Introduction

How to test the complex serial driver for BLOPEX.

This is a c program (serial_driver) which was created for testing of the complex additions to blopex_abstract. It is similar to blopex_serial_double but 1. deals only with complex matrices, 1. matmultivec, pcg_multi, and multivector routines have been modified to handle complex matrices and vectors, 1. calls lobpcg_solve_complex, and 1. the serial_driver.c has been enhanced for additional test options.

The serial_driver does not require installation of Petsc or Hypre.

This interface is for basic testing of the BLOPEX code and does not does not utilize mpirun.

Compile/Link serial_driver

For the following assume all BLOPEX source is installed under directory BLOPEX.

If BLOPEX library has not been created then do this first.

Change directory to BLOPEX abstract code directory and issue make command.

>>cd BLOPEX/blopex_abstract
>>make

Note: On some computers, one may have to edit Makefile.inc in the blopex_abstract directory and change the lines:

# cc = gcc
 cc = mex

to

 cc = gcc
# cc = mex

as the mex compiler may not be installed.

Next, change directory to the code for the complex serial_driver.

>>cd BLOPEX/blopex_serial_complex

For example:

LOBPCG_ROOT_DIR =  ../../blopex_abstract
CFLAGS = -Wall -g
LAPACKLIB = -llapack

>> make

Test Execution

To run the executable (which is in ../blopex_serial_double/driver), shift to that directory and enter command

>> ./serial_driver R 100 10 

>> ./serial_driver L test1a.txt test1x.txt

The second form loads matrix from test1a.txt, initial eigenvectors from test1x.txt and solves for the same number of eigenvalues as initial eigenvectors.

Test file creation

These files are formated as follows: ascii lines with cr. 1st line is '%d %d\n' number of rows, number of columns. All other lines, '%22.16e %22.16e\n' real part of number, imaginary part of number. Numbers are in standard fortran row rank order.

These files can be created in Matlab by 1st setting up the matrix and saving using the following (matrix in variable x and x is complex).

fid = fopen('c:\text1a.txt','w');
[row,col]=size(x);
fprintf(fid,'%d %d\n',row,col);
for j=1:4 
   for i=1:6 
      fprintf(fid,'%22.16e   %22.16e\n',real(x(i,j)),imag(x(i,j)));
   end
end
fclose(fid);

The matrix to solve should be hermitian positive definite.

The matrix of initial values should have the same number of rows as the matrix to solve and the number of columns determines the number of eivenvalues to solve for.

Test Results

test1

./serial_driver L test1a.txt test1x.txt

Eigenvalues computed

Eigenvalue lambda       Residual              
-6.7895342371693745e+00  6.9799183444283968e-07
-6.1261513439877628e+00  6.8987519527870938e-07
-5.8103745267211648e+00  6.0087666718421241e-07
-5.2386164184952140e+00  6.5674786833799589e-07
-4.8808702224124803e+00  7.5848791467513662e-07

31 iterations
CPU time used = 4.1957000000000001e-02 

test2

./serial_driver L test2a.txt test2x.txt

Eigenvalues computed

Eigenvalue lambda       Residual              
-1.1080244958656804e+01  8.7715396842764707e-07
-1.0619031629224237e+01  8.3380688568426980e-07
-1.0091481158509795e+01  9.3749198003235374e-07
-9.9117793842421449e+00  8.9288849369553002e-07
-9.7514313051991888e+00  9.7493416067963172e-07
-9.3207970608165578e+00  9.8447217142738573e-07
-8.8124520024677189e+00  6.4165207440313912e-07
-8.5728211944494994e+00  7.6010455033173900e-07
-8.3669043134042269e+00  9.8320903772396290e-07
-7.9116588662085574e+00  7.6053312739064901e-07

76 iterations
CPU time used = 6.3530100000000000e-01 

random

./serial_driver R 100 10

Eigenvalues computed

Eigenvalue lambda       Residual              
-1.1210091685718119e+01  4.7967132991532764e-07
-1.0501158146809857e+01  4.4024260766008527e-07
-9.8940023904399990e+00  5.3411187229378858e-07
-9.7773232788479785e+00  8.8890256227145167e-07
-9.5346640107774565e+00  8.0222425924588777e-07
-9.2501084356249024e+00  6.7195348176370916e-07
-8.9349365221119932e+00  8.5500400698241940e-07
-8.8367798054460582e+00  6.3309352129526724e-07
-8.3516247436515307e+00  7.9788232965401455e-07
-7.8577158442232040e+00  9.5954673989161707e-07

86 iterations
CPU time used = 5.2436099999999997e-01