|> home > research > materials > castep|
CASTEP version 4.2 (academic release 2) is currently available on the IBM Regatta for specifically licensed users. The following information is specific to the use of CASTEP on the HPCx IBM Regatta (hpcx.ac.uk).
CASTEP (CAmbridge Serial Total Energy Package) is a package for performing ab initio quantum-mechanical atomistic simulations. It employs density functional theory, pseudopotentials and a plane-wave basis set. The latest version of CASTEP includes fully parallelised implementations of recent DFT developments, including the density mixing electronic minimiser and the use of ultrasoft pseudopotentials (both real and reciprocal space).
CASTEP was originally written by Mike Payne of Cambridge University, and since then has been developed by numerous research groups throughout the UK. The code is marketed commercially by Accelrys Inc, but is freely available to UK academics through the UK Car-Parrinello (UKCP) consortium. CCLRC's Daresbury Laboratory staff, now supporting the HPCx system, have also been involved in the parallelisation and optimisation of the current version of the code.
HPCx users wishing to obtain a licence to use CASTEP should contact the UKCP Consortium (address below).
CASTEP can be run on nprocs processors of the HPCx IBM Regatta using a command such as:
poe -procs <nprocs> <castepexe> <seed> <par_strat>
<nprocs> is the required number of processors (which should agree with the number specified in the Loadleveler script)
<castepexe> is the pathname of the CASTEP executable,
<seed> is the base name used to construct CASTEP input and output files and
<par_strat> (optional) is the preferred parallel strategy (i.e., G, K or KG). If <par_strat> is omitted, CASTEP will make a sensible choice of its own.
A copy of the latest CASTEP 4.2 executable is held centrally on the HPCx IBM Regatta, at /usr/local/bin/castepexe, and so should appear in the default path of authorised users. Access to this copy of the executable is restricted; HPCx users wishing to obtain a licence to use CASTEP and to be given access to the centrally-installed copy should contact the UKCP Consortium (address below).
The parallelisation of CASTEP is such that, in most cases, the code will run on any number of processors. However, CASTEP can be a very memory-intensive application, and for large system sizes it may be necessary to use a certain minimum number of processors to ensure that sufficient total memory is available.
In general, using more processors will lead to a reduction in the wall-clock time taken for a calculation, although at a higher total CPU-hour cost. However, for all calculations there comes a point at which using additional processors will not lead to any further speed-up. Generally, the larger the system, the greater the number of processors that can usefully be employed.
Further note that, when performing calculations with more than one k-point, the greatest efficiency of the parallel code is often obtained by ensuring that the number of processors is either a divisor or multiple of the number of k-points used (more generally, one should attempt to maximise the greatest common factor of the number of k-points and the number of processors).
To obtain a licence to use CASTEP and to access to the CASTEP executable held on the HPCx IBM Regatta, or for CASTEP user support, please contact:
CCLRC Daresbury Laboratory
Warrington WA4 4AD, UK.
UKCP web pages containing further information about the purpose of CASTEP, its functionality and the science that it implements
|http://www.hpcx.ac.uk/research/materials/castep.html||contact email - firstname.lastname@example.org||© UoE HPCX Ltd|