Software and methods under the microscope – Sisu and Taito involved

An international team of collaborators, including researchers from Aalto University and Åbo Akademi University, have shed new light on the use of Density Functional Theory in the study of solids. Over 60 experts from about 30 universities and research institutes studied how results from quantum simulations of material properties varied between researchers and the software they used. The latest code and methods were more successful than their predecessors, and their results were consistent.

– A large team of experts compared the software they've been using and saw that their degree of accuracy was unprecedented, says senior lecturer Torbjörn Björkman from Åbo Akademi University, previously a post-doc in Aalto University's Centre of Excellence in Computational Nanoscience (COMP).

Reproducibility is one of the cornerstones of research. Computer-generated results can vary depending on the implementation of the theoretical model. According to a release issued by the team, this is particularly true of material property predictions.

Their collaborative project focused on Density Functional Theory (DFT), which is widely used in both the academic world and industry. There are, however, a number of different calculation programs available, and it can sometimes be difficult to determine which one would be best suited to a particular type of calculation.

The study was performed in collaboration with coding experts and developers. A total of 40 methods were examined.

– These days, electron structure calculations play a critical role in many fields of material research. They are central to the interpretation of many studies, and enable even more accurate predictions. That's why it's vital for the different methods to produce the same results to a high degree of accuracy. Discovering that the latest methods do indeed do this has therefore been a highly significant outcome, says Björkman.

These results can now be set as quality standards against which future software can be evaluated. The study's participants hope that this will lead to better code and methods being used in simulations.

"First-class services"

CSC's supercomputer Sisu and supercluster Taito were used to test a variety of solutions for constructing a basis set for a particular electron structure calculation method used to study solids. And there's more to come. 

– Methods can be further developed based on these results, and CSC's computing resources will naturally be important in our future work. My own areas of interest involve developing methods that handle interaction between electrons in solids, in particular some very difficult cases, such as systems that are known to be highly correlated or characterised by weak van der Waals interaction, says Björkman.

– As a computational physics researcher in Finland, CSC's large-scale resources are of critical importance to me. Doing new and creative things often means that you need to push the technical boundaries of your field, and during my time in Finland, CSC has provided first-class services with a minimum of bureaucracy for the individual researcher. This helps tremendously.

The results of the study were published in the journal 'Science' on 25 March 2016.