Evans-Krylov estimates for a nonconvex Monge-Ampère equation (Q289870): Difference between revisions

Evans-Krylov estimates [\textit{L. C. Evans}, Commun. Pure Appl. Math. 35, 333--363 (1982; Zbl 0469.35022); \textit{N. V. Krylov}, Izv. Akad. Nauk SSSR, Ser. Mat. 46, 487--523 (1982; Zbl 0511.35002)] for certain real and complex twisted Monge-Am\`pere equations and parabolic Monge-Ampère equations are established. In the real cases, let \(U\subset \mathbb{R}^k\times\mathbb{R}^\ell\), with coordinates \(\{x_i\}_{i=1}^k\) and \(\{y_i\}_{i=1}^\ell\) and let \(u\in C^\infty(U)\) be convex in the \(x\) variables and concave in the \(y\) variables. Then the real equations treated in this paper are \(F(u)=0\) and \(H(u)=0\), where \[ F(u)=\log\det u_{x_ix_j}-\log\det(-u_{y_iy_j}), \quad H(u)=\frac{\partial}{\partial t}u-F(u). \] In the complex case, coordinates are \(\{z_i\}_{i=1}^k\) and \(\{w_i\}_{i=1}^\ell\) and \(u\) is plurisubharmonic in the \(z\) variables and plurisuperharmonic in the \(w\) variables. The equations are \(F_\mathbb{C}(u)=0\) and \(H_\mathbb{C}(u)=0\), where \[ F_\mathbb{C}(u)=\log\det u_{z_i\bar{z}_j}-\log\det(-u_{w_i\bar{w}_j}), \quad H_\mathbb{C}(u)=\frac{\partial}{\partial t}u-F_\mathbb{C}(u). \] Let \(U\in \mathbb{R}^k\times\mathbb{R}^\ell\), or \(\in\mathbb{C}^k\times\mathbb{C}^\ell\) and \(\lambda, \Lambda>0\), and set \[ \begin{aligned} \mathcal{E}_U^{k,\ell,\mathbb{R}}&=\{u\in C^\infty(U)| D^2u|_{\mathbb{R}^k}>0, D^2u|_{\mathbb{R}^\ell}<0\}, \\ \mathcal{E}_U^{k,\ell,\mathbb{C}}&=\{u\in C^\infty(U)|\mathrm{i}\partial^bar{\partial}u|_{\mathbb{C}^k}>0, \mathrm{i}\partial\bar{\partial}u|_{\mathbb{C}^\ell}<0\}.\\ \mathcal{E}_{U,\lambda,\Lambda}^{k,\ell,\mathbb{R}}&=\{u\in\mathcal{E}_U^{k.\ell.\mathbb{R}}|\lambda I_k\leq D^2u|_{\mathbb{R}^k}\leq \Lambda I_k, \lambda I_\ell\leq(-D^2u|_{\mathbb{R}^\ell})\leq\Lambda I_\ell \}, \\ \mathcal{E}_{U,\lambda,\Lambda}^{k,\ell,\mathbb{C}}&=\{u\in\mathcal{E}_U^{k,\ell,\mathbb{C}}|\lambda I_k\leq \mathrm{i}\partial\bar{\partial}u|_{\mathbb{C}^k}\leq\Lambda I_k, \lambda I_\ell\leq(-\mathrm{i}\partial\bar{\partial}u|_{\mathbb{C}^\ell})\leq \Lambda I_\ell \}.\end{aligned}\tag{Definition 2.1} \] Then this paper proves Theorem 1.1. There exists \(C=C(k,\ell,\lambda,\Lambda)\) and \(\alpha=\alpha(k,\ell,\lambda,\Lambda)>0\) such that {\parindent=0.7cm \begin{itemize}\item[--] If \(u\in\mathcal{E}^{k,\ell,\mathbb{R}}_{B_2,\lambda.\Lambda}\) and \(F(u)=0\), then \(\|u\|_{C^{2,\alpha}(B_1)}\leq C\). \item[--] If \(u\in\mathcal{E}_{B_2,\lambda,\Lambda}^{k,\ell,\mathbb{C}}\) and \(F_\mathbb{C}(u)=0\), then \(\|u\|_{C^{2,\alpha}(B_1)}\leq C\). \item[--] If \(u\in\mathcal{E}_{Q_2,\lambda,\Lambda}^{k,\ell,\mathbb{R}}\) and \(H(u)=0\), then \(\|\dfrac{\partial}{\partial t}u\|_{C^\alpha(Q_1)}+\|u\|_{C^{2,\alpha}(Q_1)}\leq C\). \item[--] If \(u\in\mathcal{E}_{Q_2,\lambda,\Lambda}^{k,\ell,\mathbb{C}}\) and \(H_\mathbb{C}u=0\), then \(\|\dfrac{\partial}{\partial t}u\|_{C^\alpha(Q_1)}+\|u\|_{C^{2,\alpha}(Q_1)}\leq C\). \end{itemize}} Proof of real case is based on Legendre transform \(u\to w\), where \(u\) is a convex function on \(\mathbb{R}^n\) and \(w(y)=x_i\dfrac{\partial u}{\partial x_i}(x)-u(x)\), where \(y_i(x)=\dfrac{\partial u}{\partial x_i}(x)\). Because we have \[ \det\frac{\partial^2w}{\partial y_i\partial y_j}=\bigl(\det\frac{\partial^2u}{\partial x_i\partial x_j}\bigr)^{-1}. \] To adopt Legendre transform to the study of the twisted Monge-Ampère equation, partial Legendre transform \(\mathcal{PL}_{k,\ell}:\mathbb{R}_\Omega\to \mathcal{E}^{n,0,\mathbb{R}}_{\hat{\Omega}}\) is introduced (Definition 2.2). Then after computing its transformation law, it is shown if \(F(u)=0\) and \(u\in\mathcal{E}~{k,\ell}_{\mathbb{R}^n,\lambda,\Lambda}\), then \(u\) is a quadratic polynomial (Coll.2.4). In \S3, application of partial Legendre transform is described and obtain monotonicity formulas for certain combinations of second derivatives along solutions of \(H(u)=0\) and \(H_\mathbb{C}(u)=0\) (Prop.3.2). Details of the proofs are given in Appendix, because they need long calculations. The authors say there is s yet not a clear Legendre transformation defined for plurisubharmonic functions. But maximum principle estimates which correspond to usual maximum principle estimates for the complex Monge-Ampère equation, assuming the Legendre transformation were defined. Hence we obtain corresponding results in real cases (Proposition 3.3). Th.1.1 is proved by these results in \S4. In \S5, the last Section, adopting Theorem 1.1, the following a priori regularity results for the pluriharmonic flow [\textit{J. Streets}, ``Pluriclosed flow on generalized Kähler manifolds with split tangent bundle'', Preprint, \url{arXiv:1405.0727}, \textit{J. Streets} and \textit{G. Tian}, Nucl. Phys., B 858, No. 2, 366--376 (2012; Zbl 1246.53091)] is established. Theorem 1.2. Let \(M^{2n},g_0,J_A,J_B)\) be a compact generalized Kähler manifolds satisfying \([J_A,J_B]=0\). Let \(g_t\) be the solution to pluriharmonic flow with initial condiiton \(g_0\). Suppose the solution exists on \([0,\tau), \tau<\tau^\ast(g)\)(cf. Definition 5.6) and there exists and there exists constant \(\lambda,\Lambda\) such that \(\lambda g_0<g_t<\Lambda g_0\). Given \(k\in\mathbb{N}, \alpha\in[0,1)\) there exists a constant \(C=C(k,\alpha,g_0,\lambda,\Lambda,\tau)\) such that \(\|g_t\|_{C^{k,\alpha}}\leq C\).