{"id":"2d75eba3-cded-4903-919f-3c8b40b61fba","arxiv_id":"2505.07163","paper_version":1,"verdict":"REJECT","confidence":"HIGH","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":1,"one_line_summary":"The exact spin-elimination idea via Walsh-Hadamard expansion is sound, but the paper's explicit two-spin, three-spin, and other gadget formulas contain sign errors that break the claimed ground-state preservation.","lead":"This paper claims you can shrink an Ising spin system by deleting one spin and replacing it with new interactions among its neighbors, keeping the lowest-energy solution identical. The general idea is real, but several of the printed replacement formulas have sign errors, so the headline applications do not work as written.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Reader's sign-error objection does not reproduce; the central exact-elimination claim is sound, with only application-level overreach in the continuous Hopfield claims.","rationale":"The central claim is that for any spin s_a, min_{s_a} s_a P(s_nbr) = −|P(s_nbr)|, and the FWHT expansion of −|P| produces a reduced Hamiltonian with the same ground-state configurations. This is exact because the local term is linear in s_a for each fixed neighbor configuration; substituting its minimum preserves the global minimum energy and the minimizing assignments. I directly verified the reader's two counterexamples. Eq. (7) with a=1, b=1, c=0 yields β2 = −1 in the paper's convention, and Eq. (8) with b=c=d=1 yields −1/2 for all three pairwise coefficients. The reader appears to have used a different bit-to-sign mapping, producing sign flips that are not present. The exactness of the spin-elimination procedure is therefore not undermined. The weaker point in the manuscript is Section 5: the discrete ground-state preservation does not automatically transfer to the continuous Hopfield dynamics, whose Lyapunov function includes the extra term (1/2)Σ[x_i y_i − ln(1+y_i^2)]. Consequently, claims about suppressing spurious continuous attractors and doubling retrieval frequency are supported only by the reported simulations, not by the exactness theorem. There is also a minor error in the {±1,0} gadget: the substitution t_i = σ_i s_i does not map to ±1 when σ_i = 0; the intended construction should first omit zero-coefficient variables. Neither issue invalidates the central elimination result, but they justify a conditional acceptance rather than an outright rejection.","tokens_in":23058,"tokens_out":44361,"duration_ms":390901,"concrete_test":"Recompute the Walsh coefficients of −|1 + s_b| on {±1}^2 and of −|s_b + s_c + s_d| on {±1}^3 using the paper's indexing, then compare with Eqs. (7) and (8). If the coefficients match β2 = −1 and α3 = γ3 = β3 = −1/2, the reader's counterexample is refuted and the central claim is supported.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The reader's central objection is arithmetically incorrect. Using the paper's stated indexing (h_{σ1σ0} = 1/4 |a + σ1 b + σ0 c|, with +1 mapped to binary bit 1), Eq. (7) for (a,b,c) = (1,1,0) has h0 = h1 = 0 and h2 = h3 = 1/2, so β2 = h0 + h1 − h2 − h3 = −1, not +1. For Eq. (8) with b = c = d = 1, g0 = g1 = g2 = 1/4 and g3 = 3/4, giving α3 = γ3 = β3 = −1/2, not +1/2. Direct Walsh evaluation confirms these coefficients. The replacement min_{s_a} s_a P(s_nbr) = −|P(s_nbr)|, followed by the multilinear FWHT expansion, is mathematically exact for ground-state preservation. I therefore find no load-bearing defect in the central claim. The genuine soft spot is the unproven extension to continuous Hopfield dynamics in Section 5: exact preservation of discrete ground states does not by itself guarantee preservation of continuous attractors or their basins, and the reported memory-retrieval improvements rest on numerical simulation rather than on the exactness theorem. This affects an application claim, not the core elimination construction.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes an exact spin-elimination technique for k-local Ising Hamiltonians. For a spin s_a entering the Hamiltonian as s_a P(s_nbr), the author replaces s_a P(s_nbr) by the unique multilinear polynomial F(s_nbr) equal to -|P(s_nbr)| on all configurations of the neighbors, computed via the fast Walsh–Hadamard transform. This preserves the ground-state energy and allows the eliminated spin to be recovered as s_a = -sign(P(s_nbr)). The paper derives closed-form gadgets for two-spin, three-spin, symmetric n-spin, and several other local structures, and applies the method to Max-Cut on cubic graphs, the J-Möbius ladder, Hopfield associative memories, and QAOA-based integer factorization. I checked the two- and three-spin gadgets in Eqs. (7) and (8): the sign-error objection raised in the review correspondence does not reproduce, since Eq. (7) with (a,b,c)=(1,1,0) gives beta_2=-1 and Eq. (8) with b=c=d=1 gives the pairwise coefficients -1/2. The main exactness argument is sound, but I find genuine errors in the appendix gadget formulas and overreach in the Hopfield application claims.","tokens_in":23308,"tokens_out":36298,"duration_ms":297592,"significance":"The central result is mathematically sound and useful: min_{s_a} s_a P = -|P|, followed by the unique multilinear Walsh expansion, is a parameter-free exact reduction that does not rely on fitting or approximation. I verified the printed two- and three-spin gadgets and the symmetric n-spin formula in small cases; they check out. This gives a clean presentation of a known pseudo-Boolean variable-elimination idea, with concrete closed forms that may be deployable on Ising hardware supporting k-local interactions. However, the appendix formulas Eqs. (26) and (27) are incorrect, and the example built on them in Section 5 does not support the claimed preservation of 16 minima. In addition, the exactness theorem concerns discrete ground states, not the continuous Hopfield dynamics used to justify the memory-retrieval claims. With those issues fixed, the paper would be a worthwhile contribution; in its current form, the application-level claims overreach the proven result.","major_comments":[{"comment":"Equations (26) and (27) are incorrect for generic parameters. The coefficient definitions h_{sigma2 sigma1 sigma0} = (1/4)|a + sigma2 b + sigma1 c + sigma0 d| are appropriate for a linear expression a + sigma2 b + sigma1 c + sigma0 d, but the functions in Eqs. (26) and (27) contain the monomials a s1 s2 + b s1 s3 + c s2 + d s3 and a s1 s2 s3 + b s1 s2 s4 + c s3 + d s4, respectively. For example, in Eq. (26) with a=1, b=0, c=1, d=0, the true function is -|s1 s2 + s2| = -1 - s1, whereas the printed formula gives -1 - s1 s2 s3. Similarly, Eq. (27) with a=1, b=0, c=1, d=0 evaluates to -1 - s1 s2 in reality, but the printed formula gives -1 - s1 s2 s3 s4. These errors propagate into Appendix 8.4: the claimed reduced adder Hamiltonian H_a^(4) = 2 + s6 + s5(1+s6-s7-s8) + s7 s8 is not constant on {±1}^4 (for example, all spins +1 gives energy 4, while (s5,s6,s7,s8)=(1,1,-1,-1) gives energy 8), so it cannot have 16 ground states. The Section 5 claim that the network 'still exhibits 16 valid minima' is therefore unsupported and needs correction.","section":"Appendix 8.3, Eqs. (26)-(27); Section 5 and Appendix 8.4"},{"comment":"The exact-elimination theorem applies to the discrete Ising energy, not to the continuous Hopfield Lyapunov function of Eq. (20) or to basins of attraction of the continuous gradient dynamics. The statements that spin elimination 'reduces the number of spurious states,' 'suppresses spurious attractors,' and 'doubles the frequency of correctly recovering the planted pattern' are numerical observations from a single N=32 experiment (Figure 6), not consequences of the exactness theorem. Because the abstract advertises improved memory retrieval and suppression of spurious attractors, the manuscript should either prove a preservation statement for the continuous dynamics or explicitly frame these claims as empirical demonstrations.","section":"Section 5, Eqs. (19)-(20)"},{"comment":"The asymptotic claim that after n rounds node degrees reach (2n+2), couplings are (2n)-local, and 3N/(n+3) spins remain is stated without proof. After the first elimination rounds, couplings are no longer uniform, so the symmetric n-spin gadget of Eq. (11) cannot be applied directly to arbitrary subgraphs; the general FWHT of Eq. (5) must be applied to each specific P(s_nbr). The spin-count and degree bounds therefore require a precise algorithm and either a proof or an explicit numerical demonstration of the claimed scaling.","section":"Section 3, strategy (i)"}],"minor_comments":[{"comment":"The phrase 'for any fixed configurations s_nbr' should read 'for any fixed configuration s_nbr'.","section":"Section 2, after Eq. (4)"},{"comment":"The set notation in Eq. (12) is inconsistent: the summation is over 'A' but the cardinality condition uses '|S|=2k'; it should be a single symbol throughout.","section":"Eq. (12)"},{"comment":"The caption appears to be missing the word 'from' before 'the upper left to the lower right'.","section":"Figure 4 caption"},{"comment":"The binary-index conventions for h_{sigma1 sigma0}, g_{sigma1 sigma0}, and h_{sigma2 sigma1 sigma0} are terse; given that they have already caused confusion, a short worked example of the bit mapping for Eq. (7) would improve readability.","section":"Eqs. (7)-(8) and Appendix 8.3"}],"recommendation":"major_revision","confidential_remarks":"The core exact-elimination theorem is sound; the rejection based on sign errors in Eqs. (7)-(8) is not supported by direct Walsh evaluation. However, the appendix contains genuine errors in Eqs. (26)-(27), which invalidate the two-bit ripple-carry adder example used to support the Hopfield-network claims. These errors are fixable by replacing the faulty formulas with correct FWHT-based expansions or by removing the example. The manuscript should also be checked against the existing pseudo-Boolean variable-elimination literature; the novelty lies in the gadget compilation and the hardware-oriented applications rather than in the general min-projection principle."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The reader's sign-error objection does not survive a direct check. For Eq. (7) with a=1, b=1, c=0, the paper's own indexing gives h0=h1=0, h2=h3=1/2, so beta2 = h0+h1-h2-h3 = -1, not +1. For Eq. (8) with b=c=d=1, g0=g1=g2=1/4 and g3=3/4, so all three pairwise coefficients are -1/2, as the paper states. The stress-test note is right and the reader's central complaint is wrong.\n\nThe real contribution is a clean packaging of a known identity: replacing s_a P(s_nbr) with the Walsh expansion of -|P(s_nbr)| preserves ground states exactly. The explicit two-spin, three-spin, and n-spin symmetric gadgets are correct evaluations of that identity, and the closed-form n-spin formula is genuinely convenient. The Möbius ladder example is a solid illustration, and the ground-state preservation proof is straightforward and sound.\n\nThe soft spots are where the paper pushes past what the math actually supports. Section 5 claims that spin elimination reduces spurious attractors in continuous Hopfield dynamics. The discrete ground-state exactness theorem does not extend to continuous fixed points or basins, and the paper offers only a single 32-spin numerical experiment, not a proof. The language about improving memory capacity is also loose: the example reduces spurious minima, not storage capacity. The Max-Cut strategy is heuristic — the claim about solving instances roughly 50% larger is an average based on a greedy covering procedure with no worst-case guarantee, and the degree increase to 6 and beyond is a real cost that gets underweighted. The factorization section reduces a ten-qubit compiled Hamiltonian to two qubits, which is a nice demonstration, but the preprocessing or compilation cost of the original Hamiltonian is not discussed, so the \"record-size\" language is speculative.\n\nNone of these issues make the central result wrong. The math is correct, the gadgets are useful, and the applications, while overclaimed, point in a practical direction. This paper should go to peer review, not the desk. A referee should ask for a tempered Section 5, a more careful statement of the Max-Cut scaling, and a cost discussion in the factorization section.","headline":"The reader's sign-error objection is arithmetically wrong; the central exact-elimination claim is sound, but the Hopfield and Max-Cut application claims overreach.","tokens_in":23878,"tokens_out":3979,"would_cite":true,"duration_ms":37977,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A spin in a $k$-local Ising Hamiltonian can be removed exactly, preserving every ground state, by substituting the multilinear expansion of $-|P(s_{\\mathrm{nbr}})|$.","keywords":["exact spin elimination","k-local Ising Hamiltonian","ground-state preservation","Walsh-Hadamard transform","Max-Cut on cubic graphs","QAOA factorization","Hopfield networks","spin gadgets"],"falsifier":"Pick any printed gadget and enumerate the corresponding spin configurations. For example, with $b=c=d=1$, Eq. (8) must equal $-|s_b+s_c+s_d|$ on all eight assignments of $(s_b,s_c,s_d)$, with constant $-3/2$ and each pair coefficient $-1/2$; a single mismatch on one configuration shows the gadget is not the exact elimination it claims to be. The same enumeration test applies to Eqs. (7), (13), and (26).","tokens_in":1807,"feed_emoji":"🧲","tokens_out":3845,"duration_ms":169244,"temperature":0.7,"pith_summary":"The paper tries to establish that the number of spins in a $k$-local Ising Hamiltonian can be reduced exactly, one spin at a time, without changing any ground-state configuration. The terms containing a selected spin $s_a$ form a local block $s_a P(s_{\\mathrm{nbr}})$; for fixed neighbors this block is minimized by $s_a = -\\mathrm{sign}(P(s_{\\mathrm{nbr}}))$, with optimal value $-|P(s_{\\mathrm{nbr}})|$, and the paper replaces the block with the unique multilinear polynomial that equals $-|P|$ on every neighbor configuration. The paper claims this replacement is exact, introduces no auxiliary spins, and runs in a single classical pass, after which the eliminated spin is recovered by $s_a = -\\mathrm{sign}(P(s_{\\mathrm{nbr}}))$. A sympathetic reader would care because spin or qubit count is the principal hardware bottleneck for Ising solvers, quantum annealers, and QAOA circuits, so an exact way to shrink the problem could enlarge solvable instances without altering the answer.","feed_headline":"Exact spin elimination cuts Ising problem size, keeping ground states","feed_subtitle":"Removed spins become neighbor couplings, so Max-Cut, factorization, and Hopfield recall need fewer qubits.","key_machinery":"The load-bearing object is the identity $s_a P(s_{\\mathrm{nbr}}) \\to -|P(s_{\\mathrm{nbr}})| \\to F(s_{\\mathrm{nbr}})$, which turns spin elimination into a function-representation problem: $F$ is the unique multilinear polynomial on the hypercube $\\{\\pm1\\}^d$ that agrees with $-|P|$ everywhere. The fast Walsh-Hadamard transform provides those coefficients in $O(d2^d)$ time, and the paper's printed gadgets, namely the two-spin gadget of Eq. (7), the three-spin gadget of Eq. (8), the $k$-local subgraph gadgets of Eqs. (13) and (26), and the fully symmetric $n$-spin gadget of Eqs. (10)-(11), are the coefficient lists that implement this identity for the specific local blocks used in Max-Cut, Möbius-ladder, factorization, and Hopfield-network applications.","core_discovery":"On the paper's own terms, the central discovery is the replacement identity $s_a P(s_{\\mathrm{nbr}}) \\to -|P(s_{\\mathrm{nbr}})| \\to F(s_{\\mathrm{nbr}})$, where $F$ is the unique multilinear (Walsh) expansion of $-|P|$ on $\\{\\pm1\\}^d$. Because $s_a$ can always choose the sign that minimizes its local energy, the replacement captures the exact optimal response of the removed spin, and the paper argues that the ground-state configurations of the full Hamiltonian are therefore preserved exactly, including their degeneracies. The eliminated spin is recovered by $s_a = -\\mathrm{sign}(P(s_{\\mathrm{nbr}}))$; only when $P=0$ does the removed spin remain undetermined. The generality is claimed through closed-form gadgets for two-, three-, and symmetric $n$-neighbor blocks, with arbitrary local blocks handled by the fast Walsh-Hadamard transform in $O(d2^d)$ time.","pith_inferences":["A testable extension the paper leaves implicit is the choice of elimination order: since several spins can be removed in different sequences, one could search for the order that minimizes the final graph degree or interaction order for a fixed hardware locality budget, and the paper does not study that optimization.","Because $P(s_{\\mathrm{nbr}})=0$ preserves degeneracy exactly, the same reduction could serve as a preprocessing step for sampling all degenerate ground states rather than finding one optimum, although the paper demonstrates this only through preserving degeneracy, not through a sampling algorithm.","If the Walsh-expansion step is treated as a primitive, the technique suggests a general front-end for any Ising solver: eliminate a chosen fraction of spins in one deterministic pass, solve the smaller system, then back-map; all reported applications follow this pattern, but the paper does not analyze worst-case coefficient growth over many elimination rounds."],"forward_implications":["For 3-regular Max-Cut, the paper claims that eliminating one spin per four-neighbor subgraph lets a solver with $N$ spins handle roughly $3N/2$ nodes on $k$-local hardware, and that 2-local-only hardware can still remove on average more than a third of the nodes.","The three-qubit Ising Hamiltonian factorizing $291311$ reduces to two qubits, and the ten-qubit Hamiltonian from the 48-bit factorization reduces to three and then two qubits, lowering QAOA circuit depth and noise exposure.","A Hopfield network that eliminates four or more of its 32 stored-pattern spins keeps its planted global minima and their degeneracy while removing several high-lying spurious minima, which the paper reports as improved retrieval frequency in random trials.","In the Möbius-ladder example, elimination prunes higher-energy states so that transitions between the two lowest states no longer require flipping half the spins, easing the barrier that traps soft-spin solvers.","Every reduced solution maps back to the original problem by $s_a = -\\mathrm{sign}(P(s_{\\mathrm{nbr}}))$, so the smaller Hamiltonian is a true compression of the original rather than a heuristic approximation."],"supporting_citations":[{"why":"Provides the fast Walsh-Hadamard transform used to compute the multilinear coefficients of $-|P(s_{\\mathrm{nbr}})|$.","marker":"[53]"},{"why":"Establishes the pseudo-Boolean optimization and persistency background against which the paper's single-pass exact elimination is distinguished.","marker":"[38]"},{"why":"Supplies the experimental coherent-Ising-machine versus quantum-annealer Max-Cut benchmark on cubic graphs that motivates the larger-instance claim.","marker":"[58]"},{"why":"Provides the three-qubit adiabatic factorization of 291311 that the paper compresses to two qubits.","marker":"[81]"},{"why":"Provides the ten-qubit sublinear-resource factorization Hamiltonian that the paper reduces to three and then two qubits.","marker":"[82]"},{"why":"Defines the Hopfield associative-memory model whose attractors and retrieval dynamics the elimination procedure is applied to.","marker":"[67]"},{"why":"Supplies the classical Hopfield capacity bound that motivates replacing pairwise interactions with higher-order terms.","marker":"[74]"},{"why":"Introduces dense associative memories with higher-order interactions, the model class the paper connects to its $k$-local spin-elimination formulation.","marker":"[71]"}],"fun_headline_variants":["Exact spin elimination shrinks Ising problems, preserves ground states","Remove spins exactly: ground states stay, complexity shifts","Exact spin removal: fewer variables, same ground-state solutions","Exact elimination: Ising ground states preserved with fewer spins","Spin elimination exact: fewer spins, identical ground states"],"cache_read_input_tokens":25984,"weakest_assumption_plain":"The method stands on the claim that every term containing the removed spin has been collected in $s_a P(s_{\\mathrm{nbr}})$ and that the polynomial substituted for $-|P|$ really is its unique multilinear expansion on all neighbor configurations; if either fails on even one configuration, the reduced Hamiltonian's ground states need not match the original's.","fun_headline_variants_meta":{"raw":{"variants":["Exact spin elimination shrinks Ising problems, preserves ground states","Remove spins exactly: ground states stay, complexity shifts","Exact spin removal: fewer variables, same ground-state solutions","Exact elimination: Ising ground states preserved with fewer spins","Spin elimination exact: fewer spins, identical ground states"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001141,"raw_usage":{"total_tokens":4752,"prompt_tokens":981,"completion_tokens":3771,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":597,"completion_tokens_details":{"reasoning_tokens":3688}},"tokens_in":597,"tokens_out":3771,"duration_ms":24734,"temperature":1.0,"reasoning_tokens":3688,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T22:27:12.837884+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Pick any printed gadget and enumerate the corresponding spin configurations. For example, with $b=c=d=1$, Eq. (8) must equal $-|s_b+s_c+s_d|$ on all eight assignments of $(s_b,s_c,s_d)$, with constant $-3/2$ and each pair coefficient $-1/2$; a single mismatch on one configuration shows the gadget is not the exact elimination it claims to be. The same enumeration test applies to Eqs. (7), (13), and (26).","supporting_citations":[{"cited_title":"title Unified matrix treatment of the fast walsh-hadamard transform","cited_arxiv_id":null,"evidence_quote":"Provides the fast Walsh-Hadamard transform used to compute the multilinear coefficients of $-|P(s_{\\mathrm{nbr}})|$."},{"cited_title":"& author Hammer, P","cited_arxiv_id":null,"evidence_quote":"Establishes the pseudo-Boolean optimization and persistency background against which the paper's single-pass exact elimination is distinguished."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the experimental coherent-Ising-machine versus quantum-annealer Max-Cut benchmark on cubic graphs that motivates the larger-instance claim."},{"cited_title":"High-fidelity adiabatic quantum computation using the intrinsic Hamiltonian of a spin system: Application to the experimental factorization of 291311","cited_arxiv_id":"1706.08061","evidence_quote":"Provides the three-qubit adiabatic factorization of 291311 that the paper compresses to two qubits."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines the Hopfield associative-memory model whose attractors and retrieval dynamics the elimination procedure is applied to."},{"cited_title":", author Posner, E","cited_arxiv_id":null,"evidence_quote":"Supplies the classical Hopfield capacity bound that motivates replacing pairwise interactions with higher-order terms."},{"cited_title":"& author Hopfield, J","cited_arxiv_id":null,"evidence_quote":"Introduces dense associative memories with higher-order interactions, the model class the paper connects to its $k$-local spin-elimination formulation."}],"review_version":1}