Wrong Magic Is Worse Than No Magic

Plain-language explainer for doi:10.5281/zenodo.20541665

The central idea in one sentence

Three different quantum gates produce states with identical negativity and identical error-correction syndromes — yet they are maximally different in their computational usefulness, and using the wrong one actively degrades performance below the classical baseline.

Why do we care?

The standard metric for quantum magic is a single number — Wigner negativity $\mathcal{N}(\rho)$. Magic state factories accept states above a negativity threshold and discard the rest. This is analogous to checking the voltage of a battery but not which terminal is positive.

The syndrome blindness theorem (proved in this paper): the three controlled-S states

\[\mathrm{CS}_{01}|{+++}\rangle, \quad \mathrm{CS}_{02}|{+++}\rangle, \quad \mathrm{CS}_{12}|{+++}\rangle\]

satisfy:

  1. Equal total Wigner negativity: $\mathcal{N} = 1/8$ for all three
  2. Identical stabiliser syndromes: all 7 GHZ stabiliser expectation values equal $+1$
  3. Maximally separated orbit valence labels — they live in completely different orbits

A routing or calibration bug that applies $\mathrm{CS}{02}$ when $\mathrm{CS}{01}$ was intended produces a state that passes every standard quality check. Yet that state is guaranteed to fail the intended computation — and will score below the classical baseline (0.472 vs 0.500) in the Fano Line Verification Game.

Wrong magic is worse than no magic.

What is a CS gate and why does its identity matter?

The controlled-S gate $\mathrm{CS}{ij}$ applies a phase of $i$ to qubit $j$ conditioned on qubit $i$ being in state $\lvert 1\rangle$. On a 3-qubit register, there are three choices: $\mathrm{CS}{01}$, $\mathrm{CS}{02}$, $\mathrm{CS}{12}$.

These are physically distinct operations on hardware with a fixed coupling map (as on all current superconducting processors). A routing bug or miscalibration can apply the wrong one. Standard fidelity benchmarking measures total gate error rate — it cannot detect a gate that works perfectly but is the wrong gate.

The orbit valence label detects exactly this failure mode.

The CS magic composition theorem

When two CS gates are applied in sequence, their orbit valences combine via the XOR-Fano rule:

\[\mathrm{CS}_{ij} \cdot \mathrm{CS}_{kl}: \quad L_{\text{out}} = L_{ij} \oplus L_{kl}\]

where $\oplus$ is XOR and the orbit labels follow the lines of a second Fano plane on the orbit indices. This is the CS Magic Composition Theorem — the orbit valence algebra is closed under the XOR-Fano rule.

Consequences:

  • Two CS gates from the same orbit cancel: $L \oplus L = 0$ (stabiliser state, zero magic)
  • Two CS gates from adjacent orbits produce a specific third orbit
  • The full multiplication table is the 7-line structure of the second Fano plane

This means that a quantum compiler can track magic valence through a circuit algebraically, using only XOR arithmetic on 3-bit labels — no Wigner function computation needed at runtime.

The Fano Line Verification Game

The paper introduces the Fano Line Verification Game as an operational test of magic valence. Two separated players share a state and receive input questions corresponding to Fano lines. They win if their measurement results are consistent with a Fano-line structure.

Classical players (using no entanglement) win with probability at most $5/7 \approx 0.714$. Players sharing a GHZ state win with probability $1.000$ (perfect). Players sharing a wrong-valence magic state score $0.472$ — below the classical bound of $0.500$.

This is the operational meaning of “wrong magic is worse than no magic”: it actively anti-correlates the players’ answers, doing worse than independent random guessing.

Why Gottesman-Knill doesn’t see this

The Gottesman-Knill theorem says: any quantum circuit consisting entirely of Clifford gates (H, CNOT, S, CZ) can be efficiently simulated classically. A magic state (non-Clifford resource) breaks this.

But Gottesman-Knill treats all magic states as equivalent — it draws a line between “simulable” (no magic) and “not simulable” (some magic). The orbit valence structure shows this line is too coarse. Within the non-simulable region, orbit valence further partitions states by which computations they enable. A state in orbit $\mathcal{O}1$ can enable $\mathrm{CS}{01}$-type computations but is actively harmful for $\mathrm{CS}_{02}$-type computations.

The correct simulability boundary is not a binary Clifford/non-Clifford line — it is a 7-valued Fano-orbit classifier.

Practical implication: the ORBIT opcode

The 7 GHZ stabiliser expectation values $\langle ZZI\rangle, \langle ZIZ\rangle, \langle IZZ\rangle, \langle XXX\rangle, \langle YYX\rangle, \langle YXY\rangle, \langle XYY\rangle$ determine the orbit valence label directly. These are the same measurements already performed at every $[[7,1,3]]$ Steane code error correction cycle.

The ORBIT opcode = Steane syndrome extraction. These two tasks are the same 7 measurements; the QEC and magic-state resource-theory communities have been doing different post-processing on the same bits.

Adding orbit-valence post-processing to an existing Steane-code processor costs zero additional quantum circuit operations. It catches coherent gate-identity errors that syndrome decoders cannot.

For the full technical treatment, see doi:10.5281/zenodo.20541665.