Smite: A quasiclassical trajectory (QCT) program for bimolecular collisions and unimolecular dynamics on ab-initio, and machine-learned potential energy surfaces

arXiv:2609.40135 · physics.chem-ph, quant-ph · Submitted 2026-09-30 · Read on arXiv

Listen

Transcript

Introduction to the show: ident: Quantum Radio. Generated commentary on the latest quantum physics and condensed matter papers.

Kai: I'm Kai, and with me are Mira and Lev, guest researcher.

Mira: Today's paper: "Smite: A quasiclassical trajectory (QCT) program for bimolecular collisions and unimolecular dynamics on ab-initio, and machine-learned potential energy surfaces".

Kai: Smite presents a Python toolkit designed for quasiclassical trajectory (QCT) simulations encompassing reactive and inelastic bimolecular collisions, unimolecular dynamics, and molecular collisions with finite surface models.

Mira: First, who's behind it and why it matters.

Paper summary: Kai: So, to recap what we’ve discussed about Smite, the paper introduces this Python toolkit for quasiclassical trajectory simulations of reactive and inelastic bimolecular collisions, unimolecular dynamics, and molecular collisions with finite surface models. The main thesis is that it provides explicit control over reactant preparation—including harmonic normal modes, thermal populations, or ground-state Wigner sampling—and allows the nuclear propagators to use on-the-fly electronic structure calculations or user-supplied machine-learned potential energy surfaces for the trajectories themselves.

Mira: Exactly. They claim this is significant because it offers explicit control over reactant preparation, meaning users can choose exactly how they initialize the simulation, whether it's by fixing quantum numbers or sampling thermally based on a Boltzmann distribution. Furthermore, the system lets the nuclear propagators utilize either on-the-fly electronic structure calculations or user-supplied machine-learned potential energy surfaces for the trajectories themselves.

Lev: From my point of view, what’s important is that this explicit control over preparation—the input conditions—allows for a much more structured approach than purely classical molecular dynamics, which is what Smite sits between and full quantum scattering.

Kai: It’s about bridging that gap pragmatically; it lets researchers draw initial conditions from ensembles of classical states corresponding to preselected quantum-mechanical vibrational and rotational states before propagating those trajectories with classical mechanics.

Mira: And they detail the specific ways this preparation control is implemented, mentioning fixed angular momentum magnitude or thermal distributions appropriate for diatomic and polyatomic rotors, and even a specialized technique called Rotating-Morse sampling which provides a coupled anharmonic treatment of vibration and rotation.

Lev: That Rotating-Morse sampling sounds like it tackles the coupling between different degrees of freedom in a way that is necessary for accurately modeling things like vibrational and rotational motion simultaneously.

Kai: Beyond just setting up the initial state, Smite integrates modules that share common energy and derivative interfaces for geometry optimization, transition-state searches, and intrinsic-reaction-coordinate following. This makes it a cohesive system for studying a whole reaction process.

Mira: It's also important that the nuclear propagation engine supports various integrators, such as velocity Verlet and leapfrog schemes, giving flexibility in how the trajectory is advanced through time. This modularity means you can tailor the simulation to your specific needs.

Lev: If we're thinking about real hardware implementation, having these various integrators available means we can test which one offers better stability and faster convergence for the specific chemical problem we are tackling.

Kai: And crucially, as the paper shows they can operate with on-the-fly electronic structure calculations or user-supplied machine-learned potential energy surfaces, which is a huge factor for accessibility. This opens up possibilities beyond systems where you have to rely solely on pre-calculated, fixed potentials.

Mira: The core claim is that this combination of explicit input control and flexible potential energy access allows the toolkit to simulate a wide range of chemical processes, from simple molecular collisions to complex unimolecular dynamics. It’s designed to be a versatile middle ground.

Lev: It sounds like the main goal is creating a robust simulation framework that can handle the complexity of chemical environments without being completely constrained by any single theoretical assumption about the potential energy surface.

Conclusion: Kai: To wrap up, Smite is this Python toolkit designed for quasiclassical trajectory simulations of reactive and inelastic bimolecular collisions, unimolecular dynamics, and molecular collisions with finite surface models. The authors present it as a versatile tool that centers on explicit control over reactant preparation—things like harmonic normal modes or ground-state Wigner sampling—and allows the nuclear propagators to use either on-the-fly electronic structure calculations or user-supplied machine-learned potential energy surfaces for the trajectories themselves.

Mira: The implication here is that this framework gives researchers a much finer degree of control over the simulation's starting point and its underlying physics, which is valuable when studying processes where those initial conditions dictate the outcome. It’s not just a program; it’s a system that lets you choose your theoretical assumptions for initialization and propagation.

Lev: From an error correction standpoint, the ability to define those initial states so precisely means we can design better strategies for managing the initial state preparation errors before they even propagate through the dynamics. It's about controlling the noise at its source.

Kai: And that control extends to how you handle potentials, as you can use ab-initio data or machine-learned surfaces directly during the trajectory calculation. That flexibility means you aren't stuck with a single, potentially inaccurate analytical model for a complex chemical process.

Mira: Ultimately, the title of this paper reflects its ambition to be an explicit simulation method that sits between full quantum scattering and purely classical molecular dynamics. It’s aiming to deliver the full mechanistic movie of a process without needing a predefined reaction coordinate or assuming ergodicity.

Lev: The real impact on the world, if this works as intended, is providing a pathway to study complex chemical environments that were previously too computationally expensive to explore thoroughly using traditional methods. It opens up new avenues for simulating everything from combustion dynamics to other challenging molecular interactions.

Kai: So, in simple terms, Smite is a sophisticated simulation environment that lets you tailor the starting conditions and the energy function access precisely to investigate complex chemical reactions in collision or unimolecular settings. It’s about giving us the tools to look at the full mechanism of a process directly.

Mira: Exactly. It’s about moving away from relying solely on simplified models and instead using explicit control over the preparation and the potential energy source to get a more comprehensive picture of how these chemical events unfold.

Lev: And for the hardware community, it suggests that if we can build systems capable of handling those on-the-fly calculations, this framework provides a concrete way to test the performance and error propagation of such complex dynamics.

Kai: So, Smite is a significant development because it gives us a powerful way to simulate these chemical phenomena by putting the control back into the hands of the researcher during preparation and potential energy selection.

P´eter Szab´o, Jenne van Veerdeghem, J´erˆome Loreau, Jean-Fran¸cois M¨uller, Jeremy N. Harvey

Belgian Institute for Space Aeronomy (BIRA-IASB) · Department of Chemistry, KU Leuven

physics.chem-ph, quant-ph

Submitted: 2026-09-30

Updated: 2026-09-30

Code: https://github.com/peter88szabo/Smite

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 84/100

The gist: Smite presents a Python toolkit designed for quasiclassical trajectory (QCT) simulations encompassing reactive and inelastic bimolecular collisions, unimolecular dynamics, and molecular collisions

Key concepts

Quasiclassical Trajectory (QCT): QCT is a simulation method that balances full quantum scattering accuracy with classical molecular dynamics. It starts by drawing initial conditions from quantum states and then propagates these trajectories using classical mechanics, providing a practical middle ground for complex simulations.
Reactant Preparation Control: Smite gives users explicit control over how reactants are set up before simulation begins. This includes fixing vibrational modes, sampling thermal populations using Boltzmann distributions, or preparing initial conditions via Ground-state Wigner sampling to ensure accurate starting points.
On-the-fly PES Integration: The software is designed to work with various potential energy surfaces (PES). Users can feed in electronic structure calculations or machine-learned potentials during the simulation. This flexibility allows the toolkit to handle complex chemical systems that might not have pre-existing analytical potentials.

Terminology

Ab Initio — 10× in this paper · explained in 1 episode
Accuracy — 8× in this paper · explained in 1 episode
Core — 4× in this paper · explained in 2 episodes
Anisotropy — 2× in this paper · explained in 3 episodes
Consistency — 2× in this paper · explained in 3 episodes
Abundance — 1× in this paper · explained in 1 episode
Adam — 1× in this paper · explained in 2 episodes
Adaptive Sampling — 1× in this paper · explained in 1 episode
Alignment — 1× in this paper · explained in 1 episode
Archetype — 1× in this paper · explained in 1 episode
Benchmark — 1× in this paper · explained in 2 episodes
Degeneracy — 1× in this paper · explained in 3 episodes
Geometric Distribution — 1× in this paper · explained in 2 episodes
Modularity — 1× in this paper · explained in 9 episodes
Provenance — 1× in this paper · explained in 3 episodes
Quantization — 1× in this paper · explained in 12 episodes
Trace — 1× in this paper · explained in 2 episodes
Wigner function — 1× in this paper · explained in 2 episodes

Summary

Smite presents a Python toolkit designed for quasiclassical trajectory (QCT) simulations encompassing reactive and inelastic bimolecular collisions, unimolecular dynamics, and molecular collisions with finite surface models. This program is significant because it provides an explicit control over reactant preparation—including harmonic normal modes, thermal populations, and ground-state Wigner sampling—and allows the nuclear propagators to utilize on-the-fly electronic structure calculations or user-supplied machine-learned potential energy surfaces.

The gist

Smite is a Python toolkit for quasiclassical trajectory (QCT) simulations of reactive and inelastic bimolecular collisions, unimolecular dynamics, and molecular collisions with finite surface models.

How it works

The core methodology of Smite is the quasiclassical trajectory method (QCT), which serves as a pragmatic middle ground between full quantum scattering—exact but currently limited to few-atom systems—and purely classical molecular dynamics. Initial conditions are drawn from ensembles of classical states corresponding to preselected quantum-mechanical vibrational and rotational states of the reactants, after which trajectories are propagated with classical mechanics.

The software explicitly separates reactant preparation from trajectory propagation. The toolkit offers explicit control over various initial condition distributions:

Vibrational sampling can be fixed by a Fixed quantum number (Q) or sampled based on Thermal population (T) using Boltzmann-distributed integer populations, or prepared via Ground-state Wigner (W) sampling.
Rotational preparation allows for fixed angular momentum magnitude or thermal distributions appropriate to diatomic and polyatomic rotors, including the use of Rotating-Morse sampling for coupled anharmonic treatment of vibration and rotation.
For collisions, initial conditions include relative translation sampled from fixed or thermal collision energies, and impact parameter choices such as fixed, sampled uniformly in b, or sampled uniformly over the collision disk.

Core Functionality and Interfaces

Smite integrates various modules that share common energy and derivative interfaces for geometry optimization, transition-state searches, crossing-point optimization, intrinsic-reaction-coordinate following, and thermochemistry. The nuclear propagation engine supports various integrators including velocity Verlet and leapfrog schemes. Crucially, the system can operate with on-the-fly electronic structure calculations or user-supplied analytical and machine-learned potential energy surfaces.

Analysis Tools

The program provides extensive analysis tools to connect recorded trajectories to dynamical and spectroscopic observables. These include:

(41) Product identity, energy partitioning, and state assignment:

(42) Cross sections and thermal rate estimators:

(50) Energy transfer and stereodynamical correlations:

The analysis routines allow for the calculation of product energy partitioning, stereodynamical correlations, time-resolved vibrational signatures, and independent-atom x-ray scattering. For instance, cross sections are estimated using a sampling density function, where the estimator is defined as:

σbr = (1/N) Σ i 2πbi g(bi) hi r.

Advanced Modules

Smite includes specialized modules to extend its capabilities:

(E) Fewest-switches surface hopping:

This module propagates electronic amplitudes while the nuclei move on one active surface using the FSSH framework, approximating real coupling vectors from the adiabatic gap and its derivatives via a curvature construction associated with Baeck–An-type approaches.

(F) Photoionization preparation and ionic dynamics:

This module maps a neutral phase-space ensemble onto an ionic PES, allowing for initial conditions in unimolecular ionic dynamics. It implements four preparation levels, including Level 3 which accounts for photon/photoelectron recoil, by solving:

1/2 saint + b squared = Ki int (40)

(G) Stationary-point, reaction-path, and thermochemical tools:

These tools include minimum-energy optimization, transition-state searches using a partitioned rational function optimization step, and IRC following. The module also provides Normal-mode analysis which explicitly inputs Temperature, pressure, rotational symmetry number, and chirality factor.

Validation

Validation involves comparing sampled distributions against analytical targets. For example, fixed-energy harmonic draws must reproduce the requested mode energy before rotational adjustment. Production tests monitor timestep convergence, energy and momentum conservation, and sensitivity to reaction-channel thresholds. The code's modular structure allows for tracing these choices from preparation through propagation to observable analysis.

Code Availability

The source code is available at https://github.com/peter88szabo/Smite under the GNU General Public License. The repository includes examples and tests, and the Python API remains the complete interface for advanced preparation and analysis. The software supports energy and derivative interfaces from various electronic structure backends such as xTB, ORCA, PySCF, Psi4, and SCINE Sparrow.

Improvements for AI systems

Here are specific improvements to AI systems based on the Smite toolkit, detailing what an improved system could achieve:

The Smite toolkit enables a shift from purely statistical or empirical modeling of chemical dynamics to high-fidelity, mechanistic, and predictive simulations grounded in quantum mechanics. An AI system leveraging this framework can be significantly enhanced in the following ways:

A simulation platform capable of performing high-dimensional, ab-initio guided molecular dynamics (MD) simulations for reactive systems across a vast range of chemical environments (combustion, atmospheric chemistry, surface interactions).
The ability to generate mechanistic movies and state-resolved kinetic data from trajectory ensembles that explicitly capture nonstatistical behaviors and complex reaction pathways that statistical theories fail to describe.

Specific capabilities enabled by these improvements:

A system can simulate the formation of complex chemical species (e.g., HO2 addition to radicals) by directly observing site-specific branching ratios, even in barrierless or conformationally flexible capture processes where simple statistical models yield incorrect results.
The AI system could predict the roundabout mechanism of reactions (like Cl− + CH3I SN2) by tracing trajectories that avoid deep potential energy minima, revealing non-IRC (Intramolecular Reaction Coordinate) pathways that are otherwise inaccessible to standard methods.
The system can quantify how internal energy redistribution and conformational motion compete with reaction rates by analyzing the time evolution of vibrational signatures in collision complexes, providing insights into the memory retained during a collision.
It can generate highly specific, state-resolved cross-sections and rate coefficients for complex collisions (e.g., CO–CO or H2O–H2), which are crucial inputs for master equation models of thermal kinetics, moving beyond simple thermal rate coefficients to provide detailed energy transfer kernels and moments.
The system can perform rigorous validation by comparing classical trajectory results against established quantum scattering methods (like close-coupling calculations) and analytical PES calculations, allowing the AI to assess the accuracy of underlying machine-learned potentials or fitted surfaces in real-time.
It can generate structural form factors (using independent-atom x-ray scattering proxies) directly from trajectory coordinates, providing atomistic snapshots of molecular structure and dynamics that are sensitive to bond changes during a reaction or collision.
The system can model photoionization events, generating initial conditions for subsequent unimolecular ionic dynamics by explicitly accounting for energy and momentum constraints (recoil), allowing the simulation of processes like radical-O2 capture initiation.

Abstract

We present Smite, a Python toolkit for quasiclassical trajectory (QCT) simulations of reactive and inelastic bimolecular collisions, unimolecular dynamics, and molecular collisions with finite surface models. A central feature is the explicit control of reactant preparation: harmonic normal modes admit fixed quantum numbers, prescribed energies, thermal quantum populations, or ground-state Wigner sampling; rotation is prepared at fixed angular-momentum magnitude or from thermal distributions appropriate to diatomic and polyatomic rotors. Rotating-Morse sampling provides a coupled anharmonic treatment of diatomic vibration and rotation, while saved molecular-dynamics phase points provide an alternative source of initial conditions. The same nuclear propagators operate with on-the-fly electronic-structure calculations or user-supplied analytical and machine-learned potential energy surfaces. Additional modules provide constrained rigid-fragment dynamics, thermostat-based preparation, photoionization initial conditions with energy and recoil constraints, and fewest-switches surface hopping (FSSH) on multiple PESs with approximate curvature-derived couplings. Geometry optimization, transition-state searches, crossing-point optimization, intrinsic-reaction-coordinate following, and thermochemistry use the same energy and derivative interfaces. Analysis tools connect trajectories to product energy partitioning, stereodynamical correlations, time-resolved vibrational signatures, and independent-atom x-ray scattering.