SC 740 Presentation Review

By Deborah Dent

"A Scaleable Library for Pseudorandom Number Generation Theory and Practice"

Presented by

Dr. Michael Mascagni

Coordinator, Ph.D. Program in Scientific Computing

University of Southern Mississippi

August 27, 1997

Dr. Mascagni's presentation was based on the results of an research effort to develop a new standard library for scaleable pseudorandom number generation. The main objective of this effort is to create an easy to use tool for the user community, which utilizes Monte Carlo applications on scaleable platforms.

Having worked as a software developer for the Department of Army and Department of Defense, I can personally relate to the importance of incorporating the results of this research into a library accessible by the scientific user community. This type of technology transfer is always valuable to engineers, computer scientist, mathematicians and other scientist which are constantly seeking reliable, proven, easy to use, cost saving tools to assist them in performing their tasks.

Some of the research goals discussed by Dr. Mascagni include:

• Making easy accessible and usable pseudorandom generators available on parallel machines.
• Providing parallel pseudorandom generator streams with favorable statistical properties and with the ability to produce statistically significant and new results.
• Providing pseudorandom generator streams that are reproducible regardless to the number of processors present. (Valuable verification goal).
• Providing the user the capability to generate their own pseudorandom generator streams.
• Providing portability between serial and parallel platforms.
• Providing high quality cost effective pseudorandom generators.

The main emphasis of Dr. Mascagni's presentation was centered on three different parameterized mathematical methods that can be uniformly incorporated in the library:

• Default generator: additive lagged-Fibonacci
• Prime Modulus LCG
• Maximal - period shift-register generators

in which he presented detail information and proofs. What impressed me is that these parameterized methods based on seeding are extremely useful in providing the capability of producing unique seeds for every processor in multiprocessor applications.

After presenting information on the usefulness of utilizing pseudorandom generator in the Dept. of Energy Neutron Generator, Dr. Mascagni wrapped up his presentation with a brief discussion on the data structure (binary trees) used in the implementation of the pseudorandom generator code.

The web page referenced in Dr. Mascagni's presentation states that this project is currently ongoing. I am sure the user community will welcome the results.