The experiments

Fifty years of a determined sceptic being given nowhere left to stand — ending with detector settings taken from the light of quasars, so that any conspiracy would have had to be arranged before the Earth existed.

1972 – 202220 min

By the end you should be able to

Name the three loopholes and say why each is a legitimate objection
Explain what a loophole-free test has to do simultaneously, and why that is hard
Say what the results establish, and what remains logically open

The first person to actually build it was John Clauser, with Stuart Freedman at Berkeley, in 1972. He had trouble getting permission. He was told repeatedly that it was a waste of time, that the answer was known, and that it would damage his career — and it did damage his career; he had difficulty securing permanent positions for years afterwards, and eventually left academic physics. And like Bell, Clauser expected Einstein to be right. He has said since that he was very sad to find otherwise, and that he had rather hoped to overthrow quantum mechanics. Their source was calcium atoms excited to a state that decays through a two-photon cascade, the pair polarisation-correlated by conservation of angular momentum. They measured a violation of the inequality. And the experiment left two doors open.

Alain Aspect speaking, in later life. — Alain Aspect, whose Orsay experiments in 1981 and 1982 were the first to switch the analyser settings while the photons were already in flight — closing the loophole that the detectors might somehow have agreed in advance. He shared the 2022 Nobel Prize with John Clauser and Anton Zeilinger.

The three loopholes

This is where the story becomes good, because the objections are not obstruction. They are the right questions, and closing them took forty years. 1. Locality. If the detector settings are fixed in advance, or changed slowly, there is time for a light-speed signal to pass from one detector to the other during the measurement. Then ordinary communication could produce the correlations, and Bell's no-influence assumption simply fails. To close it: choose the settings so late, and separate the detectors so far, that no signal at light speed could carry the choice across in time. 2. Detection (fair sampling). Real detectors miss most of what arrives — well over 90% in the early experiments. So you measure a subset, and a local realist can reasonably suggest that whether a photon is detected depends on the very hidden variables at issue. Then the correlations are a selection effect. To close it: detect a large enough fraction that no such subset could account for the result. 3. Freedom of choice. Bell assumes the settings are independent of whatever produced the pair. What if some influence in the common past fixed both? This one cannot be closed, only pushed back — and pushing it back turns out to be worth doing spectacularly.

The experiment

Switching the analysers while the photons are in flight

Alain Aspect, with Philippe Grangier, Gérard Roger and Jean Dalibard · 1981 – 1982 · Institut d’Optique, Orsay

The question: Do entangled photons violate Bell’s inequality even when the detector settings are changed after the photons have left the source — so that no signal at light speed could carry the choice from one detector to the other?
The apparatus: A calcium-40 two-photon cascade source, pumped by lasers, producing polarisation-entangled pairs at a rate high enough for good statistics. The 1982 two-channel experiment used polarising beamsplitters measuring both outputs rather than discarding one. The switching experiment used acousto-optical switches — standing ultrasonic waves in water acting as diffraction gratings — to redirect each photon to one of two differently oriented polarisers, changing every 10 ns, with the detectors 6.5 m apart.

Theory predicted

Local realism requires S ≤ 2. Quantum mechanics predicts about 2.70 for this apparatus — below the ideal 2√2 because real sources, imperfect polarisers and finite collection angle all reduce the visibility.

They measured

Two-channel experiment: S = 2.697 ± 0.015, against a Bell bound of 2 — a violation by 46 standard deviations, and 95.4% of the maximum quantum mechanics permits. Switched-analyser experiment: S = 2.404 ± 0.080, a 5σ violation with the settings changing faster than light could cross between the wings.

How sure could they be? Light needs 21.7 ns to travel 6.5 m; the switches changed every 10 ns, so the setting on one side genuinely changed faster than the news could arrive at the other. Aspect was careful about the residual weakness: the switching was quasi-periodic rather than random, so a sufficiently determined sceptic could argue the pattern was in principle predictable. He said so in the paper.

Why it mattered

The result that made Bell tests mainstream physics rather than a curiosity. It substantially closed the locality loophole and largely removed the fair-sampling assumption, and its sheer statistical weight — 46σ — made the violation impossible to attribute to noise. Aspect shared the 2022 Nobel Prize with Clauser and Zeilinger.

Aspect measured 2.697 against a bound of 2 — 95% of the way to the quantum maximum.

Loading the detectors…

The bound is 2 for any local realist theory whatsoever. Aspect’s two-channel experiment reached 2.697 ± 0.015, which is 46 standard deviations above it. The shortfall from the ideal 2√2 is instrumental — real sources, imperfect polarisers, finite collection angle — not a hint of a smaller effect.

Aspect closed the locality loophole substantially, and was careful to say so rather than claim more. His switching was quasi-periodic rather than genuinely random, so a determined sceptic could argue the pattern was in principle predictable in advance. That was closed properly in 1998, in Innsbruck, by Weihs, Zeilinger and colleagues: genuinely random settings from quantum random number generators, with the detectors 400 metres apart and the whole choose-and-measure sequence completed well inside the light travel time. The detection loophole was closed separately, and later. First in 2001 with trapped beryllium ions at NIST, where essentially every event is registered — but the ions were only micrometres apart, so locality was wide open. Then with photons in 2013, once superconducting nanowire detectors reached efficiencies above 70%.

Why closing them separately is not enough

This is the subtle point, and it is why the story runs to 2015 rather than 1998. An experiment that closes locality while leaving detection open can be explained by a local model that exploits detection. An experiment that closes detection while leaving locality open can be explained by a local model that exploits locality. Two experiments, each answering one objection, do not add up to an answer to both — because a single local theory need only survive each experiment on its own terms. To refute local realism you must close them simultaneously, in one experiment. And the requirements pull directly against each other. Separating the detectors far enough to guarantee spacelike separation means the particles travel further, which means losing more of them, which reopens the detection loophole. Every metre of separation costs you efficiency. That tension is why it took another seventeen years.

It was achieved in 2015, by three groups independently within months of each other. Delft (Hensen and colleagues) used electron spins in nitrogen-vacancy centres in diamond, 1.3 km apart across the university campus, entangled by a scheme that swaps entanglement through photons detected at a midpoint. Detection of the spin state is near-perfect. The technique is desperately inefficient — they recorded 245 events in total — and that was enough for a violation. Vienna (Giustina and colleagues) and NIST (Shalm and colleagues) did it with photons, using the new high-efficiency superconducting detectors and fast random switching. All three closed locality and detection at the same time. There is no longer a version of local realism that survives any of them.

Settings chosen by quasars

Which leaves freedom of choice — and here the experiments become genuinely strange, in the best way. If you worry the settings might be correlated with whatever produced the pairs, take them from somewhere that source could not plausibly have influenced. 2017. A group used the colour of photons arriving from two distant stars to choose the detector settings in real time. Any common influence would have to predate the light's emission — hundreds of years ago. 2018. They went further and used quasars, one of them about 7.8 billion light-years away. For a superdeterministic conspiracy to explain the result, the correlation between that quasar's light and the properties of particles in an Austrian laboratory would have had to be arranged billions of years before the Earth formed. 2018, the BIG Bell Test. And if you distrust machine-generated randomness entirely, use people. Around 100,000 volunteers worldwide played an online game producing random bits, streamed in real time to twelve laboratories on five continents. Settings chosen by human whim. The inequality was violated.

You might think

Bell tests have closed every possible loophole, so nothing remains open.

Actually

Superdeterminism cannot be experimentally refuted, and honesty requires saying so. Any test involves choices — by a machine, a quasar or a human — and superdeterminism holds that those choices were themselves fixed by the same past that fixed the particle properties. The quasar experiments push the required conspiracy back billions of years; they cannot eliminate it, and no experiment can. A few serious physicists take this seriously, notably Gerard ’t Hooft. Most do not, on the grounds that a universe arranged to make experiments systematically misleading would undermine the possibility of doing science at all. That is an argument from consequences rather than from evidence, and it is worth noticing the difference. What can be said is that the alternative has been made extraordinarily expensive: it now requires a conspiracy older than the solar system.

1972Freedman and Clauser perform the first Bell test. Clauser is warned it will damage his career, and it does.
1982Aspect switches the analysers while the photons are in flight: S = 2.697 ± 0.015.
1998Innsbruck closes the locality loophole properly, with truly random settings at 400 m.
2001NIST closes the detection loophole with trapped ions — but at micrometre separation.
2015Delft, Vienna and NIST independently close locality and detection simultaneously.
2017–18Settings taken from starlight, then from quasars 7.8 billion light-years away.
2018The BIG Bell Test: 100,000 humans generate the settings for twelve laboratories.
2022Nobel Prize to Clauser, Aspect and Zeilinger.

Where this leaves us

The world is not locally real. That is now an experimental fact rather than a position. Which sounds like the end of Act 7, and is nearly the opposite — because within a decade of Aspect, the question changed completely. People stopped asking what entanglement means and started asking what it is for. If measuring one particle instantly fixes what you will find at the other, and no message can be sent, is there anything you can actually do with it? On the face of it, a correlation you cannot use to signal sounds useless. It is not. Between roughly 1984 and 1994 it became clear that entanglement is a resource — something you can spend to accomplish tasks that are provably impossible without it. Unbreakable key distribution, whose security rests on Bell violation rather than on any assumption about an eavesdropper's cleverness. Teleporting a quantum state. Computation that is not merely faster but differently structured. None of that required resolving what any of it means, and none of it has. The measurement problem is exactly where Act 6 left it. What changed is that the strangeness became useful.