Why are phonon calculations so slow?
This is because an automated supercell approach to computing the dynamical matrices is used. The idea is to compute the phonon modes on a relatively coarse grid (eg. 4x4x4) and interpolate to arbitrary q-points. Always choose the grid numbers to be multiples of each other i.e. 4x4x8 is efficient but 2x3x5 will be very inefficient. By far the most time consuming step is the diagonalisation of the supercell Hamiltonian. Therefore the planewave cut-off variable rgkmax should be made as small as possible (but no smaller!). Dynamical matrix calculations can also be spread across a cluster of nodes which share the same directory.
