Thesis Reference
Thesis
On the device-independent approach to quantum physics : advances in quantum nonlocality and multipartite entanglement detection
BANCAL, Jean-Daniel
Abstract
La physique quantique a participé au développement de nombreux domaines, qu'il s'agisse de l'informatique en permettant de traiter l'information électroniquement à l'aide de transistors, des communications, rendues possibles à grande échelle par la lumière laser guidée dans des fibre optique, ou bien de la médecine par les méthodes d'imagerie par résonnance magnétique nucléaire. Qui eut cru que l'hypothèse quantique formulée par Max Planck à l'aube du 20ème siècle aurait, de fil en aiguille, de telles répercussions? Malgré cela, la physique quantique reste encore passablement mystérieuse. L'un de ses aspects les plus intriguants étant sans doute son charactère nonlocal, c'est-à-dire sa capacité à violer des inégalités de Bell à l'aide de systèmes isolés les uns des autres.
Reference
BANCAL, Jean-Daniel. On the device-independent approach to quantum physics : advances in quantum nonlocality and multipartite entanglement detection. Thèse de doctorat : Univ. Genève, 2012, no. Sc. 4419
URN : urn:nbn:ch:unige-217102 DOI : 10.13097/archive-ouverte/unige:21710
Available at: http://archive-ouverte.unige.ch/unige:21710
Disclaimer: layout of this document may differ from the published version.
1 / 1 UNIVERSITE´ DE GENEVE` FACULTE´ DES SCIENCES Groupe de Physique Appliqu´ee - Optique Prof. N. Gisin
ON THE DEVICE-INDEPENDENT APPROACH TO QUANTUM PHYSICS
ADVANCES IN QUANTUM NONLOCALITY AND MULTIPARTITE ENTANGLEMENT DETECTION
THESE`
pr´esent´ee`ala Facult´edes Sciences de l’Universit´ede Gen`eve pour obtenir le grade de Docteur `esSciences, mention Physique
par
Jean-Daniel Bancal de Meyrin (GE)
Th`eseN◦ 4419
GENEVE` Atelier d’impression ReproMail 2012
R´esum´e
La physique quantique a particip´eau d´eveloppement de nombreux domaines, qu’il s’agisse de l’informatique en permettant de traiter l’information ´electroniquement `al’aide de transistors, des communications, rendues possibles `agrande ´echelle par la lumi`ere laser guid´eedans des fibre optique, ou bien de la m´edecine par les m´eth- odes d’imagerie par r´esonnance magn´etique nucl´eaire. Qui eut cru que l’hypoth`ese quantique formul´eepar Max Planck `al’aube du 20`eme si`ecle aurait, de fil en aiguille, de telles r´epercussions?
Malgr´ecela, la physique quantique reste encore passablement myst´erieuse. L’un de ses aspects les plus intriguants ´etant sans doute son charact`ere nonlocal, c’est- `a-dire sa capacit´e`avioler des in´egalit´esde Bell `al’aide de syst`emes isol´esles uns des autres. Une telle violation sugg`ere en effet que ces syst`emes sont causalement reli´es,ce qui semble contredire le fait qu’ils soient mutuellement s´epar´es.
La mani`ere directe avec laquelle la nonlocalit´equantique apparaˆıt dans des r´esultats exp´erimentaux lui permet d’ˆetre test´eeen faisant appel `aun minimum d’hypoth`eses. En particulier, aucune erreur de calibration sur des appareils de mesure individuels ne peut remettre en cause le r´esultat d’une telle exp´erience. Cette robustesse face aux erreurs d’impl´ementation, qui sont inh´erentes `atoute manipulation exp´erimentale, ouvre la voie vers de nouvelles approches exp´erimen- tales. En effet elle montre qu’il est possible de r´epondre `acertaines questions en faisant appel `avirtuellement aucune hypoth`ese du moment que les syst`emes mesur´es sont suffisamment s´epar´esles uns des autres.
A quelles questions peut-on r´epondre de cette fa¸con-l`a? A quoi peut servir la violation d’une in´egalit´ede Bell en g´en´eral? Mais aussi, comment la nature s’y prend-elle pour violer une in´egalit´ede Bell? Et quelles sont les limites de la nonlocalit´equantique? Voici quelques-unes des questions abord´ees par cette th`ese.
iii
Contents
Introduction 1
1 Bell tests in bipartite scenarios 5 1.1 No-signalling and local causality ...... 5 1.1.1 Local correlations ...... 5 1.1.2 No-signalling correlations ...... 7 1.1.3 Geometrical representation ...... 7 1.1.4 Experimental loopholes ...... 8 1.2 Bell test between an atom and an optical mode ...... 9 1.2.1 Creating atom-photon entanglement ...... 9 1.2.2 CHSH violation ...... 10 1.2.3 Space-like separation ...... 11 1.2.4 Conclusion ...... 11 1.3 Bell test with multiple pairs ...... 12 1.3.1 Two sources ...... 12 1.3.2 Noise model ...... 13 1.3.3 Bell violation ...... 14 1.4 Experimental violation of Bell inequalities with a commercial source of entanglement ...... 14 1.4.1 Experimental setup ...... 14 1.4.2 Test of several Bell inequalities ...... 15 1.4.3 Chained Bell inequality ...... 15 1.4.4 Conclusion ...... 17
2 Nonlocality with three and more parties 19 2.1 Defining genuine multipartite nonlocality ...... 19 2.2 Multipartite Bell-like inequalities ...... 21 2.2.1 A general structure for (n, m, k) scenarios ...... 21 2.2.2 Recursion relation ...... 22 2.3 Nonlocality from local marginals ...... 24 2.3.1 An inequality ...... 25 2.3.2 Conclusion ...... 25 2.4 Tripartite nonlocal boxes ...... 26 2.4.1 The tripartite nosignalling polytope ...... 26 2.4.2 Conclusion ...... 26 2.5 A tight limit on quantum nonlocality ...... 26 2.5.1 Can you guess your neighbour’s input (GYNI)? ...... 26 2.5.2 Outlook ...... 27 2.6 Simulating projective measurements on the GHZ state ...... 28
v CONTENTS
2.6.1 Nonlocal resources ...... 28 2.6.2 Simulation ...... 28 2.6.3 Conclusion ...... 29
3 Device-independent entanglement detection 31 3.1 Imperfect measurements ...... 31 3.1.1 Effects of systematic errors on tomography ...... 32 3.1.2 Effects of systematic errors on entanglement witnesses ...... 32 3.2 Witnesses insensitive to systematic errors? ...... 33 3.2.1 Device-independent witnesses for genuine tripartite entanglement . 34 3.2.2 A witness for genuine multipartite entanglement ...... 34 3.3 Experimental demonstration ...... 35 3.3.1 Experimental setup and procedure ...... 35 3.3.2 Addressing errors ...... 36 3.3.3 Experimental results ...... 37 3.4 Conclusion ...... 37
4 Quantum information put into practice 39 4.1 Memoryless attack on the 6-state QKD protocol ...... 39 4.1.1 The 6-state protocol ...... 39 4.1.2 Secret key rate ...... 40 4.1.3 Discussion ...... 41 4.2 Private database queries ...... 42 4.2.1 Sketch of the protocol ...... 42 4.2.2 Discussion ...... 43
5 Finite-speed hidden influences 45 5.1 Finite-speed propagation and v-causal theories ...... 45 5.1.1 v-causal models and experimental limitations ...... 46 5.1.2 Influences without communication? ...... 47 5.2 The hidden influence polytope ...... 47 5.2.1 Quantum violation and faster-than-light communication ...... 49 5.3 Experimental perspectives ...... 50 5.4 Conclusion ...... 51
Conclusion and outlook 53
Acknowledgements 55
Bibliography 57
A Polytopes 63 A.1 Definition and terminology ...... 63 A.2 Some operations on polytopes ...... 64 A.2.1 Projection ...... 64 A.2.2 Slice ...... 66 A.2.3 Another tasks : finding facets lying under an inequality ...... 66
B Memoryless attack on the 6-state protocol – proof 67
Papers 71
vi Introduction
From its beginning in the 1920’s quantum physics has challenged our understanding of the world. Particles that could be conceived previously as points turned out to be provided with a wave evolving in time according to a law of motion. This conceptual change allowed for previously unsuspected phenomenons to be observed, like for instance the interference of a molecule with itself demonstrated several times experimentally (e.g. with C60 molecules in [1]). If the quantum theory is recognized for its extraordinary predictive power, the picture of the world that it suggests is not the subject of a common agreement. For instance, the question of whether the wavefunction ψ , a fundamental ingredient of the theory, should | i be understood as a proper physical object, i.e. a physical property of every quantum system, or rather as a tool from the theory which is only useful to predict the evolution of physically relevant objects, is still an active subject of research [2, 3, 4, 5]. One could argue that questions about the possible interpretation of the elements of the quantum theory are of secondary importance, provided that predictions match exper- imental results. But that would be putting aside the possibility for such considerations to reveal fundamental properties of nature. For instance, the quantum measurement pro- cess is commonly understood as an instantaneous change of the wavefunction throughout all space. If this process is indeed instantaneous, and if the wavefunction is a physical object, then measurement of a quantum system is a strongly nonlocal phenomenon, and one should expect physical quantities to be the subject of such instantaneous change at a distance. On the other hand, if the wavefunction can be understood as a tool of the the- ory, without a concrete physical counterpart, then the nonlocal character of this process might just be an artifact of the theory, without direct incidence on physically relevant quantities. Since a proper understanding of the elements of the quantum theory seems difficult to reach without invoking arguable choices of additional assumptions, and since only properties of nature that can have a measurable impact are worth discussing anyway, we ask what properties of quantum physics can be detected directly from experimental data, without relying on more assumptions than the ones needed in order to make sense out of these data. In this way, we hope to be able to explore properties of quantum physics more directly. Moreover, we can expect to be able to check these properties on nature directly, because we follow an approach which fundamentally relies on experimental results. Following the seminal work of John Bell [6], we consider experimental setups charac- terized by a number n of identifiable systems, which can be measured in m possible ways, yielding each time one out of k possible values. The results of such a Bell-type experiment can be characterized by conditional probability distribution of the form P (ab xy) (here for | a scenario with n = 2 parties), which we refer for short as correlations. These correlations describe how often the results a and b are observed on two separated systems whenever measurement x and y are performed on them, respectively.
1 Introduction
An important property of correlations is that they are always accessible in principle: by sufficiently separating the systems under study, and performing enough measurements, the raw data produced during an experiment allows one to evaluate P (ab xy) directly. | Namely, if the measurements x and y are performed by the two parties Alice and Bob N times, leading to N(ab xy) N observations of the outcomes a and b, then xy | ≤ xy N(ab xy) P (ab xy) = lim | . (0.0.1) | Nxy N →∞ xy Statistical analysis can be used to infer the value of P (ab xy) with high probability when | the number of measurements performed is finite (N < ). xy ∞ Moreover the evaluation of the correlations P (ab xy) requires no knowledge about the | process creating the experimental results. It thus does not rely on any interpretation of the elements of a theory susceptible of describing these processes. Rather, all which is needed in order to make sense out of correlations is well-defined systems and indices to identify the inputs and outputs of the experiment in a reliable fashion. Since this description of an experiment requires no precise description of the working of the measurement devices we refer to it as device-independent. This makes correlations well adapted for our purpose. They are thus the central object of interest in this thesis. Note that apart from allowing to study nature with a minimum number of assumption, the device-independent hypotheses are also naturally adapted to the study of problems involving untrusted devices, such as quantum key distribution (QKD) [7], or to derive conclusions that are particularly robust with respect to practical uncertainties. While we present a possible application of the second kind in chapter 3, a significant part of this thesis is devoted to the study of correlations in multipartite scenarios.
Outline The content of this thesis is organized as follows. First, we recall Bell’s notion of local causality. This is useful for the rest of the thesis since most of it relates in some way to Bell inequalities. This leads us to discuss several studies on nonlocality in bipartite scenarios, including a proposal for a loophole-free Bell experiment combining measurement on an atom and a photon, and the analysis of Bell tests in presence of multipairs. We conclude this section with an experimental demon- stration of nonlocal correlations conducted with a commercially-available entanglement source. The second part of this thesis discusses the notion of nonlocality in scenarios involving three or more parties. We discuss the definition of genuine multipartite nonlocality, and present a family of inequalities that can detect multipartite nonlocal correlations. In this section we also study more specifically the structure of multipartite correlations by analysing the set of tripartite no-signalling correlations and questioning the constraints that relate different marginals of a single multipartite system. Finally, we provide a bound on the nonlocality of quantum correlations, and a model that simulates measurements on a GHZ state with the help of bipartite nonlocal boxes. The next section is devoted to the detection of genuine multipartite entanglement in a device-independent manner. We examine in which case genuine multipartite entangle- ment can be witnessed based solely on the observation of some correlations. This allows one to witness multipartite entanglement, a property of quantum physics, in a way that is particularly resistant to practical imperfections. These results are illustrated experi- mentally.
2 The fourth section reminds us that quantum physics allows one to perform some tasks which would be impossible or harder otherwise. It contains the analysis of a specific attack on the 6-state QKD protocol, as well as a proposal for practical secure database queries. Finally, we close this thesis by considering the possibility of relaxing Bell’s condition of local causality to recover a causal explanation of quantum nonlocal correlations with a sense of proportion. We show that this is not possible without allowing for faster-than- light communication. Apart from the beginning of the first chapter, which contains several definitions used in the rest of the text, the different parts of this thesis can be read independently of each other. While the main text is meant to be concise, all complementary information should be found in the appendices and attached papers.
3 Introduction
4 Chapter 1
Bell tests in bipartite scenarios
1.1 No-signalling and local causality
From our everyday experience, we know that any transmission of information (i.e. com- munication) must be carried by a physical support: in order to let someone know about something we can say it to him, write him an SMS or a letter about it, etc, i.e. use either acoustic waves, electrons, or paper to carry this information to our friend. This idea can be expressed in the following principle: No-signalling principle. Any transmission of information must be carried by a physical support leaving the emitter after the message is chosen. This principle is satisfied by several if not all physical theories, including classical and quantum physics [8, 9]. In fact, the no-signalling principle is tightly related to quantum physics, since it can be seen restricts both the possibility of cloning quantum systems and the possibility of discriminating between quantum states, two peculiarities of quantum physics [10, 11]. In practice, many physical supports are available for communication in nature, like the ones mentioned above, but since the advent of special relativity, it is generally admitted that none of them can carry information faster than light in vacuum. The existence of an upperbound on the speed of all communication led Bell to enunciate the principle of local causality: “The direct causes (and effects) of events are near by, and even the indirect causes (and effects) are no further away than permitted by the velocity of light.” J. S. Bell [12] m In other words, the speed of light c = 29907920458 s , taken as an upper bound on any communication speed, naturally defines the limit between events in space-time which can have a direct causal relation with each other, and events which cannot. More precisely, for every event K the set of events that can be influenced by a decision taken at K is defined by its future light-cone. Similarly, all events that can influence a process taking place at K are contained in its past light-cone (c.f. Figure 1.1a).
1.1.1 Local correlations It is of particular interest to ask which correlations P (ab xy) can be observed, according | to the principle of local causality, in the situation of Figure 1.1b in which two measure- ments are performed in a space-like manner. Indeed, in this case, no information can be
5 Bell tests in bipartite scenarios
a) time future light-cone b) time
ab
K AB
Λ1 Λ2 Λ3 past light-cone space space
Figure 1.1: a) Space-time diagram showing the regions containing events that can influ- ence, or that can be influenced by K according to Bell’s principle of local causality. b) Bipartite Bell experiment in which events creating the outcomes a and b are space-like separated. Apart from the inputs x and y which are chosen at A and B, the outcome a can depend only on the regions Λ1 and Λ2, and b only on Λ2 and Λ3.
exchanged directly between the two measurement events. Letting the (free) choices of measurement setting x and y be made locally at the time of measurement in A and B, we see that the outcome a is produced by Alice’s measurement device before any information about Bob’s choice of measurement y can reach it. Thus, a cannot depend on y. Similarly the production of b by Bob’s measurement device cannot depend on x. Still, the measurement processes at A and B can depend on more variables than only x and y. In particular, a is allowed to depend on the whole content of its past light-cone, including Λ2, a region of space-time which can also influence the process creating b, and Λ1 which cannot influence b. Let us thus denote by λ1, λ2, λ3 all variables which belong to the corresponding regions Λ1, Λ2, Λ3 and which are relevant to make predictions about a and b. The most precise prediction of a that can be given prior to measurement and in agreement with the principle of local causality can then be described by a probability distribution of the form P (a x, λ , λ ). Allowing this distribution to also depend on A | 1 2 λ , resulting in a prediction of the form P (a x, λ , λ , λ ), can only gives a more precise 3 A | 1 2 3 description of a than a locally causal theory1. Similarly, the distribution P (b y, λ , λ , λ ) B | 1 2 3 describes predictions about b at least as well as any locally causal theory. Since the processes happening at A and B cannot influence each other, they are independent. The average bipartite correlations produced in this situation must thus be of the form P (ab xy) = p(λ)P (a x, λ)P (b y, λ) (1.1.1) | A | B | Xλ where λ = (λ , λ , λ ) and p(λ) is a probability distribution, i.e. p(λ) 0, p(λ) = 1. 1 2 3 ≥ λ We refer to this decomposition as the locality condition. Any correlations which can be P decomposed in this way are called local, and conversely any correlations which admit no such decomposition are referred to as being nonlocal. Note that here the regions Λ1, Λ2, Λ3 in Figure 1.1b extend up to immemorial times and depend on the precise space-time positions of A and B. Bell showed that different regions Λ0 with nicer properties can be chosen in order to reach the same decomposition
1 Remember that x and y are only chosen at A and B, in a way that is independent of λ1, λ2 and λ3. Allowing a prediction of a to depend on λ3 thus still doesn’t allow it to depend on y.
6 1.1 No-signalling and local causality
(1.1.1). Namely, any region Λ0 that screens off the regions Λi that we considered here, i.e. that already contains the information from Λ which is relevant to make predictions about a and b [13], is good enough to reach equation (1.1.1).
1.1.2 No-signalling correlations If no decomposition of the form (1.1.1) exists for some correlations P (ab xy), some kind of | influence must have taken place between the two measurement events. Yet, this influence might not be available through the correlations to transmit a message. Indeed, users having only access to the variables a, b, x, y can only encode a message to be carried from A to B by the influences in the choice of their inputs x and y. And this message can only be decoded from the observation of the outcomes a and b. Thus, in order to be able to use some correlations to communicate, the statistics of one party’s outcome must depend on the other party’s choice of measurement. In other words they must violate one of the no-signalling conditions:
P (a xy) = P (ab xy) = P (a x) y | | | ∀ Xb (1.1.2) P (b xy) = P (ab xy) = P (b y) x. | | | ∀ a X Note that these constraints are also sufficient: if P (b xy) = P (b x y) for some x, x , y, b, | 6 | 0 0 then Alice can always send a message to Bob by choosing between x and x0, and re- peating the experiment enough times to allow for Bob to discriminate between these two probabilities. Correlations satisfying the conditions (1.1.2) are called no-signalling. Violation of the no-signalling conditions allows for communication, which is very com- mon in nature. However, violation of these constraints between space-like separated events would allow for faster-than-light communication. Assuming that no physical sup- port can carry information faster than light, this would directly contradict the principle of no-signalling.
1.1.3 Geometrical representation When talking about correlations, it is often useful to represent these probabilities in the vector space obtained by concatenating all components of P (ab xy). Let us briefly describe | a few sets of correlations in this space. For concreteness, we consider here the scenario where a, b, x, y = 0, 1 can only take binary values. Every conditional probability distribution P (ab xy) can then be represented | as the vector
16 ~p = (P (00 00),P (10 00),P (00 10),P (10 10),...,P (11 11)) R (1.1.3) | | | | | ∈ which belongs to a 16-dimensional vector space. Since probabilities satisfy the normal- ization condition P (ab xy) = 1 x, y, the space spanned by the correlation vectors ab | ∀ ~p is in fact only 12-dimensional. Moreover, probabilities are always positive and must P thus satisfy the constraints P (ab xy) 0 a, b, x, y. This restricts the set of vectors ~p | ≥ ∀ that correspond to valid correlations P (ab xy) within this 12-dimensional space. Since the | number of positivity constraints is finite, the set of valid correlation vectors is a polytope (see Appendix A), which we refer to as the positivity polytope. Similarly, no-signalling correlations are normalized and positive. Moreover they sat- isfy the no-signalling conditions (1.1.2). These linear conditions define the no-signalling
7 Bell tests in bipartite scenarios
NS Q L
Figure 1.2: Schematic representation of the set of local (L), quantum (Q), and no- signalling (NS) correlations. Note the inclusion L Q NS. ⊂ ⊂ subspace, which is of dimension 8 here. The set of all no-signalling correlations is thus the slice of the positivity polytope with this subspace. This is again a polytope (see Appendix A), which is usually called the no-signalling polytope. The set of local correlations, as defined by (1.1.1), can also be described by a polytope in the space of correlations. Indeed it is known [20] that any local correlation P (ab xy) | can be decomposed as a convex combination
P (ab xy) = p(µ)P (ab xy), p(µ) 0, p(µ) = 1 (1.1.4) | µ | ≥ µ µ X X of deterministic local strategies P (ab xy) = P (a x, µ)P (b y, µ) 0, 1 . On the other µ | A | B | ∈ { } hand any convex combination of deterministic local strategies is also local. The set of local correlations thus corresponds to the convex hull of the deterministic local strategies. Since the number of such strategies is finite, this set is also a polytope, the local polytope. Whereas any inequality satisfied by the local polytope is a valid Bell inequality, the facets of this polytope are tight Bell inequalities. Finally, it is also useful to characterize the set of quantum correlations. These corre- lations are all the ones which can be obtained by measuring a quantum state ρ with some local measurement operator Ma x and Mb y. They can thus always be written as | |
P (ab xy) = tr(Ma x Mb yρ). (1.1.5) | | ⊗ |
Where ρ 0, trρ = 1, Ma x 0 a Ma x = 11, Mb y 0 b Mb y = 11. ≥ | ≥ | | ≥ | The set of quantum correlations is convex, but admits an infinite number of extremal P P points. It is thus not a polytope. Nevertheless, it can be efficiently characterized by a hierarchy of semi-definite programs [21, 22]. While quantum correlations can violate Bell inequalities, these correlations always satisfy the no-signalling condition (1.1.2). The boundary of this set thus lies between the two preceding sets as represented in Figure 1.2. Note that since quantum correlations can be nonlocal, they can require an exchange of influences between the measurement events. However these influences remains out of reach from us because quantum correlations satisfy the no-signalling condition.
1.1.4 Experimental loopholes Knowing that quantum physics can violate the locality condition (1.1.1) is one thing. Verifying that nature violates it is another, which requires the observation of a faithful
8 1.2 Bell test between an atom and an optical mode
Bell inequality violation. In particular, such an experiment should demonstrate that no locally causal theory is able to reproduce the experimental results. Given the current technological limitations, all Bell inequality violations demonstrated so far suffer from at least one of the following two loopholes, which prevents them from strictly concluding about the nonlocal character of nature. The locality loophole. As discussed previously, space-time separation between the measurement events should be guaranteed in order to prevent any communication between the measurement devices. More precisely, one should make sure that the speed of light prevents Alice’s choice of measurement setting to reach Bob’s device before it produces its outcome. And similarly for Bob’s setting. Given the speed of light, this puts stringent constraints on the admissible duration of the measurement processes, or on the distance that should separate them. The detection loophole. If the measurement devices fail to produce outcomes a or b too often, because the systems to be measured are frequently lost along the way for instance, then there is a possibility that discarding the non-detected events can allow for a local model to reproduce the post-selected correlations [14]. The probability that the measurements produce results, given some inputs x and y, should thus not be too low. With the elements we recalled here, we can now present our contributions.
1.2 Bell test between an atom and an optical mode
If Bell experiments conducted so far have always suffered from one of the loophole de- scribed above, technological advances suggest that both the locality and the detection loophole might soon be closable within the same experiment. In order to make this hap- pen, novel proposals taking into account present capabilities are highly welcome. Here we describe a proposal for a loophole-free Bell test, and analyse its feasibility. Bell tests with photonic systems are well designed to ensure strict space-like separation between the measurement events, thanks to the high speed at which photons can travel. However optical losses are unavoidable, leaving the detection loophole open. On the other hand, atomic systems can provide very high detection efficiency, but don’t travel well enough to allow for a space-time separation between the measurements. To close both loopholes, we consider here an hybrid entangled system consisting of an atom (which can be detected very efficiently) and a photon (which can travel fast, and thus helps to close the locality loophole). We first describe how entanglement between an atom and a photon can be produced, and then discuss the constraints that an experiment would have to satisfy in order to allow an experiment on this system to demonstrate nonlocality.
1.2.1 Creating atom-photon entanglement Let us consider an atom with a lambda-type level configuration (as depicted in Fig. 1.3), initially prepared in the state g . A pump laser pulse with the Rabi frequency Ω can be | i used to partially excite the atom in such a way that it can spontaneously decay into the level s by emitting a photon. Long after the decay time of the atom, the atom-photon | i state is given by ψφ = cos(θ/2) g, 0 + eiφ sin(θ/2) s, 1 (1.2.1) | i | i | i where θ = dtΩ(t) refers to the area of the pump pulse. The phase term is defined by φ = k r k r where k (k ) corresponds to the wave vector of the pump (the p Rp − s s p s 9 Bell tests in bipartite scenarios
e | !
Ω spontaneous photon
g s | ! | !
Figure 1.3: Basic level scheme for the creation of atom-photon entanglement by partial excitation of an atom. The branching ratio is such that when the atom is excited, it decays preferentially in s . | i spontaneous photon) and rp (rs) is the atom position when the pump photon is absorbed (the spontaneous photon is emitted).
1.2.2 CHSH violation
In order to demonstrate nonlocality with the above state, we propose to test the CHSH Bell inequality [15]: S = E + E + E E 2 (1.2.2) 00 01 10 − 11 ≤ where E = p(a = b xy) p(a = b xy) is the correlation between Alice and Bob’s xy | − 6 | outcomes when they respectively perform measurements x and y. Here we consider that Alice can choose measurement bases for her qubit on the whole Bloch sphere. However, since Bob’s qubit lies in the Fock space spanned by 0 and 1 photons, we let him only choose between two kinds of natural measurements : photon counting and homodyne measurements. Since measurements on the atom can be very efficient, we assume that they always produce an outcome. Similarly, homodyne measurements can be very efficient [16] so that Bob’s homodyne measurement is considered perfectly efficient. However, we let his photon counter have a detection efficiency ηd : when a photon arrives on his detector, it thus produces a click with probability ηd. To analyse the impact of the distance between Alice and Bob, we model the channel through which the photon propagates as a lossy channel with transmission ηt:
0, 0 0, 0 | i → | i (1.2.3) 1, 0 1 η 0, 1 + √η 1, 0 | i → − t| i t| i p here the second qubit is a mode of the environment, which is not observed. Tracing out this mode, we get an effective state after the transmission line of
iφ iφ 2 ρ = (cos θ g, 0 +e sin θ√η s, 1 )(cos θ g, 0 +e− sin θ√η s, 1 )+(1 η ) sin θ s, 0 s, 0 . ηt | i t| i h | th | − t | ih | (1.2.4) Considering this state and equation (1.2.2) together, we optimized the free parameters in the state and measurements to get the largest violation for several choices of ηt and ηd. The result is plotted in Figure 1.4.
10 1.2 Bell test between an atom and an optical mode
Figure 1.4: Amount of CHSH violation achievable in an atom-photon Bell experiment. The dashed-dotted line corresponds to the case in which both measurements on the pho- tonic mode are homodyne measurement. The other curves are for one homodyne mea- surement and a photon counting. The lowest permissible transmission here is ηt = 61% and the lowest photo-detection efficiency is ηd = 39%.
1.2.3 Space-like separation From Figure 1.4, we see that the test above provides a certain robustness with respect to losses and detection inefficiency. In order to close both loopholes, these quantities should be compared with the losses expected from an experiment ensuring space-like separation of the measurements. These are ultimately determined by the time needed in order to perform the measurements on the atom or the photon. In our case, we expect the slowest measurement to be the atomic one. Still, the measurement should take about 1 µs [23]. A distance of the order of 300 m would thus be needed to ensure space-like separation. For 800 nm photons, a fiber of this length has a transmission of 93%. The scheme with double homodyne measurements is compatible with these requirements (see Figure 1.4), which attests of the potential feasibility of this experiment2.
1.2.4 Conclusion We showed that a sensible violation of the CHSH inequality can be obtained by combining measurements on an atom with photon counting and homodyne measurements on an optical mode. We also argued that the discussed quantum state could be produced with existing technology. Any practical implementation of the above scheme would involve imperfections. For instance, the branching ratio of the atom may not be perfect, meaning that the excited level e in Figure 1.3 could decay to other levels than s , which we didn’t take into | i | i account here. The movement of the atom during the application of the Ω pulse can have
2Note however that the total transmission efficiency also includes the collection efficiency, i.e. the probability with which the photon emitted by the atom is collected into a fibre. Collection efficiencies of the order of 50% have already been demonstrated using a cavity (see paper [K]).
11 Bell tests in bipartite scenarios
M independentAlice’s Bob’s measurement sources measurement ρ *
A *... B n+ n+
order loss * order loss
ρM nA nB − −
Figure 1.5: Setup of a multipair Bell experiment: here a source produces M independent pairs of entangled particles. Since the pairing between Alice’s and Bob’s particles is lost during their transmission, all particles are measured identically by each party. The total number of particles detected in both outcomes + and - are tallied on both sides. an influence on the phase φ of the produced state as well, and the transmission line of the photon should be stable enough not to loose this phase. All of these aspects can be shown not to threaten directly the main conclusion (see paper [K] for more details). This supports the idea that a Bell experiment closing both the locality and the detection loophole is within technological reach (see also [24, 25] for more proposals along these lines).
1.3 Bell test with multiple pairs
Bell experiments are often realized by measuring one pair of particles at a time. However, there are situations in which the entanglement produced is shared by many particles which cannot be addressed individually. For instance, in [26], many pairs of entangled ultracold atoms have been produced, but they cannot be measured individually. One can thus wonder whether violation of a Bell inequality could in principle be tested in such many-body systems. Here we consider systems consisting of M particles, which cannot be addressed in- dividually. The measurements thus act identically on all the particles that each part receives. For encoding in polarization, this can be modeled by a polarizing beam-splitter (PBS) followed by photon counters (c.f. Figure 1.5). In this situation, the detectors following the PBS can receive different numbers of particles. In order to recover binary outcomes allowing to test the CHSH Bell inequality [15], we introduce a post-processing of the outcomes. Namely, whenever the number of photons detected in the ’+’ port n+ is greater than or equal to a given threshold N, the outcome is set to ’+1’, otherwise it is set to ’-1’. Two particular strategies of interest here are the majority voting strategy when N = M/2, and the unanimity voting strategy when N = M.
1.3.1 Two sources Within this measurement setup, we consider two possible sources of entangled particles: a source of distinguishable particles, and a source of indistinguishable ones. The first one produces states of the form
M M ρ = ρ⊗ = ( ψ ψ )⊗ (1.3.1) M | ih | 12 1.3 Bell test with multiple pairs
0.2
0.1 Independent pairs 0.05
0.02 Indistinguishable Resistance to noise 0.01 photons M 1 2 5 10 20 50 Number of pairs
Figure 1.6: Maximal resistance to noise in the majority voting scenario (full red lines) and the unanimity scenario (dashed blue lines) for sources producing independent pairs (thick lines) or indistinguishable photons (thin lines). The unanimous vote is more robust with indistinguishable photons, but majority voting on independently produced pairs yields the most persistent violation. where ψ = 1 ( 00 + 11 ). The second source produces indistinguishable particles in | i √2 | i | i the state 1 M ΦM = (a†b† + a†b†) 0 . (1.3.2) | i M!√M + 1 0 0 1 1 | i States of this second form can be obtained for instance by parametric down conversion (PDC), which produce Poissonian distributions of such states.
1.3.2 Noise model
In order to quantify the amount of CHSH violation for each source, we introduce a noisy channel between the source and Alice. This channel consists of a random unitary U = exp( β~n ~σ) applied to the state, where the rotation axis ~n is uniformly distributed on the − · β2 2 Bloch sphere, and the angle β follows a gaussian distribution p(β) = e− 2σ2 (1 e 2σ2 )√2πσ − − centered at the origin and of variance σ2. The state after this channel is given by
ρ = p(β)(U 11)ρ (U † 11)[dU] (1.3.3) out ⊗ in ⊗ ZSU(2) where [dU] is the Haar measure on SU(2). 1 The application of this channel on the maximally-entangled state Φ = (a†b† + | 1i √2 0 0 a†b†) 0 produces the Werner state [19] 1 1 | i 11 ρ = w Φ Φ + (1 w) (1.3.4) | 1ih 1| − 4
1 2σ2 4σ2 6σ2 with w = 3 e− + e− + e− . We thus quantify violations by the largest amount of noise = 1 w 4σ2 which still allows one to find a violation of the CHSH inequality. − ' 13 Bell tests in bipartite scenarios
Figure 1.7: Sketch of the experimental setup used to test various Bell inequalities. Alice’s and Bob’s choice of settings are adjusted by rotation of linear polarizers.
1.3.3 Bell violation The best resistance to noise found after optimizing on the settings is represented as a function of the number of photons produced M in Figure 1.6. Interestingly, both sources provide a finite violation even for fairly large numbers of particles. Still, the maximum 1 violation decreases like M − in both cases, in connivance with the principle of macroscopic locality [27]. Thus, if a Bell violation can be found in multipair systems, it becomes less and less significant as the number of particles involved increases.
1.4 Experimental violation of Bell inequalities with a com- mercial source of entanglement
Back in the 1970-1980s, the first experimenters to test Bell inequalities had to put special efforts in building sources of entangled particles [29]. Since then, a lot of effort has been done to improve these sources. Today, it is possible to buy sources of entangled particles that are ready to test a Bell inequality. Here we demonstrate the violation of several Bell inequalities that we obtained with a commercially available source.
1.4.1 Experimental setup We used the QuTools source [28] to produce pairs of 810nm photons entangled in po- larization via spontaneous parametric down-conversion (SPDC). The source consists of a bulk β-barium borate (BBO) crystal, cut for Type II phase-matching, which is pumped at 405nm by a continuous wave diode laser (see Figure 1.7). The two photons produced by the source have orthogonal polarization and are emmited in cones. After selection of the spatial mode corresponding to the intersection of the two cones with pinholes and single-mode fibers, the photons collected can be described by the state
1 iφ ψ = H s V i + e V s H i . (1.4.1) | i √2 | i | i | i | i The setup allows us to measure each photon along any direction lying on the equator of the Bloch sphere. A partial tomography in this x-y plane shows that the state is close to a Werner state 11 ρ = V ψ− ψ− + (1 V ) (1.4.2) | ih | − 4 with visibility V = 94%.
14 1.4 Experimental violation of Bell inequalities with a commercial source of entanglement
1.4.2 Test of several Bell inequalities
Using this source, we tested the CHSH, I3322, AS1 and AS2 Bell inequalities [15, 30, 17, 18]. The results are represented in table 1.1. The values found are in good agreement with the values expected from the partial tomographic knowledge of the source.
exp exp tom I IL IL I I − pnoise(%) (σ units) I 2 2.731 0.015 2.683 49 27 CHSH ± I 4 4.592 0.024 4.769 25 13 3322 ± AS 6 7.747 0.026 7.750 67 23 1 ± AS 10 12.85 0.030 12.819 95 22 2 ± Table 1.1: Measurement of the CHSH inequality and of inequalities inequivalent to CHSH. exp IL is the local bound, I is the value of the Bell parameter obtained experimentally with the optimized settings, Itom is the expected value from the partial tomography, Iexp I − L is the difference between the obtained value and the local bound in terms of number of standard deviations σ and pnoise(%) is the critical level of white noise that can be added to the system without loosing a violation.
1.4.3 Chained Bell inequality On top of these inequalities, we also tested the N-settings chained Bell inequality, which can be written as
IN = E + E + E + ... + E E 2(N 1) = IN . (1.4.3) 11 12 22 NN − N1 ≤ − L The values obtained experimentally are reported for N 6 in table 1.2. The chained ≤ inequality has a number of applications which we mention below.
exp exp tom I IL N IL I I − pnoise (%) (σ units) 2 2 2.731 0.015 2.683 49 27 ± 3 4 4.907 0.019 4.925 48 18 ± 4 6 7.018 0.023 6.999 44 15 ± 5 8 8.969 0.026 8.996 37 11 ± 6 10 10.91 0.028 10.954 33 8 ± Table 1.2: Measurement of the chained inequalities with N settings per side.
Randomness certified by the no-signalling principle An interesting property of the chained Bell inequality is that the marginal probabilities P (a x) and P (b y) tend to 1/2 as the violation of the inequality increases. This allows | | one to certify that the outcomes produced by measuring the quantum system must be truly random, in the sense that no algorithm can possibly predict the measurement out- comes [31]. More precisely, the amount of true randomness that could be extracted from the ex- perimental results found by Alice can be evaluated by finding the largest marginal prob- ability P (a x) which is compatible with the measured Bell inequality violation Iexp. We ∗ | 15 Bell tests in bipartite scenarios
1
Quantum bound Noïsignaling bound for all N 0.9
0.8 (a|x) * P 0.7
0.6
N=2 N=3 N=4 N=5 N=6
0.5 2Nï2 2Nï3/2 2Nï1 2Nï1/2 2N IN
Figure 1.8: Maximum marginal probability compatible with a violation of the N-settings chained inequality. The bound implied by the no-signalling principle is identical for all N. performed this optimization over the set of quantum correlations as well as among all no- signalling correlations. The result is shown in Figure 1.8 together with the experimentally achieved values Iexp. The strongest bound imposed by the no-signalling principle is P (a x) = 0.7455 ∗ | ± 0.0057, achieved for the inequality with N = 4 settings. This allows one in principle to extract H (a x) = log P (a x) = 0.41 0.01 random bits per run. min | − 2 ∗ | ±
EPR2 local part
N Another property of the chained inequality is that its maximum quantum value IQ = π 2N cos 2N , achievable by measuring a singlet state [32], approaches the no-signalling bound IN = 2N as the number of settings N increases. This allows one to conclude that NS the singlet state has no local part in the sense of EPR2 [33]. Indeed, if a fraction pL of the measured pairs would behave locally during an experi- mental evaluation of the chained inequality (1.4.3) yielding the value Iexp, the following equation would hold: Iexp = p IN + (1 p )IN , (1.4.4) L L − L NL N N N where IL is the value of I achieved with the local pairs of particles, and INL a value of the same expression achieved on the rest of the particles. Since the following bounds hold: IN 2(N 1) and IN IN = 2N, the local part p of the measured states must L ≤ − NL ≤ NS L be bounded by Iexp p pmax = N . (1.4.5) L ≤ L − 2 N For Iexp = IN , we find pmax = N 1 cos π →∞ 0. Thus, for every number of Q L − 2N −→ settings N, testing the chained inequality can provide an upperbound on the local content of the state measured which eventually converges to 0. max In our case, the best bound on pL is found for N = 4 settings, yielding pL = 0.491 0.012. While recent work could demonstrate an even lower value [34], this simple ± 16 1.4 Experimental violation of Bell inequalities with a commercial source of entanglement experiment already shows that at least half of the photon pairs produced by the source are nonlocal.
1.4.4 Conclusion In this experiment, we relied on the fair sampling assumption because the single pho- tons detectors were not efficient enough to close the detection loophole. Moreover the detection events were not space-like separated. Yet, this experiment shows that a simple demonstration of several interesting results of quantum information theory is nowadays possible with modest equipment.
17 Bell tests in bipartite scenarios
18 Chapter 2
Nonlocality with three and more parties
In the precedent chapter, we focused on Bell-type experiments involving two parties only. While this is the simplest case, and indeed the most often discussed one, the idea of local correlations can be extended straightforwardly to multipartite scenarios involving more parties. Labeling the (output,input) of a third party Charly by (c,z), the locality condition (1.1.1) generalises to:
P (abc xyz) = p(λ)P (a x, λ)P (b y, λ)P (c z, λ), (2.0.1) | | | | Xλ and similarly for more parties. Tripartite correlations P (abc xyz) are then referred to as | nonlocal if and only if they cannot be decomposed as (2.0.1).
2.1 Defining genuine multipartite nonlocality
Just like entanglement can have more forms in a multipartite scenario than in the bipar- tite case [35], it is easy to realize that the definition (2.0.1) does not capture the whole potential of nonlocality in a tripartite scenario. Consider indeed some bipartite nonlocal correlations P (ab xy) and arbitrary statistics for Charly P (c z). The product of the AB | C | two distributions P (abc xyz) = P (ab xy)P (c z) violates (2.0.1) and is thus nonlocal. | AB | C | However it is clear that Charly plays no role in the nonlocality of these correlations. These correlations are thus not genuinely three-way nonlocal. This observation was first made by Svetlichny in 1987 [36], who proposed an inequality capable of certifying (if violated) that correlations cannot be explained by a mechanism involving fewer than 3 parties. This is the Svetlichny inequality
S = E + E + E E + E E E E 4 (2.1.1) 111 112 121 − 122 211 − 212 − 221 − 222 ≤ with E = 1 ( 1)a+b+cP (abc xyz), which is satisfied by all tripartite correlations xyz a,b,c=0 − | of the form P P (abc xyz) = p (λ)P (ab xy, λ)P (c z, λ) | 1 AB | C | Xλ (2.1.2) + p (λ)P (ac xz, λ)P (b y, λ) + p (λ)P (bc yz, λ)P (a x, λ) 2 AC | B | 3 BC | A | Xλ Xλ 19 Nonlocality with three and more parties with p 0 and p (λ) + p (λ) + p (λ) = 1. i ≥ λ 1 2 3 If being unable to decompose some tripartite correlations P (abc xyz) in the form of P | (2.1.2) is sufficient to conclude that none of the parties was separated from the other ones in the process that created these correlations, it was pointed out recently that this condition is not always necessary (see paper [N] and [37]). To understand why this is the case, let us consider the situation in which the three measurement events producing a, b, and c, are not simultaneous but follow an order: Alice measures first, then Bob, and finally Charly (A < B < C). If decompositions of the form (2.1.2) exist for the observed correlations we might want to conclude that these correlations can be reproduced by some interaction between pairs of parties. Yet, this is not possible if every such decomposition happens to contain PAB terms that are signalling from B to A, i.e. such that P (ab xy, λ) depends on y. Indeed, in the considered b AB | configuration (A < B < C), y can always be chosen freely after a. The distribution of a P thus cannot depend on y. It thus seems important, from a physical point of view, to consider decompositions (2.1.2) that are compatible with the situation in which the correlations are produced. In order to conclude something about the nature of correlations that is independent of the situation in which they appear, we suggest to require a consistent decomposition (2.1.2) to exist for all possible measurement situations. Thus, we say that correlations are Svetlichny-sequential iff they can be decomposed as
P (abc xyz) = p (λ)P TAB (ab xy, λ)P (c z, λ) | 1 | C | Xλ (2.1.3) + p (λ)P TAC (ac xz, λ)P (b y, λ) + p (λ)P TBC (bc yz, λ)P (a x, λ) 2 | B | 3 | A | Xλ Xλ for every possible ordering of the measurements. Non-Svetlichny-sequential correlations are then called genuinely tripartite nonlocal. For correlations that are not genuinely tripartite nonlocal in this sense, a biseparable model cannot be constructed coherently for all possible ordering of the measurements. Here P TAB (ab xy, λ) depends on the order | of measurement between Alice and Bob. Namely, P TAB (ab xy, λ) = P A
20 2.2 Multipartite Bell-like inequalities are performed is crucial in several situations, as shown here, and in the last chapter of this thesis.
2.2 Multipartite Bell-like inequalities
If the CHSH inequality (1.2.2) found many applications, it is certainly because of its outstanding properties, but probably also because of its simplicity. Indeed, the CHSH inequality has a lot of symmetries: it is for instance invariant under permutation of parties, so that no party plays a special role, and it only involves correlations between the parties’ outcomes, so that the specific outcomes of a party play no specific role independently of the other party’s outcome. Since the structure of the local polytope quickly gets too complicated to allow a direct computation of its facets [39] when more than two parties, inputs or outputs are consid- ered, it seems natural, when considering such scenarios, to first restrict one’s attention to Bell inequalities having properties similar to the ones we just mentioned. This allows one to simplify the analysis while leaving a hope that the results found in this way can be useful, since the CHSH inequality satisfies these constraints. In this perspective, we note that a complete description of the Bell inequalities which involve only full-correlations terms could be found for all (n, 2, 2) scenarios in [40, 41]. Here we denote by (n, m, k) the Bell scenario involving n parties, each with m possible measurement settings producing one out of k possible outcomes. On the other hand, we describe in paper [E] how the search for Bell inequalities that are symmetric under per- mutations of the parties can be simplified by considering projections of the local polytope, which are much easier to solve than the full polytopes. This allows us in turn to discover many families of Bell and Svetlichny inequalities for several scenarios (c.f. paper [E] for more details). Here, we consider a special form of Bell expressions that is both symmetric under permutation of the parties and only involves full-correlations. We show that many in- equalities presented throughout the years in the literature have this form. This allows us to propose a natural generalization of them to general (n, m, k) scenarios. Moreover, we show that several bounds on these expressions can be easily computed once some bound for the corresponding few-party inequality is known.
2.2.1 A general structure for (n, m, k) scenarios Let us consider the (n, m, k) Bell scenario. Denoting by ~s = (s , . . . , s ) 0, 1, . . . , m 1 n ∈ { − 1 n the settings of all parties, and ~r = (r , . . . , r ) 0, 1, . . . , k 1 n their results, we } 1 n ∈ { − } write the following Bell expression: s Ωn,m,k;f = f [s]m, r P (~r ~s) (2.2.1) − m k | X~s X~r h j ki where s = n s and r = n r are the sums of all parties’ inputs and outputs, x i=1 i i=1 i b c is the integer part of x, [x]y = x y x/y the modulo function, and f : 0, . . . , m 1 P P − b c { − } × 0, . . . , k 1 R is a real-valued function (defined by m k real parameters) that fully { − } → × characterizes Ωn,m,k;f . Clearly, this Bell polynomial is symmetric under permutation of the parties and only involves correlation terms, since only sums of all the parties’ settings and outcomes matter. A choice of function f(s, r) allows one to write an expression for any scenario with n parties, m inputs and k outputs. Table 2.1 shows how different choices of this function
21 Nonlocality with three and more parties
m inputs
chained DIEW BKP ?
CHSH Svetlichny CGLMP
n parties k outputs Svetlichny-CGLMP
Figure 2.1: Previously known families of Bell expressions (see [15, 42, 43, 44, 45, 46, 47, 48] and papers [F, L]) are recovered by equation (2.2.1) with the choice of parameter f(s, r) = δ r + δ [ r] . In particular, the CHSH expression is recovered for n = m = k = 2. s,0 s,1 − k This provides a natural way to extend these inequalities to a general scenario (n, m, k) (see also [49]). allow to recover several Bell-like expressions used in the literature. In particular, the choice f(s, r) = δ r + δ [ r] (2.2.2) s,0 s,1 − k allows one to generalize all the expressions represented in Figure 2.1 to the (n, m, k) sce- nario. Note that the generalization obtained in this way was also discovered independently by [49].
n m k f(s, r) Bell expression 3, odd 2 2 δ r MABK [50] ≥ s,0 3 2 2 δs,0 r DIEW [71] ≥ ≥ ∆ s 3 2 2 cos( − π) r , ∆ R DIEW [71] ≥ ≥ m ∈ 2 2 2 δ r + δ [ r] c.f. Figure 2.1 ≥ ≥ ≥ s,0 s,1 − k
Table 2.1: A summary of some known Bell expressions that can be recovered as special cases of Ωn,m,k;f .
2.2.2 Recursion relation
By performing the change of variable s = [s + s ] , r = [r + r s1+sn ] , equation 10 1 n m 10 1 n − m k (2.2.1) can be rewritten as m 1 k 1 − − Ω = Ω(sn,rn) P (r s ) , (2.2.3) n,m,k;f n 1,m,k;f n| n s =0 r =0 − Xn Xn
(sn,rn) where Ωn 1,m,k;f is equivalent upon permutation of inputs and ouputs to an (n-1)-partite − expression (2.2.1) conditionned on rn and sn. Since the (n-1)-partite polynomial is gen- erated by the same function f(s, r) as the n-partite one Ωn,m,k;f , this provides a way to relate the n-partite expression to polynomials of the same kind involving fewer parties. We describe below how this relation allows one to derive a number of bound for Ωn,m,k;f .
22 2.2 Multipartite Bell-like inequalities
a) b) c) d)
Figure 2.2: In a grouping models, n parties can be shared out into g groups. Arbitrary communication is allowed between parties belonging to the same group, but no communi- cation is allowed between different groups. a) With g = 1, every no-signalling correlation can be reproduced by the model. b) and c) g = 2: Parties are shared into two groups. This corresponds to the usual Svetlichny model. Correlations that cannot be reproduced here are genuinely multipartite nonlocal. d) For g = n, the model coincides with the usual local model. Any 2 < g < n allows one to interpolate between the local and usual Svetlichny models.
Tsirelson bounds Given, a Tsirelson bound Ω βT on a bipartite Bell-like expression, equation 2,m,k;f ≥ 2,m,k;f (2.2.3) induces the following Tsirelson bound for the n-partite polynomial:
n 2 T Ω m − β . (2.2.4) n,m,k;f ≥ 2,m,k;f Nontrivial Tsirelson bounds can thus be easily deduced for these multipartite Bell in- equalities, thanks to their special structure.
Generalized Svetlichny bounds By considering a scenario in which n parties are gathered into two groups, Svetlichny deduced an inequality which detects when interaction between all parties must have hap- pened [36]. More generally, the amount of interaction needed between n parties in order to reproduce some correlations can be quantified by the maximal number of groups g into which the parties can be separated while still being able to reproduce these correlations (c.f. Figure 2.2). Within each group, parties are allowed to communicate their inputs to each other, and to agree on which outputs they want to outcome, but no communication is allowed between the different groups1. Thanks to relation (2.2.3), the bound of any Ωn,m,k;f polynomial that can be achieved with parties distributed into g groups can be obtained from the local bound of the poly- L nomial with n = g. Denoting this bound by βg,m,k;f gives the following bound for Ωn,m,k;f upon separation of the parties into g groups:
n g L Ω m − β . (2.2.5) n,m,k;f ≥ g,m,k;f Note that similar bounds were already derived for correlations obtained by measuring quantum states that are positive under partial transposition across all partitions of the n systems into g subsystems [51].
Application to quantum states. Violation of (2.2.5) allows one to put an upper bound on the number of groups g into which the parties can be distributed in order for them to
1As discussed in section 2.1, the order in which the different parties are measured should in principle be included in the model. However one can show that this order is not important here, i.e. the different definitions discussed in section 2.1 coincide, because the inequalities we consider here only involve full- correlations.
23 Nonlocality with three and more parties be able to reproduce the observed correlations with local operation and communication within the groups. Here, we consider the above bounds in the case m = k = 2. It turns out that in order to test for an even or odd number of groups it is useful to consider two different Bell polynomials (c.f. paper [B]). For g odd, we thus choose f(s, r) = δs,0 r, and when g is even we choose f(s, r) = δ r + δ [ r] . s,0 s,1 − k Considering violations of these inequalities with measurements on the partially-entangled GHZ states GHZ = cos θ 00 ... 0 + sin θ 11 ... 1 (2.2.6) | θi | i | i shows that parties cannot be separated into more that g = 1 2 log sin(2θ) in order − b 2 c to reproduce the achieved correlations (see paper [B]). By considering measurement on the n-partite W state
1 W = ( 10 ... 0 + 01 ... 0 + 00 ... 1 ), (2.2.7) | i √n | i | i | i we could also show that letting two parties interact is not enough to reproduce the cor- relations that can be observed as soon as n 3. However, we found no violation for the ≥ bounds considered if the parties are separated into less than n 1 groups (see paper [B] − for more details).
Biseparable bounds
Finally, we note that equation (2.2.3) can also be used to deduce bounds on Ωn,m,k;f that are satisfied upon measurement of biseparable quantum states. In particular, for k = 2 ouputs and f(s, r) = g(s) r, we show in paper [L] that the following biseparable bound · holds:
m 1 m 1 1 n 2 − ηjπ − s Ωn,m,2;g r m − m g(s) max ηj csc g(s) ωj (2.2.8) · ≥ 2 − j=0,...,m 1 " 2m #! s=0 − s=0 X X
where ηj is the greatest common divisor of 2j + 1 and m, and ωj = exp(iπ(2j + 1)/m). The choice g(s) = δs,0 + δs,1 allows one to write an entanglement witness that is well adapted to detect genuine multipartite entanglement in multipartite GHZ state. We use this witness in chapter 3 of this thesis to detect genuine multipartite entanglement in a system of trapped ions.
2.3 Nonlocality from local marginals
In a multipartite system, every subset of parties constitutes a proper system in itself. The fact that these subsystems describe parts of the same total system requires them to satisfy some compatibility conditions: bipartite correlations P (ab xy) for instance are | compatible with the tripartite ones P (abc xyz) if and only if P (ab xy) = P (abc xyz). | | c | While this condition is easy to check when the global correlations are known, checking P whether several reduced states are compatible with each other is known as the marginal problem [52]. The marginal problem is trivial in the bipartite case: every single-party marginals P (a x) and P (b y) are always compatible with the joint distribution P (ab xy) = P (a x)P (b y). | | | | | However, the situation is less evident in situations involving more parties. For instance, it is known that if PAB violates the CHSH inequality, then it cannot be compatible with
24 2.3 Nonlocality from local marginals
correlations PBC violating CHSH as well; a phenomenon known as the monogamy of nonlocality [53]. Here we want to explore this multipartite constraint further. In particular, we ask whether some local marginals can witness nonlocality (or even genuine multipartite non- locality), because the only global correlations with which they are compatible is (genuinely multipartite) nonlocal.
2.3.1 An inequality In order to answer this question in the tripartite case, we consider all bipartite correlations PAB, PAC and PBC , which are compatible with Svetlichny-sequential correlations PABC . This set of correlations can be described as
Π = (P ,P ,P ) P s.t. (2.1.3) holds . (2.3.1) { AB AC BC | ∃ ABC } Since Π is convex, it can be characterized with linear Bell-type inequalities. Using the decomposition (2.1.3) and linear programming, one can show that the fol- lowing inequality is satisfied by correlations in Π:
A (1+B +B +C ) A (1+B +C +C ) B +C +B C +B C 4, (2.3.2) −h 0 0 1 0 i−h 1 0 0 1 i−h 0 0 0 0 1 1i ≤ where A = P (0 x) P (1 x), A B = P (00 xy) P (01 xy) P (10 xy) + h xi A | − A | h x yi AB | − AB | − AB | P (11 xy) and similarly for the other parties. Violation of this inequality by some AB | bipartite correlations thus implies that the only tripartite distributions with which they are compatible are genuinely-tripartite nonlocal. Interestingly, this inequality can be violated by measuring a noisy W state 11 W = p W W + (1 p) (2.3.3) | i | ih | − 8 as soon as p > 0.9548, by using the measurement operators
A = cos ασ + sin ασ ,A = cos ασ sin ασ 0 z x 1 z − x B = σ ,B = cos βσ + sin βσ (2.3.4) 0 − z 1 z x C = σ ,C = cos βσ sin βσ 0 − z 1 z − x with α = 3.6241 and β = 2.0221. However, no bipartite correlations achieved by measuring this state can violate a Bell inequality with binary inputs and outputs, because the bipartite reduced states p p 1 p ρ = ( 01 + 10 )( 01 + 10 ) + 00 00 + − 11 (2.3.5) red 3 | i | i h | h | 3| ih | 4 satisfies the Horodecki criterion for every 0 < p < 1 [54]. This shows that there ex- ist bipartite quantum correlations that are local, but only compatible with genuinely tripartite-nonlocal correlations.
2.3.2 Conclusion This illustrates the strength of the compatibility constraints that relates different parts of a system. Namely, the fact that subsystems are parts of the same system allows one to reveal properties of that system which were not apparent in its individual parts. Note that a similar result can be achieved with respect to entanglement (see paper [P] for more details).
25 Nonlocality with three and more parties
2.4 Tripartite nonlocal boxes
Another way to get some insight into the structure of multipartite correlations is by studying the largest possible set of multipartite correlations. Within the no-signalling subspace, this is defined by all correlations satisfying the positivity conditions
P (abc xyz) 0 a, b, c, x, y, z. (2.4.1) | ≥ ∀ The no-signalling polytope is thus easily characterized in terms of facets. Its extremal points, however, reveal that this polytope has a highly nontrivial geometric structure.
2.4.1 The tripartite nosignalling polytope Using the porta software [105], we could find all extremal points of the tripartite no- signalling polytope when all parties have binary inputs and outputs. After sorting these boxes into equivalence classes under permutation of inputs, outputs and parties, we could identify 46 families of extremal no-signalling boxes. A complete description of these families is given in paper [G]. Apart from the deterministic local strategies and the PR boxes, which are also extremal points of the bipartite nosignalling polytope, we found 44 families of truly tripartite boxes. Interestingly, three of them admit a decomposition of the form (2.1.3), and are thus not genuinely 3-way nonlocal in this sense. Using the 46 classes of Bell inequalities that define the local polytope in this sce- nario [60], we studied the relation of each box with respect to each Bell inequality (c.f. paper [G]).
2.4.2 Conclusion The reason why so many new no-signalling boxes appear when considering a scenario with more parties is still a mystery, which asks for further investigation. Nevertheless, the observation that the number of extremal boxes coincides with the number of Bell inequalities in this scenario was recently elucidated by Fritz [61]. It seems thus that a strong link exists between the set of nosignalling correlations and that of local correlations.
2.5 A tight limit on quantum nonlocality
As already mentioned in section 1.1, quantum correlations can be nonlocal, but not as much as imposed by the no-signalling conditions (1.1.2) alone. Indeed, while no-signalling correlations can achieve a value of CHSH equal to 4 with a PR box, it is known since Tsirelson [62] that measurements on a quantum system cannot exceed 2√2. Despite recent advances on the subject [63, 8], the question why the nonlocality of quantum correlations is limited the way it is remains open. Here, we introduce a simple game for which quantum correlations provide no advantage, whereas general no-signalling correlations do. The inequality associated to this game is shown to identify part of the boundary separating quantum correlations from more general no-signalling correlations.
2.5.1 Can you guess your neighbour’s input (GYNI)? Let n parties form a circle as represented in Figure 2.3. The game runs as follows: parties are given some input x 0, 1 , and are requested to output their right neighbour’s i ∈ { } 26 2.5 A tight limit on quantum nonlocality
x1 x2 xn
...
a1 a2 an
xn+1 = x1
Figure 2.3: Representation of the GYNI nonlocal game. The goal is that each party outputs its right-neighbour’s input: ai = xi+1. input. The success that the parties have in playing this game can be measured with the following quantity: ω = q(x)P (ai = xi+1 i x) (2.5.1) x ∀ | X where x = (x , x , . . . , x ) 0, 1 n denotes all the parties’ inputs, P (a = x i x) is 1 2 n ∈ { } i i+1∀ | the probability that the players obtain the correct outputs when they received the input string x, and q(x) is the probability that the set of inputs x is given to the parties. One can show that the maximum value of ω that the parties can achieve by using classical or quantum resources is
ωc = max[q(x) + q(x¯)] (2.5.2) x where x¯ = (¯x , x¯ ,..., x¯ ) withx ¯ = x 1 is the negation of x. Since this value is always 1 2 n i i ⊕ achievable classically, it follows that the Bell inequality ω ω can never be violated ≤ c quantum mechanically. One might think that the reason why quantum correlations give no advantage in this game, as compared to local correlations, is that signalling is required in order to guess someone’s input better than classically. But since quantum correlations are no-signalling, they cannot help it. However this is not totally true. Indeed, by choosing the distribution of inputs q(x) adequately, it is sometimes possible to achieve a violation ω > ωc of the inequality without allowing the player to guess anything on the other parties’ inputs. Consider for instance the following input distribution:
21 nˆ if x ... x = 0 q(x) = − 1 ⊕ ⊕ nˆ (2.5.3) (0 otherwise withn ˆ an odd number between 3 and n. In the case n =n ˆ = 3, the inequality ω ω ≤ c can be violated by several tripartite extremal no-signaling boxes (c.f. section 2.4). In particular, two boxes can achieve the value ω = 4/3ωc. It is thus clear that the bound ωc is not a consequence of the no-signaling condition.
2.5.2 Outlook We could check numerically that for the choicen ˆ = 2 n + 1 the inequalities ω ω are b 2 c ≤ c facets of the local polytope up to n = 7. Yet, they can always be violated by no-signalling correlations. This game thus tightly identifies part of the boundary separating quantum from supra-quantum correlations.
27 Nonlocality with three and more parties
2.6 Simulating projective measurements on the GHZ state
One way to study nonlocal correlations is to try to simulate them with a measureable amount of nonlocal resources. This allows one to put an upper bound on the power of these correlations. For instance, it is well known that correlations created upon measurement of a singlet state can be simulated by the use of shared randomness supplemented by 1 bit of communication [55], or by 1 use of a PR box [56]. Thus no correlations found upon measurement of a singlet state can achieve a task that 1 bit of communication, or 1 PR box, cannot. Here, we consider the simulation of the n-partite Greenberger-Horne-Zeilinger (GHZ) state 1 GHZ = ( 00 ... 0 + 11 ... 1 ). (2.6.1) | i √2 | i | i Since this state is genuinely tripartite nonlocal (it can violate the Svetlichny inequality (2.1.1) or its n-partite generalization [46, 47]), it cannot be simulated with just shared randomness and interaction between a subset of the parties. Nevertheless, we consider the task of simulating it with the aid of bipartite resources only.
2.6.1 Nonlocal resources Let us allow the parties to share nonlocal boxes of the following kinds in addition to pre-established randomness. PR box. A Popescu-Rohrlich (PR) box [57] is a nonlocal box that admits two bits x, y 0, 1 as inputs and produces locally random bits a, b 0, 1 , which satisfy the ∈ { } ∈ { } binary relation a + b = xy. (2.6.2) M box. A Millionaire box [58] is a nonlocal box that admits two continuous inputs x, y [0, 1[ and produces locally random bits a, b 0, 1 , such that the following relation ∈ ∈ { } is satisfied: a + b = sg(x y) (2.6.3) − where addition is modulo 2 and the sign function is defined as sg(x) = 0 if x > 0 and sg(x) = 1 if x 0. ≤ Note that none of these boxes is signaling: the outcomes produced by the boxes are locally random and thus carry no information on the other party’s choice of input.
2.6.2 Simulation A protocol to simulate measurements on the GHZ state (2.6.1) with nonlocal boxes runs as follows: before letting the parties choose their measurement settings, they are allowed to share any information, plus a number of boxes (as shown in Figure 2.4). The parties can then choose their measurement settings (which we represent by vectors ~a, ~b, ~c on the Bloch sphere). They are allowed to locally process this setting together with the pre-established shared randomness and accesses to their boxes. The parties then output the result of this process, which we denote by α, β, γ 1, 1 . ∈ {− } Using the above boxes, we considered the simulation of the correlations found by measuring the GHZ state in the equatorial plane, i.e. with ~a = (cos φa, sin φa, 0), etc. In this case the correlations take the form
α = β = ... = αβ = ... = 0 (2.6.4) h i h i h i 28 2.6 Simulating projective measurements on the GHZ state
Alice
PR box
M boxM boxM Charlie
PR box Bob
Figure 2.4: Setup for the simulation of the tripartite GHZ state in the x-y plane : two Millionaire boxes are shared between Alice and Bob and each of them shares a PR box with Charlie.
Alice PR box Charlie
PR box
M boxM boxM boxM boxM e
PR box
Bob PR box Dave
Figure 2.5: Distribution of bipartite no-signalling boxes that allows for the simulation of equatorial von Neumann measurement on the 4-partite GHZ state. for all marginal correlations, and
αβ . . . = cos(φ + φ + ...) (2.6.5) h i a b for the full n-partite correlation term. Here are the results that we could show (proofs in paper [C]): Theorem. Equatorial von Neumann measurements on the tripartite GHZ state can be simulated with 2 M boxes and 2 PR boxes distributed as in Figure 2.4. Theorem. Equatorial von Neumann measurements on the 4-partite GHZ state can be simulated with 4 M boxes and 4 PR boxes distributed as in Figure 2.5. Moreover, one can show that a PR box can be simulated with one bit of classical communication transmitted from one end of the box to the other one, and an M box with an average of 4 bits. Each of these protocols can thus be translated into communication models with a finite-average communication cost. Namely, and average of 10 bits allow for the simulation of equatorial measurements on the tripartite GHZ state, whereas 20 bits suffice on average for the four-partite case.
2.6.3 Conclusion We showed that bipartite no-signalling resources are enough to reproduce the nonlocal character of these GHZ correlations, even though these correlations are genuinely mul- tipartite nonlocal. Moreover, we provided models to reproduce these correlations with a finite amount of communication on average. Note that this latter result was recently improved for the tripartite case [59].
29 Nonlocality with three and more parties
30 Chapter 3
Device-independent entanglement detection
Entanglement is one of the most intriguing feature of quantum physics. It allows several particles to be in a state which cannot be understood as a concatenation of the sate of each particle. Experimental demonstration of entanglement is generally performed with one of the two following techniques: tomography of the full quantum state, or evaluation of an entanglement witness. In the first case, the state ρ of the system is characterized by performing a number of complementary measurements on it [64]. For instance, on two qubits, measurement of the product of all Pauli operators σ σ , with j = 0, 1, 2, 3, and σ = 11 allows one in i ⊗ j 0 principle to deduce ρ by solving the set of linear equations
tr(ρ σ σ ) = f (3.0.1) i ⊗ j ij
where fij is the observed frequency for the corresponding measurements. In practice however, experimental imperfections typically lead to a solution for the former set of equations which is unphysical so that more complicated techniques are generally used instead of the linear inversion, like maximum likelihood estimation [65]. Still, once the reconstructed state is found, theoretical analyses can be performed on it to check directly whether the quantum state is entangled or not. In contrast, an entanglement witness is an observable such that tr(ρ ) 0 when- W W ≥ ever ρ is separable [35]. Any decomposition of in terms of local observables allows one W to evaluate it by performing local measurements on the state under consideration. If a value tr(ρ ) < 0 is found, the measured state is then said to be entangled. W 3.1 Imperfect measurements
Any experimental manipulation is affected by imperfections, be it only the finiteness of the number of times measurements are repeated in order to accumulate sufficient statistics. Interestingly, the effect that statistical uncertainties on the frequencies fij can have on tomographically reconstructed states was analysed rigorously only very recently [66, 67]. Still, even in absence of statistical uncertainties, which can in principle be avoided by performing enough measurements in a random order, systematic errors in the measure- ment process can possibly affect the conclusion of a test. While this problem is known, it is seldom discussed in the literature. Let us show what kind of effects these errors can have in the detection of entanglement.
31 Device-independent entanglement detection
~m3 ~n3
~m2
~n2 ε ε
~m1 ε ~n1
Figure 3.1: Intended and actual measurement directions for tomography on a qubit. The actual measurement directions ~n are distant from the desired ones ~m by an angle smaller than ε.
3.1.1 Effects of systematic errors on tomography In order to evaluate the effect of systematic errors on the process of tomography recon- struction, we consider the situation in which each measurement can be slightly misaligned. Namely, if ~m ~σ denotes the desired measurement on a qubit, the actual measurement · performed can be written as ~n ~σ with ~m ~n cos(ε) (c.f. Figure 3.1). However, since ~n · · ≥ is unknown, results measured along ~n are interpreted during the reconstruction process as coming from measurements along ~m. Considering qubit states ψ , we looked for the maximum effect that these errors could | i have on the reconstructed state ρ by performing the following optimization :
min ψ ρ ψ ψ ,n h | | i | i i (3.1.1) subject to ~m ~n cos(ε) measurement i i · i ≥ ∀ The results of this numerical optimization are shown in Figure 3.2a. For small errors ε in the definition of the measurement bases, the uncertainty of the reconstructed n-qubit state increases at least as n ε (c.f. paper [O] for more details). Thus, if measurements √2 are done with linear polarizers having a precision of 1o in real space, for instance, the precision of the reconstructed state can decrease by 2.5% per qubit in the system. Interestingly, we found that entangled states are usually more robust to systematic errors than the worst bound shown in Figure 3.2a (see paper [O]). Nevertheless, imper- fect measurements on separable states can sometimes lead to an entangled reconstructed state [68]. Entanglement can thus be wrongly witnessed through tomography because of systematic errors.
3.1.2 Effects of systematic errors on entanglement witnesses In a similar fashion, we analysed the witness 1 = 11 GHZ GHZ (3.1.2) W 2 − | ih | which detects genuine multipartite entanglement [35]. For this we used the decomposition of in terms of local operators given in [69]. Allowing again all measurement operators W to differ from the prescribed ones by at most ε, we looked for the smallest value tr(ρ ) W that could be achieved by measuring a biseparable state ρ.
32 3.2 Witnesses insensitive to systematic errors?
min ψ ρ ψ min tr(ρbisep ) h | | i W a) 1 b) 0
0.8 n -0.05 n = 1 = 2 n = 4 0.6 n -0.1 = 2 n n = 6 0.4 = 3 -0.15 n = 8 0.2 n = 4 -0.2 0 -0.25 0 π/18 π/6 π/3 0 0.05 0.1 0.15 0.2 Systematic error ε Systematic error ε
Figure 3.2: a) Minimum fidelity of the tomographically reconstructed n-qubit state when the measurement settings deviate by at most ε from the requested ones. b) Minimum expectation value of tr( ρ) found using imperfect measurements on bispearable states. W
The results are shown in Figure 3.2b. In particular, we note that a negative value can be found, and thus entanglement wrongly detected, as soon as ε > 0. This entanglement witness measured in this manner is thus sensitive to misalignment of the measurements.
3.2 Witnesses insensitive to systematic errors?
If the entanglement detection schemes we just mentioned are sensitive to systematic errors, these kind of errors, unlike statistical errors, can be hard to evaluate in practice: how can one make sure that measurement settings are perfectly aligned? or that they are not more misaligned than some ε? How can one certify that the measurements do not act on a larger Hilbert space than expected? etc. Even in the case that these errors can be reasonably estimated, taking them into account makes the analysis of the situation complicated... It is thus worth asking whether entanglement detection can be made resistant to systematic errors. The answer to this question is already known since several years: Bell inequalities can detect entanglement without relying on any hypothesis about the kind of measurements performed or even about the nature of the measured system. Indeed, since no Bell inequal- ity can be violated by measuring a separable state, violation of a Bell inequality witnesses that the state measured is entangled. All it needs to certify entanglement through the violation of a Bell inequality, in a bipartite scenario for instance, is a proper way of mea- suring the correlations P (ab xy). That is, the inputs of each party should be indexed by | x and y, their outputs by a and b, and the systems should be well identified from each other. The conclusion is then device-independent as discussed in the introduction. Conversely, if the correlations found during an experiment violate no Bell inequal- ity, then the presence of entanglement cannot be certified based only on the device- independent hypotheses. Indeed, in this case a local strategy can reproduce the correla- tions, and thus some measurements on a separable state as well. Entanglement can thus be demonstrated in a device-independent manner if and only if measurement statistics can violate a Bell inequality. More precisely, if a Bell inequality β P (ab xy) c is satistifed by all lo- abxy abxy | ≤ cal correlations, then the measurement operators Ma x, Mb y used during the experi- P | | 33 Device-independent entanglement detection ment (which might not be the ones that we think we’re measuring) define the operator
= c11 abxy βabxyMa x Mb y, which is a witness for entanglement: it satisfies W − | ⊗ | tr(ρ ) 0 for every state ρ that is separable. A negative value tr(ρ ) < 0 is found as W ≥ P W soon as the Bell inequality is violated. When testing a Bell inequality we thus test an entanglement witness which is adapted to the measurement implemented experimentally, even if these measurements are not known to us. The conclusion is thus independent of these measurements.
3.2.1 Device-independent witnesses for genuine tripartite entanglement The equivalence between Bell inequalities and device-independent entanglement witnesses remains true in scenarios involving more than two parties. However, in these scenarios one is typically interested in witnessing genuine multipartite entanglement [35]. Indeed, it is generally easier to test entanglement between a subset of parties directly on those parties specifically. If tripartite Bell inequalities can typically be violated with states that are not genuinely multipartite entanglement, and are thus unable to witness this kind of entanglement, vio- lation of a Svetlichny inequality demonstrates genuine tripartite nonlocality (c.f. section 2.1) and is thus sufficient to demonstrate genuine tripartite entanglement as well. But it is not necessary [70]. Rather, genuinely tripartite entanglement can be demonstrated from observation of correlations P (abc xyz) as soon as these correlations cannot be obtained by | measuring a biseparable quantum state ρbisep, i.e. if the correlations are not biseparable correlations:
Pbisep(abc xyz) = tr(ρbisepMa x Mb y Mc z) (3.2.1) | | ⊗ | ⊗ | where Ma x, Mb y and Mc z are arbitrary measurement operators. The set| of tripartite| biseparable| correlations can be described in terms of a hierarchy of semidefinite programming (SDP) (see Figure 3.3 and paper [J]). It is thus possible to determine directly from the observed correlations whether a genuinely tripartite entangled state was measured, or whether these correlations are compatible with measurements per- formed on an biseparable state, in which case genuine multipartite entanglement cannot be demonstrated with the device-independent hypotheses only.
3.2.2 A witness for genuine multipartite entanglement Letting ~s 0, . . . , m 1 n be a vector denoting the choice of measurement for the ∈ { − } n parties within m possible ones, and ~r 0, 1 n be the vector of their outcomes, the ∈ { } following inequality is satisfied by all biseparable quantum correlations (c.f. papiers [J, L]):
s s 1 − n 2 π I = ( 1)b m cE + ( 1)b m cE 2m − cot = B (3.2.2) n,m − ~s − ~s ≤ 2m n,m [sX]m=0 [sX]m=1 where E = ( 1)rP (~r ~s) is the n-partite correlator, [x] = x x m and s = s , ~s ~r − | m − b m c i i r = ri are the sums of all parties’ inputs and outputs. This inequality can thus be i P P used as a device-independent witness to detect genuine multipartite entanglement. P Measuring the n-partite GHZ state GHZ = 1 ( 0 n + 1 n) with the jth measure- | i 2 | i⊗ | i⊗ ment settings of every party lying in the x-y plane as π π cos(φ )σ + sin(φ )σ , with φ = + j , for j = 0, . . . , m 1, (3.2.3) j x j y j −2mn m −
34 3.3 Experimental demonstration
I3,3 = B3,3
Q3
S2/1
Q2/1
Figure 3.3: Particular slice in the space of tripartite correlations with 3 settings and 2 out- comes representing schematically the sets of general quantum correlations (Q3), Svetlichny correlations (S2/1) and biseparable quantum correlations (Q2/1). The inequality (3.2.2) is also represented and detects correlations that are not genuine tripartite-nonlocal.
n 1 π yields the value In,m = 2m − cos 2m > Bn,m. This inequality can thus detect genuine multipartite entanglement in noisy GHZ states ρ = V GHZ GHZ + (1 V )11/2n with π 1 | ih | − visibilities down to Vc = (m sin 2m )− . As the number of settings m increases, this critical visibility decreases, tending to the value of 2/π [71]. While inequality (3.2.2) reduces to the Svetlichny inequality for m = 2, and thus also detects genuine n-partite nonlocality in this case, it does not do so anymore for more inputs. Multipartite entanglement is then detected with a lower visibility than multipartite nonlocality.
3.3 Experimental demonstration
In the group of Prof. R. Blatt in Innsbruck, we tested the inequality (3.2.2) on a system of trapped 40Ca+ ions [72].
3.3.1 Experimental setup and procedure We used a linear trap loaded with n = 3, 4 or 6 ions, the logical states 0 and 1 of each | i | i ions being encoded in the S (m = 1/2) ground state and D (m = 1/2) metastable 1/2 − 5/2 − state, respectively. After initialization of the system in the ground state of the center-of-mass motion by Doppler and sideband cooling and in the logical state 0 n by optical pumping, the ions | i⊗ can be brought to the GHZ entangled state by applying a Mølmer-Sørensen gate [73]. Measurement of the ions in the computational basis is achieved by the electron-shelving technique by scattering light on the S P transition, and detecting the fluorescence 1/2 ↔ 1/2 with a photomultiplier tube. In order to perform measurements of all ions in the x-y plane of the Bloch sphere, we first apply local phase gates exp( i φ σ ) by means of AC-stark- − 2 z shift beams focused on individual ions. The x axis of the Bloch sphere of all ions is then brought to the computational basis by applying a collective π/2 rotation on the qubit transition. While the Mølmer-Sørensen gate can yield maximally entangled states with a high fidelity, the coherence time of the GHZ state produced in this way is of about 2 ms for n = 3 ions, and it decreases quadratically with the number of entangled qubits [74]. The
35 Device-independent entanglement detection duration of the σ rotations, performed sequentially on the ions, taking 100 µs/2π, a z ∼ significant decrease in the quality of the state can take place during the application of these pulses. To avoid this effect, we inverted the state of half of the ions before doing the Mølmen-Sørensen entangling gate for n = 4, 6. This allows one to produce a decoherence- free GHZ state of the form 1 ( 0 n/2 1 n/2 + eiϕ 1 n/2 0 n/2) whose coherence time of 2 | i⊗ | i⊗ | i⊗ | i⊗ 300ms leaves enough time to manipulate all ions. Note that the measurements settings ∼ (3.2.3) need to be adapted to this new state. We thus used the following ones on n = 4, 6 π ions: for m = 2 settings, we used the phases φj = j 2 for the first half of the ions and n+1 1 j n+1 π φj0 = 12n π + −2 π for other ones; for m = 3 we used φj = 12n π + j 3 for the first ions and 2 j φj0 = −3 π for the other ones, where j = 0, . . . , m 1 denotes the different measurement − 5 7n setting of each party. The optimal state then has the phase ϕ = 24−π . Finally, to cancel the effect of eventual drifts during the experiment, measurements sets were taken by blocks of 50 identical measurement chosen in a random order.
3.3.2 Addressing errors In order to show an indisputable violation of (3.2.2), the experiment producing the corre- lations should close all loopholes that can appear in a Bell experiment (c.f. chapter 1). Of course, this is not the case in the present experiment: even though the detection loophole is closed here, measurements were not performed in a space-like manner. In fact, the different systems are not even totally isolated from each other since they are separated by only 3 5 µm. The measurements performed on the ions might thus not be put in a ∼ tensor form as assumed in section 3.2. If arbitrary joint measurements are allowed in the decomposition (3.2.1) instead of a tensor product of local measurement, any correlation can be obtained by measuring a biseparable state, and thus any value of the inequality (3.2.2) can be reached as well. No interesting bound on an inequality can thus be proven in the presence of arbitrary cross-talks, i.e. without making some assumptions about which cross-talk is present in the system. Note that this situation is similar to the fact that no interesting bound can usually be put on a standard entanglement witness in presence of arbitrary systematic errors. We thus performed a special analysis in order to estimate the amount of cross-talk in our system, and how it could influence the biseparable bound Bn,m of the inequality. In our system, we expect the strongest source of cross-talk between the ions to be due to the imperfect focusing of the AC-stark shift lasers. Indeed, it is the only action which is supposed to act on some ions specifically and which might not: leakage of this laser onto neighbouring ions can cause them to feel part of the rotation imposed on the first ion. The state of an ion, or equivalently the basis in which it is measured, can thus depend on the measurement settings of the other ions. This effect can be modeled by replacing the measurement phases φ by φ = M φ where M = 1, and 0 M if j = k is j j0 k jk k jj ≤ jk ≤ 6 the amount of cross-talk from ion k to ion j. Here is a bound on the worst addressing P error. In order to evaluate the impact that these errors can have on the bound Bn,m, we estimated the amount of addressing errors present in the experiment. This allowed us to determine an upper bound on the addressing errors which was not exceeded in the 6 experiment, except possibly with a probability smaller than 10− . This upper bound is = 0.52% for n = 3, = 5.2% for n = 4, and = 5.4% for n = 6. We then computed numerically the maximum impact ∆IAE = I I=0 that addressing errors bounded by bisep− bisep can have on the biseparable bound for the settings we intended to use in the experiment. Assuming that the maximum contribution of the addressing errors to (3.2.2) is given by
36 3.4 Conclusion
∆IAE, we update the bound B to BAE = B +∆IAE to account for the cross-talks present in the experiment. Note that even though the actual measurement settings might differ from the ones we intended to measure, the modified bound BAE remains valid in the presence of cross-talk if the measurements implemented in the lab differ (not too much) from the ideal ones, =0 because B > Ibisep.
3.3.3 Experimental results The experimental evaluation of the witness are summarized in table 3.1 In all cases the measured values are consistent with a visibility of the state of about 90%, except for the tripartite case in which the GHZ state was not decoherence-free. The inequalities with two inputs (m = 2) coincides with the Svetlichny inequalities and thus detect genuine multipartite nonlocality. The witness with three inputs (m = 3) however, is able to detect genuine multipartite entanglement even in absence of genuine nonlocality. This allows one to demonstrate stronger violations as shown in table 3.1.
Iexp BAE n m BAE Iexp Visibility − (σ units) 2 4.234 4.78(6) 84(1) 9 3 3 10.894 12.39(1) 79.5(1) 150 2 8.832 10.41(6) 92.1(5) 26 4 3 33.513 42.53(8) 90.9(2) 113 2 36.4 40(1) 89(3) 4 6 3 306.8 376(3) 89(1) 23
Table 3.1: Summary of the experimental measurement of (3.2.2). For each scenario considered, the value of (3.2.2) observed is reported as Iexp, together with the associated visibility, i.e. the ratio between this value and the one expected from optimal measurement on a perfect GHZ state. The experimental value should be compared to the bound BAE, which includes a correction due to the addressing errors observed between the ions (c.f. section 3.3.2).
3.4 Conclusion
Any measurement of a device-independent entanglement witness results in the test of a standard entanglement witness which relies on the measurement settings actually imple- mented in the lab rather than on measurement settings which might not exactly corre- spond to the experimental situation. This ensures that a violation of the inequality cannot be caused by a miscalibration of the experiment. Device-independent witnesses are thus particularly robust to (possibly unknown) measurement imperfections inherent to every experimental test. Motivated by this perspective, we constructed device-independent witnesses able to detect genuine multipartite entanglement. Since these witnesses can detect genuine multi- partite entanglement even in absence of genuine multipartite nonlocality, they can provide larger experimental violations than tests of Svetlichny inequalities, as was demonstrated in the experiment we conducted with the Innsbruck ion group of Prof. R. Blatt. Despite being robust to imperfect measurements, the bounds of these witnesses can be affected by cross-talks between subsystems if these are not perfectly isolated from each
37 Device-independent entanglement detection other. Since this problem is quite generic, and is present in many experimental setups, it deserves further investigation.
38 Chapter 4
Quantum information put into practice
Allowing information to be carried by physical systems described by the rules of quantum physics led to a deep questioning of the theory of information. While many questions remain open, the emerging field of quantum information already led to several remarkably concrete applications which would not exist otherwise. Here we present a modest contribution to the analysis of the security of Quantum Key Distribution (QKD), as well as a protocol which can be used to question a database with some level of security.
4.1 Memoryless attack on the 6-state QKD protocol
Quantum key distribution (QKD) allows two parties who share an initial secret key of finite size, to increase its size by exchanging quantum and classical signals through an untrusted environment. The new key generated in this way can then be used for any cryptographic application [80], such as secure transmission of a secret message, a task which is not known to be possible by classical means. Standard security proofs for QKD protocols aim at relying on the weakest possible assumptions. For instance, it is usually admitted that a possible eavesdropper is not constrained by technological limitations but only by the laws of physics. Such assumptions allow one to derive strong security bounds. However, if a particular circumstance happen to restrict further the possible action of an eavesdropper, more refined security analyses taking these limitations into account can allow the trusted parties to improve the efficiency of their protocol. Motivated by the effort put in several groups worldwide [81, 82, 83, 84] to implement quantum memories preserving coherence and population over more than several milisec- onds, we consider the case in which the eavesdropper has no access to a long-lasting quantum memory. Security proofs applicable in this scenario have been presented in [85] for the BB84 protocol, and more recently for the BB84, SARG and 6-state QKD protocols [86]. Here we give a tighter bound than [86] for the achievable secure key-rate of the prepare-and- measure 6-state protocol when the eavesdropper has no access to any quantum memory.
4.1.1 The 6-state protocol The 6-state protocol for quantum key distribution [87] runs in 4 parts.
39 Quantum information put into practice
Alice Eve Bob
ρi F ρ0 { k} i
Figure 4.1: Schematic representation of a Prepare and Measure QKD protocol: Alice prepares a quantum state ρi that she sends to Bob though a public quantum channel, which can be under the control of an eavesdropper.
Distribution : Alice prepares one of the six qubit states ρ = ψ ψ chosen uni- i | iih i| formly at random within
0 + 1 0 1 0 + i 1 0 i 1 ψ1 = 0 , ψ2 = 1 , ψ3 = | i | i, ψ4 = | i − | i, ψ5 = | i | i, ψ6 = | i − | i. | i | i | i | i | i √2 | i √2 | i √2 | i √2 (4.1.1) A i 1 She remembers the basis b = − corresponding to this state as well as the bit X = i 1 b 2 c − mod 2. Alice sends this state to Bob through a public quantum channel. Upon receival of the system from Alice, Bob measures it in either the x, y, or z basis. He remembers his choice of basis bB = 0, 1, 2 as well as the result of his measurement Y = 0, 1. This A step is repeated N times, allowing the parties to accumulate the strings bk , Xk and B { } { } bk , Yk . { } { } A B Sifting : Alice and Bob publicly announce their choice of bases bk and bk . Having learned the other party’s choice of basis, they discard the runs k in which bA = bB (X k 6 k k and Yk are not expected to be correlated in this case), and keep the results from the other A B runs indexed by k0. The basis information is then b = b = b . k0 k0 k0 Error correction : An error correction protocol is run from Alice to Bob1 in order to correct for expected errors between their sifted raw key strings X and Y . This { k0 } { k0 } corrects Bob’s string Y to let him hold the same sifted bit string X as Alice. This { k0 } { k0 } procedure also lets Alice and Bob evaluate the average Quantum Bit Error Rate (QBER): Q = P (X = Y ). k0 6 k0 Secure key extraction : Privacy amplification is performed on the corrected bit strings X in order to extract its secret part. { k0 } During the whole protocol, exchanges of classical information are authenticated with the initial secret key shared by the two parties in order to avoid man-in-the-middle attacks.
4.1.2 Secret key rate Here, we consider an eavesdroper, Eve, which can access the quantum channel used by Alice to send the quantum states she prepares to Bob, and which can listen to all classical transmissions taking place during the protocol. However, Eve cannot hold any quantum information. Her most general interaction with the quantum channel can thus be modeled by a POVM acting independently on each of the qubits sent by Alice (c.f. Figure 4.1). Notice that Eve’s power is greatly reduced compared to the case in which she performs a general coherent attack. In particular she cannot use any information about the basis A,B used by Alice or Bob sk to choose how to measure her system. Moreover, since each run k of the protocol is treated independently of the precedent ones by Alice and Bob, the most powerful attack that Eve can perform is an individual attack.
1Note that reverse reconciliation, in which Bob sends information to Alice for her to recover Bob’s key, or two-way reconciliation [88] is also possible, but we don’t consider this case here.
40 4.1 Memoryless attack on the 6-state QKD protocol
1
0.8 Collective attack Memoryless attack (EB) Memoryless attack (PM)
0.6
0.4 ertkey rate secret
0.2
12.6% 21.0% 0 0% 5% 10% 15% 20.4% 25% QBER
Figure 4.2: Comparison of the secret key rate of the 6-state protocol in different situations. The bound for collective attack is as given by [90]. The two bounds against adversary without a quantum memory are in the entanglement-based scheme (EB) as given by [86] and as given by equation (4.1.4) for the prepare and measure scheme (PM).
We thus use the Csisz´ar-K¨orner formula [89], which expresses the secret key rate that Alice and Bob can extract during a realization of the protocol:
r = I(A : B) min(I(A : E),I(B : E)) (4.1.2) − where A represents Alice’s sifted key (i.e. X ), B Bob’s sifted key, and E any system { k0 } hold by Eve. I(X : Y ) here stands for the mutual information between variables X and Y . The mutual information between Alice and Bob is given as usual by:
I(A : B) = 1 h(Q), (4.1.3) − where h(p) = p log p (1 p) log(1 p) is the binary entropy function of p. The following − − − − result provides a lower bound on the key rate r by upper-bounding the maximal mutual information between Alice and Eve as a function of the QBER.
Result. The maximum information that an eavesdropper without quantum memory can have in common with Alice’s bits after sifting is given by:
1 1 3Q(2 3Q) I(A : E) = 1 h − − . (4.1.4) 3 " − p 2 !# (proof in Appendix B)
Note that this bound does not refer to the length N of the raw key produced by Alice and Bob. It is thus only valid in the limit of infinite key length N . → ∞ 4.1.3 Discussion A result similar to the one presented here was recently published by Aur´elien Bocquet, Anthony Leverrier and Romain All´eaume in [86]. However, their analysis refers to the entanglement-based realization of the 6-state protocol. In this version, preparation of the state ρi by Alice is realized by letting her measure in the σx, σy or σz basis a maximally entangled state shared with Bob. Since the eavesdropper can interact with the quantum
41 Quantum information put into practice
channel during the distribution of the entangled state ρAB, she can in principle hold a purification ψ such that tr ( ψ ψ ) = ρ of this state. | ABEi E | ABEih ABE| AB A comparison between the achievable key rate in the above prepare-and-measure and in the entanglement-based scheme is shown in Figure 4.2. This shows that the key rate is slightly higher in the prepare-and-measure scheme. This contrasts with the same bounds for the BB84 QKD protocol, which are identical for both implementations.
4.2 Private database queries
While QKD allows one to secure the communication between two trustfull parties, many more cryptographic tasks can be considered. Here we consider a situation in which Alice wants to learn about an element of a database held by Bob, without letting Bob know which element she’s interested in. This task is also known as 1 out of N oblivious transfer (for a database of N ele- ments) [75]. The security of the database querries consists of two parts: Database security: Bob wants to bound the information that Alice can access on his • database. Ideally he would like this information to be restricted to 1 bit per querry. User privacy: Alice wants to bound the probability that Bob can learn which item • of his database she is interested in. Ideally, he should get no information about it. Even though it was proved that both aspects of the security cannot be fully satisfied at the same time [76, 77], Giovannetti, Lloyd and Maccone [78] recently proposed a quantum protocol that could provide a reasonable level of security for both the user and the database provider. However, in lossy situations the security of their protocol is compromised: since it requires Alice to send her question to Bob before knowing whether her system will come back with an answer or not, Bob can take advantage of the losses by requiring Alice to send her question several times, and thus learning what her question is with high probability. Here we propose a protocol for private database queries based on the SARG QKD protocol [79], which is fundamentally noise-resistant. After presenting the main ideas of our protocol, we argue about the partial security it provides to both the database provider, and the user.
4.2.1 Sketch of the protocol The protocol for private database queries presented here is based on the SARG04 QKD protocol [79], and only differs in the classical processing. We summarize here the main steps of the protocol. Distribution : Bob uniformly chooses one of the four qubit-states , , , | ↑i | →i | ↓i and sends it to Alice. Alice measures the quantum system she received from Bob | ←i either in the σx or in the σz basis and records the measured state. Sifting : If Alice didn’t receive some systems from Bob, due to losses, she tells so to Bob which discards these runs. This allows the protocol to be loss resistant since at this stage, no information about the database, or about Alice’s question has been exchanged. For the systems that Alice received, Bob announces a pair of states within the following ones which contains the state he prepared: , , , , , , {| ↑i | →i} {| →i | ↓i} {| ↓i | ←i} , . {| ←i | ↑i} Transcoding : Bob translates the state and to bits 0, and and | ↑i | ↓i | ←i | →i to bits 1. On her side, knowing her measurement results as well as the sifting sets, Alice
42 4.2 Private database queries tries to guess the bit that Bob computed. This can be summarized in the following table if her measurement result is : | ↑i
Alice’s measurement result sifting set guess of Bob’s state guess of Bob’s bit , ? ? | ↑i {| ↑i | →i} , 1 | ↑i {| →i | ↓i} | →i , 1 | ↑i {| ↓i | ←i} | ←i , ? ? | ↑i {| ←i | ↑i} Information reduction : The bit string of length k N is divided into k substrings × of length N. These substrings are added bitwise, yielding a string of length N. The bitwise addition is commutative and acts as + + = = +, + = , + ? = ? =? (c.f. ⊕ −⊕− ⊕− − ⊕ −⊕ Figure 4.3). If Alice is left with a string of question marks ?, the protocol is restarted. If this happens too often, Bob aborts the protocol to avoid that Alice keeps only cases where she has few question marks.
Figure 4.3: Alice’s information on the key is reduced by xor-ing several keys together.
Database access : Alice announces the number s = j i where j is the item of the − database that she’s interested in and i is an item of the xor-ed key Kf that she knows. Bob announces the N bits C = X Kf where X are the elements of his database. i i ⊕ i+s n Alice deduces the element she’s interested in X = C Kf . i i ⊕ j
4.2.2 Discussion As mention above, a private database query protocol must provide two kinds of security. First, the database holder needs some guarantees that little information about his full database is revealed during the protocol. To see why this is the case here, we realize that the only way for Alice to know elements of Bob’s database is by guessing bits in the key Kf . But the states that Alice needs to discriminate for this, even after having learned the sifting sets, are not orthogonal to each other. An individual attack thus never allows her to learn Bob’s bit with certainty. There is thus a bound on how much information on the database Alice has access to in this case. In paper [H] we discuss in more details how the reduction step ensures that the key hold by Alice contains many question marks, so that she cannot learn many elements of Bob’s database. Second, the user needs to make sure that the database holder has little chance of guessing the element of the database that she is interested in. In order to learn the item of the database that Alice is interested in, Bob needs to guess j, the item of Alice’s final key that is different from a question mark. He thus needs to learn about the conclusiveness of Alice’s transcoding. But the choice of Alice’s measurement bases is unknown to Bob, he can thus never be sure whether she translated her result to a question mark or not. This remains partly true even in the case that Bob sends different states than the ones prescribed by the protocol (c.f. paper [H] for more details).
43 Quantum information put into practice
The above protocol for database queries thus provides some level of security for both the user and the database provider, while being resistant to losses. The exact amount of security provided is however not very clear yet. In particular, we only considered here specific individual attacks. It would thus be interesting on one side to study more general attacks, and on the other side to develop security proofs for given classes of attacks.
44 Chapter 5
Finite-speed hidden influences
The violation of a Bell inequality with space-like separated measurements precludes the explaination of nonlocal correlations in terms of causal influences propagating slower than light. Yet, these correlations can still be explained in a causal manner if one gives up Bell’s locality condition. Indeed, this is the explanation followed when one says something like “A measurement on the singlet state ψ = 1 ( 01 10 ) yielding result ‘0’ in the | −i √2 | i − | i computational basis of Alice prepares the state 1 for Bob”. With a slightly different | i taste, Bohmian mechanics also provides a causal explanation for quantum correlations, which does not rely on quantum steering or collapse of the wavefunction. However, both of these explanations are much more nonlocal than a simple violation of Bell’s local causality condition implies: not only do they involve faster-than-light influences at a distance, but these influences also have immediate effects on distant particles no matter how far away they are. Here we question whether such a strong violation of the notion of locality is necessary or not.
5.1 Finite-speed propagation and v-causal theories
One way of violating Bell’s local causality condition while still keeping a notion of“locality” is to allow causal influences to propagate faster than light, but only up to some finite speed v < . In this way, instantaneous influence at a distance is avoided, and causal ∞ influences can still be understood as propagating in spacetime, i.e. acting locally, “de proche en proche”. Since the advent of special relativity, it might seem uncalled-for to consider faster- than-light propagations in space-time1. Indeed, it is well-known that faster-than-light in- formation transmission in a Lorentz-invariant theory can generate temporal paradoxes [9]. However, this needs not be the case if the theory describing the interaction with supra- luminal transmissions is not Lorentz-invariant. For instance, if the speed of every faster- than-light communication is defined in a unique reference frame, then a temporal order is restored. Considering thus a preferred reference frame for definiteness, we can formalize the idea of finite-speed causal influences as follows: to every event K, a past and a future v-cone can be associated in the preferred frame (c.f. Figure 5.1a). What happens at K can only influence other events lying in the future v-cone of K, and K can only depend on what is contained within its past v-cone. We denote by A < B configurations in which
1As a matter of fact, the same remark applies to instantaneous influences of the kind we just mentioned, which is rarely mentioned.
45 Finite-speed hidden influences
a) b) c) future B time
K A A B
past space
Figure 5.1: Space-time diagram in the preferred reference frame a) The past and future v- cones (hatched areas) define the sets of events that can influence, or that can be influenced by K within a v-causal theory. b) A < B: finite-speed influences can propagate from A to B. c) A B: no influence can be directly exchanged between two events which are ∼ not in each other’s v-cones.
A lies in the past v-cone of B, and A B those in which A and B lie outside each-other’s ∼ v-cones (c.f. Figure 5.1b-c). Any theory satisfying these constraints is referred to as being v-causal. Note that Bell’s condition of local causality is recovered for v = c.
5.1.1 v-causal models and experimental limitations Clearly, v-causal theories, just like locally causal ones, are fundamentally incompatible with quantum physics. Indeed, they don’t allow two parties to violate a Bell inequal- ity if their measurements are performed simultaneously in the preferred frame, whereas quantum physics predicts that such inequalities can be violated independently of the space-time location of the measurements. Provided that correlations in nature agree with the quantum predictions, one could thus expect to be able to rule out v-causal models experimentally. However this is not directly possible. Indeed, due to the finite accuracy inherent to every experimental manipulation, a v-causal model with sufficiently large speed v can always explain the experimental violation of a Bell inequality. Moreover, since quantum correlations are no-signalling, they can always be reproduced with the aid of the one-way communication available to v-causal models. Thus, quite on the contrary, if all correlations that can be observed in Bell-like experiments agree with the quantum predictions, then they can also be explained by a v-causal theory. Experiments performed so far have thus only been able to put a lower bound on the speed v that is needed for the viability of v-causality. For instance, Salart et al. [91] and Cocciaro et al. [92] have shown that, if the speed of the earth in the preferred frame is 3 less than 10− c, then v-causal theories must have a speed v larger than 10000 times the speed of light c. Given that experimental results cannot rule out v-causal models directly, we examine below in more details the potential physical consequences of these theories. In the following we distinguish between several kinds of correlations. First, correla- tions are referred to as easily accessible in an experiment if they don’t require very good synchronization between any measurements. All correlations that a v-causal model can freely choose because influence was able to propagate through all parties are of this kind. Second, hardly accessible correlations are those which require nearly perfect synchroniza- tion, the degree of synchronization required depending on the speed v of the model. Since some measurements are too simultaneous to allow influences to propagate between them in this case, v-causal models cannot produce all possible correlations of this kind. Finally,
46 5.2 The hidden influence polytope correlations are said to be not directly accessible if they require perfect synchronization between some measurements. In this case at least part of these correlations involve si- multaneous measurements and are thus even inaccessible in principle. A v-causal model is then said to be quantum if every time its correlations are easily accessible they are also in agreement with the quantum prediction. v-causal models which are not quantum can in principle be detected experimentally, whereas quantum v-causal models are experimentally indistinguishable from quantum physics without extraordinary synchronization capabilities. Even though both easily accessible and hardly accessible correlations are in principle measurable, we would like to say something about v-causal models independently of their typical speed v based only on the measurement of easily accessible correlations.
5.1.2 Influences without communication?
As mentioned in chapter 1, the fact that faster-than-light influences be needed in order to reproduce some correlations does not necessarily allow these correlations to be used to signal faster than light. Rather, a violation of the no-signalling conditions (1.1.2) is needed to allow correlations to be used for communication. The superluminal influences of a v-causal model can thus remain hidden from observers having only access to the produced correlations if these correlations are no-signalling. In particular, as long as the correlations produced agree with the quantum predictions, they are no-signalling and thus cannot be used to communicate. Since all easily accessible correlations produced by a quantum v-causal model are quantum, simultaneous measurements must be considered in order to allow quantum v- causal models with arbitrary speed v to produce correlations diverging from the quantum prediction. It was suggested in [93, 94] that the correlations predicted by a v-causal model in this situation could become signalling and allow for faster-than-light communication. Here, we investigate this question in more detail, and show that it is indeed possible to communicate faster than light in a v-causal world in which all easily accessible correlations are in agreement with quantum physics. Note that a first example of situation in which a v-causal model was shown to allow for faster than light communication was put forward recently in [95]. However, this example requires the observation of supra-quantum correlations in order to conclude. It thus doesn’t apply to quantum v-causal models, and is not likely to lead to an experimental application. We present below a general approach which allows one to test if the existence of signalling correlations can be deduced from the knowledge of potentially accessible cor- relations. We then examine whether such a test can be expected to be conclusive if correlations observed experimentally are assumed to be the ones predicted by quantum theory.
5.2 The hidden influence polytope
Following the above discussion, we consider a space-time configuration in which some measurements are simultaneous in order to open the possibility for quantum v-causal models to produce non-quantum correlations. We then examine whether the correlations
47 Finite-speed hidden influences
time B0 C0
BC
D
A space
Figure 5.2: In the four-partite Bell-type experiment characterized by the space-time or- dering R = (A < D < (B C)), no influence can be exchanged between Bob and Charly. ∼ However, if Charly delays his measurement, he can allow the configuration to recover a complete order T1 = (A < D < B < C0). Similarly, Bob can delay his measurement in order to obtain the order T2 = (A < D < C < B0). .
produced by the model in this configuration can remain no-signalling or not2. For definiteness, let us consider here the 4-partite space-time configuration R = (A < D < (B C)) shown in Figure 5.2. A v-causal model in this situation must produce ∼ BC correlations that are local, even after conditioning on what happened at A and D (see paper [M] for more details). The correlations P (bc yz, axdw) must thus satisfy all | bipartite Bell inequalities βi P (bc yz, axdw) βi (5.2.1) bcyz | ≤ 0 bcyzX where (βi , βi ) denote the coefficients of all Bell inequalities that are relevant given { 0 bcyz }i the number of inputs and outputs of each party. On the other hand, the correlations P (abcd xyzw) are no-signalling if and only if they | satisfy the 4-partite no-signalling conditions:
P (abcd xyzw) = P (bcd yzw) , P (abcd xyzw) = P (acd xzw) | | | | a X Xb (5.2.2) P (abcd xyzw) = P (abd xyw) , P (abcd xyzw) = P (abc xyz). | | | | c X Xd No-signalling correlations produced by a v-causal model in the R configuration must thus satisfy both condition (5.2.1) and (5.2.2). Since these form a finite set of linear conditions, they define a polytope in the space of correlations (c.f. Appendix A). We refer to this polytope as the hidden influence polytope associated to R. In order to test whether a v-causal model satisfies the above conditions, we need to know which correlations the model produces in the R configuration. But since B and C are measured simultaneously in R, the correlations P (abcd xyzw) are not directly accessible: | their observation requires perfect synchronization between some of the measurements. Still, given the properties of v-causal models, one can show that some parts of the 4- partite distribution P (abcd xyzw) can be deduced indirectly. | 2Note that signalling could be activated in cases where the model only produces no-signalling corre- lations as well. Indeed, if a marginal probability distribution can have different (no-signalling) values depending on the time chosen by some other party to perform his measurement, in the fashion of [93, 94], this change in the correlation can allow to guess the time of measurement chosen by a distant party. However we don’t consider this possibility here.
48 5.2 The hidden influence polytope
To see this, consider that Charly, in the experiment of Figure 5.2, could perform his measurement at C as planed initially or choose to delay it to C0 (or even to never do it). In any case, since he can in principle make his choice outside of the past v-cone of A, B and D, his choice cannot affect what happens at A, B and D. Thus, the ABD marginal produced by the model must be the same in the R configuration as in the T1 = (A < D < B < C0) configuration. Since correlations in the T1 configuration are easily accessible, the ABD marginal in the R configuration can be determined through measurements in T1. Similarly, one can show that the ACD marginal in the R configuration must equal that in the T2 = (A < D < C < B0) configuration. It is thus also easily accessible, and both the ABD and the ACD marginal in the R configuration can in principle be known. The BC marginal is however clearly inaccessible experimentally since it explicitely requires measurements to be performed simultaneously. These two 3-partite marginals thus constitute the maximum amount of information that one can hope to infer about the R configuration. In order to reach a conclusion without making assumptions on the value of the un- known marginals, we project the hidden influence polytope onto the subspace spanned by the ABD and ACD marginals. This allows to deduce the conditions that are satisfied by the v-causal model in the situation R, in terms of the known marginals only. Note that whenever an inequality satisfied by this projected polytope is violated, one of the two conditions (5.2.1) or (5.2.2) must be violated as well. Since (5.2.1) cannot be violated in the R configuration, by the definition of v-causality, the model must violate condition (5.2.2) in this configuration, i.e. produce signalling correlations. Using techniques described in Appendix A, we could find several inequalities of the projected hidden influence polytope in configuration R when all parties use binary inputs and outputs. We present one of them below.
5.2.1 Quantum violation and faster-than-light communication The following inequality is satisfied by all no-signalling correlations produced by a v-causal model in the R configuration (c.f. Figure 5.2):
S = 3 A B B C 3 D − h 0i − h 0i − h 1i − h 0i − h 0i A B A B + A C − h 1 0i − h 1 1i h 0 0i + 2 A C + A D + B D h 1 0i h 0 0i h 0 1i B1D1 C0D0 2 C1D1 − h i − h i − h i (5.2.3) + A B D + A B D + A B D h 0 0 0i h 0 0 1i h 0 1 0i A B D A B D A B D − h 0 1 1i − h 1 0 0i − h 1 1 0i + A C D + 2 A C D 2 A C D h 0 0 0i h 1 0 0i − h 0 1 1i 7, ≤ where A = P (0 x) P (1 x), A B = ( 1)a+bP (ab xy) and so on. h xi A | − A | h x yi ab − AB | Recall that by construction this inequality only involves correlations that are easily P accessible through some experiment. A quantum v-causal model can thus reproduce any value of S that is achievable with quantum correlations. Interestingly, this inequality can be violated by measuring a 4-qubit state (c.f. paper [M]). We can thus deduce that the corresponding quantum v-causal model must produce signalling correlations in the R configuration.
49 Finite-speed hidden influences
time
A0
D0
B C D
space A Figure 5.3: In the configuration of Figure 5.2, letting the parties B and C broadcast (at light-speed) their measurement results allows one to evaluate the marginal correlations BCD at the point D0, which lies outside of the future light-cone of A (shaded area). If this marginal depends on Alice’s input, it can thus be used for superluminal communication from A to D0. Similarly, if the ABC marginal correlations depend on the measurement w made at D, superluminal communication is possible from D to the point A0.
Thanks to the geometry of this configuration, any signalling obtained in the correla- tions can be used to communicate faster than light as soon as v > c. Indeed, by definition of v-causal models, signalling can neither happen from B to ACD, nor from C to ABD, which lie in the past of C. It must thus happen either from A to BCD or from D to ABC. In both cases, this signalling in the correlations can be used to send signals faster than light (see Figure 5.3). Finite-speed v-causal models for quantum correlations can thus be used to communi- cate faster than light.
5.3 Experimental perspectives
By construction, inequality (5.2.3) can be evaluated without requiring perfect synchro- nization between any of the four parties measured. It thus opens the possibility to test v-causal models experimentally in a way that is independent of the speed v, unlike prece- dent approaches. Namely, an experimental violation of equation (5.2.3) would allow to conclude that if a v-causal model is responsible for the observed correlations, then it must also allow to communicate faster than light in some situations. Note that this conclusion is also valid if other systems than the four of interest happen to be measured during the experiment, even if this makes some measurements happen simultaneously in the privileged reference frame. One way to evaluate the quantity S experimentally could be by performing measure- ments in the T1 and T2 configurations of Figure 5.2. This is possible in principle if one knows how the preferred reference frame moves with respect to earth. A simpler demonstration of the violation of S could also be performed without closing the locality loophole, as is common in many Bell experiments. Indeed, by performing measurements in a time-like manner, it is easy to ensure that the measurements are performed according to the T1 and T2 orders. However, this experiment would not be as strong as the previous ones as it would rely on a proper shielding of the measured systems
50 5.4 Conclusion in order to ensure that no communication between them happened by an exchange of some physical systems. Also, it would not allow to conclude directly (i.e. without invoking further assumptions) about the possibility to communicate faster-than-light. Rather, it would allow to conclude that slower-than-light signalling (communication without a physical support) is possible.
5.4 Conclusion
Thanks to inequality (5.2.3), we proved that the nonlocality of quantum correlations cannot be explained by superluminal finite-speed causal influences without opening the possibility to communicate faster than light. If one rejects this possibility, then one should also reject v-causal models as an attempt to keep a form of locality in causal explanations of quantum correlations. Moreover, we argued that extraordinary synchronization between measurements is not necessary in order to reach this conclusion, for all v-causal model. This contrasts with previous approaches to v-causal models which could only test models with a speed v limited by technological constraints. It thus opens the possibility for new experimental approach to these models.
51 Finite-speed hidden influences
52 Conclusion and outlook
In this thesis we presented several studies related to correlations in the context of quantum physics. First, we focused on general properties of correlations, the most important of which being the nonlocality of quantum correlations, i.e. the ability for results obtained upon measurement of a quantum system to violate a Bell inequality. While nonlocality has attracted a lot of attention in the bipartite case, our results indicate that the situation changes dramatically when a third party is considered. Indeed, several of the results presented here don’t have a bipartite analogue: it is impossible in the bipartite case to deduce that some global correlations are nonlocal by only studying their marginals (c.f. section 2.3), no tight Bell inequality for two parties is known to be impossible to violate with quantum correlations (c.f. section 2.5), and bipartite Bell experiments can only test v-causal models with a bounded speed v (c.f. chapter 5). This seems to indicate that much is still to be discovered in multipartite systems. For instance, it would be interesting to explore further the role that the relation between different marginals of a system can play. Also, since several results obtained in the bipartite case don’t extend straightfor- wardly to more parties, it could be interesting to look at these in more detail. Not much is known for instance about the possibility to simulate entangled states with classical re- sources in multipartite scenarios. Also, some physical principles like information causality, which generalizes the no-signalling principle to situations in which physical supports with bounded capacity are allowed to carry information, have not yet found a good way to be expressed in multipartite scenarios [8, 96, 97]. Further investigation on these topics can give hints as to whether the difficulty encountered here is simply technical or whether a more fundamental reasons is responsible for them. At a more technical level, given the important role played by polytopes in the charac- terization of correlations, the development of new tools to work with them would be very helpful. For instance, better ways to deal with symmetries of polytopes are highly desir- able. Indeed, symmetries typically induce a high level of redundancy in the description of polytopes, which makes several tasks on them highly inefficient.
In this thesis we also showed how working with correlations can provide robust conclu- sions in practical situations subject to uncertainties. The device-independent assumptions are indeed weak enough to potentially allow for their implementation in practice, and yet strong enough to allow for the demonstration of interesting results, like the existence of genuinely multipartite entanglement. In other words, the ability to properly separate subsystems under study, and to identify their different possible inputs and outputs can be sufficient to obtain significant results. However, experimental systems need not always meet these requirements. For in- stance, ions sharing the same trap can be hard to address individually, leading to an imperfect separation between subsystems (c.f. section 3.3). Other systems, like super-
53 Conclusion and outlook conducting qubits [98, 99], are subject to similar limitations. While this mismatch with the working assumptions could be seen as invalidating any possible conclusion, it also seems natural to expect small amounts of cross-talks between subsystems to have limited consequences. A proper way to estimate these cross-talks as well as a careful analysis of their possible impact would be welcome as it would allow one to easily apply the device- independent approach to many practical systems. However it remains to be found.
Finally, we also used correlations in this thesis as tools to study fundamental properties of nature. In particular, following Bell, we questioned the emergence of quantum nonlocal correlations in space-time. What our result suggests is that instead of asking how faster- than-light causal influences can coexist with the theory of relativity, we might just have to wonder about how infinitely-fast causal influences are at all compatible with relativity. Making the meaning of any of these questions more precise would already be a significative step forward.
54 Acknowledgements
First of all, I would like to thank Prof. Nicolas Gisin without which none of the work presented here would have been possible. I am very grateful for the opportunity he gave me to join his group, as well as for his availability for discussions, and in general for his constant support.
A long time ago, Cyril Branciard, Nicolas Brunner and Christoph Simon accompanied my first steps in the field of quantum information. Thank you!
I owe Stefano Pironio a great deal for all he gave me, from an ounce of mathematical rigor to advices on belgium chocolate.
Thanks to Yeong-Cherng Liang who has always been of great support to me, and whose complementary point of view on many subjects I very much appreciated.
My thanks also go to Nicolas Sangouard for all these ‘short’ discussions...
It was both a privilege and a pleasure to work with the Innsbruck team. Thanks a lot to Prof. Rainer Blatt for making this possible, and to Julio Barreiro for the correspondences.
I’m very grateful to Enrico Pomarico for sharing with me concerns that an physicist can face during an experiment, and for his company during conferences which I enjoyed a lot.
Thank you Bruno Sanguinetti for this marvelous time in Prague.
I am also thankful to Antonio Ac´ın who invited me several times to Castelldefels, and who suggested subjects to work on with some of his coworkers. Thanks also to Mafalda Almeida and Lars Wurflinger¨ for the nice collaborations.
I would also like to thank Tam´asV´ertesi for the work we did together, as well as the many other visitors that came to the GAP for a day or more of exchange.
Thank you Clara for your cheerfulness ;-)
Thanks also to Michael Afzelius, Denis Rosset, Charles Lim Ci Wen, Tomy Barnea, Basile Grandjean, Pavel Sekatski, Raphael Ferretti-Sch¨obitz, Markus Jakobi, Keimpe Nevenzeel and all members of the GAP which I had the chance to meet; thanks for your friendliness.
Finally, I am very grateful to my friends and family for their support. Thank you!
55 Acknowledgements
56 Bibliography
[1] M. Arndt, O. Nairz, J. Vos-Andreae, C. Keller, G. van der Zouw and A. Zeilinger, Nature 401, 680-682 (1999).
[2] M. F. Pusey, J. Barrett and T. Rudolph, Nature Phys. 8, 476 (2012).
[3] M. J. W. Hall, arXiv:1111.6304
[4] R. Colbeck and R. Renner, Phys. Rev. Lett. 108, 150402 (2012).
[5] H. F. Hofman, arXiv:1112.2446
[6] J. S. Bell, Speakable and unspeakable in quantum mechanics (Cambridge University Press, Cambridge, 1987).
[7] Ll. Masanes, S. Pironio and A. Ac´ın, Nature Comm. 2, 238 (2011).
[8] M. Paw lowski, T. Paterek, D. Kaszlikowski, V. Scarani, A. Winter and M. Zukowski,˙ Nature 461, 1101-1104 (2009).
[9] T. Mauldin, Quantum Non-Locality and Relativity: Metaphysical Intimations of Modern Physics (Blackwell Publishers, Oxford, 2002).
[10] N. Gisin, Phys. Lett. A 242, (1998).
[11] J. Bae, W.-Y. Hwang and Y.-D. Han, Phys. Rev. Lett. 107, 170403 (2011).
[12] J. S. Bell, La nouvelle cuisine, Speakable and unspeakable in quantum mechanics, second edition (Cambridge University Press, 2004).
[13] T. Norsen, arXiv:0707.0401
[14] N. Gisin and B. Gisin, Phys. Lett. A 260, 323 (1999).
[15] J. F. Clauser, M. A. Horne, A. Shimony and R. A. Holt, Phys. Rev. Lett. 23, 880 (1969).
[16] H. Nha and H. J. Carmichael, Phys. Rev. Lett. 93, 020401 (2004).
[17] N. Brunner and N. Gisin, Phys. Lett. A 372, 3162 (2008).
[18] D. Avis, H. Imai and T. Ito, J. Phys. A: Math. and Gen. 39, 11283 (2006).
[19] R. F. Werner, Phys. Rev. A 40, 4277 (1989).
[20] I. Pitowsky, Brit. J. Phil. Sci. 45, 95 (1994).
[21] M. Navascu´es, S. Pironio and A. Ac´ın, New J. Phys. 10, 073013 (2008).
57 BIBLIOGRAPHY
[22] Y.-C. Liang and A. C. Doherty, Phys. Rev. A 75, 042103 (2007).
[23] F. Henkel, M. Krug, J. Hofmann, W. Rosenfeld, M. Weber and H. Weinfurter, Phys. Rev. Lett. 105, 253001 (2010).
[24] A. Cabello and J.-øA. Larsson, Phys. Rev. Lett. 98, 220402 (2007).
[25] D. Cavalcanti, N. Brunner, P. Skrzypczyk, A. Salles and V. Scarani, Phys. Rev. A 84, 022105 (2011).
[26] M. Anderlini, P. J. Lee, B. L. Brown, J. Sebby-Strabley, W. D. Phillips and J. V. Porto, Nature 448, 452-456 (2007).
[27] M. Navascu´es and H. Wunderlich, Proc. Roy. Soc. Lond. A 466, 881-890 (2009).
[28] P. Trojek, C. Schmid, M. Bourennane, H. Weinfurter and C. Kurtsiefer, Opt. Expr. 12, 276 (2004).
[29] A. Aspect, http://arxiv.org/abs/quant-ph/0402001
[30] D. Collins and N. Gisin, J. Phys. A: Math. and Gen. 37, 1775 (2004).
[31] S. Pironio, A. Ac´ın, S. Massar, A. Boyer de la Giroday, D. N. Matsukevich, P. Maunz, S. Olmschenk, D. Hayes, L. Luo, T. A. Manning and C. Monroe, Nature 464, 1021- 1024 (2010).
[32] S. Wehner, Phys. Rev. A 73, 022110 (2006).
[33] A. Elitzur, S. Popescu and D. Rohrlich, Phys. Lett. A 162, 25 (1992).
[34] L. Aolita, R. Gallego, A. Ac´ın, A. Chiuri, G. Vallone, P. Mataloni and A. Cabello, Phys. Rev. A 85, 032107 (2012).
[35]O.G uhne¨ and G. T´oth, Physics Reports 474, 1 (2009).
[36] G. Svetlichny, Phys. Rev. D 35, 3066 (1987).
[37] R. Gallego, L. E. Wurflinger,¨ A. Ac´ınand M. Navascu´es, arXiv:1112.2647
[38] A. Ac´ın, N. Gisin and Ll. Masanes, Phys. Rev. Lett. 97, 120405 (2006).
[39] I. Pitowski, Quantum Probability – Quantum Logic (Springer, Berlin, 1989).
[40] R. F. Werner and M. M. Wolf, Phys. Rev. A 64, 032112 (2001).
[41]M. Zukowski˙ and C.ˇ Brukner, Phys. Rev. Lett. 88, 210401 (2002).
[42] P. M. Pearle, Phys. Rev. D 2, 1418 (1970); S. L. Braunstein and C. M. Caves, Ann. Phys. (N.Y.) 202, 22 (1990).
[43] D. Collins, N. Gisin, N. Linden, S. Massar S. Popescu, Phys. Rev. Lett. 88, 040404 (2002).
[44] D. Kaszilkowski, L. C. Kwek, J.-L. Chen, M. Zukowski˙ and C. H. Oh, Phys. Rev. A 65, 032118 (2002).
[45] J. Barrett, A. Kent and S. Pironio, Phys. Rev. Lett. 97, 170409 (2006).
58 BIBLIOGRAPHY
[46] D. Collins, N. Gisin, S. Popescu, D. Roberts and V. Scarani, Phys. Rev. Lett 88, 170405 (2002).
[47] M. Seevinck and G. Svetlichny, Phys. Rev. Lett, 89, 060401 (2002).
[48] J.-L. Chen, D.-L. Deng, H.-Y. Su, C. Wu and C. H. Oh, Phys. Rev. A 83, 022316 (2011).
[49] L. Aolita, R. Gallego, A. Cabello and A. Ac´ın, Phys. Rev. Lett. 108, 100401 (2012).
[50] N. D. Mermin, Phys. Rev. Lett. 65, 1838 (1990); S. M. Roy and V. Singh, Phys. Rev. Lett. 67, 2761 (1991); M. Ardehali, Phys. Rev. A 46, 5375 (1992); A. V. Belinski˘i and D. N. Klyshko, Phys. Usp. 36, 363 (1993); N. Gisin and H. Bechmann-Pasquinucci, Phys. Lett. A 246, 1 (1998).
[51] R. F. Werner and M. M. Wolf, Phys. Rev. A 61, 062102 (2000).
[52] A. Klyachko, quant-ph/0409113 (2004).
[53] J. Barrett, N. Linden, S. Massar, S. Pironio, S. Popescu and D. Roberts, Phys. Rev. A 71, 022101 (2005).
[54] R. Horodecki, P. Horodecki and M. Horodecki, Phys. Lett. A 200, 340 (1995).
[55] B. F. Toner and D. Bacon, Phys. Rev. Lett. 91, 187904 (2003).
[56] N. J. Cerf, N. Gisin, S. Massar and S. Popescu, Phys. Rev. Lett. 94, 220403 (2005).
[57] S. Popescu and D. Rohrlich, Found. Phys. 24, 379 (1994).
[58] A. C. C. Yao, in 23rd Annual Symposium on Foundations of Computer Science, Chicago, (IEEE, New York, 1982), p. 160.
[59] C. Branciard and N. Gisin, Phys. Rev. Lett. 107, 020401 (2011).
[60]C. Sliwa,´ Phys. Lett. A 317, 165 (2003).
[61] T. Fritz, arXiv:1202.0141
[62] B. S. Cirel’son, Lett. Math. Phys. 4, 93 (1980).
[63] G. Brassard, H. Buhrman, N. Linden, A. A. M´ethot, A. Tapp and F. Unger, Phys. Rev. Lett. 96, 250401 (2006).
[64] G. M. D’Ariano, M. De Laurentis, M. G. A. Paris, A. Porzio and S. Solimeno, J. Opt. B: Quantum Semiclass. Opt. 4, S127 (2002).
[65] Z. Hradil, J. Reh´e˘cek,˘ J. Fiur´a˘sek and M. Je˘zek, Lecture Notes in Physics: Quantum State Estimation, edited by M. G. A. Paris and J. Reh´e˘cek(Springer-Verlag,˘ Berlin Heidelberg, 2004), pp. 59-112.
[66] Matthias Christandl and Renato Renner, arXiv:1108.5329
[67] Robin Blume-Kohout, arXiv:1202.5270
[68] D. Rosset, private communication.
59 BIBLIOGRAPHY
[69]O.G uhne,¨ C.-Y. Lu, W.-B. Gao and J.-W. Pan, Phys. Rev. A 76, 030305(R) (2007).
[70] J. L. Cereceda, Phys. Rev. A 66, 024102 (2002).
[71] K. F. Pal and T. V´ertesi, Phys. Rev. A 83, 062123 (2011).
[72] F. Schmidt-Kaler, H. H¨affner, S. Gulde, R. Riebe, G. P. T. Lancaster, T. Deuschle, C. Becher, W. H¨ansel, J. Eschner, C. F. Roos and R. Blatt, Appl. Phys. B 77, 789-796 (2003).
[73] G. Kirchmair, J. Benhelm, F. Z¨ahringer, R. Gerritsma, C. F. Roos and R. Blatt, New J. Phys. 11, 023002 (2009).
[74] T. Monz, P. Schindler, J. T. Barreiro, M. Chwalla, D. Nigg, W. A. Coish, M. Har- lander, W. H¨ansel, M, Hennrich and R. Blatt, Phys. Rev. Lett. 106, 130506 (2011).
[75] J. Kilian, Proc. 20th STOC (ACM, New York, 1988), p. 20.
[76] B. Chor, O. Goldreich, E. Kushilevitz and M. Sudan, FOCS ’95: Proceedings of the 36th Annual Symposium on Foundations of Computer Science, p. 41 (1995).
[77] E. Kushilevitz and R. Ostrovsky, FOCS ’97. Proceedings of the 38th Annual Sym- posium on Foundations of Computer Science, p. 364 (1997).
[78] V. Giovannetti, S. Lloyd and L. Maccone, Phys. Rev. Lett. 100, 230502 (2008).
[79] V. Scarani, A. Ac´ın, G. Ribordy and N. Gisin, Phys. Rev. Lett. 92, 057901 (2004).
[80] J. Mueller-Quade and R. Renner, New J. Phys. 11, 085006 (2009).
[81] C. Clausen, I. Usmani, F. Bussi`eres, N. Sangouard, M. Afzelius, H. de Riedmatten and N. Gisin, Nature 469, 508 (2011).
[82] E. Saglamyurek, N. Sinclair, J. Jin, J. A. Slater, D. Oblak, F. Bussi`eres, M. George, R. Ricken, W. Sohler and W. Tittel, Nature 469, 512 (2011).
[83] H. P. Specht, C. N¨olleke, A. Reiserer, M. Uphoff, E. Figueroa, S. Ritter and G. Rempe, Nature 473, 190 (2011).
[84] B. Julsgaard, J. Sherson, J. I. Cirac, J. Fiur´a˘sek and E. S. Polzik, Nature 432, 482 (2004).
[85]N.L utkenhaus,¨ Phys. Rev. A 54, 97 (1996).
[86] A. Bocquet, R. All´eaume and A. Leverrier, J. Phys. A: Math. Theor. 45, 025305 (2012).
[87] D. Bruss, Phys. Rev. Lett. 81, 3018 (1998).
[88] U. Maurer, IEEE Trans. Inf. Theory 39, 733 (1993).
[89] I. Csisz´arand J. K¨orner, IEEE Trans. Inf. Theory 24, 339 (1978); R. Ahlswede, I. Csisz´ar, IEEE Trans. Inf. Theory 39, 1121 (1993).
[90] H.-K. Lo, quant-ph/0102138.
[91] D. Salart, A. Baas, C. Branciard, N. Gisin and H. Zbinden, Nature 454, 861 (2008).
60 BIBLIOGRAPHY
[92] B. Cocciaro, S. Faetti and L. Fronzoni, Phys. Lett. A 375, 379 (2011).
[93] V. Scarani and N. Gisin, Phys. Lett. A 295, 167 (2002).
[94] V. Scarani and N. Gisin, Braz. J. Phys. 35, 2A (2005).
[95] S. Coretti, E. H¨anggi and S. Wolf, Phys. Rev. Lett. 107, 100402 (2011).
[96] R. Gallego, L. E Wurflinger,¨ A. Ac´ın and M. Navascu´es,Phys. Rev. Lett. 107, 210403 (2011).
[97] L.-Y. Hsu, Phys. Rev. A 85, 032115 (2012).
[98] M. Neeley, R. C. Bialczak, M. Lenander, E. Lucero, M. Mariantoni, A. D. O’Connell, D. Sank, H. Wang, M. Weides, J. Wenner, Y. Yin, T. Yamamoto, A. N. Cleland and J. M. Martinis, Nature 467, 570 (2010).
[99] L. DiCarlo, M. D. Reed, L. Sun, B. R. Johnson, J. M. Chow, J. M. Gambetta, L. Frunzio, S. M. Girvin, M. H. Devoret and R. J. Schoelkopf, Nature 467, 574 (2010).
[100] G. M. Ziegler, Lectures on Polytopes, Graduate Texts in Mathematics 152 (Springer-Verlag New York Berlin Heidelberg, Revised edition, 1998).
[101] D. Bremner, On the complexity of vertex and facet enumeration for convex poly- topes, PhD thesis (1998).
[102] http://cgm.cs.mcgill.ca/˜avis/C/lrs.html
[103] http://www.ifor.math.ethz.ch/˜fukuda/cdd home/index.html
[104] http://www.uic.unn.ru/˜zny/skeleton/
[105] http://typo.zib.de/opt-long projects/Software/Porta/
[106] http://www-control.eng.cam.ac.uk/˜cnj22/research/projection.html
[107] Nikolai Yu. Zolotykh, private communication.
61 BIBLIOGRAPHY
62 Appendix A
Polytopes
A.1 Definition and terminology
d A polytope R is the convex hull of a finite number of points Vi = (Vi,1,Vi,2,...,Vi,d) P ⊂ ∈ Rd d = x = (x1, x2, . . . , xd) R s.t. x = qiVi, qi 0 (A.1.1) P { ∈ ≥ } Xi where we denote by x = (1, x1, x2, . . . , xd) the points x completed by an extra component to fit in a space of dimension d + 1 for convenience [100]. In general, several sets of points V can describe the same polytope through { i}i P (A.1.1). For instance, if a point V is not an extremal point of , i.e. if q 0 such that i P ∃ i ≥ Vj = i=j qiVi for some j, then Vi i=j describes the same polytope . On the other 6 { } 6 P hand, if Vj is an extremal point of , then no set of points V Vj can describe the P P { i0} 6⊃ same polytope . The description of a polytope through (A.1.1) is thus minimal when P all point in V are extremal points of the polytope, i.e. vertices. We refer to this as the { i}i extremal points description of a polytope, or V -representation. Notice that the condition for extremality of a point V is linear. The minimal set V can thus be found from i { i}i,min V with the help of linear programming. { i}i The dimension of a polytope dim( ) is given by the dimension of the smallest vector P space that contains . It can be computed from the rank of its extremal points as P rk(V ) if q 0 s.t. q V = (1, 0, 0,..., 0) dim( ) = ij ∃ i ≥ i i i (A.1.2) P (rk(Vij) 1 else. − P The main theorem on polytopes [100] tells that any polytope can also be described P as the intersection of finitely many half-spaces x H H , namely as: j j jk ≥ − 0,k
d P d = x R s.t. xjHj,k 0 k (A.1.3) P { ∈ ≥ ∀ } Xj=0 As in the extremal point description of a polytope (A.1.1), the half-spaces descrip- tion of a polytope can be made unique and minimal by requiring its inequalities to be irredundant, i.e. such that no qk 0 can satisfy Hj,k = qk Hj,k . ≥ k0=k 0 0 When an inequality is irredundant, it is called a facet of the6 polytope. Its intersection P with is then of dimension dim( ) 1. An inequality satisfied by the polytope which P P − is not a facet might still have a non-null intersection with the polytope. The intersection
63 Polytopes of this inequality with the polytope is a often called a face of the polytope, and has a dimension strictly less than dim( ) 1. P − A polytope can thus be described in two equivalent ways (A.1.1), (A.1.3). Transform- ing one representation of a polytope into its dual one is in general a difficult task [101]. Nevertheless, when the polytopes are not too complicated, it can be possible to perform this transformation exactly with the aid of a computer. Several open-source softwares are available for this, like lrs [102], cdd [103], skeleton [104] or porta [105].
A.2 Some operations on polytopes
Polytopes can be manipulated in several ways. Here we describe some of these operations. See also the appendix of [61] for more examples.
A.2.1 Projection
One way to reduce the dimensionality of a polytope is to project it onto a subspace S Rd. For this, consider the linear projection operator Π : Rd S. Without loss of ⊂ →1 generality, Π can be taken to act as Π(x) = (x1, x2, . . . , xs, 0,..., 0) , where s = dim(S) is the dimension of the projected space. The projection of a polytope described in terms of extremal points is easily computed by projecting its extremal points:
d 0 = Π( ) = x R s.t. x = qiΠ(Vi), qi 0 . (A.2.1) P P { ∈ ≥ } Xi Note that all projected vertices Π(V ) need not be extremal points of the projected { i }i polytope anymore. However all extremal points of are necessarily projections of some P0 extremal points of the original polytope. They are thus necessarily contained in the set Π(V ) . Projection of a polytope can thus only reduce the number of its extremal { i }i vertices. When a polytope is specified in terms of half-spaces, finding the H-representation of its projection is more difficult. The Fourier-Motzkin algorithm achieves this without requiring to first transform the description of the polytope into its V-form, but it becomes quickly unpractical for larger problems because of its double-exponential computational complexity.2 Still, if one is not interested in the full set of inequalities describing the projected polytope , a number of its facets can be found heuristically. Here we describe two P0 linear programs that can be used for this. The first linear program for finding facets of a projected polytope is similar to the shooting oracle described in [106]. The idea is, starting from a point that belongs to the interior of , to travel as far as possible in one direction of the subspace S, until touching P0 the boundary of . The point then reached must generically belong to a facet of . P0 P0
Find a facet of a projected polytope. Let x0 belong to the interior of , and P0 1A change of variables can be performed on order to let Π take this form if necessary. 2Note that other algorithms have been proposed, such as the ESP one [106] which is sensitive to the number of facets in the projected polytope 0 rather than in the number of facets of the full polytope . P P
64 A.2 Some operations on polytopes
let v S be a direction in the projected subspace. The linear program ∈ s 0 min Ho,kzk + xj Hj,kzk zk Xk Xk Xj=1 s subject to v H z = 1 j j,k k − (A.2.2) Xj=1 Xk H z = 0 j = s + 1, . . . , d j,k k ∀ Xk z 0 k ≥ provides the inequality s I + x I 0 (A.2.3) 0 j j ≥ Xj=1 s 0 with coefficients I0 = k Ho,kzk + k j=1 xj Hj,kzk, Ij = k Hj,kzk which is satisfied by all points x belonging to the projection of the polytope . To P P P P P0 see this one can check that the dual of this program is
max µ µ,ys+1,...,yd s d (A.2.4) subject to H + (x0 + µv )H + y H 0 k 0,k j j j,k j j,k ≥ ∀ Xj=1 j=Xs+1 which computes the largest value of µ such that x0 + µv . ∈ P0
In order to find a facet with the above linear program, one needs to choose a direction v S. Choosing an interesting direction is not necessarily obvious when nothing or just ∈ little about the projected polytope is known. Here is a linear program which can provide an interesting direction v to look for a facet when some facets of the projected polytope are known. The idea is that an interesting direction to look for a new facet of the P0 projected polytope is in the direction of a vertex of the polytope described by the known facets of this polytope: if there is a difference between the known polytope and , then P0 some of its extremal points need to be outside , otherwise the two polytopes are equal P0 and there is nothing left to be found.
Find an extremal point of a polytope. Let Hk k be a set of inequalities defining d { } a polytope 00, and let w R be a direction in space. The following linear P ∈ program yields a point x which lies on the boundary of : P00
min wjxj xj j X (A.2.5) subject to H + H x 0 k 0k jk j ≥ ∀ Xj If w is chosen at random, the point x is generically an extremal point of . P00
65 Polytopes
Application to polytope representation conversion It is possible to find the H-representation of a polytope from its V -representation by performing a polytope slice followed by a polytope projection. Indeed, the conditions in equation (A.1.1) can be understood as defining a polytope ˜ Rd+n, where n is the P ∈ number of extremal points of , and ˜ is defined by the following H-representation: P P ˜ d+n = (x, q) R s.t. x = qiVi, qi 0 (A.2.6) P { ∈ ≥ } Xi ˜ d Elimination of the variables qi, i.e. projection of onto the subspace S = R thus defines P the set of x such that the conditions in (A.1.1) hold. This projected polytope is thus , described in terms of inequalities. This is the basic technique used by the software P porta [105] to solve a polytope.
A.2.2 Slice Another way of reducing the dimensionality of a polytope is to consider a slice of it, that is, the intersection of the polytope with a subspace S Rd of dimension s. S can be P d ⊂ defined by a set of linear equations: S = x R s.t. xjEjl = 0 l = 1 . . . d s . The { ∈ j ∀ − } slice of which belongs to S is easily defined if is specified in the H-representation: P P P d 0 = x R s.t. xjMkj 0 k, xjEjl = 0 l (A.2.7) P { ∈ ≥ ∀ ∀ } Xj Xj Computing the slice of a polytope in its V -representation is more complicated. In fact, on can show through polytope duality [100] that this task is equivalent to a projection of the polytope dual to [107]. It can thus be done with the techniques described in the P precedent section.
A.2.3 Another tasks : finding facets lying under an inequality An inequality satisfied by a polytope is not a facet of the polytope if its dimension is lower than d 1. When this is the case, one can show that the inequality can be expressed as − the convex combination of a number of tighter inequalities, the tighter of which are facets of the polytope, sharing an intersection of dimension d 1 with the polytope. Thus, if a − point violates a non-facet inequality, it can only violate some of these tighter ones by a larger amount. Given a non-facet inequality, it can thus be interesting to look for these tighter facets. A method for finding the facets underlying an inequality given a V -description of the polytope is presented in paper [E]. The idea is that the rank of the set of extremal points saturating the inequality can be augmented by adding more extremal points of the polytope to this set. When the achieved rank is sufficient, and if the hyperplane passing through these points does not cut the polytope into two parts, then it describes a facet of the polytope. By construction, the intersection of this facet with the polytope coincides with the intersection of the original inequality with the polytope since it is generated by the same set of extremal points.
66 Appendix B
Memoryless attack on the 6-state protocol – proof
Here we provide a proof for the bound (4.1.4) used in the main text.
Proof. Without loss of generality, Eve’s POVM elements can be written as:
F = A† A = a 11 + (b a )P (B.0.1) k k k k k − k k
where Ak = √akUkPk + √bkUkPk is a Kraus operator associated to the element Fk, ak, bk 0, Uk is a unitary operator, Pk = 11 Pk and Pk is a one-dimensional projector.≥ − Eve’s information on Alice’s bit. We consider a given run k of the protocol for which Alice and Bob used the same basis b = bk, and denote by A Alice’s bit, B Bob’s bit and E the result of Eve’s POVM measurement. The total information gained by Eve on Alice’s bit after the sifting procedure is given by I(A :(E, b)). Since b is independent of A and E, and since the state produced by Alice ρ = ρ(A, b) is a function of A and b, we have:
I(A :(E, b)) = H(A) + H(E, b) H(A, E, b) − = H(A) + H(E) + H(b) H(A, E, b) − (B.0.2) = H(A, b) + H(E) H(A, E, b) = I((A, b): E) − = I(ρ : E) = log(6) I(ρ E) − | Where H is Shannon’s entropy, and we assumed that the six states are chosen by 1 Alice with the same probability 6 . We thus need to bound the quantity I(ρ E) which expresses the information that Eve is missing after she learns the result of| her measurement, to know which state ρ was prepared by Alice. Using the fact that Prob(E = k) = tr(Fk)/2, this quantity can be expressed as
I(ρ E) = I(E ρ) + log(tr(F )) + log(3) (B.0.3) | | k Xk where I(E ρ) = Prob(A = a, b = s, E = k) log(Prob(E = k A = a, b = s)). | − a,s,k | For every statePρ that Alice can produce, ρ = 11 ρ can also be produced by Alice. An attack described by the POVM elements F −thus provides Eve with the same { k} information as the attack F˜k where F˜k = Ak11 + (bk ak)Pk. Indeed, as shown above, this information is a{ symmetric} function of P (E =−k ρ = ρ(a, b)) and |
tr(ρFk) = ak + (bk ak)tr(ρPk) − (B.0.4) = a + (b a )tr(ρF˜ ) = tr(ρF˜ ). k k − k k k 67 Memoryless attack on the 6-state protocol – proof
Moreover, the information that can be extracted from a mixture of measurements M1 applied with probability p1 and M2 applied with probability p2 is
I([(p1,M1), (p2,M2)]) = p1I(M1) + p2I(M2), (B.0.5) where I(M1,2) is the information that can be extracted by using measurement M1,2 only. We can thus write
1 1 I( F ) = I F F˜ { k} 2 k ∪ 2 k (B.0.6) a + b F F˜ = k k I k , k 2 "(ak + bk ak + bk )# Xk where the factor 1/(ak + bk) is a normalization coefficient. Thus, the information gained by performing an arbitrary POVM measurement can also be achieved by mix- ing measurement strategies consisting of only two POVM elements. Let us thus consider a POVM measurement for Eve consisting only of the two elements Fk ak+bk ˜ and Fk . ak+bk By direct computation one finds that 1 I(E ρ) = (h(c0 ) + h(d0 ) + h(e0 )) (B.0.7) k| 3 k k k
ak+tr(ρ2Pk)(bk ak) ak+tr(ρ4Pk)(bk ak) ak+tr(ρ6Pk)(bk ak) with c0 = − , d0 = − , e0 = − . Since k ak+bk k ak+bk k ak+bk this function is convex in c0, d0, e0, its minimum lies on the boundary of the admissible region
1 2 1 2 1 2 1 tr(ρ P ) + tr(ρ P ) + tr(ρ P ) . (B.0.8) 2 k − 2 2 k − 2 2 k − 2 ≤ 4 More precisely, this is found for tr(ρ P ) 0, 1 , tr(ρ P ) 0, 1 , or tr(ρ P ) 2 k ∈ { } 4 k ∈ { } 6 k ∈ 0, 1 . In this case, since log tr Fk = 0, we find { } ak+bk bk 2 + h a +b I(ρ E ) = k k + log(3) (B.0.9) | k 3
ak where k = . bk All in all, this gives the following bound on Eve’s information about Alice’s bit:
2 + h 1 a + b 1+k I(A :(E, b)) 1 k k . (B.0.10) ≤ − 2 · 3 Xk
Perturbation on Bob’s system. The attack of Eve delivers the state ρi0 = k AkρiAk† to Bob instead of the expected ρi. This creates some errors in the outcomes of Bob, which are measured by the QBER: P
6 1 Q = 1 P (A = i, B = i) = 1 tr(ρ0 E ) (B.0.11) − − 6 i i i=1 i X X where Ei, i = 1,..., 6 are the six possible measurement operators of Bob and
ρ0 = a b U ρ U †+√a (√a b )U P ρ P U †+ b ( b √a )U P ρ P U †. i k k k i k k k− k k k i k k k k− k k k i k k k X p p p p (B.0.12)
68 Note, that the attack A has the same effect as A˜ for A˜ = √a U P + { k} { k} k k k k √bkUkP k. Indeed, for all i there is a j such that ρi = ρj and Ei = Ej, and one can check that:
tr(UkρiU †Ei) = tr(Ukρ U †Ei), k i k (B.0.13) tr(UkPkρiPkUk†Ei) = tr(UkP kρiP kUk†Ei).
So we can assume that both Ak and A˜k are present in the attack. In this case the perturbed state is
2 (√ak √bk) ρ0 = a b U ρ U † + − (U P ρ P U † + U P ρ P U †). (B.0.14) i k k k i k 2 k k i k k k k i k k k X p To bound the second part of this expression, one can show by direct computation that
tr(UkP kρiP kUk†Ei) + tr(UkPkρiPkUk†Ei) i X (B.0.15) = tr(P kρi)tr(P kUk†EiUk) + tr(Pkρi)tr(PkUk†EiUk) i X = 2((1 c)(1 c˜) + cc˜ + (1 d)(1 d˜) + dd˜+ (1 e)(1 e˜) + ee˜) − − − − − − ˜ where c = tr(Pkρ2), d = tr(Pkρ4), e = tr(Pkρ6),c ˜ = tr(UkPkUk†E2), d = tr(UkPkUk†E4) ande ˜ = tr(UkPkUk†E6). The maximum value of (B.0.15) under the constraints 2 2 2 2 2 2 c 1 + d 1 + e 1 1 , c˜ 1 + d˜ 1 + e˜ 1 1 can be − 2 − 2 − 2 ≤ 4 − 2 − 2 − 2 ≤ 4 checked to be 4.
Finally, the first part of (B.0.14) can also be bounded since tr(UkρiUk†Ei) 1. Thus we find that ≤ a + b (1 √ )2 Q k k − k . (B.0.16) ≥ 2 · 1 + k Xk Putting the two bounds together. Let us consider equations (B.0.10) and (B.0.16) together. Keeping the sum ak + bk constant for all k, we choose two values of k if possible: k1 and k2 such that k1 < k2. Following [85] one can show that increasing k1 in such a way that keeps the bound on the QBER (B.0.16) unchanged, can only decrease k2 and increase Eve’s information as given by (B.0.10). It is thus always better to have = k. In this case both bounds become: k ∀ 1 1 h 1+ (1 √)2 I(A :(E,S)) − ,Q − (B.0.17) ≤ 3 ≥ 3(1 + ) which can be summarized as
1 1 3Q(2 3Q) I(A :(E,S)) 1 h − − . (B.0.18) ≤ 3 " − p 2 !#
Tightness of the bound To show that the above bound is tight, consider the attack 1 γ in which Eve uses the two POVM elements Fk = −2 11 + γ k k for k = 0, 1. This gives P (ρ = z k) = (1 ( 1)kγ)/6, P (ρ = x k) = P (ρ|=ih |y k) = 1/6, and so 2 ± | 1 ±1 −γ ± | ± | I(A E) = 3 + log 3 + 3 h −3 . Since I(A) = 1 + log 3, this attack proved Eve with | 1 1 γ a mutual information with Alice of I(A : E) = I(A) I(A E) = 1 h − . − | 3 − 2 1 √1 γ2 − − Moreover, the QBER induced by this attack is Q = 3 . Thus Eve can choose an attack that saturates equation (4.1.4): the bound is tight.
69 Memoryless attack on the 6-state protocol – proof
70 Papers
List of published papers:
[A] Testing a Bell inequality in multipair scenarios, J.-D. Bancal, C. Branciard, N. Brun- ner, N. Gisin, S. Popescu and C. Simon, Phys. Rev. A 78, 062110 (2008).
[B] Quantifying Multipartite Nonlocality, J.-D. Bancal, C. Branciard, N. Gisin and S. Pironio, Phys. Rev. Lett. 103, 090503 (2009).
[C] Simulation of Equatorial von Neumann Measurements on GHZ States Using Non- local Resources, J.-D. Bancal, C. Branciard and N. Gisin, Adv. in Math. Phys., vol. 2010, Article ID 293245 (2010).
[D] Guess Your Neighbor’s Input: A Multipartite Nonlocal Game with No Quantum Ad- vantage, M. L. Almeida, J.-D. Bancal, N. Brunner, A. Ac´ın, N. Gisin and S. Pironio, Phys. Rev. Lett. 104, 230404 (2010).
[E] Looking for symmetric Bell inequalities, J.-D. Bancal, N. Gisin and S. Pironio, J. Phys. A: Math. Theor. 43, 385303 (2010).
[F] Detecting Genuine Multipartite Quantum Nonlocality: A Simple Approach and Gen- eralization to Arbitrary Dimensions, J.-D. Bancal, N. Brunner, N. Gisin and Y.- C. Liang, Phys. Rev. Lett. 106, 020405 (2011).
[G] Extremal correlations of the tripartite no-signaling polytope, S. Pironio, J.-D. Bancal and V. Scarani, J. Phys. A: Math. Theor. 44, 065303 (2011).
[H] Practical private database queries based on a quantum-key-distribution protocol, M. Jakobi, C. Simon, N. Gisin, C. Branciard, J.-D. Bancal, N. Walenta and H. Zbinden, Phys. Rev. A 83, 022301 (2011).
[I] Various quantum nonlocality tests with a commercial two-photon entanglement source, E. Pomarico, J.-D. Bancal, B. Sanguinetti, A. Rochdi and N. Gisin, Phys. Rev. A 83, 052104 (2011).
[J] Device-Independent Witnesses of Genuine Multipartite Entanglement, J.-D. Bancal, N. Gisin, Y.-C. Liang and S. Pironio, Phys. Rev. Lett. 106, 250404 (2011).
[K] Loophole-free Bell test with one atom and less than one photon on average, N. San- gouard, J.-D. Bancal, N. Gisin, W. Rosenfeld, P. Sekatski, M. Weber and H. Wein- furter, Phys. Rev. A 84, 052122 (2011).
[L] A framework for the study of symmetric full-correlation Bell-like inequalities, J.- D. Bancal, C. Branciard, N. Brunner, N. Gisin and Y.-C. Liang, J. Phys. A: Math. Theor 45, 125301 (2012).
71 Papers
Preprints:
[M] Quantum nonlocality based on finite-speed causal influences leads to superluminal signaling, J.-D. Bancal, S. Pironio, A. Ac´ın, Y.-C. Liang, V. Scarani and N. Gisin, arXiv:1110.3795
[N] The definition of multipartite nonlocality, J. Barrett, S. Pironio, J.-D. Bancal and N. Gisin, arXiv:1112.2626
[O] Imperfect measurements settings: implications on quantum state tomography and entanglement witnesses, D. Rosset, R. Ferretti-Sch¨obitz, J.-D. Bancal, N. Gisin and Y.-C. Liang, arXiv:1203.0911
[P] Useful multipartite correlations from useless reduced states, L. E. Wurflinger,¨ J.- D. Bancal, T. V´ertesi, A. Ac´ın and N. Gisin, arXiv:1203.4968
72 Paper A
Testing a Bell inequality in multipair scenarios
J.-D. Bancal, C. Branciard, N. Brunner, N. Gisin, S. Popescu and C. Simon
Physical Review A 78, 062110 (2008)
73
PHYSICAL REVIEW A 78, 062110 ͑2008͒
Testing a Bell inequality in multipair scenarios
Jean-Daniel Bancal,1 Cyril Branciard,1 Nicolas Brunner,1 Nicolas Gisin,1 Sandu Popescu,2,3 and Christoph Simon1 1Group of Applied Physics, University of Geneva, 20 rue de l’Ecole-de-Médecine, CH-1211 Geneva 4, Switzerland 2H.H. Wills Physics Laboratory, University of Bristol, Tyndall Avenue, Bristol BS8 1TL, United Kingdom 3Hewlett-Packard Laboratories, Stoke Gifford, Bristol BS12 6QZ, United Kingdom ͑Received 8 October 2008; published 15 December 2008͒ To date, most efforts to demonstrate quantum nonlocality have concentrated on systems of two ͑or very few͒ particles. It is, however, difficult in many experiments to address individual particles, making it hard to highlight the presence of nonlocality. We show how a natural setup with no access to individual particles allows one to violate the Clauser-Horne-Shimony-Holt inequality with many pairs, including in our analysis effects of noise and losses. We discuss the case of distinguishable and indistinguishable particles. Finally, a comparison of these two situations provides insight into the complex relation between entanglement and nonlocality.
DOI: 10.1103/PhysRevA.78.062110 PACS number͑s͒: 03.65.Ud
I. INTRODUCTION way to tell which particle is entangled with which. The cor- responding loss of entanglement has been derived in Ref. ͓5͔. Entanglement is the resource that allows one to establish In the second case, the pairs are indistinguishable; so in some quantum nonlocal correlations ͓1͔. These correlations have sense the information about the pairing is here lost in a co- been the center of a wide interest, because of their fascinat- herent way. ing nature and of their impressive power for processing in- Reid et al. ͓6͔ have considered the case of indistinguish- formation. Experimentally, quantum nonlocality has been able pairs ͑with global measurement͒ in optics. More specifi- demonstrated in so-called Bell experiments, which have to cally, these authors, extending on a previous work of Drum- date all confirmed the quantum predictions ͓2͔. mond ͓7͔, showed how Bell inequalities can be tested ͑and Most theoretical works on Bell experiments and Bell in- violated͒ when many pairs are created via PDC. In this case equalities have focused on the case where the source emits a the pairs are indistinguishable because of the process of single entangled pair of particles at a time. Indeed, this is the stimulated emission. In Ref. ͓8͔, Jones et al. have considered simplest situation to study. From the experimental point of a related scenario; there, entangled pairs are delivered via an view, most experiments have been designed in order to inept delivery service, but at the end only a single pair is match this theoretical model. For example, in photonic ex- measured. Also considering multiparticle entanglement in periments, the source, usually based on parametric down- such scenario is an interesting problem: see, for example, conversion ͑PDC͒, is set in the weak regime; i.e., when the Refs. ͓9,10͔. source emits something, it is most likely a single pair of In this paper we will study the violation of Bell inequali- entangled photons. ties in a general multipair scenario. We start by treating the However, there are experimental situations, such as in case of independent pairs ͑Sec. II͒. We argue that the resis- many-body systems, where producing single entangled pairs tance to noise is here the relevant measure of nonlocality, is rather difficult. For instance, in Ref. ͓3͔ many entangled evaluating it. The consequences of particle losses are also pairs ͑Ӎ104͒ of ultracold atoms have been created, but can- investigated. Next, we move to the case of indistinguishable not be addressed individually. So, while entanglement has pairs ͑Sec. III͒ after a brief review of the results of Ref. ͓6͔, definitely been created in this system, one still lacks an effi- we present an analysis of the influence of noise and losses in cient method for demonstrating its quantum nonlocality this case. In Sec. IV, we compare the entanglement and non- through the violation of some Bell inequality. The goal of the present paper is to discuss techniques for testing Bell in- equalities in such multipair scenarios, where the particles on Alice’s and Bob’s side cannot be individually addressed, and must therefore be measured globally ͑see Fig. 1͒. What we mean here by global measurements is that each particle is submitted to the same measurement. Note that the case of more general measurements ͑collective measurements on all particles͒ has been considered in Ref. ͓4͔. Basically one should distinguish two cases: independent FIG. 1. Setup: a source produces M independent pairs ͑or pairs and indistinguishable pairs. In the first case, the pairs equivalently M independent sources each produce a pair͒, the pair- are created independently, but cannot be addressed individu- ing between Alice’s and Bob’s particles is lost during their trans- ally; therefore, they must be measured globally ͑on both Al- mission, and each party measures all their incoming particles in the ͒ ϩ͑Ϫ͒ ice’s and Bob’s sides . During this global measurement, the same basis. The total number n+͑−͒ of particles detected in the classical information about the pairing is lost: there is no outcome is tallied on both sides.
1050-2947/2008/78͑6͒/062110͑8͒ 062110-1 ©2008 The American Physical Society BANCAL et al. PHYSICAL REVIEW A 78, 062110 ͑2008͒ locality in both cases. This leads us to a surprising result: threshold N=M /2,2M /3,...,M such that the outcome is ജ while the state of indistinguishable pairs contains more en- + iff n+ N. tanglement than the state of independent pairs ͑after the clas- At this point the two relevant questions are the following: sical mixing͒, the latest appears to be more nonlocal. In other first, is it possible to violate the CH inequality with any of words, the incoherent loss of information provides more non- these voting procedures? Second, if yes, which strategy locality, but less entanglement, than the coherent loss of in- yields the largest violation? To address these questions one formation ͑indistinguishable pairs͒. This provides a novel ex- must compute the joint and marginal probabilities entering ample ͑here in the case of multipairs͒ of the complex relation the CH inequality for each procedure. between entanglement and nonlocality. Finally we provide some experimental perspectives ͑Sec. V͒ and conclusions. B. Pure states
II. INDEPENDENT PAIRS Let us first consider the pure entangled states ͉͘ ͉ ͘ ͉ ͘ ͑ ͒ We consider a source emitting M entangled pairs, each of = cos 00 + sin 11 , 4 them being in the same entangled two-qubit state . Thus the ͑ ͒ ͉͉͗͘ ͉⌿ ͘ so that in Eq. 1 , = . We will also write M global state is =͉͘M. For detectors with a perfect efficiency =1, all M = M = ¯ . particles are detected on both Alice’s and Bob’s sides. The M marginal and joint probabilities entering the CH inequality M times ͑1͒ for a vote with given threshold N are
Each pair being independent, Alice and Bob receive M M uncorrelated particles. Since Alice and Bob are unable to M A͑ ͒ ͩ ͪ ͑ ͒n+ ͑ ͒M−n+ ͑ ͒ P+ A = ͚ p+ A p− A , 5 address single particles in their ensemble, they perform a n =N n+ global measurement on their M particles; i.e., all M particles + ͑ are measured in the same basis we shall consider here only n␣ p␣͑A,B͒ ͒ ͑ ͒ ͑ ͒ von Neumann measurements or, equivalently, in the same P++ A,B = M! ͚ ͟ , 6 direction on the Bloch sphere. After the measurement appa- n␣⌶ ␣, n␣! ratus, two detectors count the number of particles, n and n , + − where p ͑A ,B ͒= 1 tr(͑ +A ͒ ͑ +B ͉͉͒͗͘) and in each output mode ͑see Fig. 1͒. If the detectors are per- ++ i j 4 1 i 1 j p ͑A /B ͒ 1 (͑ A ͒ / ͑ B ͉͉͒͗͘) fectly efficient ͑=1͒, one has M =n +n . + i j = 2 tr 1+ i 1 1 1+ j are the quantum + − joint and marginal probabilities for a single pair. Alice and Bob’s outputs are denoted ␣,͕+,−͖, n␣ is the number of A. Testing the CH inequality pairs which gave detections ␣ and , and ⌶=͕n␣ ͉͚ A ജ B ജ ͖ Our goal is to test a Bell inequality. Here we shall focus N+ n␣=M ,n+ =n+++n+− N,n+ =n+++n−+ N is the on the simplest Bell inequality, the Clauser-Horne-Shimony- set of all events yielding the result “ϩϩ” after voting. Holt ͑CHSH͒ inequality ͓11͔, which involves two inputs on Next, one can choose the state ͉͘ and the measured set- ͑ Alice and Bob’s sides, A1, A2 and B1, B2, and two outputs tings. For the maximally entangled state of two qubits ␣,͕+,−͖. For convenience we write it under the CH form =/4͒ one may choose the standard optimal ͑for the case ͒ ͓12͔ M =1 settings for the CH inequality—i.e., A0 = z, A1 = x, x+z −x+z B = ͱ , and B = ͱ . Doing so with majority voting =−PA͑A ͒ − PB͑B ͒ + P ͑A ,B ͒ + P ͑A ,B ͒ 0 2 1 2 ICH + 1 + 1 ++ 1 1 ++ 1 2 ͑N=M /2͒, the CH inequality can be violated for any value ͑ ͒ ͑ ͒ ഛ ͑ ͒ + P++ A2,B1 − P++ A2,B2 0, 2 of M; the maximal amount of violation is numerically found to decrease with the number of emitted pairs as M−1. where P ͑A ,B ͒ is the probability for both Alice and Bob to ++ i j Remarkably, a higher violation is found for different mea- output “ϩ” when performing measurements A and B , re- i j surement settings, given by spectively. Recall that under the hypothesis of no-signaling ͓ ͔ both CH and CHSH inequalities are equivalent 13 . Now, in A0 = z, order to test inequality ͑2͒, Alice and Bob must transform their data, basically n and n , into a binary result “ϩ”or ␣ ␣ + − A1 = sin 2 x + cos 2 z, “Ϫ”. A natural way of doing it is by invoking a voting pro- cedure: for instance, ␣ ␣ B0 = sin x + cos z, ͑ ͒ ജ i Majority voting: if n+ n− “+”; → B = − sin ␣ + cos ␣ . ͑7͒ otherwise “− ”, 1 x z → ␣ϳ −1/2 With 2ͱ2 M , those settings are numerically found to be ͑ ͒ −1/2 ͓ ii Unanimous voting: if n+ = M “+”; optimal. In this case the decrease of CH is only M see → Fig. 2͑a͔͒. The state leading to the largestI violation is always otherwise “− ”, ͑3͒ → the maximally entangled one ͑=/4͒ for majority voting. or any intermediate possibility—for instance, 2/3or3/4 ma- The one-parameter planar settings ͑7͒ were already used jority. For each voting method and given M corresponds a by several authors ͓6,7͔; for example, in Bell experiments
062110-2 TESTING A BELL INEQUALITY IN MULTIPAIR SCENARIOS PHYSICAL REVIEW A 78, 062110 ͑2008͒
noise can be modeled at the level of the source, supposing that the produced pairs are not in the pure state ͉͉͗͘, but instead in a Werner state of the form
1 = w͉͉͗͘ + ͑1−w͒ . ͑8͒ 4
The resistance to noise of a given violation is then defined by the maximal amount ⑀=1−w of white noise that can be added to the pure state ͉͉͗͘ such that the resulting state still violates a Bell inequality ͑CH here͒. Considering now that the sources of Fig. 1 produce the state ͑8͒, we look for the largest value of ⑀ which still gives ͑ ͒ a positive value of CH, using settings of the form 7 and optimizing on the stateI ͑͒. For all voting strategies we find a resistance to noise decreasing like ⑀ϳM−1, the majority voting being still the best choice ͓see Fig. 2͑b͔͒. Unlike when maximizing CH in the absence of noise, here the optimal state is alwaysI the maximally entangled one ͑=/4͒, even for intermediary voting strategies, for which the CH viola- tion with this state decreases exponentially withI M. This shows that appropriate figures of merit need to be used when examining practical situations. These results are encouraging, but just as detectors might not be perfect, maybe the source cannot guarantee an exact number of pairs, M, as needed here. To show that these vio- ͑ ͒ FIG. 2. Color online CH violation and resistance to noise for a lations are relatively robust towards this issue, we now look source producing M independent pairs. The states and settings used ͑ ͒ at the case of sources producing a number of entangled pairs are discussed in the text. a Maximal CH values for various thresh- which follow a Poissonian distribution. olds: majority voting ͑solid red line͒,3/4 voting ͑dotted green line͒, and unanimity ͑dashed blue line͒. The decrease is as M−1/2 for the first two and exponential for the last one. The highest violation is D. Poisson sources thus reached using a majority vote. ͑b͒ Resistance to noise for the different thresholds ͑same colors͒. All curves decrease as M−1. The A Poissonian source produces a state M of M pairs with most resistant violation is that achieved by using majority voting. a Poissonian probability
͑ ͒ M using a state we shall look at 10 later with the unanimous p͑M͒ = e− , ͑9͒ vote for any M. M! Performing numerical optimizations, we also found that a violation can be obtained for any voting strategy with any where is the mean number of photon pairs. With such a number of emitted pairs M ͓see Fig. 2͑a͔͒. We optimized the source, a different number of pairs is created every time. So state ͑͒ and the four measurement settings, each time find- for a chosen voting assignment ͑3͒ the threshold N varies ͑ ͒ ͑ ing optimal settings of the form 7 . For the unanimous vote, with the total number of photons detected, M =n+ +n− we for instance, the optimal state is less and less entangled as the still consider perfect detectors͒, according to each realiza- number of emitted pairs, M, increases, as described in ͓14͔, tion. and the violation decreases exponentially with M. Thus for Using settings of the form ͑7͒, we optimized numerically pure entangled two-qubit states, the largest amount of viola- ␣ and the state ͑͒ for several votes, in a situation where the tion is obtained with majority voting. source is Poissonian. Doing so in order to get the largest CH violation and the highest resistance to noise, we obtained results very similar to that of the fixed M case, verifying in C. Resistance to noise −1/2 particular a decrease of CH as for the majority vote We now compute the resistance to noise that these viola- and of the resistance to noiseI as −1 ͑see Fig. 3͒. Similarly, tions could bear, which is the relevant measure of nonlocality the states yielding the largest CH values are the maximally considering experimental perspectives—the amount of viola- entangled one for the majorityI vote and partially entangled tion being basically just a number, without much significance ones for the two other votes. A difference, however, is that in the present case as we shall see. CH is found to decrease slower than exponentially for the In a practical Einstein-Podolsky-Rosen ͑EPR͒ experiment, Iunanimity vote. Ͼ imperfect detectors, noisy sources, or disturbing channels in- Note also that since it is possible to find CH 0 and the troduce noise in the measurement results. To first order, this probability to get a ϩ result vanishes for I 0, there exist → 062110-3 BANCAL et al. PHYSICAL REVIEW A 78, 062110 ͑2008͒
FIG. 4. ͑Color online͒ CH violation with inefficient detectors as a function of the probability for a photon to be detected. The upper thin curve shows the traditional case M =1 with known critical ef- ficiency =2/3 ͓15͔. The two other curves are for M =5 pairs with majority ͑solid red line͒ and unanimity voting ͑dashed blue line͒. find a Bell violation in such circumstances. Figure 4 shows the maximal CH violations obtained ͑optimizing on states and settings͒ as a function of the detection efficiency for M =1 and M =5 with majority and unanimity voting. The re- quired detector efficiency increases with the number of pairs, leaving no chance to find a violation at high M with Ӷ1. One way to deal with detector inefficiencies consists in post-selecting events in which exactly M photons are de- tected on both sides. In this way, cases in which particles FIG. 3. ͑Color online͒ Maximal Bell violation and resistance to were not detected are neglected and the Bell violation is ͑ noise with a Poissonian source of independent pairs. The red lines recovered independently of the losses. If detectors also have represents the majority voting, the green dotted lines the 3/4 vot- dark counts, noise will appear in the statistics, which can be ing, and the blue dashed lines the unanimity voting. Settings are treated with Werner states as presented in the “resistance-to- ͒ chose in the form ͑7͒; optimal states are discussed in the text. ͑a͒ noise” section. This approach is, however, not perfect as it is subject to the so-called detection loophole ͓16͔: there exist For a large mean photon number, the decrease of CH goes like −1/2 for the majority and 3/4 vote, just like for theI fixed M case. local models, exploiting detectors inefficiencies, that can vio- late a Bell inequality ͓17͔. But also, one needs to know ex- Concerning the unanimity vote, CH decreases faster than a polyno- mial, but slower than an exponential.I ͑b͒ Resistance to noise is very actly the number of pairs, M, created before measuring them. similar for all strategies, decreasing as −1 just like with as source This last condition might not be guaranteed, for example, of fixed pairs number. with Poissonian sources where knowledge of M is often in- ferred from the number of detected particles. Ӎ To estimate the impact of losses, we consider the case in an optimal 1.2–1.8 yielding a maximum CH violation. which exactly 1 of the M photons flying to Alice and 1 going But this feature is not found in the resistance to noise. to Bob are not detected. As the number of created pairs in- creases, this is a situation that must happen frequently even E. Inefficient detectors with very efficient detectors. Using the majority and unanim- ity vote in this situation, we numerically verified that the CH We now consider detectors with finite efficiency Ͻ1 inequality could not be violated, at least for M ഛ50. and look in what circumstance a Bell violation can still be A way to understand this result is by noting that the sets observed in a multipair scheme with such detectors. is to of events yielding results ϩ and Ϫ are separated by only one be understood here as the probability for a particle to be photon number. Thus, removing one photon mixes the two detected. sets. It should thus be advantageous to separate these two In general, in the presence of detector inefficiencies ͑or ജ ജ cases such that, for instance, n+ N +, n− N −, M −N particle losses͒ the total number of particles detected by Al- Ͻn ϽN . Using this particular post-selection,→ → we could ͑ A A B B͒ + → ice and Bob are different n+ +n− n+ +n− . Thus, for a given find a Bell violation in the case of one photon loss on both voting strategy, the thresholds N applied by Alice and Bob sides, with N=M −1 ͑unanimity voting͒, starting at M =5. might be different for the same event, since it depends on the For details on this post-selection, see Ref. ͓14͔. total number of photons detected by each party. Testing a Bell inequality in this situation without appealing to post- III. INDISTINGUISHABLE PHOTONS selection introduces no detection loophole, but it is not a In the first part of this work we showed how, using mul- surprise to find that high efficiencies are needed in order to tiple independent pairs together with independent global
062110-4 TESTING A BELL INEQUALITY IN MULTIPAIR SCENARIOS PHYSICAL REVIEW A 78, 062110 ͑2008͒
measurement on all the photons produced, one could find a substantial CH violation, even in the presence of lots of pairs. But how good is this compared to a source producing the 2M photons altogether? For the sake of comparison we now consider a specific example, commonly produced in many laboratories. By the same occasion it will uncover some aspects of the relation between entanglement and non- locality. The state we are discussing now can be written as M ͉⌽ ͗͘⌽ ͉ = M M with 1 ͉⌽ ͘ ͑ † † † †͒M͉ ͘ ͑ ͒ M = a0b0 + a1b1 0 , 10 M!ͱM +1
where a0 and a1 are orthogonal modes on Alice’s side and b0 ͑ and b1 orthogonal modes on Bob’s side for instance, hori- zontal and vertical polarization modes͒. A way to produce this state is with a PDC source, which gives a Poissonian distribution of such states. The same global measurements as previously performed on M photons can be realized here by just using the same setup as before: a polarizer followed by two photon counters on each side ͑same setup as represented in Fig. 1, but with a different source͒. Considering the state ͑10͒, we make a similar analysis as previously, briefly reviewing the results of ͓6,7͔ for the amount of violation achievable and presenting our own analysis for the resistance to noise. We computed the new probabilities entering the CH ex- pression for this specific state and, choosing variousI voting procedures, numerically optimized the settings according to FIG. 5. ͑Color online͒ Comparison between sources producing ␣, Eqs. ͑7͒, in order to get the largest violation. Surpris- independent pairs or indistinguishable photons, using settings of the CH ͑ ͒ ͑ ͒ ingly, for any number of photons, M,I all voting procedures form 7 . a Maximal CH violation achieved with a source of in- distinguishable photons for various voting procedure ͑superposed yield approximately the same maximum violation of CH, −1 ͓ ͑ ͔͒ I black dots͒. Compared to the previously calculated violations for decreasing as M see Fig. 5 a . This is even more surpris- ͓ ͑ ͔͒ ing as the settings needed for that are not the same for all independent pairs same curves as in Fig. 2 a , unanimity voting ͑lower blue dashed line͒ yields less violation, while majority voting voting methods. ͑upper red line͒ yield the highest values. Note that the maximal Note that Reid et al. ͓6͔ used another figure of merit: S ICH +B violation with indistinguishable photons almost does not depend on ICH ͓ A͑ ͒ B͑ ͔͒ = B with B= P+ A1 + P+ B1 , which gives different re- the voting procedure used. ͑b͒ Maximal resistance to noise in the sults for the different voting strategies. Recalling the artifacts majority voting scenario ͑solid red lines͒ and the unanimity sce- we already found in the amount of CH violation for Poisso- nario ͑dashed blue lines͒ for sources producing independent pairs nian sources, we choose to look now at an experimentally ͓thick line, same curves as in Fig. 2͑b͔͒ or indistinguishable photons meaningful figure of merit: namely, the resistance to noise. ͑thin line͒. The unanimous vote is more robust for indistinguishable photons, but majority voting on independently produced pairs yields the most persistent violation. A. Noisy symmetric state
Unlike for distinguishable photons, the effect of a noise The state after the noisy channel is thus given by map on a symmetric M-photon state does not affect each photon independently. We thus need here a more precise ͵ ͑͒͑ ͒ ͑ † ͓͒ ͔͑͒ noise model. For the sake of simplicity, we put ourselves in out = p U 1 in U 1 dU 11 ͑ ͒ an asymmetric setting, modeling the noise observed in the SU 2
state measured by Alice and Bob as coming from the imper- ϱ fection of the channel linking the source and Alice. Because 1 2 = ͵ d͵ d͵ dsin2 ͑͒sin ͑͒p͑͒ the channel slightly deteriorates the systems passing through 4 −ϱ 0 0 it, but has no preferred basis, we model it, by an average ϫ͓ ͑ ͒ ͔ ͓ †͑ ͒ ͔ ͑ ͒ over all rotation axes nជ =͑sin cos ,sin sin ,cos ͒ in U , , 1 in U , , 1 , 12 the Bloch sphere of rotations U by an angle 2, with  following a properly normalized Gaussian distribution p͑͒ where we have used the appropriate Haar measure of 2 2 SU͑2͒ in terms of the Euler angles: ͓dU͔ 2 − /2 ជ ͑ជ ជ ជ ͒ = 2 e . For any representation = J ,J ,J of 1 2 ͑ −2 ͒ͱ x y z ͑͒ ͑͒  1−e 2 = 4 sin sin d d d . We introduced the Haar mea- SU͑2͒ generators, the rotation operator is U=exp͑−nជ ·ជ ͒. sure here because it is the only measure which is invariant
062110-5 BANCAL et al. PHYSICAL REVIEW A 78, 062110 ͑2008͒ under group operations. It thus treats every rotation the same ͉0͘ ͉0͘ + ͉1͘ ͉1͘ M ͉⌿ ͘ ͩ A B A B ͪ way, reflecting the fact that the noise has no preferred rota- M = ͱ2 tion axis. ͉⌽ ͗͘⌽ ͉ M This channel applied to a single pair state in= 1 1 ͉⌽ ͘ 1 ͑ † † † †͉͒ ͘ ͑ ͒ =2−M/2͚ ͚ ͉0͘i͉1͘M−i ͉0͘i͉1͘M−i with 1 = ͱ2 a0b0 +a1b1 0 produces a Werner state 8 , al- A A B B i=0 ⌸M lowing one to make a correspondence between the usual i noise model used in the previous part of this work, in terms sym M of Werner states and this one: −M/2 ͉ ͘ ͉ ͘ ϳ͉⌽ ͘ —— 2 ͚ i,M − i A i,M − i B M , → i=0
2 2 2 ͑16͒ 3w = e−2 + e−4 + e−6 . ͑13͒ ⌸M ͑ M ͒ where i is the set of all i possible arrangements of i “0” ͑ †͒i͑ †͒M−i ͉ ͘ a0 a1 ͉ ͘ This relation allows us to interpret the amount of white noise and M −i “1”, and i,M −i A = ͱi!͑M−i!͒ 0 is the Fock state ⑀=1−w as being, to first order, the variance of the random describing i of Alice’s M photons in the “0” state and M −i in rotation angle: the “1” state. ͉⌿ ͘ ͉⌽ ͘ So the only difference between M and M is the dis- tinguishability of the M photons flying to Alice or Bob. But ⑀ =42 + O͑4͒. ͑14͒ in the setup we considered ͑as described in Fig. 1͒,wedid not take advantage of the particular pairing between some of Applying this channel to the state ͉⌽ ͘ for various M and M Alice’s photons with some of Bob’s ones. Because we ap- performing the majority and unanimity votes with settings in plied a global measurement, we could even suppose that all ␣, Eqs. ͑7͒, we found that the unanimity procedure is more photons on Alice’s ͑Bob’s͒ side were mixed before reaching robust to noise than the majority vote, scaling like ϳM−1 the beam splitter. In other words, we classically lost trace of ͓see Fig. 5͑b͔͒. the pairing between Alice’s and Bob’s photons. We are thus comparing a situation in which one explicitly chose not to B. Particle losses distinguish between photons belonging to a given set, with another one for which these photons are intrinsically indis- To compare indistinguishable and independent pairs in the tinguishable. case of losses, we consider the case in which one particle is Let us now compare the entanglement present in both lost on each side, yielding a total number of detections, states. Eisert et al. ͓5͔ calculated the amount of entanglement ͉⌿ ͘ 2͑M −1͒. In terms of modes, the state measured after the loss present in the state of distinguishable particles M after of particles can be written having forgotten the pairing of Alice’s photons with Bob’s ones. For M even,
M/2 ϳ a b ͉⌽ ͗͘⌽ ͉a†b† + a b ͉⌽ ͗͘⌽ ͉a†b† ͑2j +1͒2 M +1 M−1 0 0 M M 0 0 0 1 M M 0 1 E = E͉͑⌿ ͒͘ = ͚ ͩ ͪlog ͑2j +1͒. d M M͑ ͒ / 2 ͉⌽ ͗͘⌽ ͉ † † ͉⌽ ͗͘⌽ ͉ † † j=0 2 M +1 M 2−j + a1b0 M M a1b0 + a1b1 M M a1b1. ͑15͒ ͑17͒ ͉⌽ ͘ Concerning the state of indistinguishable particles M , ͑ ͒ Using such a state, we could find a violation for sufficiently writing it in terms of modes as in Eq. 16 , we see that its many pairs M ജ10, starting with majority voting. Thus, there entanglement is given by is no need for additional post-selection here. ͉͑⌽ ͒͘ ͑ ͒ ͑ ͒ Ei = E M = log2 M +1 , 18 since it is a maximally entangled state of two systems of IV. DISTINGUISHABLE VERSUS dimension M. Evaluating these two quantities, we find Ei Ͼ ∀ / M ϱ INDISTINGUISHABLE PAIRS Ed M, and more precisely Ei Ed ——→ 2. So more en- tanglement is present in the state where→ photons are quan- In the last sections we examined how two different mul- tumly indistinguishable, while a larger violation of the CH ͉⌿ ͘ ͑ ͒ ͉⌽ ͘ ͑ ͒ tiparticle bipartite states M , Eq. 4 , and M , Eq. 10 , inequality can be observed using a natural setup if the pho- could be used to show nonlocality using a natural setup pro- tons are in principle distinguishable, but we choose not to ducing binary outcomes. These two states are actually re- make any difference between them. Looking at how resistant lated: if one were to produce the state of independent pairs these violations are with respect to noise confirms this order. ͉⌿ ͘ M with fundamentally indistinguishable photons on both It should only be noted that compared to particle losses, the Alice and Bob’s sides, then the state created would be sym- indistinguishable case looks more resistant, since no addi- metric with respect to permutations between Alice’s photons tional post-selection was necessary to find a violation when or Bob’s ones, and we would actually have produced state both Alice and Bob lost a particle during the experiment. ͉⌽ ͘ ͉⌿ ͘ M . This can be seen by projecting M onto the corre- This is in agreement with other results ͓18͔, showing that sponding symmetric subspaces: entanglement and nonlocality are different measures.
062110-6 TESTING A BELL INEQUALITY IN MULTIPAIR SCENARIOS PHYSICAL REVIEW A 78, 062110 ͑2008͒
TABLE I. Summary of the main results of this work.
Indistinguishable Independent pairs photons
͉0͘ ͉0͘ + ͉1͘ ͉1͘ M 1 ͫ A B A B ͬ ͓a†b† + a†b†͔M͉0͘ 0 0 1 1 State ͱ2 M!ͱM +1
Entanglement 1 after particles ϳ ͑ ͓͒ ͔ log ͑M +1͒ log2 M 5 2 order loss 2
Largest CH ϳM−1/2 ϳM−1 violation ͑majority voting͒ ͑any voting procedure͒
Highest ϳM−1 ϳM−1 resistance ͑majority voting͒ ͑unanimous voting͒ to noise
Loss of one No violation without post- Violation possible particle on selection, at least for M ഛ50 for M ജ10 each side
V. EXPERIMENTAL PERSPECTIVES entanglement using PDC sources was demonstrated in ͓10͔. A careful analysis of post-selection might thus open the pos- In this section we give a brief overview of experimental sibility to feasible experiments. situation where our techniques might be applied. As mentioned previously, the experiment of Ref. ͓3͔ VI. CONCLUSION shows evidence for entanglement in ensembles of ultracold We considered Bell experiments on multiple pairs of par- atoms of 87Rb in an optical lattice. Entanglement between ticles, where the two parties are not able to address each two atomic levels is generated via a partial swap gate, an particle separately and thus call upon global measurements, entangling operation. In order to apply our techniques, the projecting all of their incoming particles in the same basis. atoms of each level should be addressed separately; that is, Votes were introduced as a natural way to produce binary Alice should hold all atoms in the ground state and Bob all outcomes from two detection numbers. This allowed us to atoms in the excited state. Note that in this experiment the test the CH inequality in the presence of both a source of M pairs are independent because they are located in different independent pairs and of M indistinguishable pairs, high- regions of the optical lattice. lighting a violation of the CH inequality for any number of Another experimental situation invoking Bose-Einstein pairs, M. Considering the resistance to noise of such viola- condensates where our techniques might be useful is super- tions, modeled as a noisy channel, we could provide an ex- radiant scattering ͓19͔. It has been argued that this process perimentally meaningful measure of nonlocality. The impact generates entanglement between the emitted photons and the of losses was also evaluated for the two situations, showing atoms of the condensates. In that case the particles would be that indistinguishable pairs are more robust against losses. indistinguishable. More detailed results are summarized in Table I. Finally, a A third possibility is the experiment discussed in Ref. comparison of the nonlocality observed for each source with ͓20͔, which is a proposal for energy-time entanglement of the entanglement of their respective states provided another quasiparticles in a solid-state device. This experiment is an example of the nonmonotonicity between these two quanti- adaptation of the Franson-type experiment ͓21͔ with en- ties. tangled electron-hole pairs. Finally it is also worth mentioning quantum optics. How- ACKNOWLEDGMENTS ever, it is not clear that our techniques will turn out useful in We thank F.S. Cataliotti for pointing out the potential ap- this field, since they require high detection efficiencies, a plication of our results to superradiant scattering. We ac- feature that still lacks generally in optics. Still, sources pro- knowledge financial support from the EU project QAP ͑No. ducing independent entangled pairs, or indistinguishable IST-FET FP6-015848͒ and the Swiss NCCR Quantum photons via PDC, are already well understood. Multiphoton Photonics.
062110-7 BANCAL et al. PHYSICAL REVIEW A 78, 062110 ͑2008͒
͓1͔ J. Bell, Physics ͑Long Island City, N.Y.͒ 1, 195 ͑1964͒. ͓11͔ J. F. Clauser, M. A. Horne, A. Shimony, and R. A. Holt, Phys. ͓2͔ A. Aspect, Nature ͑London͒ 398, 189 ͑1999͒. Rev. Lett. 23, 880 ͑1969͒. ͓3͔ M. Anderlini et al., Nature ͑London͒ 448, 452 ͑2007͒. ͓12͔ J. F. Clauser and M. A. Horne, Phys. Rev. D 10, 526 ͑1974͒. ͓ ͔ 4 Y.-C. Liang and A. C. Doherty, Phys. Rev. A 73, 052116 ͓13͔ D. Collins and N. Gisin, J. Phys. A 37, 1775 ͑2004͒. ͑ ͒ 2006 . ͓14͔ N. Brunner, C. Branciard, and N. Gisin, Phys. Rev. A 78, ͓5͔ J. Eisert, T. Felbinger, P. Papadopoulos, M. B. Plenio, and M. 052110 ͑2008͒. Wilkens, Phys. Rev. Lett. 84, 1611 ͑2000͒. ͓15͔ P. H. Eberhard, Phys. Rev. A 47, R747 ͑1993͒. ͓6͔ M. D. Reid, W. J. Munro, and F. De Martini, Phys. Rev. A 66, ͓16͔ P. Pearle, Phys. Rev. D 2, 1418 ͑1970͒. 033801 ͑2002͒. ͓ ͔ ͑ ͒ ͓7͔ P. D. Drummond, Phys. Rev. Lett. 50, 1407 ͑1983͒. 17 N. Gisin and B. Gisin, Phys. Lett. A 260, 323 1999 . ͓ ͔ ͓8͔ S. J. Jones, H. M. Wiseman, and D. T. Pope, Phys. Rev. A 72, 18 A. A. Méthot and V. Scarani, Quantum Inf. Comput. 7, 157 ͑ ͒ 022330 ͑2005͒. 2007 . ͓9͔ G. Toth, C. Knapp, O. Guhne, and H. J. Briegel, Phys. Rev. ͓19͔ S. Inouye et al., Science 285, 5427, ͑1999͒. Lett. 99, 250405 ͑2007͒. ͓20͔ V. Scarani, N. Gisin, and S. Popescu, Phys. Rev. Lett. 92, ͓10͔ H. S. Eisenberg, G. Khoury, G. A. Durkin, C. Simon, and D. 167901 ͑2004͒. Bouwmeester, Phys. Rev. Lett. 93, 193901 ͑2004͒. ͓21͔ J. D. Franson, Phys. Rev. Lett. 62, 2205 ͑1989͒.
062110-8 Paper B
Quantifying Multipartite Nonlocality
J.-D. Bancal, C. Branciard, N. Gisin and S. Pironio
Physical Review Letters 103, 090503 (2009)
83
week ending PRL 103, 090503 (2009) PHYSICAL REVIEW LETTERS 28 AUGUST 2009
Quantifying Multipartite Nonlocality
Jean-Daniel Bancal, Cyril Branciard, Nicolas Gisin, and Stefano Pironio Group of Applied Physics, University of Geneva, 20 rue de l’Ecole-de-Me´decine, CH-1211 Geneva 4, Switzerland (Received 16 March 2009; published 26 August 2009) The nonlocal correlations of multipartite entangled states can be reproduced by a classical model if sufficiently many parties join together or if sufficiently many parties broadcast their measurement inputs. The maximal number m of groups and the minimal number k of broadcasting parties that allow for the reproduction of a given set of correlations quantify their multipartite nonlocal content. We show how upper bounds on m and lower bounds on k can be computed from the violation of the Mermin-Svetlichny inequalities. While n-partite Greenberger-Horne-Zeilinger states violate these inequalities maximally, we find that W states violate them only by a very small amount.
DOI: 10.1103/PhysRevLett.103.090503 PACS numbers: 03.67.Mn, 03.65.Ud
By performing local measurements on an n-partite en- cal communication models in the manner of Svetlichny tangled state, one obtains outcomes that may be nonlocal, [10–13], where the n parties are divided into m disjoint in the sense that they violate a Bell inequality [1]. Since the subgroups. Within each group, the parties are free to seminal work of Bell, nonlocality has been a central sub- collaborate and communicate with each other but are not ject of study in the foundations of quantum theory and has allowed to do so between distinct groups. The idea is that a been supported by many experiments [2,3]. More recently, given set of correlations contains more multipartite non- it has also been realized that it plays a key role in various locality if more parties need to join to be able to reproduce quantum information applications, where it represents a these correlations (see Fig. 1). The second measure of resource different from entanglement [4]. multipartite nonlocality that we introduce is based on While nonlocality has been extensively studied in the models where k parties broadcast their measurement inputs bipartite (n ¼ 2) and to a lesser extent in the tripartite (n ¼ to all others. The idea again is that correlations that require 3) case, the general n-partite case remains much unex- more broadcasting parties to be simulated contain more plored. The physics of many-particle systems, however, multipartite nonlocality. The maximal number m of groups is well known to differ fundamentally from the one of a few and the minimal number k of broadcasting parties that particles and to give rise to new interesting phenomena, allow for the reproduction of a given set of correlations such as phase transitions or quantum computing. Entangle- thus represent two simple ways of quantifying their multi- ment theory, in particular, appears to have a much more partite nonlocal content. complex and richer structure in the n-partite case than it Given an arbitrary set of correlations, it may in general has in the bipartite setting [5,6]. This is reflected by the fact be difficult to determine the corresponding values of m and that multipartite entanglement is a very active field of k. To evaluate these quantities, we introduce a family of research that has led to important insights into our under- Bell tests based on the Mermin-Svetlichny (MS) inequal- standing of many-particle physics (see, e.g., [7,8]). In view ities [10,14]. Specifically, we compute the maximal value of this, it seems worthy to investigate also how nonlocality of the MS expressions achieved by models where n parties manifests itself in a multipartite scenario. What new fea- form m groups and where k parties broadcast their inputs. tures emerge in this context, and what are their funda- By comparing the amount by which quantum states violate mental implications? How does one characterize the non- the MS inequalities with our bounds, one thus obtains locality of experimentally realizable multiqubit states, constraints on the values of m and k necessary to reproduce such as W states, for instance? What role do n-partite their nonlocal correlations. Since these criteria are based nonlocal correlations play in quantum information proto- on Bell-like inequalities, they can be tested experimentally. cols, e.g., in measurement-based computation [9]? The vision behind the present Letter is that, in order to answer such questions and make further progress on our understanding of multipartite nonlocality, one should first find ways to quantify it. Motivated by this idea, we intro- FIG. 1. Different groupings of n ¼ 4 parties into m groups. duce two simple measures that quantify the multipartite Within each group, every party can communicate to any other extent of nonlocality. party, as indicated by the arrows. (a) If all parties join into one A natural way to characterize nonlocality is to attempt to group (m ¼ 1), they can achieve any correlations. (b),(c) If they replicate it using models where some nonlocal interactions split into m ¼ 2 groups, they can realize some nonlocal corre- (such as communication) are allowed between some par- lations but not all. (d) If they are all separated (m ¼ n), they can ties. The first measure that we consider is based on classi- only reproduce local correlations.
0031-9007=09=103(9)=090503(4) 090503-1 Ó 2009 The American Physical Society week ending PRL 103, 090503 (2009) PHYSICAL REVIEW LETTERS 28 AUGUST 2009
A Bell-like test for a given number of groups could with each other but are not allowed to do so between a priori depend on how the groups are formed, e.g., 2 þ distinct groups. 2 in Fig. 1(b) or 1 þ 3 in Fig. 1(c), and on which party (ii) Broadcasting.—Out of the n parties, k of them can belongs to which group. But the tests that we present here broadcast their input to all other parties. The remaining depend only on the total number m of groups and not on m ¼ n k parties cannot communicate their input to any how the parties are distributed within each group. other party. Furthermore, in the measurement scenario that we consider In the framework of these two communication models, in this work (restricted to ‘‘correlation functions’’), a com- the values that can be reached by the MS polynomials are munication model with m disjoint groups is less powerful bounded as follows. than a communication model with k ¼ n m broadcast- Theorem.—For both the grouping and the broadcasting ing parties. Yet we find that the bounds on the MS expres- models, sions are identical in both cases. m ðn mÞ=2 jSn j 2 : (3) As mentioned above, our results can be used to estimate the multipartite nonlocal content of quantum states. We Moreover, this bound is tight; i.e., for each model, there m ðn mÞ=2 carry out this analysis for Greenberger-Horne-Zeilinger exists a strategy that yields jSn j¼2 (in the case of (GHZ)-like and W states in the last part of this Letter. the grouping model, this is true for any possible grouping Definitions.—We consider a Bell experiment involving of the n parties into m groups). n parties which can each perform one out of two measure- Before proving our theorem, let us elaborate on some ments. The outcomes of these measurements are written aj comments. First of all, let us mention that, for m ¼ 2, the 0 and aj and can take the values 1. Letting M1 ¼ a1,we results obtained in Refs. [11,12] for the grouping model are define recursively the MS polynomials [10,11,14,15]as recovered. Note also that, since we consider correlation functions M ¼ 1ða þ a0 ÞM þ 1ða a0 ÞM0 ; n 2 n n n 1 2 n n n 1 (1) only, the grouping model is weaker than the broadcasting 1 model. Indeed, in each group, one can assume that all 0 Mn ¼ pffiffiffi ðMn MnÞ; (2) 2 parties send their inputs to one singled-out party, which decides for the correlation function of the whole group. 0 where Mn is obtained from Mn by exchanging all primed The broadcasting model clearly allows more communica- þ and nonprimed aj’s. Mn and Mn are equivalent under the tion than this. 0 0 exchange faj;ajg$f aj;ajg for any single party j, which The fact that the same bounds hold for the two models is corresponds to a relabeling of its inputs and outputs. The not trivial and is actually a special property of the MS MS polynomials are symmetric under permutations of the expressions. Indeed, we have been able to construct in- parties. equalities that distinguish between these models. We interpret these polynomials as sums of expectation A more technical remark.—As observed in Ref. [13] for values by identifying each term of the form a1 ...an with the case m ¼ 2, the structure of the MS inequalities allows the correlation coefficient ha1 ...ani, which is the expec- one to detect a stronger form of nonlocality than the one tation value of the product of the outputs a1 ...an. The induced by grouping. It is interesting to identify precisely above polynomials can thus be interpreted as Bell inequal- the most general communication model associated with M 1 ities. Their localpffiffiffi bounds are known [11]tobej nj this stronger form of nonlocality. and jMn j 2, while the algebraic bounds (the maximal The common feature of the two above models that we value achieved by an arbitrary nonlocal model) are easily exploit in our proof (see below) and that fundamentally bn=2c bðn 1Þ=2cþ1=2 Sm found to be jMnj 2 and jMn j 2 . limits the values of the n expressions is that there exists a In the remainder of this Letter, we shall be interested in special subset of m parties such that none of the n parties the following family of polynomials: knows more than one input from this subset. This is ob- vious in the broadcasting model; in the grouping model, Mn for n m even; Sm ¼ simply pick one party in each of the m groups. Let us n Mþ n m : n for odd therefore define the most general (but less natural) com- Quantifying multipartite nonlocality through communi- munication model with this property. cation models.—In a classical communication model, the n Restrained-subset model.—Among the n parties, there is parties have access to shared randomness and are allowed a subset of m parties, such that none of the n parties to communicate their inputs to some other parties. Given receives more than one input from this subset. The other the information available to them, each party then produces parties are free to communicate as they wish. Note that the a local output. Here, as explained in the introduction, we parties within the special subset of m parties cannot receive define two families of models that depend on a parameter inputs from any other party in the subset, as they already m (or k ¼ n m) which quantify the extent of multipartite know their own input. nonlocality. This model also satisfies the bound (3); for the case m ¼ (i) Grouping.—The n parties are grouped into m sub- 2, the results of Ref. [13] are recovered. This model is sets. Within each group, the parties are free to collaborate optimal for the MS expressions, in the sense that any 090503-2 week ending PRL 103, 090503 (2009) PHYSICAL REVIEW LETTERS 28 AUGUST 2009 additional communication between the parties allows them 2ðni 1Þ=2 at the same time. This is because the (tight) to violate (3). Mþ 2ni=2 algebraic bound for Gi is , and Eq. (2) tells us that Proof of (3).—It is sufficient to prove (3) for our stron- 0 in order to achieve it both MG and MG must reach their gest model, i.e., for the restrained-subset model. Since the i i algebraic limit. Similarly, there exists a strategy for groups MS inequalities are symmetric under permutations of the þ with an even number of parties ni such that jM j¼ parties, we can assume without loss of generality that the Gi jM j¼2ðni 1Þ=2 fM; M0g parties 1; ...;mare the ones in the restrained subset. Gi . We shall thus associate polyno- m þ Consider first the case of even n m, for which Sn ¼ mials to odd groups and fM ;M g to even ones. Mn. Applying twice the recursive definition (1), we get Consider two groups Gi and Gj and their union Gij ¼ Gi [ Gj. From the definitions (1) and (2), one can derive M 1 a a M0 a a0 M n ¼ 2ð n n 1 n 2 þ n n 1 n 2 the decompositions: 0 0 0 0 þ anan 1Mn 2 anan 1Mn 2Þ: (4) M 1 M M M0 M0 M M0 ; Gij ¼ 2½ Gi ð Gj þ Gj Þþ Gi ð Gj Gj Þ ð0Þ ð0Þ Using again twice (1) for Mn 2, we can replace Mn 2 as a 1 þ þ þ MG ¼ ½MG ðMG þ MG ÞþMG ðMG MG Þ ; ð0Þ n m ij 2 i j j i j j function of Mn 4 in (4). Iterating this process 2 times, 1 0 0 M ¼ ½M ðMG þ M Þ M ðMG M Þ : we end up with the following expression for Mn: Gij 2 Gi j Gj Gi j Gj
X1 0 0 1 s s s ...s M M n mþ1 n mþ1 Similar relations are also obtained for Gij since ð G Þ ¼ Mn ¼ an ...am 1Mm ; (5) 2ðn mÞ=2 þ M . Now inserting in the above relations the value sn;...;smþ1¼0 G attained by the strategies that we just mentioned for the 0 1 0 where ai ¼ ai and ai ¼ ai and where, depending on the two initial groups, one finds that their combined strategy sn;...;sm 1 s ; ...;s M þ 0 ðni 1Þ=2 ðnj 1Þ=2 value of ( n mþ1), m is equal to one of the can achieve jMG j¼jMG j¼2 2 or M ; M0 ij ij polynomials f m mg. Mþ M 2ðni 1Þ=22ðnj 1Þ=2 sn...sm 1 j G j¼j G j¼ , depending on which The MS polynomial Mm þ is a function of the out- ij ij s ...s s ...s 0 0 n mþ1 n mþ1 set of polynomials is associated to the two initial groups. puts fa1;a1 ...;am;amg, i.e., Mm ¼ Mm 0 0 Iterating this construction by joining groups succes- (a1;a1 ...;am;am). Among the parties fm þ 1; ...;ng, Q M m 2ðni 1Þ=2 there exists a (possibly empty) subset fj1; ...;jlg that do sively 2 by 2, we find j nj¼ i¼1 when there Mþ not receive any input from parties 2; ...;m but possibly isQ an even number of even groups and j n j¼ 0 m 2ðni 1Þ=2 from party 1. Define two effective outputs A1 and A1 as i¼1 otherwise. Since the parity of the number s s s s j1 jl 0 0 j1 jl n m A1 ¼ a1a ...a and A ¼ a a ...a , respectively. of even groups is the same as the parity of , there j1 jl 1 1 j1 jl S There also exist similar disjoint subsets for parties Qmust exist a strategy which achieves j nj¼ m 2ðni 1Þ=2 2ðn mÞ=2 j 2; ...;m, for which we also define effective outputs i¼1 ¼ . 0 0 Nonlocality of quantum states.—Suppose that one ob- A2;A2; ...;Am;Am. Then we can write m ðn mÞ=2 jSn j 2 s s s ...s s ...s serves a violation of the inequality . One a n ...a mþ1 M n mþ1 M n mþ1 A ;A0 ...;A ;A0 : n mþ1 m ¼ m ð 1 1 m mÞ can then conclude that, in order to reproduce the corre- s ...s sponding nonlocal correlations in the framework of our M n mþ1 A ;A0 ...;A ;A0 Formally, m ( 1 1 m m) is a MS polyno- communication models, the parties cannot be separated in m mial that involves parties isolated from each other, since more than m 1 groups or that at least k þ 1 ¼ n m þ A A0 j the outputs j and j of party do not depend on the input 1 parties must broadcast their input. Thus, the above m 1 m of any of the other parties. It can therefore not bounds on Sn give us bounds on the multipartite character exceed its local bound 1. Inserting this bound in (5), we of the observed nonlocal correlations (an upper bound on ðn mÞ=2 find jMnj 2 . m or a lower bound on k). For odd values of n m, we have to consider the poly- Here we discuss the violation of the inequalities (3) for m þ nomials Sn ¼ Mn . Using the definitions (1) and (2), one n-partite GHZ-like and W states. States in the GHZ family þ can show that Mn has a similar decomposition as Mn in are defined as jGHZ i¼cos j00...0iþsin j11...1i. þ (5). The same reasoning as before then leads to jMn j The maximal value of Mn for these states was conjectured ðn mÞ=2 ðn 1Þ=2 2 . j in Ref. [16]tobeMn ¼ maxf1; 2 sin2 g. Numerical Proof of the tightness of (3).—To prove that (3) is a tight optimizations (see Fig.pffiffiffi2) induce us to conjecture that þ ðn 1Þ=2 bound, it is sufficient to prove that it can be reached by our similarly Mn ¼ maxf 2; 2 sin2 g. Upon compari- weaker communication model, i.e., the grouping model son with the bound (3), we conclude that all n-partite (for any possible distribution of the n parties into m GHZ states with > =8 are maximally nonlocal accord- groups). ing to our criterion (i.e., all parties must be grouped to- Let Gi (i ¼ 1; ...;m) denote the m groups into which gether or n 1 parties must broadcast their input to the n parties are split. For all groups Gi having an odd reproduce their correlations). Less entangled GHZ states, number ni of parties, there exists a strategy for the parties on the other hand, cannot be simulated if the parties are G M M0 m 1 k in i to reach both algebraic bounds j Gi j¼j Gi j¼ separated in more than groups or if fewer than þ 090503-3 week ending PRL 103, 090503 (2009) PHYSICAL REVIEW LETTERS 28 AUGUST 2009
þ FIG. 3. Maximal values of Mn (solid line) and Mn (dashed þ FIG. 2. Maximal values of Mn and Mn for partially entangled line) for n-partite W states. The curves were obtained by a GHZ states for 3 n 6. The dots are values found by nu- general numerical optimization for n 9 and under the hy- merical optimization, and the solid lines are the conjectured pothesis that all parties use identical measurement settings for M Mþ 2ðn 1Þ=2 sin2 M violationpffiffiffi n ¼ n ¼ (valid only above 1 for n 10 n 19. The asymptotic values for n !1 computed as þ and 2 for Mn ). explained in the text are also shown.
1 ¼ n m þ 1 parties broadcast their inputs whenever ity of W states. Finding which one of these possibilities is ðm 1Þ=2 > c with sin2 c ¼ 2 . Interestingly, c is the the correct one is an interesting problem for future re- n search. Also, it would be interesting to analyze the non- same for all . pffiffiffi Consider now the W states jWni¼ð1= nÞ locality of other multipartite quantum states with our ðj10...0iþ...þj0...01iÞ. Numerical optimizations criteria. suggest that the maximal values of the MS polynomials As suggested by the situation in entanglement theory, we for these states are upper bounded by a small constant for do not expect our measures to be the only ways to quantify all n (see Fig. 3). To convince ourselves that this is indeed the multipartite content of nonlocality. It would thus be of the case, we analyzed analytically the case where all pairs interest to look for different measures, based on other of measurement settings are the same for all parties. This is nonlocal models than the ones considered here. justified by the results of our numerical optimizations up to Finally, let us stress that the criteria that we presented in n ¼ 9, for which the optimal measurement settings can this Letter can be tested experimentally. It would thus be always be of this form. We thus introduce for all n parties worth (re)considering experiments on multipartite non- two measurement operators A0 and A1 represented by locality in view of our results. vectors a~i ¼ðsin i cos i; sin i sin i; cos iÞ. One can We acknowledge support by the Swiss NCCR Quantum show that, as n increases, the maximal value of jMnj or Photonics and the European ERC-AG QORE. þ jMn j can be reached for i ¼ 0 and i ! 0. Assuming a n power lawpffiffiffi for ið Þ, one finds that it should be given by c = n i i at the maximum. After optimization of the [1] J. Bell, Speakable and Unspeakable in Quantum c c M Mþ constants 0 and 1 for both n and n , we found that Mechanics (Cambridge University Press, Cambridge, the asymptotic maximal values of the MS polynomials England, 1987). (under our assumptions, which we believepffiffiffiffiffiffiffiffi are not restric- [2] M. Genovese, Phys. Rep. 413, 319 (2005). þ n 1 [3] A. Aspect, Nature (London) 398, 189 (1999). tive) are jM1j’1:62 and jM1j¼2 2=e. Since Sn > 1 for n 3, letting one party broadcast his input, or letting [4] J. Barrett et al., Phys. Rev. A 71, 022101 (2005); C. two parties join to form a group, is not sufficient to repro- Brukner et al., Phys. Rev. Lett. 92, 127901 (2004); A. K. W Ekert, Phys. Rev. Lett. 67, 661 (1991); J. Barrett et al., duce the correlations of the state. However, we cannot Phys. Rev. Lett. 95, 010503 (2005); A. Acı´n et al., Phys. reach the same conclusion if more than two parties join or Rev. Lett. 98, 230501 (2007); L. Masanes, Phys. Rev. Lett. if k ¼ n m 2 parties broadcast their inputs, since the 102, 140501 (2009). criterion (3) is not violated in this case. [5] W. Du¨r et al., Phys. Rev. A 62, 062314 (2000). Conclusion.—We proposed in this Letter two simple [6] R. Horodecki et al., Rev. Mod. Phys. 81, 865 (2009). measures of multipartite nonlocality and introduced a se- [7] A. Osterloh et al., Nature (London) 416, 608 (2002). ries of Bell tests to evaluate them. This represents a pri- [8] G. Vidal, Phys. Rev. Lett. 93, 040502 (2004). mary step towards a quantitative understanding of quantum [9] R. Raussendorf et al., Phys. Rev. Lett. 86, 5188 (2001). nonlocality for an arbitrary number n of parties. [10] G. Svetlichny, Phys. Rev. D 35, 3066 (1987). While GHZ states exhibit a strong form of multipartite [11] D. Collins et al., Phys. Rev. Lett. 88, 170405 (2002). [12] M. Seevinck et al., Phys. Rev. Lett. 89, 060401 (2002). nonlocality according to our criterion, we found that W k [13] N. S. Jones et al., Phys. Rev. A 71, 042329 (2005). states violate our inequalities only for small values of . [14] N. D. Mermin, Phys. Rev. Lett. 65, 1838 (1990). This suggests that W states exhibit only a very weak form [15] R. F. Werner and M. M. Wolf, Phys. Rev. A 64, 032112 of multipartite nonlocality. Or it might be that other in- (2001). equalities are necessary to quantify properly the nonlocal- [16] V. Scarani et al., J. Phys. A 34, 6043 (2001).
090503-4 Paper C
Simulation of Equatorial von Neumann Measurements on GHZ States Using Nonlocal Resources
J.-D. Bancal, C. Branciard and N. Gisin
Advances in Mathematical Physics, volume 2010, Article ID 293245 (2010)
89
Hindawi Publishing Corporation Advances in Mathematical Physics Volume 2010, Article ID 293245, 14 pages doi:10.1155/2010/293245
Research Article Simulation of Equatorial von Neumann Measurements on GHZ States Using Nonlocal Resources
Jean-Daniel Bancal, Cyril Branciard, and Nicolas Gisin
Group of Applied Physics, University of Geneva, 20 rue de l’Ecole-de-Medecine,´ 1211 Geneva 4, Switzerland
Correspondence should be addressed to Jean-Daniel Bancal, [email protected]
Received 31 August 2009; Accepted 11 December 2009
Academic Editor: Shao-Ming Fei
Copyright q 2010 Jean-Daniel Bancal et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Reproducing with elementary resources the correlations that arise when a quantum system is measured quantum state simulation allows one to get insight on the operational and computational power of quantum correlations. We propose a family of models that can simulate von Neumann measurements in the x − y plane of the Bloch sphere on n-partite GHZ states. For the tripartite and fourpartite states, the models use only bipartite nonlocal boxes; they can be translated into classical communication schemes with finite average communication cost.
1. Introduction
Understanding the nonlocal correlations created upon measurement of some entangled quantum system is a problem which runs up against our common representation of the world, by the very definition of nonlocality, i.e. violation of a Bell inequality 1 . Indeed, no explanation one would reasonably accept as possible, like agreement prior to measurement or subluminal communication of inputs, seems to be used by nature in order to create these correlations see the numerous experimental violations of Bell inequalities 2 . Still, some insight on the power of such correlations was gained when people came out with models able to reproduce them in terms of classical resources. For instance, Toner and Bacon 3 showed how to simulate von Neumann measurements on a singlet state with one bit of communication. Such a result puts an upper bound on the required amount of nonlocal resources needed for the reproduction of singlet correlations; it guarantees also that the corresponding correlations are not a stronger resource of nonlocality than 1 bit of classical communication. Adifferent kind of resources that was also considered are the so-called nonlocal boxes 4 : these are simple nonlocal correlations which do not allow signaling. Successful 2 Advances in Mathematical Physics simulation schemes using nonlocal boxes as unique nonlocal resources include the simulation of the singlet 5 and of partially entangled two-qubit states 6 . Concerning multipartite systems, communication models reproducing Pauli mea- surements on n-partite GHZ or on graph states have also been proposed 7, 8 . For arbitrary possible measurements on the tripartite GHZ state, previous studies suggested that its simulation with bounded communication might be impossible, taking as an example correlations corresponding to measurements of this state in the x − y plane of the Bloch sphere 9 . In this paper, we construct a model which analytically reproduces these equatorial correlations, and whose only nonlocal resources are Popescu-Rohrlich PR boxes 10 and Millionaire boxes 11 . Thus a finite number of bipartite nonlocal boxes are proven to be sufficient to reproduce these genuinely tripartite nonlocal correlations. Note also that even though our model does not give an upper bound on the worst-case communication cost, it does provide a communication model with finite expected communication cost, simulating for instance the tripartite GHZ state with an average total of 10 bits of communication between the parties c.f. Appendix B . The paper is organized as follows: first, we recall the correlations of the GHZ state that we want to simulate. We then present a model for the 3-partite case, and generalize it to more parties. We discuss the construction and then conclude.
2. GHZ Correlations
Consider the n-partite GHZ state
1 ··· ··· |GHZn √ |00 0 |11 1 . 2.1 2
Our goal is to reproduce the correlations which are obtained when von Neumann measurements are performed on this state, by using other nonlocal resources such as nonlocal boxes possibly supplemented with shared randomness . For n 2, the protocol presented in 5 for the singlet state allows one to reproduce the correlations for any measurement settings, using one PR box. Here we recall the definition of a PR box:
PR Box
A Popescu-Rohrlich PR box is a nonlocal box that admits two bits x, y ∈{0, 1} as inputs and produces locally random bits a, b ∈{0, 1}, which satisfy the binary relation
a b xy. 2.2
Going to n ≥ 3, we shall only consider measurements in the x − y plane equatorial measurements , which have the nice feature of producing unbiased marginals: all correlation terms involving strictly fewer than n parties vanish. We write each party’s measurement operator as: A cos φaσX sin φaσY , B cos φbσX sin φbσY ,.... Denoting the binary result of each measurement by α,β,...∈{−1, 1}, the correlations we are interested in are given by
α β ··· αβ ··· 0 2.3 Advances in Mathematical Physics 3
φb φa φc
Alice PR Charlie Bob
α γ β | Figure 1: Simulation of GHZ3 in a Svetlichny scenario: Alice and Bob form a group and can share their information with each other, while Charlie is separated from them. In this scenario 1 PR box allows one to reproduce the equatorial correlations. for all sets of fewer than n parties, and ··· ··· αβ ω cos φa φb φz 2.4 for the full n-partite correlation term. In other words, outcomes appear to be random except when all of them are considered together, in which case their correlation takes a form reminiscent of the singlet state. To simulate such correlations, nonlocal boxes similar to the Millionnaire box will be useful, so let us recall what a Millionnaire box is.
MBox
A Millionaire box is a nonlocal box that admits two continuous inputs x, y ∈ 0, 1 and produces locally random bits a, b ∈{0, 1}, such that a b sg x − y , 2.5 where the sign function is defined as sg x 0ifx>0andsg x 1ifx ≤ 0. It is worth mentioning that even though we restrict the set of possible measurements on the GHZ states, the correlations we consider can still exhibit full n-partite nonlocality. Indeed, the Svetlichny inequality for n parties can be maximally violated with settings in the x − y plane 12, 13 . This implies that in order to simulate these correlations, any model must truly involve all n parties together 14 .
3. Simulation Model for the 3-Partite GHZ State
Let us consider the above correlations for n 3 parties, for which the outcomes of all parties need to be correlated according to αβγ cos φa φb φc . As a first step towards the simulation of these correlations, let us relax some of the constraints and allow two parties to cooperate in a Svetlichny-like scenario 15 see Figure 1 : for instance Alice and Bob would be allowed to communicate with each other, but not with Charlie who is kept isolated from them. In such a scenario, the three parties could create correlations of the desired form with one PR box by using the protocol of 5 to generate outputs α and γ that have a cosine correlation of the form αγ cos φab φc , with a fictitious measurement angle φab φa φb. By then setting either α α, β 1or 4 Advances in Mathematical Physics
α −α, β −1 each with probability 1/2 ,andγ γ , they would recover the desired tripartite correlations αβγ cos φa φb φc . Of course, letting Alice and Bob share their inputs is not satisfactory yet, as this would require signaling between them. We shall now see that it is actually possible to reseparate them, while keeping the tripartite correlation term unchanged. In order to do so, let us recall that the model used above to create the bipartite cosine correlation with a PR box works by asking the parties here, Alice-Bob together and Charlie to input in the box terms of the form 5