Large deviations and averaging for stochastic tamed 3D Navier-Stokes equations with fast oscillations (Q2155176)

Let \(\mathbb D\) denote the three-dimensional torus, \(H^m\) the Sobolev space on \(\mathbb D\), \(\mathbb H^m\) its subspace of divergence free elements, \(\mathcal P\) the orthogonal projection from \(H^0\) onto \(\mathbb H^0\), \(A\mathbf u=\mathcal P\Delta\mathbf u\), \(B(\mathbf u)=\mathcal P[(\mathbf u\cdot\nabla)\mathbf u]\) and let \(f:\mathbb H^1\times H^0\to\mathbb H^1\), \(h:\mathbb H^1\times H^0\to H^0\) be mappings that satisfy \begin{align*} \|f(\mathbf u_0,\mathbf v_0)-f(\mathbf u_1,\mathbf v_1)\|_{H^1}&\le C_0\|\mathbf u_0-\mathbf u_1\|_{H^1}+C_1\|\mathbf v_0-\mathbf v_1\|_{H^0}, \\ \|h(\mathbf u_0,\mathbf v_0)-h(\mathbf u_1,\mathbf v_1)\|_{H^0}&\le C_2\|\mathbf u_0-\mathbf u_1\|_{H^1}+C_3\|\mathbf v_0-\mathbf v_1\|_{H^0}, \end{align*} where \(C_3\) is strictly smaller than the least eigenvalue of \(-\Delta\). Let also \(g_N:\mathbb R_+\to\mathbb R_+\) be a smooth function such that \(g_N(r)=0\) on \([0,N]\), \(g_N(r)=r-N\) on \([N+1,\infty)\) and \(0\le g_N^\prime\le C\) on \(\mathbb R_+\), let \(U\) be a Hilbert space, \(\sigma_1\in L_2(U,\mathbb H^1)\), \(\sigma_2\in L_2(U,H^0)\), where \(L_2\) denotes the space of Hilbert-Schmidt operators and let \(W\) be a cylindrical Wiener process on \(U\). Under the above assumptions, given an initial condition \((\mathbf u_0,\mathbf v_0)\in\mathbb H^1\times H^0\), there exists a strong solution (both in the PDE and the probabilistic sense) of the equations \begin{align*} d\mathbf u&=[A\mathbf u-B(\mathbf u)-\mathcal P(g_N(|\mathbf u|^2)\mathbf u)+f(\mathbf u,\mathbf v)]\,dt+\sqrt\varepsilon\sigma_1\,dW, \\ d\mathbf v&=\frac 1\delta[\Delta\mathbf v+h(\mathbf u,\mathbf v)]\,dt+\frac{1}{\sqrt\delta}\sigma_2\,dW \end{align*} for every \(\delta>0\) and \(\varepsilon>0\) that we denote by \((\mathbf u_{\varepsilon,\delta},\mathbf v_{\varepsilon,\delta})\) and which satisfies \[ \lim_{\delta\to 0+}\mathbb E\left[\sup_{t\in[0,T]}\|\mathbf u_{1,\delta}(t)-\mathbf u(t)\|^p_{H^1}\right]=0 \] for every \(p>0\) and \(T>0\) where \(\mathbf u\) is the solution of the equation \[ d\mathbf u=[A\mathbf u-B(\mathbf u)-\mathcal P(g_N(|\mathbf u|^2)\mathbf u)+\bar f(\mathbf u)]\,dt+\sigma_1\,dW \] for \[ \bar f(u)=\int_{H^0}f(u,\cdot)\,d\mu^u,\qquad u\in\mathbb H^1. \] Here \(\mu^u\) is the unique invariant probability measure of the equation \begin{align*} dv&=[\Delta v+h(u,v)]\,dt+\sigma_2\,dW ,\\ v(0)&=v\in H^0,\quad u\in\mathbb H^1. \end{align*} Secondly, it is proved that the family \(\{\mathbf u_{\varepsilon,\delta_\varepsilon}\}\) satisfies the large deviation principle on \(C([0,T];\mathbb H^1)\cap L^2(0,T;\mathbb H^2)\) as \(\varepsilon\to 0+\) provided that \[ \lim_{\varepsilon\to 0+}\frac{\delta_\varepsilon}\varepsilon=0 \] for the good rate function \[ I(w)=\inf\left\{\frac 12\int_0^T\|\phi(s)\|_U^2\,ds\right\}, \] where the infimum is considered over the set \[ \left\{\phi\in L^2(0,T;U):\,w=\bar{\mathbf{u}}_\phi\right\} \] and \(\bar{\mathbf{u}}_\phi\) is the solution of the equation \begin{align*} \partial_tu&=A\mathbf u-B(\mathbf u)-\mathcal P(g_N(|\mathbf u|^2)\mathbf u)+\bar f(\mathbf u)+\sigma_1\phi, \\ \mathbf u(0)&=u. \end{align*} The existence and uniqueness of solutions to this last equation is also proved in the paper.