Stationary random fields with linear regressions

From MaRDI portal
Publication:1872197



Abstract: We analyze certain stationary fields with linear regressions and quadratic conditional variances. This classic probabilistic problem leads somewhat unexpectedly to stationary Markov processes closely tied to non-commutative probability through the q-Hermite polynomials.


Let \((X_k)\) be a square-integrable random field with the first two conditional moments given by \[ E(X_k|\dots,X_{k-2}, X_{k-1},X_{k+1},X_{k+2},\dots)= L(X_{k-1},X_{k+1}), \] \[ E(X_k^2|\dots,X_{k-2},X_{k-1},X_{k+1}, X_{k+2},\dots)= Q(X_{k-1},X_{k+1}), \] where \(L(x,y)=a(x+y)+b\), \(Q(x, y)=A(x^2+y^2)+Bxy+C+D(x+y)\). Assume that \((X_k)\) is \(L_2\)-stationary and that all one-dimensional distributions are equal. The author gives sufficient conditions under which the one-dimensional distributions are either normal or compactly supported. The proofs are based on polynomials \(Q_n(x)\) defined by \(xQ_n(x)=Q_{n+1}(x)+(1+q+\cdots+q^{n-1})Q_{n-1}(x)\) with \(Q_{-1}(x)=0\), \(Q_0(x)=1\). Finally, Markov fields with the conditional moments \(L(X_{k-1},X_{k+1})\) and \(Q(X_{k-1}, X_{k+1})\) are constructed.




Cited in
(31)








This page was built for publication: Stationary random fields with linear regressions

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q1872197)