Dear all,
I am having an "issue" using pzhegvx routine to compute eigenpairs of a complex matrix. Subsequent runs of the same software exhibits non-deterministic behavior, the problem is evident also when using 16 processes. Even if the computed eigenvalues seem to legalistically lie into the convergence interval as specified by ABSTOL parameter, I would like to ask if such a behavior must be expected, because of the inevitable inherent error due to the finite arithmetic, or not.
thanks in advance
Loriano

