def moveL(self):
# read As, Bs, Cl, Cr
hf = h5py.File(self.dataname, 'r')
As = df.readarr(hf, 'data/As_' + str(self.ix - 1))
Cl = df.readarr(hf, 'data/Cl_' + str(self.ix - 1))
Cr = df.readarr(hf, 'data/Cr_' + str(indexR(self.L, self.ix + 1)))
hf.close()
# energy of the present state
newE = mpsE(Cl, Cr, self.Ms, self.M)
self.adjust_alpha(newE)
self.E = newE
# construt single-site H and solve
E, Ms = singlesite_hamilt(Cl, Cr, self.M, self.Ms)
self.reducedE = self.E - E
# subspace expansion
P = self.alpha * obtainPL(Cr, Ms, self.M)
Ms_tilde = np.append(Ms, P, axis=1)
zeros = np.zeros((2, As.shape[1], P.shape[1]))
As_tilde = np.append(As, zeros, axis=2)
# truncation
Ms_tilde = np.swapaxes(Ms_tilde, 0, 1)
Ms_tilde = np.reshape(
Ms_tilde,
(Ms_tilde.shape[0], Ms_tilde.shape[1] * Ms_tilde.shape[2]))
Ms_tilde_dag = Ms_tilde.T.conj()
q, r = np.linalg.qr(Ms_tilde_dag)
q, r = q.T.conj(), r.T.conj()
Dl = self.Ms.shape[1]
if q.shape[0] > Dl:
q = np.compress([1] * Dl, q, axis=0)
r = np.compress([1] * Dl, r, axis=1)
# new Bs, Cr and new Ms
newBs = np.reshape(q, (q.shape[0], 2, q.shape[1] // 2))
newBs = np.swapaxes(newBs, 0, 1)
newCr = contr.contractR(Cr, newBs, self.M)
self.Ms = np.einsum('abc,cd->abd', As_tilde, r)
# update datafile
hf = h5py.File(self.dataname, 'r+')
df.savearr(hf, 'data/Bs_' + str(indexR(self.L, self.ix)), newBs)
df.savearr(hf, 'data/Cr_' + str(indexR(self.L, self.ix)), newCr)
self.ix -= 1
del hf['data/As_' + str(self.ix)]
del hf['data/Cl_' + str(self.ix)]
hf.close()
def moveR(self):
# read As, Bs, Cl, Cr
hf = h5py.File(self.dataname, 'r')
Bs = df.readarr(hf, 'data/Bs_' + str(indexR(self.L, self.ix + 1)))
Cl = df.readarr(hf, 'data/Cl_' + str(self.ix - 1))
Cr = df.readarr(hf, 'data/Cr_' + str(indexR(self.L, self.ix + 1)))
hf.close()
# energy of the present state
self.E = mpsE(Cl, Cr, self.Ms, self.M)
self.adjust_alpha(self.E)
# construt single-site H and solve
E, Ms = singlesite_hamilt(Cl, Cr, self.M, self.Ms)
self.reducedE = self.E - E
# subspace expansion
P = self.alpha * obtainPR(Cl, Ms, self.M)
Ms_tilde = np.append(Ms, P, axis=2)
zeros = np.zeros((2, P.shape[2], Bs.shape[2]))
Bs_tilde = np.append(Bs, zeros, axis=1)
# truncation
Ms_tilde = np.reshape(
Ms_tilde,
(Ms_tilde.shape[0] * Ms_tilde.shape[1], Ms_tilde.shape[2]))
q, r = np.linalg.qr(Ms_tilde)
Dr = self.Ms.shape[2]
if q.shape[1] > Dr:
q = np.compress([1] * Dr, q, axis=1)
r = np.compress([1] * Dr, r, axis=0)
# new As,Cl and new Ms
newAs = np.reshape(q, (2, q.shape[0] // 2, q.shape[1]))
newCl = contr.contractL(Cl, newAs, self.M)
self.Ms = np.einsum('bc,acd->abd', r, Bs_tilde)
# update datafile
hf = h5py.File(self.dataname, 'r+')
df.savearr(hf, 'data/As_' + str(self.ix), newAs)
df.savearr(hf, 'data/Cl_' + str(self.ix), newCl)
self.ix += 1
del hf['data/Bs_' + str(indexR(self.L, self.ix))]
del hf['data/Cr_' + str(indexR(self.L, self.ix))]
hf.close()
def __init__(self, L, D):
self.L = L
self.D = D
self.dataname = 'data/xy_dmrg_D' + str(self.D) + '.h5'
self.M = mpo.xy_mpo()
self.alpha = 0.1
self.E = 9.99e9
self.reducedE = 9.99e9
idmrgsol = idmrg.idmrg(self.D, self.L)
idmrgsol.grow_to_target()
self.ix = L // 2
hf = h5py.File(self.dataname, 'r+')
Ms = df.readarr(hf, 'data/Bs_' + str(indexR(self.L, self.ix)))
self.Ms = np.einsum('bc,acd->abd', np.diag(hf['data/sigma']), Ms)
df.savearr(hf, 'data/Cl_-1', np.reshape([1, 0, 0, 0], (1, 1, 4, 1)))
df.savearr(hf, 'data/Cr_-1', np.reshape([0, 0, 0, 1], (1, 4, 1, 1)))
del hf['data/Bs_' + str(indexR(self.L, self.ix))]
del hf['data/Cr_' + str(indexR(self.L, self.ix))]
hf.close()
def __init__(self, bondD, targetL):
self.bondD = bondD
self.M = mpo.xy_mpo()
self.targetL = targetL
if self.targetL % 2 != 0:
warnings.warn('idmrg: L must be even. L modified to L+1')
self.targetL += 1
self.dataname = 'data/xy_dmrg_D'+str(self.bondD)+'.h5'
initL = 2*int(np.log2(self.bondD)) + 2
self.currL = initL
H0 = ed.xy_hamilt(initL)
psi0 = ed.ground_state(H0)[1]
A_list, B_list, sigma = canonical(psi0, initL)
# truncate
A_list[initL//2 - 1] = np.compress([1]*self.bondD, A_list[initL//2 - 1], axis=2)
B_list[initL//2 - 1] = np.compress([1]*self.bondD, B_list[initL//2 - 1], axis=1)
sigma = np.compress([1]*self.bondD, sigma)
if not contr.check_Lnorm(A_list) or not contr.check_Rnorm(B_list):
warnings.warn('idrmg: normalization condition violated due to truncation')
self.As, self.Bs = A_list[initL//2 - 1], B_list[initL//2 - 1]
self.sigma = sigma
hf = h5py.File(self.dataname, 'w')
hf['parameters/bondD']=self.bondD
hf['parameters/currL']=self.currL
self.Cl = np.reshape([1,0,0,0], (1,1,4,1))
self.Cr = np.reshape([0,0,0,1], (1,4,1,1))
for ix in range(initL//2):
self.Cl = contr.contractL(self.Cl, A_list[ix], self.M)
self.Cr = contr.contractR(self.Cr, B_list[ix], self.M)
# save As, Bs, Cl, Cr
df.savearr(hf, 'data/As_'+str(ix), A_list[ix])
df.savearr(hf, 'data/Cl_'+str(ix), self.Cl)
df.savearr(hf, 'data/Bs_'+str(ix), B_list[ix])
df.savearr(hf, 'data/Cr_'+str(ix), self.Cr)
hf['data/sigma'] = self.sigma
hf.close()
def growby2(self):
newAs, newBs, initpsi = sadd.init_newpsi(self.sigma, self.As, self.Bs)
newCl = contr.contractL(self.Cl, newAs, self.M)
newCr = contr.contractR(self.Cr, newBs, self.M)
H2 = np.einsum('abde,cdfg->abcefg', newCl, newCr)
dimH2 = newCl.shape[1]**2
H2 = np.reshape(H2, (dimH2, dimH2))
H2s = scipy.sparse.csr_matrix(H2)
if not np.allclose(H2, H2.T.conj()):
warnings.warn(str(self.currL)+': H2 not Hermitian')
initpsi = np.reshape(initpsi, (dimH2, 1))
newpsi = scipy.sparse.linalg.eigsh(H2s, k=1, which='SA', v0=initpsi)[1]
newpsi = np.reshape(newpsi, (newCl.shape[1], newCl.shape[1]))
u, s, vdag = np.linalg.svd(newpsi)
# update
self.currL += 2
self.As = np.einsum('abc,cd->abd', newAs, u)
self.Bs = np.einsum('ac,bcd->bad', vdag, newBs)
self.sigma = s
self.Cl = np.einsum('abcd,be->aecd', newCl, u.conj())
self.Cl = np.einsum('abcd,de->abce', self.Cl, u)
self.Cr = np.einsum('ab,bcde->acde', vdag.conj(), newCr)
self.Cr = np.einsum('ad,bcde->bcae', vdag, self.Cr)
# save data
hf = h5py.File(self.dataname, 'r+')
ix = self.currL // 2 - 1
df.savearr(hf, 'data/As_'+str(ix), self.As)
df.savearr(hf, 'data/Cl_'+str(ix), self.Cl)
df.savearr(hf, 'data/Bs_'+str(ix), self.Bs)
df.savearr(hf, 'data/Cr_'+str(ix), self.Cr)
del hf['parameters/currL']
hf['parameters/currL'] = self.currL
del hf['data/sigma']
hf['data/sigma'] = self.sigma
hf.close()