Abstract:The AAA algorithm has become a popular tool for data-driven rational approximation of single variable functions, such as transfer functions of a linear dynamical system. In the setting of parametric dynamical systems appearing in many prominent applications, the underlying (transfer) function to be modeled is a multivariate function. With this in mind, we develop the AAA framework for approximating multivariate functions where the approximant is constructed in the multivariate barycentric form. The method is data-driven, in the sense that it does not require access to full state-space model and requires only function evaluations. We discuss an extension to the case of matrix-valued functions, i.e., multi-input/multi-output dynamical systems, and provide a connection to the tangential interpolation theory. Several numerical examples illustrate the effectiveness of the proposed approach.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is how to effectively approximate multivariate functions when performing data - driven modeling in parameterized dynamical systems. Specifically, the paper proposes a method, the p - AAA algorithm, which is an extension of the AAA algorithm, for approximating the transfer functions of parameterized dynamical systems. This method is especially suitable for cases where the internal description is unknown and only input / output measurement data are available. By using these data, the p - AAA algorithm can construct a low - dimensional model that can well approximate the original system, thereby significantly reducing the computational cost while maintaining accuracy.
### Background and Motivation
Many physical phenomena can be modeled by dynamical systems that depend on one or more parameters. These parameters can represent material properties, boundary conditions, system geometries, etc. For example, consider an input - output system described by a system of linear ordinary differential equations:
\[
\dot{x}(t, p)=A(p)x(t, p)+bf(t); \quad y(t, p)=c^{\top}x(t, p),
\]
where \( p\in P\subset\mathbb{R} \) represents parameter variation, \( A(p)\in\mathbb{R}^{\rho\times\rho} \), \( b, c\in\mathbb{R}^{\rho} \), \( f(t)\in\mathbb{R} \) is the input (forcing term), \( y(t, p)\in\mathbb{R} \) is the output (quantity of interest), and \( x(t, p)\in\mathbb{R}^{\rho} \) is the state (internal degrees of freedom). Assuming the initial condition is zero, i.e., \( x(0) = 0 \), the output \( y(t, p) \) can be expressed as a convolution integral:
\[
y(t, p)=\int_{0}^{t}c^{\top}e^{(t - \tau)A(p)}bf(\tau)d\tau.
\]
When the system dimension \( \rho \) is large, repeatedly calculating the quantity of interest \( y(t, p) \) for different parameter values becomes computationally very expensive. One way to solve this problem is to find a low - dimensional surrogate model, i.e., a reduced - order dynamical system, such that the computational cost of re - evaluating the system is significantly reduced while still accurately capturing \( y(t, p) \). This is the goal of parameterized model order reduction (PMoR).
### Method Overview
The paper proposes the p - AAA algorithm, which is based on the AAA algorithm and extended to the multivariate case for approximating the transfer functions of parameterized dynamical systems. The specific steps are as follows:
1. **Data Partitioning**: Divide the sampling points and the corresponding function values into two parts, one part for interpolation and the other part for least - squares fitting.
2. **Interpolation**: At the selected interpolation points, ensure that the approximate function \( \tilde{H}(s, p) \) is equal to the true function \( H(s, p) \).
3. **Least - Squares Fitting**: On the remaining data points, select coefficients by minimizing the error to make \( \tilde{H}(s, p) \) as close as possible to \( H(s, p) \).
4. **Iterative Update**: Select the next interpolation point through a greedy search, update the data partitioning, and recalculate the coefficients until the preset error tolerance or the required model order is reached.
### Applications and Advantages
The p - AAA algorithm is not only applicable to the modeling of parameterized dynamical systems but can also be applied to other multivariate function approximation problems, such as the modeling of static partial differential equations. The advantage of this method lies in its data - driven nature; it does not require a complete state - space model of the system, only function value samples are needed. In addition, this method has no restrictions on the type of parameter dependence and can flexibly handle parameter dependence in different types of system matrices.
Through numerical experiments, the paper demonstrates the effectiveness of the p - AAA algorithm and proves its superior performance in dealing with high - dimensional parameterized dynamical systems.