Precise Laplace asymptotics for singular stochastic PDEs: the case of 2D gPAM (Q2127587): Difference between revisions

The authors develop, within the scope of the regularity structures theory, Laplace asymptotics for a family of generalized parabolic Anderson model (gPAM\(_\varepsilon\)) defined on the 2D torus. The solution of the (gPAM\(_\varepsilon\)): \(\hat{u}^\varepsilon\in \mathcal{X}_T:=\{u\in \mathcal{C}([0, T]; \mathcal{C}^\eta(\mathbb{T}^2): u(0,.)=u_0(.)\}, \; T>0\), is defined as the unique limit, as \(\delta \rightarrow 0\), in \(\mathcal{X}_T\) in probability sense on \(\{T^\varepsilon>T\}\) of the family of solutions \((\hat{u}^\varepsilon_{\xi_\delta})_{\delta>0}\) of: \[ \begin{cases} (\partial_t-\Delta)\hat{u}^\varepsilon_{\xi_\delta}= g(\hat{u}^\varepsilon_{\xi_\delta})(\varepsilon\xi_\delta-\varepsilon^2\mathfrak{c}_\delta g'(\hat{u}^\varepsilon_{\xi_\delta})), \\ \hat{u}^\varepsilon_{\xi_\delta}(0, .)=u_0(.), \end{cases} \tag{1} \] where \(\varepsilon \in [0, 1]\), \(\xi\) is a spatial white noise(SWN) with Cameron-Martin space \(\mathcal{H}=L^2(\mathbb{T}^2)\), \(\xi_\delta:=\xi*\rho_\delta\), with \(\rho_\delta:=\delta^{-2}\rho(\delta^{-1})\) and \((\mathfrak{c}_\delta)_{\delta>0}\) is a family of real constants, \(u_0\in \mathcal{C}^\eta(\mathbb{T}^2)\), for a fixed \(\eta \in (1/2, 1)\) and \(g\in \mathcal{C}^{N+7}_b(\mathbb{R}), \, N\geq 0\). Hairer et al. have already applied the regularity structures theory, to prove the well-posedness and the large deviation principle for the (gPAM\(_\varepsilon\)). In the present work, the authors consider the following shifted version of (1): \[ \begin{cases} (\partial_t-\Delta)\hat{u}^\varepsilon_{\xi_\delta; \mathit{h}}= g(\hat{u}^\varepsilon_{\xi_\delta; \mathit{h}})(\varepsilon\xi_\delta+\mathit{h}-\varepsilon^2\mathfrak{c}_\delta g'(\hat{u}^\varepsilon_{\xi_\delta; \mathit{h}})),\\ \hat{u}^\varepsilon_{\xi_\delta; \mathit{h}}(0)=u_0, \end{cases} \tag{2} \] where \(\mathit{h}\in \mathcal{H}\). The authors set, by analogy to the Martin-Hairer's result, \(\hat{u}^\varepsilon_h:=\lim_{\delta\rightarrow 0}\hat{u}^\varepsilon_{\xi_\delta; h}\), (in probability- in \(\mathcal{X}_T\)) and they assume that: \begin{itemize} \item[(H1)] A given functional \(F\) is continuous and bounded from \(\mathcal{X}_T\) to \(\mathbb{R}\) (\(F\in \mathcal{C}_b(\mathcal{X}_T ;\mathbb{R})\)). \item[(H2)] The functional \(\mathfrak{F}:= (F\circ \Phi\circ \mathfrak{L})+ \mathfrak{I}: \mathcal{H}\rightarrow (-\infty, +\infty] \) attains its unique minimum at \(\mathbf{h} \in \mathcal{H}\). Here: \(\mathfrak{I}(h)= \frac12\|h\|^2_{\mathcal{H}}1_{h\in \mathcal{H}}+\infty 1_{h\notin \mathcal{H}}\) is the Schilder's rate function, \(\Phi: \mathcal{M}\rightarrow \mathcal{X}_T\) is the solution map, \(\mathcal{M}\) is the model space and \(\mathfrak{L}\) is the canonical lift. \item[(H3)] The functional \(F\) is \(N+3\)-Fréchet differentiable in a neighborhood \(\mathcal{N}\) of \((\Phi\circ \mathfrak{L})(\mathbf{h})\) and there exist \(M_1, \dots, M_{N+3}\) positive numbers such that for all \(v\in \mathcal{N}, \; y\in \mathcal{X}_T\), \[ |D^{(k)}F(v)[y, \dots, y]|\leq M_k\|y\|^k_{\mathcal{X}_T}, \; k=1, \dots, N+3. \] \item[(H4)] The minimiser \(\mathbf{h}\) is non-degenerate, i.e., \(D^2 \mathfrak{F}\mid_{\mathbf{h}}[\mathit{h}, \mathit{h}]>0, \;\; \text{ for all } \mathit{h}\in \mathcal{H}\setminus \{0\}\). \end{itemize} They prove that under the hypothesis (H1)--(H4) and for small \(\varepsilon\) \((\varepsilon \searrow 0)\), the following Laplace asymptotics holds: \[ J(\varepsilon) := \mathbb{E}\left[1_{T^\varepsilon>T} e^{-\varepsilon^{-2}F(\hat{u}^\varepsilon)}\right]= e^{-\varepsilon^{-2}\mathfrak{F}(\mathbf{h})}\left[\sum_{m=0}^{N}a_m\varepsilon^m + o(\varepsilon^N)\right], \] where \(a_m=\mathbb{E}\big[e^{-\frac12 \hat{Q}_\mathbf{h}}W_m\big]\),\(W_0=1,\; W_m:=\sum_{k=1}^N\frac{(-1)^k}{k!}\sum_{\pi\in \mathcal{G}_=(k, m)}\Pi_{i=1}^k\frac{\hat{F}^{(\pi(i))}_\mathbf{h}}{\pi(i)!}, \; m\in \{1, \dots, N\}\), for \(k\in \{1, \dots, m\}\), \(\mathcal{G}_=(k, m):=\{\text{maps}\; \pi:\{1, \dots, k\}\rightarrow \{3, \dots, m+2\},\; s.t.\; \sum_{i=1}^k(\pi(i)-2)=m\}\), \(\hat{F}^{(j)}_\mathbf{h}= \partial_\varepsilon^j\setminus_{\varepsilon=0} F(\hat{u}^\varepsilon_{\mathbf{h}})\) and \(\hat{Q}_\mathbf{h}:=\hat{F}^{(2)}_\mathbf{h}\). The authors declare that the above result remains true for the black-box, invented by Hairer, for a class of singular stochastic partial differential equations and they consider the Taylor expansion used to prove the main result as a second contribution. This last is formulated, for \(h\in \mathcal{H}, g\in \mathcal{C}_b^{l+4}, l\geq 1\) as: \(\text{ for all } T>0, \text{ exists } \varepsilon_0>0, \; s.t.\; \hat{u}_h^\bullet: [0, \varepsilon_0) \ni \varepsilon \mapsto \hat{u}_h^\varepsilon \in \mathcal{X}_T \) is \(l\) times Fréchet differentiable and \(\hat{u}_h^\varepsilon= w_h+ \sum_{m=1}^{l-1}\frac{\varepsilon^m}{m!}\hat{u}_h^{(m)}+\hat{R}^{(l)}_{h,\varepsilon}\), where \(\hat{u}_h^{(m)}:= \partial_\varepsilon^m\setminus_{\varepsilon=0}\hat{u}^\varepsilon_{h}\), \(\hat{R}^{(l)}_{h,\varepsilon}:=\int_0^1\frac{(1-s)^{l-1}}{(l-1)!}\partial_r^l\setminus_{r=s\varepsilon}\hat{u}^r_{h} ds\) and \(w_h\) is the unique solution of the deterministic initial problem: \((\partial_t-\Delta) w_h= g(w_h)h, \; w_h(0, .)=u_0(.) \). Moreover, the terms \(u_h^{(m)}(\mathbf{Z})\) and the remainder \(R^{(l)}_{h,\varepsilon}(\mathbf{Z})\) are continuous functions of the model \(\mathbf{Z}\) and satisfies some estimates, \(u_h^{(m)}(\mathbf{Z})\) are m-homogeneous w.r.t. dilatation in the model \(\mathbf{Z}\) and \(\hat{u}_h^{(m)} =\lim_{\delta\rightarrow 0} \hat{u}_{\xi_\delta, h}^{(m)}\), (the limit is considered in the sense mentioned above). The main tools of the proofs are based on the skeleton of the classical proofs and the estimation of the terms and the reminder of the Taylor expansion upon the regularity structures theory. The paper is self-contained via rich appendices.