Questions on OCPNEB calculator used in CatTSunami

1

Hi,
I have been trying to get the OCPNEB calculator to carry out TS search for a test case to crosscheck with the DFT data I got. I am following the methodology in the CatTSunami tutorial (CatTSunami Tutorial). I have got a couple of questions with regard to this:

  1. While instantiating reaction class, how can I work with a reaction that is not a part of your current database? I am trying to work with an association reaction that is not there in the current list of dissociation reactions. Is there a way I can encode my reaction so that it is readable by the calculator
  2. Can I bypass AdsorbML and still use the OCPNEB calculator? I have the initial and final state POSCARs with me and wish to use the OCPNEB calculator on them to find the transition sate.
2

Hello,

Thanks for your interest in CatTSunami.

  1. I have been meaning to make a workbook with documentation to address this. Let me work on that tomorrow.
  2. Please see this notebook. It covers the use case you are interested in. Particularly the ā€œIf you have a proposed initial and final frameā€ section https://github.com/FAIR-Chem/fairchem/blob/main/src/fairchem/applications/cattsunami/tutorial/fairchem_models_for_nebs.ipynb
2 Likes
3

I opened a PR to add the mentioned notebook. If you have any questions please let me know.

https://github.com/FAIR-Chem/fairchem/pull/882

1 Like
4

Hi Brook
Thanks for your quick response.
I tried running the code in the workbook linked here. I tried running it both ways i.e.; starting with a trajectory and also with initial and final frames. I am getting type error in both cases when OCPNEB is called: TypeError: OCPTrainer.init() got an unexpected keyword argument ā€˜loss_fnsā€™.
I havenā€™t explicitly given any loss_fns so I donā€™t understand where the problem is. Could you please help me with this.
I am also attaching a screenshot for your reference
Thank you

Screenshot 2024-10-21 at 9.23.20 AM
Screenshot 2024-10-21 at 9.23.20 AM2268Ɨ1394 319 KB

5

Hi! It looks like youā€™re on an older version of the fairchem repo. Can you try pulling the latest version and trying again. fairchem/src/fairchem/applications/cattsunami/core/ocpneb.py at 20d2798246eda4bcceec0153a059b4c51623a949 Ā· FAIR-Chem/fairchem Ā· GitHub correctly shows the updated arguments.

1 Like
6

Hi Muhammed,
Thanks! This solved the issue.

Tiny suggested correction: In the workbook shared (fairchem/src/fairchem/applications/cattsunami/tutorial/fairchem_models_for_nebs.ipynb at main Ā· FAIR-Chem/fairchem Ā· GitHub), ā€˜calcā€™, called in the visualizing block, is not defined within this code. I see that it is defined in the main CatTSunami tutorial workbook (i.e.; calc = OCPCalculator(checkpoint_path = checkpoint_path, cpu = cpu)). This can be added to the current workbook as well, to make it easier for the users to follow.

1 Like
7

For my system, the activation barriers evaluated using OCPNEB are off by about 1-1.5 eV compared to the values from DFT based CI-NEB calculations. What could be some possible reasons for this? Also, could you please suggest some things I can try to improve the accuracy for my system.
Thanks