xnn.common.models.dispersion_fast.eeq.LOW_K_VECTORS#

xnn.common.models.dispersion_fast.eeq.LOW_K_VECTORS = 256#

reciprocal vectors (smallest |G|, largest weight) in the low-rank part of the conjugate-gradient preconditioner; 0 for plain Jacobi. Measured on water boxes (A100, 5k-24k atoms): 256 cuts the operator applications of a step 3-5x (94 -> 31 at 5k, 154 -> 43 at 24k atoms in float32) and is the fastest or within 1% of it; more vectors save few iterations and cost more set-up