Modern Theory of Summation of Random Variables

DOI10.1515/9783110936537MaRDI QIDQ4210937zbMATH OpenOpenAlexFDO

Publication date 9 September 1998

Full work available at URL https://doi.org/10.1515/9783110936537

Research exposition (monographs, survey articles) pertaining to probability theory (60-02) Infinitely divisible distributions; stable distributions (60E07) Limit theorems in probability theory (60Fxx)

Review text

This book is devoted to graduate and postgraduate students. It will be of value and interest to researchers in probability theory, mathematical statistics and applications. This first volume in the new book series ``Modern Probability and Statistics contains a comprehensive description of the general mechanism of forming limit theorems for sums of independent random variables as well as refinements of limit theorems presenting convergence rate estimates. Chapter 1. Introduction to the theory of probability metrics. (Probability metrics; Minimal and protominimal metrics; Special minimal and protominimal metrics; Ideal metrics; Relationships between metrics; Linearly invariant metrics.) The chapter contains exhaustive information how to define metrics in the set of random variables in a reasonable way, what the properties of such metrics are and what the methods of their application to the problems of approximation type are. Having a great variety of metrics and general conception of their classification, structure, properties and interconnections, we rather naturally come to the conclusion that not all metrics are equivalently suitable for applications. Some of them are capable to simplify the solution essentially due to their specific properties. Being formalized the general arguments form the so-called method of probability metrics. The main statement of this method is the account of structural features of a stochastic model, in whose framework the approximation problem is solved, together with the properties of the selected metric. The better the structural singularities are accounted for in the properties of the metric, the easier the problem posed is solved. Since this assumes the choice of the metric corresponding to the problem irrespective of how much this metric is popular or suitable for further numerical computations, it appears the necessity to recalculate the obtained estimates for metrics, whose specification is dictated by extraneous reasons. This stage of solution helps us to overcome a specific range of questions and related facts. Chapter 2. The nature of limit theorems. (The history of limit theorems; A generalization of the Lindeberg-Feller theorem; Natural estimates of convergence rate in the central limit theorem; Limit theorems as stability theorems; One more mechanism forming the central limit theorem.) The chapter contains a brief review of main results of the theory of limit theorems for sums of independent random variables since the book of \textit{J. Bernoulli}, ``Ars conjectandi (Basel, 1713) saw the light up to publication of the fundamental monograph of \textit{B. V. Gnedenko} and \textit{A. N. Kolmogorov}, ``Limit distributions for sums of independent random variables (1954; Zbl 0056.36001). Since the information of historical kind is primarily of methodological importance for the author, giving the results noted as milestones of creation of the modern theory of summation of independent random variables, he affords himself to depart from the traditional formulations, if he finds it more suitable. The author introduces a new notion -- the notion of a natural estimate. To be exact let \(F_n\) and \(\Phi\) stand for the distribution function of a normalized and centered sum of independent random variables and the standard normal distribution function, respectively. An estimate \(W_n\) is called natural if \(C'W_n\leq \sup_n| F_n(x)- \Phi(x)|\leq CW_n\), \(C'\) and \(C\) are absolute constants. The notion of a natural estimate leads the author to a new point of view concerning the nature of limit theorems in the scheme of summation of independent random variables as stability theorems. The formulation of limit theorems as stability relations has a number of attractive features, in particular, it may serve as a measure of the naturalness of the adopted formulation of the problem and the completeness of the result obtained. Chapter 3. The normalization of random sequences. (General normalization problems; The choice of constants under linear normalization.) Chapter 4. Centers and scatters of random variables. (The definiton and properties of index, centers and scatters; An analog of Chebyshev inequality; Centers and scatters in the criteria of \(L\)-compactness; Centers and scatters in the criteria of \(L\)-convergence of random sequences.) These two chapters are closely connected to each other. Considering the problem of normalization of random variables, we use such numerical characteristics as medians, truncated mathematical expectations, truncated variances which are capable to take into account necessary information about the properties of sums and summands constructing these sums. Expectations and variances are suitable if they exist. In general, it is naturally to pose the question whether or not there exist characteristic of random variables which possess the main properties of expectations and variances and at the same time are free of their deficiencies. It turns out that such characteristics do exist. The author calls them centers and scatters of random variables. Chapter 5. Classical theory of limit theorems for sums of independent random variables. (Reduced model; Infinitely divisible laws; Limit theorems of the classical theory; The estimates of accuracy of approximations in limit theorems of the classical theory; On completeness of the classical theory.) It is quite evident that the classical theory of limit theorems for sums of independent random variables is characterized for the most part by the condition: the summands must be uniformly asymptotically negligible. The chapter contains a number of well-known and important limit theorems, but its exposition differs from those which have been treated in a traditional way. Here one can find a systematical use of centers and scatters as well as the method of probability metrics. Chapter 6. Generalizations of the classical theory of summation of independent random variables. (The history of the problem; Stability of compositions of distributions; Limit theorems in a non-classical setting; Limit theorems in the generalized scheme; The method of metric distances; Limit theorems of non-independent random variables.) The chapter contains a number of important theorems for sums of independent random variables without the assumption that the summands are uniformly asymptotically negligible. It should be noted that the demonstration of the main results on the general theory of limit theorems without any conditions on the summands does not give any answer to the question how universal the regularities observed in the scheme of summation of random variables are. The answer to this question is that the constructed theory of limit theorems owes its success to the specific properties of the operation of summation of random variables, and therefore the passage to other scheme is possibly only in the case where the used operation is related to the operation of summation with respect to its properties. The author demonstrates this on the so-called max-scheme -- the scheme of summation of independent random variables where the basic operation \(x+y\) is replaced by the operation \(\max(x,y)\). The notion of independence is fundamental for the probability theory in its whole, and its role for limit theorems was recognized already in the last century. The last fifteen pages of the chapter are devoted to the discussion in what ways the basic statement can be generalized for the sums of non-independent random variables.

Cited in

(89)

This page was built for publication: Modern Theory of Summation of Random Variables

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4210937)