Professional Documents
Culture Documents
Photonic Multipartite Entanglement
Photonic Multipartite Entanglement
entanglement
Generation, measurement and applications
Massimiliano Smania
Abstract
We are currently witnessing a fundamental change in the field of quantum information, whereby protocols and experiments
previously performed in university labs are now being implemented in real-world scenarios, and a strong commercial push
for new and reliable applications is contributing significantly in advancing fundamental research. In this thesis and related
included papers, I first look at a keystone of quantum science, Bell's theorem. In particular, I will expose an issue that
we call apparent signalling, which affects many current and past experiments relying on Bell tests. A statistical test of
the impact of apparent signalling is described, together with experimental approaches to successfully mitigate it. Next, I
consider one of the most refined ideas that recently emerged in quantum information, device-independent certification.
Device-independent quantum information aims at answering the question: "Assuming we trust quantum mechanics, what
can we conclude about the quantum systems or the measurement operators in a given experiment, based solely on its results,
while making minimal assumptions on the physical devices used?". In my work, the problem was successfully approached
in two different scenarios, one based on entangled photons and the other on prepare-and-measure experiments with single
photons, with the aim of certifying informationally-complete quantum measurements. Finally, I conclude by presenting
an elegant and promising approach to the experimental generation of multi-photon entanglement, which is a fundamental
prerequisite in most modern quantum information protocols.
Keywords: quantum information, entanglement, Bell tests, POVM, device-independent, self-testing, quantum optics,
prepare-and-measure.
Stockholm 2020
http://urn.kb.se/resolve?urn=urn:nbn:se:su:diva-182523
ISBN 978-91-7911-030-7
ISBN 978-91-7911-031-4
Department of Physics
Stockholm University, 106 91 Stockholm
PHOTONIC MULTIPARTITE ENTANGLEMENT
Massimiliano Smania
Photonic multipartite
entanglement
Generation, measurement and applications
Massimiliano Smania
©Massimiliano Smania, Stockholm University 2020
v
vi
Acknowledgements
vii
not be complete without the latest addition to my family: Vani, the efforts
and sacrifices you made so I could reach this goal are what made it worth
in the first place. Thank you for taking care of me as if your life depended
on it, your attention is what kept me going.
Karin and Paolo, thank you for being the best friends to live next to in
a pandemic, and so much more. Actually, thank you Karin for the Sam-
manfattning too, for giving me a place to sleep, for covering my rent when
I could not pay, for sharing apartments with me throughout my PhD, and
additional crucial help which would need a separate section just to list.
Thank you Andrea for exchanging doubts on our aptitude as researchers,
and for long and intricate discussions on all of the big themes in the world.
Both made us stronger. I am also grateful to all the friends that have made
my time outside Albanova so much fun, in particular the Desert Island team
and the ever-expanding Italian club.
Sincere thanks also to all the friends who, although not in Stockholm,
played a fundamental role during the time of my PhD studies anyway. In
particular, I would like to thank Davide, Luca, Alice, Ale, Ale, Degia, Joey,
Danel and Tri. Your friendship has time and again proven stronger than the
distance that divides us.
Last but not least, the research work carried out in this thesis was funded
by the Knut and Alice Wallenberg Foundation and the Swedish Research
Council.
viii
Contents
Abstract v
Acknowledgements vii
List of Figures xv
Sammanfattning xviii
Preface xix
ix
2 Bell tests 13
2.1 The EPR paradox . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Bell’s theorem and the CHSH inequality . . . . . . . . . . 14
2.3 Loopholes . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4 Applications . . . . . . . . . . . . . . . . . . . . . . . . . 18
3 Experimental background 21
3.1 Single photon polarisation as a qubit . . . . . . . . . . . . 21
3.2 The main actors in polarisation manipulation . . . . . . . 22
3.2.1 Wave-plates . . . . . . . . . . . . . . . . . . . . . 22
3.2.2 Polarisers . . . . . . . . . . . . . . . . . . . . . . 23
3.2.3 (Polarising) beam-splitters . . . . . . . . . . . . . 24
3.2.4 Polarisation-dependent filter . . . . . . . . . . . . 25
3.3 Polarisation qubit state preparation . . . . . . . . . . . . . 27
3.4 State analysis and detection . . . . . . . . . . . . . . . . . 27
4 Signalling 31
4.1 A definition . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.1.1 A mathematical definition . . . . . . . . . . . . . 32
4.2 Some key motivations . . . . . . . . . . . . . . . . . . . . 33
4.3 Experimental sources of signalling . . . . . . . . . . . . . 35
4.3.1 Solutions and results . . . . . . . . . . . . . . . . 36
4.4 Concerning common assumptions . . . . . . . . . . . . . 37
4.5 Additional experimental work . . . . . . . . . . . . . . . 37
4.5.1 State characterisation . . . . . . . . . . . . . . . . 38
4.5.2 Wave-plate motor hysteresis . . . . . . . . . . . . 38
4.5.3 Laser power stability . . . . . . . . . . . . . . . . 39
x
6.2 Scaling up to GHZ states . . . . . . . . . . . . . . . . . . 55
6.3 General setup characterisation . . . . . . . . . . . . . . . 58
6.3.1 Source brightness . . . . . . . . . . . . . . . . . . 60
6.3.2 Source fidelity . . . . . . . . . . . . . . . . . . . 63
6.4 Characterisation of individual sources . . . . . . . . . . . 68
6.4.1 2-photon sources . . . . . . . . . . . . . . . . . . 68
6.4.2 4-photon sources . . . . . . . . . . . . . . . . . . 70
6.4.3 6-photon source – γ 6 . . . . . . . . . . . . . . . . 72
6.5 Comparison with literature . . . . . . . . . . . . . . . . . 74
6.5.1 Future outlook . . . . . . . . . . . . . . . . . . . 75
References 79
xi
List of Papers
The following papers, referred to in the text by their Roman numerals, are
included in this thesis.
xii
Author’s contribution
PAPER I: I designed the experimental setup, built it, and performed the
measurements. I carried out a relevant part of the data anal-
ysis and participated in theoretical discussions. I contributed
to writing the paper.
PAPER II: I and M. Nawareg designed and built the experimental setup.
I performed all the measurements independently, except for
the full state tomography, which I did with M. Nawareg. I
carried out the data analysis and participated in developing
the theoretical background. I contributed to writing the paper.
PAPER III: I designed and built the experimental setup. I performed the
measurements and their analysis. I participated in discus-
sions of the theoretical ideas in the paper, and I contributed
in writing the experimental sections.
xiii
Relevant papers not included in the
thesis
xiv
List of Figures
3.1 A beam-splitter. . . . . . . . . . . . . . . . . . . . . . . . 24
3.2 Polarisation-dependent filter. . . . . . . . . . . . . . . . . 25
3.3 Single qubit state preparation. . . . . . . . . . . . . . . . 27
3.4 A measurement station for single photon polarisation qubits. 28
4.1 Simulated CHSH result for equal and very unequal overall
efficiencies. Reported from Paper I. . . . . . . . . . . . . 34
4.2 Example of experimental data showing high apparent sig-
nalling. Reported from Paper I. . . . . . . . . . . . . . . 35
4.3 Test of wave-plate motor hysteresis. . . . . . . . . . . . . 38
4.4 Test of power stability of pump laser. . . . . . . . . . . . . 39
xv
6.13 A measure of photon indistinguishability as function of pump
power. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
6.14 Experimental quantum state tomography of 2-qubits. . . . 70
6.15 Experimental quantum state tomography of 4-qubits. . . . 71
6.16 Experimental results of measurement of operators in Eq.
(6.11). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
6.17 Experimental results of measurement of operators in Eq.
(6.12). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
6.18 Expected 8-photon rates as function of pump power and
collection efficiency. . . . . . . . . . . . . . . . . . . . . 76
xvi
List of Tables
1.1 Pauli operators and their eigenstates, labelled with light po-
larisation directions. . . . . . . . . . . . . . . . . . . . . . 3
xvii
Sammanfattning
xviii
Preface
One evening around halfway through my PhD studies, I found myself demon-
strating a Bell test experiment to a group of rather enthusiastic students as
part of a modern physics course. The course was included in a general
physics program, which skimmed over theorems and equations in favour of
the more popular aspects of the discipline. After going through the optics
included in the experimental setup, I quickly performed the few measure-
ments required for the estimation of the final result. Result which, I an-
nounced in darkness, turned out to be safely beyond the boundary of clas-
sical physics, which I had previously explained on the whiteboard. A few
hands popped up, and I swiftly dispensed equation-free, dogmatic answers
to the apparent satisfaction of the class. That was, until one perplexed-
looking student asked: “What is entanglement?”. Realising that the stan-
dard definition would require knowledge of at least quantum superposition
and state separability, I scrambled for an answer which the students could
cope with. I went with the all-time favourite: two identical objects are of
two colours at the same time, and when we check the colour of one of them,
they both immediately turn the same colour, no matter how far apart they
are. The class was not satisfied. They pointed out there is nothing wrong
with the objects being one of two colours from the beginning, and no weird
instantaneous influence is required to explain why they always turn out of
the same colour. That was correct of course, but I could not possibly say
“that is the difference between a mixed quantum state and entanglement”,
could I? Unable to find a better answer, I resorted to playing the card of dog-
matic knowledge once again, and simply said that if the students’ remark
were true, our little experiment would not have violated Bell’s inequality,
and that more knowledge was of course required to really understand the
definition of entanglement.
Unsurprisingly perhaps, there really is no way to explain the concept
of entanglement without invoking some of the features which are unique to
quantum mechanics. Although the consequences of entanglement used to
puzzle some of the best minds of the previous century [1], they are now ac-
cepted and mostly understood by the scientific community. In recent years,
quantum information has been transitioning from books and labs to cata-
logues and (online) shops, by means of multi-national public and private ef-
xix
forts. While the industry shows a growing interest in investing in quantum
information and its applications, researchers are sharpening their existing
tools and exploring new paradigms in view of real-world implementations.
In this context, my PhD work aimed at both refining current experi-
mental approaches to quantum information applications, and probing some
of the more recent ideas emerging in the field. After a brief introduction
of some fundamental theoretical concepts in Chapters 1 and 2, I will dis-
cuss in the present thesis how such theoretical concepts were translated into
experiment in Chapter 3. Chapter 4 will report on how, in Paper I, we ex-
posed and characterised a potential issue common to some of the best Bell
experiments found in literature. On the other hand, in Paper II and III,
we explored the recent concept of device-independent quantum informa-
tion in relation to a type of measurement rather unusual in quantum optics,
that is, non-projective measurements. The different approaches to the prob-
lem of device-independent certification taken in the two papers are outlined
in Chapter 5. However, none of the work in the mentioned papers could
be carried out without the development and characterisation of a source of
entangled photons. Our novel approach on multi-photon entanglement gen-
eration is what took most of my time and efforts during my PhD and, since
not yet published, it is reported in details in Chapter 6. The results will
constitute part of Paper IV, which is currently in preparation.
The three introductory chapters are built upon my licentiate thesis, which
was defended on December 20, 2018, half-way through my PhD studies.
Among the papers discussed here, Paper I was included in that disserta-
tion, although in the form of a previous version. As a consequence, Chapter
4 too is partially taken from my licentiate thesis. More in detail, the contri-
butions from my previous dissertation are as follows:
Chapter 1 : Sections 1.1, 1.2, 1.3, the first part of 1.6, and 1.7 were present
in the licentiate thesis. They have been limitedly extended.
Chapter 2 was present in the licentiate thesis. It has been reviewed and
corrected where needed.
Chapter 3 was present in the licentiate thesis, except for Section 3.2.4.
Section 3.4 has been extended.
Chapter 4 : Sections 4.1, 4.2, 4.3 and 4.5 were present in my licentiate
thesis. They have been significantly reworked and extended.
Chapter 6 : Section 6.1 was present in my licentiate thesis. It has been
reviewed and limitedly extended.
All the remaining parts are new.
xx
1. Bits of quantum information
1
with 0 ≤ θ ≤ π and 0 ≤ φ ≤ 2π respectively being latitude and longitude
on the sphere.
In the same way that we can express states either as linear combina-
tions of basis states or as vectors, these two representations can be used for
operators, or observables. For example, Pauli’s σz operator is:
1 0
σz = |0i h0| − |1i h1| = . (1.4)
0 −1
Pauli operators (reported in Tab. 1.1) take on a very special role in quan-
tum information theory: their eigenvectors constitute the three mutually
unbiased bases in the qubit space. Moreover, Pauli matrices, together with
the identity matrix (sometimes referred to as σ0 ), span the 2 × 2 matrix
space. For this reason, any operation on a qubit can be expressed as a linear
combination of these four matrices. As observables, Pauli operators fulfil
Heisenberg’s uncertainty relation. They will be used throughout this work.
When acting on a qubit, Pauli operators σx , σz and σy respectively per-
form a bit-flip, a phase-flip, and both of these operations, in the computa-
tional basis (see Tab. 1.1).
The six states making up the three Pauli bases are depicted in Fig. 1.2.
2
Operator Matrix Eigenstates Operation
0 1 |0i±|1i σx |0i = |1i
σx |±i = √
1 0
2 σx |1i = |0i
0 −i |0i−i|1i |0i+i|1i σy |0i = i |1i
σy |Ri = √ , |Li = √
i 0
2 2 σy |1i = −i |0i
1 0 σz |0i = |0i
σz |Hi = |0i , |V i = |1i
0 −1 σz |1i = − |1i
Table 1.1: Pauli operators and their eigenstates, labelled with light polarisa-
tion directions.
Figure 1.2: The six eigenstates of Pauli operators σx , σy and σz are drawn in,
respectively, red, blue and green on the Bloch sphere.
1 0 0 0
0 1 0 0
|00i = |0i ⊗ |0i =
0 , |01i = 0 , |10i = 1 , |11i = 0 . (1.5)
0 0 0 1
3
Similarly to the case of one qubit in Eq. (1.1), we can write the general pure
state of two qubits as
As a final note, even though qubits are often associated with a particu-
lar physical quantum system, for example photons, to the point where the
two terms are used interchangeably (as will be the case further on in this
thesis), it is important to notice that while a qubit is a mathematical entity,
a photon or an atom are instead physical ones. Single photons or atoms can
in principle store several - infinite even - qubits [3].
4
includes any entanglement among its sub-parts, each part alone will be de-
scribed by a mixed state. The other way around is also true in a way: mixed
states can always be described as a reduction of a pure state living in a
bigger Hilbert space.
Intuitively, a single concept can explain the mixing of a pure state due
to two as seemingly different causes as decoherence and entanglement: a
system (for simplicity, we will consider a qubit) in a pure state interacting
with the environment exchanges information with it. This interaction gen-
erates entanglement, and the initial information representing the pure state
is thus spread on a far bigger system. Unfortunately, we are usually unable
to measure the environment at the same time as our qubit, and therefore
the information yielded to the former is lost. As a consequence of measur-
ing just the qubit part of the greater entangled system (which also includes
the environment), we necessarily obtain a mixed state. We will consider
entanglement at greater length in the next section.
As far as representation is concerned, the most convenient way to work
with mixed states is within the density matrix formalism, which also works
for pure states. Each matrix element is given by hi| ρ | ji, where i, j =
{1, ..N} refer to basis vectors in the N-dimensional Hilbert space where the
state is defined. The density matrix is therefore basis-dependent, although
the computational basis is normally used. As an example, the state in Eq.
(1.1) can be written as
∗
αα αβ ∗
ρ = |ψi hψ| = . (1.9)
α ∗β β β ∗
The eigenvalues of the density matrix are the probabilities pi of Eq. (1.8),
thus ρ is Hermitian and positive semi-definite. Finally, a normalised state
always has tr (ρ) = 1 and tr ρ 2 ≤ 1, where the inequality is saturated only
for pure states, therefore establishing a simple distinction between pure and
mixed states. Moreover, a density matrix represents a pure state if and only
if it is idempotent (i.e. ρ = ρ 2 ), or if it has rank one.
5
Since the eigenvectors form an orthonormal basis, any pure state |ψi can
always be expressed as a function of them of the form:
∑ Ei† Ei = 1. (1.13)
i
6
In the special case in which the operators are also self-adjoint (Ei = Ei† )
and orthogonal to each other, i.e. Ei E j = δi j Ei , the measurement is called
projective, and the operators are orthogonal projectors. Therefore, an ob-
servable1 E describing a projective measurement can be decomposed as
E = ∑ ei E i , (1.14)
i
while for mixed states Eq. (1.12) is in general used. Although being a very
special case of quantum measurements, it should be noted that projective
measurements are the most commonly implemented in quantum informa-
tion experiments, by a wide margin.
∑ Fi† Fi = ∑ Ei = 1. (1.16)
i i
The set {Ei = Fi† Fi } consists of positive operators that are in general not
orthogonal to each other. Each Ei is a POVM element, or effect, and the
whole set is simply called a POVM. Upon measuring the POVM on a pure
state, the probability of outcome i is:
interchangeably.
7
For mixed states, the probability is again estimated as
Two specific examples of POVMs that were realised for qubits in our labs
will be introduced in the following: the SIC-POVM and the Trine-POVM.
1.4.3 SIC-POVM
In the quest for the best possible POVM, two properties are highly desir-
able: first, the measurements should be able to reconstruct any quantum
state; second, it should do so in the optimal way. A measurement with
the first property is called informationally complete (IC), since its statistics
uniquely identify the state it acted upon. The property of optimality is in-
stead associated with the POVM having the minimal amount of outcomes,
and the outcomes being maximally independent. Such a POVM is called
symmetric. Symmetric informationally-complete (SIC) POVMs are thus
the ideal quantum measurements in terms of information extracted from the
measuring process. The simplest way to describe a SIC-POVM in dimen-
sion d is with a set of d 2 normalised states {|φi i} satisfying [4]
2 1
hφ j |φi i = , j 6= i. (1.19)
1+d
Starting from these states, the POVM effects can be defined as the sub-
normalised projectors:
|φi i hφi |
Ei ≡ , (1.20)
d
which are a complete set (see Eq. (1.16)). It is clear from Eq. (1.19)
that effects Ei are not orthogonal to each other.
2 The same
equation also
expresses the symmetry condition Ei E j = 1/ d (1 + d) , for i 6= j. On the
other hand, starting from Eq. (1.19) one can also conclude that SIC-POVM
effects are linearly independent, which implies that the set is IC [4].
The existence of SIC-POVMs in arbitrary dimensions is to date an open
question, although algebraic or numerical solutions in many dimensions are
known [5; 6]. In our work, we focused on the simple d = 2, where pen-and-
8
paper calculations lead to the four projectors
|φ1 i = |0i ,
√
|0i + 2 |1i
|φ2 i = √ ,
3
√
|0i + 2ei2π/3 |1i (1.21)
|φ3 i = √ ,
3
√
|0i + 2ei4π/3 |1i
|φ4 i = √ ,
3
from which the SIC-POVM effects can be calculated with the help of Eq.
(1.20). The four vectors are easily remembered thanks to their representa-
tion in the Bloch sphere, where they point to the corners of a regular tetra-
hedron, as depicted in Fig. 1.3 (left).
∣ϕ1⟩ ∣ψ1⟩
∣ϕ4⟩ ∣ϕ3⟩
∣ϕ2⟩ ∣ψ2⟩ ∣ψ3⟩
Figure 1.3: The SIC-POVM (left) and Trine-POVM (right) projectors, de-
picted on the Bloch sphere, from Eqs. (1.19) and (1.22) respectively.
1.4.4 Trine-POVM
In Paper III, we also considered a different non-projective measurement
for qubits, called Trine-POVM. The name comes from the fact that this
measurement has three outcomes, whose corresponding projectors on the
Bloch sphere form an equilateral triangle as in Fig. 1.3 (right). Due to
the normalisation condition in Eq. (1.16), the three vectors lie on the same
plane in the Bloch sphere, and can for example be written as:
|ψ1 i = |0i
√
|0i + 3 |1i
|ψ2 i = (1.22)
2√
|0i − 3 |1i
|ψ3 i = ,
2
9
for a Trine-POVM on the xz disk. Thanks to its symmetry and simplicity, the
Trine-POVM made for a good candidate for demonstrating the self-testing
methods explained in Paper III, and in particular the concept of robustness
to experimental noise.
1.6 Entanglement
Perhaps the most striking and essential concept in QM, entanglement never
fails to puzzle anyone who comes across it, from non-scientists to the great-
est physicist of the last century. The idea that quantum objects can be bound
by correlations that are stronger than any classical correlation one might
conceive of, was so alien and counter-intuitive that, after Einstein, Podol-
sky and Rosen pointed out some of its apparently unnatural consequences
in 1935 [1], it took almost thirty more years to figure out a way to test these
concepts in a lab. These tests are one of the main concerns of this work.
From a mathematical perspective, a pure quantum state is not entangled
if it can be expressed as the tensor product of its subsystems, while a mixed
state is not entangled if it is equivalent to a mixture of non-entangled pure
states. A general non-entangled mixture can be expressed as
10
where the probabilities pi sum to unity and superscripts refer to each part
of the composite system. Because entanglement is a concept that inherently
involves two parts of a system, it is fairly well studied and characterised
in bipartite quantum systems. Up to six total dimensions - thus in case of
qubit-qubit or qubit-qutrit systems - the Peres-Horodecki criterion gives a
necessary and sufficient condition for separability [8; 9], therefore clearly
indicating whether a quantum system contains any entanglement. In higher
dimensions though, the criterion only provides sufficiency for separability.
It should be noted however, that in the special case of pure states, a sim-
ple method to determine if the system is entangled consists in calculating
its partial trace. If any of its sub-partitions turns out to be mixed, then the
original system necessarily contained entanglement. A system composed
by n parts is entangled even if only one of its 2n−1 − 1 possible partitions
contains entanglement. If all such partitions are entangled the system is said
to possess genuine entanglement.
11
measurements that grows exponentially with N. For this reason, the witness
above is commonly used in multi-photon entanglement experiments [12–
15], and will be employed extensively in Chapter 6 when characterising the
multi-photon entanglement source used in our work.
12
2. Bell tests
The terminology “Bell tests” encompasses a rather large and diverse group
of experiments that since the ’70s have aimed at shedding light on some
of the most fundamental questions about reality: is Nature “real” and in-
dependent of our knowledge and perception of it? Is it local, that is, only
influenced by its surroundings? In this chapter we will go deeper into the
notion of local realism and exactly how quantum mechanics seems to give
a way-out of the contradictions that Einstein, among others, ran into.
13
fluencing them, prior to them being measured. Such a theory is said to
be local realist, because it satisfies the principle of realism, according to
which reality is independent of our description of or interaction with it, and
its physical properties are totally defined at any point in space and time.
Quantum mechanics, on the other hand, not only cannot predict values of
incompatible observables with certainty, but goes as far as suggesting that
these values are not defined before measurement, all of it while seemingly
admitting simultaneous non-local interactions. The idea that elements of re-
ality could be influenced instantaneously by far-away events was too much
to accept for the inventor of special relativity, who suggested that a theory
that does not explicitly comply with locality cannot be correct. As we know
now, quantum mechanics is indeed not local realist, but it does not violate
causality [18, p. 426-428], therefore complying - at least in principle - with
Einstein’s idea of locality.
While most interpretations of the theory, including the ever-favourite
Copenhagen Interpretation, reject realism, therefore admitting that observ-
ables are not defined prior to measurement, other interpretations - most fa-
mously De Broglie-Bohm’s - do not discard determinism as easily, prefer-
ring instead to explicitly break the locality principle. To date, no definitive
argument has closed the discussion on which principle to follow.
We thus concluded that quantum theory cannot fulfil both locality and
realism at the same time. A careful reader might think that we are missing
the important question though: what can be inferred on Nature and local
realism from real-world experiments?
14
inequality [19], as reported by Bell himself in [20, p. 36-37].
Suppose we have two qubits that are sent to two stations which perform
measurements along directions ~a and ~b respectively. The initial state of the
qubits is described entirely by (possibly hidden) variables λ , and measure-
ment outcomes can be A(~a, λ ) = ±1 and B(~b, λ ) = ±1. The fact that A and
B do not depend on ~b and ~a respectively, is the locality assumption.
The expectation value of the experiment will be
Z
E ~a,~b = A ~a, λ B ~b, λ ρ(λ ) dλ (2.1)
Using the constraints mentioned above on the outcomes and the triangle
inequality, we get
Z
|E a, b − E a, b0 | ≤ 1 ± A a0 , λ B b0 , λ ρ(λ ) dλ
Z
+ 1 ± A a0 , λ B b, λ ρ(λ ) dλ
= 2 ± E a0 , b0 + E a0 , b ,
15
Choosing suitable settings, Inequality (2.2) is comfortably violated:
√
|E 0◦ , 45◦ − E 0◦ , 135◦ | + |E 90◦ , 45◦ + E 90◦ , 135◦ | = 2 2 > 2 .
√ (2.4)
The value 2 2 is also the maximum reachable by quantum physics, and
known as Tsirelson’s bound [21].
We have therefore concluded that quantum mechanics cannot be local
realist. What about Nature?
Since the 70s, a great amount of Bell experiments have been performed,
mainly with photons as qubits [22–25], but also ions [26] and other systems
as atom condensates [27] or nitrogen-vacancies in diamonds [28]. Needless
to say, all of these experiments, and countless others, have overwhelmingly
shown that Nature is indeed not compatible with local realism. It is impor-
tant to notice that all of these experimental tests are not directed at proving
that quantum mechanics is correct or complete, but rather at demonstrating
that Nature cannot be described by any local realist theory. The fact that
some of the results come really close to Tsirelson’s bound, without over-
coming it, is at best an additional indication that quantum mechanics is the
best theory of its kind that we have at the moment. Moreover, the degree
of assurance these experiments offer changes from case to case, mainly be-
cause of assumptions that the authors had to make in order to violate the
inequality. These assumptions are often called loopholes, and we shall take
a look at them in the next section.
2.3 Loopholes
During the years, as new and more advanced Bell experiments were pro-
posed, researchers came up with weaknesses in the setups that left room for
(local) hidden variable models to invalidate results that would have other-
wise ruled out local realism. Some of these so-called loopholes are setup-
specific, but the most important and experimentally demanding ones are
found in most experimental realisations until very recently. A selection of
them includes:
Locality loophole Originally identified by Bell himself [20], this loophole
corresponds to the breaking of the main assumption in Bell’s work,
i.e. that the outcome of a measurement on one of the qubits has to be
independent from the measurement setting on the other qubit. This
is usually solved by involving relativistic causality: if measurement
settings on each qubit are chosen after the two qubits are far apart,
enough that no signal can travel between the two before the measure-
ment is completed, then there can be no causal correlation between
16
settings and outcomes. This loophole was already closed in a famous
experiment by Aspect and others in 1982 [23], when fast polarisers
were used for measurements of photon polarisation.
Free-will loophole We have already seen that Bell’s key assumption is that
outcomes A and B cannot depend on settings b and a respectively.
Suppose we go back in time to when the measurement devices and
the source of qubits were stored together in the same room, or even
much further to the Big Bang if necessary: if all future events have
been determined by an interaction that happened before the experi-
ment, then there can be no independence between the measurement
devices at all! This metaphysical concept is also known as super-
determinism. To-date, physicist have not yet figured out a solution to
this matter, and it is likely that one is not even possible. Whether this
is within the realm of Physics is obviously debatable, but considering
the astounding philosophical implications of Bell’s theorem, his reply
upon being accused of doing metaphysics, for considering this matter
in the first place, is easily shared: “Disgrace indeed, to be caught in
a metaphysical position! But it seems to me that in this matter I am
just pursuing my profession of theoretical physics.” [20].
For the first time in 2015, in a span of a few weeks, three research groups
managed to independently close both the locality and detection loopholes
at the same time. Two of them used entangled photons [24; 25], and one
combined photons and electronic spin in nitrogen-vacancies [28]. These ex-
periments have shown, covering all known and addressable loopholes until
proven otherwise, that Nature is not local realist.
17
2.4 Applications
As we have seen in the previous sections, Bell tests were originally con-
ceived as a way of determining whether Nature could be compatible with
any local realist theory of hidden variables, or if non-local theories, as quan-
tum mechanics, are the right path to a better understanding of the universe.
Besides the fundamental scope of Bell’s work, and its profound philosoph-
ical implications, researchers have, throughout the years, identified several
other fields that can take advantage of his theorem. Some interesting exam-
ples are reported here.
18
All of the above-mentioned applications require either loophole-free Bell’s
inequality violations, or additional assumptions to meet the setting/outcome
independence requirement. In particular, a very common assumption con-
sists in expecting there to be no signalling (or no communication) between
Alice’s and Bob’s measurement stations. As we shall see in Chapter 4, this
assumption is often taken for granted and not suitably tested.
19
20
3. Experimental background
21
states are derived, as reported in Tab. 1.1. Once a photon is prepared in a
given polarisation, it will maintain it as long as it propagates through air or
other non-birefringent media, while optical fibres and glass will in general
modify the polarisation state.
3.2.1 Wave-plates
Birefringent materials can present two different refractive indexes to light
propagating through them. In particular, in our work we used wave-plates
made of quartz, which is a positive (ne > no ) uni-axial crystal. Light polar-
isation components along the ordinary and extraordinary axes will experi-
ence a relative phase retardance φ which depends on wavelength λ , crystal
thickness d and refractive index difference in the following way:
2πd (ne − no )
φ= . (3.1)
λ
In matrix notation, this phase shift corresponds to
0 1 0
W (φ ) = . (3.2)
0 eiφ
There are two particular values of the phase φ that are of interest: when
φ = π, the wave-plate is called half wave-plate (HWP), while φ = π2 corre-
sponds to a quarter wave-plate (QWP). These components usually differ in
thickness d. By changing the angle θ between the optic axis and the inci-
dent light polarisation, we can vary the relative phase between the |Hi and
|V i components.
In the lab, we conventionally set θ = 0 when the vertical polarisation is
aligned to the optic axis. In this case, the effect of a wave-plate is expressed
by Eq. (3.2), for given d and λ . To describe the more interesting scenario of
θ 6= 0, one can rotate the frame of reference to the one where θ = 0, apply
matrix W 0 (φ ), and then rotate back to the lab frame.
Using the rotation matrix
cos θ sin θ
R (θ ) = , (3.3)
− sin θ cos θ
22
we can define the generic wave-plate matrix as
W (θ , φ ) = RT (θ )W 0 (φ ) R (θ )
cos2 θ + eiφ sin2 θ sin θ cos θ 1 − eiφ (3.4)
= .
sin θ cos θ 1 − eiφ sin2 θ + eiφ cos2 θ
3.2.2 Polarisers
While a wave-plate is essential in performing (ideally) unitary operations
on polarisation qubits, polarisers are very useful whenever a specific state
is required, or if we want to go from a mixed to a pure state. These are in
practice polarisation filters that only let through one particular polarisation,
depending on the orientation angle at which they are set.
The polarisers we use in the lab consist of thin parallel nano-wires that
absorb the polarisation component orthogonal to the wire orientation. By
rotating the component around the propagation axis, any linear polarisation
on the xz equator can be prepared. The action of a polariser, at an angle θ
with the vertical polarisation, can be found by applying again the same trick
as used for wave-plates, of changing frame of reference:
T 0 0 sin2 θ − sin θ cos θ
Pol (θ ) = R (θ ) R (θ ) = . (3.7)
0 1 − sin θ cos θ cos2 θ
It is worth noting that the operation of such a polariser is not unitary: the
absorbed component is “lost”.
1 Overall phases in front of operator matrices are purposefully ignored.
23
3.2.3 (Polarising) beam-splitters
Perhaps the most crucial optical component in quantum optics experiments,
a beam-splitter (BS) is a semi-transparent planar - or cubic - piece of glass
with (at least) two input and two output directions. It can be used to split,
combine, or interfere light beams or single photons. The BSs we used are
cubic glass devices consisting of two triangular halves, adhering to each
other’s hypotenuse facet through a dielectric film.
If we consider the BS in Fig. 3.1, we can describe its operation within a
second-quantisation-type of formalism [37, Section 6.2], using annihilation
operators as field amplitudes:
0
â2 t r â0
= 0 (3.8)
â3 r t â1
where (r, r0 ) and (t,t 0 ) are complex reflectance and transmittance as indi-
cated in Fig. 3.1, which are constrained by energy conservation.
Figure 3.1: A beam-splitter with two input ports (a0 , a1 ) and two output ports
(a2 , a3 ).
For a dielectric BS, the phase difference between transmitted and re-
flected beams is ±i. If we assign that phase to the reflected beam, and
further simplify to the case of a 50:50 lossless BS, we can rewrite Eq. (3.8)
as
â2 1 1 i â0
=√ . (3.9)
â3 2 i 1 â1
Polarisation-insensitive BSs as the one described are used for path-encoded
qubits and photon counting implementations, or wherever interferometry
plays a role.
In our case of polarisation qubits, a version of BS aptly called polar-
ising beam-splitter (PBS) was used. This device effectively combines the
polarisation degree of freedom with path, usually reflecting vertical polar-
isation and transmitting its horizontal component. In mathematical terms,
24
following the same notation as above
â0,H −→ â2,H
â0,V −→ iâ3,V
(3.10)
â1,H −→ â3,H
â1,V −→ iâ2,V
H2
H1
Output 2
Input 1
PBS
Output 1
Input 2
25
the path corresponding to Input 1 → clock-wise rotation → Output 1, while
|bi stands for Input 2 → counter-clock-wise rotation → Output 2. In this
notation, we can describe the PBS as
1 0 0 0
0 0 0 i
UPBS = 0 0 1 0 ,
(3.11)
0 i 0 0
while the two HWPs act jointly as
UθH1 θH2 = HWP(θH1 ) ⊕ HWP(θH2 ) =
cos 2θH1 sin 2θH1 0 0
sin 2θH1 − cos 2θH1 0 0 (3.12)
.
0 0 cos 2θH2 sin 2θH2
0 0 sin 2θH2 − cos 2θH2
The PDF operator is then found as UPDF (θH1 , θH2 ) = UPBS UθH1 θH2 UPBS . In
Dirac notation, the action of the PDF on the state |ψi = α |Hi + β |V i sent
through Input 1 is
UPDF (θH1 , θH2 )(α |H, ai + β |V, ai) =
= α cos 2θH1 |H, ai + β cos 2θH2 |V, ai (3.13)
+ iβ sin 2θH2 |H, bi + iα sin 2θH1 |V, bi .
If we were to disregard Output 2 completely, we see that the amplitudes of
the horizontal and vertical components at Output 1 can indeed be tuned ar-
bitrarily by simply setting the desired angles on the two wave plates, hence
the name PDF.
In our work reported in Papers II and III, we used both outputs of
the PDF in order to implement non-projective measurements. This was
achieved by adding a polarisation analysis setup as in Fig. 3.4 to every out-
put, and setting all the required angles and phases on the interferometer and
the measurement wave-plates. As an example, in the case of a 4-outcome
measurement we can express the Kraus operators for each outcome as
F1 = ha| UPBS QWP(θQa )HWP(θHa )W 0 (φa ) |ai ha| UPDF |ai ,
F2 = ha| UPBS QWP(θQb )HWP(θHb )W 0 (φb ) |ai hb| UPDF |ai ,
(3.14)
F3 = hb| UPBS QWP(θQb )HWP(θHb )W 0 (φb ) |ai hb| UPDF |ai ,
F4 = hb| UPBS QWP(θQa )HWP(θHa )W 0 (φa ) |ai ha| UPDF |ai ,
as functions of the PDF HWPs angles and the HWP, QWP and phase-plate
(W 0 ) angles in outcomes 1 (θHa , θQa , φa ) and 2 (θHb , θQb , φb ). From these,
POVM effects (see Section 1.4.2) are found as Ei = Fi† Fi .
26
3.3 Polarisation qubit state preparation
The simplest and cleanest way to prepare a single qubit in a desired pure
state is by using a combination of polariser and HWP plus QWP, as in Fig.
3.3.
PBS
|ψi = α |Hi + β |V i −−→ α |H, a2 i + iβ |V, a3 i . (3.16)
27
Positioning detectors at the outputs, we can count photons and derive prob-
abilities. For example, referring to Fig. 3.4 – and ignoring wave-plates for
now – we can experimentally determine the modulus squared of the coeffi-
cients in the state above:
N0
|α|2 = |hH, a2 | (α |H, a2 i + iβ |V, a3 i)i|2 =
N0 + N1
(3.17)
N1
|β |2 = |hV, a3 | (α |H, a2 i + iβ |V, a3 i)i|2 =
N0 + N1
where N0 and N1 are the number of “clicks” in detectors D0 and D1 during
the measurement.
28
so that the PBS will separate different (orthogonal) polarisations. For pro-
jective measurements in general, given the operator σ of an observable that
we would like to measure, we can find the wave plate angles experimentally
required by solving the following equation:
Wave-plate settings for the most common projective measurements are re-
ported in Tab. 3.1.
Table 3.1: Wave-plate angle settings are reported for Pauli measurement op-
erators, together with the PBS outputs associated to each projector.
29
30
4. Signalling
31
4.1 A definition
Bell’s requirement that measurement outcomes do not depend on far-away
measurement settings imposes that no communication happens between Al-
ice and Bob prior to or during measurement. Such communication, or trans-
fer of information, is also called signalling. We can therefore say that a
successful violation of a Bell inequality must be free from signalling. On
the other hand, as we have seen in Section 2.3, Bell tests have rather strict
requirements on overall experimental efficiency. For practical or techno-
logical reasons though, the greatest majority of experiments and applica-
tions to date do not meet this requirement, choosing instead to assume fair-
sampling, and taking for granted that no signalling took place in the partic-
ular scenario. After all, the idea that no communication happens between
two distinct and separate measurement devices does not seem that much of
a stretch. Still, in all of these cases it is important to rule out any dependence
between Alice’s settings and Bob’s outcomes, and vice-versa.
32
the experimental uncertainty, that is for example:
A
α+1,0 P(+1, _|0, 0) − P(+1, _|0, 1)
A
= A
, (4.2)
σ+1,0 σ+1,0
33
3.0
2.5
2.0
S parameter
1.5
1.0
0.5
0.0
0.0 0.2 0.4 0.6 0.8 1.0
p
Alice’s and Bob’s stations, and send them classically correlated sys-
tems instead. If she has access to Alice and Bob’s detectors, she could
then tamper with the efficiencies so that they get S > 2, and they will
not know about her unless they perform tests that include additional
measurements. Alternatively, the two parties can run the statistical
test of the non-signalling conditions that we presented in Paper I, on
the same data used to estimate S, which would show the failing of
those conditions, thus prompting them to discard the data. This type
of issue is particularly relevant for all applications where the CHSH
result is used as a means of certification, which could therefore be
rendered invalid.
34
40
35
25
20
15
10
0
0 200 400 600 800 1000
time (s)
For all these reasons, testing conditions (4.1) is not only a matter of
fulfilling Bell’s original assumptions in order to obtain a true violation, but
it is also crucial in order to have reliable qualitative and quantitative results
to use for applications.
1. Power drift of pump laser: if only one detector per station is used,
the two outcomes of each measurement have to be checked in se-
quence. If the total rate of photon changes between different settings,
in particular with a systematic drift, the fair-sampling assumption is
breached.
35
3. Measurement setting reproducibility: a low precision in setting the
angles of wave-plates or polarisers can lead to apparent signalling,
as it is equivalent, in a way, to Alice changing her measurement de-
pending on Bob’s setting. This is particularly problematic if only one
round of settings is performed.
While this list is of course not complete, it includes the most relevant errors
which are induced by standard procedures in CHSH experiments, especially
if one is not aware of the issue.
36
showed that repeating the four CHSH measurement settings several times
(a rigorous Bell test would also include setting randomisation), and using
more precise motors to rotate wave-plates, could significantly reduce the
effects of problem 3, assuming that the motor precision is symmetric around
a central position (more details about this are given in the next section).
Regarding problem 4, we described a simple method to verify the dif-
ference in collection efficiency, and used variable attenuators to equalise
them.
Finally, after taking care of all the above-mentioned issues, we ran a
CHSH experiment which yielded S = 2.812 ± 0.003, with a signalling sig-
nificance of 1.3 standard deviations. This is the highest violation that we
know of, free of apparent signalling.
The reader is referred to Paper I for more details on the different ex-
periments and their results.
As mentioned above, our work concerns experiments that assume fair sam-
pling, and therefore only use coincident detections for data analysis. All
works that do not suffer from the detection loophole, including in particular
loophole-free ones [24; 25; 28], calculate marginal probabilities from sin-
gle events alone, and will therefore not introduce any non-local effect in the
data processing stage. As a consequence, the test presented in Paper I will
not reveal any apparent signalling. While some of the experimental issues
introduced in Section 4.3 will of course impact the final result, for example
asymmetric efficiency, these will in general only deteriorate the Bell pa-
rameter, or should be accounted for as systematic uncertainties. Regarding
other common experimental assumptions, as locality or freedom-of-choice,
they are effectively irrelevant for the work presented above, since apparent
signalling is introduced after the measurements, due to the “non local” data
processing.
While the core results are detailed in Paper I, additional experimental work
which might give a more complete overview of what has been done in the
lab is reported here.
37
4.5.1 State characterisation
In order to saturate the CHSH inequality, a maximally entangled state is
required. Any of the Bell states in Eq. (1.26)
√ would be equally good, and we
chose to prepare |φ + i = (|00i + |11i)/ 2. Naturally, the actual prepared
state is not the ideal |φ + i, and in order to characterise it we measured its
visibility in the horizontal/vertical and diagonal bases. The results were:
16
14
12
Frequency
10
0
−0.06 −0.04 −0.02 0.00 0.02 0.04
< E > −E
As we supposed, the motor mostly went back to the same initial po-
sition, and deviations from it are distributed normally, showing a well-
38
behaved (unbiased) Gaussian behaviour.
0.16
0.14
0.12
Frequency
0.10
0.08
0.06
0.04
0.02
0.00
3272.5 3275.0 3277.5 3280.0 3282.5 3285.0 3287.5
Power (mW)
Figure 4.4: Histogram of pump laser power output as measured every 5 sec-
onds, for 7 hours. The data was fitted by a Gaussian curve that revealed an
emitted power of (3280 ± 2) mW.
39
40
5. Certifying a generalised
quantum measurement
41
5.1 Device independent quantum information
While Ekert’s original proposal for entanglement based QKD [17] already
contained the seeds of the device-independence paradigm, it took a few
years longer until Mayers and Yao explicitly proposed that observable ex-
perimental frequencies could uniquely determine the quantum state and
measurements used to obtain them [48]. Their idea was formalised in 2007,
when Acín and collegues proved that quantum non-locality, as certified by
a Bell inequality, could be used to enforce cryptographic security in pres-
ence of untrusted, black-box devices [49]. The security proof of DI-QKD
lead the way to subsequent application of the DI paradigm to other fields.
Most notably, Pironio and collaborators presented in 2010 a proof of DI
randomness expansion, along with an experimental demonstration [50].
Since those two initial proofs, the key advancements in the field can
be associated with a process of reconciling the abstract, ideal definitions of
devices in DI theory, with the very real and imperfect equipment used in
laboratories. Indeed, while DI protocols often make no theoretical a priori
assumptions on the specific functioning of devices in the lab, some basic
characterisation is usually still required, for example the isolation of these
black-boxes from each other. This reality-check type of process leads to the
analysis of scenarios where some assumptions are relaxed, and part of the
equipment is considered to be trusted and characterised, in order to lower
the extremely demanding experimental requirements imposed by fully DI
protocols. Two important scenarios that have been considered are repre-
sented by the concepts of semi-DI and measurement-DI, where the dimen-
sion of the quantum system is upper-bounded [51], or the source is trusted
[52], respectively. On the other hand, the idea of robust self-testing sprung
up, whereupon results yielded by imperfect devices lead to certifiable prop-
erties in spite of their sub-optimality, thanks to DI proofs that are tolerant to
experimental noise [53].
Papers II and III present two of these approaches, with the common
aim of certifying generalised quantum measurements.
42
on a qubit by taking advantage of a particular Bell inequality. Such an ap-
proach is characterised by the following steps:
While a similar procedure works very well when certifying a shared secure
key or random bit sequence, it runs into a delicate problem when certify-
ing non-projective measurements: because of Neumark’s dilation theorem,
a non-projective measurement on a given Hilbert space can in general be
re-cast as a projective measurement in a bigger space. As a consequence,
the third step in the procedure above relies on the assumption that the mea-
surement is performed on a qubit. As we will see further on, in Paper II we
arguably found a way around this problem, and at least partially removed
the qubit assumption. Nevertheless, this issue hints at a different way of
certifying non-projectivity. This second method is based on the Prepare-
and-measure paradigm, and it was realised in Paper III. Its most important
steps are:
2. Alice acts on the qubit with a unitary operator, thus preparing some
quantum state; she then transmits the qubit to Bob, who performs a
measurement on it. Again, relevant correlations between preparation
and measurement outcome are estimated.
4. If the experimental result turns out higher than those bounds, Bob’s
relevant measurement certifiably has four outcomes.
43
The assumption on the dimension of the quantum systems used, though
limiting, is arguably fairly natural in the scenario of generalised measure-
ment certification. The concept of non-projectivity in fact hints at a Hilbert
space of fixed dimension to begin with. On the other hand, the much sim-
pler prepare-and-measure scenario comes with several significant practical
advantages compared to entanglement-based experiments. From the exper-
imental point of view, neither space-like separation between observers nor
highly entangled states are required. On the theoretical/computational side,
the much smaller Hilbert space allows for a remarkable simplification of the
analysis, in turn enabling more advanced and versatile certification methods
to be developed. This is at least in part why we were able to carry the certifi-
cation methods much further in Paper III compared to Paper II. As a quick
overview, a comparison of some of the main features of the two approaches
used in the two papers is reported in Tab. 5.1.
Entanglement-based Prepare-and-measure
Quantum states Entangled Single qubits
Source Untrusted Partially trusted (dim bound)
Alice & Bob Space-like separated Sequential
Measurement Coincident Bob only
DI Full Partial
44
reason behind its name will become clear by the end of this section. Simi-
larly to its more famous CHSH sibling (or ancestor!), the EBI scenario in-
volves two parties, Alice and Bob, sharing a 2-qubit system. Alice performs
three binary measurements, while Bob four, out of which they each may get
outcomes +1 or −1. From the combined measurements, they calculate joint
probabilities P(a, b|x, y), where (x, y) are Alice’s and Bob’s measurement
settings, and (a, b) their respective outcomes. Using these frequencies, they
may estimate the value of the Bell parameter
45
√
bound of 4 3 as βel , and the bound is reached when the four vectors of
Alice’s A4 measurement are anti-aligned to Bob’s projective measurements,
thus forming again the SIC-POVM tetrahedron:
1 α −β (1 + i)
A4,1 = ,
2 β (−1 + i) 1−α
1 1−α β (−1 + i)
A4,2 = ,
2 −β (1 + i) α
(5.4)
1 1−α β (1 − i)
A4,3 = ,
2 β (1 + i) α
1 α β (1 + i)
A4,4 = ,
2 β (1 − i) 1 − α
√ √
with α = 3−6 3 and β = 63 .
Unfortunately, certifying the symmetric POVM with the above inequal-
ity is experimentally not feasible, since it would require to hit the optimal
quantum bound exactly. What can definitely be done though, is certifying
the arguably most interesting property of the SIC-POVM, that is, informa-
tion completeness. An IC-POVM with the minimal amount of outcomes,
that is, d 2 , is aptly called minimal informationally complete (MIC) POVM.
As we have shown in the supplementary material to Paper II, proving IC
for qubits is equivalent to showing that the measurement has four outcomes
(three independent plus normalisation). In order to check that, we have
computed the maximum value that can be obtained for a generalised variant
of Eq. (5.3), in case Alice’s A4 is a 3-outcome measurement. The calcu-
lations were carried out with Navascués, Pironio, and Acín’s semi-definite
programming method [55], also known as NPA.
46
calculated the bounds on βIC for 3- and 4-outcome measurements to be:
3-outcome 4-outcome
βIC ≤ 6.8782 ≤ 6.9883. (5.5)
Since four outcomes are IC for qubits, the higher bound also sets the limit
for quantum correlations.
Taking into account statistics from Alice’s A4 measurements, we ob-
tained the experimental result:
exp
βIC = 6.960 ± 0.007. (5.6)
The result lies more than 11 standard deviations above the 3-outcome bound,
therefore certifying that Alice’s A4 was a genuine 4-outcome, MIC measure-
ment.
47
the experimental probabilities P(b|x, y), in order to calculate the value of
witness
1
A0SIC = P(b = Sx,y |x, y), (5.7)
12 ∑
x,y
with
0 0 0
0 1 1
Sx,y =
1
. (5.8)
0 1
1 1 0
As in the entanglement-based protocol, the witness is maximised when Al-
ice’s four prepared states form a regular tetrahedron on the Bloch sphere (as
in Fig. 1.3 (left) for example), while Bob performs measurements √ σx ,σy
and σz . In this case, the theoretical maximum of ASIC = 1/2 1 + 1/ 3 ≈
0
0.7887 is achieved. The next step consists in providing Bob with an addi-
tional 4-outcome measurement (y = 4), while modifying the witness above
as
4
1
ASIC = P(b = S |x, y) − k ∑ P(b = x|x, y = 4). (5.9)
12 ∑
x,y
x,y x=1
√
The ideal result of ASIC = 1/2(1 + 1/ 3) would then certify that Bob’s last
measurement was the SIC-POVM. Due to the impossibility of eliminating
all experimental noise though, what can be concluded when a lower-than-
optimal result is obtained?
48
which is many standard deviations above both projective and 3-outcome
bounds, therefore certifying that Bob’s fourth measurement is genuinely a
4-outcome one.
Besides the qualitative conclusion just presented, the prepare-and-mea-
sure approach granted the possibility of a deeper, quantitative analysis of
the measurement. In Paper III, we introduced a method to pursuit robust
self-testing of a target non-projective measurement. Following that method,
we targeted the SIC-POVM, in order to get a quantitative estimate of the
fidelity of Bob’s fourth measurement to it. We therefore proceeded to nu-
merically approximate the worst-case fidelity of around 3 × 105 random 3-
and 4-outcome measurements, under the constraint of our experimental cor-
relations. The analysis returned a worst-case fidelity of approximately 0.98
to the qubit SIC-POVM. It is important to point out that this result is not
tight, but represents a numerical approximation made feasible by the bound
on the Hilbert space in the prepare-and-measure scenario. A tight lower
bound can in many instances be obtained with the SWAP method, although
for rather technical reasons this could not be pursued in the case of the SIC-
POVM. The reader is referred to Paper III for more details on why that was
the case, and for an example of a different non-projective measurement (the
Trine-POVM introduced in Section 1.4.4) where tight results were achieved
for both non-projectivity and fidelity to the target measurement.
49
50
6. Multi-photon entanglement
source
51
The experimental work outlined in this chapter has been carried out
in collaboration with Muhammad Sadiq, Hammad Anwer and Mohamed
Nawareg.
where ε0 is the free-space permittivity, χ (i) are the i-th order dielectric sus-
ceptibilities, and for simplicity we took both the polarisation P(t) and the
electric field E(t) to be scalar quantities.
In case of a material with large enough second order susceptibility, there
is a chance that a photon propagating through it will down-convert into two
photons of lower energy. The process has to preserve energy and momen-
tum, that is, respectively:
ω p = ωs + ωi
(6.2)
k~p = ~ks + ~ki ,
where omegas refer to frequencies of, in order, the incident field (or pump
photon), signal and idler photons. The second equation, involving mo-
menta, is also called phase-matching condition. Only the case of ωs = ωi ,
known as degenerate down-conversion, concerns this work. Because of
Eqs. (6.2), signal and idler photons will be naturally created with very
strong correlations, which pave the way for entanglement.
In our experiments we used Beta-Barium Borate (BBO), an uni-axial
birefringent non-linear material. The polarisation perpendicular to the plane
defined by ~k and the optic axis is called ordinary polarisation, and while
propagating, it will experience the ordinary refractive index no . Conversely,
a light field polarised in that plane is described as extraordinarily polarised,
and will witness the extraordinary refractive index ne . A BBO crystal can
1A
much more detailed description of non-linear processes can be found for
example in Chapters 1 and 2 of [62].
52
satisfy the phase-matching conditions in two ways, depending on the angle
of incidence of light with respect to the optic axis orientation:
6.1.1 Pump
Pump photons must have the exact spectral, temporal and polarisation prop-
erties needed for optimal down-conversion and entanglement yield. To this
end, we used a mode-locked Ti:Sapphire pulsed laser emitting 140 fs long
pulses at 780 nm, with a repetition rate frep of 80 MHz. The spectral width
of the emitted light was approximately 8 nm, while the average power was
3 W. To achieve the required degenerate down-conversion that yields 780
nm photons, the pump light underwent a non-linear process called second-
harmonic generation, which can be intuitively imagined as the opposite
of down-conversion. A 1 mm thick Bismuth triborate (BiBO) non-linear
crystal was employed in this case, producing a 390 nm – 1.1 nm wide –
deep-violet beam of around 1 W of average power after focusing the pump
on the crystal.
The out-coming laser light, collimated again, contained a significant
amount of residual 780 nm pump, which was filtered out with the help of
multi-reflection on several dichroic mirrors (see Fig. 6.1). The resulting
deep-violet pulses were then sent to the down-conversion source.
53
Figure 6.1: 780 nm red light pulses from the femto-second laser are focused
on a BiBO crystal for second harmonic generation (SHG). The up-converted
violet light at 390 nm is then collimated and sent to the down-conversion
setup, while residual red light is filtered out with the help of several dichroic
mirrors (DM).
H D
Pump polarisation
SPDC polarisation
54
propagating polarisations is left unaffected by the first crystal it encounters,
while undergoing type-I down-conversion inside the second. The down-
converted photon pairs of orthogonal polarisations, emitted in a cone whose
aperture depends on the phase-matching condition, are then recombined at
the beam-splitter, and reflected by a dichroic mirror onto a collimating lens.
Because of the very short duration of the femto-second pump pulses, the
down-converted light can be thought of as a ring. The rings coming from
the two crystals are shown as photographed by a single-photon CCD camera
in Fig. 6.3. After the collimating lens, their diameter is approximately 15
mm.
55
No filter
3 nm filter
4000
Intensity (AU)
3000
2000
1000
0
740 750 760 770 780 790 800
wave-length (nm)
56
Figure 6.5: Depiction of the gold-coated ten-facet pyramidal mirror used to
split the SDPC rings into ten different modes, from the side (left) and from
the top (right). Each mode corresponds to a single qubit in the prepared GHZ
state. The mode numbering is consistent with Fig. 6.7.
PBS
BBO BBO
390 390
Figure 6.6: Independent Bell pairs are entangled with each other by inter-
fering one photon from each pair on a PBS. Upon post-selecting the events
when one photon is obtained in each of the four output modes, the maximally
entangled |GHZ 4 i state is obtained. In our setup, different pairs correspond to
different directions around the down-conversion cone of a single BBO crystal.
pairs of entangled photons, we take the two signal photons and overlap them
on a PBS. The resulting interaction, known as Hong-Ou-Mandel (HOM)
interference, effectively entangles the four photons. Finally, to obtain a
GHZ state, we post-select the events when each PBS interaction results in
a photon at each outcome mode. Provided that the interacting photons are
indistinguishable, all which-path information prior to the PBS is erased,
though we can be certain that they will be both in |Hi or |V i states. The
resulting state after the PBSs (and post-selection) is the GHZ state of N
57
qubits
1
|GHZ N i = √ (|H1 ...HN i + |V1 ...VN i) . (6.5)
2
The above experimental configuration is also known as star configuration,
and is at the core of many recent realisations of multi-photon entangled
states [12–15].
In our labs, we realised the star configuration for up to five photon pairs,
although because of limits set by pair-generation rates, only four such pairs
can be entangled, and the remaining one is used as an independent high-
quality Bell state source. The setup configuration is depicted in Fig. 6.7.
Each mode and its respective prime constitute a source of entangled pairs,
and considering the entangling interference at the PBSs, we can identify the
nine entanglement sources reported in Tab. 6.1. In the remaining of this
chapter, I will discuss some of the key properties of the setup in general,
and then show relevant experimental characterisation of the sources from
Tab. 6.1.
58
Pump
5 5'
1' 4'
2' 3'
4 3 2 1
Filter
HWP
QWP
PBS
SMF coupler
59
• Quantitative yield Commonly referred to as source brightness, it
will in general depend on the photon collection efficiency ξ , and the
total pair generation probability (per pump pulse) RP . The resulting
N-fold coincidence probability for |GHZ N i states will be very close
to
N/2
(RP ξ N )/2N/2−1 , (6.6)
where the denominator is due to post-selection.
• Coupling of the photon spatial mode into the optical fibre before the
detectors.
• Detection efficiency.
60
setup is attenuated by a factor of approximately 3/4 from the SPDC crys-
tals to the fibre coupler, resulting in one every four photons not reaching
the fibre. The main culprits are the golden decagon mirror, the Sagnac-loop
PBS and the uncoated down-conversion crystals. Regarding detection effi-
ciency, the avalanche photo-diodes (APDs) used in the setup have a quan-
tum efficiency ranging from 0.5 to 0.6. As a side note, ξ is also inversely
proportional to the spatial walk-off in the non-linear crystal, due to optical
birefringence, which increases for longer crystals. While this is more of
an issue in type-II sources because of the opposite polarisations of signal
and idler photons, it still affects coupling in type-I SPDC, since the spatial
modes exiting the crystals will be elliptical, and the longer the crystal the
higher the ellipticity.
The overall collection efficiency ξi for mode i can be experimentally
estimated from the rate of coincident events Ci,i0 between i and its twin
mode i0 , and the rate of twin single events Si0 , as ξi = Ci,i0 /Si0 . In our setup,
ξ ranged from 0.16 to 0.21, depending on the particular mode considered.
The pair production probability RP is instead determined by:
• SPDC efficiency.
• SPDC crystal length.
• Pump power.
While the non-linear process efficiency is related to the joint spectral am-
plitude (JSA) and will not be discussed here, RP is roughly inversely pro-
portional to the crystal length [15]. A longer crystal increases the down-
conversion probability, though it also increases the distinguishability of in-
dependent pairs, since they might be created at locations far away along
the crystal, farther than the coherence length of the wave-packets set by the
spectral filters. For the above reasons, crystals of 2 mm thickness were cho-
sen for our source1 . Finally, RP is proportional to the pump power, which
could be easily controlled in our setup thanks to a HWP positioned before
the up-conversion crystal. The average power of the up-converted 390 nm
beam in our source ranged from close to zero up to around 860 mW. For
mode i in our setup, RP,i can be estimated as:
Si Si0 Si
RP,i = = . (6.7)
frep Ci,i0 frep ξi
In Fig. 6.8, the experimental dependence of RP as function of pump power
is reported, for mode 1.
1 The choice of optimal crystal type and thickness was carried out before the
author joined in the work.
61
0.10
0.08
0.06
RP
0.04
0.02
Figure 6.8: The experimental pair production probability for mode 1, as de-
termined from Eq. (6.7), is plotted against pump power. The relation is linear
for small RP , as shown by the linear fit.
500
4-photon coincidences [Hz]
2500 1500
400
Singles [kHz]
2000
300 1000
1500
200 500
1000
500 100
0
250 500 750 250 500 750 250 500 750
Power [mW] Power [mW] Power [mW]
Most crucially, using Eq. (6.6), we can predict rates of N-photon coin-
cidences as function of pump power and collection efficiency ξ . A simple
way to visualise this dual-variable relation is by removing the dependence
on ξ (since this is not an adjustable experimental parameter anyway). This
simplification can be carried out by plotting estimated rates as function of
62
measured 2-photon coincidences, which also depend on ξ . Such a plot is
reported in Fig. 6.10 for 4-, 6- and 8-photon rates.
101
100
Event rate
63
In this section, we will go through the most relevant of such imper-
fections, and describe some of the methods we followed to mitigate their
impact. In particular, we consider:
64
Higher-order SPDC events
Si S j Si S j
aci j = · frep = . (6.8)
frep frep frep
This method can be used for states of any number N of qubits, by substitut-
ing Si with the sum of all possible N − 1 coincidences. From the equation
above, we can predict the effect that accidental coincidences will have on
the visibility of the two qubit |φ + i state in σz basis. If we label Alice’s
(Bob’s) H,V outcomes as 1, 2 (3, 4), we can write the visibility explicitly
in terms of accidental coincidences, as:
To simplify, we can assume the state obtained if all higher-order events were
neglected is the ideal |φ + i Bell state (as we will see in the following, this
assumption is fairly close to reality). In this case, C13 = C24 = C, C14 =
C23 = 0, S1 = S2 = S3 = S4 = S, and ac13 = ac14 = ac23 = ac24 = ac, thus
we can rewrite:
2C 1
Vσz = = . (6.10)
2C + 4ac 1 + 2S/ξ frep
65
Since we know the relation between the rate of singles and pump power in
our setup from the linear fit in Fig. 6.9, we can plot Vσz as function of pump
power and ξ . The resulting graph is reported in Fig. 6.11.
66
to be absolutely identical, or indistinguishable, in order for the interference
process to give rise to maximally entangled states between different pairs.
In addition to the measures described above, some further conditions have
to be met for HOM interference to be optimal. First of all, the two photons
coming from different pairs have to be spatially and temporally overlapped.
Spatial overlap is achieved by optimising collection efficiency at the fibre
couplers for one photons, and then using the mirrors along the path of the
other photon to optimise collection efficiency of the latter (without moving
the couplers!). As a result, all photons entering the SMFs and reaching the
detectors will have followed the same path from the interference PBS on-
wards. Temporal overlap depends instead on relative path length, and we
adjusted it by adding two retro-reflectors on each path, one of which was
mounted on a micro-metric translation stage, longitudinal to photon propa-
gation. In addition, the spectra of independent photons also need to overlap
perfectly. This is achieved with the help of the narrow-band spectral filters
located before each measurement station (see Fig. 6.7).
Most crucially though, the interference process will result in a maxi-
mally entangled state only in case Fock states of single photons enter the
PBS inputs. Since the down-conversion process actually produces sub-
Poisson distributed squeezed states, the interference quality is disturbed by
the higher-order down-conversion events. In order to quantify this effect,
we first have to define a “quality” parameter: the HOM visibility. Given two
down-converted pairs of photons, of defined, identical polarisations, we can
send the signal photons into the same input of a PBS. If the two photons are
indistinguishable, they will always come out bunched together at one out-
put. This is precisely the HOM effect [66]. The four-fold coincidence rate
between the two trigger photons and the two detectors at the outputs of the
PBS should therefore be zero. Every time the two photons are instead distin-
guishable, they will independently pick an output, and there will be a 50/50
chance that they end up in different detectors, thus giving rise to a four-fold
coincident event. A relatively simple way to make the photons completely
distinguishable is by delaying them by an amount of time greater than their
wave-packet coherence. This is possible in our setup thanks to the retro-
reflectors on translation stages. We can then define the HOM visibility as
the difference between the rate of four-fold coincidences when the photons
are out of coherence and when they are perfectly overlapped, divided by the
former. An experimental example for the interference of modes 1 and 2 is
shown in Fig. 6.12.
In contrast with the case of Bell state visibility, there is no simple way
to predict the HOM visibility as function of pump power (as in Fig. 6.11).
Nevertheless, we can of course measure it, as shown in Fig. 6.13. From
67
100
80
40
20
0
-600 -400 -200 0 200 400 600
Relative delay [µm]
68
0.90
0.85
HOM visibility
0.80
0.75
0.70
Figure 6.13: HOM interference visibility for modes 1 and 2 at several dif-
ferent pump powers. The exponential fit (of p-value 1.0) returned an upper
bound for the visibility of approximately 91%.
since the source has in principle no preferential direction along the down-
conversion ring, and neither has the rest of the setup, the other sources can
perform very similarly to γ52 , as we will see.
Source γ52
Since Bell states form a basis for 2-qubit states, the ability to prepare perfect
Bell states effectively means that any pure 2-qubit state can be produced.
In the source, a simple lambda-half wave plate (HWP) and a phase-plate
(PP) in one of the two modes are enough to convert among any of the Bell
states. Considering this, we focused our characterisation on the fidelity√of
our experimental state to the maximally entangled |φ + i = (|00i + |11i)/ 2
state. In order to characterise it, we performed a full state tomography with
the maximum likelihood method [67]. In addition, we employed motorised
wave-plate rotation stages so that we could repeat the measurements several
times. Each setting was measured for 30 seconds, and the nine tomography
settings were repeated 35 times. The results are reported in Fig. 6.14.
The prepared state showed a median fidelity to |φ + i of (99.7 ± 0.1)%,
which is at least as good as some of the best results found in literature
[34; 64]. The uncertainty is the standard deviation of the 35 measurement
repetitions. While these results were corrected for accidental coincidences,
the measurements were performed at extremely low pump power (of the
order of ∼ 1 mW), so that the correction does not affect the final result on
state fidelity, up to the uncertainty.
69
0.5 0.5
The visibility in H/V (σz ) and +/− (σx ) bases for the remaining four
sources of photon pairs, together with the discussed results for source γ52
as a comparison, is reported in Tab. 6.2.
Table 6.2: Experimental visibility in σz and σx bases for the 2-photon sources
in the setup. The different uncertainties are mainly due to different measure-
ment duration.
70
the estimation of fidelity witness D introduced in Section 1.6.1 for source
γ14 .
Source γ24
Figure 6.15: Full quantum state tomography of the |GHZ 4 i states produced
by source γ24 . The real (left) and imaginary (right) parts of the density matrix
are plotted.
Since the fidelity is higher than 50%, we can confirm that the source
produces genuinely entangled |GHZ 4 i states.
Source γ14
71
surements in case of 4 qubits. The measurement operators are:
e ⊗4 = |Hi hH|⊗4 + |V i hV |⊗4 ,
Z
⊗4
⊗4 0 0
M0 = cos π σx + sin π σy = σx⊗4 ,
4 4
⊗4
⊗4 1 1 σx + σy ⊗4
M1 = cos π σx + sin π σy = √ ,
4 4 2 (6.11)
⊗4
⊗4 2 2
M2 = cos π σx + sin π σy = σy⊗4 ,
4 4
⊗4
⊗4 3 3 −σx + σy ⊗4
M3 = cos π σx + sin π σy = √ .
4 4 2
Measuring each setting for 600 seconds at a rate of approximately 15 4-
photon coincidences per second, we estimated the values of the operators
above as in Fig. 6.16. Through Eq. (1.25)), we derived a fidelity of our
state with |GHZ 4 i of 90.1 +0.7 +0.1
−1.1 −0.2 %, which again includes a correction
for accidental events.
1
Experimental result
0.5
-0.5
-1
Ze⊗4 M0⊗4 M1⊗4 M2⊗4 M3⊗4
The result overlaps with the corresponding one reported for source γ24 ,
within one standard deviation. As was the case for γ24 , uncertainties have
been estimated through Monte Carlo simulations and correspond, respec-
tively, to poissonian and wave-plate angle errors.
72
modes 1 and 2 further down the setup, at the following PBS. This process
enables the generation of |GHZ 6 i quantum states. In the case of six qubits,
a full state tomography would require 36 = 729 measurements. To obtain an
uncertainty of the order of 1% in the fidelity, each measurement would need
to last on the order of an hour, which clearly makes the full state tomography
approach unfeasible. Once again then, we resort to operator D and Eq.
(1.25) to estimate the state fidelity. This approach requires measuring the
following seven operators:
e ⊗6 = |Hi hH|⊗6 + |V i hV |⊗6 ,
Z
√ !⊗6
3σx + σy
M0⊗6 = σx⊗6 , M1⊗6 = ,
2
√ !⊗6
σx + 3σy (6.12)
M2⊗6 = , M3⊗6 = σy⊗6 ,
2
√ !⊗6 √ !⊗6
−σx + 3σy − 3σx + σy
M4⊗6 = , M5⊗6 = .
2 2
We have measured each setting for two hours at a rate of approximately
1.2 6-photon coincidences per√minute. The resulting fidelity with state
(|HHHHHHi + |VVVVVV i)/ 2, corrected for accidental events, was es-
timated at 76.1 +1.7 +0.2
−1.8 −0.2 %, and results of all seven operators are reported
in Fig. 6.17.
1
Experimental result
0.5
-0.5
-1
Ze⊗6 M0⊗6 M1⊗6 M2⊗6 M3⊗6 M4⊗6 M5⊗6
73
6.5 Comparison with literature
In the interest of comparing our multi-photon entanglement source with
other works in literature, I will consider the already mentioned references
[12; 14; 15], in addition to a previous source realised in our research group
and reported in references [68; 69]. While all six experiments aimed at gen-
erating entangled states, reference [68] did not target the maximally entan-
gled GHZ state, but a different rotationally invariant, non-maximally entan-
gled state. Moreover, rather than working with first-order SPDC, the setup
took advantage of third-order events to generate six entangled photons. Tab.
6.3 reports a streamlined comparison of the different setups.
First of all, it should be noted that our design employs only two non-
linear crystals, which is a fraction of those used by the other setups for
|GHZi state generation [12; 14; 15]. This is an enormous experimental ad-
vantage when initially setting up the experiment. It also makes the setup
more compact and arguably more stable in time. Moving on along Tab.
6.3, our source shows a remarkably lower overall collection efficiency com-
74
pared with references [14; 15]. The extremely high efficiency reported in
reference [14] is very likely due to the SPDC crystals emitting Gaussian-
distributed, collinear beams, which have much better mode-matching into
SMFs. According to the authors of reference [15] instead, they were able
to dramatically improve previous results thanks to the reduced walk-off of
BiBO crystals compared to BBO, and by using shorter crystals. In terms
of rate of pair production however, our source is roughly on par with, or
above, the others reported. Both brightness and collection efficiency were
improved compared to previous efforts in our lab (in reference [68]). In-
distinguishability of independent photons, as measured through HOM in-
terference, turned out to be as high as the best of the works considered in
the comparison.
With regards to fidelity of state preparation, comparing different results
is a delicate matter, since higher fidelity may be achieved by decreasing
pump power, at the cost however of an exponentially decreased generation
rate. With that in mind, the currently available results referring to |GHZ 4 i
and |GHZ 6 i show that fidelities higher than reported in other works can be
achieved in our setup, although at lower generation rates. More data needs
to be gathered in order to compare bigger states and fidelity at different
rates.
75
Figure 6.18: 8-photon rates as function of both pump power and collection
efficiency ξ , as estimated from Eq. (6.6). The dependence on RP is substituted
with that on pump power using the linear fit of experimental data in Fig. 6.8.
76
7. Summary and outlook
In this thesis, I have outlined the most important results obtained in Papers
I-III. These works underline two crucial aspects of the current landscape
in quantum information: the need for reliable and reproducible results for
applications, and the possibility of a new level of security and certification
thanks to physical constraints and minimal assumptions.
In discussing Paper I, we have seen how ever-present Bell tests can
run into troubles when experimenters assume fair-sampling. The effects of
this issue, that we named apparent signalling, include distorting results and
invalidating conclusions. We have therefore experimentally investigated the
causes behind apparent signalling, and proposed effective ways of testing
and solving them.
In Papers II and III, we have explored the recent notions of device-
independence and self-testing. These outstanding concepts aim at drawing
conclusions from experimental data with minimal assumptions on the de-
vices used to collect it. Because they rely on trusting the laws of quantum
mechanics, they are arguably the next step in quantum information’s path
to maturity. That is, after coming to terms with the rejection of classical
physics by means of Bell’s theorem. In the two papers, we employed the
techniques of device-independence and self-testing to successfully certify
non-projective and informationally complete generalised quantum measure-
ments. As we have seen in the introduction to Chapter 5, these measure-
ments are optimal in several different scenarios, and in higher dimension
are still the object of current fundamental research. In order to certify the
various properties of our experimental measures, we have taken two quite
different approaches, one relying on entangled qubits, and the other on her-
alded single qubits. These two options have different advantages and disad-
vantages which were discussed in Chapter 5.
Throughout the refinement of existing experiments and the develop-
ment of new ideas, photonic entanglement has remained an omnipresent
resource in quantum information protocols. The need for better and more
complex multipartite entangled states, for both fundamental studies and
applications as quantum computation, has pushed us towards the devel-
opment of a source of multi-photon entanglement based on spontaneous-
parametric down-conversion. The original design solutions we employed
77
were described in Chapter 6, together with thorough experimental charac-
terisation, and they will be at the core of the results in Paper IV, which
is still in the works at the time of writing. An extensive comparison with
other state-of-the-art multi-photon sources found in literature indicates our
source as a valid alternative, while being experimentally far less complex.
Finally, while more work is still needed to fully characterise the entangle-
ment source, potential improvements have already been identified and will
be investigated in the near future.
78
References
[2] M. A. N IELSEN AND I. L. C HUANG. Quantum Computation and Quantum Information, 53.
2000. 1
[3] X.-L. WANG , Y.-H. L UO , H.-L. H UANG , M.-C. C HEN , Z.-E. S U , C. L IU , C. C HEN , W. L I ,
Y.-Q. FANG , X. J IANG , J. Z HANG , L. L I , N.-L. L IU , C.-Y. L U , AND J.-W. PAN. 18-Qubit
Entanglement with Six Photons’ Three Degrees of Freedom. Phys. Rev. Lett., 120:260502,
Jun 2018. 4
[5] A. J. S COTT. SICs: Extending the list of solutions. arXiv:1703.03993, (2):1–15, Mar 2017. 8
[6] M. A PPLEBY, T.-Y. C HIEN , S. F LAMMIA , AND S. WALDRON. Constructing exact symmetric
informationally complete measurements from numerical solutions. Journal of Physics A:
Mathematical and Theoretical, 51(16):165302, Apr 2018. 8
[7] R. J OZSA. Fidelity for Mixed Quantum States. Journal of Modern Optics, 41(12):2315–2323,
Dec 1994. 10
[8] A. P ERES. Separability Criterion for Density Matrices. Phys. Rev. Lett., 77:1413–1415, Aug
1996. 11
[10] B. M. T ERHAL. Bell inequalities and the separability criterion. Physics Letters, Section A:
General, Atomic and Solid State Physics, 271(5-6):319–326, 2000. 11
[11] O. G ÜHNE , C.-Y. L U , W.-B. G AO , AND J.-W. PAN. Toolbox for entanglement detection and
fidelity estimation. Physical Review A, 76(3):030305, Jun 2007. 11
[12] Y.-F. H UANG , B.-H. L IU , L. P ENG , Y.-H. L I , L. L I , C.-F. L I , AND G.-C. G UO. Experimental
generation of an eight-photon Greenberger–Horne–Zeilinger state. Nature Communications,
2(1):546, Sep 2011. 12, 51, 56, 58, 74
[13] X.-C. YAO , T.-X. WANG , P. X U , H. L U , G.-S. PAN , X.-H. BAO , C.-Z. P ENG , C.-Y. L U ,
Y.-A. C HEN , AND J.-W. PAN. Observation of eight-photon entanglement. Nature Photonics,
6(4):225–228, Apr 2012. 51
79
[14] X.-L. WANG , L.-K. C HEN , W. L I , H.-L. H UANG , C. L IU , C. C HEN , Y.-H. L UO , Z.-E. S U ,
D. W U , Z.-D. L I , H. L U , Y. H U , X. J IANG , C.-Z. P ENG , L. L I , N.-L. L IU , Y.-A. C HEN , C.-
Y. L U , AND J.-W. PAN. Experimental Ten-Photon Entanglement. Physical Review Letters,
117(21):210502, Nov 2016. 51, 68, 74, 75
[15] L.-K. C HEN , Z.-D. L I , X.-C. YAO , M. H UANG , W. L I , H. L U , X. Y UAN , Y.-B. Z HANG ,
X. J IANG , C.-Z. P ENG , L. L I , N.-L. L IU , X. M A , C.-Y. L U , Y.-A. C HEN , AND J.-W. PAN.
Observation of ten-photon entanglement using thin BiB_3O_6 crystals. Optica, 4(1):77, Jan
2017. 12, 51, 56, 58, 61, 74, 75, 76
[16] J. S. B ELL. On the Einstein Podolsky Rosen paradox. Physics Physique Fizika, 1(3):195–200,
1964. 12, 14
[17] A. K. E KERT. Quantum cryptography based on Bell’s theorem. Physical Review Letters,
67(6):661–663, 1991. 12, 18, 31, 42
[20] J. S. B ELL. Speakable and Unspeakable in Quantum Mechanics, 1989. 15, 16, 17
[23] A. A SPECT, J. DALIBARD , AND G. ROGER. Experimental test of Bell’s inequalities using
time- varying analyzers. Physical Review Letters, 49(25):1804–1807, 1982. 17
80
[29] M. G IUSTINA , A. M ECH , S. R AMELOW, B. W ITTMANN , J. KOFLER , J. B EYER , A. L ITA ,
B. C ALKINS , T. G ERRITS , S. W. NAM , R. U RSIN , AND A. Z EILINGER. Bell violation using
entangled photons without the fair-sampling assumption. Nature, 497(7448):227–230, 2013.
17
[33] Č. B RUKNER , M. Z UKOWSKI , J. W. PAN , AND A. Z EILINGER. Bell’s inequalities and quan-
tum communication complexity. Physical Review Letters, 92(12):1–4, 2004. 18
[36] O. G ÜHNE AND G. T ÓTH. Entanglement detection. Physics Reports, 474(1-6):1–75, 2009. 18
[37] C. G ERRY AND P. K NIGHT. Introductory Quantum Optics. Cambridge University Press, 2005.
24
[39] C. A. F UCHS. Quantum Mechanics as Quantum Information (and only a little more). pages
1–59, May 2002. 41
[40] J. Ř EHÁČEK , B.-G. E NGLERT, AND D. K ASZLIKOWSKI. Minimal qubit tomography. Physi-
cal Review A, 70(5):052321, Nov 2004. 41
[41] C. A. F UCHS AND M. S ASAKI. Squeezing Quantum Information through a Classical Chan-
nel: Measuring the “Quantumness” of a Set of Quantum States. Quantum Info. Comput.,
3(5):377–404, Sep 2003. 41
[42] A. ACÍN , S. P IRONIO , T. V ÉRTESI , AND P. W ITTEK. Optimal randomness certification from
one entangled bit. Physical Review A, 93(4):040102, Apr 2016. 41, 45
[44] T. D URT, C. K URTSIEFER , A. L AMAS -L INARES , AND A. L ING. Wigner tomography of two-
qubit states and quantum cryptography. Physical Review A, 78(4):042338, Oct 2008. 41
81
[45] Z. E. D. M EDENDORP, F. A. T ORRES -RUIZ , L. K. S HALM , G. N. M. TABIA , C. A. F UCHS ,
AND A. M. S TEINBERG . Experimental characterization of qutrits using symmetric informa-
tionally complete positive operator-valued measurements. Physical Review A, 83(5):051801,
May 2011.
[48] D. M AYERS AND A. YAO. Quantum cryptography with imperfect apparatus. In Proceedings
39th Annual Symposium on Foundations of Computer Science (Cat. No.98CB36280), pages 503–
509, Nov 1998. 42
[53] M. M CKAGUE , T. H. YANG , AND V. S CARANI. Robust self-testing of the singlet. Journal of
Physics A: Mathematical and Theoretical, 45(45), 2012. 42
82
[61] P. WALTHER , K. J. R ESCH , T. RUDOLPH , E. S CHENCK , H. W EINFURTER , V. V EDRAL ,
M. A SPELMEYER , AND A. Z EILINGER. Experimental one-way quantum computing. Na-
ture, 434(7030):169–176, Mar 2005. 51
[65] A. P RADANA AND L. Y. C HEW. Quantum interference of multi-photon at beam splitter with
application in measurement-device-independent quantum key distribution. New Journal of
Physics, 21(5):053027, May 2019. 66
[68] M. R ÅDMARK. Photonic quantum information and experimental tests of foundations of quantum
mechanics. PhD thesis, Stockholm University, Department of Physics, 2010. 74, 75
83