def __init__(self, grid, beta=0.1, nmaxold=7, weight=50.0):
self.beta = beta
self.nmaxold = nmaxold
self.weight = weight
self.x_iG = [] #densities
self.R_iG = [] #Residuals
self.A_ii = np.zeros((0,0))
self.dNt = 1.0 # initial error
if self.weight == 0.0:
self.use_metric = False
else:
self.use_metric = True
self.metric = Metric(grid, self.weight)
class Mixer:
"""Adapted from GPAW Pulay/DIIS mixer."""
def __init__(self, grid, beta=0.1, nmaxold=7, weight=50.0):
self.beta = beta
self.nmaxold = nmaxold
self.weight = weight
self.x_iG = [] #densities
self.R_iG = [] #Residuals
self.A_ii = np.zeros((0,0))
self.dNt = 1.0 # initial error
if self.weight == 0.0:
self.use_metric = False
else:
self.use_metric = True
self.metric = Metric(grid, self.weight)
def reset(self):
self.x_iG = [] #densities
self.R_iG = [] #Residuals
self.A_ii = np.zeros((0,0))
self.dNt = 1.0 # initial error
def get_error(self):
return self.dNt
def mix(self, x):
iold = len(self.x_iG)
if iold > 0:
if iold > self.nmaxold:
del self.x_iG[0]
del self.R_iG[0]
iold = self.nmaxold
R_G = x - self.x_iG[-1]
self.dNt = np.linalg.norm(R_G)
self.R_iG.append(R_G)
A_ii = np.zeros((iold,iold))
i2 = iold - 1
if self.use_metric:
mR_G = self.metric.apply(R_G)
else:
mR_G = R_G
for i1, R_1G in enumerate(self.R_iG):
A_ii[i1,i2] = np.dot(R_1G, mR_G)
A_ii[i2,i1] = A_ii[i1,i2]
A_ii[:i2,:i2] = self.A_ii[-i2:, -i2:]
self.A_ii = A_ii
try:
B_ii = np.linalg.inv(A_ii)
except np.linalg.LinAlgError:
alpha_i = np.zeros(iold)
alpha_i[-1] = 1.0
else:
alpha_i = B_ii.sum(1)
try:
# Normalize:
alpha_i /= alpha_i.sum()
except ZeroDivisionError:
alpha_i[:] = 0.0
alpha_i[-1] = 1.0
x[:] = 0.0
beta = self.beta
for i, alpha in enumerate(alpha_i):
x += alpha * self.x_iG[i]
x += alpha * beta * self.R_iG[i]
self.x_iG.append(x.copy())
