{"id":"25bff653-5285-45a7-ac93-8d4ed6a98be2","arxiv_id":"2504.18683","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"DUCC effective Hamiltonians combined with ADAPT-VQE recover dynamical correlation energy outside the active space with similar ADAPT iteration counts as bare Hamiltonians, but the gains rely on classical CCSD amplitudes and were not universal.","lead":"Researchers combined two existing quantum chemistry methods, ADAPT-VQE and double unitary coupled cluster downfolding, to compress molecular Hamiltonians so fewer qubits are needed. Benchmark calculations on LiH, H6, and water show the downfolded Hamiltonians recover more correlation energy than the bare active space, with caveats about strongly correlated systems and one unconverged case.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 'without increasing load' claim rests entirely on iteration counts; per-iteration measurement cost and circuit depth are unreported, and H2O exceeds the iteration limit, so the central claim is not established.","rationale":"The reader identified the classical external amplitudes as the weakest assumption. I agree that the method inherits CCSD limitations, and the paper itself demonstrates this for stretched H6 (Section IV.A.1) and even shows non-variational DUCC results in canonical orbitals (SI Section S-I). However, the authors explicitly acknowledge these limitations, and the comparative claim that DUCC improves over the bare active-space Hamiltonian still holds in those benchmarks. The more consequential gap is the resource claim: the paper equates 'load' with ADAPT-VQE iteration count, but the actual quantum load also includes the number of Hamiltonian terms, per-iteration measurement costs, shot counts, and circuit depth. The paper presents none of these, and its own data already contain a counterexample to 'similar convergence' (H2O exceeding 200 iterations). This is an internal support gap in the central claim, not a disagreement with external consensus. The work is valuable and the issue is addressable by additional resource analysis and revised claims, so the conditional verdict stands unchanged.","tokens_in":19987,"tokens_out":6301,"duration_ms":64578,"concrete_test":"For LiH at 1Re (or another converged case), construct bare, A4, and A7 Hamiltonians and count the number of Jordan-Wigner Pauli strings and their Pauli weights for each. Run ADAPT-VQE with the same gradient threshold, operator pool, and optimizer settings; record the number of ADAPT iterations, the number and type of selected operators, ansatz CNOT depth, and the estimated shot count (using a fixed measurement-partitioning scheme) needed to reach chemical accuracy for each Hamiltonian. If the DUCC runs require materially more measurements or deeper circuits per iteration, the 'without increasing load' claim fails; if total resources are comparable or lower, it holds.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The abstract's central claim that DUCC Hamiltonians 'provide increased accuracy without increasing the load on the quantum processor' is supported only by the observation that ADAPT-VQE converges in a similar number of iterations for bare, A4, and A7 Hamiltonians (Section IV.C). Iteration count is not a complete measure of quantum load. Each ADAPT iteration requires VQE parameter optimization and gradient measurements of Eq. (6), both of which involve estimating expectation values of the Hamiltonian. The DUCC effective Hamiltonians are dense transformed integrals, and the A7(34) variant explicitly contains three- and four-body terms (Eqs. 17-18), so the qubit Hamiltonian generally has a different number of Pauli strings and different Pauli weights than the bare active-space Hamiltonian. This changes per-iteration measurement counts, shot requirements, and measurement-circuit costs. The paper reports no Pauli-term counts, no measurement groupings, no shot estimates, and no circuit-depth or CNOT comparisons. Thus the 'without increasing load' statement is unsubstantiated. The claim of 'similar convergence' is also internally contradicted: Section IV.C states that for H2O the algorithm exceeds the 200-iteration limit, so not all simulations converge in similar iteration counts. The central assertion should be scoped to iteration counts or supported by a full resource comparison.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript reports a numerical study that combines double unitary coupled cluster (DUCC) effective Hamiltonians with ADAPT-VQE for quantum chemistry. The authors benchmark two truncations of the DUCC commutator expansion (A4 and A7, plus variants including three- and four-body terms) on LiH, H6, and H2O, using external amplitudes from CCSD, MP2, CCD, and CCSD with T1 set to zero. They compare exact diagonalizations of the effective Hamiltonians against FCI (LiH) or extrapolated ASCI (H6, H2O) reference energies, and they compare ADAPT-VQE iteration counts for bare and downfolded Hamiltonians. The central claim, stated in the abstract, is that DUCC Hamiltonians provide increased accuracy without increasing the load on the quantum processor.","tokens_in":20234,"tokens_out":4159,"duration_ms":41248,"significance":"If the central claim were fully established, the work would be a useful step toward NISQ-era quantum chemistry: active-space qubit counts could recover full-basis accuracy without additional ADAPT iterations. The study's strengths are its systematic exploration of commutator truncations and external-amplitude choices, and its honest reporting of failure cases, including non-variational DUCC energies in canonical orbitals (SI S-I) and the breakdown of CCSD-based amplitudes for stretched H6 (Section IV.A.1). The benchmarks use external FCI and ASCI references, and the numerical tables are internally consistent. However, the quantum-resource part of the central claim is not established by the reported iteration counts, and one of the three molecules (H2O) explicitly violates the abstract's 'similar convergence' statement.","major_comments":[{"comment":"The statement that DUCC Hamiltonians provide increased accuracy 'without increasing the load on the quantum processor' is supported only by ADAPT iteration counts, and iteration count is not a complete measure of quantum load. Each ADAPT iteration requires gradient measurements in Eq. (6) and VQE parameter optimization, and the DUCC Hamiltonians of Eqs. (17)-(18) are generally denser and, in the A7(34) case, contain three- and four-body terms; this changes the Pauli-string count, measurement groupings, shot requirements, and circuit depth. No such resource data are reported. Moreover, Section IV.C states that for H2O the algorithm exceeds the 200-iteration limit, which contradicts the abstract's 'similar convergence' claim. The central claim should be scoped to iteration counts or supported by a full resource comparison.","section":"IV.C, Fig. 2; abstract"},{"comment":"The accuracy of the effective Hamiltonian depends on the classical external amplitudes, and the paper's own data show this dependence is load-bearing: for stretched H6, CCSD becomes non-variational and the A4/A7 errors grow substantially, with the A7 error increasing from 2.20 mHa at 1 Å to 8.25 mHa at 2 Å when CCSD amplitudes are used. The paper acknowledges this in Section IV.A.1, but the abstract and conclusions state a general accuracy improvement without a strong-correlation caveat. The general claim should be restricted to regimes where the single-reference external amplitudes are reliable, with a quantitative statement of the limitation.","section":"IV.A.1, Eq. (19), Table II"},{"comment":"The H6 and H2O reference energies are extrapolated ASCI energies rather than exact energies, and the manuscript reports no uncertainty estimate for these references. For H2O in particular, Table I shows A7 errors of 6.16 mHa at 1Re and 15.24 mHa at 2Re, both above the 1.59 mHa chemical-accuracy threshold, so the claimed accuracy improvement is not uniform across the tested systems. The conclusions should state the accuracy ceiling implied by the external-amplitude approximation and the reference uncertainty rather than presenting the method as uniformly more accurate.","section":"III, Table I"}],"minor_comments":[{"comment":"The text says these expansions are 'truncated to two body operators,' but later sections consider A4(3), A7(3), and A7(34) with three- and four-body terms; please clarify the convention and define the notation before first use.","section":"II.B, Eqs. (17)-(18)"},{"comment":"For H2O at 2Re, please state explicitly whether the ADAPT-VQE calculation eventually converged after exceeding the 200-iteration limit or was terminated at the limit; this is important for interpreting the convergence plot.","section":"IV.C, Fig. 2"},{"comment":"There is a typo in 'Grimsely' in the introduction; it should read 'Grimsley'.","section":"I"},{"comment":"Please provide a data-availability or code-availability statement, since the numerical benchmarks use an in-house ADAPT-VQE implementation and would benefit from reproducibility details.","section":"III"},{"comment":"Use 'mHa' consistently instead of 'mH' in the table headings and text.","section":"Tables I and II"}],"recommendation":"major_revision","confidential_remarks":"The manuscript leans heavily on the authors' own prior DUCC publications (Refs. 101-111), but the central benchmarks are against external references (FCI and ASCI) and the numerical data appear internally consistent; I do not see a novelty or integrity problem. The main issue is that the resource-cost claim in the abstract goes beyond what the paper measures; a revision that either supplies the missing resource comparison or rescopes the claim to iteration counts would make the paper publishable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a genuinely useful numerical benchmark, and the paper is written honestly. The headline claim in the abstract, however, goes a step beyond what the data show, and the authors should tighten it.\n\nWhat's new: DUCC downfolding has been around, and ADAPT-VQE has been around, but this is the first systematic test of feeding DUCC effective Hamiltonians into ADAPT-VQE. The paper adds real data: A4 vs A7 commutator truncations, inclusion of three- and four-body terms, and a comparison of MP2, CCD, and CCSD-derived external amplitudes. It also reports where things fail — stretched H6, non-variational energies in canonical orbitals, and the H2O case that doesn't converge within 200 iterations. That level of honesty is rare and should be acknowledged.\n\nWhere it's soft: the abstract's 'without increasing the load on the quantum processor' is not supported by the reported metrics. The only load proxy is ADAPT-VQE iteration count. But a DUCC effective Hamiltonian is a dense object; the qubit Hamiltonian generally has more Pauli strings and heavier weights than the bare active-space Hamiltonian. That changes per-iteration measurement costs, shot requirements, and circuit depth. None of that is reported. Also, the 'similar convergence' claim is contradicted by the H2O result, which exceeds the iteration limit. So the central claim should either be scoped to iteration counts or backed by a real resource comparison.\n\nThe second soft spot is the reliance on classical CCSD amplitudes to build the effective Hamiltonian. The paper acknowledges this, and it's a real limitation: in strong correlation regimes where CCSD breaks down, DUCC inherits the problem. So the method is best seen as a way to fold dynamical correlation into an active space when a single-reference classical method is still reliable, not as a universal fix.\n\nThe numerical data look internally consistent, and the references (FCI, extrapolated ASCI) are independent. No code or data are provided, which makes independent verification harder, but the numbers are plausible and the honest reporting helps.\n\nBottom line: worth a serious referee. The technical content is sound; the overclaim in the abstract is fixable. I'd recommend accepting after revision, with the resource claim either removed or properly supported, and code/data release encouraged.","headline":"A useful, honest benchmark of DUCC downfolding with ADAPT-VQE, but the abstract's 'no extra quantum load' claim is not supported by the reported iteration counts.","tokens_in":20814,"tokens_out":3595,"would_cite":true,"duration_ms":31018,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P68","81V55"],"pacs":["03.67.Ac","31.15.V-"],"model":"deepseek-v4-flash","headline":"Downfolded Hamiltonians lift VQE accuracy with no extra iterations.","keywords":["ADAPT-VQE","double unitary coupled cluster","Hamiltonian downfolding","effective Hamiltonian","active space","dynamical correlation","variational quantum eigensolver","quantum chemistry"],"falsifier":"Compute DUCC-ADAPT-VQE energies for symmetric stretched H6 at 2 Å using external amplitudes from a multireference method instead of CCSD; if the energy error stays near the CCSD-based value rather than dropping below it, the claim that the classical amplitude source is the limiting factor would be falsified. A separate test: if any molecule shows ADAPT-VQE needing significantly more iterations for a DUCC Hamiltonian than for the bare Hamiltonian, the no-extra-load claim would be falsified.","tokens_in":19750,"feed_emoji":"⚛️","tokens_out":8352,"duration_ms":72764,"temperature":0.7,"pith_summary":"This paper combines two techniques: ADAPT-VQE, an algorithm that grows a quantum trial state operator by operator, and double unitary coupled cluster (DUCC) downfolding, which classically transforms a molecular Hamiltonian so that correlation outside a chosen active space is folded into an effective active-space Hamiltonian. The central claim is that DUCC effective Hamiltonians, when used with ADAPT-VQE, converge to the ground state in essentially the same number of iterations as the bare active-space Hamiltonian, while recovering dynamical correlation energy from outside the active space. If correct, this means a quantum processor can reach accuracy comparable to a full-basis calculation using only an active-space number of qubits, with no additional algorithmic iterations. The paper tests this across LiH, H6, and H2O, examining strong correlation, truncations of the commutator expansion, higher-body terms, and the choice of classical amplitudes used to build the effective Hamiltonian.","feed_headline":"Downfolded Hamiltonians lift VQE accuracy with no extra iterations","feed_subtitle":"Classical downfolding folds external correlation into the active space, so ADAPT-VQE converges just as fast.","key_machinery":"The central object is the DUCC effective Hamiltonian, defined by similarity-transforming the bare Hamiltonian with the external cluster operator $\\hat{\\sigma}_{\\mathrm{ext}}$ and projecting into the active space. The transformation is expanded with the Baker-Campbell-Hausdorff series, truncated at the A4 or A7 levels and typically reduced to one- and two-body operators, with the option to retain three- and four-body terms. The external amplitudes that define $\\hat{\\sigma}_{\\mathrm{ext}}$ are taken from classical coupled-cluster calculations (CCSD), sometimes replaced by MP2, CCD, or CCSD-with-T1-set-to-zero amplitudes. The second key piece is ADAPT-VQE, which grows a parameterized trial state by appending exponentiated excitations selected by the largest gradient of the energy with respect to pool operators; the paper uses a generalized singles and doubles pool and the Jordan-Wigner mapping. This machinery tests whether the transformed integrals alter the convergence of ADAPT-VQE while recovering external correlation.","core_discovery":"The central claim is that DUCC effective Hamiltonians provide increased accuracy without increasing the load on the quantum processor. When the effective Hamiltonian is built with the A4 or A7 truncations of the Baker-Campbell-Hausdorff expansion and external amplitudes from CCSD, exact diagonalization of the effective Hamiltonian in the active space recovers most of the dynamical correlation energy outside that space; for LiH the A7 effective Hamiltonian reaches errors near 0.06 mHa across the dissociation curve. When ADAPT-VQE is applied to the effective Hamiltonian, it converges to the ground state of the effective Hamiltonian in nearly the same number of iterations as it does for the bare active-space Hamiltonian. Adding three- and four-body terms to the effective Hamiltonian does not change this convergence picture. The paper also finds that the accuracy of the effective Hamiltonian is sensitive to the type of external amplitudes: CCSD amplitudes give the best results, while MP2, CCD, and CCSD-with-T1-zero amplitudes are less accurate, and in strongly correlated regimes where CCSD breaks down, the DUCC approximations inherit that breakdown.","pith_inferences":["A natural next test is to replace CCSD external amplitudes with multireference or self-consistently optimized amplitudes; if the DUCC energy then stays accurate where CCSD fails, it would confirm that the classical amplitude source is the bottleneck.","The near-identical iteration counts across bare and downfolded Hamiltonians suggest that the effective Hamiltonian preserves the same easy directions for ansatz growth, so further convergence gains may come from operator-pool design rather than from better integrals.","Because DUCC yields a Hermitian effective Hamiltonian, it should combine cleanly with other resource-reduction tools such as qubit tapering or grouped Pauli measurements; the paper does not test this combination.","The finding that higher-body terms matter most when $\\hat{T}_1$ amplitudes are large suggests a practical rule: systems with a large $\\hat{T}_1$ diagnostic should retain three- and four-body terms in the effective Hamiltonian."],"forward_implications":["For molecules where single-reference coupled-cluster amplitudes are reliable, DUCC-ADAPT-VQE can reach near-full-basis accuracy with an active-space-sized qubit register.","The number of ADAPT-VQE iterations needed for a given target accuracy is nearly unchanged when the bare Hamiltonian is replaced by a DUCC effective Hamiltonian, even when three- and four-body terms are present.","A two-body operator pool remains effective for effective Hamiltonians that contain higher-body operators, so no new operator-pool design is required to use downfolded integrals.","The accuracy of the downfolded Hamiltonian is limited by the quality of the classical amplitudes; where CCSD breaks down, the effective Hamiltonian inherits that breakdown."],"supporting_citations":[{"why":"Introduces ADAPT-VQE, the adaptive ansatz-construction algorithm whose convergence this paper tests.","marker":"[54]"},{"why":"Introduces double unitary coupled cluster theory and the downfolded effective Hamiltonian construction.","marker":"[103]"},{"why":"Provides the A4 and A7 commutator-truncation definitions and labels used for the effective Hamiltonians.","marker":"[110]"},{"why":"Demonstrates DUCC effective Hamiltonians with VQE, the prior work this paper extends to adaptive ansatze.","marker":"[107]"},{"why":"Supplies the CCSD amplitudes used to approximate external cluster amplitudes in Eq. (19).","marker":"[116]"},{"why":"Supplies the MP2 amplitudes tested as an alternative external-amplitude approximation.","marker":"[117]"},{"why":"Generates the commutator expressions that define the truncated Baker-Campbell-Hausdorff expansion.","marker":"[122]"},{"why":"Provides the extrapolated ASCI reference energies used to benchmark H6 and H2O errors.","marker":"[124]"}],"fun_headline_variants":["DUCC downfolding boosts VQE accuracy at same iteration count","Effective Hamiltonians from DUCC enhance ADAPT-VQE without extra qubit load","Folding in correlation: DUCC-Hamiltonians let VQE keep its speed","ADAPT-VQE on downfolded Hamiltonians: same iterations, more accuracy"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The downfolded Hamiltonian is only as accurate as the classical amplitudes (CCSD, MP2, or CCD) used to build it, so when those amplitudes are wrong, as in stretched H6, the effective Hamiltonian inherits the error.","fun_headline_variants_meta":{"raw":{"variants":["DUCC downfolding boosts VQE accuracy at same iteration count","Effective Hamiltonians from DUCC enhance ADAPT-VQE without extra qubit load","Folding in correlation: DUCC-Hamiltonians let VQE keep its speed","ADAPT-VQE on downfolded Hamiltonians: same iterations, more accuracy"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000702,"raw_usage":{"total_tokens":3143,"prompt_tokens":894,"completion_tokens":2249,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":510,"completion_tokens_details":{"reasoning_tokens":2165}},"tokens_in":510,"tokens_out":2249,"duration_ms":15665,"temperature":1.0,"reasoning_tokens":2165,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T10:12:40.525035+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute DUCC-ADAPT-VQE energies for symmetric stretched H6 at 2 Å using external amplitudes from a multireference method instead of CCSD; if the energy error stays near the CCSD-based value rather than dropping below it, the claim that the classical amplitude source is the limiting factor would be falsified. A separate test: if any molecule shows ADAPT-VQE needing significantly more iterations for a DUCC Hamiltonian than for the bare Hamiltonian, the no-extra-load claim would be falsified.","supporting_citations":[{"cited_title":"Properties of coupled-cluster equations originating in excitation sub-algebras","cited_arxiv_id":null,"evidence_quote":"Introduces double unitary coupled cluster theory and the downfolded effective Hamiltonian construction."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the A4 and A7 commutator-truncation definitions and labels used for the effective Hamiltonians."},{"cited_title":"P.; Peng, B.; Kowalski, K","cited_arxiv_id":null,"evidence_quote":"Demonstrates DUCC effective Hamiltonians with VQE, the prior work this paper extends to adaptive ansatze."},{"cited_title":"O.; Mayhall, N","cited_arxiv_id":null,"evidence_quote":"Supplies the CCSD amplitudes used to approximate external cluster amplitudes in Eq. (19)."},{"cited_title":"A.; Chan, G","cited_arxiv_id":null,"evidence_quote":"Supplies the MP2 amplitudes tested as an alternative external-amplitude approximation."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Generates the commutator expressions that define the truncated Baker-Campbell-Hausdorff expansion."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the extrapolated ASCI reference energies used to benchmark H6 and H2O errors."}],"review_version":1}