view article
Figure 9
Time needed for one cycle of wavefunction calculation, partitioning and least-squares refinement for HAR/fragHAR as a function of the number of atoms in the refinement. Data points shown are oligopeptide and protein structures and refinements presented in this work, using r2SCAN/3-21G. Timing of partitioning has not been corrected for different resolutions or mapping onto different computational cores in the case of a low number of atoms per core. Fitted lines are shown for both wavefunction calculation types (an exponential for HAR and a linear function for fragHAR). Timing for the Gauss–Newton least-squares refinement is reported for ten cycles in olex.refine. This can be considered as a reference to how long the refinement using the IAM or any non-spherical method would take, as this is a common step for all models. Refinements of the polypeptides were run until convergence was achieved (<10 cycles in all cases). For the two protein data sets the refinements using the full least squares became stuck in oscillations. Therefore, the calculations were considered finished after five HAR cycles and a local minimum was found using Levenberg–Marquardt damping during the refinement.

IUCrJ
Volume 13| Part 1| January 2026| Pages 53-62
ISSN: 2052-2525
https://doi.org/10.1107/S2052252525010188