pith. sign in

arxiv: 2603.05531 · v2 · submitted 2026-03-02 · 🌊 nlin.CD

Adjoint-based optimization with quantized local reduced-order models for spatiotemporally chaotic systems

Defne E. Ozan , Antonio Colanera , Luca Magri This is my paper

Pith reviewed 2026-05-15 17:40 UTC · model grok-4.3

classification 🌊 nlin.CD
keywords reduced-order modelingadjoint optimizationchaotic systemsKuramoto-Sivashinsky equationdata assimilationquantized modelsvariational data assimilationspatiotemporal chaos
0
0 comments X

The pith

Quantized local reduced-order models paired with adjoint optimization reconstruct chaotic trajectories up to 0.25 Lyapunov times at 3.5 times full-model speed.

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

The paper develops a reduced-order approach that merges quantized local models with adjoint-based optimization to handle optimization tasks in systems with spatiotemporal chaos. This is demonstrated on a variational data assimilation problem for the Kuramoto-Sivashinsky equation, where the method recovers the complete trajectory from full-state data available only at the final time. The reconstruction succeeds for assimilation windows reaching 0.25 Lyapunov times. The same procedure runs 3.5 times faster than the underlying high-fidelity model. The results point to a practical route for making adjoint computations feasible in larger chaotic flows.

Core claim

We combine quantized local reduced-order models with adjoint-based optimization to perform efficient optimization in spatiotemporally chaotic systems. When applied to data assimilation in the Kuramoto-Sivashinsky equation, this method reconstructs the full trajectory for up to 0.25 Lyapunov times using final-time full-state measurements and achieves a 3.5 times speedup over the full-order model.

What carries the argument

Quantized local reduced-order models embedded inside the adjoint-based optimization loop

If this is right

  • The method reconstructs the full trajectory of the Kuramoto-Sivashinsky equation for up to 0.25 Lyapunov times from final-time full-state measurements.
  • The algorithm delivers a 3.5 times computational speedup relative to the full-order model.
  • The combination enables use of reduced-order models inside adjoint optimization loops for spatiotemporally chaotic dynamics.
  • The approach opens reduced-order modeling routes for optimization problems governed by chaotic partial differential equations.

Where Pith is reading between the lines

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

  • The same coupling of quantized local models and adjoints could be tested on other chaotic systems such as turbulent fluid flows.
  • The method might extend to cases with only partial state observations rather than full-state final measurements.
  • Further tuning of quantization levels or local model selection could shift the accuracy-speed tradeoff for longer horizons.

Load-bearing premise

The quantized local reduced-order models remain accurate enough when used repeatedly inside the adjoint optimization loop for the chaotic dynamics.

What would settle it

Applying the algorithm to the Kuramoto-Sivashinsky equation on assimilation windows longer than 0.25 Lyapunov times and finding either large trajectory reconstruction errors or disappearance of the reported speedup.

Figures

Figures reproduced from arXiv: 2603.05531 by Antonio Colanera, Defne E. Ozan, Luca Magri.

Figure 1
Figure 1. Figure 1: Overview of adjoint-based optimization with quantized local reduced-order models (ql-ROMs): (1) data collection; (2) phase-space quantization and local basis/model construction; (3) ql-ROM adjoint integrated backward in time with jump/coordinate-transformation at cluster switches; (4) variational data assimilation using the ql-ROM direct–adjoint loop to update the initial condition. where tm = m∆t with ∆t … view at source ↗
Figure 2
Figure 2. Figure 2: Sensitivity of the switching time and accuracy of the adjoint sensitivity com￾puted using the quantized local reduced model in comparison to the full-order model. (a) Estimated PDF of the normalized Ts in the ql-ROM, by perturbing the initial con￾dition with random noise with standard deviation 10−3 . (b) the relative ℓ2-norm of the error between these two gradient vectors, denoted by ϵ, and (c) the cosine… view at source ↗
Figure 3
Figure 3. Figure 3: Variational data assimilation on the Kuramoto-Sivashinsky equation using the adjoint of the quantized local reduced-order model. The method reconstructs the true trajectory from measurements at the final time T = 0.25 LT following the convergence of the optimization objective, J . seconds for 100 iterations, overall resulting in ×3.5 speed-up (results obtained on Intel Xeon(R) Gold 5218R CPU at 2.10 GHz × … view at source ↗
read the original abstract

We introduce a computationally efficient and accurate reduced order modelling approach for the optimization of spatiotemporally chaotic systems. The proposed method combines quantized local reduced order modelling with adjoint-based optimization. We employ the methodology in a variational data assimilation problem for the chaotic Kuramoto-Sivashinsky equation and show that it successfully reconstructs the full trajectory for up to 0.25 Lyapunov times given full state measurements at the final time. The proposed algorithm provides 3.5 times speed-up when compared to the full-order model. The proposed method opens up new possibilities for the reduced order modelling of spatiotemporally chaotic systems.

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 introduces a method that combines quantized local reduced-order models (ROMs) with adjoint-based optimization to enable efficient optimization and data assimilation for spatiotemporally chaotic systems. Applied to a variational data assimilation problem for the Kuramoto-Sivashinsky equation with full-state measurements at the final time, the approach reconstructs the full trajectory for up to 0.25 Lyapunov times and achieves a 3.5x speedup relative to the full-order model.

Significance. If the quantized local ROMs preserve sufficient accuracy in the adjoint gradients for the chaotic dynamics, the method could provide a practical route to reduced-order optimization in high-dimensional chaotic systems where full-order adjoint computations are prohibitive. The reported speedup and short-horizon reconstruction success are concrete empirical results that, if supported by quantitative validation, would strengthen the case for quantized local ROMs in data assimilation and control applications.

major comments (2)
  1. [Results section (data assimilation experiments)] The central claim that the quantized local ROMs remain sufficiently accurate when coupled to the adjoint-based optimization loop rests on reconstruction success at 0.25 Lyapunov times, but no direct comparison of the ROM adjoint gradients to the full-order adjoint gradients is provided, nor are error bounds or sensitivity analyses with respect to quantization parameters reported. This leaves open whether modest local approximation errors amplify under backward integration in the presence of positive Lyapunov exponents.
  2. [Numerical results and timing benchmarks] The reported 3.5x speedup is given relative to the full-order model, but the manuscript does not specify the wall-clock or flop counts for the ROM construction, quantization overhead, and adjoint solves, nor does it include baseline comparisons against other reduced-order or surrogate adjoint methods. Without these, the practical advantage for longer assimilation windows cannot be assessed.
minor comments (2)
  1. [Method section] Clarify the precise definition of the quantization operator and the locality criterion used to build the local ROMs; the notation for the quantized state and the reduced basis should be introduced with an explicit equation.
  2. [Abstract and Results] The abstract states 'successful reconstruction' without accompanying quantitative metrics (e.g., L2 trajectory error, Lyapunov-time normalized error, or convergence of the optimization objective); these should be added to the results figures or tables.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed review. We address each major comment point by point below, indicating revisions where appropriate.

read point-by-point responses

  1. Referee: [Results section (data assimilation experiments)] The central claim that the quantized local ROMs remain sufficiently accurate when coupled to the adjoint-based optimization loop rests on reconstruction success at 0.25 Lyapunov times, but no direct comparison of the ROM adjoint gradients to the full-order adjoint gradients is provided, nor are error bounds or sensitivity analyses with respect to quantization parameters reported. This leaves open whether modest local approximation errors amplify under backward integration in the presence of positive Lyapunov exponents.

    Authors: We agree that a direct comparison of ROM adjoint gradients to full-order gradients, along with error bounds and sensitivity analyses, would strengthen the validation. The trajectory reconstruction success up to 0.25 Lyapunov times provides indirect evidence that gradient inaccuracies remain tolerable for the optimization to recover the correct solution. To address the concern explicitly, we will add quantitative gradient error comparisons for representative cases and a sensitivity study with respect to quantization parameters in the revised manuscript. revision: yes

  2. Referee: [Numerical results and timing benchmarks] The reported 3.5x speedup is given relative to the full-order model, but the manuscript does not specify the wall-clock or flop counts for the ROM construction, quantization overhead, and adjoint solves, nor does it include baseline comparisons against other reduced-order or surrogate adjoint methods. Without these, the practical advantage for longer assimilation windows cannot be assessed.

    Authors: The reported 3.5x speedup measures the total wall-clock time of the complete optimization loop (including ROM construction and quantization) versus the full-order model. We will revise the manuscript to include explicit breakdowns of wall-clock times and flop counts for each component. Direct comparisons to other surrogate adjoint methods are not included, as our focus was on feasibility versus the full-order baseline; adding such baselines would require new experiments outside the current scope. We will add a discussion of this limitation. revision: partial

Circularity Check

0 steps flagged

No circularity: empirical validation of combined ROM-adjoint method on KS equation

full rationale

The paper proposes a hybrid method (quantized local reduced-order models coupled to adjoint-based optimization) and validates it through direct numerical experiments on the Kuramoto-Sivashinsky equation. Reported outcomes (trajectory reconstruction up to 0.25 Lyapunov times and 3.5x speedup) are measured results of running the algorithm, not quantities defined or fitted to be identical to the inputs. No derivation step reduces by construction to a self-definition, a renamed fit, or an unverified self-citation chain; the central claims remain independent of the method's own outputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim depends on standard assumptions in reduced-order modeling and adjoint methods for chaotic PDEs; no free parameters or invented entities are specified in the abstract.

axioms (1)
  • domain assumption The Kuramoto-Sivashinsky equation serves as a representative model for spatiotemporally chaotic systems.
    Used as the testbed for the data assimilation problem.

pith-pipeline@v0.9.0 · 5405 in / 1254 out tokens · 55312 ms · 2026-05-15T17:40:10.703372+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

15 extracted references · 15 canonical work pages

  1. [1]

    Applied Mathematical Sciences, Springer London (2008)

    di Bernardo, M., Budd, C.J., Champneys, A.R., Kowalczyk, P.: Piecewise-smooth Dynamical Systems: Theory and Applications. Applied Mathematical Sciences, Springer London (2008). https://doi.org/10.1007/978-1-84628-708-4

    work page doi:10.1007/978-1-84628-708-4 2008

  2. [2]

    ComputerMethodsinAppliedMechanicsandEngineering447,118393(Dec2025)

    Colanera,A.,Magri,L.:Quantizedlocalreduced-ordermodelingintime(ql-ROM). ComputerMethodsinAppliedMechanicsandEngineering447,118393(Dec2025). https://doi.org/10.1016/j.cma.2025.118393

    work page doi:10.1016/j.cma.2025.118393 2025

  3. [3]

    https://doi.org/10.48550/ARXIV.2511.04586

    Colanera, A., Magri, L.: Towards extreme event prediction of tur- bulent flows with quantized local reduced-order models (2025). https://doi.org/10.48550/ARXIV.2511.04586

    work page doi:10.48550/arxiv.2511.04586 2025

  4. [4]

    Guibas, J., Mardani, M., Li, Z., Tao, A., Anandkumar, A., and Catanzaro, B

    Cox, S.M., Matthews, P.C.: Exponential Time Differencing for Stiff Sys- tems. Journal of Computational Physics176(2), 430–455 (Mar 2002). https://doi.org/10.1006/jcph.2002.6995

    work page doi:10.1006/jcph.2002.6995 2002

  5. [5]

    Flow, Turbulence and Combustion65(3/4), 393–415 (2000)

    Giles, M.B., Pierce, N.A.: An Introduction to the Adjoint Approach to Design. Flow, Turbulence and Combustion65(3/4), 393–415 (2000). https://doi.org/10.1023/A:1011430410075

    work page doi:10.1023/a:1011430410075 2000

  6. [6]

    Society for Industrial and Applied Mathematics (Jan 2002)

    Gunzburger, M.D.: Perspectives in Flow Control and Optimiza- tion. Society for Industrial and Applied Mathematics (Jan 2002). https://doi.org/10.1137/1.9780898718720

    work page doi:10.1137/1.9780898718720 2002

  7. [7]

    Cambridge University Press, 1 edn

    Holmes, P., Lumley, J.L., Berkooz, G.: Turbulence, Coherent Structures, Dy- namical Systems and Symmetry. Cambridge University Press, 1 edn. (Oct 1996). https://doi.org/10.1017/CBO9780511622700

    work page doi:10.1017/cbo9780511622700 1996

  8. [8]

    URL http: //dx.doi.org/10.1109/JPROC.2024.3440211

    Kong, N.J., Payne, J.J., Zhu, J., Johnson, A.M.: Saltation matrices: The essential tool for linearizing hybrid dynamical systems. arXiv (2024). https://doi.org/10.1109/JPROC.2024.3440211

    work page doi:10.1109/jproc.2024.3440211 2024

  9. [9]

    Progress of Theoreti- cal Physics Supplement64, 346–367 (1978)

    Kuramoto, Y.: Diffusion-Induced Chaos in Reaction Systems. Progress of Theoreti- cal Physics Supplement64, 346–367 (1978). https://doi.org/10.1143/PTPS.64.346

    work page doi:10.1143/ptps.64.346 1978

  10. [10]

    Applied Mechan- ics Reviews71(2), 020801 (Mar 2019)

    Magri, L.: Adjoint Methods as Design Tools in Thermoacoustics. Applied Mechan- ics Reviews71(2), 020801 (Mar 2019). https://doi.org/10.1115/1.4042821

    work page doi:10.1115/1.4042821 2019

  11. [11]

    Computer Methods in Applied Mechanics and En- gineering418, 116502 (Jan 2024)

    Nóvoa, A., Racca, A., Magri, L.: Inferring unknown unknowns: Regularized bias- aware ensemble kalman filter. Computer Methods in Applied Mechanics and En- gineering418, 116502 (Jan 2024). https://doi.org/10.1016/j.cma.2023.116502

    work page doi:10.1016/j.cma.2023.116502 2024

  12. [12]

    Physical Review Fluids9(10), 103902 (Oct 2024)

    Ozan, D.E., Magri, L.: Data-driven computation of adjoint sensitivities without adjoint solvers: An application to thermoacoustics. Physical Review Fluids9(10), 103902 (Oct 2024). https://doi.org/10.1103/physrevfluids.9.103902

    work page doi:10.1103/physrevfluids.9.103902 2024

  13. [13]

    Journal of Mathemat- ics in Industry1(1), 3 (Dec 2011)

    Quarteroni, A., Rozza, G., Manzoni, A.: Certified reduced basis approximation for parametrized partial differential equations and applications. Journal of Mathemat- ics in Industry1(1), 3 (Dec 2011). https://doi.org/10.1186/2190-5983-1-3

    work page doi:10.1186/2190-5983-1-3 2011

  14. [14]

    SIAM Journal on Applied Mathematics39(1), 67–82 (Aug 1980)

    Sivashinsky, G.I.: On Flame Propagation Under Conditions of Stoichiom- etry. SIAM Journal on Applied Mathematics39(1), 67–82 (Aug 1980). https://doi.org/10.1137/0139007

    work page doi:10.1137/0139007 1980

  15. [15]

    Annual Review of Fluid Me- chanics57(1), 311–334 (Jan 2025)

    Zaki, T.A.: Turbulence from an Observer Perspective. Annual Review of Fluid Me- chanics57(1), 311–334 (Jan 2025). https://doi.org/10.1146/annurev-fluid-030424- 114735

    work page doi:10.1146/annurev-fluid-030424- 2025