Noise Profile

from hilbertbench.analysis import noise_profile

result = noise_profile(trace)

---

What it measures

noise_profile reads the device calibration snapshot recorded at execution time and estimates the hardware noise exposure of the run. It combines:

• Coherence times (T1, T2) from the calibration snapshot
• Readout error per qubit
• Gate error rates
• Circuit structure (depth and gate count from circuit_structure)

to produce an estimated circuit fidelity: the probability that a single execution produces the correct output under the calibrated noise model.

This is a coarse estimate using standard NISQ product formulas — not a full noise simulation. Its value is helping you correlate a loss spike or accuracy drop with device calibration rather than your model alone.

Simulator traces

Traces recorded on statevector simulators (StatevectorEstimator, default.qubit) carry no calibration snapshot. The function returns None-valued noise fields rather than raising an error — degraded mode, not failure.

---

Function signature

noise_profile(
    trace: HilbertTrace | str | Path,
) -> dict

---

Output fields

{
    'status':                   'Moderate Noise',
    'has_calibration':          True,
    'estimated_circuit_fidelity': 0.72,
    't1_us': {
        'mean': 85.3, 'min': 72.1, 'max': 98.4
    },
    't2_us': {
        'mean': 61.7, 'min': 44.2, 'max': 79.3
    },
    'readout_error': {
        'mean': 0.021, 'min': 0.008, 'max': 0.038
    },
    'gate_error': {
        'mean': 0.003, 'min': 0.001, 'max': 0.007
    },
    'num_qubits_calibrated': 2,
}

For simulator traces (no calibration):

{
    'status':                     'No Calibration Data',
    'has_calibration':            False,
    'estimated_circuit_fidelity': None,
    't1_us':        None,
    't2_us':        None,
    'readout_error': None,
    'gate_error':    None,
    'num_qubits_calibrated': 0,
}

Field	Meaning
status	Human-readable noise verdict
has_calibration	Whether a calibration snapshot was found in the trace
estimated_circuit_fidelity	Product-formula estimate of single-run fidelity
t1_us	T1 coherence time statistics (mean/min/max, microseconds)
t2_us	T2 coherence time statistics (mean/min/max, microseconds)
readout_error	Readout error statistics (mean/min/max, probability)
gate_error	Gate error statistics (mean/min/max, probability)
num_qubits_calibrated	Number of qubits with calibration data

---

Verdicts

status	Fidelity	Meaning
Low Noise	≥ 0.90	Device was well-calibrated — noise is unlikely to dominate
Moderate Noise	0.50 – 0.90	Meaningful noise exposure — consider error mitigation
High Noise	< 0.50	Fidelity below 50% — results may be dominated by noise
No Calibration Data	—	Simulator trace or backend did not expose calibration

---

Limitations

• The estimated fidelity uses the product approximation: fidelity ≈ ∏ (1 − gate_error_i) × ∏ (1 − readout_error_j). This is a lower bound under the assumption of independent, Markovian errors. Crosstalk and non-Markovian effects can make the actual fidelity worse.
• Calibration data is a snapshot from the time the job was submitted. Device calibration drifts; results from long jobs may have been collected under different conditions.
• For traces with multiple circuit structures, the dominant circuit (by gate count) is used for the fidelity estimate.

---

Example

from hilbertbench.analysis import noise_profile

result = noise_profile(trace)

if not result['has_calibration']:
    print("Simulator trace — no hardware noise data")
else:
    print(f"Noise verdict:   {result['status']}")
    print(f"Est. fidelity:   {result['estimated_circuit_fidelity']:.1%}")
    print(f"T1 (mean):       {result['t1_us']['mean']:.1f} µs")
    print(f"Readout error:   {result['readout_error']['mean']:.3f}")