QSYS

Quantum Entanglement

John M. Donohue
Institute for Quantum Computing, University of Waterloo
(Dated: July 12, 2022)

In these notes, we will discuss the phenomenon of quantum entanglement in detail. We’ll
start by reviewing Section 3.6 of the Mathematics of Quantum Mechanics textbook to un-
derstand what makes a quantum state entangled. We’ll then discuss how we can put entan-
glement to the test with Bell’s inequalities.
Prerequisites: The Mathematics of Quantum Mechanics [1].
 2

WHAT IS ENTANGLEMENT?

So far, we’ve mostly considered individual quantum systems, like single photons and electrons.
In real experiments, it is very difficult to truly work with single quantum systems. Often, we’re
interested in how multiple systems behave together, and what their relationships to each other are.
If multiple quantum systems interact with each other, their properties may become dependent
on each other, to the point where they don’t have well-defined properties of their own. When the
properties of quantum systems cannot be separated from each other, we call them entangled.
Mathematically, an entangled state is defined in reverse, as a state that is not unentangled.
Let’s consider two qubits for the moment. An unentangled, or separable state of two qubits can
be written as |Ψ⟩ = |ψ⟩ ⊗ |φ ⟩, where the Kronecker product ⊗ indicates a clear line of separation
between the two systems. An entangled state is defined simply as a state without this property.

Definition 1. An unentangled, or separable state of two qubits can be written as

Separable : |Ψ⟩ = |ψ⟩ ⊗ |φ ⟩.

If the state is not separable, it is entangled,

Entangled : |Ψ⟩ ̸= |ψ⟩ ⊗ |φ ⟩.

This means that, in a separable state, each qubit in the two-qubit state |Ψ⟩ has it’s own well-
defined quantum state, |ψ⟩ and |φ ⟩. In an entangled state, each qubit does not have its own
quantum state. The two qubits are only well-defined when we consider them both together.
In the two-qubit scenario, recall that any quantum state can be written in terms of the two-qubit
computational basis:

Definition 2. The computational basis states for a two-qubit system are:

       
1 0 0 0
       
0 1 0 0
|00⟩ =   , |01⟩ =   , |10⟩ =   , |11⟩ =   .
       
0 0 1 0
       
0 0 0 1

Note that we’re using the short-form notation here where |00⟩ = |0⟩ ⊗ |0⟩.
 3

Let’s look at a few examples of two-qubit states and determine if they are entangled or not.

Example 1. Is the state |Ψ⟩ = √1 (|00⟩ + |01⟩) entangled?

2
We can rewrite the state by grouping terms as:

1 1
|Ψ⟩ = √ |0⟩ ⊗ |0⟩ + √ |0⟩ ⊗ |1⟩
2  2
|0⟩ + |1⟩
= (|0⟩) ⊗ √
2
= |0⟩ ⊗ |+⟩

This state can be written in the form |ψ⟩ ⊗ |φ ⟩, where |ψ⟩ = |0⟩ and |φ ⟩ = |+⟩, and
therefore it is separable.

Example 2. Is the state |Ψ− ⟩ = √1 (|01⟩ − |10⟩)

2
entangled?
This state doesn’t allow us to group terms as easily. We’ll see if it’s entangled by looking
for contradictions. If the state |Ψ− ⟩ is separable, we can write it as |ψ⟩ ⊗ |φ ⟩. In
general, we can write these qubit states as |ψ⟩ = ψ0 |0⟩ − ψ1 |1⟩ and |φ ⟩ = φ0 |0⟩ + φ1 |1⟩,
where ψi and φi are complex numbers satisying |ψ0 |2 + |ψ1 |2 = 1 = |φ0 |2 + |φ1 |2 . We
can therefore expand the state and write an arbitrary two-qubit separable state in terms
of the four numbers ψ0 , ψ1 , φ0 , and φ1 as

if separable, |Ψ− ⟩ = ψ0 φ0 |00⟩ + ψ0 φ1 |01⟩ + ψ1 φ0 |10⟩ + ψ1 φ1 |11⟩.

For the state √1 (|01⟩ − |10⟩) to be separable, both (a): ψ0 φ1 = −ψ1 φ0 = √1 and (b):
2 2
ψ0 φ0 = ψ1 φ1 = 0 must be true. Statement (b) implies that at least two of the four values
are zero, but this creates a contradiction with statement (a)!
We have found a contradiction, and therefore the state is not separable. Since it is not
separable, it must be entangled.

Exercise 1. Determine if the following states are separable or entangled:

1
(a) (|00⟩ − |01⟩ − |10⟩ + |11⟩)
2
1
(b) (|00⟩ + |01⟩ + |10⟩ − |11⟩)
2
 4

Entanglement vs. correlation

Let’s say Alice and Bob each have one qubit from an entangled state. If the entangled state
is |Ψ− ⟩ = √1 (|01⟩ − |10⟩),
2
if Alice measures the first qubit in the state “0”, Bob will find the
second qubit in the state “1”. If Alice measures the first qubit in the state “1”, Bob will instead
measure the second qubit in the state “0”. The results are always anti-correlated. Each qubit’s
state is completely random individually, but a clear correlation is present if they consider both
qubits together. The qubits are in a non-local superposition of the states |01⟩ and |10⟩, and when
one qubit is measured, the other collapses at the same time.
While the idea of collapse is unique to quantum mechanics, the idea of correlation is nothing
new. Say that, instead of a pair of photons or electrons, we had a pair of shoes. Alice takes a
shoe from the pair and sends the other shoe to Bob. When Bob looks at what shoe he has, he
immediately knows which shoe Alice has. If Bob sees a left shoe, Alice must have the right one,
and vice-versa. This example models the exact same anti-correlation as |Ψ− ⟩ did.
So what separates entanglement from classical correlations? Unlike classical states, there are
many ways to measure a quantum state. Let’s say that the shoes can be measured in a quantum
way, and we label the left shoe as |0⟩ and the right as |1⟩. Once Alice sends a shoe to Bob, the state
is either |01⟩ or |10⟩. If Alice and Bob measure in the +/− basis, they’ll see completely random
outcomes instead of the anti-correlations of the 0/1 basis.
But the state |Ψ− ⟩ is very different than a 50/50 chance of |01⟩ or |10⟩. By writing it as
√1 (|01⟩ − |10⟩), we’re writing it as a superposition of |01⟩ and |10⟩. Just as a quantum coin in
2
the superposition state |+⟩ behaves differently than a classical coin if measured in different ways,
the superposition state |Ψ− ⟩ may behave differently than the anti-correlated shoes if we use the
right measurements. What happens if we measure |Ψ− ⟩ in the basis formed by the states |+⟩ and
|−⟩?
 5

Example 3. What happens when we measure the state |Ψ− ⟩ = √1 (|01⟩ − |10⟩)
2
in the
+/- superposition basis? We’ll use the relationship |0⟩ = √1 (|+⟩ + |−⟩) and |1⟩ =
2
√1 (|+⟩ − |−⟩)
2
to rewrite |Ψ− ⟩ as

 
1 1 | + +⟩ − | + −⟩ + | − +⟩ − | − −⟩ | + +⟩ − | + −⟩ + | − +⟩ − | − −⟩
√ (|01⟩ − |10⟩) = √ −
2 2 2 2
 
1 0| + +⟩ − 2| + −⟩ + 2| − +⟩ − 0| − −⟩
=√
2 2
1
= √ (| − +⟩ − | + −⟩)
2

Due to interference, the | + +⟩ and | − −⟩ terms cancel out, and we’re left with |Ψ− ⟩ =
√1 (| − +⟩ − | + −⟩). We can see that, even when making measurements in a different
2
basis, the results are still anti-correlated.

Entanglement is only measurably different than classical correlations if we measure in multiple

bases. That’s because entanglement is not just a correlation; it’s a superposition as well.
Notably, it doesn’t matter how far apart Alice and Bob are for them to see entanglement. If
Alice and Bob are lightyears apart, their qubits may still be entangled. When Alice measures
her qubit, Bob’s state will instantly collapse. Depending on how Alice measures her qubit, Bob’s
could collapse to a state in the 0/1 basis, the +/− basis, or any other basis of Alice’s choosing.
This cannot be used for communicating faster than the speed of light, since Alice’s can’t control
which state she measures (the outcome is still random).
As we’ll see later, there are ways to create and measure entangled states. However, just mea-
suring in two bases is not enough to rigourously prove that they’re entangled. To do that, we’ll
need to dive a bit deeper into Bell’s inequalities.

|01⟩+|10⟩
Exercise 2. Rewrite the other “Bell” entangled states in the +/- basis: |Ψ+ ⟩ = √
2
,
|00⟩+|11⟩ |00⟩−|11⟩
|Φ+ ⟩ = √
2
, and |Φ− ⟩ = √
2
.

Exercise 3. What correlations, if any, does the state |Ψ− ⟩ = √12 (|01⟩ − |10⟩) have in the
|0⟩+i|1⟩ |0⟩−i|1⟩
basis formed by | + i⟩ = √
2
and | − i⟩ = √
2
?
What about if Alice measures in the +/- basis and Bob measures in the ±i basis?
 6

THE BELL-CHSH INEQUALITY

Entanglement is one of the most truly “quantum” features of quantum mechanics. When trying
to understand it, many scientists weren’t convinced. What if, instead of needing these non-local
superposition states, there were other factors at play. What if, for example, the qubits shared
extra classical correlations that humans can’t see, but travel with the photons and tell them how to
behave when measured in certain ways? This kind of hidden variable theory would suggest that
the universe does behave in an intuitive way, just one that can’t be readily investigated. Maybe
quantum mechanics is just a rest stop on the way to a more satisfying theory of reality yet to be
discovered?
The Bell inequalities explicitly deny such an idea [2, 3], showing that quantum weirdness is an
essential part of the universe and putting it to the experimental test. To talk about them, we will
need to understand how to calculate expectation values and use the outer product. If you aren’t
comfortable with using these yet, read through the appendix first.

The inequality

Imagine that Alice and Bob, as seen in Fig. 1, are playing a game. In this game, they are each
asked one of two yes/no questions by a shadowy agent. Alice’s questions are labelled A0 and A1 ,
and Bob’s questions B0 and B1 . They do not know in advance what question to expect, and they
have no knowledge of what the other player is being asked. While they can plan as much as they
want before the game starts, they aren’t allowed to communicate at all once a question is asked.
The goal of the game is to agree every time they are both asked the “0” question (A0 and B0 ),
agree every time they are asked different questions (A0 and B1 , or A1 and B0 ), and disagree every
time they are both asked the “1” question (A1 and B1 ). If we assume that the questions are random,
then they can win 75% of the time by always agreeing. In this scenario, they only lose when they
are both asked the “1” question. But can they do better?
To find out, let’s formulate the problem mathematically. Instead of “YES”/“NO”, let’s say that
Alice and Bob can answer the questions with either “+1” or “-1”. If they are supposed to agree,
their answers should multiply to +1. If they are supposed to disagree, their answers should multiply
to -1. For example, in the case where they are asked the questions A0 and B0 , their goal is to make
the expectation value of their combined answer, ⟨A0 B0 ⟩, as close to 1 as possible. Simultaneously,
 7

A0 or A1 B0 or B1

+1 or -1 +1 or -1

Alice Bob

FIG. 1. The CHSH game with classical players. Alice and Bob are asked one of two questions by an agent,
and can provide yes or no (±1) answers. They are separated by a wall and not allowed to communicate
once the game starts. They win if they satisfy the requirements of the table, where their answers agree in all
cases except when they are asked questions A1 and B1 .

⟨A0 B1 ⟩ and ⟨A1 B0 ⟩ must be close to 1, and ⟨A1 B1 ⟩ must be close to -1. We can put this all together
by asking Alice and Bob to maximize the parameter S, which has an expectation value of

⟨S⟩ = ⟨A0 B0 ⟩ + ⟨A0 B1 ⟩ + ⟨A1 B0 ⟩ − ⟨A1 B1 ⟩. (1)

If Alice and Bob aren’t allowed to communicate, then Alice’s answer can’t depend on the
question Bob was asked. Therefore, the joint expectation value is just the multiplication of the
two individual expectation values, ⟨Ai Bi ⟩ = ⟨Ai ⟩⟨Bi ⟩. First, let’s assume that Alice and Bob have
pre-determined answers for each question. In this case, we can re-arrange terms and find that S
cannot be greater than 2, as

S = A0 (B0 + B1 ) + A1 (B0 − B1 ) = ±2.

Since B0 and B1 must be either +1 or -1, then one of B0 + B1 or B0 − B1 must be equal to ±2 and
the other equal to zero. Since A0 and A1 can also only be ±1, S can then only take the values ±2.
But what if Alice and Bob introduce some probabilities into the mix. Say that Alice and Bob
agree that, for every trial, they’ll use a different strategy, labelled “i” with probability λi , such that
∑i λi = 1. They co-ordinate exactly what they’ll do each time, but still don’t know what question
to expect in advance. So, for each value of i,

Si = A0,i B0,i + A0,i B1,i + A1,i B0,i − A1,i B1,i = ±2.

 8

The overall expectation value of S will be a mixture of all these strategies, and will have an absolute
value of
|⟨S⟩| = ∑ λiSi = |⟨A0 B0 ⟩ + ⟨A0 B1 ⟩ + ⟨A1 B0 ⟩ − ⟨A1 B1 ⟩| ≤ 2. (2)
i

Since the maximum value any one strategy can provide is 2, a mixture of different strategies can
only give an expectation value of less than or equal to 2. This the CHSH inequality [3], and all
classical systems must obey it. This inequality is a modification to John Bell’s original theorem
by Clauser, Horne, Shimony, and Holt, collectively known as CHSH [3].
The derivation depended principally on two assumptions about the universe. First, the universe
is local. What Alice does or measures can’t affect what Bob does in a distant lab without sending a
signal. If Alice and Bob are incredibly far apart and make their choices quickly, that signal would
have to travel faster than the speed of light. This is why we could argue that ⟨Ai Bi ⟩ = ⟨Ai ⟩⟨Bi ⟩.
The second assumption is that the values Ai and Bi have a definite value independent of being
measured, often called the assumption of realism. Say that, instead of directly answering the
question, Alice and Bob each opened a box which contained their answer. The assumption of
realism asserts that whatever was in that box would have the same answer regardless of whether
Alice and Bob opened it or not. Even if Alice chooses to open the A1 box, it should still be possible
to know exactly what she would have done if she opened the A0 box instead.
Combined, these two assumptions are called local realism. All classical theories, like the
hidden-variable theory where an unknown parameter λi controls the scenario, obey local realism.
But what about if we bring quantum mechanics back in?

Using entanglement to beat CHSH

Let’s reformulate the problem into the language of quantum measurements, as sketched in
Fig. 2. Let’s say Alice and Bob each have a qubit, and their two-qubit system is in the entangled
state |Ψ− ⟩ = √12 (|01⟩ − |10⟩). Depending on the question they’re asked by the agent, they’ll make
one of two quantum measurements, A0 or A1 for Alice, and B0 or B1 for Bob. If you need to refresh
on how to write down two-qubit states and operation matrices, see Section 3.6 of the Mathematics
Primer [1].
With the |Ψ− ⟩ state, if Alice and Bob both measure in the 0/1 or +/- basis, they will always
obtain opposite results, as seen in Example 3. We can rewrite this using the expectation values of
 9

A0 A1 B0 B1
+1
Source
-1

Alice Bob

FIG. 2. The CHSH game with quantum players. This time, Alice and Bob both share entangled particles,
and formulate the “questions” they ask in terms of the quantum measurements A0 or A1 (for Alice) and B0
or B1 (for Bob). Each measurement spits out either a +1 or -1 value, and they choose their measurement
and read their value fast enough that they could not have sent a signal to each other.

the Z and X operators, which can be expressed as

 
1 0
Z=  = (+1)|0⟩⟨0| + (−1)|1⟩⟨1|
0 −1
 
0 1
X =  = (+1)|+⟩⟨+| + (−1)|−⟩⟨−|.
1 0

Note how we re-write the Z matrix the second time. What it’s telling us is that if we measure the
state |0⟩, we assign a value of ”+1”, and if we measure the state |1⟩, we assign the value ”-1”.
The expectation value of |0⟩ would be given as ⟨0|Z|0⟩ = +1, meaning that it always gives a ”+1”
value. The expectation value of the superposition state |+⟩, for example, would be ⟨+|Z|+⟩ = 0,
as you would measure ”+1” half the time and ”-1” the other half, which would average out.
Just as in the classical case, when we look at two-qubit measurements, we effectively multiply
Alice and Bob’s results together. So if Alice and Bob both measure the same thing (both ”+1”
or both ”-1”), their answers multiply to ”+1”. If they measure opposite results, they’ll multiply
to ”-1”. We can represent this as a two-qubit quantum measurement, Z ⊗ Z for the 0/1 basis and
X ⊗ X for the +/- basis. Recalling Definition 3.6.10 from the Math Primer [1], we can write these
in matrix form to find joint expectation values.
 10

Example 4. We can write the matrix Z ⊗ Z out, following the rules of Kronecker products, as
       
1 0 1 0 1 0 0 0
 1  0    
0 −1 0 −1   0 −1 0 0 
  

Z ⊗Z =  =
   

  0 0 −1 0 
  
 1 0 1 0
 0  −1   
0 −1 0 −1 0 0 0 1

To find the expectation value ⟨Z ⊗ Z⟩ with the state |Ψ− ⟩, we find

  
1 0 0 0  0 
 
−1 0 √1
i 0 
h 0
⟨Ψ− |Z ⊗ Z|Ψ− ⟩ = 0
 
√1 − √12 0   2  = −1.
 
2 0 0 −1 0 − √12

 



0 0 0 1 0

Alternatively, we could have done the entire problem using bra-ket notation alone, by writing
Z = |0⟩⟨0| − |1⟩⟨1| and expanding Z ⊗ Z as

Z ⊗ Z = (|0⟩⟨0| − |1⟩⟨1|) ⊗ (|0⟩⟨0| − |1⟩⟨1|)

= |00⟩⟨00| − |01⟩⟨01| − |10⟩⟨10| + |11⟩⟨11|.

The expectation value for |Ψ− ⟩ can then be found as

1
⟨Ψ− |Z ⊗ Z|Ψ− ⟩ = (⟨01| − ⟨10|) (|00⟩⟨00| − |01⟩⟨01| − |10⟩⟨10| + |11⟩⟨11|) (|01⟩ − |10⟩)
2
1
= −|⟨01|01⟩|2 − |⟨10|10⟩|2 = −1


2

Exercise 4. Using either method, show that, for the state |Ψ− ⟩, ⟨X ⊗ X⟩ = −1, ⟨X ⊗ Z⟩ = 0,
and ⟨Z ⊗ X⟩ = 0.

Let’s say that Alice and Bob use the following strategy. If Alice is asked the question A0 , she’ll
perform the quantum measurement A0 = Z. If she is asked the question A1 , she’ll perform the
quantum measurement A1 = X. Bob will similarly perform quantum measurements on his half
of |Ψ− ⟩, but on his side, B0 = √1 (−Z − X)
2
and B1 = √1 (−Z + X).
2
Bob’s measurements can be
 11

thought of as tilted versions of Alice’s; if they were using polarized photons, Alice would hold her
polarizer at 0◦ and 45◦ for her two measurements, while Bob’s polarizer angles would be 22.5◦
and 67.5◦ . The factor of √1
2
is necessary to make sure the eigenvalues of the operators are +1 and
−1.
If they are both asked question “0”, their expectation value of their answer will be the expecta-
tion value of the following joint quantum operator:

1
⟨A0 B0 ⟩ = ⟨Ψ− |Z ⊗ √ (−Z − X) |Ψ− ⟩
2
1
= √ −⟨Ψ− |Z ⊗ Z|Ψ− ⟩ − ⟨Ψ− |Z ⊗ X|Ψ− ⟩


(3)
2
1 1
= √ (1 − 0) = √
2 2

If we continue this for the other three options, we find that

1
⟨A0 B1 ⟩ = √
2
1
⟨A1 B0 ⟩ = √
2
1
⟨A1 B1 ⟩ = − √
2

None of these possibilities have an expectation value of ±1, meaning that there is an element of
randomness. In the first three, we’re more likely to get +1 than −1, and the opposite for the fourth
possibility.
If we bring these all together and find the value of the parameter ⟨S⟩ from Eq. (2), we find that

√
⟨S⟩Ψ− = ⟨A0 B0 ⟩ + ⟨A0 B1 ⟩ + ⟨A1 B0 ⟩ − ⟨A1 B1 ⟩ = 2 2 ≈ 2.83. (4)

But in Eq. (2), we showed that (assuming local realism) the maximum value of ⟨S⟩ was 2! Some-
how, by using entangled quantum states and clever measurements, quantum mechanics is capable
of something that no classical strategy can achieve. Notably, quantum mechanics still does not
always win the CHSH game (instead of a 75% chance of winning, it is now an 85.3% chance of
winning), but if we can show in an experiment that this bound is beat, then we can guarantee that
our system is indeed quantum entangled.
More philosophically, it shows that our assumptions about local realism are not obeyed by
 12

quantum mechanics. To have a full appreciation of the quantum world, we have to give up one or
the other. Either we give up locality, and admit that entanglement allows distant particles to co-
ordinate with each other without needing to send physical signals, or we give up realism, accepting
that quantum outcomes are truly undefined until they are measured.

Exercise 5. If we had the state |Ψ+ ⟩ = √12 (|01⟩ + |10⟩) instead, would we still violate the Bell-
CHSH inequality with these measurement settings? How could we change our measurements
to still violate the inequality?

Exercise 6. Show that the naive choice of A0 = B0 = Z and A1 = B1 = X does not violate the
Bell-CHSH inequality.

Exercise 7. Let’s optimize! Let B0 = (cos θ )X + (sin θ )Z, B1 = −(sin θ )X + (cos θ )Z, A0 =
√
Z, and A1 = X . Show that ⟨S⟩Ψ− = 2 2 is the maximum possible violation of the Bell-CHSH
inequality. This value, called Tsirelson’s Bound, turns out to be the highest value possible
using any quantum state and quantum measurements [4].

Faster-than-light communication?

One common misconception about entanglement and it’s defiance of local realism is that it
allows for Alice and Bob to communicate faster than the speed of light. In short, entanglement
absolutely cannot be used to communicate faster than the speed of light. The difference between
the kind of non-locality that entanglement can exhibit and the kind that would allow for faster-
than-light communication is subtle, but very important. If we wanted to “fake” a violation of the
Bell-CHSH inequality, we would need some kind of information-carrying signal that travels faster
than the speed of light. But entanglement itself does not allow Alice to send a bit of information
to Bob.
The key to this is the difference between signalling and correlation. When Alice measures
her half of the entangled state, she knows with certainty what Bob will measure. But she has no
way of controlling what Bob will measure. If Alice measures “+1”, then Bob will measure “-1”
in the same basis, but there was a 50% chance of Alice measuring “-1” instead. By adding the
complication of measurements in multiple bases, we can find correlations that are impossible to
mimic with classical systems. But we still have that element of randomness that prevents us from
 13

using them to pass messages. We’ll see later when discussing superdense coding and teleportation
how we can use these correlations for different tasks, but these all require additional information
that travels slower than the speed of light.
An interesting side-note: if we were able to clone a quantum state, we would be able to use
entanglement to communicate faster than the speed of light. In fact, an experiment was proposed
in 1982 to clone quantum states to use for faster-than-light communication [5]. This proposal
was found to have a critical error, which was found almost immediately and had a wide-ranging
impact on quantum information theory. The error is simple: it is impossible to clone a quantum
state, i.e. create a unitary transformation which takes the state |ψ⟩ and outputs two copies of the
state, |ψ⟩ ⊗ |ψ⟩ [6]. The proof is among the simplest in quantum mechanics, and can fit on a
napkin.

Example 5. The No-Cloning Theorem [6].

For a quantum state cloner to be perfectly effective, it must work for any possible input state
|ψ⟩. To show that this is impossible, we’ll use a few example cases. Without loss of generality,
say that we start with two quantum states: the state we want to clone, |ψ⟩, and a blank state to
copy it to, initially set to |0⟩. The ideal cloning unitary Uclone should perform

Uclone |ψ⟩|0⟩ = |ψ⟩|ψ⟩.

If we use it with the states |0⟩ and |1⟩, it must work such that

Uclone |0⟩|0⟩ = |0⟩|0⟩,

Uclone |1⟩|0⟩ = |1⟩|1⟩.

What about a superposition state? The unitary must be linear, so we can use the previous two
results to find

1
Uclone |+⟩|0⟩ = √ (Uclone |0⟩|0⟩ +Uclone |1⟩|0⟩)
2
1
= √ (|00⟩ + |11⟩)
2
̸= |+⟩|+⟩
 14

This contradiction very simply shows that a perfect unitary cloning machine is impossible to
build in quantum mechanics.

Experiments and loopholes

To put the Bell-CHSH inequality to the test, we need a source of entangled photons and a
way to measure them in multiple bases. Early experiments used atomic cascades which emitted
polarization-entangled photons [7, 8]. Modern experiments use photon sources based on nonlinear
optics or quantum dots, which have a much higher collection efficiency and photon rate [9]. In all
of these, a photon source spits out entangled photons to Alice and Bob, who randomly choose a
measurement for each photon. The two collect data for many photons and compare their results
afterwards. They calculate an overall value of S from their measurements and using error analysis
to determine how certain they are that they violated the inequality.
In testing the Bell-CHSH inequality experimentally, it’s important to remember what exactly
we’re testing, and how we could falsify a violation. If Alice was able to communicate her mea-
surement choice and result to Bob before he made his measurement, Bob could cheat and ensure
that they always win the game. This is a kind of loophole, called the locality loophole, in that it
allows them to get around the rules of the game. To close this loophole and prevent Alice, Bob,
or nature from cheating, Alice and Bob must be far enough apart that Alice and Bob must choose
their measurement, measure their photons, and record theirs results fast enough that no signal trav-
elling at the speed of light could tell the other what measurement was performed before the photon
was detected. In the language of relativity, Alice and Bob’s actions must be outside of each other’s
light cones.
Another loophole is the detection or fair-sampling loophole. As an analogy, imagine you work
at an apple farm, and your job is to determine how many apples have gone bad. You obviously
can’t check every apple, so you have to pick a small sample of apples to test. If the farmer selects
his fifty best apples for you to test, you won’t know anything about how good the rest of the apples
are. You need to have a fair, random, and unbiased sample to make claims about the rest of the
apples.
Now let’s think about it with photons. No photon detector is perfectly efficient. Even the best
detectors today miss at least 5% of the photons that strike them. Say that the detectors chose
which photons to miss, keeping measurements that win with the CHSH game and discarding
 15

measurements that lose the CHSH game. By taking a smaller sample of the data, we can violate
the Bell-CHSH inequality, even if we wouldn’t had we used all the data. To make sure that this
loophole is closed, we need to build our experiment so we lose as few photons as possible, meaning
high-efficiency detection and carefully aligned optical fibres. In the most common experiments,
the overall efficiency has to be a minimum of 67%, which is much harder than it sounds.
Both of these loopholes could be considered conspiracy theories, where nature abuses cheat
codes to make the world look quantum when it really obeys local realism. To rule them out
completely, we need to design experiments that are both long-distance and high-efficiency simul-
taneously. In 2015, three landmark experiments managed to show this, two using polarization-
entangled photons [10, 11] and one using entangled diamonds [12].
There still remains the so-called freedom of choice loophole. In all of these experiments, Alice
and Bob’s measurements are chosen randomly for every photon, often combining many sources of
randomness together (including some quantum tools!). If the measurements aren’t truly random,
and could have been known in advance, that creates a loophole. This goes back to philosophical
arguments about whether Alice, Bob, and their equipment truly have free will. If Alice and Bob’s
choices are known in advance by some all-knowing entity, they could take advantage of that to
fake the inequality violation. However, we’d like to think that such an all-knowing entity would
have better things to do than muck around with our understanding of local realism.

Application example: Device-independent QKD

Bell’s inequalities can be used directly to make quantum key distribution even more secure. In
QKD, an eavesdropper introduces errors if they measure the quantum state Alice sends to Bob.
We could modify the protocol slightly by having Alice and Bob both measure photons from an
entangled source [13]. If the entangled state is |Φ+ ⟩ = √1 (|00⟩ + |11⟩), they will always agree on
2
the result if they both measure in the Z or X basis. However, if an eavesdropper measures a photon
in the middle, they’ll disturb the quantum state and break Alice and Bob’s correlations. By doing
the same error-correcting steps as in BB84, they’ll detect the presence of the eavesdropper.
In the real world, this isn’t so simple. If the eavesdropper controls which measurements Alice
and Bob make, they can remain undetected [14]. The idea of device-independent QKD is to
add extra complication such that, even if the eavesdropper controls Alice and Bob’s measurement
devices, they won’t be able to access the key. Alice and Bob need only to secure the actual
 16

readout of the measurement and be able to choose which button to push, but the eavesdropper can
hold onto the measurement device itself and control what each button does. Instead of just two
measurements, Alice and Bob have an extra measurement that tests the Bell-CHSH inequality. If
they find a violation of the Bell-CHSH inequality, they know that their quantum state is indeed
entangled, and that their device is making valid quantum measurements. Essentially, this allows
Alice and Bob to catch the eavesdropper messing with both their quantum state and their quantum
device. The eavesdropper can still deny service by tinkering with the measurement device, but will
once again be caught when Alice and Bob compare their error rates. In practice, the eavesdropper
can take advantage of loopholes to “fake” the Bell-CHSH violation, so we need a loophole-free
experiment for unconditional security.

CONCLUSION

Entanglement is a broad topic that underlies nearly all of quantum information theory and
experiments. Later in QCSYS, we’ll show how entanglement connects to quantum computing.
We’ll also revisit entanglement when discussing quantum teleportation, superdense coding, and
other information paradoxes.

[1] M. Laforest, The Mathematics of Quantum Mechanics (The Institute for Quantum Computing, Uni-
versity of Waterloo, 2015).
[2] J. Bell, On the Einstein-Podolsky-Rosen paradox, Physics 1, 195 (1964).
[3] J. F. Clauser et al., Proposed experiment to test local hidden-variable theories, Phys. Rev. Lett. 23, 880
(1969).
[4] B. S. Cirel’son, Quantum generalizations of Bell’s inequality, Letters in Mathematical Physics 4, 93
(1980).
[5] A. Peres, How the no-cloning theorem got its name, Fortschritte der Physik: Progress of Physics 51,
458 (2003).
[6] W. K. Wootters and W. H. Zurek, A single quantum cannot be cloned, Nature (London) 299, 802
(1982).
 17

[7] S. J. Freedman and J. F. Clauser, Experimental test of local hidden-variable theories, Phys. Rev. Lett.
28, 938 (1972).
[8] A. Aspect, P. Grangier, and G. Roger, Experimental tests of realistic local theories via Bell’s theorem,
Phys. Rev. Lett. 47, 460 (1981).
[9] P. G. Kwiat et al., Ultrabright source of polarization-entangled photons, Phys. Rev. A 60, R773 (1999).
[10] L. K. Shalm, E. Meyer-Scott, B. G. Christensen, P. Bierhorst, M. A. Wayne, M. J. Stevens, T. Gerrits,
S. Glancy, D. R. Hamel, M. S. Allman, et al., Strong loophole-free test of local realism, Phys. Rev.
Lett. 115, 250402 (2015).
[11] M. Giustina, M. A. Versteegh, S. Wengerowsky, J. Handsteiner, A. Hochrainer, K. Phelan, F. Stein-
lechner, J. Kofler, J.-Å. Larsson, C. Abellán, et al., Significant-loophole-free test of bell’s theorem
with entangled photons, Phys. Rev. Lett. 115, 250401 (2015).
[12] B. Hensen, H. Bernien, A. Dréau, A. Reiserer, N. Kalb, M. Blok, J. Ruitenberg, R. Vermeulen,
R. Schouten, C. Abellán, et al., Loophole-free Bell inequality violation using electron spins separated
by 1.3 kilometres, Nature 526, 682 (2015).
[13] A. K. Ekert, Quantum cryptography based on Bell’s theorem, Phys. Rev. Lett. 67, 661 (1991).
[14] A. Acı́n, N. Brunner, N. Gisin, S. Massar, S. Pironio, and V. Scarani, Device-independent security of
quantum cryptography against collective attacks, Phys. Rev. Lett. 98, 230501 (2007).
 18

MEASUREMENTS AND EXPECTATION VALUES

In probability problems like risk analysis, we often have a series of possible outcomes with
known probabilities. If each outcome has a value associated with it (for example, a payout), it can
be useful to know the average value to expect each time. In these cases, where we have possible
outcomes with values Vi and probabilities pi of occurring, the expectation value of V is

⟨V ⟩ = ∑ piVi . (5)
i

Example 6. You’re at a carnival, and are invited to play a game where you draw a marble out
of a jar. The jar contains 10 blue marbles, 5 red marbles, and 1 black marble. If you pick the
black marble, you’ll get $50. If you draw a red marble, you’ll get $10. If you draw a blue
marble, you’ll get nothing. The game costs $5 to play. Is it a good deal?
Most of the marbles are blue, so the most likely scenario is you’ll get nothing. But since
the other marbles are worth so much, is it worth the risk? To find out, we’ll calculate the
1 5
expectation value. The values we need are: pblack = 16 and Vblack = $50, pred = 16 and
10
Vblack = $10, , and pblue = 16 and Vblue = $0. Using Eq. (5), the expected win amount is
therefore

1 5 100
⟨V ⟩ = pblackVblack + pred Vred + pblueVblue = 50 × + 10 × = = $6.25.
16 16 16

If we play many times, we expect to win $6.25 per game. Since the game costs only $5 to
play, it’s a worthwhile risk.

Let’s adapt this concept to quantum problems. As we’ve seen many times, the outcome of a
quantum measurement is probabilistic. For example, if I measure whether a qubit in state |ψ⟩ is
in the state |0⟩ or |1⟩, I’ll find it in the state |0⟩ with probability |⟨0|ψ⟩|2 , and in the state |1⟩ with
probability |⟨1|ψ⟩|2 . This is elegantly summarized in Born’s rule, which states that the probability
of measuring |ψ⟩ in the state |φ ⟩ is given by |⟨φ |ψ⟩|2 .
Since we are talking about qubits, each basis consists of exactly two states that are orthogonal
to each other. In general during this section, we’ll call them |φ+ ⟩ and |φ− ⟩. For a measurement
in the 0/1 basis, |φ+ ⟩ = |0⟩ and |φ− ⟩ = |1⟩. For a measurement in the +/- basis, |φ+ ⟩ = |+⟩ and
|φ− ⟩ = |−⟩.
 19

What happens if we repeat our experiment many times, preparing and measuring our quan-
tum system in exactly the same way? By looking at the statistics afterward, we can infer those
probabilities and learn about the state |ψ⟩. To help tabulate the results, every experiment will spit
out a number Q, which we assign the value “Q = +1” every time we measure the state |φ+ ⟩, and
“Q = −1” every time we measure the state |φ− ⟩. By running the experiment many times, adding
these values up, and dividing by the number of times we run the experiment, we obtain the expec-
tation value of Q, which we write as ⟨Q⟩. This can be related to the probabilities obtained through
Born’s rule as
⟨Q⟩ = (+1) × |⟨φ+ |ψ⟩|2 + (−1) × |⟨φ− |ψ⟩|2 . (6)

The expectation value tells us the average value of Q over many experiments. If ⟨Q⟩ = 0.9,
for example, we’d understand that it is much more likely to obtain a value of Q = +1 rather than
Q = −1 if we were to run the experiment one more time.
√ √
1+
√ 2 |0⟩ + 1−
√ 2 |1⟩
Example 7. We have a machine that makes the state |ψ⟩ = 6 6
whenever we
push a button, which we then measure in the +/- basis. We also have a counter Q, initially set
to Q = 0. Every time we measure |+⟩, we increase Q’s value by one. Every time we measure
|−⟩, we decrease it by one. If we run the experiment 1000 times, what is the expected value
of Q?
1
First, we find the expectation of Q. Using Born’s rule, p+ = |⟨+|ψ⟩|2 = 3 and p− =
|⟨−|ψ⟩|2 = 23 . The expectation value of Q is therefore

1 2 1
⟨Q⟩ = (+1) × + (−1) × = − .
3 3 3

So, for each run of the experiment, we’ll reduce the counter by 1/3 on average. If we run the
experiment 1000 times, we expect to reduce the counter by 1000 × ⟨Q⟩ ≈ −333.

Using matrices, we can make the problem even simpler. If we multiply a row vector and a
column vector, as in the inner product, we get a number out. If we reverse the order and multiply
a column vector and a row vector, we get a matrix instead. We write this outer product matrix
Mφ as |φ ⟩⟨φ |. The matrix Mφ is an example of an observable, and has a convenient connection to
Born’s rule as
|⟨φ |ψ⟩|2 = ⟨ψ|φ ⟩⟨φ |ψ⟩ = ⟨ψ|Mφ |ψ⟩.
 20

To calculate an expectation value, we previously used Born’s rule to find the probabilities, and
multiplied those probabilities by the appropriate ±1 values. However, we can skip this step by
using the outer product to create a new observable matrix M as

M = (+1) × |φ+ ⟩⟨φ+ | + (−1) × |φ− ⟩⟨φ− | = M+ − M−

Let’s use the same example as above to find an expectation value, but with matrices this time.
√ √
Example 8. We once again have a machine that makes |ψ⟩ = 1+√6 2 |0⟩+ 1−√6 2 |1⟩, and measure
in the +/- basis. To find the expectation value, we first construct the observable for the +/- basis
as
M± = (+1) × |+⟩⟨+| + (−1) × |−⟩⟨−|
   
1 h i 1 1 h i 1
= 1 1  − 1 −1  
2 1 2 −1
   
1 1 1  1  1 −1 
=  −
2 1 1 2 −1 1
 
0 1
= 
1 0

To find the expectation value in one step, we surround the matrix with the row and column
vector corresponding to the state |ψ⟩ as
  √ 
√ √ 1+
√ 2
0 1
 = −1.
h i
⟨ψ|M± |ψ⟩ = 1+
√ 2 1−
√ 2   √6
6 6
1 0 1−
√ 2 3
6

The matrix we found should look familiar as the X, or bit-flip, matrix. It is also the matrix that
can be used to find expectation values when measuring in the +/- basis. From another point of
view, the eigenvalues of the X matrix are the states |+⟩ and |−⟩, with eigenvalues of +1 and −1,
respectively. This is why we often call the +/- basis the X basis.

Exercise 8. Show that the Z and Y matrices can be used to find expectation values in the 0/1
and ±i bases, respectively.

Now let’s see how we use these tools to derive the Bell-CHSH inequality.