Nonlocality Without Entanglement (NLWE)

• NLWE is a phenomenon where orthogonal product states, though separable, defy perfect local discrimination using LOCC.
• Canonical constructions like the domino and SHIFT ensembles demonstrate the gap between global measurements and LOCC protocols.
• Extensions include multipartite scenarios, causal order reinterpretations, and generalized theories that reveal deeper measurement and resource constraints.

Nonlocality without entanglement (NLWE) denotes the operational phenomenon in which a set of mutually orthogonal product states is perfectly distinguishable by a global measurement yet cannot be perfectly discriminated by local operations and classical communication (LOCC), despite each state being non-entangled and locally preparable. In part of the recent literature the same effect is termed quantum nonlocality without entanglement (QNLWE) (Kunjwal et al., 2022). The subject now spans the original orthogonal-state setting, multipartite strengthening via local irreducibility, causal-order reinterpretations, nonorthogonal discrimination, state exclusion, and certification paradigms that connect NLWE to Bell and network nonlocality (Zhen et al., 2022).

1. Operational definition and conceptual scope

In the standard formulation, NLWE concerns ensembles of orthogonal product states. Each state is separable, so entanglement is absent at the level of state preparation; nevertheless, the ensemble may fail to admit perfect LOCC discrimination. The nonlocal feature therefore lies not in Bell-inequality violation, but in the gap between global and locally implementable decoding of globally encoded classical information (Kunjwal et al., 2022).

This notion should be separated from Bell nonlocality. Bell nonlocality is a property of measurement statistics, while entanglement is a property of the quantum state itself; multipartite entanglement and multipartite Bell nonlocality are inequivalent resources for any number of parties (Augusiak et al., 2014). NLWE addresses a different inequivalence: product-state structure does not imply local measurability, and separable measurements need not be LOCC-implementable (Bhattacharya et al., 2019).

A recurring technical distinction is between LOCC and broader measurement classes. Several works in the area use LOCC-impossibility as the defining witness of NLWE, but also emphasize an operational gap between LOCC-restricted measurements and separable measurements. This distinction is central in both the original product-basis examples and later extensions to confidence-based, exclusion-based, and causal-order-based formulations (Kunjwal et al., 2022).

2. Canonical orthogonal constructions and proof strategies

The canonical examples are orthogonal product bases such as the two-qutrit “domino” basis and the tripartite SHIFT ensemble. In the device-independent treatment of the domino case, the relevant bipartite measurement has nine rank-one product projectors,

$$\begin{aligned} M'_{0,0|\lozenge} &= |1\rangle\langle 1|\otimes|1\rangle\langle 1|, \qquad M'_{0,1|\lozenge} = |0\rangle\langle 0|\otimes|+\rangle_{0,1}\langle +|, \ M'_{0,2|\lozenge} &= |0\rangle\langle 0|\otimes|-\rangle_{0,1}\langle -|, \qquad M'_{1,0|\lozenge} = |2\rangle\langle 2|\otimes|+\rangle_{1,2}\langle +|, \ M'_{1,1|\lozenge} &= |2\rangle\langle 2|\otimes|-\rangle_{1,2}\langle -|, \qquad M'_{1,2|\lozenge} = |+\rangle_{1,2}\langle +|\otimes|0\rangle\langle 0|, \ M'_{2,0|\lozenge} &= |-\rangle_{1,2}\langle -|\otimes|0\rangle\langle 0|, \qquad M'_{2,1|\lozenge} = |+\rangle_{0,1}\langle +|\otimes|2\rangle\langle 2|, \ M'_{2,2|\lozenge} &= |-\rangle_{0,1}\langle -|\otimes|2\rangle\langle 2| , \end{aligned}$$

with $|\pm\rangle_{j,k}=(|j\rangle\pm|k\rangle)/\sqrt2$. The eigenstates are all product states, yet the measurement is non-LOCC, making it a textbook NLWE example (Šupić et al., 2022).

The tripartite SHIFT ensemble is

$$\{ |000\rangle,\ |111\rangle,\ |+01\rangle,\ |-01\rangle,\ |1+0\rangle,\ |1-0\rangle,\ |01+\rangle,\ |01-\rangle \}.$$

These are eight mutually orthogonal three-qubit product states. They are globally distinguishable but not perfectly discriminable by LOCC under definite causal order. The obstruction is that whichever party starts, indistinguishable branches remain because the locally correct basis choice depends on information only another party can reveal later in the protocol (Kunjwal et al., 2022).

A general constructive paradigm uses a “stopper state” to force any orthogonality-preserving first local measurement to be trivial. For $(\mathbb C^d)^{\otimes n}$, one construction gives $n(d-1)+1$ orthogonal product states, consisting of $n(d-1)$ structured “edge” states plus

$$|S\rangle=\bigotimes_{k=1}^n \left(\sum_{i_k=0}^{d_k-1}|i_k\rangle\right).$$

The proof strategy is algebraic: if a local POVM element on subsystem $t$ has matrix $E_t=(m_{ij})$, Lemma 1 forces $m_{ab}=0$ for $|\pm\rangle_{j,k}=(|j\rangle\pm|k\rangle)/\sqrt2$0, while Lemma 2 forces $|\pm\rangle_{j,k}=(|j\rangle\pm|k\rangle)/\sqrt2$1. Hence $|\pm\rangle_{j,k}=(|j\rangle\pm|k\rangle)/\sqrt2$2, so no party can initiate a nontrivial orthogonality-preserving LOCC protocol. The same paper extends the construction to arbitrary multipartite dimensions $|\pm\rangle_{j,k}=(|j\rangle\pm|k\rangle)/\sqrt2$3, yielding $|\pm\rangle_{j,k}=(|j\rangle\pm|k\rangle)/\sqrt2$4 orthogonal product states that are not perfectly distinguishable by LOCC (Zhen et al., 2022).

These constructions sharpen the operational content of NLWE: the essential object is not an isolated state but a carefully engineered orthogonality pattern. This suggests that product-state nonlocality is best understood as a property of measurement geometry and protocol constraints rather than of entanglement resources.

3. Multipartite strengthening: local irreducibility, genuine nonlocality, and asymmetry

A stronger notion than local indistinguishability is local irreducibility. A set is locally irreducible if it is not possible to eliminate one or more states by orthogonality-preserving local measurements. Such a set is automatically locally indistinguishable, but the converse does not always hold. On this basis, strong nonlocality is defined for multipartite product-state sets that are locally irreducible in every bipartition (Halder et al., 2018).

Explicit orthogonal product bases with this property exist on $|\pm\rangle_{j,k}=(|j\rangle\pm|k\rangle)/\sqrt2$5 and $|\pm\rangle_{j,k}=(|j\rangle\pm|k\rangle)/\sqrt2$6. The $|\pm\rangle_{j,k}=(|j\rangle\pm|k\rangle)/\sqrt2$7 construction is the minimum dimension necessary for such product states to exist, because if any subsystem is a qubit then orthogonal product states are locally distinguishable in the corresponding qubit-versus-rest bipartition. The existence of these strongly nonlocal bases implies that local implementation of a multipartite separable measurement may require entangled resources across all bipartitions (Halder et al., 2018).

A weaker, but still genuinely multipartite, notion is local distinguishability based genuine nonlocality: a set is genuinely nonlocal if it is locally indistinguishable for every bipartition. Using an explicit bipartite locally indistinguishable set of $|\pm\rangle_{j,k}=(|j\rangle\pm|k\rangle)/\sqrt2$8 product states in $|\pm\rangle_{j,k}=(|j\rangle\pm|k\rangle)/\sqrt2$9, together with embedding and graph-connectivity arguments, one obtains genuinely nonlocal fully product-state sets in every multipartite system $$\{ |000\rangle,\ |111\rangle,\ |+01\rangle,\ |-01\rangle,\ |1+0\rangle,\ |1-0\rangle,\ |01+\rangle,\ |01-\rangle \}.$$0 with $$\{ |000\rangle,\ |111\rangle,\ |+01\rangle,\ |-01\rangle,\ |1+0\rangle,\ |1-0\rangle,\ |01+\rangle,\ |01-\rangle \}.$$1 and $$\{ |000\rangle,\ |111\rangle,\ |+01\rangle,\ |-01\rangle,\ |1+0\rangle,\ |1-0\rangle,\ |01+\rangle,\ |01-\rangle \}.$$2 (Li et al., 2020).

Tripartite constructions also exhibit party asymmetry. In $$\{ |000\rangle,\ |111\rangle,\ |+01\rangle,\ |-01\rangle,\ |1+0\rangle,\ |1-0\rangle,\ |01+\rangle,\ |01-\rangle \}.$$3, a minimal party-asymmetric genuinely nonlocal set is constructed, together with a local discrimination protocol that uses a three-qubit GHZ state as a resource. By contrast, a single copy of a two-qubit Bell state gives no advantage for that task. The same work also constructs an incomplete party-asymmetric strong nonlocal set, highlighting that multipartite NLWE can depend sharply on both bipartition structure and the type of entanglement assistance allowed (Bhunia et al., 2021).

4. Causal-order reinterpretations

One of the most significant recent reinterpretations is that some instances of NLWE depend not only on locality and classical communication, but also on definite causal order. The central claim is that LOCC is a conjunction of three constraints: local operations, classical communication, and definite causal order. In this perspective, the SHIFT ensemble is not merely “locally impossible”; it is impossible under a definite ordering of classical communication (Kunjwal et al., 2022).

The key noncausal resource is the deterministic Araújo–Feix / Baumeler–Wolf process, which maps local outputs $$\{ |000\rangle,\ |111\rangle,\ |+01\rangle,\ |-01\rangle,\ |1+0\rangle,\ |1-0\rangle,\ |01+\rangle,\ |01-\rangle \}.$$4 to local inputs $$\{ |000\rangle,\ |111\rangle,\ |+01\rangle,\ |-01\rangle,\ |1+0\rangle,\ |1-0\rangle,\ |01+\rangle,\ |01-\rangle \}.$$5 by

$$\{ |000\rangle,\ |111\rangle,\ |+01\rangle,\ |-01\rangle,\ |1+0\rangle,\ |1-0\rangle,\ |01+\rangle,\ |01-\rangle \}.$$6

Using these bits, each party conditionally applies a Hadamard and measures in the computational basis. For the SHIFT ensemble, this protocol perfectly identifies the state while using only local quantum operations and classical communication without definite causal order. Conversely, a measurement in the SHIFT basis can be used to implement the AF/BW channel. The resulting operational equivalence,

$$\{ |000\rangle,\ |111\rangle,\ |+01\rangle,\ |-01\rangle,\ |1+0\rangle,\ |1-0\rangle,\ |01+\rangle,\ |01-\rangle \}.$$7

recasts this instance of QNLWE as a manifestation of noncausality (Kunjwal et al., 2022).

This viewpoint extends beyond the tripartite Boolean case. For any Boolean $$\{ |000\rangle,\ |111\rangle,\ |+01\rangle,\ |-01\rangle,\ |1+0\rangle,\ |1-0\rangle,\ |01+\rangle,\ |01-\rangle \}.$$8-party classical process without global past, one can build an $$\{ |000\rangle,\ |111\rangle,\ |+01\rangle,\ |-01\rangle,\ |1+0\rangle,\ |1-0\rangle,\ |01+\rangle,\ |01-\rangle \}.$$9-qubit orthonormal product basis

$(\mathbb C^d)^{\otimes n}$0

where the basis choice at each site depends on the global bit string through the process function. These ensembles are orthogonal and non-entangled but not perfectly LOCC-discriminable under definite causal order (Kunjwal et al., 2022).

A broader equivalence is established for unambiguous complete product bases. If

$(\mathbb C^d)^{\otimes n}$1

is complete and unambiguous, local orthogonality satisfies

$(\mathbb C^d)^{\otimes n}$2

which induces a unique process function $(\mathbb C^d)^{\otimes n}$3. Logical consistency is captured by the unique fixed-point property

$(\mathbb C^d)^{\otimes n}$4

Within this framework, non-causality of the process function is exactly mirrored by QNLWE of the basis, while bipartite unambiguous QNLWE bases cannot exist because all bipartite process functions are causal (Dourdent et al., 29 Dec 2025). This suggests that, at least for the unambiguous complete-product subclass, NLWE can be read as a causal-structure obstruction.

5. Extensions beyond perfect orthogonal-state discrimination

NLWE is no longer confined to perfect discrimination of orthogonal product states. For nonorthogonal but linearly independent multipartite product states, the relevant task can be optimal unambiguous discrimination. In that setting the occurrence of NLWE may depend on the nonzero prior probabilities. A central criterion states that the globally optimal unambiguous measurement detects only $(\mathbb C^d)^{\otimes n}$5 iff

$(\mathbb C^d)^{\otimes n}$6

When only one reciprocal vector is a product vector, LOCC matches the global optimum exactly in that prior regime and fails outside it. This yields a three-way classification of product-state ensembles for optimal unambiguous discrimination: Type I, no NLWE for any nonzero priors; Type II, NLWE for all nonzero priors; and Type III, NLWE depending on the nonzero priors (Ha et al., 2021).

Postmeasurement information about the prepared subensemble can either suppress or reveal NLWE. In minimum-error discrimination, one two-qubit ensemble satisfies

$(\mathbb C^d)^{\otimes n}$7

so postmeasurement information annihilates NLWE; another ensemble satisfies

$(\mathbb C^d)^{\otimes n}$8

so the same resource creates NLWE (Ha et al., 2021). The analogous lock/unlock phenomenon also occurs for optimal unambiguous discrimination (Ha et al., 2021).

State exclusion provides another operational generalization. In this setting one asks whether a measurement can exclude one or more candidate states rather than identify the prepared state. Three bipartite product states can be globally antidistinguishable while failing to be LOCC antidistinguishable, proving that three is the minimal number of states required for NLWE in exclusion. Ordinary LOCC antidistinguishability of multipartite product states is symmetric with respect to the initiating party, but this symmetry fails for higher-order $(\mathbb C^d)^{\otimes n}$9-antidistinguishability; the same work also exhibits a tripartite product-state example that is globally antidistinguishable but not LOCC antidistinguishable across any bipartition, yielding genuine multipartite NLWE in the exclusion framework (Manna et al., 17 Feb 2026).

6. Certification, generalized theories, and structural limits

NLWE can be certified rather than merely postulated from a trusted Hilbert-space description. In a bilocal entanglement-swapping network with two maximally entangled qutrit sources, exact reproduction of suitable correlations self-tests Bob’s two-qutrit domino measurement up to local isometries. The central node’s measurement is thereby certified as a measurement with a product eigenbasis that is nevertheless non-LOCC, establishing a device-independent bridge between NLWE and network nonlocality (Šupić et al., 2022).

A semi-device-independent route is available through maximum-confidence discrimination. For separable-state ensembles $n(d-1)+1$0, the confidence is

$n(d-1)+1$1

For the two-qubit antiparallel SIC ensemble, the noiseless case yields

$n(d-1)+1$2

so global measurements outperform separable ones. Certification can be based on observed outcome rates $n(d-1)+1$3, and for this ensemble NLWE is certifiable only when $n(d-1)+1$4 (Lee et al., 11 Jun 2026).

The phenomenon is not specific to quantum theory. In generalized probabilistic theories, product states may also be globally distinguishable but locally indistinguishable, asymmetric local discrimination may occur, and separable but locally unimplementable measurements may exist. A comparison framework based on the local-global discrimination gap $n(d-1)+1$5 at fixed signaling dimension shows that polygon theories can exhibit stronger NLWE than quantum theory; for the analyzed three-party tasks,

$n(d-1)+1$6

(Bhattacharya et al., 2019).

Several works also connect NLWE to Bell-type notions more directly. Under controlled-NOT, a full product basis creates entangled states if and only if the full basis or some subspace becomes irreducible in LOCC discrimination, and the associated “nonlocal entropy” quantifies this entanglement potential of product ensembles (Mal et al., 2020). At the same time, recent structural results place sharp limits on hidden-NLWE activation: a complete orthogonal product basis that is initially LOCC-distinguishable remains so under all orthogonality-preserving local projective measurements, implying that incompleteness is necessary for activation of NLWE by this mechanism. In multipartite language, such complete distinguishable bases are strongly local, meaning non-activable under every bipartition (Bhunia et al., 28 May 2026).

Taken together, these developments show that NLWE is not a single anomaly tied to one product basis. It is a broad operational theme concerning the mismatch between separability of states and locality of measurements, with variants controlled by orthogonality structure, prior information, postmeasurement side information, causal order, exclusion criteria, and the surrounding physical theory itself.