Reliable structural data from Rietveld refinements via restraint consistency

Abstract

A new approach is proposed for assessment of the quality of restrained Rietveld refinement. As the first step, we require the consistency of the restraints and diffraction data: namely, the absence of outliers in the distribution of the bond length deviations from the target values. This condition is tested using a simple and robust outlier detection criterion and can be reached for any correct structure. Then, the range of restraint variation where the model remains consistent (uncertainty window) is calculated. This metric is indicative of the combined model and data quality. The average half uncertainty window (HUW) can serve as an estimate of the effective bond precision; HUW < 0.04 Å is reachable for highly crystalline powders, and HUW exceeds 0.3 Å for unacceptable models. A round-robin involving the most popular laboratory diffractometer geometries and an analysis of literature examples show that HUW, unlike 'classic' R values, is instrument independent and versatile.