Change the setting while the photons are in flight
A fixed polarizer leaves a long time for information about its setting to cross the apparatus. In 1982 Aspect, Dalibard and Roger made the measurement setting change while the photons were still in flight. Each arm held an optical switch that directed light toward one of two fixed polarizers. Changing the route changed the polarization question, without having to turn a mechanical part.
The aim was to make the setting change too late for a light-speed signal to reach the distant measurement in time. The key comparison was therefore a time against a distance: how often did the switches change, and how long would light take to travel between the two sides?
If two events are farther apart than light can travel during the interval between them, they are spacelike separated. In the simplified condition below, L is their distance, delta-t is their time difference in one inertial frame, and c is the speed of light. The paper compared switching intervals of roughly ten nanoseconds with a light-travel time of roughly forty nanoseconds between the two sides.
A real timing check has to define each event and include detector delays, path lengths and safety margins with their uncertainties. The moment an electrical command starts is not necessarily the moment a photon meets the chosen analyzer. Collapsing all these events into one convenient timestamp can reverse the conclusion.
Read the evidence: A. Aspect, J. Dalibard and G. Roger (1982), Experimental Test of Bell’s Inequalities Using Time-Varying Analyzers, Phys. Rev. Lett. 49, 1804–1807
The switching was quasiperiodic, not random
Sound waves in the optical switches diffracted the light into the selected paths. The two sides had independent drives at different frequencies. Their switching followed a quasiperiodic pattern: it varied, but followed predictable cycles. The authors explicitly called for a later experiment with more ideally random and complete switching.
Predictability matters even when the two measured switching sequences show little ordinary correlation: a model that knows the rule in advance can anticipate the next setting. Unpredictability alone is not enough either. A fast, unpredictable setting must also sit at the right place in spacetime: it needs to be chosen late enough that information about it cannot reach the distant outcome in time.
Read the evidence: A. Aspect, J. Dalibard and G. Roger (1982), Experimental Test of Bell’s Inequalities Using Time-Varying Analyzers, Phys. Rev. Lett. 49, 1804–1807Aspect (2022), Experimental tests of Bell’s inequalities: a first-hand account
Know which Bell statistic the paper used
The paper combined normalized count rates into a quantity whose local upper bound is zero, and reported 0.101 with an uncertainty of 0.020. That number cannot be compared with the CHSH statistic for two-valued outcomes, whose upper bound is two, even if the two share a symbol.
When reading, write next to each inequality the possible outcomes, the normalization and any auxiliary runs it relies on. Then note whether lost events count as an outcome, are ignored, or are handled by an extra assumption. A short list like this keeps a large violation in one convention from being mistaken for a small one in another.
Read the evidence: A. Aspect, J. Dalibard and G. Roger (1982), Experimental Test of Bell’s Inequalities Using Time-Varying Analyzers, Phys. Rev. Lett. 49, 1804–1807
Counts and corrections are part of the result
Part of the coincidence signal was background, estimated from delayed events. A correction like this needs its own measurement procedure. Subtracting background can be legitimate, but a reader should know what was subtracted and how its uncertainty carries into the final comparison.
Two possible changes answer different questions. Moving the delayed window tests the background estimate. Raising detection efficiency changes which emitted pairs are seen at all, and a check of the first says nothing about the second. Timing, optical throughput and count analysis are linked properties of one apparatus, but each answers its own question.
Read the evidence: A. Aspect, J. Dalibard and G. Roger (1982), Experimental Test of Bell’s Inequalities Using Time-Varying Analyzers, Phys. Rev. Lett. 49, 1804–1807
A real advance that left work for later
The switching experiment became one step in a longer programme of improving sources, analyzers and detection. It made the timing question experimentally concrete. Later tests combined less predictable choices with more efficient detection so that both requirements could be addressed in the same run.
The timing limits a possible causal route between the apparatus arms. It still leaves the detection assumptions and the predictability of the switching to be assessed separately. Quantum correlations observed in either arrangement provide no controllable faster-than-light message.
Read the evidence: Aspect (2022), Experimental tests of Bell’s inequalities: a first-hand account