pith. sign in

arxiv: 2602.06760 · v1 · submitted 2026-02-06 · ⚛️ physics.chem-ph

Mimyria: Machine learned vibrational spectroscopy for aqueous systems made simple

Philipp Schienbein This is my paper

Pith reviewed 2026-05-16 06:26 UTC · model grok-4.3

classification ⚛️ physics.chem-ph
keywords machine learningvibrational spectroscopyIR spectroscopyRaman spectroscopymolecular dynamicsaqueous systemsresponse tensors
0
0 comments X

The pith

Mimyria trains machine learning models on atomic response tensors to generate accurate IR and Raman spectra from molecular dynamics simulations using small datasets.

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

The paper presents mimyria as a framework that combines electronic structure calculations with machine learning to compute vibrational spectra efficiently. It introduces the polarizability gradient tensor as a new target property for Raman spectroscopy alongside the atomic polar tensor for IR. Models trained on small sets of aqueous data produce spectra that agree well with explicit ab initio references, with spectral convergence occurring faster than error metrics like RMSE suggest. This integration allows for data-efficient and reliable vibrational spectroscopy in condensed phases without needing extensive computations for each trajectory point.

Core claim

By integrating response-tensor learning, automated training, and spectral-domain validation into a unified workflow, mimyria enables data-efficient and quantitatively reliable vibrational spectroscopy for aqueous systems.

What carries the argument

The polarizability gradient tensor (PGT) as an atom-resolved machine-learning target property for Raman spectroscopy, complementing the atomic polar tensor (APT) for IR, within an automated workflow that validates at the spectral level.

If this is right

  • IR and Raman spectra can be obtained from MD trajectories using ML surrogates instead of direct ab initio calculations at every step.
  • Spectral agreement improves more rapidly than the root-mean-square error of the underlying models.
  • Practical guidelines and early-stopping criteria can be used based on relating model errors to observable accuracy.
  • Small training sets are sufficient for convergence to reference spectra on aqueous benchmarks.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Extending the approach to non-aqueous systems or more complex mixtures could broaden its applicability to biological environments.
  • Relating RMSE to spectral fidelity might inform error estimation in other machine learning applications to spectroscopy.
  • Testing on larger systems would check if the data efficiency holds beyond benchmark cases.

Load-bearing premise

Machine learning models trained on small sets of aqueous benchmark data will generalize to produce spectra matching explicit ab initio references at the level of observable accuracy.

What would settle it

A direct comparison showing that ML-generated spectra deviate significantly from ab initio spectra for an aqueous system outside the training benchmarks, even when model RMSE is low.

Figures

Figures reproduced from arXiv: 2602.06760 by Philipp Schienbein.

Figure 1
Figure 1. Figure 1: RMSE (top panels) and relative RMSE, δ P rmse, see Eq. 14 (bottom panels), when computing APTs from taking the derivative of atomic forces with respect to an electric field (a), Eq. 21, last term, or from taking the derivative of the total dipole moment with respect to an atomic displacement in space (b), Eq. 21, central term, both evaluated numerically by central finite differences using the field strengt… view at source ↗
Figure 2
Figure 2. Figure 2: RMSE (top panels) and relative RMSE, δ Q rmse, see Eq. 14 (bottom panels), when computing PGTs from taking the derivative of atomic forces with respect to an electric field (a), Eq. 22, last term, or from taking the derivative of the polarizability tensor with respect to an atomic displacement in space (b), Eq. 22, central term, both evaluated numerically by central finite differences using the displacemen… view at source ↗
Figure 3
Figure 3. Figure 3: Parity plots comparing APTs (a) and PGTs (b) obtained from numeri [PITH_FULL_IMAGE:figures/full_fig_p017_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: APTNN performance as a function of training set size for a SO [PITH_FULL_IMAGE:figures/full_fig_p018_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: (a) Comparison of the explicit ab initio spectrum (black) with machine-learning spectra obtained from APTNNs trained on 10 (red, upward triangles) and 200 training configurations (blue, downward triangles). All spectra are computed from the same 20 ps MLMD trajectory, with APTs evaluated either from DFT reference calculations or predicted by the trained models. (b) Predicted spectra obtained from 80 indepe… view at source ↗
Figure 6
Figure 6. Figure 6: PGTNN performance as a function of training set size for a SO [PITH_FULL_IMAGE:figures/full_fig_p021_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: (a) Comparison of the explicit ab initio spectrum (black) with machine-learning spectra obtained from PGTNNs trained on 10 (red, upward triangles) and 200 training configurations (blue, downward triangles). All spectra are computed from the same 20 ps MLMD trajectory, with PGTs evaluated either from DFT reference calculations or predicted by the trained models. (b) Predicted spectra obtained from 80 indepe… view at source ↗
Figure 8
Figure 8. Figure 8: Isotropic (Eq. 9, orange), parallel (Eq. 11, black), and perpendic [PITH_FULL_IMAGE:figures/full_fig_p023_8.png] view at source ↗
read the original abstract

Vibrational spectroscopy provides a powerful connection between molecular dynamics (MD) simulations and experiment, but its routine use in condensed-phase systems remains limited. We introduce mimyria, a modular and automated framework that orchestrates electronic-structure reference calculations, trains atom-resolved machine-learning response models, and generates IR and Raman spectra from MD trajectories within a unified workflow. We introduce the polarizability gradient tensor (PGT) as a novel atom-resolved machine-learning target property for Raman spectroscopy, complementing the established atomic polar tensor (APT) for IR spectroscopy. As a necessary prerequisite, we demonstrate how both PGTs and APTs can accurately be computed from electronic-structure theory, validate them across formally equivalent derivative formulations, and thereby benchmark their numerical consistency. We then employ machine learning as an efficient surrogate to represent the validated APT and PGT response functions on aqueous benchmark systems. We validate the trained models directly at the level of the spectrum against explicit ab initio reference calculations and find that IR and Raman spectra converge with surprisingly small training sets. Moreover, spectral agreement improves more rapidly than the root-mean-square error (RMSE). While RMSE is straightforward to compute, statistically converged reference spectra are generally impractical to obtain, motivating the need to relate model-level errors to observable-level accuracy. By connecting these complementary error measures, we provide practical guidelines and early-stopping criteria for achieving sufficient spectral fidelity. By integrating response-tensor learning, automated training, and spectral-domain validation into a unified workflow, mimyria enables data-efficient and quantitatively reliable vibrational spectroscopy.

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

2 major / 2 minor

Summary. The manuscript introduces mimyria, a modular automated framework that performs electronic-structure reference calculations, trains atom-resolved machine-learning models for the atomic polar tensor (APT) and the newly introduced polarizability gradient tensor (PGT), and generates IR and Raman spectra from MD trajectories. It validates APT and PGT consistency across derivative formulations from electronic structure theory, trains ML surrogates on small aqueous benchmark sets, and reports that spectra converge with small training data while spectral agreement improves faster than RMSE, supplying practical early-stopping guidelines that relate model errors to observable accuracy.

Significance. If the central claims hold, the work offers a practical route to data-efficient, quantitatively reliable vibrational spectroscopy for aqueous condensed-phase systems by replacing expensive ab initio response calculations with ML surrogates and by validating directly at the spectral level rather than only at the tensor level. The unified workflow and the emphasis on observable-level fidelity over raw RMSE are strengths that could make routine simulation-to-experiment comparison more accessible.

major comments (2)
  1. [Results and validation sections] The spectral validation is performed exclusively on the same aqueous benchmark systems used for training. No explicit out-of-distribution tests (different temperatures, ion concentrations, or larger clusters) are described, leaving the transferability of local APT/PGT predictions to global spectra untested in regimes where error accumulation could exceed the reported benchmark agreement.
  2. [Discussion of error measures and guidelines] The claim that spectra converge with 'surprisingly small' training sets and that spectral agreement improves faster than RMSE rests on the premise that benchmark agreement implies observable accuracy. Because statistically converged reference spectra are acknowledged as impractical, the early-stopping guidelines require a clearer quantitative mapping from tensor RMSE to spectral error that is demonstrated beyond the training distribution.
minor comments (2)
  1. [Methods on response-tensor computation] The abstract states that PGTs and APTs are validated 'across formally equivalent derivative formulations' but the main text should include a brief table or equation set explicitly showing the numerical consistency metrics (e.g., maximum deviations) for both tensors.
  2. [Introduction of PGT] Notation for the PGT is introduced as novel; a short comparison paragraph to existing polarizability-derivative approaches would help readers assess its incremental value.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed report. We address each major comment below and have revised the manuscript to strengthen the presentation of validation and error analysis where feasible.

read point-by-point responses

  1. Referee: [Results and validation sections] The spectral validation is performed exclusively on the same aqueous benchmark systems used for training. No explicit out-of-distribution tests (different temperatures, ion concentrations, or larger clusters) are described, leaving the transferability of local APT/PGT predictions to global spectra untested in regimes where error accumulation could exceed the reported benchmark agreement.

    Authors: We agree that explicit out-of-distribution tests would further strengthen the claims. The current work focuses on establishing the framework, validating numerical consistency of APT/PGT, and demonstrating data-efficient convergence on standard aqueous benchmarks. Because APT and PGT are strictly local, atom-centered quantities, their transferability follows from the locality of the underlying electronic response and the atom-resolved descriptors; we have added a dedicated paragraph in the Discussion section explaining this expectation and citing supporting literature on local response models. Explicit tests at varied temperatures, ion concentrations, or larger clusters are outside the scope of the present manuscript but represent a natural next step that we now flag as future work. revision: partial

  2. Referee: [Discussion of error measures and guidelines] The claim that spectra converge with 'surprisingly small' training sets and that spectral agreement improves faster than RMSE rests on the premise that benchmark agreement implies observable accuracy. Because statistically converged reference spectra are acknowledged as impractical, the early-stopping guidelines require a clearer quantitative mapping from tensor RMSE to spectral error that is demonstrated beyond the training distribution.

    Authors: We appreciate the call for a more explicit mapping. In the original manuscript we already compare spectra generated from models trained on progressively larger sets against ab initio reference spectra on the same systems, showing that key spectral features stabilize at lower RMSE than the tensor error itself. To address the referee’s point directly, we have added a new supplementary figure and accompanying text that quantifies the correlation between tensor RMSE and two spectral error metrics (integrated absolute deviation and peak-position shift) across the full range of training-set sizes. While we cannot generate statistically converged reference spectra for every condition, the observed monotonic improvement in spectral fidelity with decreasing tensor error provides the practical early-stopping guideline we advocate; we now state this limitation more explicitly in the revised Discussion. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected in derivation chain

full rationale

The paper computes APT and PGT response tensors from electronic-structure theory, trains ML surrogates on aqueous benchmark data, and validates generated IR/Raman spectra directly against independent explicit ab initio reference calculations on the same systems. This external benchmark comparison means spectral agreement is not equivalent to the training inputs by construction. No self-definitional mappings, fitted parameters renamed as predictions, or load-bearing self-citations appear in the workflow; the early-stopping guidelines derive from relating RMSE to observable fidelity using those independent references, keeping the central claim self-contained against external electronic-structure benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The central claim depends on the numerical consistency of APT and PGT from electronic structure derivatives and on the assumption that small ML training sets suffice for spectral fidelity; no free parameters are explicitly fitted in the abstract beyond training set selection.

free parameters (1)
  • training set size
    The paper states convergence occurs with surprisingly small training sets, implying this size is chosen based on observed spectral agreement rather than a fixed rule.
axioms (1)
  • domain assumption APT and PGT response functions can be accurately computed from electronic-structure theory using formally equivalent derivative formulations
    The abstract presents this computation and cross-validation as a necessary prerequisite before ML training.
invented entities (1)
  • polarizability gradient tensor (PGT) no independent evidence
    purpose: Atom-resolved machine-learning target property for generating Raman spectra
    Introduced as novel to complement the atomic polar tensor for IR; computed from theory but treated as a new construct for the ML surrogate.

pith-pipeline@v0.9.0 · 5569 in / 1372 out tokens · 37801 ms · 2026-05-16T06:26:18.650062+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

81 extracted references · 81 canonical work pages · 2 internal anchors

  1. [1]

    Vibrational Spectroscopy as a Probe of Stru cture and Dynamics in Liquid Water

    H. J. Bakker and J. L. Skinner. “Vibrational Spectroscopy as a Probe of Stru cture and Dynamics in Liquid Water”. Chem. Rev. 110 (2010), pp. 1498–1517

    work page 2010

  2. [2]

    Vibrational Spectroscopy and Dynamics of Water

    F. Perakis, L. De Marco, A. Shalit, F. Tang, Z. R. Kann, T. D. K¨ uhne, R. Torre, M. Bonn, and Y. Nagata. “Vibrational Spectroscopy and Dynamics of Water”. Chem. Rev. 116 (2016), pp. 7590–7607

    work page 2016

  3. [3]

    Ran- dom encounters dominate water-water interactions at supercritical con ditions

    K. Mauelshagen, P. Schienbein, I. Kolling, G. Schwaab, D. Marx, and M . Havenith. “Ran- dom encounters dominate water-water interactions at supercritical con ditions”. Sci. Adv. 11 (2025), eadp8614

    work page 2025

  4. [4]

    Dissecting the THz spectrum of liquid water from first principles via correlations in time and space

    M. Heyden, J. Sun, S. Funkner, G. Mathias, H. Forbert, M. Havenit h, and D. Marx. “Dissecting the THz spectrum of liquid water from first principles via correlations in time and space”. Proc. Natl. Acad. Sci. USA 107 (2010), pp. 12068–12073

    work page 2010

  5. [5]

    D. R. Galimberti, S. Bougueroua, J. Mah´ e, M. Tommasini, A. M. Rijs, an d M.-P. Gaigeot. “Conformational assignment of gas phase peptides and their H-bonded complexe s using far- IR/THz: IR-UV ion dip experiment, DFT-MD spectroscopy, and graph theor y for mode assignment”. Faraday Discuss. 217 (2019), pp. 67–97

    work page 2019

  6. [6]

    Vibrational spectroscopies in l iquid water: on tem- perature and coordination effects in Raman and infrared spectroscopies

    R. Vuilleumier and A. P. Seitsonen. “Vibrational spectroscopies in l iquid water: on tem- perature and coordination effects in Raman and infrared spectroscopies”. Condens. Matter Phys. 26 (2023), p. 33301

    work page 2023

  7. [7]

    Dissecting the hydrogen bond network of water: Charge transfer and nucl ear quantum effects

    M. Fl´ or, D. M. Wilkins, M. de la Puente, D. Laage, G. Cassone, A. Hassanal i, and S. Roke. “Dissecting the hydrogen bond network of water: Charge transfer and nucl ear quantum effects”. Science 386 (2024)

    work page 2024

  8. [8]

    THz Pump Pulse-Driven Temporal Respon se of Liquid Water Probed by Machine-Learning-Accelerated Non-Equilibrium Molecular Dy namics

    K. Joll and P. Schienbein. “THz Pump Pulse-Driven Temporal Respon se of Liquid Water Probed by Machine-Learning-Accelerated Non-Equilibrium Molecular Dy namics”. J. Phys. Chem. Lett. 16 (2025), pp. 9183–9193

    work page 2025

  9. [9]

    Observation of water dangling OH bonds around dissolved nonpolar groups

    P. N. Perera, K. R. Fega, C. Lawrence, E. J. Sundstrom, J. Tomlinson- Phillips, and D. Ben-Amotz. “Observation of water dangling OH bonds around dissolved nonpolar groups”. Proc. Natl. Acad. Sci. USA 106 (2009), pp. 12230–12234

    work page 2009

  10. [10]

    Low-frequency modes of aqueous alkali h alide solutions: Glimpsing the hydrogen bonding vibration

    I. A. Heisler and S. R. Meech. “Low-frequency modes of aqueous alkali h alide solutions: Glimpsing the hydrogen bonding vibration”. Science 327 (2010), pp. 857–860

    work page 2010

  11. [11]

    Cor relations in the Solute–Solvent Dynamics Reach Beyond the First Hydration Shell of Ions

    P. Schienbein, G. Schwaab, H. Forbert, M. Havenith, and D. Marx. “Cor relations in the Solute–Solvent Dynamics Reach Beyond the First Hydration Shell of Ions”. J. Phys. Chem. Lett. 8 (2017), pp. 2373–2380

    work page 2017

  12. [12]

    Hydration-Shell Vibrational Spectroscopy

    D. Ben-Amotz. “Hydration-Shell Vibrational Spectroscopy”. J. Am. Chem. Soc. 141 (2019), pp. 10569–10580

    work page 2019

  13. [13]

    Ion Hydration and Ion Pairing as Probed by THz Spectroscopy

    G. Schwaab, F. Sebastiani, and M. Havenith. “Ion Hydration and Ion Pairing as Probed by THz Spectroscopy”. Angew. Chem. Int. Ed. 58 (2019), pp. 3000–3013

    work page 2019

  14. [14]

    Ab Initio Molecular Dynamics Computation of the Infrared Spectrum of Aqueous Uracil

    M.-P. Gaigeot and M. Sprik. “Ab Initio Molecular Dynamics Computation of the Infrared Spectrum of Aqueous Uracil”. J. Phys. Chem. B 107 (2003), pp. 10344–10358. 24

    work page 2003

  15. [15]

    Extract ing effective normal modes from equilibrium dynamics at finite temperature

    M. Martinez, M.-P. Gaigeot, D. Borgis, and R. Vuilleumier. “Extract ing effective normal modes from equilibrium dynamics at finite temperature”. J. Chem. Phys. 125 (2006)

    work page 2006

  16. [16]

    Un- derstanding THz Spectra of Aqueous Solutions: Glycine in Light and Heavy Water

    J. Sun, G. Niehues, H. Forbert, D. Decka, G. Schwaab, D. Marx, and M. Ha venith. “Un- derstanding THz Spectra of Aqueous Solutions: Glycine in Light and Heavy Water”. J. Am. Chem. Soc. 136 (2014), pp. 5031–5038

    work page 2014

  17. [17]

    Pressure Response of the THz Spectrum of Bulk Li quid Water Revealed by Intermolecular Instantaneous Normal Mode Analysis

    S. Imoto and D. Marx. “Pressure Response of the THz Spectrum of Bulk Li quid Water Revealed by Intermolecular Instantaneous Normal Mode Analysis”. J. Chem. Phys. 150 (2019), p. 084502

    work page 2019

  18. [18]

    Dipole moment derivatives and infrared i ntensities. I. Polar tensors

    W. B. Person and J. H. Newton. “Dipole moment derivatives and infrared i ntensities. I. Polar tensors”. J. Chem. Phys. 61 (1974), pp. 1040–1049

    work page 1974

  19. [19]

    Dielectric tensor, effectiv e charges, and phonons in α -quartz by variational density-functional perturbation theory

    X. Gonze, D. C. Allan, and M. P. Teter. “Dielectric tensor, effectiv e charges, and phonons in α -quartz by variational density-functional perturbation theory”. Phys. Rev. Lett. 68 (1992), pp. 3603–3606

    work page 1992

  20. [20]

    Infrared spectr oscopy in the gas and liquid phase from first principle molecular dynamics simulations: ap plication to small pep- tides

    M.-P. Gaigeot, M. Martinez, and R. Vuilleumier. “Infrared spectr oscopy in the gas and liquid phase from first principle molecular dynamics simulations: ap plication to small pep- tides”. Mol. Phys. 105 (2007), pp. 2857–2878

    work page 2007

  21. [21]

    Sum Frequency Generation Spectra from Velocity-Velocity Correlation Functions

    R. Khatib and M. Sulpizi. “Sum Frequency Generation Spectra from Velocity-Velocity Correlation Functions”. J. Phys. Chem. Lett. 8 (2017), pp. 1310–1314

    work page 2017

  22. [22]

    Spectroscopy from Machine Learning by Accurately R epresenting the Atomic Polar Tensor

    P. Schienbein. “Spectroscopy from Machine Learning by Accurately R epresenting the Atomic Polar Tensor”. J. Chem. Theory Comput. 19 (2023), pp. 705–712

    work page 2023

  23. [23]

    Rattling in the Cage: Ions as Probe ob Sub-picosecond Wate r Network Dynamics

    D. A. Schmidt, ¨O. Birer, S. Funkner, B. Born, R. Gnanasekaran, G. Schwaab, D. M. Leitn er, and M. Havenith. “Rattling in the Cage: Ions as Probe ob Sub-picosecond Wate r Network Dynamics”. J. Am. Chem. Soc. 131 (2009), pp. 18512–18517

    work page 2009

  24. [24]

    Low-frequency dynamics of aqueous alkali chloride solutions as probed by terahertz spectroscopy

    Z. R. Kann and J. L. Skinner. “Low-frequency dynamics of aqueous alkali chloride solutions as probed by terahertz spectroscopy”. J. Chem. Phys. 144 (2016)

    work page 2016

  25. [25]

    Ab initio in frared spectrum of liquid water

    P. L. Silvestrelli, M. Bernasconi, and M. Parrinello. “Ab initio in frared spectrum of liquid water”. Chem. Phys. Lett. 277 (1997), pp. 478–482

    work page 1997

  26. [26]

    Combining Static and Dynamical Approaches for Infrared Spectra Calculations of Gas Ph ase Molecules and Clusters

    D. R. Galimberti, A. Milani, M. Tommasini, C. Castiglioni, and M.-P. Gaigeot. “Combining Static and Dynamical Approaches for Infrared Spectra Calculations of Gas Ph ase Molecules and Clusters”. J. Chem. Theory Comput. 13 (2017), pp. 3802–3813

    work page 2017

  27. [27]

    Comput ing vibrational spectra from ab initio molecular dynamics

    M. Thomas, M. Brehm, R. Fligg, P. V¨ ohringer, and B. Kirchner. “Comput ing vibrational spectra from ab initio molecular dynamics”. Phys. Chem. Chem. Phys. 15 (2013), pp. 6608– 6622

    work page 2013

  28. [28]

    Generalized Neural-Network Represen tation of High-Dimensional Potential-Energy Surfaces

    J. Behler and M. Parrinello. “Generalized Neural-Network Represen tation of High-Dimensional Potential-Energy Surfaces”. Phys. Rev. Lett. 98 (2007), p. 146401

    work page 2007

  29. [29]

    Four Generations of High-Dimensional Neural Network Potenti als

    J. Behler. “Four Generations of High-Dimensional Neural Network Potenti als”. Chem. Rev. 121 (2021), pp. 10037–10072. 25

    work page 2021

  30. [30]

    Gaussian Approxi mation Poten- tials: The Accuracy of Quantum Mechanics, without the Electrons

    A. P. Bart´ ok, M. C. Payne, R. Kondor, and G. Cs´ anyi. “Gaussian Approxi mation Poten- tials: The Accuracy of Quantum Mechanics, without the Electrons”. Phys. Rev. Lett. 104 (2010), p. 136403

    work page 2010

  31. [31]

    E(3)-equivariant graph neural networks f or data-efficient and accurate interatomic potentials

    S. Batzner, A. Musaelian, L. Sun, M. Geiger, J. P. Mailoa, M. Kornbluth , N. Molinari, T. E. Smidt, and B. Kozinsky. “E(3)-equivariant graph neural networks f or data-efficient and accurate interatomic potentials”. Nat. Commun. 13 (2022), p. 2453

    work page 2022

  32. [32]

    MACE: Hi gher Order Equiv- ariant Message Passing Neural Networks for Fast and Accurate Force Fields

    I. Batatia, D. P. Kovacs, G. Simm, C. Ortner, and G. Csanyi. “MACE: Hi gher Order Equiv- ariant Message Passing Neural Networks for Fast and Accurate Force Fields”. Advances in Neural Information Processing Systems . Ed. by S. Koyejo, S. Mohamed, A. Agarwal, D. Belgrave, K. Cho, and A. Oh. Vol. 35. Curran Associates, Inc., 2022, pp. 11423–11436

    work page 2022

  33. [33]

    Graph Atomic Cluster Expan sion for Semilocal Interactions beyond Equivariant Message Passing

    A. Bochkarev, Y. Lysogorskiy, and R. Drautz. “Graph Atomic Cluster Expan sion for Semilocal Interactions beyond Equivariant Message Passing”. Phys. Rev. X 14 (2024), p. 021036

    work page 2024

  34. [34]

    Machine learning molecu lar dynamics for the simulation of infrared spectra

    M. Gastegger, J. Behler, and P. Marquetand. “Machine learning molecu lar dynamics for the simulation of infrared spectra”. Chem. Sci. 8 (2017), pp. 6924–6935

    work page 2017

  35. [35]

    Discovering a Transferable Charge Assignment Model Using Machine Learning

    A. E. Sifain, N. Lubbers, B. T. Nebgen, J. S. Smith, A. Y. Lokhov, O. Isay ev, A. E. Roitberg, K. Barros, and S. Tretiak. “Discovering a Transferable Charge Assignment Model Using Machine Learning”. J. Phys. Chem. Lett. 9 (2018), pp. 4495–4501

    work page 2018

  36. [36]

    Accurate molecular polarizabilities with coupled cluster theory and machine l earning

    D. M. Wilkins, A. Grisafi, Y. Yang, K. U. Lao, R. A. DiStasio, and M. Ceriotti. “Accurate molecular polarizabilities with coupled cluster theory and machine l earning”. Proc. Natl. Acad. Sci. U.S.A. 116 (2019), pp. 3401–3406

    work page 2019

  37. [37]

    Using Gaussian proc ess regression to simulate the vibrational Raman spectra of molecular crystals

    N. Raimbault, A. Grisafi, M. Ceriotti, and M. Rossi. “Using Gaussian proc ess regression to simulate the vibrational Raman spectra of molecular crystals”. New J. Phys. 21 (2019), p. 105001

    work page 2019

  38. [38]

    Machine learning of solvent effects on molecular spectra and reactions

    M. Gastegger, K. T. Sch¨ utt, and K.-R. M¨ uller. “Machine learning of solvent effects on molecular spectra and reactions”. Chem. Sci. 12 (2021), pp. 11473–11483

    work page 2021

  39. [39]

    First-pri nciples spectroscopy of aqueous interfaces using machine-learned electronic and quantum nu clear effects

    V. Kapil, D. P. Kov´ acs, G. Cs´ anyi, and A. Michaelides. “First-pri nciples spectroscopy of aqueous interfaces using machine-learned electronic and quantum nu clear effects”. Faraday Discuss. 249 (2024), pp. 50–68

    work page 2024

  40. [40]

    Tensorial Properties via the Neuroevolution Pote ntial Framework: Fast Simulation of Infrared and Raman Spectra

    N. Xu, P. Rosander, C. Sch¨ afer, E. Lindgren, N. ¨Osterbacka, M. Fang, W. Chen, Y. He, Z. Fan, and P. Erhart. “Tensorial Properties via the Neuroevolution Pote ntial Framework: Fast Simulation of Infrared and Raman Spectra”. J. Chem. Theory Comput. 20 (2024), pp. 3273–3284

    work page 2024

  41. [41]

    Derivative learning of tensorial quantities—Predicting finite temperature infrared spectra from first principles

    B. Schmiedmayer and G. Kresse. “Derivative learning of tensorial quantities—Predicting finite temperature infrared spectra from first principles”. J. Chem. Phys. 161 (2024)

    work page 2024

  42. [42]

    Electric-field driven nuclear dynamics of liquids and solids from a multi-valued machine-learned dipolar model

    E. Stocco, C. Carbogno, and M. Rossi. “Electric-field driven nuclear dynamics of liquids and solids from a multi-valued machine-learned dipolar model”. npj Computational Materials 11 (2025)

    work page 2025

  43. [43]

    A beginner’s guide to the modern theory of polarization

    N. A. Spaldin. “A beginner’s guide to the modern theory of polarization”. J. Solid State. Chem. 195 (2012), pp. 2–10. 26

    work page 2012

  44. [44]

    Computing Bulk Phase IR Spectra from Fi- nite Cluster Data via Equivariant Neural Networks

    A. Jindal, P. Schienbein, B. Das, and D. Marx. “Computing Bulk Phase IR Spectra from Fi- nite Cluster Data via Equivariant Neural Networks”. J. Chem. Theory Comput. 21 (2025), pp. 5382–5388

    work page 2025

  45. [45]

    Machine learning interatom ic potential can infer electrical response

    P. Zhong, D. Kim, D. S. King, and B. Cheng. “Machine learning interatom ic potential can infer electrical response”. npj Comput. Mater. 11 (2025)

    work page 2025

  46. [46]

    St¨ ark, H

    P. St¨ ark, H. Stooß, M. F. Langer, E. Rumiantsev, A. Schlaich, M. Ceriot ti, and P. Loche. Simultaneous Learning of Static and Dynamic Charges . arXiv:2601.03656. 2026

    work page internal anchor Pith review arXiv 2026

  47. [47]

    Molecular d ynamics simulation with finite electric fields using Perturbed Neural Network Potenti als

    K. Joll, P. Schienbein, K. M. Rosso, and J. Blumberger. “Molecular d ynamics simulation with finite electric fields using Perturbed Neural Network Potenti als”. Nat. Commun. 15 (2024), p. 8192

    work page 2024

  48. [48]

    Committee neural network pot entials control generalization errors and enable active learning

    C. Schran, K. Brezina, and O. Marsalek. “Committee neural network pot entials control generalization errors and enable active learning”. J. Chem. Phys. 153 (2020), p. 104105

    work page 2020

  49. [49]

    Machine learning potentials for complex aqueous systems made simple

    C. Schran, F. L. Thiemann, P. Rowe, E. A. M¨ uller, O. Marsalek, and A. Michaelides. “Machine learning potentials for complex aqueous systems made simple”. Proc. Natl. Acad. Sci. USA 118 (2021), e2110077118

    work page 2021

  50. [50]

    Random Sampling Versus Active Learning Algorithms for Machine Learning Potentials of Quantum Liquid Water

    N. Stolte, J. Daru, H. Forbert, D. Marx, and J. Behler. “Random Sampling Versus Active Learning Algorithms for Machine Learning Potentials of Quantum Liquid Water ”. J. Chem. Theory Comput. 21 (2025), pp. 886–899

    work page 2025

  51. [51]

    McQuarrie

    D. McQuarrie. Statistical Mechanics. University Science Books, Sausalito, 2000

    work page 2000

  52. [52]

    R. Kubo, M. Toda, and N. Hashitsume. Statistical Physics II . Springer-Verlag, Berlin, 1991

    work page 1991

  53. [53]

    Quantum corrections to classi- cal time-correlation functions: Hydrogen bonding and anharmonic floppy mod es

    R. Ram ´ ırez, T. L´ opez-Ciudad, P. P. Kumar, and D. Marx. “Quantum corrections to classi- cal time-correlation functions: Hydrogen bonding and anharmonic floppy mod es.” J. Chem. Phys. 121 (2004), pp. 3973–3983

    work page 2004

  54. [54]

    Pabst, H

    F. Pabst, H. Forbert, and D. Marx. Resolving the Nature of the Lowest-Frequency Raman Mode of Liquid Water . arXiv:2511.14940. 2025

    work page arXiv 2025

  55. [55]

    Relations between Raman Spectroscopy and Nuclear Spin Re laxation

    R. G. Gordon. “Relations between Raman Spectroscopy and Nuclear Spin Re laxation”. J. Chem. Phys. 42 (1965), pp. 3658–3665

    work page 1965

  56. [56]

    Correlation Functions for Molecular Motion

    R. Gordon. “Correlation Functions for Molecular Motion”. Advances in Magnetic Res- onance. Ed. by J. S. W AUGH. Vol. 3. Advances in Magnetic and Optical Resonance. Academic Press, 1968, pp. 1–42

    work page 1968

  57. [57]

    Raman spec - trum and polarizability of liquid water from deep neural networks

    G. M. Sommers, M. F. Calegari Andrade, L. Zhang, H. Wang, and R. Car. “Raman spec - trum and polarizability of liquid water from deep neural networks”. Phys. Chem. Chem. Phys. 22 (2020), pp. 10592–10602

    work page 2020

  58. [58]

    Voronoi dipole moments for the s imulation of bulk phase vibrational spectra

    M. Thomas, M. Brehm, and B. Kirchner. “Voronoi dipole moments for the s imulation of bulk phase vibrational spectra”. Phys. Chem. Chem. Phys. 17 (2015), pp. 3207–3213

    work page 2015

  59. [59]

    Vibrational spectroscopy by means of first-pri nciples molecular dynamics simulations

    E. Ditler and S. Luber. “Vibrational spectroscopy by means of first-pri nciples molecular dynamics simulations”. WIREs Comput. Mol. Sci. 12 (2022), e1605. 27

    work page 2022

  60. [60]

    R. P. Feynman, R. B. Leighton, and M. L. Sands. Feynman Lectures on Physics Vol. 2 . Narosa Publishing House, 2016

    work page 2016

  61. [61]

    B. J. Berne and R. Pecora. Dynamic Light Scattering With Applications to Chemistry, Bi- ology, and Physics. With applications to chemistry, biology, and p hysics. Dover ed. Mineola, NY: Dover Publications, 2000

    work page 2000

  62. [62]

    D. Long. The Raman Effect: A Unified Treatment of the Theory of Raman Scatt ering by Molecules. John Wiley & Sons, 2002

    work page 2002

  63. [63]

    Analytic calculation and analysis of atomic polar tensors for molecules and materials using the Gaussian and plane waves ap proach

    E. Ditler, C. Kumar, and S. Luber. “Analytic calculation and analysis of atomic polar tensors for molecules and materials using the Gaussian and plane waves ap proach”. J. Chem. Phys. 154 (2021), p. 104121

    work page 2021

  64. [64]

    Szabo and N

    A. Szabo and N. S. Ostlund. Modern Quantum Chemistry . Dover Publications, Inc., 1996

    work page 1996

  65. [65]

    J. D. Jackson. Classical Electrodynamics - Second Edition . John Wiley & Sons, Inc., New York, 1975

    work page 1975

  66. [66]

    Geiger, T

    M. Geiger, T. Smidt, A. M., B. K. Miller, W. Boomsma, B. Dice, K. Lap chevskyi, M. Weiler, M. Tyszkiewicz, S. Batzner, D. Madisetti, M. Uhrin, J. Frellsen , N. Jung, S. Sanborn, M. Wen, J. Rackers, M. Rød, and M. Bailey. Euclidean neural networks: e3nn . Version 0.5.0. 2022

    work page 2022

  67. [67]

    PyTorch: An I mperative Style, High-Performance Deep Learning Library

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Kill een, Z. Lin, N. Gimelshein, L. Antiga, A. Desmaison, A. Kopf, E. Yang, Z. DeVito, M. Raison, A. T ejani, S. Chilamkurthy, B. Steiner, L. Fang, J. Bai, and S. Chintala. “PyTorch: An I mperative Style, High-Performance Deep Learning Library”. Advances in Neural Information Processing S...

    work page 2019

  68. [68]

    D. P. Kingma and J. Ba. Adam: A Method for Stochastic Optimization . https://arxiv. org/abs/1412.6980. 2014

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  69. [69]

    Supercritical Water is not Hydrogen Bonde d

    P. Schienbein and D. Marx. “Supercritical Water is not Hydrogen Bonde d”. Angew. Chem. Int. Ed. 59 (2020), pp. 18578–18585

    work page 2020

  70. [70]

    Improved Adsorption Energeti cs within Density-Functional Theory using revised Perdew-Burke-Ernzerhof Functionals

    B. Hammer, L. B. Hansen, and J. K. Nørskov. “Improved Adsorption Energeti cs within Density-Functional Theory using revised Perdew-Burke-Ernzerhof Functionals”. Phys. Rev. B 59 (1999), pp. 7413–7421

    work page 1999

  71. [71]

    The perform ance of the Becke—Lee—Yang—Parr (B—LYP) density functional theory with various basis s ets

    P. M. Gill, B. G. Johnson, J. A. Pople, and M. J. Frisch. “The perform ance of the Becke—Lee—Yang—Parr (B—LYP) density functional theory with various basis s ets”. Chem. Phys. Lett. 197 (1992), pp. 499–505

    work page 1992

  72. [72]

    A Consistent and Accu rate Ab Initio Parametrization of Density Functional Dispersion Correction (DFT-D) for the 94 Elements H-Pu

    S. Grimme, J. Antony, S. Ehrlich, and H. Krieg. “A Consistent and Accu rate Ab Initio Parametrization of Density Functional Dispersion Correction (DFT-D) for the 94 Elements H-Pu”. J. Chem. Phys. 132 (2010), p. 154104

    work page 2010

  73. [73]

    Quickstep: Fast and Accurate Density Functional Calculations Using a mixed Gauss ian and Plane Waves Approach

    J. VandeVondele, M. Krack, F. Mohamed, M. Parrinello, T. Chassaing, an d J. Hutter. “Quickstep: Fast and Accurate Density Functional Calculations Using a mixed Gauss ian and Plane Waves Approach”. Comput. Phys. Commun. 167 (2005), pp. 103–128. 28

    work page 2005

  74. [74]

    CP2K : An electronic structure and molecular dynamics software package - Quickstep: Efficie nt and accurate electronic structure calculations

    T. D. K¨ uhne, M. Iannuzzi, M. D. Ben, V. V. Rybkin, P. Seewald, F. Stein, T. Laino, R. Z. Khaliullin, O. Sch¨ utt, F. Schiffmann, D. Golze, J. Wilhelm , S. Chulkov, M. H. Bani- Hashemian, V. Weber, U. Borˇ stnik, M. Taillefumier, A. S. Jakobovits, A. Lazzaro, H. Pabst, T. M¨ uller, R. Schade, M. Guidon, S. Andermatt, N. Holmberg, G. K . Schenter, A. Hehn,...

    work page 2020

  75. [75]

    Anharmonic Raman Spectra in High-Pressu re Ice from Ab Initio Simulations

    A. Putrino and M. Parrinello. “Anharmonic Raman Spectra in High-Pressu re Ice from Ab Initio Simulations”. Phys. Rev. Lett. 88 (2002), p. 176401

    work page 2002

  76. [76]

    Raman spectra from ab initio mol ecular dynamics and its application to liquid S-methyloxirane

    S. Luber, M. Iannuzzi, and J. Hutter. “Raman spectra from ab initio mol ecular dynamics and its application to liquid S-methyloxirane”. J. Chem. Phys. 141 (2014), p. 094503

    work page 2014

  77. [77]

    Water structure and solvation of osmolyt es at high hydrostatic pressure: pure water and TMAO solutions at 10 kbar versus 1 bar

    S. Imoto, H. Forbert, and D. Marx. “Water structure and solvation of osmolyt es at high hydrostatic pressure: pure water and TMAO solutions at 10 kbar versus 1 bar”. Phys. Chem. Chem. Phys. 17 (2015), pp. 24224–24237

    work page 2015

  78. [78]

    Liquid-Vapor Phase Diagram of RPBE-D3 Water : Electronic Properties along the Coexistence Curve and in the Supercritical P hase

    P. Schienbein and D. Marx. “Liquid-Vapor Phase Diagram of RPBE-D3 Water : Electronic Properties along the Coexistence Curve and in the Supercritical P hase”. J. Phys. Chem. B 122 (2018), pp. 3318–3329

    work page 2018

  79. [79]

    Assessing the properties of supercri tical water in terms of structural dynamics and electronic polarization effects

    P. Schienbein and D. Marx. “Assessing the properties of supercri tical water in terms of structural dynamics and electronic polarization effects”. Phys. Chem. Chem. Phys. 22 (2020), pp. 10462–10479

    work page 2020

  80. [80]

    Raman-and infrared spe ctroscopic investigation of aqueous ZnSO 4 solutions from 8 ◦ C to 165 ◦ C: inner-and outer-sphere com- plexes

    W. W. Rudolph, M. H. Brooker, and P. Tremaine. “Raman-and infrared spe ctroscopic investigation of aqueous ZnSO 4 solutions from 8 ◦ C to 165 ◦ C: inner-and outer-sphere com- plexes”. Z. Phys. Chem. 209 (1999), pp. 181–207

    work page 1999

Showing first 80 references.