{"id":"57944291-6f49-4a6a-ae0e-cda10365be97","arxiv_id":"2501.18937","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"DOCI-based trial wave functions in ph-AFQMC match expensive CAS-based trials for single-bond breaking but underperform for strongly correlated systems.","lead":"This paper tests a cheap electrons-paired-only wave function called DOCI as the trial state inside the auxiliary-field quantum Monte Carlo method for molecules. It works well for single bond breaking but fails for strongly correlated molecules like the carbon dimer.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The polymer-additive 'outperforms CCSD(T)' claim rests on agreement with CASSCF-AFQMC, not on an exact reference, leaving the seniority-zero trial-bias assumption untested for HQ/HEMA.","rationale":"The reader's weakest assumption is that the seniority-zero trial has enough overlap with the exact ground state to keep the phaseless bias small. I agree that this is the central risk, and the paper's own H6, C2, and two-bond H2O data show where it fails. My stress-test concern is more specific to the polymer-additive portion of the central claim: the single-bond 'outperforms CCSD(T)' conclusion for HQ and HEMA is inferred from agreement with CASSCF-AFQMC rather than from an exact benchmark. Since CASSCF-AFQMC is itself approximate, this agreement does not by itself establish that the seniority-zero bias is small there. The proposed FCI-in-active-space trial comparison is a direct and feasible check: it isolates the trial's effect on the phaseless bias under identical AFQMC conditions and would show whether the residual bias is negligible for the claimed systems. The reader's CONDITIONAL verdict is appropriate; the paper should either provide this exact-reference check or clearly weaken the polymer-additive outperformance claim. I do not see a reason to move the verdict further because the limitation is stated honestly and the H2O/FCI evidence supports at least the narrower methodological conclusion.","tokens_in":15553,"tokens_out":16903,"duration_ms":175907,"concrete_test":"At the equilibrium and stretched geometries used in Fig. 9, diagonalize FCI in the (8e,8o) active space for HQ and in the (2e,2o) active space for HEMA, record the squared overlap |<Psi_FCI|Psi_OO-DOCI>|^2, and rerun ph-AFQMC with the FCI-in-active-space trial using the same AFQMC parameters. If the OO-DOCI-AFQMC energies shift by more than about 1 kcal/mol relative to the FCI-trial AFQMC energies while the CCSD(T) relative error remains larger, then the seniority-zero trial bias is the source of the claimed outperformance; if the two AFQMC energies agree, the trial-bias assumption is confirmed for these single-bond systems.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central accuracy claim for OO-DOCI-AFQMC on single O-H cleavage is load-bearing on the assumption that the seniority-zero trial leaves a small phaseless bias. This is demonstrated for H2O by comparison with FCI, but for HQ and HEMA the evidence is only that OO-DOCI-AFQMC matches CASSCF-AFQMC. That is not an exact reference: ph-AFQMC with a CASSCF trial is itself approximate, and the paper's own strongly correlated results show the seniority-zero trial bias can be large. For the H6 ring (Fig. 4), C2 (Fig. 7), and two-bond H2O dissociation (Fig. 6), the bias exceeds chemical accuracy and in some regions reaches roughly 0.1 Eh as static correlation strengthens. For HQ, the bare OO-DOCI trial energy already differs from CASSCF by about 6.5 kcal/mol in the (8e,8o) active space, and whether the AFQMC projection removes that difference for HQ/HEMA is not checked against FCI or DMRG. Therefore the claim that OO-DOCI-AFQMC outperforms CCSD(T) for the polymer additives is underdetermined by the benchmarks reported.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper proposes using doubly occupied configuration interaction (DOCI) and orbital-optimized DOCI (OO-DOCI) wave functions as trial wave functions in phaseless auxiliary-field quantum Monte Carlo (ph-AFQMC). The motivation is to reduce the exponential cost of complete active space (CAS) trial wave functions while still capturing static correlation, with AFQMC supplying dynamical correlation. The method is tested on linear H4, H6 ring, two-bond water dissociation, the carbon dimer, and single O-H bond breaking in water, hydroquinone (HQ), and 2-hydroxyethyl methacrylate (HEMA). Comparisons are made to CASSCF-AFQMC, CASCI-AFQMC, CCSD, CCSD(T), FCI (for water), and DMRG (for C2). The results show that OO-DOCI-AFQMC closely matches CASSCF-AFQMC and FCI for single O-H bond breaking in water, but deviates significantly in strongly correlated regimes such as the H6 ring dissociation limit, two-bond water dissociation at large r, and the carbon dimer. The authors conclude that seniority-zero trial wave functions are insufficient for strong correlation and suggest extending the seniority space.","tokens_in":15679,"tokens_out":8857,"duration_ms":74837,"significance":"The paper's main positive result, that OO-DOCI-AFQMC reproduces FCI for single O-H bond breaking in water, is a useful demonstration that a seniority-zero trial can work in a regime of moderate static correlation. The honest reporting of failure cases (H6 ring, two-bond water, C2) is a strength and provides clear boundaries for the method's applicability. The use of open-source tools (PySCF, ipie, DOCI module) and detailed computational parameters supports reproducibility. However, the broad promise in the abstract and conclusions, that the method 'offers a path to accurate multi-reference calculations for larger, more complex systems,' is not supported by the data, particularly because the only large-active-space example (C2) shows large errors. The polymer-additive claims also lack an exact reference. The paper is a valid contribution if the claims are recalibrated.","major_comments":[{"comment":"The claim that OO-DOCI-AFQMC 'outperforms coupled-cluster singles, doubles, and perturbative triples' for HQ and HEMA is underdetermined. The only reference for these molecules is CASSCF-AFQMC, which is itself an approximate ph-AFQMC calculation with a different trial; no FCI, DMRG, or other near-exact result is provided. The paper's own data in Figs. 4, 6, and 7 show that seniority-zero trial bias can reach roughly 0.1 Eh in strongly correlated regimes, so agreement with CASSCF-AFQMC does not by itself establish accuracy. To support the claim, the authors should provide an independent reference for HQ/HEMA or weaken the claim to agreement with CASSCF-AFQMC.","section":"Sec. IV.D, Figs. 9-10"},{"comment":"The only large-active-space calculation, OO-DOCI(12e,28o)-AFQMC for C2 (Fig. 7), shows significant deviations from DMRG, and its energy profile closely mirrors the (8e,8o) result. This directly contradicts the abstract's and conclusion's statement that the method 'offers a path to accurate multi-reference calculations for larger, more complex systems.' The cost reduction is real, but accuracy at larger active spaces is not demonstrated. The conclusions should be restricted to single-bond dissociation with modest active spaces, or rephrased as an open question.","section":"Sec. IV.C, Fig. 7; Sec. V"},{"comment":"The symmetry content of the AFQMC calculations for C2 is unspecified. Fig. 7 compares AFQMC energies to DMRG curves for three states (X 1Σ+g, B 1Δg, B′ 1Σ+g), but the text does not state which symmetry or spin each AFQMC curve corresponds to, nor whether the calculations are restricted to a particular irreducible representation. The discussion of crossings and avoided crossings at r≈3.0 Bohr implies the AFQMC may be following the lowest state irrespective of symmetry, but this is not stated. Without this information, the deviations from DMRG and the discontinuities cannot be interpreted.","section":"Sec. IV.C, Fig. 7"},{"comment":"The central assumption that the seniority-zero trial keeps the phaseless bias small is not quantified. The paper reports no overlap measure (e.g., ⟨ΨT|Ψ0⟩ or local energy variance) for any system, so the reader cannot assess a priori when the method is reliable. Given that the method fails exactly where static correlation is strongest (Figs. 4, 6, 7), a diagnostic based on the DOCI wave function (e.g., weight of the leading configuration or the DOCI–CASSCF energy gap) would strengthen the claim that DOCI captures static correlation in the successful cases.","section":"Sec. II.A, Eq. (11); Sec. IV"}],"minor_comments":[{"comment":"The notation Δθ_i for the phase of S_i is inconsistent with the definition θ_i = arg(S_i); please rename to avoid implying a difference from the previous time step.","section":"Eq. (11)"},{"comment":"The units in 'Δτ = 0.005 E−1 h' should be formatted as E_h^{-1}.","section":"Sec. III"},{"comment":"The geometries of HQ and HEMA are not provided; please include Cartesian coordinates or a reference to allow reproduction.","section":"Sec. IV.D"},{"comment":"The AFQMC curves are not distinguished by symmetry in the legend; consider using different line styles or clarifying in the caption which state is computed.","section":"Fig. 7"},{"comment":"The phrase 'reduces the configuration space to the square root' is imprecise; the number of configurations is reduced to the square root of the CAS count, not the space itself.","section":"Sec. II.B"},{"comment":"'The molecular geometries used in the Gaussian and PySCF calculations differed only by an amount on the order of round-off error' should be quantified or replaced with a statement that identical geometries were used.","section":"Sec. IV.D"}],"recommendation":"major_revision","confidential_remarks":"The paper is transparent about limitations, but the abstract and conclusions overstate the method's promise. The polymer-additive benchmark lacks an exact reference, and the C2 large-active-space result undermines the 'path to accurate multi-reference calculations' statement. The authors should be asked to recalibrate these claims. The FCI benchmark for H2O single-bond dissociation is the strongest part of the paper."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a clean numerical benchmark of DOCI and OO-DOCI trial wave functions in ph-AFQMC. The combination is new as far as the cited literature goes, and the paper is honest about where it fails. The central positive claim—that OO-DOCI-AFQMC matches CASSCF-AFQMC for single O–H bond breaking and fixes the CCSD(T) dissociation error—is supported for H2O by comparison with FCI, but for HQ/HEMA only by agreement with CASSCF-AFQMC. That last point is the main soft spot.\n\nWhat the paper does well: it tests the trial wave function across a sensible range of correlation regimes, from single-bond breaking to strongly correlated H6, C2, and two-bond water dissociation. The failures are reported as data, not explained away. The seniority-zero limitation is stated in the abstract and conclusions. Using DOCI reduces the number of configurations from (n choose k)^2 to (n choose k), which is a real practical advantage for active spaces beyond CAS limits; the C2 (12e,28o) run demonstrates this. The paper is clearly written and the ph-AFQMC formalism is standard.\n\nThe soft spots are real but not disqualifying. First, the claim that OO-DOCI-AFQMC 'outperforms' CCSD(T) for the polymer additives is underdetermined. The reference is CASSCF-AFQMC, not FCI or DMRG. For H2O, where the active space is (2e,2o), DOCI is essentially equivalent to CASSCF, so that check doesn't validate the seniority-zero approximation for larger active spaces. For HQ, the bare OO-DOCI trial already differs from CASSCF by about 6.5 kcal/mol, and the paper does not show that the AFQMC projection removes that difference. The stress-test note is right: without an exact reference, the seniority-zero trial bias remains untested for HQ/HEMA. Second, the paper does not report input coordinates or raw energies, and error bars appear only in some figures. That is a reproducibility shortfall for a numerical benchmark. Third, the abstract's promise of 'a path to accurate multi-reference calculations' overreaches given the demonstrated failures in strongly correlated regimes—though the authors immediately qualify it.\n\nThe math is standard; there are no new formal results, and the paper does not claim any. The citation pattern looks fine; the relevant AFQMC trial-function literature is covered. I would trust the H2O FCI comparison and the general direction of the conclusions. For a referee, the main requests should be: release input geometries, total energies, and error bars; provide an exact or DMRG reference for at least one single-bond-breaking case with an active space larger than (2e,2o); and soften the abstract's broad language. I'd send it to peer review.","headline":"Solid, honest benchmark of DOCI/OO-DOCI trials in ph-AFQMC; the single-bond claims mostly hold, but the polymer-additive 'outperforms CCSD(T)' claim needs an exact reference.","tokens_in":16335,"tokens_out":2098,"would_cite":true,"duration_ms":18546,"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":"Doubly occupied configuration interaction wave functions can serve as trial states for phaseless auxiliary-field quantum Monte Carlo, matching CAS-based accuracy for single-bond breaking at lower cost, while failing for strongly…","keywords":["auxiliary-field quantum Monte Carlo","seniority-zero trial wave function","doubly occupied configuration interaction","orbital-optimized DOCI","static correlation","dynamical correlation","bond dissociation","polymer additives"],"falsifier":"Compute the squared overlap between the OO-DOCI trial and a near-exact (FCI or DMRG) ground state along the H6 ring and C2 dissociation curves: the paper's account predicts this overlap stays high where OO-DOCI-AFQMC matches CASSCF-AFQMC and drops sharply where its energy error grows, so a low overlap in the failing regions would confirm the diagnosis. Alternatively, rerun the failing points with a trial that adds seniority-2 configurations and check whether the energy error disappears.","tokens_in":15239,"feed_emoji":"⚛️","tokens_out":7479,"duration_ms":63159,"temperature":0.7,"pith_summary":"The paper proposes using the doubly occupied configuration interaction (DOCI) wave function—a seniority-zero ansatz that keeps only empty or doubly occupied orbitals—as the trial wave function in phaseless auxiliary-field quantum Monte Carlo (ph-AFQMC). The idea is to let DOCI or its orbital-optimized variant (OO-DOCI) capture static correlation cheaply, and let the QMC imaginary-time propagation add dynamical correlation across all orbitals. The paper shows that for single O–H bond breaking in water and two polymer additives, OO-DOCI-AFQMC closely matches CAS-based ph-AFQMC and even beats CCSD(T), while cutting the trial-state cost from a squared binomial to a single binomial. It also reports that for strongly correlated systems—the carbon dimer, multi-bond breaking in hydrogen systems and water—seniority-zero trials lose quantitative accuracy, so configurations with unpaired electrons are needed in the trial. The net claim is a trade: DOCI trials make ph-AFQMC a cheaper multi-reference method for single-bond and weakly multi-reference problems, with a clearly identified failure mode.","feed_headline":"Pair-only trial states bring AFQMC to CAS accuracy","feed_subtitle":"A cheap pair-only trial state gives quantum Monte Carlo the static correlation it needs, matching much costlier CAS benchmarks on single…","key_machinery":"The central object is the seniority-zero trial wave function: a configuration-interaction wave function built only from electron-pair configurations, in which every spatial orbital is empty or doubly occupied and the number of unpaired electrons (the seniority) is zero. DOCI supplies this trial state, and orbital-optimized OO-DOCI rotates the orbitals to make it as good as possible. Inside ph-AFQMC the trial enters the phaseless weight update $w_i \\propto w_i |S_i| \\max(0,\\cos\\Delta\\theta_i)$ through the overlap ratio $S_i = \\langle\\Psi_T|\\Phi_i(\\tau+\\Delta\\tau)\\rangle/\\langle\\Psi_T|\\Phi_i(\\tau)\\rangle$, so it controls the sign and phase bias of the random walk. Its practical role is to supply static correlation cheaply, with $N_{DO} = \\binom{n}{k}$ configurations instead of $\\binom{n}{k}^2$ for a complete active space, while the imaginary-time propagation adds dynamical correlation.","core_discovery":"On its own terms, the paper's central claim is that the seniority-zero DOCI wave function is a viable trial state for ph-AFQMC: it supplies enough static correlation to keep the phaseless bias small in mildly multi-reference regimes, and the QMC step supplies the dynamical correlation missing from DOCI itself. The evidence is the set of potential-energy curves: OO-DOCI-AFQMC reproduces CASSCF-AFQMC for single O–H dissociation in water and polymer additives and gives dissociation energies in close agreement with FCI for H2O, while DOCI-AFQMC improves on bare DOCI everywhere. The paper also claims a cost advantage: DOCI's configuration count is binomial in the active space instead of the squared binomial of a complete active space, demonstrated by an OO-DOCI(12e,28o) trial for C2 that CAS cannot reach. The companion claim, made explicitly, is that this strategy is not enough for strongly correlated regimes: errors grow in the fully dissociated H4/H6/H2O curves and in C2, and the paper attributes the failure to the missing seniority-2 (unpaired-electron) configurations in the trial.","pith_inferences":["Editorial inference: extending the trial space to include seniority-2 configurations (unpaired-electron excitations) should restore quantitative accuracy in C2 and multi-bond dissociation, at a cost still below full CAS; the paper's seniority-zero diagnosis points directly to this as the next test.","Editorial inference: because DOCI and pair coupled cluster doubles give nearly identical correlation energies and the latter scales more cheaply, replacing DOCI with a pCCD-style trial inside ph-AFQMC could push the same strategy to much larger active spaces than the paper demonstrates.","Editorial inference: the close agreement between OO-DOCI-AFQMC and CASSCF-AFQMC for single O–H bonds suggests that ph-AFQMC's dynamical correlation correction is robust to modest differences in trial-state detail, implying the practical bottleneck is static-correlation content rather than orbital choice."],"forward_implications":["For single O–H bond breaking, OO-DOCI-AFQMC reaches CASSCF-AFQMC-level accuracy, so it can serve as a cheaper multi-reference benchmark where CAS trials are too expensive.","The trial-state configuration count drops from $\\binom{n}{k}^2$ to $\\binom{n}{k}$, letting ph-AFQMC use multi-reference trial states in active spaces beyond the usual CAS limit, as demonstrated with the (12e,28o) trial on C2.","In strongly correlated regimes—H4/H6 dissociation, two-bond H2O breaking, and C2—seniority-zero trials produce biased ph-AFQMC energies, so quantitative accuracy there requires extending the trial space beyond seniority zero.","CCSD(T)'s O–H dissociation curves for hydroquinone and HEMA degrade as the T1 diagnostic exceeds 0.02, while OO-DOCI-AFQMC stays close to the CASSCF-AFQMC reference, reinforcing the need for multi-reference trials in industrially relevant molecules."],"supporting_citations":[{"why":"Supplies the phaseless auxiliary-field quantum Monte Carlo method whose trial-state dependence the paper exploits.","marker":"[17]"},{"why":"Establishes ph-AFQMC accuracy benchmarks with CASSCF trial states that the DOCI-based trials are compared against.","marker":"[18]"},{"why":"Provides the ipie software implementation used for all ph-AFQMC calculations in the paper.","marker":"[27]"},{"why":"Defines the DOCI / generalized molecular orbital framework used to build the seniority-zero trial states.","marker":"[44]"},{"why":"Introduces the seniority hierarchy that justifies restricting the trial space to seniority zero.","marker":"[46]"},{"why":"Supplies DMRG reference potential-energy curves for the carbon dimer used to judge AFQMC accuracy.","marker":"[67]"},{"why":"Provides the T1 diagnostic used to show that CCSD(T) becomes unreliable in the O–H dissociation region.","marker":"[70]"}],"fun_headline_variants":["Pair-only trial states: CAS accuracy without CAS cost","DOCI trial wavefunction lifts AFQMC to CAS-level accuracy","Pair-only AFQMC trial states: match CAS at lower cost","Pair-only trial states: a cheap shortcut to CAS accuracy","Pair-only trial wavefunction: AFQMC on par with CAS for mild correlation"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the seniority-zero trial wave function has enough overlap with the true ground state to keep the phaseless AFQMC bias small over the entire bond-breaking curve; the paper's own results show this premise breaks down in exactly the most strongly correlated regimes.","fun_headline_variants_meta":{"raw":{"variants":["Pair-only trial states: CAS accuracy without CAS cost","DOCI trial wavefunction lifts AFQMC to CAS-level accuracy","Pair-only AFQMC trial states: match CAS at lower cost","Pair-only trial states: a cheap shortcut to CAS accuracy","Pair-only trial wavefunction: AFQMC on par with CAS for mild correlation"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000925,"raw_usage":{"total_tokens":4017,"prompt_tokens":1049,"completion_tokens":2968,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":665,"completion_tokens_details":{"reasoning_tokens":2879}},"tokens_in":665,"tokens_out":2968,"duration_ms":59424,"temperature":1.0,"reasoning_tokens":2879,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-09T21:54:16.953497+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the squared overlap between the OO-DOCI trial and a near-exact (FCI or DMRG) ground state along the H6 ring and C2 dissociation curves: the paper's account predicts this overlap stays high where OO-DOCI-AFQMC matches CASSCF-AFQMC and drops sharply where its energy error grows, so a low overlap in the failing regions would confirm the diagnosis. Alternatively, rerun the failing points with a trial that adds seniority-2 configurations and check whether the energy error disappears.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the T1 diagnostic used to show that CCSD(T) becomes unreliable in the O–H dissociation region."},{"cited_title":"Zhang and H","cited_arxiv_id":null,"evidence_quote":"Supplies the phaseless auxiliary-field quantum Monte Carlo method whose trial-state dependence the paper exploits."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes ph-AFQMC accuracy benchmarks with CASSCF trial states that the DOCI-based trials are compared against."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the ipie software implementation used for all ph-AFQMC calculations in the paper."},{"cited_title":"Couty and M","cited_arxiv_id":null,"evidence_quote":"Defines the DOCI / generalized molecular orbital framework used to build the seniority-zero trial states."},{"cited_title":"Bytautas, T","cited_arxiv_id":null,"evidence_quote":"Introduces the seniority hierarchy that justifies restricting the trial space to seniority zero."},{"cited_title":"Wouters, W","cited_arxiv_id":null,"evidence_quote":"Supplies DMRG reference potential-energy curves for the carbon dimer used to judge AFQMC accuracy."}],"review_version":1}