An \(L^p\)-view of the Bahadur-Kiefer theorem (Q2568544)

Let \((U_n)\) be an i.i.d. sequence such that \(U_1\) is uniformly distributed on \((0,1)\). Let \[ \alpha_n(t):=n^{1/2}(F_n(t)-t), \quad n\leq t\leq 1, \] be the uniform empirical process and \[ \beta_n(t):=n^{1/2}(F^{-1}_n(t)-t),\quad n\leq t\leq 1, \] the uniform empirical quantile process. Here, \(F_n(\cdot)\) is the empirical distribution function based on \(U_1,\dots,U_n\), and \[ F^{-1}_n(t):=\inf\{s>0:F_n(s)\geq t\}, \quad 0<t\leq 1, \] \((F^{-1}_n(0):=F^{-1}_n(0+))\) is the quantile function of \(F_n\). The (\((0,1)\)-uniform) Bahadur-Kiefer process is given by \[ R_n(t):=\alpha_n(t)+\beta_n(t), \quad 0\leq t \leq 1. \] The following result is well-known (\(\|\cdot \|\) denoting the sup-norm): \[ \|R_n\|/\sqrt{\|\alpha_n\|} \sim n^{-1/4}(\log n)^{1/2}\quad \text{a.s.}, \] see, e.g., \textit{P. Deheuvels} and \textit{D. M. Mason} [Ann. Probab. 18, 669-697 (1990; Zbl 0712.60028)]. In the present paper it is shown (\(\|\cdot\|_p\) denoting the \(L^p\)-norm) that \[ n^{1/4}\|R_n\|_p/\sqrt{\|\alpha_n(p/2)} \to c_0(p)\;\text{a.s.}\;(2\leq p<\infty), \] (\(c_0(p)>0\) denoting a finite (explicity known) constant). In particular, \(c_0(2)=1\).