PPM hot spots hint

[Nathan Goldbaum]

Hi all,

Several months back there was a lot of discussion on this list about seeing hot spots in PPM hydrodynamics calculations.

The problem usually manifests itself as a single zone near a refinement boundary exhibiting a temperature going to infinity and density going to zero over the course of many timesteps.  Since the temperature increases in the bad zone without bound, this causes the simulation progress to grind to a halt. I don't we ever landed on a satisfactory explanation for exactly why this happens.

Recently, I've been doing some experimenting and have found one possible way to avoid the crashes.  The solution seems to be running with first order interpolation (i.e. InterpolationMethod = 4).  With this parameter choice, the host spot crashes never occur for my production isolated galaxy simulations.

Hopefully this will be useful for others doing production calculations using the PPM solver.

Best,

Nathan Goldbaum

[David Collins]

That's interesting, thanks.  I wonder if it's due to the fact that we have a first order in time interpolation for the boundary, but might be using higher order in space in the boundary?  I'd be curious to know if only reduction in the boundary interpolation does it...

d.

--
You received this message because you are subscribed to the Google Groups "enzo-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email to enzo-dev+u...@googlegroups.com.
To post to this group, send email to enzo...@googlegroups.com.
Visit this group at http://groups.google.com/group/enzo-dev.
For more options, visit https://groups.google.com/d/optout.

--

-- Sent from a computer.

[Britton Smith]

Hi everyone,

I have a small followup on this, too.  I did a test using the AMRCosmology_Grackle sample run in enzo-dev using InterpolationMethods 1 and 4.  This is a very low resolution run (32 Mpc/h, 32^3 root grid + 5 AMR), so maybe this makes the differences more pronounced.

At z = 0, projections of density and temperature look very similar, as do density/temperature phase plots.  One bigger difference I saw was the in the final stellar mass.

InterpolationMethod = 1

3093 star particles

2.039948e+11 Msun total stellar mass

InterpolationMethod = 4

311 star particles

3.687624e+10 Msun total stellar mass

Perhaps differences will be less pronounced for a higher resolution run.  I have another set of tests running now with the same setup, but with a 64^3 root grid.  I will update when that is finished.

Britton

[Britton Smith]

Hi again,

Here are the results for simulations identical to the ones in the last email, except with a 64^3 root grid.

InterpolationMethod = 1

Total stars: 10632305.

Total stellar mass: 2.90819866913e+13 Msun.

InterpolationMethod = 4

Total stars: 13760987.

Total stellar mass: 3.93434962933e+13 Msun.

In the case of InterpolationMethod = 1, I had to also set

FluxCorrection = 0

ConservativeInterpolation = 0

in order to avoid the PPM hotspots.  Note, that this simulation has AMR but not static grids, so PPM hotspots are not exclusive to the nested static grid setup.

The only other thing of some interest is the slight, but noticeable different in the phase plots:

InterpolationMethod = 1

http://i.imgur.com/vqdq61O.png

InterpolationMethod = 4

http://i.imgur.com/RSnJ8aw.png

I'm not sure what conclusions can be drawn from this, but feel free to speculate wildly.

Britton

[Nathan Goldbaum]

Hi Britton,

For what its worth, I've done similar comparisons (unfortunately I don't have the data available since I did this several months ago) and see no significant differences in the star formation rate for my isolated galaxy simulations.

As always, you may probably want to test your parameter choices and your mileage may vary...

-Nathan