Abstract:The Legendre-based ultraspherical spectral method for ordinary differential equations is combined with a formula for the convolution of two Legendre series to produce a new technique for solving linear Fredholm and Volterra integro-differential equations with convolution-type kernels. When the kernel and coefficient functions are sufficiently smooth then the method is spectrally-accurate and the resulting almost-banded linear systems can be solved with linear complexity.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is: to develop an efficient and accurate method for solving linear Fredholm and Volterra integro - differential equations (IDE) with convolution - type kernels. Specifically, these equations are in the following forms:
For the Fredholm integro - differential equation:
\[ L_r y(t) = f(t) + g(t) \int_0^T k(t - s) h(s) y(s) \, ds, \quad 0 \leq t \leq T, \tag{1.1} \]
For the Volterra integro - differential equation:
\[ L_r y(t) = f(t) + g(t) \int_0^t k(t - s) h(s) y(s) \, ds, \quad 0 \leq t \leq T, \tag{1.2} \]
where \(L_r\) is a linear differential operator in the form of:
\[ L_r y(t) = a_r(t) y^{(r)}(t) + a_{r - 1}(t) y^{(r - 1)}(t)+\cdots + a_0(t) y(t), \]
and \(k(t)\) is the convolution kernel, and \(g(t)\) and \(h(t)\) are given smooth functions.
### Key challenges in solving the problem
1. **Infeasibility of analytical solutions**: In practical applications, analytical solutions (i.e., closed - form solutions) are usually not obtainable for this type of equations, so numerical approximation methods need to be relied on.
2. **Computational complexity**: Traditional global collocation methods (such as the collocation method on Chebyshev or Legendre points) can provide exponential convergence and high - precision solutions, but they will lead to dense matrices, and solving these matrices is very time - consuming. Moreover, when the differential order \(r\) is large, the matrix representing the differential operator is usually ill - conditioned, which further increases the difficulty of solution.
### The method proposed in the paper
The paper proposes an ultraspherical spectral method based on Legendre polynomials, combined with the discretization technique of the convolution formula, to efficiently solve the above two types of equations. The specific steps are as follows:
1. **Legendre expansion**: Expand the unknown solution \(y(t)\) and the kernel function \(k(t)\) into Legendre polynomial series.
2. **Discretization**: Use the ultraspherical spectral method to discretize the differential operator to obtain a sparse (almost banded) and well - conditioned matrix.
3. **Convolution operation**: Introduce the convolution formula to discretize the convolution term into a banded matrix, thereby maintaining the sparsity of the entire system.
4. **Linear system solution**: Finally, form an almost banded linear system, which can be efficiently solved by fast algorithms (such as Schur complement decomposition or Woodbury formula).
### Advantages of the method
- **Spectral accuracy**: For sufficiently smooth kernel functions and coefficient functions, this method has spectral accuracy, that is, as the expansion order increases, the error decreases exponentially.
- **Linear complexity**: Since the generated matrices are almost banded, the solution process has linear complexity and is suitable for large - scale problems.
In summary, this paper aims to develop an efficient, accurate and stable numerical method for solving linear Fredholm and Volterra integro - differential equations with convolution - type kernels by combining Legendre polynomials and the ultraspherical spectral method.