Funky extra flags

Optimization flags

optimize allows you to optimize for different resources, e.g. disk IO, memory usage, etc. The flags for the various optimizations you want to obtain are concatenated by underscores.

optimize flag1_flag2_flag3 (sets 3 optimization flags)

Implemented flags:

• insulator does a direct minimization of the energy functional rather than the usual scfloop. Only works for insulators! Also, the number of bands must be ONLY THE NUMBER OF OCCUPIED BANDS.
• metal uses the ensemble DFT formulation of Marzari and Vanderbilt. It can be very slow and is not recommended.
• memory optimizes for memory in the calculation of NMR shifts, at a small expense in additional CPU time.
• nostartguess Recalculates the starting guesses in the NMR subroutine every time. Using this option will generally result in SUBSTANTIALLY SLOWER execution, but will reduce memory cost.
• nocgsolveguess switches OFF the better starting guesses for the conjugate gradient solver in the NMR subroutine. Using this option will generally result in SUBSTANTIALLY SLOWER execution, but will reduce memory cost.
• nowfnreuse when relaxing with a fixed lattice, DOES NOT reuse the wave functions from the previous relaxation step.
• noprecond do not precondition in the direct energy minimization. Use this option only if you have convergence problems, since it will generally result in 2-3 times slower convergence.
• noadjustshift keeps the number of bands fixed during the calculation and should be used if you have trouble with complex charge densities.

The computational cost is generally dominated by the FFTs, and on parallel computers, a significant fraction of the FFT time is involved in communication between processors. Poor scaling results when latency dominates these communications, but the latency may be reduced (with an associated memory cost on some machines) by setting
number_bands_fft n
where n is the number of bands to FFT simultaneously (this option is disabled when using the SHMEM library).