Figure 4
Diffuse intensity statistics and data-quality indicators. (a) Merged intensities were split between halo and non-halo voxels (see main text) and binned in shells of constant scattering vector s, where 1/s is the resolution. The mean intensity (top panel) rises to a peak at ∼3 Å resolution (∼0.3 Å−1). The intensities are normalized to 1 at the non-halo maximum. The halo-containing voxels have higher intensity on average than non-halo voxels. The intensity variations of interest are quantified by the standard deviation within each resolution shell (bottom panel). The halo-containing voxels have an approximately fivefold greater signal than non-halo voxels. Within each resolution shell, the non-halo variations are less than 10% of the mean scattering, indicating that this signal is much more subtle. (b) The correlation coefficient for random half-data sets (CC1/2) between crystals 1 and 2 scaled independently (CCRep) and between half-data sets split according to Friedel symmetry (CCFriedel) are compared for halo-containing voxels (top panel) and non-halo voxels (bottom panel). Correlation coefficients are close to 1 for much of the resolution range (insets) and decay at high resolution as the signal-to-noise ratio decreases. The correlations are higher overall for halo-containing voxels, as expected from the higher signal strength. There is no significant difference between random selection of observations (CC1/2) and grouping observations based on symmetry considerations (CCFriedel), which is expected for successful scaling. The CCRep statistic closely follows CC1/2, showing that both the measurement and data-processing procedures are reproducible. |