Ab Initio MD

AIMD is computationally intensive, typically scaling as O(N³) with system size for standard DFT, though linear-scaling methods exist. Practical limits on current hardware:

• System size: ~100–1000 atoms for routine BOMD
• Timescale: Tens to hundreds of picoseconds
• Cost driver: The self-consistent field (SCF) convergence at each step This is orders of magnitude more expensive than classical MD, which can handle millions of atoms for microseconds. The trade-off is accuracy versus sampling. AIMD is therefore often used to parameterize or validate machine-learned interatomic potentials that reproduce DFT accuracy at classical cost.

The workhorse electronic structure method underpinning most Ab Initio MD simulations. DFT calculates the ground-state energy of a many-electron system by expressing it as a functional of the electron density rather than the many-body wavefunction, dramatically reducing computational cost. The Kohn-Sham equations map the interacting system onto a fictitious non-interacting system, with exchange-correlation functionals (e.g., PBE, B3LYP) approximating the complex quantum many-body effects. In AIMD, DFT provides the interatomic forces at each time step via the Hellmann-Feynman theorem.

A specific flavor of Ab Initio MD where the electronic wavefunction is fully converged to the ground state at every nuclear configuration. At each time step, a self-consistent field (SCF) optimization solves the static electronic structure problem for fixed nuclei, and the resulting forces drive the nuclear motion via Newton's equations. This adiabatic approximation decouples electronic and nuclear motion, assuming electrons instantaneously adapt. The primary computational bottleneck is the SCF convergence at every femtosecond-scale step.

An alternative Ab Initio MD approach that treats electronic degrees of freedom as fictitious dynamical variables with a small mass, propagating them alongside the nuclei using an extended Lagrangian. This avoids explicit SCF convergence at each step, keeping electrons close to the Born-Oppenheimer surface through an adiabatic decoupling. The fictitious electron kinetic energy must remain small and decoupled from nuclear motion. CPMD is efficient for large systems but requires careful tuning of the fictitious electron mass to prevent energy drift.

A machine-learned interatomic potential trained on high-level quantum mechanical data that provides ab initio accuracy at force-field cost. Unlike true AIMD, forces are predicted by a neural network rather than computed from electronic structure, enabling orders-of-magnitude speedup. Architectures like Deep Potential (DeePMD) and SchNet learn local atomic environment descriptors that respect rotational, translational, and permutational symmetries. These potentials bridge the gap between empirical force fields and full AIMD for large-scale, long-timescale simulations.

The dominant basis set choice for Ab Initio MD of periodic systems. The electronic wavefunctions are expanded in a discrete set of plane waves up to a kinetic energy cutoff, naturally respecting the periodicity of the simulation cell. Advantages include systematic convergence by increasing the cutoff, absence of Basis Set Superposition Error (BSSE), and efficient computation of the Hartree potential via Fast Fourier Transforms. The primary disadvantage is the large number of basis functions required for core electrons, mitigated by pseudopotentials.

An approximation that replaces the strong Coulomb potential of the atomic nucleus and tightly bound core electrons with a weaker effective potential acting only on valence electrons. This dramatically reduces the number of plane waves needed in Ab Initio MD calculations. Norm-conserving pseudopotentials preserve the scattering properties, while ultrasoft pseudopotentials relax the norm-conservation constraint for greater smoothness. The Projector Augmented Wave (PAW) method reconstructs the full all-electron wavefunction, combining pseudopotential efficiency with all-electron accuracy.