Why The Results of Parallel and Serial Monte Carlo Simulations May Differ

Parallel Monte Carlo simulations often expose faults in random number generators

💡 Research Summary

The paper investigates why parallel Monte Carlo simulations often yield results that differ from their serial counterparts, even though both are intended to sample the same underlying stochastic process. The authors begin by describing the fundamental distinction: a parallel simulation distributes the sampling work across multiple processors that operate concurrently, whereas a serial simulation performs the same sampling sequentially on a single processor. In theory, a formal IF‑THEN theorem guarantees that, provided the source stochastic process (the random number generator, RNG) satisfies certain properties—most notably that it produces independent, uniformly distributed numbers on the interval (0, 1)—the two resulting processes should be statistically identical.

Empirical evidence, however, repeatedly shows deviations. The paper cites two representative studies: Mahobar and Rollett (2003)