Problem with number of Materials

[ValletRomain]

Hello, @vincentchabannes , @romainhild , @prudhomm

With toolboxe cfpdes, I try to run a resolution of coupled elastic, thermic and magnetism equations on magnet with 14 Helices.

I have about 300 Materials and the mesh have 120000 nDof. I resolve with Newton solver and lu method. My simulation doesn't converge and take a lot of time : for 2 iterations with 1 core, it takes 1100s->3h !

I try to do this : I reduce the number of materials to 300 to about 40 without change the problem. I resolve with Newton solver and lu method. My simulation converges, gives good results (good physical values) and takes about 200s.

I think, the number of Materials changes the performance of solver. Is it right ?

My JSON and CFG file are here :

• version with detailled Materials : cfg, json
• version with grouped Materials : cfg, json

[vincentchabannes]

there is no reason that the solver didn't converge (if the pde is the same when you use 40 mats and the solver works). The algebraic system is identical. Probably a json error somewhere (else there is a bug in cfpdes). For the perf, indeed the time can increase with the number of mat (but normally not too much, except if you use symbol mat prop without matname).

[ValletRomain]

Hello @vincentchabannes ,

I put new version of CFG and JSON file with less variables in Materials and with section PostProcess commented.

The new results are :

• with 300 Materials : 4640.88 s for solver and 5.921897e+03s for execution time
• with 40 Materials : 88.3281 s for solver and 1.209600e+02s for execution time

My JSON and CFG files are here :

• version with detailled Materials : cfg, json
• version with grouped Materials : cfg, json

I give you the mesh.

[github-actions[bot]]

Stale issue message

[prudhomm]

@feelpp/hifimagnet could you check this testcase again ? @jermuzet?

[jermuzet]

The files don't seem to exist anymore, or I can't access them, but the elasto-thermo-magnetic model with the magnet M19061901 works well with the v109 and v110. This model has 322 materials and converges in sequential and parallel computation.