Community
cancel
Showing results for
Did you mean:
Beginner
76 Views

## Pardiso parameters : phase, pt maxfct and mnum. Matrix inverse

Hi,

I have a few questions (sorry if some of them look basic). First, is there a way to retrieve the result of the factorisation of phase 2 (the same result one would get from Lapack's posv)? I think there is something to do with parameters pt, maxfct and mnum, but I don't understand exactly how to use them. I suppose I would have to call Pardiso two times : one with phase = 12, and the second with phase = 33 (I want to mimick posv). Is this correct?

Second, is there a way to use Pardiso to find the inverse of a matrix (like finding A^-1 from Ax = b, if x and b are known)? I was hoping to use it instead of Lapack's potri.

Thanks and have a nice day!
4 Replies
Employee
76 Views
There's really no way currently to access the intermediate factors in PARDISO. This request has been made and I can add this thread to the record for tracking purposes.

There are no special settings for this if that is what you mean. You could set b to the idenity, I. I'll let others comment on whether this is a good idea or not.

-Todd
Beginner
76 Views

Hi,

Are there plans to release Lapack routines and drivers using sparse storage format? That would be really helpful as it will decrease significantly memory storage size. I'm thinking about computing matrix inverses, as they are computer intensive.

Thanks and have a nice day!

Moderator
76 Views
then it would be something like SparseLapack :). There no such plan in the nearest versions. What we have for today - Sparse BLAS Level 1 and 2 Routines and Direct and iterative Sparse Solvers for sparse storage format.
Beginner
76 Views
Sorry for digging through an old post, but I have some news about matrix inverses (for the posterity...). I tried using Pardiso to calculate the inverse, settingthe right-hand-sideas the identity matrix. It works perfectly! The only drawback is that it takes thrice the memory space as you have to allocate space for the right-hand-side matrix and the solution matrix the same size as the input matrix. So for large problems, it is not very useful. But it works. ;)

And yes, Mr. Gennady, if I may made a request about it, I would really appreciate a Sparse Lapack. (: