|
| | GeometryOptimizer (const core::Molecule &molecule, const ConvergenceCriteria &criteria={}, const OptimizerSettings &settings={}) |
| void | set_initial_hessian (const Mat &cartesian_hessian) |
| | Initial Hessian in Cartesians (Hartree/Bohr^2, 3N x 3N, atom-major xyz), e.g.
|
| void | set_followed_mode (const Vec &cartesian_mode) |
| | Saddle-point search: start along the Hessian mode overlapping most with this Cartesian displacement (3N, atom-major; e.g.
|
| void | update (double energy, const Mat3N &gradient) |
| | Energy (Hartree) and gradient (Hartree/Bohr, 3 x N) at the geometry last returned by get_next_geometry().
|
| bool | step () |
| | Accept or reject the latest point, check convergence at the accepted one, and if not converged choose the next geometry.
|
| core::Molecule | get_next_geometry () const |
| void | shift_energy_reference (double shift) |
| | Shift every stored energy, when the energy surface moves by a constant (e.g.
|
| bool | is_converged () const |
| double | current_energy () const |
| | Energy at the last accepted point (Hartree).
|
| int | current_step () const |
| double | trust_radius () const |
| const CoordinateSystem & | coordinates () const |
| const Vec & | followed_mode () const |
| | Saddle-point search: the followed mode as a unit Cartesian displacement (3N, atom-major), and the number of internal modes with negative curvature, both at the last accepted point.
|
| int | negative_modes () const |
Geometry optimizer for minima and first-order saddle points (L.
-P. Wang and C. Song, J. Chem. Phys. 144, 214108 (2016), with the step control of geomeTRIC's current releases).
For a minimum, the step is the Newton step on a positive definite Hessian, shifted (Levenberg) to the length that moves the atoms by the trust radius (RMS, after alignment), in TRIC or Cartesian coordinates; the Hessian starts from Lindh's model and is updated by BFGS. For a saddle point, the step is the restricted-step partitioned RFO step (maximising along the followed mode, minimising along the rest), and the Hessian, best computed, is updated by Bofill's formula. A step whose energy change departs from the model too far is rejected and retaken from the last accepted geometry with a smaller trust radius. Coordinates are rebuilt where they become degenerate, the Hessian carried through Cartesians. Convergence is judged on per-atom Cartesian gradients and aligned displacements.
Use: update() with the energy and gradient at get_next_geometry(), then step(), until step() returns true.