Re: Barnes-Hut treecode (was Re: Galaxy Interaction Simulations)

Wayne Hayes (
8 Mar 94 21:41:37 GMT

I'm starting my Computer Science Master's Thesis on astrophysical
N-body simulation for large N. I've started a literature survey (read
only about 25 papers so far), including Greengard, Barnes & Hut,
Sellwood's survey paper, and lots of others. I'm still not exactly
sure what kinds of things I would like to work on for it. A number of
obvious possibilities are parallelization, study of error build-up (eg,
truncation error from from using low-order integration methods, etc),
using high-order integrators (Sellwood quotes another author as saying
"integrators have approached a local optimum", but I don't believe it
:-), among others.

I'd like to solicit ideas from anybody who's actually *using* N-body
simulators for research. What are some interesting things to work on
in N-body simulation? For example, do we understand how to choose
individual time-steps for particles?

One paper I found very interesting was Jernigan & Porter's ``A tree
code with logarithmic reduction of force terms, hierarchical
regularization of all variables, and explicit accuracy controls.'' (ApJ
Sup 71:871, 1989 Dec). It seems to me their method NlogN is not at all
amenable to softening, and thus may be a much better than the Barnes
and Hut method for an ``ideal'' N-body simulation where your N is close
to the actual N of the system. (I think the only systems we're even
close to such a simulation is, just maybe, simulating a single small
globular cluster.) I've tried to contact the authors via e-mail but
haven't got a response. Anybody know where I can get ahold of Jernigan
or Porter?

"The money spent on the Mars Observer was not sent to Mars. A few tons of metal
were.  The engineering lessons and experience from building the spacecraft and
instruments will remain on Earth to benefit future missions if we choose to
undertake them." -- Steve Collins    || Wayne Hayes