Position: The Inevitable Transition to Machine Learning in Quantum Chemistry

Chao-Ping Hsu; Karen Sargsyan

REVIEW 2 major objections 4 minor 98 references

Position: The Inevitable Transition to Machine Learning in Quantum Chemistry

T0 review · 2 major / 4 minor · reviewed 2026-08-02 · deepseek-v4-flash

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

Pith's one-line read Machine learning should guide quantum chemistry's next stage, not from a proof of impossibility but as the rational choice under uncertainty.

desk verdict A sincere, well-written position paper whose decision-theoretic core is defensible, but whose quantitative 'compression ratio' support is shaky and whose title overclaims. read the letter →

arxiv 2607.18281 v2 pith:RBREF3LU submitted 2026-06-30 physics.chem-ph cs.LGquant-ph

Karen Sargsyan , Chao-Ping Hsu This is my paper · ORCID

classification physics.chem-phcs.LGquant-ph

keywords machinelearningquantumchemistrydensityfunctionaltheoryspace-timetradeoffQMA-hardnessneuralnetworkpotentialspositionpaperchemicalspace

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

This position paper tries to establish that machine learning, rather than hand-crafted analytical approximation, should be the primary direction for developing new quantum chemistry methods. It explicitly disclaims logical necessity: exact electronic structure is QMA-hard, but real chemistry might be easier; the argument is decision-theoretic. Because a trained network stores precomputed 'advice' in its weights, ML implements the classic space-time tradeoff, replacing intractable recomputation with learned, stored knowledge. The paper reads fifty years of DFT and wavefunction development as small-scale hand-crafted machine learning that has exhausted what human intuition can explore, citing stagnant electron densities, unresolved strong correlation, and hundreds of functionals without convergence toward the exact functional. If the position holds, the practical consequence is that funding, software, benchmarks, and training in quantum chemistry should be reorganized around ML.

What carries the argument

The Space-Time Tradeoff, formalized through non-uniform computation (the complexity class P/poly). A trained neural network is read as a circuit whose weights are precomputed 'advice'; training discovers specialized knowledge, converting an intractable uniform computation into stored memory plus efficient evaluation. The quantitative engine is the compression ratio: roughly 10^7 parameters reaching useful accuracy across roughly 10^33–10^60 molecules implies compression of 10^26–10^53, which the paper argues is possible only because chemistry has exploitable structure — locality, smoothness, and symmetry — that learned representations capture rather than memorize.

What would settle it

A direct falsifier, acknowledged by the paper: a non-ML approximation achieving chemical accuracy across broad chemical space with favorable scaling and systematic treatment of strong correlation, without extensive parameterization against reference data. A sharper probe: find a broad class of chemically simple systems outside any reasonable training region where a 10^7-parameter universal potential's error grows superlinearly with distance to the training data, breaking the compression prediction.

Watch

Extended reading notes

Core claim

The central claim is that, given uncertainty about the true complexity of chemically relevant problems, the Space-Time Tradeoff implemented via machine learning is the rational path forward. Training a neural network is non-uniform computation: the network is a circuit, the trained weights are stored advice, and the model compresses an estimated 10^33–10^60 chemical space into roughly 10^6–10^7 parameters. The paper's distinctive move is decision-theoretic rather than complexity-theoretic: ML degrades gracefully across scenarios — if chemistry is hard, ML becomes necessary; if it is easy but lacks simple analytical solutions, ML discovers specialized approximations; if simple formulas exist,

Load-bearing premise

The argument collapses if chemistry's exploitable structure — locality, smoothness, and symmetry — is not enough for a finite-parameter model to generalize across chemical space, because then the compression ratio is an artifact and a trained network is just a memorization device.

Editorial extensions

If this is right

If ML is the rational priority, funding and research infrastructure should shift toward datasets, equivariant architectures, uncertainty quantification, and benchmarking rather than incremental development of new functionals.
Quantum chemistry software packages should treat ML integration as a first-class capability; the DM21 adoption gap shows that benchmark success without software integration yields little practical progress.
Benchmark protocols should go beyond average accuracy to include out-of-distribution generalization and per-prediction reliability reporting, because silent failures are the main deployment risk.
Neural scaling laws imply the route to higher accuracy is predictable — larger models and more data — rather than waiting for conceptual breakthroughs in ansatz design.
Quantum computers will serve as reference-data generators for hard systems, while ML will be the generalization engine that makes that data useful for dynamics, thermodynamics, and materials discovery.

Reading between the lines

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

An implication the authors leave implicit: if the compression argument is sound, the field's data-generation hierarchy becomes strategic infrastructure — traditional high-accuracy methods matter primarily for the quality of the reference data they supply.
The argument is sharper than the paper's own falsifier: train a fixed architecture on a systematically starved subset of a well-characterized chemical class and measure error growth versus distance from the training region; a true compression of 10^26 predicts slow growth, while a sharp breakdown would falsify the locality/smoothness premise.
A corollary the paper does not state: evaluation culture should move from leaderboard averages to calibrated uncertainty plus adversarial out-of-distribution probe sets, since the decision-theoretic case depends on graceful degradation, not peak accuracy.
The position also implies an adoption roadmap: if software integration is the bottleneck, ML's success depends less on new architectures than on engineering — analytic gradients, SCF convergence, periodic boundary conditions, and package availability.

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.

Desk Editor's Note

read the letter

Colleague, you should know this paper is a sincere, well-cited position statement, not a technical breakthrough. The real novelty is the framing: traditional DFT and WFT development as 'hand-crafted machine learning'—feature engineering, hyperparameter search, small training sets. That lens is productive, and the authors use it fairly, even dedicating an appendix to the DM21 adoption gap and another to out-of-distribution failure modes. The AlphaFold analogy is apt but not new; the QMA-hardness and P/poly background is correct but standard.

The central argument is decision-theoretic: since we don't know whether chemically relevant problems are truly hard or just analytically messy, ML is the rational bet because it 'degrades gracefully.' That's plausible but asserted, not demonstrated. The asymmetry claim does a lot of work. The paper's weakest point, though, is the compression ratio in Appendix A.3. Dividing the size of drug-like chemical space (10^33–10^60) by a model's parameter count (10^6–10^7) and calling the result 'striking' conflates a discrete enumeration of molecules with the capacity of a continuous function. A neural network doesn't store molecules; it interpolates a manifold. Recompute against training configurations (1.58M for MPtrj) and the ratio collapses to about 10^1. The paper itself says 'the model does not memorize chemical space,' so using a memorization-style ratio as quantitative support is internally inconsistent. That claim should be withdrawn or recharacterized.

The bigger case doesn't collapse without it. The stagnation of functional development, the strong-correlation wall, and the scaling results are real and reasonably cited. The authors are honest about uncertainty and about ML's own limitations, both in the main text and appendices. But the title says 'Inevitable' while the body explicitly says there is no logical necessity—that mismatch should be fixed. The 'hand-crafted ML' analogy, while useful, is also pushed a bit far: traditional functionals have interpretable physical content that small learned models don't, and the analogy risks underselling that difference.

I'd bring this to a reading group as an opinion piece worth arguing with. If I were an editor, I'd send it to peer review—referees should push on the compression claims and the 'degrade gracefully' premise, and the paper would be better for it. It's the kind of strategically important but non-technical argument that deserves serious scrutiny, not desk rejection. I'd suggest the authors cut the compression-ratio argument to a qualitative remark and soften the title before publication.

Referee Report

2 major / 4 minor

Summary. This position paper argues that machine learning should be the primary direction for developing new quantum chemistry approximations, explicitly disclaiming logical necessity and instead advancing a decision-theoretic argument: given QMA-hardness of the general many-body problem and the observed saturation of hand-crafted DFT and wavefunction methods, the Space-Time Tradeoff implemented via ML is presented as the rational path forward. The paper reframes traditional method development as 'hand-crafted machine learning', reviews supporting empirical evidence (AlphaFold, DM21, MACE, neural wavefunctions), discusses challenges and adoption gaps, and proposes concrete research-priority shifts. The main quantitative support is the compression-ratio argument in Section 2.2 and Appendix A.3.

Significance. If the position holds, it implies a strategic reallocation of research effort in quantum chemistry, which is a high-impact claim. The paper is unusually honest for a position piece: it states what would falsify its position, acknowledges that the typical-case objection cannot be dismissed a priori, and includes detailed limitations appendices. Its strengths include the explicit framing of the decision-theoretic (rather than logical-necessity) case, the wide-ranging and recent literature coverage, and the concrete discussion of the adoption gap through the DM21 case study. However, the central quantitative pillar—the compression-ratio argument—is flawed as stated, and the 'graceful degradation' premise conflicts with the paper's own discussion of silent out-of-distribution failures. These need correction before the argument can carry the weight assigned to it.

major comments (2)

[§2.2, Appendix A.3] The paper's quantitative case rests on the claimed compression ratio of 10^26–10^53, obtained by dividing the estimated size of drug-like chemical space (10^33–10^60) by the parameter count of a modern ML potential (10^6–10^7). I do not think this inference is valid. A neural network is a continuous function; its parameter count bounds the complexity of the function class, not the number of molecules it can 'store' or distinguish. The model does not memorize 10^33 entries; it interpolates on a continuous manifold. Comparing a discrete enumeration of possible molecules with a parameter count is an apples-to-oranges comparison. If one instead used the 1.58M training configurations of MPtrj or the number of unique compositions, the 'compression ratio' would collapse to roughly 10^1–10^2. The paper itself acknowledges (A.3) that the model does not memorize chemical space, which directly unde
[§2.3] The decision-theoretic claim is the backbone of Position 1, but its key premise—'ML succeeds across a wide range of scenarios' and 'degrades gracefully'—is asserted rather than derived. The paper's own Appendix B.4 states that ML models can fail silently on out-of-distribution inputs, producing confident but incorrect predictions, and that reliable uncertainty quantification remains an open problem. This is the opposite of graceful degradation in the scenario where the test distribution shifts, which is precisely the regime relevant to 'chemically relevant problems' as an unknown structured subset. At minimum, the claim needs to be conditioned on progress in uncertainty quantification and OOD detection, or explicitly restricted to scenarios where the test distribution overlaps the training distribution. The AlphaFold analogy is an existence proof, not a general argument, and the 'if simp

minor comments (4)

[§6.4 / Appendix B.6] The DM21 case study is carefully done, but the generalization concern raised by Gerasimov et al. (training-set overlap) is not purely an 'engineering challenge' as the paper classifies it; it is a data and validation issue. The paper mentions this but could state more clearly that this challenge is distinct from SCF convergence, gradients, and software integration.
[Appendix A.2] The parameter-count estimate for MACE-MP-0 ('of order 10^6') is used in the central compression ratio. Please provide a precise citation or table entry for the parameter count, since the ratio is sensitive to this number.
[Appendix C.3] The statement that equivariance 'changes the exponent' of the error-versus-data curve is a strong claim based on a single comparison in NequIP. The caveats given are good, but the wording could be softened to 'in this comparison' to avoid overgeneralization.
[General] Typos and formatting: 'V ASP' in Appendix B.6 should be 'VASP'; 'Schr¨odinger' appears in the abstract and elsewhere with a LaTeX umlaut artifact; the reference list is long but a few entries (e.g., Gerasimov et al., Kirkpatrick et al. 2022) are cited only in appendix discussion and could be moved or cross-referenced.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the position is explicitly decision-theoretic and its supporting evidence is external; the single self-citation is peripheral.

full rationale

The paper's central claim is not derived from a fitted parameter or from a self-citation. Position 1 is framed as a decision-theoretic choice under uncertainty: 'Given uncertainty about the true complexity of chemically relevant problems, the Space-Time Tradeoff implemented via machine learning is the rational path forward.' The quantitative compression-ratio argument in Appendix A.3 divides externally published estimates of drug-like chemical space (Bohacek et al. 1996; Polishchuk et al. 2013) by externally reported MACE parameter counts (Batatia et al. 2025); even if that ratio is statistically contestable (parameters are not a count of stored molecules), it is not a reduction of the conclusion to the input. The paper explicitly disclaims memorization: 'The model does not memorize chemical space; it learns generalizable patterns.' The evidence for ML success (AlphaFold, MACE, DM21, FermiNet, Orbformer) comes from independent groups and is externally benchmarked. The 'hand-crafted machine learning' framing is an analogy, not a derivation, and no equation in the paper equates a claimed result with its own assumption. The only self-citation (Mazmanian et al. 2022, including author Sargsyan) appears in a list of application areas for quantum chemistry and is not load-bearing; per the review rules, it does not raise the circularity score. The paper also candidly reports challenges (e.g., DM21's training-set overlap critique from Gerasimov et al. 2022) in Appendix B.6, which further supports that the position is not being protected by circular argument.

Assumptions & free parameters 0 free parameters · 6 assumptions · 0 invented entities

The paper does not fit parameters or introduce new physical entities. It relies on established complexity results, empirical assessments of the field, and a decision-theoretic framing. The main load-bearing axioms are the existence of exploitable chemical structure and the assumption that ML generalizes across chemical space — both asserted rather than demonstrated.

assumptions (6)

domain assumption Ground-state energy of a general quantum system is QMA-hard; related complexity results (undecidability of spectral gap, NP-hardness of Hartree-Fock).
Invoked in Section 2.1 as the foundation for the intractability argument. These are established results from prior literature, not proved in this paper.
domain assumption Chemical space has exploitable structure — locality, smoothness, symmetry — that learned representations can capture.
Central to the Space-Time Tradeoff argument (Section 2.2, Appendix A). The paper assumes this structure sufficiently constrains the problem so that finite models can generalize.
ad hoc to paper A trained neural network is a non-uniform circuit with 'advice' in the P/poly sense, implementing a practical Space-Time Tradeoff.
The P/poly analogy is presented in Section 2.2 as a formal grounding, but the mapping from trained networks to advice circuits is an interpretive leap not supported by complexity theory.
domain assumption Traditional methods have demonstrably stalled: 400+ DFT functionals without convergence, density regression since 2000, strong correlation unsolved.
Empirical premise grounded in cited works (Medvedev et al., Mardirossian & Head-Gordon) but contested in the literature (Kepp, Gerasimov, Zhao). The paper acknowledges the contestation yet proceeds on this assumption.
domain assumption Neural scaling laws in chemistry will continue, allowing larger models to yield predictably higher accuracy.
Appendix A.4 cites scaling law studies and extrapolates them into a roadmap for progress. This assumes the observed trends persist beyond the tested regime.
ad hoc to paper Machine learning 'succeeds across a wide range of scenarios' and 'degrades gracefully' if the subset is easy.
The decision-theoretic core (Section 2.3) asserts this asymmetry without deriving it. It is a plausible heuristic but not a proved property of ML on chemical problems.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Position: The Inevitable Transition to Machine Learning in Quantum Chemistry." pith.science (2026). https://pith.science/paper/RBREF3LU

@misc{pith2026260718281,
  author       = {Pith},
  title        = {Pith review of: Position: The Inevitable Transition to Machine Learning in Quantum Chemistry},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RBREF3LU}},
  note         = {Machine review of arXiv:2607.18281}
}

read the original abstract

Finding exact solutions to the quantum many-body problem is computationally intractable (QMA-hard). Traditional approximations for electrons in an atom or molecule -- density functional theory and wavefunction methods -- have been indispensable, but their development shows signs of saturation: DFT functionals have proliferated without converging toward the exact functional, and strong correlation remains largely unsolved after decades of effort. This position paper argues that machine learning represents the most promising path forward -- not as a proof of logical necessity, but as a decision-theoretic argument: ML succeeds whether the underlying problems are truly hard or merely lack simple analytical solutions. We reframe recent traditional method development as ``hand-crafted machine learning'' that has exhausted the hypothesis space accessible to human intuition. Significant challenges remain, but these have clear research paths forward, unlike the fundamental barriers facing traditional approaches. ML-based approaches merit strategic priority in quantum chemistry's next phase.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

98 extracted references · 2 linked inside Pith

[1]

Nature Materials , volume =

Marzari, Nicola and Ferretti, Andrea and Wolverton, Chris , title =. Nature Materials , volume =. 2021 , doi =

2021
[2]

Energy & Environmental Materials , volume =

He, Qiang and Yu, Bin and Li, Zhaohuai and Zhao, Yan , title =. Energy & Environmental Materials , volume =. 2019 , doi =

2019
[3]

, title =

Jain, Anubhav and Ong, Shyue Ping and Hautier, Geoffroy and Chen, Wei and Richards, William Davidson and Dacek, Stephen and Cholia, Shreyas and Gunter, Dan and Skinner, David and Ceder, Gerbrand and Persson, Kristin A. , title =. APL Materials , volume =. 2013 , doi =

2013
[4]

The Journal of Physical Chemistry B , volume =

Kircheva, Nikoleta and Angelova, Silvia and Nikolova, Valya and Dudev, Todor , title =. The Journal of Physical Chemistry B , volume =. 2025 , doi =

2025
[5]

Chemical Reviews , volume =

Dudev, Todor and Lim, Carmay , title =. Chemical Reviews , volume =. 2014 , doi =

2014
[6]

and Colby, Sean M

Borges, Ricardo M. and Colby, Sean M. and Das, Susanta and Edison, Arthur S. and Fiehn, Oliver and Kind, Tobias and Lee, Jesi and Merrill, Amy T. and Merz, Kenneth M. and Metz, Thomas O. and Nunez, Jamie R. and Tantillo, Dean J. and Wang, Lee-Ping and Wang, Shunyang and Renslow, Ryan S. , title =. Chemical Reviews , volume =. 2021 , doi =

2021
[7]

WIREs Computational Molecular Science , volume =

Mazmanian, Karine and Chen, Ting and Sargsyan, Karen and Lim, Carmay , title =. WIREs Computational Molecular Science , volume =. 2022 , doi =

2022
[8]

Inorganic Chemistry , volume =

Kircheva, Nikoleta and Dobrev, Stefan and Nikolova, Valya and Yocheva, Lyubima and Angelova, Silvia and Dudev, Todor , title =. Inorganic Chemistry , volume =. 2024 , doi =

2024

Show all 98 references

[9]

Chemical Reviews , volume =

Cavalli, Andrea and Carloni, Paolo and Recanatini, Maurizio , title =. Chemical Reviews , volume =. 2006 , doi =

2006
[10]

, title =

Niazi, Sarfaraz K. , title =. International Journal of Molecular Sciences , volume =. 2025 , doi =

2025
[11]

Computational Chemistry as Applied in Environmental Research: Opportunities and Challenges , journal =

Sandoval-Pauker, Christian and Yin, Sheng and Castillo, Alexandria and Ocuane, Neidy and Puerto-Diaz, Diego and Villagr. Computational Chemistry as Applied in Environmental Research: Opportunities and Challenges , journal =. 2024 , doi =

2024
[12]

Canadian Journal of Chemistry , volume =

Siahrostami, Samira and Murray, Nicholas , title =. Canadian Journal of Chemistry , volume =. 2024 , doi =

2024
[13]

Journal of Chemical Theory and Computation , volume =

Gao, Hong and Imamura, Satoshi and Kasagi, Akihiko and Yoshida, Eiji , title =. Journal of Chemical Theory and Computation , volume =. 2024 , doi =

2024
[14]

Molecular Electronic-Structure Theory , publisher =

Helgaker, Trygve and J. Molecular Electronic-Structure Theory , publisher =
[15]

Arora, Sanjeev and Barak, Boaz , title =
[16]

and Shen, Alexander H

Kitaev, Alexei Yu. and Shen, Alexander H. and Vyalyi, Mikhail N. , title =
[17]

and Perez-Garcia, David and Wolf, Michael M

Cubitt, Toby S. and Perez-Garcia, David and Wolf, Michael M. , title =. Nature , volume =. 2015 , doi =

2015
[18]

Physical Review Letters , volume =

Liu, Yi-Kai and Christandl, Matthias and Verstraete, Frank , title =. Physical Review Letters , volume =. 2007 , doi =

2007
[19]

and Love, Peter J

Whitfield, James D. and Love, Peter J. and Aspuru-Guzik, Al. Computational Complexity in Electronic Structure , journal =. 2013 , doi =

2013
[20]

, title =

Hellman, Martin E. , title =. IEEE Transactions on Information Theory , volume =. 1980 , doi =

1980
[21]

and McMartin, Colin and Guida, Wayne C

Bohacek, Regine S. and McMartin, Colin and Guida, Wayne C. , title =. Medicinal Research Reviews , volume =. 1996 , doi =

1996
[22]

Accounts of Chemical Research , volume =

Reymond, Jean-Louis , title =. Accounts of Chemical Research , volume =. 2015 , doi =

2015
[23]

and Madzhidov, Timur I

Polishchuk, Pavel G. and Madzhidov, Timur I. and Varnek, Alexandre , title =. Journal of Computer-Aided Molecular Design , volume =. 2013 , doi =

2013
[24]

, title =

Fraenkel, Aviezri S. , title =. Bulletin of Mathematical Biology , volume =. 1993 , doi =

1993
[25]

Highly Accurate Protein Structure Prediction with

Jumper, John and Evans, Richard and Pritzel, Alexander and Green, Tim and Figurnov, Michael and Ronneberger, Olaf and Tunyasuvunakool, Kathryn and Bates, Russ and. Highly Accurate Protein Structure Prediction with. Nature , volume =. 2021 , doi =

2021
[26]

Proceedings of the 29th ACM SIGPLAN Annual Symposium on Principles and Practice of Parallel Programming , pages =

Cheng, Shenggan and Zhao, Xuanlei and Lu, Guangyang and Fang, Jiarui and Zheng, Tian and Wu, Ruidong and Zhang, Xiwen and Peng, Jian and You, Yang , title =. Proceedings of the 29th ACM SIGPLAN Annual Symposium on Principles and Practice of Parallel Programming , pages =. 2024 , doi =

2024
[27]

Journal of Chemical Information and Modeling , volume =

Skolnick, Jeffrey and Gao, Mu and Zhou, Hongyi and Singh, Suresh , title =. Journal of Chemical Information and Modeling , volume =. 2021 , doi =

2021
[28]

Lehtola, Susi and Steigemann, Conrad and Oliveira, Micael J. T. and Marques, Miguel A. L. , title =. SoftwareX , volume =. 2018 , doi =

2018
[29]

and Kaplan, Aaron D

Furness, James W. and Kaplan, Aaron D. and Ning, Jinliang and Perdew, John P. and Sun, Jianwei , title =. The Journal of Physical Chemistry Letters , volume =. 2020 , doi =

2020
[30]

and Galvelis, Raimondas and Herr, John E

Eastman, Peter and Behara, Pavan Kumar and Dotson, David L. and Galvelis, Raimondas and Herr, John E. and Horton, Josh T. and Mao, Yuezhi and Chodera, John D. and Pritchard, Benjamin P. and Wang, Yuanqing and De Fabritiis, Gianni and Markland, Thomas E. , title =. Scientific D...

2023
[31]

Karton, Amir and Sylvetsky, Nitai and Martin, Jan M. L. , title =. Journal of Computational Chemistry , volume =. 2017 , doi =

2017
[32]

and Isayev, Olexandr and Roitberg, Adrian E

Smith, Justin S. and Isayev, Olexandr and Roitberg, Adrian E. , title =. Scientific Data , volume =. 2017 , doi =

2017
[33]

and Zubatyuk, Roman and Nebgen, Benjamin and Lubbers, Nicholas and Barros, Kipton and Roitberg, Adrian E

Smith, Justin S. and Zubatyuk, Roman and Nebgen, Benjamin and Lubbers, Nicholas and Barros, Kipton and Roitberg, Adrian E. and Isayev, Olexandr and Tretiak, Sergei , title =. Scientific Data , volume =. 2020 , doi =

2020
[34]

Machine-Learning-Assisted Determination of the Global Zero-Temperature Phase Diagram of Materials , journal =

Schmidt, Jonathan and Hoffmann, Noah and Wang, Hai-Chen and Borlido, Pedro and Carri. Machine-Learning-Assisted Determination of the Global Zero-Temperature Phase Diagram of Materials , journal =. 2023 , doi =

2023
[35]

Physical Review , volume =

Hohenberg, Pierre and Kohn, Walter , title =. Physical Review , volume =. 1964 , doi =

1964
[36]

Nature Physics , volume =

Schuch, Norbert and Verstraete, Frank , title =. Nature Physics , volume =. 2009 , doi =

2009
[37]

arXiv preprint arXiv:0712.0483 , year =

Schuch, Norbert and Verstraete, Frank , title =. arXiv preprint arXiv:0712.0483 , year =

arXiv
[38]

and Schmidt, Karla , title =

Perdew, John P. and Schmidt, Karla , title =. AIP Conference Proceedings , volume =. 2001 , doi =

2001
[39]

, title =

Becke, Axel D. , title =. The Journal of Chemical Physics , volume =. 1993 , doi =

1993
[40]

and Devlin, Frank J

Stephens, Philip J. and Devlin, Frank J. and Chabalowski, Cary F. and Frisch, Michael J. , title =. The Journal of Physical Chemistry , volume =. 1994 , doi =

1994
[41]

and Bushmarinov, Ivan S

Medvedev, Michael G. and Bushmarinov, Ivan S. and Sun, Jianwei and Perdew, John P. and Lyssenko, Konstantin A. , title =. Science , volume =. 2017 , doi =

2017
[42]

, title =

Kepp, Kasper P. , title =. Science , volume =. 2017 , doi =

2017
[43]

Science , volume =

Hammes-Schiffer, Sharon , title =. Science , volume =. 2017 , doi =

2017
[44]

, title =

Karton, Amir and de Oliveira, Marcelo T. , title =. WIREs Computational Molecular Science , volume =. 2025 , doi =

2025
[45]

Molecular Physics , volume =

Mardirossian, Narbe and Head-Gordon, Martin , title =. Molecular Physics , volume =. 2017 , doi =

2017
[46]

The Journal of Chemical Physics , volume =

Mardirossian, Narbe and Head-Gordon, Martin , title =. The Journal of Chemical Physics , volume =. 2016 , doi =

2016
[47]

and Pople, John A

Raghavachari, Krishnan and Trucks, Gary W. and Pople, John A. and Head-Gordon, Martin , title =. Chemical Physics Letters , volume =. 1989 , doi =

1989
[48]

, title =

White, Steven R. , title =. Physical Review Letters , volume =. 1992 , doi =

1992
[49]

The Density-Matrix Renormalization Group in the Age of Matrix Product States , journal =

Schollw. The Density-Matrix Renormalization Group in the Age of Matrix Product States , journal =. 2011 , doi =

2011
[50]

Roothaan, Clemens C. J. , title =. Reviews of Modern Physics , volume =. 1951 , doi =

1951
[51]

, title =

Dunning, Thom H. , title =. The Journal of Chemical Physics , volume =. 1989 , doi =

1989
[52]

and Musia

Bartlett, Rodney J. and Musia. Coupled-Cluster Theory in Quantum Chemistry , journal =. 2007 , doi =

2007
[53]

and Berkelbach, Timothy C

Neufeld, Verena A. and Berkelbach, Timothy C. , title =. Physical Review Letters , volume =. 2023 , doi =

2023
[54]

and Lee, Joonho and Head-Gordon, Martin , title =

Bertels, Luke W. and Lee, Joonho and Head-Gordon, Martin , title =. Journal of Chemical Theory and Computation , volume =. 2021 , doi =

2021
[55]

and Alavi, Ali , title =

Li Manni, Giovanni and Smart, Simon D. and Alavi, Ali , title =. Journal of Chemical Theory and Computation , volume =. 2016 , doi =

2016
[56]

The Journal of Chemical Physics , volume =

Riplinger, Christoph and Neese, Frank , title =. The Journal of Chemical Physics , volume =. 2013 , doi =

2013
[57]

The Journal of Chemical Physics , volume =

Riplinger, Christoph and Sandhoefer, Barbara and Hansen, Andreas and Neese, Frank , title =. The Journal of Chemical Physics , volume =. 2013 , doi =

2013
[58]

Proceedings of the National Academy of Sciences , volume =

Prodan, Emil and Kohn, Walter , title =. Proceedings of the National Academy of Sciences , volume =. 2005 , doi =

2005
[59]

Nature Communications , volume =

Gong, Xiaoxun and Li, He and Zou, Nianlong and Xu, Runzhang and Duan, Wenhui and Xu, Yong , title =. Nature Communications , volume =. 2023 , doi =

2023
[60]

Nature Computational Science , volume =

Li, He and Wang, Zun and Zou, Nianlong and Ye, Meng and Xu, Runzhang and Gong, Xiaoxun and Duan, Wenhui and Xu, Yong , title =. Nature Computational Science , volume =. 2022 , doi =

2022
[61]

and Lee, Su-In , title =

Lundberg, Scott M. and Lee, Su-In , title =. Advances in Neural Information Processing Systems , volume =
[62]

and Rupp, Matthias and von Lilienfeld, O

Ramakrishnan, Raghunathan and Dral, Pavlo O. and Rupp, Matthias and von Lilienfeld, O. Anatole , title =. Scientific Data , volume =. 2014 , doi =

2014
[63]

and Riley, Patrick F

Gilmer, Justin and Schoenholz, Samuel S. and Riley, Patrick F. and Vinyals, Oriol and Dahl, George E. , title =. Proceedings of the 34th International Conference on Machine Learning , pages =
[64]

and Kornbluth, Mordechai and Molinari, Nicola and Smidt, Tess E

Batzner, Simon and Musaelian, Albert and Sun, Lixin and Geiger, Mario and Mailoa, Jonathan P. and Kornbluth, Mordechai and Molinari, Nicola and Smidt, Tess E. and Kozinsky, Boris , title =. Nature Communications , volume =. 2022 , doi =

2022
[65]

Equivariant Message Passing for the Prediction of Tensorial Properties and Molecular Spectra , booktitle =

Sch. Equivariant Message Passing for the Prediction of Tensorial Properties and Molecular Spectra , booktitle =
[66]

and Huddleston, Kate K

Devereux, Christian and Smith, Justin S. and Huddleston, Kate K. and Barros, Kipton and Zubatyuk, Roman and Isayev, Olexandr and Roitberg, Adrian E. , title =. Journal of Chemical Theory and Computation , volume =. 2020 , doi =

2020
[67]

Journal of the American Chemical Society , volume =

Kov. Journal of the American Chemical Society , volume =. 2025 , doi =

2025
[68]

Advances in Neural Information Processing Systems , volume =

Batatia, Ilyes and Kov. Advances in Neural Information Processing Systems , volume =
[69]

and Ceder, Gerbrand , title =

Deng, Bowen and Zhong, Peichen and Jun, KyuJung and Riebesell, Janosh and Han, Kevin and Bartel, Christopher J. and Ceder, Gerbrand , title =. Nature Machine Intelligence , volume =. 2023 , doi =

2023
[70]

and Kovács, Dávid P

Batatia, Ilyes and Benner, Philipp and Chiang, Yuan and Elena, Alin M. and Kovács, Dávid P. and Riebesell, Janosh and Advincula, Xavier R. and Asta, Mark and Avaylon, Matthew and Baldwin, William J. and Berger, Fabian and Bernstein, Noam and Bhowmik, Arghya and Bigi, Filippo a...

2025
[71]

and De Vita, Alessandro , title =

Li, Zhenwei and Kermode, James R. and De Vita, Alessandro , title =. Physical Review Letters , volume =. 2015 , doi =

2015
[72]

and Nebgen, Benjamin , title =

Kulichenko, Maksim and Barros, Kipton and Lubbers, Nicholas and Li, Ying Wai and Messerly, Richard and Tretiak, Sergei and Smith, Justin S. and Nebgen, Benjamin , title =. Nature Computational Science , volume =. 2023 , doi =

2023
[73]

Machine Learning for Molecular Simulation , journal =

No. Machine Learning for Molecular Simulation , journal =. 2020 , doi =

2020
[74]

Review of Multi-Fidelity Models , journal =

Fern. Review of Multi-Fidelity Models , journal =. 2023 , doi =

2023
[75]

Physical Chemistry Chemical Physics , volume =

Zhao, Heng and Gould, Tim and Vuckovic, Stefan , title =. Physical Chemistry Chemical Physics , volume =. 2024 , doi =

2024
[76]

and Losev, Timofey V

Gerasimov, Igor S. and Losev, Timofey V. and Epifanov, Evgeny Yu. and Rudenko, Irina and Bushmarinov, Ivan S. and Ryabov, Alexander A. and Zhilyaev, Petr A. and Medvedev, Michael G. , title =. Science , volume =. 2022 , doi =

2022
[77]

Kirkpatrick, James and McMorrow, Brendan and Turban, David H. P. and Gaunt, Alexander L. and Spencer, James S. and Matthews, Alexander G. D. G. and Obika, Annette and Thiry, Louis and Fortunato, Meire and Pfau, David and Castellanos, Lara Rom. Response to Comment on ``Pushing ...

2022
[78]

Kirkpatrick, James and McMorrow, Brendan and Turban, David H. P. and Gaunt, Alexander L. and Spencer, James S. and Matthews, Alexander G. D. G. and Obika, Annette and Thiry, Louis and Fortunato, Meire and Pfau, David and Castellanos, Lara Rom. Pushing the Frontiers of Density ...

2021
[79]

Deep-Neural-Network Solution of the Electronic

Hermann, Jan and Sch. Deep-Neural-Network Solution of the Electronic. Nature Chemistry , volume =. 2020 , doi =

2020
[80]

and Matthews, Alexander G

Pfau, David and Spencer, James S. and Matthews, Alexander G. D. G. and Foulkes, W. M. C. , title =. Physical Review Research , volume =. 2020 , doi =

2020
[81]

Nature Communications , volume =

Scherbela, Michael and Gerard, Leon and Grohs, Philipp , title =. Nature Communications , volume =. 2024 , doi =

2024
[82]

An Ab Initio Foundation Model of Wavefunctions That Accurately Describes Chemical Bond Breaking , journal =

Foster, Adam and Sch. An Ab Initio Foundation Model of Wavefunctions That Accurately Describes Chemical Bond Breaking , journal =
[83]

and Scherbela, M

Gerard, L. and Scherbela, M. and Sutterud, H. and Foulkes, W. M. C. and Grohs, P. , title =. Nature Computational Science , volume =. 2025 , doi =

2025
[84]

and Chen, Ji and He, Di and Goddard III, William A

Jiang, Du and Wen, Xuelan and Chen, Yixiao and Li, Ruichen and Fu, Weizhong and Pham, Hung Q. and Chen, Ji and He, Di and Goddard III, William A. and Wang, Liwei and Ren, Weiluo , title =. arXiv preprint arXiv:2508.02570 , year =

arXiv
[85]

Quantum Computational Chemistry , journal =

McArdle, Sam and Endo, Suguru and Aspuru-Guzik, Al. Quantum Computational Chemistry , journal =. 2020 , doi =

2020
[86]

and Boixo, Sergio and Smelyanskiy, Vadim N

McClean, Jarrod R. and Boixo, Sergio and Smelyanskiy, Vadim N. and Babbush, Ryan and Neven, Hartmut , title =. Nature Communications , volume =. 2018 , doi =

2018
[87]

Digital Discovery , volume =

A Universal Machine Learning Model for the Electronic Density of States , author =. Digital Discovery , volume =. 2026 , doi =

2026
[88]

The Journal of Physical Chemistry Letters , volume =

Ren, Fangning and Chen, Xu and Liu, Fang , title =. The Journal of Physical Chemistry Letters , volume =. 2025 , doi =

2025
[89]

and Geng, Dominik and Gerhartz, Gerrit and Ickler, Marc K

Remme, Roman and Kaczun, Tobias and Ebert, Tim and Gehrig, Christof A. and Geng, Dominik and Gerhartz, Gerrit and Ickler, Marc K. and Klockow, Manuel V. and Lippmann, Peter and Schmidt, Johannes S. and Wagner, Simon and Dreuw, Andreas and Hamprecht, Fred A. , title =. Journal ...

2025
[90]

Gao, Nicholas and Grutschus, Till and No. Excited. Proceedings of the 43rd International Conference on Machine Learning , series =
[91]

Entwistle, Michael T. and Sch. Electronic Excited States in Deep Variational. Nature Communications , volume =. 2023 , doi =

2023
[92]

and Soklaski, Ryan and Axelrod, Simon and Samsi, Siddharth and G

Frey, Nathan C. and Soklaski, Ryan and Axelrod, Simon and Samsi, Siddharth and G. Neural Scaling of Deep Chemical Models , journal =. 2023 , doi =

2023
[93]

and Kornbluth, Mordechai and Kozinsky, Boris , title =

Musaelian, Albert and Batzner, Simon and Johansson, Anders and Sun, Lixin and Owen, Cameron J. and Kornbluth, Mordechai and Kozinsky, Boris , title =. Nature Communications , volume =. 2023 , doi =

2023
[94]

Lawrence , title =

Passaro, Saro and Zitnick, C. Lawrence , title =. Proceedings of the 40th International Conference on Machine Learning , pages =
[95]

and Han, Seungwu , title =

Kim, Jaesun and You, Jinmu and Park, Yutack and Lim, Yunsung and Kang, Yujin and Kim, Jisu and Jeon, Haekwan and Ju, Suyeon and Hong, Deokgi and Lee, Seung Yul and Choi, Saerom and Kim, Yongdeok and Lee, Jae W. and Han, Seungwu , title =. Nature Communications , volume =. 2026 , doi =

2026
[96]

and Kaplan, Aaron D

Kuner, Matthew C. and Kaplan, Aaron D. and Persson, Kristin A. and Asta, Mark and Chrzan, Daryl C. , title =. npj Computational Materials , volume =. 2025 , doi =

2025
[97]

and Liu, Runze and Qi, Ji and Ko, Tsz Wai and Deng, Bowen and Riebesell, Janosh and Ceder, Gerbrand and Persson, Kristin A

Kaplan, Aaron D. and Liu, Runze and Qi, Ji and Ko, Tsz Wai and Deng, Bowen and Riebesell, Janosh and Ceder, Gerbrand and Persson, Kristin A. and Ong, Shyue Ping , title =. arXiv preprint arXiv:2503.04070 , year =. 2503.04070 , archivePrefix =
[98]

Nature Computational Science , volume =

Zhang, He and Liu, Siyuan and You, Jiacheng and Liu, Chang and Zheng, Shuxin and Lu, Ziheng and Wang, Tong and Zheng, Nanning and Shao, Bin , title =. Nature Computational Science , volume =. 2024 , doi =

2024

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.

[1] [1]

Nature Materials , volume =

Marzari, Nicola and Ferretti, Andrea and Wolverton, Chris , title =. Nature Materials , volume =. 2021 , doi =

2021

[2] [2]

Energy & Environmental Materials , volume =

He, Qiang and Yu, Bin and Li, Zhaohuai and Zhao, Yan , title =. Energy & Environmental Materials , volume =. 2019 , doi =

2019

[3] [3]

, title =

Jain, Anubhav and Ong, Shyue Ping and Hautier, Geoffroy and Chen, Wei and Richards, William Davidson and Dacek, Stephen and Cholia, Shreyas and Gunter, Dan and Skinner, David and Ceder, Gerbrand and Persson, Kristin A. , title =. APL Materials , volume =. 2013 , doi =

2013

[4] [4]

The Journal of Physical Chemistry B , volume =

Kircheva, Nikoleta and Angelova, Silvia and Nikolova, Valya and Dudev, Todor , title =. The Journal of Physical Chemistry B , volume =. 2025 , doi =

2025

[5] [5]

Chemical Reviews , volume =

Dudev, Todor and Lim, Carmay , title =. Chemical Reviews , volume =. 2014 , doi =

2014

[6] [6]

and Colby, Sean M

Borges, Ricardo M. and Colby, Sean M. and Das, Susanta and Edison, Arthur S. and Fiehn, Oliver and Kind, Tobias and Lee, Jesi and Merrill, Amy T. and Merz, Kenneth M. and Metz, Thomas O. and Nunez, Jamie R. and Tantillo, Dean J. and Wang, Lee-Ping and Wang, Shunyang and Renslow, Ryan S. , title =. Chemical Reviews , volume =. 2021 , doi =

2021

[7] [7]

WIREs Computational Molecular Science , volume =

Mazmanian, Karine and Chen, Ting and Sargsyan, Karen and Lim, Carmay , title =. WIREs Computational Molecular Science , volume =. 2022 , doi =

2022

[8] [8]

Inorganic Chemistry , volume =

Kircheva, Nikoleta and Dobrev, Stefan and Nikolova, Valya and Yocheva, Lyubima and Angelova, Silvia and Dudev, Todor , title =. Inorganic Chemistry , volume =. 2024 , doi =

2024