Bell's Inequality: The Experiment That Ended the Debate on Hidden Variables

Einstein's objection, made testable

In 1935 Einstein, Podolsky and Rosen argued that quantum mechanics must be incomplete: if two particles are entangled, measuring one instantly determines the outcome for the other, however far apart they are, and they found that troubling enough to suspect that each particle secretly carries a predetermined, "hidden" set of answers fixed at the moment they were created — a local hidden-variable theory. For nearly thirty years this was treated as a matter of philosophical taste, because both pictures seemed to predict the same measurable statistics. In 1964, John Bell proved they do not. He derived an inequality that any local hidden-variable theory must obey — and showed that quantum mechanics predicts it can be broken.

The CHSH test: four correlations, one number

The version most experiments actually run is the CHSH inequality, formulated by Clauser, Horne, Shimony and Holt in 1969. Two distant experimenters, conventionally Alice and Bob, each independently choose one of two measurement settings for every entangled pair they receive — Alice picks a or a′, Bob picks b or b′ — and record a +1 or -1 outcome. Averaged over many trials, this gives four correlation values, combined into a single quantity S:

S = E(a,b) - E(a,b') + E(a',b) + E(a',b')

E(x,y) = average of (outcome_A * outcome_B) over many trials with
         Alice's setting x and Bob's setting y

any LOCAL HIDDEN-VARIABLE theory:      |S| <= 2      (Bell / CHSH bound)
quantum mechanics, entangled qubits:   |S| up to 2*sqrt(2) ≈ 2.828   (Tsirelson's bound)

The bound of 2 for local hidden variables follows from simple algebra: if each particle's outcome for every possible setting is already fixed in advance, independent of what the distant experimenter chooses, then no combination of four such fixed, bounded numbers can push S above 2. Quantum mechanics makes no such assumption, and for a maximally entangled pair — the singlet state — measured at angles 0°, 45°, 22.5° and 67.5°, its prediction reaches 2√2, comfortably above the classical ceiling and exactly at the maximum quantum mechanics itself permits.

Closing the loopholes

Alain Aspect's 1982 experiments were the first strong evidence of a CHSH violation, but early tests left two loopholes a determined skeptic could exploit. The detection loophole: if a device only detects a fraction of the particles, maybe the undetected ones would have restored the classical bound. The locality loophole: if the measurement settings are chosen too slowly, or the detectors are too close together, maybe a slower-than-light signal could sneak between them before the measurement completes. In 2015, three independent groups — Hensen et al. using electron spins in diamonds, and teams led by NIST and the University of Vienna using photons — closed both loopholes simultaneously in the same experiment for the first time, and all three confirmed the quantum prediction. The 2022 Nobel Prize in Physics went to Alain Aspect, John Clauser and Anton Zeilinger for this line of experimental work.

What the violation does and does not mean

A confirmed CHSH violation rules out local realism — the conjunction of "outcomes are predetermined" and "no influence travels faster than light between the measurements." It does not mean information can be sent faster than light: each individual measurement outcome is still perfectly random from the observer's point of view, and the correlation is only visible once the two lists of results are brought together and compared, which requires an ordinary classical channel no faster than light. What entanglement actually buys — secure quantum key distribution, quantum teleportation of a state, speedups in certain quantum algorithms — depends on exactly this kind of correlation that classical physics cannot reproduce, which is why Bell tests are not just a historical curiosity but the empirical foundation the entire field of quantum information sits on.

Frequently asked questions

Try it live

Everything above runs in your browser — open Bell's Inequality and change the parameters while it is running. Nothing is installed, nothing is uploaded, the whole model lives in one tab.