Bell's inequality — the game you can't win classically

When you measured the Bell state in Lesson 4.4 and saw the two qubits always agree, it was tempting to shrug and say “OK, correlation. Classical coins can do that too — if I glue two pennies together heads-up, they’ll always show the same side.”

Einstein made something like this argument. His intuition was: the Bell pair isn’t doing anything strange — the two qubits just “agreed in advance” which outcome they’d produce, and that agreement was locked in when the pair was created. Nothing non-classical is happening; we just don’t know the secret plan.

This is called a local hidden variable theory. In 1964, John Bell proved that it’s wrong. And the proof takes the form of a game.

The CHSH game

Alice and Bob play a cooperative game against a referee. The rules:

1. The referee gives Alice a random bit $x \in \{0, 1\}$ and Bob a random bit $y \in \{0, 1\}$. Alice and Bob cannot communicate during the game — they are in separate rooms.
2. Alice outputs a bit $a$. Bob outputs a bit $b$.
3. They win the round if $a \oplus b = x \wedge y$ (in words: “the XOR of their outputs equals the AND of their questions”).

Unpack the win condition:

• If $x = 0$ or $y = 0$: $x \wedge y = 0$, so they need $a = b$.
• If $x = y = 1$: $x \wedge y = 1$, so they need $a \neq b$.

Alice and Bob can agree on a strategy before the game starts, but cannot communicate once it begins.

The classical limit

How well can they do? Alice and Bob might try many strategies:

• Always output 0: wins on $(0,0), (0,1), (1,0)$, loses on $(1,1)$. 3 out of 4 = 75%.
• Output your question: Alice outputs $a = x$, Bob outputs $b = y$. Check: $a \oplus b = x \oplus y$. Wins iff $x \oplus y = x \wedge y$. That’s wins on $(0,0)$ (both zero) and $(1,1)$ (0 vs 1 — lose!)… 2 out of 4 = 50%. Worse.
• Randomize: try mixing strategies. Doesn’t help — the expected win rate is still bounded by 75%.

A theorem says: no classical strategy can win the CHSH game more than 75% of the time. This is Bell’s inequality in disguise, rephrased as a game. The inequality is:

$\text{(classical win rate)} \leq \frac{3}{4}$

It holds for any strategy whatsoever, deterministic or randomized, as long as Alice and Bob can’t communicate during the game and their outputs are determined by local information (their own question plus any pre-arranged “cheat sheet”).

The quantum surprise

Now suppose, before the game, Alice and Bob each take one qubit of a Bell pair $|\Phi^+\rangle$. During the game:

• Alice, on receiving $x$, measures her qubit in a specific basis (different basis depending on $x$) and outputs the result $a$.
• Bob, on receiving $y$, measures his qubit in a specific basis (different basis depending on $y$) and outputs the result $b$.

With the right choice of measurement angles — specifically, angles that differ by $\pi/4$ between the two cases — the quantum win rate is:

$\text{(quantum win rate)} = \cos^2\!\big(\tfrac{\pi}{8}\big) \approx 85.36\%$

This is the Tsirelson bound, and it’s the maximum possible quantum win rate. It’s strictly greater than $75\%$. Alice and Bob, sharing quantum entanglement, can beat any classical team — not because they’re cheating or communicating, but because the entanglement itself produces stronger-than-classical correlations.

Play the game

The widget below runs both strategies in parallel. Play a few rounds by hand to get a feel for it, then mash “×1000” and watch the numbers settle.

The classical bar converges to exactly $75\%$. The quantum bar converges to about $85.4\%$. The gap is $\approx 10$ percentage points, reliably, over thousands of rounds. There is no classical strategy anywhere that gets above $75\%$ — and yet the quantum team routinely sits in the $85\%$ range. This is not a fluke; it’s a direct experimental consequence of entanglement, and it rules out every possible classical explanation in one shot.

Why this matters

Before Bell’s theorem, it was reasonable to think that quantum weirdness was just our classical ignorance of hidden variables: “the qubits knew the answer all along, we just don’t have access to their internal state.” After Bell, that position is untenable. If the qubits had secret pre-existing answers (and respected the no-communication rule), they could never beat 75%. The fact that they do beat 75% — experimentally confirmed over and over since the 1980s — means one of those assumptions has to go. Either:

1. Locality is false. The qubits are somehow communicating instantaneously across arbitrary distances. But this communication can never be used to send a usable signal — only to produce correlations. Weird, but consistent with relativity.
2. Realism is false. The qubits don’t have definite values at all until measured. Measurement isn’t revealing a pre-existing property; it’s creating one.

Most physicists accept (2) — that measurement outcomes genuinely don’t exist before measurement. The 2022 Nobel Prize in Physics went to Aspect, Clauser, and Zeilinger for experiments that closed loophole after loophole in Bell-type tests, leaving essentially no classical escape route.

What’s next

You have now seen the single most important foundational result in quantum information. Everything from here on uses entanglement as a resource, and Bell’s inequality is what proves the resource is genuinely non-classical.

The final lesson in this module covers two more crucial limits of entanglement: the no-cloning theorem (you cannot copy an unknown quantum state) and the monogamy of entanglement (if two qubits are maximally entangled, a third can’t share in the entanglement). These limits turn out to be features, not bugs — they’re exactly what make quantum cryptography secure.