Reference
All work totally or partially based on DLFIND should cite [16] as well as the original references of the optimization algorithms used (references given in [16]).
Description
A lightweighted description of geometry optimization in general and DLFIND in particular can be found in Frontiers 2007.
Functionality
Coordinate systems:
 Cartesian coordinates (including frozen atoms and components), massweighted Cartesian coordinates
 Internal coordinates (including all constraints):
 DLC (delocalized internal coordinates, i.e. redundant internal coordinates)
 DLCTC (total connection)
 HDLC (hybrid delocalized internal coordinates, see Phys. Chem. Chem. Phys. 2, 2177 (2000))
 HDLCTC
Combinations of coordinates (images):
 Dimer method [11]
 NEB (nudged elastic band) [12]
All of the combinations work with all versions of coordinate systems.
Optimizers:
 steepest descent
 conjugate gradient
 LBFGS
 PRFO, Hessian update mechanisms: Powell and Bofill. Hessian either by input or by finitedifference. In the latter case either in Cartesians (then the update also in Cartesians, and one can output frequencies), or in internals.
 Damped dynamics
 Algorithms for Conical intersection search:
 Penalty function
 Gradient projection method
 LagrangeNewton method
 Stochastic search methods (including a genetic algorithm) for global and local minimization. These methods optimize by calculating may energies in parallel and are thus wellsuited for massively parallel computation.
Line search algorithms:
 Simple scaling of the proposed step (covering the maximum step length)
 Trust radius based on energy decrease
 Trust radius based on the projection of the gradient on the step
The design allows new methods to be easily implemented.
Reaction rate calculations with or without tunneling contributions
 Instanton theory (aka imaginaryF theory or harmonic quantum transition state theory) to calculate tunneling rates.
 Instanton optimizations with a quadraticallyconverging optimizer [20]
 Instanton rate calculations (parallelized)
 Adaptive step size in instanton calculations [22]
 Reaction rates without tunneling
Other functionality:
 The optimizer is fully restartable.
 DLFIND can be included in ChemShell and GAMESSUK and was used in conjuction with a nuber of other codes
Authors
 Johannes Kästner, main author
 Tom W. Keal contributed conical intersection search algorithms, parallelization of NEB and finitedifference Hessian calculations and fixed many bugs.
 Joanne M. Carr is adding parallel search algorithms
 Judith B. Rommel contributed to the implementation of instanton theory
 Salomon Billeter and Alexander Turner: parts of their HDLCopt routines have been used in the coordinate transformation, by courtesy of the MaxPlanckInstitute for coal research.
 The LBFGS code by Jorge Nocedal was used
Download
DLFIND is distributed at http://ccpforge.cse.rl.ac.uk/gf/project/dlfind/ under the LGPL license. It can be checked out (svn) after registering at CCPForge. It is written in Fortran 95. The interface to the calling program is kept slim and welldefined, which should facilitate to interface DLFIND to various programs. Up to now, DLFIND is included in ChemShell, GAMESSUK, TeraChem, and a few other codes.


Trajectories of two transitionstate searches using the dimer method in DLFIND. The dimer midpoint converges to the transition state (blue sphere).


The dimer method applied to a biological system (the enzyme PHBH) in HDLC coordinates.


Converged nudgedelastic band path on an example surface (MüllerBrown potential). The green spheres indicate minima, the blue sphere indicates the climbing image which converged to the transition state.


Energy and geometries of a nudgedelastic band path of a simple chemical system, optimized with DLFIND.


