Project Summary  -  (a similar description can be found here)

Title of Project:  Numerical Simulation of Supersonic Jet Noise using High-order Shock Capturing Schemes and Realization of Peak Computational Performanc in Parallel Computations

Sponsor: Computational Research Institute (CRI) and Purdue Research Foundation (PRF)

Project PIs:    G. Blaisdell (PI), A. Lyrintzis (co-PI), A. Shame (co-PI)

Project abstract:  

Supersonic jet noise is a critical factor in the design of modern commercial and military aircraft.  The proposed project would help develop computational tools needed to increase our level of understanding of supersonic jet noise.  There are two specific goals for this project.  The first is to extend the capability of our large eddy simulation (LES) code so that it can compute noise from supersonic jets.  Over the past several years we have developed a methodology for simulating subsonic turbulent jets and predicting the far field noise produced. Extension of our work to include supersonic jets is difficult becausesuch flows contain both turbulence and shock waves.  Turbulence consists of a wide range of length scales, and high-order methods are required in order to capture the widest possible range of length scales accurately.  However, in the presence of shock waves such methods typically develop oscillations, which can lead to inaccuracies, spurious physical effects, and numerical instabilities.  There has been recent work on the development of high-order shock capturing numerical methods that can accurately resolve a wide range of length scales while providing the nonlinear dissipation needed to give smooth solutions near a shock.  We are currently testing these methods and plan to implement the most promising ones into our large eddy simulation code.

 The second goal of this project is to collaborate with experts in computer science in order to greatly improve the computational performance of our codes.  Simulation codes such as ours typically have a floating point operation rate that is only about 5\% of the peak machine performance.  This is a community-wide problem in computational science and engineering.  We believe that by rewriting the most computationally intensive portions of our LES code we may be able to significantly improve the floating point performance.  This may involve taking advantage of the machine specific cache memory size, or writing machine specific routines to perform computationally intensive tasks such as solving tridiagonal or pentadiagonal linear systems and doing matrix vector multiplication with banded
matrices.

 The outcome of this project will be the development of a large eddy simulation code capable of accurately computing turbulent flows with embedded shocks.  The specific application to be investigated is noise produced by supersonic jets.  In addition the code performance will be improved by up to an order of magnitude.  The improved computational performance will serve as an example to the computational science and engineering community of how to improve other simulation codes.