The main advantage of the BFGS minimizer (Pfrommer et al., 1997) is the ability to perform cell optimization, including optimization at fixed external stress. The BFGS scheme uses a starting Hessian which is recursively updated during optimization. The CASTEP implementation involves a Hessian in the mixed space of internal and cell degrees of freedom, so that both lattice parameters and atomic coordinates can be optimized.
When cell optimization is required, use of a finite basis set correction term is recommended. Calculating the term is not too costly (10-30% of the self-consistent electronic minimization at the first iteration) relative to the advantages that it provides. Cell optimization runs may be problematic if the finite basis set correction is not used or if the energy cutoff is so low that the correction is not accurate. In these circumstances, the optimization run may stop with a message which states that the energy has converged, but that the stress is still non-zero. The minimizer attempts to find the energy minimum rather than the stress zero-point, since the former is more meaningful under the circumstances.
Another potential pitfall is the use of the Fixed Basis Size Cell optimization setting when the starting geometry is very different from the final one. The finite basis set correction depends on the cell size and shape, although this dependence is disregarded by the minimizer. In addition, the effective cutoff energy changes when the cell geometry is modified with the above setting (it is the number of plane waves that is kept fixed). If this change takes the dEtot/d(ln Ecut) function far away from the point that was used to evaluate the finite basis set correction, the results will not be accurate. Therefore, you should compare the starting and final geometries and perform a completely new run starting from the final configuration if the difference between the two structures is large.
CASTEP can perform geometry optimization with constraints applied. The simplest type of constraint is to fix the atom positions. In CASTEP, this means fixing the fractional coordinates of the atoms.
Atoms on special positions need not be fixed manually as the symmetry treatment in CASTEP ensures the correct behavior of such atoms.
It is also possible to fix lattice parameters. Once again, it is only necessary to do this for parameters that are not fixed by symmetry. This type of constraint is often useful in phase stability studies when, for example, you want to fix the angles to 90° (Winkler and Milman, 1997).
It is also possible to impose more general linear constraints on atomic coordinates. These constraints are specified using a matrix that transforms the Cartesian coordinates of all atoms to the subspace of unconstrained coordinates. This facility is intended for fixing, for example, the z-coordinates of atoms in slab calculations of surface processes.
Non-linear constraints refer to constraints on interatomic distances (bonds), angles, and torsions. Such constraints can be imposed using the delocalized internals optimizer.
You can facilitate cell optimization convergence by making a reasonable initial estimate of the bulk modulus for the material. The initial changes to lattice vectors are calculated on the basis of this value (and, of course, based on the calculated stress tensor). When the estimated bulk modulus is too high, the steps become smaller than is ideal for the minimizer; therefore, it is conceivably possible that you could reduce the number of geometry optimization steps by reducing this value. On the other hand, too small a bulk modulus for a hard material could result in an oscillating behavior, where the first step is too big. Thus, prior knowledge of the hardness of the material could help to reduce the number of steps required for calculations. The effect of a judicious choice of the initial bulk modulus value is illustrated in the table below (times are given relative to that of the calculation using an initial value of 500 GPa).
|Compound||Estimated bulk modulus (GPa)||Number of steps||Relative time|
Damped molecular dynamics presents an alternative method for geometry minimization that involves only internal coordinates (the cell parameters have to be fixed). The method uses the critical damping regime as a way of dealing with the ground state. The regime can be implemented either by using one damping coefficient for all degrees of freedom (coupled modes) or by using different coefficients for different degrees of freedom (independent modes). The latter approach allows you to freeze all modes, both fast and slow, simultaneously. Alternatively, you can perform steepest descents damping with a fixed coefficient. However, this is a less efficient approach. Indeed, it is not, strictly speaking, a molecular dynamics technique, since it solves first-order equations of motion and not second-order equations.
Both independent modes and coupled modes damped molecular dynamics runs can be performed with a bigger time step than undamped molecular dynamics simulations. When the system gets close to equilibrium, the time step can be increased even further, since the fast modes freeze out before the slow ones. CASTEP can automatically adjust the time step, leading to increased efficiency of the algorithm. It is also recommended that you recalculate the damping coefficients periodically during a damped molecular dynamics run. A full description of the method implemented in CASTEP can be found elsewhere (Probert, 2003).
Setting up a geometry optimization