Real space refinement has had a long history of being successfully applied to the solution of protein structures. It is useful because it is the simplest way to incorporate both the magnitude and phase of the observed diffraction pattern. Because of the widespread use of reciprocal space refinement methods the real space refinement method has been overlooked in recent years.
The case where real space refinement is indicated is when phases have been determined, by some means, and a partial model has been constructed. To produce a good phase combined map, the partial model should be adjusted to fit the data as closely as possible. Refinement of a partial model in reciprocal space is difficult because it is not clear how much of each observed amplitude is supposed to be due to the missing portion of the model. The problem is solved in real space because the Fourier synthesis which produced the ``observed'' map brings the relevant density into the neighborhood of the atoms in the partial model.
The function minimized with this module is
where the
calculated electron density map is series terminated to match
the resolution limits of the diffraction data. Because the
calculated density far from any atom in the model is zero, all
terms which depend on density points far from the model are
constant and do not affect the gradient of the function.
Therefore we only need to sum over the points that are close to
the partial model. If one works out the equation for the
gradient of this function one will find that it can be
calculated using the Agarwal method of the RFACTOR module. The
only difference is in the coefficients of the difference map.
For this case the coefficients are those of the map which
results from the inversion of the Fourier coefficients
.
The program Rfactor performs all the work specific to this module. Run the program exactly as you would when running reciprocal space refinement but add a MODULE REALSPACE modifier to the command card.