Sequences of Chance Events without After-Effects

Author:
  • AKA. Ya. Khinchin
    A. Ya. Khinchin

Abstract
A sequence of similar chance events is a random process $x(t)$$(t \geqq 0)$ in which the function $x(t)$ [the number of events taking place in the time interval $(0,t)$] never decreases and can take only nonnegative integer values. We have a complete stochastic description of the given sequence if we are given, for any finite set $0 < t_1 < t_2 < \cdots < t_n $ of moments, the n-dimensional distribution law of the vector $[x(t_1 ),x(t_2 ), \cdots ,x(t_n )]$ . A sequence of events is said to be stationary if the distribution law of the vector $[x(a + t_i ) - x(a),1 \leqq i \leqq n]$ depends only on $t_1 ,t_2 , \cdots ,t_n $, and not on a and hence, for every $a \geqq 0$, coincides with the distribution law of the above-mentioned vector $x(t_i ),1 \leqq i \leqq n$. The sequence is said to be without after-effects if, for any set of pairwise nonintersecting time-intervals $(u_i ,v_i ),1 \leqq i \leqq n$, the differences $x(v_i ) - x(u_i )$ are mutually independent random quantities. If $v_k (\alpha ,\beta )$$(0 \leqq \alpha \leqq \beta ,k \geqq 0)$ denotes the probability of the equation $x(\beta ) - x(\alpha ) = k$ [i.e., the probability that k events of the given sequence occur in the course of the interval $(\alpha ,\beta )$], then a sequence without after-effects is obviously completely defined by the specification of the system of functions $v_k (\alpha ,\beta ),k = 0,1,2, \cdots $ for then, the distribution of the difference $x(\beta ) - x(\alpha )$ is known for every interval $(\alpha ,\beta )$. Moreover, if the sequence is stationary, then the quantity $v_k (\alpha ,\alpha + t)$ depends only on k and t and can be denoted by $v_k (t)$ (the probability of the occurrence of k events in an arbitrary interval of length t). If to the two requirements of stationary character and absence of after-effects we add another condition, namely that\[ \mathop {\lim }\limits_{t \to + 0} \frac{{1 - v_0 (t) - v_1 (t)}}{t} = 0 \]then, as is known ([1], Chap. 1),\[ (1)\qquad v_k (t) = e^{-\lambda t} \frac{{(\lambda t)^k }}{{k1}}\quad (k = 0,1,2, \cdots ). \]The distribution of the quantity $x(t)$ is thus completely defined up to the arbitrary positive constant $\lambda $ under the above three assumptions ($\lambda $ is called the “intensity” of the given sequence). Recently the general form of stationary sequences without after-effects, without the additional condition (1), was found by Redhef f er [2] and the author ([1], § 8). On the other hand, the form of sequences without after-effects, satisfying (1) but not in general stationary, has long been known.Lately one has found that on the one hand the investigations of sequences without after-effects of a very general type are valuable for applications. On the other hand, methods have been found making such researches possible for some special problems (see, e.g., the examination of Erlang’s well-known problem by Palm [3] or by the author ([1], § 24), and the work of Takács on the particle effect). Accordingly, the question naturally arises, what is the general form of sequences without after-effects? The present work contains a solution of this problem; the solution seems satisfactory, since all the arbitrary elements of the required structure are revealed in it with complete clarity. In addition to the absence of after-effects, let us assume also that the mathematical expectation $\Lambda (t)$ of the number of events of the given sequence occurring in the interval $(0,t)$ is finite for all $t \geqq 0$. For a given sequence a point t is said to be regular or singular according as $\Lambda (t)$ is continuous or discontinuous at t; if the sequence has only regular points, the sequence itself is said to be regular. Finally, the function \[ \Phi (x,\alpha ,\beta ) = \sum\limits_{k = 0}^\infty {v_k (\alpha ,\beta )x^k } \]is called the generating function of the given sequence. The sequence (without after-effects) is thus completely described by the specification of $\Phi (x,\alpha ,\beta )$. Theorem 1. Every regular sequence of chance events without after-effects has a generating function of the form\[ \Phi (x,\alpha ,\beta ) = \exp \left\{ {\sum\limits_{k = 0}^\infty {\left[ {\chi _k (\beta ) - \chi (\alpha )} \right]x^k } } \right\} \]where the functions$\chi _k (t)[\chi _k (0) = 0]$satisfy the following conditions: 1. $\chi _k (t)$is continuous for$t \geqq 0$; 2. $\chi _k (t)$is nonincreasing for$k = 0$and nondecreasing for$k > 0$;\[ 3.\qquad \sum\limits_{k = 0}^\infty {k\chi _k (t)} < + \infty \qquad (t \geqq 0); \]\[ 4.\qquad \sum\limits_{k = 0}^\infty {\chi _k (t)} = 0\qquad (t \geqq 0). \]Conversely, every such function is the generating function of a regular function of chance events without after-effects. Now a sequence S of chance events is said to be singular if it has the following three properties: 1. The events of the sequence S can occur only in a predetermined enumerable number of instants $t_i (i = 1,2, \cdots )$. 2. The numbers of events occurring at different steps $t_i $ are mutually independent random quantities (thus S is a sequence without after-effects). 3. Let $q_k^{(i)} (k \geqq 0,i \geqq 1)$ denote the probability that at step $t_i $ there are exactly k events of S; then for all $t > 0$\[ \sum\limits_{0 \leqq t_i < t} \sum\limits_{k = 0}^\infty kq_k^{(i)} < + \infty .\]Obviously a singular sequence is uniquely defined by the specification of the numbers $t_i $ and $q_k^{(i)} (k \geqq 0,i \geqq 1)$. For general sequences without after-effects [with a finite function $\Lambda (t)$] we have the following Theorem 2. Any sequence without after-effects can be regarded as a superposition of two mutually independent sequences o f the same type, one of which is regular and the other singular; the parameters$t_i q_k^{(i)} $ of the singular component are thereby obtained directly from the function$\Lambda (t)$. Since the nature of the regular component is completely described...

This publication has 3 references indexed in Scilit: