A mathematical preprint proposes a way to read entanglement through the geometry of bipartite quantum-state space. Its central result is a set of analytic bounds on the weight of the entangled component in a Best Separable Approximation, or BSA, together with an ordering result saying that this component has no smaller entangled-space size than the original mixed state.
ArXiv:2608.20050v1 is dated 20 Aug 2026. The manuscript studies an abstract bipartite system, written HA ⊗ HB, with subsystem dimensions dA and dB, then treats two-qubit states separately.
The paper’s method is analytic: it works with quantum-state sets, their boundaries and decompositions. Its main results are inequalities derived from that geometry, rather than estimates drawn from a sample.
A reference point inside the state space
The construction starts from the maximally mixed state, ρ0, and uses it as a reference point. From there, the analysis follows radial directions until they meet two boundaries: the boundary of the separable-state set and the boundary of the quantum-state set. That choice of reference turns a question about entanglement into a question about positions along rays.
In plain language, convex geometry here is the study of the shape made by allowed quantum states and their mixtures. The paper asks where a target state sits along a line from ρ0, where the separable region ends and where the broader quantum-state region ends.
Those locations are summarized by two quantities. The paper defines G(ρ) = [1 − L(ρ)]/L(ρ) as a geometric entangled-space quantity and Q(ρ) = [(1 − p)/p]/G(ρ) as a relative entanglement degree. The notation is specific to the paper’s radial construction.
The BSA puts the pieces on the edge
The BSA is the paper’s way of separating an entangled mixed state into a separable part and an entangled remainder. In its notation, ρ = (1 − p0)ρA + p0ρB, with ρA separable and p0 chosen to maximize the separable weight. The coefficient p0 therefore tracks the weight assigned to the remaining entangled component.
Optimality gives the decomposition a precise geometric shape. The separable component lies on the boundary of the separable-state set, while the entangled component lies on the boundary of the quantum-state set. The two pieces are therefore tied to the edges of the sets used in the construction.
That boundary picture is the bridge between the BSA and the geometric quantities. Once the target state, the separable component and the entangled component are placed within the same radial construction, the paper can compare their entangled-space sizes and constrain p0.
What the general inequalities say
For a general bipartite state, the reported bound is (1−p)LB/[p(1−LB)] ≤ p0 ≤ Q(ρ). In the paper’s notation, the left side is a geometric lower bound built from p and LB, while Q(ρ) supplies the upper bound.
The inequality does not assign one universal entangled fraction to every state. It constrains the p0 belonging to the state and its optimal BSA under the construction used in the analysis. Since the result is analytic, the supplied analysis reports no statistical uncertainty interval for it.
A second comparison is written as [1 − LR]/LR ≤ [1 − LB]/LB. The paper interprets this ordering to mean that the BSA entangled component has an entangled-space size no smaller than the original mixed state. That is a statement about the geometry assigned by the construction, not a claim that a physical process has made the state more entangled.
Two qubits sharpen the picture
The paper then specializes to two-qubit states, where the PPT criterion is necessary and sufficient for separability. PPT is a test applied to the state’s density matrix; in this two-qubit setting, it supplies the separability check used to obtain the geometric quantities.
That special case produces a structural result, not only a weight bound. In an optimal two-qubit BSA, the remaining entangled component is pure: ρB = |ψB⟩⟨ψB|. The entangled side of the decomposition therefore has a particularly simple form within this case.
For a pure two-qubit entangled component, the geometric expression is (1 − LB)/LB = 2|sin 2θ|. The reciprocal ratio obeys LB/(1 − LB) ≥ 1/2, with its minimum at θ = π/4, the value associated in the paper with a maximally entangled state.
Read together, those formulas say that the geometric entangled-space size changes with the parameter θ and is largest for the maximally entangled state. The result gives the abstract geometry a recognizable pattern: the state identified as maximally entangled occupies the largest such radial entangled-space measure.
The two-qubit weight bound also takes a simpler form: (1−p)/(2p) ≤ p0 ≤ Q(ρ). The lower limit now uses the factor 1/2, while the upper limit remains the relative quantity Q(ρ) defined earlier.
The manuscript says both sides of this two-qubit bound can be saturated for suitable states. Within the stated mathematical setting, that means appropriate cases reach the lower equality and other appropriate cases reach the upper equality.
The hard part beyond two qubits
The two-qubit formulas come with a built-in boundary. In higher dimensions, the geometric relations remain meaningful, but locating the separable boundary becomes harder because PPT is no longer sufficient for separability. The obstacle is therefore not the radial language itself; it is determining the relevant boundary.
That distinction limits how far the explicit formulas can be carried. The PPT-based derivation is tied to the two-qubit case, while the general construction leaves the separable boundary to be determined in whatever higher-dimensional system is under study.
The paper’s stated output is a set of geometric quantities and analytic BSA bounds for an abstract state space, not a sample-level estimate. Its conclusions therefore describe what follows from the convex-geometric construction and its two-qubit specialization.
The remaining technical question is how to compute the relevant separable boundaries and geometric quantities explicitly in higher-dimensional systems where PPT is insufficient.
The document is identified as arXiv:2608.20050v1 and dated 20 Aug 2026. It is therefore presented here as a preprint, and its mathematical claims should be read in that publication status.
The supplied document does not report financial funding. Its acknowledgment instead credits OpenAI’s GPT models with writing and preparation assistance.
Paper data and sources
Original title: Radial Convex Geometry of Quantum States and Its Relation to Best Separable Approximation
Authors: Haonan Qiang
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-20
DOI: Not available
Original paper · Full text