search

davydden / large-strain-matrix-free

A repository with code for the paper "A matrix-free approach for finite-strain hyperelastic problems using geometric multigrid"
GNU Lesser General Public License v2.1
2 stars 6 forks source link

more stable test output #45

Closed davydden closed 6 years ago

davydden commented 6 years ago

what I did in another continuum mechanics code is to switch to a separate blessed output which does not contain iteration counts, something along the lines:

Grid:
  Reference volume: 3.12073e-07
Triangulation:
  Number of active cells: 64
  Number of degrees of freedom: 960

Timestep 1 @ 0.1s
Converged in 3 Newton iterations
Solution @ 0 0.000333
  displacement:         -3.38783e-21 5.50048e-06
  deformation gradient: 0.998386 -9.49904e-17 1.67878e-16 1.00468

Timestep 2 @ 0.2s
Converged in 4 Newton iterations
Solution @ 0 0.000333
  displacement:         -1.10399e-21 1.10425e-05
  deformation gradient: 0.996739 -2.31032e-17 4.63221e-17 1.00941

shall we do the same here? We can still keep residual norms at convergence and alike. Or maybe output average number of iterations, like 

Average Linear solver iterations:   6xx

(putting exact number might give unstable blessed outputs again).

p.s. of course, the standard output will stay as is and we will still see the nubmer of CG iterations per step.

@masterleinad @jppelteret if you support this idea, I can quickly do this.

jppelteret commented 6 years ago

I do similar for my own codes and the elasticity based deal.II tests, so I support the idea.