0

I need to solve generalised eigenvalue problems that stem from solid mechanics (mass/stiffness matrices, (K-l*M)x=0). The systems dimension range from 500k to 15M degrees of freedom. Matrices are real and symmetric. I am currenty using Arpack.jl to solve the problem, but memory consumption simply blows up with the number of degrees of freedom and time performances are OK. Do you know if there are good Julia alternatives for systems of such dimensions (e.g. PCG Lanzcos algorithm o Supernode method)? Has anyone compared Arpack.jl with KrylovKit.jl for this type of problems?

Smoop94
  • 73
  • 3
  • 1
    can you post some code which you call and produces the errors? (MWE) and also are you making your matrices sparse to start off with? will the entries be sparse in those matrices? Does the memory problem happen before calling the libraries or within those libraries? Can you calculate that solution without the libraries? – Vass Jan 24 '23 at 20:39
  • 1
    Try asking in Julia Discourse. Sometimes better for wider algo discussions. – Dan Getz Jan 25 '23 at 02:21
  • @Vass I posed a general question but I do not have errors. Matrices are sparse and memory performaces are just those expected from Arpack. Results form Arpack are good and validated. I am searching for a better alternative when I cannot handle the big memory required by that library. – Smoop94 Jan 25 '23 at 20:15

0 Answers0