pith. sign in

arxiv: 2602.13847 · v4 · submitted 2026-02-14 · 🌊 nlin.CD · cond-mat.stat-mech· cs.LG· physics.ao-ph

Physics and causally constrained discrete-time neural models of turbulent dynamical systems

Fabrizio Falasca , Laure Zanna This is my paper

Pith reviewed 2026-05-15 22:14 UTC · model grok-4.3

classification 🌊 nlin.CD cond-mat.stat-mechcs.LGphysics.ao-ph
keywords neural networksturbulent flowscausal constraintsenergy preservationdiscrete-time modelsreduced-order modelingdynamical systemsphysics-informed learning
0
0 comments X

The pith

A finite-time flow map with energy-preserving nonlinearities and causal constraints lets neural models learn stable discrete-time representations of turbulent systems from data.

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

The paper introduces a framework that builds neural models of turbulent dynamical systems by first defining a finite-time flow map whose nonlinear terms strictly preserve energy and then adding causal constraints to block spurious cross-variable interactions. This combination produces models that remain stable over time while reproducing both the long-term statistics of the original system and its responses to added external forces of varying strength. The approach is demonstrated on the stochastic Charney-DeVore equations and a symmetry-broken Lorenz-96 system, showing that the learned models match key behaviors without introducing new instabilities. If the method works as claimed, it would enable reduced-order modeling directly from observational data for systems where full-resolution simulations remain computationally prohibitive.

Core claim

The authors formulate a discrete-time flow map whose nonlinear interactions are constructed to preserve energy exactly, then impose causal constraints on the neural architecture to eliminate non-physical couplings; the resulting models, trained on data, accurately reproduce stationary statistics and the response to both weak and strong external forcings in two canonical turbulent systems.

What carries the argument

Finite-time flow map with strict energy-preserving nonlinearities combined with causal constraints that suppress spurious interactions across degrees of freedom.

If this is right

  • The models capture stationary statistics of the underlying turbulent flow.
  • The models reproduce the system response to both small and large external forcings.
  • The models remain stable over long integration times without introducing new instabilities.
  • The framework supports reduced-order modeling directly from observational data for a range of turbulent dynamical systems.

Where Pith is reading between the lines

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

  • The same construction could be applied to observational records from real geophysical flows where governing equations are only partially known.
  • Causal constraints may become increasingly important as the number of resolved degrees of freedom grows, limiting unphysical information propagation in high-dimensional data.
  • Testing whether the energy-preservation property continues to hold under strong time-dependent forcings would clarify the limits of the discrete-time formulation.

Load-bearing premise

A finite-time flow map with strict energy-preserving nonlinearities and causal constraints can be learned from data while still reproducing the full range of turbulent behavior without new instabilities or loss of fidelity.

What would settle it

Train the model on one system and test whether it develops growing instabilities or mismatches the observed response amplitude when a large external forcing is applied to a second, independent turbulent system.

Figures

Figures reproduced from arXiv: 2602.13847 by Fabrizio Falasca, Laure Zanna.

Figure 1
Figure 1. Figure 1: FIG. 1: Causal constraining of the CdV model (Eq. (5)) [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3: Same as Figure 1, but for the L96 system [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4: Response of the L96 system to a large Gaussian [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 1
Figure 1. Figure 1: FIG. 1: Lorenz attractor simulated with the sequential mapping scheme [PITH_FULL_IMAGE:figures/full_fig_p016_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2: Stationary distributions (PDFs) of [PITH_FULL_IMAGE:figures/full_fig_p018_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3: Experiment with subsampling size 100. In the first and second rows, we show the stationary distributions [PITH_FULL_IMAGE:figures/full_fig_p019_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4: Same as Fig. 3 but after subsampling the original Charney-DeVore integration every 500 time steps [PITH_FULL_IMAGE:figures/full_fig_p020_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5: Time-dependent, responses in ensemble mean [PITH_FULL_IMAGE:figures/full_fig_p022_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6: Stationary statistics. First and second rows: stationary distributions (PDFs) of the variables [PITH_FULL_IMAGE:figures/full_fig_p025_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: FIG. 7: Time-dependent, responses in ensemble mean of variable [PITH_FULL_IMAGE:figures/full_fig_p026_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: FIG. 8: Time-dependent, responses in ensemble variance of variable [PITH_FULL_IMAGE:figures/full_fig_p027_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: FIG. 9: Perturbed statistics in the linear regime. Panel (a): Total MSE over time for the mean response, computed [PITH_FULL_IMAGE:figures/full_fig_p027_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: FIG. 10: Perturbed statistics in the nonlinear regime. Row (i) shows the repsonse in ensemble mean to a step [PITH_FULL_IMAGE:figures/full_fig_p028_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: FIG. 11: Perturbed statistics in the nonlinear regime. Row (i) shows the response in ensemble variance to a step [PITH_FULL_IMAGE:figures/full_fig_p029_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: FIG. 12: Perturbed statistics in the nonlinear regime. Panel (a): Total MSE over time for the ensemble mean [PITH_FULL_IMAGE:figures/full_fig_p029_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: FIG. 13: Stationary statistics. Stationary distributions (PDFs) of the variables [PITH_FULL_IMAGE:figures/full_fig_p030_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: FIG. 14: Autocorrelation functions (ACFs) of the variables [PITH_FULL_IMAGE:figures/full_fig_p031_14.png] view at source ↗
read the original abstract

We present a framework for constructing physics and causally constrained neural models of turbulent dynamical systems from data. We first formulate a finite-time flow map with strict energy-preserving nonlinearities for stable modeling of temporally discrete trajectories. We then impose causal constraints to suppress spurious interactions across degrees of freedom. The resulting neural models accurately capture stationary statistics and responses to both small and large external forcings. We demonstrate the framework on the stochastic Charney-DeVore equations and on a symmetry-broken Lorenz-96 system. The framework is broadly applicable to reduced-order modeling of turbulent dynamical systems from observational data.

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 / 2 minor

Summary. The paper presents a framework for constructing physics- and causally-constrained neural models of turbulent dynamical systems from data. It formulates a finite-time flow map incorporating strict energy-preserving nonlinearities for stable discrete-time trajectory modeling, then imposes causal constraints to eliminate spurious cross-degree-of-freedom interactions. The resulting models are claimed to accurately reproduce stationary statistics and responses to both small and large external forcings, with demonstrations on the stochastic Charney-DeVore equations and a symmetry-broken Lorenz-96 system. The approach is positioned as broadly applicable to reduced-order modeling of turbulent systems from observational data.

Significance. If the quantitative validation holds, the framework offers a structured way to embed physical invariants (energy conservation) and causality directly into learned discrete-time maps, which could improve long-term stability and physical fidelity in data-driven reduced-order models for chaotic and turbulent flows without post-hoc corrections.

major comments (2)
  1. [Abstract and Results] Abstract and results sections: the central claim that the models 'accurately capture stationary statistics and responses to both small and large external forcings' is load-bearing but unsupported by any reported error metrics, statistical tests (e.g., relative L2 norms on means/variances or Kolmogorov-Smirnov distances on PDFs), or ablation comparisons against unconstrained baselines in the provided description; explicit quantitative tables or figures with these values are required to substantiate the claim.
  2. [Methods] Methods on the finite-time flow map: the enforcement of 'strict energy-preserving nonlinearities' must be shown to hold exactly (e.g., via explicit parameterization or projection operator) rather than approximately, because any residual energy drift would undermine the stability argument for long-time turbulent statistics under forcing.
minor comments (2)
  1. [Methods] Notation for the causal constraint operator should be defined explicitly with an equation number rather than described only in prose to aid reproducibility.
  2. [Figures] Figure captions for the Lorenz-96 and Charney-DeVore demonstrations should specify the exact forcing amplitudes and integration lengths used to generate the reported statistics.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive report and positive recommendation for minor revision. We address each major comment below and will revise the manuscript accordingly to strengthen the quantitative support and clarify the exact enforcement of constraints.

read point-by-point responses

  1. Referee: [Abstract and Results] Abstract and results sections: the central claim that the models 'accurately capture stationary statistics and responses to both small and large external forcings' is load-bearing but unsupported by any reported error metrics, statistical tests (e.g., relative L2 norms on means/variances or Kolmogorov-Smirnov distances on PDFs), or ablation comparisons against unconstrained baselines in the provided description; explicit quantitative tables or figures with these values are required to substantiate the claim.

    Authors: We agree that explicit quantitative metrics strengthen the central claims. The manuscript already contains visual comparisons of stationary statistics and forced responses for both systems, but we will add a new table in the results section reporting relative L2 norms on means and variances, Kolmogorov-Smirnov distances between PDFs, and direct ablation comparisons against unconstrained neural baselines. These additions will be placed alongside the existing figures to substantiate the accuracy claims without altering the narrative. revision: yes

  2. Referee: [Methods] Methods on the finite-time flow map: the enforcement of 'strict energy-preserving nonlinearities' must be shown to hold exactly (e.g., via explicit parameterization or projection operator) rather than approximately, because any residual energy drift would undermine the stability argument for long-time turbulent statistics under forcing.

    Authors: We agree that exact enforcement is essential for the stability argument. The finite-time flow map is constructed via an explicit parameterization of the nonlinear term that guarantees energy preservation by design (using a skew-symmetric bilinear form that satisfies the required inner-product identity identically). We will expand the methods section with the explicit algebraic form of this parameterization and a short proof that the discrete energy is conserved exactly at every step, independent of the learned coefficients. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The derivation begins by formulating a finite-time flow map that incorporates strict energy-preserving nonlinearities by construction and then imposes causal constraints structurally to limit interactions. These choices are presented as modeling decisions rather than quantities fitted to data and relabeled as predictions. The subsequent claim that the resulting models reproduce stationary statistics and responses to forcings is an empirical outcome on the Charney-DeVore and Lorenz-96 systems, not a tautological restatement of the inputs. No self-citation is invoked as load-bearing justification for uniqueness or for smuggling an ansatz, and no step reduces an output to an input by definition. The chain therefore remains self-contained with independent content.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Framework rests on domain assumptions about energy preservation and causality being enforceable in neural architectures for turbulence; no free parameters or invented entities are explicitly introduced in the abstract.

axioms (2)
  • domain assumption Finite-time flow map can be formulated with strict energy-preserving nonlinearities for stable discrete trajectories
    Invoked to ensure stability in modeling temporally discrete trajectories of turbulent systems.
  • domain assumption Causal constraints can suppress spurious interactions across degrees of freedom while preserving essential dynamics
    Imposed to ensure physical realism in learned interactions from data.

pith-pipeline@v0.9.0 · 5397 in / 1262 out tokens · 34153 ms · 2026-05-15T22:14:46.862194+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

108 extracted references · 108 canonical work pages · 5 internal anchors

  1. [1]

    Castiglione, M

    P. Castiglione, M. Falcioni, A. Lesne, and A. Vulpi- ani,Chaos and Coarse Graining in Statistical Mechanics (Cambridge University Press, 2008)

    work page 2008

  2. [2]

    Majda,Introduction to Turbulent Dynamical Systems in Complex Systems(Springer Cham, 2016)

    A. Majda,Introduction to Turbulent Dynamical Systems in Complex Systems(Springer Cham, 2016)

    work page 2016

  3. [3]

    Izvekov and G

    S. Izvekov and G. Voth, Multiscale coarse grain- ing of liquid-state systems, J. Chem. Phys123, https://doi.org/10.1063/1.2038787 (2005)

    work page doi:10.1063/1.2038787 2005

  4. [4]

    Prinz, H

    J.-H. Prinz, H. Wu, M. Sarich, B. Keller, M. Senne, M. Held, J. Chodera, C. Sch¨ utte, and F. No´ e, Markov models of molecular kinetics: Generation and validation, J. Chem. Phys134, https://doi.org/10.1063/1.3565032 (2011)

    work page doi:10.1063/1.3565032 2011

  5. [5]

    Harrison, O

    L. Harrison, O. David, and K. Friston, Stochastic mod- els of neuronal dynamics, Phil. Trans. R. Soc. B360, 1075–1091 (2005)

    work page 2005

  6. [6]

    Rangan, L

    A. Rangan, L. Tao, G. Kovacic, and D. Cai, Multiscale modeling of the primary visual cortex, IEEE Eng Med Biol Mag.28, 19 (2009)

    work page 2009

  7. [7]

    Roberts, Normal form transforms separate slow and fast modes in stochastic dynamical systems, Physica A: Statistical Mechanics and its Applications387, 12 (2008)

    A. Roberts, Normal form transforms separate slow and fast modes in stochastic dynamical systems, Physica A: Statistical Mechanics and its Applications387, 12 (2008)

    work page 2008

  8. [8]

    A. J. Roberts, Resolving the multitude of microscale in- teractions accurately models stochastic partial differen- tial equations, LMS Journal of Computation and Math- ematics9, 193–221 (2006)

    work page 2006

  9. [9]

    Majda and D

    A. Majda and D. Qi, Strategies for Reduced- Order Models for Predicting the Statistical Re- sponses and Uncertainty Quantification in Complex Turbulent Dynamical Systems, SIAM REVIEW60, https://doi.org/10.1137/16M1104664 (2018). 7

    work page doi:10.1137/16m1104664 2018

  10. [10]

    Sanderse, Non-linearly stable reduced-order models for incompressible flow with energy-conserving finite vol- ume methods, Journal of Computational Physics421, 109736 (2020)

    B. Sanderse, Non-linearly stable reduced-order models for incompressible flow with energy-conserving finite vol- ume methods, Journal of Computational Physics421, 109736 (2020)

    work page 2020

  11. [11]

    Chen and H

    N. Chen and H. Liu, Minimum reduced-order models via causal inference, Nonlinear Dyn113, 11327–11351 (2025)

    work page 2025

  12. [12]

    Hasselmann, Stochastic climate models part i

    K. Hasselmann, Stochastic climate models part i. theory., Tellus28, 473 (1976)

    work page 1976

  13. [13]

    A. J. Majda, I. Timofeyev, and Vanden-Eijnden, Models for stochastic climate prediction, Proc. Natl. Acad. Sci. USA96, 14687–14691 (1999)

    work page 1999

  14. [14]

    Majda, C

    A. Majda, C. Franzke, and D. Crommelin, Normal forms for reduced stochastic climate models, Proc. Natl. Acad. Sci.10, 3649–3653 (2010)

    work page 2010

  15. [15]

    Ghil and V

    M. Ghil and V. Lucarini, The physics of climate vari- ability and climate change, Rev. Mod. Phys.92, 035002 (2020)

    work page 2020

  16. [16]

    Lucarini and M

    V. Lucarini and M. Chekroun, Theoretical tools for un- derstanding the climate crisis from Hasselmann’s pro- gramme and beyond, Nat Rev Phys (2023) (2023)

    work page 2023

  17. [17]

    Kitsios, L

    V. Kitsios, L. Cordier, and T. J. O’Kane, Three- dimensional proper orthogonal decomposition reduced- order model of the global atmospheric climate, Inter- national Journal of Heat and Fluid Flow119, 110253 (2026)

    work page 2026

  18. [18]

    Kitsios, L

    V. Kitsios, L. Cordier, and T. J. O’Kane, Proper orthog- onal decomposition reduced-order model of the global oceans, Theor. Comput. Fluid Dyn.38, 707–727 (2024)

    work page 2024

  19. [19]

    Penland, Random Forcing and Forecasting Using Prin- cipal Oscillation Pattern Analysis, Monthly Weather Re- view117, 2165 (1989)

    C. Penland, Random Forcing and Forecasting Using Prin- cipal Oscillation Pattern Analysis, Monthly Weather Re- view117, 2165 (1989)

    work page 1989

  20. [20]

    Horenko, Finite element approach to clustering of mul- tidimensional time series, SIAM J

    I. Horenko, Finite element approach to clustering of mul- tidimensional time series, SIAM J. Sci. Comp.32, 68–83 (2010)

    work page 2010

  21. [21]

    Pavliotis and A

    G. Pavliotis and A. Stuart, Parameter estimation for mul- tiscale diffusions, J. Stat. Phys.127, 741–781 (2007)

    work page 2007

  22. [22]

    Majda and J

    A. Majda and J. Harlim, Physics constrained nonlinear regression models for time series, NONLINEARITY26, 201–217 (2013)

    work page 2013

  23. [23]

    Kravtsov, D

    S. Kravtsov, D. Kondrashov, and M. Ghil, Multilevel Regression Modeling of Nonlinear Processes: Derivation and Applications to Climatic Variability , Journal of Cli- mate18, 4404–4424 (2005)

    work page 2005

  24. [24]

    Kondrashov, M

    D. Kondrashov, M. Chekroun, and M. Ghil, Data-driven non-markovian closure models, Physica D: Nonlinear Phenomena297, 33 (2015)

    work page 2015

  25. [25]

    J. L. Callaham, J.-C. Loiseau, G. Rigas, and S. L. Brun- ton, Nonlinear stochastic modelling with langevin regres- sion, Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences477, 20210092 (2021), https://royalsocietypublishing.org/rspa/article- pdf/doi/10.1098/rspa.2021.0092/725688/rspa.2021.0092.pdf

    work page doi:10.1098/rspa.2021.0092/725688/rspa.2021.0092.pdf 2021

  26. [26]

    Souza, Representing turbulent statistics with parti- tions of state space

    A. Souza, Representing turbulent statistics with parti- tions of state space. part 1. theory and methodology., J. Fluid Mech.997(2024)

    work page 2024

  27. [27]

    Chen and A

    N. Chen and A. J. Majda, Conditional gaussian systems for multiscale nonlinear stochastic systems: Prediction, state estimation and uncertainty quantification, Entropy 20, 10.3390/e20070509 (2018)

    work page doi:10.3390/e20070509 2018

  28. [28]

    Giorgini, Score-Based Modeling of Effective Langevin Dynamics, Arxiv https://doi.org/10.48550/arXiv.2505.01895 (2025)

    L. Giorgini, Score-Based Modeling of Effective Langevin Dynamics, Arxiv https://doi.org/10.48550/arXiv.2505.01895 (2025)

    work page doi:10.48550/arxiv.2505.01895 2025

  29. [29]

    Azencott, A

    R. Azencott, A. Beri, and I. Timofeyev, Adaptive sub- sampling for parametric estimation of gaussian diffusions, J Stat Phys139, 1066–1089 (2010)

    work page 2010

  30. [30]

    Ferretti, V

    F. Ferretti, V. Chard` es, T. Mora, A. M. Walczak, and I. Giardina, Building general langevin models from dis- crete datasets, Phys. Rev. X10, 031018 (2020)

    work page 2020

  31. [31]

    Falasca, Probing forced responses and causality in data-driven climate emulators: Conceptual limitations and the role of reduced-order models, Phys

    F. Falasca, Probing forced responses and causality in data-driven climate emulators: Conceptual limitations and the role of reduced-order models, Phys. Rev. Res. 7, 043314 (2025)

    work page 2025

  32. [32]

    Geneva and N

    N. Geneva and N. Zabaras, Modeling the dynamics of pde systems with physics-constrained deep auto-regressive networks, Journal of Computational Physics403, 109056 (2020)

    work page 2020

  33. [33]

    Milano and P

    M. Milano and P. Koumoutsakos, Neural network mod- eling for near wall turbulent flow, Journal of Computa- tional Physics182, 1 (2002)

    work page 2002

  34. [34]

    Vlachas, G

    P. Vlachas, G. Arampatzis, C. Uhler, and et al., Mul- tiscale simulations of complex systems by learning their effective dynamics, Nat Mach Intell4, 359–366 (2022)

    work page 2022

  35. [35]

    FourCastNet: A Global Data-driven High-resolution Weather Model using Adaptive Fourier Neural Operators

    J. Pathak, S. Subramanian, P. Harrington, S. Raja, A. Chattopadhyay, M. Mardani, T. Kurth, D. Hall, Z. Li, K. Azizzadenesheli, P. Hassanzadeh, K. Kashinath, and A. Anandkumar, FourCastNet: A Global Data-driven High-resolution Weather Model using Adaptive Fourier Neural Operators, arXiv e-prints , arXiv:2202.11214 (2022), arXiv:2202.11214 [physics.ao-ph]

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  36. [36]

    arXiv, ://arxiv.org/abs/2507.12144, arXiv:2507.12144 [cs], doi:10.48550/arXiv.2507.12144

    B. Bonev and et al., FourCastNet 3: A geomet- ric approach to probabilistic machine-learning weather forecasting at scale, arXiv e-prints 10.48550/arXiv.2507.12144 (2025)

    work page doi:10.48550/arxiv.2507.12144 2025

  37. [37]

    ClimateBench v1.0: A bench- mark for data-driven climate projections

    O. Watt-Meyer, G. Dresdner, J. McGibbon, S. K. Clark, B. Henn, J. Duncan, N. D. Brenowitz, K. Kashinath, M. S. Pritchard, B. Bonev, M. E. Peters, and C. S. Bretherton, ACE: A fast, skillful learned global atmo- spheric model for climate prediction, arXiv e-prints , arXiv:2310.02074 (2023), arXiv:2310.02074 [physics.ao- ph]

    work page arXiv 2023

  38. [38]

    Watt-Meyer, B

    O. Watt-Meyer, B. Henn, J. McGibbon, S. K. Clark, A. Kwa, W. A. Perkins, and C. S. Bretherton, ACE2: Accurately learning subseasonal to decadal at- mospheric variability and forced responses, arXiv e-prints https://arxiv.org/abs/2411.11268 (2024)

    work page arXiv 2024

  39. [39]

    H. Guan, T. Arcomano, A. Chattopaday, and R. Maulik, LUCIE: A Lightweight Uncoupled ClI- mate Emulator with long-term stability and physi- cal consistency for O(1000)-member ensembles, arXiv https://doi.org/10.48550/arXiv.2405.16297 (2025)

    work page doi:10.48550/arxiv.2405.16297 2025

  40. [40]

    Learning non-Gaussian spatial distributions via Bayesian transport maps with parametric shrink- age

    W. Chapman, J. Schreck, Y. Sha, D. Gagne II, D. Kim- para, L. Zanna, K. Mayer, and J. Berner, CAMulator: Fast Emulation of the Community Atmosphere Model, Arxiv https://doi.org/10.48550/arXiv.2504.06007 (2025)

    work page doi:10.48550/arxiv.2504.06007 2025

  41. [41]

    Dheeshjith, A

    S. Dheeshjith, A. Subel, A. Adcroft, J. Busecke, C. Fernandez-Granda, S. Gupta, and L. Zanna, Samu- dra: An AI global ocean emulator for climate., Geophys- ical Research Letters52, e2024GL114318 (2025)

    work page 2025

  42. [42]

    Physics-aware generative models for turbulent fluid flows through energy-consistent stochastic interpolants.arXiv preprint arXiv:2504.05852, 2025

    N. M¨ ucke and B. Sanderse, Physics-aware gener- ative models for turbulent fluid flows through energy-consistent stochastic interpolants, Arxiv https://doi.org/10.48550/arXiv.2504.05852 (2025)

    work page doi:10.48550/arxiv.2504.05852 2025

  43. [43]

    Pedersen, L

    C. Pedersen, L. Zanna, and J. Bruna, Thermalizer: Sta- ble autoregressive neural emulation of spatiotemporal chaos, Arxiv https://doi.org/10.48550/arXiv.2503.18731 8 (2025)

    work page doi:10.48550/arxiv.2503.18731 2025

  44. [44]

    X. Tian, D. Holdaway, and D. Kleist, Evaluating machine learning weather models for data assimilation: Funda- mental limitations in tangent linear and adjoint proper- ties, Geophysical Research Letters53, e2025GL119402 (2026)

    work page 2026

  45. [45]

    Van Loon, M

    S. Van Loon, M. Rugenstein, and E. A. Barnes, Reanalysis-based Global Radiative Response to Sea Surface Temperature Patterns: Evaluating the Ai2 Climate Emulator, Geophysical Research Letters52, e2025GL115432 (2025)

    work page 2025

  46. [46]

    arXiv preprint arXiv:2510.02415 , year=

    B. Zhang and T. Merlis, The Equilibrium Re- sponse of Atmospheric Machine-Learning Models to Uniform Sea Surface Temperature Warming, Arxiv https://doi.org/10.48550/arXiv.2510.02415 (2026)

    work page internal anchor Pith review doi:10.48550/arxiv.2510.02415 2026

  47. [47]

    Ismael, Reflections on the asymmetry of causation, Interface Focus12, 20220081 (2023)

    J. Ismael, Reflections on the asymmetry of causation, Interface Focus12, 20220081 (2023)

    work page 2023

  48. [48]

    Aurell and G

    E. Aurell and G. Del Ferraro, Causal analysis, correlation- response, and dynamic cavity, J. of Phys.: Conference Series699, 012002 (2016)

    work page 2016

  49. [49]

    V. Lucarini, Revising and Extending the Linear Response Theory for Statistical Mechanical Systems: Evaluating Observables as Predictors and Predictands, J Stat Phys 173, 1698–1721 (2018)

    work page 2018

  50. [50]

    Lucente, M

    D. Lucente, M. Baldovin, F. Cecconi, M. Cencini, N. Coc- ciaglia, A. Puglisi, M. Viale, and A. Vulpiani, Conceptual and practical approaches for investigating irreversible processes, New Journal of Physics27, 041201 (2025)

    work page 2025

  51. [51]

    Paus, Inferring causality in brain images: a perturba- tion approach, Philos Trans R Soc Lond B Biol Sci.29, 041201 (2005)

    T. Paus, Inferring causality in brain images: a perturba- tion approach, Philos Trans R Soc Lond B Biol Sci.29, 041201 (2005)

    work page 2005

  52. [52]

    Bloch-Johnson, M

    J. Bloch-Johnson, M. A. A. Rugenstein, M. J. Alessi, C. Proistosescu, M. Zhao, B. Zhang, A. I. L. Williams, J. M. Gregory, J. Cole, Y. Dong, M. L. Duffy, S. M. Kang, and C. Zhou, The green’s function model intercomparison project (gfmip) protocol, Journal of Advances in Model- ing Earth Systems16, e2023MS003700 (2024)

    work page 2024

  53. [53]

    R. H. Kraichnan, Classical fluctuation-relaxation theo- rem, Phys. Rev.113, 1181 (1959)

    work page 1959

  54. [54]

    A. J. Majda, R. V. Abramov, and M. J. Grote,Infor- mation Theory and Stochastics for Multiscale Nonlinear Systems(CRM Monograph Series, American Mathemat- ical Society, 2005)

    work page 2005

  55. [55]

    Lucarini, M

    V. Lucarini, M. S. Guti´ errez, J. Moroney, and N. Zagli, A general framework for linking free and forced fluctuations via koopmanism, Chaos, Solitons & Fractals202, 117540 (2026)

    work page 2026

  56. [56]

    Baldovin, F

    M. Baldovin, F. Cecconi, and A. Vulpiani, Understand- ing causation via correlations and linear response theory, Physical Review Research2, 043436 (2020)

    work page 2020

  57. [57]

    Cecconi, G

    F. Cecconi, G. Costantini, C. Guardiani, M. Baldovin, and A. Vulpiani, Correlation, response and entropy ap- proaches to allosteric behaviors: a critical comparison on the ubiquitin case, Physical Biology20, 056002 (2023)

    work page 2023

  58. [58]

    Charney and J

    J. Charney and J. DeVore, Multiple flow equilibria in the atmosphere and blocking, Journal of the atmospheric sciences36, 1205 (1979)

    work page 1979

  59. [59]

    Lorenz, Predictability: a problem partly solved, Sem- inar on Predictability, 4-8 September 19951, 1 (1995)

    E. Lorenz, Predictability: a problem partly solved, Sem- inar on Predictability, 4-8 September 19951, 1 (1995)

    work page 1995

  60. [60]

    Lorenz, Deterministic nonperiodic flow, J

    E. Lorenz, Deterministic nonperiodic flow, J. Atmo- spheric Sci.20, 130–141 (1963)

    work page 1963

  61. [61]

    Kwasniok, Reduced atmospheric models using dynam- ically motivated basis functions, Journal of the Atmo- spheric Sciences64, 3452–3474 (2007)

    F. Kwasniok, Reduced atmospheric models using dynam- ically motivated basis functions, Journal of the Atmo- spheric Sciences64, 3452–3474 (2007)

    work page 2007

  62. [62]

    van Gastelen, W

    T. van Gastelen, W. Edeling, and B. Sanderse, Energy- conserving neural network for turbulence closure mod- eling, Journal of Computational Physics508, 113003 (2024)

    work page 2024

  63. [63]

    F. Roth, D. Klein, M. Kannapinn, J. Peters, and O. Weeger, Stable Port-Hamiltonian Neural Net- works, Arxiv https://doi.org/10.48550/arXiv.2502.02480 (2025)

    work page doi:10.48550/arxiv.2502.02480 2025

  64. [64]

    Bou-Rabee and H

    N. Bou-Rabee and H. Owhadi, Long-run accuracy of variational integrators in the stochastic context, SIAM Journal on Numerical Analysis48, 278 (2010), https://doi.org/10.1137/090758842

    work page doi:10.1137/090758842 2010

  65. [65]

    Marconi, A

    U. Marconi, A. Puglisi, L. Rondoni, and A. Vulpiani, Fluctuation-dissipation: Response theory in statistical physics, Phys. Rep.461(2008)

    work page 2008

  66. [66]

    L. T. Giorgini, K. Deck, T. Bischoff, and A. Souza, Re- sponse theory via generative score modeling, Phys. Rev. Lett.133, 267302 (2024)

    work page 2024

  67. [67]

    Giorgini, F

    L. Giorgini, F. Falasca, and A. Souza, Predicting Forced Responses of Probability Distributions via the Fluctuation-Dissipation Theorem and Generative Mod- eling, Arxiv https://doi.org/10.48550/arXiv.2504.13333 (2025)

    work page doi:10.48550/arxiv.2504.13333 2025

  68. [68]

    Y. Song, J. Sohl-Dickstein, D. P. Kingma, A. Ku- mar, S. Ermon, and B. Poole, Score-based gen- erative modeling through stochastic differen- tial equations, arXiv preprint arXiv:2011.13456 https://doi.org/10.48550/arXiv.2011.13456 (2021)

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2011.13456 2011

  69. [69]

    Y. Song, S. Garg, J. Shi, and S. Ermon, Sliced score matching: A scalable approach to density and score estimation, arXiv preprint arXiv.1905.07088 https://doi.org/10.48550/arXiv.1905.07088 (2021)

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1905.07088 1905

  70. [70]

    Souza, S

    A. Souza, S. Silvestri, K. Deck, T. Bischoff, R. Ferrari, and G. Flierl, Surface to seafloor: A generative ai frame- work for decoding the ocean interior state, arXiv preprint https://doi.org/10.48550/arXiv.2504.15308 (2025)

    work page doi:10.48550/arxiv.2504.15308 2025

  71. [71]

    Price, A

    I. Price, A. Sanchez-Gonzalez, F. Alet, and et al., Prob- abilistic weather forecasting with machine learning, Na- ture637, 84–90 (2025)

    work page 2025

  72. [72]

    N. Brenowitz and et al., Climate in a bottle: To- wards a generative foundation model for the kilometer- scale global atmosphere, arXiv preprint arXiv.2505.06474 https://doi.org/10.48550/arXiv.2505.06474 (2025)

    work page doi:10.48550/arxiv.2505.06474 2025

  73. [73]

    Lobbe, D

    A. Lobbe, D. Crisan, and O. Lang, Bayesian in- ference for geophysical fluid dynamics using gen- erative models, Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences383, 20240321 (2025), https://royalsocietypublishing.org/rsta/article- pdf/doi/10.1098/rsta.2024.0321/2819094/rsta.2024.0321.pdf

    work page doi:10.1098/rsta.2024.0321/2819094/rsta.2024.0321.pdf 2025

  74. [74]

    C. E. Leith, Climate response and fluctuation dissipa- tion, Journal of The Atmospheric Science32, 2022–2026 (1975)

    work page 2022

  75. [75]

    Hyv¨ arinen, Estimation of non-normalized statistical models by score matching, Journal of Machine Learning Research6, 695–709 (2005)

    A. Hyv¨ arinen, Estimation of non-normalized statistical models by score matching, Journal of Machine Learning Research6, 695–709 (2005)

    work page 2005

  76. [76]

    Pedregosa, G

    F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cour- napeau, M. Brucher, M. Perrot, and E. Duchesnay, Scikit-learn: Machine learning in Python, Journal of Ma- chine Learning Research12, 2825 (2011)

    work page 2011

  77. [77]

    Cvitanovi´ c, R

    P. Cvitanovi´ c, R. Artuso, R. Mainieri, G. Tan- ner, and G. Vattay,Chaos: Classical and Quan- 9 tum(ChaosBook.org, Niels Bohr Institute, Copenhagen, 2016)

    work page 2016

  78. [78]

    Miranda and E

    R. Miranda and E. Stone, The proto-lorenz system, Physics Letters A178, 105 (1993)

    work page 1993

  79. [79]

    De Swart, Low-order spectral models of the atmo- spheric circulation: A survey, Acta Applicandae Mathe- matica11, 49 (1988)

    H. De Swart, Low-order spectral models of the atmo- spheric circulation: A survey, Acta Applicandae Mathe- matica11, 49 (1988)

    work page 1988

  80. [80]

    Crommelin and A

    D. Crommelin and A. Majda, Strategies for Model Re- duction: Comparing Different Optimal Bases, Journal of the Atmospheric Sciences61, 2206–2217 (2004)

    work page 2004

Showing first 80 references.