TURBOMOLE Users Forum
TURBOMOLE Modules => Jobex: Structure Optimization and Molecular Dynamics => Topic started by: e.berardo on May 15, 2012, 07:44:58 PM
-
Hi to everybody,
I am struggling to optimize the geometry of a vanadium dioxide cluster (V9O18) with DFT (B3-LYP), starting from a bulk cut.
[ I am using a def2-TZVP basis set, scfconv 7, gcart 4] - using TURBOMOLE 6.3.1
I want my cluster to be spin polarized with 9 unpaired electrons and to help the SCF convergency I decided to run an open shell calculation allowing fractional occupancy.
Also with this set of parameters I wasn't able to bring the system to convergence (going over 500 SCF iterations, with big energy changes between two different steps) and calling eiger I noticed some gaps in the electronic occupancy.
I checked the turbo-forum (thanks for the useful suggestions) and decided to use the fermi smearing (300 K) fixing the number of unpaired electrons to 9, and modify the scforbital shift from closedshell=0.5 to automatic=.1. Fermi really helped the SCF convergence of the system and the optimization ended up fine.
As in the previous optimization I exploited the c2v symmetry, I wanted to try to restart the optimization without any symmetry constrain.
So, I got the final structure, I did a clean restart with define and started the optimization with C1 symmetry.
The SCF reached convergency and the first gradient calculation ended up fine. The second SCF converged, but the grad stopped: error in gradient step (1).
I checked the gradient calculated in the first step and this doesn't seem incredibly high (|dE/dxyz| = 0.079746). I also checked the electronic occupancy which looks fine with no gaps in it.
I appreciate any suggestion or advise, as I ran out of ideas.
Thanks for the support,
Enrico
-
Hi Enrico,
Did you run the job using a queueing system? Did it produce anything in standard output (for example an ugly crash report)? Did you run the job in parallel? And which file contained the message "error in gradient step"? I'm just thinking that there should be more output somewhere that could help us to diagnose the situation.
Sometimes the TM binaries can crash due to too low stacksize, but your system is not that large so I would not expect this to be a problem.
Antti
-
Hi Antti,
Yes I used a queueing system, I ran the MPI version of TM on my local cluster (cluster build up from dual quad core xeon motherboard, linked by infiniband), running on 16 processors.
As far as I can see there are no faulty files apart from the GEO_OPT_FILES which sends me to job.2, where there is the gradient step which died.
Job.last contains the first successful GRAD step, STATPT and the second dscf (which converged in 54 iterations). The file .e is empty, and all the slave.output files show the unsuccessful gradient step.
thanks
Enrico
-
Hi,
If the queueing system stdout and stderr don't show any additional information, this is a rather nasty case to diagnose. Perhaps it would be best to contact the Turbomole support.
Best,
Antti