arXiv Science⌕ Search
Home/Archive/arXiv

arXiv · 2610.13902

Accuracy of Vibrational Properties of Machine-Learning Force Fields for Molecular Crystal Transformations

Natalia Goncharova·Oskar Schweighofer·Johannes Hoja
View original source ↗Read source document ↗Export metadata ↓

Abstract

Machine-learning force fields (MLFFs) enable the calculation of transformation pathways in molecular crystals at a fraction of the computational cost of density functional theory (DFT). However, a good agreement in energies and structures does not necessarily imply an accurate description of vibrational properties, which are essential for characterizing transition states and determining free energies at finite temperatures. Herein, we assess the accuracy of system-specific SO3krates MLFFs trained on PBE+MBD reference data for describing vibrational properties of molecular crystal transformations. As examples, we consider a polymorph transition of 1,4-diiodobenzene and a single-crystal-to-single-crystal Diels-Alder reaction. The transition state structures obtained from solid-state nudged-elastic-band (NEB) calculations with the MLFFs exhibit a single imaginary $Γ$-point frequency associated with the respective transformation. Compared with PBE+MBD, the mean absolute errors of $Γ$-point vibrational frequencies and relative harmonic free energies at 300 K are about 3 cm$^{-1}$ and 1 kJ/mol, respectively. Furthermore, we compare for the Diels-Alder reaction a collective transformation pathway with a sequential one, in which individual diene-dienophile pairs react successively, and demonstrate that PBE0+MBD single-point corrections can significantly affect the predicted reaction energetics. These results demonstrate that system-specific MLFFs can reproduce not only the energies and forces but also the local curvature of the potential-energy surface with high accuracy, enabling reliable calculations of harmonic vibrational free energies, providing a basis for extending the studies of molecular crystal transformations to larger supercells.

Explore related subjects

physics.chem-ph

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Natalia Goncharova, Oskar Schweighofer, Johannes Hoja. 2026-10-09. Accuracy of Vibrational Properties of Machine-Learning Force Fields for Molecular Crystal Transformations. https://arxiv.org/abs/2610.13902

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Assessing the Performance of CC2 for Excited-State Dynamics: A Benchmark Study with Pyrazine

During the ultrafast electronic relaxation of pyrazine, the bright state undergoes rapid decay accompanied by coupled electronic and nuclear motion. We investigate the vibrational modes associated with this process using trajectory surface hopping simulations. To enable accurate \emph{ab initio} excited-state dynamics, we apply CC2 analytical gradients and nonadiabatic coupling vectors within the Q-Chem package. Using both a reduced vibronic coupling model and full-dimensional simulations, we demonstrate that CC2 predicts that the dark $A_\text{1u}$ state actively participates in the initial relaxation and the subsequent dynamics. The full-dimensional simulations yield a $B_\text{2u}$ decay time of 25--26 fs, in good agreement with the experimental measurement of $22\pm3$ fs. Furthermore, we identify the $Q_\text{8a}$ and $Q_\text{9a}$ modes as dominant tuning coordinates whose motions are phase-locked to the population oscillations between the $A_{1u}$ and $B_{3u}$ states.

physics.chem-ph↗
arXiv2026

NucleoNet: A Nuclear-Aware Determinant Ansatz for Multi-Geometry Neural Variational Monte Carlo

Neural-network variational Monte Carlo provides an accurate route to solving the electronic Schrödinger equation, but many neural wavefunctions are optimized for a single nuclear geometry, requiring separate calculations to map out a potential energy surface. We introduce NucleoNet, a nuclear-aware determinant ansatz in which electrons and nuclei form a unified particle set propagated through a shared electron--nucleus backbone that jointly evolves one-particle and two-particle representations. Nuclei carry learned representations that evolve alongside the electronic representations and condition orbital construction, while antisymmetry is imposed only by electronic determinants. On equilibrium small-molecule benchmarks, NucleoNet achieves accuracy comparable to established neural-wavefunction baselines with substantially fewer parameters. For each molecular system, a single model optimized jointly across geometries describes the N$_2$ dissociation curve, Li displacement in LiCN, and the ethane and formamide minimum-energy paths for bond dissociation. Sparse-geometry training generalizes to unseen geometries for N$_2$ and LiCN without further optimization, and N$_2$ ablations show that nuclear streams substantially improve this interpolation. These results show that a compact wavefunction with jointly evolved nuclear representations can describe the geometries of a molecular system with a single set of parameters.

physics.chem-ph↗
arXiv2026

Quantitative modeling of acid-base equilibria

Simulating acid-base equilibria from first principles is a tremendous challenge, as it requires modeling bond-breaking events in a complex hydrogen-bonding environment. The quantum mechanical behavior of the protons is often important, complicating simulations even further. We show that by combining state-of-the-art machine-learning potentials, accelerated sampling of the dissociation process, and an efficient estimation of nuclear quantum corrections it is possible to reliably compute the dissociation constants of different species in water. We demonstrate the transferability of this workflow, applying it successfully to the autoionization of water and to the entire ladder of the aqueous dissociation of phosphoric acid. The definition of the dissociation constant, the choice of order parameters, sampling artifacts and finite-size effects, as well as quantum nuclear effects all play a non-trivial role in determining the outcome of the calculation, so that cutting corners is likely to introduce biases that can only be compensated by fortuitous cancellations of errors.

physics.chem-ph↗