[aspect-devel] ASPECT question

[Wolfgang Bangerth]

>>> If the limit is in the code then the solver fails to converge for large
>>> thermal viscosity exponents after a few hundred steps, but if I remove
>>> these bounds it appears to work fine.
>>
>> What is your viscosity range in that case? I haven't tested 'very large'
>> viscosity contrasts yet. There might be stuff that needs changing in
>> ASPECT to make it work. One thing you can try is using a constant
>> viscosity per cell for example.
>
> It starts to fail for viscosity ranges around 1e5 and above.  I don’t think
> this is because of the large range, but rather because of the sharp cutoff
> in viscosity change caused by the min(max(…)) in simple.cc:43.  If I turn
> off this limit then large viscosity ranges up to at least 1e7 work fine.

That is (i) rather weird, given that the cutoff is at least continuous, (ii) 
also good to know. I don't think either Timo or I had ever tried to solve a 
model with such large viscosity variations before and it's good to know that 
it can be done.


> For now I’ll change the limits in my own code.  I just wanted to make sure
> there wasn’t a particular reason (e.g. invalid results) the limits were
> there in the first place.

Like Timo said, these are legacy. I think for benchmarks like the one you're 
implementing, it is sensible to implement a material model specifically for 
the benchmark, based on the new and simpler 'simpler' material model.

Best
  W.

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