Pith. sign in

REVIEW 4 major objections 7 minor 56 references

Nonadiabatic Molecular Dynamics on Real-time Excited-State Surfaces via Machine Learning Hamiltonians

T0 review · 4 major / 7 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read A machine-learned Hamiltonian makes on-the-fly fewest-switches surface hopping in solids practical, removing the Classical Path Approximation.

desk verdict Convincing proof that ML Hamiltonians can make on-the-fly NAMD in solids practical, but the delta-SCF force approximation is load-bearing and only validated on silicon. read the letter →

arxiv 2608.08095 v1 pith:5GQ2JGEE submitted 2026-08-08 physics.comp-ph

Changwei Zhang , Yang Zhong , Zhi-Guo Tao , Yingzhou Li , Zhenggang Lan , Oleg V. Prezhdo , Xin-Gao Gong , Weibin Chu
show 1 more author
Hongjun Xiang
This is my paper · ORCID
classification physics.comp-ph
keywords nonadiabaticmoleculardynamicssurfacehoppingmachinelearningHamiltonianE(3)-equivariantneuralnetworkexcited-stateforcescouplingvectorsdelta-SCFpolaron
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

The paper introduces on-the-fly N$^2$AMD and claims it removes the main bottleneck in nonadiabatic molecular dynamics (NAMD) for periodic solids: the need to compute excited-state forces and nonadiabatic coupling vectors at every time step. Where the standard Classical Path Approximation fixes nuclei on a ground-state surface and misses the back-reaction of excited carriers, the new workflow uses an E(3)-equivariant neural network to predict the electronic Hamiltonian and obtains all required derivatives by automatic differentiation. This makes true fewest-switches surface hopping in solids practical at ab initio accuracy and, because the Hamiltonian can be trained on hybrid-functional data, at hybrid-functional accuracy. If the claim holds, the method corrects order-of-magnitude errors in carrier relaxation, enables simulation of photoinduced ferroelectric switching, and captures polaron formation in real time.

What carries the argument

The load-bearing object is the differentiable machine-learned Hamiltonian. An E(3)-equivariant graph neural network (HamGNN) outputs the Hamiltonian matrix $H_{\mu\nu}$ and overlap matrix $S_{\mu\nu}$ from atomic positions, so $\nabla_R H_{\mu\nu}$ is obtained by automatic differentiation and the overlap gradients $\langle\nabla_R\phi_\mu|\phi_\nu\rangle$, $\langle\phi_\mu|\nabla_R\phi_\nu\rangle$ by two-center integrals. The excited-state force is the delta-SCF expression $F \approx F_{gs} + \sum_{i\in VB}(1-f_i)\nabla_R\epsilon_i - \sum_{a\in CB}f_a\nabla_R\epsilon_a$, and the nonadiabatic coupling vector is $d_{mn} = (\epsilon_m-\epsilon_n)^{-1}\sum_{\mu\nu}c^*_{m\mu}c_{n\nu}[-\nabla_R H_{\mu\nu} + \epsilon_n\langle\nabla_R\phi_\mu|\phi_\nu\rangle + \epsilon_m\langle\phi_\mu|\nabla_R\phi_\nu\rangle]$ for $m\neq n$. Combining these with the ground-state force from the Allegro potential closes the feedback loop: the electronic state directs the nuclei, and the resulting nuclear motion feeds back into the Hamiltonian at the next time step.

What would settle it

Take the bilayer h-BN AB-BA sliding pathway at 0.027 e/u.c. and the HSE hole polaron in anatase TiO2, and recompute forces with fully self-consistent delta-SCF at the geometries visited by the NAMD trajectories; if the forces differ from Eq. (2) by more than a few tens of meV/Å, or if the h-BN barrier is no longer eliminated, the central claim fails.

Watch

Extended reading notes

Core claim

The paper's central claim is that a single learned Hamiltonian—an E(3)-equivariant neural network mapping atomic configurations to Hamiltonian and overlap matrices in a numerical atomic orbital basis—supplies every ingredient on-the-fly surface hopping needs: excited-state potential energy surfaces through the delta-SCF force expression of Eq. (2), nonadiabatic coupling vectors through the gradient formula of Eq. (4), and ground-state forces through a separate equivariant interatomic potential. On bulk silicon, the learned excited-state force agrees with self-consistent delta-SCF benchmarks to a mean absolute error of 7.6 meV/Å, while the ground-state approximation differs by 110.4 meV/Å, and the workflow runs a 96-atom cell 163 times faster than DFT. The method is then applied to three regimes the Classical Path Approximation cannot reach: carrier relaxation in a MoS2/WS2 heterostructure, where the relaxation rate is corrected by an order of magnitude; photoinduced ferroelectric switching in bilayer h-BN, with 82% of trajectories reversing polarization within 1.5 ps; and real-time hole polaron formation in anatase TiO2 at the hybrid-functional HSE level.

Load-bearing premise

The load-bearing premise is that the non-self-consistent delta-SCF force formula, benchmarked only on silicon at 0.39% valence excitation, remains accurate for bilayer h-BN at 0.027 e/u.c. and for the HSE hole polaron in TiO2; if self-consistent relaxation of the Hamiltonian under the excited density matters there, the simulated switching and polaron dynamics rest on incorrect forces.

Editorial extensions

If this is right

  • Nuclei in solid-state NAMD can evolve on excited-state potential energy surfaces, so the lattice responds to the instantaneous electronic state instead of a precomputed phonon background.
  • Hybrid-functional NAMD becomes feasible, as shown by the HSE simulation of hole polaron formation in anatase TiO2, removing self-interaction errors that PBE-based NAMD cannot capture.
  • Carrier relaxation kinetics computed with the Classical Path Approximation may be systematically revised; in the MoS2/WS2 heterostructure the correction changes the dynamics by an order of magnitude.
  • Photoinduced phase transitions such as sliding ferroelectric switching in bilayer h-BN become directly simulable, including hot-carrier cooling that constrained Born-Oppenheimer MD misses.
  • Because the learned-Hamiltonian workflow scales as $O(N^2)$ while DFT scales more steeply, the speedup over ab initio NAMD grows with system size.

Reading between the lines

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

  • The same differentiable-Hamiltonian pipeline could be trained against self-consistent delta-SCF, GW, or BSE data, which would harden the force approximation for strongly correlated systems or high excitation densities without changing the workflow.
  • Because the network returns the full Hamiltonian rather than a scalar energy, the framework could be extended to spin-orbit coupling, Berry phases, or time-dependent electron propagation within the same trajectory machinery.
  • The demonstrated transferability suggests a cheap validation protocol: before production NAMD in a new material, compute self-consistent delta-SCF forces on a handful of representative geometries to confirm Eq. (2) holds there.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 7 minor

Summary. The paper introduces on-the-fly N2AMD, a machine-learning framework for nonadiabatic molecular dynamics in periodic solids. It combines an E(3)-equivariant neural-network Hamiltonian (HamGNN) with a ground-state ML potential (Allegro) to compute excited-state forces through a delta-SCF frozen-orbital approximation (Eq. 2) and nonadiabatic coupling vectors through automatic differentiation of the learned Hamiltonian (Eq. 4). This allows fewest-switches surface hopping without the Classical Path Approximation. The authors benchmark the approach on bulk silicon (Hamiltonian MAE 0.045 meV, excited-state force MAE 7.6 meV/Å, smoothed NACV MAE 0.52 meV/Å), demonstrate transferability to twisted MoS2 bilayers and Ga_xAl_1-xAs alloys, and report three applications: hole relaxation in MoS2/WS2, photoinduced ferroelectric switching in bilayer h-BN, and hole polaron formation in anatase TiO2 at the HSE level.

Significance. If the underlying approximations hold in the application regimes, this is a substantial advance: it removes the Classical Path Approximation for solid-state NAMD, restores the back-reaction of excited carriers on the lattice, and makes hybrid-functional NAMD tractable. The paper also ships code and data on Zenodo, which is a strong reproducibility feature. The reported 163x speedup per ionic step for a 96-atom silicon cell is significant. However, the central excited-state force approximation is validated only in a narrow regime, and the headline applications reach beyond that validation; the quantitative strength of the paper therefore depends on additional benchmarks that are currently absent.

major comments (4)
  1. [Eq. (2) / Benchmarks] The delta-SCF frozen-Hamiltonian force approximation in Eq. (2) is benchmarked only on bulk silicon at 0.39% valence excitation (Fig. 2a), yet it is used without revalidation for the h-BN ferroelectric switching at 0.027 e/u.c. and for the HSE hole polaron in anatase TiO2. In the polaron application, self-trapping is driven by the self-consistent response of the Hamiltonian, including the Fock-exchange component, to the localized hole; a frozen ground-state Hamiltonian gradient cannot, by construction, generate the self-consistent feedback that creates the gap state. The paper acknowledges the approximation but reports no force-error statistics along the trajectories of either application, so the central applications rest on an unverified approximation.
  2. [Eq. (2) and surrounding derivation] The statement that "double-counting energies vanish upon differentiation, as they do not explicitly depend on atomic positions" is not generally correct: the Hartree and XC double-counting terms are density functionals, and the density depends on R through the orbitals. The correct statement is that the frozen-orbital delta-SCF approximation neglects these derivatives; presenting them as vanishing by construction misstates the derivation and could mislead readers about the accuracy of the force formula.
  3. [Fig. 2d / Benchmarks] The NACV benchmark reports only the magnitude of the smoothed couplings d^s_ij = |(epsilon_i - epsilon_j) d_ij|, not the full NACV vectors d_ij that enter the FSSH hopping probability and the velocity rescaling described after Eq. (4). Because the direction of the NACV controls the momentum adjustment at each hop, a scalar MAE on smoothed magnitudes does not establish that the learned Hamiltonian yields accurate vectorial couplings; a vector benchmark, or at least a report of the directional error, is needed to support the on-the-fly FSSH claim.
  4. [Application 2 / Fig. 4] The TiO2 polaron demonstration claims hybrid-functional accuracy, but the manuscript provides no accuracy metrics for the HSE-trained neural Hamiltonian or for HSE delta-SCF forces. Training-data MAE, force MAE, and NACV MAE should be reported for the HSE model, as done for silicon, before the "hybrid-functional level" claim can be assessed quantitatively.
minor comments (7)
  1. [Benchmarks] The phrase "migrate to issues" should be "mitigate issues".
  2. [Workflow description] The phrase "combing these key quantities" should be "combining these key quantities".
  3. [Fig. 2b] The scaling comparison would be clearer if N were defined (number of atoms versus basis functions) and if the DFT scaling curve were labeled explicitly.
  4. [Eq. (4)] The line break after the equality sign leaves the summation bracket unbalanced in the printed text; please reformat the equation.
  5. [Benchmarks] The sentence "Here 0.39% of the valence electrons excited" ends without a period.
  6. [Introduction] The acronym N2AMD is not expanded at first use in the main text; consider defining it explicitly as "Neural network NAMD".
  7. [References] Reference [30] appears to be an unpublished manuscript; please provide its publication status or a journal reference if available.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the ML Hamiltonian is fitted to ab initio matrix elements, forces and NACVs are derived from established formulas, and the central claims are benchmarked against external DFT references; only minor, non-load-bearing self-citation of building blocks appears.

full rationale

The paper's derivation chain is self-contained against external benchmarks. The central learned quantity, the neural Hamiltonian, is fitted to ab initio DFT Hamiltonian matrices (MAE 0.045 meV on silicon), not to the target dynamics, and the excited-state forces (Eq. 2) and NACVs (Eq. 4) are derived from that Hamiltonian via the standard delta-SCF and Hellmann-Feynman/Pulay formulas. The delta-SCF approximation is cited to Gavnholt (an external source), and the NACV formula is cited to Abad et al. (also external). No fitted parameter is renamed as a prediction: the network is fitted to Hamiltonian matrix elements, and the forces and couplings are differentiations of those fitted elements, validated out-of-sample in Fig. 2 against self-consistent delta-SCF references and in transfer tests to twisted MoS2 and unseen Ga/Al compositions. The benchmark in Fig. 2a (7.6 meV/Angstrom MAE) measures the combined error of the non-self-consistent approximation plus the ML surrogate against a rigorous self-consistent reference, so it is a genuine test rather than a self-fulfilling comparison. The only self-citation is transparent building-block reuse: HamGNN and Allegro are taken from the authors' prior work [22], but the present paper re-benchmarks their accuracy extensively, so the self-citation is not load-bearing. The limited validation of the delta-SCF frozen-Hamiltonian approximation (silicon, 0.39% excitation) and its application to h-BN and TiO2 polaron regimes is an accuracy and robustness concern about the approximation's validity, not a circularity: the derivation does not assume the conclusion it draws. Similarly, the statement that double-counting energies vanish upon differentiation is a physical claim about the approximation that may be debatable, but it is not an instance of a quantity being defined in terms of the quantity it predicts.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new physical entities. The central claim rests on a fitted neural Hamiltonian, a fitted ground-state potential, and the delta-SCF approximation, all of which are acknowledged in the text.

free parameters (2)
  • HamGNN neural-network weights = Trained on ab initio Hamiltonians
    The equivariant neural network is fitted to DFT/HSE Hamiltonian matrix elements for the training configurations. The central claim that excited-state forces and NACVs are accurate depends on these fitted weights.
  • Allegro ground-state potential parameters = Trained on ground-state DFT energies and forces
    The ground-state PES used in the delta-prediction architecture is provided by a fitted machine-learned interatomic potential, so its parameters are free parameters fitted to data.
assumptions (4)
  • domain assumption The non-self-consistent delta-SCF force approximation (Eq. 2) is accurate for the studied systems and excitation densities.
    The paper explicitly states that the primary approximation is the neglect of self-consistent relaxation of the electronic Hamiltonian under excitation. This is benchmarked only on silicon and is carried into the h-BN and TiO2 applications without direct revalidation.
  • domain assumption The E(3)-equivariant Hamiltonian neural network generalizes to out-of-equilibrium and unseen configurations encountered during MD trajectories.
    The network is trained on a finite set of configurations and used to propagate dynamics, including transferability tests to twisted MoS2 and mixed GaAlAs. The central results depend on the network's extrapolation being accurate.
  • domain assumption Gamma-point-only sampling in large supercells is sufficient to describe the excited-state dynamics.
    The paper states that large supercells allow restriction to a single k-point, but no explicit convergence test against denser k-point sampling is reported in the main text.
  • domain assumption Fewest-switches surface hopping with velocity rescaling along the NACV gives physically correct nonadiabatic dynamics in solids.
    The method is built on the FSSH algorithm and standard velocity-rescaling prescription; the paper does not address known FSSH limitations such as decoherence or frustrated hops.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Nonadiabatic Molecular Dynamics on Real-time Excited-State Surfaces via Machine Learning Hamiltonians." pith.science (2026). https://pith.science/paper/5GQ2JGEE

@misc{pith2026260808095,
  author       = {Pith},
  title        = {Pith review of: Nonadiabatic Molecular Dynamics on Real-time Excited-State Surfaces via Machine Learning Hamiltonians},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5GQ2JGEE}},
  note         = {Machine review of arXiv:2608.08095}
}
abstract

Simulating the coupled, nonequilibrium dynamics of electrons and nuclei is a central challenge in chemistry, physics, and materials science, governing phenomena from photocatalysis to quantum information. The primary bottleneck has been the lack of a general, accurate, and efficient method for modeling the complete excited-state landscape: the potential energy surfaces, forces, and non-adiabatic couplings for multiple electronic states. While machine learning has revolutionized ground-state simulations and shown promise for excited states in molecules, a unified framework that solves the complete multi-state problem for general condensed matter systems has remained elusive. Here we introduce on-the-fly N${^2}$AMD (Neural network NAMD), a machine learning framework that makes on-the-fly NAMD in solids a reality. By employing an equivariant neural network to predict the system Hamiltonian, the framework delivers excited-state energies, forces, and non-adiabatic coupling vectors at a fraction of the cost of ab initio calculations. Crucially, it allows simulations with hybrid functional accuracy, a level of approach previously inaccessible for NAMD. We showcase its capabilities with three topical examples: correcting order-of-magnitude errors in carrier dynamics predicted by conventional procedure in a MoS$_2$/WS$_2$ heterostructure, simulating previously inaccessible photoinduced ferroelectric switching, and capturing real-time polaron formation in TiO$_2$ at the hybrid-functional level. On-the-fly N${^2}$AMD moves beyond the limitations of equilibrium theory, establishing a new paradigm for the predictive, first-principles design of materials operating far from equilibrium.

Figures

Figures reproduced from arXiv: 2608.08095 by the authors.

Figure 1
Figure 1. FIG. 1. (a) Conventional NAMD (CPA) in solids restricts [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. Accuracy, efficiency and transferability of on-the-fly N [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3. Photo-induced ferroelectric switching in bilayer [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: FIG. 4. Real-time evolution of carrier localization in anatase [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

56 extracted references · 50 canonical work pages

  1. [1]

    Williams and A

    F. Williams and A. J. Nozik, Solid-state perspectives of the photoelectrochemistry of semiconductor–electrolyte junctions, Nature312, 21 (1984)

    work page 1984

  2. [2]

    Volkov, S

    M. Volkov, S. A. Sato, F. Schlaepfer, L. Kasmi, N. Hart- mann, M. Lucchini, L. Gallmann, A. Rubio, and U. Keller, Attosecond screening dynamics mediated by electron localization in transition metals, Nat. Phys.15, 1145 (2019)

    work page 2019

  3. [3]

    A. V. Akimov, A. J. Neukirch, and O. V. Prezhdo, The- oretical Insights into Photoinduced Charge Transfer and Catalysis at Oxide Interfaces, Chem. Rev.113, 4496 (2013)

    work page 2013

  4. [4]

    V. A. Jhalani, J.-J. Zhou, and M. Bernardi, Ultrafast Hot Carrier Dynamics in GaN and Its Impact on the Efficiency Droop, Nano Lett.17, 5012 (2017)

    work page 2017

  5. [5]

    R. Long, O. V. Prezhdo, and W. Fang, Nonadiabatic charge dynamics in novel solar cell materials, WIREs Comput Mol Sci7, e1305 (2017)

    work page 2017

  6. [6]

    M. Guan, D. Chen, S. Hu, H. Zhao, P. You, and S. Meng, Theoretical Insights into Ultrafast Dynamics in Quan- tum Materials, Ultrafast Science 10.34133/2022/9767251 (2022)

    work page doi:10.34133/2022/9767251 2022

  7. [7]

    L. Wang, J. Qiu, X. Bai, and J. Xu, Surface hop- ping methods for nonadiabatic dynamics in extended systems, WIREs Computational Molecular Science10, e1435 (2020)

    work page 2020

  8. [8]

    J. C. Tully, Molecular dynamics with electronic transi- tions, The Journal of Chemical Physics93, 1061 (1990)

    work page 1990

Show all 56 references
  1. [9]

    T. R. Nelson, A. J. White, J. A. Bjorgaard, A. E. Sifain, Y. Zhang, B. Nebgen, S. Fernandez-Alberti, D. Mozyrsky, A. E. Roitberg, and S. Tretiak, Non- adiabatic Excited-State Molecular Dynamics: Theory and Applications for Modeling Photophysics in Extended Molecular Materials,...

    2020

  2. [10]

    Polyak, L

    I. Polyak, L. Hutton, R. Crespo-Otero, M. Barbatti, and P. J. Knowles, Ultrafast Photoinduced Dynamics of 1,3- Cyclohexadiene Using XMS-CASPT2 Surface Hopping, J. Chem. Theory Comput.15, 3929 (2019)

    2019

  3. [11]

    S. Mai, P. Marquetand, and L. Gonz´ alez, Nonadiabatic dynamics: The SHARC approach, WIREs Comput Mol Sci8, e1370 (2018)

    2018

  4. [12]

    A. V. Akimov and O. V. Prezhdo, The PYXAID Pro- gram for Non-Adiabatic Molecular Dynamics in Con- densed Matter Systems, J. Chem. Theory Comput.9, 4959 (2013)

    2013

  5. [13]

    Smith and A

    B. Smith and A. V. Akimov, Modeling nonadiabatic dy- namics in condensed matter materials: some recent ad- vances and applications, J. Phys.: Condens. Matter32, 073001 (2019)

    2019

  6. [14]

    X.-Y. Liu, D. Xiao, W.-H. Fang, and G. Cui, Excited- State Wave Functions and Energies Predicted by Ma- chine Learning Based on Graph Neural Networks, J. Chem. Theory Comput. 10.1021/acs.jctc.5c00886 (2025)

    2025 doi

  7. [15]

    Shakiba and A

    M. Shakiba and A. V. Akimov, Machine-Learned Kohn–Sham Hamiltonian Mapping for Nonadiabatic Molecular Dynamics, J. Chem. Theory Comput.20, 2992 (2024)

    2024

  8. [16]

    Westermayr, M

    J. Westermayr, M. Gastegger, and P. Marquetand, Combining SchNet and SHARC: The SchNarc Machine Learning Approach for Excited-State Dynamics, J. Phys. Chem. Lett.11, 3828 (2020)

    2020

  9. [17]

    P. O. Dral, F. Ge, Y.-F. Hou, P. Zheng, Y. Chen, M. Bar- batti, O. Isayev, C. Wang, B.-X. Xue, M. Pinheiro Jr, Y. Su, Y. Dai, Y. Chen, L. Zhang, S. Zhang, A. Ul- lah, Q. Zhang, and Y. Ou, MLatom 3: A Platform for Machine Learning-Enhanced Computational Chemistry Simulations an...

    2024

  10. [18]

    Mausenberger, C

    S. Mausenberger, C. M¨ uller, A. Tkatchenko, P. Marque- tand, L. Gonz´ alez, and J. Westermayr, SPAINN: equiv- ariant message passing for excited-state nonadiabatic molecular dynamics, Chem. Sci.15, 15880 (2024)

    2024

  11. [19]

    J. Li, P. Reiser, B. R. Boswell, A. Eberhard, N. Z. Burns, P. Friederich, and S. A. Lopez, Automatic discov- ery of photoisomerization mechanisms with nanosecond machine learning photodynamics simulations, Chem. Sci. 12, 5302 (2021)

    2021

  12. [20]

    Z. Wang, J. Dong, J. Qiu, and L. Wang, All-Atom Nonadiabatic Dynamics Simulation of Hybrid Graphene Nanoribbons Based on Wannier Analysis and Machine Learning, ACS Applied Materials & Interfaces14, 22929 (2022)

    2022

  13. [21]

    D. Liu, B. Wang, Y. Wu, A. S. Vasenko, and O. V. Prezhdo, Breaking the size limitation of nonadiabatic molecular dynamics in condensed matter systems with local descriptor machine learning, Proceedings of the Na- tional Academy of Sciences121, e2403497121 (2024)

    2024

  14. [22]

    Zhang, Y

    C. Zhang, Y. Zhong, Z.-G. Tao, X. Qin, H. Shang, Z. Lan, O. V. Prezhdo, X.-G. Gong, W. Chu, and H. Xiang, Ad- vancing nonadiabatic molecular dynamics simulations in solids with E(3) equivariant deep neural hamiltonians, Nat Commun16, 2033 (2025)

    2025

  15. [23]

    Martyka, X.-Y

    M. Martyka, X.-Y. Tong, J. Jankowska, and P. O. Dral, OMNI-P2x: A Universal Neural Network Potential for Excited-State Simulations (2025)

    2025

  16. [24]

    Barrett, C

    R. Barrett, C. Ortner, and J. Westermayr, Transferable Machine Learning Potential X-MACE for Excited States using Integrated DeepSets (2025)

    2025

  17. [25]

    Martinka, L

    J. Martinka, L. Zhang, Y.-F. Hou, M. Martyka, J. Pit- tner, M. Barbatti, and P. O. Dral, A Descriptor Is All You Need: Accurate Machine Learning of Nonadiabatic Coupling Vectors (2025)

    2025

  18. [26]

    X. Li, N. Lubbers, S. Tretiak, K. Barros, and Y. Zhang, Machine Learning Framework for Modeling Exciton Po- laritons in Molecular Materials, J. Chem. Theory Com- put.20, 891 (2024)

    2024

  19. [27]

    Axelrod, E

    S. Axelrod, E. Shakhnovich, and R. G´ omez-Bombarelli, Excited state non-adiabatic dynamics of large photo- switchable molecules using a chemically transferable ma- chine learning potential, Nat Commun13, 3440 (2022)

    2022

  20. [28]

    C. F. Craig, W. R. Duncan, and O. V. Prezhdo, Tra- jectory Surface Hopping in the Time-Dependent Kohn- Sham Approach for Electron-Nuclear Dynamics, Phys. 7 Rev. Lett.95, 163001 (2005)

    2005

  21. [29]

    Gavnholt, Delta self-consistent field method to obtain potential energy surfaces of excited molecules on surfaces, Phys

    J. Gavnholt, Delta self-consistent field method to obtain potential energy surfaces of excited molecules on surfaces, Phys. Rev. B78, 10.1103/PhysRevB.78.075441 (2008)

    2008 doi

  22. [30]

    Yang and P

    W. Yang and P. W. Ayers, Foundation for theδSCF Ap- proach in Density Functional Theory (2024)

    2024

  23. [31]

    C. Gao, L. Zhang, Q. Zheng, and J. Zhao, Tuning the Lifetime of Photoexcited Small Polarons on Rutile TiO2 Surface via Molecular Adsorption, J. Phys. Chem. C125, 27275 (2021)

    2021

  24. [32]

    Vandaele, M

    E. Vandaele, M. Maliˇ s, and S. Luber, TheδSCF method for non-adiabatic dynamics of systems in the liquid phase, J. Chem. Phys.156, 10.1063/5.0083340 (2022)

    2022 doi

  25. [33]

    [22, 34-38, 40-44, 51, 53-54]

    See Supplemental Material [url] for computational and training details, additional benchmarks, a de- tailed derivation and algorithmic flowchart of on-the-fly N2AMD, a comprehensive comparison with our previous work and further discussions, which includes Refs. [22, 34-38, 40-...

  26. [34]

    Pulay, Ab initio calculation of force constants and equilibrium geometries in polyatomic molecules: I

    P. Pulay, Ab initio calculation of force constants and equilibrium geometries in polyatomic molecules: I. The- ory, Molecular Physics17, 197 (1969)

    1969

  27. [35]

    F. Chen, T. Ma, T. Zhang, Y. Zhang, and H. Huang, Atomic-Level Charge Separation Strategies in Semiconductor-Based Photocatalysts, Advanced Materi- als33, 2005256 (2021)

    2021

  28. [36]

    Ozaki, Variationally optimized atomic orbitals for large-scale electronic structures, Physical Review B67, 155108 (2003)

    T. Ozaki, Variationally optimized atomic orbitals for large-scale electronic structures, Physical Review B67, 155108 (2003)

    2003

  29. [37]

    P. Lin, X. Ren, X. Liu, and L. He, Ab initio elec- tronic structure calculations based on numerical atomic orbitals: Basic fomalisms and recent progresses, WIREs Computational Molecular Science14, e1687 (2024)

    2024

  30. [38]

    P. Lin, X. Ren, and L. He, Accuracy of Localized Reso- lution of the Identity in Periodic Hybrid Functional Cal- culations with Numerical Atomic Orbitals, The Journal of Physical Chemistry Letters11, 3082 (2020)

    2020

  31. [39]

    Zheng, W

    Q. Zheng, W. Chu, C. Zhao, L. Zhang, H. Guo, Y. Wang, X. Jiang, and J. Zhao, Ab initio nonadiabatic molecular dynamics investigations on the excited carriers in con- densed matter systems, WIREs Computational Molecu- lar Science9, e1411 (2019)

    2019

  32. [40]

    E. Abad, J. P. Lewis, V. Zobaˇ c, P. Hapala, P. Jel ´ ınek, and J. Ortega, Calculation of non-adiabatic coupling vec- tors in a local-orbital basis set, The Journal of Chemical Physics138, 154106 (2013)

    2013

  33. [41]

    Zhong, H

    Y. Zhong, H. Yu, M. Su, X. Gong, and H. Xiang, Trans- ferable equivariant graph neural networks for the Hamil- tonians of molecules and solids, npj Comput Mater9, 1 (2023)

    2023

  34. [42]

    Musaelian, S

    A. Musaelian, S. Batzner, A. Johansson, L. Sun, C. J. Owen, M. Kornbluth, and B. Kozinsky, Learning local equivariant representations for large-scale atomistic dy- namics, Nat Commun14, 579 (2023)

    2023

  35. [43]

    A. V. Akimov, A Simple Phase Correction Makes a Big Difference in Nonadiabatic Molecular Dynamics, J. Phys. Chem. Lett.9, 6096 (2018)

    2018

  36. [44]

    Zheng, W

    Q. Zheng, W. A. Saidi, Y. Xie, Z. Lan, O. V. Prezhdo, H. Petek, and J. Zhao, Phonon-Assisted Ultrafast Charge Transfer at van der Waals Heterostructure Interface, Nano Lett.17, 6435 (2017)

    2017

  37. [45]

    where ultrafast manipulation via laser pulses requires accurate theoretical simulation [46–50]. We investigate photo-induced ferroelectric switching in bilayer hexago- nal boron nitride (h-BN), focusing on how photoexcita- tion tunes the energy barrier between AB and BA stack-...

  38. [46]

    R. Bian, R. He, E. Pan, Z. Li, G. Cao, P. Meng, J. Chen, Q. Liu, Z. Zhong, W. Li, and F. Liu, Developing fatigue- resistant ferroelectrics using interlayer sliding switching, Science385, 57 (2024)

    2024

  39. [47]

    E. J. Sie, C. M. Nyby, C. D. Pemmaraju, S. J. Park, X. Shen, J. Yang, M. C. Hoffmann, B. K. Ofori-Okai, R. Li, A. H. Reid, S. Weathersby, E. Mannebach, N. Finney, D. Rhodes, D. Chenet, A. Antony, L. Bali- cas, J. Hone, T. P. Devereaux, T. F. Heinz, X. Wang, and A. M. Lindenber...

    2019

  40. [48]

    Gao and L

    L. Gao and L. Bellaiche, Large Photoinduced Tuning of Ferroelectricity in Sliding Ferroelectrics, Phys. Rev. Lett. 133, 196801 (2024)

    2024

  41. [49]

    Yang and S

    Q. Yang and S. Meng, Light-Induced Complete Rever- sal of Ferroelectric Polarization in Sliding Ferroelectrics, Phys. Rev. Lett.133, 136902 (2024)

    2024

  42. [50]

    Z.-G. Tao, S. Deng, O. V. Prezhdo, H. Xiang, W. Chu, and X.-G. Gong, Tunable Ultrafast Charge Transfer across Homojunction Interface, Journal of the American Chemical Society 10.1021/jacs.4c07454 (2024)

    2024 doi

  43. [51]

    Yao, Z.-G

    D. Yao, Z.-G. Tao, C. Zhang, L. Peng, X.-G. Gong, and W. Chu, Reprogrammable Carrier Lifetimes in 2D Mate- rials via Ultrafast Ferroelectric Switching, J. Am. Chem. Soc.147, 46197 (2025)

    2025

  44. [52]

    J. P. Perdew, K. Burke, and M. Ernzerhof, Generalized Gradient Approximation Made Simple, Phys. Rev. Lett. 77, 3865 (1996)

    1996

  45. [53]

    Franchini, M

    C. Franchini, M. Reticcioli, M. Setvin, and U. Diebold, Polarons in materials, Nat Rev Mater6, 560 (2021)

    2021

  46. [54]

    A. R. Elmaslmane, M. B. Watkins, and K. P. McKenna, First-Principles Modeling of Polaron Formation in TiO 2 Polymorphs, J. Chem. Theory Comput.14, 3740 (2018)

    2018

  47. [55]

    J. Heyd, G. E. Scuseria, and M. Ernzerhof, Hybrid func- tionals based on a screened Coulomb potential, The Jour- nal of Chemical Physics118, 8207 (2003)

    2003

  48. [56]

    Nonadiabatic Molecular Dynamics on Real-time Excited-State Sur- faces via Machine Learning Hamiltonians

    C. Zhang, Y. Zhong, Z.-G. Tao, Y. Li, Z. Lan, O. V. Prezhdo, X.-G. Gong, W. Chu, and H. Xi- ang, QDyn code and original data for “Nonadiabatic Molecular Dynamics on Real-time Excited-State Sur- faces via Machine Learning Hamiltonians”,Zenodo https://doi.org/10.5281/zenodo.2027...

    2026 doi

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.