Optimal test-configurations for toric varieties (Q998662)

In Geometric Invariant Theory, there is a notion of optimal one-parameter destabilising subgroup for a non semi-stable point. Thus, it is natural to wonder if this notion admits an analog for the notion of K-stability on a polarized manifold. In this paper, the author explores this analog for toric manifolds, proving the existence of an optimal destabilising convex function and relating it to the Calabi functional. Recall that to a compact polarized toric variety \((X,L)\) corresponds a polytope \(P\subset \mathbb{R}^n\) which is equipped with a canonical measure \(d\sigma\) on the boundary \(\partial P\). S.K. Donaldson defined a functional \(\mathcal{L}\) on the space \(C_1\) of continuous convex functions on \(P^*\) (the union of \(P\) and the interiors of its codimension one faces) integrable on \(\partial P\): \[ \mathcal{L}(f)= \int_{\partial P} f d\sigma - \hat{S} \int_P f d\mu \] with \(d\mu\) the Lebesgue measure and \(\hat{S}\) the average of the scalar curvature in \(c_1(L)\). Donaldson proved that, given a rational piecewise linear convex function \(f\) on \(P\), one can define a test-configuration for \((X,L)\) to check its K-stability. Then \(\mathcal{L}(f)\) is essentially the Futaki invariant of the test-configuration. We refer to [J. Differ. Geom. 62, No. 2, 289--349 (2002; Zbl 1074.53059)] for details. Using the compactness theorem for convex functions in \(C_1\), the author shows that there exists a convex minimiser \(\Phi\in C_1\cap L^2(P)\) for the functional \[ f\mapsto\frac{\mathcal{L}(f)}{\| f\|_{L^2}}. \] This minimizer is unique up to scaling. It is not clear that \(\Phi\) can be seen as a a test-configuration since it may not be piecewise linear, but this should be considered at least as a limit of test-configurations. This is the ``optimal destalibilizing test-configuration''. Set \[ \mathcal{L}_h(f)= \int_{\partial P} f d\sigma - \int_P h \times f d\mu \] Now, by fixing \(\Phi\) with \(\mathcal{L}(\Phi)=-\| \Phi \|_{L^2}^2\), the function \(\hat{S}- \Phi\) is the unique minimizer of the \(L^2\) norm for functions in the set \[ E=\{ h \in L^2 | \mathcal{L}_h(f)\geq 0 \text{ for all convex functions } f\}, \] which can be understood as the closure of all possible constant scalar curvature functions when one is varying the (torus invariant) Kähler metric in \(c_1(L)\). Hence, one obtains a nice relationship: \[ \inf_{h\in E} \| h - \hat{S} \|_{L^2} = \sup_{f \text{ convex} } \frac{- \mathcal{L}(f) }{\| f \|_{L^2}}. \] On another hand, if the optimal destabilizing test-configuration is a piecewise linear, then the author gets a canonical decomposition of the polytope into semi-stable pieces, that are given by the maximal subpolytopes on which \(\Phi\) is linear. Another key ingredient that plays a role in this setting is the Calabi flow. The author shows that if the flow exists for all time on the toric manifold, then it minimizes the Calabi functional. This infimum is precisely given by \(\| \Phi \|_{L^2}\), that is the supremum of the (normalized) Futaki invariants over all destabilizing test-configurations. This implies a conjecture of Donaldson under the assumption of the existence of the Calabi flow. The paper is very well written and enlightens the problem of K-stability for toric manifolds.