get_mbd_rpa_energy Interface

public interface get_mbd_rpa_energy

Get the MBD energy by the RPA method.

The energy is obtained from the eigenvalues $\mu_k(\mathrm iu)$ of the symmetrized dipole matrix $\mathbf M(\mathrm iu)=\mathbf A(\mathrm iu)^\frac12\mathbf T\mathbf A(\mathrm iu)^\frac12$, with $A_{ij}(\mathrm iu)=\alpha_i(\mathrm iu)\delta_{ij}$,

$$E=\frac1{2\pi}\int_0^\infty\mathrm du\sum_k\log(1+\mu_k(\mathrm iu)) =\frac1{2\pi}\int_0^\infty\mathrm du\,\log\det(\mathbf 1+\mathbf M(\mathrm iu))$$

The derivatives use $\partial\mu_k=\mathbf c_k^\dagger(\partial\mathbf M)\mathbf c_k$, which upon summation over the eigenvalues gives the weighted resolvent $\mathbf B(\mathrm iu)=\mathbf C\operatorname{diag}( g'(\mu_k))\mathbf C^\dagger$, where $g(\mu)=\log(1+\mu)$ is the per-eigenvalue energy contribution,

$$\partial E=\frac1{2\pi}\int_0^\infty\mathrm du \operatorname{Tr}\big(\mathbf B(\mathrm iu)\partial\mathbf M(\mathrm iu)\big)$$

Both the explicit derivatives (coordinates, lattice vectors, van der Waals radii, $\mathbf q$) and the implicit derivatives with respect to the dynamic polarizabilities are returned. The latter propagate both through the $\sqrt{\alpha_i\alpha_j}$ prefactor and, when the damping depends on it, through $\sigma_{ij}(\alpha)$. With eigenvalue rescaling (mbd_calc_params_t) the contribution becomes $g(\mu)=\log(1+\lambda(\mu))-\lambda(\mu)$, with $\lambda=-\operatorname{erf}(\tfrac{\sqrt\pi}2\mu^4)^\frac14$ for $\mu<0$, and the same weighted resolvent is used.

The real-typed version is equivalent to $\mathbf q=0$.

Module Procedures