[petsc-users] Errors from large matrices

[Choi240]

Dear Jed,

I wanted to multiply those large matrices. I thought it was fast because I can get the results by just one multiplication. I think I was wrong and so I will try to get the result by the multiplication of a matrix and each vector. Thank you very much.

Joon

-------- Original message --------
Subject: Re: [petsc-users] Errors from large matrices 
From: Jed Brown <jedbrown at mcs.anl.gov> 
To: Joon hee Choi <choi240 at purdue.edu>,Barry Smith <bsmith at mcs.anl.gov> 
CC: petsc-users at mcs.anl.gov 

Joon hee Choi <choi240 at purdue.edu> writes:

> Thank you very much. I was confused between 64 bit pointers and 64 bit
> indices. The module I am using supports 64 bit indices. I succeeded in
> setting up the matrix with 4273949 x 108965941330383 size (non-zeros:
> 143599552), but failed to set up the maij matrix with 108965941330383
> x 42739470 size and 4273947 block size using MatCreateMAIJ(Mat A,
> PetscInt dof, Mat *maij). The number of non-zeros are 25495389 x 10 x
> 4273947 = 1.1 x 10^15. Do I have the method to set up the matrix
> successfully?

What do you want to do with a matrix with this enormous number of columns?

PETSc matrix data structures are not intended to be used in this way.
You should reduce the dimension of the column space so that the
corresponding dense vector can be stored. 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20130526/5ff7f21c/attachment.html>