On discrete loop signatures and Markov loops topology

Yves Le Jan

REVIEW 3 major objections 5 minor 19 references

On discrete loop signatures and Markov loops topology

T0 review · 3 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read Second-homology counts in Markov loop ensembles are independent Poisson variables.

desk verdict The main theorem is a real step forward for loop-soup topology, but the proof skips a convergence argument exactly where the loop measure is infinite. read the letter →

arxiv 1908.05187 v2 pith:WEVLOA2J submitted 2019-08-14 math.PR math.GT

Yves Le Jan This is my paper · ORCID

classification math.PRmath.GT MSC 60J1060G5505C9920F14

keywords MarkovloopensemblesmeasuresdiscretesignatureslowercentralseriessecondhomologyPoissonpointprocessesfreeLiealgebrastwistedtransitionmatrices

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 works out how the topology of the random loops generated by a Markov chain on a finite graph is reflected in algebraic invariants of the graph's fundamental group. Loops are classified by a natural degree coming from the lower central series, and the paper determines the joint law of the second-homology coordinates of all loops of degree greater than one. The central result is that these counts are independent Poisson random variables, with expectations given by an inverse Fourier transform of a limit of log-determinants of twisted transition matrices. This matters because the second homology is the first layer where noncommutativity of the fundamental group appears, so the formula turns a topological obstruction into an explicit probability law. The same framework also yields the homotopy-class distribution and the first-homology distribution.

What carries the argument

The load-bearing object is the discrete signature of a geodesic loop: a formal power series $S(g)=\prod_i e^{n_i X_{j_i}}$ in noncommuting symbols, viewed in the tensor algebra over the free Lie algebra. Its lowest-degree nonconstant term $P_g$ is a homogeneous Lie polynomial, and the degree and coefficients of $P_g$ define the loop's higher homologies. On the probabilistic side, the loop measure $\mu$ and the Poisson ensemble $L_\alpha$ convert these algebraic invariants into counts. To reach the second homology, the paper constructs a class-2 nilpotent group $G$ over $\mathbb{Z}/p\mathbb{Z}$ with elements $(a,c)$ and product $(a,c)(a',c')=(a+a',c+c'+\tfrac12(a\wedge a'-a'\wedge a))$, together with a representation $U_h$ resembling the Schrodinger representation. The identity $\sum_l \chi_{U_h}(H_A(l))\mu(l)=-\frac1{p^r}\log\det(I-P^{A,h})$ is the bridge between loop topology and determinants; an inverse Fourier transform and the limit $p\to\infty$ extract the Poisson intensities.

What would settle it

On a graph small enough to enumerate every geodesic loop, compare the empirical joint distribution of the second-homology coordinates for the degree-$>1$ loops of $L_\alpha$ with the Poisson law given by the inverse Fourier transform of $F(u)$. For example, take the graph made of two circles sharing one vertex with unit conductances and no killing: then $r=2$ and the only second-homology coordinate is $qN_{1,2}$, so the theorem predicts a single Poisson count for each integer $m$, with parameter $-\alpha\int_0^1 F(u)e^{-2\pi i m u}\,du$. Enumerating all degree-2 geodesic loops up to a length cutoff and estimating that parameter would settle whether the formula holds; any mismatch would point to failure of the large-$p$ limit or the Fourier interchange.

Watch

Extended reading notes

Core claim

The paper's central claim is Theorem 4: for any finite weighted graph and any choice of integers $m_{i,j}$ ($1\le i<j\le r$, where $r$ is the rank of the fundamental group), the variables counting loops in the Poissonian loop ensemble $L_\alpha^{(>1)}$ with double-edge currents $qN_{i,j}=m_{i,j}$ are independent Poisson random variables. Their expectations are $-\alpha$ times the inverse Fourier transform, over the torus $[0,1]^{r(r-1)/2}$, of $F(u)=\lim_{p\to\infty} p^{-r}\log\det(I-P^{A,h(u,p)})$, where $P^{A,h(u,p)}$ is the transition matrix twisted by a nilpotent holonomy built from the prime $p$ and the skew-symmetric matrix $u$. Since the coordinates $qN_{i,j}$ determine the second homology $h_2$, this gives the exact law of the second-homology field of the loop ensemble. The derivation uses a discrete analogue of the Schrodinger representation of the Heisenberg group over $\mathbb{Z}/p\mathbb{Z}$, and takes the large-$p$ limit to pass from finite cyclic data to the integers.

Load-bearing premise

The load-bearing premise is that the limit $F(u)=\lim_{p\to\infty} p^{-r}\log\det(I-P^{A,h(u,p)})$ exists and can be interchanged with the inverse Fourier transform that extracts integer loop counts from the characteristic function; if this interchange fails, the explicit Poisson formula in Theorem 4 is not established.

Editorial extensions

If this is right

For any finite graph, the whole second-homology field of the degree-$>1$ loop ensemble is Poissonian: the counts for different coordinate vectors are independent, and their means are determined by the graph's transition matrix.
The same trace-formula method gives the homotopy-class distribution and, for regular graphs, closed-form geodesic loop intensities, so the topological classification is fully explicit in those cases.
The degree of a loop is read off from the first nonzero term of its discrete signature, linking algebraic loop invariants to probabilistic loop statistics at every level.
Because the second-homology formula is a limit over primes, finite cyclic nilpotent holonomies give a computable approximation scheme for the integer-valued counts.
The paper notes that higher homologies could in principle be treated by nilpotent groups of higher class; the same Fourier-determinant pattern would then give Poisson laws for all levels of the lower central series.

Reading between the lines

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

If Theorem 4 is correct, the total second-homology field $h_2(L_\alpha^{(>1)})$ is a compound Poisson measure on $\mathbb{Z}^{r(r-1)/2}$, a structure that could be simulated directly on small graphs to check the formula.
The limiting determinant $F(u)$ can be read as a Fredholm determinant of an operator on functions on the $r$-torus; developing that operator picture might connect the discrete formula to the Brownian-loop-soup scaling limit, where the second-homology coordinates become Levy areas.
A natural test case is the graph made of two circles sharing one vertex: the second-homology space is one-dimensional, so Theorem 4 predicts a single Poisson count that can be checked by enumerating degree-2 geodesic loops by hand.

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.

Desk Editor's Note

read the letter

The genuinely new piece here is Theorem 4: a joint distribution for the second-homology coordinates of degree-two loops in a Markov loop ensemble. That is a real addition, and the machinery behind it—the nilpotent holonomy group over Z_p, the Schr\"odinger-like representation, and the exact trace computation—is coherent and well motivated. The algebraic parts of the paper are solid, and the self-review of earlier first-homology and homotopy results is honest and useful. Proposition 4 cleanly recovers Mnëv's formula as it should.

The soft spot is exactly where the stress-test note lands: the passage from finite p to the limit defining F(u), and then to the inverse Fourier formula, is asserted with 'It follows that' and Remark (b). When the killing rate is positive, the loop measure is finite, the sets {qN_i ≡ 0 mod p} shrink to {qN_i = 0}, and a dominated convergence argument makes the limit and inversion legitimate. But the paper also treats κ = 0, where the loop measure has infinite total mass and the sums are only conditionally convergent. No uniform integrability or cancellation argument is supplied for that case, and the existence of the limit F(u) itself is not proved. This is a genuine gap in the proof of the main theorem, not a cosmetic one.

It is also fixable: either restrict the theorem to positive killing, or provide a proper limiting argument for the infinite-measure case, for instance by introducing a killing parameter and taking κ → 0 at the end. There is nothing here that suggests the conclusion is false; the computation is too clean for that. I also want to note what the paper does not do: it does not fit any parameters, it does not hide circularity, and it is explicit about the limitation for higher homologies. The citations are mostly to the author's own prior work, but that is appropriate for a line of research he largely created.

Who should read this: people working on Markov loop soups, discrete Chen signatures, and the topology of random loop ensembles. They will find a useful consolidation of known results and a plausible new theorem that needs a bit more proof hygiene before it becomes fully reliable.

I would send it to a serious referee. The main contribution deserves refereed scrutiny, and the convergence gap should be caught and fixed in revision.

Referee Report

3 major / 5 minor

Summary. The paper studies Markov loop ensembles on finite graphs and develops relations between discrete signatures, the lower central series of the free fundamental group, and the homology of loops. The first part recalls or derives algebraic identities (Propositions 1–3) expressing the degree and the first non-zero homology class of a loop in terms of signed multiple-edge crossing counts. The probabilistic part uses the loop measure and its Poissonian ensembles: Proposition 4 recovers Mnev's formula for the expected number of loops homotopic to a given closed geodesic on a regular graph, Theorem 2 gives the distribution of first-homology coordinates via Fourier inversion, and Theorem 3 is the Selberg-type trace formula for holonomy characters. The new contribution is Section 8, where a finite nilpotent holonomy group G = Z_p^r ⋉ (Z_p^r)^{∧2} with a Schrödinger-like representation is used to derive the joint distribution of the second-homology coordinates of loops of degree greater than 1. This is stated as Theorem 4, with expectations given by an inverse Fourier transform of the limit F(u) = lim_{p→∞} p^{-r} log det(I - P^{A,h(u,p)}).

Significance. If Theorem 4 is valid, it is a genuine advance: it gives an explicit formula for the exact joint distribution of the second-homology coordinates of a Markov loop ensemble, going beyond the previously known first-homology and homotopy-class results. The algebraic framework is elegant and mostly well supported: Propositions 1–3 are backed by standard references on free Lie algebras and signatures, and Proposition 4 correctly reproduces a known formula of Mnev, which is a useful check on the formalism. The trace formula in Theorem 3 is a standard and powerful tool, and the idea of using a finite nilpotent holonomy group together with a discrete Schrödinger representation to isolate degree-two homology is conceptually appealing. The paper is concise and largely self-contained, and its main difficulty is a single but load-bearing analytic passage in the proof of Theorem 4.

major comments (3)

[Section 8, equation preceding Theorem 4 and Theorem 4] The central step of the proof is the assertion, introduced by 'It follows that' and repeated in Remark (b), that one may pass to the limit p → ∞ in the identity -p^{-r} log det(I - P^{A,h(u,p)}) = ∑_{l: qN_i(l)≡0 mod p} e^{2πi⟨qN^{(2)}(l), h(u,p)⟩/p} μ(l) and then interchange the resulting limit F(u) with the inverse Fourier transform that extracts the integer-valued second-homology counts. The finite-p identity itself is a coherent consequence of Theorem 3 and the trace computation, but the right-hand side is a conditionally convergent sum over a loop measure that need not be finite (for instance when the killing rate vanishes), and for every finite p the summation set {qN_i(l)≡0 mod p} still contains loops with arbitrarily large first homology. No dominated convergence, uniform integrability, or regularization argument is supplied to justify either the pointwise convergence of these sums to the sum over loops with qN_i(l)=0 for all i, or the passage of the limit under the Fourier integral in r(r-1)/2 variables. This is load-bearing because the expectation formula in Theorem 4 is the paper's principal new result; without this limit/interchange step, that formula is not derived.
[Theorem 4 and Remark (c)] The proposed interpretation of F(u) as the logarithm of a Fredholm determinant for the infinite-dimensional representation U_u is also asserted without proof. Even if such a representation exists, the proof needs to show that F(u) is a sufficiently regular function of u for the inverse Fourier transform in Theorem 4 to be meaningful; the manuscript only establishes, conditionally on the unjustified convergence, that F is a pointwise limit. This is part of the same analytic gap as the previous comment, but it deserves separate attention because the regularity and integrability of F are necessary for the displayed integral formula to define the Poisson expectations.
[Section 8, definition of the limit] The limit defining F(u) is taken over primes p with h(u,p) the integral part of u p. The proof does not explain why the normalized logarithm p^{-r} log det(I - P^{A,h(u,p)}) has a limit, nor why the limit is independent of the particular way the integers h(u,p) approximate u. Without such a statement, the formula for F(u) is not well defined, and the subsequent Fourier inversion cannot be applied. This is a prerequisite for Theorem 4 and should be proved or replaced by a precise approximation argument.

minor comments (5)

[Introduction] The first paragraph contains a typo: 'in the context of on finite graphs' should read 'in the context of finite graphs'.
[Section 5, Proposition 3 remark] The remark after Proposition 3 says 'This gives a non self-contained proof of proposition 2'; presumably 'self-contained' is intended, and the sentence should be corrected.
[Section 8, notation] The indicator ‹₁{a=0}› appears in the source as '1 ta“0u'; it should be typeset as a proper indicator function, and the same applies to other occurrences of indicator notation.
[Abstract] The manuscript lists no MSC classification; the abstract contains an empty 'AMS 2000 subject classification' field, which should be completed or removed.
[References] Reference [8] is cited as 'To appear in Séminaire de Probabilités' without a year; the reference should be updated to its final publication data.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: Theorem 4 is derived from an established trace formula and Fourier inversion, with an unproved limit interchange that is an analytical gap, not a circular reduction.

full rationale

The paper's central new result, Theorem 4, gives the joint distribution of the second-homology coordinates of a Markov loop ensemble. Its derivation starts from Theorem 3, the trace formula sum_l chi_pi(H_A(l)) mu(l) = -1/dim(pi) log det(I - P^{A,pi}), which is stated as following directly from the based loop measure and is an established identity, not a restatement of the target distribution. The nilpotent holonomy computation then produces the exact finite-p identity expressing the Fourier transform of the loop measure restricted to loops with vanishing first homology as -p^{-r} log det(I - P^{A,U_h}). Theorem 4 is obtained by an inverse Fourier transform and a p-to-infinity limit. No parameter is fitted to the claimed Poisson expectations, and no quantity in Theorem 4 is defined in terms of the output it predicts. The main weakness is that the text says 'It follows that' before Theorem 4 and Remark (b) asserts that the inverse Fourier transform can be performed before the limit, but no dominated convergence or uniform integrability argument is supplied for interchanging the limit with the Fourier integral. That is a genuine analytical gap in the proof, but it is not circularity: it does not reduce the theorem to its own conclusion, nor does it rename an input as a prediction. The self-citations in the paper are to standard or previously established results, including the loop measure framework [7], signature theory [10], and free Lie algebra facts [15]; they are not used to forbid alternatives or to define the predicted distribution into existence. Sections 6 and 7 explicitly review and benchmark earlier results, including recovery of formulas from Mnëv and Ihara zeta, so the paper is largely self-contained against external checks. Overall, the derivation chain is independent of the claimed output, and no circular step is present.

Assumptions & free parameters 0 free parameters · 3 assumptions · 2 invented entities

The central results use standard algebraic topology and free group theory as black boxes, and introduce one unproved analytic limit in Theorem 4. No number is fitted to data; the graph weights are inputs. The main analytic assumption is the existence and well-behavedness of the determinant limit, which is ad hoc to this proof.

assumptions (3)

domain assumption The graph is finite, connected, non-oriented, with at most one edge between two vertices and no loop edges; conductances C_e and killing rates kappa_x define a lambda-symmetric Markov chain.
Section 2 and Section 6; this is the framework for all loop measures and loop ensembles in the paper.
standard math Standard combinatorial group theory facts: lower central series, Witt formula for ranks of H_n, Magnus theory of free group signatures, and shuffle product identities.
Used throughout Sections 3-5; the paper cites [10], [14], and [15] as the sources.
ad hoc to paper The limit F(u)=lim_{p->infinity} (1/p^r) log det(I-P^{A,h(u,p)}) exists and the inverse Fourier transform can be interchanged with the limit to extract integer-valued loop counts.
Asserted before Theorem 4 in Section 8 without proof; the main new result depends on this analytic step.

invented entities (2)

Nilpotent holonomy group G = Z_p^r ⋉ wedge^2 Z_p^r with Heisenberg-type product
purpose: Finite group encoding first and second homology data of a loop so that the trace formula of Theorem 3 can be applied.
Introduced in Section 8 purely as a computational device; it is a standard construction over Z_p but is not tested outside this paper.
Schrodinger-like unitary representation U_h of G on functions on Z_p^r
purpose: Provides a computable character chi_{U_h}(H_A(l)) = indicator(h1=0) exp(2 pi i <h2(l),h>/p), linking loop counts to a determinant.
Defined and verified by direct calculation in Section 8; its use is internal to the derivation of Theorem 4.

how reviews work

0 comments

Cite this review

Pith. "Pith review of On discrete loop signatures and Markov loops topology." pith.science (2026). https://pith.science/paper/WEVLOA2J

@misc{pith2026190805187,
  author       = {Pith},
  title        = {Pith review of: On discrete loop signatures and Markov loops topology},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WEVLOA2J}},
  note         = {Machine review of arXiv:1908.05187}
}

read the original abstract

Our purpose is to explore, in the context of loop ensembles on finite graphs, the relations between combinatorial group theory, loops topology, loop measures, and signatures of discrete paths. We determine the distributions of the loop homotopy class, and of the first and second homologies, defined by the lower central series of the fundamental group.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

19 extracted references · 19 canonical work pages

[1]

Some relations between the spectra of simple and non-backtracking random walks

Nalini Anantharaman. Some relations between the spectra of sim ple and non-backtracking random walks. arXiv:1703.03852

work page Pith review arXiv
[2]

Kuo-tsai Chen, Integration of paths–a faithful representa tion of paths by noncommutative formal power, Trans. Amer. Math. Soc. 156, 39 5-407 (1971)

work page 1971
[3]

Ben Hambly, Terry Lyons, Uniqueness for the signature of a pat h of bounded variation and the reduced path group, Ann. of Math. (2) , 171 109–167 (2010)

work page 2010
[4]

Zeta functions of ﬁnite grap hs

Motoko Kotani, Toshikazu Sunada. Zeta functions of ﬁnite grap hs. J. Math. Sci. Univ. Tokyo 7, 7-25 (2000)

work page 2000
[5]

Jacobian Tori Associated wit h a Finite Graph and Its Abelian Covering Graphs

Motoko Kotani, Toshikazu Sunada. Jacobian Tori Associated wit h a Finite Graph and Its Abelian Covering Graphs. Advances in Applied Mathematics 24, 89–110. (2000)

work page 2000
[6]

The Brownian loop soup

Gregory Lawler, Wendelin Werner. The Brownian loop soup. Proba bility Th. Rel. Fields 128, 565-588 (2004). 19

work page 2004
[7]

Markov paths, loops and ﬁelds

Yves Le Jan. Markov paths, loops and ﬁelds. ´Ecole d’´Et´ e de Probabilit´ es de Saint-Flour XXXVIII - 2008. Lecture Notes in Mathematics 2026 . Springer. (2011)

work page 2011
[8]

Random ﬂows deﬁned by Markov loops

Yves Le Jan. Random ﬂows deﬁned by Markov loops. To appear in St´ eminaire de Probabilit´ es

work page

Show all 19 references

[9]

Markov loops, free ﬁeld and Eulerian networks

Yves Le Jan. Markov loops, free ﬁeld and Eulerian networks. arXiv:1405.2879. J. Math. Soc. Japan 67, 1671-1680 (2015)

2015 arXiv
[10]

Wilhelm Magnus, Abraham Karass, Donald Solitar. Dover. (1976) Com- binatorial group theory. 2d Ed. Dover. (1976)

1976
[11]

Markov loops, Coverings and Fields

Yves Le Jan. Markov loops, Coverings and Fields. arXiv: 1602.02 708. Annales de la Facult´ e des Sciences de Toulouse XXVI, 401-416 (201 7)
[12]

Discrete Path Integral Approach to the Selber g Trace For- mula for Regular Graphs

Pavel Mn¨ ev. Discrete Path Integral Approach to the Selber g Trace For- mula for Regular Graphs. Comm. Math. Phys. 274, 233-241 (2007)

2007
[13]

Massey, Algebraic Topology: An Introduction Springer (1967)

William S. Massey, Algebraic Topology: An Introduction Springer (1967)

1967
[14]

Lie elements and an algebra associated with shuﬄes

Rimhak Ree. Lie elements and an algebra associated with shuﬄes. Ann. of Math. (2)68 210–220, (1958)

1958
[15]

Free Lie algebras

Christophe Reutenauer. Free Lie algebras. Clarendon Press. Oxford. (1993)

1993
[16]

Stark, Audrey A

Harold M. Stark, Audrey A. Terras, Zeta functions on ﬁnite gr aphs and coverings. Advances in Maths 121, 134-165 (1996)

1996
[17]

Scuola intenazio nale di Fisica ”Enrico Fermi”

Kurt Symanzik, Euclidean quantum ﬁeld theory. Scuola intenazio nale di Fisica ”Enrico Fermi”. XL V Corso. 152-223 Academic Press. (196 9)
[18]

Yusuke Watanabe and Kenji Fukumizu, Graph zeta function in t he Bethe free energy and loopy belief propagation. Adv. Neural Info rm. Proc. Sys. 22, 2017-2025.(2010)

2010
[19]

Appendix in Graphs on Surfaces and their Application s, by S.K

Don Zagier. Appendix in Graphs on Surfaces and their Application s, by S.K. Lando and A.K. Zvonkin. Springer. (2004). NYU-ECNU Institute of Mathematical Sciences at NYU Shanghai. D´ epartement de Math´ ematique. Universit´ e Paris-Sud. Orsay yves.lejan@math.u-psud.fr yl57@nyu.edu 20

2004

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [1]

Some relations between the spectra of simple and non-backtracking random walks

Nalini Anantharaman. Some relations between the spectra of sim ple and non-backtracking random walks. arXiv:1703.03852

work page Pith review arXiv

[2] [2]

Kuo-tsai Chen, Integration of paths–a faithful representa tion of paths by noncommutative formal power, Trans. Amer. Math. Soc. 156, 39 5-407 (1971)

work page 1971

[3] [3]

Ben Hambly, Terry Lyons, Uniqueness for the signature of a pat h of bounded variation and the reduced path group, Ann. of Math. (2) , 171 109–167 (2010)

work page 2010

[4] [4]

Zeta functions of ﬁnite grap hs

Motoko Kotani, Toshikazu Sunada. Zeta functions of ﬁnite grap hs. J. Math. Sci. Univ. Tokyo 7, 7-25 (2000)

work page 2000

[5] [5]

Jacobian Tori Associated wit h a Finite Graph and Its Abelian Covering Graphs

Motoko Kotani, Toshikazu Sunada. Jacobian Tori Associated wit h a Finite Graph and Its Abelian Covering Graphs. Advances in Applied Mathematics 24, 89–110. (2000)

work page 2000

[6] [6]

The Brownian loop soup

Gregory Lawler, Wendelin Werner. The Brownian loop soup. Proba bility Th. Rel. Fields 128, 565-588 (2004). 19

work page 2004

[7] [7]

Markov paths, loops and ﬁelds

Yves Le Jan. Markov paths, loops and ﬁelds. ´Ecole d’´Et´ e de Probabilit´ es de Saint-Flour XXXVIII - 2008. Lecture Notes in Mathematics 2026 . Springer. (2011)

work page 2011

[8] [8]

Random ﬂows deﬁned by Markov loops

Yves Le Jan. Random ﬂows deﬁned by Markov loops. To appear in St´ eminaire de Probabilit´ es

work page