{"id":"e5ba3822-d115-49e1-91c2-855d1d67df4c","arxiv_id":"2604.01983","paper_version":2,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":5.5,"correctness_risk":"low","formal_verification":"none","parameter_count":4,"one_line_summary":"SQD with LUCJ (and a new LCNot-UCCSD variant) on IQM Sirius recovers chemically accurate energies, 1D/2D PES, and DMET-embedded ligand/amantadine results versus FCI/CASCI references.","lead":"Researchers ran large quantum chemistry simulations on IQM’s 24-qubit Sirius processor, recovering ground-state energies and potential-energy surfaces for small molecules plus embedded ligand and amantadine systems to chemical accuracy versus classical FCI/CASCI. The work shows sample-based diagonalization plus embedding is already practical on today’s superconducting hardware.","discovery_kind":"extension","skeptic_critique":{"model":"grok-4.5","headline":"No significant objection identified","rationale":"The manuscript is a large-scale hardware demonstration whose central claim is empirical and carefully scoped. The data (Tables 4, 6, 8, 10; Figures 10, 12, 16–23, 25–30, 35–37) show that, under the non-factor εs=10^8 regime that spans the full symmetry space, SQD(LUCJ) recovers FCI/CASCI energies to numerical precision for the reported systems; under reduced sampling the errors grow in a documented, geometry-dependent way that the authors do not hide. The reader's identified weakest assumption correctly flags the practical boundary of the method, but that boundary is already part of the paper's own narrative (Sections 4.1.2, 4.2.1, 4.3). No internal inconsistency, circular reference, or unsupported extrapolation into classically intractable regimes is present. Consequently the ACCEPT verdict stands; the concrete test above is a straightforward reproducibility check rather than a potential falsifier of the claim.","tokens_in":55931,"tokens_out":545,"duration_ms":6288,"concrete_test":"Independently recompute the SQD energies for the three BeH2 geometries with largest |ESQD−EFCI| under εs=35 (Figure 19d) and the 2D-PES grid points exceeding 1.59 mHa under εs=21 (Figure 28d) using the published bitstring counts and configuration-recovery code; if the recomputed values still lie within chemical accuracy of FCI for the majority of points, the central claim is confirmed.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The reader's weakest assumption (trial-state overlap after noise, especially multi-reference geometries and deep LCNot circuits) is real but already treated as a reported limitation rather than a hidden premise of the central claim. The paper's strongest claim is carefully qualified (\"majority ... to within chemical accuracy for the chosen basis sets\") and is supported by direct numerical comparisons to FCI / DMET-CASCI across fixed geometries, 1D/2D PES grids, and embedded fragments (Sections 4.1–4.4). Failures of LCNot recovery on H2O/NH3, elevated errors for stretched BeH2 under reduced εs, and stochastic ηsub under εs=√|S| are disclosed with quantitative tables and heatmaps. Because the claim does not assert universal recovery or quantum advantage beyond the demonstrated regime, the overlap assumption is not load-bearing for the stated experimental result.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The manuscript reports an extensive experimental campaign of Sample-based Quantum Diagonalization (SQD) on IQM’s Sirius 24-qubit superconducting processor (up to 16 operational qubits). Using the Local Unitary Cluster Jastrow (LUCJ) ansatz (CCSD-initialized) and a newly introduced Linear-CNOT UCCSD (LCNot-UCCSD) ansatz (MP2-initialized), the authors compute ground-state energies for H2, LiH, BeH2, H2O and NH3, perform 1D potential-energy scans for H2, HeH+, LiH and BeH2 (STO-3G and, for the two-electron systems, 6-31G), and map a 32×32 2D PES for H2O in STO-3G. They further couple SQD(LUCJ) to Density Matrix Embedding Theory (DMET) for eight ligand-like molecules and for amantadine, reporting active-space energies that largely agree with FCI or DMET-CASCI to within chemical accuracy for the chosen basis sets. Circuit resources, subspace-coverage metrics (ηpost-cr, ηsub), and the effects of the sampling parameter εs are systematically tabulated and plotted.","tokens_in":56138,"tokens_out":1285,"duration_ms":11621,"significance":"If the numerical claims hold, the work is a substantial experimental contribution to near-term quantum chemistry. It provides one of the most complete hardware demonstrations of sample-based diagonalization to date: multi-molecule benchmarks, two ansätze with explicit resource and accuracy trade-offs, the first experimental 32×32 2D PES of water on a superconducting device, and DMET-SQD energies for a pharmacologically relevant molecule (amantadine) on IQM hardware. Strengths include transparent reporting of failures (LCNot recovery on H2O/NH3), controlled degradation under reduced εs, and direct comparison to independent classical FCI/CASCI/CCSD references with no fitted energy parameters. The data establish a practical baseline for hybrid SQD–embedding workflows on current superconducting platforms.","major_comments":[{"comment":"§4.1.2 and Table 6: LCNot-UCCSD configuration recovery fails entirely for H2O and NH3 (no samples in the symmetry sector S). The abstract and conclusion still present LCNot-UCCSD as a demonstrated alternative within the SQD workflow. The manuscript should state more sharply that LCNot is only validated for H2, LiH and BeH2 on this hardware, and that the depth-induced failure mode is a hard limit of the present demonstration rather than a minor caveat.","section":null},{"comment":"§4.2.1 (BeH2, εs=√|S|) and §4.3 (2D H2O, εs=21): under reduced subsampling, isolated geometries exceed chemical accuracy (≈1.6 mHa), especially where multireference character grows. The abstract’s claim that “the majority” of energies lie within chemical accuracy is supported by the tables, but the paper should quantify the fraction of grid points / runs that fail the threshold (e.g., for the 32×32 surface and for stretched BeH2) so that the qualifier “majority” is numerically precise rather than qualitative.","section":null},{"comment":"§3.3 and §4.4: all DMET impurities are truncated to a 4-HOMO–4-LUMO active space (≤16 qubits). The reported chemical accuracy is therefore relative to DMET-CASCI in the same truncated space, not to full-system FCI or larger active spaces. This is methodologically consistent, but the abstract and §5 should make explicit that the embedding results demonstrate accuracy of the SQD impurity solver inside a fixed, hardware-limited active space, not recovery of the untruncated molecular correlation energy.","section":null}],"minor_comments":[{"comment":"Table 1 is a useful algorithm comparison but is long; consider moving part of it to the SI or tightening the “Key limitations” column for readability.","section":null},{"comment":"Notation: ηsym, ηpost-cr and ηsub are defined in Eq. (28); ensure every figure caption that uses them points back to that definition (some 1D/2D captions omit it).","section":null},{"comment":"§2.3 / Fig. 2: the LCNot-UCCSD single- and double-excitation circuits are shown, but a short statement that no parity-relaxing approximations were used (as claimed later) would help readers who only skim the methods.","section":null},{"comment":"Appendix references (B–D) for geometries, calibration and QPU runtimes are essential for reproducibility; confirm that the Zenodo deposit includes the raw bitstring counts and the exact geometry JSON used for the 2D grid.","section":null},{"comment":"A few typographical inconsistencies appear (e.g., “Nitrosyl” vs “Nitrocyl” in Fig. 35 labels; “post–configuration-recovery” hyphenation). A light copy-edit pass would clean these.","section":null},{"comment":"The claim of “first experimental 2D-PES on superconducting hardware” is carefully hedged in the text; keep that hedging in the abstract so it is not over-read as absolute priority over all platforms.","section":null}],"recommendation":"minor_revision","confidential_remarks":"The experimental scope is genuine and the data appear carefully collected. The main risk for the journal is over-claiming: the abstract’s “chemically accurate and scalable” framing is stronger than the body, which correctly shows accuracy only inside small active spaces / truncated DMET impurities and documents clear failure modes. With the three major points above addressed (scope of LCNot, quantified “majority,” and active-space truncation language), the paper is suitable for acceptance. No integrity or novelty-disclosure concerns."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"This is a large, careful experimental campaign of SQD on IQM Sirius (up to 16 qubits). The pieces that are actually new are the LCNot-UCCSD variant inside SQD, the first dense 32×32 experimental 2D-PES of water on superconducting hardware, and DMET-SQD energies for a ligand set plus amantadine on this platform. Everything else (SQD, LUCJ, DMET embedding) already exists; the value is the systematic head-to-head and the scale of the hardware data.\n\nWhat they do well is documentation and honesty. Energies come from classical exact diagonalization of a hardware-sampled subspace and are checked against independent PySCF FCI/CASCI/CCSD. They report ηsub, ηpost-cr, gate counts, multiple runs, and the regimes where things break: LCNot recovery fails entirely for H2O/NH3, stretched BeH2 degrades under reduced εs, and ηsub becomes stochastic under εs=√|S|. The central claim is carefully qualified (“majority … chemical accuracy for the chosen basis sets”) and the tables support it. No circularity; free parameters (shots, εs, recovery iterations, 4HOMO–4LUMO cut) are explicit.\n\nSoft spots are real but already on the page. Single-reference CCSD/MP2 initialization plus 10k shots is a genuine limit for multi-reference geometries and deep circuits; they do not hide it. Novelty is incremental-to-moderate, not a new algorithmic paradigm. Minimal bases and active-space truncations keep the problems hardware-sized; that is fine for a NISQ demonstration but bounds how far the “scalable” language can stretch. Citation pattern looks normal for this literature.\n\nWho it is for: people building or benchmarking hybrid quantum-chemistry workflows on real devices. Math and data look solid enough for a serious referee. I would accept for peer review; the experimental record is useful even if the conceptual advance is modest. Engage if you care about near-term SQD/DMET practice; skip if you only want new theory.","headline":"Solid experimental hardware paper: dense 2D water PES and DMET-SQD on amantadine on IQM Sirius, with transparent sampling limits and no overclaim of advantage.","tokens_in":56845,"tokens_out":571,"would_cite":true,"duration_ms":7352,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"Sample-based quantum diagonalization on IQM hardware recovers chemical accuracy for molecules up through amantadine, including a full experimental 2D water surface.","keywords":["Sample-based Quantum Diagonalization","LUCJ ansatz","LCNot-UCCSD","Density Matrix Embedding Theory","potential energy surface","IQM Sirius","quantum chemistry","chemical accuracy"],"falsifier":"Repeat the BeH2 dissociation or the water 2-D surface with the same shot budget and LUCJ initialization but force the post-recovery subspace dimension well below the full symmetry space; if energies then systematically exceed chemical accuracy relative to FCI, the claim that the sampled subspace remains faithful collapses.","tokens_in":56819,"feed_emoji":"⚛️","tokens_out":762,"duration_ms":6809,"temperature":0.7,"pith_summary":"This paper reports a large hardware campaign on a 24-qubit superconducting processor (up to 16 operational qubits) that uses sample-based quantum diagonalization (SQD) rather than variational optimization. A trial circuit prepares a state; the device only samples configurations; classical exact diagonalization inside that sampled subspace returns the energy. With the Local Unitary Cluster Jastrow (LUCJ) ansatz the authors recover full-configuration-interaction (FCI) ground-state energies for H2, LiH, BeH2, H2O and NH3 to chemical accuracy, map one-dimensional dissociation curves, and produce the first experimental 32-by-32 two-dimensional potential-energy surface of water on superconducting hardware. They further embed SQD inside density-matrix embedding theory (DMET) to treat eight ligand-like molecules and the drug amantadine, again matching DMET-CASCI references within chemical accuracy. A second, deeper Linear-CNOT UCCSD ansatz is introduced for comparison: it reduces classical pre-processing but fails for larger systems once circuit depth outruns hardware noise. The work therefore argues that sample-based hybrid workflows already deliver chemically useful energies on present-day devices and that embedding extends their reach to pharmacologically relevant molecules.","feed_headline":"Quantum hardware hits chemical accuracy for water and amantadine","feed_subtitle":"Sample-based diagonalization on a 16-qubit processor maps a full 2-D water surface and drug fragments within 1 kcal/mol of exact results","key_machinery":"Sample-based Quantum Diagonalization (SQD) with the LUCJ ansatz: the quantum processor only samples a trial wave-function; classical configuration recovery and exact sparse diagonalization inside the recovered subspace produce a variational energy that is noise-free once the subspace is fixed.","core_discovery":"Across fixed-geometry benchmarks, 1-D and 2-D potential-energy surfaces, and DMET-embedded ligand and amantadine calculations, the majority of SQD(LUCJ) energies obtained on IQM Sirius agree with exact FCI or DMET-CASCI references to within chemical accuracy for the chosen basis sets, establishing that sample-based diagonalization plus classical embedding is already a practical route to chemically accurate molecular energies on current superconducting hardware.","pith_inferences":[],"forward_implications":[],"fun_headline_variants":["SQD on 16-qubit IQM chip matches FCI for water and amantadine","Full 2D water PES reaches chemical accuracy on IQM Sirius","Hybrid SQD-DMET yields chemical accuracy for amantadine and ligands","Sample-based diagonalization hits chemical accuracy across molecular surfaces","IQM hardware maps water surface and drug energies within chemical accuracy"],"cache_read_input_tokens":49280,"weakest_assumption_plain":"That a shallow LUCJ (or deeper LCNot-UCCSD) circuit, started from single-reference amplitudes and sampled with ten thousand shots, still yields a configuration pool whose recovered subspace overlaps the true ground state after hardware noise, especially near multi-reference geometries or when circuit depth becomes large.","fun_headline_variants_meta":{"raw":{"variants":["SQD on 16-qubit IQM chip matches FCI for water and amantadine","Full 2D water PES reaches chemical accuracy on IQM Sirius","Hybrid SQD-DMET yields chemical accuracy for amantadine and ligands","Sample-based diagonalization hits chemical accuracy across molecular surfaces","IQM hardware maps water surface and drug energies within chemical accuracy"]},"model":"grok-4.5","effort":"low","cost_usd":0.00753,"raw_usage":{"total_tokens":1937,"prompt_tokens":975,"num_sources_used":0,"completion_tokens":79,"cost_in_usd_ticks":75300000,"prompt_tokens_details":{"text_tokens":975,"audio_tokens":0,"image_tokens":0,"cached_tokens":128},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":883,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":975,"tokens_out":79,"duration_ms":7984,"temperature":1.0,"reasoning_tokens":883,"cache_read_input_tokens":128,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-15T11:42:10.224912+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Repeat the BeH2 dissociation or the water 2-D surface with the same shot budget and LUCJ initialization but force the post-recovery subspace dimension well below the full symmetry space; if energies then systematically exceed chemical accuracy relative to FCI, the claim that the sampled subspace remains faithful collapses.","supporting_citations":[],"review_version":1}