Skip to content

8.2 Prepare an AIMD calculation and decide when equilibration ends

These are unexecuted teaching inputs and starting settings to test. Original figures are schematics, not computed results. Use licensed VASP and PAW data, replace every placeholder, record the executable version and validate convergence.

8.2.1 Model, units and provenance

Use eV for energy, Å for length and eV/Å for force; 1 kbar = 0.1 GPa. State normalization per atom, molecule, primitive cell or simulation cell. Record PAW identifiers, release, ZVAL, ENMAX and permitted hashes; never redistribute POTCAR. SCF convergence addresses the chosen electronic problem; convergence of the target property requires separate tests.

 Prepare an AIMD calculation and decide when equilibration ends — original conceptual schematic

Original schematic. Curves explain concepts; blank data areas await verified learner results. No calculation is claimed.

8.2.2 Worked case: procedure, interpretation and checks

Level: beginner after static DFT

What problem are we solving?

A relaxation asks, “where is a nearby minimum?” AIMD asks, “what configurations does the system visit while moving?” A relaxed structure with no velocities is not yet a thermal ensemble. Conversely, a movie with lively atoms is not proof of equilibrium. Your first deliverable is an audit trail connecting composition, density, electronic settings, initial velocities, warm-up, and the frames retained for analysis.

Use an uncomplicated periodic solid as a first case, then repeat the workflow for an already packed liquid. Do not create a liquid by randomly placing atoms at arbitrary separations. A bad close contact can convert large repulsion into kinetic energy before a thermostat responds. For a molecular liquid, inspect both intermolecular distances and intramolecular bonds; for a solid, inspect defects, duplicate boundary atoms and supercell replication.

Prerequisites. You must already be able to run a converged static calculation and interpret forces. Prepare POSCAR, KPOINTS and licensed POTCAR; record the VASP version/build. Choose the physical question first: local structure, thermal expansion and diffusion require different cell sizes and trajectory lengths. A Γ-only supercell is a hypothesis to test on representative snapshots, not a right earned by having many atoms. The official silicon exercise provides a useful compact learning context, but its short teaching trajectory and inexpensive settings are not a production convergence standard. Official melting-silicon tutorial

Practical sequence

  1. Define the model. Record formula, atom count, cell vectors, mass density, charge, magnetic state and constraints. Density is total mass divided by cell volume. For a slab, volume including vacuum is not a liquid density.
  2. Inspect and repair the starting geometry. View the periodic image; check shortest pair distances by element and plausible coordination. Relax severe overlaps at the same intended Hamiltonian. Do not overinterpret a 0 K optimized liquid configuration as a finite-temperature liquid.
  3. Converge forces, not only one total energy. Select a small set of distorted or warm snapshots. Compare forces under tighter ENCUT, EDIFF and denser k sampling. Record maximum and RMS force differences in eV/Å. Pressure-sensitive work also needs stress convergence.
  4. Run a pilot. Use a short, conservative timestep; save every step at first. Examine electronic convergence, closest contacts and temperature before committing expensive allocation time.
  5. Warm and equilibrate. If a ramp is needed, keep it separate from a constant-target segment. Do not pool a temperature ramp into an equilibrium average.
  6. Select a production start based on observables. Plot potential energy per atom, temperature, pressure, and at least one structural observable. Compare consecutive block means and independent replicas. A stable temperature with continuing structural change is not equilibrated.
  7. Archive the decision. Write the discarded interval, evidence used, selected ensemble and retained time window into an analysis manifest. Save a clean restart state and proceed to timestep validation.

This requests 200 fs = 0.2 ps; it is a pilot, not a claim that equilibrium or diffusion can be established in that time. For two species, supply two friction coefficients in POSCAR species order. NSW × POTIM gives nominal simulated duration; completed steps may be fewer if the job stops early. TEBEG is not a command that every instantaneous temperature must equal 300 K. Initialization and restart velocities affect what actually starts moving. POTIM, TEBEG

Read the outputs and make a decision

Build one row per completed ionic step: step, physical time, T, potential/free-energy quantity used, kinetic energy, SCF iterations, and maximum force. Keep the labels of the original energy fields. F, E0, and the MD E field have distinct meanings; thermostat contributions may enter the MD total. Do not rename every energy column “total energy.” OSZICAR definitions

Temperature fluctuations are expected for a finite system. Large fluctuations are not automatically a failure, but a step with an SCF failure, an enormous force spike or a broken molecule that should be chemically stable deserves inspection. Structural relaxation can be much slower than thermostat relaxation. Use a property relevant to the question, for example a coordination histogram for a melt or an orientational order parameter for a molecular solid.

Sanity checks: frame count matches completed steps and output cadence; energy and time units are explicit; no unexplained restart discontinuity; no repeated electronic failures; averaging excludes ramp and transient; at least two starting conditions lead to compatible retained distributions. If not, extend or redesign the sampling rather than lowering your acceptance standard.

Novice traps: calling every moving atom a liquid; deleting inconvenient unstable frames without diagnosing them; treating the first 10% as a universal equilibration rule; assuming the thermostat repairs wrong density; comparing two trajectories with different functionals; choosing production length only from available wall time.

Exercise. Make three plots from your own pilot: raw T(t), energy per atom, and a structural coordinate. Propose a production boundary, then test whether moving that boundary changes the retained average beyond its uncertainty. Deliver the figures, a paragraph defending the boundary, and a list of tests still missing. No numerical target is supplied because the answer depends on your model.

8.2.3 Unexecuted inputs and analysis scaffolds

These are unexecuted teaching inputs and starting settings to test. Original figures are schematics, not computed results. Use licensed VASP and PAW data, replace every placeholder, record the executable version and validate convergence.

8.2.3.1 Input block 1

# UNEXECUTED MD delta: fixed-cell pilot, one chemical species
IBRION = 0
NSW = 400
POTIM = 0.5
TEBEG = 300
TEEND = 300
MDALGO = 3
LANGEVIN_GAMMA = 5.0
ISIF = 2
ISYM = 0
NBLOCK = 1
# Retain separately validated functional, ENCUT, k mesh, spin and smearing.

8.2.5 Technical sources