Hello everyone,
I am a beginner user of VTST. I'm looking for TS of interstitial in fcc systems.
To determine in one simulation the migration energy and the transition state I used vasp compiled with vtst.
I used 2 optimizers FIRE and LBFGS. Both simulations converged to the minimization criterions specified (EDIFF=1e5 and EDIFFG=1E2)
By plotting the MEP, the image with the highest energy was the TS.
But when I wanted to determine the Gvib (using phonopy) of the TS (to get the attempt frequency (..exp((Gvib(TS)Gvib(ini))/kT)) I also plot the phonon band structure.
The problem is that the bandstructure has not the characteristic of a TS (I had 2 imaginaries modes or zero imaginary mode).
You can find attached as support 2 plots and the NEB using FIRE optimizer for N diffusiong between tetra and M site in fcc.
So, here my questions :
Can vtst can failed to determine the TS when the path has some symmetries characteristics ?
Could it be that the NEB are not enough converged ? (EDIFFG = 1E3 ?)
Maybe I could use improved dimer method or lanczos ? But I didn't find any practical examples in crystal so I not really sure how I should proceed.
Any suggestion are welcome, I'm quite stuck with it.
about the determination of the transition state
Moderator: moderators
about the determination of the transition state
 Attachments

 NEB_N_tM.tar.gz
 (32.33 MiB) Downloaded 746 times

 band_dos.pdf
 bandstructure of C at TS between octa and M in a fcc crystal
 (362.17 KiB) Downloaded 528 times

 band_dos.pdf
 bandstructure of N at TS between tetra and site M a fcc crystal
 (362.26 KiB) Downloaded 518 times
Re: about the determination of the transition state
I do have a few suggestions. But first, it's good to see that you were able to converge to a saddle. I think that with some tweaking you will be able to calculate reliable phonons as well.
1. Consider if you really want to relax the cell parameters for this diffusion mechanism. Typically when looking at diffusion in a crystal, the lattice parameters are set by the bulk. It is a little odd to consider a periodic array of point defects moving in a crystal lattice where the lattice vectors can change along the reaction path. Even if you do think this is appropriate, I suggest, for a start, that you keep the lattice vectors of the perfect Ni lattice frozen for all of your calculations.
2. Turn of symmetry for all of these calculations. Repeat your initial and final state calculations using EDIFF=0.005 or 0.002, introducing N into the appropriate sites. Set up these cells so that one atom is in exactly the same position in the initial and final structure and freeze it to eliminate any translation.
3. Again, use EDIFF=0.005 or 0.002 for your CINEB calculation to find a wellconverged saddle.
4. Check, by hand, to see if your saddle is degenerate in the sense that there could be symmetrically equivalent initial and final states to which you specified. It's hard for me to tell, but I don't know the system as well as you do. It is possible that you converged to a second order saddle, possibly because this was enforced by symmetry. If there does appear to be degenerate reaction coordinates (negative modes) at the saddle, then displace the image away along one of the negative modes and see if the path breaks up into two steps or relaxes to a lower energy path. Use FIRE or QM for this, not LBFGS.
4. If you are happy with your saddle, repeat your phonon calculation using ediff=1e8 and a small displacement on the order of 0.001 Ang. Alternatively, you can use the perturbation theory method which is now in vasp.
1. Consider if you really want to relax the cell parameters for this diffusion mechanism. Typically when looking at diffusion in a crystal, the lattice parameters are set by the bulk. It is a little odd to consider a periodic array of point defects moving in a crystal lattice where the lattice vectors can change along the reaction path. Even if you do think this is appropriate, I suggest, for a start, that you keep the lattice vectors of the perfect Ni lattice frozen for all of your calculations.
2. Turn of symmetry for all of these calculations. Repeat your initial and final state calculations using EDIFF=0.005 or 0.002, introducing N into the appropriate sites. Set up these cells so that one atom is in exactly the same position in the initial and final structure and freeze it to eliminate any translation.
3. Again, use EDIFF=0.005 or 0.002 for your CINEB calculation to find a wellconverged saddle.
4. Check, by hand, to see if your saddle is degenerate in the sense that there could be symmetrically equivalent initial and final states to which you specified. It's hard for me to tell, but I don't know the system as well as you do. It is possible that you converged to a second order saddle, possibly because this was enforced by symmetry. If there does appear to be degenerate reaction coordinates (negative modes) at the saddle, then displace the image away along one of the negative modes and see if the path breaks up into two steps or relaxes to a lower energy path. Use FIRE or QM for this, not LBFGS.
4. If you are happy with your saddle, repeat your phonon calculation using ediff=1e8 and a small displacement on the order of 0.001 Ang. Alternatively, you can use the perturbation theory method which is now in vasp.
Re: about the determination of the transition state
Thanks a lot for the answer,
I'll do some simulations following your suggestions and let you know the outcomes in a next reply.
I'll do some simulations following your suggestions and let you know the outcomes in a next reply.