Large portions of the unit cell of a macromolecular crystal are filled with disordered solvent. X-rays scatter from this solvent, and the structure factors are affected. To reproduce the scattering of the crystal, both the ordered and the disordered portion must be modeled accurately.
Models of proteins often include no description of the disordered solvent, and as a consequence they are particularly poor at reproducing the low-resolution scattering. Usually the problem is simply ignored by deleting the low resolution data. While this solution avoids embarrassingly high R values it does not solve the problem.
Density maps calculated without the low resolution data will be significantly distorted, particularly near the surface of the molecule. Models that contain no description of the disordered solvent will have systematic errors in many of their parameters when they are refined against the low-resolution data. In particular, the B factors will all be underestimated. The only way to avoid these problems is to include all the data and to incorporate a model of the disordered solvent.
To be useful, a solvent model must be simple to incorporate, quick to
calculate, and must contain a minimum number of parameters. The method
used in TNT has all these properties. The review of Brünger
(1994) discusses a number of other solvent-modeling methods.