Magic Hierarchy: From Quantum Gates to Graphene

• Magic Hierarchy is an ordered classification that tiers computational resources and physical phases, as seen in quantum information (Clifford hierarchies and magic states) and condensed matter systems.
• In quantum computation, operational magic is quantified via multi-layer protocols and magic state cultivation methods that address fault tolerance and resource optimization.
• Across disciplines, magic hierarchy manifests in algebraic rescalings in twisted graphene, phase ordering in neutrino oscillations, and structured symmetries in Lie algebras and supergravity.

Searching arXiv for papers on “magic hierarchy” and closely related uses across quantum information and condensed matter. Search query: "magic hierarchy Clifford hierarchy magic state cultivation magic hierarchy twisted bilayer graphene" “Magic hierarchy” is a polysemous technical expression used in several research literatures to denote an ordered family of objects singled out by a “magic” condition. In quantum information, it most often refers to structures built around the Clifford hierarchy, magic states, and alternating Clifford/non-Clifford circuit layers; in twisted graphene it denotes exact rescalings of bilayer magic-angle sequences in alternating-twist multilayers and, in a separate usage, an ordering of correlated phases by robustness to deviations from the magic angle; in neutrino oscillations it refers to magic-baseline or magic-energy conditions that remove $\delta$-dependence; and in exceptional algebra and supergravity it appears in the Magic Square program, where Jordan algebras organize a hierarchy of symmetry groups and real forms (Chen et al., 9 Jun 2026, 1901.10485, Polski et al., 2022, Rahman et al., 2010, Cacciatori et al., 2012). This suggests a family resemblance rather than a single formal theory: a “magic hierarchy” is typically an ordered classification in which a resource, parameter, or symmetry appears in progressively structured layers.

1. Quantum-information meaning: magic as a hierarchy of computational resources

In quantum computation, the dominant usage is tied to the Clifford hierarchy. For odd prime dimension $p$, one standard recursive definition is

$$\mathcal C_k:=\{U\mid U\mathcal C_1 U^\dagger\subseteq \mathcal C_{k-1}\},$$

with $\mathcal C_1$ the Pauli group and $\mathcal C_2$ the Clifford group. The third level already contains the canonical non-stabilizer resources: single-qudit magic states of the form

$$\ket{f_{a,b,c}}=\frac{1}{\sqrt p}\sum_{k=0}^{p-1}\omega^{a k^3+b k^2+c k}\ket{k}, \qquad \omega=e^{2\pi i/p},$$

which are obtained by applying a diagonal level-3 unitary $M_{a,b,c}\in\mathcal C_3$ to $\ket{+}$. In the Pauli-restricted qudit CHSH scenario, these states maximize the Bell operator within the allowed strategy class, while among equatorial states they minimize total collision entropy over the $p$ non-computational stabilizer bases; stabilizer states occupy the opposite extreme, being the least nonlocal and most uncertain in that setting (Howard, 2015).

Prime-dimensional magic-state structure is finer than a single orbit. The Alltop vectors are produced from Ivanovic MUBs by a level-3 “magic” unitary

$M=\sum_r \omega^{r^3}\ket r\!\bra r,$

and therefore serve as prime-dimensional magic states. When the prime dimension satisfies $p$0, the set of Alltop vectors splits into three distinct Clifford orbits, so not all magic states are Clifford-equivalent. The same arithmetic condition is also tied to order-3 Clifford symmetries, Zauner subspaces, and differing mana values in explicit examples such as $p$1 (Bengtsson et al., 2014).

A separate line of work studies whether hierarchy level alone orders operational magic in early fault-tolerant quantum computing. For pure-state magic functionals built from Pauli expectation values, faithfulness, permutation invariance, tensor-product additivity, and continuity force a Rényi-type form built from Pauli moments

$p$2

Within that framework, two no-go theorems are established: hierarchy level alone cannot universally order operational magic, and no state-independent sequence of operations can guarantee monotonic magic improvement. In the shallow diagonal setting, a zero-magic mechanism exists; conversely, maximal magic strictly requires graph-state preconditioning. In the multilayer STAR-motivated setting, the Pauli spectrum evolves by an exact iterative update rule, but the resulting orthogonal transfer cannot decrease all $p$3 simultaneously for all inputs. The implication is that magic generation is state-dependent and must be treated as a control-and-optimization problem over the Pauli spectrum rather than as a monotone ladder indexed only by gate level (Lu et al., 6 May 2026).

Resource quantification also develops its own hierarchy. For the robustness of magic, symmetry reduction compresses the exact $p$4-qubit optimization to an $p$5-dimensional invariant subspace for $p$6 and $p$7, and then a finite approximation hierarchy is obtained by restricting decompositions to $p$8-partite entangled stabilizer states. Each fixed level is polynomial-time evaluable, yields a rigorous upper bound, and the hierarchy terminates at the exact robustness when $p$9 (Heinrich et al., 2018).

An explicit fermionic analogue is the matchgate hierarchy. There the role of Paulis is played by Majorana operators, level 2 is the Gaussian or matchgate class, and higher levels are defined by recursive conjugation of Majoranas. The operational interpretation parallels Gottesman–Chuang teleportation: hierarchy level measures the required depth of adaptivity and the number of matchgate-magic states consumed. For two qubits, the paper gives a complete characterization and shows that the number of resource states grows linearly with the target gate’s level (Bampounis et al., 2024).

2. Fault tolerance, gate lifting, and higher-level magic-state constructions

A central recent development is the extension of magic-state cultivation beyond the standard $$\mathcal C_k:=\{U\mid U\mathcal C_1 U^\dagger\subseteq \mathcal C_{k-1}\},$$0 resource. In the hierarchy notation

$$\mathcal C_k:=\{U\mid U\mathcal C_1 U^\dagger\subseteq \mathcal C_{k-1}\},$$1

so $$\mathcal C_k:=\{U\mid U\mathcal C_1 U^\dagger\subseteq \mathcal C_{k-1}\},$$2 is the $$\mathcal C_k:=\{U\mid U\mathcal C_1 U^\dagger\subseteq \mathcal C_{k-1}\},$$3 case and $$\mathcal C_k:=\{U\mid U\mathcal C_1 U^\dagger\subseteq \mathcal C_{k-1}\},$$4 is the $$\mathcal C_k:=\{U\mid U\mathcal C_1 U^\dagger\subseteq \mathcal C_{k-1}\},$$5 case. The cultivation protocol of “Efficient Magic State Cultivation for $$\mathcal C_k:=\{U\mid U\mathcal C_1 U^\dagger\subseteq \mathcal C_{k-1}\},$$6 Gates” generalizes phase-kickback checks to arbitrary Clifford-hierarchy levels in codes that support the needed transversal gate. The key phase relation is

$$\mathcal C_k:=\{U\mid U\mathcal C_1 U^\dagger\subseteq \mathcal C_{k-1}\},$$7

which enables double phase-kickback checks using a GHZ ancilla and $$\mathcal C_k:=\{U\mid U\mathcal C_1 U^\dagger\subseteq \mathcal C_{k-1}\},$$8 or $$\mathcal C_k:=\{U\mid U\mathcal C_1 U^\dagger\subseteq \mathcal C_{k-1}\},$$9 on the ancillas. The concrete $\mathcal C_1$0 construction uses the doubled color code, including the $\mathcal C_1$1 quantum Reed–Muller code and $\mathcal C_1$2 doubled color codes such as $\mathcal C_1$3 and $\mathcal C_1$4. The paper also gives an escape strategy to larger rotated surface codes by lattice surgery. Quantitatively, it reports that $\mathcal C_1$5 behaves very similarly to $\mathcal C_1$6 in the chosen noise models; for $\mathcal C_1$7 codes, ungrown $\mathcal C_1$8 cultivation reaches logical infidelities around $\mathcal C_1$9 with success rates around $\mathcal C_2$0; and, for end-to-end cultivation plus lattice surgery, $\mathcal C_2$1 gives about $\mathcal C_2$2 infidelity with idling noise and below $\mathcal C_2$3 without idling noise, while $\mathcal C_2$4 improves logical error by roughly two orders of magnitude at the cost of very low success rates (Chen et al., 9 Jun 2026).

A structural complement is the study of when taking square roots climbs the Clifford hierarchy. For a Hermitian involution $\mathcal C_2$5, the canonical square root

$\mathcal C_2$6

does not in general rise by one level. The complete characterization for Hermitian Clifford gates states that $\mathcal C_2$7 exactly when the associated symplectic matrix $\mathcal C_2$8 is hyperbolic and $\mathcal C_2$9. The same work shows that if a Hermitian Clifford gate climbs to level 3, then the square root of its controlled version lies in level 4. This provides a precise criterion for “climbing the hierarchy” by square roots and controls rather than by diagonal constructions alone (Bastioni et al., 12 Mar 2026).

The hierarchy can also be built into the code rather than supplied as an external injected resource. Clifford hierarchy stabilizer codes generalize Pauli stabilizer codes by allowing stabilizers to contain gates from higher levels, with explicit 2D and 3D constructions tied to twisted Dijkgraaf–Witten gauge theories. In 2D, the framework yields transversal logical $$\ket{f_{a,b,c}}=\frac{1}{\sqrt p}\sum_{k=0}^{p-1}\omega^{a k^3+b k^2+c k}\ket{k}, \qquad \omega=e^{2\pi i/p},$$0 and $$\ket{f_{a,b,c}}=\frac{1}{\sqrt p}\sum_{k=0}^{p-1}\omega^{a k^3+b k^2+c k}\ket{k}, \qquad \omega=e^{2\pi i/p},$$1 gates; in 3D, it yields a transversal logical $$\ket{f_{a,b,c}}=\frac{1}{\sqrt p}\sum_{k=0}^{p-1}\omega^{a k^3+b k^2+c k}\ket{k}, \qquad \omega=e^{2\pi i/p},$$2 gate. The same structure supports logical $$\ket{f_{a,b,c}}=\frac{1}{\sqrt p}\sum_{k=0}^{p-1}\omega^{a k^3+b k^2+c k}\ket{k}, \qquad \omega=e^{2\pi i/p},$$3-magic-state preparation by code switching in $$\ket{f_{a,b,c}}=\frac{1}{\sqrt p}\sum_{k=0}^{p-1}\omega^{a k^3+b k^2+c k}\ket{k}, \qquad \omega=e^{2\pi i/p},$$4 rounds via a just-in-time decoder, and motivates a conjectural hierarchy-versus-dimension principle in which a logical gate in the $$\ket{f_{a,b,c}}=\frac{1}{\sqrt p}\sum_{k=0}^{p-1}\omega^{a k^3+b k^2+c k}\ket{k}, \qquad \omega=e^{2\pi i/p},$$5-th Clifford-hierarchy level can be realized in $$\ket{f_{a,b,c}}=\frac{1}{\sqrt p}\sum_{k=0}^{p-1}\omega^{a k^3+b k^2+c k}\ket{k}, \qquad \omega=e^{2\pi i/p},$$6 spatial dimensions (Kobayashi et al., 4 Nov 2025).

Topological quantum field theory supplies another route. In Dijkgraaf–Witten theory with gauge group $$\ket{f_{a,b,c}}=\frac{1}{\sqrt p}\sum_{k=0}^{p-1}\omega^{a k^3+b k^2+c k}\ket{k}, \qquad \omega=e^{2\pi i/p},$$7, a single Dehn twist on the boundary torus gives the exact logical $$\ket{f_{a,b,c}}=\frac{1}{\sqrt p}\sum_{k=0}^{p-1}\omega^{a k^3+b k^2+c k}\ket{k}, \qquad \omega=e^{2\pi i/p},$$8 gate,

$$\ket{f_{a,b,c}}=\frac{1}{\sqrt p}\sum_{k=0}^{p-1}\omega^{a k^3+b k^2+c k}\ket{k}, \qquad \omega=e^{2\pi i/p},$$9

while in Chern–Simons theory the path integral can generate a continuously tunable Ising interaction gate $M_{a,b,c}\in\mathcal C_3$0, which is non-Clifford away from discrete Clifford points. The same paper shows that Toffoli is obstructed in $M_{a,b,c}\in\mathcal C_3$1 by the $M_{a,b,c}\in\mathcal C_3$2 fusion structure and identifies $M_{a,b,c}\in\mathcal C_3$3 as the minimal theory with the required conditional branching and dense mapping-class-group image (Munizzi et al., 15 Apr 2026).

3. Many-body magic classes, long-range magic, and circuit-complexity hierarchies

A distinct usage treats magic as a macroscopic many-body feature rather than a single-shot computational resource. “Magic Class and the Convolution Group” defines two states to be in the same circuit magic class if they are related by a Clifford unitary, and introduces the convolution group (CG), an iterative coarse-graining map based on quantum convolution. The CG fixed point is the mean state $M_{a,b,c}\in\mathcal C_3$4, obtained by retaining only characteristic-function components of unit modulus. For $M_{a,b,c}\in\mathcal C_3$5-qudit pure states, the CG fixed points organize states into $M_{a,b,c}\in\mathcal C_3$6 classes, with class label $M_{a,b,c}\in\mathcal C_3$7 determined by the stabilizer-group size

$M_{a,b,c}\in\mathcal C_3$8

equivalently by the entropy criterion

$M_{a,b,c}\in\mathcal C_3$9

The paper explicitly frames this as an RG-like classification in which symmetry and entropy of the fixed point play the role of phase diagnostics (Bu et al., 2024).

Long-range magic sharpens the analogy with long-range entanglement. In the first level of the many-body magic hierarchy, one considers the two orderings $\ket{+}$0 and $\ket{+}$1, where $\ket{+}$2 denotes a finite-depth local unitary circuit. Two-sided long-range magic means that a state cannot be prepared by either ordering, even approximately, and therefore lies outside the entire first level. Explicit examples include the “magical cat” state

$\ket{+}$3

and ground states of certain nonabelian topological orders; the proofs use light-cone bounds, stabilizer overlap lemmas, mutual-information arguments, approximate-code complexity bounds, and the absence of topologically transversal gates (Li, 26 Mar 2026).

The same first-level regime appears in the circuit-complexity formulation of the magic hierarchy. “Quantum circuit lower bounds in the magic hierarchy” defines alternating classes $\ket{+}$4 and $\ket{+}$5, built from arbitrary-size Clifford circuits and constant-depth $\ket{+}$6 layers, and sets

$\ket{+}$7

The model unifies $\ket{+}$8 and adaptive intermediate-measurement models, with the explicit equivalence $\ket{+}$9. At the first nontrivial level, the paper proves state-preparation lower bounds for explicit states, including ground states of some topologically ordered Hamiltonians and nonstabilizer quantum codes. A key technical idea is “infectiousness”: if even a single state in a high-distance code can be approximately prepared by such a circuit, then the entire subspace must lie close to a perturbed stabilizer code. The same paper emphasizes that stronger lower bounds higher in the hierarchy would imply classical circuit lower bounds against $p$0 beyond current techniques (Parham, 28 Apr 2025).

4. Magic-angle hierarchy and phase hierarchy in twisted graphene

In condensed-matter theory, “magic hierarchy” has a mathematically unrelated but structurally similar meaning. For alternating-twist multilayer graphene (ATMG), the layer angles are

$p$1

and the continuum Hamiltonian can be mapped exactly to decoupled bilayer Hamiltonians at different effective couplings by an SVD of the interlayer-coupling matrix $p$2. For equal nearest-neighbor couplings, the singular values are

$p$3

so the multilayer magic angles are inherited from the twisted-bilayer sequence by simple rescaling: $p$4 This is the “magic angle hierarchy” of ATMG: the multilayer problem is exactly a direct sum of TBG blocks with rescaled couplings. Consequences include the $p$5 scaling of trilayer magic angles, the golden-ratio $p$6 and $p$7 sequences in the quadrilayer, the coexistence of flat and dispersive bands in odd-layer systems, and a continuum of magic angles for $p$8 in the large-$p$9 limit (1901.10485).

A different hierarchy appears in twisted bilayer graphene itself, where correlated phases are ordered by robustness against deviations from the magic angle $M=\sum_r \omega^{r^3}\ket r\!\bra r,$0, or equivalently by the interaction strength required to stabilize them. Over the device range

$M=\sum_r \omega^{r^3}\ket r\!\bra r,$1

superconductivity near $M=\sum_r \omega^{r^3}\ket r\!\bra r,$2, high-temperature cascade transitions, and linear-in-$M=\sum_r \omega^{r^3}\ket r\!\bra r,$3 resistivity are the more robust phases, whereas correlated insulators near $M=\sum_r \omega^{r^3}\ket r\!\bra r,$4, orbital ferromagnetism or anomalous Hall effect, and incipient Chern insulating behavior are more fragile. Superconductivity is observed from $M=\sum_r \omega^{r^3}\ket r\!\bra r,$5 to $M=\sum_r \omega^{r^3}\ket r\!\bra r,$6, with the strongest $M=\sum_r \omega^{r^3}\ket r\!\bra r,$7 near the magic angle and a hole-side dome reaching $M=\sum_r \omega^{r^3}\ket r\!\bra r,$8 at $M=\sum_r \omega^{r^3}\ket r\!\bra r,$9; by contrast, well-defined $p$00 correlated-insulator gaps occur only in the narrower interval $p$01 to $p$02. Near $p$03, the metallic anomalous Hall phase in the $p$04 device exhibits hysteresis loops in $p$05, a coercive field up to about $p$06 mT, maximal $p$07 about $p$08, and Curie temperature $p$09 K (Polski et al., 2022).

The two graphene usages share the word “hierarchy” but not the same object of classification. In ATMG the hierarchy is exact and algebraic, inherited from singular values of $p$10; in TBG phase diagrams it is empirical and phenomenological, defined by which phases survive as the twist angle moves away from the magic value. This suggests two distinct condensed-matter meanings of the term: one for flat-band kinematics, another for correlated-phase stability.

5. Neutrino-oscillation usages: magic baselines, magic energies, and hierarchy determination

In neutrino phenomenology, “magic” refers to special choices of baseline or energy for which $p$11 becomes independent of the CP phase $p$12. In matter with possible non-standard interactions (NSI), the usual magic-baseline condition works only in restricted regimes. For small $p$13 and standard matter only, the classic magic baseline is

$p$14

but the condition fails in general for small $p$15 with NSI in the $p$16 and $p$17 sectors and for large $p$18 even without NSI. The proposed alternative is a magic-energy condition, obtained by solving the same phase-cancellation criterion for $p$19 instead of $p$20. In the small-$p$21 regime this gives

$p$22

and the resulting transition probability is $p$23-independent even with NSI. The same paper argues that the energy condition remains viable for both small and large $p$24, is experimentally attractive with monoenergetic beams, and can help resolve the neutrino mass hierarchy because the allowed magic energies depend on the sign of $p$25 (Rahman et al., 2010).

A complementary analysis using exact matter evolution in T2K, NO$p$26A, LBNE, and LAGUNA identifies hierarchy- and channel-dependent magic windows. In the energy interval $p$27 GeV, NO$p$28A exhibits several such windows, while LAGUNA has a magic-baseline region only for the inverse-hierarchy $p$29 channel around $p$30–$p$31 GeV. The same work finds that the clearest mass-hierarchy separation in its survey occurs at LAGUNA with $p$32 GeV in the disappearance channels: for $p$33, $p$34 for normal hierarchy versus $p$35 for inverse hierarchy; and for $p$36, $p$37 for normal hierarchy versus $p$38 for inverse hierarchy (Das et al., 2013).

Here, then, “magic hierarchy” is not a hierarchy of states or gates but a hierarchy problem in the neutrino-oscillation sense: magic conditions are used to disentangle CP effects from ordering effects in the mass spectrum. The terminological overlap with quantum-information “magic” is purely lexical.

6. Algebraic and supergravity usages: the Magic Square as a hierarchy of symmetries

In mathematical physics, the relevant “magic” structure is the Magic Square of Lie algebras associated with rank-3 Jordan algebras over division and split composition algebras. Using Tits’ formula,

$p$39

and Vinberg’s symmetric reformulation,

$p$40

the square organizes the automorphism, reduced structure, conformal, and quasi-conformal symmetry algebras associated with $p$41 and its Lorentzian analogue $p$42. “Squaring the Magic” extends the classical Euclidean picture and obtains, beyond the known Freudenthal–Rozenfeld–Tits, Günaydin–Sierra–Townsend, and Barton–Sudbery squares, 7 new Euclidean Magic Squares and 10 new Lorentzian Magic Squares (Cacciatori et al., 2012).

The Lorentzian construction is especially significant because it produces non-compact real forms such as $p$43 and $p$44, and the rows of the square acquire direct physical interpretations as $p$45-duality groups, scalar-manifold isometries, and attractor-orbit stabilizers in supergravity and Einstein–Maxwell sigma-model systems in $p$46. In the Euclidean/Lorentzian comparison, $p$47, $p$48, and $p$49 coincide between $p$50 and $p$51, whereas $p$52 differs. The paper also ties these structures to compactifications of $p$53- and $p$54-theory and to large non-BPS attractor orbits in Lorentzian spacetime (Cacciatori et al., 2012).

This usage is again formally distinct from both quantum-computational and condensed-matter meanings. Yet it retains the same organizing motif: “magic” designates a structured array whose successive rows or columns correspond to progressively larger symmetry groups. A plausible implication is that the persistence of the term across fields reflects not a shared ontology, but a recurring methodological preference for highly constrained ladders—of gates, states, angles, phases, baselines, or Lie-theoretic symmetries—where the exceptional cases admit exact classification.