Lennard-Jones Potential

The potential is defined as V(r) = 4ε[(σ/r)^12 - (σ/r)^6], where r is the interatomic distance. The r^(-12) term models Pauli repulsion at short range due to overlapping electron clouds, while the r^(-6) term captures London dispersion attraction from induced dipole interactions. The parameters ε (well depth) and σ (collision diameter) are atom-type specific and determined empirically or from quantum calculations.

The two parameters have clear physical interpretations:

• ε (epsilon): The depth of the potential well, representing the maximum attractive energy between two atoms
• σ (sigma): The finite distance at which the inter-particle potential is zero, defining the effective atomic radius

These are typically derived by fitting to experimental data like second virial coefficients or ab initio quantum mechanical calculations. Common values for carbon in the OPLS force field are σ = 3.50 Å and ε = 0.066 kcal/mol.

The Lennard-Jones potential is computationally inexpensive because it depends only on pairwise distances and requires no expensive exponential or trigonometric function evaluations. The r^(-12) and r^(-6) terms are calculated using simple multiplications. In practice, a cutoff radius (typically 10-12 Å) is applied, beyond which interactions are truncated or shifted to zero, dramatically reducing the O(N^2) computational cost of non-bonded force calculations in large systems.

For heterogeneous atom pairs (e.g., carbon interacting with nitrogen), combining rules estimate cross-interaction parameters from pure-component values:

• Lorentz-Berthelot rule: σ_ij = (σ_i + σ_j)/2 and ε_ij = √(ε_i * ε_j)
• Geometric mean rule: σ_ij = √(σ_i * σ_j)

These approximations are critical for simulating multi-component systems like protein-ligand complexes but can introduce inaccuracies for highly polarizable atoms.

The standard 12-6 Lennard-Jones potential has known shortcomings:

• The r^(-12) repulsion is steeper than physically realistic exponential repulsion
• It cannot capture many-body dispersion effects or polarization
• It poorly describes highly polarizable atoms like halogens

Extensions include the Buckingham potential (exponential repulsion), the 9-6 Lennard-Jones variant for softer repulsion, and Drude oscillator models that explicitly account for electronic polarization in advanced force fields.

The Lennard-Jones potential is a universal component of Class I additive force fields including:

• AMBER: Used for proteins and nucleic acids with atom-type specific parameters
• CHARMM: Combined with electrostatic terms for biomolecular simulation
• OPLS: Optimized to reproduce liquid-state thermodynamic properties
• GROMOS: A united-atom approach where aliphatic hydrogens are implicit

It exclusively models van der Waals interactions, working alongside Coulombic terms for electrostatics and bonded terms for covalent geometry.

The process of deriving the specific ε (well depth) and σ (collision diameter) parameters for the Lennard-Jones potential. Parameters are optimized to reproduce experimental liquid densities, heats of vaporization, or high-level ab initio data. Common force fields like AMBER, CHARMM, and OPLS-AA provide extensive libraries of atom-type-specific LJ parameters, often using Lorentz-Berthelot combining rules to calculate cross-interactions between different atom types.

A computational necessity for making LJ potential calculations tractable in large systems. The attractive r⁻⁶ dispersion term decays rapidly, so interactions beyond a cutoff radius (typically 10–14 Å) are truncated. To avoid discontinuities, a switching function smoothly brings the energy and force to zero. Long-range dispersion corrections are applied analytically to recover the truncated tail contribution to energy and pressure.

The standard algorithm for handling long-range electrostatics in periodic systems, used alongside the LJ potential which models short-range van der Waals. PME splits the Coulombic sum into a short-range real-space term (calculated with a direct cutoff) and a long-range reciprocal-space term solved via Fast Fourier Transforms. The LJ potential is computed only in real space, typically with the same cutoff as the electrostatic real-space term.

The classic 12-6 Lennard-Jones form uses an r⁻¹² repulsive wall. However, some force fields like the Halgren 9-6 potential use a softer r⁻⁹ repulsive term, which can better reproduce high-level quantum mechanical dimer curves for certain atom pairs. The choice of exponent affects the steepness of the repulsive wall and the shape of the potential well, influencing predicted densities and solvation free energies.

A modern machine learning approach that learns the entire potential energy surface, including many-body effects, directly from density functional theory (DFT) data. Unlike the pairwise-additive LJ potential, NNPs like DeepMD and ANI capture polarization, charge transfer, and bond breaking. They provide ab initio accuracy at a fraction of the cost, effectively replacing the LJ functional form with a learned, non-linear function of the local atomic environment.

A purely repulsive variant of the Lennard-Jones potential, created by truncating and shifting the LJ function at its minimum (r = 2¹/⁶σ). The WCA potential models only the excluded volume of atoms, serving as a reference system in perturbation theory for liquids. It is widely used in soft matter physics and as a building block for coarse-grained models where attractive interactions are treated separately.