CCP9 - Computational Studies of the Electronic Structure of Solids

Motivation.

The SIC-LSD method is a very powerful tool to investigate the ground state configuration and the finite temperature phase diagram of strongly correlated materials. A complete study of the system, however, can require a very large number of separate calculations, including:

• scans over different configurations of localized and delocalized states,
• scans over the lattice constant,
• scans over different magnetic structures,
• scans over the temperature.

All results must then be analysed in order to extract the relevant information.

This sort of problem is an ideal example for the use of Grid technologies:

• (Semi-) automatic generation of input data
• Metascheduling across suitable grid resources
• Matching requirements to available resources to minimise queuing time
• Simulations can be run across multiple heterogeneous compute resources, using the abstraction provided by the grid middleware.
• Both the input and output files are automatically archived
  • within the data grid, creating a complete audit trail for the calculation.
• Key data values from the calculation are harvested from the
  • output files and stored within a metadata database, which significantly facilitates the processing of data from a large number of related calculations.

Use of the e-Minerals Infrastructure

The following infrastructure was used:

• Actual calculation: Fortran90 code (SIC-LSD)
• Generation of XML output: FoX-Library (F95)
• Automatic metadata insertion: RCommands
• Storage of data files: SCommands (SRB client tools)
• Data extraction: AgentX
• Job submission: Remote My_Condor_Submit (RMCS)
• Data Analysis: RGem

Example: Antiferromagnetism in CaCuO2

CaCuO2 is a prototype of the high temperature superconductors. In its pure form, it is an antiferromagnetic insulator. Upon doping with Sr, which introduces holes in the CuO2 layers, it undergoes a Mott Metal-Insulator Transition to a non-magnetic metallic phase, which occurs at a hole concentration of about 0.15.

This cannot be explained with the local spin-density (LSD) approximation, which would predict a nonmagnetic metal also for the undoped system.

The self-interaction corrected LSD (SIC-LSD) allows for:

• the correct description of the antiferromagnetic phase through localization of all Cu d-electrons (d9 configuration).
• the description of the transition to the metallic state through the delocalization of one d-electron (d8 configuration).

Computationally, the transition can be investigated by studying the energy difference between the d8 and the d9 configurations. Here this energy difference is plotted as a function of the hole doping (through a rigid band description to mimic Sr doping) for different lattice constants.

The plot shows the energy difference between the d9 and the d8 configurations: positive values indicate the metallic phase, while negative values correspond to the antiferromagnetic, insulating phase. It can be seen that the critical hole concentration depends on the lattice constant, and can be lowered by compressing the system.

After this successful feasibility study, the Grid technology will be applied to a series of high-profile applications:

• ground state configuration of U compounds:
  • o spin-orbit coupling reduces the symmetry of the system: more configurations
• finite temperature phase diagram of Pu:
  • o at T=0K: a wealth of configurations which are close in energy

    o at T>0K: include thermal fluctuations between these configurations

• study of the stripe phases of LaCu2O4:

  • o striped phases require a large supercell: large number of d8/d9 configurations

• metal insulator transition in LaCu2O4:

  • o explore phase space spanned by hole concentration and lattice constant o use CPA to model the hole doping and a mixed valence of Cu

XML markup of the output files

The XML output from the simulation codes simplifies the automatic extraction of key information at the post-processing stage. This is then stored within a metadata database, providing a highly efficient method of summarising the results from a large number of individual calculations, e.g. during a study to explore a phase space.

In addition, XML output can be transformed to XHTML, using XSLT tools such as ccViz. This rendered output facilitates information delivery as, for example, it can include embedded SVG graphs showing the variation of key properties during the self-consistency iterations.

Data interoperability between different LMTO codes has been demonstrated:

• TB-LMTO 47 (Stuttgart code)
• SIC-LMTO (Daresbury code)
• relativistic SIC-LMTO (Daresbury/Kent code)
• TB-LMTO (LDA+U) (Ekaterinburg code)

Progress to Date

Leon Petit (Universiy of Aarhus) presented his work at Daresbury on 7/6/2009.

We investigate the potential of high throughput computing for predicting new materials from a systematic study of the groundstate properties of whole compound families. The self-interaction corrected (SIC) local spin-density approximation (LSD) is used to predict the groundstate valency configuration of both actinide and lanthanide materials. For a given compound the groundstate is determined by minimizing the SIC-LSD total energy functional, in a process that requires some 40 self-consistent runs per system. With respect to the lanthanide materials, the entire manifold of mono-pnictides and mono-chalcogenides (140 compounds) was investigated, and the corresponding calculations have been run using the NW-GRID (UK?s North-West facility). The resulting electronic phase diagram is characterized by valency transitions brought about by a complex interplay of ligand chemistry on one hand, and lanthanide contraction on the other hand. The predicted groundstates, from the trivalent ?early? pnictides to the divalent ?late? chalcogenides are in good agreement with experiment. The electronic structure of the investigated actinide carbides, nitrides, and oxides (actual/potential nuclear fuels), is characterized by a broader span in oxidation states, with a 6+ valency predicted for UC. A gradual increase in the localization of the 5f electrons (and correspondingly lower oxidation states) is observed in going from U to Cm. Overall the calculated groundstates indicate an increasing degree of ionicity from the carbides to the oxides.