Stochastic sub-additivity approach to the conditional large deviation principle (Q1872225)

Given independent stationary sequences \(X_{1},X_{2},\dots,\) \(Y_{1},Y_{2},\dots \) with values in Polish spaces and a bounded measurable function \(\rho,\) the author studies the so called ``quenched'' LDP. This means that for a.s. sequences \(x_{1},x_{2},\dots \), the distributions of \(\frac{1}{n}\sum_{k=1}^{n}\rho (x_{k},Y_{k})\) satisfy the LDP. The motivation for this study originates from data compression. Ingredients of the proof of the conditional LDP are a stochastic version of Hammerley's approximate sub-additivity together with a combination of the asymptotic value method of \textit{W. Bryc} [in: Diffusion processes and related problems in analysis, Vol. I: Diffusions in analysis and geometry. Prog. Probab. 22, 447-472 (1990; Zbl 0724.60032)] and the Gaertner-Ellis theorem.