[gmx-users] segfault with an otherwise stable system when I turn on FEP (complete decoupling)

[Christopher Neale]

Dear Michael:

I have uploaded them as http://redmine.gromacs.org/issues/1306

It does not crash immediately. The crash is stochastic, giving a segfault between 200 and 5000 integration steps. That made me think it was a simple exploding system problem, but there are other things (listed in my original post) that make me think otherwise. Most notably, the drug is fine in both water and vacuum. I have also built numerous systems and get the crash in each one. My actual production system also has a protein, but during debugging I found that the error persists in a simple and small water solution.

Thank you for your assistance.
Chris.

-- original message --

Chris, can you post a redmine on this so I can look at the files?

Also, does it crash immediately, or after a while?

On Thu, Jul 18, 2013 at 2:45 PM, Christopher Neale
<chris.neale at mail.utoronto.ca> wrote:
> Dear Users:
>
> I have a system with water and a drug (54 total atoms; 27 heavy atoms). The system is stable when I simulate it for 1 ns. However, Once I add the following options to the .mdp file, the run dies after a few ps with a segfault.
>
> free-energy = yes
> init-lambda = 1
> couple-lambda0 = vdw-q
> couple-lambda1 = none
> couple-intramol = no
> couple-moltype = drug
>
> I do not get any step files or any lincs warnings. If I look at the .xtc and .edr files, there is no indication of something blowing up before the segfault. I have also verified that the drug runs without any problem in vacuum. I get the same behaviour if I remove constraints and use a timestep of 0.5 fs. The segfault is reproducible with v4.6.1 and v4.6.3. I am using the charmm FF, but I converted all UB angles in my drug to type-1 angles and still got the segfault. I also get the segfault with particle decomposition and/or while running a single thread. I am currently using the SD integrator, but I get the same segfault with md and md-vv. Couple-intramol=yes doesn't resolve it, neither does using separate T-coupling groups for the water and drug. Neither does turning off pressure coupling.
>
> Here is the .mdp file that works fine, but gives me a segfault when I add the free energy stuff (above):
>
> constraints = all-bonds
> lincs-iter =  1
> lincs-order =  6
> constraint_algorithm =  lincs
> integrator = sd
> dt = 0.002
> tinit = 0
> nsteps = 100000
> nstcomm = 1
> nstxout = 0
> nstvout = 0
> nstfout = 0
> nstxtcout = 500
> nstenergy = 500
> nstlist = 10
> nstlog=0 ; reduce log file size
> ns_type = grid
> vdwtype = cut-off
> rlist = 0.8
> rvdw = 0.8
> rcoulomb = 0.8
> coulombtype = cut-off
> tc_grps             =  System
> tau_t               =  1.0
> ld_seed             =  -1
> ref_t = 310
> gen_temp = 310
> gen_vel = yes
> unconstrained_start = no
> gen_seed = -1
> Pcoupl = berendsen
> pcoupltype = isotropic
> tau_p = 4
> compressibility = 4.5e-5
> ref_p = 1.0
>
> I do realize that some of these settings are not ideal for a production run. I started with the real Charmm cutoffs + PME, etc, (which also gives the segfault) but this is what I am using right now for quick testing.
>
> The only thing keeping me from filing a redmine issue is that if I remove my drug and do the FEP on one of the water molecules (using the FEP code listed above), I have no segfault. Therefore it is clearly related to the drug, whose parameters I built so I may have caused the problem somehow. Nevertheless, the drug runs fine in water and in vacuum without the FEP code, so I can't imagine what could be causing this segfault (also, the fact that it's a segfault means that I don;t get any useful info from mdrun as to what might be going wrong).
>
> Thank you,
> Chris.