Sei sulla pagina 1di 88

Hybrid and enhanced strain elements

Vinh Phu Nguyen


Delft Unitversity of Technology
Falculty of Civil Engineering and Geosciences
Computational Mechanics Group

December 19, 2011


2 Chap. 0
Contents

0.1 Variational principles for linear elasticity . . . . . . . . . . . . . . . . . . . . . 1


0.1.1 Principle of minimum total potential energy . . . . . . . . . . . . . . . 2
0.1.2 Principle of minimum complementary energy . . . . . . . . . . . . . . 5
0.1.3 The Hellinger-Reissner functional . . . . . . . . . . . . . . . . . . . . 7
0.1.4 Veubeke-Hu-Washizu variational principle . . . . . . . . . . . . . . . 9
0.1.5 Assumed stress hybrid principle . . . . . . . . . . . . . . . . . . . . . 11
0.1.6 Two field stress-strain variational principle . . . . . . . . . . . . . . . 13
0.1.7 Two-field displacement-strain variational principle . . . . . . . . . . . 16
0.2 Variational principles for large deformation . . . . . . . . . . . . . . . . . . . 17
0.2.1 Summary of key equations . . . . . . . . . . . . . . . . . . . . . . . . 17
0.3 HR-based mixed elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
0.3.1 Approximations and discrete equation . . . . . . . . . . . . . . . . . . 19
0.3.2 Construction of P matrix . . . . . . . . . . . . . . . . . . . . . . . . . 19
0.3.3 A four-noded HR mixed elements . . . . . . . . . . . . . . . . . . . . 21
0.3.4 Numerical examples . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
0.4 Displacement-strain mixed finite elements . . . . . . . . . . . . . . . . . . . . 22
0.5 Hybrid stress finite elements for linear elasticity . . . . . . . . . . . . . . . . . 25
0.5.1 Equilibrium stress hybrid elements . . . . . . . . . . . . . . . . . . . . 25
0.5.2 Homogeneous Voronoi cell element . . . . . . . . . . . . . . . . . . . 33
0.5.3 Composite Voronoi cell element . . . . . . . . . . . . . . . . . . . . . 36
0.6 Hybrid stress finite elements for inelastic materials . . . . . . . . . . . . . . . 39
0.6.1 Variational principle and discrete equations . . . . . . . . . . . . . . . 39
0.6.2 Solution procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
0.7 Enhanced assumed strain elements . . . . . . . . . . . . . . . . . . . . . . . . 42
0.7.1 Assumed strain field . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
0.7.2 Discrete system of equations . . . . . . . . . . . . . . . . . . . . . . . 43
0.7.3 Construction of G matrix . . . . . . . . . . . . . . . . . . . . . . . . . 44
0.7.4 Stress recovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
0.8 EAS for small strain inelastic materials . . . . . . . . . . . . . . . . . . . . . . 49
0.9 EAS for large deformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
0.9.1 EAS based on the enhancement of E . . . . . . . . . . . . . . . . . . . 55
0.10 Modified enhanced assumed strain methods . . . . . . . . . . . . . . . . . . . 60
0.10.1 Taylor expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
0.11 Mixed-enhanced strain elements . . . . . . . . . . . . . . . . . . . . . . . . . 60
Sec. 0.1 Variational principles for linear elasticity 1

0.12 Small strain rate independent plasticity . . . . . . . . . . . . . . . . . . . . . . 62


0.12.1 Fully implicit scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
0.12.2 J2 plasticity with isotropic hardening . . . . . . . . . . . . . . . . . . 68
0.12.3 Algorithmic tangent modulii of von Mises plasticity . . . . . . . . . . 72
0.12.4 J2 plasticity with isotropic/kinematic hardening . . . . . . . . . . . . . 73
0.12.5 Matlab implementation . . . . . . . . . . . . . . . . . . . . . . . . . . 75
0.12.6 Some numerical examples . . . . . . . . . . . . . . . . . . . . . . . . 75
0.13 Large strain rate independent plasticity . . . . . . . . . . . . . . . . . . . . . . 75
0.13.1 Introduction to manifold analysis . . . . . . . . . . . . . . . . . . . . 75
0.13.2 Multiplicative hyperelastic plasticity . . . . . . . . . . . . . . . . . . . 76
0.14 Crystal plasticity finite elements . . . . . . . . . . . . . . . . . . . . . . . . . 77
0.14.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
0.14.2 Plastic deformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
0.14.3 Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
0.14.4 Plastic flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
0.14.5 Hardening rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
0.14.6 Stress update algorithms . . . . . . . . . . . . . . . . . . . . . . . . . 82

0.1 Variational principles for linear elasticity


Single-field variational principles

Principle of total potential energy: displacement elements


Principle of total complementary potential energy: equilibrium elements (stress field
is unknown) [25]

Multi-field variational principles

Mixed variational principles


Hellinger-Reissner principle: HR mixed elements
Veubeke-Hu-Washizu principle: Enhanced Assumed Strain elements
Modified principles for incompressible media
Hybrid variational principles: hybrid finite elements

Multi-field variational principles allow for the development of efficient finite elements:

improving existing conforming low-order displacement based finite elements (coarse


mesh accuracy, locking free, distorted mesh)

improving stress solution

relaxing continuity requirements (plate and shell elements)


2 CONTENTS Chap. 0

developing n-sided (n 3) polygonal elements

Remark 0.0.1. In a mixed formulation, there are more than two fields that are independently
approximated in the element domain. On the other hand, in a hybrid formulation, there is at least
one field defined in the element domain and another field defined along the element boundary.
As a matter of fact, hybrid-mixed formulations involve more than one type of internal field
variables and also contain boundary variables.
In this section all important variational principles in linear elasticity are derived in details.
Note that the construction of versatile and efficient finite elements requires the combination of
all these variational principles. Moreover, extension of these variational principles to the case
of inelasticity is also presented.
Considering a solid that is bounded by . Prescribed displacements u are imposed on the
Dirichlet boundary u and prescribed tractions t are applied on the Neumann boundary
t . Table 1 lists the governing equations of the elasto-static problem where B is the
strain-displacement matrix and D is the elasticity matrix.

Name Acr Valid Matrix form


Kinematics Equation KE = Bu
Constitutive Equa. CE = D
Balance Equation BE + b = 0
Primary BCs. PBC u u=u
Flux BCs. FBC t t=t

Table 1: Summary of elasto-static governing equations.

0.1.1 Principle of minimum total potential energy


The well known displacement finite elements, widely used in engineering analysis, are derived
from the principle of minimum total potential energy. According to this principle, the balance
equation (or equilibrium equation) and the flux BCs are satisfied in a weak sense while other
equations are identically fulfilled. The unknown field that subjects to variation is the displace-
ment field. In summary,

Master field: u

Slave fields: ,

Strong links: KE, CE, PBC. Those equations are fulfilled point by point

Weak links: BE and FBC. Those equations are fulfilled in an average sense.

Gluing links: Lagrange multiplier method


Sec. 0.1 Variational principles for linear elasticity 3

The two week links (BE and FBC) are enforced using the Lagrange multipliers defined as a
variation of the displacement field (work conjugating pairs). That is
Z Z
(ij,j + bi )ui d = 0, (ij nj ti )ui dt = 0 (1)
t

We can write

Z Z Z
ui
ij,j ui d = ij nj ui d ij d (Gauss theorem)
xj

Z  Z   
1 ui 1 uj 1 ui 1 uj
= ij nj ui d ij
+ + d
| 2 xj 2 xi 2 x j 2 xi

{z } | {z }
ij unsymmetric
Z Z
= ij nj ui d ij ij d

(2)

where in the second equality, use was made of the fact that double contract of a symmetric
tensor (stress field) and an unsymmetric tensor is zero. Hence the first equation in Eq. (1)
becomes
Z Z Z
ij nj ui d ij ij d + bi ui d = 0 (3)

Using the second equation in Eq. (1), the above becomes
Z Z Z
ij ij d bi ui d ti ui dt = 0 (4)
t

where the fact that ui = 0 on u was used. In matrix notation, the above reads
Z Z Z 
T T T
d u bd + u tdt = 0 (5)
t

This is the well known equation of virtual work and is the starting point to construct displace-
ment type finite element formulations. It is emphasized that this equation has been derived
without any assumption on the stress-strain relation i.e., it can be applied to inelastic solids.
If there exists a strain energy function W (ij ) so that
W
ij = (6)
ij
W
then, ij ij = ij
ij = W . Equation (4) can be rewritten as

TPE = 0 (7)
4 CONTENTS Chap. 0

with the total potential energy functional TPE given by


Z Z Z
TPE (ui ) = W (ij )d bi ui d ti ui dt (8)
t
| {z } | {z }
internal energy external energy

It has been shown that, if the strain energy function is positive definite, the stationary value
defined by Eq. (7) corresponds to a minimum. Therefore, it is referred to as the principle
of minimum total potential energy introduced by Kirchhoff, in 1850, in the framework of the
theory of linear elasticity. It only involves displacement fields as unknown variables and it states
that, among all kinematically admissible displacement fields, those that satisfy the equilibrium
conditions lead to a minimum value of the total potential energy functional.

Remark 0.0.2. Displacement like finite elements can be constructed using either the equation
of virtual work, see Eq. (5) or the principle of minimum potential energy given by Eqs. (7) and
(8). The numerical solutions obtained by means of these models are very often referred to as
kinematically admissible solutions, since they satisfy in strong form the kinematical differential
equations as well as the kinematical (or Dirichlet) boundary conditions of a given boundary
problem.

Remark 0.0.3. Note that, although displacement models lead to sufficiently accurate displace-
ment fields, the corresponding stress fields may be highly erroneous. This occurs since the
accuracy of the approximate displacement field rapidly deteriorates when differentiations are
required to compute other results, such as stresses or deformations. This drawback of displace-
ment models can however be overcome by using high order approximations such as meshfree
approximations and NURBS.

Euler equations associated to a variational principle are found by computing the first varia-
tion of the functional and setting it to zero. The first variation of the total potential energy Eq.
(8) is given by
Z Z Z
TPE = ij ij d bi ui d ti ui d (9)
t


Using the identity ij ij = xj
(ij ui ) ui ij,j , the above becomes
Z Z
TPE = (ij,j + bi )ui d + (ij nj ti )ui d (10)
t

Setting TPE to zero yields the following equations after the fundamental lemma of variational
calculus (see e.g., [4, 12])

ij,j + bi = 0 in
(11)
ij nj = ti on t
Sec. 0.1 Variational principles for linear elasticity 5

which are essentially the governing equations that we need to solve. In the field of variational
calculus, they are known as Euler equations.

Finite element applications In a FEM framework, the solid is discretized into a number of
non-overlapping elements e . The total potential energy functional then reads
X
TPE (ui ) = eTPE (ui ) (12)
e

with the elementary functional given by


Z   Z Z
e 1
TPE (ui ) = W (ui,j + uj,i) d bi ui d ti ui dt (13)
e 2 e et
| {z } | {z }
internal energy external energy

0.1.2 Principle of minimum complementary energy


This is a single field principle in which the primary field is the stress field. The kinematic equa-
tion and the Dirichlet BCs are enforced in a weak sense while other equations are identically
satisfied. In summary,

Master field: that satisfies the equilibrium equation

Slave fields: u,

Strong links: BE, CE, FBC. Those equations are fulfilled point by point

Weak links: KE and PBC. Those equations are fulfilled in an average sense.

Gluing links: Lagrange multiplier method

The weak enforcement of KE and PBC are given as


Z   Z
1
ij (ui,j + uj,i ) ij d = 0, (ui ui )ij nj du = 0 (14)
2 u

where the first equation is to enforce the two strain fields (one from the displacement field and
one from the stress field) to be coincident.
One can write

 
1 1 1
(ui,j + uj,i)ij = (ui,j + uj,i) + (ui,j uj,i ) ij = ui,j ij
2 2 2
(15)

= (ui ij ) ui (ij ) = (ui ij )
xj xj xj
6 CONTENTS Chap. 0

The term x j (ij ) = ij,j vanishes since the balance equation is enforced strongly (ij,j =
bi and bi does not subject to a variation). Therefore one can write
Z Z
1
(ui,j + uj,i )ij d = ui ij nj d (16)
2

Similarly, ij nj = 0 on t . The above equation thus becomes


Z Z
1
(ui,j + uj,i )ij d = ui ij nj d (17)
2 u

The first equation in Eq. (14) is then simplified to


Z Z Z Z

ij ij d ui ij nj d = ij ij d ui ij nj d = 0 (18)
u u

where in the second equality, the second equation in Eq. (14) was used. This equation is the
equation of virtual complementary work. Again, the derivation does not make use of the stress-
strain relation. Therefore, the equation of virtual complementary work can be readily applied
for inelastic materials.
Finite element models relying on the equation of virtual complementary work usually called
equilibrium models. The numerical solutions obtained through these models are very often
called statically admissible solutions.
Let us assume that there exists a function W (ij ) so that

W (ij )
ij = (19)
ij
W (ij )
W (ij ) is called the complementary energy function. We can write ij ij = ij
ij =
W . Therefore, Equation (18) is equivalent to

TCPE = 0 (20)
with the total complementary potential energy being given by
Z Z
TCPE (ij ) = W (ij )d

ui ij nj d (21)
u

Note that in this functional only the stress field appears and there is no presence of the displace-
ment field except the prescribed one.
In case of linear elastic solid, the complementary energy reads W = 1/2ij Cijkl kl , so we
have
Z Z
1
TCPE = ij Cijklkl d ui ij nj d (22)
2 u

Euler equations Again, Euler equations are found by computing the first variation of the func-
tional TCPE and setting it to zero. That is
Sec. 0.1 Variational principles for linear elasticity 7

Z Z
TCPE = ij ij d ui ij nj d
u
Z Z Z Z
(23)
= ij ij d ui ij nj d + ui ij nj d ui ij nj d
u u
| {z }
0

Replacing the last term in the above by using Eq. (17), the above becomes
Z Z Z Z
u
TCPE = ij ij d ij ij d ui ij nj d + ui ij nj d (24)
u u

By TCPE = 0 and the fundamental lemma of variation calculus, we obtain the following Euler
equations

ij = uij in
(25)
ui = ui on u

0.1.3 The Hellinger-Reissner functional


Master field: u,

Slave fields: u and

Strong links: KE, CE, PBC. Those equations are fulfilled point by point

Weak links: BE and FBC. Those equations are fulfilled in an average sense.

Gluing links: Lagrange multiplier method

The two slave fields u and are given by

u = Bu, = C (26)
where C is the compliance tensor. Thus, we have the following equations

Z
(ij,j + bi )ui d = 0
Z

(ij nj ti )ui dt = 0 (27)


t
Z
(uij ij )ij d = 0

which results in the following equation (use was made of Eq. (2))
8 CONTENTS Chap. 0

Z Z Z
  
HR = uij ij + ij uij ij ij d bi ui d ti ui dt = 0 (28)
t

The above is the variation of the following HR functional


Z   Z Z 
T 1 T T T
HR (, u) = u C d u bd + u tdt (29)
2 t

where use was made of the following identity


 
1 T
C = Cijkl kl ij = ij ij (30)
2
Proof. Using the directional derivative, one can write
 
1 T d  T

C = ( + ) C( + )
2 d =0
(31)
1 T 
= C + T C = T C
2
where use was made of the symmetry of C in the final step.
Remark 0.0.4. It can be seen that the complementary energy principle and the HR principle
require the compliance tensor C. In other words, these principles use a complementary energy
density. For linear elasticity this tensor can be easily obtained by inverting the Hooks law
= D. It is emphasized that, however, inversion of stress-strain relations is not in general
possible. Legendre transformation to construct the complementary energy density.

Z Z Z 
 T  T T
HR (, u) = u W () d u bd + u tdt (32)
t

From the functional in Eq. (29), the variational index of the displacement field is one while
the variational index of the stress field is 0 (no derivative of stresses appears in the functional).
Therefore,

Displacement shape functions must be C 0 (continuous) between elements and C 1 inside


(continuous and differentiable).

Stress shape functions can be C 1 (discontinuous) between elements, and C 0 (continu-


ous) inside.

Generalized HR functional In this so-called generalized HR functional, the Dirichlet boundary


condition is enforced in a weak sense. That is
Z
ij nj (ui ui )d = 0 (33)
u
Sec. 0.1 Variational principles for linear elasticity 9

Following the same line of derivation as done for the HR functional, we have

Z Z
  
gHR = uij ij + ij uij ij ij d bi ui d
Z Z
(34)
ij nj ui d ij nj (ui ui )d = 0
u

Splitting the surface integral on into two integrals, one on u and one on t , the above is
rewritten as

Z Z
  
gHR = uij ij + ij uij d ij ij
bi ui d

Z Z Z
ij nj ui d ij nj ui d + ij nj (ui ui )d
u t u
Z
 u  
Z Z (35)
u
= ij ij + ij ij ij ij d bi ui d ij nj ui d
t
Z Z
ij nj ui d ij nj (ui ui )d
u u

The above is the exact first variation of the following generalized HR functional
Z
g
HR = HR ij nj (ui ui )d (36)
u

0.1.4 Veubeke-Hu-Washizu variational principle


This section presents a derivation of the well known Veubeke-Hu-Washizu functional. Note that
this functional is usually referred to as the Hu-Washizu functional in literature. For a historical
note on the contribution of Fraeijs de Veubeke on this functional, see [3]. This is a three-field
mixed functional in which the displacement, strain and stress fields are all primary unknowns.

Master fields: u, and


Slave fields: u and
Weak links: BE, FBC

By matching (u , ) and ( , ) together with the standard equations for enforcing BE and FBC,
we have

Z Z
(uij ij )ij d = 0, (ij ij )ij d = 0
Z Z (37)
(ij,j + bi )ui d = 0, (ij nj ti )ui d = 0
t
10 CONTENTS Chap. 0

By using the following equation (see Eq. (2))


Z Z Z
ij,j ui d = ti ui d ij uij d (38)
t

together with the equations in Eq. (37)1 , one obtains

Z
 
VHW = (uij ij )ij + (ij ij )ij + ij uij d
Z Z 
ti ui d + bi ui d = 0 (39)
t
| {z }
ext
U
This general functional, that encompasses all the aforementioned variational principles, is a
starting point for developing many efficient finite elements. It should be emphasized that, the
original form as in the above equation is, however, not used for it would lead to elements with
too many degrees of freedom.

VHW functional for inelastic materials By introducing the strain energy function W (ij ), Eq.
(39) can be rewritten as
Z
 u 
VHW = (ij ij ) (ij ij ) + W (ij ) d U ext (40)

which is the variation of the following VHW functional


Z
VHW (u, , ) = [W () + : (u )] d U ext (u) (41)

Linear elasticity In this case, the functional reads


Z  
1 T T T
VHW (u, , ) = D + Bu d U ext (u) (42)
2
Euler equations are obtained by computing the first variation of the functional

Z Z Z
W u
VHW = : d + : ( )d + : (u )d U ext (43)

After rearrangement and some straightforward computations, we arrive at

Z   Z
W
VHW = : d + : (u )d

Z Z (44)
(div + b) ud + ( n t) ud
t
Sec. 0.1 Variational principles for linear elasticity 11

Using the identity VHW = 0 and the fundamental lemma of variational calculus yield the
following Euler equations

div + b = 0
W
= 0
(45)
u = 0
n t = 0

Type Master fields Name Usefullness


Single-field displacement TPE ++++
Single-field stress TCPE +
Single-field strain -
Two-field displacement, stress HR ++
Two-field displacement, strain -
Two-field stress, strain -
Three-field displacement, stress, strain VHW +++

Table 2: Canonical variational principles in elasticity

Displacement compatibility at interelement boundaries ui = ui

0.1.5 Assumed stress hybrid principle


The assumed stress hybrid variational principle was developed by Pian and Tong [16]. It is based
on the principle of minimum complementary energy with an enforcement of the interelement
traction continuity condition.
A
i
ta
(b)
u t
(a) tb

Figure 1: hybrid-2elems
12 CONTENTS Chap. 0

Considering a simple mesh consisting of two triangular elements as shown in Fig. (1).
Along the (only) interelement boundary AB, the traction continuity condition reads

ta + tb = 0 on AB (46)
This condition is enforced using the Lagrange multiplier method. Hence, the hybrid functional
is given by

Z
hybrid (ij , i ) = TCPE (ij ) + i (tia + tib )ds
ABZ
Z Z (47)
= W (ij )d

ui ij nj d + i (tia + tib )ds
u AB

where Eq. (21) has been used. Since the line integral over AB will be considered twice, we can
write
Z Z Z
hybrid (ij , i ) = W (ij )d

ui ij nj d + i ti d (48)
u i

where i denotes the interelement boundary. In order to find the physical interpretation of the
Lagrange multiplier i , we compute the first variation of the functional. That is,

Z Z Z
hybrid = ij ij d uij ij d
ui ij nj d
u
Z Z Z (49)
+ ui ij nj d + i ij nj d + i ij nj d
u i i

where use was made of Eq. (24). From this equation, the Lagrange multipliers are the negatives
of the displacements on the interelement boundaries i.e., i = ui . Thus, the hybrid functional
is now written as
Z Z Z
hybrid (ij , ui ) = W (ij )d

ui ti d ui ti d (50)
u i
R
It
R is convenient
R to Rtransform the last term in the above equation using the identity
d =
u
d + t
d + i
d, hence

Z Z Z Z Z
hybrid (ij , ui ) = W (ij )d

ui ti d ui ti d + ui ti d + ui ti d (51)
u u t

Using the Dirichlet boundary condition ui = ui on u and the Neumann boundary condition
ti = ti on t , the above becomes
Z Z Z
hybrid (ij , ui ) = W (ij )d ui ti d +

ui ti d (52)
t
Sec. 0.1 Variational principles for linear elasticity 13

This functional will be used in Section 0.5 to derive various assumed stress hybrid finite
elements.

Euler equations It can shown that the Euler equations associated to the assumed stress hybrid
functional include (i) Eq. (25)1 , (ii) the FBC and (iii) the interelement traction continuity. The
balance equation, the constitutive equation, the kinematic equation and the Dirichlet BC are
satisfied in a strong sense.

0.1.6 Two field stress-strain variational principle


Stress field ij identically satisfies the equilibrium condition;

Strain field ij

There are two slave fields namely strain field from the displacement uij and stress field from
the strains ij .
Z Z Z
u
(ij ij )ij d + (ij ij )ij d (ui ui )ij nj d = 0 (53)
u

With the following equation


Z Z
uij ij d = ui ij nj d (54)
u

Equation 53 thus becomes


Z Z Z

ij ij d (ij ij )d + ui ij nj d = 0 (55)
u

By introducing the strain energy function W (ij ), the above can be rewritten as follows
Z Z Z
W (ij )d (ij ij )d + ui ij nj d = 0 (56)
u

which is the variation of the following functional


Z Z
(ij , ij ) = [W (ij ) ij ij ] d + ui ij nj d (57)
u

for linear elastic materials W (ij ) = 12 Dijkl kl ij , we therefore obtain the following func-
tional
Z   Z
1
(ij , ij ) = Dijkl kl ij ij ij d + ui ij nj d (58)
2 u

In order to make a hybrid functional from this, the stress field is no longer necessarily contin-
uous across the interelement boundary. The traction continuity at these boundaries is enforced
using an additional fieldthe boundary displacement field di .
14 CONTENTS Chap. 0

Z   Z Z
1
hybrid (ij , ij , di ) = Dijkl kl ij ij ij d + di ti d di ti d (59)
2 t

Z Z Z Z Z
T T

= + d + T d + T
d td + T
d td dT td (60)
t

The stresses, strains and the boundary displacements are approximated by

= P (61a)
= M (61b)
d = Na (61c)

where P, M, N are matrices of approximation functions which are described later. The internal
stress and strain variables are denoted by and , respectively. Boundary nodal displacements
are represented by a.
Traction at the interelement boundaries is defined as

t = n = nP = R (62)

Substituting Eqs. 61 and 62 into Eq. 60 yields

Z Z
T T T T

= P M + M P d + T MT DT Md
Z Z Z (63)
T T T T
+ a N Rd + a N Rd aT NT td
t

Setting to zero giving us the following equations

Z

T PT M + T MT DT M d = 0
Z Z
M Pd + aT NT Rd = 0
T T
(64)
Z Z

NT Rd NT td = 0
t
Sec. 0.1 Variational principles for linear elasticity 15

Let us introduce the following matrices and vector

Z
G= RT Nd (65a)
Z
H= PT Md (65b)

Z
Q= MT DMd (65c)
Z
f= NT td (65d)
t

Eq. 64 becomes

HT + Q = 0 (66a)
H + Ga = 0 (66b)
GT = f (66c)

From Eq. 66a, one gets = Q1 HT , substituting this into Eq. 66b gives us

= L1 Ga, L = HQ1 HT (67)


Introducing into Eq. 66c yields
 T 1 
G L G a=f (68)
in which the expression in the bracket is the stiffness matrix K.
In 2D, the dimensions of the matrices are as follows

P : 3 n
M : 3 n
R : 2 n
(69)
G : n nd
H : n n
Q : n n
where n , n , nd represent the number of stress parameters, strain parameters and nodal dis-
placement values per element.
In order for K to be rank efficient, the number of stress parameters must satisfy the following
condition

n nd nr (70)
16 CONTENTS Chap. 0

where nr represents the number of rigid body motion modes. Furthermore, matrix L must be
invertible.

1
xx 1 y 0 0 0 x 0
2
yy = 0 0 1 x 0 0 y .. P
(71)
.
xy 0 0 0 0 1 y x
7
The strain field should satisfy the Saint-Venant compatibility equation that reads

2 xx 2 yy
2 xy
+ 2
=0 (72)
y 2 x2 xy

xx 0 1
1 0

yy = 0 10 2 = M (73)
xy 1 3
0 0


1
xx 1 0 0 y 0 2

yy = 0 1 0 0 x 3 = M
(74)

xy 0 0 1 x y 4
5

0.1.7 Two-field displacement-strain variational principle


Two master fields: u and

Equilibrium, kinematics, Dirichlet and Neumann conditions enforced in a weak sense

Constitutive relation is identically satisfied

Z Z Z
(ij uij )ij d + (ij,j + bi )ui d + (ui ui )ij nj d = 0 (75)
u

The second integral in the above can be elaborated as follows (see Eq. 3)
Z Z Z
u
ij nj ui d ij ij d + bi ui d = 0 (76)

Splitting the first surface integral into two parts, one part on u and one on t one obtains
Z Z Z Z
u
ij nj (ui ui )d + ti ui d ij ij d + bi ui d = 0 (77)
u t

Substituting of Eq. (77) into Eq. (75) yields


Sec. 0.2 Variational principles for large deformation 17

Z Z
 
ij ij (ij uij ) d + [(ui ui )ij nj ]d + ext = 0 (78)
u

Using the strain energy function W (ij ), we have


 
W W
ij ij = (ij ij ) ij ij = ij ij (79)
ij ij
With this equation, Eq. (78) is the first variation of the following functional

Z   Z
W u
(ui , ij ) = (ij ij ) W (ij ) d + (ui ui )ij nj d + ext = 0 (80)
ij u

Linear elasticity The functional in Eq. (80) becomes

Z   Z
1 u
(ui , ij ) = Dijkl ij kl Dijkl kl ij d + [(ui ui )ij nj ]d + ext (81)
2 u

Z   Z
1 1
(ui , ij ) = Dijkl ij kl Dijkl kl (ui,j + uj,i) d + [(ui ui )Dijkl kl nj ]d + ext
2 2 u
(82)

0.2 Variational principles for large deformation


0.2.1 Summary of key equations
Governing equations in the total Lagrangian form are given in Table 3. Work conjugate pairs
of stress and strain measures are shown in Table 4. Finally, relations between different stress
measures are presented in Table 5.

Kinematics F E = 1/2(FT F I)
Equilibrium DivP + 0 b = 0 v Div(FS) + 0 b = 0 v

W W
Constitutive P= S=
F E
Table 3: Governing equations in the total Lagrangian form.
18 CONTENTS Chap. 0

F E D
(Cauchy stress) - - x
S (second PK stress) - x -
P (first PK stress) x - -
(Kirchhoff stress) - - x

Table 4: Work conjugate pairs of stress and strain measures.

S P
- J 1 F S FT J 1 P FT J 1
S JF F-T
1
- F1 P F1 F-T
P JF1 FS - F1
J F S FT P FT -

Table 5: Relation between different stress measures.

x
F= = I + Grad u deformation gradient (83)
X
C = FT F right Cauchy deformation tensor (84)
T
b= FF left Cauchy deformation tensor (85)
1 1
E = (C I) = (FT F I) Green strain tensor (86)
2 2
1
e = (I b )
1
Almansi strain tensor (87)
2

L = F F1 velocity gradient (88)


1
D = F-T E F1 = (L + LT ) rate of deformation tensor (89)
2
1 T
W = (L L ) spin tensor (90)
2


d 1 
E = (F + )T (F + ) I
d =0 2

(91)
1 T 
= F + T F = FT = FT Gradu
2
where the identity F = Gradu was used in the final equality.
Sec. 0.3 HR-based mixed elements 19

0.3 HR-based mixed elements


0.3.1 Approximations and discrete equation
The displacement field and the strain field are defined in a standard manner and are given by

u = N()a, u = Ba (92)
where N() is the shape functions matrix and a denotes the nodal displacement vector.
The stress field is approximated by

= P (93)
where vector contains the stress parameters which are not nodal values i.e., not defined at the
nodes and matrix P contains the stress interpolation functions to be defined later on.
Introducing the aboves into the HR functional given in Eq.(29) yields

Z   Z Z 
T T 1 T T T T T T
HR = a B P P CP d a N bd + a N tdt (94)
2 t

Stationary of this functional with respect to a and gives us

Z  Z Z
HR T T
=0: B Pd = N bd + NT tdt (95)
a
| {z t }
ext
Z  Z f 
HR
=0: PT Bd a = PT CPd (96)

| {z } | {z }
G H
From Eq. (96), an expression is found for which is then substituted into Eq. (95). Specially,

T 1 ext
= H1 Ga, G
| H {z G} a = f (97)
K
this results in the standard element stiffness matrix with the nodal displacements as the only
unknown. The process of eliminating the stress parameters at the element level is referred to
as static condensation.

0.3.2 Construction of P matrix


Rank deficiency The dimension of P matrix must be chosen in such a way that the element
stiffness matrix has a proper rank. From the theory of linear algebra, we know that

rank(AB) min(ra , rb ), rank(A + B) ra + rb (98)


20 CONTENTS Chap. 0

where ra and rb are the ranks of matrices A and B, respectively.


The element stiffness matrix reads

|{z} GT
K = |{z} H 1
|{z} G
|{z} (99)
nu nu nu n n n n nu
where n is the number of stress parameters i.e., length of the vector , nu denotes the number
of displacement degrees of freedom. In order for K to be rank efficient, the number of stress
parameters must satisfy the following condition

n nu nr (100)
where nr represents the number of rigid body motion modes. It is emphasized that n is re-
stricted by the limitation principle of Fraeijs de Veubeke [25]. Moreover, n should be as small
as possible to reduce the computational expense.
The stress interpolation matrix P is computed from the one defined in the parent domain
using the following equation

P = FT
0 P() (101)
where the transformation matrix F0 for two dimensional problems is given by

2
J11 J12 J12 J11 J12
2
F0 = J21 J21 J22 J21 J22 (102)
2J11 J21 2J12 J22 J11 J22 + J12 J21 =0

Box 0.1 summarizes the HR based mixed finite element formulation for linear elasticity.

Box 0.1 HR mixed elements: linear elasticity

P = FT P()
Z0
He = PT CPd (n n )
Z e
Ge = PT Bd (n nd )
e
Ke = GT 1
e He Ge (nd nd )
X
K= Ke
e
ext
Ka = f
= PH1
e Ge ae
Sec. 0.3 HR-based mixed elements 21

0.3.3 A four-noded HR mixed elements


For bilinear quadrilateral elements, n 5. The most efficient element is therefore obtained
with five stress parameters. The stress interpolation matrix reads

1 0 0 0

P() = 0 1 0 0 (103)
0 0 1 0 0
from which the HR Q4 element is free of volumetric and shear locking. An explanation for this
free locking behaviour is given later.
A straightforward (not efficient) implementation of the HR four-noded quadrilateral element
is given in List 1. Note that in this implementation, the nodal displacements are ordered as
a = [ax1 ax2 . . . ay1 ay2 . . .]T .
Listing 1: Linear elastic HR four-noded quadrilateral element: a Matlab implementation.
for e=1:numelem
sctr = element(e,:); %element scatter vector
sctrB = [sctr sctr+numnode]; %assembly index
nn = length(sctr);

G = zeros(5,8);
H = zeros(5,5);

% compute the transformation matrix F0


% which is evaluated at the element center

pt0 = zeros(1,2);
[N,dNdxi] = lagrange_basis(elemType,pt0);
J0 = node(sctr,:)*dNdxi;

j11 = J0(1,1); j12 = J0(1,2);


j21 = J0(2,1); j22 = J0(2,2);

F0 = [j11*j11 j12*j12 j11*j12;


j21*j21 j22*j22 j21*j22;
2*j11*j21 2*j12*j22 j11*j22 + j12*j21];

D = F0 * C * F0; % C: compliance matrix

for q=1:size(W,1) % quadrature loop


pt = Q(q,:);
wt = W(q);
[N,dNdxi] = lagrange_basis(elemType,pt);
22 CONTENTS Chap. 0

J = node(sctr,:)*dNdxi;
invJ = inv(J);
dNdx = dNdxi*invJ;

% COMPUTE B MATRIX
B = zeros(3,2*nn);
B(1,1:nn) = dNdx(:,1);
B(2,nn+1:2*nn) = dNdx(:,2);
B(3,1:nn) = dNdx(:,2);
B(3,nn+1:2*nn) = dNdx(:,1);

% COMPUTE P MATRIX

P = [1 0 0 pt(2) 0;
0 1 0 0 pt(1);
0 0 1 0 0 ];

% COMPUTE G and H MATRICES

w = W(q)*det(J);

G = G + P * F0 * B * w;
H = H + P * D * P * w;
end % of quadrature loop

% COMPUTE ELEMENT STIFFNESS MATRIX

K(sctrB,sctrB) = K(sctrB,sctrB) + G * inv(H) * G;


end % of element loop

0.3.4 Numerical examples


Eigenvalue analysis Considering a single Q4 element of unit square geometry. The material
properties are E = 1, = 0.25 under a plane stress condition. Table 6 gives the eigenvalues of
the element stiffness matrices obtained with various element formulations.
Cook membrane problem is given in Fig. 2. A plane strain condition is assumed. The results
obtained with the standard Q4 element and the mixed element HR Q4 are depicted in Fig. 3.

0.4 Displacement-strain mixed finite elements


Recall from Section 0.1.7 the following equation
Sec. 0.4 Displacement-strain mixed finite elements 23

Q4 HS5-Q4 HS7-Q4 HR-Q4 DS9-Q4


1.3333 1.3333 1.3333 1.3333 1.3333
0.8000 0.8000 0.8000 0.8000 0.8000
0.8000 0.8000 0.8000 0.8000 0.8000
0.4889 0.3333 0.3879 0.3333 0.3556
0.4889 0.3333 0.3879 0.3333 0.3556
0.0000 0.0000 0.0000 0.0000 0.0000
0.0000 0.0000 0.0000 0.0000 0.0000
0.0000 0.0000 0.0000 0.0000 0.0000

Table 6: Eigenvalues of various elements. HSn-Q4 denotes the hybrid assumed stress Q4 ele-
ment with n stress parameters.

48 mm uy
16 mm

F
44 mm

E = 250 Mpa
v = 0.4999
F = 6.25 N/mm

Figure 2: Cook membrane problem.

DEFORMED DISPLACEMENT IN YDIRECTION


DEFORMED DISPLACEMENT IN YDIRECTION
7 60
60
2
6 50
50
5 1.5
40
40
4
30 30
3 1

20 20
2
0.5
10 1 10

0 0 0 0
10 0 10 20 30 40 50 60 10 0 10 20 30 40 50 60

Figure 3: Cook membrane problem: solution (vertical displacement of the upper-right corner
uy ) obtained with the HR Q4 element (left) and locking solution with the standard Q4 (right).
24 CONTENTS Chap. 0

Z
 T 
(u, ) = T T
u u d +
ext
=0 (104)

We introduce the following approximations for the displacement field, and the strain field

u = Na (105a)
= M (105b)

where a denotes the nodal displacement vector and represents the elementary strain vector. N
are the standard isoparametric shape functions and M contains interpolation functions specially
designed, for example, to get rid of locking.
The strains derived from the displacements and the stresses are then given by

u = Ba (106a)
= DM (106b)

Substitution of Eqs. (105) and (106) into Eq. (104) yields


Z Z
 T T T T T T

(u, ) = M DM a B DM a B DM d + aT NT td = 0

(107)
Let us introduce the following quantities
Z
G= BT DMd (108a)
Z
H= MT DMd (108b)

Z
f= NT td (108c)
t

From Eq. (107), one obtains the following equation applied to every element

H = GT a (109)
and the following equation applied to the whole mesh

G = f (110)
Introducing Eq. (109) into the above gives us

1 T
|GH{z G } a = f (111)
K
Sec. 0.5 Hybrid stress finite elements for linear elasticity 25

0.5 Hybrid stress finite elements for linear elasticity


Hybrid finite elements are for

improving existing conforming displacement based finite elements (coarse mesh accu-
racy, locking, distorted mesh)

improving stress solution

relaxing continuity requirements

The first hybrid element is the quadrilateral hybrid stress element developed in 1964 by Pian
[13]. An incomplete polynomial is used to approximate the stress field. The element is therefore
not invariant. Later on, in [14], the authors extended the formulation by deriving a hybrid stress
element that is based on the HR functional. For an account of the early development of hybrid
elements, we refer to [15].

0.5.1 Equilibrium stress hybrid elements


Two master fields: a stress field defined in the interior element and a displacement field
defined on the element boundary (interfacial displacement field)

The interfacial displacement field is introduced into the functional by adding a so-called
interface functional to the interior functional

Since the displacement field is only defined along the interelement boundaries, this hybrid for-
mulation can be used to construct n-sided polygonal elements (n > 4) such as the Voronoi Cell
FEM (VCFEM) developed by Ghosh and co-workers, see e.g., [5, 7] and the recent monograph
Ghosh [6].
The functional for the equilibrium stress hybrid elements (i.e., the stress field is chosen to
fulfill the equilibrium equation a priori) is then defined as

Z Z Z
1
hybrid = ij Cijklkl d + ui ij nj d + di ij nj d
2 u i
| {z } | {z }
interior functional interfacial functional
Z Z Z (112)
1
= ij Cijklkl d + ui ij nj d + di ij nj d
2 u i
| {z } | {z }
UC Wd

where the interior functional is the total complementary energy functional, see Eq. (21). In the
above, i represents the interelement boundary. Note that if the flux ti = ij nj is continuous
across i , then the interfacial functional vanishes (one boundary is traversed twice with normals
pointing in opposite directions).
Using the following identity
26 CONTENTS Chap. 0

Z Z Z Z
di ij nj d = di ij nj d di ij nj d di ij nj d (113)
i t u

one can rewrite Wd as

Z Z Z Z
Wd = ui ij nj d + di ij nj d di ij nj d di ij nj d
u
Z Z t u
(114)
= di ij nj d di ti d
t

where the boundary conditions di = ui on u and traction boundary condition were used.
Substituting Eq. (114) into Eq. (112) results in the final form of the hybrid functional
Z Z Z
1
hybrid (ij , di ) = ij Cijkl kl d + di ij nj d di ti d (115)
2 t

Remark 0.0.5. The derivation presented so far follows the lecture notes of Prof. C.A. Felipa.
In [16], another derivation of the hybrid functional given in Eq. (115) was presented. This
derivation used the total complementary energy functional, Eq. (21), plus an enforcement of
the interelement traction continuity, see Section 0.1.5.

Remark 0.0.6. The hybrid functional (115) can also be derived from the generalized HR func-
tional given in Eq. (36) when the stress field satisfies the homogeneous equilibrium equation
i.e., ij,j = 0.

Discretizations
The stress field is approximated by

= P (116)
where P is a matrix containing the interpolation functions (discussed later) and is a vector
containing the stress parameters (unknown to be determined).
The traction vector in 2D is given by

" # xx " #
nx 0 ny tx
yy = (117)
0 ny nx ty
xy
in compact notation (Eq. 116 was used)

t = |{z}
nP (118)
R

The interfacial displacement field is given by


Sec. 0.5 Hybrid stress finite elements for linear elasticity 27

d = L (119)

where L represents, for 2D problems, one dimensional shape functions; is the standard vector
of unknown nodal displacements.
By introducing Eqs. (116), (118) and (119) into the hybrid functional Eq. (115), one gets

Z  Z  Z
1
hybrid = T T
P CPd + T T
R Ld T
LT td (120)
2
| {z } | {z } | t {z }
H G f ext

Stationary of hybrid with respect to gives

hybrid
= 0 H + G = 0 = H1 G (121)

Substituting the above into Eq.(120) yields the following discrete energy in terms of nodal
displacements

1
hybrid = T GT T T ext
{z G} f
| H (122)
2
K
Stationary of the above functional with respect to the nodal displacement vector yields the
standard equation

K = f ext (123)

which is solved for the nodal displacement . Again, after condensing out the stress variables
at the element level, a displacement-like formulation has been obtained.
For convenience of computer implementation, the formulation of the hybrid stress element
is summarized in Box 0.2. A Matlab implementation of this Box is described later.

Construction of P matrix

Stress field is in equilibrium

Stiffness matrix K must not be rank deficient

K should be geometrically isotropic or rotationally invariant

Dimension of P is small as much as possible (performance issue).


28 CONTENTS Chap. 0

Box 0.2 Hybrid stress elements

P = P(x, y)
" #
nx 0 ny
n=
0 ny nx
Z
He = PT CPd (n n )
Ze
Ge = RT Ld, R = nP (n nd )
e
Ke = GT 1
e He Ge (nd nd )
X
K= Ke
e
ext
Kd = f
= PH1
e Ge e

n is the number of stress variables per element; nd denotes the number of nodal displacement
values per element.

A hybrid Q4 element
As a concrete example, we consider a four node quadrilateral element. In case of zero body
force, the P matrix can be chosen to be

1 0 0 y 0

P = 0 1 0 0 x (124)
0 0 1 0 0
which results in the first hybrid element in literature presented in Pian [13]. Note that this
element is, however, not frame invariant. This element is best suited for illustration purposes.
Considering an edge that connects node i having coordinates (xi , yi ) with node i + 1 having
coordinates (xi+1 , yi+1 ). By defining the following quantities
p
x = xi+1 xi , y = yi+1 yi , l= x2 + y 2 (125)
then, the unit normal vector to this edge is given by n = (y/l, x/l). Then, the R matrix
in Eq. (118) is given by

" # " #
nx 0 ny ynx 0 1 y 0 x yy 0
R = nP = = (126)
0 ny nx 0 xny l 0 x y 0 xx

The displacement field along an element edge connecting two nodes i and i + 1 is given by
Sec. 0.5 Hybrid stress finite elements for linear elasticity 29

x

" # i
y
1 s/l 0 s/l 0 i
d= x , 0 s 1 (127)
0 1 s/l 0 s/l i+1
yi+1
| {z }
L
Hence, RT L on this edge is given by

(1 s/l)y 0 s/ly 0

0 x(1 s/l) 0 xs/l
1
RT L = x(1 s/l) y(1 s/l) xs/l ys/l (128)
l


yy(1 s/l) 0 yys/l 0
0 xx(1 s/l) 0 xxs/l
Next, we write x, y (defined on the edge) in terms of s using

x = (1 s/l)xi + s/lxi+1
(129)
y = (1 s/l)yi + s/lyi+1
Finally the line integralR on the element edge (to compute matrix G) joining two nodes i and
1
i + 1 can be written as 0 RT Lds. This integral is evaluated exactly and the result is given by 1

ay 0 by 0

0 ax 0 bx
1
RT L = ax ay bx by (130)
l

ey 0 fy 0
0 ex 0 fx
where

ay = y 0.5/ly
ax = x + 0.5/lx
bx = 0.5/lx
by = 0.5/ly
  (131)
ey = y (1 1/l + 1/(3l2 ))yi + (0.5/l 1/(3l2 ))yi+1
 
fy = y (0.5/l 1/(3l2 ))yi + (1/(3l2 ))yi+1
 
ex = x (1 1/l + 1/(3l2 ))xi + (0.5/l 1/(3l2 ))xi+1
 
fx = x (0.5/l 1/(3l2 ))xi + (1/(3l2 ))xi+1
1
For more complex stress interpolations P, the symbolic mathematics package Maple or Mathematica should
be used.
30 CONTENTS Chap. 0

As the final step, the G matrix for an element e is obtained from the assembly of the contribu-
tions of its edges
Z nedge Z
X
T
Ge = R Ld = (RT L)se d (132)
e s=1 se

where nedge denotes the number of edges for an element, se is the edge s of element e. The
assembly of Gse to Ge is performed by considering the element e as a separate mesh and each
edge is a two-noded line element in this mesh.

Remark 0.0.7. For complex stress interpolations, the line integrals along the element bound-
aries have to be evaluated numerically.

Matlab implementation of this hybrid Q4 element is given in List 2. This implementation


employs the stress interpolation matrix evaluated in a scaled local coordinate system to improve
the conditioning number of H matrix (see next section for details). Besides, it implements a
numerical integration rule to evaluate G matrix. Note that when the element edge displacement
is given by Eq. (127) then the element nodal displacements are placed in this order =
[x1 y1 x2 y2 . . . xn yn ]T . This is in contrast to the implementation of mixed and enhanced
strain formulations given in this manuscript where = [x1 x2 . . . xn y1 y2 . . . yn ]T .

Listing 2: Linear elastic assumed stress hybrid Q4 element: a Matlab implementation.


U = zeros(2*numnode,1); % nodal displacement vector
f = zeros(2*numnode,1); % external load vector
K = zeros(2*numnode,2*numnode); % stiffness matrix

H = zeros(5,5);

GL = [1 2 3 4; % used to assemble G_e^s into G_e


3 4 5 6;
5 6 7 8;
7 8 1 2];

%%%%%%%%%%%% COMPUTE STIFFNESS MATRIX %%%%%%%%%%%%%%

[W,Q] = quadrature(2,GAUSS,2); %2x2 Gaussian quadrature


[W1,Q1] = quadrature(2,GAUSS,1); %2 Gaussian quadrature (1D)

for e=1:numelem % start of element loop


sctr = element(e,:);
nn = length(sctr);

for k = 1 : nn
sctrB(2*k-1) = 2*sctr(k)-1;
Sec. 0.5 Hybrid stress finite elements for linear elasticity 31

sctrB(2*k) = 2*sctr(k) ;
end

% compute the center coordinate

xe = node(sctr,:);
xc = mean(xe(:,1));
yc = mean(xe(:,2));
linv = 1/abs(xc - xe(1,1));

H = 0;
G = zeros(5,8);

% COMPUTE H MATRIX

for q=1:size(W,1) % quadrature loop


pt = Q(q,:);
wt = W(q);
[N,dNdxi] = lagrange_basis(elemType,pt);
J = node(sctr,:)*dNdxi;

ptX = node(sctr,1) * N;
ptY = node(sctr,2) * N;

% COMPUTE P MATRIX

P = [1 0 0 (ptY-yc)*linv 0;
0 1 0 0 (ptX-xc)*linv;
0 0 1 0 0 ];

H = H + P * C * P * wt * det(J);
end % of quadrature loop

% COMPUTE G MATRIX, line integral

sctrE = [sctr sctr(1)];

for ie = 1:4 % loop on element edges


sctrBe = [sctrE(ie) sctrE(ie+1)];
sctrBE = GL(ie,:);

xyE = node(sctrBe,:);
32 CONTENTS Chap. 0

% compute the normal vector

deltaX = xyE(2,1) - xyE(1,1);


deltaY = xyE(2,2) - xyE(1,2);
le = sqrt(deltaX*deltaX + deltaY*deltaY);
n = (1/le) * [deltaY 0 -deltaX;0 -deltaX deltaY];

Gs = zeros(5,4);

for gp=1:size(W1,1)
pt = Q1(gp,:);
wt = W1(gp);
[N,dNdxi] = lagrange_basis(L2,pt);
J = dNdxi * xyE;

ptX = xyE(:,1) * N;
ptY = xyE(:,2) * N;

% COMPUTE P MATRIX

P = [1 0 0 (ptY-yc)*linv 0;
0 1 0 0 (ptX-xc)*linv;
0 0 1 0 0 ];

L = [N(1) 0 N(2) 0;
0 N(1) 0 N(2)];

Gs = Gs + P * n * L * wt * norm(J);
end
G(:,sctrBE) = G(:,sctrBE) + Gs;
end

% COMPUTE ELEMENT STIFFNESS MATRIX

K(sctrB,sctrB) = K(sctrB,sctrB) + G * inv(H) * G;


end % of element loop
%%%%%%%%% END OF STIFFNESS MATRIX COMPUTATION %%%%%%%%%%%%%%

In order to validate the implementation, a rectangular plate of dimension 10 5 mm2 in


tension is studied. The left edge is fixed in both directions. Horizontal forces at right nodes are
6.25 N. Young modulus of 250 MPa and a zero Poisson ratio. Plane strain condition. Result in
Fig. (4). Fig. (5).
Sec. 0.5 Hybrid stress finite elements for linear elasticity 33

DEFORMED DISPLACEMENT IN XDIRECTION


DEFORMED DISPLACEMENT IN XDIRECTION
12 1
5
0.9
4 10 0.25
0.8
3
0.7 0.2
2 8
0.6
1
6 0.5 0.15
0
0.4
1
4 0.1
0.3
2
0.2
3 2 0.05
0.1
4
0 0
0 2 4 6 8 10 12 0 0.2 0.4 0.6 0.8 1

Figure 4: A rectangular plate in tension. Comparison between results obtained with the hybrid
formulation (left) and standard Q4 elements (right).
DEFORMED STRESS PLOT, SIGMA XX DEFORMED STRESS PLOT, SIGMA XX
1
5 13.4
0.9 40
4 13.2
0.8
3 13 30
0.7
2 12.8
0.6 20
1 12.6
0.5
0 12.4 10
0.4
1 12.2
0
0.3
2 12
0.2 10
3 11.8
0.1
4 11.6 20
0
0 2 4 6 8 10 12 0 0.2 0.4 0.6 0.8 1

Figure 5: A rectangular plate in tension. Comparison between results obtained with the hybrid
formulation (left) and standard Q4 elements (right).

0.5.2 Homogeneous Voronoi cell element


The homogeneous Voronoi cell element was first presented in Ghosh and Mallett [5]. The
characteristics of this element are as follows

A n-sided (4 n 10) convex polygonal element;

Based on the hybrid functional given in Eq.(115);

Stress interpolation P derived from Airys stress functions;

Domain integrals are achieved with a subdivision of the element into sub-domains.

The last item reduces the efficiency of the method. It is desirable to have an integration method
without subdivision of the element domain. Some common Voronoi cell elements are depicted
34 CONTENTS Chap. 0

Figure 6: A continuum discretized into a set of Voronoi cells.

Figure 7: A Voronoi cell element: 6-sided element (left) and 7-sided element (right). Filled
circles are the nodes. Note that more nodes can be placed along the boundary.

in Fig. (7).

Stress interpolations From the Airys stress function (x, y), the stress field is given by (see
e.g., [4])

2 (x, y) 2 (x, y) 2 (x, y)


xx = , yy = , xy = (133)
y 2 x2 xy
For example, with a third-order Airys stress functions, the stresses are defined as

1
xx 1 y 0 0 0 x 0
2
yy = 0 0 1 x 0 0 y .. P
(134)
.
xy 0 0 0 0 1 y x
7

A complete linear polynomial chosen for the stress interpolation ensures that the element is
frame invariant.
Sec. 0.5 Hybrid stress finite elements for linear elasticity 35

In order to avoid ill-conditioned H matrix (for elements locating far from the global coor-
dinate origin, x, y 1, thus H contains terms that are of different orders of magnitude), the
stress interpolation functions should be expressed in a scaled local coordinate system.

x
(xc , yc )
y

Figure 8: Scaled local coordinate system to construct the stress interpolations.

The scaled coordinate system, (x, y), is located at the center, denoted by xc = (xc , yc ), of a
VC element. The scaling is such that 1 x, y 1.
x xc y yc
x= , y= (135)
l l
where l is the characteristic length of a Voronoi cell element that is defined as

l = max ||x xc || , x e (136)


In terms of the local coordinate system, the stress interpolation given in Eq. (134) now reads

1 y 0 0 0 x 0

P = 0 0 1 x 0 0 y (137)
0 0 0 0 1 y x
Numerical integration Integrals over an element domain is computed by dividing the element
domain into a number of quadrilaterals, see Fig. (9). Thus, for a scalar function f (x, y), we
write

Z XZ
f (x, y)d = f (x, y)d
e quad
e
XZ (138)
= f (x(, ), y(, ))Jd


where J denotes the Jacobian of the isoparametric mapping from the parent domain  to quad
e .
Integral on the parent domain is performed in a standard manner i.e., using a Gaussian quadra-
ture.
36 CONTENTS Chap. 0


(+1, +1)

quad
e 

(1, 1)

Figure 9: Element subdivision of numerical domain integration. Black filled circles denote the
nodes while white filled circles are the vertices of quadrilaterals used for numerical integration
purpose.

0.5.3 Composite Voronoi cell element


Composite Voronoi cell elements have been developed by Ghosh and Moorthy [7]. They are
n-gon elements with arbitrary inclusions (or voids) as shown in Fig. 10a).
e
eM

eI ne
eI

nI

Figure 10: A composite Voronoi cell element. Boundary nodes and interface nodes are denoted
by filled circles and squares, respectively.

Element functional The composite Voronoi cell element is based on a modification of the
hybrid functional given in Eq. (115). For one element (Voronoi cell element as shown in Fig.
10b), the internal energy is given by

Z Z Z Z
1 T 1 T
eint = CM M d CI I d + dT
M td + dT e e
I (tM tI )d (139)
eM 2 M eI 2 I e eI

where the last term in the above is to enforce the traction continuity at the matrix/inclusion
interface eI . The stresses in the matrix and inclusion are denoted by M and I , respectively.
The elastic compliance matrices of the matrix and inclusion are represented by CM and CI . The
boundary displacement (defined on e ) is denoted by dM and dI is the matrix/inclusion interface
displacement.
Sec. 0.5 Hybrid stress finite elements for linear elasticity 37

The matrix and inclusion stress fields are approximated by

M = PMM , I = PII (140)

where PM and PI are the stress interpolations for the matrix phase and inclusion phase, respec-
tively. In 2D, PM is a 3 nM I M/I
matrix and PI is a 3 n matrix where n denotes the number
of stress parameters for the matrix/inclusion phase.
The displacement fields of the element boundary and the matrix/inclusion interface are ap-
proximated by

dM = LM M , dI = LI I (141)

where M , a vector of length ned , is the vector of nodal displacement and I , a vector of length
nId , is the vector of nodal displacement on the matrix/inclusion interface.

Remark 0.0.8. In case of circular or ellipse inclusions, a large number of nodes must be dis-
tributed on the matrix/inclusion interface to properly capture the geometry. As seen later, an
increase in the number of interface nodes leads to a higher number of stress parameters. This
increases the computational cost. B-spline and/or NURBS can be employed to represent the
inclusion geometry with a much lower number of nodes.

Substituting of Eqs. (140) and (141) into Eq. (139) yields

1 T 1
eint = M HM M IT HI I + M
T T
GM M + M GIM I IT GII I (142)
2 2
with

Z Z
HM = PT
M CM PM d, HI = PTI CI PI d
eM e
I
Z Z
GM = LT
M ne PM d, GII = LTI nI PI d (143)
e eI
Z
GIM = LT
I nI PM d
eI

The total energy of element e reads

e = eint eext , eext = T


Mf
ext
(144)

where f ext is the standard external force vector.

Discrete equations Stationary of e with respect to M and I gives us


38 CONTENTS Chap. 0

e
= 0 : HM M + GM M + GIM I = 0
M
(145)
e
= 0 : HI I GII I = 0
I

or in matrix notation
" #" # " #" #
HM 0 M GM GIM M
= (146)
0 HI I 0 GII I

which allows for the determination of the stress parameters upon the availability of the displace-
ments
" # " #1 " #" #
M HM 0 GM GIM M
= (147)
I 0 HI 0 GII I

Stationary of e with respect to M and I gives us

e
= 0 : GT
M M f
ext
=0
M
(148)
e
= 0 : GT T
IM M GII I = 0
I

or in matrix notation
" #" # " #
GT
M 0 M f ext
= (149)
GIM GT
T
II I 0

Substituting Eq. (147) into Eq. (149) yields the following system of equations to determine the
nodal displacements
" #" #1 " #" # " #
GT
M 0 HM 0 GM GIM M f ext
= (150)
GT T
IM GII 0 HI 0 GII I 0
| {z }
K
where K is the stiffness matrix which is given by
" #
GT 1
M HM GM GT 1
M HM GIM
K= (151)
GT T T 1
IM HM GM GIM HM GIM + GII HI GII
1 1
Sec. 0.6 Hybrid stress finite elements for inelastic materials 39

Stress interpolations

Displacement interpolations

Numerical integration

Eliminating rigid body modes at inclusion interfaces

0.6 Hybrid stress finite elements for inelastic materials


0.6.1 Variational principle and discrete equations
For inelastic materials, it is appropriate to write the hybrid functional in Eq. (115) as follows
Z Z Z
= W (ij )d +

di ti d di ti d (152)
t

where W (ij ) is the so-called complementary strain energy function.


The first variation of the above functional is given by
Z Z Z Z
T T T
= d + t dd + d td dT td (153)
t

which is elaborated as follows after the introduction of the stress approximation, Eq. (116), the
traction approximation Eq. (118), and interfacial displacement approximation, Eq. (119)

Z
 Z 
T T T T
= P d + R Ld

Z  Z  (154)
T T T T
+ L Rd L td
t

Setting this variation to zero with respect to and yields the system of equations

Z Z 
T T
P d = R Ld e , e = 1, . . . , ne (155)
e e
Z  Z
T
L Rd = LT td (156)
t

where by the virtue of the discontinuity of the stress parameters , the first system of equations
apply for every single element e (ne denotes the number of elements). The second system of
equation is a global one.
40 CONTENTS Chap. 0

0.6.2 Solution procedure


Equation (155) is rewritten as follows
Z Z 
T T
P ((e ))d R Ld e = 0 (157)
e e

where the nonlinear dependence of the strains in the stress parameters was explicitly written.
This system of nonlinear algebraic equations is solved for the element stress parameters itera-
tively using the Newton-Raphson method.
At iteration i + 1 from iteration i with i , we have the following linearized equation

Z Z  Z 

T T

P (ei )d T
R Ld e + P d e = 0 (158)
e e e
or
Z  Z  Z
T i T
P S Pd e = R Ld e PT (ei )d (159)
| e {z } | e {z } e

Hie (n n ) Ge (n nd )

with S =
. The procedure to compute Si and ( i ) given i is described later. This
procedure is achieved using a strain update algorithm.
The element stress parameters are updated according to

ei+1 = ei + e (160)
the process is repeated until a convergence criterion is satisfied.
Equation (156) is rewritten as follows
ne
X
GT
e e f
ext
=0 (161)
e=1

where f ext is the standard external force vector. This equation is linearized around the iteration
k as follows
ne ne  k
X X
GT k
e e f ext
+ GT
e e = 0 (162)
e=1 e=1
e


which, after substituting Eq. (159) for
, reads
ne
X ne
X

GT 1
e He Ge e = f ext
GT e e
k
(163)
| {z } | {z }
e=1 e=1
Ke feint
where Ke is the element stiffness matrix and feint denotes the element internal nodal force vector.
Sec. 0.7 Hybrid stress finite elements for inelastic materials 41

After solving Eq. (163) for the nodal displacement increments , the nodal displacements
are updated according to

k+1 = k + (164)

the process is repeated until a convergence criterion is satisfied.


For convenience of computer implementation, the overall procedure of the hybrid formu-
lation for inelastic materials is given in Box 0.3. In this box, 0 denotes values at the previous
converged load increment.

Box 0.3 Nonlinear hybrid elements.


1. Input: global nodal displacement vector 0 , elementary stress params e0 , external force
f ext

2. Loop over elements

(a) Compute Ge
(b) Compute e
i. Initialization: He = 0, fe2 = 0, e = e0
ii. Compute He by Gaussian quadrature
A. i xi xi P(i ) (i ) = P(i )e
B. Compute (i ), S(i ) corresponding to (i )
C. fe2 PT (i )(i )wi
D. He PT (i )S(i )P(i )wi
2
iii. Solution: = H1 e (Ge e fe )
iv. Update: e
v. Check convergence: yes proceed to 2c; go back to A otherwise
(c) Compute the element stiffness matrix Ke = GT
e He Ge
1

(d) Compute the element internal force vector feint = GT


e e
P P
3. Assembly: K = e Ke , f int = e feint

4. Solution: = K1 f ext f int

5. Update displacement:

6. Check convergence: yes, continue to next load increment. No, go back to step 2b.
42 CONTENTS Chap. 0

0.7 Enhanced assumed strain elements


The enhanced assumed strain (EAS) method, which is a generalization of the so-called incom-
patible modes elements [22] and pioneered by Simo and Rifai [20], is based on the following
ingredients

The Veubeke-Hu-Washizu variational principle given in Eq. (42);

Additive decomposition of the strain field into two parts namely a compatible part and an
incompatible part which is called the enhanced strain field;

The stress field is eliminated from the three-field functional on the basis of an orthogonal-
ity condition. The resulting functional involves therefore only two fields (displacements
and enhanced strains);

The enhanced strain parameters can be condensed out at the element level making the
final form very similar to the standard displacement formulation finite elements. This
makes EAS elements very appealing for inelastic materials. This is in sharp contrast to
HR-based elements.

In [19] an EAS for geometrically nonlinear problems has been presented. A review of
early developments of the EAS formulation has been given in [24]. In this section, the EAS
formulation is presented for linear elasticity.

0.7.1 Assumed strain field


The idea is to add extra strain components to the compatible strains in order to balance the
parasitic strains

Bu + |{z}
= |{z} e (165)
compatible enhanced
e
where is an independent incompatible strain field. Note that this incompatible strain field is
not subject to any interelement continuity requirement.
Substituting Eq. (165) into the VHW functional, Eq. (42), yields
Z  
e 1 T T e
(u, , ) = D d U ext (166)
2
An important step is to make the enhanced strain field orthogonal to the stress field through the
following orthogonality condition
Z
T e d = 0 (167)

so that the stresses are eliminated from the original functional and a new simplified two-field
functional is obtained
Sec. 0.7 Enhanced assumed strain elements 43

Z  
e 1 e T e
(u, ) = [Bu + ] D [Bu + ] d U ext (168)
2

0.7.2 Discrete system of equations


In addition to the standard approximation of the displacement field u = N()d, the enhanced
strain field is approximated by

e = G (169)
where denotes the enhanced strain degrees of freedom that are local to the elements and G is
the strain interpolation matrix.
Stationary of the functional given in Eq. (168) with respect to d and yields the following
system of equations
! ! !
K LT d f ext
= (170)
L H 0
with the matrices given by

Z
K= BT DBd (displacement stiffness matrix)
Z
L= GT DBd (coupling matrix) (171)
Z
H= GT DGd (strain stiffness matrix)

where K is the standard stiffness matrix i.e., the stiffness matrix in the displacement based FE
formulation. Since D is positive definite and G is constructed with an assumption that its rows
are linear independent, the matrix H is positive definite.
The enhanced strain degrees of freedom can be eliminated from the above equation by a
static condensation procedure. Thus,

(K LT H1 L)d = f ext (172)


It can be recognized from the above equation that

basic features of a displacement based FE formulation retain. This makes EAS approach
more attractive for strain-driven nonlinear material formulations like plasticity than a hy-
brid FE formulation.

the fact that the stiffness matrix equals the standard stiffness matrix (K) minus another
matrix makes the element less stiff => locking free formulation.
44 CONTENTS Chap. 0

0.7.3 Construction of G matrix


Orthogonality condition;

Columns of G are linear independent;

Stability (enhanced strains must be different from compatible strains);

Efficiency: the number of enhanced strain parameters as less as possible.

The interpolation matrix G must be constructed in such a manner that the orthogonality
condition Eq. (167) is fulfilled. In order to pass the patch test (a constant stress field should be
reproduced), the orthogonality condition is enforced only for a constant stress field . So,
Z Z Z
T e e
d = 0 d = 0 Gd = 0 (173)

The idea is now to build G in terms of G() in the following manner [20]

J0 1
G= F G() (174)
J 0
where J denotes the determinant of the Jacobian of the isoparametric mapping and J0 is that
determinant evaluated at the element center. In the above the transformation matrix F0 is,
according to tensor calculus, used to map G() to G defined in the global frame. It is given by

2
J11 J12 J12 J11 J12
2
F0 = J21 J21 J22 J21 J22 (175)
2J11 J21 2J12 J22 J11 J22 + J12 J21 =0

for two dimensions. Note that the transformation matrix is evaluated at the origin so that its
components are constant and the polynomial order of G() is therefore not increased.
The orthogonality condition Eq. (173) now becomes
Z
G()d = 0 (176)


where  denotes the parent domain (where numerical integration is performed).

Remark 0.0.9. In nonlinear solid mechanics especially for finite deformation problems, low
order finite elements (Q4 elements in 2D and H8 elements in 3D) are in favor. It is emphasized
that the enhanced strain approach does not work (no improvement is gained) for constant strain
elements (three-noded triangular elements). Therefore, most of EAS formulations (and HR-
based formulations) have been devoted to Q4 and H8 elements.

Box 0.4 summarizes the key equations of the enhanced strain method for linear elasticity
problems.
Sec. 0.7 Enhanced assumed strain elements 45

Box 0.4 Enhanced strain elements for linear elasticity

J0 1
G= F0 G()
ZJ
Ke = BT DBd (nu nu )
Z e
He = GT DGd (n n )
Z e
Le = GT DBd (n nd )
e

Ke = Ke LT 1
e He Le (nd nd )
X
K= Ke
e
ext
Kd = f

Q1E4 element (bilinear quadrilateral with 4 EAS parameters)


This is the most simple EAS element with four enhanced strain modes. The enhanced strain
interpolation matrix is given by [20]

0 0 0

G() = 0 0 0 (177)
0 0
This element is free of volumetric and shear locking. To show this, one has to inspect the
deformation modes of this element given in Table 7.

Volumetric locking is the phenomenon in which a finite element is unable to model volume
preserved motions (isochoric motions) which are occured in nearly incompressible media and
in plastically yielded materials. The predict numerical displacement field is virtually zero in
such cases. Isochoric motion means + = constant. From Table 7, we have for the standard
Q4 elements

+ = c3 + c6 + c7 + c8 (178)
which is vanished only for c7 = c8 = 0 i.e., the element locks. For the Q1E4 elements

+ = c3 + c6 + (c7 + 2 ) + (c8 + 1 ) (179)


and the condition ii = constant becomes

c3 + c6 = constant, c7 + 2 = 0, c 8 + 1 = 0 (180)
46 CONTENTS Chap. 0

which is satisfied with non-zero displacements i.e., no volumetric locking occurs.

Shear locking is present in a pure bending state with a non-zero shear stress. In order to show
that the Q1E4 element is shear locking free, let us consider the shear strain given by

= c4 + c5 + (c7 + 3 ) + (c8 + 4 ) (181)


It can be observed that for 3 = c7 , is then zero in the direction. Note that c7 and c8
represent pure bending modes.

mode c1 c2 c3 c4 c5 c6 c7 c8 1 2 3 4
u= 1 0 0 0 0
v= 0 1 0 0 0
0 0 c3 0 0 0 c7 0 1 0 0 0
0 0 0 0 0 c6 0 c8 0 2 0 0
0 0 0 c4 c5 0 c7 c8 0 0 3 4

Table 7: Deformation modes of a four-noded quadrilateral element with 4 enhanced modes.

Numerical integration is achieved with a four point Gauss rule. The Matlab implementation
of the Q1E4 element is given in Box 3. In order to validate this implementation, the Cook
membrane problem as described in Fig. 2 is considered. The result is given in Table 8.

Listing 3: Linear elastic Q1E4 element: a Matlab implementation.


for e=1:numelem
sctr = element(e,:);
sctrB = [sctr sctr+numnode];
nn = length(sctr);

Ke = zeros(8,8);
Le = zeros(4,8);
He = zeros(4,4);

% compute the transformation matrix F0


% which is evaluated at the element center

pt0 = zeros(1,2);
[N,dNdxi] = lagrange_basis(elemType,pt0);
J0 = node(sctr,:)*dNdxi;

j11 = J0(1,1); j12 = J0(1,2);


j21 = J0(2,1); j22 = J0(2,2);
Sec. 0.7 Enhanced assumed strain elements 47

F0 = [j11*j11 j12*j12 j11*j12;


j21*j21 j22*j22 j21*j22;
2*j11*j21 2*j12*j22 j11*j22 + j12*j21];

j0f0Inv = det(J0) * inv(F0);

for q=1:size(W,1) % quadrature loop


pt = Q(q,:);
wt = W(q);
[N,dNdxi] = lagrange_basis(elemType,pt);
J = node(sctr,:)*dNdxi;
invJ = inv(J);
dNdx = dNdxi*invJ;

% COMPUTE B MATRIX
B = zeros(3,2*nn);
B(1,1:nn) = dNdx(:,1);
B(2,nn+1:2*nn) = dNdx(:,2);
B(3,1:nn) = dNdx(:,2);
B(3,nn+1:2*nn) = dNdx(:,1);

% COMPUTE G MATRIX

G = [pt(1) 0 0 0;
0 pt(2) 0 0;
0 0 pt(1) pt(2) ];

% COMPUTE K, L and H MATRICES

j = det(J);
w = W(q)*j;

Gxi = (1/j) * j0f0Inv * G;

Ke = Ke + B * D * B * w;
Le = Le + Gxi * D * B * w;
He = He + Gxi * D * Gxi * w;
end % of quadrature loop

% COMPUTE ELEMENT STIFFNESS MATRIX

K(sctrB,sctrB) = K(sctrB,sctrB) + Ke - Le * inv(He) * Le;


48 CONTENTS Chap. 0

end % of element loop

mesh Q4 Q1E4 HR-Q1S5 H4


44 2.0829 6.8436 6.8466 6.5353
88 2.1401 7.4301 7.4303 7.2348
16 16 2.3114 7.6363 7.6363 7.5277
25 25 2.5849 7.6939 7.6939 7.6295
30 30 2.7601 7.7092 7.7092 7.6584

Table 8: Cook membrane problem: convergence of the vertical displacement of the upper-right
corner obtained with standard Q4, Q1E4, the mixed Q4 with 5 stress variables and the assumed
stress hybrid Q4 with 5 stress parameters.

Q1E7 element (bilinear quadrilateral with 7 EAS parameters)


The strain interpolation matrix for the Q1E7 element is given by [1]

0 0 0 0 0

G() = 0 0 0 0 0 (182)
0 0 0 0
The numerical tests show that for a rectangular or parallelogram geometry, Q1E7 and Q1E4 are
identical. The bilinear terms in Eq. (182) improve only the bending behaviour of a distorted
element. Numerical integration is achieved with a four point Gauss rule.

Remark 0.0.10. In [1], it has been stated that the HR-based elements are equivalent to EAS
elements if the trial stresses in the HR elements are complementary to the trial strains in the
EAS formulation. Table 9 illustrates such an equivalence.

Pian-Sumihara EAS-7
Interpolation of stresses interpolation of strains
11 = 1 + 4 11 = 1 + 5
22 = 2 + 5 22 = 2 + 6
12 = 3 12 = 3 + 4 + 7

Table 9: Equivalence of Pian-Sumihara elements and a EAS-7 element.


Sec. 0.8 EAS for small strain inelastic materials 49

H1E7 element (trilinear hexahedral with 7 EAS parameters)

0.7.4 Stress recovery


The stresses are computed in a standard manner from the strains Eq. (165). Although
this approach is simple it is not variationally consistent in the sense that the orthogonality
condition is not satisfied.
A variationally consistent stress recovery as given in [20], see also [1].

0.8 EAS for small strain inelastic materials


For inelastic materials, in Eq. (??) we do not substitute ij by an constitutive relation. This
results in the following
Z
 u 
VHW = (ij ij ) (ij ij ) + ij ij d U ext (183)

By using ij = uij + eij and the orthogonality, Eq. (167), the above reduces to
Z Z
   
VHW = ij (uij + eij ) d U ext = (u + e )T d U ext (184)

where in the second equality, we have switched to matrix notation which is convenient for the
subsequent derivations.
The variation of the compatible and incompatible strains are given by

u = Ba, e = G (185)
Substituting the above into Eq. (184) yields
Z
 T T 
VHW = (a B + T GT ) d aT f ext (186)

The equation VHW = 0 then gives us
Z Z
T
B d = f , ext
GT d = 0 (187)

| {z } | {z }
int eas
f f
The above system of nonlinear algebraic equations is solved with the Newton-Raphson method.
To this end, the above is linearized around the state (ak , k ) as follows

Z  Z

T T
int ext

f f k+ B d a + B d = 0
a k
Z
 Z
k (188)

fkeas + GT d a + GT d = 0
a k k
50 CONTENTS Chap. 0

The stress-strain relation is cast into a rate form as

= D = D(u + e ) = D(Ba + G) (189)


Equation (188) becomes (subscript k skipped)

Z  Z 
T T
B DBd a + B DGd = f ext f int
Z  Z  (190)
T T
G DBd a + G DGd = f eas

Putting the above in matrix notation, we have


" #" # " #
K LT a f ext f int
= (191)
L H f eas
where the matrices K, H and L are given in Eq. (171). Again, a static condensation procedure
is carried out at the element level to eliminate the strain parameters . This is done as follows.
From Eq. (191)2 , we can write

= H1 (f eas + La) (192)


Substituting the above into Eq. (191)1 yields
  
K LT H1 L a = f ext f int LT H1 f eas (193)
| {z } | {z }
int
K f
In addition to the computation of standard terms such as K, f int and f ext , one has to compute
LT H1 L and LT H1 f eas . Then, the modified element stiffness matrix K and modified internal
force vector f int are computed. Again, the bottleneck of the method is the inversion of H, a
n n matrix.
For ease of computer implementation, Box 0.5 summarizes the solution procedure for the
small strain EAS formulation applicable to inelastic materials. Note that this is a straightforward
implementation in a sense that efficiency was not taken into account. A Matlab implementa-
tion for the Q1E4 elements with an isotropic linear elastic material implemented in a nonlinear
setting is given in List 4. This describes how a nonlinear EAS element can be coded without
delving into the complexity of inelastic constitutive models such as plasticity. This implementa-
tion can be extended without major difficulties to plastic material models. The only modification
lies in the computation of stresses from strains.

Remark 0.0.11. An efficient implementation of the EAS formulation has been presented in
Simo et al. [21]. This implementation avoids the storage of matrices required in the static
condensation of the strain parameters. It is very similar to the solution procedure of the assumed
stress hybrid formulation described in Section 0.6.2.
Sec. 0.8 EAS for small strain inelastic materials 51

Box 0.5 EAS for material nonlinearity: solution procedure.


1. Newton-Raphson iterations for a given load step f ext

(a) Set initial guess: a = a0 , = 0


(b) Using a, to compute stresses at all Gauss points
(c) Compute element matrices
Z Z Z
Ke = BT DBd, Le = T
B DGd, He = GT DGd
e e e

(d) Compute element vectors


Z Z
T
feint = B d, feeas = GT d
e e

(e) Compute modified tangent stiffness matrix and internal force vector

Ke = Ke LT 1
e He Le

feint = feint LT 1 eas


e He fe

(f) Assemble and solve for nodal displacement increment


n
X n
X
int
K= Ke , f = feint
e=1
ext int
 e=1
a = K 1
f f

(g) Update displacement


a = a + a
(h) Update enhanced strain parameters (element level)
eas
e = e H1
e (fe + Le ae )

(i) Check convergence: no go back to 1b


eas

H1
e , Le and fe have to be stored at element level.

Listing 4: Linear elastic Q1E4 element in a nonlinear framework: an Matlab implementation.

% DATA STRUCTURES

u = zeros(2*numnode,1);% nodal displacement vector


52 CONTENTS Chap. 0

du = zeros(2*numnode,1);% nodal increment disp. vector


a = zeros(4,numelem); % element stress parameters alpha
fext = zeros(2*numnode,1);% external force vector
fint = zeros(2*numnode,1);% internal force vector
fea = zeros(4,1); % enhanced strain force vector
fie = zeros(8,1); % internal force vector per element
fe = zeros(4,numelem); % enhanced strain force for whole mesh
K = sparse(2*numnode,2*numnode); % stiffness matrix

invHE = zeros(4,4,numelem);
LE = zeros(4,8,numelem);

F0 = zeros(3);
Ke = zeros(8,8);
Le = zeros(4,8);
He = zeros(4,4);

% NEWTON-RAPHSON ITERATIONS

while (rnorm > epsilon)

[W,Q]=quadrature(2, GAUSS, 2);% 2x2 Gaussian quadrature

for e=1:numelem
sctr = element(e,:);
sctrB = [sctr sctr+numnode];
nn = length(sctr);

Ke = 0;
Le = 0;
He = 0;

fie = 0;
fea = 0;

% compute the transformation matrix F0


% which is evaluated at the element center

pt0 = zeros(1,2);
[N,dNdxi] = lagrange_basis(elemType,pt0);
J0 = node(sctr,:)*dNdxi;

j11 = J0(1,1); j12 = J0(1,2);


Sec. 0.8 EAS for small strain inelastic materials 53

j21 = J0(2,1); j22 = J0(2,2);

F0 = [j11*j11 j12*j12 j11*j12;


j21*j21 j22*j22 j21*j22;
2*j11*j21 2*j12*j22 j11*j22 + j12*j21];

j0f0Inv = det(J0) * inv(F0);

% GET ELEMENT DISP AND ENHANCED STRAIN PARAMS


% FROM THE GLOBAL VECTOR/MATRIX

ue = u(sctrB);
ae = a(:,e);

for q=1:size(W,1) % quadrature loop


pt = Q(q,:);
wt = W(q);
[N,dNdxi] = lagrange_basis(elemType,pt);
J = node(sctr,:)*dNdxi;
invJ = inv(J);
dNdx = dNdxi*invJ;

% COMPUTE B MATRIX

B = zeros(3,2*nn);
B(1,1:nn) = dNdx(:,1);
B(2,nn+1:2*nn) = dNdx(:,2);
B(3,1:nn) = dNdx(:,2);
B(3,nn+1:2*nn) = dNdx(:,1);

% COMPUTE G MATRIX

G = [pt(1) 0 0 0;
0 pt(2) 0 0;
0 0 pt(1) pt(2) ];

% ************************************
% COMPUTE K, L and H MATRICES
% ************************************

j = det(J);
w = W(q)*j;
54 CONTENTS Chap. 0

Gxi = (1/j) * j0f0Inv * G;

% COMPUTE STRAIN

eps = B * ue + Gxi * ae;

% COMPUTE STRESS

sigma = D * eps;

% COMPUTE STISSNESS and FORCE VECTORS

Ke = Ke + B * D * B * w;
Le = Le + Gxi * D * B * w;
He = He + Gxi * D * Gxi * w;

fie = fie + B * sigma * w ;


fea = fea + Gxi * sigma * w;

end % of quadrature loop

% COMPUTE ELEMENT STIFFNESS MATRIX

invHe = inv(He);

invHE(:,:,e) = invHe; % store these matrices at element level


LE (:,:,e) = Le;
fe (:,e) = fea;

K(sctrB,sctrB) = K(sctrB,sctrB) + Ke - Le * invHe * Le;


fint(sctrB) = fint(sctrB) + fie - Le * invHe * fea;

end % of element loop


%%%%%%%%%%% END OF STIFFNESS MATRIX COMPUTATION %%%%%%%%%

% APPLY ESSENTIAL BOUNDARY CONDITIONS

r = fext - fint; % residual vector


r = r-K(:,udofs)*uFixed;
r = r-K(:,vdofs)*vFixed;
r(udofs) = uFixed;
r(vdofs) = vFixed;
Sec. 0.9 EAS for large deformation 55

% ...

% SOLVE SYSTEM

du = K\r;

% UPDATE NODAL DISPLACEMENT and ELEMENT STRAIN PARAMS

u = u + du;

for e=1:numelem
sctr = element(e,:);
sctrB = [sctr sctr+numnode];
a(:,e)= a(:,e) - invHE(:,:,e)*(fe(:,e)+LE(:,:,e)*du(sctrB));
end

rnorm = norm(r);
in = in + 1;
end % end of Newton-Raphson iterations

0.9 EAS for large deformation


In finite deformation theory there are many different strain measures. This leads to different
EAS formulations for finite deformation problems. Basically there are two approaches

Enhancement of the displacement gradient H [19]

Enhanced of the Green-Lagrangian strains E, [2]

In this section the EAS formulation as described in [2] is presented. This formulation is a
straightforward extension of the EAS formulation for linear elasticity problems. Recent studies
follow this approach, see e.g., [23].

0.9.1 EAS based on the enhancement of E


Variational principle
The Veubeke-Hu-Washizu functional in the reference configuration reads
Z Z  
1 T
VHW (u, E, S) = W (E)d + S : (F F I) E d Uext (194)
0 0 2
where S denotes the second Piola-Kirchhoff (2nd PK) stress tensor and Uext represents the
external work. In the above, W (E) is the strain energy potential defined as
56 CONTENTS Chap. 0

W (E)
S= (195)
E
The Green-Lagrangian strains E is additively decomposed as
1 u
E = Eu + Ee , Eu = (FT F I), F = I + (196)
2 X
Introducing the above into Eq. (194) yields
Z Z
e u e
VHW (u, E , S) = W (E + E )d S : Ee d Uext (197)
0 0

Using the following orthogonality condition


Z
S : Ee d = 0 (198)
0

simplifies Eq. (197) to


Z
e
VHW (u, E ) = W (Eu + Ee )d Uext (199)
0

The first variation of the above functional reads


Z Z
e W u W
VHW (u, E ) = : E d + : Ee d Uext (200)
0 E 0 E

which is written in matrix notation as


Z Z
e u T
VHW (u, E ) = (E ) Sd + (Ee )T Sd Uext (201)
0 0

where S and Eu/e are now vectors.

Discretization
The displacement field and the enhanced strain field are approximated as

u = N()a, Ee = M() (202)


where N() denotes the standard FE shape functions; M() represents the enhanced strain
interpolations. The nodal displacements (global) are given by a whereas designates the (local)
enhanced strain parameters.
From Eq. (202), we can write

Eu = B0 a, Ee = M (203)
where the matrix B0 is the strain-displacement matrix. Its explicit form will be given later.
Substituting Eq. (203) into Eq. (201), one obtains
Sec. 0.9 EAS for large deformation 57

Z Z Z
e T
VHW (u, E ) = a BT
0 Sd + T T T
M Sd a NT td (204)
0 0 t

Setting the above variation to zero yields the following discrete equations

Z Z
BT
0 Sd NT td = 0
| 0 {z } | t {z }
f int f ext
Z (205)
T
M Sd = 0
0
| {z }
eas
f
where the first equation is the standard discrete equilibrium equation.

Linearization
Equation (205) is linearized around the state (a0 , 0 ) as

Z  Z 
S S
ru0 + BT
+K
0 d
a + T
B0 geo
d = 0
0 a

Z  a0 Z 0  0 (206)
S S
f0eas + MT d a + MT d = 0
0 a a0 0 0

where ru = f int f ext and Kgeo is the geometric tangent matrix (also known as the initial stress
matrix) of which expression is derived later.
Let

2W S
C := 2
= (207)
E E
be the tangent elastic moduli. Note that E = B0 a + M.

S S E S S E
= = CB0 , = = CM (208)
a E a E
Introducing the above into Equation (206) yields

Z  Z 

ru0 + BT
0 CB0 d +K

geo
a + B T
0 CMd = 0

0
Z  a0 Z 0  0 (209)

f0eas + MT CB0 d a + MT CMd = 0
0 a0 0 0
58 CONTENTS Chap. 0

In matrix notation, the above becomes


" #" # " #
Kmat + Kgeo L a ru0
= eas (210)
LT H f0
where the matrices are given by

Z
K mat
= BT
0 CB0 d (material tangent stiffness) (211)
0
Z " #" #
h i S
11 S12 NJ,x
Kgeo
IJ = I NI,x NI,y d (geometric tangent stiffness) (212)
0 S21 S22 NJ,y
Z
L= BT
0 CMd (coupling stiffness) (213)
0
Z
H= MT CMd (strain stiffness) (214)
0

the derivation of the geometric tangent matrix Kgeo is given in a subsequent section.
The enhanced strain parameters can be condensed out at the element level as follows.
From Eq. (210)2 , we can write

= H1 f eas + LT a (215)
Substituting the above into Eq. (210)1 yields

(Kmat + Kgeo LH1 LT )a = ru + LH1 f eas



= f ext f int LH1 f eas (216)
| {z }
int
f

Implementation
The solution procedure of the finite deformation EAS formulation is given in Box 0.6. The
derivation of B0 and Kgeo is given as follows.
1 T 
Ee = FT Gradu =
F Gradu + (Gradu)T F (217)
2
which is rewritten more conveniently in indicial notation as

 
e 1
EAB = FkA (NI akI ) + (NI akI )FkB
2 XB XA (218)
1
= [FkA NI,B + FkB NI,A ] akI
2
Sec. 0.9 EAS for large deformation 59

By computing every components of E and put it in a vector using Voigt notation, one can write

E11 F11 NI,X F21 NI,X " #
axI
E22 = F12 NI,Y F22 NI,Y (219)
ayI
2E12 F11 NI,Y + F12 NI,X F21 NI,Y + F22 NI,X
| {z }
B0I
for two dimensions.
In order to derive the geometric tangent matrix,

NI
Eiju = Fki (uk ) = Fki akI (220)
Xj Xj
hence
Z Z 
W NI
: Eu d = Sij Fki d akI (221)
0 E 0 Xj
| {z }
int
fkI
The geometric part of the rate of the internal force vector is given by
Z Z 
geo NI NJ NI
fkI = Sij Fki d = Sij d ukJ (222)
0 Xj 0 Xi Xj
In two dimensions, the above becomes
Z
" # NJ NI " #
geo Sij d 0
fxI
0 Xi Xj uxJ
geo = (223)
Z
fyI NJ NI uyJ
0 Sij d
0 Xi Xj
By defining the following quantity

Z Z " #" #
NJ NI h i S S N
11 12 J,X
GIJ = Sij d = NI,X NI,Y d (224)
0 Xi Xj 0 S21 S22 NJ,Y

one can writes


" # " #" #
fxI
geo
GIJ 0 uxJ
= = GIJ IuJ (225)
fyI
geo
0 GIJ uyJ
| {z }
geo
KIJ
where I is the 2 2 identity matrix.
60 CONTENTS Chap. 0

geo
fx1 G11 0 G12 0 G13 0 ux1
geo
fy1 0 G11 0 G12 0 G13 uy1
geo
fx2 G21 0 G22 0 G23 0 ux2
fgeo = 0 G (226)

0 G 0 G u
y2 21 22 23 y2
geo
fx3 G31 0 G32 0 G33 0 ux3
fy3
geo
0 G31 0 G32 0 G33 uy3

0.10 Modified enhanced assumed strain methods


The motivation behind the development of modified EAS formulations is the need of locking
free elements with a low number of enhanced variables (also known as internal variables).
Some works have appeared in the literature

Modified gradients of the standard shape functions Simo et al. [21].

Piltner and Taylor [17]

Korelc and Wriggers [10], KORELC and WRIGGERS [11]

Puso [18]

0.10.1 Taylor expansion


= Bu + su + e (227)

Z Z
T su
d = 0 T e d = 0 (228)

Z
1
= (Bu + su + e )T C(Bu + su + e )d U ext (229)
2
Z
1 c
c su e
(u, , , ) = ( + su + e )T C(c + su + e )d U ext (230)
2
1 1 3
2 b 4

0.11 Mixed-enhanced strain elements


Kasper and L. Taylor [8], Kasper and Taylor [9]
Sec. 0.12 Mixed-enhanced strain elements 61

Box 0.6 EAS for finite deformation: solution procedure.


1. Newton-Raphson iterations:

(a) Set initial guess: a = a0 , = 0


(b) Compute element matrices
Z
mat
Ke = BT
0 CB0 d
0e
Z " #" #
h i S S N
11 12 J,x
Kgeo
e,IJ = I NI,x NI,y d
0e S21 S22 NJ,y
Z
Le = BT
0 CMd
0e
Z
He = MT CMd
0e

(c) Compute element vectors


Z Z
feint = BT
0 Sd, feeas = MT Sd
0e 0e

(d) Compute modified tangent stiffness matrix and internal force vector
1 T
Ke = Kmat geo
e + Ke Le He Le

feint = feint Le H1 eas


e fe

(e) Assemble and solve for nodal displacement increment


n
X n
X
int
K= Ke , f = feint
e=1
 e=1
a = K1 f ext f int

(f) Update displacement


a = a + a
(g) Update enhanced strain parameters (element level)

e = e H1
e feeas + LT
e ae

(h) Check convergence: no go back to 2b


62 CONTENTS Chap. 0

0.12 Small strain rate independent plasticity


The governing equations of small strain rate independent plastic model are summarized in Box
0.7.

Z
f int
= BT n+1 d (231a)
Z
K = BT Cep Bd (231b)

At one integration point, given data of the previous converged step n including qn , pn , and
the strain at step n + 1, n+1 the role of stress update algorithms is to find the stresses at the
step n + 1, n+1 and the corresponding tangent Cep .
Stress update algorithms

Fully implicit scheme

Semi-implicit scheme (Ortiz)

Box 0.7 Small strain rate independent plasticity.


1. Additive decomposition of the strain tensor

= e + p (232)

2. Plastic strain rate


p = f, (, q) (233)

3. Evolution of internal variables


q = f,q (, q) (234)

4. Stress-strain relation
= C : ( p ) (235)

5. Yield condition
f (, q) = 0 (236)

6. Loading/unloading condition

f 0, 0, f = 0 (237)
Sec. 0.12 Small strain rate independent plasticity 63

0.12.1 Fully implicit scheme


Given the state at time step n i.e., n , pn , qn , and the strain increment , one needs to find
n+1 , pn+1 , qn+1 , n+1 . To this end, the following equations are solved

n+1 = n + (238a)
pn+1 = pn + n+1 rn+1 (238b)
qn+1 = qn + n+1 hn+1 (238c)
n+1 = C : (n+1 pn+1 ) (238d)
fn+1 = f (n+1 , qn+1 ) = 0 (238e)

We can rewrite Eq.(238d) as

n+1 = C : (n+1 pn+1 )


= C : (n + pn n+1 rn+1 )
h i
p
= C : (n n ) + C : n+1 C : rn+1
| {z } (239)
n
h i
= n + C : n+1 C : rn+1
| {z }
n+1

where Eq.(238a) and Eq.(238b) have been used. By introducing the following quantity

trial
n+1 := n + C : (240)
we can rewrite Eq.(239) as

trial
n+1 = n+1 n+1 , n+1 = C : n+1 rn+1 = C : pn+1 (241)
| {z }
pn+1
trial
It is observed that n+1 is computed solely by the total strain (the plastic strain is frozen).
trial
Therefore n+1 is the trial stress of the elastic predictor phase. If this trial state falls outside
the yield surface, then the correct state is obtained by returning (projecting) this trial stress to
the updated yield surface. This is achieved with the so-called plastic corrector n+1 . The
geometrical interpretation of this return mapping scheme is illustrated in Fig. (11).

Elastic predictor phase


trial
n+1 = n + C :
qtrial
n+1 = qn (242)
p,trial
n+1 = pn
64 CONTENTS Chap. 0

trial
n+1

elastic plastic corrector


predictor

n+1
f (n+1 , qn+1 ) = 0

f (n , qn ) = 0

Figure 11: Elastic predictor/plastic corrector or return mapping algorithm.

If the trial elastic state is elastic i.e., falls inside the yield surface, then this state is the final
state. That is,

trial
if f (n+1 , qtrial trial
n+1 ) 0 then ()n+1 = ()n+1 (243)

Plastic corrector phase


trial
If f (n+1 , qtrial
n+1 ) > 0, one has to solve the system in Eq. (238). One accurate method to solve
it is to use the Newton-Raphson method. To this end, let us define the following residuals

a = pn+1 + pn + n+1 rn+1 = 0 (244a)


b = qn+1 + qn + n+1 hn+1 = 0 (244b)
f = f (n+1 , qn+1 ) = 0 (244c)
Eq.(244a) can be rewritten as

a = pn+1 + n+1 rn+1


(245)
= C1 : n+1 + n+1 rn+1 (, q)
where we have used the second equation of Eq.(241) in the second equality and write out the
dependency of r on the stresses and internal variables to make the following linearization clear.
Its linearization is given by (subscript n + 1 has been omitted for clarity reason)

a(k) + C1 : (k) + r(k) (k) + (r(k)


:
(k)
+ r(k) (k)
q q ) = 0 (246)
b(k) q(k) + h(k) (k) + (h(k)
:
(k)
+ h(k) (k)
q q ) = 0 (247)
Sec. 0.12 Small strain rate independent plasticity 65

where k denotes the iteration number and k = (k ). In matrix notation, the above equa-
tion is given by
" #
(k) h i h i
 1
A(k) = a(k)
(k)
r(k) (248)
q(k)

where
" #(k)
 
(k) 1 C1 + r rq
A = (249)
h I + hq

and
" # " #
h i (k) h i (k)
a r
a(k) = (k) , r(k) = (k) (250)
b h

Solving Eq.(248) for the stress increment and the internal variables increment gives
" #
(k)  (k)  h (k) i (k)
 (k)  h (k) i
= A a A r (251)
q(k)

Subsituting the above in the linearization of the yield function which is given by

f (k) + f(k) : (k) + fq(k) q(k) = 0 (252)

and solving for the update of the plastic multiplier, we obtain

f (k) f (k) A(k) a(k)


(k) = (253)
f (k) : A(k) : r(k)
where we used the notation f = [f fq ]. Finally, update the plastic strain, internal varibles, the
plastic parameter and the stress

p,(k+1) p,(k)
n+1 = n+1 C1 : (k) (254a)
(k+1) (k) (k)
q = q + q (254b)
(k+1)
= (k) + (k) (254c)
(k+1) = (k) + (k) (254d)

For convenience of computer implementation, the fully implicit scheme for small strain rate
independent plasticity model is given in Box 0.8.
66 CONTENTS Chap. 0

Consistent tangent
dn+1 = Cep : dn+1 (255)
Equation 238 in rate form is given by (subscript n + 1 is skipped)

d = C : (d dp ) (256a)
dp = d()r + (r : d + rq : dq) (256b)
dq = d()h + (h : d + hq : dq) (256c)
df = f : d + fq dq = 0 (256d)

From the first three equations of the above, eliminating dp and solving for d, dq we have
" # " #
d d
= [A] d()A : r (257)
dq 0
where
" #(1)
C1 + r rq
A= (258)
h I + hq
Replacing Eq.(257) into Eq. (256d) and solving for d() gives
h iT
f : A : d 0
d() = (259)
f : A : r
Substituting the above into Eq.(257) we have
" #  " #
d (A : r) (f : A) d
= A (260)
dq f : A : r 0
which is the expression of the consistent tangent for the stress and internal variables.
We are interested in materials where rq and h vanish i.e., the coupling terms in A are
zeros. In this case a closed form expression for the consistent tangent exits. In this simplified
case, the tensor A becomes
" # " #
(C1 + r )1 0 C 0
A= = (261)
0 (I + hq )1 0 Y
Substituting Eqs.(261), (259) into Eq.(257), we get

d = Cep d (262)
where
Sec. 0.12 Small strain rate independent plasticity 67

(C : r) (f : C)
Cep = C (263)
fq Y h + f : C : r

which has the same form with the continuum elasto-plastic tangent modulus except that the
elastic modulus is replaced by C and the term fq h is replaced by fq Y h.

Box 0.8 Fully implicit scheme for small strain rate independent plasticity.
1. Input: pn , qn , n+1

2. Initialization

k = 0 : p(0) = pn , q(0) = qn , (0) = C : (n+1 p(0) ), (0) = 0

3. Check yield function and convergence


" #
(k) (k) (k) (k) a(k)
f = f ( , q ), a = (k)
b

If f (k) < T OL1 and a(k) < T OL2: converged. Else: goto step

4. Compute increments in plasticity multiplier, in stress and in internal variables


" #(k) " #
h i (k) h i
  C1 + r rq r
A (k) 1
= , r(k) = (k) , f (k) = f(k) fq(k)
h I + hq h

f (k) f (k) A(k) a(k)


(k) =
f (k) : A(k) : r(k)
" #
(k)  (k)  h (k) i (k)
 (k)  h (k) i
= A a A r
q(k)

5. Update varibles

p,(k+1) = p,(k) C1 : (k)


q(k+1) = q(k) + q(k)
(k+1) = (k) + (k)
(k+1) = (k) + (k)
k = k + 1, goto step 3
68 CONTENTS Chap. 0

0.12.2 J2 plasticity with isotropic hardening


The constitutive equations of von Mises plastic material including isotropic hardening are given
by

f = Y () (265a)
r
3 1
= s : s, s = tr1 (265b)
2 3
Y () = 0Y + H() (265c)
r
3 s
p = f = (265d)
2 ||s||
r
2
= ||p || = (265e)
3
= C : ( p ), C = Kij kl + (ik jl + il jk ) (265f)

where is the so-called equivalent or effective stress; Y () is the yield stress which is a func-
tion of the equivalent plastic strain . The second equality in Eq. (265e) has used Eq. (265d).

Implicit elastic-predictor/plastic-corrector scheme


Let first compute the trial stress which is given by

trial = n + C : (266)
Then, the stress at the end of the time step n + 1 is given by, see Eq.(241)
3
n+1 = trial Cf = trial sn+1 (267)
n+1
3
where we have used the fact that f = s and C : s = 2s. Derivation of these relations is
2
given in a subsequent section.
Since the trace of the deviatoric stress vanishes, the above indicates that

n+1 trial
kk = kk (268)
Hence, the deviatoric stress can be written as

1
sn+1
ij = ijn+1 lln+1 ij
3
3 n+1 1 trial
= ijtrial sij kk ij (269)
n+1 3
3
= strial
ij sn+1
ij
n+1
Sec. 0.12 Small strain rate independent plasticity 69

where Eq.(267) and Eq.(268) have been used in the second equality. In the above, strial
ij is the
trial deviatoric stress tensor. Thus, we have

strial
ij
sn+1
ij = (270)
3
1+
n+1
Therefore, the equivalent stress at the end of time step n + 1 is computed according to
 1/2
3 n+1 n+1 trial
n+1 = s s = (271)
2 ij ij 3
1+
n+1
with the use of Eq.(270) in the second step. In the above, trial is the trial equivalent stress.
Thus, n+1 can be written as

n+1 = trial 3 (272)


Introducing Eq.(272) into Eq.(270) gives
 

sn+1
ij = 1 3 trial strial
ij (273)

which means that, if the plastic increment can be computed, hence sn+1 ij is completely
determined. The above also indicates why the method is named radial return mapping. In the
deviatoric stress plane, the von Mises yield surface is a circle, thus the normal to it (strial ) is
radial. So the above equation means that starting from the trial deviatoric stress, the stress is
returned back radially to the yield surface. This geometric interpretation is illustrated in Fig.
(12).
It is now ready to compute the yield function at the end of the time step n + 1 and enforce
the condition f n+1 = 0 there,

fn+1 = n+1 Y (n+1 ) = 0


(274)
= trial 3 Y (n + ) = 0

where Eq.(272) has been used and the fact that = has been made use. This equation is
generally a nonlinear equation of one unknown, the plastic increment , which requires the
use of the Newton-Raphson method,

trial 3(k) Y (n + (k) )


(k+1) = (k) + , = (275)
3 + H
The iterative process is repeated until the yield function given in Eq.(274) is sufficiently closed
to zero. In the above, H is the so-called plastic modulus which is given by
Y
H= (276)

70 CONTENTS Chap. 0

strial

l
/ tria
3
sn sn+1

fn = 0
fn+1 = 0

Figure 12: Radial return mapping for J2 plasticity with isotropic hardening.

Having obtained the plastic increment, the final thing to do is to update the stress, plastic strain
and the equivalent plastic strain by

3 3 strial
pn+1 = pn + sn+1 = pn + trial (277a)
2 n+1 2
n+1
1 trial
n+1 = Ytrial strial + kk 1 (277b)
3
n+1 = n + (277c)

where the fact that at the end of the time step, f = 0 hence Y = has been used. Also,
we have divided side by side the two Eqs.(270) and (271) and used the result in the second
equation. In the above 1 is the second order identity tensor.
The radial return algorithm for isotropic hardening von Mises plasticity is summarized, for
convenience of programming, in box 0.12.2. Noting that, we have added the computation of the
algorithmic tangent matrix for completeness.

Remark 0.0.12. In case of linear isotropic hardening, there exists a closed form expression for
the plastic increment. Therefore, there is no need for iterations at the integration point level. In
that case, Eq. ( 275) simplifies to

fn+1 = trial 3 Y 0 Hn H = 0 (280)


Sec. 0.12 Small strain rate independent plasticity 71

Box 0.9 Radial return mapping for isotropic hardening J2 palsticity


1. Initialization (k = 0): set the plastic strain and internal variables to their converged values
at the previous step. The plastic increment is set to zero and compute the trial stress

(p,0) = pn , (0) = n , (0) = 0, en+1 = dev(n+1 ), strial = 2(en+1 e(p,0) )

r
trial 3 trial trial strial
= s , s = strial : strial , n=
2 ||strial ||
2. Compute the yield function and check for convergence

f (k) = trial 3(k) Y ((k) )

Check convergence f < T OL , if yes, stop. Else goto 3.

3. Compute increments of plastic parameter

f (k) Y (k)
(k+1) = (k) + , H (k) = ( )
3 + H (k)
(k+1) = n + (k+1)

4. Increase k = k + 1 and go back to step 2.

5. Update plastic strains, internal variables, plastic parameter and stresses

r
3
pn+1 = pn + n
2
r
2 n+1
n+1 = Y n + Kn+1
kk 1
3
n+1 = n +

6. Compute the algorithmic tangent matrix

1
Calg = K1 1 + 2(I 1 1) 2n n
3
r r
2 3 3 1
b = trial , =1+ b, = b, =
||s || 2 2 H
1+
3
72 CONTENTS Chap. 0

which allows to obtain analytically

trial Y 0 Hn
= (281)
3 + H
The case H = 0 corresponds to perfect plasticity.

0.12.3 Algorithmic tangent modulii of von Mises plasticity


r
2 s strial
n+1 = C : (n+1 np ) 3 nn+1 , n := = (282)
3 ||s|| ||s||trial
Differentiating the above with respect to the strain, we write

r
2
dn+1 = C : dn+1 3 (dnn+1 + dnn+1 )
3
" r r # (283)
2 n+1 2 nn+1
= C3 n +3 : dn+1
3 n+1 3 n+1

In the above, the term in the bracket is identically the algorithmic tangent modulus matrix.

fn+1 = trial 3 Y () = 0 (284)


To compute the derivative of the plastic multiplier increment w.r.t the strain, it suffices to differ-
entiate the above equation w.r.t the strain, this is given by

trial Y
3 =0 (285)
n+1 n+1 n+1
The derivative of the trial effective stress w.r.t the strain is given by
r r
trial 3 n+1 strial 3 n+1
= n : = n 2 (286)
n+1 2 n+1 2
Introducing the above into Eq.(285) gives
r !1
Y
2 n+1
= n , = 1+ (287)
n+1 3 3
We turn now our attention to the derivative of the normal vector w.r.t the strain. By applying
the chain rule, we can write

n n s
= : (288)
s
with the following results
Sec. 0.12 Small strain rate independent plasticity 73

n 1 s 1
= (I n n), = 2(I 1 1) (289)
s ||s|| 3
Now, substituting Eq.(289) into (288) gives

n 2 1
= (I n n) : (I 1 1)
||s|| 3
(290)
2 1
= (I n n 1 1)
||s|| 3

Substitution of Eqs.(287) and (290) into Eq.(283) yields the algorithmic tangent moduli matrix

r
alg 1 3 1
C = K1 1 + 2(I 1 1) 2nn+1 nn+1 + 2 b(I nn+1 nn+1 1 1)
3 2 3
1
= K1 1 + 2(I 1 1) 2nn+1 nn+1
3
(291)

with the coefficients given by


r r
2 3 3 1
b = trial , =1+ b, = b, = (292)
||s || 2 2 H
1+
3

Implementation

For convenience of implementation, the algorithmic tangent matrix is given here in matrix form
for plane strain condition

4 2
K + 3 K 3 0 n1 n1 n1 n2 n1 n3

Calg = K 2 K + 4 0 2 n2 n1 n2 n2 n2 n3

(293)

3 3


n3 n1 n3 n2 n3 n3
0 0

0.12.4 J2 plasticity with isotropic/kinematic hardening


The constitutive equations of von Mises plastic material including isotropic hardening and kine-
matic hardening are given by
74 CONTENTS Chap. 0

f = Y () (294a)
r
3 1
= : , = s , s = tr1 (294b)
2 3
Y () = 0Y + H() (294c)
r
3
p = f = (294d)
2 ||||
r
2
= ||p || = (294e)
3
r
2 k d
= H () , H k () = (294f)
3 |||| d
p
= C : ( ), C = Kij kl + (ik jl + il jk ) (294g)

where denotes the back stress which defines the translation of the center of the von Mises
yield surface; is the relative stress.

Return mapping algorithm

s = 2 (295)

r
3 n+1
sn+1 = 2(n+1 pn+1 ) = strial 2 (296)
2 ||n+1 ||

r r
trial 3 2 k n+1
n+1 = sn+1 n+1 = s 2 n H ()
2 |||| 3 ||n+1 ||
r   (297)
trial 3 2 k n+1
= 2 + H
2 3 ||n+1 ||

r  
trial 3 2 k trial
n+1 = 2 + H
2 3 || trial ||
  (298)

= 1 3 + H k trial trial

trial [3 + H k (n + )] Y (n + ) = 0 (299)

trial 3 (n + ) n Y (n + ) = 0 (300)
Sec. 0.13 Large strain rate independent plasticity 75

Consistent tangent

0.12.5 Matlab implementation


0.12.6 Some numerical examples

0.13 Large strain rate independent plasticity


Basically finite deformation plasticity models are grouped into two categories

1. Rate additive hypoelastic plasticity

D = De + Dp

2. Multiplicative hyperelastic plasticity

F = Fe Fp

Some drawbacks of the rate additive hypoelastic plasticity:

the elastic response is hypoelastic and therefore the work done in a closed cycle of defor-
mation is not exactly zero;

they are limited to small elastic strains (but large rotations) for which the hypothesis of
hypoelasticity is valid;

Expensive incrementally objective integration algorithms;

they suffer from a somewhat ad hoc choice of objective stress rates.

Multiplicative hyperelastic plastic models were born to eliminate the aforementioned drawbacks
of rate additive hypoelastic plastic models.

0.13.1 Introduction to manifold analysis


Push-forward and pull-back operations
The transformation of the Lagrangian vector dX (initial configuration) to the Eulerian vector
dx (current configuration) can be considered as a push-forward operation by F

dx = dX = F dX (301)
Similarly, the inverse transformation of dx to dX is considered as a pull-back operation by F1

dX = dx = F1 dx (302)
76 CONTENTS Chap. 0

0.13.2 Multiplicative hyperelastic plasticity


1. Basis kinematic assumption-multiplicative decomposition of the deformation gradient
tensor

F = Fe Fp (303)

2. e
w(E )
S= e (304)
E

Hyperelastic potential
e 1 e e
E = (C I), C = FeT Fe (305)
2
e
w(E )
S= e (306)
E
2w e
S = e
e
e : E = Cel : E (307)
E E
1
E e = C e = (Fe )T De Fe (308)
2

Decomposition of rate of deformation


Using the multiplicative decomposition given in Eq.(303), we can write the velocity gradient as

D e p
L = F F1 = (F F ) (Fe Fp )1
Dt (309)
= Fe (Fe )1 + Fe Fp (Fp )1 (Fe )1

Defining the elastic and plastic parts of L as

Le := Fe (Fe )1 (310a)
Lp := Fe Fp (Fp )1 (Fe )1 (310b)

So, we have

L = Le + Lp (311)
Based on these definitions, we can also define the elastic and plastic rate of deformation
tensor and spin tensor as
Sec. 0.14 Crystal plasticity finite elements 77

1 1
De = (Le + LeT ), We = (Le LeT )
2 2 (312)
1 1
Dp = (Lp + LpT ), Wp = (Lp LpT )
2 2
We have

(L Lp ) Fe = Le Fe = Fe (313)
where Eq.(311) has been used in the second step and Eq.(310)1 has been used in the second
equality. The above then gives the rate of elastic deformation gradient written as

Fe = L Fe (Dp + Wp ) Fe (314)
Velocity gradient L on is defined as the pull back of L by Fe

L = (Fe )1 L Fe
 
e 1 e e 1 e p p 1 e 1
= (F ) F (F ) + F F (F ) (F ) Fe
(315)
= (Fe )1 Fe + Fp (Fp )1
| {z } | {z }
e p
L L
where in the second equality, Eq. (309) has been used.
e = C : De = C : (D Dp )
S = Cel : E (316)
el el

p
L = Fp (Fp )1 = r(S, q) (317)

q = h(S, q) (318)

f (S, q) = 0 (319)

0.14 Crystal plasticity finite elements


0.14.1 Introduction
Crystal structure
Most solid metals have crystalline structure, which means that atoms within materials are sit-
uated in a repeating array over long range. To describe crystalline structure, atoms are imaged
as solid hard balls sitting next to each other with different patterns and the minimum repeating
pattern is defined as an unit cell.
78 CONTENTS Chap. 0

Figure 13: From left to right: body-centered cubic (bcc), face-centered cubic (fcc) and hexago-
nal closed pack(hcp).

Miller indices

Miller indices (lmn) denotes a plane that intercepts the three points a1 /l, a2 /m and a3 /n. If
one of the indices is zero, it means that the planes do not intersect that axis (the intercept is "at
infinity"). Figure 14 gives an example. By convention, negative integers are written with a bar
as in 3 for 3.

Figure 14: Examples of determining indices for a plane using intercepts with axes; left (111),
right (221).

0.14.2 Plastic deformation


The predominant mode of plastic deformation in single crystals is crystallographic slip.
Area of the slip plane is A/ cos . The shear force acting on the slip plane is F cos .
Therefore, the resolved shear stress is given by
Sec. 0.14 Crystal plasticity finite elements 79

Figure 15: Slip in single crystal.

Figure 16: Slip in single crystal.


80 CONTENTS Chap. 0

F cos F
R = = |cos {zcos } (320)
A/ cos A
Schmid factor

= ( n) m = : (m n) (321)

0.14.3 Kinematics
The deformation gradient F is multiplicatively decomposed as

F = Fe Fp (322)

where Fe and Fp represents the elastic and plastic deformation gradients, respectively. The
existence of such a multiplicative decomposition implies that there is some stress-free interme-
diate configuration which contains the deformation due to plastic slip only; lattice distortion
and rotation are presumed to be contained in Fe . The plastic deformation gradient is assumed to
be volume-conserving. These assumptions ensure that the decomposition Eq. (322) is unique.

0
n0 n m
F
m0

Fe
Fp
n0
m0

Figure 17: Schematic of the decomposition of deformation into elastic and plastic part.

The velocity gradient tensor L is defined as the spatial gradient of the velocity. It can be
defined in terms of the gradient deformation tensor as

L = FF1 (323)

Introducing Eq. (322) into Eq. (323) yields


Sec. 0.14 Crystal plasticity finite elements 81

L = (Fe Fp + Fe Fp )(Fp )1 (Fe )1


= Fe Fp (Fp )1 (Fe )1 + Fe Fp (Fp )1 (Fe )1
(324)
= Fe (Fe )1 +Fe Fp (Fp )1 (Fe )1
| {z } | {z }
Le Lp

where Lp is the plastic velocity gradient defined in the intermediate configuration.

Fp = Lp Fp (325)

P : F = P : (Fe Fp + Fe Fp )
= (PFT
p ) : Fe + P : (Fe Fp )
(326)
= (PFT T T
p ) : Fe + (Fe PFp ) : Lp
| {z }
P

where P is the first Piola-Kirchhoff stress tensor relative to the intermediate configuration.
hyperelastic formulation, S is the second Piola-Kirchhoff stress tensor
1
P = Fe S, S = C : Ee , Ee = (Ce I), Ce = FT e Fe (327)
2
where Ee is the elastic Green strain tensor and Ce is the elastic Cauchy-Green strain tensor.
The slip direction and slip normal in the current (deformed) configuration are given by

m = m0 (Fe )1 , n = Fe n0 (328)
where (m0 , n0 ) is the slip system in the reference configuration.

0.14.4 Plastic flow


n
X
Lp = (m0 n0 ) (329)
| {z }
=1 Schmid tensor

where n denotes the total number of slip systems; is the shear rate of the same plane.
The shear rate depends on the resolved shear stress and the critical shear stress g (also
called shear resistance)

= f ( , g ) (330)

Cutino-Ortiz model
"  m1 #

= 0 1 , g (331)
g
82 CONTENTS Chap.

Box 0.10 d
F = Fe Fp

Ce = FT
e Fe

n
X
Lp = m n
=1

S = C : Ee

= (Ce m )T Sn

= f ( , g )

X
g = h ()

0.14.5 Hardening rules


The evolution of the shear resistance is defined by the following hardening rule

g = (, ) (332)
For example,
n
X

g = h | | (333)
=1

where h denotes the hardening moduli; h is the self-hardening and h with 6= reflects
latent-hardening.

h = h()[q + (1 q) ] (334)

0.14.6 Stress update algorithms


Bibliography

[1] U. Andelfinger and E. Ramm. EASARelements for twoARdimensional, three-


ARdimensional, plate and shell structures and their equivalence to HRARele-
ments. International Journal for Numerical Methods in Engineering, 36(8):
13111337, April 1993. ISSN 1097-0207. doi: 10.1002/nme.1620360805. URL
http://onlinelibrary.wiley.com/doi/10.1002/nme.1620360805/abstract.

[2] M. Bischoff and E. Ramm. Shear deformable shell elements for large
strains and rotations. International Journal for Numerical Methods in En-
gineering, 40(23):44274449, December 1997. ISSN 1097-0207. doi:
10.1002/(SICI)1097-0207(19971215)40:23<4427::AID-NME268>3.0.CO;2-9. URL
http://onlinelibrary.wiley.com/doi/10.1002/(SICI)1097-0207(19971215)

[3] C. A. Felippa. On the original publication of the general canonical functional of linear
elasticity. Journal of Applied Mechanics, 67(1):217219, March 2000. doi: 10.1115/1.
321170. URL http://link.aip.org/link/?AMJ/67/217/1.

[4] Y. C. Fung and Pin Tong. Classical and Computational Solid Mechan. World Scientific
Publishing Company, July 2001. ISBN 9810241240.

[5] S. Ghosh and R.L. Mallett. Voronoi cell finite elements. Computers & Structures,
50(1):3346, March 1994. ISSN 0045-7949. doi: 16/0045-7949(94)90435-9. URL
http://www.sciencedirect.com/science/article/pii/0045794994904359.

[6] Somnath Ghosh. Micromechanical Analysis and Multi-Scale Modeling Using the Voronoi
Cell Finite Element Method. CRC Press, Har/Cdr edition, June 2011. ISBN 1420094378.

[7] Somnath Ghosh and Suresh Moorthy. Elastic-plastic analysis of arbitrary


heterogeneous materials with the voronoi cell finite element method. Com-
puter Methods in Applied Mechanics and Engineering, 121(1-4):373409,
March 1995. ISSN 0045-7825. doi: 16/0045-7825(94)00687-I. URL
http://www.sciencedirect.com/science/article/pii/004578259400687I.

[8] Eric P. Kasper and Robert L. Taylor. A mixed-enhanced strain method: Part
II: geometrically nonlinear problems. Computers & Structures, 75(3):251
260, April 2000. ISSN 0045-7949. doi: 16/S0045-7949(99)00135-2. URL
http://www.sciencedirect.com/science/article/pii/S0045794999001352.
84 BIBLIOGRAPHY Chap.

[9] Eric P. Kasper and Robert L. Taylor. A mixed-enhanced strain method:


Part i: Geometrically linear problems. Computers & Structures, 75(3):237
250, April 2000. ISSN 0045-7949. doi: 16/S0045-7949(99)00134-0. URL
http://www.sciencedirect.com/science/article/pii/S0045794999001340.

[10] J. Korelc and P. Wriggers. An efficient 3D enhanced strain element with


taylor expansion of the shape functions. Computational Mechanics, 19(2):30
40, December 1996. ISSN 0178-7675. doi: 10.1007/BF02757781. URL
http://www.springerlink.com/content/k23523057q416w83/.

[11] JOE KORELC and PETER WRIGGERS. IMPROVED ENHANCED


STRAIN FOURARNODE ELEMENT WITH TAYLOR EXPANSION OF
THE SHAPE FUNCTIONS. International Journal for Numerical Meth-
ods in Engineering, 40(3):407421, February 1997. ISSN 1097-0207. doi:
10.1002/(SICI)1097-0207(19970215)40:3<407::AID-NME70>3.0.CO;2-P. URL
http://onlinelibrary.wiley.com/doi/10.1002/(SICI)1097-0207(19970215)40:

[12] Cornelius Lanczos. The Variational Principles of Mechanics. Dover Publications, 4 edi-
tion, March 1986. ISBN 0486650677.

[13] T. H. H. Pian. Derivation of element stiffness matrices by assumed stress distribution.


AIAA Journal, 2:13331336, 1964.

[14] T. H. H Pian and K. Sumihara. Rational approach for assumed stress finite ele-
ments. International Journal for Numerical Methods in Engineering, 20(9):1685
1695, September 1984. ISSN 1097-0207. doi: 10.1002/nme.1620200911. URL
http://onlinelibrary.wiley.com/doi/10.1002/nme.1620200911/abstract.

[15] Theodore H. H Pian. Some notes on the early history of hybrid stress fi-
nite element method. International Journal for Numerical Methods in En-
gineering, 47(1AR3):419425, January 2000. ISSN 1097-0207. doi:
10.1002/(SICI)1097-0207(20000110/30)47:1/3<419::AID-NME778>3.0.CO;2-#. URL
http://onlinelibrary.wiley.com/doi/10.1002/(SICI)1097-0207(20000110/30)

[16] Theodore H. H Pian and Pin Tong. Basis of finite element methods for solid
continua. International Journal for Numerical Methods in Engineering, 1(1):
328, January 1969. ISSN 1097-0207. doi: 10.1002/nme.1620010103. URL
http://onlinelibrary.wiley.com/doi/10.1002/nme.1620010103/abstract.

[17] R. Piltner and R. L Taylor. A quadrilateral mixed finite element with two enhanced
strain modes. International Journal for Numerical Methods in Engineering, 38
(11):17831808, June 1995. ISSN 1097-0207. doi: 10.1002/nme.1620381102. URL
http://onlinelibrary.wiley.com/doi/10.1002/nme.1620381102/abstract.

[18] Michael Anthony Puso. A highly efficient enhanced assumed strain physically
stabilized hexahedral element. International Journal for Numerical Methods
Sec. .0 BIBLIOGRAPHY 85

in Engineering, 49(8):10291064, November 2000. ISSN 1097-0207. doi:


10.1002/1097-0207(20001120)49:8<1029::AID-NME990>3.0.CO;2-3. URL
http://onlinelibrary.wiley.com/doi/10.1002/1097-0207(20001120)49:8<1

[19] J. C Simo and F. Armero. Geometrically nonARlinear enhanced


strain mixed methods and the method of incompatible modes. Interna-
tional Journal for Numerical Methods in Engineering, 33(7):14131449,
May 1992. ISSN 1097-0207. doi: 10.1002/nme.1620330705. URL
http://onlinelibrary.wiley.com/doi/10.1002/nme.1620330705/abstract.

[20] J. C Simo and M. S Rifai. A class of mixed assumed strain methods and the method
of incompatible modes. International Journal for Numerical Methods in Engineering,
29(8):15951638, June 1990. ISSN 1097-0207. doi: 10.1002/nme.1620290802. URL
http://onlinelibrary.wiley.com/doi/10.1002/nme.1620290802/abstract.

[21] J. C. Simo, F. Armero, and R. L. Taylor. Improved versions of assumed en-


hanced strain tri-linear elements for 3D finite deformation problems. Com-
puter Methods in Applied Mechanics and Engineering, 110(3-4):359386, De-
cember 1993. ISSN 0045-7825. doi: 16/0045-7825(93)90215-J. URL
http://www.sciencedirect.com/science/article/pii/004578259390215J.

[22] Robert L Taylor, Peter J Beresford, and Edward L Wilson. A nonARconforming element
for stress analysis. International Journal for Numerical Methods in Engineering, 10
(6):12111219, January 1976. ISSN 1097-0207. doi: 10.1002/nme.1620100602. URL
http://onlinelibrary.wiley.com/doi/10.1002/nme.1620100602/abstract.

[23] L. Vu-Quoc and X. G. Tan. Optimal solid shells for non-linear analyses of multilayer
composites. i. statics. Computer Methods in Applied Mechanics and Engineering, 192(9-
10):9751016, February 2003. ISSN 0045-7825. doi: 16/S0045-7825(02)00435-8. URL
http://www.sciencedirect.com/science/article/pii/S0045782502004358.

[24] P. Wriggers and J. Korelc. On enhanced strain methods for small and fi-
nite deformations of solids. Computational Mechanics, 18(6):413428,
December 1996. ISSN 0178-7675. doi: 10.1007/BF00350250. URL
http://academic.research.microsoft.com/Publication/3487410/on-enhanc

[25] O. C Zienkiewicz. Displacement and equilibrium models in the finite element


method by b. fraeijs de veubeke, chapter 9, pages 145AS197 of stress analy-
sis, edited by o. c. zienkiewicz and g. s. holister, published by john wiley &
sons, 1965. International Journal for Numerical Methods in Engineering, 52(3):
287342, September 2001. ISSN 1097-0207. doi: 10.1002/nme.339. URL
http://onlinelibrary.wiley.com/doi/10.1002/nme.339/abstract.

Potrebbero piacerti anche