Quantum information theory (MAT4430) Spring 2021

Lecture 3: The theory of open quantum systems

Lecturer: Alexander Müller-Hermes

When encountering a specific quantum system, there are two problems to be solved in
order to harness it for quantum information processing:
1. We need to find a description of the system in the quantum mechanical formalism.

2. We need to engineer information theoretic protocols for the system using the quantum
mechanical formalism.
In this course, we will exclusively focus on the second problem. We will not describe any
particular quantum system, but instead study the capabilities and fundamental limits of
general quantum systems to process different kinds of information. To do so, we will need
an abstract version of the formalism of quantum theory, which can be seen as an extension
of classical probability theory.

1 The big picture

The formalism of quantum mechanics is based on three fundamental notions: States, time-
evolutions and measurements. The state of a general physical system describes all its proper-
ties, or at least all the properties we care about in some physical description. The formalism
of time-evolutions describes how the system evolves in time after being in some initial state.
To compute the time-evolution of a system we usually need to know how its constituents
interact with each other. In classical physics, we would be done now, but in quantum physics
a special role is played by the measurement process. Formally, a measurement is distinct
from time-evolutions in two aspects: While the time-evolution in quantum mechanics will
turn out to be reversible, i.e., the state at some time uniquely determines the state at prior
times, this is not true for measurements where the state undergoes an irreversible change.
The time-evolution is also deterministic in the sense that states at all times are uniquely
determined by the states at prior times. However, measurements are fundamentally proba-
bilistic, i.e., random measurement outcomes will be observed with a probability distribution
determined by the measurement and the state that is measured. The problem to reconcile
these two conflicting descriptions of reality is called the measurement problem. To this day, it
is an open problem at the foundation of quantum mechanics. While the strange predictions
of quantum theory have been replicated in experiments countless times to high degrees of
accuracy we are far away from resolving this issue.
On the way to a general formalism of quantum mechanics, we will start by the description
of closed quantum systems. A quantum system is called closed when it does not interact with
any other quantum system. Historically, this was the starting point for quantum theory, but
it was later observed that this description is unsatisfactory: Even in well-controlled labratory
settings, quantum systems will interact with the environment (e.g., through magnetic fields
from electric currents, radioactive background or cosmic radiation) and can therefore not
considered to be closed. In principle, we could consider the whole universe as a closed
quantum system, but this would be hopelessly complicated. Instead, we will develop a
general formalism of open quantum systems, i.e., quantum systems that may interact with
an environment, which include the closed quantum systems as a special case. This will lead
to the general formalism commonly used in quantum information theory and throughout the
rest of the course.

1
2 The first postulate: State space
The formalism of quantum mechanics in its modern form is build upon the theory of Hilbert
spaces, or complex Euclidean spaces if we restrict to finite dimensions. In this course, we
will restrict to finite-dimensional quantum systems and measurements with finitely many
outcomes. In practice, this is usually no restriction since any experiment will only have a
finite resolution and therefore all occurring objects can be discretized accordingly.

Postulate 1 (State space). For every quantum system, there is an associated complex Eu-
clidean space H called the state space.

The first postulate sets the stage for the general formalism and we will have to explain
how to model states, time-evolutions and measurements on general state spaces.

3 Postulates for closed quantum systems

Let us start with the postulates of quantum mechanics for closed quantum systems. In this
section, H will always denote the state space of the closed system.

Postulate 2 (Pure states). The states of a closed quantum system are represented by vectors
|ψi ∈ H satisfying hψ|ψi = 1.

As we will see below, the previous postulate is not entirely true: While every state of a
closed system can be represented by a normalized vector in Hilbert space, this representation
is redundant. This aspect is crucial, and we will discuss it further towards the end of this
section.

Postulate 3 (Time-evolution of closed systems). Interaction within the system is modelled

via a Hermitian operator H ∈ B (H)sa called the Hamiltonian. When the system is initially
prepared in a pure state |ψi ∈ H, then the time-evolution is described by the Schrödinger
equation
d
|ψ(t)i = iH|ψ(t)i, |ψ(0)i = |ψi.
dt
Let us pause for a while, to fully appreciate the previous postulate. For any Hamiltonian
H ∈ B (H)sa the Schrödinger equation can be solved by defining the evolution operator
R
Ut : H → H by Ut = exp (iHt) for each t ∈ . For an initial state |ψ(0)i = |ψi ∈ H, the
time-evolution is then described by

|ψ(t)i = Ut |ψi.

It is easy to check, that {Ut }t∈R defines a group of unitary operators with identity U0 = 1H
and Ut−1 = U−t . In fact, we have the easy describtion Ut = U t for the fixed unitary
U = exp (iH). Postulate 3 implies that closed quantum systems evolve by a unitary trans-
formation. Moreover, for any given unitary U ∈ U (H), we can always find a Hamiltonian
H ∈ B (H)sa such that U = exp (iH) and therefore, we can formulate the previous postulate
in the following alternative form:

Postulate 30 (Time-evolution of closed systems II). If the closed system is in an inital

state |ψi ∈ H and evolves in time to a state |ψ 0 i ∈ H, then there exists a unitary operator
U ∈ U (H) such that |ψ 0 i = U |ψi.

Finally, we need to introduce the notion of measurement. Note that measurements require
an interaction of the closed system with an external observer, and strictly speaking this means
that our quantum system is no longer closed when this happens. In this sense, the system

2
does not need to evolve according to a unitary transformation during the measurement
process, and indeed we will see that (in general) it does not. In fact, the measurement
process will result in an instantaneous and irreversible change of the system’s state, which
has been referred to as the collaps of the wave function.

Postulate 4 (Projection-valued measure (PVM)). For every measurement of a closed quan-

tum system there is an associated set of projections {PP N
n }n=1 ⊂ Proj (H) corresponding to
N
measurement outcomes n ∈ {1, . . . , N } and satisfying n=1 Pi = 1H . Measuring a state
|ψi ∈ H results in the outcome n with probability

pn = hψ|Pn |ψi.

After measuring a state |ψi ∈ H and obtaining the outcome n, the system is in the post-
measurement state
Pn |ψi
|ψn i = p .
hψ|Pn |ψi
Measurements are the only way to obtain classical data from a quantum state. Since
any measurement is a probabilistic process, we will usually consider the whole statistics of
measurement outcomes as the classical data obtained from the measurement. Here, it is
tacitly assumed that we can perform some sort of experiment (see Figure 1), consisting of a
preparation of the state and a measurement, many times independently to obtain the statis-
tics of outcomes. This can be seen as the most basic interpretation of quantum mechanics,
saying that the mathematical objects are just describing the statistics obtained from repeat-
ing identitcal experiments and nothing more. However, there are other interpretations and
the question whether the quantum states should be considered as “real” or just as “states
of knowledge” is the subject of ongoing debate. Another major problem at the foundation
of quantum theory is the measurement problem:

Figure 1: A quantum experiment.

Problem 3.1 (The measurement problem). What is the physical process behind the mea-
surement formalism? How can it change the state of the system instantly? Why is this
process formally different to a time-evolution, when every measurement apparatus should, in
principle, also be a quantum system?

We need to make some further remarks about PVM defined in the Postulate 4. It can be
shown that the projection operators in any PVM are orthogonal, i.e., they satisfy Pn Pm = 0
whenever n 6= m. This has an important consequence: After measuring the system and
obtaining an outcome n ∈ {1, . . . , N } measuring the system again will always result in the
same outcome n. It can therefore be said, that the system’s state does not have a well-defined

3
(“classical”) value of the quantity measured by a PVM prior to the measurement process,
but that afterwards the value is well-defined.
Postulate 4 has an important implication for our version of Postulate 2 stated above:
Note that the probabilities pn = hψ|Pn |ψi of obtaining the outcome n ∈ {1, . . . , N } are the
same when measuring any state eiα |ψi ∈ H for α ∈ R
with the PVM {Pn }N n=1 ⊂ Proj (H).
We say that the measurement statistics does not depend on any global phase factor. Since
measurement outcomes are all the information we can ever obtain from a quantum system it
makes sense to identify all vectors eiα |ψi ∈ H and refer to them as a single state. Formally,
this means that we should consider pure states as elements of a complex projective space.
There is a simple an elegant solution to this problem, which we will adopt in the general
formalism. Instead of representing pure states by vectors |ψi ∈ H, we will represent them by
rank-1 projections |ψihψ| ∈ Proj (H). Note that this takes care of the aforementioned issue
and all vectors eiα |ψi for α ∈R and some fixed |ψi ∈ H define the same projector |ψihψ|.
Moreover, we can formulate all the previous postulates by using projections instead:

• Pure states: Every state of a closed system is represented by a projection |ψihψ| ∈

Proj (H) for some |ψi ∈ H.

• Time-evolution: States of closed systems evolve unitarily, i.e., |ψihψ| 7→ U |ψihψ|U †

for some unitary U ∈ U (H).

• Measurements: Measuring a pure state |ψihψ| ∈ Proj (H) using a PVM {Pn }N n=1 ⊂
Proj (H) results in measurement outcome n ∈ {1, . . . , N } with probability

pn = hψ|Pn |ψi = Tr [|ψihψ|Pn ] .

After obtaining the outcome n ∈ {1, . . . , N }, the system is in the state

Pn |ψihψ|Pn
|ψn ihψn | = .
hψ|Pn |ψi

So far, we have only considered a single closed quantum system without any additional
structure. In quantum information theory, we will often encounter quantum systems that
are composed of simpler systems. The last postulate describes the state space of such com-
positions:

Postulate 5 (Composite systems). If a closed quantum system is composed of two quantum

systems with state spaces HA and HB , then its state space is given by the tensor product
HAB = HA ⊗ HB .

This postulate has profound consequences for the description of quantum systems. For
example, it implies that the dimension of the state space of a quantum system composed of
N quantum systems with d-dimensional state spaces is dN , i.e., exponentially large in the
number N . This fact makes it very challenging to numerically simulate quantum systems
on classical computers and it is one motivation to develop quantum computers which could
simulate quantum systems faster.
Let us close this section with an important observation: There are two distinct classes of
pure states on composite systems:

• Pure product states: These are states of the form |ψAB ihψAB | = |φA ihφA | ⊗ |τB ihτB |
for pure states |φA ihφA | ∈ Proj (HA ) and |τB ihτB | ∈ Proj (HB ).

• Pure entangled states: These are states of the form |ψAB ihψAB | ∈ Proj (HA ⊗ HB )
that are not pure product states.

4
 In the next section, we will obtain a description for the state of a system ‘A’ that is part
of a closed composite system ‘AB’ in some pure state |ψAB ihψAB |. If the composite system
‘AB’ is in a product pure state |ψAB ihψAB | = |φA ihφA | ⊗ |τB ihτB |, then this is quite easy, and
we should just take the pure state |φA ihφA | to be the state of the system ‘A’. Why does this
make sense? Consider the situation, where you have two closed quantum systems ‘A’ and
‘B’ modelled by the complex Euclidean spaces HA and HB , respectively. The formalism of
closed quantum systems can then be applied in the following way (see Figure 2):

• Pure states of a pair of closed systems: Pure product states

|ψAB ihψAB | = |φA ihφA | ⊗ |τB ihτB |.

• Product time-evolution: Time-evolution by product unitary UA ⊗ UB for UA ∈

U (HA ) and UB ∈ U (HB ).

• Product measurement: The measurement using the product PVM {PnA ⊗ Pm

B}
n,m
for PVMs {PnA }N B M
n=1 ⊂ Proj (HA ) and {Pn }m=1 ⊂ Proj (HB ).

Figure 2: Two closed quantum systems.

Note that measuring a pure product state |φA ihφA | ⊗ |τB ihτB | using a product measure-
ment {PnA ⊗ Pm B}
n,m yields the product distribution pA × pB of the outcome distributions
pA and pB obtained from measuring the pure states |φA ihφA | and |τB ihτB | with the PVMs
{PnA }N B M
n=1 and {Pn }m=1 , respectively. This is exactly, what we should expect from measuring
two closed systems, which, by definition, do not interact with each other or any other system.
Evolving the systems using product time-evolutions only leads to other product states, and
the closedness of the systems is preserved.

4 Quantum states
What is the problem with the formalism of closed quantum systems? Consider the situation
where you want to describe a quantum system contained in your laboratory, which is affected
by the environment outside of the laboratory. An example could be an atom that is affected
by the electromagnetic field of a nearby powerline, or by cosmic radiation coming from
particles crashing into earth’s atmosphere at high speed. It seems hopeless to describe the
closed system consisting of the lab, and all other quantum systems that might interact with

5
 Figure 3: The laboratory as an open quantum system.

it. The formalism of open quantum systems solves this problem and it allows us to treat the
system of interest directly.
The philosophy behind this approach assumes the universe to be split in two parts:
The system ‘A’ we are interested in (e.g., the laboratory which we can control) and the
environment ‘E’ which interacts with our system in a complicated way (and which we cannot
control!). See Figure 3 for a visualization. Together the composite system ‘AE’ will be closed
and we can use the postulates of closed quantum systems to describe its pure states, time-
evolutions and measurements. The formalism we are going to develop allows us to express
these objects on the system ‘A’ alone, which amounts to treat ‘A’ as an open quantum
system (being part of the system ‘AE’ joint with the environment). The notion of density
operators is at the basis of this formalism. We will simply call these objects quantum states,
since they generalize pure quantum states, and in the following we will see how they arise
naturally from the formalism of closed quantum systems.

Quantum states from quantum marginals: Given a composite quantum system, how
can we talk about the subsystems individually? How should we represent their state? We
can use the measurement formalism to answer this question. Imagine a setting, where there
are two systems ‘A’ and ‘E’ modelled by complex Euclidean spaces HA and HE , respectively,
such that their composition ‘AE’ is closed. Given a PVM {Pn }Nn=1 ⊂ Proj (HA ), we aim to
AE N
find a PVM {Qn }n=1 ⊂ Proj (HA ⊗ HE ) that describes the measurement process when
measuring the PVM {Pn }N n=1 on the system ‘A’ alone. A reasonable guess would be to set

QAE
n = Pn ⊗ 1HE , (1)
for every n ∈ {1, . . . , n}. We will now show that there is no other choice for this PVM:
Theorem 4.1. The PVM {QAE N
n }n=1 ⊂ Proj (HA ⊗ HE ) given by (1) is the only PVM for
which the outcome distribution when measuring the pure product state |φA ihφA | ⊗ |τB ihτB |
for any |φA ihφA | ∈ Proj (HA ) and any |τB ihτB | ∈ Proj (HB ) coincides with the outcome
distribution of measuring the pure state |φA ihφA | with the PVM {Pn }N
n=1 .

Proof. Consider the pure states

(n) (n) Pn |φA ihφA |Pn
|ψAE ihψAE | = ⊗ |τE ihτE |,
hφA |Pn |φA i
for any n ∈ {1, . . . , N }, pure states |τE ihτE | ∈ Proj (HE ) and |φA ihφA | ∈ Proj (HA ). In
each of these states, the ‘A’ system is in a post-measurement state of the PVM {Pn }N n=1 ⊂
Proj (HA ) and we know that another measurement with the same PVM leads to a determin-
istic outcome. By the assumptions on {QAE N
n }n=1 ⊂ Proj (HA ⊗ HE ) we have

Tr (Pn |φA ihφA |Pn ⊗ |τE ihτE |) QAE

 
m = hφA |Pn |φA iδnm , (2)

6
for each n, m ∈ {1, . . . , N } and any pure states |τE ihτE | ∈ Proj (HE ) and |φA ihφA | ∈
Proj (HA ). Defining the positive operator
QA AE
m [τE ] = (1A ⊗ hτE |) Qm (1A ⊗ |τE i) ,

we conclude that
hφA |Pn QA A
 
m [τE ] Pn |φA i = Tr Pn |φA ihφA |Pn Qm [τE ] = hφA |Pn |φA iδnm

for each n, m ∈ {1, . . . , N } and any pure state |φA ihφA | ∈ Proj (HA ). Since Pm QA
m [τE ] Pm
and Pm are selfadjoint operators, we have
Pm QA
m [τE ] Pm = Pm .

Moreover, we have hφA |Pn QA A

m [τE ] Pn |φA i = 0 whenever n 6= m, and using that Qm [τE ] is
positive, this implies
Pn |φA i ∈ ker QA


m [τE ] ,
for any |φA i ∈ HA . Since all involved operators are selfadjoint, we conclude that
QA A
m [τE ] Pn = Pn Qm [τE ] = 0,
P
whenever n 6= m. Finally, using 1HA − Pm = n6=m Pn shows that

QA A
m [τE ] (1HA − Pm ) = (1HA − Pm ) Qm [τE ] = 0,

and since Pm QA
m [τE ] Pm = Pm , we find that

QA A
m [τE ] = (Pm + (1HA − Pm )) Qm [τE ] (Pm + (1HA − Pm )) = Pm ,

for any m ∈ {1, . . . , N }. This identity holds independently of |τE i ∈ HE , and it is easy to
show (e.g., by expanding it in a basis) that this implies
QAE
m = Pm ⊗ 1HE ,

for any m ∈ {1, . . . , N }, and we are done.

The previous theorem implies that the measurement from (1) is the only sensible choice
for a PVM corresponding to the partial measurement of {Pn }N n=1 ⊂ Proj (HA ) on the system
‘A’ of a joint closed system ‘AE’. Of course, it might not be clear that there has to be such
a PVM, and in the end it is a physical statement that the measurement formalism extends
independently of the underlying state. We will not bother to much about this now, and we
just take (1) as the definition of a partial measurement on any closed quantum system, i.e.,
not neccessarily closed quantum systems which are composed of closed quantum systems.
Now, we can get back to our initial question: How can we represent the state of a
subsystem of a composite system? For this, consider a PVM {Pn }N n=1 ⊂ Proj (HA ). If the
system ‘AE’ is in the pure state |ψAE ihψAE | ∈ Proj (HA ⊗ HE ), then measuring the PVM
{Pn ⊗ 1HE }N n=1 ⊂ Proj (HA ⊗ HE ) gives outcome n with probability

pn = Tr [|ψAE ihψAE | (Pn ⊗ 1HE )] = Tr [ρA Pn ] , (3)

where we have introduced the operator
dE
X
ρA = (idA ⊗ Tr) (|ψAE ihψAE |) := (1A ⊗ hiE |) |ψAE ihψAE | (1A ⊗ |iE i) ,
i=1

obtained by applying the trace partially to the pure state |ψAE ihψAE |. It can be checked
that the operator ρA is positive and has unit trace. Conveniently, these properties guar-
antee that the numbers pn = Tr [ρA Pn ] define a probability distribution for any projective
measurement {Pn }N n=1 ⊂ Proj (HA ). Operators with these properties are sometimes called
density operators, but we will use the following terminology:

7
Definition 4.2 (Quantum states). A quantum state on H is a positive operator ρ ∈ B(H)+
satisfying Tr [ρ] = 1. We denote the set of quantum states by D (H)
The above reasoning suggests, that ρA is the correct object to represent the state of the
subsystem ‘A’ within the composite system ‘AE’. Let us state the following definition:
Definition 4.3 (Partial trace). The linear map TrB = idA ⊗ Tr : B(HA ⊗ HB ) → B(HA )
is called the partial trace (on the system ‘B’). When no confusion can arise, we will use
the notation TrA , TrB , TrAB , etc., to denote the partial traces for subsystems carrying the
corresponding labels.
The partial trace is analogous to taking the marginal of a classical probability distribution
(see Lecture 1). The particular operators ρA or ρE obtained by taking the partial traces of a
pure state |ψAE ihψAE | are also known as the reduced density operators of |ψAE ihψAE |. The
formalism of open quantum systems will express all measurements and time evolutions in
terms of such reduced density operators on the system ‘A’, i.e., tracing out the environment
‘E’. We will later see, that this is equivalent to considering the full set D (H) of quantum
states, i.e., every quantum state arises as the reduced density operator of some pure state.

Quantum states from ensembles: We have seen how quantum states arise naturally
as reduced density operators, but there is another natural way of thinking about them.
Consider a closed quantum system with associated Euclidean space H and imagine that we
can prepare it in any pure state from a set {|ψk ihψk |}K k=1 . Intuitively, it should then be
possible to prepare the system in a statistical mixture of pure states, where we prepare the
state |ψk ihψk | with probability qk for some probability distribution q ∈ P ({1, . . . , K}). How
can we represent such a statistical mixture using the Euclidean space H?
Remember that everything we can know about the quantum system is obtained by mea-
surements. What statistics should we expect by measuring our statistical mixture of quan-
tum states using the PVM {Pn }N n=1 ⊂ Proj (H). By combining the preparation and the
measurement we should obtain the measurement outcome n with probability
K
"K #
X X
pn = qk hψk |Pn |ψk i = Tr qk |ψk ihψk |Pn . (4)
k=1 k=1

The previous formula is suggests that we could represent the statistical mixture of the pure
states |ψk ihψk | with probability qk by the operator
K
X
ρ= qk |ψk ihψk |.
k=1

Again, this operator is positive and has unit trace, i.e., it is a quantum state. From the
spectral theorem, we conclude that each quantum state can be written as some statistical
mixture of (orthogonal) pure quantum states. In fact, there are in general many different
ways of expressing a density operator as a statistical mixture, and we will see later that it is
a crucial property of quantum theory that we cannot tell such equivalent mixtures apart.
It should not come as too much of a surprise that quantum states describe statistical
mixtures, since this is a special case of the reduced density matrix from the last chapter. If
we consider the quantum state
K
X
ρCA = qk |kihk| ⊗ |ψk ihψk |, (5)
k=1
PK
then we obtain the reduced density operator ρA = TrC [ρCA ] = k=1 qk |ψk ihψk |.

8
Quantum states, knowledge and entanglement: There is another point to make re-
duced density operators and statistical mixtures. How can we tell whether a pure quantum
state |ψAB ihψAB | is entangled or not? Using the partial trace, this is easy: A pure quantum
state |ψAB ihψAB | is a product state if and only if its partial trace ρA = TrB [|ψAB ihψAB |]
is a pure state as well. This fact has an interesting interpretation: Since non-pure density
operators arise as statistical mixtures of pure states, they represent situations where our
knowledge about the quantum system is incomplete1 (we do not know which of the pure
states has been prepared). In this sense, closed quantum system is in an entangled pure
state if and only if our knowledge about its subsystems is incomplete.

5 Time-evolutions of open systems: Quantum channels

How can we model time-evolutions on the quantum system ‘A’ without having access to
the environment ‘E’ ? To answer this question, we will first consider a general, but also
very concrete scenario where we can derive the answer from the formalism of closed systems.
Afterwards, we will reason more abstractly about these time evolutions and finish this section
with a definition.

Concrete quantum channels: Consider first the situation, where the composite system
‘AE’ is in any pure quantum state |ψAE ihψAE | ∈ Proj (HA ⊗ HB ) with reduced density
operator ρA = TrE [|ψAE ihψAE |]. If the composite quantum system ‘AE’ undergoes a time-
evolution represented by a unitary operator UAE , then the evolved reduced quantum state
is h i
†
ρ0A = TrE UAE |ψAE ihψAE |UAE .
By computing some examples, it is easy to verify that the map ρA 7→ ρ0A is not well-defined
on reduced density matrices and in general it depends on the pure state |ψAE ihψAE | realizing
ρA as a reduced density operator. This means, that there is no reduced description of this
time-evolution, and it can only be described on the whole system ‘AE’.
Which time-evolutions can be described on the system ‘A’ alone? Consider the situa-
tion, where the composite system ‘AE’ is initially in a product quantum state ρA ⊗ σE ∈
D (HA ⊗ HE ) for a fixed quantum state σE ∈ D (HE ) (not neccessarily pure). Then, the
same process as before leads to the well-defined linear map
h i
†
ρA 7→ ρ0A = TrE UAE (ρA ⊗ σE ) UAE . (6)
While time-evolutions should correspond to transformation of a system ‘A’, we could also
describe general physical processes transforming a system ‘A’ into another system ‘B’. For
example, we could consider the situation, where the composite system ‘ABE’ is in the quan-
tum state ρA ⊗ σBE for a fixed quantum state σBE ∈ D (HB ⊗ HE ) (not neccessarily a
product state). Using a unitary UABE ∈ U (HA ⊗ HB ⊗ HE ), we can obtain a well-defined
linear map by h i
†
ρA 7→ TrAE UABE (ρA ⊗ σBE ) UABE ∈ D (HB ) . (7)
The linear maps in (6) and (7) are instances of quantum channels, which are the most general
physical processes transforming a system ‘A’ into itself or into a system ‘B’ definable on the
reduced description of these systems alone. Moreover, we will later show that any general
quantum channel (see definition below) can be written as in (7) with suitable choices of
σBE and UABE . Note that, unlike unitary time-evolutions, quantum channels are in general
not reversible. Physically, the information needed to reverse the evolution is leaked to the
environment and it cannot be reversed on its output alone.
1
Note that having complete knowledge about a quantum system (i.e, it is in a pure state) does not mean
that we can predict measurement outcomes with certainty!

9
Abstract quantum channels: Let us think more abstractly about physical processes
transforming an open system ‘A’ to an open system ‘B’. How should we model such a process?
If we can build a formalism based on the notion of quantum states, then physical processes
in this formalism should be represented by maps taking quantum states on the input system
to quantum states on the output system. Moreover, we should expect such maps to be linear
since the time-evolutions on closed quantum systems are linear2 . Finally, these maps should
be applicable to subsystems of larger systems as well. Using the measurement formalism
one can again argue, that applying a linear map T : B (HA ) → B (HB ) partially on the
subsystem ‘A’ of a composite system with Euclidean space HE ⊗ HA amounts to applying
the idE ⊗ T . Again, this map should map quantum states to quantum states.

Figure 4: Partial application of a quantum channel.

Let us now properly define the general notions used to formalize quantum theory. We
start by introducing some abstract properties of linear maps, which will be crucial in the
formalism of open quantum systems. A linear map T : B (HA ) → B (HB ) is called
• positive if T (X) ∈ B (HB )+ for every X ∈ B (HA )+ .

• completely positive if (idE ⊗ T ) : B (HE ) ⊗ B (HA ) → B (HE ) ⊗ B (HB ) is positive

for every complex Euclidean space HE .

• trace-preserving if Tr [T (X)] = Tr [X] for every X ∈ B (HA ).

Note that every completely positive map is also positive since we can choose HC to be 1-
dimensional. It can be checked, that the concrete examples of quantum channels in (6) and
(7) are completely positive and trace-preserving, and we will take these properties as the
basis for a general definition.
Definition 5.1 (Quantum channels). A linear map T : B (HA ) → B (HB ) is called a
quantum channel if it is completely positive and trace-preserving.

6 Measurements of open systems: POVMs and instruments

How should we model measurements of an open quantum system? To answer this question,
we will again consider a concrete measurement setting in the formalism of closed quantum
systems. Formally, we have to distinguish to problems:
1. What is the most general way to obtain a classical measurement statistics from a
quantum system?

2. How does a general measurement process change the state of a quantum system?
2
For a better argument see...

10
We will see that these two questions have slightly unrelated answers. We will answer the first
question by stating the most general form of destructive measurements, i.e., measurements
for which the measured quantum system is (considered to be) destroyed in the process. The
answer to the second question will lead to a general formalism of non-destructive measure-
ments, which contains the destructive measurements as a special case. While it is tempting to
focus on the non-destructive case, it is not advised to do so. In many situations the destruc-
tive measurement formalism is sufficient and much easier to handle than the non-destructive
formalism. Moreover, when we are anyway only interested in classical information about
quantum systems, then there is no need to keep any quantum systems indefinitely.

Destructive measurements: In the same way as we have obtained quantum channels

as reductions of certain time-evolutions of the composite system ‘AE’, we can also consider
the reduction of PVMs on the composite system ‘AE’ to the system ‘A’. Again, we have
to be careful about the initial state since entanglement might prevent us from obtaining a
well-defined reduced description. Consider the situation, where the composite system ‘AE’
is in a product quantum state ρA ⊗ σE ∈ D (HA ⊗ HE ) for a fixed quantum state σE ∈
D (HE ) (not neccessarily pure). Measuring this product state with the PVM {PnAE }n∈N ⊂
Proj (HA ⊗ HE ) results in measurement outcome n ∈ {1, . . . , N } with probability

pn = Tr (ρA ⊗ σE ) PnAE = Tr ρA QA
   
n , (8)

for the operators

QA AE
 
n = TrE (1HA ⊗ σE ) Pn .
+ PN
It can be checked that QAn ∈ B (HA ) for every n ∈ {1, . . . , N } and
A
n=1 Qn = 1HA . These
properties guarantee that the numbers pn are positive and add up to 1, i.e., that
PN theyAdefine a
A N
probability distribution. The set of positive operators {Qn }n=1 satisfying n=1 Qn = 1HA
is a special instance of a positive operator-valued measure (POVM), which are the most
general measurements in quantum theory. The general definition is as follows:
+
Definition 6.1 (POVM). APset of operators {Qn }N
n=1 ⊂ B (H) is called a positive operator-
N
valued measure (POVM) if n=1 Qn = 1H .

Note that this definition is necessary for the numbers defined in (8) to be probabilities
for any quantum state ρA .

Non-destructive measurements: In the previous section, we have only specified the

probabilities of obtaining a given measurement outcome, but not the quantum state of the
system after the measurement. Indeed, specifying the POVM {Qn }N n=1 is in general not
sufficient to determine the post-measurement state of the system, and it is convenient to
regard such a measurement as destructive, i.e., the quantum system is considered to be
destroyed in the measurement process. What additional information is needed to determine
the post-measurement states of the measurement process outlined above? Again, consider a
quantum state σE ∈ D (HE ) and a PVM {PnAE }n∈N ⊂ Proj (HA ⊗ HE ). Then, measuring
the product state ρA ⊗ σE ∈ D (HA ⊗ HE ) leads gives the outcome n with the probability in
(8). After obtaining the outcome n the reduced density operator on the system ‘A’ will be
PdE †
TrE PnAE (ρA ⊗ σE ) PnAE
 
(n) i=1,j=1 Kn,i,j ρA Kn,i,j Tn (ρA )
τA = AE
= AE
= , (9)
Tr [(ρA ⊗ σE ) Pn ] Tr [(ρA ⊗ σE ) Pn ] Tr [Tn (ρA )]

for the operators Kn,i,j ∈ B (HA ) given by

√
Kn,i,j = ai (1HA ⊗ hjE |) PnAE (1HA ⊗ |φi i) ,

11
where σE = di=1
PE
ai |φi ihφi | is the spectral decomposition of the positive operator σE . It is
easy to check that the linear maps Tn : B(HA ) → B(HA ) given by
dE
†
X
Tn (X) = Kn,i,j XKn,i,j ,
i,j=1

(n)
are completely positive, which implies that each τA is a quantum state, i.e., it is positive
and has unit trace. Furthermore, note that
dE
†
X
Kn,i,j = TrE (1HA ⊗ σE ) PnAE = QA
 
Kn,i,j n,
i,j=1

which are the operators of the POVM considered in the previous paragraph. This implies
"N #
X
Tr Tn (X) = Tr [X] , (10)
n=1

such that the statistical mixture of post-measurement states

N N N
(n) (n)
X X X
pn τA = Tr [Tn (ρA )] τA = Tn (ρA ) ,
n=1 n=1 n=1

is a quantum state as well. The set {Tn }N n=1 represented by operators Kn,i,j as above and sat-
isfying (10) is a special instance of an instrument, which are the most general non-destructive
measurements in quantum theory. The formal definition is as follows:
Definition 6.2 (Instruments). A set {Tn }N
P of completely positive maps Tn : B (HA ) →
n=1
HB is called an instrument if the sum T = Nn=1 Tn is a quantum channel.

7 The formalism of quantum theory

We can now combine all the insights from the previous section to state the postulates of
quantum theory in their modern form. The first postulate stays the same as before, and we
merge it with the postulate for composing quantum systems:
Postulate 1 (State space).
1. For every quantum system, there is an associated complex Euclidean space H called the
state space.
2. The state space of a composite quantum systems ‘AB’ arises as the tensor product
HAB = HA ⊗ HB from the state spaces HA and HB of its subsystems.
The second postulate is modified using the notion of density operators introduced above:
Postulate 2 (Quantum states). The states of a quantum system with state space H are
represented by density operators ρ ∈ D (H), which we simply call quantum states.
As before, we will say that a quantum state ρ ∈ D (H) is pure, if it coincides with a
rank-1 projection, i.e., if ρ = |ψihψ| for some unit vector |ψi ∈ H. Note that the pure states
are the extreme points of the convex set ρ ∈ D (H).
The next definition is slightly more general than what we have discussed above. When
considering general physical transformations, i.e., all transformations that can be obtained
by combining partial traces and time-evolutions, there is no reason that the dimension of the
input system and the output system should coincide. However, it is easy to see by the same
reasoning as above that such maps still need to be completely positive and trace-preserving.
This justifies the following:

12
Postulate 3 (Quantum channels). The physical transformations of a quantum system ‘A’
into a quantum system ‘B’ is represented by a quantum channel T : B(HA ) → B(HB ), i.e.,
a trace-preserving and completely positive map.

A special case of a quantum channel is a unitary quantum channel T : B(H) → B(H)

given by T (X) = U XU † for a unitary U ∈ U (H). Therefore, the notion of quantum channel
contains the time-evolutions of closed systems as a special case. Quantum channels are, in
general, not reversible, i.e., their inverse is usually not a quantum channel3 . Examples of
irreversible quantum channels are, e.g., the constant quantum channels T : B(HA ) → B(HB )
given by T (X) = Tr [X] σ for some fixed quantum state σ ∈ D(HB ).
When discussing measurements, we have the two notions of destructive and non-destructive
measurements:

Postulate 4 (Measurements and instruments). When measuring a quantum system we dis-

tinguish the following cases:

• Destructive measurements:PThey are modelled by a POVM {Qn }N n=1 , i.e., a set

of positive operators satisfying N Q
n=1 n = 1. Measuring a quantum state ρ ∈ D (H)
results in the measurement outcome n with probability

pn = Tr [ρQn ] .

• Non-destructive measurements: They are P modelled by an instrument {Tn }N

n=1 ,
i.e., a set of completely positive maps such that N T
n=1 n is a quantum channel. Mea-
suring a quantum state ρ ∈ D (H) results in the measurement outcome n with proba-
bility
pn = Tr [Tn (ρ)] ,
and after obtaining outcome n the quantum state of the system is

Tn (ρ)
ρn = .
Tr [Tn (ρ)]

Some authors would not use the terms non-destructive measurements and instruments
as synonyms, but to us this seems like a rather arbitrary distinction and we will keep the
terminology introduced here. We will also continue to use the term PVM to mean a POVM
in which all operators are orthogonal projections.
Now, we have formulated the postulates of quantum theory. By our discussion from
the previous sections, these postulates contain all states, time-evolutions and measurements
of open quantum systems. Since our new postulates contain the old postulates for closed
quantum systems as special cases (pure quantum states are the states of closed systems,
unitary quantum channels describe their time-evolution, etc.) it is clear that they are more
general. In the next section we will see that the two sets of postulates are in fact equivalent.
Given any object in the theory of open quantum systems it is always possible to dilate it to
an object in the theory of closed quantum systems.

3
It can be shown that a quantum channel T : B(H) → B(H) is reversible if and only if it is a unitary
quantum channel.

13