Bilinear quadratic output systems and balanced truncation
Heike Faßbender, Serkan Gugercin, Till Peters · 2025
Read Bilinear quadratic output systems and balanced truncation by Heike Faßbender, Serkan Gugercin, Till Peters (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-03684.txt.
Introduction
Heike Faßbender, Serkan Gugercin, and Till Peters extended a model reduction method to systems whose output follows a quadratic law, in a mathematics paper published in 2025. Their systems are bilinear, meaning nearly linear with a mild nonlinear term. Such systems can be huge, and a small model that behaves the same way is valuable. The classic reduction method weighs a system's reachability and observability maps to cut what is not needed.
The team set up paired forms of those maps for this class of systems and derived the laws they obey. They prove when the maps exist and are the unique fit. Reduced forms of the maps, which cost less to build, feed the reduction as well. Two worked examples show that the reduced models track the true systems closely. The work gives control mathematics a clean base for this kind of output.
Catalog record
| CREATOR | Heike Faßbender, Serkan Gugercin, Till Peters |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 79 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-03684
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Bilinear quadratic output systems and balanced truncation Thanks: Submitted to the editors DATE. Heike Faßbender, Serkan Gugercin, Till Peters Abstract Dynamical systems with quadratic outputs have recently attracted significant attention. In this paper, we consider bilinear dynamical systems, a special class of weakly nonlinear systems, with a quadratic output. We develop various primal-dual formulations for these systems and define the corresponding system Gramians. Conditions for the existence and uniqueness of these Gramians are established, and the generalized Lyapunov equations they satisfy are derived. Using these Gramians and their truncated versions, which are computationally more efficient, we construct a balanced truncation framework for bilinear systems with quadratic outputs. The proposed approach is demonstrated through two numerical examples. keywords Model order reduction, balanced truncation, bilinear systems, linear quadratic output systems, Gramians, generalized Lyapunov equations Funding. Part of this work was completed during a research stay of the third author at Virginia Tech supported by a fellowship of the German Academic Exchange Service (DAAD). The work of Gugercin was supported in part by the National Science Foundation (NSF), United States under Grant No. CMMI-2130695 † † runningheads: BQO systems and balanced truncation / H. Faßbender, S. Gugercin and T. Peters MSC 93A15, 93B05, 93B07, 93C10, 93C15 1 Introduction We investigate the so-called bilinear quadratic output (BQO) dynamical systems described in the state-space form as \displaystyle\dot{x}(t) \displaystyle=Ax(t)+\sum_{k=1}^{m}N_{k}x(t)u_{k}(t)+Bu(t),\qquad x(0)=0, (1a) \displaystyle y(t) \displaystyle=Cx(t)+\begin{bmatrix}x(t)^{T}M_{1}x(t)\\ \vdots\\ x(t)^{T}M_{p}x(t)\end{bmatrix}, (1b) where A\in\mathbb{R}^{n\times n},B\in\mathbb{R}^{n\times m},C\in\mathbb{R}^{p\times n},N_{k}\in\mathbb{R}^{n\times n} for k=1,\dots,m , M_{j}\in\mathbb{R}^{n\times n} for j=1,\dots,p , and t\in T:=[0,\infty) . In ( 1 ), x(t)\in\mathbb{R}^{n} describes the state, u(t)\in\mathbb{R}^{m} the inputs, and y(t)\in\mathbb{R}^{p} the outputs of the system. It can be assumed without loss of generality that M_{j} is symmetric since x(t)^{T}M_{j}x(t)=x(t)^{T}M_{j}^{T}x(t)=\frac{1}{2}x(t)^{T}(M_{j}+M_{j}^{T})x(t) . The BQO systems represent a generalization of bilinear systems (see, e.g., [ 2 , 42 ] ) as well as of linear quadratic output (LQO) systems (see, e.g., [ 11 ] ). More specifically, in eq. 1b , setting M_{1}=\dots=M_{p}=0 leads to the output y(t)=Cx(t) , recovering the bilinear dynamical systems. On the other hand, setting N_{1}=\dots=N_{m}=0 in eq. 1a leads to the state equation \dot{x}(t)=Ax(t)+Bu(t) , recovering the linear quadratic output (LQO) systems. Bilinear state equations naturally appear in various applications or a result of Carleman bilinearization of nonlinear dynamics, see, e.g., [ 33 , 25 , 21 ] . In recent years, dynamical systems with quadratic outputs as in eq. 1b have gained significant attention as they appear in the modeling of the variance of a quantity of interest of a stochastic model or in measuring the energy of state; see, e.g., [ 30 , 29 , 41 , 32 , 38 ] and the references therein. Moreover, in the context of port-Hamiltonian dynamical systems, bilinear port-Hamiltonian dynamical systems naturally appear with a state-space form \displaystyle\dot{x}(t) \displaystyle=(J-R)Qx(t)+\left(B+Nx(t)\right)u(t), (2a) \displaystyle y(t) \displaystyle=\left(B+Nx(t)\right)^{T}Qx(t), (2b) where J,R,Q,N\in\mathbb{R}^{n\times n},B\in\mathbb{R}^{n\times 1} with J=-J^{T},R=R^{T}\geq 0,Q=Q^{T}\geq 0 , see [ 24 , 39 ] . (For simplicity, here we only considered a single-input single-output bilinear port-Hamiltonian system.) Note that eq. 2 is a special case of eq. 1 . In most cases, such as those where the dynamical system arises from the spatial discretization of an underlying PDE, the dimension of the resulting BQO system is significantly large, and the task of model order reduction arises. Thus we seek a BQO system of similar structure but with a significantly smaller dimension which should accurately describe the same input-output behavior. More precisely, we look for the reduced BQO system of order r\ll n described as \displaystyle\dot{\hat{x}}(t) \displaystyle=\hat{A}\hat{x}(t)+\sum_{k=1}^{m}\hat{N}_{k}\hat{x}(t)u_{k}(t)+\hat{B}u(t),\qquad\hat{x}(0)=0, (3a) \displaystyle\hat{y}(t) \displaystyle=\hat{C}\hat{x}(t)+\begin{bmatrix}\hat{x}(t)^{T}\hat{M}_{1}\hat{x}(t)\\ \vdots\\ \hat{x}(t)^{T}\hat{M}_{p}\hat{x}(t)\end{bmatrix}, (3b) where \hat{A}\in\mathbb{R}^{r\times r},\hat{B}\in\mathbb{R}^{r\times m},\hat{C}\in\mathbb{R}^{p\times r} , \hat{N}_{k}\in\mathbb{R}^{r\times r} for k=1,\dots,m , \hat{M}_{j}\in\mathbb{R}^{r\times r} for j=1,\dots,p , t\in T . In eq. 3 , \hat{x}(t)\in\mathbb{R}^{r} describes the reduced state and \hat{y}(t)\in\mathbb{R}^{p} the reduced output of the system. The goal is to construct eq. 3 such that the reduced output \hat{y}(t) provides a high-fidelity approximation to the original output y(t) . In this manuscript, we employ balanced truncation (BT), a model order reduction method that is well established for linear systems (originally in [ 26 , 27 ] , see also [ 3 , 6 , 13 , 20 ] ), or [ 9 , Section 7 and 8] ). BT transforms the original system into a state-space representation where the reachability and observability Gramians are equal and diagonal. The diagonal elements are called the Hankel singular values. The states corresponding to the smallest Hankel singular values are then removed because they have the least influence on the system’s behavior. In general, this results in a reduced model that provides a good approximation of the original system with a lower dimension. Theoretical foundation of balancing and balanced truncation for nonlinear systems has been introduced in [ 35 ] which requires solving Hamilton-Jacobi partial differential equations. This task is computationally challenging for large-scale systems and the reduced system, in general, will not inherit the original structure due to the state-dependent model reduction bases. For example, in our setting, this means that the reduced system will not retain the structure eq. 3 . As an alternative, one seeks algebraic Gramians, which are generalizations of the linear case and solve generalized Lyapunov equations. One of the main contributions of this work will be the development of proper and relevant algebraic Gramians for BQO systems. BQO systems and their model order reduction have been briefly discussed in the master’s thesis [ 28 , pp. 25–42] . In particular, [ 28 , Section 3.1] highlights a transformation of BQO systems into bilinear systems with linear output via lifting transformations. While this approach enables the application of existing bilinear system theory and reduction techniques, it comes at the significant cost of increasing the system matrix size from n\times n to (n+n^{2})\times(n+n^{2}) - an increase in the dimension that is clearly impractical for large-scale systems. Therefore, instead, [ 28 ] avoids this lifting technique and aims to develop reduction tools directly on the original BQO structure. To achieve this goal in the BT setting, one needs the Gramians. In [ 28 ] , the reachability Gramian of ( 1 ) is taken to be the same as in the bilinear case, that is, ( 1 ) with M_{1}=\dots=M_{p}=0 . Similarly, the construction of the observability Gramian of ( 1 ) is motivated by the observability Gramian for the bilinear case. The analysis in [ 28 ] is restricted to single-input single-output (SISO) configurations ( m=p=1 ). In contrast, our work addresses the more general case of multi-input multi-output (MIMO) BQO systems and focuses on developing the reachability and observability Gramians following the approach for quadratic bilinear systems from [ 10 ] . In quadratic bilinear systems, ( 1a ) has an additional quadratic summand H(x(t)\otimes x(t)), while in ( 1b ) all matrices M_{j},j=1,\dots,p are zero. Therefore, with H=0 , the derivation of the reachability Gramian given in [ 10 ] can be adopted here. In [ 10 , Section 2.2] , it is suggested to define the observability Gramian for quadratic bilinear systems as the reachability Gramian of the dual system , analogous to what is known for linear and bilinear systems. We will follow this approach in our derivations. As we will show in Section 3.2 , there