A general estimate in the invariance principle (Q642084)

Let \(X_1, X_2,\dots\) be independent random variables with mean zero and finite variances. Define \(S_n= X_1+\cdots+ X_n\), \(t_n= \sum_{k\leq n}\text{Var\,}X_k\). General estimates for the accuracy of an approximation \(S_n\) by a suitably constructed Wiener process are obtained. Their form depend on a moment condition for summands expressed in terms of \(H(X_i)\) with \(H\) belonging to the class of functions defined by \textit{J. Komlós}, \textit{P. Major} and \textit{G. Tusnády} [Z. Wahrscheinlichkeitstheor. Verw. Geb. 34, 33--58 (1976; Zbl 0307.60045)]. One of the corollaries states that there exists a Wiener process \(W\) defined on the same probability space as \((X_n)\) such that for all \(b\geq K_0\) and \(x\geq bx_\varepsilon\) \[ P\Biggl(\max_{k\leq n}|S_k- W(t_k)|\geq Cx\Biggr)\leq P\Biggl(\max_{k\leq n}|X_k|\geq x/b\Biggr)+ t_n x^{-2} H^{-b\varepsilon}(x)+ t_n x^{-2} e^{-2x/x_\varepsilon}, \] where \(C\) is an absolute constant while \(\varepsilon\), \(x_\varepsilon\) and \(K_0\) are defined by the moment condition. It should be noted that the Wiener process \(W\) does not depend on \(x\) and the estimate shows explicit dependence on the distributions of summands.