Matrix Iteration Repear, Modal Constirbutions and Static Correction

Truncation Errors,
Correction
Procedures
Giacomo Boffi
Rayleigh-Ritz
Example
Subspace iteration
Truncation Errors, Correction Procedures

Giacomo Boffi
Dipartimento di Ingegneria Civile e Ambientale, Politecnico di Milano
May 20, 2015
How many
eigenvectors?
Outline
Truncation Errors,
Correction
Procedures
Giacomo Boffi
Rayleigh-Ritz
Example
Subspace iteration
Rayleigh-Ritz Example
Subspace iteration
How many eigenvectors?

Modal Partecipation Factor
Dynamic magnification factor
Static Correction
How many
eigenvectors?
RR Example
m
k x5
m
k x4
m
k x3
m
k x2
m
k x1
RR Example
m
k x5
m
k x4
m
k x3
m
k x2
m
k x1
The structural
matrices
+2 1
0
0
1 +2 1
0
0
1
+2
1
K=k
0
0
1 +2
0
0
0
1
0
1
0
0
0
M = m 0
0
1
+1
0
0
1
0
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
1
RR Example
m
k x5
m
k x4
m
k x3
m
k x2
m
k x1
The structural
matrices
+2 1
0
0
0
1 0
1 +2 1
0 1
0
0
1 +2 1
0
K=k
0
M = m 0 0
0
0 0
0
1 +2 1
0
0
0
1 +1
0 0
The Ritz
and the reduced matrices,
base vectors

0.2 0.5
= k 0.2 0.2
0.4 1.0 K
0.2 2.0
=
0.6 0.5
0.8 +0.0 M
= m 2.2 0.2
0.2 2.5
1.0
1.0
0
0
1
0
0
0
0
0
1
0
0
0
0
1
RR Example
m
The structural
matrices
+2 1
0
0
0
1 0 0 0
1 +2 1
0 1 0 0
0
0
m
1 +2 1
0
K=k
0
M = m 0 0 1 0
k x4
0
0 0 0 1
0
1
+2
1
m
0
0
0
1 +1
0 0 0 0
k x3 The Ritz base vectors and the reduced matrices,
m

0.2 0.5
= k 0.2 0.2
0.4 1.0 K
k x2
m
0.2 2.0
=
0.6 0.5
0.8 +0.0 M
= m 2.2 0.2
k x1
0.2 2.5
1.0
1.0

2 22
2 2
z1
0
Red. eigenproblem ( = 2 m/k):
=
2 2 20 25 z2
0
k x5
0
0
0
1
RR Example
m
The structural
matrices
+2 1
0
0
0
1 0 0 0
1 +2 1
0 1 0 0
0
0
m
1 +2 1
0
K=k
0
M = m 0 0 1 0
k x4
0
0 0 0 1
0
1
+2
1
m
0
0
0
1 +1
0 0 0 0
k x3 The Ritz base vectors and the reduced matrices,
m

0.2 0.5
= k 0.2 0.2
0.4 1.0 K
k x2
m
0.2 2.0
=
0.6 0.5
0.8 +0.0 M
= m 2.2 0.2
k x1
0.2 2.5
1.0
1.0

2 22
2 2
z1
0
Red. eigenproblem ( = 2 m/k):
=
2 2 20 25 z2
0
The roots are
=
0.0824,
=
0.800,
the
frequencies
are
1
2
p
p
1 = 0.287 k/m [ = 0.285], 2 = 0.850 k/m [ = 0.831], while the
k/m normalized exact eigenvalues are [0.08101405, 0.69027853].
The first eigenvalue is estimated with good approximation.
k x5
0
0
0
1
Truncation Errors,
Correction
Procedures
Giacomo Boffi
1.329
0.1360
The RR eigenvector matrix, and the exact one, :
The Ritz coordinates eigenvector matrix is Z =
+0.3338
+0.6676
=
+0.8654
+1.0632
+1.1932
0.6135
1.2270
0.6008
,
+0.0254
+1.2713
+0.3338
+0.6405
=
+0.8954
+1.0779
+1.1932

0.03170
.
1.240
0.8398
1.0999
0.6008
.
+0.3131
+1.0108
The accuracy of the estimates for the 1st mode is very good, on the
contrary the 2nd mode estimates are in error starting from the second
digit.
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
Truncation Errors,
Correction
Procedures
Giacomo Boffi
1.329
0.1360
The RR eigenvector matrix, and the exact one, :
The Ritz coordinates eigenvector matrix is Z =
+0.3338
+0.6676
=
+0.8654
+1.0632
+1.1932
0.6135
1.2270
0.6008
,
+0.0254
+1.2713
+0.3338
+0.6405
=
+0.8954
+1.0779
+1.1932

0.03170
.
1.240
0.8398
1.0999
0.6008
.
+0.3131
+1.0108
The accuracy of the estimates for the 1st mode is very good, on the
contrary the 2nd mode estimates are in error starting from the second
digit.
= K1 M as a new Ritz base to get a
It may be interesting to use
new estimate of the Ritz and of the structural eigenpairs.
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
Introduction to Subspace Iteration
Truncation Errors,
Correction
Procedures
Giacomo Boffi
We have seen that the Rayleigh-Ritz procedure can offer a

good estimate for p M/2 modes, mostly because of the
arbitrariness in the choice of the Ritz reduced base .
Solving a M = 2p order eigenvalue problem to get p
eigenvalues is very onerous as the operation count is
O(M 3 ).
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
Truncation Errors,
Correction
Procedures
Giacomo Boffi

O(M 3 ).
If we could reduce the arbitrariness in the choice of the Ritz
base vectors, we could use less vectors and solve a much
smaller (in terms of operations count) eigenvalue problem.
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
Truncation Errors,
Correction
Procedures
Giacomo Boffi

O(M 3 ).
If we could reduce the arbitrariness in the choice of the Ritz
base vectors, we could use less vectors and solve a much
smaller (in terms of operations count) eigenvalue problem.
If one thinks of it, with a M = 1 base we could nevertheless
compute within arbitrary accuracy one eigenvector using
Matrix Iteration, isnt it? the trick is changing the base at
every iteration...
It happens that Matrix Iteration can be applied to a set of
trial vectors at once, under the name of Subspace Iteration.
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
Statement of the procedure
Truncation Errors,
Correction
Procedures
Giacomo Boffi
The first M eigenvalue equations can be written in matrix

algebra, in terms of an N M matrix of eigenvectors and
an M M diagonal matrix that collects the eigenvalues
K = M
NN NM
NN NM MM
Using again the hat notation for the unnormalized iterate,

from the previous equation we can write
1 = M0
K
where 0 is the matrix, N M, of the zero order trial
1 is the matrix of the non-normalized first
vectors, and
order trial vectors.
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
Orthonormalization
Truncation Errors,
Correction
Procedures
Giacomo Boffi
To proceed with iterations,

n+1 must be orthogonalized, so
1. the trial vectors in
that each trial vector converges to a different
eigenvector instead of collapsing to the first
eigenvector,
2. all the trial vectors must be normalized, so that the
ratio between the normalized vectors and the
unnormalized iterated vectors converges to the
corresponding eigenvalue.
Next week, more on Gram-Schmidt procedure
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
Orthonormalization
Truncation Errors,
Correction
Procedures
Giacomo Boffi
To proceed with iterations,

n+1 must be orthogonalized, so
1. the trial vectors in
that each trial vector converges to a different
eigenvector instead of collapsing to the first
eigenvector,
2. all the trial vectors must be normalized, so that the
ratio between the normalized vectors and the
unnormalized iterated vectors converges to the
corresponding eigenvalue.
These operations can be performed in different ways (e.g.,
ortho-normalization by Gram-Schmidt1 procedure).
Another possibility to do both at once is solving a
Rayleigh-Ritz eigenvalue problem, defined in the Ritz base
n+1 .
constituted by the vectors in
1
Next week, more on Gram-Schmidt procedure
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
Associated Eigenvalue Problem
Truncation Errors,
Correction
Procedures
Giacomo Boffi
Developing the procedure for n = 0, with the generalized matrices

1 T K
1
K?1 =
Rayleigh-Ritz
Example
Subspace iteration
and
1 T M
1
M?1 =
the Rayleigh-Ritz eigenvalue problem associated with the
1 is
orthonormalisation of
1 = M?1 Z
1 21 .
K?1 Z
1 and the
After solving for the Ritz coordinates mode shapes, Z
frequencies 21 , using any suitable procedure, it is usually convenient to
1 = I. The ortho-normalized set
1 T M?1 Z
normalize the shapes, so that Z
of trial vectors at the end of the iteration is then written as
1Z
1.
1 =
The entire process can be repeated for n = 1, then n = 2, n = . . .
until the eigenvalues converge within a prescribed tolerance.
How many
eigenvectors?
Convergence
Truncation Errors,
Correction
Procedures
Giacomo Boffi
In principle, the procedure will converge to all the M lower

eigenvalues and eigenvectors of the structural problem, but
it was found that the subspace iteration method converges
faster to the lower p eigenpairs, those required for dynamic
analysis, if there is some additional trial vector; on the
other hand, too many additional trial vectors slow down the
computation without ulterior benefits.
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
Convergence
Truncation Errors,
Correction
Procedures
Giacomo Boffi

Experience has shown that the optimal total number M of
trial vectors is the minimum of 2p and p + 8.
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
Convergence
Truncation Errors,
Correction
Procedures
Giacomo Boffi

Experience has shown that the optimal total number M of
trial vectors is the minimum of 2p and p + 8.
The subspace iteration method makes it possible to
compute simultaneosly a set of eigenpairs within any
required level of approximation, and is the preferred method
to compute the eigenpairs of a complex dynamic system.
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
Standard Form
Truncation Errors,
Correction
Procedures
In algebra textbooks, the eigenproblem is usually stated as

A y = y
Rayleigh-Ritz
Example
and all the relevant algorithms to actually compute the eigenthings

(Jacobi method, Q R method, etc) are referred to the above
statement of the problem.
Our problem is, instead, formulated as
K x = M x.
Any symmetric, definite positive matrix B can be subjected to a unique
Choleski Decomposition (CD), B = L LT where L is a lower triangular
matrix. Applying CD to M, the eigenvector equation is,
LT x.
K x = K (LT )1 LT x = L
|{z}
| {z }
M
I
Premultiplying by L1 , with y = LT x
1
L1 K(LT )1 |{z}
LT x = L
LT x
| {z L} |{z}
|
{z
}
y
y
I
A
Giacomo Boffi
Ay = y.
Subspace iteration
How many
eigenvectors?
How Many Eigenvectors
Truncation Errors,
Correction
Procedures
Giacomo Boffi
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
Dynamic magnification
factor
are needed to correctly represent the response of a MDOF

system to a time-varying load?
Static Correction
Introduction
Truncation Errors,
Correction
Procedures
Giacomo Boffi
Rayleigh-Ritz
Example
To understand how many eigenvectors we have to use in a

dynamic analysis, we must consider two aspects, the
loading shape and the excitation frequency.
In the following, well consider only external loadings whose
dependance on time and space can be separated, as in
p(x, t) = r f (t),
so that we can discuss separately the two aspects of the
problem.
Subspace iteration
How many
eigenvectors?
factor
Static Correction
Introduction
It is worth noting that earthquake loadings are precisely of
this type:
p(x, t) = Mr ug (t)
where the vector r is used to choose the structural dofs
that are excited by the ground motion component under
consideration.
Usually r is simply a vector of zeros and ones that represents the
degrees of freedom that are excited by a component of the ground
acceleration.
Truncation Errors,
Correction
Procedures
Giacomo Boffi
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
factor
Static Correction
Introduction
Truncation Errors,
Correction
Procedures
It is worth noting that earthquake loadings are precisely of

this type:
p(x, t) = Mr ug (t)
where the vector r is used to choose the structural dofs
that are excited by the ground motion component under
consideration.
Usually r is simply a vector of zeros and ones that represents the
degrees of freedom that are excited by a component of the ground
acceleration.
Multiplication of M by g (the acceleration of gravity) and

division of ug by g, serves to show a dimensional load
vector multiplied by an adimensional function.
p(x, t) = g Mr
g (t)
u
g
= rg fg (t)
Giacomo Boffi
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
factor
Static Correction

Under the assumption of separability, we can write the i -th modal
equation of motion as
T
i r f (t)
Mi
qi + 2i i q i + i2 qi =
= i f (t)
T
g i Mr f (t)
g
Mi
Truncation Errors,
Correction
Procedures
Giacomo Boffi
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
with the modal mass Mi = T

i M i .
It is apparent that the modal response amplitude depends
I
on the characteristics of the time dependency of loading,

f (t),
on the so called modal partecipation factor i ,

i = T
i r/Mi
g
= g T
r/Mi = T
i M
i r /Mi
= M1 T r
factor
Static Correction

Under the assumption of separability, we can write the i -th modal
equation of motion as
T
i r f (t)
Mi
qi + 2i i q i + i2 qi =
= i f (t)
T
g i Mr f (t)
g
Mi
Truncation Errors,
Correction
Procedures
Giacomo Boffi
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
with the modal mass Mi = T

i M i .
It is apparent that the modal response amplitude depends
I
on the characteristics of the time dependency of loading,

f (t),
on the so called modal partecipation factor i ,

i = T
i r/Mi
g
= g T
r/Mi = T
i M
i r /Mi
= M1 T r
Note that both the definitions of modal partecipation give it the
dimensions of an acceleration.
factor
Static Correction
Partecipation Factor Amplitudes
Truncation Errors,
Correction
Procedures
Giacomo Boffi
Rayleigh-Ritz
Example
Subspace iteration
For a given loading r the modal partecipation factor i is proportional

to the work done by the modal displacement qi T
i for the given
loading r:
I if the mode shape and the loading shape are approximately equal
(equal signs, component by component), the work (dot product)

is maximized,
I if the mode shape is significantly different from the loading
(different signs), there is some amount of cancellation and the

value of the s will be reduced.
How many
eigenvectors?
factor
Static Correction
Example
Truncation Errors,
Correction
Procedures
Consider a shear type building, with mass distribution

approximately constant over its height:
Giacomo Boffi
Rayleigh-Ritz
Example
Subspace iteration
r = {1, 1, . . . , 1}T
and
g Mr mg{1, 1, . . . , 1}T .
How many
eigenvectors?
an external loading and the first 3 eigenvectors as sketched

below:
gM r
factor
Static Correction
Example, cont.
Truncation Errors,
Correction
Procedures
Giacomo Boffi
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
factor
Static Correction
gM r
For EQ loading, 1 is relatively large for the first mode, as

loading components and displacements have the same sign,
with respect to other i s, where the oscillating nature of
the higher eigenvectors will lead to increasing cancellation.
On the other hand, consider the external loading, whose
peculiar shape is similar to the 3rd mode. 3 will be more
relevant than i s for lower or higher modes.
Modal Loads Expansion
Truncation Errors,
Correction
Procedures
Giacomo Boffi
We define the modal load contribution as
Rayleigh-Ritz
Example
r i = M i ai
and express the load vector as a linear combination of the modal
contributions
X
X
r=
M i ai =
ri .
i
If we premultiply by
T
j
the above equation,

X
X
T
T
M i ai =
ij Mi ai = Mj aj
j r = j
i
1. a modal load component works only for the displacements

associated with the corresponding eigenvector,
2. comparing with the definition of i = T
i r/Mi , we conclude that
ri = M i i
R = M diag()
Subspace iteration
How many
eigenvectors?
factor
Static Correction
Equivalent Static Forces

For mode i , the equation of motion is
qi + 2i i q i + i2 qi = i f (t)
Truncation Errors,
Correction
Procedures
Giacomo Boffi
Rayleigh-Ritz
Example
Subspace iteration
with qi = i Di , we can write, to single out the

dependency on the modulating function,
i + 2i i D i + 2 Di = f (t)
D
i
The modal contribution to displacement is
xi = i i Di (t)
and the modal contribution to elastic forces f i = K xi can
be written (being K i = i2 M i ) as
f i = K xi = i K i Di = i2 (i M i )Di = ri i2 Di
1
Di (dimensionally the square of a time), is called

pseudo-displacement.
How many
eigenvectors?
factor
Static Correction
Equivalent Static Forces

For mode i , the equation of motion is
qi + 2i i q i + i2 qi = i f (t)
Truncation Errors,
Correction
Procedures
Giacomo Boffi
Rayleigh-Ritz
Example
Subspace iteration
with qi = i Di , we can write, to single out the

dependency on the modulating function,
i + 2i i D i + 2 Di = f (t)
D
i
The modal contribution to displacement is
xi = i i Di (t)
and the modal contribution to elastic forces f i = K xi can
be written (being K i = i2 M i ) as
f i = K xi = i K i Di = i2 (i M i )Di = ri i2 Di
1
Di (dimensionally the square of a time), is called

pseudo-displacement.
How many
eigenvectors?
factor
Static Correction
Equivalent Static Response
Truncation Errors,
Correction
Procedures
Giacomo Boffi
The response can be determined by superposition of the effects of

these pseudo-static forces f i = ri i2 Di (t).
If a required response quantity (be it a nodal displacement, a bending
moment in a beam, the total shear force in a building storey, etc etc)
is indicated by s(t), we can compute with a static calculation (usually
using the FEM model underlying the dynamic analysis) the modal
static contribution sist and write
X st 2
X
s(t) =
si (i Di (t)) =
si (t),
where the modal contribution to response si (t) is given by
1. static analysis using ri as the static load vector,
2. dynamic amplification using the factor i2 Di (t).
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
factor
Static Correction
Equivalent Static Response
Truncation Errors,
Correction
Procedures
Giacomo Boffi
The response can be determined by superposition of the effects of

these pseudo-static forces f i = ri i2 Di (t).
If a required response quantity (be it a nodal displacement, a bending
moment in a beam, the total shear force in a building storey, etc etc)
is indicated by s(t), we can compute with a static calculation (usually
using the FEM model underlying the dynamic analysis) the modal
static contribution sist and write
X st 2
X
s(t) =
si (i Di (t)) =
si (t),
where the modal contribution to response si (t) is given by
1. static analysis using ri as the static load vector,
2. dynamic amplification using the factor i2 Di (t).
This formulation is particularly apt to our discussion of different
contributions to response components.
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
factor
Static Correction
Modal Contribution Factors (MCF)
Truncation Errors,
Correction
Procedures
Giacomo Boffi
s st ,
Say that the static response due to r is denoted by

then
si (t), the modal contribution to response s(t), can be
written
si (t) =
sist i2 Di (t)
s st
s st ist
s
s st
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
factor
i2 Di (t)
= si s
st
i2 Di (t).
We have introduced si = sist , the modal contribution factor,

the ratio of the modal static contribution to the total static
response.
The si are dimensionless, are indipendent P
on the eigenvector
si = 1.
scaling procedure and their sum is unity,
Static Correction
Maximum Response
Truncation Errors,
Correction
Procedures
Denote by Di0 the maximum absolute value (or peak) of

the pseudo displacement time history,
Di0 = max{|Di (t)|}.
t
Giacomo Boffi
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
It will be
si0 = si s
st
factor
i2 Di0
Static Correction
Eventually, the dynamic response factor for mode i , Rdi is

defined by
Di0
Rdi = st
Di0
st is the peak value of the static pseudowhere Di0
displacement
Dist =
f (t)
,
i2
st
Di0
=
f0
i2
Maximum Response
Truncation Errors,
Correction
Procedures
Giacomo Boffi
With f0 = max{|f (t)|} the peak p-displacement is

Di0 =
Rdi f0 /i2
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
and the peak of the modal contribution is

si0 = si s st i2 Di0 = f0 s st
si Rdi
factor
Static Correction
Maximum Response
Truncation Errors,
Correction
Procedures
Giacomo Boffi
With f0 = max{|f (t)|} the peak p-displacement is

Di0 =
Rdi f0 /i2
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
and the peak of the modal contribution is

si0 = si s st i2 Di0 = f0 s st
si Rdi
The first two terms are independent of the mode, the last
are independent from each other and their product is the
factor that influences the modal contributions.
Note that this product has the sign of si , as the dynamic
response factor is always positive.
factor
Static Correction
MCFs example
The following table (from Chopra, 2nd ed.) displays the
si and their
partial sums for a shear-type, 5 floors building where all the storey
masses are equal and all the storey stiffnesses are equal too.
The response quantities chosen are x5n , the MCFs to the top
n , the MCF s to the base shear, for two different
displacement and V
load shapes.
MCFs example
si and their
displacement and V
load shapes.
r = {0, 0, 0, 0, 1}T
r = {0, 0, 0, 1, 2}T
Top Disp.
Base Shear
Top Disp.
i
X
i
X
i
X
n or i
x5n
1
2
3
4
5
0.880
0.087
0.024
0.008
0.002
x5j
n
V
0.880
0.967
0.991
0.998
1.000
1.252
-0.362
0.159
-0.063
0.015
j
V
x5n
1.252
0.890
1.048
0.985
1.000
0.792
0.123
0.055
0.024
0.006
Base Shear
x5j
n
V
i
X
j
V
0.792
0.915
0.970
0.994
1.000
1.353
-0.612
0.043
-0.242
0.070
1.353
0.741
1.172
0.930
1.000
MCFs example
si and their
displacement and V
load shapes.
r = {0, 0, 0, 0, 1}T
r = {0, 0, 0, 1, 2}T
Top Disp.
Base Shear
Top Disp.
i
X
i
X
i
X
n or i
x5n
1
2
3
4
5
0.880
0.087
0.024
0.008
0.002
x5j
n
V
0.880
0.967
0.991
0.998
1.000
1.252
-0.362
0.159
-0.063
0.015
j
V
x5n
1.252
0.890
1.048
0.985
1.000
0.792
0.123
0.055
0.024
0.006
Base Shear
x5j
n
V
i
X
j
V
0.792
0.915
0.970
0.994
1.000
1.353
-0.612
0.043
-0.242
0.070
1.353
0.741
1.172
0.930
1.000
Note that:
1. for any given r, the base shear is more influenced by higher modes, and
2. for any given response quantity, the second, skewed r gives greater
modal contributions for higher modes.
Dynamic Response Ratios
Truncation Errors,
Correction
Procedures
Giacomo Boffi
Dynamic Response Ratios are the same that we have seen for SDOF
systems.
Next page, for an undamped system,
I solid line, the ratio of the modal elastic force FS,i = Ki qi sin t to
the harmonic applied modal force, Pi sin t, plotted against the

frequency ratio = /i .
For = 0 the ratio is 1, the applied load is fully balanced by the
elastic resistance.
For fixed excitation frequency, 0 for high modal frequencies.
I dashed line,the ratio of the modal inertial force, FI ,i = 2 FS,i to
the load.
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
factor
Static Correction
Dynamic Response Ratios
Truncation Errors,
Correction
Procedures
Giacomo Boffi
Dynamic Response Ratios are the same that we have seen for SDOF
systems.
Next page, for an undamped system,
I solid line, the ratio of the modal elastic force FS,i = Ki qi sin t to
the harmonic applied modal force, Pi sin t, plotted against the

frequency ratio = /i .
For = 0 the ratio is 1, the applied load is fully balanced by the
elastic resistance.
For fixed excitation frequency, 0 for high modal frequencies.
I dashed line,the ratio of the modal inertial force, FI ,i = 2 FS,i to
the load.
Note that for steady-state motion the sum of the elastic and inertial
force ratios is constant and equal to 1, as in
(FS,i + FI ,i ) sin t = Pi sin t.
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
factor
Static Correction
4
FS/Pi
Modal resistance ratios
FI/Pi
2
1
0
-1
-2
-3
0.5
1.5
2.5
Frequency ratio, =/i
I For a fixed excitation frequency and high modal frequencies the
frequency ratio 0.
4
FS/Pi
FI/Pi
2
1
0
-1
-2
-3
0.5
1.5
2.5
frequency ratio 0.
I For 0 the response is quasi-static.
4
FS/Pi
FI/Pi
2
1
0
-1
-2
-3
0.5
1.5
2.5
frequency ratio 0.
I Hence, for higher modes the response is pseudo-static.
4
FS/Pi
FI/Pi
2
1
0
-1
-2
-3
0.5
1.5
2.5
frequency ratio 0.
I Hence, for higher modes the response is pseudo-static.
I On the contrary, for excitation frequencies high enough the lower
modes respond with purely inertial forces.
Static Correction
Truncation Errors,
Correction
Procedures
Giacomo Boffi
The preceding discussion indicates that higher modes

contributions to the response could be approximated with
the static response, leading to the idea of a Static
Correction of the dynamic response
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
factor
Static Correction
Static Correction
Truncation Errors,
Correction
Procedures
Giacomo Boffi
The preceding discussion indicates that higher modes

contributions to the response could be approximated with
the static response, leading to the idea of a Static
Correction of the dynamic response
pi (t)
for i > ndy ,
For a system where qi (t)
Ki
ndy being the number of dynamically responding modes,
we can write
x(t) xdy (t) + xst (t) =
ndy
X
1
i qi (t) +
N
X
ndy +1
pi (t)
Ki
where the response for each of the first ndy modes can be
computed as usual.
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
factor
Static Correction
Static Modal Components
Truncation Errors,
Correction
Procedures
The static modal displacement component xj , j > ndy can

be written
xj (t) = j qj (t)
j T
j
Kj
p(t) = Fj p(t)
How many
eigenvectors?
Static Correction
and is used to compute the j-th mode static deflections due

to the applied load vector.
The total displacements, the dynamic contributions and the
static correction, for p(t) = r f (t), are then
j qj (t) + f (t)
Subspace iteration
factor
j T
j
Fj =
Kj
ndy
X
Rayleigh-Ritz
Example
The modal flexibility matrix is defined by
Giacomo Boffi
N
X
ndy +1
Fj r.
Alternative Formulation
Truncation Errors,
Correction
Procedures
Giacomo Boffi
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
Our last formula for static correction is

x
ndy
X
1
j qj (t) + f (t)
N
X
factor
Fj r.
ndy +1
To use the above formula all mode shapes, all modal

stiffnesses and all modal flexibility matrices must be
computed, undermining the efficiency of the procedure.
Static Correction
This problem can be obviated computing the total static
displacements and expressing it inPterms of modal
contributions: xst = K1 rf (t) = N
1 Fj rf (t).
Truncation Errors,
Correction
Procedures
Giacomo Boffi
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
factor
Static Correction
Truncation Errors,
Correction
Procedures

1 Fj rf (t).
Subtracting the static displacements due to the first ndy
modes to both members it is
Giacomo Boffi
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
N
X
Fj rf (t) = K1 rf (t)
ndy +1
ndy
X
Fj rf (t) = f (t) K1
ndy
X
The corrected total displacements have hence the

expression
!
ndy
ndy
X
X
x
i qi (t) + f (t) K1
Fi r,
1
!
Fj
factor
r.
Static Correction
Truncation Errors,
Correction
Procedures

1 Fj rf (t).
Subtracting the static displacements due to the first ndy
modes to both members it is
Giacomo Boffi
Rayleigh-Ritz
Example
Subspace iteration
How many
eigenvectors?
N
X
Fj rf (t) = K1 rf (t)
ndy +1
ndy
X
Fj rf (t) = f (t) K1
ndy
X
!
Fj
The corrected total displacements have hence the

expression
!
ndy
ndy
X
X
x
i qi (t) + f (t) K1
Fi r,
1
Note that the constant term following f (t) can be

computed with information already in our possess at the
end of the dynamic analysis.
factor
r.
Static Correction
Effectiveness of Static Correction
Truncation Errors,
Correction
Procedures
Giacomo Boffi
Rayleigh-Ritz
Example
In these circumstances, few modes with static correction

give results comparable to the results obtained using much
more modes in a straightforward modal displacement
superposition analysis.
Subspace iteration
How many
eigenvectors?
factor
Static Correction
Truncation Errors,
Correction
Procedures
Giacomo Boffi
Rayleigh-Ritz
Example

Subspace iteration
How many
eigenvectors?
factor
Static Correction
An high number of modes is required to account for

the spatial distribution of the loading but only a few
lower modes are subjected to significant dynamic
amplification.
Truncation Errors,
Correction
Procedures
Giacomo Boffi
Rayleigh-Ritz
Example

Subspace iteration
How many
eigenvectors?
factor
Static Correction
An high number of modes is required to account for

the spatial distribution of the loading but only a few
lower modes are subjected to significant dynamic
amplification.
Refined stress analysis is required even if the dynamic

response involves only a few lower modes.

Matrix Iteration Repear, Modal Constirbutions and Static Correction

Caricato da

Informazioni sul documento

Descrizione originale:

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Matrix Iteration Repear, Modal Constirbutions and Static Correction

Caricato da

Copyright:

Formati disponibili

Truncation Errors,

Truncation Errors, Correction Procedures

May 20, 2015

How many eigenvectors?

The Ritz coordinates eigenvector matrix is Z =

The Ritz coordinates eigenvector matrix is Z =

Introduction to Subspace Iteration

We have seen that the Rayleigh-Ritz procedure can offer a

Introduction to Subspace Iteration

We have seen that the Rayleigh-Ritz procedure can offer a

Introduction to Subspace Iteration

We have seen that the Rayleigh-Ritz procedure can offer a

Statement of the procedure

The first M eigenvalue equations can be written in matrix

Using again the hat notation for the unnormalized iterate,

To proceed with iterations,

Next week, more on Gram-Schmidt procedure

To proceed with iterations,

Next week, more on Gram-Schmidt procedure

Associated Eigenvalue Problem

Developing the procedure for n = 0, with the generalized matrices

In principle, the procedure will converge to all the M lower

In principle, the procedure will converge to all the M lower

In principle, the procedure will converge to all the M lower

In algebra textbooks, the eigenproblem is usually stated as

and all the relevant algorithms to actually compute the eigenthings

How Many Eigenvectors

are needed to correctly represent the response of a MDOF

To understand how many eigenvectors we have to use in a

It is worth noting that earthquake loadings are precisely of

Multiplication of M by g (the acceleration of gravity) and

Modal Partecipation Factor

with the modal mass Mi = T

on the characteristics of the time dependency of loading,

on the so called modal partecipation factor i ,

Modal Partecipation Factor

with the modal mass Mi = T

on the characteristics of the time dependency of loading,

on the so called modal partecipation factor i ,

Partecipation Factor Amplitudes

For a given loading r the modal partecipation factor i is proportional

(equal signs, component by component), the work (dot product)

(different signs), there is some amount of cancellation and the

Consider a shear type building, with mass distribution

an external loading and the first 3 eigenvectors as sketched

For EQ loading, 1 is relatively large for the first mode, as

Modal Loads Expansion

We define the modal load contribution as

the above equation,

1. a modal load component works only for the displacements

Equivalent Static Forces

with qi = i Di , we can write, to single out the

Di (dimensionally the square of a time), is called

Equivalent Static Forces

with qi = i Di , we can write, to single out the

Di (dimensionally the square of a time), is called

Equivalent Static Response

The response can be determined by superposition of the effects of

Equivalent Static Response

The response can be determined by superposition of the effects of

Modal Contribution Factors (MCF)

Say that the static response due to r is denoted by