Committor Analysis

Introduction

A coordinate can produce a smooth free-energy barrier and still be poor at predicting reaction. Two configurations with the same bond distance may have different solvent orientations or protein conformations and thus different fates. Committor analysis tests this directly by asking: from this configuration, what fraction of appropriately sampled trajectories reaches products before reactants? It links a proposed static bottleneck to actual dynamics under a specified model.

Core explanation

Define reactant basin A and product basin B with objective criteria. For a starting microscopic configuration x, the product committor pB(x) is the probability that dynamics reaches B before A. It ranges from zero for strongly reactant-committed configurations to one for strongly product-committed ones. A value near one half marks configurations equally likely to reach either basin. It is a probability under a stated dynamics and initial momentum distribution, not an inherent label printed on a molecular geometry.

To estimate pB for one configuration, launch many short shooting trajectories , often resampling velocities from the appropriate equilibrium distribution while holding positions fixed. Count the fraction reaching B before A. If 20 of 40 shots reach B, an estimate is 0.5, with substantial sampling uncertainty. Some systems require including momenta in the initial state because commitment depends strongly on velocity; a position-only committor averages over velocities by definition. Basin definitions and trajectory stopping rules must be specified.

Suppose a proposed reaction coordinate ξ has a PMF maximum at ξ . Collect independent configurations with ξ near ξ and estimate each committor. If most have pB near 0.5, the coordinate's barrier region plausibly aligns with a dynamical transition ensemble. If pB values span near zero to near one, configurations sharing ξ have very different fates; another coordinate such as solvent polarisation, torsion or donor–acceptor distance is likely missing. A broad histogram cannot be repaired by merely plotting ξ with finer resolution if the missing variable is orthogonal.

The transition-state ensemble in this dynamical sense consists of configurations near equal commitment, rather than only one minimum-energy saddle geometry. On a high-dimensional free-energy surface it can contain diverse structures. It is related to a good dividing surface with low recrossing, but pB ≈ 0.5 for a configuration is not sufficient to determine a rate constant. One also needs how often the system reaches that region and the flux across it.

Committor analysis is costly. Many configurations, each with many short trajectories, may be required, and rare events can make basin arrival slow. Statistical uncertainty must be reported, and configurations should represent the ensemble near the proposed bottleneck rather than a hand-picked favorable set. A candidate coordinate can be improved iteratively by correlating pB with additional structural variables or by fitting a model to committor outcomes and validating on independent data.

For a non-equilibrium photochemical or nonadiabatic process, the notion of commitment still applies but must specify electronic state, momentum and dynamics method carefully. A ground-state equilibrium position-only committor cannot be transplanted unchanged to a prepared excited-state wavepacket. The strength of the method lies in its explicit definition of reactant and product fates for the model actually studied.

Committor tests complement, rather than replace, path and thermodynamic analyses. IRC confirms which minima a saddle connects on one surface. A PMF estimates equilibrium probabilities under a coordinate projection. Committor shooting asks whether candidate configurations are truly uncertain between A and B in dynamical evolution. Agreement across all three makes a reaction-coordinate proposal much more convincing.

Step-by-step reasoning

Specify nonoverlapping A and B basins and a dynamics model. Sample diverse configurations near the candidate barrier or dividing surface. For each, launch repeated trajectories with velocities drawn from the intended distribution, and record which basin is reached first. Estimate pB and confidence intervals per configuration. Plot the distribution of pB at fixed proposed coordinate value and correlate outliers with omitted variables. Revise the coordinate and repeat on independent configurations if the distribution is broad.

Visual explanation

Draw a line at ξ = ξ crossing a two-dimensional landscape. Place several dots on that line, colored blue for pB near zero, purple for pB near one half and red for pB near one. A mixture of colors shows the line is not an effective dividing surface. Beside it draw a revised curved boundary passing through mostly purple dots and a histogram of their committor values concentrated near 0.5.

Real-world analogy

Two travellers can stand the same distance from a destination yet have different chances of arriving first if one faces a bridge and the other faces a blocked road. Distance alone is a weak progress measure. Launching many trial journeys from each position resembles committor shooting. Molecules add thermal momentum, fluctuating surroundings and precisely defined basin arrival rules.

Real-world example

For a solvated hydride transfer, a bond-difference coordinate may be near zero for many configurations. If committor shooting shows that some nearly always return to reactant because donor–acceptor distance is too long, adding that distance can improve the coordinate. Studies of enzyme transition states have compared one- and multidimensional coordinates with transition-path and committor analyses to judge whether extra structural variables materially improve the bottleneck description.

Why?

Why is pB = 0.5 useful? It marks equal dynamical likelihood of reaching either basin under the stated model. Why shoot multiple trajectories? One deterministic outcome cannot estimate a probability over thermal momenta or stochastic environment. Why inspect a distribution at fixed ξ? It reveals hidden variables that ξ does not distinguish. Why define basins carefully? Changing what counts as A or B changes the commitment question.

Common misconception

A configuration at the maximum of a projected PMF does not automatically have committor 0.5. Orthogonal coordinates can change its fate. Likewise, one trajectory reaching product does not assign pB = 1; probability requires repeated appropriate sampling. The committor is model and basin dependent, not a universal molecular property.

Worked example

Question: Five configurations all have the same proposed ξ value at a PMF maximum. Forty shooting trajectories from each give product counts 2, 9, 20, 34 and 39. Is ξ alone a good bottleneck coordinate?

Reasoning: Estimated committors are 0.05, 0.225, 0.50, 0.85 and 0.975. They span almost the full range despite identical ξ, far beyond what one would expect from a narrow collection of transition-state-like configurations. Some other structural or environmental variable must distinguish reactant- and product-committed states. More shots improve precision but will not make the broad systematic spread disappear.

Answer: No. The broad committor distribution shows that ξ omits important dynamical information.

Quick check

1. What does a committor near 0.5 mean under specified basins and dynamics? Answer: From that initial condition, reactant and product basins are reached first with approximately equal probability.

Exam focus

Define pB precisely as first-arrival probability. Explain shooting trajectories and velocity resampling, then use the committor distribution at a candidate barrier to assess coordinate quality. Distinguish a committor test from an IRC or a PMF calculation and state that it does not alone yield a rate.

Advanced insight

For overdamped stochastic dynamics the position committor can be an ideal two-state reaction coordinate in a formal sense, but estimating it in high dimensions is expensive. In inertial dynamics, momentum can materially affect commitment, especially immediately after a saddle. Machine-learning models can approximate committors from structural features, but they need balanced training paths and independent validation. Multiple product basins require a vector of commitments rather than one binary A-versus-B probability.

Summary

Committor analysis measures the probability that a configuration reaches products before reactants. Configurations at a good bottleneck should have commitment near one half under the chosen dynamics; a broad spread at fixed proposed coordinate signals omitted variables. Repeated shooting trajectories, careful basin definitions and uncertainty estimates are essential. The method tests coordinate quality and complements free-energy and stationary-path calculations.

Practice questions

1. What is estimated by launching many trajectories from one starting configuration? Answer: Its probability of reaching product before reactant under the stated initial-velocity and dynamics model.

2. Why is a broad pB distribution at one ξ value troubling? Answer: Configurations that the coordinate treats as equivalent have different dynamical fates.

3. Does pB ≈ 0.5 for one geometry determine the reaction rate? Answer: No. Rate also depends on how often that region is visited and the reactive flux through it.

4. How can an omitted solvent variable be identified? Answer: Correlate committor values with solvent configurations at fixed ξ and test an expanded coordinate.

Sources: Journal of Chemical Theory and Computation, committor-guided pathway analysis; Journal of Chemical Theory and Computation, enzyme transition-state coordinate tests; Chemical Reviews, committor and enhanced sampling.