Discrete transparent boundary conditions for wide angle parabolic equations in underwater acoustics (Q1268361): Difference between revisions

The objective is to construct discrete transparent boundary condition (DTBC) for a Crank-Nicolson finite difference discretization of the wide angle parabolic equation (WAPE) in the application to underwater acoustics. The standard strategy in solving underwater acoustic problem was to introduce rather thick absorbing layers below the sea bottom, and then to limit the depth-range by imposing again a Dirichlet boundary condition. Absorbing layer increases the computational cost, it is also hard to estimate the change in the real solution due to the introduction of absorbing layer. Transparent boundary condition (TBC) is used to overcome the shortcoming. The basic idea is to explicitly solve the equation in the bottom region for decaying solution, which is the exterior of the computational domain \((0,z_b)\). The TBC is nonlocal in \(r\); thus in range-marching algorithms this condition requires storing the bottom boundary data of all previous range levels. The coupling of WAPE for the ocean \((0< z< z_b)\) to the standard parabolic equation (SPE) for the sea bottom \((z>z_b)\) is performed due to the TBC corresponding to the SPE, this is equivalent to the half-space problem. While the TBC fully solves the truncated problem, numerical discretization is difficult and less accurate, and the overall numerical scheme is only conditionally stable. In this paper, a new discrete TBC is derived from the fully discretized half-space problem. It is of discrete convolution form, involving the boundary data from the whole ``past range''. The crucial point is to find the inverse \(Z\)-transformation explicitly for the \(Z\)-transformed DTBC. In the derivation of the DTBC, a uniform \(z\)-discretization for the exterior problem \((z> z_b)\) is needed. For the interior problem, however, a nonuniform discretization may be used, and this would not change the DTBC. The DTBC is reflection-free and yields an unconditionally stable scheme. Numerical results are presented on several bench-mark problems to compare with other discretization solutions; these results show superiority of the new scheme.