Quantum Data Hiding: Concepts & Constructions

• Quantum Data Hiding is the encoding of classical values into distributed quantum states such that LOCC operations yield near-random results while global measurements reveal the information perfectly.
• Key constructions employ Bell-state mixtures, Werner-type states, and group-symmetric separable ensembles to exploit the gap between LOCC and unrestricted discrimination.
• Recent advances leverage PPT relaxations, tensor-power protocols, and multipartite extensions to enhance security and practical applicability in quantum cryptographic tasks.

Searching arXiv for recent and foundational papers on quantum data hiding. Quantum data hiding (QDH) is the task of encoding a classical value into a distributed quantum state so that parties restricted to local operations and classical communication (LOCC) cannot recover the value, while a joint global measurement can recover it reliably or perfectly. In the standard bipartite formulation, one studies pairs or ensembles of states that are perfectly or nearly perfectly distinguishable under unrestricted measurements yet almost indistinguishable under LOCC; across the literature, QDH has been developed as a theory of restricted distinguishability, asymptotic coding by tensor-product repetition, geometric norm separation, and multipartite cryptographic functionality (Lami et al., 2017, Ha et al., 2023).

1. Operational formulation and basic figures of merit

A bipartite ensemble is written as

$\mathcal E=\{\eta_i,\rho_i\}_{i=0}^{n-1},$

with $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$, $\eta_i\ge 0$, and $\sum_i\eta_i=1$. The central operational quantities are the optimal global discrimination probability

$p_G(\mathcal E),$

and the optimal LOCC discrimination probability

$p_L(\mathcal E).$

An ensemble is operationally nonlocal when $p_L(\mathcal E)<p_G(\mathcal E)$, and it induces a QDH scheme when the hiding step leaves LOCC performance close to random guessing while a revealing step based on a joint measurement achieves $p_G=1$ (Ha et al., 2023).

For binary hiding pairs $(\rho,\sigma)$, the unrestricted success probability is governed by the trace norm through the Holevo–Helstrom formula, while the restricted success probability is governed by an LOCC norm. In this language, QDH is the existence of states with $\|\rho-\sigma\|_1=2$ but $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$0. A common quantitative benchmark is the data hiding ratio

$$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$1

and the maximal bipartite quantum-mechanical ratio satisfies

$$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$2

which identifies local dimension as the controlling asymptotic resource for finite-dimensional bipartite hiding (Lami et al., 2017).

Recent work also formulates QDH against a general restricted measurement class $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$3. In that setting, a pair $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$4 is $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$5-data hiding against $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$6 when

$$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$7

This formulation isolates the global decoding advantage $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$8 from the adversarial restricted advantage $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$9, and it is useful in later connections to verification and subspace testing (Akibue et al., 1 Sep 2025).

2. Restricted discrimination, PPT relaxations, and asymptotic hiding criteria

Because direct optimization over LOCC measurements is difficult, much of the modern theory passes through PPT-relaxed discrimination. For bipartite ensembles, one defines an efficiently analyzable upper bound $\eta_i\ge 0$0 by optimizing against partial transposes of the code states; it satisfies

$\eta_i\ge 0$1

For two-state ensembles,

$\eta_i\ge 0$2

which turns restricted distinguishability into a trace-norm computation on the partial transpose (Ha et al., 2023).

Ha and Kim sharpen this PPT perspective by introducing the bias operator

$\eta_i\ge 0$3

If there exists a Hermitian operator $\eta_i\ge 0$4 such that

$\eta_i\ge 0$5

then

$\eta_i\ge 0$6

This produces a sufficient local-discrimination bound directly from a partial-transpose decomposition. For tensor-power repetition, the same framework yields a sufficient condition for asymptotic hiding: if there exists $\eta_i\ge 0$7 with

$\eta_i\ge 0$8

then for the $\eta_i\ge 0$9-fold ensemble

$\sum_i\eta_i=1$0

and if $\sum_i\eta_i=1$1, then $\sum_i\eta_i=1$2 exponentially, so the PPT and therefore LOCC success probability converges to random guessing while global decoding remains perfect because of orthogonality (Ha et al., 25 Feb 2025).

The same logic extends beyond binary alphabets. For an orthogonal $\sum_i\eta_i=1$3-state ensemble $\sum_i\eta_i=1$4, if

$\sum_i\eta_i=1$5

then an $\sum_i\eta_i=1$6-ary hiding protocol follows by taking $\sum_i\eta_i=1$7-fold tensor powers, classifying strings by their mod-$\sum_i\eta_i=1$8 sum, and broadcasting a masked classical side value $\sum_i\eta_i=1$9. In the induced ensemble, the restricted success probability converges to $p_G(\mathcal E),$0, so the best LOCC strategy becomes asymptotically no better than blind guessing (Ha et al., 2023).

3. Canonical constructions and optimal scaling

The traditional QDH paradigm, associated in the data block with DiVincenzo–Terhal and related work, uses Bell-state mixtures or Werner-type states. In these constructions, orthogonality and strong nonlocality ensure that global measurements distinguish the code states while LOCC protocols do not. A canonical example is the symmetric/antisymmetric Werner pair: if $p_G(\mathcal E),$1 and $p_G(\mathcal E),$2 are the projectors onto the symmetric and antisymmetric subspaces, then

$p_G(\mathcal E),$3

so the hiding ratio already scales linearly in the local dimension parameter $p_G(\mathcal E),$4 (Lami et al., 2017).

The upper bound is controlled by teleportation arguments. For $p_G(\mathcal E),$5,

$p_G(\mathcal E),$6

which implies

$p_G(\mathcal E),$7

Together with Werner-type lower bounds, this establishes the exact quantum scaling

$p_G(\mathcal E),$8

and, in the broader GPT framework, the maximal hiding ratio is universally bounded by the minimum local real dimension (Lami et al., 2017).

More recent analysis clarifies the structural limits of earlier families. Werner states can be made either separable or globally perfectly orthogonal, but not both; random states can hide many more bits, but they are typically entangled and only approximately orthogonal. This diagnosis motivates the explicit construction of new group-symmetric hiding states that are simultaneously separable, perfectly orthogonal, and invariant under partial transpose, thereby pushing “nonlocality without entanglement” to an extremal form (Mele et al., 3 Oct 2025).

4. Orthogonal separable-state hiding

A major recent development is the construction of QDH schemes from orthogonal separable ensembles. In a $p_G(\mathcal E),$9 example due to Ha and Kim, one sets $p_L(\mathcal E).$0 and defines

$p_L(\mathcal E).$1

where $p_L(\mathcal E).$2 and $p_L(\mathcal E).$3 are orthonormal bases of $p_L(\mathcal E).$4 chosen so that $p_L(\mathcal E).$5 whenever $p_L(\mathcal E).$6. With

$p_L(\mathcal E).$7

one has $p_L(\mathcal E).$8 and

$p_L(\mathcal E).$9

Hence $p_L(\mathcal E)<p_G(\mathcal E)$0 exponentially in $p_L(\mathcal E)<p_G(\mathcal E)$1, while $p_L(\mathcal E)<p_G(\mathcal E)$2. The same paper gives an odd-$p_L(\mathcal E)<p_G(\mathcal E)$3 generalization and emphasizes that low-dimensional separable state preparation shifts experimental difficulty away from entanglement generation and toward the final global decoding measurement (Ha et al., 25 Feb 2025).

The implementation remarks attached to this framework are unusually concrete. Because the codewords are separable, the preparation stage requires only local product-vector synthesis. Repeating the same $p_L(\mathcal E)<p_G(\mathcal E)$4 separable density $p_L(\mathcal E)<p_G(\mathcal E)$5 times is experimentally simple in the sense stated by the paper, and proposed platforms include photonic qutrits, spin-1 systems such as NV centers or trapped ions, and orbital-angular-momentum modes of photons. The revealing measurement is a joint projection onto the orthogonal supports of $p_L(\mathcal E)<p_G(\mathcal E)$6 and $p_L(\mathcal E)<p_G(\mathcal E)$7, realizable in principle by sequential two-system interactions or a collective interferometer (Ha et al., 25 Feb 2025).

The separable line was extended further in a later two-qubit construction. There, for a two-state ensemble $p_L(\mathcal E)<p_G(\mathcal E)$8 with $p_L(\mathcal E)<p_G(\mathcal E)$9, if there exist an integer $p_G=1$0 and a Hermitian $p_G=1$1 such that

$p_G=1$2

then

$p_G=1$3

An explicit example uses

$p_G=1$4

$p_G=1$5

with $p_G=1$6 and $p_G=1$7. The paper states that this is the first one-bit hiding scheme using only two-qubit separable states, i.e. the smallest possible Hilbert-space dimension (Ha et al., 17 Dec 2025).

A separate explicit separable program uses group symmetry rather than low local dimension. On $p_G=1$8, the states

$p_G=1$9

are orthogonal, separable, and invariant under partial transpose. For tensor powers, the LOCC bias satisfies

$(\rho,\sigma)$0

and the resulting $(\rho,\sigma)$1-hiding states can be realized with local dimension

$(\rho,\sigma)$2

This construction was presented as an explicit alternative to Werner and random-state families (Mele et al., 3 Oct 2025).

5. Multipartite, continuous-variable, Gaussian, and noisy-channel extensions

QDH generalizes naturally from bipartite to multipartite access structures. In the $(\rho,\sigma)$3-player scheme of Ha and Kim, one begins with an orthogonal $(\rho,\sigma)$4-partite ensemble $(\rho,\sigma)$5 satisfying $(\rho,\sigma)$6 and

$(\rho,\sigma)$7

where $(\rho,\sigma)$8 ranges over nontrivial partitions and $(\rho,\sigma)$9 is a partial-transpose upper bound for $\|\rho-\sigma\|_1=2$0-LOCC discrimination. After $\|\rho-\sigma\|_1=2$1-fold repetition and mod-$\|\rho-\sigma\|_1=2$2 sum encoding, the induced hidden ensemble $\|\rho-\sigma\|_1=2$3 obeys

$\|\rho-\sigma\|_1=2$4

for every non-global partition $\|\rho-\sigma\|_1=2$5, with $\|\rho-\sigma\|_1=2$6. Thus non-global decoding becomes exponentially close to random guessing, while global decoding remains perfect. The same work gives GHZ-based examples and an explicit $\|\rho-\sigma\|_1=2$7-bit hiding construction using $\|\rho-\sigma\|_1=2$8 orthogonal states (Ha et al., 2024). A different multipartite protocol concatenates blockwise subprotocols across partitions and adds remote deletion: the sender can measure the retained subsystem so that the post-abort receiver state becomes independent of the hidden value (1804.01982).

Continuous-variable QDH requires different control parameters because local Hilbert-space dimension is infinite. Lami proved an LOCC lower bound in terms of mean photon number: for $\|\rho-\sigma\|_1=2$9 modes on one side, energy budget $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$00, global bias $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$01, and $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$02,

$$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$03

In particular, with fixed $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$04, even orthogonal states cannot be hidden below a bias $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$05. The same paper also gives a single-mode GOCC hiding example using even and odd thermal states $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$06 and $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$07, for which $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$08 but $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$09 as $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$10 (Lami, 2021). Wang and Smith develop related Gaussian constructions: one bit can be hidden against Gaussian LOCC by mixtures of displaced two-mode squeezed states, and against arbitrary Gaussian measurements by two-mode thermal states in the weak-strength limit (Wang et al., 2 Feb 2025).

In noisy-network settings, QDH is formulated as a channel coding problem. For a quantum broadcast channel $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$11, the bit-hiding capacity $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$12 is the optimal asymptotic rate at which a joint decoder can recover a classical message while every LOCC decoder sees an output distribution close to that of a fixed separable state. A regularized upper bound is

$$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$13

For pure-loss bosonic broadcast channels with coherent-state encodings of mean photon number $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$14, any coherent-state data-hiding protocol has rate at most

$$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$15

bits per use. The same work also proves a lower bound for mictodiactic channels, defined by $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$16, thereby initiating a systematic noisy-channel theory of QDH (Lupo et al., 2015).

6. Relation to masking, private states, and quantumness of correlations

QDH for classical information must be distinguished from hiding or masking arbitrary quantum information. Zhu’s results show that in complex quantum mechanics any informationally complete set $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$17 is antiscrambling and not maskable. This strengthens the Braunstein–Pati no-hiding theorem and the Modi et al. no-masking theorem from the full pure-state or full state space to every informationally complete subset, including 2-designs such as SICs and complete MUBs. By contrast, real quantum mechanics admits masking of all real states: if $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$18 is divisible by

$$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$19

one can construct an isometry

$$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$20

from a Hurwitz–Radon set $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$21 so that both marginals are constant for every real input state. Each subsystem must then have dimension at least $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$22, so the overhead is exponential in $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$23 (Zhu, 2020). This suggests that classical-bit data hiding and universal quantum-state masking are sharply different tasks.

QDH also appears as a structural ingredient in private-state cryptography. In the Bell-private representation of a private state, the shield system carries an ensemble $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$24 that hides the Bell-phase index from local observers. Kaur, Wilde, and Winter show that this hidden phase information both protects secret key and obstructs entanglement distillation; for key-correlated states, the one-way distillable entanglement is governed by a restricted relative entropy and, when the averaged shield state is separable, satisfies

$$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$25

In that sense, the key–entanglement gap is directly tied to data hiding in the shield (Christandl et al., 2016).

A complementary perspective comes from disturbance-based quantifiers of ensemble quantumness. For an equal-weight hiding ensemble $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$26, the hiding gap obeys

$$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$27

Thus, even though entanglement is not necessary for good QDH, ensemble quantumness of correlations is necessary. The paper explicitly emphasizes that a good hiding scheme may use separable or highly mixed states, but the ensemble itself must still be strongly nonclassical in the disturbance sense (Piani et al., 2014).

7. Robustness, limitations, and current directions

The robustness question asks whether auxiliary quantum resources can break hiding. A recent unified framework considers catalytic local discrimination and memory-assisted discrimination. If the hiding states $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$28 are separable, then neither an entangled catalyst returned unchanged nor reusable quantum memory can improve the optimal LOCC discrimination probability: $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$29 For some entangled hiding pairs, however, a reusable quantum memory can raise the asymptotic success rate to arbitrarily close to one: $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$30 The resulting dichotomy identifies separable encodings as a robust strategy when adversaries may possess auxiliary quantum resources (Philip et al., 6 Nov 2025).

Another recent direction connects QDH to quantum state verification. A pure state is most difficult to verify in QSV if and only if it is most secure in QDH with respect to the relevant parameters, and for extremal pure states the two fundamental quantities coincide. The same work extends the correspondence from pure-state QDH to mixed-state QDH and quantum subspace verification, showing that verification sample complexity and hiding security are controlled by the same restricted-measurement geometry (Akibue et al., 1 Sep 2025).

Several open problems remain explicit in the cited literature. One is whether the multipartite sufficient condition $$\rho_i\in\mathcal D(\mathcal H_A\otimes\mathcal H_B)$$31 is also necessary (Ha et al., 2024). Another is the classification of maximal maskable subtheories in complex quantum mechanics, beyond the real subtheory; the hyperdisk conjecture is stated as open (Zhu, 2020). In noisy settings, extending lower bounds from mictodiactic channels to arbitrary broadcast channels remains an identified problem (Lupo et al., 2015). In resource-assisted discrimination, the largest class of hiding states secure against catalysts or memory beyond the separable case is likewise open (Philip et al., 6 Nov 2025). Together these directions indicate that QDH has developed from a bipartite state-discrimination phenomenon into a broader framework connecting nonlocality without entanglement, cryptographic access structures, restricted-measurement geometry, and operational resource theory.