Deep learning collective variables from transition path ensemble.

The study of the rare transitions that take place between long lived metastable states is a major challenge in molecular dynamics simulations. Many of the methods suggested to address this problem rely on the identification of the slow modes of the system, which are referred to as collective variables. Recently, machine learning methods have been used to learn the collective variables as functions of a large number of physical descriptors. Among many such methods, Deep Targeted Discriminant Analysis has proven to be useful. This collective variable is built from data harvested from short unbiased simulations in the metastable basins. Here, we enrich the set of data on which the Deep Targeted Discriminant Analysis collective variable is built by adding data from the transition path ensemble. These are collected from a number of reactive trajectories obtained using the On-the-fly Probability Enhanced Sampling flooding method. The collective variables thus trained lead to more accurate sampling and faster convergence. The performance of these new collective variables is tested on a number of representative examples.

A unified framework for machine learning collective variables for enhanced sampling simulations: mlcolvar.

Identifying a reduced set of collective variables is critical for understanding atomistic simulations and accelerating them through enhanced sampling techniques. Recently, several methods have been proposed to learn these variables directly from atomistic data. Depending on the type of data available, the learning process can be framed as dimensionality reduction, classification of metastable states, or identification of slow modes. Here, we present mlcolvar, a Python library that simplifies the construction of these variables and their use in the context of enhanced sampling through a contributed interface to the PLUMED software. The library is organized modularly to facilitate the extension and cross-contamination of these methodologies. In this spirit, we developed a general multi-task learning framework in which multiple objective functions and data from different simulations can be combined to improve the collective variables. The library's versatility is demonstrated through simple examples that are prototypical of realistic scenarios.

Computing the committor with the committor to study the transition state ensemble.

The study of the kinetic bottlenecks that hinder the rare transitions between long-lived metastable states is a major challenge in atomistic simulations. Here we propose a method to explore the transition state ensemble, which is the distribution of configurations that the system passes through as it translocates from one metastable basin to another. We base our method on the committor function and the variational principle that it obeys. We find its minimum through a self-consistent procedure that starts from information limited to the initial and final states. Right from the start, our procedure allows the sampling of very many transition state configurations. With the help of the variational principle, we perform a detailed analysis of the transition state ensemble, ranking quantitatively the degrees of freedom mostly involved in the transition and enabling a systematic approach for the interpretation of simulation results and the construction of efficient physics-informed collective variables.

Structure and polymerization of liquid sulfur across the λ-transition.

The anomalous λ-transition of liquid sulfur, which is supposed to be related to the transformation of eight-membered sulfur rings into long polymeric chains, has attracted considerable attention. However, a detailed description of the underlying dynamical polymerization process is still missing. Here, we study the structures and the mechanism of the polymerization processes of liquid sulfur across the λ-transition as well as its reverse process of formation of the rings. We do so by performing ab initio-quality molecular dynamics simulations thanks to a combination of machine learning potentials and state-of-the-art enhanced sampling techniques. With our approach, we obtain structural results that are in good agreement with the experiments and we report precious dynamical insights into the mechanisms involved in the process.

From Enhanced Sampling to Reaction Profiles.

The determination of efficient collective variables is crucial to the success of many enhanced sampling methods. As inspired by previous discrimination approaches, we first collect a set of data from the different metastable basins. The data are then projected with the help of a neural network into a low-dimensional manifold in which data from different basins are well-discriminated. This is here guaranteed by imposing that the projected data follows a preassigned distribution. The collective variables thus obtained lead to an efficient sampling and often allow reducing the number of collective variables in a multibasin scenario. We first check the validity of the method in two-state systems. We then move to multistep chemical processes. In the latter case, at variance with previous approaches, one single collective variable suffices, leading not only to computational efficiency but also to a very clear representation of the reaction free-energy profile.