1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
|
from abc import ABC, abstractmethod
import numpy as np
# Due to the high prevalence of cyclic imports surrounding ase.optimize,
# we define the Optimizable ABC here in utils.
# Can we find a better way?
class Optimizable(ABC):
@abstractmethod
def ndofs(self) -> int:
"""Return number of degrees of freedom."""
@abstractmethod
def get_x(self) -> np.ndarray:
"""Return current coordinates as a flat ndarray."""
@abstractmethod
def set_x(self, x: np.ndarray) -> None:
"""Set flat ndarray as current coordinates."""
@abstractmethod
def get_gradient(self) -> np.ndarray:
"""Return gradient at current coordinates as flat ndarray.
NOTE: Currently this is still the (flat) "forces" i.e.
the negative gradient. This must be fixed before the optimizable
API is done."""
# Callers who want Nx3 will do ".get_gradient().reshape(-1, 3)".
# We can probably weed out most such reshapings.
# Grep for the above expression in order to find places that should
# be updated.
@abstractmethod
def get_value(self) -> float:
"""Return function value at current coordinates."""
@abstractmethod
def iterimages(self):
"""Yield domain objects that can be saved as trajectory.
For example this can yield Atoms objects if the optimizer
has a trajectory that can write Atoms objects."""
def converged(self, gradient: np.ndarray, fmax: float) -> bool:
"""Standard implementation of convergence criterion.
This assumes that forces are the actual (Nx3) forces.
We can hopefully change this."""
assert gradient.ndim == 1
return self.gradient_norm(gradient) < fmax
def gradient_norm(self, gradient):
forces = gradient.reshape(-1, 3) # XXX Cartesian
return np.linalg.norm(forces, axis=1).max()
def __ase_optimizable__(self) -> 'Optimizable':
"""Return self, being already an Optimizable."""
return self
|
