Page 1 of 1


Posted: 11 Nov 2013, 12:14
by Martina
Hi all,

I have run now the ParStokes Solver and I am comparing to the "normal" Navier-Stokes Solver (with FlowModel=Stokes) and I get in some regions rather large differences. I haven't yet played around a lot with the different parameters (for ex. convergence tolerances), but I was wondering what has already been done in terms of comparison?
Am I right assuming that the solution of the "normal" Navier-Stokes (direct solution for the linear part and fully converged for the nonlinear part) should be considered as the "real" one, so in principle I should tune the different parameters from the ParStokes to get as close as possible to the Navier-Stokes solution?


P.S. To Peter/Fab/and others from the last Elmer course: In the meanwhile my ParStokes Solver works - as long as I do restarts only from runs where the ParStokes Solver is already present in the sif-file. I don't know what error I had last week in the end when I had these additional errors, I simply started over and it worked right away.

Re: ParStokes

Posted: 13 Nov 2013, 15:40
by mika

The solutions of the standard NS solver and the ParStokes solver have been compared carefully for several problem setups (including the ISMIP-HOM benchmark cases for which reference results have also been published in the literature) and the solutions have generally been in good agreement. In cases where the finite element stability is close to become an issue (usually via introducing elements with high aspect ratios) larger (but usually localized) differences have been observed, but this may affect the accuracy of the standard NS solution as well and one may pose the question whether comparing two nearly unstable solutions is useful.

Best regards,

Re: ParStokes

Posted: 13 Nov 2013, 19:46
by Martina
Hi Mika,

thanks for your feedback.
Then I should probably play around a bit more with the parameters of the ParSolver and see what happens. I have also to add, that the biggest differences occur in the areas of fast flow caused by basal sliding, that is in areas of very fine mesh resolution (refinement automatically done with yams accordingly to derivations of velocity data) and very high velocities. So maybe I'm already at the limit of the stability? Probably that needs a careful closer look into. I'm not sure if I have time for that right now (I was mainly looking into that because of the Elmer-course), but I'll keep that in mind!