How to extract the total energy from ACFDT-RPA calculations?

[vaspGG]

Dear Admin and vasp users,

The recipe to calculate ACFDT-RPA total energy can be found on vasp website, and it is not difficult to follow these instructions. My confusion is how to get the total energy I need?

E(RPA)=E(C)+E(EXX), where E(C) means the correlation and E(EXX) means the exchange.

(1)For E(C), I assumeit is from "Fourth step" (ACFDT) calculation, in the OUTCAR, there is a line "linear regression converged value", I think the following energy is PRA correlation energy. Right?

(2) For E(EXX), it should be from the "Second step" (HF) calculation? But I do not know it is the "energy (sigma->0)" in OUTCAR, or it is just the " -exchange EXHF = 199.53197515" in the electronic steps which is positive value?

Thanks in advance for any response!

Best Wishes!

[vaspGG]

One further question, for the additional steps for metals: "Some issues particular to ACFDT-RPA calculations on metals"
http://cms.mpi.univie.ac.at/vasp/vasp/g ... rgies.html


In the OUTCAR, I got
HF-correction -197.1577315 -197.1957935 -0.0380620

So, correction energy for Eexx us -0.038 eV, am I right ?


Any responce to the above two posts are appreciated!

Thanks!

[kelum]

Dear vaspGG,

the RPA correlation energy is the first number on the "converged value" line,
the EXX energy is the "energy (sigma->0)" value obtained with ALGO=EIGENVAL on DFT input orbitals.
The "-exchange EXHF" is just one contribution to the EXX energy.
The HFc contribution for metals is small, it is the -0.038 eV in your calculations (the last number printed out on the line).

Best,
Jiri

[vaspGG]

Dear vaspGG,

the RPA correlation energy is the first number on the "converged value" line,
the EXX energy is the "energy (sigma->0)" value obtained with ALGO=EIGENVAL on DFT input orbitals.
The "-exchange EXHF" is just one contribution to the EXX energy.
The HFc contribution for metals is small, it is the -0.038 eV in your calculations (the last number printed out on the line).

Best,
Jiri

Dear Jiri,

Thanks a lot for your reply, which definitely has solved my problems.

One more thing is that for the second step, where the Exx is calculated, I think the NSW=0 should be added into the INCAR to make sure that the single electronic steps inludes the HF. Otherwise, the first three electronic steps normally are done without HF calculation even LHFCALC=T is set for VASP. Am I right?

Best

GG

[kelum]

Dear GG,

in the EXX part only single step is made, where the non-self-consistent calculation of the HF-like energy is performed using the DFT orbitals.
The ALGO=Eigenval sets NELM to 1, so that only one electronic step is made, this step uses HF, if you set LHFCALC=.TRUE. and so on,
as in the manual. In the first step the HF Hamiltonian is calculated with the input orbitals, as it would be even for self-consistent calculation.
In the few steps afterwards done in a self-consistent calculation, it is not recalculated, but kept constant to speed things up (or something
similar to this is done).
NSW gives the number of ionic steps, zero is the default value, and one should not change it since all the EXX and RPA calculations need
to be done on the same geometry.

Best,
Jiri

[vaspGG]

Dear GG,

in the EXX part only single step is made, where the non-self-consistent calculation of the HF-like energy is performed using the DFT orbitals.
The ALGO=Eigenval sets NELM to 1, so that only one electronic step is made, this step uses HF, if you set LHFCALC=.TRUE. and so on,
as in the manual. In the first step the HF Hamiltonian is calculated with the input orbitals, as it would be even for self-consistent calculation.
In the few steps afterwards done in a self-consistent calculation, it is not recalculated, but kept constant to speed things up (or something
similar to this is done).
NSW gives the number of ionic steps, zero is the default value, and one should not change it since all the EXX and RPA calculations need
to be done on the same geometry.

Best,
Jiri

Dear Jiri,

Again, thank you for the reply. I was wrong about the NSW=0 has to be set in step 2. In fact, when the relaxation is done using HF, NELMDL is defaulted to be -5, which means in the first 5 steps the hamitonian is fixed and no HF calculations is done, after that, HF will be added.

In the fourth step (ACFDT step), the SIGMA should be set as 1/4 of the KS band gap, but for the metal without gap, SIGMA = 0.05 is OK in this case? According to the website ISMEAR = -1 has to be included in the INCAR, right?

Best Wishes

GG

[juhL]

Hey everyone,

I would like to take the last question by GG concerning the smearing for metals, asking a few practical questions concerning setup testing myself:

According to the work of Harl (10.1103/PhysRevB.81.115126), the best strategy seems to be to converge the k-point grid for metals for E_EXX' and E_c together. In addition, metallic cases should be corrected with the E_HFc term, i.e., I assume that E_EXX' = E_EXX + E_HFc. This work also shows some testing with respect to the smearing parameter (sigma) for Methfessel-Paxton smearing and results seem to be quite insensitive (although the range of tested values seems to be odd, as one usually applies values around 0.2 eV and below, whereas in this work values 0.2 - 0.8 eV are shown). One can, of course, do some testing for every specific case, however, I think it's best to keep the same sigma value throughout all steps (1-5). Does anyone have some experience/arguments?

A second question is concerning the cutoffs. It is suggested to use different cutoffs during the calculation of the EXX energy and the correlation energy. Is it correct that the same setup should be employed in step 2 (the EXX energy according to HF) and in step 5 (the correction term)?