Local Excitement in CP2K

eCSE03-011

Key Personnel

PI/Co-I: Dr. Matthew Watkins - University of Lincoln, Dr. Iain Bethune - EPCC, University of Edinburgh

Technical: Dr. Sergey Chulkov - University of Lincoln

Relevant Documents

eCSE Technical Report: Local Excitement in CP2K

Project summary

A number of research and application areas - from dye-sensitized solar cells, breakdown and long-term reliability in microelectronics transistor stacks, radiation damage in insulating materials and biological materials such as DNA, through to photo-catalysis - require an accurate description of excited electronic states and how they absorb light. We have implemented a method, called Time Dependent Density Function Theory, within the open source atomistic simulation code CP2K. This method allows us to calculate optical properties - by which we principally mean what colours of light are absorbed by a sample, and how strongly. We can now perform excited state calculations for large systems (>1000 atoms) with good accuracy (the non-local functionals we support show a significantly reduced Mean Square Error in the energy of light absorbed compared to conventional approximations). The use of non-local density functionals is increasingly recognized as a requirement for qualitatively correct prediction of material properties, driven by the requirement to correctly compute electronic band-gaps, correctly describe the wavefunctions of defects, as well as the significantly improved thermochemistry. For excited states modelled using TDDFT non-local exchange includes a qualitatively correct long-range interaction between the electron and hole generated by the light excitation. Applying hybrid functionals, especially in excited states, to condensed phase materials has largely been limited to rather small and idealised systems. By coupling the Auxiliary Density Matrix Method approximation to the non-local exchange to a linear scaling with system size density functional perturbation theory, we enable routine calculations on larger systems with a more accurate level of theory than was generally practical previously.

The work in this eCSE project enables much more realistic models of systems, revealing much about mechanisms of photoreactions, material breakdown and charge transfer of interest many workers in fields from solid state physics, to materials and biochemistry. The reduced scaling (linear in system size) and smaller prefactor than previous implementations will allow excited state calculation to be performed more routinely, for larger systems, in a time resolved manner along molecular dynamics trajectories and will be suitable for high-throughput applications.

Achievement of objectives

The project has achieved several objectives:

  1. Implementation of linear response TDDFT with hybrid (non-local) density functionals, and novel algorithms to achieve a 100x speedup over a basic implementation, and oscillator strengths.

    Fully achieved: This is the main objective of the project. A full reimplementation of the TDDFT code in CP2K has been carried out, and is available in the trunk version of the code for all users. It allows the calculation of excited state energies using local and non-local density functional and the intensity of the transitions (oscillator strength). The speed up achieved depends strongly on the system, but will exceed 100x in large systems with dense basis sets.

  2. Real time propagation TDDFT projections onto ground state orbitals to allow detailed analysis of excitations, demonstration of agreement of orbital decomposition with TDDFPT code in objective 1.

    Fully achieved: Code to output the time dependent overlap coefficients has been added to CP2K. We have also added new routines for exporting electronic orbitals, provide much quicker and easier visualisation of results and easier to storage (single compressed file, and compression can be tuned for visualisation quality rather than calculation).

  3. Delta-SCF calculations via the Maximum Overlap Method (MOM)

    Fully achieved: An implementation of the MOM method has been added to the code.

  4. Implement routines to perform momentum resolved band structure calculations via Wannier interpolation - demonstrate 10x speed up relative to full k-point calculations

    Achieved: Code to output a minimal form of the Kohn-Sham and overlap matrices has been added, allowing band structure to be unfolded onto small primitive cells via post-processing. Timing targets are hard to assess clearly in this case and something of a moving target as the k-point code has evolved.

  5. Extend the Berry-phase routines to allow Wannier localization of "entangled states" allowing the approximate localization of (semi)-metallic one-electron states.

    Partly Achieved: We became aware of other developments along these lines, so rather than duplicating work, we provided assistance in the implementation of a Wannier90 interface code. This will provide the required functionality.

Summary of the software

All code is part of the CP2K software package and made available under the GPL v2. It is freely available to anyone, including the latest development code. CP2K is supported on ARCHER and an up-to-date module is ready for deployment.

The code is available from the CP2K website, SVN repositories, and a github mirror.

The majority of our code is in the 4.1 release (5 October 2016) of the software, while the final versions are in the trunk (5.0) and will be part of the 5.1 release (6-12 month cycle). A pre-release version of 5.0 is already installed as a non-default module on ARCHER.