Using eScience to calibrate our tools: parameterisation of quantum mechanical calculations with grid technologies
Kat F Austen, T. O. H. White, Richard P. Bruin, Martin T. Dove, Emilio Artacho, R. P. Tyer · Department of Earth Sciences EPrints Repository · 2006
A report is presented on the use of eScience tools to parameterise a quantum mechanical model of an environmentally important organic molecule. eScience tools are shown to enable better model parameterisation by facilitating broad parameter sweeps that would otherwise, were more conventional methods used, be prohibitive in both time required to set up, submit and evaluate the calculations, and in the volume of data storage required. In this case, the broad parameter sweeps performed highlighted the existence of a computational artefact that was not expected affect this system to such an extent, and which is unlikely to have been observed had fewer data points been taken. The better parameterisation of the model leads to more accurate results and the better identification of the applicability of aspects of the model to the system, such that great confidence can be put in the results of the research, which is of environmental importance. 1. Introduction Polychlorinated biphenyls (PCBs) have long been known to have environmental significance due to their persistence within the environment, and their high toxicity. New information on the interaction of these chemicals with common soil minerals is continually being sought to facilitate meso-scale modelling of their movement through aquifers, through assessment of the retardation effects of different minerals and their adsorption isotherms. The structure of biphenyl is shown in Figure 1. PCBs share this structure, and each hydrogen atom on the carbon rings can be substituted for chlorine atoms, in any combination. The number of possible PCB congeners, 209 different arrangements of chlorine atoms around the biphenyl rings, poses an arduous and time-consuming problem for the typical computational scientist. Were all of these congeners to be studied by hand, the time required to generate the starting structures alone would be prohibitive, without even considering the best parameterisation of the model. This paper reports the work that has been carried out, using