Various points about CASTEP & relaxation. MJR 10/96
Efficiency
----------
CASTEP uses a combined ion-electron conjugate gradients method, due to
Mike Payne et al. This is not necessarily optimal. For complex problems,
the use of BFGS (implimented on CETEP) is _much_ faster. See thesis,
R. Shah.
Unit cell relaxation
--------------------
Not done by academic code. Done by MSI's commercial code. Need to relax
at constant cut-off with Francis-Payne correction to energy and stress.
Symmetry
--------
If symmetry is imposed it cannot be broken. If symmetry is not imposed,
it might be broken due to numeric noise, but only after many iterations.
Therefore, do not start with a configuration of higher symmetry than the
expected relaxed configuration.
Number of k-points
------------------
Relaxation writes information to fort.40 and higher. K-point files start
at fort.20. Hence no more than 20 k-points can be used.
Enabling
--------
Set RMOVE in fort.15 to 1.0 for those ions which are to move, 0.0 for
those which are fixed.
Set IION to 1 in fort.14
Beware: the total number of electronic steps will be NITER*NITFIX,
and with IPRINT=0 there will be no output except on steps
which involve ionic relaxation. Even with IPRINT=1 the total
energy is printed out only on ionic relaxation, when it is
printed _thrice_. Even if IION=0, this is still true, except
that the energy is printed once per NITFIX iterations.