Abstract:We propose a quadrature-based formula for computing the exponential function of matrices with a non-oscillatory integral on an infinite interval and an oscillatory integral on a finite interval. In the literature, existing quadrature-based formulas are based on the inverse Laplace transform or the Fourier transform. We show these expressions are essentially equivalent in terms of complex integrals and choose the former as a starting point to reduce computational cost. By choosing a simple integral path, we derive an integral expression mentioned above. Then, we can easily apply the double-exponential formula and the Gauss-Legendre formula, which have rigorous error bounds. As numerical experiments show, the proposed formula outperforms the existing formulas when the imaginary parts of the eigenvalues of matrices have large absolute values.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is: how to calculate the exponential function of a matrix without using oscillatory integrals on infinite intervals. Specifically, the authors propose a method based on quadrature formulas for calculating the exponential function $\exp(A)$ of a matrix $A\in\mathbb{C}^{m\times m}$, where the eigenvalues of $A$ have negative real parts.
### Background and Problem Description
In scientific computing, the matrix exponential function $\exp(A)$ appears in many problems, such as the numerical solution of ordinary differential equations, control theory, Markov chain processes, and directed network analysis. Traditional calculation methods (such as Schur decomposition, Padé approximation, Newton's method, etc.) are effective for medium - sized matrices, but not necessarily efficient for large - scale matrices and are difficult to parallelize.
Existing methods based on quadrature formulas mainly rely on integral representations of the inverse Laplace transform or Fourier transform. Although these methods are effective, they require the calculation of oscillatory integrals on infinite intervals, which is numerically challenging, especially when the imaginary parts of the eigenvalues are large.
### Main Contributions of the Paper
To solve the above problems, this paper proposes a new quadrature formula that avoids oscillatory integrals on infinite intervals. Specifically, the authors achieve this goal through the following steps:
1. **Derivation of Non - oscillatory Integral Formula**: By choosing a specific integration path, the authors derive a formula that includes non - oscillatory integrals on infinite intervals and oscillatory integrals on finite intervals.
2. **Application of Numerical Integration Methods**: For non - oscillatory integrals on infinite intervals, the authors apply the double - exponential (DE) formula; for oscillatory integrals on finite intervals, they apply the Gauss - Legendre quadrature formula.
3. **Strict Error Bounds**: Through strict mathematical derivations, the authors obtain the error bounds of these two quadrature methods, ensuring the reliability of the numerical results.
### Mathematical Formulas
The key formulas in this paper include:
- Definition of the matrix exponential function:
\[
\exp(A)=\sum_{k = 0}^{\infty}\frac{1}{k!}A^{k}
\]
- Proposed new integral formula:
\[
\exp(z)=\frac{1}{2\pi i}\int_{0}^{\infty}\left(\frac{e^{i\alpha}}{z - i\alpha+x}-\frac{e^{-i\alpha}}{z + i\alpha+x}\right)e^{-x}\,dx+\frac{\alpha}{2\pi}\int_{- 1}^{1}\frac{e^{i\alpha x}}{i\alpha x - z}\,dx
\]
- DE formula applied to non - oscillatory integrals:
\[
\tilde{I}_{\alpha,n,h}(z):=h\sum_{k = - n}^{n}f_{\alpha}(z,\phi(kh))\phi'(kh)
\]
- Gauss - Legendre quadrature formula applied to oscillatory integrals:
\[
\tilde{J}_{\alpha,N}(z):=\sum_{i = 1}^{N}w_{i}g_{\alpha}(z,t_{i})
\]
### Conclusion
Through numerical experiments, the authors verify that the proposed method has better convergence and accuracy than existing methods when dealing with matrices with large imaginary parts of eigenvalues. In addition, this method is easy to parallelize and is suitable for the calculation of large - scale matrices.
In summary, this paper solves the key problem of how to avoid oscillatory integrals on infinite intervals when calculating the matrix exponential function and provides an efficient and reliable numerical method.