Open AccessInvestigation of pattern recognition techniques for the indentification of splitting surfaces in Monte Carlo particle transport calculations
Author(s) -
J.L. Macdonald
Publication year - 1975
Language(s) - English
Resource type - Reports
DOI - 10.2172/4157716
Subject(s) - monte carlo method , monte carlo molecular modeling , monte carlo method in statistical physics , statistical physics , hybrid monte carlo , quasi monte carlo method , dynamic monte carlo method , monte carlo integration , particle filter , computer science , variance reduction , algorithm , mathematics , markov chain monte carlo , artificial intelligence , statistics , physics , kalman filter
Statistical and deterministic pattern recognition systems are designed to classify the state space of a Monte Carlo transport problem into importance regions. The surfaces separating the regions can be used for particle splitting and Russian roulette in state space in order to reduce the variance of the Monte Carlo tally. Computer experiments are performed to evaluate the performance of the technique using one and two dimensional Monte Carlo problems. Additional experiments are performed to determine the sensitivity of the technique to various pattern recognition and Monte Carlo problem dependent parameters. A system for applying the technique to a general purpose Monte Carlo code is described. An estimate of the computer time required by the technique is made in order to determine its effectiveness as a variance reduction device. It is recommended that the technique be further investigated in a general purpose Monte Carlo code. (auth