Parabolic Monge-Ampère Equations

• Parabolic Monge-Ampère equations are nonlinear, time-dependent PDEs that generalize elliptic models and are central to complex, Hermitian, and convex geometry.
• They utilize parabolic flow methods, rigorous a priori estimates, and viscosity solutions to secure regularity, convergence, and well-posedness in diverse settings.
• These equations bridge geometric analysis with algorithmic optimal transport, offering practical tools for canonical metric problems and singularity treatment.

Parabolic Monge-Ampère equations are a class of fully nonlinear, time-dependent partial differential equations integral to complex, real, quaternionic, and optimal transport geometry. They generalize the elliptic Monge-Ampère equation, introducing an evolutionary aspect, and are structurally central to complex differential geometry, pluripotential theory, convex geometry, mirror symmetry, and algorithmic optimal transport. Parabolic Monge-Ampère flow methods yield powerful proofs and regularity results for canonical metrics on Kähler, Hermitian, and Gauduchon manifolds, and underpin several fundamental geometric and analytic flows.

1. Analytical Framework and Model Equations

The prototypical parabolic Monge-Ampère equation on a compact n-dimensional complex or real manifold $(M,g)$ evolves a real potential $u(x,t)$ via

$$\frac{\partial u}{\partial t} = \log\frac{\det( g_{i\bar j} + u_{i\bar j} )}{\omega^n} - F(x),$$

subject to an initial condition $u(0,x) = u_0(x)$ and a positivity constraint on the evolving form (e.g., $\omega + i\partial\bar\partial u > 0$ in the complex setting). In the real convex case, for $u : \Omega \subset \mathbb{R}^n \to \mathbb{R}$ and uniformly convex domain $\Omega$, the equation reads

$$-u_t + \det D^2 u = \psi(x,t),\quad u|_{\partial_p Q_T} = \phi(x,t),$$

where $Q_T = \Omega \times (0,T]$ and $\partial_p Q_T$ is the parabolic boundary. The boundary data and compatibility assumptions ensure well-posedness and regularity (Zhou et al., 2024, Gill, 2010).

In the Hermitian case, one studies generalized flows such as

$$\partial_t u = \ln S_n(X_u) - \ln S_{n-\alpha}(X_u) + \ln \chi^n - \ln \psi(x),\quad X_u = \chi + i\partial\bar\partial u,$$

for a background form $\chi$ and constants $\alpha$, unifying several classical flows such as the Kähler-Ricci flow (Cao), J-flow (Donaldson’s problem), and Chern-Ricci flow (Sun, 2013, Sun, 2015).

On Gauduchon manifolds, the natural parabolic flow becomes, with torsion and trace terms,

$$\frac{\partial u}{\partial t} = \log \frac{(\omega + \frac{1}{n-1}[(\Delta u)\omega - \sqrt{-1}\partial\bar\partial u] + Z(u))^n}{\omega^n} - F(x),$$

preserving the Gauduchon condition and leading to a parabolic proof of the Gauduchon conjecture (Zheng, 2016).

Quaternionic analogues on hyperKähler manifolds similarly admit evolution equations involving the quaternionic Hessian, such as

$$\partial_t u = \log \det\left( R_{h\,i \bar j} +\tfrac1{n-1} ((\Delta_{I,g} u)g_{i\bar j}-u_{i\bar j}) \right) - f(x),$$

effectively extending the Calabi–Yau paradigm to the HKT/quaternionic context (Fu et al., 2023).

2. A Priori Estimates and Regularity

Global existence and regularity rely on a closed hierarchy of a priori bounds:

C⁰-bound (oscillation):

$$\sup_M u( \cdot, t ) - \inf_M u( \cdot, t ) \leq C_0,$$

by maximum principle arguments for $\partial_t u$ (Gill, 2010, Chu, 2016).

C¹ (gradient) estimate: Auxiliary test functions such as $Q = e^{f(u)}|\nabla u|^2$ and a parabolic maximum principle yield

$|\nabla u|(x,t) \leq C_1.$

Second-order (Laplacian/Hessian) estimate: Using carefully chosen test quantities (e.g., $\log \mathrm{Tr}_\omega(\omega_u) - A(u - \inf u)$, or the log of the largest eigenvalue plus gradient and exponential corrections for non-integrable backgrounds), one obtains

$$\mathrm{Tr}_\omega(\omega_u) \leq C_2 \exp\left\{ A(\sup_M u - \inf_M u) \right\}.$$

This extends to the full Hessian via Legendre duality and geometric barriers in the real convex case (Zhou et al., 2024).

Higher-order and Hölder regularity: Parabolic Evans–Krylov theory applies given uniform parabolicity, providing $C^{2,\alpha}$ bounds for $u$, and parabolic Schauder estimates yield $C^k$ regularity for all $k$.

Harnack inequalities: For the time derivative $y = \partial_t u$, one derives exponential decay of oscillation and convergence: $$\sup_M y(\cdot, t_1) \le \inf_M y(\cdot, t_2) e^{C(t_2 - t_1)} + C, \quad \Rightarrow \text{osc}_M \partial_t u \le C e^{-\lambda t}.$$ Integration in time then enforces $u(t) \to u_\infty$ in $C^\infty$ (Sun, 2013, Chu, 2016).

Boundary regularity: In convex domain settings, boundary $C^2$ and $C^{2,\alpha}$-estimates are established using barrier constructions and weak-Harnack iteration (Zhou et al., 2024).

$L^\infty$-bounds under entropy conditions: For parabolic complex Monge-Ampère flows with rough twisting, entropy control is used to obtain $L^\infty$ bounds without pointwise assumptions (Zhao, 2023).

3. Long-Time Behavior and Convergence

The normalized parabolic flow, with appropriate subtraction of the evolving mean

$$\tilde u(t) = u(t) - \frac{1}{\mathrm{Vol}(M)} \int_M u(t) \omega^n,$$

yields exponential decay of $\partial_t u$ and strong $C^\infty$ convergence to the elliptic solution: $$(\omega + i\partial\bar\partial u_\infty)^n = e^{F(x) + b} \omega^n,$$ where the constant $b$ is fixed by the calibration of the total volume (Gill, 2010, Chu, 2016, Sun, 2015).

In the real, convex domain case, the convergence is to time-independent convex solutions of the classical elliptic Monge–Ampère boundary value problem (Zhou et al., 2024). For flows on Gauduchon manifolds, the limiting metric solves the prescribed Chern–Ricci equation, establishing a parabolic proof of the Gauduchon conjecture (Zheng, 2016).

4. Viscosity Solutions and Weak Theory

Viscosity methods provide a powerful framework for parabolic Monge–Ampère equations with weak or non-smooth data, especially in domains with boundary or when right-hand sides vanish or degenerate:

• Viscosity sub- and supersolutions are defined by upper- and lower-test functions $q(t,z)$ in $C^{1,2}$; the comparison principle and Perron's method can be applied under minimal regularity (Do et al., 2021, Do et al., 2019).
• Existence, uniqueness, and Holder regularity of solutions are established even in the presence of moving zero sets or strongly pseudoconvex domains (Do et al., 2021).
• Weak solutions with conical singularities along divisors (e.g., for the conical Kähler–Ricci flow) exhibit $C^{2,\alpha,\beta}$ regularity after initial smoothing steps, providing powerful tools for singular metric problems (Liu et al., 2016).

5. Algorithmic and Optimal Transport Aspects

Parabolic Monge–Ampère equations arise in large-scale limits of discrete regularizations in optimal transport, most notably as the continuous-time limit of the Sinkhorn algorithm: $$\frac{\partial \psi_t}{\partial t}(y) = -f(\nabla \psi_t(y)) + g(y) + \log \det( \nabla^2 \psi_t(y) ).$$ This mirrors iterative refinement in Brenier potential space and allows for efficient generative algorithms with no-regret guarantees, even for non-log-concave targets (Deb et al., 12 Apr 2025, Berman, 2017). The time-discretized mirror-descent/sinkhorn steps converge provably under mild convexity assumptions.

A new evolution variational inequality (EVI) adapted to the mirror-descent in Wasserstein space converts per-iterate KL-divergence improvement into telescoping Bregman divergence terms, quantifying convergence rates in both average and last-iterate sense.

Connections to geometric flows (e.g., Ricci flow on torus, reflector-antenna problem on the sphere) are explicit in this framework (Berman, 2017).

6. Geometric and Structural Significance

Parabolic Monge–Ampère equations:

• Generalize and unify classic flows: the Kähler–Ricci, Chern–Ricci, J-flow, and their extensions on non-Kähler (Hermitian, Gauduchon) backgrounds (Gill, 2010, Sun, 2013, Sun, 2015, Zheng, 2016).
• Facilitate parabolic proofs of existence and uniqueness for complex Monge–Ampère equations, often bypassing continuity method technicalities, automatically selecting the normalization constant for the limiting elliptic PDE (Gill, 2010).
• Enable existence and regularity for metrics with singularities, such as conical metrics along divisors, in both weak and strong senses, under minimal data assumptions (Liu et al., 2016).
• Provide robust frameworks for canonical metric problems, such as the Calabi–Yau theorem (Kähler case), Hermitian–Yau–Tosatti–Weinkove, and the Gauduchon conjecture (Zheng, 2016).

7. Methodological Innovations and Extensions

• Maximum principle-based a priori estimates, extended to non-integrable (almost Hermitian) and non-Kähler backgrounds, handle torsion and higher-order nonlinearity (Chu, 2016, Fu et al., 2023).
• Legendre transform and duality techniques enable global Hessian bounds in real, convex settings (Zhou et al., 2024).
• Parabolic Schauder and Evans–Krylov theory are extended to concave and generalized, non-concave flows, underpinning full regularity.
• Viscosity solution methods, stability under time/space approximation, and intrinsic comparison principles play a decisive role in the non-smooth and boundary value settings (Do et al., 2021, Do et al., 2019).
• Mirror-descent and no-regret evolutionary inequalities explicitly connect nonlinear parabolic PDE theory to practical algorithms for transport, sampling, and generative learning (Deb et al., 12 Apr 2025).

In summary, parabolic Monge–Ampère equations constitute a core analytical and geometric structure, serving as the evolutionary backbone for resolving regularity, existence, and canonical metric problems in complex, Hermitian, real, and quaternionic geometry, with significant impact in pluripotential theory, singular geometry, convex analysis, high-dimensional optimization, and computational mathematics.