20.05.2014 Views

link to my thesis

link to my thesis

link to my thesis

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

10.2. LINEARIZING A POLYNOMIAL EIGENVALUE PROBLEM 167<br />

generalized eigenvalue problem (B + ρ 1 C)w = 0 one may also consider the left generalized<br />

eigenvalue problem u ∗ (B + ρ 1 C)=0 T . Thus, linear dependence of the rows of<br />

B+ρ 1 C for various values of ρ 1 (including infinity) can be given similar meaning <strong>to</strong>o.<br />

As demonstrated in the previous section, the problem of finding solutions of the<br />

system of equations (10.2) subject <strong>to</strong> (10.3) can be cast in the form of a polynomial<br />

eigenvalue problem involving the square polynomial matrix Ãã N−1<br />

(ρ 1 ) T . The<br />

problem of computing all the values ρ 1 and corresponding non-trivial vec<strong>to</strong>rs v that<br />

satisfy Ãã N−1<br />

(ρ 1 ) T v = 0 can be approached by rewriting it as a generalized eigenvalue<br />

problem as follows.<br />

First, expand the polynomial matrix Ãã N−1<br />

(ρ 1 ) T as a matrix polynomial in ρ 1 of<br />

degree N − 1, by:<br />

ÃãN−1 (ρ 1 ) T = A 0 + ρ 1 A 1 + ...+ ρ N−1<br />

1 A N−1 . (10.12)<br />

Applying a linearization technique as mentioned, e.g., in [55], [56] or [72], makes<br />

the polynomial eigenvalue problem (10.10), where the involved matrix is written as<br />

(10.12), equivalent <strong>to</strong> a generalized eigenvalue problem:<br />

(A 0 + ρ 1 A 1 + ...+ ρ N−1<br />

1 A N−1 ) v =0 ⇐⇒ (B + ρ 1 C)ṽ = 0 (10.13)<br />

where ρ 1 occurs linearly and where the matrices B and C are (large) block-partitioned<br />

square matrices with the following structure:<br />

⎛<br />

⎞<br />

0 I 0<br />

.<br />

B = ⎜<br />

. ..<br />

⎟<br />

⎝ 0 0 I ⎠ ,<br />

A 0 A 1 ... A N−2<br />

⎛<br />

C = ⎜<br />

⎝<br />

(10.14)<br />

⎞<br />

−I 0 0<br />

. .. . ⎟<br />

0 −I 0 ⎠ .<br />

0 ... 0 A N−1<br />

The eigenvec<strong>to</strong>rs ṽ of the generalized eigenvalue problem (B+ρ 1 C)ṽ = 0 in (10.13)<br />

have a built-in structure, which is given by:<br />

⎛<br />

ṽ = ⎜<br />

⎝<br />

v<br />

ρ 1 v<br />

.<br />

ρ N−2<br />

1 v<br />

⎞<br />

⎟<br />

⎠ . (10.15)<br />

The block vec<strong>to</strong>rs v inside such an eigenvec<strong>to</strong>r ṽ are eigenvec<strong>to</strong>rs of the original<br />

polynomial eigenvalue problem (10.10). Note that the eigenvec<strong>to</strong>rs ṽ have additional

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!