Error estimates for finite element methods for a wide-angle parabolic equation (Q1344322)

The topic of this paper is the analysis of convergence of a finite difference-finite element method for solving the model third-order initial-boundary value problem \[ [1 + \sigma(\beta(z,r) + i\nu(z,r))]u_ r + \alpha u_{zzr} = i\alpha_{xx} u_{zz} + i[\beta(z,r) + i\nu(z,r)]u. \] Here \(u(z,r)\) is a complex-valued function defined on a rectangle, \(\alpha\) and \(\sigma\) are real constants, and \(\beta\) and \(\nu\) are smooth, real-valued functions. This partial differential equation occurs in problems of wave propagation as a wide-angle, parabolic approximation to the Helmholtz equation in cylindrical coordinates; \(r\) represents the range and \(z\) is the depth variable. The paper is divided into two main parts: the first deals with semidiscrete Galerkin approximations with respect to the variable \(z\). For this purpose a weak formulation is constructed. The main result of this section is one giving the \(L^ 2\) error estimate \[ \max_{0 \leq r \leq R} \| u(r) - u_ h(r)\| \leq ch^ s \] for the case in which polynomials of degree \(s-1\) are used. The second objective of the paper is to examine fully discrete approximations, in which a finite difference approach with respect to the variable \(r\) is used. Two possibilities are examined: one is a Crank- Nicolson approximation, while the other is a fourth-order Runge-Kutta scheme. These two schemes together with the finite element approximation with respect to \(z\), yield optimal estimates for the error which are respectively \(O(h^ s + k^ 2)\) and \(O(h^ s + k^ 4)\). Both schemes are unconditionally stable. The authors also propose an algorithm for implementation of the Runge- Kutta scheme, but do not give any numerical results.