On supporting affine functionals for Entanglement of Formation

AI summary

Authors: A. S. Holevo, M. E. Shirokov (Steklov Mathematical Institute, Moscow, Russia) arXiv: 2608.27363v1 [quant-ph], 27 Aug 2026

Overview

Research area: Quantum information theory, specifically the mathematical theory of entanglement measures — here the Entanglement of Formation (EoF) in bipartite quantum systems — combined with convex analysis and functional analysis on state spaces.

Technical level: Advanced. The paper relies on Fenchel–Moreau duality, convex closures, subdifferentials, weak convergence of Borel measures on pure states, and Wootters' formula for two-qubit entanglement.

Scope (one sentence): The paper settles, negatively, whether the EoF always admits a supporting affine functional at every state of a finite-dimensional bipartite system, by giving an explicit two-qubit counterexample and by characterizing when such functionals do exist, in both finite and infinite dimensions.

What This Paper Is About

The Entanglement of Formation is a convex function on the set of quantum states, and general convex analysis says it can be written as the pointwise supremum of all affine functionals that lie below it. Several articles had assumed that, because the EoF has a convex roof structure and the subsystems are finite-dimensional, this supremum is actually attained at every state — i.e., that a "supporting" affine functional exists even at degenerate states sitting on the boundary of the state set. The paper shows this assumption is false, by constructing an explicit two-qubit state where no such functional exists, and then develops general criteria that say exactly when local and global supporting functionals exist.

Key Contributions

1. A criterion (Theorem 1) linking supporting functionals to Lipschitz lower semicontinuity. For a state ρ and any subspace ℋ₀ containing its support, the existence of a supporting functional with spectral diameter D(Λ_ρ) ≤ 2C is shown equivalent to the bound E_F(ρ) − E_F(σ) ≤ C‖ρ − σ‖₁ holding for all states σ on ℋ₀, and equivalent to a third variational characterization involving an operator Λ̃_ρ with ‖Λ̃_ρ‖ ≤ C.

2. An explicit two-qubit counterexample (Proposition 3). The state ρ = ½|Φ⁺⟩⟨Φ⁺| + ½|01⟩⟨01|, of rank 2 and spectrum {1/2, 1/2, 0, 0}, has no supporting affine functional for the EoF — even in the simplest case where A and B are qubits.

3. Extension to all dimensions (Corollary 6). Embedding the two-qubit example in arbitrary two-dimensional subspaces shows that bipartite systems of any dimension contain degenerate states with no supporting affine functional.

4. Explicit Lipschitz lower semicontinuity bounds (Proposition 7, Lemma 8) for finite-rank states ρ, in terms of the minimal eigenvalue λ_min^ρ, with and without restrictions on the support of σ.

Main Findings

• The assumed guarantee fails at degenerate states. General convex analysis only guarantees a supporting functional at internal (non-degenerate) points of the state set; the EoF's special convex roof structure does not rescue degenerate states. The paper's stated motivation is to show that "all these hopes are in vain."

• The counterexample state. ρ = ½|Φ⁺⟩⟨Φ⁺| + ½|01⟩⟨01| with |Φ⁺⟩ = (1/√2)(|00⟩ + |11⟩) has C(ρ) = 1/2 and E_F(ρ) = ℰ(1/2) ≈ 0.246. Its spectrum is {1/2, 1/2, 0, 0}, so rank ρ = 2.

• Divergent one-sided derivative. Taking σ = |10⟩⟨10| and ρ_t = (1−t)ρ + tσ, the authors prove lim_{t→0⁺} [E_F(ρ) − E_F(ρ_t)]/t = +∞, which rules out any finite Lipschitz constant C_ρ and hence any supporting functional.

• The explicit decay of the concurrence. For 0 ≤ t ≤ t₀ = 1/3, C(ρ_t) = (1−t)/2 − 2√(t(1−t)/2) = 1/2 − √(2t) + O(t). The square-root singularity, not linear behaviour, is the source of the failure.

• The slope constant. ℰ is smooth and strictly increasing on (0,1), and κ := ℰ′(1/2) = ln(2+√3)/√3 ≈ 0.760 > 0. Consequently E_F(ρ) − E_F(ρ_t) = κ√2 √t + O(t), with κ√2 ≈ 1.07.

• Local versus global functionals. Theorem 1 with ℋ₀ = ℋ_AB characterizes global supporting functionals; with ℋ₀ = supp ρ it characterizes local ones. Local supporting functionals always exist at states of finite-dimensional systems, but the paper states that for some two-qubit states the local functional cannot be extended to a global one (Proposition 3 in the paper's numbering of that claim in Section 3).

• Almost-everywhere equality (Remark 2). If a supporting functional exists, then ⟨ψ|Λ_ρ|ψ⟩ = S(ψ_A) for μ_ρ-almost all ψ, for any measure μ_ρ realizing the EoF at ρ.

• Bounds in Proposition 7. With rank ρ_A = d < ∞: E_F(ρ) − E_F(σ) ≤ [(log d − E_F(ρ))/λ_min^ρ] ‖ρ − σ‖₁/2 for all σ with supp σ ⊆ supp ρ. With S(ρ_A) < ∞: E_F(ρ) − E_F(σ) ≤ [S(ρ_A)/(λ_min^ρ)² − E_F(ρ)/λ_min^ρ] ‖ρ − σ‖₁/2 under the same support condition.

• Lemma 8. If S(ψ_A) ≤ C < ∞ for all ψ in supp ρ, then the supporting operator exists on ℋ_ρ with D(Λ_ρ) ≤ (C − E_F(ρ))/λ_min^ρ.

• Infinite dimensions. The paper adopts definition (3) — an infimum over Borel probability measures on pure states with prescribed barycenter — noting that its agreement with the discrete-ensemble definition (1) in infinite dimensions is, as far as the authors know, an open question, while (1) and (3) coincide whenever min{S(ρ_A), S(ρ_B)} < +∞.

• Disclosed use of an AI assistant. The authors state that the search for the counterexample via Wootters' formula was carried out "very quickly with Claude Fable 5," and credit "the help of Claude Fable 5" in the abstract and in the statement of Proposition 3.

Methodology in Plain English

The strategy rests on a dictionary between two seemingly different questions. Asking "is there an affine functional touching the EoF exactly at ρ from below?" is shown to be the same as asking "does the EoF drop at most linearly, with some finite constant, as you move away from ρ?" That equivalence is Theorem 1, proved by writing the EoF as an infimum over ensembles (measures on pure states), swapping infimum and supremum using a minimax argument, and then reading off the optimal operator.

With that dictionary in hand, disproving the existence of a supporting functional reduces to exhibiting a single direction of approach to ρ along which the EoF falls off faster than linearly. The authors restrict to two qubits, where the EoF is given exactly by Wootters' formula in terms of the concurrence C(ρ), and pick a rank-2 state mixing a maximally entangled Bell state with a product state. Mixing in a small amount t of the state |10⟩⟨10| makes the concurrence drop like 1/2 − √(2t) rather than 1/2 − const·t. Because the binary-entropy expression ℰ has a nonzero finite derivative κ at C = 1/2, the EoF itself drops like a constant times √t, whose derivative at t = 0 is infinite. That single computation kills every candidate Lipschitz constant.

The paper then reverses direction: since the criterion is an equivalence, verifying a specific operator guess (via Wootters' formula and eigenvalue bounds) certifies the existence of a supporting functional, which yields the explicit bounds of Proposition 7 and Lemma 8.

Why This Matters

Impact on research. The paper closes off a natural but incorrect route: one cannot assume that every degenerate state of a finite-dimensional bipartite system admits a global supporting affine functional for the EoF. This bears directly on any argument that represents the EoF as a supremum over entanglement-witness-like operators and assumes attainment — for instance, in duality-based and convex-optimization approaches to computing or bounding entanglement measures. It also separates the local and global pictures: local functionals exist but may not extend.

Real-world applications (areas where EoF and its variational representations are used):

• Quantifying entanglement in quantum-computing hardware and in the resource theory of entanglement.
• Security analysis and rate calculations in quantum key distribution and other entanglement-based communication protocols.
• Certification and benchmarking of entangled state preparation, where entanglement measures must be estimated reliably from finite data.
• Numerical convex-roof optimization and semidefinite-style relaxations used to bound entanglement measures in many-body and high-dimensional settings.

Industry relevance. Any error bar or stability estimate quoted for an entanglement measure rests on continuity and Lipschitz-type bounds. The paper's √t-type behaviour and its explicit bounds in Proposition 7 tell practitioners that near certain degenerate states, estimates of the EoF are inherently more sensitive than a linear stability analysis would suggest.

Future Directions

• Characterize the full set of degenerate states lacking supporting functionals. The paper gives one explicit family and shows such states exist in every dimension; a structural description of which degenerate states fail (and which, despite degeneracy, still admit a functional) is left open.

• Tighten or extend the Lipschitz bounds. Proposition 7 gives bounds involving λ_min^ρ and log d or S(ρ_A), valid under the support restriction supp σ ⊆ supp ρ; removal or sharpening of these restrictions, and the precise optimal constants, remain to be determined.

• Settle the infinite-dimensional definitional question. The paper explicitly notes that the applicability of the discrete-ensemble formula (1) in infinite dimensions is, as far as the authors know, an open question, and that its equivalence with (3) holds when at least one marginal entropy is finite.

• Re-examine results that relied on the attainment assumption. Since the paper states that several articles assumed the global supporting functional always exists, a natural follow-up is to check which prior conclusions survive and which require non-degenerate-state hypotheses.

Target Audience

Researchers in quantum information theory and mathematical physics working on entanglement measures, convex roofs, and duality-based characterizations of quantum resource quantifiers; mathematicians working on convex analysis on infinite-dimensional state spaces; and graduate students or specialists who need to know precisely when variational representations of the EoF are attained. The paper assumes comfort with trace-class operators, von Neumann entropy, Fenchel–Moreau duality, and Wootters' formula, so a beginner-friendly read is not expected.

Authors’ abstract

In several articles, the authors assume that the convex roof structure of the EoF and finite-dimensionality of subsystems $A$ and $B$ guarantee the existence of the (global) supporting affine functional for the EoF at any state of the system $AB$. This means that for any state $ρ$ of $AB$ there is a Hermitian operator $Λ_ρ$ on $\mathcal{H}_{AB}=\mathcal{H}_A\otimes\mathcal{H}_B$ such that $E_F(ρ)=\mathrm{Tr}Λ_ρρ$ and $E_F(σ)\geq\mathrm{Tr}Λ_ρσ$ for any state $σ$ of $AB$. We present an explicit example showing that, when $ρ$ is degenerate, this is not true even in the simplest case when $A$ and $B$ are qubit systems. The construction is based on the fact that the existence of a supporting affine functional for the EoF at a state $ρ$ is equivalent to the Lipschitz lower semicontinuity of the EoF at this state $ρ$. We use Wootters' formula and the help of Claude Fable 5 to find a state $ρ$ of the system $AB$ for which the latter property does not hold. We also describe conditions for the existence the local and global supporting affine functionals for the EoF at a given state of both finite and infinite-dimensional bipartite quantum systems. These conditions allow us to find Lipschitz lower semicontinuity bounds for the EoF at a given finite rank state $ρ$ (i.e. inequalities of the form $\,E_F(ρ)-E_F(σ)\leq C_ρ\|ρ-σ\|_1$) with and without restrictions on the support of the state $σ$.