def fuzz_state(self, f=1.0):
norm = la.norm(self.A)
fac = f*(norm / (self.q * self.D**2))
R = np.empty_like(self.A)
m.randomize_cmplx(R, -fac/2.0, fac/2.0)
self.A += R
def randomize(self, do_update=True):
"""Randomizes the parameter tensors self.A.
Parameters
----------
do_update : bool (True)
Whether to perform self.update() after randomizing.
"""
m.randomize_cmplx(self.A)
self.A /= la.norm(self.A)
if do_update:
self.update()
def expand_D(self, newD, refac=100, imfac=0):
"""Expands the bond dimension in a simple way.
New matrix entries are (mostly) randomized.
Parameters
----------
newD : int
The new bond-dimension.
refac : float
Scaling factor for the real component of the added noise.
imfac : float
Scaling factor for the imaginary component of the added noise.
"""
if newD < self.D:
return False
oldD = self.D
oldA = self.A
oldl = np.asarray(self.l)
oldr = np.asarray(self.r)
self._init_arrays(newD, self.q)
realnorm = la.norm(oldA.real.ravel())
imagnorm = la.norm(oldA.imag.ravel())
realfac = (realnorm / oldA.size) * refac
imagfac = (imagnorm / oldA.size) * imfac
# m.randomize_cmplx(newA[:, self.D:, self.D:], a=-fac, b=fac)
m.randomize_cmplx(self.A[:, :oldD, oldD:], a=0, b=realfac, aj=0, bj=imagfac)
m.randomize_cmplx(self.A[:, oldD:, :oldD], a=0, b=realfac, aj=0, bj=imagfac)
self.A[:, oldD:, oldD:] = 0 #for nearest-neighbour hamiltonian
# self.A[:, :oldD, oldD:] = oldA[:, :, :(newD - oldD)]
# self.A[:, oldD:, :oldD] = oldA[:, :(newD - oldD), :]
self.A[:, :oldD, :oldD] = oldA
self.l[:oldD, :oldD] = oldl
self.l[:oldD, oldD:].fill(la.norm(oldl) / oldD**2)
self.l[oldD:, :oldD].fill(la.norm(oldl) / oldD**2)
self.l[oldD:, oldD:].fill(la.norm(oldl) / oldD**2)
self.r[:oldD, :oldD] = oldr
self.r[oldD:, :oldD].fill(la.norm(oldr) / oldD**2)
self.r[:oldD, oldD:].fill(la.norm(oldr) / oldD**2)
self.r[oldD:, oldD:].fill(la.norm(oldr) / oldD**2)
def expand_D(self, newD):
"""Expands the bond dimension in a simple way.
New matrix entries are (mostly) randomized.
"""
if newD < self.D:
return False
oldD = self.D
oldA = self.A
oldK = self.K
oldl = np.asarray(self.l)
oldr = np.asarray(self.r)
self._init_arrays(newD, self.q)
realnorm = la.norm(oldA.real)
imagnorm = la.norm(oldA.imag)
realfac = (realnorm / (self.q * oldD**2))
imagfac = (imagnorm / (self.q * oldD**2))
# m.randomize_cmplx(newA[:, self.D:, self.D:], a=-fac, b=fac)
m.randomize_cmplx(self.A[:, :oldD, oldD:], a=0, b=realfac, aj=0, bj=imagfac)
m.randomize_cmplx(self.A[:, oldD:, :oldD], a=0, b=realfac, aj=0, bj=imagfac)
self.A[:, oldD:, oldD:] = 0 #for nearest-neighbour hamiltonian
# self.A[:, :oldD, oldD:] = oldA[:, :, :(newD - oldD)]
# self.A[:, oldD:, :oldD] = oldA[:, :(newD - oldD), :]
self.A[:, :oldD, :oldD] = oldA
self.l[:oldD, :oldD] = oldl
self.l[:oldD, oldD:].fill(la.norm(oldl) / oldD**2)
self.l[oldD:, :oldD].fill(la.norm(oldl) / oldD**2)
self.l[oldD:, oldD:].fill(la.norm(oldl) / oldD**2)
self.r[:oldD, :oldD] = oldr
self.r[oldD:, :oldD].fill(la.norm(oldr) / oldD**2)
self.r[:oldD, oldD:].fill(la.norm(oldr) / oldD**2)
self.r[oldD:, oldD:].fill(la.norm(oldr) / oldD**2)
self.K[:oldD, :oldD] = oldK
self.K[oldD:, :oldD].fill(la.norm(oldK) / oldD**2)
self.K[:oldD, oldD:].fill(la.norm(oldK) / oldD**2)
self.K[oldD:, oldD:].fill(la.norm(oldK) / oldD**2)
def add_noise(self, fac=1.0, do_update=True):
"""Adds some random (white) noise of a given magnitude to the parameter
tensors A.
Parameters
----------
fac : number
A factor determining the amplitude of the random noise.
do_update : bool (True)
Whether to perform self.update() after randomizing.
"""
norm = la.norm(self.A)
fac = fac * (norm / (self.q * self.D**2))
R = np.empty_like(self.A)
m.randomize_cmplx(R, -fac / 2.0, fac / 2.0)
self.A += R
if do_update:
self.update()
def randomize(self, fac=0.5):
m.randomize_cmplx(self.A, a=-fac, b=fac)
