REVIEW 4 minor 5 references
Signed circulants at the Ramanujan bound
T0 review · 0 major / 4 minor · reviewed 2026-08-01 · deepseek-v4-flash
Pith's one-line read A parity condition on quadrilaterals drives a signed degree-4 cyclic graph to a spectral radius below 2√2.
desk verdict Explicit sub-Kesten signings of C_n(1,2); the spectral math is correct, the main conjecture is open, and the paper deserves refereeing. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing object is the signed adjacency matrix A_σ = C_1 + D C_2, where C_1 and C_2 are the unsigned adjacency matrices of the step-1 and step-2 circulants and D is the diagonal matrix with entries (-1)^j. Its invariance on the two-dimensional subspaces spanned by the pair of exponential vectors indexed by k and k+n/2 turns the spectral problem into eigenvalues of a traceless 2×2 matrix, hence ±√(a²+b²) with a=2cosθ_k, b=2cos2θ_k. The combinatorial selector is the mod-2 system s·χ_{Q_i}=1; its incidence matrix has rank n-1, so the solution set is four switching classes, coordinatized by the sign of one triangle and the sign product along the step-1 spanning cycle. For the twisted cla
What would settle it
Exhaustively enumerate all 2^{n+1} switching classes for n=20 and compare the minimum spectral radius with ρ_-(20)=2√(cos²(π/20)+cos²(π/10)); a class with smaller radius would refute Conjecture 3. This is the first even size past the paper's verification range (n≤18).
Extended reading notes
Core claim
The central discovery is that the parity rule 'every quadrilateral unbalanced' selects the near-optimal signings. For even n the solutions form four switching classes; the representative with +1 on step-1 edges and (-1)^i on step-2 edges has spectrum ±2√(cos²θ_k+cos²2θ_k), spectral radius exactly 2√2. The two classes with odd sign product around the step-1 spanning cycle have the smaller radius ρ_-(n)=2√(cos²(π/n)+cos²(2π/n))<2√2. The paper proves these identities by diagonalizing the signed adjacency matrix on the invariant subspaces spanned by paired exponential vectors, and conjectures, with exhaustive verification through n=18, that the smaller value is the minimum over all switching cla
Load-bearing premise
The proof assumes without proof that for n≥10 every quadrilateral of C_n(1,2) is one of the n cycles Q_i=(i,i+1,i+3,i+2); if another four-cycle existed for some n, the mod-2 system (1) would no longer encode the intended condition, and the four-class conclusion would not follow.
Editorial extensions
If this is right
- For every even n≥10 there is an explicit signing of C_n(1,2) with spectral radius 2√2, strictly below the degree-4 bound 2√3; for the twisted classes the radius is even smaller, 2√(cos²(π/n)+cos²(2π/n)).
- The quadrilateral parity system has exactly four switching-class solutions; every signing in the family has one of two spectral radii depending only on the sign product around the step-1 spanning cycle.
- No signing can unbalance all quadrilaterals when n is odd; at most n-1 quadrilaterals can be made unbalanced.
- If Conjecture 3 holds, the twisted classes are global optimizers among all 2^{n+1} switching classes, a flux-minimization statement for this family; this is verified exhaustively through n=18.
- At n=8, requiring all ten quadrilaterals to be unbalanced selects exactly the twisted classes and excludes the 2√2 class.
Reading between the lines
- The paper leaves implicit that the two-dimensional block structure is a recipe: for any two-step circulant with a shift-invariant quadrilateral incidence matrix, the same rank calculation would isolate a small family of candidate optimizer signings, making the method transferable.
- The phase-twisted operator A(φ) has not been optimized over φ by the paper; a natural extension is to minimize the spectral radius over φ between 0 and π/n and see whether ρ_-(n) is the endpoint minimum.
- The contrast between random signings, which concentrate above the floor, and the parity family, which sits below it, suggests a testable general mechanism: constraining all short even cycles to be unbalanced may be a deterministic way to beat random spectral concentration on other 4-regular graphs.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper studies signings of the 4-regular circulant graph C_n(1,2) subject to the F2 constraint that every quadrilateral is unbalanced. For even n≥10 it proves that this constraint system has exactly four switching classes, and it computes the spectrum of the canonical class (step-1 edges all +1, step-2 edges alternating signs) as {±2√(cos²θ_k+cos²2θ_k)}, giving spectral radius exactly 2√2, below the Kesten bound 2√3. It then shows the quadrilateral condition is equivalent to alternating triangle fluxes, parametrizes the four classes by (τ0, α), proves that the spectral radius depends only on α, and computes ρ_-(n)=2√(cos²(π/n)+cos²(2π/n))<2√2 for the two twisted classes. For odd n the system is inconsistent; the exceptional n=8 case is handled separately. The paper concludes with a conjecture, supported by exhaustive enumeration for n=8,...,18, that ρ_-(n) is the global minimum over all 2^{n+1} switching classes.
Significance. If the result stands, it supplies an explicit, parameter-free family of 4-regular signings with spectral radius strictly below the Kesten floor, and an exactly solvable instance of the companion's parity-family method. The spectral derivations are self-contained and do not rely on the companion manuscript for their main conclusions. The exhaustive enumeration is reproducible from the supplied repository and gives solid computational evidence for the conjecture. The conjecture itself, with its Lieb flux-phase analogy, is a meaningful open problem. The paper is honest in distinguishing theorem from conjecture.
minor comments (4)
- [§1 (quadrilateral classification)] The assertion that for n≥10 the quadrilaterals of C_n(1,2) are exactly Q_i=(i,i+1,i+3,i+2) is load-bearing: it converts 'every quadrilateral unbalanced' into system (1) and underlies the rank n−1 incidence argument. It is stated without proof. I checked the claim and it is correct: a 4-cycle is a closed walk with four steps in {±1,±2}; for n≥10 the total displacement has absolute value at most 8<n, so the step multiset must sum to zero over Z, and the only simple such walks are the Q_i. Please insert a short proof of this classification and flag the n=8 exception explicitly at that point.
- [Proposition 2 proof] The diagonal-unitary conjugacy criterion is invoked without proof: two Hermitian matrices with the same connected support, all off-diagonal entries of modulus one, and equal cycle holonomy are conjugate by a diagonal unitary. This is standard, but since it is essential for the isospectrality of A(φ) with the real signed adjacency matrix, a one-sentence justification should be added. It would also be helpful to state explicitly that checking the triangle holonomies and the Hamilton-cycle holonomy suffices because those cycles form a basis of the cycle space.
- [Remark 5 / Conjecture 3] The side claim that uniformly random signings concentrate above the Kesten bound 2√3 from n≈480 is stated without proof and relies on the companion manuscript [2]. Since [2] is not part of this submission, this assertion should either be proved, replaced by a precise bound, or explicitly marked as conditional on [2].
- [Proposition 2 proof] The sentence 'the maximum over the shifted lattice is attained at π/n' is terse. The full argument is that g(π/n)>1 for all even n≥8, while all other shifted momenta in [π/2, π−π/n] have g(θ)≤1 by symmetry of g and the already-established monotonicity on [0,π/2]. Spelling this out would improve readability.
Circularity Check
No significant circularity: the spectral and class-count derivations are self-contained
full rationale
The paper's central claims are derived from self-contained linear algebra and Fourier diagonalization, not from the companion manuscript [2] or from assuming the target results. The only self-reference is [2], which is used for motivation and for the general parity-family context ('This is the parity family of the companion manuscript [2]'), but Proposition 1 proves consistency and the four-class structure from the coefficient comparison of the quadrilateral constraints, and Propositions 1–2 compute spectra by explicit Fourier blocks. The twisted-class spectrum is obtained by exhibiting a Hermitian matrix A(φ) with the same holonomy as the target real signed adjacency matrix and then diagonalizing it; this is a legitimate isospectrality reduction, not a fit of the desired eigenvalue formula. The asserted quadrilateral classification for n≥10 ('its quadrilaterals, for n≥10, are exactly Q_i=(i,i+1,i+3,i+2)') is stated without proof but is an independent combinatorial fact; its truth does not depend on the results being derived, and the paper does not use the spectral conclusions to justify it. The open Conjecture 3 is explicitly conjectural and its Lieb flux-phase analogy is not used as a proof step. No target quantity is fitted, renamed, or imported from a self-citation chain, so no circular step is present.
Assumptions & free parameters
assumptions (5)
- standard math Cycle-sign pattern determines switching class; number of switching classes is 2^{m-n+1}.
- domain assumption For n≥10, the quadrilaterals of C_n(1,2) are exactly Q_i=(i,i+1,i+3,i+2), one per vertex.
- standard math Any two Hermitian weightings on a connected graph with modulus-one entries and equal cycle holonomy are conjugate by a diagonal unitary.
- standard math The Fourier basis diagonalizes C_1 and C_2, and D maps f_k to f_{k+n/2}.
- standard math Rotation by one vertex is an automorphism of C_n(1,2) preserving the spectrum.
Cite this review
Pith. "Pith review of Signed circulants at the Ramanujan bound." pith.science (2026). https://pith.science/paper/ZARYCODB
@misc{pith2026260718334,
author = {Pith},
title = {Pith review of: Signed circulants at the Ramanujan bound},
year = {2026},
howpublished = {\url{https://pith.science/paper/ZARYCODB}},
note = {Machine review of arXiv:2607.18334}
}
abstract
For the circulant graph $C_n(1,2)$ with $n\ge10$ even, the $\F_2$ system requiring every quadrilateral to be unbalanced is consistent and its solutions form exactly four switching classes. We show that the class containing the signing which is $+1$ on step-$1$ edges and $(-1)^i$ on step-$2$ edges has spectrum $\{\pm2\sqrt{\cos^2\theta_k+\cos^2 2\theta_k}\}$ and spectral radius exactly $2\sqrt2$, well below the Kesten bound $2\sqrt3$; that the quadrilateral system is equivalent to alternating triangle fluxes, so that the four classes are coordinatized by $(\tau_0,\alpha)$ and the spectral radius depends only on the Hamilton-cycle holonomy $\alpha$; and that the two twisted classes attain $\rho_-(n)=2\sqrt{\cos^2(\pi/n)+\cos^2(2\pi/n)}<2\sqrt2$. Exhaustive enumeration of all $2^{n+1}$ switching classes for $n\in\{8,10,12,14,16,18\}$ shows that $\rho_-(n)$ is the global minimum in every case, and we conjecture this for all even $n$; the lower bound is a flux-minimization statement in the sense of Lieb's flux-phase theorem. For odd $n$ the quadrilateral system is inconsistent.
Reference graph
Works this paper leans on
-
[2]
Author,Parity families and a kernel-averagedL-function for near-Ramanujan signings, com- panion manuscript
-
[1]
Y. Bilu, N. Linial,Lifts, discrepancy and nearly optimal spectral gap, Combinatorica 26 (2006), 495-519
2006
-
[3]
E. H. Lieb,Flux phase of the half-filled band, Phys. Rev. Lett. 73 (1994), 2158-2161
1994
-
[4]
A. W. Marcus, D. A. Spielman, N. Srivastava,Interlacing families I: Bipartite Ramanujan graphs of all degrees, Ann. of Math. 182 (2015), 307-325
2015
-
[5]
Z. Xu, X. Zhang,An improved upper bound for the Bilu-Linial conjecture, arXiv:2606.28797 (2026). 4
arXiv 2026
Reviewed August 1, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.