Star particles outside their own grid for one timestep?

[Nathan Goldbaum]

HI all,

I've run into a possible issue with the way star particle feedback works and I'd like to get some feedback from people who have more experience with the way this works in Enzo 2.X.  The specific use case I'm interested in is an active particle simulation in enzo 3.0 but I think this issue applies equally to star and active particles.

I've discovered that in a simulation with tens of thousands of star particles, every once in a while one of the particles ends up outside its own grid immediately after UpdateParticlePositions gets called in EvolveLevel.  That is, the particle drifted outside of its own grid after its position was updated during the evolution loop.

I guess for star particles this isn't really a huge concern since only one particle out of presumably a large number of particles is affected and if that one particle doesn't do feedback for one timestep it's not the end of the world.

Is that pretty much the state of it or is there some code somewhere I'm not seeing that handles this situation and communication the particle to the new grid it should be associated with?

Cheers,

Nathan

[Greg Bryan]

Hi Nathan -- I think that's about the state of it.  It's definitely one of the big problems with the previous star particle feedback implementation.  We could either think about ways to fix the particle ownership, or I wonder if a general distributed feedback scheme (via ActiveParticles) wouldn't also effectively fix this, since the distributed feedback has to be able to touch other grids anyway.

Greg

--
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?hl=en.
For more options, visit https://groups.google.com/groups/opt_out.
 
 

[Nathan Goldbaum]

Ho Greg,

I think that's right - I've looked and my sink particles will still properly do feedback even if they've been temporarily misassigned to grids.

Of course the feedback algorithm they use:

for each grid

   for each particle on this grid

         construct feedback zone

         do feedback

         distribute feedback zone

isn't terribly efficient since it needs to be done in lockstep on all cores.  For my purposes it's fine but if you had tens of millions of star particles it wouldn't work.  I think it's possible to do it efficiently, so long as it doesn't matter which order the particles do feedback, but that will have to wait on someone else to implement - I need to finish my PhD!

Nathan

[Geoffrey So]

"in a simulation with tens of thousands of star particles, every once in a while one of the particles ends up outside its own grid immediately after UpdateParticlePositions gets called in EvolveLevel."

Just curious is this different than the error of particle outside of their grid we saw last time in Georgia Tech?  I recall our test problem with 3.0 was tiny but I don't recall the number of star particles.

From

G.S.

[Nathan Goldbaum]

Hi Geoffrey,

It's possible that's the case, we'd have to reproduce your tests using the version of the code from back then to be sure.

If you (or anyone else) are interested in testing out active particles, the AMR cosmology test bundled with enzo should work correctly now.  This is thanks to Britton's work that fixed up the Cen & Ostriker star particles.  Since we haven't reenabled the ParticleType arrays yet, you'll still have to clone my private fork of enzo 3.0 to get Britton's changes.  Hopefully these changes will be included in the main enzo-3.0 soon.

We don't get bitwise identical results compared to enzo 2, but we've concluded that should be expected given the fact that the particle masses are not being summed in the same order when they are deposited on the baryon mass field.  Even that roundoff error level of inaccuracy is enough to perturb the solution noticeably once a large number of star particles have formed.

-Nathan