What Bell Actually Proved
In 1964, John Bell asked a simple question: if quantum mechanics is wrong about entanglement — if there are hidden local variables that determine measurement outcomes in advance — what would the statistics look like?
The answer is an inequality. Any local hidden variable theory predicts:
\[S = |E(a,b) - E(a,b')| + |E(a',b) + E(a',b')| \leq 2\]
where \(E(a,b)\) is the correlation between Alice’s measurement in direction \(a\) and Bob’s in direction \(b\). This is the CHSH inequality (Clauser, Horne, Shimony, Holt, 1969).
Quantum mechanics predicts \(S = 2\sqrt{2} \approx 2.828\) for a maximally entangled Bell state measured at optimal angles. Aspect’s experiments (1982), and every loophole-free test since (Hensen et al. 2015, Giustina et al. 2015), confirm the quantum prediction.
\(S > 2\) is not a curiosity. It is a proof that the world is not locally real. The measurement outcomes were not pre-determined. The correlations cannot be explained by any information Alice and Bob shared before separating.
In QRL:
result = chsh_test(n_trials=10000)
print(result.S) # 2.828 ± 0.02
print(result.classical_bound) # 2.0
print(result.tsirelson_bound) # 2.828...
print(result.violates_classical) # Truetheory = theoretical_chsh()
print(theory) # 2.8284271247... exactly 2√2The CHSH parameter \(S\) is directly related to the entanglement of the state via concurrence \(C\):
\[S = 2\sqrt{1 + C^2}\]
For \(C = 0\) (no entanglement): \(S = 2\) — exactly at the classical bound. For \(C = 1\) (Bell state): \(S = 2\sqrt{2}\) — Tsirelson’s bound, the quantum maximum.
Entanglement is not a feature of quantum mechanics. It is the reason the CHSH inequality is violated. They are the same thing.
The Bell State
The canonical two-qubit entangled state is the \(|\Phi^+\rangle\) Bell state:
\[|\Phi^+\rangle = \frac{1}{\sqrt{2}}\left(|00\rangle + |11\rangle\right)\]
This is a superposition of both qubits being 0 and both being 1. There is no classical analogue. You cannot write it as a product \(|\psi_A\rangle \otimes |\psi_B\rangle\) — it is genuinely two-particle.
The optimal CHSH measurement angles are \(0°, 45°, 90°, 135°\): Alice measures at \(0°\) or \(90°\), Bob at \(45°\) or \(135°\). The quantum correlation \(E(a,b) = -\cos(\theta_{ab})\) where \(\theta_{ab}\) is the angle between measurement directions.
test = BellTest(n_trials=5000)
test.set_state("phi_plus")
test.set_angles(a=0, a_prime=90, b=45, b_prime=135)
result = test.run()
print(result.S) # ≈ 2.82
print(result.p_value) # ≪ 0.001 — statistically decisiveThe GHZ Paradox
The CHSH violation is statistical — you need many trials to see it clearly. The GHZ paradox (Greenberger, Horne, Zeilinger, 1989) achieves something stronger: it proves local realism is wrong in a single measurement.
Take three qubits in the GHZ state:
\[|\text{GHZ}\rangle = \frac{1}{\sqrt{2}}\left(|000\rangle + |111\rangle\right)\]
Now consider measuring each qubit in either the \(X\) or \(Y\) basis. The GHZ state satisfies:
\[XXX|\text{GHZ}\rangle = +|\text{GHZ}\rangle\] \[XYY|\text{GHZ}\rangle = -|\text{GHZ}\rangle\] \[YXY|\text{GHZ}\rangle = -|\text{GHZ}\rangle\] \[YYX|\text{GHZ}\rangle = -|\text{GHZ}\rangle\]
The product of all four left-hand sides is \(+1\) (each qubit appears twice in the Pauli operators). But the product of the right-hand sides is \(+1 \times (-1)^3 = -1\).
If outcomes were pre-determined, the product of all outcomes would be the same regardless of measurement order. Quantum mechanics gives a contradiction: the product must be both \(+1\) and \(-1\). Local realism is algebraically impossible — not just statistically improbable.
In QRL:
result = ghz_paradox_test(trials=10)Output:
Trial 1: XXX → outcomes (+1,+1,+1), product = +1 ✓ [QM predicts +1]
Trial 2: XYY → outcomes (+1,-1,+1), product = -1 ✓ [QM predicts -1]
Trial 3: YXY → outcomes (-1,+1,+1), product = -1 ✓ [QM predicts -1]
Trial 4: YYX → outcomes (+1,+1,-1), product = -1 ✓ [QM predicts -1]
...
Local realism requires product(XYY)·product(YXY)·product(YYX) = product(XXX).
Quantum mechanics: (−1)·(−1)·(−1) = −1 ≠ +1.
Logical contradiction. Local realism is refuted.Every single trial confirms the contradiction. This is not a matter of statistics.
Mermin’s Generalisation
The GHZ paradox generalises to \(n\) qubits. For the \(n\)-qubit GHZ state, the Mermin inequality gives:
\[M_{\text{quantum}} = 2^{n/2}\]
while the classical bound remains \(M \leq 2\) for all \(n\).
for n in [3, 5, 7, 10]:
result = mermin_test(n_qubits=n, trials=1000)
print(f"n={n}: M = {result.M:.2f} (classical ≤ 2.0)")Output:
n=3: M = 2.83 (classical ≤ 2.0)
n=5: M = 5.66 (classical ≤ 2.0)
n=7: M = 11.31 (classical ≤ 2.0)
n=10: M = 32.00 (classical ≤ 2.0)The quantum violation grows exponentially with the number of qubits. With 10 entangled qubits, the Mermin parameter is 16 times the classical bound. This exponential separation is what makes entanglement a computational resource — it is not just a philosophical curiosity about the nature of reality.
Hardware Verification
QRL was verified on real photonic hardware — the Quandela qpu:belenos processor — as part of the QPL 2026 paper. A dual-rail Bell state was generated using a heralded single-photon source and a linear optical beamsplitter network.
Results from 1000 shots: - 423 valid dual-rail coincidence events (42.3% yield) - 577 Hong-Ou-Mandel bunching events (57.7%) — expected for indistinguishable photons - CHSH parameter \(S = 2.61 \pm 0.08\) — violates the classical bound by more than 7 standard deviations
The Hong-Ou-Mandel bunching is itself a quantum effect: two indistinguishable photons entering a 50:50 beamsplitter always exit the same port. Classical waves do not do this. The 57.7% bunching rate is consistent with the theoretical 50% HOM effect plus detector efficiency losses.
result = hardware_bell_test()
print(result.S) # 2.61
print(result.shots) # 423 valid events
print(result.hom_fraction) # 0.577Why This Is the Foundation
The reason this matters beyond philosophy:
CHSH violation = entanglement = quantum computational advantage.
Every quantum speedup — Shor’s algorithm, Grover’s algorithm, quantum error correction, quantum sensing, molecular ground states — ultimately traces back to the same resource: the correlations that cannot exist classically. Bell’s theorem proves those correlations are real.
The domains Bell covers — chemistry, sensing, biology — all use entanglement as their working resource:
- Chemistry: orbital entanglement determines correlation energy. FCI wavefunction requires superposition of configurations.
- Sensing: GHZ states achieve Heisenberg scaling \(\Delta\theta \geq 1/n\) because they saturate the maximum \(F_Q = n^2\).
- Biology: chromophore entanglement in the FMO complex enables near-perfect energy transfer. Radical pair spin correlation enables magnetic navigation.
The physics layer — qrl.physics.bell and qrl.physics.ghz — is not one application among many. It is the reason the applications work.