NEB Convergence - forces not going down

[mmg016]

Dear VTST developers and users,

I am running a NEB calculation on a relatively simple system (CH2 plus H on a Cu overlayer coated Nickel (100) slab). My convergence criteria is EDIFFG = -0.01 (0.01 ev/A). I am using IOPT = 2 optimiser. I see that the forces are initially going down until 0.1 eV/A, but then they are not going down further on some images. They are maintaining steadily around that point.

I have attached the .tar.gz of my calculation. I would greatly appreciate any pointers/help to bring the forces down to convergence.

Best regards,
mmg016

[graeme]

Your initial path is relatively far from a minimum energy path. It involves breaking a bond well away from the surface.

Try starting the path again using IOPT=3 and TIMESTEP=0.01. Run for about 20-50 iterations and then take a look at what the path is converging towards. Look for intermediate minima and indications that it is converging towards a different mechanism. Once the forces are below ~1 eV/Ang, you can use a TIMESTEP of 0.1 or go back to IOPT=2. Then again, monitor your run and make sure that the path is smooth and converging steadily. If intermediate minima show up, minimize them and break the path into separate NEB calculations.

The path you are trying to converge now is very long and looks to me to involve a number of different elementary steps so that there is insufficient resolution to properly converge. Try my suggestions and it should work better.

[mmg016]

Dear Prof. Henkelman,

Thank you for your kind suggestions.

I have tried to implement them. Initially, as suggested, I restarted this NEB with IOPT = 3 and TIMESTEP = 0.01. It took 56 ionic steps for the forces to come down to 1 eV/Angs. At this stage, I restarted from these geometries with IOPT = 3 and TIMESTEP = 0.1. The convergence criteria for this calculation was set to be 0.1 eV/Angs. However, I see that forces on some images are going down until 0.2 eV/Angs but not further down to convergence. They are infact going back up from that point. When I see the path of convergence, it is rather smooth with no other minima.

Could you please take a look at the attached. Would you like for me to continue for few more steps with TIMESTEP = 0.01 or may be switch to IOPT=2.

Best regards,
mmg016

[graeme]

Try now minimizing images 01, where the methyl molecule has simply rotated, and then 03, where the C-H bond has clearly broken. Assuming these minimize to different initial and final states, use them for a new band to resolve this specific reaction path.