In 1964, John Bell first derived the original Bell inequality, $|E(a,b)-E(a,c)|\leq1+E(b,c)$. Here $a,b,c$ are three different possible spin measurement directions, and $E$ is the measured correlation, meaning the expectation value of the product of the signs of the two outcomes, given the specified measurement settings. (For example : there are two detectors A and B, and spin is measured in units of $\hbar/2$,now if for a particular measurement A measures +1 and B measures -1 then their product is -1, if we consider many runs of the experiment and caculate the average of the product for directions $a$ and $b$ which is $P(a,b)$, then quantum mechanics predict $P(a,b)= -a.b$.)

From this form of Bell's inequality, consider starting with a singlet spin state, and defining these three measurement directions such that the angle between each of the vectors of spin measurements are 120°. Then, I find that the correlations are $E(a,b)=-a.b= 1/2=E(a,c)=E(b,c)$. Thus $|\frac{1}{2}-\frac{1}{2}|\leq 1+\frac{1}{2}$ means $0\leq3/2$ which satisfies Bell's inequality, instead of violating it.

And yet this example is often used to show a violation of Bell's inequality. In the linked article, the calculation agrees with mine, showing that for any two different measurements, the outcomes have the same sign 75% of the time and a different sign 25% of the time. So the correlation functions $E(a,b)$, etc., are all indeed 0.5. Please help explain in what way this system violates the Bell inequality.

  • $\begingroup$ Of interest: this question is a general discussion of cases where the inequality is not violated: physics.stackexchange.com/q/732247 $\endgroup$
    – rghome
    Commented Oct 13, 2023 at 7:40
  • 1
    $\begingroup$ @TobiasFünke: See Mermin's "Bringing home the atomic world: Quantum mysteries for anyone" (1981), which considers entangled photons encountering a pair of detectors having with three settings 120° apart. This illustration of Bell's inequality was later picked up by other authors, including (off the top of my head) Moore in Six Ideas that Shaped Physics and Maudlin's Quantum Nonlocality and Relativity. $\endgroup$ Commented Oct 13, 2023 at 11:30
  • $\begingroup$ @TobiasFünke Preskill in his lecture notes theory.caltech.edu/~preskill/ph229 also uses that Bell inequality (which afaik is the original one by Bell) to explain this. (Chapter 4.2) $\endgroup$ Commented Oct 13, 2023 at 11:36
  • $\begingroup$ @SubhajitPodder This is pure guessing (partly for the reasons explained above), but: In the standard Bell inequality, the measurements must be in opposite directions for the two parties. It is hard to tell since you don't explain your formulas, but could it be that in that case, there is an additional minus sign, so that P = -1/2 instead of 1/2, which would violate the inequality? $\endgroup$ Commented Oct 13, 2023 at 11:43
  • 1
    $\begingroup$ @DrChinese The OP only ever used P. (In fact, checking the edits, only a minor fraction of the post is by the OP, so it is really hard to say what they think.) $\endgroup$ Commented Oct 15, 2023 at 21:11

3 Answers 3


Bell's inequality is not necessarily violated for all possible angles between measurement directions - it may be that measurements at some angles meet Bell's inequality, as in the example you have given.

However, we know that Bell's inequality is violated for some angles between measurements. A common example is when the measurements are at $0$, $45$ and $90$ degrees - see this Wikipedia article.

Your confusion may arise from the fact that some explanations of Bell's theorem use a set up of detectors in which $P(a,b) = \cos^2 (\frac {\theta} 2)$; this set up is different from that in Bell's original paper, where $P(a,b) = - \cos \theta$.

  • $\begingroup$ The standard Bell inequality is violated for the 120-degree alignment. There seems to be an error in the OPs calculation (or definitions), which needs to be clarified. $\endgroup$ Commented Oct 13, 2023 at 11:28
  • $\begingroup$ yes, but I remember that for 120 angles it should be violated , also see quantamagazine.org/… $\endgroup$ Commented Oct 13, 2023 at 11:59
  • $\begingroup$ @NorbertSchuch Can you show the calculation which demonstrates that Bell's inequality is violated when the detectors are set an angles $0, 120, 240$ degrees, so $P(a,b) = P(a,c) = P(b,c) = - \cos 120^o = \frac 1 2$ ? $\endgroup$
    – gandalf61
    Commented Oct 13, 2023 at 12:10
  • 1
    $\begingroup$ @NorbertSchuch No, no, no. The OP says that Bell's inequality is not violated in the $0,120,240$ case and they think it should be. But their calculation is correct - the inequality is not violated in the $0,120,240$ degrees case, but it does not need to be because Bell's theorem does not say the inequality is violated for all possible settings of the detectors, but only for some of them. $\endgroup$
    – gandalf61
    Commented Oct 13, 2023 at 12:22
  • 1
    $\begingroup$ @Mauricio see my comment above. Indeed the violations occur at different angles for electrons vs. photons. The 120/120/120 case is not violated for electron spin, but is violated for photon polarization. $\endgroup$
    – DrChinese
    Commented Oct 15, 2023 at 19:16

The correct answer, already given in a comment, isn't that there is a factor of two in the angles; rather, it's the fact that the quanta magazine used a different inequality than the one originally formulated by Bell. More specifically, they used an inequality due to Mermin, which you can also read more about here.

In Mermin's scenario, we have three random variables for Alice and Bob each: $A_j$, $B_k$, and $j,k = 1,2,3$. These variables correspond to the different settings for Alice and Bob, and take on the values $\pm 1$. We are interested in how often Alice and Bob give the same answer, which is given by the product of their random variables. The sum of these products is lower bounded:

$$ \left|\sum_{jk}A_jB_k\right| \geq 1 $$

since there are nine terms in the sum, and the smallest value is when four (five) are +1 (-1) or vice versa. Let's look at the case where

$$ \sum_{jk}A_jB_k \leq -1. $$

Let the joint expectation values be $E(A_j,B_k) = E(A_j B_k)$. Then

$$ E(\sum_{jk}A_jB_k) = \sum_{j}E(A_j,B_j) + \sum_{j\neq k} E(A_j,B_k) \leq -1. $$

If we require anti-correlated outcomes when Alice and Bob choose the same setting ($E(A_j,B_j) = -1$) this can be written

$$ \sum_{j\neq k} E(A_j,B_k) \leq 2 $$

If you do the same for the other sign in the absolute value you get $\sum_{j\neq k} E(A_j,B_k) \geq 4$. Taking the settings $A_i = B_i = \{0^{\circ},120^{\circ},240^{\circ}\}$ we see that the quantum prediction violates the inequality

$$ 4 \nleq \sum_{j\neq k} E(A_j,B_k) = 6\cdot\frac{1}{2} = 3 \nleq 2. $$

If you want to write the inequality on a form even more similar to the original Bell inequality you can use $E(a,b) = E(b,a)$ (this follows from the requirement of anti-correlation for same settings) to simplify it to

$$ E(a,b) + E(a,c) + E(b,c) \leq 1 \ \text{or}\ 2\leq E(a,b) + E(a,c) + E(b,c) \\ \iff |E(a,b) + E(a,c) + E(b,c) - \frac{3}{2}| \geq \frac{1}{2}. $$

Note that neither this form of the inequality, nor Bell's original one, are experimentally testable since they use perfect anti-correlation for equal settings as part of the derivation.

Edit: I actually don't think the bound $E(a,b) + E(a,c) + E(b,c) \geq 2$ is necessary. It's easy to show that there is a strategy that satisfies $E(a,b) + E(a,c) + E(b,c) = 1$, and if there was one that satisfied $E(a,b) + E(a,c) + E(b,c) = 2$ a convex combination of them could violate the inequality. Maybe someone can show this in a more direct way.

  • 1
    $\begingroup$ Right, that's precisely the version of the inequality (and the storyline provided, cf. your second link) by Preskill in his lecture notes, Sec. 4.2! $\endgroup$ Commented Oct 16, 2023 at 20:04
  • 1
    $\begingroup$ ... Your last inequality looks strange. I mean ... it says 2<=1. $\endgroup$ Commented Oct 16, 2023 at 21:11
  • $\begingroup$ @NorbertSchuch thanks, that was a stupid way to write what I meant. $\endgroup$
    – fulis
    Commented Oct 16, 2023 at 22:20
  • $\begingroup$ I think the curly bracket is still confusing: What you mean is "or", what at least I would implicitly think is "and". $\endgroup$ Commented Oct 16, 2023 at 22:24
  • 1
    $\begingroup$ Fully agree with all you say, except that 2<=1 ;) $\endgroup$ Commented Oct 17, 2023 at 14:40

This experiment you describe can indeed be used to show that something non-classical is going on. However, as you noticed, you can't see that non-classicality by looking at Bell's original inequality from his 1964 paper. (Note that this is an inequality which is barely used anymore; these days the gold standard is the CHSH inequality.) True, unlike CHSH, that original inequality had three settings, as does your example. But the non-classicality of your example isn't apparent until you also analyze the additional three (perfect) anti-correlations $E(a,a)=-1$, $E(b,b)=-1$ and $E(c,c)=-1$. With these perfect anti-correlations in the mix, you can indeed show that there is no classical account of these experiments.

If you're asking about the relevant inequality which these correlations violate, the original point of this very example was to avoid that sort of technical argument. Instead, Mermin wanted to show that the assumptions behind Bell's original Inequality (and the CHSH, and others) must be violated without using inequalities. The argument simply shows that there is no classical explanation for the experiment. In other words, given the usual assumptions behind Bell's Theorem, there is simply no set of hidden states which will allow both the correlations you calculated as well as the perfect anti-correlations $E=-1$ for the same settings.

EDIT: The wrong paper was cited before; this set-up was from Mermin, and indeed the experiment has its own wikipedia page.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.