Concurrent processing applied to the Exodus Monte Carlo technique

✍ Scribed by V. F. Fusco

• Publisher: John Wiley and Sons
• Year: 1991
• Tongue: English
• Weight: 579 KB
• Volume: 4
• Category: Article
• ISSN: 0894-3370
• DOI: 10.1002/jnm.1660040205

✦ Synopsis

This paper illustrates how problems based on the Probabilistic Potential Method can be solved using a parallelized form of the Exodus Monte Carlo technique. The Exodus technique which does not use random number generation is shown to be more efficient for moderately large problems than the conventional fixed random walk method when programmed in both sequential and concurrent forms.

The technique is illustrated for an electrostatic field problem by using as an example internal potential evaluation for a rectangular coaxial line. The problem geometry is torn into subsections which are then analysed independently using concurrent programming methods. The resulting algorithm is compared with the same problem executed using a conventional sequential algorithm. Computation is achieved by use of an array of transputers. The design of the concurrent algorithm is described in detail and typical results are compared with the fixed random walk Monte Carlo method and also with the finite difference relaxation method. Parallel Fortran is used throughout as the programming language.