-
Deep Learning of the Evolution Operator Enables Forecasting of Out-of-Training Dynamics in Chaotic Systems
Authors:
Ira J. S. Shokar,
Peter H. Haynes,
Rich R. Kerswell
Abstract:
We demonstrate that a deep learning emulator for chaotic systems can forecast phenomena absent from training data. Using the Kuramoto-Sivashinsky and beta-plane turbulence models, we evaluate the emulator through scenarios probing the fundamental phenomena of both systems: forecasting spontaneous relaminarisation, capturing initialisation of arbitrary chaotic states, zero-shot prediction of dynami…
▽ More
We demonstrate that a deep learning emulator for chaotic systems can forecast phenomena absent from training data. Using the Kuramoto-Sivashinsky and beta-plane turbulence models, we evaluate the emulator through scenarios probing the fundamental phenomena of both systems: forecasting spontaneous relaminarisation, capturing initialisation of arbitrary chaotic states, zero-shot prediction of dynamics with parameter values outside of the training range, and characterisation of dynamical statistics from artificially restricted training datasets. Our results show that deep learning emulators can uncover emergent behaviours and rare events in complex systems by learning underlying mathematical rules, rather than merely mimicking observed patterns.
△ Less
Submitted 27 February, 2025;
originally announced February 2025.
-
Stochastic Latent Transformer: Efficient Modelling of Stochastically Forced Zonal Jets
Authors:
Ira J. S. Shokar,
Rich R. Kerswell,
Peter H. Haynes
Abstract:
We present a novel probabilistic deep learning approach, the 'Stochastic Latent Transformer' (SLT), designed for the efficient reduced-order modelling of stochastic partial differential equations. Stochastically driven flow models are pertinent to a diverse range of natural phenomena, including jets on giant planets, ocean circulation, and the variability of midlatitude weather. However, much of t…
▽ More
We present a novel probabilistic deep learning approach, the 'Stochastic Latent Transformer' (SLT), designed for the efficient reduced-order modelling of stochastic partial differential equations. Stochastically driven flow models are pertinent to a diverse range of natural phenomena, including jets on giant planets, ocean circulation, and the variability of midlatitude weather. However, much of the recent progress in deep learning has predominantly focused on deterministic systems. The SLT comprises a stochastically-forced transformer paired with a translation-equivariant autoencoder, trained towards the Continuous Ranked Probability Score. We showcase its effectiveness by applying it to a well-researched zonal jet system, where the interaction between stochastically forced eddies and the zonal mean flow results in a rich low-frequency variability. The SLT accurately reproduces system dynamics across various integration periods, validated through quantitative diagnostics that include spectral properties and the rate of transitions between distinct states. The SLT achieves a five-order-of-magnitude speedup in emulating the zonally-averaged flow compared to direct numerical simulations. This acceleration facilitates the cost-effective generation of large ensembles, enabling the exploration of statistical questions concerning the probabilities of spontaneous transition events.
△ Less
Submitted 18 December, 2023; v1 submitted 25 October, 2023;
originally announced October 2023.
-
Dispersion in the large-deviation regime. Part II: cellular flow at large Péclet number
Authors:
P. H. Haynes,
J. Vanneste
Abstract:
A standard model for the study of scalar dispersion through advection and molecular diffusion is a two-dimensional periodic flow with closed streamlines inside periodic cells. Over long time scales, the dispersion of a scalar in this flow can be characterised by an effective diffusivity that is a factor $\mathrm{Pe}^{1/2}$ larger than molecular diffusivity when the Péclet number $\mathrm{Pe}$ is l…
▽ More
A standard model for the study of scalar dispersion through advection and molecular diffusion is a two-dimensional periodic flow with closed streamlines inside periodic cells. Over long time scales, the dispersion of a scalar in this flow can be characterised by an effective diffusivity that is a factor $\mathrm{Pe}^{1/2}$ larger than molecular diffusivity when the Péclet number $\mathrm{Pe}$ is large. Here we provide a more complete description of dispersion in this regime by applying the large-deviation theory developed in Part I of this paper. We derive approximations to the rate function governing the scalar concentration at large time $t$ by carrying out an asymptotic analysis of the relevant family of eigenvalue problems. We identify two asymptotic regimes and make predictions for the rate function and spatial structure of the scalar. Regime I applies to distances from the release point that satisfy $|\boldsymbol{x}| = O(\mathrm{Pe}^{1/4} t)$ . The concentration in this regime is isotropic at large scales, is uniform along streamlines within each cell, and varies rapidly in boundary layers surrounding the separatrices between adjacent cells. The results of homogenisation theory are recovered from our analysis. Regime II applies when $|\boldsymbol{x}|=O(\mathrm{Pe} \, t/\log \mathrm{Pe})$ and is characterised by an anisotropic concentration distribution that is localised around the separatrices. A novel feature of this regime is the crucial role played by the dynamics near the hyperbolic stagnation points. A consequence is that in part of the regime the dispersion can be interpreted as resulting from a random walk on the lattice of stagnation points. The two regimes overlap so that our asymptotic results describe the scalar concentration over a large range of distances. They are verified against numerical solutions of the family of eigenvalue problems yielding the rate function.
△ Less
Submitted 26 January, 2014;
originally announced January 2014.
-
Dispersion in the large-deviation regime. Part I: shear flows and periodic flows
Authors:
P. H. Haynes,
J. Vanneste
Abstract:
The dispersion of a passive scalar in a fluid through the combined action of advection and molecular diffusion is often described as a diffusive process, with an effective diffusivity that is enhanced compared to the molecular value. However, this description fails to capture the tails of the scalar concentration distribution in initial-value problems. To remedy this, we develop a large-deviation…
▽ More
The dispersion of a passive scalar in a fluid through the combined action of advection and molecular diffusion is often described as a diffusive process, with an effective diffusivity that is enhanced compared to the molecular value. However, this description fails to capture the tails of the scalar concentration distribution in initial-value problems. To remedy this, we develop a large-deviation theory of scalar dispersion that provides an approximation to the scalar concentration valid at much larger distances away from the centre of mass, specifically distances that are $O(t)$ rather than $O(t^{1/2})$, where $t \gg 1$ is the time from the scalar release. The theory centres on the calculation of a rate function obtained by solving a one-parameter family of eigenvalue problems which we derive using two alternative approaches, one asymptotic, the other probabilistic. We emphasise the connection between large deviations and homogenisation: a perturbative solution of the eigenvalue problems reduces at leading order to the cell problem of homogenisation theory. We consider two classes of flows in some detail: shear flows and cellular flows. In both cases, large deviation generalises classical results on effective diffusivity and captures new phenomena relevant to the tails of the scalar distribution. These include approximately finite dispersion speeds arising at large Péclet number $\mathrm{Pe}$ (corresponding to small molecular diffusivity) and, for two-dimensional cellular flows, anisotropic dispersion. Explicit asymptotic results are obtained for shear flows in the limit of large $\mathrm{Pe}$. (A companion paper, Part II, is devoted to the large-$\mathrm{Pe}$ asymptotic treatment of cellular flows.) The predictions of large-deviation theory are compared with Monte Carlo simulations that estimate the tails of concentration accurately using importance sampling.
△ Less
Submitted 26 January, 2014;
originally announced January 2014.
-
Homogenization for advection-diffusion in a perforated domain
Authors:
P. H. Haynes,
V. H. Hoang,
J. R. Norris,
K. C. Zygalakis
Abstract:
The volume of a Wiener sausage constructed from a diffusion process with periodic, mean-zero, divergence-free velocity field, in dimension 3 or more, is shown to have a non-random and positive asymptotic rate of growth. This is used to establish the existence of a homogenized limit for such a diffusion when subject to Dirichlet conditions on the boundaries of a sparse and independent array of obst…
▽ More
The volume of a Wiener sausage constructed from a diffusion process with periodic, mean-zero, divergence-free velocity field, in dimension 3 or more, is shown to have a non-random and positive asymptotic rate of growth. This is used to establish the existence of a homogenized limit for such a diffusion when subject to Dirichlet conditions on the boundaries of a sparse and independent array of obstacles. There is a constant effective long-time loss rate at the obstacles. The dependence of this rate on the form and intensity of the obstacles and on the velocity field is investigated. A Monte Carlo algorithm for the computation of the volume growth rate of the sausage is introduced and some numerical results are presented for the Taylor--Green velocity field.
△ Less
Submitted 25 March, 2010; v1 submitted 21 March, 2010;
originally announced March 2010.
-
The role of a delay time on the spatial structure of chaotically advected reactive scalars
Authors:
Alexandra Tzella,
Peter H. Haynes
Abstract:
The stationary-state spatial structure of reacting scalar fields, chaotically advected by a two-dimensional large-scale flow, is examined for the case for which the reaction equations contain delay terms. Previous theoretical investigations have shown that, in the absence of delay terms and in a regime where diffusion can be neglected (large Péclet number), the emergent spatial structures are fi…
▽ More
The stationary-state spatial structure of reacting scalar fields, chaotically advected by a two-dimensional large-scale flow, is examined for the case for which the reaction equations contain delay terms. Previous theoretical investigations have shown that, in the absence of delay terms and in a regime where diffusion can be neglected (large Péclet number), the emergent spatial structures are filamental and characterized by a single scaling regime with a Hölder exponent that depends on the rate of convergence of the reactive processes and the strength of the stirring measured by the average stretching rate. In the presence of delay terms, we show that for sufficiently small scales all interacting fields should share the same spatial structure, as found in the absence of delay terms. Depending on the strength of the stirring and the magnitude of the delay time, two further scaling regimes that are unique to the delay system may appear at intermediate length scales. An expression for the transition length scale dividing small-scale and intermediate-scale regimes is obtained and the scaling behavior of the scalar field is explained. The theoretical results are illustrated by numerical calculations for two types of reaction models, both based on delay differential equations, coupled to a two-dimensional chaotic advection flow. The first corresponds to a single reactive scalar and the second to a nonlinear biological model that includes nutrients, phytoplankton and zooplankton. As in the no-delay case, the presence of asymmetrical couplings among the biological species results in a non-generic scaling behavior.
△ Less
Submitted 27 May, 2009;
originally announced May 2009.
-
Diffusion of passive scalar in a finite-scale random flow
Authors:
A. A. Schekochihin,
P. H. Haynes,
S. C. Cowley
Abstract:
We consider a solvable model of the decay of scalar variance in a single-scale random velocity field. We show that if there is a separation between the flow scale k_flow^{-1} and the box size k_box^{-1}, the decay rate lambda ~ (k_box/k_flow)^2 is determined by the turbulent diffusion of the box-scale mode. Exponential decay at the rate lambda is preceded by a transient powerlike decay (the tota…
▽ More
We consider a solvable model of the decay of scalar variance in a single-scale random velocity field. We show that if there is a separation between the flow scale k_flow^{-1} and the box size k_box^{-1}, the decay rate lambda ~ (k_box/k_flow)^2 is determined by the turbulent diffusion of the box-scale mode. Exponential decay at the rate lambda is preceded by a transient powerlike decay (the total scalar variance ~ t^{-5/2} if the Corrsin invariant is zero, t^{-3/2} otherwise) that lasts a time t~1/λ. Spectra are sharply peaked at k=k_box. The box-scale peak acts as a slowly decaying source to a secondary peak at the flow scale. The variance spectrum at scales intermediate between the two peaks (k_box<<k<<k_flow) is ~ k + a k^2 + ... (a>0). The mixing of the flow-scale modes by the random flow produces, for the case of large Peclet number, a k^{-1+delta} spectrum at k>>k_flow, where delta ~ lambda is a small correction. Our solution thus elucidates the spectral make up of the ``strange mode,'' combining small-scale structure and a decay law set by the largest scales.
△ Less
Submitted 1 December, 2004; v1 submitted 8 April, 2004;
originally announced April 2004.
-
Chaotic mixing induced transitions in reaction-diffusion systems
Authors:
Zoltan Neufeld,
Peter H. Haynes,
Tamas Tel
Abstract:
We study the evolution of a localized perturbation in a chemical system with multiple homogeneous steady states, in the presence of stirring by a fluid flow. Two distinct regimes are found as the rate of stirring is varied relative to the rate of the chemical reaction. When the stirring is fast localized perturbations decay towards a spatially homogeneous state. When the stirring is slow (or fas…
▽ More
We study the evolution of a localized perturbation in a chemical system with multiple homogeneous steady states, in the presence of stirring by a fluid flow. Two distinct regimes are found as the rate of stirring is varied relative to the rate of the chemical reaction. When the stirring is fast localized perturbations decay towards a spatially homogeneous state. When the stirring is slow (or fast reaction) localized perturbations propagate by advection in form of a filament with a roughly constant width and exponentially increasing length. The width of the filament depends on the stirring rate and reaction rate but is independent of the initial perturbation. We investigate this problem numerically in both closed and open flow systems and explain the results using a one-dimensional "mean-strain" model for the transverse profile of the filament that captures the interplay between the propagation of the reaction-diffusion front and the stretching due to chaotic advection.
△ Less
Submitted 25 February, 2002;
originally announced February 2002.
-
The effect of forcing on the spatial structure and spectra of chaotically advected passive scalars
Authors:
Zoltan Neufeld,
Peter H. Haynes,
Guillemette Picard
Abstract:
The stationary distribution of passive tracers chaotically advected by a two-dimensional large-scale flow is investigated. The tracer field is force by resetting the value of the tracer in certain localised regions. This problem is mathematically equivalent to advection in open flows and results in a fractal tracer structure. The spectral exponent of the tracer field is different from that for a…
▽ More
The stationary distribution of passive tracers chaotically advected by a two-dimensional large-scale flow is investigated. The tracer field is force by resetting the value of the tracer in certain localised regions. This problem is mathematically equivalent to advection in open flows and results in a fractal tracer structure. The spectral exponent of the tracer field is different from that for a passive tracer with the usual additive forcing (the so called Batchelor spectrum) and is related to the fractal dimension of the set of points that have never visited the forcing regions. We illustrate this behaviour by considering a time-periodic flow whose effect is equivalent to a simple two-dimensional area-preserving map. We also show that similar structure in the tracer field is found when the flow is aperiodic in time.
△ Less
Submitted 9 December, 1999;
originally announced December 1999.
-
Chaotic advection of reacting substances: Plankton dynamics on a meandering jet
Authors:
Cristobal Lopez,
Zoltan Neufeld,
Emilio Hernandez-Garcia,
Peter H. Haynes
Abstract:
We study the spatial patterns formed by interacting populations or reacting chemicals under the influence of chaotic flows. In particular, we have considered a three-component model of plankton dynamics advected by a meandering jet. We report general results, stressing the existence of a smooth-filamental transition in the concentration patterns depending on the relative strength of the stirring…
▽ More
We study the spatial patterns formed by interacting populations or reacting chemicals under the influence of chaotic flows. In particular, we have considered a three-component model of plankton dynamics advected by a meandering jet. We report general results, stressing the existence of a smooth-filamental transition in the concentration patterns depending on the relative strength of the stirring by the chaotic flow and the relaxation properties of planktonic dynamical system. Patterns obtained in open and closed flows are compared.
△ Less
Submitted 17 June, 1999;
originally announced June 1999.
-
Smooth-filamental transition of active tracer fields stirred by chaotic advection
Authors:
Zoltan Neufeld,
Cristobal Lopez,
Peter H. Haynes
Abstract:
The spatial distribution of interacting chemical fields is investigated in the non-diffusive limit. The evolution of fluid parcels is described by independent dynamical systems driven by chaotic advection. The distribution can be filamental or smooth depending on the relative strength of the dispersion due to chaotic advection and the stability of the chemical dynamics. We give the condition for…
▽ More
The spatial distribution of interacting chemical fields is investigated in the non-diffusive limit. The evolution of fluid parcels is described by independent dynamical systems driven by chaotic advection. The distribution can be filamental or smooth depending on the relative strength of the dispersion due to chaotic advection and the stability of the chemical dynamics. We give the condition for the smooth-filamental transition and relate the Hölder exponent of the filamental structure to the Lyapunov exponents. Theoretical findings are illustrated by numerical experiments.
△ Less
Submitted 10 June, 1999;
originally announced June 1999.
