Abstract:When $0<\alpha<1$, the approximation for the Caputo derivative $$y^{(\alpha)}(x) = \frac{1}{\Gamma(2-\alpha)h^\alpha}\sum_{k=0}^n \sigma_k^{(\alpha)} y(x-kh)+O\bigl(h^{2-\alpha}\bigr),$$ where $\sigma_0^{(\alpha)} = 1, \sigma_n^{(\alpha)} = (n-1)^{1-a}-n^{1-a}$ and $$\sigma_k^{(\alpha)} = (k-1)^{1-\alpha}-2k^{1-a}+(k+1)^{1-\alpha},\quad (k=1...,n-1),$$ has accuracy $O\bigl(h^{2-\alpha}\bigr)$. We use the expansion of $\sum_{k=0}^n k^\alpha$ to determine an approximation for the fractional integral of order $2-\alpha$ and the second order approximation for the Caputo derivative $$y^{(\alpha)}(x) = \frac{1}{\Gamma(2-\alpha)h^\alpha}\sum_{k=0}^n \delta_k^{(\alpha)} y(x-kh)+O\bigl(h^{2}\bigr),$$ where $\delta_k^{(\alpha)} = \sigma_k^{(\alpha)}$ for $2\leq k\leq n$, $$\delta_0^{(\alpha)} = \sigma_0^{(\alpha)}-\zeta(\alpha-1), \delta_1^{(\alpha)} = \sigma_1^{(\alpha)}+2\zeta(\alpha-1),\delta_2^{(\alpha)} = \sigma_2^{(\alpha)}-\zeta(\alpha-1),$$ and $\zeta(s)$ is the Riemann zeta function. The numerical solutions of the fractional relaxation and subdiffusion equations are computed.
What problem does this paper attempt to address?
The main problem that this paper attempts to solve is to improve the accuracy of the numerical approximation of the Caputo fractional - order derivative, especially when the fractional - order derivative order \(\alpha\) is close to 1. Specifically:
- **Existing problems**: For the case of the fractional - order derivative \(0 < \alpha < 1\), the commonly used Caputo fractional - order derivative discretization methods (such as formula (1)) will have the accuracy degraded to \(O(h)\) when \(\alpha\) is close to 1, which leads to a large amount of calculation and insufficient accuracy when solving multi - dimensional fractional - order partial differential equations.
- **Research objective**: By improving the existing approximation methods, a new approximation method with second - order accuracy \(O(h^2)\) is proposed, thereby improving the accuracy and efficiency of the numerical solution, especially in the case where \(\alpha\) is close to 1.
### Main content
1. **Background introduction**:
- The paper first introduces the definition of the Caputo fractional - order derivative and its importance in modeling complex diffusion processes.
- The first - order approximation formula of the commonly used Caputo fractional - order derivative is proposed, and its limitations are pointed out.
2. **Theoretical derivation**:
- Using the power series expansion and the Riemann zeta function, the author derives a new second - order approximation formula (formula (4)), which improves the accuracy by adjusting the first three coefficients.
- The second - order approximation of the fractional - order integral is derived in detail, and the second - order approximation of the Caputo fractional - order derivative is obtained by using these results.
3. **Numerical experiments**:
- The effectiveness of the new method is verified through numerical experiments, especially when \(\alpha\) is close to 1, the new method significantly improves the accuracy of the numerical solution.
- The performance of different approximation methods in the fractional - order relaxation equation and the time - fractional - order sub - diffusion equation is compared, showing the advantages of the new method.
### Conclusion
The paper successfully proposes a new second - order approximation method, which effectively solves the problem of the decrease in the approximation accuracy of the Caputo fractional - order derivative when \(\alpha\) is close to 1. This improvement not only improves the accuracy of the numerical solution, but also reduces the amount of calculation, providing a more effective tool for solving complex fractional - order differential equations.
### Formula summary
- **First - order approximation formula**:
\[
y^{(\alpha)}_n \approx \frac{1}{\Gamma(2-\alpha) h^\alpha} \sum_{k = 0}^n \sigma_k^{(\alpha)} y_{n - k}+O(h^{2-\alpha})
\]
where,
\[
\sigma_0^{(\alpha)}=1,\quad \sigma_n^{(\alpha)}=(n - 1)^{1-\alpha}-n^{1-\alpha},\quad \sigma_k^{(\alpha)}=(k + 1)^{1-\alpha}-2k^{1-\alpha}+(k - 1)^{1-\alpha}
\]
- **Second - order approximation formula**:
\[
y^{(\alpha)}_n \approx \frac{1}{\Gamma(2-\alpha) h^\alpha} \sum_{k = 0}^n \delta_k^{(\alpha)} y_{n - k}+O(h^2)
\]
where,
\[
\delta_0^{(\alpha)}=\sigma_0^{(\alpha)}-\zeta(\alpha - 1),\quad \delta_1^{(\alpha)}=\sigma_1^{(\alpha)}+2\zeta(\alpha - 1),\quad \delta_2^{(\alpha)}=\sigma_2^{(\alpha)}-\zeta(\alpha - 1),\quad \delta_k^{(\alpha)}=\sigma_k^{(\alpha)}