How Do Thermostats Change Sampling and Dynamics?

Created on July 14, 2026 by Sungsoo Ahn , last updated on July 14, 2026

2026   ·   molecular-dynamics   thermostats   langevin   sampling   kups   ·   science

Note: This is an early draft page for the executable kUPS MD tutorial series. It is intentionally hidden from site navigation while the simulations, notebooks, figures, and review artifacts mature. This post builds on the initialization and integrator diagnostics by asking what changes once a thermostat is added. Corrections and replication issues should be tracked in sungsoo-ahn/kups-md-tutorials.

Introduction

A thermostat is often described as “keeping the temperature fixed.” That phrase is too vague for molecular simulation. A thermostat changes the equations being integrated, the distribution being sampled, and usually the dynamical memory of the trajectory.

For ML researchers, the practical question is not only whether the kinetic temperature is close to the target. It is whether the sampled configurational and kinetic moments match the intended ensemble, and whether the coupling is so strong that time-correlation functions are no longer interpretable as physical dynamics.

This draft demonstrates the executable slice of the fourth tutorial with a controlled harmonic oscillator and BAOAB Langevin dynamics. The oscillator is a small microscope for the sampling/dynamics distinction; the final article still needs an argon/kUPS thermostat diagnostic before publication.

What Does Coupling Strength Change?

The current diagnostic fixes the oscillator, timestep, target temperature, and BAOAB splitting, then changes the Langevin friction:

Choice Full value Why it matters
system harmonic oscillator canonical targets are known
thermostat BAOAB Langevin explicit stochastic splitting
target temperature 1.0 dimensionless (kT)
timestep 0.02 shared across coupling strengths
friction values 0.1, 1.0, 5.0 weak, moderate, and strong coupling
samples 3500 per run enough for compact moment checks

The check is deliberately two-sided. If the moments are wrong, the thermostat is not sampling the intended canonical distribution. If the moments are right but autocorrelation changes sharply, the thermostat may still be unsuitable for dynamical observables.

What Should The Diagnostic Show?

The full run compares observed position and velocity variances to their canonical targets. It also compares mean kinetic energy to the (0.5kT) target for one degree of freedom. Finally, it reports the position integrated autocorrelation time to show that stronger coupling can change dynamical memory.

Thermostat diagnostics for the committed full profile. The BAOAB Langevin cases sample near the same canonical moment targets, but strong coupling substantially increases position autocorrelation time.

Reproduction

The current executable path is:

git clone https://github.com/sungsoo-ahn/kups-md-tutorials
cd kups-md-tutorials
uv sync
uv run kups-tutorial run 04 --profile smoke
uv run kups-tutorial verify 04 --profile smoke
uv run kups-tutorial run 04 --profile full
uv run kups-tutorial verify 04 --profile full
uv run jupyter execute notebooks/post-04-thermostats.ipynb --inplace

The notebook is deliberately not the implementation source. It imports the configuration loader, thermostat diagnostics, and figure generator from src/kups_md_tutorials/.

Current Status

This page is not the final article. The implemented pieces are:

  • smoke and full controlled BAOAB Langevin workflows
  • committed compact summaries and downsampled samples
  • executable notebook
  • generated SVG/PNG figure and snapshot review
  • self-review note covering code, science, notebook, and figure feedback

The missing pieces are:

  • final article prose
  • citations for Langevin dynamics, BAOAB splitting, canonical sampling, and thermostat-induced dynamical distortion
  • rendered desktop and mobile page snapshots
  • argon/kUPS thermostat diagnostics that connect this controlled testbed to the production trajectory workflow

The rule for this series is simple: a result is not ready because the code ran. It is ready only after the code, data, figure, prose, and rendered page have all been reviewed against the same reproducibility contract.

References

  • Leimkuhler, B. & Matthews, C. (2013). Rational construction of stochastic numerical methods for molecular sampling. Applied Mathematics Research eXpress, 2013(1), 34-56.
  • Bussi, G., Donadio, D. & Parrinello, M. (2007). Canonical sampling through velocity rescaling. Journal of Chemical Physics, 126, 014101.
  • Tuckerman, M. E. (2010). Statistical Mechanics: Theory and Molecular Simulation. Oxford University Press.