Parameter-shift optimization of the minimal function is used as the default phase-refinement method because tests with known data sets have shown it to be more reliable (within the context of SnB) for large structures. In most cases, a maximum of two 90-degree shifts are tested for each phase, and three passes (iterations) are made through the list of phases sorted in decreasing order according to their related |E| values. However, parameter-shift is performed differently for P1 structures. Only a single iteration is made, and the size of the shift angle is increased for large structures (>100 unique non-H atoms).

In difficult cases, it may be worth trying tangent-formula refinement as an alternative.

SnB may run more efficiently on very small structures if conventional tangent refinement is used without the dual-space cycling scheme. In this case, the number of tangent-refinement iterations should be on the order of 1-2 dozen with more iterations being performed for larger structures. In addition, the number of Shake-and-Bake cycles is reduced to 1.