search

BriefFiniteElementNet / BriefFiniteElement.Net

BriefFiniteElementDotNET (BFE.NET) is a library for linear-static Finite Element Method (FEM) analysis of solids and structures in .NET
https://www.bfe-framework.net
GNU Lesser General Public License v3.0
154 stars 58 forks source link

Question about available Solvers #99

Closed thaletob closed 3 years ago

thaletob commented 3 years ago

Hello again, I am actually trying out the two different solvers available. In my case I've got structures with about 11000 Quad-Shell-Elements. In contrast to your documentation the Cholesky-Solver is much faster (about 2x faster) than the Gradient Solver. (Cholesky takes about 16.5 to 17.5 s to solve, the gradient-solver needs about 28 to 41 s)

Do you have any ideas why? I am just wondering because I never got a problem where the direct solver is faster than the iterative one.

Greetings and thanks for the great work!

epsi1on commented 3 years ago

Can you please send the code snippet to reproduce the issue? Not sure what is the reason...

thaletob commented 3 years ago

When I'm testing it for smaller models, then the gradient solver is much faster. But in this case I am using it in an optimization-procedure with many solver-calls. The model-properties do not change. I just changed the solver type by adding BuiltInSolverType.ConjugateGradient to the model.Solve() call. As I mentioned... this leads to a smaller calculation time for small structures but a larger calculation time for big models. So it's more like a question than an issue...

epsi1on commented 3 years ago

It could be a bug, but cannot exactly tell you what is the reason unless trace the code...