QUBO solver - Haiqu

haiqu.solve_qubo() This notebook demonstrates the haiqu.solve_qubo() API that automatically handles quantum optimization end-to-end. For a 20-variable 6-regular Max-Cut problem (loaded from a deterministic reference graph), the raw quantum results achieve a suboptimal solution. With built-in post-processing, haiqu.solve_qubo() achieves the optimal solution—no additional quantum circuits required. Having a bug or an issue? Submit feedback haiqu.solve_qubo() What does it do? Provides a high-level interface for quantum optimization that automatically builds LR-QAOA circuits, runs them on quantum backends, applies post-processing, and calculates relevant metrics. How do I use it? Pass a QUBO problem, select a backend, configure circuit parameters (number of layers, shots), and optionally enable compression. The function returns comprehensive optimization results. What are the options? device_id – choose backend (simulators or real hardware). p – number of LR-QAOA layers. postprocess_iterations – number of post-processing passes. compression – enable circuit compression. Which options do you recommend? Start with aer_simulator for quick results. To improve solution quality: postprocess_iterations (default 5) – increase for better results at the cost of runtime. p (default 10) – adjust the number of LR-QAOA layers to improve solution quality. compression=True – try enabling when running on hardware to reduce circuit depth and see how that improves the solution quality Initialize the benchmark Import the necessary libraries, initialize the Haiqu SDK, and create a 20-variable optimization problem. We’ll demonstrate the value of built-in post-processing and CVaR analysis.

from haiqu.sdk import haiqu
from haiqu.sdk.optimization import QUBO
from qiskit_optimization.applications import Maxcut
import networkx as nx
import pandas as pd
import json

# Set pandas display options to show full column width
pd.set_option('display.max_colwidth', None)

haiqu.login()
haiqu.init("QUBO Solve Tutorial")

# Load Max-Cut graph from reference JSON file for deterministic results
graph_file = 'maxcut_graph_reference.json'

print(f"Loading reference graph from {graph_file}...")
with open(graph_file, 'r') as f:
    graph_data = json.load(f)

# Reconstruct the graph
G = nx.Graph()
G.add_nodes_from(graph_data['nodes'])
for edge in graph_data['edges']:
    G.add_edge(edge['source'], edge['target'], weight=edge['weight'])


print(f"Max-Cut Problem: {G.number_of_nodes()} nodes, {G.number_of_edges()} edges ({graph_data['parameters']['degree_graph']}-regular graph)")

# Create QUBO
maxcut = Maxcut(G)
qp = maxcut.to_quadratic_program()
problem = QUBO.from_quadratic_program(qp)

Run benchmark scenarios Run haiqu.solve_qubo() once and analyze results with and without post-processing.

# Run QAOA with shallow circuit
p = 1 # number of LR-QAOA layers
shots = 1_000
print(f"Running haiqu.solve_qubo() with p={p}, shots={shots}...\n")

result = haiqu.solve_qubo(
    problem=problem,
    device_id="aer_simulator",
    options={"method": "matrix_product_state"},
    p=p,
    shots=shots,
)

raw_best_cost = min(result.raw_costs.values())
processed_best_cost = min(result.processed_costs.values())

# Create scenarios
scenarios = [
    {
        "description": "Raw Quantum Results",
        "postprocess": False,
        "best_cost": raw_best_cost,
    },
    {
        "description": "With Postprocessing", 
        "postprocess": True,
        "best_cost": processed_best_cost,
    }
]

The haiqu.solve_qubo() function provides comprehensive optimization results. Summary of scenarios:

# baseline optimal solution: note that we always use the qiskit's little-endian convention for bitstrings where for example '01' means q1=0 and q0=1.
optimal_bitstring = '10001110010010101100'

# Calculate optimal cost
optimal_cost = problem.cost(optimal_bitstring)

print(f"True optimal cost (QUBO minimization): {optimal_cost:.2f}\n")

# Build results table
results_data = []

for scenario in scenarios:
    cost = scenario["best_cost"]

    gap = (cost - optimal_cost) / abs(optimal_cost)

    # Calculate normalized approximation ratio (for minimization problems)
    # ρ = 1 + (f(x) - f*) / |f*|
    normalized_approx_ratio = 1 + gap
    
    if gap < 1e-6:
        ratio_with_interpretation = f"{normalized_approx_ratio:.4f} (Optimal)"
    else:
        ratio_with_interpretation = f"{normalized_approx_ratio:.4f} ({gap*100:.1f}% worse than optimal)"
    
    # Determine business impact
    if scenario["description"] == "Raw Quantum Results":
        business_impact = "Demonstrates baseline quantum performance on the optimization problem"
    else:
        business_impact = "Achieves optimal solution at no extra quantum cost"

    results_data.append({
        "Scenario": scenario["description"],
        "Best Cost": f"{cost:.2f}",
        "Normalized Approximation Ratio": ratio_with_interpretation,
        "Business Impact": business_impact,
    })

pd.DataFrame(results_data)

💡 Good to Know: The haiqu.solve_qubo() function automatically applies post-processing and provides advanced metrics such as CVar, providing both raw and refined results in a single call—no additional quantum runs required. Get in Touch Documentation portal docs.haiqu.ai Contact Support feedback.haiqu.ai Follow Us on LinkedIn latest news on LinkedIn Visit Our Website Learn more about Haiqu Inc. on haiqu.ai Business Inquiries Contact us at info@haiqu.ai