Machine learning prediction of a chemical reaction over 8 decades of energy

Daniel Julian; Jesús Pérez-Ríos

Home/Archive/arXiv

arXiv · 2507.01793

Machine learning prediction of a chemical reaction over 8 decades of energy

Daniel Julian·Jesús Pérez-Ríos

View original source ↗Read source document ↗Export metadata ↓

Abstract

Recent progress in machine learning has sparked increased interest in utilizing this technology to predict the outcomes of chemical reactions. The ultimate aim of such endeavors is to develop a universal model that can predict products for any chemical reaction given reactants and physical conditions. In pursuit of ever more universal chemical predictors, machine learning models for atom-diatom and diatom-diatom reactions have been developed, yet no such models exist for termolecular reactions. Accordingly, we introduce neural networks trained to predict opacity functions of atom recombination reactions. Our models predict the recombination of Sr$^+$ + Cs + Cs $\rightarrow$ SrCs$^+$ + Cs and Sr$^+$ + Cs + Cs $\rightarrow$ Cs$_2$ + Sr$^+$ over multiple orders of magnitude of energy, yielding overall results with a relative error $\lesssim 10\%$. Even far beyond the range of energies seen during training, our models predict the atom recombination reaction rate accurately. As a result, the machine is capable of learning the physics behind the atom recombination reaction dynamics.

Explore related subjects

physics.chem-ph physics.atom-ph

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Daniel Julian, Jesús Pérez-Ríos. 2025-07-02. Machine learning prediction of a chemical reaction over 8 decades of energy. https://arxiv.org/abs/2507.01793

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

KEEP EXPLORING

Related papers

arXiv2026

A Task-Based Framework for Evaluating Raman Spectral Quality Measures

Raman spectral preprocessing and enhancement are often evaluated by comparing output spectra with a reference. Interpreting these comparisons requires evidence that spectral quality measures reflect downstream task performance. We present a controlled-perturbation framework for testing this relationship. Five perturbation types (baseline distortion, independent noise, correlated noise, a global wavenumber shift, and nonlinear axis warping) generate paired changes in a spectral measure (metric harm) and in downstream performance (task harm). An alignment gap (AG) quantifies how much the relationship between metric harm and task harm changes with perturbation type. Ordering concordance (OC) measures how often a metric correctly ranks two conditions by their task harm. The framework evaluates thirteen outputs (MSE, RMSE, MAE, NMSE, spectral angle, Pearson correlation, Wasserstein distance, a structure-to-noise ratio, peak precision, recall, F1, artifact ratio, and missing ratio). Three public datasets provide bacterial classification, sugar-mixture quantification, and mineral identification tasks. PCA with logistic regression, partial least squares regression, and cosine library matching supply the task outcomes. Classifiers and calibrations are fitted either to unperturbed training spectra or to each perturbed training condition, then evaluated on the same perturbed test spectra. Mineral queries are compared with an unchanged or correspondingly perturbed library. The resulting comparisons identify task-specific strengths and limitations, including cases where better ordering does not accompany a smaller AG. Removing axis perturbations and comparing spectra on a common physical grid test how these findings depend on the evaluation design. The framework provides a reproducible procedure for assessing existing measures and testing new candidates against downstream task performance.

physics.chem-ph↗

arXiv2026

A System-Independent Metadynamics Strategy for Reactive Training Data: Application to Gas-Phase Organic Reactions

General-purpose machine-learning interatomic potentials (MLIPs) for organic reactions need to be accurate on both the minimum energy path (MEP) for static evaluation of basic properties and the broader configurational space for simulating reaction dynamics. Existing general datasets for gas-phase organic reactions rely on quasi-static relaxation that confines configurations to the MEP vicinity, so models trained on them could fail on direct molecular-dynamics trajectories; the gap is methodological, not a question of dataset size. We introduce a spatiotemporally resolved, system-independent collective variable (CV): Cartesian RMSD within randomly partitioned local domains against an expanding list of time-averaged reference geometries. The CV drives metadynamics as the main exploration engine, supplemented by structural relaxation towards transition state (TS) to augment the coverage around TS. Within a concurrent-learning workflow, this produces OpenRxn26, a dataset of 1.8~M DFT-labeled configurations covering neutral singlet unimolecular reactions in the H/C/N/O chemical space ($N_\mathrm{heavy} \leq 30$), containing reactive atomic environments underrepresented in community datasets. Trained on OpenRxn26, a DPA3 model (denoted DPA3_rxn) achieves transferable accuracy on barrier heights and reaction energies. On off-MEP reactive trajectories, DPA3_rxn is the only model in the benchmark suite to reach 1.0 kcal/mol energy accuracy compared with the labeling method, where the domain MLIP leading on static benchmarks degrades several-fold (e.g. MACE_OMol25), showing the insufficiency of quasi-static sampling and MEP-anchored benchmarks for guaranteeing dynamics reliability of MLIPs. OpenRxn26 thus provides MD-ready reactive training data for gas-phase neutral singlet organic reactions, verifying the generality and efficiency of the sampling strategy.

physics.chem-ph↗

arXiv2026

Full-frequency GW from Cayley-transformed self-energy moments

The dynamical GW self-energy approximation is a key computational tool to provide the fundamental spectrum of electronic systems. We reformulate this approximation, representing the particle and hole parts of the GW self-energy through a highly compact set of Cayley-transformed moment constraints. The Cayley transformation maps real frequencies to the unit circle, keeping the moments bounded as their order increases, ensuring numerical stability and allowing resolution to be focused on an energy range of interest. We calculate these Cayley-transformed moments via an efficient O[N$^4$] scaling contour integration, and from them, construct a Hermitian upfolded Hamiltonian with a linearly scaling dimensionality with system size. A single-shot diagonalization of this effective Hamiltonian gives an explicit full-frequency G0W0 Green's function with manifestly real poles and non-negative spectral weights. This enables quasiparticle energies, satellite features, and their spectral weights to be obtained across the full G0W0 spectrum. Comparisons with exact G0W0 calculations and convergence across the GW100 test set and the larger Chlorophyll A molecule demonstrate substantially faster and more reliable convergence with moment order than an earlier monomial-moment approach. These Cayley moment representations therefore provide a stable, compact, and systematically improvable route to the complete spectral information of zero-temperature GW, without explicit frequency grids, plasmon-pole models and other common approximations, or analytic continuation.

physics.chem-ph↗