Weighted balanced truncation method for approximating kernel functions by exponentials.

Area of Science:

• Computational mathematics
• Scientific computing
• Numerical analysis

Background:

• Kernel approximation using exponentials is crucial for solving problems involving convolution quadrature and particle interactions.
• Applications include integral-differential equations, molecular dynamics simulations, and machine learning algorithms.
• Accurate kernel representation is essential for the efficiency and reliability of these computational methods.

Purpose of the Study:

• To introduce a novel weighted balanced truncation method for kernel approximation.
• To reduce the number of exponential terms required for accurate kernel representation.
• To enhance the approximation of long-range kernels.

Main Methods:

• Development of a weighted balanced truncation algorithm.
• Application of the method to approximate Ewald splitting and inverse power kernels.
• Comparison of the new method against classical balanced truncation.

Main Results:

• The weighted balanced truncation method significantly reduces the number of required exponential terms.
• Achieved over four digits of accuracy improvement for Ewald splitting and inverse power kernels.
• Demonstrated superior performance compared to classical balanced truncation for long-range kernels.

Conclusions:

• The weighted balanced truncation method offers a more efficient and accurate approach to kernel approximation.
• The method shows significant promise for practical applications in scientific computing.
• This advancement can lead to faster and more reliable simulations in various fields.