Gauge Fixing

GLU by default tries to get you to run the fastest gauge fixing routines the FACG of my paper: https://arxiv.org/abs/1405.5812 and the supporting proceedings https://arxiv.org/abs/1412.2807

To get the full benefit of the routine I strongly recommend using FFTW or MKL to Fourier Accelerate the gradients. In principle this is just something like a Jacobi preconditioner but it is unreasonably effective.

If you dislike doing things quickly you can always configure the (Stochastic) Over-relaxation just to prove to yourself that it is a terrible option.

The Landau routines are more memory expensive compared to the Coulomb (which operates timeslice-by-timeslice as this is most efficient) and typically require more iterations and require a 4D FFT compared to 3D, which scale better.

The conjugate gradient routines are demonstrably better than the vanilla steepest-descent at reducing the iteration count but the cost per iteration is a little higher, however this is still really worth it as the paper shows. For larger volumes (as we need fewer FFTs and more operations that scale linearly with the volume) the CG routines are much better as we begin to be dominated by the FFTs.