[gmx-developers] Reaction Filed crash!

[Justin A. Lemkul]

Miguel Machuqueiro wrote:
> 
> Hi to all developers,
> 
> We have been using GMX here in the group for almost a decade and we are 
> familiar with almost all versions of the package.
> Recently, when testing the newer versions (4.5.x) using Reaction Field 
> in a simple lysozyme system (4LZT), my system crashed constantly.
> 
> System: 4LZT (129 a.a.) directly from PDB databank.
> Water: ~6250 molecules
> GMX: 4.5.x (all versions tested)
> ForceField: Gromos 53A6
> Long range elect.: RF, RF-nec, GRF or simply "Cutoff" (all with same 
> fate.. crash!)
> Ensemble: NPT or NVT
> 
> So, what do we already know about the problem:
>  - The crash always starts in the NH bond present in the amides (main 
> chain or side chains of GLN and ASN). In the stepxxxx.pdb, we observe 
> the hydrogen jumping out of place.
>  - The stability of the simulation is dependent on the quality of the 
> minimization/initialization procedure. Simple/regular procedures leads 
> to crashs within 10 ns of simulation. More elaborate protocols of 
> minimization/initialization can delay the crashs a little longer.
>  - The system crashes independently of the number of "threads" used (1-12).
>  - The instability only leads to crashs when dealing with proteins >100 
> a.a.. Simulating a 15 a.a. peptide for over 1 microsecond does not lead 
> to a crash.
>  - No problem/crash is observed when using PME.
>  - No problem/crash is observed when using GMX versions prior to 4.0.7.
> 
> My question is this, what might have happen with the RF code between 
> versions 4.0.x and 4.5.x that leads to this problem? Or is there any 
> specific parameter to be used with RF/Cutoff to avoid these problems?
> 
> Even though RF is not very popular with GMX developers, it has been 
> essential for the development of our in house Constant-pH MD method 
> (http://www.gromacs.org/Documentation/How-tos/Constant_pH_Simulation 
> <http://www.gromacs.org/Documentation/How-tos/Constant_pH_Simulation?highlight=constant+pH>).
> 
> On a typical MD run we use the following MDP file, or a similar one.
> 
> ________________
> integrator               = md
> tinit                    = 0
> dt                       = 0.002
> nsteps                   = 5000000
> comm-mode                = Linear
> nstcomm                  = 1
> nstxout                  = 0
> nstvout                  = 0
> nstfout                  = 0
> nstlog                   = 5000
> nstenergy                = 5000
> nstxtcout                = 5000
> xtc-precision            = 1000
> energygrps               = Protein SOL
> nstlist                  = 5
> ns_type                  = grid
> pbc                      = xyz
> periodic_molecules       = no
> rlist                    = 0.9
> coulombtype              = Reaction-Field
> rcoulomb-switch          = 0
> rcoulomb                 = 1.4
> epsilon_r                = 1
> epsilon_rf               = 54
> vdwtype                  = Cut-off
> rvdw-switch              = 0
> rvdw                     = 1.4
> DispCorr                 = No
> tcoupl                   = berendsen
> tc-grps                  = Protein SOL
> tau-t                    = 0.1 0.1
> ref-t                    = 298.15 298.15
> Pcoupl                   = berendsen
> Pcoupltype               = Isotropic
> tau-p                    = 0.5
> compressibility          = 4.5e-5
> ref-p                    = 1
> refcoord_scaling         = No
> gen_vel                  = no
> constraints              = all-bonds
> constraint_algorithm     = lincs
> lincs-order              = 4
> lincs-iter               = 1
> lincs-warnangle          = 90
> ___________________________
> 
> Thank you for any help you can assist us.
> 

Sporadic crashes have been reported for systems that are stable under 4.0.x and 
not 4.5.x.  The solution that seems to have worked is to adjust the nsttcouple 
and nstpcouple values.  See the discussion here (and subsequent posts in the 
thread):

http://lists.gromacs.org/pipermail/gmx-users/2011-September/064232.html

-Justin

> Regards,
> Miguel
>  
> 
> -- 
> ============================================
> Miguel Machuqueiro
> Department of Chemistry and Biochemistry
> Faculty of Sciences, University of Lisbon
> Campo Grande, Edifício C8 (sala 8.5.47)
> 1749-016 Lisboa, Portugal
> Tel.  : +351 217500112 (int.ext.28547)
> Mobile: +351 967562285
> E-mail: machuque at fc.ul.pt
> www1: http://webpages.fc.ul.pt/~mamachuqueiro
> www2: http://intheochem.fc.ul.pt
> ______________________________________________
> 
> 

-- 
========================================

Justin A. Lemkul
Ph.D. Candidate
ICTAS Doctoral Scholar
MILES-IGERT Trainee
Department of Biochemistry
Virginia Tech
Blacksburg, VA
jalemkul[at]vt.edu | (540) 231-9080
http://www.bevanlab.biochem.vt.edu/Pages/Personal/justin

========================================