Lin-Lu-Yau Curvature on Graphs

• Lin-Lu-Yau curvature is a discrete Ricci-type measure defined on graph edges via lazy random walk measures and Laplacian comparisons.
• It enforces local-to-global geometric constraints, yielding rigidity, finiteness, and connectivity bounds across various graph classes.
• Extensions to weighted, directed graphs and hypergraphs broaden its applications in discrete Ricci flows and spectral graph analysis.

Lin-Lu-Yau Curvature

The Lin-Lu-Yau (LLY) curvature is a discrete Ricci-type curvature notion defined for graphs (and recently for hypergraphs and directed graphs) as an analytic and combinatorial adaptation of Ollivier’s Ricci curvature. Introduced by Lin, Lu, and Yau in 2011, LLY curvature exhibits both metric-measure and classical combinatorial features, with fundamental rigidity, finiteness, and extremal properties across graph classes. Core to the theory are optimal transport, Markov random walks, and Laplacian inequalities, yielding both local and global geometric constraints.

1. Formal Definition and Characterizations

For a simple connected graph $G=(V,E)$ with graph metric $d(.,.)$, the LLY curvature operates on edges using discrete transportation metrics and lazy random walks. For each $x\in V$ and $\alpha\in[0,1)$, define the $\alpha$-lazy random walk measure

$$m_x^\alpha(v) = \begin{cases} \alpha & v = x \ \frac{1-\alpha}{\deg(x)} & v \in N(x) \ 0 & \text{otherwise} \end{cases}$$

where $N(x)$ is the neighborhood of $x$. The 1-Wasserstein distance between probability measures $m_1, m_2$ is

$$W(m_1, m_2) = \inf_{A} \sum_{u,v \in V} A(u,v)\, d(u,v)$$

with the infimum over all couplings $d(.,.)$0 of $d(.,.)$1.

The $d(.,.)$2-Ollivier-Ricci curvature along $d(.,.)$3 is

$d(.,.)$4

The LLY curvature is the first derivative at $d(.,.)$5: $d(.,.)$6

A fundamental limit-free formulation is in terms of normalized combinatorial Laplacian: $d(.,.)$7 where $d(.,.)$8 and $d(.,.)$9 indicates $x\in V$0 is 1-Lipschitz (Lu et al., 2020).

On regular graphs, transport reduces to an optimal assignment between asymmetrically paired neighborhoods: $x\in V$1 where $x\in V$2 and the minimum is over all bijections $x\in V$3 (Hehl, 2024, Li et al., 2024, Mou, 16 Jun 2025).

2. Core Structural and Rigidity Results

LLY curvature is highly sensitive to local graph structure and yields strong global constraints.

Planar and Outerplanar Graphs

For planar graphs with minimum degree at least 3, strict positivity of LLY curvature on every edge implies finiteness and a maximal degree bound: $x\in V$4 with an explicit (large) bound on order $x\in V$5, and similar but sharper bounds for outerplanar graphs ($x\in V$6, $x\in V$7) (Lu et al., 2020, Brooks et al., 2024). These results depend on combinatorial lemmas linking positive curvature to forbidden local configurations (caps, minors, and neighborhood overlaps).

Connectivity and Degree Constraints

Finiteness plus non-negative LLY curvature (resp. strictly positive) enforces optimal edge-connectivity: $x\in V$8 where $x\in V$9 is minimum degree and $\alpha\in[0,1)$0 is edge-connectivity (Liu et al., 28 Aug 2025, Chen et al., 19 Apr 2025). Conversely, finite connected graphs admitting $\alpha\in[0,1)$1 with $\alpha\in[0,1)$2 must be infinite. Lower bounds on vertex-connectivity scale with minimum degree times the LLY curvature, and for high connectivity $\alpha\in[0,1)$3, curvature is uniformly positive.

A minimum-degree threshold also ensures non-negativity: $\alpha\in[0,1)$4 and this bound is sharp (Hehl, 6 Feb 2025).

Diameter and Volume Bounds

Positive curvature bounds global geometry: $\alpha\in[0,1)$5 and for $\alpha\in[0,1)$6-free graphs, a tight order bound applies: $\alpha\in[0,1)$7 with equality if and only if $\alpha\in[0,1)$8 is a hypercube (Gamlath et al., 2023). Regular graphs with positive LLY curvature also satisfy refined Moore-type bounds.

3. Curvature in Regular and Strongly Regular Graphs

Amply regular graphs, including strongly regular and distance-regular graphs, allow explicit curvature computations via neighborhood-matching methods. For an amply regular graph with degree $\alpha\in[0,1)$9, parameters $\alpha$0, girth 3, and $\alpha$1, the sharp bound is

$\alpha$2

and all conference graphs ($\alpha$3) have strictly positive curvature (Huang et al., 2022, Li et al., 2021).

Matching-based transport plans via Hall’s theorem underlie these results, providing immediate lower bounds for graph diameter and Laplacian eigenvalues: $\alpha$4

Explicit formulas for LLY curvature also characterize and distinguish cocktail-party graphs ($\alpha$5) by the property $\alpha$6, hypercubes ($\alpha$7), and bones for bone-idle graphs where all edges are Ricci-flat ($\alpha$8 for all $\alpha$9) (Hehl, 2024, Hehl, 2024, Li et al., 2024).

4. Extremal Examples, Classification, and Forbidden Structures

Graph families saturating degree or order bounds include hypercubes for the $$m_x^\alpha(v) = \begin{cases} \alpha & v = x \ \frac{1-\alpha}{\deg(x)} & v \in N(x) \ 0 & \text{otherwise} \end{cases}$$0-free bound, cocktail-party graphs for the maximal unit curvature result, and specialized exceptional Halin and outerplanar graphs for positive curvature with bounded order (Chen et al., 7 May 2025, Chen et al., 6 Mar 2025, Lu et al., 2020).

The structure of positively curved graphs is heavily restricted by local girth and neighborhood overlap:

• Triangles or squares are often forced in local neighborhoods to avoid negative curvature.
• Planar graphs with $$m_x^\alpha(v) = \begin{cases} \alpha & v = x \ \frac{1-\alpha}{\deg(x)} & v \in N(x) \ 0 & \text{otherwise} \end{cases}$$1 and positive LLY curvature must be finite; Halin-type and outerplanar graphs are classified completely up to small order.
• In regular graphs of girth four, the assignment formulation detects bone-idle behavior tied to matchings at distance three (Hehl, 2024, Hehl, 2024).

5. Lin-Lu-Yau Curvature in Products, Weighted, Directed, and Hypergraphs

Recent developments extend LLY curvature to products, weighted graphs, and hypergraphs:

• Product graphs: The curvature of strong and Cartesian products of regular graphs admits explicit formulas, with matching-type decompositions yielding exact assignment-based LLY curvature for all horizontal/vertical edges and certain diagonal types (Mou, 16 Jun 2025).
• Weighted graphs and Ricci flow: LLY curvature is defined for weighted graphs with vertex and edge measures. The associated LLY Ricci flow for edge-weights,

$$m_x^\alpha(v) = \begin{cases} \alpha & v = x \ \frac{1-\alpha}{\deg(x)} & v \in N(x) \ 0 & \text{otherwise} \end{cases}$$2

evolves toward metrics with constant curvature on trees, coinciding there with the Forman curvature flow (Bai et al., 6 Jan 2026).

• Hypergraphs: The LLY approach generalizes to both undirected and directed hypergraphs, with properly adapted random walk kernels and Wasserstein transport. Key results include existence of the curvature limit, diameter bounds, monotonicity properties, and explicit examples demonstrating the geometric and analytic reach of the notion in higher-order combinatorics (Tian et al., 5 Jul 2025, Ikeda et al., 2021).
• Directed graphs: LLY curvature for directed graphs is defined using symmetrized Markov kernels and an adapted Laplacian, yielding Bonnet–Myers and Cheng-type rigidity theorems and product formulas. Positivity implies structural constraints analogous to the undirected theory, with special attention to spherically suspended graphs and extremal diameter cases (Ozawa et al., 2020).

6. Connections and Contrasts With Other Curvature Notions

LLY curvature is closely related but quantitatively and qualitatively distinct from:

• Ollivier-Ricci curvature (finite $$m_x^\alpha(v) = \begin{cases} \alpha & v = x \ \frac{1-\alpha}{\deg(x)} & v \in N(x) \ 0 & \text{otherwise} \end{cases}$$3), which is concave in $$m_x^\alpha(v) = \begin{cases} \alpha & v = x \ \frac{1-\alpha}{\deg(x)} & v \in N(x) \ 0 & \text{otherwise} \end{cases}$$4, while LLY uses the linear scaling as $$m_x^\alpha(v) = \begin{cases} \alpha & v = x \ \frac{1-\alpha}{\deg(x)} & v \in N(x) \ 0 & \text{otherwise} \end{cases}$$5.
• Combinatorial curvature (angle-deficit or Euler–Gauss–Bonnet), which is vertex-based and relies on planar embeddings and face structure, unlike LLY's edge-based, metric-measure formulation.
• Bakry–Émery curvature, which in connectivity questions produces weaker bounds compared to LLY (cf. $$m_x^\alpha(v) = \begin{cases} \alpha & v = x \ \frac{1-\alpha}{\deg(x)} & v \in N(x) \ 0 & \text{otherwise} \end{cases}$$6 vs. $$m_x^\alpha(v) = \begin{cases} \alpha & v = x \ \frac{1-\alpha}{\deg(x)} & v \in N(x) \ 0 & \text{otherwise} \end{cases}$$7).

LLY curvature's metric-measure foundation underpins its analytic properties: Laplacian comparison principles, spectral gap and heat kernel decay, optimal transport interpretations, and tight Bonnet–Myers-type theorems (Lu et al., 2020, Hehl, 2024, Li et al., 2024, Bai et al., 6 Jan 2026).

7. Open Problems and Research Directions

Active areas include:

• Determining tight order/degree/extremal bounds for planar graphs and their subclasses with positive LLY curvature.
• Understanding bone-idle phenomena and Ricci-flat graph classification beyond low degree.
• Generalization to higher-order complexes (hypergraphs), with attention to nonlinearity and analytic subtleties.
• Extension and analysis of curvature flows (discrete Ricci flows), their convergence, normalization, and application to random walks and diffusion on complex networks.
• Full classification of graphs attaining Bonnet–Myers equality, especially in the directed setting, and investigation of volume growth and entropy bounds.

The cumulative body of results demonstrates the LLY curvature as a robust framework for detecting and enforcing local-to-global phenomena in discrete geometric analysis, crossing combinatorial, probabilistic, and analytic perspectives.