Pith. sign in

REVIEW 4 major objections 4 minor 1 cited by

Lin-Lu-Yau Ricci curvature on hypergraphs

T0 review · 4 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read This paper defines Lin-Lu-Yau Ricci curvature on undirected and directed hypergraphs as the α→1 limit of normalized Ollivier curvature, and proves a Bonnet-Myers diameter bound from positive curvature.

desk verdict The paper's headline hyperedge curvature is ill-defined because kappa_1(h) is negative for hyperedges of size >2, making the defining limit blow up, though the pairwise curvature and directed extension are worth saving. read the letter →

arxiv 2507.04109 v1 pith:K3PI2HE5 submitted 2025-07-05 math.DG

classification math.DG MSC 05C6505C1005C81
keywords Lin-Lu-YauRiccicurvaturehypergraphOllivierBonnet-Myerstheoremdirected1-Wassersteindistancerandomwalkbound
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper sets out to define Lin-Lu-Yau (LLY) Ricci curvature for hypergraphs, covering both undirected and directed hyperedges, through the same random-walk and optimal-transport route that yields LLY curvature on graphs. The curvature of a hyperedge $h$ is the limit, as a laziness parameter $\alpha$ tends to 1, of the normalized Ollivier curvature $\kappa_\alpha(h)/(1-\alpha)$, where $\kappa_\alpha$ compares the Wasserstein distance between $\alpha$-lazy random walks from the vertices of $h$ with the hyperedge length $L(h)$. The paper argues that this limit is well defined and that positive lower bounds on the curvature force the diameter bound $\operatorname{diam}(H)\le 2\max_h w_h/\kappa$, an analogue of the Bonnet-Myers theorem. The payoff is a computable, definitionally unified curvature for higher-order network structures, where edges can connect more than two vertices at once.

What carries the argument

The carrying object is the normalized quotient $g(\alpha)=\kappa_\alpha/(1-\alpha)$ built from the $\alpha$-lazy random walk. At $\alpha=1$ the walk stays put; as $\alpha$ decreases, probability mass spreads through hyperedge incidence, so the Wasserstein distance between walks measures how much transport cost shrinks as laziness is removed. The paper's argument has two pillars: concavity of $\kappa_\alpha$ in $\alpha$, which it uses to claim that $g$ is increasing, and an upper bound $\kappa_\alpha\le C(1-\alpha)$, which bounds $g$; together these are meant to force a finite $\alpha\to 1$ limit. In the directed case, Kantorovich-Rubinstein duality enters through lower bounds on Wasserstein quasi-distances using 1-Lipschitz test functions.

What would settle it

Take three vertices $x_1,x_2,x_3$ joined by a single hyperedge $h$ of weight 1, so every pairwise distance is 1. Then at $\alpha=1$, $W_1(h)=W(\delta_{x_1},\delta_{x_2})+W(\delta_{x_1},\delta_{x_3})+W(\delta_{x_2},\delta_{x_3})=3$ and $L(h)=1$, so $\kappa_1(h)=1-3=-2$. By continuity, $\kappa_\alpha(h)/(1-\alpha)\to -\infty$ as $\alpha\to 1$, which is a concrete place where the claimed finite limit fails. Computing this one example settles the question.

Watch

Extended reading notes

Core claim

The paper's central claim is that Lin-Lu-Yau curvature is not confined to ordinary graphs: for a hyperedge $h$ in a weighted undirected hypergraph, the quotient $g_h(\alpha)=\kappa_\alpha(h)/(1-\alpha)$, where $\kappa_\alpha(h)=1-W_\alpha(h)/L(h)$ and $W_\alpha(h)$ is the sum of 1-Wasserstein distances between $\alpha$-lazy random walks from all pairs of vertices of $h$, has a finite limit $\kappa(h)=\lim_{\alpha\to 1}g_h(\alpha)$; the same limit with $d(u,v)$ in place of $L(h)$ defines $\kappa(u,v)$ for any vertex pair. For directed hypergraphs $h=(A_h,B_h)$, the analogous limit uses the Wasserstein quasi-distance between the input measure $\mu^\alpha_{A_h}$ and the output measure $\mu^\alpha_{B_h}$, normalized by $L(h)=\min_{x\in A_h,\,y\in B_h} d(x,y)$. Concavity of $\kappa_\alpha$ in $\alpha$ and linear upper bounds of order $1-\alpha$ are the supporting estimates, and on this definition the paper builds a Bonnet-Myers-type diameter bound, $\operatorname{diam}(H)\le 2\max_h w_h/\kappa$, plus a vertex-count bound for oriented hypergraphs.

Load-bearing premise

The load-bearing premise is that $\kappa_\alpha(h)/(1-\alpha)$ is increasing on $[0,1)$ and bounded above, so the limit at $\alpha=1$ is finite; this holds for vertex pairs and graph edges because $\kappa_1(h)=0$, but for hyperedges of size three or more $\kappa_1(h)$ is generally negative, which would make the quotient diverge instead.

Editorial extensions

If this is right

  • If the limit exists, LLY curvature of a hyperedge is directly computable from the random-walk couplings, offering a practical alternative to definitions based on Kantorovich differences and submodular Laplacians.
  • A uniform lower bound $\kappa>0$ on the curvature of well-transported vertex pairs propagates to all vertex pairs, so checking a special class of pairs suffices.
  • Positive LLY curvature on every relevant hyperedge implies $\operatorname{diam}(H)\le 2\max_h w_h/\kappa$ for undirected hypergraphs, and the analogous bound holds for oriented hypergraphs.
  • For oriented hypergraphs with unit weights, the curvature bound yields a quantitative ceiling on the vertex count, $N\le 1+\sum_{k=1}^{\lfloor 2/\kappa\rfloor}\Delta^k\prod_{i=1}^{k-1}\frac{B_H}{1+B_H}(1+B_H-i\kappa)$.
  • Directed hyperedge curvature is not controlled above by the maximum curvature of the constituent directed edges, as the paper's example with $\kappa(h)=3/10$ and negative edge curvatures shows.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A reader extending the paper's logic should check the endpoint $\alpha=1$ before relying on the limit: for a hyperedge with at least three vertices, $\kappa_1(h)=1-W_1(h)/L(h)$, and $W_1(h)$ sums all pairwise vertex distances while $L(h)$ is their minimum, so $\kappa_1(h)$ is typically negative and $\kappa_\alpha(h)/(1-\alpha)$ then diverges to $-\infty$ as $\alpha\to 1$; the claimed monotonicity
  • A natural repair compatible with the graph case is to define $\kappa(h)$ as the slope $\lim_{\alpha\to 1}(\kappa_\alpha(h)-\kappa_1(h))/(1-\alpha)$, which reduces to the paper's quotient when $\kappa_1(h)=0$ and remains finite in the same examples.
  • Under that repair, the Bonnet-Myers bound would likely need an additional term depending on $\kappa_1(h)$, so the diameter estimate would not follow from the paper's current proof as stated.
  • The same endpoint check applies to directed hyperedges, because at $\alpha=1$ the cost is the Wasserstein distance between uniform distributions on the input and output sets, which need not equal the minimum pairwise distance $L(h)$.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The paper proposes a framework for Lin-Lu-Yau (LLY) Ricci curvature on undirected, directed, and oriented hypergraphs. It defines α-dependent Ollivier-type curvature κ_α for hyperedges and vertex pairs using hypergraph random walks and the 1-Wasserstein distance, then defines LLY curvature as the limit κ = lim_{α→1} κ_α/(1−α). The paper claims concavity and upper bounds make these limits exist, derives monotonicity and upper bounds on vertex-pair curvature, and proves Bonnet-Myers-type diameter and vertex-count estimates. The pairwise vertex-curvature part is a standard variant of the graph construction, but the hyperedge-curvature definition, which is the paper's central advertised contribution, is not well-posed: for hyperedges of size at least three the required limit is −∞ rather than a finite number.

Significance. If the hyperedge limit were well-defined, the paper would provide a computable, unified LLY-type curvature for hypergraphs together with Bonnet-Myers bounds, a potentially useful addition to the hypergraph-curvature literature. The paper also usefully emphasizes computability relative to the Laplacian-based definitions of earlier work, and the pairwise vertex-curvature analysis in Section 2, including the upper bound (4), appears sound. However, the central object of the paper — hyperedge curvature — is undefined as written, and the paper's own Example 2.4 contradicts the definition. The surviving pairwise results are standard variants rather than the advertised hyperedge theory, so the main claims of the abstract and introduction are not established.

major comments (4)
  1. [Section 2, Eq. (5) and the paragraph after Lemma 2.2] Concavity of κ_α(h) does not imply that g(α)=κ_α(h)/(1−α) is increasing on [0,1). For 0≤α<β<1, concavity gives g(β) ≥ g(α) + [(β−α)/((1−α)(1−β))] κ_1(h), which is an increase only if κ_1(h)≥0. At α=1, μ_1^{x_i}=δ_{x_i}, so W_1(h)=Σ_{i<j} d(x_{h_i},x_{h_j}) ≥ L(h)=min_{i<j} d(x_{h_i},x_{h_j}), and for a hyperedge with at least three distinct vertices the inequality is strict. Hence κ_1(h)=1−W_1(h)/L(h)<0. By continuity, κ_α(h)→κ_1(h)<0, so κ_α(h)/(1−α)→−∞ as α→1, not a finite LLY curvature. Lemma 2.2's upper bound cannot rescue the argument: an upper bound on a non-monotone function does not produce convergence. Thus Eq. (5) defines κ(h) only in the graph case |h|=2, not for hyperedges of size at least three.
  2. [Section 3, Lemmas 3.1–3.2 and Definition 3.5] The same logical gap invalidates Definition 3.5. The claim that concavity makes g(α) increasing is repeated after Lemma 3.1, but at α=1 the directed measures are μ_1^{A_h} and μ_1^{B_h}, uniform on A_h and B_h respectively, so W_1(μ_1^{A_h}, μ_1^{B_h}) ≥ L(h) and κ_1(h)≤0. Whenever the uniform coupling cannot be realized entirely at cost L(h) — which is the generic case, for example when the vertices realizing the minimum distance do not form a perfect matching with the uniform marginals — κ_1(h)<0 and the quotient in Eq. (20) diverges to −∞. The upper bound in Lemma 3.2 is beside the point: without monotonicity it does not establish a finite limit. The directed hyperedge curvature is therefore not well-posed for general directed hypergraphs.
  3. [Example 2.4] The example contradicts the definition. For h_1={x_1,x_2,x_3} with unit weights, all three pairwise distances are 1 because each pair lies in h_1, so L(h_1)=1 and W_1(h_1)=3. Therefore κ_1(h_1)=−2, and by Eq. (5) the limit κ(h_1)=lim_{α→1} κ_α(h_1)/(1−α) diverges to −∞. The text reports the finite value 5/6 for κ(h_1). No computation is shown for the displayed values, and they are not consistent with Definition 2.6. This concrete mismatch confirms that the reported numbers are not computations of the object defined in Eq. (5).
  4. [Theorems 2.6 and 3.6] Because hyperedge curvature is undefined, the statements that invoke κ(h) are vacuous or invalid. Theorem 3.6's bound L(h)≤2 max_h w_h/κ(h) presupposes a positive finite κ(h), which Definition 3.5 does not provide for directed hyperedges. The pair-curvature parts of Theorems 2.6 and 3.6 depend only on the pair upper bound (4) and the pair limit, and they remain plausible; however, this does not recover the paper's hyperedge-level claims, which are the stated main contribution.
minor comments (4)
  1. [Example 2.4] The displayed values are written as κ_α(x_2,x_3)=3/2, κ_α(x_1,x_2)=1/2, and κ_α(h_1)=5/6, but the text seems to intend the limiting κ rather than the α-dependent κ_α; the notation should be corrected and a full computation supplied.
  2. [Definition 2.4, Eq. (3)] The random-walk formula divides by |h′|−1, so the paper should state explicitly that all hyperedges have cardinality at least two; otherwise the Markov chain is undefined for singleton hyperedges.
  3. [Section 3, Proposition 3.1] Proposition 3.1 states only the inequality W(μ_α^{A_h}, μ_α^{B_h}) ≥ sup over 1-Lipschitz functions, whereas the undirected section uses the Kantorovich-Rubinstein equality; the paper should explain why the directed setting does not require the reverse inequality for the later limiting arguments.
  4. [Throughout] There are several typographical and grammatical errors that should be fixed, including 'Our results extends' in the abstract, 'the proof if completed' at the end of Theorem 3.6, and the inconsistent use of Hdi versus the arrow-notation in Section 3.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the derivation is self-contained; the well-posedness gap is a mathematical error, not a circular one.

full rationale

The paper's construction follows Ollivier and Lin-Lu-Yau: it defines a parameterized curvature κα via Eq. (2) and the random-walk measures (3), proves concavity (Lemma 2.1) and upper bounds (Lemma 2.2), and then takes the limit in (5) to define LLY curvature. No parameter is fitted to the quantities being predicted, and the example values are direct computations from the definition. The cited external results ([23], [30], [16]) are not by the present authors and are used as standard building blocks, not as a substitute for the paper's central argument. The main defect the reader identifies is that Lemma 2.1 plus Lemma 2.2 do not guarantee the limit exists: concavity alone does not make κα(h)/(1−α) increasing unless κα(1) ≥ 0, while for hyperedges of size greater than two, W1(h) ≥ L(h) with strict inequality typical, so κα(1) is usually negative and the quotient may diverge to −∞. This is a serious mathematical gap in the claimed well-posedness and the Bonnet–Myers-type conclusions depend on it, but it is not circular reasoning: the limit is not defined in terms of itself, nor is any conclusion imported from the paper's own prior work. The derivation chain is self-contained and not equivalent to its inputs by construction, so the circularity score is 0.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

No free parameters fitted to data and no newly postulated entities. The critical unsupported input is the monotonicity of the quotient used to define hyperedge curvature.

assumptions (3)
  • ad hoc to paper The quotient g(alpha) = kappa_alpha(h)/(1-alpha) is increasing on [0,1) for hyperedges, and this follows from concavity of kappa_alpha.
    Used to define hyperedge LLY curvature via a limit in Definition 2.6 and Definition 3.5. It is false for hyperedges of size greater than 2 because kappa_alpha(1) = 1 - W_1/L <= 0; the quotient is decreasing and its limit is -infinity.
  • domain assumption Every optimal hyperpath between x and y can be chosen so that each consecutive vertex pair is well-transported with d(x_{gamma_j}, x_{gamma_j+1}) = w_{h_{gamma_j}} and distances add along the path.
    Needed for Proposition 2.5, which extends a lower curvature bound from well-transported pairs to all pairs. This is not proved and may fail when a hyperedge's internal distance is smaller than its weight.
  • standard math Kantorovich-Rubinstein duality and convexity of the Wasserstein distance in the marginal pair hold on finite hypergraphs.
    Standard optimal transport results used in Propositions 2.1 and 3.1 and in the concavity lemmas.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Lin-Lu-Yau Ricci curvature on hypergraphs." pith.science (2026). https://pith.science/paper/K3PI2HE5

@misc{pith2026250704109,
  author       = {Pith},
  title        = {Pith review of: Lin-Lu-Yau Ricci curvature on hypergraphs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/K3PI2HE5}},
  note         = {Machine review of arXiv:2507.04109}
}
abstract

In this paper, we introduce a unified framework for defining Lin-Lu-Yau (LLY) Ricci curvature on both undirected and directed hypergraphs. By establishing upper bounds and monotonicity properties for the parameterized curvature $\kappa_\alpha$, we justify the well-posedness and compatibility of our definitions. Furthermore, we prove a Bonnet-Myers-type theorem for hypergraphs, which highlights the potential of LLY Ricci curvature in hypergraph analysis, particularly in studying geometric and structural properties. Our results extends the foundational definitions of graph Ricci curvature by Ollivier and Lin-Lu-Yau to the hypergraph setting.

Figures

Figures reproduced from arXiv: 2507.04109 by the authors.

Figure 1
Figure 1. The hypergraph H4 When investigating lower bounds of the LLY curvature for arbitrary vertex pairs {x, y}, it suffices to consider a special class of pairs defined below, which we term well-transported pairs. Definition 2.7. For a hypergraph Hun = (V, H,w), any pair of vertices {u, v} containing in a hyper￾edge h is called well-transported if the distance between u and v satisfies d(u, v) = wh. Proposition 2.5. Suppo… view at source ↗
Figure 2
Figure 2. The hypergraph Hxyz Example 3.3. Consider the hypergraph Hxyz shown in [PITH_FULL_IMAGE:figures/full_fig_p015_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Edge-connectivity and LLY curvature of hypergraphs

    math.CO 2026-08 conditional novelty 6.0 of 10

    For r-uniform linear hypergraphs with nonnegative Lin-Lu-Yau curvature, edge-connectivity equals minimum incidence degree, while nonlinear examples violate this with arbitrarily large gaps.

Reference graph

Works this paper leans on

35 extracted references · 31 canonical work pages · cited by 1 Pith paper

  1. [1]

    Akamatsu, A New Transport Distance and Its Associated Ricci Curvature of Hypergraphs, Analysis and Geometry in Metric Spaces 10 (1) (2022) 90–108

    T. Akamatsu, A New Transport Distance and Its Associated Ricci Curvature of Hypergraphs, Analysis and Geometry in Metric Spaces 10 (1) (2022) 90–108

  2. [2]

    Weak Kantorovich difference and associated Ricci curvature of hypergraphs

    T. Akamatsu, Weak Kantorovich di fference and associated Ricci curvature of hypergraphs, arXiv: 2306.14084

  3. [3]

    Signless Normalized Laplacian for Hypergraphs

    E. Andreotti, R. Mulas, Signless Normalized Laplacian for Hypergraphs, arXiv: 2005.14484 9

  4. [4]

    Asoodeh, T

    S. Asoodeh, T. Gao, J. Evans, Curvature of Hypergraphs via Multi-Marginal Optimal Transport, in: 2018 IEEE Confer- ence on Decision and Control (CDC), 1180–1185, 2018

  5. [5]

    J. Bai, B. Gong, Y . Zhao, F. Lei, C. Yan, Y . Gao, Multi-Scale Representation Learning on Hypergraph for 3D Shape Retrieval and Recognition, IEEE Transactions on Image Processing 30 (2021) 5327–5338

  6. [6]

    S. Bai, A. Huang, L. Lu, S.-T. Yau, On the sum of Ricci-curvatures for weighted graphs, Pure Appl. Math. Q. 17 (5) (2021) 1599–1617

  7. [7]

    Bauer, J

    F. Bauer, J. Jost, S. Liu, Ollivier–Ricci curvature and the spectrum of the normalized graph Laplace operator, Math. Res. Lett. 19 (6) (2012) 1185–1205

  8. [8]

    Benson, P

    B. Benson, P. Ralli, P. Tetali, V olume growth, curvature, and Buser-type inequalities in graphs, Int. Math. Res. Not. IMRN 2021 (22) (2021) 17091–17139

Show all 35 references
  1. [9]

    D. P. Bourne, D. Cushing, S.Liu, F.M ¨unch, N. Peyerimho ff, Ollivier–Ricci Idleness Functions of Graphs, SIAM J. Discrete Math. 32 (2) (2018) 1408–1424

  2. [10]

    Brezis, Remarks on the Monge–Kantorovich problem in the discrete setting, C

    H. Brezis, Remarks on the Monge–Kantorovich problem in the discrete setting, C. R. Acad. Sci. Paris S ´er. I 356 (2) (2018) 207–213

  3. [11]

    Brezis, J.-M

    H. Brezis, J.-M. Coron, E. H. Lieb, Harmonic maps with defects, Commun. Math. Phys. 107 (1986) 649–705

  4. [12]

    Coupette, S

    C. Coupette, S. Dalleiger, B. Rieck, Ollivier-Ricci curvature for hypergraphs: A unified framework, arXiv: 2210.12048

  5. [13]

    G. F. de Arruda, G. Petri, Y . Moreno, Social contagion models on hypergraphs, Phys. Rev. Res. 2 (2020) 023032

  6. [14]

    Devriendt, R

    K. Devriendt, R. Lambiotte, Discrete curvature on graphs from the e ffective resistance, J. Phys. Complex. 3 (2) (2022) 025008

  7. [15]

    M. Eidi, A. Farzam, W. Leal, A. Samal, J. Jost, Edge-based analysis of networks: curvatures of graphs and hypergraphs, Theory Biosci. 139 (4) (2020) 337–348

  8. [16]

    M. Eidi, J. Jost, Ollivier Ricci curvature of directed hypergraphs, Sci. Rep. 10 (1) (2020) 12466. 21

  9. [17]

    Y . Han, P. Wang, S. Kundu, Y . Ding, Z. Wang, Vision HGNN: An Image is More than a Graph of Nodes, in: 2023 IEEE/CVF International Conference on Computer Vision (ICCV), 19821–19831, 2023

  10. [18]

    Ikeda, Y

    M. Ikeda, Y . Kitabeppu, Y . Takai, T. Ueharam, Coarse Ricci curvature of hypergraphs and its generalization, arXiv: 2102.00698

  11. [19]

    J. Jost, S. Liu, Ollivier’s Ricci curvature, local clustering and curvature-dimension inequalities on graphs, Discrete Comput. Geom. 51 (2) (2014) 300–322

  12. [20]

    J. Jost, R. Mulas, Hypergraph Laplace operators for chemical reaction networks, Adv. Math. 351 (2019) 870–896

  13. [21]

    J. Jost, R. Mulas, Normalized Laplace operators for hypergraphs with real coe fficients, J. Complex Netw. 9 (1) (2021) cnab009

  14. [22]

    W. Leal, G. Restrepo, P. F. Stadler, J. Jost, Forman–Ricci curvature for hypergraphs, Adv. Complex Syst. 24 (01) (2021) 2150003

  15. [23]

    Y . Lin, L. Lu, S.-T. Yau, Ricci curvature of graphs, Tohoku Math. J. 63 (4) (2011) 605–627

  16. [24]

    Lin, S.-T

    Y . Lin, S.-T. Yau, Ricci curvature and eigenvalue estimate on locally finite graphs, Math. Res. Lett. 17 (2) (2010) 343–356

  17. [25]

    X. Liu, C. Song, S. Liu, M. Li, X. Zhou, W. Zhang, Multi-way relation-enhanced hypergraph representation learning for anti-cancer drug synergy prediction, Bioinformatics 38 (20) (2022) 4782–4789

  18. [26]

    J. Ma, Y . Yang, Evolution of weights on a connected finite graph, arXiv: 2411.06393

  19. [27]

    Montrucchio, G

    L. Montrucchio, G. Pistone, Kantorovich distance on a weighted graph, arXiv: 1905.07547 1420

  20. [28]

    Mulas, Sharp bounds for the largest eigenvalue, Math

    R. Mulas, Sharp bounds for the largest eigenvalue, Math. Notes 109 (1) (2021) 102–109

  21. [29]

    M ¨unch, R

    F. M ¨unch, R. K. Wojciechowski, Ollivier Ricci curvature for general graph Laplacians: heat equation, Laplacian com- parison, non-explosion and diameter bounds, Adv. Math. 356 (2019) 106759

  22. [30]

    Ollivier, Ricci curvature of Markov chains on metric spaces, J

    Y . Ollivier, Ricci curvature of Markov chains on metric spaces, J. Funct. Anal. 256 (3) (2009) 810–864

  23. [31]

    Ozawa, Y

    R. Ozawa, Y . Sakurai, T. Yamada, Geometric and spectral properties of directed graphs under a lower Ricci curvature bound, Calc. Var. Partial. Differ. Equ. 59 (2020) 1–39

  24. [32]

    K. M. Saifuddin, B. Bumgardner, F. Tanvir, E. Akbas, HyGNN: Drug-Drug Interaction Prediction via Hypergraph Neural Network, in: 2023 IEEE 39th International Conference on Data Engineering (ICDE), 1503–1516, 2023

  25. [33]

    Villani, Topics in optimal transportation, vol

    C. Villani, Topics in optimal transportation, vol. 58, American Mathematical Soc., 2021

  26. [34]

    Yamada, The Ricci curvature on directed graphs, J

    T. Yamada, The Ricci curvature on directed graphs, J. Korean Math. Soc. 56 (1) (2019) 113–125

  27. [35]

    J. L. Yu, H. Z. Yin, X. Xia, T. Chen, J. D. Li, Z. Huang, Self-supervised learning for recommender systems: A survey, IEEE Transactions on Knowledge and Data Engineering 36 (1) (2023) 335–355. 22

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.