Linear Algebra

A Russian README is here.

Usually, when solving large linear systems of algebraic equations,
determinants are not necessary, but sometimes you still want them to know
(or at least their signs).

Here I present a few well tested packages, familiar to me, which are capable of computing determinants for various matrices (including the sparse ones).

The package was assembled partly through
the new Trefor , but here is a pure Fortran version.

After unpacking the archive determ.tgz
one has to go inside the new directory determinant/ and type there:

make -f ifcdeterm.mak

-- this is for ifort, or

make -f mgfdeterm.mak

-- this is for gfortran

One can embed other compilers in the same makefile (see, for example, the commented variants, uncomment and edit them).

To test the minors of a matrix type:

make -f ifcdeterm.mak alargematr

make -f mgfdeterm.mak alargematr

You will get xalargematr.exe and let it run.

Specify the size of the matrix N when prompted. Normally, it works OK up to N ~ 200. For larger N the determinants may become too large, but if the matrix is good and normalized a much larger N can be taken. One has to press "enter" many times (edit the source tLargematrAlone.f90 to remove "call mypause" if you do not want to pause).

When it stops, analyze the file

compDet015.res if N = 15
compDet125.res if N = 125, etc.

In this file, the first three columns give the diagonal minors of the
matrix computed in three different ways (the first two columns are almost one way via lapack).
All three answers must match.

I have used different versions of  lapack libraries (usually  they are installed on
linux machines, but here I collected all the necessary source code from lapack).

There is also a large file test_mat.f90 from the site

- in my package only a small part of this code is used, but you
can play with other tests (I have inserted only a

There is also a file ma28Timmes.f90 from

The main program in the package tLargematrAlone.f90 -- after the row
238 there the matrix amat (i, k) is filled with random numbers and unities.

I.e. this is a package for an arbitrary matrix. It should work as well for
symmetric matrices, but I have not investigated here special packages for symmetric matrices. The idea is that the special packages
should be faster, but here for N< 200 everything is not bad. 

Thus, you need to  substitute your matrix in place of  my random one if you want to find its minors.

The same J.Burkardt has collected programs for symmetric matrices:

They are based on the book by
Alan George, Joseph W.-H. Liu,
Computer Solution of Large Sparse Positive Definite Matrices
Prentice Hall, 1981,

but I do not mastered these programs.

I used these books :

S. Pissanetsky: Sparse Matrix Technology

O. Oesterby, Z. Zlatev: Direct methods for sparse matrices

J. Rice - Matrix computations and mathematical software

SeB - Sergei I. Blinnikov