README
1arpackmm: utility to test arpack with matrix market files.
2
3This is meant to run arpack with different options to find
4what are the ones that are the best for your particular
5problem.
6
7Typically: computing small eigen values may breakdown and/or
8be slow (arpack is good at finding large eigen values only).
9To compute small eigen values, it may be a better choice
10to look for large eigen values with invert or shift+invert.
11
12How to use this utility ?
13Start simple, then, increase complexity if/when needed:
14- defaults to: standard real symmetric eigen problem to be
15 solved accurately (double precision).
16- list all possible options with --help.
17- add --verbose or --debug to get more informations.
18- if solve breaks down, you may:
19 - play with --nbCV (increase workspace size).
20 - play with --shiftReal/Imag and/or --invert.
21 - if arpack mode > 1, change solver with --slv.
22 - if arpack mode > 1, change solver parameters with --slv
23 (typically: pivoting threshold, solver tolerance, ...).
24 - try --restart (restart from previous eigen basis
25 approximation computed by previous run).
26 - etc...
27
28Note: using openblas/mkl/atlas instead of netlib blas/lapack
29may impact results. In some cases, it may also fail checks.
30Use --noCheck to skip checks.
31
32To build this utility, you need:
33- to use a fortran compiler which supports iso_c_binding.
34- to have installed eigen3 (to deal with the RCI).
35- to configure arpack-ng this way:
36 - autotools:
37 ~arpack-ng> ./configure --enable-icb-exmm; make all check
38 - cmake:
39 ~arpack-ng/build> cmake -D ICBEXMM=ON ..; make all test
40