Sampling phonon spectra from molecular-dynamics simulations

Sampling phonon DOS from molecular dynamics simulation

The phonon density of states can be obtained as the power spectrum from the normalized velocity auto correlation function. The normalized velocity auto correlation function for a $N$-particle system is given by \begin{equation} f(t)=\sum_{s=1}^{types}f_{s}(t)=\frac{\langle \sum_{s=1}^{types}\sum_{i=1}^{N_{s}}\mathbf{v}_{i}(\Delta T)\mathbf{v}_{i}(\Delta T+t) \rangle}{\mathbf{v}_{i}(\Delta T)\mathbf{v}_{i}(\Delta T)}. \end{equation} The brackets $\langle ,\rangle$ denotes a thermal average which has to be computed over different trajectories and and starting times $\Delta T$ within each trajectory. The sum over $i$ runs over the atoms within each species and the sum $s$ is over all atomic species contained in the simulated system. From this the phonon density of states is obtained by computing the power spectrum of $f_{s}(t)$: \begin{equation} g(\omega)=g_{s}(\omega)=\sum_{s=1}^{types}g_{s}(\omega)=\left| \int_{-\infty}^{\infty}\sum_{s=1}^{types}f_{s}(t)e^{-i\omega t}\right|^{2}. \end{equation} To properly sample the phonon density of states from molecular dynamics simulations the following the steps have to be accomplished:

code	publication
step 1	generate starting structures in NVT simulation
step 2	generate a NVE simulation for every starting structure to obtain velocity fields
step 3	compute normalized velocity auto correlation function for every NVE simulation
step 4	compute power spectrum for every normalized velocity auto correlation function
step 5	compute averages and check for convergence

Related tags and articles

Molecular-dynamics calculations, Computing the phonon dispersion and DOS, IBRION, MDALGO, ISIF, TEBEG, NSW, POTIM, ANDERSEN_PROB, QPOINTS, LPHON_DISPERSION, PHON_NWRITE, LPHON_POLAR, PHON_DIELECTRIC, PHON_BORN_CHARGES,PHON_G_CUTOFF