Page 1 of 1

NEB calc stalls on ionic relaxation steps

Posted: Tue Jun 09, 2009 2:30 pm
by chucker
I'm doing a 3 image NEB calculation and the calculations hang/stall if each image doesn't finish on the same ionic relaxation step. For example, here is a tail of my OSZICAR files.

==> 01/OSZICAR <==
DAV: 1 -0.549766659224E+03 -0.54977E+03 -0.10444E-01 11968 0.698E-01 0.132E-01
DAV: 2 -0.549766739042E+03 -0.79818E-04 -0.85038E-03 13592 0.208E-01 0.898E-02
DAV: 3 -0.549766605105E+03 0.13394E-03 -0.15784E-03 13264 0.610E-02 0.625E-02
DAV: 4 -0.549766566731E+03 0.38374E-04 -0.82698E-04 12104 0.431E-02
2 F= -.54976657E+03 E0= -.54976215E+03 d E =0.178017E-02 mag= 0.1788
N E dE d eps ncg rms rms(c)
DAV: 1 -0.549768552297E+03 -0.54977E+03 -0.19289E-02 11752 0.296E-01 0.544E-02
DAV: 2 -0.549768583956E+03 -0.31659E-04 -0.16604E-03 13632 0.913E-02 0.390E-02
DAV: 3 -0.549768564766E+03 0.19190E-04 -0.26958E-04 8176 0.255E-02
3 F= -.54976856E+03 E0= -.54976424E+03 d E =-.217864E-03 mag= 0.1845

==> 02/OSZICAR <==
DAV: 2 -0.550219234906E+03 -0.32937E-04 -0.15049E-03 13720 0.874E-02 0.314E-02
DAV: 3 -0.550219211044E+03 0.23862E-04 -0.95851E-05 8264 0.241E-02
2 F= -.55021921E+03 E0= -.55021409E+03 d E =0.853309E-02 mag= -0.0101
N E dE d eps ncg rms rms(c)
DAV: 1 -0.550223000412E+03 -0.55022E+03 -0.48273E-03 11888 0.142E-01 0.219E-02
DAV: 2 -0.550223007280E+03 -0.68679E-05 -0.32719E-04 11008 0.425E-02
3 F= -.55022301E+03 E0= -.55021789E+03 d E =0.473685E-02 mag= -0.0078
N E dE d eps ncg rms rms(c)
DAV: 1 -0.550214802333E+03 -0.55021E+03 -0.14789E-02 11760 0.239E-01 0.370E-02
DAV: 2 -0.550214828944E+03 -0.26611E-04 -0.80419E-04 13528 0.652E-02

==> 03/OSZICAR <==
DAV: 1 -0.551188239810E+03 -0.55119E+03 -0.12459E-02 11912 0.234E-01 0.389E-02
DAV: 2 -0.551188253385E+03 -0.13575E-04 -0.58371E-04 13640 0.568E-02
2 F= -.55118825E+03 E0= -.55118293E+03 d E =0.503279E-02 mag= 0.0408
N E dE d eps ncg rms rms(c)
DAV: 1 -0.551190463427E+03 -0.55119E+03 -0.24110E-03 11832 0.104E-01 0.142E-02
DAV: 2 -0.551190470415E+03 -0.69881E-05 -0.11822E-04 8192 0.258E-02
3 F= -.55119047E+03 E0= -.55118516E+03 d E =0.281576E-02 mag= 0.0294
N E dE d eps ncg rms rms(c)
DAV: 1 -0.551185451424E+03 -0.55119E+03 -0.12250E-02 11584 0.232E-01 0.353E-02
DAV: 2 -0.551185460374E+03 -0.89498E-05 -0.55013E-04 13456 0.555E-02


Image 01 had converged, but image 02 and 03 want to continue, but don't ever finish the next step. Any ideas on what the problem can be?

Here is my INCAR

System = SiC
ISTART = 0
ENCUT = 450
ISMEAR = 0
SIGMA = 0.05

ISPIN = 2
LREAL = AUTO

IMAGES = 3
SPRING = -5
LCLIMB = .TRUE.

IBRION = 2
NSW = 20

Thanks.

Re: NEB calc stalls on ionic relaxation steps

Posted: Tue Jun 09, 2009 3:25 pm
by graeme
It sounds like an MPI problem related to the connection to idle images. I have never seen this problem and don't know how to fix it. Typical calculations spend more time optimizing points near the saddle which are further from equilibrium, so spending different amounts of time on each image is very common.

I assume that you get the same problem with the built-in NEB with vasp.

Re: NEB calc stalls on ionic relaxation steps

Posted: Tue Jun 09, 2009 3:55 pm
by chucker
Thanks for the reply. I suspected it was mpi related but I wanted to make sure it wasn't on the NEB side of things.