A coupling approach to randomly forced nonlinear PDE’s. I. (English) Zbl 0991.60056

Let \(H\) be a separable Hilbert space, \(\{e_{k}\}\) its orthonormal basis, and \(S: H\to H\) a mapping Lipschitz continuous on bounded sets. Let \(\eta_{k}\), \(k\geq 1\), be independent identically distributed \(H\)-valued random variables of the form \(\eta_k = \sum^\infty_{j=1} b_{j}\xi_{jk}e_{j}\), where \(b_{j}\geq 0\) satisfy \(\sum^\infty_{j=1} b^2_{j} <\infty\), and for each \(j\geq 1\) the law of the real-valued random variable \(\xi_{jk}\) has a density \(p_{j}\) with respect to Lebesgue measure, \(p_{j}\) being a function of bounded variation, zero outside \([-1,1]\), and such that \(\int_{[-\varepsilon,\varepsilon]} p_{j}(r)dr >0\) for all \(\varepsilon>0\). Long-time behaviour of a Markov chain \((u_{n},\mathbb P_{x})\), defined recursively by \(u_{k} = S(u_{k-1}) + \eta_{k}\), \(k\geq 1\), \(\mathbb P_{x}\{u_0 = x\} =1\), is studied.
Suppose that for any \(R>r>0\) there exist \(a<1\) and \(n_0\in\mathbb N\) such that \(\|S^{n}(y)\|\leq \max(a\|y\|, r)\) for all \(y\in H\), \(\|y\|\leq R\), and all \(n\geq n_0\). Denote by \(K\) the set \(\{\sum^\infty_{j=1} f_{j}e_{j}\), \(|f_{j} |\leq b_{j}\) for all \(j\geq 1\}\), and for any bounded set \(B\) define \(\mathcal A_{0}(B) = B\), \(\mathcal A_{k}(B) = S(\mathcal A_{k-1} (B)) + K\), \(k\geq 1\). Assume that there exists \(\rho >0\) such that for every bounded set \(B\) a \(k_0\in \mathbb N\) may be found such that \(\mathcal A_{k}(B)\) is contained in a ball with radius \(\rho\) centered at \(0\) for all \(k\geq k_0\). Let \(Q_{N}\) be the orthogonal projection onto the orthogonal complement of the space spanned by \(e_1,\ldots,e_{N}\). For each \(R>0\), let there exist \(\gamma_{N}(R)\), \(\gamma_{N}(R)\searrow 0\) as \(N\to\infty\), such that \[ \|Q_{N}S(y_1)-Q_{N}S(y_2)\|\leq \gamma_{N}(R)\|y_1 - y_2\| \] for all \(y_1,y_2\in H\), \(\|y_{i}\|\leq R\). Under the above hypotheses it is proven by using the coupling method that there exists an \(N\in\mathbb N\) such that the chain \(\{u_{n}\}\) has a unique invariant measure \(\mu\) whenever \(b_{j}\neq 0\), \(j=1,\ldots, N\). Moreover, \[ |\mathbb E_{x}f(u_{k}) - \mu(f)|\leq C_{R} \exp(-ck^{1/2}) \Bigl(\sup_{H}|f|+ \text{Lip}(f)\Bigr) \] for some constant \(c>0\), all bounded Lipschitz continuous functions \(f\), all \(R>0\) and \(x\in H\), \(\|x\|\leq R\); with a constant \(C_{R}\) dependent only on \(R\). Markov chains of the above type have arisen in the authors’ investigations concerning two-dimensional Navier-Stokes equations perturbed by a random kick-force [Commun. Math. Phys. 213, No. 2, 291-330 (2000; Zbl 0974.60046)].


60J05 Discrete-time Markov processes on general state spaces
35R60 PDEs with randomness, stochastic partial differential equations
37H10 Generation, random and stochastic difference and differential equations
76F55 Statistical turbulence modeling


Zbl 0974.60046
Full Text: DOI