def penalty(f, gradf, g, gradg, x0, sigma_exp=2, tol=0.1):
x = x0.copy()
path = [ x.copy()]
converged = False
step = 1.0
while not converged:
sigma = .1*step**2.0
# Funktion + aktuelle Straffunktionen
def q(x):
res = f(x)
for lg in g(x):
res += minimum(0, sigma*lg)**2
return res
def gradq(x):
res = gradf(x)
for lg, lgg in zip(g(x), gradg(x)):
res += 2*minimum(0, sigma*lg)*sigma*lgg
return res
# steilster Abstieg mit Armijo-Schrittweite
x = armijo_steepest_descent(q, gradq, x, tol)
path.append(x.copy())
if min(g(x)) > -tol:
converged = True
step += 1.0
return array(path).transpose()
