A central limit theorem for generalized multilinear forms (Q921705)

Let \(X_ i\), \(i\geq 1\), be independent random variables, and, for each \(I\subset \{1,2,...,n\}\), let \(W_ I\) be a function of \(\{X_ i\), \(i\in I\}\). Suppose further that the \(W_ I\) have mean zero and finite fourth moments, and that \({\mathbb{E}} W_ IW_ J=0\) if \(I\neq J\). Such random variables arise naturally in the decomposition of \textit{W. Hoeffding} [Ann. Math. Stat. 19, 293-325 (1948; Zbl 0032.04101)]. The author is interested in conditions under which the d-homogeneous sum \(W=\sum_{| I| =d}W_ I\), standardized to have variance 1, converges to \(N(0,1)\) as \(n\to \infty\). This is not necessarily to be expected: if \(d\geq 2\), a weighted mixture of centred chi-squared distributions is usual, but the setting is general enough to allow the possibility that all the weights become small as n increases. The main result is that under an asymptotic negligibility condition, it is sufficient that \({\mathbb{E}} W^ 4\to 3\).