Perfect Reversal of Rejection Sampling Methods for First-Passage-Time and Similar Probability Distributions
Abstract
We present a perfectly reversible method for bi-directional generation of samples from computationally complex probability distributions. While the previously best-known procedures consume memory proportional to the length of execution between changes of execution direction, here we present a scheme to completely eliminate the memory overhead. Our solution affords two important features, namely determinism and repeatability, across arbitrarily spaced changes of direction (and arbitrary number of samples) along the sample stream. We illustrate the perfect reversal method with first passage time distributions that appear in physical system models, and present its implementation and verification in FORTRAN.
[Pub 103]
Kalyan Perumalla
As a Federal Program Manager in Advanced Scientific Computing Research at the U.S. Dept. of Energy, Office of Science, Kalyan Perumalla manages a $100-million R&D portfolio covering AI, HPC, Quantum, SciDAC, and Basic Computer Science. In his 25-year R&D leadership experience, he previously led advanced R&D as Distinguished Research Staff Member at the Oak Ridge National Laboratory (ORNL) developing scalable software and applications on the world’s largest supercomputers for 17 years, including as a line manager and a founding group leader. He has held senior faculty and adjunct appointments at UTK, GT, and UNL, and was an IAS Fellow at Durham University.