トロッター誤差を低減する多製品式

import warnings

import numpy as np
import matplotlib.pyplot as plt
from functools import partial
from copy import deepcopy

from qiskit import QuantumCircuit
from qiskit.quantum_info import Pauli, SparsePauliOp, Statevector
from qiskit.synthesis import SuzukiTrotter
from qiskit.transpiler import CouplingMap, PassManager
from qiskit.transpiler.preset_passmanagers import generate_preset_pass_manager
from qiskit.circuit.library import XXPlusYYGate
from qiskit.transpiler.passes.optimization.collect_and_collapse import (
    CollectAndCollapse,
    collect_using_filter_function,
    collapse_to_operation,
)

from qiskit_aer import AerSimulator
from qiskit_ibm_runtime import EstimatorV2 as Estimator, QiskitRuntimeService

from qiskit_addon_utils.problem_generators import (
    generate_xyz_hamiltonian,
    generate_time_evolution_circuit,
)
from qiskit_addon_utils.slicing import slice_by_depth
from qiskit_addon_mpf.static import setup_static_lse
from qiskit_addon_mpf.dynamic import setup_dynamic_lse
from qiskit_addon_mpf.costs import (
    setup_exact_problem,
    setup_sum_of_squares_problem,
    setup_frobenius_problem,
)
from qiskit_addon_mpf.backends.tenpy_layers import (
    LayerModel,
    LayerwiseEvolver,
)
from qiskit_addon_mpf.backends.tenpy_tebd import MPOState, MPS_neel_state

from scipy.linalg import expm

# Suppress TeNPy's `unit_cell_width` future-API warning. The default
# (`unit_cell_width=len(sites)`) is correct for Chain lattices, which is what
# `CouplingMap.from_line(...)` produces here, so the warning is informational.
warnings.filterwarnings(
    "ignore",
    message=r".*unit_cell_width.*",
    category=UserWarning,
)


# --- Helper: collect XX + YY rotations into a single gate ---
def filter_function(node):
    return node.op.name in {"rxx", "ryy"}


collect_function = partial(
    collect_using_filter_function,
    filter_function=filter_function,
    split_blocks=True,
    min_block_size=1,
)


def collapse_to_xx_plus_yy(block):
    param = 0.0
    for node in block.data:
        param += node.operation.params[0]
    return XXPlusYYGate(param)


collapse_function = partial(
    collapse_to_operation,
    collapse_function=collapse_to_xx_plus_yy,
)

pm = PassManager()
pm.append(CollectAndCollapse(collect_function, collapse_function))

L = 10

# Generate coupling map and Hamiltonian
coupling_map = CouplingMap.from_line(L, bidirectional=False)

hamiltonian = generate_xyz_hamiltonian(
    coupling_map,
    coupling_constants=(1.0, 1.0, 1.0),
    ext_magnetic_field=(0.0, 0.0, 0.0),
)
print(hamiltonian)

# Observable: ZZ on the middle pair of qubits
observable = SparsePauliOp.from_sparse_list(
    [("ZZ", (L // 2 - 1, L // 2), 1.0)], num_qubits=L
)
print(observable)

# MPF parameters
mpf_trotter_steps = [1, 2, 4]
order = 2
symmetric = False

trotter_times = np.arange(0.5, 1.55, 0.1)
exact_evolution_times = np.arange(trotter_times[0], 1.55, 0.05)

# Initial Neel state preparation
initial_state_circ = QuantumCircuit(L)
initial_state_circ.x([i for i in range(L) if i % 2 != 0])


all_circs = []
for total_time in trotter_times:
    mpf_trotter_circs = [
        generate_time_evolution_circuit(
            hamiltonian,
            time=total_time,
            synthesis=SuzukiTrotter(reps=num_steps, order=order),
        )
        for num_steps in mpf_trotter_steps
    ]

    mpf_trotter_circs = pm.run(
        mpf_trotter_circs
    )  # Collect XX and YY into XX + YY

    mpf_circuits = [
        initial_state_circ.compose(circuit) for circuit in mpf_trotter_circs
    ]
    all_circs.append(mpf_circuits)

mpf_circuits[-1].draw("mpl", fold=-1)

aer_sim = AerSimulator()
pm_sim = generate_preset_pass_manager(backend=aer_sim, optimization_level=3)

isa_circs_all_times = [
    pm_sim.run([deepcopy(c) for c in mpf_circuits])
    for mpf_circuits in all_circs
]

estimator = Estimator(mode=aer_sim)

mpf_expvals_all_times, mpf_stds_all_times = [], []
for isa_circuits in isa_circs_all_times:
    result = estimator.run(
        [(circuit, observable) for circuit in isa_circuits], precision=0.005
    ).result()
    mpf_expvals_all_times.append([res.data.evs for res in result])
    mpf_stds_all_times.append([res.data.stds for res in result])

exact_expvals = []
for t in exact_evolution_times:
    exp_H = expm(-1j * t * hamiltonian.to_matrix())
    initial_state = Statevector(initial_state_circ).data
    time_evolved_state = exp_H @ initial_state

    exact_obs = (
        time_evolved_state.conj()
        @ observable.to_matrix()
        @ time_evolved_state
    ).real
    exact_expvals.append(exact_obs)

lse = setup_static_lse(mpf_trotter_steps, order=order, symmetric=symmetric)

lse.A

lse.b

mpf_coeffs = lse.solve()
print(
    f"The static coefficients associated with the ansatze are: {mpf_coeffs}"
)

model_exact, coeffs_exact = setup_exact_problem(lse)
model_exact.solve()
print(coeffs_exact.value)

print(
    "L1 norm of the exact coefficients:",
    np.linalg.norm(coeffs_exact.value, ord=1),
)

model_approx, coeffs_approx = setup_sum_of_squares_problem(
    lse, max_l1_norm=1.5
)
model_approx.solve()
print(coeffs_approx.value)
print(
    "L1 norm of the approximate coefficients:",
    np.linalg.norm(coeffs_approx.value, ord=1),
)

# Create approximate time-evolution circuits
single_2nd_order_circ = generate_time_evolution_circuit(
    hamiltonian, time=1.0, synthesis=SuzukiTrotter(reps=1, order=order)
)
single_2nd_order_circ = pm.run(single_2nd_order_circ)  # collect XX and YY

# Find layers in the circuit
layers = slice_by_depth(single_2nd_order_circ, max_slice_depth=1)

# Create tensor network models
models = [
    LayerModel.from_quantum_circuit(layer, conserve="Sz") for layer in layers
]

# Create the time-evolution object
approx_factory = partial(
    LayerwiseEvolver,
    layers=models,
    options={
        "preserve_norm": False,
        "trunc_params": {
            "chi_max": 64,
            "svd_min": 1e-8,
            "trunc_cut": None,
        },
        "max_delta_t": 2,
    },
)

single_4th_order_circ = generate_time_evolution_circuit(
    hamiltonian, time=1.0, synthesis=SuzukiTrotter(reps=1, order=4)
)
single_4th_order_circ = pm.run(single_4th_order_circ)
exact_model_layers = [
    LayerModel.from_quantum_circuit(layer, conserve="Sz")
    for layer in slice_by_depth(single_4th_order_circ, max_slice_depth=1)
]

exact_factory = partial(
    LayerwiseEvolver,
    layers=exact_model_layers,
    dt=0.1,
    options={
        "preserve_norm": False,
        "trunc_params": {
            "chi_max": 64,
            "svd_min": 1e-8,
            "trunc_cut": None,
        },
        "max_delta_t": 2,
    },
)

def identity_factory():
    return MPOState.initialize_from_lattice(models[0].lat, conserve=True)


mps_initial_state = MPS_neel_state(models[0].lat)

mpf_dynamic_coeffs_list = []
for t in trotter_times:
    print(f"Computing dynamic coefficients for time={t}")
    lse = setup_dynamic_lse(
        mpf_trotter_steps,
        t,
        identity_factory,
        exact_factory,
        approx_factory,
        mps_initial_state,
    )
    problem, coeffs = setup_frobenius_problem(lse)
    try:
        problem.solve()
        mpf_dynamic_coeffs_list.append(coeffs.value)
    except Exception as error:
        mpf_dynamic_coeffs_list.append(np.zeros(len(mpf_trotter_steps)))
        print(error, "Calculation Failed for time", t)
    print("")

sym = {1: "^", 2: "s", 4: "p"}
# Get expectation values at all times for each Trotter step
for k, step in enumerate(mpf_trotter_steps):
    trotter_curve, trotter_curve_error = [], []
    for trotter_expvals, trotter_stds in zip(
        mpf_expvals_all_times, mpf_stds_all_times
    ):
        trotter_curve.append(trotter_expvals[k])
        trotter_curve_error.append(trotter_stds[k])

    plt.errorbar(
        trotter_times,
        trotter_curve,
        yerr=trotter_curve_error,
        alpha=0.5,
        markersize=4,
        marker=sym[step],
        color="grey",
        label=f"{mpf_trotter_steps[k]} Trotter steps",
    )

# Get expectation values at all times for the static MPF with exact coeffs
exact_mpf_curve, exact_mpf_curve_error = [], []
for trotter_expvals, trotter_stds in zip(
    mpf_expvals_all_times, mpf_stds_all_times
):
    mpf_std = np.sqrt(
        sum(
            [
                (coeff**2) * (std**2)
                for coeff, std in zip(coeffs_exact.value, trotter_stds)
            ]
        )
    )
    exact_mpf_curve_error.append(mpf_std)
    exact_mpf_curve.append(trotter_expvals @ coeffs_exact.value)

plt.errorbar(
    trotter_times,
    exact_mpf_curve,
    yerr=exact_mpf_curve_error,
    markersize=4,
    marker="o",
    label="Static MPF - Exact",
    color="purple",
)


# Get expectation values at all times for the static MPF with approximate coeffs
approx_mpf_curve, approx_mpf_curve_error = [], []
for trotter_expvals, trotter_stds in zip(
    mpf_expvals_all_times, mpf_stds_all_times
):
    mpf_std = np.sqrt(
        sum(
            [
                (coeff**2) * (std**2)
                for coeff, std in zip(coeffs_approx.value, trotter_stds)
            ]
        )
    )
    approx_mpf_curve_error.append(mpf_std)
    approx_mpf_curve.append(trotter_expvals @ coeffs_approx.value)

plt.errorbar(
    trotter_times,
    approx_mpf_curve,
    yerr=approx_mpf_curve_error,
    markersize=4,
    marker="o",
    label="Static MPF - Approx",
    color="orange",
)


# Get expectation values at all times for the dynamic MPF
dynamic_mpf_curve, dynamic_mpf_curve_error = [], []
for trotter_expvals, trotter_stds, dynamic_coeffs in zip(
    mpf_expvals_all_times, mpf_stds_all_times, mpf_dynamic_coeffs_list
):
    mpf_std = np.sqrt(
        sum(
            [
                (coeff**2) * (std**2)
                for coeff, std in zip(dynamic_coeffs, trotter_stds)
            ]
        )
    )
    dynamic_mpf_curve_error.append(mpf_std)
    dynamic_mpf_curve.append(trotter_expvals @ dynamic_coeffs)

plt.errorbar(
    trotter_times,
    dynamic_mpf_curve,
    yerr=dynamic_mpf_curve_error,
    markersize=4,
    marker="o",
    label="Dynamic MPF",
    color="pink",
)


# Exact expectation values
plt.plot(
    exact_evolution_times,
    exact_expvals,
    color="red",
    linestyle="--",
    label="Exact time-evolution",
)

plt.title(f"$\\langle Z_{{{L//2-1}}} Z_{{{L//2}}} \\rangle$ vs time")
plt.xlabel("Time")
plt.ylabel("Expectation Value")
plt.legend(loc="upper center", bbox_to_anchor=(0.5, -0.2), ncol=2)
plt.grid(alpha=0.1)
plt.tight_layout()
plt.show()

# -------------------------Step 1-------------------------
L = 50
coupling_map = CouplingMap.from_line(L, bidirectional=False)

# XXZ Hamiltonian with random couplings (Ref. [3])
np.random.seed(0)
even_edges = list(coupling_map.get_edges())[::2]
odd_edges = list(coupling_map.get_edges())[1::2]

Js = np.random.uniform(0.5, 1.5, size=L)
hamiltonian = SparsePauliOp(Pauli("I" * L))
for i, edge in enumerate(even_edges + odd_edges):
    hamiltonian += SparsePauliOp.from_sparse_list(
        [
            ("XX", (edge), 2 * Js[i]),
            ("YY", (edge), 2 * Js[i]),
            ("ZZ", (edge), 4 * Js[i]),
        ],
        num_qubits=L,
    )

observable = SparsePauliOp.from_sparse_list(
    [("ZZ", (L // 2 - 1, L // 2), 1.0)], num_qubits=L
)

total_time = 3
mpf_trotter_steps = [3, 4, 6]
order = 2
symmetric = True

# Static coefficients
lse = setup_static_lse(mpf_trotter_steps, order=order, symmetric=symmetric)
mpf_coeffs = lse.solve()
print(f"Static coefficients: {mpf_coeffs}")
print(f"L1 norm: {np.linalg.norm(mpf_coeffs, ord=1)}")

model_approx, coeffs_approx = setup_sum_of_squares_problem(
    lse, max_l1_norm=2.0
)
model_approx.solve()
print(f"Approximate coefficients: {coeffs_approx.value}")
print(f"L1 norm (approx): {np.linalg.norm(coeffs_approx.value, ord=1)}")

# -------------------------Dynamic coefficients-------------------------
single_2nd_order_circ = generate_time_evolution_circuit(
    hamiltonian, time=1.0, synthesis=SuzukiTrotter(reps=1, order=order)
)
single_2nd_order_circ = pm.run(single_2nd_order_circ)

layers = slice_by_depth(single_2nd_order_circ, max_slice_depth=1)
models = [
    LayerModel.from_quantum_circuit(layer, conserve="Sz") for layer in layers
]

approx_factory = partial(
    LayerwiseEvolver,
    layers=models,
    options={
        "preserve_norm": False,
        "trunc_params": {"chi_max": 64, "svd_min": 1e-8, "trunc_cut": None},
        "max_delta_t": 4,
    },
)

single_4th_order_circ = generate_time_evolution_circuit(
    hamiltonian, time=1.0, synthesis=SuzukiTrotter(reps=1, order=4)
)
single_4th_order_circ = pm.run(single_4th_order_circ)
exact_model_layers = [
    LayerModel.from_quantum_circuit(layer, conserve="Sz")
    for layer in slice_by_depth(single_4th_order_circ, max_slice_depth=1)
]

exact_factory = partial(
    LayerwiseEvolver,
    layers=exact_model_layers,
    dt=0.1,
    options={
        "preserve_norm": False,
        "trunc_params": {"chi_max": 64, "svd_min": 1e-8, "trunc_cut": None},
        "max_delta_t": 3,
    },
)


def identity_factory():
    return MPOState.initialize_from_lattice(models[0].lat, conserve=True)


mps_initial_state = MPS_neel_state(models[0].lat)

print(f"Computing dynamic coefficients for time={total_time}")
lse_dyn = setup_dynamic_lse(
    mpf_trotter_steps,
    total_time,
    identity_factory,
    exact_factory,
    approx_factory,
    mps_initial_state,
)
problem, coeffs_dyn = setup_frobenius_problem(lse_dyn)
try:
    problem.solve()
    mpf_dynamic_coeffs = coeffs_dyn.value
except Exception as error:
    mpf_dynamic_coeffs = np.zeros(len(mpf_trotter_steps))
    print(error, "Calculation Failed")

# -------------------------Step 1 (cont): Build circuits-------------------------
mpf_circuits = []
for k in mpf_trotter_steps:
    circuit = QuantumCircuit(L)
    circuit.x([i for i in range(L) if i % 2])
    trotter_circ = generate_time_evolution_circuit(
        hamiltonian,
        synthesis=SuzukiTrotter(reps=k, order=order),
        time=total_time,
    )
    circuit.compose(trotter_circ, qubits=range(L), inplace=True)
    mpf_circuits.append(circuit)

# Baseline "single deep circuit" comparison run with k=10 Trotter steps.
# Its two-qubit depth is deeper than the deepest MPF constituent (k_max=6) plus
# the overhead of running multiple circuits, pushing it into the noise-limited
# regime where MPF is expected to outperform. It does NOT target the MPF's effective
# Trotter error (which would require many more steps).
comp_circuit = QuantumCircuit(L)
comp_circuit.x([i for i in range(L) if i % 2])
trotter_circ = generate_time_evolution_circuit(
    hamiltonian,
    synthesis=SuzukiTrotter(reps=10, order=order),
    time=total_time,
)
comp_circuit.compose(trotter_circ, qubits=range(L), inplace=True)
mpf_circuits.append(comp_circuit)

# -------------------------Step 2-------------------------
service = QiskitRuntimeService()
# backend = service.least_busy(operational=True, simulator=False, min_num_qubits=L)
backend = service.backend("ibm_fez")
print(backend)

transpiler = generate_preset_pass_manager(
    optimization_level=3, backend=backend
)
transpiled_circuits = [transpiler.run(circ) for circ in mpf_circuits]

isa_observables = [
    observable.apply_layout(circ.layout) for circ in transpiled_circuits
]

# -------------------------Step 3-------------------------
estimator = Estimator(mode=backend)
estimator.options.default_shots = 30000

# Error suppression/mitigation
estimator.options.dynamical_decoupling.enable = True
estimator.options.twirling.enable_gates = True
estimator.options.twirling.enable_measure = True
estimator.options.twirling.num_randomizations = "auto"
estimator.options.twirling.strategy = "active-accum"
estimator.options.resilience.measure_mitigation = True
estimator.options.experimental.execution_path = "gen3-turbo"

estimator.options.resilience.zne_mitigation = True
estimator.options.resilience.zne.noise_factors = (1, 1.2, 1.4)
estimator.options.resilience.zne.extrapolator = "linear"

estimator.options.environment.job_tags = ["TUT_MPF"]

job_50 = estimator.run(
    [
        (circ, observable)
        for circ, observable in zip(transpiled_circuits, isa_observables)
    ]
)

# -------------------------Step 4-------------------------
result = job_50.result()
evs = [res.data.evs for res in result]
std = [res.data.stds for res in result]

print(evs)
print(std)

exact_mpf_std = np.sqrt(
    sum([(coeff**2) * (std**2) for coeff, std in zip(mpf_coeffs, std[:3])])
)
print(
    "Exact static MPF expectation value: ",
    evs[:3] @ mpf_coeffs,
    "+-",
    exact_mpf_std,
)
approx_mpf_std = np.sqrt(
    sum(
        [
            (coeff**2) * (std**2)
            for coeff, std in zip(coeffs_approx.value, std[:3])
        ]
    )
)
print(
    "Approximate static MPF expectation value: ",
    evs[:3] @ coeffs_approx.value,
    "+-",
    approx_mpf_std,
)
dynamic_mpf_std = np.sqrt(
    sum(
        [
            (coeff**2) * (std**2)
            for coeff, std in zip(mpf_dynamic_coeffs, std[:3])
        ]
    )
)
print(
    "Dynamic MPF expectation value: ",
    evs[:3] @ mpf_dynamic_coeffs,
    "+-",
    dynamic_mpf_std,
)

sym = {3: "^", 4: "s", 6: "p"}
for k, step in enumerate(mpf_trotter_steps):
    plt.errorbar(
        k,
        evs[k],
        yerr=std[k],
        alpha=0.5,
        markersize=4,
        marker=sym[step],
        color="grey",
        label=f"{mpf_trotter_steps[k]} Trotter steps",
    )

plt.errorbar(
    3,
    evs[-1],
    yerr=std[-1],
    alpha=0.5,
    markersize=8,
    marker="x",
    color="blue",
    label="10 Trotter steps",
)

plt.errorbar(
    4,
    evs[:3] @ mpf_coeffs,
    yerr=exact_mpf_std,
    markersize=4,
    marker="o",
    color="purple",
    label="Static MPF",
)

plt.errorbar(
    5,
    evs[:3] @ coeffs_approx.value,
    yerr=approx_mpf_std,
    markersize=4,
    marker="o",
    color="orange",
    label="Approximate static MPF",
)

plt.errorbar(
    6,
    evs[:3] @ mpf_dynamic_coeffs,
    yerr=dynamic_mpf_std,
    markersize=4,
    marker="o",
    color="pink",
    label="Dynamic MPF",
)

exact_obs = -0.24384471447172074  # Calculated via Tensor Network calculation
plt.axhline(
    y=exact_obs, linestyle="--", color="red", label="Exact time-evolution"
)

plt.title(
    f"$\\langle Z_{{{L//2-1}}} Z_{{{L//2}}} \\rangle$ at time {total_time} for the different methods"
)
plt.xlabel("Method")
plt.ylabel("Expectation Value")
plt.legend(loc="upper center", bbox_to_anchor=(0.5, -0.2), ncol=2)
plt.grid(alpha=0.1)
plt.tight_layout()
plt.show()