GHZ States and Genuine Multipartite Entanglement

The founding question of multipartite entanglement is deceptively simple: when three or more quantum systems are correlated, is that correlation just a web of pairwise bonds, or something irreducibly collective — a whole that lives nowhere in its parts? The Greenberger–Horne–Zeilinger state is the sharpest answer that collective entanglement is real. Its canonical three-qubit form is $|\mathrm{GHZ}\rangle=\tfrac{1}{\sqrt2}(|000\rangle+|111\rangle)$ , an equal coherent superposition of two globally distinct configurations — a microscopic Schrödinger cat with exactly two branches, all-zeros and all-ones. This one state does three deep things at once, and this article is their synthesis. First, it is genuinely tripartite: every single qubit is maximally mixed, hence maximally entangled with the other two, yet every pair alone is separable with zero entanglement — the entanglement lives only in the whole and dies when one qubit is discarded. Second, it refutes local realism with a single deterministic self-contradiction — Mermin's "all-versus-nothing" argument, where local hidden variables predict the opposite sign of a product that quantum mechanics fixes with certainty. Third, it anchors the modern classification of entanglement: under stochastic local operations, three-qubit pure states split into exactly two genuine classes, GHZ and W, distinguished by the SLOCC-invariant three-tangle, maximal for GHZ and zero for W. The GHZ state is where entanglement stops being a two-body idea and becomes a theory of the many.

Key takeaways

The GHZ state $|\mathrm{GHZ}\rangle=\tfrac{1}{\sqrt2}(|000\rangle+|111\rangle)$ is the canonical genuinely tripartite state: no bipartition is separable, and its correlations cannot be reduced to bipartite ones.
It is fragile: tracing out any one qubit leaves the separable mixture $\tfrac12(|00\rangle\langle00|+|11\rangle\langle11|)$ ( $C=0$ ), while each single qubit alone is $I/2$ — maximal entanglement with the rest, zero within any pair.
It gives an all-versus-nothing refutation of local hidden variables: a single run of correlated Pauli measurements forces $+1\ne-1$ , needing no inequality, so the Mermin operator reaches its algebraic maximum $\langle\mathcal M\rangle=4$ against the classical bound $2$ .
Under SLOCC, three-qubit genuine entanglement splits into exactly two classes, GHZ and W; the three-tangle is $\tau_{ABC}=1$ for GHZ and $0$ for W, and they cannot be interconverted even with vanishing probability.
Key idea: A GHZ state is the maximal coherent superposition of two orthogonal global classical configurations; its single-party reductions are maximally mixed while its two-party reductions are separable, so its entanglement lives only in the full state and dies with the loss of one particle.

The core insight

A GHZ state is the maximal coherent superposition of two orthogonal global classical configurations: its single-party reductions are maximally mixed (maximal entanglement with the rest) while its two-party reductions are separable — genuine multipartite entanglement that lives only in the full state and dies with the loss of any one particle.

Where this lives today

GHZ states have graduated from foundations to engineering. Superconducting, trapped-ion and neutral-atom processors now certify genuine multipartite entanglement across dozens of qubits, and photonic experiments push multiphoton GHZ states past ten — all using the fidelity-above-one-half witness of equation (6) as the acceptance test. In metrology an $N$ -qubit GHZ state is the ideal Heisenberg-limited probe, its phase sensitivity scaling as $1/N$ rather than the standard-quantum-limit $1/\sqrt N$ — tempered by fragility, since the entanglement that sharpens the estimate also makes it decohere $N$ times faster. In networking, GHZ states are the native resource for conference key agreement, anonymous transmission, secret sharing, and clock synchronization, and the glue of measurement-based computation and repeaters. And at the frontier, loophole-free GHZ-type tests remain the benchmark for device-independent certification of multipartite nonlocality — the deepest statement that nature is not locally real.

The physical picture

Imagine a single quantum coin all three qubits read at once. In one branch every qubit reports $0$ ; in the other, $1$ . The GHZ state is the equal, phase-coherent superposition of these two whole-system branches. No qubit can tell locally which branch it is in — its own record is a fair coin — because the "which-branch" information is written collectively across all three, nowhere on any one. This is what genuine tripartite entanglement is: a correlation with no local shadow. Now discard a qubit. Trace out $C$ and the surviving pair collapses to $\tfrac12(|00\rangle\langle00|+|11\rangle\langle11|)$ — two bits that always agree, but with no coherence: a classical correlation reproducible with a shared random bit. Trace out two qubits and the last is $I/2$ , maximally mixed. The coherence that makes GHZ a quantum resource lives only in the three-body amplitude and evaporates the instant one particle is lost. This is the fragility the first figure makes concrete, with the invariants attached.

Figure

The GHZ state is the equal superposition of the all-zero and all-one configurations with amplitude 1/√2. Discard any single qubit and the surviving pair is an equal classical mixture of 00 and 11 with concurrence C = 0; each qubit alone is I/2 with entropy S = 1 bit — every qubit maximally entangled with the rest yet no pair entangled, and three-tangle τ = 1

The second thing the GHZ state does is refute local realism with a single stroke. Measure the four commuting observables $XXX$ , $XYY$ , $YXY$ , $YYX$ ; on the GHZ state each has a definite value, $+1,-1,-1,-1$ . A local realist must assign every qubit a predetermined outcome $m_x,m_y=\pm1$ for each Pauli, fixed before any distant choice. Multiply the four predictions: because each $Y$ appears twice, every hidden value is squared and the local-realist product is forced to $+1$ — while quantum mechanics forces the same product to $-1$ . No inequality, no average: the disagreement is a bare sign on a single run. The second figure lays this out, with the Mermin value $\langle\mathcal M\rangle=4$ .

Figure

The GHZ all-versus-nothing paradox. On the GHZ state the four joint Pauli observables take definite values XXX = +1, XYY = −1, YXY = −1, YYX = −1. Multiplying the four, every hidden value m = ±1 appears squared, so local realism forces the product to +1 while quantum mechanics forces −1. The same relations give the Mermin operator its algebraic maximum ⟨M⟩ = 4 versus the local bound 2

The third thing the GHZ state does is stake out one pole of tripartite entanglement. Its opposite is the W state, $\tfrac{1}{\sqrt3}(|001\rangle+|010\rangle+|100\rangle)$ , also genuinely tripartite but inverse in behaviour: lose one qubit from W and the remaining pair stays entangled. GHZ concentrates its entanglement in an irreducible three-body resource that dies with one loss; W distributes robust pairwise entanglement that survives. No local protocol converts one into the other, even with vanishing probability.

The core equations

The starting point is the state itself, generalized to $N$ parties: the equal superposition of the two extreme basis strings.

|\mathrm{GHZ}_N\rangle=\frac{1}{\sqrt2}\big(|0\rangle^{\otimes N}+|1\rangle^{\otimes N}\big) \qquad (1)

where: $N$ — number of qubits (canonically $N=3$ ); $|0\rangle,|1\rangle$ — computational-basis states of a qubit; $\otimes N$ — $N$ -fold tensor product. Reading: support on just two of the $2^N$ strings, but the coherence between them is what makes it entangled, not a classical mixture. The single-party marginal is the first diagnostic. Trace out all but one qubit and you get the maximally mixed state — the sign that a qubit knows nothing alone, hence is maximally correlated with the others.

\rho_A=\mathrm{Tr}_{BC}\,|\mathrm{GHZ}\rangle\langle\mathrm{GHZ}|=\frac{I}{2},\qquad S(\rho_A)=1\ \text{bit} \qquad (2)

where: $\mathrm{Tr}_{BC}$ — partial trace over $B,C$ ; $\rho_A$ — reduced state of qubit $A$ ; $S(\rho)=-\mathrm{Tr}(\rho\log_2\rho)$ — von Neumann entropy, the entanglement entropy across the cut. Reading: $S=1$ bit is maximal, so the $A|BC$ cut carries one full ebit. The two-party marginal is the surprise. Trace out one qubit and the remaining pair is separable — a classical mixture of product states, the fragility that sets GHZ apart.

\rho_{AB}=\mathrm{Tr}_C\,|\mathrm{GHZ}\rangle\langle\mathrm{GHZ}|=\tfrac12\big(|00\rangle\langle00|+|11\rangle\langle11|\big),\qquad C(\rho_{AB})=0 \qquad (3)

where: $\rho_{AB}$ — reduced state of $A,B$ after discarding $C$ ; $|00\rangle,|11\rangle$ — correlated product states with no coherence; $C$ — Wootters concurrence. Reading: two bits that always agree but share no coherence — reproducible with a shared random bit, hence concurrence zero. The Mermin operator packages the all-versus-nothing argument into one Hermitian observable separating quantum mechanics from local realism.

\mathcal{M}=XXX-XYY-YXY-YYX \qquad (4)

where: $X,Y$ — Pauli operators $\sigma_x,\sigma_y$ on the indicated qubit; $XYY=\sigma_x\otimes\sigma_y\otimes\sigma_y$ ; $\mathcal M$ — Mermin operator, local bound $|\langle\mathcal M\rangle|\le2$ , algebraic maximum $4$ . Reading: each term is a signed stabilizer of $|\mathrm{GHZ}\rangle$ , so all four saturate at once. The three-tangle quantifies how much of the entanglement is irreducibly tripartite — the residual of the Coffman–Kundu–Wootters monogamy relation: one qubit's entanglement with the rest, minus its pairwise bonds.

\tau_{ABC}=C_{A(BC)}^2-C_{AB}^2-C_{AC}^2 \qquad (5)

where: $\tau_{ABC}$ — three-tangle, an SLOCC invariant of genuine tripartite entanglement; $C_{A(BC)}$ — concurrence of $A$ against $BC$ ; $C_{AB},C_{AC}$ — pairwise concurrences. Reading: for GHZ, $C^2_{A(BC)}=1$ and $C_{AB}=C_{AC}=0$ , so $\tau=1$ — all entanglement tripartite, none pairwise. Finally, the two genuine classes are SLOCC-inequivalent, and one operator certifies GHZ-type entanglement.

\tau_{ABC}(\mathrm{GHZ})=1\ne0=\tau_{ABC}(W),\qquad \mathcal{W}_{\mathrm{GHZ}}=\tfrac12 I-|\mathrm{GHZ}\rangle\langle\mathrm{GHZ}| \qquad (6)

where: $W=\tfrac{1}{\sqrt3}(|001\rangle+|010\rangle+|100\rangle)$ — the other genuine class; $\mathcal W_{\mathrm{GHZ}}$ — genuine-multipartite witness; $\tfrac12$ — the maximal biseparable overlap $\max_{|\phi\rangle}|\langle\phi|\mathrm{GHZ}\rangle|^2$ . Reading: $\mathrm{Tr}(\mathcal W_{\mathrm{GHZ}}\rho)<0$ , i.e. fidelity $F>\tfrac12$ , certifies genuine tripartite entanglement.

Derivation AThe GHZ marginals — maximal mixing above, classical correlation below

The goal is to compute the one- and two-qubit reduced states of the GHZ state directly from the partial trace, and read off the two facts that define genuine multipartite entanglement: every single qubit is maximally entangled with the pair it leaves behind, while every isolated pair is separable. We group the state across each relevant cut, trace out the complement term by term, and see which cross terms survive.

Step 1. Group $A$ against $BC$ : $|\mathrm{GHZ}\rangle=\tfrac{1}{\sqrt2}(|0\rangle_A|00\rangle_{BC}+|1\rangle_A|11\rangle_{BC})$ , with the two $BC$ branches $|00\rangle,|11\rangle$ orthonormal. Why: this exposes the $A|BC$ cut as a Schmidt decomposition with two equal weights $\tfrac12$ — already the whole story of the single-party marginal.

Step 2. Trace over $BC$ : the cross terms carry $\langle00|11\rangle=0$ and vanish, so $\rho_A=\tfrac12(|0\rangle\langle0|+|1\rangle\langle1|)=I/2$ . Why: orthogonality of the branches erases all coherence when $BC$ is discarded, leaving a maximally mixed qubit, $S=1$ bit — maximal $A|BC$ entanglement.

Step 3. Regroup across $AB$ against $C$ : $|\mathrm{GHZ}\rangle=\tfrac{1}{\sqrt2}(|00\rangle_{AB}|0\rangle_C+|11\rangle_{AB}|1\rangle_C)$ , and trace over $C$ . The off-diagonal terms carry $\langle0|1\rangle_C=0$ , so $\rho_{AB}=\tfrac12(|00\rangle\langle00|+|11\rangle\langle11|)$ . Why: discarding one qubit again kills the coherence, but what remains is a mixture of two product states — a classical correlation, not an entangled one.

Step 4. The concurrence of $\rho_{AB}$ : diagonal in the product basis with no coherence, so its spin-flip eigenvalues vanish and $C(\rho_{AB})=0$ — separable. Why: the fragility — GHZ entanglement is stored in no pair, so losing the third qubit leaves classically-correlated bits with nothing to distill.

Go deeper · Derivation B · 1 minThe all-versus-nothing contradiction with local realism

Derivation BThe all-versus-nothing contradiction with local realism

The goal is to show that a single perfectly-correlated set of GHZ measurements is logically incompatible with any local hidden-variable model — no inequality, no statistics, no averaging. We assume local realism in its strongest form: each qubit $j$ carries predetermined outcomes $m_x^{(j)},m_y^{(j)}\in\{+1,-1\}$ for $\sigma_x,\sigma_y$ , fixed independently of the distant measurement settings, and we derive a flat contradiction between what quantum mechanics assigns to a certain product of observables and what any such assignment must give.

Step 1. Record the quantum predictions. From the stabilizer relations of $|\mathrm{GHZ}\rangle$ (next box), $XXX=+1$ and $XYY=YXY=YYX=-1$ with certainty on every copy. Why: these are eigenvalue equations, so each joint observable has a definite outcome — no randomness for a hidden-variable model to hide behind.

Step 2. Impose locality: a local-realist assignment factors each joint outcome into single-site values, $XYY\to m_x^{(1)}m_y^{(2)}m_y^{(3)}$ , and so on. Why: locality forbids a qubit's readout from depending on distant settings, so the joint value splits into a product of pre-existing single-qubit values.

Step 3. Multiply the four factorized predictions: $(m_x^{(1)}m_x^{(2)}m_x^{(3)})(m_x^{(1)}m_y^{(2)}m_y^{(3)})(m_y^{(1)}m_x^{(2)}m_y^{(3)})(m_y^{(1)}m_y^{(2)}m_x^{(3)})$ . Every hidden value appears exactly twice. Why: across the four settings each site contributes one $X$ and two $Y$ 's, forcing every variable to a square.

Step 4. Since each $m=\pm1$ , every squared factor is $+1$ , so the local-realist product is forced to $+1$ . But quantum mechanics gives $(+1)(-1)(-1)(-1)=-1$ : a contradiction on a single run, not a violated average. Why: this all-versus-nothing gap is stronger than Bell/CHSH — no classical model even approximately reproduces these deterministic correlations.

Go deeper · Derivation C · 1 minGHZ and W as distinct SLOCC classes via the three-tangle

Derivation CGHZ and W as distinct SLOCC classes via the three-tangle

The goal is to prove GHZ and W are genuinely different tripartite resources — not interconvertible by local operations even with vanishing probability — by exhibiting an SLOCC invariant, the three-tangle $\tau_{ABC}$ , nonzero for one and zero for the other. Since $\tau$ transforms by the positive factor $|\det g_A\det g_B\det g_C|^2$ under invertible local filtering $g\in\mathrm{GL}(2,\mathbb C)^{\otimes3}$ , the property " $\tau=0$ " survives every SLOCC operation. We compute $\tau$ for both and invoke that invariance.

Step 1. For GHZ, the marginals give $C_{A(BC)}^2=4\det\rho_A=1$ and $C_{AB}=C_{AC}=0$ , so $\tau_{ABC}(\mathrm{GHZ})=1$ . Why: GHZ saturates CKW monogamy with all entanglement in the tripartite residual and none in any pair — the maximal genuinely tripartite state.

Step 2. For $W=\tfrac{1}{\sqrt3}(|001\rangle+|010\rangle+|100\rangle)$ , $\rho_A=\mathrm{diag}(\tfrac23,\tfrac13)$ gives $C_{A(BC)}^2=4\det\rho_A=\tfrac89$ , while each pair has concurrence $\tfrac23$ , so $C_{AB}^2=C_{AC}^2=\tfrac49$ . Why: W stores its entanglement pairwise — a genuinely tripartite state whose correlations are carried by two-body bonds.

Step 3. The W residual: $\tau_{ABC}(W)=\tfrac89-\tfrac49-\tfrac49=0$ . Why: monogamy is saturated by the pairwise terms alone, leaving no tripartite residual — the opposite balance from GHZ.

Step 4. Because SLOCC sends $\tau\to|\det g_A\det g_B\det g_C|^2\,\tau$ , a positive rescaling, it can never turn a nonzero $\tau$ into zero. A GHZ state ( $\tau=1$ ) can therefore never be mapped onto W ( $\tau=0$ ), even probabilistically, or conversely. Why: this is the precise sense in which GHZ and W are two inequivalent classes of genuine tripartite entanglement — the SLOCC classification of Dür, Vidal and Cirac.

Limiting cases

The GHZ state sits at several extremes of the multipartite landscape at once; the contrasts below sharpen exactly what is special about it.

Case	Result	Meaning
Single-qubit reduction	$\rho_A=\mathrm{Tr}_{BC}\,\vert\mathrm{GHZ}\rangle\langle\mathrm{GHZ}\vert=\tfrac{I}{2}$	Maximally mixed: each qubit is maximally entangled with the remaining pair, entanglement entropy $S=1$ bit across every $1$ -vs- $2$ cut.
Two-qubit reduction	$\rho_{AB}=\tfrac12(\vert00\rangle\langle00\vert+\vert11\rangle\langle11\vert)$	Separable and classically correlated, concurrence $C=0$ : loss of one qubit destroys all entanglement — GHZ fragility.
Three-tangle, GHZ vs W	$\tau_{ABC}(\mathrm{GHZ})=1,\ \tau_{ABC}(W)=0$	GHZ maximizes the tripartite residual; W has none, though both are genuinely tripartite — two inequivalent SLOCC classes.
Mermin expectation	$\langle\mathcal M\rangle_{\mathrm{GHZ}}=4$ , bound $2$ , algebraic max $4$	GHZ hits the algebraic ceiling of the Mermin inequality — maximal multipartite nonlocality, unlike CHSH's $2\sqrt2$ .
Large $N$ (GHZ $_N$ )	Mermin-type violation grows as $2^{(N-1)/2}$ ; one lost qubit $\Rightarrow$ separable rest	Nonlocality sharpens with party number, yet a single decohered party classicalizes the whole register — the core obstacle to scaling GHZ.

Common trapEvery qubit is maximally entangled, so every pair must be entangled too

This is the most common error about GHZ states, and it is false. Each single qubit is maximally entangled with the other two ( $\rho_A=I/2$ ), yet each pair is the separable mixture $\tfrac12(|00\rangle\langle00|+|11\rangle\langle11|)$ with concurrence exactly zero. Maximal entanglement across the $1$ -vs- $2$ cut does not imply, and here excludes, pairwise entanglement: the correlation is three-body with no two-body shadow. Two related traps: GHZ and W are both genuinely tripartite yet SLOCC-inequivalent (you cannot convert one to the other even with vanishing probability, since $\tau=1$ for GHZ and $0$ for W); and the GHZ–Mermin refutation is a logical, single-shot contradiction ( $+1\ne-1$ ), not a statistical CHSH violation.

Worked examples

1 · Full GHZ audit — marginals, entropy, tangle, and the Mermin value (fully worked)

For $|\mathrm{GHZ}\rangle=\tfrac{1}{\sqrt2}(|000\rangle+|111\rangle)$ , compute the single- and two-qubit reduced states, the entanglement entropy of one qubit, the three-tangle, and the Mermin expectation value $\langle\mathcal M\rangle$ — and interpret each.

Step 1. Single-qubit marginal. Grouping $A$ against $BC$ and tracing out the orthogonal branches gives $\rho_A=\tfrac12(|0\rangle\langle0|+|1\rangle\langle1|)=I/2$ , eigenvalues $\{\tfrac12,\tfrac12\}$ , so $S(\rho_A)=-2\cdot\tfrac12\log_2\tfrac12=\mathbf{1\ bit}$ — each qubit maximally entangled with the pair it leaves behind.

Step 2. Two-qubit marginal. Tracing out $C$ gives $\rho_{AB}=\tfrac12(|00\rangle\langle00|+|11\rangle\langle11|)$ , diagonal in the product basis with no coherence, so its Wootters concurrence is $\mathbf{C(\rho_{AB})=0}$ — separable (likewise $\rho_{AC},\rho_{BC}$ ).

Step 3. Three-tangle. The one-vs-two concurrence is $C_{A(BC)}^2=4\det\rho_A=1$ , while $C_{AB}^2=C_{AC}^2=0$ . Hence $\tau_{ABC}=1-0-0=\mathbf{1}$ : entanglement entirely tripartite, saturating CKW monogamy with nothing left for the pairs.

Step 4. Mermin value. From the stabilizer relations $XXX|\mathrm{GHZ}\rangle=+|\mathrm{GHZ}\rangle$ , $XYY|\mathrm{GHZ}\rangle=YXY|\mathrm{GHZ}\rangle=YYX|\mathrm{GHZ}\rangle=-|\mathrm{GHZ}\rangle$ , $\langle\mathcal M\rangle=1-(-1)-(-1)-(-1)=\mathbf{4}$ — the algebraic maximum, double the local-realist bound $2$ .

Reading. The four numbers — $S=1$ bit, $C=0$ , $\tau=1$ , $\langle\mathcal M\rangle=4$ — are the complete fingerprint of GHZ: maximal one-vs-two entanglement, zero pairwise entanglement, maximal tripartite tangle, maximal nonlocality. Every qubit is maximally entangled with the rest, no pair is entangled at all, and the state saturates the Mermin ceiling.

2 · The W state, for contrast — where the tangle goes to zero

Repeat the tangle computation for $W=\tfrac{1}{\sqrt3}(|001\rangle+|010\rangle+|100\rangle)$ and explain why it lands in a different SLOCC class from GHZ.

Step 1. Single-qubit marginal. Qubit $A$ is $|1\rangle$ in one of three terms and $|0\rangle$ in the other two, so $\rho_A=\mathrm{diag}(\tfrac23,\tfrac13)$ and $C_{A(BC)}^2=4\det\rho_A=\mathbf{8/9}$ — large, but below the GHZ value $1$ .

Step 2. Pairwise concurrence. Tracing out $C$ leaves a rank-two entangled state on $AB$ with $C_{AB}=\tfrac23$ , so $C_{AB}^2=\mathbf{4/9}$ , and likewise $C_{AC}^2=4/9$ . Unlike GHZ, the pairs of W are entangled and survive loss of the third qubit.

Step 3. Three-tangle: $\tau_{ABC}(W)=\tfrac89-\tfrac49-\tfrac49=\mathbf{0}$ — all of W's entanglement is pairwise, with no three-body residual. Because SLOCC can only rescale $\tau$ by a positive factor, no invertible local operation maps GHZ ( $\tau=1$ ) to W ( $\tau=0$ ) or back, even with vanishing probability — they are the two distinct genuine-tripartite classes.

3 · Certifying GHZ in a noisy lab — the witness threshold

A source prepares $\rho=p\,|\mathrm{GHZ}\rangle\langle\mathrm{GHZ}|+(1-p)\,\tfrac{I}{8}$ (GHZ plus white noise). How pure must it be for the witness $\mathcal W_{\mathrm{GHZ}}=\tfrac12I-|\mathrm{GHZ}\rangle\langle\mathrm{GHZ}|$ to certify genuine tripartite entanglement?

Step 1. The GHZ fidelity is $F=\langle\mathrm{GHZ}|\rho|\mathrm{GHZ}\rangle=p+(1-p)\tfrac18=\tfrac18+\tfrac78 p$ , so $\mathrm{Tr}(\mathcal W_{\mathrm{GHZ}}\rho)=\tfrac12-F$ is negative exactly when $\mathbf{F>1/2}$ .

Step 2. Setting $\tfrac18+\tfrac78p>\tfrac12$ gives $p>\tfrac{3}{7}\approx\mathbf{0.43}$ : the state need be only $43\%$ GHZ by weight for the witness to fire. That fidelity, read from $XXX$ and the $Z$ -parities, is the standard lab acceptance test. Fidelity above 1/2 certifies the whole register is genuinely, not merely biseparably, entangled.

Physics of the portrait

The GHZ state's arc runs from a thought experiment through a teachable paradox to a Nobel-winning laboratory reality — a foundations argument that became a working piece of quantum technology within a single career.

Origins · 1 minAnton Zeilinger and how this was discovered

Anton Zeilinger

b. 1945, Austrian experimental physicist, Nobel Prize in Physics 2022 · experimental quantum foundations and multiphoton entanglement

Anton Zeilinger is the experimentalist who turned the GHZ argument from a chalkboard paradox into laboratory fact. The state and its theorem were born in 1989, when Daniel Greenberger, Michael Horne and Zeilinger showed that three or more entangled particles refute local hidden variables deterministically, not statistically — a result so counterintuitive that it took David Mermin's crisp 1990 reformulation, the "all-versus-nothing" contradiction $+1\ne-1$ , to make it teachable. For a decade the three-particle state stayed a theorist's object, since producing three simultaneously entangled photons lay beyond the two-photon sources of the day. Zeilinger's groups closed that gap: in 1999–2000 they reported the first experimental GHZ states and the first confirmation of the GHZ-Mermin predictions, observing the very sign flip local realism forbids. That work, with his teleportation and entanglement-swapping demonstrations, earned him a share of the 2022 Nobel Prize with John Clauser and Alain Aspect. The GHZ state sits at the centre of that citation: the sharpest weapon ever devised against local realism, and Zeilinger is the one who fired it.

History

1935

— Einstein, Podolsky and Rosen, and independently Schrödinger, frame entanglement and the cat paradox

1964

— Bell derives his inequality, turning local realism into a testable statistical claim

1989

— Greenberger, Horne and Zeilinger publish the three-particle all-versus-nothing argument

1990

— Mermin recasts GHZ as a single-run logical contradiction and introduces the Mermin operator

2000

— Coffman, Kundu and Wootters define the three-tangle and prove the CKW monogamy relation

2000

— Dür, Vidal and Cirac prove GHZ and W are the only two genuine tripartite SLOCC classes

2022

— Zeilinger shares the Nobel Prize for experimental tests of entanglement and quantum information

Going further

Texts. Nielsen & Chuang, Quantum Computation and Quantum Information, §2.6, §12.3 · Preskill, Lecture Notes on Quantum Computation, Ch. 4 · Wilde, Quantum Information Theory, Ch. 11 · Watrous, The Theory of Quantum Information, Ch. 6. Primary sources. Greenberger, Horne & Zeilinger, in Bell's Theorem, Quantum Theory and Conceptions of the Universe (1989) · Mermin, Phys. Rev. Lett. 65, 1838 (1990) · Coffman, Kundu & Wootters, Phys. Rev. A 61, 052306 (2000) · Dür, Vidal & Cirac, Phys. Rev. A 62, 062314 (2000).

Provenance. All results were re-derived and numerically checked from $|\mathrm{GHZ}\rangle=\tfrac{1}{\sqrt2}(|000\rangle+|111\rangle)$ : marginals $\rho_A=I/2$ ( $S=1$ bit) and $\rho_{AB}$ separable ( $C=0$ ); stabilizer eigenvalues $XXX=+1$ , $XYY=YXY=YYX=-1$ giving $\langle\mathcal M\rangle=4$ vs the LHV bound $2$ ; $\tau_{\mathrm{GHZ}}=4\det\rho_A=1$ versus $\tau_W=0$ ; and the biseparable overlap $\tfrac12$ fixing the witness. One correction to older drafts: the witness is $\tfrac12I-|\mathrm{GHZ}\rangle\langle\mathrm{GHZ}|$ (threshold $\tfrac12$ ), not $\tfrac34$ — the maximal biseparable overlap for the three-qubit GHZ is $1/2$ , by Cauchy–Schwarz on the $A|BC$ Schmidt structure.

Builds on: Bell States and EPR Pairs · Separable versus Entangled States · Schmidt Decomposition · Bell's Theorem and Local Hidden Variables. Leads to: Entanglement Witnesses · Concurrence and Wootters' Formula · CHSH Inequality and Its Quantum Violation · Nonlocality and the No-Signalling Principle.

Ask the tutor: Ask me to walk through the all-versus-nothing sign flip line by line, or to compute the three-tangle of any three-qubit state.

Why is the GHZ state called "genuinely" multipartite entangled?

A three-qubit state is biseparable if it is a mixture of states each product across some bipartition — $A|BC$ , $B|AC$ , or $C|AB$ . Genuine multipartite entanglement means biseparable across no bipartition, even as a mixture. GHZ qualifies: whichever qubit you try to split off, it does not factor, and the whole cannot be built from parts entangled only two-at-a-time. The witness $\tfrac12I-|\mathrm{GHZ}\rangle\langle\mathrm{GHZ}|$ makes this operational — its expectation is negative only for states no biseparable mixture can reach.

How can every qubit be maximally entangled with the rest while no pair is entangled?

This feels paradoxical only if you carry two-qubit intuition into three. Entanglement is a property of a cut, and GHZ has two different cuts. Across the $1$ -vs- $2$ cut the state is maximally entangled, $\rho_A=I/2$ . But the $AB$ pair alone comes from discarding $C$ , and that partial trace destroys the coherence linking the branches, leaving a classical mixture with zero entanglement. No contradiction — the entanglement is a three-body correlation with no two-body projection, which is what "genuine tripartite" means.

What makes the GHZ argument different from Bell's or CHSH's?

Bell/CHSH is statistical: local hidden variables are bounded by an average ( $|S|\le2$ ), and quantum mechanics beats it only on average, so you accumulate many runs. GHZ is logical: the four observables $XXX,XYY,YXY,YYX$ have definite values, and any local-realist assignment forces their product to $+1$ while quantum mechanics forces $-1$ — a contradiction on a single run, no inequality, no statistics. Mermin called it "all versus nothing" because local realism gets the answer exactly backwards, not just too small on average.

What exactly is the three-tangle, and why does it distinguish GHZ from W?

The three-tangle $\tau_{ABC}=C_{A(BC)}^2-C_{AB}^2-C_{AC}^2$ is the residual of the Coffman–Kundu–Wootters monogamy relation: how much of qubit $A$ 's entanglement with the rest is not explained by its pairwise bonds. It is symmetric in the three qubits and an SLOCC invariant. For GHZ, $C_{A(BC)}^2=1$ and the pairwise concurrences vanish, so $\tau=1$ — pure tripartite. For W the pairwise terms $4/9+4/9$ exactly absorb the $8/9$ one-vs-two term, so $\tau=0$ — pure pairwise. Since $\tau=0$ is SLOCC-invariant, the two can never be interconverted locally.

Why is the GHZ state described as "fragile," and does that limit its usefulness?

Fragile means losing or decohering one qubit collapses the entire global correlation into a classical mixture — once a party is traced out, no entanglement survives in any subset. That is a liability for storage and metrology (a GHZ probe decoheres $N$ times faster than a single qubit) and the reason large GHZ states are hard to scale. But the same fragility is a feature for security: in quantum secret sharing no proper subset of parties can extract any information, precisely because their reduced state is separable. Whether it helps or hurts depends on whether you want to protect the correlation or exploit its all-or-nothing character.

Do GHZ states exist for more than three qubits, and does anything change?

Yes — equation (1) defines $|\mathrm{GHZ}_N\rangle$ for any $N$ , and the features persist: single-qubit marginals are $I/2$ , the state is genuinely $N$ -partite, and it is stabilized by $X^{\otimes N}$ with the $Z_iZ_{i+1}$ parities. The nonlocality strengthens with $N$ : the Mermin–Klyshko inequalities are violated by a factor growing as $2^{(N-1)/2}$ , so the quantum–classical gap widens with party number — the opposite of the fixed CHSH gap. So does the fragility: one lost qubit still classicalizes the whole register, the central obstacle to scaling GHZ states.