Uniform \(K\)-stability, Duistermaat-Heckman measures and singularities of pairs (Q1687881): Difference between revisions

Let \(X\) be a complex projective manifold endowed with an ample line bundle \(L\). The pair \((X, L)\) is said to be \(K\)-stable if for every test configuration the Donaldson-Futaki invariant \(\text{DF}(X,L)\) is non-negative and equality holds only if the configuration is (almost) trivial. The relation between the existence of constant scalar curvature Kähler metrics and \(K\)-stability has been studied intensively in the last years, culminating in the work of \textit{X. Chen} et al. [J. Am. Math. Soc. 28, No. 1, 183--197 (2015; Zbl 1312.53096)] for the case where \(L\) is a multiple of \(-K_X\). In this paper the authors give an approach to \(K\)-stability inspired by non-Archimedean geometry. \newline Given an arbitrary polarized scheme \((X, L)\) with a \(\mathbb G_m\)-action, let \(H^0(X,mL)=\bigoplus_{\lambda \in\mathbb Z}H^0(X,mL)_\lambda\) be the weight decomposition. Set \(N_m:=\dim H^0(X,mL)\) for every \(m \in \mathbb N\). The Duistermaat-Heckman measure \[ \text{DH}_{(X,L)}:=\lim_{m\to\infty}\frac{1}{N_m}\sum_{\lambda \in \mathbb Z} \dim H^0(X,mL)_\lambda \delta_{m^{-1}\lambda} \] is a probability measure with compact support in \(\mathbb R\) and the Donaldson-Futaki invariant \(\text{DF}(X,L)\) appears in the subdominant term of the expansion \[ \frac{1}{N_m}\sum_{\lambda\in \mathbb Z}m^{-1}\lambda\dim H^0(X,mL)_\lambda=\int_\mathbb R\lambda\,\text{DH}_{(X,L)}(d\lambda)-(2m)^{-1}\text{DF}(X,L)+O(m^{-2}). \] For a test configuration \((\mathcal X,\mathcal L)\) such that \(\mathcal L\) is relatively ample, the Duistermaat-Heckman measure and Donaldson-Futaki invariant are defined to be those of the central fibre. The first main result of this paper is that the absolutely continuous part of \(\text{DH}(\mathcal X, \mathcal L)\) has piecewise polynomial density, and its singular part is a finite sum of point masses. Moreover \(\text{DH}(\mathcal X, \mathcal L)\) is a finite sum of point masses if and only if \((\mathcal X, \mathcal L)\) is almost trivial. If \((X,L)\) is a polarized variety (i.e. a reduced, irreducible scheme), the measure \(\text{DH}(\mathcal X, \mathcal L)\) is a Dirac mass if and only if for some \(p \in [1, \infty[\) one has \(\|(\mathcal X,\mathcal L)\|_p=0\) (cf.[\textit{S. K. Donaldson}, J. Differ. Geom. 70, No. 3, 453--472 (2005; Zbl 1149.53042)]). Suppose now that \(X\) is a normal, irreducible variety. Then \((X,L)\) is said to be \(L^p\)-uniformly K-stable if \(\text{DH}(\mathcal X, \mathcal L) \geq \delta \|(\mathcal X,\mathcal L)\|_p\) for some uniform constant \(\delta>0\). The authors study \(L^1\)-uniform stability in terms of the non-Archimedean counterparts of classical functionals in Kähler geometry. More precisely, the non-Archimedean Monge-Ampère energy is \[ E^{\text{ NA}}(\mathcal X,\mathcal L)=\frac{(\bar{\mathcal L}^{n+1})}{(n+1)V}=\int_{\mathbb R}\lambda\,\text{DH}_{(\mathcal X,\mathcal L)}(d\lambda), \] where \(V=(L^n)\) and \((\bar{\mathcal X},\bar{\mathcal L})\) is the natural \(\mathbb G_m\)-equivariant compactification of \((\mathcal X,\mathcal L)\) over \(\mathbb P^1\). The non-Archimedean J-energy is \[ J^{\text{ NA}}(\mathcal X,\mathcal L) =\lambda_{\max}-E^{\text{ NA}}(\mathcal X,\mathcal L) =\lambda_{\max}-\int_{\mathbb R} \lambda\,\text{DH}_{(\mathcal X,\mathcal L)}(d\lambda)\geq 0, \] with \(\lambda_{\max}\) the upper bound of the support of \(\text{DH}_{(\mathcal X,\mathcal L)}\). The authors show that \[ c_n J^{\text{ NA}}(\mathcal X,\mathcal L)\leq \|(\mathcal X,\mathcal L)\|_1\leq 2 J^{\text{ NA}}(\mathcal X,\mathcal L) \] for some constant \(c_n>0\). This topic is further developed in the recent paper [the authors, ``Uniform K-stability and asymptotics of energy functionals in Kähler geometry'', Preprint, \url{arXiv:1603.01026}]. \newline The last part of the paper treats the relation between the birational geometry of \(X\) and its test configurations using the language of valuations. Given a boundary divisor \(B\) on \(X\), one defines the non-Archimedean entropy of a test configuration \((\mathcal X , \mathcal L)\) as \[ H^{\text{ NA}}_B(\mathcal X,\mathcal L)=V^{-1}\sum_E A_{(X,B)}(r(\text{ord}_E))(E\cdot\mathcal L^n), \] where the sum runs over the irreducible components of \(\mathcal X_0\) and \(A_{(X,B)}(v)\) denotes the log discrepancy of a divisorial valuation \(v\) with respect the pair \((X,B)\). The authors prove that this functional can be used to characterize the singularities of \(X\): let \(B\) be an effective boundary divisor on \(X\). Then the pair \((X,B)\) is log-canonical (resp.\ klt) if and only if \(H_B^{\text{ NA}}(\mathcal X,\mathcal L)\geq 0\) (resp.\ \(>0\)) for every non-trivial normal test configuration \((\mathcal X,\mathcal L)\) such that \(\mathcal L\) is relatively ample. If \((X,B)\) is klt, there exists a \(\delta>0\) such that \(H_B^{\text{ NA}}(\mathcal X,\mathcal L)\geq \delta J^{\text{ NA}}(\mathcal X,\mathcal L)\) for all test configurations.