Point processes, regular variation and weak convergence (Q3730739)

From MaRDI portal

!

This is the item page for this Wikibase entity, intended for internal use and editing purposes. Please use the normal view instead:

scientific article; zbMATH DE number 3962884
Language Label Description Also known as
default for all languages
No label defined
    English
    Point processes, regular variation and weak convergence
    scientific article; zbMATH DE number 3962884

      Statements

      Point processes, regular variation and weak convergence (English)
      0 references
      1986
      0 references
      functional limit theorems
      0 references
      weak convergence
      0 references
      vague convergence
      0 references
      Radon measure
      0 references
      regularly varying tails
      0 references
      extremal process
      0 references
      regular variation of the tail of a convolution
      0 references
      two-parameter stable processes
      0 references
      0 references
      The central subject of this interesting paper is a technique to derive limit theorems, mostly functional limit theorems for stochastic processes, by proving weak convergence (in distribution) of point processes to a limiting point process. Let E be a locally compact space with denumerable topology base and \(M_ p(E)\) the set of point measures on E, i.e. measures \(\sum_{i}\epsilon (x_ i)\), where \(\epsilon\) (a) is the unit mass at a, such that any relatively compact set contains finitely many \(x_ i\). The set \(M_ p(E)\) is given the topology of vague convergence, which makes it a complete separable metric space. A point process on E is a measurable mapping \(\xi\) from a probability space to \(M_ p(E)\). It could be the identity on \(M_ p(E)\) [see \textit{J. Neveu}, École d'été Probab. Saint-Flour VI-1976, Lect. Notes Math. 598, 249-447 (1977; Zbl 0439.60044)]. A sequence \(\{\xi_ n\}\) of point processes on E converges weakly to a point process \(\xi\) on E if \(E(f\circ \xi_ n)\to E(f\circ \xi)\), with \(E=\exp ectation\), for every bounded continuous \(f: M_ p(E)\to {\mathbb{R}}.\) The starting point is the following theorem (*). Let \(X_{nj}\), \(j=1,2,...\), be independent E-valued random variables with common distribution \(P_ n\). Then the sequence of point processes \(\sum_{j}\epsilon ((jn^{-1},X_{nj}))\), \(n=1,2,...\), on (0,\(\infty)\times E\) converges weakly to a Poisson process on (0,\(\infty)\times E\) with mean measure dt\(\times d\mu\), where \(\mu\) is a Radon measure on E, iff \(nP_ n\) converges vaguely to \(\mu\). A corollary for \(E={\mathbb{R}}^ d:\) Let \(X_ n\), \(n=1,2,...\), be i.i.d. random vectors in \({\mathbb{R}}^ d\) and \(0<a_ n\to \infty\). Then the point process \(\sum_{k}\epsilon ((kn^{-1},a_ n^{-1}X_ n))\) converges weakly to a Poisson process on \((0,\infty)\times {\mathbb{R}}^ d\) with mean measure dt\(\times d\nu\), where \(\nu (\{\| x\| \geq r)<\infty\), \(r>0\), iff \(nP(a_ 1^{-1}X_ n\in \cdot)\) converges vaguely to \(\nu\). Here the sets \(\{\| x\| \geq r\}\) are taken to be compact, i.e. we transform \(x\to \| x\|^{-2}x\). The condition is one of the equivalent definitions of \(X_ 1\) having regularly varying tails. We illustrate the technique at the simplest example. Let \(X_ n\), \(n=1,2,...\), be i.i.d. random variables with \(P(X_ 1\leq x)=F(x)\) such that \(P(M_ n\leq a_ nx+b_ n)\to G(x)\) or, equivalently, \(n(1-F(a_ nx+b_ n))\to \log G(x)\), where \(M_ n=\max (X_ 1,...,X_ n)\). Then (*) with \(X_{nj}=(X_ j-b_ n)/a_ n\) if \((X_ j-b_ n)/a_ n\in S=\sup p(G)\) and \(X_{nj}=\inf S\) otherwise, gives that \(\sum_ k\epsilon ((kn^{-1},X_{nk}))\) converges weakly to a Poisson process on (0,\(\infty)\times S\) with mean measure dt\(\times d\nu\), \(\nu [x,\infty)=-\log G(x)\). By considering the (a.s. continuous) functional \(T: M_ p((0,\infty)\times S)\to D[0,\infty)\) with T(\(\sum \epsilon (x_ k,y_ k))(t)=\max \{y_ k| x_ k\leq t\}\) it is proved that the process \(Y_ n(t)=(M_{[nt]}-b_ n)/a_ n\) converges weakly in D[0,\(\infty)\) to the extremal process generated by G. The technique is applied to: convergence to a stable process, regular variation of the tail of a convolution and of the distribution of XY, \(X>0\), \(Y\in {\mathbb{R}}^ d\) (by the ''only if'' part), last exit times sup\(\{\) \(k\geq 0:\) \(X_ k>\beta k\}\) as \(\beta\) \(\to 0\), joint limit laws for sample mean and variance and for sum and maximum, self-normalized sums and maxima, two-parameter stable processes, slowly varying tails and subordinated distributions and randomly stopped sums. Since often here the conditions for convergence involve regular variations of tails, the paper also serves as a useful survey of the connections between regular variation and probability limit theorems.
      0 references

      Identifiers

      0 references
      0 references
      0 references
      0 references
      0 references
      0 references