Monitoring Convergence of Molecular Simulations in the Presence of Kinetic Trapping

Convergence to equilibrium is an essential requirement for molecular simulation output to be accurate and reproducible. Yet testing for convergence is challenging, especially in instances where kinetic trapping and the consequent plateauing of ensemble quantities can falsely suggest premature equilibrium. Here we introduce a quantitative method for monitoring convergence based on multiple independent simulations started from diverse initial configurations. Our approach addresses common pitfalls of convergence testing, and allows specification of convergence criteria at a hierarchy of resolutions. We demonstrate the utility of this approach in monitoring simulations of several peptide systems of varying sizes, and use it to quantify the efficiency improvements realized by advanced simulation methods. ∗To whom correspondence should be addressed †Department of Bioinformatics and Biostatistics, Duke University, Durham, NC 27708, United States ‡Departments of Statistical Science and Computer Science, Program in Computational Biology and Bioinformatics, Program in Structural Biology and Biophysics, Duke University, Durham, NC 27708-0251, United States