For a given set of F(hkl) values, one wants to determine which one to give phases to in the initial pass, and how many in total to use. Depending upon this choice the later analysis with the genetic algorithm will converge fast or slow. The program div2D attempts to find the best ones to pick. The actual algorithm used is quite complicated, and is a weighted choice depending both upon the magnitude of the unitary structure factors and how well each one is defined by those reflections already chosen to date.

This program takes a set of hkl, F and (if available) phases and works out which are the most useful ones to permute for a subsequent search using the genetic algorithm. (You can specify origin defining reflections yourself or allow the code to do it for you via the ORIGIN command in name.ins.) It will output name.Fsort, name.Usort and name.Dsort. These are, in order, sorted by the size of F, the unitary structure factors U and by the divergence analysis. In general, you should copy the resorted name.Dsort file back to name.hkl. It will also output in control.suggested a prototype of the control file for the genetic algorithm. This should be copied to a file called control.  To run div2D, use the command:

div2D name

in the directory containing the name.ins and name.hkl files. Even simpler, a small shell script div2D.sh can be run as

which will automaticaly copy control.suggested to control and name.Dsort to name.hkl.

In the process of doing the analysis, the program will have a look at the statistics of the reflections, estimate the magnitude of unmeasured reflections and look for Sigma₁ relationships. The statistics can be used as a guide to see if the structure is centro-symmetric or not; this is not to be trusted that much. The estimate of the unmeasured reflections is relatively good in many cases, and can tell you ahead of time whether you may have problems solving the structure. The Sigma₁ relationships can sometimes be used to fix ahead of times certain reflections. You should only use these if the predicted probability is more than 0.9 as a general rule, if at all.

 Back