Quantum State Exclusion Overview

• Quantum state exclusion is a quantum inference task that rules out potential preparation labels with certainty or minimal average error.
• It is formulated as a semidefinite program and analyzed via convex optimization, symmetry methods, and resource-theoretic measures like the weight of informativeness.
• Recent research extends its scope to many-copy activation, asymptotic error exponents, and foundational implications in nonlocality, contextuality, and channel exclusion.

Quantum state exclusion is the operational task of ruling out hypotheses about the preparation of a quantum system with certainty or with minimal average error. It is complementary to quantum state discrimination: in discrimination one attempts to identify which state was prepared, whereas in exclusion one outputs a label, or more generally a subset of labels, that is guaranteed not to be the true preparation label in the conclusive setting, or is optimized against the event that the excluded label coincides with the truth in the minimum-error setting (1908.10347, Bandyopadhyay et al., 2013). The subject is closely tied to antidistinguishability, the Pusey–Barrett–Rudolph construction, semidefinite optimization, group covariance, resource theories, and, more recently, asymptotic error exponents, many-copy activation, and LOCC separations (Bandyopadhyay et al., 2013, Roy et al., 20 Jan 2026).

1. Operational task and core variants

A state exclusion game is specified by an ensemble $E=\{p(x),\rho_x\}$, with labels $x\in\{1,\dots,k\}$. In the minimum-error formulation, a POVM $\{M_x\}$ is interpreted so that outcome $x$ means “exclude $x$,” and the error event is precisely that the excluded label equals the true label. For a fixed exclusion POVM, the error probability is

$$P_{\mathrm{err}}^{\mathrm{excl}}=\sum_x p_x\,\mathrm{Tr}[\rho_x M_x],$$

and the success probability is $$P_{\mathrm{succ}}^{\mathrm{excl}}=1-P_{\mathrm{err}}^{\mathrm{excl}}$$ (Uola et al., 2019). In the formulation that allows a measurement $M=\{M_a\}$ followed by classical post-processing to a $k$-outcome exclusion POVM $N=\{N_x\}$, the minimum-error quantum state exclusion cost is

$x\in\{1,\dots,k\}$0

with $x\in\{1,\dots,k\}$1 for a stochastic map $x\in\{1,\dots,k\}$2 (1908.10347).

The conclusive, or perfect, version requires zero probability of excluding the true state. For single-state exclusion this means a POVM $x\in\{1,\dots,k\}$3 satisfying

$x\in\{1,\dots,k\}$4

together with $x\in\{1,\dots,k\}$5 and $x\in\{1,\dots,k\}$6 (Bandyopadhyay et al., 2013, Stratton et al., 2024). For pure states $x\in\{1,\dots,k\}$7, the condition $x\in\{1,\dots,k\}$8 implies $x\in\{1,\dots,k\}$9 (Molina, 2017). This perfect form is also called antidistinguishability (Stratton et al., 2024, Roy et al., 20 Jan 2026).

A further variant is unambiguous exclusion, which permits an inconclusive outcome. Then the POVM is $\{M_x\}$0 with $\{M_x\}$1 and $\{M_x\}$2, subject to the no-error constraints $\{M_x\}$3 for all $\{M_x\}$4, while the optimization minimizes the failure probability

$\{M_x\}$5

(Uola et al., 2019, Bandyopadhyay et al., 2013).

The task admits higher-order generalizations. In $\{M_x\}$6-state exclusion, one excludes $\{M_x\}$7 labels per outcome using POVM elements $\{M_x\}$8 indexed by $\{M_x\}$9-subsets $x$0, with conclusive constraints

$x$1

(Stratton et al., 2024). Recent work also distinguishes weak and strong exclusion. Weak exclusion requires feasible conclusive outcomes but does not require that all states or all $x$2-subsets are exhaustively excluded; strong exclusion requires exhaustive coverage and nonzero relevant POVM elements (Stratton et al., 2024, Manna et al., 17 Feb 2026).

A persistent point of confusion is the relation to discrimination. Exclusion is weaker than identification in general, but not always strictly so. For two states, state exclusion is equivalent to perfect discrimination: both are feasible if and only if the states are orthogonal (Roy et al., 20 Jan 2026). For three or more pure states, perfect exclusion can hold even for nonorthogonal sets (Roy et al., 20 Jan 2026, Bandyopadhyay et al., 2013).

2. Convex optimization, optimality conditions, and perfect-exclusion criteria

Quantum state exclusion is naturally formulated as an SDP. For single-state minimum-error exclusion, with weighted states $x$3, the primal problem is

$x$4

while the dual is

$x$5

(Bandyopadhyay et al., 2013). Strong duality holds by Slater’s theorem, and optimality is characterized by the condition that

$x$6

is Hermitian and satisfies $x$7 for all $x$8 (Bandyopadhyay et al., 2013). A later formulation gives the dual SDP for one-shot exclusion as

$x$9

(Ji et al., 2024).

Several general criteria constrain perfect exclusion. A necessary condition for conclusive exclusion of an ensemble $x$0 is

$x$1

where $x$2 is the fidelity (Bandyopadhyay et al., 2013). For $x$3-state conclusive exclusion, a necessary condition in terms of support projectors $x$4 is

$x$5

and this immediately limits the number of labels that can be excluded with certainty (Stratton et al., 2024).

For three pure states, there are exact criteria. If $x$6, then perfect antidistinguishability holds iff

$x$7

(Roy et al., 20 Jan 2026, Manna et al., 17 Feb 2026). In the special case of three pure states in three dimensions, POVMs do not outperform projective measurements for perfect exclusion: if a POVM perfectly excludes the triple, then there also exists an orthonormal-basis measurement that does so (Molina, 2017). Equivalently, with overlaps

$x$8

perfect exclusion is possible iff

$x$9

(Molina, 2017).

A plausible implication is that the geometry of supports and Gram spectra, rather than the mere distinction between POVMs and projections, governs much of the perfect-exclusion landscape. This is explicit in later spectral and group-covariant treatments (Molina, 2017, Diebra et al., 4 Mar 2025).

3. Minimum-error exclusion, informativeness, and exclusion-based information measures

Minimum-error exclusion admits an exact operational interpretation within the quantum resource theory of measurement informativeness. For a POVM $$P_{\mathrm{err}}^{\mathrm{excl}}=\sum_x p_x\,\mathrm{Tr}[\rho_x M_x],$$0, informativeness is quantified by the weight of informativeness

$$P_{\mathrm{err}}^{\mathrm{excl}}=\sum_x p_x\,\mathrm{Tr}[\rho_x M_x],$$1

with closed form

$$P_{\mathrm{err}}^{\mathrm{excl}}=\sum_x p_x\,\mathrm{Tr}[\rho_x M_x],$$2

(1908.10347). This quantifier is faithful, convex, monotone under simulation, and satisfies $$P_{\mathrm{err}}^{\mathrm{excl}}=\sum_x p_x\,\mathrm{Tr}[\rho_x M_x],$$3, with rank-1 projective measurements achieving $$P_{\mathrm{err}}^{\mathrm{excl}}=\sum_x p_x\,\mathrm{Tr}[\rho_x M_x],$$4 (1908.10347).

Its operational meaning is exact. For any measurement $$P_{\mathrm{err}}^{\mathrm{excl}}=\sum_x p_x\,\mathrm{Tr}[\rho_x M_x],$$5,

$$P_{\mathrm{err}}^{\mathrm{excl}}=\sum_x p_x\,\mathrm{Tr}[\rho_x M_x],$$6

where the classical baseline is

$$P_{\mathrm{err}}^{\mathrm{excl}}=\sum_x p_x\,\mathrm{Tr}[\rho_x M_x],$$7

(1908.10347). Thus the optimal relative exclusion advantage of a measurement over an uninformative strategy is precisely the weight-based resource measure. The same work proves that the family $$P_{\mathrm{err}}^{\mathrm{excl}}=\sum_x p_x\,\mathrm{Tr}[\rho_x M_x],$$8 is a complete set of monotones for the simulation preorder on measurements: $$P_{\mathrm{err}}^{\mathrm{excl}}=\sum_x p_x\,\mathrm{Tr}[\rho_x M_x],$$9 can simulate $$P_{\mathrm{succ}}^{\mathrm{excl}}=1-P_{\mathrm{err}}^{\mathrm{excl}}$$0 iff

$$P_{\mathrm{succ}}^{\mathrm{excl}}=1-P_{\mathrm{err}}^{\mathrm{excl}}$$1

(1908.10347).

The information-theoretic counterpart is exclusion entropy and excludible information. For a classical variable $$P_{\mathrm{succ}}^{\mathrm{excl}}=1-P_{\mathrm{err}}^{\mathrm{excl}}$$2,

$$P_{\mathrm{succ}}^{\mathrm{excl}}=1-P_{\mathrm{err}}^{\mathrm{excl}}$$3

and for a classical channel $$P_{\mathrm{succ}}^{\mathrm{excl}}=1-P_{\mathrm{err}}^{\mathrm{excl}}$$4 one defines

$$P_{\mathrm{succ}}^{\mathrm{excl}}=1-P_{\mathrm{err}}^{\mathrm{excl}}$$5

together with the single-shot mutual exclusion information

$$P_{\mathrm{succ}}^{\mathrm{excl}}=1-P_{\mathrm{err}}^{\mathrm{excl}}$$6

(1908.10347). For a measurement channel $$P_{\mathrm{succ}}^{\mathrm{excl}}=1-P_{\mathrm{err}}^{\mathrm{excl}}$$7, the excludible information is

$$P_{\mathrm{succ}}^{\mathrm{excl}}=1-P_{\mathrm{err}}^{\mathrm{excl}}$$8

(1908.10347). This yields a three-way correspondence: $$P_{\mathrm{succ}}^{\mathrm{excl}}=1-P_{\mathrm{err}}^{\mathrm{excl}}$$9 (1908.10347).

A broader resource-theoretic generalization shows that convex weight plays the same role for arbitrary convex quantum resources. For state assemblages, measurement assemblages, and channels, the optimal canonical ratio of exclusion performance achieved by a resource device versus the best free device is exactly $M=\{M_a\}$0 (Uola et al., 2019). This suggests that exclusion tasks are the natural operational partners of weight-like, rather than robustness-like, resource measures (1908.10347, Uola et al., 2019).

4. Symmetry, group-generated ensembles, and explicit solvability

Finite-group symmetry yields one of the most complete analytical pictures presently available. For an ensemble generated from a fiducial pure state by a finite group action,

$M=\{M_a\}$1

the problem reduces to the Gram matrix

$M=\{M_a\}$2

and to covariant POVMs of the form $M=\{M_a\}$3 (Diebra et al., 4 Mar 2025). For arbitrary sets of pure states generated by finite groups, perfect exclusion is possible iff the eigenvalues $M=\{M_a\}$4 of the Gram matrix satisfy

$M=\{M_a\}$5

(Diebra et al., 4 Mar 2025). When perfect exclusion fails, the minimum-error and unambiguous failures are still explicit: $M=\{M_a\}$6

$M=\{M_a\}$7

(Diebra et al., 4 Mar 2025). The optimal POVMs are covariant and rank-1 (Diebra et al., 4 Mar 2025).

A parallel, representation-theoretic treatment derives explicit criteria for conclusive exclusion under finite groups and compact Lie groups. With an isotypic decomposition and amplitudes $M=\{M_a\}$8 across irreducible sectors, a sufficient condition is

$M=\{M_a\}$9

and for finite Abelian groups this becomes necessary and sufficient: $k$0 (Yao et al., 6 Mar 2025). In this formulation, exclusion feasibility becomes a polygon-closure condition in the complex plane, weighted by irrep dimensions (Yao et al., 6 Mar 2025).

The same symmetry machinery reproduces and generalizes the PBR threshold. For the orbit $k$1 generated from $k$2 by $k$3, conclusive exclusion is possible iff

$k$4

(Yao et al., 6 Mar 2025). The group-generated perspective also yields consequences for zero-error communication: if conclusive exclusion of the orbit is feasible, then the feedback-assisted and non-signalling-assisted zero-error capacities satisfy

$k$5

(Yao et al., 6 Mar 2025).

This suggests that symmetry does more than simplify optimization: it exposes the exclusion problem as a spectral feasibility condition on the orbit itself. In the finite-group setting, that condition is complete (Diebra et al., 4 Mar 2025).

5. Many copies, asymptotic exponents, and channel exclusion

The many-copy regime reveals a sharp activation phenomenon. For any finite set of $k$6 pure states, there exists a finite $k$7 such that the tensor-power set

$k$8

is antidistinguishable (Roy et al., 20 Jan 2026). A sufficient bound is obtained from the maximal pairwise overlap $k$9: $N=\{N_x\}$0 (Roy et al., 20 Jan 2026). At the same time, there is no uniform finite bound: for every natural number $N=\{N_x\}$1, there exist pure-state sets for which exclusion remains impossible with $N=\{N_x\}$2 or fewer copies (Roy et al., 20 Jan 2026). For two states, by contrast, no finite number of copies helps unless the states are already orthogonal (Roy et al., 20 Jan 2026).

On the asymptotic minimum-error side, the central quantity is the exclusion error exponent

$N=\{N_x\}$3

with $N=\{N_x\}$4 and $N=\{N_x\}$5 (Ji et al., 2024). A single-letter upper bound is given by the multivariate log-Euclidean Chernoff divergence

$N=\{N_x\}$6

and

$N=\{N_x\}$7

(Ji et al., 2024). This improves the previously known efficiently computable bound based on $N=\{N_x\}$8 (Ji et al., 2024). A companion work re-derives the same asymptotic converse by a divergence-radius method, showing

$N=\{N_x\}$9

(Ji et al., 16 Jan 2025).

These methods extend to channel exclusion. For channels $x\in\{1,\dots,k\}$00, with adaptive strategies permitted, the asymptotic exponent is upper bounded by a reverse divergence radius: $x\in\{1,\dots,k\}$01 where $x\in\{1,\dots,k\}$02 is the Belavkin–Staszewski channel divergence (Ji et al., 16 Jan 2025). The 2024 analysis further gives a single-letter, efficiently computable upper bound on channel-exclusion error exponents even under adaptive strategies, and for classical channels the bound is achievable by a nonadaptive strategy, yielding the exact exponent (Ji et al., 2024).

A distinct channel-level development concerns conclusive $x\in\{1,\dots,k\}$03-state exclusion with entanglement assistance. If Alice encodes into one half of a maximally entangled state and the other half passes through a noisy channel $x\in\{1,\dots,k\}$04 with Choi rank $x\in\{1,\dots,k\}$05, then the maximum number of bit strings that Bob can conclusively exclude obeys

$x\in\{1,\dots,k\}$06

(Stratton et al., 2024). For dephasing channels this bound is tight, while for full-Choi-rank depolarizing channels it gives $x\in\{1,\dots,k\}$07, so no bit string can be excluded with certainty (Stratton et al., 2024).

6. Foundational, nonlocal, and contextual aspects

Quantum state exclusion entered the modern literature in part through the PBR theorem, where conclusive exclusion of product states is used to constrain hidden-variable models (Bandyopadhyay et al., 2013). The SDP analysis of conclusive exclusion yields an analogue of Tsirelson’s bound for the PBR experiment and proves the optimality of the Hadamard-basis measurement used there (Bandyopadhyay et al., 2013).

More recent work shows that exclusion captures genuinely nonclassical operational phenomena that are distinct from those seen in discrimination. One such result is a contextual advantage for conclusive exclusion. In a two-qubit PBR-inspired scenario involving four exclusion tasks, the quantity

$x\in\{1,\dots,k\}$08

satisfies the noncontextuality inequality

$x\in\{1,\dots,k\}$09

while the quantum realization attains

$x\in\{1,\dots,k\}$10

(Yīng et al., 3 Dec 2025). With white noise of visibility $x\in\{1,\dots,k\}$11, the quantum value becomes $x\in\{1,\dots,k\}$12, so violation of the noncontextual bound requires $x\in\{1,\dots,k\}$13 (Yīng et al., 3 Dec 2025). The same bound also functions as a classical bilocal causal-compatibility inequality (Yīng et al., 3 Dec 2025).

Another line concerns LOCC and nonlocality without entanglement. Three bipartite product states can be globally antidistinguishable yet fail to be LOCC-antidistinguishable, and three is the minimal number of states for this phenomenon (Manna et al., 17 Feb 2026). The same work establishes global-versus-LOCC separations for $x\in\{1,\dots,k\}$14-antidistinguishability and gives a tripartite product-state example that is globally antidistinguishable but not LOCC-antidistinguishable across any bipartition, thereby demonstrating genuine nonlocality without entanglement in the exclusion setting (Manna et al., 17 Feb 2026). It also proves a symmetry theorem for LOCC antidistinguishability of product states: if such a set is LOCC-antidistinguishable, then it remains LOCC-antidistinguishable irrespective of the initiating party, although this symmetry can break down for higher-order $x\in\{1,\dots,k\}$15-antidistinguishability (Manna et al., 17 Feb 2026).

These developments correct a common misconception that exclusion is merely a reformulation of discrimination. The data indicate the opposite: exclusion has its own resource-theoretic monotones, its own exact information quantity, different asymptotic converse structure, and distinct LOCC and contextuality phenomena (1908.10347, Ji et al., 2024, Manna et al., 17 Feb 2026).

7. Broader scope and current directions

The contemporary theory of quantum state exclusion spans several mutually reinforcing regimes. In single-shot optimization, SDPs and dual certificates give exact formulations, optimality conditions, and concrete perfect-exclusion criteria (Bandyopadhyay et al., 2013, Molina, 2017). In resource theory, exclusion identifies the operational role of convex-weight and weight-of-informativeness quantifiers (1908.10347, Uola et al., 2019). Under symmetry, finite-group-generated pure-state ensembles admit complete spectral solutions and explicit optimal POVMs (Diebra et al., 4 Mar 2025, Yao et al., 6 Mar 2025). In the asymptotic regime, the best efficiently computable converse bounds are multivariate and barycentric, rather than pairwise, and are expressed through log-Euclidean Chernoff divergences or reverse divergence radii (Ji et al., 2024, Ji et al., 16 Jan 2025). In many-copy settings, exclusion is universally activated for every set of at least three pure states, though the required copy number can be arbitrarily large (Roy et al., 20 Jan 2026).

Several limitations remain explicit in the literature. Exact asymptotic exponents for general nonclassical state ensembles remain open (Ji et al., 2024). The universal many-copy activation theorem is presently proved for pure states, not mixed states (Roy et al., 20 Jan 2026). For non-Abelian group actions, some available criteria are sufficient but not necessary (Yao et al., 6 Mar 2025). Beyond the $x\in\{1,\dots,k\}$16-states-in-$x\in\{1,\dots,k\}$17-dimensions case, the precise boundary between projective and general POVM power in perfect exclusion is not settled in general (Molina, 2017). And while channel exclusion now admits sharp upper bounds and operational Choi-rank limits, structural characterizations of optimal adaptive exclusion protocols are still incomplete (Ji et al., 2024, Stratton et al., 2024).

Taken together, these results place quantum state exclusion alongside discrimination as a distinct inference primitive in quantum information theory: weaker than full identification, but rich enough to support exact SDP characterizations, nontrivial resource-theoretic correspondences, sharp symmetry reductions, asymptotic converse theory, and foundational separations unavailable from discrimination alone (1908.10347, Bandyopadhyay et al., 2013).