Discrete Ricci Curvature via Convexity of the Entropy

on YouTube ↗56m

" summary

Shows W2 fails on discrete graphs (two‑point case gives no W2 geodesics), so builds a new Riemannian‑type transport metric using a logarithmic mean to pair vertex densities with edge gradients. Proves the continuous‑time heat semigroup is the gradient flow of entropy for that metric, derives modified log‑Sobolev with constant = curvature lower bound, proves tensorization and recovers sharp constants for the hypercube, and connects to JKO/EVI methods.

AI-written summary · about

" chapters (9)

1. 0:00Introduction to optimal transport
2. 1:24The Wasserstein metric
3. 6:41McCann's theorem and geodesics
4. 12:24Generalizing curvature bounds
5. 18:12Challenges in discrete settings
6. 23:54Heat flow as a gradient flow
7. 28:51Defining a discrete metric
8. 36:12Logarithmic mean and chain rule
9. 44:20Functional inequalities and results