Categories: Physics, Quantum information, Quantum mechanics.

GHZ paradox

The Greenberger-Horne-Zeilinger or GHZ paradox is an alternative proof of Bell’s theorem that does not use inequalities, but instead the three-particle entangled GHZ state GHZ\ket{\mathrm{GHZ}}:

GHZ12(000+111)\begin{aligned} \boxed{ \ket{\mathrm{GHZ}} \equiv \frac{1}{\sqrt{2}} \Big( \ket{000} + \ket{111} \Big) } \end{aligned}

Where 0\ket{0} and 1\ket{1} are qubit states, specifically the eigenvalues of the Pauli matrix σ^z\hat{\sigma}_z.

If we now apply certain products of the Pauli matrices σ^x\hat{\sigma}_x and σ^y\hat{\sigma}_y as quantum gates to this three-particle state, we find:

σ^xσ^xσ^xGHZ=12(σ^x0σ^x0σ^x0+σ^x1σ^x1σ^x1)=12(111+000)=GHZσ^xσ^yσ^yGHZ=12(σ^x0σ^y0σ^y0+σ^x1σ^y1σ^y1)=12(1i1i1+0i0i0)=GHZ\begin{aligned} \hat{\sigma}_x \otimes \hat{\sigma}_x \otimes \hat{\sigma}_x \ket{\mathrm{GHZ}} &= \frac{1}{\sqrt{2}} \Big( \hat{\sigma}_x \ket{0} \otimes \hat{\sigma}_x \ket{0} \otimes \hat{\sigma}_x \ket{0} + \hat{\sigma}_x \ket{1} \otimes \hat{\sigma}_x \ket{1} \otimes \hat{\sigma}_x \ket{1} \Big) \\ &= \frac{1}{\sqrt{2}} \Big( \ket{1} \otimes \ket{1} \otimes \ket{1} + \ket{0} \otimes \ket{0} \otimes \ket{0} \Big) \\ &= \ket{\mathrm{GHZ}} \\ \hat{\sigma}_x \otimes \hat{\sigma}_y \otimes \hat{\sigma}_y \ket{\mathrm{GHZ}} &= \frac{1}{\sqrt{2}} \Big( \hat{\sigma}_x \ket{0} \otimes \hat{\sigma}_y \ket{0} \otimes \hat{\sigma}_y \ket{0} + \hat{\sigma}_x \ket{1} \otimes \hat{\sigma}_y \ket{1} \otimes \hat{\sigma}_y \ket{1} \Big) \\ &= \frac{1}{\sqrt{2}} \Big( \ket{1} \otimes i \ket{1} \otimes i \ket{1} + \ket{0} \otimes i \ket{0} \otimes i \ket{0} \Big) \\ &= - \ket{\mathrm{GHZ}} \end{aligned}

In other words, the GHZ state is a simultaneous eigenstate of these composite operators, with eigenvalues +1+1 and 1-1, respectively. Let us do the same for two more operators, so that we have a set of four observables for which GHZ\ket{\mathrm{GHZ}} gives these eigenvalues:

σ^xσ^xσ^x    +1σ^xσ^yσ^y    1σ^yσ^xσ^y    1σ^yσ^yσ^x    1\begin{aligned} \hat{\sigma}_x \otimes \hat{\sigma}_x \otimes \hat{\sigma}_x \quad &\implies \quad +1 \\ \hat{\sigma}_x \otimes \hat{\sigma}_y \otimes \hat{\sigma}_y \quad &\implies \quad -1 \\ \hat{\sigma}_y \otimes \hat{\sigma}_x \otimes \hat{\sigma}_y \quad &\implies \quad -1 \\ \hat{\sigma}_y \otimes \hat{\sigma}_y \otimes \hat{\sigma}_x \quad &\implies \quad -1 \end{aligned}

According to any local hidden variable (LHV) theory, the measurement outcomes of the operators are predetermined, and the three particles AA, BB and CC can be measured separately, or in other words, the eigenvalues can be factorized:

σ^xσ^xσ^x    +1=mxAmxBmxCσ^xσ^yσ^y    1=mxAmyBmyCσ^yσ^xσ^y    1=myAmxBmyCσ^yσ^yσ^x    1=myAmyBmxC\begin{aligned} \hat{\sigma}_x \otimes \hat{\sigma}_x \otimes \hat{\sigma}_x \quad &\implies \quad +1 = m_x^A m_x^B m_x^C \\ \hat{\sigma}_x \otimes \hat{\sigma}_y \otimes \hat{\sigma}_y \quad &\implies \quad -1 = m_x^A m_y^B m_y^C \\ \hat{\sigma}_y \otimes \hat{\sigma}_x \otimes \hat{\sigma}_y \quad &\implies \quad -1 = m_y^A m_x^B m_y^C \\ \hat{\sigma}_y \otimes \hat{\sigma}_y \otimes \hat{\sigma}_x \quad &\implies \quad -1 = m_y^A m_y^B m_x^C \end{aligned}

Where mxA=±1m_x^A = \pm 1 etc. Let us now multiply both sides of these four equations together:

(+1)(1)(1)(1)=(mxAmxBmxC)(mxAmyBmyC)(myAmxBmyC)(myAmyBmxC)1=(mxA)2(mxB)2(mxC)2(myA)2(myB)2(myC)2\begin{aligned} (+1) (-1) (-1) (-1) &= (m_x^A m_x^B m_x^C) (m_x^A m_y^B m_y^C) (m_y^A m_x^B m_y^C) (m_y^A m_y^B m_x^C) \\ -1 &= (m_x^A)^2 (m_x^B)^2 (m_x^C)^2 (m_y^A)^2 (m_y^B)^2 (m_y^C)^2 \end{aligned}

This is a contradiction: the left-hand side is 1-1, but all six factors on the right are +1+1. This means that we must have made an incorrect assumption along the way.

Our only assumption was that we could factorize the eigenvalues, so that e.g. particle AA could be measured on its own without an “action-at-a-distance” effect on BB or CC. However, because that leads us to a contradiction, we must conclude that action-at-a-distance exists, and that therefore all LHV-based theories are invalid.


  1. N. Brunner, Quantum information theory: lecture notes, 2019, unpublished.
  2. J.B. Brask, Quantum information: lecture notes, 2021, unpublished.