The DLFIND manual is attached. For convenience, a non-exhaustive list of default parameter values is provided here. The "under-the-hood" algorithms for choosing default parameters are in some cases more complicated than those presented here: e.g., multistate optimizations sometimes change defaults relative to their single-state congeners.
Printout level.
- 0: none
- 2: some (default)
- 4: verbose
- 6: debug
Type of coordinate system.
- 0: Cartesian (default)
- 1: hybrid delocalized internal coordinates, primitive internal coordinate scheme
- 2: hybrid delocalized internal coordinates, total connection scheme
- 3: delocalized internal coordinates, primitive internal coordinate scheme
- 4: delocalized internal coordinates, total connection scheme
- 10–14: Lagrange–Newton conical intersection search, with 2nd digit referring to above options
- 100+: nudged elastic band, quantum TS search, dimer method, chain search (see DLFIND manual).
Controls optimization algorithm.
- 0: steepest descent
- 1: Polak-Ribiere conjugate gradient w/ automatic restart
- 2: Polak-Ribiere conjugate gradient w/ restart every 10 steps
- 3: L-BGFS (default)
- 10: P-RFO, for transition state searches
Type of line search or trust radius.
- 0: simple scaling (default for
iopt
> 3) - 1: trust radius based on energy criterion (default when
iopt
is 3) - 2: trust radius based on gradient criterion (default when
iopt
is 0, 1, or 2)
Multistate calculations.
- 0: single-state calculation (default)
- 1: conical intersection optimization w/ penalty function algorithm
- 2: conical intersection optimization w/ gradient projection algorithm
- 3: conical intersection optimization w/ Lagrange–Newton algorithm
Microiterative optimization for multilayer systems. Layers assigned via spec
.
- 0: standard (non-microiterative) optimization (default)
- 1: microiterative optimization
Method used to generate initial Hessian matrix.
- 0: external calculation using
dlf_get_hessian
(default ifdlf_get_hessian
defined) - 1: build with one-point finite difference of gradient
- 2: build with two-point finite difference of gradient (default otherwise)
- 3: build diagonal Hessian with one-point finite difference
- 4: set Hessian to be an identity matrix
How the Hessian is updated.
- 0: no update
- 1: Powell update
- 2: Bofill update (default)
- 3: BGFS update
Convergence criterion on max gradient component. Default is 4.5E-4.
Convergence criterion on max energy change. Default is 1E-6.
Max number of cycles. Default is 100.
Memory for L-BGFS algorithm. Defaults to nvarin
, with min value 5 and max value 50.
Max number of Hessian updates. Default is 50.
Maximum stepsize (in internal coordinates). Default is 0.5.
Delta for finite-difference Hessian calculations (in internal coordinates). Default is 0.01.
Freezing atoms, etc for optimization. (spec
is an array with an entry for each atom.)
- >0: active, treated normally. The value can be used to encode residue/fragment number.
- 0: active in optimization, but treated in Cartesian coordinates no matter what.
- -1: frozen
- -2: x-coordinate frozen
- -3: y-coordinate frozen
- -4: z-coordinate frozen
- -23: x-coordinate and y-coordinate frozen
- -24: x-coordinate and z-coordinate frozen
- -34: y-coordinate and z-coordinate frozen
spec
also contains a way to specify constraints. See manual for more information.