PVM-GRACE:PVM based automatic Feynman diagram calculation system

Paper: 226
Session: A (poster)
Presenter: Yuasa, Fukuko, KEK, Tsukuba
Keywords: networking, parallelization, simulation, symbolic computing, massive parallel systems

PVM-GRACE:PVM based automatic Feynman diagram calculation system

Fukuko YUASA, Setsuya Kawabata, Tadashi Ishikawa

KEK, National Laboratory for High Energy Physics
1-1 OHO, Tsukuba City, Japan 305

Denis Perret-Gallix
LAPP, Laboratoire d'Annecy-le-Vieux de Physique des Particules
74941 Annecy-le-Vieux CEDEX, FRANCE


We will present a PVM[1] based automatic Feynman diagrams calculation system,
PVM-GRACE. GRACE[2] is a powerful software package for the calculation of High
energy physics process cross-section and for the event generation.

Using the Feynman diagrams technique, all observables can be computed for
any process at any order of the perturbative theory. However, the calculation
becomes rapidly tedious and involved. Automatic computation systems have been
developed to cover the increasing need for such calculations.

The MINAMI-TATEYA group has developed the GRACE program package. When the
initial and final states of the physics process, the physics model and the
kinematics are set by input data-cards, GRACE generates automatically the
FORTRAN code corresponding to the matrix element of the physics process.
Integration of the matrix element over the multi-dimensional phase space leads
to the cross-section and the event generation. For instance, grc4f[3] is one of
the event generators for LEP-II experiments generated by GRACE.

For the next linear collider experiments at the higher energy region,
GRACE will play an important role in calculating the cross-section and
generating physics events. However, the larger the centers of mass energy and
the number of final state particles, the larger the CPU time involved.
For example, there are over 200 Feynman diagrams to be calculated for the
final 6-body physics process. At centers of mass energy 500 GeV, it takes
over 28 hours on the HITACHI 3500/850PS (269 Spec float 92) to get the good
accuracy for the numerical integration.

We have implemented PVM in the GRACE system (PVM-GRACE) to reduce both the
actual execution time and the required memory space per processor. The basic
idea behind the PVM-GRACE project is to distribute amplitude calculations
over several PVM slaves which are spawned by the PVM master. This simultaneous
calculation of amplitudes reduces the total execution time of GRACE drastically
as well as the required memory allocation on each processor.

The performance of PVM-GRACE for several physics process have beed measured
on both the workstation cluster and the massive parallel computer. It is
clearly shown that PVM-GRACE gives excellent performance on both type of
processors in term of reduction of actual execution time when the network
overhead is kept small as compared to the processing time. It is also shown
when the physics process becomes more complex and its processing time longer,
the performance of PVM-GRACE becomes more efficient. PVM-GRACE will become more
important for the future experiments like JLC in addition to LEP-II.

[1]PVM: Al Geist et al., PVM3 user's guide and reference manual.

[2]GRACE: MINAMI-TATEYA group, KEK Report 92-19.
A Feynman-graph generator for any order of coupling constants.
Computer Physics Communication 92 (1995) 127-152.

[3]grc4f: J.Fujimoto et al.,
to be appeared in Computer Physics Communication.