Evaluation of a phase diagram with a novel technique in an HPC-Europa3 funded collaboration
Posted: 19 Jul 2019 | 09:22
György Hantal was an HPC-Europa3 visitor from 18th March to 17th May 2019.
I am György Hantal (pictured here with Durham castle in the background), a postdoctoral researcher from the University of Vienna. I was an HPC-Europa3 visitor for two months, hosted by Dr. Lívia Bartók-Pártay at the University of Reading's Department of Chemistry and supported by EPCC in Edinburgh.
The collaboration with Dr Bartók-Pártay was motivated by the fascinating natural phenomenon of biomineralisation: many living organisms use inorganic compounds to strengthen their bodies by either mineralising tissues (eg bone) or secreting minerals around themselves (eg the shell of a snail). One of the really amazing aspects of this phenomenon is the precision by which living organisms can control the morphology as well as the polymorphism of the crystallites which they incorporate into very complex structures. The mechanism of this intriguing phenomenon is still mostly unknown.
We focused on CaCO3 (calcium carbonate), the most abundant mineralizing compound in invertebrate organisms, which has received continued attention over the last few decades. In order to study this phenomenon computationally, one needs a reliable computer model that correctly reproduces the relative stability of the various polymorphs in a wide range of representative conditions. Surprisingly enough, CaCO3 turns out to be a notoriously difficult system in this regard. Indeed, most computer models fail to reproduce the subtle stability difference between aragonite and calcite in ambient conditions where the latter is more stable, though aragonite has the lower enthalpy.
Figure 1: the structure of the three anhydrous polymorphs of CaCO3: a) calcite, b) aragonite, and c) vaterite.
In the framework of this collaboration, our goal was to compute the phase diagram of the presumably best CaCO3 computer model (Raiteri, P., et al. J. Phys. Chem. C 119, 2015, 24447) with a novel technique called nested sampling (Pártay, L.B. et al. J. Phys. Chem. B 114, 2010, 10502). I used pymatnest, an open-source, highly parallel computer code, to perform nested sampling calculations in different conditions. Nested sampling is a Bayesian sampling technique that can be used to efficiently sample very complex multidimensional functions (such as the potential energy surface defined in the configurational space) without any prior knowledge of the object to be sampled. The method operates with independent samplers, called ‘walkers’ that explore the function to be sampled concurrently but without any need for information from each other. This is why the method is relatively easy to parallelise and can, due to the typically large number of walkers, considerably benefit from massive HPC facilities. The great strength of the technique is that it makes it possible to estimate directly the partition function from which a lot of important thermodynamic quantities can be derived.
My HPC-Europa fellowship gave a unique opportunity to come to the UK and work with one of main developers of the nested sampling technique as well as to use ARCHER and in part also Cirrus, two of the largest modern HPC facilities in the UK. I am especially thankful for the adept technical and administrative support provided by Mario Antonioletti and Catherine Inglis who made my stay carefree, allowing me to dedicate most of my attention to my project. Without this fellowship, I also wouldn’t have had the chance to discover the historical and natural beauty of the UK on my weekends (as it is illustrated by the photo at the top of this article).
HPC-Europa3 website: www.hpc-europa.eu/
Next Call closes on 19 September 2019.
György Hantal, University of Vienna