非参数回归函数核估计的一致强收敛速度

设(X,Y)为d×1随机向量,f(x,y)为其概率密度函数,(X_i,Y_i) i=1,2,…,n为抽自f的i. i. d. 样本,m(x)(?)E(Y|X=x)称Y对X的回归函数。Watson (1964),Nagaraya (1964)提出用m_n(x)=sum from i=1 to n (Y_iK(?))/sum from i=1 to n (K((x-X_i)/h_n))估计m(x),其中K(x)为R~d上的概率密度,h_n>0,h_n→0(n→∞),这种估计称核估计。引入记号:ω(x)(?) integral from R~1 to ∞(yf(x,y)dy),g(x)(?) integral from R~1 to ∞(f(x,y)dy),又ω_n(x)(?)1/(nh_n~d) sum from i=1 to n (Y_iK)((x-X_i)/h_n),g_n(x)(?)1/(nh_n~d) sum from i=1 to n (K((x-X_i)/h_n)),它们分别是ω(x)和g(x)的估计。则m(x)=ω(x)/g(x),m_n(x)=ω_n(x)/g_n(x)(约定0/0=0)。当d=1时,E. Schuster和S. Yakowitz(1979)证明了在一组条件下,存在常数c>0,他对(?)ε>0,当n充分大时,其中,