{"id":"44cf33b3-ffc1-4c89-b684-33f2b2a1c229","arxiv_id":"2501.13862","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A classical-reservoir-derived variational circuit achieves at least 0.99 ground-state fidelity for 8- and 10-site 2D Fermi-Hubbard models at U/t=8, using fewer parameters and no fermionic swap network.","lead":"This paper proposes a new way to prepare the ground state of the Fermi-Hubbard model on a quantum computer, by coupling the system to a classical reservoir and using the resulting time evolution as a variational circuit. On small 2D lattices, the approach reaches very high ground-state fidelities with fewer parameters than previous methods, including in the strongly correlated regime.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 'consistent ≥0.99 fidelity' claim rests on single optimization runs and one disorder realization, while Appendix B reports run-to-run energy variation up to 1e-2; without multi-seed statistics the consistency claim is not yet supported.","rationale":"The central claim is empirical: not that the ansatz is exact or universal, but that it consistently prepares high-fidelity ground states in the cases examined. The qualifier 'consistently' is load-bearing, and the reported evidence is one run per system/spin plus one disorder realization. The authors themselves report seed-dependent final-energy variation up to order 1e-2, which is the same scale as the 1e-2 infidelity threshold for the advertised 0.99 fidelity, particularly in near-degenerate spin sectors. Thus the data as presented underdetermine the consistency claim. This concern is concrete and testable, and it does not attack the theoretical motivation or the validity of the small-system numerics. If the multi-seed and multi-disorder test passes, the paper's conclusion is strengthened; if it fails, the claim should be weakened to 'in the runs we examined' and the disorder robustness statement should be removed or qualified. The data-availability reference [52] is not resolvable from the manuscript, which makes independent verification harder and reinforces the need for the proposed replication. I recommend keeping the reader's CONDITIONAL verdict: the method is promising and internally coherent, but the central consistency assertion needs additional statistical support before acceptance.","tokens_in":11053,"tokens_out":12358,"duration_ms":116217,"concrete_test":"Re-run the optimization in Eq. (7) for every system and spin sector reported (2×4 and 2×5 ladders at U/t=2 and 8; √8×√8 cluster at U/t=8; open and periodic boundary conditions) with at least 20 L-BFGS seeds from the same initialization distribution, and repeat the disordered calculation for at least 10 independent disorder realizations. Record the distribution of infidelities: if any seed or realization that falls under 'all cases examined' yields infidelity >1e-2, the consistency claim is falsified; if all runs reach ≥0.99 fidelity, the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central empirical claim—'consistently achieve ≥0.99 fidelity in all the cases we examined' (abstract and §IV)—is supported only by single successful optimization trajectories per system and spin sector. Figure 4 and the ladder plots show one curve per configuration, with no seed ensemble. Appendix B concedes that 'the variation in the final energy due to the initial guess is case-dependent... typically at most on the order of 10^{-2}.' A relative energy variation of 1e-2 is the same order as the error budget needed to guarantee infidelity below 1e-2, especially in the small-gap S=1 sector shown in Table I. The disorder robustness claim is based on 'one explicit representation' of a disordered Hamiltonian (Table I caption), i.e., a single sample. Without a multi-seed success-rate analysis and a disorder ensemble, the word 'consistently' is not established. This is not a mathematical contradiction; it is a missing statistical support for the central empirical claim. The Lie-algebra closure argument in §II covers only the free-fermion hopping part and does not certify the interacting ansatz, so the numerics are the only evidence for expressibility. Since the advertised consistency is the load-bearing part of the claim, this gap is significant.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a ground-state preparation method for the Fermi-Hubbard model based on coupling to a classical reservoir in the instantaneous-response limit. In this limit the evolution reduces to a layered Hamiltonian ansatz built from spin-adapted nearest-neighbor hopping terms along a Jordan-Wigner snake ordering plus on-site potential terms, without enlarging the Hilbert space. Starting from a product state with fixed total spin, the parameters are optimized by L-BFGS energy minimization. Numerical results on 8- and 10-site two-dimensional clusters at U/t=8 report infidelities in the range 1e-3 to 1e-6, and the method is compared with the number-preserving ansatz in terms of variational parameters needed to reach 0.99 fidelity. The central claim is that the method can consistently achieve at least 0.99 fidelity in all examined cases with lower resource counts than previous ansatze.","tokens_in":11349,"tokens_out":6827,"duration_ms":64912,"significance":"If the fidelity and resource claims hold, this is a useful and practical simplification of the Hamiltonian variational ansatz: it preserves total spin, avoids fermionic swap networks, uses only M=2N-1 parameters per layer, and reproduces exact ground states on the tested strongly correlated clusters. The authors provide open data and code, which is a clear strength. However, the headline 'consistently achieve >=0.99 fidelity' claim currently rests on single optimization runs and a single disorder realization, while the paper itself reports run-to-run energy variation of order 1e-2. The comparison with the NP ansatz is also made at U/t=2, not at the strong-coupling value U/t=8 that motivates the work. The core idea is sound and the numerical demonstrations are consistent with exact ground states, but the statistical support for the central claim needs strengthening.","major_comments":[{"comment":"The claim that the method can 'consistently achieve >=0.99 fidelity in all the cases we examined' is not supported by the data as presented. Figures 4 and 5 show a single optimization trajectory per spin sector and system, with no ensemble over initial parameters, and Appendix B states that the variation in the final energy due to the initial guess is 'typically at most on the order of 10^{-2}'. That variation is comparable to the error budget needed to guarantee an infidelity below 1e-2, particularly in the small-gap S=1 sector of Table I. Please report the success rate and the distribution of fidelities over many random initializations (and, where relevant, over disorder samples) so that the consistency claim can be evaluated quantitatively.","section":"Abstract and Section IV"},{"comment":"The disorder-robustness conclusion is drawn from 'one explicit representation' of a disordered Hubbard Hamiltonian. A single random realization cannot establish robustness against disorder. The authors should provide ensemble-averaged infidelities or relative energy errors over multiple disorder samples, including the spread across samples, before claiming that disorder does not significantly affect performance.","section":"Table I and Section III.A"},{"comment":"The resource comparison with the number-preserving ansatz is performed only at U/t=2, whereas the motivation and conclusions emphasize the strongly correlated regime U/t=8. At U/t=2 the system is weakly correlated, so Table II does not substantiate the claim of improved performance in the strong-correlation regime where prior ansatze are said to plateau. Please provide a comparison at U/t=8 or explicitly limit the comparison claim to the parameter regime actually studied.","section":"Table II and Section III.B"}],"minor_comments":[{"comment":"The sentence 'The resulting time evolution operator consist of...' should read 'consists of'.","section":"Abstract"},{"comment":"The phrase 'The noininteracting state' appears to be a typo for 'noninteracting state'; please correct it.","section":"Section IV"},{"comment":"The text uses 'Langrangian-based dynamics' and 'Langrangian' in the figure caption; these should be 'Lagrangian'.","section":"Section II and Figure 1"},{"comment":"The column header 'Fermionic swap network' would be clearer as 'Uses fermionic swap network?' and the NP parameter counts should be accompanied by the number of layers used to reach the 0.99 fidelity threshold, so the comparison is reproducible.","section":"Table II"},{"comment":"The notation in Eq. (2) involving a time-ordered exponential of beta-weighted operators with a two-time action S(lambda) is standard in path-integral contexts but is unusual here; a brief definition of the trace and the ordering would improve readability.","section":"Section II, Eq. (2)"}],"recommendation":"major_revision","confidential_remarks":"To the editor: the manuscript is within scope for a quantum simulation journal and the proposed method is a reasonable variant of the Hamiltonian variational ansatz. The main issue is that the headline consistency claim is not yet backed by multi-seed and multi-disorder statistics, and the strong-coupling resource comparison is missing. These are fixable within the manuscript's existing framework, so I do not see grounds for rejection. A major revision addressing the statistical support and the U/t=8 comparison would make the paper's claims solid."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a legitimate ansatz paper with a real resource improvement, but the headline claim that it 'consistently achieves ≥0.99 fidelity' is not yet backed by the statistics. The numerical demonstrations on 8- and 10-site clusters are clean, the resource comparison against the NP ansatz is concrete, and the authors are honest about the limitations of their theoretical arguments.\n\nWhat is actually new: the classical-reservoir framing provides a fresh perspective on the HVA, and the concrete circuit construction is a genuine improvement over previous work. Spin-adapted hopping restricted to Jordan-Wigner-nearest-neighbor pairs removes the need for fSWAP networks, product-state initialization with target total spin avoids the U=0 ground state preparation, and each layer has only 2N-1 parameters. The Lie-algebra closure argument for the hopping part is a nice theoretical justification for the JW-path restriction, and the authors correctly note that this argument only covers the Gaussian part, not the full interacting expressibility.\n\nThe numerical results are consistent with the central fidelity claim for the specific cases shown: infidelities reach 1e-3 to 1e-6 on the √8×√8 cluster and 2×4/2×5 ladders at U/t=8, and the parameter count is substantially lower than the NP ansatz. The data and code are promised openly, which I take as a positive sign for reproducibility.\n\nThe soft spots are real but addressable. The main one is the word 'consistently.' The paper shows one optimization curve per spin sector per system, and the disorder robustness rests on a single explicit disorder realization. Appendix B concedes that the run-to-run variation in final energy is 'typically at most on the order of 10^{-2}'—which is exactly the same order as the error budget required to guarantee infidelity below 1e-2, especially in the small-gap S=1 sector. Without multi-seed initialization statistics and a disorder ensemble, the consistency claim is not established. This is missing statistical support, not a mathematical contradiction, but it is load-bearing because the abstract emphasizes consistency. The scaling behavior is also only conjectural: two small system sizes, no scaling analysis, and the interacting expressibility is entirely empirical.\n\nWho this is for: researchers working on variational quantum simulation of strongly correlated fermions, near-term algorithms, and Hubbard model resource counts. It deserves a serious referee: the core idea is well-motivated, the resource reduction is concrete, and the numerics are relevant. My recommendation is to engage with it and send it to peer review, but to require multi-seed statistics and a disorder ensemble before the 'consistently' claim is accepted.","headline":"A legitimate ansatz paper with a real resource reduction, but the 'consistently ≥0.99 fidelity' claim needs multi-seed statistics before it can be taken at face value.","tokens_in":11872,"tokens_out":2949,"would_cite":true,"duration_ms":24766,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A classical-reservoir-inspired variational circuit prepares Fermi-Hubbard ground states with at least 0.99 fidelity on all tested two-dimensional clusters, using fewer parameters than previous ansatze.","keywords":["Fermi-Hubbard model","ground-state preparation","classical reservoir","Hamiltonian variational ansatz","Jordan-Wigner transformation","total spin conservation","strong correlations","variational quantum circuit"],"falsifier":"Compute the exact ground state of a larger two-dimensional Hubbard cluster, such as a $4\\times4$ or 20-site lattice at $U/t=8$ in the $S=0$ sector, run the same ansatz with the paper's optimization procedure, and check whether the optimized fidelity reaches 0.99; if the infidelity plateaus above 0.01 for any such cluster, the claim that the method consistently achieves 0.99 fidelity would be refuted.","tokens_in":10864,"feed_emoji":"⚛️","tokens_out":17901,"duration_ms":132187,"temperature":0.7,"pith_summary":"The Fermi-Hubbard model (electrons hopping on a lattice with on-site repulsion) is a standard testbed for quantum ground-state preparation, but its exponentially large Hilbert space makes variational paths hard to find. This paper claims that a variational circuit built from spin-adapted (total-spin-conserving) nearest-neighbor hopping and on-site interaction steps can drive a simple product state to the interacting ground state with at least 0.99 fidelity, including at strong coupling $U/t=8$. The circuit is motivated as a classical reservoir coupled to the system in the instantaneous-response limit, which turns the cooling dynamics into Hamiltonian evolution that never expands the Hilbert space. On an 8-site periodic cluster and on $2\\times4$ and $2\\times5$ ladders, the optimized ansatz reaches infidelities between $10^{-3}$ and $10^{-6}$, and on open-boundary ladders it needs roughly half the parameters of the number-preserving ansatz (a prior particle-conserving variational ansatz). If the results extend beyond these small clusters, the method offers a resource-efficient, spin-conserving route to strongly correlated ground states on near-term quantum hardware, with no fermionic swap network and no Z-string operators in state preparation.","feed_headline":"Reservoir circuit prepares Hubbard ground states at 99% fidelity","feed_subtitle":"Spin-symmetric hopping circuits reach the same ground states with far fewer parameters than previous ansatze.","key_machinery":"The load-bearing mechanism is a layered ansatz $|\\psi_f\\rangle = \\prod_{l=1}^L e^{i\\hat U_l(\\lambda')} e^{-i\\hat T_l(\\lambda)} e^{i\\hat T'_l(\\lambda)} |\\psi_0\\rangle$, built from spin-adapted nearest-neighbor hopping terms $\\hat T = \\sum_{\\langle i,j\\rangle,\\sigma} \\lambda (\\hat c^\\dagger_{i\\sigma}\\hat c_{j\\sigma} + \\mathrm{h.c.})$ and on-site potential terms $\\hat U = \\sum_i \\lambda' \\hat n_{i\\uparrow}\\hat n_{i\\downarrow}$; each layer carries $M=2N-1$ parameters. The argument has three supporting parts: pairing the spin-up and spin-down hoppings conserves total spin and its $z$-component, so optimization stays in one spin sector; the Lie-algebra closure $[\\hat c^\\dagger_p \\hat c_q, \\hat c^\\dagger_a \\hat c_b] = \\hat c^\\dagger_p \\delta_{qa}\\hat c_b - \\hat c^\\dagger_a \\delta_{bp}\\hat c_q$ lets adjacent hoppings along the one-dimensional Jordan-Wigner path generate all effective hoppings, so no fermionic swap network and no Z-string operators are needed in state preparation; and the instantaneous-response limit of a classical reservoir justifies Hamiltonian time evolution in the canonical ensemble, keeping the state pure and the Hilbert-space dimension unchanged.","core_discovery":"The paper's central claim is that a layered circuit of spin-adapted hopping and on-site potential terms, with parameters found by minimizing the final-state energy, can prepare the Hubbard ground state with high fidelity without leaving the original Hilbert space. The construction starts from a classical reservoir action and takes the instantaneous-response limit, which removes the two-time dependence and allows a canonical Hamiltonian description; the state remains pure and total particle number and spin are conserved. In the numerical tests on the $\\sqrt{8}\\times\\sqrt{8}$ periodic cluster at $U/t=8$ (total spin sectors $S=0,1,2,3$, pure and disordered) and on $2\\times4$ and $2\\times5$ ladders at $U/t=2$ and $U/t=8$, the method consistently reaches fidelities of at least 0.99, with infidelities spanning $10^{-3}$ to $10^{-6}$. Against the number-preserving ansatz on open-boundary ladders, it achieves the same fidelity with 180 and 285 parameters instead of 392 and 432, and with one layer of $X$ gates for initialization instead of $N-1$ layers.","pith_inferences":["Editorial inference: if the 0.99-fidelity performance persists on 16- and 20-site clusters, this ansatz becomes a leading candidate for near-term Hubbard-model simulation, but that scaling is not established in the paper.","Editorial inference: the same layer structure may transfer to other lattice fermion models by replacing the on-site potential with their interaction terms; the spin-adapted hopping pairing and the one-dimensional hopping closure would remain valid only for interactions that preserve total spin.","Editorial inference: applying the ansatz away from half-filling, in three dimensions, or with a different optimizer would test the paper's conjecture that preparation difficulty is essentially independent of interaction strength; those regimes are not covered by the reported numerical evidence."],"forward_implications":["Ground states of strongly correlated Hubbard clusters can be prepared with at least 0.99 fidelity using circuits whose per-layer cost is linear ($M=2N-1$ parameters), with no fermionic swap network and no Z-string operators in the state-preparation circuit.","Initialization reduces to a single layer of $X$ gates on a simple product state, so the method avoids the more involved preparation of the noninteracting $U=0$ ground state used by other variational algorithms.","On the open $2\\times4$ and $2\\times5$ ladders at $U/t=2$, the same fidelity is reached with 180 and 285 parameters respectively, versus 392 and 432 for the number-preserving ansatz, and with $N-1$ initialization layers reduced to one.","Restricting to one total-spin sector allows the ansatz to target the ground state of each spin sector separately, and the numerical results show stability against random hopping and on-site disorder."],"supporting_citations":[{"why":"Provides the number-preserving ansatz used as the benchmark; its parameter counts (392 and 432 for 2x4 and 2x5 open ladders) are the resource targets this work improves on.","marker":"[34]"},{"why":"Documents the plateauing energy error and sub-0.99 fidelity (~0.7 in the 3x3 case) in prior ansatze that this work aims to fix.","marker":"[36]"},{"why":"Introduces the Hamiltonian variational ansatz whose layered structure the classical reservoir ansatz closely resembles.","marker":"[33]"},{"why":"Shows that pairing spin-up and spin-down hopping terms conserves total spin, the symmetry constraint at the heart of the ansatz.","marker":"[38]"},{"why":"Jordan-Wigner transformation used to encode fermionic operators on qubits; adjacent hoppings along the string need no Z-string operators.","marker":"[39]"},{"why":"Free-fermion compression results supporting the claim that hopping along the one-dimensional Jordan-Wigner path generates all effective hoppings.","marker":"[40, 41]"},{"why":"Describes the fermionic swap network approach that this work avoids, and supplies the N-layer bound on initial-state preparation used in the comparison.","marker":"[48]"},{"why":"SU(2) identity used to implement the exponential of each hopping term efficiently in the classical numerical simulation.","marker":"[56]"}],"fun_headline_variants":["Reservoir-inspired circuit hits 99% Hubbard ground state fidelity","Classical reservoir trick prepares Hubbard ground states with fewer parameters","Instantaneous reservoir gives Hamiltonian path to Hubbard ground state","Spin-adapted reservoir ansatz reaches 99% fidelity with fewer parameters"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The claim rests on the ansatz being expressive enough to reach the exact ground state with at least 99 percent fidelity on the studied clusters, and on the numerical optimizer reliably finding that minimum; this is demonstrated for 8- and 10-site systems but not proven for larger sizes or for the interacting, non-Gaussian part of the ansatz.","fun_headline_variants_meta":{"raw":{"variants":["Reservoir-inspired circuit hits 99% Hubbard ground state fidelity","Classical reservoir trick prepares Hubbard ground states with fewer parameters","Instantaneous reservoir gives Hamiltonian path to Hubbard ground state","Spin-adapted reservoir ansatz reaches 99% fidelity with fewer parameters"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000595,"raw_usage":{"total_tokens":2775,"prompt_tokens":925,"completion_tokens":1850,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":541,"completion_tokens_details":{"reasoning_tokens":1779}},"tokens_in":541,"tokens_out":1850,"duration_ms":12597,"temperature":1.0,"reasoning_tokens":1779,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T15:31:43.814614+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the exact ground state of a larger two-dimensional Hubbard cluster, such as a $4\\times4$ or 20-site lattice at $U/t=8$ in the $S=0$ sector, run the same ansatz with the paper's optimization procedure, and check whether the optimized fidelity reaches 0.99; if the infidelity plateaus above 0.01 for any such cluster, the claim that the method consistently achieves 0.99 fidelity would be refuted.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the number-preserving ansatz used as the benchmark; its parameter counts (392 and 432 for 2x4 and 2x5 open ladders) are the resource targets this work improves on."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Documents the plateauing energy error and sub-0.99 fidelity (~0.7 in the 3x3 case) in prior ansatze that this work aims to fix."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces the Hamiltonian variational ansatz whose layered structure the classical reservoir ansatz closely resembles."},{"cited_title":"Reiner, F","cited_arxiv_id":null,"evidence_quote":"Shows that pairing spin-up and spin-down hopping terms conserves total spin, the symmetry constraint at the heart of the ansatz."},{"cited_title":"Jiang, K","cited_arxiv_id":null,"evidence_quote":"SU(2) identity used to implement the exponential of each hopping term efficiently in the classical numerical simulation."}],"review_version":1}