Entanglement Matrix in Quantum Systems

• Entanglement Matrix is a collection of matrix-valued constructs that encode quantum correlations through properties like rank, spectrum, and block structure.
• It spans multiple frameworks including reduced density matrices for bipartite states, bond spaces in MPS, covariance criteria, and graph-theoretic adjacency matrices.
• These matrices enable quantification of entanglement entropy, detection of nonseparability, and analysis of emergent geometry in quantum many-body systems.

“Entanglement matrix” does not denote a single universally standardized object in quantum information or many-body theory. In the literature surveyed here, it refers to several matrix-valued constructions that encode entanglement in different ways: the reduced density matrix $\rho_A$, the finite bond space and transfer matrices of a matrix product state (MPS), covariance and commutation matrices used in separability tests, generalized adjacency matrices that reproduce subsystem entropies, and, in some explicitly named graph-state formalisms, midpoint-indexed matrices defined from geometric cuts. A plausible unifying description is that an entanglement matrix is any matrix-valued object whose rank, spectrum, singular values, positivity, or block structure encodes quantum correlations across a partition (0712.1976, Tripathi et al., 2018, Roy et al., 2019).

1. Terminology and recurrent forms

Several distinct matrix objects are called, or function as, an entanglement matrix in different subfields. Some papers explicitly state that they do not introduce an object literally called an entanglement matrix, while nevertheless using a matrix that plays that role operationally.

Setting	Matrix object	Role
Bipartite quantum states	$$\rho_A=\operatorname{Tr}_{\bar A}|\psi\rangle\langle\psi|$$	Canonical carrier of entanglement entropy
MPS and tensor networks	Bond space, transfer matrix, overlap transfer matrix	Finite-rank entanglement support and propagation
Covariance criteria	$V+\frac{i}{2}\Omega$, or its partially transposed form	Entanglement detection from moments
Entanglement geometry	$J_{ij}$	Reconstructs $S_A$ as a cut weight
Graph-state classification	Midpoint-indexed symmetric matrix	Paper-specific bookkeeping of graph-state entanglement

In one line of work, the “entanglement adjacency matrix” $J$ is defined by

$S_A=\sum_{i\in A}\sum_{j\in \bar A}J_{ij},$

so subsystem entropy is represented as the weight of a graph cut (Roy et al., 2019). In another, finite-$\chi$ MPS are interpreted as finite-rank approximations to the half-chain reduced density matrix, with the bond indices acting as the effective entanglement-carrying matrix space (0712.1976). In covariance-based detection, the matrix

$$V_{jk}=\frac{1}{2}\langle\{\xi_j,\xi_k\}\rangle-\langle\xi_j\rangle\langle\xi_k\rangle$$

and its partially transposed counterpart become the relevant entanglement witness (Tripathi et al., 2018). A separate graph-state paper explicitly introduces an “Entanglement Matrix,” but its midpoint-based definition is nonstandard and specific to that formalism (Sharma, 15 Jul 2025).

2. Reduced density matrices and spectral support

The canonical matrix formulation of bipartite entanglement begins with the Schmidt decomposition

$$\ket{\Psi}=\sum_{\alpha=1}^{\min(\dim {\mathcal H}_A,\dim {\mathcal H}_B)} \lambda_{\alpha} \ket{\varphi_{\alpha}^A} \ket{\varphi_{\alpha}^B}, \qquad \sum_\alpha |\lambda_\alpha|^2=1,$$

and the reduced density matrix $$\rho_A=\operatorname{Tr}_{\bar A}|\psi\rangle\langle\psi|$$0, whose eigenvalues are $$\rho_A=\operatorname{Tr}_{\bar A}|\psi\rangle\langle\psi|$$1. The von Neumann entropy is

$$\rho_A=\operatorname{Tr}_{\bar A}|\psi\rangle\langle\psi|$$2

so the reduced density matrix is the most direct entanglement matrix in standard bipartite theory (0712.1976).

This remains the operative matrix object in many-body eigenstate studies. For the spin-$$\rho_A=\operatorname{Tr}_{\bar A}|\psi\rangle\langle\psi|$$3 XXZ chain, the entanglement of an eigenstate $$\rho_A=\operatorname{Tr}_{\bar A}|\psi\rangle\langle\psi|$$4 is defined by

$$\rho_A=\operatorname{Tr}_{\bar A}|\psi\rangle\langle\psi|$$5

That work emphasizes that $$\rho_A=\operatorname{Tr}_{\bar A}|\psi\rangle\langle\psi|$$6 is the only explicit entanglement-related matrix used there; no entanglement spectrum or modular Hamiltonian is introduced (LeBlond et al., 2019).

A different but closely related matrix appears in entanglement negativity. Negativity is defined from the partial transpose,

$$\rho_A=\operatorname{Tr}_{\bar A}|\psi\rangle\langle\psi|$$7

so the entanglement content is encoded not by $$\rho_A=\operatorname{Tr}_{\bar A}|\psi\rangle\langle\psi|$$8 but by the singular-value structure of $$\rho_A=\operatorname{Tr}_{\bar A}|\psi\rangle\langle\psi|$$9. The partial transpose is represented explicitly by a fixed permutation matrix $V+\frac{i}{2}\Omega$0 acting on $V+\frac{i}{2}\Omega$1,

$V+\frac{i}{2}\Omega$2

and the derivative of $V+\frac{i}{2}\Omega$3 is controlled by the matrix sign of $V+\frac{i}{2}\Omega$4 (Cresswell et al., 2018). In this formulation, the entanglement matrix is the partially transposed density matrix together with the linear superoperator that implements partial transposition.

3. Matrix product states and finite entanglement support

In MPS theory, entanglement is encoded concretely in the bond indices linking local tensors rather than in an abstract auxiliary object. For

$V+\frac{i}{2}\Omega$5

with each $V+\frac{i}{2}\Omega$6 a $V+\frac{i}{2}\Omega$7 matrix, the Schmidt rank across any bond cut is at most $V+\frac{i}{2}\Omega$8. Consequently, the reduced density matrix has rank at most $V+\frac{i}{2}\Omega$9, and

$J_{ij}$0

The bond space therefore acts as a finite-dimensional entanglement channel, or, equivalently, as the effective finite-rank support of the reduced density matrix (0712.1976).

At criticality this finite support produces a characteristic finite-entanglement scaling. For the quantum Ising chain the half-chain entropy satisfies

$J_{ij}$1

while the induced correlation length obeys

$J_{ij}$2

for Ising and $J_{ij}$3 for Heisenberg (0712.1976). The transfer matrix built from local tensors supplies the effective correlation length through the ratio of its two leading eigenvalues, so the same finite-dimensional matrix structure governs both entanglement support and long-range correlations.

A related matrix notion appears in geometric entanglement. For periodic finite chains, the overlap between an MPS ground state $J_{ij}$4 and a separable state $J_{ij}$5 is encoded in the transfer matrix

$J_{ij}$6

whose dominant eigenpairs determine fidelities and gradients in the optimization of geometric entanglement per lattice site (Hu et al., 2011). This is not a reduced-density-matrix formalism; the relevant matrix is an overlap transfer matrix.

In symmetric-state entanglement classification, the local matrices themselves become the classifier. A symmetric $J_{ij}$7-qubit state can be written as

$J_{ij}$8

equivalently as a diagonal MPS with diagonal Kraus operators $J_{ij}$9. The minimal diagonal bond dimension $S_A$0 is then used to define entanglement families compatible with SLOCC, and reduced-density-matrix rank and kernel data determine parent-Hamiltonian interaction length (Sanz et al., 2015). Here the entanglement matrix is not a single matrix but the MPS matrix structure itself.

4. Covariance matrices, positive maps, and matrix inequalities

A second major usage treats entanglement matrices as witness matrices constructed from moments. For an arbitrary set of Hermitian observables $S_A$1, the covariance and commutation matrices are

$S_A$2

Every physical state satisfies the generalized uncertainty relation

$S_A$3

For separable bipartite states, the partially transposed matrices must satisfy

$S_A$4

Violation implies entanglement (Tripathi et al., 2018). In this setting the entanglement matrix is the Hermitian matrix $S_A$5, whose negative eigenvalues certify nonseparability.

A broader matrix-inequality framework starts from immanant inequalities for positive semidefinite matrices $S_A$6, promotes them to Young-projector operator inequalities, and then to positive maps and witnesses. The key operator-valued matrices are the Young projectors $S_A$7, the filtered multilinear maps

$S_A$8

and the witness matrices

$S_A$9

Entanglement is detected by failure of positivity of the map output or negativity of $J$0 (Rico, 25 Feb 2025). This usage treats an entanglement matrix as a positivity-test matrix derived from symmetric matrix inequalities.

Tripartite detection based on principal basis matrix representations uses the correlation coefficients

$J$1

to build matrix slices $J$2, and then linear combinations such as

$J$3

Necessary conditions for separability are then expressed as bounds on $J$4 for qubits, or on phase-weighted sums $J$5 for qudits (Zhao et al., 2022). These are explicitly entanglement-sensitive matrices extracted from a three-body correlation tensor.

5. Geometric and graph-theoretic adjacency matrices

A geometrical use of the term appears in the entanglement adjacency matrix $J$6, a symmetric weighted matrix on the set of sites, defined by the exact or approximate relation

$J$7

When this representation is exact and $J$8, one has

$J$9

and more generally

$S_A=\sum_{i\in A}\sum_{j\in \bar A}J_{ij},$0

Thus $S_A=\sum_{i\in A}\sum_{j\in \bar A}J_{ij},$1 functions as a generalized adjacency matrix whose cut weights reproduce entanglement entropies (Roy et al., 2019).

This representation yields a natural entanglement contour,

$S_A=\sum_{i\in A}\sum_{j\in \bar A}J_{ij},$2

and, for translationally invariant one-dimensional systems, a recursion relation

$S_A=\sum_{i\in A}\sum_{j\in \bar A}J_{ij},$3

For a CFT interval on the infinite line, the continuum kernel is

$S_A=\sum_{i\in A}\sum_{j\in \bar A}J_{ij},$4

which the paper interprets formally as the two-point function of an entanglement current (Roy et al., 2019). In this literature the entanglement matrix is literal: it is the weighted graph matrix from which all cut entropies are reconstructed.

A distinct, explicitly named but nonstandard graph-state formalism defines the Entanglement Matrix as a symmetric matrix indexed by edge midpoints rather than by qubits. Primary midpoints, associated with adjacent qubits, contribute to diagonal and off-diagonal entries; secondary midpoints contribute only to diagonal entries. The total entanglement is then the sum of the diagonal entries and one triangular half of the matrix. For complete odd-$S_A=\sum_{i\in A}\sum_{j\in \bar A}J_{ij},$5 graph states the paper reports

$S_A=\sum_{i\in A}\sum_{j\in \bar A}J_{ij},$6

while for even $S_A=\sum_{i\in A}\sum_{j\in \bar A}J_{ij},$7 it gives piecewise formulas, including

$S_A=\sum_{i\in A}\sum_{j\in \bar A}J_{ij},$8

and, for even multiples of $S_A=\sum_{i\in A}\sum_{j\in \bar A}J_{ij},$9,

$\chi$0

That work also states that “The reason why this happens is still not clear,” so these midpoint formulas should be read as properties of that paper’s custom bookkeeping scheme rather than as standard graph-state entanglement theory (Sharma, 15 Jul 2025).

6. Matrix degrees of freedom, gauge constraints, and emergent geometry

In matrix-model and gauge/gravity settings, the phrase shifts meaning again: the relevant subsystem is a subset of matrix degrees of freedom. In gauged multi-matrix models, diagonal matrix entries encode D-brane positions and off-diagonal entries encode strings stretched between branes. After gauge fixing adapted to a planar target-space cut, the problem maps to a $\chi$1 gauge theory on a complete graph with extended Hilbert space

$\chi$2

Tracing out hidden vertices and links yields a reduced state block-diagonal in charge sectors,

$\chi$3

with entropy

$\chi$4

Here the “entanglement matrix” is the reduced density matrix for a matrix-defined subsystem, together with the matrix block structure induced by gauge constraints (Hampapura et al., 2020).

A related proposal called matrix entanglement defines subsystems directly in the extended Hilbert space of matrix/color degrees of freedom,

$\chi$5

with physical meaning supplied by wave packets $\chi$6 whose matrix centers encode emergent D-brane or black-hole geometry (Gautam et al., 2022). In that setting the “matrix” is not a matrix of entropies but the underlying matrix-valued dynamical variable whose blocks define the subsystem.

The Quantum Hall Matrix Model provides an explicit emergent-space calculation. A cut through the droplet yields two contributions: a boundary term from the collective eigenvalue mode,

$\chi$7

and a bulk term from duplicated off-diagonal matrix elements crossing the cut,

$\chi$8

so that $\chi$9 (Frenkel et al., 2021). In this example the singular values of the cut-crossing off-diagonal block play the decisive entanglement role, making the singular-value structure of a matrix block directly geometric.

7. Kernel-based formulations and conceptual boundaries

Some recent work treats an operator kernel itself as the entangling matrix. In Matrix theory, integrating out heavy off-diagonal modes leads to a Gaussian kernel

$$V_{jk}=\frac{1}{2}\langle\{\xi_j,\xi_k\}\rangle-\langle\xi_j\rangle\langle\xi_k\rangle$$0

from which the reduced density matrix of the probe is controlled by

$$V_{jk}=\frac{1}{2}\langle\{\xi_j,\xi_k\}\rangle-\langle\xi_j\rangle\langle\xi_k\rangle$$1

The entanglement entropy is then a function of $$V_{jk}=\frac{1}{2}\langle\{\xi_j,\xi_k\}\rangle-\langle\xi_j\rangle\langle\xi_k\rangle$$2, and in the weak-coupling regime

$$V_{jk}=\frac{1}{2}\langle\{\xi_j,\xi_k\}\rangle-\langle\xi_j\rangle\langle\xi_k\rangle$$3

The same paper also introduces a covariant $$V_{jk}=\frac{1}{2}\langle\{\xi_j,\xi_k\}\rangle-\langle\xi_j\rangle\langle\xi_k\rangle$$4-tensor,

$$V_{jk}=\frac{1}{2}\langle\{\xi_j,\xi_k\}\rangle-\langle\xi_j\rangle\langle\xi_k\rangle$$5

as the geometric counterpart of the entanglement kernel (Gray et al., 2021).

In an operator-algebraic scattering formulation, the two-body scattering kernel

$$V_{jk}=\frac{1}{2}\langle\{\xi_j,\xi_k\}\rangle-\langle\xi_j\rangle\langle\xi_k\rangle$$6

acts on a bipartite representation space, while a product input $$V_{jk}=\frac{1}{2}\langle\{\xi_j,\xi_k\}\rangle-\langle\xi_j\rangle\langle\xi_k\rangle$$7 produces the outgoing coefficient matrix

$$V_{jk}=\frac{1}{2}\langle\{\xi_j,\xi_k\}\rangle-\langle\xi_j\rangle\langle\xi_k\rangle$$8

The reduced density matrix is

$$V_{jk}=\frac{1}{2}\langle\{\xi_j,\xi_k\}\rangle-\langle\xi_j\rangle\langle\xi_k\rangle$$9

and the entanglement is computed from the singular values of $$\ket{\Psi}=\sum_{\alpha=1}^{\min(\dim {\mathcal H}_A,\dim {\mathcal H}_B)} \lambda_{\alpha} \ket{\varphi_{\alpha}^A} \ket{\varphi_{\alpha}^B}, \qquad \sum_\alpha |\lambda_\alpha|^2=1,$$0 (Lyu et al., 12 Nov 2025). This again uses a matrix whose singular-value structure is the entanglement carrier, but it is neither a reduced density matrix nor a covariance matrix.

A recent review of matrix quantum mechanics makes the broader terminological point explicit: MQM does not introduce a standard object literally called an “entanglement matrix,” and the closest constructions are reduced density matrices for matrix-defined subsystems, projector/block decompositions

$$\ket{\Psi}=\sum_{\alpha=1}^{\min(\dim {\mathcal H}_A,\dim {\mathcal H}_B)} \lambda_{\alpha} \ket{\varphi_{\alpha}^A} \ket{\varphi_{\alpha}^B}, \qquad \sum_\alpha |\lambda_\alpha|^2=1,$$1

Gaussian correlation kernels, and representation-block decompositions of $$\ket{\Psi}=\sum_{\alpha=1}^{\min(\dim {\mathcal H}_A,\dim {\mathcal H}_B)} \lambda_{\alpha} \ket{\varphi_{\alpha}^A} \ket{\varphi_{\alpha}^B}, \qquad \sum_\alpha |\lambda_\alpha|^2=1,$$2 (Fliss et al., 2 Dec 2025). That observation captures the general state of the term. “Entanglement matrix” is best understood as a context-dependent label for matrix-valued encodings of entanglement rather than as a single invariant object shared across quantum information, tensor networks, gauge theory, graph states, and emergent-geometry models.