moldesign.min package¶

Submodules¶

moldesign.min.base module¶

class moldesign.min.base.MinimizerBase(mol, nsteps=20, force_tolerance=<Quantity(0.00514220566583, 'eV / ang')>, frame_interval=None, _restart_from=0, _restart_energy=None)[source]¶

Bases: object

callback(*args)[source]¶

To be called after each minimization step

Parameters:	*args – ignored

grad(vector)[source]¶

Callback function to calculate the objective’s gradient

objective(vector)[source]¶

Callback function to calculate the objective function

run()[source]¶

moldesign.min.descent module¶

class moldesign.min.descent.GradientDescent(mol, max_atom_move=<Quantity(0.05, 'ang')>, scaling=<Quantity(0.01, 'ang ** 2 / eV')>, gamma=0.4, control=0.25, **kwargs)[source]¶

Bases: moldesign.min.base.MinimizerBase

A careful (perhaps overly careful) gradient descent implementation designed to relax structures far from equilibrium.

A backtracking line search is performed along the steepest gradient direction.

The maximum move for any single atom is also limited by max_atom_move

Note

This algorithm is good at stably removing large forces, but it’s very poorly suited to

locating any type of critical point; don’t use this to find a minimum!

References

https://www.math.washington.edu/~burke/crs/408/lectures/L7-line-search.pdf

Parameters:

mol (moldesign.Molecule) – molecule to minimize max_atom_move (Scalar[length]) – maximum displacement of a single atom scaling (Scalar[length/force]) – unit of displacement per unit force gamma (float) – number between 0 and 1 indicating scale factor for backtracking search control (float) – threshold for terminating line search; this is a proportion (0<=``control``<=1) of the expected function decrease **kwargs (dict) – kwargs from MinimizerBase

run()[source]¶

scale_move(grad)[source]¶

moldesign.min.descent.gradient_descent(*args, **kwargs)¶

A careful (perhaps overly careful) gradient descent implementation designed to relax structures far from equilibrium.

A backtracking line search is performed along the steepest gradient direction.

The maximum move for any single atom is also limited by max_atom_move

Note

This algorithm is good at stably removing large forces, but it’s very poorly suited to

locating any type of critical point; don’t use this to find a minimum!

References

https://www.math.washington.edu/~burke/crs/408/lectures/L7-line-search.pdf

Parameters:

mol (moldesign.Molecule) – molecule to minimize max_atom_move (Scalar[length]) – maximum displacement of a single atom scaling (Scalar[length/force]) – unit of displacement per unit force gamma (float) – number between 0 and 1 indicating scale factor for backtracking search control (float) – threshold for terminating line search; this is a proportion (0<=``control``<=1) of the expected function decrease **kwargs (dict) – kwargs from MinimizerBase

moldesign.min.scipy module¶

class moldesign.min.scipy.BFGS(mol, nsteps=20, force_tolerance=<Quantity(0.00514220566583, 'eV / ang')>, frame_interval=None, _restart_from=0, _restart_energy=None)[source]¶

Bases: moldesign.min.scipy.ScipyMinimizer

moldesign.min.scipy.bfgs(*args, **kwargs)¶

class moldesign.min.scipy.SequentialLeastSquares(mol, nsteps=20, force_tolerance=<Quantity(0.00514220566583, 'eV / ang')>, frame_interval=None, _restart_from=0, _restart_energy=None)[source]¶

Bases: moldesign.min.scipy.ScipyMinimizer

moldesign.min.scipy.sequential_least_squares(*args, **kwargs)¶

moldesign.min.smart module¶

class moldesign.min.smart.SmartMin(*args, **kwargs)[source]¶

Bases: moldesign.min.base.MinimizerBase

Uses gradient descent until forces fall below a threshold, then switches to BFGS (unconstrained) or SLSQP (constrained).

Parameters:	gd_threshold (u.Scalar[force]) – Use gradient descent if there are any forces larger than this; use an approximate hessian method (BFGS or SLSQP) otherwise

Note

Not really that smart.

run()[source]¶

moldesign.min.smart.minimize(*args, **kwargs)¶

Uses gradient descent until forces fall below a threshold, then switches to BFGS (unconstrained) or SLSQP (constrained).

Parameters:	gd_threshold (u.Scalar[force]) – Use gradient descent if there are any forces larger than this; use an approximate hessian method (BFGS or SLSQP) otherwise

Note

Not really that smart.