Pith. sign in

REVIEW 3 major objections 5 minor 108 references

Machine learning for sample-based quantum diagonalization: generative configuration recovery and the classical-simulability frontier

T0 review · 3 major / 5 minor · reviewed 2026-08-08 · deepseek-v4-flash

Pith's one-line read Sample-based quantum diagonalization has no demonstrated quantum advantage over classical selected configuration interaction, on the published record as of mid-2026.

desk verdict A careful, honest critical review of SQD/QSCI whose negative verdict is solid on the classical-baseline critique but leans on an unverified dequantization preprint for its sharpest claim. read the letter →

arxiv 2608.05314 v1 pith:CZPTSUPG submitted 2026-08-05 quant-ph physics.chem-ph

classification quant-phphysics.chem-ph
keywords sample-basedquantumdiagonalizationquantum-selectedconfigurationinteractionselectedrecoverygenerativeflownetworksadvantagedequantizationmachinelearningforchemistry
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Sample-based quantum diagonalization (SQD, also quantum-selected configuration interaction or QSCI) lets a quantum processor propose electronic configurations, which a classical computer then uses to diagonalize the Hamiltonian in the sampled space; the paper argues that the field's central question is whether that quantum-selected subspace can beat classical selected configuration interaction. The paper's answer, carefully scoped to the published record as of mid-2026, is no: in every reproducible same-active-space comparison, strong classical selected-CI methods match or beat the quantum-sampled subspace, and the single-layer LUCJ circuits used in the flagship demonstrations now admit polynomial-time classical energy estimation. It offers a structured taxonomy of machine-learning proposers, identifies a conspicuous gap (a reward-proportional generative-flow-network proposer for tail discovery has not been built), and proposes a ten-point benchmarking standard. The paper's own FCI-exact experiments confirm that the cheap Epstein-Nesbet prior's rank correlation with the exact weights degrades with multireference character, but refute the idea that the one observed generative advantage (robustness to shot starvation under noise) is multireference-specific: it is generic. A sympathetic reader is meant to take away a map: quantum or generative advantage can survive only where the classical signal that makes an SQD subspace verifiable also fails to construct it cheaply, or where the classical lower bound is an unconditional theorem.

What carries the argument

At the centre is the SQD/QSCI loop: a shallow LUCJ circuit proposes bitstrings, a self-consistent configuration-recovery step repairs noise-corrupted samples, and the Hamiltonian is diagonalized exactly in the recovered determinant subspace, giving a variational upper bound. The load-bearing identity is the coupon-collector law for the marginal weights $w_\alpha(i)=\sum_\beta |c_{i\beta}|^2$, which makes discovering the heavy-tailed correlation-energy tail exponentially expensive in shots; the cheap classical signal is the Epstein-Nesbet first-order coefficient $c^{(1)}_i=\langle D_i|\hat H|D_{\rm HF}\rangle/(E_{\rm HF}-\langle D_i|\hat H|D_i\rangle)$, whose squared value is used as a reward. The paper's proposed method, left as a gap for others to fill, is a generative flow network trained by trajectory balance so that terminal determinants are sampled proportionally to reward, with particle number enforced by construction. The verdict itself is carried by the imported dequantization of single-layer LUCJ circuits, which treats the layer as a free-fermionic orbital rotation dressed by a low-rank diagonal-Coulomb factor, making the energy classically computable.

What would settle it

Reproduce the flagship 77-qubit [4Fe-4S] energy with an independent implementation of the single-layer LUCJ classical-simulation algorithm; if the algorithm fails to match or beat the published hardware energy, or if a multi-layer extension is shown not to admit the same Pfaffian-style evaluation, the central negative claim loses its strongest pillar. Independently, a single reproducible same-active-space benchmark in which the raw quantum-sampled subspace at matched dimension and shot budget beats iterative heat-bath CI would refute the overall verdict.

Watch

Extended reading notes

Core claim

The paper's central claim is a scoped negative: across the published same-active-space comparisons, no reproducible demonstration shows that the SQD/QSCI quantum sampler beats SHCI, CIPSI, DMRG, or AFQMC on molecular electronic structure, and the flagship sampling circuits have been classically reproduced at the energy (weak-simulation) level. The mechanism behind the verdict is that the same projected-Hamiltonian structure that makes an SQD subspace classically verifiable also makes a cheap classical importance heuristic available for constructing it; the single-layer LUCJ dequantization is the extreme case, where the sampler itself is classically simulable. The paper distinguishes scale demonstrations (77 qubits, 12,635-atom embeddings) from advantage claims and argues the burden of proof sits on the quantum side. Its own controlled experiments add two results: the Spearman rank correlation between the Epstein-Nesbet prior and exact FCI weights falls monotonically with multireference character (from 0.72 to 0.60 along stretched N2), and the one generative advantage found under device noise, valid-by-construction sampling when valid shots are starved, is generic rather than a multireference-specific or quantum edge. The paper leaves two genuine openings: deep (multi-layer) LUCJ circuits are not covered by the dequantization, and learning from quantum experiments has an unconditional classical sample-complexity lower bound that cannot be dequantized, though its bridge to chemistry is unbuilt.

Load-bearing premise

The strongest part of the negative verdict, that the flagship single-layer LUCJ circuits can be simulated classically in polynomial time, is imported from a 2026 preprint and is not derived or independently verified in this paper.

Editorial extensions

If this is right

  • Any future claim that SQD or an ML-augmented proposer beats classical selected CI will need to clear the paper's ten-point bar: exact-FCI ground truth, strong classical baselines, a dequantization baseline, shot-based cost accounting, multi-seed error bars, calibrated asymmetric noise, active-space and orbital convergence, energy differences, multiple geometries, and pre-registered resource accoun
  • If the dequantization holds, the 77-qubit [4Fe-4S] and related hardware results are demonstrations of scale and noise tolerance, not evidence of quantum advantage.
  • A learned proposer trained on a cheap classical reward inherits that reward's ceiling; the paper's own five-seed N2 experiment shows a GFlowNet on the Epstein-Nesbet reward (192 ± 19 mHa at D=120) does not beat deterministic greedy top-K (41.3 mHa).
  • The only demonstrated generative advantage, robustness to valid-shot starvation under noise, is a generic constraint-satisfaction effect available to any symmetry-valid classical sampler, so it is not a quantum or chemistry advantage.
  • The two remaining openings are the depth corridor (deep multi-layer LUCJ not covered by the single-layer dequantization) and the unconditional advantage in learning from experiments, which the paper flags as the highest-value bridge to chemistry.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the regime-map logic is right, the rank correlation $\rho$ between a cheap prior and exact FCI weights is a cheap pre-screening coordinate: compute it for a proposed system before running hardware, and expect learned or quantum proposers to help only where it degrades.
  • A testable extension the paper leaves open: train a GFlowNet proposer with a reward that fuses the Epstein-Nesbet prior with quantum-sample frequencies, and compare against the same reward used greedily; the paper's data suggest proportional sampling will not overtake top-K unless the fused reward is genuinely better.
  • The single-layer dequantization points toward an isomorphic dichotomy to the barren-plateau/simulability tension from variational quantum eigensolvers: shallow, structured, noise-tolerant circuits tend to be classically simulable, while the depth that would defeat simulation may also overwhelm recovery. Whether a usable corridor exists is unresolved.
  • For learning from experiments, a concrete bridge would use quantum-memory learning to characterize an effective device error model and feed that model into the SQD recovery step; the paper mentions this direction but does not test it.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. This manuscript is a critical review of sample-based quantum diagonalization (SQD/QSCI) and its machine-learning augmentation, organized around a taxonomy of generative and learned configuration selectors, a proposed benchmarking standard, and an original set of FCI-verifiable experiments. The paper identifies a GFlowNet gap for determinant selection, argues that no reproducible same-active-space demonstration currently shows the quantum sampler beating classical selected CI, and reports that the flagship single-layer LUCJ circuits admit polynomial-time classical energy estimation, attributing this dequantization to an external 2026 preprint. The author's own experiments show that a cheap Epstein–Nesbet prior correlates only partially with exact FCI weights and degrades monotonically with multireference character, that a GFlowNet trained on this cheap reward beats blind sampling but not greedy top-K selection, and that the generative advantage under simulated device noise is a generic shot-starvation effect rather than a multireference-specific one. The manuscript ships reproducible code for all original figures.

Significance. If the verdict holds, the paper provides a valuable service to a fast-moving field: it consolidates the critique literature, states a concrete falsifiable benchmark standard, and supplies original exact-FCI experiments with multi-seed statistics and open code. The self-correction of the single-seed GFlowNet result under the five-seed standard is an exemplary practice. The central negative conclusion, however, rests in its strongest form on an externally imported dequantization claim that is not independently verified in the manuscript, which makes the significance of the flagship claim conditional on the correctness and applicability of that external theorem. The taxonomy and the regime map remain useful even if that external pillar weakens, because the broader no-advantage verdict is also supported by the compactness and random-recovery literature quoted in §5.2–§5.3.

major comments (3)
  1. [§5.3, §5.5, §6(iii)] The strongest concrete form of the negative verdict—that the flagship 77-qubit [4Fe-4S] sampling circuits have been classically reproduced at the energy level—is imported entirely from the external preprint [26] and is not re-derived, implemented, or independently checked in this manuscript. Section 5.3 calls this dequantization 'decisive,' and §5.5 states it as a fact, while §6(iii) turns it into a mandatory baseline. If the theorem in [26] is wrong, or if the actual hardware circuits contain an additional entangling layer, orbital-rotation schedule, or recovery-dependent modification that violates its single-layer LUCJ assumptions, then this particular pillar of the verdict collapses. The broader negative conclusion could still survive on the other cited literature (e.g., [14], [24], [25]), but the manuscript should either reproduce the dequantization on the specific published circuits or explicitly re-label the flagship claim as conditional on an unverified external result, separating it from the independently supported parts of the verdict.
  2. [§4.6, Fig. 6] The GFlowNet experiment, while honestly executed, evaluates only the cheap static Epstein–Nesbet reward and does not test the full proposal described in §4.6, which includes a fused reward combining the EN prior with recovered sample statistics and a learnable reward temperature. The conclusion that the GFlowNet 'inherits the ceiling of the signal it is trained on' is appropriately scoped to this static-reward setting, but the manuscript should state more explicitly that the open question for the proposed GFlowNet is therefore not addressed by its own Fig. 6, only by the noise and multireference experiments of §7. As written, a reader could mistake Fig. 6 for a test of the proposed method rather than a test of its cheap-reward component.
  3. [§7, Fig. 10] The claim that the noise crossover is 'universal' is supported only for a single molecule (N2) at a fixed subspace dimension D=120, with five seeds per geometry. The paper correctly notes the sign-test floor of p=0.0625 and that per-geometry tests certify seed reproducibility rather than generalization across chemical space, but the word 'universal' appears in the figure caption and in the abstract's framing of a 'controlled single-molecule noise sweep.' This is a wording issue rather than a statistical one, yet it matters because the conclusion 'available in principle across chemistry' is an extrapolation beyond the data. I recommend replacing 'universal' with 'consistent across the N2 dissociation curve' and making the single-molecule scope prominent in the abstract.
minor comments (5)
  1. [Abstract and §1] The abstract contains the typo 'whichconfigurations' and the full text uses inconsistent em-dash spacing; these should be cleaned up.
  2. [§4.6, Eq. (5)] The trajectory-balance objective uses the notation P_F(a_t|s_{t-1};θ) and P^⊤_θ(x) without fully defining the forward policy's action space or the relationship between trajectories and terminal determinants; a sentence defining these objects would improve readability.
  3. [§5.2] The text says 'Two peer-reviewed-track critiques frame the negative case,' but reference [24] is a 2026 preprint; the wording should distinguish peer-reviewed works from preprints consistently, as the paper itself does elsewhere.
  4. [§7, Fig. 9 caption] The caption calls ρ an 'order parameter,' which is reasonable, but the text correctly notes that ρ is tie-laden because 69% of α-strings have zero Epstein–Nesbet reward; the caption should carry that caveat as well.
  5. [§6(v)] The sign-test floor of p=0.0625 at n=5 is correctly reported, but the text says 'a significance statement on any claimed separation should be mandatory' while the paper's own Fig. 10 uses parametric t4 values as 'reproducibility diagnostics'; a brief explanation of how to combine the two under the proposed standard would remove the apparent tension.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's verdict rests on independent literature, its own experiments are falsifiable and self-correcting, and no load-bearing step reduces to its inputs by construction.

full rationale

The paper is a critical review whose central negative verdict is assembled from independent sources (Reinholdt et al., Gaberle and Jattana, Vaquero-Sabater et al., and Belagali et al. [26]), none authored by the present author, so the self-citation patterns that ground circularity findings do not occur. The paper's own computations—the GFlowNet compactness study, the Epstein–Nesbet rank-correlation sweep, and the noise-crossover experiment—are presented with exact-FCI ground truth and multi-seed statistics, and they explicitly refute the author's own initial multireference-specificity hypothesis; a finding that contradicts its author's expectation is not circular. The strongest claim that single-layer LUCJ circuits are classically reproducible at the energy level is imported from the external preprint [26] rather than derived here, but that is an external dependency and a verifiability/correctness risk, not a circular reduction: the paper does not fit [26]'s result to its own data, and the broader no-advantage verdict also stands on independent critique literature. No equation in the paper is defined in terms of its conclusion, no fitted parameter is renamed as a prediction, and no uniqueness claim is imported from the author's prior work. Accordingly the circularity score is 0.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The paper's own quantitative claims rest on small exact active spaces, simulated noise, an imported dequantization theorem, and a chosen reward. The central negative verdict mostly imports external literature and does not depend on the free parameters, but the secondary 'generic' noise-robustness claim does.

free parameters (3)
  • subspace dimension D = 120
    Fixed by hand for the compactness and noise-crossover studies (Figs. 6, 7, 10); all reported gaps are at this dimension.
  • GFlowNet reward temperature beta = 0.5
    Tempering exponent for the cheap Epstein-Nesbet reward in the compactness study (Fig. 6); no temperature sweep is reported, so the comparison to greedy top-K holds at one temperature.
  • noise scale multiplier = 0x, 1x, 2x, 3x FakeTorino/Heron
    Swept, not fitted; the crossover conclusion is conditional on this noise model and on the chosen scale range.
assumptions (5)
  • domain assumption Exact FCI in CAS(10e,12o) N2 and CAS(8e,12o) H2O is representative of the regime where SQD operates.
    All original energies are errors against exact FCI in these small spaces; if they do not capture the heavy-tail regime of larger hardware runs, the regime map may not transfer.
  • domain assumption FakeTorino/Heron backend-calibrated noise models faithfully represent real-device readout and depolarizing errors.
    The noise-robustness crossover (Fig. 10) is entirely simulated with local noise models; no real-hardware confirmation is included.
  • domain assumption The single-layer LUCJ dequantization theorem of [26] is correct and applies to the flagship circuits.
    The strongest form of the negative verdict is imported from this preprint without independent derivation here.
  • domain assumption The Epstein-Nesbet first-order coefficient is an adequate proxy for the importance signal used by strong classical selected-CI heuristics.
    It defines the order parameter rho and the GFlowNet reward; if classical selectors use substantially richer signals, the regime map coordinate may be mis-specified.
  • standard math Slater-Condon rules and the Hylleraas-Undheim-MacDonald variational theorem hold.
    Used to justify the projected Hamiltonian energies and upper-bound convergence.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Machine learning for sample-based quantum diagonalization: generative configuration recovery and the classical-simulability frontier." pith.science (2026). https://pith.science/paper/CZPTSUPG

@misc{pith2026260805314,
  author       = {Pith},
  title        = {Pith review of: Machine learning for sample-based quantum diagonalization: generative configuration recovery and the classical-simulability frontier},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CZPTSUPG}},
  note         = {Machine review of arXiv:2608.05314}
}
read the original abstract

Sample-based quantum diagonalization (SQD), equivalently quantum-selected configuration interaction (QSCI), has in two years become a pragmatic centre of gravity of pre-fault-tolerant quantum chemistry: a quantum processor samples electronic configurations, and the many-electron Hamiltonian is diagonalized classically in the resulting determinant subspace. Its accuracy is set entirely by which configurations enter that subspace, a selection problem for machine learning made acute by a coupon-collector bottleneck. We critically review the ecosystem of generative and learned selectors, organizing it by the object each method generates and the importance signal it exploits, and expose one conspicuous gap: a reward-proportional generative-flow-network proposer built for tail discovery. We then confront the field's central question -- whether the quantum sampler beats classical selected configuration interaction -- and report a carefully scoped negative: across published same-active-space comparisons, strong classical selected CI matches or beats the quantum-sampled subspace, and the flagship single-layer circuits now admit polynomial-time classical energy estimation. We distil a benchmarking standard and turn the negative into a regime map, then test it with FCI-exact experiments that confirm one prediction and refute another: the cheap prior's rank correlation with the exact weights declines with multireference character (a usable coordinate), but a controlled single-molecule noise sweep shows the one generative advantage we find, robustness to valid-shot starvation, to be generic rather than the multireference-specific effect a confounded contrast first suggested. Finally, we flag learning from quantum experiments, whose classical sample-complexity lower bound is an unconditional theorem, as the one adjacent frontier where a quantum advantage is provable but not yet bridged to chemistry.

Figures

Figures reproduced from arXiv: 2608.05314 by the authors.

Figure 1
Figure 1. The sample-based quantum diagonalization workflow, with real data at every stage [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. The molecular systems, their active-space orbital-energy structure, and the frontier [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. The coupon-collector bottleneck, computed exactly. [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Configuration recovery under backend-calibrated noise [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: A taxonomy of machine-learning methods for determinant and subspace selection in [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: Compactness under a cheap, FCI-free reward [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]
Figure 7
Figure 7. Figure 7: Classical selected CI versus a learned proposer at FCI-verifiable scale [PITH_FULL_IMAGE:figures/full_fig_p019_7.png]
Figure 8
Figure 8. Figure 8: The advantage landscape of §5–§7. The structure that makes SQD classically verifiable is the structure that makes it classically constructible; advantage survives only where that equivalence breaks. Colour-blind-safe (Okabe–Ito) palette; zones are also distinguished by…
Figure 9
Figure 9. Figure 9: The cheap prior degrades with multireference character (N [PITH_FULL_IMAGE:figures/full_fig_p022_9.png]
Figure 10
Figure 10. Figure 10: A controlled single-molecule test refutes multireference-specificity of the noise [PITH_FULL_IMAGE:figures/full_fig_p023_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

108 extracted references · 28 canonical work pages

  1. [26]

    Belagali, H., Van Camp, T., Pradeep, R., Das, S., Anand, N., LaRose, R.,Efficient classical simulation of large-scale unitary cluster Jastrow circuits, arXiv:2607.21337 (2026)

  2. [14]

    M., Kjellgren, E

    Reinholdt, P., Ziems, K. M., Kjellgren, E. R., Coriani, S., Sauer, S. P. A., Kongsted, J.,Critical limitations in quantum-selected configuration interaction methods,J. Chem. Theory Comput. 21, 6811 (2025); arXiv:2501.07231

  3. [24]

    A Critical Assessment of the Sample-Based Quantum Diagonalization for Heisenberg and Hubbard Models

    Gaberle, C., Jattana, M. S.,A critical assessment of sample-based quantum diagonalization for Heisenberg and Hubbard models, arXiv:2605.02494 (2026)

  4. [25]

    Vaquero-Sabater, N., Carreras, A., Broers, L., Shirakawa, T., Yunoki, S., Casanova, D.,Noise and configuration- recovery impact on quantum-selected configuration interaction, arXiv:2605.23697 (2026)

  5. [1]

    Kempe, J., Kitaev, A., Regev, O.,The complexity of the local Hamiltonian problem,SIAM J. Comput. 35, 1070 (2006)

  6. [2]

    P., Rancurel, P.,Iterative perturbation calculations of ground and excited state energies (CIPSI),J

    Huron, B., Malrieu, J. P., Rancurel, P.,Iterative perturbation calculations of ground and excited state energies (CIPSI),J. Chem. Phys. 58, 5745 (1973)

  7. [3]

    A., Tubman, N

    Holmes, A. A., Tubman, N. M., Umrigar, C. J.,Heat-bath configuration interaction (HCI),J. Chem. Theory Comput. 12, 3674 (2016); arXiv:1606.07453

  8. [4]

    A., Jeanmairet, G., Alavi, A., Umrigar, C

    Sharma, S., Holmes, A. A., Jeanmairet, G., Alavi, A., Umrigar, C. J.,Semistochastic heat-bath configuration interaction (SHCI),J. Chem. Theory Comput. 13, 1595 (2017); arXiv:1610.06660

Show all 108 references
  1. [5]

    M., et al., Head-Gordon, M., Whaley, K

    Tubman, N. M., et al., Head-Gordon, M., Whaley, K. B.,Adaptive sampling configuration interaction (ASCI),J. Chem. Phys. 145, 044112 (2016)

  2. [6]

    L.,A variational eigenvalue solver on a photonic quantum processor (VQE),Nat

    Peruzzo, A., McClean, J., Shadbolt, P., et al., O’Brien, J. L.,A variational eigenvalue solver on a photonic quantum processor (VQE),Nat. Commun. 5, 4213 (2014)

  3. [7]

    R., Boixo, S., Smelyanskiy, V

    McClean, J. R., Boixo, S., Smelyanskiy, V. N., Babbush, R., Neven, H.,Barren plateaus in quantum neural network training landscapes,Nat. Commun. 9, 4812 (2018)

  4. [8]

    Larocca, M., et al.,A review of barren plateaus in variational quantum computing, arXiv:2405.00781 (2025)

  5. [9]

    Cerezo, M., Larocca, M., García-Martín, D., et al.,Does provable absence of barren plateaus imply classical simulability?,Nat. Commun. (2025); arXiv:2312.09121

  6. [10]

    O.,Quantum-selected configuration interaction (QSCI),Phys

    Kanno, K., Kohda, M., Imai, R., Koh, S., Mitarai, K., Mizukami, W., Nakagawa, Y. O.,Quantum-selected configuration interaction (QSCI),Phys. Rev. Research 8, 023268 (2026); arXiv:2302.11320 (2023)

  7. [11]

    Robledo-Moreno, J., Motta, M., et al.,Chemistry beyond the scale of exact diagonalization on a quantum- centric supercomputer (SQD),Science Advances 11, eadu9991 (2025), DOI 10.1126/sciadv.adu9991; arXiv:2405.05068

  8. [12]

    Shirakawa, T., et al., Yunoki, S., Motta, M.,Full-scale closed-loop electronic-structure calculation on a QPU coupled to Fugaku, arXiv:2511.00224 (2025)

  9. [13]

    Wang, Q., Motta, M., et al., Gagliardi, L.,Localized sample-based quantum diagonalization for strongly correlated chemistry (LAS-SQD),PNAS 123, e2603914123 (2026); arXiv:2512.14936

  10. [15]

    Patra, C., Mondal, D., Halder, S., et al., Maitra, R.,Physics-informed generative ML for accelerated quantum- centric supercomputing (PIGen-SQD), arXiv:2512.06858 (2025)

  11. [16]

    K., Anurag, K

    Patra, A. K., Anurag, K. S. V., Bhat, R., et al., Maitra, R., Jaiganesh, G.,Machine-learned compact subspace generation for QSCI within DMET (DMET-QSCI-RBM), arXiv:2607.20585 (2026)

  12. [17]

    Chang, J.-Y., Kuo, E.-J., et al.,Iterative dual-channel neural-quantum-state algorithm for selected configuration interaction (HI-NQS), arXiv:2606.26760 (2026)

  13. [18]

    J., Ding, L., Reiher, M.,Neural quantum states based on selected configurations (NQS-SC),J

    Solanki, M. J., Ding, L., Reiher, M.,Neural quantum states based on selected configurations (NQS-SC),J. Phys. Chem. Lett. (2026); arXiv:2602.12993

  14. [19]

    SC’25 (2025)

    Shang, H., et al.,NNQS-SCI: tackling trillion-dimensional Hilbert space with adaptive neural quantum states, Proc. SC’25 (2025). 26

  15. [20]

    Zeni, S., Varutti, G., Nespolo, J., Trypogeorgos, D.,Enhancing quantum-classical configuration interaction with a neural-network classifier (cSQD), arXiv:2606.24332 (2026)

  16. [21]

    Zhang, H., Zeng, X., Li, Z., Zhou, Y.,Hamiltonian-guided autoregressive selected-CI (HAAR-SCI),J. Chem. Theory Comput. (2025), DOI 10.1021/acs.jctc.5c01415

  17. [22]

    Nie, W., et al., Yang, J.,Learning to rank for selected configuration interaction (RCI), arXiv:2605.10348 (2026)

  18. [23]

    Kemmoku, R., Gao, Q., Kanno, S., et al., Nakaji, K.,Generative circuit design for quantum-selected configuration interaction, arXiv:2604.09756 (2026)

  19. [27]

    Hafid, A., Iwakiri, H., Tsubouchi, K., Yoshioka, N., Kohda, M.,On the hardness of classically sampling quantum- chemistry circuits, arXiv:2504.12893 (2025)

  20. [28]

    Kirby, W., et al., Yunoki, S., Mezzacapo, A.,Observation of improved accuracy over classical sparse ground-state solvers using a quantum computer, arXiv:2603.03496 (2026)

  21. [29]

    J., Whaley, K

    Motta, M., Sung, K. J., Whaley, K. B., Head-Gordon, M., Shee, J.,The local unitary cluster Jastrow (LUCJ) ansatz,Chem. Sci. 14 (2023), DOI 10.1039/D3SC02516K

  22. [30]

    O., Kamoshita, M., Mizukami, W., et al.,ADAPT-QSCI: adaptive construction of an input state for quantum-selected configuration interaction,J

    Nakagawa, Y. O., Kamoshita, M., Mizukami, W., et al.,ADAPT-QSCI: adaptive construction of an input state for quantum-selected configuration interaction,J. Chem. Theory Comput. 20, 10817 (2024); arXiv:2311.01105

  23. [31]

    O.,Quantum-selected configuration interaction with a time-evolved state (TE-QSCI), Phys

    Mikkelsen, M., Nakagawa, Y. O.,Quantum-selected configuration interaction with a time-evolved state (TE-QSCI), Phys. Rev. Research 7, 043043 (2025); arXiv:2412.13839

  24. [32]

    Sugisaki, K., Kanno, S., Itoko, T., Sakuma, R., Yamamoto, N.,Hamiltonian-simulation-based quantum-selected configuration interaction (HSB-QSCI),Phys. Chem. Chem. Phys. 27, 20869 (2025); arXiv:2412.07218

  25. [33]

    Yu, J., Robledo-Moreno, J., et al., Mezzacapo, A., Sharma, K.,Quantum-centric algorithm for sample-based Krylov diagonalization (SKQD), arXiv:2501.09702 (2025)

  26. [34]

    Piccinelli, S., Baiardi, A., Barison, S., et al., Tavernelli, I.,Provable-convergence randomized sample-based Krylov quantum diagonalization (SqDRIFT), arXiv:2508.02578 (2025)

  27. [35]

    Stockinger, N., et al.,Sample-based quantum diagonalization with amplitude amplification of unseen bitstrings (SQD-AA), arXiv:2605.02565 (2026)

  28. [36]

    Miura, R., et al.,Active-sampling sample-based quantum diagonalization (AS-SQD), arXiv:2603.13536 (2026)

  29. [37]

    McFarthing, S., Pellow-Jarman, A., Petruccione, F.,Noise-resilient quantum chemistry with half the qubits (HCI-HSQD), arXiv:2602.01165 (2026)

  30. [38]

    Yoshida, Y., et al., Mizukami, W.,Doubling the size of QSCI via the seniority-zero space (DOCI-QSCI), arXiv:2602.07912 (2026)

  31. [39]

    Smith, T., et al.,Entanglement-forging sample-based quantum diagonalization, arXiv:2508.08229 (2025)

  32. [40]

    Chen, Y., et al.,Neural fermionic compression for qubit-reduced SQD (Lossy-QSCI), arXiv:2505.17846 (2025)

  33. [41]

    Q., Kapourniotis, T., Georgopoulos, K.,Resource-efficient quantum algorithms for selected Hamiltonian subspace diagonalization (CIM-QSCI / QSHCI), arXiv:2603.13160 (2026)

    Graves, V., Hlatshwayo, M. Q., Kapourniotis, T., Georgopoulos, K.,Resource-efficient quantum algorithms for selected Hamiltonian subspace diagonalization (CIM-QSCI / QSHCI), arXiv:2603.13160 (2026)

  34. [42]

    Shajan, A., et al.,Sample-based quantum diagonalization within density-matrix embedding theory (DMET-SQD), J. Chem. Theory Comput. (2025)

  35. [43]

    J., Motta, M., Shee, J.,Sample-based quantum diagonalization enhanced by phaseless auxiliary-field quantum Monte Carlo (AFQMC-SQD),J

    Danilov, D., Robledo-Moreno, J., Sung, K. J., Motta, M., Shee, J.,Sample-based quantum diagonalization enhanced by phaseless auxiliary-field quantum Monte Carlo (AFQMC-SQD),J. Chem. Theory Comput. (2025); arXiv:2503.05967

  36. [44]

    Erhart, L., Yoshida, Y., Mizukami, W.,Coupled cluster method tailored by quantum-selected configuration interaction (TCC-QSCI), arXiv:2506.16911 (2025)

  37. [45]

    Barison, S., Robledo-Moreno, J., Motta, M.,Extended sample-based quantum diagonalization for excited states, Quantum Sci. Technol. 10, 025034 (2025); arXiv:2411.00468

  38. [46]

    V.,Towards compact wavefunctions from quantum-selected configuration interaction, arXiv:2509.02525 (2025)

    Weaving, T., Mingare, A., Ralli, A., Coveney, P. V.,Towards compact wavefunctions from quantum-selected configuration interaction, arXiv:2509.02525 (2025)

  39. [47]

    Nogaki, K., Backes, S., Shirakawa, T., Yunoki, S., Arita, R.,Symmetry-adapted sample-based quantum diagonalization: application to lattice model,Phys. Rev. B 114, 065124 (2026); arXiv:2505.00914

  40. [48]

    Huang, H.-Y., Kueng, R., Preskill, J.,Predicting many properties of a quantum system from very few measurements (classical shadows),Nat. Phys. 16, 1050 (2020); arXiv:2002.08953

  41. [49]

    Bhuiyan, A., Chu, C., Lou, Q., Zheng, M.,On the hardware robustness of sample-based quantum diagonalization, arXiv:2607.18196 (2026). 27

  42. [50]

    P.,Machine learning configuration interaction,J

    Coe, J. P.,Machine learning configuration interaction,J. Chem. Theory Comput. 14, 5739 (2018); arXiv:1808.05787

  43. [51]

    P.,Machine-learning configuration interaction for excited states and potential-energy curves,J

    Coe, J. P.,Machine-learning configuration interaction for excited states and potential-energy curves,J. Chem. Theory Comput. 15 (2019)

  44. [52]

    Sun, D., et al., Shang, H., Tao, D.,GPU-accelerated CI selection with neural quantum states (QiankunNet-cuSCI), arXiv:2604.15768 (2026)

  45. [53]

    Thompson, S., Gunlycke, D.,Auto-regressive neural-quantum-state sampling for selected configuration interaction (ARNN-SCI), arXiv:2603.24728 (2026)

  46. [54]

    J., Hu, H., Yang, C., Li, X.,Reinforcement learning configuration interaction (RL-CI),J

    Goings, J. J., Hu, H., Yang, C., Li, X.,Reinforcement learning configuration interaction (RL-CI),J. Chem. Theory Comput. 17, 5482 (2021), DOI 10.1021/acs.jctc.1c00010

  47. [55]

    Zeng, X., Gong, M., Kan, B., et al., Pan, J.-W., Yang, J.,Transformer-refined quantum sampling for strongly correlated electronic structure (QiankunNet-QSCI on Zuchongzhi 3.1), arXiv:2605.24617 (2026)

  48. [56]

    Herzog, B., Casier, B., Lebègue, S., Rocca, D.,Solving the Schrödinger equation in configuration space with generative machine learning, arXiv:2208.06708 (2022)

  49. [57]

    I., Mendez-Vazquez, A., et al.,Configuration-interaction-guided sampling with an interpretable restricted Boltzmann machine,Mach

    Hernandez-Martinez, J. I., Mendez-Vazquez, A., et al.,Configuration-interaction-guided sampling with an interpretable restricted Boltzmann machine,Mach. Learn.: Sci. Technol. (2025), DOI 10.1088/2632- 2153/ae0fd5; arXiv:2409.06146

  50. [58]

    Carleo, G., Troyer, M.,Solving the quantum many-body problem with artificial neural networks,Science 355, 602 (2017)

  51. [59]

    Thirion, L., Schmerwitz, Y. L. A., Kroesbergen, M., Levi, G., Jónsson, E. Ö., Bilous, P., Jónsson, H., Hansmann, P.,Natural-orbital-based neural-network configuration interaction (NNCI), arXiv:2510.27665 (2025)

  52. [60]

    Schmerwitz, Y. L. A., Thirion, L., Levi, G., Jónsson, E. Ö., Bilous, P., Jónsson, H., Hansmann, P.,A neural- network-based selective configuration interaction approach to molecular electronic structure (NNCI),J. Chem. Theory Comput. 21, 2301 (2025); arXiv:2406.08154

  53. [61]

    Nakaji, K., et al., Aspuru-Guzik, A.,The generative quantum eigensolver (GQE / GPT-QE), arXiv:2401.09253 (2024)

  54. [62]

    Bengio, E., Jain, M., Korablyov, M., Precup, D., Bengio, Y.,Flow network based generative models for non-iterative diverse candidate generation,NeurIPS 34 (2021); arXiv:2106.04399

  55. [63]

    Malkin, N., Jain, M., Bengio, E., Sun, C., Bengio, Y.,Trajectory balance: improved credit assignment in GFlowNets, NeurIPS 35 (2022); arXiv:2201.13259

  56. [64]

    Malkin, N., et al., Bengio, Y.,GFlowNets and variational inference,ICLR (2023); arXiv:2210.00580

  57. [65]

    12, 5 (2026), DOI 10.1038/s41534-025-01159-x; arXiv:2507.01726

    Zou, H., Rahm, M., Frisk Kockum, A., Olsson, S.,Generative flow-based warm start of the variational quantum eigensolver (Flow-VQE),npj Quantum Inf. 12, 5 (2026), DOI 10.1038/s41534-025-01159-x; arXiv:2507.01726

  58. [66]

    Huidobro-Meezs,I.L.,Dai,J.,Rabusseau,G.,Vargas-Hernández,R.A.,GFlowNets for Hamiltonian decomposition in groups of compatible operators, NeurIPS 2024 Machine Learning and the Physical Sciences Workshop, arXiv:2410.16041 (2024)

  59. [67]

    L., Dai, J., Vargas-Hernández, R

    Huidobro-Meezs, I. L., Dai, J., Vargas-Hernández, R. A.,Discrete flow-based generative models for measurement optimization in quantum computing, arXiv:2509.15486 (2025)

  60. [68]

    Zhang, D., Malkin, N., et al., Bengio, Y.,Generative flow networks for discrete probabilistic modeling (EB-GFN), ICML (2022); arXiv:2202.01361

  61. [69]

    Pandey, M., Subbaraj, G., Bengio, E.,GFlowNet pretraining with inexpensive rewards, arXiv:2409.09702 (2024)

  62. [70]

    Kim, M., Ko, J., Yun, T., Zhang, D., Pan, L., Kim, W., Park, J., Bengio, E., Bengio, Y.,Learning to scale logits for temperature-conditional GFlowNets,ICML (2024); arXiv:2310.02823

  63. [71]

    Sun, Z., Yang, Y.,DIFUSCO: graph-based diffusion solvers for combinatorial optimization,NeurIPS (2023)

  64. [72]

    S., Matthews, A

    Pfau, D., Spencer, J. S., Matthews, A. G. D. G., Foulkes, W. M. C.,Ab initio solution of the many-electron Schrödinger equation with deep neural networks (FermiNet),Phys. Rev. Research 2, 033429 (2020)

  65. [73]

    Hermann, J., Schätzle, Z., Noé, F.,Deep-neural-network solution of the electronic Schrödinger equation (PauliNet), Nat. Chem. 12, 891 (2020)

  66. [74]

    S., Pfau, D.,A self-attention ansatz for ab initio quantum chemistry (Psiformer), ICLR (2023)

    von Glehn, I., Spencer, J. S., Pfau, D.,A self-attention ansatz for ab initio quantum chemistry (Psiformer), ICLR (2023)

  67. [75]

    Gao, N., Günnemann, S.,Neural Pfaffians: solving many-electron Schrödinger equations with neural networks, NeurIPS (2024)

  68. [76]

    Foster, A., Schätzle, Z., et al., Noé, F., Hermann, J.,Orbformer: an ab initio foundation model of electronic wavefunctions, arXiv:2506.19960 (2025)

  69. [77]

    K.,Neural network backflow for ab initio quantum chemistry (NNBF),Phys

    Liu, A.-J., Clark, B. K.,Neural network backflow for ab initio quantum chemistry (NNBF),Phys. Rev. B 110, 115137 (2024). 28

  70. [78]

    K.,Efficient optimization of neural network backflow for ab initio quantum chemistry (opt-NNBF), arXiv:2502.18843 (2025)

    Liu, A.-J., Clark, B. K.,Efficient optimization of neural network backflow for ab initio quantum chemistry (opt-NNBF), arXiv:2502.18843 (2025)

  71. [79]

    Giuliani, C., Nys, J., Martinazzo, R., Carleo, G., Rossi, R.,Precise quantum chemistry calculations with few Slater determinants,Nat. Commun. (2026), DOI 10.1038/s41467-026-70255-z; arXiv:2503.14502

  72. [80]

    Hermann, J., et al., Noé, F.,Ab initio quantum chemistry with neural-network wavefunctions(review),Nat. Rev. Chem. 7, 692 (2023); arXiv:2208.12590

  73. [81]

    D., Malyshev, A., Lvovsky, A

    Barrett, T. D., Malyshev, A., Lvovsky, A. I.,Autoregressive neural-network wavefunctions for ab initio quantum chemistry (NAQS),Nat. Mach. Intell. 4, 351 (2022); arXiv:2109.12606

  74. [82]

    Garniron, Y., et al., Scemama, A., Loos, P.-F.,Quantum Package 2.0: a CIPSI/selected-CI open-source program, J. Chem. Theory Comput. 15, 3591 (2019); arXiv:1902.08154

  75. [83]

    Loos, P.-F., Damour, Y., Ammar, A., Caffarel, M., Kossoski, F., Scemama, A.,Go green: selected configuration interaction as a sustainable route to high accuracy, arXiv:2402.13111 (2024)

  76. [84]

    R.,Density matrix formulation for quantum renormalization groups (DMRG),Phys

    White, S. R.,Density matrix formulation for quantum renormalization groups (DMRG),Phys. Rev. Lett. 69, 2863 (1992)

  77. [85]

    K.-L., Head-Gordon, M.,Highly correlated calculations with a polynomial cost algorithm (DMRG),J

    Chan, G. K.-L., Head-Gordon, M.,Highly correlated calculations with a polynomial cost algorithm (DMRG),J. Chem. Phys. 116, 4462 (2002)

  78. [86]

    Zhang, S., Krakauer, H.,Quantum Monte Carlo method using phase-free random walks (phaseless AFQMC), Phys. Rev. Lett. 90, 136401 (2003)

  79. [87]

    H., Thom, A

    Booth, G. H., Thom, A. J. W., Alavi, A.,Fermion Monte Carlo without fixed nodes (FCIQMC),J. Chem. Phys. 131, 054106 (2009)

  80. [88]

    J., et al.,The ground-state electronic energy of benzene,J

    Eriksen, J. J., et al.,The ground-state electronic energy of benzene,J. Phys. Chem. Lett. 11, 8922 (2020); arXiv:2008.02678

  81. [89]

    M., Chan, G

    Motta, M., Ceperley, D. M., Chan, G. K.-L., et al.,Towards the solution of the many-electron problem: hydrogen chain benchmark,Phys. Rev. X 7, 031059 (2017)

  82. [90]

    T., Yao, Y., et al., Umrigar, C

    Williams, K. T., Yao, Y., et al., Umrigar, C. J., Chan, G. K.-L., Zhang, S., Wagner, L. K. (Simons Collaboration on the Many-Electron Problem),Direct comparison of many-body methods for realistic electronic Hamiltonians, Phys. Rev. X 10, 011041 (2020); arXiv:1910.00045

  83. [91]

    P., Abraham, V., Peng, B., Asthana, A.,Chemically decisive benchmarks on the path to quantum utility, arXiv:2601.10813 (2026)

    Sundar, S. P., Abraham, V., Peng, B., Asthana, A.,Chemically decisive benchmarks on the path to quantum utility, arXiv:2601.10813 (2026)

  84. [92]

    M., Zhang, H., Motta, M., Faulstich, F

    Raisuddin, O. M., Zhang, H., Motta, M., Faulstich, F. M.,From promise to practice: benchmarking quantum chemistry on quantum hardware, arXiv:2512.01012 (2025)

  85. [93]

    Zhang, H., Otten, M.,From random determinants to the ground state (TrimCI), arXiv:2511.14734 (2025)

  86. [94]

    S., Chong, F

    Hassman, Z., Reardon-Smith, O., Ravi, G. S., Chong, F. T., Sung, K. J.,ExtraFerm: an extended matchgate simulator, arXiv:2511.12416 (2025)

  87. [95]

    E., Motta, M., Robledo-Moreno, J., Mezzacapo, A., Chuang, I

    Chin, Z. E., Motta, M., Robledo-Moreno, J., Mezzacapo, A., Chuang, I. L., Kirby, W.,Polynomial-time exact diagonalization via sparse guided eigenwalks, arXiv:2606.23967 (2026)

  88. [96]

    K.-L.,Classical computational simulation of the FeMo-cofactor model to chemical accuracy, arXiv:2601.04621 (2026)

    Zhai, H., Li, C., Zhang, X., Li, Z., Lee, S., Chan, G. K.-L.,Classical computational simulation of the FeMo-cofactor model to chemical accuracy, arXiv:2601.04621 (2026)

  89. [97]

    Ma, H., Kan, B., Shang, H., Yang, J.,Transformer-based neural-network backflow for strongly correlated molecules, arXiv:2509.25720 (2025)

  90. [98]

    K.-L., Babbush, R.,Evaluating the evidence for exponential quantum advantage in ground-state quantum chemistry,Nat

    Lee, S., Lee, J., Zhai, H., et al., Chan, G. K.-L., Babbush, R.,Evaluating the evidence for exponential quantum advantage in ground-state quantum chemistry,Nat. Commun. 14, 1952 (2023); arXiv:2208.02199

  91. [99]

    C., et al.,Quantum advantage in computational chemistry?, arXiv:2508.20972 (2025)

    Gundlach, H., Thompson, N. C., et al.,Quantum advantage in computational chemistry?, arXiv:2508.20972 (2025)

  92. [100]

    A., Xantheas, S

    Legeza, Ö., Menczer, A., Werner, M. A., Xantheas, S. S., Neese, F., Ganahl, M., Brower, C., Rodriguez Bernabeu, S., Hammond, J., Gunnels, J.,Hunting for quantum advantage in electronic structure calculations is a highly non-trivial task, arXiv:2603.28648 (2026)

  93. [101]

    F., Mayhall, N

    Shrikhande, C., Bachhar, A., Rodriguez Jimenez, A., Valeev, E. F., Mayhall, N. J.,Pauli propagation enables fast classical simulation of strongly correlated quantum systems, arXiv:2511.21651 (2025)

  94. [102]

    Chen, S., Cotler, J., Huang, H.-Y., Li, J.,Exponential separations between learning with and without quantum memory,FOCS (2021); arXiv:2111.05881

  95. [103]

    Huang, H.-Y., Broughton, M., Cotler, J., et al.,Quantum advantage in learning from experiments,Science 376, 1182 (2022); arXiv:2112.00778

  96. [104]

    L., Preskill, J.,Quantum learning advantage on a scalable photonic platform, Science (2025); arXiv:2502.07770

    Liu, Z.-H., et al., Andersen, U. L., Preskill, J.,Quantum learning advantage on a scalable photonic platform, Science (2025); arXiv:2502.07770

  97. [105]

    Huang, H.-Y., Choi, S., McClean, J., Preskill, J.,The vast world of quantum advantage,Phys. Rev. X (2025); arXiv:2508.05720. 29

  98. [106]

    Cotler, J., Gong, W., Kannan, I.,Noisy quantum learning theory, arXiv:2512.10929 (2025)

  99. [107]

    Seif, A., et al., Minev, Z.,Entanglement-enhanced learning of quantum processes at scale, arXiv:2408.03376 (2024)

  100. [108]

    S., Sitdikov, I., Salcedo, C., Seif, A., Minev, Z

    Liao, H., Wang, D. S., Sitdikov, I., Salcedo, C., Seif, A., Minev, Z. K.,Machine learning for practical quantum error mitigation (ML-QEM),Nat. Mach. Intell. 6, 1478 (2024); arXiv:2309.17368. 30

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.