sparse equation system

Ion_C_ — Fri, 14 Feb 2014 07:10:33 GMT

Hi!

In the below code, I have declared one matrix and one vector: A[3][3] and B[3]. The matrix A is symmetric from main diagonal. I converted the matrix A from dense format to CSR format, and from CSR format to coordinate format using the function mkl_dcsrcoo. All conversions are ok, but I can not solve correctly the system AxY=B. I obtained a wrong solution. The correct solution is: Y=[0.5 0.5 0.5]

How can I obtain the correct solution?

The code is following:

#include "stdafx.h"
#include <stdlib.h>
#include <stdio.h>
#include<conio.h>
#include<math.h>
#include "mkl.h"
#include "mkl_lapacke.h"

int _tmain(int argc, _TCHAR* argv[])
{
	double A[3][3] = {{2.5,-0.5,0.},{-0.5,5,-0.5},{0,-0.5,2.5}};
	double B[3]={1.0,2.0,1.0};
	MKL_INT  rows = 3, cols = 3, nnz = 7,info;
	MKL_INT job[6]={0,0,0,2,nnz,2}; 
	//for conversion of matrix A to CSR format
	double  *arrV      = new double[nnz]; 
    MKL_INT *Columns   = new MKL_INT[nnz];
    MKL_INT *Rows      = new MKL_INT[rows+1];

	mkl_ddnscsr(job,&rows, &cols, A[0], &rows, arrV, Columns, Rows,&info);

	printf("\Values\n");
	for(int i=0;i<nnz;i++)
		printf("%.3f ",arrV);
	printf("\nColumns\n");
	for(int i=0;i<nnz;i++)
		printf("%d ",Columns);
 	printf("\nRows\n");
	for(int i=0;i<rows+1;i++)
		printf("%d ",Rows);
	printf("\n");
	
	//convert to column format
	double  *acoo      = new double[nnz]; 
	int     *rowind    = new int[nnz]; 
	int     *colind	   = new int[nnz]; 
	
	MKL_INT job1[6]={0,0,0,0,nnz,3}; 
	mkl_dcsrcoo(job1,&rows,arrV,Columns,Rows,&nnz,acoo,rowind,colind,&info);
	printf("info mkl_dcsrcoo=%d\nacoo\n",info);
	for(int i=0;i<nnz;i++)
		printf("%.3f ",acoo);
	printf("\n");
	for(int i=0;i<nnz;i++)
		printf("%d ",rowind);
	printf("\n");
	for(int i=0;i<nnz;i++)
		printf("%d ",colind);

	MKL_INT colsC=3;
	double alpha=1;
	double *Y=new double[3];
	mkl_dcoosm ("n",&rows,&colsC,&alpha,"TUNC",acoo,rowind,colind,&nnz,B,&rows,Y,&colsC);
	printf("\n");
	for(int i=0;i<3;i++)
		printf("%.3f ",Y);

	delete[] arrV;
	delete[] Columns;
	delete[] Rows;
	delete[] acoo;
	delete[] rowind;
	delete[] colind;
	_getch();
	return 0;
}

For the MATDESCRA argument,

mecej4 — Fri, 14 Feb 2014 20:42:14 GMT

For the MATDESCRA argument, you gave "TUNC". The first letter signifies that the matrix being passed is upper triangular. This would make sense only if your matrix, which is a square matrix, has already been factorized into an LU or LDU product. Please note that with some Lapack/MKL routines a symmetric matrix is passed with the convention of including only the upper triangular part. However, this part is NOT the U factor that I just mentioned.

topic sparse equation system in Intel® oneAPI Math Kernel Library

sparse equation system

For the MATDESCRA argument,