Device-Independent Certification

Tsirelson's bound is not only a ceiling — it is a rigidity statement. Reaching $S=2\sqrt2$ is possible in essentially one way, so the observed value alone tells you what state and measurements were used, even if you treat the devices as sealed black boxes. This is the foundation of device-independent quantum information.

What "device-independent" means

In a device-independent (DI) analysis you assume almost nothing about the hardware. The boxes have inputs (settings $x,y$ ) and outputs (results $a,b$ ); you do not trust their Hilbert-space dimension, the state they share, or what their measurements actually are. The only data are the observed statistics $p(a,b\mid x,y)$ . A conclusion is DI if it follows from those statistics plus the no-signalling/locality structure, regardless of the internal model.

The standing assumptions are minimal but essential: the devices are spatially separated and do not communicate during a round, the inputs are freely chosen (measurement independence), and the runs are independent. Under just these, a Bell violation is a hardware-agnostic certificate.

Self-testing: the CHSH rigidity theorem

Self-testing is the strongest form of DI certification: a particular value of a Bell expression pins down the state and measurements up to a local isometry. For CHSH the statement is:

If two black boxes achieve $S = 2\sqrt2$ , then there exist local isometries $\Phi_A\otimes\Phi_B$ mapping the unknown shared state $|\psi\rangle$ to $|\Phi^+\rangle\otimes|\text{junk}\rangle$ , and mapping the unknown observables $A_0,A_1$ to the Pauli operators $Z,X$ (and $B_0,B_1$ to $\tfrac{Z\pm X}{\sqrt2}$ ) on the relevant factor.

In words: maximal violation forces an embedded Bell pair measured along orthogonal axes. The "junk" register and the isometry capture the unavoidable freedoms — local basis changes, added ancillas, complex conjugation — that no statistics can ever resolve. Everything else is fixed.

Robustness

Real experiments see $S = 2\sqrt2 - \varepsilon$ , never the exact value. Robust self-testing theorems make the conclusion quantitative: the certified state is $O(\sqrt\varepsilon)$ -close (in trace distance, after the isometry) to a perfect Bell pair. The certificate degrades gracefully, so a near-maximal violation still guarantees a near-ideal Bell pair — which is what lets DI protocols tolerate noise.

Why it is powerful

Self-testing turns a number measured on untrusted hardware into a statement about the quantum systems inside. A vendor could ship a malicious or simply faulty device; if it nonetheless reports $S\approx 2\sqrt2$ across freely chosen, space-like-separated settings, it must contain a near-perfect entangled pair measured almost optimally. That guarantee — entanglement certified without trusting the box — is exactly what the next lessons exploit to extract certified randomness and, ultimately, to secure key distribution.

The takeaway

Device independence draws conclusions from statistics alone. CHSH is self-testing: the maximal value $2\sqrt2$ certifies a Bell pair and orthogonal Pauli measurements up to a local isometry, and robust versions keep that guarantee under realistic noise. Rigidity, not just the bound, is what makes nonlocality a usable resource.