Bell inequalities, CHSH inequality, and the Tsirelson bound

Intuition Beginner

Quantum mechanics says that two particles can be entangled in such a way that measuring one immediately fixes what the other will show, even when the two particles are far apart. Einstein, Podolsky, and Rosen in 1935 found this so unsettling that they argued it could not be the final story. Their idea was that each particle must really carry hidden instructions, set at the moment the pair was created, which dictate the outcomes of every possible measurement. Quantum mechanics would then be an incomplete statistical description of a deeper deterministic theory. Einstein called the alternative "spooky action at a distance".

For nearly thirty years this was a philosophical dispute. Then in 1964, John Bell showed that the question can be settled by experiment. Bell took the EPR picture seriously and asked what statistical patterns it predicted, then compared with what quantum mechanics predicted. He proved a remarkable fact. Any theory in which (a) each particle carries pre-set instructions, and (b) no influence travels faster than light, must satisfy a specific numerical inequality. Quantum mechanics violates the inequality. The two pictures are not equivalent, and an experiment can choose between them.

The cleanest version of Bell's argument is the CHSH inequality, named for Clauser, Horne, Shimony, and Holt, who published it in 1969. The setup is this. Alice and Bob each receive one particle from an entangled pair. Alice chooses between two measurement settings, call them $a$ and $a^{\prime }$. Bob chooses between his two settings $b$ and $b^{\prime }$. Each measurement returns one of two outcomes, which we label $+1$ and $-1$. They repeat the experiment many times, keeping a running tally of the correlations between their outcomes for each pair of settings.

From the tally they compute four average products: $E(a,b)$, $E(a,b^{\prime })$, $E(a^{\prime },b)$, $E(a^{\prime },b^{\prime })$. They combine these into a single number called the CHSH quantity: $S=E(a,b)+E(a,b^{\prime })+E(a^{\prime },b)-E(a^{\prime },b^{\prime })$. The CHSH inequality says that if Alice's and Bob's outcomes are determined by hidden instructions carried by their particles, and no signal can travel from one side to the other during the experiment, then $S$ cannot exceed $2$ in absolute value. This is a hard ceiling that local hidden-variable theories cannot surpass.

Quantum mechanics breaks the ceiling. For the entangled Bell state and the right choice of measurement angles, quantum mechanics predicts $S=2\sqrt{2}$, about $2.828$. This is larger than $2$. The excess, sitting between $2$ and $2\sqrt{2}$, is the quantitative signature of quantum nonlocality. Boris Tsirelson proved in 1980 that quantum mechanics itself imposes the ceiling $2\sqrt{2}$. A hypothetical "super-quantum" theory could in principle reach $S=4$, but nature never has.

The concrete calculation uses the Bell state, the maximally entangled two-qubit state shared between Alice and Bob. The optimal CHSH measurement angles are $0$ degrees and $90$ degrees on Alice's side, and $45$ degrees and $135$ degrees on Bob's side. For these angles, the correlation between Alice and Bob obeys $E({\theta }_A,{\theta }_B)=\cos (2({\theta }_A-{\theta }_B))$. Plug in the four pairs: $\cos (2\cdot (-45))=\cos (90)/\sqrt{2}$. Working through, the four cosines yield $1/\sqrt{2}$, $1/\sqrt{2}$, $1/\sqrt{2}$, and $-1/\sqrt{2}$, so $S=4/\sqrt{2}=2\sqrt{2}$.

The experimental story closed the loop. Alain Aspect and collaborators in Orsay performed the first decisive Bell experiments in 1981 and 1982, observing CHSH values well above $2$ and consistent with the quantum prediction $2\sqrt{2}$. By 2015, three independent groups (Delft, Vienna, Boulder) had performed loophole-free Bell tests that simultaneously closed every conceivable evasion route. The 2022 Nobel Prize in Physics went to Aspect, Clauser, and Zeilinger for this body of work.

The takeaway is that nature is not described by any theory in which each particle carries pre-set local instructions. Entanglement is a fundamentally non-classical resource. Whatever description is correct, it must allow quantum mechanics's $2\sqrt{2}$ CHSH value, which is exactly what the experiments measure. Bell inequalities and their violations are the operational foundation of every modern quantum-information protocol that exploits entanglement, from device-independent cryptography to quantum random-number generation to the certification of quantum supremacy.

Visual Beginner

The first picture shows the CHSH measurement geometry. Alice's two measurement axes are at $0$ and $90$ degrees; Bob's two are at $45$ and $135$ degrees. The angle between each Alice setting and each Bob setting is $\pm 45$ degrees, except for the $(a^{\prime },b^{\prime })$ pair, where it is $135$ degrees. The four correlation cosines combine to give $S=2\sqrt{2}$.

The second picture compares the three regimes of CHSH values. Local hidden-variable theories are bounded by $2$. Quantum mechanics achieves up to $2\sqrt{2}$, the Tsirelson bound. Hypothetical no-signalling theories can in principle reach $4$, the algebraic maximum, realised by the Popescu-Rohrlich box. Nature lies inside the quantum region, never violating Tsirelson.

The third picture shows the historical experimental loophole-closure story. The original Aspect experiments in 1981-82 closed the locality loophole at light-speed level by switching analyzer settings during photon flight. The 1990s detection-efficiency experiments by Rowe et al. closed the detection loophole using ion traps with near-perfect detection. The 2015 Delft, Vienna, and Boulder experiments closed every loophole simultaneously.

Worked example Beginner

Alice and Bob share the Bell state $|{\Phi }^{+}⟩=(|00⟩+|11⟩)/\sqrt{2}$. Alice measures along axis $a$ at angle ${\theta }_A$ and Bob measures along axis $b$ at angle ${\theta }_B$, each obtaining outcome $+1$ or $-1$. The quantum-mechanical prediction for the correlation between Alice's and Bob's outcomes is $$E({\theta }_A,{\theta }_B)=\cos (2({\theta }_A-{\theta }_B)).$$ Compute the CHSH value $S=E(a,b)+E(a,b^{\prime })+E(a^{\prime },b)-E(a^{\prime },b^{\prime })$ for the standard choice ${\theta }_a=0$, ${\theta }_{a^{\prime }}=90$ degrees, ${\theta }_b=45$ degrees, ${\theta }_{b^{\prime }}=135$ degrees.

Step 1. Compute each correlation using $E=\cos (2\Delta \theta )$ with $\Delta \theta ={\theta }_A-{\theta }_B$.

For $(a,b)$: $\Delta \theta =0-45=-45$ degrees, so $2\Delta \theta =-90$ degrees, and $\cos (-90)=0$.

Wait, that gives zero. Recompute: the angle convention in spin-1/2 measurement is that the outcome correlation for axes at relative angle $\Delta$ is $-\cos (\Delta )$ for the singlet state and $+\cos (\Delta )$ for the $|{\Phi }^{+}⟩$ Bell state, where $\Delta$ is the actual angle between the spin measurement axes. For photon polarisation Bell-state measurements, the convention has the factor of $2$ built in because polarisation axes have period $180$ degrees. Use $E({\theta }_A,{\theta }_B)=\cos (2({\theta }_A-{\theta }_B))$ throughout for polarisation.

Step 2. Recompute with the polarisation convention. The four pairs:

$E(a,b)=\cos (2(0-45))=\cos (-90)$. This is $0$. The arithmetic gives the wrong CHSH value, suggesting a different angle assignment is needed for the polarisation convention. Standard convention for CHSH maximum: ${\theta }_a=0$, ${\theta }_{a^{\prime }}=45$, ${\theta }_b=22.5$, ${\theta }_{b^{\prime }}=67.5$ degrees for photon polarisation Bell-state measurements, giving $\Delta \theta =\pm 22.5$ degrees for the three good correlations and $\Delta \theta =67.5$ degrees for the bad one.

Step 3. With ${\theta }_a=0$, ${\theta }_{a^{\prime }}=45$, ${\theta }_b=22.5$, ${\theta }_{b^{\prime }}=67.5$ degrees, compute:

$E(a,b)=\cos (2(0-22.5))=\cos (-45)=1/\sqrt{2}\approx 0.707$.

$E(a,b^{\prime })=\cos (2(0-67.5))=\cos (-135)=-1/\sqrt{2}\approx -0.707$.

$E(a^{\prime },b)=\cos (2(45-22.5))=\cos (45)=1/\sqrt{2}\approx 0.707$.

$E(a^{\prime },b^{\prime })=\cos (2(45-67.5))=\cos (-45)=1/\sqrt{2}\approx 0.707$.

Step 4. Combine using the CHSH formula. With the sign convention used here ($+$ on three terms and $-$ on $E(a^{\prime },b^{\prime })$), and noting that the asymmetric assignment of correlations places the negative sign on the small-angle pair, the CHSH form requires careful sign tracking. Use the equivalent CHSH form $S^{\prime }=E(a,b)-E(a,b^{\prime })+E(a^{\prime },b)+E(a^{\prime },b^{\prime })$, giving $S^{\prime }=0.707+0.707+0.707+0.707=4/\sqrt{2}=2\sqrt{2}\approx 2.828$.

What this tells us: the quantum-mechanical CHSH value $2\sqrt{2}$ exceeds the classical bound $2$ by about $40$ percent, decisively distinguishing quantum mechanics from any local hidden-variable theory. The Bell state and these specific measurement angles saturate the Tsirelson bound, the maximum possible CHSH value allowed by quantum mechanics.

Check your understanding Beginner

Formal definition Intermediate+

Let ${\mathcal{H}}_A$ and ${\mathcal{H}}_B$ be finite-dimensional complex Hilbert spaces, and let $\rho \in \mathcal{S}({\mathcal{H}}_A\otimes {\mathcal{H}}_B)$ be a bipartite quantum state. A bipartite measurement scenario consists of two parties, Alice and Bob, who each choose one of two measurement settings labelled $\{a,a^{\prime }\}$ for Alice and $\{b,b^{\prime }\}$ for Bob. Each measurement is dichotomic, returning one of two outcomes labelled $\pm 1$.

Definition (CHSH observables). A CHSH measurement scenario on $\rho$ is specified by four Hermitian operators $A,A^{\prime }$ on ${\mathcal{H}}_A$ and $B,B^{\prime }$ on ${\mathcal{H}}_B$, each with spectrum contained in $\{-1,+1\}$ (equivalently, each squares to the identity: $A^2=A^{\prime 2}=I_A$ and $B^2=B^{\prime 2}=I_B$).

Definition (correlation). For a state $\rho$ and dichotomic observables $X$ on Alice and $Y$ on Bob, the correlation is $$E(X,Y):=\mathrm{tr}(\rho \cdot (X\otimes Y))\in [-1,+1].$$

Definition (CHSH operator and CHSH quantity). The CHSH operator on ${\mathcal{H}}_A\otimes {\mathcal{H}}_B$ is $$\mathcal{S}:=A\otimes B+A\otimes B^{\prime }+A^{\prime }\otimes B-A^{\prime }\otimes B^{\prime }.$$ The CHSH quantity in state $\rho$ is $$S(\rho ;A,A^{\prime },B,B^{\prime }):=\mathrm{tr}(\rho \cdot \mathcal{S})=E(A,B)+E(A,B^{\prime })+E(A^{\prime },B)-E(A^{\prime },B^{\prime }).$$

Definition (local hidden-variable model). A local hidden-variable (LHV) model for the correlations $\{E(X,Y){\}}_{X\in \{A,A^{\prime }\},Y\in \{B,B^{\prime }\}}$ consists of a probability space $(\Lambda ,\mu )$ together with functions $A,A^{\prime },B,B^{\prime }:\Lambda \to \{-1,+1\}$ such that, for each pair $(X,Y)\in \{A,A^{\prime }\}\times \{B,B^{\prime }\}$, $$E(X,Y)={\int }_{\Lambda }X(\lambda )Y(\lambda )d\mu (\lambda ).$$ The functions $A,A^{\prime }$ depend only on Alice's local data (and the hidden variable $\lambda$); similarly for $B,B^{\prime }$. The integration variable $\lambda$ models the "hidden state" carried by the entangled pair, sampled with distribution $\mu$.

Definition (Bell state and Pauli observables). The maximally entangled Bell state on two qubits is $$|{\Phi }^{+}⟩:=\frac{1}{\sqrt{2}}(|00⟩+|11⟩)\in {\mathbb{C}}^2\otimes {\mathbb{C}}^2.$$ The Pauli matrices ${\sigma }_x,{\sigma }_y,{\sigma }_z$ are the standard $2\times 2$ Hermitian unitaries with ${\sigma }_i^2=I$ and $\{{\sigma }_i,{\sigma }_j\}=2{\delta }_{ij}I$. The optimal CHSH observables on the Bell state are $$A={\sigma }_z,A^{\prime }={\sigma }_x,B=\frac{{\sigma }_z+{\sigma }_x}{\sqrt{2}},B^{\prime }=\frac{{\sigma }_z-{\sigma }_x}{\sqrt{2}}.$$ Each of these is Hermitian with $\pm 1$ spectrum, hence a valid dichotomic observable.

Counterexamples to common slips Intermediate+

• Dichotomic does not mean rank-one projector. A dichotomic observable $X$ with $X^2=I$ has two eigenvalues $\pm 1$ but can have any pair of eigenspace dimensions consistent with $\dim {\mathcal{H}}_A$. For example, the projector $|0⟩⟨0|-|1⟩⟨1|$ on a qubit is dichotomic with rank-one eigenspaces; on ${\mathbb{C}}^3$, the operator $\mathrm{diag}(1,1,-1)$ is dichotomic with eigenspace dimensions $2$ and $1$.

• The CHSH operator depends on which term carries the minus sign. Permuting which of the four products carries the minus sign gives an equivalent inequality up to relabelling of measurement settings. The four operators $\pm A\otimes B^{\prime }\mp A^{\prime }\otimes B^{\prime }$ etc. give the eight "CHSH-equivalent" inequalities, all with classical bound $2$ and Tsirelson bound $2\sqrt{2}$.

• Tsirelson is not the algebraic maximum. The four-term sum of operators with eigenvalues in $[-1,+1]$ can algebraically reach $4$ in absolute value. Tsirelson's bound $|⟨\mathcal{S}⟩|\le 2\sqrt{2}$ is strictly tighter than the algebraic bound $4$, because the operators are constrained to be tensor products of dichotomic Hermitians, not arbitrary $\pm 1$-bounded operators. The gap $[2\sqrt{2},4)$ is precisely the no-signalling-but-non-quantum region.

• Bell violation is not faster-than-light signalling. The marginal probability for Alice's outcome alone is independent of Bob's measurement setting (no-signalling theorem). Bell violation is a correlation phenomenon visible only when Alice and Bob compare results after the fact via a classical channel.

Key theorem with proof Intermediate+

Theorem (CHSH inequality, Clauser-Horne-Shimony-Holt 1969). Let $E(X,Y)$ for $X\in \{A,A^{\prime }\},Y\in \{B,B^{\prime }\}$ be correlations admitting a local hidden-variable model. Then $$|E(A,B)+E(A,B^{\prime })+E(A^{\prime },B)-E(A^{\prime },B^{\prime })|\le 2.$$

Proof. Let $(\Lambda ,\mu )$ be the hidden-variable space and let $A,A^{\prime },B,B^{\prime }:\Lambda \to \{-1,+1\}$ be the LHV outcome functions. Compute the integrand of $S$ at a fixed $\lambda$: $$A(\lambda )B(\lambda )+A(\lambda )B^{\prime }(\lambda )+A^{\prime }(\lambda )B(\lambda )-A^{\prime }(\lambda )B^{\prime }(\lambda )=A(\lambda )(B(\lambda )+B^{\prime }(\lambda ))+A^{\prime }(\lambda )(B(\lambda )-B^{\prime }(\lambda )).$$

Since $B(\lambda ),B^{\prime }(\lambda )\in \{-1,+1\}$, exactly one of $B(\lambda )+B^{\prime }(\lambda )$ and $B(\lambda )-B^{\prime }(\lambda )$ is zero, and the other is $\pm 2$. Hence $$A(\lambda )(B(\lambda )+B^{\prime }(\lambda ))+A^{\prime }(\lambda )(B(\lambda )-B^{\prime }(\lambda ))\in \{-2,+2\}$$ at each $\lambda$. Integrating against $\mu$, $$S={\int }_{\Lambda }[A(\lambda )(B+B^{\prime })(\lambda )+A^{\prime }(\lambda )(B-B^{\prime })(\lambda )]d\mu (\lambda ),$$ and the integrand is bounded pointwise by $2$ in absolute value. Hence $|S|\le {\int }_{\Lambda }2d\mu =2$.

$\square$

Bridge. The CHSH bound builds toward the Tsirelson bound proved in Master Theorem 1 below: the quantum CHSH value can exceed $2$ but cannot exceed $2\sqrt{2}$, and this is exactly the bridge between classical and quantum correlation structures. The foundational reason the bound is $2$ rather than $4$ is the pointwise factorisation $A(B+B^{\prime })+A^{\prime }(B-B^{\prime })$, which forces one of the two sums-of-Bob-outcomes to vanish when outcomes are deterministic; the relaxation to non-commuting quantum observables produces the $\sqrt{2}$ improvement. The central insight is that LHV correlations form a convex polytope (the local polytope) inside the cube $[-1,+1{]}^4$, with vertices given by the $2^4=16$ deterministic strategies, and CHSH is the facet inequality of this polytope. Bell inequalities generalise: each facet of each local polytope is a Bell inequality, and the full classification appears again in 12.17.07 device-independent protocols, where every Bell-type violation certifies a quantum advantage.

Exercises Intermediate+

Advanced results Master

The Tsirelson bound and the operator identity

Theorem 1 (Tsirelson bound, Cirel'son 1980). For any state $\rho$ on ${\mathcal{H}}_A\otimes {\mathcal{H}}_B$ and any dichotomic Hermitian observables $A,A^{\prime }$ on ${\mathcal{H}}_A$, $B,B^{\prime }$ on ${\mathcal{H}}_B$, the CHSH operator $\mathcal{S}=A\otimes B+A\otimes B^{\prime }+A^{\prime }\otimes B-A^{\prime }\otimes B^{\prime }$ satisfies $\Vert \mathcal{S}\Vert \le 2\sqrt{2}$, hence $|\mathrm{tr}(\rho \cdot \mathcal{S})|\le 2\sqrt{2}$.

The proof, as in Exercise 7 above, uses the operator identity $${\mathcal{S}}^2=4I-[A,A^{\prime }]\otimes [B,B^{\prime }]$$ combined with the commutator bound $\Vert [X,X^{\prime }]\Vert \le 2$ for any dichotomic Hermitian pair. The result is a universal upper bound on quantum CHSH, independent of dimension and of the choice of state. The bound is tight: Theorem 2 below shows that it is saturated by the Bell state and the canonical Pauli-axis observables.

Saturation by the Bell state and the canonical observables

Theorem 2 (Bell-state saturation). Let $|{\Phi }^{+}⟩=(|00⟩+|11⟩)/\sqrt{2}$ be the Bell state on ${\mathbb{C}}^2\otimes {\mathbb{C}}^2$. With observables $A={\sigma }_z$, $A^{\prime }={\sigma }_x$, $B=({\sigma }_z+{\sigma }_x)/\sqrt{2}$, $B^{\prime }=({\sigma }_z-{\sigma }_x)/\sqrt{2}$, the CHSH value satisfies $⟨{\Phi }^{+}|\mathcal{S}|{\Phi }^{+}⟩=2\sqrt{2}$.

Proof. Compute each of the four correlations. For the Bell state $|{\Phi }^{+}⟩$ and Pauli operators, $⟨{\Phi }^{+}|{\sigma }_z\otimes {\sigma }_z|{\Phi }^{+}⟩=1$, $⟨{\Phi }^{+}|{\sigma }_x\otimes {\sigma }_x|{\Phi }^{+}⟩=1$, $⟨{\Phi }^{+}|{\sigma }_y\otimes {\sigma }_y|{\Phi }^{+}⟩=-1$, and cross terms $⟨{\Phi }^{+}|{\sigma }_i\otimes {\sigma }_j|{\Phi }^{+}⟩=0$ for $i\ne j$.

Hence $⟨A\otimes B⟩=⟨{\sigma }_z\otimes ({\sigma }_z+{\sigma }_x)/\sqrt{2}⟩=(1+0)/\sqrt{2}=1/\sqrt{2}$. Similarly $⟨A\otimes B^{\prime }⟩=(1-0)/\sqrt{2}=1/\sqrt{2}$. And $⟨A^{\prime }\otimes B⟩=(0+1)/\sqrt{2}=1/\sqrt{2}$ and $⟨A^{\prime }\otimes B^{\prime }⟩=(0-1)/\sqrt{2}=-1/\sqrt{2}$.

Combine: $S=1/\sqrt{2}+1/\sqrt{2}+1/\sqrt{2}-(-1/\sqrt{2})=4/\sqrt{2}=2\sqrt{2}$. $\square$

This achieves the Tsirelson bound exactly. By Theorem 1, no quantum state and no quantum observables can do better.

Loophole-free Bell tests and the 2022 Nobel Prize

The path from theoretical inequality to definitive experimental refutation of local realism spanned five decades.

Aspect's experiments (1981-1982). Aspect, Grangier, and Roger (1982 Phys. Rev. Lett. 49, 91) measured CHSH on calcium-atom polarisation-entangled photon pairs, obtaining $S=2.697\pm 0.015$, in agreement with the quantum prediction and decisively violating the LHV bound. The follow-up Aspect-Dalibard-Roger experiment (1982 Phys. Rev. Lett. 49, 1804) added time-varying analyzers switching during photon flight to close the locality loophole at the level of light-speed-bounded influence.

The detection-efficiency challenge. The Aspect experiments left open the detection loophole: only a small fraction of photons were detected, and the fair-sampling assumption (that detected pairs are statistically representative) had to be invoked. Rowe et al. (2001 Nature 409, 791) used Be${}^{+}$ ions in a trap with $>90\%$ detection efficiency to close this loophole in an internal-state Bell test, at the cost of nearby-particle physics not having the spacelike-separation property.

The 2015 loophole-free trifecta. Three independent groups closed all loopholes simultaneously in 2015. Hensen et al. (2015 Nature 526, 682) at Delft used electron-spin states in two nitrogen-vacancy (NV) centres in diamond, separated by $1.3$ km, entangled via photon-mediated heralding; the experiment achieved CHSH $S=2.42\pm 0.20$ with spacelike separation enforced by the $1.3$ km distance and nanosecond-scale measurement timing. Giustina et al. (2015 Phys. Rev. Lett. 115, 250401) at Vienna used polarisation-entangled photon pairs with superconducting detectors of $\sim 75\%$ efficiency. Shalm et al. (2015 Phys. Rev. Lett. 115, 250402) at NIST Boulder used a similar photon-pair setup. All three obtained statistical rejection of local realism at $>8\sigma$ confidence.

The 2022 Nobel Prize. The Royal Swedish Academy awarded the 2022 Nobel Prize in Physics to John F. Clauser (UC Berkeley, for the 1972 Freedman-Clauser experiment and the CHSH derivation), Alain Aspect (Université Paris-Saclay, for the 1981-82 locality-closing experiments), and Anton Zeilinger (University of Vienna, for the 2015 photon loophole-free test and the long-distance teleportation programme) "for experiments with entangled photons, establishing the violation of Bell inequalities and pioneering quantum information science."

Generalisations: GHZ, KCBS, and multipartite Bell

Theorem 3 (GHZ paradox, Mermin 1990). The three-qubit GHZ state $|GHZ⟩=(|000⟩+|111⟩)/\sqrt{2}$ satisfies ${\sigma }_x\otimes {\sigma }_x\otimes {\sigma }_x|GHZ⟩=+|GHZ⟩$ and ${\sigma }_x\otimes {\sigma }_y\otimes {\sigma }_y|GHZ⟩={\sigma }_y\otimes {\sigma }_x\otimes {\sigma }_y|GHZ⟩={\sigma }_y\otimes {\sigma }_y\otimes {\sigma }_x|GHZ⟩=-|GHZ⟩$. No deterministic LHV value assignment can reproduce these four eigenvalue equations simultaneously.

Mermin (1990 Phys. Rev. Lett. 65, 1838) observed that the four GHZ eigenvalue equations imply a parity contradiction in any LHV model: multiplying all four gives ${\sigma }_x^4\otimes {\sigma }_x{\sigma }_y^2\otimes {\sigma }_x{\sigma }_y^2\cdot (\text{LHV signs})=-1$, but the LHV product is $(+1)(-1)(-1)(-1)=-1$ on the right-hand side while the operator product is $+I$ on the left, an algebraic inconsistency. The GHZ paradox achieves the Bell-like contradiction with deterministic eigenvalue equations, no statistical averaging needed, making it stronger than CHSH (which is statistical).

Theorem 4 (KCBS contextuality inequality, Klyachko-Can-Binicioglu-Shumovsky 2008). On a single spin-1 system, there exist five projective measurements $\{P_i{\}}_{i=1}^5$ with $P_i{P}_{i+1}=0$ (indices mod 5) such that the expectation ${\sum }_i⟨P_i⟩\le 2$ in any non-contextual hidden-variable model, while the quantum maximum is $\sqrt{5}\approx 2.236$.

Klyachko et al. (2008 Phys. Rev. Lett. 101, 020403) gave the simplest contextuality inequality for a single-particle system: no entanglement is required, but quantum mechanics still violates a "no-contextual-hidden-variable" bound. This generalises Bell to the single-system Kochen-Specker setting.

Magic-square games and pseudo-telepathy

The Mermin-Peres magic square (Exercise 8 above) is the prototype of quantum pseudo-telepathy: a cooperative game where the quantum strategy succeeds with certainty while no classical strategy can. Cleve-Hoyer-Toner-Watrous (2004 J. Comput. Syst. Sci. 69, 137) systematised pseudo-telepathy games and showed many examples. The general structure: a referee distributes inputs $(x,y)$ to Alice and Bob, asking each for an output; the winning condition depends jointly on $(x,y,a,b)$; the quantum strategy uses a shared entangled state plus local measurements; pseudo-telepathy means the winning probability is $1$ quantumly while the classical maximum is $<1$. The CHSH game has classical winning probability $3/4$ and quantum winning probability ${\cos }^2(\pi /8)=(1+1/\sqrt{2})/2\approx 0.854$, demonstrating that even non-pseudo-telepathy games can show quantum advantage.

Device-independent cryptography and the Ekert protocol

Theorem 5 (Ekert 1991). A loophole-free CHSH-violation experiment between Alice and Bob certifies that their shared bipartite state is genuinely entangled and that an eavesdropper cannot have classical pre-recorded knowledge of all measurement outcomes. This enables device-independent quantum key distribution: keys generated from CHSH-violating measurements are provably secure against any attack consistent with quantum mechanics and no-signalling, with no trust assumptions on the devices.

Ekert (1991 Phys. Rev. Lett. 67, 661) introduced quantum cryptography based on Bell's theorem (the "E91 protocol"), where the security argument reduces to the experimental measurement of CHSH violation. Mayers-Yao (1998 + 2003 Quant. Inf. Comp.) developed the device-independent framework in full generality. The modern device-independent QKD machinery (Vazirani-Vidick 2014 Phys. Rev. Lett. 113, 140501; Arnon-Friedman-Dupuis-Fawzi-Renner-Vidick 2018 Nat. Commun. 9, 459) provides composable security from CHSH violation alone.

Popescu-Rohrlich boxes and the no-signalling polytope

Theorem 6 (Popescu-Rohrlich 1994). There exist hypothetical two-input, two-output correlations $\{p(a,b|x,y){\}}_{a,b,x,y\in \{0,1\}}$ that respect the no-signalling condition (Alice's marginal is independent of Bob's input and vice versa) but achieve CHSH $S=4$, the algebraic maximum.

Popescu and Rohrlich (1994 Found. Phys. 24, 379) constructed the explicit PR-box correlation $$p(a,b|x,y)=\{\begin{array}{ll}1/2& \text{if }a\oplus b=x\cdot y,\\ 0& \text{otherwise.}\end{array}$$ The PR box is no-signalling (each marginal is uniform), but CHSH-saturates at $S=4$, exceeding Tsirelson by a factor of $\sqrt{2}$. Quantum mechanics does not realise PR boxes; the gap between $2\sqrt{2}$ and $4$ is a structural fact about quantum mechanics that no-signalling alone does not explain.

Theorem 7 (information causality principle, Pawlowski et al. 2009). No physical theory respecting no-signalling and a natural information-causality axiom can exceed CHSH $S=2\sqrt{2}$.

Pawlowski et al. (2009 Nature 461, 1101) showed that the information-causality principle — that the mutual information accessible to Bob is bounded by the classical communication received — is satisfied by quantum mechanics but violated by PR boxes. This identifies Tsirelson's bound as a consequence of an information-theoretic principle rather than just a property of Hilbert-space tensor products.

MIP* = RE and the structural depth of quantum correlations

Theorem 8 (MIP = RE, Ji-Natarajan-Vidick-Wright-Yuen 2020).* The class of decision problems solvable by multi-prover interactive proofs with shared entanglement (MIP) equals the class of recursively enumerable problems (RE). As a corollary, the set of finite-dimensional quantum correlations is not equal to its closure (Slofstra 2019 Forum Math. Pi 7, e1), and the Connes embedding conjecture from operator algebras is refuted.*

Ji, Natarajan, Vidick, Wright, and Yuen (2020 arXiv:2001.04383) proved that the entangled-prover interactive proof system can decide undecidable problems, by encoding the halting problem into a CHSH-like game. Slofstra (2019) had earlier shown that the set of bipartite quantum correlations is not closed: finite-dimensional limits of finite-dimensional quantum correlations can require infinite-dimensional Hilbert spaces to realise. This refutes Connes's 1976 embedding conjecture from operator algebras and identifies the quantum correlation set with a strictly larger object than initially expected. The depth of CHSH-type structures extends into logic and operator algebras.

Synthesis. Bell inequalities and the CHSH inequality are the foundational reason quantum mechanics cannot be a local hidden-variable theory. The central insight is that the classical bound $|S|\le 2$ follows from the pointwise factorisation $A(B+B^{\prime })+A^{\prime }(B-B^{\prime })$ in any LHV model, while the quantum bound $|S|\le 2\sqrt{2}$ follows from the operator identity ${\mathcal{S}}^2=4I-[A,A^{\prime }]\otimes [B,B^{\prime }]$ combined with the dichotomic-Hermitian commutator bound $\Vert [A,A^{\prime }]\Vert \le 2$. Putting these together identifies the Bell-state-with-Pauli-axis configuration as the unique saturating quantum strategy: Tsirelson's bound is tight, and the Bell state with optimal observables hits it exactly.

This is exactly the bridge between mathematical inequality and experimental physics. The Aspect 1981-82 experiments first established the inequality is violated; the 2015 Hensen-Giustina-Shalm loophole-free trifecta closed every conceivable evasion route; the 2022 Nobel Prize crystallised the body of work into recognition. The pattern recurs across the modern theory: GHZ paradoxes give deterministic LHV contradictions in tripartite settings, the magic-square game gives perfect pseudo-telepathy that no classical strategy can match, and KCBS gives a single-particle contextuality inequality with no entanglement requirement. The structural depth extends through Slofstra's 2019 demonstration that the quantum correlation set is not closed and the Ji-Natarajan-Vidick-Wright-Yuen MIP* = RE theorem identifying entangled-prover interactive proofs with the full recursively-enumerable class, refuting Connes's embedding conjecture.

The pattern generalises into operational quantum information. The Ekert 1991 device-independent QKD protocol identifies CHSH violation as a security certificate, and the modern Vazirani-Vidick framework gives composable security from CHSH alone. Popescu-Rohrlich boxes saturate the algebraic CHSH maximum $S=4$ while respecting no-signalling, identifying the Tsirelson gap $[2\sqrt{2},4)$ as a structural property of quantum mechanics that no-signalling alone cannot explain. The Pawlowski-et-al. 2009 information-causality principle identifies Tsirelson's bound as the consequence of an information-theoretic axiom. The structure builds toward 12.17.07 quantum teleportation, 12.17.04 pending EPR paradox, and the broader programme of certifying quantum advantage through Bell-type violations. Across all these layers, the CHSH inequality and its quantum saturation are the canonical operational signature of entanglement, and the Tsirelson bound is the canonical quantitative limit on how non-classical nature is permitted to be.

Full proof set Master

Proposition 1 (LHV bound for CHSH). In any local hidden-variable model with outcome functions $A,A^{\prime },B,B^{\prime }:\Lambda \to \{-1,+1\}$ and probability measure $\mu$, the CHSH quantity satisfies $|S|\le 2$.

Proof. At each $\lambda \in \Lambda$, compute the integrand $A(\lambda )B(\lambda )+A(\lambda )B^{\prime }(\lambda )+A^{\prime }(\lambda )B(\lambda )-A^{\prime }(\lambda )B^{\prime }(\lambda )$ by factoring: $$A(\lambda )(B(\lambda )+B^{\prime }(\lambda ))+A^{\prime }(\lambda )(B(\lambda )-B^{\prime }(\lambda )).$$ Since $B(\lambda ),B^{\prime }(\lambda )\in \{-1,+1\}$, exactly one of the two sums is $\pm 2$ and the other is $0$. The integrand is in $\{-2,+2\}$ at each $\lambda$. Integrating over $\mu$: $$|S|=\left|{\int }_{\Lambda }\right[A(B+B^{\prime })+A^{\prime }(B-B^{\prime })](\lambda )d\mu (\lambda )|\le {\int }_{\Lambda }2d\mu =2.\square$$

Proposition 2 (Tsirelson bound). For any state $\rho$ on ${\mathcal{H}}_A\otimes {\mathcal{H}}_B$ and any dichotomic Hermitian $A,A^{\prime },B,B^{\prime }$, $|\mathrm{tr}(\rho \cdot \mathcal{S})|\le 2\sqrt{2}$.

Proof. Use the operator identity (Exercise 3 corrected sign): ${\mathcal{S}}^2=4I-[A,A^{\prime }]\otimes [B,B^{\prime }]$. Take operator norm: $\Vert {\mathcal{S}}^2\Vert \le 4+\Vert [A,A^{\prime }]\Vert \cdot \Vert [B,B^{\prime }]\Vert \le 4+4=8$, using $\Vert [X,X^{\prime }]\Vert \le 2$ for dichotomic Hermitian pairs (Exercise 6).

Since $\mathcal{S}$ is Hermitian, $\Vert \mathcal{S}{\Vert }^2=\Vert {\mathcal{S}}^2\Vert$, hence $\Vert \mathcal{S}\Vert \le 2\sqrt{2}$. For any state, $|\mathrm{tr}(\rho \cdot \mathcal{S})|\le \Vert \mathcal{S}\Vert \cdot \Vert \rho {\Vert }_1=2\sqrt{2}$.

$\square$

Proposition 3 (saturation by Bell state). The Bell state $|{\Phi }^{+}⟩$ and observables $A={\sigma }_z$, $A^{\prime }={\sigma }_x$, $B=({\sigma }_z+{\sigma }_x)/\sqrt{2}$, $B^{\prime }=({\sigma }_z-{\sigma }_x)/\sqrt{2}$ achieve $⟨{\Phi }^{+}|\mathcal{S}|{\Phi }^{+}⟩=2\sqrt{2}$.

Proof. Direct computation using $⟨{\Phi }^{+}|{\sigma }_i\otimes {\sigma }_j|{\Phi }^{+}⟩={\delta }_{ix}{\delta }_{jx}+{\delta }_{iz}{\delta }_{jz}-{\delta }_{iy}{\delta }_{jy}$ (computed in Theorem 2 above; the Bell state $|{\Phi }^{+}⟩$ is the simultaneous $+1$ eigenstate of ${\sigma }_x\otimes {\sigma }_x$ and ${\sigma }_z\otimes {\sigma }_z$ and the $-1$ eigenstate of ${\sigma }_y\otimes {\sigma }_y$). Substituting and adding, $$⟨\mathcal{S}⟩=\frac{1}{\sqrt{2}}(1+1+1+1)=2\sqrt{2}.\square$$

Proposition 4 (Werner-state threshold). The Werner state ${\rho }_W=p|{\Phi }^{+}⟩⟨{\Phi }^{+}|+(1-p)I/4$ has CHSH value $S({\rho }_W)=2\sqrt{2}p$ with the optimal observables, hence violates CHSH iff $p>1/\sqrt{2}$.

Proof. By linearity, $\mathrm{tr}({\rho }_W\cdot \mathcal{S})=p\cdot \mathrm{tr}(|{\Phi }^{+}⟩⟨{\Phi }^{+}|\cdot \mathcal{S})+(1-p)\cdot \mathrm{tr}((I/4)\mathcal{S})$. The first term equals $p\cdot 2\sqrt{2}$ by Proposition 3. For the second, $\mathrm{tr}((I/4)\mathcal{S})=\mathrm{tr}(\mathcal{S})/4=0$ since $\mathcal{S}$ is a sum of tensor products of traceless Pauli operators (each ${\sigma }_z,{\sigma }_x$ has trace zero, and the basis $B,B^{\prime }$ chosen as linear combinations of traceless operators is also traceless). Hence $S({\rho }_W)=2\sqrt{2}p$. The threshold $p^{\ast }$ with $S=2$ is $p^{\ast }=1/\sqrt{2}$.

$\square$

Proposition 5 (commutator bound). For any dichotomic Hermitian $A$ with $A^2=I$ and any dichotomic Hermitian $A^{\prime }$ with $A^{\prime 2}=I$, $\Vert [A,A^{\prime }]\Vert \le 2$, with equality iff $\{A,A^{\prime }\}=0$ (anticommutation).

Proof. By the spectral theorem and $A^2=I$, $A$ has operator norm $\Vert A\Vert =1$, and similarly $\Vert A^{\prime }\Vert =1$. Then $$\Vert [A,A^{\prime }]\Vert =\Vert A{A}^{\prime }-A^{\prime }A\Vert \le 2\Vert A\Vert \Vert A^{\prime }\Vert =2.$$ Equality requires $\Vert A{A}^{\prime }\Vert =\Vert A\Vert \Vert A^{\prime }\Vert$ and $A{A}^{\prime }=-A^{\prime }A$. The first is automatic by submultiplicativity equality, and the second is anticommutation: $\{A,A^{\prime }\}=A{A}^{\prime }+A^{\prime }A=0$. For example, $A={\sigma }_z$ and $A^{\prime }={\sigma }_x$ anticommute and give $[{\sigma }_z,{\sigma }_x]=2i{\sigma }_y$ with norm $2$. $\square$

Connections Master

• Density matrix and pure/mixed states 12.17.01. The direct prerequisite. The Bell state $|{\Phi }^{+}⟩$ is a pure state of the bipartite system ${\mathbb{C}}^2\otimes {\mathbb{C}}^2$, but its reduced density matrix on either subsystem is ${\rho }_A={\rho }_B=I/2$, the maximally mixed qubit state. The pure-yet-locally-mixed structure is exactly the entanglement that enables CHSH violation, and the computation of expectation values $\mathrm{tr}(\rho \cdot (X\otimes Y))$ uses the density-matrix formalism developed in 12.17.01.

• Entanglement and Schmidt decomposition 12.17.02. The other direct prerequisite. The Bell state has Schmidt rank $2$ on ${\mathbb{C}}^2\otimes {\mathbb{C}}^2$ — the maximum possible — and Schmidt coefficients $(1/\sqrt{2},1/\sqrt{2})$. This maximally-entangled structure is what saturates the Tsirelson bound: less-entangled states give CHSH strictly less than $2\sqrt{2}$, and the Werner-state threshold $p^{\ast }=1/\sqrt{2}$ identifies the entanglement-strength regime where CHSH violation begins.

• Operators and observables 12.02.02. The framework prerequisite. The CHSH operator $\mathcal{S}$ is a Hermitian observable on ${\mathcal{H}}_A\otimes {\mathcal{H}}_B$, and the dichotomic constraints $A^2=A^{\prime 2}=I$ encode the $\pm 1$-outcome structure. The Pauli observables ${\sigma }_z,{\sigma }_x$ used in Theorem 2 are the canonical dichotomic Hermitian on a qubit, and the operator-norm bound $\Vert [A,A^{\prime }]\Vert \le 2$ is a spectral fact established in 12.02.02.

• Angular momentum and SU(2) 12.05.01. The spin-observable foundation. The Pauli observables ${\sigma }_x,{\sigma }_y,{\sigma }_z$ are the spin-1/2 generators, and the Bell state $|{\Phi }^{+}⟩$ is closely related to spin-singlet and spin-triplet states of two spin-1/2 systems. The optimal CHSH measurement axes correspond to spin measurements along the $z$ and $x$ axes for Alice and along the bisector axes $(z+x)/\sqrt{2}$ and $(z-x)/\sqrt{2}$ for Bob, a configuration motivated by the SU(2) geometry.

• EPR paradox 12.17.04 pending. The historical seed. Einstein, Podolsky, and Rosen (1935) framed the question that Bell's theorem ultimately answered: their thought experiment on perfectly-correlated entangled pairs argued that quantum mechanics must be incomplete, leaving room for hidden variables. Bell's 1964 inequality and the CHSH 1969 refinement converted the philosophical dispute into an experimentally testable inequality; the 1982 Aspect experiments and 2015 loophole-free tests delivered the empirical verdict.

• Quantum teleportation 12.17.07. The direct downstream protocol. Quantum teleportation consumes one Bell pair (one "ebit" of entanglement) to transmit one unknown qubit state from Alice to Bob, using only classical communication. The Bell-basis projective measurement on Alice's side and the Pauli-correction step on Bob's side are direct operational consequences of the Bell-state structure analysed here. Device-independent versions of teleportation (Lim-Christensen-Plenio 2010 + later) use CHSH violation to certify that the shared resource is genuinely a Bell pair.

• Kullback-Leibler divergence 26.11.04 pending. The classical-information shadow. The relative entropy $D(\rho \Vert \sigma )$ in quantum information theory specialises to the classical Kullback-Leibler divergence when $\rho$ and $\sigma$ are classical probability distributions. The Pawlowski-et-al. information-causality principle (Theorem 7 above) is formulated using mutual information, the classical-channel analogue of which is built from KL divergence; the proof that information causality implies the Tsirelson bound uses KL-divergence-based information bounds in the no-signalling polytope.

Historical and philosophical context Master

Einstein, Podolsky, and Rosen (1935) ^[EPR1935] in Physical Review 47 posed the foundational thought experiment: two particles entangled at a common origin and then separated to spacelike separation. EPR argued that since measurements on one particle can predict outcomes on the other with certainty (and without disturbing the other), each particle must have a definite "element of reality" prior to measurement, contradicting the quantum description in which neither particle has definite properties until measured. The EPR conclusion was that quantum mechanics is incomplete, and a deeper hidden-variable theory must underlie it. Bohr's 1935 reply ^[EPR1935] in the same volume defended quantum mechanics by arguing that the EPR notion of "element of reality" was too narrow.

For thirty years the dispute was metaphysical. Then John Bell (1964) ^[Bell1964] in the obscure journal Physics (volume 1) showed it is empirical. Bell took the EPR hidden-variable proposal seriously and derived statistical inequalities that any such theory must satisfy. He then showed quantum mechanics violates the inequalities. Bell's original paper used a different inequality from the modern CHSH form. Clauser, Horne, Shimony, and Holt (1969) ^[CHSH1969] in Physical Review Letters 23 reformulated Bell's argument as a testable bound on a particular four-correlation quantity, the CHSH inequality $|S|\le 2$.

Cirel'son (Tsirelson) (1980) ^{[Cirelson1980]} in Letters in Mathematical Physics 4 proved the quantum upper bound: no quantum state can give $S>2\sqrt{2}$. The proof used the operator identity ${\mathcal{S}}^2=4I-[A,A^{\prime }]\otimes [B,B^{\prime }]$ and the commutator bound for dichotomic Hermitians, in a remarkably compact two-page argument. Tsirelson's 1987 follow-up extended the bound to higher-dimensional Bell-type inequalities. The Cirel'son bound has stood for forty-five years as the quantum ceiling on CHSH.

Alain Aspect's experiments in Orsay (1981-1982) ^[Aspect1982] gave the first decisive confirmation of CHSH violation. Aspect-Grangier-Roger (1982 Phys. Rev. Lett. 49, 91) measured $S=2.697\pm 0.015$ on calcium polarisation-entangled photon pairs, decisively above the LHV bound and consistent with the quantum prediction. The Aspect-Dalibard-Roger 1982 follow-up (Phys. Rev. Lett. 49, 1804) added rapid analyzer switching during photon flight, closing the locality loophole at the level of light-speed-bounded influence. These experiments established the empirical foundation of nonlocality.

The 2015 loophole-free trifecta closed every remaining evasion route. Hensen et al. (2015 Nature 526, 682) ^[Hensen2015] at Delft used electron spins in nitrogen-vacancy centres in diamond separated by 1.3 kilometres, photon-mediated entanglement heralding, and nanosecond-scale measurement timing to enforce simultaneous closure of locality, detection, and free-choice loopholes. Giustina et al. (2015 Phys. Rev. Lett. 115, 250401) ^{[Giustina2015]} at Vienna and Shalm et al. (2015 Phys. Rev. Lett. 115, 250402) ^[Shalm2015] at NIST Boulder achieved the same with high-efficiency photon detection. All three obtained statistical rejection of local realism at $>8\sigma$ confidence.

The 2022 Nobel Prize in Physics was awarded to John F. Clauser, Alain Aspect, and Anton Zeilinger "for experiments with entangled photons, establishing the violation of Bell inequalities and pioneering quantum information science." The body of work covered the original 1972 Freedman-Clauser experiment, the 1981-82 Aspect locality-closing experiments, the 2015 loophole-free photon test, and the long-distance teleportation programme that demonstrated entanglement-based quantum communication at scale.

The theoretical lineage continues. Mermin (1990) ^[Mermin1990] introduced the GHZ paradox, demonstrating deterministic LHV contradictions for tripartite states. Ekert (1991) ^[Ekert1991] proposed the first Bell-inequality-based quantum cryptography protocol, founding device-independent quantum key distribution. Popescu and Rohrlich (1994) ^[PR1994] showed that the algebraic maximum $S=4$ is consistent with no-signalling, identifying the Tsirelson gap $[2\sqrt{2},4)$ as a structural property of quantum mechanics that no-signalling alone does not explain. Pawlowski et al. (2009) ^{[Pawlowski2009]} proved that the information-causality principle implies the Tsirelson bound, providing an information-theoretic derivation of the quantum CHSH ceiling.

The most recent structural developments are Slofstra (2019) ^{[Slofstra2019]} and Ji-Natarajan-Vidick-Wright-Yuen (2020) ^[JNVWY2020]. Slofstra proved that the set of quantum correlations is not closed: finite-dimensional limits can require infinite-dimensional Hilbert spaces. JNVWY proved MIP* = RE, demonstrating that the quantum interactive proof system can decide undecidable problems (refuting Connes's 1976 embedding conjecture from operator algebras). The Brunner-Cavalcanti-Pironio-Scarani-Wehner Reviews of Modern Physics 86 (2014) ^[BCPSW2014] survey gathers the modern theoretical structure of Bell nonlocality across all these strands. The CHSH inequality that Clauser-Horne-Shimony-Holt formulated in 1969 has grown into the foundational operational signature of quantum mechanics's break from classical local realism, an empirical verdict delivered by Aspect, Hensen, Giustina, Shalm, Zeilinger, and many others, and a structural depth extending into operator algebras, computational complexity, and quantum cryptography.

Bibliography Master

@article{EPR1935,
  author = {Einstein, Albert and Podolsky, Boris and Rosen, Nathan},
  title = {Can quantum-mechanical description of physical reality be considered complete?},
  journal = {Physical Review},
  volume = {47},
  pages = {777-780},
  year = {1935},
}

@article{Bell1964,
  author = {Bell, John S.},
  title = {On the {Einstein-Podolsky-Rosen} paradox},
  journal = {Physics},
  volume = {1},
  pages = {195-200},
  year = {1964},
}

@article{CHSH1969,
  author = {Clauser, John F. and Horne, Michael A. and Shimony, Abner and Holt, Richard A.},
  title = {Proposed experiment to test local hidden-variable theories},
  journal = {Physical Review Letters},
  volume = {23},
  pages = {880-884},
  year = {1969},
}

@article{FreedmanClauser1972,
  author = {Freedman, Stuart J. and Clauser, John F.},
  title = {Experimental test of local hidden-variable theories},
  journal = {Physical Review Letters},
  volume = {28},
  pages = {938-941},
  year = {1972},
}

@article{Cirelson1980,
  author = {Cirel'son, Boris S.},
  title = {Quantum generalizations of {Bell}'s inequality},
  journal = {Letters in Mathematical Physics},
  volume = {4},
  pages = {93-100},
  year = {1980},
}

@article{Aspect1982a,
  author = {Aspect, Alain and Grangier, Philippe and Roger, Gérard},
  title = {Experimental realization of {Einstein-Podolsky-Rosen-Bohm} {Gedankenexperiment}: A new violation of {Bell}'s inequalities},
  journal = {Physical Review Letters},
  volume = {49},
  pages = {91-94},
  year = {1982},
}

@article{Aspect1982b,
  author = {Aspect, Alain and Dalibard, Jean and Roger, Gérard},
  title = {Experimental test of {Bell}'s inequalities using time-varying analyzers},
  journal = {Physical Review Letters},
  volume = {49},
  pages = {1804-1807},
  year = {1982},
}

@article{Mermin1985,
  author = {Mermin, N. David},
  title = {Is the moon there when nobody looks? Reality and the quantum theory},
  journal = {Physics Today},
  volume = {38},
  number = {4},
  pages = {38-47},
  year = {1985},
}

@article{Mermin1990,
  author = {Mermin, N. David},
  title = {Extreme quantum entanglement in a superposition of macroscopically distinct states},
  journal = {Physical Review Letters},
  volume = {65},
  pages = {1838-1840},
  year = {1990},
}

@article{Ekert1991,
  author = {Ekert, Artur K.},
  title = {Quantum cryptography based on {Bell}'s theorem},
  journal = {Physical Review Letters},
  volume = {67},
  pages = {661-663},
  year = {1991},
}

@article{PR1994,
  author = {Popescu, Sandu and Rohrlich, Daniel},
  title = {Quantum nonlocality as an axiom},
  journal = {Foundations of Physics},
  volume = {24},
  pages = {379-385},
  year = {1994},
}

@article{KCBS2008,
  author = {Klyachko, Alexander A. and Can, M. Ali and Binicioglu, Sinem and Shumovsky, Alexander S.},
  title = {Simple test for hidden variables in spin-1 systems},
  journal = {Physical Review Letters},
  volume = {101},
  pages = {020403},
  year = {2008},
}

@article{Pawlowski2009,
  author = {Pawlowski, Marcin and Paterek, Tomasz and Kaszlikowski, Dagomir and Scarani, Valerio and Winter, Andreas and Żukowski, Marek},
  title = {Information causality as a physical principle},
  journal = {Nature},
  volume = {461},
  pages = {1101-1104},
  year = {2009},
}

@article{Hall2010,
  author = {Hall, Michael J. W.},
  title = {Local deterministic model of singlet state correlations based on relaxing measurement independence},
  journal = {Physical Review Letters},
  volume = {105},
  pages = {250404},
  year = {2010},
}

@article{BCPSW2014,
  author = {Brunner, Nicolas and Cavalcanti, Daniel and Pironio, Stefano and Scarani, Valerio and Wehner, Stephanie},
  title = {{Bell} nonlocality},
  journal = {Reviews of Modern Physics},
  volume = {86},
  pages = {419-478},
  year = {2014},
}

@article{Hensen2015,
  author = {Hensen, Bas and Bernien, H. and Dréau, A. E. and Reiserer, A. and Kalb, N. and Blok, M. S. and Ruitenberg, J. and Vermeulen, R. F. L. and Schouten, R. N. and Abellán, C. and Amaya, W. and Pruneri, V. and Mitchell, M. W. and Markham, M. and Twitchen, D. J. and Elkouss, D. and Wehner, S. and Taminiau, T. H. and Hanson, R.},
  title = {Loophole-free {Bell} inequality violation using electron spins separated by 1.3 kilometres},
  journal = {Nature},
  volume = {526},
  pages = {682-686},
  year = {2015},
}

@article{Giustina2015,
  author = {Giustina, Marissa and Versteegh, Marijn A. M. and Wengerowsky, Sören and Handsteiner, Johannes and Hochrainer, Armin and Phelan, Kevin and Steinlechner, Fabian and Kofler, Johannes and Larsson, Jan-Åke and Abellán, Carlos and Amaya, Waldimar and Pruneri, Valerio and Mitchell, Morgan W. and Beyer, Jörn and Gerrits, Thomas and Lita, Adriana E. and Shalm, Lynden K. and Nam, Sae Woo and Scheidl, Thomas and Ursin, Rupert and Wittmann, Bernhard and Zeilinger, Anton},
  title = {Significant-loophole-free test of {Bell}'s theorem with entangled photons},
  journal = {Physical Review Letters},
  volume = {115},
  pages = {250401},
  year = {2015},
}

@article{Shalm2015,
  author = {Shalm, Lynden K. and Meyer-Scott, Evan and Christensen, Bradley G. and Bierhorst, Peter and Wayne, Michael A. and Stevens, Martin J. and Gerrits, Thomas and Glancy, Scott and Hamel, Deny R. and Allman, Michael S. and Coakley, Kevin J. and Dyer, Shellee D. and Hodge, Carson and Lita, Adriana E. and Verma, Varun B. and Lambrocco, Camilla and Tortorici, Edward and Migdall, Alan L. and Zhang, Yanbao and Kumor, Daniel R. and Farr, William H. and Marsili, Francesco and Shaw, Matthew D. and Stern, Jeffrey A. and Abellán, Carlos and Amaya, Waldimar and Pruneri, Valerio and Jennewein, Thomas and Mitchell, Morgan W. and Kwiat, Paul G. and Bienfang, Joshua C. and Mirin, Richard P. and Knill, Emanuel and Nam, Sae Woo},
  title = {Strong loophole-free test of local realism},
  journal = {Physical Review Letters},
  volume = {115},
  pages = {250402},
  year = {2015},
}

@article{Handsteiner2017,
  author = {Handsteiner, Johannes and Friedman, Andrew S. and Rauch, Dominik and Gallicchio, Jason and Liu, Bo and Hosp, Hannes and Kofler, Johannes and Bricher, Daniel and Fink, Matthias and Leung, Calvin and Mark, Anthony and Nguyen, Hien T. and Sanders, Isabella and Steinlechner, Fabian and Ursin, Rupert and Wengerowsky, Sören and Guth, Alan H. and Kaiser, David I. and Scheidl, Thomas and Zeilinger, Anton},
  title = {Cosmic {Bell} test: Measurement settings from {Milky Way} stars},
  journal = {Physical Review Letters},
  volume = {118},
  pages = {060401},
  year = {2017},
}

@article{Slofstra2019,
  author = {Slofstra, William},
  title = {The set of quantum correlations is not closed},
  journal = {Forum of Mathematics, Pi},
  volume = {7},
  pages = {e1},
  year = {2019},
}

@article{JNVWY2020,
  author = {Ji, Zhengfeng and Natarajan, Anand and Vidick, Thomas and Wright, John and Yuen, Henry},
  title = {{MIP}* = {RE}},
  journal = {arXiv preprint arXiv:2001.04383},
  year = {2020},
}

@article{Rowe2001,
  author = {Rowe, M. A. and Kielpinski, D. and Meyer, V. and Sackett, C. A. and Itano, W. M. and Monroe, C. and Wineland, D. J.},
  title = {Experimental violation of a {Bell}'s inequality with efficient detection},
  journal = {Nature},
  volume = {409},
  pages = {791-794},
  year = {2001},
}

@article{CHTW2004,
  author = {Cleve, Richard and Hoyer, Peter and Toner, Benjamin and Watrous, John},
  title = {Consequences and limits of nonlocal strategies},
  journal = {Journal of Computer and System Sciences},
  volume = {69},
  pages = {137-178},
  year = {2004},
}

@book{NielsenChuang2010,
  author = {Nielsen, Michael A. and Chuang, Isaac L.},
  title = {Quantum Computation and Quantum Information},
  edition = {10th Anniversary},
  publisher = {Cambridge University Press},
  year = {2010},
}

@book{Susskind2014,
  author = {Susskind, Leonard and Friedman, Art},
  title = {Quantum Mechanics: The Theoretical Minimum},
  publisher = {Basic Books},
  year = {2014},
}

@book{Preskill,
  author = {Preskill, John},
  title = {Lecture Notes on Quantum Computation},
  publisher = {Caltech Ph219},
  note = {Available online at theory.caltech.edu/people/preskill/ph229},
}