Skip to content

Fix eigenvector computation for zero matrix

Carsten Gräser requested to merge feature/eigenvalues-for-zeromatrix into master

For a zero matrix eigenValuesVectorsImpl() currently still divides by the infinity norm for preconditioning. While the computed values seem to be correct, but in view of nan entries I consider this a lucky coincidence. One could avoid the scaling for a zero matrix, but if this is checked anyway, one can directly return the exact result. This is what's implemented here.

Merge request reports