8.4 Choose an NVT thermostat without confusing sampling and dynamics
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.
8.4.1 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.
Original schematic. Curves explain concepts; blank data areas await verified learner results. No calculation is claimed.
8.4.2 Worked case: procedure, interpretation and checks
Intuition and prerequisites
A thermostat exchanges energy with an imagined heat bath. The purpose of canonical sampling is a correct equilibrium distribution, not a perfectly flat temperature trace. A strongly damped system may have a satisfactory structural distribution while its diffusion or velocity correlations are altered. Decide first whether you want equilibrium structure or real-time dynamics.
Start from the preparation and NVE lessons. Use a fixed volume appropriate to the physical state; a thermostat cannot turn an incorrect density into the intended pressure. Know the species order and whether any atoms are frozen. Temperature derives from kinetic energy and the active degrees of freedom; constraints and removed collective motion matter. Avoid manually rescaling velocities every saved frame to make a prettier graph.
For an accessible starting point use Langevin NVT. Friction introduces a timescale roughly γ⁻¹: γ=5 ps⁻¹ corresponds to 0.2 ps. This is dimensional interpretation, not a guarantee of thermalization time. Assign a coefficient to every species in POSCAR order. Zero means that species is not directly thermostatted. LANGEVIN_GAMMA, Langevin thermostat
Practical comparison
- Take independent equilibrated states and run candidate coupling strengths at the same temperature, volume and Hamiltonian. One illustrative sensitivity set is 1, 5 and 10 ps⁻¹. Do not present it as a recommended physical range for every system.
- Compare mean temperature with uncertainty, energy distribution, relevant RDF or order parameter, and species-resolved kinetic temperatures where meaningful.
- Plot autocorrelation of the observable you intend to average. Fast temperature equilibration does not guarantee that a slow torsion, defect jump or phase change is sampled.
- If reporting diffusion, vibrational spectra or residence times, assess coupling dependence. Consider NVE production from an equilibrated state where appropriate, with energy validation. Retain enough independent starts to estimate variability.
- Keep equilibration and production settings clearly separated. If the thermostat changes, document the transition and discard a justified re-equilibration interval.
Nosé–Hoover uses deterministic extended dynamics and can be useful, but a single thermostat is not automatically ergodic for small or stiff systems. An alternative branch may use MDALGO=2 and an explicitly tested SMASS, removing Langevin tags. SMASS=0 is a special VASP choice, not a zero-coupling limit; consult its definition before tuning. SMASS
Do not mix settings for different algorithms in one “universal” INCAR. More recent thermostat options have version constraints. In particular, the official MDALGO page identifies CSVR (MDALGO=5) as available from VASP 6.4.3. Confirm support in your executable before adopting a copied input. The reference manual, executable version and active input values should travel with your results. MDALGO
What counts as success?
A successful structural-sampling test shows compatible structural averages over justified analysis windows, plausible temperature fluctuations and no hidden numerical failure. Equal mean temperatures alone are insufficient. Plot raw data and block means; label the burn-in interval. For a two-temperature physical problem, do not accidentally create separate species baths and interpret the result as ordinary equilibrium.
Common traps: choosing huge friction to hide energy drift; regarding Nosé thermostat energies as physical lattice energy; copying a species-friction list after reordering POSCAR; treating identical TEBEG/TEEND as proof of equilibrium; using an obsolete tag from old lecture notes without checking current semantics.
Exercise. Design a comparison where the RDF is the primary output and the velocity autocorrelation is secondary. Predict which will be more directly disturbed by strong damping. Run only after writing the accepted precision, independent-start strategy and cost limit. Explain any disagreement without automatically declaring one thermostat “wrong.”
8.4.3 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.
8.4.3.1 Input block 1
# UNEXECUTED NVT delta for a TWO-species example
IBRION = 0
MDALGO = 3
ISIF = 2
TEBEG = 350
TEEND = 350
LANGEVIN_GAMMA = 5.0 5.0
POTIM = 0.5
NSW = 2000
ISYM = 0
8.4.4 Related learning paths
- 1.1 Four input files, one physical question
- 1.2 A convergence laboratory with an error budget
- 8.3 Validate NVE timestep and energy conservation
- 8.5 NpT dynamics, pressure units and a stable simulation cell