Exponential mixing for the 3D stochastic Navier-Stokes equations (Q883037): Difference between revisions

Let \(|\cdot|\) and \((\cdot, \cdot)\) denote the norm and the inner product of \(L^2(D; {\mathbb R}^3)\), respectively, and \(|\cdot|_p\) the norm of \(L^p(D; {\mathbb R}^3)\), where \(D\) is an open bounded domain of \({\mathbb R}^3\) with smooth boundary \(\partial D\) or \(D = (0,1)^3\). \(H^p(D; {\mathbb R}^3)\) denotes the Sobolev space relative to \(L^2(D; {\mathbb R}^3)\) with \(p \in {\mathbb N}\), and its norm \(| X |_{H^p}^2=\sum_{| \alpha | \leqslant p} | \partial_{\alpha} X |^2\). The Sobolev space \(H_0^1(D; {\mathbb R}^3)\) is the closure of the space of smooth functions on \(D\) with compact support by \(| \cdot |_{H^1}\). Note that \(( H_0^1(D; {\mathbb R}^3),\| \cdot \|)\) is a Hilbert space for \(\| X \| = | \nabla X |\). Let \(H\) and \(V\) be the closure of the space of smooth functions on \(D\) with compact support and free divergence for the norm \(| \cdot |\) and \(\| \cdot \|\), respectively. Let \(\pi\) be the orthogonal projection in \(L^2(D; {\mathbb R}^3)\) onto the space \(H\). The purpose of this paper is to study the stochastic Navier-Stokes equations on a three-dimensional bounded domain (NS3D): \[ \begin{cases} d X + \nu A X d t + B(X) d t = \tilde{\varphi}(X) d W + \tilde{f} d t \\ X(0) = x_0 \end{cases} \tag{1} \] where \(W\) is a cylindrical Wiener process on \(H\), \(A :=\pi ( - \varDelta)\) with \(D(A) :=V \cap H^2 (D ; {\mathbb R}^3)\), \(B(u,v):=\pi ( (u, \nabla) v)\), and \(B(u) =B(u,u)\). These equations describe the time evolution of an incompressible fluid subjected to a deterministic and a random exterior force. Here \(X\) means the velocity and \(\nu\) denotes the viscosity, but as it is classical the author gets rid of the pressure \(p\). The external force field acting on the fluid is the sum of a random force field of white noise type \(\widetilde{\varphi}(X) d W\) and a deterministic one \(\widetilde{f} dt\), where symbols \(\widetilde{\varphi}\) and \(\widetilde{f}\) denote the projections of \(\varphi\) and \(f\), but, henceforth, we use the same \(\varphi, f\) with a slight abuse of notation just for simplicity. In the deterministic case \(( \varphi \equiv 0)\), there exists a global weak solution in the PDE sense of (1) when \(x_0\) is square integrable, but uniqueness of such a solution is not known. On the other hand there exists a unique local strong solution when \(x_0\) is smooth, but global existence is an open problem. In the stochastic case, there exists a global weak solution of the martingale problem, but pathwise uniqueness or uniqueness in law remain open problems. The goal of this paper is to establish that, if \(\phi\) is at the same time sufficiently smooth and non-degenerate, then the weak solutions converge exponentially fast to equilibrium: more precisely, given a solution, there exists a stationary solution (which might depends on the given solution), such that the total variation distance between the laws of the given solution and the stationary solution converges to zero exponentially fast. Since the arguments used in dimension two are not applicable to this case because of lack of uniqueness, the author employs the coupling methods, which have been introduced recently in the context of stochastic partial differential equations, by \textit{M. Hairer} [Probab.\ Theory Relat. Fields 124, No. 3, 345--380 (2002; Zbl 1032.60056)] and \textit{S. Kuksin} and \textit{A. Shirikyan} [Commun. Math. Phys. 221, No. 2, 351--366 (2001; Zbl 0991.60056) and J. Math. Pures Appl. (9) 81, No. 6, 567--602 (2002; Zbl 1021.37044)]. In this paper new ideas are introduced, however, the author requires that the noise is sufficiently non-degenerate, and many difficulties disappear. The principal idea consists in the point that coupling of solutions can be achieved for initial data which are small in a sufficiently smooth norm. A coupling satisfying good properties is constructed thanks to the Bismut-Elworthy-Li formula. Another important ingredient in the proof of the present paper is that any weak solution enters a small ball in the smooth norm and also that the time of entering this ball admits an exponential moment. Moreover, the author overcomes the lack of uniqueness of solutions by working with Galerkin approximations. Let us consider \(( e_n)_n\), an eigenbasis of \(H\), associated to the increasing sequence \((\mu_n)_n\) of eigenvalues of \(( A, \text{Dom}(A))\). \(( A^s, \text{Dom}(A^s))\) is the fractional power of the operator \((A, \text{Dom}(A))\) for \(s \in {\mathbb R}\), and set \(\| X \|_s:= | A^{s/2} X |\), and \({\mathbb H}_s:= \text{Dom} ( A^{s/2})\). Define \({\mathcal X} := L_{\text{loc}}^{\infty}({\mathbb R}^+; H)\cap L_{\text{loc}}^2({\mathbb R}^+, V) \cap C({\mathbb R}^+ ; {\mathbb H}_s)\), \({\mathcal W}\) \(:=\) \(C({\mathbb R}^+; {\mathbb H}_{-2})\) and \(\Omega_*\) \(=\) \({\mathcal X} \times {\mathcal W}\), where \(s\) is any fixed negative number. Let \(X_*\) (resp., \(W_*\)) be the projector \(\Omega_* \to {\mathcal X}\) (resp., \(\Omega_* \to {\mathcal W}\)). For a Polish space \(E\), \(P(E)\) denotes the set of probability measures on \(E\) endowed with the Borel-\(\sigma\)-algebra. A probability measure \({\mathbb P}_{\lambda}\) on \(( \Omega_*, {\mathcal F}^*)\) is said to be a weak solution of (1) with initial law \(\lambda \in P(H)\) if \[ \begin{multlined} ( X_* (t), \psi) + \nu \int_0^t ( X_*(s), A \psi) \,ds + \int_0^t ( B( X_*(s)), \psi) \,ds \\ = ( X_*(0), \psi) + t ( f, \psi) + \int_0^t ( \psi, \varphi( X_*(s)) \,d W_*(s)), \quad {\mathbb P}_{\lambda}\text{-a.s.} \end{multlined}\tag{2} \] holds for any \(t \in {\mathbb R}^+\) and any \(\psi\) smooth mapping on \(D\) with compact support and divergence zero. For any \(N \in {\mathbb N}\), \(P_N\) is the eigenprojector of \(A\) associated to the first \(N\) eigenvalues. The following is the approximation of (1): \[ d X_N + \nu A X_N dt + P_N B(X_N) dt = P_N \varphi(X_N) d W + P_N f dt, \quad\text{with } X_N(0) = P_N x_0.\tag{3} \] In order to assure the existence of a weak solution, the following is assumed: (A.1) The mapping \(\varphi\) is bounded Lipschitz \(H \to {\mathcal L}_2(H; {\mathbb H}_1)\) and \(f \in H\), where \({\mathcal L}_2(H; {\mathbb H}_1)\) denotes the space of Hilbert-Schmidt linear operators from \(H\) to \({\mathbb H}_1\). Then it can be shown that, for a subsequence \((N_k)_k\), \(( X_{N_k}, W)\) converges in law to \({\mathbb P}_{x_0}\) a weak solution of (1). Hence the existence of the weak solution of (1) is obtained, but uniqueness remains an open problem. Assume now: (A.2) There exist \(\varepsilon > 0\) and a family \(( \varphi_n)_n\) of continuous mappings \(H \to {\mathbb R}\) with continuous derivatives such that \[ \varphi(x) d W= \sum_{n=1}^{\infty} \varphi_n(x) e_n d W_n \] with \(W =\sum_n W_n e_n\) where \(\sum_{n=1}^{\infty} \sup_{x \in H} | \varphi_n(x) |^2 \mu_n^{1 + \varepsilon}< \infty\). Moreover, for any \(x, y \in {\mathbb H}_2\), \(\sum_{n=1}^{\infty} | \varphi'_n(x) \eta |^2 \mu_n^2<K_1 \| \eta \|_2^2\), \((\exists K_1 > 0)\), \(\varphi_n(x) > 0\) and \(\sup_{x \in H} | \varphi^{-1}(x) |_{ {\mathcal L}({\mathbb H_3}; H)}^2\) \(< \infty\). Here is the main result of the present paper: Theorem. There exist \(\delta^0, C\) and \(\gamma > 0\) (only depending on \(\varphi\), \(D\), \(\varepsilon\) and \(\nu\)) such that, for any weak solution \({\mathbb P}_{\lambda}\) with initial law \(\lambda \in P(H)\) which is a limit of solutions of (3), there exists a weak stationary solution \({\mathbb P}_{\lambda}\) with initial law \(\mu\) such that \[ \| {\mathcal D}_{ {\mathbb P}_{\lambda} }( X_*(t)) - \mu \|_{\text{var}} \leqslant C e^{- \gamma t} \left( 1 + \int_H | x |^2 \lambda(dx) \right), \tag{4} \] provided \(\| f \|_{\varepsilon}^2 \leqslant \delta^0\), where \(\| \cdot \|_{\text{var}}\) is the total variation norm associated to the space \({\mathbb H}_s\) for \(s < 0\). Moreover, for a given \({\mathbb P}_{\lambda}\), \(\mu\) is unique and \({\mathbb P}_{\lambda}\) is a limit of solutions of (3). Due to the lack of uniqueness, it is not straightforward to define a Markov evolution associated to (1). Some recent progress has been obtained in this direction; see \textit{G. Da Prato} and \textit{A. Debussche} [J. Math. Pures Appl. (9) 82, No. 8, 877--947 (2003; Zbl 1109.60047)] and \textit{A. Debussche} and \textit{C. Odasso} [J. Evol. Equ. 6, No. 2, 305--324 (2006; Zbl 1110.35110)], where under similar conditions on \(\varphi\) and \(f\) it is shown that every solution of (1) with Dirichlet boundary condition, being limit of Galerkin approximations, verifies the weak Markov property.