Classical Trajectory Simulations

Introduction

A minimum-energy path describes where a molecule can travel on an energy landscape, but it does not say how a moving molecule actually travels. Classical trajectory simulations assign positions and velocities to nuclei and integrate their motion under forces from a potential-energy surface. Repeating this for many initial conditions yields reaction probabilities, scattering patterns, product branching and travel times. The method gives valuable dynamical insight while approximating quantum nuclear behavior.

Core explanation

For each nucleus i, classical motion follows Newton's equation mi d²ri/dt² = −∇iV(r), where V is a potential-energy surface for the chosen electronic state. Numerical integration advances positions and velocities in small time steps. If the force field is accurate and the step is sufficiently small, total energy should remain approximately conserved in an isolated simulation. With thermostats or environmental models, energy exchange must instead follow the specified model.

Reactive trajectories need a potential that remains meaningful when bonds form and break. A conventional fixed-bond force field may fail for this purpose. One strategy fits a high-dimensional function to many electronic-structure energies and forces, making force evaluations fast enough for thousands of trajectories. Its speed is valuable, but accuracy is only as good as the training coverage. A fit that looks excellent near reactant and product minima can still be wrong in a rarely sampled saddle or high-energy collision region.

Initial conditions are just as important as the surface. In scattering, sample collision energy, impact parameter, molecular orientation and reactant vibrational and rotational states according to the experiment. In a thermal ensemble, sample positions and momenta from an appropriate distribution. For transition-state-launched trajectories, sample transverse modes and velocity signs consistently with a flux distribution. One visually striking trajectory is an anecdote, not a product ratio or rate constant.

After propagation, define objective outcomes: whether reactants returned, which bonds formed, product identity, internal energy and scattering angle. For a sample of N equally weighted trajectories, an estimated reaction probability is Nreactive/N, with statistical uncertainty that shrinks only as the sample grows. Weighted sampling requires weighted estimators. Rare channels may need many trajectories, especially if their probability is small. A finite observation time can mistake a delayed reaction for no reaction.

Quasiclassical trajectories often initialise vibrational energies to mimic quantum levels or zero-point motion, then propagate classically. This is practical but cannot enforce all quantum rules. Classical energy flow can move nominal zero-point energy out of one mode, a problem called zero-point-energy leakage. Trajectories also omit wave interference, quantised product states and tunnelling unless augmented with special approximations. Their predictions should be compared with quantum calculations or experiment where those effects are significant.

The dynamical picture can contradict a static interpretation. After crossing one saddle, trajectories may recross, wander around a flat region, form a short-lived complex or divide between product basins. These outcomes can depend on momentum in modes perpendicular to the IRC. Trajectory simulations thus test whether simple transition-state assumptions are valid, not merely illustrate a proposed arrow-pushing mechanism.

Step-by-step reasoning

Choose a validated potential surface covering all likely regions of configuration space. Define an initial ensemble corresponding to the question: thermal reaction, beam collision or transition-state launch. Integrate equations of motion with convergence checks on time step and energy conservation. Apply predeclared criteria for reactants, intermediates and products. Repeat over enough trajectories to estimate probabilities and uncertainty. Compare energy, angular and isotope-dependent observables with experiment, and test sensitivity to the surface fit and sampling.

Visual explanation

Draw one contour map with several arrows starting at slightly different positions and velocities near a saddle. Some return to reactants and others split toward products. Beside it show a flowchart: sample initial conditions → calculate force → step positions and momenta → classify outcome → repeat. Mark the fitted surface's training points along wells, barrier and scattering regions to emphasise coverage.

Real-world analogy

A topographic trail map gives possible routes, but throwing many balls across it with different speeds and starting directions shows which valleys they actually enter. Repeating throws estimates a distribution, not one guaranteed path. Molecules differ because their nuclei can tunnel and interfere as quantum waves, and a fitted potential can be inaccurate outside its sampled region.

Real-world example

A reactive collision between an ion and a neutral molecule may depend strongly on approach orientation and impact parameter. Quasiclassical trajectory work can sample these factors and predict cross sections and product energy disposal for comparison with beam data. One published direct-dynamics study of formaldehyde cation with D2 found strong orientation dependence and examined how vibrational preparation changed reactivity, illustrating both the value and limitations of trajectory modelling.

Why?

Why integrate many trajectories? Different initial conditions yield different outcomes and an experimental sample contains a distribution. Why check energy conservation? Numerical drift can create or remove apparent reaction channels. Why fit a surface? Direct electronic force calculations at every small time step can be expensive. Why validate high-energy regions? Reactive trajectories can visit configurations absent from equilibrium training data.

Common misconception

A trajectory is not the IRC followed in time. An IRC is a steepest-descent geometric construction, while a trajectory carries momentum and can deviate, recross or branch. Also, assigning initial quantum zero-point energy does not make classical propagation fully quantum mechanical; energy leakage and missing interference remain possible.

Worked example

Question: Of 500 equally sampled collision trajectories at one energy, 75 reach product P1, 25 reach P2 and 400 scatter without reaction. Estimate overall reaction probability and conditional product fractions among reactive trajectories.

Reasoning: Reactive count is 75+25 = 100, so the total reaction probability is 100/500 = 0.20. Conditional on reaction, P1 is 75/100 = 0.75 and P2 is 25/100 = 0.25. These estimates apply to the selected initial-condition distribution; a thermal rate would require sampling or averaging over thermal collision energies and states.

Answer: Reaction probability is 0.20; among reactive events, P1:P2 fractions are 0.75:0.25.

Quick check

1. Why is one trajectory insufficient for a predicted branching ratio? Answer: Branching is a statistical property of an ensemble with varied starting positions, velocities and states.

Exam focus

Write Newton's force equation and explain numerical propagation on a PES. Specify how initial-condition sampling and product classification determine the meaning of reported fractions. Contrast a trajectory with an IRC and identify fitted-surface error and missing quantum effects as key limits.

Advanced insight

Rare-event fractions have binomial or weighted-sampling uncertainty, so hundreds of trajectories may still be inadequate for a one-percent branch. A fitted potential may require active learning: run trial trajectories, identify configurations where model uncertainty is high, calculate new electronic reference points and refit. For nonadiabatic dynamics, one surface and Newtonian motion are insufficient; electronic-state transitions need an additional model. Statistical convergence and model validation are separate requirements.

Summary

Classical trajectories propagate nuclei in time using forces from a potential-energy surface. Ensembles of carefully sampled starting states yield reaction probabilities, product branching and scattering observables. Fitted surfaces enable extensive sampling but must cover reactive geometries. Quantum tunnelling, interference and zero-point constraints are not fully represented, so comparison with other theory and experiment remains essential.

Practice questions

1. What force determines nuclear acceleration in an adiabatic classical trajectory? Answer: The negative gradient of the chosen potential-energy surface with respect to each nuclear coordinate.

2. Why can a fitted surface fail despite good agreement at stable minima? Answer: It may be inaccurate in barriers, collisions or product-branching regions absent from its training data.

3. What is zero-point-energy leakage in quasiclassical trajectories? Answer: Classical dynamics can transfer nominal quantum zero-point energy out of a mode below its allowed quantum ground-state amount.

4. How should a collision trajectory ensemble be selected for comparison with a beam experiment? Answer: Sample the beam's collision energies, impact parameters, orientations and internal states, then model detection conditions.

Sources: Journal of Chemical Theory and Computation, fitting reactive surfaces; Journal of Physical Chemistry A, direct-dynamics collision trajectories; Accounts of Chemical Research, ab initio direct dynamics.