Hi Mark, Thanks for your reply. Below is the output if I call KSPComputeEigenvalues
0.330475 -0.0485014 0.521211 0.417409 0.684726 -0.377126 0.885941 0.354342 0.957845 -0.0508471 0.964676 -0.241642 1.05921 0.0742963 1.82065 -0.0209096 I have the following questions: * These eigenvalues are sorted according to the magnitudes. so "lowest" means smallest magnitude and "highest" means largest magnitude in your previous email? * I understand that if the preconditioner is perfect, all the eigenvalues should be (1,0). Since my preconditioner is not perfect, to understand its performance, is it correct to say that, I need to keep an eye on the eigenvalues whose distance to (1,0) are the furthest? * How does petsc decides how many eigenvalues to output in KSPComputeEigenvalues. I am solving a set of linear systems, sometimes KSPComputeEigenvalues outputs 8 eigenvalues, sometimes it outputs just 2 eigenvalues. * In the output which I showed above, are these the ones with the smallest magnitude and also the ones with the largest magnitudes? and what's between are all ignored? If this is the case, which ones are the "lowest" and which ones are the "highest"? Thanks for your help and sorry for so many questions, Feng ________________________________ From: Mark Adams <[email protected]> Sent: 04 October 2022 17:18 To: feng wang <[email protected]> Cc: [email protected] <[email protected]> Subject: Re: [petsc-users] clarification on extreme eigenvalues from KSPComputeEigenvalues The extreme eigenvalues are the lowest and highest. A perfect preconditioner would give all eigenvalues = 1.0 Mark On Tue, Oct 4, 2022 at 1:03 PM feng wang <[email protected]<mailto:[email protected]>> wrote: Dear All, I am using the KSPComputeEigenvalues to understand the performance of my preconditioner, and I am using the right-preconditioned GMRES with ASM. In the user guide, it says this routine computes the extreme eigenvalues of the preconditioned operator. If I understand it correctly, these eigenvalues are the ones furthest away from (1,0)? If the preconditioning is perfect, all the eigenvalues should be (1,0). Thanks, Feng
