arpack-ng
arpack-ng copied to clipboard
Consecutive identical eigenproblems give different results
Expected behavior
If 2 identical consecutive eigenproblems are solved the results should be identical
Actual behavior
They are not because the lapack random number generator is initialized only once for the first eigenproblem and never again. Hence the starting vector for the 2 identical consecutive eigenproblems are not the same and the final result differes.
Where/how to reproduce the problem
- arpack-ng: release
- OS: ubuntu
- compiler: gcc
Steps to reproduce the problem
Solve 2 identical consecutive eigenvalue problems.
No solution for now. Duplicate : https://github.com/opencollab/arpack-ng/issues/218
and the final result differes.
BTW, how much in norm for instance? 1.e-6 in double? 1.e-3 in float? more?