arxiv: 2605.07052 · v1 · submitted 2026-05-08 · 📡 eess.SY · cs.LG· cs.SY

Recognition: 2 theorem links

· Lean Theorem

A Behavioral Framework for Data-Driven Modeling of Nonlinear Systems in Vector-Valued Reproducing Kernel Hilbert Spaces

Boya Hou, Maxim Raginsky

Authors on Pith no claims yet

Pith reviewed 2026-05-11 01:01 UTC · model grok-4.3

classification 📡 eess.SY cs.LGcs.SY

keywords behavioral approachdata-driven modelingnonlinear systemsreproducing kernel Hilbert spaceVolterra seriesHammerstein systemssubspace identificationminimum-norm interpolation

0 comments

The pith

Behavioral approach generalizes to nonlinear systems in vector-valued RKHS for data-driven modeling

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

This paper generalizes Jan Willems' behavioral approach from linear time-invariant systems to a class of discrete-time nonlinear systems embedded in vector-valued reproducing kernel Hilbert spaces. The covered systems include those modeled by Volterra series and autoregressive variants, plus those with Hammerstein-type state-space realizations. The framework links behavioral properties directly to data-driven techniques such as minimum-norm interpolation and subspace identification. A sympathetic reader would care because it supports simulation and control tasks on unknown nonlinear systems straight from data, skipping the usual explicit identification step.

Core claim

The paper establishes that behavioral properties carry over when the target nonlinear systems are represented in a vector-valued RKHS, so that data-driven modeling objectives can be achieved without an intermediate explicit system model.

What carries the argument

The embedding of system trajectories into a vector-valued reproducing kernel Hilbert space, which preserves key behavioral relations and interfaces them with kernel-based interpolation and identification algorithms.

If this is right

Simulation and control of Volterra and Hammerstein systems become feasible directly from data.
Minimum-norm interpolation in the RKHS serves as a modeling tool under the behavioral framework.
Subspace identification techniques extend to these nonlinear systems via the RKHS embedding.
Data-driven objectives avoid the need for an explicit intermediate model identification.

Where Pith is reading between the lines

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

The same embedding strategy could apply to other nonlinear classes that admit RKHS representations.
Links to kernel methods may support extensions to uncertain or partially observed nonlinear dynamics.

Load-bearing premise

That the nonlinear systems of interest can be faithfully embedded and operated upon inside a vector-valued RKHS so that behavioral properties and the associated data-driven links remain valid.

What would settle it

A concrete calculation showing that minimum-norm interpolation or subspace methods in the RKHS produce output trajectories that differ measurably from the true responses of a known Volterra or Hammerstein system driven by the same inputs.

read the original abstract

We generalize Jan Willems' behavioral approach to a class of discrete-time nonlinear systems in a vector-valued reproducing kernel Hilbert space (RKHS). Apart from linear time-invariant systems, this class covers nonlinear systems modeled by Volterra series and their autoregressive variants, as well as systems admitting Hammerstein-type state-space realizations. We apply the proposed framework to the problem of data-driven modeling of such systems, i.e., when simulation or control objectives for an unknown system are carried out without an explicit system identification step. To that end, we link the behavioral approach to two data-driven modeling methods in a vector-valued RKHS: (1) minimum-norm interpolation and (2) subspace identification.

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.

Desk Editor's Note private letter to a colleague

This paper extends Willems' behavioral approach to nonlinear systems via vector-valued RKHS embeddings, with direct derivations to interpolation and subspace methods that hold up internally for the covered classes.

read the letter

This paper generalizes Jan Willems' behavioral approach to discrete-time nonlinear systems by placing them in a vector-valued reproducing kernel Hilbert space. It covers Volterra series and autoregressive variants along with Hammerstein realizations, then uses that setup to connect behavioral ideas to data-driven modeling through minimum-norm interpolation and subspace identification. The main strength is that these connections follow from the reproducing property, keeping the framework consistent for the systems that embed properly. The work does well in providing a unified lens without forcing linearity or other restrictions within its scope. The stress-test confirms that the derivations avoid circularity and hidden assumptions, so the argument stands on its own terms for the targeted class. This is a genuine extension rather than a minor tweak, as the abstract positions the RKHS embedding as the key step that enables the data-driven links. Soft spots are present but not central. The class of systems is specific, so the framework requires that the nonlinear dynamics admit the embedding; if a system doesn't fit neatly, the behavioral properties may not transfer directly. The paper should spell out those embedding conditions clearly in the full text. Also, since we only have the abstract here, the actual proofs and examples would need review to confirm no extra fitting is involved in the predictions. The citation pattern looks standard, referencing Willems and kernel methods without obvious gaps. This is for readers in systems theory, control, and machine learning who deal with data-driven approaches to nonlinear dynamics. A colleague working on behavioral theory or kernel-based identification would get value from seeing how the two areas link up. It is not for someone looking for a general nonlinear theory that covers everything. I would bring this to a reading group to walk through the embeddings and see how the subspace method adapts. I probably would not cite it in my next paper unless the details prove robust, but the paper shows clear thinking and deserves a serious referee. I recommend sending it to peer review.

Referee Report

0 major / 4 minor

Summary. The manuscript generalizes Jan Willems' behavioral approach to discrete-time nonlinear systems represented in vector-valued reproducing kernel Hilbert spaces (RKHS). The covered class includes Volterra series, their autoregressive variants, and systems with Hammerstein-type state-space realizations. The framework is applied to data-driven modeling by linking behavioral properties directly to minimum-norm interpolation and subspace identification methods in the RKHS, enabling simulation and control of unknown systems without an explicit identification step.

Significance. If the embeddings and derivations hold, the result is significant: it extends the behavioral framework, long influential for linear systems, to a practically relevant class of nonlinear systems via RKHS tools. The direct connections to interpolation and subspace methods provide a unified data-driven route that avoids model identification, with potential impact on nonlinear control and identification. The internal consistency of the trajectory-space construction and reproducing-property arguments is a clear strength.

minor comments (4)

The abstract states that the class covers Volterra, AR, and Hammerstein systems, but the introduction would benefit from a short forward reference to the specific embedding constructions (likely in §3 or §4) so readers can immediately locate the supporting arguments.
Notation for the vector-valued RKHS and the associated trajectory space should be introduced once in a dedicated preliminary section and then used uniformly; occasional redefinitions in later sections reduce readability.
Figure captions for any illustrative trajectory plots or kernel matrices should explicitly state the kernel choice and the dimension of the output space to allow immediate assessment of the numerical examples.
A brief discussion of computational complexity or scalability of the minimum-norm interpolation step relative to classical subspace methods would strengthen the practical claims without altering the theoretical contribution.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive summary, significance assessment, and recommendation of minor revision. The report affirms the internal consistency of the trajectory-space construction and the links to interpolation and subspace methods. No specific major comments were raised in the report.

Circularity Check

0 steps flagged

No significant circularity in the derivation chain

full rationale

The paper generalizes Willems' behavioral approach by embedding specified nonlinear classes (Volterra, AR variants, Hammerstein) into a vector-valued RKHS and derives data-driven links to minimum-norm interpolation and subspace methods from the reproducing property and trajectory space. No self-definitional reductions, fitted inputs renamed as predictions, or load-bearing self-citation chains appear; the central claims rest on the stated embedding assumption and RKHS axioms without circular collapse to inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract supplies no explicit free parameters, axioms, or invented entities; full manuscript required to audit.

pith-pipeline@v0.9.0 · 5417 in / 971 out tokens · 36164 ms · 2026-05-11T01:01:23.256494+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

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

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear
We generalize Jan Willems' behavioral approach to a class of discrete-time nonlinear systems in a vector-valued reproducing kernel Hilbert space (RKHS). ... link the behavioral approach to two data-driven modeling methods ... (1) minimum-norm interpolation and (2) subspace identification.
IndisputableMonolith/Foundation/ArithmeticFromLogic.lean embed_injective unclear
Theorem 6 (Behavior Representer Theorem). ... If ΣT(z0)=0 then y0+=kT(z0)K†T YT

Reference graph

Works this paper leans on

26 extracted references

[1]

From time series to linear system—part i. finite dimensional linear time invariant systems,

J. C. Willems, “From time series to linear system—part i. finite dimensional linear time invariant systems,”Automatica, vol. 22, no. 5, pp. 561–580, 1986

1986
[2]

Paradigms and puzzles in the theory of dynamical systems,

——, “Paradigms and puzzles in the theory of dynamical systems,”IEEE Transactions on Automatic Control, vol. 36, no. 3, pp. 259–294, 1991

1991
[3]

A note on persistency of excitation,

J. C. Willems, P. Rapisarda, I. Markovsky, and B. L. De Moor, “A note on persistency of excitation,”Systems & Control Letters, vol. 54, no. 4, pp. 325–329, 2005

2005
[4]

Markovsky, J

I. Markovsky, J. C. Willems, S. Van Huffel, and B. De Moor, Exact and approximate modeling of linear systems: A behavioral approach. SIAM, 2006

2006
[5]

Data-driven simulation and control,

I. Markovsky and P. Rapisarda, “Data-driven simulation and control,”International Journal of Control, vol. 81, no. 12, pp. 1946–1959, 2008

1946
[6]

Data-enabled predictive control: In the shallows of the deepc,

J. Coulson, J. Lygeros, and F. D ¨orfler, “Data-enabled predictive control: In the shallows of the deepc,” in2019 18th European control conference (ECC). IEEE, 2019, pp. 307–312

2019
[7]

Bridging direct and indirect data-driven control formulations via regularizations and relaxations,

F. D¨orfler, J. Coulson, and I. Markovsky, “Bridging direct and indirect data-driven control formulations via regularizations and relaxations,”IEEE Transactions on Automatic Control, vol. 68, no. 2, pp. 883–897, 2022

2022
[8]

The informativity approach to data-driven analysis and control,

H. J. Van Waarde, J. Eising, M. K. Camlibel, and H. L. Trentelman, “The informativity approach to data-driven analysis and control,”IEEE Control Systems Magazine, vol. 43, no. 6, pp. 32–66, 2023

2023
[9]

Data-driven dynamic interpolation and approximation,

I. Markovsky and F. D ¨orfler, “Data-driven dynamic interpolation and approximation,”Automatica, vol. 135, p. 110008, 2022

2022
[10]

Data-driven internal model control of second-order discrete volterra systems,

J. G. Rueda-Escobedo and J. Schiffer, “Data-driven internal model control of second-order discrete volterra systems,” in 2020 59th IEEE Conference on Decision and Control (CDC). IEEE, 2020, pp. 4572–4579

2020
[11]

Data-based system analysis and control of flat nonlinear systems,

M. Alsalti, J. Berberich, V . G. Lopez, F. Allg ¨ower, and M. A. M¨uller, “Data-based system analysis and control of flat nonlinear systems,” in2021 60th IEEE Conference on Decision and Control (CDC). IEEE, 2021, pp. 1484–1489

2021
[12]

A trajectory-based framework for data-driven system analysis and control,

J. Berberich and F. Allg ¨ower, “A trajectory-based framework for data-driven system analysis and control,” in2020 European Control Conference (ECC). IEEE, 2020, pp. 1365–1370

2020
[13]

Willems’ fundamental lemma for nonlinear systems with koopman linear embedding,

X. Shang, J. Cort ´es, and Y . Zheng, “Willems’ fundamental lemma for nonlinear systems with koopman linear embedding,” IEEE Control Systems Letters, 2024

2024
[14]

Robust and kernelized data-enabled predictive control for nonlinear systems,

L. Huang, J. Lygeros, and F. D ¨orfler, “Robust and kernelized data-enabled predictive control for nonlinear systems,”IEEE Transactions on Control Systems Technology, vol. 32, no. 2, pp. 611–624, 2023

2023
[15]

Exploring the links between the fundamental lemma and kernel regression,

O. Molodchyk and T. Faulwasser, “Exploring the links between the fundamental lemma and kernel regression,”IEEE Control Systems Letters, vol. 8, pp. 2045–2050, 2024

2045
[16]

A generalized rep- resenter theorem,

B. Sch ¨olkopf, R. Herbrich, and A. J. Smola, “A generalized rep- resenter theorem,” inInternational conference on computational learning theory. Springer, 2001, pp. 416–426

2001
[17]

Van Overschee and B

P. Van Overschee and B. De Moor,Subspace identification for linear systems: Theory—Implementation—Applications. Kluwer, 1996

1996
[18]

Berlinet and C

A. Berlinet and C. Thomas-Agnan,Reproducing kernel Hilbert spaces in probability and statistics. Springer Science & Business Media, 2011

2011
[19]

Theory of reproducing kernels,

N. Aronszajn, “Theory of reproducing kernels,”Transactions of the American mathematical society, vol. 68, no. 3, pp. 337–404, 1950

1950
[20]

Vector valued reproducing kernel Hilbert spaces of integrable functions and Mercer theorem,

C. Carmeli, E. De Vito, and A. Toigo, “Vector valued reproducing kernel Hilbert spaces of integrable functions and Mercer theorem,” Analysis and Applications, vol. 04, no. 04, pp. 377–408, Oct. 2006

2006
[21]

Persistence of excitation in extended least squares,

J. Moore, “Persistence of excitation in extended least squares,” IEEE Transactions on Automatic Control, vol. 28, no. 1, pp. 60–68, 1983

1983
[22]

Persistence of excitation in linear systems,

M. Green and J. B. Moore, “Persistence of excitation in linear systems,”Systems & Control Letters, vol. 7, no. 5, pp. 351–360, 1986

1986
[23]

A best approximation frame- work and implementation for simulation of large-scale nonlinear systems,

R. De Figueiredo and T. Dwyer, “A best approximation frame- work and implementation for simulation of large-scale nonlinear systems,”IEEE Transactions on circuits and systems, vol. 27, no. 11, pp. 1005–1014, 1980

1980
[24]

On learning vector-valued functions,

C. A. Micchelli and M. Pontil, “On learning vector-valued functions,”Neural computation, vol. 17, no. 1, pp. 177–204, 2005

2005
[25]

Interpolating classifiers make few mistakes,

T. Liang and B. Recht, “Interpolating classifiers make few mistakes,”Journal of Machine Learning Research, vol. 24, no. 20, pp. 1–27, 2023

2023
[26]

Willems’ fundamental lemma for state-space systems and its extension to multiple datasets,

H. J. Van Waarde, C. De Persis, M. K. Camlibel, and P. Tesi, “Willems’ fundamental lemma for state-space systems and its extension to multiple datasets,”IEEE Control Systems Letters, vol. 4, no. 3, pp. 602–607, 2020

2020