The small ball problem for the Brownian sheet (Q1345602)

Let \(B_{s,t}\), \(s,t \in \mathbb{R}^+\), be a Brownian sheet, i.e. the centered Gaussian process with continuous realizations and covariance \(EB_{s,t} B_{s',t'} = \min (s,s') \min (t,t')\). Denote by \(\lambda\) the Lebesgue measure on \([0,1]^2\), by \(|\cdot |_2\) the usual norm in \(L^2 (\lambda)\), and by \(|\cdot |_{\psi_\alpha}\) the Orlicz norm \[ |f |_{\psi_\alpha} = \inf \Bigl \{ c > 0 : \int \exp \bigl \{ |f |^\alpha/c \bigr \} d \lambda \leq 2 \Bigr\}, \quad \alpha \geq 2. \] The following results are obtained: For a universal constant \(C\) and \(0 < \varepsilon \leq 1/2\), \[ \exp \left \{ - {C \over \varepsilon^2} \left( \log {1 \over \varepsilon} \right)^3 \right \} \leq P \Bigl[ \sup_{0 \leq s,t \leq 1} |B_{s,t} |\leq \varepsilon \Bigr] \leq \exp \left \{ - {1 \over C \varepsilon^2} \left( \log {1 \over \varepsilon} \right)^3 \right \}, \] \[ P \bigl[ |B_{s,t} |_2 \leq \varepsilon \bigr] \leq \exp \left \{ - {C \over \varepsilon^2} \left( \log {1 \over \varepsilon} \right)^2 \right\}, \] and, given \(2 \leq \alpha < \infty\), there exists a constant \(K (\alpha)\) such that for all \(0 < \varepsilon \leq 1/2\), \[ \left \{ - {K (\alpha) \over \varepsilon^2} \left( \log {1 \over \varepsilon} \right)^{3 - 2/ \alpha} \right \} \leq P \bigl[ |B_{s,t} |_{\psi_\alpha} \leq \varepsilon \bigr] \leq \left \{- {1 \over K (\alpha) \varepsilon^2} \left( \log {1 \over \varepsilon} \right)^{3 - 2/ \alpha} \right\}. \] The most interesting part of the article, namely the proof of upper bounds for sup and Orlicz norms, is based on some combinatorial estimates for linear operations formed by means of a special orthogonal system in \(L^2 (\lambda)\) similar to Haar functions. The results for sup norm are also generalized to the case of Gaussian processes \(B^\mu_{s,t}\) such that \[ EB^\mu_{s,t} B^\mu_{s',t'} = \mu \biggl \{ \bigl[ 0, \min (s,s') \bigr] \times \bigl[ 0, \min (t,t') \bigr] \biggr\}, \] where \(\mu\) is a positive measure on \([0,1]^2\).