Pseudo-Spectral Methods

Pseudo-spectral methods compute derivatives in spectral space but evaluate nonlinear products on a physical grid, joined by fast transforms.

Transform between two representations

A pure spectral treatment of a nonlinear term requires convolution sums over modes, which are expensive. The pseudo-spectral approach avoids them by keeping two representations of the field: the spectral coefficients, where derivatives are cheap and exact, and the physical grid values, where pointwise products are cheap. The fast Fourier transform (FFT) moves between them in O(N log N) operations.

A typical step transforms fields to physical space, multiplies them pointwise to form the nonlinear term, transforms the product back to spectral space, and there applies derivatives and time advancement. This collocation-based handling of nonlinearity is what distinguishes pseudo-spectral from fully spectral Galerkin methods.

Aliasing and dealiasing

Evaluating a product on a finite grid folds high-wavenumber content back into resolved modes, an error called aliasing. It can drive instabilities in turbulence simulations. The standard cure is the 3/2 rule: pad the transforms to 3/2 the resolution before forming products so the aliased content lands in modes that are then discarded. The 2/3 truncation rule is an equivalent simpler variant.

Strengths and constraints

Pseudo-spectral methods retain near-spectral accuracy while making nonlinear terms affordable, so they are standard for direct numerical simulation of turbulence. They inherit the periodicity and smoothness assumptions of spectral methods and depend on efficient parallel FFTs, whose global communication can dominate at large scale.

• FFT bridges spectral and physical representations
• Nonlinear products formed pointwise in physical space
• Dealiasing via 3/2 padding or 2/3 truncation
• Global FFT communication is the parallel bottleneck

This method underlies many plasma turbulence codes, where resolving the nonlinear energy cascade accurately across scales is the central numerical challenge.