Keywords: eigenvalues, eigenvectors, test-matrix, scipy, python
See SuiteSparse Matrix Collection / Norris, "Finite element modelling of human body parts, Laplace equation on a 2D mesh".
What does this plot tell us ? Without knowing the 2d structure, not a lot -- just a pretty picture. (The top 10 eigenvectors are for the smallest eigenvalues, the bottom 10 for the biggest. The closely-spaced big eigenvalues are tough for Arpack.)
Fwiw, numpy.linalg.eigh
took ~ 2 minutes realtime
to compute all 9604 eigenvalues and eigenvectors,
on my old 4-core iMac.
Sparse random matrices like this make tough test cases for
scipy.sparse solvers
of Ax = b
.
Shifting the eigenvalues away from 0 -- A + 1e-4 I, A + 1 I, A + 1.2 I
--
changes the runtimes and accuracy |b - Ax|
of *GMRES by large factors.
A wiki of test matrices, with plots of their spectra and/or eigenvectors, would be useful. It could help people to find "a problem roughly like mine", and collect suggestions on How-to-find-eigenmodes of several different kinds of problems.
Circular law
"Shift-invert-in-pictures" under my gists
cheers
-- denis 20 June 2020