CO Oxidation NEB Help - Jobs won't converge

[cswang]

Hi all,

I'm really desperate because my hours expire in March.

I am trying to test out the CO oxidation reaction, which is enhanced by a single atom catalyst - Pt adsorbed on single-vacancy graphene. Basically, I want to emulate the results in the following paper: http://pubs.rsc.org/-/content/articlepd ... c2cp41441d (see Figure 5b)

However, about after a month, I'm still struggling to get this reaction to occur.
- One issue is that the job still hasn't finished even though it's been running for multiple days.
- Another issue is that despite starting with inputs such that the transition state should be one of the middle images, it keeps getting pushed to either the first/second image or the last/second-to-last one. I actually turned off LCLIMB with the latest attempt because I read that I can turn this parameter on when I am closer to reaching convergence, but I've had this set to .TRUE. for the first 48 hrs of the job...
- Sometimes an atom will move in a wacky way and the energy of one image will suddenly jump 10 eV... I keep having to stop it when one of the images goes wrong, and restart the job entirely.

I tried to simplify this reaction and just look CO + O2 --> CO2 + O. This reaction was also taking forever, so I coarsened the job by using Selective Dynamics and fixing the XYZ of both oxygen atoms int he O2 molecule, but this has still been running for 25 hrs, and the end doesn't seem to be in sight. I'm just so frustrated...

If it makes a difference, I'm running on XSEDE Stampede 2. I am also a little bit confused about the job submission (I submit with the file 'test.job') because it seems different to me than the original Stampede...

I'm attaching all relevant files. I'd really appreciate any help I can get on the matter.

Thanks,

CSW

[graeme]

Please include the OUTCAR files in your calculations. In fact, include all files except the WAV* CHG* files so I can see what is going on.

[cswang]

Wow, thanks so much for taking a look.

Attached are the files for both jobs. The OUTCARs are only the most recent ones since I've reset the jobs after running into the time limits and overwrote the old files.

[cswang]

Sorry to for the trouble, but I realized that the very original POSCAR files weren't included in the last post.

The attachment here shows what I originally started off with. However, the CO molecule kept spinning in vacuum in the early images and then suddenly forming CO2 in the latter images. Because of this, I tried deleting some of the repetitive images and interpolating between the images that were closer to the transition point, which is what is actually attached in the post above (Pt_SVG_CO_ox.zip).

[graeme]

Your initial path has a lot of unnecessary molecular motion. Start again with a smooth path connecting your initial and final states. If atoms move to close to each other (e.g. <1 Ang), either move them apart manually, or use our pushapart script to do this. Make sure that you have the optimal assignment of O atoms between your initial and final states. A poor choice will result in, for example, unnecessary rotation of the CO2 or CO molecule.

Next, using the IOPT=3 TIMESTEP=0.02 settings until the forces drop below 1 eV/Ang. Then, either continue using a larger TIMESTEP of 0.1, or you can try switching to a more efficient optimizer. Check for smooth convergence. If you ever see jumps in the force above 1 eV/Ang (as you can see in the attached calculation) stop and restart that run with more conservative optimizer settings.

[cswang]

Thanks for the help, graeme. Originally the path was somewhat straight but a lot of this spinning at the beginning occurred during NEB runs.

Were you able to take a look at the CO + O2 --> CO2 + O job (CO_ox.zip)? This one was much more straightforward but still taking forever to finish...

At any rate, I think it's a good idea to start the more complicated job from scratch. I will run the attached job when XSEDE is back up from maintenance. Do you mind taking a look if the input seems reasonable?

I really appreciate all the help.

[graeme]

You need 00 and 06 directories with your endpoints.

[cswang]

Hi, sorry for forgetting to include those. Here's the full job, which I ran after starting from scratch. As you can see, I'm still having issues. I'm not sure what's wrong with my input such that the molecules keep rotating the way that they do. None of these images look like they are anywhere near finding the transition state...

Any suggestions?

[attachment=0]forum_help.zip[/attachment]