Geometry optimization refers to the determination of stationary points, principally minima and transition states, on molecular potential energy surfaces. It is an iterative process, requiring the repeated calculation of energies, gradients and (possibly) Hessians at each optimization cycle until convergence is attained. The optimization step involves modifying the current geometry, utilizing current and previous energy, gradient and Hessian information to produce a revised geometry which is closer to the target stationary point than its predecessor was. The art of geometry optimization lies in calculating the step $\mathbf{h}$, the displacement from the starting geometry on that cycle, so as to converge in as few cycles as possible.
There are four main factors that influence the rate of convergence. These are:
Initial starting geometry.
Algorithm used to determine the step $\mathbf{h}$.
Quality of the Hessian (second derivative) matrix.
Coordinate system chosen.
The first of these factors is obvious: the closer the initial geometry is to the final converged geometry the fewer optimization cycles it will take to reach it. The second factor is again obvious: if a poor step $\mathbf{h}$ is predicted, this will obviously slow down the rate of convergence. The third factor is related to the second: the best algorithms make use of second derivative (curvature) information in calculating $\mathbf{h}$, and the better this information is, the better will be the predicted step. The importance of the fourth factor (the coordinate system) has been generally appreciated later on: a good choice of coordinates can enhance the convergence rate by an order of magnitude (a factor of 10) or more, depending on the molecule being optimized.
Q-Chem includes a powerful suite of algorithms for geometry optimization written by Jon Baker and known collectively as Optimize. These algorithms have been developed and perfected over the past ten years and the code is robust and has been well tested. Optimize is a general geometry optimization package for locating both minima and transition states. It can optimize using Cartesian, Z-matrix coordinates or delocalized internal coordinates. The last of these are generated automatically from the Cartesian coordinates and are often found to be particularly effective. It also handles fixed constraints on distances, angles, torsions and out-of-plane bends, between any atoms in the molecule, whether or not the desired constraint is satisfied in the starting geometry. Finally it can freeze atomic positions, or any $x$, $y$, $z$ Cartesian atomic coordinates.
Optimize is designed to operate with minimal user input. All that is required is the initial guess geometry, either in Cartesian coordinates (e.g., from a suitable model builder such as IQmol) or as a Z-matrix, the type of stationary point being sought (minimum or transition state) and details of any imposed constraints. All decisions as to the optimization strategy (what algorithm to use, what coordinate system to choose, how to handle the constraints) are made by Optimize.
Note particularly, that although the starting geometry is input in a particular coordinate system (as a Z-matrix, for example) these coordinates are not necessarily used during the actual optimization. The best coordinates for the majority of geometry optimizations are delocalized internals, and these will be tried first. Only if delocalized internals fail for some reason, or if conditions prevent them being used (e.g., frozen atoms) will other coordinate systems be tried. If all else fails the default is to switch to Cartesian coordinates. Similar defaults hold for the optimization algorithm, maximum step size, convergence criteria, etc. You may of course override the default choices and force a particular optimization strategy, but it is not normally necessary to provide Optimize with anything other than the minimal information outlined above.
The heart of the Optimize package (for both minima and transition states) is
Baker’s eigenvector-following (EF) algorithm.
^{
51
}
J. Comput. Chem.
(1986),
7,
pp. 385.
Link
This was developed following the work of Cerjan and Miller,
^{
177
}
J. Chem. Phys.
(1981),
75,
pp. 2800.
Link
and of Simons and co-workers.
^{
1001
}
J. Phys. Chem.
(1983),
87,
pp. 2745.
Link
^{,}
^{
57
}
J. Phys. Chem.
(1985),
89,
pp. 52.
Link
The Hessian
mode-following option incorporated into this algorithm is capable of locating
transition states by walking uphill from the associated minima. By following
the lowest Hessian mode, the EF algorithm can locate transition states starting
from any reasonable input geometry and Hessian.
An additional option available for minimization is Pulay’s GDIIS
algorithm,
^{
231
}
J. Mol. Struct. (Theochem)
(1984),
114,
pp. 31.
Link
which is based on the well known DIIS technique for
accelerating SCF convergence.
^{
908
}
J. Comput. Chem.
(1982),
3,
pp. 556.
Link
GDIIS must be specifically
requested, as the EF algorithm is the default.
Although optimizations can be carried out in Cartesian or Z-matrix
coordinates, the best choice, as noted above, is usually delocalized internal
coordinates. These coordinates were developed by Baker
et al.,
^{
50
}
J. Chem. Phys.
(1996),
105,
pp. 192.
Link
and can be considered as a further extension of the natural
internal coordinates developed by Pulay et al.
^{
904
}
J. Am. Chem. Soc.
(1979),
101,
pp. 2550.
Link
^{,}
^{
318
}
J. Am. Chem. Soc.
(1992),
114,
pp. 8191.
Link
and the redundant optimization method of Pulay and Fogarasi.
^{
905
}
J. Chem. Phys.
(1992),
96,
pp. 2856.
Link
Optimize incorporates a very accurate and efficient Lagrange multiplier
algorithm for constrained optimization. This was originally developed for use
with Cartesian coordinates
^{
52
}
J. Comput. Chem.
(1992),
13,
pp. 240.
Link
^{,}
^{
48
}
J. Comput. Chem.
(1993),
14,
pp. 1339.
Link
and can handle
constraints that are not satisfied in the starting geometry. The
Lagrange multiplier approach has been modified for use with delocalized
internals;
^{
53
}
J. Comput. Chem.
(1997),
18,
pp. 1079.
Link
this is much more efficient and is now the
default. The Lagrange multiplier code can locate constrained transition states
as well as minima.