Entanglement:

definition and first applications

•
Phys 402, Spring Term 2023

1 Entanglement
1.1 Tensor product Hilbert spaces
Tensor product Hilbert spaces describe physical systems with a number of independent
degrees of freedom, and are thus very important. As an example, consider a chain of n
spin-1/2 particles. We already know that the state of each individual particle lives in a
Hilbert space H1 = C2 . But the combined system is also quantum mechanical, hence must
live in a Hilbert space Hn . But how is Hn constructed?
To approach this question, we choose for each of the n spins 1/2 a basis, say the local
eigenbasis of Ŝz . Every spin i may point up or down which we write as |0ii and |1ii
respectively. The special quantum states where each spin is either pointing up or down we
may then described by an n-component binary vector s such that the state of the i-th spin
is |si ii . The state of all n spins in this configuration is |si which we write as

|si = |s1 i1 ⊗ |s2 i2 ⊗ ... ⊗ |sn in . (1)

Therein, the symbol “⊗” represents the tensor product between states. It signifies that the
states joined by it refer to different degrees of freedom. Between the tensor product and the
addition of Hilbert space vectors holds a distributive law,

|αi1 ⊗ (|βi2 + |γi2 ) = |αi1 ⊗ |βi2 + |αi1 ⊗ |γi2 .

Now, the n-particle Hilbert space is the linear span of the basis states Eq. (1),

Hn = span |si, s ∈ (Z2 )n .

Even if we used a special basis to define it, the n-particle Hilbert space Hn is a basis-
independent construct (show it!), and we write
(1) (2) (n)
Hn = H1 ⊗ H1 ⊗ ... ⊗ H1 .

1
Observation. The dimension of Hn in the example just discussed is dim(Hn ) = 2n . As a
result, the size of the matrices representing operators acting on Hn grows exponentially fast
in the number n of degrees of freedom. This illustrates why quantum mechanics problems
are so hard to put on a computer.
Example. To make all this more concrete, let’s consider the special case of two tensor
(1,2) (1) (2)
factors, i.e. H2 = H1 ⊗ H1 , where each tensor factor is still a two-dimensional Hilbert
space. There are 2 × 2 = 4 basis states |si, namely |(0, 0)i = |0i1 ⊗ |0i2 , |(0, 1)i = |0i1 ⊗ |1i2 ,
(1,2) (1) (2)
|(1, 0)i = |1i1 ⊗ |0i2 , and |(1, 1)i = |1i1 ⊗ |1i2 . A general state |Ψi ∈ H2 = H1 ⊗ H1
has the form

|Ψi1,2 = c00 |0i1 ⊗ |0i2 + c01 |0i1 ⊗ |1i2 + c10 |1i1 ⊗ |0i2 + c11 |1i1 ⊗ |1i2 .

In the above, the local Hilbert space dimension of 2 was chosen only for simplicity. Tensor
products can be formed between Hilbert spaces of any dimension, and the dimensions do
not need to be the same in different factors. If dim(HA ) = a and dim(HB ) = b then
dim(HA ⊗ HB ) = ab.

1.2 Definition of “Entanglement”

The basis states |si in Eq. (1) by their very construction have the property of being tensor
product states, |si = |s1 i1 ⊗ |s2 i2 ⊗ ... ⊗ |sn in . General states |ψi ∈ Hn = (C2 )n will not
have this property. This observation provides reason to make an important definition.

Definition 1. A pure quantum state |Ψi shared between parties 1,2, .. , n is called entangled
if
|Ψi =
6 |ψ1 i1 ⊗ |ψ2 i2 ⊗ .. ⊗ |ψn in , (2)
for any set of local states {|ψ1 i, |ψ2 i, .., |ψn i}.

We will soon learn how a “pure” state is to be distinguished from a “mixed” state. A
more general definition of “entanglement” will follow. However, for now it is more useful to
specialize the above definition than to generalize, namely to just two parties.

Definition 2. A pure bipartitie quantum state |ΨiAB shared between parties A and B is
called entangled if
|ΨiAB 6= |ψiA ⊗ |φiB , (3)
for any pair of local states |ψi, |φi.

A prototypical entangled state between two spin-1/2 systems A and B is

| ↑iA ⊗ | ↓iB − | ↓iA ⊗ | ↑iB

|Φi = √ . (4)
2
This state is known as a singlet state, because it carries a total spin of 0 (if you ask a
condensed matter or particle physicist), or alternatively as a “Bell state” (if you ask someone
in the domain of quantum optics or quantum information).

2
1.3 Physical significance of entanglement
The physical manifestation of entanglement is that it leads to correlations that are stronger
than what is classically possible. For example, consider the Bell state of Eq. (4), and
measurement of particle 1 in any of the bases

B(α, β) = {|rα,β i := α| ↑i + β| ↓i, |rα,β i := β ∗ | ↑i − α∗ | ↓i, |α|2 + |β|2 = 1}.

By straightforward calculation one can now establish the following two facts:
1. The state |Φi of Eq. (4) has no local properties. Whatever the parameters α, β ∈ C,
the measurement of particle 1 (or, equivalently, of particle 2) in the basis B(α, β) leads
to a 50/50 distribution of outcome.
2. Measurement of particle 1 B(α, β) leads to the post-measurement state

outcome “+”: |Φi =⇒ |rα,β i1 ⊗ |rα,β i2 ,

outcome “-”: |Φi =⇒ |rα,β i1 ⊗ |rα,β i2 .
That is, after measurement, the spin states of particle 1 and 2 are perfectly anti-
correlated, irrespective of the measurement outcome and the measurement basis.
Let’s compare the second finding with correlations that are possible classically. Classical
spins still can point in any direction in space, but can be correlated only in one direction.
For example, with a 50/50 probability, spin 1 can point up and spin 2 down, or the other
way around. Those classical spins are perfectly anticorrelated in the z-direction. But unlike
in the quantum case, they are uncorrelated in the x- and in the y-direction.

1.4 Applications
1.4.1 A non-application: superluminal communication
The observation of item 2 in Section 1.3 was dubbed by Albert Einstein “spooky action at
a distance”. The measurement of particle 1 has an effect on both particles, namely they
are both projected into a fully specified local state. This holds irrespective of whether both
particles are located in the same laboratory, or one particle is on earth and the other on the
moon.
And so it does indeed seem as if a non-local influence was at work here; and this puzzled
Einstein. A question one may ask at this point is whether this apparent non-local influence
is so weird that it contradicts the laws of special relativity. Concretely, one may ask: Can
spooky action at a distance be used for super-luminal (i.e., instantaneous) communication?
Here is a protocol once proposed1 . To transmit one bit of information from a sender A
to a receiver B, apply the following protocol:
1. In advance of the actual communication, the parties share Bell states |Φi between
themselves, such that for each Bell state, party A holds one qubit of it, and party B
holds the other. Each transmission of a bit from A to B uses up one Bell state |Φi.
2. The communication begins. If the bit b to be transmitted has the value b = 0, then
party A measures their qubit of |Φi in the eigenbasis of Sz , i.e., in Bz = {| ↑i, | ↓i}.
If b = 1, then party A measures their qubit of |Φi in the eigenbasis of Sx , i.e., in
Bx = {| →i, | ←i}.
1
.. and which was indeed granted a patent

3
 3. Party B copies their state a large number of times, i.e.,
| ↑i −→ | ↑i ⊗ | ↑i ⊗ | ↑i ⊗ ... ⊗ | ↑i,
| ↓i −→ | ↓i ⊗ | ↓i ⊗ | ↓i ⊗ ... ⊗ | ↓i,
(5)
| →i −→ | →i ⊗ | →i ⊗ | →i ⊗ ... ⊗ | →i,
| ←i −→ | ←i ⊗ | ←i ⊗ | ←i ⊗ ... ⊗ | ←i.

4. From the state created in the copying step of Eq. (5), party B now identifies the state
received in the transmission, through the measurement of the multiple copies.
• If the state received was | ↑i or | ↓i, B outputs b = 0.
• If the state received was | →i or | ←i, B outputs b = 1.
Let’s now investigate whether this is a viable protocol. Clearly, steps 1 and 2 are viable, so
we only need to look at steps 3 and 4. The measurement in step 4 of the above protocol is
viable. The four states on the r.h.s. of Eq. (5) are near-orthogonal if the number of copies
is large, hence can be near-perfectly distinguished.
If you want a more explicit protocol, here is one: Measure all even-numbered copies in
the z-basis and all odd-numbered copies in the x-basis. Either the measurement outcomes on
the even- or on the odd-numbered qubits are perfectly correlated, and the respective others
are completely uncorrelated. If the outcomes on the even-numbered qubits are perfectly
correlated then output b = 0, and if the outcomes of the odd-numbered qubits are perfectly
correlated then output b = 1. The probability for both measurement records appearing
perfectly correlated and thus leading to an inconclusive result drops exponentially with the
number of copies as 21−n .
The problem is with step 3 of the protocol. The copying operation written down in
Eq. (5) is not a viable quantum operation. This is the content of the no-cloning theorem.
Theorem 1 (No cloning). Be |ψi ∈ Hd an unknown quantum state, and |0i ∈ Hd a fixed
known quantum state. Then, the copying (cloning) operation C defined by

C : |ψi ⊗ |0i −→ |ψi ⊗ |ψi, ∀|ψi ∈ Hd ,

cannot be realized in quantum mechanics.

So next time you stand in front of a copying machine waiting for page 50 of your docu-
ment to emerge, marvel at the fantastic opportunities afforded by our classical world!
What matters most from the perspective we have taken is that the coexistence of quan-
tum mechanics and special relativity is saved fo now. Note, though, that the no-cloning
theorem only rules out one particular suggested protocol for superluminal communication,
not all. However, we will rule out all such protocols in Section 2, after establishing a bit
more notational infrastructure.
The proof of the no-cloning theorem is very simple, but before we can approach it we
must return to our descriptions of quantum evolutions. They have one property that we
have so far not made explicit—linearity. Namely we observe
∂
1. If, for a given Hamiltonian H, the Schrödinger equation i~ ∂t |Ψ(t)i = H|Ψ(t)i has a
solution |Ψ1 (t)i and a solution |Ψ2 (t)i, then for all a, b ∈ C,

|Ψ0a,b (t)i = a|Ψ1 (t)i + b|Ψ2 (t)i

is also a solution of this Schrödinger equation.

4
 2. Consider a quantum state |Ψ0 i = a|Ψ1 i + b|Ψ2 i subjected to a measurement described
by the projectors {Πi , i = 1..n}. Then, for the post-measurement states |m0 i, m1 i
and m2 i resulting from |Ψ0 i, |Ψ1 i, |Ψ2 i, respectively, it holds that |m1 i ∼ Πi |Ψ1 i,
|m2 i ∼ Πi |Ψ2 i, and

|m0 i ∝ Πi |Ψ0 i = a (Πi |Ψ1 i) + b (Πi |Ψ2 i) , ∀i.

That is, measurement is linear up to normalization.

Thus, up to normalization, all quantum evolution is linear.

Proof of Theorem 1. Wlog. consider a Hilbert space of local dimension d = 2, i.e.,
|ψi ∈ C2 . The cloning maps C are sequences of projections (dependent on measurement
outcomes) and unitaries. When leaving out the final normalization, those operations are all
linear.
By assumption, we have to have

C(|0i ⊗ |0i) ∝ |0i ⊗ |0i,

(6)
C(|1i ⊗ |0i) ∝ |1i ⊗ |1i,

and
C(|+i ⊗ |0i) ∝ |+i ⊗ |+i, (7)
√
where |+i := (|0i + |1i)/ 2. On the other hand, by linearity, we also have
 
|0i+|1i
C(|+i ⊗ |0i) = C √
2
⊗ |0i
= √2 C (|0i ⊗ |0i) + √12 C (|1i ⊗ |0i)
1

= √12 (a|0i ⊗ |0i + b|1i ⊗ |1i) .

Therein, in the second line we have used the linearity of C, and in the third line Eq. (6).
Whatever the values a, b ∈ C in the last line of the above equation, it is in contradiction
with Eq. (7). 

1.4.2 Dense coding

It’s high time to introduce the qubit. The quantum bit, or qubit for short, is the quantum
counterpart of the bit. I has two basis states, |0i and |1i, which correspond to the states that
a classical bit can assume. But different from the classical bit, the qubit can also represent
all superpositions
α|0i + β|1i.
That is, the possible states of a qubit are all states in the 2-dimensional Hilbert space C2 .
While, as was hinted at in the last section (and we will prove this in full generality
in Section 2), quantum mechanics does not break special relativity, it does break through
another barrier. Namely, in order to transmit one bit worth of information from A to B, at
least one bit worth of carrier has to be sent. This is intuitively clear, but for the skeptics,
it also arises as a limit of Shannon’s source coding theorem.
Now it turns out that one qubit can be used to send two classical bits [2]. The proof
is constructive–we present a protocol that achieves this rate. The following protocol [2]
transmits two classical bits by sending one qubit:
√
1. A Bell state |Φ00 i = (|00i + |11i)/ 2 is prepared between sender A and receiver B.

5
 2. A encodes a 2-bit message by applying one of the four operations on his end of the
state: I, σx , σy , σz .
3. A sends their qubit to B.
4. B measures in the Bell basis B and thus retrieves the 2-bit message.
Historical note on dense coding. The above protocol of dense coding are the exact
opposite of technically hard. Yet it is published in Physical Review Letters – how come?
Their value is not in mastering a tedious and long-standing problem. Rather they moved
into new territory, saw some unexpected phenomenology, and opened up a new field.
David Mermin, whom you may know from a condensed matter physics class, was a
referee for the dense coding paper. He made his reviews publicly available [1] in 2003, and
I have reprinted it here from [1].
Bennett and Wiesner, ‘‘Communication via one-and two-particle...’’ LT4749
Your question was: Does this qualify as ‘‘strikingly different’’ enough to
publish? I have never read anything like it, and I have read a lot on EPR,
though far from everything ever written. So as far as I know it is different.
But strikingly? The argument is very simple, so shouldn’t the point be
obvious? After reading the paper I put it aside and spent the next week
working hard on something totally unrelated. Every now and then I would
introspect to see if some way of looking at the argument had germinated that
reduced it to a triviality. None had. Last night I woke up at 3am, fascinated
and obsessed with it. Couldn’t get back to sleep. That’s my definition of
‘‘striking’’.

1.4.3 Quantum teleportation

This section is optional reading

The teleportation protocol2 is, in a sense, the reverse of dense coding. Consuming one Bell
pair worth of entanglement (a so-called ebit), an unknown quantum state can be transferred
from A to B by sending two classical bits.
To prepare for the statement of the protocol, we observe that the Bell state Eq. (4) has
three cousins who are just as entangled, and with which it forms an orthonormal basis B,

|Φ00 i = |0iA ⊗|0iB√+|1i
2
A ⊗|1iB
, 


|0iA ⊗|0iB −|1iA ⊗|1iB 
|Φ01 i = √
2
, 
B (8)
|Φ10 i = |0iA ⊗|1iB√+|1i
2
A ⊗|0iB
, 


|Φ i = |0iA ⊗|1iB√−|1iA ⊗|0iB . 

11 2

We note the property σz |φ00 i = |Φ01 i, σx |φ00 i = |Φ10 i, σy |φ00 i ∼ |Φ11 i, or, in short
i  j
|Φij i = σx(2) σz(2) |Φ00 i.

(9)
Note that the Bell state |Φ00 i is (up to a global, hence physically irrelevant, phase) the only
state satisfying the two eigenvalue equations
(1) (2)
σx ⊗ σx |Φ00 i = |Φ00 i,
(1) (2) (10)
σz ⊗ σz |Φ00 i = |Φ00 i.
2
In fact, the teleportation paper is now a PRL milestone, and a “free to read” article.

6
 ψ Bell measurement

outcome (i,j)

Φ00 site A
site B
σij ψ

Figure 1: The teleportation circuit. |Φ00 i is a Bell state as defined in Eq. (8). In the Bell
measurement, the output (i, j) means that the post-measurement state |Φij i was obtained.
For details of the Pauli correction σij see text.

Combining Eqs. (9) with (10), we find that we can generalize the latter. Namely, for all
four Bell states it holds (show it!) that
(1) (2)
σx ⊗ σx |Φij i = (−1)j |Φij i,
(1) (2) (11)
σz ⊗ σz |Φij i = (−1)i |Φij i.

This is the first time we encounter the description of a quantum state by a set of eigenvalue
equations it satisfies. In many subsequent examples we will find that that this is a very
efficient way to describe certain quantum states. Namely, when applicable, n eigenvalue
equations suffice to fully specify an n-qubit quantum states. Compare this to the 2n am-
plitudes one might otherwise write down, and appreciate the efficiency of the method. In
fact, there is a whole formalism around this, called the stabilizer formalism, and we will
introduce its basics in the passing.

The teleportation protocol. We now turn to the question of how many classical bits
need to be sent to transmit the state of one qubit. The answer again turns out to be 2, if
we are permitted the use of entanglement. The protocol [3] goes like this:

1. A sender A and a receiver B prepare a joint Bell state |Φ00 i.

2. Party A now puts in place the state |ψi to transmit (A holds two qubits at this point).

3. A performs a measurement in the Bell basis B on their two-qubit system, and thereby
obtains the two-bit outcome o = (i, j).

4. A transmits the two-bit measurement outcome o to B.

5. Depending on the value of o, B applies one of the four correction operations I, σx , σy ,

σz to its qubit, and thereby retrieves |ψi.

An important point about the above protocol is that the teleported quantum state |ψi can
be unknown to party A at the beginning of the protocol. The protocol works regardless.
Proof of the correctness of the protocol. We teleport a 1-qubit state |ψi = α |0i + β |1i.
We label the qubits on site A as 1,2, and on site B as 3. The three qubit state after the

7
Bell measurement is, up to normalization
 h ij h ii 
(2) (2)
(|Φij i12 hΦij | ⊗ I3 ) |ψi1 ⊗ |Φ00 i23 = |Φij i12 hΦ00 | σz σx ⊗ I3 |ψi1 ⊗ |Φ00 i23
h ii h ij
(3) (3)
= (|Φij i12 hΦ00 | ⊗ I3 ) |ψi1 ⊗ σx σz |Φ00 i23
h ii h ij
(3) (3)
= σx σz (|Φij i12 hΦ00 | ⊗ I3 ) ×
 
α α β β
√ |000i123 + 2 |011i123 + 2 |100i123 + 2 |111i123
√ √ √
h 2 ii h ij
(3) (3)
= 12 σx σz |Φij i12 ⊗ (α|0i3 + β|1i3 )
h ii h ij
(3) (3)
= 21 σx σz |Φij i12 ⊗ |ψi3 .

Therein, in the first line we have used Eq. (9), and in the third line Eq. (10). Thus, the
unknown input state |Ψi is indeed teleported to qubit 3, and the correction operation σij is
j  i
σij = σz(3) σx(3) .


2 Density operators
2.1 Motivation
Motivation 1. Consider a stream of spin-1/2 particles where each spin is pointing up
(wrt the z-axis) with a probability of 1/2, and pointing down with a probability of 1/2.
How do we describe this physical situation?
A suggestion one might come up with is to choose the state with same amplitudes for
then spin-up and spin-down components,
| ↑i + | ↓i
√ . (12)
2
Indeed, the probabilities for finding the spin pointing up or down are indeed both equal to
1/2, as required.
However, this is not enough. The suggested state must give the right expectation val-
ues and probabilities for every possible observable on the given physical system. So let us
consider the measurement of Ŝx . For a 50/50 mixture of | ↑i and | ↓i—or any classical
mixture—the rules of classical probability theory apply, in particular Bayes’ rules for con-
ditional probabilities. The probability for finding the “up” result in a measurement of Ŝx
equals the probability of having | ↑iz times the conditional probability of finding the “X-up”
result in the measurement of | ↑iz plus the probability of having | ↓iz times the conditional
probability of finding the “X-up” result in the measurement of | ↓iz . Thus, the probability
of finding “X-up” for the given mixture should be 1/2 ∗ 1/2 + 1/2 ∗ 1/2 = 1/2. However, for
the state in (12) we find that the same probability should be unity. Contradiction.
We might seek to avoid this contradiction by changing the relative phase in (12) from
“+” to some eiφ , but it will not help. While the discrepancy may cease for Ŝx , there will
always be an observable for which it persists (can you show that?).

Motivation 2. Consider a Bell state, with one spin 1/2 particle on earth, and one on the
moon. Only the particle on earth is available for measurement. Is there a description for
the state of the particle on earth alone, from which the statistics of all viable measurements
can be derived?

8
2.2 Math interlude — trace and partial trace
Consider a Hilbert space Hd of dimension d, and an orthonormal basis B = {|ii, i = 1, .., d}.
Then for any linear operator A on Hd we can define the trace Tr(A) via
d
X
Tr(A) = hi|A|ii. (13)
i=1

The trace is independent of the particular ONB chosen. Be B 0 = {|i0 i, i = 1, .., d} another
ONB, related to the former via the relation |i0 i = U |ii, with U a unitary matrix (recall
that we have shown in class that all ONBs for a given Hilbert space are related by unitary
transformation). Then,
Pd
Tr0 (A) := hi0 |A|i0 i
Pi=1
d
= hi|U † AU |ii
Pi=1
d
= hi|U † |jihj|A|kihk|U |ii
Pi,j,k=1
d
= [U † ]ij [U ]ki hj|A|ki
Pi,j,k=1
d
= δ hj|A|ki
Pj,k=1 jk
= j hj|A|ji
= Tr(A).

Another property of the trace is “cyclicity”. It means that

Tr(ABC) = Tr(BCA) = Tr(CAB). (14)

Partial trace. The partial trace generalizes the above defined notion of trace. It can be
applied to tensor product Hilbert spaces HAB = HA ⊗ HB . Be BA = {|ii, i = 1, .., dA } a
basis for the subsystem A, and BAB a product basis for HAB , i.e., BAB = {|iiA ⊗ |jiB , 1 ≤
i ≤ dA , 1 ≤ j ≤ dB }. The partial trace TrA (X) of an operator X acting on HAB is defined
as
dA
X
TrA (X) := A hi|X|iiA . (15)
i=1

This definition requires explanation, as it is not a priori clear how an operator X defined
on the joint system AB acts on a basis ket |ii defined on A. In this regard, we observe that
dA X
X dB
X= |kiA hk| ⊗ |liB hl|X|riA hr| ⊗ |siB hs|,
k,r=1 l,s=1

and
A hi| (|kiA ⊗ |liB ) = δik |liB .
Therefore, w.r.t. those bases
X
TrA (X) = |liB (A hi| ⊗ B hl|X|iiA ⊗ |siB ) B hs|.
i,l,s

The partial traces satisfy the property that

Tr(X) = TrB (TrA (X)) = TrA (TrB (X)) . (16)

This property will be very useful when discussing the scenario of Motivation 2 above.

9
2.3 Definition of the density operator
We now slightly generalize the setting of Motivation 1. Consider the ensemble E of quantum
states, 
E = pi , |φi i .
This means that states |φi i are drawn randomly from a set, with probabilities pi . For this
ensemble and for any operator A, the expectation hAiE is
X
hAiE = pi hφi |A|φi i.
i

We may now manipulate this expression, as follows

P P
hAiE = p hφ |jihj|A|φi i
Pi i Pj i
= p hj|A|φi ihφi |ji
Pi i j P
= j hj|A ( i pi |φi ihφi |) |ji
= Tr Aρ,

where P
ρ := i pi |φi ihφi | . (17)
 P
Definition 3. For an ensemble E = pi , |φi i , ρ := i pi |φi ihφi | is the corresponding
density operator.

If a density operator has an eigenvalue of 1 then it is called pure. Otherwise, a density

operator is called mixed. All pure density operators are of the form ρ = |ψihψ| for some
state |ψi ∈ H (can you show this?). Thus, they have an ensemble representation with just
one state in it. That’s the reason for calling such states pure.
Let’s see what we have achieved. The just-derived expression

hAiE = Tr Aρ (18)

for the expectation value of A is the counterpart and generalization of our earlier formula
hAiψ = hψ|A|ψi. Our new expression necessitated the definition of a novel object describing
the state of quantum systems, namely the density operator. With the new construct in hand,
we can now efficiently describe classical mixtures of quantum states. This was previously
cumbersome. The density operator ρ is thus a useful generation of the notion of the quantum
state |ψi. As Eq. (18) makes clear, all the properties of a given quantum system are encoded
in its density operator ρ.
We may now specialize to the exact situation of the Motivation section. The 50/50
mixture encountered there was
1 1
ρ50/50 = | ↑ih↑ | + | ↓ih↓ |.
2 2
Remark. While the density operator ρ for a given ensemble E is unique, the converse is
not true. A given density operator will in general have multiple representations as an
ensemble. For example, the above density operator ρ50/50 is also described by an ensemble
E 0 = {(1/2, | ←i), (1/2, | →i)} of spins pointing up and down in the x-direction,
1 1
ρ50/50 = | ←ih← | + | →ih→ |.
2 2

10
Properties of density operators. Irrespective of any further specification of the system
under consideration, all density operators ρ have the following three important properties:

1. Hermiticity: ρ† = ρ.

2. Normalization: Tr ρ = 1.

3. Non-negativity: hψ|ρ|ψi ≥ 0, ∀ |ψi ∈ H.

Problem 1. Demonstrate the above three properties, starting from Definition 3 (where
all states in the ensemble are normalized to unity).

Entanglement for mixed states. We can now generalize the earlier definition of entan-
glement to mixed states.

Definition 4. A quantum state (density operator) ρ shared between parties 1,2, .. , n is

called separable if it can be written in the form
X
ρ= pi ρ1 (i) ⊗ ρ2 (i) ⊗ .. ⊗ ρn (i), (19)
i

Pall ρk (i) are valid density operators local to the respective parties k, pi ≥ 0 for all i,
where
and i pi = 1. Otherwise ρ is called entangled.

2.4 Quantum evolution of density matrices

There is little new here, actually. We just need to reformulate and adapt what we already
know.

2.4.1 Unitary evolution according to the Schrödinger equation

We may consider the time dependent Schrödinger equation as a PDE for evolution operators
U (t, t0 ),
∂
i~ U (t, t0 ) = HU (t, t0 ),
∂t
with the boundary condition U (t0 , t0 ) = I. The time evolution operator describes the
evolution of quantum states |Ψ(t)i according to the above Schrödinger equation via |Ψ(t)i =
U (t, t0 )|Ψ(t0 )i. Now, as we see in Eq. (17)/ Definition 3, the density operator ρ is made of
outer products of kets and bras, and therefore it evolves according to

ρ(t) = U (t, t0 )ρ(t0 )U (t, t0 )† . (20)

If you prefer a PDE, the same can be stated as

∂
i~ ρ(t) = [H, ρ(t)].
∂t

11
2.4.2 Measurement according to the Born rule and Dirac projection postulate
Based on the laws of measurement we have already established for state kets, the corre-
sponding formulation for density operators ρ is as follows. Consider a quantum measurement
described by the set {Πi , i = 1, .., k} of orthogonal projectors.
Then, the probability pi for obtaining the outcome i given the density operator ρ is

pi = Tr(Πi ρ). (21)

This is the Born rule formulated at the level of density operators.

The state after measurement, given the outcome i was obtained, is
Πi ρΠi
ρfinal = . (22)
Tr(Πi ρ)
This is the Dirac projection postulate, formulated at the level of density operators.

2.5 Reduced density operators

We now return to Motivation 2. Consider a Bell state, with one spin 1/2 particle sitting in
your lab on earth, and the other spin 1/2 particle who knows where. Maybe moon, maybe
Mars, not certain. In this situation, the only measurements that can be performed on the
Bell state |B11 iAB (A is your lab) are of observables OA ⊗ IB . That is, observables that act
non-trivially only on the local Hilbert space associated with party A. In such a situation,
is there a “reduced” description of the Bell state |B11 i that describes what the Bell state
“looks like” from the perspective of A alone?
It turns out that such a local description does indeed exist, and it is in terms of density
operators. Namely, consider the expectation value of any observable of form OA ⊗ IB , on a
general quantum state |ΦiAB (of Bell type or not)

AB hΦ|OA ⊗ IB |ΦiAB = Tr (OA ⊗ IB |ΦiAB hΦ|)

= TrA TrB (OA ⊗ IB |ΦiAB hΦ|)
= TrA (OA TrB (|ΦiAB hΦ|))
= TrA (OA ρA ) ,

with
ρA := TrB (|ΦiAB hΦ|) . (23)
The state ρA is called a “reduced” state, because it arose from the reduction of a bipartite
state |ΦiAB for which particle B is inaccessible for measurement.
The above is, in fact, a rather generic situation in physics, in particular condensed
matter physics; no need to invoke the moon. It often happens that an object of interest–
the “system” S—interacts with the degrees of freedom around it, collectively denoted the
“environment” E. We thus typically have a joint Hilbert space HSE = HS ⊗ HE , wherein
the environment we have no control over, and only the system S is observable. In such a
setting it is adequate to describe the joint system–environment state that in principle exists
by a reduced density operator on S.

2.6 Impossibility of superluminal communication

We now have the tools in place to prove that quantum mechanics does not provide for
superluminal communication. Recall that the no-cloning theorem we discussed earlier only

12
rules out one specific protocol that people have discussed. However, we need to find an
argument that applies to all conceivable protocols.
Consider any state |Φi ∈ HA ⊗ HB shared between parties A and B. Can the sender
A send signals to the receiver B by manipulating only its own local subsystem? In other
words, the question is whether a local operation at A can change the reduced density matrix

ρB := TrA (|ΦiAB hΦ|)

at location B. If it can, then it is possible to instantaneously transmit information; and if

it cannot then such transformation is impossible.
A priori, there are two types of evolutions we need to take into account, (a) unitary
evolution according to the Schrödinger equation and (b) measurement according to the
Dirac projection postulate and the Born rule.
(a) Unitaries. The local unitaries that A can perform are UA ⊗ IB . The reduced density
operator at B after such an operation, ρ0B , is
 
ρ0B = TrA UA ⊗ IB |ΦiAB hΦ| UA† ⊗ IB |
 
†
= TrA IA ⊗ IB |ΦiAB hΦ| UA UA ⊗ IB
= TrA (IA ⊗ IB |ΦiAB hΦ| IA ⊗ IB )
= TrA (|ΦiAB hΦ|)
= ρB .
Thus, the reduced state on B does not change when any unitary is applied at A.
(b) Measurements. Let’s recap how the measurement {Πi , i = ..} looks from the per-
spective of A,
(Πi )A |ΦiAB hΦ|(Πi )A
i : |ΦiAB hΦ| 7→ ,
Tr((Πi )A |ΦiAB hΦ|)
and the probability for obtaining the outcome i is p(i) = Tr((Πi )A |ΦiAB hΦ|).
The perspective of party B is different. Namely, B doesn’t learn the outcome i of the
measurement (at least not at an instant; the outcome could surely be transmitted, which
however would defy the purpose of the conjectured protocol). In that situation, B retains
the density operator resulting from the proper average over all measurement outcomes,
X (Πi )A |ΦiAB hΦ|(Πi )A
|ΦiAB hΦ| 7→ Tr((Πi )A |ΦiAB hΦ|)
iX
Tr((Πi )A |ΦiAB hΦ|)
= (Πi )A |ΦiAB hΦ|(Πi )A .
i

Therefore, the reduced density operator ρ00B on B after a measurement on A is

!
X
ρ00B = TrA (Πi )A |ΦiAB hΦ|(Πi )A
i !
X
= TrA |ΦiAB hΦ| (Πi )A (Πi )A
i !
X
= TrA |ΦiAB hΦ| (Πi )A
i
= TrA (|ΦiAB hΦ|)
= ρB .

13
Thus ρB doesn’t change under measurement on A either.
Since all quantum operations are composed of unitaries and measurements, nothing the
sender A can do on their quantum system influences the reduced state at the receiver’s end.
Remark: The argument above is actually from a research paper [4]!

References
[1] N.D. Mermin, Copenhagen Computation: How I Stop Worrying and Love Bohr,
arXiv:0305088 (quant-ph); IBM Journal of Research and Development, Vol. 48, No. 1,
53-62 (2004).

[2] C. Bennett and S.J. Wiesner. Communication via one- and two-particle operators on
Einstein-Podolsky-Rosen states. Phys. Rev. Lett. 69, 2881 (1992).

[3] C.H. Bennett, G. Brassard, C. Crepeau, R. Jozsa, A. Peres and W.K. Wootters, Tele-
porting an Unknown Quantum State via Dual Classical and Einstein-Podolsky-Rosen
Channels, Phys. Rev. Lett. 70, 1895 (1993).

[4] D. Bruss, G.M. D’Ariano, C. Macchiavello, and M.F. Sacchi, Approximate quantum
cloning and the impossibility of superluminal information transfer, Phys. Rev. A 62
062302 (2000).

14