scientific article; zbMATH DE number 1145157

Authors Benjamin Melamed, Reuven Y. Rubinstein

Publication date 26 April 1998

output analysis perturbation analysis random number generation rare events simulation response surface methodology sensitivity analysis stochastic optimization variance reduction

Mathematics Subject Classification ID

Introductory exposition (textbooks, tutorial papers, etc.) pertaining to probability theory (60-01) Introductory exposition (textbooks, tutorial papers, etc.) pertaining to operations research and mathematical programming (90-01) Introductory exposition (textbooks, tutorial papers, etc.) pertaining to systems and control theory (93-01)

Review text

This book is divided into two parts, I: Conventional simulation and II: Modern simulation. As the title indicates, the first part treats the most basic parts of simulation analysis like random number generation, output analysis with particular emphasis on steady-state simulation (batch means, regenerative simulation etc.) and variance reduction techniques (common random numbers, antithetic sampling, control variates, conditional Monte Carlo and stratification). In addition, the TES (transform-expand-sample) methodology for generating correlated sequences as developed by the second author [e.g. ORSA J. Comput. 3, No. 4, 317-329 (1991; Zbl 0764.65002)] is exposed. The second part deals with more recent developments, in particular sensitivity analysis: how does one quantify by simulation the effects of small perturbations of systems parameters. The exposition concentrates here on the score function method, to which the first author has made numerous contributions; see in particular the earlier book by the first author and \textit{A. Shapiro} [``Discrete events systems. Sensitivity analysis and stochastic optimization by the score function method (1993; Zbl 0805.93002)], henceforth abbreviated as RS. The idea is to express the performance measure in terms of likelihood ratios and differentiate under the integral sign to get the sensitivities, which contrast the sample path differentiation used in infinitesimal perturbation analysis, e.g. \textit{P. Glasserman} [``Gradient estimation via perturbation analysis (1991; Zbl 0746.90024)]. The advantage of the score function method is largely that it applies to a larger class of problems. In the book, static problems (e.g., mean shortest paths in PERT networks) and dynamic problems (e.g., queueing performance and manufacturing systems) are treated separately. A further distinction is made between distributional parameters like exponential rates, normal means and variances etc., and structural parameters like the levels \(s\), \(S\) in an \((s,S)\) inventory model. In the case of structural parameters, the basic absolute continuity requirements for the score function method are often violated, and a method (`push-out') for solving this problem is described. Part II further deals with response surface methodology (how to obtain smooth estimates of performance measures for not one but several sets of parameters), stochastic optimization, and rare events simulation (how does one obtain simulation estimators of small probabilities which have good complexity properties in the sense that the relative error remains almost bounded as the rare events probability goes to zero). The book has considerable overlap with the RS book quoted above, and in fact, large parts are very close. Compared to RS, the present book puts less emphasis on stochastic optimization and is considerably broader in its choice of topics. Because of this broadness, the book will serve as a valuable textbook for students in operations research, industrial engineering and management science.

Recommendations

Cited in

(31)

This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4384475)