An efficient and fast parallel method for Volterra integral equations of Abel type (Q818204): Difference between revisions

For the solution of Volterra integral equations of Abel type as \[ y(t)=f(t) + \int\limits^t_0 \frac{A}{(t-s)^\alpha}y(s)\,ds, \quad t\in [0,T], \quad A \in \mathbb R^{n \times n}, \quad n \geq q, \tag{1} \] where the matrix \(A\) has eigenvalues laying in the open left half complex plan, well known nonstationary Richardson waveform relaxation method which has the form \[ y^{(i+1)}(t) + \int\limits^t_0 \frac{\mu_i}{(t-s)^\alpha}Iy^{(i+1)}(s)ds + \int\limits^t_0 \frac{(A-\mu_i I)}{(t-s)^\alpha}y^{(i)}(s)\,ds \] can be applied. Here \(\mu_i \in R,\) I is the identity matrix of order \(n.\) The authors present an efficient and fast parallel waveform relaxation method for Volterra integral equations of Abel type, obtained by reformulating a nonstationary waveform relaxation method for systems of equations with linear coefficient constant kernel as (1). To this aim they consider the Laplace transform of the equation and apply the recurrence relation given by the Chebyshev polynomial acceleration for linear algebraic systems. Back in the time domain, they obtain a three term recursion which requires, at each iteration, the evaluation of convolution integrals, where only the Laplace transform of the kernel is known. For this calculation one can use a fast convolution algorithm. Numerical experiments are done also on problems where it is not possible to use the original nonstationary method, obtaining good results in terms of improvement of the rate of convergence with respect the stationary method.