kim-convergence Utility Module

How Do You Automatically Estimate the Length of the Simulation Required?

These visualizations demonstrate why manual estimation of simulation length is difficult. The animations show trajectories from an NPT molecular dynamics simulation tracking three key properties: temperature (T), pressure (P), and box volume (V).

Top row: 10 ps simulation | Bottom row: 50 ps simulation

Key observations:

Different properties converge at different rates
Visual inspection cannot determine statistical reliability
Manual estimation is therefore unreliable

It is desirable to simulate the minimum amount of time necessary to reach an acceptable amount of uncertainty in the quantity of interest. The kim-convergence package addresses this by providing tools for automatic equilibration detection and run length control in simulations.

Key Features

Equilibration Detection: Identifies when your simulation reaches steady-state using methods like Marginal Standard Error Rule (MSER).
Run Length Control: Extends simulations adaptively until user-specified accuracy is achieved, based on confidence intervals and statistical inefficiency.
Upper Confidence Limits (UCL): Methods like Heidelberger-Welch, N-SKART, and MSER-m for estimating uncertainty.
Time Series Analysis: Tools for statistical inefficiency, autocorrelation, and sampling uncorrelated data.
Integration-Friendly: Works with LAMMPS, OpenMM, and custom simulators via a callback API.