The HECToR Service is now closed and has been superceded by ARCHER.

Hybrid Time-Dependent Density Functional Theory in CASTEP Part 1

This Distributed Computational Science and Engineering (dCSE) project was to implement Time-dependent density functional theory (TDDFT) in the density functional theory code CASTEP. TDDFT has become a well-established technique for modelling excited state properties in molecular systems, and has been implemented in several quantum-chemistry codes, e.g. CPMD. This project also forms the first part of the work to implement excited-state force calculations for geometry optimisations and molecular dynamics in CASTEP, for the second part of the work please see here.

An implementation of TDDFT in CASTEP will now give the UK electronic structure community an opportunity to address cutting-edge scientific problems in areas such as inorganic and organic photovoltaic materials, catalytic reactions at surfaces, light-emitting polymer materials for optical displays, and femtosecond laser chemistry. Including the use of hybrid functionals promises to address some of the limitations that have previously hindered the application of TDDFT to extended systems.

The individual aims of the project were to :

  • Implement a straightforward scheme to compute the electronic response to an external electric field of a set frequency. This is based on the existing DFPT module code in CASTEP and provides a reference calculation against which the following calculations may be benchmarked.
  • Produce a releasable implementation of TDDFT using Hutter’s published method and compare the performance of two 'off-the-shelf' eigensolvers, namely ARPACK and EA19.
  • Port the implementation to HECToR and follow more extensive testing, debugging and benchmarking against previous calculations. On the 4 way SMP nodes of HECToR Phase 2a 80% parallel efficiency was achievable on up to 256 processing cores. On the 24 core nodes of Phase 2b (with each node consisting of 2x12 core sockets and each socket having 2xhex core dies) comparable calculation times could only be achieved by using 4 cores per node i.e. 1 core per hex core die.

Please see PDF or HTML for a report which summarises this project.