Abstract

Linear-scaling formalisms of density functional theory (DFT) are becoming increasingly popular due to their ability to overcome the size limitations of standard cubic scaling implementations of DFT , thereby enabling simulations of tens of thousands of atoms. One approach, which is implemented in the wavelet-based BigDFT code, uses localised support functions, which are optimised to reflect their local chemical environment and thus constitute an accurate minimal basis set. Beyond opening up a path to linear scaling, this localised description also offers an opportunity for combining with different fragment based approaches. In BigDFT fragments are exploited both to further reduce the computational cost for systems containing repetition, and for defining a complexity reduction reduction framework for analysing the electronic structure of large systems, for example by enabling a graph-based description of interactions between fragments. In this talk I will describe how linear-scaling BigDFT and the related fragment approaches are used simulate large systems, giving examples of the corresponding new opportunities for performing and analysing first principles simulations of many thousand atom systems.