Resistance-Based Curvature

• Resistance-based curvature is a framework that defines discrete geometric invariants via effective resistance metrics derived from electrical network theory.
• It bridges local and global graph properties by providing discrete analogues to Ricci and scalar curvature, influencing spectral, probabilistic, and combinatorial analyses.
• This approach underpins scalable algorithms for tasks like community detection and graph neural network design, leveraging efficient Laplacian-based computations.

Resistance-Based Curvature is a family of discrete geometric invariants for graphs, networks, and related structures, defined using effective resistance metrics arising from electrical network theory. These curvatures interpolate between highly local and global properties, provide rigorous analogues to Ricci curvature and scalar curvature in Riemannian geometry, and are tightly coupled to spectral, probabilistic, and combinatorial graph-theoretic quantities. They play a central role in modern geometric data analysis, discrete geometric flows, and network science.

1. Electrical Network Foundations and Effective Resistance

Let $G=(V,E)$ be a finite, connected graph. Assign to each edge $e=uv$ a positive resistance $\ell_e>0$, equivalently a conductance $w_{uv}=1/\ell_{uv}$, yielding the weighted (combinatorial) Laplacian $L$ with entries:

• $L_{uu} = \sum_{v \sim u} w_{uv}$,
• $L_{uv} = -w_{uv}$ for $u \neq v$.

For vertices $u,v \in V$, the effective resistance $\omega_{uv}$ is defined as the voltage drop across $e=uv$0 and $e=uv$1 under unit current injection at $e=uv$2 and extraction at $e=uv$3, computed as

$e=uv$4

where $e=uv$5 is the Moore–Penrose pseudoinverse of $e=uv$6 (Dawkins et al., 2024, Devriendt et al., 2022, Yadav et al., 26 Oct 2025).

For edge $e=uv$7, the relative resistance is $e=uv$8. By the Kirchhoff/Matrix-Tree theorem, $e=uv$9 admits both a spanning-tree formula and a recursive series-parallel law.

The effective resistance metric has the following essential properties (Devriendt et al., 2022):

• $\ell_e>0$0 is a metric: nonnegative, symmetric, and satisfies the triangle inequality,
• $\ell_e>0$1 iff $\ell_e>0$2,
• For trees, $\ell_e>0$3 equals the shortest-path (graph) distance.

2. Resistance-Based Curvature: Definitions

2.1 Vertex (Scalar) Curvature

For weighted, undirected $\ell_e>0$4, the resistance curvature at vertex $\ell_e>0$5 is

$\ell_e>0$6

This quantity aggregates the 'distance' from $\ell_e>0$7 to its neighbors measured by effective resistance. In unweighted graphs, $\ell_e>0$8.

On trees, $\ell_e>0$9, so leaf vertices have $w_{uv}=1/\ell_{uv}$0 and internal vertices of degree two have $w_{uv}=1/\ell_{uv}$1 (Devriendt et al., 2022). In cycles ($w_{uv}=1/\ell_{uv}$2), $w_{uv}=1/\ell_{uv}$3 for all $w_{uv}=1/\ell_{uv}$4.

2.2 Edge (Ricci-Type) Curvature

For an edge $w_{uv}=1/\ell_{uv}$5, several forms exist:

• Ricci–Foster curvature (primary in (Dawkins et al., 2024)):

$w_{uv}=1/\ell_{uv}$6

• Alternatively, averaging vertex scalar curvatures:

$w_{uv}=1/\ell_{uv}$7

Both definitions measure the deviation of the effective (electrical) distance from the local degree and are normalized to facilitate comparisons across network topologies (Devriendt et al., 2022, Yadav et al., 26 Oct 2025).

2.3 Matrix-based and Spectral Formulations

Resistance-based (scalar) curvature also admits a global definition: let $w_{uv}=1/\ell_{uv}$8 be the effective resistance matrix with entries $w_{uv}=1/\ell_{uv}$9. Then the curvature vector $L$0 solves the linear system

$L$1

That is, $L$2, and $L$3 is interpreted as the "curvature" at vertex $L$4 (Sun et al., 29 Apr 2025, Devriendt et al., 2023). In constant-curvature graphs (e.g., cycles, complete graphs), all $L$5 coincide, and this value directly controls spectral and diameter bounds.

3. Theoretical Properties and Curvature Evolution

3.1 Structural and Combinatorial Results

Foster’s Theorem: The sum of all edge Ricci–Foster curvatures $L$6, regardless of edge weights (Dawkins et al., 2024).

Sum rules for vertex curvature: $L$7; this follows from the spanning tree structure or the total expected degree in a random tree (Devriendt, 2024, Devriendt et al., 2022).

Degree bounds: For all $L$8, $L$9 (Devriendt et al., 2022).

Extremal graphs:

• Cycles $L_{uu} = \sum_{v \sim u} w_{uv}$0 minimize constant resistance curvature: $L_{uu} = \sum_{v \sim u} w_{uv}$1 (Sun et al., 29 Apr 2025).
• Complete graphs $L_{uu} = \sum_{v \sim u} w_{uv}$2 maximize: $L_{uu} = \sum_{v \sim u} w_{uv}$3.

Curvature and diameter/spectral gap: Graphs with resistance curvature bounded below by $L_{uu} = \sum_{v \sim u} w_{uv}$4 enjoy upper bounds on diameter and lower bounds on spectral gaps (Lichnerowicz), e.g., $L_{uu} = \sum_{v \sim u} w_{uv}$5 (Devriendt et al., 2023).

Curvature and random walks: Commute times are proportional to resistance, and mixing times admit quantitative estimates in terms of minimum resistance curvature.

3.2 Ricci–Foster Curvature Flow

The Ricci–Foster flow evolves edge resistances $L_{uu} = \sum_{v \sim u} w_{uv}$6 by

$L_{uu} = \sum_{v \sim u} w_{uv}$7

Positive curvature edges contract, negative expand, and flat edges remain unchanged (Dawkins et al., 2024). Existence and uniqueness of short-time solutions are guaranteed; the flow preserves nonnegative curvature—if initially $L_{uu} = \sum_{v \sim u} w_{uv}$8, then $L_{uu} = \sum_{v \sim u} w_{uv}$9 throughout the evolution.

On a regular $L_{uv} = -w_{uv}$0-cycle ($L_{uv} = -w_{uv}$1), $L_{uv} = -w_{uv}$2 for all $L_{uv} = -w_{uv}$3, with curvature $L_{uv} = -w_{uv}$4 remaining constant. This serves as a discrete analogue of the Einstein condition in smooth geometry.

4. Computational Aspects and Algorithmic Implications

4.1 Matrix Computation

• Effective resistance matrix: $L_{uv} = -w_{uv}$5 via Laplacian pseudoinverse, or $L_{uv} = -w_{uv}$6 with fast Laplacian-system solvers for large sparse graphs.
• Local curvature aggregation: Once $L_{uv} = -w_{uv}$7 is computed for relevant edges, $L_{uv} = -w_{uv}$8 and $L_{uv} = -w_{uv}$9 evaluation is $u \neq v$0 (Devriendt et al., 2022, Fei et al., 3 Nov 2025).
• Community detection: Iterative Ricci–Foster curvature flow plus Gaussian Mixture Model separation efficiently detects planted communities with greater speed than Ollivier–Ricci based flows, scaling nearly linearly in the number of edges (Onuchin et al., 12 Nov 2025).

4.2 Extensions

Resistance-based curvatures extend to:

• Directed graphs: via symmetrized Laplacian pseudoinverses and generalized Fiedler–Bapat identities (Kajiura et al., 31 Mar 2026).
• Signed graphs: providing a strict-negative-type metric, and hence a rigorous resistance-based geometry for signed interaction networks.
• Product graphs: explicit additivity and averaging rules for Cartesian products, with known phenomena in grid graphs (e.g., negative interior curvature, positive boundary curvature in $u \neq v$1 grids) (Dawkins et al., 2024).

5. Relation to Other Discrete Curvature Notions

• Ollivier–Ricci curvature: Based on optimal transport and Wasserstein distance between vertex-centered probability measures. The resistance-based Ricci curvature satisfies $u \neq v$2, with equality on bridges (Devriendt et al., 2022, Fei et al., 3 Nov 2025).
• Forman–Ricci curvature: Combinatorially $u \neq v$3 for unit weights; the resistance curvature always bounds below Forman–Ricci on edges and interpolates between Forman–Ricci and Ollivier–Ricci (Devriendt et al., 2022).
• Combinatorial curvature (angular defect): The resistance curvature at a vertex coincides with the expectation of combinatorial curvature over random spanning trees.
• Spectral/metric grounding: Resistance-based curvature is fundamentally spectral/combinatorial, encoding global connectivity and random walk properties, unlike transport-based curvatures which focus on local measure comparison.

6. Applications and Open Directions

• Topological and combinatorial graph theory: Resistance curvature characterizes Hamiltonicity, toughness, and matching properties, and determines extremal properties of graphs (e.g., minimum constant curvature graphs are cycles) (Devriendt, 2024, Sun et al., 29 Apr 2025).
• Geometric data analysis: Curvature-based pipelines for clustering, manifold learning, and network comparison leverage resistance curvature for capturing both local and global geometry (Yadav et al., 26 Oct 2025, Fei et al., 13 Jan 2026).
• Graph neural networks and learning frameworks: Effective resistance curvature provides geometric priors for GNN architectures, enabling structure-aware message passing, enhancing robustness, and matching Ollivier–Ricci's expressiveness at vastly lower computational cost (Fei et al., 3 Nov 2025).
• Community detection and geometric flows: Algorithms based on Ricci–Foster flow and resistance curvature gradient descent offer highly scalable methods for unsupervised structure discovery (Onuchin et al., 12 Nov 2025, Fei et al., 13 Jan 2026).
• Physical models: Resistance-based concepts appear in modeling transport properties in multiphase systems, including the curvature dependence of interfacial resistance in nucleation theory (Glavatskiy et al., 2013).
• Structural graph geometry: The characterization of graphs with nonnegative (or constant/positive) resistance curvature reveals strong links to spanning-tree polytopes, matching polytopes, and convex polytopal structure, with positive resistance curvature being rare and highly constrained (Devriendt, 2024, Loera et al., 13 Oct 2025).

Open problems include sharpening spectral and combinatorial characterizations, understanding curvature behavior in product and higher-dimensional graph families, expanding theory for signed/directed/time-evolving structures, and further integrating resistance curvature into deep graph learning pipelines (Dawkins et al., 2024, Fei et al., 13 Jan 2026, Yadav et al., 26 Oct 2025).

---

Resistance-based curvature stands as a central, unifying concept in discrete differential geometry and geometric network science, offering a rigorous, computable, and geometrically transparent framework that directly ties combinatorial, spectral, and probabilistic phenomena on graphs (Dawkins et al., 2024, Devriendt et al., 2022, Sun et al., 29 Apr 2025, Devriendt, 2024, Devriendt et al., 2023, Yadav et al., 26 Oct 2025, Fei et al., 3 Nov 2025, Fei et al., 13 Jan 2026, Onuchin et al., 12 Nov 2025).