You have exceeded the array bounds for the data type molecular transformations. You must increase the parameter SYCMAX and recompile the affected programs.
The most common source of this problem is the addition of water molecules without consideration of the choice of symmetry equivalent positions. If you add a water molecule but the particular coordinates you choose to add are that of a symmetry image far from the rest of the atoms you will cause TNT to attempt to generate all the symmetry images between the protein and the distant water molecule. (Even though the R-value and difference map are perfectly happy with your choice.)
The solution is to make sure that the procedure you use to find water always picks the symmetry image closest to the protein. TNT's peak picker does this automatically but some other programs do not. Once the error has been made it can be hard to track down which water molecule is the outlier. What I have done in the past is to place the water in a separate file and peak pick the protein without water. Then I put the water and the peaks in the program Geometry and look for bad contacts. Each peak should have a very bad contact with either the direct or symmetry image of a water molecule. Then you simply relabel the peaks to match you old convention and use them as your water in the future.
Of course, you will have to recompile TNT with a large enough value for sycmax to get Geometry working in the first place.
This problem can also arise if you have a multi-chain protein but have not chosen the particular images of the chains which are next to each other. If the A chain is at the bottom of the unit cell and the B chain at the top you might encounter this problem. In general, you should try to keep your chains as consolidated as possible. Doing so will avoid this problem and save you some fights with the PDB when it comes time to deposit the final model.
For the life of me, I cannot imagine how knowledge of the height of peak in rms tells you whether that peak is caused by a water molecule or phase error. A water molecule has a certain height in electrons/A^3 and that height tells you how good a water molecule would explain away the peak.
Contouring a 2Fo-Fc map at 1 rms, as seems to be commonly done, is pure nonsense. Since all protein is pretty much the same the rms of the protein region of the unit cell will be constant from project to project, given a particular resolution. Therefore the rms of the 2Fo-Fc map will simply be a function of how much of the unit cell is occupied by protein and how much by solvent (which has a much smaller rms). Why would one choose to contour maps at a lower level when there is a 80% solvent and higher when there is 30%. It only leads to confusion.
I always contour 2Fo-Fc maps at 0.35 electrons/A^3 by default but look at lower levels when the density is weak. By being consistent from project to project I can use the experience from one project to help with another.
Groups which have no analog in proteins are tougher. I have spent some time looking at the BCORREL library and believe the following rules can be used.
If your crystal diffracts to a high enough resolution that the B factors of the protein part behave properly (agree with the Bcorrel library even without restraint), there should be no need to construct restraints for rest of the model.
Other programs will, however. The convention for them is to declare a trans proline as type PRO and a cis proline as type CPR. TNT will accept either name and treat them the same. You may want to change the name in case you run across a program that cares.