Sc. Carmona et al., A LEVEL-1 LARGE-DEVIATION PRINCIPLE FOR THE AUTOCOVARIANCES OF UNIQUELY ERGODIC TRANSFORMATIONS WITH ADDITIVE NOISE, Journal of statistical physics, 91(1-2), 1998, pp. 395-421
A large-deviation principle (LDP) at level 1 for random means of the t
ype M-n = 1/n Sigma(f = 0)(n - 1) Z(j)Z(j + 1), n = 1, 2,... is establ
ished. The random process {Z(n)}(n greater than or equal to 0) is give
n by Z(n) = Phi(X-n) + xi(n), n = 0, 1, 2,..., where {X-n}(n greater t
han or equal to 0) and {xi(n)}(n greater than or equal to 0) are indep
endent random sequences: the former is a stationary process defined by
X-n = T-n(X-0), X-0 is uniformly distributed on the circle S-1, T: S-
1 --> S-1 is a continuous, uniquely ergodic transformation preserving
the Lebesgue measure on S-1, and {xi(n)}(n greater than or equal to 0)
is a random sequence of independent and identically distributed rando
m variables on S-1; Phi is a continuous real function. The LDP at leve
l 1 for the means M-n is obtained by using the level 2 LDP for the Mar
kov process {V-n = (X-n, xi(n), xi(n + 1))}(n greater than or equal to
0) and the contraction principle. For establishing this level 2 LDP,
one can consider a more general setting: T: [0, 1) --> [0, 1) is a mea
sure-preserving Lebesgue measure, Phi: [0, 1) --> R is a real measurab
le function, and xi(n) are independent and identically distributed ran
dom variables on R (for instance, they could have a Gaussian distribut
ion with mean zero and Variance sigma(2)). The analogous result for th
e case of autocovariance of order ic is also true.