Download as pdf or txt
Download as pdf or txt
You are on page 1of 128

Perspectives on Entanglement:

Gregorious Quantum Information


(Session 1?)

Christian Ferko

July 21, 2022

Christian Ferko Entanglement July 21, 2022 1 / 64


Prologue: why quantum information?

Christian Ferko Entanglement July 21, 2022 2 / 64


Three vignettes (plus math bonus).

Before I attempt to tell you about quantum information, I should try to


convince you that it’s worth caring about.

I’ll provide a few examples – from physics, quantum computing, and high
energy theory – to illustrate why quantum information is interesting.

1 The EPR “paradox” (physics).


2 Quantum teleportation (quantum computing).
3 Ryu-Takayanagi (hep-th).
4 Bonus: entanglement as cohomology (mathematics).

Christian Ferko Entanglement July 21, 2022 3 / 64


Vignette 1: The EPR “paradox” (physics).

Christian Ferko Entanglement July 21, 2022 4 / 64


Locality in physics.

In the gauge theory talk I spoke about locality: the physics affecting some
object should only care about conditions near that object.

Part of what makes entanglement, and therefore quantum information,


interesting is that it appears to badly violate this principle.

This apparent violation is made especially sharp in the so-called “EPR


paradox” named for Einstein, Podolsky, and Rosen.

Christian Ferko Entanglement July 21, 2022 5 / 64


EPR setup.

Suppose that a spin-0 particle decays into two spin- 12 particles, which then
fly off to far-separated detectors on the right and left.

The initial state has spin zero, and angular momentum is conserved. Thus
one of the decay products is spin down and the other is spin up.

If I measure spin-up (+ 21 ) at the right detector, then I instantaneously


know that someone will measure spin-down (− 21 ) at the left detector.

When I first learned this, I found it very boring. Let’s see why I was wrong.

Christian Ferko Entanglement July 21, 2022 6 / 64


A non-quantum EPR.
Consider a different experiment: a lab assistant places two socks, one red
and one blue, into boxes at random. I don’t see which sock goes into which
box. The boxes are then carried away to far-separated experimenters.

If I open the box at the right side and find a red sock, then I
instantaneously know that someone will find the blue sock at the left.

This is expected and totally unremarkable. I claim that the quantum


version of this experiment is fundamentally different.
Christian Ferko Entanglement July 21, 2022 7 / 64
Violating Bell.

It is possible to prove that the outcome of any experiment wherein the


particles both had well-defined, independent values for their spins through
the experiment satisfy certain inequalities. For example,

|P(⃗sL , ⃗sR ) − P(⃗sL , ⃗s0 )| ≤ 1 + P(⃗sR , ⃗s0 ) ,

where ⃗sL , ⃗sR are the spin vectors of the left and right particles, and ⃗s0 is
any fixed reference vector.

The predictions which quantum mechanics makes for the outcomes


of the EPR scenario violate these inequalities. Experiments have
confirmed these predicted violations.

We are forced to conclude that the quantum EPR experiment is not like
the sock example. The two spins had not “already decided” which was up
and which was down at the time of the decay.

Christian Ferko Entanglement July 21, 2022 8 / 64


The particles were entangled.
The resolution is that the spin- 12 particles did not have independent spins
before reaching the detectors. As we will explain, they were in an
entangled state, which we might write as
1
|Ψ⟩ = √ (|↑⟩ ⊗ |↓⟩ − |↓⟩ ⊗ |↑⟩) .
2
The preceding inequalities can now be violated because there is no way to
speak about individual properties of each particle in this state. We may
only speak about the properties of the combined system.

When a measurement is performed, the state collapses into a state where


each particle has well-defined spin. This happens immediately when either
particle reaches a detector.

Conclusion: The collapse of this state is transmitted instantaneously from


one particle to another, regardless of the physical distance, upon a
measurement. There is no speed-of-light delay. So long for locality.
Christian Ferko Entanglement July 21, 2022 9 / 64
Vignette 2: Quantum teleportation
(quantum computing).

Christian Ferko Entanglement July 21, 2022 10 / 64


Quantum teleportation.
The previous discussion makes it seem as though “entanglement” is just a
strange quirk of probability distributions for quantum measurements. In
applications, however, entanglement is viewed as a concrete resource, like
energy, which you can use up to do useful things.

A striking example of this is quantum teleporation. Suppose Greg is on


Titan and I’m on Earth, but we each possess one of the particles in an EPR
pair (e.g. from a source, or one which Greg took with him 5 years ago).
Christian Ferko Entanglement July 21, 2022 11 / 64
How to transmit a state?

Say my friend has a different electron, separate from the ones in the EPR
pair, and I want to transmit the state of that electron to Greg. How?

I cannot simply measure the state of the electron and send the data
to him. Measurement of any property collapses the state.

I cannot make a copy of the quantum state to perform more


measurements (or to send the copy) by the no-cloning theorem.

I cannot encode the state of the electron in some radio signal and
send this. (Theorem: A quantum state cannot be transferred by
solely classical communication.)

Is there any way for Greg to recreate the state of my electron in his lab,
short of physically sending my electron to him? Assume we can send radio
signals, and use our shared EPR pair, but nothing else.

Christian Ferko Entanglement July 21, 2022 12 / 64


The teleportation procedure.
First I bring together the friend’s electron, which I want to transmit, and
my particle in our EPR pair. Then I perform a joint measurement which
yields one of four results:
1 1
1 : √ (|↑⟩ ⊗ |↓⟩ − |↓⟩ ⊗ |↑⟩) , 2 : √ (|↑⟩ ⊗ |↑⟩ − |↓⟩ ⊗ |↓⟩)
2 2
1 1
3 : √ (|↑⟩ ⊗ |↑⟩ + |↓⟩ ⊗ |↓⟩) , 4 : √ (|↑⟩ ⊗ |↓⟩ + |↓⟩ ⊗ |↑⟩) .
2 2
I send the result of the measurement to Greg (2 bits of data), and he
performs the following rotation on his member of the EPR pair:

1 : nothing , 2 : 180◦ around x axis ,


3 : 180◦ around y axis , 4 : 180◦ around z axis .

After doing this, Greg’s electron is in the precise state that my


friend’s electron was in before the procedure. This only works because
our particles were entangled. On my end, the friend’s state is “destroyed.”
Christian Ferko Entanglement July 21, 2022 13 / 64
Quantum communication.

We have now transmitted the full state


of a quantum system – more than any amount of
data that can be sent by classical communication
alone, and thus sometimes called “ℵ0 bits”
– by using an EPR pair and sending only 2 bits.

The EPR pair is used up by this process! We


have “spent” one unit of entanglement to send
the state to Greg.

This is the sense in which entanglement is a


resource. Many remarkable things are possible
with quantum computation and quantum
communication that cannot be done classically.
But entanglement is a key ingredient, and we
would thus like to quantify how much we have.
Christian Ferko Entanglement July 21, 2022 14 / 64
At this stage entanglement may appear to be such an abstract
notion that the need to quantify it does not seem to be urgent
but then, once upon a time, ‘energy’ must have seemed a very ab-
stract notion indeed, and now there are thriving industries whose
role is to deliver it in precisely quantified amounts. Perhaps our
governments will eventually have special Departments of Entan-
glement to deal with these things.
Vignette 3: Ryu-Takayanagi (hep-th).

Christian Ferko Entanglement July 21, 2022 15 / 64


The gauge-gravity duality.
An exciting – and very unexpected – perspective on entanglement comes
from my field of high energy theory, and in particular holography.

One of the great insights we’ve learned from string theory is the
AdS/CFT correspondence: a theory of quantum gravity on a negatively
curved spacetime manifold, called anti-de Sitter space, is equivalent to a
conformal field theory which lives on the boundary of the spacetime.
Christian Ferko Entanglement July 21, 2022 16 / 64
Entanglement entropy.

Fix a subregion A of the boundary. In a given quantum state |Ψ⟩ of the


conformal field theory, there can be entanglement between the degrees of
freedom in A and those in AC .
The degree of entanglement is quantified by the von Neumann entropy:

SvN = − Tr (ρA log(ρA )) , ρA = TrAC (|Ψ⟩⟨Ψ|) .

Because the boundary theory is equivalent to the bulk gravitational theory


by AdS/CFT, we expect that SvN corresponds to some quantity that can
be computed in the AdS spacetime.

Gravity is a theory of geometry. Can we find a geometrical object in the


bulk which encodes the entanglement entropy SvN ?

Christian Ferko Entanglement July 21, 2022 17 / 64


The Ryu-Takayanagi formula.
Answer: The entropy is given by SvN = Area(γA)
4πG , where γA is a surface
homologous to A, whose boundary is ∂A, and which extremizes the area.

This is remarkable! An information-theoretic quantity in the field theory,


SvN , is related to a purely geometric quantity in the bulk gravity theory,
namely the area of γA !
Christian Ferko Entanglement July 21, 2022 18 / 64
Bonus Vignette: entanglement as cohomology
(mathematics).

Christian Ferko Entanglement July 21, 2022 19 / 64


Extra motivation for Greg .

Much of this talk is aimed at giving Greg some broader background for a
project involving entanglement and homological algebra.

The idea: entanglement is morally an obstruction to describing a “global”


composite system using “local” descriptions of each subsystem.

This sounds a lot like cohomology – for instance, every closed form is
locally exact, and the obstruction to describing a closed form as a globally
exact form is encoded by cohomology. Indeed, there is a cochain complex

E d −1 E d0
0 −→ C −→ Com(ρA ) × Com(ρB ) −→ Com(ρAB ) −→ 0 .

ker(dE0 )
A bipartite system is entangled if and only H 0 = im(dE−1 )
is non-trivial.

Christian Ferko Entanglement July 21, 2022 20 / 64


Back to the prologue.

Christian Ferko Entanglement July 21, 2022 21 / 64


This is a hot field.
Quantum information is an extraordinarily interdisciplinary and active
research area. Massive amounts of funding go to physicists, computer
scientists, mathematicians, and engineers working in this field.

My goal in this talk (or these talks) is humble: I want to develop some of
the shared language and notation which people use when working on this
subject, and to present some simple examples to build intuition.
In short: I aim to give you some of the prerequisite background and
context needed to read papers in this area. I apologize that much of
this is standard textbook material, but the payoff is worthwhile.
Christian Ferko Entanglement July 21, 2022 22 / 64
Roadmap.

We’ve now seen four ways to think about entanglement: as correlations, as


a resource, as area, and as cohomology.

If I have not persuaded you that this is an interesting subject, then you
have no soul.

Here is the plan for the remainder of the talk:

□ Prologue: why quantum information? A few vignettes.

□ Act 1: Welcome to Hilbert space.

□ Act 2: Endomorphisms and observables.

□ Act 3: A toy example involving hard, black electrons.

□ Epilogue: Preview of tensor products, factorizability, and more.

Christian Ferko Entanglement July 21, 2022 23 / 64


Act 1: Welcome to Hilbert space.

Christian Ferko Entanglement July 21, 2022 24 / 64


What is Hilbert space?

The stage on which quantum mechanics unfolds is called Hilbert space.

A (complex) Hilbert space H is a vector space over


p C, equipped with an
inner product ⟨x, y ⟩ which induces a norm ∥x∥ = ⟨x, x⟩, and which is
complete with respect to this norm. Completeness means that every
Cauchy sequence in H converges to an element of H.

Note that (⟨x, y ⟩)∗ = ⟨y , x⟩. We choose conventions where

⟨ax1 + bx2 , y ⟩ = a∗ ⟨x1 , y ⟩ + b ∗ ⟨x2 , y ⟩ .

We will call the notation ⟨x, y ⟩ for the inner product “math notation.”
Importantly, in this expression, both x and y are elements of H (not H∗ ).

The usual axioms for a vector space, inner product, norm, etc. hold.*
*
To wit, a vector space is an abelian group under addition, and scalar multiplication
defines a ring homomorphism from C into the endomorphism ring of this group.
Christian Ferko Entanglement July 21, 2022 25 / 64
Infinite vs. finite dimensions.
Often in physics we care about infinite-dimensional Hilbert spaces.

We usually restrict to separable Hilbert spaces, i.e. those which have a


countable orthonormal basis (equivalently, a countable dense subset).

One way of representing infinite-dimensional Hilbert space is as ℓ2 , the


vector space of all square-summable sequences of complex numbers:

( )
X
2 2
ℓ = (z1 , z2 , z3 , · · · ) zi ∈ C , |zi | < ∞ .
i=1

All infinite-dimensional separable Hilbert spaces are isomorphic. You may


hear people say “there is only one Hilbert space”* to refer to this fact.

However, for this talk (and related talks/projects) we will only consider
finite-dimensional Hilbert spaces. That is, H = Cn for some n.
*
I am confused about a discussion in the server where Tetra wanted to distinguish
between ℓ2 and L2 . Perhaps I’m missing some turtle-esque subtlety.
Christian Ferko Entanglement July 21, 2022 26 / 64
States are rays in Hilbert space.

The state of a quantum system is determined by a ray in Hilbert space,


i.e. an element of H/∼ where ∼ is the equivalence relation

v ∼ λv , λ ∈ C∗ ≡ C \ {0} .

Two states are viewed as physically equivalent if they differ only by


multiplication by a non-zero complex number. We often remove part of
this ambiguity by demanding that states be normalized, that is,

∥v ∥ = 1 .

Two normalized states are still viewed as physically equivalent if they differ
by a phase, or a complex number z of unit norm, z = e iθ for θ ∈ R.

Christian Ferko Entanglement July 21, 2022 27 / 64


Dirac’s bra-ket notation.
We adopt the convention that an element of Hilbert space is written as a
ket, |ψ⟩ ∈ H. This is no different from a statement like v ∈ V , where V is
a vector space; we simply decorate the symbol v with | · ⟩.

Example. The Hilbert space H = C2 is spanned by basis vectors


   
1 0
|e1 ⟩ = , |e2 ⟩ = ,
0 1

and a general element |v ⟩ ∈ H can therefore be written


 
a
|v ⟩ = a1 |e1 ⟩ + a2 |e2 ⟩ = 1 , a1 , a2 ∈ C .
a2

This Hilbert space H = C2 models the state of a single quantum bit or


qubit. In that context, we often write |e1 ⟩ as |0⟩ and |e2 ⟩ as |1⟩, which is
called the computational basis. Note that |0⟩ = ̸ 0 is not the zero vector.
Christian Ferko Entanglement July 21, 2022 28 / 64
Bras live in the dual space.

Let H∗ (also written H∨ ) be the dual space to H, which is the space of


linear functions φ : H −→ C. By the Riesz representation theorem,

if φ ∈ H∗ there is a yφ ∈ H so that φ (x) = ⟨yφ , x⟩ for all x ∈ H .

That is: the action of any dual vector φ can be reproduced by taking the
inner product with an appropriate vector yφ . We therefore adopt the
notation ⟨φ| ∈ H∗ for elements of the dual space, and define

⟨φ | x⟩ ≡ φ(x) = ⟨yφ , x⟩ .

To be clear: on the left side, an element ⟨φ| ∈ H∗ eats an element


|x⟩ ∈ H and returns a number (“physics notation”). The right side is the
inner product between two elements x, yφ ∈ H (“math notation”).

Christian Ferko Entanglement July 21, 2022 29 / 64


An isomorphism between H and H∗ .
A vector defines a linear map. In the “math” notation, we would say that
for each y ∈ H, there is a linear map φy ∈ H∗ defined by

φy (x) = ⟨y , x⟩ for all x ∈ H .

Conversely, a linear map defines a vector. By Riesz, for each φ ∈ H∗ there


is an yφ ∈ H so that

φ(x) = ⟨yφ , x⟩ for all x ∈ H .

Thus there is a one-to-one correspondence between elements of H and


elements of H∗ . Under this correspondence, |x⟩ ←→ ⟨x|.

Furthermore, it is redundant to consider both inner products ⟨y , x⟩ of two


elements of H, and pairings φ(y ) of an element of H∗ with an element of
H – they are different descriptions of the same thing!

We will prefer the pairing ⟨φ|x⟩ over the explicit inner product ⟨yφ , x⟩.
Christian Ferko Entanglement July 21, 2022 30 / 64
Finite-dimensional bras and kets.

We are specializing to H = Cn , which has the natural inner product

⟨v , w ⟩ = v1∗ w1 + v2∗ w2 + · · · + vn∗ wn .

Here vi , wj are the components of v , w in an orthonormal basis {ei }. For


instance, v = v1 e1 + · · · + vn en . We will represent v as a column vector:
 
v1
 .. 
|v ⟩ = v1 |e1 ⟩ + · · · + vn |en ⟩ =  .  .
vn

It is now natural to write bras as row vectors:

⟨v | = v1∗ · · · vn∗ .
 

In going from a ket to a bra, we must complex-conjugate all of the entries.

Christian Ferko Entanglement July 21, 2022 31 / 64


This notation evokes the dot product.
The reason for using row and column vectors is that the pairing of a bra
⟨v | ∈ H∗ and a ket |w ⟩ ∈ H now looks like the usual matrix product:
 
w1
 . 
⟨v |w ⟩ = v1 · · · vn  ..  = v1∗ w1 + · · · + vn∗ wn = ⟨v , w ⟩ .
 ∗ ∗

wn

We introduce the symbol † which satisfies |v ⟩† = ⟨v |, ⟨v |† = |v ⟩. That is,


the † operation transposes a vector (from row vector to column vector, or
from column to row) and conjugates each entry.

This is by analogy with a similar operation for operators which we will


introduce later. One can view a ket or bra as an n × 1 or 1 × n matrix,
respectively, with entries wi1 = wi for a ket or v1j = vj∗ for a bra. Then

vij† = vji∗ .

Christian Ferko Entanglement July 21, 2022 32 / 64


A numerical example.
To finish this act, let’s look at an explicit calculation for the sake of
exposure. The computational basis states are
   
1 0
|0⟩ = , |1⟩ = .
0 1
 
i i 1
Given another state |ψ⟩ = √2 (|0⟩ + |1⟩) = √2 , we have
1
i i  
⟨ψ| = − √ (⟨0| + ⟨1|) = − √ 1 1 .
2 2
Then ⟨ψ|ψ⟩, which is the norm of |ψ⟩, is
 
i i
⟨ψ|ψ⟩ = − √ (⟨0| + ⟨1|) · √ (|0⟩ + |1⟩)
2 2
1
= (⟨0|0⟩ + ⟨0|1⟩ + ⟨1|0⟩ + ⟨1|1⟩)
2
1
= (1 + 0 + 0 + 1) = 1 .
2
Christian Ferko Entanglement July 21, 2022 33 / 64
Act 2: Endomorphisms and observables.

Christian Ferko Entanglement July 21, 2022 34 / 64


Endomorphisms are linear maps.
Consider the collection End(H) of endomorphisms from H to itself. The
underlying structure of H is a vector space, and an object O ∈ End(H) is
a linear map on this vector space:

O (a1 |v1 ⟩ + a2 |v2 ⟩) = a1 O (|v1 ⟩) + a2 O (|v2 ⟩) .

In the infinite-dimensional setting, one typically restricts attention to


bounded maps. But for us, H = Cn , so all linear maps are bounded.

In particular, when H = Cn , any linear map can be represented as a matrix


whose columns are the images of the basis vectors under the map.
 
2 2 o00 o01
O : C −→ C , O= .
o10 o11

We will use the terms “(linear) map,” “operator,” and “endomorphism”


interchangeably to refer to elements of End(H).
Christian Ferko Entanglement July 21, 2022 35 / 64
Outer product notation.

Consider the formal expression |v ⟩ ⟨w |, where |v ⟩ ∈ H and ⟨w | ∈ H∗ . It is


natural to interpret this object as a linear map from H to H. Why?
 
|x⟩ ∈ H =⇒ |v ⟩ ⟨w | |x⟩ = |v ⟩ ⟨w | x⟩ = z |v ⟩ .
| {z }
=z∈C

That is, the object |v ⟩ ⟨w | eats a vector |x⟩ ∈ H and spits out some other
vector z |v ⟩ ∈ H. This is exactly what a function from H to H should do.

The function defined in this way is linear because ⟨w | ∈ H∗ is, by


definition, a linear map from H to C. So |v ⟩ ⟨w | ∈ End(H).

The operation which takes a ket |v ⟩ and bra ⟨w | and creates a linear map
is called the outer product. That is, the outer product is a map from
H × H∗ to End(H).

Christian Ferko Entanglement July 21, 2022 36 / 64


The outer product in a basis.
The notation |v ⟩ ⟨w | immediately suggests how to evaluate it when we
represent |v ⟩ as a column vector and ⟨w | as a row vector. For instance, if
 
v
|v ⟩ = 1 , ⟨w | = w1∗ w2∗ ,
 
v2

then their outer product is evaluated using the usual matrix multiplication,

v1 w1∗ v1 w2∗
   
v1  ∗ ∗

|v ⟩ ⟨w | = w1 w2 = .
v2 v2 w1∗ v2 w2∗

In particular, a general 2 × 2 matrix can be written as


 
o00 o01
= o00 |0⟩ ⟨0| + o01 |0⟩ ⟨1| + o10 |1⟩ ⟨0| + o11 |1⟩ ⟨1| .
o10 o11
n
Any linear map O : Cn → Cn can be written as O =
P
oij |ei ⟩ ⟨ej |.
i,j=1
Christian Ferko Entanglement July 21, 2022 37 / 64
The Hermitian adjoint.
Let O ∈ End(H). There exists a unique operator O† such that

⟨x, Oy ⟩ = ⟨O† x, y ⟩ for all x, y ∈ H .‘

Note the formal similarity to adjoint functors. In bra-ket notation,


 † D E†
⟨x | O | y ⟩ = ⟨x| (O |y ⟩) = O† |x⟩ |y ⟩ = y O† x .

That is, one can do either of the following:


1 act with O on |y ⟩, and evaluate the function ⟨x| ∈ H∗ on the
resulting ket O |y ⟩; or
2 find the element of H∗ dual to the ket O† |x⟩ ∈ H under the
ismorphism between H and H∗ , and evaluate that function on the ket
|y ⟩ ∈ H;

and the results are the same.


Christian Ferko Entanglement July 21, 2022 38 / 64
The adjoint in components.
Note that we are overloading the symbol †. When acting on a bra or ket,
it behaves as |v ⟩† = ⟨v |, ⟨v |† = |v ⟩. But when acting on an operator, it
acts as the Hermitian adjoint.*

We now argue that this overloading makes sense. In components, the


action of † on a matrix is to transpose the matrix and conjugate its entries:
 †  ∗ ∗

o00 o01 o00 o10
= ∗ ∗ ,
o10 o11 o01 o11

or more concisely, Oij† = Oji∗ . This is the same as the action of † on a bra
or ket, thought of as a 1 × n or n × 1 matrix, respectively.

The two definitions are both compatible and consistent, assuming that the
Hermitian conjugate of a product reverses order (just like the matrix
transpose). For instance, (O1 O2 |v ⟩)† = ⟨v | O2† O1† .
*
Further, when acting on a complex number z we define z † = z ∗
Christian Ferko Entanglement July 21, 2022 39 / 64
Hermitian operators have real eigenvalues.
We say that an operator H ∈ End(H) is Hermitian if H † = H.

Recall that v ̸= 0 is an eigenvector of H with eigenvalue λ if Hv = λv .


A Hermitian operator has real eigenvalues. If H † = H, note that
⟨v , Hv ⟩ = ⟨H † v , v ⟩ = ⟨Hv , v ⟩ .
If v is an eigenvector, this means
⟨v , λv ⟩ = ⟨λv , v ⟩ =⇒ λ∥v ∥2 = λ∗ ∥v ∥2 ,
so λ = λ∗ and λ is real.

More generally, an operator A is normal if AA† = A† A. Theorem: an


operator A is normal if and only if it admits a representation
n
X
A= λi |vi ⟩ ⟨vi |
i=1

where the λi are eigenvalues of A and {|vi ⟩} is an orthonormal basis of Cn .


Christian Ferko Entanglement July 21, 2022 40 / 64
Observables are associated with Hermitian operators.
Quantities we measure in the real world are real numbers, so eigenvalues
of Hermitian operators are good candidates for observable quantities.
Fact 1. The state of a physical system is represented by a ray |ψ⟩ ∈ H/∼.
Fact 2. Every observable A is represented by Hermitian operator Â.
Fact 3. The only possible result of a measurement of an observable A is
one of the eigenvalues of Â.
Fact 4. If a state |ψ⟩ is written as a sum of eigenvectors of Â,
n
X
|ψ⟩ = ai |vi ⟩ , Â |vi ⟩ = λi |vi ⟩ ,
i=1

then the probability of measuring eigenvalue λi is proportional to |ai |2 .


Fact 5. Immediately after a measurement of observable A has yielded a
value λi , the state of the system is the eigenvector |vi ⟩.
Christian Ferko Entanglement July 21, 2022 41 / 64
The operator algebra.
We write A(H) for the operator algebra on H.

An algebra A over C is just a vector space over C, along with an operation


A × A → A for multiplying two vectors which satisfies some axioms.

We claim that End(H) already has a natural algebra structure. We know


how to add and scalar-multiply operators,

z1 O1 + z2 O2 ,

which makes End(H) a vector space. Multiplication of two operators is


defined as composition of the endomorphisms:

O1 O2 ≡ O1 ◦ O2 ∈ End(H) .

End(H) with these operations defines A(H), the operator algebra.*


*
The subset of A(H) consisting of Hermitian operators A† = A, which we write as
As.a. (H), is not a closed algebra with this product, but it is a Jordan C ∗ algebra.
Christian Ferko Entanglement July 21, 2022 42 / 64
Operator algebra for Cn .

In our case, H = Cn , and the algebra A(Cn ) is especially simple.

Recall that we write Matn (C) for the algebra of all n × n matrices with
entries in C. The addition operation is just elementwise addition of the
matrix entries and multiplication by scalars is elementwise multiplication:
     
1 −1 3 0 2 · 1 + 3 2 · (−1) + 0
2 + = ,
0 5 2 −2 2·0+2 2·5−2

and “vector multiplication” in the algebra is the matrix product:


     
2 1 −1 1 2 · (−1) + 1 · 2 2·1+1·4
· = .
−3 0 2 4 (−3) · (−1) + 0 · 2 −3 · 1 + 0 · 4

Therefore, A(Cn ) is just the matrix algebra Matn (C).

Christian Ferko Entanglement July 21, 2022 43 / 64


Act 3: A Toy Example Involving
Hard, Black Electrons

Christian Ferko Entanglement July 21, 2022 44 / 64


Electrons, color, and hardness.
Let’s set formalism aside for a moment and discuss some observations
which this machinery was invented to explain.

The experiments we consider involve properties of electrons (or any spin- 12


particle). To avoid having to explain spin, we will assume that electrons
have two properties called color and hardness.*

An electron’s color can only be either black or white. Its hardness can only
be either hard or soft. We can measure color or hardness by feeding
electrons into a box which spits out the electrons from one of two ports.

*
These are of course not actual terms; real electrons do not have color or hardness.
We are just using different words for real properties to simplify the discussion.
Christian Ferko Entanglement July 21, 2022 45 / 64
Repeatable.

Measuring a property twice in immediate succession gives the same result.

Christian Ferko Entanglement July 21, 2022 46 / 64


Uncorrelated.

Measuring color gives no information about hardness and vice-versa.

Christian Ferko Entanglement July 21, 2022 47 / 64


Make a prediction.
Question. Predict the fraction of electrons that will exit the final box at
the top as black and as white.

Christian Ferko Entanglement July 21, 2022 48 / 64


A surprise.
In this experiment, half of the electrons exit the final color box as black
and half as white, even though we previously measured all of the input
electrons as black!

Christian Ferko Entanglement July 21, 2022 49 / 64


Mirrors and combiners.
We now introduce two additional ingredients for these experiments.
Mirrors re-direct the electron beam without changing any properties.
Combiner boxes merge two beams without changing any properties.

Christian Ferko Entanglement July 21, 2022 50 / 64


Mirror experiment 1.
What happens here?

Christian Ferko Entanglement July 21, 2022 51 / 64


Mirror experiment 2.
What happens here?

Christian Ferko Entanglement July 21, 2022 52 / 64


Mirror experiment 3.
What happens here?

Christian Ferko Entanglement July 21, 2022 53 / 64


Wall.

Suppose we put a wall in the way of the soft path.

Christian Ferko Entanglement July 21, 2022 54 / 64


Very confusing.

An electron in mirror experiment 3 exits as white 100% of the time.

Did it take the hard path? No; with the wall, half of the hard-path
electrons are black.

Did it take the soft path? No; with the wall, half of the soft-path electrons
are black.

Did it take both? No; we can always measure the electron to be on one
path or the other, but not both.

Did it take neither? No; putting walls in both paths removes all output.

What other logical possibility is there?

Christian Ferko Entanglement July 21, 2022 55 / 64


Superposition.
This is the reason that we are forced to adopt a framework where the state
of a system is described by a vector. Crucially, vectors can be added:
1
e − = √ (|hard path⟩ + |soft path⟩) .
2
Physically, an object in this state is not really taking both paths; when
measured, it is always found in one or the other. Instead we say that the
electron is in a superposition of states corresponding to the two paths.

If we measure which path the electron takes – e.g. by putting a wall in one
of the paths, or installing cameras along the paths – then the state
collapses to |hard path⟩ or |soft path⟩, each with probability 12 .

However, if we do not measure which path the electron takes, it remains in


this superposition, which happens to be an eigenstate of the color operator
with “eigenvalue” white.
Christian Ferko Entanglement July 21, 2022 56 / 64
Hardness example.

Example. Let’s suppose


 that the “hardness” property is associated with
1 0
the matrix σz = . The eigenvalues and eigenvectors are
0 −1
   
1 0
λ1 = 1 with v1 = , λ2 = −1 with v2 = .
0 1

If I prepare an electron in the state


 
1 1+i 1+i 1−i
u= = v1 + v2
2 1−i 2 2

and measure its hardness, I will get “hard” (λ1 = 1) half of the time and
“soft” (λ2 = −1) half of the time.

Christian Ferko Entanglement July 21, 2022 57 / 64


Hardness exercise.

Question. If I prepare an electron in the state


 
1 1+i
u=√
6 2

and measure its hardness, what are the probabilities of getting “hard”
(λ1 = 1) and “soft” (λ2 = −1)?

Christian Ferko Entanglement July 21, 2022 58 / 64


Color measurement.
 
1
Say I prepare an electron in a state of definite hardness, u = .
0
   
0 1 1 1
Color is associated with σx = , with eigenvectors v1 = √2 and
1 0 1
 
1
v2 = √12 , eigenvalues λ1 = 1 (“black”) and λ2 = −1 (“white”).
−1

When I measure the color in state u, what do I get?


  1  1 
1 1 1
u= = 21 + 21 = √ v1 + √ v2 .
0 2 − 2 2 2
|{z} |{z}
a1 a2

So I measure “black” half of the time and “white” half of the time.

Christian Ferko Entanglement July 21, 2022 59 / 64


Repeated measurement exercise.
 
0
Exercise. I prepare an electron in u = (definitely soft).
1

(1) If I measure color, what are the possible outcomes and probabilities?

(2) Suppose the measurement in (1) yields “black”. What state is the
electron in now?

(3) After measuring “black” in (2), I now measure hardness. What are the
outcomes and probabilities?

Christian Ferko Entanglement July 21, 2022 60 / 64


“Wavefunction collapse.”

Punchline: measuring a property of the electron changes its state (Fact 5).

When you measure “color” and get “black”, the electron collapses to
1
u = √12 . Now it no longer has a definite hardness.
1

We can never simultaneously know the “color” and “hardness” of an


electron since no vector is both an eigenstate of σx and σz .

This is not a “tree falls in the forest” question – i.e. it is not the case that
a black electron has some hardness, but we simply don’t know it.

Instead, it does not even make sense to speak about the hardness of a
black electron. Hardness is not a property which a black electron can have.

Christian Ferko Entanglement July 21, 2022 61 / 64


Epilogue: Preview of tensor products, factorizability,
and more.

Christian Ferko Entanglement July 21, 2022 62 / 64


Taking stock and looking ahead.

We have learned a good deal of quantum mechanics in this (long) talk,


focusing on the Hilbert space where our states live, and the Hermitian
operators which correspond to measurable quantities.

However, all of our discussion involved only a single particle Hilbert space.
To understand entanglement, we must consider Hilbert spaces like

HAB = HA ⊗ HB ,

where ⊗ is the tensor product.

A state |Ψ⟩ ∈ HAB is said to be entangled if |Ψ⟩ cannot be written as


|ΨA ⟩ ⊗ |ΨB ⟩ for any pair of states |ΨA ⟩ ∈ HA , |ΨB ⟩ ∈ HB .

That is, a state is entangled if it does not factorize as a tensor product of


states on each of the factor Hilbert spaces.

Christian Ferko Entanglement July 21, 2022 63 / 64


Mixed states and density matrices.

There is another fact that we have not discussed. An ensemble of 100


quantum particles, all in the state √12 (|0⟩ + |1⟩), is not the same as an
ensemble of 100 quantum particles of which 50 are in the state |0⟩ and 50
of which are in the state |1⟩.

The former ensemble corresponds to a pure state and the latter is a


mixed state. If we continue these talks, we will learn that an entangled
pure state on HAB gives rise to mixed states on HA and on HB .

Beautifully, an information-theoretic quantity called the von Neumann


entropy of these mixed states is what measures the entanglement of the
parent pure state. This is the beginning of quantum information proper.

Thank you for your attention!

Christian Ferko Entanglement July 21, 2022 64 / 64

You might also like