Musing

When I was studying Physics at university, it was, if I can actually remember correctly, common to say that the Bell Inequalities require us to give up one of three things:

  1. Locality: distant, spacelike separated, events can not influence one another.
  2. Counterfactual definiteness: observables have a definite value even when I am not measuring them.
  3. Scientific induction: the basis of all empiricism.

I am now convinced (to beyond reasonable doubt) that in fact we must give up locality. Furthermore, I think the focus on “counter factuals” (“what if I measured that instead of this?”) is largely a red herring. The 1964 Bell paper, which builds on top of EPR, inherited perfect anti-correlations which then brought about outcome determinism. Deterministic outcomes could be seen as providing definite values for counterfactuals. But future Bell papers relaxed the perfect anti-correlation requirement, worked for stochastic outcomes, and thereby no longer touched counterfactuals. Long story short, I don’t believe you can bypass Bell’s theorem by (simply) denying counterfactual definiteness.

Bell’s theorem and their violation in experiment form the core difficulty in understanding Quantum Mechanics. Before Bell’s inequalities were violated in the lab (and subsequent experiments closed the relevant loopholes), we could still hope that our classical intuitions (“classical” here includes relativity theory) on matter and causality were valid. We need merely believe that Quantum Mechanics is incomplete (or experiment would show it to be wrong). Once Bell’s inequalities were shown to be violated in the lab, we knew that we have to give up something deeply precious to our “normal” intuitions. In this way, it is similar to having to give up our naive notions of absolute time and space to accommodate relativity.

At its core, Bell’s theorem is quite simple. It states that no locally causal theory (in the words of Bell, see 1) can fully reproduce the results of Quantum Mechanics. If all causes of an event are restricted to the past light cone of that event, we can not specify a complete set of variables describing the probability distribution of that event yielding the same results as quantum theory. This result essentially forces us to give up on locality. There are some loop holes around this argument which we will discuss in denials-of-bells-theorem.

Musing

This page is mostly taken from Bell’s original papers. Bell’s writing is terse, but very readable and often quite enjoyable. I would recommend reading some of his actual essays.

The Einstein Podolsky Rosen Argument

EPR, in 1935 2, made an argument for the case that the quantum mechanical description of the world was incomplete. In making this argument, they gave two important definitions:

  1. Criterion of reality: “If, without in any way disturbing a system, we can predict with certainty (i.e., with probability equal to unity) the value of a physical quantity, then there exists an element of physical reality corresponding to this physical quantity.” This criterion is a sufficient, but not necessary criterion.
  2. Condition of completeness: “every element of the physical reality must have a counterpart in the physical theory.” This condition is necessary.

Given these two definitions, EPR “proved” that Quantum Mechanics is incomplete. It was Bell who later recognized that EPR snuck in a hidden assumption in their proof: locality.

Musing

Niels Bohr provided a supposed refutation of the EPR argument in his paper of the same title 3. I have read this refutation. I can not understand this refutation.

The Setup

There’s 2 experimenters, Alice and Bob, who are far away from each other. So far away, that it takes light a decent amount of time to get from Alice to Bob. Halfway between them, we set up a device that can send a spin-1/2 particle to each of Alice and Bob in a special correlated state, which we model in quantum mechanics as a singlet state (explained below). We imagine, for example, a spin-0 particle which decays into two spin-1/2 particles. Alice and Bob each perform a spin measurement on this particle, e.g. using Stern-Gerlach magnets, and record the results. We perform this procedure many times and look at correlations in Alice’s and Bob’s tabulated results.

Most discussions of this thought experiment prior to Bell assumed Alice and Bob would measure the spin in parallel, anti-parallel, or orthogonal directions. One of the insights of Bell was that, indeed, there was no reason Alice and Bob needed to stick to those particular measurement settings. Therefore, in our setup, Alice chooses to measure the spin of her particle along the -axis while Bob measures the spin of his particle along the -axis.

The Quantum Prediction

The quantum mechanical analysis is easy. We have a singlet state:

Where, for convenience we wrote e.g. , and we label Alice’s and Bob’s operators and correspondingly. A measurement along some axis (with ) is represented by where are the 3 Pauli matrices packaged into a vector.

Musing

I’m going to drop a lot of symbols because I am lazy. I’ll only include it if I think it looks pretty. Live with it.

Hence, the expectation value of the product of the outcomes given by quantum mechanics is and we shall do this in stages. First, we will look at the action of Pauli matrices on our basis vectors:

Then we notice that will become 9 terms that look like: and in theory we can just grind through all the algebra to get finally:

Musing

It really is just a grind through algebra. I live in the school of “trust me bro” so you can take equation (1) as fact. But if it helps, this is the standard textbook result and is also the result Bell presents in his original 1964 paper.

Hint

At the end of the day, we will find that only 3 final terms survive (if our result is to be believed at all): . We can check one particular term that should disappear:

and then for the rest you do it yourself.

The Local Causal Prediction and The Inequality

Before we start, we should first note that there have been several versions of inequalities that are all called “Bell inequalities”. Bell himself is most famous for formulating the original version in a publication “On the Einstein Podolsky Rosen Paradox” 19644. This is the version of the inequalities that we will begin our argument with. However, we should note that there are several assumptions in this original version which are somewhat more stringent than they needed to be. Future publications, some from Bell himself, softened these assumptions to make “Bell’s Theorem” stronger. In particular, this 1964 version of the theorem appeals to perfect anti correlation, i.e. it requires that we are sure our state is prepared such that if Alice and Bob measure along the same axis they always get opposite results. This requirement is difficult to realize in the laboratory and, when combined with locality, gives rise to outcome determinism (explained in more detail below) which is sometimes the target of attack.

Musing

I did read a paper https://arxiv.org/pdf/1308.5290 from Englert which seems to quite miss the point of Bell’s theorem (all versions of it it seems). The claim made in the paper is quoted below. In sections 7 and 8 of the paper, he makes several claims criticising Bell’s theorem for apparently not dealing with the stochastic nature of the world. It’s unclear if the point is trying to attack outcome determinism or some strawman version of Bell’s theorem.

Berthold-Georg Englert

Quantum theory is a well-defined local theory with a clear interpretation. No “measurement problem” or any other foundational matters are waiting to be settled.

Let us begin by following Bell’s argument in his 1964 paper. The core hypothesis under test here is given by a quote.

John S. Bell

Now we make the hypothesis, and it seems one at least worth considering, that if the two measurements [of the singlet state using Stern-Gerlach magnets] are made at places remote from one another the orientation of one magnet does not influence the result obtained with the other.

Bell then goes on to give a quick (1 sentence actually, but I don’t want this page to just all be quotes) argument that the perfect (anti) correlation of the singlet state, combined with the EPR argument, means outcome determinism. This leads to his equation 1:

Let’s unpack this a little because Bell’s exposition is short but his argumentation packs a punch. and are functions which, given their inputs, will tell you the outcomes recorded by Alice and Bob respectively after they measure their particle along the and directions respectively. is the (or a set of) “hidden variable(s)“. The fact that functions and are deterministic functions of their inputs, unattached to any probability distribution over them, is the statement of outcome determinism. Why do I keep saying “outcome determinism” and not just “determinism”? Well, if we take seriously the fact that is a hidden variable which we can assign some probability distribution to, there is already a “lack of knowledge” notion of probability introduced in this picture.

Next, we note that core assumption in this equation, which Bell points out himself, is that is not a function of and is not a function of . This is the mathematical manifestation of that quoted hypothesis above and sometimes called “parameter independence” in the literature.

Determinism

One thing that Bell liked to emphasize (because his 1964 paper seems to get attacked on this point a lot) is that (outcome) determinism is a consequence of locality + perfect anti-correlation. For Bell’s original 1964 paper, he assumed a perfect anti-correlation and therefore outcome determinism came along for the ride. It happens that some believe we can bypass Bell’s theorem if only we accept that the world is not deterministic. This is false. In fact, Bell himself gave proofs of the CHSH inequalities 5 that don’t make use of outcome determinism at all. For an example of such an attack on Bell’s theorem, see the quote below.

Berthold-Georg Englert

…experimental data contradict Bell’s theorem, which implies that — as a statement about physical systems — the theorem is wrong. Since there is no error in the reasoning that establishes the theorem from its assumptions, the flaw must be in the assumptions. Specifically, it is the assumption that a mechanism exists that determines which detector will click for the next photon registered by [a polarization-measuring apparatus]. There is no such deterministic mechanism — quantum processes are fundamentally probabilistic, events are randomly realized — and the violation of Bell’s theorem by actual data confirms that.

Given a probability distribution over , we arrive immediately at the expectation value of joint measurements (i.e. the expectation of the product ):

Bell used the notation but I think this is confusing since P evokes probability. What we have here is an expectation value, so I will use the symbol instead. Note here that by choosing to not be dependent on measurement settings we are assuming statistical independence (sometimes also called measurement independence). If the distribution of itself depends on your measurement settings, the rest of the proof will not go through. This is the loop hole exploited by the so-called “superdeterminism”.

Our task would be to show that this can not actually reproduce (be made equal to) Quantum Mechanics result given earlier of .

The actual steps of the proof is quite simple. First, we notice that if then the QM prediction gives an expectation value . The only way to satisfy is to set (except possibly for a set of points in of zero probability). Note that we only invoked to get this equality, but we have not assumed what actually is, so we can, in particular, also say so that our expectation value equation can be rewritten:

Now we look at the quantity and do some algebra:

That second line is a bit slick, and Bell just says “using (1)” so let’s unpack it a little. First, we absorb the sign into the integral and reverse the order of the subtraction inside the solid brackets. Second, note that since that .

Now just take the absolute value of both sides:

Again, the second line is a bit slick. That line comes from moving the absolute value into the integral. The less than or equal sign appears of course because integrating an absolute value will be greater than or equal to the absolute value of the integral and since we should reverse the order of the difference inside the bracket when we take the absolute value. So let’s notice that the second line actually contains to get us the final Bell inequality:

So what have we (actually Bell) proven here? Bell proved that, given his original hypothesis and a few auxiliary assumptions, a quite broad class of local hidden variable theories will obey this inequality. Future versions of Bell’s inequalities, as discussed below, will broaden the argument to apply to essentially all local hidden variable theories. But let’s just try out this inequality. Let’s put on a single plane and angle them such that with bisecting that angle so that . If we plug in the quantum mechanical predictions for the expectation values into the inequality, we get:

An obvious contradiction. Quantum mechanics doesn’t care about maintaining this inequality. Bell devotes about two more pages to proving that essentially the form of the expectation value obtained via a local hidden variable theory can not reproduce, or even approximately reproduce, the form of the expectation value given by quantum mechanics. These arguments are a bit more mathematically dense. Instead of focusing on those arguments here, we will give a somewhat more intuitive view of Bell Inequalities later when we discuss the CHSH version of Bell’s theorem. Finally, regardless of what quantum mechanics, the theory, says, Bell inequality violations were confirmed in the lab (and indeed the 2022 Nobel Prize in physics was given to Aspect, Clauser, and Zeilinger for experimentally showing these violations). One could consider the proof of Bell inequality violations as the point in time in history when we definitively knew we didn’t live in a classically intuitive world.

Non-Conflicting Cases

If EPR made their argument against the completeness of QM in 1935 using entanglement, and Schroedinger already noted the importance of entanglement in the same year (see quote below), why did it take until 1964 for John Bell to carry that torch forward into a no-go theorem with eventual experimental consequences? The initial reason was that EPR’s entangled state is not amenable to actual experimental tests. Their exposition was based on continuous representations of the wave function, in position and momenta bases, which make 100% anti-correlation practically impossible. It was David Bohm, in the 1950’s, who gave the spin-based formulation of the EPR argument. Spin, being a discrete space, is much simpler to reason about as well as perform experiments on.

However, even after David Bohm reformulated the argument based on spin, most people only discussed the implications of the correlations for measurement axes fixed to perfectly aligned or orthogonal to each other. For such a simplistic setting, Bell’s inequalities don’t come into the picture and a local hidden variable theory could be formulated which reproduces the QM predictions for those particular settings. Bell goes over this in section 3 of his original 1964 paper4.

Erwin Schroedinger

I would not call [entanglement] one but rather the characteristic trait of quantum mechanics, the one that enforces its entire departure from classical lines of thought.

The CHSH Version

There’s a few practical issues with the original Bell Inequalities presented in The Local Causal Prediction and The Inequality when it comes to testing them in a lab. Furthermore, Bell would later sharpen his argument, especially by weakening the requirement of perfect anti-correlation that were inherited from the EPR argument, in several future papers after 1964. In each of those papers, e.g. “The Theory of Local Beables” 1, “Bertlmann’s Socks and the Nature of Reality” 5, and “La Nouvelle Cuisine” 6, Bell uses a version of the inequalities developed by Clauser-Horne-Shimony-Holt (CHSH). The argument proving the CHSH version (which we will give shortly) does not require the perfect anti-correlations and, therefore, does not lead to outcome determinism as was the case in the 1964 version. The requirement in the 1964 argument (specifically when we imposed the condition except on a set of measure zero) of a perfectly anti-correlated state is impossible to guarantee in the lab. We need a way to deal with some imperfection since in practice we will always encounter some noise and errors. However, instead of trying to perform a spot fix for this one issue, we will rederive the CHSH version of Bell’s inequality using the criterion of local causality as formulated by Bell.

Deriving the CHSH inequality mathematically is actually just a few steps and relatively simple. The meat is in the arguments of what it means. Suppose we set up the experiment exactly as how we have been setting it up, with Alice’s and Bob’s measurements of the spins at spacelike separation. Now, instead of looking at deterministic outcomes and as we have previously looked at before, we look at probability distributions over the outcomes: and . Deterministic outcomes won’t be forced on us because we won’t assume the underlying state is perfectly (anti) correlated.

Info

The actual outcomes, though, will always be a measurement of up or down. That’s just how a Stern-Gerlach experiment works for a spin-½ particle. So these and should be thought of, for our current purposes, simply as the Bernoulli distribution (a distribution over binary outcomes), e.g.

Let be any set of extra variables which would complete quantum mechanics in the sense of EPR. Let be any set of variables allowed by standard quantum mechanics to describe the experimental set up. The important thing here is that and are variables which fully describe a region of spacetime that blocks both Alice and Bob’s backward light cones, including blocking their overlaps, so that there is no other way for causal connections, in the sense of relativity theory, to “sneak in”. For an illustration, see Bell’s La Nouvelle Cuisine6.

The the joint probability distribution over the outcomes can be expressed as . There is a well known rule in probability called the chain rule which allows us to break down joint probabilities into conditional ones: (and one algebraic manipulation from here gets you to Bayes’ theorem). We apply the chain rule to our probability distribution to get:

It is here that we apply the condition of locality (as Bell defined it) and say and . This locality condition has been further analyzed and broken down into outcome independence, which is the dropping of the conditioning on (the outcome on the far side) that happened—and we would have dropped if we applied the chain rule the other way, and parameter independence, which is the dropping of the conditioning on from and on from . We arrive at the local causality condition, sometimes also called the factorizability condition or Bell locality condition:

Reichenbach's Principle of Common Causes

Imagine two events A and B often occur together so that . We are told of course that correlation does not mean causation so we can’t necessarily conclude that one is the cause of the other. Still, isn’t it a little peculiar if these events keep being correlated with no apparent link between them? Could it always be a coincidence? Reichenbach didn’t think it likely so he came up with his principle of common cause7. The principle says that when such a correlation exists, the two events have some common cause and when that common cause is accounted for, the probabilities become independent again, i.e.:

In such a way, this common cause “screens off” the apparent correlations. We can see how similar this principle is to Bell’s factorizability condition, with the role of being taken by . It is unclear if Bell knew of Reichenbach’s work; however, it does appear that Bell always considered factorizability to be a simple combination of the chain rule of probability and locality.

John S. Bell

The scientific attitude is that correlations cry out for explanation5.

Just like before, we consider the expectation values of the product of outcomes:

Here we have assumed that which is the Statistical Independence condition. Further, we have defined by taking the sum over first. All we need of those expressions is the fact that . Note here the difference from the earlier derivation of Bell’s inequality where were shown to take definite values .

The CHSH inequalities are just a few steps of algebra away. We perform algebraic manipulations on the following:

The first two equalities are simply plugging in variables and algebra. The first inequality comes because we used , (so we can carry the absolute value “through”), and the absolute value of an integral is smaller than the integral of the absolute value (convince yourself of this using sums). The final inequality is simply due to if . Note where we used the statistical independence condition so that becomes a single factor.

Finally we write the full CHSH inequality, while dropping the conditional for notational convenience:

What happens if we use the QM calculations for the singlet state? In that case, recall that we derived earlier . Let’s choose confined to a plane with the following settings (we give measurements relative to a):

Plugging these in, we arrive at:

Which obviously, and maximally (to Tsirelson’s bound) violates the CHSH inequality.

The nice thing about this derivation of the essential inequalities is several-fold. Firstly, we didn’t assume perfect anti-correlation anywhere (other than to show QM’s maximum violation) and so we were not forced to use outcome determinism as before. Secondly, the maximal quantum mechanical violation of the CHSH inequality is a factor of which gives experimentalists some space for error when confirming an inequality violation. Experimenters don’t need to achieve exactly a correlation of , they simply need to show that the value is above 2. Bell also spent time sharpening his arguments, mostly using the CHSH inequality rather than his original one, so that by the time of the Socks and Cuisine papers his arguments were well practiced and he understood what kind of objections people might have about his theorems. He significantly sharpened his local causality definition and related notions.

Footnotes

  1. Bell, J. S. “The Theory of Local Beables.” CERN preprint TH-2053, 1975. Reprinted in Speakable and Unspeakable in Quantum Mechanics, Cambridge University Press, 1987. ↩︎ ↩︎2

  2. Einstein, A., Podolsky, B., and Rosen, N. “Can Quantum-Mechanical Description of Physical Reality Be Considered Complete?” Physical Review, vol. 47, 1935, pp. 777–780. ↩︎

  3. Bohr, N. “Can Quantum-Mechanical Description of Physical Reality be Considered Complete?” Physical Review, vol. 48, 1935, pp. 696–702. ↩︎

  4. Bell, J. S. “On the Einstein Podolsky Rosen Paradox.” Physics Physique Физика, vol. 1, no. 3, 1964, pp. 195–200. Reprinted in Speakable and Unspeakable in Quantum Mechanics, Cambridge University Press, 1987. ↩︎ ↩︎2

  5. Bell, J. S. “Bertlmann’s Socks and the Nature of Reality.” Journal de Physique Colloques, vol. 42, no. C2, 1981, pp. C2-41–C2-62. Reprinted in Speakable and Unspeakable in Quantum Mechanics, Cambridge University Press, 1987. ↩︎ ↩︎2 ↩︎3

  6. Bell, J. S. “La Nouvelle Cuisine.” In Speakable and Unspeakable in Quantum Mechanics, 2nd ed., Cambridge University Press, 2004, pp. 232–248. ↩︎ ↩︎2

  7. Hitchcock, C., and Rédei, M. “Reichenbach’s Common Cause Principle.” The Stanford Encyclopedia of Philosophy, edited by E. N. Zalta, 2020. https://plato.stanford.edu/entries/physics-Rpcc/ ↩︎