{"id":"a592383e-9b64-479a-b26f-a3f4faf3500c","arxiv_id":"2505.08994","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Quantum annealing on fullerene Ising cages tracks a perturbatively selected quantum superposition, and the newer D-Wave processor achieves higher fidelity to that state than the older one.","lead":"The authors ran quantum annealing experiments on D-Wave processors using frustrated Ising spin models shaped like fullerenes, showing that quantum fluctuations split the degeneracy of the classical ground states. The fullerenes are proposed as a new benchmark that separates the simulation quality of older and newer quantum annealer hardware.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The QPU time axis is fixed by assuming negligible decoherence at t_a=5 ns and calibrating F' at that point; if 5 ns is not coherent, the quantitative QPU/ED agreement in Figs. 3-4 is partly an artifact.","rationale":"The paper is careful and mostly honest: it reports MPS convergence limits, shares the BP-TNS failure, and states the calibration assumption explicitly. The central theoretical construction (perturbative tunneling matrix with -1 entries for floppy-dimer flips) is standard degenerate perturbation theory and is parameter-free. The qualitative separation between ADV1 and ADV2 is plausible and robust to calibration details because both devices are mapped through the same procedure and the newer device has a higher energy scale and lower noise. The load-bearing weakness is that the quantitative comparison to |ψ_epsilon> relies on an untested equality between the 5 ns QPU state and the ideal closed-system state. The calibration is performed on the same observable (F') that is later used to claim agreement, and the J-to-time rescaling is only approximately equivalent to time rescaling. A direct simulation at |J|=0.95 would settle this; the paper did not report such a run. This does not invalidate the paper but supports keeping the CONDITIONAL verdict: the devices appear to produce dynamics close to the ideal TFIM in some regime, but the exact time scale and the quantitative fidelity to |ψ_epsilon> still need an independent check.","tokens_in":10296,"tokens_out":7775,"duration_ms":75884,"concrete_test":"Run ED/MPS simulations directly at the QPU conditions |J_ij|=0.95, physical t_a=5 ns, using the actual ADV1 and ADV2 schedules, with no time rescaling, and compare the resulting F', D, and δE to the QPU outputs at those same settings. If the directly simulated values disagree with the QPU values by more than the sampling error, the negligible-decoherence assumption at 5 ns fails. Alternatively, recalibrate the effective time using δE instead of F' at the same 5 ns anchor; if the inferred equivalent time changes by more than ~20%, the calibration is observable-dependent and cannot be interpreted as establishing a physical time axis.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Supplementary 'Calibrating effective time' states: 'we assume that for t_a = 5 ns the QPU dynamics closely reflect the ideal dynamics of a closed quantum system with negligible decoherence.' Under that assumption, all QPU runs are mapped onto the |J|=1 ED/MPS time axis by matching the binned fidelity F' at t_a=5 ns for N=20; the ADV2 time axis is likewise fixed by matching MPS F' at t_a=3 ns. This creates an exact overlap at the anchor point by construction. The apparent tracking of QPU and classical curves elsewhere (e.g., the nonmonotonic D in Fig. 4a and the F' minimum in Fig. 4b) is not an independent test of the closed-system hypothesis, because the time coordinate of every QPU point is set by the same F'-matched rescaling, and runs at t_a>5 ns with |J_ij|=0.95 inherit the J-to-time rescaling calibrated at 5 ns. The paper's own MPS convergence data show the largest QPU/classical deviations near t_a≈1 ns, a region inaccessible to direct QPU measurements (physical t_a≥5 ns); the claimed coherent-regime agreement therefore rests on an extrapolation of the calibration. If decoherence is already significant at 5 ns, the calibrated effective time is biased, and the quantitative match of D and F' to |ψ_epsilon> is weakened, even though the qualitative ADV2>ADV1 ordering would likely survive. The N=60 data provide some out-of-sample support (the calibration was fixed on N=20), but the time mapping is the same and no independent consistency check is reported.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes frustrated Ising fullerenes (the 20-spin dodecahedron and two 60-spin buckyball variants) as boundary-free benchmark systems for quantum simulators. For these systems, the authors derive a perturbative target state |ψ_ε> from degenerate perturbation theory: it is the principal eigenvector of the tunneling matrix among classical ground states, with 'floppy dimers' as the tunneling objects. They report quantum annealing experiments on two D-Wave processors (ADV1 and ADV2), comparing residual energy density δE, average floppy-dimer density D, and a binned fidelity F' against exact diagonalization for N=20 and matrix-product-state simulations for N=60. The central claim is that the QPU dynamics reproduce the quantum-fluctuation-selected state |ψ_ε> in the coherent regime, and that the newer ADV2 processor achieves higher fidelity to this state than the older ADV1 processor.","tokens_in":10648,"tokens_out":4838,"duration_ms":45926,"significance":"If the result holds, this is a valuable benchmark proposal: the target state |ψ_ε> is derived from first principles with no fitted physics parameters, and the N=20 exact-diagonalization comparison is rigorous. The paper also includes careful MPS convergence checks, reports reproducible simulator settings, and connects the observed order-by-disorder phenomenon to an experimentally testable observable. The qualitative ordering ADV2 > ADV1 in fidelity is a useful, hardware-relevant finding. However, the quantitative agreement between QPU and classical curves is weakened by the effective-time calibration, as detailed in the major comments.","major_comments":[{"comment":"The effective-time calibration is the load-bearing element of the quantitative comparison. The assumption that t_a=5 ns QPU dynamics are those of a closed quantum system with negligible decoherence, followed by matching F' at the anchor, means that the agreement between QPU and classical curves in Figs. 3 and 4 is enforced at one point and only tests the shape of the curves through the global time rescaling. If decoherence is already significant at 5 ns, all calibrated times are biased and the quantitative values of D and F' reported for the 'coherent regime' are not meaningful. Please provide a robustness analysis: for example, vary the anchor time if the hardware allows, or validate the mapping using an observable not involved in the calibration (e.g., δE or D), and state explicitly what deviations would falsify the closed-system assumption.","section":"Supplementary, 'Calibrating effective time'"},{"comment":"For N=60, the MPS benchmark is not converged in the region t_a≈1 ns on the calibrated axis, where bond-dimension deviations are largest and where the QPU data are mapped. The main-text claim that QPU data 'track' the classical curves in the coherent regime for N=60 is therefore supported by unconverged reference data. The quantitative N=60 comparison should be either converged further (e.g., higher χ in the short-t_a region) or downgraded to a qualitative statement, with the rigorous quantitative comparison restricted to N=20 where ED is exact.","section":"Supplementary, Fig. S2 and 'Classical simulation methods' (MPS)"},{"comment":"The binned fidelity F' is used both as the calibration observable and as the central quality metric. This creates a partial circularity: the time rescaling is chosen to make F' match at the anchor, and the subsequent agreement of the 1-F' minimum is not fully independent. Please show the raw, uncalibrated QPU data (physical t_a and |J_ij| values) alongside the rescaled curves, or calibrate on one observable (e.g., δE) and test F' out-of-sample, so that the reader can assess which features of the data are robust.","section":"Results, Fig. 4"}],"minor_comments":[{"comment":"The spelled name 'Battachyyara' is a typo; it should be 'Bhattacharyya'.","section":"Fig. 4b caption"},{"comment":"The operator D is presented without derivation; a short explanation of how the projectors enforce the floppy-dimer condition would improve readability.","section":"Eq. (6)"},{"comment":"The relation between the 'resonating dimers' of the abstract and the perturbative tunneling matrix is stated only informally; a few sentences connecting the dimer flip to the −1 off-diagonal matrix element would help.","section":"Introduction and 'Frustrated Ising fullerenes'"}],"recommendation":"major_revision","confidential_remarks":"The manuscript relies heavily on the authors' own Ref. 30 for the MPS methodology and time-calibration approach; this is appropriate but means the calibration assumption is not independently benchmarked within this manuscript. The paper fits the journal's scope as a quantum-simulation benchmark study. The central qualitative finding (order-by-disorder effect, ADV2 outperforming ADV1) is likely robust, but the quantitative claim needs the robustness analysis described in the major comments."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things up front. The paper introduces genuinely useful benchmark problems—boundary-free frustrated Ising fullerenes—and a discriminating observable, floppy-dimer density D, plus a binned fidelity F′ that exploits the lattices' automorphism structure. The two-generation D-Wave comparison is clean and the newer processor clearly wins. The second thing: the main caveat is not the physics, it's the time calibration. The QPU time axis is set by assuming negligible decoherence at t_a = 5 ns and matching F′ at that point, so the agreement between QPU and ED/MPS curves is partly anchored by construction. The reader's conditional verdict is the right one.\n\nWhat's genuinely new: the specific lattices (AFM dodecahedron, buckyball, and fully frustrated mixed FM/AFM buckyball), the perturbative target state |ψ_ε⟩ from degenerate perturbation theory with no fitted physics parameters, and the floppy-dimer density observable. The N=20 exact diagonalization is exact; the MPS bond-dimension convergence for N=60 is honestly reported, including the failure near t_a ≈ 1 ns. The BP-TNS failure is disclosed rather than hidden. The N=24 supplement gives an out-of-sample consistency check. Credit where due: the qualitative structure—nonmonotonic D in the coherent regime, D approaching |ψ_0⟩ for long anneals, ADV2 outperforming ADV1 in fidelity—is likely robust.\n\nSoft spots in proportion. The effective-time calibration is load-bearing. The supplementary section 'Calibrating effective time' says 'we assume that for t_a = 5 ns the QPU dynamics closely reflect the ideal dynamics of a closed quantum system with negligible decoherence.' All QPU runs are mapped onto the classical time axis by matching F′ at 5 ns for N=20. That makes the overlap at the anchor exact by construction, and longer-time or reduced-|J| runs inherit the same rescaling. If decoherence is non-negligible at 5 ns, the quantitative match to |ψ_ε⟩ is stronger than the data actually demonstrate. The N=60 data provide some out-of-sample support (the calibration was fixed on N=20), but no independent consistency check is reported. The main figures also omit explicit error bars, though the paper states the sampling floor is below axis limits. These are serious but not disqualifying: the qualitative claims don't depend on the precise time axis.\n\nThe citation pattern is fine. Refs. 25–28 are appropriate for unfair ground-state sampling; Ref. 30 is the same group's earlier method paper, which is reasonable since the MPS methods are built on it. No sign of hiding prior work.\n\nWho gets value: anyone benchmarking quantum annealers or analogue simulators, and the quantum simulation community more broadly. This deserves a serious referee. I'd ask for a sensitivity analysis of the calibration assumption and error bars in the main figures, but I would not desk-reject.","headline":"Worth engaging: genuinely useful frustrated-fullerene benchmarks and a clear newer-vs-older annealer comparison, with the effective-time calibration as the main soft spot.","tokens_in":11201,"tokens_out":3010,"would_cite":false,"duration_ms":31021,"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":"This paper reports the experimental observation that quantum fluctuations lift the classical ground-state degeneracy of Ising models on fullerene graphs, producing a superposition state that two generations of superconducting quantum…","keywords":["frustrated Ising fullerenes","quantum annealing","order by disorder","floppy dimers","ground-state degeneracy","quantum simulation benchmark","buckyball"],"falsifier":"Check the calibration anchor: simulate the ideal closed-system dynamics (exact diagonalization for N=20) at t_a = 5 ns with |J|=1 and compare directly, without any rescaling, to the QPU output at 5 ns under the actual annealing schedule. If the binned fidelity or dimer density differ by more than the statistical sampling error, the assumption of negligible decoherence at 5 ns fails. A second check: recalibrate by matching the fidelity at a different anchor time (e.g., 10 ns) and test whether the qualitative ADV1-versus-ADV2 separation in the residual energy and dimer density curves survives.","tokens_in":10098,"feed_emoji":"⚛️","tokens_out":11534,"duration_ms":103057,"temperature":0.7,"pith_summary":"This paper attempts to show that quantum fluctuations, not thermal effects, select a specific superposition state out of the heavily degenerate classical ground states of antiferromagnetic Ising models defined on fullerene graphs — a 20-spin dodecahedron and 60-spin buckyballs with two different coupling patterns. Because any two classical ground states differ by at least one floppy-dimer flip, a small transverse field generates an effective tunneling matrix on the ground-state manifold, and the paper identifies the target state |ψε⟩ as that matrix's principal eigenvector. The paper then reports that two generations of superconducting quantum annealers reproduce the predicted out-of-equilibrium dynamics in the coherent regime: they match exact-diagonalization and matrix-product-state simulations for short anneal times, show a characteristic infidelity minimum, and only later relax toward the uniform superposition over classical ground states. The clearer separation between the two processors in the fidelity data suggests that these boundary-free frustrated lattices can serve as a practical, shot-efficient benchmark for quantum-simulation hardware.","feed_headline":"Quantum fluctuations lift the frozen spin degeneracy of Ising fullerenes","feed_subtitle":"Floppy-dimer density and binned fidelity separate two annealer generations in the coherent regime.","key_machinery":"The load-bearing mechanism is the floppy dimer: a pair of spins joined by a bond whose simultaneous flip connects two classical ground states at Hamming distance 2. Under a small transverse field each floppy dimer acts as a second-order tunneling resonance, contributing an off-diagonal −1 to the effective tunneling matrix on the ground-state manifold; the perturbative ground state |ψε⟩ is the principal eigenvector of that matrix, which plays the role of a dimer-flip adjacency graph on the degenerate space. The key experimental observable is the binned fidelity F′, the Bhattacharyya coefficient between sampled output probabilities and the |ψε⟩ probabilities averaged over automorphism orbits of the fullerene; binning by symmetry suppresses the large sampling error that would otherwise drown out the signal, and together with the floppy-dimer density operator D it turns a highly degenerate manifold into a sensitive and shot-efficient benchmark of a simulator's adherence to the coherent quantum target state.","core_discovery":"On a three-regular fullerene graph with ±1 couplings, the classical Ising model has 250, 16,000, or 1,125,000 degenerate ground states depending on the coupling pattern. In the perturbative limit Γ=ε→0, the paper claims that the ground state of εH_D + H_I is determined entirely by the tunneling matrix whose rows and columns index the classical ground states: the entry is −1 for pairs of states that differ by a floppy-dimer flip (the simultaneous flip of two coupled spins) and 0 otherwise, and |ψε⟩ is the principal eigenvector of this matrix. The paper simulates the nonequilibrium approach toward |ψε⟩ with exact diagonalization and MPS time evolution and on two generations of superconducting quantum annealers, tracking residual energy density, the average floppy-dimer density D, and a binned fidelity F′. In the coherent short-time regime the QPU measurements reproduce the nonmonotonic D curve and the infidelity minimum of the ideal closed-system dynamics; at long anneal times they thermalize toward the uniform-superposition statistics |ψ0⟩ rather than |ψε⟩, and the lower-noise second-generation processor (ADV2) reaches lower residual energy, later thermalization onset, and higher peak fidelity than its predecessor (ADV1). The authors take this as experimental demonstration that quantum fluctuations lift the classical ground-state degeneracy in these fullerenes, with the dynamics precisely enough resolved to distinguish simulator quality.","pith_inferences":["The tunneling matrix on the ground-state manifold is the adjacency matrix of a graph whose vertices are classical ground states; for fullerenes with larger degeneracy this graph likely fragments, and the principal-eigenvector selection would then single out one connected component, making the benchmark's dynamics sensitive to the fullerenes' dimer-cover topology — a sharper discriminator than tota","A cross-platform test could use the same calibration trick on neutral-atom or trapped-ion simulators: any platform whose output at short times matches exact diagonalization should collapse onto the same universal curves of D versus effective time, giving a direct apples-to-apples comparison of simulation quality across hardware types.","The calibration is anchored at a single point (t_a=5 ns) using only the N=20 dodecahedron; extending the same fidelity-matching to N=60 or to the mixed-coupler fullerene would test whether the effective-time mapping is a genuine dynamical equivalence, and any failure would point to schedule-dependent decoherence models rather than a single timescale.","Because binned fidelity uses automorphism orbits, the same symmetry reduction could be applied to other observables or to noisy intermediate-scale processors with fewer shots, potentially making such benchmarks accessible to hardware with low repetition rates."],"forward_implications":["Frustrated fullerenes provide a systematic benchmark family: the degeneracy N_GS can be tuned from 250 to over 10^6 by choosing the dodecahedron, the all-antiferromagnetic buckyball, or the mixed ferromagnet/antiferromagnet buckyball, and the same graphs embed directly in current annealer topologies.","A simulator's quality is read off from the plateau value of the floppy-dimer density D and the minimum of the binned infidelity 1−F′: coherent platforms settle near |ψε⟩ (D≈5.45 for the dodecahedron) while thermalizing platforms drift toward |ψ0⟩ (D≈4.8), so D itself is a single-number coherence proxy.","The benchmark distinguishes hardware generations without needing full state tomography: ADV2's later thermalization onset and higher peak fidelity relative to ADV1 demonstrate that the observables are sensitive to noise levels.","Because the target state |ψε⟩ differs from the uniform classical superposition |ψ0⟩, the same experiment separates coherent quantum evolution from classical sampling: a thermal sampler would produce |ψ0⟩ statistics, not |ψε⟩, at long times."],"supporting_citations":[{"why":"Provides the quantum dimer model on fullerenes that motivates the resonating floppy-dimer physics central to the paper's tunneling picture.","marker":"[24]"},{"why":"Earlier comparison of the perturbative tunneling ground state with the uniform superposition of classical ground states, which the present work extends to fullerenes.","marker":"[29]"},{"why":"Supplies the matrix-product-state time-evolution methods, the ADV1/ADV2 processor labels, and the effective-time calibration approach used throughout.","marker":"[30]"},{"why":"Establishes the coherent-annealing regime and decoherence expectations in programmable Ising chains that the paper relies on for interpreting short-anneal QPU data.","marker":"[31]"},{"why":"Provides the subgraph-embedding solver used to map the fullerene graphs onto the QPU connectivity graphs for the hardware experiments.","marker":"[35]"},{"why":"Supplies the shimming and calibration refinement procedure that balances qubit offsets and couplers in the QPU runs.","marker":"[36]"},{"why":"Cited to argue that neutral-atom platforms could assemble the same three-dimensional fullerene graphs, supporting the cross-platform benchmark claim.","marker":"[21]"}],"fun_headline_variants":["Quantum tunneling lifts spin degeneracy in Ising fullerenes","Frustrated fullerenes benchmark quantum annealer generations","Ising fullerene dynamics separate two annealer generations","Quantum fluctuations break classical spin ground-state ties","Fullerene qubit test sharpens across annealer versions"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole comparison of QPU data against classical simulations stands on the claim that at the shortest anneal time t_a = 5 ns the QPU output is a faithful closed-system evolution with negligible decoherence; if already decohered at 5 ns, the time rescaling that produces the apparent agreement would be anchoring the data to the wrong curve.","fun_headline_variants_meta":{"raw":{"variants":["Quantum tunneling lifts spin degeneracy in Ising fullerenes","Frustrated fullerenes benchmark quantum annealer generations","Ising fullerene dynamics separate two annealer generations","Quantum fluctuations break classical spin ground-state ties","Fullerene qubit test sharpens across annealer versions"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000168,"raw_usage":{"total_tokens":1255,"prompt_tokens":937,"completion_tokens":318,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":553,"completion_tokens_details":{"reasoning_tokens":239}},"tokens_in":553,"tokens_out":318,"duration_ms":3591,"temperature":1.0,"reasoning_tokens":239,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T21:44:09.532421+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Check the calibration anchor: simulate the ideal closed-system dynamics (exact diagonalization for N=20) at t_a = 5 ns with |J|=1 and compare directly, without any rescaling, to the QPU output at 5 ns under the actual annealing schedule. If the binned fidelity or dimer density differ by more than the statistical sampling error, the assumption of negligible decoherence at 5 ns fails. A second check: recalibrate by matching the fidelity at a different anchor time (e.g., 10 ns) and test whether the qualitative ADV1-versus-ADV2 separation in the residual energy and dimer density curves survives.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the quantum dimer model on fullerenes that motivates the resonating floppy-dimer physics central to the paper's tunneling picture."},{"cited_title":"Matsuda, H","cited_arxiv_id":null,"evidence_quote":"Earlier comparison of the perturbative tunneling ground state with the uniform superposition of classical ground states, which the present work extends to fullerenes."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes the coherent-annealing regime and decoherence expectations in programmable Ising chains that the paper relies on for interpreting short-anneal QPU data."},{"cited_title":"McCreesh, P","cited_arxiv_id":null,"evidence_quote":"Provides the subgraph-embedding solver used to map the fullerene graphs onto the QPU connectivity graphs for the hardware experiments."},{"cited_title":"Barredo, V","cited_arxiv_id":null,"evidence_quote":"Cited to argue that neutral-atom platforms could assemble the same three-dimensional fullerene graphs, supporting the cross-platform benchmark claim."}],"review_version":1}