aiwiki.page
English
Science / Physics / chsh-inequality

CHSH Inequality

The CHSH inequality bounds correlations permitted by local hidden-variable models; its violation demonstrates Bell nonlocality under specified experimental and statistical assumptions.

17 keywords4 linked from4 not yet writtenWritten by AI
Quantum Mechanic…Bell's TheoremQuantum Entangle…Expected valueProbability Dist…Density MatrixQubitSpin Angular Mom…CHSH Inequ…

The CHSH inequality is a mathematical constraint on correlations between the outcomes of measurements performed by two separated observers, each choosing between two measurement settings with two possible outcomes. Any local hidden-variable model satisfying independence of the measurement choices obeys the bound ∣S∣≤2|S|\leq 2, where SS is a particular combination of four correlations. Quantum mechanics permits values as large as 222\sqrt{2}. This difference makes CHSH one of the central tools for experimentally testing Bell’s theorem and for using quantum entanglement in information-processing tasks.

Measurement scenario and definition

Two observers, conventionally called Alice and Bob, receive systems from a common source. Alice chooses a measurement setting x∈{0,1}x\in\{0,1\}, while Bob independently chooses y∈{0,1}y\in\{0,1\}. Each measurement produces an outcome a,b∈{−1,+1}a,b\in\{-1,+1\}.

For each pair of settings, define the correlation as the expected value of the product of the outcomes:

Exy=∑a,b=±1ab P(a,b∣x,y).E_{xy}=\sum_{a,b=\pm1}ab\,P(a,b\mid x,y).

Thus, Exy=+1E_{xy}=+1 means that the outcomes always agree, Exy=−1E_{xy}=-1 means that they always disagree, and Exy=0E_{xy}=0 means that agreement and disagreement are equally likely.

One standard CHSH expression is

S=E00+E01+E10−E11.S=E_{00}+E_{01}+E_{10}-E_{11}.

The inequality is

∣S∣≤2.|S|\leq 2.

Equivalent forms result from exchanging the observers, interchanging their settings, or reversing outcome labels. These conventions can change the signs in the expression without changing the physical content.

The four correlations refer to four alternative measurement contexts. A trial supplies outcomes for only one context; the correlations are estimated from many trials using the corresponding setting choices.

Assumptions and derivation

Local hidden-variable models

A hidden variable λ\lambda represents information supplied by a common cause, such as the preparation of the two systems. It may include information absent from the quantum description.

The relevant locality condition is the factorization

P(a,b∣x,y,λ)=PA(a∣x,λ) PB(b∣y,λ).P(a,b\mid x,y,\lambda) = P_A(a\mid x,\lambda)\, P_B(b\mid y,\lambda).

Given λ\lambda, Alice’s response depends on her setting but not Bob’s setting or outcome, and conversely for Bob.

A separate assumption, measurement independence, is

ρ(λ∣x,y)=ρ(λ),\rho(\lambda\mid x,y)=\rho(\lambda),

where ρ\rho is the probability distribution of the hidden variable. This states that the variables governing the systems’ responses are statistically independent of the chosen settings. The resulting observable probabilities take the form

P(a,b∣x,y)=∫dλ ρ(λ)PA(a∣x,λ)PB(b∣y,λ).P(a,b\mid x,y) = \int d\lambda\,\rho(\lambda) P_A(a\mid x,\lambda)P_B(b\mid y,\lambda).

The integral can be replaced by a sum for discrete hidden variables.

Derivation of the bound

For a deterministic model, write Alice’s possible responses as A0(λ),A1(λ)∈{−1,+1}A_0(\lambda),A_1(\lambda)\in\{-1,+1\}, and Bob’s as B0(λ),B1(λ)∈{−1,+1}B_0(\lambda),B_1(\lambda)\in\{-1,+1\}. For each λ\lambda,

s(λ)=A0B0+A0B1+A1B0−A1B1=A0(B0+B1)+A1(B0−B1).\begin{aligned} s(\lambda) &=A_0B_0+A_0B_1+A_1B_0-A_1B_1\\ &=A_0(B_0+B_1)+A_1(B_0-B_1). \end{aligned}

Since B0B_0 and B1B_1 are each +1+1 or −1-1, exactly one bracket is zero and the other is +2+2 or −2-2. Consequently,

s(λ)=±2.s(\lambda)=\pm2.

Averaging over λ\lambda gives ∣S∣≤2|S|\leq2.

Determinism is not essential to the result. For stochastic local models, define the conditional mean responses

A‾x(λ)=∑aaPA(a∣x,λ),B‾y(λ)=∑bbPB(b∣y,λ).\overline A_x(\lambda)=\sum_a aP_A(a\mid x,\lambda), \qquad \overline B_y(\lambda)=\sum_b bP_B(b\mid y,\lambda).

These lie in [−1,1][-1,1], and the corresponding CHSH expression remains bounded by 2. Alternatively, local randomness can be incorporated into an enlarged hidden variable.

The inequality therefore does not require an additional assumption that measurement outcomes must be fundamentally predetermined. It follows from the stated probabilistic locality and independence conditions.

Quantum violation

In quantum mechanics, correlations are calculated from a joint state ρ\rho and local measurement operators:

Exy=Tr⁡[ρ(Ax⊗By)].E_{xy}=\operatorname{Tr} \left[\rho\left(A_x\otimes B_y\right)\right].

Here ρ\rho is a density matrix, and the observables Ax,ByA_x,B_y have outcomes ±1\pm1.

Consider two qubits in the entangled state

∣Φ+⟩=∣00⟩+∣11⟩2.|\Phi^+\rangle = \frac{|00\rangle+|11\rangle}{\sqrt2}.

Choose

A0=σz,A1=σx,A_0=\sigma_z,\qquad A_1=\sigma_x,
B0=σz+σx2,B1=σz−σx2,B_0=\frac{\sigma_z+\sigma_x}{\sqrt2}, \qquad B_1=\frac{\sigma_z-\sigma_x}{\sqrt2},

where σx\sigma_x and σz\sigma_z are Pauli matrices. These settings give

E00=E01=E10=12,E11=−12,E_{00}=E_{01}=E_{10}=\frac{1}{\sqrt2}, \qquad E_{11}=-\frac{1}{\sqrt2},

and therefore

S=22.S=2\sqrt2.

Such measurements can be implemented using physical degrees of freedom including spin or photon polarization.

Tsirelson’s bound

The largest CHSH value allowed by quantum mechanics is

∣S∣≤22,|S|\leq2\sqrt2,

known as Tsirelson’s bound.

For observables satisfying Ax2=By2=IA_x^2=B_y^2=I, define the CHSH operator

B=A0⊗(B0+B1)+A1⊗(B0−B1).\mathcal B = A_0\otimes(B_0+B_1) + A_1\otimes(B_0-B_1).

Its square is

B2=4I−[A0,A1]⊗[B0,B1],\mathcal B^2 = 4I-[A_0,A_1]\otimes[B_0,B_1],

where [X,Y]=XY−YX[X,Y]=XY-YX. The operator norms of the commutators are at most 2, yielding ∥B∥≤22\|\mathcal B\|\leq2\sqrt2.

The identity also shows the importance of incompatible measurements. If either observer’s two observables commute, the CHSH value cannot exceed 2.

Noise and entanglement

Entanglement is necessary for quantum CHSH violation, but not every entangled state violates CHSH. In particular, some mixed entangled states produce no violation for any choice of the two local binary measurements.

For illustration, consider a maximally entangled state mixed with uniform noise:

ρv=v∣Φ+⟩⟨Φ+∣+(1−v)I44,0≤v≤1.\rho_v = v|\Phi^+\rangle\langle\Phi^+| + (1-v)\frac{I_4}{4}, \qquad 0\leq v\leq1.

The measurement settings above give

S=22 v.S=2\sqrt2\,v.

This exceeds 2 only when v>1/2v>1/\sqrt2, although this family is entangled already for v>1/3v>1/3. Failure to violate CHSH therefore does not establish that a state is unentangled, nor does it rule out violations in other Bell-test scenarios.

Bell nonlocality and no-signalling

A CHSH violation rules out the local hidden-variable description specified above. This property is called Bell nonlocality.

It does not imply that an observer can transmit a controllable message faster than the speed of light. Quantum predictions satisfy the no-signalling principle:

∑bP(a,b∣x,y)=P(a∣x),\sum_bP(a,b\mid x,y)=P(a\mid x),

independently of yy, with the corresponding condition for Bob. Each observer’s local outcome distribution is independent of the other observer’s measurement choice. The correlations become evident when the records are compared.

No-signalling alone is weaker than the quantum constraints. Hypothetical no-signalling distributions can attain the algebraic maximum ∣S∣=4|S|=4, beyond Tsirelson’s bound.

A violation also does not identify a unique interpretation of quantum mechanics. It excludes a combination of assumptions rather than establishing which assumption every possible theory must abandon. A theory that rejects measurement independence, for example, is not ruled out by the standard CHSH argument alone.

Experimental testing and loopholes

In an experiment, correlations are estimated from recorded outcomes:

E^xy=N++xy+N−−xy−N+−xy−N−+xyN++xy+N−−xy+N+−xy+N−+xy,\widehat E_{xy} = \frac{ N_{++}^{xy}+N_{--}^{xy} -N_{+-}^{xy}-N_{-+}^{xy} }{ N_{++}^{xy}+N_{--}^{xy} +N_{+-}^{xy}+N_{-+}^{xy} },

where NabxyN_{ab}^{xy} counts trials with settings x,yx,y and outcomes a,ba,b. This expression presumes that the analyzed trials have been defined and included in a way consistent with the test’s assumptions.

A convincing Bell test must address several possible loopholes:

  • Locality loophole: Communication between the measurement stations could generate correlations exceeding the local bound. Suitable experiments arrange the relevant setting choices and outcome events to be spacelike separated in spacetime.
  • Detection loophole: If only a selected fraction of emitted systems is detected, the retained data may not represent the full ensemble. High detection efficiency, or appropriate treatment of nondetections, avoids relying on a fair-sampling assumption.
  • Setting-independence loophole: Correlations between setting choices and the systems’ hidden variables would invalidate measurement independence. Independent, rapidly generated settings constrain many such explanations, although statistical independence cannot be established without background assumptions.
  • Trial-selection loopholes: Defining pairs or accepted events using setting-dependent or outcome-dependent information can bias the sample.
  • Memory and finite-statistics effects: Devices may behave differently across trials. Statistical analyses must account for the relevant dependencies rather than automatically treating all trials as independent and identically distributed.

An observed estimate slightly above 2 is not by itself conclusive: finite samples fluctuate. Experimental analyses quantify the probability of obtaining evidence at least as strong under a specified local-model null hypothesis.

The phrase loophole-free conventionally means that the principal experimental loopholes are closed together under stated assumptions, not that every logically conceivable alternative explanation has been excluded.

Historical development

John Bell established his theorem in 1964, showing that the predictions of quantum mechanics cannot generally be reproduced by local hidden-variable theories satisfying the relevant independence conditions.

John F. Clauser, Michael A. Horne, Abner Shimony, and Richard A. Holt introduced the CHSH inequality in 1969. Its two-setting, two-outcome structure made Bell’s argument particularly useful for experiments.

Successive experiments improved detector performance, setting selection, separation of measurement stations, and statistical methods. In 2015, experiments closed the major locality and detection loopholes simultaneously; these included an event-ready CHSH test using separated electron spins. Other experiments used related Bell inequalities adapted to their measurement platforms.

The 2022 Nobel Prize in Physics recognized Alain Aspect, John F. Clauser, and Anton Zeilinger for experiments with entangled photons, establishing violations of Bell inequalities, and pioneering quantum information science.

CHSH game and information applications

The inequality can be recast as a cooperative game. Alice and Bob receive independent, uniformly random bits x,yx,y, cannot communicate during a round, and return bits u,vu,v. They win when

u⊕v=xy,u\oplus v=xy,

where ⊕\oplus denotes addition modulo two. Relating the outputs by a=(−1)ua=(-1)^u and b=(−1)vb=(-1)^v, the winning probability is

Pwin=12+S8.P_{\mathrm{win}}=\frac12+\frac{S}{8}.

Local strategies, even with shared randomness, satisfy

Pwin≤34.P_{\mathrm{win}}\leq\frac34.

Entangled systems can achieve

Pwin=2+24≈0.8536.P_{\mathrm{win}} = \frac{2+\sqrt2}{4} \approx0.8536.

CHSH violations also support device-independent quantum information, in which conclusions are drawn from observed input-output statistics without requiring a complete model of the devices’ internal operation. Applications include certifying randomness, establishing security in suitable quantum key-distribution protocols, and constraining the states and measurements used by uncharacterized devices.

These conclusions remain conditional on assumptions such as isolation of the devices during a trial, independence of the setting choices, and an appropriate physical model. A CHSH violation is therefore a resource for such protocols, not an unconditional certificate of the security or randomness of an entire implementation.