KERNEL_TRUNCATION for charged system

[jess_white]

Hi Henrique,

Yes sorry about the large amount of data I have given you. So the the diopside one I saw from a previous forum post (https://vasp.at/forum/viewtopic.php?p=30167) that it suggested to increase the vacuum (this was before 6.5.1) as there was a bug so I decided to see if that was the case (it was not). I have just looked at my INCARs for diopside surface and they both have the inclusion of GGA and IVDW. Its because for this particular supercomputer I use its dependent on where I put the tags for dispersion forces or if gets an error. That's why there can be two GGA or IVDW and one of each is cancelled out. My apologies for that confusion.

That is correct I want to calculate the adsorption energy using that formula

For all the other minerals the vacuum is not as large - it was just for diopside so I don't think that is the issue and I also did diopside with the smaller vacuum previously and gained the same results. What I find interesting is lizaradite showed great results which is making me think it is related to the z axis.

Thanks so much for all your help!

[jess_white]

Hi Henrique,

Sorry for the large and overwhelming amount of information. So the IVDW and GGA tags should be the same for both - in the INCAR I just have to put it in specific places for them to work when I use one of the supercomputers or I get an error. So there may be multiple GGA and IVDW tags with some cancelled out (I know I should probably just delete the commented ones out to make it less messy ). For the vacuum I decided to increase it for diopside to see if it would help as I saw this previous thread (https://vasp.at/forum/viewtopic.php?p=30167). I have also done diopside with normal vacuum and gain same results. All the other minerals have the normal vacuum. What I find interesting is that lizaradite gains normal results, which leads me to believe the problem lies with the z axis.

That is correct with the adsorption energy and how I am calculating it

Sorry for this super hard problem and I appreciate all the help!

[henrique_miranda]

No problem, I am trying to help

Maybe you can try, using the exact same INCAR:

1. Take the supercell with 26A vaccum with the molecule and compute total energy without ionic relaxation (IBRION=1) with and without kernel truncation.
2. Remove the molecule from that same POSCAR and compute total energy without ionic relaxation (IBRION=-1) with and without kernel truncation.

From these 4 values of the total energy you should be able to get a value for the adsorption energy with and without kernel truncation.
Are they similar or are they orders of magnitude different?
Then you could try to look at the convergence of the adsorption energy with and without kernel truncation as a function of the vaccum in the supercell.

After this is done then you could try activating ionic relaxation for each different vaccum distance and look at this convergence plot again.
Only once you activate ionic relaxation you will get a realistic value for the adsorption energy.

[jess_white]

Hi Henrique,

Thank you for the suggestions So I have run the ones with the carbonate adsorption (1 and 1 kernel) with an IBRION = 1 and the surface with and without kernel with IBRION = -1. The addition of kernel leads to odd results again. Are you suggesting I keep increasing the vacuum calculations and see if I will get a decent result? For the current ones, yes they haven't reached convergence (especially the 1_kernel) but I still think the adsorption energy will be around -200 eV.

[jess_white]

and here are the surface calculations

[henrique_miranda]

There was a typo in my previous post.
I wrote "without ionic relaxation" but in one of the cases I accidentally wrote IBRION=1 while I mean IBRION=-1.

[jess_white]

Hi Henrique,

My bad I should of spotted that out! Here are the calculations all done with IBRION = -1.

Surf = -1132.01
Surf_kernel = -10767.84
adsorption = -1169.87
adsorption_kernel = -11499.90

Adsorption energy without kernel = -2.37 eV
Adsorption energy with kernel = -703.5 eV

I have attached the files as well But do you want me to increase the vacuum of the kernel slowly and see if it will get better results?

[jess_white]

and here are the surface files

[henrique_miranda]

Ah perfect!
These are exactly the calculations I had in mind. Thanks for sharing them.
Now, I would say that the results are not completely crazy.
They are not good but that is to be expected because the relaxation was deactivated.

The problem is that one needs to use truncation in both calculations to be able to compute energy differences as is required to get adsorption energy.

I think the next step is indeed plotting the two values of the adsorption energy (with and wihout coulomb truncation) as a function of vaccum distance.
I would maybe try these values: 35A 40A 50A 55A 60A.
I believe the 26.35A is still small for the kernel truncation (the truncation length is z/2 which means 26.35/2 = 13.175 but that is too close to the width of your layer. I think the adsorption energy with the coulomb truncation will stabilize w.r.t vaccum at around 50A. The calculation without truncation might converge very slowly but if everything is correct you can extrapolate it to infinite vaccum and you should get the same value as when you use coulomb truncation.

Once this is clear don't forget that to obtain physical results you need to activate the ionic relaxation. I would suggest that you do this relaxation with coulomb truncation at the smallest possible vaccum such that the adsorption energy does not depend on it anymore.

[jess_white]

hi Henrique,

Thank you for the suggestions - I am currently still running them. Just wondering if the calculations need to be converged completely? Only because the kernel/adsorbate with differing vacuum size will take a while from the early calculations. I haven't calculate the adsorption energy yet but I may have a try without the calculations not converging just to see if I can already see a general trend or not.

[henrique_miranda]

Hi Jess,

What do you mean the calculations need to be converged completely?
I would suggest that you first do a test with varying vaccum sizes WITHOUT doing ionic relaxation (IBRION=-1).
The structure should be sensible i.e. close to relaxed, otherwise you cannot interpret the value of the adsorption energy converged with respect to vaccum.
To be more precise: if you take the surface, introduce the molecule in some position without relaxation you are not sure that the adsorption energy will be any close to the experimentally measured one. You should realax it but it does not need to be done very strictly.
Or do you mean converged with respect to k-points, encut etc?
I would say no. You can converge the vaccum and assume the other parameters of the calculation can be converged separately.