Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

In Reverie Together: Ten Years of Mathematical Discovery with a Machine Collaborator

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read The authors present four open graph-theoretic conjectures generated by TxGraffiti over a decade—simple, sharp, and empirically validated—and they argue these are genuine products of human-machine collaboration.

desk verdict The paper's flagship claim is false: Conjecture 1 fails on K2, so the four open conjectures are not all open as stated. read the letter →

arxiv 2507.17780 v1 pith:YEJDSNEG submitted 2025-07-23 cs.DM cs.AImath.CO

classification cs.DMcs.AImath.CO MSC 05C6905C7005C07
keywords automatedconjecturingTxGraffitigraphtheoryindependencenumberzeroforcingmaximalmatchingharmonicindexopenproblems
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

These authors report four open conjectures in graph theory that were generated by TxGraffiti, an automated conjecturing program, and have resisted proof and counterexample for years. Each conjecture is a short inequality pairing natural graph invariants: independence number against a combination of annihilation number and residue; zero forcing number against independence number; independent domination number against minimum maximal matching; and minimum maximal matching against harmonic index. The paper's evidence is empirical—all four hold on the hundreds of graphs in the system's dataset, with equality on structured families—and the authors argue that their simplicity, sharpness, and durability make them worthwhile targets for human and machine mathematicians. If the conjectures are true, they would show that a machine's pattern search can produce statements that sit at the edge of current knowledge.

What carries the argument

The carrier of the argument is TxGraffiti's generate-and-filter loop: from precomputed tables of graph invariants, the system combines expressions symbolically and keeps only empirically valid inequalities, ranking them by sharpness—how often equality is attained in its dataset of hundreds of graphs. The four conjectures are the surviving objects, and each pairs invariants of distinct flavors: the independence number $\alpha$ with the degree-sequence invariants $a$ and $R$; the zero forcing number $Z$ with $\alpha$; the independent domination number $i$ with the minimum maximal matching number $\mu^*$; and $\mu^*$ with the harmonic index $H$. The paper also uses the identity $\mu^*(G)=i(L(G))$ to connect the matching-based parameters to domination in line graphs, and it treats equality examples as structural hints for future proofs.

What would settle it

Enumerate all connected simple graphs up to order 10 (or 11) and compute, for each, the invariants $\alpha$, $a$, $R$, $Z$, $i$, $\mu^*$, and $H$; a single graph violating any of the four inequalities would refute that conjecture, and for Conjecture 2 the restricted family $\Delta(G) \le 3$, $G \not\simeq K_4$ could be decided exactly by this census.

Watch

Extended reading notes

Core claim

At the center of the paper is the claim that four inequalities, each produced by TxGraffiti between 2016 and 2023, are genuine open problems. For a nontrivial connected graph $G$, the conjectures state: $\alpha(G) \ge (a(G)+R(G))/\Delta(G)$, and for connected graphs with $\Delta(G) \le 3$ other than $K_4$, $Z(G) \le \alpha(G)+1$; for $r$-regular graphs with $r>0$, $i(G) \le \mu^*(G)$; and for every nontrivial connected graph, $\mu^*(G) \le H(G)$. The authors give partial evidence: regular bipartite graphs and certain cubic graphs satisfy the first, claw-free cubic graphs satisfy the second, 2-regular graphs satisfy the third, and the fourth holds with equality on structured families in the dataset. They stress that every statement is simple, sharp, empirically valid on hundreds of graphs, and still open despite sustained attention.

Load-bearing premise

The case rests on the assumption that the hundreds of graphs in TxGraffiti's dataset are a representative test bed, so the absence of counterexamples there is meaningful evidence for the four inequalities, and that the authors' claim that the statements remain open is accurate.

Editorial extensions

If this is right

  • If Conjecture 1 is true, then every nontrivial connected graph has a lower bound on $\alpha(G)$ computable directly from its degree sequence, strengthening the classical $\alpha \ge R$ inequality.
  • If Conjecture 2 is true, then for connected subcubic graphs, the zero forcing number is determined within one of the independence number (except at $K_4$), giving a tight bridge between a linear-algebraic parameter and a packing parameter.
  • If Conjecture 3 is true, then in every regular graph the smallest maximal independent set is no larger than the smallest maximal matching, extending the known $\alpha(G) \le \mu(G)$ phenomenon from maximum packings to saturated packings.
  • If Conjecture 4 is true, then the discrete minimum maximal matching number is bounded by the continuous harmonic index for every connected graph, opening a quantitative link between chemical graph indices and domination-theoretic parameters.
  • Because all four are formalized in a proof assistant, any one of them being proved or refuted would give a concrete data point on whether machine-generated conjectures accelerate mathematical research.

Reading between the lines

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

  • If the four conjectures all hold, it would suggest that simple arithmetic combinations of classical invariants are an abundant source of open problems, and that the bottleneck will shift from generating conjectures to resolving them.
  • A natural stress test is to run the same inequality search on graph families the current dataset undersamples—highly irregular graphs for Conjecture 1, subcubic graphs with many degree-2 vertices for Conjecture 2, and high-degree regular graphs for Conjecture 3.
  • The authors' narrative implies a new criterion for evaluating conjecturing software: not only theorems proved, but the durability and perceived naturalness of the open problems it leaves behind. A longer-term test is whether these four conjectures, or descendants of them, eventually fall to human or machine proof.
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

3 major / 5 minor

Summary. The manuscript presents four graph-theoretic inequalities generated by the TxGraffiti automated conjecturing system between 2016 and 2023, describes them as open conjectures, gives partial validation for Conjecture 1, displays an empirical plot for Conjecture 4, and provides Lean 4 statements in an appendix. The paper also situates these conjectures in the longer history of TxGraffiti and lists previously published theorems that originated from the system. The central claims are that each conjecture is concise, grounded in natural graph invariants, empirically validated across hundreds of graphs, and resistant to both proof and counterexample despite extensive effort.

Significance. The project has a credible track record: Table 1 lists nine published or in-press theorems that originated from TxGraffiti, which gives the conjectures a genuine pedigree. If the four statements were correct and open, they would be attractive test cases for automated and human proving, and the Lean-translation appendix would add value. However, the mathematical core is currently compromised by a concrete counterexample to Conjecture 1, and the empirical support is largely circular and underdocumented. The paper needs correction and additional evidence before the central claims can be accepted.

major comments (3)
  1. [Section 2.1, Conjecture 1] As stated, Conjecture 1 is false. For G = K2, we have n = 2, m = 1, degree sequence [1,1], so a(G) = 1 (since 1 <= 1 but 2 > 1), R(G) = 1 (the Havel-Hakimi process sends [1,1] to [0]), Delta(G) = 1, and alpha(G) = 1. The claimed inequality reads 1 >= (1+1)/1 = 2, which is false. K2 is a nontrivial connected graph under the standard convention, and the statement contains no exclusions. The subsequent partial result in the same section explicitly assumes Delta(G) >= 2 in the regular bipartite case, suggesting the exceptional case was recognized. Please add the missing hypothesis (e.g., Delta(G) >= 2 or |V(G)| >= 3), update the abstract, Conjecture 1, and the Lean statement in Appendix A, and re-run the empirical validation under the corrected hypothesis.
  2. [Section 2.4, Figure 1] The empirical validation is selection-circular and not fully documented. The conjectures were generated by TxGraffiti from precomputed datasets, and Figure 1 validates Conjecture 4 on the same 335 graphs from the TxGraffiti dataset; a search that selects inequalities on the basis of holding on a dataset cannot be independently confirmed on that same dataset. The paper provides no dataset, no search bounds, and no validation counts for Conjectures 1-3. Please provide the dataset or a reproducible generator, report validation statistics for all four conjectures, and include an out-of-sample check (for example, exhaustive enumeration of all connected graphs up to order 10, or random graphs held out from the generation process).
  3. [Appendix A, Listing 1] The Lean statements do not match the conjectures stated in the text. In conjecture_two, hypotheses h2 and h3 require max_degree G = min_degree G and max_degree G = 3, which restricts the statement to cubic graphs, whereas Conjecture 2 is for all connected graphs with Delta(G) <= 3. In conjecture_one, h2 : order G >= 1 is weaker than the intended nontrivial condition (order at least 2) and omits the Delta(G) >= 2 hypothesis needed to avoid the K2 counterexample. As written, the appendix does not support the claim that the conjectures have been accurately translated into Lean 4.
minor comments (5)
  1. [Section 2.1] The term 'nontrivial connected graph' is ambiguous; define it explicitly (e.g., |V(G)| >= 2, or after the revision, |V(G)| >= 3 with Delta(G) >= 2) so that the K2 case is treated unambiguously.
  2. [Section 2.1] The term 'König-Egerváry graph' is used without definition or reference; add a one-sentence definition or citation.
  3. [Section 2.3] The phrase 'If G is an r-regular graph G with r > 0' contains a duplicated 'G'; rephrase to 'If G is an r-regular graph with r > 0'.
  4. [Figure 1] The axes appear to show only numeric scales without clear titles; add explicit axis titles for H(G) and mu*(G) to make the plot self-contained.
  5. [Title page] The title contains a stray space in 'Ten Y ears'; correct it to 'Ten Years'.

Circularity Check

2 steps flagged · score 6.0 of 10

Same-dataset validation is presented as empirical support; the conjectures are outputs of the very fitting process used to 'validate' them.

  1. fitted input called prediction [Section 2.1 (Conjecture 1 selection criterion)]
    "At the time, candidate inequalities were ranked solely by sharpness, measured as the number of graphs in the dataset for which the inequality held with equality."

    This sentence states TxGraffiti's selection rule for Conjecture 1: the inequality was chosen because it held with equality on many graphs in the TxGraffiti dataset. The abstract then presents the conjecture as 'empirically validated across hundreds of graphs.' Since the selection and the validation use the same precomputed dataset, any inequality that survived the sharpness ranking is guaranteed, by construction, to hold on that dataset. Reporting this as empirical validation is therefore a check of the fitting data, not independent evidence; no hold-out dataset is provided.

  2. fitted input called prediction [Section 2.4, Figure 1 caption]
    "Figure 1 illustrates the relationship between µ∗(G) and H(G) across 335 simple connected graphs in the TxGraffiti dataset. ... All data points lie on or below the red dashed line H(G) = µ∗(G)."

    Conjecture 4 was generated by TxGraffiti, which operates on 'precomputed tabular data of graph invariants' and searches for inequalities that are 'empirically valid' on that data. The figure plots the same 335 TxGraffiti graphs used in that search and observes that all points lie below the line. Thus the plot restates the search filter that selected the conjecture. The caption calls this 'Empirical validation,' but because generation and evaluation share the same dataset, the evidence is circular by construction and there is no independent test set.

full rationale

The paper's circularity is confined to its empirical-validation claims. TxGraffiti generated each conjecture by searching its precomputed dataset for inequalities that held, often ranking by sharpness or equality count on that dataset. The paper then reports these same-data observations as 'empirically validated across hundreds of graphs.' For Conjecture 1, the selection criterion is explicitly the number of dataset graphs with equality; for Conjecture 4, Figure 1 plots the very same TxGraffiti dataset used for generation. This is fitted-input-as-validation: the evidence is forced by the selection procedure. The open-problem status and the mathematical content are not themselves circular—the conjectures could still be true or false—but the paper's only empirical support reduces to a restatement of the data-mining filter. The self-citations in Table 1 are to peer-reviewed publications and are not load-bearing in the same way; they provide external support for earlier TxGraffiti outputs. Separately, Conjecture 1 appears false for K2 under the paper's own definitions; that is a correctness concern, not a circularity one, and does not affect the score. Overall, partial circularity in the central empirical-support claim warrants a 6.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

No free parameters appear in the conjectures themselves; however, the selection and validation process relies on the TxGraffiti dataset and the authors' assertion of openness, which are ad hoc to this paper.

assumptions (4)
  • standard math Standard graph theory invariants (independence number, annihilation number, residue, zero forcing number, independent domination number, saturation number, harmonic index) are well-defined for finite simple graphs.
    The paper relies on these definitions from cited literature.
  • standard math The Havel-Hakimi theorem states that a sequence is graphic if and only if the process terminates in all zeros.
    Invoked in the definition of residue (Section 2.1).
  • domain assumption The authors' claim that these conjectures have resisted proof and counterexample, based on extensive collaborations and literature knowledge.
    This is not independently verifiable from the paper and underlies the 'open' status.
  • ad hoc to paper The TxGraffiti dataset is representative of graph structures of interest.
    The empirical validation and selection are based on this dataset; no independent external test set is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of In Reverie Together: Ten Years of Mathematical Discovery with a Machine Collaborator." pith.science (2026). https://pith.science/paper/YEJDSNEG

@misc{pith2026250717780,
  author       = {Pith},
  title        = {Pith review of: In Reverie Together: Ten Years of Mathematical Discovery with a Machine Collaborator},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YEJDSNEG}},
  note         = {Machine review of arXiv:2507.17780}
}
read the original abstract

We present four open conjectures in graph theory generated by the automated conjecturing system \texttt{TxGraffiti}. Each conjecture is concise, grounded in natural graph invariants, and empirically validated across hundreds of graphs. Despite extensive effort, these statements remain unresolved--defying both proof and counterexample. They are not only mathematical challenges but creative expressions--born of symbolic pattern recognition and mathematician-defined heuristics, refined through years of human dialogue, and now offered back to the community as collaborative artifacts. These conjectures invite not only formal proof, but also reflection on how machines can evoke wonder, spark curiosity, and contribute to the raw material of discovery. By highlighting these problems, we aim to inspire both human mathematicians and AI systems to engage with them--not only to solve them, but to reflect on what it means when machines participate meaningfully in the creative process of mathematical thought.

Figures

Figures reproduced from arXiv: 2507.17780 by the authors.

Figure 1
Figure 1. Empirical validation of the conjectured inequality [PITH_FULL_IMAGE:figures/full_fig_p006_1.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. A counterexample to the zero forcing versus independence conjecture for cubic and subcubic graphs

    math.CO 2026-07 accept novelty 7.0 of 10

    Explicit connected subcubic and cubic graphs attain Z = α + 2, refuting the TxGraffiti conjecture that Z ≤ α + 1 for connected graphs with Δ ≤ 3.

Reference graph

Works this paper leans on

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

  1. [1]

    Barioli, W

    F. Barioli, W. Barrett, S. Butler, S. M Cioabă, D. Cvetković, S. M. Fallat, C. Godsil, W. Haemers, L. Hogben, R. Mikkelson, S. Narayan, O. Pryporova, I. Sciriha, W. So, D. Stevanović, H. van der Holst, K. Vander Meulen, and A. Wangsness Wehe. Zero forcing sets and the minimum rank of graphs. Linear Algebra Appl., 428 0 (7): 0 1628--1648, 2008

  2. [2]

    Brimkov, R

    B. Brimkov, R. Davila, H. Schuerger, and M. Young. On a conjecture of T x G raffiti: Relating zero forcing and vertex covers in graphs. Discrete Appl. Math., 359: 0 290--302, 2024. doi:10.1016/j.dam.2023.11.009

  3. [3]

    Caro and R

    Y. Caro and R. Pepper. Degree sequence index strategy. Australas. J. Comb., 59 0 (1): 0 1--23, 2014

  4. [4]

    Y. Caro, R. Davila, M. A. Henning, and R. Pepper. Conjectures of T x G raffiti: Independence, domination, and matchings. Australas. J. Comb, 84 0 (2): 0 258--274, 2022 a

  5. [5]

    Y. Caro, R. Davila, and R. Pepper. New results relating independence and matchings. Discuss. Math. Graph Theory, 42 0 (3): 0 921--935, 2022 b . doi:10.7151/dmgt.2353

  6. [6]

    R. Davila. Total and Zero Forcing in Graphs. Ph D D issertation, University of Johannesburg, 2019

  7. [7]

    R. Davila. Another conjecture of T x G raffiti concerning zero forcing and domination in graphs. arXiv preprint arXiv:2406.19231, 2024 a . https://arxiv.org/abs/2406.19231

  8. [8]

    R. Davila. The optimist: Towards fully automated graph theory research. arXiv preprint arXiv:2411.09158, 2024 b

Show all 32 references
  1. [9]

    R. Davila. T x G raffiti: I nteractive conjecturing website. https://txgraffiti.streamlit.app, 2025. Accessed: July 20, 2025

  2. [10]

    Davila and J

    R. Davila and J. Eddy. Tx G raffiti : A python package for automated mathematical conjecturing. https://pypi.org/project/txgraffiti, 2025. Version 0.4.1

  3. [11]

    Davila and M

    R. Davila and M. A. Henning. Total forcing versus total domination in cubic graphs. Applied Mathematics and Computation, 354: 0 385--395, 2019. doi:10.1016/j.amc.2019.02.020

  4. [12]

    Davila and M

    R. Davila and M. A. Henning. Zero forcing in claw-free cubic graphs. Bull. Malays. Math. Sci. Soc., 43: 0 673--688, 2020. doi:10.1007/s40840-019-00800-w

  5. [13]

    Davila and M

    R. Davila and M. A. Henning. Relating zero forcing and domination in cubic graphs. J. Comb. Optim., 41: 0 553--577, 2021. doi:10.1007/s10878-020-00573-5

  6. [14]

    Davila, H

    R. Davila, H. Schuerger, and B. Small. A characterization of claw-free graphs using zero forcing invariants. J. Graph Theory, (accepted with revisions) 2025. arXiv preprint at https://doi.org/10.48550/arXiv.2412.03463

  7. [15]

    DeLaViña

    E. DeLaViña. Graffiti.pc: A variant of G raffiti. In Graphs and Discovery, DIMACS Ser. Discret. Math. Theor. Comput. Sci., volume 69, pages 71--79, Providence, RI, 2005 a . Amer. Math. Soc

  8. [16]

    DeLaViña

    E. DeLaViña. Some history of the development of graffiti. In Graphs and Discovery, DIMACS Ser. Discret. Math. Theor. Comput. Sci., volume 69, pages 81--118, Providence, RI, 2005 b . Amer. Math. Soc

  9. [17]

    DeLaViña, Q

    E. DeLaViña, Q. Liu, R. Pepper, B. Waller, and D. B. West. Some conjectures of G raffiti.pc on total domination. Congr. Numer., 185: 0 pp. 81--95, 2007

  10. [18]

    Du and B

    Z. Du and B. Zhou. Sharp bounds for the harmonic index of graphs. Applied Mathematics Letters, 26 0 (5): 0 464--467, 2013

  11. [19]

    Fajtlowicz

    S. Fajtlowicz. On the conjectures of G raffiti III . Congr. Numer., 66: 0 23--32, 1988

  12. [20]

    Fajtlowicz

    S. Fajtlowicz. On conjectures of G raffiti IV . Congr. Numer., pages 231--240, 1990

  13. [21]

    Favaron, M

    O. Favaron, M. Mah \'e o, and J.-F. Sacl \'e . On the residue of a graph. J. Graph Theory, 15: 0 39--64, 1991

  14. [22]

    J. R. Griggs and D. J. Kleitman. Independence and the havel--hakimi residue. Discrete Math., 127: 0 241--249, 1999

  15. [23]

    S. L. Hakimi. On the realizability of a set of integers as degrees of the vertices of a linear graph, i. SIAM J. Appl. Math., 10: 0 496--506, 1962

  16. [24]

    V. Havel. A remark on the existence of finite graphs. Casopis Pest. Mat., 80: 0 477--480, 1955. (in Czech)

  17. [25]

    M. He, H. Li, N. Song, and S. Ji. The zero forcing number of claw-free cubic graphs. Discrete Appl. Math., 359: 0 321--330, 2024. ISSN 0166-218X. doi:https://doi.org/10.1016/j.dam.2024.08.011. URL https://www.sciencedirect.com/science/article/pii/S0166218X24003639

  18. [26]

    Hogben, J

    L. Hogben, J. C.-H. Lin, and B. L. Shader. Inverse Problems and Zero Forcing for Graphs, volume 270 of Mathematical Surveys and Monographs. American Mathematical Society, 2022. ISBN 978-1-4704-6655-8. URL https://bookstore.ams.org/surv-270

  19. [27]

    Ilić and D

    A. Ilić and D. Stevanović. On the harmonic index of graphs. MATCH Commun. Math. Comput. Chem., 65 0 (3): 0 731--744, 2011

  20. [28]

    R. Pepper. Bounding the independence number of a graph using its annihilation number. Bull. Inst. Combin. Appl., 41: 0 89--94, 2004

  21. [29]

    R. Pepper. On the annihilation number of a graph. In Proceedings of the 15th American Conference on Applied Mathematics (ACAM), volume 1, pages 217--220. WSEAS, 2009

  22. [30]

    Schuerger, N

    H. Schuerger, N. Warnberg, and M. Young. Zero forcing and vertex independence number on cubic and subcubic graphs, 2024. URL https://doi.org/10.48550/arXiv.2410.21724. arXiv:2410.21724 [math.CO]

  23. [31]

    Tavakoli and B

    M. Tavakoli and B. Samadi. On the saturation number of graphs and their line graphs. Util. Math., 119: 0 241--256, 2022

  24. [32]

    E. Triesch. Degree sequences of graphs and dominance order. J. Graph Theory, 22: 0 89--93, 1996

Pith tools

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