ETH Polymer Physics seminar


2016-11-07
10:15 at HCP F 43.4

Molecular dice

Santosh Ansumali

Jawaharlal Nehru Centre for Advanced Scientific Research (JNCASR), Bangalore, India

Good quality pseudo-random numbers are at the heart of many computer simulation methodologies in science and engineering. Starting from pseudo-random number generators, one often defines appropriate inversion schemes to generate desired distributions. As Gaussian statistics appears very often in physical processes, the inversion scheme for generating Gaussian statistics, Box-Muller method, plays an important role in scientific simulations. However, this method requires 100-1000 floating point operations to generate a single Gaussian random number. Our ability to do large scale simulations can be greatly enhanced if an analog of pseudo-random number generators can be found for direct generation of Gaussian random sequence. We propose a new algorithm ``Molecular dice'' to directly generate Gaussian random number. This new algorithm is able to generate Gaussian sequence as fast as uniform number sequence itself. We explore possible O(N) extension of this algorithm for multivariate normal distribution. Finally, we comment on possibility of generating other statistics such as Poisson distribution using molecular dice idea.


© Mar 2024 mk     719 entries