Either the topology is bad or the starting structure has clashes that cannot be adequately addressed using EM. Visual inspection should tell you what is causing the clash. Further minimization, with a smaller step size or different algorithm, may be useful, but usually there is something more fundamentally wrong going on when this happens.

by visual inspection , I didn't find any clashes .

note that this coordinates are the coordinates of the crystallized structure . so I don't think that there are any clashes .

but I think that due to bad topology but how I can solve this problem .

note that the topology obtained from ATB server .

Proper parametrization is a difficult task that can take days or weeks of effort per molecule. You need to understand how the parent force field was optimized and validated and repeat those procedures for any functional group or molecule. Gromos force fields have, in one sense, the benefit of having transferable functional groups, but the downside is that introducing totally new groups is very difficult because the methods are not always clear.

The biggest questions at this point are:

1. What is the molecule you're working with?

2. How do the ATB parameters for known functional groups compare to those in the parent force field? For instance, are the charges and atom types in the ligand for any alcohols, amines, acids, aromatic groups, etc. consistent between the parent force field and that topology that ATB produced?

3. Are there any new functional groups in the ligand that are not present in the parent force field? If so, how have you evaluated their parameters?

Dear Dr. Justin

where I can find the library of Gromos forcefeild ? Is there is a one or I need to collect the data from letratures ?

Is there is a code to assign each charge of the parent FF to its atom or I need to make it manually ?

Charges are assigned based on what the functional group is. You can find out what functional groups are covered by reading the paper associated with the parameter set you're working with (something hopefully you have already done!) and by looking through the force field's .rtp file for examples related to known residues.

Just use double precision version for the first optimization. I think then you are good to switch back to single precision. Also don't forget to do a CG step every 10 SD steps.

In such cases,as Justin mentioned earlier, ligand topology is the primary culprit, so you need to be careful while preparing it. Another thing is to run the simulation of ligand alone in water to check whether its parameters are stable or not.

I had the same problem, and upon visual inspection, parameters/coordinates look fine, but in EM, they acquire an unusual bonding pattern. Later, I refined these things and simulated the complex successfully...