Computational Distillable Entanglement

• Computational Distillable Entanglement is the rate at which entanglement can be efficiently extracted using polynomial-time LOCC protocols under resource constraints.
• It reveals significant separations from information-theoretic measures, illustrating cases where maximal entanglement is theoretically present but computationally inaccessible.
• The concept underpins realistic quantum cryptography and network protocols by modeling entanglement as a practical resource limited by computational power.

Computational Distillable Entanglement is a restriction of the traditional concept of distillable entanglement that accounts for the computational resources available for entanglement manipulation protocols. Unlike the information-theoretic distillable entanglement—defined as the optimal asymptotic rate at which maximally entangled pairs can be extracted from many copies of a bipartite quantum state under unrestricted LOCC—computational distillable entanglement evaluates the rate achievable by families of LOCC protocols that are constrained to run in polynomial time (with respect to the state size) or meet similar resource limitations. This framework has become central to accurately quantifying entanglement as a practical resource in realistic settings, such as quantum networks or cryptographic tasks, where the efficiency of classical and quantum processing is fundamentally limited (Leone et al., 17 Feb 2025, Arnon et al., 2023, Ryzov et al., 26 Sep 2025).

1. Formal Definition and Main Properties

Let $\{\psi_{A,B;n}\}_n$ be a family of $n$-qubit bipartite pure states, and let $k(n) = O(\mathrm{poly}\,n)$. A family of LOCC protocols $\{\Gamma_n\}$ is computationally efficient if each $\Gamma_n$ runs in time $\mathrm{poly}(n)$ and acts on at most $k(n)$ copies. A $(k, R, \varepsilon, p)$ distillation protocol consists of an LOCC

$$\Gamma : \ket{\psi_{AB;n}}^{\otimes k} \rightarrow p \ket{0}_X \otimes \omega_{AB} + (1-p) \ket{1}_X \otimes \sigma_{AB}$$

such that $$\tfrac12\|\omega_{AB} - \phi_{AB}^{+\otimes \lfloor kR \rfloor}\|_1 \le \varepsilon$$. The computational distillable entanglement, $\hat E_D^{(k)}(\psi_{AB;n})$, is the supremum over all such achievable rates $R$ (Leone et al., 17 Feb 2025).

Key structural properties (Ryzov et al., 26 Sep 2025):

• Lower-bound convexity: For an ensemble average, the efficient distillation rate is convex in lower bounds.
• Superadditivity: Lower bound on $\hat E_D$ for tensor products is additive across efficiently distillable pairs.
• Monotonicity: Monotonic only under efficient LOCC (not arbitrary LOCC); the measure fails to be an LOCC monotone under inefficient maps.
• Local Unitary Invariance: Invariant only under efficiently implementable local unitaries; not invariant under arbitrary unitaries.

2. Separation from Information-Theoretic Measures

Computational distillable entanglement exhibits strong, sometimes maximal, separations from information-theoretic (Shannon- or von Neumann-entropy based) entanglement measures.

For pure states, classical theory says the asymptotic rate is the von Neumann entropy of the reduced state, $S(\rho_A)$. Under computational constraints: $$R_{\text{comp}}^{\text{distill}}(\psi_{A,B;n}) = \frac{1}{n} H_{\min}(\rho_{A;n})$$ where $H_{\min}(\rho_{A;n}) = -\log \|\rho_{A;n}\|$ is the min-entropy (Leone et al., 17 Feb 2025). This can represent an exponential separation. Families of states exist with $\tilde{\Omega}(n)$ von Neumann entropy but $S_{\min} = o(1)$, so computationally efficient protocols distill essentially nothing, while information-theoretically, they are maximally entangled.

For noisy or mixed states, results show even preparation or distinguishability barriers: there exist large families of states (e.g., those embedded with computationally hard-to-decode structures such as quantum ciphertext) for which efficient LOCC distillation protocols cannot extract more than negligible entanglement, while the information-theoretic measure (with unrestricted resources) is maximal (Arnon et al., 2023).

The table below summarizes key differences:

Scenario	Information-Theoretic $E_D$	Computational $\hat E_D$	Reference
Pure states, large $n$	$S(\rho_A)$	$H_{\min}(\rho_A)$	(Leone et al., 17 Feb 2025)
Crypto-hiding states	Maximal (e.g., $n$ ebits)	$0$	(Arnon et al., 2023)
Generic mixed	Regularization needed	Achievable rate via $\mathrm{poly}$-time LOCC only	(Ryzov et al., 26 Sep 2025)

3. Axiomatic Structure and Operational Criteria

Computational distillable entanglement departs from traditional entanglement monotones in several respects (Ryzov et al., 26 Sep 2025):

• Additivity (lower-bound superadditivity): For block families, the lower bound on the rate is additive under efficiency constraints.
• Monotonicity: Only under efficient LOCC. Pseudo-entanglement and crypto-induced structures break monotonicity for general LOCC.
• Local Unitary Invariance: Only holds under efficiently computable unitaries; fails for arbitrary unitaries due to the exponential size of the unitary group compared to the set of poly-size protocols.
• Extension to uniform/asymptotic scenario: Uniform rates are defined with an $\inf_\epsilon$ over the error and a $\liminf$ over $n$.

4. Protocolic and Complexity-Theoretic Barriers

Results establish that when constrained to polynomial-time protocols, the task of learning or manipulating fine spectral data (eigenvalues) of the state is intractable in general (Leone et al., 17 Feb 2025). In particular:

• The min-entropy dominates achievable rates for distillation; higher Rényi or von Neumann entropies are inaccessible without superpolynomial resources.
• Sample complexity for entropy learning is exponential: $\Omega(2^{n/2})$ for estimating $S_1(\rho_A)$ within constant error, even if $S_1$ itself is constant.
• Pseudo-entangled states (editor's term) constructed from cryptographic primitives (post-quantum one-way functions, commitment schemes) can be efficiently prepared, but their computational distillable entanglement is rigorously zero (Arnon et al., 2023).

5. Implications for Quantum Networks, Cryptography, and Theory

The gap between computational and traditional distillable entanglement has direct implications:

• Quantum cryptography: Pseudo-entanglement, where states are computationally indistinguishable from maximally entangled but distillation is infeasible, enables new security primitives and underlies computational notions of secrecy.
• Quantum networks: Actual usability of entangled links is limited by computationally tractable rates, not just theoretical ones.
• Holography and AdS/CFT: The study of computational bottlenecks in entanglement distillation offers nuanced insights into boundary/bulk dualities, as complexity constraints must be modeled explicitly (Arnon et al., 2023).
• Resource theory: The computational measures do not, in general, yield monotones under the standard symmetry group (all LUs or LOCC), requiring restriction to efficient families.

6. Explicit Examples and Separation Mechanisms

Two explicit constructions exemplify maximal separation:

1. Existential separation: Families of maximally entangled states indexed over an $\epsilon$-net in $U(d)$ admit no poly-size LOCC protocol for nontrivial distillation from all, due to the exponential covering number (Arnon et al., 2023, Ryzov et al., 26 Sep 2025).
2. Cryptographic separation: Using a quantum commitment $C$, the state $(I\otimes C)\Phi^{\otimes n}$ is efficiently preparable, but all efficient protocols leak exponentially little about the commitment, so no efficient LOCC protocol can distill entanglement beyond negligible bounds (Arnon et al., 2023).

These constructions demonstrate that computational inefficiency, not just ignorance of spectral data, fundamentally limits distillation rates.

7. Outlook and Open Problems

Avenues for research highlighted in the literature include:

• Development of computational analogues to entropy duality relations, e.g., a computational version of $H_{\min}(A|E) = -H_{\max}(A|B)$.
• Determining the complexity class of the computational Uhlmann transform and its relation to efficient distillation (Arnon et al., 2023).
• Delineating connections to quantum pseudoentropy and HILL entropy (Arnon et al., 2023).
• Exploring the extension to general (non-pure, correlated, or distributed) states and the implications for distributed quantum computing and error correction.

Open problems focus on characterizing the landscape of computational entanglement monotones, identifying new classes of states with large or subtle computational-information-theoretic separations, and clarifying the resource-theoretic axiomatization under operational constraints.

---

References:

• "Entanglement theory with limited computational resources" (Leone et al., 17 Feb 2025)
• "Computational Entanglement Theory" (Arnon et al., 2023)
• "Properties of computational entanglement measures" (Ryzov et al., 26 Sep 2025)