Learn

Force field parameterization

A classical force field is a sum of simple terms for bonds, angles, torsions, Lennard-Jones interactions, and charges. Each term needs numbers. Some come from quantum chemistry, some from experiments, and many are copied from similar molecules. This page shows how the fitting works for a few of them.

    A bond from a QM scan

    You stretch the bond in small steps in a quantum chemistry code, record the energy, and fit a harmonic term near the minimum. I used ORCA for the Si-C bond in alkylsilanes. Note that CHARMM writes the term without the factor of 1/2.

    V(r) = Kb (r − r0)2

    The real curve is not harmonic, so the result depends on how wide the fitting window is. Points close to the minimum give the right curvature, which is what matters for vibrations at room temperature.

    Try: widen the window, and watch Kb drop as the anharmonic part creeps in.

    Fitting a torsion

    Torsions are fitted to a dihedral scan. You rotate around the bond, relax everything else, and take the QM energy minus what the rest of the force field already gives. The remainder is fitted with a short cosine series.

    E(φ) = Σn Kn [1 + cos(nφ − δn)]

    Writing cos(nφ − δ) as a cos(nφ) + b sin(nφ) makes this a linear least squares problem.

    Try: one term cannot fit the profile, three terms can, and more terms only fit the noise.

    Combining rules

    Force fields give LJ parameters per atom type, not per pair. For unlike pairs, a combining rule fills in the gap. CHARMM and AMBER use Lorentz-Berthelot (the arithmetic mean for σ and the geometric mean for ε), while OPLS uses geometric means for both.

    The choice matters for mixtures and interfaces, and this is why mixing two force fields needs care.

    Try: make the two atoms very different in size, and compare the two rules.

    Where the rest comes from

    Charges are usually fitted to the electrostatic potential around the molecule from QM (RESP in AMBER), or adjusted to interaction energies with water (CHARMM). LJ parameters are tuned so that the model reproduces the liquid density and the heat of vaporization. Validation then checks properties that were not fitted, such as diffusion, surface tension, or hydration free energy. Coarse-grained parameters follow the same logic one level up, as shown on the coarse-graining and DPD pages.