Pith. sign in

REVIEW 5 major objections 6 minor 1 cited by

The 11-loop graph cohomology

T0 review · 5 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The 11-loop graph cohomology is computed, and the vanishing of a key bracket refutes a strong conjecture.

desk verdict The 11-loop computation is solid and worth taking seriously; the counterexample to Brown's strong conjecture is the headline, but it leans on an unpublished spectral-sequence vanishing statement that the paper does not reprove. read the letter →

arxiv 2508.13724 v1 pith:VJNCTZQL submitted 2025-08-19 math.QA math.AT

Thomas Willwacher This is my paper · ORCID
classification math.QAmath.AT
keywords graphcohomologylooporder11spectralsequenceGrothendieck-TeichmüllerLiealgebratriconnectedcomplexrankcomputationfinitefieldfreeconjecture
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

The paper computes the rational cohomology of the graph complex at loop order 11, for all degrees when $n=3$ and for degrees 5 through 8 when $n=2$, and fixes the remaining low-degree entries by spectral-sequence and degree arguments. A sympathetic reader should care because the graph complex is a known source of invariants in knot theory and of the Grothendieck-Teichmüller Lie algebra, so exact dimensions constrain those areas. The paper's main negative claim is that the bracket $[\sigma_3,X_{10}]$ vanishes in $H^7(GC_2)$ at loop order 13, contradicting the strong form of a conjecture that predicted a larger free Lie algebra generated by such classes. If the computation stands, that strong conjecture is false, and the Lie algebra structure of graph cohomology is more complicated than the conjecture allowed.

What carries the argument

The driving object is the loop-order spectral sequence of the twisted graph complex $(GC_0, \delta+[L_1,-])$, whose first page is $H(GC_0)$ and which converges to the one-dimensional span of the loop $L_1$. The paper relies on the result that the differential on pages 1, 3, and 5 vanishes, so a class can cancel only against specific partners; this bookkeeping fixes the low-degree entries in Table 1 once neighboring cohomology groups are known. For the counterexample, the same twisted complex supplies explicit lifts $\hat L_{4k+1}$ and $\hat X_{10}$, and the proof that $[\sigma_3,\hat X_{10}]$ is a $\delta$-cocycle that can be extended to a $\delta+\nabla$-cocycle $\hat Z$ is what makes the bracket vanish.

What would settle it

Evaluate the rank of the differential on the triconnected graph complex at the exact spot $(g,k)=(11,6)$ for $n=2$ over the rationals (or over a second large prime); if the rank exceeds the reported finite-field value, the entry $H^6(GC_2^{11})=1$ would drop to 0. For the counterexample, compute $H^7(GC_2^{13})$ independently over another prime; a nonzero class would contradict Proposition 6.1.

Watch

Extended reading notes

Core claim

At loop order 11 the cohomology of the commutative graph complex is as listed in Tables 1 and 2: for $GC_2$ the nonzero dimensions are $H^0\cong \mathbb{Q}^2$, $H^3\cong\mathbb{Q}^2$, and $H^6\cong\mathbb{Q}$, with all other computed degrees zero; for $GC_3$ the top-degree entries $H^{-3}$ follow the known sequence up to loop order 14, and the entries at $(11,-6)$ and $(11,-7)$ are upper bounds of dimension 7 and 1 that would drop to 6 and 0 if the rank bounds are not tight. The paper also proves $[\sigma_3,X_{10}]=0$ in $H^7(GC_2^{13})$, where $\sigma_3$ is the degree-zero generator and $X_{10}$ the nonzero class at $(g,k)=(10,7)$. This vanishing is shown by extending the two classes to elements of a twisted graph complex whose combined differential cancels exactly, so the bracket is a boundary; the strong version of the conjecture that these classes freely generate a larger Lie algebra is therefore false.

Load-bearing premise

The results stand or fall on the claim that the spectral sequence of the twisted graph complex converges to the expected answer and that its differential is zero on pages 1, 3, and 5; if those properties fail, the degree-counting arguments that pin down several table entries and the vanishing of $[\sigma_3,X_{10}]$ would no longer be valid.

Editorial extensions

If this is right

  • The 11-loop row of the even graph cohomology table is now entirely determined: $H^0$ is 2-dimensional, $H^3$ is 2-dimensional, $H^6$ is 1-dimensional, and all other computed degrees vanish.
  • The top-degree cohomology of the odd complex $H^{-3}(GC_3)$ is known exactly through loop order 14, because the new upper bounds match known lower bounds.
  • The strong free-generation conjecture is false: $\sigma_3$ and $X_{10}$ do not generate a free Lie subalgebra, since their bracket vanishes at loop order 13.
  • In the spectral sequence, $X_{10}$ survives to the $E_4$ page and kills $[\sigma_5,\sigma_9]$ in degree $(14,0)$, which forces $H^3(GC_2^{12}) \cong \mathbb{Q}^2$.

Reading between the lines

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

  • The proof of the vanishing bracket suggests a general descent mechanism: a degree-zero class bracketed with a higher-loop class often lands, after the spectral sequence, in a higher degree at a larger loop order rather than dying immediately; this could be tested by computing $H^3(GC_2^{15})$, where the paper itself points as the natural landing spot.
  • If Conjecture 5.1 holds for all $k$, the classes $X_{4k+2}$ form an infinite periodic family, and the spectral-sequence bookkeeping would force a predictable cascade of cancellations; checking the next member $X_{14}$ at loop order 14 would sharpen that picture.
  • The finite-field rank computations yield exact rational dimensions only when the resulting upper bound is zero; running the same GPU-accelerated rank algorithm at loop order 12 and beyond could settle Conjecture 3.7 and decide whether the top-degree upper bounds for $g=13,14$ are tight.
  • A repaired version of the disproved conjecture may exist in which the generators are replaced by their descendants in higher loop order; the paper leaves the formulation open, and the 11-loop data give a concrete target for such a revision.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 6 minor

Summary. The paper reports numerical computations of Kontsevich graph cohomology H(GC_n^{g-loop}) for loop order g = 11, with some additional higher-loop entries, and uses these results to discuss a conjecture of Brown. The main tables give dimensions for n = 2 and n = 3; the top-degree entries for g = 11–14 come from upper bounds via a barrel-graph reduction, and several low-degree entries for n = 2 are derived from a spectral sequence. The paper's headline theoretical claim is Proposition 6.1, which asserts that [σ3, X10] = 0 in H^7(GC_2^{13-loop}), providing a counterexample to a strong version of Brown's conjecture.

Significance. If the computations and the spectral-sequence input are correct, this is the first determination of the 11-loop graph cohomology and provides a concrete counterexample to a widely discussed conjecture, which would be of real interest to the graph complex community. The paper is careful in several respects: it distinguishes exact zero entries from upper bounds, explains why zero upper bounds are exact, and makes the dependence on finite-field rank computations explicit. However, the central theoretical and numerical claims depend on unpublished results and on a proof sketch, so the significance is presently conditional.

major comments (5)
  1. [§3, Proposition 3.2] Proposition 3.2 is load-bearing for the entire top-degree computation, since Corollary 3.4 and all upper bounds in Figure 1 rely on it, yet its proof is only a sketch. The final case is self-referential: Case 6 states that after using IHX relations to move all vertices onto the base loop, 'the graph is necessarily of the form of Case 6.' Case 3 similarly asserts a reduction of k without specifying the induction. This needs to be replaced by a complete argument or by a machine-checked verification; without it, the new top-degree entries for g = 11–14 are not established.
  2. [§6, Proposition 6.1] The final step of the proof of Proposition 6.1 ('not possible by degree reasons') silently uses the vanishing of differentials for i = 1, 3, 5 from [20, Corollary 4.4] and the convergence statement from [14]. The i = 1 case is essential: if the differential from bidegree (12,8) were nonzero, an unknown class in H^8(GC_2^{12-loop}) (a question-mark entry in Table 1) could kill Z, and the conclusion would not follow. [20] is an unpublished preprint by the same author, and neither the convergence theorem nor the specific vanishing differentials are proved or independently verified in this paper. Please supply a proof or an independent check of the needed vanishing before the counterexample can be considered established.
  3. [§5, Corollary 5.2] Corollary 5.2 is used in Section 2 to derive H^3(GC_2^{12-loop}) ≅ Q^2, but its proof relies on the sentence 'By computer calculation we also checked that it holds for k = 2,' with no details, code, or reproducibility data. Since this corollary is a new input to the table, the computation behind it should be documented, or the derivation should be made independent of it.
  4. [§4, Table 2] The entries (11,-6) and (11,-7) in Table 2 are explicitly only upper bounds: the text says that if they are not tight, the dimensions would be 6 and 0 rather than 7 and 1. This is incompatible with the abstract's unqualified statement that the paper computes the graph cohomology in loop order 11. The table and abstract should state that the loop order 11 computation is complete only up to two upper bounds, or the missing lower bounds should be supplied.
  5. [§6, first paragraph] The paper never states the strong version of Brown's conjecture that it claims to disprove. It refers to [7, Conjecture 2.5] and says the construction of the higher-degree generators is 'partially conjectural,' but the reader cannot verify that the vanishing [σ3, X10] = 0 contradicts the intended statement unless that statement is formulated precisely. Please include the precise conjecture and identify exactly which of its assertions Proposition 6.1 refutes.
minor comments (6)
  1. [Title page] The author line 'Thomas Will W acher' contains spacing artifacts; please correct the name to 'Thomas Willwacher'.
  2. [§2, first paragraph] The sentence 'It is known by an earlier work of the author that H^0(GC_2) ∼= grt1 is identified with the Grothendieck-Teichmüller Lie algebra' is grammatically awkward and should be rephrased.
  3. [Table 2] Table 2 contains two rows labeled '-16'; the row labels should be renumbered.
  4. [§4, last paragraph] The phrase 'we used used values of N' contains a duplicated word; it should read 'we used values of N'.
  5. [§4, paragraph before Eq. (6)] The text says the Deligne-Drinfeld conjecture has been verified 'as far'; this should be 'as far as', and the sentence should be completed.
  6. [§3, proof of Proposition 3.2] The proof says it uses 'a variation of Kneissler's argument ... with a modified last step,' but the modified step is not identified; please say explicitly which case or argument differs from [13, Section 5.1].

Circularity Check

1 steps flagged · score 4.0 of 10

Proposition 6.1's counterexample is forced by an unpublished same-author differential-vanishing result; the 11-loop computations themselves are independent.

  1. uniqueness imported from authors [Section 2 (p. 4) and Section 6, proof of Proposition 6.1]
    "Furthermore, as shown in [20, Corollary 4.4], no cancellation is possible for i = 1, 3, 5, since the differential vanishes on the respective pages of the spectral sequence. / However, if Z were a non-trivial cohomology class, this would imply that it is killed by some other class in our spectral sequence. However, this is not possible by degree reasons, see Table 1: The first possible position (11, 9) from where it could be killed is already in the vanishing range."

    The class Z = [σ3,X10] lives in bidegree (13,7). The only potential killer not excluded by the paper's own degree bounds is the i=1 differential source in bidegree (12,8); Table 1 explicitly leaves H^8(GC2^{12-loop}) unknown. The proof silently passes from (12,8) to the 'first possible position (11,9)' because [20, Cor. 4.4]—a result from the author's own unpublished preprint—declares the i=1 (and i=3,5) differentials zero. That vanishing is the load-bearing premise; it is neither proved nor machine-checked here. The conclusion [σ3,X10]=0 is therefore not derived from first principles in this paper but imported from the same-author citation chain.

full rationale

The paper's main numerical content—the 11-loop dimensions in Tables 1 and 2—is not circular: Section 4 computes ranks of the triconnected differentials over F_p, and exactness is certified by the rule that zero upper bounds (or entries with two adjacent zero bounds) force the lower-bound ranks to be tight. The top-degree entries in Section 3 are genuine upper bounds from the barrel-diagram reduction (Proposition 3.2), and for odd n they are matched to independent lower bounds of Vogel/Broadhurst; the even-n entries for g=13,14 are explicitly presented as conjectural upper bounds, not as predictions. Section 2's theoretical entries are derived from the published spectral sequence of [14], the Euler characteristic of [21]/[4], and degree bounds, not fitted to the target. The only step that reduces to a same-author prior result is the vanishing of the i=1,3,5 differentials imported from [20, Cor. 4.4]. That corollary is load-bearing for Proposition 6.1: it is the sole reason the class Z at (13,7) cannot be killed from (12,8), whose cohomology is listed as unknown in Table 1. Since [20] is an unpublished same-author preprint and the present paper neither proves nor independently verifies the vanishing statement, the counterexample to Brown's strong conjecture is partly forced by a self-citation. The paper is otherwise transparent about its limitations (Table 2 dagger entries and the 0* entries from [17]).

Assumptions & free parameters 2 free parameters · 7 assumptions · 0 invented entities

The central claims rest on a large body of prior theory (spectral sequences, quasi-isomorphisms, Grothendieck-Teichmüller computations) and on algorithmic parameters (finite field characteristic, number of random vectors). No new physical or algebraic entities are introduced.

free parameters (2)
  • Finite field characteristic p = 3323
    Chosen prime for rank computations. The F_p rank is a lower bound for the Q-rank, so the final dimensions are upper bounds. The choice is arbitrary but affects the probability that the randomized rank algorithm returns the true rank; no systematic scan over primes is reported.
  • Number of random vectors N in Coppersmith-Wiedemann = 4 to 32
    Algorithmic parameter controlling the probability of correctness in the block Wiedemann rank estimate; larger N increases reliability but costs more computation.
assumptions (7)
  • domain assumption The spectral sequence of the loop-order filtration on the twisted graph complex satisfies E1 = H(GC_0) and converges to Q L1, and no cancellations occur for i=1,3,5.
    Used throughout Section 2 to infer entries of Tables 1 and 2; cited to [14] and [20, Cor 4.4].
  • domain assumption The triconnected graph complex Gtri_n is quasi-isomorphic to GC_n.
    Used in Section 4 to replace the full graph complex by the smaller triconnected one; cited to [20].
  • domain assumption The dimensions of H^0(GC_2^{g-loop}) (the weight-g part of grt1) are known up to g=29.
    Used to fill the first row of Table 1 and to compute the bottom cohomology of GC_3; cited to [16] (in preparation) and [3].
  • standard math Any top-degree element of GC_3 is cohomologous to a linear combination of barrel diagrams (Kneissler's Proposition 3.1).
    Used as the base case for the extension to even n in Proposition 3.2; cited to [13].
  • domain assumption Known lower bounds on the top cohomology of GC_3 from Vogel and Broadhurst are correct.
    Used to convert the upper bounds into exact dimensions in Table 2; cited to [18] and [5].
  • standard math The degree vanishing bounds H^k(GC_2)=0 for k<0 or k>g-3 and corresponding bounds for GC_3 hold.
    Used in Section 2 and in the proof of Proposition 6.1 to rule out possible killing positions; cited to [19].
  • domain assumption The Coppersmith-Wiedemann implementation, including the GPU step, correctly returns a lower bound for the F_p-rank.
    Tooling assumption for the numerical ranks in Section 4; a bug could change the tables.

how reviews work

0 comments
Cite this review

Pith. "Pith review of The 11-loop graph cohomology." pith.science (2026). https://pith.science/paper/VJNCTZQL

@misc{pith2026250813724,
  author       = {Pith},
  title        = {Pith review of: The 11-loop graph cohomology},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VJNCTZQL}},
  note         = {Machine review of arXiv:2508.13724}
}
read the original abstract

We compute the Kontsevich graph cohomology in loop order 11, and for some degrees in higher loop order. We apply our results to discuss a conjecture of F. Brown, providing a counterexample to a strong version of the conjecture.

Figures

Figures reproduced from arXiv: 2508.13724 by the authors.

Figure 1
Figure 1. Numerical results for the upper bound on Htop(GCg-loop n ) according to Corollary 3.4. For g ≥ 11 the rank of d is computed over a finite field (F3323). We nevertheless get valid bounds on the dimension of the cohomology with rational coefficients as discussed in the text. Conjecture 3.7. For the choice of subspace V = Vg as in (4) the dimension bound of Corollary 3.4 is sharp. 4. Details on the numerical computatio… 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. Graph integrals, Feynman periods, and single-valued multiple zeta values

    math.QA 2026-07 conditional novelty 7.0 of 10

    Canonical integrals of graphs with E=2V−2 equal RW integrals and evaluate to single-valued multiple zeta values, which are shown to lie in the space of Feynman periods.

Reference graph

Works this paper leans on

22 extracted references · 17 canonical work pages · cited by 1 Pith paper

  1. [14]

    Preprint arXiv:q-alg/9706022, 1997

    arXiv 1997

  2. [20]

    Thomas Willwacher. M. Kontsevich’s graph complex and the Grothendieck- Teichm¨ uller Lie algebra.Invent. Math., 2014. arXiv:1009.1654. 16 THOMAS WILL W ACHER

    arXiv 2014

  3. [1]

    Alekseev and C

    A. Alekseev and C. Torossian, The Kashiwara-Vergne conjecture and Drinfeld’s associators, Ann. of Math. (2) 175, 2 (2012) 415–463; arXiv:0802.4300

    arXiv 2012

  4. [2]

    Graph Cohomology - An Overview and Some Computations

    Dror Bar Natan and Brendan McKay. Graph Cohomology - An Overview and Some Computations. Available at http://www.math.toronto.edu/~drorbn/ papers/GCOC/GCOC.ps

    work page

  5. [3]

    Bl¨ umlein, D.J

    J. Bl¨ umlein, D.J. Broadhurst, J.A.M. Vermaseren. The Multiple Zeta Value Data Mine, Comput. Phys. Commun. 181, 582–625, (2010)

    work page 2010

  6. [4]

    On the Euler characteristic of the commutative graph complex and the top weight cohomology of $\mathcal M_g$

    Michael Borinsky and Don Zagier. On the Euler characteristic of the com- mutative graph complex and the top weight cohomology of Mg. Preprint arXiv:2405.04190, 2024

    work page Pith review arXiv 2024

  7. [5]

    D. J. Broadhurst. Conjectured enumeration of Vassiliev invariants Preprint arXiv:q-alg/9709031, 1997

    work page Pith review arXiv 1997

  8. [6]

    Mixed Tate motives over Z

    Francis Brown. Mixed Tate motives over Z. Ann. of Math. (2), 175(2):949–976, 2012

    work page 2012

Show all 22 references
  1. [7]

    Invariant differential forms on complexes of graphs and Feyn- man integrals

    Francis Brown. Invariant differential forms on complexes of graphs and Feyn- man integrals. SIGMA Symmetry Integrability Geom. Methods Appl.17 (2021), Paper No. 103, 54 pp

    2021

  2. [8]

    Graph homology computations

    Simon Brun and Thomas Willwacher. Graph homology computations. New York J. Math. 30 (2024), 58–92

    2024

  3. [9]

    Tropical curves, graph com- plexes, and top weight cohomology of Mg, J

    Melody Chan, Søren Galatius and Sam Payne. Tropical curves, graph com- plexes, and top weight cohomology of Mg, J. Amer. Math. Soc. 34 (2021), no. 2, 565–594

    2021

  4. [10]

    Drinfeld, On quasitriangular quasi-Hopf algebras and on a group that is closely connected with Gal( Q/Q)

    V.G. Drinfeld, On quasitriangular quasi-Hopf algebras and on a group that is closely connected with Gal( Q/Q). (Russian) Algebra i Analiz 2, 4 (1990) 149–181; translation in Leningrad Math. J. 2, 4 (1991) 829–860

    1990

  5. [11]

    Parallel computation of the rank of large sparse matrices from algebraic K-theory

    Jean-Guillaume Dumas, Philippe Elbaz-Vincent, Pascal Giorgi, Anna Urb´ anska. Parallel computation of the rank of large sparse matrices from algebraic K-theory. PASCO ’07: Proceedings of the 2007 international work- shop on Parallel symbolic computation, pages 43 – 52, https:/...

    2007

  6. [12]

    The rational homotopy of mapping spaces of En operads

    Benoit Fresse, Victor Turchin and Thomas Willwacher. The rational homotopy of mapping spaces of En operads. Preprint, arXiv:1703.06123, 2017

    2017 arXiv

  7. [13]

    Kneissler

    Jan A. Kneissler. The number of primitive Vassiliev invariants up to degree

  8. [15]

    Differentials on graph complexes

    Anton Khoroshkin, Thomas Willwacher and Marko ˇZivkovi´ c. Differentials on graph complexes. Adv. Math., 307:1184–1214, 2017

    2017

  9. [16]

    ( http://linalg.org)

    The LinBox Group, LinBox – Exact Linear Algebra over the Integers and Finite Rings, Version 1.1.6; 2008. ( http://linalg.org)

    2008

  10. [17]

    Naef and T

    F. Naef and T. Willwacher. Numerical computation of linearized KV and the Deligne-Drinfeld and Broadhurst-Kreimer conjectures. in preparation, 2025

    2025

  11. [18]

    Naef and T

    F. Naef and T. Willwacher. An odd version of the Grothendieck-Teichm¨ uller and Kashiwara-Vergne Lie algebras. in preparation, 2025

    2025

  12. [19]

    Algebraic structures on modules of diagrams

    Pierre Vogel. Algebraic structures on modules of diagrams. J. Pure Appl. Algebra, 215(6): 1292–1339, 2011

    2011

  13. [21]

    The triconnected Kontsevich graph complex

    Thomas Willwacher. The triconnected Kontsevich graph complex. Preprint arXiv:2503.17131, 2025

    2025 arXiv

  14. [22]

    Multiple edges in M

    Thomas Willwacher and Marko ˇZivkovi´ c. Multiple edges in M. Kontsevich’s graph complexes and computations of the dimensions and Euler characteristics. Adv. Math. 272 (2015), 553–578. Department of Mathematics, ETH Zurich, R ¨amistrasse 101, 8092 Zurich, Switzerland

    2015

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.