Genuine Multipartite Entanglement (GME)

• GME is defined as an N-partite quantum state that cannot be decomposed into a convex mixture of bipartition-separable states, representing the strongest form of multipartite entanglement.
• Recent research quantifies GME using geometric mean constructions and concurrence-based methods, clarifying the distinction between full inseparability and biseparability.
• Experimental detection of GME employs macroscopic observables, phase-space witnesses, and multiple-copy activation, bridging theoretical measures with practical quantum systems.

Genuine multipartite entanglement (GME) is the strongest standard notion of multipartite entanglement for a quantum state: an $N$-partite state is genuinely multipartite entangled if it is not biseparable, i.e., if it cannot be written as a convex mixture of states that are separable with respect to some bipartition. In current research, GME functions simultaneously as a structural classification of multipartite correlations, a target for entanglement quantification, and a benchmark for experiments ranging from continuous-variable optics to many-qubit devices and many-electron systems (Gerke et al., 2016, Palazuelos et al., 2022).

1. Definition, biseparability, and adjacent notions

The standard mixed-state definition is formulated through biseparable decompositions. For an $N$-partite system with Hilbert space $$\mathcal{H}=\mathcal{H}_1\otimes\cdots\otimes\mathcal{H}_N$$, the set of non-GME states contains all convex mixtures of bipartition-separable states. In the notation used for continuous-variable quantum optics,

$$\hat\sigma_2 = \sum_{\text{bipartitions } \mathcal{I}_1:\mathcal{I}_2} p_{\mathcal{I}_1:\mathcal{I}_2}\,\hat\sigma_{\mathcal{I}_1:\mathcal{I}_2},$$

where each $\hat\sigma_{\mathcal{I}_1:\mathcal{I}_2}$ is separable with respect to $\mathcal{I}_1:\mathcal{I}_2$. A state is genuinely $N$-partite entangled precisely when it cannot be written in this form (Gerke et al., 2016). An equivalent formulation used in the multiple-copy setting is that a biseparable state admits a convex mixture over all possible bipartitions,

$\rho=\sum_{M\subsetneq [n]} p_M \chi_M,$

with each $\chi_M$ separable across $M\mid \bar M$ (Palazuelos et al., 2022).

This definition sharply separates GME from weaker notions. A state may be entangled across every individual bipartition and still fail to be GME if it remains a convex mixture of bipartition-separable states. The literature therefore distinguishes at least three levels: partition entanglement, full inseparability, and genuine multipartite entanglement. The six-mode continuous-variable experiment on a state generated by parametric downconversion of a femtosecond frequency comb is a concrete example: the measured $N$0 covariance matrix showed entanglement across all individual partitions and strong evidence for $N$1-partite entanglement for all $N$2, but no violation of the $N$3 criterion, so no GME was detected (Gerke et al., 2016).

This distinction matters conceptually. Failure to certify GME does not imply absence of multipartite quantum correlations. The continuous-variable results explicitly show that a state can be entangled across all bipartitions yet still be biseparable in the strict convex-geometric sense (Gerke et al., 2016). A plausible implication is that GME should be viewed as one particularly stringent layer inside a broader hierarchy of multipartite correlation structures rather than as an exhaustive descriptor of multipartite entanglement.

2. Quantification: geometric means, concurrence constructions, and AME-sensitive measures

A major line of recent work constructs GME measures from bipartition entanglement data. A unified framework defines, for a pure $N$4-partite state $N$5, a GME measure based on the geometric mean over all bipartitions,

$N$6

where $N$7 is an entanglement quantifier for the bipartition $N$8 induced by a symmetric concave function $N$9, and $$\mathcal{H}=\mathcal{H}_1\otimes\cdots\otimes\mathcal{H}_N$$0 is the total number of bipartitions. With suitable choices of $$\mathcal{H}=\mathcal{H}_1\otimes\cdots\otimes\mathcal{H}_N$$1, this reproduces GME concurrence, GME negativity, the geometric measure of GME, and GME G-concurrence. The construction is reported to satisfy zero-on-biseparable states, convexity, LOCC monotonicity, smoothness, and scalability (Wang et al., 14 Aug 2025).

A closely related program develops explicitly geometric measures. For arbitrary multipartite pure states, a geometric GME measure based on the volume of a concurrence regular polygonal pyramid is defined by

$$\mathcal{H}=\mathcal{H}_1\otimes\cdots\otimes\mathcal{H}_N$$2

where $$\mathcal{H}=\mathcal{H}_1\otimes\cdots\otimes\mathcal{H}_N$$3 is built from the geometric mean of all $$\mathcal{H}=\mathcal{H}_1\otimes\cdots\otimes\mathcal{H}_N$$4-to-rest concurrences and $$\mathcal{H}=\mathcal{H}_1\otimes\cdots\otimes\mathcal{H}_N$$5 from the geometric mean of the remaining bipartition concurrences. The measure is positive if and only if the state is genuinely $$\mathcal{H}=\mathcal{H}_1\otimes\cdots\otimes\mathcal{H}_N$$6-partite entangled, is monotonic under LOCC, and reduces to the four-partite pyramid measure for $$\mathcal{H}=\mathcal{H}_1\otimes\cdots\otimes\mathcal{H}_N$$7. In the four-qubit examples reported there, $$\mathcal{H}=\mathcal{H}_1\otimes\cdots\otimes\mathcal{H}_N$$8 and $$\mathcal{H}=\mathcal{H}_1\otimes\cdots\otimes\mathcal{H}_N$$9, so the GHZ state is assigned greater GME than the W state (Zhao et al., 2024).

Concurrence-based geometric constructions also appear in triangle and quadrilateral forms. For three parties, the concurrence-triangle approach normalizes the area of a triangle with side lengths given by one-versus-rest concurrences,

$$\hat\sigma_2 = \sum_{\text{bipartitions } \mathcal{I}_1:\mathcal{I}_2} p_{\mathcal{I}_1:\mathcal{I}_2}\,\hat\sigma_{\mathcal{I}_1:\mathcal{I}_2},$$0

with $$\hat\sigma_2 = \sum_{\text{bipartitions } \mathcal{I}_1:\mathcal{I}_2} p_{\mathcal{I}_1:\mathcal{I}_2}\,\hat\sigma_{\mathcal{I}_1:\mathcal{I}_2},$$1. For general $$\hat\sigma_2 = \sum_{\text{bipartitions } \mathcal{I}_1:\mathcal{I}_2} p_{\mathcal{I}_1:\mathcal{I}_2}\,\hat\sigma_{\mathcal{I}_1:\mathcal{I}_2},$$2, the construction uses geometric means of areas of “concurrence triangles” and extends to mixed states through the convex roof (Jin et al., 2022). For four-qubit systems, an alternative 2D construction uses three quadrilaterals built from one-to-three and two-to-two bipartition concurrences; this was proposed partly in response to the observation that “concurrence fill” for three parties is not monotonic under LOCC and hence not a faithful entanglement measure (Mishra et al., 2022).

Parameterized generalizations have also appeared. The G$$\hat\sigma_2 = \sum_{\text{bipartitions } \mathcal{I}_1:\mathcal{I}_2} p_{\mathcal{I}_1:\mathcal{I}_2}\,\hat\sigma_{\mathcal{I}_1:\mathcal{I}_2},$$3C family defines

$$\hat\sigma_2 = \sum_{\text{bipartitions } \mathcal{I}_1:\mathcal{I}_2} p_{\mathcal{I}_1:\mathcal{I}_2}\,\hat\sigma_{\mathcal{I}_1:\mathcal{I}_2},$$4

the geometric mean of bipartite $$\hat\sigma_2 = \sum_{\text{bipartitions } \mathcal{I}_1:\mathcal{I}_2} p_{\mathcal{I}_1:\mathcal{I}_2}\,\hat\sigma_{\mathcal{I}_1:\mathcal{I}_2},$$5-concurrences over all bipartitions. It is reported to satisfy non-negativity, convexity, LOCC monotonicity, local-unitary invariance, strong monotonicity, and continuity for pure states. Analytical expressions are given for $$\hat\sigma_2 = \sum_{\text{bipartitions } \mathcal{I}_1:\mathcal{I}_2} p_{\mathcal{I}_1:\mathcal{I}_2}\,\hat\sigma_{\mathcal{I}_1:\mathcal{I}_2},$$6-qubit GHZ and W states, again with the GHZ states more genuinely entangled than the W states according to this family (Wang et al., 9 May 2025).

A distinct direction is to interpolate between GME and absolutely maximally entangled (AME) states. The GME-AME multipartite entanglement measure

$$\hat\sigma_2 = \sum_{\text{bipartitions } \mathcal{I}_1:\mathcal{I}_2} p_{\mathcal{I}_1:\mathcal{I}_2}\,\hat\sigma_{\mathcal{I}_1:\mathcal{I}_2},$$7

is zero for biseparable states, positive for GME states, and equal to $$\hat\sigma_2 = \sum_{\text{bipartitions } \mathcal{I}_1:\mathcal{I}_2} p_{\mathcal{I}_1:\mathcal{I}_2}\,\hat\sigma_{\mathcal{I}_1:\mathcal{I}_2},$$8 if and only if the state is AME. For four-partite qutrit permutation states, 33 distinct classes by measure value were reported, including 72 states with $$\hat\sigma_2 = \sum_{\text{bipartitions } \mathcal{I}_1:\mathcal{I}_2} p_{\mathcal{I}_1:\mathcal{I}_2}\,\hat\sigma_{\mathcal{I}_1:\mathcal{I}_2},$$9 and 72 with $\hat\sigma_{\mathcal{I}_1:\mathcal{I}_2}$0 (V et al., 2024).

3. Detection and certification without full tomography

Because mixed-state GME is difficult to quantify exactly, much of the literature focuses on sufficient criteria and experimentally feasible witnesses. One macroscopic approach generalizes the entanglement gap to multipartite settings. Given a Hamiltonian $\hat\sigma_{\mathcal{I}_1:\mathcal{I}_2}$1, one defines

$\hat\sigma_{\mathcal{I}_1:\mathcal{I}_2}$2

and if $\hat\sigma_{\mathcal{I}_1:\mathcal{I}_2}$3, then $\hat\sigma_{\mathcal{I}_1:\mathcal{I}_2}$4 is $\hat\sigma_{\mathcal{I}_1:\mathcal{I}_2}$5-inseparable; for $\hat\sigma_{\mathcal{I}_1:\mathcal{I}_2}$6, this is the “GME gap” (Gabriel et al., 2012). The method uses only macroscopic observables such as energy and was shown on Heisenberg and spin-1 models to detect regions not detected by the $\hat\sigma_{\mathcal{I}_1:\mathcal{I}_2}$7 criteria (Gabriel et al., 2012).

Local-observable criteria provide another route. A sufficient condition based on local sum uncertainty relations was developed for arbitrary multipartite systems. For noisy $\hat\sigma_{\mathcal{I}_1:\mathcal{I}_2}$8-qubit W states,

$\hat\sigma_{\mathcal{I}_1:\mathcal{I}_2}$9

the criterion was applied for $\mathcal{I}_1:\mathcal{I}_2$0, and the reported thresholds were $\mathcal{I}_1:\mathcal{I}_2$1, $\mathcal{I}_1:\mathcal{I}_2$2, $\mathcal{I}_1:\mathcal{I}_2$3, and $\mathcal{I}_1:\mathcal{I}_2$4, respectively; it was also applied to a three-qutrit state and was reported stronger than criteria based on GME concurrence and Fisher information (Li et al., 2021).

Positive-map methods have recently been reformulated in terms of truncated moments. For a multipartite state $\mathcal{I}_1:\mathcal{I}_2$5, one considers moments

$\mathcal{I}_1:\mathcal{I}_2$6

of a GME map built from positive but not completely positive maps across bipartitions, and constructs Hankel matrices from a finite set of these moments. For all biseparable states the Hankel matrices are positive semidefinite; a negative determinant therefore witnesses GME. The proposal is explicitly designed to avoid full tomography and to permit polynomial-scaling experimental estimation through SWAP-type observables and shadow-tomography-style protocols (Mukherjee et al., 30 May 2025).

Measurement restrictions have driven further specialized criteria. For graph states, a family of witnesses based on a small subset of stabilizers requires only $\mathcal{I}_1:\mathcal{I}_2$7 out of $\mathcal{I}_1:\mathcal{I}_2$8 stabilizers, and only $\mathcal{I}_1:\mathcal{I}_2$9-body correlators with

$N$0

This was proposed for platforms with restricted connectivity and analyzed analytically for white-noise graph states and numerically for microwave photonic qubits (Li et al., 29 Apr 2025). A different operational framework uses correlations in mutually unbiased bases (MUBs): for tripartite and quadripartite qubit systems, all biseparable states satisfy $N$1 and $N$2, whereas GME states can violate these bounds (Nandi, 28 Sep 2025).

Data-driven detection has also emerged. Convolutional neural networks and CNNs enhanced with squeeze-and-excitation modules were trained on SDP-labeled GME data and GHZ-diagonal states. For GHZ-diagonal states from 4 to 20 qubits, both CNN and CNN-SE achieved $N$3 accuracy, with CNN-SE consistently $N$4 even up to 20 qubits; the work also analyzed false positives and false negatives and emphasized reduction of false positives (Luo et al., 19 Aug 2025).

4. Multiple-copy activation and the resource-theoretic tension

In the multiple-copy regime, GME exhibits a nontrivial activation phenomenon. A state is called GME-activatable if there exists $N$5 such that $N$6 is GME. A central theorem states that an $N$7-partite state is GME-activatable if and only if it is not partially separable across one bipartition (Palazuelos et al., 2022). The same work proves that there is no universal upper bound on the number of copies needed for activation: for any $N$8 and any $N$9, there exist fixed-local-dimension GME-activatable states whose first $\rho=\sum_{M\subsetneq [n]} p_M \chi_M,$0 copies remain biseparable (Palazuelos et al., 2022).

In continuous-variable systems, this phenomenon extends beyond finite dimensions. Examples were constructed of non-Gaussian states that are biseparable but fully inseparable and whose two-copy state is detected as GME. For Gaussian states, the covariance-matrix biseparability criterion was shown not to be sufficient even within the Gaussian sector: there exist Gaussian states satisfying the criterion yet being GME, or at least GME-activatable (Baksová et al., 2023).

These results create a resource-theoretic tension. If biseparable states can yield GME after copying, then biseparability is not tensor stable. One analysis argues that activation of GME from multiple copies of GME-free states necessarily involves entangling operations, even when those operations are “local” with respect to enlarged laboratories containing multiple subsystems (Wieśniak, 2024). This suggests that the physical meaning of “party” and the allowed operation set are indispensable when interpreting GME as a resource.

The activation phenomenon has now been demonstrated experimentally. In a trapped-ion processor, two copies of a biseparable but fully inseparable three-qubit state were prepared and shown to possess GME through a semidefinite-program-derived witness. The measured expectation value was

$\rho=\sum_{M\subsetneq [n]} p_M \chi_M,$1

more than 11 standard deviations below zero, while the individual copies were separately certified as biseparable (Stárek et al., 14 Oct 2025).

5. Continuous variables, Gaussian criteria, and phase-space witnesses

Continuous-variable platforms have sharpened the distinction between GME and more general multipartite entanglement. For Gaussian states described by covariance matrices, entanglement witnesses quadratic in quadratures can test $\rho=\sum_{M\subsetneq [n]} p_M \chi_M,$2-separability via

$\rho=\sum_{M\subsetneq [n]} p_M \chi_M,$3

with negative $\rho=\sum_{M\subsetneq [n]} p_M \chi_M,$4 indicating significant entanglement. In the six-mode SPOPO experiment, no GME was detected at $\rho=\sum_{M\subsetneq [n]} p_M \chi_M,$5, but inseparability with respect to convex combinations of $\rho=\sum_{M\subsetneq [n]} p_M \chi_M,$6-partitions was found for $\rho=\sum_{M\subsetneq [n]} p_M \chi_M,$7 (Gerke et al., 2016).

The covariance-matrix perspective is powerful but incomplete. The multi-copy activation analysis explicitly showed that no criterion based only on first and second moments can detect all GME, even for Gaussian states, because Gaussian states with the same covariance data can differ in whether they are GME or merely biseparable or activatable (Baksová et al., 2023). This limitation has motivated phase-space criteria that access non-Gaussianity more directly.

A recent development connects GME to Wigner negativity. For an $\rho=\sum_{M\subsetneq [n]} p_M \chi_M,$8-mode continuous-variable system, sufficiently large negativity volume along a suitable two-dimensional slice of the Wigner function certifies GME, and sufficiently large nonclassicality depth of the center-of-mass mode also certifies GME. In particular,

$\rho=\sum_{M\subsetneq [n]} p_M \chi_M,$9

Violations of the corresponding smoothed-Wigner criterion provide lower bounds on the trace distance to the set of non-GME states, and the criteria are designed for native phase-space measurements in circuit/cavity QED, trapped ions and atoms, and circuit quantum acoustodynamic systems (Zaw et al., 30 Oct 2025).

This suggests a shift in continuous-variable GME detection from covariance-only methods toward hybrid criteria that incorporate higher-order or quasiprobability information. The practical implication stated in this line of work is that finite-region Wigner sampling or a finite number of characteristic-function measurements can be sufficient for certification (Zaw et al., 30 Oct 2025).

6. Physical generation, dynamics, and applications

GME is increasingly studied as a physically extractable or dynamically generated resource rather than only as an abstract classification. In many-electron systems, it was shown that high GME can be extracted from closed-shell molecular states even in the non-interacting limit, using projections onto localized orbitals. For benzene, the extracted six-spin $\chi_M$0-electron state was reported to have $\chi_M$1, and this value was practically unchanged between mean-field and CASSCF descriptions; by contrast, linear dimerized molecules such as hexatriene and decapentaene displayed reduced extracted GME, with strong pairwise singlet structure competing against global multipartite entanglement (Troiani et al., 2024).

Open-system generation has also been studied in explicitly thermal settings. In a chain of $\chi_M$2 LC resonators coupled optomechanically to a common thermal acoustic reservoir with a frequency-comb spectrum, exact analytical evolution predicts periodic generation of non-Gaussian multipartite entanglement, including GME, even though the system is in a heat bath. At stroboscopic times the model also generates multipartite cat states of the GHZ form with high fidelity, and quantum Fisher information is used as the main certification tool (Qiu et al., 2024).

Unified geometric-mean measures have further been applied to dynamics and relativistic settings. Fidelity-based lower bounds, designed to avoid full tomography, were used to analyze genuine quadripartite entanglement sudden death