Three parts composes this thesis. Firstly the convergence rates of conditional moments of random variables were discussed as the distribution function belongs to the domains of attraction of D(Φα),D(Ψα) and -D(Λ) respectively. The main results are:Theorem A If is p - order conditional moments of X, let ,sgn(A0t) is constant for large t, satisfying (2.4) and (2.5), thenTheorem B If F ,let x0 = sup{x : F(x) < 1},B0(t) = p +α, sgn(B0(t))is constant for large t, satisfying (2.9) and (2.10), thenTheorem C If F ,let P(t) = - log Jp(t) satisfying (2.14), thenwhereIn the second part,we give the large quantile (xPn) estimation of fixed smoothing parameter:Theorem D If F∈D(Gγ) (γ> 0),there exists a regularly varying function and lim (t) =∞, such that lim∞locally uniformly for x > 0. m is a fixed constant,npn→∞, pn→0 as n→∞, then Theorem E If F∈D(Gγ) (γ< 0),there exists a regularly varying function ,and lim sup locally uniformly for x > 0. m is a fixed constant, npn→∞,pn→0 as n→∞,thenTheorem F If F∈D(Gγ) (γ= 0),there exists a regularly varying functionÏ(t), lim ,and lim sup for all x > 0, where dy. m is a fixed constant,npn→c as n→∞,c∈R+, thenIn the thrid part, almost sure central limit theorem of maximum and minima on dependent Gaussian sequence was analyzed, which wasTheorem G suppose be a standardized stationary Gaussion sequences with covariance rn→0 as n→∞,andwhereε> 0,γ> 2. And(1). If there exist constants ofμn,vn and 0≤τ< 00,0≤θ<∞, such that and are bounded,then...
|