Bell's Theorem

This disproves Einstein, Podolsky, and Rosen (1935)

Let Alice have two possible measurements A1A_1 or A2A_2. She picks one per experimental run. Bob has two possible measurements B1B_1 or B2B_2. He also picks one per run.

Each measurements returns ±1\pm 1

Ai,Bj{±1}A_i,B_j\in\{\pm 1\}

Local realism assumes that before every measurement, every observable has a pre-existing Definite value determined by some hidden variable λ\lambda. So in any given run, A1,A2,B1,B2A_1,A_2,B_1,B_2 all have predetermined definite values.

Let QQ be a combination observable where

Q=A1(B1B2)+A2(B1+B2)Q=A_1(B_1-B_2)+A_2(B_1+B_2)

Given that the measurement of Ai,BjA_i,B_j returns ±1\pm 1 we just defined before, then Q can only be either ±2\pm 2.

Therefore, QQ's average on each run must statisfy

2E[Q]+2-2\leq \mathbb{E}[Q]\leq +2

expand QQ using linearity

E[Q]=E[A1B1]E[A1B2]+E[A2B1]+E[A2B2]\mathbb{E}[Q] = \mathbb{E}[A_1 B_1] - \mathbb{E}[A_1 B_2] + \mathbb{E}[A_2 B_1] + \mathbb{E}[A_2 B_2] E[A1B1]E[A1B2]+E[A2B1]+E[A2B2]2\big|\mathbb{E}[A_1 B_1] - \mathbb{E}[A_1 B_2] + \mathbb{E}[A_2 B_1] + \mathbb{E}[A_2 B_2]\big| \leq 2

This is CHSH inequality Clauser, Horne, Shimony, and Holt (1969).

Consider a rotated observable Let

Wθ(A)=σAXsinθ+σAZcosθW^{(A)}_\theta = \sigma_A^X \sin\theta + \sigma_A^Z\cos\theta Wθ(B)=σBXsinθ+σBZcosθW^{(B)}_{\theta'} = \sigma_B^X \sin{\theta'} + \sigma_B^Z\cos{\theta'}

Combining gives

E[Wθ(A)Wθ(B)]=E[(σAXsinθ+σAZcosθ)(σBXsinθ+σBZcosθ)]\mathbb{E}[W^{(A)}_\theta W^{(B)}_{\theta'}] = \mathbb{E}[(\sigma^X_A\sin\theta + \sigma^Z_A\cos\theta)(\sigma^X_B\sin\theta' + \sigma^Z_B\cos\theta')] =sinθsinθE[σAXσBX]+cosθcosθE[σAZσBZ]+sinθcosθE[σAXσBZ]+cosθsinθE[σAZσBX]= \sin\theta\sin\theta'\,\mathbb{E}[\sigma^X_A\sigma^X_B] + \cos\theta\cos\theta'\,\mathbb{E}[\sigma^Z_A\sigma^Z_B] + \sin\theta\cos\theta'\,\mathbb{E}[\sigma^X_A\sigma^Z_B] + \cos\theta\sin\theta'\,\mathbb{E}[\sigma^Z_A\sigma^X_B]

See Singlet

=sinθsinθcosθcosθ= -\sin\theta\sin\theta' - \cos\theta\cos\theta' =cos(θθ)= -\cos(\theta - \theta')

Suppose Alice and Bob chose

A1=W0,A2=Wπ/2,B1=Wπ/4,B2=W3π/4A_1 = W_0, \quad A_2 = W_{\pi/2}, \quad B_1 = W_{\pi/4}, \quad B_2 = W_{3\pi/4}

Where θ=0\theta=0 is measuring on σZ\sigma^Z and θ=π/2\theta=\pi/2 is measuring on σX\sigma^X

E[A1B1]=12,E[A1B2]=+12,E[A2B1]=12,E[A2B2]=12\mathbb{E}[A_1 B_1] = -\tfrac{1}{\sqrt{2}}, \quad \mathbb{E}[A_1 B_2] = +\tfrac{1}{\sqrt{2}}, \quad \mathbb{E}[A_2 B_1] = -\tfrac{1}{\sqrt{2}}, \quad \mathbb{E}[A_2 B_2] = -\tfrac{1}{\sqrt{2}}

Substitute into Clauser, Horne, Shimony, and Holt (1969) to get

=12121212=22= \big|{-\tfrac{1}{\sqrt{2}}} - \tfrac{1}{\sqrt{2}} - \tfrac{1}{\sqrt{2}} - \tfrac{1}{\sqrt{2}}\big| = 2\sqrt{2}

However, this contradicts as

22≰22\sqrt{2} \not\leq 2

The shows that the two are mathematically incompatible. Bell's theorem in concrete form.

222\sqrt{2} is known a Tsirelson's bound. The maximum Clauser, Horne, Shimony, and Holt (1969) value a state can get.

See how it is exploited with Greenberger, Horne, and Zeilinger (1989)