Clifford Non-Stabilizerness: Quantum Magic Resource

• Clifford non-stabilizerness is the measure of quantum 'magic' that quantifies how far states or operations are from the efficient, stabilizer (Clifford) framework.
• It plays a critical role in enabling universal quantum computation by marking the transition from easily simulable stabilizer dynamics to complex, non-Clifford processes.
• Recent quantitative methods like stabilizer Rényi entropies and NsEE diagnose many-body phases, dynamics, and chaos, offering tools for fault tolerance and metrology.

Clifford non-stabilizerness, usually called magic, denotes the part of a quantum state or operation that lies outside the stabilizer/Clifford subtheory and therefore outside the regime efficiently simulable by Gottesman–Knill-type methods. In contemporary usage it is both a resource-theoretic notion and a many-body diagnostic: it quantifies departure from stabilizer structure, not merely departure from product structure or the presence of large entanglement, and it appears in studies of phases of matter, monitored and open dynamics, metrology, fault tolerance, and classical-simulation complexity (Tirrito et al., 2023, Huang et al., 2024, Korbany et al., 26 Feb 2025).

1. Resource-theoretic meaning

In the stabilizer framework, stabilizer states are states generated from computational-basis product states by Clifford unitaries, and Clifford operations are the free operations because they preserve Pauli structure under conjugation. A pure state is therefore stabilizer if it lies in the Clifford orbit of a computational-basis product state, while a non-stabilizer state carries a beyond-Clifford resource. This is the sense in which magic measures “how far” a state is from stabilizer structure (Tirrito et al., 2023).

This distinction is not equivalent to either separability or entanglement. Product states such as $|T\rangle^{\otimes N}$, with

$$|T\rangle=\frac{|0\rangle+e^{i\pi/4}|1\rangle}{\sqrt{2}},$$

have zero entanglement entropy but nonzero magic, while GHZ states can be highly entangled yet have zero non-stabilizerness in the stabilizer sense. Recent work on permutation-symmetric metrological states makes the same point from a different direction: GHZ states maximize certain Bell-correlation diagnostics while still having $\mathcal M_q=0$ (Huang et al., 2024, Hernández-Yanes et al., 1 Oct 2025).

The same resource-theoretic language extends naturally from states to logical operations. Gates such as $T$, $CS$, and $\sqrt{T}$ are non-Clifford and hence non-stabilizer resources; in fault-tolerant settings they mark the transition from stabilizer computation to universal computation. Clifford-hierarchy stabilizer codes make this explicit by constructing topological codes with transversal logical $T$, $CS$, and $\sqrt{T}$ gates, thereby embedding non-stabilizer functionality directly into code symmetries rather than treating it solely as an externally injected ancilla resource (Kobayashi et al., 4 Nov 2025).

2. Quantifiers and structural diagnostics

The dominant quantitative language is based on stabilizer Rényi entropies. For a pure $L$-qubit state $$|T\rangle=\frac{|0\rangle+e^{i\pi/4}|1\rangle}{\sqrt{2}},$$0,

$$|T\rangle=\frac{|0\rangle+e^{i\pi/4}|1\rangle}{\sqrt{2}},$$1

where $$|T\rangle=\frac{|0\rangle+e^{i\pi/4}|1\rangle}{\sqrt{2}},$$2 is the Pauli-string set. These quantities are faithful, Clifford invariant, and additive. For mixed states and open-system density matrices, a common choice is

$$|T\rangle=\frac{|0\rangle+e^{i\pi/4}|1\rangle}{\sqrt{2}},$$3

which interprets magic as delocalization of the Pauli spectrum (Grabarits et al., 9 Mar 2026, Sticlet et al., 15 Apr 2025).

Several related measures refine different aspects of Clifford non-stabilizerness. The stabilizer linear entropy

$$|T\rangle=\frac{|0\rangle+e^{i\pi/4}|1\rangle}{\sqrt{2}},$$4

obeys

$$|T\rangle=\frac{|0\rangle+e^{i\pi/4}|1\rangle}{\sqrt{2}},$$5

making it a convenient proxy for second-order stabilizer entropy. Stabilizer nullity

$$|T\rangle=\frac{|0\rangle+e^{i\pi/4}|1\rangle}{\sqrt{2}},$$6

counts the loss of exact Pauli stabilizers and functions as a coarse-grained magic monotone. CSS entropies generalize this perspective to qudits through defect-subspace projectors in the Clifford commutant; for qubits the first nontrivial CSS entropy reduces exactly to the second Rényi stabilizer entropy (Tirrito et al., 2023, Scocco et al., 15 Jul 2025, Turkeshi et al., 2024).

A persistent theme is that magic is not exhausted by any single diagnostic. One proposal aimed explicitly at classical simulation hardness is Non-stabilizerness Entanglement Entropy (NsEE),

$$|T\rangle=\frac{|0\rangle+e^{i\pi/4}|1\rangle}{\sqrt{2}},$$7

the minimum residual entanglement after optimization over Clifford circuits. This quantity was introduced because large entanglement can occur within the classically easy stabilizer sector, while nonzero stabilizer Rényi entropy can occur in states that remain easy to represent because they are unentangled product states. NsEE is therefore intended to quantify the interplay of entanglement and beyond-Clifford structure, rather than magic alone (Huang et al., 2024).

Two structural characterizations tie magic to notions of flatness. In one direction, multifractal flatness

$$|T\rangle=\frac{|0\rangle+e^{i\pi/4}|1\rangle}{\sqrt{2}},$$8

is nonnegative, vanishes iff the computational-basis participation distribution is flat, and its Clifford-orbit average satisfies

$$|T\rangle=\frac{|0\rangle+e^{i\pi/4}|1\rangle}{\sqrt{2}},$$9

In another direction, the anti-flatness of the entanglement spectrum,

$\mathcal M_q=0$0

has a Clifford-orbit average proportional to $\mathcal M_q=0$1. Both results recast non-stabilizerness as a flatness problem—either of basis probabilities across a Clifford orbit or of reduced-state spectra under Clifford scrambling (Turkeshi et al., 2023, Tirrito et al., 2023).

The literature also stresses that magic can be basis dependent. In non-Hermitian systems, for example, the quantity actually analyzed is the right-right stabilizer Rényi entropy

$\mathcal M_q=0$2

computed from a reduced density matrix built only from the right eigenvector. In that setting real-space and momentum-space magic can display opposite extrema at the same exceptional line, reflecting basis dependence without destroying diagnostic usefulness (Moca et al., 20 Oct 2025).

3. Many-body structure, phases, and long-range organization

In many-body systems, Clifford non-stabilizerness is often split into local and nonlocal components. A canonical example is the $\mathcal M_q=0$3-state, whose order-2 stabilizer Rényi entropy is

$\mathcal M_q=0$4

so that $\mathcal M_q=0$5 at large $\mathcal M_q=0$6. This logarithmic scaling was identified as a form of non-local non-stabilizerness: each component of the superposition is stabilizer-like, while the magic originates from the delocalized coherent superposition itself. In topologically frustrated spin chains, the relevant kink superposition is Clifford-equivalent to a $\mathcal M_q=0$7-state, yielding a decomposition into an extensive local contribution plus the subdominant logarithmic $\mathcal M_q=0$8-contribution (Odavić et al., 2022).

This idea was sharpened into the notion of long-range nonstabilizerness, defined as the component of magic that cannot be removed by shallow local quantum circuits. For one-dimensional translation-invariant MPS ground states, a sufficient criterion is obtained from the renormalization-group fixed point: if the asymptotic Shannon entropy of the sector weights, equivalently the distant-region mutual information of the fixed-point decomposition, approaches a non-integer value, then no polylog-depth circuit can map the state close to a stabilizer state. Stabilizer fixed points force this mutual information to be quantized, so non-integer limits obstruct shallow-circuit trivialization (Korbany et al., 26 Feb 2025).

Global symmetries impose another layer of structure. For $\mathcal M_q=0$9-constrained Haar-random states, the average stabilizer entropy is suppressed relative to the unconstrained benchmark. Unconstrained Haar-random states satisfy $T$0, whereas at zero magnetization in the standard $T$1 sector the constrained benchmark becomes $T$2. Exact results furthermore show strong agreement with midspectrum eigenstates of the nonlocal complex-fermion SYK model and systematic $T$3 deficits in local XXZ chains, highlighting the role of locality in preventing full randomization within the symmetry sector (Iannotti et al., 30 Mar 2026).

Non-Hermitian many-body systems exhibit a further reorganization of magic around criticality and exceptional points. In the $T$4-symmetric non-Hermitian transverse-field Ising chain, $T$5 peaks along the Hermitian-like Ising transition line

$T$6

but vanishes at the exceptional line $T$7, where a similarity-transform argument maps the problem to a stabilizer-like ferromagnetic ground state. In the non-Hermitian XX chain the pattern reverses in real space: $T$8 is maximized at the exceptional line $T$9, while the momentum-space magic density has a local minimum there and even vanishes at the exact exceptional momentum $CS$0. The shared message is that exceptional physics is marked by extremal, but basis-dependent, non-stabilizerness (Moca et al., 20 Oct 2025).

4. Dynamics, criticality, and chaos

Generic dynamics can generate non-stabilizerness much faster than they generate entanglement. In one-dimensional brick-wall random unitary circuits, CSS entropies equilibrate with

$CS$1

whereas entanglement entropy saturates only on $CS$2 timescales. The deviation from Haar-typical magic relaxes as

$CS$3

placing magic spreading in the same fast class as anti-concentration and Hilbert-space delocalization rather than ballistic entanglement growth (Turkeshi et al., 2024).

Interspersed random Clifford layers profoundly reshape this dynamics without creating magic on their own. For the linear stabilizer-entropy-based non-stabilizing power $CS$4, averaging over an intervening random Clifford $CS$5 yields

$CS$6

which leads to exponential thermalization toward the Haar-averaged value in long Clifford-interlaced circuits. The same framework extends to an operator-space non-stabilizing power and shows that chaos in brick-wall circuits depends jointly on non-stabilizing power, entangling power, and gate typicality rather than on any single resource (Varikuti et al., 20 May 2025).

Measurement can either suppress or sustain magic. In monitored circuits built from random Clifford unitaries and local projective measurements, computational-basis measurements cannot create magic and reduce stabilizer nullity only in quantized unit steps, with transition probability

$CS$7

at large $CS$8. Full removal of magic then requires exponentially many measurements, $CS$9, reflecting protection by Clifford scrambling. By contrast, measurements in rotated non-Clifford bases both create and destroy non-stabilizerness, driving the system to a universal extensive-nullity steady state; the stabilizer Rényi entropies then reveal angle dependence and a nonzero steady-state magic density even when nullity appears almost angle insensitive (Scocco et al., 15 Jul 2025).

Open-system and driven many-body settings display additional universal behavior. In boundary-driven open XXZ chains,

$\sqrt{T}$0

with ballistic $\sqrt{T}$1, KPZ $\sqrt{T}$2, and diffusive $\sqrt{T}$3 regimes. Under bulk dephasing, dissipation can transiently enhance magic before suppressing it, and in magnetization-conserving sectors the nonequilibrium steady state can retain nonzero magic density. In slow sweeps across quantum critical points, both $\sqrt{T}$4 and the cumulants of the logarithmic Pauli spectrum satisfy Kibble–Zurek-type power laws

$\sqrt{T}$5

while the logarithmic Pauli spectrum becomes asymptotically Gaussian, so the Pauli spectrum itself is lognormal (Sticlet et al., 15 Apr 2025, Grabarits et al., 9 Mar 2026).

The relation to chaos diagnostics depends on the probe. In isospectral-twirled ensembles interpolating between Clifford eigenbases and Haar-random eigenbases through $\sqrt{T}$6-doped circuits, Loschmidt echoes and OTOCs clearly distinguish stabilizer from Haar behavior, while tripartite information, entanglement-entropy bounds, coherence, and WYD skew information show only finite-size or transient differences. This sharpens the distinction between probes sensitive to magic in the eigenbasis and probes largely insensitive to it (Cusumano et al., 31 Mar 2026).

5. Numerical methods and experimental access

Much of the recent development has been driven by scalable numerical methods. In non-Hermitian spin chains, ground states were obtained by non-Hermitian DMRG/MPS with open boundary conditions, right eigenstates as MPS, and the ground state defined by the smallest real part of the energy; $\sqrt{T}$7 was then estimated by Metropolis–Hastings sampling over Pauli strings weighted by $\sqrt{T}$8. For open XXZ chains, an MPO/MPS algorithm computes $\sqrt{T}$9 from vectorized density matrices while keeping the bond dimension constant, avoiding the bond-dimension proliferation usually associated with mixed-state tensor-network calculations (Moca et al., 20 Oct 2025, Sticlet et al., 15 Apr 2025).

For conventional MPS studies, perfect Pauli sampling, Pauli-Markov chains, and replica Pauli-MPS methods have made full-state magic and mutual magic accessible at system sizes far beyond direct enumeration. In spin-1 XXZ chains, the full-state SRE density was observed to converge with

$T$0

at critical points and to saturate extremely rapidly in gapped phases, often more easily than entanglement entropy itself. Pauli-Markov chains were also shown to provide especially efficient estimators for mutual information and mutual magic of disconnected subsystems (Frau et al., 2024).

A distinct strategy is to avoid simulating non-Clifford dynamics directly. Iterative Clifford Circuit Renormalization (ICCR) rewrites a circuit with measurements or $T$1-gates into a Clifford-only circuit acting on a renormalized initial state, pushing non-stabilizing effects backward and compressing the effective state by MPS methods. This gave access to systems up to $T$2 qubits and revealed a measurement-induced magic-purification transition near $T$3 in a monitored random Clifford circuit (Paviglianiti et al., 2024).

Several proposals translate magic quantification into experimentally tractable observables. The multifractal-flatness approach samples random Clifford circuits, measures computational-basis probabilities, and infers $T$4 from an orbit-averaged flatness witness, providing a practical certification protocol even though the cost grows exponentially for precise estimation. The entanglement-spectrum-flatness approach measures

$T$5

after Clifford scrambling and uses it as a local witness of non-stabilizerness, with explicit discussion of coherent gate noise and applications to cold-atom and solid-state platforms. For permutation-symmetric states, the SRE itself simplifies drastically: in the large-$T$6 limit it depends only on six overlaps with the coherent stabilizer states $T$7, which can be accessed by interaction-based readout or twist-echo methods (Turkeshi et al., 2023, Tirrito et al., 2023, Hernández-Yanes et al., 1 Oct 2025).

6. Operational roles: simulation hardness, metrology, estimation, and control

A central operational question is what non-stabilizerness actually measures. Ordinary entanglement entropy is insufficient because stabilizer states can have large entanglement while remaining classically easy, and magic alone is insufficient because product states such as $T$8 have nonzero SRE but zero entanglement and are still easy to represent classically. NsEE was introduced precisely to isolate the part of entanglement that survives optimal Clifford simplification, and numerical studies with Clifford-circuits-augmented MPS showed that it tracks easy-to-hard transitions in random Clifford+$T$9 circuits more faithfully than either entanglement entropy or SRE alone (Huang et al., 2024).

In quantum metrology, non-stabilizerness differentiates distinct forms of useful many-body structure. Under one-axis twisting,

$CS$0

optimal squeezing occurs at $CS$1 with $CS$2, and the corresponding SRE grows logarithmically with system size. At later times, the dynamics produces kitten states—superpositions of rotated GHZ states—with $CS$3-independent SRE that decreases as Bell-correlation strength increases. In the limiting GHZ case, Bell correlations are maximal while magic vanishes, establishing that Bell nonlocality and Clifford non-stabilizerness are distinct resources (Hernández-Yanes et al., 1 Oct 2025).

In quantum state estimation, magic becomes a metrological resource in a direct algebraic sense. For single-setting protocols consisting of ancillas, a circuit, and a fixed projective readout, stabilizer-only resources are always informationally equivalent to projective measurement in a stabilizer basis and hence never informationally complete, regardless of the number of ancillas. Introducing $CS$4-gates enlarges the accessible operator span: at least

$CS$5

$CS$6-gates are necessary for informational completeness, while $CS$7 suffice, leading to the conjecture that $CS$8 are both necessary and sufficient (Monaco et al., 30 Sep 2025).

Fault-tolerant computation and state control provide two further operational incarnations. Clifford-hierarchy stabilizer codes realize logical $CS$9, $\sqrt{T}$0, and $\sqrt{T}$1 gates transversally and support fault-tolerant preparation of logical $\sqrt{T}$2 magic states via code switching and just-in-time decoding, showing that non-stabilizer resources can be generated from topological automorphism symmetries rather than only consumed as external ancillas (Kobayashi et al., 4 Nov 2025). Conversely, recent work introduced the dismagicker, a non-Clifford unitary variationally optimized to reduce $\sqrt{T}$3. Interleaving such gates with Clifford disentanglers in MPS sweeps was found to suppress both non-stabilizerness and entanglement more effectively than either strategy alone, improving fixed-bond-dimension simulation accuracy and rotating target states toward more classically tractable representations (Huang et al., 5 Apr 2026).