27.07.2013 Views

AIR Tools - A MATLAB Package for Algebraic Iterative ...

AIR Tools - A MATLAB Package for Algebraic Iterative ...

AIR Tools - A MATLAB Package for Algebraic Iterative ...

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.

3.4 Block-<strong>Iterative</strong> Methods 31<br />

where ǫ is an arbitrarily small but fixed constant and M t(k) are given symmetric<br />

and positive definite matrices with the control t(k), then any sequence generated<br />

by (3.11) converges to a solution <strong>for</strong> (2.1). If in addition x 0 ∈ R(A T ), then x k<br />

converges to the solution of minimum 2-norm.<br />

The BICAV method has the property that <strong>for</strong> p = 1 the method becomes fully<br />

simultaneous, i.e. it becomes the CAV method. For p = m we on the other<br />

hand have that BICAV becomes the well-known Kaczmarz’s method.<br />

3.4.4 Block-<strong>Iterative</strong> Diagonally Relaxed Orthogonal Projections<br />

(BIDROP)<br />

For the general SIRT methods we described a method called DROP and we<br />

will now introduce its block-iterative generalization, which we will call Block-<br />

<strong>Iterative</strong> Diagonally Relaxed Orthogonal Projections (BIDROP).<br />

If we let Wt be positive definite diagonal matrices and Ut be symmetric positive<br />

definite matrices <strong>for</strong> t = 1, 2, . . ., p, then the algorithm <strong>for</strong> the BIDROP method<br />

looks as follows:<br />

x k+1 = x k + λkU t(k)A T t(k) W t(k)(b t(k) − Atkx k ), (3.12)<br />

where t(k) = (k mod p) + 1.<br />

The following convergence theorem is derived <strong>for</strong> the BIDROP method:<br />

Theorem 3.4 Let U be a given symmetric and positive definite matrix, and let<br />

Wt be given positive definite diagonal matrices. If <strong>for</strong> all k ≥ 0,<br />

0 < ǫ ≤ λk ≤ (2 − ǫ)/ρ(UA T t(k) W t(k)A t(k)),<br />

where ǫ is an arbitrarily small but fixed constant, then any sequence generated<br />

by (3.12) converges to a solution. If in addition x 0 ∈ R(UA T ), then the solution<br />

has minimal U −1 -norm.<br />

With only one block, i.e. p = 1, and U1 = S and W1 = W, then we have the<br />

standard DROP method.<br />

The BIDROP method is a general method since Ut and Wt is not specific given.<br />

One of the variants of BIDROP is introduced in [5] and is called BIDROP1.

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

Saved successfully!

Ooh no, something went wrong!