>If you stipulate that you already know the correct answer then it ceases to be >a useful analogy to solving a crystal structure.
But we do know all of the functions we are trying to fit in crystallography--we are trying to estimate their parameters. >The usual problem we face is fixing or replacing an imperfect model. This is >where it is important to improve the observation/parameter ratio. Would I >prefer high quality data to low quality data? Sure! But you typically don't >have that option. You do have the option of choosing a model that optimizes the obs/param ratio. The problem is that not all o:p ratios are created equal. With regard to your paper: I know there are always a thousand things to do, but one could repeat your study starting from a high-multiplicity, high-resolution data set, and see what effect artificially lowering multiplicity (removing measurements randomly) would have on the optimal cutoff for anisotropic B refinement. The cutoff would certainly shift towards lower resolution, but I would be curious by how much. I would guess changing from a multiplicity of 30 to 4 would have a significant if not dramatic effect. JPK