The computational prediction of polymorphs

At the empirical end of the spectrum of methods employed for the theoret-ical/computational generation of possible polymorphic crystal structures is the approach taken by Gavezzotti (1991, 1994a, 1996, 1997) and co-workers (1995, 

Another approach (common to many strategies of structure generation and polymorph prediction ) is to attempt to match the powder diffraction pattern calculated from the generated trial structure with an experimental one, although this can no longer be justified as a truly predictive procedure. 

A more purist approach, which in principle ignores the accumulated structural data and associated structural trends, involves placing each independent molecule with a defined conformation at a certain location (X, K, Z) with a certain orientation (p, 6, (p) m a. unit cell defined by its six parameters (a, b, c, a, p, y). Each additional independent molecule in the asymmetric unit is defined with six additional location and orientation parameters. Each variation of conformation for one or all of the individual molecules constitutes a new starting model. 

In the approach of Chin et al., the simulated annealing method was employed on molecules that tend to form hydrogen-bonded tapes (Schwiebert et al. 1995), thus providing a well-defined (but necessarily limited) starting model for the subsequent trial structures. 

How well do these methods work individually and collectively In 1999 the CCDC conducted a blind test of structure (including polymorph) prediction, in which most of the groups mentioned in this section participated (Lommerse et al. 2000). Such tests should serve as a challenge, a model and a landmark for the development of these computational techniques. 

---