(1g) Quantifying Uncertainty in Molecular Simulations

The collection of averages by molecular simulation is a stochastic process, and consequently it yields results that are not perfectly reproducible. To use these averages or to compare them to other quantities it is essential to understand the uncertainty associated with them, i.e., by how much might they differ if the calculation were repeated with different initial conditions? The means to quantify this uncertainty is well established and cuts across all fields of science, but the reporting of uncertainties with molecular simulation data is still not a universal practice. Moreover, there are pitfalls to avoid in applying error analysis to the results of a molecular simulation, as it is frequently necessary to manipulate and combine the direct simulation results to obtain the property of interest. Proper attention is needed to correlation both across averages and within subaverages to ensure that the simulation uncertainty is correctly propagated to the final result.

In this workshop we review these issues and provide several instances demonstrating the need for attention to uncertainty calculation. We guide the participants through example calculations demonstrating techniques for computing and propagating uncertainties.