Large \(N\) limit of the \(O(N)\) linear sigma model via stochastic quantization (Q2119209)

The study concerns the \(N\) system of stochastic partial differential equations on \(\mathbb{T}^2\) known as \(O(N)\) linear sigma model and given for \(i=1,\dots, N\) by \[ \begin{cases} (\partial_t-\Delta +m)\Phi^N_i=-\frac{1}{N}\sum_{j=1}^N(\Phi^N_j)^2 \Phi^N_i+\xi_i,\\ \qquad\qquad\;\,\Phi^N_i(0)=\phi^N_i, \end{cases}\tag{1} \] where \(N\in \mathbb{N}\), \(m\geq 0\) fixed, \((\xi_i)_{i=1}^N\) is a finite sequence of independent space-time white noises defined on a probability basis, \((\phi^N_i)_{i=1}^N\) are random initial data independent of \((\xi_i)_{i=1}^N\), the product in the nonlinear term is understood as a Wick type, canonical or renormalized and it is denoted below by \(:\cdot :\) (see (\(\ast\)) below). The first main result is the convergence in probability in the space \(C([0,T], C^{-1}(\mathbb{T}^2)), \;\; \text{ for all } \; T>0\), of \( \Phi_i^N, \;\;\text{ for all } \; i,\) as \(N\rightarrow \infty\), to the unique solution \(\Psi_i\) of the mean field SPDE: \[ \begin{cases} (\partial_t-\Delta +m)\Psi_i=-\mathbb{E}[(\Psi_i)^2-Z_i^2] \Psi_i+\xi_i\\ \qquad\qquad\;\,\Psi_i(0)=\psi_i, \end{cases}\tag{2} \] with \((\psi_i)_{i=1}^N\) are random initial data and \((Z_i)_{i=1}^N\) are the stationary solutions for: \[ (\partial_t-\Delta +m)Z_i=\xi_i, \;\;\; Z_i(0)=z_i. \tag{3} \] The assumptions imposed are : \begin{itemize} \item The initial data \((\phi_i^N, \psi_i )_{i=1}^N\) are finite moment random variables with components in Besov space \( C^{-\kappa}\), for some small (but without giving a quantified upper bound) \( \kappa>0\), \item \((\psi_i )_{i=1}^N\) are i.i.d, \item \(\text{ for all } p>1, i=1,\ldots,N, \phi_i^N\) converges in \( L^p(\Omega, C^{-\kappa})\) to \( \psi_i\), \item \(\frac{1}{N}\sum_{j=1}^N||\phi_i^N- \psi_i||^p_{C^{-\kappa}}\) converges in probability to zero. \end{itemize} If moreover \((\phi_i^N, \psi_i )_{i=1}^N\) are exchangeable then \[ \lim_{N\rightarrow \infty}\mathbb{E}||\Phi_i^N(t)- \Psi_i(t)||^2_{L^2(\mathbb{T}^2)} =0, \;\;\; \text{ for all } t>0.\] The solutions are defined via the decomposition: \( \Psi_i = Z_i+X_i\) and \( \Phi_i^N = Z_i^N+Y_i^N\), with \(X_i\) and \(Y_i^N\) solving: \[ (\partial_t-\Delta +m)X_i= -(\mathbb{E}[X_i^2]+2\mathbb{E}[X_iZ_i]+\mathbb{E}[:Z_i^2:])(X_i+Z_i), \;\;\; X_i(0)=\eta_i, \tag{3}\] \[\left\{ \begin{array}{lr} (\partial_t-\Delta +m)Y_i= -\frac{1}{N}\sum_{j=1}^N(Y_j^2Y_i + Y_j^2Z_i +2Y_jY_iZ_j+ 2Y_j:Z_iZ_j:+ :Z_j^2:Y_i+:Z_iZ_j^2:),\nonumber\\ Y_i(0)=y_i, \end{array} \right.\tag{4} \] The renormalized Wick product is defined as: \[ \begin{cases} :Z_iZ_j: = :\tilde{Z}_i\tilde{Z}_j: + S_t \tilde{z}_i\tilde{Z}_j+ S_t \tilde{z}_j\tilde{Z}_i+ S_t \tilde{z}_iS_t \tilde{z}_j \\ :Z_iZ^2_j:= :\tilde{Z}_i\tilde{Z}^2_j: + S_t \tilde{z}_i:\tilde{Z}^2_j:+ 2S_t \tilde{z}_j:\tilde{Z}_i\tilde{Z}_j:+ 2S_t \tilde{z}_iS_t \tilde{z}_j\tilde{Z}_j+ (S_t \tilde{z}_j)^2\tilde{Z}_i+ S_t \tilde{z}_i(S_t \tilde{z}_j)^2, \end{cases}\tag{\(\ast\)} \] where \( Z_i= \tilde{Z}_i+ S_t \tilde{z}_i\), \(\tilde{z}_i:= z_i- \tilde{Z}_i(0)\), \((S_t)_t\) is the heat semigroup, \(:\tilde{Z}_i\tilde{Z}_j:\) and \(:\tilde{Z}_i\tilde{Z}_j\tilde{Z}_k:\) in the left hand sides are the canonical Wick product defined as the limit, in \(C_TC^{-\kappa}\) of the corresponding products of the stationary solution of the linear SPDE: \((\partial_t-\Delta +m)\tilde{Z}_{i, \varepsilon}=\xi_{i, \varepsilon}\), with \(\xi_{i, \varepsilon}\) being a space-time mollification of \(\xi_i\). The authors prove the existence and the uniqueness of the solution \(\Psi_i\) under the additional conditions that \(\mathbb{E}||\eta_i||^4_{L^4}<\infty\) and \(\mathbb{E}||z_i||^p_{C^{-\kappa}}<\infty\). The solution satisfies, for \(\beta>3\kappa>0\) small enough (but again without giving a quantified upper bound) \[\mathbb{E}[\sup_{t\in[0, T]}t^{\beta+1/2}||\Psi_i- Z_i||^2_{C^\beta}]<\infty. \] The second main result is that the family \((\nu^{N}_i)_N\) of invariant measures for (1) is tight on \(C^{-\kappa}, \kappa >0\) and there exists \(C_0>0,\) such that \[ \text{ for all } \;\; m\geq C_0 (\mathbb{E}||Z_1||^{\frac{2}{2-s}}_{C^{-s}}+ \mathbb{E}||:Z_1:||^{\frac{2}{2-s}}_{C^{-s}}+\mathbb{E}||:Z_1Z_2:||^2_{C^{-s}}+1 ),\] with \( s\in [2\kappa, 1/4)\), there exists \(C>0\), for which the following estimate holds: \[ \mathbb{W}(\nu^{N}_i, \mathcal{N}(0, \frac12(m-\Delta)^{-1})\leq CN^{-\frac12},\] where \( \mathcal{N}(0, \frac12(m-\Delta)^{-1})\) is the law of the Gaussian free field, \(\mathbb{W}\) is the \(C^{-\kappa}-\)Wasserstein distance. The marginal law of the first \(k\) components; \(\nu^{N, k}\) converges to \(\mathcal{N}(0, \frac12(m-\Delta)^{-1})\times\dots \times \mathcal{N}(0, \frac12(m-\Delta)^{-1})\). Moreover, there exists \(C_1>0\), such that \[ \text{ for all } \;\; m\geq C_1 (\mathbb{E}||:Z_2Z_1:||^2_{C^{-s}}+ (\mathbb{E}||Z_1||^2_{C^{-s}})^{\frac{1}{1-s}}+1),\] the law of the Gaussian free field is the unique invariant measure to Equation (2), supported on \(C^{-\kappa}\), with \(X_i\) satisfying: \[(\partial_t-\Delta +m)X_i= -\mathbb{E}[X_i^2+2X_iZ_i](X_i+Z_i), \;\;\; X_i(0)=\Psi_i(0)-Z_i(0). \tag{5}\] In the third part, the interest is on observables defined by: \[ \begin{aligned} \frac{1}{N}:(\sum_{i=1}^N \Phi_i^2)^2:&= \frac{1}{N}\sum_{i, j=1}^N (Y_i^2Y_j^2+ 4 Y_i^2Y_jZ_j+2 Y_i^2:Z_i^2:\\ &+ :Z_i^2Z_j^2: +4 Y_i: Z_iZ_j^2: +4 Y_iY_j:Z_iZ_j:),\\ \frac{1}{\sqrt{N}}\sum_{i=1}^N :\Phi_i^2:& = \frac{1}{\sqrt{N}}\sum_{i=1}^N (Y_i^2+ 2 Y_iZ_i+:Z_i^2:). \end{aligned} \] Here, the couple \((Y_i, Z_i) \) is stationary and the Wick product is canonical. The main result is that there exists \(m_0\), such that \(\text{ for all }\;\; m\geq m_0\), the laws of \( \frac{1}{\sqrt{N}}\sum_{i=1}^N :\Phi_i^2:\) respectively of \( \frac{1}{N}:(\sum_{i=1}^N \Phi_i)^2:\) are tight on \( B_{2,2}^{-2\kappa}\) respectively \( B_{1,1}^{-3\kappa}\) and \[ \lim_{N\rightarrow \infty} \hat{G}_N=2\hat{T^2}/(1+2\hat{T^2}),\] \[ \lim_{N\rightarrow \infty}\lim_{\varepsilon\rightarrow 0}\mathbb{E}\frac{1}{N}\langle :(\sum_{i=1}^N \Phi_i^2)^2:, \rho_x^\varepsilon:\rangle= -4\sum_{k\in \mathbb{Z}^2} \hat{T^2}(k)^2/(1+2\hat{T^2}(k)), \] where \(G_n(x-z):=\mathbb{E}[ \frac{1}{\sqrt{N}} :\sum_{i=1}^N \Phi_i^2:(x) \frac{1}{\sqrt{N}} :(\sum_{i=1}^N \Phi_i^2):(z)]\), the hat represents the Fourier transform, \(\rho_x^\varepsilon\) is a mollifier and \(T=\frac12(m-\Delta)^{-1}\). The usual techniques are used to prove the results but with relevant boundedness of the quantities and the norms. For example, the existence and the uniqueness of \( \Psi_i = Z_i+X_i\) yields from the existence and uniqueness \(X_i\) given \(Z_i\). A local mild solution \( X_i\in L^2(\Omega; C_{T^*}L^4\cap C((0, T^*];C^\beta)\) has been proved than thanks to the uniform boundedness of \(\mathbb{E}[\sup_{t\in[0, T]}t^{\beta+1/2}||X_i||^2_{C^\beta}]\), it becomes global. From the computations, one can deduce that at least \(\beta\) and \(\kappa\) have to satisfy \(\beta< 2/3-\kappa\) and \(\kappa<1/6\).