Force accuracy at high starting redshift

From: Manodeep Sinha <manodeep.sinha_at_Vanderbilt.Edu>
Date: Fri, 17 Jun 2011 12:18:56 -0500

Hi everyone,

I would like to start a cosmological dark matter-only simulation at an
initial redshift of ~800. From what I understand, round-off errors can
give erroneous forces/positions at such high redshift, particularly from
the tree part. Before I delve into running a series of smaller
simulations to test for convergence/optimal integration parameters, does
anyone here already have experience with such high starting z?

Ultimately I want to avoid these two situations:

i) net force dominated by numerical errors from the short-range part
ii) x_old + v * dt = x_old

The simulation boxsize is 130 Mpc/h with 12963 particles. I plan to use
PMGRID=1296 option to compute the long-range forces, the relative
opening criteria for the tree and a double-precision Gadget executable.

Is this even feasible with the public Gadget-2 code? And if so, does
anyone have recommendations for the right combination of parameters --
the ones I am looking at are: ErrTolIntAccuracy, ErrTolForceAcc, RCUT,
ASMTH and MinSizeTimestep (hopefully, I am not missing any other
important one).

Thanks in advance,
Manodeep Sinha,

-----------------------------
6902 Stevenson Center,
Dept. of Physics& Astronomy,
Vanderbilt University,
Nashville, TN - 37221, USA.
-----------------------------
Received on 2011-06-17 19:19:00