A Summary Guide to the Latin Hypercube Sampling (LHS) Utility

Latin Hypercube Sampling (LHS) is one method of Monte Carlo-type sampling, which is useful for limiting sample size yet maximizing the range of sampling of the underlying distributions. The LHS utility, for which this document describes the usage, also allows for user-specified correlations between two or more of the sampled distributions. The LHS utility described herein is a full re-coding using C/C++ of the original LHS utility—developed at Sandia National Labs (Swiler and Wyss (2004)), written in FORTRAN and freely available. The re-coding hones close to the original FORTRAN code, but allows for significantly more flexibility. For example, dynamic memory allocation is used for all internal variables and hence there are no pre-determined dimensions. The new utility has additional features compared to the original FORTRAN code: (1) it includes 10 new statistical distributions; (2) it has four additional output formats; and (3) it has an alternative random number generator. This guide provides a summary of the full features of the LHS utility. For a complete reference, with the exception of the new features, as well as a description of the intuition behind the LHS algorithm users are referred to Swiler and Wyss (2004).