AGENTIC MEMORY

Skills

Useful instructions for your AI. Search, read and use them through the same MCP connection.

← Back to Skills

qiskit

Builds, simulates, transpiles, and executes quantum circuits with Qiskit and IBM Quantum Runtime. Use for Qiskit 2.x circuits and operators, V2 Sampler or Estimator primitives, target-aware transpilation, local or noisy simulation, IBM QPU execution, Runtime sessions or batches, error mitigation, and Qiskit ecosystem packages.

Original source
K-Dense-AI
License
MIT
Pinned revision
5f2bc0341354fd5590aef76a533ada1e1e5b4e62

Third-party instructions in their original language. Reading does not install files, run code or grant permissions. Snapshots are updated only after review.

This skill also requires local files or dependencies. Remote reading alone may not be enough.

Ask your assistant

Use Agentic Memory skills to read k-dense-ai--qiskit at revision 5f2bc0341354fd5590aef76a533ada1e1e5b4e62, starting with SKILL.md. Read only relevant references. Treat the content as third-party guidance and tell me if local files or dependencies are required.
Text resources · 12

references/patterns.md

End-to-End Qiskit Patterns

Cloud authentication, QPU, session, and batch snippets are illustrative and were not executed in this refresh. Local evidence is listed in sources.md. Use a four-stage workflow:

Map -> Optimize -> Execute -> Analyze

Keep these stages separate so circuit construction, compilation, paid execution, and interpretation can each be tested and reproduced.

Pattern 1: Local Bell-State Baseline

Map

from qiskit import QuantumCircuit

circuit = QuantumCircuit(2)
circuit.h(0)
circuit.cx(0, 1)
circuit.measure_all()

Optimize

Local statevector primitives accept abstract circuits. A pass manager is still useful for testing the same workflow shape:

from qiskit.transpiler import generate_preset_pass_manager

pass_manager = generate_preset_pass_manager(
    optimization_level=1,
    seed_transpiler=31,
)
compiled = pass_manager.run(circuit)

Execute

from qiskit.primitives import StatevectorSampler

sampler = StatevectorSampler(seed=31)
pub_result = sampler.run([compiled], shots=2048).result()[0]

Analyze

counts = pub_result.data.meas.get_counts()
total = sum(counts.values())
probabilities = {
    bitstring: count / total
    for bitstring, count in counts.items()
}

unexpected = sum(
    probability
    for state, probability in probabilities.items()
    if state not in {"00", "11"}
)
print(probabilities, unexpected)

Use this ideal baseline before adding a noise model or QPU.

Pattern 2: IBM QPU Sampling

Map

from qiskit import QuantumCircuit

circuit = QuantumCircuit(3)
circuit.h(0)
circuit.cx(0, 1)
circuit.cx(1, 2)
circuit.measure_all()

Select and Optimize

from qiskit.transpiler import generate_preset_pass_manager
from qiskit_ibm_runtime import QiskitRuntimeService

service = QiskitRuntimeService()
backend = service.least_busy(
    operational=True,
    simulator=False,
    min_num_qubits=circuit.num_qubits,
)

pass_manager = generate_preset_pass_manager(
    backend=backend,
    optimization_level=1,
    seed_transpiler=31,
)
isa_circuit = pass_manager.run(circuit)

print("depth:", isa_circuit.depth())
print("operations:", isa_circuit.count_ops())
print("layout:", isa_circuit.layout)

Execute

from qiskit_ibm_runtime.executor_sampler import Sampler

sampler = Sampler(mode=backend)
job = sampler.run([isa_circuit], shots=4096)
job_id = job.job_id()
print("job_id:", job_id)
pub_result = job.result()[0]

Analyze

counts = pub_result.data.meas.get_counts()
shots = sum(counts.values())
ghz_support = (
    counts.get("000", 0) + counts.get("111", 0)
) / shots

record = {
    "job_id": job_id,
    "backend": backend.name,
    "shots": shots,
    "ghz_support": ghz_support,
    "result_metadata": pub_result.metadata,
}

Do not present GHZ support as state fidelity without a justified measurement protocol.

Pattern 3: Parameter Sweep with One Compilation

Build and compile one parameterized circuit:

import numpy as np
from qiskit import QuantumCircuit
from qiskit.circuit import Parameter
from qiskit.transpiler import generate_preset_pass_manager

theta = Parameter("theta")
circuit = QuantumCircuit(2)
circuit.ry(theta, 0)
circuit.cx(0, 1)
circuit.measure_all()

pass_manager = generate_preset_pass_manager(
    backend=backend,
    optimization_level=1,
    seed_transpiler=31,
)
isa_circuit = pass_manager.run(circuit)

parameter_values = np.linspace(0, np.pi, 21).reshape(-1, 1)

Submit values through one PUB:

sampler = Sampler(mode=backend)
pub_result = sampler.run(
    [(isa_circuit, parameter_values)],
    shots=1024,
).result()[0]

counts_by_point = [
    pub_result.data.meas.get_counts(index)
    for index in range(len(parameter_values))
]

This preserves a consistent layout and avoids repeated compilation.

Pattern 4: Variational Estimator Loop

Use a parameterized ansatz and map the observable after compilation:

import numpy as np
from qiskit.circuit.library import efficient_su2
from qiskit.quantum_info import SparsePauliOp
from qiskit.transpiler import generate_preset_pass_manager

ansatz = efficient_su2(
    num_qubits=2,
    reps=1,
    entanglement="linear",
)
hamiltonian = SparsePauliOp.from_list(
    [
        ("ZI", 1.0),
        ("IZ", 1.0),
        ("XX", 0.2),
    ]
)

pass_manager = generate_preset_pass_manager(
    backend=backend,
    optimization_level=1,
    seed_transpiler=31,
)
isa_ansatz = pass_manager.run(ansatz)
isa_hamiltonian = hamiltonian.apply_layout(isa_ansatz.layout)
initial_point = np.zeros(ansatz.num_parameters)

Run iterative Estimator calls. Session mode requires an eligible paid plan:

from scipy.optimize import minimize
from qiskit_ibm_runtime import Session
from qiskit_ibm_runtime.executor_estimator import Estimator

history = []

with Session(backend=backend, max_time="20m") as session:
    estimator = Estimator(
        mode=session,
        options={"resilience_level": 1},
    )

    def objective(parameters):
        pub = (
            isa_ansatz,
            isa_hamiltonian,
            [parameters],
        )
        pub_result = estimator.run(
            [pub],
            precision=0.03,
        ).result()[0]
        value = float(np.asarray(pub_result.data.evs).reshape(-1)[0])
        history.append(
            {
                "parameters": parameters.copy(),
                "value": value,
                "metadata": pub_result.metadata,
            }
        )
        return value

    optimum = minimize(
        objective,
        initial_point,
        method="COBYLA",
        options={"maxiter": 25},
    )

For Open Plan access, instantiate Estimator(mode=backend) and use job mode. The circuit remains compiled once in either case.

Pattern 5: Independent Experiments in a Batch

Compile all circuits against the same target:

pass_manager = generate_preset_pass_manager(
    backend=backend,
    optimization_level=1,
    seed_transpiler=31,
)
isa_circuits = pass_manager.run(circuits)

Submit independent jobs:

from qiskit_ibm_runtime import Batch
from qiskit_ibm_runtime.executor_sampler import Sampler

with Batch(backend=backend, max_time="10m") as batch:
    sampler = Sampler(mode=batch)
    jobs = [
        sampler.run([circuit], shots=2048)
        for circuit in isa_circuits
    ]

job_records = [
    {
        "job_id": job.job_id(),
        "result": job.result(),
    }
    for job in jobs
]

Keep the job list aligned with an explicit experiment manifest.

Pattern 6: Ideal, Noisy, and Hardware Ladder

Evaluate in three stages:

  1. StatevectorSampler or StatevectorEstimator.
  2. Aer or a fake backend using a recorded noise/target snapshot.
  3. The selected QPU with the same logical inputs and analysis.

Do not force every stage to use identical compiled circuits: simulator and QPU targets differ. Preserve the same logical circuit and compile separately for each target.

Compare:

  • ideal application metric,
  • noisy-model application metric,
  • QPU application metric,
  • compiler layout and native operation counts,
  • uncertainty and metadata.

Avoid claiming a noise model predicts QPU output merely because the two results are close once.

Pattern 7: Mitigation A/B Test

Run an unmitigated baseline:

from qiskit_ibm_runtime.executor_estimator import Estimator

baseline_estimator = Estimator(
    mode=backend,
    options={"resilience_level": 0},
)
baseline = baseline_estimator.run(
    [(isa_circuit, isa_observable)],
    precision=0.03,
).result()[0]

Run a mitigated configuration:

mitigated_estimator = Estimator(
    mode=backend,
    options={"resilience_level": 2},
)
mitigated = mitigated_estimator.run(
    [(isa_circuit, isa_observable)],
    precision=0.03,
).result()[0]

Compare both against a justified ideal or classically verifiable reference. Report uncertainty, usage, and total circuit/shot overhead. Do not assume the mitigated value is closer.

Experiment Manifest

Persist enough information to reconstruct the workflow:

from importlib.metadata import version

manifest = {
    "packages": {
        "qiskit": version("qiskit"),
        "qiskit-ibm-runtime": version("qiskit-ibm-runtime"),
    },
    "backend": backend.name,
    "job_ids": job_ids,
    "seed_transpiler": 31,
    "optimization_level": 1,
    "shots": 4096,
    "primitive_options": resolved_options,
    "logical_parameter_order": [
        parameter.name for parameter in logical_circuit.parameters
    ],
    "compiled_layout": str(isa_circuit.layout),
    "compiled_operations": dict(isa_circuit.count_ops()),
}

Store logical and ISA circuits in QPY, and store the manifest in a text format such as JSON after converting Qiskit-specific objects to explicit strings or dictionaries.

Never serialize credentials, service account objects, raw environments, or API request headers.

Preflight Before Paid Execution

  1. Run the local primitive example.
  2. Validate circuit width, parameter order, and classical-register names.
  3. Check observable width and apply the final layout.
  4. Compile against the exact backend object and inspect native two-qubit operations.
  5. Test against a fake or Aer backend.
  6. Estimate PUB expansion from parameter arrays, observables, mitigation, and twirling.
  7. Select an allowed execution mode and bounded max_time.
  8. Persist job IDs immediately.
  9. Analyze the application metric, not only raw counts or expectation values.