"""Contains all interacting functionality and solvers."""
import os
import copy
import string
import itertools
import functools
from tqdm import tqdm
import numpy as np
import scipy.sparse as sps
import scipy.sparse.linalg as spsla
import iDEA.system
import iDEA.state
import iDEA.methods.non_interacting
if os.environ.get("iDEA_GPU") == "True":
import cupy as cnp
import cupyx.scipy.sparse as csps
import cupyx.scipy.sparse.linalg as cspsla
name = "interacting"
[docs]def kinetic_energy_operator(s: iDEA.system.System) -> sps.dia_matrix:
r"""
Compute many-particle kinetic energy operator as a matrix.
This is built using a given number of finite differences to represent the second derivative.
The number of differences taken is defined in s.stencil.
| Args:
| s: iDEA.system.System, System object.
| Returns:
| K: sps.dia_matrix, Kintetic energy operator.
"""
k = iDEA.methods.non_interacting.kinetic_energy_operator(s)
k = sps.dia_matrix(k)
I = sps.identity(s.x.shape[0], format="dia")
partial_operators = lambda A, B, k, n: (
A if i + k == n - 1 else B for i in range(n)
)
fold_partial_operators = lambda f, po: functools.reduce(
lambda acc, val: f(val, acc, format="dia"), po
)
generate_terms = lambda f, A, B, n: (
fold_partial_operators(f, partial_operators(A, B, k, n)) for k in range(n)
)
terms = generate_terms(sps.kron, h, I, s.count)
K = sps.dia_matrix((s.x.shape[0] ** s.count,) * 2, dtype=float)
for term in terms:
K += term
return K
[docs]def external_potential_operator(s: iDEA.system.System) -> sps.dia_matrix:
r"""
Compute many-particle external potential energy operator as a matrix.
| Args:
| s: iDEA.system.System, System object.
| Returns:
| Vext: sps.dia_matrix, External potential operator.
"""
vext = iDEA.methods.non_interacting.external_potential_operator(s)
vext = sps.dia_matrix(vext)
I = sps.identity(s.x.shape[0], format="dia")
partial_operators = lambda A, B, k, n: (
A if i + k == n - 1 else B for i in range(n)
)
fold_partial_operators = lambda f, po: functools.reduce(
lambda acc, val: f(val, acc, format="dia"), po
)
generate_terms = lambda f, A, B, n: (
fold_partial_operators(f, partial_operators(A, B, k, n)) for k in range(n)
)
terms = generate_terms(sps.kron, h, I, s.count)
Vext = sps.dia_matrix((s.x.shape[0] ** s.count,) * 2, dtype=float)
for term in terms:
Vext += term
return Vext
[docs]def hamiltonian(s: iDEA.system.System) -> sps.dia_matrix:
r"""
Compute the many-body Hamiltonian.
| Args:
| s: iDEA.system.System, System object.
| Returns:
| H: sps.dia_matrix, Hamiltonian.
"""
# Construct the non-interacting part of the many-body Hamiltonian
h = iDEA.methods.non_interacting.hamiltonian(s)[0]
h = sps.dia_matrix(h)
I = sps.identity(s.x.shape[0], format="dia")
partial_operators = lambda A, B, k, n: (
A if i + k == n - 1 else B for i in range(n)
)
fold_partial_operators = lambda f, po: functools.reduce(
lambda acc, val: f(val, acc, format="dia"), po
)
generate_terms = lambda f, A, B, n: (
fold_partial_operators(f, partial_operators(A, B, k, n)) for k in range(n)
)
terms = generate_terms(sps.kron, h, I, s.count)
H0 = sps.dia_matrix((s.x.shape[0] ** s.count,) * 2, dtype=float)
for term in terms:
H0 += term
# Add the interaction part of the many-body Hamiltonian
symbols = string.ascii_lowercase + string.ascii_uppercase
if s.count > 1:
indices = ",".join(
["".join(c) for c in itertools.combinations(symbols[: s.count], 2)]
)
U = np.log(
np.einsum(
indices + "->" + symbols[: s.count],
*(np.exp(s.v_int),) * int(s.count * (s.count - 1) / 2)
)
)
U = sps.diags(U.reshape((H0.shape[0])), format="dia")
else:
U = 0.0
# Construct the total many-body Hamiltonian
H = H0 + U
return H
[docs]def total_energy(s: iDEA.system.System, state: iDEA.state.ManyBodyState) -> float:
r"""
Compute the total energy of an interacting state.
| Args:
| s: iDEA.system.System, System object.
| state: iDEA.state.ManyBodyState, State.
| Returns:
| E: float, Total energy.
"""
return state.energy
def _permutation_parity(p):
r"""
Compute the permulation paritiy of a given permutation.
| Args:
| p: tuple, Permutation.
| Returns:
| parity: float, Permutation parity.
"""
p = list(p)
parity = 1
for i in range(0, len(p) - 1):
if p[i] != i:
parity *= -1
mn = min(range(i, len(p)), key=p.__getitem__)
p[i], p[mn] = p[mn], p[i]
return parity
[docs]def antisymmetrize(s, spaces, spins, energies):
r"""
Antisymmetrize the solution to the Schrodinger equation.
| Args:
| s: iDEA.system.System, System object.
| spaces: np.ndarray, Spatial parts of the wavefunction.
| spins: np.ndarray, Spin parts of the wavefunction.
| energies: np.ndarray, Energies.
| Returns:
| fulls: np.ndarray, Full anantisymmetrized wavefunction.
| spaces: np.ndarray, Spatial parts of the wavefunction.
| spins: np.ndarray, Spin parts of the wavefunction.
| energies: np.ndarray, Energies.
"""
# Perform antisymmetrization.
l = string.ascii_lowercase[: s.count]
L = string.ascii_uppercase[: s.count]
st = (
l
+ "Y,"
+ L
+ "Y->"
+ "".join([i for sub in list(zip(l, L)) for i in sub])
+ "Y"
)
fulls = np.einsum(st, spaces, spins)
L = list(zip(list(range(0, s.count * 2, 2)), list(range(1, s.count * 2, 2))))
perms = itertools.permutations(list(range(s.count)))
fulls_copy = copy.deepcopy(fulls)
fulls = np.zeros_like(fulls)
for p in perms:
indices = list(itertools.chain(*[L[e] for e in p]))
fulls += _permutation_parity(p) * np.moveaxis(
fulls_copy, list(range(s.count * 2)), indices
)
# Filter out zeros.
allowed_fulls = []
allowed_energies = []
allowed_spaces = []
allowed_spins = []
for n in range(fulls.shape[-1]):
if np.allclose(fulls[..., n], np.zeros(fulls.shape[:-1])):
pass
else:
allowed_fulls.append(fulls[..., n])
allowed_energies.append(energies[n])
allowed_spaces.append(spaces[..., n])
allowed_spins.append(spins[..., n])
fulls = np.moveaxis(np.array(allowed_fulls), 0, -1)
spaces = np.moveaxis(np.array(allowed_spaces), 0, -1)
spins = np.moveaxis(np.array(allowed_spins), 0, -1)
energies = np.array(allowed_energies)
# Normalise.
for k in range(fulls.shape[-1]):
fulls[..., k] = fulls[..., k] / np.sqrt(
np.sum(abs(fulls[..., k]) ** 2) * s.dx**s.count
)
# Filter out duplicates.
allowed_fulls = []
allowed_energies = []
for n in range(fulls.shape[-1] - 1):
if np.allclose(abs(fulls[..., n]), abs(fulls[..., n + 1])):
pass
else:
allowed_fulls.append(fulls[..., n])
allowed_energies.append(energies[n])
allowed_fulls.append(fulls[..., -1])
allowed_energies.append(energies[-1])
fulls = np.moveaxis(np.array(allowed_fulls), 0, -1)
spaces = spaces[..., : fulls.shape[-1]]
spins = spins[..., : fulls.shape[-1]]
energies = np.array(allowed_energies)
return fulls, spaces, spins, energies
def _estimate_level(s: iDEA.system.System, k: int) -> int:
r"""
Estimate the solution to the Schrodinger equation needed to eachive given antisymetric energy state.
| Args:
| s: iDEA.system.System, System object.
| k: int, Target energy state.
| Returns:
| level: int, Extimate of level of excitement.
"""
return (abs(s.up_count - s.down_count) + 1) ** 2 * s.count * (k + 1)
def _solve_on_gpu(H: np.ndarray, k: int) -> tuple:
r"""
Solves the eigenproblem on the GPU.
| Args:
| H: np.ndarray, Hamiltonian.
| k: int, Eigenstate to solve for.
| Returns:
| eigenvalues_gpu, eigenstates_gpu: tuple, Solved eigenvalues and eigenstates.
"""
sigma = 0
which = "LA"
H_gpu_shifted = csps.csr_matrix(H - sigma * csps.csr_matrix(sps.eye(H.shape[0])))
H_gpu_LU = cspsla.splu(H_gpu_shifted)
H_gpu_LO = cspsla.LinearOperator(H_gpu_shifted.shape, H_gpu_LU.solve)
eigenvalues_gpu, eigenstates_gpu = cspsla.eigsh(H_gpu_LO, k=k, which=which)
eigenvalues_gpu = eigenvalues_gpu
eigenstates_gpu = eigenstates_gpu
eigenvalues_gpu = (1 + eigenvalues_gpu * sigma) / eigenvalues_gpu
idx = np.argsort(eigenvalues_gpu)
eigenstates_gpu = cnp.transpose(eigenstates_gpu)
eigenvalues_gpu = eigenvalues_gpu[idx]
eigenstates_gpu = cnp.transpose(eigenstates_gpu[idx])
return eigenvalues_gpu, eigenstates_gpu
[docs]def solve(
s: iDEA.system.System, H: np.ndarray = None, k: int = 0, level=None
) -> iDEA.state.ManyBodyState:
r"""
Solves the interacting Schrodinger equation of the given system.
| Args:
| s: iDEA.system.System, System object.
| H: np.ndarray, Hamiltonian [If None this will be computed from s]. (default = None)
| k: int, Energy state to solve for. (default = 0, the ground-state)
| level: int. Max level of excitation to use when solving the Schrodinger equation.
| Returns:
| state: iDEA.state.ManyBodyState, Solved state.
"""
# Construct the many-body state.
state = iDEA.state.ManyBodyState()
# Construct the Hamiltonian.
if H is None:
H = hamiltonian(s)
# Estimate the level of excitation.
if level is None:
level = _estimate_level(s, k)
# Solve the many-body Schrodinger equation.
print("iDEA.methods.interacting.solve: solving eigenproblem...")
if os.environ.get("iDEA_GPU") == "True":
H_gpu = csps.csr_matrix(H)
energies, spaces = _solve_on_gpu(H_gpu, level)
energies = energies.get()
spaces = spaces.get()
else:
energies, spaces = spsla.eigsh(H.tocsr(), k=level, which="SA")
# Reshape and normalise the solutions.
spaces = spaces.reshape((s.x.shape[0],) * s.count + (spaces.shape[-1],))
for j in range(spaces.shape[-1]):
spaces[..., j] = spaces[..., j] / np.sqrt(
np.sum(abs(spaces[..., j]) ** 2) * s.dx**s.count
)
# Construct the spin part.
symbols = string.ascii_lowercase + string.ascii_uppercase
u = np.array([1, 0])
d = np.array([0, 1])
spin_state = tuple([u if spin == "u" else d for spin in s.electrons])
spin = np.einsum(
",".join(symbols[: s.count]) + "->" + "".join(symbols[: s.count]), *spin_state
)
spins = np.zeros(shape=((2,) * s.count + (spaces.shape[-1],)))
for i in range(spaces.shape[-1]):
spins[..., i] = spin
# Antisymmetrize.
fulls, spaces, spins, energies = antisymmetrize(s, spaces, spins, energies)
# Populate the state.
state.space = spaces[..., k]
state.spin = spins[..., k]
state.full = fulls[..., k]
state.energy = energies[k]
return state
[docs]def propagate_step(
s: iDEA.system.System,
evolution: iDEA.state.ManyBodyEvolution,
H: sps.dia_matrix,
v_ptrb: np.ndarray,
j: int,
dt: float,
objs: tuple,
) -> iDEA.state.ManyBodyEvolution:
r"""
Propagate a many body state forward in time, one time-step, due to a local pertubation.
| Args:
| s: iDEA.system.System, System object.
| evolution: iDEA.state.ManyBodyEvolution, time-dependent evolution.
| H: np.ndarray, Static Hamiltonian [If None this will be computed from s]. (default = None)
| v_ptrb: np.ndarray, Local perturbing potential on the grid of t and x values, indexed as v_ptrb[time,space].
| j: int, Time index.
| dt: float, Time-step.
| objs: tuple. Tuple of objects needed to construct many-body operator (I, generate_terms).
| Returns:
| evolution: iDEA.state.ManyBodyEvolution, time-dependent evolution one time-step evolved.
"""
# Construct the pertubation potential.
vptrb = sps.dia_matrix(np.diag(v_ptrb[j, :]))
terms = objs[1](sps.kron, vptrb, objs[0], s.count)
Vptrb = sps.dia_matrix((s.x.shape[0] ** s.count,) * 2, dtype=float)
for term in terms:
Vptrb += term
# Contruct the perturbed Hamiltonian.
Hp = H + Vptrb
# Evolve.
wavefunction = evolution.td_space[j - 1, ...].reshape((s.x.shape[0] ** s.count))
wavefunction = spsla.expm_multiply(-1.0j * dt * Hp, wavefunction)
evolution.td_space[j, ...] = wavefunction.reshape((s.x.shape[0],) * s.count)
return evolution
[docs]def propagate(
s: iDEA.system.System,
state: iDEA.state.ManyBodyState,
v_ptrb: np.ndarray,
t: np.ndarray,
H: sps.dia_matrix = None,
) -> iDEA.state.ManyBodyEvolution:
r"""
Propagate a many body state forward in time due to a local pertubation.
| Args:
| s: iDEA.system.System, System object.
| state: iDEA.state.ManyBodyState, State to be propigated.
| v_ptrb: np.ndarray, Local perturbing potential on the grid of t and x values, indexed as v_ptrb[time,space].
| t: np.ndarray, Grid of time values.
| H: np.ndarray, Static Hamiltonian [If None this will be computed from s]. (default = None)
| Returns:
| evolution: iDEA.state.ManyBodyEvolution, Solved time-dependent evolution.
"""
# Construct the unperturbed Hamiltonian.
if H is None:
H = hamiltonian(s)
# Compute timestep.
dt = t[1] - t[0]
# Initilise time-dependent wavefunction.
evolution = iDEA.state.ManyBodyEvolution(initial_state=state)
evolution.td_space = np.zeros(shape=t.shape + state.space.shape, dtype=complex)
evolution.td_space[0, ...] = copy.deepcopy(evolution.space)
# Construct objects needed to update potential.
I = sps.identity(s.x.shape[0], format="dia")
partial_operators = lambda A, B, k, n: (
A if i + k == n - 1 else B for i in range(n)
)
fold_partial_operators = lambda f, po: functools.reduce(
lambda acc, val: f(val, acc, format="dia"), po
)
generate_terms = lambda f, A, B, n: (
fold_partial_operators(f, partial_operators(A, B, k, n)) for k in range(n)
)
objs = (I, generate_terms)
# Propagate.
for j, ti in enumerate(
tqdm(t, desc="iDEA.methods.interacting.propagate: propagating state")
):
if j != 0:
propagate_step(s, evolution, H, v_ptrb, j, dt, objs)
# Populate the many-body time-dependent evolution.
evolution.v_ptrb = v_ptrb
evolution.t = t
return evolution