{"id":"ab80a21a-a684-4fc1-89aa-37e74aad952a","arxiv_id":"2501.02681","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Simulations of small Coherent Ising Machines suggest non-classical initial states and time-varying couplings can improve ground-state search success, but the effect is not separated from a classical amplitude advantage.","lead":"This paper simulates a small quantum optical Coherent Ising Machine with 3 to 5 coupled light pulses, and finds that non-classical initial states and time-varying couplings can reach the correct Ising ground state faster in low-noise regimes. The result is a candidate source of quantum advantage for optimization hardware, but the comparison lacks a classical benchmark with equal photon number, so the advantage is not yet proven.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The quantum-superposition advantage claim is untested against an equal-photon control: cat-state speed-up is compared only to vacuum, not to coherent states or classical mixtures of the same amplitude, and no classical CIM comparison is reported.","rationale":"The reader's weakest_assumption identifies exactly the load-bearing concern: the contrast between high-photon-number nonclassical states and zero-photon vacuum cannot separate quantum superposition effects from amplitude effects. The abstract's promised 'comparisons with classical CIM models' are absent from the text, making the baseline gap even more central. I give credit where the paper has independent support: two separately coded MCWF implementations, explicit truncation checks, and honest limitation statements in the introduction. These make the numerical method credible, so the issue is not soundness of the simulation itself. The missing controls are, however, decisive for the central claim: without an equal-amplitude coherent-state or classical-mixture arm, and without a mean-field classical CIM simulation, the reported speed-up cannot be attributed to quantum coherence. This is an addressable gap, not a demonstrated fatal error, so the appropriate outcome is a conditional verdict pending the baseline simulations.","tokens_in":12365,"tokens_out":4731,"duration_ms":52893,"concrete_test":"Run the same MCWF simulations used for Figs. 3 and 4 (same lambda, g, J, T, Nsteps, Ncut, and 10^4 trajectories) with three additional initial conditions at alpha = 2.582: (i) coherent state |alpha>^⊗M, (ii) coherent state |-alpha>^⊗M, and (iii) the classical mixture (|alpha><alpha| + |-alpha><-alpha|)/2 per mode. Also integrate the mean-field equation Eq. 8 from alpha_i(0) = 2.582 with the same coupling schedules and compare the success-rate curves. If any of these match the cat-state curves, the superposition-specific advantage is not established; if all fall clearly below, the claim survives.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim (abstract; Section IV) is that quantum superpositions and time-varying couplings improve CIM success rates and time-to-solution in a low-dissipation regime. The superposition evidence, however, compares only two extremes: vacuum |0>^M with zero photons against cat/entangled states with alpha = 2.582, i.e. 6.667 mean photons per mode (Section III A, Eq. 29). This confounds 'quantum superposition' with initial photon number and amplitude. A coherent state |alpha> or the classical mixture (|alpha><alpha| + |-alpha><-alpha|)/2 has the same photon number with no inter-component coherence; if either reproduces the speed-up, the quantum-advantage claim collapses into an amplitude effect. The abstract also explicitly promises 'comparisons with classical CIM models,' but the full text only quotes the mean-field equation Eq. 8 and never simulates it. Without these baseline controls, the statement that 'quantum tunneling effects in this strong coupling limit can overcome trapping in false minima' is unsupported. The numerical methods themselves (two independent MCWF implementations in xSPDE4 and QuTiP, truncation checks in Fig. 2) look sound; the gap is in experimental design and baseline selection, not in the simulation machinery.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents Monte Carlo wave-function simulations of a coherent Ising machine (CIM) with up to five modes, using both xSPDE4 and QuTiP implementations. The authors compare initial vacuum states with quantum superposition states (cat product and entangled states) and study time-dependent coupling and dissipation schedules. They report improved success rates and faster time-to-solution for non-classical initial states in a low-dissipation regime, and interpret the results as evidence for potential quantum computational advantage. They also compute purity to analyze decoherence.","tokens_in":12604,"tokens_out":6169,"duration_ms":60525,"significance":"If the central claim were fully supported, the paper would provide indicative evidence that quantum coherence in the initial state can improve small-scale CIM optimization. The manuscript has clear strengths: the MCWF method is standard and implemented with two independent codes, the photon-number cutoff is checked (Fig. 2), and sampling and time-step errors are reported for the 4-mode case. However, the significance is currently limited because the claimed quantum advantage rests on a comparison that confounds quantum superpositions with initial photon number, and the abstract promises classical CIM comparisons that are not actually presented.","major_comments":[{"comment":"The central comparison conflates quantum superposition with initial photon number. The cat product state ψ_sup uses α = 2.582, giving mean photon number 6.667 per mode, whereas the vacuum state ψ_vac has zero photons. A coherent state with the same amplitude, or the classical mixture (|α⟩⟨α| + |-α⟩⟨-α|)/2, would have the same photon number without inter-component coherence. If either control reproduces the observed speed-up, the claim that quantum superpositions are responsible collapses into an amplitude or photon-number effect. This control is essential to the abstract's claim of quantum computational advantage and is presently missing.","section":"Section III A, Eq. (29), Fig. 3"},{"comment":"The abstract states that 'comparisons with classical CIM models give evidence that quantum tunneling effects in this strong coupling limit can overcome trapping in false minima,' but the manuscript never simulates the classical mean-field equation Eq. (8) or any other classical CIM model. The only benchmark used is the vacuum initial state, which is a quantum state, not a classical CIM model. Without these simulations, the claim about overcoming trapping in false minima is unsupported.","section":"Abstract and Section II, Eq. (8)"},{"comment":"The success-rate differences between initial states are reported without error bars or statistical significance tests. For the M = 5 case the improvement is described as marginal, and for the low-dissipation case the advantage is short-lived. Since the success rates in Fig. 4 are only about 0.13-0.16 and the sampling error is ~4e-3, the reader cannot assess whether the differences in Figs. 3 and 5 are statistically meaningful. Confidence intervals or a statistical test should be provided for the main comparisons.","section":"Fig. 3 and Fig. 5"},{"comment":"The time-dependent coupling speed-up in Fig. 7(a) is presented without sampling error or statistical significance analysis. The claim that a linear increase of J_coef(t) improves the maximum success rate is a central part of the paper's message, and the absence of error estimates weakens the reliability of this conclusion.","section":"Section IV, Fig. 7(a)"}],"minor_comments":[{"comment":"The notation for the Hermite integrals is inconsistent: Eqs. (21)-(22) write Λ(m_i, m_i') as the integral of Hermite polynomials without the Gaussian weight or normalization, while Eq. (25) gives a formula that apparently includes the e^{-x^2} weight. The text should clarify that the Hermite functions (including the Gaussian factor) are used in the quadrature probability, or define the integrals consistently.","section":"Section II B, Eqs. (21)-(25)"},{"comment":"The purity estimator in Eq. (28) includes the i = j terms in the double sum, which contribute unity and can bias the sample purity upward, especially for small trajectory numbers. Excluding self-terms or using a standard unbiased estimator would be more appropriate.","section":"Section II C, Eq. (28)"},{"comment":"The claim that the simulations involve Hilbert spaces exceeding 10^7 dimensions is not demonstrated for the presented cases: with M = 5 and N = 16 the dimension is 17^5 ≈ 1.4 × 10^6. If an M = 6 case was simulated, it should be shown; otherwise the statement is misleading.","section":"Introduction and Section III B"},{"comment":"The success-rate curves for the max-cut problem with different g(t) schedules are shown without error estimates, and the text does not report the number of trajectories or time steps used. This information is needed to judge the significance of the observed improvement.","section":"Section IV A, Fig. 9"}],"recommendation":"major_revision","confidential_remarks":"The paper makes a strong claim of quantum computational advantage, but the evidence is confounded by the lack of equal-photon controls and the absence of the promised classical CIM simulations. The simulation machinery appears sound, and the issues are fixable by adding the missing baselines and error analysis. This is a borderline case between major revision and reject; I lean toward major revision because the technical core is standard and the missing controls are well-defined additions."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know two things about this paper. First, it is a careful MCWF simulation of small coherently coupled CIMs, and the numerical work looks solid. Second, the central claim—that quantum superpositions give a computational advantage—is not actually tested, because the cat states are only compared to vacuum, not to equal-photon coherent states or classical mixtures. The speed-up might just be an amplitude effect.\n\nWhat is new: applying MCWF to non-Gaussian CIM states with M=3–5, with Hilbert spaces exceeding 10^7, using two independent implementations (xSPDE4 and QuTiP). The truncation checks and reported sampling/time-step errors in Fig. 2 and the text are good practice. The purity analysis is a useful addition, showing that the initial cat states decohere quickly in the high-dissipation regime. As a simulation methods paper, this is competent and readable.\n\nThe soft spot is the baseline. The abstract explicitly promises 'comparisons with classical CIM models,' but the full text only quotes the mean-field equation (8) and never simulates it. The cat product state in Eq. (29) uses alpha=2.582, i.e., 6.667 mean photons per mode, while the vacuum has zero. Nothing in the paper rules out a coherent state with the same amplitude producing the same transient success-rate enhancement. That is a load-bearing gap. The hand-picked J12 values near degeneracy and the time-dependent schedules are a minor concern about generality, but the paper is honest that these are transitional studies.\n\nWho should read it: people working on CIM simulation methods or on whether non-classical initial states can help quantum optimization. It is not a demonstration of quantum advantage. The method section could be useful to cite, but I would not cite the advantage claim.\n\nFor peer review: yes, this deserves refereeing. The simulation machinery is solid, the missing controls are specific and addressable, and the paper's own framing is appropriately cautious. But an editor should send it with the expectation that the authors add an equal-photon coherent-state control and actual classical CIM simulations before the advantage claim can stand.","headline":"Solid MCWF simulations of small CIMs, but the quantum-advantage claim is untested against an equal-photon control.","tokens_in":13179,"tokens_out":2526,"would_cite":false,"duration_ms":23075,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P68","81V80"],"pacs":["42.65.Yj","03.67.Lx"],"model":"deepseek-v4-flash","headline":"This paper claims that starting a Coherent Ising Machine from quantum superposition states and ramping its couplings can find Ising ground states faster than the standard vacuum-start protocol, as seen in Monte Carlo wave-function…","keywords":["coherent Ising machine","Monte Carlo wave-function method","quantum computational advantage","optical parametric oscillator","Ising ground state","max-cut problem","quantum decoherence","time-dependent coupling"],"falsifier":"Simulate the same three-, four-, and five-mode problems with each mode initialized in a classical coherent state $\\lvert\\alpha\\rangle$ with $\\alpha = 2.582$ (the amplitude used for the cat states) and compare the time to maximum success probability with the paper's Fig. 3; if the coherent-state curves match the cat-state curves, the quantum-advantage claim collapses. A second check is to run a classical (mean-field or positive-P) CIM simulation with the same time-dependent couplings and see whether the same success-rate improvements appear without any non-classical initial state.","tokens_in":12113,"feed_emoji":"⚛️","tokens_out":10008,"duration_ms":82886,"temperature":0.7,"pith_summary":"This paper claims that a Coherent Ising Machine (CIM) — an optical network of parametric oscillators that searches for Ising-model ground states — can reach the correct solution faster when it starts from quantum superposition (cat) states and when the oscillator couplings are ramped up over time, rather than starting from the classical vacuum state with fixed couplings. To test this in a regime where Gaussian approximations are invalid, the authors simulate the full quantum master equation with Monte Carlo wave-function (quantum jump) methods, handling Hilbert spaces exceeding $10^7$ dimensions. In their small three- to five-mode simulations, the superposition-initialized machine reaches higher success probabilities sooner, and time-dependent couplings provide a further speed-up, which they interpret as evidence that quantum effects can help overcome trapping in false minima. Their purity analysis shows that decoherence is the main obstacle, so the advantage appears only in a low-dissipation regime with time-varying parameters.","feed_headline":"Superposition states speed up Ising-machine simulations","feed_subtitle":"Monte Carlo wave-function runs in 10-million-dimensional Hilbert spaces show faster success with cat states and time-varying couplings.","key_machinery":"The load-bearing object is the Monte Carlo wave-function (MCWF) method, a quantum-jump algorithm that evolves a stochastic wave function under a non-Hermitian effective Hamiltonian and applies Lindblad jump operators at random times, so the computational cost scales with the Hilbert-space dimension rather than its square, as in master-equation methods. The simulations use number-state cutoffs (up to $16$ photons per mode, truncation checked by varying the cutoff) and evaluate the success rate from quadrature probability distributions via Hermite integrals; they also compute the state purity as a decoherence diagnostic. The system is the standard CIM master equation: $M$ degenerate optical parametric oscillators with one- and two-photon damping and coherent coupling, initialized either in vacuum, in a product of cat states, or in an $M$-partite entangled state, with time-dependent coupling coefficient $J_{\\mathrm{coef}}(t)$ and time-dependent nonlinear dissipation $g(t)$.","core_discovery":"The central claim is that the Coherent Ising Machine can exhibit a quantum advantage in the speed of finding Ising ground states when it is initialized in a non-classical superposition of coherent states (a cat product state) and operated with time-varying couplings in a low-dissipation regime. Using Monte Carlo wave-function simulations that do not assume Gaussianity, the authors compare vacuum, cat-product, and entangled initial states on frustrated three-, four-, and five-spin problems and on a five-node max-cut problem. They find that the cat-product state (coherent amplitude $\\alpha = 2.582$, mean photon number $6.667$ per mode) reaches maximum success probability faster than the vacuum state, while the entangled state gives mixed results; increasing the coupling strength fourfold via a linear ramp, or decreasing the two-photon dissipation from $0.9$ to $0.6$, further raises the success rate. The purity analysis shows that the initial quantum state decoheres quickly under strong dissipation, so the advantage survives only in the low-dissipation regime where time-dependent couplings are used.","pith_inferences":["The paper does not compare the cat-product initial state with a classical coherent state of the same amplitude, so the observed speed-up could in principle come from the higher initial photon number rather than from quantum superposition; a coherent-state baseline would separate these.","The time-dependent coupling strategies resemble quantum-annealing schedules, suggesting a testable connection: if the CIM speed-up mirrors transverse-field annealing on the same small instances, the effect may be generic to ramped interactions rather than specific to optical parametric oscillators.","The success-rate measure uses only the sign of the quadrature, discarding amplitude information; a threshold-based readout using the full joint distribution might change the apparent advantage and is worth testing.","Because the paper's own caveat limits conclusions to small systems, a natural extension is to simulate larger $M$ with sparse or tensor-network wave-function methods to see whether the advantage grows, saturates, or reverses with system size."],"forward_implications":["If the speed-up persists at larger sizes, initializing a CIM in a cat product state and ramping the coupling could reduce time-to-solution for max-cut and other Ising-type problems compared with vacuum-initialized machines.","The finding that a linear ramp of the coupling improves success rates suggests that the control schedule itself is a resource, analogous to annealing schedules in other quantum optimizers.","The purity measurements give an operationally measurable correlate: high success rates appear when purity is preserved, so reducing absorption in the coupling path should directly boost solver performance.","The MCWF simulations, scaling linearly in Hilbert-space dimension, provide a benchmark method for testing non-Gaussian quantum effects in CIMs of three to five modes, beyond what master equations can reach."],"supporting_citations":[{"why":"Supplies the Monte Carlo wave-function (quantum jump) method used for all simulations.","marker":"[25]"},{"why":"Gives the master equation and mean-field model of the coherently coupled CIM that the simulations solve.","marker":"[35]"},{"why":"Provides the exact DOPO steady state and the cat-state transient that motivates the non-classical initial states.","marker":"[27]"},{"why":"Shows why the Gaussian approximation fails for non-Gaussian superposition states, justifying the full wave-function approach.","marker":"[24]"},{"why":"Reports the first small-scale DOPO network experiment that defines the CIM architecture being simulated.","marker":"[17]"}],"fun_headline_variants":["Cat states give Ising machine a quantum speedup","Quantum advantage in Ising machine via cat states","Time-varying couplings boost Ising machine optimization","Monte Carlo runs show Ising machine quantum edge"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The claimed quantum advantage is measured against a vacuum initial state, but the cat state that wins also carries many more photons ($\\alpha = 2.582$ per mode); without a coherent-state baseline at the same amplitude, the speed-up cannot be pinned to quantum superposition.","fun_headline_variants_meta":{"raw":{"variants":["Cat states give Ising machine a quantum speedup","Quantum advantage in Ising machine via cat states","Time-varying couplings boost Ising machine optimization","Monte Carlo runs show Ising machine quantum edge"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000661,"raw_usage":{"total_tokens":3068,"prompt_tokens":1036,"completion_tokens":2032,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":652,"completion_tokens_details":{"reasoning_tokens":1984}},"tokens_in":652,"tokens_out":2032,"duration_ms":13637,"temperature":1.0,"reasoning_tokens":1984,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T22:07:32.671524+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate the same three-, four-, and five-mode problems with each mode initialized in a classical coherent state $\\lvert\\alpha\\rangle$ with $\\alpha = 2.582$ (the amplitude used for the cat states) and compare the time to maximum success probability with the paper's Fig. 3; if the coherent-state curves match the cat-state curves, the quantum-advantage claim collapses. A second check is to run a classical (mean-field or positive-P) CIM simulation with the same time-dependent couplings and see whether the same success-rate improvements appear without any non-classical initial state.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the Monte Carlo wave-function (quantum jump) method used for all simulations."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the master equation and mean-field model of the coherently coupled CIM that the simulations solve."},{"cited_title":"Drummond, K","cited_arxiv_id":null,"evidence_quote":"Provides the exact DOPO steady state and the cat-state transient that motivates the non-classical initial states."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Shows why the Gaussian approximation fails for non-Gaussian superposition states, justifying the full wave-function approach."},{"cited_title":"Yamamoto, K","cited_arxiv_id":null,"evidence_quote":"Reports the first small-scale DOPO network experiment that defines the CIM architecture being simulated."}],"review_version":1}