A new efficient algorithm for computational aeroacoustics on parallel processors

One of the great challenges in computational physics is the prediction of flow associated noise, where the quantities of interest, namely the sound waves can be at high frequencies and are usually orders of magnitude smaller in magnitude than the mean quantities. In order to numerically resolve such small scales governed by the fluid dynamics equations, high resolution schemes are required. Thus solutions of flow noise problems are computationally intensive. An efficient, hybrid, data parallel computational aeroacoustics algorithm has been developed for the prediction of noise radiation and scattering from three-dimensional geometries. The algorithm solves the Euler/Navier-Stokes equations in the interior and nonreflecting boundary conditions on the outer boundaries. A moving surface Kirchhoff method is coupled to the flow solver for far-field predictions. The algorithm uses standard time and spatial discretization techniques but utilizes several new optimization strategies that are highly suitable for single zone solutions on data parallel processors. One strategy, for example, enables simultaneous residual evaluations of the interior and far-field nonreflecting boundary conditions equations, reducing the computational effort spent on them by approximately 60% CPU time savings. The algorithms for the flow solver and the Kirchhoff method and their coupling are described in this paper, and results for some example radiation and scattering problems are presented.