Let (Ω,cal A,P) be a probability space and assume that
the probability measure P has the following representation:
P(·)
= intΘ P(θ,·)dm(θ)
where (Θ,cal M,m) is a
probability space (with Θ a first countable topological space)
and P(·,·) a probability transition function on
Θ×cal A. aLet also, Yjspj=1infty be a sequence of random
variables defined on (Ω,cal A,P) and taking values in an
arbitrary measurable space (S,cal S) and assume that for each
θinΘ, Yjspj=1infty is an i.i.d. sequence under
P(θ,·). Let the empirical measures of Yjspj=1infty be
defined by:
Ln=1over nsumspj=1nδYj
with δx the Dirac
measure at the point x, and let νn=cal LP(Ln). a
Define:
Mn=nover bn(Ln-μ)
with μ=P o (Y1)-1 and nspn=1infty
a positive real sequence such that:
bnover
n1over2uildreln→inftyoverlongrightarrowinfty, bnover
nuildreln→inftyoverlongrightarrow 0eqno(*)
and let
ildeνn=cal LP(Mn).
In chapter 2 of this dissertation, we
study Large Deviations for the sequence of probability measures
νnspn=1infty. In chapter 3, a Moderate Deviations result with
normalizing constants spn2over nspn=1infty, for the sequence
ildeνspn=1infty is proved aNow, let (S,cal S) (Rd,cal Bd,dge1 and define
eqalign S0 & = 0cr Sj &=
sumspi=1jYi, j=1,2,3,···
and denote by sn(t),t in [0,1] the
polygonal line in Rd determined by the points (jover n, Sjover xn),
j=0,1,2,···, n (trajectories of Yjspj=1infty), with xnspn=1infty
a positive real sequence and let μn=cal LP=(sn(·)).
In
chapter 4, we study Large Deviations for the sequence of
probability measures μnspn=1infty, when xn = n. Finally, in
chapter 5 we prove a Moderate Deviations result with normalizing
constants xspn2over nspn=1infty, for the sequence μnspn=1infty
when xn=bn and bn is as in (*)