# choice of iparms in Pardiso

## choice of iparms in Pardiso

hello,

I would like to solve a linear system of type Ax = b with A,B and x in the file joined.

The matrix A is well-conditioned, so I must be wrong with the iparms :

357         iparm[0] = 1; /* No solver default */
358         iparm[1] = 2; /* Fill-in reordering from METIS */
359         iparm[2] = 8;
360         iparm[3] = 0; /* CGS */
361         iparm[4] = 0; /* No user fill-in reducing permutation */
362         iparm[5] = 0; /* Write solution into x */
363         iparm[6] = 0; /* Not in use */
364         iparm[7] = 0; /* Max numbers of iterative refinement steps */
365         iparm[8] = 0; /* Not in use */
366         iparm[9] = 13; /* Perturb the pivot elements with 1E-13 */
367         iparm[10] = 1; /* Use nonsymmetric permutation and scaling MPS */
368         iparm[11] = 0; /* Not in use */
369         iparm[12] = 0; /* Not in use */
370         iparm[13] = 0; /* Output: Number of perturbed pivots */
371         iparm[14] = 0; /* Not in use */
372         iparm[15] = 0; /* Not in use */
373         iparm[16] = 0; /* Not in use */
374         iparm[17] = -1; /* Output: Number of nonzeros in the factor LU */
375         iparm[18] = -1; /* Output: Mflops for LU factorization */
376         iparm[19] = 0; /* Output: Numbers of CG Iterations */
377         iparm[27] = 1; /* check the data structure */
378         iparm[31] = 1; /* iterative solver*/
380         maxfct = 1; /* Maximum number of numerical factorizations. */
381         mnum = 1; /* Which factorization to use. */

The Package ID of mkl is : l_mkl_p_10.0.011

Thanks for any help you can give me
Antoine

ВложениеРазмер
752.77 КБ
5 сообщений / 0 новое
Пожалуйста, обратитесь к странице Уведомление об оптимизации для более подробной информации относительно производительности и оптимизации в программных продуктах компании Intel.

In producing the attachment, you seem to have used a version of tar whose output my Cygwin version cannot process:

tar: Ignoring unknown extended header keyword `SCHILY.dev'
tar: Ignoring unknown extended header keyword `SCHILY.ino'
tar: Ignoring unknown extended header keyword `SCHILY.nlink'

ok, you can try this

## Вложения:

ВложениеРазмер
752.97 КБ

I can now extract and view the files, thanks.

However, it is not clear what the files contain. Does 36_A.mat contain the upper triangle of a symmetric matrix? Or does it contain a triangular matrix? Do you expect the matrix to be positive-definite?

What is the reason behind giving values for the vector x? Your matrix appears to be 252 X 252, but the files for x and b have 1296 lines each. How do you explain this?

Sorry, you are right, I should have specified my problem.

The matrix I sent you was a 1296 x 1296 matrix in COO format.
I send it to you in CSR format, it will be easier.
The new file for A is composed of 4 lines : the size of data (row column val), the ranks of the first element of the row, the columns, and the values.

This matrix has no particularity : non-symmetrical, not positive-definite...

I give 2 vectors x here : one obtained with Matlab (very good acuracy), and one calculated by Pardiso. Maybe you need it to compare, or just seeing how wrong is Pardiso with my parameters.

ВложениеРазмер
78.09 КБ