Lipschitz constants for some approximation operators of a Lipschitz continuous function (Q908458)

Let \(I_ k\subset {\mathbb{R}}^ k\) be a convex subset and let f be a real function with \(f\in C(I_ k)\). Suppose that \(S_{n,x}\), \(n\geq 1\), is a sequence of k-dimensional random vectors over \(I_ k\) with mean \(E(S_{n,x})=nx\) and with \(Cov(S_{n,x}S_{n,x})=\sigma^ 2_ n(x)\) a nonsingular variance-covariance matrix, where x is a k-dimensional vector of parameters. The authors study the multivariate version of the Feller operator, namely \[ L_ n(f,x):=E\{f(S_{n,x}/n)\}:=\int^{\infty}_{- \infty}...\int^{\infty}_{-\infty}f(t_ 1/n,...,t_ k/n)dF_{n,x}(t_ 1,...,t_ k), \] where \(F_{n,x}\) is the joint distribution function of \(S_{n,x}\) and \(E| f(S_{n,x}/n)| <\infty\). A ``splitting property'' of \(S_{n,x}\) is introduced; it means that for every \(x,y\in I_ k\) there exist random vectors U, \(R_ x\), and \(R_ y\) such that U, \(R_ x\), and \(R_ y\) are defined over the same probability space with \(S_{n,x}\sim U+R_ x\) and \(S_{n,y}\sim U+R_ y\), and \(E\| R_ x-R_ y\| \leq n\| x-y\|\) \((\| \cdot \|\) is the \(\ell_ 1\) norm in \({\mathbb{R}}^ k)\). Here \(R_ x\) and \(R_ y\) need not be independent. Further, \(f\in Lip_ A \mu (I_ k)\) means that \(| f(x)-f(y)| \leq A\| x-y\|^{\mu},\) \(x,y\in I_ k\) and \(\mu\in (0,1]\). The main theorem then reads: Assume that \(S_{n,x}\) has the splitting property. If \(f\in Lip_ A \mu (I_ k)\), then \(L_ n(f)\in Lip_ A \mu (I_ k)\). Finally, the authors show that a ``probability interpretation'' of many well-known operators is the splitting property (Bernstein operators, Szász operators, Baskakov operators, Gamma operators, etc.) Thus this theorem can be applied to such operators.