Very large numbers of particles can be employed at little cost since most cpu time is taken up by the determination of the potential on the grid. [See Sellwood (1997, in "Computational Astrophysics" ed Clarke & West, ASP Conf series v123, p215) for a performance comparison with other codes.]
The version of the algorithm used here requires the individual grid cells to be cubic, but the overall grid need not be cubic. The FFTs supplied require the number of mesh spaces in each direction to be (2**n + 1), where the exponent n may be chosen independently for each coordinate direction. As large grids require a great deal of memory (c400 MB for a 257**3 grid), it is recommended that the parameters set in the include file ’rjinclude.h’ be no larger than necessary, although the code will function correctly as long as the actual dimensions used do not exceed those set by the parameter statement at compile time.
Time integration follows the standard 2nd order time-centered leap-frog, with the velocities one half a time step out of synchrony with positions. This difference is maintained in the internally stored coordinates and is created and can be undone, by a call to subroutine TMCENT. For output of the particle coordinates at a particular instant, the velocities need to be the average of those before and after the time for the positions.
Results, in this public version, are simply the phase space coordinates of all the particles as often as requested, which can create a very large file. The authors therefore do not employ this scheme themselves, preferring instead to measure and save properties of the model as the simulation evolves. An example of this "on the fly" analysis is provided in the routines ICHECK and MEASURE to determine the global integrals (energy, momentum, etc)
The grid is set up in subroutine GRDSET using parameters read in from a short ASCII input file (galaxy.dat). The positions and velocities of the particles are read in subroutine LOADUP, from a local file galaxy.ini. The gravitational field is determined by a call to FINDF. The model is integrated forward by a call to STEP. After the desired evolution is completed, the positions and velocities of the particles are saved by a call to UNLOAD. The new 2014 public release of galaxy contains an updated code, including versions of SCF (see scfm(1NEMO) ) and BHTREE (see hackcode1(1NEMO) ).
$NEMO/src/nbody/evolve/sellwood source code tree (w/ res2snap & snap2ini) $NEMO/usr/sellwood/GALAXY15 source code tree for GALAXY15The main files associated with the run are (notice the basename ’galaxy’ is fixed by the code):
galaxy.dat ASCII input: grid parameters, length and time scales galaxy.ini ASCII input: initial coordinates of all the particles galaxy.lis ASCII output: a brief summary of progress (appended) galaxy.fin ASCII output: final coordinates of all the particles galaxy.res binary output: coordinates and potentials at intervals (appended) galaxy.tmp short ASCII file (deleted when closed) galaxy.aux large binary file (deleted when closed)
Sample galaxy.dat initialization file:
33 33 33 # number of grid cells in (x,y,z) 15.0 # number of grid cells per length unit 0.05 # time step length 0.5 # time between particle outputs 0.1 # time between integral checks 1.00 # end time
The format of the ASCII galaxy.ini (and also galaxy.fin) files is:
Time Mass Nbody
X_1 Y_1 Z_1 VX_1 VY_1 VZ_1
....
X_n Y_n Z_n VX_n VY_n VZ_n
The galaxy.fin can also be used for a restart. Note that the timestep is
one more then the last requested time, to prevent that the galaxy.res file
will contain a datadump on the restart timestep twice.
The new release V15 does not depend on commercial (e.g. NAG) software, and can be compiled using open source tools and libraries. PGPLOT must be installed independantly.
#! /bin/csh -f
if ($#argv != 2) then
echo Usage: res2snap FILE N
echo Converts the N-th snapshot from FILE to NEMOs snapshot format
exit 0
endif
# set command line parameters
set file=$1
set n=$2
# get header info
set tsnap=‘unfio $file "$n*2-1" float | awk ’{if (NR==1) print $1}’‘
set nbody=‘unfio $file "$n*2-1" int | awk ’{if (NR==2) print $1}’‘
# dump data and convert to snapshot
unfio $file "$n*2" float maxbuf=100000 |\
tabtos - ${file:r}.$n.snap "" pos,vel,phi options=wrap times=$tsnap nbody=$nbody
2021 update: 200 steps on an i5-1135G7 took 3.7 13 speedup from the 2004 results. However, compiling with flang (FFLAGS=-O3),
200 steps took 0.22
James & Sellwood (1978, MNRAS v182, p331) James (1977, J Comp Phys v25, p71). Sellwood (1997, in "Computational Astrophysics" ed Clarke & West, ASP Conf series v123, p215) http://www.physics.rutgers.edu/galaxy (full 2014 version; V1.5 and later)
9-jun-97 V1.0 Sellwood public release/adopted for NEMO JS/PJT 24-jun-97 V1.1 added ICHECK/MEASURE; dtlog to galaxy.dat JS 18-mar-04 fixed bad usage line; refer to rungalaxy now PJT 8-mar-06 V1.3 now installs by default into NEMOBIN PJT 26-jun-2014 notes on the official full public release PJT 10-mar-2017 notes on the new V15 release PJT 6-feb-2018 notes on the new V1.5.2 release PJT