How to split a large job into a set of smaller jobs. Utility sum

If your CASTEP job is estimated to be too large to run with all $\mathbf{k}$-points simultaneously, you can split it into parts. Final files band.out and psi2.[option] are then compiled together by a special utility sum (hopefully, it works although it has been long time ago!).

Let us consider this case in detail:

• First of all, you should split the k-point sampling into subsets. For example, a 10 k-points job can be split into a set of 10 jobs with 1 k-point each or into a set of 3 jobs with 3, 3 and 4 k-points each, etc.
• Run CASTEP for each k-point subset and then rename all the output files, namely band.out and psi2.[option], as follows: band.out.1, band.out.2, etc. and psi2.DOS.1, psi2.DOS.2, etc.
• Compile the sum routine using a shell-script sum.comp (provided) from your working directory and then run it as

   sum  [number of subsets]
  

  The number of subsets is optional and will be asked if not given. To compile the sum utility you can use any of the param.inc files used to compile any of your CASTEP codes since the number of k-points, NKPTS, does not matter; only the grid size and the number of bands do matter.

In the case that the option MAP is on, you should be a bit more careful. It is a good idea to refrain from running all CASTEP jobs one after another because this may lead to a collection of very large {psi2.MAP.N} files. To save the disk space, you can run the first two jobs, then sum them, delete files with the extensions .1 and .2 and move files psi2.[option] and band.out into psi2.[option].1 and band.out.1, respectively. Then, you can run CASTEP for the 3rd subset, rename the files produced by attaching the extension .2 and then run the sum again. This process must be repeated as many times as necessary for each sequential pair of subsets.

The sum-ming can be done automatically by constructing a shell-script. Suppose, we have 5 subsets of the k-points and the corresponding fort.15 files for them are fort.15.1, fort.15.2, ..., fort.15.5, while the corresponding CASTEP codes are castepx.1, castepx.2, ..., castepx.5 (note that some of them may coincide but should have different names anyway; use symbolic link in this case). Then, the following script can be used:

 #! /bin/csh -f
foreach num (1 2 3 4 5)
if( $num == '1' ) then
  set ext='1'
elseif
  set ext='2'
endif
 /bin/cp fort.15.$num fort.15
castepx.$num > problem.$num
if( -e band.out )  mv band.out band.out.$ext 
if ( -e psi2.DOS ) mv psi2.DOS psi2.DOS.$ext
if ( -e psi2.DOp ) mv psi2.DOp psi2.DOp.$ext
if ( -e psi2.SRF ) mv psi2.SRF psi2.SRF.$ext
if ( -e psi2.MAP ) mv psi2.MAP psi2.MAP.$ext
if( $ext == '2') then
  sum 2 > sum.$num
  mv band.out band.out.1
  if ( -e psi2.DOS ) mv psi2.DOS psi2.DOS.1
  if ( -e psi2.DOp ) mv psi2.DOp psi2.DOp.1
  if ( -e psi2.SRF ) mv psi2.SRF psi2.SRF.1
  if ( -e psi2.MAP ) mv psi2.MAP psi2.MAP.1
  if ( -e psi2.DOS.2 ) /bin/rm psi2.DOS.2
  if ( -e psi2.DOp.2 ) /bin/rm psi2.DOp.2
  if ( -e psi2.MAP.2 ) /bin/rm psi2.MAP.2
  if ( -e psi2.SRF.2 ) /bin/rm psi2.SRF.2
endif
end
#
mv band.out.1 band.out
if ( -e psi2.DOS.1 ) mv psi2.DOS.1 psi2.DOS
if ( -e psi2.DOp.1 ) mv psi2.DOp.1 psi2.DOp
if ( -e psi2.SRF.1 ) mv psi2.SRF.1 psi2.SRF
if ( -e psi2.MAP.1 ) mv psi2.MAP.1 psi2.MAP