Activation energy increases at each step of VCNEB calculation

[Mehmet Dogan]

Dear forum,

I'm new to USPEX, and want to do some VCNEB calculations. I tried to run them with Quantum Espresso, but for the reasons I explained here, I haven't been able to. So I'm using VASP, to which I'm also new.

I'm calculating the energy barrier between two phases of solid hydrogen at 500 GPa. I've done this calculation up for ten steps. In the zeroth step (where the intermediate images are just linear combinations of the initial and the final images), the activation energy is 0.2 eV. At each step, it increases instead of decreasing, and reaches 10.0 eV at step 10. This is the opposite behavior to what we would expect, but I couldn't find any reasonable explanation for it.

I've attached the relevant files, but if you need more information, please let me know. Thank you very much in advance for any comments.

Sincerely,
Mehmet Dogan

[Mehmet Dogan]

Dear forum,

I fixed the number of images to 15 and reduced dt to 0.01. However, the barrier still goes up at each step, albeit more slowly (see the attached files). In fact, the enthalpy of each image monotonically goes up. Can you make sense of these results? Any suggestions will be much appreciated.

[Konstantin Rushchanskii]

Hi Mehmet,

Please include INCAR file too. It is possible, that with insufficient basis and settings you have bad (not converged) forces in VASP.

Best,

K

[Mehmet Dogan]

Hi Konstantin,

Thank you for your reply. I've tried the calculation again. First I made sure the starting and the ending images are well-converged, which led me to more stringent calculation parameters (ENCUT, Kresol). However, I still see increasing energy barriers. I'm pretty sure the forces are calculated correctly, but please check the attached files to see if there's anything else I should change.

I'm wondering if this may be due to the fact that during the phase transition I'm computing, very strong bonds are broken so the forces are indeed large, and the code somehow doesn't work very well with large forces? Would that be a possibility? Thank you very much for your time.

Best,

Mehmet

[Konstantin Rushchanskii]

Some extra hints:

1. Make sure you use "hard" H_h pseudopotential. With standard H potential at extreme pressures the overlap of rigid cores could be large, which leads to errors in forces and energies. 

2. You use VCNEB in USPEX. For that you should have information about pressure in OUTCAR. In INCAR I don't see flag ISIF = 2. Use it together with  NSW = 0.

Check the last block here:

https://uspex-team.org/online_utilities/uspex_manual_release/EnglishVersion/uspex_manual_english/sect0015.html

Best,

K

[Mehmet Dogan]

Dear Konstantin,

Thank you very much for your suggestions. I will try (1) and see if there's any significant changes. I'm quite confident about the two endpoints of calculation because I got very similar structures to the ones I got from QE with a norm-conserving pseudopotential. Please see OUTCAR_last for the results of the VASP run that corresponds to Image 10 of the VCNEB calculation (I couldn't attach OUTCAR_first for Image 1 because it was too large). Regarding (2), please see OUTCAR_old which corresponds to image 5 at step 10. It looks like although I didn't define NSW or ISIF, NSW=0 and ISIF=2 are set by default. Thanks for sending me that link though, I will refer to it in later calculations.

I want to get your opinion about a few issues. I looked at my OUTPUT.txt from the USPEX run (which has information on steps 0, 5 and 10) in detail, and I'm confused about two things:

1) Only the external pressure of the first and the last images are fixed at 5000 kbar, and the external pressures of the intermediate images seem to decrease over time, and their volumes increase. For instance, the external pressure of Image 5 goes from 4856 to 4261 to 3794 to kbar (steps 0 to 5 to 10). Its volume goes from 4.71 to 4.95 to 5.17 Angstrom3. I think what USPEX prints as external pressure is the sum of external pressure and Pulay stress (=PSTRESS) in VASP. I thought the VCNEB method kept the pressure constant for all images, which I interpret as the following: Expand or shrink the cell so that the external pressure computed in VASP approaches zero (or external pressure reported in the USPEX output approaches the user-defined pressure, in our case 500 GPa). However, here, it looks like the cells keep expanding and moving away from that pressure.

While writing this, I realized that there is one image (Image 9) for which both step 0 and step 10 OUTCAR files are saved. So I checked how its enthalpy was computed:

Step 0: energy = -3.8869 eV; volume = 4.61; external pressure = -61+5000 = 4939 kbar; enthalpy = 10.5052 eV

Step 10: energy = -5.2754 eV; volume = 5.13; external pressure = -1235+5000 = 3765 kbar; enthalpy = 10.7369 eV

So if I calculate P=(enthalpy-energy)/V, I get P=5000kbar in both cases. So the enthalpy is computed not with the external pressure but with PSTRESS. This may be why the enthalpy is increasing at each step (P is fixed while V is increasing). If I compute the enthalpy with the external pressure instead, I get 10.3306 eV and 6.7832 eV at step 0 and step 10, respectively, but this is not what I need either (since the pressure is not kept constant).

2) How are the "VCNEB forces on lattice" calculated? When I read the VCNEB paper, my understanding was that the VCNEB forces on the lattice and the atoms are computed at constant pressure. So since the first and the last image are fully relaxed at that pressure (enthalpy at 500 GPa has local minima at those configurations), shouldn't the VCNEB forces on the lattice be zero for these images? The VCNEB forces on the intermediate images seem to go down with each step, which makes sense, but this is happening while the pressure is going down instead of staying constant. Perhaps if I let this run continue, at some point the intermediate steps will reach zero external pressure and the VCNEB forces will approach zero, but the final transition state will be computed at zero pressure.

So I think there's something wrong with how pressure is taken into account, but I'm not sure how to fix that, since all VASP runs say PSTRESS=5000 and in the USPEX input we have "500.0 : ExternalPressure". There may be a deeper issue here. Let me know what you think, and thank you very much for your time.

Best,
Mehmet

[sourav guha]

Hello,

Did you manage to resolve the issue with high energy barrier? My calculations shows a significant rise in barrier at around 60/70th step and stays there for atleast the 100thstep.

Regards