Software for Chemistry & Materials
SCM is an Amsterdam-based computational chemistry software company, focused on atomistic materials simulation. Originally spinning out from the VrijeUniversiteit as Scientific Computing & Modelling N.V. in 1995, the SCM team supports and develops the Amsterdam Modeling Suite, centeredaround the flagship program Amsterdam Density Functional (ADF), which was originally developed in the 1970s in the theoretical chemistry department. Over a 140 authors have contributed to our software, we collaborate with academic development groups around the world, and participate in many EU and Dutch projects, such as S4CE. SCM’s computational chemistry tools are used worldwide by modellers in academia, government labs and industry studying various fields of chemistry and materials science.
Our main role within S4CE is to develop modelling tools to explore the molecular mechanisms for CO2 fixation in sub-surface formations (Task 5.2 – Development of models for CO2 carbonation). The long-term goal is to develop a theoretical understanding of the carbonation of CO2 in different substrates and geological situations. SCM’s portfolio of modules represents an excellent starting point towards that goal. In particular, the molecular dynamics ReaxFF method is well suited to the study of reaction dynamics in large complex systems, such as those relevant for S4CE.
Figure 1: Typical ReaxFF simulation, showing a fluid reacting with a solid surface.
ReaxFF is a program for modelling chemical reactions with atomistic potentials based on the reactive force field approach. This is a simplified method where the potential energy of a system of atomsis described in terms of a functional form and parameter sets. The parameters of the energy functions may be derived from experiments in physics or chemistry, calculations in quantum mechanics, or both. In our case, we use our ADF Density Functional Theory (DFT) code to provide high-quality reference data. Thanks to such simplified approach, ReaxFF can be used to model reactions in complex chemical mixtures totalling hundreds of thousands of atoms, on a modern desktop computer, and with a relatively high accuracy. ReaxFF has been used over the past decade in various studies of complicated reactive systems, including solvent environments, interfaces, and molecules on metal (oxide) surfaces.
However, despite its strengths, ReaxFF presents two limitations that need to be overcome in order to apply it successfully to S4CE systems: (1) it needs to be drastically sped up, to be able to describe slow processes, and (2) new sets of parameters need to be generated, suited to describe the chemical compositions and reactions involved in CO2 carbonation in different rocks.We have been working on both fronts, improving and extending ReaxFF so that it can be used to understand the molecular mechanisms for CO2 carbonation in various environments and in the presence of H2O and H2S.
We have accelerated ReaxFF implementing hybrid parallelism, i.e., splitting the calculations across parallel processes on distributed memory devices (using the MPI protocol), as well as across parallel threads on shared memory devices (OpenMP). And we have also implemented the so-called Collective Variable-Driven Hyperdynamics (CVHD) method, a statistical “enhanced sampling” approach that saves computational time replicating the system evolution across different pathways and focusing on the regions where reactions actually take place. That yields speed-up factors of up to 109, and the combination of these developments brings us from the nanosecond to the microsecondor even secondtimescale, much closer to the CO2 fixation timescales.
Regarding the generation of specific force-field parameters, we have been implementing algorithms toautomate the process, such as an on-the-fly reaction detection scheme in ReaxFF that detects new species and provides fully automatic reaction networks and rate constants. This facilitates the generation of previously unavailable force-field parameters, with minimal user input.
Figure 2: Automatic generation of complex reaction networks with ReaxFF.
After these developments, the next steps ahead consist in generating DFT reference data with ADF, to fit ReaxFF parameters for all possible reactive pathways encountered by CO2 as it carbonates in contact within basalt rocks. We will then be able to run ReaxFF simulations and derive predictions, at sufficiently long time scales to be experimentally relevant, of what is happening at the atomistic level in the carbonation process, as a function of the host material and of other substances present (such as salts and H2S).
Authors: Sergio Lopez