pith. sign in

arxiv: 2604.27991 · v1 · submitted 2026-04-30 · 🧮 math.NA · cs.NA

Noise-induced enhancement of regime lifetimes -- A data-driven approach using deterministic trajectories

Henry Schoeller , Robin Chemnitz , P\'eter Koltai , Maximilian Engel , Stephan Pfahl This is my paper

Pith reviewed 2026-05-07 05:31 UTC · model grok-4.3

classification 🧮 math.NA cs.NA
keywords stochastic inertiaregime lifetimesnoise-induced effectsMarkov chain approximationdeterministic trajectoriesatmospheric dynamicsdynamical systemsdata-driven method
0
0 comments X

The pith

Small noise increases the average lifetime of dynamical regimes compared to the deterministic case.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows that adding a small amount of noise to certain dynamical systems causes them to remain in specific regimes longer on average than they would without noise. This counterintuitive effect, called stochastic inertia, is demonstrated in both simple toy models and reduced models of atmospheric dynamics. To study it efficiently, the authors build a Markov chain using only a long trajectory from the noise-free system, which then approximates how small noise affects regime transitions. This method correctly predicts the enhancement in regime lifetimes for small noise levels without requiring simulations of the noisy dynamics itself. The finding matters because it challenges the common assumption that noise always disrupts persistence and offers a practical way to analyze it in complex systems like weather models.

Core claim

As long as the noise is sufficiently small, the noisy system stays in regimes of interest on average longer than its deterministic counterpart. This phenomenon, termed stochastic inertia, is observed through numerical simulations and can be predicted by constructing a Markov chain on points from a deterministic trajectory to approximate the transition behavior under noise.

What carries the argument

A Markov chain constructed from points along a sufficiently long deterministic trajectory, with transitions estimated to model the effect of small noise on regime exits.

If this is right

  • The average time spent in regimes increases under small noise, leading to longer persistence.
  • The Markov chain method allows prediction of stochastic inertia using only deterministic data.
  • This holds across phenomenological toy models and reduced atmospheric dynamics models.
  • Extensive numerical simulations confirm the effect for different small noise levels.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • This approach could be extended to analyze noise effects in higher-dimensional or real-world systems where full noisy simulations are computationally expensive.
  • Implications for improved modeling of persistent weather patterns or climate regimes where small perturbations might enhance stability.
  • If the Markov approximation holds more generally, it might reduce the need for ensemble simulations in studying stochastic influences on dynamics.

Load-bearing premise

The Markov chain built from the deterministic trajectory accurately approximates the transition probabilities of the noisy system when noise is small.

What would settle it

Directly simulate the noisy system for small noise levels and measure average regime lifetimes; if they do not exceed the deterministic lifetimes or match the Markov predictions, the claim of stochastic inertia would be falsified.

Figures

Figures reproduced from arXiv: 2604.27991 by Henry Schoeller, Maximilian Engel, P\'eter Koltai, Robin Chemnitz, Stephan Pfahl.

Figure 1
Figure 1. Figure 1: Example trajectories initialized in x0 = 5 × 10−2 for the one-dimensional toy model (15) for three different noise strengths σ, 103 steps of ∆t = 10−2 and a reinsertion rule with ϵ = 5 × 10−2 . Also shown are the boundaries defining the regime M and the system’s boundaries at whose crossing reinsertion takes place. tions. In view at source ↗
Figure 2
Figure 2. Figure 2: Mean regime lifetimes for each noise strength view at source ↗
Figure 3
Figure 3. Figure 3: Mean escape times for the toy system (15) initialized in various points over a range of noise strengths σ. The lines indicate analytic solutions according to Eqn. (19). The full markers indicate empirical mean escape times of Monte Carlo simulations of N ≈ 104 realisations with time steps of ∆t = 10−2 . The hollow markers indicate the expected mean escape times θ σ obtained using the Markov chain method an… view at source ↗
Figure 4
Figure 4. Figure 4: Normalized sum of affinities according to Eqn. ( view at source ↗
Figure 5
Figure 5. Figure 5: Mean regime life times for each noise strength view at source ↗
Figure 6
Figure 6. Figure 6: Two-dimensional projections of the deterministic CdV attractor based on a view at source ↗
Figure 7
Figure 7. Figure 7: Same as Fig view at source ↗
Figure 8
Figure 8. Figure 8: Assignment of 2 × 104 points in a principal component projection from a deterministic trajectory with ∆t = 10 that has been partitioned according to the k-means clustering in the embedding explained in the text based on σ = 10−2 . Also indicated are the “blocking” fixed point (black) and the “zonal” fixed point (red). tering on the embedded data to obtain a partition into clusters. Usually, the number of c… view at source ↗
Figure 9
Figure 9. Figure 9: Normalized sum of affinities according to Eqn. ( view at source ↗
Figure 10
Figure 10. Figure 10: For each point in the blocking regime M, its maximum relative increase in escape time (left y-axis) is shown against its deterministic escape time. A small random value is added to the x-location of each point to improve visibility. The points are coloured according to the noise strength σmax at which the maximal escape time is achieved. Data for points within one deterministic step until escape are not s… view at source ↗
Figure 11
Figure 11. Figure 11: Pointwise escape times from the center of the blocking regime of the CdV view at source ↗
Figure 12
Figure 12. Figure 12: Expected escape times θ σ across noise strength σ calculated according to Eqn. (12) for points of the deterministic trajectory that have just entered the blocking regime M. Only every tenth entry point is shown for better visibility. The curves are coloured according to their deterministic escape time (σ = 0). The location of these entry points are also shown in the inset along with all the other points i… view at source ↗
Figure 13
Figure 13. Figure 13: Mean regime life times for each noise strength view at source ↗
Figure 14
Figure 14. Figure 14: Same as Fig view at source ↗
Figure 15
Figure 15. Figure 15: Same as Fig view at source ↗
read the original abstract

We investigate the lifetime of dynamical regimes under the impact of noise motivated by low-dimensional models of the atmosphere. One may expect that the inclusion of noise tends to make the system leave prescribed regions of the state space faster. However, for relevant systems with complexities ranging from phenomenological toy models to reduced models of atmospheric dynamics, this intuition has proven misleading. As long as the noise is sufficiently small, the noisy system stays in regimes of interest on average longer than its deterministic counterpart, an effect we call ``stochastic inertia''. This phenomenon has been observed through extensive numerical simulations for different noise levels. We propose a numerical technique for testing the occurrence of stochastic inertia, constructing, for any fixed noise level, a Markov chain on the set of points given by a sufficiently long trajectory of the system without noise. The method is shown to correctly predict the presence of stochastic inertia in simple systems, and its utility is demonstrated on a paradigm model of atmospheric dynamics.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 3 minor

Summary. The paper claims that sufficiently small noise can increase the average lifetime of dynamical regimes relative to the deterministic flow (termed 'stochastic inertia'), contrary to the naive expectation that noise accelerates exits. The authors support the claim via extensive simulations on toy models and a reduced atmospheric dynamics model, and introduce a data-driven numerical method that builds a Markov chain whose states are the points of a long deterministic trajectory, with transitions defined by applying the noise kernel to the deterministic flow map. The method is reported to correctly detect the presence of stochastic inertia in simple systems and is demonstrated on the atmospheric paradigm model.

Significance. If the central claim and the associated numerical method hold under scrutiny, the work would be of moderate significance for stochastic dynamics and reduced-order modeling in atmospheric science. The data-driven construction that re-uses a single deterministic trajectory to probe noisy regime lifetimes is a practical strength, potentially reducing the cost of Monte-Carlo sampling for small noise. The observation challenges the common intuition that additive noise always shortens residence times and could inform ensemble forecasting or regime-detection algorithms, provided the approximation error is controlled.

major comments (2)
  1. [§3] §3 (Markov-chain construction): the discrete chain is defined by applying the noise kernel to the deterministic flow map evaluated at the trajectory points, yet no a-priori error bound or convergence statement is given that guarantees the chain's mean sojourn times converge to those of the continuous noisy process as the trajectory length N→∞ and noise amplitude σ→0. This is load-bearing for the small-noise claim, because when σ is small the true process remains in a thin tubular neighborhood of the orbit; the finite-point coarse-graining can distort exit rates near saddles or narrow passages between regimes.
  2. [§4.1–4.2] §4.1–4.2 (validation on toy models): while the abstract states that the method 'correctly predicts' stochastic inertia in simple systems, the manuscript does not report quantitative error metrics (e.g., relative difference in mean residence time between the Markov chain and direct Euler–Maruyama integration of the SDE) as a function of σ, N, or the number of points retained. Without such diagnostics it is impossible to assess whether the observed enhancement survives the discretization error that the skeptic correctly flags.
minor comments (3)
  1. [Introduction] The definition of 'regimes of interest' (how the sets are chosen or detected) is stated only informally in the introduction; an explicit algorithmic description or reference to a standard clustering/partitioning procedure would improve reproducibility.
  2. [Figures] Figure captions should state the exact noise amplitudes, integration time step, and trajectory length used for each panel so that the reader can judge the small-noise regime directly from the graphics.
  3. [§5] A short discussion of how the method scales with state-space dimension would be useful, given that the target application is reduced atmospheric models that may still be moderately high-dimensional.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for the careful reading and constructive comments on our manuscript. We address each major point below, indicating where revisions will be made to improve clarity and validation while remaining within the scope of the numerical, data-driven study.

read point-by-point responses

  1. Referee: §3 (Markov-chain construction): the discrete chain is defined by applying the noise kernel to the deterministic flow map evaluated at the trajectory points, yet no a-priori error bound or convergence statement is given that guarantees the chain's mean sojourn times converge to those of the continuous noisy process as the trajectory length N→∞ and noise amplitude σ→0. This is load-bearing for the small-noise claim, because when σ is small the true process remains in a thin tubular neighborhood of the orbit; the finite-point coarse-graining can distort exit rates near saddles or narrow passages between regimes.

    Authors: We acknowledge that the manuscript provides no rigorous a priori error bound or convergence theorem for the mean sojourn times of the Markov chain to those of the underlying SDE as N→∞ and σ→0. The construction is presented as a practical data-driven surrogate that reuses a single deterministic trajectory to estimate transition probabilities under small noise, motivated by the observation that noisy paths remain close to the deterministic orbit. We agree that finite sampling can introduce distortion near saddles or narrow passages. In the revised manuscript we will expand the discussion in §3 with a heuristic argument based on the density of the trajectory and standard results on stochastic perturbations of deterministic flows, together with a brief reference to related approximation literature. A complete rigorous convergence analysis, however, lies beyond the scope of this primarily numerical work. revision: partial

  2. Referee: §4.1–4.2 (validation on toy models): while the abstract states that the method 'correctly predicts' stochastic inertia in simple systems, the manuscript does not report quantitative error metrics (e.g., relative difference in mean residence time between the Markov chain and direct Euler–Maruyama integration of the SDE) as a function of σ, N, or the number of points retained. Without such diagnostics it is impossible to assess whether the observed enhancement survives the discretization error that the skeptic correctly flags.

    Authors: We accept the referee's observation that quantitative error diagnostics are absent. Although the manuscript shows qualitative agreement between the Markov-chain predictions and direct SDE integrations for the toy models, we did not include relative-error tables or plots versus σ and N. In the revised version we will add, in §4.1 and §4.2, a new figure (or table) reporting the relative difference in computed mean regime lifetimes between the Markov chain and Monte-Carlo Euler–Maruyama simulations, for several values of σ and N. This will allow readers to judge the magnitude of the discretization error and confirm that the reported stochastic-inertia effect persists under the observed numerical tolerances. revision: yes

standing simulated objections not resolved
  • A full rigorous a priori error bound and convergence proof for the Markov chain's sojourn times to the continuous SDE process (as N→∞ and σ→0) is not provided and cannot be supplied within the present numerical study.

Circularity Check

0 steps flagged

No significant circularity; data-driven Markov chain validated externally on toy models

full rationale

The paper's core contribution is an empirical observation of 'stochastic inertia' (longer regime lifetimes under small noise) obtained via direct numerical simulation of both deterministic and noisy systems on toy models and a reduced atmospheric dynamics model. The proposed Markov-chain construction on a long deterministic trajectory is introduced as a practical, data-driven approximation tool for testing the effect at fixed noise levels without repeated noisy integrations. Validation consists of showing that the chain reproduces the presence of the inertia effect in simple systems where independent noisy simulations provide ground truth; this comparison is external to the chain construction itself. No equation or step reduces a claimed prediction to a fitted parameter, self-definition, or load-bearing self-citation. The method is presented as an approximation whose accuracy is demonstrated numerically rather than derived tautologically from its inputs. The skeptic concern about possible distortion of exit rates for very small noise is a question of approximation quality, not circularity in the derivation.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on the validity of the Markov chain approximation for small noise and on the representativeness of deterministic trajectories for the systems studied. No explicit free parameters are mentioned in the abstract.

axioms (1)
  • domain assumption A sufficiently long deterministic trajectory densely samples the relevant state-space regions so that a Markov chain on those points approximates noisy transitions.
    Invoked when constructing the Markov chain from the deterministic trajectory to test stochastic inertia.
invented entities (1)
  • stochastic inertia no independent evidence
    purpose: Label for the observed noise-induced increase in average regime lifetimes.
    Newly introduced term for the counterintuitive numerical finding.

pith-pipeline@v0.9.0 · 5471 in / 1277 out tokens · 52091 ms · 2026-05-07T05:31:43.030452+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

3 extracted references · 3 canonical work pages

  1. [1]

    Koltai and D

    [KR18] P. Koltai and D. M. Renger. From large deviations to semidistances of trans- port and mixing: Coherence analysis for finite Lagrangian data.Journal of Nonlinear Science, 28(5):1915–1957,

    work page 1915

  2. [2]

    Kwasniok

    [Kwa14] F. Kwasniok. Enhanced regime predictability in atmospheric low-order mod- els due to stochastic forcing.Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 372(2018):20130286,

    work page 2018

  3. [3]

    Weisheimer, S

    [WCPV14] A. Weisheimer, S. Corti, T. Palmer, and F. Vitart. Addressing model error through atmospheric stochastic physical parametrizations: Impact on the coupled ECMWF seasonal forecasting system.Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 372(2018):20130290, June

    work page 2018