L2-L2-gain bounds for quadratic output systems
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Mathematics. The day in mathematics: new theorems, new proofs, and what they open.
Ada: We're looking at the paper: "L2-L2-gain bounds for quadratic output systems".
Emmy: Here's the gist of the paper. L2-L2-gain bounds for quadratic output systems derive an explicit bound for the L2-L2 gain of linear time-invariant systems whose output is a quadratic function of the state and the input.
Ada: This result provides a practical tool for assessing and reducing quadratic-output models.
Emmy: System Formulation and Motivation The paper investigates linear time-invariant systems with quadratic outputs (LQO systems) of the form defined in equation (1.1).
Ada: These systems appear naturally in several application areas, such as port-Hamiltonian systems, optimal control, and stochastic problems.
Paper summary: Ada: To wrap up this talk on "L2-L2-gain bounds for quadratic output systems", we've covered how they tackle the challenge of finding a concrete L2-L2 gain number for systems whose outputs are quadratic functions of both the state and input.
Emmy: The authors, Birgit Hillebrecht, focus on deriving this explicit bound and showing how it can be computed using a set of Lyapunov and Sylvester matrix equations, which is a practical computational approach.
Kolya: What does this all mean for the bigger picture? It suggests that we can get a reliable way to assess how much the output will deviate from what we expect based on the input, even in these more complex quadratic settings.
Ada: The implication is that H2 oriented model reduction algorithms can be adapted for these quadratic output models to find an optimal reduced order model in terms of minimizing the L2 error of the output relative to the L2 magnitude of your input.
Emmy: And while they provide a bound, they also point out that adapting balanced truncation for this purpose is still difficult because of how complicated the inner product formulation is for finding an observability energy Gramian.
Kolya: So, in simple terms, it's about getting a computable measure of performance for these systems without having to do heavy numerical integration over the frequency axis.
Ada: That's right. The paper gives a pathway to move from abstract quadratic output models to concrete error bounds that you can actually use in design and analysis.
Conclusion: Ada: So we’re wrapping up this look at these L2-L2 gain bounds for quadratic output systems, focusing on Birgit Hillebrecht and her team.
Emmy: It’s about finding a way to get a concrete number—an explicit bound—for how well these kinds of linear time invariant systems perform in the L2 sense.
Kolya: What does that actually mean for us? Does it give us a tool we can actually use when we’re trying to design or reduce these models?
Ada: It suggests that instead of just knowing the general rules for linear systems, we now have a specific formula for quadratic output models.
Emmy: The authors show how you get this bound by looking at the system in terms of bivariate transfer functions and using some specific inner product setup.
Kolya: And that inner product thing you mentioned earlier, how do you actually compute it without having to integrate over every single frequency point? Is that a real computational win?
Ada: It is. They use a set of Lyapunov and Sylvester matrix equations, which means the computation becomes solving those matrices instead of doing messy integration.
Emmy: That structure behind the computation is interesting; it shows how geometry helps us define this error measure cleanly.
Kolya: But what about when you move to systems that have both bilinear and quadratic terms? Does this same approach still work there, or do things get much messier?
Ada: They do extend the thinking to bilinear and quadratic throughput terms, but they use Hölder’s inequality on an adaptation of a formula from earlier in the paper.
Emmy: That leads to a bound involving the H infinity norm of an LTI system that’s determined by some specific matrices A, B, R and P.
Kolya: So what's the practical implication for someone just trying to run a simulation? Does this mean faster model reduction algorithms are possible now?
Ada: The first two results point toward adapting H2 oriented model reduction algorithms from linear systems to these quadratic ones.
Emmy: The goal is finding an optimal reduced order model that minimizes the L2 error of the output relative to the L2 magnitude of your input.
Kolya: And they also mention that a balanced truncation method for this still seems elusive because defining that observability energy Gramian is super involved.
Ada: So, while we got a bound, adapting standard balanced truncation to match it isn't as straightforward as you might hope.
Emmy: It means the structure of the quadratic output adds complexity to the model reduction process itself.
Kolya: In short, they give us a concrete error measure and a way to calculate it, but it’s still more work than just plugging in some numbers.
Birgit Hillebrecht
Karlsruhe Institute of Technology
math.OC, math.DS
Submitted: 2026-07-01
Updated: 2026-10-31
License: http://creativecommons.org/licenses/by/4.0/
Importance score: 80/100
The gist: L2-L2-gain bounds for quadratic output systems derive an explicit bound for the L2-L2 gain of linear time-invariant systems whose output is a quadratic function of the state and the input.
Key concepts
- L2-L2-gain
- This measures the worst-case amplification of an LTI system's output energy relative to its input energy over all time. A lower L2-L2 gain indicates a more stable and less noisy system when dealing with signals in the L2 space, which is crucial for control applications.
- LQO systems
- These are Linear Time-Invariant systems where the output signal is not just a linear function of the state but involves quadratic terms (e.g., $y(t) = M(x(t) imes x(t))$). They arise in various fields like optimal control and port-Hamiltonian systems.
- Inner Product for Computability
- To find a usable error measure between two LQO systems, the paper introduces a specific inner product. This allows the complex output error to be calculated by solving a set of Lyapunov and Sylvester matrix equations instead of needing difficult numerical integration over frequency.
Terminology
Summary
L2-L2-gain bounds for quadratic output systems derive an explicit bound for the L2-L2 gain of linear time-invariant systems whose output is a quadratic function of the state and the input. p. 0 edu
Abstract. We derive an explicit bound for the L2-L2 gain of linear time-invariant
systems whose output is a quadratic function of the state and the input. Further, we show how
the bound can be computed by solving linear matrix equations. This result provides a
practical tool for assessing and reducing quadratic-output models.Where the paper says it
Where the paper says it
System Formulation and Motivation
The paper investigates linear time-invariant systems with quadratic outputs (LQO systems) of the form defined in equation (1.1). These systems appear naturally in several application areas, such as port-Hamiltonian systems, optimal control, and stochastic problems. p. 0 1)
extend the standard linear time-invariant (LTI) systems to a mildly nonlinear form. Such
quadratic outputs arise naturally in several application areas: In port-Hamiltonian systems
they reflect the input-energy relation [10], in linear-quadratic optimal control they are part
of the optimization objective [7], and in stochastic problems the variance is a quadratic
function of the states [9]. Such
quadratic outputs arise naturally in several application areas: In port-Hamiltonian systems
they reflect the input-energy relation [10], in linear-quadratic optimal control they are part
of the optimization objective [7], and in stochastic problems the variance is a quadratic
function of the states [9]. Motivated by these applications and the relevance of L2
-L2 gains for robust and optimal
control [14, 17] as well as for model order reduction (MOR) [3], the central aim of this
manuscript is to derive a bound for the L2
-L2 gain of the system (1. Motivated by these applications and the relevance of L2
-L2 gains for robust and optimal
control [14, 17] as well as for model order reduction (MOR) [3], the central aim of this
manuscript is to derive a bound for the L2
-L2 gain of the system (1. It is well-known
that, for stable LTI systems, the L2
-L2 gain is bounded by the H∞-norm of the transfer
function [3]. It is well-known
that, for stable LTI systems, the L2
-L2 gain is bounded by the H∞-norm of the transfer
function [3]. Nevertheless, an analogous expression for LQO systems is not yet known.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Frequency Domain Representation and Bounds
For systems whose output depends solely quadratically on the state, i.e., y(t) = M(x(t) ⊗ x(t)), the bivariate transfer function and output are given by (2. p. 2 ,
ΣM
(
x˙(t) = Ax(t) + Bu(t),
y(t) = M(x(t) ⊗ x(t)). 1)
For this restricted case, the bivariate transfer function and output are given by [6]
Ge (s1
, s2
) = M
h
(s1I − A)
−1B ⊗ (s2I − A)
−1B
i
∈ C
p×m
2
, (2. ,
ΣM
(
x˙(t) = Ax(t) + Bu(t),
y(t) = M(x(t) ⊗ x(t)). 1)
For this restricted case, the bivariate transfer function and output are given by [6]
Ge (s1
, s2
) = M
h
(s1I − A)
−1B ⊗ (s2I − A)
−1B
i
∈ C
p×m
2
, (2. Such systems
appear naturally in many areas, for example for port-Hamiltonian systems, optimalcontrol, and stochastic problems. In case the output is purely quadratic in the state,
the bound equals the L2-norm of the bivariate transfer function evaluated along the
anti-diagonal {(s, −s) | s ∈ iR} of the iR × iR frequency domain.Where the paper says it
Where the paper says it
Where the paper says it
Inner Product and Computability
To obtain a computable expression for the output error between two LQO-systems, a suitable inner product is introduced in Definition 3.1. Theorem 3.3 then shows how this inner product can be evaluated by solving a set of Lyapunov and Sylvester matrix equations. p. 6 Inner product and Sylvester-equation-based computation of the gain
bound
To obtain a computable expression for the output error between two LQO-systems we
first introduce a suitable inner product in Definition 3. 3 then shows how
this inner product can be evaluated by solving a set of Lyapunov and Sylvester matrix
equations. The paper concludes with a brief numerical example in Section 7 and an
outlook in Section 8. From now on we follow the convention that bivariate auxiliary functions
carry a tilde ˜·, that frequency-domain quantities are written with uppercase letters, and
that time-domain quantities are written with lowercase letters.Where the paper says it
Where the paper says it
General Gain Bound Structure
The full L2-L2-gain bound for an LQO system (1.1) without quadratic throughput, i. p. 2 3 that the inner product can be evaluated by solving a set of
Lyapunov and Sylvester equations. In the second stage, we investigate the bilinear and
quadratic throughput terms in Theorem 4. edu
Abstract. We derive an explicit bound for the L2-L2 gain of linear time-invariant
systems whose output is a quadratic function of the state and the input.Where the paper says it
Where the paper says it
Bilinear and Throughput Considerations
For systems with only bilinear outputs, i.e., y(t) = R(u(t) ⊗ x(t)), the L2-L2 gain can be bounded by a formula involving the transfer function and the L2-norm of the input. p. 0 Motivated by these applications and the relevance of L2
-L2 gains for robust and optimal
control [14, 17] as well as for model order reduction (MOR) [3], the central aim of this
manuscript is to derive a bound for the L2
-L2 gain of the system (1. It is well-known
that, for stable LTI systems, the L2
-L2 gain is bounded by the H∞-norm of the transfer
function [3]. Motivated by these applications and the relevance of L2
-L2 gains for robust and optimal
control [14, 17] as well as for model order reduction (MOR) [3], the central aim of this
manuscript is to derive a bound for the L2
-L2 gain of the system (1. It is well-known
that, for stable LTI systems, the L2
-L2 gain is bounded by the H∞-norm of the transfer
function [3].Where the paper says it
Where the paper says it
Conclusion and Implications
The first two results suggest that H2-oriented MOR algorithms from the LTI setting can be adapted for quadratic-output models yielding an optimal reduced order model in the sense that it minimizes the L2-error of the output relative to the L2-magnitude of the input. In turn, an adaptation of balanced truncation that yields a similar a priori bound as for the LTI case remains elusive since the inner product formulation is highly involved and does not allow for an immediate identification of a suitable Gramian related to the observability energy. p. 8 Inspired by the H2
-bounding balanced truncation algorithm for LQO systems
in [2], the Sylvester-based formulation in Theorem 3. 3 can yield an L2
-L2
-oriented BT in
case the observability energy can be expressed through the matrices Q and Yi
. 1 and Section 5 to finally summarize the results
for general systems including bilinear terms as well as the known results for LTI systems in
Theorem 6. The paper concludes with a brief numerical example in Section 7 and an
outlook in Section 8.Where the paper says it
Where the paper says it
Example Validation
A small numerical example illustrates the new bound with a state dimension n = 3. From here, we can compute the gain bound using the Sylvester-equation formulation. p. 1 1 that the L2
-L2
-gain bound can be computed using a suitable L2
-norm in
the frequency space of the bivariate transfer function. This L2
-formulation leads to a
corresponding inner product definition (see Definition 3. The main contributions of this paper are grouped into two stages: In the first stage,
we consider systems with M ̸= 0 but C = 0, D = 0, R = 0 and P = 0 and show in
Theorem 2. 1 that the L2
-L2
-gain bound can be computed using a suitable L2
-norm in
the frequency space of the bivariate transfer function. We derive an explicit bound for the L2-L2 gain of linear time-invariant
systems whose output is a quadratic function of the state and the input. Such systems
appear naturally in many areas, for example for port-Hamiltonian systems, optimalcontrol, and stochastic problems.Where the paper says it
Where the paper says it
Where the paper says it
References
[1] A. C. Antoulas, Approximation of large-scale dynamical systems, Society for Industrial and Applied Mathematics, 2005
[2] P. Benner, P. Goyal, and I. P. Duff, Gramians, energy functionals, and balanced truncation for linear dynamical systems with quadratic outputs, IEEE Trans. p. 8 Inspired by the H2
-bounding balanced truncation algorithm for LQO systems
in [2], the Sylvester-based formulation in Theorem 3. 3 can yield an L2
-L2
-oriented BT in
case the observability energy can be expressed through the matrices Q and Yi
.Where the paper says it
[3] P. Benner, W. Schilders, S. Grivet-Talocia, A. Quarteroni, G. Rozza, and L. M. Silveira, Volume 1: System- and Data-Driven Methods and Algorithms, Number 1 in Model Order Reduction
[4] C. Chen and R. F. Chiu, New theorems of association of variables in multiple dimensional Laplace transform, Internat. p. 1 4a)
ye(t1
, t2
) = Z t1
0
Z t2
0
h(τ1
, τ2
) (u(t1 − τ1
) ⊗ u(t2 − τ2
)) dτ1 dτ2
. 4b)
The idea in [6, 13] is to then apply the bivariate Laplace transform L2
[5] to the kernel
yielding the bivariate transfer function
Ge (s1
, s2
) = M
h
(s1I − A)
−1B ⊗ (s2I − A)
−1B
i
+ R
h
(s1I − A)
−1B ⊗ I
i
+ P ∈ C
p×m
2
.Where the paper says it
[5] L. Debnath, The double Laplace transforms and their properties with applications to functional, integral and partial differential equations, Internat. J. Appl. Comput. Math., 2(2):223–241, 2015
[6] A. N. Diaz, M. Heinkenschloss, I. V. Gosea, and A. C. Antoulas, Interpolatory model reduction of quadratic-bilinear dynamical systems with quadratic-bilinear outputs, Adv. p. 0 Such
quadratic outputs arise naturally in several application areas: In port-Hamiltonian systems
they reflect the input-energy relation [10], in linear-quadratic optimal control they are part
of the optimization objective [7], and in stochastic problems the variance is a quadratic
function of the states [9]. Motivated by these applications and the relevance of L2
-L2 gains for robust and optimal
control [14, 17] as well as for model order reduction (MOR) [3], the central aim of this
manuscript is to derive a bound for the L2
-L2 gain of the system (1.Where the paper says it
[7] J. Engwerda and A. Weeren, A result on output feedback linear quadratic control, Automatica, 44(1):265–271, 2008
[8] S. Fuller, B. Greiner, J. Moore, R. Murray, R. van Paassen, and R. Yorke, The Python control systems library (python-control). p. 0 1)
extend the standard linear time-invariant (LTI) systems to a mildly nonlinear form. Such
quadratic outputs arise naturally in several application areas: In port-Hamiltonian systems
they reflect the input-energy relation [10], in linear-quadratic optimal control they are part
of the optimization objective [7], and in stochastic problems the variance is a quadratic
function of the states [9].Where the paper says it
[9] B. Haasdonk, K. Urban, and B. Wieland, Reduced basis methods for parameterized partial differential equations with stochastic influences using the Karhunen–Loe‘ve expansion, SIAM-ASA J. Uncertain. Quantif., 1(1):79–105, 2013
[10] T. Holicki, J. Nicodemus, P. Schwerdtner, and B. Unger, Energy matching in reduced passive and port-Hamiltonian systems, SIAM J. p. 0 1)
extend the standard linear time-invariant (LTI) systems to a mildly nonlinear form. Such
quadratic outputs arise naturally in several application areas: In port-Hamiltonian systems
they reflect the input-energy relation [10], in linear-quadratic optimal control they are part
of the optimization objective [7], and in stochastic problems the variance is a quadratic
function of the states [9].Where the paper says it
[11] R. A. Horn and C. R. Johnson, Topics in Matrix Analysis, Cambridge University Press, 1991
[12] K. B. Petersen and M. S. Pedersen, The matrix cookbook, 2012 Version 20121115
[13] S. Reiter, I. V. Gosea, I. P. Duff, and S. Gugercin, H2-optimal model reduction of linear quadratic-output systems by multivariate rational interpolation, arXiv e-print 2505. p. 0 Such
quadratic outputs arise naturally in several application areas: In port-Hamiltonian systems
they reflect the input-energy relation [10], in linear-quadratic optimal control they are part
of the optimization objective [7], and in stochastic problems the variance is a quadratic
function of the states [9]. Motivated by these applications and the relevance of L2
-L2 gains for robust and optimal
control [14, 17] as well as for model order reduction (MOR) [3], the central aim of this
manuscript is to derive a bound for the L2
-L2 gain of the system (1.Where the paper says it
[14] A. van der Schaft, l2-gain analysis of nonlinear systems and nonlinear state feedback H∞ control, IEEE Trans. p. 0 1)
extend the standard linear time-invariant (LTI) systems to a mildly nonlinear form. Such
quadratic outputs arise naturally in several application areas: In port-Hamiltonian systems
they reflect the input-energy relation [10], in linear-quadratic optimal control they are part
of the optimization objective [7], and in stochastic problems the variance is a quadratic
function of the states [9].Where the paper says it
[15] H. K. Wimmer, Contour integral solutions of Sylvester-type matrix equations, Linear Algebra Appl. p. 6 Inner product and Sylvester-equation-based computation of the gain
bound
To obtain a computable expression for the output error between two LQO-systems we
first introduce a suitable inner product in Definition 3. 3 then shows how
this inner product can be evaluated by solving a set of Lyapunov and Sylvester matrix
equations.Where the paper says it
[16] Y. Zhang and N. Wong, Compact model order reduction of weakly nonlinear systems by associated transform, Int J. p. 0 This result provides a
practical tool for assessing and reducing quadratic-output models. Linear quadratic-output systems, L2-L2-gain, model order reduction
Mathematics subject classification.Where the paper says it
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Improved Model Order Reduction for Quadratic Output Systems: The paper provides a method to compute an explicit bound for quadratic-output systems by solving linear matrix equations, which offers a practical tool for assessing and reducing quadratic-output models. This allows AI systems to produce smaller, yet accurate, models of complex physical processes like port-Hamiltonian systems or optimal control problems.
-
Enhanced Robust Control Design: The derivation of the L2-L2 gain bound provides a quantifiable measure of system robustness against input disturbances. This can be used to design controllers that guarantee specific performance levels for systems where the output is quadratic in the state and input, improving stability and performance in uncertain environments.
-
Computation of Output Error Between Models: The paper introduces an inner product definition for two LQO systems, allowing the computation of the output error between them by solving a set of Lyapunov and Sylvester equations. This capability enables AI systems to quantify how much a reduced-order model deviates from the full system's behavior, which is crucial for model order reduction in control applications.
Abstract
We derive an explicit bound for the L2-L2-gain of linear time-invariant systems whose output is a quadratic function of the state and the input. Such systems appear naturally in many areas, for example for port-Hamiltonian systems, optimal-control, and stochastic problems. In case the output is purely quadratic in the state, the bound is determined by the supremum over ω in R of the L2-norms of the bivariate transfer function restricted to the anti-diagonals (s,, iω-s) s in i R of the frequency domain i R times i R. Further, we show how the bound can be computed by solving linear matrix equations. This result provides a practical tool for assessing and reducing quadratic-output models.
Sources
Related papers
- An Accelerated Proximal Gradient Method for Bures-Wasserstein Space
- FOMTools: Development and AI-Assisted Code Generation for First-Order Methods
- Lions and Muons: Optimization via Stochastic Frank-Wolfe under Heavy-Tailed Noise
- Dynamic Service Recommendation with Congestion-Dependent Joining: Near-Optimal and Constant-Factor Approximation Algorithms
- Integrating Learning and Optimization for Open Transport Capacity Pooling under Disruption
- Zeroth-Order Gradient Estimators under Decision-Dependent Noise