Source-linked AI summary
Simulating rare events in dynamical processes
Cristian Giardina, Jorge Kurchan, Vivien Lecomte, Julien Tailleur
TL;DR
Rare trajectories are important in chemical transformations, planetary systems, turbulence, waves, transport, and glassy dynamics, but direct simulation can be infeasible. The review presents population dynamics that replicates or kills copies to favor atypical histories, and illustrates the approach across stochastic and Hamiltonian settings. Its examples show that the method can recover rare-event statistics and probe the configurations or phases responsible for anomalous behavior.
Problem
Rare trajectories govern important dynamical phenomena, but repeatedly simulating the true dynamics to observe them can become infeasible.
Method
The review evolves a population of system copies under the original dynamics while replicating or killing clones according to a trajectory observable.
Results
The method recovers large-deviation quantities and observables among atypical histories, while examples probe anomalous currents, dynamical phases, and planetary-system fluctuations.
Takeaways & Limitations
Population-based cloning provides an accelerated route to studying rare dynamical events and the configurations responsible for them.
Takeaways & Limitations
For hydrodynamic fluctuations, the standard fluctuation theorem involves temperatures that are irrelevant to the macroscopic fluctuations, motivating an effective-temperature interpretation.
Abstract
from arXiv · showhide
Atypical, rare trajectories of dynamical systems are important: they are often the paths for chemical reactions, the haven of (relative) stability of planetary systems, the rogue waves that are detected in oil platforms, the structures that are responsible for intermittency in a turbulent liquid, the active regions that allow a supercooled liquid to flow... Simulating them in an efficient, accelerated way, is in fact quite simple. In this paper we review a computational technique to study such rare events in both stochastic and Hamiltonian systems. The method is based on the evolution of a family of copies of the system which are replicated or killed in such a way as to favor the realization of the atypical trajectories. We illustrate this with various examples.
1 Introduction
Rare trajectories matter across chemical, planetary, turbulent, wave, transport, and glassy systems, yet direct simulation becomes infeasible because the relevant events occur infrequently. The review focuses on population-based cloning, which biases configuration-space dynamics toward atypical trajectories.
- Motivating examples: Rare trajectories can govern chemical transformations, planetary-system behavior, turbulent intermittency, rogue waves, transport, and dynamic heterogeneity.Examples include activation events, exceptional ballistic paths, unusually large waves, and atypically mobile or immobile regions.
- Motivating examples: In molecular dynamics, long metastable vibrations are interrupted by infrequent activation events that produce major rearrangements and chemical transformations.The review emphasizes accelerating their simulation rather than waiting for spontaneous occurrence.
- Motivating examples: For externally forced systems, trajectory-dependent injected power or entropy production can exhibit rare fluctuations beyond the limits imposed on their expectation values.Work extraction while lowering total entropy becomes exponentially less probable with system size and time interval.
- Computational challenge: Repeated or long simulations of the true dynamics soon become infeasible for studying these events.The review contrasts direct simulation with path sampling and a configuration-space population strategy.
- Computational strategy: The reviewed configuration-space method evolves copies of the system while replication or killing applies a controlled Darwinian pressure favoring atypical trajectories.The paper concentrates on this second class of rare-event methods rather than Monte Carlo dynamics in trajectory space.
2 Population dynamics
Population dynamics transforms trajectory large deviations into an evolution of independent clones whose replication or killing biases histories according to an observable. The resulting clone population estimates the Laplace transform and supports measurements within the selected atypical histories.
- Large deviations: For a time-averaged observable, the probability typically scales as p(A_o) ∼ e^−tI(A_o), where I(A_o) is obtained from a Legendre transform.The practical calculation uses the Laplace transform rather than estimating the probability directly.
- Clone dynamics: The method considers independent clones following the original noisy dynamics and applies replication or killing at each interval.Each clone is replaced on average by exp(αA(x_a)δt) clones, implementing the trajectory bias.
- Large deviations: The cloning dynamics yields the large-time generating quantity Z_t(α) through the average cloning or pruning rate.A finite population, typically in the hundreds, is used in practice instead of infinitely many clones.
- Implementation: The specific population-dynamics algorithm depends on whether state space and time are continuous or discrete.The review specifies the corresponding implementation separately for each example.
- Atypical histories: The algorithms also estimate observables among histories weighted by e^αF, enabling questions about other properties of unusually selected trajectories.For example, vorticity can be studied where energy dissipation is unusually large.
3 Biasing the stationary distribution: drift versus cloning
The review separates drift and cloning as complementary ways to bias trajectory sampling and shows that a change of basis can redistribute their relative contribution. This dynamic importance sampling differs from ordinary equilibrium reweighting.
- Bias decomposition: The operator H_α contains diffusion, drift, and cloning terms, corresponding to distinct components of the biased dynamics.The decomposition identifies which parts of the population evolution alter motion and which alter clone number.
- Bias decomposition: Dynamic importance sampling reshuffles the relative importance of drift and cloning through a change of basis.The review notes that no universally optimal choice of the field φ exists.
- Reweighting: The basis transformation reweights trajectories according to initial and final configurations while relating modified-dynamics averages to original many-time expectations.The transformed initial distribution is proportional to e^φP_o(x), and the final observable carries e^φ(t_n).
- Reweighting: This approach is not ordinary equilibrium Monte Carlo importance sampling, which modifies the energy and compensates through reweighted averages.The distinction matters because the modified dynamics here target trajectory histories rather than only equilibrium configurations.
- Example: turbulence: For driven turbulence, cloning or pruning based on the time derivative of a structure-function-related quantity can efficiently estimate its moments.The modified dynamics emphasize configurations associated with the selected intermittency statistics.
4 Transport
The transport examples apply cloning to rare current fluctuations in exclusion processes, using population dynamics to estimate large-deviation quantities and reveal atypical configurations. Results agree with analytic predictions and expose shocks whose motion depends on the bias and density.
- Population dynamics: Cloning estimates the large-deviation function for trajectory observables such as current in high-dimensional Markov transport models.The method biases transitions through cloning and a probability-conserving transition matrix, with Z_t(α) recovered from population growth or renormalization factors.
- Population dynamics: The algorithm keeps a fixed clone population by applying cloning or pruning followed by renormalization, while sequential implementations improve parallelization at the cost of population control.Finite-population simulations typically use hundreds of clones; sequential execution reduces parallel overhead but makes the total number harder to control.
- TASEP: For the TASEP, cloning rates are proportional to N times the number of particles with a free site to their right, targeting anomalous low-current configurations.The method runs independent copies and replicates configurations according to the relevant jump structure.
- TASEP: The computed Z_t(α) agrees excellently with analytic results using modest numerical effort, while the simulations reveal shocks and track a moving shock’s second-class particle.The method also exposes configurations responsible for anomalously small currents.
- TASEP: At N = 100 and density 0.5 with α = −50/N, the shock does not drift; at density 0.3 with α = −30/N, it drifts right as theory predicts.The simulations used L = 1000 clones initialized with random occupancy at density 0.5 for the first case.
- Continuous-time dynamics: Continuous-time cloning avoids discretization and rejection problems when rare-event trajectories span widely varying waiting times, although it requires two random numbers per configuration change.Discrete-time simulation can suffer many rejection events when rapid and slow configurations coexist, whereas continuous time is more cumbersome to implement.
5 Fluctuations of Dynamical Activity
Dynamical activity provides a way to study atypically mobile or immobile trajectories in glassy systems. In kinetically constrained models, these trajectories can exhibit active–inactive coexistence and a first-order dynamical phase transition.
- Dynamical activity K counts configuration changes and is used to identify trajectories that are faster or slower than average.Histories are weighted by e^-sK, favoring active histories for s < 0 and inactive histories for s > 0.
- The one-dimensional FA model imposes neighbor-dependent flipping, so active regions facilitate activity nearby.Inactive sites activate at rate c and active sites deactivate at rate 1 − c, provided at least one neighboring site is active.
- FA-model space-time histories show dynamical coexistence of active and inactive regions, analogous to liquid–solid coexistence in a static first-order transition.The comparison concerns atypical histories and treats time as one diagram direction.
- The continuous-time cloning algorithm computes the dynamical partition function, whose large-system non-analyticities signal a dynamical phase transition.The relevant averages are taken over histories of duration t in the large-t limit at fixed system size L.
- Dynamical phase coexistence: In the large-system limit, the active phase s ≤ 0 has finite dynamical free energy and active-site density, whereas both characterize an inactive phase for s > 0.For s > 0, the dynamical free energy is identically zero and the active-site density tends to zero; the density is discontinuous at s = 0.
- Dynamical phase coexistence: Whether molecular glass models share this transition remains unresolved because the inactive phase and the limiting behavior of the finite-size critical point are still being characterized.The open issue is whether sc(L) tends to 0 as L tends to infinity, determining whether standard dynamics at s = 0 is exactly critical.
6 Fluctuation of chaoticity in dynamical systems
The section develops Lyapunov Weighted Dynamics for sampling atypical chaoticity in deterministic and noisy Hamiltonian systems. Applied to the standard map and FPU chains, the method reveals coexistence and distinct trajectory structures associated with regular and chaotic dynamics.
- Chaoticity and large deviations: Lyapunov exponents quantify chaoticity through the growth of infinitesimal perturbations and can be studied using large-deviation methods.The resulting dynamical partition function and topological pressure characterize fluctuations of finite-time Lyapunov exponents.
- Population dynamics: Small stochastic noise is added to deterministic dynamics so clones can diversify and effectively sample trajectory space, especially when phase space contains disconnected regular and chaotic regions.The noise is used as a shortcut for the dependence of deterministic fluctuations on initial conditions.
- Population dynamics: Lyapunov Weighted Dynamics evolves cloned systems and tangent vectors, then replicates or prunes clones according to their tangent-vector renormalization factors.A fixed population is restored after each cloning step while normalization factors provide the dynamical partition function.
- The Standard Map: For the standard map, α < 0 localizes trajectories on integrable islands, whereas α > 0 detects chaotic layers surrounding those islands.The map transitions from integrable behavior at k = 0 toward increasingly chaotic behavior as k increases.
- The Standard Map: The standard map’s topological pressure indicates a critical point where chaotic and integrable trajectories coexist as in a first-order phase transition.The calculation uses the dynamical free energy and average Lyapunov exponent.
- FPU chains: In the FPU chain, regular-trajectory bias produces a long-lived ballistic gas of solitons, whereas chaotic-trajectory bias localizes long-lived chaotic breathers.The regular-bias result requires setting the center-of-mass velocity to zero; whether the chain has a corresponding critical point remains uncomputed.
7 Work and entropy production
The section studies fluctuations of work, entropy production, and dissipated power in driven systems. Cloning provides a way to probe large deviations and verify fluctuation-theorem symmetries, including in the Sinai billiard.
- Motivation: Work and entropy production fluctuate with microscopic configurations and bath conditions, while the Second Law constrains their averages rather than the full range of rare fluctuations.Rare negative-entropy or atypical-work trajectories can therefore be exponentially unlikely while remaining relevant to fluctuation studies.
- Fluctuation relations: Transient, Jarzynski, stationary, and Gallavotti-Cohen fluctuation relations connect probabilities or generating functions of opposite work or entropy-production fluctuations.The Gallavotti-Cohen case concerns deterministic dynamics and depends on ergodic properties involving attractor and repellor sets.
- Open regimes: The method is motivated as a tool for exploring regimes where strong forcing excites macroscopic structures and produces fluctuations much larger than microscopic thermal scales.The text identifies the limits of rigorous fluctuation relations in hydrodynamic systems as open questions.
- Sinai billiard: Cloning was applied to the Sinai billiard to calculate dissipated-power fluctuations and test the fluctuation-theorem symmetry in a driven chaotic system.The billiard uses periodic boundaries, an external field, and a deterministic thermostat fixing the velocity modulus.
- Sinai billiard: For the Sinai billiard, clone diversity is maintained with random kicks and macroscopic cloning intervals chosen according to the system’s chaotic properties.The reported settings allow a few collisions and support clone separation for noise intensities Δ = 10^-3 to 10^-4.
8 Planetary systems
The section presents planetary dynamics as a rare-event problem because observationally compatible initial conditions can generate widely different past and future trajectories. Cloning could sample these atypical orbital histories, but the proposed detailed applications remain future work.
- Motivation: Planetary systems require statistical trajectory analysis because observationally allowed initial conditions can produce widely varying inferences about their past and future.The paper emphasizes that many discovered systems are stable only over limited past or future intervals.
- Sensitivity of planetary trajectories: Small changes in present planetary positions can produce dramatically different orbital outcomes, including possible intersections between Mercury’s and Venus’s orbits.An example considers displacing Earth’s position by approximately 150 meters.
- Rare-event sampling: Cloning based on an observable such as eccentricity change could generate the full probability distribution of that observable at later times.The procedure tracks the total clone population or its normalization factors during the evolution.
- Rare-event sampling: In chaotic planetary dynamics, negligible displacements or noise can diversify clones and produce a range of trajectories.The perturbations may be comparable to other neglected external sources of displacement.
- Prospects: A large-deviation analysis of planetary-system histories is proposed as a way to investigate possible past and future evolution and the self-organization of solar-system stability.The text presents these applications as interesting future studies rather than completed calculations.