intermediate 8 min read
Physical sciences · Topic
Molecular Mechanics
differential equations · optimization · numerical methods · linear algebra
Molecular mechanics (MM) replaces quantum mechanical electrons with empirical potential energy functions parameterised to reproduce experimental or high-level quantum data. By avoiding explicit treatment of electrons, MM enables simulations of systems containing millions of atoms over nanosecond to microsecond timescales. Force fields such as AMBER, CHARMM, and OPLS are the backbone of computational biology and drug discovery.

The Force Field Energy Function

The total potential energy is a sum of bonded and non-bonded contributions:

\[E_{total} = \underbrace{\sum_{bonds}k_b(r-r_0)^2 + \sum_{angles}k_\theta(\theta-\theta_0)^2 + \sum_{torsions}V_n[1+\cos(n\phi-\gamma)]}_{bonded} + \underbrace{\sum_{i<j}\left[\frac{A_{ij}}{r_{ij}^{12}}-\frac{B_{ij}}{r_{ij}^6}+\frac{q_iq_j}{4\pi\varepsilon_0 r_{ij}}\right]}_{non\text{-}bonded}\]

The Lennard-Jones $r^{-12}$ repulsion models Pauli exclusion while the $r^{-6}$ term captures London dispersion. Partial charges $q_i$ generate the electrostatic interactions.

Parameterisation

Force field parameters are derived from a combination of:

  • High-level QM calculations (bond lengths, torsion profiles, ESP charges)
  • Experimental thermodynamic data (heats of vaporisation, densities)
  • Spectroscopic observables (IR frequencies)
Force Field Primary Application Charge Scheme
AMBER Proteins, nucleic acids RESP
CHARMM Membrane proteins, lipids CGenFF
OPLS-AA Organic liquids, drug-like CM5
GROMOS Carbohydrates Partial equalisation

Strengths and Limitations

MM geometry optimisation minimises $E_{total}$ with gradient-based methods such as conjugate gradient or L-BFGS, converging in $O(10^3)$–$O(10^5)$ steps. The computational cost scales as $O(N\log N)$ with particle mesh Ewald for long-range electrostatics.

Key limitations include the inability to model bond breaking or formation, fixed charge models that ignore electronic polarisation, and transferability issues outside the parameterisation set. QM/MM hybrid methods address the first two by treating a reactive centre with DFT or semiempirical quantum mechanics while the environment is described by MM.