1
$\begingroup$

I need to compute the inverse matrix:

$$(a_i A+B)^{-1}, \qquad i=1,\ldots,N$$

where $N$ is a large number.

$A$ and $B$ are general $M\times M$ matrices independent of $i$. The only thing that changes with $i$ is the scalar $a_{i}$.

How can I compute this efficiently?

Btw. I am using Matlab.

$\endgroup$
1
$\begingroup$

If approximate inverses are enough (and it sounds like they are, since you're doing this by computer), then you can use Neumann series. Assuming everything converges, you have: $$(a_iA+B)^{-1}=\sum_{n\ge 0} (I-a_iA-B)^{n}=\sum_{n\ge 0}\sum_{k=0}^n{n\choose k}(-a_iA)^k(I-B)^{n-k}=$$ $$=\sum_{n\ge 0}\sum_{k=0}^na_i^kM(n,k)=\sum_{k\ge 0}a_i^k\sum_{n\ge k}M(n,k)=\sum_{k\ge 0}a_i^kN(k)$$ where $$N(k)=\sum_{n\ge k}{n\choose k}(-A)^k(I-B)^{n-k}$$

Hence, you can precompute the $N(k)$ matrices, and for each $i$ just compute the series $\sum_{k\ge 0}a_i^kN(k)$. In both cases, you compute "enough" terms to have a good approximate answer.

Note: In order for the $N(k)$ matrices to be well-defined, it's probably necessary that $B$ be invertible.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.