Local Optimizers

TSASE holds the following local optimization routines:

Local Optimization

The local optimization algorithms available in TSASE are: SDLBFGS and MDmin

All local optimization classes have the following structure:

class Optimizer:
        def __init__(self, atoms, restart=None, logfile=None):
        def run(self, fmax=0.05, emax= 0.001, steps=1000000,optimizer='L2',maximize=False):
        def get_number_of_steps():

The structure of the local optimization classes is very similar to ASE implementation with a few additional features. Below describes the additional options in TSASE implementation:

optimizerIn ASEs implementation the convergence criteria is only the maximum per atom force

In TSASE implementation there are additional options:

‘L2’ : L2-norm for the entire force vector

‘maxatom’ : maximum per atom force

‘energy’ : value of potential energy

‘bgsd’a special case convergence criteria for the bgsd saddle search;

a trajectory is converged when the L2 norm of the force or the potential energy is below a specified value

fmax : specified force convergence criteria

emax : specified energy convergence criteria

Steepest Descent Limited memory BFGS

A limited memory version of the bfgs algorithm. Unlike the bfgs algorithm used in bfgs.py, the inverse of Hessian matrix is updated. The inverse Hessian is represented only as a diagonal matrix to save memor.y

This version of LBFGS is based off of ASE implementation with a few improvements. The initial guess for the inverse Hessian is updated at every step by estimating the curvature along the previous step direction. If a negative curvature is calculated or if the angle between the force and the LBFGS direction is greater than 90, then the memory is reset.

Below is an example script of how to use this optimizer:

al = tsase.calculators.al()
p = tsase.io.read_con('al.con')
p.set_calculator(al)
min = tsase.optimize.SDLBFGS(p, maxstep=0.1, memory=100)
min.run()

MDMin

ASE implementation of MDMin (See ASE for details) with the additional option to have a maxmimum step distance.

Below is and example of how to use this optimizer

min = tsase.optimize.MDMin(p,dt=0.1,maxstep=1.0)
min.run()

• MushyBox

  Relaxes a variable number of degrees of freedom which may include full cell relaxation