What Is Quantum Entanglement?

Erwin Schrödinger needed a word. In 1935, writing in English for a Cambridge journal, he was describing something quantum mechanics kept doing that classical physics never did, and he settled on entanglement. He called it the characteristic trait of the theory. Ninety years on, that verdict has survived every test thrown at it. Quantum entanglement is now a measured laboratory resource, a line item in engineering budgets, and the reason physicists no longer believe the world can be both local and real.

What quantum entanglement actually is

Start with two qubits. Each one on its own can sit in a superposition of 0 and 1. Write down the state of the pair and you usually get a product: this qubit is doing one thing, that qubit is doing another, and the two descriptions stay separate.

Sometimes the state refuses to factorise. The best known example is a Bell state, an equal mix of both-zero and both-one with nothing else in it. Ask what the first qubit is doing by itself and quantum mechanics offers nothing better than a coin flip. Ask what the pair is doing and the answer is exact. The whole is sharply defined while its parts are not.

That inversion has no classical counterpart. It is the definition of entanglement, and everything else follows from it. Measure both qubits along the same axis and the outcomes always agree. Each outcome, taken alone, is random. The correlation is perfect. The individual results are not.

Why gloves in boxes do not explain it

There is an obvious classical story. Put a left glove in one box and a right glove in another. Post them to opposite ends of the country. Opening one box tells you instantly what is in the other, and no physics travels anywhere. The correlation was fixed when somebody packed the boxes.

Albert Einstein, Boris Podolsky and Nathan Rosen published a sharper version of that argument in 1935. If you can predict a distant measurement with certainty without touching the distant system, they reasoned, that measurement must already have a definite value. Quantum mechanics assigns no such value. So quantum mechanics is incomplete, and some hidden variable carries the missing answer.

This was a good argument. Einstein later described the alternative as spooky action at a distance, and he meant it as a complaint rather than a slogan. For nearly thirty years the dispute looked untestable. It read as a matter of taste about what a physical theory ought to contain.

Bell turned the argument into arithmetic

John Bell broke the deadlock in 1964 with four pages in an obscure journal. He asked what any theory with local hidden variables must predict, whatever those hidden variables turn out to be. The answer took the form of an inequality.

The usable version arrived in 1969 from John Clauser, Michael Horne, Abner Shimony and Richard Holt. Two observers each pick one of two measurement settings at random. Combine the four resulting correlations into a single number, S. Any local hidden-variable theory obeys S less than or equal to 2. Quantum mechanics predicts up to 2 times the square root of 2, roughly 2.828, at the right measurement angles. Boris Tsirelson proved in 1980 that quantum theory cannot do better than that.

The loopholes took forty-three years to close

Measuring S is easy. Measuring it in a way that admits no escape is hard. Three loopholes stood in the way.

Detection. Real detectors miss most photons, so the sample you analyse may not represent the sample you created. Locality. If one detector can signal the other before the second measurement finishes, the correlation needs no quantum explanation at all. Freedom of choice. If the measurement settings share a common cause with the source, the test is rigged before it starts.

Stuart Freedman and John Clauser reported the first violation in 1972. Alain Aspect and colleagues switched their analyser settings while the photons were in flight in 1982, going after locality. Both experiments left detection wide open, and everyone involved knew it.

Three groups closed all three loopholes at once in 2015. Ronald Hanson’s team in Delft entangled electron spins in two diamonds 1.3 kilometres apart and measured S = 2.42. Two photon experiments, from Vienna and from NIST, followed within months. Aspect, Clauser and Anton Zeilinger shared the 2022 Nobel Prize for the work that led there.

How hard the tests have been pushed since

Delft was not the end of it. In 2023 Andreas Wallraff’s group at ETH Zurich ran a loophole-free Bell test on superconducting qubits, the same technology inside Google and IBM processors.

Separating those qubits far enough meant building a cryogenic link 30 metres long. More than 1.3 tonnes of radiation shielding sits below 80 kelvin, around 90 kilograms of it below 50 millikelvin, and the qubits themselves run near 15 millikelvin. Across more than a million trials the team found S = 2.0747, plus or minus 0.0033. The violation is small. The p-value is smaller than 10 to the power minus 108.

Others attacked freedom of choice from a stranger direction. In 2018 a Vienna-led group used light from two quasars to set the detector angles. Any conspiracy between settings and source must therefore date back at least 7.8 billion years. That rules it out of 96 percent of the past light cone of the experiment, on the assumption of fair sampling. The same year, about 100,000 volunteers generated 97,347,490 binary choices by hand for the BIG Bell Test, feeding 13 experiments across five continents on the theory that human whim is the hardest thing to conspire with.

What entanglement cannot do

Perfect correlation across any distance sounds like a communication channel. It is not one, and the proof is short.

Whatever measurement Alice performs, the statistics Bob sees on his own qubit do not change. His outcomes are 50:50 before she measures and 50:50 afterwards. The correlation only shows up when you compare the two lists of results, and that comparison takes an ordinary classical message travelling no faster than light. That is the no-communication theorem. It is the reason entanglement sits alongside relativity instead of breaking it.

Why physicists call it a resource

The field stopped treating entanglement as a paradox and started treating it as inventory. It can be quantified, produced, degraded and spent.

It is also jealous. Monogamy of entanglement says that two qubits maximally entangled with each other can share none at all with a third. That is not an engineering nuisance. It is the security argument for quantum key distribution, because any eavesdropper who correlates with the signal necessarily weakens the correlation the legitimate parties are checking. It sits on a different footing from post-quantum cryptography, which defends ordinary classical channels with harder mathematics instead.

Entanglement is fragile. Any stray interaction leaks it into the environment, and that leakage is what decoherence means in practice.

The applications fall out of the accounting. Error correction spends entanglement to buy reliability. It holds one logical qubit non-locally across many physical ones, so no local error can reach what is encoded. Teleportation shifts states between modules. Entangled sensors beat the standard quantum limit. Vendors now benchmark processors largely on how much entanglement they can create and hold.

Entanglement at 13 teraelectronvolts

The effect is not confined to cold, quiet, heavily shielded apparatus. In 2024 the ATLAS collaboration reported entanglement between top quarks and antiquarks made in proton collisions at the Large Hadron Collider. It is the highest energy at which anyone has seen it.

A top quark decays before it can bind into a hadron, so its spin passes to its decay products, and the angle between the two charged leptons carries the spin correlation. ATLAS measured an entanglement marker D of minus 0.537, with a statistical uncertainty of 0.002 and a systematic uncertainty of 0.019, for top pairs of invariant mass between 340 and 380 GeV. That narrow window matters: averaged over all production angles, entanglement in this system survives only near threshold. Significance exceeded five standard deviations, and CMS has since seen it too. The Standard Model predicted the effect; the measurement confirmed it.

One caveat belongs on that number. ATLAS quotes D in a fiducial phase space defined with stable particles, chosen specifically to limit how much the result leans on Monte Carlo generators and parton-shower modelling. It is a measurement read out inside a simulation-dependent frame, not a raw count of events.

And across 420 kilometres of fibre

The practical frontier is entanglement between matter, not between photons in flight. Photons are easy to make and easy to lose. Atoms remember.

In August 2026 a team led by Jian-Wei Pan at the University of Science and Technology of China entangled two clouds of laser-cooled atoms through 420 kilometres of optical fibre, reported in Physical Review Letters. They ran the DLCZ protocol, converted the emitted photons into the telecom S band where fibre loss is lowest, and locked the relative phase across a link long enough that vibration and temperature drift would otherwise wreck it.

The headline is not the distance. It is that the entangling rate beat the repeaterless bound, the theoretical ceiling on how fast entanglement can cross a direct fibre channel with no memory in the middle. Beating that bound is the defining job of a quantum repeater. Memory-based networks now do something that direct transmission demonstrably cannot.

Established, contested, unproven

Established. Experiments violate Bell inequalities. The violations hold in loophole-free tests on photons, trapped ions, neutral atoms, nitrogen-vacancy centres in diamond and superconducting circuits. Local hidden-variable theories are dead. Entanglement is a countable resource with working applications in computing, cryptography and sensing.

Contested. What the violation means. Copenhagen, many-worlds, QBism and objective-collapse models all reproduce the same numbers and disagree sharply about what happens underneath. A Bell violation rules out a class of theories. It does not pick an interpretation, and physicists who agree on every measured value still argue about this one.

Unproven. Superdeterminism, the last surviving escape, holds that the settings and the particles were correlated from the outset. No experiment can rule it out, which is why most physicists treat it as unproductive rather than refuted. Also unproven: that any of this scales into a quantum internet. A 420-kilometre link is one link. A network is a different problem.