Information Technology Reference
In-Depth Information
2.4 Krylov Subspace
Given a square matrix A
q
and a vector b
q
,
the spanned by the vectors
A
m
1
q
f
b
q
;
A
q
b
q
; ...;
b
q
g
is called a standard Krylov subspace of dimension m denoted
K
m
q
f
A
q
;
b
q
g
for each sub-matrix (Awais et al.
2007
; Heyouni and Jbilou
2006
):
K
m
q
f
A
q
;
b
q
g
¼span
f
b
q
;
A
q
b
q
; ...;
A
m
1
b
q
g
ð
18
Þ
q
An effective reduced model in the form (
2
) by the projection onto the Krylov
subspace of the states matrices of system (
1
) is obtained. But there is another
method to generate the Krylov subspace, which is more ef
cient that called rational
Krylov subspace de
ned as :
Y
m
q
j
q
ð
s
1
q
I
q
Þ
1
b
q
; ...;
s
j
q
I
q
Þ
1
b
q
g
K
m
q
f
A
q
;
b
q
;
s
q
g
¼
span
fð
A
q
A
q
ð
19
Þ
where, s
q
¼
ð
s
1
q
;
s
2
q
; ...;
s
m
q
Þ
3 Dual Rational Krylov for Switched Linear System
In this section, the details of the Dual Rational Krylov algorithm for computing of
two projection matrices V
r
q
and Z
r
q
for each subsystem according to switching
signal q are brie
y recalled. Dual Rational Krylov is among the best approaches to
reduce the large-scale linear switched systems. It is easy to implemented, numer-
ically stable and to avoid the dif
fl
culties in the constructing of the two projections
matrices. V
r
q
and Z
r
q
are constructed column by column during the iteration process
using a Gram Schmidt techniques in orthogonalization procedure, such as the
condition of biorthogonalithy is satis
ed Z
r
q
V
r
q
¼
I
r
q
. Take a switched linear sys-
tem as a form (
1
) and assume that a sequence of expansion points
s
r
q
g
is given, with r is the order of reduced subsystem. These expansion points are
interspersed. For each expansion point of each subsystem a two column vectors are
generated, i.e in the first iteration uses s
1
q
, the second iteration uses s
2
q
until
rth iteration.
The details of the Dual Rational Krylov algorithm for switched linear system can
be found in Table
1
(Antoulas
2009
; Druskin and Simoncini
2011
; Flagg et al.
2012
; Zhanga et al.
2008
).
The main steps of this method are:
f
s
1
q
;
s
2
q
; ...;
Step 1: Choose the interpolation points for each subsystem by the use of the
eigenvalues criterion (Gugercin
2008
).
Search WWH ::
Custom Search