Fekete points and convergence towards equilibrium measures on complex manifolds (Q666288): Difference between revisions

Let \(L\) be a holomorphic line bundle on a compact complex manifold \(X\) of complex dimension \(n\). Let \((K,\phi)\) be a weighted compact subset, i.e., \(K\) is a non-pluripolar compact subset \(K\) of \(X\) and \(\phi\) is a continuous Hermitian metric on \(L\). Let \(\mu\) be a probability measure on \(K\). If \(s\) is a section of \(kL=L^{\otimes k}\), we denote the corresponding pointwise length function by \[ |s|_{k\phi} = |s| e^{-k\phi}. \] It is then an interesting and intensively discussed problem to study the asymptotic behavior of the spaces of global sections \(s\in H^0=(X,kL)\) endowed with either the \(L^2\)-norm \[ \|s\|^2_{L^2(\mu,k\phi)} := \int_X |s|^2_{k\phi} d\mu \] or the \(L^\infty\)-norm \[ \|s\|_{L^\infty(K,k\phi)} := \sup_K |s|_{k\phi}. \] We introduce the equilibrium weight of \((K,\phi)\) as \[ \phi_K := \sup \big\{\psi\text{ psh weight on }L: \psi \leq \phi\text{ on }K\big\}, \] and denote by \(\phi_K^*\) its upper semi-continuous regularization which is a psh weight on \(L\) since \(K\) is non-pluripolar. The volume \(\mathrm{vol}(L)\) of \(L\) is defined by \[ N_k := \dim H^0(X,kL) = \mathrm{vol}(L) \frac{k^n}{n!} + o(k^n). \] Assume that \(L\) is a big line bundle, i.e., assume that \(\mathrm{vol}(L)>0\). Then the equilibrium measure of \((K,\phi)\) is defined as the Monge-Ampère measure of \(\phi_K^*\) normalized to unit mass: \[ \mu_{eq}(K,\phi):=\frac{1}{\mathrm{vol}(L)} (dd^c \phi_K^*)^n. \] This makes sense as it was shown by the first two authors in [``Growth of balls of holomorphic sections and energy at equilibrium'', Invent. Math. 181, No. 2, 337--394 (2010; Zbl 1208.32020)] that \[ \mathrm{vol}(L) = \int_X (dd^c \phi_K^*)^n. \] The main goal of the paper under review is to give a general criterion involving spaces of global sections that ensures convergence of probability measures on \(K\) of Bergman-type towards the equilibrium measure \(\mu_{eq}(K,\phi)\). To describe that, we also need the notion of a Fekete configuration, that is a finite subset of points in \(K\) maximizing the interpolation problem. Let \(N:=\dim H^0(L)\) and \(P=(x_1, \dots, x_N)\in K^N\) be a configuration of points in \(K\). Then \(P\) is said to be a Fekete configuration for \((K,\phi)\) if it maximizes the determinant of the evaluation operator \[ \mathrm{ev}_P: H^0(L) \rightarrow \bigoplus_{j=1}^N L_{x_j} \] with respect to a given basis \(s_1, \dots, s_N\) of \(H^0(L)\), i.e., the determinant \[ \big|\det\big(s_i(x_j)\big)\big| e^{-(\phi(x_1) + \dots + \phi(x_N))}. \] This condition does not depend on the choice of the basis \((s_i)\). For each \(P=(x_1, \dots, x_N)\in X^N\) let \[ \delta_P := \frac{1}{N}\sum_{j=1}^N \delta_{x_j} \] be the averaging measure along \(P\). The first main result of the paper under review is an equidistribution theorem for Fekete configurations: Theorem A. Let \((X,L)\) be a compact complex manifold equipped with a big line bundle. Let \(K\) be a non-pluripolar compact subset of \(X\) and \(\phi\) a continuous weight on \(L\). For each \(k\) let \(P_k\in K^{N_k}\) be a Fekete configuration for \((K,k\phi)\). Then the sequence \(P_k\) equidistributes towards the equilibrium measure as \(k\rightarrow \infty\), that is \[ \lim_{k\rightarrow \infty} \delta_{P_k} = \mu_{eq}(K,\phi) \] holds in the weak topology of measures. There is another way to represent \(\mu_{eq}(K,\phi)\) in terms of Bernstein-Markov measures. We first need the notion of the Bergman measure. The distortion between the natural \(L^2\) and \(L^\infty\)-norms on \(H^0(L)\) is locally accounted for by the distortion function \(\rho(\mu,\phi)\) whose value at \(x\in X\) is defined by \[ \rho(\mu,\phi)(x) = \sup_{\|s\|_{L^2(\mu,\phi)}=1} |s(x)|^2_\phi, \] the squared norm of the evaluation operator at \(x\in X\). Then the probability measure \[ \beta(\mu,\phi) := \frac{1}{N} \rho(\mu,\phi)\mu, \] where \(N=\dim H^0(L)\), is called the Bergman measure. The measure \(\mu\) is called Bernstein-Markov if the growth of the distortion between \(L^2(\mu,k\phi)\) and \(L^\infty(K,k\phi)\) norms is subexponential as \(k\rightarrow \infty\), that is \[ \sup_K \rho(\mu,k\phi) = O\big(e^{\epsilon k}\big)\text{ for all } \epsilon>0. \] Now then, the Bergman measures converge to the equilibrium measure: Theorem B. Let \((X,L)\) be a compact complex manifold equipped with a big line bundle. Let \(K\) be a non-pluripolar compact subset of \(X\) and \(\phi\) be a continuous weight on \(L\). Let \(\mu\) be a Bernstein-Markov measure for \((K,\phi)\). Then \[ \lim_{k\rightarrow \infty} \beta(\mu,k\phi) = \mu_{eq}(K,\phi) \] holds in the weak topology of measures. Both Theorem A and B are obtained as special cases of Theorem C below which gives a more general criterion ensuring convergence of Bergman measures to equilibrium in terms of \(\mathcal{L}\)-functionals, first introduced by \textit{S. K. Donaldson} [``Scalar curvature and projective embeddings. II'', Q. J. Math. 56, No. 3, 345--356 (2005; Zbl 1159.32012); ``Some numerical results in complex differential geometry'', Pure Appl. Math. Q. 5, No. 2, 571--618 (2009; Zbl 1178.32018)]. The \(L^2\) and \(L^\infty\) norms on \(H^0(kL)\) are described geometrically by their unit balls, which will be denoted by \[ \mathcal{B}^\infty(K,k\phi) \subset \mathcal{B}^2(\mu,k\phi) \subset H^0(kL). \] Fix a reference weighted compact set \((K_0,\phi_0)\) and a probability measure \(\mu_0\) on \(K_0\) which is Bernstein-Markov with respect to \((K_0,\phi_0)\). Normalize the Haar measure vol on \(H^0(kL)\) by \[ \mathrm{vol}\mathcal{B}^2(K_0,k\phi_0)=1, \] and introduce the \(\mathcal{L}\)-functionals \[ \mathcal{L}_k(\mu,\phi) := \frac{1}{2kN_k} \log\mathrm{vol} \mathcal{B}^2(\mu,k\phi) \] and \[ \mathcal{L}_k(K,\phi) := \frac{1}{2 k N_k} \log\mathrm{vol} \mathcal{B}^\infty(K,k\phi). \] For a psh function with minimal singularities \(\psi\), let \(\mathcal{E}(\psi)\) be the Monge-Ampère energy, characterized as the primitive of the Monge-Ampère operator \[ \frac{d}{dt}|_{t=0^+} \mathcal{E}(t\psi_1 + (1-t)\psi_2)= \int_X (\psi_1 -\psi_2) (dd^c\psi_2)^n \] normalized by \(\mathcal{E}(\phi^*_{0,K_0})\). Then the energy at equilibrium of \((K,\phi)\) is \[ \mathcal{E}_{eq}(K,\phi):= \frac{1}{\mathrm{vol}(L)} \mathcal{E}(\phi_K^*) \] and by Theorem A in [the first two authors, loc. cit.] we have: \[ \lim_{k\rightarrow \infty} \mathcal{L}_k(K,\phi) = \mathcal{E}_{eq}(K,\phi). \] If \(\mu\) satisfies the Bernstein-Markov property described above, then also \[ \lim_{k\rightarrow \infty} \mathcal{L}_k(\mu,\phi) = \mathcal{E}_{eq}(K,\phi). \] The key result (implying Theorem A and Theorem B) of the paper under review is as follows: Theorem C. Let \((\mu_k)\) be a sequence of probability measures on \(K\) such that \[ \lim_{k\rightarrow \infty} \mathcal{L}_k(\mu_k,\phi) = \mathcal{E}_{eq}(K,\phi). \] Then the associated Bergman measures satisfy \[ \lim_{k\rightarrow \infty} \beta(\mu_k,k\phi) = \mu_{eq}(K,\phi) \] in the weak topology of measures. This pretty general statement comprises a lot of classical and more special results and yields some nice applications to interpolation. We refer to the introduction of the paper under review for more details.