Three-Qubit Nonlocality Paradoxes

• The topic defines three-qubit paradoxes as logical contradictions where quantum correlations violate all possible local-hidden-variable assignments.
• It reviews formulations including the GHZ all-versus-nothing, Hardy, and Cabello approaches, emphasizing measurement protocols and success probabilities.
• A combinatorial and graph-theoretic classification framework is introduced to systematically organize paradox families and guide experimental implementations.

Three-qubit nonlocality paradoxes are logical contradictions between the possibilistic structure of quantum correlations for three spatially separated qubits and any local-hidden-variable description based on preassigned outcomes. In the modern sheaf-theoretic formulation, a three-qubit scenario $(|\psi\rangle,\mathcal M)$ is paradoxical when its empirical model is strongly nonlocal: no global assignment of binary outcomes to all local measurements remains compatible with the support of the Born-rule distributions across all contexts (Silva et al., 20 Aug 2025). This notion includes the Greenberger-Horne-Zeilinger (GHZ) all-versus-nothing contradiction, Hardy and Cabello forms of nonlocality without inequalities, projected-coloring-graph and quantum pigeonhole variants, and recent structural classification results showing that the space of three-qubit paradoxes is much broader than the standard GHZ construction alone (Silva et al., 1 Jul 2026).

1. Formal framework and state-measurement structure

A three-qubit measurement scenario is a Bell scenario $M=(M_1,M_2,M_3)$, where each $M_i$ is a finite set of local measurements on qubit $i$, outcomes are binary $\mathcal O=\{0,1\}$, and contexts are triples $C=(A,B,C)\in M_1\times M_2\times M_3$. For a state $|\psi\rangle$, the quantum model assigns to each context the Born-rule distribution

$$P_C(o_a,o_b,o_c) = \bigl|\langle A\!\to\!o_a\otimes B\!\to\!o_b\otimes C\!\to\!o_c\mid\psi\rangle\bigr|^2.$$

An event is impossible precisely when its amplitude vanishes: $$\langle A\!\to\!o_a,\;B\!\to\!o_b,\;C\!\to\!o_c\mid\psi\rangle=0.$$ Strong nonlocality means that every global assignment $g$ fails on at least one context: $M=(M_1,M_2,M_3)$0 Equivalently, there is no assignment $M=(M_1,M_2,M_3)$1 consistent with the support of the empirical model (Silva et al., 20 Aug 2025).

For three qubits, the structural state constraint is already restrictive. A key theorem specialized to this setting states that any three-qubit state admitting a paradox lies in the SLOCC class of GHZ and, up to local unitaries, has a balanced form; moreover, equatorial measurements

$M=(M_1,M_2,M_3)$2

suffice to witness the paradox (Silva et al., 20 Aug 2025). This places three-qubit paradoxes in a narrow but highly structured sector of multipartite nonlocality: they are not arbitrary support obstructions, but arise from GHZ-type entanglement together with equatorial measurement geometry.

2. GHZ all-versus-nothing contradictions

The conventional GHZ paradox is the canonical three-qubit all-versus-nothing proof. In one convenient phase convention, the state is

$M=(M_1,M_2,M_3)$3

and the relevant local observables are the Pauli operators $M=(M_1,M_2,M_3)$4 and $M=(M_1,M_2,M_3)$5 on qubit $M=(M_1,M_2,M_3)$6. Direct action on the state gives the operator constraints

$M=(M_1,M_2,M_3)$7

$M=(M_1,M_2,M_3)$8

If a local-hidden-variable model assigns predetermined values $M=(M_1,M_2,M_3)$9 to each $M_i$0, then multiplying the last three equations yields $M_i$1, while the first requires $M_i$2. No assignment satisfies all four simultaneously, producing the traditional GHZ contradiction (Chakraborty et al., 2024).

In the bit-valued formulation used in later classification work, the GHZ paradox becomes a system of parity constraints. With outcomes $M_i$3, the four contexts

$M_i$4

impose

$M_i$5

for $M_i$6, and

$M_i$7

for each of $M_i$8, $M_i$9, and $i$0. Summing over $i$1 gives $i$2, which is the parity form of the contradiction (Silva et al., 20 Aug 2025).

Later work identified a more refined conditional structure inside GHZ-type paradoxes. For interpolant states, conditioning on the third qubit’s measurement and outcome yields Alice-Bob XOR systems of the form

$i$3

and paradoxicality is equivalent to inconsistency of the union of these conditioned systems for every total assignment of the third party’s outcomes (Silva et al., 20 Aug 2025). This conditional viewpoint became the basis for the broader classification of non-GHZ three-qubit paradoxes.

3. Hardy and generalized Hardy paradoxes

Hardy’s paradox is a nonlocality-without-inequalities argument based on zero-probability constraints together with one strictly positive event. In the generalized multipartite framework, the three-qubit case uses the GHZ state

$i$4

with two binary-outcome measurements per party, $i$5 and $i$6. Measurement $i$7 projects onto the $i$8 eigenstate of $i$9,

$\mathcal O=\{0,1\}$0

while $\mathcal O=\{0,1\}$1 projects onto the $\mathcal O=\{0,1\}$2 eigenstate of $\mathcal O=\{0,1\}$3,

$\mathcal O=\{0,1\}$4

Writing $\mathcal O=\{0,1\}$5 and $\mathcal O=\{0,1\}$6, the three-qubit generalized Hardy paradox in the $\mathcal O=\{0,1\}$7 scenario is

$\mathcal O=\{0,1\}$8

In a deterministic local-hidden-variable model, each $\mathcal O=\{0,1\}$9 is preassigned. The six zero-probability constraints force $C=(A,B,C)\in M_1\times M_2\times M_3$0, whereas quantum mechanics yields a nonzero value, giving a direct all-versus-nothing contradiction (Jiang et al., 2017).

For the specified measurements on the GHZ state,

$C=(A,B,C)\in M_1\times M_2\times M_3$1

Thus the paradox succeeds with probability $C=(A,B,C)\in M_1\times M_2\times M_3$2. This is higher than the standard three-qubit Hardy $C=(A,B,C)\in M_1\times M_2\times M_3$3 scenario, whose success probability is $C=(A,B,C)\in M_1\times M_2\times M_3$4; the generalized construction therefore doubles the three-qubit success rate and, according to the paper, is stronger in showing the conflict of quantum mechanics with local realism (Jiang et al., 2017).

The same seven probabilities define the Bell-type inequality

$C=(A,B,C)\in M_1\times M_2\times M_3$5

Every local-hidden-variable model satisfies $C=(A,B,C)\in M_1\times M_2\times M_3$6, while quantum mechanics gives $C=(A,B,C)\in M_1\times M_2\times M_3$7. For the proposed photonic implementation, if each of the six nominally zero probabilities is at most $C=(A,B,C)\in M_1\times M_2\times M_3$8, violation persists whenever

$C=(A,B,C)\in M_1\times M_2\times M_3$9

that is, for $|\psi\rangle$0 (Jiang et al., 2017).

More generally, the generalized Hardy framework covers Hardy’s original construction and Cereceda’s extension as special cases, and for any $|\psi\rangle$1 there exist generalized paradoxes with success probability as high as $|\psi\rangle$2. The three-qubit case is the first nontrivial instance of this strengthened multipartite family (Jiang et al., 2017).

4. Cabello, quantum pigeonhole, and other Hardy-like variants

Cabello’s nonlocality argument (CNA) is another three-qubit nonlocality-without-inequalities scheme. It uses the generalized GHZ state

$|\psi\rangle$3

equivalently

$|\psi\rangle$4

Each qubit $|\psi\rangle$5 has two dichotomic observables, $|\psi\rangle$6 and $|\psi\rangle$7, with outcomes $|\psi\rangle$8. The Cabello conditions are: $|\psi\rangle$9

$$P_C(o_a,o_b,o_c) = \bigl|\langle A\!\to\!o_a\otimes B\!\to\!o_b\otimes C\!\to\!o_c\mid\psi\rangle\bigr|^2.$$0

$$P_C(o_a,o_b,o_c) = \bigl|\langle A\!\to\!o_a\otimes B\!\to\!o_b\otimes C\!\to\!o_c\mid\psi\rangle\bigr|^2.$$1

Local realism forces $$P_C(o_a,o_b,o_c) = \bigl|\langle A\!\to\!o_a\otimes B\!\to\!o_b\otimes C\!\to\!o_c\mid\psi\rangle\bigr|^2.$$2, so $$P_C(o_a,o_b,o_c) = \bigl|\langle A\!\to\!o_a\otimes B\!\to\!o_b\otimes C\!\to\!o_c\mid\psi\rangle\bigr|^2.$$3 yields a contradiction. The success quantity is

$$P_C(o_a,o_b,o_c) = \bigl|\langle A\!\to\!o_a\otimes B\!\to\!o_b\otimes C\!\to\!o_c\mid\psi\rangle\bigr|^2.$$4

In the special case $$P_C(o_a,o_b,o_c) = \bigl|\langle A\!\to\!o_a\otimes B\!\to\!o_b\otimes C\!\to\!o_c\mid\psi\rangle\bigr|^2.$$5, one recovers Hardy’s argument (Cereceda, 2016).

For generalized three-qubit GHZ states, CNA runs for almost all entangled states, and its quantum maximum occurs at the maximally entangled point $$P_C(o_a,o_b,o_c) = \bigl|\langle A\!\to\!o_a\otimes B\!\to\!o_b\otimes C\!\to\!o_c\mid\psi\rangle\bigr|^2.$$6. Numerically, the optimum is attained at

$$P_C(o_a,o_b,o_c) = \bigl|\langle A\!\to\!o_a\otimes B\!\to\!o_b\otimes C\!\to\!o_c\mid\psi\rangle\bigr|^2.$$7

yielding

$$P_C(o_a,o_b,o_c) = \bigl|\langle A\!\to\!o_a\otimes B\!\to\!o_b\otimes C\!\to\!o_c\mid\psi\rangle\bigr|^2.$$8

and hence

$$P_C(o_a,o_b,o_c) = \bigl|\langle A\!\to\!o_a\otimes B\!\to\!o_b\otimes C\!\to\!o_c\mid\psi\rangle\bigr|^2.$$9

By comparison, the standard three-qubit Hardy argument has

$$\langle A\!\to\!o_a,\;B\!\to\!o_b,\;C\!\to\!o_c\mid\psi\rangle=0.$$0

Within generalized no-signaling theory, both HNA and CNA can reach $$\langle A\!\to\!o_a,\;B\!\to\!o_b,\;C\!\to\!o_c\mid\psi\rangle=0.$$1 (Cereceda, 2016).

A distinct Hardy-like branch is provided by the quantum pigeonhole paradoxes derived from projected-coloring graphs (PCGs). The two minimal un-colorable three-vertex PCGs correspond to the states

$$\langle A\!\to\!o_a,\;B\!\to\!o_b,\;C\!\to\!o_c\mid\psi\rangle=0.$$2

and

$$\langle A\!\to\!o_a,\;B\!\to\!o_b,\;C\!\to\!o_c\mid\psi\rangle=0.$$3

The relevant observables are single-qubit $$\langle A\!\to\!o_a,\;B\!\to\!o_b,\;C\!\to\!o_c\mid\psi\rangle=0.$$4 and pairwise products $$\langle A\!\to\!o_a,\;B\!\to\!o_b,\;C\!\to\!o_c\mid\psi\rangle=0.$$5. For $$\langle A\!\to\!o_a,\;B\!\to\!o_b,\;C\!\to\!o_c\mid\psi\rangle=0.$$6, quantum theory predicts

$$\langle A\!\to\!o_a,\;B\!\to\!o_b,\;C\!\to\!o_c\mid\psi\rangle=0.$$7

together with

$$\langle A\!\to\!o_a,\;B\!\to\!o_b,\;C\!\to\!o_c\mid\psi\rangle=0.$$8

Classically, if $$\langle A\!\to\!o_a,\;B\!\to\!o_b,\;C\!\to\!o_c\mid\psi\rangle=0.$$9 in a deterministic model, then all three pairwise products $g$0 must be $g$1, which contradicts the classical pigeonhole rule for three objects in two boxes. The same logical pattern applies to $g$2, except that two conditional constraints have $g$3 instead of $g$4. In both cases the success probability is $g$5 (Ru et al., 2022).

These Hardy-like constructions show that three-qubit paradoxes need not be exhausted by the GHZ parity contradiction or by the standard Hardy support pattern. They can also be organized by graph-theoretic uncolorability, conditional pairwise parities, and post-selection structure (Ru et al., 2022).

5. Classification beyond the conventional GHZ construction

A major recent development is the structural classification of three-qubit paradoxes. The central notion is the biconditional parity proof (BPP). In this framework, Charlie has exactly two measurements $g$6, and conditioning on Charlie’s choice and outcome $g$7, the Alice-Bob possibilistic structure reduces to a single XOR equation: $g$8 for a parity function $g$9 defined on a subset $M=(M_1,M_2,M_3)$00. If $M=(M_1,M_2,M_3)$01, all four Alice-Bob outcomes are possible; if $M=(M_1,M_2,M_3)$02, exactly the opposite parity class is impossible. The paradox arises when, for each $M=(M_1,M_2,M_3)$03, the pair of conditioned XOR systems has no global solution $M=(M_1,M_2,M_3)$04 (Silva et al., 1 Jul 2026).

The classification machinery is combinatorial and logical. Zero-probability events are encoded as clauses in a 2-CNF formula $M=(M_1,M_2,M_3)$05, and the resulting implication graph $M=(M_1,M_2,M_3)$06 is unsatisfiable precisely when a strongly connected component contains a literal and its complement. For interpolant states, impossibility pairs come in complementary pairs, which makes $M=(M_1,M_2,M_3)$07 bidirected; then unsatisfiability reduces to the existence of a witness path from a literal to its complement. From such paths one extracts a finite “Charlie clock”

$M=(M_1,M_2,M_3)$08

together with a canonical Alice-Bob completion $M=(M_1,M_2,M_3)$09 and a tuple of layer shifts $M=(M_1,M_2,M_3)$10. The main theorem states that every minimal three-qubit paradox admitting a biconditional parity proof is uniquely classified—up to local unitary and permutation equivalence—by the finite datum $M=(M_1,M_2,M_3)$11 (Silva et al., 1 Jul 2026).

This framework encompasses all earlier-known examples and also produces genuinely new families. The examples include the Abramsky-Barbosa family with $M=(M_1,M_2,M_3)$12 equatorial measurements on Alice and Bob and two measurements on Charlie, new interpolant families with three Charlie measurements in symmetric $M=(M_1,M_2,M_3)$13 and asymmetric $M=(M_1,M_2,M_3)$14 forms, and exotic infinite families with four Charlie measurements such as $M=(M_1,M_2,M_3)$15, $M=(M_1,M_2,M_3)$16, and $M=(M_1,M_2,M_3)$17 scenarios. These violate regularity conditions assumed in earlier constructions, including: use of an interpolant state, restriction to two Charlie measurements, equality of Alice and Bob’s numbers of measurements, maximal logico-combinatorial strength, and contradiction by summing all parity constraints (Silva et al., 1 Jul 2026).

A complementary 2025 classification program established a partial classification under explicit N-regularity conditions. In that setting, $M=(M_1,M_2,M_3)$18, $M=(M_1,M_2,M_3)$19, the paradox is maximally impossible, and each of the four conditioned systems has coefficient-matrix rank $M=(M_1,M_2,M_3)$20. Every $M=(M_1,M_2,M_3)$21-regular paradox is then equivalent to exactly one of four infinite families, including the GHZ case $M=(M_1,M_2,M_3)$22, the Abramsky et al. family, and further families with shifted or asymmetric equatorial measurement sets. The same work exhibited a highly exotic non-$M=(M_1,M_2,M_3)$23-regular paradox and conjectured that all three-qubit paradoxes arise from a one-parameter interpolant family $M=(M_1,M_2,M_3)$24; the conjecture is supported by structural constraints on impossible events but was not proved there (Silva et al., 20 Aug 2025).

Taken together, these results shift the subject from isolated constructions to a genuine classification theory. The standard GHZ paradox is now understood as one corner of a larger combinatorial landscape of conditioned parity obstructions, witness paths, clocks, cosets, and layer structures.

6. Experiments, operational variants, and robustness

Three-qubit paradoxes have been treated not only as logical no-go results but also as experimentally implementable and operationally useful correlation structures. For the generalized Hardy paradox, the proposed three-photon implementation uses a polarization-entangled GHZ source generated by cascaded spontaneous parametric down-conversion or a single-crystal interferometric scheme. Measurement in the $M=(M_1,M_2,M_3)$25 basis is realized by a half-wave plate at $M=(M_1,M_2,M_3)$26 followed by a polarizing beam splitter, while the $M=(M_1,M_2,M_3)$27 basis uses a quarter-wave plate at $M=(M_1,M_2,M_3)$28 together with a half-wave plate. The six forbidden joint settings should vanish within error, and the $M=(M_1,M_2,M_3)$29 outcome should appear with probability about $M=(M_1,M_2,M_3)$30; the white-noise tolerance condition is $M=(M_1,M_2,M_3)$31 (Jiang et al., 2017).

The projected-coloring-graph paradoxes have already been implemented optically. The experiment employed 390 nm, 140 fs pulses at 76 MHz, two beam-like compound BBO crystals in sandwich geometry to generate two independent $M=(M_1,M_2,M_3)$32 pairs, spatial and temporal compensation giving Bell-pair visibility $M=(M_1,M_2,M_3)$33, and Hong-Ou-Mandel visibility $M=(M_1,M_2,M_3)$34 after overlap on a PBS. By wave-plate rotations, the raw GHZ state was converted into $M=(M_1,M_2,M_3)$35 or $M=(M_1,M_2,M_3)$36, with a single QWP change from $M=(M_1,M_2,M_3)$37 to $M=(M_1,M_2,M_3)$38 toggling between them. For $M=(M_1,M_2,M_3)$39, the reported values were $M=(M_1,M_2,M_3)$40, $M=(M_1,M_2,M_3)$41, $M=(M_1,M_2,M_3)$42, and $M=(M_1,M_2,M_3)$43. For $M=(M_1,M_2,M_3)$44, the corresponding values were $M=(M_1,M_2,M_3)$45, $M=(M_1,M_2,M_3)$46, $M=(M_1,M_2,M_3)$47, and $M=(M_1,M_2,M_3)$48 (Ru et al., 2022).

A common misconception is that, because the GHZ paradox is already an all-versus-nothing contradiction, no stronger three-qubit form can exist. A recent answer is the randomized GHZ game $M=(M_1,M_2,M_3)$49, where the promise condition is itself randomly chosen and revealed only to one party. The shared resource is still a GHZ state, the local measurements remain $M=(M_1,M_2,M_3)$50-type, and the quantum strategy wins with probability $M=(M_1,M_2,M_3)$51. However, any deterministic local-hidden-variable strategy wins at most $M=(M_1,M_2,M_3)$52. When lifted to the communication-complexity task $M=(M_1,M_2,M_3)$53, the GHZ-assisted protocol computes the target exactly with one bit from Alice and one from Bob, whereas any classical protocol with perfect success requires at least three bits of communication. In that operational sense, the randomized variant is stronger than the conventional GHZ game, even though both have perfect quantum success (Chakraborty et al., 2024).

Robustness against white noise is usually quantified by Bell inequalities rather than possibilistic paradoxes, but the connection is direct for GHZ correlations. For the noisy state

$M=(M_1,M_2,M_3)$54

the $M=(M_1,M_2,M_3)$55 Mermin inequality gives the familiar threshold $M=(M_1,M_2,M_3)$56. Systematic facet searches with more settings lowered this threshold below $M=(M_1,M_2,M_3)$57: the tight $M=(M_1,M_2,M_3)$58 inequality $M=(M_1,M_2,M_3)$59 gives $M=(M_1,M_2,M_3)$60, while the most economical nontrivial tight facet with $M=(M_1,M_2,M_3)$61 settings gives $M=(M_1,M_2,M_3)$62. These results show that noisy three-qubit GHZ nonlocality persists slightly below the Mermin threshold once additional settings are allowed (Vertesi et al., 2011).

Three-qubit nonlocality paradoxes therefore occupy a dual role. They are, at one level, sharp logical obstructions to global assignments and local realism; at another, they are structured resources for nonlocal games, communication tasks, measurement-based quantum computation, randomness certification, device-independent key distribution, and self-testing. Recent work has made both roles more precise by extending the family of paradoxes beyond GHZ, classifying broad subclasses, and tying their logical form to operational advantage (Silva et al., 1 Jul 2026).