Qiskit Magic Metrics

resource-metrics
entanglement
stabilizer-formalism
Exact, stabilizer-accelerated magic and entanglement metrics for Qiskit, as composable PassManager passes
Published

September 2, 2026

STATUS · In Review · 23 passing tests

What it does

Three exact resource/entanglement metrics for Qiskit circuits — stabilizer Rényi entropy (magic), the Meyer-Wallach measure (global multipartite entanglement), and von Neumann entanglement entropy across an arbitrary bipartition — each implemented as a PassManager AnalysisPass that also works as a plain callable. Every metric has two computation paths: a general statevector path for arbitrary circuits, and a stabilizer fast path for Clifford circuits that answers exactly, in polynomial time, using only the stabilizer tableau — no statevector construction, no exponential blowup.

The Meyer-Wallach fast path is the piece I actually derived myself, not sourced from a paper: a qubit’s reduced state is pure if and only if some GF(2) combination of stabilizer generators cancels out on every other qubit, found via a null-space computation over the binary field rather than the usual reduced-density-matrix route.

Why exact, stabilizer-accelerated metrics

Before writing any code, I checked what already existed rather than assume a clean gap. Qurrium/Qurry is a real, actively maintained package that measures magic-adjacent quantities via randomized measurement on real or simulated hardware — the right tool when you have hardware access and want a shot-sampled estimate. qLEET implements expressibility and Meyer-Wallach entangling capability, but looks unmaintained and isn’t PassManager-integrated. Neither combines exact, deterministic computation with stabilizer-tableau shortcuts and PassManager-native composability, which is the actual gap this package fills — narrower than “no ecosystem equivalent exists,” and stated that way in the repo’s own prior-art writeup rather than glossed over.

A Lean 4 proof of the Meyer-Wallach null-space claim — the one derivation above that isn’t already published elsewhere — is planned but not yet built. Tracked openly in the repo’s roadmap, not implied as already done.