mod committor_pop¶
- module committor_pop¶
Population resampled by estimated probability of improvement. A population resampled by estimated probability of improvement.
Measured negative. This module is kept as the reproducibility record of a mechanism that was built, measured against paired controls, and refuted; the numbers and the mechanism of failure are in
docs/derivations/. It is not part of the recommended configuration and ships as evidence, not as advice.The committor of rare-event theory is the probability that a trajectory from a state reaches the product before returning to the reactant. Without naming any product, the search-facing analogue is the probability that a walk segment from a state produces a new lower minimum before the segment ends: defined by energy and history alone, no order parameter, no target.
Each walker carries a Beta posterior over that probability, updated by one Bernoulli observation per segment. Resampling is Thompson: the walker with the worst posterior draw is replaced by a clone of the one with the best. The distinction from diffusion Monte Carlo weighting is the point. Boltzmann weights
exp(-beta E)clone the deepest walkers, and the measured failure mode of this landscape is precisely a walker pinned at the deepest wrong minimum; an improvement-committor posterior culls a deep-but-dead walker that depth weighting would multiply.Segments are ordinary runs of the cluster driver with whatever mechanism stack the caller configured, so everything measured to work (the composed excursion, the depth bandit) rides along unchanged.
Functions
- fn committor_population(cfg: &Config, walkers: usize, segment_budget: usize, ledger: &mut Ledger, relax: Relax<'_>, seed: u64) -> CommittorOutcome¶
Runs
walkersshort chains undercfg, resampling by improvement posterior after each round, until the ledger refuses.
Structs and Unions