Conversation

Notices

1. Tomas SA2TMS (tomas@social.umeahackerspace.se)'s status on Tuesday, 16-Oct-2018 17:09:36 EDT Tomas SA2TMS
   Evening project: implementing a solver in GNU Octave for Laplace's equation, to have a simple libre system for FEM:ing electrostatics. Using 1.7x1.7x1.7 mm elements in a 150x150x150 mm domain results in a 516600x516600 sparse equation system (nnz=3553440), which UMFPACK quite happily solves in a few minutes
   1. Tomas SA2TMS (tomas@social.umeahackerspace.se)'s status on Tuesday, 16-Oct-2018 17:14:08 EDT Tomas SA2TMS

      in reply to
      Forgot a picture. This is a slice showing voltages in the XZ plane https://social.umeahackerspace.se/attachment/456969
   2. Tomas SA2TMS (tomas@social.umeahackerspace.se)'s status on Wednesday, 17-Oct-2018 06:07:42 EDT Tomas SA2TMS

      in reply to
      Trying to go down to 1.0 mm elements, but alas:
      
      warning: warning -3, at line 683 in file ../Supernodal/cholmod_super_symbolic.c: problem too large
      
      I probably have to switch to something like a biconjugate solver
      1. Tomas SA2TMS (tomas@social.umeahackerspace.se)'s status on Wednesday, 17-Oct-2018 14:46:08 EDT Tomas SA2TMS

         in reply to
         Turns out Octave has an implementation of the biconjugate gradient stabilized method (BiCGSTAB) which is much much faster and memory efficient than good old A\rhs or LU factorization