Convexity of the extended K-energy and the large time behaviour of the weak Calabi flow

TL;DR

This paper proves the convexity of the extended K-energy on a metric completion of K"ahler potentials, analyzes the long-term behavior of the weak Calabi flow, and explores conditions for convergence or divergence on K"ahler manifolds.

Contribution

It establishes convexity of the K-energy on $ ext{E}^p$, studies the asymptotics of the weak Calabi flow in a CAT(0) space, and provides new insights into flow convergence and destabilizing rays.

Findings

The extended K-energy is convex and lower semi-continuous on $ ext{E}^p$.

The weak Calabi flow either diverges or converges to a minimizer of the K-energy.

The flow's long-term behavior confirms a conjecture of Donaldson in certain cases.

Abstract

Let $(X,\omega)$ be a compact connected K\"ahler manifold and denote by $(\mathcal E^p,d_p)$ the metric completion of the space of K\"ahler potentials $\mathcal H_\omega$ with respect to the $L^p$-type path length metric $d_p$. First, we show that the natural analytic extension of the (twisted) Mabuchi K-energy to $\mathcal E^p$ is a $d_p$-lsc functional that is convex along finite energy geodesics. Second, following the program of J. Streets, we use this to study the asymptotics of the weak (twisted) Calabi flow inside the CAT(0) metric space $(\mathcal E^2,d_2)$. This flow exists for all times and coincides with the usual smooth (twisted) Calabi flow whenever the latter exists. We show that the weak (twisted) Calabi flow either diverges with respect to the $d_2$-metric or it $d_1$-converges to some minimizer of the K-energy inside $\mathcal E^2$. This gives the first concrete result about the long time convergence of this flow on general K\"ahler manifolds, partially confirming a conjecture of Donaldson. Finally, we investigate the possibility of constructing destabilizing geodesic rays asymptotic to diverging weak (twisted) Calabi trajectories, and give a result in the case when the twisting form is K\"ahler. If the twisting form is only smooth, we reduce this problem to a conjecture on the regularity of minimizers of the K-energy on $\mathcal E^1$, known to hold in case of Fano manifolds.