Approximation methods in probability theory

Research exposition (monographs, survey articles) pertaining to probability theory (60-02) Probability distributions: general theory (60E05) Infinitely divisible distributions; stable distributions (60E07) Characteristic functions; other transforms (60E10) Central limit and other weak theorems (60F05) Sums of independent random variables; random walks (60G50) Approximations to statistical distributions (nonasymptotic) (62E17) Asymptotic distribution theory in statistics (62E20)

Review text

This book gives an introduction to a wide range of methods for assessing the quality of approximations in probability theory, that is, the distance between given distributions. Each technique is presented with all of the details necessary to apply it, including proofs of the majority of the results used and useful illustrations and examples throughout. Clear bibliographical notes give a starting point for further reading, more recent developments and any proofs that have been omitted. The text and exposition is clear and well laid-out. In particular, each of the methods discussed is accompanied by a summary which highlights each step in a typical application of that method, alongside some of its advantages and drawbacks. These summaries are of great benefit to the reader wishing to compare the techniques under discussion or select a technique for a particular application. Chapter 1 collects many standard definitions, results and bounds used in the remainder of the book, and ensures that the prerequisites for reading the text are as minimal as possible for a book at this level. The remaining chapters then each focus on a particular technique or setting, giving bounds in terms of the total variation, uniform, local and Wasserstein metrics. Chapter 2 discusses the method of convolutions; taking advantage of convolution structure and properties of norms in order to bound distances between compound random variables. As with most of the book, the focus is on proving upper bounds. Chapters 3--6 consider a variety of techniques applicable to lattice random variables, including bounds taking advantage of non-negativity or symmetry of the random variable in question. Each of these chapters details bounds in a different metric: Chapter 3 gives bounds in the local metric, Chapter 4 uses the uniform metric, Chapter 5 gives results in both the total variation and Wasserstein metrics, and Chapter 6 gives non-uniform estimates. Chapter 7 is a brief discussion of Arak's lemma for non-lattice discrete distributions. Absolutely continuous approximations are considered in Chapter 8, which includes estimates in several of the metrics already discussed, non-uniform estimates, and results under assumptions of bounded density. Esseen-type results employing characteristic functions are detailed in Chapter 9. This includes a proof of the famous Berry-Esseen theorem in the case of finite third moments, and also results under significantly weaker moment assumptions, including bounds for distributions with no finite moments. Chapter 10 gives lower bounds in various metrics using characteristic function techniques. Some further lower bounds are also given as part of the following chapter. Stein's method, based on a characterisation of the target distribution and a suitable differential/difference equation, is the subject of Chapter 11. The treatment here is deliberately not exhaustive, but is an excellent starting point for learning the technique and to be introduced to the relevant literature. The material presented here is interesting, and complementary to that in more comprehensive book-length treatments of Stein's method such as [\textit{A. D. Barbour} et al., Poisson approximation. Oxford: Clarendon Press (1992; Zbl 0746.60002)] and [\textit{L. H. Y. Chen} et al., Normal approximation by Stein's method. Berlin: Springer (2011; Zbl 1213.62027)]. Chapter 12 demonstrates the triangle function method, involving a certain decomposition of the measure of interest, and then controlling the behaviour of indicator functions by suitably bounding closely related triangular functions. Heinrich's method, based on an extension of characteristic function techniques to the dependent setting, is employed in Chapter 13 to treat sums of \(m\)-dependent random variables, including an application to a runs problem. The final chapter gives shorter introductions to six further approximation techniques, along with a discussion of asymptotically sharp constants in compound Poisson and compound binomial approximations. Techniques discussed here are {\parindent=0.7cm\begin{itemize}\item[--] the method of compositions, which takes advantage of smoothing properties of the normal distribution, \item[--] a very brief introduction to coupling techniques, \item[--] Bentkus' approach of constructing a smooth parametric curve between the random variables whose distance is to be estimated, \item[--] a discrete version of Lindeberg's method, \item[--] Tikhomirov's method, which has some commonality with Stein's method, but uses characteristic functions, \item[--] bounds employing integrals of a concentration function. \end{itemize}} Each chapter concludes with a selection of exercises, for the majority of which solutions are given at the end of the book. In conclusion, this book is an excellent starting point for those new to the area, and a useful reference for more experienced researchers. It brings together a wide range of techniques for probabilistic approximations in a way which is accessible to a reader with a modest background in probability. The presentation is user-friendly, giving the reader all that is needed to select and apply appropriate methods for a range of approximation problems.

Recommendations

Cited in

(31)

This page was built for publication: Approximation methods in probability theory

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q289667)