MoREST & MoREAT: from atomistic dynamics to reaction models

From atomistic simulation to reaction understanding

In many reaction and materials problems, the relevant question is what a system does under specific temperature, pressure, composition and collision conditions: which states it visits, which pathways compete, what products form, and how those processes appear in experiment. For unknown or complex reactions, this information is difficult to obtain from static structures or from a reaction pathway specified in advance.

MoREST and MoREAT were developed as a connected simulation-and-analysis framework. MoREST generates trajectories under defined physical conditions. MoREAT turns those trajectories into structural states, reaction networks and pathway models, and can derive kinetic and spectroscopic observables.

Simulating behaviour under different physical conditions

Atomistic systems undergo thermal motion, collisions, structural rearrangements and transitions among many configurations. Adequate sampling is needed to identify which states matter, which processes are accessible and how the result changes with temperature, pressure or collision conditions.

Different questions require different simulation strategies. Finite-temperature reaction studies need broad exploration of state space; short-time collision calculations focus on transient dynamics; rovibrational simulations need to preserve the motion that carries spectral information. MoREST combines phase-space sampling, trajectory scattering, molecular rovibration, structure searching, enhanced sampling, external potentials, thermostats and barostats in a modular workflow.

Multicomponent systems introduce an additional environmental constraint. A gas phase, a solid surface and other components may need to respond to different conditions, so controlling only the total pressure is not always sufficient. A partial-pressure barostat developed within MoREST allows selected components to be coupled to independent pressure controls while other parts of the system follow different dynamical or thermodynamic settings. This provides a route to gas adsorption, surface processes and other multicomponent simulations under specified partial pressures.

Energy and force evaluation is separated from the dynamical workflow. The same simulation layer can therefore use ab initio electronic-structure calculations or machine-learning potentials. Machine-learning potentials can also be coupled to external electronic-structure calculations for active learning or on-the-fly training, allowing the underlying force model to change without rebuilding the higher-level simulation workflow.

Discovering states and pathways in unknown reactions

Complex reactions often begin without a complete reaction network. Relevant intermediates, metastable structures and competing channels may only appear during finite-temperature dynamics, and an actual trajectory does not have to remain on a single minimum-energy pathway.

MoREST–MoREAT can therefore sample the dynamics first and construct the state representation afterwards, rather than requiring the complete network to be specified before the calculation begins. Enhanced-sampling strategies that do not depend on a predefined reaction coordinate can be used when the pathway itself is part of the unknown problem.

MoREAT then represents, clusters and discretizes the configurations visited in the trajectories and analyses transitions between the resulting states. Intermediate structures and competing pathways can be identified directly from the sampled dynamics, including metastable and short-lived configurations that are actually visited at finite temperature.

Building reaction models from trajectories

Once states and transitions have been identified, the appropriate reaction model depends on the physical regime. For systems close to thermodynamic equilibrium, trajectories can be discretized into states and described with a Markov state model. Combined with transition path theory, this provides state populations, free energies, reaction fluxes, reaction rates and the relative contributions of competing pathways.

Short-time collisions and strongly non-equilibrium processes require a different description because the order of events carries mechanistic information. A path-dependent reaction-tree model was developed to retain the sequence of microscopic reaction events and to estimate the probabilities of bond-forming and bond-breaking processes.

The two descriptions address different questions: equilibrium-like systems are treated through statistical state-to-state transitions, while path-resolved models retain the sequence of events needed to analyse microscopic reaction mechanisms.

Connecting atomistic motion to experimental observables

Experiments usually measure rates, lifetimes, spectra and their dependence on environmental conditions rather than the atomic coordinates at every simulation step. Trajectories therefore need to be converted into quantities that can be compared directly with measurements.

Dynamics can be used to estimate complex lifetimes across temperature and to include thermal motion and environmental effects directly in the calculation. Molecular vibration and rotation can also be analysed from time-dependent trajectories. Correlation functions of dipole moments, velocities and rotational quantities provide routes to infrared and rovibrational spectra.

Because these spectra are obtained from finite-temperature molecular motion, anharmonic effects enter naturally rather than requiring the dynamics to be reduced to a purely harmonic treatment. The same trajectory data can therefore support reaction-network analysis, mechanistic interpretation, kinetics and experimentally comparable observables.