Magic (Nonstabilizerness)

Definition

Magic (or nonstabilizerness) is the quantum-information resource that quantifies how far a quantum state is from the set of stabilizer states. Stabilizer states are the states preparable from a reference state using only Clifford operations (the normalizer of the Pauli group); by the Gottesman–Knill theorem they can be simulated classically in polynomial time. Magic is therefore the ingredient a state must have to enable a genuine quantum-computational advantage / universal quantum computation (Bravyi–Kitaev): Clifford + magic = universality.

Magic is a property of a single state, not a correlation between subsystems — this is what distinguishes it sharply from entanglement. A stabilizer state can be maximally entangled (a Bell pair is a stabilizer state) yet has zero magic; conversely magic can be large in a product state. The two resources are independent in general, but linked through one structural fact used heavily in QFT:

Flatness ⟺ stabilizer (necessary condition). For a pure stabilizer state, the reduced density matrix on any region is proportional to a projector, so all its Rényi entropies coincide: the entanglement spectrum is flat, $S_n(A)=\log {\lambda }_A$ independent of the Rényi index $n$, and stays flat under Clifford operations. Hence a non-flat entanglement spectrum is a witness of magic (Tirrito et al. 2024).

Anti-flatness. A convenient $n$-resolved witness is $F_n({\rho }_A)=\mathrm{Tr}({\rho }_A^{n+1})\mathrm{Tr}({\rho }_A^{n-1})-\mathrm{Tr}({\rho }_A^n{)}^2,$ which vanishes iff the spectrum is flat. The $n=2$ case is the most informative: after averaging over a Clifford orbit it reproduces the linearized stabilizer entropy, and it lower-bounds the distance to the stabilizer set, $F_2({\rho }_A)/8\le {\min }_{\sigma \in \mathrm{STAB}}\Vert \psi -\sigma \Vert$ (Cao et al. 2024).

The resource theory in one line

Magic is the entanglement-theory machine with different gears: free states $\to$ the stabilizer polytope $\mathrm{STAB}=\mathrm{conv}\{\text{pure stabilizer states}\}$ (in place of separable states), free operations $\to$ stabilizer protocols (Cliffords, Pauli measurements, stabilizer ancillas, partial trace, feed-forward — exactly the Gottesman–Knill-simulable set, in place of LOCC), and a monotone is any $M$ vanishing on the free set and non-increasing under the free operations. The full datum is therefore not $(\mathcal{M},\rho )$ but $(\mathcal{M},\mathcal{P},{\mathfrak{F}}_{\mathcal{P}},{\mathfrak{O}}_{\mathcal{P}},\rho ),$ and the whole difficulty of exporting magic to QFT is that a net supplies the first and last entries but not the middle three (canonical-free-structure-for-magic-in-qft).

Measure	Domain	Additivity	Computability	Operational meaning
Relative entropy of magic $M_{\mathrm{rel}}={\min }_{\sigma \in \mathrm{STAB}}D(\rho \Vert \sigma )$	mixed	non-additive	hard (convex opt.)	distinguishability from the free set — the modular-native one
Robustness $\mathrm{RoM}$ (${\ell }_1$ of a signed stabilizer expansion)	mixed	sub­multiplicative	LP, few qubits	quasiprobability simulation cost, $N_{\text{samples}}\sim {\mathrm{RoM}}^2$
Stabilizer extent $\xi$	pure	multiplicative (few-qubit)	LP, few qubits	stabilizer-rank simulation cost
Mana $\mathcal{M}=\log \Vert W_{\rho }{\Vert }_1$	odd qudits	additive	easy	Wigner negativity; distillation bounds
Stabilizer Rényi entropy $M_{\alpha }$	pure	additive	easy ($4^n$ Pauli moments)	chaos diagnostics; measurable by randomized measurements

There is no best measure, only a division of labour: $M_{\alpha }$ is the workhorse (tractable, replica-friendly, field-theory portable), $\mathrm{RoM}$/$\xi$/mana carry the sharpest operational meanings, and $M_{\mathrm{rel}}$ is the one that survives the passage to type III. For qubits mana is unavailable — no qubit Wigner function is simultaneously Clifford-covariant and non-negative exactly on the stabilizer states.

$M_{\alpha }$ deserves its own gloss, because it is the one that reaches field theory. With $q_{\psi }(P)=\frac{1}{d}⟨\psi |P|\psi {⟩}^2$ the Pauli probability distribution (normalized by Parseval, ${\sum }_P⟨P{⟩}^2=d$), $M_{\alpha }(\psi )=H_{\alpha }(q_{\psi })-\log d$; a stabilizer state has $q_{\psi }$ uniform on the $d$ elements of its stabilizer group, so $M_{\alpha }=0$. Magic is thus the delocalization of the state in the Pauli frame — operator-space participation — which is why it couples naturally to operator spreading and chaos, and why Hoshino–Oshikawa–Ashida can write it as a replicated path integral with a line defect whose universal constant is an Affleck–Ludwig $g$ factor.

Role in Research

Magic entered this wiki through Benedetti–Dabholkar–Dalmonte (2026), which proves that every physically relevant state of a local QFT necessarily has non-zero magic. The result is not a lattice statement dressed up — it is forced by the same algebraic structure the group’s whole program relies on:

• By the Reeh–Schlieder theorem, the vacuum and states resembling it at short distances are cyclic and separating for every local algebra, hence define faithful normal states.
• By the Bisognano–Wichmann theorem, the vacuum wedge modular operator is ${\Delta }_{\Omega }=e^{-2\pi K}$ ($K$ = boost generator), whose spectrum is the continuous real line.
• Because local algebras are type III₁, the Connes invariant $S(\mathcal{A})=\{0\}\cup {\mathbb{R}}^{+}$ is continuous.

A flat entanglement spectrum requires a modular operator proportional to the identity (trivial modular flow) — a type-I / discrete feature incompatible with a faithful normal state on a type III₁ factor. So magic in QFT is a shadow of type III₁, exactly the property that also makes vacuum Bell-CHSH violation natural and makes the Araki–Uhlmann relative entropy the right entropy. This places magic as a fourth vertex on the type III₁ common-thread diagram alongside Bell violation, relative entropy, and embezzlement.

Two features make it immediately actionable for the group:

1. It is computable with tools already in hand. The anti-flatness $F_2$ is read off the same Rindler-wedge reduced density matrices and free-field Bogoliubov coefficients used in the relative-entropy and Bell-CHSH computations — so the group’s coherent / squeezed / cat states can be assigned a magic content directly (magic-of-coherent-squeezed-cat-states).
2. It is a phase diagnostic. Magic sharply separates local QFT (magical, type III₁) from TQFT (can be stabilizer: the toric code has a flat spectrum, $Z[{\mathcal{M}}_n]=(Z[{\mathcal{M}}_1]{)}^n{D}^{n-1}$, and is not a net of type III₁ factors). This is a different cut than non-Gaussianity (free vs. interacting), and it links to topological-order, string-net-condensation, and confinement / higher-form symmetry (entanglement-as-confinement-probe).

Relations

• type-iii-von-neumann-algebras — the continuous Connes spectrum of type III₁ is precisely what forbids the flat spectrum of a stabilizer state; magic is forced by the algebraic type
• bisognano-wichmann-theorem — fixes the continuous-spectrum vacuum modular operator that makes the spectrum non-flat
• reeh-schlieder-theorem — guarantees physical states are cyclic-separating (faithful normal), the hypothesis of the argument
• tomita-takesaki-modular-theory — trivial modular flow ⟺ flat spectrum ⟺ stabilizer; non-trivial modular flow ⟺ magic
• araki-uhlmann-relative-entropy — the other III₁-native quantity, and the one that carries the relative modular operator ${\Delta }_{\sigma |\rho }$ in which $M_{\mathrm{rel}}$ is exactly expressible; a putative magic ↔ relative-entropy inequality is relative-entropy-magic-inequality
• canonical-free-structure-for-magic-in-qft — the structural blockage between the witness and a quantitative measure; quantum-magic-modular-qft-study is the internal study that establishes it
• weyl-operators — the continuum Weyl system that the Pauli group is the finite model of; the stabilizer formalism is finite Heisenberg–Weyl theory, with $(\Gamma ,\sigma )$ replaced by $({\mathbb{Z}}_2^{2n},[\cdot ,\cdot ])$ and stabilizer states = Lagrangean subspace + sign pattern
• bell-chsh-inequality — shares the cyclic-separating / BW root; both are readings of the continuous modular spectrum
• entanglement-embezzlement — sibling III₁-rooted resource in the common-thread program
• rindler-wedges, squeezed-states, cat-states — the free-field setting where anti-flatness is explicitly computable
• replica-trick-gravity — the replica/conical-defect route to the same non-flatness conclusion
• topological-order, toric-code, string-net-condensation — the TQFT sector where stabilizer (flat) states are allowed

Papers

• 2026-benedetti-magic-in-qft — universality of magic in local QFT (the source; full references therein).
• quantum-magic-modular-qft-study — internal study (Aug 2026): the modular formulation, the frame-dependence no-go, the three-resource taxonomy, and verified single-qubit benchmarks.
• Foundational QI: Bravyi–Kitaev (2005), Howard–Wallman–Veitch–Emerson (2014); resource theory: Veitch–Mousavian–Gottesman–Emerson (2014, $M_{\mathrm{rel}}$ and mana), Howard–Campbell (2017) and Heinrich–Gross (2019, robustness), Bravyi–Gosset (2016) and Bravyi et al. (2019, extent), Leone–Oliviero–Hamma (2022, SRE) with monotonicity for $\alpha \ge 2$ by Leone–Bittel (2024), Deckers et al. (2026, closest stabilizer states and nonadditivity of $M_{\mathrm{rel}}$).
• Witnesses: Tirrito et al. (2024, anti-flatness), Cao et al. (2024, stabilizer-distance bound and holographic magic); CFT/lattice: White–Cao–Swingle (2021), Hoshino–Oshikawa–Ashida (2026), Tarabunga–Tirrito–Dalmonte (2023), Matsuda–Hoshino–Ashida (2026, convolution MRE unifying spins/bosons/fermions); chaos/gravity: García-García–Liu–Zheng (2026, fermionic anti-flatness in SYK with a gravity dual).

Notes

• Frame-dependence, stated precisely. Stabilizerness depends on the choice of Pauli/Weyl frame, and this is not a lattice artefact that the algebraic description washes out — it is a no-go. No functional of the abstract pair $(B(\mathcal{H}),\rho )$ alone can be a faithful magic measure: every automorphism of a finite type-I factor is inner, so $\rho$ and $U\rho U^{†}$ are abstractly isomorphic, while stabilizerness is invariant only under the Clifford normalizer. One qubit settles it — ${\rho }_{ϵ}^{+}$ and $T{\rho }_{ϵ}^{+}{T}^{†}$ have identical spectra and unitarily equivalent modular operators, yet the first is free and the second is magic for $0<ϵ<1-1/\sqrt{2}$. The right reading of the QFT statements is therefore: the witness (non-flatness) is frame-independent and rests only on modular structure; a quantitative monotone is not, and still requires the frame as declared data. See canonical-free-structure-for-magic-in-qft and quantum-magic-modular-qft-study.
• Three resources, one name — and they are provably inequivalent. (i) Stabilizer magic, free set = the stabilizer polytope; (ii) non-Gaussianity, free set = quasifree/Gaussian states; (iii) modular anti-flatness, where flat spectra are the excluded free-like behaviour. The separating example is the free-field vacuum itself: it is quasifree, so any quasifree-distance measure returns zero — yet it has nontrivial local modular structure and is non-stabilizer by the anti-flatness criterion. So “distance to Gaussian” is not QFT magic, and the Weyl $\to$ Pauli, quasifree $\to$ stabilizer dictionary is an analogy, not an equality. Entanglement is a fourth, independent axis (a Bell pair is a maximally entangled stabilizer state).
• Large-$N$ particle states flatten. Benedetti et al. find the anti-flatness of Unruh $N$-particle states decreases as $N\to \infty$, seemingly approaching a flat (stabilizer-like) spectrum — but that limit is a singular state outside Fock space, so it does not violate the universality result. Whether superposition states (cat states) run the opposite way is an open computation (magic-of-coherent-squeezed-cat-states).
• Measure vs. witness. Flatness/anti-flatness is a witness (necessary condition). Turning it into an intrinsic measure of magic in QFT — via averaging over Clifford orbits — is flagged by the authors as not yet developed in a QFT context, and is itself a research target. The 2026-08 study makes the gap structural rather than merely undeveloped: type III₁ supplies the obstruction but no Pauli group, no Clifford normalizer, no convex free set, no free operations, and no operational task; and if every faithful normal state shares the same full positive modular spectral support, that support cannot order states by magic at all.
• What does transfer to type III. The divergence machinery. In Type-I standard form, $M_{\mathrm{rel}}(\rho )={\inf }_{\sigma \in \mathrm{STAB}}[-⟨{\rho }^{1/2},\log {\Delta }_{\sigma |\rho }{\rho }^{1/2}⟩]$ with ${\Delta }_{\sigma |\rho }=L_{\sigma }{R}_{{\rho }^{-1}}$ — an exact rewrite whose form is unchanged on any von Neumann algebra once a free set is named. Stabilizer Rényi entropy likewise: $M_{\alpha }=\log d-D_{\alpha }(q_{\psi }\Vert u)$ is a classical Rényi divergence on the commutative ${\ell }^{\infty }({\mathcal{P}}_d)$. Both are reformulations, not new monotones, and the second has a hard continuum obstruction — an infinite Weyl group has no uniform counting measure.