Reverse Monte Carlo

Reverse Monte Carlo (RMC) ^[1] is a variation of the standard Metropolis Monte Carlo method. It is used to produce a 3 dimensional atomic model that fits a set of measurements (neutron diffraction, X-ray-diffraction, EXAFS etc.). In addition to measured data a number of constraints based on prior knowledge of the system (for example, chemical bonding etc.) can be applied. Some examples are:

Closest approach between atoms (hard sphere potential)
Coordination numbers.
Angles in triplets of atoms.

The 3 dimensional structure that is produced by reverse Monte Carlo is not unique; it is a model consistent with the data and constraints provided.

The algorithm for reverse Monte Carlo can be written as follows:

Start with a configuration of atoms with periodic boundary conditions. This can be a random or a crystalline configuration from a different simulation or model.
Calculate the total radial distribution function $g_{o}^{C}(r)$ for this old configuration (C=Calculated, o=Old).
Transform to the total structure factor:
$S_{o}^{C}(Q)-1={\frac {4\pi \rho }{Q}}\int \limits _{0}^{\infty }r(g_{o}^{C}(r)-1)\sin(Qr)\,dr$

where Q is the momentum transfer and $\rho$ the number density.
Calculate the difference between the measured structure factor $S^{E}(Q)$ (E=Experimental) and the one calculated from the configuration $S_{o}^{C}(Q)$ :
$\chi _{o}^{2}=\sum _{i}(S_{o}^{C}(Q_{i})-S^{E}(Q_{i}))^{2}/\sigma (Q_{i})^{2}$

this sum is taken over all experimental points $\sigma$ is the experimental error.
Select and move one atom at random and calculate the new (n=New) distribution function, structure factor and:
$\chi _{n}^{2}=\sum _{i}(S_{n}^{C}(Q_{i})-S^{E}(Q_{i}))^{2}/\sigma (Q_{i})^{2}$
If $\chi _{n}^{2}<\chi _{o}^{2}$ accept the move and let the new configuration become the old. If $\chi _{n}^{2}\geq \chi _{o}^{2}$ then the move is accepted with probability $\exp(-(\chi _{n}^{2}-\chi _{0}^{2})/2)$ otherwise it is rejected.
repeat from step 5.