[aspect-devel] Refinement Strategy

Timo Heister heister at math.tamu.edu
Wed Jan 25 07:43:11 PST 2012

Hi everyone,

Please update deal.II to r24920 or higher. It contains a bug fix for
adaptive mesh refinement.

I changed and enhanced the refinement strategies in aspect today.
Sorry Thomas, I ripped some of your stuff out. You all might need to
update your .prms, but the default should be sensible.

The options for "Strategy" are:

Temperature: Kelly Estimator for the temperature. This is this old
model that does not see density jumps, but should be fine if you run
with a smooth density.

Normalized density and temperature: this is the first version that
looks at jumps in the density and temperature and was the version used
until today. The big problem is, that both indictators are normalized
first before the maximum between them is taken. This gives strange
results if you don't have a density jump and it can not see how
important the density jump is, if it is there. It also looks like it
normally keeps the cells where the density jump is always on the
finest level. This might be desirable or not...

Weighted density and temperature: This method does not normalize the
density and temperature errors but tries to find a sensible balance
between them. This contains a tuning parameter (not exposed as of
now), that might need to be adapted. Advantage: should work even if
there is no density jump.

Density c_p temperature: new indicator that just looks at the quantity
rho*c_p*temperature and tries to resolve it. This is currently the
default and is hopefully superior to the other options. It does not
need a tuning parameter, works in all geometries, and also if the
density is smooth.

Thomas (or others): it would be great, if you could test some of these options.

Timo Heister

More information about the Aspect-devel mailing list