Large deviations for Gibbs measures with singular Hamiltonians and emergence of Kähler-Einstein metrics (Q2014275): Difference between revisions

Let \(X\) be a compact Riemannian manifold and \(dV\) a volume form on \(X.\) Given a positive integer \(N\), consider the \(N\)-fold product \(X^N\) and a lower symmetric semi-continuous functions \(H^{(N)}\) on \(X^N\) and a constant \(\beta \in (0, \infty)\). The Gibbs measure \(\mu^{(N)}_\beta\) associated to these data is given by \[ \mu^{(N)}_\beta:= e^{-\beta H^{(N)}} d V^{\otimes N}/ Z_{N,\beta}, \] where \(Z_{N,\beta}\) is a normalizing constant so that \(\mu^{(N)}_\beta\) is a probability measure on \(X^N\). By definition, the couple \((X^N, \mu^{(N)}_\beta)\) is a random point process with \(N\) particles on \(X\). From the point of view of statistical mechanics, it models \(N\) identical particles on \(X\) interacting by the Hamiltonian \(H^{(N)}\) in thermal equilibrium at inverse temperature \(\beta\). We write \(x^{(N)}=(x^{(N)}_1,\dots, x^{(N)}_N)\) for a general point in \(X^N.\) A classical problem is to study the conditions under which we have (almost surely) equidistribution of these \(N\) particles as \(N \to \infty\). To be precise, we want to know whether the random empirical measure \[ \delta_N(x^{(N)}):= \frac{1}{N} \sum_{j=1}^N x^{(N)}_j \] converges to a deterministic probability measure on \(X\) as \(N \to \infty\) for almost every sequence \(\mathbf{x}=\big(x^{(N)}\big)_{N \in \mathbb{N}^*}\) in \(\prod_{N \in \mathbb{N}^*} X^N\) with respect to the probability product measure \(\prod_{N \in \mathbb{N}^*}\mu^{(N)}_\beta.\) Another way to formulate this question is to consider the map \(\delta_N\) from \(X^N\) to the space \(\mathcal{M}(X)\) of probability measures on \(X\) defined by \(x^{(N)} \longmapsto \delta_N(x^{(N)})\) as above. We want to study necessary conditions for which the law \(\Gamma_N:= (\delta_N)_* \mu^{(N)}\) of \(\delta_N\) converges to some Dirac mass on \(\mathcal{M}(X).\) The first main result of the paper under review is that if \(H^{(N)}\) satisfies a certain normalizing condition and \(H^{(N)}\) is uniformly super-harmonic, i.e, \(\Delta_{x_1} H^{(N)}(x_1, \dots,x_N) \) is bounded above by a constant on \(X^N\) (in the sense of distributions), then \(\Gamma_N\) satisfies a large deviation principle which in particular implies that \(\Gamma_N\) converges to a unique minimizer \(\mu_\beta\) of a certain functional on \(\mathcal{M}(X)\). The key feature of this theorem is that \(H^{(N)}\) can be singular. The second part of the paper is devoted to an application of the first main result to the study of Kähler-Einstein metrics on a compact Kähler manifold \(X\). It is well known by a celebrated result of Yau that \(X\) admits a Kähler-Einstein metric \(\omega_{KE}\) if and only if the canonical line bundle \(K_X\) is ample. Since the proof of the existence of \(\omega_{KE}\) is not constructive, we don't have much information on explicit approximations of this Kähler-Einstein metric. The author shows that the normalized volume form of the metric in question is the limit of the sequence of the random empirical measures of a suitable random point process and \(\omega_{KE}\) itself can be approximated by a sequence of Kähler forms defined by (quasi-)explicit formulae associated to the last random point process. Let us enter the details now. For any positive integer \(k,\) let \((s_1, \dots, s_{N_{k}})\) be a basis of the vector space \(H^0(X, K_X^k)\) of holomorphic sections of the \(k^{th}\) power \(K_X^k\) of \(K_X\). Observe that \(|s_j|^{2/k}\) is a well-defined volume form on \(X.\) Hence, \[ \det S^{(k)}(x_1, \dots, x_{N_k}):= \big|\det [s_j(x_l)]_{1\leq j, l \leq N_k}\big|^{2/(k)} \] is a well-defined measure on \(X^{N_k}.\) Let \(Z_{N_k}\) be a normalizing constant such that \(\mu^{(N_k)}:= 1/Z_{N_k} \det S^{(k)}\) is a probability measure. The last measure is independent of the choice of the basis of \(H^0(X, K_X^k)\) and is symmetric in \((x_1,\dots, x_{N_k}).\) Hence, \((X^{N_k}, \mu^{(N_k)})\) defines a canonical random point process associated to \(X\) with the Hamiltonians \[ H^{(N_k)}(x_1, \dots,x_{N_k}):= - k^{-1} \log \bigg| \frac{ \det S^{(k)}(x_1, \dots, x_{N_k})}{dV^{\otimes N_k}} \bigg|^2, \] where \(dV:= \omega^{\dim X}\) and \(\omega\) is a fixed Kähler form on \(X. \) The second main result of this paper says that \(\mu^{N_k}\) converges to \(\omega_{KE}^{\dim X}\) and \[ dd^c \log \int_{X^{N_k-1}} \big | \det S^{(k)}(\dot, x_2, \dots, x_{N_k})\big|^{2/k} \to \,\omega_{KE} \] in the sense of currents. The author also studies a more general case where \(K_X\) is replaced by a positive line bundle or any big line bundle. At the end of the paper, he formulates some conjectures around possible approximations of the solution of the complex Monge-Ampère equation by means of random point process.