Rényi Entanglement of Purification Is Non-additive

• Additivity determines whether the asymptotic quantity has a single-letter formula instead of requiring optimization over exponentially growing tensor powers.Source: §I · p. 1
• The paper targets the unresolved additivity of entanglement of purification for classical states using simple, low-dimensional deterministic constructions.Source: §I · p. 1
• Entanglement of purification measures total classical and quantum correlations, and its regularized form governs asymptotic correlation preparation and visible compression.Source: §I · p. 1
• The canonical purification is solved exactly for every Rényi order for classical two-qubit states.Source: §I · p. 1
• For every 0 ≤ α < 1, two copies of a single classical two-qubit state witness strict nonadditivity, while α ≥ 2 admits additivity through a 2 × 2 norm-compression argument.Source: §I · pp. 1–2
• The interval 1 ≤ α < 2 remains open, with available indications motivating a conjecture of additivity throughout that interval.Source: §I · pp. 1–2

• The exact one-copy optimization is controlled by a single scalar parameter playing the role of concurrence in the two-qubit formula.Source: §II · p. 2
• The canonical purification simultaneously minimizes Rényi entropy for every α ∈ [0, ∞], including the von Neumann and endpoint orders.Source: §II · p. 2
• The canonical purification has a rank-at-most-two reduced spectrum whose eigenvalues are determined by the classical concurrence parameter.Source: §II · p. 2
• The canonical spectrum majorizes the Schmidt spectrum of every competing purification.Source: §II · p. 2
• Schur concavity, together with the rank and largest-eigenvalue endpoints, shows that no competing purification has lower Rényi entropy.Source: §II · p. 2
• The same strategy extends to binary classical–quantum states, where an Uhlmann purification is globally optimal for all Rényi orders.Source: §II · pp. 2–3

Signed-canonical purifications

• Every sign table s_ij = ±1 yields a purification of the same classical state, while nonfactorizing signs can change the Schmidt spectrum and Rényi entropies.Local-factorizable signs are removable by local diagonal unitaries; general patterns are not.Source: §III · p. 3

Numerical evidence

• At two copies, 512 inequivalent sign patterns were exhaustively searched across 5 × 10^4 uniformly sampled classical two-qubit states.The witness beats the product value for a fraction of sampled states, with violations becoming rare near α = 1.Source: §III · p. 3, Figure 1 · p. 3

Explicit construction

• Theorem 2 proves that for every 0 ≤ α < 1, some classical two-qubit state violates additivity using two copies.This replaces the numerical evidence with an explicit construction.Source: §III · p. 3
• For the one-parameter family, changing only the all-zero coefficient in the two-copy canonical purification produces the signed witness.For 0 < α < 1, sufficiently small r gives the strict entropy decrease; α = 0 is handled separately.Source: §III · pp. 3–4

Strict violation

• The signed two-copy purification has strictly lower Rényi entropy than the product purification, proving E_P^(α)(ρ_r^⊗2) < 2E_P^(α)(ρ_r).At α = 0, the signed coefficient matrix has rank three, while the one-copy theorem supplies the product value.Source: §III · p. 4

Theorem

• For every classical two-qubit state and finite-dimensional classical auxiliary state, Rényi entanglement of purification is additive for α ≥ 2.The result includes tensoring with an arbitrary finite classical state.Source: §IV · p. 4

Norm compression

• The proof converts purification optimization into maximizing a Schatten p norm with p = 2α, then compresses a 2 × 2 block operator to scalar norms.Product purifications saturate the resulting bound, making the optimized norm multiplicative.Source: §IV · p. 4

Rényi-order phase diagram

• The phase diagram distinguishes red orders with classical nonadditive examples, blue orders with additivity for every classical two-qubit state, and orange unsettled orders.The diagram summarizes the order-dependent status established by the paper.Source: Figure 2 · p. 4

Threshold

• The proof threshold α = 2 comes from applying a positive 2 × 2 block inequality at exponent p/2, whose required direction begins at p/2 = 2.The argument is established for every real p ≥ 4 and also treats α = ∞.Source: §IV · p. 4

Open interval

• The interval 1 ≤ α < 2 remains open; Audenaert’s conjectured norm inequality would extend the proof to α > 1 and, by continuity, to α = 1.The authors therefore conjecture additivity throughout α ≥ 1.Source: §IV · p. 4

Central conjecture

• The authors conjecture that classical two-qubit Rényi entanglement of purification is additive, and hence single-letter, for all α ≥ 1.This is motivated by the meeting point between the vanishing subunit-order witness and the conjectured extension of norm compression.Source: §V · p. 4

Open problem

• Proving the von Neumann case α = 1 is identified as the central remaining step.Possible routes are norm compression plus continuity or a direct entropic inequality.Source: §V · p. 4

Next testing ground

• If the conjecture holds, a von Neumann counterexample must lie outside the classical two-qubit family.Binary classical–quantum states are proposed as a next testing ground because they retain an exact one-copy formula.Source: §V · p. 4

Extensions

• The signed Z2 witness suggests a hierarchy toward continuous U(1) phases and restricted unitaries for larger classical or classical–quantum states.These extensions generalize the available phase freedom beyond sign choices.Source: §V · pp. 4–5

Broader connections

• The construction connects EoP with reflected entropy, tensor-network or geometric interpretations, and constrained minimum-output-entropy problems.The paper presents these as broader settings for the small witnesses developed here.Source: §V · p. 5

Purification structure

• An arbitrary purification must attach orthonormal ancillary states to distinct occupied classical outcomes when the reduced state is diagonal.This structural constraint enables the overlap analysis.Source: Appendix A · p. 7

Overlap bound

• The largest eigenvalue is bounded by maximizing the overlap of the purification with arbitrary normalized product kets across AA′ : BB′.The test kets are resolved according to the classical bits A and B.Source: Appendix A · p. 7

Norm reduction

• Triangle inequality and normalization bound each product-overlap contribution by the norms of its classical-bit components.The proof then retains only these four norms to construct normalized logical kets.Source: Appendix A · p. 7

Canonical reduction

• Replacing arbitrary components by normalized logical kets reduces the overlap problem to the scalar classical coefficient table.The preceding bound is combined with the variational characterization of the largest eigenvalue.Source: Appendix A · p. 7

Classical–quantum extension

• The same rank-two outer structure supports an exact optimization for binary classical–quantum states at every Rényi order.The appendix states this as Proposition 1 for 0 ≤ α ≤ ∞.Source: Appendix B · pp. 7–8

• Uhlmann’s theorem provides purifications of the two branch states with maximal overlap on a common ancillary system.Source: §B.1 · p. 8
• Orthogonal local flags combine the branch purifications into a purification of the classical state.Source: §B.1 · p. 8
• The resulting marginal has the same nonzero spectrum as G, establishing the achievable entropy bound.Source: §B.1 · p. 8

• Any purification is resolved into the same two classical branches, reducing the comparison to branch-dependent ancillary states.Source: §B.2 · p. 8
• Because G has rank two, controlling the largest eigenvalue suffices to obtain the full majorization relation.Source: §B.2 · pp. 8–9
• The only remaining freedom is the overlap between the two right-side branch kets, which is bounded by the original fidelity.Source: §B.2 · pp. 8–9
• Schur concavity converts majorization into Sα(ΦAA′) ≥ Sα(G) for every 0 < α < ∞.Source: §B.2 · p. 9
• The exact one-copy optimum underpins the subsequent two-copy nonadditivity proof.Source: Appendix C · p. 9

• The construction flips only the amplitude associated with physical outcome 00 in both copies.Source: §C.1 · p. 10
• All sixteen squared amplitudes remain unchanged, so the sign-flipped ket is an admissible purification of ρr⊗2.Source: §C.1 · p. 10
• Only the |00⟩R branch changes the left marginal under this sign flip.Source: §C.1 · p. 10
• The diagonal contribution cancels exactly, leaving only coherence terms in the marginal difference.Source: §C.1 · p. 10
• Equation (C12) isolates the complete collective effect needed for the two-copy witness.Source: §C.1 · p. 10

C.2. Strict entropy gain for every fixed 0 < α < 1

• For 0 < α < 1, the interpolation Xr(t) between the product marginal and the signed marginal remains full rank for sufficiently small r.Source: §C.2 · pp. 10–11
• Concavity of Tr Xα places the Rényi-trace curve below its tangent at the product point, reducing the comparison to its derivative.Source: §C.2 · p. 11
• The exact derivative identity and the sign analysis imply a strict entropy decrease for every sufficiently small r > 0.Source: §C.2 · pp. 11–12
• The required smallness of r depends on α, so the argument is not uniform as α approaches 1 and does not settle the von Neumann case.Source: §C.2 · p. 12

C.3. The rank endpoint α = 0

• At α = 0, the signed marginal has rank three rather than the product rank four, yielding log2 3 < log2 4.Source: §C.3 · pp. 12–13

Higher Rényi orders

• The proof compresses operator blocks to a nonnegative 2 × 2 scalar matrix and applies norm monotonicity at the threshold p ≥ 4.Source: §D.1 · p. 14, §D.2 · p. 14
• King’s inequality supplies the required direction only for s = p/2 ≥ 2, explaining the order threshold.Source: §D.2 · p. 14, §D.3 · p. 15

Purification decomposition

• A general purification of ρP ⊗ ρQ decomposes into outer P-blocks whose inner leaf matrices are admissible purification amplitudes for Q.The construction preserves Hilbert–Schmidt orthonormality within each fixed P-block.Source: §D.4 · p. 16, §D.5 · p. 17

Norm compression

• For p ≥4, block compression and entrywise monotonicity bound the joint purification norm by the corresponding one-copy quantity.Taking the supremum over joint purifications yields the required upper bound.Source: §D.4 · p. 16

One-copy and converse bounds

• The one-copy optimization is exact: when Q is trivial, the canonical purification attains the reverse inequality, identifying the optimization with ∥A(P)∥p.Tensor products of admissible amplitudes remain admissible and Schatten norms multiply, supplying the converse direction.Source: §D.4 · p. 16

Rényi-order conclusion

• Setting p = 2α proves strong additivity for every finite real α ≥2, while an operator-norm block estimate handles the min-entropy endpoint.The endpoint argument bounds each output block-row norm by the corresponding component of a matrix of operator norms.Source: §D.4 · p. 16

Block structure

• The resulting coefficient matrix has a 2 × 2 outer structure, with each block carrying a 2 × 2 inner structure suited to the compression lemma.Outer indices identify the P-block, while inner indices locate entries within that block.Source: §D.5 · p. 17