{"id":"c4f628ec-1b1b-4724-81d6-3083b04c84f8","arxiv_id":"2501.13371","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"LAS-nuVQE, which applies a non-unitary Jastrow operator on top of a LASSCF initial state plus a hardware-efficient ansatz, reaches chemical accuracy in noiseless simulations of H4 and square cyclobutadiene with under 70 gates.","lead":"This paper combines a non-unitary Jastrow-style correction with a localized active space initial state for the variational quantum eigensolver, testing it on 8-qubit molecular models. It reports chemical accuracy with fewer than 70 circuit gates and shows how Pauli grouping and shot-frugal sampling cut measurement costs.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Resource claim counts only HEA gates; direct initialization of the LASSCF state is omitted from the reported '<70 total gates' totals, so the hardware-feasibility conclusion is not yet supported.","rationale":"The numerical energy results in the paper are likely correct for the noiseless simulations reported, and the method is a reasonable combination of existing ideas. The load-bearing weakness is resource accounting: the paper explicitly says DI is used for state preparation, yet every quoted gate total counts only HEA layers. Because the central claim is about affordability on near-term hardware, omitting the state preparation circuit makes the '<70 total gates' statement unsupported. The reader's weakest_assumption identifies exactly this issue, and the condition they place on acceptance is appropriate. A secondary concern is that the spin-constrained results in Table 2 do not appear to reach chemical accuracy, so the chemical-accuracy claim for cyclobutadiene may rest on the spin-contaminated states in Table 1; this is substantive but secondary to the direct resource-accounting problem. The recommendation is therefore to keep the reader's conditional verdict: require the authors to report total circuit counts including DI, or to explicitly limit their resource claims to HEA-layer counts and qualify the hardware-feasibility conclusion.","tokens_in":17807,"tokens_out":7193,"duration_ms":739161,"concrete_test":"Use the authors' GitHub code to reproduce the H4 (2-layer) and C4H4 (4-layer) LAS-nuVQE calculations, then transpile the full circuit including Qiskit's initialize for the LASSCF state with the same basis gates used in the paper (e.g., Rz, SX, CNOT). Compare the total CNOT count of the entire circuit to the reported 21 and 28 CNOTs. If including DI raises the CNOT count by more than roughly 50%, the '<70 total gates' claim must be revised; if the transpiled DI circuit is negligible, the claim stands.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that LAS-nuVQE reaches chemical accuracy with fewer than 70 total gates for H4 and C4H4. However, the computational details state that all simulations use DI state preparation with Qiskit, and the gate counts quoted in Section III.B (24 SQG + 21 CNOT for H4 at 2 layers; 40 SQG + 28 CNOT for C4H4 at 4 layers) count only the HEA layers. The DI circuit that loads the LASSCF fragment-product wave function is not included. For an 8-qubit active space, Qiskit's initialize of a generic 8-qubit state requires many controlled rotations and is far from negligible; even exploiting the product structure of two (2e,2o) fragments, state preparation is unlikely to cost less than the 21-28 CNOTs reported. Since the abstract and conclusion claim affordable quantum resources and '<70 total gates', the quoted counts understate the true circuit cost. The energy comparisons themselves are noiseless simulation results and are not invalidated, but the resource comparison against vanilla VQE/HEA is unfair unless the DI overhead is included. The omission is not acknowledged in the paper: Algorithm 1 says 'Prepare LASSCF ... with DI' without reporting its cost.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript introduces LAS-nuVQE, a variational quantum eigensolver variant in which a hardware-efficient ansatz is applied to a LASSCF initial state and a linearized qubit Jastrow operator J = 1 - sum_i alpha_i Z_i - sum_ij lambda_ij Z_i Z_j is folded into the objective function E = <Psi|J^dagger H J|Psi> / <Psi|J^dagger J|Psi>. Noiseless simulations are reported for H4 and square cyclobutadiene, with claims of chemical accuracy using fewer than 70 gates, spin-constrained calculations using a penalty term, measurement-cost mitigation via Pauli grouping and shot-frugal sampling, and wall-clock scaling estimates for up to 100 qubits. The paper argues that LAS-nuVQE is a more accurate and cheaper alternative to vanilla VQE with HEA for multireference systems.","tokens_in":18122,"tokens_out":5779,"duration_ms":51135,"significance":"If the resource claims held, this would be a practically useful recipe for near-term multireference quantum chemistry, because it combines a fragment-based multireference initial state with shallow HEA circuits and avoids the ancilla overhead of implementing nonunitary operations directly. The paper has genuine strengths: the energy expression in Eq. (10) is the exact expectation value of the non-unitary state rather than a fitted surrogate, comparisons are made to external FCI/CASCI references, the code is openly available, and the noiseless simulations consistently show LAS-nuVQE improving on LAS-VQE at equal HEA depth. However, the quantitative headline claims are not yet supported: the quoted gate counts exclude the LASSCF state-preparation circuit, and the cyclobutadiene chemical-accuracy result rests on heavily spin-contaminated states. The scalability conclusion likewise depends on strong, unverified assumptions.","major_comments":[{"comment":"The quoted gate counts (24 SQGs + 21 CNOTs for H4 and 40 SQGs + 28 CNOTs for cyclobutadiene) count only the HEA layers. The Computational Details state that all simulations use DI state preparation with Qiskit, and Algorithm 1 step 1 explicitly prepares the LASSCF wave function with DI. The DI cost is not included in the abstract's '<70 total gates' claim, in Section III.B, or in the conclusion. Please report the full circuit including the LASSCF initialization, or revise the headline claim to refer to HEA layers only; without this, the hardware-feasibility comparison is incomplete.","section":"Computational details and Algorithm 1; Sec. III.B"},{"comment":"The cyclobutadiene chemical-accuracy result in Fig. 5(a) is obtained from unconstrained runs whose spin expectation values range from 0.67 to 1.52 (Table 1), i.e., states that are not the targeted singlet. The spin-constrained energies in Table 2 are several mEh higher at the same layer counts (e.g., -153.6758 Ha at 4 layers versus -153.7023 Ha unconstrained) and are not compared to the CASCI reference, so it is unclear whether any spin-pure state reaches chemical accuracy. The headline claim for cyclobutadiene should therefore be restricted to the unconstrained states, or the spin-constrained data should be shown to meet chemical accuracy.","section":"Sec. III.B and III.C, Tables 1 and 2"},{"comment":"The wall-clock comparison relies on three unqualified assumptions: N_l = N_q, one Jastrow parameter saves one single-qubit gate, and shot counts drop by two orders of magnitude. The text provides no evidence or sensitivity analysis for these assumptions. Because the conclusion that nuVQE with grouping and shot-frugal sampling is cheaper than HEA at 100 qubits follows directly from these assumptions, the scalability conclusion should be presented as conditional and accompanied by a sensitivity study over layer counts and shot-reduction factors.","section":"Sec. III.E, Figs. 9 and 10"}],"minor_comments":[{"comment":"The word 'respectivity' should be 'respectively'.","section":"Sec. III.D"},{"comment":"The text contains 'walk clock estimation' and 'natural atom/trapped ion platform'; these should read 'wall clock estimation' and 'neutral atom/trapped ion platform'.","section":"Sec. III.E"},{"comment":"The quantity '103−4' appears to be a formatting error for 10^3-4 and should be typeset consistently.","section":"Abstract"},{"comment":"In the definition of WRS, M is described as 'the sum of the coefficient'; for the probability p_i = |c_i|/M to be normalized, M should be the sum of absolute values of the coefficients.","section":"Sec. II.B, Eq. (16)"}],"recommendation":"major_revision","confidential_remarks":"The DI gate-count omission is the most serious issue: the '<70 gates' phrase appears in the abstract and conclusion and would be misleading if interpreted as the total circuit cost. The spin-contamination problem is equally important for the cyclobutadiene claim. The paper's strengths, including the open code and the exact nonunitary objective, justify a revision rather than rejection. I would recommend that the editor require a corrected resource accounting and a clear separation between the unconstrained and spin-pure claims before publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The core finding holds up: in noiseless simulation, LAS-nuVQE consistently beats LAS-VQE at the same HEA depth, and it recovers most of the interfragment correlation missing from LASSCF. The measurement-mitigation analysis is careful and useful. That said, the paper's headline resource claim is not yet supported. The reported '<70 total gates' counts only the HEA layers; the LASSCF initial state is prepared with Qiskit's direct initialization, whose cost is not included. For an 8-qubit active space that is not negligible, and the claim that the method is affordable on hardware as-is is therefore overstated. The comparison against HF-VQE is also not apples-to-apples, since HF starts from a computational basis state at no extra cost.\n\nThe chemical-accuracy claim for cyclobutadiene is also narrower than it appears. The unconstrained LAS-nuVQE states that reach chemical accuracy have spin expectation values between 0.67 and 1.52, so they are not the correct singlet. The spin-constrained variant produces spin-pure states, but its energies are higher and do not reach chemical accuracy. The paper is transparent about this, but the abstract and conclusion lean on the unconstrained numbers.\n\nThe scalability model is a rough estimate built on hand-set assumptions, such as layers equal to qubits and one Jastrow parameter saving one SQG, with no sensitivity analysis. The wall-clock extrapolation is illustrative, not predictive. The shot-frugal results are limited to H2 with five runs, fine as a demonstration, but not a strong statistical case.\n\nNone of this is fatal. The combination of LASSCF with nuVQE is a natural and sensible step, and the noiseless energy comparisons are likely correct as reported. The code and data are available, which is good practice. The problems are addressable: include state-preparation cost, report spin-constrained chemical accuracy if it can be achieved, and soften or better justify the scalability claims.\n\nThis paper deserves a serious referee. It is not transformative, but it is a competent benchmark that could be useful to the quantum-chemistry NISQ community after revision. I would send it to review, with the expectation of major revision.","headline":"A solid noiseless benchmark combining LASSCF with nuVQE, but the resource claim omits state preparation and the chemical-accuracy result is spin-contaminated.","tokens_in":18665,"tokens_out":3794,"would_cite":true,"duration_ms":32267,"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":"LAS-nuVQE couples a cheap classical fragment wave function with a non-unitary Jastrow correction, reaching chemical accuracy on both test molecules with under 70 total circuit gates.","keywords":["variational quantum eigensolver","localized active space self-consistent field","non-unitary Jastrow operator","hardware-efficient ansatz","fragment-based quantum chemistry","measurement mitigation","Pauli grouping","shot-frugal sampling"],"falsifier":"Compile the full circuit for square cyclobutadiene including direct-initialization state preparation of the LASSCF wave function, count every gate, and run the optimization on a noisy processor; if the total gate count exceeds the coherence limits or the energy no longer stays within chemical accuracy of the CASCI reference, the paper's core resource claim fails. A noiseless equivalent is to repeat the simulations with the state-preparation circuit included in every gate count and wall-clock estimate.","tokens_in":17601,"feed_emoji":"⚛️","tokens_out":5477,"duration_ms":46990,"temperature":0.7,"pith_summary":"The paper proposes combining the localized active space self-consistent field (LASSCF) initial state with a non-unitary variational quantum eigensolver (nuVQE), in which a Jastrow-inspired operator is folded into the measured expectation value instead of the circuit. The aim is to recover the interfragment electron correlation that LASSCF drops while keeping circuits shallow enough for near-term hardware. The paper reports chemical accuracy for H4 with two hardware-efficient ansatz layers (24 single-qubit and 21 CNOT gates) and for square cyclobutadiene with four layers (40 single-qubit and 28 CNOT gates), and shows that at equal circuit depth the non-unitary correction beats plain VQE with a hardware-efficient ansatz. If these resource counts survive real hardware accounting, the method offers a route to multireference quantum chemistry without deep chemically inspired circuits.","feed_headline":"Chemical accuracy with under 70 quantum gates, via LAS-nuVQE","feed_subtitle":"A non-unitary Jastrow correction on a shallow circuit recovers electron correlation that cheap fragment methods miss.","key_machinery":"The load-bearing object is the linearized qubit Jastrow operator J(α,λ) = 1 − Σ_i α_i Z_i − Σ_{i<j} λ_ij Z_i Z_j, applied through the modified expectation value E = ⟨Ψ|J†H_QJ|Ψ⟩ / ⟨Ψ|J†J|Ψ⟩. Because J contains only diagonal Pauli-Z strings, it never enters the circuit; its classical parameters are optimized alongside the hardware-efficient ansatz angles, and the denominator accounts for the non-unitary normalization. This is what lets nuVQE add correlation without adding gates.","core_discovery":"The central claim is that the non-unitary operator J = 1 − Σ_i α_i Z_i − Σ_{i<j} λ_ij Z_i Z_j, applied not as a circuit but through the expectation values of J†H_QJ and J†J, restores interfragment correlation on top of a LASSCF reference while leaving the circuit depth unchanged. Using a linear-connectivity hardware-efficient ansatz built on a directly initialized LASSCF state, the authors find that LAS-nuVQE reaches chemical accuracy in the active-space Hamiltonians of H4 and square cyclobutadiene with fewer than 70 gates total, and that it recovers over 95% of the interfragment correlation energy missing from LASSCF at two layers and above 99.9% at four layers. They further show that adding a spin-penalty term produces spin-pure energies for square cyclobutadiene, and that Pauli grouping together with shot-frugal sampling cuts the measurement cost of the J-transformed Hamiltonian by one to two orders of magnitude, bringing estimated one-query wall-clock times below those of vanilla VQE with HEA.","pith_inferences":["Because the Jastrow operator is diagonal in the computational basis, its measurement overhead is concentrated in diagonal Pauli strings; combining LAS-nuVQE with readout-error mitigation or sparse Pauli sampling may further cut the cost beyond the paper's grouping and shot-frugal results.","The same non-unitary correction could be applied on top of any fragment-based initial state, not only LASSCF; the paper's mechanism suggests cluster mean-field or density-matrix-embedding references would benefit similarly, but this is not tested.","The gate-count assumption for direct initialization is the natural next thing to check: if fragment-product states can be prepared by fewer gates than one HEA layer, the under-70-gates regime generalizes to larger fragments.","Spin constraints in the paper address S^2; extending the penalty to particle-number or spatial-symmetry operators could make HEA applicable to excited states, a testable extension."],"forward_implications":["At equal HEA depth, LAS-nuVQE yields lower energy error than VQE with the same HEA for H4 at both geometries and for square cyclobutadiene at every tested number of layers.","The non-unitary correction recovers more than 95% of the LASSCF-missing interfragment correlation at two layers and over 99.9% at four or more layers for square cyclobutadiene.","The spin-constrained variant restores the correct S^2 = 0 state for square cyclobutadiene while retaining an accuracy edge over spin-constrained plain VQE.","With qubit-wise or fully-commuting Pauli grouping plus shot-frugal sampling, the measurement overhead of the J-transformed Hamiltonian drops by up to two orders of magnitude, making one query of nuVQE faster in wall-clock estimates than one query of HEA-based VQE at 100 qubits on the modeled superconducting platform.","LAS-nuVQE reaches chemical accuracy with under 70 total gates (24 single-qubit plus 21 CNOT gates for H4; 40 single-qubit plus 28 CNOT gates for square cyclobutadiene)."],"supporting_citations":[{"why":"Supplies the nuVQE method: non-unitary Jastrow-inspired qubit operators with HEA, evaluated through modified expectation values.","marker":"23"},{"why":"Define the LASSCF wave function and its variational optimization, the initial state on which the paper builds.","marker":"25,26"},{"why":"Earlier LAS-UCC method that restores interfragment correlation with unitary coupled cluster; the paper's cost comparison target.","marker":"35"},{"why":"LAS-USCC, the selective coupled-cluster alternative that motivated the search for cheaper post-LASSCF correlation.","marker":"19"},{"why":"Addresses state preparation for fragment-based quantum chemistry, backing the direct-initialization assumption behind the gate counts.","marker":"52"},{"why":"Provides the spin-penalty approach used for the spin-constrained LAS-nuVQE variant.","marker":"62"},{"why":"Ground the shot-frugal sampling and latency model used to estimate measurement costs and wall-clock times.","marker":"73,74"},{"why":"Foundational VQE and measurement-grouping formalism used for Pauli grouping and Hamiltonian measurement.","marker":"6,9"}],"fun_headline_variants":["LAS-nuVQE hits chemical accuracy with under 70 gates","Under 70 gates, LAS-nuVQE recovers correlation","LAS-nuVQE: <70 gates to chemical accuracy","LAS-nuVQE: chemical accuracy, spin purity, low measurement cost","Shallow circuits, deep chemistry: LAS-nuVQE passes chemical accuracy"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The quoted gate counts count only the hardware-efficient ansatz layers on top of the initial state, assuming the LASSCF fragment wave function can be prepared by direct initialization at a cost that does not affect the resource numbers; if that state preparation requires circuits comparable to or deeper than the ansatz itself, the under-70-gates claim would not hold on hardware.","fun_headline_variants_meta":{"raw":{"variants":["LAS-nuVQE hits chemical accuracy with under 70 gates","Under 70 gates, LAS-nuVQE recovers correlation","LAS-nuVQE: <70 gates to chemical accuracy","LAS-nuVQE: chemical accuracy, spin purity, low measurement cost","Shallow circuits, deep chemistry: LAS-nuVQE passes chemical accuracy"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000703,"raw_usage":{"total_tokens":3265,"prompt_tokens":1135,"completion_tokens":2130,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":751,"completion_tokens_details":{"reasoning_tokens":2046}},"tokens_in":751,"tokens_out":2130,"duration_ms":16334,"temperature":1.0,"reasoning_tokens":2046,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T16:01:10.688716+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compile the full circuit for square cyclobutadiene including direct-initialization state preparation of the LASSCF wave function, count every gate, and run the optimization on a noisy processor; if the total gate count exceeds the coherence limits or the energy no longer stays within chemical accuracy of the CASCI reference, the paper's core resource claim fails. A noiseless equivalent is to repeat the simulations with the state-preparation circuit included in every gate count and wall-clock estimate.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the nuVQE method: non-unitary Jastrow-inspired qubit operators with HEA, evaluated through modified expectation values."}],"review_version":1}