An important first choice is which method to use for optimisation of the ionic positions. The default option is the
Broyden-Fletcher-Goldfarb-Shannon (BFGS) method and this is used throughout the rest of this user guide. For fixed cell
calculations there are two other options, damped molecular dynamics
geom_method : dampedmd
and delocalised internal coordinates
geom_method : delocalized
The user is free to learn about these methods from the literature and try them out if they suit a particular problem.
Since we will be doing variable cell calculations later on we will stick with BFGS.
The verbosity of information in the output files is controlled by
iprint : value
value can be 0,1,2 or 3 in order of increasing verbosity with a default value of 1, which in most cases will be sufficient to
understand how the calculation has progressed. iprint : 3 is for full debugging. For users who wish to know a
little bit more about what occurs during a calculation it might be interesting to set
iprint : 2.
If a calculation is unexpectedly interrupted many hours of work might be lost. CASTEP automatically backs up the calculation
after every 5 BFGS steps. The parameter
num_backup_iter : value
allows the backup interval to be user specified. For very slow jobs it is sensible to back up the results after every BFGS step
num_backup_iter : 1
When a calculation that was stopped is resumed the parameter
continuation : default
must be included so that the calculation will read the .check file and use the last completed BFGS step as a starting point
for the calculation. This feature is essential when the calculation is run on large shared computational facilities where
the run time per job is restricted.
It is possible to choose whether the algorithm should favour a speedier calculation at the cost of using more
RAM during the calculation or whether to to slow down the calculation and make it less RAM intensive.
opt_strategy : value
value can be 'MEMORY' or 'SPEED'. The default is for a balance between memory usage and
performance. If the calculation is at all likely to be RAM limited then MEMORY should be chosen. On RAM
extensive architectures SPEED should be chosen. SPEED should also be used for most clusters, multiple
processors or supercomputers because these write data temporarily to disk so as to conserve RAM which significantly
hampers performance.