Re: Galaxy Interaction Simulations

Chris Mihos (hos@corona.UCSC.EDU)
2 Mar 1994 23:15:05 GMT

In article <2l1v15$>,
Ross Chandler <> wrote:
> (Pat) writes:
>>galaxy-galaxy collision while interesting, requires more computer time
>>then you can get at present. in another 5 years, it'll be cheap
>>but for now, you neeed a less intensive problem..
>You don't need huge numbers of particles to get realistic looking results.
>With 2000 particles and a split hamiltonian to reduce the number of force
>calculations a simulation need only take a hour or two on a 100 mips
>workstation. I attend lectures in between the simulations to fill up the time.
>p.s. If anyone wants to contradict me then please back up what you say with the
>appropriate formula, algorithm or argument so i may take your view on board and
>improve my simulations.

I wont contradict you until we understand each other's terms & accuracy

again, as I said in a post a while back, for pretty pictures a few thousand
particles are fine, provided you dont go straight Nbody. However, to
self-constently model disk galaxies you need > 10^4 particles per
disk galaxy in order to a) make them stable, and b) keep them dynamically
cool (reference: see Sellwood ARAA a few years back, or, more specifically,
Sellwood, J.A. 1989, \mnras, 238, 115).

If you want to keep disk galaxies cool over more than a few Gyr,
**doing self-consistent (ie no rigid halos, three body models, etc) work**
then you need > 5x10^5 particles.

The post which originally started this thread was (I think) by someone
who wanted to do entry-level exploratory calculations, and as we said
previously, a few thousand particles with tricks such as 3 body will
give nice pictures. but dont trust the detailed dynamics in these
types of simulations -- we find discrepancies in merging times for
galaxies modeled with 5e4 and 5e5 particles.

5e5 particles on a treecode takes ~ 20 Cray C90 hours to run a 1 Gyr
encounter (approx 8 halfmass rotation periods).

so to, galaxy interactions are in the realm of
current computing power, but you need fast machines (ie, again,
NOT PCs! IBM RS6000s, Dec Alphas, fast Suns, etc. are what you want)

and to, you're right small nice looking simulations
can be run, but don't trust the detailed dynamics.

> My opinions are wrong? - Well then please send me the correct
> neural net weightings.

"neural net weightings"? i'm just a lowly astronomer, man!

Chris Mihos