{"id":"e4872453-2aaf-46ff-838f-edc4311a9c32","arxiv_id":"2411.19604","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"For couplings J_ij=(i^d+j^d)/N^d, the ground state is a two-block configuration whose boundary is set by an algebraic equation, giving an exact benchmark for Ising machines.","lead":"This paper introduces a family of fully connected Ising models where couplings grow with spin indices, and presents a mathematical solution for their ground states. It then uses this exactly solvable family to test Ising machines, reporting that D-Wave hardware deviates from the correct answers at larger sizes.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The two-cluster exactness proof rests on a continuous-limit error bound (Eq. 22) that does not apply to the discontinuous sign-product integrand in Eq. 23, leaving finite-N exactness unproven; the d≤-1 case is asserted via an ad hoc regularization.","rationale":"The reader's weakest assumption correctly identifies the continuous-limit proof as the hinge. This is genuinely load-bearing because Section IV is the only argument excluding multi-domain configurations; the exact formula Eq. (8) merely evaluates two-cluster candidates. The error-control gap is concrete: the sign product is discontinuous, and Eq. (22)'s stated error bound is for smooth integrands. The N^2 prefactor amplifies the quadrature error to O(N), while the discrete energy differences between candidate M values near the optimum are O(1); thus the continuous minimization cannot by itself identify the exact finite-N minimizer. However, the claim is not refuted: the rank-2 identity H = S·V − Σ_u_i u_i suggests an elementary rearrangement proof exists, and the brute-force agreement in Fig. 3 (d=1..5, N≤28) and SimCIM agreement in Fig. 4 are positive evidence. The lack of exhaustive checks for d<0 and the ad hoc regularization for d≤-1 (Eq. D6) strengthen the need for a conditional verdict rather than rejection. Therefore the reader's CONDITIONAL verdict is appropriate and unchanged.","tokens_in":19548,"tokens_out":15699,"duration_ms":128223,"concrete_test":"Re-derive the ground state for finite N using the exact rank-2 structure of J^{(N,d)}: for any configuration, H = S·V − Σ_i u_i with u_i=(i/N)^d, S=Σ_i s_i, and V=Σ_i u_i s_i. For fixed M (fixed S), the rearrangement inequality with monotone u_i shows the energy is extremized by a block of M consecutive up spins. If this exact proof works for all real d, the two-cluster claim and Eq. (8) are established without the continuous-limit error control; if it only works for d>0, the remaining d regimes need exhaustive checks (e.g., N≤24, d∈{−2,−1.5,−0.5,0.5,1.5}).","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that the global ground state of J^{(N,d)}_{ij} in Eq. (2) is always the two-cluster state Eq. (4), with M obtained from Eq. (8). The only general argument for the two-cluster pattern is the continuous approximation of Section IV. The discrete Hamiltonian is replaced by HΛ in Eq. (23) using the Riemann-sum error bound Eq. (22), which the authors state holds for smooth integrands. The integrand in Eq. (23) contains Λ factors of sgn(x−qα) and sgn(y−qα) and is therefore discontinuous. The paper itself warns (Section IV) that the method cannot be used when the derivative of the integrand does not exist, yet it is applied here. Because the prefactor N^2 multiplies the integration, an O(1/N) quadrature error becomes an O(N) absolute error in HΛ, which can exceed the O(1) energy differences between adjacent integer M values near the continuous minimum. Hence the continuous minimization does not rigorously certify the finite-N exactness of Eqs. (8)-(9). For d≤-1, the integral in Eq. (23) diverges and the modified lower limit in Eq. (D6) is an asserted regularization, not a derivation. Figure 3 checks only d=1..5 and N≤28; the d<0 and d≤-1 regimes are not brute-force verified.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces the deterministic fully connected Ising coupling matrix J_{ij}^{(N,d)}=(i^d+j^d)(1-\\delta_{ij})/N^d and claims that, for every real d and every N, the ground state has the ordered two-cluster form of Eq. (4), with the number M of up spins obtained by minimizing the one-parameter function H(M,N,d) in Eq. (8). In the large-N limit the ratio q=M/N is claimed to satisfy the algebraic equation 1+(1+d)q^d-2(2+d)q^{d+1}=0, Eq. (17), with an additional regularized branch, Eq. (D6), for d\\le -1. The proof strategy is to approximate the discrete spin configuration by a continuous sign-product function S(x,q), pass from the Hamiltonian to the continuous functional H_\\Lambda of Eq. (23), and argue that the minimum of H_\\Lambda occurs at \\Lambda=1. The paper then uses this class as a fidelity benchmark: brute-force enumeration for N\\le 28 and SimCIM runs for N=1000 agree with the predicted energies and ratios, while a D-Wave QPU deviates for N>20.","tokens_in":19930,"tokens_out":12585,"duration_ms":105559,"significance":"If the finite-N exactness claim is correct, the paper provides a nontrivial class of fully connected Ising models solvable in polynomial time, with no fitted parameters in the central derivation. This would give the community a useful, deterministic benchmark for testing Ising machines, and the paper's numerical checks against brute-force enumeration and SimCIM are a genuine strength. The post-hoc power-law fit q(d)=0.61 d^{0.13} is descriptive and does not feed back into the derivation, so the circularity burden is low. However, the proof of the two-cluster pattern currently rests on a continuous-limit argument whose error control does not apply to the discontinuous integrand actually used, and the d\\le -1 branch is asserted rather than derived; these gaps are load-bearing for the central claim.","major_comments":[{"comment":"The proof that the ground state has the two-cluster form rests on replacing the discrete sum by H_\\Lambda and on the Riemann-sum error bound in Eq. (22). That bound is stated for smooth integrands, and the text itself warns that the method cannot be used when the derivative of the integrand does not exist. The integrand in Eq. (23) contains \\Lambda factors of sgn(x-q_\\alpha) and sgn(y-q_\\alpha) and is discontinuous, so the O(1/N) quadrature error is not justified. Moreover, because H_\\Lambda carries a prefactor N^2, a quadrature error of order 1/N becomes an absolute energy error of order N, which can exceed the O(1) energy differences between adjacent integer values of M near the continuous minimum. Consequently, the exactness of the finite-N ground-state pattern Eq. (4) is not rigorously established by the continuous-limit argument; it currently rests on the numerical checks in Figure 3 and on heuristic SimCIM runs.","section":"Section IV, Eqs. (21)-(23)"},{"comment":"For d\\le -1 the integral in Eq. (23) diverges, and replacing the lower integration limit by 1/N is an asserted regularization rather than a derivation from the discrete Hamiltonian Eq. (8). The sentence 'The proof for this case follows a similar line of reasoning' is a placeholder, not a proof. This matters because Figure 4 reports q(d) from Eq. (D6) over the whole d\\le -1 range and Section V B explicitly discusses d<0 instances; as written, the d\\le -1 branch of the central claim is unsupported.","section":"Appendix D, Eq. (D6)"},{"comment":"The algebraic reduction contains a sign error in the description of the objective. From Eq. (8), H(M,N,d)=N^{-d}((N-1)F_d(N)-2\\tilde H) with \\tilde H=M F_d(N)+(N-2M)F_d(M), so minimizing the Ising energy is equivalent to maximizing \\tilde H, not minimizing it as stated before Eq. (11). The first-order condition Eq. (12) is unchanged and Eq. (17) is numerically correct, but the argument that the stationary point is the global energy minimum is incomplete: the convexity discussion in Appendix C is for H_1, not for the discrete objective \\tilde H, and the boundary cases M=0,N are not compared there.","section":"Section III, Eqs. (8)-(11)"}],"minor_comments":[{"comment":"'Bernouli' should be 'Bernoulli'.","section":"Eq. (7)"},{"comment":"The displayed error bound (f(b)-f(a))(b-a)/N is not a general quadrature error bound for smooth f; for smooth functions one expects O(N^{-2}) for trapezoidal or midpoint rules, and for monotone f the displayed form needs a separate derivation. Please replace it with a correct statement or a citation.","section":"Section IV, Eq. (22)"},{"comment":"The claim that H_1 is 'always negative for q_1\\neq 1/2 and d>-1' has an exception at d=0, where the critical value is H_1=0. The statement should be restricted to the critical branch and to d>0, or d=0 should be handled separately.","section":"Appendix C"},{"comment":"The phrase 'Without any loss of generality' before changing the lower limit of a divergent integral is inaccurate; the regularization changes the model and should be labeled as such.","section":"Appendix D"},{"comment":"Since brute-force enumeration up to N=28 involves 2^28 configurations, the caption should state explicitly whether the enumeration is exhaustive and how the computational cost was handled.","section":"Figure 3 caption"}],"recommendation":"major_revision","confidential_remarks":"The numerical evidence suggests that the two-cluster claim is likely true at least for d>0, and the class is potentially useful as a benchmark. The main theorem, however, is currently supported by a continuous-limit proof that the paper itself flags as inapplicable to discontinuous integrands, and the d\\le -1 branch is only asserted. These are fixable with a direct discrete exchange argument or by honestly restricting the exactness claim, so major revision rather than rejection seems appropriate."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nQuick take: the central claim is right as far as I can tell - the ground state of J_ij=(i^d+j^d)/N^d is the sorted two-cluster state, M from Eq. (8), and q=M/N satisfies Eq. (17) in the large-N limit. Brute force up to N=28 and SimCIM at N=1000 agree; the D-Wave benchmark is a genuinely nice application. This is a new family, absent from the cited literature, and it gives practitioners a tunable, polynomial-time-checkable benchmark. The paper deserves a serious referee.\n\nWhat it does well: Eq. (8) is exact for two-cluster states, and the numerical checks are credible. The landscape analysis (local minima counts vs. Wishart, DOS) adds useful context.\n\nWhere the soft spots are: the proof that the ground state is always two-cluster is not rigorous. The continuous-limit argument in Section IV uses a Riemann-sum error bound, Eq. (22), stated for smooth integrands, but the sign-product integrand in Eq. (23) is discontinuous. The N^2 prefactor turns an O(1/N) quadrature error into an O(N) absolute error, which is the same order as the energy spacing between adjacent M values near the minimum. So the continuous minimization does not certify the finite-N exactness of Eq. (8). That is a legitimate gap. For d<=-1, Eq. (23) diverges; Eq. (D6) is a lower-cutoff regularization asserted without a proper derivation, and Figure 3 only checks d=1..5, so the d<0 regime is essentially unverified for exactness.\n\nI checked, and there is a much simpler exact discrete proof: because the coupling is additive, H depends only on S = sum_i s_i and A = sum_i i^d s_i. For fixed S>0, the rearrangement inequality puts the up spins on the smallest indices, i.e., the sorted two-cluster is optimal. Then minimizing over M gives Eq. (8). This works for all d>-1 (after the usual spin-flip for S<0), and it removes the need for the continuous argument entirely. The authors should replace their Section IV proof with this, or cite a rigorous bounded-variation error bound if they insist on the continuum.\n\nThe d<=-1 extension needs more work: define the regime, derive the cutoff, and verify it against exact enumeration for small N.\n\nBottom line: right result, flawed proof. I'd send it to peer review, but require the proof fix before acceptance. No code or data was provided, which is worth asking for.\n\nYes, I'd cite it if I work on Ising benchmarks.","headline":"A useful benchmark class of Ising instances with a likely-correct closed-form ground state, but the paper's proof of the two-cluster pattern has a real gap and the d<=-1 extension is asserted rather than derived.","tokens_in":20409,"tokens_out":6682,"would_cite":true,"duration_ms":54625,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["82B20","82B80"],"pacs":["05.50.+q","75.10.Hk"],"model":"deepseek-v4-flash","headline":"A deterministic family of fully connected Ising models has exact two-cluster ground states, found in polynomial time.","keywords":["Ising model","ground state","continuous approximation","two-cluster ansatz","fully connected couplings","Ising machine fidelity","energy landscape","quantum annealer benchmark"],"falsifier":"Run an exact search for $N=32$ at fractional $d$ values such as $d=0.3$ and compare the true ground state with the two-cluster minimizer from Eq. (8): any ground state with more than two spin clusters, or any mismatch in the minimizing $M$, would falsify the transfer from the continuous functional to the discrete Hamiltonian.","tokens_in":19378,"feed_emoji":"🧲","tokens_out":8602,"duration_ms":65241,"temperature":0.7,"pith_summary":"The paper introduces a deterministic family of fully connected Ising couplings $J^{(N,d)}_{ij}=(i^d+j^d)(1-\\delta_{ij})/N^d$ and claims that for every $d>-1$ the ground state is always two adjacent clusters, one of up spins and one of down spins. This reduces the search over $2^N$ spin configurations to a one-variable minimization over the cluster size $M$, taking $O(N)$ time. The authors derive a closed-form equation for the large-$N$ cluster ratio $q=M/N$, and they report that the prediction matches brute-force enumeration and a simulated coherent Ising machine, while a physical quantum annealer deviates for larger problem sizes. The practical point is that this class is an exact, polynomial-time-verifiable benchmark for testing how faithfully Ising hardware encodes a problem.","feed_headline":"Exact ground states found for a class of Ising models","feed_subtitle":"One variable gives the ground state, and the benchmark exposes where a quantum annealer loses fidelity.","key_machinery":"The load-bearing object is the continuous spin function $S(x,\\mathbf q)=(-1)^\\Lambda \\prod_{\\alpha=1}^\\Lambda \\mathrm{sgn}(x-q_\\alpha)$, which encodes all domain-wall positions as discontinuities and turns the discrete Hamiltonian into a Riemann-sum integral. The factorization $H_\\Lambda=\\frac{N^2}{1+d}QP$ is what makes the argument work: it separates the geometry of the boundaries from the $d$-dependent weights, so the energetics reduces to comparing signs of two alternating sums. A second ingredient is the power-sum identity $F_d(N)=\\sum_{i=1}^N i^d$, which converts the two-cluster energy into a closed algebraic form and yields the transcendental equation for $q$.","core_discovery":"The central discovery is that the energy of any configuration in this coupling family can be represented, in the continuous limit, by a functional $H_\\Lambda(d,\\mathbf q)$ built from sign functions at the domain-wall positions $\\mathbf q=(q_1,\\dots,q_\\Lambda)$. This functional factorizes as $H_\\Lambda=\\frac{N^2}{1+d}QP$, where $Q$ and $P$ are alternating sums over the boundaries, and the paper argues that only the two-cluster case $\\Lambda=1$ can make the product negative while remaining stable; configurations with more domain walls either have zero energy at their critical points or relax to fewer walls. The ground state is therefore the two-cluster configuration, with the cluster split obtained from $1+(1+d)q^d-2(2+d)q^{d+1}=0$ in the large-$N$ limit. For finite $N$, the closed expression $H(M,N,d)=N^{-d}\\big((N-2M-1)F_d(N)+(4M-2N)F_d(M)\\big)$ with $F_d(N)=\\sum_{i=1}^N i^d$ gives an $O(N)$ recipe that the authors verify against exhaustive enumeration.","pith_inferences":["The continuous-to-discrete transfer is the step that would benefit from a rigorous finite-$N$ error bound for the discontinuous sign-product integrand; without it the exactness of the two-cluster ansatz for every finite $N$ remains an asymptotic inference.","The same construction may work for other monotone coupling functions $f(i)+f(j)$, and the two-sum factorization suggests a route to additional exactly solvable fully connected families.","Because tuning $d$ moves the landscape from many local minima to a smoother spectrum, the family could serve as a controlled ruggedness knob in experiments on annealing dynamics.","A direct finite-$N$ check of the minimal $M$ formula at, say, $N=32$ for several fractional $d$ values would cost little and would test whether the transfer already holds in the regime where quantum hardware begins to fail."],"forward_implications":["For any $d>-1$ and large $N$, the ground-state magnetization fraction $q$ is determined by a single equation, so no heuristic search is needed for this coupling class.","Brute-force validation is replaced by an $O(N)$ check, making $J^{(N,d)}$ a scalable exact-reference benchmark for Ising minimizers.","A physical quantum annealer showed measurable deviation from the analytical ground state starting near $N=20$, while a simulated coherent Ising machine stayed on the predicted ground state up to $N=1000$, isolating encoding fidelity as the source of the gap.","Permutation invariance lets one choose $d$ so that a prescribed spin configuration with a given up-spin fraction $q$ is the ground state, turning the class into a generator of instances with known answers."],"supporting_citations":[{"why":"Supplies the power-sum identity that turns the two-cluster Hamiltonian into the closed form $H(M,N,d)$.","marker":"[35]"},{"why":"Provides the simulated coherent Ising machine algorithm used to validate predicted ground states at large $N$.","marker":"[31–34]"},{"why":"Documents the quantum annealer hardware whose outputs are compared against the analytical benchmark.","marker":"[37]"},{"why":"Supplies the Wishart planted ensemble used as a comparison baseline for local-minima counts and density of states.","marker":"[29]"},{"why":"Establishes NP-hardness of general Ising ground-state search, giving context for why an exactly solvable class is useful.","marker":"[4]"}],"fun_headline_variants":["Continuous reformulation yields exact ground states for Ising class","Exact Ising ground states via continuous limit, benchmark for annealers","New analytic route to Ising ground states reveals machine fidelity","Continuous trick solves some Ising models exactly, tests quantum annealers","Ising ground states from continuous approximation, with fidelity test"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof leans on replacing the discrete spin sum by a Riemann integral whose integrand contains a discontinuous sign product, and the stated $O(1/N)$ error bound is for smooth integrands; if that transfer fails at finite $N$, the exactness of the two-cluster ground state for fixed $N$ is not established.","fun_headline_variants_meta":{"raw":{"variants":["Continuous reformulation yields exact ground states for Ising class","Exact Ising ground states via continuous limit, benchmark for annealers","New analytic route to Ising ground states reveals machine fidelity","Continuous trick solves some Ising models exactly, tests quantum annealers","Ising ground states from continuous approximation, with fidelity test"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000725,"raw_usage":{"total_tokens":3238,"prompt_tokens":918,"completion_tokens":2320,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":534,"completion_tokens_details":{"reasoning_tokens":2234}},"tokens_in":534,"tokens_out":2320,"duration_ms":13235,"temperature":1.0,"reasoning_tokens":2234,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T06:02:23.746269+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run an exact search for $N=32$ at fractional $d$ values such as $d=0.3$ and compare the true ground state with the two-cluster minimizer from Eq. (8): any ground state with more than two spin clusters, or any mismatch in the minimizing $M$, would falsify the transfer from the continuous functional to the discrete Hamiltonian.","supporting_citations":[{"cited_title":"Yamamoto, K","cited_arxiv_id":null,"evidence_quote":"Supplies the power-sum identity that turns the two-cluster Hamiltonian into the closed form $H(M,N,d)$."},{"cited_title":"Zeng, X.-P","cited_arxiv_id":null,"evidence_quote":"Documents the quantum annealer hardware whose outputs are compared against the analytical benchmark."},{"cited_title":"Bak and R","cited_arxiv_id":null,"evidence_quote":"Supplies the Wishart planted ensemble used as a comparison baseline for local-minima counts and density of states."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes NP-hardness of general Ising ground-state search, giving context for why an exactly solvable class is useful."}],"review_version":1}