AIMD mixing enthalpy - wrong results.

[wojtek_gierlotka]

Dear All,

I try to calculate mixing enthalpy of the liquid alloys Cu-Mg using AIMD. The application of the AIMD to this kind of problem is well documented in literature and looks the method works...but not for me. The enthalpy is simply calculated by the equation deltaH = H(alloy_A-B)-n*H(A)-m*H(B) where n, m are stoichiometric coefficients, and enthalpy H is approximated by energies. After calculating energies for Cu, Mg, and the alloy Cu06Mg04 I got mixing enthalpy 2 times bigger than it appears from experiment. I use a supercell with 125 atoms, the initial atom displacement for the alloy was prepared by SQS method (using SQSgenerator code). Before the final calculations the size of supercell was adjusted to have the external pressure close to 0.
The INCAR is listed below:
ISMEAR = 1
IBRION = 0
MDALGO = 2
ISIF = 2
SMASS = 0.0
SIGMA = 0.2
LREAL = Auto
ALGO = VeryFast
PREC = Accurate
ISYM = -1
TEBEG = 1400
NSW = 15000
POTIM = 2
NCORE = 24
LWAVE = FALSE
LCHARG = FALSE
ADDGRID = FALSE
EDIFF = 1E-4
ENCUT = 400 eV
NELM = 100

The calculations were done in Gamma point only. I will be thankful for all suggestions that can improve the results.

cheers
wojtek

[svijay]

Dear wojtek,

There are several reasons why the results of an ab-initio molecular dynamics calculation do not always match experiments. For instance, have you checked the convergence of your NVT calculation (particularly the energy) with the timestep? Another possible reason for a mismatch with experiments could be your DFT computational setup (choice of k-points, electronic convergence through EDIFF, etc). If you would like something more specific, please feel free to share an example with outputs (as discussed here: https://www.vasp.at/wiki/index.php/Mini ... 20possible.)

Sudarshan

[wojtek_gierlotka]

Dear Sudarshan,

thank you for your answer. As you can see from the INCAR the calculation was set to 30ps with a step 2fs. I'm attaching the OUTCAR, POSCAR, and KPOINTS. Due to the very big size of the OUTCAR I've remove part of the file and have left only first and last 200 iterations. I've used the PAW_PBE Cu 22Jun2005 and PAW_PBE Mg 13Apr2007 pseudo potentials. Thank you for your help.

cheers
wojtek

[svijay]

Dear wojtek,

Thanks for sharing these inputs - I can't see the convergence of the calculation based on a shortened OUTCAR file. I suggest plotting the energies TOTEN, EKIN and ETOTAL against the total number of iterations and checking if the cumulative averages converges to within some threshold. If it does, that would rule out one of many possible issues with matching with experiments.

[wojtek_gierlotka]

Thank you for the answer. Looks like the total energy didn't converged. Toten and kinetic energy are fluctuating. Please, take a look at the attached pictures. I'll be very thankful for suggestions.

cheers
wojtek

[svijay]

The cumulative average of these quantities must converge, the quantities themselves are expected to fluctuate (https://www.vasp.at/wiki/index.php/Category:Ensembles might provide some background).

[wojtek_gierlotka]

Thank you. I'll work on that.

cheers
wojtek

[wojtek_gierlotka]

To solve the problem with convergence of total energy I've changed the size of the system to 64 atoms and I've tried to change some tags (like increasing encut or change the POTIM) but I'm still not able to get correct results in a case of supercell of 64 Cu atoms. During the first run I found that there is a band crossing between old and new wave function so I've added the IWAVPR = 11 tag. I've ran the system for 22 ps (4, 6, 12 ps with copying CONTCAR to POSCAR) and the total energy still decreases. Do you have any suggestion what else I can do? Thank you.

INCAR:
ISMEAR = 0
IBRION = 0
MDALGO = 2
ISIF = 2
SMASS = 0.0
SIGMA = 0.1
LREAL = Auto
ALGO = VeryFast
PREC = Low
ISYM = 0
TEBEG = 1800
TEND = 1800
NSW = 12000
POTIM = 3.0
NCORE = 24
ENCUT = 400
IWAVPR = 11
NELMIN = 6
NELM = 100


cheers
wojtek

[svijay]

Dear wojtek,

Based on the INCAR file that you provided
(1) It appears that you are taking a large time step (POTIM) with a perhaps inaccurate forces (PREC). Have you tested convergence of your forces?
(2) Do the cumulative averages decrease? I would image the system is not yet equilibriated

Sudarshan