juDFT: Hands-on DFT codes from Jülich
Location: Forschungszentrum Jülich, Germany
For details on the organization of the workshop see: http://www.juDFT.de
Density-functional theory (DFT) provides the most efficient and practical framework to compute atomistic properties of solids, liquids, clusters and molecules from the basic laws of quantum mechanics and has spread to various applications in physics, chemistry, materials science, biology, mineralogy, engineering etc. Atomistic computations have achieved a high degree of sophistication. In many cases one can predict the observable properties ``ab-initio'' without any adjustable parameters. A diverse spectrum of versatile, complex and rather sophisticated electronic structure methods has been developed to compute an unmanageably large variety of properties in a multitude of rather different systems. The use and application of these electronic structure methods require a thorough training at a location where users meet developers of such methods.
There is an increasing request from users to provide a tutorial on the DFT codes from Jülich(www.judft.de). Well established is the FLEUR code, a publicly available FLAPW package, that was developed in the recent years along the direction of treating electronically complex bulk materials and materials in two and one dimension. It is able to deal with complex non-collinear magnetism, excitations on the basis of many-body perturbation theory and spin-dependent transport properties. Additional codes from Jülich include the massively parallellized finite-difference code juRS, suitable for structural relaxations and calculations of large unit-cells. The Jülich supercomputing center (JSC) provides unique capabilities for such applications.
The tutorial aims at introducing the various DFT codes developed in Jülich to interested PhD students and postdocs. Besides the basic theory, the underlying numerical algorithms and physical approximations will be discussed in a series of lectures by the developers of the codes. In detail the tutorial will focus on the following methods and codes:
i) In condensed matter physics and materials science the full-potential linearized augmented planewave (FLAPW) method has emerged as a widely used very robust and precise state-of-the-art ab initio electronic structure technique with reasonable computational efficiency to simulate the electronic properties of materials on the basis of density-functional theory. Due to the high precision it is widely accepted that it provides the DFT answer to the problem. FLAPW is an all-electron method which is universally applicable to all atoms of the periodic table in particular to transition metals and rare-earths and to multi-atomic systems with compact as well as open structures. Due to the all-electron nature of the method, magnetism is included rigorously and nuclear quantities e.g. isomer shift, hyperfine field, electric field gradient (EFG), and core level shift are calculated routinely. The FLEUR code implements this FLAPW method and its specifics will be discussed in detail in the tutorial.
ii) While standard DFT still is the method for the simulation of many systems and physical properties due to its relatively modest computational demands and its well-proven successes, it is cannot give an answer to all problems in electronic structure theory and more sophisticated additional schemes have been introduced to deal with materials with strong local correlation effects or to obtain bandgaps in semiconductors in better agreement with experiment. Hence the class of orbital dependent (hybrid) functionals as implemented on top of the FLEUR code will be discussed as well as the SPEX code which includes an implementation of the GW-approximation in the FLAPW method.
iii) The development of computer architectures towards extreme multicore systems impose a significant challenge on DFT implementations. Many of the established methods suffer from bad scaling properties and hence can not easily take advantage of current supercomputers and future massively parallel systems. Here the numerical solution of the Kohn-Sham equations on a regular real-space grid offers advantages by exploiting domain-decomposition algorithms which can be easily scaled up to several thousands of cores. The juRS code uses such a real-space finite-difference approach and its features and usage will also be introduced within the tutorial.
iv) In much less detail the tutorial will summarize additional methods and codes developed in Jülich to enable the students to gain some insight into the various Green function methods, in particular the KKR methods and the embedding approach. Here the focus will be on the introduction of the general methology, the calculation of transport and scattering properties and the possibilities the codes offer without trying to introduce these codes up to the level at which further independent calculations would
Three lecture of 1h each are planed on every morning introducing the various methods used in the codes as well as the basics of the physical effects described. The entire afternoon is devoted to practical sessions in which the codes are introduced and their capabilities can be explored under close guidance of the developers.
09:00 Density functional theory
10:15 Electronic structure methods
11:30 Full-potential linearized augmented plane wave method
14:00 Hands-on tutorial I
Installation of FLEUR and basic examples
10:15 Spin orbit coupling
11:30 Wannier functions
14:00 Hands-on tutorial II
Magnetic properties of the bulk and surfaces
09.00 Hybrid functionals and Optimized effective potentials
10:15 The GW-approximation
11:30 Constrained RPA
14:00 Hands-on tutorial III
Approaches beyond LDA/GGA
09:00 The PAW approximation
10:15 Real-space DFT
11:30 Massively parallel DFT calculations
14:00 Hands-on tutorial IV
The juRS code
09:00 The KKR Green function method
10:15 The Green function embedding technique
11:30 Electronic transport
14:00 Hands-on tutorial V
Green function methods
Gustav Bihlmayer (Forschungszentrum Jülich) - Organiser
Stefan Blügel (Forschungszentrum Jülich) - Organiser
Daniel Wortmann (Forschungszentrum Jülich) - Organiser