pith. sign in

arxiv: 1906.10033 · v1 · pith:PQIJXI7Pnew · submitted 2019-06-24 · ⚛️ physics.chem-ph · stat.ML

Unifying machine learning and quantum chemistry -- a deep neural network for molecular wavefunctions

K. T. Sch\"utt , M. Gastegger , A. Tkatchenko , K.-R. M\"uller , R. J. Maurer This is my paper

Pith reviewed 2026-05-25 16:55 UTC · model grok-4.3

classification ⚛️ physics.chem-ph stat.ML
keywords deep learningquantum chemistrymolecular wavefunctionselectronic structureneural networksatomic orbitalsproperty predictioninverse design
0
0 comments X

The pith

A deep neural network predicts the quantum mechanical wavefunction of molecules in a local atomic orbital basis.

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

The paper introduces a machine learning model trained to output the electronic wavefunction of a molecule directly from its atomic geometry. This wavefunction, expressed in a local basis of atomic orbitals, serves as the starting point from which energies, forces, and all other ground-state properties are derived. The model operates at speeds comparable to classical force fields while remaining fully differentiable with respect to atomic positions. By retaining explicit access to the electronic degrees of freedom, the approach aims to support both large-scale screening and detailed chemical analysis that standard property-prediction models cannot provide. The authors illustrate the framework on example molecules to indicate its utility for tasks such as inverse molecular design.

Core claim

The central claim is that a deep learning framework can predict the quantum mechanical wavefunction in a local basis of atomic orbitals, from which all other ground-state properties can be derived. This approach retains full access to the electronic structure via the wavefunction at force field-like efficiency and captures quantum mechanics in an analytically differentiable representation.

What carries the argument

Deep neural network that maps molecular geometry to wavefunction coefficients in a local atomic orbital basis.

Load-bearing premise

The neural network trained on quantum chemical calculations can accurately predict the wavefunction for molecules outside the training set so that properties derived from the prediction remain reliable.

What would settle it

Train the network on one set of molecules, then compare its predicted wavefunction and all derived properties against reference quantum chemistry results on a held-out molecule; large systematic deviations would falsify the central claim.

Figures

Figures reproduced from arXiv: 1906.10033 by A. Tkatchenko, K.-R. M\"uller, K. T. Sch\"utt, M. Gastegger, R. J. Maurer.

Figure 1
Figure 1. Figure 1: FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: b depicts the forces the different MOs exert onto the hydrogen atom exchanged during the proton trans￾fer. All forces are projected onto the reaction coordinate, where positive values correspond to a force driving the proton towards the product state. In the initial config￾uration I, most forces lead to attraction of the hydrogen atom to the right oxygen. In the intermediate config￾uration II, orbital rear… view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4 [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
read the original abstract

Machine learning advances chemistry and materials science by enabling large-scale exploration of chemical space based on quantum chemical calculations. While these models supply fast and accurate predictions of atomistic chemical properties, they do not explicitly capture the electronic degrees of freedom of a molecule, which limits their applicability for reactive chemistry and chemical analysis. Here we present a deep learning framework for the prediction of the quantum mechanical wavefunction in a local basis of atomic orbitals from which all other ground-state properties can be derived. This approach retains full access to the electronic structure via the wavefunction at force field-like efficiency and captures quantum mechanics in an analytically differentiable representation. On several examples, we demonstrate that this opens promising avenues to perform inverse design of molecular structures for target electronic property optimisation and a clear path towards increased synergy of machine learning and quantum chemistry.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 0 minor

Summary. The manuscript presents a deep neural network framework to predict the quantum mechanical wavefunction of a molecule in a local atomic-orbital basis; all ground-state properties are then obtained by post-processing this wavefunction, achieving force-field-like efficiency while retaining analytic access to the electronic structure.

Significance. If the generalization claim holds with quantitative accuracy, the work would provide a differentiable, wavefunction-level interface between machine learning and quantum chemistry, enabling inverse design and large-scale reactive simulations that are currently inaccessible to either pure ML property predictors or conventional QC methods.

major comments (1)
  1. [Abstract] Abstract: the statement that 'demonstrations on several examples support the claim' is presented without any quantitative accuracy metrics, validation protocols, baseline comparisons, or error analysis for either the predicted wavefunction or the derived properties. This information is load-bearing for the central assertion that the model generalizes to unseen molecules while preserving reliable derived quantities.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the positive assessment of the work's potential significance and for the constructive comment on the abstract. We address the point below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the statement that 'demonstrations on several examples support the claim' is presented without any quantitative accuracy metrics, validation protocols, baseline comparisons, or error analysis for either the predicted wavefunction or the derived properties. This information is load-bearing for the central assertion that the model generalizes to unseen molecules while preserving reliable derived quantities.

    Authors: We agree that the abstract would benefit from a concise quantitative statement to support the generalization claim. In the revised manuscript we will update the final sentence of the abstract to include brief, representative metrics drawn from the results (e.g., mean absolute errors on wavefunction coefficients and on derived energies/forces for held-out molecules, together with a short description of the train/test protocol). These numbers will be chosen to be representative of the quantitative accuracy reported in the main text while remaining within the abstract's length constraints. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper trains a neural network on external quantum chemical reference calculations to predict the wavefunction coefficients in a local atomic-orbital basis; derived properties are then obtained by standard quantum-chemical post-processing of that predicted wavefunction. No equation in the supplied material defines a target quantity in terms of itself, renames a fitted parameter as a prediction, or relies on a load-bearing self-citation whose own justification is internal to the present work. The generalization premise is an empirical claim that can be tested against held-out QC data and is therefore not circular by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Review is based solely on the abstract; no specific free parameters, invented entities, or detailed axioms are described in the provided text.

axioms (1)
  • domain assumption Quantum chemical calculations supply accurate wavefunction training targets
    The framework is trained on data from quantum chemical calculations.

pith-pipeline@v0.9.0 · 5691 in / 1072 out tokens · 27796 ms · 2026-05-25T16:55:15.509596+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

48 extracted references · 48 canonical work pages · 4 internal anchors

  1. [1]

    Behler \ and\ author M

    author author J. Behler \ and\ author M. Parrinello ,\ @noop journal journal Phys. Rev. Lett. \ volume 98 ,\ pages 146401 ( year 2007 ) NoStop

    work page 2007

  2. [2]

    author author B. J. \ Braams \ and\ author J. M. \ Bowman ,\ @noop journal journal International Reviews in Physical Chemistry \ volume 28 ,\ pages 577 ( year 2009 ) NoStop

    work page 2009

  3. [3]

    author author A. P. \ Bart \'o k , author M. C. \ Payne , author R. Kondor , \ and\ author G. Cs \'a nyi ,\ @noop journal journal Phys. Rev. Lett. \ volume 104 ,\ pages 136403 ( year 2010 ) NoStop

    work page 2010

  4. [4]

    author author J. S. \ Smith , author O. Isayev , \ and\ author A. E. \ Roitberg ,\ @noop journal journal Chem. Sci. \ volume 8 ,\ pages 3192 ( year 2017 ) NoStop

    work page 2017

  5. [5]

    author author E. V. \ Podryabinkin \ and\ author A. V. \ Shapeev ,\ @noop journal journal Computational Materials Science \ volume 140 ,\ pages 171 ( year 2017 ) NoStop

    work page 2017

  6. [6]

    author author E. V. \ Podryabinkin , author E. V. \ Tikhonov , author A. V. \ Shapeev , \ and\ author A. R. \ Oganov ,\ @noop journal journal Physical Review B \ volume 99 ,\ pages 064114 ( year 2019 ) NoStop

    work page 2019

  7. [7]

    u tt , \ and\ author K.-R. \ M \

    author author S. Chmiela , author A. Tkatchenko , author H. E. \ Sauceda , author I. Poltavsky , author K. T. \ Sch \"u tt , \ and\ author K.-R. \ M \"u ller ,\ @noop journal journal Sci. Adv. \ volume 3 ,\ pages e1603015 ( year 2017 ) NoStop

    work page 2017

  8. [8]

    Chmiela , author H

    author author S. Chmiela , author H. E. \ Sauceda , author K.-R. \ M \"u ller , \ and\ author A. Tkatchenko ,\ @noop journal journal Nature communications \ volume 9 ,\ pages 3887 ( year 2018 ) NoStop

    work page 2018

  9. [9]

    Ryczko , author D

    author author K. Ryczko , author D. Strubbe , \ and\ author I. Tamblyn ,\ http://arxiv.org/abs/1811.08928 \ ( year 2018 ) ,\ http://arxiv.org/abs/1811.08928 arXiv:1811.08928 NoStop

    work page arXiv 2018

  10. [10]

    Brockherde , author L

    author author F. Brockherde , author L. Voigt , author L. Li , author M. E. \ Tuckerman , author K. Burke , \ and\ author K.-R. \ M \"u ller ,\ @noop journal journal Nat. Commun. \ volume 8 ,\ pages 872 ( year 2017 ) NoStop

    work page 2017

  11. [11]

    author author D. M. \ Wilkins , author A. Grisafi , author Y. Yang , author K. U. \ Lao , author R. A. \ DiStasio , \ and\ author M. Ceriotti ,\ 10.1073/pnas.1816132116 journal journal Proceedings of the National Academy of Sciences of the United States of America \ volume 116 ,\ pages 3401 ( year 2019 ) NoStop

    work page doi:10.1073/pnas.1816132116 2019

  12. [12]

    Gastegger , author J

    author author M. Gastegger , author J. Behler , \ and\ author P. Marquetand ,\ @noop journal journal Chem. Sci. \ volume 8 ,\ pages 6924 ( year 2017 ) NoStop

    work page 2017

  13. [13]

    Rupp , author A

    author author M. Rupp , author A. Tkatchenko , author K.-R. \ M \"u ller , \ and\ author O. A. \ Von Lilienfeld ,\ @noop journal journal Phys. Rev. Lett. \ volume 108 ,\ pages 058301 ( year 2012 ) NoStop

    work page 2012

  14. [14]

    Eickenberg , author G

    author author M. Eickenberg , author G. Exarchakis , author M. Hirn , \ and\ author S. Mallat ,\ in\ @noop booktitle Advances in Neural Information Processing Systems 30 \ ( publisher Curran Associates, Inc. ,\ year 2017 )\ pp.\ pages 6543--6552 NoStop

    work page 2017

  15. [15]

    author author O. A. \ von Lilienfeld ,\ 10.1002/anie.201709686 journal journal Angewandte Chemie International Edition \ volume 57 ,\ pages 4164 ( year 2018 ) NoStop

    work page doi:10.1002/anie.201709686 2018

  16. [16]

    Gilmer , author S

    author author J. Gilmer , author S. S. \ Schoenholz , author P. F. \ Riley , author O. Vinyals , \ and\ author G. E. \ Dahl ,\ in\ @noop booktitle Proceedings of the 34th International Conference on Machine Learning ,\ pp.\ pages 1263--1272 NoStop

    work page

  17. [17]

    Jha , author L

    author author D. Jha , author L. Ward , author A. Paul , author W.-k. \ Liao , author A. Choudhary , author C. Wolverton , \ and\ author A. Agrawal ,\ @noop journal journal Scientific reports \ volume 8 ,\ pages 17593 ( year 2018 ) NoStop

    work page 2018

  18. [18]

    author author J. R. \ Kitchin ,\ 10.1038/s41929-018-0056-y journal journal Nature Catalysis \ volume 1 ,\ pages 230 ( year 2018 ) NoStop

    work page doi:10.1038/s41929-018-0056-y 2018

  19. [19]

    Maryasin , author P

    author author B. Maryasin , author P. Marquetand , \ and\ author N. Maulide ,\ 10.1002/anie.201803562 journal journal Angewandte Chemie International Edition \ volume 57 ,\ pages 6978 ( year 2018 ) NoStop

    work page doi:10.1002/anie.201803562 2018

  20. [20]

    Li , author C

    author author H. Li , author C. Collins , author M. Tanha , author G. J. \ Gordon , \ and\ author D. J. \ Yaron ,\ 10.1021/acs.jctc.8b00873 journal journal Journal of Chemical Theory and Computation \ volume 14 ,\ pages 5764 ( year 2018 ) ,\ note pMID: 30351008 NoStop

    work page doi:10.1021/acs.jctc.8b00873 2018

  21. [21]

    Hegde \ and\ author R

    author author G. Hegde \ and\ author R. C. \ Bowen ,\ @noop journal journal Scientific reports \ volume 7 ,\ pages 42669 ( year 2017 ) NoStop

    work page 2017

  22. [22]

    Grisafi , author D

    author author A. Grisafi , author D. M. \ Wilkins , author G. Cs \' a nyi , \ and\ author M. Ceriotti ,\ 10.1103/PhysRevLett.120.036002 journal journal Physical Review Letters \ volume 120 ,\ pages 036002 ( year 2018 ) NoStop

    work page doi:10.1103/physrevlett.120.036002 2018

  23. [23]

    Tensor field networks: Rotation- and translation-equivariant neural networks for 3D point clouds

    author author N. Thomas , author T. Smidt , author S. Kearnes , author L. Yang , author L. Li , author K. Kohlhoff , \ and\ author P. Riley ,\ http://arxiv.org/abs/1802.08219 \ ( year 2018 ) ,\ http://arxiv.org/abs/1802.08219 arXiv:1802.08219 NoStop

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  24. [24]

    u tt , author H. E. \ Sauceda , author P.-J. \ Kindermans , author A. Tkatchenko , \ and\ author K.-R. \ M \

    author author K. T. \ Sch \"u tt , author H. E. \ Sauceda , author P.-J. \ Kindermans , author A. Tkatchenko , \ and\ author K.-R. \ M \"u ller ,\ @noop journal journal The Journal of Chemical Physics \ volume 148 ,\ pages 241722 ( year 2018 a ) NoStop

    work page 2018

  25. [25]

    author author C. J. \ Cramer ,\ @noop title Essentials of computational chemistry: theories and models \ ( publisher John Wiley & Sons ,\ year 2004 ) NoStop

    work page 2004

  26. [26]

    u tt , author F. Arbabzadah , author S. Chmiela , author K.-R. \ M \

    author author K. T. \ Sch \"u tt , author F. Arbabzadah , author S. Chmiela , author K.-R. \ M \"u ller , \ and\ author A. Tkatchenko ,\ @noop journal journal Nat. Commun. \ volume 8 ,\ eid 13890 ( year 2017 ) NoStop

    work page 2017

  27. [27]

    u tt , author P.-J. \ Kindermans , author H. E. \ Sauceda , author S. Chmiela , author A. Tkatchenko , \ and\ author K.-R. \ M\

    author author K. T. \ Sch\" u tt , author P.-J. \ Kindermans , author H. E. \ Sauceda , author S. Chmiela , author A. Tkatchenko , \ and\ author K.-R. \ M\" u ller ,\ in\ @noop booktitle Advances in Neural Information Processing Systems 30 \ ( year 2017 )\ pp.\ pages 992--1002 NoStop

    work page 2017

  28. [28]

    author author J. P. \ Perdew , author K. Burke , \ and\ author M. Ernzerhof ,\ @noop journal journal Phys. Rev. Lett. \ volume 77 ,\ pages 3865 ( year 1996 a ) NoStop

    work page 1996

  29. [29]

    Bach , author A

    author author S. Bach , author A. Binder , author G. Montavon , author F. Klauschen , author K.-R. \ M \"u ller , \ and\ author W. Samek ,\ 10.1371/journal.pone.0130140 journal journal PloS one \ volume 10 ,\ pages e0130140 ( year 2015 ) NoStop

    work page doi:10.1371/journal.pone.0130140 2015

  30. [30]

    Montavon , author W

    author author G. Montavon , author W. Samek , \ and\ author K.-R. \ M\"uller ,\ https://doi.org/10.1016/j.dsp.2017.10.011 journal journal Digital Signal Processing \ volume 73 ,\ pages 1 ( year 2018 ) NoStop

    work page doi:10.1016/j.dsp.2017.10.011 2017

  31. [31]

    u tt , author M. Alber , author K.-R. \ M \

    author author P.-J. \ Kindermans , author K. T. \ Sch \"u tt , author M. Alber , author K.-R. \ M \"u ller , author D. Erhan , author B. Kim , \ and\ author S. D \"a hne ,\ in\ @noop booktitle International Conference on Learning Representations (ICLR) \ ( year 2018 ) NoStop

    work page 2018

  32. [32]

    Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV)

    author author B. Kim , author M. Wattenberg , author J. Gilmer , author C. Cai , author J. Wexler , author F. Viegas , \ and\ author R. Sayres ,\ @noop journal journal arXiv preprint arXiv:1711.11279 \ ( year 2017 ) NoStop

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  33. [33]

    a ldchen , author A. Binder , author G. Montavon , author W. Samek , \ and\ author K.-R. \ M \

    author author S. Lapuschkin , author S. W \"a ldchen , author A. Binder , author G. Montavon , author W. Samek , \ and\ author K.-R. \ M \"u ller ,\ @noop journal journal Nature communications \ volume 10 ,\ pages 1096 ( year 2019 ) NoStop

    work page 2019

  34. [34]

    De , author A

    author author S. De , author A. P. \ Bart \'o k , author G. Cs \'a nyi , \ and\ author M. Ceriotti ,\ @noop journal journal Physical Chemistry Chemical Physics \ volume 18 ,\ pages 13754 ( year 2016 ) NoStop

    work page 2016

  35. [35]

    Quantum-chemical insights from interpretable atomistic neural networks

    author author K. T. \ Sch \"u tt , author M. Gastegger , author A. Tkatchenko , \ and\ author K.-R. \ M \"u ller ,\ @noop journal journal arXiv preprint arXiv:1806.10349 \ ( year 2018 b ) NoStop

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  36. [36]

    To Baben , author J

    author author M. To Baben , author J. Achenbach , \ and\ author O. Von Lilienfeld ,\ @noop journal journal The Journal of chemical physics \ volume 144 ,\ pages 104103 ( year 2016 ) NoStop

    work page 2016

  37. [37]

    You , author B

    author author J. You , author B. Liu , author Z. Ying , author V. Pande , \ and\ author J. Leskovec ,\ in\ @noop booktitle Advances in Neural Information Processing Systems \ ( year 2018 )\ pp.\ pages 6410--6421 NoStop

    work page 2018

  38. [38]

    author author R. J. \ Maurer , author M. Askerka , author V. S. \ Batista , \ and\ author J. C. \ Tully ,\ 10.1103/PhysRevB.94.115432 journal journal Phys. Rev. B \ volume 94 ,\ pages 115432 ( year 2016 ) NoStop

    work page doi:10.1103/physrevb.94.115432 2016

  39. [39]

    author author W. C. \ Lu , author C. Z. \ Wang , author M. W. \ Schmidt , author L. Bytautas , author K. M. \ Ho , \ and\ author K. Ruedenberg ,\ 10.1063/1.1638731 journal journal The Journal of chemical physics \ volume 120 ,\ pages 2629 ( year 2004 ) NoStop

    work page doi:10.1063/1.1638731 2004

  40. [40]

    Neese ,\ 10.1002/wcms.81 journal journal WIREs Comput

    author author F. Neese ,\ 10.1002/wcms.81 journal journal WIREs Comput. Mol. Sci. \ volume 2 ,\ pages 73 ( year 2012 ) NoStop

    work page doi:10.1002/wcms.81 2012

  41. [41]

    Weigend \ and\ author R

    author author F. Weigend \ and\ author R. Ahlrichs ,\ 10.1039/B508541A journal journal Phys. Chem. Chem. Phys. \ volume 7 ,\ pages 3297 ( year 2005 ) NoStop

    work page doi:10.1039/b508541a 2005

  42. [42]

    Pulay ,\ https://doi.org/10.1016/0009-2614(80)80396-4 journal journal Chemical Physics Letters \ volume 73 ,\ pages 393 ( year 1980 ) NoStop

    author author P. Pulay ,\ https://doi.org/10.1016/0009-2614(80)80396-4 journal journal Chemical Physics Letters \ volume 73 ,\ pages 393 ( year 1980 ) NoStop

    work page doi:10.1016/0009-2614(80)80396-4 1980

  43. [43]

    author author J. P. \ Perdew , author K. Burke , \ and\ author M. Ernzerhof ,\ @noop journal journal Phys. Rev. Lett. \ volume 77 ,\ pages 3865 ( year 1996 b ) NoStop

    work page 1996

  44. [44]

    u tt , author P. Kessel , author M. Gastegger , author K. Nicoli , author A. Tkatchenko , \ and\ author K.-R. \ M \

    author author K. Sch \"u tt , author P. Kessel , author M. Gastegger , author K. Nicoli , author A. Tkatchenko , \ and\ author K.-R. \ M \"u ller ,\ @noop journal journal Journal of chemical theory and computation \ volume 15 ,\ pages 448 ( year 2018 c ) NoStop

    work page 2018

  45. [45]

    Bussi \ and\ author M

    author author G. Bussi \ and\ author M. Parrinello ,\ 10.1103/PhysRevE.75.056707 journal journal Phys. Rev. E \ volume 75 ,\ pages 056707 ( year 2007 ) NoStop

    work page doi:10.1103/physreve.75.056707 2007

  46. [46]

    Sutskever , author J

    author author I. Sutskever , author J. Martens , \ and\ author G. E. \ Hinton ,\ in\ @noop booktitle Proceedings of the 28th International Conference on Machine Learning \ ( year 2011 )\ pp.\ pages 1017--1024 NoStop

    work page 2011

  47. [47]

    Schober , author K

    author author C. Schober , author K. Reuter , \ and\ author H. Oberhofer ,\ 10.1063/1.4940920 journal journal The Journal of Chemical Physics \ volume 144 ,\ pages 054103 ( year 2016 ) NoStop

    work page doi:10.1063/1.4940920 2016

  48. [48]

    author author D. P. \ Kingma \ and\ author J. Ba ,\ @noop journal journal arXiv preprint arXiv:1412.6980 \ ( year 2014 ) NoStop

    work page internal anchor Pith review Pith/arXiv arXiv 2014