[gmx-users] On the use of SD and CG

[Martina Bertsch, PhD]

Greetings.

First of all, thank you, David and Anton, for all your kind advice. I 
know that I speak for many GMX users when I say that we have learnt a 
lot from you. Special thanks to Peter and Graham for their modification 
of mdrun and all the other contributions.

I have just had a very long day trying to prepare my system for the 
PR-MD by SD- and CG-EM. The system was originally composed of 1 protein 
molecule, 114 POPC lipids and 2460 SPC waters. First I applied the SD 
algorithm (max 10000 steps, emtol = 1000 kJ mol^-1 nm^-1, initial step 
size: 0.01 nm), which after ~200 steps returned a well known error message:

Stepsize too small (... nm) Converged to machine precision, but not to the requested precision (...)

At that point, I used the structure resulting from the SD as an input 
for the CG-EM run (max 10000 steps, emtol = 1000 kJ mol^-1 nm^-1, 
initial step size: 0.01 nm, nstcgsteep=50). This time, after ~30 steps, 
I obtained another familiar message:

Fatal error: ci = -2147483648 should be in 0 .. 7999 [FILE nsgrid.c, LINE 210]

The atoms from 2 POPC and 3 waters that had large forces acting on them, 
appeared to have clashes with the protein, so I deleted the 
corresponding molecules from the *gro file resulting from the 
interrupted CG, updated the number of atoms on line 2 of the same file, 
renumbered with editconf and updated the *top file. 

Then, I entered that data set into another iteration of SD and CG, 
followed by deletion of overlapping lipid and solvent molecules. I 
noticed that the forces were weaker, but still well above the emtol 
value. Neither algorithm was achieving convergence.

I reminimized the protein itself in Insight II without any problem. Upon 
reinsertion in the membrane, the forces were weaker, but eventually it 
took me 9 cycles (!) of SD and CG (gradually decreasing emtol) and 
deletion of unwanted molecules of POPC and water. The system eventually 
converged after about 600 steps SD and 150 steps CG (combined over all 
iterations) at emtol = 100. There were 108 POPCs and 2439 SOL molecules 
left in the system. The potential energy was ~ -2.2*10^5.

Next, I wanted to run a PR-MD, with positional restraints imposed on the 
protein, using PME, separate T coupling for each group and no p coupling 
(see one of my previous messages for *pr.mdp). Since many messages in 
the gmx archive stress the importance of neutralizing the system before 
a PR-MD run with PME, I added 10 Cl anions to my system. But now, as you 
might guess, my PR-MD simulation crashes, giving the following error:

LINCS warning: relative constraint deviation after LINCS between certain 
protein and/or POPC and/or water atoms

So, I try minimizing this new system with SD, examine the resulting 
structure, delete a Cl ion that got inside the protein transmembrane 
cavity, reminimize using CG, delete some more POPC and water molecules, 
but the system won't converge. The forces are on the order of 2000-3000. 
I tried using the resulting structure as an input for PR-MD or full MD, 
but could not even produce a single step.

I have a feeling this adventure should not be quite so eventful: I am 
getting every single error message from the book. Any kind suggestions?

Best regards,

Martina