Is G-SSNEB available now?

[shilendra]

Dear Prof. Graeme,

I implemented all your above suggestions. Now I am getting same results with standard pseudopotential PBE_Na and and PBE_Na_pv. But only problem is that I gave 5 images 6 is the final images. I am getting image 5 is more stable than final image 6.

Why is it so ? I have relaxed image 5 individual also but I am getting same result.

My other query is that sometime I have seen results are reasonable with default ENCUT value but if we give ENCUT 1.3 times then results becomes opposite.

How does choose ENCUT for ssNEB ?

I have attached here graph of image energies.

Thank You
Shilendra

[graeme]

You may have to repeat the restart process until the ssNEB finishes after one iteration. This will guarantee that the cutoff is consistent for all of the images. You can always go to a higher cutoff (or accurate with the 1.3 multiplier) which will improve the convergence rate (but also make each calculation slower). As with any DFT calculation, you need to make sure that your results are converged with respect to the calculation parameters.

[shilendra]

Dear Prof. Graeme,

As you said in above post repeat the restart process until the ssNEB finishes after one iteration.
I want to ask here, is this repeat restart process necessary every time for ssNEB calculations or when we are not getting reasonable result then we should repeat the restart process until the ssNEB finishes after one iteration.

If programme stopped after the message reached required accuracy stopping structural energy minimization . Still we have to check convergence criteria which we gave has reached or not. How does check it by looking dE value in OSZICAR file. Any other way to check convergence criteria for EDIFF and EDIFFG.

Thank You
Shilendra

[graeme]

It is only when the cell changes shape during the calculation that you need to restart in order to compare relative energies. So, for example, when you found an intermediate point lower than an endpoint after a calculation, you can see if this is real by doing a single electronic structure calculation (NSW=1) and see if the energies change.

I would use a force criterion for all structural optimization (setting ediffg negative)

[shilendra]

Dear Prof. Graeme, thank you for your help and support,

I have cleared so many things about ssNEB calculation, Now I am trying it for my system of interest. For rerun calculations, I tried vfin.pl target_dir script but I could not do. I kept vtst script folder in main directory with INACR POTCAR and KPOINTS then ran command
perl vtst/vfin.pl target.

It was showing no error and it is generating empty target dir. How does use this script ?

Thank You
Shilendra

[graeme]

*Make sure that the scripts are in your path*. (you should not have to call them from a specific directory). Then, in the main directory of your calculation, run

vfin.pl dir

where dir is the name of the sub directory which will contain the calculation information. Then, you should be set to continue your calculation from the same main directory where you run the command. In this way, you can conveniently run many continuations and have the information from each run in a subdirectory that you name appropriately.

[shilendra]

Dear Prof. Graeme,
As per above discussion I was facing problem one intermediate image having lower energy than endpoint energy. As you suggested repeat the restart process until the ssNEB finishes after one iteration. I repeated calculation 5 times after that job finished in 2 iteration (NSW=2). After that it is again increasing NSW for required accuracy.

I am confused here which result I should consider because in each repeat calculation I am getting different activation energy and it is increasing with repeatation.

Upto 3 time repeat calculation I am getting images 3 having climbing image and after that image 2 is having climbing image i.e. image 2 is having more energy among all images.

I am getting activation as 0.01, 0.035, 0.053, 0.064, 0.071, 0.075 eV etc..

I have attached here all compiled results of energies and graphs.

Thank You
Shilendra

[graeme]

Good to see these results. And thank you for posting them so that other can see this clear example.

The barrier will change as you approach the converged result, but once the calculation is converged in one iteration, it will not change any more. This is due to the simple fact that the geometry will not move in a single iteration and so any subsequent calculation will be identical.

As I mentioned before, doing accurate calculations with high cutoffs and kpoint meshes will make this repetition of the calculation much less important, but at the expense of a higher computational cost for each optimization of the band.

[shilendra]

Dear Prof. Graeme thank you for your help and support !

As I understood from your post final converged result (NSW=1) should be consider in case cell is changing shape (i.e. POSCAR and CONTCAR having much difference or any other criteria for knowing cell is changing shape) and calculation can be done for default ENCUT.

In my case on repetition I got converged result at NSW=2 and further repetition it is converging at NSW=26. So I have to repeat calculation upto converged result at NSW=1.

Thank You
Shilendra

[shilendra]

Dear Prof. Graeme,

I learned ssNEB calculation on Na system. Now I am doing ssNEB calculation on my system of interest rare earth vanadates. My unit cell is having 24 atoms monoclinic and tetragonal geometry. I am taking 5 images but calculation is not converging. In my machine user has limit one job for 96 hours after that machine kills job. I have submitted 3 time job after CONTCAR to POSCAR still My job is not converging.

Where is problem ? For restart job I did only CONTCAR to POSCAR nothing I changed.

Thank You
Shilendra

[graeme]

I would need to see the calculation files to understand what is going on; if you post them (without CHG* WAV*) I'll take a look.

[shilendra]

Dear Prof. Graeme,

My files after compressing (removing CHG and WAV) are 368 Mb but when I am attaching it is showing error. What should I do ?

Thank You
Shilendra

[graeme]

I'm curious what error you are getting. Anyway, I upped the upload limit from 512MB to 1GB. Otherwise, a dropbox link would work.

[shilendra]

Dear Prof. Graeme,

I have shared you files on DROPBOX.

https://www.dropbox.com/home

Thank You
Shilendra

[graeme]

You will need to take a very different approach with this calculation.

The first thing that I noticed is how many atoms are moving in a complex fashion in your initial path. Do you really believe that this is the simplest path between your initial and final states? Consider reordering the atoms in your final state with respect to your initial state so that the atoms do not have to move as far in your initial path.

The poor initial path is reflected in the initial forces -- about 3000 eV/Ang for image 3! If you look at which atoms have high forces, I expect that you will see at least two atoms running into each other in your initial path. Again, see if you can improve it.

The optimizers can not easily deal with such high forces. If you really want to try converging such a path, you will need to drop your time step to about TIMSTEP=0.001. Then, do much shorter calculations. The one you sent me goes crazy after the first two iterations; the calculation can not recover after this so that the remaining 4998 iterations are wasted. Instead, just run about 100 iterations with a small time step and check for a systematic lowering of the forces. Check that your path is becoming reasonable (visually). You may have to include more images for a smooth path of such a complex transformation. Look for ways in which the path is converging which might give you clues to lower energy mechanisms. If intermediate minima show up, minimize them and consider dividing your path into two (or more) elementary steps. Once the forces are lower than 1 eV/Ang, you can use a larger time step to speed up convergence.

In summary, the potential energy surface for solid-solid phase transitions can be complex and you need to navigate it with care. Do careful small steps and make sure that everything is reasonable (continuous paths, sensible energy, forces and geometries, and a systematic lowering of the forces upon optimization). Look at the paths and adjust your endpoints or atomic identities if a lower energy and simpler path becomes apparent.