Li–Yau Gradient Estimate on Manifolds

• Li–Yau Gradient Estimate is a differential inequality that provides bounds on the normalized gradient of positive solutions to elliptic and parabolic equations on manifolds.
• It explicitly relates curvature via the Bakry–Émery Ricci tensor and nonlinear interaction terms, yielding both local and global gradient estimates.
• The methodology employs maximum principles and Weitzenböck formulas, leading to practical applications in Harnack inequalities and Liouville-type rigidity theorems.

A Li–Yau gradient estimate, or Li–Yau inequality, is a fundamental type of pointwise differential inequality for positive solutions to elliptic or parabolic equations on manifolds (and their generalizations). First introduced in the context of the heat equation on Riemannian manifolds, Li–Yau estimates give quantitative bounds on the size of the (normalized) gradient of a solution in terms of curvature and other geometric data, often enabling Harnack inequalities and rigidity results. In the context of nonlinear elliptic equations with the Witten Laplacian on smooth metric measure spaces, these estimates unify, extend, and strengthen classical results to a broader class of geometries and operators, particularly those involving the Bakry–Émery Ricci curvature tensor and a broad class of nonlinearities.

1. Setting and Main Gradient Estimates

Consider a smooth metric measure space $(M^n, g, e^{-f} dv_g)$, where $g$ is a Riemannian metric and $f \in C^\infty(M)$ is a weight function. The relevant differential operator is the Witten (or drifting) Laplacian

$$A_f u = \Delta u - \langle \nabla f, \nabla u \rangle,$$

and the $m$-Bakry–Émery Ricci tensor

$\Ric_f^m = \Ric + \Hess f - \frac{1}{m-n} df \otimes df, \qquad m \geq n.$

Focus on positive solutions $u>0$ to nonlinear elliptic equations of the form

$A_f u(x) + \mathcal{E}(x, u(x)) = 0,$

where $\mathcal{E}(x, u)$ is a smooth nonlinearity.

Local (Ball) Estimate

On $B_{2R}(p)$ (geodesic ball in $g$0), assuming $g$1 ($g$2), the main local Li–Yau type gradient estimate is

$g$3

where

• (I) is a local geometry term involving $g$4, and $g$5,
• (II) is a nonlinear interaction term depending on suprema of various derivatives of $g$6,
• (III) captures the "bad part" of the nonlinearity via $g$7.

All constants are given explicitly in (Taheri et al., 2023), Theorem 2.1.

Global Estimate

If curvature assumptions hold globally,

$g$8

with (I) (local geometry) vanishing as $g$9.

2. Dependence on Geometric and Nonlinear Data

The gradient bound exhibits explicit dependence on:

• Curvature: Only via a Bakry–Émery lower bound $f \in C^\infty(M)$0, entering constants in a dimensionally sharp way.
• Dimension: Both real $f \in C^\infty(M)$1 and Bakry–Émery dimension $f \in C^\infty(M)$2 appear, controlling the influence of curvature.
• Nonlinearity: All terms—$f \in C^\infty(M)$3, its first ($f \in C^\infty(M)$4, $f \in C^\infty(M)$5) and second derivatives ($f \in C^\infty(M)$6), and their suprema—enter explicitly into the estimate, capturing both growth and structural aspects.
• Cutoff Radii: Local estimates account for ball radii via geometric constants, while global estimates drop these terms.

This enables sharp control in environments with weighted Ricci curvature and diverse types of nonlinear drift or reaction terms.

3. Methodology and Proof Outline

The proof strategy adapts classical maximum principle and Bochner–Weitzenböck arguments to the smooth metric measure space and drifting Laplacian context.

• Change of Variables: Set $f \in C^\infty(M)$7 to translate the nonlinear PDE into a suitable form for differential manipulation.
• Weitzenböck Formula: For the Witten Laplacian, compute $f \in C^\infty(M)$8, which involves second derivatives, drift interactions, and curvature.
• Auxiliary Function: Introduce a Li–Yau type quantity $f \in C^\infty(M)$9, with $$A_f u = \Delta u - \langle \nabla f, \nabla u \rangle,$$0, to localize the argument.
• Differential Inequality: Establish a lower bound for $$A_f u = \Delta u - \langle \nabla f, \nabla u \rangle,$$1 up to gradient and nonlinearity interaction terms, using curvature-dimension bounds.
• Cut-off Function and Maximum Principle: Multiply by a cutoff $$A_f u = \Delta u - \langle \nabla f, \nabla u \rangle,$$2, apply the maximum principle at an interior point, and control all error terms with explicit constants.
• Globalization: Passing $$A_f u = \Delta u - \langle \nabla f, \nabla u \rangle,$$3 removes local terms yielding the global inequality.

This method is robust to the inclusion of general nonlinearities, the drift term from $$A_f u = \Delta u - \langle \nabla f, \nabla u \rangle,$$4, and weaker curvature lower bounds.

4. Comparison to Classical Li–Yau Estimates

The original Li–Yau estimate applies to the heat equation,

$$A_f u = \Delta u - \langle \nabla f, \nabla u \rangle,$$5

assuming $$A_f u = \Delta u - \langle \nabla f, \nabla u \rangle,$$6. In the static (elliptic) case, for harmonic functions, this gives the classic Yau gradient bound.

The advances in (Taheri et al., 2023) generalize this in several dimensions:

• Witten Laplacian: The inclusion of a weighted drift $$A_f u = \Delta u - \langle \nabla f, \nabla u \rangle,$$7 and the Bakry–Émery Ricci tensor.
• General Nonlinearity: The class $$A_f u = \Delta u - \langle \nabla f, \nabla u \rangle,$$8 covers log-linear, power-type, and other nonlinear reaction terms, not just linear equations.
• Elliptic and Fully Nonlinear Context: Both local and global results, independent of time, and valid on complete noncompact manifolds provided curvature-dimension hypotheses.
• Dimensionally Sharp Structure: The constants precisely capture $$A_f u = \Delta u - \langle \nabla f, \nabla u \rangle,$$9, drift, and curvature influences.

The methodology leverages Weitzenböck formula extensions and refined maximum principle arguments characteristic of Li–Yau-type theory but adapted to a broader analytic and geometric framework.

5. Principal Applications: Harnack Inequalities and Liouville-Type Theorems

Elliptic Harnack Inequalities

The Li–Yau gradient bound enables immediate two-point (elliptic) Harnack inequalities:

• Local: For $m$0 solving (1.1) in $m$1, for $m$2,

$m$3

where $m$4 is the right-hand side of the gradient estimate.

• Global: On the entire manifold,

$m$5

Thus, uniform gradient control translates into sharp pointwise upper/lower bounds.

Liouville-Type Rigidity Theorems

Under $m$6 and appropriate structural conditions on $m$7, the gradient inequality forces

$m$8

and hence $m$9.

Typical cases cover:

• Power-type nonlinearities: If $\Ric_f^m = \Ric + \Hess f - \frac{1}{m-n} df \otimes df, \qquad m \geq n.$0 with all $\Ric_f^m = \Ric + \Hess f - \frac{1}{m-n} df \otimes df, \qquad m \geq n.$1, $\Ric_f^m = \Ric + \Hess f - \frac{1}{m-n} df \otimes df, \qquad m \geq n.$2, then any positive solution must be constant.
• Log-power mixtures: For $\Ric_f^m = \Ric + \Hess f - \frac{1}{m-n} df \otimes df, \qquad m \geq n.$3, with conditions on $\Ric_f^m = \Ric + \Hess f - \frac{1}{m-n} df \otimes df, \qquad m \geq n.$4 and $\Ric_f^m = \Ric + \Hess f - \frac{1}{m-n} df \otimes df, \qquad m \geq n.$5, constancy is again forced.

These generalize and extend classical non-existence and constant-solution results to highly general elliptic PDEs on weighted manifolds.

6. Broader Significance and Unification

The gradient estimates synthesized in (Taheri et al., 2023) function as a general framework subsuming:

• Linear theory (Laplace/heat equations under Ricci curvature lower bounds).
• Drifting Laplacians and Bakry–Émery geometry.
• Nonlinear equations with a broad spectrum of reaction terms, including applications to the geometry of Ricci solitons and their generalizations.
• Rigorous handling of both local and global estimates, noncompact settings, and curvature dimension conditions.

These results provide robust analytic tools that propagate into applications such as eigenvalue bounds, heat kernel estimates, and rigidity phenomena on geometric-analytic and probabilistic PDEs, emphasizing the unifying role of Li–Yau gradient bounds in contemporary geometric analysis (Taheri et al., 2023).