On PPT entanglement distillation

arXiv:2610.12454 · quant-ph · Submitted 2026-10-08 · Read on arXiv

Listen

Transcript

Introduction to the show: ident: Quantum Radio. Generated commentary on the latest quantum physics and condensed matter papers.

Kai: Today's paper: "On PPT entanglement distillation".

Mira: The gist A regularised formula for PPT distillable entanglement in terms of a measured-relativeentropy-like quantity and two different converse constructions are presented,

Kai: First, who's behind it and why it matters.

Paper summary: Kai: So to summarize what the authors are trying to do here, they've presented this regularised formula for E d, PPT(rho) using that measured-relative entropy quantity L(rho), and they also provide two distinct converse constructions. This is a way of bounding the entanglement you can actually distill from these PPT states.

Mira: What’s really important is that these converses are different in nature; one relies on operator quadratic forms, and the other uses Hirschman’s strengthening of the Hadamard three-line theorem. These tools allow them to upgrade the best computable upper bounds they had on E d, LOCC(rho) to bounds involving R infinity(rho), Esq(rho), EQ(rho), and EA(rho).

Lev: From an error correction standpoint, this means that if you were trying to run an entanglement distillation protocol on real hardware, these converses give us a tighter way to understand the limits of what's physically achievable with LOCC. It connects the theoretical bounds back to operational reality.

Kai: The paper applies this immediately by showing that for three times three Werner states, specifically at the antisymmetric weight of twenty-five/twenty-six the certified upper bound of zero point six two one zero seven ebits is strictly below the regularised Rains bound, which is about zero point six four seven six six ebits <ref:2610.12454#pg1,certified upper bound of $0.62107$ ebits>.

Mira: That strict separation means that for these specific states, the regularised Rains bound isn't a tight limit for PPT distillation; it's too high. The authors found this gap specifically for three times three Werner states with that weight, and they show this applies to the bounds from both quadratic forms and analytic families <ref:2610.12454#pg1>.

Lev: That separation is significant because it means the theoretical upper limits derived from these specific tools are actually lower than what we previously thought was achievable, which helps constrain where we can look for new protocols.

Kai: So, in short, the paper gives a systematic way to calculate a lower bound on PPT distillable entanglement and then provides two different ways to establish upper bounds that are stronger than what was previously known. This is about getting a clearer picture of the limits of entanglement manipulation under these constraints.

Conclusion: Mira: Thinking back on "On PPT entanglement distillation," the authors are essentially giving us a regularised formula for E d, PPT(rho) and two specific converse constructions based on quadratic forms and analytic functions. This is a way to quantify what you can distill from these states.

Kai: What this means for the broader context is that they’ve resolved an open problem in the theory of entanglement manipulation by showing that the PPT distillable entanglement can be strictly smaller than the regularised Rains bound for certain states.

Lev: For anyone listening who only cares about whether this changes anything practically, it tells us that while we have these strong theoretical upper bounds, they might not always be tight for every state we consider. It means there are specific quantum states where the distillation rate is lower than the best general limit suggests.

Mira: And it opens up questions about how to calculate E d, PPT(rho) by means of a single-letter formula, or at least whether it’s computable for general states, because while they give us converging lower bounds through their regularised characterisation, they don't guarantee how far those bounds are from the actual limit.

Kai: That gap in certainty is what makes this work hard; we have these mathematical tools that give us lower bounds, but we haven't fully mapped out the distance to the true distillable entanglement. It’s a step toward understanding those ultimate operational constraints in the asymptotic regime where you have infinite copies of a state.

Lev: So, it’s less about finding a single answer right now and more about establishing these robust theoretical frameworks—the regularised formula and those two strong converses—to understand the landscape of what's possible with PPT channels.

Mira: That's the essence of the paper; providing a rigorous framework for bounding entanglement in this specific class of channels. We need to see how these tools are used by experimentalists to test these bounds on actual quantum systems.

Kai: Exactly, and that’s what we’ll be looking at next—how this stuff translates into what you can actually cool and measure in a lab setup.

Scuola Normale Superiore

quant-ph

Submitted: 2026-10-08

Updated: 2026-10-08

Comments: 31+7 pages, 1 figure

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 82/100

The gist: The gist A regularised formula for PPT distillable entanglement in terms of a measured-relativeentropy-like quantity and two different converse constructions are presented, which show that the PPT

Key concepts

PPT Distillable Entanglement ($E_{d, PPT}$): This is the maximum rate at which copies of a noisy quantum state can be distilled into 'ebits' using only local operations and classical communication (LOCC). The paper focuses on the PPT version, which relates to entanglement that can be detected by positive partial transpose. Calculating this rate for simple states is mathematically very difficult.
Regularised Formula ($L_ ext{inf}( ho)$): This is a regularized formula for $E_{d, PPT}$ defined as the limit of a sequence involving the supremum over finite measurements and their relative entropies. It provides a computable lower bound for the true distillable entanglement, allowing researchers to systematically find increasing rational bounds.
Converse Constructions: These are upper bounds on $E_{d, PPT}$ derived using two distinct mathematical frameworks: one based on operator quadratic forms and another based on the theory of analytic functions. These converses help establish rigorous limits for the distillable entanglement.
Regularised Rains Bound ($R_ ext{inf}$): This is a known upper bound for PPT distillable entanglement. The paper demonstrates that its regularized version can be strictly exceeded by certain upper bounds derived from quadratic forms and analytic methods, proving that the true rate can be lower than this bound.

Terminology

Entanglement — 71× in this paper · explained in 4 episodes
Distillation — 28× in this paper · explained in 2 episodes
Trace — 15× in this paper · explained in 2 episodes
Entanglement Distillation — 10× in this paper · explained in 2 episodes
Additivity — 5× in this paper · explained in 1 episode
Superadditivity — 3× in this paper · explained in 2 episodes
Benchmark — 2× in this paper · explained in 2 episodes
Transformer — 2× in this paper · explained in 2 episodes
Verification — 2× in this paper · explained in 2 episodes
Alignment — 1× in this paper
Analyzer — 1× in this paper
Fidelity — 1× in this paper · explained in 2 episodes
Generalization — 1× in this paper · explained in 13 episodes
Hilbert Space — 1× in this paper · explained in 2 episodes
Homogeneity — 1× in this paper · explained in 2 episodes

Summary

The gist A regularised formula for PPT distillable entanglement in terms of a measured-relativeentropy-like quantity and two different converse constructions are presented, which show that the PPT distillable entanglement can be strictly smaller than the regularised Rains bound for certain states

Introduction and Background

Understanding asymptotic entanglement manipulation is a truly daring endeavour <ref:2610.12454#pg3> For how central the phenomenon is for foundations [1–5] as well as applications [6–10], we only have a limited knowledge of the ultimate operational constraints on entanglement transformation in the asymptotic regime, where arbitrarily many copies of a given source state are available <ref:2610.12454#pg3> A particularly challenging task, in this sense, is entanglement distillation [11–14], which consists of turning copies of a noisy, possibly mixed quantum state into as many ‘ebits’ as possible with asymptotically vanishing error <ref:2610.12454#pg3> The ultimate rate of ebit production that is achievable from copies of a state ρ = ρAB through local operations and classical communication (LOCC) is called the LOCC distillable entanglement of ρ, and denoted E d, LOCC(ρ) <ref:2610.12454#pg3> Calculating E d, LOCC(ρ) for even very simple mixed states ρ is typically an extremely challenging task <ref:2610.12454#pg3> Roughly speaking, the nature of this difficulty can be traced back to three different obstacles (1) The set of LOCC operations is mathematically complicated to describe and analyse [15] (2) The presence of the approximation error makes the output of the distillation protocol a moving target <ref:2610.12454#pg3> (3) The asymptotic limit of an arbitrarily high number of input copies is analytically intractable, as we have no guarantee on how many copies may be needed to enable distillation [16>

Main Results

The main results presented are: (a) a regularised formula for the PPT distillable entanglement in terms of a measured-relativeentropy-like quantity <ref:2610.12454#pg3>, and (b) two different single-letter converses, one based on operator quadratic forms and the other on Hirschman’s strengthening of the Hadamard three-line theorem <ref:2610.12454#pg3> As an immediate application of (b), we show that the PPT distillable entanglement can be strictly smaller than the regularised Rains bound, thereby resolving an open problem in the theory of entanglement manipulation [Regula et al., NJP 21:103017, 2019>

Regularised Formula for PPT Distillable Entanglement

The first result expresses the distillation rate through a measurement optimisation <ref:2610.12454#pg5> Let M be a finite measurement channel represented by the POVM (E j)j, and define L(ρ) B sup M Õ j: Tr[ρE j]>0 Tr[ρE j]log Tr[ρE j] E Γ j ∞, L∞(rho) B lim n→∞ 1/n L ρ⊗n <ref:2610.12454#pg6> Theorem 1 (Regularised formula for the PPT distillable entanglement). For every finite-dimensional bipartite state, with L defined by (9), E d, PPT(ρ) = L∞(ρ) = sup n∈N+ 1/n L ρ⊗n <ref:2610.12454#pg7> For a state with computable entries, Theorem 1 gives a computable increasing sequence of lower bounds converging to E d, PPT(ρ): finite rational POVMs and positive rational weights can be enumerated, and the corresponding scores can be approximated from below. Thus E d, PPT is lower semicomputable <ref:2610.12454#pg7>

Converse Constructions

The paper introduces two upper bounds on E d, PPT that are valid for any state <ref:2610.12454#pg3> These are quite different in nature: the one in Theorem 2 is based on quadratic forms, while that in Theorem 3 is based on the theory of analytic functions <ref:2610.12454#pg7> Since every LOCC operation is PPT, these converses also upgrade the best available (computable) upper bounds on the LOCC distillable entanglement to E d, LOCC(ρ) ≤ min R∞(ρ), Esq(ρ), EQ(ρ), EA(ρ) <ref:2610.12454#pg7>

Werner States and Strict Separation

The paper investigates Werner states for explicit bounds and certified separation <ref:2610.12454#pg8> Theorem 4 (Strict qutrit Werner separation). For d = 3 and p = 25/26, 0.5836 ≤ E d, PPT(ρ25/26) ≤ 0.6211 Each of E Q, U and E A, defined in (12), (100) and (119), respectively, is strictly below R∞ on this state <ref:2610.12454#pg9>

Conclusion

The main problem left open is to calculate the PPT distillable entanglement by means of a single-letter formula, or at least to decide whether it is computable at all for general states <ref:2610.12454#pg4> Our regularised characterisation provides converging lower bounds, but no guarantee on their distance from the limit <ref:2610.12454#pg4>

How it works

The regularised formula is constructed by defining a functional e l(rho,M) and then taking its regularisation L(ρ) = sup M e l(rho,M), with its regularisation L∞(ρ) = lim n→∞ 1/n L ρ⊗n <ref:2610.12454#pg12>
The conditional quadratic converse (Theorem 2) is based on an infimum over all finite-dimensional extensions and all strictly positive super-operators K, J satisfying a specific inequality <ref:2610.12454#pg14>
The analytic family strong converse (Theorem 3) uses an annulus O and an analytic operator Z(z) to define a functional J Z,O,z∗(R), which yields an exponential strong-converse rate <ref:2610.12454#pg20>

Operational Interpretation

The paper identifies the regularised functions with the operational distillation rate <ref:2610.12454#pg17> The achievability argument applies a fixed measurement repeatedly and selects its typical weighted outcomes; the reverse inequality follows by pulling back the Bell test of a distillation protocol <ref:2610.12454#pg17>

Converse via Quadratic Forms on Operator Spaces

Theorem 2 (Conditional quadratic converse). Let ρ = ρA:B be a bipartite state. With Vα defined by (11), put E Q(ρ) = E Q(ρA:B) B infrhoA A':B', α,K,J Vα[ρA A':B', K] − Vα[ρA':B', J] <ref:2610.12454#pg21> Every rate strictly above E Q(rho) has an exponential strong converse <ref:2610.12454#pg21>

Converse via Hirschman’s Extension of the Hadamard Three-Line Theorem

Theorem 3 (Analytic-family strong converse). Let ρ = ρA:B be a finite-dimensional bipartite state, let O = O = c + a e i i B b < C and let Z(z) be an operator on A B, analytic in a neighbourhood of O, with Tr[Z(z)] = 1. Suppose that z∗ ∈ O, σ = Z(z∗) is a state, and supp ρ ⊆ supp σ. With the density in (18), built from (17), define J Z,O,z∗(R) B ∫ dO min log Z(z) 1 log Z(z)Γ 1−R kO,z∗(z) d z <ref:2610.12454#pg26> If R ≥ 0 and D(ρ∥σ) + J Z,O,z∗(R) < 0, then R is an exponential strong-converse rate for PPT distillation of ρ <ref:2610.12454#pg26>

Gap from the Regularised Rains Bound

Theorem 4 (Strict qutrit Werner separation). For d = 3 and p = 25/26, 0.5836 ≤ E d, PPT(ρ25/26) ≤ 0.6211 Each of E Q, U and E A, defined in (12), (100) and (119), respectively, is strictly below R∞ on this state <ref:2610.12454#pg31>

How it works

(a) A regularised formula for the PPT distillable entanglement:

The formula is given by E d, PPT(ρ) = L∞(ρ) = sup n∈N+ 1/n L ρ⊗n <ref:2610.12454#pg7> This can be used to calculate achievable rates (i.e. lower bounds on E d, PPT) systematically <ref:2610.12454#pg7>

(b) Two converse constructions:

The two converses are based on quadratic forms and analytic families, both of which improve on the regularised Rains bound already for two-qutrit Werner states <ref:2610.

Improvements for AI systems

AI systems can calculate a regularised formula for PPT distillable entanglement in terms of a measured-relative entropy-like quantity (Theorem 1), providing a concrete way to approach the optimal distillation rate from below.
AI can establish an explicit faithful lower bound in terms of the entanglement negativity (Theorem 1), which makes the qualitative statement (4) any NPT state is PPT distillable "quantitative by supplying a strictly positive rate whenever N(rho) > 0."
AI systems can derive two different upper bounds for PPT distillable entanglement, one based on quadratic forms on operator spaces (Theorem 2) and another based on the theory of analytic functions (Theorem 3).
AI can verify that the conjectured equality (5) between the PPT distillable entanglement and the regularised Rains bound is, alas, false, by showing that for a "3 × 3 Werner state with antisymmetric weight 25/26, we have Ed, PPT(rho25/26) ≤ 0.62107 < R∞(rho25/26)."
AI can implement a conditional quadratic converse (Theorem 14), which provides an upper bound on distillable entanglement: ED, PPT(rhoA:B) ≤ ED†, PPT(rhoA:B) ≤ EQ(rhoA:B).
AI can compute the exact one-copy POVM bound for Werner states (Proposition 19), which provides a specific achievable rate based on an optimum measurement channel.
AI systems can generate a certified 24-copy qutrit effect with an achievable curve expansion, which strictly improves the quadratic coefficient of the one-copy bound for near-threshold parameters.
AI can apply Hirschman’s extension of the Hadamard three-line theorem to derive a different quadratic converse (Theorem 3), providing another upper bound on distillable entanglement based on analytic families.
AI can compute the combined general converse (Corollary 18), which provides an overall upper bound by taking the minimum of R∞(rho), EQ(rho), and EA(rho).