3.21 Many-Body Dispersion (MBD) method

The following two keywords activate either the MBD@rsSCS (many_body_dispersion) or the MBD-NL (many_body_dispersion_nl) method. The optional arguments are shared by both methods.

• •

  beta=<real> [default: depends on XC functional] sets the damping parameter $\beta$.

• •

  k_grid=<integer>:<integer>:<integer> [default: taken from k_grid] specifies the $k$-point grid used for sampling the first Brillouin zone in the MBD calculation. The grid is shifted by half a distance between the $k$-points from the $\mathrm{\Gamma }$-point. Note that if either generalized_regular_k_grid or k_points_external are used, then specification of this k_grid option is mandatory. [only for periodic systems]

• •

  freq_grid=<integer> [default: 15] controls the size of the imaginary-frequency grid used for the Casimir–Polder integral.

• •

  self_consistent=<logical> [default: .false.] turns on self-consistency as explained under vdw_correction_hirshfeld_sc. Currently, this works only for MBD@rsSCS.

• •

  vdw_params_kind=<string> [default: "ts"] specifies the set of free-atom vdW parameters (${\alpha }_0$, $C_6$, $R_{\text{vdW}}$) used. "ts" uses the original set of parameters, "tssurf" uses values from the so-called vdW${}^{\text{surf}}$ approach [267] for some of the elements. The used parameters are listed here. This option should be left unchanged with MBD-NL, which is designed to work with the default set of vdW parameters in all cases.

Examples:

• •

  many_body_dispersion (this uses the default settings)

• •

  many_body_dispersion_nl beta=0.8 k_grid=3:3:3 (explicit settings of the damping parameter and of a $3\times 3\times 3$ $k$-point grid)

Caveats:

• •

  The many_body_dispersion approach, i.e., the MBD@rsSCS approach, should work reliably for dispersion bonded systems. However, the associated Hamiltonian can have negative eigenvalues for systems that are more strongly bonded, typically ionic or metallic systems. If this happens, this would be the result of too high polarizability as predicted by the Hirshfeld volumes. The code will stop if this happens.

• •

  The many_body_dispersion_nl is a more advanced scheme and is expected to do a much better job of not overpolarizing parts of the system. Negative eigenvalues should occur much less frequently (hopefully only in corner cases).

• •

  If negative eigenvalues of the MBD Hamiltonians occur, the best path forward is an assessment if this particular approach to include dispersion contributions is physically appropriate for the system in question.