Dear VASP users,

I am currently looking at some structures of transition metal fluorides (e.g. TiF3, VF3) and want to extract the frequencies of the compounds to confirm local minima. With the compounds having Mott insulator characteristics similar to e.g. NiO, I typically apply DFT+U when optimizing these structures and calculating the DOS. I'm interested in using the DFPT formalism for getting the frequencies of my optimized structures, partially because I find it more elegant, but mostly since I could efficiently use the results in junction with the Phonopy software (using a supercell) to get non-Gamma-point frequencies.

When doing test calculations on TiF3, I often get more or less large non-acoustic imaginary frequencies when using IBRION = 8. In other calculations I only get the typical 3 small acoustic imaginary freq's, but some real freq's differ from my reference finite displacement calculation (IBRION = 6) by up to a 100 cm^-1.

In the past, I have found the SCF convergence of TiF3 quite tricky, often ending up in local minima. So, I could keep on playing with different parameters like I've done for a couple of days now, but then I had a look at this paper by Coccocioni and coworkers where they needed to specifically implement DFPT+U, in their case in Quantum Espresso (https://journals.aps.org/prb/pdf/10.110 ... .84.161102). This brought up the possibility that the linear response calculations neglect some parts of the Hubbard correction to the Hamiltonian also in VASP. While googling around, I also didn't stumble upon any examples of people using DFPT+U

So my question is simply, will DFPT+U work at all in VASP?

Cheers,

Stefan