Spectral properties of the zero temperature Edwards-Anderson model

Mriganka Basu Roy Chowdhury; Shirshendu Ganguly

REVIEW 3 major objections 4 minor 19 references

Spectral properties of the zero temperature Edwards-Anderson model

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read In the Edwards-Anderson spin glass, the Fourier spectrum is forced off the shortest path by a new barrier argument.

desk verdict Promising barrier idea for quantitative spectral bounds in the EA model, but Lemma 3.1 has an unaddressed proof gap that takes down Theorems 2.2 and 2.3 as written. read the letter →

arxiv 2507.10507 v1 pith:Y4WHV546 submitted 2025-07-14 math.PR cond-mat.dis-nnmath-phmath.MP

Mriganka Basu Roy Chowdhury , Shirshendu Ganguly This is my paper · ORCID

classification math.PRcond-mat.dis-nnmath-phmath.MP MSC 60K3582B44

keywords Edwards-AndersonmodelspinglasszerotemperatureFourierspectrumspectralsamplenoisesensitivitydisorderchaospercolationbarriers

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 studies the zero-temperature Edwards-Anderson spin glass on an n by n grid and asks how large the spectral sample of the two-point spin correlation must be. It proves that for two horizontally opposite vertices, the spectral measure assigns exponentially small mass to the straight path between them, and more generally to every edge set of size at most (1+epsilon)n. The proof introduces a percolation-type barrier: a local pattern of very strong and very weak couplings around an edge that forces the ground-state spin product to be insensitive to the sign of that edge. If correct, this is evidence that the Fourier spectrum of the EA ground state is genuinely larger than the shortest path, consistent with numerics suggesting a fractal dimension above one.

What carries the argument

The carrying object is the barrier configuration around an edge e: a box around e whose boundary cycle consists of nine 'blue' edges with |J| at least 100 and whose interior 'red' edges have |J| at most 1. Lemma 3.1 asserts that in any ground state, the spin product along the boundary path must satisfy sigma_{w_i} sigma_{w_{i+1}} J_{(w_i,w_{i+1})} > 0 for every boundary edge, so the boundary spins are pinned relative to one corner and the relative spin sigma_u sigma_v is independent of the sign of J_e. Lemma 3.2 then flips the signs of all other edges in a vertical cutset through a barrier; because barriers occur independently with constant probability along the line, an exponentially small failure probability plus cutset symmetry forces the conditional expectation given the line couplings to vanish.

What would settle it

Run the barrier configuration around the middle edge e of the line in an n by n grid, set all exterior couplings adjacent to the box boundary to +1000 with outside spins free, and compare ground-state sigma_u sigma_v for J_e=+1 and J_e=-1; if the product changes, Lemma 3.1's pinning claim is false.

Watch

Extended reading notes

Core claim

The paper's central claim is Theorem 2.3: for u=(1,n/2) and v=(n,n/2) in the n by n grid, there exist epsilon>0 and c>0, independent of n, such that the total spectral mass of all edge sets S connecting u and v with |S| at most (1+epsilon)n is at most exp(-cn). Interpreted probabilistically, a sample from the spectral measure has size strictly larger than (1+epsilon) times the L1 distance with probability exponentially close to 1. The weaker Theorem 2.2 gives the same exponential decay for the single straight-line set L. The argument builds on Chatterjee's observation that any Fourier block whose support does not connect the two vertices has zero coefficient; the new work is a quantitative, geometrically local mechanism that kills the conditional expectation of the relative spin on a large conditioning set.

Load-bearing premise

The argument depends on Lemma 3.1's assumption that a local box of prescribed high and low couplings pins the boundary spins of the box no matter what the couplings outside the box do, even though those exterior couplings are not part of the energy comparison in (3.5).

Editorial extensions

If this is right

The straight line L carries exponentially small spectral mass, so the spectral sample is not concentrated on the unique geodesic between u and v.
All small sets of size at most (1+epsilon)n together carry exponentially small mass, so with probability exponentially close to 1 the spectral sample has size at least (1+epsilon)n.
The same statements hold in any dimension, with columns replaced by slabs of width W.
A matching lower bound up to logarithmic factors shows the exponential decay in Theorem 2.2 is essentially sharp.
The barrier-plus-entropy method may be reused for other conditioning sets, since the barrier events are independent across disjoint straight segments.

Reading between the lines

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

The method should transfer to positive temperature if the barrier energy comparison is augmented by an entropy penalty for flipping interior spins; that would turn Fourier-spectrum lower bounds into bounds on the Ornstein-Uhlenbeck decorrelation time.
Because the barrier event is defined only through absolute values, the same lower bound should hold for any continuous symmetric coupling distribution; the Gaussian assumption is used for the Hermite framework, not for the barrier mechanism.
One could probe the conjectured fractal dimension by computing the expected size of the spectral sample; the paper's bounds imply E|S| is at least (1+epsilon)n minus an exponentially small term, but not a power-law exponent.

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

Colleague,

The short version: the paper has a genuinely new idea — a local barrier configuration that would decouple a conditioned coupling from the relative spin — and the intended application to lower bounds on spectral sample size is a sensible next step after Chatterjee. But the main lemma has a proof gap that the current text does not fix, and until that is repaired the theorems are not established.

What is new and good: the barrier construction itself, and the way the authors use it to get an exponential bound on µ(L) and then a union-bound upgrade to all sets of size at most (1+ε)n. The counting part of Lemma 3.5 is careful and the estimates look right, conditional on having the barrier lemma. The paper is also honest: it states clearly that the improvement over Chatterjee's qualitative result is modest, and it discusses the limitations of the method in Remarks 2.4 and 2.5. Citation practice is fine; the main external input is Chatterjee's Proposition 2.1, which is clearly cited.

The soft spot is Lemma 3.1. The energy comparison in (3.5) only counts the blue path edges and the Low edges inside the box. It omits every edge from the boundary vertices w_i to the exterior. The barrier event does not constrain those exterior couplings, so they can be arbitrarily large. When the competitor flips a boundary vertex, each such exterior edge contributes ±2|J_e| to the energy difference, and a single large exterior coupling can wipe out the claimed 160 gain. The lemma as stated is therefore not proven, and it may well be false as a deterministic statement. Since Lemma 3.2 uses Lemma 3.1 to flip e* without changing σ_uσ_v, both Theorem 2.2 and Theorem 2.3 rest on this step. The fix is plausible — add the exterior edges incident to the boundary to the Low set, or otherwise bound them — but that is not a cosmetic change; it alters the barrier event and the probability estimate.

I don't think the paper is salvageable in its present form, but the idea deserves a careful referee. The structural parts — the philosophy of using barriers to get quantitative connectivity control, and the combinatorial reduction — are worth engaging with. I'd send it to review rather than desk-reject, but the referee should focus on whether Lemma 3.1 can be repaired.

For your purposes: would I cite it now? No, because the main results are not established. Would I bring it to reading group? Maybe; it is a good discussion piece about where noise-sensitivity arguments for spin glasses stand, and the gap is instructive.

Referee Report

3 major / 4 minor

Summary. The paper studies the zero-temperature Edwards-Anderson model on the n×n grid with i.i.d. standard Gaussian couplings. It defines a spectral probability measure μ for the two-point ground-state correlation σ_u σ_v via the Hermite expansion, and uses Chatterjee's Proposition 2.1 that the spectral sample must connect u and v. The main results are exponential upper bounds: Theorem 2.2 states μ(L) ≤ exp(−cn) for the straight horizontal line L connecting u=(1,n/2) and v=(n,n/2), and Theorem 2.3 states that the total spectral mass of sets of size at most (1+ε)n is also exp(−cn). The proof introduces local “barrier” events of high and low couplings and claims in Lemma 3.1 that a barrier forces alignment of the boundary spins of a small box, independently of all couplings outside. Lemma 3.2 then uses this to implement a cut-set sign-flipping symmetry and conclude that the conditional expectation of σ_u σ_v is zero on the good event, yielding the exponential decay. Theorem 2.3 is obtained by combining the barrier argument with a combinatorial classification of small connecting sets, with a union bound over the classes.

Significance. The intended contribution is a new percolation-type estimate on the size of the spectral sample in a lattice spin glass, a direction where rigorous results are scarce and where numerical work predicts fractal dimension strictly above one. The paper is transparent about its modest quantitative gain and explicitly builds on an external result of Chatterjee rather than on fitted parameters or circular reasoning; I see no circularity in the use of Proposition 2.1. If the barrier mechanism were valid, it would be a clean and potentially transferable idea. However, the central mechanism is not established: the proof of Lemma 3.1 omits uncontrolled exterior edges, and the main theorems depend directly on that lemma. The exposition is otherwise clear, and the combinatorial part of Theorem 2.3 is reasonable, but the probabilistic core of the argument needs substantial repair.

major comments (3)

[Section 3, Lemma 3.1 and equation (3.5)] The energy comparison in (3.5) is incomplete. The configuration σ̂ flips the ten boundary vertices w_1,...,w_10, so every edge from a boundary vertex to the exterior of the box changes its energy by ±2|J_e|. These exterior couplings are not constrained by the event Barrier(e), and they can be arbitrarily large. A single exterior edge with |J_e| ≥ 1000 incident to a boundary vertex can dominate the claimed lower bound of 200 − 40 = 160 and pin that boundary spin, so the lemma's conclusion that σ_{w_i}σ_{w_{i+1}}J_{(w_i,w_{i+1})} > 0 for all i, irrespective of all couplings outside the box, is false as stated. The domain Markov property invoked at the end of the proof does not help because the boundary spins are not conditioned; the claim 'fixing σ_{w_i} for 1 ≤ i ≤ 10' assumes the very alignment that the lemma is supposed to prove. Since this lemma is the load-bearing step for Lemma 3.2 and hence for Theorems 2.2 and 2.3, the main results are not established by the present proof.
[Section 3, Lemma 3.2 and equation (3.7)] Lemma 3.2 asserts E[σ_u σ_v | J_L, |J|, Good] = 0 by arguing that on Barrier(e*) the value of σ_u σ_v does not depend on the sign of J_{e*}, so the cut-set flipping symmetry applies. This step relies entirely on Lemma 3.1. Given the gap in Lemma 3.1, the sign of σ_u σ_v may depend on J_{e*} through large exterior couplings incident to the boundary of the barrier box, and the conditioned law of σ_u σ_v need not be uniform on {±1}. Consequently the conditional-expectation bound (3.3), and with it the exponential bound μ(L) ≤ exp(−cn), does not follow.
[Section 3, Lemma 3.5 part 2 (proof of Theorem 2.3)] The proof of part 2 of Lemma 3.5 repeats the barrier argument on straight segments of the sets in each class G and again concludes that the conditional expectation is zero on the good event. No new mechanism is introduced to control exterior edges incident to the boundary vertices of the barrier boxes. Thus the same omission that invalidates Lemma 3.1 propagates to the proof of Theorem 2.3, and the claimed exponential bound on the total spectral mass of small sets is not proved as written.

minor comments (4)

[Section 3, Figure 3.2] The vertices w_1,...,w_10 are not labeled in Figure 3.2, which makes it difficult to verify the traversal order, the set of Low edges, and the top-right corner mentioned in the lemma statement.
[Section 3, Remark 3.3] The sentence 'for all the vertices w including v, incident on edges e∈L, it must be must be the case, provided that σ_u is pinned to be 1, that σ_w = 1' contains a duplicated 'must be' and is grammatically unclear; the intended set of vertices should be specified precisely.
[Section 3, proof of Theorem 2.3] The statement 'The theorem is an immediate consequence of the following lemma by a union bound' would benefit from spelling out the union bound over the collection C, since the balance between the entropic factor exp(n·ε·(log(1/ε)+c₂)) and the spectral bound exp(−cn) is the crux of the argument and is only implicit in the text.
[Section 1 and Remark 3.4] The paper advertises that the arguments are not planar and hold in any dimension, but the barrier construction is drawn and described only in two dimensions; the spanning-tree generalization in Remark 3.4 is only a sketch and would need a precise statement to be verifiable.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the derivation is self-contained except for an external, non-self-cited lemma from Chatterjee.

full rationale

The paper's derivation chain is not circular. The spectral measure is defined directly from the Fourier coefficients of the two-point spin correlation, and the main results are lower bounds on the size of its support. The load-bearing input is Proposition 2.1, quoted from Chatterjee [Cha23], which is an external result by a different author, not a self-citation; the paper even reviews its proof before use. The conditional-expectation identity (3.1) follows from Hermite orthonormality and is a direct computation, not an assumed conclusion. The barrier construction in Lemma 3.1 is an independent geometric/energetic argument that does not presuppose the spectral statements of Theorems 2.2 or 2.3. There are no fitted parameters renamed as predictions, no self-citation chain establishing uniqueness, no ansatz smuggled in via the paper's own prior work, and no re-labeling of a known result as a new one. Even if Lemma 3.1's energy comparison were mathematically incomplete, as a correctness criticism of the exterior-edge terms, that would be a proof gap rather than a circularity: the lemma is not defined in terms of the target theorem. The paper's reliance on external results and its honest remarks about open problems further support a non-circular finding. Thus the appropriate circularity score is 0.

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

The paper introduces no new physical entities: the barrier is a proof device, not a new object. The main unstated assumption is the local dominance of the barrier over unconstrained exterior edges.

assumptions (4)

domain assumption The Gaussian couplings are i.i.d. standard, and the ground state is unique up to global sign flip.
Sets the model; standard for the EA model. Used in Section 1.
domain assumption Chatterjee's Proposition 2.1: if the support E_k does not connect u and v, then α_k = 0.
External theorem from [Cha23], used as the starting point for Theorems 2.2 and 2.3.
standard math The Hermite polynomials form an orthonormal basis and are eigenfunctions of the OU generator with eigenvalue -k.
Standard Fourier analysis for Gaussian functions; used in Section 1.
ad hoc to paper The local barrier configuration (blue edges ≥100, red edges ≤1) forces the ground state alignment on the boundary path independently of exterior couplings and of J_e.
This is the content of Lemma 3.1; the proof omits exterior edges, so the axiom is not established. Appears in the proof of Lemma 3.1, equation (3.5).

how reviews work

0 comments

Cite this review

Pith. "Pith review of Spectral properties of the zero temperature Edwards-Anderson model." pith.science (2026). https://pith.science/paper/Y4WHV546

@misc{pith2026250710507,
  author       = {Pith},
  title        = {Pith review of: Spectral properties of the zero temperature Edwards-Anderson model},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Y4WHV546}},
  note         = {Machine review of arXiv:2507.10507}
}

read the original abstract

An Ising model with random couplings on a graph is a model of a spin glass. While the mean field case of the Sherrington-Kirkpatrick model is very well studied, the more realistic lattice setting, known as the Edwards-Anderson (EA) model, has witnessed rather limited progress. In (Chatterjee,'23) chaotic properties of the ground state in the EA model were established via the study of the Fourier spectrum of the two-point spin correlation. A natural direction of research concerns fractal properties of the Fourier spectrum in analogy with critical percolation. In particular, numerical findings (Bray, Moore,'87) seem to support the belief that the fractal dimension of the associated spectral sample drawn according to the Fourier spectrum is strictly bigger than one. Towards this, in this note we introduce a percolation-type argument, relying on the construction of ``barriers'', to obtain new probabilistic lower bounds on the size of the spectral sample.

Figures

Figures reproduced from arXiv: 2507.10507 by the authors.

**Figure 3.1.** The setup for Theorem 2.2. The blue edges form a “vertical” cutset separating u and v. Since the blue edges form a cut-set separating u and v, flipping the signs of the couplings along these edges will flip the sign of the relative spin σuσv. However, since we condition on JL, which freezes Je ⋆ for the blue edge e ⋆ on L, its sign cannot be flipped. This brings us to the following key new observation. With a positi… view at source ↗

**Figure 3.2.** The barrier configuration around e. The couplings along all the dashed red edges are ⩽ 1 in absolute value, and those along the solid blue edges are > 100 in absolute value. We refer to the horizontal axis containing e as the central horizontal axis. The top-right corner which will be a useful reference point in the arguments is made solid. We first specify what a barrier configuration will be for us. Referring to … view at source ↗

**Figure 3.3.** Quantities involved in the proof of Lemma 3.5. The solid red edges form the set S. The columns I3 and I5 (shaded in orange) are “straight” (note that I3 is straight even though its left-boundary has some edges in S). The group G for this set is defined by J = {3, 5} and Y = {y3, y5} as shown in the figure. Note that S is not stipulated to itself be connected and hence may contain additional edges outside the path co… view at source ↗

Figures from the paper (1 more)

**Figure 3.4.** Figure 3.4: Illustration of SG where G corresponds to the one in [PITH_FULL_IMAGE:figures/full_fig_p014_3_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

19 extracted references · 16 canonical work pages

[1]

Universality of chaos and ultrametricity in mixed p-spin models

Antonio Auffinger and Wei-Kuo Chen. Universality of chaos and ultrametricity in mixed p-spin models. Communications on Pure and Applied Mathematics , 69(11):2107--2130, 2016

work page 2016
[2]

On absence of disorder chaos for spin glasses on z\^ d

Louis-Pierre Arguin and Jack Hanson. On absence of disorder chaos for spin glasses on z\^ d. 2020

work page 2020
[3]

A relation between disorder chaos and incongruent states in spin glasses on z\^ d z d

L-P Arguin, CM Newman, and DL Stein. A relation between disorder chaos and incongruent states in spin glasses on z\^ d z d. Communications in Mathematical Physics , 367:1019--1043, 2019

work page 2019
[4]

Noise sensitivity of boolean functions and applications to percolation

Itai Benjamini, Gil Kalai, and Oded Schramm. Noise sensitivity of boolean functions and applications to percolation. Publications Math \'e matiques de l'Institut des Hautes \'E tudes Scientifiques , 90:5--43, 1999

work page 1999
[5]

Critical behavior of the three-dimensional ising spin glass

AJ Bray and MA Moore. Critical behavior of the three-dimensional ising spin glass. Physical Review B , 31(1):631, 1985

work page 1985
[6]

Chaotic nature of the spin-glass phase

Alan J Bray and Michael A Moore. Chaotic nature of the spin-glass phase. Physical review letters , 58(1):57, 1987

work page 1987
[7]

Disorder chaos and multiple valleys in spin glasses

Sourav Chatterjee. Disorder chaos and multiple valleys in spin glasses. arXiv preprint arXiv:0907.3381 , 2009

arXiv 2009
[8]

Superconcentration and related topics , volume 15

Sourav Chatterjee. Superconcentration and related topics , volume 15. Springer, 2014

work page 2014

Show all 19 references

[9]

Spin glass phase at zero temperature in the edwards-anderson model

Sourav Chatterjee. Spin glass phase at zero temperature in the edwards-anderson model. arXiv preprint arXiv:2301.04112 , 2023

2023 arXiv
[10]

Disorder chaos in the sherrington--kirkpatrick model with external field

Wei-Kuo Chen. Disorder chaos in the sherrington--kirkpatrick model with external field. 2013

2013
[11]

Chaos in the mixed even-spin models

Wei-Kuo Chen. Chaos in the mixed even-spin models. Communications in Mathematical Physics , 328:867--901, 2014

2014
[12]

Disorder chaos in some diluted spin glass models

Wei-Kuo Chen and Dmitry Panchenko. Disorder chaos in some diluted spin glass models. 2018

2018
[13]

Theory of spin glasses

Samuel Frederick Edwards and Phil W Anderson. Theory of spin glasses. Journal of Physics F: Metal Physics , 5(5):965, 1975

1975
[14]

A simple approach to chaos for p-spin models

Ronen Eldan. A simple approach to chaos for p-spin models. Journal of Statistical Physics , 181(4):1266--1276, 2020

2020
[15]

Ordered phase of short-range ising spin-glasses

Daniel S Fisher and David A Huse. Ordered phase of short-range ising spin-glasses. Physical review letters , 56(15):1601, 1986

1986
[16]

The fourier spectrum of critical percolation

Christophe Garban, G \'a bor Pete, and Oded Schramm. The fourier spectrum of critical percolation. 2010

2010
[17]

Ordering and broken symmetry in short-ranged spin glasses

Charles M Newman and Daniel L Stein. Ordering and broken symmetry in short-ranged spin glasses. Journal of Physics: Condensed Matter , 15(32):R1319, 2003

2003
[18]

The sherrington-kirkpatrick model

Dmitry Panchenko. The sherrington-kirkpatrick model . Springer Science & Business Media, 2013

2013
[19]

Solvable model of a spin-glass

David Sherrington and Scott Kirkpatrick. Solvable model of a spin-glass. Physical review letters , 35(26):1792, 1975

1975

Pith tools

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

[1] [1]

Universality of chaos and ultrametricity in mixed p-spin models

Antonio Auffinger and Wei-Kuo Chen. Universality of chaos and ultrametricity in mixed p-spin models. Communications on Pure and Applied Mathematics , 69(11):2107--2130, 2016

work page 2016

[2] [2]

On absence of disorder chaos for spin glasses on z\^ d

Louis-Pierre Arguin and Jack Hanson. On absence of disorder chaos for spin glasses on z\^ d. 2020

work page 2020

[3] [3]

A relation between disorder chaos and incongruent states in spin glasses on z\^ d z d

L-P Arguin, CM Newman, and DL Stein. A relation between disorder chaos and incongruent states in spin glasses on z\^ d z d. Communications in Mathematical Physics , 367:1019--1043, 2019

work page 2019

[4] [4]

Noise sensitivity of boolean functions and applications to percolation

Itai Benjamini, Gil Kalai, and Oded Schramm. Noise sensitivity of boolean functions and applications to percolation. Publications Math \'e matiques de l'Institut des Hautes \'E tudes Scientifiques , 90:5--43, 1999

work page 1999

[5] [5]

Critical behavior of the three-dimensional ising spin glass

AJ Bray and MA Moore. Critical behavior of the three-dimensional ising spin glass. Physical Review B , 31(1):631, 1985

work page 1985

[6] [6]

Chaotic nature of the spin-glass phase

Alan J Bray and Michael A Moore. Chaotic nature of the spin-glass phase. Physical review letters , 58(1):57, 1987

work page 1987

[7] [7]

Disorder chaos and multiple valleys in spin glasses

Sourav Chatterjee. Disorder chaos and multiple valleys in spin glasses. arXiv preprint arXiv:0907.3381 , 2009

arXiv 2009

[8] [8]

Superconcentration and related topics , volume 15

Sourav Chatterjee. Superconcentration and related topics , volume 15. Springer, 2014

work page 2014