constraints & restraints¶
The implementation in xopt has the following limitations: Constraints are only possible for Cartesian coordinates, while restraints can be defined for internal (primitive) coordinates, e.g. bond lengths and angles. Of course, a very restrictive restraint on an internal coordinates leads to a practical constraint. Thus, using high values for the restraining potential, also internal coordinates can be ‘constrained’.
The restraining potential is in the form of a simple, quadratic function with the restraining energy $R$ being defined as
with \(k_r\) as the force constant of the restraint, $p$ the type of internal (primitive) coordinate (bond, angle or torsion) and its reference value \(p_{ref}\).
The actual value of \(k_r\) needs to be chooses carefully and is structure and level of theory specific. The reference value \(x_{ref}\) is taken from the initial input geometry (or from the xopt.restart file if requested).
xopt checks for the file “xopt.control” at the startup. Its content will overwrite any commandline options!
- legend::
- i,j,k,l = <integer>, atomic numbers starting from 1 F = <floating point> eg. “1.0”
Restrained optimizations:
$restr
atom i F
bond i j F
angle i j k F
dihed i j k l F
vol sphere [vdw,<float>]
An additional gradient from the restraining potential (\(G_r\)) will be added to the total gradient: \(G_total\): \(G + G_r\). gnorm and all other values are calculated using \(G_total\) ! Thus is it better to rely on energy convergence for restrained optimizations.
One can add a target value for dihedral restraints with the keyword target:
dihed i j k l F target T
, where \(T\) states the target value in degree.
Cartesian constrains:
$freeze
Hopt #just optimize hydrogen positions, freeze the rest
atom i #freeze atom number i
elem <string> #freeze all elements <string> (lower case !)
list <list> #freeze a list of integers, eg. 'list 1-3,8,9,33-34'
region [sphere,box] [atom,space] [<int>,<float>,<float>,<float>] # freeze a region centered at an atom or a point in Cartesian space.
For constrains the gradient components (and respective entries in the B-matrix) are set to zero.