Skip to content

GitLab

Commit 6a811870 authored by Jacob August Davison's avatar Jacob August Davison
Browse files

refactored to make computations exclusively with Node objects

parent 748fe497
Expand all Hide whitespace changes
Inline Side-by-side
main.py
View file @ 6a811870
......@@ -10,16 +10,18 @@ from scipy.integrate import odeint, ode
import numpy as np
import time
import pickle
import tensorflow as tf
#import tensorflow as tf
#tf.enable_v2_behavior()
print("GPU available: ",tf.test.is_gpu_available())
#print("GPU available: ",tf.test.is_gpu_available())
import tracemalloc
import os, sys
#from memory_profiler import profile
import itertools
import random
sess = tf.compat.v1.Session(config=tf.compat.v1.ConfigProto(log_device_placement=True))
sess.close()
#import tensornetwork as tn
#tn.set_default_backend("tensorflow")
#sess = tf.compat.v1.Session(config=tf.compat.v1.ConfigProto(log_device_placement=True))
#sess.close()
# user files
# sys.path.append('C:\\Users\\davison\\Research\\exact_diagonalization\\')
......@@ -58,12 +60,25 @@ def derivative(t, y, hamiltonian, occ_tensors, generator, flow):
E, f, G = ravel(y, hamiltonian.n_sp_states)
generator.f = f
generator.G = G
# timefi = time.time()
generator.f = tn.Node(f)
# timeff = time.time()
# timegi = time.time()
generator.G = tn.Node(G)
# timegf = time.time()
# timefli = time.time()
dE, df, dG = flow.flow(generator)
# timeflf = time.time()
# print("""
# f time: {:2.4f} s
# G time: {:2.4f} s
# Flow time: {:2.4f} s
# """.format(timeff-timefi, timegf-timegi, timeflf-timefli))
dy = unravel(dE, df, dG)
dy = unravel(dE.tensor, df.tensor, dG.tensor)
return dy
......@@ -105,6 +120,7 @@ def ravel(y, bas_len):
ravel_f = np.reshape(y[1:bas_len**2+1], (bas_len, bas_len))
ravel_G = np.reshape(y[bas_len**2+1:bas_len**2+1+bas_len**4],
(bas_len, bas_len, bas_len, bas_len))
return(ravel_E, ravel_f, ravel_G)
......@@ -119,7 +135,7 @@ def main(n_holes, n_particles, ref=None, d=1.0, g=0.5, pb=0.0):
if ref == None:
ha = PairingHamiltonian2B(n_holes, n_particles, d=d, g=g, pb=pb)
ref = [1,1,1,1,0,0,0,0] # this is just for printing
ref = ha.reference # this is just for printing
else:
ha = PairingHamiltonian2B(n_holes, n_particles, ref=ref, d=d, g=g, pb=pb)
......@@ -146,8 +162,8 @@ def main(n_holes, n_particles, ref=None, d=1.0, g=0.5, pb=0.0):
flowi = time.time()
# --- Solve the IM-SRG flow
y0 = unravel(ha.E, ha.f, ha.G)
y0 = unravel(ha.E.tensor, ha.f.tensor, ha.G.tensor)
#print(y0)
solver = ode(derivative,jac=None)
solver.set_integrator('vode', method='bdf', order=5, nsteps=500)
solver.set_f_params(ha, ot, wg, fl)
......@@ -216,7 +232,7 @@ if __name__ == '__main__':
# with open('occE_nonzero.txt', 'a') as fi:
# fi.write('%s %s %s %s %s %s -- %s\n' % (a,b,c,d,e,f,val))
test_exact('plots_exact_2b/', main)
#test_exact('plots_exact_2b/', main)
main(4,4)
......
main3b.py
View file @ 6a811870
......@@ -15,7 +15,7 @@ import os, sys
#from memory_profiler import profile
import itertools
import random
import tensornetwork
import tensornetwork as tn
# user files
# sys.path.append('C:\\Users\\davison\\Research\\exact_diagonalization\\')
......@@ -258,7 +258,8 @@ def test_exact(plots_dir):
if __name__ == '__main__':
test_exact("plots3b\\")
#test_exact("plots3b/")
main3b(4,4)
#test = main3b(4,4)
#tracemalloc.start()
......
oop_imsrg/flow.py
View file @ 6a811870
This diff is collapsed.
oop_imsrg/generator.py
View file @ 6a811870
This diff is collapsed.
oop_imsrg/hamiltonian.py
View file @ 6a811870
......@@ -2,7 +2,8 @@ import numpy as np
#import tensorflow as tf
# tf.enable_v2_behavior()
import tensornetwork as tn
tn.set_default_backend("tensorflow")
tn.set_default_backend("numpy")
class Hamiltonian(object):
"""Parent class for organization purposes. Ideally, all Hamiltonian
classes should inherit from this class. In this way, AssertionErrors
......@@ -238,8 +239,8 @@ class PairingHamiltonian2B(Hamiltonian):
G) -- two-body piece"""
bas1B = self.sp_basis # get the single particle basis
H1B_t = self.H1B # get the 1B tensor
H2B_t = self.H2B # get the 2B tensor
H1B_t = self.H1B.astype(np.float32) # get the 1B tensor
H2B_t = self.H2B.astype(np.float32) # get the 2B tensor
holes = self.holes # get hole states
particles = self.particles # get particle states
......@@ -248,246 +249,44 @@ class PairingHamiltonian2B(Hamiltonian):
# - Calculate 0B piece
H1B_holes = H1B_t[np.ix_(holes,holes)]
H2B_holes = H2B_t[np.ix_(holes,holes,holes,holes)]
ob_node0b = tn.Node(H1B_holes)
tb_node0b = tn.Node(H2B_holes)
ob_ii = tn.connect(ob_node0b[0],ob_node0b[1])
tb_ijij1 = tn.connect(tb_node0b[0], tb_node0b[2])
tb_ijij2 = tn.connect(tb_node0b[1], tb_node0b[3])
ob_node0b[0] ^ ob_node0b[1]
ob_contract = ob_node0b @ ob_node0b # optimized contraction
tb_node0b[0] ^ tb_node0b[2]
tb_node0b[1] ^ tb_node0b[3]
tb_contract = tb_node0b @ tb_node0b
# ob_ii = tn.connect(ob_node0b[0],ob_node0b[1])
# tb_ijij1 = tn.connect(tb_node0b[0], tb_node0b[2])
# tb_ijij2 = tn.connect(tb_node0b[1], tb_node0b[3])
flatten = tn.flatten_edges([tb_ijij1, tb_ijij2])
ob_contract = tn.contract(ob_ii).tensor.numpy()
tb_contract = 0.5*tn.contract(flatten).tensor.numpy()
# flatten = tn.flatten_edges([tb_ijij1, tb_ijij2])
# ob_contract = tn.contract(ob_ii).tensor#.tensor.numpy()
# tb_contract = 0.5*tn.contract(flatten).tensor#.tensor.numpy()
E = ob_contract + tb_contract
E = E.astype(np.float32)
E = ob_contract + tn.Node(0.5)*tb_contract
#E = E.astype(np.float32)
# - Calculate 1B piece
ob_node1b = tn.Node(H1B_t)
tb_node1b = tn.Node(H2B_t[np.ix_(bas1B,holes,bas1B,holes)])
tb_ihjh = tn.connect(tb_node1b[1], tb_node1b[3])
tb_contract = tn.contract(tb_ihjh)
tb_node1b[1] ^ tb_node1b[3]
tb_contract = tb_node1b @ tb_node1b
f = ob_node1b.tensor.numpy() + tb_contract.tensor.numpy()
f = f.astype(np.float32)
G = H2B_t
# tb_ihjh = tn.connect(tb_node1b[1], tb_node1b[3])
# tb_contract = tn.contract(tb_ihjh)
#f = ob_node1b.tensor.numpy() + tb_contract.tensor.numpy()
f = (ob_node1b + tb_contract)
#f = f.astype(np.float32)
# - Calculate 2B piece
G = tn.Node(H2B_t)
return (E, f, G)
# class PairingHamiltonian3B(PairingHamiltonian2B):
#
# def __init__(self, n_hole_states, n_particle_states, d=1.0, g=0.5, pb=0.0):
# """Class constructor. Instantiate PairingHamiltonian3B object.
#
# Arguments:
#
# n_hole_states -- number of holes states in the single particle basis
# n_particle_states -- number of particles states in the single particle basis
#
# Keyword arguments:
#
# d -- the energy level spacing (default: 1.0)
# g -- the pairing strength (default: 0.5)
# pb -- strength of the pair-breaking term (operates in double particle basis) (default: 0.0)"""
#
# self._d = d
# self._g = g
# self._pb = pb
# self._reference = np.append(np.ones(n_hole_states), np.zeros(n_particle_states))
# self._holes = np.arange(n_hole_states, dtype=np.int64)
#
# self._n_sp_states = n_hole_states + n_particle_states
# self._particles = np.arange(n_hole_states,self.n_sp_states, dtype=np.int64)
# self._sp_basis = np.append(self.holes, self.particles)
#
# self._H1B, self._H2B, self._H3B = self.construct()
# self._E, self._f, self._G = self.normal_order()
#
# @property
# def H3B(self):
# """Returns:
#
# H2B -- three-body (rank 6) tensor defined by construct()."""
# return self._H3B
#
# def delta2B(self, p,q,r,s):
# """Inherited from parent."""
# return super().delta2B(p,q,r,s)
#
# def deltaPB(self, p,q,r,s):
# """Inherited from parent."""
# return super().deltaPB(p,q,r,s)
#
# def delta3B(self, p,q,r,s,t,u):
# """Determines if a two-body tensor elements should be zero,
# positive, or negative. This behavior is dictated by the three-
# body excitation term in pairing Hamiltonian.
#
# Arguments:
#
# p,q,r,s,t,u -- indices in single-particle basis"""
#
# pp = np.floor_divide(p,2)
# qp = np.floor_divide(q,2)
# rp = np.floor_divide(r,2)
# sp = np.floor_divide(s,2)
# tp = np.floor_divide(t,2)
# up = np.floor_divide(u,2)
#
# ps = 1 if p%2==0 else -1
# qs = 1 if q%2==0 else -1
# rs = 1 if r%2==0 else -1
# ss = 1 if s%2==0 else -1
# ts = 1 if t%2==0 else -1
# us = 1 if u%2==0 else -1
#
# lhs_p = np.array([pp, qp, rp])
# lhs_s = np.array([ps, qs, rs])
# rhs_p = np.array([sp, tp, up])
# rhs_s = np.array([ss, ts, us])
#
# p_equiv = np.equal(lhs_p, rhs_p)
# s_equiv = np.equal(lhs_s, rhs_s)
# p_where = np.where(p_equiv==True)[0]
# s_where = np.where(s_equiv==True)[0]
# # print(p_where)
# result = 0
# result_str ='0'
# if len(p_where) == 2 and (lhs_p[p_where[0]] == lhs_p[p_where[1]] and rhs_p[p_where[0]] == rhs_p[p_where[1]]):
# # print("in")
# result = 1
#
# p_where_n = np.where(p_equiv==False)[0]
# l_equiv = np.equal(lhs_p[p_where_n], lhs_p[p_where])
# r_equiv = np.equal(rhs_p[p_where_n], rhs_p[p_where])
# # print(l_equiv, r_equiv)
# if(l_equiv.any() or r_equiv.any()):
# result *= 0
#
# l_spins = lhs_s[p_where]
# r_spins = rhs_s[p_where]
# # print(l_spins, r_spins)
#
# if l_spins[0] == l_spins[1] or r_spins[0] == r_spins[1]:
# result *= 0
# result_str += '0'
# if l_spins[0] == r_spins[0] and l_spins[1] == r_spins[1]:
# result *= 1
# result_str += '1'
# if l_spins[0] == r_spins[1] and l_spins[1] == r_spins[0]:
# result *= -1
# result_str += '-1'
#
# if lhs_p[p_where_n] != rhs_p[p_where_n] and lhs_s[p_where_n] != rhs_p[p_where_n]:
# result *= 1
# result_str += '1'
# else:
# result *= 0
# result_str += '0'
#
# return result
#
# def construct(self):
# """Constructs the one- and two-body pieces of the pairing
# Hamiltonian.
#
# Returns:
#
# (H1B, -- one-body tensor elements (defined by one-body operator)
# H2B) -- two-body tensor elements (defined by two-body operator)"""
#
# bas1B = self.sp_basis # get the single particle basis
#
# # one body part of Hamiltonian is floor-division of basis index
# # matrix elements are (P-1) where P is energy level
# H1B = np.diag(self.d*np.floor_divide(bas1B,2))
#
# # two body part of Hamiltonian constructed from four indices
# # with non-zero elements defined by pairing term
# H2B = np.zeros(np.ones(4, dtype=np.int64)*self.n_sp_states)
# for p in bas1B:
# for q in bas1B:
# for r in bas1B:
# for s in bas1B:
# H2B[p,q,r,s] += self.g*0.5*self.delta2B(p,q,r,s)
# H2B[p,q,r,s] += self.pb*0.5*self.deltaPB(p,q,r,s)
#
# W=1
# H3B = np.zeros(np.ones(6, dtype=np.int64)*self.n_sp_states)
# for p in bas1B:
# for q in bas1B:
# for r in bas1B:
# for s in bas1B:
# for t in bas1B:
# for u in bas1B:
# H3B[p,q,r,s,t,u] += W*(1./6.)*self.delta3B(p,q,r,s,t,u)
# # print(H3B[0,1,2,3,4,5])
# return (H1B, H2B, H3B)
# # if lhs_p[p_where_n] != rhs_p[p_where_n]:
# # if np.array_equal(s_where, [])
#
# def normal_order(self):
# """Normal-orders the pairing Hamiltonian.
#
# Returns:
#
# (E, -- zero-body piece
# f, -- one-body piece
# G) -- two-body piece"""
#
# bas1B = self.sp_basis # get the single particle basis
# H1B_t = self.H1B # get the 1B tensor
# H2B_t = self.H2B # get the 2B tensor
# H3B_t = self.H3B
# holes = self.holes # get hole states
# particles = self.particles # get particle states
#
# net = TensorNetwork()
#
# # - Calculate 0B piece
# H1B_holes = H1B_t[np.ix_(holes,holes)]
# H2B_holes = H2B_t[np.ix_(holes,holes,holes,holes)]
# H3B_holes = H3B_t[np.ix_(holes,holes,holes,holes,holes,holes)]
#
# ob_node0b = net.add_node(H1B_holes)
# tb_node0b = net.add_node(H2B_holes)
#
# ob_ii = net.connect(ob_node0b[0],ob_node0b[1])
# tb_ijij1 = net.connect(tb_node0b[0], tb_node0b[2])
# tb_ijij2 = net.connect(tb_node0b[1], tb_node0b[3])
#
# flatten = net.flatten_edges([tb_ijij1, tb_ijij2])
# ob_contract = net.contract(ob_ii).tensor.numpy()
# tb_contract = 0.5*net.contract(flatten).tensor.numpy()
#
# E = ob_contract + tb_contract
#
#
# # - Calculate 1B piece
# ob_node1b = net.add_node(H1B_t)
# tb_node1b = net.add_node(H2B_t[np.ix_(bas1B,holes,bas1B,holes)])
#
# tb_ihjh = net.connect(tb_node1b[1], tb_node1b[3])
# tb_contract = net.contract(tb_ihjh)
#
# f = ob_node1b.tensor.numpy() + tb_contract.tensor.numpy()
#
# G = H2B_t
#
# del net
#
# return (E, f, G)
#
# # test = PairingHamiltonian3B(4,4)
# # bas1B = test.sp_basis
# # for p in bas1B:
# # for q in bas1B:
# # for r in bas1B:
# # for s in bas1B:
# # for t in bas1B:
# # for u in bas1B:
# # if test.H3B[p,q,r,s,t,u] != 0:
# # print('{:0.4f} | {:d} {:d} {:d} {:d} {:d} {:d}'.format(test.H3B[p,q,r,s,t,u], p,q,r,s,t,u))
return (E, f, G)
oop_imsrg/occupation_tensors.py
View file @ 6a811870
import numpy as np
import os
from oop_imsrg.hamiltonian import *
import sys
import numba
class OccupationTensors(object):
"""Functions as a container for important occupation tensors
......@@ -31,6 +35,11 @@ class OccupationTensors(object):
self._occI = self.__get_occI()
self._occJ = self.__get_occJ()
if not os.path.exists("occ_storage/"):
os.mkdir("occ_storage/")
@property
def occA(self):
"""Returns:
......@@ -112,8 +121,9 @@ class OccupationTensors(object):
return self._occJ
# ---- BUILD OCCUPATION TENSORS ---
# ---- BUILD OCCUPATION TENSORS ---
#@jit#(nopython=True)
def __get_occA(self, flag=0):
"""Builds the occupation tensor occA.
......@@ -130,11 +140,48 @@ class OccupationTensors(object):
n = len(bas1B)
if flag == 0: # default
occA = np.zeros((n,n,n,n),dtype=np.float32)
for a in bas1B:
for b in bas1B:
occA[a,b,a,b] = ref[a] - ref[b]
# occA = np.zeros((n,n,n,n),dtype=np.float32)
# for a in bas1B:
# for b in bas1B:
# occA[a,b,a,b] = ref[a] - ref[b]
# print(sys.getsizeof(occA)/10**6)
# TENSOR TRAIN DECOMPOSITION
# We find a TT-decomposition by hand, because the the tensor we want
# to decompose is small enough to do so. This is an exact decomposition
# of the rank 2 tensor described by n_a-n_b.
#Ga = tn.Node(np.array([ [1,1], [1,1], [1,1], [1,1], [0,1], [0,1], [0,1], [0,1] ]))
#Gb = tn.Node(np.transpose(np.array([ [1,-1], [1,-1], [1,-1], [1,-1], [1,0], [1,0], [1,0], [1,0] ])))
Ga = tn.Node(np.append(ref[:,np.newaxis], np.ones((n,1)),axis=1).astype(int))
Gb = tn.Node(np.transpose(np.append(np.ones((n,1)), -1*ref[:,np.newaxis],axis=1).astype(int)))
Gab = tn.ncon([Ga,Gb], [(-1,1),(1,-2)])
final = tn.outer_product(Gab, tn.Node(np.ones((n,n)))).tensor
# PARALLELIZE NESTED LOOPS FOR BETTER PERFORMANCE
@numba.jit(nopython=True)
def enforce_delta(n, tensor):
bas1B = range(n)
for a in bas1B:
for b in bas1B:
for c in bas1B:
for d in bas1B:
if not(a == c and b == d):
tensor[a,b,c,d] = 0
return tensor
occA = tn.Node(enforce_delta(n, final))
# print(sys.getsizeof(occA)/10**6)
if flag == 1:
occA = np.zeros((n,n),dtype=np.float32)
......@@ -161,11 +208,41 @@ class OccupationTensors(object):
n = len(bas1B)
if flag == 0: # default
occB = np.zeros((n,n,n,n),dtype=np.float32)
# occB = np.zeros((n,n,n,n),dtype=np.float32)
for a in bas1B:
for b in bas1B:
occB[a,b,a,b] = 1 - ref[a] - ref[b]
# for a in bas1B:
# for b in bas1B:
# occB[a,b,a,b] = 1 - ref[a] - ref[b]
#occB = tn.Node(1 - self.occA.tensor)
# TENSOR TRAIN DECOMPOSITION of rank 4 tensor with elements given
# by 1 - n_a - n_b.
#Ga = tn.Node(np.array([ [0,1], [0,1], [0,1], [0,1], [1,1], [1,1], [1,1], [1,1] ]))
#Gb = tn.Node(np.transpose(np.array([ [1,-1], [1,-1], [1,-1], [1,-1], [1,0], [1,0], [1,0], [1,0] ])))
Ga = tn.Node(np.append(1-ref[:,np.newaxis], np.ones((n,1)),axis=1).astype(int))
Gb = tn.Node(np.transpose(np.append(np.ones((n,1)), -1*ref[:,np.newaxis],axis=1).astype(int)))
Gab = tn.ncon([Ga,Gb], [(-1,1),(1,-2)])
final = tn.outer_product(Gab, tn.Node(np.ones((n,n)))).tensor
# PARALLELIZE NESTED LOOPS FOR BETTER PERFORMANCE
@numba.jit(nopython=True)
def enforce_delta(n, tensor):
bas1B = range(n)
for a in bas1B:
for b in bas1B:
for c in bas1B:
for d in bas1B:
if not(a == c and b == d):
tensor[a,b,c,d] = 0
return tensor
occB = tn.Node(enforce_delta(n, final))
if flag == 1:
occB = np.zeros((n,n),dtype=np.float32)
......@@ -191,27 +268,66 @@ class OccupationTensors(object):
ref = self._reference
n = len(bas1B)
# occC1 = np.einsum('i,j,k->ijk', ref,ref,(1-ref))
# occC2 = np.einsum('i,j,k->ijk', (1-ref),(1-ref),ref)
# occC3 = occC1 + occC2
# if flag == 0: # default
# occC = np.einsum('ijk,lmn->ijklmn', occC3, occC3)
# return occC
# if flag == 1:
# occC = occC3
# return occC
if flag == 0: # default
occC = np.zeros((n,n,n,n,n,n),dtype=np.float32)
for a in bas1B:
for b in bas1B:
for c in bas1B:
occC[a,b,c,a,b,c] = ref[a]*ref[b]*(1-ref[c]) + \
(1-ref[a])*(1-ref[b])*ref[c]
# occC = np.zeros((n,n,n,n,n,n),dtype=np.float32)
# for a in bas1B:
# for b in bas1B:
# for c in bas1B:
# occC[a,b,c,a,b,c] = ref[a]*ref[b]*(1-ref[c]) + \
# (1-ref[a])*(1-ref[b])*ref[c]
# print(sys.getsizeof(occC)/10**6)
# TENSOR TRAIN DECOMPOSITION of rank 6 tensor with elements
# given by n_a*n_b + (1 - n_b - n_a)*n_c.
#Ga = tn.Node(np.array([ [1,1,1,1], [1,1,1,1], [1,1,1,1], [1,1,1,1], [0,1,1,0], [0,1,1,0], [0,1,1,0], [0,1,1,0] ]))
# Gb = tn.Node(np.array([ [[1,0,0,0], [0,1,0,0],[0,0,-1,0],[0,0,0,-1]],
# [[1,0,0,0], [0,1,0,0],[0,0,-1,0],[0,0,0,-1]],
# [[1,0,0,0], [0,1,0,0],[0,0,-1,0],[0,0,0,-1]],
# [[1,0,0,0], [0,1,0,0],[0,0,-1,0],[0,0,0,-1]],
# [[0,0,0,0], [0,1,0,0],[0,0,0,0],[0,0,0,-1]],
# [[0,0,0,0], [0,1,0,0],[0,0,0,0],[0,0,0,-1]],
# [[0,0,0,0], [0,1,0,0],[0,0,0,0],[0,0,0,-1]],
# [[0,0,0,0], [0,1,0,0],[0,0,0,0],[0,0,0,-1]] ]))
#Gc = tn.Node(np.transpose(np.array([ [1,1,1,1], [1,1,1,1], [1,1,1,1], [1,1,1,1], [1,0,0,0], [1,0,0,0], [1,0,0,0], [1,0,0,0] ])))
Ga1 = np.append(ref[:,np.newaxis], np.ones((n,2)),axis=1).astype(int)
Ga2 = np.append(Ga1, ref[:,np.newaxis],axis=1).astype(int)
Ga = tn.Node(Ga2)
Gb1= np.append(ref[np.newaxis,np.newaxis,:],np.zeros((1,1,n)),axis=1).astype(int)
Gb2= np.append(Gb1, np.append(np.zeros((1,1,n)), np.ones((1,1,n)),axis=1), axis=0).astype(int)
Gb3 = np.array([[1,0],[0,-1]])
Gb = tn.Node(np.kron(Gb3, np.transpose(Gb2)))
Gc1 = np.append(np.ones((n,1)), np.repeat(ref[:,np.newaxis],3,axis=1), axis=1).astype(int)
Gc = tn.Node(np.transpose(Gc1))
Gabc = tn.ncon([Ga, Gb, Gc], [(-1, 1), (-2, 1, 2), (2, -3)])
final = tn.outer_product(Gabc, tn.Node(np.ones((n,n,n)))).tensor
@numba.jit(nopython=True)#, parallel=True)
def enforce_delta(n, tensor):
bas1B = range(n)
for a in bas1B:
for b in bas1B:
for c in bas1B:
for d in bas1B:
for e in bas1B:
for f in bas1B:
if not(a == d and b == e and c == f):
tensor[a,b,c,d,e,f] = 0
return tensor