It is widely accepted that bonded parameters for molecular mechanics force fields can be fit to quantum-mechanically derived potential energy surfaces (PES). Performing these fits for small molecules is a straightforward task because fits can be made directly to PES scans for each unique conformational degree of freedom. This is not the case, however, in complex molecules. For example, highly branched structures are plagued with interdependent degrees of freedom; torsional displacement can lead to significant changes in nearby angles and bond lengths. Additionally, these molecules tend to have a large number of unique bond length, bending angle, and dihedral types. For these reasons, the best approach for bonded parameter fitting it is not immediately obvious.
In this work, an iterative configurational-bias simulated annealing (ICBSA) approach is proposed wherein (i) a large number of molecular conformations is generated through loose configurational-bias Monte Carlo (CBMC) sampling of the molecules of interest, (ii) the bonded parameters are optimized via simulating annealing to the energy differences between conformations obtained from the full quantum-mechanically derived energies corrected by the non-bonded molecular mechanics energies, and (iii) the new set of bonded parameters is used for a subsequent round of CBMC sampling and simulating annealing. This approach is then used to re-fit bonded parameters for the Transferable Potentials for Phase Equilibra (TraPPE) force field for linear and branched alkanes, allowing for extension to heavily branched alkanes and introduction of flexible bonds.