[Paper Review] Equivalent Semigroup Properties for Curvature-Dimension Condition

Feng‐Yu Wang|arXiv (Cornell University)|Dec 28, 2010

Geometric Analysis and Curvature Flows13 references4 citations

TL;DR

This paper establishes equivalent semigroup inequalities—such as gradient, Harnack, and log-Harnack inequalities—for the curvature-dimension condition (CD) with finite dimension $ n $, extending known results from the $ n = ∞ $ case. The key contribution is proving that these inequalities are equivalent to the Bakry-Émery CD condition, enabling new heat kernel estimates, HWI inequalities, and transportation cost bounds under general curvature and dimension constraints.

ABSTRACT

Some equivalent gradient and Harnack inequalities of a diffusion semigroup are presented for the curvature-dimension condition of the associated generator. As applications, the first eigenvalue, the log-Harnack inequality, the heat kernel estimates, and the HWI inequality are derived by using the curvature-dimension condition. The transportation inequality for diffusion semigroups is also investigated.

Motivation & Objective

• To identify analytic inequalities of the diffusion semigroup $ P_t $ that are equivalent to the curvature-dimension condition (CD) with finite dimension $ n $.
• To extend known semigroup inequalities—previously valid only for $ n = ∞ $—to the finite $ n $ case.
• To derive new quantitative estimates for the heat kernel, first eigenvalue, and Wasserstein distances using the equivalent inequalities.
• To establish transportation-cost inequalities for diffusion semigroups under the CD condition with finite $ n $.

Proposed method

• Derives six equivalent inequalities (1)-(6) for the semigroup $ P_t $, including gradient, Harnack, and log-Harnack type bounds, under the CD condition.
• Uses coupling by reflection and parallel displacement of diffusion processes to estimate the distance process $ \rho(X_t, Y_t) $, linking it to the curvature and drift $ Z $.
• Applies Itô's formula to the distance process and derives bounds on the drift term $ I_Z(x,y) $ using Jacobi field estimates and curvature assumptions.
• Constructs explicit test functions $ f(s) $ along geodesics to bound the index form and derive sharp estimates for $ I_Z(x,y) $.
• Uses the coupling construction to derive moment bounds on $ \mathbb{E}[\tilde{\rho}(X_t, Y_t)^p] $, leading to exponential contraction in Wasserstein distance.
• Derives explicit heat kernel inequalities (1.6) and (1.7) from the log-Harnack inequality (6), linking them to relative entropy and $ L^2 $-norms.

Experimental results

Research questions

• RQ1 Are there equivalent semigroup inequalities for the curvature-dimension condition with finite $ n $, beyond the $ n = \infty $ case?
• RQ2 Can the log-Harnack inequality (6) be used to derive explicit heat kernel bounds and relative entropy estimates?
• RQ3 What is the optimal contraction rate of the diffusion semigroup in Wasserstein distance under the CD condition with finite $ n $?
• RQ4 How do the coupling by reflection and parallel displacement methods yield sharp estimates for the distance process under curvature and drift constraints?
• RQ5 What is the precise dependence of the HWI inequality and transportation cost bounds on the curvature $ K $, dimension $ n $, and Ricci curvature?

Key findings

• The curvature-dimension condition (1.1) is equivalent to six semigroup inequalities (1)-(6), including gradient, Harnack, and log-Harnack inequalities.
• The log-Harnack inequality (6) implies the heat kernel inequality (1.6) and (1.7), which bound the relative entropy of $ p_{t+s}^\nu $ and $ p_t^\nu $ in terms of $ \rho(x,y)^2 $, $ K $, and $ n $.
• For $ K > 0 $, the Wasserstein distance $ W_p(\mu_1 P_t, \mu_2 P_t) \leq e^{Kt} W_p(\mu_1, \mu_2) $, showing exponential contraction.
• For $ K < 0 $, the contraction rate is $ \exp\left(\frac{nK}{n-1}t\right) $, which is stronger than the $ e^{Kt} $ rate when $ n < \infty $.
• The coupling by reflection yields the bound $ \mathbb{E}[\tilde{\rho}(X_t, Y_t)] \leq e^{\frac{nK}{n-1}t} \mathbb{E}[\tilde{\rho}(X_0, Y_0)] $, implying exponential contraction in $ W_1 $.
• The explicit form of $ I_Z(x,y) $ is $ 2\sqrt{K(n-1)}\tanh(\frac{\rho}{2}\sqrt{K/(n-1)}) $ for $ K > 0 $, and $ -2\sqrt{-K(n-1)}\tan(\frac{\rho}{2}\sqrt{-K/(n-1)}) $ for $ K < 0 $, which is sharp.