The almost sure invariance principles of degenerate \(U\)-statistics of degree two for stationary random variables (Q1316605)

The authors prove an almost sure invariance principle for bivariate degenerate \(U\)-statistics \(U_ n= \sum_{1\leq i\neq j\leq n} h(\xi_ i, \xi_ j)\) in the case where the underlying observations \((\xi_ i)\) are weakly dependent. Strong mixing, absolutely regular and \(\varphi\)- mixing processes with suitable rates are studied. The limiting process can be represented as a weighted sum of dependent chi-square processes. The proof is based on the Hilbert-Schmidt expansion of the kernel \(h(x,y)= \sum_ i \lambda_ i g_ i(x) g_ i(y)\). This makes it possible to express the \(U\)-statistic as a functional of partial sums of the vectors \((g_ 1(\xi_ i), g_ 2(\xi_ i),\dots)\). To the latter an a.s. invariance principle in a suitable Hilbert space is applied.