elements:

−div θ(x)Ae(u) = 0 in Ω
+Boundary Conditions
where θ(x) is a piecewise constant field defined by

⎧
⎨ θ = ε (≈ 10−3 ) if ψ > 0 for all the nodes of the element
θ = ad-hoc proportion if the 0 level set goes through the element
⎩
θ = 1 si ψ < 0 for all the nodes of the element
3.9 Transport scheme for structured meshes

The Hamilton-Jacobi equation
∂ψ
+ V |∇ψ| = 0 in Ω
∂t
on a structured mesh is classically solved by an explicit upwind finite dif-

ferences scheme of first order:
ψin+1 − ψin
+ min(Vin , 0) g + (Dx+ ψin , Dx− ψin )
Δt
+ max(Vin , 0) g − (Dx+ ψin , Dx− ψin ) = 0
n
ψi+1 − ψin ψ n − ψi−1
n
with Dx+ ψin = , Dx− ψin = i , and
Δx Δx

g − (d+ , d− ) = min(d+ , 0)2 + max(d− , 0)2 ,

g + (d+ , d− ) = max(d+ , 0)2 + min(d− , 0)2 .
Second order schemes are also possible.
3.10 Complex geometries

To deal with complex geometries of the design domain (i.e. non rect-
angular geometries) there are two possibilities: either use an unstructured
mesh (triangular or tetrahedral) and special schemes for Hamilton-Jacobi
equation, or use a large rectangular bounding box and a structured rect-
angular mesh. Then introduce an additional level set to define the fixed
domain as shown on Figure below.
166 F. Jouve
3.11 Reinitialization of the level set

The level set function has to be periodically reinitialized to avoid it to
be too flat (→ poor precision on ψ) or too steep (→ poor precision on ∇ψ
i.e. the normal) after some transport steps. It is done by solving
∂ψ
+ sign(ψ) |∇ψ| − 1 = 0 in Ω,
∂t
whose stationary solution is the signed distance to the interface.
Figure 10. Example of 3d optimisation of a chair by the level set method

3.12 Concluding remarks

We have presented a short overview of two different approaches to tackle
structural shape and topology optimization problems. The first one – using
the relaxation by homogenization – is more rigorous from a mathematical
point of view, but more difficult to adapt to real problems coming from
industry. It is very efficient to treat the compliance problem but cannot
handle complex (nonlinear) direct problems, nor a large variety of objective
functions and constraints. On the other side, the level set method is more
versatile. It can be applied to almost any state equation and cost function.
Addressing complex set of constraints – for example geometric ones, like
thickness control – is also possible. It involves the use of different optimiza-
tion schemes that would have been beyond the scope of this chapter. The
algorithms presented are quite easy to implement in an existing numerical
code, since the necessary changes could be almost entirely contained in ex-
ternal modules, with only small interferences with the core of the program.
The global cost of one optimization, in term of computation time, is al-
ways a few tens of the cost of one evaluation of the direct problem, since
the optimization time is negligible for large 3d problems compared to the
time needed to solve the direct problem. It is a challenging issue to spread
these techniques outside the academic world, and implement them into com-
mercial codes used by engineers. It is an ongoing work, hopefully successful.
Acknowledgement: all this work has been done in collaboration with

Grégoire Allaire. Many other persons have also been involved at differ-
ent stages of the work. Among them I must cite, in alphabetic order:
E.Bonnetier, G.Francfort, F.de Gournay, A.M.Toader and N.Van Goethem.
Thanks to all of them.
168 F. Jouve
Bibliography
[1] Abgrall R., Numerical discretization of the first-order Hamilton-
Jacobi equation on triangular meshes, Com. Pure Applied Math.,
XLIX, 1339-1373 (1996).
[2] Achtziger W., Topology Optimization Subject to Design-Dependent
Validity of Constraints, in Topology optimization of structures and
composite continua, G.I.N. Rozvany and N. Olhoff eds., 177-191,
Kluwer Academic Publishers (2000).
[3] Alexandrov 0., Santosa F., A topology-preserving level set method for
shape optimization, J. Comp. Phys., 204/1, 121-130 (2005).
[4] Allaire G., Explicit lamination parameters for three-dimensional
shape optimization, Control and Cybernetics, 23, 309-326 (1994).
[5] Allaire G., Shape optimization by the homogenization method,
Springer Verlag, New York (2001).
[6] Allaire G., Conception optimale de structures, Collection:
Mathématiques et Applications, Vol. 58, Springer (2007).
[7] Allaire G., Bonnetier E., Francfort G., Jouve F., Shape optimization
by the homogenization method, Nüm. Math., 76, 27-68 (1997).
[8] Allaire G., De Gournay F., Jouve F., Toader A.-M., Structural opti-
mization using topological and shape sensitivity via a level set method,
Control and Cybernetics, 34, 59-80 (2005).
[9] Allaire G., Francfort G., A numerical algorithm for topology and
shape optimization, in Topology design of structures, Nato ASI Se-
ries E, M. Bendsoe et al. eds., 239-248, Kluwer, Dordrecht (1993).
[10] Allaire G., Jouve F., Minimum stress optimal design with the level
set method, Eng. Anal. with Bound. Elem., 32, 909-918 (2008).
[11] Allaire G., Jouve F., A level set method for vibration and multiple
loads structural optimization, Comp. Meth. in App. Mech. and Engrg.,
194, 3269-3290 (2005).
[12] Allaire G., Jouve F., Toader A.M., Structural optimization using sen-
sitivity analysis and a level-set method, J. Comp. Phys., 194/1, 363-
393 (2004).
[13] Allaire G., Jouve F., Toader A.M., A level-set method for shape op-
timization, C. R. Acad. Sci., Paris, Série I, 334, 1125-1130 (2002).
[14] Allaire G., Jouve F., Van Goethem N., Damage and fracture evolution
in brittle materials by shape optimization methods, J. Comp. Phys.,
230/11, 5010-5044 (2011).
[15] Allaire G., Kohn R.V., Optimal bounds on the effective behavior of a
mixture of two well-ordered elastic materials, Quat. Appl. Math., 51,
643-674 (1993).
[16] Allaire G., Kohn R.V., Optimal design for minimum weight and
compliance in plane stress using extremal microstructures, Europ. J.
Mech. A/Solids, 12(6), 839-878 (1993).
[17] Ambrosio L., Buttazzo G., An optimal design problem with perimeter
penalization, Calc. Var., 1, 55-69 (1993).
[18] Ammari H., Kang H., Reconstruction of small inhomogeneities
from boundary measurements, Lecture notes in mathematics, 1846.
Springer, (2000).
[19] Avellaneda M., Optimal bounds and microgeometries for elastic two-
phase composites, SIAM J. Appl. Math., 47(6), 1216-1228 (1987).
[20] Bendsoe M., Optimal shape design as a material distribution problem,
Struct. Optim., 1, 193-202 (1989).
[21] Bendsoe M., Methods for optimization of structural topology, shape
and material, Springer Verlag, New York (1995).
[22] Bendsoe M., Kikuchi N., Generating optimal topologies in structural
design using a homogenization method, Comp. Meth. Appl. Mech.
Eng., 71, 197-224 (1988).
[23] Bendsoe M., Sigmund O., Topology Optimization. Theory, Methods,
and Applications, Springer Verlag, New York (2003).
[24] Bensoussan A., Lions J.L., Papanicolaou G., Asymptotic Analysis for
Periodic Structures, North Holland, Amsterdam (1978).
[25] Bittencourt T.N., Ingraffea A.R., Wawrzynek P.A., Sousa J.L., Quasi-
automatic simulation of crack propagation for 2D LEFM problems,
Engineering Fracture Mechanics 55, no. 2, 321-324 (1996).
[26] Bourdin B., Numerical implementation of the variational formulation
for quasi-static brittle fracture, Interfaces and free boundaries, 9, 411-
430 (2007).
[27] Bourdin B., Francfort G. A., Marigo J.-J., Numerical experiments in
revisited brittle fracture, J. Mech. Phys. Solids 48, no. 4, 797-826
(2000).
[28] Bourdin B., Francfort G. A., Marigo J.-J., The variational approach
to fracture, Journal of Elasticity, 91, 5 148 (2008).
[29] Buhl T., Pedersen C., Sigmund O., Stiffness design of geometrically
nonlinear structures using topology optimization, Structural and Mul-
tidisciplinary Optimization, 19, 2, 93-104 (2000).
[30] Burger M., A framework for the construction of level set methods for
shape optimization and reconstruction, Interfaces and Free Bound-
aries, 5, 301-329 (2003).
[31] Burger M., Hackl B., Ring W., Incorporating topological derivatives
into level set methods, J. Comp. Phys., 194(1), 344-362 (2004).
170 F. Jouve
[32] Céa J., Conception optimale ou identification de formes, calcul rapide

de la dérivée directionnelle de la fonction coût, Math. Model. Num.
Anal., 20, 3, 371-402 (1986).
[33] Céa J., Garreau S., Guillaume P., Masmoudi M., The shape and topo-
logical optimizations connection, IV WCCM, Part II (Buenos Aires,
1998), Comput. Methods Appl. Mech. Engrg. 188, 713-726 (2000).
[34] Chambolle A., Larsen C.J., C ∞ -regularity of the free boundary for
a two-dimensional optimal compliance problem, Calc. of Variations
and Partial Diff. Eq., 18(1), (2003).
[35] Chenais D., On the existence of a solution in a domain identification
problem, J. Math. Anal. Appl., 52, 189-289 (1975).
[36] Cherkaev A., Variational Methods for Structural Optimization,
Springer Verlag, New York (2000).
[37] Dal Maso G., Kohn R., The local character of G-closure, unpublished
work.
[38] De Gournay F., Allaire G., Jouve F., Shape and topology optimization
of the robust compliance via the level set method, ESAIM COCV,
4/1, 43-70 (2008).
[39] Diaz A., Sigmund O., Checkerboard patterns in layout optimization.
Structural Optimization, 10, 40-45 (1995).
[40] Duysinx P., Bendsoe M., Topology Optimization of Continuum Struc-
tures with Local Stress Constraints, Int. J. Num. Meth. Engng., 43,
1453-1478 (1998).
[41] Eschenauer H., Kobelev V., Schumacher A., Bubble method for topol-
ogy and shape optimization of structures, Structural Optimization, 8,
42-51 (1994).
[42] Francfort G., Marigo J.-J., Stable damage evolution in a brittle con-
tinuous medium, Eur. J. Mech., A/Solids, 12 (2), 149-189 (1993).
[43] Francfort G., Marigo J.-J., Revisiting brittle fracture as an energy
minimization problem, J. Mech. Phys. Solids, 46 (8), 1319-1342
(1998).
[44] Francfort G., Murat F., Homogenization and optimal bounds in linear
elasticity, Arch. Rat. Mech. Anal., 94, 307-334 (1986).
[45] Francfort G., Murat F., Tartar L., Fourth order moments of a non-
negative measure on S 2 and application, Arch. Rat. Mech. Anal.,
131(4), 305-333 (1995).
[46] Garreau S., Guillaume P., Masmoudi M., The topological asymptotic
for PDE systems: the elasticity case, SIAM J. Control Optim., 39,
no. 6, 1756-1778 (2001).
[47] Gibianski L., Cherkaev A., Design of composite plates of extremal
rigidity, Ioffe Physicotechnical Institute preprint, (1984). (English
translation in Topics in the mathematical modeling of composite mate-

rials, A. Cherkaev and R.V. Kohn eds., series “Progress in Nonlinear
Differential Equations and their Applications”, Birkhaüser, Boston
(1997).)
[48] Hamda H., Jouve F., Lutton E., Schoenauer M., Sebag M., Compact
unstructured representations for Evolutionary Design, Applied Intel-
ligence, 16, 139-155 (2002).
[49] Hashin Z., The elastic moduli of heterogeneous materials, J. Appl.
Mech., 29, 143-150 (1963).
[50] Hashin Z., Shtrikman S., A variational approach to the theory of the
elastic behavior of multiphase materials, J. Mech. Phys. Solids, 11,
127-140 (1963).
[51] Kohn R., Lipton R., Optimal bounds for the effective energy of a mix-
ture of isotropic, incompressible, elastic materials, Arch. Rat. Mech.
Anal., 102(4), 331-350 (1988).
[52] Kohn R., Strang G., Optimal Design and Relaxation of Variational
Problems I-II-III, Comm. Pure Appl. Math., 39, 113-137, 139-182,
353-377 (1986).
[53] Larsen C., Richardson C., Sarkis M., A level set method for the
Mumford-Shah functional and fracture, SIAM J. Imag. Sci., to ap-
pear.
[54] Lurie K., Cherkaev A., Effective characteristics of composite materials
and the optimal design of structural elements, Uspekhi Mekhaniki 9,
3-81 (1986). (English translation in Topics in the mathematical mod-
eling of composite materials, A. Cherkaev and R.V. Kohn eds., series
“Progress in Nonlinear Differential Equations and their Applications”,
Birkhaüser, Boston (1997).)
[55] Michell A., The limits of economy of material in frame-structures,
Phil. Mag., 8, 589-597 (1904).
[56] Milton G., The theory of composites, Cambridge University Press
(2001).
[57] Murat F., Contre-exemples pour divers problèmes où le contrôle inter-
vient dans les coefficients, Ann. Mat. Pura Appl., 112, 49-68 (1977).
[58] Murat F., Simon S., Études de problèmes d’optimal design, Lecture
Notes in Computer Science 41, 54-62, Springer Verlag, Berlin (1976).
[59] Murat F., Tartar L., H-convergence, in Topics in the mathematical
modeling of composite materials, A. Cherkaev and R.V. Kohn eds.,
series: “Progress in Nonlinear Differential Equations and their Appli-
cations”, Birkhaüser, Boston (1997). (French version: mimeographed
notes, séminaire d’Analyse Fonctionnelle et Numérique de l’Université
d’Alger (1978)).
172 F. Jouve
[60] Murat F., Tartar L., Calcul des Variations et Homogénéisation, Les
Méthodes de l’Homogénéisation Théorie et Applications en Physique,
Coll. Dir. Etudes et Recherches EDF, Eyrolles, 319-369 (1985). (En-
glish translation in Topics in the mathematical modeling of compos-
ite materials, A. Cherkaev and R.V. Kohn eds., series “Progress in
Nonlinear Differential Equations and their Applications”, Birkhaüser,
Boston, (1997)).
[61] Neves M., Rodrigues H., Guedes J., Generalized topology design of
structures with a buckling load criterion, Struct. Optim., 10, 71-78
(1995).
[62] Olhoff N., Seyranian A.P., Lund E., Multiple eigenvalues in structural
optimization problems, Struct. Optim., 8, 207-227 (1994).
[63] Osher S., Fedkiw R., Level set methods and dynamic implicit sur-
faces, Applied Mathematical Sciences, 153, Springer-Verlag, New
York (2003).
[64] Osher S., Santosa F., Level set methods for optimization problems
involving geometry and constraints: frequencies of a two-density in-
homogeneous drum, J. Comp. Phys., 171, 272-288 (2001).
[65] Osher S., Sethian J.A., Front propagating with curvature dependent
speed: algorithms based on Hamilton-Jacobi formulations, J. Comp.
Phys., 78, 12-49 (1988).
[66] Pironneau O., Optimal shape design for elliptic systems, Springer-
Verlag, New York (1984).
[67] Rozvany G., Bendsoe M., Kirsch U., Layout optimization of struc-
tures, Appl. Mech. Reviews, 48, 41-118 (1995).
[68] Sanchez-Palencia E., Non homogeneous media and vibration theory,
Lecture Notes in physics 127, Springer Verlag (1980).
[69] Kane C., Schoenauer M., Topological optimum design using genetic
algorithms, Control and Cybernetics, 25, 1059-1088 (1996).
[70] Sethian J.A., Level set Methods and fast marching methods: evolv-
ing interfaces in computational geometry, fluid mechanics, computer
vision and materials science, Cambridge University Press (1999).
[71] Sethian J., Wiegmann A., Structural boundary design via level set and
immersed interface methods, J. Comp. Phys., 163, 489-528 (2000).
[72] Sigmund O., On the design of compliant mechanisms using topology
optimization, Mech. Struct. Mach., 25, 493-524 (1997).
[73] Sigmund O., Petersson J., Numerical instabilities in topology opti-
mization: A survey on procedures dealing with checkerboards, mesh-
dependencies and local minima, Struct. Optim., 16, 68-75 (1998).
[74] Simon J., Differentiation with respect to the domain in boundary
value problems, Num. Funct. Anal. Optimz., 2, 649-687 (1980).
[75] Sokolowski J., Żochowski A., On the topological derivative in shape

optimization, SIAM J. Control Optim., 37, 1251-1272 (1999).
[76] Sokolowski J., Żochowski A., Topological derivatives of shape func-
tionals for elasticity systems, Mech. Structures Mach., 29, no. 3, 331-
349 (2001).
[77] Sokolowski J., Zolesio J.P., Introduction to shape optimization: shape
sensitivity analysis, Springer Series in Computational Mathematics,
Vol. 10, Springer, Berlin (1992).
[78] Suzuki K., Kikuchi N., A homogenization method for shape and topol-
ogy optimization, Comp. Meth. Appl. Mech. Eng., 93, 291-318 (1991).
[79] Tartar L., Estimation de coefficients homogénéisés, Computing meth-
ods in applied sciences and engineering, R. Glowinski, J.L. Lions eds.,
Lecture Notes in Math., 704, Springer Verlag, 364-373 (1978).
[80] Tartar L., Estimations fines des coefficients Homogénéisés, Ennio de
Giorgi colloquium, P. Krée ed., Pitman Research Notes in Math., 125,
168-187 (1985).
[81] Tartar L., Remarks on optimal design problems, in Calculus of varia-
tions, homogenization and continuum mechanics, G. Bouchitté et al.
eds., Series on Adv. in Math. for Appl. Sci, 18, 279-296, World Sci-
entific, Singapore (1994).
[82] Van Goethem N., Novotny A., Crack nucleation sensitivity analysis
Math. Meth. Appl. Sc., 33, no. 16, 1978 1994 (2010).
[83] Vigdergauz S., Effective elastic parameters of a plate with a regular
system of equal-strength holes, Mech. Solids, 21, 162-166 (1986).
[84] Wang M.Y., Wang X., Guo D., A level set method for structural
topology optimization, Comput. Methods Appl. Mech. Engrg., 192,
227-246 (2003).
[85] Wang M.Y., Wang X., Guo D., Structural shape and topology op-
timization in a level set based framework of region representation,
Struct. and Multidisc. Opt. , 27(1-2), 1-19 (2004).
[86] Zhou M., Rozvany G., The COC algorithm, Part II: Topological,
geometrical and generalized shape optimization, Comp. Meth. App.
Mech. Engrg., 89, 309-336 (1991).
Compliance Minimization
of Two-Material Elastic Structures
Grzegorz Dzierżanowski and Tomasz Lewiński

Warsaw University of Technology, Faculty of Civil Engineering,
al. Armii Ludowej 16, 00-637 Warsaw, Poland
e-mail: {gd, tle}@il.pw.edu.pl
Abstract Minimum compliance problem set for a structure made

of two elastic, isotropic materials is tackled in this paper. The relax-
ation by homogenization technique is used for obtaining its mathe-
matically well-posed formulation. The problem is first discussed in
general two-material context. Derivation of main results is recalled
and supplemented with some explanations and remarks. Next, an
important topic of one-material layout optimization (or shape op-
timization) is addressed. It is hampered by the non-smoothness of
formula for relaxed stress energy hence its approximation is pro-
posed which in turn makes the FEM easier to apply in solving the
equilibrium problem. Shape optimization is then linked to a well-
known Michell problem of the lightest, fully stressed structures.
Possible extension of the relaxation by homogenization method to
other structures like thin or moderately thick plates in bending as
well as thin plates or shells submerged to simultaneous in-plane and
bending load are also commented.
1 Introduction
The paper concerns classical topic of topology optimization: given two elas-
tic isotropic materials whose amounts are fixed, fill up an a priori assumed
design domain, in this way forming a least compliant structure which trans-
mits a given load to a given part of the boundary of this domain.
This subject is discussed as the plane elasticity problem, or the problem
of thin elastic plate loaded in plane, which falls into broader category of
vectorial problems in topology optimization. Its achievements have been
preceded by solutions to simpler, scalar elliptic case specified as e.g. sta-
tionary heat conduction, torsion of bars, or pre-stressed membranes. De-
velopment of this branch of optimization has been extensively outlined in

DOI 10.1007/978-3-7091-1643-2_8, © CISM, Udine 2014
176 G. Dzierżanowski and T. Lewiński
(Tartar, 1980), see also (Allaire, 2002, Sec. 3) and (Cherkaev, 2000, Part II)
for main results.
Main aim of the lecture is to derive the compliance minimization problem
in its relaxed version. Exposition of the topic given in the sequel is alter-
native to the one in (Allaire, 2002, Sec. 4.2.1) where the Hashin-Shtrikman
bound, see (2.142) ibidem, is used for estimating the stress energy density.
It is shown that the bound may be found in a different way, namely by ap-
plying the translation method, see relevant articles in (Cherkaev and Kohn,
1997) or (Cherkaev, 2000, Sec. 8, 15) for details. This lecture can be treated
as an addendum to the results mentioned above.
Key issues of the relaxed minimum compliance problem in two-dimensio-
nal setting are presented in Section 2.3. At this point, it is worth pointing
out that the passage from (32) to (57) is relatively short. Its counterpart
for Kirchhoff plates has been explained in (Lewiński and Telega, 2000, Sec.
23.6). The extension of both formulations to the shallow shell case is shown
in (Dzierżanowski, 2012a). Main steps of the derivation in plane elasticity
setting are recalled in the sequel for making the lecture easier to follow. They
also serve as a basis for approximate treatment of the shape optimization
problem in which case the relaxed potential (57) appears to be non-smooth.
The relaxed formulation of shape optimization admits a direct neglect-
ing of stiffness of the weaker material. It allows for certain simplifications
leading to the analytical solution of (34) but it is a source of essential nu-
merical difficulties at the same time. Various approaches are proposed in
the context of shape optimization to overcome the above-mentioned obsta-
cles. They are thoroughly discussed in (Allaire et al., 1997) and (Bendsøe
and Sigmund, 2003), see also (Stolpe and Svanberg, 2001), (Bruggi, 2008)
or (Talischi et al., 2009). In the present article, the novel approach, called
GRAMP, is put forward. It is based directly on the exact result (57), see
(Dzierżanowski, 2012b). The main advantage of GRAMP is its uniform ap-
plicability to the problems of isotropic materials with various possible ratios
between shear and bulk moduli.
Compliance Minimization of Two-Material Elastic Structures 177
2 Two-Material Layout Optimization in Plane

Elasticity
2.1 Notation and Basic Equations
Consider an Euclidean space R2 with its Cartesian basis (e1 , e2 ) and
choose
1 1
E1 = √ (e1 ⊗ e1 + e2 ⊗ e2 ), E2 = √ (e1 ⊗ e1 − e2 ⊗ e2 ),
2 2
(1)
1
E3 = √ (e1 ⊗ e2 + e2 ⊗ e1 )
2
for the basis in the space of symmetric second order tensors, denoted in the
sequel by E2s . With this notation, any tensor ξ ∈ E2s can be thought of as a
vector (ξ1 ξ2 ξ3 )T in R3 whose components are given by
1 1 √
ξ1 = √ (ξ11 + ξ22 ), ξ2 = √ (ξ11 − ξ22 ), ξ3 = 2 ξ12 . (2)
2 2
Let E4s stand for the space of symmetric fourth order tensors acting on
elements from E2s . Each C ∈ E4s is thus represented in (1) by a matrix (Cij ),
i, j = 1, 2, 3, such that Cij = Cji . In the sequel, elements of E4s are referred
to as constitutive tensors, or Hooke’s tensors, of 2D elasticity. To shorten
the notation, it is convenient to define contraction operators

2
3
a·b= aα b α , a, b ∈ R2 , ξ:η= ξi ηi , ξ, η ∈ E2s ,
α=1 i=1
(3)

3
(Cξ)i = Cij ξj , C ∈ E4s , ξ ∈ E2s ,
j=1
and to introduce identity tensors I2 ∈ E2s and I4 ∈ E4s given by
I2 = E1 , I4 = E1 ⊗ E1 + E2 ⊗ E2 + E3 ⊗ E3 . (4)
Set Ω ∈ R2 for a middle plane of a plate whose thickness h is constant and

significantly smaller than the dimensions of Ω. Assume that Ω has a bound-
ary ∂Ω parametrized by a natural parameter s and write p = (p1 , p2 )T ,
pα = pα (s), α = 1, 2, for the density of loading prescribed on ∂Ωσ ⊂ ∂Ω.
Let the stress resultants form a tensor field σ = σ(x), x = (x1 , x2 ) ∈ Ω,
σ(x) ∈ E2s , where

σ11 σ12
σ= , σαβ = σβα , α, β = 1, 2. (5)
σ21 σ22
Displacement field u = (u1 , u2 ), uα = uα (x), determines the in-plane

deformations of a plate. They are given by a strain tensor field ε = ε(u(x))
such that ε ∈ E2s and

ε11 ε12 1 ∂uα ∂uβ
ε= , εαβ (u) = + . (6)
ε21 ε22 2 ∂xβ ∂xα
The plate is supported along ∂Ωu = ∂Ω\∂Ωσ . From now on, V denotes
the space of kinematically admissible functions, i.e. satisfying the boundary
conditions prescribed on ∂Ωu .
Requirement imposed on σ and p is that the equlibrium equation

σ : ε(v) dx = p · v ds (7)
Ω ∂Ωσ
is satisfied for any v ∈ V . Formula (7) determines the set of statically

admissible stress resultant fields

Σ(Ω) = σ ∈ L2 (Ω, E2s ) div σ ∈ L2 (Ω, R2 ),

div σ = 0 in Ω, σn = p on ∂Ωσ (8)
where n = n(s) stands for a vector normal to ∂Ωσ .
Set A, B ∈ E4s for constitutive tensors related to cylindrical domains
ΩA × {−h/2, h/2} and ΩB × {−h/2, h/2}. It is assumed that the materials
are homogeneously distributed along the thickness of a plate. By this, tensor
field of constitutive properties is defined in Ω by

Aχ (x) = χ(x) A + 1 − χ(x) B (9)

with ⎧
⎨1 if x ∈ ΩA ,
χ(x) = (10)
⎩
0 if x ∈ ΩB .
In further discussion it is required that

1
χ(x) dx = mA (11)
|Ω|
Ω
where mA ∈ (0, 1) hence the amounts of materials A and B in the volume

of a plate are fixed by mA h and (1 − mA )h respectively.
Tensor field (9) links the deformation measures ε and stress resultants σ
by linear formula
σχ = Aχ ε(uχ ). (12)
Here uχ ∈ V denotes the solution of the equilibrium problem (7) with

σ = σχ expressed by uχ through (12). Compliance of the plate is defined
as a functional depending on χ by

Υ(uχ ) = p · uχ ds. (13)
∂Ωσ
Two-dimensional elasticity theory includes two cases: plane stress and

plane strain. Their mathematical description is identical up to the interpre-
tation of elastic moduli in (12). Hence, without loss of generality one may
refer to the former formulation in discussing the optimal design problem of
elastic plates made of two materials and subjected to the in-plane external
loading.
Tensors A and B are assumed isotropic in the sequel. Their constitutive
properties are represented in basis (1) by
A = 2kA I2 ⊗ I2 + 2μA (I4 − I2 ⊗ I2 ),

(14)
B = 2kB I2 ⊗ I2 + 2μB (I4 − I2 ⊗ I2 )
where k and μ respectively denote the Kelvin and Kirchhoff moduli which
are related to Young’s modulus E and Poisson ratio ν through
Eh Eh
k= , μ= . (15)
2(1 − ν) 2(1 + ν)
In the course of this study, materials are considered well-ordered, i.e.
kA > kB , μA > μB . (16)
This restriction is imposed in order to enable the formulation of the shape

optimization problem where the weaker material B degenerates to void and
kB = μB = 0. It has to be pointed out that the discussion in case of two
badly-ordered non-degenerated materials poses no additional problem, see
(Allaire and Kohn, 1994) for further details.
2.2 Formulation of the Optimization Problem

The overall stiffness of the plate can be defined as the inverse of its
compliance (13) hence maximization of the stiffness is equivalent to the
minimization of Υ(χ). Finding the minimal value of structural compliance
over all possible distributions of materials χ ∈ (Ω, {0, 1}) with the restriction
imposed by the so-called isoperimetric condition (11) is the main concern
of the layout optimization problem discussed in this paper. According to

Castigliano’s theorem, see (Nečas and Hlavaček, 1981), (Duvaut and Lions,
1976), the compliance can be expressed by

Υ(χ) = min σ : A−1 σ dx σ ∈ Σ(Ω) (17)
χ
Ω
where, due to assumed isotropy,

Aχ (x)−1 = χ(x) A−1 + 1 − χ(x) B −1 , (18)
and
1 1
A−1 = KA I2 ⊗ I2 + LA (I4 − I2 ⊗ I2 ),
2 2
(19)
1 1
B −1 = KB I2 ⊗ I2 + LB (I4 − I2 ⊗ I2 )
2 2
with K = k −1 , L = μ−1 . From (16) it follows that KA < KB , LA < LB . A
search for the stiffest plate loaded in-plane is realized by

∞
Jmin = min J(χ) χ ∈ L Ω; {0, 1} (20)
where
J(χ) = Υ(χ) + χ(x) dx, (21)
Ω
and ∈ R denotes a Lagrange multiplier for the constraint set in (11).

The problem set by (20) and (21) is equivalent to minimizing Υ(χ) over
χ satisfying Ω χ(x) dx ≤ mA |Ω|, see (Lipton, 1994). However, in most
cases, this slack isoperimetric condition is satisfied as equality hence (11) is
assumed in further considerations.
The optimization problem above is ill-posed due to the fact that se-
quences {χn } ∈ L∞ (Ω; {0, 1}) are in general non-convergent in this space.
Thus, (20) needs to be given a form in which the existence of solutions is a
priori guaranteed.
Reformulation starts from the observation
χn L∞ (Ω) ≤ 1. (22)
It follows that any sequence of characteristic functions converges weakly-∗ to

some θ ∈ L∞ (Ω, [0, 1]), see (Allaire, 2002, Sec. 1.2.1). From the mechanical
point of view, it means that the classical layouts of materials A and B,
determined by (9), are now supplemented with their arbitrary mixtures.

The amounts of A and B at given x ∈ Ω are measured by θ(x) and 1 − θ(x)
respectively, where θ(x) admits any value in the interval [0, 1]. Constitutive
law (9) is replaced by function A∗ = A∗ (x). Proper characterization of A∗
combines micro- and macroscale description as shown in the sequel.
Reformulation of (20) is called “relaxation by homogenization”. Under-
lying mathematical theory is being developed simultaneously with its me-
chanical applications from the 1970s. The key definition is briefly recalled
below, but detailed exposition of homogenization lies beyond the scope of
this paper. The reader is thus referred to (Tartar, 1980) or (Allaire, 2002,
Sec. 1.4.2) for further explanations.
The sequence of tensorial constitutive functions Aχ generated by the
∗
sequence of characteristic functions such that χ θ is convergent in the
sense of homogenization (or H-convergent) to the function A∗ if for any
q = [q1 , q2 ]T , qα = qα (x), α = 1, 2, x ∈ Ω, representing the load acting in
Ω, the sequence uχ solving

div (Aχ ∇uχ ) = −q in Ω,
(23)
uχ ∈ H01 (Ω),
satisfies
uχ u in H01 (Ω),
(24)
Aχ ∇uχ A∗ ∇u in L2 (Ω, E2s ),
where u solves the local homogenized problem

div (A∗ ∇u) = −q in Ω,
(25)
u ∈ H01 (Ω).
It also can be proved that if Aχ is H-convergent to A∗ then the solutions

to (23), (25) satisfy

∇uχ : (Aχ ∇uχ ) dx → ∇u : (A∗ ∇u) dx. (26)
Ω Ω
It follows that stress and strain energy sequences are convergent in the sense
of (26).
Important property of homogenization in RN , N > 1 stems from the fol-
lowing fact: different {χm }, {χn } ∈ L∞ (Ω; {0, 1}) may converge to the same
limit θ ∈ L∞ (Ω, [0, 1]), but sequences {Aχm }, {Aχn } can tend to different
limits A∗1 , A∗2 ∈ L∞ (Ω; E4s ). Hence, taking all sequences {χn } weakly-∗ con-
verging to a given limit θ results in forming a set of all possible homogenized
Hooke’s laws A∗ , i.e. all compositions of A and B taken in proportions θ,

1 − θ, θ ∈ [0, 1]. This set, denoted in the sequel by Gθ , is stable under
homogenization, i.e. any H-limit of its elements also belongs to Gθ .
Full characterization of Gθ is still unknown. It can be proved that con-
sidering all H-limits of sequences of periodic characteristic functions is suffi-
cient to obtain Gθ . However, explicit formula determining the representation
of a composite Hooke’s tensor correpsonding to arbitrary sequence {χn } is
not yet concluded, except for some special cases. One of them is investigated
in subsequent sections.
From the definition of H-convergence it follows that functions A∗ can
be determined locally by computing homogenized constitutive properties
A∗ (x) independently at each x ∈ Ω. This in turn means that Gθ can be fully
characterized as a union of sets Gθ(x) where the latter denotes a closure of
a two-element set {A, B}. Each Gθ(x) contains all homogenized Hooke’s
tensors related to all possible mixtures of A and B taken in proportions
θ(x) and 1 − θ(x) respectively. This property of Gθ is referred to as the local
character of G-closure.
By virtue of the remarks above, any function θ ∈ L∞ (Ω, [0, 1]) deter-
mines a set

Gθ = A∗ ∈ L∞ (Ω, E4s ) A∗ (x) ∈ Gθ(x) . (27)
Relaxed counterpart of (20) involves minimization over a pair of unknowns

(θ, A∗ ) and it reads

∗ ∗ ∗
∞ ∗
Jmin = min J (θ, A ) θ ∈ L (Ω; [0, 1]), A ∈ Gθ (28)
where
∗ ∗ ∗
J (θ, A ) = Υ (θ) + θ(x) dx, (29)
Ω
and

Υ∗ (θ) = min σ : A∗ −1 σ dx σ ∈ Σ(Ω) . (30)
Ω
Function θ is restricted to obey the isoperimetric condition (11).

The order of minimizations in (28) is arbitrary. Moreover, the local
character of G-closure allows for the following reformulation of (28)

∗
Jmin = min 2W ∗ (σ, θ) + θ dx σ ∈ Σ(Ω), θ ∈ L∞ (Ω; [0, 1]) (31)
Ω
where

2W ∗ (σ, θ) = min σ : A∗ −1 σ A∗ ∈ Gθ (32)
stands for the (doubled) stress energy density of a homogenized plate, and
A∗ = A∗ (x), Gθ = Gθ(x) .
Further modification of the problem can be achieved by swapping the
minimization over θ ∈ L∞ (Ω; {0, 1} with integration in (31). This is pos-
sible due to the Rockafellar Theorem, see (Rockafellar, 1976). For detailed
justification of the application of Rockafellar’s Theorem in the context of a
problem tackled in this paper, the reader is referred to (Allaire, 2002, Sec.
4.1.4), see also (Dzierżanowski, 2012b) for its brief recall.
In this way, the relaxed material layout optimization problem for mini-
mum compliance takes the form

∗
Jmin = min F ∗ (σ) dx σ ∈ Σ(Ω) (33)
Ω
where

F ∗ (σ) = min 2W ∗ (σ, θ) + θ θ ∈ [0, 1] . (34)
2.3 Lower Estimation of the Homogenized Stress Energy

Calculating the stress energy density W ∗ in (32) is crucial for the solution
of the optimization problem (31). As mentioned in the preceding Section,
the search for optimal A∗ ∈ Gθ can be narrowed to the H-limits of Aχn
determined by sequences of periodic characteristic functions {χn }. To this
end, introduce a basic periodicity cell Y = (0, a) × (0, b) with a and b being
dimensionless and such that |Y | = 1. The coordinate system y = (y1 , y2 ) ∈
Y is linked with x ∈ Ω by δy = x, where δ denotes a very small positive
number which can be understood as a parameter determining the fineness
of the periodicity mesh in Ω. In this sense, the dimensions of Y and Ω are
separated by several length scales, see (Sanchez-Palencia, 1980). Symbol ·
used in the sequel stands for the averaging over Y , i.e. g = Y g(y)dy with
|Y | = 1.
Assume that x ∈ Ω is arbitrary but fixed and the values σ(x) and θ(x)
are given. Next, associate cell Y with x and write

Σ(Y ) = τ ∈ L2 (Y, E2s ) div τ = 0 in Y,

τ n takes opposite values on opposite sides of Y (35)
for the set of periodic stress resultants that are statically admissible in
Y . Here, n represents a unit vector outward normal to ∂Y . Furthermore,
expand the periodicity mesh to the whole space R2 . Given materials A and
B are distributed in Y according to the function χ ∈ L∞ (Y, {0, 1}) such
that χ = θ. The constitutive tensor field Aχ in Y is expressed similarly
to (9). Mean stress energy in Y is measured by

2 W (τ, χ) = τ : A−1 χ τ . (36)
The homogenized energy density (32) is now expressed in the form

Wχhom (σ) = min W (τ, χ) τ ∈ Σ(Y ), τ = σ (37)

W ∗ (σ, θ) = min Wχhom (σ) χ ∈ L∞ (Y, {0, 1}), χ = θ (38)
where the minimum is, in general, attained on a hierarchical microstructure

which gives rise to the so-called “composite materials of a rank” whose
examples are shown in Figures 1, 3.
From (38) it follows that W ∗ ≤ W and the task now is to determine
a lower estimation QW ≤ W ∗ in order to bound the values of W ∗ . Conse-
quently, explicit calculation of W ∗ consists of two steps: (i) determination
of function QW , (ii) calculation of minimizers χopt , χopt = θ, and τopt ,
τopt = σ, for which QW = W . The former step is briefly described
below and the latter is dealt with in the following section.
Lower bound QW on the energy density W ∗ can be derived by making
use of a mathematical technique named “the translation method”, see (Mil-
ton, 2002), (Cherkaev, 2000). In this method, the constitutive tensor A−1 χ
is shifted (translated) by a certain isotropic tensor α T , where

1
T = I4 − I2 ⊗ I2 (39)
2
and α ∈ R denotes the translation parameter. Tensor T realizes
det τ = −τ : (T τ ). (40)
Key ingredient of the translation method is the quasiaffinity property of

function det τ , i.e.
det τ = det σ, τ ∈ Σ(Y ), σ = τ . (41)
Proof of the latter relies on the periodicity of τ . Despite this fact, there
is no loss in generality of further discussion. Full justification of (41) can
be found in the above-mentioned references, see also (Lewiński and Telega,
2000), hence it is omitted here.
The next step of the method consists in rewritting (36) in the form

2 W (τ, χ) = 2W (τ, χ) + α det τ − α det τ (42)
or explicitly, by making use of (40),

2 W (τ, χ) = τ : A−1 χ − αT τ + α τ : Tτ . (43)
Optimal adjustment of the translation parameter α is the next goal in search

for QW .
Neglecting the differential constraint divτ = 0 imposed on the elements
of Σ(Y ) such that τ = σ allows for estimating the first term on the r.h.s.
of (43) by
−1
−1
−1 −1
τ : Aχ − α T τ ≥ σ : Aχ − α T σ (44)

which is valid for any τ ∈ L2 (Y, E2s ) and for positive definite tensors A−1
χ −

α T ∈ E4s . The latter requirement restricts possible values of translation

parameter to
−KA ≤ α ≤ LA . (45)
By applying (41) and (44) respectively to the second and first terms on
the r.h.s. of (43) one obtains the family of lower bounds on the homogenized
stress energy energy density parameterized by the scalar α restricted by (45).
The best lower estimation of QW ≤ W ∗ can be thus expressed in the form

2 QW (σ, θ) = max σ : C(α, θ) σ α as in (45) (46)
where

−1
−1 −1
C(α, θ) = θ A−1 − α T + (1 − θ) B −1 − α T + α T. (47)
The “max” operation in (46) can be performed analytically. Necessary

calculations pose no particular problem hence only their general outline is
presented here.
For simplicity of notation in the sequel, write
Δf = |fB − fA |, f θ = θfA + (1 − θ)fB ,

(48)
[f ]θ = (1 − θ)fA + θfB ,
and
σI − σII

ζ= (49)
σI + σII
where σI , σII denote the principal values of σ; |σI | ≥ |σII | by convention and
σI + σII = 0. Next, rewrite (47) in the form
⎛ ⎞
C11 0 0
1⎜⎜
⎟
⎟
C= ⎜ 0 C22 0 ⎟ (50)
2⎝ ⎠
0 0 C33
where
(KA + α)(KB + α) (LA − α)(LB − α)
C11 = − α, C22 = C33 = + α.
[K]θ + α [L]θ − α
(51)
By taking the eigenbasis of vector σ as the physical basis in (1) it is possible
to write ⎛ ⎞
σI + σII
1 ⎜⎜
⎟
⎟
σ = √ ⎜σI − σII ⎟ , (52)
2⎝ ⎠
0
and consequently

σ : C(α, θ) σ = (σI )2 C11 (α, θ) + C22 (α, θ) ζ 2 . (53)
Investigating the necessary condition of optimality with respect to α

and combining it with the requirement (45) leads to the optimal values of
α given by ⎧
⎪
⎪ LA 0 ≤ ζ ≤ ζ2 ,
⎪
⎪
⎪
⎨
α = ΔK[L]θ − ζ ΔL[K]θ ζ2 ≤ ζ ≤ ζ1 , (54)
⎪
⎪ ΔK + ζ ΔL
⎪
⎪
⎪
⎩
−KA ζ1 ≤ ζ
where
KA + [L]θ θΔK
ζ1 = , ζ2 = , (55)
θΔL LA + [K]θ
define three regions (regimes) of optimality
Z3 : ζ ∈ (0, ζ2 ), Z2 : ζ ∈ (ζ2 , ζ1 ), Z1 : ζ ∈ (ζ1 , +∞). (56)
The lower bound on W ∗ takes the explicit form

2 2
8 QW (σ, θ) = K∗ (θ) (σI + σII ) + L∗ (θ) (σI − σII ) (57)
where
⎧
⎪ KA KB + LA Kθ
⎪
⎪ 0 ≤ ζ ≤ ζ2 ,
⎪
⎪ LA + [K]θ
⎨
K∗ (θ) = KA KB + [L]θ Kθ − (1 − θ)θΔKΔL ζ (58)
⎪
⎪ ζ2 ≤ ζ ≤ ζ1 ,
⎪
⎪ [K + L]θ
⎪
⎩
KA ζ1 ≤ ζ,
and
⎧
⎪
⎪ LA 0 ≤ ζ ≤ ζ2 ,
⎪
⎪
⎪
⎪
⎨ LA LB + [K]θ Lθ − (1 − θ)θΔKΔL ζ −1
L∗ (θ) = ζ2 ≤ ζ ≤ ζ1 , (59)
⎪
⎪
[K + L]θ
⎪
⎪
⎪ LA LB + KA Lθ
⎪
⎩ ζ1 ≤ ζ,
KA + [L]θ
denote the reciprocals of optimal effective moduli. One may observe that
the case of σI + σII = 0 falls into region ζ1 ≤ ζ.
Values of K∗ and L∗ depend continuously on ζ hence QW is also contin-
uous in this variable. Moreover, QW is smooth hence differentiable in the
whole interval ζ ∈ (0, +∞). Consequently, formula
∂
ε= QW (60)
∂σ
determines the constitutive equations of optimal material for the whole

range of σ.
Inverting K∗ and L∗ to the form k∗ , μ∗ is necessary for implementation
of the results in typical numerical codes. This operation is possible in each
region of optimality, see (Lewiński, 2004b). It gives
⎧
⎪ kA kB + μA kθ
⎪
⎪
⎪ 0 ≤ ζ! ≤ ζ!2 ,
⎪
⎪ μA + [k]θ
⎨
k∗ (θ) = kA kB + [μ]θ kθ + (1 − θ)θΔkΔμ ζ! (61)
⎪
⎪ ζ!2 ≤ ζ! ≤ ζ!1 ,
⎪
⎪ [k + μ]θ
⎪
⎪
⎩
kA !
ζ!1 ≤ ζ,
and
⎧
⎪
⎪
⎪ μA 0 ≤ ζ! ≤ ζ!2 ,
⎪
⎪
⎪
⎪
⎨ μ μ + [k] μ + (1 − θ)θΔkΔμ ζ!−1
ζ!2 ≤ ζ! ≤ ζ!1 ,
A B θ θ
μ∗ (θ) = (62)
⎪
⎪ [k + μ]θ
⎪
⎪
⎪
⎪ μ μ + k μ
⎪ ζ!1 ≤ ζ!
A B A θ
⎩
kA + [μ]θ
where
εI − εII
!
ζ = (63)
εI + εII
depends on principal values εI , εII , |εI | ≥ |εII |, of deformation tensor ε and
kA + [μ]θ θΔk
ζ!1 = , ζ!2 = . (64)
θΔμ μA + [k]θ
Case of εI + εII = 0 falls into region ζ!1 ≤ ζ.

!
2.4 Optimal Microstructures

Sequential Laminates In the search for optimal τopt and χopt in (38) it is
first conjectured that χopt describes a layout of materials in Y taking a form
of the so-called strong sequential laminate. Such laminates are obtained
by repetitive homogenization of basic materials according to the following
scheme: (i) A and B are mixed in proportions θ1 , 1 − θ1 respectively and
in this way the homogenized material (rank-1 laminate) of moduli A∗1 is
obtained, (ii) A is mixed with A∗1 in a new direction and with proportions
θ2 , 1 − θ2 and this operation results in a rank-2 laminate of moduli A∗2 .
Sequential process of laminating A∗p−1 , p > 1, with the strong basic material
A can be performed infinitely many times. The set of laminates where the
weaker material B (core) is wrapped by the stronger one A (coating) in
overall proportions 1 − θ, θ, is denoted by L+
θ (A, B). An example of a rank-
3 sequential laminate is shown in Figure 1. Note that a word “laminate” is
used to describe a composite obtained by recursive layering of materials on
the middle plane Ω. Such terminology may confuse the reader familiar with
laminates understood as a stack of laminae in a direction perpendicular to
Ω. Hence, it has to be stressed that the latter theory is not discussed in
this paper.
The role of a core and coating can be swapped over in the sequence of
laminations. It gives rise to a set L− + −
θ . Both Lθ and Lθ are the subsets of
the general set of laminates Lθ . The latter is obtained when all materials
t3
t1
n1
A t2 1
q1 n3
*
n2 1 A A3
A 1*
q1
*
1- q A A2 q3
1-
B
1- q 3
1
2
q2
Figure 1. Rank-3 sequential laminate.
constructed at the previous step of lamination are allowed in the next step,
i.e.
– materials A and B are available for the first lamination,
– materials A, B and L(A, B) are available for the second one,
– materials A, B, L(A, B), L(A, L(A, B)), L(B, L(A, B))
and L(L(A, B), L(A, B)) are available for the third one,
– etc.
It is shown in (Cherkaev, 2000, Sec. 10) that in this way one may characterize
the whole set Gθ .
t*,ktt t Att
t Ant
t*,knn t Ann 1
*,k
t nt tk+1
nk+1
Gk+1
*
Ak A
1- qk+1 qk+1
Figure 2. Cell Y filled with a strong sequential laminate A∗k+1 with an

interface line Γk+1 separating materials A∗k and A. Stress components τnn
and τnt are required to be continuous on Γk+1 in order to assure equilibrium
of the indicated segment of the cell .
Stress components related to both sides of interfaces separating the mate-

rials within the cell Y should fulfill certain equilibrium requirements which
assume the form of continuity conditions. In the sequel these conditions

are referred to as compatibility conditions of the stress state. This remark
explains a decomposition of the stress field in the form
τ = Pc qc + Pd qd (65)
where qc , qd respectively stand for the continuous components of τ and those

whose values may jump across Γk+1 . Field τ has to obey the compatibility
condition
[τnn n + τtn t]Γ = 0. (66)
It follows that τtt is the only component whose continuity on Γk+1 is not
required. Projectors Pc , Pd and vectors qc , qd take the form

1 √
T
Pc = n ⊗ n √ (n ⊗ t + t ⊗ n) , qc = τnn 2 τnt ,
2 (3×2) (67)
Pd = (t ⊗ t)(3×1) , qd = (τtt ) ,
with the representations of Pc = Pc (ϕ(n)) and Pd = Pd (ϕ(n)) in the basis

(1) given by
⎛ 1 ⎞ ⎛ 1 ⎞
√ 0 √
⎜ 2 ⎟ ⎜ 2 ⎟
⎜ ⎟ ⎜ ⎟
⎜ 1 ⎟ ⎜ 1 ⎟
Pc = ⎜
⎜ √2 cos(2 ϕ(n)) − sin(2 ϕ(n))⎟
⎟,
⎜
Pd = ⎜− √ cos(2 ϕ(n))⎟
⎟
⎜ ⎟ ⎜ 2 ⎟
⎝ 1 ⎠ ⎝ 1 ⎠
√ sin(2 ϕ(n)) cos(2 ϕ(n)) − √ sin(2 ϕ(n))
2 2
(68)
where ϕ(n) stands for an angle of a lamination direction in the assumed
Cartesian basis (e1 , e2 ). Both tensors appear in the lamination formulae of
a rank-p strong sequential laminate,
" $−1
#
p
A∗−1
p = A−1 + (1 − θ) (B −1 − A−1 )−1 + θ mk fAc (ϕ(nk )) (69)
k=1
where −1 T
fAc (ψ) = Pd (ψ) PdT (ψ)A−1 Pd (ψ) Pd (ψ) (70)
is defined through standard matrix operations in the basis (1). Derivation
of the formula for A∗−1
p involves straightforward but lengthy algebraic cal-
culations hence they are omitted here. Different, albeit equivalent form of
(70) is reported in e.g. (Allaire, 1997).
Lamination parameters
θ1 θ2 (1 − θ1 )
m1 = , m2 = , ...
θ θ
(71)
θp (1 − θp−1 )(1 − θp−2 ) . . . (1 − θ1 )
mp = ,
θ
p
such that k=1 mk = 1, measure the relative fraction of material A in
subsequent laminations. The overall amount of A in a rank-p composite is
thus given by
%p
1−θ = (1 − θk ). (72)
k=1
The counterpart of formula (69) in terms of stiffness tensors reads

" $−1
#
p
A∗p = A + (1 − θ) (A − B) −1
−θ mk fA (ϕ(nk )) , (73)
k=1
with
−1 T
fA (ψ) = Pc (ψ) PcT (ψ)APc (ψ) Pc (ψ). (74)
It turns out that rank-2 sequential strong laminates with orthogonal
directions of layerings (so-called T-structures, see Figure 3) are sufficient to
prove the attainability QW = W ∗ , see (38). Effective constitutive tensor
A∗2 associated with this composite possesses a property
2 QW (σ, θ) = σ : (A∗−1
2 σ), A∗2 ∈ Gθ . (75)
At this point of discussion it is worth pointing out that the definition of

tensor C(α, θ) in (47) involves only isotropic tensors A, B and T . By this,
C(α, θ) and QW (σ, θ), see (57), are also isotropic. On the other hand, one
may check that formula (69) represents an orthotropic tensor A∗−1 p hence
the equality in (75) may seem unavailable in general. However, it has to be
understood that the only concern in the search of optimal laminate material
is the equality of the scalars QW = W ∗ . Full match between the represen-
tations of tensors A∗−1
2 and C(α, θ) is not required. Consequently, there is
no contradiction in treating both tensors as optimal: (i) tensor C(α, θ) may
be referred to as optimal isotropic effective tensor whose components locally
depend on the pair (σ, θ) through the relations determining α in (54); (ii)
tensor A∗−1
2 describes optimal orthotropic properties locally dependent on
θ only.
sII
1- q1
sI sI
B
q1
A A
(a) sII (b) 1- q2 q2
Figure 3. T-structure obtained by a sequential lamination of material A

(coating, line filling) around material B (core, white color): (a) construction
of a laminate, (b) accumulated fractions θ1 , θ2 measuring the amount of
material A in each lamination. Directions of laminations are determined by
σI and σII denoting the principal values of tensor σ.
3 Plane Shape Optimization and Optimal Porous

Composites
3.1 Explicit Formula for Stress Energy Estimation
Very important topic in structural design deals with optimal layout of
one material in the design area Ω. It is referred to as “shape optimization
problem” and can be discussed in the general two-material framework set-
ting presented in previous Section of this paper. Roughly speaking, formulae
related to shape optimization are obtained by letting kB → 0, μB → 0 (or,
equivalently KB → +∞, LB → +∞) in their two-material counterparts.
The physical meaning of this purely mathematical operation is understood
as replacing the weaker (more compliant) material with void.
Formuale in (55) are replaced with ζ1 = ζ2 = 1 and (56) becomes now
ζ ∈ Z3 = (0, 1) if σI σII > 0 and ζ ∈ Z1 = (1, +∞) if σI σII < 0. Here
II
ζ = σσII −σ
+σII . Note that region Z2 is excluded from the analysis.
Moreover, (54) now reads
⎧
⎨L det σ > 0,
α= (76)
⎩
−K det σ < 0
where K ≡ KA and L ≡ LA . From (58) and (59) it follows that
K + α(1 − θ) L − α(1 − θ)
K∗ (θ) = , L∗ (θ) = . (77)
θ θ
Let us now set ρ = σσIII with |σI | ≥ |σII |. Hence ρ ∈ [−1, 0) if σI σII < 0
and ρ ∈ (0, 1] if σI σII > 0. We also assume that h = 1. One may check that
(57) is now given by
⎧
⎪ 1 1−ν
⎪
⎨ (1 − ρ)2 + 2 ρ for ρ < 0,
θE E
2 QW (σ, θ) = (σI )2 (78)
⎪
⎪
⎩ 1 (1 + ρ)2 − 2 1 + ν ρ for ρ > 0,
θE E
where E ≡ EA and ν ≡ νA . Equivalently
& '
1 1 + sgn(ρ) ν
2 QW (σ, θ) = (σI )2
1 + |ρ|) − 2
2
|ρ| . (79)
θE E
From (79) it follows that QW is continuous for any σ but it is not smooth
for det σ = 0. Thus, QW is not differentiable in the whole range of σ. This
fact poses a particular obstacle in developing the FEM-based numerical
scheme for shape optimization problems. Possible remedy is proposed in
the subsequent Section.
Conditions for attainability of QW on certain microstructures are dis-
cussed in Sec. 2.3 and 2.4. Hence in further considerations we assume that
QW = W ∗ . Consequently, in case of shape optimization, solution to (34)
can be explicitly found. It reads

|σI | + |σII |
θopt (σ) = min 1, √ (80)
E
hence the integrand in (33) takes the form:
⎧ 2 (√

)
⎪
⎪ E |σI | + |σII | − 1 + sgn(σI σII )ν |σI σII | if θ < 1,
⎨E
F ∗ (σ) = & '
⎪
⎪ 2 1 1 1
⎩ (σI ) + (σII ) − νσI σII + E
2 2
if θ = 1.
E 2 2 2
(81)
Formula (69) becomes
" p $−1
1−θ #
A∗−1
p =A −1
+ c
mk fA (ϕ(nk )) , (82)
θ
k=1
see (70) for fAc , and (73) takes the form

" $−1
#
p
A∗p = A + (1 − θ) A −1
−θ mk fA (ϕ(nk )) , (83)
k=1
with fA defined through (74).
4 Solid-Void Interpolation Schemes

4.1 Scheme Based on the Properties of Optimal Stress Energy
For given σ, relation (79) sets the mapping θ → QW (σ, θ) which is
optimal in the sense of minimizing the stress energy accumulated in an
arbitrary mixture of material and void taken in proportions θ and 1 − θ
respectively. By rewriting (79) in the form
(σI )2
2 Wopt (σ, θ) = Uopt (ρ, θ),
θE (84)

Uopt (ρ, θ) = aopt (θ) 1 + ρ2 + 2bopt (θ) ρ,

σII
where ρ = σI and
aopt (ρ, θ) = 1, bopt (ρ, θ) = −θν + (1 − θ) sgn(ρ), (85)
one may define the optimal solid-void interpolation scheme of material prop-
erties as
θE bopt (ρ, θ)
Eopt (ρ, θ) = , νopt (ρ, θ) = − . (86)
aopt (ρ, θ) aopt (ρ, θ)
Numerical implementation of the above results is hampered by the non-
smoothness of stress energy functional Uopt (ρ, θ) for ρ = 0. Namely, if this
is the case, then the constitutive formula is not unique, hence it cannot be
directly inverted into the stress-displacement form which serves as a nat-
ural basis for a numerical implementation by FEM. As a possible remedy,
one may either make use of the solution to the problem of optimal lay-
out formulated for a two-material structure, see (61) and (62), or seek the
smooth, thus invertible, approximation Uapp of Uopt . In this Section, the
latter option is investigated.
Function Uopt is taken as a reference expression in determining aapp (θ)
and bapp (θ). Both coefficients are required to be independent of the sign of
ρ and such that resulting Uapp is isotropic, i.e. it takes the form

Uapp (ρ, θ) = aapp (θ) 1 + ρ2 + 2bapp (θ) ρ. (87)

It should also satisfy

Uapp (ρ, θ) ≥ Uopt (ρ, θ) (88)
for any pair (ρ, θ) in order. It has to be pointed out that both restrictions
are heuristic, nevertheless they lead to acceptable results as it is shown in
the sequel. The sought approximation with the above required properties
is realized at Uapp sharing the common tangent with Uopt at two distinct
points ρ1 and ρ2 . Setting ρ1 = 1/q1 and ρ2 = −1/q2 followed by

∂Uapp 1 ∂Uopt 1
,θ = ,θ ,
∂ρ q1 ∂ρ q1
(89)
∂Uapp 1 ∂Uopt 1
− ,θ = − ,θ ,
∂ρ q2 ∂ρ q2
leads to
q1 q2 q1 − q2
aapp (θ) = 1 + 2(1 − θ) , bapp (θ) = (1 − θ) − θν. (90)
q1 + q2 q1 + q2
Requirement (88) restricts the range of qα to
qα ∈ [1, +∞). (91)
Moreover, Uapp = Uopt at two distinct points iff q1 = q2 = 1. In this way,
the two-parameter solid-void interpolation scheme is established with the
approximate stress energy function Uapp defined through (87) and (90).
The simplified, one-parameter variant of the above is obtained by setting
q1 = q2 = q in (90). This choice leads to the uniform approximation of the
functional Uopt as it is shown in the sequel. As a result one may claim the
stress energy approximation in the form (87) with coefficients
aapp (θ) = 1 + q (1 − θ), bapp (θ) = −θν. (92)
It turns out that with the approach discussed above, one may justify
and generalize the RAMP (Rational Approximation of Material Proper-
ties) scheme proposed by Stolpe and Svanberg (2001). Hence the nickname
Generalized RAMP (GRAMP) is adopted for the solid-void interpolation
model presented in this Section. For full derivation of the GRAMP model,
the reader is referred to (Dzierżanowski, 2012b).
Similarly to (86), in the GRAMP scheme, Young’s modulus and the
Poisson ratio are calculated as
EGRAMP (θ) = fq (θ) E, νGRAMP (θ) = fq (θ) ν,
θ (93)
fq (θ) =
1 + q(1 − θ)
where fq (θ) denotes the material properties interpolation function. The set
of plots in Figure 4 shows fq for different values of parameter q.
fq(q)
Figure 4. Interpolation function fq plotted for q = 1 (solid line), q = 3

(dashed line) and q = 10 (dotted line).
Assuming the solid-void interpolation scheme in the GRAMP form has

a great impact on the quality of approximation Uapp . For given ρ0 ∈ (0, 1],
it can be measured as
0
−ρ
|Uapp (ρ, θ) − Uopt (ρ, θ)| dρ
Δ(Uapp ) = ρ00 , (94)
0 |Uapp (ρ, θ) − Uopt (ρ, θ)| dρ
thus expressing a simple requirement of approximation uniformity for dif-

ferent signs of ρ. One may calculate that Δ(UGRAMP ) = 1 which is not the
case for neither of approximation schemes briefly recalled in the sequel.
4.2 Other Interpolation Schemes

Among other solid-void material interpolation schemes, the most popu-
lar and widely used is the SIMP (Solid Isotropic Material with Penalization)
model. In this approach, the intermediate material densities are penalized
by introducing the penalization factor p in the definition of material prop-

erties. Function USIMP is determined through substituting
1 ν
aSIMP (θ) = , bSIMP(θ) = − , (95)
θp−1 θp−1
in (87) hence Young’s modulus and the Poisson ratio interpolations are
given by
ESIMP (θ) = θp E, νSIMP (θ) = ν. (96)
The RAMP scheme mentioned in the previous Section admits the func-
tion URAMP described by (87) with
aRAMP (θ) = 1 + q (1 − θ), bRAMP (θ) = −ν [1 + q (1 − θ)] (97)
hence
ERAMP (θ) = EGRAMP (θ), νRAMP (θ) = ν = const. (98)
Functions USIMP and URAMP have to obey (88) as any other approximate
stress energy function Uapp . Hence, the ranges of SIMP or RAMP parame-
ters (p and q respectively) are limited. Even more restrictive is the require-
ment of macrostructural isotropy of the solid-void composite. Bounds on
the effective isotropic properties of a two-material mixture were introduced
by Hashin and Shtrikman (1963) and improved by Cherkaev and Gibiansky
(1993). It is worth pointing out, however, that both results coincide for
the mixture of material and void hence, in the sequel, the requirement of
effective isotropy is denoted by UHS ≤ Uapp . Formula determining UHS can
be expressed by (87) with
aHS (θ) = 3 − 2 θ, bHS (θ) = −θν − (1 − θ), (99)
hence
θE θν + (1 − θ)
EHS (θ) = , νHS (θ) = . (100)
3−2θ 3−2θ
The latter can be independently derived from the Hashin-Shtrikman upper
bounds on Kelvin’s and Kirchhoff’s moduli of a mixture of two isotropic
materials given by
−1
1 θ
kHS = kA − (1 − θ) − ,
kA − kB kA + μA
−1 (101)
1 θ(kA + 2μA )
μHS = μA − (1 − θ) − .
μ A − μB 2μA (kA + μA )
The first step in justifying this fact is to apply kB = 0, μB = 0 to (101).
Next, thus obtained result is rewritten in terms of EA , νA , and EHS , νHS , see
(15) for h = 1, as both material A and the effective composite material are
isotropic. In this way, (100) follows with subscript A dropped for simplicity
of notation.
Table 1 presents lower bounds imposed on the parameters of GRAMP,
RAMP and SIMP approximations. They stem from two requirements which
has to be fulfilled by any approximate stress energy functional Uapp . In-
equality Uapp − Uopt ≥ 0, see (88), restricts the values of parameters to
those for which the approximate energy takes the form of an isotropic func-
tion (87) while Uapp − UHS ≥ 0 imposes the effective isotropy of a composite
material as an additional constraint. Details of the results are reported in
e.g. (Dzierżanowski, 2012b) and (Bendsøe and Sigmund, 1999; Stolpe and
Svanberg, 2001). It is worth pointing out that in the GRAMP scheme pa-
rameters are independent of the basic material’s Poisson ratio value ν and
density θ. On the contrary, in the course of calculations related to RAMP
and SIMP schemes one has to set θ = 1 in order to assure the validity of
expressions in Table 1 for any θ ∈ [0, 1].
The comparison of Uopt , UHS and UGRAMP , URAMP , USIMP with param-
eters of lowest possible values are shown for ν = 1/3 (Figure 5(a)) and
ν = −1/3 (Figure 5(b)). In the latter case of an auxetic material, values of
parameters in the RAMP and SIMP schemes are significantly higher than
for the solid with positive Poisson’s ratio. This in turn, reflects in higher
values of the predicted stress energy and may influence the structural com-
pliance and optimal solid-void layout.
Table 1. Requirements on the parameters in GRAMP, RAMP and SIMP

interpolation schemes imposed by the conditions of optimality (Uapp ≥ Uopt )
and material isotropy (Uapp ≥ UHS ).
Uapp ≥ Uopt Uapp ≥ UHS
Uapp = UGRAMP1 q≥1 q≥3
1+ν 1+ν
Uapp = URAMP2 q ≥ max 1−ν
1−ν , 1+ν q ≥ max 3−ν
1−ν , 1+ν

Uapp = USIMP2 p ≥ max 2 2
1−ν , 1+ν p ≥ max 2 4
1−ν , 1+ν
1
independent of the basic material’s Poisson ratio ν and density θ,
2
calculated for θ = 1 in order to enforce validity for any θ ∈ [0, 1].
U (r,q)
r
(a)
-1 1
U (r,q)
r
(b)
-1 1
Figure 5. Plots of functions Uopt (lower solid line), UHS (upper solid line),
UGRAMP with q = 1 (lower dotted line), UGRAMP with q = 3 (upper dotted
line) and:
(a) USIMP with q = 3 (dashdotted line), URAMP with q = 2. Plots match
up with ν = 1/3 (in this unique case URAMP = UHS ) and θ = 0.7;
(b) URAMP with q = 5 (dashed line), USIMP with q = 6 (dashdotted line).
Plots match up with ν = −1/3 and θ = 0.7.
5 Examples of Material Layouts

In this Section, the optimal layout of materials in a cantilever is considered,
see Figure 6. Dimensions of a rectangular design area Ω are given by l × h =
2 × 1. Isotropic properties of basic materials are given by EA = 1, νA = 0.3,
EB = 0.5, νB = 0.15, and the fraction of material A in Ω is set to mA = 0.5.
The structure is due to a concentrated load P = 1. Subsequent calculations
are FEM-based with the mesh of 200 × 100 square, four-node elements with
two degrees of freedom per node and one integration point per element.
h
P
Figure 6. Geometry, load and boundary conditions for a cantilever.
Algorithmical Issues Let Ne stand for the number of elements in the

FE model, NG for the number of integration points within one element
and write n = NG Ne for the dimension of the space in which the discretized
minimum compliance problem is formulated. Next, set e = 1, . . . , Ne for the
number of arbitrary element. All symbols denoting fields in the preceding
Sections are now understood as n-dimensional vectors. Typical algorithm
of calculations in the considered optimization problem is incremental. The
m-th step, m ∈ N, of the algorithm consists of repetitive application of the
following operations:
1. Assuming that θ(m−1) and σ (m−1) are fixed, determine the constitutive
(m) (m)
properties k∗ and μ∗ in each element.
2. Solve the equilibrium equation of a structure. Calculate σ (m) .
3. Find θ(m) such that the isoperimetric condition is satisfied. For this,
the inner loop is required for the adjustment of the Lagrange multiplier
. It consists of determination of the vector
∂W ∗ (m) (m−1)
J (m) () = 2 (σ , θ )+I (102)
∂θ
where I denotes the vector whose all components are equal to 1. Note
2
that W ∗ in (102) should be replaced by certain Wapp = (σθE I)
Uapp
reported in Section 4 if one seeks the approximate solution.
Next,
J (m)
Δθ() = − (m) θ , (103)
J
is calculated with θ denoting the parameter ensuring stability of the
material density updating scheme. The negative sign in (103) stems
from the assumption that the compliance of a plate (or, equivalently,
its stress energy) decreases for increasing ratio of stronger material in
a composite. The loop is rerun until reaches the value for which

θ(m) = θ(m−1) + Δθ(), 0 ≤ θ(m) e ≤ 1 (104)
satisfies the isoperimetric condition. Determination of may be im-

plicit in the minimum compliance algorithm if one makes use of a
third-party numerical subroutine capable of performing constrained
minimization in Rn .
The algorithm stops when
σ (m) − σ (m−1) < σ (105)
where σ is fixed, small positive number.
Optimal Layout of Two Materials Optimal layout of stronger material

in a cantilever is shown in Figure 7(a). Figure 7(b) displays the division of
Ω into optimality regions Z1 , Z2 . Thorough check of the former shows that
tr σ ≈ 0 in that part of Ω which is marked in deep grey in Figure 7(b). Note
that the region Z3 is absent in the optimal layout.
Optimal Layout of One Material and Density Filtering for “0-1”

Designs The comparison of material layouts based on the exact, homog-
enization approach and those resulting from the application of simplified
solid-void scheme (GRAMP) are shown in Figure 8. However, for practi-
cal purposes it is desirable that the intermediate densities are filtered out
from the final design. Application of such filter allows not only for ob-
taining nearly black-and-white material layouts but also helps in avoiding
the checkerboard instability of numerical calculations. Roughly speaking,
filtering the density θ(x) at given x ∈ Ω works by averaging it over the
neighborhood of given radius r. By this, the tendency to reproduce the
fine-scale arrangement of solid and void is bypassed and composite regions
(a) (b)
Figure 7. Optimal design of a two-material cantilever: (a) layout of

stronger material, (b) regions of optimality – Z1 (deep grey), Z2 (grey).
(a) compliance = 55,95 (b) compliance = 56,21
Figure 8. Comparison of compliances for optimal design of a one-material

cantilever: (a) homogenization-based approach, (b) approach by GRAMP
for q = 1. In both cases, application of elements with four integration points
led to significant reduction of checkerboard-like patterns.
are introduced in Ω. In turn, these regions are penalized by proper pa-

rameter adjustment in the material interpolation formula. As a result, the
topology of final design becomes nearly “0-1” and the width of transition
between void and solid areas depends on the filtering radius r.
According to e.g. (Bendsøe and Sigmund, 2003), implementing a filtering
technique in the formulation of optimal design problem at hand does not
impose any additional limit on the space L∞ (Ω, [0, 1]), i.e. the space of
material distribution functions. Density filter can be applied at each x ∈ Ω
independently as a part of the material redistribution procedure.
The idea of filtering stems from the papers of Bruns and Tortorelli (2001)
and Bourdin (2001) and numerous filtering methods are reviewed in (Sig-
mund, 2007). The cone-shaped density filter of Bruns and Tortorelli may be
(a) compliance = 61,49 (b) compliance = 64,02
(c) compliance = 67,81 (d) compliance = 64,69
Figure 9. Comparison of unfiltered and filtered designs: (a) approach by

GRAMP q = 3 (unfiltered), (b) approach by GRAMP q = 3 (filtered), (c)
approach by SIMP for p = 3 (unfiltered), (d) approach by SIMP for p = 3
(filtered). In either case, elements with one integration point were used in
the disretization of the problem. Application of the density filter removed
the checkerboard-like patterns from the solutions.
coded by making use of the built-in MATLAB convolution function conv2.

For simplicity, assume that the design area Ω is mimicked by a k × l reg-
ular mesh of square elements with unit side length. Let the indices (i, j)
respectively determine the horizontal and vertical location of the element
mod
and write θ(i,j) for the material density. Density θ(i,j) modified by making
use of conv2 function is computed as the weighted average

h(m,n) θ(i+m,j+n)
m,n
mod
θ(i,j) = ,
h(m,n)
m,n (106)
m ∈ {−i + 1, −i + 2, . . . , −1, 0, 1, . . . , k − i},
n ∈ {−j + 1, −j + 2, . . . , −1, 0, 1, . . . , l − j}
where
h(m,n) = max{0, rmin − d(m,n) } (107)
and d(m,n) denotes the distance between the centers of given element (i, j)
and the one located at (i+m, j +n) while rmin stands for the filtering radius.
The comparison of unfiltered and filtered design obtained by GRAMP
and SIMP schemes is shown in Figure 9.
6 Optimal Plates of Small Volume

Consider the generalized shape optimization as discussed in Section 3. The
weaker material is degenerated to void. Minimum compliance problem has
been rearranged to the form

∗
Jmin = min ∗
F (σ) dx σ ∈ Σ(Ω) (108)
Ω
where
⎧2 (√

)
⎪
⎪ E |σI | + |σII | − 1 + sgn(σI σII )ν |σI σII | if θ < 1,
⎨E
∗
F (σ) = & '
⎪
⎪ 2 1
⎩ (σI )2 + (σII )2 − 2νσI σII + E if θ = 1.
E 2
(109)
Here denotes the Lagrangian multiplier for the isoperimetric condition

1
θ(x) dx = m (110)
|Ω|
Ω
with m ∈ (0, 1) and

1
θopt (σ) = min 1, √ |σI | + |σII | (111)

E
Consider the case of m being small. Then the multiplier becomes a
big number. From (109), in the regions where θ is small, it follows that
F ∗ (σ) = G(σ) and *

G(σ) = 2 |σI | + |σII | . (112)

E
∗
+
Hence Jmin = 2 J0 /E and

J0 = min |σI | + |σII | dx σ ∈ Σ(Ω) (113)
Ω
The function: σ1 = |σI | + |σII | satisfies all conditions for a norm in E2s .
Define a norm dual to · 1 by

|ε : τ |
ε−1 = sup τ ∈ E2
s , ε ∈ E2s . (114)
τ 1
The next step is to prove that
ε−1 = max(|εI |, |εII |) (115)
where εI , εII stand for the eigenvalues of ε.

Assume that εI > εII and τI > τII and use the following formula, see
(Lewiński, 2004a),
ε : τ = εI τI + εII τII − sin2 (αε − ατ )(εI − εII )(τI − τII ) (116)
with αε denoting an angle between e1 and first eigendirection of ε; ατ is

defined in a similar manner. Thus supremum in (114) is attained for the
case of αε = ατ and
⎧ ⎫
⎪ ⎪
⎨ ε τ +ε τ ⎬
I I II II
ε−p = sup
p τI , τII ∈ R (117)
⎪
⎩
1
⎪
⎭
|τI |p + |τII |p
for p = 1. It is known that for p ≥ 1

1 1
ε−p = εq , + = 1. (118)
p q
Hence for p = 1 we have q = ∞ which confirms (115).
This result paves the way to the formulation dual to (113). According
to the technique of Lagrange multipliers, see (Strang and Kohn, 1983), the
problem
⎧ ⎫
⎨ ⎬
J0 = min max σ 1 dx + f (v) − σ : ε(v) dx (119)
σ∈L2 (Ω,E2s ) v∈V ⎩ ⎭
div σ∈L2 (Ω,R2 ) Ω Ω
where
f (v) = p · v ds, (120)
∂Ωσ
is equivalent to
J0 = max J1 (v) + f (v) (121)
v∈V
with
J1 (v) = min σ1 − σ : ε(v) dx. (122)
σ∈L2 (Ω,E2s )
div σ∈L2 (Ω,R2 ) Ω
The next step is to localize this minimum by noting that

⎧
⎨0 if ε ∈ B−1 ,
min2 (σ1 − σ : ε) = (123)
σ∈Es ⎩
−∞ otherwise
where

B−1 = ε ∈ E2s ε−1 ≤ 1 (124)
or

B−1 = ε ∈ E2s |εI | ≤ 1, |εII | ≤ 1 . (125)
In order to justify (123) note that
σ : ε ≤ σ1 ε−1 (126)
hence
σ : ε ≤ σ1 for ε ∈ B−1 (127)
and
σ: ε
g(σ) = 1 − ≥0 (128)
σ1
for any ε ∈ B−1 . It follows that
σ1 − σ : ε = σ1 g(σ) (129)
hence
min (σ1 − σ : ε) = 0 (130)
σ∈E2s
for ε ∈ B−1 .
Assume now that ε ∈ / B−1 and ε1 = 1 + δ > 1, δ > 0. Then, for
such ε one obtains σ : ε ≤ (1 + δ)σ1 and g(σ) ≥ −δ. Since there exists σ
for which the last relation becomes equality, the minimum in (130) is not
bounded from below. Consequently, (123) is confirmed.
Now turn back to (121), (122) and find

J0 = max f (v) v ∈ V, ε(v(x)) ∈ B−1 for a.e. x ∈ Ω (131)
or, equivalently,

J0 = max f (v) v ∈ V, |εI (v(x))| ≤ 1, |εII (v(x))| ≤ 1

for a.e. x ∈ Ω . (132)
If in (111) is a big number, then
|σI | + |σII |
θopt = √ (133)
E
and θopt < 1. Consequently the isoperimetric condition (110) gives
*
J0
= (134)
E E m |Ω|
where J0 is defined by (113). Eventually
∗ 2
Jmin = (J0 )2 . (135)
E m |Ω|
Assume that v ∗ denotes the maximizer in (132). It follows that
∗ 2
Jmin = (f (v ∗ ))2 . (136)
E m |Ω|
The quantity J0 /σ0 with J0 defined by (113) can be interpreted as the

volume V0 of the fully stressed structure in which the absolute values of
both principal stresses attain the extreme value σ0 , see (Lewiński, 2004a,
Sec. 4). Thus V0 = J0 /σ0 .
From (113) and (132) it follows that the volume of the lightest, fully
stressed structure is given equivalently by
1
V0 = f (v ∗ ). (137)
σ0
The above formula has been derived already in 1904 in a remarkable paper
by Michell (1904) written under impression of the ideas of Clerk Maxwell,
see (Cox, 1965), (Maxwell, 1870). The contemporary point of view on
Michell structures has been set by Hemp (1973), where number of solu-
tions to the problems (113) or (132) are presented. The history of Michell’s
theory is discussed in this volume by G.I.N. Rozvany in an independent
article dealing with William Prager’s approach to the layout optimization
and to solid and structural mechanics in general, see (Rozvany, 1989).
Formulae (135) and (136) provide the following relationship between the
∗
minimum compliance Jmin and the minimum volume V0
∗ 2
Jmin = (σ0 V0 )2 . (138)
E m |Ω|
The equality m |Ω| = V0 is not used here, since the structure of minimal
compliance and the one corresponding to (113) are different. Indeed, struc-
ture solving the problem set by (108) and (109) is constructed of a mate-
rial with rank-2 microstructure, while the microstructure corresponding to
(113) is rank-1, forming the curvilinear strips along trajectories of σ being
the minimizer of (113).
A specific form of the problem (132) where the conditions |εI | ≤ 1,
|εII | ≤ 1 are expressed pointwise for a.e. x ∈ Ω stems from the fact that
integrand in (113) has a linear growth. Both problems (113) and (132) can
be interpreted as equilibrium problems of an effective body made from a
locking material. The mechanics of locking materials has been developed
in the works by Čyras (1972), Borkowski (2004a,b), Demengel and Suquet
(1986) or Telega and Jemio lo (1998). According to this theory, field σ in
(113) should be interpreted as a stress rate and not as a stress. The Michell
problem (113) has a similar mathematical structure as the problem

J1 = min |||σ||| dx σ ∈ Σ(Ω (139)
Ω
which occurs in two other subjects of optimization. It is a key problem in

the optimum design of thin plates of varying thickness loaded in plane (or
“a variable thickness sheet problem”). It also emerges in the free material
design problem with one load condition and with the trace constraint, see
the accompanying article by S. Czarnecki and T. Lewiński in the same
volume. In both problems | · ||| is a norm in E2s . Thus both (113) and
(139) involve integrands of linear growth. Consequently, the problem dual
to (139) takes a form similar to (131) with B1 replaced by a ball in E2s with
respect to the norm ||| · |||∗ dual to ||| · ||| in the sense of (114).
7 Comments on Related Topics and Final Results

The paper has shown that the relaxed version of the minimum compliance
problem of two-material plates loaded in plane, i.e. the problem (33)-(34),
with an abstract definition of the relaxed potential W ∗ , see (32), can be
put in an explicit way. This potential, constructed by (57)-(59), serves as a
convenient starting point for developing the numerical methods of solutions.
A passage from the initial formulation of the minimum compliance prob-

lem (20) to its relaxed form (33)-(34), (57)-(59) is probably one of the great-
est achievements of solid mechanics in the second half of the 20th century.
Its history has been outlined in various settings by Tartar (1980), Cherkaev
(2000), Allaire (2002), Milton (2002). It is worth pointing out here that
techniques of solving the main problem (38) - to find the relaxed functional
of the original problem – involve differential equations of equilibrium con-
cealed in the conditions τ ∈ Σ(Y ). This is why, the problem (33) upon
its relaxation assures the equilibrium not only of the body, but also of the
underlying microstructure at each point of the microstructural cell Y cor-
responding to each point of the body. The only weak point of the approach
is that the underlying microstructures are not unique.
The lecture can be extended, yet with essential changes, to the prob-
lem of optimum design of the two-material thin plates in bending, within
the known Kirchhoff’s description. The first results on this topic were pub-
lished by K. Lurie and A.V. Cherkaev in 1980s. A detailed exposition of this
topic can be found in (Lewiński and Telega, 2000, Ch. VI). This chapter is
based on the original results on homogenization of thin plates obtained by
G. Duvaut (1976-1979) and on the papers by A.V. Cherkaev and K. Lurie
(1984-1986) on optimum design of thin plates made of two isotropic materi-
als, cf. (Cherkaev and Kohn, 1997). The main tool used in the cited papers
on optimum design of thin plates is the translation method which is similar,
yet not identical, with the method used in Section 2.3. In the Kirchhoff
problem the bending moments Mij satisfy only one differential equation
of local equilibrium, which is reflected in the properties of the translation
tensor α T . In case of Kirchhoff’s plate, function det M = −M : (T M ) is
quasiconvex and not quasiaffine, see (41), which leads to α ≥ 0.
A common feature of the in-plane and bending problems of plates is
that only one stress tensor and one strain tensor are involved, linked by the
elasticity tensor. The difference lies in the mathematical structure of the
field equations. This simplicity can also be noted in the theory of membrane
shells, see (Dzierżanowski and Lewiński, 2003). In the theory of unbalanced
(viz. transversely asymmetric) thin plates and shells at least two stress
fields and two strain fields are involved. In the case of shallow shells the
equilibrium equations are weakly coupled; the results discussed here were
extended to the case of two-material transversely symmetric shallow shells
in (Dzierżanowski, 2012a).
In the theory of moderately thick plates the density of elastic energy is
a sum of the bending energy and the transverse shear energy. Two stiff-
ness tensors are involved, describing both kinds of deformations. The same
mathematical model applies to sandwich plates with soft core, see (Lewiński,
1991). This topic has been discussed in (Krog and Olhoff, 1997), (Dı́az et
al., 1995), (Czarnecki et al., 2008). Due to the presence of two stiffness ten-
sors the problem can be set in different ways, hence the relaxed formulation
is expected to be non-unique.
The theory of relaxation of the minimum compliance problem serves
as a pattern for more complicated problems concerning optimum design
of structures made of composites. Recalled results are the inspiration for
contemporary research on numerical homogenization and hierarchical mod-
eling.
Bibliography
G. Allaire, R.V. Kohn, Optimal lower bounds on the elastic energy of a
composite made from two non well-ordered isotropic materials, Quart.
Appl. Math., 52:311–333, 1994.
G. Allaire, The homogenization method for topology and shape optimiza-
tion, In: G.I.N. Rozvany (Ed.) Topology optimization in structural me-
chanics, pp. 101–134, CISM Courses and Lectures 374, Springer, Wien,
1997.
G. Allaire, Shape Optimization by the Homogenization Method, Springer,
New York, 2002.
G. Allaire, E. Bonnetier, G. Francfort, F. Jouve, Shape optimization by the
homogenization method, Numer. Math., 76:27–68, 1997.
M.P. Bendsøe,O. Sigmund, Material interpolation schemes in topology op-
timization, Arch. Appl. Mech., 69:635–654, 1999.
M.P. Bendsøe,O. Sigmund, Topology Optimization. Theory, Methods and
Applications, Springer, Berlin/Heidelberg, 2003.
A. Borkowski, On dual approach to piecewise-linear elasto-plasticity. Part I:
Continuum models, Bull. Pol. Acad. Sci. Tech. Sciences, 52(4):329–343,
2004.
A. Borkowski, On dual approach to piecewise-linear elasto-plasticity. Part
II: Discrete models, Bull. Pol. Acad. Sci. Tech. Sciences, 52(4):345–352,
2004.
B. Bourdin, Filters in topology optimization, Int. J. Numer. Meth. Eng.,
50:2143–2158, 2001.
M. Bruggi, On the solution of the checkerboard problem in mixed-FEM
topology optimization, Computers and Structures, 86:1819–1829, 2008.
T.E. Bruns, D.A. Tortorelli, Topology optimization of nonlinear elastic
structures and compliant mechanisms, Comput. Methods Appl. Mech.
Eng., 190(26-27):3443-3459, 2001.
A.V. Cherkaev, Variational Methods for Structural Optimization, Springer,
New York, 2000.
A.V. Cherkaev, L.V. Gibiansky, Coupled estimates for the bulk and shear
moduli of a two-dimensional isotropic elastic composite, J. Mech. Phys.
Solids, 41(5):937–980, 1993.
A.V. Cherkaev, R.V. Kohn (Eds.), Topics in the Mathematical Modeling of
Composite Materials, Birkhäuser, Boston, 1997.
H.L. Cox, The design of structures of least weight, Pergamon Press, Oxford,
1965.
A. Čyras, Optimization theory of perfectly locking bodies, Arch. Mech.,
24(2):203–210, 1972.
S. Czarnecki, M. Kursa, T. Lewiński, Sandwich plates of minimal compli-
ance, Comp. Meth. Appl. Mech. Engng., 197:4866–4881, 2008.
F. Demengel, P. Suquet, On locking materials, Acta Appl. Math., 6:185–211,
1986.
A.R. Dı́az, R. Lipton, C.A. Soto, A New Formulation of the Problem of
Optimum Reinforcement of Reissner-Mindlin Plates, Comp. Meth. Appl.
Mech. Engng., 123:121–139, 1995.
G. Duvaut, J.-L. Lions, Inequalities in Mechanics and Physics, Springer-
Verlag, Berlin, 1976.
G. Dzierżanowski, Stress energy minimization as a tool in the material lay-
out design of shallow shells, Int. J. Solids Struct., 49(11-12):1343–1354,
2012.
G. Dzierżanowski, On the comparison of material interpolation schemes
and optimal composite properties in plane shape optimization, Struct.
Multdisc. Optim., 2012. DOI: dx.doi.org/10.1007/s00158-012-0788-2
G. Dzierżanowski, T. Lewiński, Layout optimization of two isotropic mate-
rials in elastic shells, J. Theor. Appl. Mech., 41(3): 459–472, 2003.
Z. Hashin, S. Shtrikman, A variational approach to the theory of the elastic
behaviour of multiphase materials, J. Mech. Phys. Solids, 11:127–140,
1963.
W. Hemp, Optimum Structures, Clarendon Press, Oxford, 1973.
L.A. Krog, N. Olhoff, Topology and reinforcement layout optimization of
disk, plate, and shell structures, In: G.I.N. Rozvany (Ed.), Topology Op-
timization in Structural Mechanics, pp. 237–322, Springer, Wien, 1997.
T. Lewiński, On displacement-based theories of sandwich plates with soft
core, J. Eng. Math., 25:223–241, 1991.
T. Lewiński, Michell structures formed on surfaces of revolution, Struct.
Multidisc. Optim., 28:20–30, 2004.
T. Lewiński, Homogenization and optimal design in structural mechanics.
In: P.P.Castañeda, J.J. Telega, B. Gambin (Eds.), Nonlinear Homoge-
nization and its Application to Composites, Polycrystals and Smart Ma-
terials, NATO Science Series II, Mathematics, Physics and Chemistry
170, pp. 139–168, Kluwer, Dordrecht, 2004.
T. Lewiński, J.J. Telega, Plates, Laminates and Shells. Asymptotic Anal-

ysis and Homogenization, World Scientific Publishing, Singapore/New
Jersey/London/Hong Kong, 2000.
R. Lipton, On a saddle-point theorem with application to structural opti-
mization. J. Optim. Theory. Appl., 81:549–568, 1994.
C. Maxwell, On reciprocal figures, frames and diagrams of forces, Trans.
Royal Soc. Edinburgh, Scientific Papers II, 26:161-207, 1870.
A. Michell, The limits of economy of material in frame structures, Phil.
Mag., 8:589-597, 1904.
G.W. Milton, The Theory of Composites, University Press, Cambridge,
2002.
J. Nečas, I. Hlavaček, Mathematical Theory of Elastic and Elasto-Plastic
Bodies. An Introduction, Elsevier, Amsterdam, 1981.
R.T. Rockafellar, Integral functionals, normal integrands and measurable
selections. In: L. Waelbroeck (Ed.) Nonlinear Operators and the Cal-
culus of Variations, Lecture Notes in Mathematics (543), pp. 157–207,
Springer, New York, 1976.
G.I.N. Rozvany, Structural Design via Optimality Criteria, Kluwer Aca-
demic Publishers, Dordrecht, 1989.
E. Sanchez-Palencia, Non-Homogeneous Media and Vibration Theory,
Springer-Verlag, Berlin, 1980.
O. Sigmund, Morphology-based black and white filters for topology opti-
mization, Struct. Multidisc. Optim., 33:401-424, 2007.
M. Stolpe, K. Svanberg, An alternative interpolation scheme for minimum
compliance topology optimization, Struct. Multidisc. Optim., 22:116–
124, 2001.
G. Strang, R.V. Kohn, Hencky-Prandtl nets and constrained Michell trusses.
Comp. Meth. Appl. Mech. Engng., 36:207–222, 1983.
C. Talischi, G.H. Paulino, C.H. Le, Honeycomb Wachspress finite elements
for structural topology optimization, Struct. Multidisc. Optim., 37:569–
583, 2009.
L. Tartar, An introduction to the homogenization method in optimal design,
In: B. Kawohl, O. Pironneau, L. Tartar and J.-P. Zolésio (Eds.), Optimal
Shape Design, Lecture Notes in Mathematics 1740, pp. 47–156, Springer,
Berlin/Heidelberg, 2000.
J.J. Telega, S. Jemio lo, Macroscopic behaviour of locking materials with
microstructure. Part I. Primal and dual elastic-locking potential. Relax-
ation, Bull. Pol. Acad. Sci. Tech. Sciences, 46(3):265–276, 1998.
The Free Material Design in linear elasticity
Slawomir Czarnecki and Tomasz Lewiński

Warsaw Univeristy of Technology, Faculty of Civil Engineering,
al. Armii ludowej 16; 00-637 Warsaw, Poland
Abstract The Free Material Design (FMD) is a branch of topology

optimization. In the present article the FMD formulation is confined
to the minimum compliance problem within the linear elasticity
setting. The design variables are all elastic moduli, forming a Hooke
tensor C at each point of the design domain. The isoperimetric
condition concerns the integral of the p-norm of the vector of the
eigenvalues of the tensor C. The most important version refers to
p = 1, imposing the condition on the integral of the trace of C.
The paper delivers explicit stress-based formulations and numerical
solutions of the FMD problems in the case of a single load case as
well as for a general case of a finite number of load conditions.
1 Introduction
The physical properties of the elastic bodies are determined by the distribu-
tion of the elasticity moduli Cijkl ; i, j, k, l = 1, 2, 3. Due to known symmetry
properties the number of independent moduli equals : 6+5+4+3+2+1 = 21.
The contemporary technology provides means for controlling the distribu-
tion of these moduli within the design domain. A prerequisite for the opti-
mum design of the layouts of the moduli within the body is a proper insight
into the algebraic structure of the Hooke tensor C. The explicit algebraic
representation of this tensor is delivered in (Rychlewski, 1984; Mehrabadi
and Cowin, 1990; Sutcliffe, 1992; Moakher and Norris, 1996). The two-
dimensional case is discussed in (Blinowski et al., 1996). The papers men-
tioned indicate that the Hooke tensor can be decomposed as follows
C = λ1 ω1 ⊗ ω 1 + . . . + λm ω m ⊗ ω m (1.1)
where m = d(d + 1)/2; d equals 3 in the spatial case, then m = 6, and

d = 2 in the plane case, and then m = 3. The eigenvalues λK , K =
1, . . . , m of C will be called Kelvin moduli, as suggested in (Rychlewski,
1984). The tensors ω K are symmetric tensors of rank 2, mutually orthogonal

DOI 10.1007/978-3-7091-1643-2_9, © CISM, Udine 2014
214 S. Czarnecki and T. Lewiński
and normalized. These tensors can be interpreted as vectors in Rm . The

problem of optimal choice of the moduli Cijkl is replaced by the problem of
optimal choice of the Kelvin moduli and of the vectors ω K of the orthogonal
basis in Rm at each point of the design domain, see (Czarnecki and Lewiński,
2011,2012).
The most important merit function in the optimum design of elastic
structures is the compliance. In case of a single load condition the compli-
ance equals the work done by the loading on the displacement due to this
loading. In case of n loading conditions the merit function is usually taken
as a weigthed sum of the compliances with weights η1 , . . . , ηn such that their
sum is 1, treating the loads as acting independently. A natural definition of
the compliance arises upon defining the virtual work of the loading, which
is a linear functional whose arguments are elements of the space of virtual
displacements. Thus the compliance is the value of this functional on the
solution to the problem referred to the same loading. The compliance thus
defined is equal to the doubled value of the elastic energy stored in the body.
The minimal compliance is referred to the same loading but to a different,
optimized structure.
Under the term Free Material Design (FMD) we understand here the
problem of minimization of the compliance over all possible distributions
of the elastic moduli, the components of the tensor field C, within a given
design domain. The shape of the design domain is not subject to changes
during the optimization process. Moreover, the mass density plays no role
here, in particular, it does not parameterize the elastic moduli, in contrast
to many popular optimum layout schemes. Consequently, the isoperimetric
condition cannot be expressed in terms of the mass density distribution to
bound the total mass of the body. Not only the mass density, but no other
parameterization of the elastic moduli distribution will be used. This means
that the isoperimetric condition in the FMD problem must be expressed in
terms of the invariants of C. In the papers by Bendsøe et al. (1994, 1995)
originating the subject in its current formulation the two isoperimetric con-
ditions have been suggested, expressing the boundedness of the integral of
the p-norm ||λ||p , p = 1, 2, of the vector λ = [λ1 , . . . , λm ] composed of
the Kelvin moduli. Note that ||λ||1 = tr C. The quantity ||λ||p can be
interpreted as a unit cost of the material, proportional to the values of the
Kelvin moduli. The FMD problem with the mentioned isoperimetric con-
dition has been the subject of numerous studies, see (Kočvara et al. 2007,
2008), (Turteltaub and Washabaugh, 1999), (Hörnlein et al., 2001), (Bar-
barosie and Lopes, 2008), (Haslinger et al., 2010), which put emphasis of
various aspects of the problem. The Thesis of Werner (2000) summarizes
the research on the topic up to 2000. All the papers mentioned above have
The Free Material Design in Linear Elasticity 215
a common feature: they are based on the approach in which the displace-
ment field is the main state (or behavioral) variable. Consequently, in each
case the FMD is led to a saddle point problem. On the other hand, the
theory of minimum compliance of two-material structures, along with its
relaxation theory, see Cherkaev (2000) and the article (Dzierżanowski and
Lewiński, 2013) in this book, teaches us that the stress-based approach is
the most convenient point of departure, since the compliance is expressed
as minimum of the complementary energy over the stress states being stat-
ically admissible. Since two minima: over the design variables and over
the state variables can be switched, the minimum operation over the de-
sign variables becomes an internal operation for which the stress states are
viewed as fixed. This paves the way to an analytical solution of the nested
optimization problem. The theorem behind this rearrangement is due to
Rockafellar’s paper (1976) which teaches us that the minimization of the
integral functionals over the fields can be replaced by integration of the
function defined through a pointwise minimization.
The present paper is organized as follows. Sec. 2 recalls the equilibrium
problem of a linear elastic body. The algebraic structure of Hooke tensors
is recalled in Sec. 3. Why the compliance can be expressed by the comple-
mentary energy is cleared up in Sec. 4. Further part of the lecture concerns
the FMD problem: for the single load case (Sec. 5–Sec. 8), for the two load
case in 2D (Sec. 9, Sec. 10) and for the multi-load cases in Secs 11–13.
The results of the FMD in the one load case in 2D look similar to
Michell’s solution. The reason of this similarity is explained in Sec. 6; prob-
lem (6.28) is similar to Michell’s kinematic formulation; the difference is in
the shape of the locking locus B. In (6.28) this is a ball with respect to the
Euclidean norm || · ||2 while in Michell’s formulation the ball is defined by
∞-norm. Thus the lecture shows why the FMD problem with a single load
conditions is expressed by both primal and dual problems of an effective
medium with locking, see (Demengel and Suquet, 1986).
The numerical methods used are based on new concepts of interpolating
statically admissible stresses; the underdeterminate algebraic systems are
solved by using the SVD decompositions.
It is remarkable, that the singular values of matrices appear twice: a)
as a tool to express the integrand of the effective locking complementary
potential in (13.4) and b) in the mentioned SVD based numerical method.
Therefore, prior to study this article it is recommended to study the related
chapters of the textbooks on linear algebra which concern the singular val-
ues.
The summation convention for repeated indices is adopted. The eigen-
values of a square symmetric matrix A are denoted by μi (A). The singular
values of an arbitrary matrix A are denoted by si (A). By ijk we denote

the permutation symbol, defined as usual, see (Fung, 1965). The set of
symmetric tensors of rank 2 is denoted by E2s . Let σ, τ ∈ E2s ; the scalar
product σ · τ is defined by σij τij , i, j = 1, 2, . . . , d; d is the dimension of
the problem, hence d = 2 (plane case) or d = 3 (three-dimensional case).
The paper does not introduce regularity conditions of the fields in the
design domain. That is why we do not distinguish between inf and min
and between sup and max operations. Not all the mathematical questions
are resolved; in particular we shall not provide the conditions under which
minimum in (6.10) is attained. The theoretical discussion in (Demengel and
Suquet, 1986) on the locking materials problems suggest that the minimum
in (6.10) is attained in the space of measures. Thus the mathematical
questions can be solved only by using very subtle tools of the contemporary
calculus of variations.
2 Equilibrium
2.1 Plane stress case
Consider an elastic plate of unit thickness, loaded in-plane, whose middle
plane Ω is a symmetry plane of the boundary loading (tractions) of intensity
T (s), acted along the segment Γ1 of the contour Γ of the domain Ω; s is the
natural parameter of the contour. Let us assume that along the segment
Γ2 of the contour Γ the plate is fixed, hence the in-plane displacement
vector vanishes along the contour Γ2 . The plane Ω is parameterized by
a Cartesian system (x1 , x2 ) with the orthonormal basis (e1 , e2 ). Points
in Ω are denoted by x and are identified with the pair (x1 , x2 ). In the
plane stress state the deformation state is determined by the displacement
field u = (u1 , u2 ) referred to the middle plane. We shall not recall here
the standard arguments leading to the plane stress model. The state of
deformation is characterized by the components εij = (ui,j + uj,i )/2 where
∂
( ),j = ; i, j = 1, 2. The right-hand side of this definition will be
∂xj
denoted by εij (u) hence ε(u) is a symmetric part of the gradient of u. The
virtual displacement field will be denoted by v = (v1 , v2 ); these fields are
assumed to satisfy the same kinematic conditions as the unknown fields u1 ,
u2 : they vanish along Γ2 . By writing v ∈ V (Ω) we understand that v is
sufficiently regular and vanish along Γ2 . Usually, the regularity assumptions
should assure that εij ∈ L2 (Ω) or V (Ω) ⊂ H 1 (Ω, R2 ), see (Duvaut and
Lions, 1976), (Nečas and Hlavaček, 1981), (Ciarlet, 1988). One can show
that if εij (v) = 0, i, j = 1, 2, then v is composed of translations and
infinitesimal rigid rotations around the axis normal to the plane Ω. The
in-plane stress resultants in the plate are denoted by: τ = (τij ), i, j = 1, 2;

they will be called stresses, for simplicity. The components τij are referred
to the middle plane Ω. At each point x ∈ Ω the components τij (x), εij (x)
are symmetric; they form symmetric tensors, which will be denoted by:
τ (x), ε(x) ∈ E2s . The body forces will be omitted. Thus the stresses satisfy
the equilibrium conditions within Ω
τij,j = 0 (2.1)
and the equilibrium conditions along the loaded contour Γ1 :
τij nj = Ti on Γ1 (2.2)
where n = (n1 , n2 ) is an outward normal vector to Γ. Eq. (2.2) is written

under the condition of the symmetry of τ . The conditions (2.1), (2.2) are
equivalent to the variational equilibrium equation:

τ · ε(v)dx = T · vds ∀ v ∈ V (Ω) (2.3)
Ω Γ1
The regularity conditions of the tractions T are specified in (Duvaut and

Lions, 1976). The proof of the implication: (2.3) implies (2.1) and (2.2) is a
good exercise of the variational calculus. The right hand side of (2.3) will be
treated as a value of a linear form f (·) on the virtual field v ∈ V (Ω); it will
be denoted by f (v). Note that τ · ε = τij εij is the scalar product of τ and
ε in E2s while T · v is the scalar product in R. The set of the fields τ on Ω
satisfying (2.3), or the set of statically admissible trial stress fields, will be
denoted by Σ(Ω). The set Σ(Ω) is a convex subset of the space L2 (Ω, E2s ).
If the components τij are appropriately smooth, the conditions (2.1) will be
satisfied identically if
τ11 = F,22 , τ22 = F,11 , τ12 = −F,12 (2.4)
where F (x) is an Airy function. This function should satisfy the boundary
conditions due to (2.2). Let us write (2.4) in the brief manner
τij = ik3 jl3 F,kl (2.5)
and re-write (2.2) in the form
ik3 jl3 F,kl nj = Ti on Γ1 (2.6)
Thus the function F should satisfy (2.6) and be appropriately regular.

2.2 The spatial case

Consider a 3D body occupying the domain Ω in R3 and whose segment
Γ2 of the boundary is fixed. The segment Γ1 of the boundary is subject
to the tractions of intensity T . The domain Ω is parameterized by the
Cartesian system (x1 , x2 , x3 ) with the orthonormal basis e1 , e2 , e3 . The
vector field of displacements is denoted by u = (u1 , u2 , u3 ). The strain
components are defined as previously by εij (u); now i, j = 1, 2, 3. The
trial displacement field is denoted by v = (v1 , v2 , v3 ). The field v is
kinematically admissible if is appropriately regular in Ω and if vanishes on
Γ2 ; we write v ∈ V (Ω). The statically admissible trial stress fields satisfy
(2.1), (2.2) for i, j = 1, 2, 3 while n is outward normal to the surface
Γ1 . The set Σ(Ω) of such fields τ on Ω is alternatively defined as the set
of fields satisfying (2.3), where the integrals concern now Ω ⊂ R3 and the
surface Γ1 . In 3D case the general integral of equations (2.1) can be written
in the Finzi form, see (Fung, 1965, Sec. 9.2, Eq. (9.2.23)):
τij = imr jns φrs,mn (2.7)
where φrs = φsr and ijk is the permutation symbol. Thus six independent
stress functions determine the stresses satisfying the equilibrium conditions
within the body. On the surface Γ1 the following conditions should be
fulfilled:
imr jps φrs,mp nj = Ti (2.8)
3 The Hooke law

3.1 Plane case
The linear law which links stresses and strains in the plane stress case
reads:
σij = Cijkl εkl (3.1)
where i, j, k, l = 1, 2. The components Cijkl satisfy the symmetry condi-
tions:
Cijkl = Cjikl = Cklij (3.2)
and the conditions of positive definiteness
Cijkl ξij ξkl α||ξ||2 ∀ ξ ∈ E2s (3.3)
where α > 0. The set of such tensors C will be denoted by E4s .

The plane stress description applies to the plate-like bodies in which the
planes parallel to the middle plane Ω are the planes of material symmetry.
Then the assumption σ33 ≈ 0 makes it possible to eliminate ε33 which leads
to the equation of the form (3.1). The moduli Cijkl are called reduced
moduli of the generalized plane stress state. In the case of isotropy these
moduli can be written with using the effective Kelvin modulus k and the
Kirchhoff modulus μ as follows
C1111 = C2222 = k + μ, C1122 = k − μ, C2211 = C1122 , C1212 = μ,

C2121 = C2112 = C1221 = C1212 (3.4)
while the other Cijkl components vanish; k > 0, μ > 0. If the Young
modulus equals E and the Poisson ratio equals ν, then
E E
k= , μ= (3.5)
2(1 − ν) 2(1 + ν)
Let us note that & '−1
1 1 1
E= + (3.6)
2 2k 2μ
or the Young modulus is a harmonic mean of the moduli 2k and 2μ.
Let us introduce the auxiliary tensors:
1
E 1 = √ (e1 ⊗ e1 + e2 ⊗ e2 )
2
1
E 2 = √ (e1 ⊗ e1 − e2 ⊗ e2 ) (3.7)
2
1
E 3 = √ (e1 ⊗ e2 + e2 ⊗ e1 )
2
They form an orthonormal basis. Instead of the conventional decomposition:
C = Cijkl ei ⊗ ej ⊗ ek ⊗ el (3.8)
we consider the decomposition:
C = CKL E K ⊗ E L , K, L = 1, 2, 3 (3.9)
In case of isotropy the coefficients CKL form the diagonal matrix

⎡ ⎤
2k 0 0
⎢ ⎥
[CKL ] = ⎣ 0 2μ 0 ⎦ (3.10)
0 0 2μ
We note that
λ1 = 2k, λ2 = λ3 = 2μ (3.11)
are the eigenvalues of the tensor C in case of isotropy. The corresponding

eigentensors read
P 1 = E1 ⊗ E1, P 2 = E2 ⊗ E2, P 3 = E3 ⊗ E3 (3.12)
One can easy check that P K are projection operators or P K P K = P K ,
P L P K = 0 for K = L. In the isotropic case the representation (3.9) can
be re-written as
C = λ1 P 1 + λ2 P 2 + λ3 P 3 (3.13)
where λK are given by (3.11) and P K are given by (3.12). However, the
representation (3.13) is valid in general, for each anisotropy in the plane
case. The projection operators have always the form
P 1 = ω 1 ⊗ ω1 , P 2 = ω2 ⊗ ω 2 , P 3 = ω3 ⊗ ω3 (3.14)
where ωK ∈ E2s and
ωK · ω L = δKL , K, L = 1, 2, 3 (3.15)
and the dot means the scalar product in E2s . The tensors ω K are called,
after (Rychlewski, 1984, Blinowski et al. 1996), the eigenstates. The moduli
λK will be called Kelvin moduli; we shall assume that λ1 ≥ λ2 ≥ λ3 .
In the case of isotropy, two Kelvin moduli coincide, see (3.11). The cor-
responding eigenstates describe two possible modes of pure shear. Isotropy
implies that the shear modulus is the same for both the shear modes.
The tensor inverse to C has the same projectors, or
1 1 1
C −1 = P1 + P2 + P3 (3.16)
λ1 λ2 λ3
The optimization process can lead to a degeneration of (3.13), to the forms
C = λ1 P 1 , C = λ1 P 1 + λ2 P 2 (3.17)
Then the condition (3.3) is not satisfied; the question of correctness of the
boundary value problem must be reconsidered, as will be seen later.
Remark 3.1
Algebra of tensors of E2s class can be replaced by the algebra of vectors in
R3 with using the basis (3.7). A tensor a ∈ Es2 can be decomposed as:
a = ãK E K , K = 1, 2, 3. Thus the tensor a is treated as a vector in R3 of
components ãK .
The basis (3.7) is used rather rarely. Usually it is sufficient to use the
basis
B 1 = e1 ⊗ e1 , B 2 = e2 ⊗ e2 , B3 = E3 (3.18)
which is also orthonormal. By equating the decompositions:

a = aij ei ⊗ ek , a ∈ E2s , j, j = 1, 2
(3.19)
a = aK B K , K ∈ 1, 2, 3
we arrive at the rules
√
a1 = a11 , a2 = a22 ,
a3 = 2a12 (3.20)
√
and further we can treat the triple [a11 a22 2a12 ] as a vector in R3 ,
T
which obeys the known rules of rotations. In particular the fields ωK can
be treated as vectors in R3 of components which form an orthonormal basis.
3.2 The spatial case

The stresses are linked with strains by (3.1) where now i, j, k, l =
1, 2, 3. The assumptions of symmetry (3.2) and positive definiteness (3.3)
hold. Let us introduce the basis in Es2
B 1 = e1 ⊗ e1 , B 2 = e2 ⊗ e2 , B 3 = e3 ⊗ e3 ,
1 1
B 4 = √ (e2 ⊗ e3 + e3 ⊗ e2 ), B 5 = √ (e1 ⊗ e3 + e3 ⊗ e1 ), (3.21)
2 2
1
B 6 = √ (e1 ⊗ e2 + e2 ⊗ e1 )
2
and represent tensor C in this basis
C = CKL B K ⊗ B L , K, L = 1, 2, . . . , 6 (3.22)
The components CKL are determined by the components Cijkl by
C11 = C1111 , C12 = C1122 , C12 = C1133 ,
√ √ √
C14 = 2C1123 , C15 = 2C1113 , C16 = 2C1112 ,
√
C22 = C2222 , C23 = C2233 , C24 = 2C2223 ,
√ √
C25 = 2C2213 , C26 = 2C2212 , (3.23)
√ √ √
C33 = C3333 , C34 = 2C3323 , C35 = 2C3313 , C36 = 2C3321 ,
C44 = 2C2323 , C45 = 2C2313 , C46 = 2C2312 ,
C55 = 2C1313 , C56 = 2C1312 , C66 = 2C1212 ,
Due to the representation (3.22) one can define correctly the eigenvalues of
C, according to the rules of linear algebra. This is the crucial point of the
present paper. Let us represent tensors σ and ε in the basis (3.21):
σ = σK B K , ε = εK B K , K = 1, . . . , 6 (3.24)
where
σ1 = σ11 , σ2 = σ22 , σ3 = σ33 ,
√ √ √ (3.25)
σ4 = 2σ23 , σ5 = 2σ13 , σ6 = 2σ12
and the equations for the strain components are similar. The equation
σ = Cε can now be interpreted as the equation linking two vectors in R6 ,
or: σK = CKL εL . The eigenvalues of tensor C are the eigenvalues of the
matrix [CKL ].
The isotropic material is characterized by the bulk and shear moduli are
expressed by
E E
k= , μ= (3.26)
3(1 − 2ν) 2(1 + ν)
and E, ν are Young modulus and Poisson’s ratio. The moduli Cijkl read
4
C1111 = C2222 = C3333 = k + μ
3
2
C1122 = C1133 = C2233 = k − μ (3.27)
3
C1212 = C1313 = C2323 = μ
and the other moduli follow from symmetry rules (3.2) or vanish. The
components CKL in the basis B K ⊗ B L can be found by (3.23). The rules
of transformation due to rotations are given in (Mehrabadi and Cowin,
1990), see (Czarnecki and Lewiński, 2006). We note that the moduli
λ1 = 3k, λ2 = 2μ, λ3 = 2μ, λ4 = 2μ, λ5 = 2μ, λ6 = 2μ (3.28)
are eigenvalues of the tensor C.
In case of an arbitrary anisotropy the spectral decomposition of the
tensor C has the form
#
m
C= λK ωK ⊗ ωK , m=6 (3.29)
K=1
where λK > 0, K = 1, . . . , 6. We assume that λ1 ≥ λ2 ≥ . . . ≥ λ6 . The

projectors P 1 = ω 1 ⊗ω1 , P 2 = ω2 ⊗ω2 , . . . , P 6 = ω6 ⊗ω6 are determined
by six tensors ωK ∈ E2s satisfying (3.15) for K, L = 1, . . . , 6. The tensor
C −1 has the representation
#
m
1
C −1 = ωK ⊗ ωK , m=6 (3.30)
λK
K=1
where m = 6 and ωK are the same as in (3.29). The moduli λK will be

called Kelvin moduli, according to Rychlewski’s (1984) suggestion.
4 The boundary value problem of elasticity.

The compliance and the stiffness
The problem reads: find u ∈ V (Ω) such that
σ = Cε(u) and σ ∈ Σ(Ω) (4.1)
The formulation above refers to both 2D (d = 2) and 3D (d = 3) problems.
Thus among τ ∈ Σ(Ω) one should find the stress state which is linked by
Hooke’s law with the strain ε(u) and the field u vanishes on Γ2 .
The compliance of the elastic body is defined by
Υ = f (u) (4.2)
Thus Υ is a positive number, equal to the doubled value of the elastic energy
stored in the deformed body. The meaning of the compliance is directly
given by the formula f (u), hence this is the work done by the loading on
the displacements of the body caused by the same loading. The stiffness of
1
the body is defined as . Thus maximization of the stiffness amounts
f (u)
to minimization of the compliance.
In the theory of optimum design of elastic bodies the compliance can be
expressed by Castigliano’s variational formula, which is crucial in the theory
of optimum design. To write down this formula in a compact manner we
introduce the scalar product in the space L2 (Ω, E2s )

(σ, τ )C −1 = σ · (C −1 τ )dx (4.3)
Ω
where the dot means the scalar product in Es2 . The scalar product (4.3)
defines the norm +
||τ ||C −1 = (τ , τ )C −1 (4.4)
in L2 (Ω, E2s ).
According to the Castigliano theorem the compliance (4.2) is expressed
by
Υ = min ||τ ||2C −1 τ ∈ Σ(Ω) (4.5)
The proof of (4.5) can be found in (Duvaut and Lions, 1976).
5 The FMD problem for the case of the Kelvin

moduli being fixed. A single load case
The task is to fill up the domain Ω in Rd (d = 2 or 3) by a linear elastic
material of arbitrary eigenstates ω K and given Kelvin moduli λK , K =
1, . . . , m; m = d(d + 1)/2. Among such fields of elastic moduli we are

looking for those which make the compliance Υ minimal. The loading T on
Γ1 is fixed. The domain Γ1 of its application is not subject to optimization.
Thus the values λK , K = 1, . . . , m are given at each point x. Let λ1 ≥
λ2 ≥ . . . ≥ λm ≥ λ0 > 0. The assumption λ0 > 0 is equivalent to (3.3). By
(3.16) and (3.14) we can write
#
m
1
τ · (C −1 τ ) = (ω K · τ )2 (5.1)
λK
K=1
We decompose the unit tensor I 4 of E4s
#
m
I4 = ωK ⊗ ωK (5.2)
K=1
and eliminate ω1 from (5.1)
1 # n
−1
τ · (C )τ = ||τ ||2 + νL (ω L · τ )2 (5.3)
λ1
L=2
where ||τ ||2 = τ · τ and

1 1
νL = − (5.4)
λL λ1
Note that νL+1 > νL , L = 2, . . . , m − 1. We re-write (4.5)
⎧ ⎡ ⎤ ⎫
⎨ # ⎬
1
m

Υ(ω2 , . . . , ω m ) = min ⎣ ||τ ||2 + νL (ω L · τ )2 ⎦ dx τ ∈ Σ(Ω)
⎩ Ω λ1 ⎭
L=2
(5.5)
The left hand side will be treated as a functional with arguments: ω L ,
L = 2, . . . , m. The fields ω 2 , . . . , ω m satisfy the orthogonality and
normality conditions (3.15). Minimization of Υ over these fields:
I = min { Υ(ω 2 , . . . , ωm ) | ω L · ω K = δLK } (5.6)
imposes
ω L (x) · τ (x) = 0, L = 2, . . . , m (5.7)
since the terms underlined in (5.5) are non-negative. Thus the choice
1
ω1 = τ, ω L · ω 1 = δL1 (5.8)
||τ ||
is optimal. The optimal compliance is expressed by

1
I = min ||τ ||2 dx τ ∈ Σ(Ω) (5.9)
Ω λ1
Let τ = σ̃ be the minimizer of this problem. The eigenstates are given

by (5.8), where τ = σ̃. The conditions (5.7) determine the vectors ω L =,
L = 2, . . . , m, up to one parameter fixing the position of the vector ω2
with respect to the vector ω1 . Thus the optimal components of the tensor
C are given by (3.29); they are determined up to one parameter.
6 The FMD problem with the isoperimetric condition

expressed in terms of the Kelvin moduli. The case of a
single load
In its original formulation put forward in (Bendsøe et al., 1994) two FMD
problems have been dealt with, with the isoperimetric conditions of the
form:
||λ||p = Eo , p = 1, 2 (6.1)
where
1
f = f dx (6.2)
|Ω| Ω
is the mean value of a function f defined on Ω ⊂ Rd and
5 m 6 p1
#
||λ||p = |λK |p (6.3)
K=1
d(d + 1)
where λ = (λ1 , . . . , λm ), m = ; the constant E0 is a referential
2
elastic modulus. The condition (6.1) can be interpreted as a condition of
bounding the cost of the material to be used. To arrive at a relatively simple
final result we neglect the conditions: λK ≥ λmin > 0 and λK ≤ λmax . We
admit zero values of the Kelvin moduli λK , K = 2, . . . , m. Thus we
consider the FMD problem with the condition (6.1) or
Yp = min { Jp (τ ) | τ ∈ Σ(Ω)} (6.4)
with

1
Jp (τ ) = min ||τ || dx λ1 > 0,
2
λK 0, K = 2, . . . , m;
Ω λ1

||λ||p dx = Λ (6.5)
Ω
where Λ = |Ω|Eb . Since λ2 , . . . , λm do not enter the minimized functional,

all these design variables should vanish to make the feasible domain for λ1
as big as possible. This remark makes Jp p-independent: Jp = J and

1
J(τ ) = min ||τ ||2 dx λ1 > 0, λ1 dx = Λ (6.6)
Ω λ1 Ω
If ||τ || > 0 in Ω the latter problem can be solved:

2
1
J(τ ) = ||τ ||dx (6.7)
Λ Ω
and the minimizer in (6.6) equals

Λ
λ1 (x) = ||τ (x)|| (6.8)
||τ ||dx
Ω
We note that Yp = Y is p-independent and

2
1
Y = min ||τ ||dx τ ∈ Σ(Ω) (6.9)
Λ Ω
We assume now that the results (6.7), (6.8) can be accepted even if τ (x) = 0
for some x ∈ Ω. Let τ = π be the minimizer of the problem:

Z = min ||τ ||dx τ ∈ Σ(Ω) (6.10)
Ω
Then the optimal λ1 and ω 1 are expressed by

||π(x)|| 1
λ1 (x) = Eo , ω1 = π (6.11)
||π|| ||π||
Other ω K , K = 2, . . . , m are determined by (5.8), up to some free param-
eters. The optimal Hooke tensor has components
||π(x)||
Cijkl (x) = Eo π̂ij (x)π̂kl (x) (6.12)
||π||
π
with π̂ = .
||π||
The optimal stress components are expressed by
||π(x)||
σij (x) = Eo (π̂ · ε)π̂ij (6.13)
||π||
which proves that σ and π are collinear and form common trajectories of
principal stresses. The optimal tensor (6.12) satisfies the symmetry condi-
tions (3.2).
Note that the functional in (6.10) has a convex integrand (of linear
growth) and the field runs over a convex set in L2 (Ω, E2s ). Thus the prob-
lem (6.10) is well posed. The existence question lies outside the present
paper. We mention only that keeping the field τ within Σel (Ω) would be
artificial, since the minimizing sequencies for (6.10) tend to limits which are
not solutions to any well posed problems of linear elasticity.
Some aspects of the problem (6.10) will be discussed in Sec. 8. Now let us
stress that due to linear growth of the integrand of (6.10) there can appear
subdomains in Ω where π = 0. Then λ1 = 0 in these subdomains, which
goes beyond the assumptions in the formulation (6.6). We note, however,
||π||2
that in the domains where π = 0, the quotient is an indeterminate
λ1
expression (0/0). Therefore, the appearance of the domains where π = 0 is
not an argument to reject the formulation (6.10).
The solution to problem (6.10) determines not only the λ1 distribution,
but, which is the most important, determines the domain, where λ1 = 0 or
the domain where the material should be removed. One can alternatively
say that the shape of the optimal body is defined by the effective domain
of ||π||−1 . To detect the holes in Ω we do not need special techniques
of expanding small holes. We should only be equipped with a numerical
method by means of which the problem (6.10) can be solved with high
accuracy, to determine the boundary line of the effective domain of ||π||−1 .
In this manner we arrive at a new, optimized shape of the plate. Let us
name it Ωef f . For x ∈ Ωef f we have ||π(x)|| > 0. Even for x ∈ Ωef f the
condition (3.3) is not satisfied. Indeed, let us compute
||π(x)||
Cijkl (x)ξij ξkl = Eo (π̂(x) · ξ)2 (6.14)
||π||
For ξ ⊥ π̂ this expression vanishes. Although the material does not satisfy
(3.3), its moduli are distributed in such a way that the optimal anisotropic
body can transmit the given load T to the given support Γ2 . Indeed, the
solution of problem (6.10) determines a new domain Ωef f cut out from
Ω in which ||π(x)|| > 0 where the optimum design method applies. This
new plate is capable of transmitting the load T to the support Γ2 . We
substitute (6.13) into (6.10) to arrive at the elasticity problem which governs
the optimal plate behaviour:
a(u, v) = f (v) ∀ v ∈ V (Ωef f ) (6.15)

with

Eo
a(u, v) = ||π(x)|| (π̂ · ε(u)) (π̂ · ε(v)) dx (6.16)
||π|| Ωef f
where f (·), defined by the r.h.s. of (2.3), can also be interpreted as

f (v) = π · ε(v)dx (6.17)
Ωef f
since π truncated to Ωef f belongs to Σ(Ωef f ). We note that a(·, ·) is

symmetric and non-negative, while a(v, v) = 0 only if ε(v) ⊥ π. Let us
define
Vo = v ∈ (H 1 (Ωef f ))d a(v, v) = 0 (6.18)
The problem (6.15) is solvable if
f (v) = 0 ∀ v ∈ Vo (6.19)
We note that this condition holds since f (v) = 0 means that π ⊥ ε(v) and
just such v constitute the set V0 .
Assume that u and u solve (6.15). Then a(u − u , v) = 0 for v ∈
V (Ωef f ). We take v = u − u as kinematically admissible to arrive at the
condition u − u ∈ V0 . The solution u is determined up to terms from V0
which vanish on Γ2 . Let us look more closely on the fields in V0 . If v ∈ V0
then ε(v) · π = 0 or
πij vi,j = 0 (6.20)
Noting that div π = 0 we rearrange (6.20) to the form
(πij vi ),j = 0 (6.21)
or div(πv) = 0. Assume d = 3; the divergence free functions are represented

by
πv = rot ψ (6.22)
and ψ = (ψ1 , ψ2 , ψ3 ), ψi are arbitrary, differentiable functions. The set V0
is composed of functions v such that πv = rot ψ and v vanishes on Γ2 .
Thus the problem (6.15) does not determine the field u in a unique
manner. But the stress field (6.13) is unique, the same for each u + v,
v ∈ V0 which solves (6.15); indeed π̂ · (ε(u + v)) = π̂ · ε(u) if v ∈ V0 , see
(6.13).
It is worth noting that this is a feature of many optimal structures,
among them the funicular structures are worth mentioned. The funiculars
are geometrically variable but the forces in bars are uniquely determined.
The criterion of the correctness of the statics solution concerning the geo-
metrically variable structures is that the virtual work of the loading on all
zero-energy modes (rigid body modes, zero strain modes) vanishes. This
condition here assumed the form (6.19).
Let us come back to the problem (6.10), of crucial importance for the
FMD. Let us disclose the condition (2.4) by writing:

Z= min
2 2
max ||τ ||dx + f (v) − τ · ε(v)dx (6.23)
τ ∈L (Ω, Es ) v∈V (Ω) Ω Ω
The operations max and min can be switched (by arguments similar to those
in Strang and Kohn, 1983):
Z = max {f (v) + R(v)} (6.24)

v∈V (Ω)
with

R(v) = min 2 2
[||τ || − τ · ε(v)] dx τ ∈ L (Ω, Es ) (6.25)
Ω
One can prove that

0 if ε(v(x)) ∈ B a.e. in Ω
R(v) = (6.26)
−∞ otherwise
where
B = ε ∈ E2s | ||ε|| 1 (6.27)
Thus we find
Z = max { f (v) | v ∈ V (Ω), ε(v(x)) ∈ B, for a.e. x ∈ Ω} (6.28)
The formulation above can be interpreted as embedding of the body Ω into

Rd . In this infinite domain one should form a displacement field v vanishing
on Γ2 such that at each point x : ||ε(v(x))|| ≤ 1. Among such fields v we
look for the one which maximizes the virtual work f (v).
If Z given by (6.10) and by (6.28) coincide, we are sure that both the
problems have been correctly solved. Note that the solution to (6.28) deliv-
ers the trajectories of the virtual strain which coincide with the trajectories
of the stress minimizing (6.10).
Remark 6.1
The problems (6.10) and (6.28) are similar to Michell’s problem formula-
tions, discussed in the article by Lewiński and Sokól (2013) in this book.
Figure 1. A rectangular plate subject to a self-equilibrated load
Due to this similarity the optimal Michell layouts and the FMD layouts
have much in common.
A simple benchmark
Consider a plane domain shown in Fig. 1, of the sides (2c + b) by a, divided
b
into the subdomains Ω1 , Ω2 , Ω3 by the vertical lines x1 = ± . The load
2
is selfequilibrated and no kinematic conditions are imposed.
We start from considering (6.28). We take the field v of components
v1 = 0, v2 = x2 .
We note that ||ε(v)|| = 1 in the whole domain Ω. Let us compute
a a
f (v) = qbv2 x1 , − qbv2 x1 , −
2 2
f (v) = qba.
Let us consider now the problem (6.10). We take τ = 0 in Ω1 , Ω3 and
τ11 = 0, τ12 = 0, τ22 = q in Ω2 .
We compute ||τ || = q in Ω2 and then

||τ ||dx = abq
Ω
which concides with f (v). This proves that the duality gap is zero and
both the problems (6.10) and (6.28) are solved exactly; τ = π. The Kelvin
modulus λ1 is
λ1 = 0 in Ω1 , Ω3 ,
2c + b
λ1 = E0 in Ω2
b
By (5.8)
(ω 1 )11 = 0, (ω 1 )22 = 1, (ω 1 )12 = 0.
The only nonzero modulus Cijkl is C2222 = λ1 . The optimal material
in Ω2 can be viewed as fibrous along x2 of no lateral and shear stiffnesses.
The domains Ω1 , Ω3 appear to be empty.
7 A direct numerical approach to the FMD problem

with the trace constraint. A single load case
The aim of this section is to show the layouts of the optimal Kelvin modulus
λ1 and optimal optimal elastic moduli Cijkl for single loading condition. The
examples concern a rectangular plate of length Lx = 2.0 [m] and height Ly =
1.0 [m]. The finite element mesh is defined by nx × ny = 60 × 30 = 1800 4-
node, quadrilateral, isoparametric finite elements C2D4 with bilinear shape
functions interpolating the four stress fields τij = τij (x) (i, j = 1, 2) and
two displacement (kinematically admissible) test fields vi = vi (x) (i = 1, 2),
x ∈ Ω. The total number of nodes N = (nx + 1) × (ny + 1) = 1891 gives
the total number 3N = 5673 of the unknown nodal, stress parameters and
the total number 2N = 3782 of degrees of freedom equal to the number of
columns and rows in the equilibrium matrix B, respectively. The rectan-
gular matrix B = B 2N ×3N , together with the 2N dimensional (and only
partially known) vector of nodal forces Θ = Θ2N fulfilling the equilibrium
conditions defined by a rectangular system of linear algebraic equations
BΞ = Θ, where Ξ is the 3N dimensional vector Ξ = Ξ3N of the un-
known nodal stress parameters, can be easily derived immediately from the
variational equilibrium equation (2.3) after the implementation of the finite
element interpolation of the stress τ = τ (x) and kinematically admissi-
ble displacement field v = v(x) (described shortly above). In the case of
the non-self-equilibrating external boundary loading (tractions) of intensity
T = T (s), s ∈ Γ1 , the kinematic boundary conditions should be addition-
ally introduced, by proper changing the order of the rows in the matrix B
and in the vector Θ, respectively. The complete description of the com-
putational algorithm of finding the approximation Σ̂(Ω) of the statically
admissible set Σ(Ω) of stress fields τ together with the presentation of the
numerical method of finding the minimizer τ = π ∈ Σ̂(Ω) of the prob-
lem (6.10) is thoroughly discussed in the paper Czarnecki, Lewiński (2012).
In accordance with the formulae (6.11), (6.12) the distribution of optimal
λ1 (x), ω 1 (x) and optimal moduli Cijkl (x) can be found on the ground of
optimal field π.
The plate rests on two non-sliding supports on the lower edge (left and
right node) and is subject to a vertical or/and horizontal load at the lower
or/and upper edge, see Fig. 2. The load traction T = T (s) = Ti (x)ei ,
s ∈ Γ1 is modeled by the weight function
s−s0 2
Ti = Ti (s)Tmax e−( w ) (i = 1, 2)
where
Lx Lx 2Lx
s0 = , ,
2 3 3
depending on the load position.
The values Tmax , w are assumed to be equal Tmax ≈ 3.761, w = 0.15
(emulation of the unit forces). We have assumed that E0 = 1.0 [N/m2 ] –
see (6.1).
Figure 2. The body Ω – rectangular plate Lx × Ly and four variants a, b,

c, d of loading T
λ1 (π)
The distribution of optimal for four variants of loading T are
E0
shown in Fig. 3.
λ1 (π)
Figure 3. The distribution of optimal (scatter, contours and height
E0
field plots in TieDie color map in the first, second and third column, re-
spectively) for the various variants of loading T (a, b, c and d variant in the
first, second, third and fourth row, respectively)
8 The stress function method

8.1 The plane problem
We apply Airy’s method (2.4)–(2.6) in (6.10). Let us express the norm
of τ in terms of the Airy function F
7
||τ || = (F,11 )2 + (F,22 )2 + (F,12 )2 + (F,21 )2 (8.1)
or
||τ || = ||∇2 F || (8.2)
2
where ∇ F = [F,ij ] i, j = 1, 2.
Thus the right hand side of (8.2) is the Frobenius norm of the matrix
∇2 F . The condition τ ∈ Σ(Ω) will be replaced by (2.7). Hence

Z = min ||∇2 F ||dx F ∈ S(Ω), ik3 jl3 F,kl nj = Ti on Γ1 (8.3)
Ω
where S(Ω) is the space of sufficiently smooth scalar functions in Ω. The

extremals of the functional

F → ||∇2 F ||dx (8.4)
Ω
satisfy Euler’s equation:

1
F,ij =0 (8.5)
||∇2 F || ,ij
or
∇2 F
div div =0 (8.6)
||∇2 F ||
The equation above can be written in the form:
Δ2 F + nonlinear terms = 0 (8.7)
which proves its ellipticity. We note that F is not biharmonic (in general),
which discloses a discrepancy between the images of stresses in the isotropic
body in 2D and the images of the stresses which solve (6.10).
Remark 8.1
The problem (6.10) and its reformulation (8.3) resembles the minimal sur-
face problem in which the minimized functional has the form
+
f→ 1 + ||∇f ||2 dx (8.8)
Ω
where
||∇f ||2 = (f,1 )2 + (f,2 )2 (8.9)
The Euler equation for (8.8) reads

∇f
div =0 (8.10)
||∇f ||
The minimal surface problem has become an important part of the calculus
of variations, see (Nitsche, 1989). Note that the equation (8.10) is elliptic,
since it is equivalent to
Δf + (f,2 )2 f,11 + (f,1 )2 f,22 − 2f,1 f,2 f,12 = 0 (8.11)
and the first term proves ellipticity of the equation.
8.2 The spatial problem

We use the Finzi representation (2.8) in the problem (6.10). Let us
compute the norm:
||τ || = M (∇2 φ) (8.12)
or +
M (∇2 φ) = imr jns φrs,mn ikl jpq φlq,kp (8.13)
where φ ∈ E2s ,
φ = (φij ).
The unknowns are fields φij ∈ S(Ω), φij = φji , minimizing the functional

M (∇2 φ)dx (8.14)
Ω
with the condition (2.8) on Γ1 .

The stationarity conditions of (8.14) are of the form

∂M (∇2 φ)
=0 (8.15)
∂φij,mn ,mn
Thus in the spatial case discussed the number of stationarity conditions is

six, while in 2D only one condition is to be satisfied.
9 The plane FMD problem under the conditions of

fixed values of the Kelvin moduli. Vector optimization
for the two-load condition
The plate of middle plane Ω is viewed as being loaded either by
1) T (1) given on Γ1 or 2) T (2) given on Γ1
The loads cannot act simultaneously. Let the load T (α) causes the stress
field τ (α) , α = 1, 2. The equilibrium conditions read
(α)
τ α · ε(v)dx = f (v) ∀ v ∈ V (Ω) (9.1)
Ω
with
(α)
f (v) = T α ·v ds (9.2)
Γ1
The field τ (α) satisfying the above equation is statically admissible; the set
of such stress fields is denoted by Σα (Ω) Let C be tensor of the reduced
moduli, cf. Sec. 3. The compliance corresponding to the load indexed by α
is given by (4.5) or

Υ (α)
= min τ· C −1
τ dx τ ∈ Σα (Ω) (9.3)
Ω
If the field u(α) is the displacement field referring to the load of number α
then the compliance is denoted by Υ(α) = f (α) (u(α) ).
Tensor C −1 is decomposed as in (3.16). We assume that the moduli
are ordered: λ1 ≥ . . . ≥ λ3 ≥ λ0 > 0 and fixed. The design variables are
the eigenstates ω 1 , ω 2 , ω 3 satisfying the orthonormality conditions (3.15).
Since ω 1 is determined by ω2 and ω 3 the unknowns will be only two last
tensors. The set of pairs (ω 2 , ω 3 ) defined in Ω, satisfying (3.15) is denoted
by Q(Ω). The merit function is
Fη (ω 2 , ω 3 ) = ηΥ(1) (ω 2 , ω3 ) + (1 − η)Υ(2) (ω 2 , ω3 ) (9.4)
where, according to (5.5)

α
Υ (ω 2 , ω3 ) = min k(τ , ω2 , ω 3 )dx τ ∈ Σα (Ω), α = 1, 2 (9.5)

Ω
and
1
k(τ , ω 2 , ω 3 ) = ||τ ||2 + ν2 (ω 2 · τ )2 + ν3 (ω 3 · τ )2 (9.6)
λ1
the constants ν2 , ν3 are given by (5.4). The aim is to solve the family of
problems
Iη = min { Fη (ω 2 , ω 3 ) | (ω 2 , ω3 ) ∈ Q(Ω)} (9.7)
indexed by η ∈ [0, 1]. Combining (9.4)–(9.7) we arrive at

Iη = min Wη τ 1 (x), τ 2 (x) dx τ 1 ∈ Σ1 (Ω), τ 2 ∈ Σ2 (Ω) (9.8)
Ω
where
Wη (σ , σ ) = min { ηk(σ , a, b) + (1 − η)k(σ , a, b) | (a, b) ∈ Q}

(9.9)
and σ , σ ∈ E2s ; Q is the set of pairs (a, b) such that a · a = 1, b · b = 1,
a · b = 0.
The passage from (9.7) to (9.8) has been done by replacing minimization
over the fields with pointwise minimization over the numbers. The theoret-
ical background of such passage is the Theorem 3A in (Rockafellar, 1976).
The local minimization problem (9.9) has been for the first time solved in
(Czarnecki and Lewiński, 2011). It has the form
√ +
Wη (σ , σ ) = Wλ ( ησ , 1 − ησ ) (9.10)
where

1 1 1
Wλ (σ, τ ) = + ||σ||2 + ||τ ||2 −

2 λ2 λ1
7
1 1 1
− − (||σ||2 − ||τ ||2 )2 + 4(σ · τ )2 (9.11)
2 λ2 λ1
An alternative, much shorter derivation is given in this article in Sec. 12. It

leads to the equivalent result
1 1
Wλ (σ, τ ) = μ1 (g(σ, τ )) + μ2 (g(σ, τ )) (9.12)
λ1 λ2
where
μ1 (g(σ, τ )) ≥ μ2 (g(σ, τ ))
are eigenvalues of the matrix
" $
σ·σ σ·τ
g(σ, τ ) = (9.13)
τ ·σ τ ·τ
Let us stress here, this is not a Gram matrix, since it is defined by two
vectors, and the vectors belong to R3 . The Gram matrix will appear in the
case of three independent loads, see Sec. 11. Note that λ3 does not enter
(9.11). The analytical solution of (9.9) determines the optimal eigenstates
ω 2 , ω3 depending on the stress fields τ 1 , τ 2 which solve (9.8). Assume now
that for the fixed η the problem (9.8) is solved and the fields τ 1 , τ 2 are
known. We introduce
√ +
σ = ητ 1 , τ = 1 − ητ 2 (9.14)
Let at a given point x ∈ Ω, ||τ 1 || = 0, ||τ 2 || = 0 and define

2
||τ ||
ζ(σ, τ ) = (9.15)
||σ||
( π)
Let α ∈ 0, and
2
|σ · τ |
cos α = (9.16)
||σ|| ||τ ||
We compute

ζ sin 2α 1
ϕ̂ = − arctan ϕ∗ = (π − ϕ̂) (9.17)
1 + ζ cos 2α 2
Tensors ω1 , ω 2 are given by the rules

cos ϕ∗ sign(σ · τ ) cos(ϕ∗ − α) 1
ω1 = − τ+ σ
sin α ||τ || sin α ||σ||
(9.18)
sin ϕ∗ sign(σ · τ ) sin(α − ϕ∗ ) 1
ω2 = · τ+ σ
sin α ||τ || sin α ||σ||
The tensors ω 1 , ω 2 ∈ E2s wil be treated as vectors decomposable in the

basis B K , see (3.19), (3.20). Such interpretation makes it possible to treat
ω 1 , ω2 as vectors in R3 . We define ω 3 = ω1 × ω 2 by the rules in R3 and
come back to the components ω 3ij in the basis ei ⊗ ej by (3.20). Having
the components ωKij , K = 1, 2, 3 we determine the components of the
optimal tensor C by (3.13), (3.14) or
3
#
Cijkl = λK ωKij ωKkl (9.19)
K=1
If one of tensors τ 1 or τ 2 is zero, the components Cijkl are not uniquely

determined, see (116) in (Czarnecki, Lewiński, 2011).
Concluding, the optimal design goes as follows. First, the problem (9.8)
must be solved, using the explicit expressions (9.10), (9.11). Having the
fields τ 1 , τ 2 in Ω we follow the rules (9.14)–(9.19) and construct the tensor
C at each point x of Ω. The potential (9.11) is not convex, see (Czar-
necki, Lewiński, 2011). The correctness of the problem (9.8) follows from
the properties of its dual form. The problem (9.8) can be rearranged to
the formulation (P ) in which the displacement fields u(α) are the main un-
knowns:
find u(α) ∈ V (Ω) such that the equilibrium equations (9.1) are satisfied by
the stress fields τ (α) = σ α which are linked with strains εα = ε(u(α) ) by
the constitutive equations of the form
& 1 '
1 λ1 + λ2 ε1 λ1 − λ2 ε ε2
σ1 = √ √ + L √ , √
η 4 η 4 η 1−η
(9.20)
& 2
2 1
'
1 λ1 + λ2 ε λ1 − λ 2 ε ε
σ2 = √ √ + L √ , √
1−η 4 1−η 4 1−η η
with
||ε||2 − ||κ||2 2(ε · κ)
L(ε, κ) = ε+ κ (9.21)
G(ε, κ) G(ε, κ)
and 7
2
G(ε, κ) = (||ε||2 − ||κ||2 ) + 4(ε · κ)2 (9.22)
One can prove that the strain potential Wη∗ (ε1 , ε2 ) exists such that
α ∂Wη∗ (ε1 , ε2 )
σ ij = i, j = 1, 2 (9.23)
∂εαij
This potential assumes the form

ε1 ε2
Wη∗ (ε1 , 2
ε )= Wλ∗ √ , √ (9.24)
η 1−η
with
λ1 + λ2
λ1 − λ2
Wλ∗ (ε, κ) = ||ε||2 + ||κ||2 + G(ε, κ) (9.25)
8 8
The potential is
a) a homogeneous function of degree 2, or
Wλ∗ (αε, ακ) = α2 Wλ∗ (ε, κ) (9.26)
b) strictly convex, while the constitutive equations (9.23) are strictly
monotone
c) satisfies Euler’s equation
2Wη∗ (ε1 , ε2 ) = σ 1 · ε1 + σ 2 · ε2 (9.27)

The (a), (b) properties imply correctness of the (P ) formulation: the
fields u(α) are uniquely determined. The strict convexity property of Wλ∗
has been proved in (Dzierżanowski and Lewiński, 2012b). The strict mono-
tonicity follows from strict convexity, see Minty (1964), which means that
for each pairs (ε1 , ε2 ), (η 1 , η 2 )
1 1 2
σ (ε , ε ) − σ 1 (η 1 , η 2 ) · (ε1 − η 1 ) +

+ σ 2 (ε1 , ε2 ) − σ 2 (η 1 , η 2 ) · (ε2 − η 2 ) 0 (9.28)
and the equality holds if and only if ε1 = η 1 , ε2 = η 2 . The potentials

Wλ (σ, τ ), see (9.11) and Wλ∗ (ε, κ) see (9.25) are interrelated by the duality
relation

Wλ∗ (ε, κ) = max σ · ε + τ · κ − Wλ (σ, τ ) | σ, τ ∈ E2s (9.29)
which implies convexity of Wλ∗ . Note that Wλ∗ (ε, κ) can be expressed by
the formula similar to (9.12) as
λ1 λ2
Wλ∗ (ε, κ) = μ1 (g(ε, κ)) + μ2 (g(ε, κ)) (9.30)
4 4
where g(ε, κ) is the matrix of the form (9.13) of arguments ε and κ while
μα are its eigenvalues.
10 The plane FMD problem with the trace

constraint. Two load case
10.1 Problem setting
In the conventional FMD setting the functional (9.4) is minimized over
all the parameters which determine the Hooke tensor field in Ω. The design
variables are
λ1 ≥ λ2 ≥ λ3 ≥ λ03 > 0, ω 1 , ω2 , ω 3 ∈ E2s
and additionally a resource condition is imposed, written for instance in the
form
||λ||p dx = Λ (10.1)
Ω
where || · ||p is the Hölder norm of the vector λ(λ1 , λ2 , λ3 ).
In this article we confine attention to the case of p = 1, or to the isoperi-
metric condition of the form

(λ2 + λ2 + λ3 )dx = Λ (10.2)
Ω
Note that
||λ|| = |λ1 | + |λ2 | + |λ3 | and λK ≥ λ03 > 0.
The subject of consideration is minimization of the functional (9.7) over the
Kelvin moduli

Jη = min Iη (λ) | λK ∈ L(Ω, R+ ), λK λ03 > 0, ||λ||1 dx = Λ

Ω
(10.3)
where K = 1, 2, 3; L(Ω, R+ ) is the space of variation of Kelvin moduli;
the regularity condtions will note be written explicitly. Note that λ3 does
not enter Iη (λ) see Sec. 9. Thus in the process of minimization over λ1 ,
λ2 one should assume that λ3 takes the smallest value or λ3 = λ03 . This
assumption makes further discussion complicated. Thus it is reasonable to
admit λ03 = 0 to make the final result as clear as possible.
Thus in problem (10.3) two design variables λ1 , λ2 are present. We
prepare now the procedure of minimization over these design variables. We
introduce the auxiliary variable
ρ(x) = λ1 (x) + λ2 (x), ρ ∈ L(Ω, R+ ) (10.4)
and write (10.3) in the form

Jη = min Sη (ρ) | ρ ∈ L(Ω, R+ ), ρdx = Λ (10.5)

Ω
where

Sη (ρ) = min Tη (τ 1 (x), τ 2 (x), ρ(x))dx τ α ∈ Σα (Ω), α = 1, 2
Ω
(10.6)
and Tη is given as the solution to the local minimization problem

Tη (τ 1 , τ 2 , ρ) = min Wη (τ 1 , τ 2 ) λ1 + λ2 = ρ, λ1 λ2 , λ1 , λ2 ∈ R+
(10.7)
We make use now of the result (9.12): we express Wη explicitly by λ1 , λ2
1 1
Wη (τ 1 , τ 2 ) = a1 + a2 , (10.8)
λ1 λ2
where √ +
aα = μα g( ητ 1 , 1 − ητ 2 ) (10.9)
and g(·, ·) is the matrix (9.13).
One should solve the problem for given a1 > 0, a2 > 0:

a1 a2
T (a1 , a2 , ρ) = min + λ + λ = ρ λ λ > 0 (10.10)
λ2
1 2 1 2
λ1
The solution of the above problem reads:
√
aα
λ∗α = ρ √ √ (10.11)
a1 + a2
1 √ √ 2
T (a1 , a2 , ρ) = ( a1 + a2 ) (10.12)
ρ
We insert now (10.12) into (10.6) and write (10.5) in the form:
&7
1 √ +
Jη = α min min μ1 (g( ητ 1 , 1 − ητ 2 )+
τ ∈Σα (Ω) Ω ρ(x)
7 '2
√ 1 +

+ μ2 (g( ητ , 1 − ητ 2 ) dx ρ ∈ L(Ω, R+ ),

ρdx = Λ (10.13)
Ω
The nested minimization problem in (10.13) can be solved analytically. The

final result is
1
Jη = (Zη )2 (10.14)
Λ

√ 1 + α
Zη = min ητ (x), 2
1 − ητ (x) 1 dx τ ∈ Σα (Ω)
Ω 2
(10.15)
where
+ +
|||(σ, τ )||| 1 = μ1 (g(σ, τ )) + μ2 (g(σ, τ )) (10.16a)
2
or
7 +
|||(σ, τ )||| 1 = ||σ||2 + ||τ ||2 + 2 ||σ||2 ||τ ||2 − (σ · τ )2 (10.16b)
2
The minimizer ρ = ρ∗ of the nested problem in (10.13) has the form

√ +

ητ 1 (x), 1 − ητ 2 (x) 1
ρ∗ (x) = Λ +
2
(10.17)
√ 1 2
ητ (x), 1 − ητ (x) 1 dx
Ω 2
Thus the main problem is to solve (10.15). Having the minimizers τ ∗1 ,

τ ∗2 we compute: ρ∗ by (10.17), λ∗1 , λ∗2 by (10.11), (10.9) and λ∗3 = 0; the
eigenstates ω∗K are given by (9.14)–(9.18). One can prove that ||| · ||| 12 is
a norm in R6 Thus the functional in (10.15) has a convex (but not strictly
convex) integrand of linear growth.
The problem (6.10) had its dual counterpart (6.28) and similarly the
problem (10.15) can be transformed to its dual form; it reads

√ 1 1 +
Zη = max ηT · v + 1 − ηT · v ds v α ∈ V (Ω),
2 2
Γ1

1 2

ε(v ), ε(v ) ∈ B− 12 (10.18)
where

B− 12 = (ε, κ) ∈ E2s × E2s |||(ε, κ)|||− 1 1 (10.19)
2
and
σ·ε+τ ·κ
|||(ε, κ)|||− 1 = max 2 (10.20)
2 σ, τ ∈Es |||(σ, τ )||| 1
2
The geometry of the ball B− 12 is the subject of the current research.

Note that both the minimizers τ ∗1 , τ ∗2 of the problem (10.15) can be

zero in a certain subdomain of Ω. Thus according to (10.11) the moduli
λ∗α will be zero and the material is there not necessary. In this manner
the algorithm removes a part of the material from the domain reducing the
domain to Ωef f The solution to problem (10.18) has also this property that
removes the material from the domains in which ε(v ∗1 ) = 0, ε(v ∗2 ) = 0.
10.2 Numerical treatment of (10.15)

The aim of this section is to show the layouts of the optimal Kelvin
moduli λ1 , λ2 and optimal optimal elastic moduli Cijkl for two loading
conditions. The examples concern the same rectangular plate as in Sec. 7.
The two variants of two load cases are shown in Fig. 4. The computational
algorithm of finding the both approximations Σ̂α (Ω) (α = 1, 2) of the
statically admissible sets Σα (Ω) of stress fields τ α (α = 1, 2) and the
numerical method of finding the minimizers τ α = τ ∗α ∈ Σ̂α (Ω) (α = 1, 2)
of the problem (10.15) for fixed but arbitrary parameter η ∈ [0, 1] are
similar to the procedures applied in (Czarnecki and Lewiński 2012). In
accordance with the formulae (10.17), (10.9), (10.11), (10.12), (9.14)–(9.18),
the distribution of Pareto optimal fields λ∗1 (x), λ∗2 (x), ω ∗K (x) and Pareto
Figure 4. The body Ω and two variants a and b of two-case loadings T α

for α = 1 (first column) and α = 2 (second column)
∗
optimal moduli Cijkl (x) can be found on the ground of Pareto optimal
∗α
stress fields τ (α = 1, 2) see Figs 5, 6, 7 and 8.
Figure 5. The first variant a of the two-case loadings T α (α = 1, 2).

The distribution of the Pareto optimal λ∗1 , λ∗2 (λ∗1 ≥ λ∗2 ) (contours plots
in TieDie color map) in the first and second row, respectively) for η = 0.9
(first column) and η = 0.1 (second column)
Figure 6. The second variant b of the two-case loadings T α (α = 1, 2).

The distribution of the Pareto optimal λ∗1 , λ∗2 (λ∗1 ≥ λ∗2 ) (contours plots in
TieDie color map) for η = 0.9 (first column) and η = 0.1 (second column)
Figure 7. The first variant a of the two-case loadings T α (α = 1, 2). The

∗
distribution of the Pareto optimal moduli Cijkl for η = 0.5 (scatter plots in
TieDie color map)
Figure 8. The second variant b of the two-case loadings T α (α = 1, 2).

∗
The distribution of the Pareto optimal moduli Cijkl for η = 0.5 (scatter
plots in TieDie color map)
11 The plane FMD problem with the trace condition.

Three load case
11.1 The Kelvin moduli being fixed
Let us assume that λK = const and the fields ω 2 , ω 3 are design variables.
We come back to Sec. 9 and assume three kinds of loads: T (K) , K = 1, 2, 3.
The merit function is of the form

#
n
Fη (ω 1 , ω 2 , ω3 ) = ηi Υ(i) (ω 1 , ω 2 , ω3 ) (11.1)
i=1
and η1 + . . . + ηn = 1; ηi > 0 are fixed; here n = 3. The tensors ω K are

linked by (3.15). The minimization of (11.1) over ω 1 , ω 2 , ω 3 is reduced to
the form

1

Iη = min Wη τ (x), τ (x), τ (x) dx τ K ∈ ΣK (Ω)
2 3
(11.2)
Ω
where

#3 3
1 # √
1 2 3
Wη τ , τ , τ = min ( η L τ L · ω K )2 ω K ∈ R3 ,
λK
K=1 L=1
8
ωK · ω L = δKL (11.3)
We shall prove that the solution to the above problem has the form

√ √ √
Wη τ 1 , τ 2 , τ 3 = Wλ η1 τ 1 , η 2 τ 2 , η 3 τ 3 (11.4)
3

# 1
Wλ σ 1 , σ 2 , σ 3 = μK g σ 1 , σ 2 , σ 3 (11.5)
λK
K=1
where μ1 ≥ μ2 ≥ μ3 are the eigenvalues of the Gram matrix:

g σ 1 , σ 2 , σ 3 = σ i · σ j i, j=1, 2, 3 (11.6)
Thus μK > 0, K = 1, 2, 3 and the equality holds if σ K are linearly

dependent. Let us prove (11.5). Introduce the matrices 3 × 3

S = σ1 , σ2, σ3 , Q = [ω 1 , ω 2 , ω 3 ] (11.7)
In the matrices S and Q the tensors σ K , ω K are interpreted (see Re-

mark 3.1), as column vectors of dimensions 3 by 1. Note that Q is an
orthogonal matrix. Introduce the diagonal matrix:

− 21 1 1 1
Λ = diag √ , √ , √ (11.8)
λ1 λ2 λ3
Let 1
A = S T QΛ− 2 , Ŝ = SS T (11.9)
Note that
3
# 3
1 #
(ω K · σ L )2 = tr (AT A) (11.10)
λK
K=1 L=1
√
where σ L = ηL τ L and
3
#
tr (AT A) = xK · (ŜxK ) (11.11)
K=1
√
where Ŝ is expressed by σ L = ηL τ L and xK is the Kth column of the
1
matrix QΛ− 2 . Let eK be the orthonormal basis in R3 . Then
1
xK = √ QeK (11.12a)
λK
or
1
xK = √ ω K (11.12b)
λK
Note that g given by (11.6) is equal to S T S. We know that the eigenvalues
of S T S and SS T are identical. Thus the quantities μK are eigenvalues of
of the matrix Ŝ; we sort them as previously: μ1 ≥ μ2 ≥ μ3 .
The problem (11.3) can now be written as below
3
1 2 3
#

Wη τ , τ , τ = min xK · (ŜxK ) xK · xL = δKL ,

K=1
8
1
for K = L and ||xK || = √ (11.13)
λK
In the first step we minimize the term in which the norm ||xK || is the
biggest, or for K = 3. We compute

1
J3 = min x3 · (Ŝx3 ) ||x3 || = √ = x∗3 · (Ŝx∗3 ) (11.14)
λ3
where x∗3 is the eigenvector corresponding to the smallest eigenvalues of Ŝ:
Ŝx∗3 = μ3 x∗3 (11.15)
Thus
μ3
J3 = μ3 ||x∗3 ||2 = (11.16)
λ3
Then we minimize the second term:

1
J2 = min x2 · (Ŝx2 ) x2 ⊥ x∗3 , ||x2 || = √ = x∗2 · (Ŝx∗2 ) (11.17)
λ2
where x∗2 is the eigenvector corresponding to the eigenvalue μ2
Sx∗2 = μ2 x∗2 , x∗2 ⊥ x∗3 (11.18)
or
μ2
J2 = μ2 ||x∗2 ||2 = (11.19)
λ2
The last term corresponds to the vector x1 which is orthogonal to both the
vectors x∗2 and x∗3 , or this is the eigenvector corresponding to μ1 : Ŝx∗1 =
μ1 x∗1 ; we compute
μ1
J1 = x1 · (Ŝx1 ) = (11.20)
λ1
The sum J1 + J2 + J3 gives the result (11.5). In the proof we have tacitly
made use of the r earrangement inequality, which is used in the form (cf.
Hardy et al.,1999):
#
n
aK # n
aσ (K )
(11.21)
bK bK
K=1 K=1
where σ(K) is the permutation of indices {1, 2, . . . , n} while the numbers

aK and bK are sorted as below
1 1 1
a1 a2 . . . a n ; ... (11.22)
b1 b2 bn
In the 2D elasticity problem discussed here n = 3, aK = μK , bK = λK . The
inequality (11.21) clears up why in the result (11.5) the indices at μ and λ
are identical. Because of an exceptional importance of the result (11.5) we
deliver below an alternative explanation of the result.
Let us note that the minimizers x∗K in (11.3) had to be the eigenvectors
of Ŝ, since they are extremal points of the lagrangian for (11.13):
3
# 3
#
L= xK · (ŜxK ) + lK (λK − xK · xK ) (11.23)
K=1 K=1
where the orthogonality conditions have been omitted. Assume that xK +

δxK satisfy the orthogonality conditions. The variation with respect to xK
gives
δL = ŜxK − lK xK · δxK = 0 (11.24)
which confirms that the extremal points of L include the eigenproblems

for the matrix Ŝ. Thus the result of minimization in (11.13) must have
the form equal to the l.h.s. of (11.21); the result of minimization over the
permutation of indices confirms (11.5). Thus the problem (11.2) involves
the integrand given explicitly in terms of the stress fields τ 1 , τ 2 , τ 3 , see
(11.4), (11.5). This result extends the result (9.12), where the modulus λ3
was absent.
Assume that τ ∗K are minimizers of (11.2). Having them we determine
√
the fields σ ∗K = ηK τ ∗K and compute the matrix S and then compute the
eigenvalues μ∗K of the matrix Ŝ.
The process of the proof of (11.5) delivers the algorithm of computing
the eigenstates ω ∗K which make (11.3) minimal. According to (11.12b) the
vector ω ∗K is colinear with xK , hence the vector ω∗K is the eigenvector of Ŝ
of unit length, corresponding to the eigenvalue μK . Alternatively, ω∗K is the
T √
eigenvector of the Gram matrix Ŝ = g(σ ∗1 , σ ∗2 , σ ∗3 ), σ ∗K = ηK τ ∗K .
The Hooke tensor is determined by (9.19).
A direct expression of (11.5) by σ K · σ L is not easy, since the Cardan
equations are complex. Hence it is not easy to find the counterpart of (9.11)
concerning 3 loads. By analogy, we do not expect that the potential (11.5) is
convex. The displacement based version of the problem (11.2) necessitates
inversion of the equations

∂Wλ σ 1 , σ 2 , σ 3
εK = (11.25)
∂σ K
or performing the maximum operation:

Wλ∗ ε1 , ε2 , ε3 =

(11.26)
max σ 1 · ε1 + σ 2 · ε2 + σ 3 · ε3 − Wλ σ 1 , σ 2 , σ 3 σ K ∈ E2s
yet both of these operations seem difficult. The formulation (11.2) is im-
portant as a starting point for minimization over the moduli λK , which is
the subject of the subsequent section.
11.2 Releasing the Kelvin moduli

The subject of the research is the problem (10.3) where Iη is given as
in Sec. 11.1. Proceeding as in Sec. 10.1 we arrive at the equation (10.14)
where
3 8
#7 √ √ √
K
Zη = min μK g η 1 τ 1 , η 2 τ 2 , η 3 τ 3 dx τ ∈ ΣK (Ω)
Ω K=1
(11.27)
Let us remind the definition

7 of the K-th singular value of an arbitrary
matrix A: sK (A) = μK (AAT ) where μK is the eigenvalue; the matrix
AAT has always non-negative eigenvalues, hence the definition of sK has
sense. In particular, if A is a vector n by 1, then A has only one non-negative
singular value equal to its length: ||A||. If A is a square symmetric and
positive definite matrix, then its singular values are equal to its eigenvalues.
Let
√ √ √
S(x) = η 1 τ 1 (x), η 2 τ 2 (x), η 3 τ 3 (x) (11.28)
see (11.7). The problem (11.27) can be written as follows

8
#
m
K
Zη = min sK (S(x))dx τ ∈ ΣK (Ω), K = 1, . . . , m (11.29)
Ω K=1
where m=3. Upon solving this problem one can find the layouts of the
Kelvin moduli:
sK (S(x))
λ∗K = Λ 3 (11.30)
#
sK (S(x))dx
Ω K=1
Therefore λ∗1 ≥ λ∗2 ≥ λ∗3 .

Let (μK (x), ω ∗K (x)), ||ω ∗K || = 1 is the K-th solution to the eigenvalue
problem of the matrix
√ √ √
g η1 τ 1 (x), η2 τ 2 (x), η3 τ 3 (x)
or of the matrix Ŝ. The optimal Hooke tensor has the form (9.19), where
λ∗K are given by (11.30) and ω ∗K are determined as above. In contrast to
the case of two loads (Sec. 10.1) the modulus λ∗K is in general non-zero. It is
however possible, that in some subdomains the matrix S vanishes and there
the material can be removed. The function of three arguments from E2s .
3
1
#
σ , σ 2 , σ 3 = sK σ 1 , σ 2 , σ 3 (11.31)

K=1
is homogeneous of degree 1. We conjecture that this function has properties

of a norm; if this is true, the problem (11.29) can be rearranged to the dual
(kinematic) form similar to (6.28) and (10.18). The norm dual to (11.31)
determines geometry of the locking locus in the kinematic formulation.
12 The plane FMD problem with the trace

constraint. The case of n loading conditions
12.1 The Kelvin moduli being fixed
The merit function has the form (11.1). The problem (9.7) reads now

Iη = min Wη τ 1 (x), . . . , τ n (x) dx τ i ∈ Σi (Ω) (12.1)
Ω
where

#
m
1 # √
n

2
1
Wη τ , . . . , τ n
= min ηiτ i · ωK (12.2)
λK i=1
K=1
ωL · ω K = δKL , K, L = 1, 2, . . . , m; ω K ∈ Rm }
and d = 2, m = 3, n ≥ 3. We shall prove that

√ √
Wη τ 1 , . . . , τ n = Wλ η1 τ 1 , . . . , ηn τ n (12.3)

#m
1

Wλ σ 1 , σ 2 , . . . , σ n = μK Ŝ σ 1 , . . . , σ n (12.4)
λK
K=1
where m = 3 and μK are eigenvalues of the matrix Ŝ 3×3 = SS T where

S = σ 1 , . . . , σ n 3×n (12.5)
and σ i are set as columns; σ i ∈ E2s are interpreted as vectors in R3 according

to the Remark 3.1. The matrix

S T S = σ i · σ j i, j=1, ..., n (12.6)
has 3 positive eigenvalues μK (the same as the matrix Ŝ = SS T ) and n − 3

zero eigenvalues. If n > 3 the Gram matrix S T S ceases to play an important
1
role. We define the matrices Q and Λ− 2 as in Sec. 11. Now the matrix
An×3 has the form (11.9), where S is given by (12.5). The counterpart of
(11.10) is of the form
#
m
1 #
n

2
ω K · σ i = tr (AT A) (12.7)
λK i=1
K=1
√
where m = 3 and σ i = ηi τ i while (11.11) holds, with a new matrix Ŝ,
of the same dimensions 3 by 3 as in Sec. 11. As in Sec. 11 we note that:
1
xK = √ ω K , K = 1, 2, 3.
λK
Thus the problem (11.13) holds with a new matrix Ŝ, which confirms
(12.4). The eigenstates minimizing (12.2) are the eigevectors of the ma-
trix Ŝ. The optimal tensor C is given by (9.19), where ω K = ω ∗K . The
integrand of the functional in (12.1) has a quadratic growth, which makes it
possible to make use of the non-linear FEM, as in the theory of hyperelastic
bodies.
12.2 Releasing the Kelvin moduli with the trace constraint

Consider the problem (10.3) where Iη is given as in Sec. 12.1. The
equation (10.14) holds, where Zη is given by (11.29) with the matrix S(x)
given by
√ √
S(x) = η 1 τ 1 (x), . . . , η n τ n (x) (12.8)
Upon solving (11.29) one can find the Kelvin moduli by (11.30). The optimal
eigenstates ω ∗K are the eigenvectors of the matrix Ŝ. All the optimal Kelvin
moduli will be, in general, non-zero. Since the integrand in (11.29) is of
linear growth, there can appear the subdomains where S = 0. There the
material is not necessary. Thus the algorithm given determines not only the
layout of Cijkl but also predicts cutting the domain Ω to the domain where
the material is necessary, due to the load applied.
In case of n = 2 the result (12.4) still holds, since the matrix Ŝ has then
two positive eigenvalues μ1 , μ2 , the same as the matrix S T S. The equation
(12.4) reduces to (9.12), found in a different way. Similarly, Eq. (11.29)
reduces to (10.15).
13 The FMD problem in 3D

13.1 The case of the Kelvin moduli being fixed
We refer to the representation (3.29); m = 6. At each point we look for
the eigenstates ω K , K = 1, . . . , 6 to minimize
#
n
Fη (ω 1 , . . . , ω6 ) = ηi Υ(i) (ω 1 , . . . , ω 6 ) (13.1)
i=1
where η1 + . . . + ηn = 1. The tensors ω K ∈ E2s will be treated as vectors

in R6 see (3.25). The problem (9.7) has the form (12.1) and Wη is given
by (12.2), where now m = 6. One can prove (12.3), (12.4) where now the
summation runs over K = 1, . . . , m = 6. The matrix S given by (12.5)
has the dimension 6 by n. The matrix Ŝ = SS T has dimensions 6 by 6;
μK (Ŝ) ≥ 0 are eigenvalues of Ŝ, K = 1, . . . , 6.
a) Case of n ≥ 6
If σ i are nonzero, then μK (Ŝ) > 0, K = 1, . . . , 6. The matrix
(12.6) has positive eigenvalues, also equal to μK (Ŝ) and n − 6 zero
eigenvalues. One should solve (12.1), where Wη is given by (12.3),
(12.4) where m = 6.
b) Case of n = 5
Now the matrix S has dimensions 6 by 5 while the matrix Ŝ = SS T
is 6 by 6 and has one zero eigenvalues; according to the convention:
μ6 = 0. Thus Wλ is given by (12.4) with summation up to m =
5. Thus the eigenvalue λ6 ceases to have an effect on the solution
of (12.1). Upon solving (12.1) for m = 5 we find τ ∗1 , . . . , τ ∗5 .
Having these fields we solve 5 eigenvalue problems to find (ω K , μK ),
K = 1, . . . , 5, for the matrix Ŝ. The vector ω 6 is orthogonal to ω K ,
K = 1, . . . , 5. Tensor C is given by (3.29) for m = 6 and is defined
in a unique way.
c) Case of n = 4
The matrix S has dimensions 6 by 4 while Ŝ has two zero eigen-
values; μ5 = 0, μ6 = 0. Thus the summation in (12.4) is up to m = 4.
Upon solving (12.1) for m = 4 we find the fields τ ∗1 , . . . , τ ∗4 . We
compute the positive eigenvalues μK and eigenvectors μK of the ma-
trix Ŝ, K = 1, . . . , 4. Additionally we determine the vectors ω 5 , ω6
as orthogonal to ω 1 , . . . , ω 4 but this choice is not unique. Thus the
tensor C given by (3.29) for m = 6 will not have uniquely defined
projectors P 5 and P 6 .
d) Case of n = 1
e) The matrix S has dimensions 6 by 1 while the matrix Ŝ has 5 zero
eigenvalues:
μ2 = μ3 = μ4 = μ5 = μ6 = 0; μ1 > 0; η1 = 1
Now
μ1 (Ŝ) = ||σ 1 ||2
and
1
Wλ (σ 1 ) = ||σ 1 ||2 (13.2)
λ1
The vector ω1 is an eigenvector of the matrix Ŝ = σ 1 (σ 1 )T .
One can check that
1
ω1 = σ1 (13.3)
||σ 1 ||
We have arrived at the results analogous to (5.8), (5.9), where n = 1, d = 2.
Thus in the representation (3.29) only the first term is uniquely determined.
The cases of n = 2, 3 are left to the reader.
13.2 The Kelvin moduli released

a) n ≥ 6
We consider (10.3) with the functional Iη given by (12.1). Then
(10.14) holds. Thus Zη is given by (see (11.29))
m
# √ 1 √
i
Zη = min sK η 1 τ (x), . . . , η n τ (x) dx τ ∈ Σi (Ω)
n
Ω K=1
8
i = 1, . . . , n (13.4)
Since τ i (x) ∈ Rm , hence the matrix S(x) of dimensions m by n,

is given by (12.8), it has m non-negative singular values sK , K =
1, . . . , m. If n ≥ 6 then in 3D problem considered here m = 6.
Further steps are similar as in Sec. 12.2. It is worth emphasing that
the solution S can be zero in a subdomain, and there the material is
not necessary. The Kelvin moduli vanish in these regions.
b) n = 5
Summation in (13.4) runs up to m = 5, since one of the singular
values (of index 6) of the matrix S of dimensions 6 by 5 is zero.
Consequently λ∗6 = 0 by (11.30).
c) n = 4
Now s5 (S) = 0, s6 (S) = 0; λ∗5 = 0, λ∗6 = 0.
The eigenstates ω 5 , ω 6 will not be uniquely determined.
d) n = 1
The matrix S has dimensions 6 by 1. Now s1 (S) = ||σ 1 ||, σ 1 = τ 1 ;
ω 1 is given by (13.3). The result is similar to (6.10), (6.12).
14 Final remarks
1. The main conclusion is: the minimal number of the load conditions
which is indispensable to fix correctly (such that (3.3) holds) the com-
ponents of the tensor C equals m, or equals m = 3 in 2D and m = 6
in 3D. Thus this minimal number of loads is equal to the number of
components of strain (or stress). If the number of the load conditions
is smaller, then, in the whole design domain some of the Kelvin moduli
are zero. In the extreme case of a single load only one Kelvin modulus
is non-zero.
2. The FMD procedure is two-stage. To find the optimal moduli at
each point of the design domain one should: a) solve an auxiliary
problem, which, in general, has the form (13.4). There could appear
subdomains where all the minimizing fields τ ∗1 , . . . , τ ∗n vanish. The
remaining domain, denoted by Ωef f will be filled up by the optimal
non-homogeneous and anisotropic material of moduli Cijkl . b) we
solve the elasticity problem in the effective domain of given elastic
moduli. The state of stress will be determined uniquely.
3. The FMD concept can be applied not only to the minimization of the
compliance with the isoperimetric condition imposed on the integral
of the p-norm ||λ||p . Other admissible isoperimetric conditions are
discussed by (Barbarosie and Lopes, 2008). The possible FMD exten-
sions to other functionals are overviewed in (Haslinger et al. 2010).
15 Bibliography
C. Barbarosie, S. Lopes, Study of the cost functional for free material design
problems. Numer. Funct. Anal. Optimiz. 29: 115-125, 2008.
M.P. Bendsøe, J.M. Guedes, R.B. Haber, P. Pedersen and J.E. Taylor, An
Analytical Model to Predict Optimal Material Properties in the Context
of Optimal Structural Design, J. Appl. Mech. Trans. ASME 61, 930-937,
1994.
M.P. Bendsøe, A.R. Diaz, R. Lipton, J.E. Taylor, Optimal design of material
properties and material distribution for multiple loading conditions. Int. J.
Numer. Meth. Eng. 38: 1149–1170, 1995.
A. Blinowski, J. Ostrowska-Maciejewska, J. Rychlewski, Two-Dimensional
Hooke’s Tensors-Isotropic Decomposition, Effective Symmetry Criteria. Arch.
Mech., 48, 325-345, 1996.
A.V. Cherkaev, Variational Methods for Structural Optimization, Springer,
New York, 2000.
Ph. G. Ciarlet, Mathematical Elasticity, vol. I, Three dimensional elasticity.
North Holland, Amsterdam 1988.
S. Czarnecki, T. Lewiński, Shaping the stiffest three-dimensional structures

from two given isotropic materials. Computer Assisted Mechanics and En-
gineering Sciences, 13 (2006) 53-83, 2006.
S. Czarnecki, T. Lewiński, A stress-based formulation of the free material
design problem with the trace constraint and single loading condition. Bull.
Polish Ac. Sciences. Tech. Sci. 60: 191-204, 2012.
S. Czarnecki, T. Lewiński, The stiffest designs of elastic plates. Vector
optimization for two loading conditions. Computer Methods in Applied Me-
chanics and Engineering 200, 1708-1728, 2011.
F. Demengel and P. Suquet, On locking materials, Acta Appl. Math. 6,
185-211 ,1986.
G. Duvaut, J.-L. Lions, Inequalities in Mechanics and Physics. Springer-
Verlag, Berlin, 1976.
G. Dzierżanowski, T. Lewiński, Compliance minimization of thin plates
made of material with predefined Kelvin moduli. Part I. Solving the local
optimization problem, Arch. Mech. 64, 21-40, 2012a.
G. Dzierżanowski, T. Lewiński, Compliance minimization of thin plates
made of material with predefined Kelvin moduli. Part II. The effective
boundary value problem and exemplary solutions, Arch. Mech. 64: 111-
135, 2012b.
G. Dzierżanowski, T. Lewiński, Compliance Minimization of Two-Material
Elastic Structures. This volume of the CISM Lectures 2013.
Y.C. Fung, Foundations of Solid Mechanics, Prentice-Hall, Inc., New Jersey,
U.S.A. 1965.
G.H. Hardy, J.E. Littlewood, G. Polya, Inequalities, Cambridge Mathemat-
ical Library, University Press, Cambridge 1999.
J. Haslinger, M. Kočvara, G. Leugering and M. Stingl, Multidisciplinary
free material optimization, SIAM J. Appl. Math. 70(7): 2709-2728, 2010.
H.R.E.M. Hörnlein, M. Kočvara, R. Werner, Material optimization: bridg-
ing the gap between conceptual and preliminary design. Aeorosp. Sci.
Technol. 5: 541-554., 2001.
M. Kočvara, M. Stingl, Free Material Optimization for Stress Constraints.
Struct. Multidisc. Optim. 33, 323-335, 2007.
M. Kočvara, M. Stingl, and J. Zowe, Free material optimization: recent
progress. Optimization 57: 79-100, 2008.
T. Lewiński, J.J. Telega, Michell-like grillages and structures with locking,
Arch. Mech. 53: 457-485, 2001.
T. Lewiński, T. Sokól, On basic properties of Michell’s structures, this vol-

ume of the CISM Lectures 2013.
M.M. Mehrabadi, S.C. Cowin, Eigentensors of linear anistropic elastic ma-
terials. Q. J. Mech. Appl. Math. 43: 15-41, 1990.
G.J. Minty, On the monotonicity of the gradient of a convex function. Pa-
cific J. of Math. 14: 243-247, 1964.
M. Moakher, A.N. Norris, The closest elastic tensor of arbitrary symmetry
to an elasticity tensor of lower symmetry. J. Elasticity 85: 215-263, 2006.
J. Nečas, I. Hlavaček, Mathematical Theory of Elastic and Elasto-Plastic
Bodies. An Introduction. Elsevier, Rotterdam, 1981.
J.C.C. Nitsche, Lectures on Minimal Surfaces, vol. 1, Cambridge University
Press 1989.
R.T. Rockafellar, Integral functionals, normal integrands and measurable
selections. In: Nonlinear Operators and the Calculus of Variations.
L.Waelbroeck (ed.), Lecture Notes in Mathematics 9543), pp 157-207,
Springer 1976.
J. Rychlewski, On Hooke’s Law. Prikl. Mat. Mekh., 48: 420-435 , 1984 (in
Russian).
G. Strang and R.V. Kohn, Hencky-Prandtl nets and constrained Michell
trusses, Comp. Meth. Appl. Mech. Eng. 36: 207-222 ,1983.
S. Sutcliffe, Spectral Decomposition of the Elasticity Tensor. J. Appl. Mech.
Trans. ASME 59: 762-773, 1992.
S. Turteltaub, P. Washabaugh, Optimal distribution of material properties
for an elastic continuum with structure-dependent body force, Int. J. Solids.
Struct. 36: 4587-4608, 1999.
R. Werner, Free Material Optimization. Mathematical Analysis and Nu-
merical Simulation. Ph.D Thesis, Institute of Applied Mathematics II,
Friedrich-Alexander University of Erlangen-Nuernberg, Erlangen, Germany,
147 pp., 2000.
Introductory Notes on Topological Design
Optimization of Vibrating Continuum Structures
Niels Olhoff 1 and Jianbin Du2

1
Department of Mechanical and Manufacturing Engineering, Aalborg University,
DK-9220 Aalborg East, Denmark
2
School of Aerospace, Tsinghua University, Beijing 100084, P.R.China
Abstract This paper presents a brief introduction to topological design

optimization, and together with five sequential papers gives an overview
of the application of this rather novel method to problems of design of
linearly elastic continuum-type structures against vibration and noise. The
objective of such problems is often to drive the structural eigenfrequencies
of vibration as far away as possible from a prescribed external excitation
frequency - or band of excitation frequencies - in order to avoid resonance
phenomena with high vibration and noise levels. This objective may, e.g.,
be achieved by (i) maximizing the fundamental eigenfrequency of the
structure, (ii) maximizing the distance (gap) between two consecutive
eigenfrequencies, (iii) maximizing the dynamic stiffness of the structure
subject to forced vibration, or by (iv) minimizing the sound power flow
radiated from the structural surface into an acoustic medium. The
mathematical formulations of these optimization problems and several
illustrative examples are presented in this series of papers.
1 Introduction
The current paper

x ‘Introductory Notes on Topological Design Optimization of Vibrating
Continuum Structures’ (Olhoff and Du, 2013A),

DOI 10.1007/978-3-7091-1643-2_10, © CISM, Udine 2014
260 N. Olhoff and J. Du
and the five sequential papers entitled

x ‘Structural Topology Optimization with Respect to Eigenfrequencies of
Vibration’ (Olhoff and Du, 2013B)
x ‘On Optimum Design and Periodicity of Band-gap Structures’ (Olhoff
and Niu, 2013C)
x ‘Topological Design for Minimum Dynamic Compliance of Structures
under Forced Vibration’ (Olhoff and Du, 2013D)
x ‘Topological Design for Minimum Sound Emission from Structures
under Forced Vibration’ (Olhoff and Du, 2013E)
x ‘Discrete Material Optimization of Vibrating Laminated Composite
Plates for Minimum Sound Emission’ (Olhoff and Niu, 2013F),
were presented in six lectures at the Advanced School ‘Topology Optimization
in Structural and Continuum Mechanics’ held at the International Centre for
Mechanical Sciences (CISM) in Udine, Italy, June 18-22, 2012. The School was
coordinated by Professors George Rozvany, Budapest, Hungary, and Tomasz
Lewinski, Warsaw, Poland.
The method of topology optimization of continuum structures first appeared
in the literature in 1988, and was originally developed for determining the
distribution of an elastic material within an admissible design domain that yields
the stiffest possible structure for a prescribed weight, see Bendsøe and Kikuchi
(1988) and Bendsøe (1989). Since usual sizing and shape optimization methods
generally cannot change the structural topology, the development of the method
of topology optimization was a remarkable break-through in the field of
optimum design, as the choice of the best topology generally has the most decisive
impact on the gain that can be achieved by optimization. Topology optimization is
therefore an important pre-processing tool for sizing and shape optimization, see
Olhoff et al. (1991). During the last decade, the method has been extended to handle
several other design objectives and constraints. Topology optimization has therefore
become a standard tool for synthesis of parts or whole structures in the automotive
and aerospace industries, and it is rapidly spreading into other mechanical design
disciplines. The reader is referred to the exhaustive textbook by Bendsøe and
Sigmund (2003), the IUTAM Symposium proceedings edited by Bendsøe et al.
(2006), and the review article by Eschenauer and Olhoff (2001) for further
developments and publications.
Passive design against vibrations and noise was first undertaken by Olhoff (1976,
1977) in the form of shape optimization with respect to eigenfrequencies of freely,
transversely vibrating beams. By maximizing the fundamental eigenfrequency for
given beam volume, optimum cost designs against vibration resonance were
Introductory Notes on Topological Design Optimization… 261
obtained subject to all external excitation frequencies within the large range from
zero and up to the fundamental eigenfrequency. Optimization with respect to a
higher order eigenfrequency was found to produce a large gap between the subject
eigenfrequency and the adjacent lower eigenfrequency, and offered even more
competitive designs for avoidance of resonance in problems where external
excitation frequencies are confined within a large interval with finite lower and
upper limits. In subsequent beam shape optimization papers by Olhoff and Parbery
(1984), Bendsøe and Olhoff (1985), and Olhoff et al. (2012), the design objective
was directly formulated as maximization of the separation (gap) between two
consecutive eigenfrequencies of prescribed orders. The study in Olhoff et al. (2012)
yields the interesting result that, except for beam segments adjacent to the beam
ends (whose designs are characteristic for the specific boundary conditions
considered), the entire inner parts of the optimized beam designs exhibit a
significant periodicity in terms of repeated inner beam segments, the number of
which increases rapidly with increasing values of the orders of the consecutive
upper and lower frequencies of the maximized gaps.
It should be noted that the separation of adjacent eigenfrequencies as considered
by Olhoff et al. (2012) and Jensen and Pedersen (2006) is closely related to the
existence of so-called phononic (or acoustic) band gaps, i.e., gaps in the wave band
structure for periodic materials implying that elastic waves cannot propagate in
certain frequency ranges. Sigmund (2001) applied topology optimization to
maximize phononic band gaps in periodic materials (see also Diaz et al., 2005, and
Halkjær et al., 2006). Moreover, Sigmund and Jensen (2003), Jensen (2003), and
Jensen and Sigmund (2005) performed minimization of the response of band gap
structures (wave damping).
In recent papers, Bruggi and Taliercio (2012) and Niu et al. (2009) performed
topology optimization for maximum fundamental eigenfrequency of structures
composed of micropolar solids and cellular material, respectively, and Yoon (2010a)
applied a parameterisation based on element connectivity. However, topology
optimization with respect to eigenfrequencies of structural vibration was first
considered by Dias and Kikuchi (1992), who dealt with single frequency design of
plane disks. Subsequently, Ma et al. (1994), Dias et al. (1994), and Kosaka and
Swan (1999) presented different formulations for simultaneous maximization of
several frequencies of free vibration of disk and plate structures, defining the
objective function as a scalar weighted function of the eigenfrequencies. The paper
(Pedersen, 2000) dealt with maximum fundamental eigenfrequency design of plates,
and included a technique to avoid spurious localized modes. In contrast to the earlier
work, Krog and Olhoff (1999), Jensen and Pedersen (2006), and Du and Olhoff
(2007b) applied a variable bound formulation (see Bendsøe et al., 1983) which
facilitates proper treatment of multiple eigenfrequencies that very often result from
the optimization. The first of these papers treats optimization of fundamental and
higher order eigenfrequencies of disk and plate structures, while Jensen and
Pedersen (2006), and Du and Olhoff (2007b), also deal with maximization of the
separation of adjacent eigenfrequencies for single- and bi-material plates.
Topology optimization with the objective of maximizing the dynamic stiffness
(minimizing the dynamic compliance) of elastic structures subjected to time-
harmonic external loading of given frequency and amplitude are, e.g., studied by
Ma et al. (1995), Min et al. (1999), Jog (2002), Jensen and Sigmund (2005), Olhoff
and Du (2005, 2013), Kang et al. (2012), and Yang and Li (2013). Similar work on
structural topology optimization for minimum vibration amplitude response over a
range of excitation frequencies has been carried out by Calvel and Mongeau (2005)
and Jensen (2007). Recent papers on topology optimization for minimum frequency
response have been published by Yoon (2010b) and Shu et al. (2011).
Optimization of structural-acoustic systems against sound and noise emission
has benefited from textbooks by Koopmann and Fahnline (1997) and Kollmann
(2000), and proceedings by Munjal (2002) and Bendsøe et al. (2006) from two
IUTAM Symposia. During recent years, topology optimization based acoustic
design of elastic structures subjected to time-harmonic external mechanical
loading of given excitation frequency or frequency range, amplitude, and spatial
distribution has attracted significant attention, and minimization (or
maximization) of the acoustic power radiated from the structural surface(s) into a
surrounding or interior acoustic medium like air have been frequent design
objectives, see, e.g., Christensen et al. (1998), Luo and Gea (2003), Wadbro and
Berggren (2006), Sorokin et al. (2006), Bös (2006), Olhoff and Du (2006), Yoon
et al. (2007), Du and Olhoff (2007a, 2010), Dühring et al. (2008), Yamamoto et
al. (2009), Niu et al. (2010), Nandy and Jog (2011), Du et al. (2011), Kook et al.
(2012), and Yang and Du (2013).
In terms of optimization of composite structures with respect to acoustic
criteria, we may refer the reader to the review article Denli and Sun (2007) and
the bibliography Mackerle (2003), and a large number of papers cited therein. As
examples of various types of problems of optimum structural-acoustic design
with composite materials, we may refer to Hufenbach et al. (2001), Thamburaj
and Sun (2002), Chen et al. (2005); Yamamoto et al. (2008), Jensen (2009) and
Niu et al. (2010).
In Niu et al. (2010), the novel topology optimization based method termed
the Discrete Material Optimization (DMO) method (see Stegmann and Lund
2005, Lund and Stegmann 2005) is applied to furnish the simultaneous design
optimization of fiber angles, stacking sequence and selection of material for
vibrating laminated composite plates with minimum sound emission.
2 Topology Optimization and Material Interpolations
Contrary to shape optimization, problems of topology optimization are defined

on a fixed domain of space called the admissible design domain (see, e.g.,
Bendsøe and Sigmund, 2003, and Eschenauer and Olhoff, 2001). The topology
problem is basically one of discrete optimization, but this difficulty is avoided by
introducing relationships between stiffness components and the volumetric
density of material U over the admissible design domain.
Fig. 1 illustrates some basic concepts for a topology optimization problem for
a continuum structure to be made of a single material. Given are the admissible
design domain (indicated by grey in Fig. 1a), the boundary conditions, loading,
and the volume of solid, elastic material for the structure. As indicated in Fig. 1,
usually a fixed finite
Initial Design element mesh is
embedded in the entire
admissible design do-
Evenly main. Typically, the
distributed mesh is a uniform,
material rectangular partition of
space, and the material
density U is assumed to
be constant within each
(a) finite element. For the
initial design, the given
Optimum Design amount of material
normally is distributed
uniformly over the
Void admissible design
domain as indicated in
Solid Fig. 1a.
material To determine the op-
timum structural topo-
(b) logy, the densities Ue of
material in each of the
Figure 1. Illustration of a topology design process
finite elements are used
from the initial (a) to the optimum design (b).
as design variables de-
fined between limits 0
(corresponding to void as shown by white in Fig. 1b) and 1 (corresponding to
solid elastic material shown in black). The aim of the optimization process is to
find out, for each of the finite elements in the admissible design domain, whether
it should contain solid material or not. In this process (of successive iterations),
each of the design variables tend to attain one of their limiting values as
explained below, thereby forming a design with aggregations of finite elements
with solid material and void, respectively, see Fig. 1b. The result is a rough
description of outer as well as inner boundaries of the design that represents the
overall optimum topology. This topological design may subsequently be used as
a basis for refined shape optimization, see Olhoff et al. (1991).
2.1 SIMP model for topology optimization of single-material structures

As mentioned above, it is the aim of the optimization process to determine the
optimum zero(void)-one(solid) distribution of a prescribed amount of the given
material over the admissible design domain. To achieve this goal, many different
material models have been developed (see, e.g., Bendsøe and Sigmund, 2003,
and Eschenauer and Olhoff, 2001), among which the SIMP (Solid Isotropic
Microstructure with Penalty) model proposed by Bendsøe (1989), Rozvany and
Zhou (1991) and Rozvany et al. (1992) is a simple and effective one which is
widely used in optimum topology design. The SIMP model is normally applied
together with a filtering technique, see Sigmund (1997), as this prevents
checkerboard formation and dependency of optimum topology solutions on finite
element mesh-refinement. According to the SIMP model, the finite element
elasticity matrix Ee is expressed in terms of the element volumetric material
density Ue, 0 d Ue d 1, in a power p, p 1, as
E e ( U e ) U ep E*e (1)
where E*e is the elasticity matrix of a corresponding element with the fully solid
elastic material the structure is to be made of. The power p in (1), which is
termed the penalization power, is introduced with a view to yield distinctive “0-
1” designs, and is normally assigned values increasing from 1 to 3 during the
optimization process. Such values of p have the desired effect of penalizing
intermediate densities 0 < Ue < 1 since the element material volume is
proportional to Ue while the interpolation (1) implies that the element stiffness is
less than proportional. Note also that the interpolation (1) satisfies E e (0) = 0 and
E e (1) = E*e , implying that if a final design has density 0 and 1 in all elements,
this is a design for which the structural response has been evaluated with a
correct physical model.
By analogy with (1), for a vibrating structure the finite element mass matrix
may be expressed as
M e ( U e ) U eq M *e (2)
where M *e represents the element mass matrix corresponding to fully solid
material, and the power q 1. Apart from exceptions briefly discussed in the
following section, normally q = 1 is chosen.
The global stiffness matrix K and mass matrix M for the finite element based
structural response analyses behind the optimization, can now be calculated by
NE NE
p
K ¦U e K *e , M ¦U q
e M *e (3)
e 1 e 1
Here, K *e is the stiffness matrix of a finite element with the fully solid material
for the structure, and NE denotes the total number of finite elements in the
admissible design domain.
In the problem formulations in Chapters 3, 5 and 7, Ve, e = 1,…,NE, denotes
the volumes of the finite elements, V0 is the total volume of the admissible
design domain, and for single material design, V* denotes the total volume
NE
¦U V e e of solid elastic material which is available for the structure.

e 1
2.2 Localized eigenmodes

With values assigned to p and q as stated above, application of the SIMP model
for problems of topology optimization with respect to eigenfrequencies may lead
to the occurrence of spurious, localized eigenmodes associated with very low
values of corresponding eigenfrequencies. The localized eigenmodes may occur
in sub-regions of the design domain with low values of the material density (e.g.
Ue d 0.1), where the ratio between the stiffness (with, say, p = 3 in the
interpolation formula) and the mass (with q = 1) is very small. To eliminate these
spurious eigenmodes, we may use the method of Pedersen (2000) of linearizing
the element stiffness or the approach of Tcherniak (2002) of setting the element
mass to zero in sub-regions with low material density. Thus, following Tcherniak
(2002) with a slight modification to avoid numerical singularity, the interpolation
formula (2) for the finite element mass matrix was modified as
U M * , U e ! 0.1
M e ( U e ) ® er e* . (4)
¯ U e M e , U e d 0.1
Here, the mass is set very low via a high value of the penalization power r in
sub-regions with low material density. Thus, r is chosen to be about r = 6, i.e.,
much larger than the penalization power p for the stiffness, which is kept
unchanged at a value about p = 3.
It is noted that Eq. (4) is discontinuous at the low value Ue = 0.1 of the
material density. Numerically this is not a serious problem since the
discontinuity only occurs at a single point. However, we can always improve (4)
by generating a continuous interpolation model for the mass with respect to any
value of the material density between 0 and 1. For example, to achieve continuity
of the interpolation model, we may introduce the following revised form of Eq.
(4),
U M* , U e ! 0.1
M e (Ue ) ® e 6 e * . (4a)
¯c0 U e M e , U e d 0.1
where the coefficient c0 105 enforces the C0 continuity at the value Ue = 0.1 of
the material density. In several of the examples presented later in this paper, for
comparison, we have applied each of the interpolation models (4) and (4a) in the
numerical solution scheme and only found negligible differences in the final
results. The reason is that in both models, the region with lower density has a
very small contribution to the first several eigenfrequencies of the structure.
Furthermore, all intermediate values of the material density will approach zero or
one during the design process, which implies that the change of the interpolation
model in regions with lower density as shown in (4a) must have very limited
influence on the final zero-one design.
2.3 SIMP model for topology optimization of bi-material structures

The SIMP model for topology optimization of structures made of two different
solid elastic materials can be easily obtained by an extension of the SIMP model
for single-material design. Following Bendsøe and Sigmund (1999), the finite
element elasticity matrix for the bi-material problem can be expressed as
E ( U ) U p E*1 (1 U p )E*2 (5)
e e e e e e
*1 *2
where E and E denote the element elasticity matrices corresponding to the
e e
two given solid, elastic materials *1 and *2. Here, material *1 is assumed to be
the stiffer one. The penalization power p in (5) has generally been assigned the
value 3 which resulted in distinctive optimum topology designs in the examples
of bi-material design considered in the sequential papers Olhoff and Du
(2013B,E). It follows from (5) that for a given element, Ue = 1 implies that the
element fully consists of the solid material *1, while Ue = 0 means that the
element fully consists of the solid material *2.
The element mass matrix of the bi-material model may be stated as the
simple linear interpolation
M e ( U e ) U e M *e1 (1 U e )M *e2 (6)
where M *e1 and M *e2 are the element mass matrices corresponding to the two
different, given solid elastic materials *1 and *2.
The SIMP model formulated by (1) and (2) (or (5) and (6)) may be regarded
as an interpolation scheme for the structural stiffness and mass with respect to
material volume density. Recently, a generalized material model based on a
polynomial interpolation was proposed by Jensen and Pedersen (2006), and it
was shown how proper polynomials corresponding to different design objectives
can be easily obtained.
When bi-material design is treated via the problem formulations in the
sequential papers Olhoff and Du (2013B,E), then V* denotes the total volume
NE
¦U V
e 1
e e of the stiffer material *1 available for the structure, while the total
volume of material *2 is given by V0 - V*, where V0 is the volume of the

admissible design domain. In figures in Olhoff and Du (2013B,E) presenting
optimum topologies of bi-material structures, material *1 is shown in black and
material *2 in grey.
2.4 DMO model for optimization of laminated composite plates

The parameterization for discrete material optimization (DMO) (see Stegmann
and Lund 2005, Lund and Stegmann 2005), and the introduction of penalty
functions for DMO of laminated composite plate structures are discussed in the
papers Niu and Olhoff (2012, 2013F) .
References
Bendsøe, M.P., 1989. Optimal shape design as a material distribution problem. Struct
Optim 1: 193-202.
Bendsøe, M.P., Olhoff, N., Taylor, J.E., 1983. A variational formulation for multicriteria
structural optimization. J Struct Mech 11: 523-544.
Bendsøe, M.P., Olhoff, N., 1985. A method of design against vibration resonance of
beams and shafts. Optim Control Appl Meth 6: 191-200.
Bendsøe, M.P., Kikuchi, N., 1988. Generating optimal topologies in structural design
using a homogenization method. Comput Methods Appl Mech Eng 71(2): 197-224.
Bendsøe, M.P., 1989. Optimal shape design as a material distribution problem. Struct
Optim 1: 193-202.
Bendsøe, M.P., Sigmund, O., 1999. Material interpolation schemes in topology
optimization. Arch Appl Mech 69: 635-654.
Bendsøe, M.P., Sigmund, O., 2003. Topology Optimization: Theory, Methods and
Applications. Berlin, Heidelberg, New York: Springer.
Bendsøe, M.P., Olhoff, N., Sigmund, O. (Eds.), 2006. Proc. IUTAM Symposium on
Topological Design Optimization of Structures, Machines and Materials – Status and
Perspectives. Copenhagen, Denmark, Oct. 26-29, 2005. Dordrecht, The Netherlands:

Springer.
Bratus, A.S., Seyranian, A.P., 1983. Bimodal solutions in eigenvalue optimization
problems. Appl Math Mech 47: 451-457.
Bruggi, M., Taliercio, A., 2012. Maximization of the fundamental eigenfrequency of
micropolar solids through topology optimization. Struct Multidisc Optim 46(4): 549-
560.
Bös, J., 2006. Numerical optimization of the thickness distribution of three-dimensional
structures with respect to their structural acoustic properties. Struct Optim 32: 12-30.
Calvel, S., Mongeau, M., 2005. Topology optimization of a mechanical component
subject to dynamic constraints. Technical Report LAAS N0 05374. Available from
http://mip.ups-tlse.fr/perso/mongeau.
Chen, S.H., Song, X.W., Yang, Z.J., 2005. Modal optimal control of cabin noise of
vehicles. Smart Mater Struct 14: 257-264.
Cheng, G., Olhoff, N., 1982. Regularized formulation for optimal design of
axisymmetric plates. Int J Solids Struct 18: 153-169.
Christensen, S.T., Sorokin, S.V., Olhoff, N., 1998. On analysis and optimization in
structural acoustics – Part I: Problem formulation and solution techniques. Struct
Optim 16: 83-95.
Courant, R., Hilbert, D., 1953. Methods of Mathematical Physics, Vol. 1. Interscience
Publishers, New York.
Denli, H., Sun, J.Q., 2007. Structural-acoustic optimization of composite sandwich
structures: a review. The Shock and Vibration Digest 39 (3): 189-200.
Diaz, A.R., Kikuchi, N., 1992. Solutions to shape and topology eigenvalue optimization
problems using a homogenization method. Int J Num Meth Engng 35: 1487-1502.
Diaz, A.R., Lipton, R., Soto, C.A. (1994). A new formulation of the problem of optimum
reinforcement of Reissner-Mindlin plates. Comp Meth Appl Mechs Eng 123: 121-139.
Diaz, A.R., Haddow, A.G., Ma, L., 2005. Design of band-gap grid structures. Struct
Multidisc Optim. 29(6): 418-431.
Du, J., Olhoff, N., 2004a. Topological optimization of continuum structures with design-
dependent surface loading - Part I: New computational approach for 2D problems.
Struct Multidisc Optim 27: 151-165.
Du, J., Olhoff, N., 2004b. Topological optimization of continuum structures with design-
dependent surface loading - Part II: Algorithm and examples for 3D problems. Struct
Multidisc Optim 27: 166-177.
Du, J., Olhoff, N., 2007a. Minimization of sound radiation from vibrating bi-material
structures using topology optimization. Struct Multidisc Optim 33: 305-321.
Du, J., Olhoff, N., 2007b. Topological design of freely vibrating continuum structures
for maximum values of simple and multiple eigenfrequencies and frequency gaps.
Struct Multidisc Optim 34:91-110. See also Publisher’s Erratum in Struct Multidisc
Optim (2007) 34:545.
Du J., Olhoff N., 2010. Topological design of vibrating structures with respect to
optimum sound pressure characteristics in a surrounding acoustic medium. Struct
Multidisc Optim 42, 43-54.
Du, J., Song, X., Olhoff, N., 2011. Topological design of acoustic structure based on the
BEM-FEM format and the mixed formulation. In: Proc. 9th World Congress on
Structural and Multidisciplinary Optimization (WCSMO-9), June 13-17, 2011,

Shizuoka, Japan.
Dühring M.B., Jensen J.S., Sigmund O., 2008. Acoustic design by topology optimization.
J Sound Vib 317: 557-575.
Eschenauer, H.A., Olhoff, N., 2001. Topology optimization of continuum structures: A
review. Appl Mech Rev 54(4): 331-389.
Haftka, R.T., Gurdal, Z., Kamat, M.P., 1990. Elements of Structural Optimization.
Dordrecht: Kluwer.
Halkjær S., Sigmund, O., Jensen J.S., 2006. Maximizing band gaps in plate structures.
Struct Multidisc Optim 32:263-275.
Hammer, V.B., Olhoff, N., 1999. Topology optimization with design dependent loads. In
Proc. WCSMO-3, Third World Congress of Structural and Multidisciplinary
Optimization. Buffalo, NY, May 17-21, 1999.
Hammer, V.B., Olhoff, N., 2000. Topology optimization of continuum structures
subjected to pressure loading. Struct Multidisc Optim 19: 85-92.
Haug, E.J., Choi, K.K., Komkov, V., 1986. Design Sensitivity Analysis of Structural
Systems. New York: Academic Press.
Herrin, D.W., Martinus, F., Wu, T.W., Seybert, A.F., 2003. A New Look at the High
Frequency Boundary Element and Rayleigh Integral Approximations. Society of
Automotive Engineers, Inc.
Hufenbach, W., Kroll, L., Holste, C., Täger, O., Barkanov, E., 2001. Design of
dynamically loaded fiber-reinforced structures with account of their vibro-acoustic
behavior. Mech Composite Mater 37(2): 145-152.
Jensen, J.S., 2003. Phononic band gaps and vibrations in one- and two-dimensional
mass-spring structures. J. Sound and Vibration 266: 1053-1078.
Jensen J.S., 2007. Topology optimization of dynamic problems with Padé approximants.
Int J Numer Meth Engng 72: 1605-1630.
Jensen, J.S., 2009. Space-time topology optimization for one-dimensional wave
propagation. Comput Methods Appl Mech Engrg 198: 705-715.
Jensen, J.S., Sigmund, O., 2005. Topology optimization of photonic crystal structures: a
high-bandwidth low-loss T-junction waveguide. J. Opt. Soc. Am. B 22(6): 1191-1198.
Jensen, J.S., Pedersen, N.L., 2006. On maximal eigenfrequency separation in two-
material structures: the 1D and 2D scalar cases. J Sound and Vibration 289: 967-986.
Jog, C.S., 2002. Topology design of structures subjected to periodic loading. J Sound
and Vibration 253(3): 687-709.
Kang, Z., Zhang, X., Jiang, S., Cheng, G., 2012. On topology optimization of damping
layer in shell structures under harmonic excitations. Struct Multidisc Optim 46(1): 51-
67.
Kollmann, F.G., 2000. Machine Acoustics – Basics, Measurement Techniques,
Computation, Control (in German). 2nd Rev. Edition, Springer: Berlin, Heidelberg,
New York.
Kook, J., Koo, K., Hyun, J., Jensen, J.S., Wang, S., 2012. Acoustical topology
optimization for Zwicker’s loudness model – Application to noise barriers. Computer
Methods in Applied Mechanics and Engineering 237-240: 130-151.
Koopmann, G.H., Fahnline, J.B., 1997. Designing Quiet Structures: A Sound Power
Minimization Approach. Academic Publishers: London.
Kosaka, I., Swan, C.C., 1999. A symmetry reduction method for continuum structural
topology optimization. Computers & Structures 70: 47-61.
Krog, L.A., Olhoff, N., 1999. Optimum topology and reinforcement design of disk and
plate structures with multiple stiffness and eigenfrequency objectives. Computers &
Structures 72: 535-563.
Lancaster, P., 1964. On eigenvalues of matrices dependent on a parameter. Numerische
Mathematik 6: 377-387.
Lax, M., Feshbach, H., 1947. On the radiation problem at high frequencies. Journal of
the Acoustical Society of America 19: 682-690.
Lund, E., 1994. Finite Element Based Design Sensitivity Analysis and Optimization.
Ph.D. Thesis, Special Report No. 23, Dept. of Mechanical Engineering, Aalborg
University, Denmark.
Lund, E., 2009. Buckling topology optimization of laminated multi-material composite
shell structures. Composite Structures 91: 158-167.
Lund E., Stegmann J., 2005. On structural optimization of composite shell structures
using a discrete constitutive parametrization. Wind Energy 8: 109-124.
Luo J.H., Gea H.C., 2003. Optimal stiffener design for interior sound reduction using a
topology optimization based approach. J Vib Acoust 125: 267-273.
Ma, Z.D., Cheng, H.C., Kikuchi, N., 1994. Structural design for obtaining desired
eigenfrequencies by using the topology and shape optimization method. Comput
Systems Engng 5(1): 77-89.
Ma, Z.D., Kikuchi, N., Cheng, H.C., 1995. Topological design for vibrating structures.
Comput Methods Appl Mech Engrg 121: 259-280.
Mackerle, J., 2003. Topology and shape optimization of structures using FEM and BEM
– a bibliography (1999-2001). Finite elements in Analysis and design 39: 243-253.
Masur, E.F., 1984. Optimal structural design under multiple eigenvalue constraints. Int J
Solids Struct 20: 211-231.
Masur, E.F., 1985. Some additional comments on optimal structural design under
multiple eigenvalue constraints. Int J Solids Struct 21: 117-120.
Min, S., Kikuchi, N., Park, Y.C., Kim, S., Chang, S., 1999. Optimal topology design of
structures under dynamic loads. Struct Optim 17: 208-218.
Mroz, Z., Rozvany, G.I.N., 1975. Optimal design of structures with variable support
conditions. Journal of Optimization theory and Applications 15: 85-101.
Munjal, M.L., (Ed) 2002: Proc. IUTAM Symp. Designing for Quietness. Bangalore,
India, December 12-14, 2000. Kluwer: Dordrecht, The Netherlands.
Nandy, A.K., Jog, C.S., 2011. Optimization of vibrating structures to reduce radiated
noise. Struct Multidisc Optim 45(5): 717-728.
Niu, B., Yan, J., Cheng, G., 2009. Optimum structure with homogeneous optimum
cellular material for maximum fundamental frequency. Struct Multidisc Optim 39(2):
115-132.
Niu B., Olhoff N., Lund E., Cheng G., 2010. Discrete material optimization of vibrating
laminated composite plates for minimum sound radiation. Int J Solids Struct 47: 2097-
2114.
Olhoff, N., 1976. Optimization of vibrating beams with respect to higher order natural
frequencies. J Struct Mech 4: 87-122
Olhoff, N., 1977. Maximizing higher order eigenfrequencies of beams with constraints
on the design geometry. J Struct Mech 5: 107-134.

Olhoff, N., 1989. Multicriterion structural optimization via bound formulation and
mathematical programming. Struct Optim 1: 11-17.
Olhoff, N., Parbery, R., 1984. Designing vibrating beams and rotating shafts for
maximum difference between adjacent natural frequencies. Int J Solids Structures 20:
63-75.
Olhoff, N., Bendsøe, M.P., Rasmussen, J., 1991. On CAD-intergrated structural
topology and design optimization. Comput Methods Appl Mech Eng 89: 259-279.
Olhoff, N., Du, J., 2005. Topological design of continuum structures subjected to forced
vibration. In Proc. 6th World Congress of Structural and Multidisciplinary
Optimization (WCSMO-6), Rio de Janeiro, Brazil 2005, 8 pp.
Olhoff, N., Du, J., 2006. Topology optimization of vibrating bi-material structures with
respect to sound radiation. In Proc. IUTAM Symposium on Topological Design
Optimization of Structures, Machines and Materials – Status and Perspectives (held
in Copenhagen 2005). Dordrecht, The Netherlands: Springer, p. 43-52.
Olhoff, N., Niu, B., Cheng, G., 2012. Optimum design of band-gap beam structures. Int
J Solids Structures, 49(22), 3158-3169.
Olhoff, N., Du, J., 2013. Topological design for minimum dynamic compliance of
continuum structures subjected to forced vibration. Struct Multidisc Optim. (accepted).
Olhoff, N., Du, J., 2013A. Introductory notes on topological design optimization of
vibrating continuum structures. In: Rozvany, G., Lewinski, T., Eds., Topology
Optimization in Structural and Continuum Mechanics, Int. Centre for Mechanical
Sciences, Udine, Italy, June 18-22, 2012. Springer-Verlag, Vienna, 2013, 15 pp.
Olhoff, N., Du, J., 2013B. Structural topology optimization with respect to
eigenfrequencies of vibration. In: Rozvany, G., Lewinski, T., Eds., Topology
Olhoff, N., Niu, B., 2013C. On optimum design and periodicity of band-gap structures.
In: Rozvany, G., Lewinski, T., Eds., Topology Optimization in Structural and
Continuum Mechanics, Int. Centre for Mechanical Sciences, Udine, Italy, June 18-22,
2012. Springer-Verlag, Vienna, 2013, 26 pp.
Olhoff, N., Du, J., 2013D. Topological design for minimum dynamic compliance of
structures under forced vibration. In: Rozvany, G., Lewinski, T., Eds., Topology
Olhoff, N., Du, J., 2013E. Topological design for minimum sound emission from
Olhoff, N., Niu, B., 2013F. Discrete Material Optimization of Vibrating Laminated
Composite Plates for Minimum Sound Emission. In: Rozvany, G., Lewinski, T., Eds.,
Topology Optimization in Structural and Continuum Mechanics, Int. Centre for
Mechanical Sciences, Udine, Italy, June 18-22, 2012. Springer-Verlag, Vienna, 2013,
29 pp.
Overton, M.L., 1988. On minimizing the maximum eigenvalue of a symmetric matrix.
SIAM J. Matrix Anal Appl 9(2): 256-268.
Pedersen, N.L., 2000. Maximization of eigenvalues using topology optimization. Struct

Multidisc Optim 20. 2-11.
Rozvany, G., Zhou, M., 1991. Applications of the COC method in layout optimization.
In Eschenauer, H.; Mattheck, C.; Olhoff, N., Eds., Proc. Int. Conf. on Engineering
Optimization in Design Processes (held in Karlsruhe 1990), 59-70. Berlin,
Heidelberg, New York: Springer.
Rozvany, G., Zhou, M., Birker, T., 1992. Generalized shape optimization without
homogenization. Struct Optim 4: 250-252.
Seyranian, A.P., 1987. Multiple eigenvalues in optimization problems. Appl Math Mech
51: 272-275.
Seyranian, A.P., Lund, E., Olhoff, N., 1994. Multiple eigenvalues in structural
optimization problems. Struct Optim 8(4): 207-227.
Shu, L., Wang, M.Y., Fang, Z., Ma, Z., Wei, P., 2011. Level set based structural
topology optimization for minimizing frequency response. J. Sound and Vibration
330(24): 5820-5834.
Sigmund, O., 1997. On the design of compliant mechanisms using topology
optimization. Mech Struct Mach 25: 493-524.
Sigmund, O., 2001. Microstructural design of elastic band gap structures. In: G.D.
Cheng, et al., (Eds.), Proceedings of the 4th World Congress of Structrual and
Multidisciplinary Optimization WCSMO4, Dalian, China, Liaoning Electronic Press, 6
pp.
Sigmund, O., Jensen, J.S., 2003. Systematic design of phononic band-gap materials and
structures by topology optimization. Philosophical Transactions of the Royal Society
London, Series A (Mathematical, Physical and Engineering Sciences) 361: 1001-1019.
Sorokin, S.V., 2005. Private communication.
Stegmann J., Lund E., 2005. Discrete material optimization of general composite shell
structures. International Journal for Numerical Methods in Engineering 62: 2009-
2027.
Svanberg, K., 1987. The method of moving asymptotes a new method for structural
optimization. Int J Numer Meth Engng 24: 359-373.
Taylor, J.E., Bendsøe, M.P., 1984. An interpretation of min-max structural design
problems including a method for relaxing constraints. Int J Solids Struct 20: 301-314.
Tcherniak, D., 2002. Topology optimization of resonating structures using SIMP
method. Int J Numer Meth Engng 54: 1605-1622.
Thamburaj, P., Sun, J.Q., 2002. Optimization of anisotropic sandwich beams for higher
sound transmission loss. J Sound Vib 254(1): 23-36.
Tortorelli, D., Michaleris, P., 1994. Design sensitivity analysis: overview and review.
Inverse Problems in Engineering 1: 71-105.
Wadbro E., Berggren M., 2006. Topology optimization of an acoustic horn. Comput
Methods Appl Mech Eng 196: 420-436.
Wittrick, W.H., 1962. Rates of change of eigenvalues, with reference to buckling and
vibration problems. Journal of the Royal Aeronautical Society 66: 590-599.
Yamamoto, T., Maruyama, S., Nishiwaki, S., Yoshimura, M., 2008. Thickness
optimization of a multilayered structure on the coupling surface between a structure
and an acoustic cavity. J Sound Vib 318: 109-130.
Yamamoto, T., Maruyama, S., Nishiwaki, S., Yoshimura, M. 2009. Topology design of
multi-material soundproof structures including poroelastic media to minimize sound

pressure levels. Computer Methods in Applied Mechanics and Engineering 198(17-
20): 1439-1455.
Yang, R., Du, J., 2013. Microstructural topology optimization with respect to sound
power radiation. Struct Multidisc Optim 47(2): 191-206.
Yang, X., Li, Y., 2013 Topology optimization to minimize the dynamic compliance of a
bi-material plate in a thermal environment. Struct Multidisc Optim 47(3): 399-408.
Yoon, G.H., Jensen, J.S., Sigmund, O., 2007. Topology optimization of acoustic-
structure interaction problems using a mixed finite element formulation. Int J Numer
Methods Eng 70: 1049-1075.
Yoon, G.H., 2010a. Maximizing the fundamental eigenfrequency of geometrically
nonlinear structures by topology optimization based on element connectivity
parameterization. Computers & Structures 88(1-2): 120-133.
Yoon, G.H., 2010b. Structural topology optimization for frequency response problem
using model reduction schemes. Computer Methods in Applied Mechanics and
Engineering 199(25-28): 1744-1763.
Structural Topology Optimization with Respect to
Eigenfrequencies of Vibration

1
2
School of Aerospace, Tsinghua University, Beijing 100084, P.R. China
Abstract A frequent goal of the design of vibrating structures is to avoid resonance of the
structure in a given interval for external excitation frequencies. This can be achieved by,
e.g., maximizing the fundamental eigenfrequency, an eigenfrequency of higher order, or
the gap between two consecutive eigenfrequencies of given order, subject to a given
amount of structural material and prescribed boundary conditions. Mathematical
formulations and methods of numerical solution of these topology optimization problems
are presented for linearly elastic structures without damping in this paper, and several
illustrative results are shown.
1 Introduction
Methods for optimization of simple (unimodal) eigenvalues/eigenfrequencies in
shape and sizing design problems are well established and can be implemented
directly in topology optimization. The formulation for topology optimization
with respect to a simple, fundamental eigenfrequency is presented in Sub-section
2.1, and the sensitivity analysis of a simple eigenfrequency subject to change of a
design variable Ue is outlined in Sub-section 2.2. However, particularly in
topology optimization it is often found that, although an eigenfrequency is
simple during the initial stage of the iterative design procedure, later it may
become multiple due to coincidence with one or more of its adjacent
eigenfrequencies. In order to capture this behaviour, it is necessary to apply a
more general solution procedure that allows for multiplicity of the
eigenfrequency because a multiple eigenfrequency does not possess usual
differentiability properties.
In Sub-section 2.3, the abovementioned eigenfrequency optimization
problems are conveniently formulated by a so-called bound formulation

DOI 10.1007/978-3-7091-1643-2_11, © CISM, Udine 2014
(Bendsøe et al. 1983, Taylor and Bendsøe 1984, Olhoff 1989). Sub-section 2.4
then presents design sensitivity results for multiple eigenvalues derived in
(Seyranian et al. 1994, and Du and Olhoff 2007), and by usage of these results,
the problems can be solved efficiently by mathematical programming (see, e.g.,
Overton 1988, and Olhoff 1989) or by the MMA method (Svanberg 1987).
Moreover, the procedure of treating the multiple eigenvalues can be greatly
simplified by using the increments of the design variables as unknowns (see
Krog and Olhoff 1999, and Du and Olhoff 2007). Sub-section 2.5 presents the
iterative, numerical solution procedure which is developed such that it is
applicable independently of whether the subject eigenfrequencies are uni- or
multimodal.
Finally, Section 3 presents several numerical examples of topology
optimization of single- and bi-material beam- and plate-like structures, and
Section 4 concludes this paper, which lends itself to (Du and Olhoff 2007).
2 Eigenfrequency Optimization Problems
2.1 Maximization of the fundamental eigenfrequency

Problems of topology design for maximization of fundamental eigenfrequencies
of vibrating elastic structures have, e.g., been considered in the papers (Diaz and
Kikuchi 1992, Ma et al. 1994, 1995, Diaz et al. 1994, Kosaka and Swan 1999,
Krog and Olhoff 1999, Pedersen 2000). Assuming that damping can be neglected,
such a design problem can be formulated as a max-min problem as follows,
max { min {Z 2j }}
U1,, UN E j 1,J (1a)
Subject to :
Kĳ j Ȧ 2j Mĳ j , j 1, , J , (1b)
ĳ Tj Mĳ k G jk , j t k, k , j 1, , J , (1c)
NE
¦U V e e V * d 0 , V * DV0 , (1d)
e 1
0 U d Ue d 1 , e 1, , N E . (1e)
Here Zj is the j-th eigenfrequency and Mj the corresponding eigenvector, and

K and M are the symmetric and positive definite global stiffness and mass
matrices of the finite element based, generalized structural eigenvalue problem in
Structural Topology Optimization… 277
the constraint (1b). The J candidate eigenfrequencies considered will all be real
and can be numbered such that
0 Z1 d Z2 d d Z J , (2)
and it will be assumed that the corresponding eigenvectors are M-
orthonormalized, cf. (1c) where Gjk is Kronecker’s delta. In problem (1a-e), the
symbol NE denotes the total number of finite elements in the admissible design
domain. The design variables Ue, e = 1,…,NE, represent the volumetric material
densities of the finite elements, and (1e) specify lower and upper limits U and 1
for Ue. To avoid singularity of the stiffness matrix, U is not zero, but taken to be
a small positive value like U = 10-3. In (1d), the symbol D defines the volume
fraction V * /V0 , where V0 is the volume of the admissible design domain, and V *
the given available volume of solid material and of solid material *1,
respectively, for a single-material and a bi-material design problem, cf. Sub-
sections 2.1 and 2.3 in the preceding paper Olhoff and Du (2013A).
2.2 Sensitivity analysis of a simple eigenfrequency

If the jth eigenfrequency Zj is simple (also called unimodal or distinct), i.e.,
Z j 1 Z j Z j 1 , then the corresponding eigenvector ĳ j will be unique (up to a
sign) and differentiable with respect to the design variables Ue, e = 1,…,NE. To
determine the sensitivity (derivative) (O j )cUe of the eigenvalue O j Z 2j with
respect to a particular design variable Ue, we differentiate the vibration equation
(1b) with respect to Ue, and get
(K O j M ) (ĳ j )cUe (K cUe O j M cUe (O j )cUe M ) ĳ j 0 , e 1, , N E (3)
T
where ( )cUe w ( ) wU e . Pre-multiplying (3) by ĳ j and using the vibration
equation (1b) and the normalization of ĳ j included in (1c) then gives (see also
Wittrick 1962, Lancaster 1964, or Haftka et al. 1990),
(O j )cUe ĳ Tj (K cUe O j M cUe ) ĳ j , e 1,, N E (4)
The derivatives of the matrices K and M can be calculated explicitly from the
material models in Section 2 of the preceding paper Olhoff and Du (2013A).
Considering, e.g., the single-material model in Eq. (3) of that paper, the
sensitivity of the eigenvalue O j Z 2j with respect to the design variable U e
becomes
(O j )cUe ĳ Tj ( pU e( p 1) K *e O j qU e( q 1) M *e ) ĳ j , e 1, , N E (5)

The optimality condition for the maximization of a unimodal eigenvalue
O j Z 2j of given order j, j = 1, 2, …, now follows from (4) (or (5)) and usage of
the Lagrange multiplier method, and takes the form
(O j )cUe J 0Ve 0 , e 1, , N E (6)
where J0 (t 0) is the Lagrange multiplier corresponding to the material volume
constraint, and the side constraints for Ue have been ignored. With this sensitivity
result and optimality condition, the design problem (1a-e) may be solved for a
unimodal optimum eigenfrequency by using an OC (Optimality Criterion) based
method, e.g., the fixed point method (see Cheng and Olhoff 1982), or a
mathematical programming method, e.g., MMA (Svanberg 1987).
We may also wish to apply a gradient based method of solution. It is then
essential that the jth eigenvalue O j Z 2j is simple and differentiable, and hereby
admits linearization with respect to the design variables Ue, e = 1, …, NE. Hence,
if all the design variables are changed simultaneously, the linear increment 'O j
of O j Z 2j is given by the scalar product
'O j OTj ǻȡ (7)
where ǻȡ ^'U ,, 'U `

1 NE
T
is the vector of changes of the design variables, and
O j ^ĳ T
j (K cU1 O j M cU1 )ĳ j ,, ĳ Tj (K cU N O j M cU N )ĳ j
E E
`T
(8)
is the vector of sensitivities (or gradients) of the eigenvalue O j with respect to

the design variables Ue, e = 1, …, NE.
2.3 Bound formulations for maximization of the n-th eigenfrequency or

the distance between two consecutive eigenfrequencies
In this section, we first consider the more general problem of maximizing the n-
th eigenfrequency Zn of given order of a vibrating structure (cf. Olhoff 1976,
1977), i.e., the fundamental eigenfrequency (n = 1) or a higher order
eigenfrequency (n > 1). Employing a bound formulation (Bendsøe et al. 1983,
Taylor and Bendsøe 1984, and Olhoff 1989) involving a scalar variable E which
plays both the role of an objective function to be maximized and at the same time
a variable lower bound for the n-th and higher order eigenfrequencies (counted
with possible multiplicity), the above problem can be formulated as
max {E }
E , U1,, U N E (9a)
Subject to :
E Z 2j d 0 , j n, n 1,, J , (9b)
Constraints:1(b-e) (9c)
Here, as well as in Eqs. (10) below, J is assumed to be larger than the highest
order of an eigenfrequency to be considered a candidate to exchange its order
with the n-th eigenfrequency or to coalesce with this eigenfrequency during the
design process.
The problem of maximizing the distance (gap) between two consecutive
eigenfrequencies of given orders n and n – 1 with n > 1, (see Olhoff 1976, Olhoff
and Parbery 1984, Bendsøe and Olhoff 1985, Jensen and Pedersen 2005, Olhoff
et al. 2012) may be written in the following extended bound formulation, where
two bound parameters are used:
max {E 2 E1} (10a)
E 1, E 2, U1,, U N E
Subject to :
E 2 Z 2j d 0 , j n, n 1, , J , (10b)
Z 2j E1 d 0 , j 1, , n 1, (10c)
Constraints: 1(b-e). (10d)
Note that if in (10) we remove the bound variable E1 and the corresponding set of
constraints (10c) from the formulation, then the eigenfrequency gap
maximization problem (10) reduces to the n-th eigenfrequency maximization
problem (9), and in particular, for n = 1, to the problem of maximizing the
fundamental eigenfrequency in (1).
In problem (9) the eigenfrequency Zn , and in problem (10) both the
eigenfrequencies Zn and Zn1 of the optimum solution may very well be
multiple, and the bound formulations in (9) and (10) are tailored to facilitate
handling of such difficulties.
It is also worth noting that the introduction of the scalar bound variables E
in (9) and E1 and E 2 in (10) implies that even if multiple eigenfrequencies are
present, the optimization problems (9) and (10) are both differentiable if they are
considered as problems in all variables, i.e. the bound parameter(s) E (or
E1 , E 2 ), design variables U e , e 1,, N E , as well as the eigenfrequencies Z j
and eigenvectors ĳ j , j 1,, J , (implying that all these variables should have
been included under the ‘max’ signs in (9a) and (10a)). This type of problem is
referred to as one of “Simultaneous analysis and design” (SAND), and is a very
large problem in the present context. Therefore, we refrain from solving the
current topology optimization problems in this form.
In the form written above, where only the design variables U e , e 1,, N E ,
and the bound parameters E and E1 , E 2 are included under the ‘max’ signs in
(9a) and (10a), the topology optimization problems (9) and (10) are non-
differentiable because the eigenfrequencies Z j , j 1,, J , are considered as
functions of the design variables U e , e 1,, N E . This is a ‘nested’ formulation
which provides the basis for numerical solution by a scheme of successive
iterations where, in each iteration, the eigenfrequencies Z j and eigenvectors ĳ j ,
j 1,, J , are established for known design, U e , e 1,, N E , by solution of the
generalized eigenvalue problem (1b) and implementation of the orthonormality
conditions (1c).
To accommodate for occurrence of multiple eigenfrequencies, we in the
subsequent Sub-section 2.4 consider some important sensitivity results for such
eigenfrequencies. In Sub-section 2.5, we make use of these results in the
development of incremental forms of problems (9) and (10) which provide the
basis for construction of a highly efficient scheme for numerical solution of the
topology optimization problems under study.
2.4 Sensitivity analysis of multiple eigenfrequencies

Multiple eigenfrequencies may manifest themselves in different ways in
structural optimization problems. One possibility is that an eigenfrequency
subject to optimization is multiple from the beginning of the design process, e.g.,
due to structural symmetry, but an originally unimodal eigenfrequency may also
become multiple during the optimization process due to coalescence with one or
more of its adjacent eigenfrequencies. In this case, sensitivities of the multiple
eigenfrequency cannot be calculated straightforwardly from (4) (or (5)) due to
lack of usual differentiability properties of the sub-space spanned by the
eigenvectors associated with the multiple eigenfrequency. Investigations of
sensitivity analysis of multiple eigenvalues (like eigenfrequencies or buckling
loads) are available in many papers (see, e.g., Bratus and Seyranian 1983, Masur
1984, 1985, Haug et al. 1986, Overton 1988, Seyranian et al. 1994, and papers
cited therein).
Following Seyranian et al. (1994) and Du and Olhoff (2007), let us assume
that the solution of the generalized eigenvalue problem (1b) included in

~
problems (9) or (10) yields a N-fold multiple eigenvalue O ,
~
O O j Z 2j , j n, , n N 1 (11)
associated with the N (N > 1) lowest eigenfrequencies Z j appearing in the bound
constraints (9b) and (10b) *1. Here we shall assume n N 1 J , i.e., that the
total number J of eigenfrequencies (counted with multiplicity), that is considered
in problems (9) and (10) is chosen such that the J-th eigenfrequency ZJ is larger
~
than the multiple eigenfrequency corresponding to O in (11). The multiplicity of
~
the eigenvalue O in (11) implies that any linear combination of the eigenvectors
~
ĳ j , j n,, n N 1 , corresponding to O will satisfy the generalized
eigenvalue problem (1b) in (9) and (10), which implies that the eigenvectors are
not unique.
In Seyranian et al. (1994) and Du and Olhoff (2007), the sensitivity analysis
is based on a mathematical perturbation analysis of the multiple eigenvalue and
the corresponding eigenvectors. This analysis involves directional derivatives in
the design space and leads to the result that the increments 'O j of a multiple
~
eigenvalue O O j Z 2j , j n,, n N 1 , as in (11) are eigenvalues of a N-
dimensional algebraic sub-eigenvalue problem of the form
>
det f skT ǻȡ G sk ǻO @ 0 , s, k = n, …, n+N-1, (12)
where Gsk is Kronecker’s delta, and fsk denote generalized gradient vectors of the
form
~ ~
^ `T
f sk ĳ Ts (K cU1 O M cU1 )ĳ k , , ĳ Ts (K cU N O M cU N )ĳ k , s, k = n, …, n+N-1. (13)
E E
According to the definition in (13), each fsk is a NE-dimensional vector, which

means that f skT ǻȡ in (12) is a scalar product. The label ‘generalized gradient
vector’ for fsk becomes apparent when comparing (13) with the expression for the
gradient vector O j of a simple eigenvalue O j in (8). Note also that fsk = fks due
to the symmetry of the matrices K and M, and that the two subscripts s and k
refer to the orthonormalized eigenmodes from which fsk is calculated.
*1
Similarly, the eigenvalue problem (1b) contained in problem (10) may yield
another R-fold eigenvalue Oˆ O Z 2 , j n R,, n 1 , which corresponds to
j j
the R largest eigenfrequencies Z j in (10c). This case (for which we assume that
1 d n R ), is completely analogous to (11).
~
Assuming that we know the multiple eigenvalue O , the associated sub-set of
orthonormalized eigenmodes, and have computed the derivatives of the matrices
K and M, we can construct the generalized gradient vectors fsk, s, k = n, …, n+N-
1, from (13). Solving the algebraic sub-eigenvalue problem in (12) for 'O then
yields the increments 'O 'O j , j n,, n N 1 , of the multiple eigenvalue
~
^ `
O subject to a given vector ǻȡ 'U1 , , 'U N E of increments of the design
variables.
The N increments 'O j , j n,, n N 1, constitute the eigenvalues of the
sub-eigenvalue problem (12), and represent the directional derivatives of the
~
multiple eigenvalue O O j Z 2j , j n,, n N 1 , with respect to change
'U e of the design variables U e , e 1, , N E . Attention should be drawn to the
fact that the increments 'O j , j n,, n N 1 of the multiple eigenvalue are
generally non-linear functions of the direction of the design increment vector ǻȡ .
Thus, unlike simple eigenvalues, multiple eigenvalues do not admit a usual
linearization in terms of the design variables.
Finally, two important special cases should be observed.
Case of simple eigenfrequency. As is to be expected, for N = 1, i.e., j = s = k =

n, (11) and (12) reduce to the case of a simple eigenvalue On Zn2 . Eq. (12)
reduces to the simple equation
f nnT ǻȡ ǻOn 0 (14)
where, according to (7), (13) and (8), we have
f nn On (15)
i.e., f nn is simply the vector of sensitivities of the unimodal eigenvalue On with
respect to the design variables Ue, e = 1, …, NE, cf. (4) and (8).
Case of vanishing off-diagonal terms. For the case of multiple eigenvalues, cf.
(17) with N > 1, a very important observation can be made. If in (12) all off-
diagonal scalar products are zero, i.e. if
f skT ǻȡ 0 , s z k , s, k n, , n N 1, (16)
then the increment 'O j of an eigenvalue O j Z 2j becomes determined as
'O j f jjT 'ȡ , j n,, n N 1, (17)
where according to (11) and (13)

f jj ^ĳ (K c
T
j U1 E E
`T
O j M cU1 )ĳ j , , ĳ Tj (K cU N O j M cU N )ĳ j , j n,, n N 1. (18)
Hence, if the design increment vector ǻȡ fulfils (16), then f jj has precisely the
same form as the gradient vector O j in (8) for a simple eigenvalue, and the
eigenvalue increments 'O j in (17) are uniquely determined on the basis of the
eigenmodes ĳ j , j n,, n N 1 . The formulas for design sensitivity analysis
of multiple eigenvalues then become precisely the same as those for simple
eigenvalues.
2.5 Computational procedure

The topology optimization problems (9) and (10) can be efficiently solved by
an iterative procedure indicated in Fig. 1, which can be used for solution of
problems with multiple as well as simple eigenfrequencies.
The procedure is based on the results of the sensitivity analysis in the
preceding section, and is seen to consist of a main (outer) loop and an inner loop.
While steps 1, 2 and 4 of the main loop are pretty straight-forward, the third step
(the inner loop) needs to be briefly discussed. (The interested reader is referred to
Du and Olhoff (2007) for more details about the iterative procedure.)
The purpose of the third step (the inner loop) in Fig. 1 is to determine
optimum values of the increments 'U e , e 1, , N E , of the design variables,
subject to known values of iterates that have been determined in steps 1 and 2
and are fixed in the third step. To enable this, we rewrite the bound formulations
(9) and (10) in terms of the vector ǻȡ of increments 'U e , e 1, , N E , of the
design variables and corresponding increments of the squared eigenfrequencies
'O j '(Z 2j ), j n,, n N 1 , (and j n R,, n 1 , for problem (10)).
Hereby, we obtain the following sets of sub-problems to be solved for
optimum increments in the third step of the main loop of the computational
procedure for
0. Problem initialization.
Define value of n and
initialize design variables U e
1. Solution of the generalized eigenvalue problem (1b,c)

for eigenfrequencies and -modes by FE-analysis.
Detect possible multiplicity N of Zn (and R of Z n1 )
2. Computation of generalized gradients fsk, if N>1 (and

R>1) or usual gradients if N=1 (and R=1)
3. Iterative solution of optimization sub-problem (19)

(or (20)) for increments 'U e of the design variables
Main
Inner loop loop
Increments No
'U e converged ?
Yes
4. Update values of the design variables

U e : U e 'U e .
U e converged ? No
i.e., 'ȡ H ?
Yes
Stop
Figure 1. Flow chart of iterative solution procedure.

(a) Maximization of the n-th eigenfrequency:
max {E } (19a)
E , 'U1 , , 'U N E
Subject to :
E ª¬Z 2j f jjT ǻȡ º¼ d 0 , for j J n N, (19b)
E ª¬Z 2j '(Z 2j ) º¼ d 0 , j n,..., n N 1, (19c)
det ª¬f skT ǻȡ G sk ǻ(Z 2 ) º¼ 0, s, k n,..., n N 1, (19d)

NE
'Ue )Ve V * d 0, V * (19e)
¦ (U
e 1
e D V0 ,
0 U d Ue 'Ue d 1, e 1, , N E , (19f)
(b) Maximization of the gap (distance) between the n-th and (n-1)-th
eigenfrequencies:
max {E 2 E1}
E 1, E 2, 'U1,, 'U N E (20a)
Subject to :
>
E 2 Z 2j f Tjj ǻȡ d 0 , @ for j J n N, (20b)
E2 >Z 2
j @
' (Z 2j ) d 0 , j n, ..., n N 1, (20c)
>Z 2
j @
'(Z 2j ) E1 d 0 , j n R, ..., n 1, ( R d n 1) (20d)
>Z 2
j @
f Tjj ǻȡ E1 d 0 , for j n R 1, (if R d n 2) (20e)
>
det f skT ǻȡ G sk ǻ(Z 2 ) @ 0, s, k n, ..., n N 1, (20f)
det >f T
sk ǻȡ G sk ǻ (Z
2
)@ 0, s, k n R, ..., n 1, (20g)
NE
¦ (U e 'U e )Ve V * d 0, V * D V0 , (20h)

e 1
0 U d U e 'U e d 1, e 1, , N E . (20i)
Note that in the sub-optimization problems (19) and (20), the only unknowns
are the bound variables E and E1 , E 2 and the increments of the design
variables 'U e , e 1, , N E , which play the role as independent variables. The

dependent variables are the increments '(Z 2j ), j n,, n N 1 , of the N-fold
eigenfrequency Z n2 Z n2 N 1 (in problems (19) and (20)), together with the
increments '(Z 2j ), j n R, , n 1 , of the R-fold eigenfrequency
2 2
Z Z (in problem (20)). All other iterates in (19) and (20), i.e. the
n R n 1
material valume densities U e , the eigenfrequencies Z j , the generalized gradient

vectors f sk and the multiplicities N and R have been determined in step 1 and 2
of the main iteration loop, and are kept fixed in the current step 3 of this loop.
Problems (19) and (20) can be solved using the MMA method (Svanberg
1987) or a linear programming algorithm.
Finally, it is interesting to note that if we introduce the additional constraints
T
f sk ǻȡ 0 , for s z k , s, k = n, …, n+N-1, i.e. force the off-diagonal terms in (12)
to vanish, then the increments 'O j are determined in a linearized form with
respect to the increments 'U e of the material volume densities for both simple
and multiple eigenvalues, and as a result, the sub-optimization problems (19) and
(20) both reduce to linear programming problems (see Krog and Olhoff 1999).
3 Numerical Examples of Eigenfrequency Optimization
3.1 Maximization of the fundamental eigenfrequency of beam-like 2D

structures
As a first example, we consider the topology optimization of a single-material
beam-like structure modeled by 2D plane stress elements. The admissible design
domain is specified, and three different cases (a), (b) and (c) of boundary
conditions as shown in Fig. 2 and defined in the caption, are considered. The
design objective is to maximize the fundamental eigenfrequency for a prescribed
material volume fraction D = 50%, and in the initial design the available material
is uniformly distributed over the admissible design domain. The material is
isotropic with Young’s modulus E = 107, Poisson’s ratio X = 0.3 and mass
density Um = 1 (SI units are used throughout).
The fundamental eigenfrequencies of the initial designs with the three cases
(a), (b) and (c) of boundary conditions are given in the caption of Fig. 2.
a
b Admissible design domain
(a)
Admissible design domain
(b)
Admissible design domain
(c)
Figure 2(a-c). Admissible design domains (a = 8, b = 1) of beam-like 2D structures with

three different sets of boundary conditions. (a) Simply supported ends. (b) One end
clamped, the other simply supported. (c) Clamped ends. The fundamental
eigenfrequencies of the 3 initial designs (uniform distribution of material with density U =
0.5) are all unimodal with values Z10a 68.7 , Z10b 104.1 and Z10c 146.1 .
The optimized topologies are shown in Figs. 3(a-c), and the corresponding
optimum fundamental eigenfrequencies are all found to be bimodal with values
given in the caption of the figure. Fig. 4 shows the iteration history for the first 3
eigenfrequencies of the optimum bimodal design with simply supported ends in
Fig. 3(a). The iteration histories for the optimum designs with the two other
cases of boundary conditions in Figs. 3(b,c) are qualitatively similar. Figs. 5(a-c)
depict the first 3 eigenmodes of the optimized beam-like structure with simply
supported ends in Fig. 3(a), and the results show that the first 2 eigenmodes
(corresponding to the bimodal fundamental eigenfrequency) of the structure are
typical simply supported beam-type vibration modes, while the 3rd one is a more
general 2D vibration mode.
(a)
(b)
(c)
Figure 3(a-c). Optimized single-material topologies (50% volume fraction) for the three
different sets of boundary conditions defined in Figs. 2(a-c). The optimum fundamental
eigenfrequencies are all found to be bimodal and have the values (a) Z1opt
a 174.7 , (b)
Z1opt
b 288.7 , and (c) Z1opt
c 456.4 , implying that they are increased by (a) 154%, (b)
177% and (c) 212% relative to the initial designs.
600
500
Eigenfrequencies
400
Z3
300
200 Z2
(Maximized) Z1
100
0
0 20 40 60 80
Iteration number
Figure 4. Iteration history of the first 3 eigenfrequencies associated with the design
process leading to the optimum simply supported beam-like structure in Fig. 3(a). It is
seen that the fundamental eigenfrequency is simple for the initial design, but soon
coalesces with the second eigenfrequency, and the maximum fundamental eigenfrequency
is bimodal.
(a) Z1opt
a 174.7
(b) Z 2 a Z1opt
a 174.7
(c) Z3a 284.9

Figure 5(a-c). The three first eigenmodes of the simply supported beam-like structure in
Fig. 4(a) with a bimodal optimum fundamental eigenfrequency. (a) and (b) depict the two
modes associated with the optimum fundamental eigenfrequency, and (c) shows the
subsequent mode.
3.2 Maximization of the second eigenfrequency of beam-like 2D

structures
We now present an example of topology optimization of single material beam-
like structures for maximum value of the second eigenfrequency. The initial data
and the three sets of boundary conditions in this example are the same as for the
first example in Sub-section 3.1. The resulting topologies are shown in Figs. 6(a-
c).
(a)
(b)
(c)
different sets of boundary conditions in Figs. 2(a-c). The optimum second
eigenfrequencies are found to be (a) Z2opta 598.3 , (b) Z2optb 732.8 , and (c) Z2optc 849.0 ,
and are all bimodal.
3.3 Maximization of the distance (gap) between two consequtive

eigenfrequencies of beam-like 2D structure
In this example, we consider the design objective of maximizing the distance
(gap) between two consecutive eigenfrequencies (the 2nd and the 3rd
eigenfrequencies) of the clamped beam-like structure in Fig. 7(a). A concentrated
mass mc is attached at the mid-point of the lower edge of the beam-like structure
as shown in Fig. 7(a), which has the value mc = 1/2mb (Here mb is the total mass
of the initial design). We use the same admissible design domain, materials and
volume fractions as in the previous example (see Fig. 2(c)). The optimum
topology and the corresponding iteration histories of the eigenfrequencies are
given in Figs. 7(b) and 7(c). It can be seen that the 2nd eigenfrequency is
decreased and the 3rd eigenfrequency is increased. As a result, the design ends
up with a maximized gap between the 2nd and the 3rd eigenfrequencies that is
equal to 810, which is 548 % higher than the difference between the
corresponding eigenfrequencies of the initial design. Note that in Fig. 7(c) the
3rd, 4th and 5th eigenfrequencies form a tri-modal eigenfrequency of the final
optimized design.
(a)
(b)
1200
Z5
1000
Z4
Eigenfrequencies
Z3
800
600
(Maximized) gap: ZZ

3 2
400
Z2
200
Z1
0
0 20 40 60 80 100
Iteration number
(c)
Figure 7. (a) Clamped beam-like 2D structure with a concentrated mass attached at the
mid-point of the lower edge. (b) Optimized topology of the beam-like structure. The gap
between the 2nd and the 3rd eigenfrequencies is maximized. (c) Iteration history for the
first five eigenfrequencies associated with the process leading to the optimized topology.
Notice that the 3rd, 4th and 5th eigenvalues have coalesced to a tri-modal eigenfrequency
for the optimized topology.
3.4 Maximization of the fundamental eigenfrequency of single-material

plate structures
In this example, we consider the topology optimization of a single-material plate-

like structure modeled by 8-node 3D brick elements with Wilson incompatible
displacement models to improve precision. The admissible design domain is
specified, and three different cases (a), (b) and (c) of boundary conditions and
attached concentrated, nonstructural masses as shown in Fig. 8 and defined in the
caption, are considered. The design objective is to maximize the fundamental
eigenfrequency for a prescribed material volume fraction D = 50%, and in the
initial design the available material is uniformly distributed over the admissible
design domain. The material is isotropic with Young’s modulus E = 1011,
Poisson’s ratio X = 0.3 and mass density Um = 7800 (SI units are used throughout).
The fundamental eigenfrequencies of the initial designs with the three cases (a),
(b) and (c) of boundary/mass conditions are given in the caption of Fig. 8. The
optimized plate topologies are shown in Figs. 9(a-c), and the corresponding
optimum fundamental eigenfrequencies are all unimodal with values given in the
caption of Fig. 9.
(a) (b) (c)

Figure 8. Plate-like 3D structure (a=20, b= 20 and t=1) with three different cases of
boundary conditions and attachment of a concentrated nonstructural mass. (a) Simple
supports at four corners and concentrated mass mc at the center of the structure
( mc m0 / 3 , m0 the total structural mass of the plate). (b) Four edges clamped and
concentrated mass mc at the center ( mc m0 / 10 ). (c) One edge clamped, other edges
free, and concentrated mass mc attached at the mid-point of the edge opposite to the
clamped one ( mc m0 / 10 ). The first eigenfrequencies for the 3 initial designs (uniform
distribution of material with density U = 0.5) are Z10a 8.13 , Z10b 31.07 , Z10c 3.46 .
(a) (b) (c)

different cases of boundary conditions and mass attachment in Figs. 8(a-c). The optimum
fundamental eigenfrequencies are found to be (a) Z1opt
a 16.38 , (b) Z1opt
b 65.42 , and (c)
Z1opt
c 9.66 , implying that they are increased by (a) 101%, (b) 111% and (c) 179%
relative to the initial designs.
As a second example, single-material topology optimization of an initially

quadratic plate-like structure with simple supports at its four corners and center
is considered (Fig. 10(a)). The admissible design domain and the material are the
same as in the foregoing example. Due to the structural symmetry, the
fundamental eigenfrequency of the initial design is bimodal with modes shown
in Figs. 10(b-c).
(a) (b) (c)
Figure 10. Plate-like 3D structure (a=20, b= 20 and t=1) with simple supports at its four
corners and center. (a) Admissible design domain. (b-c) The eigenmodes of the initial
design associated with the bimodal fundamental eigenfrequency Z10 Z20 24.56 .
The optimized topology is shown in Fig. 11(a) (50% volume fraction), and
the corresponding optimum fundamental eigenfrequency is also bimodal.
90
Z3
80
Eigenfrequency
70
Z2
60
Z1
50 (Maximized)
40
30 Multiple eigenfrequency
20
0 10 20 30 40 50
Iteration number
(a) (b)
Figure 11. (a) Optimized topology (50% volume fraction, single-material design)
associated with the maximum fundmental eigenfrequency Z1opt 60.32 , which is
bimodal. (b) Iteration history for the first three eigenfrequencies.
3.5 Maximization of higher order eigenfrequencies of single- and bi-

material plate structures
Here we first present an example of topology optimization of single-material
plate-like structures with respect to the second eigenfrequency. The initial data
for the example are the same as for the first example in Section 3.4. Thus, we
choose the same volume and type of available material, the same admissible
design domain, and again consider the three different cases (a), (b) and (c) of
boundary conditions and attached concentrated masses as shown in Fig. 8, but
we now maximize the second eigenfrequencies. The resulting optimum
topologies and the frequency iteration histories for the three cases of boundary
conditions and mass attachment in Fig. 8 are given in Figs. 12 and 13.
(a) (b) (c)

Figure 12 (a-c). Optimized single-material topologies (50% volume fraction)
corresponding to the three different cases of boundary conditions and mass attachment in
Figs. 8(a-c). The values and multiplicities of the optimum second eigenfrequencies are:
(a) Z2opta 46.03 (trimodal), (b) Z2optb 155.43 (bimodal), (c) Z2optc 39.77 (bimodal).
60 250 120
Z5 Z4 Z5
50 Z4 100
200 Z4
Eigenfrequencies
Eigenfrequencies
Eigenfrequencies
Z3
40 (Maximized) Z2 Z3 80
150 Z2
30 (Maximized) 60
100 Z3
20 40
Z1 Z2
Z1 (Maximized)
50
10 20
Z1
0 0 0
0 20 40 60 80 0 20 40 60 80 0 20 40 60 80
Iteration number Iteration number Iteration number
(a) (b) (c)

Figure 13. Iteration histories of eigenfrequencies associated with the design process
leading to the results in Figs. 12(a-c). For case (c) it is seen that the second
eigenfrequency is simple for the initial design, but soon coalesces with the third
eigenfrequency.
In the next example, we consider topology optimization of bi-material

structures with respect to higher order eigenfrequencies. Both of the two
materials are isotropic. The stiffer material *1 with elasticity and mass matrices
E*e1 , M *e1 (see Sub-section 2.3 in the preceding paper Olhoff and Du 2013A), is
indicated by black in Fig. 14, and is the same as that used for optimization with a
single-material in the preceding examples. The weaker material *2 is indicated
by grey in Fig. 14, and has the properties E*e2 0.1E*e1 and M *e2 0.1M *e1 . We
take the volume fraction of material *1 to be 50%, and present results of
optimizing the topologies of a bi-material quadratic plate with the same
boundary conditions and attachment of a concentrated mass as shown in Fig.
8(b). Figs. 14(a-c) present the optimized plate topologies associated with
maximum values of the 4th, 5th and 6th eigenfrequencies.
(a) (b) (c)

Figure 14. Optimized topologies of bi-material plate with all edges clamped and a
concentrated mass attached to the center, cf. Fig. 8(b). The topologies correspond to
maximum values of the (a) 4th, (b) 5th and (c) 6th eigenfrequency. The stiffer and the
weaker material are indicated by black and grey, respectively, and the volume fraction of
the stiffer material *1 is 50%.
3.6 Maximization of the distance (gap) between two consecutive

eigenfrequencies of bi-material plate structures
This example also concerns topology optimization of bi-material plate structures,

and we use the same materials and volume fractions as in the previous example.
The design objective considered is to maximize the distance (gap) between the
2nd and 3rd eigenfrequencies of the structure. We use the same admissible
design domain as in Fig. 8 for the plate structure, and choose the cases (a) and (c)
of boundary conditions and concentrated mass attachment as shown in Fig. 8.
The results are given in Fig. 15.
80
Z5
70 Z4
Eigenfrequencies
Z3
60
50
(Maximized) Gap: ZZ
3 2
40
30 Z2
20
Z1
10
0 10 20 30 40 50
Iteration number
(a) (b) (c)
Figure 15. Optimized topology of the plate-like structure with simple supports at four
corners and a concentrated mass at the center, cf. Fig. 8(a). The gap between the 2nd and
the 3rd eigenfrequencies is maximized. (b) Iteration histories for the first five
eigenfrequencies associated with the process leading to the design (a). It shows that the
second and the third eigenfrequencies form a double eigenfrequency for the initial design,
but that they split as the design process proceeds, and the 3rd and the 4th eigenfrequencies
end up being a double eigenfrequency of the final design. (c) Optimized topology of the
plate-like structure with the upper horizontal edge clamped, other edges free, and a
concentrated mass attached at the mid-point of the lower horizontal edge, cf. Fig. 8(c).
The gap between the 2nd and the 3rd eigenfrequencies is maximized.
4 Conclusions
Problems of topology optimization with respect to structural eigenfrequencies of
free vibrations were studied and presented in Sections 2 and 3 of this paper. The
design objectives were maximization of eigenfrequencies of given order, and
distances (gaps) between two consecutive eigenfrequencies of the structures.
It was necessary to develop and apply special iterative numerical procedures to
handle topology optimization problems associated with both simple and multiple
eigenfrequencies. Thus, particularly in topology optimization, where the design

space is very large, it is often found that although an eigenfrequency may be
simple during the initial stage of the iterative design procedure, later it may
become multiple due to coincidence with one or more of its adjacent
eigenfrequencies. In order to capture this behaviour, it is necessary to apply an
extended mathematical formulation and solution procedure that allows for
multiplicity of the eigenfrequency because a multiple eigenfrequency – in
contrast to a simple eigenfrequency – does not possess usual differentiability
properties with respect to design.
Several numerical examples of topology optimization of single- and bi-material
beam- and plate-like structures were carried out with the abovementioned design
objectives and validated the approaches presented. The results demonstrate that
multiplicity of optimum eigenfrequencies is the rule rather than the exception in
topology optimization of freely vibrating structures and that this needs careful
attention. The results also indicate that the techniques enable us, in a most cost-
efficient manner, to move structural resonance frequencies far away from external
excitation frequencies with a view to avoid high vibration and noise levels.
References
Bendsøe, M.P., Olhoff, N., Taylor, J.E., 1983. A variational formulation for multicriteria
structural optimization. J Struct Mech 11: 523-544.
beams and shafts. Optim Control Appl Meth 6: 191-200.
Bratus, A.S., Seyranian, A.P., 1983. Bimodal solutions in eigenvalue optimization
problems. Appl Math Mech 47: 451-457.
problems using a homogenization method. Int J Num Meth Engng 35: 1487-1502.
Diaz, A.R., Lipton, R., Soto, C.A. (1994). A new formulation of the problem of optimum
reinforcement of Reissner-Midlin plates. Comp Meth Appl Mechs Eng 123: 121-139.
Du, J., Olhoff, N., 2007. Topological design of freely vibrating continuum structures for
maximum values of simple and multiple eigenfrequencies and frequency gaps. Struct
Multidisc Optim 34:91-110. See also Publisher’s Erratum in Struct Multidisc Optim
(2007) 34:545.
Haftka, R.T., Gurdal, Z., Kamat, M.P., 1990. Elements of Structural Optimization.
Dordrecht: Kluwer.
Haug, E.J., Choi, K.K., Komkov, V. 1986. Design Sensitivity Analysis of Structural
Systems. New York: Academic Press.
On Optimum Design and Periodicity
of Band-gap Structures
Niels Olhoff and Bin Niu

Abstract A band-gap structure usually consists of a periodic distribution

of elastic materials or segments, where the propagation of waves is
impeded or significantly suppressed for a range of external excitation
frequencies. Maximization of the band-gap is therefore an obvious
objective for optimum design. This problem is sometimes formulated by
optimizing a parameterized design model which assumes multiple
periodicity in the design. However, it is shown in the present paper that
such an a priori assumption is not necessary since, in general, just the
maximization of the gap between two consecutive eigenfrequencies leads
to significant design periodicity.
Hence, it is the aim of this paper to apply the method presented in the
preceding paper Olhoff and Du (2013B) to maximize gaps between two
consecutive eigenfrequencies by shape optimization of transversely
vibrating Bernoulli-Euler beams without damping, and to present and
study the associated creation of periodicity in the optimized beam designs.
In the end of the present paper, in order to study the band-gap for
travelling waves, a repeated inner segment of the optimized beams is
analyzed using Floquet theory and the waveguide finite element (WFE)
method. Finally, the frequency response is computed for the optimized
beams when these are subjected to an external time-harmonic loading with
different excitation frequencies, in order to investigate the attenuation
levels in prescribed frequency band-gaps. The results demonstrate that
there is almost perfect correlation between the band-gap size/location of
the emerging band structure and the size/location of the corresponding
eigenfrequency gap in the finite structure.
1 Introduction
A band-gap structure can quench vibrations and significantly suppress the
propagation of waves for a certain range of frequencies. Such a frequency range

DOI 10.1007/978-3-7091-1643-2_12, © CISM, Udine 2014
300 N. Olhoff and B. Niu
is termed a band-gap or stop-band. The phenomenon may occur for elastic,

acoustic or electromagnetic waves (Brillouin 1953, Sigalas and Economou 1992,
Mead 1996, Jensen et al. 2002, Jensen 2003, Hussein et al. 2006a). Except for
regions close to the boundaries, a band-gap structure usually consists of a
periodic distribution of different elastic materials, or repeated identical segments
if a single elastic material is prescribed for the structure. Due to a wealth of
potential applications in vibration protection, noise isolation, waveguiding etc.,
the study and development of band-gap rod, mass-spring, beam, grillage, disk
and plate structures, in most cases by topology optimization, have attracted
increasing attention in recent years, see e.g. (Jensen 2003, Sigmund and Jensen
2003, Halkjær and Sigmund 2004, Diaz et al. 2005, Hussein et al. 2006b,
Halkjær et al. 2006, Jensen and Pedersen 2006, Jensen 2007, Hussein et al. 2007,
Du and Olhoff 2007a,b, Larsen et al. 2009, Søe-Knudsen 2011).
The problem of design optimization of various types of structures for
maximum value of a natural frequency or maximum gap between two adjacent
natural frequencies is extensively studied, see, e.g., Olhoff (1976, 1977), Olhoff
and Parbery (1984), and Bendsøe and Olhoff (1985) for shape optimization of
beam structures, and reference may be given to, e.g., Diaz and Kikuchi (1992),
Jensen and Pedersen (2006), and Du and Olhoff (2007a,b) for topology
optimization of continuum structures, and to Niu et al. (2009) for two-scale
topology optimization of continuum structures with microstructures. Other
references are available in the exhaustive textbook (Bendsøe and Sigmund
2003).
This paper is based on recent work (Olhoff et al., 2012). Instead of
maximizing band-gaps between frequencies of propagating waves or forced
vibration excited by external time-harmonic loads, we consider the closely
related problem of maximizing the gap (also called the separation or difference)
between two adjacent eigenfrequencies) prescribed orders, and demonstrate that
maximization of the frequency gap leads to significant design periodicity.
To the authors’ best knowledge, Olhoff (1976), Olhoff and Parbery (1984)
and Bendsøe and Olhoff (1985) were the first publications on problems of
optimizing vibrating structures for maximum frequency gap – albeit the term
difference (or separation) between adjacent natural frequencies
(eigenfrequencies) was used rather than the term frequency gap in these papers.
As in the current paper, the structures considered in the papers just cited are
thin, elastic, transversely vibrating Bernoulli-Euler beams without damping, and
the problems are considered in non-dimensional form. The beams are subjected
to shape optimization with the cross-sectional area function as design variable,
and no assumption of periodicity is imposed. The cross-sections are assumed to
be geometrically similar (e.g. circular), and the total volume, length and
boundary conditions of the beams are assumed to be given.
Olhoff (1976) considered the problem of optimizing Bernoulli-Euler beams
with any combination of free, simply supported or clamped ends for maximum
On Optimum Design and Periodicity… 301
value of an eigenfrequency Zn of any prescribed order n, n = 1, 2, 3, …., without

specifying a minimum constraint for the variable cross-sectional area, thereby
allowing vanishing cross-section of the optimized beams. The latter implies that
a beam optimized with respect to a higher-order eigenfrequency Zn (n >1) will
turn out to possess n-1 degrees of inner kinematic freedom to perform rigid-body
motions due to the formation of points of vanishing beam cross-section. At these
points, either inner beam separations with both zero shear force and bending
moment, or inner hinges of zero bending moment (but finite shear force) are
created in such a way that all the n-1 rigid-body eigenfrequencies Z1 ,…., Zn 1
below the maximized n-th eigenfrequency Zn reduce to zero.
Thus, without the specification of a minimum cross-sectional area constraint,
Olhoff (1976) presents results that simultaneously constitute solutions to the
problem of maximizing the n-th eigenfrequency Zn and the problem of
maximizing the gap Zn - Zn 1 between the n-th and the (n-1)-th eigenfrequencies
of the beams. It should be borne in mind that these solutions must be considered
as optimum, limiting solutions from the point of view of practical design.
In Olhoff (1976), the governing equations are derived by the calculus of
variations and solved numerically by a successive finite difference technique
based on a formal integration of the problem for relatively low values of the
given order n of the eigenfrequency Zn . For any higher value of n, the inner
beam separations make it possible to solve very easily the maximum Zn and the
maximum Zn - Zn 1 problems with the aid of a very simple quasi-analytical
method of “Scaled Optimum Beam Elements” developed in Olhoff (1976), and
briefly described in Section 4 of the current paper. Thus, it may be stated that for
beams with any combination of the classical beam end boundary conditions
mentioned above, the optimum solutions corresponding to any given value of n
are presented in Olhoff (1976). These early frequency gap beam results show
that already starting at moderate values of n, say n=5, the optimum beam designs
exhibit a periodicity that increases significantly with increasing values of n.
In contrast to the paper just discussed, a minimum cross-sectional area
constraint (prohibiting creation of inner beam separations and hinges), was taken
into account in the follow-up papers Olhoff and Parbery (1984) and Bendsøe and
Olhoff (1985) which present two slightly different mathematical formulations of
the problem of directly maximizing the eigenfrequency gap Zn - Zn 1 for
cantilever beams. The beams are optimized with and without attached non-
structural masses, and numerical results are presented for values of n up to 5.
The present paper attempts to highlight and extend design results obtained in
Olhoff (1976), Olhoff and Parbery (1984), and Bendsøe and Olhoff (1985) by
determining and presenting new optimum frequency gap beam structures in non-
dimensional form for (i) different combinations of classical boundary conditions,
(ii) much larger values of the orders n and n-1 of the adjacent upper and lower
eigenfrequencies of maximized frequency gaps, and for (iii) different values of a
positive minimum cross-sectional area constraint. The new results are obtained
by the approach of finite element and gradient based optimization using
analytical sensitivity analysis, as described in the preceding paper Olhoff and Du
(2013B). The new solutions are compared with corresponding limiting optimum
solutions obtained without minimum cross-sectional area constraint by usage of
the aforementioned method of “Scaled Optimum Beam Elements” developed in
Olhoff (1976).
2 Formulation for Maximizing Gaps Between Adjacent Eigenfrequencies

of Beam Structures
Shape design optimization of Bernoulli-Euler beams of given length L and
volume V for maximum separation between two adjacent, free transverse
vibration frequencies (eigenfrequencies) shall now be considered. The beams are
assumed to be made of a linearly elastic material with Young’s modulus E and
mass density J , and to have variable, but geometrically similar (e.g., circular)
cross-sections with the relation I cA 2 between the area moment of inertia I
and cross-sectional area A . The constant c is given by the cross-sectional
geometry.
By introducing a dimensionless coordinate x X L , 0 d x d 1 and cross-
sectional area function D x A x L V along the beam, the dimensionless n-
2
th eigenvalue On ( On Z ), where Zn is the dimensionless circular
n
eigenfrequency) associated with free, transverse vibrations takes the form
(Olhoff, 1976),
Zn2J L5
On Zn2 , (1)
cEV
where Zn is the dimensional n-th circular eigenfrequency of the beam.
In a dimensionless finite element setting where the non-dimensional length
and volume of the beam are both assigned unit value, the problem of shape
optimization with the objective of maximizing the gap between two adjacent
frequencies Zn and Zn 1 of given orders n and n-1, can be formulated as
follows:
^
max ' Z 2 Zn2 Zn21
De
` (a)
subject to
KI j Z 2j MIj , j 1, , J , (b)
ITj MIk G jk , j t k, j, k 1, , J , (c ) (2)
NE
¦D l
e 1
e e 1 d 0, (d )
0 D min d D e , e 1, , N E . ( e)
Here, Z j and Ij are the dimensionless j-th eigenfrequency and corresponding
eigenvector, respectively, and ' Z 2 is the difference between the squares of
two consecutive eigenfrequencies of given orders n and n-1 (n = 2, 3, …). In
Eq. (2b), K and M are symmetric positive definite global stiffness and mass
matrices (with corresponding beam element matrices available in (Petyt 2010))
for the generalized structural eigenvalue problem for the vibrating beam structure.
Thus, the J candidate eigenfrequencies ( J ! n ) considered in the optimization
problem will all be real and can be ordered as follows by magnitude:
0 Z1 d Z2 d d Z J (3)
Eq. (2c) imposes the conditions of M -orthonormalization of the corresponding
eigenvectors, where G jk denotes Kronecker’s delta.
The dimensionless optimization problem (2) is discretized by subdividing the
beam into N E finite elements of equal lengths le = 1/ N E with individual cross-
sectional areas D e e 1, , N E , which play the role as design variables of the
discretized problem. Hence, Eq. (2d) expresses the non-dimensional (unit)
volume constraint for the problem, and in Eq. (2e) a positive minimum cross-
sectional area constraint value D min is prescribed for the design variables
D e e 1, , N E . The value of D min is to be chosen less than the mean (unit)
value of the cross-sectional area of the dimensionless beam, and larger than zero
to avoid singularity of the stiffness matrix.
Using an extended bound formulation (Bendsøe et al. 1983, Olhoff 1989,
Jensen and Pedersen 2006), the optimization problem in Eq. (2) can be
reformulated as in Eq. (4) where two scalar variables E1 and E 2 are introduced
in order to facilitate handling of possible multiple eigenfrequencies Zn and Zn 1 .
Note that E1 and E 2 are upper and lower bound parameters in the constraint
equations (4b) and (4c), respectively, and that the difference between them in the
objective function will be maximized. At the same time, E1 and E 2 serve as
design variables together with the cross-sectional areas.
max
E1 , E 2 ,D1 ,,D N E
Ê 2 E1` (a)
subject to
E 2 Z 2j d 0, j n, n 1, , J , (b)
Z 2j E1 d 0, j 1, , n 1, (c )
KI j 2
Z MIj , j 1, , J , (d ) (4)
j
ITj MIk G jk , j t k, j, k 1, , J , ( e)
NE
¦D l
e 1
e e 1 d 0, (f)
0 D min d D e , e 1, , N E . (g)
Not surprisingly, we can observe that the mathematical formulation of the

beam shape optimization problem in Eqs. (4) with E1 , E 2 and the cross-
sectional areas D e e 1, , N E as design variables, is analogous to the
mathematical formulation of the topology design optimization problem with E1 ,
E 2 and the volumetric material densities Ue e 1, , N E as design variables
in Eqs. (10a-d) of the preceding paper Olhoff and Du (2013B).
It is easily shown (Olhoff et al. 2012) that corresponding sensitivity results
with respect to the design variables D e e 1, , N E of the current maximum
frequency gap beam optimization problem can be readily established by
replacing symbols Ue e 1, , N E by D e e 1, , N E , Mj by Ij , and U by
D min in the sensitivity results derived for the maximum frequency gap topology
design optimization problem in Sub-sections 2.2 and 2.4 of the preceding paper.
This implies that the present problem (4) can be solved by the computational
procedure outlined in Fig. 1 and Eq. (20) of the preceding paper.
Note that this procedure is applicable independently of whether the upper
and/or lower eigenfrequencies that define the gap, are members of a multiple
eigenfrequency or are just a simple eigenfrequency.
3 Numerical Examples
3.1 Cantilever beams
Here, examples of cantilever beam designs for maximized higher order
eigenfrequency gaps 'Z3 Z3 Z2 , 'Z4 Z4 Z3 , 'Z9 Z9 Z8 ,
'Z10 Z10 Z9 , 'Z19 Z

19 Z18 and 'Z20 Z 20 Z19 will be presented.
For reasons of accuracy, an initial finite element study was carried out to ensure
that calculated natural frequencies were convergent with respect to the number of
elements applied along the length of the beam. Based on this study, 200 elements
were adopted for the first two beam designs, and 1000 elements for the last four
designs where relatively higher order natural frequencies are considered. The
optimized beam designs are shown in Figs. 1, 2, 3(b), 4(d), 6(b) and 7(b). Each
of these and subsequent designs in this paper are illustrated to suitable scale by
their shape (contour) curves r D x , 0 d x d 1 , after calculation of their linear
dimensions r D e perpendicular to the beam axis. A lower limit D min =0.05 is
prescribed for the non-dimensional cross-sectional area in all the examples
presented in the present and the two sequential sub-sections.
In order to have a convenient reference for evaluation and discussion of the
optimization results, we choose a non-dimensional comparison beam with
uniform cross-section and the same material, boundary conditions, (unit) volume,
(unit) length, and cross-section parameter c as the optimized non-dimensional
beams. The gap between the n-th and (n-1)-th natural frequencies of this uniform
beam is denoted as 'Znu , and for the frequency gaps to be considered here, we
find that 'Z3u =39.66, 'Z4u =59.20, 'Z9u =157.91, 'Z10u =177.65, 'Z19u =355.31,
u
and 'Z20 = 375.04.
The corresponding frequency gaps obtained by optimization are
significantly larger. Thus, the optimized cantilever beam designs in Figs. 1, 2,
3(b), 4(d), 6(b) and 7(b) have the following frequency gaps, 'Z3 =129.72,
'Z4 =195.15, 'Z9 =1144.81, 'Z10 =1332.25, 'Z19 = 5141.39 and 'Z20 =
5542.40.
The designs optimized for maximum frequency gaps 'Z3 and 'Z4 subject to
D min =0.05 in Figs. 1 and 2, respectively, are almost indistinguishable from
optimum designs obtained and illustrated in Olhoff (1976) for the similar
problems of maximizing the higher order natural frequencies Z3 and Z4
without a minimum constraint for the cross-sectional area of the beams; compare
Figs. 1 and 2 in the present paper with Figs. 5 and 6 in Olhoff (1976), where the
latter type of problem is treated. The designs in Figs. 1 and 2 are both obtained
by using the uniform comparison beam as an unbiased initial design.
As a matter of fact, the number of local optimum designs increases with
increasing values of the orders n and n-1 of the eigenfrequencies that define the
frequency gap to be maximized, and with the small lower limit D min =0.05
prescribed for the cross-sectional area in the current examples, it turned out that
for values of n > 4, possible optimum designs could no longer be obtained by

using the uniform comparison beam as an initial design for the iterative
computational procedure. Now, in order to obtain presumed optimum designs
maximizing the frequency gap 'Zn , for given values of n > 4, we assumed that
– as was found for the cases of maximizing the frequency gaps 'Z3 and
'Z4 described above – the design subject to the minimum allowable cross-
sectional area D min =0.05 would be very similar to the design solution to the
problem of maximizing the higher order natural frequency Zn without
specification of a minimum cross-sectional area constraint (which, in the current
context, corresponds to setting D min =0). As already indicated in the Introduction,
this problem is quite extensively covered in Olhoff (1976), and in Section 4 of
the present paper it is briefly described how the “Method of Scaled Optimum
Beam Elements” developed in that reference, may be used to determine the
design of a vibrating cantilever beam that maximizes an eigenfrequency of given,
higher order when D min =0.
We shall now present examples of optimizing cantilever beam designs for
maximum values of the natural frequency gaps 'Z9 , 'Z10 , 'Z19 and 'Z20
subject to the minimum cross-sectional area constraint value D min =0.05, i.e.,
examples where the possible optimum beam designs could not be obtained by
applying the uniform comparison beam as an initial design in the iterative
computational procedure presented in this paper.
Thus, in order to obtain the presumed optimum design maximizing the
frequency gap 'Z9 subject to D min =0.05, shown in Fig. 3(b), we applied a
biased initial design which was very similar to the optimized design obtained by
application of the “Method of Scaled Optimum Beam Elements” as described in
Section 4 by way of the example of maximizing the 9-th natural frequency of a
transversely vibrating cantilever with inner points of zero cross-sectional area
and beam separation (due to D min =0). The design obtained by this approach was
only modified by changing the cross-sectional area to be nowhere less than value
of D min , i.e., D min =0.05, before it was applied as a biased initial design for the
maximization of 'Z9 subject to this constraint value.
The resulting optimized design in Fig. 3(b) distinctly exhibits periodicity.
The two natural frequencies defining the gap 'Z9 are both unimodal, but very
close to neighbouring frequencies. The maximized frequency gap 'Z9 of the
design, cf. the caption of Fig. 3(b), is found to be substantially larger than the
corresponding frequency gap obtained when using the uniform comparison beam
as an initial design for the optimization. This may indicate that the design in Fig.
Figure 1. Cantilever with maximized frequency gap 'Z3 = 129.72. 'Z3u =39.66 for the
comparison design
Figure 2. Cantilever with maximized frequency gap 'Z4 =195.15. 'Z4u =59.20 for the
comparison design
(a)
(b)
Figure 3. Cantilever with maximized frequency gap 'Z9 =1144.81. 'Z9u =157.91 for the
comparison design. (a) Mode shapes, (b) Optimized design
3(b) is the “best” optimum solution to the problem considered. Fig. 3(a) shows to
suitable scale the free vibration modes I 9 x and I 8 x corresponding to the
normalized and mutually orthogonal mode shape vectors I 9 and I 8 associated
with the eigenfrequencies Z9 and Z8 that define the maximized frequency gap
'Z9 Z 9 Z8 of the presumed optimum design in Fig. 3(b).

With a view to obtain a physical understanding of the frequency gap
mechanism, we take a closer look at the variations of the vibration modes I 9 x ,
I 8 x and beam shape r D x along the length 0 d x d 1 of the optimized
dimensionless beam in Fig. 3. Since I 9 x and I 8 x are normalized
eigenfunctions associated with the eigenfrequencies Z9 and Z8 for the beam
with the cross-sectional area function D x , then, according to Rayleigh’s
principle, the two eigenfrequencies are given by
1
'' 2
Z i2 ³ D I
2
i dx , i 8, 9 . (5)
0
In this equation, the integral is proportional to the elastic bending energy
corresponding to the mode I i x .
The physical mechanism behind the maximized gap between the natural
frequencies Z9 and Z8 is as follows. When vibrating with the mode I 9 , see Fig.
3(a), a large value of the associated frequency Z9 is obtained by absorption of
large elastic bending energy in the segments with the larger, non-uniform cross-
sectional areas, cf. Eq. (5), whereas negligible bending energy is absorbed in the
uniform segments with high flexibility. When the beam vibrates with the mode
I 8 , see Fig. 3(a), the associated frequency Z8 becomes very small (and the
frequencies of lower order even smaller) because only the beam segment
adjacent to the clamped beam end receives a small amount of bending energy.
However, here the notable feature is that each of the other segments of the beam
essentially perform rigid body motions in a piecewise linear mode without
bending, and therefore support a low value of Z8 , as the curvatures of the kinks
of the mode all can be neglected in Eq. (5).
Figs. 4(a)-(d) depict four design solutions obtained for the problem of
maximizing the natural frequency gap 'Z10 . Although the values of the
frequency gaps 'Z10 given for the designs in Fig. 4 are very close to each other,
the four designs are seen to be distinctly different, and the eigenfrequencies
defining the gaps 'Z10 are all found to be unimodal. Since the values of the
frequency gaps 'Z10 are smaller for the designs in Figs. 4(a-c) than the value of
the gap 'Z10 for the design in Fig. 4(d), the designs in Figs 4(a-c) must be
considered to be local optimum designs, and we presume that the design in Fig.
4(d) is the global optimum design for the current problem. We managed to
obtain the four different designs by applying a biased initial design for each of
(a) 'Z10 = 1329.66
(b) 'Z10 = 1325.71
(c) 'Z10 = 1325.41
(d) 'Z10 = 1332.25
Figure 4. Cantilevers with maximized frequency gaps 'Z10 . 'Z10u =177.65 for the
comparison design. (a), (b) and (c) Local optimum solutions. (d) Presumed global
optimum solution
Figure 5. Local optimum cantilever associated with frequency gap 'Z10 =901.09
them that was very similar to one of four alternative, presumed global optimum
designs available in Olhoff (1976) for the problem of maximizing the 10-th
natural frequency when inner points of vanishing cross-sectional area are
allowed.
Fig. 5 shows the design that resulted from applying the uniform comparison
beam as an unbiased initial design when attempting to maximize the frequency
gap 'Z10 . As is seen from the caption of Fig. 5, the value of the frequency gap
(a)
(b)
Figure 6. Cantilever with maximized frequency gap 'Z19 = 5141.39. 'Z19u =355.31 for
the comparison design. (a) Mode shapes, (b) Optimized design
'Z10 for this distinctly different design is much lower than that of the design in
Fig. 4(d), so the design in Fig. 5 is only a local optimum solution.
Next, we present examples of cantilever beams with maximized gaps
between adjacent frequencies of higher orders, i.e., 'Z19 Z19 Z18 and
'Z20 Z 20 Z19 . The optimized designs are shown in Figs. 6(b) and 7(b).
Both beam designs are distinct, and the eigenfrequencies defining the maximized
frequency gaps of the designs are both found to be unimodal, albeit very close to
neighbouring eigenfrequencies.
Figs. 6(b) and 7(b) clearly show the important result that except for beam
segments adjacent to the beam ends (whose designs are characteristic for the
specific boundary conditions considered), the entire inner part of each of the
optimum beam designs exhibit a significant periodicity in terms of repeated
beam segments of the same type. By comparing the optimized designs in Figs.
6(b) and 3(b) and in Figs. 7(b) and 4(d), respectively, it may be concluded that
the degree of this inner periodicity increases with increasing values of the orders
n and n-1 of the natural frequencies that define the frequency gap subjected to
maximization. The free vibration modes I19 and I18 are drawn on the basis of the
mode shape vectors I19 and I18 corresponding to the natural frequencies Z19 and
(a)
(b)
u
Figure 7. Cantilever with maximized frequency gap 'Z20 = 5542.40. 'Z20 =375.04 for
the comparison design. (a) Mode shapes, (b) Optimized design
Z18 that define the frequency gap 'Z19 Z 19 Z18 , and are shown in Fig. 6(a).
Similarly, the free vibration modes I20 and I19 corresponding to the natural
frequencies Z20 and Z19 defining the gap 'Z20 Z 20 Z19 are shown in Fig.
7(a). Here, it is interesting to study the influence on the modes of the inner dip in
the (new) beam segment at the free end of the design in Fig. 7(b). Note finally
that it is obvious from Figs. 6 and 7 that the physical frequency gap mechanism
described in connection with Fig. 3, also manifests itself in the current examples.
3.2 Clamped-clamped beams

Here, we present a few examples of optimizing clamped-clamped Bernoulli-
Euler beams, still assuming a lower limit D min =0.05 to be prescribed for the non-
dimensional cross-sectional area of the beams. The frequency gaps considered
are 'Z4 , 'Z9 and 'Z10 , and their maximized values are found to be
'Z4 =309.74, 'Z9 =1411.34 and 'Z10 =1617.91, which are significantly larger
Figure 8. Clamped-clamped beam with maximized frequency gap 'Z4 =309.74.

'Z4u =78.96 for the comparison design

than the corresponding values 'Z4u =78.96, 'Z9u =177.65, and 'Z10u =197.39 for
the uniform, clamped-clamped comparison beam.
The optimized beam designs are shown in Figs. 8, 9 and 10, and it is
interesting to compare the design in Fig. 8 in the present paper with that in the
bottom of Fig. 13 in Olhoff (1976). We note that – as in Figs. 3(b) and 4(d) – it is
seen in Figs. 9 and 10 that periodicity, i.e., repetition of segments of the same
type, already appears in the inner part of the beam designs with maximized
frequency gaps that correspond to relatively low orders of the respective natural
frequencies. However, the segments adjacent to the beam ends are generally
different due to different characteristics of the specific boundary conditions
considered.
3.3 Clamped-simply supported beams

In this section, a few examples of Bernoulli-Euler beams clamped at one end and
Figure 11. Clamped-simply supported beam with maximized frequency gap 'Z4 =274.72.
Figure 12. Clamped-simply supported beam with maximized frequency gap

'Z9 =1226.18. 'Z9u = 172.72 for the comparison design
Figure 13. Clamped-simply supported beam with maximized frequency gap 'Z10 =
1541.79. 'Z10u = 192.46 for the comparison design
simply supported at the other end will be optimized. The same lower limit
D min =0.05 as above is prescribed. The frequency gaps considered are 'Z4 , 'Z9
and 'Z10 , and their maximized values are found to be 'Z4 =274.72,
'Z9 =1226.18 and 'Z10 =1541.79, i.e., they are significantly larger than the
corresponding values 'Z4u =74.02, 'Z9u =172.72, and 'Z10u = 192.46 for the
uniform, clamped-simply supported comparison beam.
The optimized beam designs are shown in Figs. 11, 12 and 13. Periodicity
can be observed in the inner part of the beam designs with maximized band-gaps
'Z9 and 'Z10 , see Figs. 12 and 13. The segments adjacent to the simply
supported end are generally different from those adjacent to the free or clamped
end. Two different kinds of segments adjacent to the simple end support are seen
in Figs. 12 and 13 that depict the beams maximizing the frequency gaps 'Z9
and 'Z10 , respectively. This point is discussed in Section 4.

Figure 14. Cantilever with maximized frequency gap 'Z9 =377.91 subject to a
minimum cross-sectional area constraint value 0.5. 'Z9u =157.91 for the comparison
design
Figure 15. Clamped-clamped beam with maximized frequency gap 'Z9 =460.64 subject
to a minimum cross-sectional area constraint value 0.5. 'Z9u =177.65 for the comparison
design
Figure 16. Clamped-simply supported beam with maximized frequency gap 'Z9 =
431.06 subject to a minimum cross-sectional area constraint value 0.5. 'Z9u = 172.72 for
the comparison design
3.4 Optimum design with larger minimum area constraint value

Figs. 14, 15 and 16 depict the designs obtained by maximizing the frequency gap
'Z9 of a cantilever, a clamped-clamped, and a clamped-simply supported
Bernoulli-Euler beam when applying a larger cross-sectional area constraint
value, namely D min =0.5. The values of the frequency gaps 'Z9 for the
optimized beams and the gaps 'Z9u for the corresponding uniform comparison
beams are given in the captions of Figs. 14, 15 and 16.
When comparing the optimized beams in Figs. 14, 15 and 16 with the
corresponding ones (same order of frequency gap and same boundary conditions)
in Figs. 3(b), 9 and 12, respectively, we make the important observation that
pronounced repetitions of similar segments are found in the inner parts of all
these beams, and that the same degrees of periodicity in the beams in Figs. 3, 9
and 12 are obtained in the beams in Figs. 14, 15 and 16, although the latter are
optimized with the considerably larger value D min =0.5 of the minimum cross-
sectional area constraint.
By the comparison of the above-mentioned figures, we also verify that due to
their larger design freedom, the beams optimized with the small value D min =0.05
of the lower cross-sectional area limit, are associated with significantly larger
increases of the maximized frequency gaps.
4 Discussion
Up to now, we have considered optimum design of Bernoulli-Euler beams with
the objective of maximizing, for a specified value of n, the separation (gap)
between the (higher-order) n-th and (n-1)-th natural frequencies, subject to a
prescribed positive value of a non-dimensional minimum allowable cross-
sectional area D min which has been chosen as D min =0.05 in Sub-sections 3.1, 3.2
and 3.3, and D min =0.5 in Sub-section 3.4.
In this section, we shall briefly discuss the characteristics of this natural
frequency gap optimization problem in the limiting case where the cross-
sectional area function is geometrically unconstrained (except for the given
volume). This means that no minimum constraint is specified for the cross-
sectional area of the beam, i.e., the cross-sectional area is allowed to attain zero
value in discrete points on the beam axis. In this special case (that corresponds
to D min = 0 in the context of this paper), the solutions to our problem of
maximizing the gap between the natural frequencies Zn and Zn 1 are the same
as the solutions to the problem of maximizing a single, higher order natural
frequency Zn of given order n for specified volume, length, and boundary
conditions of the beam. The latter problem is treated in (Olhoff, 1976) where a
large number of optimum designs are available.
The reason why the two different beam optimization problems have identical
solutions, can be explained as follows. When a single natural frequency Zn of
given higher order n is maximized without specification of a minimum constraint
on the cross-sectional area, the optimized beam turns out to possess n-1 degrees
of kinematic freedom to perform rigid motions, since the cross-section vanishes
at inner singular points of the beam. At these points, either inner hinges of zero
bending moment and finite shear force, or, predominantly, inner separations
with both zero bending moment and zero shear force, are created by the
optimization of the n-th natural frequency. This has the effect that
simultaneously with the maximization of the n-th natural frequency, the n-1
degrees of kinematic freedom of the beam turn all the n-1 modes associated with
the lower order natural frequencies into rigid body motions, and all these
frequencies (including that of order n-1) are therefore equal to zero.
Thus, besides maximizing the n-th natural frequency of the beam, the
problem formulation in Olhoff (1976) covers the current problem of maximizing
the difference (gap) between the n-th and the (n-1)-th natural frequency of the
beam, if D min = 0.
The optimized beams associated with the small minimum cross-sectional area
constraint value D min =0.05 in Sub-sections 3.1, 3.2, and 3.3 strongly indicate the
locations of formation of inner hinges and inner separations in the limiting case
of D min = 0. In Fig. 1, for example, the comparatively large inner beam segment
with active minimum cross-sectional area constraint will shrink to a single point
with the formation of an inner separation between the two parts of the beam in
the limiting case of D min = 0. In Fig. 2, the narrow “dip” in the cross-sectional
area function indicates a point where an inner hinge with zero bending moment
will be created in the case of D min = 0. A large number of similar points with
formation of inner beam separations and hinges in the case of D min = 0 are easily
identified in the figures. It is worth noting that in each of the optimum designs
shown, there is no more than a single inner point with a narrow “dip” in the area
function indicating formation of an inner hinge in the case of D min = 0, but an
increasing number of points that indicates creation of inner separations, when the
given order of the frequency gap is increased. An exception is the design in Fig.
5 which contains several narrow “dips”, but this is a local, and not global
optimum solution.
In the optimum designs with one end clamped and the other end simply
supported, see Figs. 12 and 13, we observe two kinds of segments adjacent to the
simply supported end. Quite surprisingly, for the design shown in Fig. 12, the
beam segment at the simply supported end will shrink to a separation in the
limiting case of D min = 0, where both the bending moment and shear force are
zero at the end point of the beam. Hence, as is discussed in Olhoff (1976), the
beam disconnects from the support, i.e., the separation makes the simple support
superfluous. This implies that the optimum solution in the limiting the case of
D min = 0 will be the same as that with a free end, instead of the original simply
supported end. Contrary to this, the beam segment adjacent to the simply
supported end in Fig. 13 will not shrink to a separation in the limiting case of
D min = 0, but remain connected to the hinge at the end point. Thus, in this case,
the optimum solution behaves as expected and takes advantage of the simple
support.
In Olhoff (1976), the dimensionless beam optimization problem is first
solved for small values of n for various combinations of classical boundary
conditions. This is done by successive iterations based on a numerical integration
of the governing equations, where singularities in the n-th eigenmode and its
derivatives are isolated at points of zero beam cross-sectional area for reasons of
accuracy and convergence. Hereby a small number of different types of
optimally designed, non-dimensional beam elements are produced. By proper
scaling, these beam elements can be used as building blocks for optimum beams
associated with much larger values of the prescribed order of their n-th
eigenfrequency.
Since, as mentioned above, at most one inner hinge will appear in a global
optimum beam with D min = 0, and an inner hinge can be included in optimized
non-dimensional beam elements mentioned above, then all other inner points of
vanishing cross-sectional area in an optimum beam associated with a sufficiently
large value of n, will be inner separations between optimized beam elements.
This is very important because the inner separations provide the means to solve
very easily the optimization problem for a sufficiently large value of n by an
optimum scaling of the optimized non-dimensional beam elements by means of
very simple analytical formulas derived in Olhoff (1976).
As an example, let us consider the approach of determining the design of a
vibrating cantilever beam that maximizes, say, the 9-th eigenfrequency Z9 of the
beam, when no minimum cross-sectional area constraint is prescribed, i.e.,
D min =0. (Note that the optimized beam with D min =0.05 is shown in Fig. 3.)
The first step of the method consists in applying Table 2 and Eq. (30) in
Olhoff (1976) which easily yields that the optimum cantilever design associated
with n=9 will have four inner separations and be composed of (or assembled as)
five dimensionless, optimized elements (or segments) along the length of the
beam: an element “a” consisting of a cantilever optimized for n=1 (see Fig. 1 in
the paper), followed by four elements “c” (see Fig. 11 in the paper), each
consisting of a free-free beam optimized for n=3 (the order of the lowest non-
vanishing natural frequency for such a beam). The four (identical) “c” elements
will endow the resulting optimum beam design with periodicity. The beam
elements “a” and “c”, together with no more than four other elements, are
necessary for the optimization of non-dimensional Bernoulli-Euler beams for any
value of n and any combination of classical boundary conditions (clamped,
simply supported and free ends). These elements are all optimized with their
designs shown in the first part of Olhoff (1976), and the elements are listed
together with their optimum characteristics in Table 1 of the paper. Finally, very
simple explicit algebraic expressions [(30), (57) (63) and (64)] are derived and
presented in the paper, for computation of the maximum value of the n-th
eigenfrequency (in the current example Z9 ), of the optimized, assembled beam,
and for the proper scaling of the lengths and volumes of the individual,
optimized beam elements, such that each of these elements will vibrate at the
same frequency as the assembled, optimum beam.
The approach is coined the “Method of Scaled Optimum Beam Elements”,

and it should be mentioned that for Bernoulli-Euler beams subject to any
combination of clamped, simply supported and free end boundary conditions, the
geometrically unconstrained optimum solutions to the problem of maximizing
the n-th eigenfrequency or frequency band-gap corresponding to any value of n
are available in the paper.
5 Free Wave Propagation and Forced Vibration in the Optimized

Periodic Beam Structure
It is interesting to examine the transverse wave propagation and vibration filter
effect in the optimized periodic beam structures obtained in the preceding
sections. First, the wave propagation in an infinitely long periodic beam is
analyzed, where the infinite beam is constructed by repeated translation of an
inner beam segment obtained in the frequency gap optimization above.
Subsequently, as an example, a frequency response analysis is conducted for the
optimized beam in Fig. 6(b), when the beam is subjected to an external time-
harmonic excitation with a view to investigate the attenuation levels in the
frequency gap 'Z19 Z19 Z18 .
From the Floquet theory (Brillouin, 1953) and waveguide finite element
(WFE) method (Mace et al., 2005),the wave propagation through the entire
infinite periodic beam mentioned above can be determined by analyzing the
wave motion within a single repeated beam segment, which is called a unit cell.
The band-gaps can be explored by analyzing the unit cell. The transfer matrix T
of the unit cell can be defined from the dynamic stiffness matrix of the
conventional finite element analysis. Detailed derivation of the transfer matrix is
available in Mace et al. (2005). The eigenvalues O of the transfer matrix T are
defined by the propagation constant K (Bloch parameter) as (Mead, 1996; Søe-
Knudsen and Sorokin, 2010)
cos K i sin K ,
Im Re Im
O eiK e K eiK e K Re Re
(6)
Re Im
where K and K represent the real and the imaginary parts of K ,
respectively. A stop band is found when all eigenvalues O of T fulfil the
condition O z 1 , i.e., K Im z 0 . In this stop band, free wave propagation is
prohibited. Due to two degrees of freedom at each end node of the unit cell, there
are four eigenvalues O1 , O2 , O3 , and O4 of the transfer matrix.
One of the inner repeated segments in Fig. 3(b) is analyzed as a unit cell in
the infinite periodic beam. The magnitude O of eigenvalues is plotted as a
function of the non-dimensional circular frequency defined in Eq. (1), as shown
in Fig. 17, where three stop bands can be identified from the frequency range
with O z 1 , indicated by three grey domains in Fig. 17. The 8th and 9th non-
dimensional circular eigenfrequencies of the uniform, comparison beam and the
optimized beam are also shown in Fig. 17. As is well known, no stop band exists
in the infinite uniform beam in absence of damping. However, a relatively large
stop band for the infinite periodic beam is observed, where bending waves
cannot propagate. Similarly, stop bands can be seen in Figs. 18 and 19 for two
other examples. These figures demonstrate that there is almost perfect correlation
between the band-gap size/location of the emerging band structure and the
size/location of the corresponding maximized natural frequency gap in the finite
structure.
Figure 17. Variation of O versus excitation frequency obtained by employing Floquet

theory for an inner repeated segment in Fig. 3(b), where the frequency gap
'Z9 Z9 Z8 of the cantilever beam is maximized. The grey domains indicate Floquet-
predicted stop bands. The 8th and 9th eigenfrequencies Z8u and Z9u of the comparison
beam with uniform cross-section, and Z8 and Z9 of the optimized beam are shown in the
figure.
Figure 18. Variation of O versus excitation frequency obtained by employing Floquet

theory for an inner repeated segment in 6(b), where the frequency gap 'Z19 Z19 Z18 of
the cantilever beam is maximized. The grey domains indicate Floquet-predicted stop
bands. The 18th and 19th eigenfrequencies Z18u and Z19u of the comparison beam with
uniform cross-section, and Z18 and Z19 of the optimized beam are shown in the figure.
Figure 19. Variation of O versus excitation frequency by employing Floquet theory for
an inner repeated segment in Fig. 9, where the frequency gap 'Z9 Z9 Z8 of the
clamped-clamped beam is maximized. The grey domains indicate Floquet-predicted stop
bands. The 8th and 9th eigenfrequencies Z8u and Z9u of the comparison beam, and Z8 and
Z9 of the optimized beam are shown in the figure.
It has been demonstrated in many papers, see e.g., Jensen (2003), Søe-
Knudsen and Sorokin (2010), that a structure with a finite number of repeated
unit cells may significantly suppress propagation of waves with frequencies in
the stop band. Fig. 20 shows the displacement response at the right hand end of
the optimized beam shown in Fig. 6(b), when the beam is subjected to a time-
harmonic base excitation in the transverse direction at the left hand end. The base
motion is prescribed with a given displacement amplitude u0 relative to the
fixed reference axis. The transverse displacement u at the right hand end is
u2
indicated in Fig. 20 in the form 10 log10 2 dB.
u0
Figure 20. Displacement response at the right hand end from the flexural vibration of the
optimized beam in Fig. 6(b) when the beam is subjected to time-harmonic base excitation
in the transverse direction at the left hand end. No damping is assumed. The 18th and 19th
eigenfrequencies Z18 and Z19 of the optimized beam are indicated in the figure.
It is seen from Fig. 20 that there is a large drop in the response in the stop
band frequency range. The stop band calculated from the corresponding unit cell
is given in Fig. 18. It demonstrates that the stop band may exist in the optimized
beam obtained by maximization of a frequency gap. It is observed from Figs. 18
and 20 that there is a correlation between the value of O representing the
strength of attenuation in a band gap, and the level of attenuation in the
frequency response function for a finite structure composed of the same periodic
unit cell. The many resonance peaks observed in Fig. 20 are due to the fact that
no damping is included. The resonance peaks can be removed or reduced by
including some damping, and we also found that there is no significant change of
the band-gap behavior for relatively small damping. The effect of smoothing by
including damping is often used in the topology optimization of band-gap
structures (Sigmund and Jensen, 2003).
6 Conclusions
Maximizing gaps between two adjacent natural frequencies (eigenfrequencies) of
free transverse vibrations of prescribed order is investigated in this paper which
lends itself to Du and Olhoff (2007a,b) and Olhoff et al. (2012). The results are
obtained by finite element and gradient based optimization using analytical
sensitivity analysis. An incremental optimization formulation with consideration
of multiple eigenvalues is applied, which can be used for problems with any mix
of multiple and simple eigenfrequencies. Non-dimensional optimum solutions
are presented for different classical boundary conditions, different orders of the
upper and lower eigenfrequencies of maximized gaps, and values of a minimum
cross-sectional area constraint. However, geometrically unconstrained optimum
solutions obtained in Olhoff (1976) are also discussed and utilized in this paper.
The results show that, except for beam segments adjacent to the beam ends
whose designs are characteristic for the specific boundary conditions considered,
all the inner part of the optimum beam designs exhibits a significant periodicity
in terms of repeated beam segments, the number of which increases with
increasing orders of the upper and lower frequencies of the maximized gaps.
When small values of the minimum cross-sectional area are prescribed,
solutions to the current problems are very close to corresponding solutions
obtainable by simple non-dimensional analytical expressions for limiting
optimum solutions that were derived earlier by a “method of scaled beam
elements” (Olhoff, 1976) in which inner points of vanishing cross-sectional area
in the beams were allowed and exploited.
In wave propagation problems, band-gap is found in an infinite beam
structure constructed by repeated translation of an inner beam segment obtained
by eigenfrequency gap optimization. The band-gap size/location of the emerging
band structure is matching very well with the size/location of the corresponding
maximized natural frequency gap in the finite structure. For the optimized
structures composed of a finite number of repeated segments in the inner part,
the motion transmitted from one end will be significantly suppressed by the
periodic segments. For the beam structures studied here, it can be concluded that
the optimum design maximizing the gap between two adjacent eigenfrequencies
of free transverse vibration of given higher order is periodic. It is also
demonstrated that the approach tailors a band-gap which is matching very well
the maximized frequency gap in the periodic structure characterizing elastic or
acoustic wave propagation.
References
Bendsøe, M. P., Olhoff, N., Taylor, J. E., 1983. A variational formulation for
multicriteria structural optimization. J Struct Mech, 11 (4): 523-544.
beams and shafts. Optim Control Applications & Methods, 6 (3): 191-200.
Applications. Berlin: Springer-Verlag, 2003.
Brillouin, L., 1953. Wave Propagation in Periodic Structures, 2nd edition. Dover
Publication, New York.
problems using a homogenization method. International Journal for Numerical
Methods in Engineering, 35 (7): 1487-1502.
Diaz, A.R., Haddow, A.G., Ma, L., 2005. Design of band-gap grid structures. Structural
and Multidisciplinary Optimization, 29 (6): 418-431.
Du, J., Olhoff, N., 2007a. Topological design of freely vibrating continuum structures
Structural and Multidisciplinary Optimization, 34 (2): 91-110.
for maximum values of simple and multiple eigenfrequencies and frequency gaps
(V.ol 34, pg 91, 2007). Editors’s Erratum, Structural and Multidisciplinary
Optimization, 34 (6), 545.
Halkjær, S., Sigmund, O., 2004. Optimization of beam properties with respect to
maximum band-gap. Mechanics of the 21st Century, Procedings of 21st International
Congress of Theoretical and Applied Mechanics. Gutkowski, W., Kowalewski, T.A.
(Eds.), IUTAM, Warsaw, Poland (ISBN: 978-1-4020-3456-5).
Halkjaer, S., Sigmund, O., Jensen, J.S., 2006. Maximizing band gaps in plate structures.
Structural and Multidisciplinary Optimization, 32 (4): 263-275.
Hladky-Hennion, A.C., Allan, G., deBilly, M., 2005. Localized modes in a one-
dimensional diatomic chain of coupled spheres. J Applied Physics, 98 (5): 054909.
Hussein, M.I., Hulbert, G.M. and Scott, R.A., 2006a. Dispersive elastodynamics of 1D
banded materials and structures: Analysis. Journal of Sound and Vibration, 289 (4-5):
779-806.
Hussein, M.I., Hamza, K., Hulbert, G.M., Scott, R.A., Saitou, K., 2006b. Multiobjective
evolutionary optimization of periodic layered materials for desired wave dispersion
characteristics. Structural and Multidisciplinary Optimization, 31 (1): 60-75.
Hussein, M.I., Hulbert, G.M., Scott, R.A., 2007. Dispersive elastodynamics of 1D
banded materials and structures: Design. Journal of Sound and Vibration, 307 (3-5):
865-893.
Jensen, J.S., 2003. Phononic band gaps and vibrations in one- and two-dimensional
mass-spring structures. Journal of Sound and Vibration, 266 (5): 1053-1078.
Jensen J.S., Sigmund O., Thomsen J.J., Bendsøe M.P., 2002. Design of multi-phase
structures with optimized vibrational and wave-transmitting properties. Proc. 15th
Nordic Seminar on Computational Mechanics, Aalborg University. Lund, E., Olhoff,
N., Stegmann, J. (Eds), Institute of Mechanical Engineering, Aalborg University,
Denmark, 2002, p. 63-66 (ISBN: 87-89206-67-3).
Jensen, J.S., Pedersen, N.L., 2006. On maximal eigenfrequency separation in two-
material structures: The 1D and 2D scalar cases. Journal of Sound and Vibration, 289:
(4-5), 967-986.
Jensen, J.S., 2007. Topology optimization problems for reflection and dissipation of
elastic waves. Journal of Sound and Vibration, 301 (1-2): 319-340.
Larsen, A.A., Laksafoss, B., Jensen, J.S., Sigmund O., 2009. Topological material layout
in plates for vibration suppression and wave propagation control. Structural and
Multidisciplinary Optimization, 37 (6): 585-594.
Mace, B.R., Duhamel, D., Brennan, M.J., Hinke, L., 2005. Finite element prediction of
wave motion in structural waveguides. Journal of the Acoustical Society of America,
117 (5): 2835-2843.
Mead, D.J., 1996. Wave propagation in continuous periodic structures: Research
contributions from Southampton, 1964-1995. Journal of Sound and Vibration, 190
(3): 495-524.
Niu, B., Yan J., Cheng, G., 2009. Optimum structure with homogeneous optimum
cellular material for maximum fundamental frequency. Structural and Multi-
disciplinary Optimization, 39 (2): 115-132.
Olhoff, N., 1976. Optimization of vibrating beams with respect to higher-order natural
frequencies. Journal of Structural Mechanics, 4 (1): 87-122.
Olhoff, N., 1977. Maximizing higher-order eigenfrequencies of beams with constraints
on design geometry. Journal of Structural Mechanics, 5 (2): 107-134.
Olhoff, N., 1989. Multicriterion structural optimization via bound formulation and
mathematical programming. Structural and Multidisciplinary Optimization, 1 (1): 11-
17.
maximum difference between adjacent natural frequencies. Int. J. Solids and
Structures, 20 (1): 63-75.
Olhoff, N., Niu, B., Cheng, G., 2012. Optimum design of band-gap structures. Int. J.
Solids and Structures, 49: 3158-3169.
Olhoff, N., Du, J., 2013B. Structural topology optimization with respect to
eigenfrequencies of vibration. In: Rozvany, G., Lewinski, T., Eds., Topology
Sciences (CISM), Udine, Italy, June 18-22, 2012. Springer-Verlag, Vienna, 2013, 24
pp.
Petyt, M., 2010. Introduction to finite element vibration analysis, 2nd Ed. Cambridge
University Press, Cambridge, UK.
Sigalas, M.M., Economou, E.N., 1992. Elastic and acoustic-wave band-structure.
Journal of Sound and Vibration, 158 (2): 377-382.
Sigmund, O., Jensen, J.S., 2003. Systematic design of phononic band-gap materials and
structures by topology optimization. Philosophical Transactions of the Royal Society
of London Series A-Mathematical Physical and Engineering Sciences, 361 (1806):
1001-1019.
Søe-Knudsen, A., 2011. Design of stop-band filter by use of pipe segments and shape
optimization. Structural and Multidisciplinary Optimization (DOI: 10.1007/s00158-
011-0691-2).
Søe-Knudsen, A., Sorokin, S.V., 2010. Modelling of linear wave propagation in spatial
fluid filled pipe systems consisting of elastic curved and straight elements. Journal of
Sound and Vibration, 329 (24): 5116-5146.
Topological Design for Minimum Dynamic
Compliance of Structures under Forced Vibration

1
2
Abstract This paper deals with topology optimization of elastic,

continuum structures without damping that are subjected to time-harmonic,
dynamic loading with prescribed excitation frequency and amplitude. An
important objective of such a design problem is often to drive the
eigenfrequencies of the structure as far away as possible from the
excitation frequency in order to avoid resonance and reduce the vibration
level of the structure. The total structural volume, the boundary conditions,
and the material are given.
1 Introduction
Topology optimization for minimum dynamic compliance is equivalent to

maximizing the dynamic stiffness of structures subjected to time-harmonic
external loading of given frequency and amplitude, and has, e.g., been studied by
Ma et al. (1995), Jog (2002), Olhoff and Du (2005, 2013), Yoon (2010), Shu et
al. (2011), Kang et al. (2012), and Yang and Li (2013). Similar problems studied
by Calvel and Mongeau (2005) and Jensen (2007) concern topology optimization
of continuum structures subjected to dynamic constraints (e.g. the amplitude of
displacement response) when a range of excitation frequencies is considered.
Topology design subject to transient external loading where the dynamic
compliance is defined relative to a specified time interval, was studied by Min et
al. (1999) and reviewed by Kang et al. (2006) where several other references are
available.

DOI 10.1007/978-3-7091-1643-2_13, © CISM, Udine 2014
A problem that may arise in structural topology optimization under time-

harmonic dynamic loading is that the static compliance (corresponding to the
same loading amplitude, but zero frequency) may increase to a very high level
(see e.g. Tcherniak, 2002) during the process where the dynamic compliance of
the structure is optimized. In extreme cases the static compliance actually tends
to infinity, see Olhoff and Du (2005, 2013), which reflects that a disintegration
of the structure is being created during the design process. However, it was
demonstrated by Olhoff and Du (2005, 2013) that the design objective of the
dynamic problem can be implemented along different optimization paths, and
that it is possible to avoid the problem mentioned above by selection of the
proper path. Thus, in the present paper, an approach is presented in which the
static compliance of the structure is constrained or decreased during the process
of optimizing the dynamic properties. An algorithm developed for this handles
the optimum design problem by a continuation technique where the loading
frequency is sequentially increased (or decreased) from a sufficiently low (or
high) initial value up to (or down to) its prescribed value. Numerical examples
presented in Section 3 demonstrate the validity of the approach, and conclusions
are drawn in Section 4. The present paper lends itself to Olhoff and Du (2013).
2 Formulation of the Problem of Minimizing the Dynamic Compliance
The problem of optimizing the topology of a continuum structure (without

damping) for minimum value of the integral dynamic structural compliance can
be formulated in a discrete form as follows:
min {F Cd2 }
Ue , e 1 , , N E
Subject to :
Cd | PTU | ,
(1)
(K Z p2 M )U P,
NE
* (V *
¦ U eVe V d 0 , DV0 ) ,
e 1
0 U d Ue d 1 , (e 1 , , N E ) .
In (1), the symbol Cd stands for the dynamic compliance defined as
Cd | P T U | . Here, P denotes the vector of amplitudes of a given external time-
iZ p t
harmonic mechanical surface loading vector p(t ) Pe with the prescribed
excitation frequency Z p , and U represents the vector of magnitudes of the
Topological Design for Minimum Dynamic Compliance… 327
iZ t
corresponding structural displacement response vector a(t ) Ue p . Thus, U
and P satisfy the dynamic equilibrium equation included in (1) for the steady-
state vibration at the prescribed frequency Z p , with K and M representing the N
dimensional global structural stiffness and mass matrices, where N is the number
of DOFs. We note that the above expression for the dynamic compliance Cd
represents the numerical mean value of the magnitudes of the surface
displacements weighted by the values of the amplitudes of the corresponding
time-harmonic surface loading. For the case of static loading ( Z p = 0), the
expression directly reduces to the traditional definition of static compliance, i.e.,
the work done by the external forces against corresponding displacements at
equilibrium.
In (1), NE denotes the total number of finite elements in the admissible design
domain for the topology optimization problem. The symbols Ue, e = 1,…,NE, play
the role of design variables of the problem and represent the volumetric material
densities of the finite elements, with lower and upper limits U and 1 specified
for Ue. To avoid singularity of the stiffness matrix, U is not zero, but taken to be
a small positive value like U = 10-3. In the second but last constraint in (1), the
symbol D defines the volume fraction V * /V0 , where V0 is the volume of the
admissible design domain, and V * is the given available volume, respectively, of
the solid material for a single-material design problem and of the solid material
*1 for a bi-material design problem, cf. Sub-sections 2.1 and 2.3 in the paper
Olhoff and Du (2013A).
It is noted from (1) that the global dynamic stiffness matrix K d defined as
Kd K Z p2 M may be negative definite when the prescribed external excitation
frequency Z p has a high value, e.g. higher than the fundamental eigenfrequency
of the structure. In this case, the scalar product of the transpose of the vector of
amplitudes of the external surface loading and the vector of amplitudes of the
structural displacement response may become negative, and in order to include
this possibility in our problem formulation, we apply the absolute value of this
scalar product as the dynamic compliance Cd, see (1). Moreover, to render the
problem differentiable, we choose the objective function F as the square of the
dynamic compliance. The dynamic equilibrium equation in (1) is solved in a
direct way by Gauss elimination in this paper.
2.1 Sensitivity analysis

The sensitivity of the objective function F in problem (1) with respect to the
design variables Ue is given by
F c (C 2 )c 2(P T U) (P c T U P T U c ) , (2)
d
where prime denotes partial derivative with respect to Ue. The sensitivity Pc of
the load vector will be zero if it is design-independent, otherwise it can be
handled using the method described by Hammer and Olhoff (1999, 2000), and
also by Du and Olhoff (2004a,b). The sensitivity Uc of the displacement vector
is given by
(K Z p2 M ) Uc f { Pc (K c Z p2 Mc) U , (3)
where the sensitivities of the stiffness and mass matrices can be directly obtained
from the SIMP material model in Eqs. (4) of the paper Olhoff and Du (2013A). The
vector f is known as the pseudo load and is defined by the term on the right-hand
side of Eq. (3). Instead of solving Eq. (3), the adjoint method (see e.g. Tortorelli and
Michaleris 1994) may be used to calculate the sensitivity of the objective function in
a more efficient manner, which gives the following result
(4)
>
F c 2(P T U) 2U T Pc U T (K c Z p2 M c) U . @
Accordingly, the optimality condition for problem (1) can be expressed in the
following form by means of the method of Lagrange multipliers,
(5)
> @
2(P T U) 2U T Pc U T (K c Z 2p Mc) U /Ve 0 ,
where / is the Lagrange multiplier corresponding to the material volume constraint,
and the side constraints for Ue have been ignored. The optimization problem (1) can
be solved by using the well-known MMA method (Svanberg 1987) or an optimality
criterion method, e.g. the fixed point method, as devised by Cheng and Olhoff
(1982).
2.2 An approach for minimum dynamic compliance design

The dynamic compliance defined in the first constraint of problem (1) may
alternatively be written as follows by using the modal superposition technique
(without damping)
1 2
2
ª §Z
I
· º § P Tĳ i · I
T
| P U | ¦ «1 ¨ p ¸ » (6)
Cd
Zi ¹ »
¨
¨ Z
¸
¸ ¦ M i ci ,
¬ ©
i 1« ¼ © i ¹ i 1
where Zi represents the ith eigenfrequency and Mi the corresponding eigenmode

of the structure, and
2 1
2
§ P Tĳ i · ª §Z · º
p
ci ¨¨ ¸
¸ and M i « 1 ¨ ¸ » (7)
© Zi ¹ «¬ © Zi ¹ »¼
can be interpreted as the contribution of the ith eigenmode of the structure to the
dynamic compliance and the corresponding magnification factor (as defined for a
single degree of freedom system). It is noted that the lower eigenfrequency
normally gives more contribution (implying a larger value of ci) to the dynamic
compliance of the structure if the corresponding eigenmode is not orthogonal to
the external loading mode. On the other hand, when the loading frequency is
close to an eigenfrequency of the structure, the absolute value of the
magnification factor corresponding to this eigenfrequency will increase very
quickly, which indicates the occurrence of a resonance.
Not surprisingly, minimization of the integral dynamic compliance normally
yields a structure whose eigenfrequencies are far from the prescribed excitation
frequency Zp of the dynamic load; this structural behaviour implies efficient
avoidance of resonance phenomena with large displacement amplitudes and low
dynamic stiffness. In the present topology optimization problem, the initial
design (cf. D1 in Fig. 1(a)), which is normally chosen to have a uniform
distribution of material with intermediate density over the admissible design
domain, may have a fundamental eigenfrequency (resonance frequency) : = :1
that is smaller than the given loading frequency Zp. In this case, a decrease of the
dynamic compliance corresponding to Zp normally implies an increase of the
static compliance (that corresponds to the same loading amplitude but zero
frequency Z = 0), due to a decrease of the fundamental eigenfrequency :
(thereby avoiding resonance), see Fig. 1(a). As a result of this, in particular in
single-material problems, the structure may become very weak at the (local)
optimum of the dynamic compliance that is obtained. In order to prevent this,
one may introduce an upper bound constraint on the static compliance. However,
this will delimit the gain of the optimization of the dynamic compliance.
In fact, much lower values of the dynamic compliance can be obtained if we
can start out the optimization procedure using a value Z of the loading frequency
that is lower than the value of the fundamental eigenfrequency (resonance
frequency) :1 for the initial design, and then sequentially increase Z up to its
originally prescribed value Zp , see Fig. 1(b). This procedure has the desirable
effect of generating a series of topologies with increasing values of both the
fundamental frequency : and the static and dynamic stiffnesses for the sequence
of structures produced (we may call this technique a “continuation technique”).
Finally, the procedure delivers the optimum dynamic compliance topology
Cd Initial design D1
Path 2
Cs2 D2 D1
Cd1
Cs1
Cd2
Z=0 :2 :1 Zp Z
(a)
Initial design D1
Cd
Path 1
D1 D3
Cd1
Cs1 Cd3
Cs3
Z=0 Zp :1 :3 Z
(b)
Figure 1. Principle sketch of the dependence of the dynamic compliance Cd on the

loading frequency Z for the case where the prescribed loading frequency Zp is close to the
fundamental eigenfrequency :1 of the initial design D1. Symbols :2 and :3 represent the
fundamental eigenfrequencies of the designs D2 and D3. (a) If the fundamental

eigenfrequency :1 of the initial design D1 is less than the prescribed loading frequency
Zp, i.e. :1 Z p , then the design will proceed along path 2 (see (a)) to decrease the
dynamic compliance, and as a result, the dynamic compliance corresponding to Zp
becomes smaller, i.e. C d 2 C d 1 , but the static compliance Cs (corresponding to Z =0) for
the same design increases, C s 2 ! C s1 . (b) If :1 ! Z p , then the design will proceed along
path 1 (see (b)), and as a result, both the dynamic compliance (corresponding to Zp) and
the static compliance (corresponding to Z =0) for the same design decrease, i.e.
C d 3 C d 1 and C s 3 C s1 .
solution subject to the originally prescribed loading frequency, Zp. The

procedure automatically avoids resonance, and works very well as long as the
prescribed loading frequency Zp is lower than the maximum obtainable value
:opt of the fundamental eigenfrequency, i.e. Z p : opt . Moreover, since the
fundamental eigenfrequency of the structure maintains a value higher than the
loading frequency, the dynamic stiffness matrix K d K Z p2 M , see problem
(1), remains positive definite during the design process. This implies a very good
feature embedded in the dynamics design, i.e., the global structural response
approaches zero when the dynamic compliance of the structure approaches zero.
Now, if the prescribed value of Zp is such that Z p ! : opt (but sufficiently
smaller than the second resonance frequency of the design associated with the
maximum eigenfrequency :opt ), the minimization of the dynamic compliance
will drive the fundamental eigenfrequency : of the design towards zero (for
single material design). At the same time, the static displacements of the
structure become very large, which means that the static stiffness tends to zero.
The physical reason for this behaviour is that, in the limit, a disintegration is
created in the structure. In this limit, the zero value of the fundamental
eigenfrequency is associated with a rigid body vibration mode of the structure,
and the static displacements of the disintegrated part of the structure become
infinite, as the structure cannot sustain the static load.
A straight-forward way of avoiding this unwanted structural behaviour is to
include an upper bound constraint on the static compliance in the mathematical
formulation of the problem of minimizing the dynamic compliance. We have
found that such a constraint is extremely effective and well-chosen when
minimizing the dynamic compliance at a value of Zp somewhat larger than :opt .
3.1 Minimum dynamic compliance design of a plate-like structure

100
Prescribed loading frequency Zp

80
Frequencies
60 First eigenfrequency :1
40
20
0
0 5 10 15 20 25 30
Iteration number
(a) (b)
-6
x 10
2
Dynamic and static compliance
1.5 Static compliance Cs Z
Dynamic compliance Cd for

Z Z
p
0.5
0
0 5 10 15 20 25 30
Iteration number
(c) (d)
Figure 2. (a) Admissible design domain (a = 3, b = 2 and c = 0.1) with loading and
support conditions. (b) Iteration history for the first eigenfrequency of the plate (:1 < Zp =
80). (c) Iteration histories for the dynamic and static structural compliance (the latter
corresponds to the same loading amplitude but frequency Z = 0). (d) Material distribution
at iteration step 30.
This example concerns optimum topology design of a single-material 3D plate-

like structure with support conditions as shown in Fig. 2(a). A time-harmonic,
concentrated transverse external load p(t) = PcosZ t with P=1 is applied to the
center of the plate, and the objective is to minimize the dynamic compliance of
the plate for a prescribed loading frequency Z = Zp = 80 and a volume fraction of
50% for the given solid material. (SI-units are used throughout.) The
fundamental eigenfrequency of the plate in the initial design (see Fig. 2(a)) is :1
= 61.6, i.e., less than the given loading frequency. Minimization of the dynamic
compliance drives the design away from the resonance point which implies a
continual decrease of :1 as shown in Fig. 2(b). As a result, the static compliance
of the structure increases very quickly (Fig. 2(c)). Fig. 2(d) shows that at
iteration step 30, the plate has become very weak at the two fixed supports. This
indicates creation of a rigid body vibration mode in association with the first
eigenfrequency, and that the structure cannot effectively sustain the static load
associated with Z = 0.
-7
130 x 10
5
120
First eigenfrequency :1 Dynamic compliance Cd for
Structural compliance
4 loading frequency Z Z

110 p
Frequencies
100
3
90
Prescribed loading frequency 2 Static compliance Cs Z
80
Z Zp
70
1
60
Initial loading frequency Z 0
50 0
0 5 10 15 20 25 30 5 10 15 20 25 30
Iteration number Iteration number
(a) (b)
(c) Zp = 80 (d) Zp =150 ( C S d C S )
Figure 3. (a), (b) Iteration histories for the first eigenfrequency of the plate, the loading
frequency, and the dynamic and static compliances. (c), (d) Optimum topologies (50%
volume fraction, single-material design) for Zp=80 and Zp=150 (with an upper bound on
the static compliance, i.e. C S d C S ).
In order to avoid such a statically weak design, a more expedient approach is

adopted for solution of the above problem. Thus, we use a continuation approach
and start out the design problem with a value Z = Z0 of the loading frequency
that is lower than the first eigenfrequency :1 of the initial design, and we then
sequentially increase Z up to its originally prescribed value Z = Zp = 80 (Fig.
3(a)). In the converged result, a structure with minimized dynamic compliance
and improved static stiffness is now obtained (see Fig. 3(b,c)).
Finally, let us consider a case with a prescribed, higher value of the loading
frequency, Zp=150. We have computed the optimum value of the first
eigenfrequency of the plate to be :opt = 127.6, i.e., lower than the given loading
frequency in this case. Then, to ensure a reasonable static stiffness of the design,
we introduce an upper bound C S d C S 5 u 10 8 for the static compliance CS in
the formulation of the problem. The optimum topology result for this problem is
shown in Fig. 3(d).
-7 -7
x 10 x 10
4
Dynamic compliance of structure
6
Dynamic compliance of structure
Dynamic compliances Cd for

3.5
Dynamic compliances Cd for 5
Z Z
p
3 Z Z
p
Z Z
p
Z Z
p
2.5
and different upper bounds on Cs 4
Z Z
p
Cs = 0.5x10-7 with constraint
2 3
Cs = 1x10-7 Cs <= Cs = 0.5x10 -7
1.5 Cs = 3x10-7
2
Cs = 4x10-7
1
1
0.5
0 0
0 50 100 150 200 0 50 100 150 200
Iteration number Iteration number
(a) (b)
Figure 4. (a) Iteration histories of the dynamic compliances of the plate subject to a high
loading frequency (Z = Zp = 150 > :opt = 127.6) and four different upper bound
constraints on the static compliance Cs, i.e. C s d C s . (b) Iteration histories of the dynamic
compliances of the plate subject to a given upper bound constraint on Cs
( C s d C s 0.5 u 10 7 ), for four different loading frequencies Zp = 130, Zp = 150, Zp = 180
and Zp = 200, all of which are higher than the optimum value of the fundamental
eigenfrequency :opt .
Fig. 4(a) shows the iteration histories of the dynamic compliance of the plate
subject to the higher loading frequency ( Z = Zp = 150) and four different upper
bound constraints on the static compliance Cs (associated with the same loading
amplitude but zero frequency). The graphs show that the optimum dynamic
compliance decreases as the upper bound constraint on Cs is increased. In Fig.
4(b), iteration histories are shown for minimum compliance topology design of
the plate subject to a given upper bound constraint on the static compliance
( C s d C s 0.5 u 10 7 ) for four different higher loading frequencies. These
graphs show that for the higher loading frequency designs, the dynamic
compliance of the structure decreases as the prescribed loading frequency is
increased. This feature is opposite to that obtained by minimum compliance
topology design subject to prescribed lower or medium loading frequencies. As a
conclusion, variations of the minimum dynamic compliance with respect to
different loading frequencies are depicted in Fig. 5(a), and Fig. 5(b) presents the
static compliances associated with the minimum dynamic compliance designs
subject to different prescribed loading frequencies.
-7 -7
x 10 x 10
1.5 1.5
Minimum dynamic compliance Cd
Dynamic
Design with Dynamic design without
minimum dynamic compliance designs
constraint constraint on Cs Dynamic

design with
Cs <= Cs
1 -7
1 constraint
Cs = 0.5x10
Cs <= Cs
Static compliance Cs of
-7
Cs = 0.5x10
0.5 0.5
Dynamic design without
constraint on Cs
0
:opt 0
:opt
0 50 100 150 200 0 50 100 150 200
Loading frequency Zp Loading frequency Zp
(a) (b)
Figure 5. (a) Minimum dynamic compliances Cd vs. different loading frequencies. (b)
Static compliances Cs (correspond to the same loading amplitude but zero frequency)
associated with the designs in Fig. 5(a) vs. different loading frequencies. Note that if the
prescribed loading frequency is close to or higher than the optimum value :opt = 127.6 of
the fundamental eigenfrequency for the corresponding problem of free vibrations of the
plate, an upper bound constraint C s d C s is prescribed for the static compliance in order to
avoid obtaining a statically too weak structure from the dynamic design.
3.2 Topology design of a 2D inlet subjected to hydrodynamic pressure

loading
Figure 6. Optimized single-material topologies (40% volume fraction) of 2D inlet for

three different loading frequencies. (a) Admissible design domain. (b) Zp = 0. (c)
Z p 800 . (d) Z p 1000 .
As an extension relative to traditional topology optimization with design-

independent loading, we now consider an example where the dynamic loading is
design-dependent, i.e., both the locations and directions of the loading change as
the structural topology changes. A method developed by Du and Olhoff (2004a,
2004b) (see also Hammer and Olhoff, 1999, 2000) is employed to handle the
design problems associated with this type of loading.
The example deals with optimum topology design of a single-material inlet
for fluid flow. The fluid flow in the channel of the initial inlet is as shown in Fig.
6(a), and is assumed to exert a uniform hydrodynamic pressure loading of given
frequency and amplitude on the inner surface of the inlet. Note that the loading
from the hydrodynamic pressure is design-dependent, as it changes with changes
of the inner surface of the inlet. The material of the inlet is isotropic with
Young’s modulus E = 107, Poisson’s ratio X = 0.3 and the specific mass Jm = 1.
The design objective is to minimize the dynamic compliance of the inlet. Fig. 6(a)
shows the admissible design domain and the initial loading boundary. Figs. 6(b-d)
show optimized topologies and the associated loading boundaries for three given
loading frequencies Z p 0 (static loading), Z p 800 and Z p 1000 . These
loading frequencies are all lower than the maximum fundamental eigenfrequency
of free vibrations of the inlet, which was found to be :opt = 1328. The
corresponding optimum topology of the inlet (with the same material volume
fraction as before) is shown in Fig. 7.
Figure 7. Optimum topology of the 2D inlet (for 40% volume fraction) obtained by
maximizing the fundamental eigenfrequency of free vibrations of the inlet. The maximum
fundamental eigenfrequency is :opt = 1328.
By comparing the optimum topologies in Figs. 6 and 7 we find some

interesting features. Thus, when the loading frequency is much lower than the
optimum fundamental eigenfrequency of the structure, the resulting topology
(see Fig. 6(c)) obtained by the dynamic design of the present paper is similar to
the static design that sustains the amplitude of the loading at zero frequency, see
Fig. 6(b), which implies that the dynamic design is dominated by the spatial
distribution of the amplitude of the external loading vector. However, if the
loading frequency is closer to the value of the optimum fundamental
eigenfrequency of the structure, the design is dominated by the dynamic
requirement, and drives the fundamental eigenfrequency of the structure as far
away as possible from the prescribed loading frequency. For an intermediate
value of the loading frequency, the optimum topology of the inlet is a kind of
compromise between the loading amplitude dominated design and the
eigenfrequency dominated design (cf. Figs. 6(b-d) and Fig. 7).
4 Conclusions
Topological design with the objective of minimizing the dynamic compliance

(maximizing the integral dynamic stiffness) of continuum structures subjected to
time-harmonic forced vibration with prescribed frequency and amplitude of the
dynamic loading, was studied in Sections 2 and 3. The results in Section 3 show
that the design objective yields structures whose eigenfrequencies of free
vibration are generally far from the given excitation frequency of the dynamic
loading, which implies efficient avoidance of resonance phenomena and
reduction of the vibration level of the structure.
It was demonstrated that the design objective may be implemented along
different optimization paths according to different levels of the given external
excitation frequency Zp. For cases where the loading has a lower or medium
value of Zp, the minimum dynamic compliance design process may be driven by
a continuation technique where the loading frequency is sequentially increased
from a sufficiently low initial value up to its prescribed value, Zp. This procedure
delivers the desired result that the optimum structure is associated with minimum
dynamic compliance subject to the prescribed loading frequency, and also
implies an effective improvement (decrease) of the static compliance of the
structure. On the other hand, if the excitation frequency was prescribed to be
somewhat larger than the maximum obtainable value :opt of the fundamental
eigenfrequency of the corresponding free vibration problem, we found that the
increase of the dynamic compliance was accompanied by a drastic decrease of
the static compliance of the structure. Thus, especially when minimizing the
dynamic compliance for single-material structures, we found it to be expedient to
introduce an upper bound constraint on the static compliance in order to maintain
a reasonable static stiffness of the design.
References
Calvel, S., Mongeau, M., 2005. Topology optimization of a mechanical component
subject to dynamic constraints. Technical Report LAAS N0 05374. Available from
http://mip.ups-tlse.fr/perso/mongeau.
Du, J., Olhoff, N., 2004a. Topological optimization of continuum structures with design-
dependent surface loading - Part I: New computational approach for 2D problems.
Struct Multidisc Optim 27: 151-165.
Du, J., Olhoff, N., 2004b. Topological optimization of continuum structures with design-
dependent surface loading - Part II: Algorithm and examples for 3D problems. Struct

Hammer, V.B., Olhoff, N., 1999. Topology optimization with design dependent loads. In
Proc. WCSMO-3, Third World Congress of Structural and Multidisciplinary
Optimization. Buffalo, NY, May 17-21, 1999.
Hammer, V.B., Olhoff, N., 2000. Topology optimization of continuum structures
subjected to pressure loading. Struct Multidisc Optim 19: 85-92.
Jensen J.S., 2007. Topology optimization of dynamic problems with Padé approximants.
Int J Numer Meth Engng 72: 1605-1630.
Kang, B.-S., Park, G.-J., Arora J.S., 2006. A review of optimization of structures
subjected to transient loads. Struct Multidisc Optim 31: 81-95.
Kang, Z., Zhang, X., Jiang, S., Cheng, G., 2012. On topology optimization of damping
layer in shell structures under harmonic excitations. Struct Multidisc Optim 46(1): 51-
67.
Ma, Z.D., Kikuchi, N., Cheng, H.C., 1995. Topological design for vibrating structures.
Comput Methods Appl Mech Engrg 121: 259-280.
Min, S., Kikuchi, N., Park, Y.C., Kim, S., Chang, S., 1999. Optimal topology design of
structures under dynamic loads. Struct Optim 17: 208-218.
Optimization WCSMO4, Rio de Janeiro, Brazil 2005, 8 pp.
Shu, L., Wang, M.Y., Fang, Z., Ma, Z., Wei, P., 2011. Level set based structural
topology optimization for minimizing frequency response. J. Sound and Vibration
330(24): 5820-5834.
Tcherniak, D., 2002. Topology optimization of resonating structures using SIMP
method. Int J Numer Meth Engng 54: 1605-1622.
Yang, X., Li, Y., 2013 Topology optimization to minimize the dynamic compliance of a
bi-material plate in a thermal environment. Struct Multidisc Optim 47(3): 399-408.
Yoon, G.H., 2010. Structural topology optimization for frequency response problem
using model reduction schemes. Computer Methods in Applied Mechanics and
Engineering 199(25-28): 1744-1763.
Topological Design for Minimum Sound Emission
from Structures under Forced Vibration

1
2
Abstract This paper is devoted to topology optimization problems formulated with the
design objective of minimizing the sound power radiated from the structural surface(s)
into a surrounding acoustic medium. Bi-material elastic continuum structures without
material damping are considered. The structural vibrations are excited by time-harmonic
external mechanical loading with prescribed excitation frequency, amplitude, and spatial
distribution. Several numerical results are presented and discussed for bi-material plate-
like structures with different sets of boundary and loading conditions.
1 Introduction
In the literature on structural acoustic and vibration design, a large number of

studies has been devoted to optimization of the size, shape, topology, and
material parameters of structures (see, e.g., Koopmann and Fahnline 1997,
Christensen et al. 1998, Kollmann 2000, Munjal 2002, and papers cited therein).
Specifically, during the last decade, topology optimization was found to be a
very powerful tool for structural design with respect to criteria from acoustics
(Luo and Gea 2003, Wadbro and Berggren 2006, Olhoff and Du 2006, Du and
Olhoff 2007, 2010, Dühring et al. 2008, Yamamoto et al. 2009, Niu et al. 2010,
Nandy and Jog 2011, Du et al. 2011, Yang and Du 2013, Kook et al. 2012).
The present paper, which lends itself to Olhoff and Du (2006) and Du and
Olhoff (2007), is devoted to topological design optimization of vibrating bi-
material elastic structures of given volume, domain and boundary conditions,
with the objective of minimizing the sound power radiated from the structural
surfaces into a surrounding acoustic medium. As in the preceding paper Olhoff
and Du (2013D) and in Olhoff and Du (2013), the structural vibrations are
excited by a time-harmonic mechanical loading with prescribed excitation
frequency and amplitude, and structural damping is not considered. It is assumed
that air is the acoustic medium and that a feedback coupling between the acoustic
medium and the structure can be neglected. Certain conditions are assumed,

DOI 10.1007/978-3-7091-1643-2_14, © CISM, Udine 2014
where the sound power radiated from the structural surface can be estimated by
using a simplified approach instead of solving the Helmholtz integral equation.
This reduces the computational cost of the structural-acoustical analysis
considerably.
2 Formulation for Minimization of Sound Radiation

In this section, we consider topological design optimization of a vibrating bi-
material elastic structure with the objective of minimizing the total sound power
(energy flux) radiated from the structural surface S into a surrounding acoustic
medium. The structural vibrations are assumed to be excited by a time-harmonic
iZ t
mechanical surface loading vector p(t ) Pe p with prescribed forcing
frequency Zp and amplitude vector P on S or part thereof. Assuming that
damping can be neglected, the corresponding structural displacement response
iZ t
vector can be stated as Ue p , and the problem of minimizing the sound power
can be formulated as follows,
° 1 ½°
min ®3 ³ I n dS ³ Re( p f vn* )dS ¾
Ue °̄ S S
2 °¿
subject to :
(K Z p2 M )U P LP f , (1)
CD P f GU HP f ,
NE
¦ U eVe V *1 d 0 , (V *1 DV0 ) ,
e 1
0 d Ue d 1 , (e 1,, N E ) .
Here, the symbols pf and vn* in the expression for represent the acoustic
pressure and the complex conjugate of the normal velocity of the structural
surface, and Pf denotes the corresponding vector of amplitudes of the acoustic
pressure on the structural surface S. The symbol L represents the fluid-structural
coupling matrix and the symbols K and M denote the N dimensional structural
stiffness and mass matrices, where N is the number of DOFs. The expression
K Z p2 M in (1) represents the dynamic stiffness matrix which we may denote
by KD. The matrices G, H and CD can be generated by the discretized Helmholtz
integral and calculation of the spatial angle along the structural surface (see, e.g.,
Christensen et al. 1998). We consider a bi-material design problem (see Sub-
section 2.3 in Olhoff and Du 2013A) where NE is the total number of finite
elements and the symbol Ue denotes the volumetric density of the stiffer material
in element e and plays the role of the design variable in the problem. The symbol
D denotes the fraction of the given volume V *1 of the stiffer material (material *1)
and is given by V *1 /V0 , where V0 is the volume of the admissible design domain.
The remaining part of the total volume V0 is occupied by a softer material
(material *2) as explained in Sub-section 2.3 of Olhoff and Du 2013A.
2.1 Calculation of sound power flow from the surface of a vibrating

structure
The first two constraint equations in (1) denote the structural-acoustic coupling
equations (without incoming acoustic waves) and imply quite complicated
computations since these equations must be solved in each iterative step of the
solution procedure. For simplification, one may consider a special case where the
vibration frequency of the structure has a sufficiently high value. In this case, the
radiation impedance at the boundary of the structure is approximately the same
as the characteristic impedance of the acoustic medium (see Herrin et al. 2003,
and Lax and Feshbach 1947), which implies that the acoustic pressure pf and
normal velocity vn of the structural surface approximately satisfy the following
linear relationship
p f J f cvn (2)
where c is the sound speed and J f is the specific mass (mass density) of the
acoustic medium. Tests performed by Sorokin (2005) for simple beam and
sphere examples show that the accuracy of (2) depends on not only the frequency
level but also on the size of the structure and the shape of the vibration mode of
the structure. Generally speaking, the accuracy of the approximation increases
with increasing values of the frequency, but may decrease with a change of the
vibration mode. Nevertheless, the tests also show that even for lower frequencies,
(2) may still yield a good approximation of the distribution (up to a multiplying
factor) of the sound pressure along the structural surface. This is useful for our
problem of optimizing the global sound radiation, because even a scaled
distribution of the sound pressure along the structural surface can yield a
topology design which is close to the optimum one.
If we further assume weak coupling, i.e., ignore the acoustic pressure in the
structural equation, the first constraint in (1) will be simplified to the equation of
a vibrating structure subjected only to the external mechanical loading P (see the
third equation of Eqs. (1) and the paper Olhoff and Du, 2006),
(Z p2 M K )U P (3)
With the above simplification, the first two constraint equations in problem
(1) may be replaced by Eqs. (2) and (3), and the sound power flow from the
structural surface can now be calculated in a very efficient way, which is

illustrated briefly as follows.
iZ t
Substituting pf by (2) and noting that vn n u(iZ p e p ) , where n is the
unit normal and u is the amplitude of the displacement, the sound power flow in
(1) may be restated as
1 (4)
3 ³ J f cZ p2 (n u)(n u) dS
S
2
Using the finite element interpolation u NU e , where N is the shape

function and Ue is the nodal displacement vector of element e, Eq. (4) may be
rewritten in matrix form as
1 (5)
3 J f cZ p2 U T S n U
2
NE NE § ·
where S n ¦ S ne ¦ ¨ ³ N T nn T N dS ¸
may be termed the surface normal matrix,
e 1 e 1
¨ ¸
© Se ¹
and U is the global nodal displacement vector of the structure as in (1).
2.2 Sensitivity analysis

The sensitivity of the objective function (i.e. the sound power flow) in problem
(1) with respect to the design variables Ue is given by
3 c J f cZ p2 U T S n Uc (6)
where prime denotes partial derivative with respect to Ue. Using Eq. (3) and
applying the adjoint method, see Tortorelli and Michaleris (1994), the sensitivity
(6) of the sound power flow may be calculated in a more efficient manner, which
gives the following result
>
3 c J f cZ p2 U Ts Pc U Ts (K c Z p2 M c) U @ (7)
Here Us is the solution to the equation (K Z p2 M ) U s S n U { f s , where fs may be

regarded as a pseudo surface load vector. Specifically, we only consider the case
of design-independent mechanical load in the present paper, so the sensitivity
Pc of the mechanical load in (7) will be zero. The sensitivities of the stiffness
and mass matrices, i.e. K c and M c , can be derived by introducing the material
models (see Sub-section 2.3 in Olhoff and Du 2013A).
Topological Design for Minimum Sound Emission… 345
Based on the above sensitivity results, the optimization problem (1) may be
solved by using the well-known MMA method (see Svanberg 1987) or an
optimality criterion method, e.g. the fixed point method (see Cheng and Olhoff
1982).
3.1 Minimization of sound power radiation from a clamped bi-material

plate excited by uniform harmonic pressure loading
(a) (b)
Figure 1. Plate-like structure (a=20, b=20, t=1) subjected to uniformly distributed
harmonic pressure loading on its upper surface. All edges of the plate are clamped.
The first example concerns optimum topology design of a bi-material plate-like

structure with clamped edges (see Fig. 1). A time-harmonic, uniformly
distributed, transverse external load p(t) = PcosZt (with P equal to unity) is
applied to the upper surface of the plate. The design objective is to minimize the
total sound power radiated from the surface of the plate to its surrounding
acoustic medium, i.e. air, for a prescribed loading frequency Z = Zp and a
volume fraction of up to 50% for the given stiffer material *1, which has the
Young’s modulus E *1 1011 , Poisson’s ratio X = 0.3 and the specific mass J m*1 =
7800 (SI units are used throughout). The soft material *2 has the properties
E *2 0.1E *1 and J m*2 = 0.1 J m*1 , and X = 0.3. The specific mass of the fluid (i.e.
air) is J f = 1.2 and the sound speed c = 343.4.
(a) Zp = 10 (b) Zp = 100 (c) Zp = 200
(d) Zp = 300 (e) Zp = 400 (f) Zp = 500
(g) Zp = 600 (h) Zp = 700 (i) Zp = 1000

Figure 2. Optimum topologies of clamped bi-material plate-like structures obtained by
minimization of the total sound power radiation subject to nine different loading
frequencies. (a) Zp = 10. (b) Zp = 100. (c) Zp = 200. (d) Zp = 300. (e) Zp = 400. (f) Zp =
500. (g) Zp = 600. (h) Zp = 700. (i) Zp = 1000. A uniformly distributed time-harmonic
pressure loading is applied to the upper surface of the plate.
The plate is modeled by 3D 8-node isoparametric elements in a 40u40u1

mesh. Nine prescribed different loading frequencies, Zp = 10, 100, 200, 300, 500
and 1000 are considered, which cover the designs from a lower frequency level
to a high frequency level. (The fundamental eigenfrequency of the initial uniform

design in Fig. 1(b) is Z1 = 95.) The finite element mesh used here ensures that
the computational results have sufficient accuracy in the frequency range (10 to
1000) considered.
The corresponding optimum topologies of the plate are presented in Figs.
2(a-i), where the stiffer material *1 is represented by black and the soft material
*2 is represented by grey. It can be seen that, as the loading frequency increases,
the optimum topology of the structure shows a more and more complicated
periodicity. In comparison to the initial uniform design, the total sound power
flow into the acoustic medium is reduced from 3.78u10-8 to 3.61u10-9 for Zp =
10, from 3.50u10-4 to 4.01u10-7 for Zp = 100, from 1.31u10-6 to 6.49u10-8 for Zp
= 200, from 2.10u10-6 to 7.43u10-8 for Zp = 300, from 1.06u10-6 to 2.57u10-8 for
Zp = 400, from 3.68u10-7 to 1.84u10-8 for Zp = 500, from 2.47u10-7 to 1.69u10-8
for Zp = 600, from 5.52u10-7 to 1.06u10-8 for Zp = 700, and from 1.03u10-7 to
5.61u10-9 for Zp = 1000. It is noted that the sound power has a remarkable
decrease in the design for the given frequency Zp = 100. The reason is that the
loading frequency Zp = 100 is very close to the first resonance point that
corresponds to the value Z1int 95 of the fundamental eigenfrequency of the
initial design (Fig. 1(b)). For the optimum design (see Fig. 2(b)), we have found
that its first two eigenfrequencies are Z1opt 59 and Z2opt 131 (see Figs. 3(a)-
3(c)), which are far away from the given loading frequency Zp = 100. This
implies that resonance has been avoided effectively by the optimum design and
explains the large decrease of the sound radiation for this design.
(a) Z1opt 59 (b) Z2opt 131 (c) Z3opt Z 2opt 131
Figure 3. The first three eigenmodes of free vibration of the optimum topology design
subject to the given loading frequency Zp = 100. The second and the third
eigenfrequencies constitute a bi-modal eigenfrequency due to the symmetry of the plate.
3.2 Comparison between topology designs of minimum sound radiation

and designs of minimum dynamic compliance
As is evident from (1), the acoustic design objective of minimum sound radiation
considered in the present paper is a global criterion. Now an interesting question
arises, namely, for the same given excitation frequency Z p and distribution of
the external mechanical loading, how much will the design result change if we
instead consider a “comparison problem” where we apply the global design
objective of minimizing the dynamic compliance, and assume that the structure
is elastic, has no material damping, and is subjected to forced vibration in
vacuum?
Optimum topology solutions to this “comparison problem” can be directly
obtained by the method developed in the papers Olhoff and Du (2005, 2013) and
described in Olhoff and Du 2013D. Thus, for the investigation in the present
section, we define the dynamic compliance as Cd | P T U | for the “comparison
problem”. Here, P denotes the amplitude vector of the given time-harmonic
mechanical surface loading (cf. the first paragraph of Section 2), and U
represents the vector of magnitudes of the corresponding structural displacement
response vector that satisfies the dynamic equilibrium equation (3) for the
steady-state vibration at the frequency Z p . The above expression for the
dynamic compliance of our “comparison problem” represents the numerical
mean value of the magnitudes of the surface displacements weighted by the
values of the amplitudes of the corresponding time-harmonic surface loading.
For the case of static loading ( Z p = 0), the expression directly reduces to the
traditional definition of static compliance, namely the work done by the external
forces against corresponding displacements at equilibrium.
It is worth mentioning at this point that it is customary to define the dynamic
compliance differently, namely as the average input power from the external
time-harmonic loading over a load cycle, if the structure possesses material
damping and is subjected to forced vibration in a light acoustic medium without
feedback (see, e.g., Koopmann and Fahnline 1997 and Jog 2002). By the law of
conservation of energy, this input power is equal to the sum of the sound power
radiated from the structure to the light acoustic medium and the power dissipated
due to the material damping in the structure. This means that if the structure is
subjected to forced vibration in vacuum and the material damping is ignored (as
is the case for our “comparison problem”), then the dynamic compliance defined
as the input power from the external loading, will be equal to zero and hence
inapplicable as an objective function for our “comparison problem”.
We now consider the initial structure and the same boundary and loading
conditions as in Fig. 1, but instead of minimizing the total sound power
(a) Zp = 10 (b) Zp = 100 (c) Zp = 200
(d) Zp = 300 (e) Zp = 500 (f) Zp = 1000

Figure 4. Optimum topologies of clamped, quadratic bi-material plate-like structures
obtained by minimizing the dynamic compliance in vacuum for six different given
excitation frequencies. (a) Zp = 10. (b) Zp = 100. (c) Zp = 200. (d) Zp = 300. (e) Zp = 500.
(f) Zp = 1000. A uniformly distributed harmonic pressure loading of constant amplitude is
applied to the upper surface of the plate.
emitted from the structural surface, we use the method described in Section 2 of
the paper Olhoff and Du 2013D) to minimize the dynamic compliance
C d | P T U | of the plate-like structure in vacuum (and assuming vanishing
structural damping). Optimum topologies for the latter problem corresponding to
the six prescribed loading frequencies Zp = 10, 100, 200, 300, 500 and 1000 are
shown in Figs. 4(a-f).
It is seen that for the low excitation frequencies Zp = 10 and Zp = 100, the
optimum topologies shown in Figs. 4(a) and 4(b) are almost indistinguishable
from the corresponding ones in Figs. 2(a) and 2(b). However, as the value of the
excitation frequency is increased, the differences between the topologies become
more pronounced (see Figs. 4 and 2 for Zp = 200, 300, 500 and 1000).
For further comparison we have calculated the values of both the sound
power and the dynamic compliance of the initial structure and the optimum
structures corresponding to the two different design objectives behind Figs. 2 and
4 (assuming the same loading and boundary conditions), and the results are
presented in Table 1.
These results provide numerical evidence that topology optimization with
respect to either of the two design objectives has a strongly improving effect on
the other objective.

Thus, substantial reductions in sound power radiated from a structure
immersed in a light acoustic medium like air can be achieved by using the
method of topology optimization described in Section 2 of Olhoff and Du 2013D
to minimize the dynamic compliance defined above for a structure in vacuum.
This approach may sometimes be attractive because it circumvents the need to
solve the acoustical equations, thereby achieving saving in computational cost.
On the other hand, the results in Table 1 show that for the higher values of the
excitation frequency Z p , significant further reductions can be obtained by
directly minimizing the sound power radiation, and the expense in computational
cost for this is indeed very limited, if one adopts the simplified approach applied
in this paper for calculation of the sound power flow from the surface of the
vibrating structure.
Table 1 Comparisons between designs of minimum sound power and designs of minimum
dynamic compliance.
Optimum designs of
Optimum designs of
Designs minimum sound
Initial uniform design minimum dynamic
power flow from the
(see Figure 1) compliance in vacuum
Excit- structural surface
(see Figure 4)
ation (see Figure 2)
Frequency Cd opt Cd Cdopt
-8 -5 -9 -6 -9
Zp = 10 3.78u10 1.39u10 3.61u10 4.43u10 3.68u10 4.42u10-6
Zp = 100 3.50u10-4 1.29u10-4 4.01u10-7 4.37u10-6 4.02u10-7 4.36u10-6
Zp = 200 1.31u10-6 3.17u10-6 6.49u10-8 3.01u10-7 1.28u10-7 1.47u10-8
Zp = 300 2.10u10-6 4.76u10-7 7.43u10-8 3.08u10-7 3.56u10-7 5.74u10-10
Zp = 500 3.68u10-7 6.84u10-7 1.84u10-8 8.49u10-8 4.46u10-8 7.85u10-9
Zp = 1000 1.03u10-7 2.09u10-7 5.61u10-9 4.18u10-8 7.41u10-8 9.29u10-10
3.3 Minimization of sound power radiation from a corner-supported bi-

material plate excited by concentrated harmonic loading
In order to investigate the influence of the boundary conditions and load
distribution on the optimum topologies, we consider again the design problem of
the bi-material plate as in Fig. 1, but in combination with another set of boundary
and loading conditions. Thus, the plate is now simply-supported at its four
corners and subjected to a concentrated time-harmonic load at the center of the
upper surface (see Fig. 5). Six prescribed values of the loading frequency, Zp =
10, 100, 300, 500, 700 and 1000 are considered separately here to generate the
corresponding optimum topologies of the bi-material plate. The optimum
topology results subject to the six different frequencies are presented in Figs.
6(a-f).
Figure 5. Corner-supported plate subjected to a concentrated harmonic load at the center

of its upper surface.
(a) Zp = 10 (b) Zp = 100 (c) Zp = 300
(d) Zp = 500 (e) Zp = 700 (f) Zp = 1000
Figure 6. Optimum topologies of the corner-supported bi-material plate-like structure (cf.

Fig. 28) for six different loading frequencies. (a) Zp = 10. (b) Zp = 100. (c) Zp = 300. (d)
Zp = 500. (e) Zp = 700. (f) Zp = 1000. A concentrated harmonic load is applied at the
center of the upper surface of the plate.
It is seen that the optimum topologies associated with the same loading
frequencies as in Figs. 2 and 6 are quite different, which implies that the
boundary and loading conditions may have large influence on the resulting
topology of the plate-like structure.
It is noteworthy that in the optimum topologies subject to the concentrated
harmonic load (see Fig. 6), the central part of the plate is always filled out with
the stiffer material *1 which also has a higher mass density. This local layout is
very efficient in counteracting the concentrated load which has not only a given
frequency but also a prescribed amplitude. The mass assembly surrounding the
point of action of the force yields a large local inertia, which effectively reduces
the displacement amplitude at the point of action of the force, and thereby
reduces the vibration amplitudes and the density of sound power emission all
over the plate.
Figs. 7 and 8 show comparisons of the power flow distribution between the
initial designs (subject to the two different boundary and loading conditions in
Figs. 1 and 5) and the corresponding optimum designs for the prescribed
excitation frequency Zp = 1000. Similar comparisons for corresponding designs
subject to the prescribed frequency Zp = 500 are given in Figs. 9 and 10. It is
seen that the distribution of the sound power in the optimum designs subject to
the concentrated load show features that are very different from those in the
optimum designs subjected to the uniform load.
(a) (b)
Figure 7. Distribution of the power flow from the structural surface of the initial design.
(a) and (b) correspond to the two different sets of boundary and loading conditions shown
in Figs. 1 and 5, respectively. The loading frequency has the prescribed value Zp = 1000.
(a) (b)
Figure 8. Distribution of the power flow from the structural surface of the optimum
design. (a) and (b) correspond to the two different sets of boundary and loading conditions
shown in Figs. 1 and 5, respectively. The loading frequency has the prescribed value Zp =
1000.
Thus, we find that for the higher excitation frequencies, the main part of the
sound power emitted from the optimum designs subjected to the concentrated
load (see Figures 8(b) and 19(b)) is limited within a small annular-like area
surrounding the mass assembly in the vicinity of the concentrated load. This
implies that the optimum designs (see Figures 6(d) and 6(f)) create an efficient
isolation of vibration and sound power radiation that to a large extent terminates
the transmission of bending waves to the boundary of the plate-like structures at
the prescribed frequencies. The features discussed here reveal very strong
similarities between the present optimum designs and so-called band-gap
structural designs (see, e.g., Halkjær et al., 2006).
(a) (b)
Figure 9. Distribution of the power flow from the structural surface of the initial design.
(a) and (b) correspond to the two different sets of boundary and loading conditions shown
in Figs. 1 and 5, respectively. The loading frequency has the prescribed value Zp = 500.
(a) (b)
Figure 10. Distribution of the power flow from the structural surface of the optimum
design. (a) and (b) correspond to the two different sets of boundary and loading conditions
shown in Figs. 1 and 5, respectively. The loading frequency has the prescribed value Zp =
500.
4 Conclusions
Problems of topological design of vibrating bi-material elastic structures with the

objective of minimizing the sound power radiated from a structural surface into a
light acoustic medium, were considered in this paper. As in the preceding paper,
the structures were subjected to forced vibration of prescribed excitation
frequency and force amplitude distribution.
For sufficiently high excitation frequencies, instead of solving the Helmholtz
integral equation, the sound pressure at the structural surface is determined
approximately by the product of the characteristic impedance of the acoustic
medium and the normal velocity of the structural surface. This simplifies the
structural-acoustic analysis and substantially reduces the computational cost.
Actually, even for relatively low frequencies, the above simplification may yield
a good approximation of the distribution of the sound pressure over the structural
surface, which is very useful in the present context. An extended SIMP model
was used for the topology design of the bi-material structures, where the same
penalization was applied for the stiffness and mass of the structural volume
elements. Sensitivities of the design objective were derived by an adjoint
method, and the optimization problem was solved by using the MMA method.
Numerical results are presented for bi-material plate structures with different
loading and boundary conditions, and interesting features of the optimum
structural topologies, power flow distributions and sound pressure waves, are
revealed and discussed. Main conclusions are:
The sound power radiation from structures subjected to forced vibration can
be considerably reduced by topology optimization. Just as was demonstrated in

(Olhoff and Du 2013, 2013D) on topological design for minimum dynamic
compliance of structures in vacuum, this is achieved by creation of an optimum
design with a large gap between two adjacent eigenfrequencies of free vibrations,
with the given excitation frequency placed approximately in the middle of the
gap.
Along these lines, topology optimization of structures with respect to either
of the two objectives (i) minimum sound radiation into a surrounding light
acoustic medium and (ii) minimum dynamic compliance in vacuum, has a
strongly improving effect on the other objective. For structures subjected to the
same excitation frequency, spatial distribution of the external mechanical loading,
and the same boundary conditions, the optimum topologies associated with the
design objectives (i) and (ii) are almost indistinguishable at lower excitation
frequencies but become more different at higher frequencies.
Independently of the spatial distribution of the external dynamic loading and
the boundary conditions, the optimum topologies of the bi-material plate
structures exhibit different types of periodicities when the excitation frequency is
increased to values above a number of the lowest resonance frequencies of the
structures.
When subjected to a concentrated harmonic load at such a value of the
excitation frequency, the optimum structural topology shows a local assembly of
the heavier material around the point of action of the load. Due to the given
amplitude of the load, the inertia of this local mass assembly does not only
reduce the displacement amplitude of the point of load action, but also the
general level of vibration and sound power radiation of the structure. Almost all
the (reduced) sound power radiation from the optimum structure is limited to
take place from within a small ring-shaped area of the weaker material around
the aforementioned local mass assembly in the case of plates. This means that
almost all of the power input from the external harmonic load is radiated as
acoustic power from these locations close to the point of action of the load,
whereby only a very small part of the power input is transmitted to the plate
boundary.
Overall, the problem studied in this paper has several features in common
with problems of topological design optimization of band-gap structures.
References
Christensen, S.T., Sorokin, S.V., Olhoff, N., 1998. On analysis and optimization in
structural acoustics – Part I: Problem formulation and solution techniques. Struct

Optim 16: 83-95.
Du, J., Olhoff, N., 2007. Minimization of sound radiation from vibrating bi-material
Du, J., Song, X., Olhoff, N., 2011. Topological design of acoustic structure based on the
BEM-FEM format and the mixed formulation. In: Proc. 9th World Congress on
Structural and Multidisciplinary Optimization (WCSMO-9), June 13-17, 2011,
Shizuoka, Japan.
Dühring M.B., Jensen J.S., Sigmund O., 2008. Acoustic design by topology optimization.
J Sound Vib 317: 557-575.
Herrin, D.W., Martinus, F., Wu, T.W., Seybert, A.F., 2003. A New Look at the High
Frequency Boundary Element and Rayleigh Integral Approximations. Society of
Automotive Engineers, Inc.
Kook, J., Koo, K., Hyun, J., Jensen, J.S., Wang, S., 2012. Acoustical topology
optimization for Zwicker’s loudness model – Application to noise barriers. Computer
Methods in Applied Mechanics and Engineering 237-240: 130-151.
New York.
Lax, M., Feshbach, H., 1947. On the radiation problem at high frequencies. Journal of
the Acoustical Society of America 19: 682-690.
Luo J.H., Gea H.C., 2003. Optimal stiffener design for interior sound reduction using a
Nandy, A.K., Jog, C.S., 2011. Optimization of vibrating structures to reduce radiated
noise. Struct Multidisc Optim 45(5): 717-728.
Niu B., Olhoff N., Lund E., Cheng G., 2010. Discrete material optimization of vibrating
laminated composite plates for minimum sound radiation. Int J Solids Struct 47: 2097-
2114.
Optimization WCSMO4, Rio de Janeiro, Brazil 2005, 8 pp.
Olhoff, N., Du, J., 2006. Topology optimization of vibrating bi-material structures with
respect to sound radiation. In Proc. IUTAM Symposium on Topological Design
Optimization of Structures, Machines and Materials – Status and Perspectives (held
in Copenhagen 2005). Dordrecht, The Netherlands: Springer, p. 43-52.
Olhoff, N., Du, J., 2013D. Topological design for minimum dynamic compliance of
Sorokin, S.V., 2005. Private communication.
Wadbro E., Berggren M., 2006. Topology optimization of an acoustic horn. Comput
Yamamoto, T., Maruyama, S., Nishiwaki, S., Yoshimura, M. 2009. Topology design of
multi-material soundproof structures including poroelastic media to minimize sound
pressure levels. Computer Methods in Applied Mechanics and Engineering 198(17-
20): 1439-1455.
Discrete Material Optimization of Vibrating
Laminated Composite Plates for Minimum Sound
Emission
Niels Olhoff and Bin Niu

DK-9220 Aalborg, Denmark
Abstract This paper deals with vibro-acoustic optimization of laminated

composite plates without damping. The vibration is excited by
time-harmonic external mechanical loading with prescribed frequency and
amplitude, and the design objective is to minimize the total sound power
radiated from the surface of the laminated plate to the surrounding
acoustic medium. Instead of solving the Helmholtz equation for evaluation
of the sound power, advantage is taken of the fact that the surface of the
laminated plate is flat, which implies that Rayleigh’s integral
approximation can be used to evaluate the sound power radiated from the
surface of the plate. The novel Discrete Material Optimization (DMO)
formulation has been applied to achieve the design optimization of fiber
angles, stacking sequence and selection of material for laminated
composite plates. Several numerical examples are presented in order to
illustrate this approach.
1 Introduction
Composite materials like fiber reinforced polymers (FRPs) are being used
increasingly in aerospace vehicles, maritime carriers, wind turbine blades, and
various mechanical equipment where high strength, high stiffness and low
weight are important properties. In such applications, the FRPs are usually
stacked in a number of layers, each consisting of strong fibers bonded together
by a resin, to form a laminate. In addition, laminated sandwich structures may
also consist of layers made of foam material. When these composite structures
are used in dynamic environments, vibration control and noise reduction become

DOI 10.1007/978-3-7091-1643-2_15, © CISM, Udine 2014
of great technical significance.

The present paper lends itself to (Niu et al. 2010), and considers the design
objective of minimizing the sound radiation from laminated composite plates
subjected to forced, time-harmonic vibration. This objective is accomplished by
optimizing simultaneously the laminates in terms of proper choice of material,
stacking sequence and fiber orientation. Such advanced optimization of
laminates has only recently become possible via the development of the method
of Discrete Material Optimization (DMO) in the papers (Stegmann and Lund
2005, Lund and Stegmann 2005), and practically feasible via new developments
in the manufacture of composite materials and structures. The DMO method is
based on ideas from multiphase topology optimization (Sigmund and Torquato
1997) and applies a parameterization where a predefined set of discrete candidate
materials and fiber angles are chosen as design variables at element level.
As in the preceding paper (Olhoff and Du 2013E), we shall assume that air
is the acoustic medium, and that a feedback coupling between the acoustic
medium and the structure can be neglected. Rayleigh’s integral is used for
computation of the sound power radiated from the structural surface instead of
solving the Helmholtz integral equation. This implies that the computational cost
of the structural-acoustic analysis can be considerably reduced, which is very
efficient for the design optimization. It has been proved that Rayleigh’s integral
provides a very good approximation of the sound pressure distribution along a
relatively flat structural surface, see Lax and Feshbach (1947), Herrin et al.
(2003), and Du and Olhoff (2007a).
Textbooks on optimization for passive design of structural-acoustic systems
against vibration and noise have been published by (Koopmann and Fahnline
1997, Kollmann 2000), and proceedings (Munjal 2002, Bendsøe et al. 2006) are
available from two IUTAM Symposia. These publications as well as papers like
(Pedersen 1982, Olhoff and Parbery 1984, Bendsøe and Olhoff 1985,
Christensen et al. 1998a,b, Luo and Gea 2003, Sorokin et al. 2006, Bös 2006,
Wadbro and Berggren 2006, Yoon et al 2007, Du and Olhoff 2007a,b, Dühring et
al. 2008, Du and Olhoff 2010, Yang and Du 2013) did not specially address
composite structure applications, but provide an overview and contain references
to the area of design optimization with respect to acoustic criteria in general.
In terms of optimization of composite structures with respect to acoustic
criteria, we may refer the reader to the review article (Denli and Sun 2007) and
the bibliography (Mackerle 2003), and a large number of papers cited therein. As
examples of various types of problems of optimum structural-acoustic design
with composite materials, we may refer to (Hufenbach et al. 2001, Thamburaj
and Sun 2002, Chen et al. 2005; Yamamoto et al. 2008, Jensen 2009, Niu et al.
2010).
The organization of the remainder of this paper is as follows. In Section 2,
the problem of structural topology optimization of laminated composite plates
subject to given amounts of the constituents is formulated for the objective of
Discrete Material Optimization of Vibrating Laminated… 361
minimizing the sound power radiated from the vibrating structural surface into
the acoustic medium. Then Rayleigh’s integral is introduced to calculate the
sound power flow from the structural surface, which leads to a simplified
optimization formulation of the current problem. Section 3 introduces the
parameterization for discrete material optimization (DMO) and discusses penalty
functions for DMO. Section 4 presents a DMO convergence measure. Section 5
deals with the sensitivity analysis required for the numerical optimization
algorithm. Section 6 presents several numerical examples with different
excitation frequencies in order to validate the proposed method, including
single-layer, multi-layer laminated composite plates, and laminated sandwich
plates consisting of layers made of FRPs and foam material. Finally, a section
with conclusions closes the paper.
2 Minimization of Sound Power Radiation for Laminated Composite

Plate using Discrete Material Optimization (DMO)
We consider design optimization of vibrating laminated composite plates with
the objective of minimizing the total sound power (energy flux) 3 radiated
from the structural surface S into a surrounding acoustic medium. The same
objective has been applied in Du and Olhoff (2007a) for a vibrating isotropic
bi-material elastic plate. The present work aims to realize this objective for a
laminated composite plate by design optimization of stacking sequence, fiber
orientations, and selection of layer materials. This problem is solved by using the
so-called Discrete Material Optimization (DMO) approach in the sense that the
structural constituents are chosen from among a given set of different candidate
materials (Stegmann and Lund 2005), which may be regarded as an extension of
classical topology optimization (see, e.g., Bendsøe and Sigmund 2003, Rozvany
et al. 1992) with a constraint on the total volume of material.
The mathematical formulation of the problem is as follows.
° 1 °½
min ®3 ³ I n dS ³ Re p f vn* dS ¾
x
¯° S S
2 ¿°
S .t. : K Z p2 M U P LP f
ǂǂCD P f GU - HP f (1)
Ne § Nl nl ·
¦ ¨ ¦¦ [ c t ¸A e d R
eli i l
©l 1 i 1
e 1 ¹
0 xmin d x j d xmax 1, j 1, N dv
In the expression for the total radiated sound power 3 in (1), the symbols p f
and vn* represent the acoustic pressure and the complex conjugate of the normal
velocity of the structural surface, and P f denotes the corresponding vector of
amplitudes of the acoustic pressure on the structural surface S. The symbol L
represents the fluid-structural coupling matrix. The matrices G , H and CD
can be generated by the discretized Helmholtz integral and calculation of the
spatial angle along the structural surface (see, e.g., Christensen et al. 1998 a, b),
and P denotes the vector of amplitudes of a given external time-harmonic
iZ t
mechanical loading vector p t = Pe p with the prescribed excitation
frequency Z p . The symbols K and M represent the global structural
stiffness and mass matrices, U denotes the vector of magnitudes of the
iZ p t
corresponding structural displacement response vector a t = Ue , and
Kd K Z M
2
p is defined as the global dynamic stiffness matrix. Here, M
depends on the mass density J i of each of the candidate materials, and it is
assumed that damping can be neglected.
Since laminated composite plates are considered, more design parameters
need to be introduced as compared to single material solid plates. Thus, l
denotes ‘layer’, e denotes ‘element’, N l is the number of layers, N e is the
number of elements, and nl is the number of candidate materials (design
variables) per layer for each element; the number of element design variables
n e for multi-layered elements (with N l layers) is the sum of the number of
Nl
design variables per layer, nl , over all N l layers, such that n e ¦ k 1
nkl , and
e
N
the total number of design variables in the problem is therefore N dv ¦ e
i 1 i
n .
It is emphasized that the ‘classical’ topology optimization formulation has one
design variable per element by setting n e =1 for single material 0/1 design. For
single-layer plates, the number of candidate materials is also the number of
element design variables. Here we consider a more general condition with
multiple layers, where the number of design variables must be summed over all
layers for this element. The symbol tl and Ae represent the thickness of the
l-th layer and area of the e-th element, respectively. In the DMO approach, the
variable xi (i 1, 2, , nl ) per layer for each element can be seen as a local
density variable that indicates possible selection of the i-th material, i.e., with
xi xmax meaning that the i-th material is chosen and xi xmin meaning that
the i-th material is not chosen. The design variables x for all layers in all
elements are denoted as
x {x j j e u l u i, where e 1, 2,..., N e ; l 1, 2,..., N l ; i 1, 2,..., nl } (2)

Thus, in comparison with classical topology optimization of a single material
structure, an extended parameterization is invoked in the DMO approach which
enables penalization of intermediate values of the design variables such that a
distinct choice of candidate material may be made. The symbol [ eli denotes the
weighting function of the i-th candidate material in the l-th layer of the e-th
element in the parameterization of the discrete material optimization (DMO)
formulation, see the subsequent section for a detailed definition. For the sake of
brevity, we use subsequently in Sections 3, 4 and 5 the symbol [i instead of
[ eli to denote the weighting functions in a certain layer of a certain element
without special statements. More discussion on weighting functions of DMO can
be found in Stegmann and Lund (2005).
The symbol R in (1) denotes the given resource, which enforces an upper
bound for the total cost or mass of the structure. The symbol ui is defined as a
so-called “unit cost factor” of the i-th candidate material. Accordingly, for a cost
or a mass constrained problem, ui represents the cost ci or the mass J i per
unit volume of the i-th candidate material in this paper. For the problem of pure
fiber angle optimization, the resource constraint may be left out entirely when
the unit cost factors for all unidirectional fiber materials are assumed equal,
because a change in fiber angle does not influence the total cost or mass.
However, this constraint on resource needs to be considered when multiple
material optimization is implemented, e.g, foam material introduced as candidate
material together with unidirectional fiber material. The notion of unit cost
factors has been used earlier in the context of other structural optimization
problems; see, for example, Taylor (1975), Mroz and Rozvany (1975), Prager
(1977), and Olhoff and Taylor (1978, 1979).
In the problem formulation (1), the structural-acoustic coupling occurs due
to the appearance of the acoustic surface pressure vector P f and the structural
displacement response vector U in both the first and second constraint
equation in (1), i.e., the equation for forced structural vibration without damping
and the discretized Helmholtz integral equation for the acoustic medium. Direct
solution of the Helmholtz integral equation implies large computational cost, and
the equation has to be solved together with the structural equation in each
iterative step of the optimization process. Therefore, for simplification,
Rayleigh’s integral approximation is adopted for determining the acoustic
pressure distribution and sound power radiation.
The conditions for this are that the structure is flat, which is clearly met by
the laminated plates considered, and that the prescribed excitation frequency Z p
is sufficiently high or the observation points are sufficiently far away from the
laminated plate, such that the product kr is much larger than 1, where k= Z p /c is
the wave number, c the speed of sound in the acoustic medium, and r is the
distance between a source point on the structural surface and an observation
point in the acoustic domain, cf. Herrin et al. (2003). Following (Du and Olhoff
2007a), assuming a sufficiently high value of the structural vibration frequency
Z p , the radiation impedance pf /vn at the structural surface will be approximately
equal to the characteristic impedance J f /c of the acoustic medium (Lax and
Feshbach 1947). Thus, the acoustic pressure pf and the normal velocity vn of the
structural surface are approximatively related by the simplified equation
p f J f cvn (3)
where J f is the mass density and c the sound speed in the acoustic medium. The
accuracy of the approximation is discussed in the papers (Du and Olhoff 2007a,
Herrin et al. 2003).
The normal velocity of the surface in (3) can be obtained as
vn n u iZ p (4)
where n is the unit normal and u the amplitude of the displacements of the
structural surface after interpolation based on finite element analysis, i.e., using
the finite element interpolation u = NU e , where N is the shape function and
U e is the nodal displacement vector of element e.
Thus, substituting (3) with vn given by (4) into the expression for the sound
power (objective function) 3 in (1), we after simple algebra obtain the
simplified expression
1
3 ³ J f cZ p2 n u n u dS (5)
S
2
which upon use of the discretized formulations can be written in the matrix form
1
3 J f cZ p2 UT S n U (6)
2
where superscript T stands for transpose, U denotes the global displacement
vector of the structure, and S n defined as
Ne Ne § ·
T
Sn ¦S ne ¦ ¨¨ ³ N nnT NdS ¸ (7)
e 1 e 1 © Se
¸
¹
is termed the surface normal matrix of the structure.
Finally, in this paper we shall consider the acoustic medium to be light, i.e.,
air with the mass density J f =1.2kg/m3 and sound speed c = 343.4m/s. This
means that we may assume weak coupling, i.e., ignore the acoustic pressure on
the structure, and this implies a further simplification of the current optimization
problem. All in all, with the simplifications made, the original formulation (1) of
our optimization problem can be re-written in the very convenient form
1
min 3 J f cZ p2 UT S n U
x 2
S .t. : K Z p2 M U P
N e § N l nl
(8)
·
¦ ¨ ¦¦ [ eli ci tl ¸Ae d R
e 1© l 1 i 1 ¹
ǂǂ0 xmin d x j d xmax 1, j 1, N dv
When comparing (8) with (1), let us first note that the assumption of weak
coupling implies that the feedback acoustic pressure LP f on the structure in (1)
vanishes such that the first constraint in (8) is simply the standard equation for a
vibrating structure without damping, that is subjected to the given external
dynamic loading P, only. At the same time, in (8) the Rayleigh approximation (3)
has dismissed the discretized Helmholtz integral equation in the second
constraint in (1) and instead taken over itself the delivery of acoustic surface
pressures to 3 (cf. the expression for 3 in (1)), and finally substituted these
pressures by surface normal velocities converted into global structural
displacements U in the expression for 3 in (8). We notice that these
displacements U are simply obtained by solution of the standard equation for
forced vibration of the structure in the first constraint of (8).
We may conclude that the application of Rayleigh’s approximation and the
assumption of weak coupling have furnished a formulation (8) of our
optimization that is much simpler than the formulation in (1) based on
Helmholtz’ integral equation and full structural-acoustic coupling. Thus, problem
(8) does not require a system of coupled equations (the first and second
constraint equations in (1)) to be assembled and solved, and is therefore much
easier to implement, much faster to solve numerically, and hence requires
considerably less computer resources than problem (1). Similar advantages must
be expected for the sensitivity analysis of problem (8).
3 DMO Parameterization Model for Topology Optimization of

Multi-layer Structures
The DMO method (Stegmann and Lund 2005, see the thesis Stegmann 2004 for
further details) achieves a parameterization as an extension of the ideas used in
structural topology optimization (Bendsøe and Kikuchi 1988). Thus, instead of
choosing between solid and void, the DMO method realizes the ability to choose
from among a given number of different candidate materials. Hereby, the DMO
formulation extends the scope of application of multiphase topology
optimization (Sigmund and Torquato 1997) and allows the structures to be
multi-layered and made of orthotropic materials.

The element constitutive matrix per layer is expressed as a weighted sum of
the matrices of the candidate materials. Various parameterization schemes have
been developed in Stegmann and Lund (2005) and Stegmann (2004) for discrete
material optimization (DMO), see also Lund and Stegmann (2005) and Lund
(2009). For multi-layer structures, the interpolation method must be implemented
layer-wise for each element, i.e. for all layers in all elements. Consequently, the
interpolation scheme is written by layer, and the constitutive relation for the l-th
layer has the form
nl
Ql ¦[ Q i i [1Q1 [ 2 Q 2 [ n Q n l l (9)
i 1
where each candidate material is characterized by a constitutive matrix Qi . The

weighting functions [i must all attain values between 0 and 1 in order to be
physically acceptable. Furthermore, it is found to be necessary that the sum of
nl
the weighting functions is always one, i.e. ¦ k 1 k
[ 1 , which is important for a
physical interpretation of the designs and for correct evaluation of quantities
such as the mass, weight and cost, etc. In Eq. (9), the parameterization model is
realized element-wise for single-layers and layer-wise for multiple layers for a
large number of candidate materials.
In this paper, a DMO parameterization is used which enforces unit value of
the sum of the weighting functions. The weighting functions can be expressed as
nl
[i
[i ¦ nl
(10)
i 1
¦ k
[
1 k
where
nl
p
[i x
l ª1 x l p º (11)
i «¬ j »¼
j 1; j z i
Here the unity demand is realized by normalizing each weighting function

initially computed by Eq.(11). It has been noted that the effect of the penalized
intermediate values of the weighting functions is slightly reduced when
introducing the normalization. The weighting functions [1 and [ 2 are
illustrated in Figs. 1(a) and (b) for unit value of the penalty factor p, and the sum
of them is shown in Fig. 1(c). It is found that the sum of the weighting functions
attains unit value for any combination of intermediate values of design variables.
For a value of the penalty factor p larger than unity, e.g., p=3, the weighting
functions [1 and [ 2 are shown in Fig. 2(a) and (b), and the sum of them again
attains unit value as indicated in Fig. 2(c). The increase of the penalty factor
slightly increases the size of the flat triangle-like plateau and the slope in the
center is steeper. This implies that the increase of the penalty factor will not help
too much to penalize intermediate values of the design variables. Fig. 3(a), (b)
and (c) depict the weighting functions [1 , [ 2 and their sum if the penalty
factor is further increased up to the value p=10. The differences between Fig. 1
and Fig. 3 are seen to be more significant. We note that the two plateau domains
in Figs. 3(a) and (b) are not favourable for penalization of design variables with
values located in these two domains. Due to this, the power p is typically
increased gradually from 1 to 3 only, and not up to a higher value during the
continuation process.
Similarly, the element mass density and element cost per layer, i.e. for the
l-th layer, are also expressed as a weighted sum for the candidate materials,
respectively,
nl nl
[i
Jl ¦ [i J i ¦ nl
Ji (12)
i 1 i 1
¦ k
[
1 k
nl nl
[i
cl ¦[ c ¦ i i nl
ci (13)
i 1 i 1
¦ k
[
1 k
where J i denotes the mass density and ci the unit cost of the i-th candidate
material. The weighting functions [i use the same interpolation formulae in
Eq. (10).
(a) (b) (c)

Figure 1. Weighting functions, [1 and [ 2 , and the sum of them for two materials with
the penalty factor p=1
(a) (b) (c)

(a) (b) (c)

For the resource constraint, linear interpolation is used, which means that
the penalty power p=1. However, for the stiffness and mass matrices, nonlinear
interpolation is used, and the penalty power p is typically increased gradually
from 1 up to 3 during the iterations.
Furthermore, the element stiffness and mass matrices can be obtained on the
basis of first-order laminated composite plate theory, and the global stiffness and
mass matrices can be obtained by assembling the element matrices.
4 Evaluation of Convergence
A convergence measure given in Stegmann and Lund (2005) is adopted to
describe whether the optimization has converged to a satisfactory result. This
convergence measure is described briefly here, while the reader is referred to the
original paper (Stegmann and Lund 2005) for a detailed discussion. For each
layer in each element the inequality is evaluated for all weighting functions, [i
[i t H [12 [ 22 [ n2 (14)
where H is a tolerance level, typically 95-99.5% suggested by Stegmann and
Lund (2005). If the inequality is satisfied for one of the weighting functions in
the layer, it is flagged as converged. The DMO convergence measure hH is
defined as the ratio between the number of converged layers N cl ,tot in all
elements and the total number of layers in all elements N l ,tot Nl Ne
N cl ,tot
hH (15)
N l ,tot
The DMO convergence measure is denoted as h95 if the tolerance level is
95% and full convergence, h95 1 , means that all layers in all elements have a
single weighting function contributing more than 95% to the Euclidian norm of
the weighting functions.
5 Design Sensitivity Analysis

The sensitivity of the objective function (i.e., the sound power 3 ) in
formulation (8) with respect to the design variable xk is given by
§1 ·
w ¨ J f cZ p2 UT S n U ¸
w3 ©2 ¹ (16)
wxk wxk
Since the surface normal matrix S n is independent of the design variables,
we obtain
w3 w U
J f cZ p2 UT S n (17)
wxk wxk
Using the adjoint method (see, e.g. Tortorelli and Michaleris 1994, Du and
Olhoff 2007a), the sensitivity of the objective function can be obtained in a more
efficient manner as
w3 w U ª § wK wM · º
J f cZ p2 UT S n J f cZ p2 «-UTs ¨ Z p2 ¸ U» (18)
wxk wxk ¬« w
© kx wxk ¹ ¼»
Here U s is the solution to the equation
K Z M U
2
p s S n U = Ps (19)
where Ps may be regarded as a pseudo surface load vector.
The sensitivities of the stiffness and mass matrices can be derived by the
DMO parameterization model introduced in the previous section.
The global resource constraint in (8) that specifies an upper bound value R
for the total cost or mass of the structure, can be transformed into
N e § N l nl ·
¦ ¨ ¦¦ [ eli ci tl ¸Ae
e 1© l 1 i 1 ¹ d1
g (20)
R
and the sensitivities of the corresponding constraint function g with respect to the
design variables are easily obtained as
l
wg tk Ak n § w[i ·
wxk
¦ ¨ ci ¸
R i 1 © wxk ¹
(21)
It should be reiterated that linear interpolation with the penalty factor p=1 is
adopted for the resource constraint. With these sensitivity results, the design
problem (8) may be solved by a mathematical programming method, e.g., MMA
developed by Svanberg (1987). A broad account of finite element based design
sensitivity analysis and optimization is available in Lund (1994).
6 Numerical Experiments of Laminated Composite Plates: Discrete

Material Optimization of Single-layer, Multi-layer and Sandwich
Plates
Uniform
harmonic
pressure load
l=Nl
l=Nl-1
l=Nl-2
FRPs(Tl) l=Nl-3
Foam
l=2
l=1
Figure 4. Quadratic laminated plate structure subjected to uniformly distributed

time-harmonic pressure loading. All four plate edges are clamped
6.1 Single-layer plate

Firstly, the problem of a single-layer clamped quadratic plate subjected to
uniformly distributed time-harmonic pressure loading p t P cos Z p t with
P 105 is considered for minimum sound radiation by Discrete Material
Optimization (DMO). The side lengths of the plate are 1m and the thickness t is
0.01m. All quantities are given in SI units in this paper. Laminated Mindlin plate
elements based on first order shear deformation theory are used in the following
numerical experiments. A nine-node Mindlin laminated plate element with five
degrees of freedom at each node is adopted. An element with this number of
degrees of freedom per node is chosen because in several of the examples in this
paper, the design parameterization allows for coupling between in-plane and
bending deformation, such that the B-matrix in an A-B-D description of the
laminate is non-zero. A 20u20 mesh discretization is applied for the plate and it
should be noted that symmetry is not invoked in the analysis. We have tested that
the finite element mesh used yields sufficiently accurate computational results in
the frequency ranges considered. As candidate materials we use a glass/epoxy
composite with Ex 54 GPa , E y Ez 18GPa , Gxy Gyz Gzx 9GPa ,
Poisson’s ratio vxy 0.25 and mass density J 1900 kg m3 . The fiber angles
are taken to be ª¬90o , r45o , 0o º¼ .
For all the examples in this paper, unbiased initial values of design variables
were set to the value corresponding to uniform distribution of all candidate
materials (in this sub-section 0.25). The corresponding initial design provides a
convenient reference for evaluation and discussion of the vibration and sound
power radiation characteristics of the designs optimized by usage of DMO in the
following. In this section the initial design corresponds to a quasi-isotropic layup
of the glass/epoxy material. Here, we start out computing a lower spectrum of
eigenfrequencies Zi of the initial design and find the results given in Table 1,
where Z=Z=884. is a bimodal (double) eigenfrequency.
Table 1 Eigenfrequencies Zi and resonance frequencies :i of the initial design

Z Z Z Z Z Z
434.=: 884. 884. 1302. 1583. 1591.=:
Here, Z1 and Z6 identify the first and second resonance frequency,

respectively, denoted as :1 and :2. Thus, it is easily shown that the
eigenfrequencies Z2, Z3, …, Z5 in both Table 1 and Table 2 below, are not
resonance frequencies for the forced time-harmonic vibration problems
considered in this section. We recall that the amplitudes of the pressure loading
are assumed to be the same all over the plate domain, and when computing the
scalar products PT I i for i=2, …, 5, where P is the vector of load amplitudes
and I i are the vectors of the eigenmodes associated with the eigenfrequencies
Zi, i=2, …, 5, we find that (within computational accuracy) the scalar product
PT I i 0 for i=2, …, 5. This means that the eigenvectors I i , i=2, …, 5, are
orthogonal to the vector of the load amplitude P , and implies that these
eigenmodes are not excited by the dynamic loading. Hence, the corresponding
eigenfrequencies Z2, …, Z5 are not resonance frequencies of the forced vibration
problem, and will not affect the sound power radiation from the structure. Both
in Table 1 and Table 2, the eigenmodes corresponding to the first and sixth
eigenfrequency Z1 and Z6, respectively, are found not to be orthogonal to the
amplitudes of the dynamic loading, which means that Z1 and Z6 constitute the
first and the second resonance frequency :1 and :2, respectively, of the forced
vibration problem.
For the optimization six prescribed different loading frequencies, Zp =10,
100, 200, 300, 500, and 1000 are considered, which cover designs from the low
frequency to a lower-medium frequency level (note that the fundamental
resonance frequency of the initial design is :1=434.). The optimized designs

with their fiber orientations are shown in Fig. 5. The optimum design for
minimum sound power at the low excitation frequency Zp = 10 resembles the
optimum design for maximum stiffness in Lund and Stegmann (2005). The
DMO convergence measures in Eq. (15) of these six results with different
excitation frequencies are h95 0.82 for Zp =10, h95 0.82 for Zp =100,
h95 0.80 for Zp =200, h95 0.80 for Zp =300, h95 0.99 for Zp =500, and
h95 0.98 for Zp =1000. The DMO convergence measure is seen to be not very
high for the cases with lower values of the excitation frequency. For these cases,
it is observed that the not-fully converged elements are mostly located in the
narrow central region of the optimized plate. In this region, it is found that the
weighting functions associated with unidirectional fiber material oriented at
r45o , 90o and 0o are very close to each other. This issue will be taken up in
Section 7. In the optimized designs in Fig. 5, the material with the highest
associated weighting function is plotted.
(a)Zp=10 (b) Zp=100 (c) Zp=200
(d) Zp=300 (e) Zp=500 (f) Zp=1000

Figure 5. Single-layer plate: Optimum designs for different excitation frequencies
Table 2 Eigenfrequencies Zi and resonance frequencies :i of the optimized design

associated with the excitation frequency Zp=500
Z Z Z Z Z Z
354.=: 834. 834. 1181. 1415. 1492.=:
The optimum designs in Fig. 5 are observed to be similar to each other for
the excitation frequencies 10, 100, 200 and 300, which are all below the first
resonance frequency :1 of the initial design as well as the resulting optimized
design. However, different topologies are found when higher excitation
frequencies are considered. The sound power radiation has generally been
decreased quite substantially for the optimized design relative to the initial
design in all the examples, see Table 3.
Table 3 Single-layer plate: Comparison of the total sound power radiation from the initial
design and the designs optimized for different excitation frequencies Zp
Excitation Initial design Optimized design Relative decrease
frequency Zp Sound power 3 Sound power 3 Fig. |33|/ 3
10 15.61 7.69 5(a) 50.77%
100 1738. 830. 5(b) 52.24%
200 10023. 4263. 5(c) 57.47%
300 51001. 15846. 5(d) 68.93%
500 375728. 91434. 5(e) 75.66%
1000 9310. 7844. 5(f) 15.77%
As can be seen from Table 3, the largest decrease (75.66%) of the sound
power emission 3 relative to that of the initial design, is obtained for the
optimized design in Fig. 5(e) with the prescribed external excitation frequency
Zp =500. The reason for this large decrease of 3 is that Zp =500 is quite close
to (slightly larger than) the first resonance frequency Z1=434. of the initial
design (as is also reflected by the high value of the sound power radiation 3 0
from the initial design at Zp =500 in Table 3). Thus, taking the optimized design
with Zp =500 in Fig. 5(e) to be given, we have computed its six lowest
eigenfrequencies and obtained the results shown in Table 2.
By comparison of results in Table 1 and Table 2, we see that the
optimization with the excitation frequency prescribed as Zp =500 has decreased
quite significantly the nearest (first) resonance frequency :1 from the value 434.
for the initial design to the value 354. for the optimized design. This implies that
large displacement amplitudes have been considerably reduced at the excitation
frequency Zp =500 by the discrete material optimization, and explains the
significant reduction of the sound radiation by 75.66%. At the same time, the
second resonance frequency :2 is slightly decreased by the optimization for the
case of Zp =500, see Table 1 and Table 2, but it is so much larger than the
excitation frequency Zp =500 that it has only marginally affected the sound
power radiation at Zp.
Hence, for the case of Zp =500, the mechanical cause of the substantial
reduction of the sound power radiation achieved by the optimization is that the
optimization has driven the nearest (first) resonance frequency as far away as
possible (i.e., downward) from the prescribed excitation frequency Zp.
Let us then consider the case of optimization for Zp =1000, where the
excitation frequency Zp is roughly located in the middle of the interval between
the first and second resonance frequencies :1=434. and :6=1591. of the initial
design, see Table 1. This is generally a favourable location of the excitation
frequency for a given design, and is also seen to lead to relatively low sound
radiation with a comparatively small difference between the values of the initial
and the optimized design (cf. Table 3). With Zp=1000, we find that the
optimization yields a design (shown in Fig. 5(f)), where the first resonance
frequency is decreased from 434. to 398. and the second resonance frequency is
increased from 1591. to 1756.
Hence, in this case with Zp =1000, the mechanical cause of the reduction of
the sound radiation obtained by the optimization (see Table 3) is that both of the
neighbouring resonance frequencies have been driven away from the excitation
frequency Zp, and thereby created an enlarged gap between these two resonance
frequencies with a reduced level of sound radiation at Zp and in its vicinity.
6.2 Multi-layer plate

A four-layer laminated plate with equal layer thicknesses and the same uniformly
distributed, time-harmonic pressure loading as in Sub-section 6.1 is considered
for Zp=100, Zp=500 and Zp=1000. These examples aim to demonstrate that the
approach can be also used for multi-layer laminated composite plate structures,
see Fig. 4. The geometry and boundary conditions of the plate are the same as in
Sub-section 6.1, and the glass/epoxy material with the same moduli of elasticity
as in Sub-section 6.1 is used. When a 20u20 mesh discretization is applied, the
total number of design variables for this four-layer plate is 6400. The layers are
numbered from bottom to top, i.e. layer 4 is the upper layer of the plate.
The optimized plates subject to Zp=100, 500, and 1000 are shown in Figs.
6(a), (b) and (c), but only by their upper layer (4) in order to save space. The
total power flow 3 from the optimized plates is reduced from 1,738. to 830.
for Zp=100, from 375,728. to 91,434. for Zp=500, and from 9,310. to 7,844. for
Zp=1000. Thus, for the given excitation frequencies Zp=100, 500 and 1000, the
same values of the objective function 3 are obtained by optimization of the
multi-layer laminated plates considered here and the single-layer plates in

Sub-section 6.1 with the same boundary conditions and structural dimensions, cf.
the results given in Table 3. For the two higher values of the excitation frequency,
Zp=500 and Zp=1000, the fiber orientations in the layers of each of the optimized
multi-layer plates are found to be the same as in the upper layers shown in Figs.
6(b) and 6(c), respectively, and these fiber orientations correspond precisely to
those in the optimized single-layer plates associated with the same excitation
frequencies, see Figs. 5(e) and 5(f). For the lower excitation frequency Zp=100,
some very few differences can be observed in the fiber orientations when
comparing the upper layer depicted in Fig. 6(a) with the other (non-shown)
layers of the optimized multi-layer plate and with the corresponding optimized
single-layer plate in Fig. 5(b). These very few differences are found in the
narrow central region of the plates where some weighting functions for
unidirectional fiber material oriented at ª¬90o , r45o , 0o º¼ appear to be very close
to each other. In agreement with the preceding discussion of the plate designs
obtained, the DMO convergence measures (see Section 4) for the optimized
multi-layer plate designs with Zp =100, 500 and 1000 are found to be h95 0.83 ,
h95 0.99 and h95 0.98 , where the value of the convergence measure
associated with Zp =100 is somewhat smaller than the very high values
associated with Zp =500 and 1000.
(a) Layer 4, Zp=100 (b) Layer 4, Zp=500 (c) Layer 4, Zp=1000

Figure 6. Multi-layer plate: Optimized design for different excitation frequencies
6.3 Laminated sandwich structure

Here and in the following Sub-section 6.4, we consider discrete material
optimization for the given excitation frequency Zp=100 and 1000 of a laminated
structure with the same size and boundary conditions as in the preceding
sub-section, but now consisting of eight layers made of glass/epoxy composite
and foam material that enable creation of a sandwich structure. The same
orthotropic glass/epoxy composite with permissible fiber angles ª¬90o , r45o , 0o º¼
is used as before, but in addition an isotropic polymeric foam material with
Young’s modulus E 125 MPa , Poisson’s ratio v 0.3 and mass density
J 100 kg m3 is assumed to be available for the structure.
In this and the next Sub-section 6.4, we choose the given global resource
constraint value R to represent the total mass of the structural material.
Accordingly, we define the unit cost factors u1 and u2 (see Section 2) of the
unidirectional fiber composite material and the isotropic foam material,
respectively, as the mass densities J i and J i of these two materials, and take
them to be c1 =1900 and c2 =100 in (8). The allowable total material mass
resource constraint value is taken to be R =10.0. In view of the data given, this
means that the foam must constitute at least 50% of the total volume.
In the current sub-section, we consider the case where the upper and the
lower layers are not allowed to choose the polymeric foam, while the inner 6
layers can locally consist of either the foam or glass/epoxy composite material.
This implies 38 design variables per element, distributed as [4, 5, 5, 5, 5, 5, 5, 4],
and brings the total number of design variables for the whole plate up to 15200,
when a 20u20 finite element mesh is used.
The results of the optimization for Zp=100 and 1000 are presented in Fig. 7
and Fig. 8, respectively, where layer elements with fibers indicate that the
glass/epoxy material is selected with the fiber orientations shown, and elements
in white indicate selection of isotropic foam material. The DMO convergence
measures for the designs with Zp =100 and 1000 are h95 0.93 and h95 0.98 ,
respectively. It is clearly seen from the figures that sandwich-like plates have
resulted from both cases of optimization, in particular for the design with
Zp=100 shown in Fig. 7, where almost all the available foam material is found in
the 4 innermost layers (core) of the 8-layer plate. In the design obtained for
Zp=1000 in Fig, 8, the available foam material is almost entirely placed in the 6
innermost layers, where it surrounds a short, approximately circular cylinder
consisting of the composite material. This will be discussed below. It is also
interesting to note that the upper and the lower layer of the optimized designs in
both of the two cases considered here, are very similar to the optimized designs
of the single-layer plates at the same excitation frequencies, as is seen by
comparing the designs of layers 1 and 8 in Fig. 7 and Fig. 8 with the design in
Fig. 5(b) forZp=100 and the design in Fig. 5(f) for Zp=1000, respectively.
Layer 1 Layer 2 Layer 3
Layer 7 Layer 8
Figure 7. Sandwich plate: Optimized design for the excitation frequencyZp=100 when no
foam material is allowed in the lower layer 1 and upper layer 8
Layer 7 Layer 8
Figure 8. Sandwich plate: Optimized design for the excitation frequency Zp=1000 when
no foam material is allowed in the lower layer 1 and upper layer 8
Table 4 Results for the initial and optimized designs of sandwich plates when no foam
material is allowed in the surface layers
1st 2nd
Sound Sound
resonance resonance
radiation for radiation for
freq. freq.
Zp=100 Zp=1000
: :
Initial design 449. 1656. 2059. 12808.
Design
optimized with 666. 2114. 967.
Zp=100, Fig. 7
Design
optimized with 369. 1984. 7969.
Zp=1000, Fig. 8
As stated in Table 4, we have found that, in comparison with the initial

design, the discrete material optimization has reduced the total power flow from
2059. to 967. for Zp=100 and from 12808. to 7969. for Zp=1000. Table 4 also
lists values of the first and second resonance frequencies :1 and :2 which we, in
a similar fashion as described in Sub-section 6.1, have computed for the initial
and the two optimized designs in Fig. 7 and Fig. 8. These results reveal, like in
Sub-section 6.1, that the mechanical explanation for the low values obtained for
the sound radiation is that the optimization has either driven the nearest
resonance frequency as far away as possible from the prescribed excitation
frequency Zp, or has increased the gap between two neighbouring resonance
frequencies as much as possible. Thus, for the design with Zp=100 in Fig. 7, the
first resonance frequency :1 has been increased quite considerably compared to
the first resonance frequency of the initial design, and for the design with
Zp=1000 in Fig. 8, the gap between the first and second resonance frequencies
has been increased by both decreasing the first resonance frequency and
increasing the second resonance frequency.
Let us finally discuss the quite remarkable difference between the overall
layouts of the inner layers of the optimized plates in Fig. 7 and Fig. 8. The plate
in Fig. 7 is optimized for an excitation frequency Zp =100 that is substantially
smaller than the first resonance frequency :1=666. of the plate, cf. Table 4. In
consistency with minimization of the total sound radiation, the plate may
therefore with good approximation be considered to have been subjected to
integral static stiffness maximization (i.e., compliance minimization) for a
uniformly distributed static loading (corresponding to Zp =0) that equals the
given amplitude P of the (otherwise) harmonic loading. Thus the design
should be mainly (bending) stiffness driven. This is confirmed by the design in
Fig. 7 where almost all the available stiffer composite material is found in the
outermost layers 1, 2 and 7, 8, while almost all the weaker foam material is
found in the inner layers 3-6.
Contrary to Fig. 7, the design in Fig. 8 with Zp =1000 has been significantly
driven by dynamics, essentially because inertia forces are proportional to the
square of the frequency of harmonic vibration of mass. Thus, as is seen, the
central part of the plate in Fig. 8 is filled-out by an approximately circular
through-the-thickness cylinder consisting of the stiffer composite material that
has a much higher mass density than the foam material. This local design of the
plate is very efficient in counteracting the time-harmonic external loading that
has not only a given frequency (Zp =1000) but also a prescribed amplitude, and
is found to be in anti-phase with the central part of the forced vibration mode,
whose shape is as depicted in Fig. 11(b). Clearly, the mass assembly in the
central part of the plate yields a large inertia that very effectively reduces the
displacement amplitudes over the central part of the plate, and, thereby, reduces
the vibration amplitudes and the density of sound power emission all over the
plate.
6.4 Laminated sandwich structure without restriction on the selection of

material in the surface layers
Here, we consider the same problem as in the preceding sub-section, but with the
exception that the upper and the lower layer are now allowed to choose freely
between the polymeric foam and the glass/epoxy composite material like the
inner 6 layers. This implies 40 design variables per element, distributed as [5, 5,
5, 5, 5, 5, 5, 5] and bringing the total number of design variables for the whole
plate up to 16 000, when a 20u20 finite element mesh is used.
The plate designs obtained by the discrete material optimization are
presented in Fig. 9 with Zp=100 and Fig. 10 with Zp=1000. The DMO
convergence measures for the designs with Zp =100 and Zp=1000 are
h95 0.93 and h95 0.98 , respectively, and the resonance frequency and sound
radiation characteristics determined for the initial design and the optimized
designs in the same way as described earlier, are listed in Table 5. Here it is seen,
by comparison with the results given in Table 4 for the optimized designs shown
in Fig. 7 and Fig. 8, that the total power flow has been slightly reduced from 967.
to 952. for the design in Fig. 9 with Zp=100, and, more significantly, from 7969.
to 6952. for the design in Fig. 10 with Zp=1000. Thus, for the same value of the
excitation frequency, the designs optimized without restriction on the selection of
material in the surface layers in the current sub-section are “more optimal” than
those obtained with the restriction in the preceding Sub-section 6.3. This is of
course to be expected since the restriction on the selection of material for the
surface layers in Sub-section 6.3 implies that the design space is smaller. The
mechanical causes of the reductions of the sound radiations from the designs in
Fig. 9 and Fig. 10 are the same as described for the designs in Figs. 7 and 8,
respectively, in Sub-section 6.3.
Layer 7 Layer 8
Figure 9. Sandwich plate: Optimized design for the excitation frequency Zp=100 and
without restriction for the surface layers
Layer 7 Layer 8
Figure 10. Sandwich plate: Optimized design for the excitation frequency Zp=1000 and
without restriction for the surface layers
By comparing the optimized designs in Fig. 9 and Fig. 10 with the

corresponding ones in Fig. 7 and Fig. 8, respectively, it appears that the removal
of the restriction against use of foam material in the surface layers has only given
rise to minor changes in the surface layers. This is particularly the case for the
design in Fig. 9 with Zp=100, where the surface layers 1 and 8 consist of
composite material except for very small regions with foam material in the
corners. The inner layers of this design also only exhibit very small changes. The
forced vibration mode of the optimized design in Fig. 9 excited at Zp=100 has
the same phase as the uniformly distributed dynamic loading, and is shown in
Fig. 11(a).
Table 5 Results for the initial and optimized designs of sandwich plates without
restriction for the surface layers
1st 2nd Sound Sound
resonance resonance radiation radiation
freq. freq. for for
: : Zp=100 Zp=1000
Initial design 432. 1594. 2713. 14152.
Design optimized with
623. 2029. 952.
Zp=100, Fig. 9
Design optimized with
281. 1963. 6952.
Zp=1000, Fig. 10
The surface layers of the design optimized with Zp=1000 in Fig. 10 also
predominantly consist of composite material, and the inner layers of the plate are
seen to exhibit only minor changes relative to the corresponding layers in Fig. 8.
However, it is noteworthy that in the surface layers of the plate in Fig. 10, a thin,
ring-like shaped region consisting of foam material is found between the large
central part of each of the surface layers and the edges of the plate. Moreover, a
close inspection of all the layers of the plate in Fig. 10 reveals that (with some
small, unimportant exceptions in the innermost layers 4 and 5), the entire plate is
equipped with an inner, through-the thickness zone of foam material that
emanates from the thin, ring-shaped regions with foam in the surface layers, and
follows these regions all the way around the large central part of the plate. A
mechanical explanation of this feature of the plate design in Fig. 10, which is
optimized for and excited by Zp=1000, may be given by considering the
corresponding forced vibration mode in Fig. 11(b), where the outer part of the
plate along the clamped edges is found to be in-phase, and the central part of the
plate to be in anti-phase with the uniformly distributed harmonic dynamic
loading. Guided by Fig. 11(b), we have found that within the thin, ring-like plate
region with foam material in the surface layers, there exists a closed curve along
which the two principal bending curvatures are zero and very small, respectively.
This means that from the point of view of optimization, the
through-the-thickness application of the isotropic, low-stiffness foam material is
“optimal” in this region of the plate, and that it saves composite material here for
other regions where high stiffnesses are useful.
(a)Zp=100 (b) Zp=1000

Figure 11. Forced vibration modes of the plate designs (a) optimized for and excited at
Zp=100 in Fig. 9, and (b) optimized for and excited at Zp=1000 in Fig. 10
7 Conclusions
Minimum sound emission from vibrating laminated composite plates without
damping is considered in this paper. The plates are subjected to uniformly
distributed time-harmonic pressure loading with prescribed frequency and
amplitude. Since the plate surfaces are flat, instead of solving the Helmholtz
equation, Rayleigh’s integral approximation is used for computing the total
sound power radiated from the plate into a light acoustic medium such as air.
This substantially reduces the computational cost of the structural-acoustic
analysis and design optimization. Optimization of fiber orientations, stacking
sequence and material selection is performed by Discrete Material Optimization
(DMO) for quadratic, clamped single-layer, multi-layer and sandwich plates.
Interesting features of the optimized designs are observed in numerical
examples.
Numerical results for single-layer and multi-layer plates show that the fiber
orientations of the layers in each of the optimized multi-layer plates are found to
be the same, and to correspond precisely to those in the optimized design of the
corresponding single-layer plate at the same excitation frequency. In the design
of laminated plates with polymeric foam and glass/epoxy composite material as
candidate materials, sandwich-like plates have been obtained for the excitation
frequencies considered. The influence of a restriction on the selection of
candidate materials for the surface layers on the optimum topologies is discussed.
For the same type of plate structure and excitation frequency, the design
optimized without restriction on candidate materials for the surface layers is
consistently better than the design optimized subject to the restriction.
The total sound power radiation from the vibrating laminated composite
plates is generally significantly decreased by the discrete material optimization.
To minimize the sound radiation, the optimization has either driven the nearest
resonance frequency as far away as possible from the prescribed excitation
frequency, or has increased the gap between two neighbouring resonance
frequencies as much as possible. For an excitation frequency smaller than the

first resonance frequency, the design is mainly (bending) stiffness driven, while
the design is driven by dynamics for higher values of the excitation frequency,
and this furnishes quite remarkable differences in the optimized designs.
There is still a need to further investigate weighting functions in the DMO
approach for the general case with a large number of candidate materials in order
to both fulfil the unity demand and penalize intermediate densities more
effectively. Furthermore, optimization with respect to a band of excitation
frequencies taking the structural damping into account may be considered in
future work.
References
beams and shafts. Optim Control Appl Methods 6: 191-200.
Bendsøe, M.P., Kikuchi, N., 1988. Generating optimal topologies in structural design
using a homogenization method. Computer Methods in Applied Mechanics and
Engineering 71: 197-224.
Applications. Springer-Verlag, Berlin.
Bendsøe, M.P., Olhoff, N., Sigmund, O., (Eds.), 2006. Proc. IUTAM Symp. Topological
Design Optimization of Structures, Machines and Materials—Status and Perspectives,
Copenhagen, Denmark, October 26–29, 2005. Springer, Berlin.
Bös, J., 2006. Numerical optimization of the thickness distribution of three-dimensional
structures with respect to their structural acoustic properties. Struct Optim 32: 12-30.
Chen, S.H., Song, X.W., Yang, Z.J., 2005. Modal optimal control of cabin noise of
vehicles. Smart Mater Struct 14: 257-264.
Christensen, S.T., Sorokin, S.V., Olhoff, N., 1998a. On analysis and optimization in
structural acoustics - Part I: Problem formulation and solution techniques. Structural
Optimization 16: 83-95.
Christensen, S.T., Sorokin, S.V., Olhoff, N., 1998b. On analysis and optimization in
structural acoustics - Part II: Exemplifications for axisymmetric structures. Structural
Optimization 16: 96-107.
Denli, H., Sun, J.Q., 2007. Structural-acoustic optimization of composite sandwich
structures: a review. The Shock and Vibration Digest 39 (3): 189-200.
Du, J., Olhoff, N., 2007a. Minimization of sound radiation from vibrating bi-material
Struct Multidisc Optim 34:91-110. See also Publisher’s Erratum in Struct Multidisc
Optim (2007) 34:545.
Du, J., Olhoff, N., 2010. Topological design of vibrating structures with respect to
Dühring, M.B., Jensen, J.S., Sigmund, O., 2008. Acoustic design by topology
optimization. J Sound Vib 317: 557-575.

Herrin, D.W., Martinus, F., Wu, T.W., Seybert, A.F., 2003. A new look at the high
frequency boundary element and Rayleigh integral approximations. Society of
Automotive Engineers, Warrendale, PA, USA.
Hufenbach, W., Kroll, L., Holste, C., Täger, O., Barkanov, E., 2001. Design of
dynamically loaded fiber-reinforced structures with account of their vibro-acoustic
behavior. Mech Composite Mater 37(2): 145-152.
Jensen, J.S., 2009. Space-time topology optimization for one-dimensional wave
propagation. Comput Methods Appl Mech Engrg 198: 705-715.
New York.
Lax, M., Feshbach, H., 1947. On the radiation problem at high frequencies. J Acoust Soc
Am 19: 682-690.
Lund, E., 1994. Finite Element Based Design Sensitivity Analysis and Optimization.
Ph.D. Thesis, Special Report No. 23, Dept. of Mechanical Engineering, Aalborg
University, Denmark.
Lund, E., 2009. Buckling topology optimization of laminated multi-material composite
shell structures. Composite Structures 91: 158-167.
Lund, E., Stegmann, J., 2005. On structural optimization of composite shell structures
using a discrete constitutive parametrization. Wind Energy 8: 109-124.
Luo, J.H., Gea, H.C., 2003. Optimal stiffener design for interior sound reduction using a
Mackerle, J., 2003. Topology and shape optimization of structures using FEM and BEM
– a bibliography (1999-2001). Finite elements in Analysis and design 39: 243-253.
Mroz, Z., Rozvany, G.I.N., 1975. Optimal design of structures with variable support
conditions. Journal of Optimization theory and Applications 15: 85-101.
Niu, B., Olhoff, N., Lund, E., Cheng, G., 2010. Discrete material optimization of
vibrating laminated composite plates for minimum sound radiation. Int J Solids Struct
47: 2097-2114.
Olhoff, N., Taylor, J.E., 1978. Designing continuous columns for minimum total cost of
material and interior supports. Journal of Structural Mechanics 6: 367-382.
maximum difference between adjacent natural frequencies. Int J Solids Struct 20:
63-75.
Olhoff, N., Du, J., 2013E. Topological design for minimum sound emission from
Sciences (CISM), Udine, Italy, June 18-22, 2012. Springer-Verlag, Vienna, 2013, 17
pp.
Pedersen, P., 1982. Design with several eigenvalue constraints by finite elements and
linear programming. J Struct Mech 10: 243-371.
Prager, W., 1977. Optimal layout of cantilever trusses. Journal of Optimization Theory
and Applications 23: 111-117.

Rozvany, G.I.N., Zhou, M., Birker, T., 1992. Generalized shape optimization without
homogenization. Structural Optimization 4: 250–252.
Sigmund, O., Torquato, S., 1997. Design of materials with extreme thermal expansion
using a three-phase topology optimization method. Journal of the Mechanics and
Physics of Solids 45: 1037-1067.
Sorokin, S.V., Olhoff, N., Ershova, O., 2006. The energy generation and transmission in
compound eleastic cylindrical shells with heavy internal fluid loading – from
parametric studies to optimization. Struct Multidisc Optim 32: 85-98.
Stegmann, J., 2004. Analysis and Optimization of Laminated Composite Shell Structures.
Ph.D. Thesis, Dept. of Mechanical Engineering, Aalborg University, Aalborg,
Denmark.
Stegmann, J., Lund, E., 2005. Discrete material optimization of general composite shell
structures. International Journal for Numerical Methods in Engineering 62:
2009-2027.
Svanberg, K., 1987. The method of moving asymptotes - a new method for structural
optimization. International Journal for Numerical Methods in Engineering 24: 359 -
373.
Taylor, J.E., 1975. Prediction of structural layout for maximum stiffness. Journal of
Optimization Theory and Applications 15: 145-155.
Thamburaj, P., Sun, J.Q., 2002. Optimization of anisotropic sandwich beams for higher
sound transmission loss. J Sound Vib 254(1): 23-36.
Tortorelli, D.A., Michaleris, P., 1994. Design sensitivity analysis: overview and review.
Wadbro, E., Berggren, M., 2006. Topology optimization of an acoustic horn. Comput
Yamamoto, T., Maruyama, S., Nishiwaki, S., Yoshimura, M., 2008. Thickness
optimization of a multilayered structure on the coupling surface between a structure
and an acoustic cavity. J Sound Vib 318: 109-130.
Yoon, G.H., Jensen, J.S., Sigmund, O., 2007. Topology optimization of acoustic-structure
interaction problems using a mixed finite element formulation. Int J Numer Methods
Eng 70: 1049-1075.
Topology Optimization
of Diffusive Transport Problems
*
Kurt Maute
*
Department of Aerospace Engineering Sciences,
University of Colorado Boulder, Boulder, CO, USA
Abstract. This lecture covers topology optimization methods for

solving diffusive transport problems, in particular heat conduction
in solids. The importance of understanding the underlying physical
phenomena to properly define the optimization problem is empha-
sized. Numerical issues arising from large gradients in the diffusion
coefficients along with remedies for mitigating theses problems will
be discussed.
1 Introduction
Besides convection and radiation, diffusive transport problems play an im-
portant role in many engineering applications. For example, the transport of
species, electrons, and thermal energy in solids is often modeled by Fickian
diffusion. Therefore, topology optimization of diffusive transport problems
has been studied frequently; see for example the work by Li et al. (1999,
2004); Gersborg-Hansen et al. (2006); Bruns (2007); Kim et al. (2009); Iga
et al. (2009); Zhuang et al. (2010). In comparison to topology optimiza-
tion in elasticity, diffusive transport problems are in general easier to treat
numerically as they only involve a scalar field (e.g. concentration, electric
potential, temperature). However, the physical phenomena, which arise as

DOI 10.1007/978-3-7091-1643-2_16, © CISM, Udine 2014
390 K. Maute
the design is changed, are often less intuitive and need careful considera-
tion. In addition, numerical issues particular to steady-state and transient
diffusive problems need to be addressed.
The common goal of topology optimization methods is to find the ge-
ometry and material layout of a body such that an objective is minimized
or maximized, subject to a set of equality and inequality constraints. In
general, the geometry and the material layout can be defined by the dis-
tribution of two or more material phases. In the simplest case of the body
being made of one homogeneous material, this definition leads to a two-
phase material layout problem where one phase is the bulk material and
the other phase represents void. The material phase at a point in a given
design domain can be defined via an integer-valued indicator function or by
either explicitly or implicitly describing the geometry of the material inter-
faces along with a binary variable that defines whether the point is inside
or outside a domain of a particular phase. The well-known density meth-
ods in topology optimization use a relaxed, continuous formulation of the
indicator function, leading to the concept of a continuously varying density
function of a real or fictitious material. The reader is referred to the excel-
lent book by Bendsøe and Sigmund (2003) for a comprehensive overview of
density methods. Level set methods describe implicitly the geometry of the
interfaces; the reader is referred to the book by Sethian (1999) and a recent
review article by van Dijk et al. (2013). The particular features of both, the
density and level set approaches, will be discussed in other chapters of this
book.
Here we will discuss topology optimization of diffusive transport prob-
lems following a density method. Further we will limit our discussion to a
two-phase, void-solid design problem. Thus, we will describe the geometry
of the body via density distribution and interpolate the material properties
as smooth continuous functions of the local density. Assuming an appropri-
ate parameterization of the density distribution and a suitable discretization
of the physical field (e.g. concentration, temperature, etc.) the objective
and constraints will be smooth functions of the optimization variables and
the resulting parameter optimization problem can be solved by standard
nonlinear programming techniques. For simple problems, such as problems
with just one linear constraint and monotonous objectives, optimality cri-
teria methods may be applicable as well.
2 Model Problem
To discuss the basic features of topology optimization of diffusive transport
problems, consider the following configuration, which can be thought of as
Topology Optimization of Diffusive Transport Problems 391
(1)
(2) 1
1
(12)
11
(2)
2
(12)
12
Figure 1. Model Problem.
either an initial design or an intermediate configuration in the course of the

optimization process.
The design domain Ω can be decomposed into non-overlapping subdo-
(k)
mains Ωi associated with the phases k = 1, 2. The interface between the
(1) (2) (12)
subdomains Ωi and Ωj is denoted by Γij . We define Ω(k) as the union
(k) (12)
of all subdomains Ωi and Γ(12) as the union of all interfaces Γij . The
external boundary is denoted by Γ.
The physical response of interest, denoted by T (xi ), in the design domain
Ω is governed by the following diffusion-reaction equation along with the
associated boundary, interface, and initial conditions:

dT ∂ ∂T
c − dij + aT − q = 0 in Ω (1)
dt ∂xj ∂xi
with
∂T
kij nj + bT − Q = 0 on ΓQ (2)
∂xi
T − T̂ = 0 on ΓT (3)
where c is a capacity coefficient, dij the diffusivity tensor, a a volumetric
production coefficient, and q a volumetric source term. The flux at the
surface ΓQ is denoted by Q with b being a surface convection coefficient.
The vector ni denotes the outward pointing normal. The field is prescribed
at ΓT with T̂ being the prescribed value.
For the sake of simplicity we assume that the diffusivity tensor is isotropic
and can be written as: dij = k δij where k is a diffusion coefficient and
392 K. Maute
δij denotes the Kronecker symbol. The material properties at a point xi

depend on the associated subdomain. In the case of a two-phase material,
the diffusion coefficient can be written as follows with I(xi ) being a binary
indicator function.

(1) (2) (1) 0 xi ∈ Ω(1)
k(xi ) = k + (k − k ) I(xi ) I(xi ) = (4)
1 xi ∈ Ω(2)
Note: In the above equation the indicator function is defined given a decom-
position of the design domain Ω into Ω(1) and Ω(2) . However, as the relation
is invertible one can use the indicator function to define the decomposition:
Ω(1) : xi ∈ Ω(1) |I(xi ) = 0 Ω(2) : xi ∈ Ω(2) |I(xi ) = 1 (5)
In addition, at the interface Γ(12) we enforce that the solution and the
fluxes are continuous:
T (1) − T (2) = 0 (6)
and
(1) (2)
(1) ∂T (12) (2) ∂T (21)
dij nj + dij nj =0 (7)
∂xi ∂xi
(kl)
where nj is the normal pointing from phase k to phase l. The physical
variable T (k) along Γ(12) is defined as follows:
(kl)
T (k) (xi ) = T (xi − nj ) →0 (8)
(12) (21)
Note: Using ni = ni one can write the interface flux condition (7)
(l)
for an isotropic material, i.e. dij = k (l) δij , as follows:
(12)
k (1) ∂T (2) /∂xi nj δij
= (9)
k (2) (12)
∂T (1) /∂xi nj δij
For a constant diffusivity k (2) and as k (1) → 0, the gradient of T (1) along
(12) (21)
nj also needs to decrease and/or the gradient of T (2) along nj needs
(1)
to increase. The physically correct solution in the extreme case of k = 0,
(21)
i.e. phase 1 is void, T (2) along nj vanishes, which represents an adiabatic
boundary condition.
For solving the diffusive transport problem defined above, it is convenient
to rewrite the governing equation in the weak form as follows, with RT being
the residual:

dT ∂T ∂T
RT = δT c dΩ + δ dij dΩ + δT (aT − q) dΩ
Ω dt Ω ∂xj ∂xi Ω

∂T
− δT kij nj dΓ
Γ ∂x
i (10)

Q−bT
(1) (2)

(1) ∂T (12) (2) ∂T (21)
− δT dij nj + dij nj dΓ = 0
Γ(12) ∂xi ∂xi

0
where we assume that T satisfies the boundary and interface conditions, i.e.
T − T̂ on ΓT and T (1) = T (2) on Γ(12) , and δT is an admissible test function,
i.e. it vanishes on ΓT .
Given the two-phase description discussed above, we are interested in
optimizing the material layout which corresponds to optimizing the geome-
try of phase (2) with phase (1) being the void phase. Regardless of choosing
the indicator function or the subdomain boundaries as primary variables,
the optimization problem reads:
minΓ(12) |I J(T )
s.t. Gi (T ) ≤ 0 i = 1, . . . , NG (11)
RT (T ) = 0
where J denotes the objective and Gi the inequality constraints.

To solve numerically the optimization problem (11) the interface Γ(12)
or the indicator function needs to be parametrized. The challenge when us-
ing the geometry of Γ(12) as the primary optimization variable is to choose
a parametrization that allows for topological changes, i.e. subdomains can
merge, vanish, and emerge. Parametrizing Γ(12) via surface interpolations,
as often used in CAD, does not allow for topological changes. Well-known
parametrization approaches that allow for topological changes are implicit
dynamic functions, such as level sets Osher and Fedkiw (2002). In the
following, we focus on topology optimization approaches that treat the in-
dicator function as the primary variable.
3 Density Methods
Treating directly the indicator function I(xi ) as the primary variable leads
to theoretical and practical issues. The indicator function may alternate
394 K. Maute
adiabatic
TL TR
adiabatic
Figure 2. Thin sheet subject to thermal gradient.
from point to point between zero and one. The associate functional space is
I(xi ) ∈ L∞ (Ω). The L∞ space contains all bounded but otherwise arbitrar-
ily varying functions. For such functions, the optimization problem (11) is
ill-posed and does not converge as the discretization of I(xi ) and the physi-
cal field T (xi ) is refined. This manifests itself in a strong dependency of the
optimization results on the discretization of the design domain. From an
engineering perspective, one may argue that such a dependency is accept-
able as one is just interested in the solution of a particular discretization.
However, due the ill-posedness of the problem there is little meaning to this
solution, except that it may represent an improvement over some other de-
sign. Furthermore, since often the size of the geometric features decrease as
the mesh is refined, the mesh resolution is likely to be inadequate to pre-
dict the physical solution with sufficient accuracy. Finally, from a practical
point of view, directly parametrizing the indicator function and using these
parameters as optimization variables leads to significant numerical costs as
it requires solving potentially large integer optimization problems. While
the efficiency of integer programming methods, such as branch-and-bound
methods, has increased, solving large integer optimization problems is still
computationally challenging and in most cases impractical.
Resolving the ill-posedness for diffusion problems follows both, from
a mathematical and physical perspective, the same ideas as in elasticity
Bendsøe and Sigmund (2003). Therefore we will not further dive into this
topic, but rather outline the basics. From an engineering view one can ex-
plain the ill-posedness of the topology optimization problem with the role
of the indicator function in the governing equations. Consider the following
heat conduction example depicted in Figure 2.
At the left edge ΓT L of a thin sheet, the temperature is prescribed to
T̂L and at the right edge ΓT R to T̂R . Adiabatic boundary conditions are
imposed on the upper and lower edges. For the sake of simplicity we ignore
volumetric fluxes and surface convection. The objective is to maximize the
layout of unit cell
periodic arrangement of unit cells
Figure 3. Optimum micro-structures for 2-phase materials with maximum

diffusivity.
total heat flux across the sheet by varying the distribution of the material
phases (1) and (2) in the design domain with the heat conductivities k (1) <
k (2) . The design domain is the entire sheet. The objective can be determined
as follows:
∂T ∂T
J= dij nj dΩ = dij nj dΩ (12)
ΩL ∂xi ΩR ∂xi
Without additional constraints, the solution of the problem is trivial: the
design domain is occupied by phase (2) as it has the higher conductivity.
When limiting the volume fraction of phase (2), f2 = |Ω2 |/|Ω|, the problem
has a non-trivial solution.
Intuitively, the heat flux at a point is maximum if the constitutive tensor,
di j, is optimum for a given temperature gradient and local volume fraction.
The heat flux across the sheet can then be maximized by finding the op-
timum distribution of local volume fraction and the resulting temperature
gradient field. In general, the local problem is finding the optimum diffu-
sivity tensor over all tensors of heterogeneous materials that are composed
of the phases (1) and (2) with a given volume fraction, f2 . For diffusive
problems, this optimization problem has attracted a significant amount of
research. One can show that the optimum diffusivity tensor can be derived
via homogenization from an optimum micro-structure that consists only of
the (1) and (2) and satisfies the volume fraction constraint. Such a micro-
structure is shown in Fig. 3. Note the orientation of the micro-structure
with respect to a macroscopic reference frame is part of the solution of the
local problem. Interestingly, optimum micro-structures of diffusive prob-
lems have a minimum interface area.
The ill-posedness of the original formulation of the topology optimiza-
tion problem is rooted in the dilemma that the indicator function I, defined
at the macro-scale, can never converge to the optimum continuum solu-
396 K. Maute
tion, which is made of locally optimum micro-structures. Indeed, as the

discretization of the design domain is refined, the optimum material dis-
tribution I(xi )∗ attempts to locally mimic the properties of the optimum
diffusivity tensor. However, since the material distribution yielding the opti-
mum diffusivity is defined on a lower length scale, the macroscopic problem
does converge, but features increasingly rapid spatial oscillations.
Note this fundamental issue is not due to the fact that the indicator
function is integer-valued. The same issue exists if one treats, for example,
the local volume fraction as the primary variable. The volume fraction
of phase 2, denoted by f2 , is defined via the equivalence of the following
integrals over an arbitrary control volume Ωc :

IdΩc = f2 dΩc (13)
For two-phase, solid-void problems this approach is typically referred to as

density method and requires some form of material interpolation, defining
a relationship between the physical property and the volume fraction. For
example, for steady-state diffusion problems the diffusivity tensor needs to
be defined as a function of the density. For explicit interpolations, such as
polynomial interpolation, using densities may mitigate some symptoms of
the ill-posedness, but does not resolve it. Density approaches only overcome
the ill-posedness if the local optimization problem is also solved, i.e. the
interpolation function is designed such that it interpolates the optimum
diffusivity tensor as a function of density. However, whether or not the ill-
posedness is resolved, in general the optimum density distribution does not
converge to a “0-1” distribution, i.e. at a point there is either phase (1) or
phase (2), and it may contain a significant amount of intermediate volume
fractions, i.e. 0 < f2 < 1.
To overcome both the issue of ill-posedness and the lack of convergence
to “0-1” solutions, a two-tier approach is typically used in density methods.
To suppress increasingly rapid spatial oscillations in the material distribu-
tion as the mesh is refined, the density distribution is restricted to smooth
functions. The smoothness is controlled via a mesh-independent length scale
parameter, similar to approaches for dealing with localization effects in non-
linear mechanics Sigmund and Maute (2012). Popular smoothing techniques
include geometric and PDE-based filtering of the density distribution and
enforcing smoothness via design constraints or penalty formulations of the
optimization problem; these techniques are reviewed in Sigmund and Pe-
tersson (1998) and Wang et al. (2011). To promote convergence to a “0-1”
solution, intermediate densities are penalized, either implicitly through the
material interpolation or explicitly through a penalty formulation. Either
adiabatic
(2) (1) (2)

TL 1 1 2
adiabatic
Figure 4. Configuration to study the influence of boundary conditions and

heat conductivity ratios on temperature distribution.
approach will, in most cases, lead to a non-convex optimization problem.

This non-convexity causes the optimization results to depend on the initial
design, the search directions, and step size used in the optimization algo-
rithm. To mitigate these phenomena, continuation methods may be used
Bendsøe and Sigmund (2003). Most density methods operate on fixed or
adaptively refined meshes, as used for example by Maute and Ramm (1995)
and Bruggi and Verani (2011). In both cases, elements are typically not
eliminated from the computational domain if they are entirely in the void
domain. This requires that the material properties representing void need
to be chosen such that the discretized weak form of the governing equations
is solvable. Thus, in the case of steady-state diffusion, the diffusivity needs
to remain finite in the void phase.
4 Physical Phenomena
In general, formulating optimization problems whose solution provide guid-
ance to engineers in the design process, requires deep insight into the physi-
cal phenomena that may dominate the response of both the converged design
as well as intermediate configurations during the design process. The latter
is of great importance for density based topology optimization, in particu-
lar as intermediate designs may involve non-conventional material layouts
and, depending on the material interpolation used, fictitious material prop-
erties. As the following simple study will show this may impact the proper
formulation of objective and constraints and may require the inclusion of
real or fictitious phenomena that otherwise would play no role in the design
problem.
Here we consider a thin sheet subdivided into three subdomains as shown
in Figure 4. The temperature being prescribed at the left edge ΓT L to T̂L
and adiabatic boundary conditions are imposed on the upper and lower
398 K. Maute
edges. We consider the problem with three different boundary conditions:

(a) The temperature is prescribed at the right edge ΓT R to T̂R .
(b) A constant heat flux QR is applied at the right edge ΓQR .
(c) Heat is convected at the upper surface of the sheet.
Assuming that the temperature gradient in thickness direction is negligible,
we can use a two-dimensional model of the governing equations, converting
the surface convection in to a volumetric one. Further, considering steady-
state conditions, the weak form of the governing equations is:

∂T ∂T
RT = δ dij dΩ + δT ã (T − Tinf ty ) dΩ − δT QdΓ = 0
Ω ∂xj ∂xi Ω Γ
(14)
where ã is an effective convection coefficient accounting for the surface-to-
volume ratio: ã = dΓupper /dΩ a. Again, we assume an isotropic material
with dij = k δij .
We study the above configurations for three different diffusivity ratios
k (1) /k (2) = [1, 0.001, 0]. The first ratio represents a homogenous sheet with
a diffusivity k (1) , the second ratio a sheet with a low conductive inclusion,
which would typically be considered “void” in topology optimization, and
the third ratio a sheet with a hole at the center. We further assume that the
convection coefficient does not depend on the diffusivity. The qualitative
temperature distributions at steady-state are depicted in Figure 5. If the
temperature is prescribed at both vertical edges and the conductivity of the
center section is reduced, the temperature fields in phase (2) becomes more
flat. In the extreme case, k (1) /k (2) = 0, the temperature is constant in the
outer sections. Thus, the total heat flux through the sheet is reduced as the
conductivity of the center section decreases. In the case of a constant flux
at the right edge, again the temperature distribution in the outer sections
becomes more flat relative to the distribution in the center section as the
conductivity in phase (1) is reduced. However, to maintain a constant heat
flux the maximum temperature increases. Furthermore, the steady-state
problem is ill-posed for k (1) = 0. In the case of a convective heat flux, an
overall similar behavior can be observed. For T̂L < T∞ the temperature
at the right edge approaches the ambient temperature and the total heat
flux at the left edge decreases as the conductivity of the center section is
reduced. In the extreme case, k (1) /k (2) = 0, both the center and outer right
section are at ambient temperature.
In properly formulating optimization problems, the main lesson from the
study above is that in general the temperature at a point is not a suitable
design criterion. Only in the case where a convective heat flux is applied at a
point, its temperature takes on a physical value even if the local conductivity
approaches zero. The convection boundary condition provides a simple but
k (1) / k (2) 1 k (1) / k (2) 10 3

k (1) / k (2) 0
TR TR TR
TL TL TL
(2) (1) (2) (2) (1) (2) (2) (1) (2)
1 1 2 1 1 2 1 1 2
(a) prescribed temperature at right edge
k (1) / k (2) 1 k (1) / k (2) 10 3

k (1) / k (2) 0
TR
ill-posed
TL TL TL
(2) (1) (2) (2) (1) (2) (2) (1) (2)
1 1 2 1 1 2 1 1 2
(b) prescribed flux at right edge
k (1) / k (2) 1 k (1) / k (2) 10 3

k (1) / k (2) 0
TR TR TR
TL TL TL
(2) (1) (2) (2) (1) (2) (2) (1) (2)
1 1 2 1 1 2 1 1 2
(c) convective heat flux at upper surface

Figure 5. Temperature distributions for different boundary conditions and
diffusivity of the center section.
400 K. Maute
40 m
20
m
qA
Figure 6. Convective cooling of rectangular plate.
physical mechanism to push the temperature field toward a unique solution,

i.e. the ambient temperature. In contrast to using the local temperature,
optimization criteria that involve the heat flux at point or through a surface
are typically well suited.
5 Example
As mentioned above, considering convective boundary conditions leads to
interesting topology optimization problems. Since density methods require
interpolating physical properties as a function of the density, one can con-
sider different interpolation schemes for the convection coefficient. To il-
lustrate the profound effect of the interpolation approach on the topol-
ogy optimization result, we revisit the problem of Bruns (2007), depicted
in Figure 6. A two-dimensional design domain is subject to a heat flux
qA = 1 pW/μm at point A. The plate is cooled via natural convection; the
convection coefficient of the bulk material and the surrounding media is
b0 = 10−3 W/μm2 K; the ambient temperature is set to T∞ = 0 K. Assum-
ing an isotropic material behavior, the conduction coefficient of the bulk
material is k0 = 1 pW/μmK. The design problem is to find the optimum
material layout that minimizes the temperature at point A; the volume of
the bulk material may not exceed 30 % of the volume of the design space.
Here we apply a standard SIMP model to interpolate the heat conduction
coefficient:
k = k0 f β (15)
with β = 3 where f denotes the volume fraction, or density. The nonlin-
ear interpolation of the conduction coefficient along with volume constraint
leads to an implicit penalization of intermediate densities. We consider two
A.1 h 2.0 P m B.1
A.2 h 0.5P m B.2
A.3 h 0.1P m B.3
Figure 7. Material layouts for different mesh refinement levels and convec-
tion interpolations.
options for interpolating the convection coefficient: in option (A) we keep

the convection coefficient constant, and in option (B) we apply a SIMP
interpolation to the convection coefficient: b = b0 f β with β = 3.
We solve the optimization problem for three levels of mesh refinement.
The density distribution is discretized by the same mesh as used for the
thermal analysis. The optimization variables are the nodal densities. The
conduction and convection coefficients are constant within one element and,
if design dependent, computed via a SIMP interpolation using the aver-
age of the nodal densities. As we operate on fixed meshes, the conduc-
tion in the void face needs to remain finite. Therefore, we set the min-
imum density to fmin = 10−3 which leads to a minimum conduction of
kmin = 10−9 pW/μmK. The gradients of the objective function with re-
spect to the optimization variables are computed by an adjoint approach.
The optimization problem is solved via the globally convergence method of
moving asymptotes (GCMMA) of Svanberg (2002).
The results of the optimization problem for options (A) and (B) and dif-
ferent mesh refinement levels are shown in Figure 7. The size of elements is
denoted by h = 2.0/0.5/0.1 μm. The difference between the results for the
different convection interpolation approaches is striking. When reducing
the connectivity with the density the designs converge to a simple, circular
shape; the conductive material is as close to the heat source at point A as
402 K. Maute
A.3 h 0.1P m B.3
8K
0K
Figure 8. Temperature contours for optimized designs.
Case A.1 A.2 A.3 B.1 B.2 B.3

Temperature [K] 3.48 3.69 3.98 5.82 5.77 6.23
Table 1. Temperature at point A for different mesh refinement levels and
convection models.
possible. Keeping the convection coefficient constant, complex heat con-

duction paths are generated, distributing the heat across the entire design
domain. The temperature contours of the designs obtained for the finest
mesh are shown in Figure 8, again illustrating the significant differences due
to different convection models.
The temperatures at point A for the different cases studied above are
summarized in Table 1. Interestingly, as the mesh is refined and the number
of optimization variables is increased, the objective increases. This some-
what unexpected behavior is due to increased accuracy of the finite element
predictions; the temperature is typically underestimated on coarse meshes.
One could argue that assuming a constant convection is not correctly
modeling the physics as the finite conduction causes heat to artificially dis-
sipating into the void phase which is then convected. To understand the
impact of this assumption on the performance of the optimized designs,
we compute the temperature at point A of design (A.3) using the design
dependent convection coefficient and of design (B.3) for a constant convec-
tion. The predicted temperatures are TA = 8.10 K for design (A.3) and
TA = 6.11 K for design (B.3). In both cases, the temperatures are above
the values obtained for the designs that were optimized for the particular
convection model. In particular, the poor performance of design (A.3) for
design dependent convection clearly shows the sensitivity of the optimiza-
tion results with respect to the assumption of the convection model.
It is also interesting that when using a constant convection model the
material layout does not converge as the mesh is refined. This indicates that
the problem is ill-posed and regularization techniques need to be applied to
h 0.5P m h 0.1P m
Figure 9. Influence of perimeter constraint on optimized material layout.
mitigate the mesh dependency. To illustrate the effect of such techniques,

here we use a global regularization approach and constrain the norm of the
density gradient integrated over the design domain:

∂f
P =

dΩ (16)

Ω ∂xi
As the density distribution converges to a “0-1” solution the integral

(16) approximates the perimeter of the body. Constraining the perimeter
is a global regularization technique as it controls the density distribution
in an integral sense rather than locally. Sensitivity filters, for example, lo-
cally control the density distribution. Figure 9 shows the optimized density
distribution for the two finer levels of mesh refinement, augmenting the op-
timization problem with an additional perimeter constraint: P ≤ Pmax with
Pmax = 160. The constraint value was chosen to be approximately the
perimeter of the material layout (A.1) in Figure 7.
The temperatures at point A for the mesh with h = 0.5 μm is TA =
3.38 K and for the mesh with h = 0.1 μm is TA = 4.43 K. For both mesh
refinement levels the performance is decreased in comparison to the results
without perimeter constraint (see Table 1). In addition, while the perimeter
constraint prevents the formation of ever smaller geometric features as the
mesh is refined, the volume of material with intermediate densities, i.e.
0 < f < 1, increases. This issue can be mitigated by increasing the SIMP
penalty factor β or by using projection schemes. The latter approach is
discussed in detail by Guest et al. (2004) and Wang et al. (2011).
6 Numerical Issues
Despite the simplicity of the partial differential equations describing diffu-
sive transport problems, their numerical treatment is not trivial. In the
transient case, instability phenomena may occur, independent of whether
implicit or explicit time integrations schemes are used, if the time step and
404 K. Maute
spatial discretization are not appropriately chosen. This type of instability

will not be discussed further in the following. Instead we will discuss an-
other instability problem which occurs in density-based topology optimiza-
tion methods of steady-state heat transfer with surface convection. Using
a constant convection coefficient, i.e. it does not vary with density, the
temperature distribution may rapidly oscillate as the local conductivity is
lower than a threshold.
This effect can be understood by examination of the properties of a
simple two-node, one-dimensional heat transfer element of length Le and
cross-section A. The thermal conduction matrix for this element is:

k A 1 −1
K= (17)
L −1 1
A spectral decomposition of this matrix yields the following eigenvalues λki

and vectors eki :
1 1
ek1 = √ , λk1 = 0 (18)
2 1

1 1 kA
ek2 = √ , λk2 = 2 (19)
2 −1 L
The thermal convection matrix for this element is:

bS 2 1
C= (20)
6 1 2
The eigenvalues λci and eigenvectors eci of this matrix are:

1 1 bS
ec1 = √ , λc1 = (21)
2 1 2

1 1 bS
ec2 = √ , λc2 = (22)
2 −1 6
Both the thermal conduction and convection matrices have the same eigen-
vectors; however the lower modes are reversed. When convection dominates
over conduction, the antisymmetric mode of heat transfer dominates and
the resulting temperature fields exhibit an oscillatory behavior above and
below the ambient temperature. We revisit case (c) of the example in Fig-
ure 2 but use a one-dimensional bar model. For the sake of simplicity, we
assume that k A = 1, b S = 1, and the length of the bar L = 1. The tem-
perature at the right end for different levels of mesh refinements are given
in Table 6.
elements Temperature
2 1.035 101
3 −1.22 100
4 5.97 10−2
5 −5.55 10−4
6 3.60 10−8
7 2.52 10−9
8 2.28 10−7
9 1.14 10−6
Table 2. Temperature versus mesh discretization for a simple thermal bar
subject to convection.
The oscillation of the temperature above and below the ambient value
is evident for low mesh densities. Because the convection influence is pro-
portional to the convecting area while the conduction influence is inversely
proportional to the element size, mesh refinement will mitigate the oscil-
lations. This can be seen by examining the eigenvalues of the conduction
matrix. For the symmetric mode to remain the lower eigenvalue, the fol-
lowing condition needs to be satisfied:
λk1 + λc1 < λk2 + λc2 (23)
which is equivalent to
b S L2
≤1 (24)
6kA
This stability condition shows that oscillations do not occur if the mesh is
sufficiently refined.
Following Bruns (2007), the stability issue for convection dominated
problems can be circumvented if the convection matrix is lumped and ap-
proximated by a diagonal matrix:

bS 1 0
C̃ = . (25)
2 0 1
As the eigenvectors of the lumped convection matrix are:

1 bS
ẽc1 = , λ̃c1 = , (26)
0 2

c 0 bS
ẽ2 = , λ̃c2 = , (27)
1 2
406 K. Maute
the antisymmetric mode is not dominant, independent of the conduction

and convection coefficients.
A final option is to link the convection coefficient to the optimization
variable, such that condition (23) is satisfied.
7 Acknowledgment
The author acknowledges the support of the National Science Foundation
under grants EFRI–1038305 and EFRI–1240374. The opinions and conclu-
sions presented in this paper are those of the authors and do not necessarily
reflect the views of the sponsoring organization. The author would also like
to thank Mr. Peter Coffin for his help in preparing the numerical results.
Bibliography
M. P. Bendsøe and O. Sigmund. Topology Optimization: Theory, Methods
and Applications. Springer, 2003.
M. Bruggi and M. Verani. A fully adaptive topology optimization algorithm
with goal-oriented error control. Computers & Structures, 89(15–16):
1481 – 1493, 2011.
T. Bruns. Topology optimization of convection-dominated, steady-state
heat transfer problems. International Journal of Heat and Mass Transfer,
50(15-16):2859 – 2873, 2007.
A. Gersborg-Hansen, M. P. Bendsøe, and O. Sigmund. Topology optimiza-
tion of heat conduction problems using the finite volume method. Struc-
tural and multidisciplinary optimization, 31(4):251–259, 2006.
J. Guest, J. Prévost, and T. Belytschko. Achieving minimum length scale
in topology optimization using nodal design variables and projection
functions. International Journal for Numerical Methods in Engineering,
61(2):238–254, 2004.
A. Iga, S. Nishiwaki, K. Izui, and M. Yoshimura. Topology optimization for
thermal conductors considering design-dependent effects, including heat
conduction and convection. International Journal of Heat and Mass
Transfer, 52(11-12):2721–2732, 2009.
M. Kim, S. Ha, and S. Cho. Level set-based topological shape optimiza-
tion of nonlinear heat conduction problems using topological derivatives.
Mechanics based design of structures and machines, 37(4):550–582, 2009.
Q. Li, G. P. Steven, O. M. Querin, and Y. Xie. Shape and topology design for
heat conduction by evolutionary structural optimization. International
Journal of Heat and Mass Transfer, 42(17):3361 – 3371, 1999.
Q. Li, G. P. Steven, Y. Xie, and O. M. Querin. Evolutionary topology

optimization for temperature reduction of heat conducting fields. Inter-
national Journal of Heat and Mass Transfer, 47(23):5071–5083, 2004.
K. Maute and E. Ramm. Adaptive topology optimization. Structural and
Multidisciplinary Optimization, 10(2):100–112, 1995.
S. Osher and R. Fedkiw. Level set methods and dynamic implicit surfaces,
volume 153. Springer, 2002.
J. Sethian. Level set methods and fast marching methods: Evolving inter-
faces in computational geometry, fluid mechanics, computer vision, and
materials science. Cambridge University Press, Cambridge, 1999.
O. Sigmund and K. Maute. Sensitivity filtering from a continuum mechanics
perspective. Structural and Multidisciplinary Optimization, pages 1–5,
2012.
O. Sigmund and J. Petersson. Numerical instabilities in topology opti-
dependencies and local minima. Structural and Multidisciplinary Opti-
mization, 16(1):68–75, August 1998.
K. Svanberg. A class of globally convergent optimization methods based on
conservative convex separable approximations. SIAM J. on Optimiza-
tion, 12(2):555–573, 2002. ISSN 1052-6234.
N. van Dijk, K. Maute, M. Langelaar, and F. van Keulen. Level-set methods
for structural topology optimization: A review. Structural and Multidis-
ciplinary Optimization, accepted for publication, 2013.
F. Wang, B. S. Lazarov, and O. Sigmund. On projection methods, conver-
gence and robust formulations in topology optimization. Structural and
C. Zhuang, Z. Xiong, and H. Ding. Topology optimization of multi-material
for the heat conduction problem based on the level set method. Engi-
neering Optimization, 42(9):811–831, 2010.
Topology Optimization of Flows:
Stokes and Navier-Stokes Models
*
Kurt Maute
*
University of Colorado, Boulder, USA
Abstract. This lecture will discuss density-based approaches for

solving flow topology optimization problems. The focus is on low-
Reynolds number fluid models, namely Stokes and laminar Navier-
Stokes models, at steady-state conditions.
1 Introduction
Finding the geometry of systems to optimize the performance of internal
and external flows is of great importance across a wide spectrum of applica-
tions. For example, fluids apply forces on bodies, such as lift and drag, and
transport species and thermal energy. Manipulating these phenomena is a
central issue for a large number of engineering systems, including aircraft
aerodynamics, injection molding, and liquid cooling. Due to the complexity
of modeling and predicting flows, historically the design of flow problems
was driven by experimental studies. With the advent of computational fluid
dynamics (CFD) in the late 1960s, the flow about more complex geometries
could be simulated and analyzed numerically. Improved numerical schemes

DOI 10.1007/978-3-7091-1643-2_17, © CISM, Udine 2014
410 K. Maute
and more powerful computing platforms enabled the systematic optimiza-

tion of flow problems via mathematical optimization methods. Until re-
cently, research on optimizing the geometry of flow problems solely focused
on shape optimization, accounting for ever complex flow phenomena (see,
for example, Gunzburger (2003) and Mohammadi and Pironneau (2001)).
Fluid topology optimization was pioneered by Borrvall and Petersson
(2003) adopting the concept of density methods, originally developed for
problems in solid mechanics, to Stokes flows. For a general introduction
to topology optimization, the reader is referred to the book by Bendsøe
and Sigmund (2003). Guest and Prévost (2006) conducted fluid topology
optimization using a Darcy-Stokes flow model. Duan et al. (2008b) and
Challis and Guest (2009) used a level-set parametrization of the material
distribution to solve Stokes flow problems. The work on Stokes models was
extended to Navier-Stokes models, for example by Gersborg-Hansen et al.
(2005); Okkels et al. (2005); Olesen et al. (2006); Evgrafov et al. (2007).
Kreissl and Maute (2011) recently presented an approach to optimize the
transient response of flows predicted by the incompressible Navier-Stokes
equations. Othmer (2006); Othmer et al. (2007) optimized the layout of 3D
air-duct manifolds for automotive applications, employing an incompressible
Navier-Stokes model. As an alternative to the Navier-Stokes flow model,
Pingen et al. (2007a, 2010) used the lattice Boltzmann method (LBM) for
solving fluid topology optimization problems. Kreissl et al. (2011) employed
the LBM in combination with a level-set based geometric interface repre-
sentation for generalized shape optimization of fluids.
In this lecture we will discuss the basic formulations and numerical tech-
niques for density-based topology optimization methods, focusing on steady-
state flows at low Reynolds number. We will study optimization problems
that, in some form or another, aim at reducing the energy dissipation in
the design domain. For example, reducing the total pressure drop between
the inlet and outlet of a channel belongs to this class of problems, as does
reducing the drag of an immersed body.
2 Flow Modeling
Assuming that compressibility effects can be neglected, the incompressible
Navier-Stokes equations are valid across a broad range of flow regimes. For
Topology Optimization of Flows: Stokes and Navier-Stokes Models 411
steady-state conditions the governing equations are:

∂v̂i ∂ σ̂ij
Momentum equation: ρ̂ v̂j = + fîB in Ω, (1)
∂ x̂j ∂ x̂j
∂v̂i
Incompressibility condition: = 0 in Ω, (2)
∂ x̂i
Surface traction condition: σ̂ij nj = F̂i on Γσ , (3)
Dirichlet condition: v̂i = v̄î on Γv and p̂ = p̄ˆ on Γp , (4)
where ρ̂ and vî describe the density and velocity, respectively. The sub-
scripts i, j define the spatial directions, ˆ indicates dimensional quantities
ˆ
and a prescribed value. The external body forces are denoted by fîB and
¯
the external traction forces by F̂i . The stress tensor σ̂ij is defined as:

∂v̂i ∂v̂j
σ̂ij = −p̂δij + μ̂ + , (5)
∂ x̂j ∂ x̂i
where p̂ and μ̂ describe the pressure and dynamic viscosity, respectively. In
the following, we assume that the dynamic viscosity is constant, i.e. it does
not depend of the flow variables.
Alternatively, the incompressible Navier-Stokes equations (1)-(2) at steady
state can be written in weak form with non-dimensional variables as follows:

∂vi 1 ∂wi ∂wj
R= wi ρ vj dΩ − wi fi dΩ −
B
+ pδij dΩ
Ω ∂xj Ω Ω 2 ∂xj ∂xi

1 ∂wi ∂wj ∂vi ∂vj ∂vi
+ + μ + dΩ + q dΩ
Ω 2 ∂xj ∂xi ∂xj ∂xi Ω ∂xi

∂vi ∂vj
− wi nj −pδij + μ + dΓ = 0, (6)
Γ ∂xj ∂xi
where R is a residual, wi an admissible velocity test function and q an
admissible pressure test function. Note, the divergence of the stress tensor,
∂ σ̂ij /∂ x̂j in Eq. (1), has been integrated by parts. The non-dimensional
dynamic viscosity is defined as:
μ̂ 1
μ= = , (7)
L̂ref v̂ ρ̂ Re
where L̂ref , v̂, ρ̂ are the dimensional reference length, reference velocity,
reference density, and Re is the Reynolds number.
For small Reynolds numbers (Re << 1) the advective and inertial forces
are small compared with viscous forces and can be neglected. In such flow
412 K. Maute
regimes the fluid velocities are small, the viscosity is large, or the length-
scales of the flow are small. In this case the strong form of the flow equations
is:
∂ σ̂ij
Momentum equation: + fîB = 0, (8)
∂ x̂j
∂v̂i
Incompressibility condition: = 0, (9)
∂ x̂i
Surface traction condition: σ̂ij nj = F̂i on Γσ , (10)
Dirichlet condition: v̂i = v̄î on Γv and p̂ = p̄ˆ on Γp . (11)
The above equations are the Stokes flow model. Their weak form using
non-dimensional variables is given by:

1 ∂wi ∂wj 1 ∂vi ∂vj
R=+ + + dΩ
Ω 2 ∂xj ∂xi Re ∂xj ∂xi

∂vi 1 ∂wi ∂wj
+ q dΩ − wi fi dΩ −
B
+ pδij dΩ
Ω ∂xi Ω Ω 2 ∂xj ∂xi

1 ∂vi ∂vj
− wi nj −pδij + + dΓ = 0, (12)
Γ μ ∂x j ∂xi
From a mathematical perspective, neglecting the advection term in the
Navier-Stokes equation simplifies the problem significantly for three reasons:
(a) the Stokes equation are linear, (b) there are no numerical instability is-
sues caused by the advection term, and (c) the resolution requirements of
the flow field, i.e. mesh density, are reduced.
3 Density Methods
As discussed for example in the chapter on topology optimization of diffusive
transport problems, one can manipulate the geometry by directly varying
the shape of the interface between fluid and solid sub-domains. In the con-
text of flow optimization, the solid can be considered the “void phase.”
Such approaches are typically restricted to shape variations and topological
changes are not possible unless special interface descriptions are used, e.g.
level set methods. Level set methods for flow topology optimization were
presented by Cunha (2004); Pingen et al. (2007b); Mohammadi and Piron-
neau (2008); Duan et al. (2008a,b,c); Challis and Guest (2009); Kreissl and
Maute (2012).
The most common approach in flow topology optimization follows a den-
sity concept. The integer-valued indicator function, which defines whether
there is fluid at a point xi in the design domain, is relaxed by introduc-

ing the volume fraction, or density, as the design variable. To model flow
through porous media with the Stokes equations the momentum equations
are typically modified either using:

∂ p̂ ∂ ∂v̂i ∂v̂j
Darcy’s law (1): κ̂ = μ̂ + = 0, (13)
∂ x̂i ∂ x̂j ∂ x̂j ∂ x̂i
or
∂ p̂ ∂ μ̂ ∂v̂i ∂v̂j
Darcy’s law (2): = + =0 (14)
∂ x̂i ∂ x̂j γ̂ ∂ x̂j ∂ x̂i
or

∂ p̂ ∂ ∂v̂i ∂v̂j
Brinkman penalization: = μ̂ + + α̂v̂i = 0 (15)
∂ x̂i ∂ x̂j ∂ x̂j ∂ x̂i
where κ̂ and γ̂ characterize the permeability of the material and α̂ is a
penalization factor which characterizes the impermeability of the material.
For κ̂ = 1, γ̂ = 1 and α̂ = 0 one recovers the original Stokes equations.
In the absence of external forces, for κ̂ = 0 the flow will stagnate as the
driving pressure gradient vanishes, i.e. the flow velocity is zero as it is in
solid material. For γ̂ → 0 the viscosity becomes infinite, again mimicking
solid material. As the Brinkman penalization factor, α̂, increases the flow
velocity also needs to vanish in order to satisfy the modified momentum
equation (15). Note the Brinkman penalization can be also used together
with one of two formulations of Darcy’s law.
In density-based flow topology optimization the permeability coefficients
and the Brinkman penalization factor are defined as functions of the density.
While both versions of Darcy’s law, (13) and (14) seem equivalent, they
behave differently in the context of density-based topology optimization.
The density dependent permeability κ̂ enters the governing equations as a
value while the permeability γ̂ enters them as a spatial derivative. Thus,
the formulation (14) shares great similarity with the governing equations
used in topology optimization of diffusive transport and elasticity. Indeed,
formulation (14) also leads to an ill-posed problem. For more details on this
issue, the reader is referred to the chapter on diffusive transport problems;
the topic is discussed in detail by Bendsøe and Sigmund (2003). In contrast,
the formulation of Darcy’s law (13) and Brinkman penalization lead to well-
posed problems and the optimization results converge as the mesh is refined.
Given the popularity of the Brinkman approach for both Stokes and
Navier-Stokes models, we will consider only this approach in the following
discussions. For Navier-Stokes models, the Brinkman term is added to the
momentum equations (1) in the same way as it was shown for the Stokes
414 K. Maute
equations. Again ignoring external volume forces, the strong form of the
momentum equations reads:
∂v̂i ∂ σ̂ij
ρ̂ v̂j = + α̂v̂i (16)
∂ x̂j ∂ x̂j
4 Flow through Porous Media

First we seek to understand the influence of the physical phenomena dom-
inating the flow through porous media and study their influence on the
formulation and solution of flow topology optimization problems. We focus
on design problems where the energy dissipation in the design domain is to
be minimized. For example, let us consider the two-dimensional channel in
Figure 1a, with a circular body of homogenous and variable density placed
at its center. The dissipated energy is obviously minimum if the body van-
ishes. Therefore, to avoid trivial solutions, the volume that can be occupied
by fluid needs to be constraint.
a. b.
D
c. d. 0 1 U
D̂ D
decreasing q.
0 1 U 0 1 U
Figure 1. Influence of material interpolation on drag on cylinder.
Consider a simple linear interpolation to define the Brinkman penaliza-

tion factor α̂ as a function of the normalized density ρ̃ : 0 ≤ ρ̃ ≤ 1, with
ρ̃ = 1 representing solid material:
α̂ = α̂U ρ̃, (17)
where α̂U is a sufficiently large, upper bound on the penalization parameter.

In Figure 1b, the drag is plotted over the density. As the density increases
the drag rapidly increases to the value of the solid cylinder. Note these
results are qualitative, but represent a wide variety of flow regimes and
boundary conditions.
When imposing a constraint on the volume that can be occupied by fluid,
the strong non-linearity observed above leads to an inherent penalization
and favors “0-1” solutions. However, the large gradients of the dissipated
energy (and the flow field in general) with respect to the density may cause
numerical issues and the optimization process often converges to a local
minimum. To reduce the gradients, the penalization coefficient is typically
interpolated such that the sensitivity of the flow field with respect to the
density is reduced. Borrvall and Petersson (2003) proposed the following
interpolation:
1 − ρ̃
α̂ = α̂U + (v α̂L − α̂U ) (1 + q) . (18)
1 − ρ̃ + q
The influence of the interpolation penalty parameter q is illustrated in Fig-
ure 1c. As shown in Figure 1d, the gradients of the drag are significantly
reduced by the interpolation (18) when compared to a linear interpolation
which is marked by the dashed line.
Based on the interpolation (18) two-dimensional and three-dimensional
flow topology optimization problems have been solved using either Stokes or
Navier-Stokes flow models. However, the parameters of the optimization al-
gorithms and the interpolation penalty q in (18) need to be chosen carefully.
To reduce the tendency of the optimization process converging to a local
minimum, the enforcement of the volume constraint needs to be controlled
such that constraint violations are tolerated initially in the optimization
process. The interpolation penalty should be as low as possible but high
enough to prevent elements with intermediate porosities from appearing in
the final design.
To illustrate the potential of topology optimization applied to flow problems,
Figure 2 displays some frequently studied problems. In all examples, the
objective is to minimize the pressure drop while constraining the volume
that can be occupied by fluid. The pressure drop is predicted by a finite
element model of the incompressible Navier-Stokes equations augmented by
the Brinkman penalization. The flow field is discretized by standard 4-node
(2D) and 8-node (3D) elements. The SUPG and PSPG stabilization schemes
of Tezduyar et al. (1992) are used to prevent node-to-node oscillations. At
the inlet the flow is prescribed and the outlet is assumed to be traction-free.
416 K. Maute
no design area - fluid
inlet
outlet
Figure 2. Examples of topology optimization of flow problems.
The density distribution is discretized by the same mesh as used for the flow
analysis. The optimization variables are the nodal densities; the Brinkman
penalization parameter is constant within one element and computed via the
interpolation scheme (18) using the averaged nodal densities. The gradients
of the pressure drop with respect to the optimization variables are computed
by an adjoint approach. The optimization problem is solved via the globally
convergence method of moving asymptotes (GCMMA) of Svanberg (2002).
Details on these examples can be found in Kreissl and Maute (2011) and
Kreissl (2011).
The figure at the beginning of this chapter shows yet another interesting
application of topology optimization; here the design problem is to create a
so-called leaky valve. The objective is to minimize the pressure drop for a
flow entering the design domain at the top of the left edge and exiting the
domain at the bottom of the right edge and to maximize the pressure drop
for the reverse flow direction. More details on the formulation and solution
of this problem can be found in Pingen et al. (2008) and Pingen (2008).
6 Concluding Remarks
While topology optimization applied to flow problems has advanced sig-
nificantly, there are numerous unresolved issues that need to be addressed
in future research. Most studies focus on low-Reynolds number applica-
tions with laminar flows at steady-state conditions. Most engineering flow
problems, however, involve high-Reynolds number, turbulent, and unsteady
flows. A few of the many issues one encounters when applying established
techniques to unsteady flows are discussed in Kreissl and Maute (2011). So
far, the objective is typically a measure of the energy dissipation and ad-
ditional constraints on the fluid volume are needed to obtain “0-1” results.
This particular combination of objective and constraints is overly restrictive
and does not allow for tackling a broad range of flow design problems. An
example of this issue is discussed in Kreissl and Maute (2012).
7 Acknowledgment
under grants CMMI–1235532 and CBET–1246854. The opinions and con-
clusions presented in this paper are those of the authors and do not neces-
sarily reflect the views of the sponsoring organization.
Bibliography
T. Borrvall and J. Petersson. Topology optimization of fluids in Stokes
flow. International Journal for Numerical Methods in Fluids, 41(1):77–
107, 2003. ISSN 0271-2091.
V. Challis and J. Guest. Level set topology optimization of fluids in stokes
flow. International Journal for Numerical Methods in Engineering, 79
(10):1284–1308, 2009.
A. L. Cunha. A Fully Eulerian Method for Shape Optimization with Appli-
cation to Navier-Stokes Flows. PhD thesis, Carnegie Mellon University,
Pittsburgh, PA 15213, September 2004.
X. Duan, Y. Ma, and R. Zhang. Optimal shape control of fluid flow us-
ing variational level set method. Physics Letters A, 372(9):1374—1379,
2008a.
X. Duan, Y. Ma, and R. Zhang. Shape-topology optimization for navier-
stokes problem using variational level set method. Journal of computa-
tional and applied mathematics, 222(2):487–499, 2008b.
418 K. Maute
X. Duan, Y. Ma, and R. Zhang. Shape-topology optimization of stokes flow

via variational level set method. Applied Mathematics and Computation,
202(1):200–209, 2008c.
A. Evgrafov, G. Pingen, and K. Maute. Topology optimization of fluid
domains: Kinetic theory approach. In In proceedings of the 7th World
Congress on Structural and Multidisciplinary Optimization, May 21-25,
Seoul, Korea. ISSMO, 2007.
A. Gersborg-Hansen, O. Sigmund, and R. Haber. Topology optimization of
channel flow problems. Structural and Multidisciplinary Optimization,
30(3):181–192, 2005.
J. K. Guest and J. H. Prévost. Optimizing multifunctional materials: Design
of microstructures for maximized stiffness and fluid permeability. Inter-
national Journal of Solids and Structures, 43(22–23):7028–7047, 2006.
M. Gunzburger. Advances in Design and Control. SIAM, 2003.
S. Kreissl and K. Maute. Topology optimization for unsteady flow. Inter-
national Journal for Numerical Methods in Engineering, 87:1229–1253,
2011.
S. Kreissl, G. Pingen, and K. Maute. An explicit level set approach for gen-
eralized shape optimization of fluids with the lattice boltzmann method.
International Journal for Numerical Methods in Fluids, 65(5):496–519,
2011.
S. Kreissl. Topology Optimization of Flow Problems Modelled by the Incom-
pressible Navier-Stokes Equations. PhD thesis, University of Colorado,
2011.
S. Kreissl and K. Maute. Levelset based fluid topology optimization using
the extended finite element method. Structural and Multidisciplinary
Optimization, pages 1–16, 2012.
B. Mohammadi and O. Pironneau. Applied Shape Optimization for Fluids.
Oxford University Press, 2001.
B. Mohammadi and O. Pironneau. Theory and practice of optimal shape
design. European Journal of Computational Mechanics, 17(1-2):13–30,
2008.
F. Okkels, L. H. Olesen, and H. Bruus. Applications of topology optimiza-
tion in the design of micro-and nanofluidic systems. In Technical Pro-
ceedings of the 2005 NSTI Nanotechnology Conference and Trade Show,
pages 575–578, 2005.
L. H. Olesen, F. Okkels, and H. Bruus. A high-level programming-language
implementation of topology optimization applied to steady-state Navier–
Stokes flow. International Journal for Numerical Methods in Engineer-
ing, 65(7):975–1001, 2006.
C. Othmer. CFD topology and shape optimization with adjoint methods.

In VDI Fahrzeug- und Verkehrstechnik, 13. Internationaler Kongress,
Berechnung und Simulation im Fahrzeugbau, Würzburg, Germany,
September 2006.
C. Othmer, E. de Villiers, and H. G. Weller. Implementation of a continuous
adjoint for topology optimization of ducted flows. In Proceedings of
the 18th AIAA Computational Fluid Dynamics Conference, Miami, FL.
AIAA, 2007.
G. Pingen, M. Waidmann, A. Evgrafov, and K. Maute. A parametric level-
set approach for topology optimization of flow domains. Structural and
G. Pingen. Topology optimization and shape optimization with the lattice
boltzmann method. Ph.d. thesis, University of Colorado at Boulder, 2008.
G. Pingen, A. Evgrafov, and K. Maute. Topology optimization of flow
domains using the lattice boltzmann method. Structural and Multidisci-
plinary Optimization, 36(6):507–524, 2007a. ISSN 1615-147X.
G. Pingen, M. Waidmann, A. Evgrafov, and K. Maute. Application of a
parametric level-set approach to topology optimization fluid with the
navie-stokes and lattice boltzmann equations. In In proceedings of the
7th World Congress on Structural and Multidisciplinary Optimization,
May 21-25, Seoul, Korea. ISSMO, 2007b.
G. Pingen, A. Evgrafov, and K. Maute. A parallel schur complement solver
for the solution of the adjoint steady-state lattice boltzmann equations:
application to design optimisation. International Journal of Computa-
tional Fluid Dynamics, 22(7):457–464, 2008.
tion, 12(2):555–573, 2002. ISSN 1052-6234.
T. E. Tezduyar, S. Mittal, S. E. Ray, and R. Shih. Incompressible flow com-
putations with stabilized bilinear and linear equal-order-interpolation
velocity-pressure elements. Computer Methods in Applied Mechanics and
Engineering, 95:221–242, 1992.
Topology Optimization
of Coupled Multi-Physics Problems
*
Kurt Maute
*
Abstract. Topology optimization provides a promising approach

to systematically design multi-physics problems, such as thermo-
mechanical, electro-static, and fluid-structure interaction problems.
This class of design problems is often dominated by nonlinear phe-
nomena and is not well suited for intuitive design strategies. In
this lecture we will discuss applications of topology optimization
methods to coupled multi-physics problems, emphasizing the dif-
ferences between volumetric and interface coupling in the context
of topology optimization. Focusing on density methods, topology
optimization of piezo-electric devices and fluid-structure interaction
problems will be studied.
1 Introduction
The performance of many engineering systems often depends on multiple
physical phenomena belonging to different engineering disciplines, such as
solid mechanics, fluid mechanics, and heat transfer. Topology optimization
methods have been developed and applied to problems that are dominated

DOI 10.1007/978-3-7091-1643-2_18, © CISM, Udine 2014
422 K. Maute
by a single phenomenon, such as elastic deformations in structural mechan-

ics and the flow of liquids and gases in fluid mechanics. The reader is referred
to the textbook by Bendsøe and Sigmund (2003) for an overview of topology
optimization. In this lecture we will discuss topology optimization methods
for coupled problems where the interaction of multiple physical phenomena
needs to be accounted for. Such problems are often labeled “multi-physics”
and this terminology will be used subsequently. Before discussing individ-
ual applications, first we will examine the basic modes in which different
physical fields interact and how this interaction is modeled in the context
of topology optimization.
1.1 Volumetric and Surface Coupling

Let us assume two scalar physical fields, u and v, which are governed by
two partial differential equations, here linear diffusion equations. For the
sake of simplicity, we study these fields for steady state conditions, i.e. they
do not vary with time. First we consider the case where both fields are
defined over the same domain Ω.
∂ u
j + q u = 0 in Ω jiu ni = Qu on Γqu u = û on Γu , (1)
∂xi i
∂ v
j + q v = 0 in Ω jiv ni = Qv on Γqv v = v̂ on Γv , (2)
∂xi i
with ji denotes the diffusive fluxes, q are external volumetric fluxes, and
Q are external surface fluxes. The fields are coupled via the constitutive
laws and/or the external volumetric fluxes:
jiu = jiu (uj , vj ) and jiv = jiv (uj , vj ) , (3)

q u = q u (uj , vj ) and v v
q = q (vj , uj ) . (4)
In both cases, the fields are coupled in each material point or a subset of
points within the volume of the body. Examples for this type of coupling
include Joule heating for electro-thermal problems, thermal expansion in
thermo-elastic problems, and piezo-electric coupling, among others.
The other basic form of interaction between multiple physical fields is
the coupling through interface conditions. In this case the physical fields
are only defined over non-overlapping subsets Ωu and Ωv with Ω = Ωu ∪ Ωv .
The interface Γuv is defined by Γuv = Ωu ∩ Ωv with nuv being the normal
vector on Γuv pointing from Ωu to Ωv . The governing equations for this
Topology Optimization of Coupled Multi-Physics Problems 423
form of coupling are:

∂ u
j + q u = 0 in Ωu jiu ni = Qu on Γqu u = û on Γu (5)
∂xi i
∂ v
j + q v = 0 in Ωv jiv ni = Qv on Γqv v = v̂ on Γv (6)
∂xi i
j uv = 0 on Γuvj and J(u, v) = 0 on ΓuvJ , (7)
where j uv = 0 represents a flux interface condition and J(u, v) = 0 an in-
terface condition operating directly on the values of the physical variables,
rather than their spatial derivatives. Note the physical fields are only cou-
pled along the interface Γuv . To satisfy the interface conditions, in general,
one needs to know the geometry of the interface, its normal, and in some
cases even higher order spatial derivative information, such as the curva-
ture. Examples for this class of coupled problems include fluid-structure
coupling and the interaction of structural deformations with electrostatic,
magnetic, and electro-magnetic fields.
There are numerous problems in engineering where both volumetric and
surface coupling phenomena are present. For example, coupling phenomena
that involve multiple scales may involve both forms of coupling: at the
micro-scale, fields may be coupled through surface interactions while, at
the macro-scale, the micro-scale coupling phenomena are described through
homogenized constitutive models. Examples for such multi-scale, multi-
physics coupling phenomena are frequently found in transport models for
heterogeneous materials.
In general, the responses of the physical fields depend on each other. In
special cases the dependency of one field on the other can be neglected. One
typically distinguishes between “two-way” and “one-way” coupling. Fluid
structure interaction is typically modeled as a “two-way” coupled problem
as the flow field changes with the structural deformations and in turn the
structural deformations depend on the forces that the fluid exerts on the
structure. Electro-thermo coupling can be often simplified to a “one-way”
coupled problem where Joule heat due to electrical current acts a volumetric
source term for the temperature field but the temperature does not influence
the governing equations of the electric field. This simplification is only
valid when the temperature dependency of the electric conductivity can
be neglected. Another well-known “one-way” coupled problem is advective
heat transport of incompressible fluids. Here the flow just advects thermal
energy but does not depend on the temperature. This simplification is only
valid if buoyancy effects can be neglected.
While not immediately obvious, in most cases coupling along interfaces
is significantly more complex than coupling through volumetric terms, both
424 K. Maute
with regards to the solution of the governing equations and the parametriza-
tion of the geometry in topology optimization. This is in particular the
case for problems involving deforming structures, which represents a class
of problems that is of great interest to the structural design optimization
community. The main issues of surface coupled problems can be summa-
rized as follows:
Formulation: Typically when physical fields are coupled via volumetric

terms both fields are described with respect to the same observer
frame, i.e. the formulations of both fields are either Eulerian, La-
grangian, or Arbitrary Lagrangian Eulerian (ALE). For surface cou-
pled problems involving elastic deformations it is however common to
use a Lagrangian description for the structure and express the exter-
nal field, such as the flow field or the electro-static field, with respect
to an Eulerian or ALE observer frame.
Nonlinearity: Even if the individual governing equations are linear with
respect with their own state variables, accounting for surface interac-
tions typically leads to nonlinear terms in the set of governing equa-
tions. While in the case of volumetric coupling through constitutive
models and volumetric source terms such nonlinearities may arise too,
such as in pressure-driven diffusion models, most volumetric coupling
phenomena considered in topology optimization are linear, such as
the thermo-mechanical coupling and piezo-electric models studied in
Sigmund (2001b), Sigmund (2001a), Silva et al. (1999a), Silva et al.
(1999b), and Silva and Kikuchi (1999). In contrast, surface coupling
phenomena involving elastic deformations require modeling the de-
formed state of the structure, which inherently leads to nonlinear
coupling terms.
Geometry: A particular difficulty arises in density based topology opti-
mization as the geometry of the interface is not explicitly defined,
but needs to be reconstructed from the density distribution. While
related problems have been studied in the context of design depen-
dent loads, for example by Hammer and Olhoff (2000) and Kikuchi
and Chen (2001), methods developed for such problems have limited
applicability for the “two-way” surface interaction problems.
In the remainder of this chapter, we will study two multi-physics prob-

lems: piezo-electric problems as an example for volumetric coupling and
fluid-structure interaction as a representative for surface coupling. In both
cases we will focus on the application of density methods to multi-physics
problems.
2 Topology Optimization of Piezo-Electric Structures

Piezo-electricity is the electro-mechanical interaction between the mechani-
cal and the electrical state in crystalline materials. Under the application of
an electric field the material deforms and under mechanical loads an electric
field is generated. This effect is used in numerous engineering applications,
including actuators, sensors, transducers, and energy harvesting devices.
Due to its great importance for a broad range of technologies, the design of
piezo-electric devices by topology optimization has been addressed in many
studies: Silva et al. (1999a), Silva et al. (1999b), Silva and Kikuchi (1999),
Ha and Cho (2006), Wang et al. (2006), Carbonari et al. (2007), Donoso and
Sigmund (2008), Zheng et al. (2008), Kang and Tong (2008), Rupp et al.
(2009), Luo et al. (2009), and Rupp et al. (2010).
Piezo-electricity couples the structural deformations and the electric field
via constitutive laws. The governing equations are:
∂
elasticity: σij = 0 in Ω, (8)
∂xj
∂
electric field: dj = 0 in Ω, (9)
∂xj
where σij is the elastic stress tensor and dj is the electric displacement
vector. For the sake of a compact presentation of the aspects relevant
to topology optimization, we consider only the static case, neglect body
force in the structural equilibrium equations, and omit the definition of the
boundary conditions. The level of difficulty of extending the static case to
free and forced vibration problems depends on the particular application,
in particular whether the dynamics of the electric circuitry attached to the
piezo-electric device needs to be included in the model; see, for example,
Rupp et al. (2009).
The constitutive equations accounting for piezo-electric coupling are:
σij = Cijkl εkl − Eijk ek , (10)

dj = Ejkl εkl + Djk ek , (11)
where Cijkl is the elastic tensor and εkl is the strain tensor. The dielectric
tensor is denoted by Djk and the electric field by ek . The coupling of the
mechanical and electric problems is described by the piezo-electric coupling
tensor Eijk . In electrostatics, the electric field is ek = −(∂φ)/(∂xk ), with φ
being the electric potential.
Applying a density method, the material properties need to be inter-
polated as functions of the density. This can be done via homogenization
426 K. Maute
of micro-structural architectures: the micro-structural geometry determines

macroscopic density and effective material properties. Over the past years,
algebraic interpolation rules, such as the Solid Isotropic Material with Pe-
nalization (SIMP) method of Bendsøe (1989) and Zhou and Rozvany (1991),
have become popular. Following this approach, we interpolate the coeffi-
cients of the elastic tensor, dielectric tensor, and the piezo-electric coupling
tensor as functions of the density. For example, assuming that the elastic
and dielectric tensors are isotropic the following interpolations need to be
defined:
em = em(ρ) de = de(ρ) Eijk = Eijk (ρ), (12)
where em denotes the Young’s modulus and de the dielectric constant. Note
it may not be necessary to interpolate all material properties; in the example
above, the Poisson ratio was chosen to be independent of the density.
The interpolation rules need to be designed such that the material distri-
bution converges to a discrete material distribution; for example, to “0-1”
solution in the case of two-phase, solid-void design problems. In density
methods, a SIMP interpolation scheme is often adopted and an additional
mass constraint is introduced. Optimizing the stiffness of a structure, the
interplay between an nonlinear interpolation of the material stiffness and
the linear interpolation of the density leads to an implicit penalization of
intermediate densities and favors a “0-1” design. However, optimizing piezo-
electric devices, the SIMP approach may or not lead to an implicit penal-
ization of intermediate densities, depending on the design objective. As
the physical phenomena become more complex it is typically not trivial to
find interpolation rules for all parameters such that a convergence to a “0-
1” solution can be achieved. If no appropriate interpolation rules can be
found, formulations explicitly penalizing intermediate densities may need to
be considered and used in combination with continuation methods. How-
ever, for multi-physics problems, one needs to ensure that the interpolations
are designed such that for intermediate densities the response of the prob-
lem is still “physical”, driving the optimization process toward a physically
meaningful solution.
2.1 Example: Design of an Energy Harvester

In Figure 1 we show the topology optimization results for energy har-
vesting device, accounting for the dynamics of simple resistive circuit. A
square plate is clamped at the middle two-fifths of one side and is subject
to a harmonic excitation at the clamped location normal to the plane of the
plate. The plate consists for four layers: a layer of transversely isotropic
piezo-electric material is sandwiched between an upper and lower electrode
Figure 1. Topology optimization of a piezo-electric energy harvesting de-

vice.
and mounted on an elastic substrate. For details on the material properties

the reader is referred to Rupp et al. (2009).
The objective of the design problem is to determine the optimal layout of
piezoelectric material such that the power dissipated by a constant resistor
is maximized for a given excitation frequency:
2
(Δϕ)
z= , (13)
2R
where Δϕ is the difference between the voltage at the upper and lower
electrodes and R is the resistivity of the external circuit.
As shown by Rupp et al. (2009) for forced vibration problems, inter-
polating the properties of the piezo-electric material by linear functions of
the density leads to a crisp “0-1” material layout. In contrast, for static
piezo-electric problems, nonlinear interpolation schemes along with a vol-
428 K. Maute
ume constraints is typically needed to obtain “0-1” material layout.

This optimization problem is solved by the Globally Convergent Method
of Moving Asymptotes (GCMMA) of Svanberg (1995) for six different thick-
ness ratios of piezoelectric to substrate materials. The initial design is a
plate fully covered with piezoelectric material. It was found that for un-
damped structures, the sudden phase change in voltage and displacement
at resonance creates a non-smooth design space that cannot be traversed ef-
ficiently by gradient-based algorithms. In order to alleviate this problem, as
well as make it more practical, mass proportional damping of 1% is included
in the problem.
The resulting optimal material layouts are shown at the bottom of Fig-
ure 1. As the thickness ratio increases the optimal design gradually changes.
The designs optimized for the largest and smallest thickness ratios signif-
icantly differ. For a thin piezoelectric layer the material layout closely
matches the signs of the strain distribution computed for a plate with-
out piezoelectric layer. In this case, piezo-electric material is only placed
in locations with positive strain as adding piezo-electric material in areas
with negative strain would cause charge cancellation and reduce the elec-
tric power generated by the device. With increasing thickness the material
layout differs increasingly from this strain distribution as the piezoelectric
layer adds stiffness and mass, thereby changing the dynamic response of the
structure.
The figure at the beginning of this chapter shows a topology optimiza-
tion result of a piezo-electric wave guide, bending an elastic wave by 90 deg.
Here the optimization variables describe the polarity of the piezo-electric
material. Spatially varying the polarity allows placing piezo-electric mate-
rial everywhere in the design domain while avoiding charge cancellation. For
further details on this application of topology optimization to piezo-electric
problems the reader is referred to Rupp et al. (2010).
3 Topology Optimization of Fluid-Structure

Interaction Problems
One of the most challenging problems in computational mechanics in gen-
eral, and in topology optimization in particular, is the analysis and design
optimization of fluid-structure interaction (FSI) problems. We distinguish
between FSI problems where either the acoustic response of the fluid is of
primary interest or the convective transport of energy and momentum is the
dominating phenomenon. In the context of topology optimization, acous-
tic FSI problems have been studied, for example, by Yoon et al. (2007).
Here we focus on the second class of FSI problem. Their complexity is due
to (a) the problems associated with simulating flows, in particular at high

Reynolds numbers, (b) the coupling along interfaces, and (c) the interaction
of two models which are typically defined in different observer frames, have
radically different discretization requirements, and call for different numer-
ical solution schemes. For a general overview of computational approaches
to fluid-structure interaction problems, the reader is referred to the review
articles by Dowell and Hall (2001) and Kamakoti and Shyy (2004).
To somewhat simplify the problem, here we will consider low Reynolds
number, incompressible flows at steady-state. The governing equations of
the “two-way” coupled fluid-structure interaction problem are:
S
∂ σ̂ij
static equilibrium: − = 0 in ΩS , (14)
∂ x̂j
F
∂v̂i ∂ σ̂ij
fluid momentum equation: ρ̂ v̂j − − α̂vî = 0 in ΩF , (15)
∂ x̂j ∂ x̂j
∂v̂i
incompressibility condition: = 0 in ΩF , (16)
∂ x̂i
where ρ̂ and vî denote the fluid density and velocity, respectively. The
S F
stress tensor of the solid is σ̂ij and in the fluid σ̂ij . The subscripts i, j
define the spatial directions, and ˆ indicates dimensional quantities. The
last term in the fluid momentum equations (15) stems from a Brinkman
model, approximating the flow through porous material. This term can be
used to enforce a stick condition by increasing the impermeability coefficient
α̂. The reader is referred to the chapter on flow topology optimization for
details of the Brinkman penalization method and its application to topology
optimization. Note the structural equilibrium equations are defined in a
Lagrangian observer frame while the fluid equations in an Eulerian one.
The constitutive equations for the solid and fluid are:

F 1 ∂ ûk ∂ ûl
Solid: σ̂ij = Cijkl + (17)
2 ∂xl ∂xk

∂v̂ i ∂v̂ j
F
Fluid: σ̂ij = −p̂δij + μ̂ + , (18)
∂ x̂j ∂ x̂i
where ûi denote the structural displacements, p̂ the fluid pressure, and μ̂
the dynamic viscosity. For the sake of simplicity, we assume a linear elastic
behavior of the solid material and a constant dynamic viscosity.
430 K. Maute
The interface conditions at the fluid-structure interface are:

S SF F FS
σ̂ij nj + σ̂ij nj = 0 on ΓF S , (19)
FS
vî = 0 on Γ , (20)
FS FS
xΓi − XiΓ − ûi = 0 on ΓF S . (21)
(22)
The first condition states that the traction forces generated by the fluid
need to be in equilibrium with the surface traction in the solid. The second
condition enforces the stick condition. The third equation describes the
location of the fluid-structure interface as a function of the structural dis-
FS FS
placements, with xΓi and XiΓ denoting the position of the interface in
the deformed and undeformed configurations, respectively. Again we omit
the definition of the boundary conditions along the external interfaces as
they play no role in the following discussion.
An intuitive approach to enforce the above interface conditions is to con-
sider the following three steps: (a) the fluid exerts a force on the structure,
(b) the structure deforms moving the fluid-structure interface, and (c) the
stick conditions are enforced at the deformed interface and a new flow so-
lution is computed. These steps identify the main functionality a coupling
method needs to provide. The reader is referred to Felippa et al. (2001) for
an in-depth discussion of coupling methods.
First, let us assume that the location of the interface in the undeformed
configuration is explicitly defined. In this case, evaluating the fluid forces
and computing the resulting structural displacements are straight forward.
The interface geometry is updated and the stick conditions are enforced.
In general, there are two formulations to project the structural deforma-
tions onto the fluid model: (a) The fluid is formulated in an ALE observer
frame which allows using a Lagrangian formulation on the interface. The
fluid mesh is deformed along with the structure and the stick conditions
can be satisfied at ΓF S . The weak form of the governing equations of the
fluid can be integrated either in the deformed configuration or, using a
push-back projection, in the undeformed configuration. (b) Alternatively,
an immersed boundary technique can be used. In this case, the flow formu-
lation remains strictly Eulerian but the stick conditions are enforced at the
deformed structure interface via either interpolation methods or by formu-
lating interface constraints. In contrast to ALE formulation and deforming
mesh approach, immersed boundary techniques are less common for simu-
lating fluid-structure interaction problems; examples can be found in Zhu
and Peskin (2002), Mittal and Iaccarino (2005), and Gerstenberger and Wall
(2008).
Both coupling concepts, the ALE formulation of the fluid as well as im-
mersed boundary techniques have been integrated into topology optimiza-
tion approaches; see, for example, the work by Maute and Allen (2004),
Maute and Reich (2006), James et al. (2008),Stanford and Ifju (2009), Yoon
(2010), and Kreissl et al. (2010). These approaches can be further classified
by the fluid and structural models used and by the design changes possi-
ble in the optimization process. Most often a linear elastic model assuming
small deformations is adopted to model the structural response, and the flow
is predicted by either a potential flow, an Euler or a Navier-Stokes model.
Depending on the combination of models, computing the design sensitivities
may pose a major challenge; see for example the work by Maute et al. (2003)
and Barcelos and Maute (2008) on adjoint sensitivity analysis methods for
compressible Navier-Stokes models. The complexity of the structural and
fluid models may further complicate the extension onto transient problems.
3.1 Topology Optimization with fixed Fluid-Structure Interface

In most topology optimization methods for FSI problems, the topology
of the fluid-structure interface is prescribed and cannot be varied in the op-
timization process. Only the layout of the structure in a design domain en-
closed by a given interface is optimized. Examples for this approach can be
found in the work of Maute and Allen (2004), Maute and Reich (2006), and
Stanford and Ifju (2009). Following the basic concept of density methods
for structural topology optimization, the density distribution describes the
geometry of the structure in the undeformed configuration. Homogenization
based or algebraic interpolation schemes are used to define the properties
of the solid material as function of the density. The interface conditions
along the fluid-structure interface are enforced in a standard manner as the
interface geometry is well defined and discretized. The primary challenges
of these approaches are the computation of the design sensitivity, the for-
mulation of suitable objectives and constraints, and the fine-tuning of the
material interpolation scheme to avoid an excessive amount of intermediate
densities.
In Figure 2 the topology optimization result for the design of internal
mechanism of a morphing airfoil is shown. The design domain is restricted
to the interior of the airfoil, enclosed by a skin. The goal is to find the
layout of a compliant mechanism that converts two discrete actuation forces
into a deformation of the skin such that the lift-to-drag ratio of the airfoil
in the deformed configuration is maximized. The volume of material in
design domain and the maximum stroke of the actuators are constrained.
The reader is referred to Maute and Reich (2006) for details regarding the
432 K. Maute
actuator skin
pressure
Figure 2. Topology optimization of a morphing airfoil.
formulation and numerical treatment of the design problem. The optimized

material layout in the undeformed configuration along with the fluid mesh
and in deformed configurations together with fluid pressure contours are
shown in Figure 2.
3.2 Topology Optimization with variable Fluid-Structure Inter-

face
The complexity of FSI topology optimization increases significantly if
the shape and topology of the fluid-structure interface is not prescribed but
evolves in the optimization process. Therefore, the fluid and structure equa-
tions need to be modeled in the entire design domain Ω. Adopting standard
techniques from structural and fluid topology optimization, a low stiffness
is assigned to material within the fluid and, using a Brinkman penaliza-
tion, a high impermeability is assigned to material with the structure. The
main challenge is to enforce the interface constraints as the fluid-structure
boundary is not explicitly defined but only implicitly through the density
distribution.
In the studies of Yoon (2010) and Kreissl et al. (2010), a Brinkman pe-
nalization approach is used to enforce the stick conditions (20). In contrast
to standard flow topology optimization, here the penalization coefficient α̂

in (15) depends on the density distribution in the deformed configuration.
In other words, the density distribution that is relevant for enforcing the
flow boundary conditions depends on the structural response. Note apply-
ing a stick condition (20) along the fluid-structure interface is only correct
for steady-state problems. For transient problems, the fluid velocities need
to equal the structural velocities along the interface.
Using an ALE formulation where the same triangulation is used to dis-
cretize the weak forms of the fluid and solid equations, the density dis-
tribution ρ(xi ) can be easily mapped into a distribution of the Brinkman
penalty factor α̂(xi ). Here the fluid and solid meshes coincide in both the
undeformed and deformed configurations. This approach has been proposed
by Yoon (2010). Using non-matching meshes or when the fluid mesh is not
deformed and immersed boundary techniques are used, the density distri-
bution ρ(xi ) defined on the deformed structural mesh needs to be mapped
onto the fluid mesh. Such an approach was introduced by Kreissl et al.
(2010), modeling the fluid by a lattice Boltzmann method and describing
the structural response by a finite element model. Note mapping the density
distribution of the deformed structure onto the fluid satisfies the interface
condition (21).
To satisfy the interface condition on the surface traction, (19), the surface
forces can be converted into an equivalent volumetric force using Stoke’s
divergence theorem.
F
F FS
∂ σ̂ij
σ̂ij nj dΓ = κ(ρ) dΩ, (23)
ΓF S Ω ∂ x̂j
where the indicator function κ depends on the material distribution and

vanishes in the fluid phase. Yoon (2010) interpolates κ by a power law.
The volumetric forces are applied to the structure. Note this approach re-
quires evaluating the fluid stress within the structural domain and therefore
relies on the “accuracy” of the Brinkman model to predict the pressure dis-
tribution within the solid domain. As shown by Kreissl and Maute (2011),
however, the Brinkman model may allow for incorrect pressure diffusion, in
particular for slender structures and coarse discretizations. Furthermore,
interpolating the indicator function as a function of the density distribution
smears out the zone over which the volumetric fluid forces are applied and
may lead to numerical artifacts.
434 K. Maute
Topology optimization provides an exciting tools for systematically study-
ing multi-physics design problems as these problems are often dominated by
nonlinear phenomena and benefit from non-intuitive design solutions. Den-
sity methods are well-suited for problems that are coupled via constitutive
equations, but face largely unresolved challenges when the physical domains
interact along interfaces. Robust and efficient topology optimization meth-
ods for complex multi-physics problems that require an accurate resolution
of the individual physical fields along the interfaces, such as transient FSI
problems at higher Reynolds numbers, are currently lacking. While level-
set methods have been applied successfully to multi-physics problems, they
have not expanded yet the range of design problems that can be solved by
topology optimization.
5 Acknowledgment
under grants CMMI–1235532 and CBET–1246854. The opinions and con-
Bibliography
M. Barcelos and K. Maute. Aeroelastic design optimization for laminar and
turbulent flows. Computer Methods in Applied Mechanics and Engineer-
ing, 197(19):1813–1832, 2008.
M. Bendsøe. Optimal shape design as a material distribution problem.
Structural and Multidisciplinary Optimization, 1(4):193–202, 1989.
R. C. Carbonari, E. C. N. Silva, and G. H. Paulino. Topology optimiza-
tion design of functionally graded bimorph-type piezoelectric actuators.
Smart Materials and Structures, 16:2605–2620, 2007.
A. Donoso and O. Sigmund. Optimization of piezoelectric bimorph actua-
tors with active damping for static and dynamic loads. Structural and
Multidisciplinary Optimization, 2008.
E. H. Dowell and K. C. Hall. Modeling of fluid-structure interaction. Annual
Review of Fluid Mechanics, 33(1):445–490, 2001.
C. A. Felippa, K. Park, and C. Farhat. Partitioned analysis of coupled
mechanical systems. Computer methods in applied mechanics and engi-
neering, 190(24):3247–3270, 2001.
A. Gerstenberger and W. A. Wall. An extended finite element

method/Lagrange multiplier based approach for fluid-structure interac-
tion. Computer Methods in Applied Mechanics and Engineering, 197:
1699–1714, 2008.
Y. Ha and S. Cho. Design sensitivity analysis and topology optimization
of eigenvalue problems for piezoelectric resonators. Smart Materials and
Structures, 15:1513–1524, 2006.
V. Hammer and N. Olhoff. Topology optimization of continuum structures
subjected to pressure loading. Struct. Multidisc. Optim., 19:85–92, 2000.
K. A. James, J. Martins, and J. Hansen. Three-dimensional structural
topology optimization of an aircraft wing using level set methods. In
Proceedings of the 12th AIAA/ISSMO Multidisciplinary and Optimiza-
tion Conference, Victoria, British Columbia, Canada, pages 1–12, 2008.
R. Kamakoti and W. Shyy. Fluid–structure interaction for aeroelastic ap-
plications. Progress in Aerospace Sciences, 40(8):535–558, 2004.
Z. Kang and L. Tong. Integrated optimization of material layout and control
voltage for piezoelectric laminated plates. Journal of Intelligent Material
Systems and Structures, 19(8):889–904, 2008.
N. Kikuchi and B.-C. Chen. Topology optimization with design-dependent
loads. Finite Elements in Analysis and Design, 37(1):57–70, January
2001.
S. Kreissl and K. Maute. Topology optimization for unsteady flow. Inter-
national Journal for Numerical Methods in Engineering, 87:1229–1253,
2011.
S. Kreissl, G. Pingen, A. Evgrafov, and K. Maute. Topology optimization
of flexible micro-fluidic devices. Structural and Multidisciplinary Opti-
mization, 42(4):495–516, 2010.
Z. Luo, L. Tong, J. Luo, P. Wei, and M. Wang. Design of piezoelectric
actuators using a multiphase level set method of piecewise constants.
Journal of Computational Physics, 228(7):2643–2659, 2009.
K. Maute and M. Allen. Conceptual design of aeroelastic structures by
topology optimization. Structural and Multidisciplinary Optimization,
27:27–42, 2004.
K. Maute and G. Reich. Integrated multidisciplinary topology optimization
approach to adaptive wing design. AIAA Journal of Aircraft, 43(1):
253–263, 2006.
K. Maute, M. Nikbay, and C. Farhat. Sensitivity analysis and design op-
timization of three-dimensional nonlinear aeroelastic systems by the ad-
joint method. International Journal for Numerical Methods in Engineer-
ing, 56(6):911–933, 2003.
R. Mittal and G. Iaccarino. Immersed boundary methods. Annual Review
Of Fluid Mechanics, 37:239–261, 2005.
436 K. Maute
C. J. Rupp, A. Evgrafov, K. Maute, and M. L. Dunn. Design of piezoelectric

energy harvesting systems: a topology optimization approach based on
multilayer plates and shells. Journal of Intelligent Material Systems and
Structures, 20(16):1923–1939, 2009.
C. J. Rupp, M. L. Dunn, and K. Maute. Switchable phononic wave filtering,
guiding, harvesting, and actuating in polarization-patterned piezoelectric
solids. Applied Physics Letters, 96(11):111902–111902, 2010.
O. Sigmund. Design of multiphysics actuators using topology optimiza-
tion - part II: Two-material structures. Computer Methods in Applied
Mechanics and Engineering, 190(49-50):6605–6627, 2001a.
O. Sigmund. Design of multiphysics actuators using topology optimiza-
tion – Part I: One–material structures. Computer Methods in Applied
Mechanics and Engineering, 190(49–50):6577–6604, 2001b.
E. C. N. Silva and N. Kikuchi. Design of piezocomposite materials and
piezoelectric transducers using topology optimization - part iii. Archives
of Computational Methods in Engineering, 6(4):305–329, 1999.
E. C. N. Silva, J. S. O. Fonseca, F. M. de Espinosa, A. T. Crumm, G. A.
Brady, J. W. Halloran, and N. Kikuchi. Design of piezocomposite mate-
rials and piezoelectric transducers using topology optimization - part i.
Archives of Computational Methods in Engineering, 6(2):117–182, 1999a.
E. C. N. Silva, S. Nishiwaki, and N. Kikuchi. Design of piezocomposite
materials and piezoelectric transducers using topology optimization -
part ii. Archives of Computational Methods in Engineering, 6(3):191–
222, 1999b.
B. Stanford and P. Ifju. Aeroelastic topology optimization of membrane
structures for micro air vehicles. Structural and Multidisciplinary Opti-
mization, 38(3):301–316, 2009.
K. Svanberg. A globally convergent version of MMA without linesearch.
In Proceedings of the First World Congress of Structural and Multidis-
ciplinary Optimization, 28 May - 2 June 1995, pages 9–16, Goslar, Ger-
many, 1995.
S. Y. Wang, K. Tai, and S. T. Quek. Topology optimization of piezoelec-
tric sensors/actuators for torsional vibration control of composite plates.
Smart Materials and Structures, 15:253–269, 2006.
G. Yoon, J. S. Jensen, and O. Sigmund. Topology optimization of acoustic-
structure interaction problems using a mixed finite element formula-
tion. International Journal for Numerical Methods in Engineering, 70
(9):1049–1075, 2007.
G. H. Yoon. Topology optimization for stationary fluid–structure interaction
problems using a new monolithic formulation. International journal for
numerical methods in engineering, 82(5):591–616, 2010.
B. Zheng, C.-J. Chang, and H. C. Gea. Topology optimization of energy

harvesting devices using piezoelectric materials. Structural and Multi-
disciplinary Optimization, 2008.
M. Zhou and G. I. N. Rozvany. The COC algorithm, part II: Topological,
geometrical and generalized shape optimization. Computer Methods in
Applied Mechanics and Engineering, 89(1-3):309–336, 1991.
L. Zhu and C. S. Peskin. Simulation of a flapping flexible filament in a
flowing soap film by the immersed boundary method. Journal of Com-
putational Physics, 179(2):452–468, 2002.
The Extended Finite Element Method
*
Kurt Maute
*
University of Colorado Boulder, Boulder, USA
Abstract. This chapter discusses level-set topology optimization

methods where the governing state equations are discretized by the
Extended Finite Element Method (XFEM). In contrast to conven-
tional methods that map the level-set function into a density distri-
bution of a fictitious material, the XFEM allow preserving the crisp
geometry information of level-sets in the mechanical model. A brief
introduction into the XFEM will be provided and its application
to topology optimization will be illustrated with problems in fluid
mechanics.
1 Introduction
Topology optimization is concerned with finding the geometry and the ma-
terial layout of a body such that an objective is minimized (or maximized)
and a set of constraints is satisfied. As discussed in other chapters, the
geometry within a design domain Ω can be defined in two ways, which
are illustrated in Figure 1. The geometry and material layout is uniquely
defined by describing the phase boundaries and by determining whether a
point is inside or outside a domain, which is enclosed by an interface and ex-
ternal boundaries. Alternatively, one can define an integer-valued indicator
function I(xi ) at all points xi ∈ Ω, with the value of the indicator func-
tion being the phase id. While both approaches are equivalent and allow

DOI 10.1007/978-3-7091-1643-2_19, © CISM, Udine 2014
440 K. Maute
Geometry of interface Indicator function

& in/outside
(1)
1 x
I (x ) (2)
(2)
2 x
(12)
parameterization parameterization
(1)
(12)h
(12)
(x) (p, x) I (x) I h (p, x )
Figure 1: Approaches to describe the geometry and material layout of a

body.
describing arbitrary complex geometries and material layouts, they require

different parametrization methods and lead to seemingly different topology
optimization methods, such as level-set and density methods. However,
when evaluating the physical response for a given design, both approaches
face similar challenges.
For the sake of simplicity, we revisit the diffusive transport problem,
which is discussed in a separate chapter. Ignoring convection, the weak
form of the governing equation for steady-state conditions is:

∂δT ∂T
dij dΩ − δT q dΩ − δT Q dΓ = 0, (1)
Ω ∂xj ∂xi Ω ΓQ
where T denotes a scalar state variable, such as the temperature, and δT

is an admissible test function. The external volumetric flux is denoted by q
and the external surface flux by Q.
We consider a two-phase “void-solid” design problem and assume an
isotropic diffusivity tensor dij = k δij . The diffusivity is set to k(xi ) = 0
in the void phase, xi ∈ Ω(1) , and k(xi ) = k0 in the solid phase, xi ∈ Ω(2) .
Depending on the geometry description, one can re-write the weak form of
the governing equation (1) in two ways. Using an indicator function I(xi ),
with
0 ∀ xi ∈ Ω(1)
I(xi ) = , (2)
1 ∀ xi ∈ Ω(2)
we obtain:

∂δT ∂T
I k0 δij dΩ − δT I q dΩ − δT I Q dΓ = 0. (3)
Ω ∂xj ∂xi Ω ΓQ
Note the volumetric and surfaces fluxes can only be applied to the solid
domain; otherwise the analysis problem is ill-posed.
The Extended Finite Element Method 441
Alternatively, in the case the geometry of the solid-void interface is ex-

plicitly defined, we can formulate the governing equations as follows:

∂δT ∂T
k0 δij dΩ − δT q dΩ − δT Q dΓ
Ω(2) ∂xj ∂xi (2)
Ω(2) ΓQ

∂T
− δT k0 δij nj dΓ = 0, (4)
Γ (12) ∂x i
where nj denotes the normal on the phase boundaries Γ(12) , pointing from
phase 2 to phase 1. Note the latter formulation allows for modeling Neu-
mann boundary conditions and applying Dirichlet boundary conditions on
the phase boundaries Γ(12) .
For now we assume that no Dirichlet or non-homogeneous Neumann in-
terface conditions on Γ(12) exist, and we study the application of discretiza-
tion schemes to the two formulations, (3) and (4). Subsequently, we focus
in the particular on the finite element method. We consider an element
that is intersected by Γ(12) as shown in Figure 2. If we do not have explicit
information of the phase boundary, but the interface geometry is defined
implicitly via the indicator function, we need to approximate the response
within the element by standard shape functions and integrate numerically
the weak form of the governing equation over the element, evaluating at
each integration point the integrand (see Figure 2a). For the particular
case of “void-solid” in the absence of Dirichlet and non-homogenous Neu-
mann boundary conditions, standard finite element interpolation schemes
are sufficient to approximate the response as it is smooth within the solid
phase and along the phase boundary. However, it cannot be guaranteed
that standard integration schemes are sufficient to properly integrate the
governing equation, which may result in singular systems (see Figure 2b).
phase 1 phase 2 Integration point
a. b. c.
Figure 2: Numerical integration of finite element intersect by phase bound-

ary.
442 K. Maute
Even if the number of integration points is significantly increased, this is-

sue is not guaranteed to be resolved (see Figure 2c). One possible remedy
is to use “adaptive” integration schemes. However since we do not know
explicitly the geometry of the interface, devising such schemes is non-trivial.
In topology optimization, the issues above are typically resolved by
smearing out the material distribution across the element. Well-aligned
with the concept of density methods, a solid volume fraction for the element
is computed. Using material interpolation schemes, the material properties
are modeled as a function of the solid volume fractions. This approach can
be considered “natural” when using volume fraction as design variables,
as done in density methods. However, most level-set methods follow this
approach too, using the concept of Ersatz materials. For Ersatz materials
linear interpolation schemes are typically used. Thus, from a finite element
analysis point of view, there is no difference between density methods and
level-set methods using Ersatz materials. From an optimization point view,
however, the methods differ with respect to the models for evaluating the
volume fractions as functions of the design variables, among other aspects.
The reader is referred to the review article on level-set methods by van
Dijk et al. (2013) and the review article by Sigmund and Maute (2013),
comparing density and level-set methods.
We shift the focus to formulations and methods that take advantage
of explicitly defining the interface geometry via level-sets. In this case, one
option is to discretize the weak form of the governing equations (4) by a stan-
dard finite element formulation. However, this requires locally or globally
re-meshing the solid domain for each design in the course of the optimization
process. This approach is not only costly but also leads to severe robustness
issues, in particular in three dimensions. Therefore, methods operating on
fixed grids are typically preferred. As the discussion of Figure 2 suggests,
it is straight forward to use adaptive integration techniques if the interface
geometry is explicitly defined. For example, one can triangulate the solid
phase within each intersected element and perform the integration of the
weak form of the governing equation with controllable accuracy. Wang and
Wang (2006) use the super-imposed finite element method of Fish (1992)
for predicting the structural response within a level-set method. Xia et al.
(2012) follow a similar approach and refine a triangulated background mesh
for intersected elements. Note these approaches are different from standard
adaptive mesh refinement methods as they can be performed solely on an
element level.
So far the discussion was limited to “void-solid” problems where neither
non-homogeneous Neumann nor Dirichlet boundary conditions need to be
satisfied on the interface Γ(12) . While this class of problem has received
significant attention in topology optimization, from a broader engineering

perspective, the majority of design problems involves multiple non-void ma-
terial phases and requires applying boundary conditions at the interface.
The latter poses a great challenge for methods that rely on the concept of
volume fractions for modeling the physical response. In the remainder of
this chapter, we discuss an alternative approach for topology optimization
that takes full advantage of an explicit interface geometry description for
predicting the physical response and for evaluating the design sensitivities.
This approach will employ level-sets for describing the evolution of the inter-
face in the optimization process and the Extended Finite Element Method
(XFEM) for predicting the physical response. To highlight some interesting
features of the XFEM for topology optimization, we will focus on topology
optimization of flow problems.
2 Level-set method
To keep this chapter as self-contained as possible, we will first outline the
particular formulation of the level-set method, which is used in the subse-
quent discussions. Note other level-set methods can be used together with
the XFEM.
The level-set method, first developed by Osher and Sethian (1988), uses
the level-set function Φ to describe the interface within a design space; see
Figure 3a. The interface Γ(f s) , here separating the fluid and solid subdo-
mains, is implicitly defined through the zero level-set contour:
Γ(f s) = {x : Φ(x) = 0} . (5)
The value of the scalar function Φ defines whether a point xi is fluid, i.e. xi ∈
Ω(f ) , solid, i.e. xi ∈ Ω(s) , or located on the fluid-structure interface, i.e. xi ∈
Γ(f s) :
Φ(x) < 0 ∀ x ∈ Ω(f ) , (6)

Φ(x) = 0 ∀ x ∈ δΩ = Γ(f s) , (7)
(s)
Φ(x) > 0 ∀x∈Ω . (8)
The level-set optimization scheme employed in this study follows the idea
by de Ruiter and van Keulen (2004). Contrary to classical level-set opti-
mization methods that advance the level-set field via the solution of the
Hamilton-Jacobi equation, here the parameters of the discretized level-set
field are explicitly defined as functions of the optimization variables sk :
Φ = Φ (sk ). The main advantage of this approach is that the resulting
optimization problem can be solved by standard nonlinear programming
444 K. Maute
©d y
©c
f
Á(x)
z
s
s
©a ¡i
f
± = ¡i x
± = ¡i
©b
(a) Interface Γ(f s) implicitly defined through (b) Elemental interface po-
Φ(x) = 0 sition, based on the nodal
Φi -values.
Figure 3: Description of a fluid-solid interface via level-sets.
methods. However, additional regularization techniques need to be applied,

such as smoothing operators proposed by Kreissl et al. (2011) and perimeter
constraints studied by Maute et al. (2011).
In the examples presented in this chapter, the level-set field is discretized
by finite elements with the same resolution, which is used to approximate
the physical response. As Kreissl et al. (2011) have shown, smoothing the
level-set field improves the rate of convergence of the optimization process
and often mitigates spatial oscillations of the interface. Therefore each nodal
level-set value Φi is defined as a function of a set of nodal design variables
sk :
Φi = Φi (sk ) . (9)
In the numerical examples discussed below, the level-set field is smoothed by
a distance based filter following the ideas of Sigmund and Petersson (1998)
and Daoud et al. (2005):
s
j (r h − rij ) sj
Φi = ∀ j : dij ≤ rs h, (10)
i,j (r h − rij )
s
where rij is the distance between the i-th and the j-th node, rs is the relative
smoothing radius and h denotes the edge length of one element. Note, in
contrast to density methods, smoothing the level-set function is insufficient
to control locally the geometry and guarantee that the design converges as
the discretization of the level-set field is refined; Sigmund and Maute (2013)
provide further details on this issue.
For the optimization problems considered in this study, the interface
geometry is varied by manipulating the level-set field in the optimization
process. The interface geometry is only dependent on the design variables
that lie within a band of width 2rs around the interface; see (10). The design
sensitivities of variables outside of this band vanish. Therefore the design
can only grow across existing boundaries, i.e. boundaries can merge, but no
new domains of solid material completely surrounded by fluid (islands) can

develop. Likewise, it is not possible to create new holes in the solid domain.
However, the inability to create new islands can be compensated in most
cases by initializing the level-set field with a large number of disconnected
solid features. The creation of holes within the solid domain is insignificant
for fluid optimization as these do not affect the flow solution. In structural
optimization the issue of creating holes is usually handled by topological
derivatives. This technique is described, for example, in Sokolowski and
Zochowski (1999), Burger et al. (2004), de Ruiter (2005), and Amstutz and
Andrä (2006).
3 Finite Element Formulation of the Incompressible

Navier-Stokes Equations
We will briefly review the flow model used in the following examples. Here
we describe the flow by the steady-state incompressible Navier-Stokes equa-
tions, which can be written in weak form with non-dimensional variables as
follows:

∂vi 1 ∂wi ∂wj
R= wi ρ vj dΩ − + pδij dΩ
Ω ∂xj Ω 2 ∂xj ∂xi

1 ∂wi ∂wj 1 ∂vi ∂vj
+ + 2μ + dΩ (11)
Ω 2 ∂xj ∂xi 2 ∂xj ∂xi

∂vi 1 ∂vi ∂vj
+ q dΩ − wi nj −pδij + 2μ + dΓ = 0,
Ω ∂xi Γ 2 ∂xj ∂xi
where R is the residual, vi the velocity, wi a velocity test function, p the

pressure, and q a pressure test function. Note the divergence of the stress
tensor, ∂σij /∂xj , has been integrated by parts. The non-dimensional dy-
namic viscosity is defined as:
μ̂ 1
μ= = , (12)
L̂ref v̂ ρ̂ Re
where L̂ref , v̂, ρ̂ are the dimensional reference length, reference velocity, and
reference density, respectively. The Reynolds number is denoted by Re.
4 Extended Finite Element Method

The basic concept of the XFEM is to enrich the shape functions such that
discontinuities in the solution across the interface Γ(f s) can be modeled
446 K. Maute
within an intersected element. The XFEM augments the standard finite

element interpolation space by introducing additional shape functions and
their associated degrees of freedom, called “enriched degrees of freedom”.
The physical field is described by the superposition of the standard and
enriched shape functions and may include kinks or jumps in the field vari-
ables within elements. The XFEM was originally developed by Daux et al.
(2000) to simulate crack propagation in solids and has since been applied
to a broad range of problems. A general overview of the XFEM is given by
Fries and Belytschko (2010).
The XFEM is attractive for level-set topology optimization methods be-
cause of its applicability to a broad range of engineering problems and its
rigorous mathematical foundation. For the particular case of two material
phases with one of them representing “void”, additional enriched degrees
of freedom may not be needed. If in addition only homogeneous Neumann
boundary conditions are applied at the phase boundaries, the XFEM differs
from the standard finite element method only with respect to the domain
of integration. As the response does not need to be modeled in the “void”
phase, the weak form of the governing equations is only integrated over
the material phase in each element. This simplified version of XFEM was
applied to shape optimization of structures by van Miegroet et al. (2005),
Duysinx et al. (2006), and van Miegroet and Duysinx (2007). If the phys-
ical model specifies Dirichlet boundary conditions along phase boundaries,
the XFEM needs to be augmented by interface constraint methods, such
as stabilized Lagrange multiplier or penalty methods. For example, Kreissl
and Maute (2012) solved flow topology optimization problems by enforcing
the stick boundary conditions along the fluid-solid interface via a stabilized
Lagrange multiplier formulation. Subsequently, this approach will be dis-
cussed in detail. The extension of the XFEM for topology optimization of
two-phase problems is discussed in Makhija and Maute (2013), emphasizing
the need for a generalized enrichment strategy to properly resolve physical
phenomena at the phase boundaries and to mitigate numerical instabilities.
For “fluid-solid” problems, a Heaviside function is chosen to enrich the
standard shape function. This allows to setting the velocity and pressure
fields to zero in the solid phase; see Figure 4:

f (x) = Ni (x) f˜i (x) + ψ (x) f¯i (x) (13)
i
with
1 ∀ x ∈ Ω(f ) ,
ψ (x) = (14)
0 ∀ x ∈ Ω(s) ,
¡i ¡i
1 1 1 1 1 1
0.5 0.8 0.5 0.8 0.5 0.8
0.6 0.6 0.6

0 0 0
1 0.4 1 0.4 1 0.4
0.8 0.8 0.8
0.6 0.2 0.6 0.2 0.6 0.2
0.4 0.4 0.4
x x x
0.2 0.2 0.2
0 0 0
0 0 0
(a) Standard shape func- (b) Heaviside enrichment (c) Product of standard
tion, N function, ψ shape and enrichment
function, N ψ
Figure 4: Regular bilinear shape function and Heaviside enrichment func-

tion; the meshes in this figure only visualize the shape functions and are
not computational meshes.
where Ni (xi ) is the standard shape function and ψ (xi ) is the enrichment
function; f˜i denotes regular and f¯i enriched degrees of freedom. The en-
riched degrees of freedom are used to describe the flow field in intersected
elements; the location of the intersection is defined by the zero level-set.
Note enriching the approximation of the flow field by no or only one
additional set of shape functions, as suggested by (13), may be insufficient
and lead to an artificial coupling of geometrically disconnected flows, which
may cause numerical instabilities and artifacts in the optimized geometry.
To properly approximate the flows in domains separated by less than two
“void” elements, additional enrichment functions may need to be intro-
duced. The reader is referred to Makhija and Maute (2013) for details on
a generalized enrichment strategy and its application to topology optimiza-
tion.
For two-phase problems, such as “solid-void” or “fluid-void” configura-
tions, the XFEM may also suffer from ill-conditioning of the discretized
governing equations if the size of the physical domain in intersected ele-
ments approaches zero; see Figure 2c. This problem can be mitigated by
eliminating the degrees of freedom, which cause the ill-conditioning, from
the discrete system and/or by applying appropriate projections schemes.
The reader is referred to Lang et al. (2013) for further details on this issue.
5 Interface Conditions
To enforce the stick condition along the fluid-solid interface,
v (x) = 0 ∀ x ∈ Γ(f s) , (15)

448 K. Maute
we adopt a stabilized Lagrange multiplier formulation. Alternatively, penalty

and Nitsche methods can be used. Following the idea of Gerstenberger and
Wall (2010), we introduce an additional stress field as the Lagrange multi-
plier for enforcing the boundary condition along the fluid-solid interface.
Starting from the weak form of the incompressible Navier-Stokes equa-
tions (11), the residual R defined in (11) is augmented as follows:

(f s)
R̃ = R − γij nj (vi − v̂i ) dΓ − L
wi σij nj dΓ(f s)
(f s) (f s)
Γ Γ
constraint term Lagrange multiplier term

1 (f )
−k γij (σij + pδij ) − L
ij dΩ = 0, (16)
Ωf 2μ

consistency term
L
where the contraction of the assumed stress field σij and the normal nj
L
serves as a Lagrange multiplier. The test function corresponding to σij is
L
denoted by γij . The strain rate tensor, ij , associated with the assumed
L
stress field, σij , is defined as:
1 L
L
ij = σij + δij p . (17)
2μ
The first term in (16) enforces the constraint, the second one represents
the contribution of the surface traction associated with the assumed stress
field to the flow momentum equation, and the third term ensures the com-
patibility of the true and assumed strain fields. The Lagrange multiplier
degrees of freedom can be eliminated on an elemental level if element-wise
continuous shape functions for Lagrange multiplier are chosen. Thus, the
interface condition is enforced without introducing additional unknowns to
the residual equations.
L
In this study we chose constant shape function for both γij and σij .
In addition, we introduce the scaling factor k in the compatibility term in
(16). Setting k to zero, leads to a classical Lagrange multiplier formula-
tion, which would require finding an appropriate Lagrange multiplier space.
While increasing k improves the numerical stability, it leads to a less ac-
curate enforcement of the boundary condition. Due to the structure of the
compatibility term, boundary conditions are less strongly enforced as the
Reynolds number (see (12)), and the elemental fluid area increase. There-
fore, we define the scaling parameter k as follows:
Ωe − Ωef
k=η , (18)
Ωe
where Ωef defines the fluid area and Ωe the total area of one element. The
scalar η depends on the Reynolds number and the spatial discretization.
6 Example
To demonstrate the utility of the combination of level-set topology opti-
mization and the XFEM for optimizing the geometry of two-dimensional
incompressible flow problems at steady-state, we consider the following two
problems. The first example shows the ability of the proposed approach to
reproduce results found by density methods using a Brinkman penalization
approach. The second example illustrates the advantage of the proposed
XFEM approach over the material interpolation approach.
In the following examples, the weak form of the incompressible Navier-
Stokes equations (11) is discretized by four-node finite elements, i.e. the
velocity and pressure fields are approximated piecewise by bilinear, equal-
order interpolations. To avoid numerical instabilities we employ the SUPG/PSPG-
stabilization scheme introduced by Tezduyar et al. (1992). All examples are
solved by the Globally Convergent Method of Moving Asymptotes (GCMMA)
of Svanberg (1995). The design sensitivities are computed by an adjoint
method. Both examples are described in detail in Kreissl and Maute (2012).
6.1 Pipe-bend
The pipe-bend problem was originally introduced by Borrvall and Pe-
tersson (2003) and is a standard example for fluid topology optimization. It
was also studied by Duan et al. (2008) and Pingen et al. (2007), among oth-
ers. The design domain as well as the initial design are depicted in Figures
5a and 5b. The objective is to minimize the difference in total pressure,
2
ptot = 1/2ρ v + p, between the inlet and outlet, subject to a constraint
that allows only 25% of the area of the design domain to be fluid:
in − pout ),
min z = (ptot tot
sk
Af luid (19)
s.t. g = 0.25 − ≥ 0,
L2
smin ≤ sk ≤ smax .
The total inlet and outlet pressures, ptot tot
in and pout , are averaged over the
inlet and outlet ports. The boundary conditions, depicted in Figure 5a, are
a parabolic inlet velocity, vin , and a constant static outlet pressure, pout .
A stick condition is enforced at the domain boundaries. The initial design
consists of an all-fluid design domain with four solid circles, as shown in
Figure 5b.
450 K. Maute
5L
L
vin
design
domain 5L
3.5L
3.5L L
pout
(a) Design domain (b) Initialization
Figure 5: Pipe-bend example.
©=0 ©=0
(a) Initial (b) Optimized
Figure 6: Level-set distributions for pipe-bend.
In order to guarantee proper flow through the inlet and outlet ports,
the level-set values coinciding with the nodes on the inlet and outlet are
fixed at Φi = −1 and are omitted from the smoothing operation defined
in (10). Discretizing the design domain by 40 × 40 elements, the level-set
distributions of the initial and optimized designs are depicted in Figure 6.
The optimized geometry is shown in Fig. 7a.
For comparison we optimize the problem (19) using a density approach
for two different levels of mesh refinement: 40 × 40 and 80 × 80 elements.
Here, the material distribution defines the penalty coefficient of the Brinkman
model. Following the work by Borrvall and Petersson (2003), the penalty
coefficient is interpolated via a convex function (see chapter on flow topology
optimization).
Figures 7b and 7c show the zero level-set of the optimized XFEM ge-
ometry on top of the optimized porosity distribution. Both designs agree
well with the one obtained from the XFEM based optimization. Figure 7b
illustrates that for the same mesh refinement the proposed XFEM approach
has higher boundary resolution compared to the porosity approach.
0 s 1 0 s 1
(a) XFEM (40 × 40 mesh) (b) Zero level-set contour (c) Zero level-set contour
on 40 × 40 porosity distri- on 80 × 80 porosity distri-
bution bution
Figure 7: Optimized pipe-bend geometries.
The figure shown at the beginning of this chapter shows the initial, an
intermediate, and the final result of a topology optimization process that
minimizes the pressure drop of a diffuser via a level-set approach. As in the
pipe-bend example above, the flow is predicted by the XFEM. The diffuser
problem was originally introduced by Borrvall and Petersson (2003) and
studied by a density method. Challis and Guest (2009) and Pingen et al.
(2007) solved the problem by level-set methods.
6.2 Nozzle with target total pressure ratio

The XFEM topology optimization approach does not rely on a material
interpolation scheme and therefore does not require a tailored combination
of objective and constraint to converge to a “0-1” material distribution.
This feature is illustrated with the example depicted in Fig. 8a. Again, we
are comparing the results of the XFEM and the porosity approach.
The goal of this optimization problem is to find a design that results in
a specified total pressure ratio between inlet and outlet:
tot 2
in − 4pout ) ,
min z = (ptot
sk
(20)
s.t. smin ≤ sk ≤ smax ,
i.e. we want to find a design for which the total pressure at the inlet is four
times the total pressure at the outlet, subject to box constraints on the
design variables.
The initial design for the XFEM case is shown in Figure 8b, the porosity
distribution is initialized as 100% fluid; both cases are discretized by 60 × 30
elements. The XFEM based optimization problem converges to the design
452 K. Maute
L
vin design
L L pout
domain
L
6L
(a) Design domain (b) Initialization (for XFEM)
Figure 8: Nozzle example.
0.787
(a) XFEM (b) Porosity approach
Figure 9: Optimized nozzle geometries.
depicted in Fig. 9a. This geometry results in an objective of z = 7.95×10−4 .

Solving the optimization problem with the porosity approach leads to the
porosity distribution in Fig. 9b which is of little value for identifying a
reasonable geometry.
While the XFEM approach leads to a somewhat intuitive design that
generates the desired pressure drop, we point out that the solution of the
optimization problem (20) strongly depends on the initial level-set field. For
example, initializing the level-set fields with small circular solid areas at the
bottom and top corner either at the inlet (case A) or the outlet (case B),
as shown in Fig. 10a, leads to the designs depicted in Fig. 10b. In case A
the objective is z = 1.48 × 10−3 and in case B z = 2.37 × 10−3 , thus both
designs match the desired total pressure ratio well.
The XFEM is a promising approach for analyzing designs within level-set
topology optimization approaches. In contrast to Ersatz material tech-
niques, the XFEM preserves the crispness of the geometry description pro-
vided by level-sets. Furthermore, it allows enforcing Dirichlet and non-
Case A Case B Case A Case B
(a) Initial designs (b) Optimized geometries
Figure 10: Influence of initial design.
homogeneous boundary conditions by standard approaches, such as stabi-

lized Lagrange multiplier and penalty methods.
However, a generalized enrichment strategy is needed to robustly and
accurately analyze arbitrary geometries and to prevent the formation of
numerical artifacts. Furthermore, additional techniques, such eliminating
degrees of freedom and projection methods, are needed to mitigate ill-
conditioning issues of the discretized governing equations.
The XFEM has not widely been integrated into topology optimization
problems methods. XFEM is not available yet as a general analysis approach
in most commercial finite element solvers, but is only offered for special
analysis problems, such as crack propagation. Furthermore, the extensions
of the standard XFEM, such as a generalized enrichment strategy, are still
in its infancy.
From a broader perspective, the XFEM shares common features with
local re-meshing schemes. As local or even global re-meshing algorithms
become more robust and efficient, it remains to be seen whether XFEM
or re-meshing will become more popular, or Ersatz material methods will
prevail.
8 Acknowledgment
under grants CMMI–1201207 and EFRI–1038305. The opinions and conclu-
sions presented in this paper are those of the authors and do not necessarily
reflect the views of the sponsoring organization.
Bibliography
S. Amstutz and H. Andrä. A new algorithm for topology optimization using
a level-set method. Journal of Computational Physics, 216(2):573–588,
454 K. Maute
2006.
T. Borrvall and J. Petersson. Topology optimization of fluids in Stokes
flow. International Journal for Numerical Methods in Fluids, 41(1):77–
107, 2003.
M. Burger, B. Hackl, and W. Ring. Incorporating topological derivatives
into level set methods. Journal of Computational Physics, 194(1):344–
362, 2004.
V. Challis and J. Guest. Level set topology optimization of fluids in stokes
flow. International Journal for Numerical Methods in Engineering, 79
(10):1284–1308, 2009.
F. Daoud, M. Firl, and K.-U. Bletzinger. Filter techniques in shape opti-
mization with cad-free parametrization. In Proceedings of the 6th World
Congresses of Structural and Multidisciplinary Optimization, May 30-
June 3, Rio de Janeiro, Brazil, 2005.
C. Daux, N. Moes, J. Dolbow, N. Sukumark, and T. Belytschko. Arbitrary
branched and intersecting cracks with the extended nite element method.
Int. J. Numer. Meth. Engng, 48:1741–1760, 2000.
M. J. de Ruiter. Topology Optimization Using a Topology Description Func-
tion Approach. PhD thesis, Technische Universiteit Delft, 2005.
M. J. de Ruiter and F. van Keulen. Topology optimization using a topology
description function. Structural and Multidisciplinary Optimization, 26
(6):406–416, 2004.
X. Duan, Y. Ma, and R. Zhang. Optimal shape control of fluid flow using
variational level set method. Physics Letters A, 372(9):1374—1379, 2008.
P. Duysinx, L. Miegroet, T. Jacobs, and C. Fleury. Generalized shape opti-
mization using x-fem and level set methods. In IUTAM Symposium on
Topological Design Optimization of Structures, Machines and Materials,
pages 23–32. Springer, 2006.
J. Fish. The s-version of the finite element method. Computers & Structures,
43(3):539 – 547, 1992.
T. Fries and T. Belytschko. The extended/generalized finite element
method: an overview of the method and its applications. International
Journal for Numerical Methods in Engineering, 84(3):253–304, 2010.
A. Gerstenberger and W. A. Wall. An embedded dirichlet formulation for 3d
continua. International Journal for Numerical Methods in Engineering,
82(5):537–563, 2010.
S. Kreissl, G. Pingen, and K. Maute. An explicit level set approach for gen-
eralized shape optimization of fluids with the lattice boltzmann method.
International Journal for Numerical Methods in Fluids, 65(5):496–519,
2011.
S. Kreissl and K. Maute. Levelset based fluid topology optimization using

the extended finite element method. Structural and Multidisciplinary
Optimization, pages 1–16, 2012.
C. Lang, D. Makhija, A. Doostan, and K. Maute. A simple and efficient
preconditioning scheme for xfem with heaviside enrichments. Computa-
tional Mechancis, 2013. to be submitted.
D. Makhija and K. Maute. Numerical instabilities in level set topology
optimization with the extended finite element method. Structural and
Multidisciplinary Optimization, 2013. submitted.
K. Maute, S. Kreissl, D. Makhija, and R. Yang. Topology optimization of
heat conduction in nano-composites. In 9th World Congress on Struc-
tural and Multidisciplinary Optimization, Shizuoka, Japan, 2011.
S. J. Osher and J. A. Sethian. Fronts propagating with curvature dependent
speed: algorithms based on Hamilton-Jacobi formulations. Journal of
Computational Physics, 79:12—49, 1988.
G. Pingen, A. Evgrafov, and K. Maute. Topology optimization of flow
domains using the lattice boltzmann method. Structural and Multidisci-
plinary Optimization, 36(6):507–524, 2007.
O. Sigmund and K. Maute. Topology optimization approaches: A com-
parative review. Structural and Multidisciplinary Optimization, 2013.
submitted.
O. Sigmund and J. Petersson. Numerical instabilities in topology opti-
dependencies and local minima. Structural and Multidisciplinary Opti-
mization, 16(1):168–75, 1998.
J. Sokolowski and A. Zochowski. Topological derivatives for elliptic prob-
lems. Inverse problems, 15:123, 1999.
K. Svanberg. A globally convergent version of MMA without linesearch.
In Proceedings of the First World Congress of Structural and Multidis-
ciplinary Optimization, 28 May - 2 June 1995, pages 9–16, Goslar, Ger-
many, 1995.
T. E. Tezduyar, S. Mittal, S. E. Ray, and R. Shih. Incompressible flow com-
putations with stabilized bilinear and linear equal-order-interpolation
velocity-pressure elements. Computer Methods in Applied Mechanics and
Engineering, 95:221–242, 1992.
N. van Dijk, K. Maute, M. Langelaar, and F. van Keulen. Level-set methods
for structural topology optimization: A review. Structural and Multidis-
ciplinary Optimization, accepted for publication, 2013.
L. van Miegroet, N. Moës, C. Fleury, and P. Duysinx. Generalized shape
optimization based on the level set method. In 6th World Congress of
Structural and Multidisciplinary Optimization, 2005.
456 K. Maute
L. van Miegroet and P. Duysinx. Stress concentration minimization of 2d

filets using x-fem and level set description. Structural and Multidisci-
plinary Optimization, 33(4-5):425–438, April 2007.
S. Wang and M. Y. Wang. A moving superimposed finite element method
for structural topology optimization. Int. J. Numer. Meth. Engng, 65:
18921922, 2006.
Q. Xia, T. Shi, S. Liu, and M. Y. Wang. A level set solution to the stress-
based structural shape and topology optimization. Computers & Struc-
tures, 9091:55 – 64, 2012.
Topology Optimization under Uncertainty
*
Kurt Maute
*
Abstract. Considering stochastic variations in material parame-

ters, geometry, and boundary conditions is for the majority of en-
gineering design problems of pivotal importance, in order to obtain
robust and reliable designs. While design optimization under un-
certainty has matured for sizing and shape optimization over the
past two decades, accounting for stochastic variations in topology
optimization is still in its infancy. This lecture will introduce basic
approaches to include uncertainty models and predictions into the
topology optimization process.
1 Introduction
The general goal of engineering design is maximizing the utility of a sys-
tem or a device while simultaneously minimizing its life-cycle costs. The
latter includes the costs for developing, manufacturing, and maintaining
the system. This task is significantly complicated by the inherently non-
deterministic nature of the system and the conditions under which it oper-
ates. Since in general simultaneously maximizing the utility and minimizing
the life-cycle costs lead to a conflict, a compromise needs to be found which
balances the utility and cost over the life-time of the system. The reader
is referred to the textbook by Hazelrigg (1996) for an introduction into the
fundamentals of engineering design.

DOI 10.1007/978-3-7091-1643-2_20, © CISM, Udine 2014
458 K. Maute
A design, which meets the goals defined above, is often labeled “robust”
and/or “reliable”. These terms have different flavors in the context of en-
gineering design. On a conceptual level, the robustness or reliability of a
system can be improved through identifying, understanding, and, if pos-
sible, eliminating basic failure mechanisms. Focusing on uncertainty, the
reliability can be associated with the probability that a failure occurs, and
optimizing for robustness can be understood as minimizing the sensitivity
of the utility and costs with respect to random variations. Identifying and
understanding failure mechanisms is at the core of traditional engineering
disciplines, such as structural and fluid mechanics. Optimizing for minimum
failure probability or maximizing the robustness of non-deterministic sys-
tems is an increasingly important field in design optimization and discussed
in this chapter in the context of topology optimization.
In recent decades, numerous optimization methods have been developed
to address the above design challenge. At the beginning of this development,
the non-deterministic, uncertain nature of engineering design problems was
often ignored and a perfect system under clearly predictable operation con-
ditions assumed. However, designs found by deterministic approaches are of-
ten sensitive to variations of system and operating parameters, and therefore
of limited value in practice. To mitigate this issue, safety factors are tradi-
tionally introduced into the formulation of the design optimization problem,
often leading to unknowingly unsafe or overly conservative designs.
Today, it is widely recognized that optimization methodologies should
account for the stochastic nature of engineering systems. These methodolo-
gies can be classified into two groups: robust design optimization (RDO)
and reliability-based design optimization (RBDO). Based on mainly deter-
ministic models, RDO methods attempt to simultaneously maximize the
deterministic performance and to minimize the sensitivity of the perfor-
mance with respect to random parameters. This approach leads to a multi-
objective optimization problem, often capturing the impact of uncertainties
only in a qualitative sense. Unlike RDO approaches, RBDO methods aim
at designing for a specific risk and target reliability level, accounting for
various sources of uncertainty in a quantitative sense. Furthermore, RBDO
approaches are based on stochastic analysis methods and, therefore, from
a theoretical and an algorithmic perspective, more challenging and com-
putationally more expensive than deterministic approaches. The reader is
referred to the books by Ang and Tang (1984), Melchers (1985), Kleiber and
Hien (1992), Ghanem and Spanos (1991), and Nikolaidis et al. (2005) for a
detailed introduction into stochastic modeling and analysis techniques. For
an overview of RBDO and RDO methods the following review papers are
recommended: Frangopol and Maute (2003), Frangopol et al. (2007), Beyer
Topology Optimization Under Uncertainty 459
and Sendhoff (2007), and Valdebenito and Schuëller (2010).

While RDO and RBDO approaches have been frequently studied in sys-
tems design as well as for sizing and shape optimization, incorporating un-
certainty into topology optimization is still in its infancy, in particular in
regards to RBDO. The main challenges in applying RDO and RBDO to
topology optimization are (a) the efficient modeling of uncertainty for a
body whose geometry only evolves during the design process, and (b) the
solution of optimization problems with a large number of design variables
and uncertain parameters at acceptable costs. Both challenges are closely
interconnected as the computational cost typically increases with the num-
ber of uncertain parameters used to parametrize the stochastic model. This
chapter will give an overview of stochastic modeling approaches and their
integration into design optimization methods, focusing on topology opti-
mization.
2 Stochastic Modeling
A stochastic model describes the uncertainty in a system, which can be
either aleatory or epistemic. The first refers to inherent, irreducible un-
certainty, such as random variations in geometry, material properties, and
operating conditions. Epistemic uncertainty is due to a lack of knowledge
about the system, which can be reduced as more information becomes avail-
able. Depending on the type of uncertainty and the amount of information,
different modeling approaches are used, such as interval methods, fuzzy
theory, Dempster-Shafer theory, and the probability theory.
In the context of topology optimization, interval and probabilistic meth-
ods are likely the most relevant approaches and therefore we will focus on
these modeling methods. Interval methods simply require knowledge about
the upper and lower bounds of the stochastic variations of a quantity of
interest. The worst and best case scenarios for given intervals are computed
to characterize the impact of uncertainty on the system. Typically interval
arithmetic or optimization techniques are used for this purpose. Proba-
bilistic methods require significantly more information about the system in
order to construct probability distributions, such as uniform and Gaussian
distributions that describe the randomness in fields and parameters. The
complexity of developing accurate probabilistic models is often underesti-
mated, in particular the need to derive such models from sufficiently rich
data sets, i.e. observations and measurements. If there is not sufficient data
to accurately construct a probabilistic model, a stochastic analysis based
on inaccurate models might lead to coarsely incorrect results. As stochas-
tic analysis methods for probabilistic models are often significantly cheaper
460 K. Maute
than interval methods with respect to their computational costs, interval

models are frequently replaced with “made-up” probabilistic models, using
for example uniform distributions. In general, such stochastic analysis re-
sults are meaningless as they neither represent a worst/best case scenario
nor characterize correctly the probabilistic behavior of the system.
Subsequently, we will briefly discuss the basic elements of probabilistic
modeling and analysis methods in the context of RBDO and RDO. This
discussion will provide insight into the applicability of these methods to
topology optimization.
2.1 Formulation of RBDO and RDO Optimization Problems

A generic reliability-based optimization problem can be formulated as
follows, with si being the optimization variables and rj being the random
parameters:
min P (c(si , rj ) ≥ c̄t ) (1)

si
subject to : P̄m − P (fm (si , rj ) < 0) ≥ 0 (2)

gnD (si ) ≥ 0 (3)
slower
i ≤ si ≤ supper
i . (4)
The objective is to minimize the probability of the cost function c(si , rj )

being larger than a target value c̄t . One set of constraints limits the fail-
ure probability associated with the limit state functions fm (si , rj ) defining
failure as fm < 0. P̄m is the maximum acceptable failure probability. A
second set of constraints, gnD , contains all deterministic constraints.
A robust design optimization problem is typically formulated as a multi-
objective problem. While more sophisticated multi-objective formulations
can be used, for the sake of simplicity, here we write the RDO problem by
mapping the individual components into one scalar objective via a weighted-
sum approach:
⎛ ⎞

min ⎝w0 c(si , rj0 ) + wj Δc(si , rj0 + Δrj )⎠ (5)
si
j
subject to : gjD (si ) ≥ 0 (6)
slower
i ≤ si ≤ supper
i , (7)
where rj0 represents the mean or nominal configuration and Δc a pertur-

bation of the cost function due to a variation of the random parameter
Δrj . Often the perturbation Δc is approximated by a first-order Taylor

expansion:
∂c
Δc = Δrj , (8)
∂rj
where the perturbation size Δrj depends on the variability of a random
parameter, such as its standard deviation.
In general, the state equations also depend on the random parameters:
Rl (si , rj , uk (si , rj )) = 0, (9)
where uk denotes the state variables. The state equations (9) can be either
added to the set of constraints of the optimization problems (1) and (5)
or explicitly satisfied when evaluating the objective and constraints. In
the following presentation, the stochastic state equations (9) are treated
separately as their mathematical properties and meaning differ from other
design constraints. The stochastic response and the design criteria need to
be evaluated at each iteration of the design optimization process. Although
this approach allows calling the stochastic analysis method in a black-box
fashion, great care and insight is needed to select the appropriate reliability
analysis method, depending on the formulation of the optimization problem
and the type of stochastic design criteria used.
2.2 Probabilistic Measures

In the case that the uncertainty of a system can be modeled by a prob-
abilistic model, the stochastic response of a system can be characterized by
stochastic moments and probabilities of particular events occurring. The
first and second moments, which are the mean and the standard deviation,
are often used to formulate engineering design problems. The nth moment
of the cost function c is:

E [cn ] = cn (si , rj , uk (si , rj )) h(rj )dΩr , (10)
Ωr
where h(rj ) denotes the joint probability density function, Ωr the Nr -

dimensional space spanned by the random parameters, and E[·] the ex-
pectation operator.
Of particular interest in RBDO is the probability of failure Pf , which is
defined by the occurrence of the event f (si , rj , uk ) < 0. A negative value of
the limit state function, f < 0, defines the failure region. The probability
of failure Pf can be computed by evaluating the following Nr -dimensional
convolution integral over the failure domain:

Pf = h(rj )dΩr . (11)
f <0
462 K. Maute
Only in special cases, this convolution integral has a closed-form solution, for
example if the random variables rj are normal (Gaussian) and the limit state
function f is linear in the random variables. In this case, the probability of
failure Pf is:
Pf = Φ(−β), (12)
where Φ is the cumulative distribution function and β the reliability in-

dex, which is the minimum distance of the iso-contour surface of the limit
state function for f = 0 to the origin of the space spanned by the random
variables.
The fundamental reliability analysis problem introduced above involves
a single failure mode of a single component. However, the reliability of
a system can only be correctly assessed by considering all failure modes
and their interactions. In addition, loads and operating conditions and
system properties often vary in time. For example, the strength of materials
may degrade due to chemical reactions. Therefore, reliability problems are
typically time-variant. While of pivotal importance for designing reliable
and robust systems, in the following discussions we ignore time-variant and
system aspects and focus on the analysis and optimization of time-invariant
and single failure mode problems.
2.3 Probabilistic Analysis Methods

Computing stochastic moments requires integration the function of in-
terest, such as the cost function, over the Nr dimensional space of random
parameters. Determining the probability of occurrence of a single failure
mode requires the evaluation of the convolution integrals (11) in some form
or the other. In general, both types of integrations need to be performed
numerically and typically require a large number of evaluations of the cost
function or the limit state function, which are in general nonlinear implicit
functions of the random variables. For computing the expectations in (10),
sampling, quadrature and other integration schemes are typically used. Ba-
sic reliability analysis methods can be divided into sampling approaches
(e.g. Monte Carlo sampling), approximate reliability analysis methods (e.g.
first and second order reliability methods), and stochastic perturbation and
projection schemes for partial differential equations (e.g. the stochastic
finite element and the spectral stochastic finite element method). These
methods can be either directly applied to the model predicting the limit
state g(si , rj , uk ) for a specific realization of the random variables rj or to
a meta-model approximating the limit state. Meta-models, which include
response surface and reduced order models, are frequently used in combi-
nation with sampling methods when the numerical simulation models are
large, the number of random parameters is small, and direct sampling lead
to unacceptable computations costs.
In general, a compromise between accuracy and numerical efficiency
needs to be found. In addition, if the design optimization procedure is
driven by a gradient-based optimization algorithm, probabilistic analysis
methods need to be chosen that allow for evaluating efficiently the design
sensitivities of probabilistic design criteria. For example, first and second
order reliability methods and, more recently, stochastic projections schemes
are frequently used within design optimization schemes. The reader is re-
ferred to the studies of Hohenbichler and Rackwitz (1986), Chiralaksanakul
and Mahadevan (2005), Maute et al. (2009), Zhao et al. (2011), and Tootk-
aboni et al. (2012).
3 Applications to Topology Optimization

Methods for accounting for uncertainty in topology optimization can be
grouped into approaches that consider uncertainty in design dependent
and design independent fields and parameters. In the latter case, generic
Optimization Under Uncertainty (OUU) methods can be integrated into
standard topology optimization methods without substantial modification,
as long as these methods allow for an efficient sensitivity analysis of the
stochastic moments or probabilities of failure. Design independent uncer-
tainty has been studied, for example, by Ben-Tal and Nemirovski (2002),
Sandgren and Cameron (2002), Maute and Frangopol (2003), Jung and Cho
(2004), Kogiso et al. (2008), Lógó et al. (2009), Asadpoure et al. (2011),
and Tootkaboni et al. (2012). Note the majority of these studies focus on
topology optimization of truss structures using a ground structure approach.
Considering uncertainty in design dependent quantities, in particular the ge-
ometry, is significantly more challenging and still in its infancy. The reader
is referred to the work by Sigmund (2009), Chen et al. (2010), Schevenels
et al. (2011), Wang et al. (2011), and Lazarov et al. (2012). In the following
we discuss briefly one example for each case.
3.1 Structural Topology Optimization with Uncertain Loads

The example shown in Figure 1 is taken from Rozvany and Maute (2011)
and serves as a benchmark problem for topology optimization with design
independent loads. At the upper center of a rectangular design space Ω an
uncertain load H is acting in horizontal direction. The bottom of the design
space is clamped. The objective is to minimize the mass of the structure
such that the probability of the strain energy Π exceeding a given threshold
464 K. Maute
Figure 1: Structural topology optimization problem under uncertain load-

ing.
P¯i is below a desired failure probability P̄ . Using a density approach this

RBDO problem reads:

min ρdΩ (13)
ρ
subject to : P̄ − P (Π > P¯i) ≥ 0 (14)

ρlower
i ≤ ρi ≤ 1, (15)
where ρ denotes the non-dimensional density. We interpolate the material

properties via a standard Solid Isotropic Material with Penalization (SIMP)
method, introduced by Bendsøe (1989) and Zhou and Rozvany (1991), with
an exponent of three.
We assume that observations of the load H are such that the load can
be model by a Gaussian distribution with zero mean, i.e. μH = 0 and a
standard deviation of σH :
H(r) = μH + σH r, (16)
where the uncertain variable r has a standard normal distribution, i.e. zero
mean and a standard deviation of one.
The strain energy can be expressed as an explicit function of the load:
1
Π(r) = F H 2, (17)
2
where F is the coefficient of the inverse of the stiffness matrix associated
with the degree of freedom at which the load is applied. The limit state
function can be written for a given design as:
1
f (r) = Π̄ − F H 2. (18)
2
2 3 4
(a) PF 10 (b) PF 10 (c) PF 10
5 6 7
(a) PF 10 (b) PF 10 (c) PF 10
Figure 2: Optimum solutions for different reliability requirements.
The most probable point (MPP) of failure is the point in the standard
normal space at which the limit state function vanishes:

1
f (r) = 0 → rM P P = (2/F )Π̄. (19)
σH
The associated probability of failure can then be easily computed by:
∞ −rM P P
Pf = h(r)dr = h(r)dr + h(r)dr = 2Φ (−rM P P ) . (20)
f <0 rM P P −∞
We need to account for the fact that the failure region is both r > rM P P
and r < −rM P P . As the probability of failure needs to be less than P̄ we
can re-write the probabilistic constraint as follows:

−1 P̄
Pf ≤ P̄ → rM P P ≥ −Φ , (21)
2
where Φ−1 is the inverse cumulative distribution function. Note Φx < 0∀0 <
x < 1.
As we have solved the reliability problem analytically, we can convert
the RBDO problem into a standard deterministic topology optimization
problem as follows:

min ρdΩ (22)
ρ

1 P̄
subject to : (2/F )Π̄ + Φ−1 ≥0 (23)
σH 2
ρlower
i ≤ ρi ≤ 1 (24)
This problem can be solved in the same way as standard deterministic prob-
lems with a mass objective and a constraint on maximum strain energy.
9
466 K. Maute
Figure 2 shows solutions for the above problem for different reliability re-
quirements, i.e. maximum feasible failure probabilities. The results were
obtained with a SIMP model using nodal densities as design variables. As
the reliability requirements increase the angle between the bars and the
thickness of the bars increase. Rozvany and Maute (2011) show that these
results are in excellent agreement with analytical results obtained for a two-
bar truss problem.
In general, the reliability problem cannot be solved analytical and nu-
merical reliability analysis methods need to be used. However, as shown,
for example, in Maute and Frangopol (2003) for a First Order Reliability
Method (FORM), the RBDO method can be integrated into the topology
optimization procedure without the need for modifying the geometry de-
scription or the overall optimization procedure of the optimization method.
3.2 Structural Topology Optimization with Uncertain Geometry

For sizing and shape optimization under uncertainty, the challenges of
finding and discretizing a stochastic model, which describes random shape
variations, are quite similar to the ones faced in a standard stochastic anal-
ysis. In topology optimization, however, modeling uncertainty in the geom-
etry poses a significant challenge as the geometry only evolves in the course
of the optimization process. As the references provided at the beginning
of this section show, so far there are only a few topology optimization ap-
proaches that account for uncertainty in the geometry. Note, when using
a ground structure approach for truss topology optimization, considering
uncertainty in the position of the nodes leads to some form of shape uncer-
tainty. This form of uncertainty can be model by standard techniques and
does not lead to the challenges discussed in this subsection.
As level-set methods provide an explicit description of the boundaries,
considering stochastic shape variations does not present a conceptual prob-
lem. Chen and Chen (2011) have presented a RDO for topology optimiza-
tion of structures using level-sets. The random geometry variations are
modeled through a stochastic velocity field in the Hamilton-Jacobi equa-
tions. The stochastic moments of the design criteria are evaluated using an
efficient quadrature rule. The shape sensitivities for the uncertain geome-
try are computed by solving the analytically derived adjoint equations. The
latter requires the computation of an additional field that tracks the mo-
tion of material points when perturbing the level-set function. This method
presents an elegant and innovative approach for integrating geometric un-
certainty into topology optimization.
The results of a simpler, less sophisticated approach are shown in Figure
a. b.
2/3 L
L
c. d.
Figure 3: Level-set topology optimization with uncertain geometry.
3. The objective is to minimize the compliance of structure while constrain-

ing the maximum volume occupied by material. Here the level-set field is
perturbed uniformly, and the objective and constraints are averaged over a
set of perturbations, assuming a uniform distribution. Here a parametric
level-set method is employed and the nodal level-set values are defined as
functions of the optimization variables. A linear filter is used to smooth the
level-set field in order to improve the convergence rate of the optimization
process. The structural response is predicted by a generalized version of
the extended finite element method (XFEM). The reader is referred to the
chapter on the XFEM for details and references. The optimization prob-
lem is solved by the Globally Convergent Method of Moving Asymptotes
by Svanberg Svanberg (2002). The initial geometry is shown in Figure 3b.
Neglecting any random shape variations leads to the geometry in Figure 3c.
Assuming a uniform shape variation of ΔX = ±2 % relative to the length
of the domain, L, results in a design with less but bulkier struts, shown in
Figure 3d.
To account for stochastic shape variations, density methods face an ad-
ditional challenge as they lack an explicit definition of boundaries. Simply
varying the density field in a traditional density approach will lead to geom-
etry variations, but these variations are arbitrary and, in general, cannot be
associated with a proper stochastic model. Sigmund (2009) and Schevenels
et al. (2011) presented a series of filtering and density projection schemes
that create a particular form of density distributions that allow varying the
geometry in a more meaningful manner. As shown in Figure 4 the main
468 K. Maute
U non-filtered/projected U filtered/projected
1 1
0 0
x x
shape variation
Figure 4: Projection of density distribution for imposing stochastic shape

variations.
concept is to create, through filtering and projection, a density distribution

that resembles a smoothed Heaviside function. The geometry is altered by
manipulating filter parameters. Interestingly, the density distributions ob-
tained this way show great similarities with the ones generated by level-set
methods, when mapping the level-set function into a material distribution.
The approach has been used along with sampling methods and in the spirit
of a stochastic interval model, examining the influence of uniform expansion
and contraction of the geometry. Following a robust design optimization ap-
proach and using a multi-objective formulation, the density distribution is
optimized for the worst case scenario which is selected form the nominal de-
sign, the maximum expansion and maximum contraction. The same overall
concept was applied to more refined stochastic models by Wang et al. (2011)
and Lazarov et al. (2012). Similar to the results shown in Figure 3, account-
ing for uncertainty leads to bulkier structural members and increases the
minimum member size.
The ability to account for uncertainty and imperfections and to increase the
robustness of the design is an important feature for any design optimization
tool. Integrating topology optimization into RDO and RBDO frameworks
is straight forward as long as the random fields and parameters are de-
sign independent. In this case and if the optimization process is driven
by gradients, RDO and RBDO approaches should be preferred which allow
computing the design sensitivities efficiently by adjoint methods.
Accounting for uncertainty in the geometry of continuous structures is
more challenging, in particular for density methods. Today’s topology meth-
ods that account for geometric uncertainty are promising. However, it is not
clear yet whether the stochastic models these methods are based on accu-
rately capture the shape variability. Reliability-based topology optimization
with geometric uncertainty is a large unexplored field.
5 Acknowledgment
under grants CMMI–1201207 and CMMI–1235532. The opinions and con-
Bibliography
A.-S. Ang and W. Tang. Probability Concepts in Engineering Planning and
Design, volume 2. John Wiley & Sons, New York, 1984.
A. Asadpoure, M. Tootkaboni, and J. K. Guest. Robust topology opti-
mization of structures with uncertainties in stiffness–application to truss
structures. Computers & Structures, 89(11):1131–1141, 2011.
A. Ben-Tal and A. Nemirovski. Robust optimization–methodology and ap-
plications. Mathematical Programming, 92(3):453–480, 2002.
M. Bendsøe. Optimal shape design as a material distribution problem.
Structural and Multidisciplinary Optimization, 1(4):193–202, 1989.
H.-G. Beyer and B. Sendhoff. Robust optimization–a comprehensive survey.
Computer methods in applied mechanics and engineering, 196(33):3190–
3218, 2007.
S. Chen and W. Chen. A new level-set based approach to shape and topol-
ogy optimization under geometric uncertainty. Structural and Multidis-
ciplinary Optimization, 44(1):1–18, 2011.
S. Chen, W. Chen, and S. Lee. Level set based robust shape and topology
optimization under random field uncertainties. Structural and Multidis-
ciplinary Optimization, 41(4):507–524, 2010.
A. Chiralaksanakul and S. Mahadevan. First-order approximation methods
in reliability-based design optimization. Journal of Mechanical Design,
127:851, 2005.
D. M. Frangopol and K. Maute. Life-cycle reliability-based optimization
of civil and aerospace structures. Computers & Structures, 81:397–410,
2003.
D. M. Frangopol, M. Liu, and K. Maute. Optimization of Structural and
Mechanical Systems, chapter 11, pages 291–329. World Scientific Pub-
lishing, 2007. Desing optimization with Uncertainty, Life-Cycle Perfor-
mance and Cost Considerations.
470 K. Maute
R. G. Ghanem and P. D. Spanos. Stochastic Finite Elements, A Spectral

Approach. Dover Publications, Inc, revised edition, 1991. ISBN 0-486-
42818-4.
G. A. Hazelrigg. Systems engineering: an approach to information-based
design. Prentice Hall Upper Saddle River, 1996.
M. Hohenbichler and R. Rackwitz. Sensitivities and importance measures
in structural reliability. Civil Engineering Systems, 3:203–209, 1986.
H.-S. Jung and S. Cho. Reliability-based topology optimization of geometri-
cally nonlinear structures with loading and material uncertainties. Finite
Elements in Analysis and Design, 41(3):311–331, 2004.
M. Kleiber and T. Hien. The Stochastic Finite Element Method, Basic
Perturbation Technique and Computer Implementation. Wiley, 1992.
N. Kogiso, W. Ahn, S. Nishiwaki, K. Izui, and M. Yoshimura. Robust
topology optimization for compliant mechanisms considering uncertainty
of applied loads. Journal of Advanced Mechanical Design, Systems, and
Manufacturing, 2(1):96–107, 2008.
B. S. Lazarov, M. Schevenels, and O. Sigmund. Topology optimization
with geometric uncertainties by perturbation techniques. International
Journal for Numerical Methods in Engineering, 90(11):1321–1336, 2012.
J. Lógó, M. Ghaemi, and M. M. Rad. Optimal topologies in case of proba-
bilistic loading: the influence of load correlation. Mechanics based design
of structures and machines, 37(3):327–348, 2009.
K. Maute and D. M. Frangopol. Reliability-based design of mems mecha-
nisms by topology optimization. Computers & Structures, 81:813–824,
2003.
K. Maute, G. Weickum, and M. Eldred. A reduced-order stochastic finite
element approach for design optimization under uncertainty. Structural
Safety, 31(6):450–459, 2009.
R. E. Melchers. Structural Reliability Analysis and Prediction, 2nd Edition.
Wiley, 2nd edition, 1985. ISBN 978-0-471-98771-0.
E. Nikolaidis, D. M. Ghiocel, and S. Singhal. Engineering design reliability
handbook. CRC Press, 2005.
G. I. Rozvany and K. Maute. Analytical and numerical solutions for a
reliability-based benchmark example. Structural and Multidisciplinary
Optimization, 43(6):745–753, 2011.
E. Sandgren and T. Cameron. Robust design optimization of structures
through consideration of variation. Computers & structures, 80(20):
1605–1613, 2002.
M. Schevenels, B. S. Lazarov, and O. Sigmund. Robust topology opti-
mization accounting for spatially varying manufacturing errors. Com-
puter Methods in Applied Mechanics and Engineering, 200(49):3613–
3627, 2011.
O. Sigmund. Manufacturing tolerant topology optimization. Acta Mechan-

ica Sinica/Lixue Xuebao, 25(2):227–239, 2009.
tion, 12(2):555–573, 2002.
M. Tootkaboni, A. Asadpoure, and J. K. Guest. Topology optimization of
continuum structures under uncertainty–a polynomial chaos approach.
Computer Methods in Applied Mechanics and Engineering, 201:263–275,
2012.
M. A. Valdebenito and G. I. Schuëller. A survey on approaches for
reliability-based optimization. Structural and Multidisciplinary Opti-
mization, 42(5):645–663, 2010.
F. Wang, B. S. Lazarov, and O. Sigmund. On projection methods, conver-
gence and robust formulations in topology optimization. Structural and
W. Zhao, J. K. Liu, and Z. R. Lu. Structural optimization based on stochas-
tic finite element analysis and optimality criteria methods. Advanced
Materials Research, 295:2526–2530, 2011.
M. Zhou and G. I. N. Rozvany. The COC algorithm, part II: Topological,
geometrical and generalized shape optimization. Computer Methods in
Applied Mechanics and Engineering, 89(1-3):309–336, 1991.

2014 Book TopologyOptimizationInStructur PDF

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

2014 Book TopologyOptimizationInStructur PDF

Caricato da

Copyright:

Formati disponibili

CISM International Centre for Mechanical Sciences 549

Courses and Lectures

The Secretary General

The series presents lecture notes, monographs, edited works and

For further volumes:

© CISM, Udine 2014

All contributions have been typeset by the authors

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

Structural Topology Optimization (STO) is a relatively new, but ra-

G.I.N. Rozvany and T. Lewiński

Structural Topology Optimization (STO) – Exact Analyt-

The Extended Finite Element Method

6WUXFWXUDO7RSRORJ\2SWLPL]DWLRQ 672 ±([DFW

G. Rozvany, T. Lewiński (Eds.), Topology Optimization in Structural

/D\RXW RSWLPL]DWLRQ /2 GHDOV ZLWK JULGOLNH ORZ YROXPH IUDFWLRQ VWUXFWXUHV

ZKHUHVLVWKHORQJLWXGLQDOVWUHVVLQDQ\WUXVVPHPEHU V & LVWKHSHUPLVVLEOHVWUHVV

ZKHUH İ LVD VPDOO NLQHPDWLFDOO\DGPLVVLEOHVWUDLQDQGNLVDSRVLWLYHFRQVWDQW

5HVHDUFKHUV¶5HVSRQVHWR0LFKHOO¶V  3DSHU

 ([WHQVLRQ RI 0LFKHOO¶V 7KHRU\ WR 6WDWLFDOO\ ,QGHWHUPLQDWH

)RU V 7 V& V S ZHKDYH

 ([DPSOHV ,OOXVWUDWLQJ WKH /LPLWHG 5DQJH RI 9DOLGLW\ RI

ZKHUH X DQG Y DUHWKHDGMRLQWGLVSODFHPHQWVLQGLUHFWLRQV[DQG\

8VLQJWKHGXDOIRUPXODWLRQLQ  LWFDQEHVKRZQ HJ5R]YDQ\HWDO WKDWWKH

6XEVWLWXWLQJ H T N  H U N  U D DQG T S   LQWR   ZH ILQG WKDW 3'

DQGWKHQ  DQG  FRQILUPWKHYROXPHYDOXHLQ  

DW rq WRWKHD[LV[7KLVVWUDLQILHOGLVNLQHPDWLFDOO\DGPLVVLEOHEHFDXVHLWVDWLVILHV

)RUWKHGXDOIRUPXODLQ  WKHDGMRLQWGLVSODFHPHQWDWWKHSRLQW*LQ)LJELV E\WKH

9 G 3  V 7  G  3   V 7  V 7 | G 3V 7 

 2SWLPDO/D\RXW7KHRU\ 3UDJHUDQG5R]YDQ\D

G. Rozvany, T. Lewiński (Eds.), Topology Optimization in Structural

VWDWLFHTXDWLRQVRQO\LJQRULQJNLQHPDWLF FRPSDWLELOLW\ UHTXLUHPHQWVDQGWKHUHIRUH

 $6LPSOH([DPSOHRI$SSO\LQJWKH3UDJHU5R]YDQ\ D /D\RXW7KH

,WFDQEHUHDGLO\VHHQWKDWDORQJWKHOLQHVXSSRUWVZLWK[ DQG\ ZHKDYH

,QVSLWHRIWKHVHDGYDQWDJHVRQO\6LJPXQG 6LJPXQGHWDO KDVFRPSDUHG

 &RJQLWLYH 3URFHVVHV LQ 'HULYLQJ ([DFW $QDO\WLFDO 6WUXFWXUDO

N FRV D H    FRV D  H  H\ N   WDQ D 

,WFDQEHVHHQWKDWIRU D  DQG D q  JLYHVWKHFRUUHFW H  YDOXHVIRUWKH

H N D H % H N H H[  D H % H  H\

ZKHUH H LN DQG H LN DUHNLQHPDWLFDOO\DGPLVVLEOHUHDO HODVWLF DQGDGMRLQWVWUDLQVLQ

6LQFHWKLVSUREOHPLVV\PPHWULFLQWKHRQO\UDQGRPYDULDEOH + FRQGLWLRQ  LV

,Q   5 LVXVHGVLQFHWKHIDLOXUHPD\RFFXUDWERWKHQGVRIWKHLQWHUYDORI

 3DUWLDO 5HOD[DWLRQ RI WKH 2UWKRJRQDOLW\ 5HTXLUHPHQW IRU

G. Rozvany, T. Lewiński (Eds.), Topology Optimization in Structural

E WKHRULJLQDO0LFKHOOWUXVV LQFOXGLQJLWVDFWLYHVXSSRUWV LVFRQWDLQHGLQWKH

PXOWLSOH2%)6¶V ZLWKWKHH[FHSWLRQRI6UHJLRQV 7KLVLVEHFDXVHDORDGRQDUHJLRQ

([DPSOHV,Q)LJWKHWZRRSWLPDOWRSRORJLHVLQ E DQG F DUHV\PPHWULF,Q)LJ

([DPSOH VNHZV\PPHWULF ORDG RQO\ RQH VNHZV\PPHWULF 2%)6 0LFKHOO¶V

([DPSOH VNHZV\PPHWULF ORDG DQ LQILQLWH QXPEHU RI VNHZV\PPHWULF DQG

2XWOLQHRISURRI L DQG LL IROORZIURPWKHVNHZV\PPHWU\RIWKHRSWLPDOVWUHVV

([DPSOH $UHPDUNDEOHH[DPSOHRIDV\PPHWULF GLVFUHWL]HG RSWLPDO WRSRORJ\ E\

 1RQ6\PPHWULF([DFW2SWLPDO6ROXWLRQV IRU 6\PPHWULF 7RSRORJ\ 3URE

1RQFRQYH[ FRVW IXQFWLRQV )RUQRQFRQYH[FRVWIXQFWLRQVWKHRSWLPDOWRSRORJ\

IRU ) !  $ N)  IRU )   $ N   ) 

4 5  4 5 

F G H

 $Q,OOXVWUDWLYH([DPSOH±0RGLILHG0LFKHOO  3UREOHP

7KHRSWLPDOWRSRORJ\ )LJD FRQVLVWVRIDFLUFXODUIDQDQGWZRVWUDLJKWEDUV

 9DOLGDWLRQ RI 1XPHULFDO 0HWKRGV LQ 6WUXFWXUDO 7RSRORJ\

G. Rozvany, T. Lewiński (Eds.), Topology Optimization in Structural

HOHPHQWV ,Q )LJ  ( &090&9 HIILFLHQF\ RI D QXPHULFDO VROXWLRQ

VWUXFWXUHDSSURDFKKDVDYHU\LPSRUWDQWIXWXUH'XHWRWKH IL[HG QRGHV ZLWK SUH

VHFWLRQDODUHDVDUH]HURVR7L &L IRUDSSURSULDWHLQGLFHVL7KXVPRVWGHVLJQ

6WUXFWXUDO7RSRORJ\2SWLPL]DWLRQ672±([DFW

/D\RXW RSWLPL]DWLRQ /2 GHDOV ZLWK JULGOLNH ORZ YROXPH IUDFWLRQ VWUXFWXUHV

ZKHUHVLVWKHORQJLWXGLQDOVWUHVVLQDQ\WUXVVPHPEHU V & LVWKHSHUPLVVLEOHVWUHVV

ZKHUH İ LVDVPDOONLQHPDWLFDOO\DGPLVVLEOHVWUDLQDQGNLVDSRVLWLYHFRQVWDQW

5HVHDUFKHUV¶5HVSRQVHWR0LFKHOO¶V3DSHU

([WHQVLRQ RI 0LFKHOO¶V 7KHRU\ WR 6WDWLFDOO\ ,QGHWHUPLQDWH

([DPSOHV ,OOXVWUDWLQJ WKH /LPLWHG 5DQJH RI 9DOLGLW\ RI

ZKHUH X DQG Y DUHWKHDGMRLQWGLVSODFHPHQWVLQGLUHFWLRQV[DQG\

8VLQJWKHGXDOIRUPXODWLRQLQLWFDQEHVKRZQHJ5R]YDQ\HWDOWKDWWKH

6XEVWLWXWLQJ H T N H U N U D DQG T S LQWR ZH ILQG WKDW 3'

DQGWKHQDQGFRQILUPWKHYROXPHYDOXHLQ

DW rq WRWKHD[LV[7KLVVWUDLQILHOGLVNLQHPDWLFDOO\DGPLVVLEOHEHFDXVHLWVDWLVILHV

)RUWKHGXDOIRUPXODLQWKHDGMRLQWGLVSODFHPHQWDWWKHSRLQW*LQ)LJELVE\WKH

9 G 3 V 7 G 3 V 7 V 7 | G 3V 7

2SWLPDO/D\RXW7KHRU\3UDJHUDQG5R]YDQ\D

VWDWLFHTXDWLRQVRQO\LJQRULQJNLQHPDWLFFRPSDWLELOLW\UHTXLUHPHQWVDQGWKHUHIRUH

$6LPSOH([DPSOHRI$SSO\LQJWKH3UDJHU5R]YDQ\D/D\RXW7KH

,WFDQEHUHDGLO\VHHQWKDWDORQJWKHOLQHVXSSRUWVZLWK[ DQG\ ZHKDYH

,QVSLWHRIWKHVHDGYDQWDJHVRQO\6LJPXQG6LJPXQGHWDOKDVFRPSDUHG

&RJQLWLYH 3URFHVVHV LQ 'HULYLQJ ([DFW $QDO\WLFDO 6WUXFWXUDO

N FRVD H FRVD H H\ N WDQ D

,WFDQEHVHHQWKDWIRU D DQG D q JLYHVWKHFRUUHFW H YDOXHVIRUWKH

H N D H % H N H H[ D H % H H\

ZKHUH H LN DQG H LN DUHNLQHPDWLFDOO\DGPLVVLEOHUHDOHODVWLFDQGDGMRLQWVWUDLQVLQ

6LQFHWKLVSUREOHPLVV\PPHWULFLQWKHRQO\UDQGRPYDULDEOH+FRQGLWLRQLV

,Q5LVXVHGVLQFHWKHIDLOXUHPD\RFFXUDWERWKHQGVRIWKHLQWHUYDORI

3DUWLDO 5HOD[DWLRQ RI WKH 2UWKRJRQDOLW\ 5HTXLUHPHQW IRU

EWKHRULJLQDO0LFKHOOWUXVVLQFOXGLQJLWVDFWLYHVXSSRUWVLVFRQWDLQHGLQWKH

PXOWLSOH2%)6¶VZLWKWKHH[FHSWLRQRI6UHJLRQV7KLVLVEHFDXVHDORDGRQDUHJLRQ

([DPSOHV,Q)LJWKHWZRRSWLPDOWRSRORJLHVLQEDQGFDUHV\PPHWULF,Q)LJ

([DPSOH VNHZV\PPHWULF ORDG RQO\ RQH VNHZV\PPHWULF 2%)6 0LFKHOO¶V

([DPSOH VNHZV\PPHWULF ORDG DQ LQILQLWH QXPEHU RI VNHZV\PPHWULF DQG

2XWOLQHRISURRILDQGLLIROORZIURPWKHVNHZV\PPHWU\RIWKHRSWLPDOVWUHVV

([DPSOH $UHPDUNDEOHH[DPSOHRIDV\PPHWULF GLVFUHWL]HG RSWLPDO WRSRORJ\ E\

1RQ6\PPHWULF([DFW2SWLPDO6ROXWLRQV IRU 6\PPHWULF 7RSRORJ\ 3URE

1RQFRQYH[ FRVW IXQFWLRQV )RUQRQFRQYH[FRVWIXQFWLRQVWKHRSWLPDOWRSRORJ\

IRU ) ! $ N) IRU ) $ N )

4 5 4 5

FGH

$Q,OOXVWUDWLYH([DPSOH±0RGLILHG0LFKHOO3UREOHP

7KHRSWLPDOWRSRORJ\)LJDFRQVLVWVRIDFLUFXODUIDQDQGWZRVWUDLJKWEDUV

9DOLGDWLRQ RI 1XPHULFDO 0HWKRGV LQ 6WUXFWXUDO 7RSRORJ\

HOHPHQWV ,Q )LJ ( &090&9 HIILFLHQF\ RI D QXPHULFDO VROXWLRQ

VWUXFWXUHDSSURDFKKDVDYHU\LPSRUWDQWIXWXUH'XHWRWKH IL[HG QRGHV ZLWK SUH

VHFWLRQDODUHDVDUH]HURVR7L &L IRUDSSURSULDWHLQGLFHVL7KXVPRVWGHVLJQ

7KHH[DFWRSWLPDOYROXPHIRUV7 V&DQGWRSKDOISODLQGRPDLQLVHTXDOWRVHH

ZKLOHWKHVDPHYROXPHIRUIXOOSODQHRUERWWRPKDOISODQHGRPDLQLVHTXDOWR

2OKRIIHWDODOVRGHWHUPLQHGWKHRSWLPDOPLFURVWUXFWXUHVDQGRSWLPDOUH

,PSRUWDQW &RQWULEXWLRQV E\ &KLQHVH 6FLHQWLVWV WR 1XPHULFDO 7RSRORJ\

6LJPXQG¶V 0RVW 6LJQLILFDQW )RUXP $UWLFOH RQ 1RQJUDGLHQW

%LUNHU7/HZLQVNL75R]YDQ\*,1$ZHOOSRVHGQRQVHOIDGMRLQWOD\RXWSUREOHP

0F.HRZQ--8SSHUOLPLWVRQWKHQXPEHURIHOHPHQWVLQHODVWLFVWUXFWXUHVRIPLQ

5R]YDQ\,14XHULQ20DVSDU=3RPH]DQVNL9:HLJKWLQFUHDVLQJHIIHFWRI

$OUHDG\LQWKHRULJLQDOSDSHUE\0LFKHOOLWZDVGLVFORVHGWKDWWKHRSWLPXP

ZKHUH'NLVOLQNHGZLWKYDE\7KHDERYHHTXDWLRQVLPSO\

/HW V V 7 V & 7KHYROXPHRIWKHWUXVVLVHVWLPDWHGIURPEHORZE\

ZKHUHWKHUHODWLYHHORQJDWLRQVDUHOLQNHGZLWKYE\7KHRUGHURIPLQDQGPD[

LI H . V V7 DQG H / V V & D

7KHPD[RSHUDWLRQLQH[FOXGHVWKHFDVHF+HQFH

ZKHUH HNGHSHQGVRQYE\/HWXVFRPHEDFNWR7KHPLQLPDOYROXPHLV

5HFDOOWKDW%NDLQYROYHGLQDUHGHWHUPLQHGE\UD

ZKHUH L w w[L DQGYLLVWKHLWKFRPSRQHQWRIY7KHFRPSRQHQWV HLMIRUPWKH

/HWQ Q,EHWKHPD[LPL]HUDQGQ Q,,,WKHPLQLPL]HURIWKHDERYHIRUPXODH7KHVH

7KHFRQGLWLRQVRUZLOOEHZULWWHQHH%V&V7ZKHUH%V&V7LVWKHVHW

/HWXVGLVFXVVVHOHFWHGVROXWLRQVWRWKH0LFKHOOSUREOHPJLYHQLQ)LJ,Q

7KH3UREOHPDVD3UREOHP'XDOWRD6WUHVVEDVHG)RU

1İ 1 , İ, 1 ,, İ ,, VLQ M H , H ,, 1 , 1 ,,

$SDUWIURPWKHGRPDLQV257ZHLQWURGXFHWKHGRPDLQV6ZKHUHH, H, RU

7KHUHDUHVROXWLRQVLQZKLFK:7 :7KHFRQGLWLRQVGHWHUPLQHWKH+HQFN\

ZKHUH[ [[ \7KHQZHLQWURGXFHWKHEDVLVYHFWRUV J L wU w[ L WDQJHQWWRWKH

$[ [ J [ [ % [ [ J [ [

FRPSRQHQWV H H Z ZE\PHDQVRIWKHSK\VLFDOFRPSRQHQWVXYRIWKHGLV

$FFRUGLQJWRZHKDYH

ZKHUH I I D E +DYLQJ I I D E $D E % D E RQH FDQ UHFRYHU WKH

GI LIGI H LI G HLI I

ZHUHDUUDQJHWRWKHIRUP

G ¬ª HLI X LY ¼º HLI ª¬ $ H LZ GD % Z LH G E º¼

WKHWKHRU\RI0LFKHOOVWUXFWXUHV,QSDUWLFXODUWKH/DPpFRHIILFLHQWV$DE%DE

G ª¬ HLI X LX º¼ HLI ¬ª $G D L%G E LZ $G D L%G E ¼º

G ª¬ HLI X LX º¼ HLI ª¬ $ LZ G D % L Z G E º¼