[aspect-devel] aspect: error related to changing number of processors

[Magali Billen]

Hi Timo,

I just learned a bunch of svn commands from Eric, that allowed me to check for other files
that were different, and I found two that I had forgotten.  And, I found the problem.

First, I forgot that I had changed the temperature initial conditions, this is why when
you ran the model it finished in two time-steps.

Here's the change I made:
--- source/initial_conditions/box.cc	(revision 1118)
+++ source/initial_conditions/box.cc	(working copy)
@@ -44,7 +44,8 @@
       double perturbation = 1;
       for (unsigned int d=0; d<dim; ++d)
         perturbation *= std::sin(numbers::PI*position[d]/geometry->get_extents()[d]);
-      return 1 + perturbation/10;
+      // MIB return 1 + perturbation/10;
+	return 1400*(1 + perturbation/10);

(Note, I had to "hard-wire" this 1400 value, because as I understand it, given how Aspect is set-up right now,
there's no easy way to use information about the boundary conditions in defining the initial conditions. It
would certainly be helpful to make this easier in the future.).

Second, in Stokes solver I had changed the accuracy from 1e-12 to 1e-15, (this was leftover from looking
for ways to decrease the heat flux on the zero heat-flux side-walls, but didn't have any effect).

- When I change this back to 1e-12, the model runs fine on 4 processors. Its interesting that
using this smaller accuracy is only a problem on 4 processors, does that make any sense to you?

Thanks for responding to my e-mail…and sorry to have sent you on a wild goose-chase.
Magali


On Aug 30, 2012, at 1:53 PM, Timo Heister wrote:

> Dear Magali,
> 
>> Exception on processing:
>> Iterative method reported convergence failure in step 7551 with residual 3.15628e+07
>> Aborting!
>> 
>> (note it was only on step 59, but is stating it has an error in step 7551 ??)
> 
> This means that the linear solver did not converge after 7551 iterations.
> 
> There can be several reasons for this happening. Does this happen in
> the same time step every time you run it with the same number of
> cores? Can you increase the number of graphical outputs and look at a
> solution directly before this crash?
> 
>> I've attached the run output file (where the error is), the parameter file,
>> and the c and h files for the temperature boundary conditions for the box convection
>> (same as earlier this summer).
> 
> I tried it here and it runs fine for me with 4 processors. Do you have
> any other changes by any chance? The code finished in 2 time steps for
> me:
> *** Timestep 1:  t=3.35488e+09 years
> *** Timestep 2:  t=1.1674e+10 years
> 
> I increased the final time and it worked for a couple hundred time steps.
> 
> Best,
> Timo
> 
> -- 
> Timo Heister
> http://www.math.tamu.edu/~heister/
> _______________________________________________
> Aspect-devel mailing list
> Aspect-devel at geodynamics.org
> http://geodynamics.org/cgi-bin/mailman/listinfo/aspect-devel

--------------------------------------------------
Associate Professor & Chancellor Fellow, U.C. Davis 
Chair, Geology Graduate Program
Department of Geology & KeckCAVES
2129 Earth & Physical Sciences Bldg
Davis, CA 95616

E-mail: mibillen at ucdavis.edu
Phone: no extension, please e-mail
--------------------------------------------------







-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://geodynamics.org/pipermail/aspect-devel/attachments/20120830/d6b86b66/attachment.htm