[aspect-devel] Onset of convection, time step choice

Wolfgang Bangerth bangerth at tamu.edu
Mon Oct 10 08:50:22 PDT 2016


> I am teaching geodynamic modeling this term, and asked my students as an
> introductory assignment to try to use ASPECT to determine experimentally the
> critical Rayleigh number for the onset of convection for an isoviscous fluid,
> heated from below with free-slip boundary conditions. The students did this by
> modifying cookbooks/convection-box.prm. One student found that convection
> began at Ra less then half the critical Ra for the chosen aspect ratio of 4:3.
> Of course, it turned out that too large a timestep was allowed. If the
> timestep size is limited to, say 100 Myr, we recover the correct value of
> Ra_{cr} within ~5% even for relatively coarse meshes (refinement level 4). I
> was a bit surprised, however, that the default timestep taken by ASPECT was so
> large. In fact, we had convergence problems for higher refinement levels for
> the default maximum timestep. For a problem like the one that we were running,
> a reasonable maximum timestep seems to be L^2/kappa (thermal diffusion
> timescale), where L is a characteristic length scale for the domain. Does it
> seem reasonable to change the way that ASPECT computes the maximum time step
> based on this idea rather than using 10^300 (the current default value, which
> seems entirely arbitrary)? I am happy to submit a pull request to do this, but
> don't want to waste my time if it's unlikely to be accepted.

As Juliane already pointed out, (i) there is already an option to do that, but 
(ii) as you mentioned you don't want to make it the default.

I don't actually have much of an opinion (I'm going to blame it on jetlag) but 
if your opinion is that this shouldn't be the default, the making L^2/kappa 
the maximum time step also seems like the wrong solution. Or do I misunderstand?

One option I see is to make the work you and your students did into a 
cookbook. I think that would be great. That would also provide a platform 
where we could talk about the issue in a highlighted comment.


Wolfgang Bangerth          email:                 bangerth at colostate.edu
                            www: http://www.math.colostate.edu/~bangerth/

More information about the Aspect-devel mailing list