def reduceUBM(ubm, nh_factor=1, nsteps=300, nsweeps=200, learning_rate=1e-2):
# create and initialize rbm wave function
rbm = nqs.brbm_nqs(ubm.nv, int(ubm.nh*nh_factor))
rbm.setRandomParams()
rbm.a = np.real(ubm.a) + 0j
rbm.b[:ubm.nh] = np.real(ubm.b) + 0j
rbm.W[:ubm.nh,:ubm.nv] = np.real(ubm.W) + 0j
rbm.Y[:ubm.nv,:ubm.nv] = np.real(ubm.Y) + 0j
print(np.real(ubm.Y))
# optimize with real parameters to adjust the absolute value
# of each coefficient in the ubm
S = sampler.metropolis(wf=rbm)
O = optimize.fidelity_GA(S, ubm)
O.opt(nsteps,nsweeps,learning_rate)
# now I have to take care of the phases of the coefficients
# a, Y = optimize.opt_phases(ubm.logPsi_dict, ubm.nv, 1000, 1e-2)
# create a new ubm
new = rbm.asUBM()
# new.Y = 1j*Y
# new.a += 1j*a
return new, O.fidelity[-1]
def det_reduceUBM(ubm, nsteps=60, learning_rate=.1):
# create and initialize rbm wave function
rbm = nqs.brbm_nqs(ubm.nv, ubm.nh)
rbm.setRandomParams()
rbm.a = ubm.a
rbm.b = ubm.b
rbm.W = ubm.W
rbm.Y = ubm.Y
# rbm.a = np.real(ubm.a) + 0j
# rbm.b[:ubm.nh] = np.real(ubm.b) + 0j
# rbm.W[:ubm.nh,:ubm.nv] = np.real(ubm.W) + 0j
# rbm.Y[:ubm.nv,:ubm.nv] = np.real(ubm.Y) + 0j
# optimization
O = optimize.det_fidelity_GA(rbm, ubm)
#O = optimize.det_rel_entropy_GD(rbm, ubm)
O.opt(nsteps,learning_rate)
new = rbm.asUBM()
# now I have to take care of the phases of the coefficients
# a, Y = optimize.opt_phases(ubm.logPsi_dict, ubm.nv, 5000, 1e-2,1)
#
# new.Y += 1j*Y
# new.a += 1j*a
# return new, O.rel_entropy[-1], a, Y, ubm
return new, O.fidelity[-1]
def run_inf_trotter(N, h, st, dt, init_wf=None):
trotter_st = st
trotter_dt = dt
decay_rate = 1
mod = models.TransIsing1D(N, h, pbc=True)
if init_wf == None:
wf = nqs.brbm_nqs(N, N)
wf.W[:N, :] = .1 * np.eye(N, dtype=complex)
else:
wf = init_wf
meanS = []
stdS = []
print("Applying Trotter Evolution...")
for k in range(trotter_st):
print(k)
if True: #k%5==0:#True:#k==trotter_st-1:
print("Measuring SigmaX...")
M = {'E': 0, 'E2': 0}
S = sampler.metropolis(wf=wf)
S.afterSweep = lambda S: measure_sigmax(M, mod, S)
S.run(nsweeps=5000, output=False)
print(M['E'], np.sqrt(M['E2'] - M['E']**2))
meanS.append(M['E'][0, 0])
stdS.append(np.sqrt(M['E2'] - M['E']**2)[0, 0])
tao = trotter_dt * (decay_rate**(k / trotter_st))
mod.applyTrotterStep(wf, tao, inf=True)
return mod, wf, meanS, stdS
def run_inf_trotter(N, h, st, dt, init_wf=None):
trotter_st = st
trotter_dt = dt
decay_rate = 1
mod = models.TransIsing1D(N, h, pbc=True)
if init_wf == None:
wf = nqs.brbm_nqs(N, N)
#wf.W = .1*np.random.rand(N,N).astype(complex)
wf.W[:N, :] = .1 * np.eye(N, dtype=complex)
# for k in range(N):
# wf.W[N+k,k] = .2
# for k in range(N-1):
# wf.W[N+k,k+1] = .2
else:
wf = init_wf
meanE = []
stdE = []
print("Applying Trotter Evolution...")
for k in range(trotter_st):
# print('a', wf.nh)
# wf.rmVisibleInt()
# print('b', wf.nh)
# wf.rmHiddenNodes(keep=4*N)
print(k)
if k % 5 == 0: #True:#k==trotter_st-1:
print("Measuring Energy...")
M = {'E': 0, 'E2': 0, 'C': 0}
S = sampler.metropolis(wf=wf)
S.afterSweep = lambda S: measure(M, mod, S)
S.run(nsweeps=1000, output=False)
print(M['E'], np.sqrt(M['E2'] - M['E']**2))
meanE.append(M['E'][0, 0])
stdE.append(np.sqrt(M['E2'] - M['E']**2)[0, 0])
tao = trotter_dt * (decay_rate**(k / trotter_st))
mod.applyTrotterStep(wf, -1j * tao, inf=True)
return mod, wf, meanE, stdE
def run_inf_trotter2(N, h, st, dt, init_wf=None):
trotter_st = st
trotter_dt = dt
decay_rate = 1
mod = models.TransIsing1D(N, h, pbc=True)
if init_wf == None:
wf = nqs.brbm_nqs(N, 0)
else:
wf = init_wf
meanE = []
stdE = []
print("Applying Trotter Evolution...")
for k in range(trotter_st):
# wf.rmHiddenNodes(keep=4*N)
print(k)
if True: #k==trotter_st-1:
print("Measuring Energy...")
M = {'E': 0, 'E2': 0, 'C': 0}
S = sampler.metropolis(wf=wf)
S.afterSweep = lambda S: measure(M, mod, S)
S.run(nsweeps=1000, output=False)
print(M['E'], np.sqrt(M['E2'] - M['E']**2))
meanE.append(M['E'][0, 0])
stdE.append(np.sqrt(M['E2'] - M['E']**2)[0, 0])
tao = trotter_dt * (decay_rate**(k / trotter_st))
mod.applyTrotterStep(wf, -1j * tao, inf=True, hidden_crz=True)
return mod, wf, meanE, stdE
import numpy as np import nqs import models import sampler import optimize #wf = nqs.ti_brbm_nqs(nv=10,nf=1) wf = nqs.brbm_nqs(nv=10, nh=10) wf.setRandomParams() TI = models.TransIsing1D(nspins=10, hfield=.1) S = sampler.metropolis(wf=wf) O = optimize.energy_GD(S, TI)