Simulation software advancing the discovery of new catalysts
6 November 2015
Dr Michail Stamatakis from the UCL Department of Chemical Engineering developed Zacros for simulating molecular phenomena on catalytic surfaces, but further work was needed for the code to run effectively on high performance computing platforms. A collaboration with the UCL Research Software Development Group has brought about significant improvements in the performance and usability of the code, which has become a popular tool with researchers around the world.
Catalysts play a vital role in diverse applications in science and in everyday life – one of the most well-known examples being the treatment of gas pollutants such as carbon monoxide (CO) from car exhausts by catalytic converters. Finding good and economical catalysts for reactions such as CO oxidation is not trivial, and UCL's Dr Michail Stamatakis is undertaking research on trying to understand and predict – before even doing experiments – which materials may make good catalysts.
The approach known as 'Kinetic Monte Carlo' (KMC) simulation, which has been applied for many years to computationally model physical-chemical interactions, can accelerate the search for new catalysts. The KMC methodology was adopted in computational studies of catalytic surface chemistries in the 1980s, but even 1990s research was constrained by the use of over-simplified chemical mechanisms. There remained a requirement for a more detailed and flexible description, of all the kinetics of the chemical reactions and the interaction of reacting substances (adsorbates) with the catalyst and with each other, while being bound to the catalytic surface.
The essential information about such mechanisms lies in the energies of gaseous and surface-bound chemical species, as well as the 'forces' of pushing or pulling between these surface species. Earlier studies were reliant upon restricted details of such interactions: for instance, forces only between nearest neighbours were accounted for, whereas long range interactions were neglected. The computational challenge was to study wider, longer and realistic reaction schemes – producing higher quality simulations of chemical mechanisms. Achieving better predictions of reactions in the molecular-scale processes required a new computational code, capable of making the most of modern computer architecture.
What we did
Zacros, a software package developed by Dr Michail Stamatakis from UCL’s Department of Chemical Engineering, enables researchers to perform high quality simulations of molecular phenomena on catalytic surfaces; however, as with any computer simulation, greater detail requires more computational resources. To develop Zacros into a more powerful research tool, Dr Stamatakis worked with UCL’s Research Software Development Group (RSDG) to allow the code to make best use of high performance computing platforms such as UCL’s Legion cluster.
The initial collaboration came about through a call from the RSDG for small projects that could benefit from free software development effort. As a result, Dr Jens Nielsen and Dr Mayeul d’Avezac, software developers with backgrounds in physics, began working with Dr Stamatakis to increase the applicability of kinetic modelling to complex reactive events, and improve the computational efficiency of the simulation code.
There were many problems to overcome – calculations involving diverse (long-range) interactions are computationally intensive – the algorithm has to deal with updating the rates of existing reaction events and adding new reactions in the queue after analysing and executing the original events. - Jens Nielsen, Research Software Developer
Supporting this increased complexity required the ability for the software to perform calculations in parallel across multiple CPU cores. Parallelising the code in this way significantly accelerates the speed at which the impact of individual reaction events upon the wider system can be calculated, making it practical to carry out such detailed simulations. This effort formed the bulk of the initial free project, and following that RSDG have continued to work with Dr Stamatakis on an ARCHER eCSE grant by the Edinburgh Parallel Computing Centre to make further improvements to the code. This has included enabling Zacros to run on ARCHER - the UK's national high performance computing resource, and further work is ongoing to allow Zacros to operate in parallel across multiple compute nodes using the MPI protocol.
About the improvements delivered by Zacros, Dr Stamatakis says:
We found in earlier studies, that in some instances chemical interactions were neglected altogether or modelled by nearest-neighbour contributions, this was due to the large computational expenses needed for implementing more accurate models. Zacros offers higher fidelity and accuracy, with wider applicability to chemistries that could not be studied before. At the same time the simulations run faster at lower cost. The advantages are significant.
Whilst making these improvements, the team have also worked to improve Zacros’s reliability, the readability of the code, and the user experience by refactoring the code, and adding additional functionality such as a command line interface. Whilst this does not give any boost to performance, these improvements are an important part of the RSDG’s ethos of good software development practice, which they aim to share with their collaborators.
Zacros has developed into a powerful research tool that is becoming increasingly popular with computational catalysis and surface science researchers. It has been realised as a licensed software product and the impact of Zacros beyond UCL can already be measured in numbers: 58 non-UCL Zacros licenses have been issued to users in 21 countries and one commercial licence has been sold. Zacros is also being developed and customised for other non-catalyst research fields such as thin-film growth and under EU2020 proposals, in an atomic layer-deposition research project.
The RSDG team have continued to collaborate with Dr Stamatakis and they are working together to investigate alternative algorithms for Zacros, which may further improve the accuracy and efficiency of the simulations.