[aspect-devel] steady state problem

[Scott King]

You can use a large CFL value, since Aspect uses an implicit solver.   We’ve used 5 even 10 or more for low Ra problems such as in Bankenbach et al.  But I suspect you already know this.

It should be possible/easy to modify the implicit solver such that it uses Picard iteration so that the dt evolution is an iteration to steady state (make the dT/dt term go to zero).   This is discussed in various places (Peter van Keken’s thesis being one — I would suspect you can get a hold of it given your location).   From my implicit solver in ConMan this was only a few minor lines of code change.   You don’t arrive after 1 step (not sure that is easy to achieve without a different solver) but it is orders of magnitude faster than matching forward in time using the standard CFL condition.

Best,

Scott


> On Dec 1, 2017, at 8:50 AM, cedric thieulot <c.thieulot at uu.nl> wrote:
> 
> Dear all, 
> 
> we are currently looking at benchmark-type problems and we are only interested in the steady state of the system. 
> Long story short, we would like for instance to run the Blankenbach experiment (i.e. convection-box.prm) and arrive at steady state 
> as fast as possible since we are not interested in the time evolution of the model: ideally we would like to arrive after 1 timestep at the 
> steady state solution, or in other words we would like to zero the \rho c_p dT/dt term in the heat transport equation. 
> We have (naively) tried this approach by zeroing the term   bdf2_factor * scratch.phi_field[i] * scratch.phi_field[j])in advection.cc <http://advection.cc/> .
> 
> Unsurprisingly, this did not work and the code aborted. Instead of zeroing the term, we tried placing in front of it a coefficient alpha=0.5
> but this lead to a temperature field which on average was twice as large as the initial one. (if alpha=0.25, then it is approx 4times larger).
> We suspect that this alpha coefficient of ours probably should come in somewhere else too (preconditioner?) but after many hours of trial
> and error we could not figure it out. 
> What are we missing here, is there a simple way to achieve what we want? 
> 
> Cheers, and good week-end to all, 
> 
> Ludovic & Cedric
> 
> _______________________________________________
> Aspect-devel mailing list
> Aspect-devel at geodynamics.org
> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/aspect-devel

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.geodynamics.org/pipermail/aspect-devel/attachments/20171201/d84e5b4e/attachment-0001.html>