Técnicas avanzadas para QAOA

Estimación de uso: 3 minutos en un procesador r2 Heron (NOTA: Esto es solo una estimación). El tiempo de ejecución puede variar.

Requisitos previos

Conceptos básicos del algoritmo cuántico de optimización aproximada (QAOA)
Tutorial sobre la resolución de problemas de optimización cuántica a escala industrial
La lección sobre QAOA a escala industrial (que forma parte del curso «La computación cuántica en la práctica ») en IBM Quantum® Learning

Resultados del aprendizaje

Utilizar técnicas avanzadas que contribuyan a mejorar la transpilación de QAOA y ofrezcan un medio para mejorar el rendimiento de QAOA

Para ver una explicación detallada del contenido de este tutorial, consulta este vídeo de Qiskit.

En segundo plano

Este tutorial presenta dos técnicas avanzadas para mejorar el rendimiento del algoritmo de optimización cuántica aproximada (QAOA) con un gran número de qubits.

Las técnicas avanzadas de este cuaderno incluyen:

Estrategia SWAP con asignación inicial SAT : se trata de una pasada del transpilador diseñada específicamente para QAOA que combina una estrategia SWAP con un solucionador SAT con el fin de mejorar la selección de los qubits físicos que se utilizarán en la QPU. La estrategia SWAP aprovecha la conmutatividad de los operadores QAOA para reordenar las puertas, de modo que las capas de puertas SWAP puedan ejecutarse simultáneamente, reduciendo así la profundidad del circuito [1]. El solucionador SAT se utiliza para encontrar una asignación inicial que minimice el número de operaciones SWAP necesarias para asignar los qubits del circuito a los qubits físicos del dispositivo [2].
CVaR función de coste : Normalmente, se utiliza el valor esperado del hamiltoniano de coste como función de coste para el QAOA, pero, tal y como se demostró en [3], centrarse en la cola de la distribución, en lugar de en el valor esperado, puede mejorar el rendimiento del QAOA en problemas de optimización combinatoria. El « CVaR » cumple con este objetivo. Para un conjunto dado de soluciones con los valores objetivos correspondientes del problema de optimización considerado, el valor en riesgo condicional ( CVaR ) con un nivel de confianza del $\alpha \in [0, 1]$ se define como la media de las soluciones óptimas $\alpha$ [3]. Por lo tanto, $\alpha = 1$ corresponde al valor esperado estándar, mientras que $\alpha=0$ corresponde al mínimo de los disparos dados, y $\alpha \in (0, 1)$ supone un término medio entre centrarse en los mejores disparos y aplicar al mismo tiempo un cierto promedio para suavizar el panorama de optimización. Además, el método de estimación de la media a través de la media ( CVaR ) puede utilizarse como técnica de mitigación de errores para mejorar la calidad de la estimación del valor objetivo [4].

Al finalizar este tutorial, deberías ser capaz de utilizar estas técnicas para obtener los mejores resultados al ejecutar QAOA en tus problemas de optimización.

Requisitos

Antes de empezar este tutorial, asegúrate de tener instalado lo siguiente:

Qiskit SDK v2.0 o posterior, con soporte de visualización
v0.43Qiskit Runtime o posterior (pip install qiskit-ibm-runtime)
Biblioteca de gráficos Rustworkx (pip install rustworkx)
Python SAT (pip install python-sat)
Complemento «Optimization Mapper» para Qiskit (pip install qiskit-addon-opt-mapper)

Configuración

from __future__ import annotations

import numpy as np
import rustworkx as rx
from dataclasses import dataclass
from itertools import combinations
from threading import Timer
from collections.abc import Callable, Iterable
from pysat.formula import CNF, IDPool
from pysat.solvers import Solver
from scipy.optimize import minimize
from rustworkx.visualization import mpl_draw as draw_graph

from qiskit.quantum_info import SparsePauliOp
from qiskit.circuit.library import QAOAAnsatz
from qiskit.circuit import QuantumCircuit, ParameterVector
from qiskit.transpiler import CouplingMap, PassManager
from qiskit.transpiler.preset_passmanagers import generate_preset_pass_manager
from qiskit.transpiler.passes.routing.commuting_2q_gate_routing import (
    SwapStrategy,
    FindCommutingPauliEvolutions,
    Commuting2qGateRouter,
)

from qiskit_ibm_runtime import QiskitRuntimeService, Session
from qiskit_ibm_runtime import SamplerV2 as Sampler

from qiskit_addon_opt_mapper.applications import Maxcut
from qiskit_addon_opt_mapper.converters import OptimizationProblemToQubo
from qiskit_addon_opt_mapper.translators import to_ising

Problema del corte máximo

Analicemos cómo resolver el problema del corte máximo en un grafo de 100 nodos utilizando QAOA. El problema del corte máximo es un problema de optimización combinatoria que se define en un grafo $G = (V, E)$ , donde $V$ es el conjunto de vértices y $E$ es el conjunto de aristas. El objetivo es dividir los vértices en dos conjuntos, $S$ y $V \setminus S$ , de tal manera que se maximice el número de aristas entre ambos conjuntos. En este ejemplo, utilizamos un grafo de 100 nodos basado en un mapa de acoplamiento de hardware.

Ejemplo de simulador a pequeña escala

Dado que el objetivo de este tutorial es mostrar el rendimiento de QAOA a escalas que superan la capacidad de un simulador, omitiremos este paso.

Si quieres probar un flujo de trabajo de QAOA basado en un simulador, echa un vistazo al tutorial sobre el algoritmo de optimización aproximada cuántica.

Ejemplo de hardware a gran escala

Paso 1: Asignar entradas clásicas a un problema cuántico

Gráfico → Hamiltoniano

En primer lugar, representa el problema en un circuito cuántico adecuado para el QAOA. Al igual que en el tutorial introductorio de QAOA, el complemento Optimization Mapper de Qiskit construye el «max-cut» OptimizationProblem a partir del grafo, lo convierte en un QUBO y traduce el QUBO al hamiltoniano de coste de Ising. El traductor también devuelve una constante tal offset que el valor de corte de una cadena de bits es $-(\langle H_C \rangle + \text{offset})$ .

# Instantiate runtime to access backend
service = QiskitRuntimeService()
backend = service.least_busy(
    min_num_qubits=100, operational=True, simulator=False
)
print(backend)

Output:

<IBMBackend('ibm_pittsburgh')>

# Check if the coupling map is symmetric. We will add a conditional below
# to avoid over-counting edges for symmetric/bi-directional coupling maps.

backend.coupling_map.is_symmetric

Output:

True

n = 100
graph_100 = rx.PyGraph()
graph_100.add_nodes_from((np.arange(0, n, 1)))
w = 1.0
elist = []

for edge in backend.coupling_map:
    if (edge[0] < n) and (edge[1] < n):
        # Conditional to avoid over-counting edges
        if (
            edge[1],
            edge[0],
            w,
        ) not in elist:
            elist.append((edge[0], edge[1], w))

graph_100.add_edges_from(elist)
draw_graph(graph_100, with_labels=True)

Output:

# Construct the cost Hamiltonian:
# graph -> OptimizationProblem -> QUBO -> Ising Hamiltonian
maxcut_problem = Maxcut(graph_100).to_optimization_problem()
qubo = OptimizationProblemToQubo().convert(maxcut_problem)
cost_hamiltonian, offset = to_ising(qubo)
print("Cost Function Hamiltonian:", cost_hamiltonian)
print("Offset:", offset)

Output:

Cost Function Hamiltonian: SparsePauliOp(['IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZ', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZI', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIIIIZIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIZIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIZIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIZIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIZIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIZIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIIIIZIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIIIIIIIZIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIIIIIIZIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIZIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIZIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIZIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIZIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIZIIIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIZIIIIIIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIZIIIIIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIZIIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IZIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'ZIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII'],
              coeffs=[0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j,
 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j,
 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j,
 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j,
 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j,
 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j,
 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j,
 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j,
 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j,
 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j,
 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j,
 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j,
 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j,
 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j])
Offset: -55.5

Paso 2: Optimizar el problema para la ejecución en hardware cuántico

Estrategia SWAP con el mapeo inicial SAT

Demostraremos cómo construir y optimizar circuitos QAOA utilizando la estrategia SWAP con mapeo inicial SAT, un pase transpilador específicamente diseñado para QAOA aplicado a problemas cuadráticos.

En este ejemplo, elegimos una estrategia de inserción SWAP para bloques de puertas de dos qubits conmutables, que aplica capas de puertas SWAP que se pueden ejecutar simultáneamente en el mapa de acoplamiento. Esta estrategia se presenta en [1] y se pasa a Commuting2qGateRouter, que se expone como una pasada de transpilador Qiskit estandarizada (véase Commuting2qGateRouter). En este ejemplo utilizamos una estrategia de intercambio de líneas.

# Extract longest path with no repeated nodes
nodes = rx.longest_simple_path(graph_100)

# Collect even edges and odd edges
even_edges = [
    (nodes[i], nodes[i + 1])
    if nodes[i] < nodes[i + 1]
    else (nodes[i + 1], nodes[i])
    for i in range(0, len(nodes) - 1, 2)
]
odd_edges = [
    (nodes[i], nodes[i + 1])
    if nodes[i] < nodes[i + 1]
    else (nodes[i + 1], nodes[i])
    for i in range(1, len(nodes) - 1, 2)
]
edge_list = [
    (edge[0], edge[1]) if edge[0] < edge[1] else (edge[1], edge[0])
    for edge in graph_100.edge_list()
]

swap_strategy = SwapStrategy(CouplingMap(edge_list), (even_edges, odd_edges))

Reasignar el gráfico utilizando un mapeador SAT

Incluso cuando un circuito está formado por puertas de conmutación (como ocurre con el circuito QAOA, pero también con las simulaciones de Hamiltonianos de Ising mediante el método de Trotter), encontrar una buena configuración inicial es una tarea compleja. Cuando utilizamos el enfoque basado en SAT presentado en [2], podemos descubrir mapeos iniciales eficaces para circuitos con puertas conmutativas, lo que se traduce en una reducción significativa del número de capas SWAP necesarias. Se ha demostrado que este enfoque es escalable hasta 500 qubits, tal y como se ilustra en el artículo.

El siguiente código muestra cómo utilizar SATMapper de Matsuo et al. para redistribuir el gráfico. Este proceso permite mapear el problema a un estado inicial más óptimo para una estrategia SWAP especificada, lo que se traduce en una reducción significativa del número de capas SWAP necesarias para ejecutar el circuito.

En SATMapper, el problema de encontrar una buena cartografía inicial se formula como un problema SAT. Se utiliza un solucionador SAT para encontrar dicho mapeo inicial para el circuito QAOA. python-sat (pysat para abreviar) es una biblioteca Python para un solucionador SAT, y la utilizaremos para resolver el problema SAT en este ejemplo.

"""A class to solve the SWAP gate insertion initial mapping problem
using the SAT approach from https://arxiv.org/abs/2212.05666.
"""


@dataclass
class SATResult:
    """A data class to hold the result of a SAT solver."""

    satisfiable: bool  # Satisfiable is True if the SAT model could be solved
    # in a given time.
    solution: dict  # The solution to the SAT problem if it is satisfiable.
    mapping: list  # The mapping of nodes in the pattern graph to nodes in the
    # target graph.
    elapsed_time: float  # The time it took to solve the SAT model.


class SATMapper:
    r"""A class to introduce a SAT-approach to solve
    the initial mapping problem in SWAP gate insertion for commuting gates.

    When this pass is run on a DAG it will look for the first instance of
    :class:`.Commuting2qBlock` and use the program graph :math:`P` of this block
    of gates to find a layout for a given swap strategy. This layout is found
    with a binary search over the layers :math:`l` of the swap strategy. At each
    considered layer a subgraph isomorphism problem formulated as a SAT is solved
    by a SAT solver. Each instance is whether it is possible to embed the program
    graph :math:`P` into the effective connectivity graph :math:`C_l` that is
    achieved by applying :math:`l` layers of the swap strategy to the coupling map
    :math:`C_0` of the backend. Since solving SAT problems can be hard, a
    ``time_out`` fixes the maximum time allotted to the SAT solver for each
    instance. If this time is exceeded the considered problem is deemed
    unsatisfiable and the binary search proceeds to the next number of swap
    layers :math:``l``.
    """

    def __init__(self, timeout: int = 60):
        """Initialize the SATMapping.

        Args:
            timeout: The allowed time in seconds for each iteration of the SAT
                solver. This variable defaults to 60 seconds.
        """
        self.timeout = timeout

    def find_initial_mappings(
        self,
        program_graph: rx.PyGraph,
        swap_strategy: SwapStrategy,
        min_layers: int | None = None,
        max_layers: int | None = None,
    ) -> dict[int, SATResult]:
        r"""Find an initial mapping for a given swap strategy. Perform a
        binary search over the number of swap layers, and for each number
        of swap layers solve a subgraph isomorphism problem formulated as
        a SAT problem.

        Args:
            program_graph (rx.PyGraph): The program graph with commuting gates, where
                                        each edge represents a two-qubit gate.
            swap_strategy (SwapStrategy): The swap strategy to use to find the
                                        initial mapping.
            min_layers (int): The minimum number of swap layers to consider.
                                        Defaults to the maximum degree of the
                                        program graph - 2.
            max_layers (int): The maximum number of swap layers to consider.
                                        Defaults to the number of qubits in the
                                        swap strategy - 2.

        Returns:
            dict[int, SATResult]: A dictionary containing the results of the SAT
                                    solver for each number of swap layers.
        """
        num_nodes_g1 = len(program_graph.nodes())
        num_nodes_g2 = swap_strategy.distance_matrix.shape[0]
        if num_nodes_g1 > num_nodes_g2:
            return SATResult(False, [], [], 0)
        if min_layers is None:
            # use the maximum degree of the program graph - 2
            # as the lower bound.
            min_layers = max((d for _, d in program_graph.degree)) - 2
        if max_layers is None:
            max_layers = num_nodes_g2 - 1

        variable_pool = IDPool(start_from=1)
        variables = np.array(
            [
                [variable_pool.id(f"v_{i}_{j}") for j in range(num_nodes_g2)]
                for i in range(num_nodes_g1)
            ],
            dtype=int,
        )
        vid2mapping = {v: idx for idx, v in np.ndenumerate(variables)}
        binary_search_results = {}

        def interrupt(solver):
            # This function is called to interrupt the solver when the
            # timeout is reached.
            solver.interrupt()

        # Make a cnf (conjunctive normal form) for the one-to-one
        # mapping constraint
        cnf1 = []
        for i in range(num_nodes_g1):
            clause = variables[i, :].tolist()
            cnf1.append(clause)
            for k, m in combinations(clause, 2):
                cnf1.append([-1 * k, -1 * m])
        for j in range(num_nodes_g2):
            clause = variables[:, j].tolist()
            for k, m in combinations(clause, 2):
                cnf1.append([-1 * k, -1 * m])

        # Perform a binary search over the number of swap layers to find the
        # minimum number of swap layers that satisfies the subgraph isomorphism
        # problem.
        while min_layers < max_layers:
            num_layers = (min_layers + max_layers) // 2

            # Create the connectivity matrix. Note that if the swap strategy
            # cannot reach full connectivity then its distance matrix will have
            # entries with -1. These entries must be treated as False.
            d_matrix = swap_strategy.distance_matrix
            connectivity_matrix = (
                (-1 < d_matrix) & (d_matrix <= num_layers)
            ).astype(int)
            # Make a cnf for the adjacency constraint
            cnf2 = []
            for e_0, e_1 in list(program_graph.edge_list()):
                clause_matrix = np.multiply(
                    connectivity_matrix, variables[e_1, :]
                )
                clause = np.concatenate(
                    (
                        [[-variables[e_0, i]] for i in range(num_nodes_g2)],
                        clause_matrix,
                    ),
                    axis=1,
                )
                # Remove 0s from each clause
                cnf2.extend([c[c != 0].tolist() for c in clause])

            cnf = CNF(from_clauses=cnf1 + cnf2)

            with Solver(bootstrap_with=cnf, use_timer=True) as solver:
                # Solve the SAT problem with a timeout.
                # Timer is used to interrupt the solver when the
                # timeout is reached.
                timer = Timer(self.timeout, interrupt, [solver])
                timer.start()
                status = solver.solve_limited(expect_interrupt=True)
                timer.cancel()
                # Get the solution and the elapsed time.
                sol = solver.get_model()
                e_time = solver.time()

                print(
                    f"Layers: {num_layers}, Status: {status}, Time: {e_time}"
                )
                if status:
                    # If the SAT problem is satisfiable, convert the solution
                    # to a mapping.
                    mapping = [vid2mapping[idx] for idx in sol if idx > 0]
                    binary_search_results[num_layers] = SATResult(
                        status, sol, mapping, e_time
                    )
                    max_layers = num_layers
                else:
                    # If the SAT problem is unsatisfiable, return the last
                    # satisfiable solution.
                    binary_search_results[num_layers] = SATResult(
                        status, sol, [], e_time
                    )
                    min_layers = num_layers + 1

        return binary_search_results

    def remap_graph_with_sat(
        self, graph: rx.PyGraph, swap_strategy, max_layers
    ):
        """Applies the SAT mapping.

        Args:
            graph (nx.Graph): The graph to remap.
            swap_strategy (SwapStrategy): The swap strategy to use
                                            to find the initial mapping.

        Returns:
            tuple: A tuple containing the remapped graph, the edge map, and the
            number of layers of the swap strategy that was used to find the
            initial mapping. If no solution is found then the tuple contains
            None for each element. Note the returned edge map `{k: v}` means that
            node `k` in the original graph gets mapped to node `v` in the
            Pauli strings.
        """
        num_nodes = len(graph.nodes())
        results = self.find_initial_mappings(
            graph, swap_strategy, 0, max_layers
        )
        solutions = [k for k, v in results.items() if v.satisfiable]

        if len(solutions):
            min_k = min(solutions)
            edge_map = dict(results[min_k].mapping)
            # Create the remapped graph
            remapped_graph = rx.PyGraph()
            remapped_graph.add_nodes_from(range(num_nodes))
            mapping = dict(results[min_k].mapping)
            for i, graph_edge in enumerate(list(graph.edge_list())):
                remapped_edge = tuple(mapping[node] for node in graph_edge)
                remapped_graph.add_edge(*remapped_edge, graph.edges()[i])
            return remapped_graph, edge_map, min_k
        else:
            return None, None, None

sm = SATMapper(timeout=10)
remapped_graph, edge_map, min_swap_layers = sm.remap_graph_with_sat(
    graph=graph_100, swap_strategy=swap_strategy, max_layers=1
)
print("Map from old to new nodes: ", edge_map)
print("Min SWAP layers:", min_swap_layers)
draw_graph(remapped_graph, node_size=200, with_labels=True, width=1)

Output:

Layers: 0, Status: True, Time: 0.028200000000000003
Map from old to new nodes:  {0: 0, 1: 1, 2: 2, 3: 3, 4: 4, 5: 5, 6: 6, 7: 7, 8: 8, 9: 9, 10: 10, 11: 11, 12: 12, 13: 13, 14: 14, 15: 15, 16: 16, 17: 17, 18: 18, 19: 19, 20: 20, 21: 21, 22: 22, 23: 23, 24: 24, 25: 25, 26: 26, 27: 27, 28: 28, 29: 29, 30: 30, 31: 31, 32: 32, 33: 33, 34: 34, 35: 35, 36: 36, 37: 37, 38: 38, 39: 39, 40: 40, 41: 41, 42: 42, 43: 43, 44: 44, 45: 45, 46: 46, 47: 47, 48: 48, 49: 49, 50: 50, 51: 51, 52: 52, 53: 53, 54: 54, 55: 55, 56: 56, 57: 57, 58: 58, 59: 59, 60: 60, 61: 61, 62: 62, 63: 63, 64: 64, 65: 65, 66: 66, 67: 67, 68: 68, 69: 69, 70: 70, 71: 71, 72: 72, 73: 73, 74: 74, 75: 75, 76: 76, 77: 77, 78: 78, 79: 79, 80: 80, 81: 81, 82: 82, 83: 83, 84: 84, 85: 85, 86: 86, 87: 87, 88: 88, 89: 89, 90: 90, 91: 91, 92: 92, 93: 93, 94: 94, 95: 95, 96: 96, 97: 97, 98: 98, 99: 99}
Min SWAP layers: 0

# Repeat the mapping for the remapped graph
remapped_problem = Maxcut(remapped_graph).to_optimization_problem()
remapped_qubo = OptimizationProblemToQubo().convert(remapped_problem)
remapped_cost_operator, remapped_offset = to_ising(remapped_qubo)
print(remapped_cost_operator)

Output:

SparsePauliOp(['IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZ', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZI', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIIIIZIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIZIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIZIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIZIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIZIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIZIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIIIIZIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIIIIIIIZIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIIIIIIZIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIZIIIIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIZIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIZIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIZIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIZIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIZIIIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIZIIIIIIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIZIIIIIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIZIIIIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IZIIIIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'IIIIZZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII', 'ZIIIZIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII'],
              coeffs=[0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j,
 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j,
 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j,
 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j,
 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j,
 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j,
 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j,
 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j,
 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j,
 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j,
 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j,
 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j,
 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j,
 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j, 0.5+0.j])

Construye un circuito QAOA con la estrategia SWAP y el mapeo SAT

Sólo queremos aplicar las estrategias SWAP a la capa del operador de costes, así que empezamos creando el bloque aislado que luego transformaremos y añadiremos al circuito QAOA final.

Para ello, podemos utilizar la QAOAAnsatz clase de Qiskit. Introducimos un circuito vacío en los initial_state campos y mixer_operator para asegurarnos de que estamos creando una capa de operadores de coste aislada. También definimos la edge_coloring mapa de manera que las puertas RZZ queden situadas junto a las puertas SWAP. Esta ubicación estratégica nos permite aprovechar las cancelaciones de CX, optimizando el circuito para obtener un mejor rendimiento. Este proceso se ejecuta dentro de la create_qaoa_swap_circuit función.

def make_meas_map(circuit: QuantumCircuit) -> dict:
    """Return a mapping from qubit index (the key) to classical bit (the value).

    This allows us to account for the swapping order introduced by the
    SWAP strategy.
    """
    creg = circuit.cregs[0]
    qreg = circuit.qregs[0]

    meas_map = {}
    for inst in circuit.data:
        if inst.operation.name == "measure":
            meas_map[qreg.index(inst.qubits[0])] = creg.index(inst.clbits[0])

    return meas_map


def apply_swap_strategy(
    circuit: QuantumCircuit,
    swap_strategy: SwapStrategy,
    edge_coloring: dict[tuple[int, int], int] | None = None,
) -> QuantumCircuit:
    """Transpile with a SWAP strategy.

    Returns:
        A quantum circuit transpiled with the given swap strategy.
    """

    pm_pre = PassManager(
        [
            FindCommutingPauliEvolutions(),
            Commuting2qGateRouter(
                swap_strategy,
                edge_coloring,
            ),
        ]
    )
    return pm_pre.run(circuit)


def apply_qaoa_layers(
    cost_layer: QuantumCircuit,
    meas_map: dict,
    num_layers: int,
    gamma: list[float] | ParameterVector = None,
    beta: list[float] | ParameterVector = None,
    initial_state: QuantumCircuit = None,
    mixer: QuantumCircuit = None,
):
    """Applies QAOA layers to construct circuit.

    First, the initial state is applied. If `initial_state` is None, we begin in
    the initial superposition state. Next, we alternate between layers of the
    cost operator and the mixer. The cost operator is alternatively applied in
    order and in reverse instruction order. This allows us to apply the swap
    strategy on odd `p` layers and undo the swap strategy on even `p` layers.
    """

    num_qubits = cost_layer.num_qubits
    new_circuit = QuantumCircuit(num_qubits, num_qubits)

    if initial_state is not None:
        new_circuit.append(initial_state, range(num_qubits))
    else:
        # all h state by default
        new_circuit.h(range(num_qubits))

    if gamma is None or beta is None:
        gamma = ParameterVector("γ'", num_layers)
        if mixer is None or mixer.num_parameters == 0:
            beta = ParameterVector("β'", num_layers)
        else:
            beta = ParameterVector("β'", num_layers * mixer.num_parameters)

    if mixer is not None:
        mixer_layer = mixer
    else:
        mixer_layer = QuantumCircuit(num_qubits)
        mixer_layer.rx(-2 * beta[0], range(num_qubits))

    for layer in range(num_layers):
        bind_dict = {cost_layer.parameters[0]: gamma[layer]}
        cost_layer_ = cost_layer.assign_parameters(bind_dict)
        bind_dict = {
            mixer_layer.parameters[i]: beta[layer + i]
            for i in range(mixer_layer.num_parameters)
        }
        layer_mixer = mixer_layer.assign_parameters(bind_dict)

        if layer % 2 == 0:
            new_circuit.append(cost_layer_, range(num_qubits))
        else:
            new_circuit.append(cost_layer_.reverse_ops(), range(num_qubits))

        new_circuit.append(layer_mixer, range(num_qubits))

    for qidx, cidx in meas_map.items():
        new_circuit.measure(qidx, cidx)

    return new_circuit


def create_qaoa_swap_circuit(
    cost_operator: SparsePauliOp,
    swap_strategy: SwapStrategy,
    edge_coloring: dict = None,
    theta: list[float] = None,
    qaoa_layers: int = 1,
    initial_state: QuantumCircuit = None,
    mixer: QuantumCircuit = None,
):
    """Create the circuit for QAOA.

    Notes: This circuit construction for QAOA works for quadratic terms in `Z`
    and will be extended to first-order terms in `Z`.
    Higher-orders are not supported.

    Args:
        cost_operator: the cost operator.
        swap_strategy: selected swap strategy
        edge_coloring: A coloring of edges that should correspond to the
            coupling map of the hardware. It defines the order in which
            we apply the Rzz gates. This allows us to choose an ordering
            such that `Rzz` gates will immediately precede SWAP gates
            to leverage CNOT cancellation.
        theta: The QAOA angles.
        qaoa_layers: The number of layers of the cost operator and the
        mixer operator.
        initial_state: The initial state on which we apply layers of
            cost operator and mixer.
        mixer: The QAOA mixer. It will be applied as is onto the QAOA
            circuit. Therefore, its output must have the same ordering
            of qubits as its input.
    """

    num_qubits = cost_operator.num_qubits

    if theta is not None:
        gamma = theta[: len(theta) // 2]
        beta = theta[len(theta) // 2 :]
        qaoa_layers = len(theta) // 2
    else:
        gamma = beta = None

    # First, create the ansatz of one layer of QAOA without mixer
    cost_layer = QAOAAnsatz(
        cost_operator,
        reps=1,
        initial_state=QuantumCircuit(num_qubits),
        mixer_operator=QuantumCircuit(num_qubits),
    ).decompose()

    # This will allow us to recover the permutation of the measurements
    # that the swaps introduce.
    cost_layer.measure_all()

    # Now, apply the swap strategy for commuting gates
    cost_layer = apply_swap_strategy(cost_layer, swap_strategy, edge_coloring)

    # Compute the measurement map (qubit to classical bit).
    # We will apply this for odd layers where the swaps were inserted.
    if qaoa_layers % 2 == 1:
        meas_map = make_meas_map(cost_layer)
    else:
        meas_map = {idx: idx for idx in range(num_qubits)}

    cost_layer.remove_final_measurements()

    # Finally, introduce the mixer circuit and add measurements
    # following measurement map
    circuit = apply_qaoa_layers(
        cost_layer, meas_map, qaoa_layers, gamma, beta, initial_state, mixer
    )

    return circuit

# We can define the edge_coloring map so that RZZ gates are positioned
# right before SWAP gates to exploit CX cancellations
# We use greedy edge coloring from rustworkx to color the edges of
# the graph. This coloring is used to order the RZZ gates
# in the circuit.

edge_coloring_idx = rx.graph_greedy_edge_color(graph_100)
edge_coloring = {
    edge: edge_coloring_idx[idx]
    for idx, edge in enumerate(list(graph_100.edge_list()))
}
edge_coloring = {tuple(sorted(k)): v for k, v in edge_coloring.items()}

qaoa_circ = create_qaoa_swap_circuit(
    remapped_cost_operator,
    swap_strategy,
    edge_coloring=edge_coloring,
    qaoa_layers=1,
)
qaoa_circ.draw(output="mpl", fold=False)

Output:

Paso 3: Ejecutar utilizando primitivas de « IBM Quantum »

Ahora vamos a prepararnos para la ejecución en hardware. Nuestro primer paso consistirá en definir una función de coste del valor en riesgo condicional ( CVaR ), que se introdujo en [3] para su uso en el marco de los algoritmos de optimización cuántica variacional.

La CVaR de una variable aleatoria $X$ para un nivel de confianza $α ∈ (0, 1]$ se define como $CVaR_{\alpha}(X) = \mathbb{E} \lbrack X | X \leq F_X^{-1}(\alpha) \rbrack$ donde $F_X^{-1}(p)$ es la función de distribución acumulativa inversa de $X$ . En otras palabras, CVaR es el valor esperado de la cola inferior $\alpha$ de la distribución de $X$ .

pass_manager = generate_preset_pass_manager(
    backend=backend,
    optimization_level=3,
)

transpiled_qaoa_circ = pass_manager.run(qaoa_circ)

# Utility functions for the evaluation of the expectation value of a measured state
# In this code, for optimization, the measured state is converted into a bit string,
# and the sign of the value is determined by taking the exclusive OR of the bits
# corresponding to Pauli Z.

_PARITY = np.array(
    [-1 if bin(i).count("1") % 2 else 1 for i in range(256)],
    dtype=np.complex128,
)


def evaluate_sparse_pauli(state: int, observable: SparsePauliOp) -> complex:
    """Utility for the evaluation of the expectation value
        of a measured state.

    Args:
        state (int): The measured state.
        observable (SparsePauliOp): The observable to evaluate the
        expectation value for.

    Returns:
        complex: The expectation value of the measured state.
    """
    packed_uint8 = np.packbits(
        observable.paulis.z, axis=1, bitorder="little"
    )  # convert observable to array with 8 bit integer
    state_bytes = np.frombuffer(
        state.to_bytes(packed_uint8.shape[1], "little"),
        dtype=np.uint8,  # convert bitstring to array with 8 bit integer
    )
    reduced = np.bitwise_xor.reduce(packed_uint8 & state_bytes, axis=1)
    # take bitwise xor of the result of 'and' conditional on the
    # above two, return 0 or 1
    return np.sum(observable.coeffs * _PARITY[reduced])

def qaoa_sampler_cost_fun(
    params, ansatz, hamiltonian, sampler, aggregation=None
):
    """Standard sampler-based QAOA cost function to be plugged into
        optimizer routines.

    Args:
        params (np.ndarray): Parameters for the ansatz.
        ansatz (QuantumCircuit): Ansatz circuit.
        hamiltonian (SparsePauliOp): Hamiltonian to be minimized.
        sampler (QAOASampler): Sampler to be used.
        aggregation (Callable | float | None): Aggregation function
            to be applied to the sampled results. If None, the sum
            of the expectation values is returned.
            If float, the CVaR with the given alpha is used.
    """
    # Run the circuit
    job = sampler.run([(ansatz, params)])
    sampler_result = job.result()
    sampled_int_counts = sampler_result[
        0
    ].data.c.get_int_counts()  # bitstrings are stored as integers
    shots = sum(sampled_int_counts.values())
    int_count_distribution = {
        key: val / shots for key, val in sampled_int_counts.items()
    }

    # a dictionary containing: {state: (measurement probability, value)}
    evaluated = {
        state: (
            probability,
            np.real(evaluate_sparse_pauli(state, hamiltonian)),
        )
        for state, probability in int_count_distribution.items()
    }

    # If aggregation is None, return the sum of the expectation values.
    # If aggregation is a float, return the CVaR with the given alpha.
    # Otherwise, use the aggregation function.
    if aggregation is None:
        result = sum(
            probability * value for probability, value in evaluated.values()
        )
    elif isinstance(aggregation, float):
        cvar_aggregation = _get_cvar_aggregation(aggregation)
        result = cvar_aggregation(evaluated.values())
    else:
        result = aggregation(evaluated.values())

    global iter_counts, result_dict
    iter_counts += 1
    temp_dict = {}
    temp_dict["params"] = params.tolist()
    temp_dict["cvar_fval"] = result
    temp_dict["fval"] = sum(
        probability * value for probability, value in evaluated.values()
    )
    temp_dict["distribution"] = sampled_int_counts
    temp_dict["evaluated"] = evaluated
    result_dict[iter_counts] = temp_dict
    print(f"Iteration {iter_counts}: {result}")

    return result


def _get_cvar_aggregation(alpha: float | None) -> Callable:
    """Return the CVaR aggregation function with the given alpha.

    Args:
        alpha (float | None): Alpha value for the CVaR aggregation.
            If None, 1 is used by default.
    Raises:
        ValueError: If alpha is not in [0, 1].
    """
    if alpha is None:
        alpha = 1
    elif not 0 <= alpha <= 1:
        raise ValueError(f"alpha must be in [0, 1], but {alpha} was given.")

    def cvar_aggregation(
        objective_dict: Iterable[tuple[float, float]],
    ) -> float:
        """Return the CVaR of the given measurements.
        Args:
            objective_dict (Iterable[tuple[float, float]]): An iterable
                of tuples containing the measured bit string and the
                objective value based on the bit string.

        """
        sorted_measurements = sorted(objective_dict, key=lambda x: x[1])
        # accumulate the probabilities until alpha is reached
        accumulated_percent = 0.0
        cvar = 0.0
        for probability, value in sorted_measurements:
            cvar += value * min(probability, alpha - accumulated_percent)
            accumulated_percent += probability
            if accumulated_percent >= alpha:
                break
        return cvar / alpha

    return cvar_aggregation

El método « CVaR » puede utilizarse como técnica de mitigación de errores, tal y como se ha comentado anteriormente [4]. En este ejemplo, determinamos el valor de « $\alpha$ » y el número de disparos en función del error por puerta en capas (EPLG) asociado al circuito.

num_2q_ops = transpiled_qaoa_circ.count_ops()[
    "cz"
]  # the two qubit gates on our backend are cz's.

for el in backend.properties().general:
    if el.name[:2] == "lf" and el.name[3:] == str(
        n
    ):  # pick out lf_100, lf of the best 100q chain
        lf = el.value  # layer fidelity
        print("layer fidelity", lf)
        eplg = 1 - lf ** (1 / (n - 1))  # error per layered gate (EPLG)
        fid_cz = 1 - eplg
        gamma_cz = 1 / fid_cz**2
        gamma_circ = gamma_cz**num_2q_ops

cvar_aggregation = 1 / np.sqrt(gamma_circ)
print("")
print("The corresponding CVaR aggregation value is: ", cvar_aggregation)
print(
    "To mitigate the twirled noise, increase shots by a factor of",
    np.sqrt(gamma_circ),
)

Output:

layer fidelity 0.6564844994021044

The corresponding CVaR aggregation value is:  0.3891703406279992
To mitigate the twirled noise, increase shots by a factor of 2.569568889515868

iter_counts = 0
result_dict = {}
init_params = [np.pi, np.pi / 2]

with Session(backend=backend) as session:
    sampler = Sampler(mode=session)
    sampler.options.default_shots = int(1000 / cvar_aggregation)
    sampler.options.dynamical_decoupling.enable = True
    sampler.options.dynamical_decoupling.sequence_type = "XY4"
    sampler.options.twirling.enable_gates = True
    sampler.options.twirling.enable_measure = True
    sampler.options.environment.job_tags = [
        "TUT_AQAOA"
    ]  # add tag for your job execution

    result = minimize(
        qaoa_sampler_cost_fun,
        init_params,
        args=(
            transpiled_qaoa_circ,
            remapped_cost_operator,
            sampler,
            cvar_aggregation,
        ),
        method="COBYLA",
        tol=1e-2,
    )
print(result)

Output:

Iteration 1: -5.0037757181682485

Iteration 2: -3.6969291788278587

Iteration 3: -5.124802512886587

Iteration 4: -3.2428286432730484

Iteration 5: -2.299841265578376

Iteration 6: -8.069790333468909

Iteration 7: -5.9319812181566265

Iteration 8: -7.16681181353234

Iteration 9: -7.060788340473292

Iteration 10: -8.50588688303254

Iteration 11: -8.968767967629661

Iteration 12: -8.461877139498597

Iteration 13: -9.42786961040421

Iteration 14: -10.0667896691371

Iteration 15: -10.140806055989648

Iteration 16: -9.700708620650207

Iteration 17: -10.226825100169627

Iteration 18: -10.124802512886392

Iteration 19: -10.02678081137897

Iteration 20: -10.384860088314245

Iteration 21: -10.111799634115

Iteration 22: -9.73571637118857

Iteration 23: -10.269834622259618

Iteration 24: -10.152808713317091

Iteration 25: -10.036783025818504

 message: Return from COBYLA because the trust region radius reaches its lower bound.
 success: True
  status: 0
     fun: -10.384860088314245
       x: [ 2.708e+00  2.304e+00]
    nfev: 25
   maxcv: 0.0

Paso 4: Procesamiento posterior y devolución del resultado en el formato clásico deseado

Ahora visualicemos nuestros resultados y, a continuación, procesémoslos para hallar el valor del corte.

from matplotlib import pyplot as plt

plt.figure(figsize=(12, 6))
plt.plot(
    [result_dict[i]["cvar_fval"] for i in range(1, iter_counts + 1)],
    label="CVaR",
)
plt.plot(
    [result_dict[i]["fval"] for i in range(1, iter_counts + 1)],
    label="Standard",
)
plt.legend()
plt.xlabel("Iteration")
plt.ylabel("Cost")
plt.show()

Output:

Lo siguiente permite obtener la mejor solución a partir de las cadenas de bits muestreadas:

# Sort the result_dict[iter_counts]['evaluated'] by the CVaR value
sorted_result_dict = [
    (k, v)
    for k, v in sorted(
        result_dict[iter_counts]["evaluated"].items(),
        key=lambda item: item[1][1],
    )
]
print(
    f"bitstring (int): {sorted_result_dict[0][0]}, "
    f"probability: {sorted_result_dict[0][1][0]}, "
    f"objective value: {sorted_result_dict[0][1][1]}"
)

Output:

bitstring (int): 759911855122641592311074830125, probability: 0.00038925652004671076, objective value: -22.5

Consideremos el hamiltoniano $H_C$ para el problema del corte máximo. Supongamos que cada vértice del grafo se asocia a un qubit en el estado $|0\rangle$ o $|1\rangle$ , donde el valor indica el conjunto al que pertenece el vértice. El objetivo del problema es maximizar el número de aristas $(v_1, v_2)$ para las que se cumple que $v_1 = |0\rangle$ y $v_2 = |1\rangle$ , o viceversa. Si asociamos el operador « $Z$ » a cada qubit, donde

Z|0\rangle = |0\rangle \qquad Z|1\rangle = -|1\rangle,

Entonces, un borde $(v_1, v_2)$ pertenece al corte si el valor propio de $(Z_1|v_1\rangle) \cdot (Z_2|v_2\rangle) = -1$ ; en otras palabras, los qubits asociados a $v_1$ y $v_2$ son diferentes. Del mismo modo, $(v_1, v_2)$ no forma parte del corte si el valor propio de $(Z_1|v_1\rangle) \cdot (Z_2|v_2\rangle) = 1$ .

El código generado OptimizationProblem por Optimization Mapper evalúa directamente el valor de corte de una cadena de bits, por lo que no es necesario escribir este cálculo manualmente. De forma equivalente, el valor de corte es $-(\langle H_C \rangle + \text{offset})$ .

def to_bitstring(integer, num_bits):
    result = np.binary_repr(integer, width=num_bits)
    return [int(digit) for digit in result]


bitstring = to_bitstring(
    sorted_result_dict[0][0], remapped_problem.get_num_vars()
)
bitstring = bitstring[::-1]
print(f"Result bitstring (binary) : {bitstring}")

cut_value = remapped_problem.objective.evaluate(np.array(bitstring))
print(f"The value of the cut is: {cut_value}")
print(
    "Cut value from the objective value:",
    -(sorted_result_dict[0][1][1] + remapped_offset),
)

Output:

Result bitstring (binary) : [1, 0, 1, 1, 0, 1, 0, 0, 1, 1, 1, 0, 1, 1, 0, 1, 0, 0, 1, 0, 0, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 1, 0, 1, 0, 1, 1, 1, 1, 0, 0, 0, 1, 1, 0, 0, 1, 0, 1, 0, 0, 1, 0, 1, 0, 1, 0, 1, 0, 0, 1, 0, 0, 0, 1, 1, 0, 0, 1, 0, 0, 1, 0, 1, 0, 1, 1, 0, 1, 0, 0, 0, 0, 1, 0, 1, 1, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 0, 0, 1]
The value of the cut is: 78.0
Cut value from the objective value: 78.0

Por último, dibujemos un gráfico basado en el resultado de CVaR. Dividimos los nodos del grafo en dos conjuntos en función del resultado de CVaR. Los nodos del primer conjunto están coloreados en gris, y los del segundo, en morado. Las aristas entre los dos conjuntos son las aristas cortadas por la partición.

def plot_result(G, x):
    colors = ["tab:grey" if i == 0 else "tab:purple" for i in x]
    pos, _default_axes = rx.spring_layout(G), plt.axes(frameon=True)
    rx.visualization.mpl_draw(
        G,
        node_color=colors,
        node_size=150,
        alpha=0.8,
        pos=pos,
        with_labels=True,
        width=1,
    )


plot_result(graph_100, to_bitstring(sorted_result_dict[0][0], 100)[::-1])

Output:

Referencias

[1] Weidenfeller, J., Valor, L. C., Gacon, J., Tornow, C., Bello, L., Woerner, S., y Egger, D. J. (2022). Escalado del algoritmo de optimización aproximada cuántica en hardware basado en qubits superconductores. Quantum, 6, 870.

[2] Matsuo, A., Yamashita, S., y Egger, D. J. (2023). Una aproximación SAT al problema de mapeo inicial en la inserción de puertas SWAP para puertas conmutantes. IEICE Transactions on Fundamentals of Electronics, Communications and Computer Sciences, 106(11), 1424-1431.

[3] Barkoutsos, P. K., Nannicini, G., Robert, A., Tavernelli, I., y Woerner, S. (2020). Mejora de la optimización cuántica variacional mediante CVaR. Quantum, 4, 256.

[4] Barron, S. V., Egger, D. J., Pelofske, E., Bärtschi, A., Eidenbenz, S., Lehmkuehler, M., y Woerner, S. (2023). Límites demostrables para valores de expectativas sin ruido calculados a partir de muestras ruidosas. arXiv preimpresión arXiv:2312.00733.

Próximos pasos

Recomendaciones

Si te ha parecido interesante este trabajo, quizá te interese el siguiente material:

El decatlón de los problemas difíciles : una lista de diez problemas de optimización que resultan complicados para los algoritmos de optimización clásicos y que pueden constituir buenos casos prácticos para poner a prueba las técnicas presentadas en este tutorial.
Un repositorio de buenas prácticas para la optimización cuántica, con el fin de mejorar aún más los resultados de tu flujo de trabajo basado en QAOA.

¿Le ha resultado útil esta página?

Informe de un error, de una errata o solicite contenido en GitHub.