qiskit_noise_learning.analysis.LegacySolve

class qiskit_noise_learning.analysis.LegacySolve

Bases: AnalysisStage

Solves for the ModelData using the legacy pair-fidelity method.

This class is more specialized than other model solvers. Where the others are set up to work for general models and path sets, this stage has the following restrictions and limitations:

• It assumes the gate set consists of pure preparation, measurement, and self-inverse unitary gates.
• It assumes the model is a PauliLindbladModel for which the unitary gate noise is 2-local, the measurement noise is 1-local, and preparation is assumed to be perfect. Furthermore, for a given gate, the fidelities of “conjugate Paulis” (Paulis that are mapped to each other under the gate) are assumed to be equal in the fitting process.
• It accepts only EvenDepthVanillaPaths and SPAMPaths paths to analyze.
• The unitary gate model fitting is based exclusively on exponential decay data, and so this stage must run after CurveFitObservables.

The fitting process proceeds by analyzing decay curves for each unitary gate independently, and likewise for any SPAM experiments.

• For the unitary gates a more restricted design matrix, constructed under the equality of fidelities with conjugate Paulis, is constructed and solved. Any fidelities of weight > 2 are not considered.
• For SPAM, as preparation is assumed perfect, all noise is attributed to measurement: the rate of the corresponding single-qubit $X$ generator is calculated directly as $-\ln(F)/2$ for a measured fidelity $F$, clipped at zero, with its variance propagated from that fidelity’s uncertainty.

In both cases, the set of returned generators fit and returned is constructed based on what Paulis appear in the fidelities encountered in the paths.

Further notes:

• Unitary gate rates are reported with zero covariance.
• When the same Pauli appears in more than one path, the fidelity estimates must be equal, otherwise this solver raises an error.

__init__

Methods

input_level

output_level