Hybrid Classical-Quantum Workflows

Hybrid Classical-Quantum Workflows

The "quantum-classical computer" of the NISQ era is a heterogeneous system: a classical processor orchestrates quantum circuit execution on a quantum processing unit (QPU). Designing this split correctly is critical for performance.

The Hybrid Computing Model

The QPU is a specialized accelerator, use it only for the operations that are intractable classically (state preparation, entangling operations, interference).

Latency Budget

Real quantum cloud access adds significant overhead:

Implication: VQE with 1000 optimization iterations × 10 Pauli terms × 2 gradient shifts × 2 min queue time ≈ 667 hours (about 28 days) on real hardware. This is why classical simulators are essential.

Workflow Patterns

Pattern 1: Batch Execution Submit multiple circuits simultaneously (different parameters, different Pauli terms).

# Batch submission, submit all circuits at once
jobs = []
for params in parameter_sweep:
    circuit = build_circuit(params)
    job = backend.run(circuit, shots=1000)
    jobs.append(job)

# Collect results
results = [job.result().get_counts() for job in jobs]

Best for: Computing Pauli term expectation values in VQE (all terms can be batched).

Pattern 2: Asynchronous with Callbacks Non-blocking job submission, classical computation continues while quantum jobs run.

async def run_vqe_step(params):
    # Submit quantum job
    job = await backend.run_async(circuit(params))
    # Classical work while waiting...
    classical_gradient_estimate = classical_heuristic(params)
    # Quantum result arrives
    energy = await collect_energy(job)
    return energy

Best for: Long queue times, can parallelize classical preprocessing.

Pattern 3: Recursive/Adaptive Classical algorithm decides next circuit based on previous quantum results.

# ADAPT-VQE: grow ansatz based on gradient screening
while max_gradient > threshold:
    # Quantum: measure gradients for all candidate operators
    gradients = measure_gradients(current_ansatz)
    # Classical: select operator with largest gradient
    best_op = max(gradients, key=gradients.get)
    # Quantum: add operator to ansatz, optimize
    current_ansatz.append(best_op)
    current_params = optimize_vqe(current_ansatz)

Best for: Adaptive algorithms (ADAPT-VQE, iterative phase estimation).

Data Encoding Strategies

A critical bottleneck: loading classical data into quantum states.