[Paper Review] Mass transport and variants of the logarithmic Sobolev inequality
This paper establishes new sufficient conditions for modified logarithmic Sobolev inequalities and isoperimetric inequalities using optimal transport methods, particularly focusing on measures with non-convex potentials that still satisfy functional inequalities due to strong integrability. The key contribution is a unified framework that generalizes known criteria and proves that Riemannian measures with lower Ricci curvature bounds and specific decaying potentials satisfy modified log-Sobolev and $F$-inequalities for $1 < \alpha \leq 2$, extending results to non-Gaussian and non-log-concave measures.
We develop the optimal transportation approach to modified log-Sobolev inequalities and to isoperimetric inequalities. Various sufficient conditions for such inequalities are given. Some of them are new even in the classical log-Sobolev case. The idea behind many of these conditions is that measures with a non-convex potential may enjoy such functional inequalities provided they have a strong integrability property that balances the lack of convexity. In addition, several known criteria are recovered in a simple unified way by transportation methods and generalized to the Riemannian setting.
Motivation & Objective
- To develop a systematic optimal transport approach to modified logarithmic Sobolev and isoperimetric inequalities.
- To identify sufficient conditions under which measures with non-convex potentials still satisfy functional inequalities, relying on strong integrability to counteract lack of convexity.
- To generalize known criteria for logarithmic Sobolev inequalities to the Riemannian setting and unify existing results via transportation techniques.
- To establish new functional inequalities—specifically $I(\tau)$, $F_\tau$, and modified log-Sobolev—for measures with power-type decay $e^{-|x|^\alpha}$, $1 < \alpha \leq 2$, on manifolds with Ricci curvature bounded below.
Proposed method
- Utilizes optimal transport theory to derive functional inequalities from isoperimetric properties, particularly through the use of cost functions and their convex conjugates.
- Applies 'tightening' techniques to reduce general inequalities to simpler forms, enabling the use of known inequalities on perturbed measures.
- Employs integration by parts and the Calabi lemma to handle non-smooth potentials (e.g., distance functions) on Riemannian manifolds, especially around cut loci.
- Uses perturbation arguments and boundedness of derivatives of logarithmic functions to control error terms in gradient estimates.
- Applies the Euclidean logarithmic Sobolev inequality as a base, then extends it to weighted Riemannian measures via a change of measure and perturbation of the potential.
- Relies on the bound $\Delta p \leq C_1 + C_2 r^{\alpha-1}$ and $|\nabla p|^2 \sim r^{2(\alpha-1)}$ to dominate curvature effects and apply Cauchy-Schwarz and Young’s inequality in the final estimates.
Experimental results
Research questions
- RQ1Under what conditions does a measure with a non-convex potential still satisfy a modified logarithmic Sobolev inequality, despite lacking strong convexity?
- RQ2Can optimal transport methods unify and generalize existing criteria for logarithmic Sobolev and $F$-Sobolev inequalities in both Euclidean and Riemannian settings?
- RQ3How do isoperimetric inequalities relate to modified log-Sobolev and $F$-Sobolev inequalities in the context of measures with power-type tails?
- RQ4What is the role of integrability properties in balancing the lack of convexity in the potential for functional inequalities to hold?
- RQ5Can the Euclidean logarithmic Sobolev inequality be extended to Riemannian manifolds with lower Ricci curvature bounds and non-Gaussian measures of the form $e^{-N\rho^\alpha}$?
Key findings
- The paper proves that for $1 < \alpha \leq 2$, the measure $\mu = \frac{1}{Z_{\alpha,N}} \exp(-N\rho^\alpha(x,x_0))\,dv$ on a Riemannian manifold with $\mathrm{Ric} \geq K\,\mathrm{Id}$ satisfies the $I(\tau)$-inequality with $\tau = 2(1 - 1/\alpha)$.
- Such measures also satisfy the $F_\tau$-inequality and a modified log-Sobolev inequality with cost function $c_{\alpha}$, which implies improved concentration and semigroup properties.
- The proof relies on transforming the Euclidean log-Sobolev inequality into a weighted form via a change of measure and applying perturbation techniques to handle non-smooth potentials.
- The method shows that even when the potential is not convex, strong integrability (e.g., $e^{\varepsilon r^\alpha} \in L^1(\nu)$) allows the functional inequalities to hold.
- The result generalizes previous findings for $\alpha \geq 2$ to the critical range $1 < \alpha < 2$, where the potential is less regular but still allows for functional inequalities.
- The framework unifies known criteria for $F$-Sobolev and super-Poincaré inequalities and recovers them via transportation methods, including Wang’s correspondence between $F$-inequalities and super-Poincaré inequalities.
Better researchstarts right now
From reading papers to final review, dramatically reduce your research time.
No credit card · Free plan available
This review was created by AI and reviewed by human editors.