Optimizer
The geometry optimizer parameters can be set in this block. Some parameters are used by all methods, while others apply only to specific optimizers.
geopt parameters
general geopt parameters
method: (string) Optimization method.
default:
No default value.options:
sd: steepest descent method with energy feedback.
sdcg: steepest descent method followed by the conjugate gradient method.
sddiis: steepest descent method followed by the direct inversion in the iterative subspace (DIIS) method.
nlbfgs: Nocedal’s implementation of the limited momory Broyden–Fletcher–Goldfarb–Shanno (LBFGS) method [27].
bfgs: Ghasemi’s implementation the BFGS method with preferential moves along soft modes.
qbfgs: (only withing taskminhocao) Quantum Espresso’s version of BFGS for variable cell shapes [21].
fire: (only withing taskminhocao) The Fast Inertial Relaxation Engine (FIRE) method [28].
fmaxtol: (real) Convergence parameter. The optimization will terminate as soon as the maximum absolute value of all force vector components falls below this threshold. In units of Ha/Bohr. Must be specified.
default:
No default value.
alphax: (real) Sandard step size used in most optimizers. Optimal value depends on the system and method.
default:
No default value.
lprint: (logical) Verbosity setting. If True, detailed information at each iteration is printed.
default:
False
nit: (integer) Maximum number of iterations.
default:
1000
dxmax: (real) Maximum displacement for each atomic component. Units in Bohr. Not yet implemented for all methods.
default:
1.d-1
funits: (real) Factor to scale energy and force units. Currently only useful for Lennard-Jones potential, where the units can be scaled to be comparable to other potentials.
default:
1.d0
cellrelax: (logical) Activates variable cell shape relaxations if set to True. Not yet implemented in all methods.
default:
False
strfact: (real) The stress tensor is scaled by this factor and treated like forces in the process of optimization. Note that internally the unit of the stress tensor is \({\textrm{Ha}}/{\textrm{Bohr}^{3}}\). Only relevant for variable cell shape relaxations.
default:
1.d2
geoext: (logical) Some atomic simulation packages (e.g., LAMMPS, GULP, etc.) come with their
own implementations of geometry optimizers. If True, these external optimizers
will be used. Not yet implemented for all external codes.
default:
False
bfgs parameters
condnum: (real) Predetermined condition number of the system.
default:
1.d1
precaution: (string) Weighting of the quasi-Hessian. In Ghasemi’s controlled BFGS, moves are gradually affected by the quasi-Hessian instead of the initial, diagonal matrix. The parameter precaution controls the rate of this gradual transition.
default:
normaloptions:
high: A high precaution is employed and the quasi-Hessian is applied with a high rate.
normal: A normal precaution is employed and the quasi-Hessian is applied with a medium rate.
low: A normal precaution is employed and the quasi-Hessian is applied with a low rate.
fire parameters
dt_start: (real) Initial time step. Arbitrary units.
default:
No default value.
dt_min: (real) Minimal time step. Arbitrary units.
default:
1.d0
dt_max: (real) Maximal time step. Arbitrary units.
default:
8.d1
qbfgs parameters
qbfgsndim: (integer) Number of old forces and displacements vector used in the PULAY mixing of the residual vectors obtained on the basis of the inverse hessian matrix given by the BFGS algorithm. When bfgs_ndim = 1, the standard quasi-Newton BFGS method is used.
default:
1
qbfgstri: (real) Initial ionic displacement in the structural relaxation.
default:
5.d-1
qbfgstrmin: (real) Minimum ionic displacement in the structural relaxation. BFGS is reset when trust_radius < trust_radius_min.
default:
1.d-3
qbfgstrmax: (real) Maximum ionic displacement in the structural relaxation.
default:
8.d-1
qbfgsw1: (real) Parameter used in line search based on the Wolfe conditions.
default:
1.d-2
qbfgsw2: (real) Parameter used in line search based on the Wolfe conditions.
default:
5.d-1