Elsevier Science Home
Computer Physics Communications Program Library
Full text online from Science Direct
Programs in Physics & Physical Chemistry
CPC Home

[Licence| Download | New Version Template] admc_v1_0.tar.gz(246 Kbytes)
Manuscript Title: Foam: multi-dimensional general purpose Monte Carlo generator with self-adapting simplical grid.
Authors: S. Jadach
Program title: Foam, version 1.01
Catalogue identifier: ADMC_v1_0
Distribution format: tar.gz
Journal reference: Comput. Phys. Commun. 130(2000)244
Programming language: Fortran.
Computer: IBM PowerPC M43P240.
Operating system: UNIX AIX 4.x, HP-UX 10.x, Linux.
RAM: 10M words
Keywords: General purpose, Utility, Phase space, Monte Carlo (MC) simulation and generation, Elementary particle physics.
Classification: 4.14, 11.2.

Nature of problem:
Monte Carlo (MC) simulation or generation of unweighted (weight equal one) events, is a standard exercise in the particle physics, and in many other areas of the research. It is often necessary to generate MC events according to a probability density with strong peaks (singularities) spanned along complicated hyperspaces of not a very well known shape. It is highly desirable to have in the program library a general-purpose numerical tool (program) with a MC generation algorithm featuring built-in capability of adjusting automatically the generation procedure to an arbitrary pattern of singularities in the probability distribution.

Solution method:
In the algorithm a simplical grid of vertices forming a "foam of cells" is built, which adapts automatically to the integrand in such a way that the resulting ratio of average weight to maximum weight, i.e. efficiency, is arbitrarily good. In the subsequent MC generation the foam of cells is used to generate one cell and a point within this cell. The above algorithm is not based on the factorizability assumption of the integrand function like VEGAS of P. Lepage [1]. It can therefore generate efficiently distributions with complicated patterns of the singularities like diagonal ridges, voids, spheres etc.

Restrictions:
The program is memory-hungry and therefore limited to small dimensions n < 10. The actual implementation is limited to n <= 5, but only because of the function calculating the determinant.

Running time:
The CPU time necessary to build up the simplical grid depends strongly on the number of dimensions and the requested size of the grid. On the IBM PowerPC M43P240 installation (266 MHz, 65 CERN units) it takes about 30 secs to build a grid of 5000 vertices, for a simple 2-dimensional distribution.

References:
[1] G.P. Lepage, J. Comput. Phys. 27, 192 (1978).