Asymptotic properties and Hausdorff dimensions of fractional Brownian sheets (Q818504): Difference between revisions

Let \(B^{H}=\{ B_{k}^{H}( t)\), \(t\in R^{N}\), \(1\leq k\leq d\} \) be an \((N,d)\)-fractional Brownian sheet of the index \(H=( H_{1},\dots ,H_{N}) \in (0,1) ^{N}\) with independent components \(\{ B_{k}^{H}(t)\), \(t\in R^{N}\} \) having the correlation functions \(E( B_{k}^{H}(s)B_{k}^{H} (t))= \prod_{j=1}^{N}\frac{1}{2}(| s_{j}| ^{2H_{j}}+| t_{j}| ^{2H_{j}}-| s_{j}-t_{j}| ^{2H_{j}}) \), \(s,t\in R^{N} \). It follows that \(B_{k}^{H}\) is an anisotropic Gaussian random field and \( B_{k}^{H}=0\) a.s. for every \(t\in R^{N}\) with at last one zero coordinate. Some asymptotic and fractal properties of the random sheet \(B^{H}\) are discussed. The uniform and local modulus of continuity and the law of the iterated logarithm for \(B^{H}\) are considered in Section 2 as \(t\in [0,1] ^{N}\) or \(| t| \rightarrow \infty .\) The proofs of the Lemarié-Meyer wavelet basis for \(L^{2}(R)\) are given. This approach enables to prove that, with probability 1, the sample function of \(B^{H}(t)\) does not satisfy any local Hölder conditions with respect to the function \(\sigma ^{1+\varepsilon} (t+h,t)\) for any \(\varepsilon >0\), where \(\sigma ^{2}( t+h,t) =E(B^{H}(t+h)- B^{H}(t)) ^{2}.\) For simplicity, the proofs are presented for \(d=1\). In Section 3, the Hausdorff and packing dimensions of the range \(B^{H}[0,1] ^{N}\), \(1\leq d\), the graph \(\operatorname{Gr}B^{H} [0,1] \) and the level set \(L_{x}=\{t\in [0,\infty)^{N}\), \(B^{H}(t)= x\}\), \(x\in R^{d}\), are obtained. Namely, with probability 1, \[ \dim _{H}B^{H}[ 0,1] ^{N}= \min \left( d,\sum_{j=1}^{N}1/H_{j}\right) ; \] \[ \dim _{H} \operatorname{Gr}B^{H}([ 0,1]^{N})= \min \left\{ \sum_{j=1}^{k}H_{k}/H_{j}+ N-k+ (1-H_{k})d;\;\sum_{j=1}^{N} 1/H_{j},\;1\leq k\leq N\right\} . \] The level set \(L_{x}\) is empty if \(\sum_{j=1}^{N} 1/H_{j}<d\) for every \(x\in R^{d}.\) If \(\sum_{j=1}^{N} 1/H_{j}>d\), then, for any \(x\in R^{d}\) and \(0<\varepsilon <1\), with positive probability \[ \dim _{H}( L_{x}\cap [ \varepsilon ,1]^{N})= \min \left\{ \sum_{j=1}^{k}H_{k}/H_{j}+ N-k-H_{k}d,\;1\leq k\leq N\right\} . \] The case \(\sum_{j=1}^{N}1/H_{j}=d\) is unproved. The packing dimension is considered as well and the formulas are similar. The previous investigations on similar problems are reviewed. The list of references consists of 36 items.