{"id":"9e341b20-fb6c-494c-86c6-fb5edab8b8ef","arxiv_id":"2412.07933","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A quantum chemistry feasibility study concludes that simulating the oxygen reduction reaction on copper with VQE or QPE is impractical with near-term and near-future quantum hardware.","lead":"This paper builds a hybrid quantum-classical workflow to model the first step of oxygen reduction, a key corrosion trigger on copper-rich aluminum alloys. It finds that even for tiny model systems, quantum computers would need hours of runtime and over a hundred thousand physical qubits, so real corrosion simulation remains far out of reach.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The apparent multiconfigurational character may be an artifact of active spaces with only one virtual orbital; if larger active spaces restore single-reference character, the quantum-computing motivation loses its chemical basis.","rationale":"The reader's CONDITIONAL verdict is, in my view, the right one, and the condition I would impose is essentially the first part of the reader's weakest assumption. The paper is otherwise honest: it labels the bottleneck choice as anticipation, it verifies VQE convergence against exact diagonalization, and its resource estimates are generated with standard tools and stated assumptions. I do not see an internal inconsistency that overturns the negative feasibility conclusion for the specific Hamiltonians constructed. The load-bearing weakness is upstream: the active spaces are the only evidence that the ORR step is strongly correlated, and by construction they contain a single virtual orbital, so the FCI can only mix the HF configuration with configurations that populate sigma*(O-O). That makes the strong-correlation claim vulnerable to active-space enlargement. I therefore agree with the reader's active-space concern but do not elevate the bottleneck issue to the same level, because the text explicitly says the bottleneck choice is anticipated rather than computed. The recommended verdict remains CONDITIONAL, unchanged from the reader's assessment.","tokens_in":19629,"tokens_out":9413,"duration_ms":100138,"concrete_test":"Rerun the FCI/CASSCF at the TS_diss geometry with an expanded active space that contains at least one additional virtual orbital selected by the same AVAS procedure (e.g., (10,8) or (12,10), adding the next O 2p/Cu 3d virtual), and compare natural orbital occupations and the leading CI coefficient with the current (10/6) result. If the leading determinant weight remains below about 0.9 and the sigma* natural occupation stays near 1 in the larger space, the multiconfigurational claim survives; if the weight rises above about 0.95 and the extra virtuals remain nearly empty or fully occupied, the one-virtual active space created the apparent strong correlation and the quantum-computing motivation for this ORR step is not established.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing premise for choosing this chemistry is that the reductive dissociation of O2 on Cu has a strongly correlated ground state. The only evidence is the FCI analysis in Sec. 2.2.2, and Sec. 2.2.2 itself discloses that all three active spaces contain exactly one virtual orbital, the sigma*(O-O), so at most double excitations are possible. In such a truncated space, the FCI wavefunction can lower energy only by populating that single antibonding orbital; the large FCI-HF correlation energy and the leading double-excitation coefficient in Fig. 3C/D therefore measure the response of a deliberately restricted Hilbert space, not a robust property of the embedded O2/Cu Hamiltonian. If an expanded active space or a multireference benchmark with dynamical correlation shows the TS_diss wavefunction remains dominated by the HF configuration, then the 'strongly correlated' characterization is an artifact of the one-virtual restriction. The resource estimates in Sec. 2.3 would still be valid for the constructed Hamiltonians, but the central motivation—that this ORR step is a natural quantum-computing target—would fail, and the industrial relevance claim would lose its chemical basis. The bottleneck identification is labeled 'anticipate' and 'speculate' in Sec. 2 and is therefore secondary.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper develops and demonstrates a hybrid classical-quantum workflow for the initial steps of the oxygen reduction reaction (ORR) on a copper surface, used as a model for Cu-rich intermetallic particles in aerospace aluminum alloys. The authors combine DFT (PBE), Hartree-Fock, and FCI in AVAS-constructed active spaces of (2/2), (6/4), and (10/6) electrons/orbitals to map the reductive dissociation pathway, and they report multiconfigurational character near the reactant and at the dissociation transition state. Using the resulting embedding Hamiltonians, they run VQE (UCCSD) simulations and fault-tolerant QPE resource estimations with the Microsoft Azure Quantum Resource Estimator, concluding that even these small active spaces demand prohibitively large resources: roughly two hours and about 135,000 physical qubits for QPE on the (10/6) active space, and over one hour per VQE iteration with error correction. The paper is explicitly negative in its central conclusion, and it discloses several limitations, including the use of classical diagonalization to set the QPE parameters and the restriction of all active spaces to a single virtual orbital.","tokens_in":19887,"tokens_out":9709,"duration_ms":96828,"significance":"If the chemical premise holds, this is a valuable and unusually honest feasibility study: it is, to my knowledge, the first corrosion-specific workflow that carries a concrete molecular problem from DFT/NEB through embedding, VQE, and QPE to physical resource counts, using reproducible standard toolchains (Quantum ESPRESSO, PySCF, Qiskit, Azure Quantum Resource Estimator). The resource estimates rest on explicitly stated physical assumptions (gate times, error rates, Trotter steps, ancilla counts) and on a careful shot-count derivation in the supplementary information whose worst-case bound is parameter-free apart from the chemical-accuracy threshold. The paper deserves explicit credit for flagging the circular use of the classically computed spectrum in the QPE parameter choice (SI Sec. 1: \"defeats the purpose\") and for disclosing that every active space contains exactly one virtual orbital (Sec. 2.2.1).","major_comments":[{"comment":"The central chemical finding, namely strong multiconfigurational character at the reactant and TSdiss geometries, is supported only by FCI calculations in which every active space contains a single virtual orbital (sigma* O-O). As the authors themselves note, this limits excitations to doubles, so the large FCI-HF correlation energies in Fig. 3C and the leading double-excitation coefficients in Fig. 3D measure the response of a deliberately truncated Hilbert space rather than a robustness property of the embedded O2/Cu Hamiltonian. This is load-bearing for the paper's motivation, and I therefore ask for a robustness test: enlarge at least the (10/6) space with additional virtual orbitals, or benchmark with a multireference method including dynamical correlation (CASPT2/NEVPT2 or DMRG). The reversed thermodynamic driving force in the (6/4)/(10/6) FCI profiles of Fig. 3B should also be revisited in that test, since it currently contradicts the DFT profile and may be an artifact of the missing correlation space.","section":"Sec. 2.2.1, Fig. 3"},{"comment":"The QPE resource numbers are conditioned on parameters chosen with knowledge of the exact eigenspectrum obtained by classical diagonalization, which SI Sec. 1 explicitly concedes \"defeats the purpose of using the QPE algorithm.\" This matters quantitatively because the QPE runtime scales exponentially with the number of ancilla qubits (the circuit contains on the order of 2^m controlled evolutions): if the (10/6) estimate required 16 instead of 15 ancilla qubits, or 14 instead of 13 for (6/4), the reported runtimes would roughly double or quadruple. The convergence analysis in the SI covers only the (2/2) and (6/4) Hamiltonians, and for (6/4) the claimed-sufficient 13 ancilla setting was not actually simulated (the text reports only that 12 ancilla \"fall very close\" to the chemical-accuracy band); no convergence data are given for (10/6). The authors should provide the (10/6) convergence data or an explicit extrapolation procedure, reconcile the (2/2) inconsistency described in the minor comments, and add a sensitivity analysis of runtime to the ancilla and Trotter-step choices. This does not overturn the qualitative impracticality conclusion, but it is load-bearing for the specific headline numbers (about two hours, 135,000 qubits).","section":"Sec. 2.3.2, Methods, SI Sec. 1"},{"comment":"The paper identifies the reductive dissociation (step 2) as the bottleneck of the overall ORR and justifies the choice of this reaction step by \"anticipat[ing]\" it to be rate-determining and \"speculat[ing]\" about strong correlation. The competing steps, in particular the first electron transfer, are not computed, so the bottleneck claim is asserted rather than demonstrated. Because the industrial-relevance framing of the paper rests on this step being the controlling reaction, I ask the authors either to compute or cite barriers for the competing ORR steps, or to reformulate the bottleneck language so that the selection of step 2 is presented as a motivated model choice rather than an established mechanistic conclusion.","section":"Sec. 2, first paragraph"}],"minor_comments":[{"comment":"The sentence \"the QPE algorithm provides only a polynomial advantage with respect to classical methods – due to the exponential reduction of the overlap between the HF wavefunction and the true ground state\" is internally inconsistent: an exponentially small HF overlap implies an exponential QPE runtime starting from the HF state, not a polynomial advantage. Please rephrase using the argument of reference [35].","section":"Sec. 2.3.2"},{"comment":"The Methods section states that 3 and 1 Trotter steps with 10 and 13 ancilla qubits were sufficient for (2/2) and (6/4), respectively, but the SI text for (2/2) says that for n=3 and n=4 the results with 10 and 12 ancilla qubits remain outside the chemical-accuracy band, and for (6/4) it acknowledges that the 13-ancilla point was not simulated. These statements should be reconciled.","section":"Methods vs. SI Figs. SI2-SI3"},{"comment":"The sentence \"Fig. SI 1 demonstrates that the choice of the number of Trotter steps, n, is a critical parameter\" appears to refer to Fig. SI2, not the eigendecomposition figure; please correct the cross-reference.","section":"SI Sec. 1, text near Fig. SI1"},{"comment":"There are typographical errors that should be fixed: \"his reaction is catalyzed\" should be \"this reaction is catalyzed\" in Sec. 2.1, and the abstract's reaction equation \"O 2 + 4 e- 2 O2-\" is garbled.","section":"Sec. 2.1 and Abstract"},{"comment":"The caption reads \"Empty stars are the estimated denote calculations without considering the overhead of QEC\"; the wording is incomplete and should be rewritten.","section":"Fig. 4A caption"},{"comment":"The SI reference list duplicates and renumbers main-text references (for example, SI ref. [7] is the same Qiskit paper as main-text ref. [48]); if the SI is meant to be read standalone, the numbering should be made consistent.","section":"Supplementary references"},{"comment":"Given the emphasis on resource estimation, the authors should consider providing the embedding Hamiltonian matrix elements and the Azure Quantum Resource Estimator input parameters as supplementary machine-readable files, rather than only \"upon request,\" to make the runtime and qubit counts fully reproducible.","section":"Data availability"}],"recommendation":"major_revision","confidential_remarks":"This is a transparent, industry-relevant feasibility study with an honestly negative conclusion, and the authors' disclosure of the one-virtual active-space restriction and of the circular QPE parameter setting is commendable. The main risk is that the strong-correlation premise for the ORR step is not yet validated; the requested robustness test (larger active space or a multireference benchmark with dynamical correlation) and a sensitivity analysis of the QPE runtime would materially strengthen the paper. The missing (10/6) QPE convergence documentation and the (2/2) 10-ancilla inconsistency should be fixed in the same revision. If the authors complete these checks, the paper would be a solid contribution to applied quantum computing for corrosion and materials science and would fit the journal's scope well."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This is a competent and unusually transparent feasibility study. The headline result—that even a six-orbital active space of the oxygen reduction reaction on copper would take about two hours of QPE runtime on a fault-tolerant machine with ~135k physical qubits—holds up under the stated assumptions, and it matches the broader consensus that quantum chemistry advantage is far off. The authors deserve credit for stating their assumptions clearly, for openly admitting the circular use of classical diagonalization to set QPE parameters, and for not overselling the negative result.\n\nWhat is actually new is modest but real: the specific application to O2 reductive dissociation on a Cu slab as a corrosion-relevant process, with VQE and QPE resource estimates for that system. The workflow itself reuses established tools (DFT/NEB, AVAS, embedding Hamiltonians, Azure estimator). The novelty claim of being 'first in this context' is weakened by the same group's earlier platinum ORR paper (ref 22), which covers a closely related reaction.\n\nThe soft spots are real and mostly in the chemistry, not the resource estimation. The stress-test concern is valid: all three active spaces contain exactly one virtual orbital (the sigma* O-O), so FCI can only produce double excitations. The large correlation energy and leading double-excitation coefficient at the reactant and TSdiss geometries may therefore be an artifact of the truncated space, not a robust property of the embedded O2/Cu Hamiltonian. The paper discloses the one-virtual restriction explicitly, but then overinterprets the result as evidence of strong correlation suitable for quantum computing. An expanded active space or a multireference benchmark with dynamical correlation (CASPT2, DMRG) is needed to confirm the wavefunction character. If the character disappears, the quantum-computing motivation loses its chemical basis. Separately, the claim that this step is the ORR bottleneck is labeled 'anticipate' and 'speculate'—fine as a hypothesis, but it should be presented that way throughout.\n\nOn the resource side, the use of exact eigenspectra to choose the scaling factor epsilon and Trotter steps is a circular step, but the authors admit it ('defeats the purpose of using the QPE algorithm'). For a resource estimate, this is actually an optimistic assumption: in practice you would not know the spectrum, so the real runtime could be worse. It does not change the qualitative conclusion.\n\nWho gets value from this paper: industry people deciding whether to invest in quantum for corrosion modeling, and quantum resource-estimation researchers looking for concrete worked examples. It deserves a serious referee; the central negative result is solid, but the chemistry claim needs scrutiny. I would recommend conditional acceptance: require validation of the active-space truncation, relabel the bottleneck claim as speculative, and temper the novelty statement relative to ref 22.","headline":"An honest, well-executed negative resource-estimation study for quantum chemistry on a corrosion-relevant reaction, but the chemical motivation rests on an active-space artifact that needs independent validation.","tokens_in":20439,"tokens_out":2082,"would_cite":false,"duration_ms":23493,"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":"Modeling one corrosion step on a quantum computer needs roughly 135,000 physical qubits and about two hours for a six-orbital active space, while classical exact diagonalization of the same Hamiltonian takes milliseconds.","keywords":["corrosion modeling","oxygen reduction reaction","quantum resource estimation","variational quantum eigensolver","quantum phase estimation","active space embedding","surface code overhead","aerospace aluminum alloys"],"falsifier":"Run DMRG or CASPT2 on the same five geometries with an active space that includes at least one additional virtual orbital beyond the sigma-star orbital; if the ground state becomes single-reference dominant and the correlation-energy difference at the dissociation transition state shrinks to noise, the paper's identification of a strongly correlated target is falsified.","tokens_in":19430,"feed_emoji":"⚛️","tokens_out":7495,"duration_ms":71297,"temperature":0.7,"pith_summary":"This paper tries to establish where quantum computing could and could not help industrial corrosion modeling. It builds a classical workflow—DFT, Hartree-Fock, and full configuration interaction—for the reductive dissociation of oxygen on a copper surface, identifies the geometries whose ground states have strong multiconfigurational character, and then asks what it would cost to solve those small electronic-structure problems with variational quantum eigensolver and quantum phase estimation algorithms. Its resource estimates say the cost is prohibitive: a (10/6) active space requires roughly 135,000 physical qubits and about two hours for QPE, while VQE with error correction exceeds one hour per iteration; exact diagonalization of the same Hamiltonian is done classically in milliseconds. The conclusion a sympathetic reader should take is that quantum computing will not contribute to corrosion simulation without major algorithmic and error-correction breakthroughs, and the paper also fixes a concrete benchmark problem and workflow for testing those advances.","feed_headline":"Small corrosion model needs 135,000 qubits, 2-hour runtime","feed_subtitle":"Even the smallest corrosion-relevant active space runs hours; classical exact diagonalization needs milliseconds.","key_machinery":"The machinery is the embedded active-space Hamiltonian: automated atomic-valence active-space selection picks the oxygen 2p and copper 3d orbitals, Hartree-Fock provides the reference, and the remaining electrons are folded into effective one-body interactions. The qubit form is obtained by Jordan-Wigner mapping, and the two quantum algorithms are UCCSD-based VQE and Trotterized QPE. The resource counts come from surface-code quantum error correction with an assumed physical error rate of $10^{-4}$ and all-to-all connectivity, whose dominant term is the overhead of T factories—the blocks that produce magic states needed for fault-tolerant T gates.","core_discovery":"The central discovery is a negative-resource result embedded in a positive workflow. For the reductive dissociation step of the oxygen reduction reaction on a copper surface, the authors construct embedding Hamiltonians for three active spaces, all containing only one virtual orbital—the sigma-star O-O orbital—so the FCI expansion stops at double excitations. They find that the reactant and transition-state geometries have a strong doubly excited configuration in the ground state, which is what makes the problem a candidate for quantum simulation. They then estimate the cost of solving the same eigenproblem with VQE and QPE under surface-code error correction: the (10/6) active space requires about 135,000 physical qubits, mostly in T factories, and about two hours of QPE runtime, while VQE with quantum error correction exceeds one hour per iteration. Because the same Hamiltonian is diagonalized exactly on a laptop-class calculation in milliseconds, the paper's argued conclusion is that quantum advantage is not in reach for this problem class, and that hardware improvements alone will not close the gap.","pith_inferences":["Beyond the paper's explicit claims, a natural next test is to run a higher-level multireference benchmark such as DMRG or CASPT2 with a larger active space that includes additional virtual orbitals; if the multiconfigurational weight of the ground state falls sharply, the case for quantum simulation of this reaction becomes weaker than the paper's active-space choice implies.","The paper assumes, rather than demonstrates, that reductive dissociation is the rate-limiting step of the oxygen reduction reaction; a microkinetic or experimental comparison of the three electron-transfer steps would determine whether this bottleneck premise holds.","If the resource estimates are correct, the near-term route for this problem class is algorithmic, not hardware-only: the paper's own survey of qubitization, quantum walks, double factorization, and improved error correction suggests those ideas must be brought to bear before the predicted hour-scale runtimes can shrink.","One could test the shot-count model directly by measuring the Hamiltonian coefficient distribution on a real device; the paper's observation that large coefficients become relatively less numerous as the active space grows implies that larger spaces may not suffer from shot noise as severely as naive term counting predicts."],"forward_implications":["If correct, even a small six-orbital corrosion-relevant Hamiltonian is hours away from being solved on a fault-tolerant quantum computer, so realistic models with dozens of active orbitals are far beyond any near-term device.","VQE's dominant cost is the shot count needed to sample Hamiltonian terms to chemical accuracy; with surface-code error correction this pushes per-iteration runtime past one hour for the (10/6) active space.","QPE's runtime is set by Trotter steps and ancilla qubits, and its physical-qubit count is dominated by T factories, reaching about 135,000 qubits for the largest active space tested.","For both algorithms, improving physical qubit error rates reduces qubit count but does not substantially reduce runtime, so algorithm and error-correction design, not just hardware, are the bottleneck.","The workflow supplies a ready-made benchmark: a set of geometries, active spaces, and Hamiltonians whose classical ground states are known, on which future quantum algorithms can be tested."],"supporting_citations":[{"why":"Supplies the AVAS orbital-selection method that defines the (2/2), (6/4), and (10/6) active spaces.","marker":"[31]"},{"why":"Provides the embedding-Hamiltonian construction that folds non-active electrons into effective one-body interactions.","marker":"[32]"},{"why":"Powers the argument that QPE offers at best polynomial advantage here because the overlap of the Hartree-Fock state with the true ground state decays exponentially.","marker":"[35]"},{"why":"Underlies the resource-estimation scaling and the comparison to practical quantum advantage used for the physical-qubit and runtime counts.","marker":"[41]"},{"why":"Defines the VQE algorithm whose per-iteration shot count drives the NISQ runtime estimates.","marker":"[46]"},{"why":"Provides the resource-estimation toolchain that produces the physical qubit and runtime numbers reported for VQE and QPE.","marker":"[52]"}],"fun_headline_variants":["Corrosion quantum run: 135k qubits, 2h; classical: ms","Quantum needs 135k qubits for model classical solves in ms","Even tiny corrosion model demands 135k qubits, 2h runtime","No quantum advantage: 135k qubits for ms-scale classical problem"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The entire case for a quantum-computing target rests on the claim that the reactant and transition-state geometries are genuinely multiconfigurational, which is established only inside active spaces containing a single virtual orbital—so a larger active space or a multireference benchmark could dissolve the motivation.","fun_headline_variants_meta":{"raw":{"variants":["Corrosion quantum run: 135k qubits, 2h; classical: ms","Quantum needs 135k qubits for model classical solves in ms","Even tiny corrosion model demands 135k qubits, 2h runtime","No quantum advantage: 135k qubits for ms-scale classical problem"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00028,"raw_usage":{"total_tokens":1695,"prompt_tokens":1015,"completion_tokens":680,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":631,"completion_tokens_details":{"reasoning_tokens":597}},"tokens_in":631,"tokens_out":680,"duration_ms":23277,"temperature":1.0,"reasoning_tokens":597,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T18:24:01.778648+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run DMRG or CASPT2 on the same five geometries with an active space that includes at least one additional virtual orbital beyond the sigma-star orbital; if the ground state becomes single-reference dominant and the correlation-energy difference at the dissociation transition state shrinks to noise, the paper's identification of a strongly correlated target is falsified.","supporting_citations":[],"review_version":1}