Core: shape determination of high-resolution peaks is bad
From a simulated data set, the instrument states (detector position, sample position and orientation and incident wavevector) are known, and perfectly defined, so the only thing that needs refining is the unit cell, removing many unknowns. Here is a high resolution peak (11, 7, 14)
:
vs a low resolution peak (1, -2, 11)
:
The issue seems to be in the way the shape is determined. The high resolution peaks lack sufficient neighbouring strong peaks to accurately determine the shape, and since they tend to have a large angular width (mosaicity), they intersect many frames and the "precession" across the image is not captured.
One possible method of resolving this is "bootstrapping" to determine peak shapes. We first use the peaks from the image analysis step to build an initial ShapeCollection
, and use these shapes to integrate the predicted peaks. Then we build a new ShapeCollection
from the predicted peaks, in the hope that we have more good shapes in the high resolution regime.