The geometry of absolute separability and other convex matrix properties from spectrum

AI summary

Overview

Research area: Quantum information theory, specifically the geometry of entanglement criteria, combined with convex geometry and real algebraic geometry.

Technical level: Advanced. The paper assumes familiarity with convex cones, faces and exposed points, spectrahedra, semialgebraic sets, and the separability/PPT formalism for bipartite quantum states.

Scope (one sentence): The paper studies the sets of eigenvalues of bipartite absolutely separable states (ASEP${m,n}$) and absolutely PPT states (APPT${m,n}$), establishing when these sets are spectrahedra or only semialgebraic, characterising their faces and extreme points, and bounding their maximal purity, minimal von Neumann entropy and relative volume.

What This Paper Is About

Deciding whether a bipartite quantum state is entangled is NP-hard in general, so researchers study "absolute" versions of separability criteria that depend only on a state's eigenvalues rather than on its eigenbasis. The absolute PPT condition has a known complete characterisation, but it is written as a finite system of linear matrix inequalities whose number grows exponentially with dimension, which obscures the symmetry of the problem. The paper's goal is to describe the geometry of the resulting eigenvalue sets directly, using permutation symmetry, convex geometry and real algebraic geometry, and to extract quantitative statements about purity, entropy and volume.

Key Contributions

1. A general theory of absolute convex sets of matrices. The authors define, for a closed convex cone $M$ of Hermitian matrices, the absolute set $AM = \bigcap_{U \in \mathcal{U}(d)} UMU^\dagger$, and prove that its diagonal part equals its spectrum, that its dual diagonal part is the dual of the diagonal part, and that diagonal extreme and exposed points of $AM$ are exactly the extreme and exposed points of the diagonal set (Theorems 3.1–3.5). They also show that if $M$ is a closed convex semialgebraic cone then $AM$, its diagonal part and its spectrum are semialgebraic (Theorem 3.6).

2. A permutation-symmetric reformulation of the absolute PPT criterion, and a spectrahedrality result. The absolute PPT criterion is rewritten using permutation orbits of a single base matrix, and this is used to show that APPT${m,n}$ is a spectrahedron for all $m \leq n$, so that all its faces are exposed. ASEP${2,n}$ is shown to be a spectrahedron as well, while for ASEP$_{m,n}$ with $m,n \geq 3$ the authors can in general only establish that it is a semialgebraic set.

3. A complete kernel-based description of the facial structure of APPT$_{m,n}$. Faces and extreme points are characterised by a constraint matrix with linear entries built from kernel subspaces, the dimension of a face is shown to be determined by the rank of that matrix, and the maximal dimension of a proper face is proved to be $mn - m - 1$.

4. Quantitative bounds via an inscribed polytope. An inscribed permutation-invariant polytope $\mathcal{P}{m,n} \subseteq \mathrm{APPT}{m,n}$ yields a rigorous lower bound on the maximal attainable purity and a rigorous upper bound on the minimal von Neumann entropy of APPT${m,n}$, together with an exponential decay statement for the relative spectral volume of APPT${m,n}$ in $n$.

Main Findings

• APPT$_{m,n}$ is a spectrahedron: for all $m \leq n$, the set of spectra of absolute PPT states admits a spectrahedral description, which implies that every face is exposed.
• ASEP is spectrahedral only in special cases so far: ASEP${2,n}$ is a spectrahedron, but for $m,n \geq 3$ the authors state that in general they can only show ASEP${m,n}$ is a semialgebraic set.
• Boundary is determinantal: because of the spectrahedral description, the topological boundary of APPT$_{m,n}$ is expressed as the feasible part of $(mn-2)$-dimensional determinantal hypersurfaces, and the paper bounds the number of irreducible determinant polynomials needed to describe that boundary.
• Face dimensions are rank-controlled: the dimension of a face of APPT$_{m,n}$ is fixed by the rank of a certain constraint matrix with linear entries, and the largest proper faces have dimension $mn - m - 1$.
• Purity lower bound and a conjecture: the inscribed polytope $\mathcal{P}{m,n}$ gives a rigorous lower bound on the maximal purity of APPT${m,n}$; the authors conjecture that the maximal purity (and its spectrum) coincides with that of the polytope in arbitrary dimensions except when $m = n = 2$, and they verify this numerically for $m = 2, 3, 4$.
• Entropy upper bound: a rigorous upper bound on the minimal von Neumann entropy of APPT${m,n}$ is derived, and numerical evidence indicates that the minimum entropy eventually coincides with that of $\mathcal{P}{m,n}$ as the local dimension $n$ increases.
• Volume decay: the relative spectral volume of APPT${m,n}$ decays exponentially in $n$, by a constant multiplicative factor of the relative volume of the inscribed polytope $\mathcal{P}{m,n}$; this is contrasted with the separable ball, which the paper states decays much faster.
• General structural identities: for absolute cones, the diagonal of the set of extreme rays equals the extreme rays of the diagonal set, and the analogous statement holds for exposed extreme rays, with both sets recovered as unitary orbits of their diagonal representatives.

Methodology in Plain English

The authors work with convex cones of Hermitian matrices rather than with normalised states, because the unnormalised versions are closed under addition and scaling. For a given cone such as the PPT cone or the separable cone, they take the intersection over all global unitary rotations, producing the "absolute" cone, and then observe that membership in this intersection depends only on the eigenvalue vector. This reduces the problem to studying a subset of the space of diagonal matrices, that is, a subset of possible spectra.

The key technical move is to replace the ordering-dependent formulation of the absolute PPT criterion (eigenvalues sorted in decreasing order, with a separate matrix inequality for each constraint) by a formulation built from permutations of one base matrix. This makes the permutation symmetry of the problem explicit and turns the question of whether the eigenvalue set is a spectrahedron into a question about a single affine matrix-valued map. To describe faces and extreme points, the authors construct constraint matrices from kernel subspaces of the boundary matrices, so that face dimensions become ranks of explicitly given matrices. For the quantitative part, they inscribe a permutation-invariant polytope inside APPT$_{m,n}$, compute its purity and entropy exactly, and use it as a comparison object for rigorous bounds and for numerical experiments at $m = 2, 3, 4$. Semialgebraicity statements are obtained by writing absolute-set membership as a first-order formula over the reals and applying quantifier elimination via the Tarski–Seidenberg principle.

Why This Matters

The absolute separability problem is open, and the conjecture that absolutely separable states coincide with absolute PPT states is supported by a proof for qubit-qudit systems and by numerical evidence in general. A precise geometric description of the two eigenvalue sets gives a structural language for attacking that conjecture: spectrahedrality implies exposed faces and tractable optimisation, whereas a merely semialgebraic set carries no such guarantee. The results also let researchers replace an exponentially large family of matrix inequalities with one permutation-symmetric description.

The paper does not report specific applications. The areas where the underlying problem is directly relevant include:

• Entanglement detection: certifying that a state is not entangled is the core primitive behind entanglement-based protocols.
• Quantum cryptography: the paper cites quantum cryptography as an application of entanglement, and security arguments rely on knowing which states are separable.
• Quantum teleportation: also cited as an entanglement application, with state preparation constrained by what can be certified as separable.
• Benchmarking and simulation of quantum devices: verifying that a prepared bipartite state has no distillable entanglement is a routine validation task, and spectrum-only criteria are cheaper to check than full separability.

Industry relevance: the results bear on semidefinite-programming practice, since spectrahedra are exactly the feasible regions of SDPs, and on any workflow that screens large families of states by eigenvalue alone. Quantum hardware and quantum communication groups that need scalable, basis-independent separability tests are the closest industrial audience; the paper itself makes no industrial claims.

Future Directions

• Settle the ASEP = APPT question in general dimensions. The equivalence is known for qubit-qudit systems and conjectured otherwise; the geometric descriptions here are proposed as a route toward a proof.
• Prove or refute the purity conjecture. The conjecture that maximal purity of APPT$_{m,n}$ matches the inscribed polytope for all dimensions except $m = n = 2$ is supported only by numerical checks at $m = 2, 3, 4$.
• Determine the exact spectral status of ASEP$_{m,n}$ for $m,n \geq 3$. It is currently only known to be semialgebraic; whether it is a spectrahedron in these cases is left open.
• Turn the numerical entropy observation into a theorem. The convergence of the minimal von Neumann entropy to the polytope value as $n$ grows is demonstrated numerically, but the rate and the exact threshold are not established.

Target Audience

Quantum information theorists working on entanglement theory and separability criteria; researchers in convex geometry and real algebraic geometry interested in spectrahedra and semialgebraic sets arising from unitary orbits; practitioners of semidefinite programming who want concrete examples of spectrahedral versus merely semialgebraic feasible sets; and graduate students with a background in quantum states and convex analysis who want to see the absolute separability problem treated as a geometric object.

Authors’ abstract

We investigate the geometric structure of the set of spectra of bipartite absolute separable states ($\mathrm{ASEP}_{m,n}$) and absolute positive partial transpose states ($\mathrm{APPT}_{m,n}$), \emph{i.e.}, bipartite quantum states that remain separable or PPT, respectively, under all global unitary transformations. First, we establish general geometric properties of absolute convex sets of matrices, their spectra and extreme points. Regarding absolute separability, we present a permutation-symmetric reformulation of the absolute PPT criterion and use it to demonstrate that $\mathrm{APPT}_{m,n}$ is a spectrahedron for all $m\leq n$: in particular, all its faces are exposed. While $\mathrm{ASEP}_{2,n}$ is a spectrahedron, too, for $\mathrm{ASEP}_{m,n}$ we can in general only show that it is a semialgebraic set for all $m,n\geq 3$. Furthermore, we provide a complete characterization of the faces and extreme points of $\mathrm{APPT}_{m,n}$ and demonstrate that the dimension of a face is determined by the rank of a certain matrix, with maximal proper faces having dimension $mn-m-1$. In the quantitative setting, we provide a rigorous lower bound on the maximal attainable purity of $\mathrm{APPT}_{m,n}$ via an inscribed polytope $\mathcal{P}_{m,n}$ and conjecture that the maximal purity of $\mathrm{APPT}_{m,n}$ (along with its spectra) coincides with the polytope for arbitrary dimensions except when $m=n=2$. Additionally, we also provide a rigorous upper bound on the minimal von Neumann entropy of $\mathrm{APPT}_{m,n}$ and demonstrate numerically that the minimum entropy eventually coincides with the polytope $\mathcal{P}_{m,n}$ as the local system dimension $n$ increases. Finally, we show that the relative spectral volume of $\mathrm{APPT}_{m,n}$ decays exponentially in $n$ by a constant multiplicative factor of the relative volume of the inscribed polytope $\mathcal{P}_{m,n}$.