{"id":"f6142d2b-b48b-4700-83fe-66a4edf6b520","arxiv_id":"2512.04801","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A hybrid VQE-CVQE scheme using a few-step 'diabatic' evolution to build a guiding state, followed by classical diagonalization in the sampled subspace, yields chemically accurate ground-state energies in toy-model and 50-qubit hardware tests.","lead":"A new hybrid algorithm merges a coarse quantum time-evolution step with classical subspace diagonalization to estimate ground-state energies. The authors demonstrate it on an 8-orbital toy model and a 50-qubit IBM processor, reporting near-exact energies for a non-interacting test case.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central claim rests on an unproven subspace heuristic: for Nτ=O(1), measured bitstrings from U(Nτ,Δτ)|Φ0⟩ plus H-couplings are assumed to span the significant part of the ground state; the V=0 hardware run cannot test this, and no interacting demonstration establishes it.","rationale":"The paper is trying to show that a small number of Trotter steps in a diabatic evolution produces a guiding state whose measurement outcomes define a subspace containing the relevant part of the ground state. This is exactly the condition that must hold for the algorithm's central claim. The reader identified this as weakest, and I agree. The Appendix A argument is the only analytical support, and it is heuristic: comparing the Dyson series of U(1,T) and U_A(T) shows same operator support, but weights differ and 'similar ordering' is not quantified. More importantly, the actual subspace B is generated by thresholded bitstrings, not by operator strings; the passage from operator support to bitstring support is not made. Appendix E proves something different — that for an eigenstate, the high-weight basis states capture its energy — and it does not address whether the exact ground state has sufficient weight in B. The V=0 hardware demonstration is a useful engineering benchmark but does not exercise the interacting part of the method, and the small interacting simulation is in a regime where the subspace is almost complete. So the strongest claim is plausible but not established. The reader's CONDITIONAL verdict is appropriate; my stress test does not move it. The proposed exact-diagonalization scan would settle the subspace heuristic directly.","tokens_in":22844,"tokens_out":9778,"duration_ms":93770,"concrete_test":"Run exact-diagonalization scans on the same spinless model (Eq. 4) with Q=10–12, Ne=Q/2, V/t ∈ {0.5,1,2,4}, and Δµ/t ∈ {0.1,0.2,0.5,1.0}. For Nτ=1 and Δτ=1/15τ0, construct the ideal B from all bitstrings with |⟨n|U|Φ0⟩|² above a realistic shot threshold (e.g., 1/8192) plus their H-couplings, then compute E_B−E_gs and the overlap |⟨ψ_gs|P_B|ψ_gs⟩|². Repeat for Nτ=2,4 and for the exact adiabatic U_A(T). If the overlap falls below about 0.9 or the energy error exceeds chemical accuracy for any moderate-V point, Appendix A's 'same excitations' heuristic fails; if it holds across the scan, the central concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Load-bearing concern: the central guarantee of the algorithm is that span(B) contains a chemically accurate approximation to the exact ground state. For Nτ=O(1) this is an assumption, not a consequence. Appendix A expands U(1,T) and U_A(T) (Eqs. A3–A6) and observes that both contain the same operator strings with different weights; the claim that the weights are 'similar in ordering' is asserted and used to conclude that the same excitations are generated. But the subspace B is not built from operator strings; it is built from bitstrings whose measured amplitudes exceed a threshold (Appendix C, ε-cutoff). A term with small weight in U(1,T) can correspond to a basis state essential to the exact ground state but below the shot threshold; if the ground state has small but collectively important amplitudes on many dropped states, the diagonalization of H_B misses them. Appendix E does not cover this: it argues only that expectation values of an eigenstate are approximated by its high-weight basis states, not that the exact ground state has high weight in B. The 50-orbital hardware experiment uses V=0, a noninteracting, classically solvable model, so it cannot validate the interacting subspace heuristic. The only interacting demonstration is the Q=8 example, where Fig. 2c shows the subspace dimension approaches the full 70-state space, so it does not probe the scalable regime.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes a hybrid quantum-classical eigensolver ('VQE-CVQE') in which the quantum circuit implements a highly discretized diabatic evolution U(Nτ,Δτ)=T∏ exp(−iH(τ_i)Δτ) from a trivial Hamiltonian H0 to the target H (Eqs. 1–2). The resulting guiding state is measured in the computational basis; the observed bitstrings B0, together with their couplings B1 under H, define a subspace V=span(B) (Table I, steps 3–4). The target Hamiltonian is projected onto V and diagonalized classically (Eq. 3), yielding a variational upper bound E_B, which is then optimized by scanning (Nτ,Δτ). The authors demonstrate the method by classical simulation on an 8-orbital, 4-electron interacting model (Fig. 2), identifying three Nτ regimes, and on the IBM Brisbane 50-qubit device for a 50-orbital, 25-electron noninteracting model (V=0; Figs. 3–4), reporting small energy errors. They further argue that the method is resilient to hardware noise and that it interpolates between NISQ and fault-tolerant operation. Appendices provide the gate decomposition, a shot-collection comparison, a demonstration that the method reduces to CVQE for fixed parameters, and an argument that eigenstate expectation values are well approximated by high-weight basis states.","tokens_in":23230,"tokens_out":17439,"duration_ms":162842,"significance":"If the subspace heuristic holds, the algorithm is a useful addition to the guided-sampling/subspace-expansion family: it combines a physically motivated diabatic ansatz with a classical diagonalization step that is guaranteed by the variational principle (Appendix D) to be no worse than the guiding-state expectation value, and it offers a concrete bridge between NISQ (small Nτ) and FTQ (large Nτ) implementations. The paper's strengths are its transparency: the 8-orbital classical simulation explicitly compares E_B with the guiding-state energy; the choice V=0 for the Brisbane run is openly justified as enabling comparison with known values; and the relationship to the authors' earlier CVQE method is stated plainly in Appendix D. The 50-qubit hardware experiment is a substantive resource, and the reported raw errors are encouraging. However, the strength of the paper's claims currently exceeds the evidence: the scalable interacting regime is untested, and the hardware results are reported without statistics or an explicit energy scale.","major_comments":[{"comment":"The load-bearing claim that for Nτ=O(1), span(B) approximates the exact ground state is not established. Appendix A compares operator strings of U(1,T) and U_A(T) (Eqs. A3–A6) and asserts a 'similar ordering of the weights,' but B is formed by thresholding measured bitstring amplitudes (Eq. C8) and one H-application (Eq. C9); a bitstring essential to the exact ground state can have small weight under U(1,T) and fall below the shot cutoff. Appendix E shows only that a high-weight component expansion approximates expectation values of an eigenstate, not that the exact ground state has high weight in B (and the step from Eqs. E12 to E13 neglects phase cancellations). Fig. 2c shows the 8-orbital subspace approaching the full 70-state space, so it does not test the scalable regime. Recommend a solvable interacting test with |B|≪dim F (e.g., Q=12–16, V≠0) and a cutoff/shot study, or constrain","section":"Sec. II / Appendix A / Appendix C / Appendix E"},{"comment":"The headline claim 'well within chemical accuracy' (abstract, conclusion) is not quantified. Energy errors are plotted in units of t, but no value of t is stated for the 50-orbital run (the t=1/15 Ha value appears only in the Fig. 2 caption). No shot counts, repetitions, or error bars are reported for any hardware point; the scatter in Fig. 4a is attributed to hardware noise purely qualitatively. Please specify the Hamiltonian parameters used, report absolute errors in Hartree with an explicit chemical-accuracy definition, and provide error bars (repeated runs or shot bootstrapping) for the claimed operating point (Nτ=1, Δτ=1/15τ0) and for at least a subset of the Fig. 3 data.","section":"Sec. III C, Figs. 3–4"},{"comment":"The hardware run is not a demonstration on an interacting system. For V=0 the Hamiltonian is quadratic, so the ideal U(Nτ,Δτ) maps the initial Slater determinant to another Slater determinant, and a single H-application from configurations near the Fermi sea generates mostly single-excitation (and a few double-excitation) states; the reported 2000–2500-state subspace therefore contains significant noise-inflated contributions. The experiment is better described as a test of whether the noisy device produces a subspace containing the relevant single-particle excitations of a solvable one-body model. The abstract's claim that the algorithm is demonstrated 'on a system of interacting electrons' is supported only by the 8-orbital classical simulation (Fig. 2), where the subspace saturates the full space (Fig. 2c). Please rephrase abstract and conclusion to separate hardware and simulation cl","section":"Sec. III C / Abstract"},{"comment":"The statement that the method is 'resilient to small fluctuations' is asserted, not derived. The argument ignores that hardware noise systematically distorts the guiding-state distribution (weight shift to high-energy bitstrings) rather than merely adding shot noise, and the shot cutoff ε of Appendix C is never specified for the hardware runs. Fig. 4 itself shows the landscape is noise-dominated (Nτ=1 optimal; Δτ=0 already accurate), so the CVQE optimization over (Nτ,Δτ) is not meaningfully exercised. Please either supply a quantitative error model (e.g., probability flux into/out of B0 per gate given the quoted 1.9% EPLG) and the measurement protocol (shots, cutoff), or soften the resilience claim.","section":"Sec. III C (noise-resilience paragraph)"}],"minor_comments":[{"comment":"Typos: 'an typical' (Sec. III B); 'qubic' (Sec. III C); 'Hailtonian' (Appendix D); 'cuttoff' (Appendix C); 'wieghts' (Appendix A, twice); 'gor' (Eq. C10); 'Variation Quantum Eigensolver' (Sec. I); 'state states' (Fig. 1 caption).","section":"Throughout"},{"comment":"Define λ(θ) explicitly (Hermitian operator?) and clarify whether the equivalence to CVQE for fixed (Nτ,Δτ) carries any practical distinction from Refs. [24–26] beyond the choice of ansatz.","section":"Appendix D, Eqs. (D1)–(D4)"},{"comment":"The color legend ('thick blue,' 'medium orange,' 'thin green') does not match the visible grayscale figure; label the curves directly or provide a matching legend.","section":"Fig. 2 caption"},{"comment":"The medium-regime plateau is described as 'nearly independent of the variational parameters'; consider stating E_B relative to the exact ground-state energy (e.g., E_B/E_exact−1) so the quality of the plateau can be assessed without reading the axis.","section":"Sec. III B / Fig. 2a"}],"recommendation":"major_revision","confidential_remarks":"The reader's conditional assessment matches mine. The variational logic is sound; the issue is the gap between the evidence and the claimed guarantees. The V=0 hardware run is essentially a noisy single-particle (or CIS-level) subspace construction, so the claim that the algorithm handles interacting systems at scale rests entirely on the unproven subspace heuristic. A revision with a genuine interacting demonstration at nontrivial subspace compression, plus proper statistics and Hartree-unit reporting, would substantially strengthen the paper. The self-citation pattern is proportionate for a direct continuation of the authors' CVQE line."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Ben, quick take on 2512.04801. The paper proposes a hybrid VQE-CVQE where the guiding state is generated by a few-step Trotterized diabatic evolution, and the variational parameters are (Nτ, Δτ). The post-processing uses the measured bitstrings to build a subspace, close it under the Hamiltonian, and diagonalize classically. That specific integration is new; for fixed parameters it is just CVQE with the guided-sampling ansatz, as they state in Appendix D. The three-regime characterization is a nice observation, and the classical Q=8 simulation shows the regimes clearly. The hardware run on 50 qubits is a genuine machine demo, and they provide full device specifications, so reproducibility is decent.\n\nThe soft spots, in order of importance. First, the load-bearing premise — that span(B) contains a chemically accurate approximation of the ground state for Nτ=O(1) — is asserted, not proven. Appendix A shows that U(1,T) and the full adiabatic evolution generate the same operator strings with similar weight ordering, and then says 'we expect' the subspaces are similar. But B is built from measured bitstrings above a threshold, not from operator strings. A term with small weight in U(1,T) can correspond to a basis state that nonetheless carries important ground-state amplitude. Appendix E only proves that expectation values of an eigenstate are approximated by its high-weight basis states; it doesn't prove the ground state has high weight in B. The V=0 hardware run cannot test this, because a noninteracting system is trivial for adiabatic preparation. The only interacting test is the Q=8 example, and there the subspace dimension approaches the full 70-state space, so it doesn't probe the scalable regime.\n\nSecond, the hardware results lack error bars, shot counts, and baseline comparisons with VQE or CVQE. The claim that the method is 'better' than either alone isn't demonstrated. The noise-resilience argument is a one-paragraph assertion. Third, the conclusion extrapolates the three regimes to 'many physically relevant models' without evidence. These are real gaps, but the paper is transparent about them — it repeatedly says 'we expect' — and the variational upper-bound structure means the output is not forced by construction. The math checks out.\n\nWho is this for? Anyone working on quantum subspace diagonalization or hybrid VQE variants. It's a reasonable incremental contribution, not a breakthrough. It deserves a serious referee, but the referee should ask for an interacting demonstration, shot statistics, and a baseline comparison before the strong claims are accepted.","headline":"A plausible hybrid subspace method with a real hardware demo, but the central subspace heuristic is unproven and the only interacting test is a toy model.","tokens_in":23723,"tokens_out":2503,"would_cite":true,"duration_ms":22904,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A single coarse Trotter step of a diabatic interpolation yields a measured subspace whose lowest eigenvalue is the ground-state energy to chemical accuracy, even on noisy hardware.","keywords":["variational quantum eigensolver","cascaded variational quantum eigensolver","diabatic state preparation","subspace diagonalization","guided sampling ansatz","NISQ","ground-state energy","Jordan-Wigner transformation"],"falsifier":"In a noiseless classical simulation of the 50-orbital non-interacting model, set Nτ = 1 and choose parameters where the guiding state has essentially zero overlap with the exact ground state (for example, Δμ ≪ t where the initial filling is not the adiabatic ground state). If the lowest eigenvalue of H_B misses chemical accuracy, the central claim is false. More directly, compute the squared overlap between the exact ground state and V = span(B0 ∪ H B0): the claim predicts this overlap is close to 1, and a value much below 1 falsifies it.","tokens_in":22701,"feed_emoji":"⚛️","tokens_out":4609,"duration_ms":40988,"temperature":0.7,"pith_summary":"This paper proposes a hybrid quantum-classical algorithm that computes ground-state energies without ever preparing the ground state itself. The quantum circuit applies a very short, heavily discretized version of adiabatic state preparation—in the demonstrated case, a single time step—starting from a simple initial Hamiltonian. The measured computational basis states, together with all states they couple to through the system Hamiltonian, define a small subspace; the paper's claim is that this subspace contains enough of the true ground state that diagonalizing the Hamiltonian restricted to it yields the ground-state energy to chemical accuracy. The authors demonstrate this on an 8-orbital interacting model in classical simulation and on a 50-orbital, 25-electron non-interacting system run on IBM Brisbane, obtaining energies within chemical accuracy despite roughly two percent error per gate layer. The algorithm is designed to work in three regimes—small, intermediate, and large numbers of time steps—corresponding to near-term noisy, intermediate, and fault-tolerant quantum computers.","feed_headline":"One coarse time step reaches chemical accuracy on 50 orbitals","feed_subtitle":"A tiny subspace built from measured bitstrings yields ground-state energies that survive real hardware noise.","key_machinery":"The machinery is the diabatic evolution operator U(Nτ, Δτ) = T ∏ e^{-i H(τ_i Δτ) Δτ}, interpolating linearly between a trivial Hamiltonian H0 and the target H. With Nτ as small as one, U produces a guiding state |Ψ0⟩; repeated measurements yield basis states B0, and applying H to B0 yields coupled states B1. The effective Hamiltonian H_B is the projection of H onto V = span(B0 ∪ B1), and its lowest eigenvalue E_B is computed classically. The load-bearing identity in Appendix A compares U(1,T) with the full adiabatic evolution U_A(T) and shows the two contain the same operator strings with similar weight ordering, which is why the authors expect the single-step subspace to capture the same lo","core_discovery":"On the paper's own terms, the central discovery is that a highly discretized Trotterized evolution—even Nτ = 1—generates a guiding state whose computational-basis measurement outcomes, when closed under application of the Hamiltonian, span a subspace in which the true ground state is well represented. The lowest eigenvalue of the projected Hamiltonian H_B is then a chemically accurate estimate of the ground-state energy. Because the subspace dimension grows polynomially and the circuit depth is set by Nτ, the method fits near-term hardware; the authors further show numerically and on IBM Brisbane that small fluctuations in the measured probability distribution barely affect the final energy,","pith_inferences":["This is effectively a measurement-guided selected configuration-interaction performed classically; a natural extension is to use the same subspace construction with other cheap guiding states (e.g., Hartree-Fock plus a few excitations) to benchmark how much diabatic preparation adds.","The error-bound intuition in Appendix E suggests accuracy is tied to eigenstate weight concentration: if the true ground state has a long tail of small-amplitude, high-energy components, more shots or a larger B may be required; this predicts a testable relationship between shot count and energy error.","The authors demonstrate only V = 0 on hardware; for interacting systems the Hamiltonian connects more states, so the subspace could grow faster, and the Nτ = 1 heuristic would need to be re-tested—a concrete next experiment.","The three-regime picture implies a simple operational rule: choose Nτ as large as hardware noise allows, then let the classical diagonalization absorb the remaining error; this could inform resource estimates for near-term quantum chemistry."],"forward_implications":["The same algorithm spans hardware generations: with Nτ small it runs on today's noisy devices, with Nτ moderate it behaves as CVQE with no parameter updates, and with Nτ large it approaches adiabatic state preparation for fault-tolerant machines.","On current hardware, Nτ = 1 is optimal for large systems: the accuracy gain from more Trotter steps is outweighed by additional gate noise, so the variational parameter update can be skipped.","The energy error is insensitive to small measurement fluctuations, because tiny changes in sampled probabilities do not add or drop basis states; the method therefore tolerates gate error rates around one to two percent per layer.","When the guiding state is a poor approximation (the small-Nτ regime), E_B can still be a good approximation, so the classical diagonalization step does useful work the quantum circuit cannot.","For the 50-orbital system the subspace has only about 2,000–2,500 states, versus roughly 10^14 states in the full electron-preserving Fock space, making the classical diagonalization tractable."],"fun_headline_variants":["One coarse Trotter step still yields chemical accuracy","Single step diabatic state prepares chemically accurate eigenstate","Tiny subspace from measured bits gives ground-state energy","A single Trotter step plus subspace beats hardware noise","Hybrid VQE-CVQE one step to chemical accuracy"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The key premise is the expectation, stated in the main text and Appendix A, that even a single coarse Trotter step produces a guiding state whose measured bitstrings, closed under the Hamiltonian, span a subspace containing a good approximation to the true ground state; if that fails, no classical diagonalization of the subspace can recover the correct energy.","fun_headline_variants_meta":{"raw":{"variants":["One coarse Trotter step still yields chemical accuracy","Single step diabatic state prepares chemically accurate eigenstate","Tiny subspace from measured bits gives ground-state energy","A single Trotter step plus subspace beats hardware noise","Hybrid VQE-CVQE one step to chemical accuracy"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000613,"raw_usage":{"total_tokens":2621,"prompt_tokens":614,"completion_tokens":2007,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":358,"completion_tokens_details":{"reasoning_tokens":1930}},"tokens_in":358,"tokens_out":2007,"duration_ms":13890,"temperature":1.0,"reasoning_tokens":1930,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T18:30:59.186300+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"In a noiseless classical simulation of the 50-orbital non-interacting model, set Nτ = 1 and choose parameters where the guiding state has essentially zero overlap with the exact ground state (for example, Δμ ≪ t where the initial filling is not the adiabatic ground state). If the lowest eigenvalue of H_B misses chemical accuracy, the central claim is false. More directly, compute the squared overlap between the exact ground state and V = span(B0 ∪ H B0): the claim predicts this overlap is close to 1, and a value much below 1 falsifies it.","supporting_citations":[],"review_version":1}