This section describes one of the most important features of any
quantum chemical software package; locating equilibrium
geometries and
transition structures of molecules. In
DALTON this is done using either
second-order trust-region
optimizations [7] or a
variety of first-order
methods [8]. These methods
have been implemented for SCF and MCSCF wave functions. For
non-variational wave functions -- like MP2 or CI -- the program can
only do first-order geometry optimizations using a
numerically
calculated gradient. This will be invoked automatically by the program
in case of a non-variational wave function. Some comments connected to
such geometry optimizations are collected in Sec. 
.
One of the strengths of the DALTON program package is the stable algorithms for locating first-order transition states . As described below, this is done by one of three algorithms: by trust-region second-order image surface minimization [9] , by gradient extremal walks [10] , or by following a specific mode [11] . These options will be discussed in more detail below.
Another feature of the DALTON program system is the options for calculating essential information about the potential energy surface so that subsequent molecular dynamics analysis can be made. There are two options for doing this in the program: One can either follow an Intrinsic Reaction Coordinate (IRC) [10] , or solve Newtons equations for the nuclei under the potential put up by the electronic cloud [12] . Whereas the first option can be used for modeling the immediate surroundings of a molecular reaction path, the second will give a precise description of one molecular trajectory. Chemical reactions may thus be monitored in time from an initial set of starting conditions. The program will automatically generate a complete description of the energy distribution into internal energies and relative translational energies. By calculating a large number of trajectories, a more complete description of the chemical reaction may be obtained [13].
Please note that for geometry optimizations using the Self-Consistent
Reaction Field model, certain restrictions apply, as discussed in
Sec. .