What is entanglement and what does it NOT allow?

Quantum entanglement is the most heavily verified correlation in physics — and the most heavily oversold word in quantum marketing. This entry of the Rosetta Quantum dictionary puts the two halves side by side: the measured record (a loophole-free Bell violation in 2015, a satellite link over 1,203 km in 2017, a 120-qubit entangled state in 2025 — sources below) and the no-go theorems that the marketing omits. The shortest version of the second half: entanglement, by itself, transmits exactly zero information (no-communication theorem; Ghirardi, Rimini & Weber, Lettere al Nuovo Cimento 27, 293, 1980). This is the sequel to our entry on what a qubit is and how a quantum computer works.

What is entanglement, exactly?

Entanglement is a joint state of two or more quantum systems that cannot be written as any list of separate, local states — the technical name is a non-separable state (Nielsen & Chuang, Quantum Computation and Quantum Information, ch. 2). Measured alone, each entangled particle behaves like a fair coin: irreducibly random. Measured jointly — and only when the two measurement records are later brought together — the outcomes show correlations stronger than any local classical model can produce.

That last clause is not rhetoric; it is a number. In 1964, John Bell proved that every local hidden-variable model obeys a hard bound — in the CHSH form, S ≤ 2 (Bell, Physics Physique Fizika 1, 195, 1964). Quantum mechanics predicts up to S = 2√2 ≈ 2.828, and no more — the Tsirelson bound (Tsirelson, Lett. Math. Phys. 4, 93, 1980). The window between 2 and 2.828 is where entanglement lives experimentally: every Bell test in the table below is a measurement of S inside that window.

One myth dies here: entangled particles do not «talk» to each other. The correlation sits in the joint statistics, and joint statistics are only visible after the two lab notebooks travel — classically, at light speed or slower — to the same table.

How do we know it is real?

Because the Bell window has been measured for five decades, with the escape hatches progressively welded shut. Freedman and Clauser produced the first laboratory violation in 1972 (Phys. Rev. Lett. 28, 938). Aspect's Orsay experiments added fast-switched analyzers in 1982 (Phys. Rev. Lett. 49, 1804). In 2015 the Delft group closed the detection and locality loopholes simultaneously: electron spins 1.3 km apart, S = 2.42 ± 0.20, p = 0.039 (Hensen et al., Nature 526, 682, 2015; arXiv:1508.05949). In 2017 the Micius satellite distributed entangled photon pairs to ground stations 1,203 km apart and measured a Bell violation of 2.37 ± 0.09 «under strict Einstein locality conditions» (Yin et al., Science 356, 1140, 2017; arXiv:1707.01339). The 2022 Nobel Prize in Physics went to Aspect, Clauser and Zeilinger «for experiments with entangled photons, establishing the violation of Bell inequalities and pioneering quantum information science» (Royal Swedish Academy of Sciences, October 2022).

The record has an engineering branch too. In December 2024, Quantinuum reported a GHZ state — the canonical many-party entangled state — across 50 logical qubits on its H2 system, with logical fidelities reported between 98.09% and 99.06%; that figure is vendor-reported in a blog post, not a peer-reviewed paper (The Quantum Insider, December 13, 2024). In October 2025, IBM reported the largest entangled state published to date: a 120-qubit GHZ state at fidelity 0.56(3), above the 0.5 line that certifies genuine multipartite entanglement — with the authors' own fine print in the same abstract: a 28% post-selection rate (IBM, «Big cats: entanglement in 120 qubits and beyond», arXiv:2510.09520). The caveat sat in the primary source, as it usually does — that pattern has its own entry in our ledger.

Why can't entanglement send a message?

Because the statistics at your end do not change when your partner measures — that is the no-communication theorem (Ghirardi, Rimini & Weber, Lettere al Nuovo Cimento 27, 293, 1980; textbook form in Nielsen & Chuang, ch. 2), and it is a theorem of quantum mechanics itself, not an engineering gap waiting for better hardware. Locally you see a fair coin before, during and after the remote measurement. The correlation only appears when the two measurement records are compared — over a classical channel, at light speed or slower. Relativity is not under threat from any row in the table above.

Teleportation is the cleanest exhibit. Moving one unknown qubit state from Alice to Bob consumes one entangled pair plus two classical bits sent the ordinary way (Bennett et al., Phys. Rev. Lett. 70, 1895, 1993); the first experiment followed in 1997 (Bouwmeester et al., Nature 390, 575). Withhold the two classical bits and Bob holds noise. The protocol's own accounting is the proof that entanglement is a resource you spend alongside communication — never a substitute for it.

What else does entanglement forbid?

Three more no-go results define the term as sharply as the Bell tests do. You cannot copy an unknown quantum state — the no-cloning theorem (Wootters & Zurek, Nature 299, 802, 1982; independently Dieks, Phys. Lett. A 92, 271, 1982) — which is why entanglement cannot be «amplified» like a radio signal and why quantum repeaters are hard engineering rather than a firmware update. Maximal entanglement cannot be shared freely: if two qubits are maximally entangled with each other, neither can be entangled with a third at all — the monogamy of entanglement (Coffman, Kundu & Wootters, Phys. Rev. A 61, 052306, 2000). And even quantum correlations have a ceiling: no experiment has ever produced S above Tsirelson's 2√2 ≈ 2.828 (Tsirelson, 1980).

What is entanglement actually useful for, then?

As a resource you spend, not a channel you talk through. Entanglement-based key distribution is a working protocol class — E91 (Ekert, Phys. Rev. Lett. 67, 661, 1991) — though on deployment our verdict stands: 0 of 6 security agencies with a published position recommend QKD as a primary control today. Inside a quantum computer, every two-qubit gate manufactures entanglement; that is where the resource does computational work, and it is why GHZ records are read as control benchmarks — IBM's own abstract frames its 120-qubit cat state as «a demonstration of good control and suppression of noise», not as a computation (arXiv:2510.09520). And entanglement does not repeal the ceilings: the problem classes quantum computing will not accelerate stay unaccelerated however many qubits are entangled. The Rosetta Quantum marker stands where it stood: verified advantaged solves attributable to entangled hardware, in our ledger to date, 0.

What we don't know

• Whether GHZ-scale control records translate into computational advantage on any real workload — our advantaged-solve marker remains at 0, and a fidelity-0.56 cat state is a control demo, not a solve (arXiv:2510.09520).
• The Quantinuum 50-logical-qubit figure is vendor-reported in a blog post; at our production date we found no peer-reviewed version to cite (The Quantum Insider, Dec 2024).
• Why nature saturates at Tsirelson's 2√2 and not at the higher no-signaling limit — a genuinely open foundations question, with «why not more?» still unanswered in the literature we cite.
• Whether satellite entanglement links scale into repeater networks with useful key rates — 1,203 km is a measured distribution record (Science 356, 2017), not a deployed network.

Rosetta Quantum publishes verdicts with raw, reproducible data. This is educational content, not a product claim.