1
$\begingroup$

Let

$$X_i = \left(I - P\left(I - t_it_i^T\right)\right)^{-1}$$

where $P$ is an $N\times N$ matrix and $t_i$ is a vector of $N$ elements.

Is there a way to simplify this expression in order to calculate the inverse matrix only once for all $i$ values?

$\endgroup$
2

2 Answers 2

4
$\begingroup$

I think I found a solution to this problem. As Federico Poloni also noted, we can use the Shelman-Morisson formula.

We can write $X_i=(I-P+Pt_it_i^T)^{-1}$

Let $A=I-P$, $u=Pt_i$ and $v=t_i$, then:

$X_i=(A+uv^{T})^{-1}$

We can use the Sheman-Morisson formulla:

$(A+uv^T)^{-1}=A^{-1}-\frac{A^{-1}uv^{T}A^{-1}}{1+v^{T}A^{-1}u}$

This way we can calculate $(I-P)^{-1}$(provided $I-P$ is invertible) only once for all values of $i$.

$\endgroup$
1
$\begingroup$

If $\|P\|<1$ then $X_i = \sum_{k\ge 0} (P-Pt_i^{\top}t_i)^k$. This is analytic in $t_i$ and in $P$. Maybe this helps.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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