forces in finite-temperature RPA

[shixubenjamin1]

Dear VASP forum,

I am interested in running finite temperature RPA and had some enquiries about utilising it in VASP:

• Is is possible to compute forces with finite temperature RPA?

• What is the cost of finite temperature RPA relative to standard RPA?

• Does the spline interpolation to speed up k-point convergence in standard RPA work with finite-temperature RPA as well?

Cheers,
Benjamin

[merzuk.kaltak]

Dear Benjamin,

you can compute RPA forces using the finite temperature formalism in vasp as of version 6.3.2, as mentioned elsewhere.
However, only systems with a finite band gap (i.e. cold electronic temperatures) are supported.
In fact, the finite-temperature formalism is particularly recommended for systems characterized by very small, yet finite, band gaps.
However, for metallic systems with partial occupancies (high temperatures) the forces are still under development.

The computational cost of the finite-temperature RPA is comparable to that of standard RPA, as detailed in our publication.

The spline interpolation for accelerating k-point convergence is compatible with the finite-temperature formalism.
For systems with small band gaps we advise using this method in preference to the k-p perturbation theory utilizing WAVEDER.

[shixubenjamin1]

Thanks for the information. If my aim is to study a range of systems (i.e., insulators and metals) within a consistent manner with RPA, I suppose the finite-temperature formalism is the only means to do this? Can you comment whether there are additional errors which arise within large band gap insulators (where standard RPA is appropriate) for using the finite-temperature formalism over the standard RPA?

[merzuk.kaltak]

The finite temperature RPA formalism yields the same results as the T=0 formalism for large band gap insulators and small electronic temperatures (e.g. SIGMA of the order of 0.05 and smaller).
If this is not the case, one should lower the value of SIGMA and inspect the occupancies to make sure they are integers (either 1 or 0). Bear in mind, lower values of SIGMA require more NOMEGA frequency points for the same integration error. This error is typically reported by following lines:

Code: Select all

 quadrature errors minimized for energies in [ 0.000E+00, 0.435E+03 ]
 time grid (T>0) determined with error: 0.5507E-07
 bosonic grid (T>0, re) determined with error: 0.1395E-06