pith. sign in

arxiv: 2604.18783 · v2 · submitted 2026-04-20 · 🧮 math.OC · cs.SY· eess.SY

A Dynamic Mode Decomposition Approach to Parameter Identification

Moad Abudia , Opeyemi Owolabi , Joel A. Rosenfeld , Rushikesh Kamalapurkar This is my paper

Pith reviewed 2026-05-10 03:45 UTC · model grok-4.3

classification 🧮 math.OC cs.SYeess.SY
keywords dynamic mode decompositionparameter identificationsystem identificationnonlinear control systemsDuffing oscillatordata-driven modelingcontrol-affine systems
0
0 comments X

The pith

A data-driven method trains a predictive model on known parameter values to recover unknown parameters like damping and stiffness from new trajectories in nonlinear systems.

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

The paper develops an algorithm that performs simultaneous system identification and parameter estimation for control-affine nonlinear systems using only measured state-action data. It first builds a predictive model from trajectories generated under several known values of the parameters of interest. The trained model is then applied to fresh state-action data to infer the unknown parameter values. A sympathetic reader would care because this approach could estimate physical quantities such as damping or stiffness directly from observed behavior without an analytical model of the full system dynamics. Numerical tests on the controlled Duffing oscillator recover both the trajectories and the unknown coefficients with good accuracy under open-loop conditions.

Core claim

By training a data-driven predictive model on state-action measurements taken at various known values of the parameters of interest, the model can subsequently be used with new state-action data to estimate the unknown parameter values in control-affine nonlinear systems, as demonstrated by accurate recovery of both trajectories and parameters in experiments on the controlled Duffing oscillator with unknown damping, stiffness, and nonlinearity coefficients under open-loop excitation.

What carries the argument

Dynamic mode decomposition builds the data-driven predictive model that maps state-action pairs to future states while embedding dependence on the parameters of interest.

If this is right

  • System trajectories can be predicted accurately even when parameters are unknown.
  • Parameters such as damping and stiffness are recoverable from open-loop excitation data.
  • The approach applies to control-affine nonlinear systems without requiring closed-form equations.
  • Parameter estimation occurs simultaneously with learning the system dynamics from data.

Where Pith is reading between the lines

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

  • If the model generalizes across parameter ranges, the technique could support online adaptation in systems whose physical properties change over time.
  • The same training-and-inference pattern might extend to identification tasks in other nonlinear oscillators or chaotic systems.
  • Incorporating closed-loop trajectories into the training set could reduce estimation error when excitation is limited.

Load-bearing premise

A predictive model trained exclusively on state-action data with known parameter values will generalize accurately enough to recover the correct unknown parameter values from new state-action trajectories.

What would settle it

Large mismatches between the estimated and true values of damping, stiffness, or nonlinearity coefficients in the controlled Duffing oscillator numerical experiments would show that the method fails to identify the parameters correctly.

Figures

Figures reproduced from arXiv: 2604.18783 by Joel A. Rosenfeld, Moad Abudia, Opeyemi Owolabi, Rushikesh Kamalapurkar.

Figure 1
Figure 1. Figure 1: A comparison between the true system trajectories [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 4
Figure 4. Figure 4: The MSE computed using different parameter values on the mesh [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 3
Figure 3. Figure 3: The MSE computed using different parameter values on the [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
read the original abstract

This paper presents a data-driven algorithm for simultaneous system identification and parameter estimation in control-affine nonlinear systems. Parameter estimation is achieved by training a data-driven predictive model using state-action measurements and various known values at the parameters of interest. The predictive model is then used in conjunction with state-action data corresponding to unknown values of the parameters to estimate the said unknown value. Numerical experiments on the controlled Duffing oscillator with unknown damping, stiffness, and nonlinearity coefficients demonstrate accurate recovery of both the system trajectories and the unknown parameter values from data collected under open-loop excitation.

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 data-driven algorithm for simultaneous system identification and parameter estimation in control-affine nonlinear systems. Parameter estimation is achieved by training a DMD-based predictive model on state-action measurements collected at various known values of the parameters of interest. The trained model is then applied to new state-action trajectories corresponding to unknown parameter values to recover those values. Numerical experiments on the controlled Duffing oscillator with unknown damping, stiffness, and nonlinearity coefficients are reported to demonstrate accurate recovery of both trajectories and parameter values from open-loop excitation data.

Significance. If the central claim holds with proper quantitative validation, the work offers a practical parametric extension of DMD for joint trajectory prediction and parameter recovery in nonlinear control-affine systems. This could be useful in settings where only input-output data are available and parameters must be inferred without an explicit model structure. The reliance on open-loop data is a positive feature that avoids circularity from closed-loop feedback. However, the absence of error metrics, rank-selection details, and baselines in the abstract leaves the practical significance difficult to assess at present.

major comments (2)
  1. [Abstract] Abstract: the central claim of 'accurate recovery' of both trajectories and unknown parameter values is stated without any quantitative error metrics (e.g., RMSE, parameter estimation error), description of the training procedure across known parameter values, DMD rank selection method, or comparison to existing parameter-identification baselines. These omissions make it impossible to evaluate whether the numerical results on the Duffing oscillator actually support the claim.
  2. [Numerical experiments] Numerical experiments section: the procedure relies on a predictive model trained at discrete known parameter values generalizing to continuous unknown values; no analysis is provided of the interpolation/extrapolation properties, the required richness of the open-loop excitation, or sensitivity to DMD rank, all of which are load-bearing for the parameter-recovery claim.
minor comments (2)
  1. Clarify the precise definition of the control-affine structure assumed and how the DMD operator is constructed to incorporate the known parameter values during training.
  2. Add a brief discussion of computational cost and scalability with state dimension, as DMD-based methods can become expensive for high-dimensional systems.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed review. We address each major comment below and have revised the manuscript to incorporate quantitative support and additional analysis where the original version was lacking.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim of 'accurate recovery' of both trajectories and unknown parameter values is stated without any quantitative error metrics (e.g., RMSE, parameter estimation error), description of the training procedure across known parameter values, DMD rank selection method, or comparison to existing parameter-identification baselines. These omissions make it impossible to evaluate whether the numerical results on the Duffing oscillator actually support the claim.

    Authors: We agree that the abstract would be strengthened by including quantitative metrics and methodological details. In the revised version we have added the average RMSE for trajectory prediction (0.012) and mean relative error for parameter recovery (under 3% across the three coefficients). We now briefly describe the training procedure (data collected on a uniform grid of known parameter values) and note that DMD rank is chosen by inspecting the singular-value spectrum with a 99% energy threshold. A short comparison to a direct least-squares parameter-fitting baseline has also been included. revision: yes

  2. Referee: [Numerical experiments] Numerical experiments section: the procedure relies on a predictive model trained at discrete known parameter values generalizing to continuous unknown values; no analysis is provided of the interpolation/extrapolation properties, the required richness of the open-loop excitation, or sensitivity to DMD rank, all of which are load-bearing for the parameter-recovery claim.

    Authors: We acknowledge that a dedicated study of generalization and robustness was missing. The revised manuscript adds a new subsection that reports interpolation and extrapolation results on a finer parameter grid (including values outside the training set) together with the corresponding parameter-estimation errors. We also include a sensitivity plot versus DMD rank and a brief discussion of the persistence-of-excitation conditions satisfied by the chosen open-loop inputs. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper trains a DMD-based predictive model exclusively on state-action trajectories paired with known parameter values, then applies the model to separate trajectories with unknown parameter values to recover those values. This separation between training data (known parameters) and inference data (unknown parameters) prevents any reduction of the claimed prediction to a tautology or self-definition. No load-bearing self-citations, uniqueness theorems imported from prior author work, or ansatzes smuggled via citation are described. The numerical validation on the Duffing oscillator uses external open-loop data and does not rely on internal fitting loops that would force the result by construction. The method is a parametric extension of standard DMD and remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the system belonging to the control-affine nonlinear class and on the DMD model trained at known parameter values being sufficiently expressive to allow inversion for unknown values; no free parameters or invented entities are mentioned in the abstract.

axioms (1)
  • domain assumption The dynamical system is control-affine nonlinear
    Explicitly stated as the class of systems for which the algorithm is designed.

pith-pipeline@v0.9.0 · 5397 in / 1270 out tokens · 31727 ms · 2026-05-10T03:45:02.629392+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

21 extracted references · 21 canonical work pages

  1. [1]

    System identification—a survey,

    K. J. ˚Astr¨om and P. Eykhoff, “System identification—a survey,” Automatica, vol. 7, no. 2, pp. 123–162, 1971

    work page 1971

  2. [2]

    Ljung,System identification — theory for the user

    L. Ljung,System identification — theory for the user. Prentice Hall, 1999

    work page 1999

  3. [3]

    Ioannou and J

    P. Ioannou and J. Sun,Robust adaptive control. Prentice Hall, 1996

    work page 1996

  4. [4]

    On the adaptive control of robot manipu- lators,

    J.-J. E. Slotine and W. Li, “On the adaptive control of robot manipu- lators,”Int. J. Robot. Res., vol. 6, no. 3, pp. 49–59, 1987

    work page 1987

  5. [5]

    State and parameter estimation for affine nonlinear systems,

    T. E. Ogri, Z. I. Bell, and R. Kamalapurkar, “State and parameter estimation for affine nonlinear systems,” inProc. IEEE Conf. Decis. Control, Singapore, 2023, pp. 1517–1522

    work page 2023

  6. [6]

    Physics-informed neural networks: a deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations,

    M. Raissi, P. Perdikaris, and G. E. Karniadakis, “Physics-informed neural networks: a deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations,” J. Comput. Phys., vol. 378, pp. 686–707, 2019

    work page 2019

  7. [7]

    Parameter estimation for a class of radial basis function-based nonlinear time-series models with moving average noises,

    Y . Zhou, Y . Wang, F. Ma, F. Ding, and T. Hayat, “Parameter estimation for a class of radial basis function-based nonlinear time-series models with moving average noises,”J. Franklin Inst., vol. 358, no. 4, pp. 2576–2595, 2021

    work page 2021

  8. [8]

    Adaptive control of a flexible-link robotic manipulator with unknown payload dynamics,

    L. J. Alder and S. M. Rock, “Adaptive control of a flexible-link robotic manipulator with unknown payload dynamics,” inProc. Amer. Control Conf., San Francisco, CA, USA, 1993, p. 467

    work page 1993

  9. [9]

    Adaptive control of a quadrotor UA V transporting a cable-suspended load with unknown mass,

    S. Dai, T. Lee, and D. S. Bernstein, “Adaptive control of a quadrotor UA V transporting a cable-suspended load with unknown mass,” in Proc. IEEE Conf. Decis. Control, Los Angeles, CA, USA, 2014, pp. 6149–6154

    work page 2014

  10. [10]

    Grey-box model for model predictive control of buildings,

    P. Klanatsky, F. Veynandt, and C. Heschl, “Grey-box model for model predictive control of buildings,”Energy Build., vol. 300, no. 1, p. 113624, 2023

    work page 2023

  11. [11]

    Occupancy prediction for building energy systems with latent force models,

    T. Wietzke, J. Gall, and K. Graichen, “Occupancy prediction for building energy systems with latent force models,”Energy Build., vol. 307, p. 113968, 2024

    work page 2024

  12. [12]

    Robust Bayesian state and parameter estimation framework for stochastic dynamical systems with combined time- varying and time-invariant parameters,

    P. Bisaillon, B. Robinson, M. Khalil, C. L. Pettit, D. Poirel, and A. Sarkar, “Robust Bayesian state and parameter estimation framework for stochastic dynamical systems with combined time- varying and time-invariant parameters,”J. Sound Vib., vol. 575, p. 118106, 2024

    work page 2024

  13. [13]

    A sparse Bayesian framework for discovering interpretable nonlinear stochastic dynamical systems with Gaussian white noise,

    T. Tripura and S. Chakraborty, “A sparse Bayesian framework for discovering interpretable nonlinear stochastic dynamical systems with Gaussian white noise,”Mech. Syst. Signal Process., vol. 187, p. 109939, 2023

    work page 2023

  14. [14]

    Deep state space models for nonlinear system identification,

    D. Gedon, N. Wahlstr ¨om, T. B. Sch ¨on, and L. Ljung, “Deep state space models for nonlinear system identification,”IFAC- PapersOnLine, vol. 54, no. 7, pp. 481–486, 2021

    work page 2021

  15. [15]

    Dynamic mode decomposition of control-affine nonlinear systems using discrete control liouville operators,

    Z. Morrison, M. Abudia, J. A. Rosenfeld, and R. Kamalapurkar, “Dynamic mode decomposition of control-affine nonlinear systems using discrete control liouville operators,”IEEE Control Systems Letters, vol. 8, pp. 79–84, 2024

    work page 2024

  16. [16]

    Koopman-based lifting techniques for nonlinear systems identification,

    A. Mauroy and J. Goncalves, “Koopman-based lifting techniques for nonlinear systems identification,”IEEE Trans. Autom. Control, vol. 65, no. 6, pp. 2550–2565, 2020

    work page 2020

  17. [17]

    Dynamic mode decomposition for continuous time systems with the Liouville operator,

    J. A. Rosenfeld, R. Kamalapurkar, L. F. Gruss, and T. T. Johnson, “Dynamic mode decomposition for continuous time systems with the Liouville operator,”J. Nonlinear Sci., vol. 32, no. 1, pp. 1–30, 2022

    work page 2022

  18. [18]

    The kernel perspective on dynamic mode decomposition,

    E. Gonzalez, M. Abudia, M. Jury, R. Kamalapurkar, and J. A. Rosenfeld, “The kernel perspective on dynamic mode decomposition,” Trans. Mach. Learn. Res., 2024

    work page 2024

  19. [19]

    On convergent dynamic mode decomposition and its equivalence with occupation kernel regression,

    M. Abudia, J. A. Rosenfeld, and R. Kamalapurkar, “On convergent dynamic mode decomposition and its equivalence with occupation kernel regression,” inIFAC-PapersOnLine, vol. 58, no. 17, 2024, pp. 103–108

    work page 2024

  20. [20]

    Vector valued reproducing kernel Hilbert spaces and universality,

    C. Carmeli, E. De Vito, A. Toigo, and V . Umanit ´a., “Vector valued reproducing kernel Hilbert spaces and universality,”Anal. Appl., vol. 08, no. 01, pp. 19–61, 2010

    work page 2010

  21. [21]

    Dynamic mode decomposition with control Liouville operators,

    J. A. Rosenfeld and R. Kamalapurkar, “Dynamic mode decomposition with control Liouville operators,” inIFAC-PapersOnLine, vol. 54, no. 9, 2021, pp. 707–712

    work page 2021