Exact Rosenthal-type bounds (Q888535)

Rosenthal inequalities provide upper bounds on moments (of order \(p>2\)) of partial sums of sequences of independent random variables (in terms of the variance of these partial sums), and have found applications in several areas, for example in proving almost sure invariance principles or convergence rates in the strong law of large numbers. The main result of the present paper is a tight upper bound on \(\mathbb{E}|X+S_\mathbf{X}|^q\), where \(5\leq q\leq p\), \(X\) is a mean-zero random variable (with finite \(q\)th moment) independent of all else, and \[ S_\mathbf{X}=X_1+\dots+X_n, \] where \(X_1,\dots,X_n\) are independent mean-zero random variables such that \[ \sum_{i=1}^n\mathbb{E}|X_i|^p\leq A,\qquad\sum_{i=1}^n\mathbb{E}X_i^2\leq B. \] The author proves that, in this situation, \[ \mathbb{E}|X+S_\mathbf{X}|^q\leq\max\left\{\mathbb{E}|X+c\tilde{\Pi}_\lambda|^q,\mathbb{E}|X-c\tilde{\Pi}_\lambda|^q\right\}, \] where \(\tilde{\Pi}_\lambda\) has Poisson distribution (independent of \(X\) and with mean \(\lambda\)) which has been centred, and \((\lambda,c)\) is the unique solution to \(c^p\lambda=A\) and \(c^2\lambda=B\). In the special case where \(X=0\) and \(q=p\), this generalises (for \(p\geq5\)) a result of \textit{E. V. Bestsennaya} and \textit{S. A. Utev} [Sib. Math. J. 32, No. 1, 139--141 (1991; Zbl 0731.60014); translation from Sib. Mat. Zh. 32, No. 1(185), 171--173 (1991)]. This main result is complemented by a similar bound on \(\mathbb{E}|X+S_\mathbf{X}|^p\) for \(2<p\leq3\), and by an upper bound under the additional restriction that each of the \(X_i\) is symmetrically distributed. The proofs use a calculus of variations of moments of infinitely divisible distributions with respect to variations in the Lévy characteristics. The author also details a more general class of moment functions under which such exact Rosenthal-type inequalities may be proved using the techniques of the present work.