[gmx-users] mdrun_mpi segmentation fault for run in vacuum

[Justin A. Lemkul]

On 6/12/12 7:46 AM, reisingere at rostlab.informatik.tu-muenchen.de wrote:
>>
>>
>> On 6/12/12 7:34 AM, reisingere at rostlab.informatik.tu-muenchen.de wrote:
>>>>
>>>>
>>>> On 6/12/12 7:05 AM, reisingere at rostlab.informatik.tu-muenchen.de wrote:
>>>>>>
>>>>>>
>>>>>> On 6/12/12 5:54 AM, reisingere at rostlab.informatik.tu-muenchen.de
>>>>>> wrote:
>>>>>>> Hi everybody,
>>>>>>> I tried to run a minimization just of the hydrogen of a membrane
>>>>>>> protein.
>>>>>>> I want to do this in vacuum.
>>>>>>>
>>>>>>> But when I started the run with
>>>>>>>
>>>>>>> mpirun mdrun_mpi -deffnm protein -v -nt 2
>>>>>>>
>>>>>>> I get the error that there is a segmentation fault.
>>>>>>
>>>>>> Threading and MPI parallelization are independent.  You can't use
>>>>>> both.
>>>>>> If
>>>>>> you've compiled with MPI support, you can't invoke the -nt option.
>>>>>
>>>>> okey, I'll try it without the -nt option. But I still get the error
>>>>>>
>>>>>>> But when I only type
>>>>>>>
>>>>>>> mpirun mdrun_mpi
>>>>>>>
>>>>>>> there is no problem so I guess that my already produced input files
>>>>>>> are
>>>>>>> the problem. For example I am not completely sure about the .mdp
>>>>>>> file.
>>>>>>> Can
>>>>>>> you please give me an example for a .mdp file for a minimization of
>>>>>>> only
>>>>>>> the hydrogen but not the whole protein in a vacuum.
>>>>>>>
>>>>>>
>>>>>> It would be far more useful for you to post what you're using so we
>>>>>> can
>>>>>> provide
>>>>>
>>>>>
>>>>> The .mpd file I use looks like this:
>>>>>
>>>>> define          = -DPOSRES
>>>>> integrator      = md
>>>>> tinit           = 0
>>>>> dt              = 0.005
>>>>> nsteps          = 20000
>>>>> nstxout         = 5000
>>>>> nstvout         = 5000
>>>>> nstfout         = 0
>>>>> nstlog          = 1000
>>>>> nstxtcout       = 1000
>>>>> nstenergy       = 1000
>>>>> energygrps      = Protein Non-Protein
>>>>> nstcalcenergy   = 5
>>>>> nstlist         = 5
>>>>> ns-type         = Grid
>>>>> pbc             = xyz
>>>>> rlist           = 0.9
>>>>> coulombtype     = Cut-off
>>>>> rcoulomb        = 0.9
>>>>> rvdw            = 0.9
>>>>> fourierspacing  = 0.12
>>>>> pme_order       = 4
>>>>> ewald_rtol      = 1e-5
>>>>> tcoupl          = V-rescale
>>>>> tc-grps         = Protein  Non-Protein
>>>>> tau_t           = 0.1      0.1
>>>>> ref_t           = 298      298
>>>>> Pcoupltype      = Isotropic
>>>>> tau_p           = 2.0
>>>>> compressibility = 4.5e-5
>>>>> ref_p           = 1.0
>>>>> gen_vel         = no
>>>>> constraints     = all-bonds
>>>>> constraint-algorithm = Lincs
>>>>> unconstrained-start  = yes
>>>>> lincs-order     = 4
>>>>> lincs-iter      = 1
>>>>> lincs-warnangle = 30
>>>>> implicit_solvent = GBSA
>>>>> gb_algorithm    = HCT
>>>>> nstgbradii      = 1.0
>>>>> rgbradii        = 0.9
>>>>> gb_epsilon_solvent = 80
>>>>> gb_dielectric_offset = 0.009
>>>>> sa_algorithm    = Ace-approximation
>>>>>
>>>>
>>>> Your goal is energy minimization of H atoms in a vacuum, correct?  Your
>>>> .mdp
>>>> file is for a full MD simulation using an NVT ensemble in implicit
>>>> solvent.
>>>> What you want is something more along the lines of:
>>>>
>>>> define	    = -DPOSRES
>>>> integrator  = steep
>>>> emtol       = 1000.0
>>>> emstep      = 0.01
>>>> nsteps      = 50000
>>>> energygrps  = system
>>>> nstlist     = 1
>>>> ns_type     = simple
>>>> rlist       = 0
>>>> coulombtype = cutoff
>>>> rcoulomb    = 0
>>>> rvdw        = 0
>>>> pbc         = no
>>>
>>> but how does the minimization "know" that it should be in vacuum.
>>
>> There's nothing particularly special about running in vacuum except that
>> there
>> is no solvent.  Infinite cutoffs and a nonperiodic simulation cell are
>> common
>> conventions for simulating in vacuum, so that's what the .mdp file above
>> does.
>
> okey, thanks a lot!!
> But when I want to do a whole md run after this minimization my first .mdp
> file was correct, right?
>

For running in implicit solvent, that .mdp file has a number of problems.

By restraining the protein, you don't really accomplish anything.  Also note 
that for using implicit solvent, you generally need infinite cutoffs (like in 
the case of the "vacuum" .mdp file I posted before) and no pressure coupling.  I 
have found that with finite cutoffs, energy conservation is poor and structures 
are unstable when using implicit solvent.

There are a number of useful discussions in the list archive about such 
considerations; I would encourage you to spend some time reading before diving 
headfirst into something that may not turn out well.

-Justin

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

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

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