Pith. sign in

REVIEW 4 major objections 5 minor 39 references

Sparse Quantum Voxel Encoding for Readout-Efficient Molecular Geometry Reconstruction on NISQ Devices

T0 review · 4 major / 5 minor · reviewed 2026-07-30 · grok-4.5

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Encoding molecules as sparse equal superpositions turns geometry readout into a coupon-collector problem needing only O(A log A) shots.

desk verdict Solid NISQ encoding-and-demo paper: coupon-collector readout works on hardware for oracle recall, but unsupervised support recovery under real device bias is not shown and state prep is left open. read the letter →

arxiv 2607.26925 v1 pith:2I3VT7MB submitted 2026-07-29 quant-ph cs.DS

Eros De Simone , Giuseppe Bifulco , Lorenza Di Mauro , Antonio Policicchio , Raoul Heese This is my paper · ORCID
classification quant-phcs.DS
keywords quantumvoxelencodingmoleculargeometryNISQcouponcollectorsupportrecoverycomputational-basissamplingreadoutefficiencystatetomographyalternative
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

Reading a full molecular geometry out of a quantum state usually means expensive tomography that scales exponentially with qubit number. This paper argues that if you discretize space into voxels and put each atom’s position and type into its own computational-basis label, then store the molecule as an equal superposition over those few occupied labels, reconstruction collapses to sampling until every atom appears at least once. In the ideal case that takes only on the order of A log A shots for an A-atom molecule, not exponential tomography. On real IBM hardware the authors reconstruct a 10-atom ethylamine molecule from an 8-qubit circuit with mean recall near 1 using a few hundred shots, even though most shots are noise. The point is a readout-friendly representation for near-term quantum models that already output quantum states, at the cost of spatial quantization to the voxel grid.

What carries the argument

Sparse quantum voxel encoding: each atom maps to one basis state via a combined voxel-index and atom-type index; the molecule is the equal superposition over those occupied states, so reconstruction is classical coupon collection on measurement outcomes.

What would settle it

Prepare the same 8-qubit ethylamine support state on comparable hardware and check whether mean reconstruction recall stays near 0.94 at ~116 shots and near 0.98 at ~200 shots; failure to reach high recall at those budgets, or circuits so deep that the support is destroyed before sampling, would refute the practical claim.

Watch

Extended reading notes

Core claim

Once a molecule is prepared as an equal superposition over the A computational-basis states that encode its voxelized atom positions and types, complete support recovery by computational-basis sampling requires O(A log A) shots in the noise-free case, and on noisy NISQ hardware a 10-atom molecule can still be recovered at high mean recall with only O(10^2) shots—orders of magnitude fewer measurements than full state tomography.

Load-bearing premise

The paper assumes a usable way already exists to prepare or learn that sparse equal-superposition state; without it the cheap readout does not yield a practical end-to-end method.

Editorial extensions

If this is right

  • Quantum generative models that emit states in this encoding can hand molecular geometries to classical post-processing with coupon-collector shot budgets instead of tomography.
  • Measurement cost at readout scales with atom count A, not with 3^n Pauli settings, for fixed voxel resolution.
  • Spatial precision is capped by voxel size, forcing an explicit qubits-versus-resolution trade-off when targeting sub-ångström work.
  • Signal-to-noise over empty voxels can serve as a ground-truth-free proxy to decide which basis states are real atoms in a generative setting.

Reading between the lines

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

  • If state-preparation depth cannot be cut well below the demonstrated ~1200-gate regime, the encoding’s readout win may stay confined to proof-of-concept circuits rather than larger generative models.
  • Adaptive or symmetry-aware grids (hinted in the paper’s outlook) would be the natural next test: same coupon-collector math, less orientation dependence and fewer wasted basis states.
  • The method is closest in spirit to basis encoding with a deliberately sparse support; neighbouring quantum ML tasks that only need set-like outputs could reuse the same support-recovery pattern.
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

4 major / 5 minor

Summary. The paper proposes quantum voxelization encoding: molecular geometries are discretized onto a 3D voxel grid, each atom’s voxel index and chemical type are packed into a single computational-basis label c_a, and the molecule is represented as an equal superposition over the A occupied basis states. Once that sparse support state is prepared, reconstruction reduces to classical coupon-collector sampling in the computational basis, requiring O(A log A) shots in the noise-free case (Sec. 3.2, App. C). The authors assume state preparation exists and focus on readout. They encode 10-atom ethylamine into 8 qubits (V=4, T=3, C=192), run on IBM Kingston and a noisy simulator, and report mean reconstruction recall R̄=0.94 at 116 shots and R̄=0.98 at 200 shots despite ~70% wasted shots, arguing this is a practical, readout-efficient representation for NISQ molecular generative models.

Significance. If the encoding is usable end-to-end, converting geometry readout from exponential tomography to polynomial support sampling is a genuine and useful contribution for quantum generative chemistry on NISQ hardware. The coupon-collector analysis is standard and correctly applied; the hardware campaign is concrete (10 repeats per shot budget, wasted-shot and noise breakdowns, simulator baseline, device appendix). Explicit acknowledgment that preparation is out of scope and that spatial precision is lost to the grid is appropriate. The work is best read as a readout-compatible representation and a small-scale NISQ feasibility demo, not as a complete data-loading or generative pipeline.

major comments (4)
  1. [Section 4.3, Eq. (15)] Sec. 4.3 and Eq. (15): reconstruction recall R is an oracle metric—it scores only whether known atoms in M* appear at least once and does not penalize false positives. The intended generative use case (Sec. 5 pipeline; end of Sec. 4.3) requires inferring an unknown support from the histogram alone. The paper never reports unsupervised recovery (e.g., top-A frequency ranking, thresholding, or clustering) of the true indices in Table 1. With the strong low-index/ground-state bias in Fig. 2a from T1 leakage, high oracle R at O(10^2) shots does not establish that an unknown molecule can be reconstructed. This is load-bearing for the “practical generative readout” claim.
  2. [Section 4.3, Eq. (16)] Sec. 4.3, Eq. (16): SNR is introduced as a “ground-truth-free proxy,” but it is defined with labeled partitions ⟨N_valid⟩ over the 10 true atoms and ⟨N_empty⟩ over the 182 empties. That measures separability given M*, not recovery without M*. Either redefine SNR from unlabeled histogram statistics only, or drop the “ground-truth-free” language and add an actual unsupervised recovery experiment under the observed hardware bias.
  3. [Section 3.1; Appendix E; Abstract/Conclusions] Sec. 3.1 and App. E: preparation of |ψ_M*⟩ is assumed to exist; the demo uses isometry initialization yielding ~1213-gate 8-qubit circuits, already near coherence-limited depth. The abstract and conclusions frame the scheme as a “practical… representation for molecular geometries on near-term devices” and a building block for QAE/QGAN. Without either a scalable preparation path or a clear scope restriction to “readout of already-prepared sparse support states,” the NISQ practicality claim overreaches. Tighten claims to readout-only, or quantify when preparation remains feasible.
  4. [Section 3.1; Section 5] Sec. 3.1 collision-free grid: the worst-case condition √3 s_voxel < d_min would require s_voxel < 0.63 Å, but experiments use 0.95 Å and rely on molecule-specific orientation/centering. For multi-molecule datasets the paper defers “optimal collision-free grid design.” For the claimed use in generative libraries this is not a minor engineering detail—it determines qubit count and whether the encoding is universal. At minimum, state the qubit/resolution scaling for a fixed universal grid over a standard set (e.g., QM9 subset) or clearly limit claims to single-molecule, orientation-fixed encoding.
minor comments (5)
  1. [Abstract; Section 2] Abstract and Intro compare to O(3^n × 10^{2–3}) full tomography; for fairness also cite classical shadows / Pauli grouping limits when the task is full geometry recovery, as partially done in Sec. 2, so the baseline is not only naive tomography.
  2. [Figure 2] Fig. 2: mark the 10 true c_a values from Table 1 on the histogram (or an inset) so readers can see whether true atoms sit above the noise floor under hardware bias.
  3. [Appendix C.4] App. C.4 η̂ estimates (0.19–0.24) are lower than the direct valid-shot fraction ~0.29 in Fig. 3; a one-sentence reconciliation would help.
  4. [Introduction; Section 5] Typos/notation: “V AEs” spacing in Intro; consistent use of M* vs M^∗; “angstr¨ om” encoding in Sec. 5.
  5. [Title page; Section 4.1] Dates say July 30, 2026 / experiments April 2026—fine for arXiv, but confirm consistency before journal submission.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: coupon-collector bound is classical theorem on a constructed equal superposition; hardware recall is external measurement.

full rationale

The paper’s central readout claim is that once a molecule is prepared as an equal superposition over A occupied computational-basis labels (Eq. 9), support recovery reduces to the classical coupon-collector problem, giving O(A log A) shots in the noise-free case (Eqs. 12–13, Appendix C). That bound is not fitted from the ethylamine hardware outcomes; it is the standard coupon-collector expectation/union-bound applied to a uniform support of size A by construction of the encoding. Reconstruction recall R (Eq. 15) and the shot-budget table are evaluated against that external classical formula and against measured bitstrings on IBM Kingston and a noisy simulator; they do not redefine a fitted parameter as a prediction. State preparation is explicitly assumed out of scope and implemented only via a generic isometry for the demo—there is no circular claim that preparation efficiency follows from the readout analysis. No load-bearing self-citation, uniqueness theorem from the same authors, or ansatz smuggled via prior work appears in the derivation chain. Concerns that R is an oracle metric and that SNR (Eq. 16) still partitions on known M* are validity/scope issues for the generative use-case, not circular reductions of outputs to inputs. Score 0.

Assumptions & free parameters 3 free parameters · 4 assumptions · 1 invented entities

The readout claim rests on standard probability (coupon collector), standard basis encoding, and several domain choices: collision-free voxelization, equal superposition, and existence of state preparation. Free parameters are the grid and experimental shot budget chosen for one molecule. No new physical entities are postulated; 'voxelization encoding' is a representation, not an ontological addition.

free parameters (3)
  • grid resolution V and voxel size s_voxel = V=4, s_voxel=0.95 Å
    Chosen as V=4, s_voxel=0.95 Å for ethylamine so that atoms occupy distinct voxels under the chosen orientation; coarser than the paper’s own worst-case √3 s_voxel < d_min bound.
  • shot budget schedule S = 47–300 shots
    S ∈ {47,70,116,200,300} set from ideal coupon-collector confidence levels plus extra shots for noise; directly determines reported recall points.
  • atom-type set size T = T=3
    T=3 {C,H,N} fixed by the demo molecule; enters qubit count n=⌈log2(V³T)⌉.
assumptions (4)
  • ad hoc to paper A sparse equal-superposition support state |ψ_M*⟩ over the A occupied basis states can be prepared (existence assumed; preparation algorithm out of scope).
    Stated explicitly in abstract and Sec. 3.1; load-bearing for any claim of practical molecular representation on hardware.
  • domain assumption At most one atom per voxel after discretization (collisions resolved by refining the grid).
    Sec. 3.1; required for injective map from atoms to basis states and for the support size to equal A.
  • standard math Ideal measurements sample uniformly from the A occupied basis states (coupon collector model).
    Sec. 3.2 Eqs. (11)–(13); standard probability, with noisy non-uniform extension in App. C.
  • domain assumption Reconstruction quality is measured by recall of known support (and SNR as ground-truth-free proxy), not by continuous geometry or bond recovery.
    Sec. 4.3; defines success as recovering discretized atom-type point cloud only.
invented entities (1)
  • quantum voxelization encoding (sparse product index c_a = v_a·T + τ_a as computational-basis support)
    purpose: Map molecular geometry to a readout-friendly sparse basis support so tomography is replaced by support recovery.
    Named representation combining known basis encoding with classical voxelization; not a new physical object. independent_evidence false as a postulated entity, though the hardware demo is real.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Sparse Quantum Voxel Encoding for Readout-Efficient Molecular Geometry Reconstruction on NISQ Devices." pith.science (2026). https://pith.science/paper/2I3VT7MB

@misc{pith2026260726925,
  author       = {Pith},
  title        = {Pith review of: Sparse Quantum Voxel Encoding for Readout-Efficient Molecular Geometry Reconstruction on NISQ Devices},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2I3VT7MB}},
  note         = {Machine review of arXiv:2607.26925}
}
abstract

We propose a sparse computational-basis encoding of voxelized molecular geometries that converts molecular reconstruction from full-state tomography into support recovery by computational-basis sampling. To realize the encoding scheme, the molecular space is discretized into a 3D grid, and each atom's position and chemical species is mapped to a single computational basis state. This discretization introduces spatial quantization at the voxel-resolution scale. The molecule is then encoded as an equal superposition over this sparse set of occupied states, where we assume that a suitable state preparation method exists. In contrast to full state tomography, which requires on the order of $\mathcal{O}(3^n \times 10^{2\text{--}3})$ measurement shots, where $n$ is the number of qubits, our proposed encoding scheme reduces to a coupon-collector sampling problem in the computational basis. Complete recovery of an $A$-atom molecule requires $\mathcal{O}(A\log A)$ shots on noise-free hardware. On noisy hardware, the required number of shots increases. We demonstrate the method on the 156-qubit IBM Kingston device using 8-qubit circuits to reconstruct the discretized geometry of a 10-atom ethylamine molecule with high mean reconstruction recall using only $\mathcal{O}(10^2)$ shots despite substantial hardware noise. These results demonstrate that our proposed encoding scheme is a practical, readout-efficient representation for molecular geometries on near-term devices.

Figures

Figures reproduced from arXiv: 2607.26925 by the authors.

Figure 1
Figure 1. Ethylamine (C2H7N) encoded in a 4 × 4 × 4 voxel grid (svoxel = 0.95 ˚A). Carbon atoms (dark gray), nitrogen (blue), and hydrogen (light gray) are shown at their MMFF-optimized 3D positions after centering. Each atom is labeled with the corresponding voxel indices. Shaded voxel boxes highlight the cells containing atoms (e.g., N is in the voxel (0, 2, 1)). Complementary to the outcome distribution, we also present th… view at source ↗
Figure 2
Figure 2. Distribution of measurement outcomes. Green bars indicate the 10 valid atomic positions, orange bars rep￾resent empty voxel detections across 182 unoccupied grid in￾dices. Red highlights mark the 5 most frequently observed empty voxels. 4.3 Reconstruction Recall To quantify the quality at which a given quantum state |ψM∗ ⟩ has been decoded with a sequence of measure￾ments, we introduce the reconstruction recall R :=… view at source ↗
Figure 4
Figure 4. Wasted shots versus total shot count. Error bars show standard deviation across 10 experiments per shot count. The relatively constant wasted shot fraction indi￾cates that noise sources scale proportionally with measure￾ment budget. noise-induced empty voxel detections, since all A atoms share equal probability 1/A under the ideal equal super￾position, while noise is spread across the remaining C−A unoccupied states… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 5 canonical work pages

  1. [1]

    Estimation of the size of drug- like chemical space based on GDB-17 data

    Pavel G Polishchuk, Timur I Madzhidov, and Alexandre Varnek. “Estimation of the size of drug- like chemical space based on GDB-17 data”. In: Journal of computer-aided molecular design27.8 (2013), pp. 675–679

    2013

  2. [2]

    Neural message passing for quantum chemistry

    Justin Gilmer et al. “Neural message passing for quantum chemistry”. In:International conference on machine learning (ICML)(2017), pp. 1263– 1272

    2017

  3. [3]

    Convolutional networks on graphs for learning molecular fingerprints

    David K Duvenaud et al. “Convolutional networks on graphs for learning molecular fingerprints”. In: Advances in neural information processing systems 28 (2015)

    2015

  4. [4]

    Automatic chemi- cal design using a data-driven continuous represen- tation of molecules

    Rafael G´ omez-Bombarelli et al. “Automatic chemi- cal design using a data-driven continuous represen- tation of molecules”. In:ACS central science4.2 (2018), pp. 268–276. 13

    2018

  5. [5]

    Grammar variational autoen- coder

    Matt J Kusner, Brooks Paige, and Jos´ e Miguel Hern´ andez-Lobato. “Grammar variational autoen- coder”. In:International conference on machine learning (ICML)(2017), pp. 1945–1954

    2017

  6. [6]

    Equivariant diffusion for molecule generation in 3d

    Emiel Hoogeboom et al. “Equivariant diffusion for molecule generation in 3d”. In:International conference on machine learning (ICML)(2022), pp. 8867–8887

    2022

  7. [7]

    Geometric latent diffusion mod- els for 3d molecule generation

    Minkai Xu et al. “Geometric latent diffusion mod- els for 3d molecule generation”. In:International conference on machine learning (ICML)(2023), pp. 38592–38610

    2023

  8. [8]

    Nielsen and Isaac L

    Michael A. Nielsen and Isaac L. Chuang.Quan- tum Computation and Quantum Information: 10th Anniversary Edition. Cambridge University Press, 2010

    2010

Show all 39 references
  1. [9]

    Quantum chemistry in the age of quantum computing

    Yudong Cao et al. “Quantum chemistry in the age of quantum computing”. In:Chemical Reviews 119.19 (2019), pp. 10856–10915.doi:10 . 1021 / acs.chemrev.8b00803

    2019

  2. [10]

    Quantum computational chemistry

    Sam McArdle et al. “Quantum computational chemistry”. In:Reviews of Modern Physics92.1 (2020), p. 015003

    2020

  3. [11]

    A Survey of Quan- tum Machine Learning: Foundations, Algorithms, Frameworks, Data and Applications

    Francesc Rodr ´ ıguez-D ´ ıaz et al. “A Survey of Quan- tum Machine Learning: Foundations, Algorithms, Frameworks, Data and Applications”. In:ACM Comput. Surv.58.4 (Oct. 2025), pp. 1–35.issn: 0360-0300.doi:10.1145/3764582

    2025 doi

  4. [12]

    Haimeng Zhao et al.Exponential quantum advan- tage in processing massive classical data. 2026. arXiv:2604 . 07639 [quant-ph].url:https : / / arxiv.org/abs/2604.07639

    2026 arXiv

  5. [13]

    Quantum autoencoders for effi- cient compression of quantum data

    Jonathan Romero, Jonathan P. Olson, and Al´ an Aspuru-Guzik. “Quantum autoencoders for effi- cient compression of quantum data”. In:Quantum Science and Technology2.4 (2017), p. 045001.doi: 10.1088/2058-9565/aa8072

    2017 doi

  6. [14]

    Parameterized quantum circuits as machine learning models

    Marcello Benedetti et al. “Parameterized quantum circuits as machine learning models”. In:Quantum Science and Technology4.4 (2019), p. 043001.doi: 10.1088/2058-9565/ab4eb5

    2019 doi

  7. [15]

    QV AE- Mole: The Quantum V AE with Spherical Latent Variable Learning for 3-D Molecule Generation

    Huaijin Wu, Xinyu Ye, and Junchi Yan. “QV AE- Mole: The Quantum V AE with Spherical Latent Variable Learning for 3-D Molecule Generation”. In:The Thirty-eighth Annual Conference on Neu- ral Information Processing Systems. 2024

    2024

  8. [16]

    Experimental Single-Setting Quantum State Tomography

    Roman Stricker et al. “Experimental Single-Setting Quantum State Tomography”. In:PRX Quan- tum3 (4 Oct. 2022), p. 040310.doi:10 . 1103 / PRXQuantum.3.040310

    2022

  9. [17]

    A tailor-made quantum state tomography approach

    D. Binosi et al. “A tailor-made quantum state tomography approach”. In:APL Quantum1.3 (2024), p. 036112.doi:10.1063/5.0219143

    2024 doi

  10. [18]

    Yi Pan et al.MolQAE: Quantum Autoencoder for Molecular Representation Learning. 2025. arXiv: 2505.01875 [quant-ph].url:https://arxiv. org/abs/2505.01875

    2025 arXiv

  11. [19]

    Noisy intermediate-scale quantum algorithm for semidefinite program- ming

    Kishor Bharti et al. “Noisy intermediate-scale quantum algorithm for semidefinite program- ming”. In:Phys. Rev. A105 (5 May 2022), p. 052445.doi:10.1103/PhysRevA.105.052445

    2022 doi

  12. [20]

    Quantum Computing in the NISQ era and beyond

    John Preskill. “Quantum Computing in the NISQ era and beyond”. In:Quantum2 (Aug. 2018), p. 79. issn: 2521-327X.doi:10.22331/q-2018-08-06- 79

    2018 doi

  13. [21]

    Maria Schuld and Francesco Petruccione.Machine Learning with Quantum Computers. 2nd. Springer, 2021

    2021

  14. [22]

    Robust data en- codings for quantum classifiers

    Ryan LaRose and Brian Coyle. “Robust data en- codings for quantum classifiers”. In:Physical Re- view A102.3 (2020), p. 032420

    2020

  15. [23]

    Data encoding patterns for quantum computing

    Manuela Weigold et al. “Data encoding patterns for quantum computing”. In:Proceedings of the 27th Conference on Pattern Languages of Pro- grams. PLoP ’20. Virtual Event: The Hillside Group, 2022.isbn: 9781941652169

    2022

  16. [24]

    SMILES, a chemical lan- guage and information system. 1. Introduction to methodology and encoding rules

    David Weininger. “SMILES, a chemical lan- guage and information system. 1. Introduction to methodology and encoding rules”. In:Journal of Chemical Information and Computer Sciences28.1 (1988), pp. 31–36.doi:10.1021/ci00057a005

    1988 doi

  17. [25]

    Predicting many properties of a quantum system from very few measurements

    Hsin-Yuan Huang, Richard Kueng, and John Preskill. “Predicting many properties of a quantum system from very few measurements”. In:Nature Physics16.10 (June 2020), pp. 1050–1057.issn: 1745-2481.doi:10.1038/s41567-020-0932-7

    2020 doi

  18. [26]

    Measurement optimization in the variational quantum eigensolver using a minimum clique cover

    Vladyslav Verteletskyi, Tzu-Ching Yen, and Ar- tur F. Izmaylov. “Measurement optimization in the variational quantum eigensolver using a minimum clique cover”. In:The Journal of Chemical Physics 152.12 (Mar. 2020), p. 124114.issn: 0021-9606. doi:10.1063/1.5141458

    2020 doi

  19. [27]

    Neural-network quantum state tomography

    Giacomo Torlai et al. “Neural-network quantum state tomography”. In:Nature Physics14.5 (2018), pp. 447–450

    2018

  20. [28]

    Quantum circuits for isome- tries

    Raban Iten et al. “Quantum circuits for isome- tries”. In:Phys. Rev. A93 (3 Mar. 2016), p. 032318. doi:10.1103/PhysRevA.93.032318

    2016 doi

  21. [29]

    The quantum measurement problem: a review of recent trends

    Anderson A. Tomaz, Rafael S. Mattos, and Mario Barbatti. “The quantum measurement problem: a review of recent trends”. In:Philosophical Maga- zine(Dec. 2025), pp. 1–38.issn: 1478-6443.doi: 10.1080/14786435.2025.2601922. 14

    2025

  22. [30]

    Problems and Snapshots from the World of Prob- ability

    Gunnar Blom, Lars Holst, and Dennis Sandell. Problems and Snapshots from the World of Prob- ability. eng. New York, NY: Springer New York, 1994.isbn: 1-4612-4304-1

    1994

  23. [31]

    Qiskit Community.Qiskit: An Open-Source Frame- work for Quantum Computing. Mar. 2017.doi:10. 5281 / zenodo . 2562110.url:https : / / github . com/Qiskit/qiskit

    2017

  24. [32]

    A quantum engineer’s guide to superconducting qubits

    P. Krantz et al. “A quantum engineer’s guide to superconducting qubits”. In:Applied Physics Re- views6.2 (June 2019), p. 021318.issn: 1931-9401. doi:10.1063/1.5089550

    2019 doi

  25. [33]

    Quantum generative adversarial learning

    Seth Lloyd and Christian Weedbrook. “Quantum generative adversarial learning”. In:Physical Re- view Letters121.4 (2018), p. 040502

    2018

  26. [34]

    Quantum chemistry structures and properties of 134 kilo molecules

    Raghunathan Ramakrishnan et al. “Quantum chemistry structures and properties of 134 kilo molecules”. In:Scientific Data1 (2014), p. 140022. doi:10.1038/sdata.2014.22

    2014 doi

  27. [35]

    ZINC – A Free Database of Commercially Available Com- pounds for Virtual Screening

    John J. Irwin and Brian K. Shoichet. “ZINC – A Free Database of Commercially Available Com- pounds for Virtual Screening”. In:Journal of Chemical Information and Modeling45.1 (2005), pp. 177–182.doi:10.1021/ci049714+

    2005 doi

  28. [36]

    Exploiting symmetry in variational quantum machine learning

    Johannes Jakob Meyer et al. “Exploiting symmetry in variational quantum machine learning”. In:PRX Quantum4.1 (2023), p. 010328

    2023

  29. [37]

    Theory for equivariant quantum neural networks

    Quynh T. Nguyen et al. “Theory for equivariant quantum neural networks”. In:PRX Quantum5.2 (2024), p. 020328

    2024

  30. [38]

    Better informed distance geometry: using what we know to improve conformation generation

    Sereina Riniker and Gregory A. Landrum. “Better informed distance geometry: using what we know to improve conformation generation”. In:Jour- nal of Chemical Information and Modeling55.12 (2015), pp. 2562–2574

    2015

  31. [39]

    Merck molecular force field. I. Basis, form, scope, parameterization, and perfor- mance of MMFF94

    Thomas A Halgren. “Merck molecular force field. I. Basis, form, scope, parameterization, and perfor- mance of MMFF94”. In:Journal of Computational Chemistry17.5-6 (1996), pp. 490–519. 15

    1996

Pith tools

Reviewed July 30, 2026 · model on record in the stance chip above.