[Yt-dev] RMS mass overdensity method

[Matthew Turk]

> to follow up, I realized that I can define a pf.h.sphere outside of the box, so I am finding particles using .sphere, but I am checking the spheres to see if they overlap with a box side. If they do, I reflect around the domain and make a sphere with a center outside the box, and get the particles that way.

Interesting!  I think this is a good method of implementing a parallel
sphere object!  We can wrap what you've done into the cutting methods.

> I've run this in serial and in parallel (2 procs) on a small dataset, and with 10,000 samples the two give answers different by six parts in one thousand. I think they would be identical if I pre-computed the random centers. Do we think I should do that?

I think if you utilize a single random state between teh two, like
what Britton does in the light cone, that would work great.  :)

> Comments welcome! I'd like to eventually commit this, so I'd like to know what I should fix before I do that, if you can catch any bugs. Also, I haven't decided on a good name for this method. What I have is not very good.

This would be a great addition.  I think this would go well in
extensions or in Derived Quantities (but not as a DQ, because I am not
sure we can implement the DQ protocol unless we use the second method
I mentioned above), but let's see if we can come up with some
streamlining first.  I am glad that the MPI calls are kept to a
minimum here, but do you mind if I make some suggestions before it
gets added?

-Matt