[gmx-users] increasing cut-off and PME grid spacing

[Justin A. Lemkul]

Hsin-Lin Chiang wrote:
> **Hi,
> 
> I want to run mdrun -nt 12 on our cluster.
> 
> When I execute,
> grompp -f test.mdp -o test.tpr -c ../3eq/eq.gro -t ../3eq/eq.trr -p 
> ../initial/insulin.top -n ../initial/index.ndx
> 
> I got the message,
> 
> Estimate for the relative computational load of the PME mesh part: 1.00
> 
> NOTE 1 [file test.mdp]:
>   The optimal PME mesh load for parallel simulations is below 0.5
>   and for highly parallel simulations between 0.25 and 0.33,
>   for higher performance, increase the cut-off and the PME grid spacing
> 
> 
> Then I changed my mdp to get
> 
> Estimate for the relative computational load of the PME mesh part: 0.20
> 
> Belos is my new mdp file
> I change rlist, rcoulumb, and rvdw from 1 to 12 and fourierspacing from 
> 0.2 to 2 to get the lower relative computational load.
> I'm afraid these extreme high value caused some bad effect.
> 

If the simulation is even stable, it will be horribly inaccurate.  12 nm cutoffs 
are unheard of and 2-nm grid spacing is about 20 times too large.

Without seeing the original .mdp file that gave the high PME load, and without a 
further description about how large the system is (number of atoms), it is hard 
to say what you should do.  Some system do not parallelize well, but I imagine 
you should be able to get better performance.

-Justin

> title        = ttt
> cpp                 =  /lib/cpp
> constraints         =  hbonds
> ;define              =  -DFLEX_SPC
> integrator          =  md
> emtol               =  100.0
> emstep              =  0.005
> dt                  =  0.002    ; ps !
> nsteps              =  25000000  ; total 50 ns
> nstcomm             =  5000
> nstxout             =  5000
> nstvout             =  5000
> nstfout             =  5000
> nstlog              =  5000
> nstenergy           =  5000
> nstlist             =  10
> ns_type             =  grid
> rlist               =  12
> rcoulomb            =  12
> rvdw                =  12
> coulombtype         =  PME
> fourierspacing      =  2
> pme_order           =  6
> optimize_fft        =  yes
> Tcoupl              =  v-rescale
> tc-grps             =  Protein Non-Protein
> ;tau_t               =  0.1  0.1
> tau_t               =  0.2 0.2
> ref_t               =  300 300
> energygrps          =  A-chain B-chain SOL NA
> Pcoupl              =  berendsen
> Pcoupltype          =  isotropic
> ;tau_p               =  0.1
> tau_p               =  0.25
> compressibility     =  5.4e-5
> ref_p               =  1.0
> gen_vel             =  yes
> gen_temp            =  300
> gen_seed            =  173529
> 
> 
> Sincerely yours,
> Hsin-Lin
> 

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

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

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