---
title: Tsirelson Inequality in Quantum Nonlocality
url: https://www.emergentmind.com/topics/tsirelson-inequality
type: topic
---

# Tsirelson Inequality in Quantum Nonlocality

Tsirelson inequality is the precise quantitative statement of how much quantum mechanics can violate the CHSH Bell inequality, and no more. In the standard bipartite scenario with two dichotomic measurements per party, local hidden-variable models satisfy \(|S|\le 2\), quantum theory satisfies \(|S|\le 2\sqrt{2}\), and no-signaling correlations can reach the algebraic maximum \(|S|\le 4\); the inequality therefore locates the quantum boundary between classical Bell correlations and the larger no-signaling set in the CHSH scenario [2306.12535] [1003.0616].

## 1. CHSH form and the standard quantum bound

In the CHSH setting, Alice chooses between two measurements \(A\) and \(A'\), Bob between \(B\) and \(B'\), and each outcome is represented by \(\pm 1\). The correlators are expectation values of outcome products, \(E(A_xB_y)=E[a_xb_y]\). A standard CHSH combination is
\[
S=E(AB)+E(AB')+E(A'B)-E(A'B').
\]
Equivalent sign conventions are common; in a formal Isabelle/HOL treatment the corresponding CHSH operator is written as \(A_0B_1-A_0B_0+A_1B_0+A_1B_1\), which differs only by relabeling and sign choices [2306.12535].

The classical Bell–CHSH inequality is
\[
|S|\le 2.
\]
Quantum theory permits stronger correlations but imposes the Tsirelson bound
\[
|S|\le 2\sqrt{2}.
\]
In the same scenario, no-signaling theories can attain the algebraic maximum \(|S|\le 4\), so Tsirelson inequality is neither a classical nor a purely algebraic constraint; it is specifically quantum [1003.0616] [1108.4549].

A useful equivalent form employs the CHSH “success probability”
\[
S:=p(a=b|00)+p(a=b|01)+p(a=b|10)+p(a\neq b|11),
\]
which is linearly related to the correlator form. In that parameterization, Tsirelson’s range is \(2-\sqrt{2}\le S\le 2+\sqrt{2}\) [1108.4549].

## 2. Operator formulation and proof structure

In finite-dimensional Hilbert-space form, a bipartite quantum state is represented by a density operator \(\rho\), i.e. a positive semidefinite Hermitian operator with \(\mathrm{Tr}(\rho)=1\). Projective measurements yield probabilities \(p_i=\mathrm{Tr}(\rho P_i)\), and for an observable \(M\) the expectation value is \(\langle M\rangle_\rho=\mathrm{Tr}(M\rho)\). For bipartite observables,
\[
E[a_xb_y]=\mathrm{Tr}\bigl((A_x\otimes B_y)\rho\bigr).
\]
In the CHSH context one introduces the CHSH operator
\[
\mathcal{B}=A_0\otimes B_1-A_0\otimes B_0+A_1\otimes B_0+A_1\otimes B_1,
\]
with assumptions that \(A_i\) and \(B_j\) are Hermitian, satisfy \(A_i^2=B_j^2=I\), and commute across subsystems, \(A_iB_j=B_jA_i\) [2306.12535].

Under these assumptions the crucial identity is
\[
\mathcal{B}^2=4I-[A_0,A_1][B_0,B_1].
\]
Since \(\|[A_0,A_1]\|\le 2\|A_0\|\|A_1\|=2\) and similarly for Bob’s operators, one obtains \(\|\mathcal{B}^2\|\le 8\), hence
\[
\|\mathcal{B}\|\le 2\sqrt{2}.
\]
Because \(\mathrm{Tr}(\mathcal{B}\rho)\le \|\mathcal{B}\|\) for density operators, this yields Tsirelson’s inequality [2306.12535].

This derivation has been formalized mechanically in Isabelle/HOL. In that formalization, `CHSH-expect-gen-leq` proves the quantum upper bound \(2\sqrt{2}\), while `CHSH-expect-lhv-leq` proves the local hidden-variable bound \(2\). The same paper formalizes density matrices, projective measurements, tensor products, and the measure-theoretic structure of hidden-variable models, making the contradiction between local hidden variables and quantum predictions a checked theorem rather than an informal calculation [2306.12535].

The same analytic pattern has been generalized to operator inequalities for bipartite tensor sums \(B=\sum_i x_i\otimes y_i\), with dimension-free bounds controlled by commutator and anticommutator norms. In that framework, Tsirelson’s CHSH estimate appears as a special case of a more general family of graph-structured operator inequalities [2511.01525].

## 3. Tightness, canonical realizations, and experiment

Tsirelson’s bound is tight: there are quantum states and measurements for which the CHSH value is exactly \(2\sqrt{2}\). In the formal Isabelle/HOL development, the saturating example is the singlet Bell state
\[
|\Psi^-\rangle=\frac{1}{\sqrt{2}}\bigl(|01\rangle-|10\rangle\bigr),
\]
with
\[
A_0=Z\otimes I,\qquad A_1=X\otimes I,
\]
\[
B_0=I\otimes ZmX,\qquad B_1=I\otimes XpZ,
\]
where
\[
XpZ=-\frac{1}{\sqrt{2}}(X+Z),\qquad ZmX=\frac{1}{\sqrt{2}}(Z-X).
\]
For these observables and \(\rho_{\Psi^-}=|\Psi^-\rangle\langle\Psi^-|\), the formalized CHSH expectation is exactly \(2\sqrt{2}\) [2306.12535].

Experimental work with polarization-entangled photons has approached this limit closely. A Sagnac polarization-entangled source reported
\[
2\sqrt{2}-S=(5.65\pm0.57)\times 10^{-3},
\]
together with brightness \((4660\pm70)\) pairs/s/mW, concurrence \(0.9953\pm0.0003\), and fidelity to an ideal Bell state \(0.99743\pm0.00014\). The same study attributes the dominant reduction in \(S\) to the precision of the collection focal point inside the crystal, and notes that some individual runs gave \(S>2\sqrt{2}\) only within statistical error bars, not as evidence for super-quantum correlations [2008.01575].

In the two-qubit Bell–CHSH setting, that experimental paper states that only maximally entangled states can achieve \(S=2\sqrt{2}\) [2008.01575]. At the same time, the operator-algebraic mechanism behind saturation is broader than the standard Pauli-matrix example. Large random dichotomic observables and freely independent “free observables” also yield \(\|B_{\text{CHSH}}\|\to 2\sqrt{2}\); in that sense, near-maximal CHSH violation is not restricted to explicit anticommuting qubit observables, although anticommutation remains the textbook construction [1512.00223].

## 4. Generalizations beyond binary-outcome CHSH

Tsirelson-type bounds extend beyond CHSH. In the \(2\times2\times d\) CGLMP setting, a simplified equivalent inequality can be written as
\[
A_d=P(A_2<B_2)+P(B_2<A_1)+P(A_1<B_1)+P(B_1\le A_2).
\]
For local realistic models, \(A_d\ge 1\). In the infinite-outcome limit \(d\to\infty\), the quantum problem becomes minimizing \(A_d\), and the resulting tight quantum inequality is
\[
A\ge 0.
\]
Thus the local bound remains \(1\), while the quantum bound in the \(2\times2\times\infty\) setting is \(0\) and is asymptotically attainable [1003.0616].

A notable feature of that many-outcome setting is that the maximal quantum violation is attained by a pure but not maximally entangled state. An approximate near-optimal family is
\[
|\psi_d\rangle_{\text{approx}}\sim \sum_{k=0}^{d-1}\sqrt{(k+1)(d-k)}\,|kk\rangle,
\]
and its entanglement entropy satisfies
\[
\lim_{d\to\infty}\frac{E(\psi_d)}{\log d}=\frac{1}{2}.
\]
This sharply contrasts with CHSH, where the maximal value is attained by the standard maximally entangled two-qubit realization [1003.0616].

In multipartite Bell theory, refined Tsirelson bounds need not be fixed constants. For Svetlichny operators \(\mathcal{S}_N^\pm\), one obtains
\[
\langle \mathcal{S}_N^\pm\rangle \le 2^{N-1}\sqrt{1+\sqrt{1-\eta^{(n)}}},
\qquad
\eta^{(n)}=\left(\frac{1}{2}\langle\{A^{(n)}_0,A^{(n)}_1\}\rangle\right)^2,
\]
so the quantum bound depends on a local anticommutator for one party. For odd-\(N\) Mermin–Klyshko operators,
\[
\langle \mathcal{M}_N\rangle \le
2^{N-3}\Big(\sqrt{2+\chi^{(n,m)}_+}+\sqrt{2-\chi^{(n,m)}_-}\Big),
\]
with \(\chi^{(n,m)}_\pm\) determined by bipartite anticommutators. These formulas make the multipartite quantum limit depend explicitly on lower-order subsystem correlations, and the paper interprets them as complementarity relations between local or bipartite correlations and higher-order nonlocality [2406.16202].

## 5. Dual geometry, extremal inequalities, and self-testing

In the \((2,2,2)\) Bell scenario, Tsirelson inequality can be studied from the dual perspective of the quantum set \(\mathcal{Q}\). The dual set
\[
\mathcal{Q}^*=\{\beta:\beta\cdot \bm P\le 1\ \forall\,\bm P\in\mathcal{Q}\}
\]
consists of Bell expressions normalized by their quantum bound. In that picture, the standard Tsirelson point
\[
\bm P_T=
\begin{array}{c|c|c}
1 & 0 & 0\\
0 & \tfrac{1}{\sqrt{2}} & \tfrac{1}{\sqrt{2}}\\
0 & \tfrac{1}{\sqrt{2}} & -\tfrac{1}{\sqrt{2}}
\end{array}
\]
is exposed not by a unique Bell functional but by a two-dimensional face of \(\mathcal{Q}^*\) whose intersection with a natural affine slice is a regular octagon [2401.12791].

Within that slice, the CHSH expression is not extremal. Rather, it decomposes into extremal Tsirelson inequalities; the paper shows explicitly that CHSH is the average of two such extremal inequalities. This gives a finer description of the local geometry of the quantum boundary than the usual statement that CHSH alone defines the Tsirelson point [2401.12791].

The same dual analysis identifies all Bell expressions in that octagonal region that self-test the Tsirelson realization
\[
|\phi^+\rangle=\frac{1}{\sqrt{2}}(|00\rangle+|11\rangle),
\qquad
A_x=\frac{Z_A+(-1)^xX_A}{\sqrt{2}},
\qquad
B_0=Z_B,\ B_1=X_B.
\]
Any Bell expression in the octagon interior has quantum bound \(1\) and is strictly maximized only by \(\bm P_T\), hence self-tests the Tsirelson realization [2401.12791].

For general bipartite correlation-type Bell inequalities with coefficient matrix \(g\), a broad Tsirelson-type estimate is
\[
Q\le \sqrt{M_1M_2}\,\|g\|_2,
\]
where \(\|g\|_2\) is the maximal singular value of \(g\). The paper giving this bound also provides a tightness criterion in terms of the singular vectors of \(g\) and a common ellipsoid condition for their row vectors. In this way, Tsirelson-type bounds become a constructive tool for designing Bell inequalities and dimension witnesses, rather than only upper bounds on pre-existing expressions [1306.3805].

## 6. Principles, reinterpretations, and nonstandard proposals

One information-theoretic route to Tsirelson inequality proceeds through generalized entropy. In the framework of convex probabilistic theories, if an entropy \(H\) satisfies three properties—agreement with Shannon entropy on classical systems, conditional entropy defined by \(H(A|B)=H(AB)-H(B)\), and a generalized data processing inequality stating that conditional entropy cannot decrease under local operations—then the theory necessarily obeys Tsirelson’s bound for CHSH. In that sense, Tsirelson’s inequality can be derived from a generalized DPI rather than directly from Hilbert-space operator algebra [1108.4549].

Other papers recast the bound in alternative conceptual terms. One geometric-algebra treatment argues that the derivation of the CHSH and Tsirelson inequalities depends crucially on the assumption that values of physical magnitudes are scalars, and reformulates the \(2\sqrt{2}\) limit as a geometric constraint for vector-valued magnitudes. This is presented there as a reinterpretation of the origin of the bound, not as a modification of its numerical value within ordinary quantum mechanics [2007.05301].

Several papers also investigate generalized theories in which the standard \(2\sqrt{2}\) ceiling is modified. In minimal-length quantum mechanics, the physical spin operator becomes momentum dependent,
\[
\hat s_i=f(\hat\pi^2)\hat S_i,
\]
and the CHSH maximum becomes
\[
\mathcal{S}^{\max}_{\text{MLQM}}=2\sqrt{2}\,\langle f^2(\hat\pi^2)\rangle,
\]
which exceeds the ordinary Tsirelson bound for suitable momentum states. In the pre-quantum theory of trace dynamics, the paper derives
\[
F_{\text{TD}}=2\sqrt{2}+\frac{1}{\sqrt{2}\,\mathrm{Re}(P_0)},
\]
again allowing values larger than \(2\sqrt{2}\) while preserving no-signaling under the probability prescription adopted there [2207.10418] [2208.02209].

Within standard quantum theory, however, the status of Tsirelson inequality remains unchanged. It is the sharp quantitative limit on CHSH nonlocality under the assumptions of Hermitian \(\pm1\)-valued observables, tensor-product bipartite structure, and quantum state expectations. Its importance lies precisely in this role: it fixes the maximum nonlocality of quantum theory, distinguishes the quantum set from both local and no-signaling models, and continues to organize work on Bell inequalities, self-testing, formal verification, many-outcome generalizations, and multipartite nonlocality [2306.12535] [2401.12791].

Source: https://www.emergentmind.com/topics/tsirelson-inequality