Applying enhanced fault localization technology to Monte Carlo simulations

This paper describes and explores applications of several new methods for explaining unexpected behavior in Monte Carlo simulations: (1) the use of fuzzy logic to represent the extent to which a program behaves as expected, (2) the analysis of variable value density distributions, and (3) the geometric treatment of predicate lists as vectors when comparing simulation runs with normal and unexpected outputs. These methods build on previous attempts to localize faults in computer programs. They address weaknesses of existing techniques in cases where programs contain real-valued random variables. The new methods were able to locate a source of error in a Monte Carlo simulation and find faults in benchmarks used by the fault localization community.