Convergence of the empirical two-sample \(U\)-statistics with \(\beta\)-mixing data (Q2043739)

The authors investigate a two-parameter empirical \(U\)-process, when the underlying data are generated by a stationary mixing stochastic process \({({X_i})_{i \geqslant 1}}\). They analyze the asymptotic distribution of the centered and normalized empirical distribution function of the data \(g({X_i},{X_j}),1 \leqslant i \leqslant \left[ {nt} \right],\left[ {nt} \right] + 1 \leqslant j \leqslant n\), given by \({e_n}(s,t): = \frac{1}{{{n^{{3 \mathord{\left/ {\vphantom {3 2}} \right. \kern-\nulldelimiterspace} 2}}}}}\sum\limits_{i = 1}^{\left[ {nt} \right]} {\sum\limits_{j = \left[ {nt} \right] + 1}^n {\left( {1\left\{ {g({X_i},{X_j}) \leqslant s} \right\} - P\left\{ {g({X_i},{X_j}) \leqslant s} \right\}} \right)} } \), \(0 \leqslant t \leqslant 1,s \in R\), and show that this process converges as \(n \to \infty \) to a Gaussian limit process if the underlying process \({({X_i})_{i \geqslant 1}}\) is short range dependent.