Nanoreactivity at drastically Extended Timescales

Modelling is a proven powerful tool in materials research, providing key information for the design of new materials and materials processes. In particular, molecular simulations provide a bridge between the microscopic properties of individual atoms and molecules and the macroscopic bulk properties of materials. However, we are not yet in a situation in which industries can effectively design new materials or materials processes based purely on molecular modelling. Present simulation techniques cannot reach the time and length scales that are required for complex chemical processes, or are based on inaccurate oversimplified models that make them unreliable. Traditional molecular simulations can, without special techniques, simulate events efficiently up to the pico- or nanosecond scale. However, many industrially relevant processes, such as diffusions and chemical reactions, involve physical and chemical events that happen at the time scale of milliseconds, seconds or even hours or days.
Current modelling platforms are either unable to address such long scales, or require expensive supercomputer facilities. This provides a business opportunity for methodological developments that would make possible running such simulations. The ReaxFF method is an ideal candidate to address that need. It is an approximate, fast methods, capable of dealing with relatively long time scales and large system sizes. It is a force field method that employs a series of empirical relations to describe the energy and forces of the materials, also describing bond-breaking reactions. ReaxFF is arguably the most transferable reactive empirical force field method (it has been applied to virtually all classes of materials and its current development covers most of the elements in the periodic table) and its balance of accuracy and speed makes it the computational method of choice for atomistic-scale dynamical simulations of chemical reactions. However, meeting the demands of materials modellers will require important developments. Namely, it will require extending ReaxFF to drastically longer time scales, in order to achieve large time and length scales with high-accuracy atomistic resolution, currently a major bottleneck for industrial modellers. SCM has addressed such demand by extending ReaxFF into a robust, accurate and predictive method for modelling reaction dynamics at industrially relevant scales, without the need for supercomputing resources. This required the implementation of acceleration techniques, coupling molecular dynamics and statistical mechanics models.

On the one hand, the molecular dynamics method at the heart of ReaxFF is limited to simulations of time and lengths around the nanoscale, whereas many relevant processes occur infrequently, beyond the microsecond time scale. On the other hand, statistical mechanics models, such as (kinetic) Monte Carlo molecular models, allow the calculation of macroscopic observables without necessarily following the dynamics of the systems, deriving them from an accurate sampling of all equivalent microscopic configurations of the system. Therefore, the time-scale problem of the MD approach can be solved by coupling it to statistical mechanics models, on which SCM has expertise. The first task in the project was thus to assess the available acceleration schemes, choosing the most suitable one. This proved to be the collective variable-driven hyperdynamics (CVHD) method, described by Prof. Dr. Erik Neyts and collaborators. Such method yields speedup factors of up to nine orders of magnitude, reaching a time scale of seconds while still accurately reproducing correct dynamics. Crucially, such scheme requires almost no user input, thanks to its self-learning nature, which makes it especially accessible for non-experts.
The CVHD scheme was subsequently implemented into ReaxFF, coupling statistical mechanics with molecular dynamics and incorporating feedback from test-users. The result was released commercially in September 2018, as part of SCM’s Amsterdam Modeling Suite.

The speed-up obtained thanks to this work effectively extends the time scale of the simulations from nanoseconds to the (micro)seconds range. That represent a dramatic acceleration, opening the door to the modelling of materials processes previously out of reach. Such a powerful tool should lower the barrier for manufacturing companies to use modelling as a means to becoming more competitive, enabling a rational design approach and minimizing expensive and time-consuming experiments. In particular, manufacturing SMEs without the resources for traditional supercomputing resources should benefit from such an accessible, cost-effective solution.

Periodic Reporting for period 1 - NET (Nanoreactivity at drastically Extended Timescales)

Partager cette page

Télécharger