---
title: Nonlocal Magic Resource
url: https://www.emergentmind.com/topics/nonlocal-magic-resource
type: topic
---

# Nonlocal Magic Resource

Searching arXiv for recent papers on nonlocal magic to ground the article in current literature.
Nonlocal magic resource denotes a family of closely related notions for irreducible non-stabilizerness that cannot be removed by local basis changes, or that becomes a genuinely global property under local many-body dynamics, or that characterizes distributed nonsignaling devices beyond local operations and shared randomness. In all of these usages, the central claim is the same: entanglement alone does not capture the beyond-Clifford structure relevant to universal quantum computation, classical intractability, or distributed nonclassicality; the decisive resource is non-stabilizerness that is either locally irreducible, spatially delocalized, or operationally nonfree relative to an explicitly chosen resource theory [2502.06393][2407.03929][1909.04065].

## 1. Resource-theoretic meanings and terminological scope

Within the stabilizer–Clifford framework, magic is non-stabilizerness relative to the Clifford group. Stabilizer states are Clifford images of computational-basis product states, and Clifford circuits acting on stabilizer inputs remain efficiently classically simulable by the Gottesman–Knill theorem. This motivates magic as the resource measuring the non-Clifford, non-stabilizer structure needed for universal quantum computation and for the breakdown of stabilizer-based classical simulation [2407.03929].

A first source of ambiguity is the choice of free operations. One line of work distinguishes operationally defined stabilizer operations, built from stabilizer state preparation, Clifford unitaries, Pauli measurements, discarding, and classical control, from axiomatically defined completely stabilizer-preserving channels. These coincide only for a single qudit and satisfy \(\mathrm{SO}_n(d)\subsetneq \mathrm{CSP}_n(d)\) for \(n\ge 2\), in direct analogy with the gap between LOCC and separable operations in entanglement theory [2011.11651]. This distinction matters because any “nonlocal” version of magic depends not only on the state class being studied but also on which operations are regarded as free.

A second source of ambiguity is that some literature uses “nonlocal magic resource” in a type-independent distributed-information sense. In that language, any nonsignaling resource not simulable by local operations and shared randomness is a nonlocal magic resource, independent of whether the resource is a state, a Bell box, a steering assemblage, a teleportage, or a general channel. The shared theme is again irreducible nonclassicality, but the free set is LOSR rather than stabilizer operations, and semiquantum channels become a universal encoding type for the full LOSR preorder [1909.04065].

## 2. Bipartite nonlocal magic and its spectral structure

For bipartite states, the standard definition minimizes a magic monotone over local unitaries. For a bipartite pure or mixed state \(\rho_{AB}\), one defines
\[
\mathcal{M}_{AB}(\rho_{AB})=\min_{U_A\otimes U_B} M\!\left((U_A\otimes U_B)\rho_{AB}(U_A^\dagger\otimes U_B^\dagger)\right),
\]
with \(M\) typically chosen as the second stabilizer Rényi entropy. This removes all local-basis-dependent magic and retains only the residual non-stabilizerness that cannot be erased by local basis changes. For two-qubit pure states with Schmidt form \(|\psi(\theta)\rangle=\cos\theta\,|00\rangle+\sin\theta\,|11\rangle\), the resulting nonlocal magic is
\[
\mathcal{M}(|\psi(\theta)\rangle)=\log\!\left(\frac{8}{7+\cos(8\theta)}\right).
\]
It vanishes for product states and Bell stabilizer states, while generic entangled non-stabilizer states have \(\mathcal{M}>0\). The same framework extends to mixed states, and in that setting separable states can already have nonzero \(\mathcal{M}_{AB}\) [2502.06393].

A stronger structural result is that pure-state bipartite nonlocal magic is an intrinsic function of the nonzero Schmidt spectrum. For a Schmidt rank-\(r\) state, a local encoding compresses the state into a minimal \(k=\lceil\log_2 r\rceil\)-qubit core on each side, with all remaining local degrees of freedom factored into stabilizer ancillas. This yields
\[
M_{\mathrm{bp}}(|\Psi\rangle_{AB})=M_{\mathrm{bp}}\!\left(|\phi_r\rangle_{\mathrm{code}}\right),
\]
so BNMR depends only on the ordered nonzero Schmidt coefficients, not on the ambient Hilbert-space dimension. The zero set is correspondingly spectral: BNMR vanishes exactly for flat Schmidt spectra of rank \(r=2^n\), i.e. the spectra realized by bipartite stabilizer states. Around such flat, stabilizer-compatible points, the leading response is quadratic,
\[
M_{\mathrm{bp}}(\boldsymbol{\lambda})=r\,\|\boldsymbol{\eta}\|_2^2+O(\|\boldsymbol{\eta}\|_2^3),
\]
for \(\boldsymbol{\lambda}=\boldsymbol{u}_r+\boldsymbol{\eta}\) with \(\sum_i\eta_i=0\). For Schmidt rank 2, BNMR has the closed form
\[
M_{\mathrm{bp}}=\ln\!\left(\frac{8}{7+\cos 8\theta}\right),
\]
and for generalized GHZ states one finds a hierarchy collapse
\[
M_{\mathrm{bp}}=M_{\mathrm{NL}}=M,
\]
so bipartite, global, and total nonlocal magic coincide exactly [2606.24368].

A complementary spectral formulation rewrites Schmidt-gauged nonlocal magic in terms of Walsh–Hadamard autocorrelations of the entanglement spectrum. For a bipartition with Schmidt-gauged representative \(|\psi\rangle_{\rm Sch}\), the paper defines
\[
M_2^{\rm Sch}=-\log_2\!\left[2^{-m}\sum_{s,k\in\mathbb F_2^m}A_s(k)^4\right],
\]
with \(A_s(k)\) built from autocorrelations \(\sqrt{\lambda_x\lambda_{x\oplus s}}\). This makes explicit that nonlocal magic is governed not only by entropy but by the harmonic organization of the entanglement spectrum. The universal bound
\[
M_2^{\rm Sch}(\psi)\le 2S_2(\rho_A)
\]
shows that entanglement constrains but does not determine nonlocal magic. Flat spectra give \(M_2^{\rm Sch}=0\), Haar-random states have \(M_2^{\rm Sch}=O(1)\), product entanglement spectra can yield extensive \(M_2^{\rm Sch}\), and one-dimensional critical free-fermion ground states produce logarithmic growth \(M_2^{\rm Sch}\sim \kappa\log_2 L\) [2607.07808].

## 3. Delocalized many-body magic in local random dynamics

A distinct usage of “nonlocal magic” emphasizes not a bipartite LU optimization, but the fact that magic generated by strictly local gates rapidly becomes a global many-body property. In brick-wall Haar-random circuits on \(N\) qudits, this is quantified by the Calderbank–Shor–Steane entropy
\[
Y_d(\Psi)=-\log\!\left[\frac{1}{d^N}\sum_{P\in\mathcal P_N(d)}\langle\Psi|P|\Psi\rangle^D\right],
\]
with \(D=2d\) for even \(d\) and \(D=d\) for odd \(d\). \(Y_d\) is a faithful, Clifford-invariant, additive magic monotone, and its MPO structure permits simulations up to \(N\simeq 1024\) qudits. In one-dimensional Haar-random circuits starting from \(|0\rangle^{\otimes N}\), the deficit from Haar saturation obeys
\[
\Delta Y_d(t)=a_d N e^{-\alpha_d t},
\]
with \(\alpha_2\approx 0.43\) for qubits and \(\alpha_3\approx 0.98\) for qutrits, yielding a saturation depth
\[
t_{\mathrm{sat}}^{(Y)}=\frac{1}{\alpha_d}\log N+O(1).
\]
This logarithmic equilibration is qualitatively different from bipartite entanglement entropy, which saturates on times \(t_{\mathrm{sat}}^{(\mathrm{ENT})}\propto N\) [2407.03929].

The same separation appears in tensor-network simulations. For Haar-random qubit circuits, the deviation of the stabilizer Rényi entropy at finite MPS bond dimension obeys
\[
\Delta M_n(\chi)=\lambda_n N e^{-\alpha_n\chi},
\]
so the bond dimension needed to converge \(M_n\) scales only as
\[
\chi_{\mathrm{SRE}}\sim O(\ln N).
\]
By contrast, the entanglement of Haar-random states requires \(\chi\sim 2^{N/2}\). The time-dependent approach to the SRE saturation value likewise follows
\[
\Delta M_n(t)\propto N e^{-\gamma_n t},
\qquad
t_{\mathrm{SRE}}\propto \ln N.
\]
This establishes an information separation: nonlocal magic, as diagnosed by global stabilizer Rényi entropies, is captured at logarithmic bond dimension even when the entanglement structure remains exponentially hard for MPS methods [2509.26342].

These results make precise a common misconception. Nonlocal magic in many-body dynamics is not simply “large entanglement by another name.” The papers instead show that states can become Haar-typical relative to Clifford structure long before long-range entanglement saturates, and that magic growth can remain accurately trackable even when most entanglement has been truncated away [2407.03929][2509.26342].

## 4. Generation, extraction, and computational architectures

At the unitary level, nonlocal magic generation admits an exact operator-space characterization. A unitary map generates nonlocal magic if and only if it generates operator entanglement on Pauli strings. Guided by this equivalence, one can define an average Pauli-entangling power as a proxy for nonlocal magic generation and derive analytical formulas, typical values, and upper bounds in terms of the nonstabilizerness properties of the evolution [2504.09360].

In measurement-based quantum computation, the same resource appears as a graph-wide capacity. The relevant magic monotone is again stabilizer Rényi entropy, but the paper distinguishes three quantities: invested magic \(M_\alpha(U)\), reserved magic \(R([M]|G\rangle)\), and potential magic \(\mathcal P(|G\rangle)=\max_{[M]}M_2([M]|G\rangle)\). Non-Pauli measurements are the only source of magic, yet graph entanglement makes the effect nonlocal. The linear chain and GHZ graph states have
\[
\mathcal P(|\mathrm{Linear}\rangle)=1T,\qquad \mathcal P(|\mathrm{GHZ}\rangle)=1T,
\]
whereas for \(d\)-dimensional graph families the lower bound scales as
\[
\mathcal P(|G\rangle)_{\min}=O\!\big(n^{(d-1)/d}\big).
\]
For the \(n\)-qubit QFT the invested magic obeys
\[
M_2(\mathrm{QFT}_n)\approx 3.4619\,n-5.3388.
\]
In this formulation, nonlocal magic is not the initial graph state itself, which is stabilizer, but the graph’s capacity to convert local non-Pauli measurement angles into globally distributed non-stabilizerness [2408.01980].

Measurements can also reorganize nonlocal magic dynamically. For two-site reduced states in critical many-body systems, the bipartite nonlocal-magic measure \(\mathcal M(r)\) can decay algebraically, and the measurement-induced quantity
\[
\widetilde{\mathcal M}(r)=\sum_i p_i\,\mathcal M(|\Psi_{1,r}^i\rangle)
\]
can decay more slowly than pre-measurement correlations. In monitored Haar-random circuits, this behavior is described as “nonstabilizerness swapping,” directly analogous to entanglement swapping but for non-stabilizer correlations [2502.06393].

Open-system dynamics provides an even stronger example. Under local amplitude damping, the \(n\)-qubit GHZ family \(\alpha|0^n\rangle+\beta|1^n\rangle\) loses magic at \(\gamma_-^{(n)}\), regains it at \(\gamma_+^{(n)}\), and loses entanglement at \(\gamma_e^{(n)}\), with
\[
\gamma_e^{(n)}+\gamma_+^{(n)}=1.
\]
For small \(\alpha\), the reborn magic resides in a fully separable state whose proper marginals are all stabilizer. Yet parity-syndrome extraction followed by Clifford decoding concentrates this global magic onto a single qubit, with a lossless-in-expectation robustness relation
\[
(P_0+P_n)\bigl[\mathcal R(\tilde\rho)-1\bigr]=\mathcal R(\rho_n(\gamma))-1.
\]
This demonstrates that globally distributed nonlocal magic can be operationally harvested and converted into single-qubit distillable magic [2605.22603].

## 5. Holography, scattering theory, and experiment

In holographic gauge theory, nonlocal magic has been linked to antiflatness of the entanglement spectrum. For Schwinger pair creation in a strongly coupled large-\(N_c\) theory, the relevant diagnostic is the refined Rényi entropy \(\widetilde S_n\) and its derivative,
\[
\left.\partial_n\widetilde S_n\right|_{n=1}=-C_E(\rho_A),
\]
where \(C_E\) is the capacity of entanglement. Using Lemma 1 of Cao et al. as quoted in that work,
\[
C_E(\rho_A)=0\Longleftrightarrow \mathcal M^{(\mathrm{NL})}(\psi_{AB})=0.
\]
For the probe-string contribution in the holographic dual, the capacity is
\[
C_E=\frac{\sqrt{\lambda}(d-2)}{(d-1)^3},
\]
so \(C_E>0\) for all boundary dimensions \(d>2\). The same quantity is also the heat capacity extracted from the probe free energy. In this setting, nonlocal magic is encoded in the nonflat entanglement spectrum across a spherical bipartition and in the \(n\)-dependence of the topological black-hole horizon [2605.04210].

A different high-energy realization appears in gluon and graviton scattering. There the outgoing two-particle helicity state defines a two-qubit pure state, and nonlocal magic is the minimal second stabilizer Rényi entropy over local unitary basis changes. For many initial states, including states produced with polarised beams, the physically natural helicity basis already coincides with a basis in which nonlocal magic is manifest. This provides an information-theoretic justification for the helicity basis in scattering-based magic analyses. The property is not universal, however: adding an \(F^3\) operator to the Yang–Mills Lagrangian breaks the coincidence between helicity-basis magic and basis-independent nonlocal magic [2603.04148].

The first direct experimental demonstration of non-local magic was performed on a superconducting quantum processing unit. The experiment used the bipartite definition
\[
M_2^{\rm NL}(\ket\psi)=\min_{U_A,U_B}M_2\big((U_A\otimes U_B)\ket\psi\big)
\]
and, for pure two-qubit states, the reduced-purity formula
\[
M^{\rm NL}(\ket\psi)=-\log_2\!\left(4P(\psi_A)^2-6P(\psi_A)+3\right).
\]
This enabled a direct separation between local and non-local magic, together with a noise model based only on independently characterized readout noise and a depolarizing CZ channel, with no free fitting parameter [2511.15576].

## 6. Distributed nonclassicality, certification, and conceptual boundaries

In the LOSR resource theory, a nonlocal magic resource is any nonsignaling channel that is not simulable by local operations and shared randomness. The framework is type-independent: states, boxes, assemblages, teleportages, semiquantum channels, and general bipartite channels are all treated as resources of different types. A key structural result is that semiquantum channels are universal for this preorder: every nonsignaling resource can be losslessly encoded into semiquantum type, and the family of all semiquantum games completely characterizes the LOSR nonclassicality of every resource in a measurement-device-independent manner [1909.04065].

This distributed perspective connects directly to computational separations. In shallow-circuit complexity, magic is necessary for particular pseudo-telepathic correlations arising from a linear binary constraint system. The paper proves that for suitable instances there exist perfect quantum strategies requiring non-Clifford gates, while every Clifford strategy is bounded away from unit success. This yields an unconditional separation
\[
\mathrm{ClifNC}^0 \neq \mathrm{QNC}^0,
\]
so “magic” becomes a nonlocal computational resource in the precise sense that perfect nonlocal-game correlations cannot be generated by magic-free shallow circuits [2402.12246].

The same theme appears in Bell certification from tomography. Using only Pauli-basis measurements already present in standard state tomography, one can construct tailored Bell inequalities and evaluate them directly from the correlation tensor. For some of these inequalities, the stabilizer maximum lies below the quantum value attained by the target state, so a violation of the stabilizer bound witnesses magic with no additional experimental cost. This establishes a three-layer interpretation of the same dataset: state characterization, Bell-nonlocality certification, and magic witnessing [2512.25068].

Across these literatures, three misconceptions are repeatedly corrected. First, nonlocal magic is not synonymous with entanglement: Bell pairs and flat entanglement spectra can have zero nonlocal magic. Second, it is not reducible to local non-stabilizerness: LU optimization, Schmidt-gauge constructions, and experimental local-erasure protocols all isolate a residual component that survives local basis changes. Third, it is not a single universally fixed quantity: the phrase can denote LU-irreducible bipartite non-stabilizerness, globally delocalized many-body magic, or LOSR-nonfree distributed resources, depending on the underlying resource theory and operational question. The common invariant across these meanings is irreducible beyond-stabilizer structure that cannot be captured by entanglement or Bell nonlocality alone [2502.06393][2606.24368][2407.03929].

Source: https://www.emergentmind.com/topics/nonlocal-magic-resource