Searcharxiv⌕ Search

arXiv subjects

Timo Schorlepp

Publications and source records attributed to Timo Schorlepp.

17 recordsLinked to original sources

Computing rare events in Lévy-driven systems via optimal control

We extend the optimal control formulation of sample path large deviation theory, which has proven effective for numerical computation of instantons and sharp probability estimates in stochastic differential equations with Gaussian noise, to systems with light-tailed Lévy noise. In the scaling limit of small jumps and high intensity, such systems satisfy a large deviation principle of Freidlin-Wentzell type, with a running cost given by the Legendre transform of the noise cumulant generating function. We propose to numerically minimize this cost over noise realizations subject to the system dynamics, which yields the instanton without having to solve Hamilton's equations, and naturally handles degenerate and multiplicative noise. When the cumulant generating function has a bounded domain, the cost grows only linearly for large noise, and instantons can contain jumps. We analyze when jumps are not only admissible but optimal, depending on the subleading behavior of the cost and on the drift, and show how jumps can be resolved numerically through a time reparameterization that freezes physical time. For continuous instantons, we derive Riccati and Fredholm determinant expressions for the leading-order prefactor, and we discuss the modifications needed when the instanton jumps. We illustrate and validate our methods against Monte Carlo simulations in several examples: a two-dimensional nonlinear model, an active run-and-tumble particle with non-Gaussian tumbling noise, and the Ghil-Sellers energy balance climate model with Lévy-perturbed solar forcing.

stat.CO↗

Large deviations of spatial moments of the Kipnis-Marchioro-Presutti lattice gas following a localized energy release

We study the long-time statistics of spatial moments of the time-dependent energy distribution in a one-dimensional Kipnis-Marchioro-Presutti model following an instantaneous release of energy on a single lattice site. The moments are $M_n(T)=\sum_i x_i^n u_i(T)$, where $u_i(T)$ is the energy at site $i$ at the observation time $T$. Employing macroscopic fluctuation theory (MFT), we calculate the variance of $M_n$ for every positive integer $n$ and determine the non-Gaussian distribution tails for the representative cases $n=1$ and $2$. For the first moment, we establish an exact duality, within MFT, to integrated-current statistics for a deterministic flat initial density. For both $n=1$ and $n=2$, the large-moment tails are dominated by a single adiabatically evolving doublon: a rapidly propagating pair of narrow solitary pulses of the energy density and of the gradient of the conjugate field. The small-second-moment tail is associated with concentration of the energy near the origin at time $T$. Comparing single- and two-doublon histories, we predict a dynamical phase transition at a critical value of $M_2$ at which the optimal path breaks reflection symmetry $x\to -x$. Numerical solutions obtained by direct minimization of the MFT action establish the second-order character of this transition and support our analytical predictions. Finally, we extend calculations of the large- and small-moment tails to arbitrary $n$ and numerically find a transition of the same type for $n=4$.

cond-mat.stat-mech↗

Beyond exponential tilting: sharp asymptotics and improved importance sampling for rare events

A classical approach for importance sampling (IS) of a rare event is to tilt the measure using its large deviation minimizer, thereby shifting the proposal mean towards the event set. While this tilt is log-efficient, we show that the coefficient of variation of the resulting estimator still grows like $\varepsilon^{-1/4}$ in the small-noise limit $\varepsilon\downarrow0$. Motivated by sharp large deviation asymptotics and the local geometry of the event set at the minimizer, we propose improvements using a control variate or an anisotropic, non-Gaussian proposal. We prove, up to truncations that hold with high probability, and confirm empirically in high-dimensional examples, that the former bounds the coefficient of variation and the latter makes it vanish, so that a bounded number of samples suffices. Our analysis addresses both the coefficient of variation of the different proposals and their necessary and sufficient sample size based on the Kullback--Leibler divergence of the IS target from the proposal; both yield the same rates in $\varepsilon$. We demonstrate the resulting samplers, including a variant that blends the classical and the new proposal across noise levels, for a stochastic Korteweg--de Vries equation with thousands of random inputs.

stat.CO↗

Large Deviations in Switching Diffusion: Dynamical Condensation and Diffusivity BBP Transition

We consider a Brownian particle with position $x(t)$ whose diffusivity changes at Poissonian switching times, with each new value drawn independently from a bounded distribution $W(D)$. Assuming $W(D)\sim C(D_{\max}-D)^ν$ near its upper edge, we study the large-deviation regime $x(t)\simeq yt$ with $y = O(1)$ at large times, far beyond the typical diffusive scale. In a recent work Guéneau et al., Phys. Rev. Lett. 135, 067102 (2025), the rate function of the displacement was obtained through a renewal approach and it was shown to display several singularities. In this paper, we carry out a detailed study of the corresponding dynamical phase transitions and trajectory condensation. For moderate $y$, the imposed displacement is shared among $O(t)$ displacement increments accumulated during successive intervals between switches. Beyond a first threshold $y_-$, one interval produces an increment of order $t$ carrying a finite fraction of the total displacement, signaling dynamical condensation. For $ν>1$, a second threshold $y_+>y_-$ is reached when this macroscopic increment carries asymptotically the entire imposed displacement, whereas for $0<ν\leq1$ the two thresholds merge, $y_-=y_+$. In turn, the moment-generating function of $x(t)$ displays a complementary transition from collective contributions of many diffusivities to selection of states near $D_{\max}$. We show that this transition is governed by the same edge mechanism as the Baik-Ben Arous-Péché (BBP) transition, indicating that diffusivity selection and trajectory condensation are two manifestations of the same phenomenon. We validate our predictions for the dynamical condensation against numerical results from rare event sampling.

cond-mat.stat-mech↗

How turbulent flows grow vorticity at a point

We describe how strong vorticity fluctuations appear in three-dimensional incompressible turbulence driven by a large-scale random force. By combining analytical and numerical methods, we construct a vorticity instanton: the most probable force history driving the flow to a prescribed large vorticity at a given point and time. We show how the smooth, large-scale force produces a long vortex filament with a thin, viscous-scale core. Our main finding is that the standard mechanism of amplifying a vortex tube through radial compression and axial stretching is only effective up to a threshold value. Producing vorticity above the threshold requires an additional ingredient: vorticity waves propagating along the tube. The final vorticity is the result of two colliding waves. This mechanism has striking similarities to the one recently suggested by OpenAI for creating infinite vorticity in finite-time Navier--Stokes blow-up.

physics.flu-dyn↗

Intermittency from instanton calculus at the transition to turbulence and fusion rules

Understanding intermittency of turbulent systems from the underlying differential equations is an outstanding problem in fluid dynamics. Here, in the example of Burgers turbulence as a stringent test, we introduce a semi-analytical method that yields high-order structure function exponents by combining instanton calculus, fusion rule predictions, and low-order statistical inputs from direct numerical simulations (DNS). We use instanton predictions, calibrated by DNS, to evaluate high velocity gradient (VG) moments at the onset of intermittency, and then infer scaling exponents in fully developed turbulence via fusion rules. We show that the method captures the crossover at $\mathrm{Re}_λ\approx 1$ in the VG moment scaling, highlight the necessity of including fluctuations around instantons, and discuss future extensions.

physics.flu-dyn↗

First-Order Trajectory Matching: Fast Ensemble Predictions of Chaotic, Turbulent, Stochastic Systems

We introduce First-Order Trajectory Matching (FTM), a surrogate-modeling method that learns the first-order local transport of probability mass from trajectories of stochastic systems. By matching the symmetric first-order motion of trajectories, FTM learns the probability current velocity, whose flow preserves time marginals to match ensemble averages, while also capturing current-like trajectory quantities such as fluxes, circulations, and barrier-crossing currents. FTM learns the current velocity directly from trajectories, avoiding drift, diffusion, and score estimation. Our stability analysis separates discretization error from sampling variance and shows that the one-step simulation-free FTM loss is stable when temporal resolution and sample size are properly balanced. Across stochastic dynamical systems and PDE examples, we empirically demonstrate that FTM provides trajectory-aware ensemble predictions at low, deterministic-rollout cost.

cs.LG↗

Scalability of the second-order reliability method for stochastic differential equations with multiplicative noise

We show how to efficiently compute asymptotically sharp estimates of extreme event probabilities in stochastic differential equations (SDEs) with small multiplicative Brownian noise. The underlying approximation is known as sharp large deviation theory or precise Laplace asymptotics in mathematics, the second-order reliability method (SORM) in reliability engineering, and the instanton or optimal fluctuation method with 1-loop corrections in physics. It is based on approximating the tail probability in question with the most probable realization of the stochastic process, and local perturbations around this realization. We first recall and contextualize the relevant classical theoretical result on precise Laplace asymptotics of diffusion processes [Ben Arous (1988), Stochastics, 25(3), 125-153], and then show how to compute the involved infinite-dimensional quantities - operator traces and Carleman-Fredholm determinants - numerically in a way that is scalable with respect to the time discretization and remains feasible in high spatial dimensions. Using tools from automatic differentiation, we achieve a straightforward black-box numerical computation of the SORM estimates in JAX. The method is illustrated in examples of SDEs and stochastic partial differential equations, including a two-dimensional random advection-diffusion model of a passive scalar. We thereby demonstrate that it is possible to obtain efficient and accurate SORM estimates for very high-dimensional problems, as long as the infinite-dimensional structure of the problem is correctly taken into account. Our JAX implementation of the method is made publicly available.

stat.CO↗

Precise asymptotic analysis of Sobolev training for random feature models

Gradient information is widely useful and available in applications, and is therefore natural to include in the training of neural networks. Yet little is known theoretically about the impact of Sobolev training -- regression with both function and gradient data -- on the generalization error of highly overparameterized predictive models in high dimensions. In this paper, we obtain a precise characterization of this training modality for random feature (RF) models in the limit where the number of trainable parameters, input dimensions, and training data tend proportionally to infinity. Our model for Sobolev training reflects practical implementations by sketching gradient data onto finite dimensional subspaces. By combining the replica method from statistical physics with linearizations in operator-valued free probability theory, we derive a closed-form description for the generalization errors of the trained RF models. For target functions described by single-index models, we demonstrate that supplementing function data with additional gradient data does not universally improve predictive performance. Rather, the degree of overparameterization should inform the choice of training method. More broadly, our results identify settings where models perform optimally by interpolating noisy function and gradient data.

stat.ML↗

Synthetic Turbulence via an Instanton Gas Approximation

Sampling synthetic turbulent fields as a computationally tractable surrogate for direct numerical simulations (DNS) is an important practical problem in various applications, and allows to test our physical understanding of the main features of real turbulent flows. Reproducing higher-order Eulerian correlation functions, as well as Lagrangian particle statistics, requires an accurate representation of coherent structures of the flow in the synthetic turbulent fields. To this end, we propose in this paper a systematic coherent-structure based method for sampling synthetic random fields, based on a superposition of instanton configurations - an instanton gas - from the field-theoretic formulation of turbulence. We discuss sampling strategies for ensembles of instantons, both with and without interactions and including Gaussian fluctuations around them. The resulting Eulerian and Lagrangian statistics are evaluated numerically and compared against DNS results, as well as Gaussian and log-normal cascade models that lack coherent structures. The instanton gas approach is illustrated via the example of one-dimensional Burgers turbulence throughout this paper, and we show that already a canonical ensemble of non-interacting instantons without fluctuations reproduces DNS statistics very well. Finally, we outline extensions of the method to higher dimensions, in particular to magnetohydrodynamic turbulence for future applications to cosmic ray propagation.

physics.flu-dyn↗

Systematic analysis of critical exponents in continuous dynamical phase transitions of weak noise theories

Dynamical phase transitions are nonequilibrium counterparts of thermodynamic phase transitions and share many similarities with their equilibrium analogs. In continuous phase transitions, critical exponents play a key role in characterizing the physics near criticality. This study aims to systematically analyze the set of possible critical exponents in weak noise statistical field theories in 1+1 dimensions, focusing on cases with a single fluctuating field. To achieve this, we develop and apply the Gaussian fluctuation method, avoiding reliance on constructing a Landau theory based on system symmetries. Our analysis reveals that the critical exponents can be categorized into a limited set of distinct cases, suggesting a constrained universality in weak noise-induced dynamical phase transitions. We illustrate our findings in two examples: short-time large deviations of the Kardar-Parisi-Zhang equation, and the weakly asymmetric exclusion process on a ring within the framework of the macroscopic fluctuation theory.

cond-mat.stat-mech↗

Scalable Methods for Computing Sharp Extreme Event Probabilities in Infinite-Dimensional Stochastic Systems

We introduce and compare computational techniques for sharp extreme event probability estimates in stochastic differential equations with small additive Gaussian noise. In particular, we focus on strategies that are scalable, i.e. their efficiency does not degrade upon temporal and possibly spatial refinement. For that purpose, we extend algorithms based on the Laplace method for estimating the probability of an extreme event to infinite dimensional path space. The method estimates the limiting exponential scaling using a single realization of the random variable, the large deviation minimizer. Finding this minimizer amounts to solving an optimization problem governed by a differential equation. The probability estimate becomes sharp when it additionally includes prefactor information, which necessitates computing the determinant of a second derivative operator to evaluate a Gaussian integral around the minimizer. We present an approach in infinite dimensions based on Fredholm determinants, and develop numerical algorithms to compute these determinants efficiently for the high-dimensional systems that arise upon discretization. We also give an interpretation of this approach using Gaussian process covariances and transition tubes. An example model problem, for which we provide an open-source python implementation, is used throughout the paper to illustrate all methods discussed. To study the performance of the methods, we consider examples of stochastic differential and stochastic partial differential equations, including the randomly forced incompressible three-dimensional Navier-Stokes equations.

stat.CO↗

Short-time large deviations of the spatially averaged height of a KPZ interface on a ring

Using the optimal fluctuation method, we evaluate the short-time probability distribution $P (\bar{H}, L, t=T)$ of the spatially averaged height $\bar{H} = (1/L) \int_0^L h(x, t=T) \, dx$ of a one-dimensional interface $h(x, t)$ governed by the Kardar-Parisi-Zhang equation $$ \partial_th=ν\partial_x^2h+\fracλ{2} \left(\partial_xh\right)^2+\sqrt{D}ξ(x,t) $$ on a ring of length $L$. The process starts from a flat interface, $h(x,t=0)=0$. Both at $λ\bar{H} < 0$, and at sufficiently small positive $λ\bar{H}$ the optimal (that is, the least-action) path $h(x,t)$ of the interface, conditioned on $\bar{H}$, is uniform in space, and the distribution $P (\bar{H}, L, T)$ is Gaussian. However, at sufficiently large $λ\bar{H} > 0$ the spatially uniform solution becomes sub-optimal and gives way to non-uniform optimal paths. We study them, and the resulting non-Gaussian distribution $P (\bar{H}, L, T)$, analytically and numerically. The loss of optimality of the uniform solution occurs via a dynamical phase transition of either first, or second order, depending on the rescaled system size $\ell = L/\sqrt{νT}$, at a critical value $\bar{H}=\bar{H}_{\text{c}}(\ell)$. At large but finite $\ell$ the transition is of first order. Remarkably, it becomes an "accidental" second-order transition in the limit of $\ell \to \infty$, where a large-deviation behavior $-\ln P (\bar{H}, L, T) \simeq (L/T) f(\bar{H})$ (in the units $λ=ν=D=1$) is observed. At small $\ell$ the transition is of second order, while at $\ell =O(1)$ transitions of both types occur.

cond-mat.stat-mech↗

Symmetries and zero modes in sample path large deviations

Sharp large deviation estimates for stochastic differential equations with small noise, based on minimizing the Freidlin-Wentzell action functional under appropriate boundary conditions, can be obtained by integrating certain matrix Riccati differential equations along the large deviation minimizers or instantons, either forward or backward in time. Previous works in this direction often rely on the existence of isolated minimizers with positive definite second variation. By adopting techniques from field theory and explicitly evaluating the large deviation prefactors as functional determinant ratios using Forman's theorem, we extend the approach to general systems where degenerate submanifolds of minimizers exist. The key technique for this is a boundary-type regularization of the second variation operator. This extension is particularly relevant if the system possesses continuous symmetries that are broken by the instantons. We find that removing the vanishing eigenvalues associated with the zero modes is possible within the Riccati formulation and amounts to modifying the initial or final conditions and evaluation of the Riccati matrices. We apply our results in multiple examples including a dynamical phase transition for the average surface height in short-time large deviations of the one-dimensional Kardar-Parisi-Zhang equation with flat initial profile.

cond-mat.stat-mech↗

Spontaneous Symmetry Breaking for Extreme Vorticity and Strain in the 3D Navier-Stokes Equations

We investigate the spatio-temporal structure of the most likely configurations realising extremely high vorticity or strain in the stochastically forced 3D incompressible Navier-Stokes equations. Most likely configurations are computed by numerically finding the highest probability velocity field realising an extreme constraint as solution of a large optimisation problem. High-vorticity configurations are identified as pinched vortex filaments with swirl, while high-strain configurations correspond to counter-rotating vortex rings. We additionally observe that the most likely configurations for vorticity and strain spontaneously break their rotational symmetry for extremely high observable values. Instanton calculus and large deviation theory allow us to show that these maximum likelihood realisations determine the tail probabilities of the observed quantities. In particular, we are able to demonstrate that artificially enforcing rotational symmetry for large strain configurations leads to a severe underestimate of their probability, as it is dominated in likelihood by an exponentially more likely symmetry broken vortex-sheet configuration.

physics.flu-dyn↗

Gel'fand-Yaglom type equations for calculating fluctuations around Instantons in stochastic systems

In recent years, instanton calculus has successfully been employed to estimate tail probabilities of rare events in various stochastic dynamical systems. Without further corrections, however, these estimates can only capture the exponential scaling. In this paper, we derive a general, closed form expression for the leading prefactor contribution of the fluctuations around the instanton trajectory for the computation of probability density functions of general observables. The key technique is applying the Gel'fand-Yaglom recursive evaluation method to the suitably discretized Gaussian path integral of the fluctuations, in order to obtain matrix evolution equations that yield the fluctuation determinant. We demonstrate agreement between these predictions and direct sampling for examples motivated from turbulence theory.

cond-mat.stat-mech↗

Solenoidal Improvements for the JF12 Galactic Magnetic Field Model

The popular JF12 analytic model by Jansson & Farrar provides a quantitative description of the Galaxy's large-scale magnetic field, which is widely used in various astrophysical applications. However, both the poloidal X-type component and the spiral disk component of JF12 exhibit regions in which the magnetic divergence constraint is violated. We first propose a cure for this problem, resulting in a truly solenoidal large-scale spiral field. Second, the otherwise straight field lines of the X-type component exhibit kinks in the Galactic plane that, in addition to implying the presence of a singular current sheet, may pose difficulties for e.g., numerical tracing of cosmic-ray particles. We propose and discuss two possible strategies to mitigate this problem. Although all corrections are kept as minimal as possible, the extended set of model parameters will have to be carefully readjusted in order to fully restore the agreement to observational data that the unmodified JF12 field is based on. Furthermore, the performance of our improved version of the field model is quantitatively assessed by test simulations using the CRPropa Galactic cosmic-ray propagation code.

astro-ph.GA↗