Skip to content
DraconiticResearch
Search Sources API
194,230 studies31 sourcesUpdated 56 min ago
Peer-reviewedCC BYNature Communications

An ab initio foundation model of wavefunctions that accurately describes chemical bond breaking

  • Adam Foster1
  • Zeno Schätzle2,3
  • P. Bernát Szabó2,3
  • Lixue Cheng2,3,4
  • Jonas Köhler5,3
  • Gino Cassella6,3
  • Nicholas Gao5,3
  • Jiawei Li3,7
  • Frank Noé8,9
  • Jan Hermann10
10 affiliations
  1. 1Microsoft Research AI for Science, Cambridge, UK. [email protected]
  2. 2Freie Universität, Berlin, Germany
  3. 3Microsoft Research AI for Science, Cambridge, UK
  4. 4The Hong Kong University of Science and Technology, Hong Kong, China
  5. 5CuspAI, Cambridge, UK
  6. 6Imperial College London, London, UK
  7. 7Tsinghua University, Beijing, China
  8. 8Freie Universität, Berlin, Germany. [email protected]
  9. 9Microsoft Research AI for Science, Cambridge, UK. [email protected]
  10. 10Microsoft Research AI for Science, Cambridge, UK. [email protected]
Open at sourcePDFDOICite

Abstract

Reliable description of bond breaking remains a major challenge for quantum chemistry due to the multireference character of the electronic structure in dissociating species. Multireference methods in particular suffer from large computational cost, which under the normal paradigm has to be paid anew for each system at a full price, ignoring commonalities in electronic structure across molecules. Quantum Monte Carlo with deep neural networks uniquely offers to exploit such commonalities by pretraining transferable wavefunction models, but all such attempts were so far limited in scope. Here, we bring this paradigm to fruition with Orbformer, a transferable wavefunction model pretrained on 22,000 equilibrium and dissociating structures that can be fine-tuned on unseen molecules reaching an accuracy-cost ratio rivalling classical multireference methods. On established benchmarks as well as more challenging bond dissociations and Diels-Alder reactions, Orbformer is the only method that consistently converges to chemical accuracy (1 kcal/mol). This work turns the idea of amortizing the cost of solving the Schrödinger equation over many molecules into a practical approach in quantum chemistry.

Related

  1. Discretization-Aware Fine-Tuning for Quantum Machine Learning with Chemical Foundation Models
    Shunji Matsuura, Sonika Johri
    arXivPreprintCC BY 4.0Full textQuantum physics & quantum informationarXiv
  2. When the gravity path integral describes a statistical average
    Gabriele Di Ubaldo, Luca V. Iliesiu, Henry W. Lin +1
    arXivPreprintCC BY 4.0Full textGravitation & general relativityHigh-energy & particle physicsarXiv
  3. Ginzburg-Landau Theory with Elastic Incompatibility Describes the Liquid-G-phase-Crystal State Diagram
    Alessio Zaccone, Konrad Samwer
    arXivPreprintarXiv licenceFull textApplied physics & instrumentationCondensed matterarXiv
  4. Spin-adapted neural network wavefunctions in real space
    Ruichen Li, Yuzhi Liu, Du Jiang +7
    Nature CommunicationsPeer-reviewedCC BY-NC-ND 4.0Full text2 citedAtomic, molecular & optical physicsCrossref
  5. All-plus Yang--Mills and gravity wavefunctions in de Sitter space
    Song He, Jiajie Mei, Yuyu Mo
    arXivPreprintCC BY 4.0Full textHigh-energy & particle physicsarXiv

Cite

BibTeXRIS
BibTeX
@article{foster2026initio,
  title = {{An ab initio foundation model of wavefunctions that accurately describes chemical bond breaking}},
  author = {Adam Foster and Zeno Schätzle and P. Bernát Szabó and Lixue Cheng and Jonas Köhler and Gino Cassella and Nicholas Gao and Jiawei Li and Frank Noé and Jan Hermann},
  journal = {Nature Communications},
  year = {2026},
  month = aug,
  doi = {10.1038/s41467-026-76604-2},
  url = {https://doi.org/10.1038/s41467-026-76604-2},
  note = {CC BY},
}
RIS
TY  - JOUR
TI  - An ab initio foundation model of wavefunctions that accurately describes chemical bond breaking
AU  - Adam Foster
AU  - Zeno Schätzle
AU  - P. Bernát Szabó
AU  - Lixue Cheng
AU  - Jonas Köhler
AU  - Gino Cassella
AU  - Nicholas Gao
AU  - Jiawei Li
AU  - Frank Noé
AU  - Jan Hermann
PY  - 2026
DA  - 2026/08/21
JO  - Nature Communications
DO  - 10.1038/s41467-026-76604-2
UR  - https://doi.org/10.1038/s41467-026-76604-2
AB  - Reliable description of bond breaking remains a major challenge for quantum chemistry due to the multireference character of the electronic structure in dissociating species. Multireference methods in particular suffer from large computational cost, which under the normal paradigm has to be paid anew for each system at a full price, ignoring commonalities in electronic structure across molecules. Quantum Monte Carlo with deep neural networks uniquely offers to exploit such commonalities by pretraining transferable wavefunction models, but all such attempts were so far limited in scope. Here, we bring this paradigm to fruition with Orbformer, a transferable wavefunction model pretrained on 22,000 equilibrium and dissociating structures that can be fine-tuned on unseen molecules reaching an accuracy-cost ratio rivalling classical multireference methods. On established benchmarks as well as more challenging bond dissociations and Diels-Alder reactions, Orbformer is the only method that consistently converges to chemical accuracy (1 kcal/mol). This work turns the idea of amortizing the cost of solving the Schrödinger equation over many molecules into a practical approach in quantum chemistry.
SN  - 2041-1723
ER  -