The Mathematics of Entanglement

Summer School at Universidad de los Andes

Fernando G. S. L. Brandao, Matthias Christandl, Aram W. Harrow and Michael Walter
27-31 May, 2013

Contents
Lecture 1 - Quantum States
1.1 Probability theory and Tensor products
1.2 Quantum Mechanics . . . . . . . . . . .
1.2.1 Measurements . . . . . . . . . .
1.3 Mixed states . . . . . . . . . . . . . . .
1.4 Composite systems and Entanglement .
1.4.1 Partial trace . . . . . . . . . . .

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

3
3
4
4
5
5
6

Lecture 2 - Quantum Operations

8
2.1 Measurements, POVMs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2 Unitary Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3 General Time Evolutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Lecture 3 - Quantum Entropy
3.1 Shannon Entropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2 Typical . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.3 Quantum compression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

11
11
11
13

Lecture 4 - Teleportation and entanglement transformations

4.1 Teleportation . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2 LOCC entanglement manipulation . . . . . . . . . . . . . . .
4.3 Distinguishing quantum states . . . . . . . . . . . . . . . . .
4.4 Entanglement dilution . . . . . . . . . . . . . . . . . . . . . .

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

14
14
15
16
16

Lecture 5 - Introduction to the Quantum Marginal Problem

5.1 The Quantum Marginal Problem or Quantum Representability Problem
5.1.1 Physical Motivation . . . . . . . . . . . . . . . . . . . . . . . . .
5.2 Quantum Marginals for 3 Parties . . . . . . . . . . . . . . . . . . . . . .
5.2.1 Warm-Up: 2 Parties . . . . . . . . . . . . . . . . . . . . . . . . .
5.2.2 3 Parties of Qubits . . . . . . . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

17
17
17
18
18
19

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

Lecture 6 - Monogamy of Entanglement

20
6.1 Symmetric Subspace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
6.2 Application to Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

Lecture 7 - Separable states, PPT and

7.1 Mixed-state entanglement . . . . .
7.2 The PPT test . . . . . . . . . . . .
7.2.1 Bound entanglement . . . .
7.3 Entanglement witnesses . . . . . .
7.4 CHSH game . . . . . . . . . . . . .

Bell inequalities
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .

Lecture 8 - Exact Entanglement Transformations

8.1 Three qubit subsystems, part 2 . . . . . . . . .
8.2 Exact Entanglement Transformation . . . . . .
8.2.1 Quantum Instrument . . . . . . . . . . .
8.2.2 LOCC as a Quantum Operation . . . .
8.2.3 SLOCC: Stochastic LOCC . . . . . . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

23
23
23
24
25
26

.
.
.
.
.

27
27
28
28
28
28

Lecture 9 - Quantum de Finetti theorem

30
9.1 de Finetti . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
9.2 Quantum Key Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Lecture 10 - Computational complexity of entanglement
32
10.1 More on CHSH games . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
10.2 Computational complexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
Lecture 11 - Three Qubit Entanglement Polytopes
37
11.1 Quantum Marginal Problem for Three Parties . . . . . . . . . . . . . . . . . . . . . . 37
Lecture 12 - High dimensional entanglement

39

Lecture 13 - LOCC distinguishability

42
13.1 Data Hiding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
13.2 Better de Finetti theorems for 1-LOCC measurements . . . . . . . . . . . . . . . . . 43
Lecture 14 - Representation Theory and Spectrum
14.1 Representation Theory . . . . . . . . . . . . . . .
14.1.1 Schur-Weyl Duality . . . . . . . . . . . .
14.1.2 Computing mj . . . . . . . . . . . . . . .
14.2 Spectrum Estimation . . . . . . . . . . . . . . . .
14.2.1 Application Keyl-Werner Relation . . . .

Estimation
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

46
46
47
47
48
48

Lecture 15 - Proof of the 1-LOCC quantum de Finetti theorem

49
15.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
15.2 Entropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

The Mathematics of Entanglement - Summer 2013

27 May, 2013

Quantum States
Lecturer: Fernando G.S.L. Brand
ao

Lecture 1

Entanglement is a quantum mechanical form of correlation, which appears in many areas, such
as condensed matter physics, quantum chemistry, and other areas of physics. This week we will
discuss a perspective from quantum information, which means we will abstract away the underlying
physics, and make statements about entanglement that apply independent of the underlying physical system. This will also allow us to discuss information-processing applications, such as quantum
cryptography.

1.1

Probability theory and Tensor products

Before discussing quantum states, we explain some aspects of probability theory, which turns out
to have many similar features.
Suppose we have a system with d possible states, for some integer d, which we label by 1, . . . , d.
Thus a deterministic state is simply an element of the set {1, . . . , d}. The probabilistic states are
probability distributions over this set, i.e. vectors in Rd+ whose entries sum to 1. The notation
Rd+ means that the entries are nonnegative. Thus, a probability distribution p = (p(1), . . . , p(d))
P
satisfies dx=1 p(x) = 1 and p(x) 0 for each x. Note that we can think of a deterministic state
x {1, . . . , d} as the probability distribution where p(x) = 1 and all other probabilities are zero.
n
Composition. Suppose we are given p Rm
+ and q R+ which correspond to independent
probability distributions. Their joint distribution is given by the vector

p(1)q(1)
p(1)q(2)

..

p q :=
p(1)q(n) .

..

.
p(m)q(n)
We have introduced the notation to denote the tensor product, which in general maps a pair of
vectors with dimensions m, n to a single vector with dimension mn. Later we will also consider
the tensor product of matrices. If Mn denotes n n matrices, and we have A Mn , B Mm then
A B Mnm is the matrix whose entries
are all

 possible products of an entry of A and an entry
a11 a12
of B. For example, if n = 2 and A =
then A B is the block matrix
a21 a22


a11 B a12 B
.
a21 B a22 B
One useful fact about tensor products, which simplifies many calculations, is that
(A B)(C D) = AC BD.
We also define the tensor product of two vector space V W to be the span of all v w for
v V and w W . In particular, observe that Cm Cn = Cmn .
3

1.2

Quantum Mechanics

We will use Dirac notation in which a ket |i denote a column vector in a complex vector space,
i.e.

1
2

|i = . Cd .
..
d
The bra h| denotes the conjugate transpose, i.e.
h| = 1 2

d .

Combining a bra and a ket gives a bra[c]ket, meaning an inner product

h|i =

d
X

i i .

i=1

In this notation the norm is

v
u d
uX
p
|i |2 .
kk2 = h|i = t
i=1

Now we can define a quantum state. The quantum analogue of a system with d states is the
d-dimensional Hilbert space Cd . For example, a quantum system with d = 2 is called a qubit. Unit
vectors |i Cd , where h|i = 1, are called pure states. They are the analogue of deterministic
states in classical probability theory. For example, we might define the following pure states of a
qubit:
 
 
1
1
0
1
, |+i = (|0i + |1i), |i = (|0i + |1i).
, |1i =
|0i =
1
0
2
2
Note that both pairs |0i, |1i and |+i, |i form orthonormal bases of a qubit.

1.2.1

Measurements

A projective measurement
P is a collection of projectors {Pk } such that Pk Md for each k, Pk = Pk ,
Pk Pk0 = k,k0 Pk and k Pk = I. For example, we might measure in the computational basis, which
consists of the unit vectors |ki with a one in the k th position and zeros elsewhere. Thus define

0
0

..

1
Pk = |kihk| =

.
.

.
0
0

which is the projector onto the one-dimensional subspace spanned by |ki.

4

Borns rule states that Pr[k], the probability of measurement outcome k, is given by
Pr[k] = h|Pk |i.
As an exercise, verify that this is equal to tr(Pk |ih|). In our example, this is simply |k |2 .
Example. If we perform the measurement {|0ih0|, |1ih1|} on |+i, then Pr[0] = Pr[1] = 1/2. If
we perform the measurement {|+ih+|, |ih|}, then Pr[+] = 1 and Pr[] = 0.

1.3

Mixed states

Mixed states are a common generalization of probability theory and pure quantum mechanics. In
general, if we have an ensemble of pure quantum states |x i with probabilities p(x), then define
the density matrix to be
X
=
p(x)|x ihx |.
x

The vectors |x i do not have to be orthogonal.

Note that is always Hermitian, meaning = . Here denotes the conjugate transpose, so
that (A )i,j = Aj,i . In fact, is positive semi-definite (PSD). This is also denoted 0. Two
equivalent definitions (assuming that = ) are:
1. For all |i, h||i 0.
2. All the eigenvalues of are nonnegative. That is,
X
=
i |i ihi |

(1.1)

for an orthonormal basis {|1 i, . . . , |d i} with each i 0.

Exercise: Prove that these definitions are equivalent.
P
P
A density matrix should also have trace one, since tr = x p(x)hx |x i =
P x p(x) = 1.
Conversely, any PSD matrix with trace one can be written in the form x p(x)|x ihx | for
some probability distribution p and some unit vectors {|x i}, and hence is a valid density matrix.
This is just based on the eigenvalue decomposition: we can always take p(x) = x in (1.1).
Note that this decomposition
isnot unique in general. For example, consider the maximally

1/2 0
mixed state = I/2 =
. This can be decomposed either as 21 |0ih0| + 12 |1ih1| or as
0 1/2
1
1
1
1
2 |+ih+| + 2 |ih|, or indeed as 2 |uihu| + 2 |vihv| for any orthonormal basis {|ui, |vi}.
Borns rule. For pure states if we measured {Pk } then the probability of outcome k would be
tr(Pk |ih|). For mixed states this becomes Pr[k] = tr(Pk ) by linearity.

1.4

Composite systems and Entanglement

We will work always with distinguishable particles. A pure state of two quantum systems is given
by a unit vector |i in the tensor product Hilbert space Cn Cm
= Cnm .
For example, if particle A is in the pure state |iA and particle B is in the pure state |iB
then their joint state is |iAB = |iA |iB . If |iA Cn and |iB Cm , then we will have
|iAB Cmn .
5

This should have the property that if we measure one system, say A, then we should obtain
the same result in this new formalism that we would have had if we treated the states separately.
If we perform the projective measurement {Pk } on system A then this is equivalent to performing
the measurement {Pk I} on the joint system. We can then calculate
Pr[k] = h|Pk I|i
=A h|B |i(Pk I)|iA |iB
= h|Pk |ih|i
= h|Pk |i
Just as there are joint distributions for which two random variables are not independent (e.g.,
...), there are quantum states which are cannot be written as a tensor product |i |i for any
choice of |i, |i. In quantum mechanics, this situation can even occur for pure states of the system.
For example, consider the EPR pair (we will see the reason for the name later):
|+ i =

|0i |0i + |1i |1i

.
2

We say that a pure state |i is entangled if, for any |i, |i, we have |i =
6 |i |i.
Entangled states have many counterintuitive properties. For example, suppose we measure the
state |+ i using the projectors {Pj,k = |jihj| |kihk|}. Then we can calculate
Pr[(0, 0)] = h+ |P0,0 |+ i = h+ | |0ih0| |0ih0| |+ i =

1
2

1
2
Pr[(0, 1)] = 0
Pr[(1, 1)] =

Pr[(1, 0)] = 0
The outcomes are perfectly correlated.
However, observe that if we measure in a different basis, we will also get perfect correlation.
Consider the measurement with outcomes
{| + +ih+ + |, | + ih+ |, | +ih + |, | ih |,
where we have used the shorthand | + +i := |+i |+i, and similarly for the other three. Then one
can calculate (and doing so is a good exercise) that, given the state |+ i, we have
1
Pr[(+, +)] = Pr[(, )] = ,
2
meaning again there is perfect correlation.

1.4.1

Partial trace

Suppose we have AB D(Cn Cm ). We would like a quantum analogue of the notion of a marginal
distribution in probability theory.

Define the reduced state of A to be

A := trB AB
X
trB (AB ) =
(I hk|)AB (I ketk),
k

where {|ki} is any orthonormal basis on B. The operation trB is called a partial trace.
We observe that if we perform a measurement {Pj } on A, then we have
Pr[j] = tr(Pj I)AB = tr(Pk trB (AB )) = tr(Pk A ).
Thus the reduced state A perfectly reproduces the statistics of any measurement on the system A.

The Mathematics of Entanglement - Summer 2013

27 May, 2013

Quantum Operations
Lecturer: Matthias Christandl

Lecture 2

In this lecture we will talk about dynamics in quantum mechanics. First stating from measurements again, and then going to unitary evolutions and general quantum dynamical processes.

2.1

Measurements, POVMs

FIXME: Insert three figures from Matthias slides.

Consider a quantum measurement as a box, applied to a mixed quantum state , with possible
outcomes labelled by i: In the previous lecture, we considered projective measurements given by
orthogonal projectors {Pi }, with Borns rule Pr(i) = tr(Pi ).
Another common way to think about these is the following:
PWe can associate to any projective
measurement an observable A with eigendecomposition A = i ai Pi , where we think of the ai as
the values that the observable attains for each outcome (e.g., the value the measurement device
displays,
the position of a pointer, . . . ). Then the expectation value of A in the state is tr(A) =
P
i ai tr(Pi ).
But is this the most general measurement allowed in quantum mechanics? It turns out that
this is not the case. Suppose we have a quantum state A on Cd and we consider the joint state
A |0ih0|B , with |0iB Cd the state of an ancillary particle. Let us perform a projective
0
0
measurement {Pi } on the joint system (i.e., the Pi are orthogonal projectors on Cd Cd = Cdd ).
Then the probability of measuring i is
Pr(i) = tr (Pi (A |0ih0|B )) .
Using the partial trace, we can rewrite this as follows
Pr(i) = trA (trB (Pi (A |0ih0|B )))
= trA (h0|B Pi |0iB A )
= tr(Qi A ),
where Qi := h0|B Pi |0iB . Thus the operators (Qi ) allow us to describe the measurement statistics
without having to consider the state of the ancillary system. What are the properties of Qi ? First,
it is PSD:
h|Qi |i = h|A h0|B Pi |0iN |iA 0,
since Pi 0. Second the Qi sum up to the identity:
X
X
X
Qi =
h0|B Pi |0iB = h0|B
Pi |0iB = h0|B I|0iB = IA .
i

The
P converse of the above is also true: Whenever we are given a set of PSD matrices Qi 0
with i Qi = I, we can always find a projective measurement {Pi } on a larger system A B such
that
tr(Qi A ) = tr(Pi (A |0ih0|B )).
8

The generalized quantum measurements we obtain in this way are called positive operator-valued
measure(ment)s (POVMs). Note that since the Qi s are not necessarily orthogonal projections,
there is no upper bound on the number of elements in a POVM.
Example 1. Consider two projective measurements, e.g. {|0ih0|, |1ih1|} and {|+ih+|, |ih|}.
Then we can define a POVM as a mixture of these two:
1
Q0 = |0ih0|,
2
1
Q1 = |1ih1|,
2
1
Q2 = |+ih+|,
2
1
Q3 = |ih|.
2
P
It is clear that k Qk = I. One way of thinking about this POVM is that with probability 1/2 we
measure in the computational basis, and with probability 1/2 in the |i basis.
Example 2. The quantum state of a qubit can always be written in the form
= (~r) =

1
(I + rx x + ry y + rz z ) ,
2

with the Pauli matrices

x =







0 i
1 0
0 1
.
, y =
, z =
i 0
0 1
1 0

Note that the Pauli matrices are traceless, so that the state has indeed trace one (its normalized).
We can the describe the state by a 3-dimensional vector ~r = (rx , ry , rz ) R3 . It turns out that
is PSD if, and only if, krk2 1. Therefore every quantum state of a qubit corresponds to a point
in a 3-dimensional sphere, called the Bloch sphere. A state is pure if, and only if, krk2 = 1.
Let us consider a collection of four pure statesP
{|ai ihai |}4i=1 that form a tetrahedron on the Bloch
sphere. Then, by symmetry of the tetrahedron, i |ai ihai | = I, so they form indeed a POVM.
FIXME: Insert tetrahedron figure.

2.2

Unitary Dynamics

Let |i be a quantum state and consider its time evolution according to the Schrodinger equation
for a time-independent Hamiltonian H. Then the state after some time t is given by
|t i = eiHt |i,
where we have set ~ = 1. The matrix U = eiHt describing the evolution of the system is a unitary
matrix, i.e. U U = U U = I.
Example 1. Ut = eit~e~ /2 with ~e R3 a unit vector and ~ = (x , y , z ) the vector of Pauli
matrices. We have
Ut (~r)Ut = (Rt~r),
where Rt rotates by an angle t around the axis ~e.
9

Example 2. The Hadamard unitary is given by H =

1
2



1 1
. Its action on the computa1 1

tional basis vectors is

H|0i = |+i,
H|1i = |i.

2.3

General Time Evolutions

There are more general possible dynamics in quantum mechanics than unitary evolution. One
possibility is that we add an acilla state |0ih0|B to A and consider a unitary dynamics UAB :
A B A0 B 0 on the joint state. The state of the A0 B 0 system is
U A |0ih0|B U

(2.1)

Suppose now we are only interested in the final state of the subsystem A0 . Then


A0 = trB 0 U A |0ih0|B U ,

(2.2)

where we traced out over subsystem B 0 . We can associate a map to this evolution as


(A ) = 0A = trB 0 U A |0ih0|B U .

(2.3)

What are the properties of ? First it maps PSD matrices to PSD matrices.We call this property
positivity. Second, it preserves the trace. We say the map is trace preserving.
Another more interesting property is that even the map Id, where Id is the identity map on
an auxiliary space of arbitrary dimension, is positive. We call this property completely positivity.
An important theorem (sometimes called Stinespring dilation) is that the converse also holds:
Any which is compeltely positive and trace preserving can be written as
(X = trB 0 (U X |0ih0|B U ) ,

(2.4)

with a unitary U : A B A0 B 0 . Therefore every quantum dynamics are in a one-to-one

correspondence to completely positive trace preserving maps, also called quantum operations.
Example: An example of a quantum operation is the so-called depolarization map: () =
(1 p) + pI/2. With probability 1 p the identity map is implemented, and with probability p
the state is destroyed and replaced by the maximally mixed one.

10

The Mathematics of Entanglement - Summer 2013

27 May, 2013

Quantum Entropy
Lecturer: Aram Harrow

3.1

Lecture 3

Shannon Entropy

In this part, we want to understand quantum information in a quantitative way. One of the
important concepts is entropy. But let us first look
P at classical entropy.
d
Given is a probability distribution p R+ , i pi = 1. The Shannon entropy of p is
H(p) =

pi log pi

(log is always to base to as we are talking about bits and units, convention limx70 x log x = 0).
Entropy quantifies uncertainty. We have maximal certainty for a deterministic distribution, e.g.
p = (1, 0, . . . , 0), H(p) = 0. The distribution with maximal uncertainty is p = ( d1 , . . . , d1 ), H(p) =
log d.
In the following we want to give Shannon entropy an operational meaning with help of the
problem of data compression. For this imagine you have a binary alphabet (d = 2) and you sample
n times independently and identically distribution from the distribution p = (, 1 ); we write
X1 , . . . , Xn i.i.d. {0, 1}. (Prob[Xi = 0] = Prob[Xi = 1] = 1 )

Typically, the number of 0s in the string n O( n and the number of 1s n(1 ) O( n. In

order to see why this is the case, consider the sum S = X1 + . . . , Xn (this equals the number of 1s
in the string). The expectation of this random variable is E[S] = E[X1 ] + . . . + E[Xn ] = n(1 ),
where we used the linearity of the expectation value. Furthermore the variance of S is V ar[S] =
V ar[X1 ] + . . . + V ar[Xn ] = nV ar[X1 ] = n(1 ) 41 . Here we used the independence of the
2 = (1)
random variable in the first equation and V ar[X1 ] = E[X12 ]E[X1 ]2 = (1)(1)

in the third. This implies that the standard deviation of S is smaller than 2n .
What does this have to do with compression? The
of possible n-bit
strings is

total number
(n/e)n
n
n!
n
n
|{0, 1} | = 2 . The number of strings with n 0s is n = (n)!((1)n)! = (n/e)n ((1)n/e)(1)n =
(1/)n (1/(1))n(1) where we used Stirlings approximation. We can rewrite this as exp(n log 1/+
(1 ) log 1/(1 )) = exp(nH(p)). Hence, we only need to store around exp(nH(p)) possible
strings, which we can do in a memory having nH(p) bits. (Note we ignored the fluctuations. If we

took them into account, we would only need additional O( n) bits.) This analysis easily generalises
to arbitrary alphabets (not only binary).

3.2

Typical

I now want to give you a different way of looking at this problem, a way that is both more rigorous
and will more easily generalise to the quantum case. This we will do with help of typical sets.
Again let X1 , . . . , Xn be i.i.d distributed with distribution p. The probability of a string is then
given by
11

Prob[X1 . . . Xn = x1 , . . . , xn ] = p(x1 )p(x2 ) p(xn ) where we used the notation (random variables are in capital letters and values in small letters)
p (n) = p p p (n times).
xn = (x1 , . . . , xn ) n
where = {1, . . . , d} is the alphabet and n denotes strings of length n over that alphabet
Note that

log pn (xn ) =

n
X

log p(xi ) nE[log p(xi )]

p
n V ar[log p(xi ] = nH(p) O( n)

i=1

where we used
E[log p(xi )] =

p(xi ) log p(xi ) = H(p)

Let us now define the typical set as the set of strings whose
Tp,n, = {xn | log pn (xn ) + nH(p)| n}
Then
> 0

lim pn Tp,n, = 1

Our compression algorithm simply keeps all the strings in the typical set and throws away all
others. Hence, all we need to know the size of the typical set. This is easy. Note that
xn Tp,n, = exp(nH(p) n) pn (xn ) exp(nH(p) + n)
Note
1 pn (Tp,n, ) |Tp,n, | min pn (xn )
where the minimum is over all strings in the typical set. This implies
1 |Tp,n, | exp(nH(p) n)
which is equivalent to
log |Tp,n, | nH(p) + n
Exercise: Show this is optimal. More precisely, show that we cannot compress to nR bits for
R < H(p) unless the error does not go to zero. Hint: Use Chebycheff ienquality: Let Z be a random
variable P rob[|Z E[Z]| kSD[Z]] 1/k 2 Possible simplifications: 1) pretend all strings to be
typical 2) use exactly nR bits.

12

3.3

Quantum compression

Probability distributions are replaced by density matrices n = (n times) If is a state of

a qubit then this state lives on a 2n dimensional space. The goal of quantum data compression is to
represent this state on a smaller dimensional subspace. Just as before in bits, we now measure the
size in terms of the number of qubits needed to represent that subspace, the log of the dimension.
It turns out to P
be possible (and optimal) to do this in nS() n where S is the von Neumann
entropy S() = i log i = H() = tr log , where the i are the eigenvalues of the density
operator.

13

The Mathematics of Entanglement - Summer 2013

28 May, 2013

Teleportation and entanglement transformations

Lecturer: Fernando G.S.L. Brand
ao

Lecture 4

Prologue: Post-measurement states

One loose thread from the previous lecture is to explain what happens to a quantum state after
we measurement. Consider a projective measurement {Pk }. (We saw yesterday in lecture 2 that
in fact these can simulate even generalized measurements.) Recall that outcome k occurs with
probability Pr[k] = tr(Pk ). Then if this measurement outcome occurs, we are left with the state
Pk Pk
.
tr(Pk )

(4.1)

Observe that this has the property that repeated measurements always produce the same answer
(although the same is not necessarily true of generalized measurements).
For a pure state |i, the post-measurement state is
Pk |i
.
kPk |ik

(4.2)

Equivalently, we can write Pk |i = p|i, where |i is a unit vector representing the postmeasurement state, and p is the probability of that outcome.

4.1

Teleportation

Suppose that Alice has a qubit |iA0 = c0 |0i + c1 |1i that she would like to transmit to Bob. If they
have access to a quantum channel, such as an optical fiber, she can of course simply give Bob the
physical system A0 whose state is |i. This approach is referred to as quantum communication.
However, if they have access to shared entanglement, then this communication can be replaced
with classical communication (while using up the entanglement). This is called teleportation.
The procedure is as follows. Suppose Alice and Bob share the state
|+ iAB =

|0, 0i + |1, 1i

,
2

and Alice wants to transmit |iA0 to Bob. Then Alice first measures systems AA0 in the basis
{|+ i, | i, | + i, | i}, defined as
|0, 0i |1, 1i

2
|0,
1i

|1, 0i

| i =
2

| i =

For ease of notation, define {|0 i, |1 i, |2 i, |3 i} := {|+ i, | i, | + i, | i}.

Outcome 0 corresponds to the unnormalized state


1
|+ ih+ |A0 A IB (|iA0 |+ iAB ) = |+ iA0 A |iB ,
2
14

meaning the outcome occurs with probability 1/4 and when it does, Bob gets |i.
One can show (calculation omitted) that outcome i (for i {0, 1, 2, 3}) corresponds to
1
(|i ihi | IB ) |iA0 |+ iAB = |i iA0 A i |iB ,
2
where {0 , 1 , 2 , 3 } denote the four Pauli matrices {I, x , y , z }. The 1/2 means that each
outcome occurs with probability 1/4. Thus, transmitting the outcome i to Bob allows him to apply
the correction i and recover the state |i.
This protocol has achieved the following transformation of resources:
1bit entanglement + 2 bits classical communication 1 qubit quantum communication.
As a sanity check, we should verify that entanglement alone cannot be used to communicate.
To check this, the joint state after the measurement is
3

A0 AB =

1X
|i ihi |A0 A i |ih|i .
4
i=0

Bobs state specifically is

3

B =

I
1X
i |ih|i = .
4
2
i=0

Teleporting entanglement. This protocol also works if applied to qubits that are entangled
with other states. For example, Alice might locally prepare an entangled state |iRA0 and then
teleport qubit A0 to Bob. Then the state |i will be shared between Alices system R and Bobs
system B. Thus, teleportation can be used to create shared entanglement. Of course, it consumes
entanglement at the same rate, so we are not getting anything for free here.

4.2

LOCC entanglement manipulation

Suppose that Alice and Bob can freely communicate classically and can manipulate quantum systems under their control, but are limited in their ability to communicate quantumly. This class
of operations is called LOCC, meaning local operations and classical communication. It often
makes sense to study entanglement in this setting, since LOCC can modify entanglement from one
type to another, but cannot create it where it didnt exist before. What types of entanglement
manipulations are possible with LOCC?
One example is to map a pure state |iAB to (UA VB )|iAB , for some choice of unitaries
UA , VB .
A more complicated example is that Alice might measure her state with a projective measurement {Pk } and transmit the oucome to Bob, who performs a unitary Uk depending on the outcome.
This is essentially the structure of teleportation. The resulting map is
X
AB 7
(Pk Uk )(Pk Uk ).
k

One task for which we might like to use LOCC is to extract pure entangled states from a
noisy state. For example, we might want to map AB to |+ ih+ |. This problem is in general
15

called entanglement distillation since we are distilling pure entanglement out of noisy entanglement.
However, we typically consider it with a few variations. First, as with many information-theoretic
m , and
problems, we will consider asymptotic transformations in which we map n
AB to |+ ih+ |
seek to maximize the ratio m/n as n . Additionally, we will allow a small error (to be
formalized later) that goes to zero as n . Semi-formally, the distillable entanglement of is
nm
o
LOCC
ED (AB ) = lim max
: n m |+ ih+ |m .
n
n

4.3

Distinguishing quantum states

The maximum distinguishing bias that any measurement can achieve between a pair of states ,
is
D(, ) = max | tr(M ( ))|.
{M,IM }
0M I

It turns out that D(, ) = 21 kk1 , where kXk1 is the trace norm, defined as kXk1 = tr( X X).
For this reason, the distance D(, ) is also called the trace distance.
Using this language, we can define ED properly as
nm
o
LOCC
ED (AB ) = lim lim max
: n m , km |+ ih+ |m k1  .
0 n
n

4.4

Entanglement dilution

Suppose we wish to create a general entangled state AB out of pure EPR pairs. As with distillation,
we will aim to maximize the asymptotic ratio achievable while the error goes to zero. Define the
entanglement cost
nm
o
Ec (AB ) = lim lim min
k


: LOCC, k(|+ ih+ |m ) n
AB 1
0 n
n
In general, Ec and ED are both hard to compute. However, if AB is pure then there is a simple
beautiful formula.
Theorem 4.1. For any pure state |iAB ,
Ec (|ih|AB ) = ED (|ih|AB ) = S(A ) = S(B ),
where S() = tr log .

16

The Mathematics of Entanglement - Summer 2013

28 May, 2013

Introduction to the Quantum Marginal Problem

Lecturer: Matthias Christandl

Lecture 5

In the picture is the cover of a a book by Godel, Escher and Bach. You see it projects B, G or
E depending from where you project the light on. Is it possible to project any triple of letters in
this way? It turns out that the answer is no. For example, by geometric considerations there is no
way of projecting A everywhere.
The goal of this lecture is to introduce a quantum version of this problem!

5.1

The Quantum Marginal Problem or Quantum Representability Problem

Consider a set of n particles each with d-dimensions. The state lives in (Cd )n . We consider
different subsets of the particles Si {1, . . . , N } and suppose we are given quantum states in each
of this sets: Si . The question we want to address is whether they are compatible, i.e. does there
exist a quantum state {1,...,n} such that for all Si ,
trSic () = Si ,

(5.1)

with Sic the complement of Si in {1, . . . , n}.

5.1.1

Physical Motivation

This is a interesting problem from a mathematical point of view, but it is also a prominent problem
in the context of condensed P
matter physics and quantum chemistry. Consider a nearest-neighbours
Hamiltonian on a line H = i hi,i+1 , where hi,i+1 := hi,i+1 I{1,...,n} i,i+1 only acts on qubits i and
i + 1. A quantity of interest is the groundenergy of the model, given by the minimum eigenvalue
of H. We can write it variationally as
Eg = minh|H|i =
|i

min
{1,...,n}

tr({1,...,n} )

(5.2)

since the set of quantum states is convex and the extremal points are the pure states. Continuing,
Eg =

min tr({1,...,n} )
X
= min
tr hi,i+1 I{1,...,n}
{1,...,n}

= min

i,i+1

tr(hi,i+1 i,i+1 ).

(5.3)

Therefore
Eg =

min

{i,i+1 }compatible

X
i

17

tr(hi,i+1 i,i+1 ),

(5.4)

where the minimization is over {i,i+1 } which are compatible.

n
Observe that the initial maximization is over |i Cd
, which has dimension dn . In contrast,
the minimization in Eq. (5.4) is over nd variables. Therefore if we could solve the compatibility
problem we could solve the original one in a much more efficient way. Unfortunately this is not
a good strategy and in fact one can show that the compatibility problem is computationally hard
(NP-hard and even QMA-hard).
There is an interesting connection of the representability problem with quantum entropies. A
important inequality of the von Neumann entropy of three subsystems ABC is strong subadditivity:
S(AB) + S(BC) S(B) + S(ABC).

(5.5)

Clearly this inequality puts restrictions on compatible states. More interestingly, one can also use
results from the quantum marginal problem to give a proof of the inequality.

5.2

Quantum Marginals for 3 Parties

A particular case of the marginal problem is the following: given three quantum states A , B and
C , are they compatible? In this case it is that the answer is yes, just consider ABC = A B C .
But what if we require that the global state ABC is pure? I.e. we would like to have a pure
state |iABC such that
trAB (|ih|ABC ) = C , trAC (|ih|ABC ) = B , trBC (|ih|ABC ) = A .

(5.6)

Then just taking the tensor product of the reduced states is not an option any more.

Example 1: A = B = C = I/2 are compatible, with the GHZ state (|0, 0, 0i + |1, 1, 1i)/ 2
being a possible
extension. A = B = C = I/2 are compatible, with the GHZ state (|0, 0, 0i +

|1, 1, 1i)/ 2 being a possible extension.

Example 2: Suppose A , B and C are compatible. Are 0A = UA A UA , 0B = UB B UB , and
0C = UC C UC compatible too? The answer is yes. Indeed if |iABC was an extension of A , B
and C , then UA UB UC |iABC is an extension of 0A , 0B and 0C .
Therefore we see that the property of A , B , C being compatible only depends on the spectra
A , B and C of A , B and C . Here A := (A,1 , . . . , A,d ) with A,i the eigenvalues of A .

5.2.1

Warm-Up: 2 Parties

Given A and B , are they compatible?

A useful way of writing a bipartite pure state |iAB is in its Schmidt form:
X
|iAB =
si |ei i |fi i

(5.7)

for orthogonal basis {|ei i} and {|fi i} o A and B, respectively. The numbers {si } are called Schmidt
values of |iAB . The reductions of |iAB are
X
A =
s2i |ei ihei |
(5.8)
i

18

and
A =

s2i |fi ihfi |

(5.9)

Therefore we see that the eigenvalues of A and B are equal and given by {s2i }.
Going back to the compatibility question, we then see from the discussion above that A and
B are compatible if, and only if, they have the same spectrum.

5.2.2

3 Parties of Qubits

Consider A , B and C each acting on C2 . Then since A = (A,1 , 1 A,1 ), the compatible region
is a subset of R3 . This have a simple algebraic characterization (shown in []):
B
C
A
max + max 1 + max

plus all possible cyclic permutations of the labels.

19

(5.10)

The Mathematics of Entanglement - Summer 2013

28 May, 2013

Monogamy of Entanglement
Lecturer: Aram Harrow

Lecture 6

Today, I will discuss a property of entanglement known as monogamy.

Consider a Hamiltonian that has two-body interactions
H=

Hij

<i,j>

where < i, j > are the edges in the interaction graph.

We will consider the rather crude approximation that every particle interacts with any other
particle in the same way; this approximation is known as mean field approximation
H

1
n

X
1i<jn

It is then folklore that the ground state has the form n

Example: Hij = Fij , where F is the swap operator defined by
F |i|i = |i|i
The +1 eigenstates are | i, | i and | i + | i. The -1 eigenstate is | i | i, the singlet.
The Hamiltonian would like that every two particle reduced density matrix is in the singlet
state. Now if the global state |iABC has a reduced state AB = trC |ih|ABC that is indeed a
pure state in singlet form, then the global state has to be of the form
|iABC = (| i | i)AB |iC .
Then we immediately see that the other pairs of particles are not entangled.
This turns out to be a general feature of such systems.
R
Theorem (quantum de Finetti): |i is symmetric state on (C D )k+n then trn |ih| d() k ,
where is a distribution over density matrices on C D
This is a quantum version of de Finettis theorem from statistics. The important consequence
of this theorem is that the remaining k particles are not entangled. Since the ground states of mean
field systems are permutation invariant this means that these ground states are not entangled, and
hence in some sense classical.
We will now introduce some mathematical tools needed to prove this theorem. The first is the
symmetic subspace

6.1

Symmetric Subspace

Let Sn be the group of permutations of n objects. Note that it contains n! elements. Now fix D
and a permutation Sn . Let P act on (C D )n :
P |i1 i |in i = |i1 (1) i |i1 (n) i.
20

The symmetric subspace is defined as the set of vectors that are invariant under the action of the
symmetric group
Symn (C D ) = {|i (C D )n : P |i = |i Sn }.
Example:D = 2, n = 2
Sym2 (C 2 ) = span{|00i, |11i, |01 + 10i}
D = 2, n = 2
Sym2 (C 2 ) = span{|000i, |111i, |001 + 010 + 100i, |101 + 011 + 110i}
n
The general
P construction is as follows. For this define the type of a string x = (x1 , . . . , xn ) as
n
type(x ) = i exi , where ej is the basis vector with a 1 in the jth position. t = (t1 , . . . , td ) is a
type if t1 + t2 + td = n and the ti are natural numbers. For every type t the vector
 1/2 X
n
|t i =
|xn i
t


where nt is the multinomial coefficient. Symn (C D ) = span{|t i} We can now compute the dimension of the symmetric subspace. Note that we can interpret this number
as the number of ways


in which you can arrange n balls into D buckets. There are n+D1
ways
of doing this, which is
n
therefore the dimension.
A useful way for calculations involving the symmetric subspace are the following two characterisations of the P
projector onto the symmetric subspace
1
1) sym = n!
P where the sum is over all permutations
R
sym
2) tr sym = d|ih|n where we integrate over the unit vectors in C D with the uniform


R
n+D1
n
D
measure d normalised
n
R to d = 1. Note that tr sym = dim Sym (C ) =
Example:
R n = 1: d|ih| = I/D
n = 2: d|ih|2 = Psym /(D(D + 1)/2) = (I + SW AP )/(D(D + 1))
We can prove 2) either by representation theory (using Schurs lemma) or by rewriting the
integral over unit vectors as an integral over Gaussian vectors and then using Wicks theorem to
solve the integral.
Note that sym projector onto the symmetric subspace, hence
a) |i Symn (C D ), sym |i = |i b) |i (C D )n , sym |i Symn (C D ) c) sym = sym

6.2

Application to Estimation

that approximates |i. We could now use different

Given |in , we want to output an estimate |i
notions of approximation. Here, we want to maximise the overlap
2k

E(|h|i|
)

R
= sym . In fact, we will
In order to do this, we will use the continuous POVM {Q }, dQ

|
n . The normalisation constant can be worked out as follows
choose Q = c|ih


Z
n+D1
n

c|ih| = csym /
,
n
21



|
n .
|ih
hence Q = n+D1
n
We now use this to solve our estimation problem:
Z
2k
2k

E(|h|i| = |h|i|
Pr[|]

|
n |ih|n )
where P rob[|]
= tr(|ih
This equals
Z

where we used

tr |ih

n+k
sym
n+k+D1
n+k

n+k

n+k

|ih|



n+D1
=
n

d|ih|n+k .

22

n+D1
n


n+kD1
n

1 Dk/n

The Mathematics of Entanglement - Summer 2013

29 May, 2013

Separable states, PPT and Bell inequalities

Lecturer: Fernando G.S.L. Brand
ao

Lecture 7

Recall from yesterday this theorem.

Theorem 7.1. For any pure state |iAB ,
Ec (|ih|AB ) = ED (|ih|AB ) = S(A ) = S(B ),
where S() = tr log .
As a result, many copies of a pure entangled state can be (approximately) reversibly transformed
into EPR pairs and back again. Up to a small approximation error and inefficiency, we have
|in
LOCC |+ inS(A ) .
AB

7.1

Mixed-state entanglement

For pure states, a state is entangled if its not a product state. This is easy to check, and we can
even quantify the amount of entanglement (using Theorem 7.1) by looking at the entropy of one of
the reduced density matrices.
But what about for mixed states? Here the situation is more complicated.
Definition 1. Define the set of separable states Sep to be the set of all AB that can be written as
X
pi |i ihi |A |i ihi |B .
(7.1)
i

Definition 2. A state is entangled if it is not separable.

We should check that this notion of entanglement makes sense in terms of LOCC. And indeed,
separable states can be created using LOCC: Alice samples i according to p, creates |i i and sends
i to Bob, who uses it to create |i i. On the other hand, entangled states cannot be created from a
separable state by using LOCC. In other words, the set Sep is closed under LOCC.

7.2

The PPT test

It is in general hard to test, given a state AB , whether Sep. Naively we would have to check
for all possible decompositions of the form in (7.1). So it is desirable to find efficient tests that
work at least some of the time.
One such test is the Positive Partial Transpose, or PPT, test. The partial transpose can be
thought of as (T id), where T is the transpose map. More concretely, if
X
XAB =
ci,j,k,l |iihj|A |kihl|B
i,j,k,l

23

then the partial transpose is

TA
XAB
=

ci,j,k,l (|iihj|A )T |kihl|B

i,j,k,l

ci,j,k,l |jihi|A |kihl|B

i,j,k,l

The PPT test asks whether TA is positive semidefinite. If so, we say

Pthat is PPT.
Observe that all separable states are PPT. This is because if = i pi |i ihi |A |i ihi |B ,
then
X
TA =
pi |i ihi |A |i ihi |.
i

This is still a valid density matrix and in particular is positive semidefinite (indeed, it is also in
Sep).
Thus, Sep implies PPT. The contrapositive is that 6 PPT implies 6 Sep. This
gives us an efficient test that will detect entanglement in some cases.
Are there in fact any states that are not in PPT? Otherwise this would not be a very interesting
test.
Examples
1. |+ iAB =

|0,0i+|1,1i

.
2

Then
1
(|0, 0ih0, 0| + |1, 0ih0, 1| + |0, 1ih1, 0| + |1, 1ih1, 1|)
2

1/2 0
0
0
0
0 1/2 0
= 1 SWAP.
=
0 1/2 0
0 2
0
0
0 1/2

A
|+ ih+ |TAB
=

This has eigenvalues (1/2, 1/2, 1/2, 1/2), meaning that |+ ih+ |AB 6 PPT. Of course, we
already knew that |+ i was entangled.
2. Lets try an example where we dont already know the answer, like a noisy version of |+ i.
Let
I
= p|+ ih+ | + (1 p) .
4
Then one can calculate min (TA ) = p2 +

1p
4

which is < 0 if and only if p > 1/3.

Maybe PPT = Sep? Unfortunately not. In D(C2 C3 ) (i.e. density matrices in which one
system has 2 dimensions and the other has 3) then all PPT states are separable. But for larger
systems, e.g. 3x3 or 2x4, then there exist PPT states that are not separable.

7.2.1

Bound entanglement

Theorem 7.2. If AB PPT then ED () = 0.

To prove this we will establish two properties of the set PPT:
24

1. PPT is closed under LOCC.

Consider a general LOCC protocol. This can be thought of as Alice and Bob alternating general measurements and sending each other the outcomes. When Alice makes a measurement,
this transformation is
(M I)AB (M I)
AB 7
.
tr((M M I)AB )
After Bob makes a measurement as well, depending on the outcome, the state is proportional
to
(M N )AB (M N ),
and so on. The class SLOCC (stochastic LOCC) consists of outcomes that can be obtained
with some positive probability, and we will see later that this can be characterized in terms
of (M N )AB (M N ).
We claim that if AB PPT then (M N )AB (M N ) PPT. Indeed
A
((M N )AB (M N ))TA = (M T N )TAB
(M T N ) .
A
Now TAB
0 and XY X 0 whenever Y 0, implying that ((M N )AB (M N ))TA 0.

2. PPT is closed under tensor product. If AB , A0 B 0 PPT, then (AB A0 B 0 ) PPT. Why?
Because
T 0
A
AA0 B 0 0.
(AB A0 B 0 )TAA0 = TAB
Proof of Theorem 7.2. Assume towards a contradiction that PPT and ED () > 0. Then for
+
any  > 0 there exists n such that n
AB can be transformed to | i using LOCC up to error . Since
n
PPT, is also PPT and so is the output of the protocol, which we call . Then TA 0 and
k |+ ih+ |k1 . If we had  = 0, then this would be a contradiction, because is in PPT
and |+ ih+ | is not. We can use an argument based on continuity (details omitted) to show that
a contradiction must appear even for some sufficiently small  > 0.
If is entangled but ED () = 0, then we say that has bound entanglement meaning that it
is entangled, but no pure entanglement can be extracted from it. By Theorem 7.2, we know that
any state in PPT but not Sep must be bound entangled.
Open question: A major open question (the NPT bound entanglement question) is whether
there exist bound entangled states that have a non-positive partial transpose.

7.3

Entanglement witnesses

Sep is a convex set, meaning that if , Sep and 0 1 then + (1 ) Sep. Thus the
separating hyperplane theorem implies that for any 6 Sep, there exists a Hermitian matrix W
such that
1. For all Sep, tr(W ) 0
2. tr(W ) < 0.
Example: consider the state = |+ ih+ |. Let W = I 2|+ ih+ |. As an exercise, show that
tr(W ) 0 for all Sep. We can also check that tr(W ) = 1.
Observe that an entanglement witness W needs to be chosen with a specific in mind. An an
exercise, show that no W can be a witness for all entangled states of a particular dimension.
25

7.4

CHSH game

One very famous type of entanglement witness is called a Bell inequality. In fact, these bounds rule
out not only separable states but even classically correlated distributions over states that could
be from a theory more general than quantum mechanics. Historically, Bell inequalities have been
important in showing that entanglement is an inescapable, and experimentally testable, part of
quantum mechanics.
The game is played by two players, Alice and Bob, together with a Referee. The Referee choose
bits r, s at random and sends r to Alice and s to Bob. Alice then sends a bit a back to the Referee
and Bob sends the bit b to the Referee.
R
r

s
a

Alice and Bob win if a b = r s, i.e. they want a b to be chosen according to this table:
r
0
0
1
1

s
0
1
0
1

desired a b
0
0
0
1

One can show that if Alice and Bob use a deterministic strategy, their success probability will
be 3/4. However, using entanglement they can achieve a success probability of cos2 (/8)
0.854 . . . > 3/4. This strategy, together with the payoff function (+1 if they win, -1 if they lose),
yields an entanglement witness, and one that can be implemented only with local measurements.

26

The Mathematics of Entanglement - Summer 2013

29 May, 2013

Exact Entanglement Transformations

Lecturer: Matthias Christandl

8.1

Lecture 8

Three qubit subsystems, part 2

Last lecture we looked at quantum states of three parties |iABC C2 C2 C2 . We say that if
A , B , and C are the reductions of |iABC then
B
C
A
max + max 1 + max .

(8.1)

Let us prove it. We have

B
A
max + max =

h|A |i +

max

|i,k|ik2 =1

max

h|B |i

|i,k|ik2 =1

= max tr(A |ih|A ) + max tr(B |ih|B )

A

=
=

max tr (AB (|ih|A IB + IA |ih|B ))

A ,B

max tr AB (|ih|A (|ih|B + |0 ih0 |B ) + (|ih|A + |0 ih0 |A ) |ih|B )

A ,B

max tr(A B(IA IB + |ih|A |ih|B ))

A ,B

= 1 + max tr(A B(|ih|A |ih|B ))

A ,B

1 + max tr(A B(|ih|AB ))

|iAB

= =

AB
max

= C
max ,

(8.2)

where in the last equality we used that |iABC is pure.

To show that Eq. (8.1) is sufficient for a triple of states A , B , and C to be compatible, let
us consider the following Ansatz:
|iABC = a|0, 0, 0i + b|0, 1, 1i + c|1, 0, 1i + d|1, 1, 0i.

(8.3)

The A local density matrix for this state is

A = (a2 + b2 )|0ih0| + (c2 + d2 )|1ih1|.

(8.4)

B = (a2 + c2 )|0ih0| + (b2 + d2 )|1ih1|

(8.5)

C = (a2 + d2 )|0ih0| + (b2 + c2 )|1ih1|

(8.6)

Likewise,
and
need picture... (add explanation later)
Next lecture we will see how the mathematics of representation theory is useful for generalizing
this result to higher dimensions.
27

8.2

Exact Entanglement Transformation

In Fernandos lecture he considered asymptotic and approximate entanglement transformations.

Here we will consider the different regime of single-copy and exact transformations. Given two
multipartite states, |i and |i, which one is more entangled? One way to out an order on the
set of quantum states is to say |i is more entangled than |i if we can transform |i into |i by
LOCC.
A LOCC protocol is given by a sequence of measurements by one of the parties and classical
communication of the outcome obtained to the other. (see fig ....).

8.2.1

Quantum Instrument

Consider a quantum operation

(A ) = trB 0 (U (A |0ih0|B )U ) =

X
X
hi|B 0 U |0iB 0 A h0|B U |iiB =
Ei A Ei ,
i

(8.7)

with Ei := h0|B U |iiB . The Ei s are called Kraus operators of the .

Note that the partial trace is the same as performing a projective measurement on B 0 and
forgetting the outcome obtained. Suppose now that we would record the outcome instead. Then
conditioned on outcome i, the state is Ei A Ei . We can associate the following measurement to it:
X
(A ) =
Ei A Ei |iihi|.
(8.8)
i

The operation is also called a quantum instrument.

8.2.2

LOCC as a Quantum Operation

Going back to the LOCC protocol, Alice first measurement can be modelled by a set of Kraus
operators {Ai1 }. Then Bobs measurement, which can depends on Alices outcome, will be given
by {Bi1 ,i2 }, and so on. In terms of a quantum operation a n-round LOCC protocol can be written
as
X
() =
(Ai1 ,...,in . . . Ai1 Bi1 ,...,in . . . Bi1 i2 )(Ai1 ,...,in . . . Ai1 Bi1 ,...,in . . . Bi1 i2 )
i1 ...,in

|i1 , . . . , in ihi1 , . . . , in |A0 |i1 , . . . , in ihi1 , . . . , in |B 0

8.2.3

(8.9)

SLOCC: Stochastic LOCC

The general form of a LOCC operation (Eq. (8.9)) is daunting. It turns out that the whole picture
simplifies if we condition on measurement outcomes, which corresponds on only considering one of
the terms in the sum of Eq. (8.9). We call this operation Stochastic LOCC and it is an operation
which can only be implemented with some non-zero probability. A general form of SLOCC (for
three parties) is then
() = (A B C)(A B C) .
(8.10)
Note that p := tr(()) gives the probability that is implemented (with probability 1 p the
protocol fails and the state is transformed into something else).
28

Definition 3. |i SLOCC |i if there exists matrices A, B, C such that |i = (A B C)|i. We

say that |i and |i have the same type of entanglement if |i SLOCC |i and |i SLOCC |i.
Since we do not care about normalization, we can w.l.o.g. take A, B, C to be SL(d) matrices
(with SL(d) the group of d d matrices of unit determinant). Therefore we see that the problem
of characterizing different entanglement classes is equivalent to the problem of classifying the orbit
classes of SL(d) SL(d) SL(d).
In general the number of orbits is very big. Indeed the dimension of the space scales by d3 , but
the group only has 3d2 parameters. But the case of three qubits turns out to be simple and we
only have 6 different classes.
One is the class of fully separable states, with representative state
|0, 0, 0i.

(8.11)

The three following ones are the classes where only two parties are entangled, with representative
states
|+ iAB |0iC
(8.12)
and likewise for AC and BC.
The 5th class if the so-called GHZ class, with representative state
1
(|0, 0, 0i + |1, 1, 1i).
2

(8.13)

The final class is the so-called W class, with representative states

1
(|0, 0, 1i + |0, 1, 0i + |1, 0, 0i).
3

29

(8.14)

The Mathematics of Entanglement - Summer 2013

29 May, 2013

Quantum de Finetti theorem

Lecturer: Aram Harrow

9.1

Lecture 9

de Finetti

Let us remind ourselves

that the quantum de Finetti theorem states that for all |i Symn+k (C d ),
R
then trn |ih| d() k .
The intuition here is that measuring the last n systems and finding that they are each in state
implies that the remaining k systems are also in state .
Let us now do the math. Recall


Z
D+m1
m
D,m
d|phiihphi|
=
/
D1
then
trn |ih| = trn idk D,n |ih|idk D,n




Z
D+n1
k
n
k
n
0 D+n1
(id |ih| |ih|(id |ih| )
= trn dd
D1
D1
Z Z
=
dd0 |
v ih
v ||h||0 i|n
Z
d|
v ih
v |

(9.1)
(9.2)
(9.3)
(9.4)
(9.5)

D+n1

where we defined D1 (idk h|n )|i =: |

v i and where in the approximation we assume
that n is large.
If we had not inserted the projector on the left, in fact we obtain the equality
Z
Z
trn |ih| = d|
v ih
v | = dp |v ihv |
(9.6)
We now claim that |v i |ik on average:
Z

dp |hv ||ik |2 =

d|h
v ||ik |2
(9.7)

Z
D+n1
=
d|h|(idk |in )|ik |2
(9.8)
D1


Z
D+n1
=
tr(|ih| d|ih|n+k )
(9.9)
D1

 


 

D+n1
D+n+k1
D+n1
D+n+k1
=
/
tr(|ih|) =
/
D1
D1
D1
D1
(9.10)
30

We can now get a good lower bound from this by expanding the binomial coefficients into
factorials:

 

D+n1
D+n+k1
(n + 1) (n + D 1)
/
=
(1k/(n+1))D1 1kD/n.
(n + k + 1) (n + k + D 1)
D1
D1
Note that this bound is polynomial in n. This is tight. There exists, however, an improvement
to an exponential dependence in n at the cost of replacing product states by almost product states.
In order to conclude the proof of the quantum de Finetti theorem, we need to relate the trace
distance to the average we computed.
For this, we consider
the fidelity |h||i| between states |i and |i. If now |h||i| = 1  and

we expand |i = 1 |i + |i

|||ih| |ih|||1 = ||id ....||1 = 2 

9.2

Quantum Key Distribution

A surprising application of entanglement is quantum key distribution. Suppose Alice and Bob share
an EPR pair |i = 12 |00 + 11i, then the joint state of Alice Bob and a potential eavesdropper Eve
is |iABE s.th. trE |ih|ABE = |ih|AB it follows that |iABE = |iAB |iE
By measuring in their standard basis, Alice and Bob thus obtain a secret random bit r. They
can use this bit to send a bit securely with help of the Vernam one-time pad cipher: Lets call
Alices message m. Alice sends the cipher c = m r to Bob. Bob then recovers the message by
adding r: c r = m r r = m.
How can we establish shared entanglement between Alice and Bob? Alice could for instance
create the state locally and send it to Bob using a quantum channel (i.e. a glas fibre).
But how can we now verify that the joint state that Alice and Bob have after the transmission
is an EPR state?
Protocol 1) Alice sends halves of n EPR pairs to Bob
2) They choose randomly half of them and perform the CHSH tests. 3) They get key from the
remaining halves
There are many technical details that I am glossing over here. One is, how can you be confident
that the other halves are in this state? The de Finetti theorem! (the choice was permutation
invariant)
In order to make this applicable in actual implementations, one may use the exponential de
Finetti theorem (Renner) or the post-selection technique (Christandl, Konig, Renner).
Other issues: there may be noise on the line. It is indeed possible to quantum key distribution
even in this case, but here one needs some other tools mainly relating to classical information theory
(information reconciliation or privacy amplification).

31

The Mathematics of Entanglement - Summer 2013

30 May, 2013

Computational complexity of entanglement

Lecturer: Fernando G.S.L. Brand
ao

10.1

Lecture 10

More on CHSH games

We continue our discussion of the CHSH game.

R
r

s
a

Alice and Bob win if a b = r s, i.e. they want a b to be chosen according to this table:
r
0
0
1
1

s
0
1
0
1

desired a b
0
0
0
1

Deterministic strategies. Consider a deterministic strategy. This means that if Alice receives
r = 0, she outputs the bit a0 and if she receives r = 1, she outputs the bit a1 . Similarly, Bob
outputs b0 if he receives s = 0 and b1 if he receives s = 1.
There are four possible inputs. If they set a0 = a1 = b0 = b1 = 0, then they will succeed with
probability 3/4. Can they do better? For a deterministic strategy this can only mean winning with
probability 1. But this implies that
a0 b0 = 0
a0 b1 = 0
a1 b0 = 0
a1 b1 = 1
Adding this up (and using x x = 0) we find 0 = 1, a contradiction.
Randomized strategies. What if Alice and Bob share some correlated random variable and
choose a deterministic strategy based on this? Then the payoff is the average of the payoffs of each
of the deterministic strategies. Thus, there must always be at least one deterministic strategy that
does at least as well as the average. So we can assume that an optimal strategy does not make use
of randomness. (Exercise: what if they use uncorrelated randomness? Can this help?)
Quantum strategies. Now suppose they share the state |+ i. Define
|0 ()i = cos()|0i + sin()|1i
|1 ()i = sin()|0i + cos()|1i
32

Observe that {|0 ()i, |1 ()i} is an orthonormal basis for any choice of .
The strategy is as follows. Alice and Bob will each measure their half of the entangled state in
the basis {|0 ()i, |1 ()i} for some choice of that depends on their inputs. They will output 0
or 1, depending on their measurement outcome. The choices of are
Alice

r=0
r=1

=0
= /4

Bob

s=0
s=1

= /8
= /8

1
As an exercise show that Pr[win] = cos2 (/8) = 21 + 2
> 3/4.
2
Another way to look at the quantum strategy is in terms of local observables. Alice and Bobs
strategy can be described in terms of the matrices




1 0
1 0
A0 =
A0 =
0 1
0 1




1 1 1
1
1 1
B0 =
B1 =
2 1 1
2 1 1

Given a state |i, the value of the game can be expressed in terms of
1
h|(A0 B0 + A0 B1 + A1 B0 A1 B1 )|i = Pr[win] Pr[lose] = 2 Pr[win] 1.
4
We can define a Hermitian matrix W 0 by
1
W = (A0 B0 + A0 B1 + A1 B0 A1 B1 ).
4
Then, for any Sep,
1
3
1
tr(W 0 ) 2 max Pr[win] 1 = 1 =
4
2
2
Define W = I2 14 W 0 . Then for all Sep, tr(W ) 0, while tr(W |+ ih+ |) = 12 < 0.
Thus Bell inequalities define entanglement witnesses; moreover, ones that distinguish an entangled state even from separable states over unbounded dimension that are measured with possibly
different measurement operators.
There has been some exciting recent work on the CHSH game. One recent line of work has
been on the rigidity property, which states that any quantum strategy that comes within  of the
1
must be within 0 of the ideal strategy (up to some trivial changes). This
optimal value 21 + 2
2
is relevant to the field of device-independent quantum information processing, which attempts to
draw conclusions about an untrusted quantum device based only on local measurement outcomes.
(For more references see arXiv:1203.2976 and arXiv:1303.3081.)

33

10.2

Computational complexity

Problem 1. Weak membership.

Given AB D(Cn Cm ),  > 0, and the promise that either
1. AB Sep, or
2. D(, Sep) = minSep D(, ) ,
decide which is the case.
This is called weak membership because of the  > 0 parameter, which means we dont have
to worry too much about numerical precision.
There are many choices of distance measure D(, ). We could take D(, ) = 21 k k1 , as we
p
did earlier. Or we should use k k2 , where kXk2 := tr(X X).
Another important problem related to Sep is called the support function. Like weak membership,
it can be defined for any set, but we will focus on the case of Sep.
Problem 2. Support function of Sep.
Given M H(Cn Cm ) and  > 0, compute hSep (M ) , where
hSep (M ) := max tr(M ).
Sep

There is a sense in which problem 1

= problem 2, meaning that an efficient solution for one can
be turned into an efficient solution to the other. We omit the proof of this fact, which is a classic
result in convex optimization [M. Gr
otschel, L. Lovasz, A. Schrijver. Geometric Algorithms and
Combinatorial Optimization. 1988].
Efficiency. What does it mean for a problem to be efficiently solvable? If we parametrize a
problem by the size of the input, then we say a problem is efficient if n-bit inputs can be solved
in time polynomial in n, i.e. in time c1 nc2 for some constants c1 , c2 . This class of problems is
called P, which stands for Polynomial time. Examples include multiplication, finding eigenvalues,
solving linear systems of equations, etc.
Another important class of problems are those where the solution can be efficiently checked. This
is called NP, which stands for Nondeterministic Polynomial time. (The term nondeterministic is
somewhat archaic, and refers to an imaginary computer that randomly checks a possible solution
and needs only to succeed with some positive, possibly infinitesimal, probability.)
One example of a problem in NP is called 3-SAT. A 3-SAT instance is a formula over variables
x1 , . . . , xn {0, 1} consisting of an AND of m clauses, where each clause is an OR of three variables
or their negations. Denoting OR with , AND with , and NOT xi with x
i , an example of a formula
would be
(x1 , . . . , xn ) = (x1 x
4 x1 7) (
x2 x
7 x1 0) . . . .
Given a formula , it is not a priori obvious how we can figure out if it is satisfiable. One option is
to check all possible values of x1 , . . . , xn . But there are 2n assignments to check, so this approach
requires exponential time. Better algorithms are known, but none has been proven to run in time
better than cn for various constants c > 1. However, 3-SAT is in NP because if is satisfiable, then
there exists a short witness proving this fact that we can quickly verify. This witness is simply
a satisfying assignment x1 , . . . , xn . Given and x1 , . . . , xn together, it is easy to verify whether
indeed (x1 , . . . , xn ) = 1.
34

Figure 1: This figure is taken from the wikipedia article http://en.wikipedia.org/wiki/Clique (graph
theory). The 42 2-cliques are the edges, the 19 3-cliques are the triangles colored light blue and the
2 4-cliques are colored dark blue. There are no 5-cliques.
NP-hardness. It is generally very difficult to prove that a problem cannot be solved efficiently.
For example, it is strongly believed that 3-SAT is not in P, but there is no proof of this conjecture.
Instead, to establish hardness we need to settle for finding evidence that falls short of a proof.
Some of the strongest evidence we are able to obtain for this is to show that a problem is
NP-hard, which means that any problem in NP be efficiently reduced to it. For example, 3-SAT is
NP-hard. This means that if we could solve 3-SAT instances of length n in time T (n), then any
other problem in NP could be solved in time poly(T (poly(n))). In particular, if 3-SAT were in
P then it would follow that P = NP.
It is conjectured that P 6= NP, because it seems harder to find a solution in general than to
recognize a solution. This is one of the biggest open problems in mathematics, and all partial results
in this direction are much much weaker. However, if we assume for now that P 6= NP, then showing
a problem is NP-hard implies that it is not in P. And since thousands of problems are known to be
NP-hard1 it suffices to show a reduction from any NP-hard problem in order to show that a new
problem is also NP-hard. Thus, this can be an effective method of showing that a problem is likely
to be hard.
Theorem 10.1. Problems 1 and 2 are NP-hard for  = 1/ poly(n, m).
We will give only a sketch of the proof.
1. Argue that MAX-CLIQUE is NP-hard. This is a classical result that we will not reproduce
here. Given a graph G = (V, E) with vertices V and edges E, a clique is a subset S V such
that (i, j) E for each i, j S, i 6= j. An example is given in Fig. 1. The MAX-CLIQUE
problem asks for the size of the largest clique in a given graph.
2.
1

See this list: http://en.wikipedia.org/wiki/List of NP-complete problems. The terminology NP-complete refers to
problems that are both NP-hard and in NP.

35

Theorem 10.2 (Motzkin-Straus). Let G = (V, E) be a graph, with maximum clique of size
W . Then
X
1
pi pj ,
(10.1)
1
= 2 max
W
(i,j)E

where the max is taken over all probability distributions p.

3. Given a graph, define
M=

|i, jihi, j|.

(i,j)E

Then
maxhx| hx|M |xi |xi = max
|xi

Defining pi = |xi

|2 ,

kxk2 =1

|xi |2 |xj |2 .

(i,j)E

we have recovered the RHS of (10.1).

4. We argue that
hSep (M ) = max ha, b|M |a, bi.
|ai,|bi

This is because Sep is a convex set, its extreme points are of the form |a, biha, b|, and the
maximum of any linear function over a convex set can be achieved by an extreme point.
5. Finally, we argue that maximizing over |a, bi is equivalent in difficulty to maximizing over
|a, ai.
What accuracy do we need here? If we want to distinguish a clique of size n (where there are n
1
vertices) from size n 1, then we need accuracy (1 n1
) (1 n1 ) 1/n2 . Thus, we have shown
that problem 2 is NP-hard for  = 1/n2 .

36

The Mathematics of Entanglement - Summer 2013

30 May, 2013

Three Qubit Entanglement Polytopes

Lecturer: Michael Walter

Lecture 11

Last time we talked about SLOCC (stochastic LOCC), where we can post-select on particular
outcomes
Given a class of states that can be interconverted by SLOCC into |i other by SLOCC, C =
{|i : |i |i}, a result by Dur-Vidal-Cirac says that
C := {(A B C)|i/k . . . k : A, B, C SL(d)}

(11.1)

For three qubits there is a simple classification of all possible types of entanglement. Apart
from product states, and states with only bipartite entanglement, the two classes have the following
representative states:
1
|GHZi = (|000i + |111i)
(11.2)
2
and
1
|W i = (|001i + |010i + |100i)
(11.3)
2
The class of SLOCC operations forms a group:
G = {A B C : A, B, C SL(d)}.

(11.4)

A easy to check fact is that SL(d) = {eX : tr(X) = 0}. Therefore

G = {eA eB eC = eAII+IBI+IIC : tr(A) = tr(B) = tr(C) = 0}.

(11.5)

We denote CABC by G.|iABC .

11.1

Quantum Marginal Problem for Three Parties

What are the possible A , B , C compatible with a pure state |iABC CABC ? We say before
that this only depends on the spectra A , B and C , as one can always apply local unitaries and
change the basis.
For example, for the W class the set of compatible spectra is given by the equation A
max +
B
max + C

2.
max
Let us start with a simpler problem, namely given a state |iABC , does there exist a state in
G.|iABC with A = B = C = I/d? This is equivalent to
tr(A A) = tr(B B) = tr(C C) = 0,

(11.6)

for all Hermitian traceless matrices A, B, C, which in turn is equivalent to

tr(A A) + tr(B B) + tr(C C) = 0,

(11.7)

for all Hermitian traceless matrices A, B, C. We can write it as

hABC |A I I + I B I + I I C|ABC i = 0.
37

(11.8)

Thus the norm of the state |ABC i should not change (to 1st order) when we apply an infinitesimal
SLOCC operation.
Let us look at

tA
ke etB etC |ABC ik =
hABC |etA etB etC |ABC i
t
t
= 2hABC |A I I + I B I + I I C|ABC i. (11.9)
So from Eq. (11.8), if |ABC i is the closest point to the origin in G.|ABC i, then A = B =
C = I/d.
What happens when there is no point in the class with A = B = C = I/d. That seems
strange, as it implies by the above that there is no closest point to the origin. But indeed this is
the case for the |W i class, for example. Consider
(0; 01/) (00 01/) (00 01/)|W i = |W i

(11.10)

and when  goes to zero, one approaches the origin. However the limit is not in G.|W i.
In general, we have:
Theorem 11.1. (Kempf-Ness) The following are equivalent:
There exists a closest point to 0 in G.|ABC i.
There exists a quantum state in G.|ABC i with A = B = C = I/d.
G.|ABC i is closed.
The theorem says that there is no point in G.|W i which is maximally mixed.
How about if we look at the closure of G.|W i?
A fact is that for the closure of G.|ABC i, for every |ABC i, contains a unique closed orbit.
Corollary 11.2. There exists a quantum state in the closure of G.|i with maximally mixed reductions if, and only if, 0
/ G.|ABC i.
We saw before that 0 is in the closure of the |W i class. Therefore we cannot approximate (to
arbitrary accuracy) states in the |W i class by a state with maximally mixed reductions.
If we are given a class G.|i and we would like to show it does not contain the origin, we can
find a function which separates the sets. It turns out we can always choose a polynomial for that
P , such that P (0) = 0 and P (|i) 6= 0 for all |i G.|i. We can choose the polynomial P to be
G-invariant (i.e. P (|i) = P (g.|i) for all g) and homogeneous. The converse is also true, so we
find that:
Theorem 11.3. 0
/ G.|ABC i if, and only if, there exists a G-invariant homogeneous polynomial
such that P (0) = 0 and P (|ABC i) 6= 0.
This new characterization doesnt look particularly useful at first sight since we have to check
all homogeneous G-invariant polynomials. However it turns out that we only have to check a
finite number of polynomials since the set of G-invariant polynomials is finitely generated (ref). A
particular case of this result to three qubits state is that any G-invariant polynomial is a sum of
powers of the Cayleys hyperdeterminant.
Next lecture we will see how we can use representation theory to study the G-invariant polynomials.
38

The Mathematics of Entanglement - Summer 2013

30 May, 2013

High dimensional entanglement

Lecturer: Aram Harrow

Lecture 12

Today, I will tell you about bizarre things that can happen with entanglement of high dimensional quantum states. Recall from Fernandos lecture that
hSEP (M ) = max tr M
SEP

where {M, id M } are the yes/no outcomes of a POVM. He also showed that it is NP-hard to
compute this quantity exactly in general. So, here we want to consider approximations to this
quantity that we can compute easier.
For this we introduce approximations to the set of separable states based on the concept of
n-extendibility.
Definition 4. Let AB D(C dA CdB ). We say that AB is n-extendible if there exists a state
AB1 Bn D(CdA Symn (CdB )) s.th. AB = trB2 Bn AB1 Bn .
It turns out that the set of n-extendible states is a good outer approximation to the set of
separable states that gets better and better as n increases. But let us first check that the set of
separable states is contained in it, that is, that every
P separable AB is n-extendible. This can be seen
by writing the separable state AB in the P
form i pi |i ihi | |i ihi |. A symmetric extension is
then easily seen to be given by AB1 Bn = i pi |i ihi | |i ihi |n . The approximation statement
is then contained in the following theorem.
Theorem 12.1. If AB is n-extendible, then there is a separable state with 12 || ||1 nd .
The proof of this theorem is very similar to the proof of the quantum de Finetti theorem which
we did yesterday (in fact, you could adapt the proof as an exercise if you wish).
As a corollary it now follows that we can approximate hSEP (M ) by
hnext (M ) := max tr M .
next

Corollary 12.2. For all 0 M id:

hsep (M ) hnext (M ) hsep (M ) +

d
.
n

The lower bound follows directly from the fact that the set of separable states is contained in
the set of n-extendible states (it even holds for all hermitian M without the restriction 0 M id.
For the upper bound, we use the theorem the observation that
1
max tr M ( ) = || ||1
0M id
2
and obtain
hnext (M ) = max tr M max tr M +
sep

next

39

d
.
n

We now want to see how difficult it is to compute hnext (M ). We rewrite hnext (M ) in the
form
hnext (M ) =
=

max

h|M idn1 |i

(12.1)

max

tr M trn1 |ih|

(12.2)

|iCdA Symn (CdB )

|iCdA Symn (CdB )

= max (sym M idn1 sym ).

(12.3)

Hence, the effort to compute hnext (M ) is polynomial in dn+1 . In order to obtain an  approximation to hsep (M ) we have to choose  = d/n according to the corollary. Hence the effort to
approximate up to accuracy  then the effort scales as dn/ .
Actually this is optimal for general M . In order to see why, we are going to employ a quantum
states known as the antisymmetric states (it is also known to the the universal counter example to
any conjecture in entanglement theory which you may have). The antisymmetric state comes in a
pair with the symmetric state:
The symmetric state is
d,2
id +F
sym
sym =
=
.
d(d + 1)/2
d(d + 1)
d,2
R
sym
It is indeed separable, because d(d+1)/2
= d|ih|2 .
The antisymmetric state is
anti =

id d,2
id F
sym
=
d(d 1)/2
d(d 1)

This antisymmetric state it funny because

1. it is very far from separable; for all separable :

1
2 ||anti

||1

1
2

2. it is very extendible; more precisely, two copies anti anti are d 1-extendible.
Let us first see why 1. holds. For this note that for M = d,2
sym : tr M anti = 0. On the other
hand
1 1
tr M = tr(/2 + F /2) = + tr F .
2 2
In order to bound tr F note that
X
X
X
tr F (X Y ) =
hi, j|F (X Y )|i, ji =
hj, i|(X Y )|i, ji =
Xji Yij = tr XY.
i,j

i,j

ij

Hence
1 1X
+
pi tr F |i ihi | |i ihi |
2 2
i
1 1X
1
= +
pi |hi |i i|2 .
2 2
2

tr M =

40

(12.4)
(12.5)

In order to see 2. holds, note that

2
d(d + 1)

anti =

|iji |jii hij| hji|

.
2
2
1i<jd
X

Consider now the following state (known as a Slater determinant)

|i =

1 X
sgn()|(1)i |(n)i.
d!
Sd

where we introduced the sign of a permutation

transpositions

sgn() = (1)

n
X
= det(
|(i)ihi|).
i

Let us now verify that the Slater determinant extends anti , i.e. that
anti = tr3d |ih|.
This can be done by a quick direct calculation of the partial trace using the formula
X
tr3d |ih| =
id hi3 in ||ih| id |i3 in i.
i3 in

resulting in
X
1
|jk kjihjk kj|.
d(d 1)
j<k

Note that the extension we constructed was actually antisymmetric and not symmetric. But if
we take two copies of the antisymmetric state, the negative signs cancel out:
|i|i Symn (C d C d )
and we thus have the desired d 1 extension of anti anti .

41

The Mathematics of Entanglement - Summer 2013

31 May, 2013

LOCC distinguishability
Lecturer: Fernando G.S.L. Brand
ao

13.1

Lecture 13

Data Hiding

Review of bad news from yesterday.

1. The weak membership problem (determining whether AB Sep or D(AB , Sep)  given
the promise that one of these holds) is NP-hard for  = 1/ poly(dim).
2. k-extendability does not give a good approximation in trace norm until k d, which corresponds to an algorithm that takes time exponential in d.
Lets look more closely at what went wrong with using k-extendable states to approximate Sep.
The state is
I F

=
WAB
D(Cd Cd ).
d(d 1)

AB k1 = 12 .
It is k-extendable for k = d 1, and satisfies minSep 21 kWAB
Here the trace distance describes our ability to distinguish W and using arbitrary twooutcome measurements {M, I M } satisfying only 0 M I. However, since arbitrary measurements can be hard to implement, it is often reasonable to consider the smaller class of measurements
that can be implemented with LOCC.

Locality-restricted measurements
Define the LOCC norm to be
1
kAB AB kLOCC :=
2

max

0M I
{M,IM }LOCC

| tr(M ( ))|.

Define the 1-LOCC norm to be analogous, but with LOCC replaced with 1-LOCC. This stands
for one-way LOCC. This means that one party (by convention, Bob) makes a measurement,
sends the outcome to Alice and she makes a measurement based on this message. The resulting
measurements always have the form
X
M=
Ak B k
k

0 Ak I

for each k

0 Bk
X
Bk = I

for each k

42

Is W still far from Sep in the LOCC norm? Observe that

min

Sep

1
1

+
kWAB
AB kLOCC kWAB
WAB
kLOCC ,
2
2

R
I+F
= d|i |ih|2 is separable.
since W + := d(d+1)
P
P
Observe that if {M, I M } LOCC then we have M = k Ak Bk and I M = k A0k Bk0
with each Ak , Bk , A0k , Bk0 0. Thus
0 M TA I.
(13.1)
We can then further relax
1

+
kWAB
WAB
kLOCC
2
=

max

tr(M (W + W ))

max

tr(M TA ((W + )TA (W )TA ))

0M I
0M TA I
0M I
0M TA I

1
k(W + )TA (W )TA k1 .
2
P
To evaluate this last quantity, observe that if F = di,j=1 |i, jihi, j|, then F TA = d+ where
P
+ := |+ ih+ | and |+ i = 1d di=1 |i, ii. Then
TA

(W )

+ TA

(W )


=

=

I F
d(d 1)

TA

I +F
d(d + 1)

TA

I F TA
I d+
=
d(d 1)
d(d 1)

I + F TA
I + d+
=
d(d + 1)
d(d + 1)

Now we can calculate

+



1
(W )TA (W + )TA = 1 I d I + d+

1
2
2 d(d 1)
d(d + 1) 1



1
+
I


=

2 d(d 1)(d + 1) (d 1)(d + 1) 1

1
.
d
This is an example of data hiding. The states W + , W are perfectly distinguishable with global
measurements, but can only be distinguished with bias 1/d using LOCC measurements.

13.2

Better de Finetti theorems for 1-LOCC measurements

This data hiding example raises the hope that a more useful version of the de Finetti theorem might
hold when we look at 1-LOCC measurements. Indeed, we will see that the following improved de
Finetti theorem does hold:
Theorem 13.1. If AB D(CdA CdB ) is k-extendable then
r
2 ln(2) log(dA )
.
min kAB AB k1-LOCC
Sep
k
43

(13.2)

This was first proved in [Brand

ao, Christandl, Yard; arXiv:1010.1750] but in Arams lecture
you will see a simpler proof, from [Brandao, Harrow; arXiv:1210.6367].
It can be shown [Matthews, Wehner, Winter; arXiv:0810.2327] that
kAB AB k1-LOCC

1
1 p
k k2 :=
tr( )2 .
127
127

Thus, Theorem 13.1 also gives a good approximation in the 2-norm.

Application to weak membership. Lets consider the weak membership problem for Sep, but
now with the distance measure given by 1-LOCC norm:
D(, Sep) := min k k1-LOCC .
Sep

We will solve this problem using semidefinite programming (SDP), which means optimizing
a linear function over matrices subject to semidefinite constraints (i.e. constraints that a given
matrix is positive semidefinite). Algorithms are known that can solve SDPs in time polynomial
in the number of variables. The SDP for checking whether AB is k-extendable is to search for a
AB1 ,...,Bk satisfying
AB1 ,...,Bk 0
j

ABj = AB .

A)
The algorithm is to run the SDP for k = 4 ln(2)log(d
. If Sep then the SDP will be
2
feasible because is also k-extendable. The harder case is to show that the SDP is infeasible when
D(AB , Sep) . But this follows from Theorem 13.1.
k
The run time is polynomial in dA dk+1
B , which is dominated by the dB term. This is

exp(c log(dA ) log(dB )/2 ),

which is slightly more than polynomial-time. It is called quasi-polynomial, meaning that it is
exp(poly(log(input size))).
Idea behind proof of Theorem 13.1. Suppose we had a magic entanglement measure:
E : D(CdA CdB ) 7 R+ with the following properties.
1. normalization: E(AB ) min(log(dA ), log(dB ))
2. monogamy: E(A:B1 B2 ) E(A:B1 ) + E(A:B2 ).
3. faithfulness: E(AB )  implies that D(, Sep) f () where f 0 as  0, e.g.

f () = c .
If we had such a measure, the proof would be very easy.
log(dA ) E(A:B1 ...Bk )

by normalization

E(A:B1 ) + E(A:B2 ...Bk )

k
X

E(A:Bi )

by monogamy
repeating the argument

i=1

= kE()
44

Rearranging, we have E() log(dA )/k, and finally we use faithfulness to argue that D(, Sep)
A)
).
f ( log(d
k
Such a measure does exist! It is called squashed entanglement and was introduced in 2003
by our very own Matthias Christandl and Andreas Winter [quant-ph/0308088]. Normalization
and monogamy are straightforward to prove for it (and were proved in the original paper), but
faithfulness was not proved until 2010 [Brandao, Christandl, Yard; arXiv:1010.1750].

45

The Mathematics of Entanglement - Summer 2013

31 May, 2013

Representation Theory and Spectrum Estimation

Lecturer: Matthias Christandl

14.1

Lecture 14

Representation Theory

Given a group G a representation of G in U (V ), the set of unitaries over the vector space V = C
is a mapping g 7 U (g) U (V ) which is a homomorphism (...).
Definition 5. We say a representation is irreducible if for all such that
W V, U (g)|wig, q W

(14.1)

we have W = {0, V }.
(g), we say they are equivalent if
Definition 6. Given two representations g 7 U (g) and g 7 U
1
(g) for all g.
there exists a invertible A such that AU (g)A = U
We have the following important theorem
Theorem 14.1. For G a finite or Lie group, U (g) = U1 (g) . . . Ul (g) with U1 , . . . , Ul irreducible
representations. Analogously,
M
V =
Vi Cmi ,
(14.2)
=

iG

is the set of equivalent classes of irreps (irreducible representations), with G

with mi N. Here G
acting irreducible in each of the Vi .
This shows that irreducible representations are building blocks of general representations
Example: Let us consider SU (2). In this case Vj is the space of spin j, of dimension dim(Vj ) =
2j + 1. A basis for Vj is {|j, mi}jm=j . Let us consider the su(2) Lie algebra
z |j, mi = m|j, mi,

(14.3)

|j, mi = c |j, mi.

(14.4)

and
Then the action of the group in Vj is obtained by exponentiating the Lie algebra.
We can write Vj = Sym2j (C2 ) (C2 )2j and the action of the group in Vj as g 7 sym g
. . . gsym .

46

14.1.1

Schur-Weyl Duality

An important theorem in representation theory of SU (d) and the symmetric group is the so-called
Schur-Weyl duality.
Consider the following representation of SU (2): g 7 g . . . g. The associated vector space
is V n = (C2 )n . We can decompose this representation into irreps as
M
V (n) =
Vj Cmj .
(14.5)
j

For Sn , the symmetric group of order n, there is a natural representation in V n , given by

|i1 , . . . , in i = | 1 (1), . . . , 1 (n)i, for Sn . This representation clearly commutes with the
one for SU (d). Schur Weyl duality gives a convenient representation into irreps for both groups
simultaneously. We will not need the details of this presentation in this lecture though. Instead
out goal is to compute the mj in (14.5).

14.1.2

Computing mj

Let us consider mnn/2 . We have:

(n)
(1)
V (n+1)
= V
V

M
n
=
Vj Cmj V1/2
j

M
n
=
Vj Cmj V1/2
j

(Vj+1/2 Vj1/2 ) Cnj

Vj 0 C

(mn
+mn
)
j 0 +1/2
j 0 1/2

j0

= ....

(14.6)

We can derive the following recursion relation: mn+1

= mnj+1/2 + mnj1/2 . The solution of this
j
recursion relation can be checked to be

 

n
n
n
mj =

2nh(1/2j/n)
(14.7)
n/2 j
n/2 j 1
with h the binary entropy.
But actually, Vj Cmj (C2 )n = span{(|01i|10i)(n/2j) |j, mi}, with Sn . The vectors
|j, mi C2j are equal to to
|j, mi = c(|0, . . . , 0, 1, . . . , 1i + permutations of 0s and 1s),
with j + m 0s and j m 1s.

47

(14.8)

14.2

Spectrum Estimation

Let us for a moment forget about representation theory and consider the problem of spectrum
estimation. In this problem we are given a source of quantum states which gives n copies of an
unknown density matrix: n . The goal is to perform a measurement which gives a estimate of .
Suppose we are only interested in estimating the eigenvalues of .
For a qubit state , the eigenvalues of we can be written as (1/2 + r, 1/2 r), with r
[0, 1/2]. The problem of estimating r was considered by Keyl and Werner [], who provided an
interesting connection of the problem to representation theory of SU (2). The measurement consists
of projecting the n copies into the spaces Vj Cmj introduced in the last section.
Informally, the claim is that with high probability j/m r. More precisely,
Pr(j) = tr(Pj n ) constenD(1/2+j/n||1/2+r) ,

(14.9)

where D(x||y) is the relative entropy of x and y, given by D(x||y) = x log xx log y +(1x) log(1
x) (1 x) log(1 y).
Let us now sketch the proof. We can compute
(h0, 1| h1, 0|) (|0, 1i |1, 0i) = (1/2 + r)(1/2 r).

(14.10)

Then using Eq. (14.7),

n

tr(Pj

 X
j
n
)
(1/2 r)n/2j (1 r)n/2+j (1/2 + r)j+m (1/2 r)jm
n(1/2 j/n)
m=j



X
(1/2 r)k
n

(1/2 + r)n/2+j (1 r)n/2j

.
(14.11)
n(1/2 j/n)
(1/2 + r)k


k=0

Noting that

14.2.1

(1/2r)k
k=0 (1/2+r)k

is a constant we obtain the claim.

Application Keyl-Werner Relation

Let us finish mentioning one application. Suppose we have |in

ABC and we measure {Pj,A }, {Pj,B },
and {Pj,C }. We just learned we will obtain with high probability outcomes jA , jB , jC such that
jA /n rA , jB /n rB , and jC /n rC . Therefore we must have


tr Pj,A Pj,B Pj,C |ih|n 6= 0.
(14.12)
j
j
j
j
j
j
Therefore {hA
| hB
| hC
|g|in } =
6 0 for all g G and bases |A
i |B
i |C
i for the
subspaces associated to Pj,A , Pj,B , Pj,C .

48

The Mathematics of Entanglement - Summer 2013

31 May, 2013

Proof of the 1-LOCC quantum de Finetti theorem

Lecturer: Aram Harrow

15.1

Lecture 15

Introduction

in this lecture, I will give a proof of the following theorem, first mentioned by Fernando, on the
way introducing useful properties of von Neumann and Shannon entropy.
Theorem 15.1. Let AB be k-extendible and M 0 a 1-LOCC measurement. Then there exists a
separable state such that
r
log dA
0
| tr M ( )| const
.
k
It is possible to swap the quantifiers with help of von Neumanns minimax theorem.
Pm
0
Recall that M 0 can be written in the
Pform M = i=1 Ai Bk for 0 Ai id and 0 Bi id.
Define the measurement M : M () = i tr Bi |iihi|. Note that AB extendible implies that there
exists a state AB1 Bn s.th. AB = ABi for all i. It is the goal to show that
id M (AB ) id M (AB )
for separable :
AB =

|i ihi | |i ihi |

Writing AB1 =

x
x px AB1 ,

we want to show that

x
id M (AB
) x M (x ),
1

where we use the shorthand x = |x ihx |. We now consider the state

AB1 Ak := (id M k )(AB1 Bk )
It is our goal to show that
AB1 =

x
px AB
1

AB1 x M (x )
x
x
AB1 A
B

In order to show this, there are two cases to consider.

Case 1: AB1 A B1 . Then we are done. Caes 1: AB1 is far from A B1 . Then we
condition on B1 and are looking at system B2 , having reduced the uncertainty about that system.
This way we get a little closer to case 1. When continuing to B3 etc, this will prove the theorem.
This was the high level view. We will now make this precise by using a measure of correlation
based on entropy. Since the maximum of an entropy is log d, this will give the bound of theorem
(as opposed to the linear scaling in d that we encountered in the trace norm quantum de Finetti
theorem.
49

15.2

Entropy

Recall that the Shannon entropy of a probability distribution p of a random variable X is given by
X
px log px = H(X)p .
H(p) =
x

P
When we have
joint
distributions
p(xy),
we
can
look
at
the
marginal
distributions
p(x)
=
y p(xy)
P
and p(y) = x p(xy) and their entropies. The conditional entropy of X given Y is defined as
X
p(y)H(X)p(x|y)
H(X|Y )p =
y

where p(x|y) = p(xy)/p(y). Writing the conditional entropy out explicitly we find the formula
H(X|Y )p = H(XY )p H(Y )p .
This gives us a beautiful interpretation of the conditional entropy: it is just the entropy of the joint
distribution minus the entropy of Y .
We can measure the correlation between two random variables X and Y by looking at the
difference between the entropy H(X) and H(X|Y )
I(X; Y ) = H(X) H(X|Y ) = H(X) + H(Y ) H(XY )
Note that this quantity is symmetric with respect to interchange of X and Y . This is known as
the mutual information and quantifies the by how much our uncertainty about X changes when we
are given Y (and vice versa, of course). It is also the amount of bits that you save by compressing
XY together as opposed to compressing X and Y separately.
The mutual information has a few nice properties:
I(X : Y ) 0
I(X : Y ) log |X| and I(X : Y ) log |Y |, where |X| denotes the number of symbols in X.

(Pinskers inequality) I(X : Y )

1
2 ln 2

2
|p(xy)

p(x)p(y)|
x,y

P

Let us now look at the quantum version of all this. I will use the notation that S(A) = S(A )
and we have a joint state AB , then S(A) = S(trB AB ). Note that it is not immediately clear
how to define the conditional entropy in the quantum case, since we cannot condition on a value of
system B. Luckily we had a second way of writing the conditional entropy and we are just going
to define the conditional quantum entropy as
S(A|B) := S(AB) S(B)
and the mutual information as
I(A : B) = S(A) + S(B) S(AB).
It has the following properties
50

 S(A : B) 0
I(A : B) 2 log dA and I(A : B) 2 log dB (note the factor of two!)
(Pinskers inequality) I(A : B)

1
2 ln 2 ||AB

A B ||21

This last property looks like it could be useful in proving the theorem; but note that it all
cannot be that easy, because we know that we should use the 1-LOCC norm and not the trace
norm...In order to proceed, we need the conditional mutual information:
I(A : B|C) = S(A|C) + S(B|C) S(AB|C) = S(AC) + S(BC) S(ABC) S(C)
This formula is a little difficult to grasp and it is difficult to get an intuition for this quantity. The
conditional mutual information has a nice property, though, it satisfies the chain rule:
I(A : BC) = I(A : C) I(A : B|C)
which you can easily check. It is called the chain rule, in part, because we can iterate it: Lets
assume we have a k-extendible state and we measure all of the B systems. How much does A know
about the Bs? The chain rule gives us
I(A : B1 Bk ) = I(A : B1 ) + I(A : B2 Bk |B1 )

(15.1)

= I(A : B1 ) + I(A : B2 |B1 ) + I(A : B3 Bk |B1 B2 )

(15.2)

= I(A : B1 ) + I(A : B2 |B1 ) + + I(A : Bk |B1 B2 Bk1 )

(15.3)

There are k terms and the sum is small than log d, hence there is a j such that I(A : Bk |B1 B2 Bk1
log d
k . This then immediately gives a proof of our theorem (see arXiv:1210.6367 for more details).

51