Phonons from density-functional-perturbation theory
This approach determines the Hessian matrix (matrix of second derivatives with respect to the position of the ions) using density-functional-perturbation theory (DFPT). IBRION=7 does not apply symmetry, whereas IBRION=8 uses symmetry to reduce the number of displacements. The output is similar as for IBRION=5 and 6. Specifically, the second derivates of the total energy with respect to ionic displacements (interatomic force constants) and the mixed second derivative with respect to the strain and the ionic displacement (internal strain tensor) are evaluated. Although the contributions from the ionic relaxations to the elastic tensor are calculated, the ion-clamped elastic tensor (rigid ion) is not determined. Born effective charges, piezoelectric constants, and the ionic contributions to the dielectric tensor are calculated if LEPSILON=.TRUE. is specified in the INCAR file.
In general, the DFPT routines in VASP are somewhat rudimentary and only support displacements commensurate with the supercell, i.e., so-called q=0 phonons. In other words, VASP can only determine phonon frequencies at the Gamma point of the supercell. Therefore, the code offers few advantages over the finite differences methods discussed above. In particular, the linear response is limited to LDA and GGA functionals and it does not determine the elastic tensors, since the linear response with respect to the strain tensor is not implemented. The only advantage of the linear response routines is that they eliminate the need to choose the magnitude of the finite displacement. Therefore, it might be helpful to first calculate phonon frequencies using linear response and then switch to finite differences and determine the largest displacement that will produce results compatible with the linear response routines.
A few technical comments are in order at this point. VASP solves the linear Sternheimer equation to determine the linear response of the orbitals. Hence, unoccupied orbitals are not required. Internally, the VASP routines for linear response rely on finite differences in two places:
- The first place is the determination of the second derivative of the exchange-correlation functional: Since most functionals do not support an algebraic determination of second derivatives, VASP always resorts to finite differences to determine the second-order change of the exchange correlation-potential and the PAW one-center terms for each atomic displacement.
- Second, after VASP has determined the first-order change of the orbitals, it computes all second derivatives using finite displacements.
To do this, VASP displaces the selected atom in the selected directions adds the calculated linear response to the orbitals, and finally determines the differences in the forces and the stress tensor for positive and negative displacements. It can be shown that this yields exactly the second-order force constants and the internal strain tensor, respectively.
Furthermore, the Born effective charges are determined "analytically" by contracting the linear response of the orbitals over the "polarization" vector Eq. (30) in Ref. [1]. These should agree well with the Born effective charges that were previously determined when the linear response with respect to external fields LEPSILON=.TRUE. was calculated (there are two different routes to calculate mixed derivatives). The final summary output towards the end of the OUTCAR file writes the Born effective charges determined from the linear response with respect to external fields.
It is possible to obtain the phonon dispersion at different q points by computing the force constants on a sufficiently large supercell and Fourier interpolating the dynamical matrices in the primitive cell.
It is also possible to use phonopy[2] to post-process the results of a finite differences calculation done with VASP.[3]
| Mind: IBRION=7 and IBRION=8 are supported from VASP.5.1 and later versions. |