amrita-ebook Google Group

Re: calculation hangs without error message after few time steps

j...@galcit.caltech.edu (James Quirk) — Mon, 29 Sep 2008 13:24:15 UT

Matei,
Yes. Any region of strongly expanding flow is a candidate.
It's not as simple as switching to a more robust scheme, for you'll
likely find its resolution is poor. A better way forward is to
employ two schemes and switch between them dynamically. But BCG
is not set up to deliver such a scheme.

Re: calculation hangs without error message after few time steps

ma...@uottawa.ca (Matei Radulescu) — Mon, 29 Sep 2008 12:59:43 UT

James,
Thanks, I'm glad the problem is resolved.
Regarding the possibility of negative pressures, that was originally
my concern. However, looking at the profiles, I did not encounter
this problem, yet. What part of the flow is likely to suffer from
this type of problem, the steady expansion originating from the

Re: calculation hangs without error message after few time steps

j...@galcit.caltech.edu (James Quirk) — Sun, 28 Sep 2008 15:42:26 UT

Matei,
Using AMRITAv3.03, I've been able to run your script out to 1000
iterations, see the attached tmp.pdf . And so it looks like the upgrade I
mentioned has done the trick. Of course, I would not be surprised that if
you ran a full-blow parameter study for a detonation diffracting around a
cyclinder, you would sooner or later run into problems, for the

Re: calculation hangs without error message after few time steps

j...@galcit.caltech.edu (James Quirk) — Fri, 26 Sep 2008 14:21:52 UT

Yes, thermal runaway processes could present a problem.
No. But there is nothing to prevent the savvy user from building
one into his or her patch-integrator.
I'm currently running your case in serial mode and have just got
to phase 32. Therefore I should know later today whether or not
v3.03 fixes the problem.

Re: calculation hangs without error message after few time steps

ma...@uottawa.ca (Matei Radulescu) — Fri, 26 Sep 2008 13:42:16 UT

Does that mean that I could run into the possibility that the CFL
condition may not be verified during the time which dt is fixed?
Say all of a sudden, something ignites & gives large characteristic
velocities?
Do you have a flag somewhere to see if indeed the CFL condition is not
broken during the fixed time steps on the finest grids necessary to

Re: calculation hangs without error message after few time steps

j...@galcit.caltech.edu (James Quirk) — Thu, 25 Sep 2008 15:34:29 UT

Sorry for the brain fade, but I'm suffering with a bad head cold
at the moment. If you take a look in the file:
$AMRITA/include/f77/AMR_SOL/AM RITA
you will see a variable MASK2 which is three nibbles wide i.e. 0xFFF.
The time step is decided by looking at the most restrictive case
from all the grid levels. But because of the way the temporal refinement

Re: calculation hangs without error message after few time steps

ma...@uottawa.ca (Matei Radulescu) — Thu, 25 Sep 2008 15:01:44 UT

The culprit for the lack of error message reported in my original
post, upon crashing, was due to my missing bracket indicated by James,
perhaps causing the error not be captured appropriately. Adding the
"}" at the critical location makes the error capturing work, and i
get::
Error at line 216 of file run_circle_detonation_simple:

Re: calculation hangs without error message after few time steps

ma...@uottawa.ca (Matei Radulescu) — Thu, 25 Sep 2008 14:57:22 UT

you mean 2^12=4096?

The reason here was to be able to do an appropriate resolution study
and not have to compare apples with oranges.

Isn't the time step always adjusted based on your most resolved Dx?
In that sense, should it not matter how many levels "above" that there
are?

If indeed the max no. of patches is 2^12=4096, than indeed, at a

Re: calculation hangs without error message after few time steps

j...@galcit.caltech.edu (James Quirk) — Thu, 25 Sep 2008 01:00:56 UT

Matei,
The discrepancy arose because I misread your original e-mail. You said
that the problem occured after the 39th output, but I took that to mean
the 39 iteration. Hence when I ran the script I reduced the march down
from 20 steps to 5 so that I would have enough checkpointed solutions to
look at what was going on.

Re: calculation hangs without error message after few time steps

ma...@uottawa.ca (Matei Radulescu) — Thu, 25 Sep 2008 00:00:00 UT

James,
Just to set the clock straight, after marching 10 steps with my script
with cfl=0.5
i get to the time 0.4303065231255236
I have uploaded the corresponding mailit entitled
run_circle_detonation_test.mai lit
which contains the ps files for the schlieren, etc...

matei

Re: calculation hangs without error message after few time steps

ma...@uottawa.ca (Matei Radulescu) — Wed, 24 Sep 2008 23:02:20 UT

HMMM. Something is very odd about the time steps the solver takes.
The time step at which my run fails is after the detonation diffracts
around the cylinder and a new detonative mach stem is forming on the
bottom horizontal boundary. It seems that your counter runs faster
than mine !?
In any case, can you run the script and see if the detonation makes it

Re: calculation hangs without error message after few time steps

j...@galcit.caltech.edu (James Quirk) — Wed, 24 Sep 2008 22:42:10 UT

Matei,

Sorry. See the attachment here.

It could be something in your .bashrc/.cshrc file, or it could
be a bug in your version of gcc, which has happened before.

I am gearing up to release AMRITAv3.03 and so I'm unlikely
to have the time to track the problem down. What I suggest is that

Re: calculation hangs without error message after few time steps

ma...@uottawa.ca (Matei Radulescu) — Wed, 24 Sep 2008 22:03:23 UT

James, Gary,
I'll try your recommendations and will get back to you regarding my
success, shortly, hopefully.
Thanks for your help.
matei
(You did not attach the file). However, I do not understand why it
bugs on my system and not on yours.
I can give you access to my machine if you feel that checking the

Re: calculation hangs without error message after few time steps

j...@galcit.caltech.edu (James Quirk) — Wed, 24 Sep 2008 11:31:00 UT

Matei,
As described in the VKI notes, AMRITA has a front-end and a back-end.
The Perl job is the AMRITA interpreter that parses your script and the
back-end is the mesh refinement engine that carries out the
actual simulation. The two ends runs as separate UNIX processes
that communicate via named pipes. Change directory to

Re: calculation hangs without error message after few time steps

men...@leeds.ac.uk (Gary Sharpe) — Wed, 24 Sep 2008 09:49:02 UT

Matei,
have you tried restarting the calculation from the 39th output and
seeing
if it gets any further. I sometimes (but not reproducibly) encounter
this problem
running in parallel on a cluster. However, I can always restart from a
saved model
and it will keep going beyond where it stopped.