def setUp(self): #TODO: Test rectangular x as well
self.d = [2, 3]
self.D = [2, 4, 3]
self.l0 = sp.rand(self.D[0],
self.D[0]) + 1.j * sp.rand(self.D[0], self.D[0])
self.r1 = sp.rand(self.D[1],
self.D[1]) + 1.j * sp.rand(self.D[1], self.D[1])
self.r2 = sp.rand(self.D[2],
self.D[2]) + 1.j * sp.rand(self.D[2], self.D[2])
self.ld0 = mm.simple_diag_matrix(
sp.rand(self.D[0]) + 1.j * sp.rand(self.D[0]))
self.rd1 = mm.simple_diag_matrix(
sp.rand(self.D[1]) + 1.j * sp.rand(self.D[1]))
self.rd2 = mm.simple_diag_matrix(
sp.rand(self.D[2]) + 1.j * sp.rand(self.D[2]))
self.eye0 = mm.eyemat(self.D[0], dtype=sp.complex128)
self.eye1 = mm.eyemat(self.D[1], dtype=sp.complex128)
self.eye2 = mm.eyemat(self.D[2], dtype=sp.complex128)
self.A0 = sp.rand(self.d[0], self.D[0], self.D[0]) + 1.j * sp.rand(
self.d[0], self.D[0], self.D[0])
self.A1 = sp.rand(self.d[0], self.D[0], self.D[1]) + 1.j * sp.rand(
self.d[0], self.D[0], self.D[1])
self.A2 = sp.rand(self.d[1], self.D[1], self.D[2]) + 1.j * sp.rand(
self.d[1], self.D[1], self.D[2])
self.A3 = sp.rand(self.d[1], self.D[2], self.D[2]) + 1.j * sp.rand(
self.d[1], self.D[2], self.D[2])
self.B1 = sp.rand(self.d[0], self.D[0], self.D[1]) + 1.j * sp.rand(
self.d[0], self.D[0], self.D[1])
self.B2 = sp.rand(self.d[1], self.D[1], self.D[2]) + 1.j * sp.rand(
self.d[1], self.D[1], self.D[2])
self.E1_AB = make_E_noop(self.A1, self.B1)
self.E2_AB = make_E_noop(self.A2, self.B2)
self.op1s_1 = sp.rand(self.d[0],
self.d[0]) + 1.j * sp.rand(self.d[0], self.d[0])
self.E1_op_AB = make_E_1s(self.A1, self.B1, self.op1s_1)
self.op2s = sp.rand(self.d[0], self.d[1],
self.d[0], self.d[1]) + 1.j * sp.rand(
self.d[0], self.d[1], self.d[0], self.d[1])
self.E12_op = make_E_2s(self.A1, self.A2, self.B1, self.B2, self.op2s)
self.AA12 = tc.calc_AA(self.A1, self.A2)
self.BB12 = tc.calc_AA(self.B1, self.B2)
self.C_A12 = tc.calc_C_mat_op_AA(self.op2s, self.AA12)
self.C_conj_B12 = tc.calc_C_conj_mat_op_AA(self.op2s, self.BB12)
self.C01 = sp.rand(self.d[0], self.d[0],
self.D[0], self.D[1]) + 1.j * sp.rand(
self.d[0], self.d[0], self.D[0], self.D[1])
def restore_LCF_l(A, lm1, Gm1, sanity_checks=False, zero_tol=1E-15):
if Gm1 is None:
GhGm1 = lm1
else:
GhGm1 = Gm1.conj().T.dot(lm1.dot(Gm1))
M = eps_l_noop(GhGm1, A, A)
G, Gi, new_D = herm_fac_with_inv(M,
zero_tol=zero_tol,
return_rank=True,
sanity_checks=sanity_checks)
if Gm1 is None:
Gm1 = G
if sanity_checks:
if new_D == A.shape[2]:
eye = sp.eye(A.shape[2])
else:
eye = mm.simple_diag_matrix(np.append(np.zeros(A.shape[2] - new_D),
np.ones(new_D)),
dtype=A.dtype)
if not sp.allclose(G.dot(Gi), eye, atol=1E-13, rtol=1E-13):
log.warning("Sanity Fail in restore_LCF_l!: Bad GT!")
for s in xrange(A.shape[0]):
A[s] = Gm1.dot(A[s]).dot(Gi)
if new_D == A.shape[2]:
l = mm.eyemat(A.shape[2], A.dtype)
else:
l = mm.simple_diag_matrix(np.append(np.zeros(A.shape[2] - new_D),
np.ones(new_D)),
dtype=A.dtype)
if sanity_checks:
lm1_ = mm.eyemat(A.shape[1], A.dtype)
l_ = eps_l_noop(lm1_, A, A)
if not sp.allclose(l_, l.A, atol=1E-13, rtol=1E-13):
log.warning("Sanity Fail in restore_LCF_l!: l is bad")
log.warning(la.norm(l_ - l))
return l, G, Gi
def restore_LCF_l(A, lm1, Gm1, sanity_checks=False, zero_tol=1E-15):
if Gm1 is None:
GhGm1 = lm1
else:
GhGm1 = Gm1.conj().T.dot(lm1.dot(Gm1))
M = eps_l_noop(GhGm1, A, A)
G, Gi, new_D = herm_fac_with_inv(M, zero_tol=zero_tol, return_rank=True,
sanity_checks=sanity_checks)
if Gm1 is None:
Gm1 = G
if sanity_checks:
if new_D == A.shape[2]:
eye = sp.eye(A.shape[2])
else:
eye = mm.simple_diag_matrix(np.append(np.zeros(A.shape[2] - new_D),
np.ones(new_D)), dtype=A.dtype)
if not sp.allclose(G.dot(Gi), eye, atol=1E-13, rtol=1E-13):
log.warning("Sanity Fail in restore_LCF_l!: Bad GT!")
for s in xrange(A.shape[0]):
A[s] = Gm1.dot(A[s]).dot(Gi)
if new_D == A.shape[2]:
l = mm.eyemat(A.shape[2], A.dtype)
else:
l = mm.simple_diag_matrix(np.append(np.zeros(A.shape[2] - new_D),
np.ones(new_D)), dtype=A.dtype)
if sanity_checks:
lm1_ = mm.eyemat(A.shape[1], A.dtype)
l_ = eps_l_noop(lm1_, A, A)
if not sp.allclose(l_, l.A, atol=1E-13, rtol=1E-13):
log.warning("Sanity Fail in restore_LCF_l!: l is bad")
log.warning(la.norm(l_ - l))
return l, G, Gi
def setUp(self): #TODO: Test rectangular x as well
self.d = [2, 3]
self.D = [2, 4, 3]
self.l0 = sp.rand(self.D[0], self.D[0]) + 1.j * sp.rand(self.D[0], self.D[0])
self.r1 = sp.rand(self.D[1], self.D[1]) + 1.j * sp.rand(self.D[1], self.D[1])
self.r2 = sp.rand(self.D[2], self.D[2]) + 1.j * sp.rand(self.D[2], self.D[2])
self.ld0 = mm.simple_diag_matrix(sp.rand(self.D[0]) + 1.j * sp.rand(self.D[0]))
self.rd1 = mm.simple_diag_matrix(sp.rand(self.D[1]) + 1.j * sp.rand(self.D[1]))
self.rd2 = mm.simple_diag_matrix(sp.rand(self.D[2]) + 1.j * sp.rand(self.D[2]))
self.eye0 = mm.eyemat(self.D[0], dtype=sp.complex128)
self.eye1 = mm.eyemat(self.D[1], dtype=sp.complex128)
self.eye2 = mm.eyemat(self.D[2], dtype=sp.complex128)
self.A0 = sp.rand(self.d[0], self.D[0], self.D[0]) + 1.j * sp.rand(self.d[0], self.D[0], self.D[0])
self.A1 = sp.rand(self.d[0], self.D[0], self.D[1]) + 1.j * sp.rand(self.d[0], self.D[0], self.D[1])
self.A2 = sp.rand(self.d[1], self.D[1], self.D[2]) + 1.j * sp.rand(self.d[1], self.D[1], self.D[2])
self.A3 = sp.rand(self.d[1], self.D[2], self.D[2]) + 1.j * sp.rand(self.d[1], self.D[2], self.D[2])
self.B1 = sp.rand(self.d[0], self.D[0], self.D[1]) + 1.j * sp.rand(self.d[0], self.D[0], self.D[1])
self.B2 = sp.rand(self.d[1], self.D[1], self.D[2]) + 1.j * sp.rand(self.d[1], self.D[1], self.D[2])
self.E1_AB = make_E_noop(self.A1, self.B1)
self.E2_AB = make_E_noop(self.A2, self.B2)
self.op1s_1 = sp.rand(self.d[0], self.d[0]) + 1.j * sp.rand(self.d[0], self.d[0])
self.E1_op_AB = make_E_1s(self.A1, self.B1, self.op1s_1)
self.op2s = sp.rand(self.d[0], self.d[1], self.d[0], self.d[1]) + 1.j * sp.rand(self.d[0], self.d[1], self.d[0], self.d[1])
self.E12_op = make_E_2s(self.A1, self.A2, self.B1, self.B2, self.op2s)
self.AA12 = tc.calc_AA(self.A1, self.A2)
self.BB12 = tc.calc_AA(self.B1, self.B2)
self.C_A12 = tc.calc_C_mat_op_AA(self.op2s, self.AA12)
self.C_conj_B12 = tc.calc_C_conj_mat_op_AA(self.op2s, self.BB12)
self.C01 = sp.rand(self.d[0], self.d[0], self.D[0], self.D[1]) + 1.j * sp.rand(self.d[0], self.d[0], self.D[0], self.D[1])
def restore_LCF_r(A, r, Gi, sanity_checks=False):
if Gi is None:
x = r
else:
x = Gi.dot(r.dot(Gi.conj().T))
M = eps_r_noop(x, A, A)
ev, EV = la.eigh(
M) #wraps lapack routines, which return eigenvalues in ascending order
if sanity_checks:
assert np.all(ev == np.sort(ev)), "unexpected eigenvalue ordering"
rm1 = mm.simple_diag_matrix(ev, dtype=A.dtype)
Gm1 = EV.conj().T
if Gi is None:
Gi = EV #for left uniform case
r = rm1 #for sanity check
for s in xrange(A.shape[0]):
A[s] = Gm1.dot(A[s].dot(Gi))
if sanity_checks:
rm1_ = eps_r_noop(r, A, A)
if not sp.allclose(rm1_, rm1, atol=1E-12, rtol=1E-12):
log.warning("Sanity Fail in restore_LCF_r!: r is bad!")
log.warning(la.norm(rm1_ - rm1))
Gm1_i = EV
if sanity_checks:
eye = sp.eye(A.shape[1])
if not sp.allclose(sp.dot(Gm1, Gm1_i), eye, atol=1E-12, rtol=1E-12):
log.warning("Sanity Fail in restore_LCF_r!: Bad GT! (off by %g)",
la.norm(sp.dot(Gm1, Gm1_i) - eye))
return rm1, Gm1, Gm1_i
def restore_LCF_r(A, r, Gi, sanity_checks=False):
if Gi is None:
x = r
else:
x = Gi.dot(r.dot(Gi.conj().T))
M = eps_r_noop(x, A, A)
ev, EV = la.eigh(M) #wraps lapack routines, which return eigenvalues in ascending order
if sanity_checks:
assert np.all(ev == np.sort(ev)), "unexpected eigenvalue ordering"
rm1 = mm.simple_diag_matrix(ev, dtype=A.dtype)
Gm1 = EV.conj().T
if Gi is None:
Gi = EV #for left uniform case
r = rm1 #for sanity check
for s in xrange(A.shape[0]):
A[s] = Gm1.dot(A[s].dot(Gi))
if sanity_checks:
rm1_ = eps_r_noop(r, A, A)
if not sp.allclose(rm1_, rm1, atol=1E-12, rtol=1E-12):
log.warning("Sanity Fail in restore_LCF_r!: r is bad!")
log.warning(la.norm(rm1_ - rm1))
Gm1_i = EV
if sanity_checks:
eye = sp.eye(A.shape[1])
if not sp.allclose(sp.dot(Gm1, Gm1_i), eye,
atol=1E-12, rtol=1E-12):
log.warning("Sanity Fail in restore_LCF_r!: Bad GT! (off by %g)", la.norm(sp.dot(Gm1, Gm1_i) - eye))
return rm1, Gm1, Gm1_i
def restore_RCF_r(A,
r,
G_n_i,
zero_tol=1E-15,
sanity_checks=False,
sc_data=''):
"""Transforms a single A[n] to obtain r[n - 1] = eye(D).
Implements the condition for right-orthonormalization from sub-section
3.1, theorem 1 of arXiv:quant-ph/0608197v2.
This function must be called for each n in turn, starting at N + 1,
passing the gauge transformation matrix from the previous step
as an argument.
Finds a G[n-1] such that orthonormalization is fulfilled for n.
If rank-deficiency is encountered, the result fulfills the orthonormality
condition in the occupied subspace with the zeros at the top-left
(for example r = diag([0, 0, 1, 1, 1, 1, 1])).
Parameters
----------
A : ndarray
The parameter tensor for the nth site A[n].
r : ndarray or object with array interface
The matrix r[n].
G_n_i : ndarray
The inverse gauge transform matrix for site n obtained in the previous step (for n + 1).
sanity_checks : bool (False)
Whether to perform additional sanity checks.
zero_tol : float
Tolerance for detecting zeros.
Returns
-------
r_nm1 : ndarray or simple_diag_matrix or eyemat
The new matrix r[n - 1].
G_nm1 : ndarray
The gauge transformation matrix for the site n - 1.
G_n_m1_i : ndarray
The inverse gauge transformation matrix for the site n - 1.
"""
if G_n_i is None:
GGh_n_i = r
else:
GGh_n_i = G_n_i.dot(r.dot(G_n_i.conj().T))
M = eps_r_noop(GGh_n_i, A, A)
X, Xi, new_D = herm_fac_with_inv(M,
zero_tol=zero_tol,
return_rank=True,
sanity_checks=sanity_checks)
G_nm1 = Xi.conj().T
G_nm1_i = X.conj().T
if G_n_i is None:
G_n_i = G_nm1_i
if sanity_checks:
#GiG may not be equal to eye in the case of rank-deficiency,
#but the rest should lie in the null space of A.
GiG = G_nm1_i.dot(G_nm1)
As = np.sum(A, axis=0)
if not sp.allclose(
GiG.dot(As).dot(G_n_i), As.dot(G_n_i), atol=1E-13, rtol=1E-13):
log.warning("Sanity Fail in restore_RCF_r!: Bad GT! %s %s",
la.norm(GiG.dot(As).dot(G_n_i) - As.dot(G_n_i)),
sc_data)
for s in xrange(A.shape[0]):
A[s] = G_nm1.dot(A[s]).dot(G_n_i)
if new_D == A.shape[1]:
r_nm1 = mm.eyemat(A.shape[1], A.dtype)
else:
r_nm1 = sp.zeros((A.shape[1]), dtype=A.dtype)
r_nm1[-new_D:] = 1
r_nm1 = mm.simple_diag_matrix(r_nm1, dtype=A.dtype)
if sanity_checks:
r_nm1_ = G_nm1.dot(M).dot(G_nm1.conj().T)
if not sp.allclose(r_nm1_, r_nm1.A, atol=1E-13, rtol=1E-13):
log.warning(
"Sanity Fail in restore_RCF_r!: r != g old_r gH! %s %s",
la.norm(r_nm1_ - r_nm1), sc_data)
r_nm1_ = eps_r_noop(r, A, A)
if not sp.allclose(r_nm1_, r_nm1.A, atol=1E-13, rtol=1E-13):
log.warning("Sanity Fail in restore_RCF_r!: r is bad! %s %s",
la.norm(r_nm1_ - r_nm1), sc_data)
return r_nm1, G_nm1, G_nm1_i
def herm_fac_with_inv(A,
lower=False,
zero_tol=1E-15,
return_rank=False,
force_evd=False,
sanity_checks=False,
sc_data=''):
"""Factorizes a Hermitian matrix using either Cholesky or eigenvalue decomposition.
Decomposes a Hermitian A as A = X*X or, if lower == True, A = XX*.
Tries Cholesky first by default, then falls back to EVD if the matrix is
not positive-definite. If Cholesky decomposition is used, X is upper (or lower)
triangular. For the EVD decomposition, the inverse becomes a pseudo-inverse
and all eigenvalues below the zero-tolerance are set to zero.
Parameters
----------
A : ndarray
The Hermitian matrix to be factorized.
lower : bool
Refers to Cholesky factorization. If True, factorize as A = XX*, otherwise as A = X*X
zero_tol : float
Tolerance for detection of zeros in EVD case.
return_rank : bool
Whether to return the rank of A. The detected rank is affected by zero_tol.
force_evd : bool
Whether to force eigenvalue instead of Cholesky decomposition.
sanity_checks : bool
Whether to perform soem basic sanity checks.
"""
if not force_evd:
try:
x = la.cholesky(A, lower=lower)
xi = mm.invtr(x, lower=lower)
nonzeros = A.shape[0]
except sp.linalg.LinAlgError: #this usually means a is not pos. def.
force_evd = True
if force_evd:
ev, EV = la.eigh(
A, turbo=True
) #wraps lapack routines, which return eigenvalues in ascending order
if sanity_checks:
assert np.all(
ev == np.sort(ev)
), "Sanity fail in herm_fac_with_inv(): Unexpected eigenvalue ordering"
if ev.min() < -zero_tol:
log.warning(
"Sanity fail in herm_fac_with_inv(): Discarding negative eigenvalues! %s %s",
ev.min(), sc_data)
nonzeros = np.count_nonzero(ev > zero_tol)
ev_sq = sp.zeros_like(ev, dtype=A.dtype)
ev_sq[-nonzeros:] = sp.sqrt(ev[-nonzeros:])
#Replace almost-zero values with zero and perform a pseudo-inverse
ev_sq_i = sp.zeros_like(ev, dtype=A.dtype)
ev_sq_i[-nonzeros:] = 1. / ev_sq[-nonzeros:]
ev_sq_i = mm.simple_diag_matrix(ev_sq_i, dtype=A.dtype)
ev_sq = mm.simple_diag_matrix(ev_sq, dtype=A.dtype)
if lower:
x = ev_sq.dot_left(EV)
xi = ev_sq_i.dot(EV.conj().T)
else:
x = ev_sq.dot(EV.conj().T)
xi = ev_sq_i.dot_left(EV)
if sanity_checks:
if not sp.allclose(A, A.conj().T, atol=1E-13, rtol=1E-13):
log.warning(
"Sanity fail in herm_fac_with_inv(): A is not Hermitian! %s %s",
la.norm(A - A.conj().T), sc_data)
eye = sp.zeros((A.shape[0]), dtype=A.dtype)
eye[-nonzeros:] = 1
eye = mm.simple_diag_matrix(eye)
if lower:
if not sp.allclose(xi.dot(x), eye, atol=1E-13, rtol=1E-13):
log.warning(
"Sanity fail in herm_fac_with_inv(): Bad left inverse! %s %s",
la.norm(xi.dot(x) - eye), sc_data)
if not sp.allclose(x.dot(x.conj().T), A, atol=1E-13, rtol=1E-13):
log.warning(
"Sanity fail in herm_fac_with_inv(): Bad decomp! %s %s",
la.norm(x.dot(x.conj().T) - A), sc_data)
if not sp.allclose(
xi.dot(A).dot(xi.conj().T), eye, atol=1E-13, rtol=1E-13):
log.warning(
"Sanity fail in herm_fac_with_inv(): Bad A inverse! %s %s",
la.norm(xi.conj().T.dot(A).dot(xi) - eye), sc_data)
else:
if not sp.allclose(x.dot(xi), eye, atol=1E-13, rtol=1E-13):
log.warning(
"Sanity fail in herm_fac_with_inv(): Bad right inverse! %s %s",
la.norm(x.dot(xi) - eye), sc_data)
if not sp.allclose(x.conj().T.dot(x), A, atol=1E-13, rtol=1E-13):
log.warning(
"Sanity fail in herm_fac_with_inv(): Bad decomp! %s %s",
la.norm(x.conj().T.dot(x) - A), sc_data)
if not sp.allclose(
xi.conj().T.dot(A).dot(xi), eye, atol=1E-13, rtol=1E-13):
log.warning(
"Sanity fail in herm_fac_with_inv(): Bad A inverse! %s %s",
la.norm(xi.conj().T.dot(A).dot(xi) - eye), sc_data)
if return_rank:
return x, xi, nonzeros
else:
return x, xi
def herm_sqrt_inv(x,
zero_tol=1E-15,
sanity_checks=False,
return_rank=False,
sc_data=''):
if isinstance(x, mm.eyemat):
x_sqrt = x
x_sqrt_i = x
rank = x.shape[0]
else:
try:
ev = x.diag #simple_diag_matrix
EV = None
except AttributeError:
ev, EV = la.eigh(x)
zeros = ev <= zero_tol #throw away negative results too!
ev_sqrt = sp.sqrt(ev)
err = sp.seterr(divide='ignore', invalid='ignore')
try:
ev_sqrt_i = 1 / ev_sqrt
ev_sqrt[zeros] = 0
ev_sqrt_i[zeros] = 0
finally:
sp.seterr(divide=err['divide'], invalid=err['invalid'])
if EV is None:
x_sqrt = mm.simple_diag_matrix(ev_sqrt, dtype=x.dtype)
x_sqrt_i = mm.simple_diag_matrix(ev_sqrt_i, dtype=x.dtype)
else:
B = mm.mmul_diag(ev_sqrt, EV.conj().T)
x_sqrt = EV.dot(B)
B = mm.mmul_diag(ev_sqrt_i, EV.conj().T)
x_sqrt_i = EV.dot(B)
rank = x.shape[0] - np.count_nonzero(zeros)
if sanity_checks:
if ev.min() < -zero_tol:
log.warning(
"Sanity Fail in herm_sqrt_inv(): Throwing away negative eigenvalues! %s %s",
ev.min(), sc_data)
if not np.allclose(x_sqrt.dot(x_sqrt), x):
log.warning(
"Sanity Fail in herm_sqrt_inv(): x_sqrt is bad! %s %s",
la.norm(x_sqrt.dot(x_sqrt) - x), sc_data)
if EV is None:
nulls = sp.zeros(x.shape[0])
nulls[zeros] = 1
nulls = sp.diag(nulls)
else: #if we did an EVD then we use the eigenvectors
nulls = EV.copy()
nulls[:, sp.invert(zeros)] = 0
nulls = nulls.dot(nulls.conj().T)
eye = np.eye(x.shape[0])
if not np.allclose(x_sqrt.dot(x_sqrt_i), eye - nulls):
log.warning(
"Sanity Fail in herm_sqrt_inv(): x_sqrt_i is bad! %s %s",
la.norm(x_sqrt.dot(x_sqrt_i) - eye + nulls), sc_data)
if return_rank:
return x_sqrt, x_sqrt_i, rank
else:
return x_sqrt, x_sqrt_i
def restore_RCF_l(A, lm1, Gm1, sanity_checks=False):
"""Transforms a single A[n] to obtain diagonal l[n].
Applied after restore_RCF_r(), this completes the full canonical form
of sub-section 3.1, theorem 1 of arXiv:quant-ph/0608197v2.
This function must be called for each n in turn, starting at 1,
passing the gauge transformation matrix from the previous step
as an argument.
Finds a G[n] such that orthonormalization is fulfilled for n.
The diagonal entries of l[n] are sorted in
ascending order (for example l[n] = diag([0, 0, 0.1, 0.2, ...])).
Parameters
----------
A : ndarray
The parameter tensor for the nth site A[n].
lm1 : ndarray or object with array interface
The matrix l[n - 1].
Gm1 : ndarray
The gauge transform matrix for site n obtained in the previous step (for n - 1).
sanity_checks : bool (False)
Whether to perform additional sanity checks.
Returns
-------
l : ndarray or simple_diag_matrix
The new, diagonal matrix l[n].
G : ndarray
The gauge transformation matrix for site n.
G_i : ndarray
Inverse of G.
"""
if Gm1 is None:
x = lm1
else:
x = Gm1.conj().T.dot(lm1.dot(Gm1))
M = eps_l_noop(x, A, A)
ev, EV = la.eigh(
M) #wraps lapack routines, which return eigenvalues in ascending order
if sanity_checks:
assert np.all(ev == np.sort(ev)), "unexpected eigenvalue ordering"
l = mm.simple_diag_matrix(ev, dtype=A.dtype)
G_i = EV
if Gm1 is None:
Gm1 = EV.conj().T #for left uniform case
lm1 = l #for sanity check
for s in xrange(A.shape[0]):
A[s] = Gm1.dot(A[s].dot(G_i))
if sanity_checks:
l_ = eps_l_noop(lm1, A, A)
if not sp.allclose(l_, l, atol=1E-12, rtol=1E-12):
log.warning("Sanity Fail in restore_RCF_l!: l is bad!")
log.warning(la.norm(l_ - l))
G = EV.conj().T
if sanity_checks:
eye = sp.eye(A.shape[2])
if not sp.allclose(sp.dot(G, G_i), eye, atol=1E-12, rtol=1E-12):
log.warning("Sanity Fail in restore_RCF_l!: Bad GT! (off by %g)",
la.norm(sp.dot(G, G_i) - eye))
return l, G, G_i
def restore_RCF_r(A, r, G_n_i, zero_tol=1E-15, sanity_checks=False, sc_data=''):
"""Transforms a single A[n] to obtain r[n - 1] = eye(D).
Implements the condition for right-orthonormalization from sub-section
3.1, theorem 1 of arXiv:quant-ph/0608197v2.
This function must be called for each n in turn, starting at N + 1,
passing the gauge transformation matrix from the previous step
as an argument.
Finds a G[n-1] such that orthonormalization is fulfilled for n.
If rank-deficiency is encountered, the result fulfills the orthonormality
condition in the occupied subspace with the zeros at the top-left
(for example r = diag([0, 0, 1, 1, 1, 1, 1])).
Parameters
----------
A : ndarray
The parameter tensor for the nth site A[n].
r : ndarray or object with array interface
The matrix r[n].
G_n_i : ndarray
The inverse gauge transform matrix for site n obtained in the previous step (for n + 1).
sanity_checks : bool (False)
Whether to perform additional sanity checks.
zero_tol : float
Tolerance for detecting zeros.
Returns
-------
r_nm1 : ndarray or simple_diag_matrix or eyemat
The new matrix r[n - 1].
G_nm1 : ndarray
The gauge transformation matrix for the site n - 1.
G_n_m1_i : ndarray
The inverse gauge transformation matrix for the site n - 1.
"""
if G_n_i is None:
GGh_n_i = r
else:
GGh_n_i = G_n_i.dot(r.dot(G_n_i.conj().T))
M = eps_r_noop(GGh_n_i, A, A)
X, Xi, new_D = herm_fac_with_inv(M, zero_tol=zero_tol, return_rank=True,
sanity_checks=sanity_checks)
G_nm1 = Xi.conj().T
G_nm1_i = X.conj().T
if G_n_i is None:
G_n_i = G_nm1_i
if sanity_checks:
#GiG may not be equal to eye in the case of rank-deficiency,
#but the rest should lie in the null space of A.
GiG = G_nm1_i.dot(G_nm1)
As = np.sum(A, axis=0)
if not sp.allclose(GiG.dot(As).dot(G_n_i),
As.dot(G_n_i), atol=1E-13, rtol=1E-13):
log.warning("Sanity Fail in restore_RCF_r!: Bad GT! %s %s",
la.norm(GiG.dot(As).dot(G_n_i) - As.dot(G_n_i)), sc_data)
for s in xrange(A.shape[0]):
A[s] = G_nm1.dot(A[s]).dot(G_n_i)
if new_D == A.shape[1]:
r_nm1 = mm.eyemat(A.shape[1], A.dtype)
else:
r_nm1 = sp.zeros((A.shape[1]), dtype=A.dtype)
r_nm1[-new_D:] = 1
r_nm1 = mm.simple_diag_matrix(r_nm1, dtype=A.dtype)
if sanity_checks:
r_nm1_ = G_nm1.dot(M).dot(G_nm1.conj().T)
if not sp.allclose(r_nm1_, r_nm1.A, atol=1E-13, rtol=1E-13):
log.warning("Sanity Fail in restore_RCF_r!: r != g old_r gH! %s %s",
la.norm(r_nm1_ - r_nm1), sc_data)
r_nm1_ = eps_r_noop(r, A, A)
if not sp.allclose(r_nm1_, r_nm1.A, atol=1E-13, rtol=1E-13):
log.warning("Sanity Fail in restore_RCF_r!: r is bad! %s %s",
la.norm(r_nm1_ - r_nm1), sc_data)
return r_nm1, G_nm1, G_nm1_i
def herm_fac_with_inv(A, lower=False, zero_tol=1E-15, return_rank=False,
force_evd=False, sanity_checks=False, sc_data=''):
"""Factorizes a Hermitian matrix using either Cholesky or eigenvalue decomposition.
Decomposes a Hermitian A as A = X*X or, if lower == True, A = XX*.
Tries Cholesky first by default, then falls back to EVD if the matrix is
not positive-definite. If Cholesky decomposition is used, X is upper (or lower)
triangular. For the EVD decomposition, the inverse becomes a pseudo-inverse
and all eigenvalues below the zero-tolerance are set to zero.
Parameters
----------
A : ndarray
The Hermitian matrix to be factorized.
lower : bool
Refers to Cholesky factorization. If True, factorize as A = XX*, otherwise as A = X*X
zero_tol : float
Tolerance for detection of zeros in EVD case.
return_rank : bool
Whether to return the rank of A. The detected rank is affected by zero_tol.
force_evd : bool
Whether to force eigenvalue instead of Cholesky decomposition.
sanity_checks : bool
Whether to perform soem basic sanity checks.
"""
if not force_evd:
try:
x = la.cholesky(A, lower=lower)
xi = mm.invtr(x, lower=lower)
nonzeros = A.shape[0]
except sp.linalg.LinAlgError: #this usually means a is not pos. def.
force_evd = True
if force_evd:
ev, EV = la.eigh(A, turbo=True) #wraps lapack routines, which return eigenvalues in ascending order
if sanity_checks:
assert np.all(ev == np.sort(ev)), "Sanity fail in herm_fac_with_inv(): Unexpected eigenvalue ordering"
if ev.min() < -zero_tol:
log.warning("Sanity fail in herm_fac_with_inv(): Discarding negative eigenvalues! %s %s",
ev.min(), sc_data)
nonzeros = np.count_nonzero(ev > zero_tol)
ev_sq = sp.zeros_like(ev, dtype=A.dtype)
ev_sq[-nonzeros:] = sp.sqrt(ev[-nonzeros:])
#Replace almost-zero values with zero and perform a pseudo-inverse
ev_sq_i = sp.zeros_like(ev, dtype=A.dtype)
ev_sq_i[-nonzeros:] = 1. / ev_sq[-nonzeros:]
ev_sq_i = mm.simple_diag_matrix(ev_sq_i, dtype=A.dtype)
ev_sq = mm.simple_diag_matrix(ev_sq, dtype=A.dtype)
if lower:
x = ev_sq.dot_left(EV)
xi = ev_sq_i.dot(EV.conj().T)
else:
x = ev_sq.dot(EV.conj().T)
xi = ev_sq_i.dot_left(EV)
if sanity_checks:
if not sp.allclose(A, A.conj().T, atol=1E-13, rtol=1E-13):
log.warning("Sanity fail in herm_fac_with_inv(): A is not Hermitian! %s %s",
la.norm(A - A.conj().T), sc_data)
eye = sp.zeros((A.shape[0]), dtype=A.dtype)
eye[-nonzeros:] = 1
eye = mm.simple_diag_matrix(eye)
if lower:
if not sp.allclose(xi.dot(x), eye, atol=1E-13, rtol=1E-13):
log.warning("Sanity fail in herm_fac_with_inv(): Bad left inverse! %s %s",
la.norm(xi.dot(x) - eye), sc_data)
if not sp.allclose(x.dot(x.conj().T), A, atol=1E-13, rtol=1E-13):
log.warning("Sanity fail in herm_fac_with_inv(): Bad decomp! %s %s",
la.norm(x.dot(x.conj().T) - A), sc_data)
if not sp.allclose(xi.dot(A).dot(xi.conj().T), eye, atol=1E-13, rtol=1E-13):
log.warning("Sanity fail in herm_fac_with_inv(): Bad A inverse! %s %s",
la.norm(xi.conj().T.dot(A).dot(xi) - eye), sc_data)
else:
if not sp.allclose(x.dot(xi), eye, atol=1E-13, rtol=1E-13):
log.warning("Sanity fail in herm_fac_with_inv(): Bad right inverse! %s %s",
la.norm(x.dot(xi) - eye), sc_data)
if not sp.allclose(x.conj().T.dot(x), A, atol=1E-13, rtol=1E-13):
log.warning("Sanity fail in herm_fac_with_inv(): Bad decomp! %s %s",
la.norm(x.conj().T.dot(x) - A), sc_data)
if not sp.allclose(xi.conj().T.dot(A).dot(xi), eye, atol=1E-13, rtol=1E-13):
log.warning("Sanity fail in herm_fac_with_inv(): Bad A inverse! %s %s",
la.norm(xi.conj().T.dot(A).dot(xi) - eye), sc_data)
if return_rank:
return x, xi, nonzeros
else:
return x, xi
def herm_sqrt_inv(x, zero_tol=1E-15, sanity_checks=False, return_rank=False, sc_data=''):
if isinstance(x, mm.eyemat):
x_sqrt = x
x_sqrt_i = x
rank = x.shape[0]
else:
try:
ev = x.diag #simple_diag_matrix
EV = None
except AttributeError:
ev, EV = la.eigh(x)
zeros = ev <= zero_tol #throw away negative results too!
ev_sqrt = sp.sqrt(ev)
err = sp.seterr(divide='ignore', invalid='ignore')
try:
ev_sqrt_i = 1 / ev_sqrt
ev_sqrt[zeros] = 0
ev_sqrt_i[zeros] = 0
finally:
sp.seterr(divide=err['divide'], invalid=err['invalid'])
if EV is None:
x_sqrt = mm.simple_diag_matrix(ev_sqrt, dtype=x.dtype)
x_sqrt_i = mm.simple_diag_matrix(ev_sqrt_i, dtype=x.dtype)
else:
B = mm.mmul_diag(ev_sqrt, EV.conj().T)
x_sqrt = EV.dot(B)
B = mm.mmul_diag(ev_sqrt_i, EV.conj().T)
x_sqrt_i = EV.dot(B)
rank = x.shape[0] - np.count_nonzero(zeros)
if sanity_checks:
if ev.min() < -zero_tol:
log.warning("Sanity Fail in herm_sqrt_inv(): Throwing away negative eigenvalues! %s %s",
ev.min(), sc_data)
if not np.allclose(x_sqrt.dot(x_sqrt), x):
log.warning("Sanity Fail in herm_sqrt_inv(): x_sqrt is bad! %s %s",
la.norm(x_sqrt.dot(x_sqrt) - x), sc_data)
if EV is None:
nulls = sp.zeros(x.shape[0])
nulls[zeros] = 1
nulls = sp.diag(nulls)
else: #if we did an EVD then we use the eigenvectors
nulls = EV.copy()
nulls[:, sp.invert(zeros)] = 0
nulls = nulls.dot(nulls.conj().T)
eye = np.eye(x.shape[0])
if not np.allclose(x_sqrt.dot(x_sqrt_i), eye - nulls):
log.warning("Sanity Fail in herm_sqrt_inv(): x_sqrt_i is bad! %s %s",
la.norm(x_sqrt.dot(x_sqrt_i) - eye + nulls), sc_data)
if return_rank:
return x_sqrt, x_sqrt_i, rank
else:
return x_sqrt, x_sqrt_i
def restore_RCF_l(A, lm1, Gm1, sanity_checks=False):
"""Transforms a single A[n] to obtain diagonal l[n].
Applied after restore_RCF_r(), this completes the full canonical form
of sub-section 3.1, theorem 1 of arXiv:quant-ph/0608197v2.
This function must be called for each n in turn, starting at 1,
passing the gauge transformation matrix from the previous step
as an argument.
Finds a G[n] such that orthonormalization is fulfilled for n.
The diagonal entries of l[n] are sorted in
ascending order (for example l[n] = diag([0, 0, 0.1, 0.2, ...])).
Parameters
----------
A : ndarray
The parameter tensor for the nth site A[n].
lm1 : ndarray or object with array interface
The matrix l[n - 1].
Gm1 : ndarray
The gauge transform matrix for site n obtained in the previous step (for n - 1).
sanity_checks : bool (False)
Whether to perform additional sanity checks.
Returns
-------
l : ndarray or simple_diag_matrix
The new, diagonal matrix l[n].
G : ndarray
The gauge transformation matrix for site n.
G_i : ndarray
Inverse of G.
"""
if Gm1 is None:
x = lm1
else:
x = Gm1.conj().T.dot(lm1.dot(Gm1))
M = eps_l_noop(x, A, A)
ev, EV = la.eigh(M) #wraps lapack routines, which return eigenvalues in ascending order
if sanity_checks:
assert np.all(ev == np.sort(ev)), "unexpected eigenvalue ordering"
l = mm.simple_diag_matrix(ev, dtype=A.dtype)
G_i = EV
if Gm1 is None:
Gm1 = EV.conj().T #for left uniform case
lm1 = l #for sanity check
for s in xrange(A.shape[0]):
A[s] = Gm1.dot(A[s].dot(G_i))
if sanity_checks:
l_ = eps_l_noop(lm1, A, A)
if not sp.allclose(l_, l, atol=1E-12, rtol=1E-12):
log.warning("Sanity Fail in restore_RCF_l!: l is bad!")
log.warning(la.norm(l_ - l))
G = EV.conj().T
if sanity_checks:
eye = sp.eye(A.shape[2])
if not sp.allclose(sp.dot(G, G_i), eye,
atol=1E-12, rtol=1E-12):
log.warning("Sanity Fail in restore_RCF_l!: Bad GT! (off by %g)", la.norm(sp.dot(G, G_i) - eye))
return l, G, G_i
