Multi-Party Quantum Sequence Discrimination

• Multi-party quantum sequence discrimination is a set of quantum information tasks where multiple parties identify sequences of nonorthogonal quantum states under operational constraints.
• Techniques such as product POVMs, adaptive LOCC, and sequential measurements optimize success probabilities while balancing information gain against disturbance.
• The framework distinguishes distinct settings—from independent ensembles to sequential protocols—with significant implications for communication, cryptography, and resource preservation.

Searching arXiv for papers on multi-party quantum sequence discrimination and closely related sequential/state-discrimination results. Multi-party quantum sequence discrimination is the family of quantum-information tasks in which several parties, sites, or time slots must identify a state sequence drawn from a known description, subject to quantum nonorthogonality and operational constraints. The phrase covers at least three distinct but related settings: independent sequence ensembles, where each site receives a state drawn secretly and independently from a local ensemble; distributed multi-copy discrimination, where several parties share $N$ copies of the same unknown state; and sequential protocols, where a single carrier is measured successively by multiple observers. Across these settings, the central structural question is whether optimal performance requires collective measurements, adaptive LOCC, or only fixed local measurements, and the principal figures of merit are minimum-error success probability, unambiguous success probability, and maximum confidence (Gupta et al., 2024, Hillery et al., 2017, Achenbach et al., 29 Apr 2026, Lee et al., 2024).

1. Operational scope and formal models

The literature uses closely related terminology for operationally different discrimination problems. In all cases, the task is specified by a known ensemble and a constrained measurement class, but the tensor-product structure plays different roles.

Setting	State structure	Representative structural result
Independent sequence ensemble	$$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$ with product priors	Product POVMs attain the optimum for minimum-error and unambiguous discrimination
Distributed multi-copy discrimination	$\rho_i^{\otimes N}$	GLOBAL, LOCC, classical-bit, and GPT-restricted strategies can differ
Sequential reuse of one system	One state is passed through $K$ observers	Information gain is traded against disturbance across observers
LOCC sequence discrimination	Product of multi-party ensembles under LOCC constraints	Factorization holds under explicit LOCC and separable-duality conditions

For minimum-error multi-copy discrimination, one starts from an ensemble

$\mathcal{E}=\{(p_i,\rho_i)\}_{i=1}^M,$

receives $N$ identical copies $\rho_i^{\otimes N}$, and maximizes

$P_{\rm succ} =\sum_{i=1}^M p_i\,\Tr\!\bigl[\Pi_i\,\rho_i^{\otimes N}\bigr]$

over POVMs $\{\Pi_i\}$ on the $N$-copy Hilbert space. Equivalently one minimizes $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$0 (Achenbach et al., 29 Apr 2026).

For independent sequence ensembles, the $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$1-th site holds a state from a known local ensemble

$$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$2

with product priors over the global sequence

$$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$3

The global objective is to identify the entire index tuple $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$4, either by minimizing average error or by maximizing conclusive success under no-error constraints (Gupta et al., 2024).

Sequential protocols differ fundamentally from both. Here a single nonorthogonal state is passed through a chain of observers, each of whom extracts partial information while leaving a post-measurement state for the next observer. This makes disturbance, rather than tensor-product measurement complexity, the primary limitation (Hillery et al., 2017, Lee et al., 2024).

2. Factorization for independently drawn sequences

For independent quantum sequences, the basic structural theorem is exceptionally strong: in both minimum-error and unambiguous discrimination, the optimal success probability factorizes,

$$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$5

and is achieved by the product POVM

$$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$6

where each local POVM is itself optimal for the corresponding single-site ensemble (Gupta et al., 2024).

The minimum-error proof is based on the Holevo–Yuen–Kennedy–Lax optimality conditions together with the positivity lemma that $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$7 and $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$8 imply $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$9 for PSD operators. The unambiguous case is cast as an SDP; tensor products of local primal and dual optimizers give matching feasible global solutions, and strong duality forces optimality. A direct consequence is that no entangling or joint quantum operation across the $\rho_i^{\otimes N}$0 parties is needed for independent sequences: each party can perform its locally optimal measurement and broadcast the classical outcome (Gupta et al., 2024).

For two-state qubit examples, if site $\rho_i^{\otimes N}$1 receives one of two pure states with priors $\rho_i^{\otimes N}$2, the local Helstrom error is

$\rho_i^{\otimes N}$3

and the sequence error becomes

$\rho_i^{\otimes N}$4

For unambiguous discrimination, the global conclusive success probability is likewise the product of the local optima (Gupta et al., 2024).

Under LOCC restrictions on multipartite states, factorization is subtler. For an ensemble $\rho_i^{\otimes N}$5, the success probabilities satisfy

$\rho_i^{\otimes N}$6

and for a sequence $\rho_i^{\otimes N}$7 one says that LOCC discrimination is factorizable when

$\rho_i^{\otimes N}$8

A necessary and sufficient condition for the single-shot “guess the most likely state” strategy is

$\rho_i^{\otimes N}$9

where $K$0. In the sequence setting, LOCC can achieve global minimum-error discrimination on the full sequence if and only if it can do so at each step individually. The same work also exhibits both a non-factorizable GHZ-type sequence and a factorizable “one-entangled” sequence, showing that LOCC factorization is neither automatic nor equivalent to trivial guessing (Ha et al., 7 Aug 2025).

A recurrent misconception is therefore incorrect in both directions. Collective measurements are not generically necessary for independent sequences, but locality constraints can become genuinely nontrivial once the local states themselves are multipartite and LOCC-restricted (Gupta et al., 2024, Ha et al., 7 Aug 2025).

3. Sequential discrimination by multiple observers

In sequential state discrimination, a single quantum system is passed through multiple observers, each of whom attempts to identify the preparation while preserving enough residual distinguishability for later observers. The standard unambiguous version was developed for $K$1-dimensional qudits prepared in one of $K$2 nonorthogonal states $K$3, including both equal-overlap and two-set-overlap geometries (Hillery et al., 2017).

For equal overlap $K$4 with $K$5, the single-observer unambiguous POVM has conclusive elements

$K$6

with $K$7. In the sequential extension to $K$8 observers $K$9, if $\mathcal{E}=\{(p_i,\rho_i)\}_{i=1}^M,$0 denotes the overlap after the $\mathcal{E}=\{(p_i,\rho_i)\}_{i=1}^M,$1-th observer and $\mathcal{E}=\{(p_i,\rho_i)\}_{i=1}^M,$2 that observer’s failure probability, the recursion is

$\mathcal{E}=\{(p_i,\rho_i)\}_{i=1}^M,$3

The all-success probability is

$\mathcal{E}=\{(p_i,\rho_i)\}_{i=1}^M,$4

while the probability that at least one observer succeeds is

$\mathcal{E}=\{(p_i,\rho_i)\}_{i=1}^M,$5

For the symmetric choice $\mathcal{E}=\{(p_i,\rho_i)\}_{i=1}^M,$6, one obtains

$\mathcal{E}=\{(p_i,\rho_i)\}_{i=1}^M,$7

The same work states that there is no penalty in failure probabilities for going from qubits to $\mathcal{E}=\{(p_i,\rho_i)\}_{i=1}^M,$8-dimensional qudits, while higher dimension increases the information per carrier through the $\mathcal{E}=\{(p_i,\rho_i)\}_{i=1}^M,$9 factor in the channel capacity $N$0 (Hillery et al., 2017).

A distinct instrument-based framework treats conclusive minimum-error discrimination by $N$1 sequential receivers for arbitrary $N$2 states, pure or mixed, in arbitrary dimension. If $N$3 is the instrument used by the $N$4-th receiver, the composed instrument is

$N$5

and the all-correct success probability is

$N$6

Its central “no-loss” theorem states that under an explicit instrument-decomposition condition the optimal success probability for $N$7 receivers equals that of the first receiver. In particular, for $N$8 this condition is always satisfiable, so for any two states one has

$N$9

for all $\rho_i^{\otimes N}$0, i.e. the Helstrom bound persists for arbitrarily many sequential receivers (Loubenets et al., 2021).

The experimental two-observer realization with polarization single photons implements the unambiguous version on an optical network. In that setting, the optimal joint success probability with no classical communication during the protocol is

$\rho_i^{\otimes N}$1

and for the Bob–Charlie case this gives $\rho_i^{\otimes N}$2 (Solís-Prosser et al., 2015).

4. Maximum-confidence discrimination and shared uncertainty

Maximum-confidence (MC) discrimination interpolates between minimum-error and unambiguous discrimination by maximizing, for each conclusive outcome $\rho_i^{\otimes N}$3, the conditional confidence

$\rho_i^{\otimes N}$4

The single-party optimality conditions are

$\rho_i^{\otimes N}$5

and the average guessing probability is

$\rho_i^{\otimes N}$6

In the sequential setting, the sharp structural statement is that equally high confidence can be maintained across all parties if and only if the conclusive POVM elements of the optimal single-party MC measurement are linearly independent (Lee et al., 2024).

For two-state sequential MC discrimination, the disturbance–information relation can be written as

$\rho_i^{\otimes N}$7

where $\rho_i^{\otimes N}$8 is the overlap of the two rank-one conclusive POVM elements at stage $\rho_i^{\otimes N}$9, $P_{\rm succ} =\sum_{i=1}^M p_i\,\Tr\!\bigl[\Pi_i\,\rho_i^{\otimes N}\bigr]$0 is the inconclusive rate, and $P_{\rm succ} =\sum_{i=1}^M p_i\,\Tr\!\bigl[\Pi_i\,\rho_i^{\otimes N}\bigr]$1 is the common confidence. This gives a logarithmic bound on how many parties can all retain confidence at least $P_{\rm succ} =\sum_{i=1}^M p_i\,\Tr\!\bigl[\Pi_i\,\rho_i^{\otimes N}\bigr]$2 (Lee et al., 2024).

A related sequential maximum-confidence scheme is built from weakened conclusive elements

$P_{\rm succ} =\sum_{i=1}^M p_i\,\Tr\!\bigl[\Pi_i\,\rho_i^{\otimes N}\bigr]$3

which preserve the confidence $P_{\rm succ} =\sum_{i=1}^M p_i\,\Tr\!\bigl[\Pi_i\,\rho_i^{\otimes N}\bigr]$4 but increase the inconclusive probability. The corresponding Kraus operators can be chosen as

$P_{\rm succ} =\sum_{i=1}^M p_i\,\Tr\!\bigl[\Pi_i\,\rho_i^{\otimes N}\bigr]$5

or, in the rank-one case,

$P_{\rm succ} =\sum_{i=1}^M p_i\,\Tr\!\bigl[\Pi_i\,\rho_i^{\otimes N}\bigr]$6

This formulation links the confidence to max-relative entropy through

$P_{\rm succ} =\sum_{i=1}^M p_i\,\Tr\!\bigl[\Pi_i\,\rho_i^{\otimes N}\bigr]$7

In the special two-state example with overlap $P_{\rm succ} =\sum_{i=1}^M p_i\,\Tr\!\bigl[\Pi_i\,\rho_i^{\otimes N}\bigr]$8, the optimum joint success is

$P_{\rm succ} =\sum_{i=1}^M p_i\,\Tr\!\bigl[\Pi_i\,\rho_i^{\otimes N}\bigr]$9

subject to equal sharing of the information gain. The same work states that, for linearly dependent examples such as trine, symmetric geometrically uniform, and mirror-symmetric ensembles, the post-measurement ensemble converges to a single fixed state under repeated sequential MC measurements (Wang et al., 17 Dec 2025).

5. Multi-copy discrimination, measurement hierarchies, and nonclassical traits

Distributed multi-copy discrimination asks how well one can identify $\{\Pi_i\}$0 when the $\{\Pi_i\}$1 copies are measured by multiple parties under different restrictions. The measurement classes considered include a single joint POVM on all copies (GLOBAL), LOCC with adaptive classical communication, classical bit strategies using commuting states diagonal in $\{\Pi_i\}$2 plus local projective measurements and classical post-processing, and generalized bit-like GPTs built from other two-dimensional convex state spaces such as regular polygons (Achenbach et al., 29 Apr 2026).

For two equiprobable pure qubit states $\{\Pi_i\}$3 with overlap

$\{\Pi_i\}$4

the optimal global discrimination on $\{\Pi_i\}$5 copies is

$\{\Pi_i\}$6

By contrast, classical bit discrimination obeys explicit bounds. For $\{\Pi_i\}$7 equiprobable bit states and $\{\Pi_i\}$8 copies,

$\{\Pi_i\}$9

with equality for $N$0 by choosing $N$1 classical probability vectors uniformly spaced in $N$2 and using majority-vote decoding. In particular,

$N$3

For two equiprobable pure qubit states with $N$4,

$N$5

as soon as $N$6 (Achenbach et al., 29 Apr 2026).

The trine ensemble gives a sharper comparison. For the cyclic geometrically uniform trine states $N$7, the pretty-good measurement is optimal on $N$8 copies and

$N$9

As $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$00, this rises exponentially to $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$01, whereas the classical bit benchmark $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$02 lags behind (Achenbach et al., 29 Apr 2026).

The same analysis shows that quantum theory is not maximal among all bit-like operational theories. In regular polygon GPTs, specifically the square and hexagon cases, a fixed local measurement on each copy plus one bit of communication can perfectly discriminate two copies of three vertices, which is impossible in qubit or bit theory. The double-trine ensemble $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$03 furnishes an explicit instance of nonlocality without entanglement: $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$04 so the gap is approximately $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$05 even though none of the input states is entangled. Two general theorems organize these phenomena. First, in any GPT of operational dimension $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$06, a fixed product measurement on $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$07 copies can attain $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$08 for $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$09 states iff $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$10. Second, perfect LOCC discrimination requires adaptive exclusion of the ensemble down to at most $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$11 perfectly distinguishable states. For qubit and bit theories this rules out perfect LOCC discrimination of nonorthogonal triples, whereas certain hexagon GPTs evade the obstruction (Achenbach et al., 29 Apr 2026).

6. Communication, cryptography, and experimental realization

Sequential discrimination has been developed not only as a measurement-theoretic problem but also as a communication primitive. In the qudit unambiguous scheme, the probability that at least one observer succeeds is $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$12, independently of the number of observers, and the channel capacity of the Alice-to-Bob link is $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$13. The same source states that higher-dimensional qudits lower a simple eavesdropper’s success probability relative to qubits, with

$$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$14

and $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$15 as $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$16 (Hillery et al., 2017).

A security analysis for quantum key distribution based on sequential discrimination introduces a unified model with Alice, Bob, and an eavesdropper Eve. Bob uses an unambiguous three-outcome POVM with strengths $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$17, Eve measures the post-Bob state, and the asymptotic key rate is taken to be the Csiszár–Körner quantity

$$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$18

The work reports a non-zero secret key rate despite eavesdropping. In the symmetric case $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$19, where $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$20, the optimized eavesdropping probability is

$$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$21

It further reports that the eavesdropping success probability is smaller for colored noise than for white noise at the same damping level, and that the maximum secret-key rate is higher under colored noise than under white noise (Namkung et al., 2023).

Experimental realization has been carried out with polarization single-photon states. Alice prepares

$$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$22

using a heralded single-photon source based on type-I SPDC in BiBO. Bob and Charlie implement sequential unambiguous discrimination with Sagnac-style interferometers; Bob performs a non-optimal USD that leaves post-measurement overlap $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$23, and Charlie performs the optimal USD for the resulting pair. The measured joint-success curve follows

$$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$24

in agreement with the underlying two-observer theory (Solís-Prosser et al., 2015).

7. Entanglement retention and structural limits

Sequential discrimination is often associated with irreversible depletion of the quantum resource being interrogated, but this is not universally the case. For two orthogonal entangled two-qubit pure states with equal priors, a one-way-LOCC protocol allows an arbitrary number of pairs of observers $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$25 to discriminate better than random guessing while ensuring that each post-measurement ensemble state retains finite entanglement (Saha et al., 6 Jun 2025).

In that protocol, the $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$26-th pair uses local unsharp POVMs with sharpness $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$27. For a special two-parameter family, the success probability at round $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$28 is

$$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$29

so every round satisfies $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$30. Entanglement is quantified by logarithmic negativity,

$$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$31

and for the same family

$$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$32

As long as $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$33 for every round, one has $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$34 and therefore $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$35. For any fixed finite sequence length $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$36, the parameters $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$37 can be chosen arbitrarily close to $$\rho_{i_1}^{(1)} \otimes \cdots \otimes \rho_{i_N}^{(N)}$$38, making the success probability arbitrarily close to unity while keeping the logarithmic negativity strictly positive (Saha et al., 6 Jun 2025).

This result complements the broader literature. Independent sequences may require no collective measurement at all; distributed multi-copy problems can exhibit strict GLOBAL–LOCC gaps and nonlocality without entanglement; sequential protocols can distribute minimum-error, unambiguous, or maximum-confidence information across many parties; and, in at least one orthogonal entangled setting, this distribution need not eliminate entanglement step by step. The combined picture is that “multi-party quantum sequence discrimination” is not a single theorem or protocol but a structured domain in which factorization, disturbance, locality, operational dimension, and resource preservation are all model-dependent constraints rather than universal laws.