Extensions of results of Komlós, Major, and Tusnády to the multivariate case (Q1122862)

The purpose of this excellent paper is to establish extensions of the well-known Komlós-Major-Tusnády strong approximations to the multivariate case. One of the main results can be stated as follows: Let H be a continuous, nonnegative function on [0,\(\infty)\) such that \(H(t)/t^{3+r}\) is eventually increasing for some \(r>0\) and log H(t)/t\({}^{1/2}\) is eventually non-increasing. Suppose X is a d- dimensional random variable with mean 0, covariance matrix \(\Sigma\) and E H(\(| X|)<\infty\), where \(| \cdot |\) denotes the Euclidean norm. Then i.i.d. sequences \(\{X_ n\}\), \(\{Y_ n\}\) can be constructed in such a way that \(X_ n=^{D}X\), \(Y_ n\) is N(0,\(\Sigma)\)- distributed and \[ T_ n=\sum^{n}_{1}X_ k-\sum^{n}_{1}Y_ k=O(H^{-1}(n))\quad a.s. \] If the moment generating function of X exists and satisfies a mild smoothness condition then the rate O(log n) can be achieved. Simultaneously with the above, anologues of KMT-type inequalities (e.g. exponential inequalities) for \(T_ n\) are also obtained. The basic tool in the proofs is an extension of the quantile transformation method of KMT to the multidimensional case. To get this a large deviation theorem for conditional distribution functions is first proved.