[aspect-devel] Benchmark

[I. van Zelst]

Dear Timo,

Problem solved! I set the tolerance of the solver to 1e-9, instead of the
default 1e-7 and now, the system is solved correctly, as you can see in the
attached figures. I am still wondering though, why I can't run the post
processor in parallel. Do you happen to know that?

Kind regards,
Iris


2014-07-02 14:27 GMT+02:00 I. van Zelst <i.vanzelst at students.uu.nl>:

> Dear Timo,
>
> I also think I have eliminated the warning, you drew my attention to. The
> new .cc file and input file are attached.
>
> Kind regards,
> Iris
>
>
> 2014-07-02 14:17 GMT+02:00 I. van Zelst <i.vanzelst at students.uu.nl>:
>
> Dear Timo,
>>
>> I ran the benchmark with only 1 core, and I got the same results as you!
>> Which makes me wonder, why it can only be run on 1 core. Do you know that?
>> However, besides running your test, I also ran a model with increased
>> resolution of ncellx = 32 or ref = 5. As soon as I did that, the errors and
>> maximum and minimum pressure were all wrong again, as you can see in the
>> enclosed figures. The reason for this is probably that the corners of the
>> domain show very extreme pressures, almost as if the model is not well
>> resolved there. Of course, these excesses in pressure lead automatically to
>> an increase in error. Could the reason for these badly resolved pressures
>> in the corners of the domain be that the solver can not handle the
>> resulting matrix very well, because of the many boundary conditions at the
>> corner elements?
>> Thanks in advance!
>>
>> Kind regards,
>> Iris
>>
>>
>> 2014-07-01 20:56 GMT+02:00 Timo Heister <heister at clemson.edu>:
>>
>> > the updated figures that are attached. I also enclose the new .cc file.
>>> Do
>>> > you have any other ideas for this behaviour?
>>>
>>> Maybe:
>>> ./ This warning highlights a problem in your code:
>>> burstedde.cc: In member function ‘dealii::Tensor<1, dim>
>>> aspect::BursteddeBenchmark::BursteddeGravity<dim>::gravity_vector(const
>>> dealii::Point<dim, double>&) const [with int dim = 2]’:
>>> burstedde.cc:512:69: warning: ‘beta’ is used uninitialized in this
>>> function [-Wuninitialized]
>>>        const double mu=exp(1. - beta * (x*(1.-x)+y*(1.-y) + z*(1.-z)));
>>> ./ Your error computations won't work correctly as is when using more
>>> than one MPI process. Are you?
>>> ./ This doesn't look too bad:
>>>      Errors u_L1, p_L1, u_L2, p_L2: 2.781543e-04, 1.430527e-03,
>>> 3.116033e-04, 2.626537e-03 ref=2
>>>      Errors u_L1, p_L1, u_L2, p_L2: 3.434312e-05, 3.123564e-04,
>>> 3.892204e-05, 6.297030e-04 ref=3
>>>      Errors u_L1, p_L1, u_L2, p_L2: 4.505512e-06, 8.855139e-05,
>>> 4.880252e-06, 1.904190e-04 ref=4
>>>
>>>
>>> --
>>> Timo Heister
>>> http://www.math.clemson.edu/~heister/
>>> _______________________________________________
>>> 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/20140702/03333f3b/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: errors.pdf
Type: application/pdf
Size: 15777 bytes
Desc: not available
URL: <http://lists.geodynamics.org/pipermail/aspect-devel/attachments/20140702/03333f3b/attachment-0003.pdf>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: maximum_pressure.pdf
Type: application/pdf
Size: 9424 bytes
Desc: not available
URL: <http://lists.geodynamics.org/pipermail/aspect-devel/attachments/20140702/03333f3b/attachment-0004.pdf>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: minimum_pressure.pdf
Type: application/pdf
Size: 9153 bytes
Desc: not available
URL: <http://lists.geodynamics.org/pipermail/aspect-devel/attachments/20140702/03333f3b/attachment-0005.pdf>