ABINIT Discussion Forums

Hi,
suppose I have two GGA functionals 'A' and 'B'. 'A' yields better cell parameters and bulk modulus than 'B', while 'B' is better for energetics (energy differences). What would be the best choice ('A' or 'B') to treat vibrational properties ?
regards,

Hi

I would say B, because you want to calculate second derivatives to the total energies, so I guess it would be more suitable to have a functional that correctly reproduce energetics.

I think, however, that the best thing to do would be to run the calculation with both functionals and compare the results obtained.

Boris

hi boris,

I think, however, that the best thing to do would be to run the calculation with both functionals and compare the results obtained.

I agree ... but suppose that you are working in a predictif way without any experimental data for such properties, or suppose that you just want to have, at least, a qualitatif idea of what calculations would yield from a physical point of view.

I would say B, because you want to calculate second derivatives to the total energies, so I guess it would be more suitable to have a functional that correctly reproduce energetics.

for my part, I would say 'A' for the following:

1 - As you pointed out, we are interested in energy second derivatives and the bulk modulus is just one of those energy second derivatives. One would say that the accuracy of, for example, phonon frequencies would be similar to the accuracy of the bulk modulus.

2 - The phonon frequencies are closely related to the cell parameter in a way that if a functional, like LDA, underestimate it the phonon frequencies will be overestimated systematically and vice-versa. This systematic error may be reduced if we are closer to the experimental cell parameter.

corrections, warnings and comments are welcome.
regards.

Hi

Your points make sense.

Have you calculated the bulk modulus by fitting to the Murnaghan equation of state or by calculating the elastic tensor using linear response ?

Hi,

I am just fitting to some form of EOS.
For linear response I think it's not yet implemented wiht PAW (I am wondering when it will be available !)

regards.