Concave elliptic equations and generalized Khovanskii-Teissier inequalities (Q2031899)

The Khovanskii-Teissier inequality for real \((1,1)\)-classes on a compact Kähler manifold \(X\), which can be viewed as a generalisation of the Hodge index theorem, is a consequence of the \textit{concavity} of the volume function \(\mathcal K \ni \alpha \mapsto \operatorname{Vol}(\alpha)^{\frac{1}{n}} =\left(\int_X \alpha^n \right)^{\frac 1n}\) on the Kähler cone \(\mathcal K \subset H^{1,1}(X,\mathbb R)\) (obtained by replacing \(\alpha\) by \(\alpha +t\beta\) for \(\beta \in H^{1,1}(X,\mathbb R)\) and \(t\) sufficiently close to \(0\) and considering the \(2^{\text{nd}}\) derivative with respect to \(t\)). On a compact Hermitian manifold \((X,\omega)\) of complex dimension \(n\), by identifying any (smooth) real \((1,1)\)-form \(\alpha\) with its \(n\)-tuple of eigenvalues \(\lambda(\alpha) := (\lambda_1(\alpha), \dots, \lambda_n(\alpha)) \in \mathcal C^0(X, \mathbb R^n)\) with respect to \(\omega\), a smooth, symmetric, concave function \(f \colon \Gamma \to \mathbb R\) on an open set \(\Gamma \subset \mathbb R^n\) which is strictly increasing in each variable gives rise to a concave elliptic operator \(F\), also denoted by \((f,\Gamma)\), on the set of real \((1,1)\)-forms \(\alpha\) such that \(\lambda(\alpha) \in \Gamma\). The paper under review generalises the Khovanskii-Teissier inequality by replacing \(\operatorname{Vol}^{\frac 1n}\) by some concave elliptic operators \((f,\Gamma)\) on some subspaces of Bott-Chern \((1,1)\)-classes (or more generally \(W\)-classes defined below) on some compact Hermitian manifolds \((X,\omega)\). Indeed, if \(c \in \mathbb R\) is a constant and \(\alpha\) is a smooth real \((1,1)\)-form with \(\lambda(\alpha) \in \Gamma\) on \(X\) such that \(F(\alpha) = f(\lambda(\alpha)) = c\), it can be shown that \(F\) is constant on the Bott-Chern class of \(\alpha\) (assuming that \(d\alpha=0\)) or more generally on the \(W\)-class of \(\alpha\) given by \[ [\alpha]_W := \left\{ \alpha +\sqrt{-1} \partial\overline\partial\varphi +W(d\varphi) \: : \: \varphi \in \mathcal C^\infty(X,\mathbb R) \right\} \; , \] where \(W \colon \Lambda^{1}_{\mathbb R} \to \Lambda^{1,1}_{\mathbb R}\) is some given smooth linear map from real \(1\)-forms to real \((1,1)\)-forms. Letting \(\operatorname{Sol}_\omega(f,\Gamma,W)\) be the set of \(W\)-classes \([\alpha]_W\) such that there is an \(\widetilde\alpha \in [\alpha]_W\) which satisfies \(\lambda(\widetilde\alpha) \in \Gamma\) and \(F(\widetilde\alpha) = c\) for some constant \(c \in \mathbb R\), it is shown in this article that \(\operatorname{Sol}_\omega(f,\Gamma,W)\) is open in the space of \(W\)-classes (under the topology induced from the \(\mathcal C^4\) topology on \(\Lambda^{1,1}_{\mathbb R}\)) and the map \[ \operatorname{Sol}_\omega(f,\Gamma,W) \ni [\alpha]_W \; \mapsto \; F(\widetilde\alpha) = c(\omega,[\alpha]_W) \in \mathbb R \] is well defined and \textit{concave} on \(\operatorname{Sol}_\omega(f,\Gamma,W)\). The generalised Khovanskii-Teissier inequality for \(W\)-classes can then be obtained, provided that \(\operatorname{Sol}_\omega(f,\Gamma,W)\) is non-empty. Let \(\sigma_\ell(\lambda) = \sigma_{\ell}(\lambda_1, \dots, \lambda_n)\) be the \(\ell\)-th elementary symmetric polynomial in the variables \(\lambda_1,\dots,\lambda_n\) with the convention that \(\sigma_0 \equiv 1\). The author applies the above result to the following families of concave elliptic operators \((f,\Gamma)\) which arise in various works: \begin{itemize} \item Hessian (Quotient) operators: For \(0 \leq \ell < k \leq n\), \[ f(\lambda) := \left(\frac{\binom{n}{k}^{-1} \sigma_k(\lambda)} {\binom{n}{\ell}^{-1} \sigma_\ell(\lambda)} \right)^{\frac{1}{k-\ell}} \; , \quad \Gamma := \Gamma_k := \left\{ \lambda \in \mathbb R^n \: : \: \sigma_j(\lambda) > 0 \;\text{ for all }\; 1\leq j\leq k \right\} \; ; \] \item Lagrangian phase operator: \[ f(\lambda) := \Theta(\lambda) :=\sum_{i=1}^n \arctan(\lambda_i) \; , \quad \Gamma := \left\{ \lambda \in \mathbb R^n \: : \: \Theta(\lambda) \geq (n-1)\frac{\pi}{2} \right\} \; ; \] and \item \((n-1,n-1)\) \(k\)-Hessian operators: \[ f(\lambda) := \left(\binom{n}{k}^{-1} \sigma_k(P(\lambda)) \right)^{\frac{1}{k}} \; , \quad \Gamma := P^{-1}(\Gamma_k), \] where \(P \colon \mathbb R^n \to \mathbb R^n\) is given by \(P(\lambda) = (\mu_1,\dots, \mu_n)\) with \(\mu_i := \frac{1}{n-1} \sum_{j \neq i} \lambda_j\). \end{itemize} In the case where \(W=0\), the author also discusses a sufficient condition, based on the study of the above examples, such that the values \(c(\omega,[\alpha])\) depend only on the cohomology class \([\omega]\).