The TNT package uses gradient descent methods to optimize the parameters of the molecular model. For methods of this type the structure of a cycle of refinement is the same regardless of the particular method used (Agarwal, R.C., Acta Cryst. (1978). A34, 791-809). Think of each parameter of the model as a component of a very long vector. First a new vector is chosen to add to the original vector, this is the shift vector. Only a fraction of this vector, however, is to be added. The second part of each cycle is to determine how much.
If the model vector is called and the shift vector , the new model vector is
where is the fraction of the shift to be applied. The two steps of a cycle of refinement are first to calculate and then to determine . The different minimization methods available in TNT alter how is calculated. The calculation of is always done the same way.