Pseudorandom Number Generators for Scientific Computing in the 21st century
University of Canterbury, New Zealand
Time & Place
Wed, 18 Apr 2018 11:00:00 NZST in Erskine 340
All are welcome
Probabilistic paradigm of scientific investigations finds more and more applications as science becomes more and more computational. Scientific investigations of complex stochastic processes and phenomena conducted by means of computers can require an access to very long sequences of random numbers. Extremely long sequences of random numbers are needed for example during simulation studies of long-range dependent processes occurring in the Internet, or studies of rare events occurring in telecommunication networks or on financial markets.
We will focus on algorithmic generators of independent and uniformly distributed numbers, that are used as primary sources of randomness for computer-based scientific investigations. The current state of art, and the requirements that such generators have to satisfy, will be considered. These generators produce numbers in cycles. Thus, as the computing technology accelerates, the cycles need to be sufficient long, for avoiding deterministic repetitions of randomness in longer scientific experiments, as this could introduce unexpected correlations.
About 20 years ago, using algorithmic generators of (pseudo-)random numbers with cycles of the order between 231 and 248 was generally acceptable. However, these cycles became insufficient for modern, fast multiprocessor computers of the 21st century. Pseudo-random number generators (PRNGs) with cycles longer than of 2600,000 have been proposed. However, such huge sequences of numbers cannot be exhaustively tested against possible undesirable properties. How long cycle should a PRNG have if one takes into account the fast advances of electronic and computer technologies in next 20-30 years? Having answered this question, we will propose a new class of pseudorandom numbers generators, that pass the most rigorous statistical tests and are able to generate numbers very fast, but in not excessively long cycles.
Krys Pawlikowski is a Professor Emeritus of University of Canterbury, where he lectured in CSSE department for 30 years and supervised/co-supervised 16 PhD students. He has been the leader of Akaroa project that resulted in the first automated full-scale grid processing system Akaroa2 for conducting quantitative distributed discrete-event simulation. Akaroa2 received an international commendation in Science category in the Computerworld Smithsonian Award, USA, in 1993 (documentation of Akaroa is kept in archives of the Smithsonian Museum of American History in Washington D.C.) and has been used widely around the world.
He received a Ph.D. degree in Computer Engineering (with Distinction) from Gdansk University of Technology, Poland, and worked there until 1983. The author of over 200 journal and conference papers and four books. Awarded the Alexander-von-Humboldt Research Fellowship (Germany) in 1983-84 and 1999. Visiting Professor at University of Vienna (Austria); NSW University (Australia); University of Genova and Technical University of Torino (Italy); Aachen Academy of Technology-RWTH, University of Wuerzburg and Technical University of Berlin (Germany); Blekinge Institute of Technology-BTH (Sweden) and Clemson University (USA). Gave over 170 invited research seminars at over 100 universities and research institutes in Asia, Australia, Europe and North America.
His current research interests include stochastic discrete-event computer simulation and its applications in performance evaluation of dynamic systems, simulation of rare events, as well as statistical analysis and modelling of data traffic.