GHOST Development

The Geophysical High Order Suite for Turbulence (GHOST) is a highly scalable Fortran 90-95 pseudo–spectral code that solves a variety of PDEs that are often encountered in turbulence studies. The use of pseudo–spectral methods has a long and distinguished history in the turbulence community that hardly needs expounding upon. The numerics employ a GalerkinFourier spectral method and nonlinear terms are solved using a parallel fast Fourier transform (FFT). For continuous, infinitely differentiable and periodic functions, numerical convergence is exponential (spectral). These methods have found widespread use in the turbulence community primarily because of their spectral accuracy, which often makes them useful–indeed even indispensible–for direct numerical simulation (DNS; XXX), but also increases the accuracy of turbulence models (XXX). Typically, pseudo–spectral methods require only 18 th or fewer of the spatial points required by many low (e.g., 2nd-4th) order finite difference schemes for comparable accuracy, and are optimal in periodic geometry in terms of computational and memory costs. GHOST contains solvers to numerically integrate the various hydrodynamic, magnetohydrodynamic, or Hall-magnetohydrodynamic equations in three dimensions with periodic boundary conditions. Somewhat more specifically, solvers are provided to study neutral and conducting fluids, rotating flows, conducting flows with external magnetic fields, and stratified flows. Subgrid models for turbulence are also available, including Lagrangian average models for hydrodynamics and magnetohydrodynamics, the Leray (tensor-diffusivity) model, and the Clark model (4)-(7) and recent extensions of spectral eddy-viscosity closures (incorporating helicity and eddy noise) (8; 9) could be included as well. The pseudo-spectral method is used to compute spatial derivatives, while a Runge-Kutta method of adjustable order is used to evolve the system in the time domain (10). A non-blocking parallel fast Fourier transform (FFTP) is provided with the code, which uses serial FFT libraries to compute FFTs in each node. Several FFT libraries are supported, including versions 2.X and 3.X of FFTW. The code is highly portable and compiles with several compilers in many platforms including the Portland suite, the XLF compilers, Compaq/HP, ABSoft, INTEL, GNU, and G95 compilers. Several flavors of MPI are regularly tested for compatibility. Both POSIX and MPI I/O are supported. Technical details of the code and of the parallelization techniques used can be found in (1; 2).