PaperlayerView exact revision on arXiv

Source-linked AI summary

Strong Converse Exponent of Quantum State Merging

Mario Berta, Hao-Chung Cheng, Roberto Rubboli, Marco Tomamichel

arXiv:2608.27202v1quant-phcs.IT

TL;DR

The paper addresses the open exact squared-fidelity decay rate in the strong converse regime of quantum state merging. It determines this exponent using optimized α-z conditional Rényi entropies along z=α/2, and derives the corresponding partially smoothed conditional min-entropy exponent governed by club-sandwiched entropies.

  • Problem

    The exact squared-fidelity decay rate for quantum state merging in the strong converse regime r < H(A|B)ψ remained open.

  • Method

    The paper analyzes one-way quantum state-merging protocols and proves converse bounds using a fully quantum noncommutative Hölder inequality.

  • Results

    The exact exponent is governed by optimized α-z conditional Rényi entropies with z=α/2 and, equivalently on the purifying AR system, by club-sandwiched conditional entropies.

  • Takeaways & Limitations

    The same club-sandwiched quantity governs partial smoothing, while global smoothing generally has a different sandwiched exponent despite sharing the same first-order threshold.

Abstract

from arXiv · show

We determine the strong converse exponent for the entanglement cost of quantum state merging, showing that it is characterized by the optimized $α$-$z$ conditional Rényi entropies with $z=α/2\in[1/2,1]$. This contrasts with the sandwiched conditional Rényi entropies that typically govern strong converse exponents in quantum information theory. As a consequence, we derive the strong converse exponent of the partially smoothed conditional min-entropy in purified distance. This exponent is governed by club-sandwiched conditional entropies, whereas global smoothing leads to a sandwiched expression.

I. INTRODUCTION

This paper determines the exact strong converse exponent for quantum state merging and identifies an α-z Rényi structure distinct from the commonly occurring sandwiched quantities. It also connects partially smoothed conditional min-entropy to the same club-sandwiched exponent structure.

  • Quantum state merging: Quantum state merging transfers Alice’s subsystem to Bob while preserving coherence with an inaccessible reference.Its optimal asymptotic net entanglement cost is the conditional entropy H(A|B)ψ; negative conditional entropy indicates entanglement generation.
  • Open problem: The paper studies the strong converse regime r < H(A|B)ψ, where optimal fidelity converges exponentially to zero and the exact squared-fidelity decay rate had remained open.Prior work established exponential strong converses and exact exponents only in the low-entanglement-cost direct regime.
  • Main result: The exact exponent is governed by the optimized α-z conditional Rényi entropy with z = α/2 for α ∈ [1,2].For 1 < α < 2, this path belongs neither to the Petz family nor to the sandwiched family, providing an operational interpretation of this α-z path.
  • Proof strategy: The converse reduces state merging to partially smoothed conditional min-entropy and uses noncommutative Hölder methods, auxiliary-state optimization, and club-sandwiched duality.Achievability uses a two-pinching universal-state architecture adapted from quantum randomness extraction while preserving the state-merging marginal.
  • Smoothing comparison: Partial smoothing and state merging have club-sandwiched strong converse exponents, whereas global smoothing has a non-optimized arrow-down sandwiched exponent.In the direct low-entanglement-cost regime, global smoothing, partial smoothing, and state merging share the same error exponent.
  • Scope and organization: The paper’s analysis relies on finite-dimensional Hilbert spaces and compares exponents under the purified-distance criterion.The results are organized around entropic quantities, one-shot reductions, converse bounds, achievability, and the comparison of global and partial smoothing.

B. Rényi divergences and conditional entropies

This section introduces the Rényi divergences and conditional entropies used in the exponent formulas, including α-z, sandwiched, arrow-down, and club-sandwiched forms. Their endpoint conventions, support conditions, and structural properties are specified.

  • The α-z Rényi divergence generates the α-z conditional Rényi entropy, with the sandwiched divergence recovered at z = α.
  • The arrow-down conditional entropy fixes the conditioning operator to the marginal ρR.
  • At α = 1, the conditional Rényi quantities are defined by continuity and equal H(A|B)ρ.
  • For 0 < α < 1 and λ ≤ 0, the club-sandwiched conditional entropy is defined with support constraints and generalized-inverse conventions.

C. Smoothing and auxiliary results

This section defines the smoothing quantities and auxiliary results connecting conditional entropies to one-way state merging. The toolkit includes partially smoothed converses, decoupling achievability, duality, pinching, and optimizer symmetry.

  • A partially smoothed conditional min-entropy imposes the marginal constraint τR ≤ ρR, which enters the merging converse.
  • The one-shot merging converse bounds protocols with purified-distance error using initial and final Schmidt ranks and their net entanglement cost.
  • Fixed-output relative-entropy decoupling supplies one-Kraus-per-outcome instruments achieving coherent gain q below H(A|R)τ.
  • For pure ψABR, club-sandwiched duality relates conditional entropies on complementary systems and is used in the exponent analysis.
  • Universal states and permutation symmetry provide commuting optimizers and permit permutation-invariant choices in the relevant conditional-entropy optimization.
  • Pinching estimates control fidelity loss when the second operator is invariant under the pinching.

III. QUANTUM STATE MERGING AND THE STRONG CONVERSE EXPONENT

The paper formalizes one-way quantum state merging through protocols constrained by net entanglement cost and optimal squared fidelity. It establishes an exact strong converse exponent and expresses it equivalently through club-sandwiched duality and α-z conditional Rényi entropies.

  • A one-way protocol starts with shared entanglement, applies Alice’s instrument, communicates its outcome, and lets Bob produce the merged output.
  • Only the difference log Kn − log Ln is constrained, defining the protocol’s net entanglement cost per copy.
  • Theorem 3.1 establishes the exact strong converse exponent for every finite-dimensional pure ψABR and every rate r.
  • With q := −r, club-sandwiched conditional-entropy duality gives an equivalent representation of the exponent.
  • The strong converse exponent is positive when the entanglement cost satisfies r < H(A|B)ψ.
  • The converse uses a one-shot partially smoothed min-entropy bound, noncommutative Hölder inequalities, and optimization over the α range [1, 2].

V. ACHIEVABILITY

The achievability proof reduces state merging to fixed-output decoupling and first obtains a Log-Euclidean exponent. Pinching and variational arguments then connect the construction to the target entropy expression while controlling fidelity and rate limits.

  • Achievability reduces merging to fixed-output decoupling, converts fixed-output fidelity by controlled Uhlmann reasoning, and applies double pinching.
  • The variational identity rewrites the exponent objective as a minimization over τAR involving relative entropies and [q − H(A|R)τ]+.
  • A minimizing τAR exists by compactness and lower semicontinuity, enabling the construction to use the corresponding decoupling instrument.
  • The first decoupling term converges to zero exponentially, providing the vanishing-error component of the construction.
  • Padding and isometric embeddings reconcile physical and declared output dimensions while preserving the required asymptotic gain.
  • The resulting declared gain is at least q and converges to q, while the fidelity bound yields the claimed achievability exponent.

B. Controlled Uhlmann recovery

Controlled Uhlmann recovery converts a one-Kraus-per-outcome instrument into a state-merging protocol using outcome-dependent isometries on Bob. The construction preserves the relevant Schmidt-rank accounting and provides a fidelity lower bound after discarding auxiliary registers.

  • Controlled Uhlmann recovery: Outcome-dependent recovery isometries for Bob turn the instrument into a state-merging protocol.The recovery follows from Uhlmann’s theorem applied to each subnormalized post-measurement branch.
  • Controlled Uhlmann recovery: The protocol’s initial and final Schmidt ranks are |KA| and |LA|, respectively.This identifies the entanglement dimensions before and after the recovery construction.
  • Controlled Uhlmann recovery: The construction remains valid when Alice’s physical output is isometrically embedded into a larger declared output space.The embedding does not invalidate the recovery statement.
  • Controlled Uhlmann recovery: The squared fidelity is bounded from below by summing the branch overlaps, with the factor N arising from normalized branch targets.The final inequality uses Cauchy–Schwarz after outcome registers and decoder environments are discarded.

C. Double pinching

The double-pinching construction makes the joint state, its marginal, and an auxiliary universal state simultaneously compatible with the commutative analysis needed for the converse and achievability bounds. Log-Euclidean expressions can then be compared with club-sandwiched quantities while controlling pinching losses.

  • C. Double pinching: Double pinching preserves the quantum marginal while simultaneously commuting the joint state, its marginal, and an auxiliary universal state.This addresses the noncommutative obstacle that a single pinching cannot resolve.
  • C. Double pinching: The universal state ωRm is permutation-invariant and commutes with ρ⊗mR, so their spectral projections commute.These commutation properties enable the corresponding pinching maps to be applied together.
  • C. Double pinching: For α ∈ [1/2,1), the Log-Euclidean-to-club-sandwiched comparison uses a permutation-invariant optimizer.The resulting candidate is suitable for the commutative reduction.
  • C. Double pinching: Trace-exponential monotonicity permits replacing the optimizer by ωRm with an additive correction, after which simultaneous diagonalization yields the common trace expression.The argument relies on trace monotonicity rather than operator monotonicity of the matrix exponential.
  • C. Double pinching: Data processing and additivity of the club-sandwiched conditional entropy control the effect of both pinchings.The instrument acts trivially on the reference system and therefore commutes with the blockwise pinchings.
  • C. Double pinching: Singular optimizers are handled by standard full-rank regularization.This extends the comparison beyond full-rank auxiliary states.

D. The achievability proof

The achievability proof builds state-merging protocols from block instruments, double pinching, and controlled recovery, then removes blocklength slack by teleporting the remainder. The lower and upper bounds converge to the same exponent, which is positive exactly below the conditional-entropy threshold.

  • D. The achievability proof: A block instrument at coherent gain mq is applied to k copies, followed by double pinching and controlled Uhlmann recovery.This yields a state-merging protocol with cost at most mkr.
  • D. The achievability proof: For n = mk + t with 0 ≤ t < m, the remainder is teleported exactly, enforcing the rate without asymptotic slack.The remainder consumes t log |A| ebits.
  • D. The achievability proof: Canonical isometric embeddings enlarge the preliminary output registers while preserving the exact rate constraint.The enlarged final registers have dimension Ln := MnL′n.
  • D. The achievability proof: Teleportation of the remainder introduces no additional fidelity loss, and the asymptotic rate and fidelity exponent remain unchanged as k → ∞.For fixed m, the remainder and logarithmic penalty are bounded independently of k.
  • D. The achievability proof: Theorem 4.3 and Theorem 5.6 match, so the strong converse exponent limit exists and equals the common expression.The lower-bound liminf and upper-bound limsup coincide.
  • D. The achievability proof: The exponent is positive when r < H(A|B)ψ, while rates r ≥ H(A|B)ψ yield zero contribution at the endpoint.Purity relates the threshold to −H(A|R)ψ in the proof.

VI. CONCLUSION AND DISCUSSION

The paper identifies exact strong converse exponents for state merging and partially smoothed conditional min-entropy. State merging follows the club-sandwiched form, while global smoothing follows the sandwiched form because physical successful branches preserve the reference marginal.

  • VI. CONCLUSION AND DISCUSSION: For every finite-dimensional pure state and net entanglement-cost rate, the state-merging strong converse exponent is exact.Its positivity occurs precisely below H(A|B)ψ.
  • VI. CONCLUSION AND DISCUSSION: On AB, state merging is governed by optimized α-z conditional Rényi entropy along z = α/2, with α ∈ [1,2].On the purifying AR system, the equivalent form is club-sandwiched conditional entropy.
  • VI. CONCLUSION AND DISCUSSION: The partially smoothed conditional min-entropy also has an exact strong converse exponent for every finite-dimensional bipartite state.Partial smoothing imposes the additional constraint τR ≤ ρR.
  • VI. CONCLUSION AND DISCUSSION: Partial smoothing makes a conditional min-entropy threshold harder to attain, and its success quantity decays at least as fast as under global smoothing.Both smoothing notions retain the same first-order threshold H(A|R)ρ.
  • VI. CONCLUSION AND DISCUSSION: Global smoothing permits a marginal mismatch to be offset by increasing H(A|R)η, whereas partial smoothing separately pays the full marginal divergence cost.The coefficient constraint selects the boundary club-sandwiched entropy for partial smoothing and the ordinary sandwiched entropy for global smoothing.
  • VI. CONCLUSION AND DISCUSSION: State merging inherits partial smoothing because every physical successful component obeys the fixed reference-marginal constraint τR ≤ ψR.This follows from trace-nonincreasing operations on Alice’s and Bob’s systems.

Appendix A: Exact strong converse exponent of the partially smoothed conditional min-entropy

The appendix proves an exact strong converse exponent for partially smoothed conditional min-entropy of arbitrary finite-dimensional bipartite states, using matching converse and achievability bounds. The exponent is positive exactly when q exceeds the conditional entropy H(A|R)ρ.

  • Theorem A.1 gives the exact strong converse exponent for partially smoothed conditional min-entropy for every finite-dimensional bipartite state ρAR and threshold q.
  • The exponent is positive if and only if q > H(A|R)ρ, with the α = 1 contribution defined as zero by continuity.
  • The converse follows from Lemma 4.1, while achievability uses a Log-Euclidean change-of-measure and pinching construction with joint and marginal soft caps.
  • The construction restricts R to supp(ρR) and selects ηAR supported within supp(ρAR), with positive slack δ > 0.
  • The proof verifies feasibility of the constructed operator τn and controls the remaining terms through data processing, Petz Rényi divergences, additivity, and exponential convergence.
  • The feasible optimization attains its maximum because the feasible set is closed, compact, and strictly positive.
  • Tensor-product feasibility and multiplicativity of squared fidelity make −log Γn(ρ,q) subadditive, so Fekete’s lemma establishes existence of the asymptotic limit.
  • The lower bound follows from Lemma 4.1 and additivity of the club-sandwiched conditional entropy; blockwise double pinching yields the matching upper bound.
Loading 2608.27202v1…