Abstract:<p>We consider fast solving a class of spatial fractional diffusion equations where the fractional differential operators are comprised of Riemann–Liouville and Caputo fractional derivatives. A circulant-based approximate inverse preconditioner is established for the discrete linear systems resulted from the finite difference discretization of this kind of fractional diffusion equations. By sufficiently exploring the Toeplitz-like structure and the rapid decay properties of the internal sub-matrices in the coefficient matrix, we show that the spectrum of the preconditioned matrix is clustered around one. Numerical experiments are performed to demonstrate the effectiveness of the proposed preconditioner.</p>
What problem does this paper attempt to address?
The problem that this paper attempts to solve is how to effectively improve computational efficiency and numerical stability when solving a class of spatial fractional - order diffusion equations. Specifically, the paper focuses on spatial fractional - order diffusion equations containing Riemann - Liouville and Caputo fractional - order derivatives. After discretization, such equations will generate full - rank linear systems, resulting in a computational cost of direct solution methods as high as \( \mathcal{O}(n^3) \) and a storage requirement of \( \mathcal{O}(n^2) \), where \( n \) is the number of grid points. Therefore, the key to solving this problem lies in designing effective pre - processing techniques to accelerate the iterative solution process.
To meet this challenge, the paper proposes a Circulant - based Approximate Inverse Preconditioner. By making full use of the Toeplitz - like structure of the coefficient matrix and the fast - decay property of the internal sub - matrices, the paper proves that the spectrum of the pre - processed matrix is clustered around 1. Numerical experiments also verify the effectiveness of the proposed pre - conditioner, indicating that it can significantly accelerate the convergence rate of the Krylov subspace method, thereby efficiently solving the discrete linear systems of large - scale fractional - order diffusion equations.
### Specific problems and solutions
1. **Problem background**:
- **Fractional - order diffusion equations**: Such equations can better describe many non - local behaviors and have more extensive applications compared to integer - order diffusion equations, such as in the fields of physics, control systems, chemistry, biology, finance, and image restoration.
- **Computational bottleneck**: Due to the non - local nature of fractional - order derivatives, the linear systems after discretization are usually full - rank, resulting in very high computational costs and storage requirements for direct solution methods.
2. **Solutions**:
- **Pre - processing techniques**: The paper proposes a Circulant - based Approximate Inverse Preconditioner, and by using the Toeplitz - like structure of the coefficient matrix and the fast - decay property of the internal sub - matrices, an efficient pre - conditioner is designed.
- **Theoretical analysis**: By fully exploring and using the diagonal dominance and fast - decay properties of the coefficient matrix, it is proved that the spectrum of the pre - processed matrix is clustered around 1, which helps to accelerate the convergence of the Krylov subspace method.
- **Numerical experiments**: Through numerical experiments, the effectiveness of the proposed pre - conditioner is verified, showing its superior performance in solving large - scale fractional - order diffusion equations.
### Main contributions
- **Efficient pre - conditioner design**: A Circulant - based Approximate Inverse Preconditioner is proposed, which can significantly reduce the computational cost of solving large - scale fractional - order diffusion equations.
- **Theoretical analysis**: Through strict mathematical analysis, it is proved that the spectrum of the pre - processed matrix is clustered around 1, providing theoretical support for the effectiveness of the pre - conditioner.
- **Numerical verification**: Through numerical experiments, the actual effect of the pre - conditioner is verified, showing its feasibility and advantages in practical applications.
In summary, through designing an efficient pre - conditioner, this paper solves the problems of computational efficiency and numerical stability when solving fractional - order diffusion equations, providing strong support for research and applications in related fields.