Development of methods for strongly correlated electron systems
Title: Development of methods for strongly correlated electron systems
DNr: SNIC 2016/1-6
Project Type: SNIC Medium Compute
Principal Investigator: Oscar Grånäs <oscar.granas@physics.uu.se>
Affiliation: Uppsala universitet
Duration: 2016-01-22 – 2017-02-01
Classification: 10304 10302
Homepage: http://www.physics.uu.se/en/mattheo/research/magnetism
Keywords:

Abstract

Magnetism often occurs in compounds with strongly interacting electrons. Traditionally computational materials science employs so called density functional theory (DFT). DFT works with an effective mean field treatment for describing the electron-electron interaction though an exchange correlation functional dependent only on the density and its gradients. Currently available functionals are inadequate for treating compounds with strong electron correlation. For this we have developed an implementation of the so called DFT+Dynamic Mean Field Theory (DMFT), where we have a explicit treatment of the subset of electrons which exhibits strong correlation. The DFT+DMFT methodology has proven to be very efficient to describe the electronic structure of compounds with strong electron correlation, including magnetism and optical properties. The DMFT implementation is currently running efficiently in the Full-Potential Linearized Muffin-Tin Orbital DFT code RSPt. Some of the algoritms in the code are suitable for porting to GPU's, while more communication intensive parts are more suitable for execution on the CPU. In this project we propose to port parts of the RSPt code to GPU execution, while maintaing the CPU (MPI) parallelism. If successful this will open up the possibility to tackle materials scientific problems at a large scale at a smaller of the cost. The implementation will be used to investigate ad-atoms on transition metal, as well as transition-metal oxide surfaces. Here strong electron correlation in addition to the spin-orbit coupling phenomena arising on the surface do to electric field gradients gives rise to a number of interesting phenomena. On a non-magnetic metallic surfaces the magnetic moment on a ad-atom looses it's intrinsic moment in the so-called Kondo effect. Upon oxygenation of the metal surface it undergoes a distinct change of the electronic structure, resulting in a semiconducting/insulating state. In this case the ad-atom keeps it's magnetic moment to much higher temperature. Hence we can study the full evolution of the many-body electron-electron interaction under the influence of spin-orbit coupling. Another class of material which exhibits both complex magnetic ordering and strong electron correlation is the Strontium Ruthenates in the Perovskite-like Ruddlesden-Popper phases. SrRuO3 is an itinerant ferromagnet, Sr2RuO4 is a ferromagnetic superconductor and Sr3Ru2O7 is metamagnetic. We indend to investigate how the crystal structure of these materials influence the electronic structure, including local dynamical electronic excitations using DMFT, while mapping long range interaction on a phenomenological model to treat spin-wave excitations. Running DFT+DMFT on these classes of materials are extremely demanding, hence the porting of the code to hybrid GPU/MPI parallelism holds great promises. All implementations will be shared with other users of the code, hence the outcome will be of greater benefit for the Swedish electronic structure community.