Arxiv:1903.06663V3 [Quant-Ph] 19 Mar 2020 3

Quantum Steering

Roope Uola,1, 2 Ana C. S. Costa,1, 3 H. Chau Nguyen,1 and Otfried Gühne1 1Naturwissenschaftlich-Technische Fakultät, UniversitätSiegen, Walter-Flex-Straße 3, 57068 Siegen, Germany 2Département de Physique Appliquée, Universitéde Genève, CH-1211 Genève, Switzerland 3Department of Physics, Federal University of Paraná, 81531-980 Curitiba, PR, Brazil (Dated: March 20, 2020)

Quantum correlations between two parties are essential for the argument of Einstein, Podolsky, and Rosen in favour of the incompleteness of quantum mechanics. Schrödinger noted that an essential point is the fact that one party can influence the wave function of the other party by performing suitable measurements. He called this phenomenon quantum steering and studied its properties, but only in the last years this kind of quantum correlation attracted significant interest in quantum information theory. In this paper the theory of quantum steering is reviewed. First, the basic concepts of steering and local hidden state models are presented and their relation to entanglement and Bell nonlocality is explained. Then various criteria for characterizing steerability and structural results on the phenomenon are described. A detailed discussion is given on the connections between steering and incompatibility of quantum measurements. Finally, applications of steering in quantum information processing and further related topics are reviewed.

CONTENTS F. Steering maps and dimension-bounded steering 20 G. Superactivation of steering 21 I. Introduction 2 A. Overview 2 IV. Joint measurability and steering 22 B. Steering as a formalization of the EPR argument 2 A. Measurement incompatibility 22 C. Steering, Bell nonlocality and entanglement 4 B. Joint measurability on Alice’s side 23 C. Joint measurability on Bob’s side 23 II. Detection of steering 5 D. Incompatibility breaking quantum channels 25 A. Steering detection from correlations 5 E. Further topics on incompatibility 25 1. Linear and nonlinear steering criteria 5 2. Steering criteria from uncertainty relations 6 V. Further topics and applications of steering 26 3. Steering and the CHSH inequality 7 A. Multipartite steering 26 4. Moment matrix approach 7 1. Steering across a bipartition 27 5. Steering criteria based on local uncertainty 2. Different approaches towards multipartite steering 27 relations 8 B. The steering ellipsoid 28 B. Steering detection from state assemblages 9 C. Gaussian steering 29 1. Formulation of the semidefinite program 9 1. A criterion for steering of Gaussian states 29 2. Steering inequalities from the SDP 9 2. Refining Gaussian steering with EPR-type 3. Quantification of steerability with SDPs 10 observables 29 4. From finite to infinite number of measurements 11 D. Temporal and channel steering 30 C. Steering detection from full information 11 E. Quantum key distribution 31 1. Two-qubit states and projective measurements 11 F. Randomness certification 31 2. Steering of higher-dimensional systems and with G. Subchannel discrimination 32 generalized measurements 13 H. One-sided device-independent and device-independent arXiv:1903.06663v3 [quant-ph] 19 Mar 2020 3. Full information steering inequality 14 quantification of measurement incompatibility 33 I. Secret sharing 34 III. Conceptual aspects of steering 14 J. Quantum teleportation 34 A. Hierarchy of correlations 14 K. Resource theory of steering 35 B. Special states and their local hidden state models 15 L. Post-quantum steering 36 1. Werner states 15 M. Historical aspects of steering 36 2. Isotropic states 16 1. Discussions between Schrödingerand Einstein 36 3. LHS model for POVMs 17 2. The two papers by Schrödinger 37 C. Steering and local filtering 18 3. Impact of these papers 38 D. One-way steerable states 18 E. Steering with bound entangled states 19 VI. Conclusion 38 2

References 39 on the finite-dimensional case and mention experiments only very shortly. For discussing quantum steering, the tool of semidefinite programming has turned out to be I. INTRODUCTION useful. Concerning this, we only discuss the main formu- lations, concrete examples and algorithms can be found A. Overview in a different review article (Cavalcanti and Skrzypczyk, 2017). In 1935, Einstein, Podolsky and Rosen (EPR) pre- As mentioned, quantum steering is related to several sented their famous argument against the completeness other central concepts in quantum theory, so it may be of quantum mechanics (Einstein et al., 1935). In this useful to the reader to mention related relevant literature argument, a two-particle state is considered, where one here. First, a review on the quantitative aspects of the party can measure the position or momentum and the EPR argument can be found in (Reid et al., 2009). The correlations of the state allow one to predict the results phenomenon of entanglement is extensively discussed in of these measurements on the other party if the same (Horodecki et al., 2009) and methods to characterize it measurement is performed there. The EPR argument in (Gühneand Tóth,2009). A detailed overview on Bell led to long-lasting discussions, but already directly af- inequalities and their applications is given in (Brunner ter its publication, Schrödingernoted an interesting phe- et al., 2014). Finally, the theory of quantum measure- nomenon in the argument: The first party can, by choos- ments is in depth developed in (Busch et al., 2016). ing the measurement, steer the state on the other side The structure of the current article is the following: In into an eigenstate of position or momentum. This can the remainder of this introduction we explain the idea not be used to transmit information, but still Schrödinger of quantum steering and the main definitions. We also considered it to be magic. provide a short comparison with quantum entanglement The early works of Schrödinger(Schrödinger,1935; and Bell nonlocality, as this is central for the further Schrödinger,1936) did not receive much attention (see discussion. also Section V.M). This changed in 2007, when a formu- Section II presents different methods for the detection lation in the language of quantum information processing of quantum steering. We discuss in detail how steerabil- was given and systematic criteria were developed (Wise- ity can be inferred, if some correlations or the complete man et al., 2007). In the modern view, steering denotes quantum state is known. These methods are then used in the impossibility to describe the conditional states at one Section III, where we describe key conceptual aspects of party by a local hidden state model. As such, steering steering. This includes the discussion of one-way steer- denotes a quantum correlation situated between entan- ing, the superactivation of steering, the steerability of glement and Bell nonlocality. In the following years, the bound entangled states and the construction of steering theory of steering developed rapidly. It was noted that maps. In addition, we can then present the relation to steering provides the natural formulation for discussing other types of quantum correlations in detail. quantum information processing, if for some of the par- Section IV deals with the connections between steering ties the measurement devices are not well characterized. and the joint measurability of observables. We explain Also, the concept of steering helped to understand and the concept of joint measurability and its various connec- answer open questions in quantum information theory. tions to steering. These connections allow one to transfer An important example is here the construction of coun- results from one topic to the other. Section V describes terexamples to the Peres conjecture, which states that different applications of steering as well as further top- certain weakly entangled states do not violate any Bell ics. This includes applications in quantum key distri- inequality. Finally, steering turned out to be closely re- bution and randomness certification, but also topics like lated to the concept of joint measurability of generalized multiparticle steering, steering of Gaussian states, the measurements in quantum mechanics. More precisely, steering ellipsoid and historical aspects of steering. Fi- measurements that are not jointly measurable are exactly nally, Section VI presents the conclusion and some open the measurements that are useful to reveal the steering questions. phenomenon. This has sparked interest in the question in which sense measurements in quantum mechanics can be considered as a resource. B. Steering as a formalization of the EPR argument This review article aims to give an introduction to the concept and applications of quantum steering. Starting Let us start by recalling the EPR argument. Origi- from the basic definitions, we explain steering criteria nally, EPR used the position and momentum of two par- and structural results on quantum steering. We also dis- ticles to explain their line of reasoning (Einstein et al., cuss in some detail related concepts, such as quantum 1935), but in the simplest setting, the argument can be 1 entanglement or the joint measurability of observables. explained with two spin- 2 particles or qubits (Bohm, We focus on the conceptual and theoretical issues and 1951). Consider two particles that are in different lo- 3

which do not need to be projective. For each of Alice’s measurement setting x and result a, Bob remains with an unnormalized conditional state %a|x. The set of these states is called the steering assemblage, and the condi- P tional states obey the condition a %a|x = %B, meaning that the reduced state %B = TrA(%AB) on Bob’s side is independent of Alice’s choice of measurements. FIG. 1 Schematic description of the steering phenomenon: A After characterizing the states %a|x, Bob may try to state %AB is distributed between two parties. Alice performs explain their appearance as follows: He assumes that ini- a measurement (labeled by x ∈ {1, 2}) on her particle and tially his particle was in some hidden state σλ with prob- obtains the result a = ±. Bob receives the corresponding ability p(λ), parametrized by some parameter (or hidden unnormalized conditional states % . If Bob cannot explain a|x variable) λ. Then, Alice’s measurement and result just this assemblage of states by assuming pre-existing states at his location, he has to believe that Alice can influence his gave him additional information on the probability of the state from a distance. states. This leads to states of the form (Wiseman et al., 2007) Z cations and are controlled by Alice and Bob, see Fig. 1. %a|x = p(a|x) dλp(λ|a, x)σλ They are in the so-called singlet state, Z 1 = dλp(λ)p(a|x, λ)σλ. (2) |ψiAB = √ (|01i − |10i), (1) 2 The equivalence between these two expressions is easy to where |0i = |z+i and |1i = |z−i denote the two pos- verify if the setting x can be chosen freely and does not sible spin orientations in the z-direction. If Alice mea- depend on the parameter λ, i.e., p(x, λ) = p(x)p(λ). The sures the spin of her particle in the z-direction and ob- two representations, however, point at different interpre- tains the result +1 (or −1) then, due to the perfect anti- tations. correlations of the singlet state, Bob’s state will be ei- The first representation can be interpreted as if the ther in the state |1i (or |0i). Similarly, if Alice measures probability distribution p(λ) is just updated to p(λ|a, x), the spin in the x-direction,√ Bob’s conditional states are depending on the classical information about the result + given by |x i = (|0√i +|1i)/ 2, if Alice’s result is −1 and a and setting x. Here, Bob does not need to believe that |x−i = (|0i − |1i)/ 2 for the result +1. Alice has control over his state, her measurements and So, by choosing her measurement setting, Alice results just gave him additional information about the can predict with certainty the values of a z- or x- distribution of the states σλ. measurement on Bob’s side. According to EPR, this The second representation can be interpreted as a sim- means that both observables must correspond to “ele- ulation task. Here, Alice can simulate the state %a|x by ments of reality”, as each of them can be predicted in drawing the states σλ according to the distribution p(λ) principle with certainty and without disturbing the sys- and, at the same time, announcing the result a depend- tem. This raises problems if one assumes that the wave ing on the known setting x and the parameter λ. Con- function is a complete description of the physical situa- sequently, Bob does not need to believe that the initial tion, since the corresponding observables do not commute state shared by him with Alice was entangled. and the quantum mechanical formalism does not allow Generally, if a representation as in Eq. (2) exists, Bob one to assign simultaneously definite values to both of does not need to assume any kind of action at a dis- them. Consequently, EPR concluded that quantum me- tance to explain the post-measurement states %a|x. Con- chanics is incomplete. sequently, he does not need to believe that Alice can steer Alice cannot transfer any information to Bob by choos- his state by her measurements and one also says that the ing her measurement directions since Bob’s reduced state state %AB is unsteerable or has a local hidden state (LHS) is independent of this choice. But, as Schrödingernoted, model. If such a model does not exist, Bob is required she can determine whether the wave function on his side to believe that Alice can steer the state in his laboratory is in an eigenstate of the Pauli matrix σx or σz. This by some “action at a distance”. In this case, the state steering of the wave function is, in Schrödinger’sown is said to be steerable. Note that steerability is an in- words, “magic”, as it forces Bob to believe that Alice herently asymmetric correlation, there are states where can influence his particle from a distance, see also Sec- Alice can steer Bob but not the other way round, see also tion V.M for details. Section III.D. The situation for general quantum states other than For the wave function in Eq. (1) the corresponding the singlet state can be formalized as follows (Wiseman assemblage is formed by the states |0ih0|/2, |1ih1|/2, et al., 2007): Alice and Bob share a bipartite quantum |x+ihx+|/2, and |x−ihx−|/2 and one can directly see state %AB and Alice performs different measurements, that no LHS model exists: The four conditional states 4 are, up to normalization, pure and thus cannot be mixtures of other states. Thus, the occurring normalized σλ have to be proportional to the four conditional states. So Eq. (2) implies that |ηihη|/2 = R dλp(λ)p(a|x, λ)σλ for all four |ηihη| and σλ coming from the set {|0ih0|, |1ih1|, |x+ihx+|, |x−ihx−|}. As mixtures are excluded, one must have p(a|x, λ) = 1 if the σλ corresponds to %a|x and therefore p(λ) = 1/2 for all λ. But then the probability distribution p(λ) cannot be normalized. For general states and measurements, however, the ex- FIG. 2 Inclusion relation between entanglement, steering, and Bell inequality violations. The set of all states is con- istence of an LHS model is not straightforward to decide. vex. The states which have an LHV model and therefore do This leads to the question of how one can decide for a not violate any Bell inequality form a convex subset. The given state %AB or a given assemblage {%a|x} whether it states with an LHS model are unsteerable and form a convex is steerable or not, and this is one of the central questions subset of the LHV states. Finally, the separable states are a of the present review article. convex subset of the LHS states.

C. Steering, Bell nonlocality and entanglement For our discussion, it is important that the measurements on separable states clearly can be explained by an There is another way to motivate the deﬁnition of LHV model. A general measurement Mx is given by a steering and steerable correlations as in Eq. (2). For that, positive operator-valued measure (POVM). This means we shortly have to explain the notions of local hidden that one considers a set of eﬀects Ea|x that are posi- variable (LHV) models and entanglement. tive operators, Ea|x ≥ 0, summing up to the identity P In a general Bell experiment, Alice and Bob perform a Ea|x = 11. The probability of the result a in a state measurements on their particles, denoted by Ax and By % is computed according to p(a) = Tr(%Ea|x). Applying and labeled by x and y. For the obtained results a, b one this to a separable state, one directly sees that the prob- asks whether their probabilities can be written as abilities of distributed measurements can be written as

Such a description is known as an LHV model: The hid- This is clearly an LHV model as in Eq. (3), with the ex- den variable λ occurs with probability p(λ) and Alice tra condition that the response functions p(a|x, λ) and and Bob can compute the occurring joint probabilities p(b|y, λ) are coming from the quantum mechanical de- with local response functions p(a|x, λ) and p(b|y, λ). For scription of measurements. a given ﬁnite number of settings x, y and outcomes a, b Having Eqs. (3) and (5) in mind, one may ask whether the probabilities that can be written as in Eq. (3) form the probabilities can also be described by a hybrid model, a high-dimensional polytope. The facets of the poly- where Alice has a general response function, while Bob’s tope are described by linear inequalities, the so-called function is derived from the quantum mechanical mea- Bell inequalities. Quantum states can result in proba- surement rule. That is, one considers probabilities of the bilities that violate the Bell inequalities, but deciding, form whether a given state violates a Bell inequality or not Z is not straightforward and subject of an entire ﬁeld of B p(a, b|x, y) = dλp(λ)p(a|x, λ)Tr(Eb|yσλ ). (6) research (Brunner et al., 2014). Let us now describe the notion of entanglement. In The point is that such probabilities are exactly the ones general, a state on a two-particle system is called sep- that occur in the steering scenario. By linearity we can arable, if it can be written as a convex combination of rewrite Eq. (6) as product states,

X A B p(a, b|x, y) = Tr(Eb|y%a|x), (7) %AB = pk%k ⊗ %k , (4) k R B where %a|x = dλp(λ)p(a|x, λ)σλ are the conditional otherwise it is called entangled. The separability of a states, allowing for an LHS model as in Eq. (2). quantum state is not easy to decide, except for systems We can conclude that the steering phenomenon relies consisting of two qubits or one qubit and a qutrit, where on quantum correlations which are between entanglement the method of the partial transposition gives a necessary and violation of a Bell inequality. In fact, any state that and sufficient criterion, see also Section III.E. violates a Bell inequality can be used for steering, and 5 any steerable state is entangled (see also Fig. 2). These This approach of detecting steerability has a natu- inclusions are strict in the sense that there are entangled ral connection to the task of entanglement verification states that cannot be used for steering, and there are (Gühneand Tóth,2009) and many concepts are similar steerable states that do not violate any Bell inequality. to the case of entanglement detection. This includes lin- In Section III.A we will discuss in detail the relation and ear criteria that are similar to entanglement witnesses, the known examples of states in the various subsets. criteria based on variances or entropic uncertainty rela- It is important to note that the indicated hierarchy rep- tions, and criteria similar to Bell inequalities. resents also different levels of trust in the measurement devices in entanglement verification. In general, in quantum information processing tasks such as cryptography, 1. Linear and nonlinear steering criteria it makes a difference whether or not one assumes that the measurement devices are well characterized. Com- Some of the typical ideas for deriving steering criteria pletely uncharacterized devices can be seen as a black are best explained with an example. Consider two qubits box, giving just some measurement results without any and the operator knowledge about the quantum description. There can Q = σ ⊗ σ + σ ⊗ σ + σ ⊗ σ . (8) also be situations where the devices are partly charac- x x y y z z terized, e.g., if the dimension of the quantum system is The question is which values hQi can have for separable known, but not the precise form of the measurement op- states. If one tries to maximize or minimize hQi over erators. separable states, it suffices to consider product states of Entanglement, steering, and Bell nonlocality corre- the form %A ⊗ %B, as these are the extreme points of spond to different levels of trust in the following sense. the separable states. But for product states the single The usual schemes of entanglement verification, such as expectation values factorize and one has (Tóth,2005) quantum state tomography or entanglement witnesses, require well-characterized measurement devices. The vi- |hQi| = |hσxiAhσxiB + hσyiAhσyiB + hσziAhσziB| olation of a Bell inequality, however, certifies the presence ~ of entanglement without any assumption on the measure- ≤ k~akkbk ≤ 1, (9) ments or dimension of the system. Steering is between ~ the two scenarios: If a state is steerable, its entangle- with ~a = (hσxiA, hσyiA, hσziA) and b defined analogously. ment can be verified in a one-sided device-independent Here, first the Cauchy-Schwarz inequality was used, and 2 2 scenario, where Bob’s measurements are characterized, then the fact that for single qubit states hσxi + hσyi + 2 but Alice’s not. In some cases, the assumptions on Bob’s hσzi ≤ 1 holds. For the singlet state, however, hQi = system can even be relaxed, see Section III.F for an ex- −3. So the operator W = 11 + Q is an entanglement ample. witness, as it has a positive mean value on all separable states, but is negative on some entangled states. If one wishes to estimate hQi for unsteerable states, II. DETECTION OF STEERING then, in view of Eq. (6), it suffices to consider product distributions again. This time, however, only Bob’s re- In this part of the review, we discuss how steering can sults are described by quantum mechanics,√ so only the ~ be verified in different scenarios. Mainly three cases can norm kbk ≤√1 is bounded, while k~ak = 3 is possible. be distinguished. First, given some expectation values of So, |hQi| ≤ 3 is a valid steering inequality that allows detecting the steerability of the singlet state (Cavalcanti the form hAi⊗Bji, one can ask whether these correlations can prove steerability. Second, one can consider the case et al., 2009). A possible modification and generalization is the fol- where Bob’s assemblage {%a|x} is given, and ask whether or not it can be explained by an LHS model. Finally, one lowing: Consider N measurements Ak on Alice’s side which can take the two values ±1 and arbitrary observ- can take a complete state %AB and ask whether this state allows seeing the phenomenon of steering if Alice makes ables Bk on Bob’s side. Then, for unsteerable states appropriate measurements. (Saunders et al., 2010)

N N X h X i |hAk ⊗ Bki| ≤ max λmax akBk , (10) A. Steering detection from correlations {ak} k=1 k=1

The simplest way to detect steering is to formulate where λmax(X) denotes the largest eigenvalue of X and criteria for the correlations between Alice’s and Bob’s ak = ±1. To prove this bound, it suﬃces to consider measurement statistics. These can then be directly eval- a product distribution as above, then each Ak can just uated in experiments, without the need of reconstructing change the signs of the Bk, and the mean value of the the whole assemblage. resulting sum is bounded by the maximal eigenvalue. 6

A diﬀerent kind of generalization uses expectation val- arising from ineﬃcient detectors. Theoretical aspects of ues on Bob’s side that are conditioned on Alice’s out- this issue are in detail discussed in (Evans et al., 2013; come. If Alice makes a measurement Ak with possible Jeon and Jeong, 2019; Vallone, 2013) and experimentally results labeled by a, one can denote with hBi|a the mean studied in (Bennet et al., 2012; Smith et al., 2012; Weston value of a measurement on Bob’s side, conditioned on the et al., 2018; Wittmann et al., 2012). outcome a. Then one can consider the nonlinear expression 2. Steering criteria from uncertainty relations (k) X 2 Tx = p(a|k)(hσxi|a) (11) a Steering inequalities based on uncertainty relations have been proposed already long before the formal def- and summing this up for three Pauli measurements gives inition of steerability in the context of the EPR argu- the bound for unsteerable states (Wittmann et al., 2012) ment (Reid, 1989; Reid et al., 2009). Also the criterion

(1) (2) (3) in Eqs. (11, 12) can be seen as a criterion in terms of Tx + Ty + Tz ≤ 1. (12) conditional variances. A systematic approach using entropic uncertainty re- This follows by considering product distributions and lations (EURs) has been proposed in (Walborn et al., hσ i2 + hσ i2 + hσ i2 ≤ 1. Note that similar bounds on x y z 2011) for continuous variable systems and tailored to dis- sums of squared mean values are known for many cases crete systems by (Schneeloch et al., 2013). Here, we fo- of anticommuting observables or mutually unbiased bases cus on the discrete version. In general, a measurement (Tóthand Gühne,2005; Wehner and Winter, 2010; Wu M results in a probability distribution P = (p , ··· , p ) et al., 2009), so one can directly generalize the criteria 1 n of the outcomes, for which one can consider the Shan- from above to broader classes of observables on Bob’s non entropy S(P) = − P p log(p ) as the entropy of side (Evans et al., 2013). With mutually unbiased bases i i i the measurement S(M). For two projective measure- as a generalization of the Pauli matrices one can even ments given by the corresponding hermitian operators find steering inequalities with an unbounded violation B = P λ |v ihv | and B = P µ |w ihw | on Bob’s (Marciniak et al., 2015; Rutkowski et al., 2017). 1 i i i i 2 i i i i side, one has the general EUR (Maassen and Uffink, For all the criteria from above the question arises, 1988) which are the best measurement directions for a given state in order to detect steering. For the criterion in S(B1) + S(B2) ≥ − ln(ΩB), (13) Eq. (10) this has been studied in (Evans and Wiseman, 2 2014). For criteria using Pauli matrices, one can still where ΩB = maxi,j(|hvi|wji| ) is the maximal overlap be- ask for the best orientation of the coordinate system. In tween the eigenstates. This and similar EURs are central (McCloskey et al., 2017) it has been observed that of- to quantum information theory and quantum cryptogra- ten, but not always, the measurements that correspond phy (Coles et al., 2017). to the semiaxes of the so-called steering ellipsoid (i.e., For product measurements A⊗B on two particles, one the ellipsoid of the potential conditional states in Bob’s can consider the joint distribution and the conditional Bloch sphere considering all possible measurements for Shannon entropy S(B|A) = S(A, B) − S(A). Then, for Alice, see Section V.B) give strong criteria. For higher- unsteerable states the relation dimensional systems, a systematic study of optimal mea- S(B |A ) + S(B |A ) ≥ − ln(Ω ) (14) surements in restrictive scenarios, i.e., in the context 1 1 2 2 B of N measurements of k outcomes, has been performed holds. The intuition behind this criterion is that if Al- in (Bavaresco et al., 2017). ice can predict from her measurement data Bob’s mea- So far, we have considered criteria that were motivated surement results better than the EUR allows, then there by concepts in entanglement theory. A different method cannot be local quantum states for Bob that reproduce to design steering inequalities for a given special scenario such measurement results. comes from the theory of semidefinite programs (SDPs). A generalization of this criterion to other entropies has As we will see in Section II.B, the question of whether a been developed (Costa et al., 2018a,b). The general ap- given assemblage {%a|x} is steerable can be decided via proach works for any entropy with the following prop- an SDP. The corresponding dual problem can then be erties: (i) the entropy is (pseudo-)additive for indepen- considered as a linear steering inequality. Further details dent distributions; (ii) one has a state independent EUR; are given in Section II.B.2. and (iii) the corresponding relative entropy is jointly con- The discussed criteria or small variations thereof have vex. The resulting criteria based on Tsallis entropy are been used in several experiments (Bennet et al., 2012; typically stronger than the ones from Shannon entropy. Saunders et al., 2010; Smith et al., 2012; Weston et al., In addition, Kriváchy et al. (2018) obtained tight steer- 2018; Wittmann et al., 2012). In the experimental works, ing inequalities in terms of the Rényi entropy for sce- it is also important to close loopholes, such as the one narios with two measurements per party, and Jia et al. 7

(2017) developed methods of detecting entanglement and For quantum steering one can consider also two mea- steering based on universal uncertainty relations and fine- surements with two outcomes per party, where only the grained uncertainty relations using majorization. measurements of Bob may be characterized. First, one In the case of continuous variable systems, the entropic can consider the case that Bob has a qubit and per- criteria proposed in (Walborn et al., 2011) are connected forms two mutually unbiased measurements (e.g., two to one of the first criteria mentioned above. In (Reid, Pauli measurements). For this scenario, it was shown 1989) the question is addressed to which extent Alice by (Cavalcanti et al., 2015b) that the full correlations can infer the value of Bob’s position XB or momentum hAiBji admit an LHS model, iff the inequality

PB by measuring her own canonical variables. The best p 2 2 h(A1 + A2)B1i + h(A1 + A2)B2i estimator of XB has as uncertainty the minimal variance 2 R 2 2 p 2 2 δmin(XB) = dxAp(xA)δ (xB|xA), where δ (xB|xA) is + h(A1 − A2)B1i + h(A1 − A2)B2i ≤ 2 (17) the variance of the conditional probability distribution. holds. Note that the resulting inequality and the un- Then, for quantum states that do not give rise to an EPR derlying problem has some similarity to Bell inequalities argument, the condition for orthogonal measurement directions for one or both 1 parties (Uffink and Seevinck, 2008). δ2 (X )δ2 (P ) ≥ (15) min B min B 4 For the more general scenario, one has to distinguish carefully whether the LHS model should explain the full holds. Later, Walborn et al. (2011) showed the criterion correlations hAiBji only, or in addition the marginal dis- S(XB|XA) + S(PB|PA) ≥ ln(πe) and demonstrated that tributions hAii and hBji. this implies Eq. (15). In addition, Walborn et al. (2011) Concerning full correlations, in (Girdhar and Caval- reports the experimental observation of states, which can canti, 2016) the case of uncharacterized projective mea- be detected by the entropic criterion, but not by Eq. (15). surements on Bob’s qubit has been considered. First, two Other experiments involving steering criteria from un- projective measurements B1 and B2 on a qubit define a certainty relations have been reported in the case of con- plane on the Bloch sphere, and in this plane one can al- tinuous variable systems in (Bowen et al., 2003), and re- ways find a third measurement B3 such that B1 and B3 cently in the case of discrete systems (Wollmann et al., are mutually unbiased; moreover, the mean values of B1 2019; Yang et al., 2019). and B2 can be obtained from the mean values of B1 and B3 and vice versa. Then, it was shown that B1 and B2 allow an LHS model iff Eq. (17) holds for B1 and B3. In 3. Steering and the CHSH inequality addition, it was shown that if Eq. (17) is violated, then the state violates also the original CHSH inequality and is Given a certain set of measurements, one can ask for thus nonlocal, but possibly for a different set of measure- the optimal inequalities for detecting quantum correla- ments (see also (Quan et al., 2017) for an independent tions. For Bell nonlocality and the case of two mea- proof). Finally, also a characterization of POVMs with surements with two outcomes each, it is known that the two outcomes has been given in (Girdhar and Cavalcanti, probabilities allow an LHV description, if and only if the 2016). Clauser-Horne-Shimony-Holt (CHSH) inequality At the same time, Quan et al. (2017) considered the question of whether full correlations and marginals of hA1B1i + hA1B2i + hA2B1i − hA2B2i ≤ 2 (16) two dichotomic measurements can be explained via an LHS model. For this case, the equivalence is not true holds (Fine, 1982), where also permutations of the mea- anymore: There are two-qubit states, which do not vi- surements and outcomes have to be taken into account. olate any CHSH inequality, nevertheless no LHS model More precisely, the inequality implies that the probabil- can explain the full correlations and marginals of certain ities of all outcomes for the measurements AiBj can be A1,A2,B1,B2. One can also find two-qubit states, for explained by a LHV model. Note that these probabili- which steerability from Alice to Bob can be proved by ties include more information than the full correlations two measurements on each side, but steering from Bob hAiBji only, as the marginals hAii and hBji are inde- to Alice is not possible (see also Section III.D). Finally, pendent of the full correlations. In other words, if the the interplay between steering and Bell inequality viola- CHSH inequality is fulfilled, there is also no two-setting tion for specific families of states has been discussed in Bell inequality with marginal terms that is violated. (Costa and Angelo, 2016; Quan et al., 2016). Similar statements are known from entanglement theory. For instance, one can consider the situation of two qubits, where Alice and Bob perform each the two mea- 4. Moment matrix approach surements σx and σz only, and not full tomography. For this scenario, all relevant entanglement witnesses have Another method that can be used for the characteriza- been characterized (Curty et al., 2004). tion of quantum correlations consists of moment matrices 8 or expectation value matrices. In general, one considers 5. Steering criteria based on local uncertainty relations a set of operators of the form Mk = {Aik ⊗ Bjk } and builds the matrix of expectation values Local uncertainty relations (LURs) are a common tool for entanglement detection and the underlying idea can † directly be generalized to steering detection. For the case Γkl = hMk Mli. (18) of entanglement, the idea is the following: Consider ob- The remaining task is to characterize the possible ma- servables Ak on Alice’s side, obeying an uncertainty re- P 2 2 2 2 trices Γ that origin from unsteerable or separable states. lation k δ (Ak) ≥ CA, where δ (X) = hX i + hXi Clearly, Γ ≥ 0, i.e., it has no negative eigenvalues. denotes the variance. An example of such a relation is P δ2(σ ) ≥ 2, for general observables such bounds This approach of moment matrices is a well-known tool i=x,y,z i can be computed systematically (Huang, 2012; Mac- in entanglement theory (Häseler et al., 2008; Miranowicz cone and Pati, 2014; Schwonnek et al., 2017). Simi- et al., 2009; Moroder et al., 2008; Shchukin and Vogel, larly, one can consider observables B for Bob, fulfill- 2005). There one can argue that the matrix Γ inherits a k ing P δ2(B ) ≥ C , and the global observables M = separable structure so that approaches using the partial k k B k A ⊗ 11 + 11 ⊗ B . Then, for separable states, the bound transposition can be applied. This also allows charac- k k P δ2(M ) ≥ C + C holds (Hofmann and Takeuchi, terization of entanglement if some of the entries Γ are k k A B kl 2003). This is a very strong entanglement criterion and not known or if the measurement devices are not trusted its properties have been studied in detail (Gittsovich (Moroder et al., 2013). et al., 2008; Gühne et al., 2006; Zhang et al., 2008). Concerning steering, it follows from Eqs. (5, 6) that For steering detection, we can use the same construc- the correlations of unsteerable states can be explained by tion, the only difference is that Alice’s measurements are an underlying separable state, where the measurements not characterized, so no uncertainty relation for them is of Alice are commuting (Kogias et al., 2015b; Moroder available (Ji et al., 2015b; Zhen et al., 2016). Conse- et al., 2016). The commutativity of Alice’s measure- quently, unsteerable states obey ments together with possible exploitation of the structure of Bob’s characterised measurements (e.g., an algebraic X 2 δ (Mk) ≥ CB. (20) structure such as the one of the Pauli spin operators) re- k sults in constraints on the moment matrix. In the end, for a given set of product operators {Mk}, one needs to The criterion of the LURs can be formulated in terms check whether there exist (complex) parameters for the of covariance matrices (Gühne et al., 2007), and this also unknown entries of the moment matrix (such as squares works for steering (Ji et al., 2015b). For a given quantum of Alice’s measurements) that make the matrix positive. state % and observables {Xk} the symmetric covariance As any moment matrix is positive, proving that such an matrix γ is defined by their elements γij = (hXiXji + assignment of parameters is not possible implies that the hXjXii)/2 − hXiihXji. If one considers in a composite underlying state is steerable. The main result of (Kogias system the set of observables {Xk} = {Aik ⊗ 11, 11 ⊗ Bjk } et al., 2015b) can be then formulated as follows. For any then the covariance matrix has a block structure unsteerable correlation experiment AC γ = , (21) AB CT B ΓR ≥ 0 for some R, (19) where A = γ(%A, {Ai}) and B = γ(%B, {Bj}) are covariance matrices for the reduced states and C is the corre- where ΓR is the moment matrix Γ for a set of parameters R (fulfilling the requirements inherited from commutativ- lation matrix with elements Cij = hAi ⊗ Bji − hAiihBji. ity on Alice’s side and possible further structure on Bob’s Given this type of covariance matrices for unsteerable side) as the unknown entries. In (Kogias et al., 2015b) states it holds that the authors further pointed out that checking the exis- γ ≥ 0 ⊕ κ , (22) tence of such parameters forms a semidefinite program AB A B and provided various examples. Note that this approach P with κB = k pkγ(|bkihbk|) being a convex combination can still be augmented by using the separable structure of covariance matrices of pure states on Bob’s system. of Γ. Here, 0A is a m × m null matrix, where m is the number In (Chen et al., 2016a) the concept of a moment ma- of observables on Alice’s side. The characterization of the trix has been used to characterize steerability in a more possible κB has been discussed for typical cases, such as refined way. Namely, one can also consider the moment qubit states or local orthogonal observables (Gittsovich matrices Γa|x for each state in the assemblage {%a|x}. Us- et al., 2008). ing the methods from (Moroder et al., 2013) this allows Finally, it should be noted that the criterion in Eq. (22) then to characterize and quantify steerability in a device is the discrete analog to a criterion for the continuous independent way. variable case, see also Section V.C. 9

B. Steering detection from state assemblages The latter means that there exists a set of (unnormalized) operators σλ satisfying When full knowledge of the unnormalized conditional X ensembles %a|x on Bob’s side is available, steerability can %a|x = D(a|x, λ)σλ for all a, x, be detected more efficiently. As already mentioned, a λ set of conditional ensembles on Bob’s side corresponding s.t.: σλ ≥ 0 for all λ. (23) to certain measurement settings from Alice is called a steering assemblage. As Alice’s choice of measurement Writing with the explicit definitions of the hidden vari- cannot be detected on Bob’s side, such assemblages are able λ and the deterministic response function D(a|x, λ), P P non-signalling in that a %a|x = a %a|x0 for different the equality in the above equation is simply 0 settings x, x . It is interesting to note that, for the bipar- X tite case, any non-signalling assemblage can be prepared %a|x = δa,ax σa1,a2,...,am . (24) with some shared state and some measurements on Al- {ai} ice’s side (Schrödinger,1936), see also Sections V.L and Intuitively, one can think of the hidden states σ V.M. Also, any unsteerable assemblage can be prepared a1,a2,...,am using a separable state and commuting measurements on as being indexed by m variables. The conditional state Alice’s side (Kogias et al., 2015b; Moroder et al., 2016). %a|x is obtained by summing the function over the values of all variables except for the x-th one, which is fixed The main point for steering detection is that for fi- a = a. nite steering assemblages, the question whether there ex- x To give an example, if one considers the case where Al- ists an LHS model described by Eq. (2) can be decided ice performs two measurements x ∈ {1, 2} with two pos- via the so-called semidefinite programming (SDP) tech- sible outcomes a ∈ {±}, the steering assemblage {% } nique (Pusey, 2013). The SDP approach also allows one a|x is unsteerable if and only if it is possible to find four pos- to derive steering inequalities and to quantify the steer- itive semidefinite operators ω , with i, j = ± such that ability of finite assemblages. ij

%+|1 = ω++ + ω+−,%+|2 = ω++ + ω−+, % = ω + ω ,% = ω + ω . (25) 1. Formulation of the semidefinite program −|1 −+ −− −|2 +− −− It is remarkable that in passing from Eq. (2) to The crucial idea is that for a finite steering assemblage Eq. (23), we have passed from an arbitrary hidden vari- {%a|x}a,x, it is sufficient to consider a finite LHS ensem- able to a finite discrete hidden variable and, at the same ble σλ (Ali et al., 2009; Pusey, 2013). Moreover, the time, fixed the response functions. One notices that the response functions in Eq. (2) can also be chosen to be finiteness of the set of measurements plays a crucial role fixed. So the remaining problem is to construct a finite in this approach. number of positive operators σλ. We focus on the concep- Given an assemblage {% } , determining the exis- tual aspects of the SDP formulation, a detailed review on a|x a,x tence of σλ satisfying Eq. (23) is in fact a well-known computational aspects can be found in (Cavalcanti and problem in convex optimization. More precisely, it is Skrzypczyk, 2017). known as a feasibility problem in semidefinite program- Consider a set of m measurement settings on Al- ming (SDP) (Boyd and Vandenberghe, 2004), which can ice’s side, x ∈ {1, 2, . . . , m}, each has q outcomes a ∈ be solved straightforwardly by an appropriate ready-to- {1, 2, . . . , q}. Given the shared state %, this gives rise use software. Furthermore, it has been shown that using to an assemblage {%a|x}a,x of m ensembles, each con- the so-called order-monotonic functions, the SDPs can sisting of q conditional states. The space of the hid- be approximated by simpler ones (Zhu et al., 2016). den variables λ can be constructed as follows. The variable λ can take qm values, each can be thought of as a string of outcomes ordered according to the measure- 2. Steering inequalities from the SDP ments, (ax=1, ax=2, ··· , ax=m). For such a string λ, we denote by λ(x) the value of the outcome at position x. The feasibility SDP (23) can be used to construct steer- Then D(a|x, λ) denotes the deterministic response func- ing inequalities. First, one can convert such a feasibility tion defined by D(a|x, λ) = δa,λ(x). This means that problem to an explicit convex maximization, D(a|x, λ) equals one for strings λ which predict the outcome a for the measurement x and zero otherwise. max µ

The crucial statement is the following: a ﬁnite steering w.r.t µ, {σλ} assemblage admits an LHS model described by Eq. (2) if X s.t. % = D(a|x, λ)σ for all a, x and only if it also admits an LHS model with the con- a|x λ structed set of strings λ as the LHV and the ﬁxed deter- λ 1 ministic functions D(a|x, λ) as the response functions. σλ ≥ µ for all λ. (26) 10

If the optimal value of µ turns to be negative, then the Q Q problem in Eq. (23) is infeasible, indicating that the as- P semblage is steerable. unsteerable unsteerable O To analyze this maximization, there is a powerful tool assemblages assemblages O P in convex optimization known as duality theory. In a all assemblages nutshell, the maximization problem in Eq. (26) is coupled all assemblages to a so-called dual minimization problem, FIG. 3 Geometrical illustrations of the steering weight (left) X and the steering robustness (right). Here P denotes the as- min Tr Fa|x%a|x semblage {%a|x} under consideration, Q denotes a general re- a,x alizable assemblage {γa|x} and O denotes an unsteerable as- w.r.t {F } LHS a|x semblage {%a|x }. The steering weight (left) seeks to minimize X p = P O/OQ for varying Q and O. The steering robustness s.t. F D(a|x, λ) ≥ 0 for all λ a|x (right) seeks to minimize t/(1 + t) = P O/P Q for varying Q a,x and O. X Tr Fa|xD(a|x, λ) = 1. (27) a,x,λ the considered assemblage, the boundary of unsteerable The two problems are dual in the sense that the optimal assemblages and the boundary of all assemblages through value of the minimization in Eq. (27) is an upper bound a certain line. Different ratios constructed from these dis- for the maximization in Eq. (26). This is known as weak tances give rise to different steerability quantifiers. duality. Under weak additional conditions, strong duality The first quantification of steerability of an assemblage also holds: the two optimal values are equal (Boyd and was proposed by (Skrzypczyk et al., 2014), known as Vandenberghe, 2004). steering weight. An assemblage {%a|x} is first written The duality theory implies that if there exists a col- as a convex combination of an unsteerable assemblage lection of observables {Fa|x} satisfying the constraints in {%LHS} and a general assemblage {γ }, P a|x a|x problem (27) and if Tr a,x Fa|x%a|x ≤ 0 then the assemblage is steerable. So, the dual problem naturally defines LHS a steering inequality. The minimizer of the problem (27) %a|x = pγa|x + (1 − p)%a|x for all a, x. (28) thus yields optimal steering inequalities for a steerable assemblage. Such steering inequalities found applications The steering weight of {%a|x}, denoted by SW({%a|x}), in several scenarios, see also Sections III.E and V.F. is the minimal weight p in such decomposition with respect to all possible choices of the general assemblage LHS {γa|x} and the unsteerable assemblage {%a|x }. A ge- 3. Quantification of steerability with SDPs ometrical illustration of the steering weight is given in Fig. 3 (left). As the set of all assemblages and unsteerable The SDP approach also allows one to quantify the assemblages can be characterized via SDPs, the steering steerability of an assemblage {%a|x}. There are several weight can also be determined by an SDP. More precisely, different quantification schemes. We selectively discuss SW({%a|x}) is given by some of those; for an extensive discussion we refer to (Cavalcanti and Skrzypczyk, 2017). X min 1 − Tr σλ The idea of quantifying the steerability of an assem- λ blage is as follows. Let us fix the number of measure- w.r.t. {σλ} ments m and the number of outcomes q per measurement X at Alice’s side. Then the space of all assemblages, i.e., all s.t. %a|x − D(a|x, λ)σλ ≥ 0 for all a, x different m decompositions of Bob’s reduced states to q- λ component ensembles, admits a natural convex structure. σλ ≥ 0 for all λ. (29) To be more precise, let {%a|x} and {%ã|x} be two assemblages, then for 0 ≤ p ≤ 1, the set {p%a|x + (1−p)˜%a|x} is A similar quantification of steerability is steering robust- also an assemblage. Within the set of all assemblages, the ness, first defined in the context of subchannel discrimi- unsteerable assemblages form a subset, which is clearly nation (Piani and Watrous, 2015); see also Section V.G. convex. Now, how much steerable an assemblage is can Here, the steering robustness SR({%a|x}) is given by the be measured by some kind of relative distance to the set minimal weight on a general assemblage {γa|x} consid- of unsteerable assemblages. More precisely, as long as ered as noise one needs to mix to the assemblage {%a|x} only the linear structure of the state assemblages is con- so that it becomes unsteerable. The geometrical illustra- cerned, the absolute distance is not meaningful, and one tion is given in Fig. 3 (right). Like the steering weight, can only consider relative ratios of distances on a line. In the steering robustness SR({%a|x}) can be computed via practice, one therefore compares the distances between a simple SDP (Cavalcanti and Skrzypczyk, 2016; Piani 11 and Watrous, 2015), convex hull of the original finite set of measurements, for which we have an LHS model. One thus also has an LHS X min Tr σλ − 1 model for the set of noisy measurements. Alternatively, λ the noise in the measurements can be also put onto the w.r.t. {σλ} state instead of the measurement set (Cavalcanti et al., X 2016; Hirsch et al., 2016b). Thus one can conclude an s.t. D(a|x, λ)σ − % ≥ 0 for all a, x λ a|x LHS model for the set of all measurements, but for a λ noisier version of the considered state. This construction σλ ≥ 0 for all λ. (30) works similarly for Bell nonlocality (Cavalcanti et al., 2016; Hirsch et al., 2016b). To compare the two quantifiers, we note that whereas the steering weight measures the unsteerable fraction in While the SDP approach has proven to be useful in a given assemblage, the generalized robustness measures algorithmically constructing certain LHS and LHV mod- the noise tolerance of an assemblage in terms of mixing. els (Cavalcanti et al., 2016; Cavalcanti and Skrzypczyk, Also, it can be shown that whereas robustness relates 2017; Fillettaz et al., 2018; Hirsch et al., 2016b), it has a to the task of subchannel discrimination (Section V.G), significant computational drawback. To reduce the noise steering weight relates to the task of subchannel exclu- needed to add to the state, the original finite set of mea- sion (Uola et al., 2019a). It also turns out that the steer- surements needs to be sufficiently large. However, the ing weight can also be understood as the maximal vio- size of the SDP, as one observes, increases exponentially lation of all possible steering inequalities with an appro- with respect to the number of measurement settings. As priate normalization (Hsieh et al., 2016). Moreover, one clearly illustrated in a systematic study (Fillettaz et al., can make a general comment on the relation between the 2018), this often imposes a significant computational dif- quantifiers that applies to general resource theories: any ficulty on the problem of deciding the steerability with extremal point has maximal weight, but the robustnesses high accuracy even for two-qubit states. can vary among different extremal points. Beyond the steering weight and steering robustness, Ku et al. (2018a) defined a geometric quantifier C. Steering detection from full information based on the trace distance between a given assemblage and its corresponding closest assemblage admitting an When the complete density matrix %AB is exploited, LHS model. Also, a device-independent quantification of one might expect to have a more complete characteriza- steerability was proposed by Chen et al. (2016a). This tion, i.e., a necessary and sufficient condition for steer- method is based on assemblage moment matrices, a col- ability. Like entanglement detection or Bell nonlocality lection of matrices of expectation values, each associated detection, this is a difficult question. There were exact re- with a conditional quantum state; see also Section II.A.4. sults only for entanglement detection of low-dimensional Finally, a different quantifier is given by the critical ra- or special states. It is thus very encouraging that some dius, as explained in Section II.C.1. exact results can also be derived for quantum steering. It was recognized already by Wiseman et al. (2007) that for certain highly symmetric states, the problem of de- 4. From finite to infinite number of measurements termining steerability with projective measurements can be solved completely, see Section III.B. Recently, a com- While the SDP approach was originally designed to plete characterization of quantum steering has been also construct LHS models when Alice is limited to a finite set achieved for two-qubit states and projective measure- of measurements, one can also draw certain conclusions ments (Jevtic et al., 2015; Nguyen et al., 2019; Nguyen for the case where Alice has an infinite number of mea- and Vu, 2016b). surements (Cavalcanti et al., 2016; Hirsch et al., 2016b). The idea is as follows. One starts with a finite set of measurements on Alice’s side and constructs an LHS model 1. Two-qubit states and projective measurements as described above. In fact, one obtains a bit more. The outcome is an LHS model not only for the original fi- From Eq. (2), one sees that in order to determine the nite set of measurements but for all measurements in its steerability of a given state one has to consider all pos- convex hull. sible LHS ensembles {p(λ), σλ}, and for each measure- Then, one considers the set of all measurements, typ- ment, solve for the response functions p(a|x, λ). The ically limited to projective ones. One can add certain source of difficulty is that the possible choice of the in- noise to the set of measurements, e.g., by sending them dexing hidden variable λ seems to be arbitrary: it can through a depolarizing channel and obtains a new set of be a discrete variable, a real-valued variable or a multi- noisy measurements. For a certain level of noise added, dimensional variable, etc. It is now worth re-examining the set of noisy measurements will shrink to fit inside the how the SDP approach discussed in Section II.B works: 12 one assumes that Alice can only make a finite number of measurements, which implies the finiteness of a necessary LHS ensemble – a unique choice of the hidden variable is thus singled out. When Alice’s set of measurements is not finite, this approach breaks down. Fortunately, one can show that (Nguyen et al., 2018, 2019) for quantum steering, there is a canonical choice of the indexing hidden variable, namely Bob’s pure states. This is also true for higher-dimensional systems. In fact, an LHS ensemble can be identified with a probability distribution (or to be more precise, a probability measure) µ over Bob’s pure states SB. Our above discussion implies that the LHS model FIG. 4 The operational meaning of the critical radius: 1 − Eq. (2) can be written as R(%AB ) measures the distance from % to the surface of unsteerable/steerable states relatively to (11A ⊗ %B )/2. Figure Z taken from (Nguyen et al., 2019). %a|x = dµ(σ)˜p(a|x, σ)σ, (31) SB for a certain choice ofp ˜(a|x, σ) (Nguyen et al., 2018). R(%AB) measures the distance from the given state to the Note that the response functionp ˜(a|x, σ) may have no surface that separates steerable states from unsteerable simple relation to p(a|x, λ) in Eq. (2) if the association states; see Fig. 4. As a consequence, one can in fact equivalently define the critical radius as λ to σλ is not injective; see (Nguyen et al., 2018) for the detailed discussions. (α) R(% ) = max{α ≥ 0 : % is unsteerable}, (34) Consider now a system of two qubits. To proceed, let AB AB us for now fix an LHS ensemble µ. Given an LHS ensem- where unsteerability is considered with respect to projec- ble µ, one still faces with the problem of solving Eq. (31) tive measurements. We will see that this definition can be forp ˜(a|x, σ) for all possible measurements x. The next naturally generalized to generalized measurements and step is to abandon this constructive approach; instead, higher-dimensional systems. one only determines a condition for this equation to have The definition of the critical radius by Eq. (33) also a solution. To this end, for a given LHS ensemble µ, one allows for its evaluation. It is shown that (Nguyen et al., defines (Nguyen et al., 2019; Nguyen and Vu, 2016a,b) 2018) by replacing Bob’s Bloch sphere by polytopes from R inside and from outside, one obtains rigorous upper and dµ(σ)| TrB(Cσ)| SB lower bounds for the critical radius. Both the compu- r(%AB, µ) = min √ , (32) C 2kTrB[¯%AB(11A ⊗ C)]k tation of the upper and lower bounds are linear programs, of which the sizes scale cubically with respect where% ¯AB = %AB − (11A ⊗ %B)/2, the norm is given by to the numbers of vertices of the polytopes. Upon in- p kXk = Tr(X†X), and the minimization runs over all creasing the numbers of vertices, the two bounds quickly single-qubit observables C on Bob’s space. In fact, the converge to the actual value of the critical radius. This quantity r(%AB, µ) characterizes the geometry of the set approach has been used to access the geometry of the of conditional states Alice can simulate from the LHS en- set of unsteerable states via its two-dimensional random semble µ, and is called the principal radius of µ (Nguyen cross-sections (Nguyen et al., 2019); see Fig. 5. and Vu, 2016a). It can then be shown that (Nguyen Notably, for the so-called Bell-diagonal states, or T - et al., 2019; Nguyen and Vu, 2016a) Eq. (31) has a so- states, an explicit formula for the critical radius has been lutionp ˜(a|x, σ) for x running over all possible projective obtained, measurements if and only if r(%AB, µ) ≥ 1. So far, a fixed choice of LHS ensemble µ is made. One R(%T ) = 2πNT | det(T )|, (35) now defines the critical radius as the maximum of the where T is the correlation matrix of the T -state, T = principal radius (32) over all LHS ensembles, ij Tr(%T σi ⊗σj) for i, j = 1, 2, 3, and the normalization factor NT is given by an integration over the Bloch sphere R(%AB) = max r(%AB, µ). (33) µ −1 R T −2 −2 NT = dS(~n)[~n T ~n] (Jevtic et al., 2015; Nguyen and Vu, 2016b). Based on this solution for T -states, an- Then a two-qubit state is unsteerable if and only if alytical bounds for the critical radius of a general state R(%AB) ≥ 1. can also be derived (Nguyen et al., 2019). (α) Remarkably, let %AB = α%AB + (1 − α)(11A ⊗ %B)/2, For further discussions on LHS models for two-qubit (α) −1 then R(%AB) = α R(%AB). This relation also gives an states in special cases, see (Miller et al., 2018; Yu et al., operational meaning to the critical radius, namely 1 − 2018a,b; Zhang and Zhang, 2019). 13

tain class of measurements in the same way as Eq. (34). In particular, considering the set of generalized measurements (POVMs) of n outcomes, one can deﬁne the critical radius for a bipartite state ρAB of dimension dA × dB by

(α) Rn(%AB) = max{α ≥ 0 : %AB is unsteerable}, (37)

(α) with %AB = α%AB + (1 − α)(11A ⊗ %B)/dA, %B = TrA(%AB) and unsteerability being considered with respect to POVMs of n outcomes on Alice’s side. Defined separable entangled but two-way unsteerable in this way, 1 − Rn(%AB) can still be interpreted as mea- one-way steerable two-way steerable numerical uncertainty suring the distance from %AB to the surface separating unsteerable/steerable states, here defined with respect to FIG. 5 Two two-dimensional random cross-sections of the set POVMs of n outcomes on Alice’s side; see again Fig. 4. of all two-qubit states. From the inner most to the outer most, However, direct evaluation of the critical radius from the different areas with different colors denote the set of separable definition Eq. (37) is clearly not possible. states characterized by the partial transposition (Horodecki Interestingly, an alternative formula for the critical ra- et al., 1996; Peres, 1996b), entangled states that are unsteerable, one-way steerable states (Alice to Bob or vice versa), dius in similarity to Eq. (32) and Eq. (33) (Nguyen et al., and two-way steerable states (Alice to Bob and vice versa). 2018, 2019) can also be found for high dimensional sys- The very thin grey lines at the two boundaries of the area cor- tems. To this end, for a finite dimensional bipartite state responding to the one-way steerable states denote those states %AB, one can define the principal radius for a given LHS where the used numerical precision was not sufficient to make ensemble µ by an unambiguous decision. Figure taken from (Nguyen et al., 2019). −1 −1 rn (%AB, µ) = sup F (%AB, µ, Z, E), (38) Z,E

−1 2. Steering of higher-dimensional systems and with generalized with F (%AB, µ, Z, E) defined to be measurements Pn 1 Pn Tr[%AB(Ei ⊗ Zi)] − Tr(Ei)Tr(%BZi) i=1 dA i=1 R 1 Pn , Characterizing quantum steering of higher- dµ(σ) maxi{hZi, σi} − Tr(Ei)Tr(%BZi) dA i=1 dimensional systems and with generalized measurements (39) (POVMs) is difficult. Most of the results on quantum where the supremum is taken over all possible n-POVMs steering, in this case, rely on the idea of adding suffi- E = (E1,E2,...,En) on Alice’s side and all possible n cient noise to the state such that an LHS for simpler observables Z = (Z1,Z2,...,Zn) on Bob’s side. The measurements (e.g., projective measurements) can be critical radius as defined by Eq. (37) can be computed turned into an LHS model for POVMs (Hirsch et al., as (Nguyen and Gühne,2020; Nguyen et al., 2019) 2013; Quintino et al., 2015; Tischler et al., 2018). More −1 −1 specifically, if a state %AB of dimension dA × dB is Rn (%AB) = min rn (%AB, µ). (40) µ unsteerable with respect to two-outcome POVMs, then the state In this way, the problem of computing the critical radius and the princpal radius is in principle an optimization 1 dA − 1 %ÃB = %AB + σA ⊗ %B (36) problem. Unfortunately, even in this form, a determin- d d A A istic algorithm to compute the principal radius and the with an arbitrary choice of state σA and %B = TrA(%AB) critical radius with n ≥ 3 is still unknown, and one has is unsteerable for arbitrary POVMs. One observes that to invoke heuristic techniques in practice (Nguyen et al., in Eq. (36), the weight (dA −1)/dA of the separable noise 2018, 2019). σA ⊗ %B is close to 1 if the dimension is high. Yet, this One observes that to study quantum steering, the set technique has played an important role in demonstrating of generalized measurements was stratified according to the hierarchy of nonlocality under generalized measure- their number of outcomes. This calls for an investiga- ments (see Section III.A), superactivation of nonlocality tion of the relation between them. Since POVMs of n by local filtering (see Section III.G), and one-way steering outcomes form a natural subset of POVMs of n + 1 out- with POVMs (see Section III.D). comes, one has a decreasing chain R2(%AB) ≥ R3(%AB) ≥ 2 Remarkably, the critical radius approach explained in R4(%AB) ≥ · · · . As extreme POVMs have at most dA the previous subsection gives a promising framework to outcomes with dA being Alice’s dimension, this chain 2 generally analyze quantum steering with POVMs and turns to equality at n = dA. Using the evidence from higher-dimensional systems. In fact, in any dimension, a heuristic computation for the principal radius, it has one can define the critical radius with respect to a cer- been conjectured (Nguyen et al., 2019) that for two-qubit 14

2 states (dA = 4), the chain in fact consists of a single num- Note that the steering inequalities (42) and (43) ber, namely R2(%AB) = R3(%AB) = R4(%AB). In words, are different from various inequalities discussed in Sec- this conjecture implies that measurements of two out- tion II.A in the sense that they exploit the full informa- comes (dichotomic measurements) are sufficient to fully tion about the state. demonstrate the quantum steerability of a two-qubit system; measurements of more outcomes are not necessary. Unfortunately, extrapolating this conjecture to higher- III. CONCEPTUAL ASPECTS OF STEERING dimensional systems fails; it is later shown (Nguyen and Gühne,2020) that this equality breaks down already for In this section we review results on the general prop- a system of two qutrits (see also Section III.B). erties and structures of quantum steering. We start with a detailed discussion on the connection between steering, entanglement, and Bell nonlocality. We also present in 3. Full information steering inequality some detail LHS models for different families of states. Then, we explain properties like one-way steering, steer- As we discussed, for high-dimensional systems, even ing of bound entangled states, steering maps and the su- when full information about a state is available, a com- peractivation of steering. putable necessary and sufficient condition for quantum steerability is not available. In this case, the detection of steerability still relies on steering inequalities. An ex- A. Hierarchy of correlations ample of steering inequalities based on full information of the state is given by Zhen et al. (2016) in terms of We explained already in Section I.C that there is a the so-called local orthogonal observables. Embedding hierarchy between Bell nonlocality, steering, and entan- in a higher-dimensional space if necessary, we can as- glement in the sense that one implies the other, but not sume that Alice and Bob have the same local dimension the other ways around. In this section, we first discuss in 2 d. One can choose a set of d orthogonal operators {Gk} some detail the known examples of states where the no- which serves as a basis for the local observable space, tions differ. Then we explain how the relations between i.e., Tr(GiGj) = δij and {Gk} spans the space of opera- the three concepts can be exploited to characterize one tors (Yu and Liu, 2005). The Pauli matrices are a familiar via another. Detailed LHS models are discussed in Sec- example of such orthogonal operators for a qubit system. tion III.B. By means of the Schmidt decomposition in the operator When discussing the existence of an LHV or LHS space, one can choose the orthonormal observables for model for a given quantum state, one has to distinguish A B the local spaces at Alice and Bob, {Gk } and {Gk }, such whether the model should explain the results for all pro- that the joint state %AB can be written as jective measurements, or, more generally, for all POVMs. Let us start our discussion with projective measurements. d2 The inequivalence between the notion of entanglement X A B %AB = λkGk ⊗ Gk , (41) and Bell nonlocality was, in fact, one of the starting k=1 points of entanglement theory (Werner, 1989). For that, one may consider the so-called two-qubit Werner state where λk ≥ 0. Then using the local uncertainty relations (see Section II.A), Zhen et al. (2016) showed that the 11 %(p) = p|ψ−ihψ−| + (1 − p) (44) state %AB is steerable from A to B if 4

2 √ d where |ψ−i = (|01i − |10i)/ 2 is the singlet state. Using X 2 A B δ (gkGk + Gk ) < d − 1 (42) the PPT criterion (see Section III.E) one can directly ver- k=1 ify that this state is entangled iff p > 1/3. In (Werner, for some choice of g , where δ2(X) denotes the variance 1989), however, an LHS model for projective measure- k ments was constructed for all values p ≤ 1/2. Moreover, of operator X. By a particular choice of gk, one can easily show that if in (Ac´ın et al., 2006; Hirsch et al., 2017) it was shown that an LHV model exists up to p ≤ 1/K (3) ≈ 0.6829, where √ G X KG(3) is the Grothendieck constant of order three, so up λk > d, (43) to this value no Bell inequality can be violated. These re- k sults demonstrate that there are entangled states which the state is steerable (Zhen et al., 2016). This elegant in- do not show Bell nonlocality. Using the fact that the equality resembles the familiar computable cross norm or Werner states is steerable for p > 1/2 (Wiseman et al., realignment (CCNR) entanglement criterion (Chen and 2007), this also proves that steering and Bell nonlocality P Wu, 2003; Rudolph, 2005), where k λk > 1 implies that are inequivalent for projective measurements. States of the state is entangled. this type have been prepared experimentally and their 15 steerability has been demonstrated in (Saunders et al., B. Special states and their local hidden state models 2010). It remains to discuss the more general case of POVMs. As already highlighted in Section II.C, the fact that First, in (Barrett, 2002) an LHV model for Werner states for quantum steering, there is a canonical choice for the with p ≤ 5/12 has been constructed which explains all hidden variable, namely Bob’s pure states, turns out to the measurement probabilities for arbitrary POVMs. In have far-reaching consequences. The point is that, hav- fact, this model can directly be converted into an LHS ing simplified the LHS model from Eq. (2) to the form model (Quintino et al., 2015). Consequently, there are of Eq. (31), the symmetry of the state has stronger im- entangled states for which all correlations for POVMs can plications on the choice of the LHS ensemble (Nguyen be explained by an LHV model. In addition, (Quintino et al., 2018). For certain highly symmetric states such as et al., 2015) presented examples of states in a 3 × 3 sys- the Werner states and the isotropic states, the symmetry tem which are steerable in both directions, but neverthe- is then enough to uniquely single out an optimal choice less an LHV model for all POVMs can be found. This of the LHS ensemble, rendering their exact characteriza- proves the inequivalence of Bell nonlocality and steering tions of quantum steering with projective measurements for POVMs. possible (Jones et al., 2007; Wiseman et al., 2007). This As mentioned, a local model that explains the results of is in contrast to Bell nonlocality: in Eq. (3), no canonical all projective measurements does not necessarily explain choice of the LHV is possible. Thus even for highly sym- all the correlations for POVMs. It is not clear, however, metric states such as the isotropic states and the Werner that POVMs give an advantage in the detection of steer- states, no exact characterization of Bell nonlocality is ing or Bell nonlocality. As discussed in Section II.C.2, known. there is numerical evidence that two-qubit states which are unsteerable for projective measurements are also unsteerable for POVMs (Nguyen et al., 2019). Concerning 1. Werner states Bell nonlocality, in (Gómez et al., 2016; Vértesi and Bene, 2010) Bell inequalities have been presented, for which the Suppose Alice and Bob share the Werner state of di- maximal violation requires POVMs, but this does not im- mension d × d (Werner, 1989), defined by ply that the states leading to this violation do not violate η d − 1 + η 11 η V also some Bell inequality for projective measurements. W = − , (47) d d − 1 d2 d − 1 d Given the similarity in the definitions of Bell nonlocality, quantum steerability and nonseparability, one may where 11 is the bipartite identity operator and V is the expect that some methods to characterize the different flip operator given by V |φ, ψi = |ψ, φi. Here we follow notions can be related to each other. Specifically, given the parameterization by Wiseman et al. (2007) so that η a state that admits an LHV model as in Eq. (3), one Wd is a product state if the mixing parameter η = 0 may expect that by adding suitable separable noise to and is a state at all only if η ≤ 1. The Werner state is 1 the state, one can obtain a state that admits an LHS entangled if and only if η > d+1 (Werner, 1989). model (2). This has been shown to be the case (Chen In fact, the Werner state was constructed in a way et al., 2018a). The authors showed that if a bipartite such that it is invariant under the same local unitary qudit-qubit state %AB admits an LHV model, then the transformation at Alice’s and Bob’s side (Werner, 1989), state that is, for any unitary operator U acting in dimension η η † † 11 d, W = (U ⊗ U)W (U ⊗ U ). This implies that the %˜ = µ% + (1 − µ)% ⊗ B , (45) d d AB AB A 2 optimal LHS ensemble on Bob’s Bloch sphere can be cho- √ sen to be symmetric under the unitary group U(d), i.e., with µ = 1/ 3 is unsteerable from Alice to Bob. Turning the Haar measure (Nguyen et al., 2018; Wiseman et al., the logic around, if% ÃB is steerable, then %AB must be 2007). Bell nonlocal. A similar statement between steerability Identifying the hidden variable λ indexing the LHS and nonseparability has also been obtained (Chen et al., with Bob’s pure states |λi, it remains to construct the 2018a; Das et al., 2019). Namely, if a bipartite qubit- response function p(a|x, λ) for a projective measurement qudit state %AB is unsteerable from Alice to Bob, then {E } = {P } to complete an LHS model. Note that the state a|x a|x for a projection outcome Pa|x = |aiha| at Alice’s side, 11 %˜ = µ% + (1 − µ) A ⊗ % , (46) Bob’s conditional state is AB AB 2 B √ d − 1 + η 11 η % = − |aiha|. (48) with µ = 1/ 3 is separable. Or, if the latter is entan- a|x d(d − 1) d d(d − 1) gled, the former is steerable. Detailed applications of this approach to detecting different nonlocality notions from The minus sign in front of the last term indicates that the others can be found in (Chen et al., 2018a, 2016; Das the two parties in the Werner state are anti-correlated. et al., 2019). To construct the response function, it is natural then to 16

1 0.6 These are thus concrete examples illustrating that quantum steering with dichotomic measurements is strictly 0.4 weaker than that of measurements with more outcomes η0.5 for higher-dimensional systems, contrasting with the con- 0.2 jecture on their equivalence for two-qubit systems (see Section II.C). 0 0.0 5 10 15 20 5 10 15 20 d d 2. Isotropic states FIG. 6 The exact noise thresholds for for the Werner states (left) and the isotropic states (right) to be steerable with Another important family of states that allows for dichotomic measurements (square, violet, upper), projective measurements (square, green, middle), and to be separable exact characterization of quantum steering is that of (square, orange, lower). The lower bounds for the noise isotropic states (Wiseman et al., 2007). The isotropic thresholds for them to be unsteerable with all generalized state of dimension d×d at mixing parameter η, 0 ≤ η ≤ 1, measurements obtained from Eqs. (58) and (58) are also pre- is defined by sented (triangle, red). Figure adapted from (Nguyen and Gühne,2020). 11 Sη = (1 − η) + η|ψ ihψ |, (52) d d2 + + √ Pd associate a pure state |λi to the outcome that has the where |ψ+i = 1/ d i=1 |i, ii. From the definition, one least overlap with Pa|x. The resulting response function notes that the isotropic state is defined with respect to a is particular choice of basis. The isotropic state is entangled if and only if η > 1/(d + 1) (Horodecki and Horodecki, 1 if |hλ|ai| < |hλ|a0i| ∀a0 6= a p(a|x, λ) = (49) 1999). In similarity to the Werner state, the isotropic 0 otherwise. η ¯ state also has a unitary symmetry, namely, Sd = (U ⊗ η ¯ † † With the LHS ensemble and this choice of the response U)Sq (U ⊗ U ) for any d × d unitary matrix U, with ¯ function, it is straightforward (Wiseman et al., 2007) to U being its complex conjugate. This again implies that show that the Werner state is unsteerable for the optimal choice of LHS ensemble is the uniform Haar measure over Bob’s Bloch sphere. 1 η ≤ 1 − . (50) For a projection outcome Pa|x = |aiha| on her side, d with an isotropic state, Alice steers Bob’s system to the It can also be shown that for the mixing parameter conditional state 1 η > 1 − d , no construction of response function is pos- 1 − η 11 η % = + |a¯iha¯|, (53) sible (Wiseman et al., 2007). This threshold together a|x d d d with the threshold for the Werner state to be separable is presented in Fig. 6 (left). Thus, whereas the exact where |a¯i is the complex conjugate of state |ai. In con- threshold of η for which the Werner state is Bell nonlocal trast to Eq. (48), the plus sign in front of the last term is still unknown even in dimension d = 2, the threshold in the above equation indicates that parties sharing an for steerability has an analytical expression in all dimen- isotropic state are correlated upto a complex conjuga- sions. Wiseman et al. (2007) also noted that the con- tion. This motivates the following choice of the response struction given above was actually the original construc- function tion by Werner (1989) to show that Bell nonlocality and 1 if |hλ|a¯i| > |hλ|a¯0i| ∀a0 6= a entanglement are distinct notions. p(a|x, λ) = (54) 0 otherwise, Projective measurements are however not the only case where the quantum steerability of the Werner states where we have also again identified the hidden variable can be exactly characterised. Specifically, it was shown λ indexing the LHS ensemble with Bob’s pure states |λi. that (Nguyen and Gühne,2020) when Alice is limited This construction leads to an LHS model for the isotropic to making dichotomic measurements, the threshold upto state with which the Werner state is unsteerable can also be derived H − 1 in practically closed form, η ≤ d , (55) d − 1 η ≤ (d − 1)2[1 − (1 − 1/d)1/(d−1)], (51) where Hd = 1 + 1/2 + 1/3 + ··· + 1/d. It can for d ≤ 105; see Fig. 6 (left). The threshold is also con- again be shown that this threshold is optimal; for jectured to hold for all dimensions (Nguyen and Gühne, η > (Hd − 1)/(d − 1) no construction for the response 2020). Interestingly, for dimension d ≥ 3, the thresh- function is possible (Wiseman et al., 2007). Remark- old Eq. (50) is strictly stronger than that of Eq. (51). ably, Almeida et al. (2007) also obtained this threshold in 17 an attempt to construct an LHV model for the isotropic arbitrary POVMs on Alice’s side if (Almeida et al., 2007; states before learning of the definition of quantum steer- Barrett, 2002) ing. This threshold is presented in Fig. 6 (right) together with that for separability. 3d − 1 η ≤ (d − 1)d−1d−d. (58) Like the Werner state, the quantum steerability of the d + 1 isotropic states with dichotomic measurements can also be exactly characterized. It was shown in (Nguyen and As we mentioned, this construction was originally sug- 5 Gühne,2020) that for d ≤ 10 , if gested as an LHS model for Werner states and the same η ≤ 1 − d−1/(d−1), (56) threshold Eq. (58) was found (Barrett, 2002; Quintino et al., 2015). However, it was shown (Nguyen and Gühne, the isotropic state is unsteerable when Alice’s measure- 2020) that for the Werner state, a better choice of the re- ments are limited to dichotomic ones; otherwise, it is sponse functions is possible, namely steerable (see Fig. 6 (right)). The threshold is also conjectured to hold for all dimensions (Nguyen and Gühne, αa|x 2 2 p(a|x, λ) = (1 − |hλ|ai| )Θ(1/d − |hλ|ai| ) 2020). d − 1 ! αa|x X αa|x 2 2 + 1 − (1 − |hλ|ai| )Θ(1/d − |hλ|ai| ) . d d − 1 3. LHS model for POVMs a (59) As quantum steering with projective and dichotomic measurements is well-understood for the Werner states The Werner state was then shown to be unsteerable for and the isotropic states, one may hope that certain LHS arbitrary POVMs on Alice’s side if (Nguyen and Gühne, models with general POVMs for them can also be con- 2020) structed. This is indeed the case. By an explicit construction, Barrett (2002) demonstrated that sufficiently 1 + (d − 1)d+1d−d weakly entangled Werner states do admit an LHV model η ≤ . (60) for all POVMs. Under the light of the formal defini- d + 1 tion of quantum steering (Wiseman et al., 2007), the LHV model turns out to be an LHS model (Quintino The two bounds Eq. (58) and Eq. (60) are also pre- et al., 2015). The model was revised recently (Nguyen sented in Fig. 6. For the Werner states, the bound and Gühne,2020), and it can be shown that Barrett’s Eq. (60) is strictly better than the bound given by original construction works best for the isotropic states; Eq. (58) for d ≥ 3. However both bounds Eq. (58) for the Werner states, a better model can be constructed. and Eq. (60) are strictly within the respective thresh- Further, for two-qubit systems, the construction can also olds for the isotropic states and the Werner states to be be extended to Bell-diagonal states (Nguyen and Gühne, unsteerable with projective measurements, Eq. (55) and 2019). Eq. (50). On the other hand, the constructions Eq. (57) To construct the model, it is sufficient to consider and Eq. (59) are by no means optimal; in fact, it is ex- only POVMs with rank-1 effects, {Ea|x} = {αa|x|aiha|}, pected to be not optimal (Nguyen and Gühne,2019). where |aiha| are rank-1 projections and 0 ≤ αa|x ≤ Thus it is still unclear at the moment if steering with 1 (Barrett, 2002). This is because other POVMs can projective measurements is equivalent to steering with be post-processed from these (see also Section IV). The generalized measurements even for these highly symmet- optimal choice for the LHS ensemble is again the uni- ric states. form distribution over Bob’s Bloch sphere (Nguyen et al., The case of two-qubit Werner states (d = 2) is slightly 2018; Wiseman et al., 2007). It is then left to construct better understood. In this case, both the bounds Eq. (58) 5 the response functions p(a|x, λ) for the mentioned mea- and Eq. (60) show that for η ≤ 12 , the Werner state is surements. unsteerable for arbitrary POVMs on Alice’s side (Barrett, 5 1 For the isotropic state, the response function can be 2002; Quintino et al., 2015). In the range 12 ≤ p ≤ 2 , given as (Almeida et al., 2007; Barrett, 2002) the state is also known to be unsteerable if the POVMs

2 2 are limited to those with 3 outcomes (Werner, 2014). For p(a|x, λ) = αa|x|hλ|a¯i| Θ(|hλ|a¯i| − 1/d) most general POVMs, numerical evidences based on the ! critical radius approach are available, which indicate that αa|x X 2 2 + 1 − α |hλ|a¯i| Θ(|hλ|a¯i| − 1/d) . d a|x the state is also unsteerable in this range (Nguyen et al., a 2018, 2019). (57) To conclude this section, we refer the interested readers With this choice of the response function, direct compu- to (Augusiak et al., 2014) for further constructions of tation shows that the isotropic state is unsteerable for LHV and LHS models. 18

C. Steering and local ﬁltering

For characterizing steerability and other correlations in quantum states, it is relevant to study their behaviour under local operations. Given a general quantum state %AB one can consider states of the type

1 %˜ = (T ⊗ T )% (T † ⊗ T † ), (61) AB N A B AB A B where TA/B are some transformation matrices and N denotes a potential renormalization. Then, one can ask whether the correlations in the state %AB are related to those of the state% ÃB. Clearly, this depends on the properties of the matrices T . For FIG. 7 The border of the one-way steerable area for the fam- A/B ily of states given by Eq. (63). The thickness of the border them there are mainly two possible choices: Either one for steering from B to A indicates the uncertain area. The restricts them to be unitary TA/B = UA/B and therefore inner bound for the border of steering from B to A with the one considers local unitary transformations. Or one con- uniform LHS ensemble as an ansatz is also included (dotted siders general invertible matrices TA/B = FA/B, which line). are the so-called local filtering operations and are more general than local unitaries. D. One-way steerable states For the case of entanglement one can directly see from the definition in Eq. (4) that local filtering oper- The asymmetry between the two parties in the defi- ations keep the property of a state being separable or nition of quantum steering immediately strikes one with entangled. Filtering operations can, nevertheless, change the question whether there is a state where Alice can the amount of entanglement. Any full rank state can steer Bob, but not the other way around (Wiseman et al., be brought into a normal form under filtering opera- 2007). Such one-way steerable states were first con- tions, where the reduced states% Ã and% ˜B are maxi- structed for continuous variable systems (Midgley et al., mally mixed (Leinaas et al., 2006; Verstraete et al., 2003). 2010; Olsen, 2013). One-way steerable states for discrete In this form, certain entanglement measures are maxi- systems were studied later by Bowles et al. (2014); Evans mized (Verstraete et al., 2003) and bringing a state in and Wiseman (2014); Skrzypczyk et al. (2014). More re- this normal form can improve many entanglement crite- cently a simpler family of one-way steerable two-qubit ria (Gittsovich et al., 2008). For Bell nonlocality, it can states was identified in (Bowles et al., 2016). This family be seen that local unitary transformations keep the prop- of states is given by erty of a state having an LHV model. But local filtering operations are already too general, there are two-qubit 11 %AB(α, θ) = α|ψθihψθ| + (1 − α) ⊗ %B, (63) states that do not violate the CHSH inequality, but after 2 local filtering, they do (Gisin, 1996; Popescu, 1995). where |ψθi = cos θ|00i + sin θ|11i and %B = TrA|ψθihψθ|, Steering is a notion between entanglement and nonlo- with 0 ≤ α ≤ 1 and 0 < θ ≤ π/4. The states can cality, so a mixed behaviour under local transformations be brought into the two-qubit Werner states with the can be expected. In fact, it was noted in (Gallego and same mixing probability α by a local filtering on Bob’s Aolita, 2015; Quintino et al., 2015; Uola et al., 2014) that side and a local unitary on Alice’s side. Therefore it local unitaries on Alice’s side and local filtering on Bob’s is steerable from Alice to Bob if and only if α > 1/2 side, (see Section III.C). Using the uniform distribution as an ansatz for the LHS ensemble, (Bowles et al., 2016) 1 † † showed that the state is unsteerable from Bob to Alice %ÃB = (UA ⊗ FB)%AB(UA ⊗ FB), (62) 2 2α−1 N for cos (2θ) ≥ (2−α)α3 . With the complete characterization of steerability for two-qubit states described in do not change the steerability of a state. Also the criti- Section II.C.1, the boundary of the set of unsteerable cal radius as a steering parameter (see Section II.C.1) is states from Bob to Alice has been obtained with high not affected. With these transformations one can achieve accuracy (Nguyen et al., 2019); see Fig. 7. It is clearly that% ˜B is maximally mixed on its support. If a state is in visible from the figure that %AB(α, θ) is one- way steer- this form, this can simplify calculations and therefore it able for a large range of parameters. is good starting point to study the steerability of a state The one-way steering phenomenon can also be shown (Nguyen et al., 2019). to persist when the measurements are extended to 19

POVMs (Quintino et al., 2015). The idea to construct %TB ≥ 0, such states are also called PPT states. It an example is as follows. One first embeds a state which was further proven that for systems consisting of two is unsteerable from Alice to Bob with respect to projec- qubits (2×2-systems) or one qubit and one qutrit (2×3- tive measurements, but steerable for the other direction, system) this criterion is sufficient for separability and all into a higher dimension on Alice’s side. One then con- PPT states are separable. For all other dimensions, PPT structs a state that admits an LHS model for all POVMs entangled states exist, these states are, in some sense, performed on Alice’s side using Eq. (36) with the state weakly entangled, as they cannot be used for certain σA chosen to be supported only in the extended dimen- quantum information tasks. sion on Alice’s side. With this choice of σA, it is easy The main quantum information task where PPT en- to show that the state is still steerable from Bob to Al- tangled states are useless is the task of entanglement dis- ice (Quintino et al., 2015). The constructed state is thus tillation. Entanglement distillation is the process where one-way steerable also when one considers all POVMs. many copies of some noisy entangled state are distilled to The one-way steering phenomenon also attracts at- few highly entangled pure states via local operations and tention from the experimental side. Early experi- classical communication (Horodecki et al., 1998). Sur- ments demonstrating one-way steering were carried out prisingly, not all entangled states can be used for dis- for continuous variable systems and Gaussian measure- tillation, and these undistillable states are called bound ments (Händchen et al., 2012). The effects of various entangled. It was shown that PPT entangled states are types of noise on the direction of steering were later an- bound entangled, but there are some NPT states, for alyzed and probed experimentally by Qin et al. (2017). which it has been conjectured that they are also bound Experiments demonstrating one-way steering for discrete entangled (Pankowski et al., 2010). systems were performed by Sun et al. (2016); Wollmann In 1999, Peres formulated the conjecture that bound et al. (2016); Xiao et al. (2017). Sun et al. (2016) entangled states do not violate any Bell inequality (Peres, and Xiao et al. (2017) concentrated on demonstrating 1999). This conjecture was based on an analogy between one-way steering when measurements are limited to two a general distillation protocol and Bell inequalities for and three settings. Wollmann et al. (2016) also demon- many observers, but for a long time no proof could be strated the persistence of the phenomena for POVMs. found. In 2013 the conjecture was made that bound en- Most recently it was realized (Baker et al., 2018) that ex- tangled states are also useless for steering (Pusey, 2013; isting experiments demonstrating one-way steering com- Skrzypczyk et al., 2014). This so-called stronger Peres mitted certain assumptions on the states or the mea- conjecture could potentially open a way to prove the orig- surements and were therefore inconclusive. A conclusive inal Peres conjecture, especially as the PPT criterion and experiment (Tischler et al., 2018) was then performed the question of steerability are closely related to SDPs. shortly after. It was shown by (Moroder et al., 2014), however, that some bound entangled states can be used for steering, and an explicit example for two qutrits was given. The E. Steering with bound entangled states idea to find the counterexample is the following. For a given state of two qutrits and two measurements with Now we discuss the steerability of so-called bound en- three outcomes each one can decide the steerability of the tangled states. This provides a relevant example for the assemblage {%a|x} with an SDP (see also Section II.B.1). fact that the characterization of steering gives new in- Considering the dual formulation of the SDP, one finds sights into old problems in entanglement theory. that the operator Before presenting the result, we have to recall some facts about the entanglement criterion of the positivity W =A1|1 ⊗ Z13 + A2|1 ⊗ Z23 + A1|2 ⊗ Z31 + A2|2 ⊗ Z32 of the partial transpose (PPT criterion) and entangle- + (A + A − 11) ⊗ Z (65) ment distillation. Let us start with the PPT criterion 3|1 3|1 33 (Horodecki et al., 1996; Peres, 1996b). Generally, for a defines a steering inequality, that is, Tr(%W ) ≥ 0 two-particle state % = P % |iihj|⊗|kihl| the partial ij,kl ij,kl for unsteerable states. Here, the A are arbi- transposition with respect to Bob is defined as a|x trary measurement operators for Alice and the set

TB X {Z13,Z23,Z31,Z32,Z33} consists of five positive opera- % = %ij,lk|iihj| ⊗ |kihl|. (64) tors, obeying the four semidefinite constraints Zi3 +Z3j − ij,kl Z33 ≥ 0 for i, j ∈ {1, 2}. Similarly, one can define a partial transposition with re- Given this steering inequality, one can look for steer- spect to Alice which obeys %TA = (%TB )T . Note that the able PPT states by an iteration of SDPs: One starts partial transposition may change the eigenvalues of a ma- with a random initial steerable state % and fixes Alice’s trix, contrary to the full transposition. measurements Aa|x to be measurements in two mutually The PPT criterion states that for separable states unbiased bases. Then, by optimizing the Zij via an SDP the partial transposition has no negative eigenvalues, one can minimize the mean value Tr(%W ) and find the 20

relations ω+− = %+|1 − ω++, ω−+ = %+|2 − ω++, and ω−− = %B − %+|1 − %+|2 + ω++. Of course, this is only a valid solution if all ωij are positive semideﬁnite. Then, one takes four positive deﬁnite operators Zij with i, j = ±, which obey the relation Z++ = Z+− + Z−+ − Z−−, and considers the bipartite operator X ΣAB = Zij ⊗ ωij. (66) ij

FIG. 8 Inclusion relation between the PPT states and entan- This is, after appropriate normalization, a separable state glement, steering, and Bell inequality violations. Separable as in Eq. (4). The point is that with the given relations states are PPT, but some entangled states are PPT as well. on the ωij and Zij this state can be written as PPT entangled states are bound entangled, as no pure state entanglement can be distilled from them. There exist, how- ΣAB = Z+− ⊗%+|1 +Z−+ ⊗%+|2 +Z−− ⊗(%B −%+|1 −%+|2) ever, PPT states that can be used for steering and also PPT (67) states that violate Bell inequalities. These states are coun- as can be verified by direct inspection. Here, all the de- terexamples to the Peres conjecture. pendencies on the ωij dropped out, so ΣAB is uniquely determined by the assemblage and the Zij only. Also the optimal steering inequality W . Given this W , one can required normalization follows directly from Eq. (67). ask for the minimal expectation value of it with respect From this the desired connection to the separability to PPT states, this is again an SDP. Having found the problem follows: Given an unsteerable assemblage and PPT state with the smallest Tr(%W ) one can optimize operators Zij obeying the conditions from above, the state ΣAB in Eq. (67) is separable. Moreover, one can over the Zij again and then iterate. In practice this procedure converges quickly towards PPT states which are show the opposite direction: If the assemblage is steer- steerable, delivering the desired counterexamples to the able, then there exist a set of operators Zij such that the stronger Peres conjecture. state ΣAB is entangled. In this case, the entanglement Having found the counterexamples, it is a natural ques- of ΣAB can even be detected by a special entanglement tion whether these states also violate a Bell inequality. witness, namely the flip operator. Indeed, as has been shown by (Vértesiand Brunner, The statement can be generalized to an arbitrary num- 2014), these states are also counterexamples to the orig- ber of measurements and outcomes (Moroder et al., inal Peres conjecture. Finally, (Yu and Oh, 2017) pre- 2016). In fact, it is related to the dual of the original sented an analytical approach, giving explicit families of SDP. PPT entangled states in any dimension d ≥ 3 which, for This reformulation of the steerability problem can give appropriate parameters, violate Bell inequalities or can insights in the detection of steering, if the measurements be used for steering, see Fig. 8. on Bob’s side are not fully characterized, but only the dimension of the space where the measurements act on is known. The core idea is the following: For any bipar- F. Steering maps and dimension-bounded steering tite state %AB and sets of local orthonormal observables A B X X Gk and Gl [that is, Tr(Gi Gj ) = δij for X ∈ {A, B}] A B In this section, we describe how the steering problem one can build the matrix Λkl = Tr(%ABGk ⊗ Gl ). Then, can be viewed as a certain kind of separability problem the computable cross-norm or realignment (CCNR) cri- (Moroder et al., 2016). This allows to apply the power- terion states that if %AB is separable, then the trace ful techniques of entanglement theory (Gühneand Tóth, norm is bounded by one, kΛk1 ≤ 1 (Chen and Wu, 2009; Horodecki et al., 2009), and study problems such 2003; Rudolph, 2005), see also Section II.C.3. This crite- as the detection of steering, if Bob’s system is not well rion has already been used to detect entanglement with characterized and only its dimension is known. uncharacterized devices, if the dimension is known: If To formulate the main idea it suffices to consider the Alice and Bob make uncharacterized measurements Ak case of two measurements (x ∈ {1, 2}) with two out- and Bl they can build the expectation value matrix comes (±) on Alice’s side. As discussed in Section ∆kl = Tr(%ABAk ⊗Bl) and, using the dimension assump- II.B.1, steerability of the assemblage {%a|x} can be de- tion, from that estimate the trace norm kΛk1 (Moroder cided by an SDP. More precisely, Eqs. (25) states that and Gittsovich, 2012). the assemblage is unsteerable, if one finds four posi- A similar approach can be used for steering (Moroder tive semidefinite operators ωij with i, j = ± such that et al., 2016). For a choice of Zij one considers the state %+|1 = ω+++ω+−,%+|2 = ω+++ω−+,%−|1 = ω−++ω−−, ΣAB. Then, on Alice’s side one takes a set of local A and %−|2 = ω+− + ω−−. These equations are not inde- orthogonal observables Gk and for Bob’s side unchar- pendent. If one takes ω++ as a free variable, one has the acterized measurements Bl and builds an expectation 21 value matrix, which can be used to estimate whether standing of the phenomenon (Hsieh et al., 2016; Quintino ΣAB violates the CCNR criterion. If this is the case, et al., 2016). In fact, Quintino et al. (2016) and Hsieh then the original assemblage was steerable. The re- et al. (2016) extended the results of (Cavalcanti et al., sulting criteria are strong: For two-qubit Werner states 2013a) to show that the steerability of all unsteerable − − %(p) = p|ψ ihψ | + (1 − p)11/4 and Pauli measurements states %AB that satisfy the so-called reduction criterion σx, σy and σz one can evaluate from the data the steer- for entanglement (Horodecki and Horodecki, 1999) can ing inequality√ in Eq. (12). It detects steerability for be superactivated. The reduction criterion states that p > 1/ 3, which is the same threshold as the steering if 11A ⊗ %B − %AB is not positive, then the state %AB inequality. So, for this case the approach allows to draw is nonseparable. As satisfying the reduction criterion is the same conclusion from the resulting data, but with- a necessary and sufficient condition for a two-qubit or out assuming that the measurements were correct Pauli a qubit-qutrit state to be nonseparable (Horodecki and measurements, the only assumption that is made is that Horodecki, 1999), the steerability of all entangled states Bob’s space is a qubit. of dimension 2 × 2 or 2 × 3 can be superactivated. Their idea is based on the exact threshold for quantum steering of the isotropic state in Eq. (55). For con- G. Superactivation of steering venience, here the isotropic state is reparametrized as 11 − |φ ihφ | Let us come back to the formulation of quantum steer- Sf = f|φ ihφ | + (1 − f) + + , (68) d + + d2 − 1 ing as a simulation task where Alice tries to convince Bob that she can steer his system from a distance as with the same notation as defined in Eq. (53) and discussed in Section I. Note that in this protocol, Alice f = 1 − (1 − 1/d2)(1 − η). According to Eq. (55), the has to prepare a large number of pairs of particles in a isotropic state in Eq. (68) is unsteerable if and only if 2 certain state. One of the particles in each pair is then f ≤ [(1 + d)Hd − d]/d . It is known (Horodecki and sent to Bob. Note that it is crucial for Alice to prepare Horodecki, 1999) that a state that violates the reduction many copies of the state, so that Bob later on can do criterion can be brought into an entangled isotropic state tomography to verify the steered states on his side. Al- with f > 1/d by local filtering on Bob’s side and the so- ice then declares the set of measurements she can make, called isotropic twirling operation. As we mentioned in or equivalently the assemblage she can steer Bob’s sys- Section III.C, local filtering on Bob’s side does not change tem to. To maximize her steering ability, Alice clearly the steerability of the state. The isotropic twirling opera- should choose the largest set of measurements. Most tion consists of averaging the state under certain random often, Alice’s measurements are assumed to be projec- local unitary transformations, thus also does not increase tive measurements (or POVMs) on separated particles the steerability. So it is sufficient to show that the steer- f on her side. This, however, is not yet the maximal set ability of the isotropic state Sd with f > 1/d can be of measurements she can do. As Alice has prepared a superactivated. This again can be shown by observing f ⊗n large number of bipartite states, she can actually make that the isotropic twirling on (Sd ) yields the isotropic n f n n f ⊗n collective measurements on several particles on her side. state Sdn of dimension d ×d . Thus (Sd ) is steerable We will see that when such collective measurements are if considered, the steerability of a state may change. More n n 1/n [(1 + d )Hdn − d ] precisely, for an unsteerable (but entangled) state %AB, f > . (69) one asks whether there exists a finite number n such that d2 ⊗n %AB is steerable. In this case, we say that the quantum At large n, the right-hand side asymptotically approaches steerability of %AB can be superactivated. 1/d. Therefore, whenever f > 1/d, there exists n such For nonseparability, a similar question is answered triv- that the inequality is satisfied, or equivalently the steer- f ially negative for any states, but for Bell nonlocality, it ability of Sd can be superactivated. has been extensively investigated since (Peres, 1996a). Beyond states that violate the reduction criterion, one For Bell nonlocality, the confirmative answer was first may ask if quantum steerability, or more generally, Bell obtained by Palazuelos (2012) and later refined by Cav- nonlocality can always be superactivated for arbitrary alcanti et al. (2013a). The authors showed that indeed entangled states. This question remains as a challenge for a certain state %AB which admits an LHV model, for future research. If this were the case, the hierarchy ⊗n for sufficiently large n, %AB can violate some Bell in- of quantum nonlocality would be unified into a single equality. Note that this is distinct from the notion of su- concept (Cavalcanti et al., 2013a). peractivation of bound entanglement (Shor et al., 2003). Besides the notion of superactivation of quantum non- While the superactivation of Bell nonlocality investigated locality via collective measurements on multiple copies by the mentioned authors also implies the ability to su- of the state as described above, there is also the notion peractivate quantum steering, exact characterizations of of superactivation of quantum nonlocality via local fil- quantum steering also significantly simplify the under- tering (on both sides). The phenomenon is dated back 22 to (Popescu, 1995), who showed that the Werner states sically processed to match those of the original pair. Fur- in dimension d ≥ 5 that admit LHS models for projec- ther fine-tunings of measurement incompatibility have tive measurements can violate a Bell inequality after ap- been presented in the literature, e.g. coexistence, broad- propriate local filtering. Recently, Hirsch et al. (2013) castability, and non-disturbance (Busch et al., 2016; showed that there are states which admit an LHS model Heinosaari, 2016; Heinosaari and Wolf, 2010), see also for POVMs but become Bell nonlocal after appropri- Section IV.E. Typically all the incompatibility related ate local filtering. However, Hirsch et al. (2016a) later extensions of non-commutativity (on a single system) co- showed that there are also entangled states whose quan- incide with non-commutativity for projective measure- tum nonlocality cannot be superactivated by local filter- ments, but for the case of POVMs they form a strict ing. hierarchy (Heinosaari and Wolf, 2010). It is worth mentioning that in the process matrix formulation of POVMs, even commuting process POVMs can be incompatible IV. JOINT MEASURABILITY AND STEERING (Sedlák et al., 2016). For investigating steering from the measurement per- In this section we discuss the problem of joint mea- spective, the notion of joint measurability appears most surability to which steering is related in a many-to-one fitting. A set {Aa|x}a,x of POVMs (i.e. positive oper- manner (Kiukas et al., 2017; Quintino et al., 2014; Uola ators summing up to the identity for every x) is said et al., 2015, 2014). Joint measurability is a natural ex- to be jointly measurable if there exists a POVM {Gλ}λ tension of commutativity for general measurements. Op- together with classical post-processings {p(a|x, λ)}a,x,λ erationally it corresponds to the possibility of deducing such that the statistics of several measurements from the statistics X of a single one. The connection between the concepts Aa|x = p(a|x, λ)Gλ. (70) of joint measurability and steering unlocks the technical λ machinery developed within the framework of quantum The POVM {Gλ}λ is called a joint observable or a joint measurement theory to be used in the context of quantum measurement of the set {Aa|x}a,x. correlations. More precisely, joint measurability has been To give an example of a set of jointly measurable studied extensively already a few decades before steering POVMs, one could use a mutually commuting pair of was formulated in its modern form. We review the con- POVMs in which case a joint measurement is given by a nection on three levels: joint measurability on Alice’s side POVM whose elements are products of the original ones. (pure states), on Bob’s side (mixed states), and on the For a more insightful example, we take a pair of noisy level of the incompatibility breaking quantum channels Pauli measurements defined as (Choi isomorphism). Moreover, we discuss in detail how known results on one field can be mapped to new ones µ 1 S := (11 ± µσx) (71) on the other. ±|x 2 1 Sµ := (11 ± µσ ), (72) ±|z 2 z A. Measurement incompatibility where 0 < µ ≤ 1. The question is now how to find can- didates for a joint measurement. For the above pair, an Measurement incompatibility manifests itself in var- educated guess [i.e. a candidate with similar symmetry ious operationally motivated forms in quantum the- as the pair (Sµ ,Sµ )] gives ory. Maybe the best-known notion is that of non- ±|x ±|z commutativity. Here by non-commutativity we mean 1 Gµ := (11 + iµσ + jµσ ), (73) the mutual non-commutativity of the POVM elements i,j 4 x z of two POVMs, i.e. for POVMs {Aa}a and {Bb}b we ask whether [Aa,Bb] = 0 for all a, b or not. From text books where i, j ∈ {−, +}. One notices straight away that on quantum mechanics we know that non-commutativity Sµ = Gµ + Gµ (74) of observables places certain restrictions on the variances ±|x ±,+ ±,− of the measured observables. Such restrictions do not, µ µ µ S±|z = G+,± + G−,±, (75) however, give any further operational insight into the involved measurements, they just follow from simple math- i.e. there exist (deterministic) post-processings that give ematics. the original measurements. The last thing to check is µ One possible operationally motivated extension of that {Gi,j}i,j forms a POVM. As the normalisation fol- commutativity is that of joint measurability. Namely, lows from the definition, one is left with checking the pos-√ one can ask whether two measurements can be per- itivity of the elements, which is equivalent to µ ≤ 1/ 2. formed simultaneously (or jointly), i.e. whether there It can be shown that this is indeed the optimal threshold exists a third measurement whose statistics can be clas- for joint measurability in our example, i.e. beyond this 23

µ µ threshold the POVMs {S±|x} and {S±|z} do not admit To demonstrate a possible use of this result, one can a joint measurement (Busch, 1986). consider a steering scenario where Alice performs mea- The above example shows that joint measurability is surements on a noisy isotropic state. This noise in the indeed a proper generalization of commutativity. In the state can be translated to Alice’s measurements by writ- literature, many such examples have been discussed in ing finite and continuous variable quantum systems (Busch 1 µ µ 1 et al., 2016). A typical question is as above: how much tr[(Aa|x ⊗ 1)%AB] = tr[(Aa|x ⊗ 1)%AB], (80) noise can be added until measurements become jointly µ + + (1−µ) µ measurable. For small numbers of measurements and where %AB = µ|ψ ihψ | + d2 11 and Aa|x = µAa|x + outcomes, this can be efficiently checked with SDP (Uola (1−µ) d tr[Aa|x]11 with µ ∈ [0, 1]. For different sets of mea- et al., 2015; Wolf et al., 2009). For more complicated sce- surements on Alice’s side one can either solve the steer- narios various optimal and semi-optimal analytical and ability by using known incompatibility results or vice numerical techniques have been developed (Bavaresco versa. To give an example, consider the known (Wise- et al., 2017; Designolle et al., 2019; Heinosaari et al., man et al., 2007) steerability threshold for the noisy 2016; Kunjwal et al., 2014; Uola et al., 2016). isotropic state (with projective measurements) µ∗ = Pd 1 ( n=1 n − 1)/(d − 1). Using Eq. (80) one sees that for any µ > µ∗ there exists a set of projective measurements B. Joint measurability on Alice’s side that remains incompatible with the amount µ of white noise. On the contrary, any set of projective measure- Comparing the definition of joint measurability with ∗ that of unsteerability, one recognizes similarities. In- ments with an amount µ ≤ µ of white noise results in deed, joint measurability is a question about the exis- an unsteerable state assemblage (with the isotropic state) tence of suitable post-processings and a common POVM, and, hence, such a set is jointly measurable. whereas unsteerability asks the existence of suitable re- It is worth noting that the connection between incom- sponse functions and a common state ensemble. To patibility of Alice’s measurements and steerability of the make the connection exact, we recall the main result of resulting assemblage is strongly motivated by a similar (Quintino et al., 2014; Uola et al., 2014): work on non-locality. In (Wolf et al., 2009) incompatibility of Alice’s measurements was proven to be equivalent A set of measurements {Aa|x}a,x is not jointly measurable if and only if it can be used to demonstrate steering to the ability of violating the CHSH inequality (when op- with some shared state. timizing over Bob’s measurements and the shared state). To be more precise, using a jointly measurable set of This connection, however, is known not to be true in P general. In (Bene and Vértesi,2018; Hirsch et al., 2018) observables on Alice’s side, i.e. Aa|x = λ p(a|x, λ)Gλ, and a shared state % results in a state assemblage counterexamples for the non-locality connection in sce- AB narios with more measurement settings are presented, X i.e. there exist sets of measurements that are not jointly % = p(a|x, λ)trA[(Gλ ⊗ 11)%AB] (76) a|x measurable but always lead to local correlations. In con- λ X trast, joint measurability and steering can both be de- = p(a|x, λ)σλ, (77) scribed in terms of operational contextuality (Tavakoli λ and Uola, 2019). The CHSH inequality is a criterion for this type of contextuality. It is an open question whether where we have written σλ = trA[(Gλ ⊗ 11)%AB]. Hence, the existence of a joint observable for Alice’s measure- there are other contextuality inequalities that fully char- ments implies the existence of a local hidden state model. acterise incompatibility. In (Tavakoli and Uola, 2019) For the other direction, using a full (finite) Schmidt rank numerical evidence is provided that a specific family of P contextuality criteria generalising the CHSH inequality state |ψi = λi|iii one has i characterise the incompatibility of sets of binary qubit T %a|x := trA[(Aa|x ⊗ 11)|ψihψ|] = CAa|xC, (78) measurements. Such characterisation, among any other, is directly applicable to steerability of state assemblages P T by the use of the techniques presented in the following where C = j λj|jihj| and X is the transpose of the operator X in the basis {|ii}i. Assuming that the as- subsection. semblage {%a|x}a,x has a local hidden state model one gets C. Joint measurability on Bob’s side X −1 −1 Aa|x = p(a|x, λ)C σλC , (79) λ The connection between steering and joint measurements presented in the above section is based on the use −1 −1 from which it is clear that {C σλC }λ forms the de- of full Schmidt rank states (on finite-dimensional sys- sired joint measurement of {Aa|x}a,x. tems). To loosen the assumption on purity of the state, 24 we recall the main result of (Uola et al., 2015): The question of steerability (of a state assemblage) is a non-normalised version of the joint measurement problem. More precisely, by normalising a state assem- ˜ blage {%a|x}a,x one gets abstract POVMs Ba|x := −1/2 −1/2 P %B %a|x%B , where %B = a %a|x and a pseudo- inverse is used when necessary. Note that we use tilde to distinguish between Bob’s actual measurements and the normalised state assemblage (which consists of abstract POVMs on a possibly smaller dimensional system than the one Bob’s measurements act on). It is straight-forward to show (Uola et al., 2015) that the state assemblage {%a|x}a,x is steerable if and only ˜ if the abstract POVMs {Ba|x}a,x are not jointly measurable. Namely, as the normalisation keeps the post- processing functions fixed, the local hidden states map to joint measurements of the normalised assemblage and FIG. 9 Regions of the parameters λ, r, θ allowing for steering, joint measurements map to local hidden states. detected by the inequality (82) (inner region) and inequality + Such connection broadens the set of techniques that (84) (outer region), with r = k~r k and θ being the angle + + are translatable between the fields of joint measurability between ~r and the z axis, and β = 0.45 (fixed). Inset: representation in the Bloch sphere of the reduced states % and steering. In general, joint measurability criteria map ±|1 (green points one upon the other) and %+|2 (red point on into steering criteria and vice versa. To give an example, + the right). The normalization factor β = Tr[%+|2] is not we take a well-known joint measurability characterisation represented. The figure is taken from (Uola et al., 2015). of two qubit POVMs (Busch, 1986). Namely, take two qubit POVMs of the form 1 (11 ± λσ ) and % = β±11 ± r±σ we present a com- 1 4 z ±|2 z A := (11 ± ~a · ~σ), (81) parison of these criteria in Fig. 9 (Uola et al., 2015). ±|x 2 x As another example, we demonstrate how steering ro- where x = 1, 2. This pair is jointly measurable if and bustness defined in Eq. (30) translates to an incompat- only if ibility robustness (Uola et al., 2015). Recall that the steering robustness SR(%a|x) of an assemblage {%a|x}a,x k~a1 + ~a2k + k~a1 − ~a2k ≤ 2. (82) can be written as

Note that this criterion is necessary for joint measura- min t ≥ 0 bility in the more general case, i.e. for pairs of POVMs %a|x + tγa|x given as s.t. unsteerable for all a, x. (85) 1 + t 1 A+|x := ((1 + αx)11 ± ~ax · ~σ),A−|x = 11 − A+|x, (83) Here the optimization is over assemblages {γa|x}a,x and 2 positive numbers t, see also Fig. 3. Mapping a state where α ∈ [−1, 1] and k~a k ≤ 1 + α . In (Busch and assemblage into a set of POVMs, one can deﬁne an in- x x x ˜ ˜ Schmidt, 2010; Stano et al., 2008; Yu et al., 2010) a nec- compatibility robustness IR(Ba|x) for a set {Ba|x}a,x as essary and suﬃcient criterion for joint measurability of (Uola et al., 2015) such pairs is given as min t ≥ 0 α2 α2 ˜ 2 2 1 2 2 Ba|x + tTa|x (1 − F1 − F2 ) 1 − 2 − 2 ≤ (~a1 · ~a2 − α1α2) , s.t. jointly measurable for all a, x. F1 F2 1 + t (84) (86)

1 p 2 2 p 2 2 with Fi = 2 ( (1 + αi) − k~aik + (1 − αi) − k~aik ), Here the optimisation is performed over POVMs for i = 1, 2. {Ta|x}a,x and positive numbers t. Note that this defi- The criteria in Eq. (82) and Eq. (84) are both steer- nition works also for a generic set of POVMs, i.e. the ing inequalities. The latter of them characterises all POVMs do not need to originate explicitly from a steer- pairs of binary unsteerable assemblages in the qubit ing problem. To give the incompatibility robustness case. Labelling members of such assemblages by %±|1 = IR(Aa|x) of a finite set of POVMs {Aa|x}a,x in the stan- 25 dard SDP form one writes (Uola et al., 2015) correspondence can be extended to infinite-dimensional systems (Kiukas et al., 2017) resulting in a fully general P G0 min tr[ λ λ ] connection between steering and joint measurements: d A state assemblage {%a|x}a,x given by Alice’s measure- X 0 s.t. D(a|x, λ)Gλ ≥ Aa|x for all a, x ments {Aa|x}a,x and a state % is steerable if and only if † λ the POVMs {Λ (Aa|x)}a,x are not jointly measurable. 0 Gλ ≥ 0 for all λ Note that although the notation here is adapted to the X 0 X 0 case of discrete POVMs (to avoid technicalities), the con- Gλ = 11 tr[ Gλ]/d, (87) nection works also for POVMs with continuous outcome λ λ sets. where {D(a|x, λ)}a,x,λ are deterministic post- To demonstrate the power of the above result, we list processings. some of its implications (Kiukas et al., 2017). First, the To finish this section, we stress out that even though connection is quantitative in the sense that the incompat- † the normalised state assemblages appear as abstract ibility robustness of {Λ (Aa|x)}a,x coincides with the so- POVMs that do not have a direct connection to the called consistent steering robustness (i.e. a special case POVMs actually measured in the correlation experiment, of steering robustness: one allows mixing only with as- the normalised state assemblages can be interpreted as semblages that have the same total state as the original the POVMs Alice should measure on the canonical pu- assemblage) of {%a|x}a,x. Second, for pure states the cor- P responding channel Λ† is unitary, hence, extending the rification of a %a|x in order to prepare {%a|x}a,x. Hence, we see that the results on steering and joint measure- main results of (Quintino et al., 2014; Uola et al., 2014) ments stated in the Section IV.B (i.e. non-jointly mea- presented in Section IV.B to the infinite-dimensional surable POVMs allow steering with some shared state) case. Third, the result characterises unsteerable states as are closely related to the connection presented here. those whose corresponding Choi-Jamiolkowski channel is incompatibility breaking (i.e. outputs only jointly measurable observables). Finally, seemingly different steer- D. Incompatibility breaking quantum channels ing problems (such as NOON states subjected to photon loss and systems with amplitude damping dynamics) can The extension of the connection between steering and have the same channel Λ, hence, making it possible to joint measurements to infinite-dimensional systems and solve many steering problems in one go. measurements with possibly continuous outcome sets has been done in (Kiukas et al., 2017). The point is to generalize the Choi-Jamiolkowski isomorphism to arbitrary E. Further topics on incompatibility shared states (i.e. not only ones with maximally mixed marginals). The generalisation gives a one-to-one con- As mentioned in the beginning of this section, mea- nection between states % with a fixed full-rank marginal surement incompatibility manifests itself in various ways %B (on Bob’s side) and channels Λ from Bob to Alice in quantum theory. In this section we review briefly through two well-known fine-tunings of commutativity and discuss their relation to quantum correlations.

% = (Λ ⊗ 11)(|ψ%B ihψ%B |), (88) First, in the case of joint measurability one asks for the existence of a common POVM and a set of post- Pd √ where |ψ%B i = i=1 si|iii is a puriﬁcation of %B = processings. One can relax this concept by dropping the Pd i=1 si|iihi|. To see a connection to incompatibility, one assumption on post-processings. Namely, we say that a writes the channel corresponding to a given state (in the set of POVMs {Aa|x}a,x is coexistent if there exists a Heisenberg picture) as POVM {Cλ}λ such that

1/2 † 1/2 T X %B Λ (A)%B = trA[(A ⊗ 11)%] , (89) AX|x = Cλ, (90)

λ∈τX|x T where is a transpose in the eigenbasis of %B. In- putting sets of POVMs {Aa|x}a,x on the left-hand side where τX|x is a subset of outcomes of {Cλ}λ for every of Eq. (89) results in transposed state assemblages on pair (X, x). Here we have used the notation X to em- the right-hand side. From this it is clear (at least in phasize that the definition is not only required to hold the finite-dimensional case in which %B can be inverted) for all POVM elements of {Aa|x}a,x, but also for sums † that the Heisenberg channel Λ sends Alice’s POVMs of outcomes, e.g. for X = {a1, a2} one has AX|x = to normalised state assemblages (i.e. POVMs) on Bob’s Aa1|x + Aa2|x. To give the concept a physical interpreta- side. Joint measurability of these POVMs is equivalent tion, in (Heinosaari et al., 2016) the authors noted that to the unsteerability of {%a|x}a,x. It turns out that this the definition is equivalent to the joint measurability of 26 the set of all binarizations (i.e. coarse-grainings to two- As some disturbing measurements can be jointly mea- valued ones) of the involved measurements. Whereas it surable, measurement disturbance is only necessary but is clear that joint measurability implies coexistence (by not sufficient for steering. One could, however, ask if the use of deterministic post-processings), the other di- there exist other types of quantum correlations or tasks rection does not hold in general (Pellonpää,2014; Reeb for which disturbance is necessary and sufficient. It turns et al., 2013). out that the question can be answered in positive and As coexistence is closely related to joint measurabil- one answer is given by violations of typical (i.e. choose ity, one can ask if anything can be learned from using between measuring or not measuring) models of macro- this concept in the realm of steering. As pointed out realism (Uola et al., 2019). More precisely, the authors in (Uola et al., 2014) one can reach steering with co- of (Uola et al., 2019) have shown that when all classical existent measurements on the uncharacterised side (pro- disturbance (i.e. clumsy measurement implementation) vided that the measurements are not jointly measurable). is isolated from a quantum system, the system can vio- What has not appeared in the literature so far, but we late macrorealism with some initial state if and only if wish to point out here, is that when using this concept the involved measurements do not fulfil the definition of on the characterized side, one can find examples of steer- non-disturbance. able assemblages that nevertheless form one ensemble. Motivated by the strong connections between quan- Consider the example of coexistent but not jointly mea- tum measurement theory and quantum correlations pre- surable POVMs given in (Reeb et al., 2013) by defining sented in this section (see also Section V.D and Sec- a vector |ϕi = √1 (|1i + |2i + |3i) and the POVMs 3 tion V.H), it will be an interesting question for future research to isolate the measurement resources behind 1 Ai|1 := (11 − |iihi|), i = 1, 2, 3 (91) other quantum tasks. Conversely, it will be of interest 2 to see if other concepts of incompatibility such as broad- 1 A+|2 := |ϕihϕ|,A−|2 := 11 − A+|2. (92) castability (Heinosaari, 2016), incompatibility on many 2 copies (Carmeli et al., 2016), and measurement simu- To see that these POVMs are coexistent, one can define lability (Oszmaniec et al., 2017) will find counterparts a POVM {Cλ}λ through the elements in the realm of quantum correlations. To conclude, we ! note that whereas further connections between measure- 1 1 1 1 1 |1ih1|, |2ih2|, |3ih3|, |ϕihϕ|, (11 − |ϕihϕ|) . ment theory and correlations remain unknown, jointly 2 2 2 2 2 measurable sets (Carmeli et al., 2019; Skrzypczyk et al., (93) 2019), or more generally all convex subsets of measurements (Uola et al., 2019b), can be characterized through For the proof that these POVMs are not jointly measur- state discrimination tasks. able we refer to (Reeb et al., 2013). Applying the mapping between measurement assemblages and state assemblages (with a full rank state %B, see also Section IV.C) to the above (or any similar) example, one ends up with V. FURTHER TOPICS AND APPLICATIONS OF a steerable state assemblage that nevertheless fits into a STEERING single ensemble. As another example, we consider the concept of mea- In this section we discuss further aspects and appli- surement disturbance. A POVM {Aa}a is called non- cations of steering. We start with multipartite steering, disturbing with respect to a POVM {Bb}b if there steering of Gaussian states and temporal steering. Then exists an instrument {Ia}a implementing {Aa}a (i.e. we discuss applications of steering such as quantum key tr[Ia(%)] = tr[Aa%] for all states %) such that distribution, randomness certification or channel discrim- X ination. Finally, we review the resource theory of steering tr[Ia(%)Bb] = tr[%Bb] (94) and the phenomenon of post-quantum steering. a holds for all states % and all outcomes b. Non-disturbance is located in between commutativity and joint measurability. Clearly commutativity implies A. Multipartite steering non-disturbance by the use of the Lüdersrule and non- disturbance implies joint measurability by defining for a The extension of steering to multipartite systems is an † non-disturbing scenario Ga,b = Ia(Bb), where the dagger emerging field of research and different approaches for refers to the Heisenberg picture. For a proof that the defining multipartite steering exist. Before explaining implications can not be reversed in general, and for more them, we would like to point out some peculiarities of detailed analysis on when the implications are reversible, the multipartite scenario, if one considers steering across we refer to (Heinosaari and Wolf, 2010). a bipartition. 27

1. Steering across a bipartition and the state is locally unsteerable even with the restric- tion to a factorizing p(a, b|x, y, λ). However, because of In order to discuss the different effects that play a role the superactivation phenomenon, this state is steerable for steering in the multipartite scenario, consider a tri- with global measurements. partite state %ABC and investigate steering across a given Another possibility is to consider the bipartition bipartition, say AB|C for definiteness. Then, there are A|BC, where Alice wants to steer Bob and Charlie. Here, BC different scenarios that have to be distinguished, where one has to consider the ensemble %a|x and ask whether in all of them Alice and Bob want to steer Charlie. R BC it can be written as %a|x = dλp(λ)p(a|x, λ)σλ . In • Global steering: In the simplest case, Alice and this case, two possible scenarios emerge, one where the Bob make global measurements on their two parti- state of Bob and Charlie is a single system, reducing cles and steer Charlie. This reduces to a bipartite to a bipartite steering problem, and another one where steering problem for % and all the usual meth- the local hidden state of Bob and Charlie factorizes, i.e. AB|C BC B C ods can be applied. σλ =σ ˜λ ⊗ σˆλ . From the above picture, one can see that the exten- • Reduced steering: Another simple case arises, if Al- sion of steering to multipartite systems leads to different ice (or Bob) try to steer Charlie, without needing scenarios, making its characterization even more difficult the help of the other. If the action of one of them than the ones for entanglement and nonlocality. is not required, this reduces to the bipartite steerability of the reduced state %A|C or %B|C . Again, all the bipartite steering theory can be applied. 2. Different approaches towards multipartite steering

• Local steering: The interesting case arises, if Alice The existing works on multipartite steering can be di- and Bob try to steer Charlie by local measurements vided into two different approaches. The first approach on their respective parties. In this case, one has to sees steering as an one-sided device-independent entan- consider the assemblage %ab|xy and ask whether its glement verification and translates this to the multipar- elements can be written as tite scenario. The second approach asks for a multipar- Z tite system whether steering is possible for a given bipar- % = dλp(λ)p(a, b|x, y, λ)σC . (95) ab|xy λ tition. To discuss the first approach, we need to recall the ba- Here, we can distinguish among several cases, de- sic definitions of the different entanglement classes for pending on the properties of p(a, b|A, B, λ). It may multipartite systems (Gühneand Tóth, 2009). For a be a general probability distribution, it may obey three-partite system % one calls the state fully sepa- the non-signaling constraint, or it may factorize, ABC rable, if it can be written as p(a, b|x, y, λ) = p(a|x, λ)p(b|y, λ). (96) fs X A B C % = pk% ⊗ % ⊗ % , (98) As p(a, b|x, y, λ) has the interpretation of a simu- ABC k k k k lation strategy [see Eq. (2)] the latter means that Alice and Bob play an independent strategy. where the pk form a probability distribution. If a state is not of this form, it is entangled, but not all particles A simple example of the difference between global are necessarily entangled. For instance, a state of the and local steering can be constructed from the phe- bs P A BC form %A|BC = k pk%k ⊗%k may contain entanglement nomenon of superactivation of steering (Quintino et al., between B and C, but it is separable for the bipartition 2016) (see also Section III.G). For certain states, one A|BC and therefore called biseparable. More generally, copy of the state is unsteerable, but many copies of mixtures of biseparable states for the different partitions the same state may become steerable. So one can con- are also biseparable sider a state %ABCC0 = %AC ⊗ %BC0 , where %BC0 is a copy of %AC , and %AC is unsteerable, but its steerability bs bs bs bs %ABC = p1%A|BC + p2%B|AC + p3%C|AB, (99) can be superactivated [where only two copies are already enough (Quintino et al., 2016)]. For this state, local mea- and states which are not biseparable are genuine mul- surements give an unsteerable state assemblage as tipartite entangled. These definitions can straightfor- CC0 wardly be extended to more than three particles. % = Tr [(A ⊗ B ⊗ 1 0 )% 0 ] ab|xy AB a|x b|y CC ABCC One-sided device-independent entanglement detection 1 1 = TrAB[(Aa|x ⊗ C )%AC ⊗ (Bb|y ⊗ C0 )%BC0 ] in the multipartite scenario has first been discussed by Z 0 (Cavalcanti et al., 2011). There criteria for full sep- = dλdµp(λ)p(µ)p(a|x, λ)p(b|y, µ)σC ⊗ σ˜C λ µ arability in the form of Mermin-type inequalities have Z CC0 been given, which hold for k trusted sites and N − k un- = dνp(ν)p(a, b|x, y, ν)ˆσν , (97) trusted sites. These criteria can be violated in quantum 28 mechanics, and the possible violation increases exponen- B. The steering ellipsoid tially with the number of parties. Inequalities for higher- dimensional systems have been derived in (He et al., Note that the definition of quantum steering Eq. (2) 2011). requires considering the ensembles of unnormalized con- In general, one if has a quantum network of N parties ditional states at Bob’s side. However, one can expect where some of the parties perform untrusted measure- that important insights can be gained by simply study- ments, the parties which trust their measurement appa- ing the normalized version of these conditional states. ratus can perform quantum state tomography, and recon- Note that in doing so, two things are lost: the steering struct the conditional state after the untrusted parties ensemble to which a conditional state belongs, and the announce their measurement choices and outcomes. For probability with which the conditional state is steered to. three parties there are two one-sided device-independent For two-qubit states, the normalized conditional states scenarios: when only one party’s device is untrusted, with Alice can steer Bob’s system to form an ellipsoid in- state assemblage side Bob’s Bloch sphere, referred to as the steering ellipsoid (Jevtic et al., 2014; Shi et al., 2011, 2012; Verstraete, 2002). Detailed analysis of their geometry has led to the %BC = Tr (A ⊗ 1 ⊗ 1 % ), (100) a|x A a|x B C ABC proposal to use them as a tool to represent two-qubit quantum states, in a way similar to the Bloch represen- and when two of them are untrusted tation of states of a single qubit. In particular, given the reduced states of both parties, a steering ellipsoid on C one side allows recovering of the density operator upto % = TrAB(Aa|x ⊗ Bb|y ⊗ 1C %ABC ). (101) ab|xy a certain local unitary or anti-unitary operation on the other side (Jevtic et al., 2014). Special attention later If %ABC is biseparable, this condition imposes constraints on was attracted to the volumes of the steering ellip- on the assemblages. Then, for a given state, to test soids (Cheng et al., 2016; Jevtic et al., 2014; McCloskey whether the assemblages of the form (100) or (101) obey et al., 2017; Milne et al., 2014; Zhang et al., 2019). In the conditions, one can use SDPs (Cavalcanti et al., particular, Milne et al. (2014, 2015) showed that the vol- 2015a). Also entropic conditions for this scenario have umes of the steering ellipsoids give upper bounds for the been studied (Costa et al., 2018a; Riccardi et al., 2018). entanglement of the state in terms of its concurrence. The second approach uses steering between the bipar- Even more interestingly, it is shown that the volumes titions to define genuine multipartite steering (He and of the steering ellipsoids obey certain monogamy rela- Reid, 2013). First, for two parties one can say that they tions (Cheng et al., 2016; Milne et al., 2014, 2015), which share steering if the first one can steer the other or vice will be discussed in the following. versa. Then, for three parties one can define genuine mul- Consider a system of three qubits ABC. Denote the tipartite steerability as the impossibility of describing a volume of the steering ellipsoids for steering from A to B state with a model where steering is shared between two and A to C by VB|A and VC|A, respectively. Milne et al. parties only. This means that the state cannot be de- (2014, 2015) showed that for all pure states of the system scribed by mixtures of bipartitions as in Eq. (99), where of three qubits ABC, one has for each partition (e.g., A|BC) the two-party state (e.g., BC) is allowed to be steerable. q q p VB|A + VC|A ≤ 4π/3. (102) One can then directly see that for checking this criterion, it is sufficient to consider the bipartitions AB|C, The authors also showed that the famous Coffman- AC|B, and BC|A, where the two-party sites are unchar- Kundu-Wootters monogamous inequality for entangle- acterized and the single-party site obeys quantum me- ment (Coffman et al., 2000) can be derived from this chanics. In addition, on the two-party site only local inequality. However, Cheng et al. (2016) showed that measurements are allowed, but the results only have to the monogamy relation Eq. (102) is violated when the obey the non-signaling condition. For proving genuine three qubits are in certain mixed states. Instead, the au- multipartite steering in this sense, several methods are thors showed that a weaker monogamy relation can be possible. If the state is pure, it suffices to check the steer- derived for all possible states over the three qubits, ability for the mentioned bipartitions, as a pure state 2/3 2/3 2/3 convex combinations into different bipartitions are not VB|A + VC|A ≤ (4π/3) . (103) possible (He and Reid, 2013). Otherwise, one may derive a linear (or convex) inequality that holds for unsteerable Recently, both the monogamy relation Eq. (103) and states of all relevant bipartitions. Due to linearity, it also the violation of Eq. (102) were illustrated experimen- holds for convex combinations and violation rules out the tally (Zhang et al., 2019). model mentioned above. This approach has been experi- It is in fact the analysis of the geometry of the steer- mentally used in (Armstrong et al., 2015; Li et al., 2015). ing ellipsoids for Bell-diagonal states that leads to the 29 exact characterization of quantum steering for this fam- With these definitions we are ready to state the char- ily of states (Jevtic et al., 2015; Nguyen and Vu, 2016a); acterisation of steerable states in Gaussian systems orig- see also Section II.C. Beyond the Bell-diagonal states, inally given in (Wiseman et al., 2007). little is known about the extent to which the steering A bipartite Gaussian state with covariance matrix VAB ellipsoids, or their volumes, can characterize the quan- and displacement rAB is unsteerable with Gaussian mea- tum steerability of the state. In particular, an interest- surements if and only if ing question for future research might be whether the monogamy relations Eq. (102) and Eq. (103) can induce VAB + i(0A ⊕ ΩB) ≥ 0. (108) certain monogamy relations between some measures of steering such as the critical radii as defined in Section Here 0A is a zero matrix on Alice’s side and ΩB is the N 0 1 II.C. matrix ⊕j=1 −1 0 on Bob’s side. In contrast to other steering scenarios, the Gaussian case appears special in that the steerability of a state C. Gaussian steering can be characterized through an easy to evaluate inequality. This is, however, not the only special feature for 1. A criterion for steering of Gaussian states Gaussian steering. Namely, within the Gaussian regime one can also prove monogamy relations for steering with Before discussing Gaussian steering some prelimi- more than two parties (Adesso and Simon, 2016; Ji et al., nary notions are needed. First, Gaussian systems re- 2015a; Lami et al., 2016; Reid, 2013). One should note fer to a special class of continuous variable scenarios. that the monogamy can break when one is allowed to Hence, one deals with infinite-dimensional Hilbert spaces N 2 perform non-Gaussian measurements (Ji et al., 2016). ⊗j=1L (R), where the index N refers to the number of modes. Every Gaussian state is described by a real symmetric matrix, the so-called covariance matrix V satisfy- 2. Refining Gaussian steering with EPR-type observables ing As a special case of interest in the Gaussian regime V + iΩ ≥ 0, (104) we discuss steering with canonical quadratures. It was N 0 1 shown in (Kiukas et al., 2017) that steerability of a given where Ω = ⊕j=1 −1 0 . More precisely, the covari- state in the Gaussian scenario can be readily detected by ance matrix of a quantum state % is given as (V )ij = T a pair of quadrature observables. Tr[%{Ri − ri,Rj − rj}], where R = (Q1,P1,...,Qn,Pn) with quadrature operators Qi and Pj satisfying [Qi,Pj] = To be more concrete, we sketch the construction of the iδij11 and [Qi,Qj] = [Pi,Pj] = 0, and rj = tr[%Rj]. More- quadratures from (Kiukas et al., 2017). First, a channel over, every real symmetric matrix satisfying Eq. (104) is called Gaussian if it maps Gaussian states to Gaussian defines a Gaussian state. The use of the word Gaussian states. Gaussian channels between systems of n and m in this context originates from the fact that the above degrees of freedom correspond to triples (M, N, c) with described states correspond to the ones whose charac- M being a real 2n × 2m matrix, N being a real 2m × 2m teristic function% ˆ(x) := tr[W (x)%] is Gaussian. Here matrix, and c being the displacement, that satisfy T W (x) = e−ix R with x = (q , p , ..., q , p )T and 1 1 n n N − iM T ΩM + iΩ ≥ 0. (109) − 1 xT V x−irT x %ˆ(x) = e 4 . (105) The transformation of Gaussian states on the level of covariance matrices is given as Second, a Gaussian measurement is a POVM Ma (with d values in a ∈ R ) whose outcome distribution for any T T Gaussian state is Gaussian. Such POVMs correspond to V 7→ M VM + N, r 7→ M r + c. (110) triples (K, L, m) satisfying Given that a bipartite Gaussian state has a corresponding L − iKT ΩK ≥ 0, (106) Choi-Jamiolkowski channel with parameters (M, N, c), one first notes that the state is unsteerable with Gaussian where K is an N × d matrix, L is an d × d matrix, and measurements if and only if the channel parameters de- m is a displacement vector. The correspondence between fine also a Gaussian measurement (Kiukas et al., 2017). the POVM Ma and the triple (K, L, m) is given through Hence, for a steerable state there exists two vectors x the operator-valued characteristic function as and y such that (yT − ixT )(N − iM T ΩM)(y + ix) < 0. As the triple (M, N, c) also fulfils Eq. (109), we have Z ipT a r := xT Ωy > 0 and Mˆ (p) := dae Ma

− 1 pT Lp−imT p 1 =W (Kp)e 4 . (107) (Mx˜)T ΩMy˜ > (˜xT Nx˜ +y ˜T Ny˜), (111) 2 30 wherex ˜ = r−1/2x andy ˜ = r−1/2y. From here one The concept of channel steering relates to the coher- T can construct two canonical quadratures as Qx˜ =x ˜ R ence of the channel extension. Namely, a channel exten- T C→A⊗B and Py˜ =y ˜ R. These are canonical as by definition sion Λ is coherent if it can not be written as x˜T Ω˜y = 1. To see that the state is indeed steerable with C→A⊗B X these measurements we refer to (Kiukas et al., 2017). To Λ (·) = Iλ(·) ⊗ σλ (114) summarise, we state the following refined characterisa- λ tion of Gaussian steering (Kiukas et al., 2017): for some instrument {Iλ}λ and states {σλ}λ. Any exten- For a bipartite Gaussian state %AB with a covari- sion that is of this form is called incoherent. One can ance matrix VAB and displacement rAB the following are show that incoherent extensions always lead to unsteer- equivalent: able instrument assemblages and any unsteerable instru- (i) %AB is steerable with Gaussian measurements ment assemblage can be prepared through some incoher- (ii) %AB is steerable with some pair of canonical quadra- ent extension (Piani, 2015). Note that in spatial steering tures any separable state leads to an unsteerable assemblage (iii) VAB + i(0A ⊕ ΩB) is not positive semi-definite. and any unsteerable assemblage can be prepared with a separable state (Moroder et al., 2016). In the original paper defining channel steering (Piani, D. Temporal and channel steering 2015), the concept is mainly probed through channel extensions as above. This leads to some connections with So far we have concentrated on steering in spatial sce- state-based correlations. For example, an extension can narios, i.e. scenarios where Alice and Bob are space- lead to a steerable instrument assemblage if and only if like separated. Some efforts for defining similar con- its Choi state allows Alice to steer Bob, and an extension cepts in temporal scenarios, i.e. scenarios where Alice is incoherent if and only if the Choi state is separable in and Bob form a prepare-and-measure type scenario, and the cut A|BC0, where C0 is the extra input system from on the level of quantum channels have also been pur- the isomorphism. sued in the literature (Chen et al., 2017, 2016b, 2014; One can investigate the channel protocol by replac- Piani, 2015). In a temporal scenario (consisting of two ing the extension with a (minimal) dilation. For com- measurement times), one can ask whether a state assem- pleteness, we note that a minimal dilation of a channel blage resulting from measurements at the first time step † Λ: L(H) 7→ L(H) can be written as Λ(·) = trA[V (·)V ], allows a local hidden state model on the second time Pn where V |ψi = k=1 |ϕki ⊗ (Kk|ψi) for all |ψi ∈ H, step. Of course, in temporal scenarios signalling is possi- n {Kk} form a linearly independent Kraus decomposi- ble and, hence, such models are sometimes trivially vio- k=1 tion of Λ, and {|ϕki}k is an orthonormal basis of the an- lated. Despite signalling, temporal steering has found cillary system. In this case, the correspondence between applications in non-Markovianity (Chen et al., 2016b) instruments and POVMs on the dilation is one-to-one and in QKD (Bartkiewicz et al., 2016), and some criteria and is given through (Chen et al., 2014) and quantifiers (Bartkiewicz et al., † 2016) have been developed. As the criteria and quanti- Ia|x(·) = trA[(Aa|x ⊗ 11)V (·)V ]. (115) fiers resemble strongly those of spatial steering presented in Sections II.A and II.B, we don’t wish to go through This generalises directly the connection between joint them in detail. measurements and spatial steering to the level of chan- In channel steering (Piani, 2015) one is interested nel steering (Uola et al., 2018). Namely, a measure- in instrument assemblages instead of state assemblages. ment assemblage {Aa|x}a,x on the minimal dilation is Namely, given a quantum channel ΛC→B from Charlie jointly measurable if and only if the corresponding in- to Bob and its extension ΛC→A⊗B, one asks if an assem- strument assemblage is unsteerable. By noticing, further- blage defined as more, that channel steering with trivial inputs (i.e. one- dimensional input system) corresponds to spatial steer- C→A⊗B Ia|x(·) := trA[(Aa|x ⊗ 11)Λ (·)], (112) ing, and that in this case a dilation corresponds to a purification of the total state of the assemblage, one re- where {Aa|x}a,x is a set of POVMs, can be written as covers the connection between joint measurements and spatial steering. X The dilation technique can also be used to prove that Ia|x(·) = p(a|x, λ)Iλ(·) (113) λ any non-signalling state assemblage originates from a set of non-signalling instruments (Uola et al., 2018), hence, for some instrument {Iλ}λ (i.e. a collection of CP maps showing that channel steering captures non-trivial (i.e. summing up to a quantum channel) and classical post- non-signalling) instances of temporal steering (with two processings {p(a|x, λ)}a,x,λ. Whenever this is the case, time steps), and that in this case a connection between the instrument assemblage {Ia|x}a,x is called unsteerable. temporal steering and joint measurements follows di- 31 rectly from the one between channel steering and incom- 1.0 r patibility. Moreover, using the channel framework one can translate concepts from spatial to temporal scenar- 0.8 ios. One example of this is given in (Uola et al., 2018) showing that temporal steering and violations of macro- 0.6 realism respect a similar strict hierarchy as spatial steering and non-locality. Note that in (Ku et al., 2018b) the 0.4 hierarchy was proven independently. 0.2 Bound on the secret key rate 0.0 E. Quantum key distribution 0.65 0.70 0.75 0.80 0.85 0.90 0.95 1.00

Alice's detection efficiency ΗA In quantum key distribution (QKD) two main types of protocols can be distinguished (Scarani et al., 2009). FIG. 10 Key rates for the QKD based on steering. For dif- In prepare & measure (PM) schemes, such as the BB84 ferent visibilities of the initial state (V ∈ {1, 0.99, 0.98, 0.95}) lower bounds on the key rate are shown. For perfect vis- protocol, Alice prepares some quantum states and sends ibility (solid blue line) a key can be extracted for detector them to Bob, who performs measurements on them. Us- efficiencies of η ≥ 0.659 for Alice. The dashed line shows a ing classical communication, Alice and Bob can then try bound (obtained with the same methods) for the fully device- to generate a secret key from the measurement data. In independent scenario for the case of perfect visibility. The entanglement-based (EB) schemes, such as the E91 pro- figure is taken from (Branciard et al., 2012). tocol, an entangled quantum state is distributed to Alice and Bob, and both make measurements on their part of the state. The source of the state might be under control device-independent QKD using this protocol against at- of an eavesdropper Eve. Again, the measurement data tacks where Eve has no quantum memory has been stud- are then used to generate a secret key. ied (see also Fig. 10). It has been shown that only the A central result concerns the role of entanglement for detector efficiency of the untrusted party matters and security. In (Curty et al., 2004) it was proved that en- that already for detector efficiencies η ≥ 0.659 a secret tanglement is a necessary precondition for security. For key can be distilled, and a non-zero key rate proves that EB schemes, this means that if the measurement data the underlying states are steerable. The obtainable key can be explained by a separable state, then no secret key rates are higher and the required detector efficiencies are can be distilled. For PM schemes, one can consider an lower than in the fully device-independent case. equivalent EB scheme, then the same statement holds. In addition, results on steering and PM schemes for It should be noted, however, that the provable presence QKD have been obtained in (Branciard et al., 2012; Ma of entanglement was not shown to be sufficient for secret and Lütkenhaus, 2012). An analysis for finite key length key generation. The question of whether entanglement can be found in (Wang et al., 2013; Zhou et al., 2017), can be verified depends on the measurement data taken and upper bounds on the key rate in one-sided device- and the assumptions made on the measurements. In a independent QKD have been obtained in (Kaur et al., device-independent scheme, where no assumptions about 2018). Finally, it should be noted that similar ideas have the measurements are made, only Bell inequalities can be also been studied and implemented for QKD with contin- used to test the presence of entanglement. Still, device- uous variables (Gehring et al., 2015; Walk et al., 2016). independent QKD can be proved to be secure against certain attacks (Ac´ın et al., 2007). One can also consider an asymmetric situation, where F. Randomness certification one party trusts its devices and the other one does not. This can be realistic, e.g. if Alice corresponds to a client The task of randomness certification can be defined of a bank having only a cheap device, while Bob repre- as follows (Ac´ın et al., 2012; Law et al., 2014). On a sents the bank itself. Clearly, in such a situation QKD quantum system % a measurement labeled by z is made can only work if the underlying state is steerable. In and the result c is obtained. Depending on the situa- (Branciard et al., 2012) this problem has been consid- tion, the measurement may be a joint measurement on ered for the BBM92 protocol (Bennett et al., 1992). In two parties of an entangled state; then the labels for the this protocol, Alice and Bob share a two-qubit Bell state measurement and result can be written as z = (x, y) and and measure either A1 = B1 = σz or A2 = B2 = σx. The c = (a, b), as in the Bell scenario in Eq. (3). The task is to correlations for the measurement A1 ⊗ B1 are used for quantify the extent to which an external adversary Eve the key generation, while the correlations in the A2 ⊗ B2 can predict the outcome c of the probability distribution measurement are used to estimate Eve’s information. p(c|z). Clearly, this depends on the assumptions made In (Branciard et al., 2012) the security of one-sided about Eve: For instance, one can distinguish the case 32 where the state % is fixed and only known to Eve from But here mainly the randomness for a single measure- the case where Eve indeed provides the state. In the for- ment setting of Alice was considered. It has been shown mer case, one can furthermore distinguish the knowledge that this can directly be computed with an SDP, without Eve has. She may have only classical information about the need of a convergent hierarchy of SDPs. This is then the state or she may hold a purification of it, see (Law shown to hold also for the scenario considered in (Law et al., 2014) for a detailed discussion. et al., 2014). In (Skrzypczyk and Cavalcanti, 2018) the In the simplest case, the state % = |ψihψ| is pure and problem has been considered for two d-dimensional sys- the measurement z is characterized. Then, the best strat- tems. A steering inequality has been derived, such that egy for Eve is to guess the c with the maximal probability, the maximal violation guarantees log(d) random bits for and the probability of guessing correctly is given by Alice’s outcomes in the one-sided device-independent scenario. Furthermore, any pure entangled state with full G(z, ψ) = max p(c|z, ψ). (116) Schmidt rank can be used to generate this amount of c randomness. If the state % is mixed then Eve may hold a purifica- Finally (Curchod et al., 2017) showed that if one con- tion of it and she may know the exact decomposition siders the Bell scenario, then sequential measurements on P one party can lead to an unbounded generation of ran- % = k pk|φkihφk| into pure states. Consequently, the maximal guessing probability is domness. The extension of this to the steering scenario is discussed in (Coyle et al., 2018). X G(z, %) = max pkG(z, φk), (117) pk,φk k G. Subchannel discrimination where the maximization runs over all decompositions of %. If the measurements z are not characterized, one has In (Piani and Watrous, 2015) an operational character- to optimize over all possible quantum realizations of the isation of steerable quantum states is provided. The idea classical probability distribution p(c|z). So, the maximal is similar to the main result of (Piani and Watrous, 2009) guessing probability is stating that every entangled state provides an advantage over separable ones in some channel discrimination task G(z, p(c|z)) = max G(z, %), (118) to the realm of steering. In the case of steering, the re- %,Mc|z lated task turns out to be that of subchannel discrimina- where the maximization runs over all quantum realization, i.e. discriminating different branches of a quantum tions, described by a state % and measurement operators evolution. Namely, take an instrument I = {Ia}a (i.e. a collection of completely positive maps summing up to a Mc|z with p(c|z) = Tr(Mc|z%). Optimizations over this set can be carried out by hierarchies of SDPs (Navascués quantum channel), a POVM B = {Bb}b, an input state et al., 2007, 2008). In all cases, the number of random % and define the probability of correctly identifying the bits that one can extract from p(c|z) is given by the min- subchannel (i.e. instrument element) as entropy, Hmin(G) = − log (G). X 2 p (I,B,%) := tr[I (%)B ]. (119) Initially, the task of randomness certification was cor a a a mainly studied in the Bell scenario, where z = (x, y) and c = (a, b) describe measurements on an entangled To find the best strategy for the task, one maximises over state (Ac´ın et al., 2012). Here, the devices are not char- input states and POVMs on the output. acterized and as soon as a Bell inequality is violated, As mentioned above, entanglement provides an advan- one can prove that the results of a fixed setting can- tage in channel discrimination tasks, i.e. tasks of discrim- not be predicted, so the randomness is certified. In (Law inating between subchannels of the form Ia = p(a)Λa, et al., 2014) randomness certification has been studied for where {Λa}a are quantum channels. To prove a similar the steering scenario: Again, one makes local measure- result for general subchannels, the authors of (Piani and ments on an entangled state, but this time the devices Watrous, 2015) limit the set of allowed measurements on Bob’s side are characterized. This leads to additional between the system (i.e. outputs of the instruments) constraints in the SDP hierarchy (Navascués et al., 2007) and the ancilla to local measurements supported by for- and consequently more randomness can be extracted. In- ward communication from output to ancilla (i.e. one-way terestingly, also for states that are not steerable the ran- LOCC measurements). Such measurements have POVM out→anc P domness can be certified; so the violation of a steering elements of the form Ca = x Aa|x ⊗ Bx, where inequality is not necessary for randomness certification {Bx}x is a POVM on the output system and {Aa|x}a in the one-sided device-independent scenario. is a POVM on the ancilla for every x. The probabil- Also in (Passaro et al., 2015) the task of random- ity of correctly identifying the branch of the evolution ness certification in the distributed one-sided device- with such measurements and a shared state %AB is given P † independent scenario between two parties was studied. as pcor(I, 1-LOCC,%AB) = a,x trout[Ia(Bx)%a|x]. Note 33 that any unsteerable state can perform at most as good mentioning that the hierarchy presented here corresponds as some single system state. One sees this by using an to one choice of quantifiers. Analogous results are possi- LHS model for the assemblage in the above equation and ble for various fine-tuned quantifiers. by choosing the best performing hidden state as the sin- To write down the result, recall the definitions of gle system state. incompatibility, steering and non-locality robustness. To prove the main result of the paper, the authors Steering robustness is defined in Eq. (30) and analogously define a quantity called steering robustness of a bipartite to that one defines incompatibility robustness IR(Aa|x) state %AB by maximising the steering robustness of all of a set {Aa|x}a,x of POVMs as possible assemblages the state allows. More formally min t A→B Rsteer (%AB) = sup{R(A)|{Aa|x}a,x}, (120) Aa|x + tNa|x X s.t. = D(a|x, λ)G for all a, x 1 + t λ where R(A) is the steering robustness of the assem- λ blage %a|x = trA[(Aa|x ⊗ 11)%AB]. Clearly the quantity t ≥ 0 A→B Rsteer (%AB) is zero if and only if the state %AB is un- X steerable. The main result now reads: Na|x ≥ 0 for all a, x, Na|x = 11 for all x For any steerable state there exists a subchannel dis- a X crimination task (with forward communication from the Gλ ≥ 0 for all λ, Gλ = 11. (122) output to the ancilla) in which the state performs better λ than any unsteerable one, i.e. Note that here D(·|x, λ) ∈ {0, 1} is a deterministic assignment for every x and λ. The interpretation of this pcor(I, 1-LOCC,%AB) A→B sup NE = 1 + Rsteer (%AB). (121) robustness is that one mixes the POVMs {Ma|x}a,x with pcor (I) {Na|x}a,x until they become jointly measurable. Here the supremum is taken over all instruments and one- Now, if Alice’s measurements {Aa|x}a,x in a steering way LOCC measurements from the output to the ancilla. scenario with a state %AB have incompatibility robust- Note that this result gives the set of steerable states an ness t, then replacing Alice’s measurements with the operational characterisation. Experimental demonstra- Aa|x+tNa|x jointly measurable POVMs 1+t shows that t is tion of this result has been presented in (Sun et al., 2018). an upper bound for the steering robustness of σa|x := We note that the result on steering robustness can be trA[(Aa|x⊗11)%AB]. In other words, steering robustness of generalised. One can define a robustness measure for any a given assemblage lower bounds the incompatibility ro- convex and closed subset of assemblages, and reach a sim- bustness of the measurements on the steering party. This ilar conclusion as above using conic programming (Uola bound, moreover, is one-sided device-independent. No- et al., 2019b). Moreover, one can show that a related tice that with a fine-tuned steering quantifier called con- measure called convex weight or free fraction has a simi- sistent steering robustness the aforementioned inequal- lar interpretation: whereas robustness measures discrim- ity is tight for full Schmidt rank states (Cavalcanti and ination power, the free fraction is a measure of exclusivity Skrzypczyk, 2016), see also (Kiukas et al., 2017). (Uola et al., 2019a). For the device-independent quantification of steering and incompatibility one can use the nonlocality robustness NR[p(a, b|x, y)] of a probability table H. One-sided device-independent and device-independent {p(a, b|x, y)} given as (Cavalcanti and Skrzypczyk, quantification of measurement incompatibility a,b,x,y 2016) As steering is closely related to joint measurability and min r non-locality, some works (Cavalcanti and Skrzypczyk, p(a, b|x, y) + rq(a, b|x, y) 2016; Chen et al., 2016a) have pursued ways of deriving s.t. 1 + r one-sided device-independent and device-independent X bounds on measurement incompatibility. The idea is = p(λ, µ)D(a|x, λ)D(b|y, µ) for all a, b, x, y to show that quantifiers of incompatibility (e.g. in- λ,µ compatibility robustness) are lower bounded by quan- r ≥ 0, p(λ, µ) ≥ 0 tifiers of steering (e.g. steering robustness), which in q(a, b|x, y) ∈ Q, (123) turn are lower bounded by non-locality quantifiers (e.g. non-locality robustness). Hence, on top of the one- where Q is the set of all possible quantum correlations sided device-independent and fully device-independent defined as lower bounds on incompatibility, one gets also a device- Q = {tr[(A ⊗ B )% ]| independent lower bound on a quantifier of steering. a|x b|y AB We follow (Cavalcanti and Skrzypczyk, 2016) to make {Aa|x}a,x, {Bb|y}b,y POVMs ,%AB a state}. the aforementioned hierarchy more concrete. It is worth (124) 34

Similarly to the one-sided device-independent quantifica- and Reid, 2013; Xiang et al., 2017); see also Section V.A. tion of incompatibility above, one sees that for a given This similarity has been made precise in analyzing the se- state assemblage {σa|x}a,x the non-locality robustness curity of secret sharing (He and Reid, 2013; Kogias et al., of any probability table originating from this assem- 2017; Xiang et al., 2017). Specifically, Xiang et al. (2017) blage, i.e. p(a, b|x, y) = tr[σa|xBb|y] with {Bb|y}b,y be- computed the secrete key-rate bound which guarantees ing POVMs on Bob’s side, gives a lower bound for the unconditional security of the protocol against eavesdrop- steering robustness of the assemblage. In other words pers and dishonest players (Kogias et al., 2017) for three- mode Gaussian states and found it to be essentially the IR(Aa|x) ≥ SR(σa|x) ≥ NR[p(a, b|x, y)]. (125) quantification of the difference between collective steering and local steering from Bob and Charlie to Alice (Kogias Hence, using the machinery of (Cavalcanti and et al., 2015a). To our knowledge, whether this quanti- Skrzypczyk, 2016; Chen et al., 2016a) one finds one-sided tative relation between quantum steering and quantum device-independent and fully device-independent lower secret sharing extends beyond Gaussian states is at the bounds for quantifiers of measurement incompatibility moment unknown. and device-independent lower bounds for quantifiers of steering. J. Quantum teleportation

I. Secret sharing Steering shares a close conceptual similarity with state teleportation (Bennett et al., 1993). We follow (Caval- Secret sharing is a cryptography protocol that allows canti et al., 2017) and consider the following abstract a dealer (Alice) to send a message to players (Bob and teleportation protocol. Alice and Bob share a bipartite Charlie) in a way such that the message can only be quantum state %AB. Charlie, perceived as the verifier, decoded if when the players work together—neither of draws a pure state ωx from a certain set of |x| states in- them can decode it by himself. If Alice shared a secret dexed by x, gives it to Alice and asks her to teleport it key (see Section V.E) with Bob and another with Char- to Bob. Without knowing the state ωx, Alice makes a CA lie, she can simply encode the message twice with the two measurement with POVM elements {Ea }a jointly on keys to ensure that only Bob and Charlie together can the received state and her particle that is entangled with decode the message. Thus, normal quantum key distri- Bob’s system. Depending on Alice’s outcome a, Bob’s bution protocols already provide one with protocols for system is then ‘steered’ to a conditional state, quantum secret sharing. But one can do it more straight- Tr (ECA ⊗ 11B)(ω ⊗ %AB) forwardly with multipartite entanglement (Hillery et al., B CA a x %a (ωx) = , (127) 1999); see also (Zukowski˙ et al., 1998) for a related pro- p(a|ωx) tocol. Take the case where Alice prepares a large number where the normalization p(a|ωx) = of the Greenberger-Horne-Zeilinger (GHZ) states (Hillery CA B AB Tr (Ea ⊗ 11 )(ωx ⊗ % ) is the probability for et al., 1999), Alice to get outcome a in her protocol given she re- 1 ceived state ωx from Charlie. Alice communicates her |GHZi = √ (|000i + |111i). (126) measurement outcome a to Bob, who then makes some 2 appropriate local unitary operation; by the end of the procedure, the state of his system is U %B U †. The Alice keeps one particle, and sends the other two to Bob a a|ωx a and Charlie. Each then measures their particles in ran- design of Alice’s measurement and Bob’s local unitary dom directions, x or y. After communicating via a clas- operations is such that the final state at Bob’s side sical public channel, they can identify the triplets where resembles Charlie’s original state ωx as much as possible. they have measured in directions xxx, xyy, yxy, yyx; The quality of the teleportation protocol can be assessed the other triplets are discarded. As one can check, the by the so-called average fidelity, GHZ state is an eigenstate of the retained measurement 1 X F¯ = p(a|ω )Tr[ω U %B(ω )U †]. (128) operators. In these triplets, Bob and Charlie can use tel |x| x x a a x a their outcomes to predict Alice’s outcomes. However, a,x the outcomes at Bob’s or Charlie’s sides separately are If the teleportation is perfect, the average fidelity is 1 for not enough to infer her outcomes. Thus, Alice can use pure states ωx. the series of outcomes of her measurements to encode the One can easily observe the similarity of the teleporta- message in her secret sharing protocol. tion protocol with that of quantum steering: instead of The fact that Bob and Charlie have to collaborate to receiving a classical input x, Alice received a quantum infer the measurement outcomes at Alice’s side resem- state from Charlie ωx as an input; see Fig. 11. The idea bles the distinction between the concepts of local and of receiving quantum inputs from a verifier (Charlie) in- global steering in the multipartite steering scenarios (He stead of a classical input has been previously considered 35

be entangled with Charlie. Instead of monogamy of entanglement, Reid (2013) then used the monogamy of a certain steering inequality to demonstrate the security of quantum teleportation.

K. Resource theory of steering

A resource theory is typically seen as consisting of two basic components: free states and free operations. Free Alice Bob states constitute a set which remains unchanged under FIG. 11 Quantum steering and quantum teleportation. In the actions of free operations. In this sense, one could quantum steering (top) Alice receives a classical input x, per- define a resource theory merely from the free operations. forms a measurement and communicates the output a to Bob; Consequently, any state that is not free has some resource Bob’s system is steered to %a|x. In teleportation (bottom) Al- in it, as it can not be created from the set of free states ice receives a quantum state ωx as input, performs a measure- with free operations. As an example, in the case of en- ment and communicates the output to Bob; Bob’s system is tanglement, free states are given by the set of separable steered to %a|x. Bob further makes a local unitary evolution on his system depending on the outcome he received to obtain states and free operations are local operations assisted by the final state. classical communication (LOCC). Another important as- pect of a resource theory are resource measures or monotones. A proper measure should not increase under free for entanglement (Branciard et al., 2013; Buscemi, 2012), operations, i.e. free operations can not create the re- and later for quantum steering both theoretically and source, should be faithful, i.e. equal to zero only for free experimentally (Cavalcanti et al., 2013b; Kocsis et al., states, and should be convex, i.e. randomisation should 2015). The benefit of allowing for the quantum inputs is not create resource either. that the verifier can now verify that Alice and Bob share In the case of steering a resource theory has been pro- a quantum correlation (entanglement, quantum steering) posed (Gallego and Aolita, 2015). The free states in this without trusting their measurement devices or their ac- theory are the unsteerable assemblages and free oper- tual measurements (Hall, 2018); see also Section I and ations can be any operations on the assemblages that further discussion in (Hall and Rivas, 2019). Utilizing do not map unsteerable assemblages into steerable ones. the similarity, Cavalcanti et al. (2017) showed that all In (Gallego and Aolita, 2015) one-way stochastic LOCC entangled states can demonstrate nonclassical teleporta- operations are shown to be free operations. In order tion in certain sense. One should, however, note that the to introduce one-way (stochastic) LOCC operations, we introduced notion of nonclassical teleportation does not adapt the notation of (Gallego and Aolita, 2015) for imply high teleportation average fidelity, which has been state assemblages. Namely, any state ensemble {%a}a a standard figure of merit. can be embedded into larger space via the correspon- P There is another line of works which attempt to relate dence {%â}a := a |aiha| ⊗ %a. To make a similar correspondence for state assemblages, one can define a map quantum steering with the security of quantum telepor- P tation. In teleporting a state to Bob, Alice does not want %Â|X (x) := a |aiha| ⊗ %a|x, where X and A label the an eavesdropper to also obtain some version of the state. sets of Alice’s inputs and outputs respectively. Now, a Certain security is guaranteed when the average fidelity one-way (stochastic) LOCC operation M is defined on of teleportation in Eq. (128) is high enough (Pirandola the assemblage% Â|X as et al., 2015). It is then shown that for certain family of bi- X † partite states, to obtain the required fidelity, the state is M(ˆ%A|X ) := (11 ⊗ Kω)Wω(ˆ%A|X )(11 ⊗ Kω), (129) not only entangled but necessarily two-way steerable (He ω et al., 2015). where {Kω}ω are Kraus operators and {Wω}ω are wiring Another way to investigate the security of teleporta- maps defined pointwise as tion is to study its sister protocol known as entanglement X X swapping. In this protocol, the state given to Alice by Wω(ˆ%A|X )(xf ) := p(x|xf , ω) p(af |a, x, xf , ω)

Charlie is priorly entangled with another particle which x af ,a Charlie keeps. By the end of the teleportation protocol (|af iha| ⊗ 11)ˆ%A|X (x)|aihaf | ⊗ 11. performed by Alice and Bob, the entanglement between (130) Charlie and Alice is transferred to that between Char- lie and Bob. In this case, the teleportation can be se- Here af and xf refer to the inputs and outputs of the cured by the monogamy of entanglement: if Charlie is final assemblage. Physically such transformations corre- sufficiently entangled with Bob, an eavesdropper cannot spond to performing an operation on the characterized 36 party, communicating the information about which op- This is not the case for the tripartite scenario (Sainz eration (i.e. ω) was performed, and the uncharacterized et al., 2015). Here, one considers the scenario where Alice party applying the corresponding classical pre- and post- and Bob make local measurements in order to steer Char- processings [i.e. p(·|xf , ω) and p(·|a, x, xf , ω)] on their lie’s state. So, Charlie has an assemblage {%ab|xy}, where side. Note that the use of one-way (stochastic) LOCC x and a (y and b) denote the measurement setting and operations as free operations has also an interesting phys- outcome of Alice (Bob). Besides being positive and the P ical motivation: they can be seen as safe operations in normalization constraint Tr( ab %ab|xy) = Tr(%C ) = 1 one-sided device-independent quantum key distribution this assemblage should fulfill that neither Alice nor Bob (Gallego and Aolita, 2015). can signal to the other parties, that is A typical resource theory aims at quantifying the re- X X 0 source in hand. For this purpose, one wishes to find a %ab|xy = %ab|x0y for all x, x , mapping (or monotone) f from the set of states of the a a resource theory to the set of non-negative real numbers X X 0 %ab|xy = %ab|xy0 for all y, y . (131) that fulfils certain requirements. In the case of steering b b the following requirements are considered (Gallego and Aolita, 2015) One can directly check that these conditions imply that also Alice and Bob jointly cannot signal to Charlie by • f(ˆ%A|X ) = 0 if and only if% Â|X is unsteerable the choice of their measurements. Contrary to the bipartite case, an assemblage obeying • f is non-increasing on average under deterministic these constraints does not need to have a quantum re- one-way LOCC alization. A simple counterexample can be derived from If in addition the mapping f is convex, it is called a con- the PR box mentioned above: If the conditional states vex steering monotone. In (Gallego and Aolita, 2015) are of the form %ab|xy = p(ab|xy)|0ih0|C , where p(ab|xy) the typical steering quantifiers, i.e. steerable weight and is the probability table of the PR box, the assemblage robustness of steering (see Section II.B), are shown to is clearly non-signaling, but cannot be realized within be convex steering monotones. Furthermore, the authors quantum mechanics. In (Sainz et al., 2015) more inter- introduce a novel convex steering monotone called rela- esting examples of this behavior were provided. Using tive entropy of steering, see also (Kaur et al., 2017; Kaur iterations of SDPs the authors found an example of a and Wilde, 2017) for further monotones and alternative qutrit assemblage %ab|xy with the properties that for any definitions of relative entropy of steering. possible measurement Ec|z of Charlie, the resulting probability distribution can be explained by a fully local hidden variable model, which is even a stronger requirement L. Post-quantum steering than being non-signaling. Still, the assemblage has no quantum realization, so there is no state %ABC such that Post-quantum steering is the phenomenon that cer- %ab|xy = TrAB(Ea|x ⊗ Eb|y%ABC ). tain assemblages {%a|x} may not be realizable by quan- In further works the theory of post-quantum steering tum mechanics, although no signaling between the par- has been extended. (Sainz et al., 2018) provided general ties is possible. For the case of Bell inequalities, it is methods to construct examples of post-quantum steering known that there are probability distributions which are and defined a quantifier of this phenomenon, (Hoban and non-signaling, but cannot come from a quantum state. Sainz, 2018) established a connection to the theory of The most prominent example is the Popescu-Rohrlich quantum channels, and (Chen et al., 2018b) used moment (PR) box (Popescu and Rohrlich, 1994), which is a non- matrices to characterize the phenomenon. signaling distribution for two parties with two measurements having two outcomes, which leads to a violation of the CHSH inequality with a value hSCHSHi =√ 4 while M. Historical aspects of steering in quantum mechanics only values hSCHSHi ≤ 2 2 can occur. The analogous question for steering highlights the 1. Discussions between Schrödingerand Einstein difference between steering in the bipartite and the multipartite case. As mentioned already in the introduction, the first For the bipartite case, one may consider an assemblage observation of the steering phenomenon dates back to P {%a|x}, obeying the no-signaling constraint a %a|x = Schrödinger’sdiscussions of the EPR argument with Ein- P 0 a %a|x0 = %B for all x, x . As already mentioned in stein. Schrödingercorresponded with several physicists Sections II.B and V.M, any such assemblage can be re- on this problem, the letters have been edited by (von alized by quantum mechanics. This means that there is Meyenn, 2011). a state %AB and measurements Ea|x such that %a|x = To understand the origin of Schrödinger’sidea, it is TrA(Ea|x%AB). important to note that Einstein did not like the way the 37

EPR paper was written and how the argument was for- It must be added, of course, that the view of the steer- mulated, for detailed discussions see (Kiefer, 2015). In- ing phenomenenon as “nonlocal” is based on a certain stead, Einstein preferred a somehow simpler version. He interpretation of the wave function, which is not shared published this version much later (Einstein, 1948), but he by everyone, see (Griffiths, 2019) for a discussion. In also explained the basic idea in a letter to Schrödinger any case, the question remains for which states this phe- on June 19th, 1935. nomenon can be observed and which states on Bob’s sys- The argument (in the formulation of (Einstein, 1948)) tem can be reached by performing measurements on Al- goes as follows. First, one considers position X and mo- ice’s side. This was also part of the discussion between mentum P as non-commuting observables. Then there Schrödingerand other physicists (such as von Laue) and are, according to Einstein, two possibilities: Schrödingerpresented his results in two subsequent pa- (i) One can assume that position and momentum have pers. definite values before a measurement of them is carried out. Then one has to admit that the wave function |ψi is not a complete description. 2. The two papers by Schrödinger (ii) One can assume that the values of position or momentum are created during a measurement. This is com- The first paper entitled “Discussion of probability re- patible with the assumption that the wave function |ψi is lations between separated systems” (Schrödinger, 1935) a complete description. If |ψi is a complete description, was submitted in August 1935. Schrödingerstates that it follows, according to Einstein, that two different wave he finds it “rather discomforting” that quantum mechan- functions describe two different physical situations. ics allows a system to be steered by performing measure- These two ways of thinking cannot be distinguished ments in a different location. He then presents several without additional assumptions. Here, Einstein intro- results on this phenomenon. duces a locality principle, stating that if one considers a First, he shows that every bipartite pure state can be bipartite system, the real physical situation at one side written as is independent of what happens on the other side. X |ψi = s |a i|b i (132) In order to conclude the incompleteness of the quan- k k k k tum mechanical description, one can consider a pure entangled wave function, as in the usual EPR argument. where the vectors |aki and |bki form orthogonal sets. Then, the conditional wave function |φiB on Bob’s side This is nowadays called the Schmidt decomposition. He depends on the choice of the measurement on Alice’s side. proves that this is unique if the coefficients sk are differ- According to the locality principle, however, the physical ent. He also recognizes that this implies that for generic reality on Bob’s side cannot change. So, one arrives at states there is one measurement for Alice (defined by the a contradiction to (ii) and the incompleteness follows. It eigenvectors |aki) which is perfectly correlated with a is interesting to note that for this argument the (perfect) measurement on Bob’s side (defined by the |bki). correlations between measurements on both sides are not Then, he discusses in some more detail the EPR state relevant. As Einstein formulated it: “I couldn’t care less from the 1935 argument (Einstein et al., 1935), which ¯ whether or not |φiB and |φiB are eigenstates of some is not a generic state, as all the Schmidt coefficients co- observables.” incide. He proves that for any observable F (X2,P2) on In the direct reply to this letter (on July 13th, 1935) Bob’s side the value can be predicted by making a suit- Schrödingerspelled out that the dependence of the con- able measurement Fˆ(X1,P1) on Alice’s side. This fact ditional state |φiB includes some “steering” from a dis- appeared already in a letter from Schrödingerto von tance. Although this phenomenon does not allow signal- Laue, and it demonstrates the surprising effect that one ing between the parties, he considers it to be magic. Re- system seems to know the answers to all possible ques- calling discussions with Einstein and colleagues in Berlin tions on the other system. during the 1920s, he writes: The second paper entitled “Probability relations be- “All the others told me that there is no incredible tween separated systems” was submitted in April 1936 magic in the sense that the system in America gives (Schrödinger,1936). Schrödingerfirst states that the X = 6 if I perform in the European system nothing or a essence of the previous work was the observation that certain action (you see, we put emphasis on spatial sepa- in quantum mechanics one can not only determine the ration), while it gives X = 5 if I perform another action; wave function at one party by making measurement on but I only repeated myself: It does not have to be so the other, but one can also control the state at one side bad in order to be silly. I can, by maltreating the Euro- by choosing the measurements on the other side. So the pean system, steer the American system deliberately into question arises, to which extent the wave function can be a state where either X is sharp, or into a state which is controlled. certainly not of this class, for example where P is sharp. In order to answer this, he first proves a statement This is also magic!” on density matrices. A given density matrix may have 38 different decompositions into pure states proof and an extension to infinite-dimensional systems X X (Hadjisavvas, 1981). % = pk|ψkihψk| = qi|φiihφi|, (133) The second mathematical statement (below Eq. (135)) k i was derived by Gisin in the context of modifications of the Schrödingerdynamics (Gisin, 1989). Here, the question and the question arises, which conditions the |φii and arises whether the modified dynamics for pure states ex- |ψki have to fulfill. Schrödingerproves, that two ensembles give the same density matrix, if and only if there is tends uniquely to mixed states. If it were different for two a unitary matrix U such that decompositions such as the ones in Eq. (133), then ensembles will become distinguishable at some point. Given √ X √ pk|ψki = Uki qi|φii. (134) the fact that both ensembles can be prepared by mea- i surements on a distant system, this would lead to a violation of the non-signaling condition, enforced by special This implies that any |φii in the range of the space relativity. Finally, both mathematical statements from spanned by the |ψki can be an element of a suitable en- Schrödingerhave also been rederived independently by semble. (Hughston et al., 1993). Then, Schrödingerapplies this to the bipartite state in Besides these mathematical results, the notion of steer- Eq. (132). Here, the reduced state ing as a kind of quantum correlation was not discussed X X for a long time. The situation changed in the eight- % = s2 |b ihb |= q |β ihβ | (135) B k k k i i i ies of the last century. Then, Bell inequalities started k i to attract more attention (Clauser and Shimony, 1978) has different decompositions. As already mentioned, the and the mathematical notions of entangled and separable 2 ensemble {sk, |bki} can be reached by making the mea- states were studied (Primas, 1983; Werner, 1984; Werner, surement defined by the orthogonal states |aki on Alice’s 1989). side, and the question arises, whether any other ensemble The paper (Vujiˇcićand Herbut, 1988) was the first {qi, |βii} can be reached. Schrödingerproves that this is to give a clear summary of Schrödinger’s ideas and an the case. Especially, if the state has full Schmidt rank extension of his results for continuous variable systems. and |bki span the whole space, any state |βii on Bob’s Also, Vujiˇcićand Herbut argue that steering is differ- side can be prepared by making a suitable measurement ent from Bell nonlocality, as it is based on the formal- on Alice’s side. ism of quantum mechanics. Independently, Reid (1989) Finally, Schrödingerstresses again that he finds the presented quantitative conditions for continuous variable phenomenon of controlling a distant state repugnant and systems to lead to an EPR-type argument. Verstraete suggests that quantum mechanics may be modified to noted in his dissertation (Verstraete, 2002) the connec- avoid it. As a potential modification, he suggests that tion of Schrödinger’s ideas to quantum teleportation and for a state as in Eq. (132) the phase relations between entanglement transformations, as in both cases one aims the sk may be lost. This means that instead of taking at preparing a quantum state on one side by making mea- the pure state % = |ψihψ| the two-particle system should surements on the other. Also, the notion of the steer- be described by a diagonal mixed state ing ellipsoid was introduced there. Shortly thereafter, steering was recognized to be relevant for foundational X 2 % = sk|akbkihakbk|, (136) questions of quantum mechanics (Clifton et al., 2003; k Spekkens, 2007). Finally, Wiseman and coworkers intro- which he considers to be a possible modification, not con- duced the notion of local hidden state models (Wiseman tradicting the experimental evidence at that time. et al., 2007), laying the foundation for the modern notion of quantum steering.

3. Impact of these papers VI. CONCLUSION In the following years, Schrödinger’sideas on steering were not further considered in the literature. His math- The notion of quantum steering is motivated by the ematical results from the second paper, however, were Einstein-Podolsky-Rosen argument and it took seven several times rederived without any reference to him. In decades until a precise formulation was given. Since then, the following, we give a short overview, a detailed discus- quantum steering has initiated a new surge of results in sion can be found in (Kirkpatrick, 2006). quantum information and the foundations of quantum A first rediscovery was presented by Jaynes (Jaynes, mechanics: Old concepts were put into the new light; 1957). He derived the first statement in Eq. (134) while long-standing problems gained progress and some were studying general properties of density matrices. Based on resolved; connections between areas were established, Jaynes’ paper, Hadjisavvas presented later a simplified and novel problems were formulated. In this review, we 39 have sketched the dynamic development of the field over For a given directed network one may ask whether the last ten years. Yet, future research is facing many there is a quantum state allowing steering along the challenges. So, to close the review, we summarize some directed edges. of the open problems: • For applications, it would be interesting to identify • As a complete characterization of quantum steer- tasks in quantum information processing, where the ability has been obtained for two-qubit systems and assumptions that can be made are highly asymmet- projective measurements, it is desirable to extend ric. Then, the methods developed in steering the- such a characterization to higher-dimensional sys- ory may be useful to study the role of correlations tems. Although there is an indication that such an therein. extension is possible, much remains to be worked • In experiments, quantum steering is verified by out in details. a finite number of measurement settings. There • The question, whether there are states that are are only few works on optimizing the measure- unsteerable with projective measurements, but are ment settings (when having a fixed number of in- steerable with POVMs is also relevant. The anal- puts/outputs) and clearly more research is needed ogous question in the context of Bell nonlocality to serve as inputs for experiments. has been a long-standing problem without any evidence whether such a state exists. With quan- This is only a small list of problems, and clearly further tum steering, one now has evidence indicating that interesting challenges remain. Given the current inter- there might be no such state for a two-qubit sys- est in the steering phenomenon, we expect that the old tem. Yet, to date there is no rigorous proof of their observations from Schrödingercan still influence current non-existence. In particular, one might still expect and future discussions on the foundations of quantum that such a state exists in high dimensions. mechanics. We thank R. M. Angelo, N. Brunner, C. Bu- • We have discussed in Section III.G that the oper- droni, S.-L. Chen, B. Coyle, S. Designolle, U. R. Fis- ational definition of steerability requires multiple cher, O. Gittsovich, R. B. Griffiths, E. Haapasalo, copies of the considered state, which implies the M. J. W. Hall, T. Heinosaari, F. Hirsch, Y. Huang, possibility of making collective measurements on M. Huber, S. Jevtic, J. Kiukas, T. Kraft, L. Lami, Alice’s side. Thus, apart from being fundamental, F. Lever, Y.-C. Liang, K. Luoma, A. Milne, T. Moroder, the question whether all entangled states become H.-V. Nguyen, J.-P. Pellonpää,M. Piani, T. Pramanik, steerable upon making collective measurements on Z. Qin, M. T. Quintino, J. Shang, G. Tóth,F. Verstraete, Alice’s side is also important to the intrinsic con- T. Vu, M. M. Wilde, H. M. Wiseman, J.-S. Xu, and X.- sistency of the concept. D. Yu for discussions and collaborations on the topic of steering or comments on the manuscript. In addition, we • The connection between quantum steering and in- thank the anonymous referees for their help. compatibility has initiated further questions: are This work has been supported by the DFG, the ERC there other connections between the different no- (Consolidator Grant 683107/TempoQ), the Finnish Cul- tions of incompatibility and different forms of quan- tural Foundation, the CNPq-Brazil (process number: tum correlations? 153436/2018-2), and the CAPES-Brazil. • A further open question asks whether there are other physically motivated properties of state as- REFERENCES semblages than that of having a local hidden state model. For instance, one may want to deduce some- Ac´ın,A., N. Brunner, N. Gisin, S. Massar, S. Pironio, and thing more than entanglement and incompatibility V. Scarani (2007), Phys. Rev. Lett. 98, 230501. from such properties. Examples are the prepara- Ac´ın,A., N. Gisin, and B. Toner (2006), Phys. Rev. A 73, bility from states with a given Schmidt number 062105. or properties motivated by incompatibility, such as Ac´ın,A., S. Massar, and S. Pironio (2012), Phys. Rev. Lett. 108, 100402. compatibility on many copies, coexistence or simu- Adesso, G., and R. Simon (2016), J. Phys. A: Math. Theor. lability. 49, 34LT02. Ali, S. T., C. Carmeli, T. Heinosaari, and A. Toigo (2009), • The study of multipartite steering is in its infancy. Found. Phys. 39, 593. A systematic investigation and comparison between Almeida, M. L., S. Pionio, J. Barrett, G. Tóth, and A. Ac´ın different definitions is necessary in the future. (2007), Phys. Rev. Lett. 99, 040403. Armstrong, S., M. Wang, R. Y. Teh, Q. Gong, Q. He, • A closely related research direction is to study J. Janousek, H.-A. Bachor, M. D. Reid, and P. K. Lam steering of parties who are connected in networks. (2015), Nat. Phys. 11, 167. 40

Augusiak, R., M. Demianowicz, and A. Ac´ın(2014), J. Phys. Reid (2009), Phys. Rev. A 80, 032112. A: Math. Theor. 47, 424002. Chen, C., R. C., X.-J. Ye, and J.-L. Chen (2018a), Phys. Baker, T. J., S. Wollmann, G. J. Pryde, and H. M. Wiseman Rev. A. 98, 052114. (2018), J. Opt. 20, 034008. Chen, J.-L., C. Ren, C. Chen, Y. X.Y., and A. K. Pati (2016), Barrett, J. (2002), Phys. Rev. A 65, 042302. Sci. Rep. 6, 39063. Bartkiewicz, K., A. Cernoch,ˇ K. Lemr, A. Miranowicz, and Chen, K., and L.-A. Wu (2003), Quant. Inf. Comp. 3, 193. F. Nori (2016), Phys. Rev. A 93, 062345. Chen, S.-L., C. Budroni, Y.-C. Liang, and Y.-N. Chen Bavaresco, J., M. T. Quintino, L. Guerini, T. O. Maciel, (2016a), Phys. Rev. Lett. 116, 240401. D. Cavalcanti, and M. T. Cunha (2017), Phys. Rev. A Chen, S.-L., C. Budroni, Y.-C. Liang, and Y.-N. Chen 96, 022110. (2018b), Phys. Rev. A 98, 042127. Bene, E., and T. Vértesi(2018), New J. Phys. 20, 013021. Chen, S.-L., N. Lambert, C.-M. Li, G.-Y. Chen, Y.-N. Chen, Bennet, A. J., D. A. Evans, D. J. Saunders, C. Branciard, A. Miranowicz, and F. Nori (2017), Sci. Rep. 7, 3728. E. G. Cavalcanti, H. M. Wiseman, and G. J. Pryde (2012), Chen, S.-L., N. Lambert, C.-M. Li, A. Miranowicz, Y.-N. Phys. Rev. X 2, 031003. Chen, and F. Nori (2016b), Phys. Rev. Lett. 116, 020503. Bennett, C. H., G. Brassard, C. Crépeau, R. Jozsa, A. Peres, Chen, Y.-N., C.-M. Li, N. Lambert, S.-L. Chen, Y. Ota, G.-Y. and W. K. Wootters (1993), Phys. Rev. Lett. 70, 1895. Chen, and F. Nori (2014), Phys. Rev. A 89, 032112. Bennett, C. H., G. Brassard, and N. D. Mermin (1992), Phys. Cheng, S., A. Milne, M. J. W. Hall, and H. M. Wiseman Rev. Lett. 68, 557. (2016), Phys. Rev. A 94, 042105. Bohm, D. (1951), Quantum Theory (Prentice Hall). Clauser, J. F., and A. Shimony (1978), Rep. Progr. Phys. 41, Bowen, W. P., R. Schnabel, P. K. Lam, and T. C. Ralph 1881. (2003), Phys. Rev. Lett. 90, 043601. Clifton, R., J. Bub, and H. Halvorson (2003), Found. Phys. Bowles, J., T. Vértesi, M. T. Quintino, and N. Brunner 33, 1561. (2014), Phys. Rev. Lett. 112, 200402. Coffman, V., J. Kundu, and W. K. Wootters (2000), Phys. Bowles, J., T. Vértesi, M. T. Quintino, and N. Brunner Rev. A 61, 052306. (2016), Phys. Rev. A 93, 022121. Coles, P. J., M. Berta, M. Tomamichel, and S. Wehner (2017), Boyd, S., and L. Vandenberghe (2004), Convex Optimization Rev. Mod. Phys. 89, 015002. (Cambridge University Press). Costa, A. C. S., and R. M. Angelo (2016), Phys. Rev. A 93, Branciard, C., E. G. Cavalcanti, S. P. Walborn, V. Scarani, 020103(R). and H. M. Wiseman (2012), Phys. Rev. A 85, 010301(R). Costa, A. C. S., R. Uola, and O. Gühne(2018a), Entropy Branciard, C., D. Rosset, Y.-C. Liang, and N. Gisin (2013), 20, 763. Phys. Rev. Lett. 110, 060405. Costa, A. C. S., R. Uola, and O. Gühne(2018b), Phys. Rev. Brunner, N., D. Cavalcanti, S. Pironio, V. Scarani, and A 98, 050104(R). S. Wehner (2014), Rev. Mod. Phys. 86, 419. Coyle, B., M. J. Hoban, and E. Kashefi (2018), Buscemi, F. (2012), Phys. Rev. Lett. 108, 200401. arXiv:1806.10565. Busch, P. (1986), Phys. Rev. D 33, 2253. Curchod , F. J., M. Johansson, R. Augusiak, M. J. Hoban, Busch, P., P. Lahti, J.-P. Pellonpää, and K.Ylinen (2016), P. Wittek, and A. Ac´ın(2017), Phys. Rev. A 95, 020102. Quantum Measurement (Theoretical and Mathematical Curty, M., M. Lewenstein, and N. Lütkenhaus (2004), Phys. Physics) (Springer). Rev. Lett. 92, 217903. Busch, P., and H.-J. Schmidt (2010), Quantum Inf. Process. Das, D., S. Sasmal, and S. Roy (2019), Phys. Rev. A 99, 9, 143. 052109. Carmeli, C., T. Heinosaari, D. Reitzner, J. Schultz, and Designolle, S., P. Skrzypczyk, F. Fröwis, and N. Brunner A. Toigo (2016), Mathematics 4, 54. (2019), Phys. Rev. Lett. 122, 050402. Carmeli, C., T. Heinosaari, and A. Toigo (2019), Phys. Rev. Einstein, A. (1948), Dialectica 2, 320. Lett. 122, 130402. Einstein, A., B. Podolsky, and N. Rosen (1935), Phys. Rev. Cavalcanti, D., A. Ac´ın,N. Brunner, and T. Vértesi(2013a), 47, 777. Phys. Rev. A 87, 042104. Evans, D. A., E. G. Cavalcanti, and H. M. Wiseman (2013), Cavalcanti, D., L. Guerini, R. Rabelo, and P. Skrzypczyk Phys. Rev. A 88, 022106. (2016), Phys. Rev. Lett. 117, 190401. Evans, D. A., and H. M. Wiseman (2014), Phys. Rev. A 90, Cavalcanti, D., and P. Skrzypczyk (2016), Phys. Rev. A 93, 012114. 052112. Fillettaz, M., F. Hirsch, S. Designolle, and N. Brunner (2018), Cavalcanti, D., and P. Skrzypczyk (2017), Rep. Prog. Phys. Phys. Rev. A 98, 022115. 80, 024001. Fine, A. (1982), Phys. Rev. Lett. 48, 291. Cavalcanti, D., P. Skrzypczyk, G. H. Aguilar, R. V. Nery, Gallego, R., and L. Aolita (2015), Phys. Rev. X 5, 041008. P. H. Souto Ribeiro, and S. P. Walborn (2015a), Nat. Com- Gehring, T., V. Händchen, J. Duhme, F. Furrer, T. Franz, mun. 6, 7941. C. Pacher, R. F. Werner, and R. Schnabel (2015), Nat. Cavalcanti, D., P. Skrzypczyk, and I. Supić(2017),ˇ Phys. Comm. 6, 8795. Rev. Lett. 119, 110501. Girdhar, P., and E. G. Cavalcanti (2016), Phys. Rev. A 94, Cavalcanti, E. G., C. J. Foster, M. Fuwa, and H. M. Wiseman 032317. (2015b), J. Opt. Soc. Am. B 32, A74. Gisin, N. (1989), Helvetica Physica Acta 62, 363. Cavalcanti, E. G., M. J. W. Hall, and H. M. Wiseman Gisin, N. (1996), Phys. Lett. A 210, 151. (2013b), Phys. Rev. A 87, 032306. Gittsovich, O., O. Gühne,P. Hyllus, and J. Eisert (2008), Cavalcanti, E. G., Q. Y. He, M. D. Reid, and H. M. Wiseman Phys. Rev. A 78, 052319. (2011), Phys. Rev. A 84, 032115. Gómez,E. S., S. Gómez,P. González,G. Cañas,J. F. Barra, Cavalcanti, E. G., S. J. Jones, H. M. Wiseman, and M. D. A. Delgado, G. B. Xavier, A. Cabello, M. Kleinmann, 41

T. Vértesi, and G. Lima (2016), Phys. Rev. Lett. 117, Jevtic, S., M. Pusey, D. Jennings, and T. Rudolph (2014), 260401. Phys. Rev. Lett. 113, 020402. Griffiths, R. B. (2019), arXiv:1901.07050. Ji, S.-W., M. Kim, and H. Nha (2015a), J. Phys. A: Math. Gühne,O., P. Hyllus, O. Gittsovich, and J. Eisert (2007), Theor. 48, 135301. Phys. Rev. Lett. 99, 130504. Ji, S.-W., J. Lee, J. Park, and H. Nha (2015b), Phys. Rev. Gühne,O., M. Mechler, G. Tóth,and P. Adam (2006), Phys. A 92, 062130. Rev. A 74, 010301. Ji, S.-W., J. Lee, J. Park, and H. Nha (2016), Sci. Rep. 6, Gühne,O., and G. Tóth(2009), Phys. Rep. 474, 1. 29729. Hadjisavvas, N. (1981), Lett. Math. Phys. 5, 327. Jia, Z.-A., Y.-C. Wu, and G.-C. Guo (2017), Phys. Rev. A Hall, M. J. W. (2018), in Reality and Measurement in Alge- 96, 032122. braic Quantum Theory, edited by M. Ozawa, J. Butterfield, Jones, S. J., H. M. Wiseman, and A. C. Doherty (2007), H. Halvorson, M. Rédei,Y. Kitajima, and F. Buscemi Phys. Rev. A 76, 052116. (Springer Singapore, Singapore) pp. 161–178. Kaur, E., X. Wang, and M. M. Wilde (2017), Phys. Rev. A Hall, M. J. W., and A. Rivas (2019), Phys. Rev. A 100, 96, 022332. 062105. Kaur, E., and M. M. Wilde (2017), J. Phys. A: Math. Theor. Händchen, V., T. Eberle, S. Steinlechner, A. Samblowski, 50, 465301. T. Franz, R. F. Werner, and R. Schnabel (2012), Nat. Kaur, E., M. M. Wilde, and A. Winter (2018), Photonics 6, 596. arXiv:1810.05627. Häseler,H., T. Moroder, and N. Lütkenhaus (2008), Phys. Kiefer, C. (2015), Albert Einstein, Boris Podolsky, Nathan Rev. A 77, 032303. Rosen: Kann die quantenmechanische Beschreibung der He, Q., L. Rosales-Zárate,G. Adesso, and M. D. Reid (2015), physikalischen Realitätals vollständigbetrachtet werden? Phys. Rev. Lett. 115, 180502. Springer. He, Q. Y., P. D. Drummond, and M. D. Reid (2011), Phys. Kirkpatrick, K. A. (2006), Found. Phys. Lett. 19, 95. Rev. A 83, 032120. Kiukas, J., C. Budroni, R. Uola, and J.-P. Pellonpää(2017), He, Q. Y., and M. D. Reid (2013), Phys. Rev. Lett. 111, Phys. Rev. A 96, 042331. 250403. Kocsis, S., M. J. W. Hall, A. J. Bennet, D. J. Sauders, and Heinosaari, T. (2016), Phys. Rev. A 93, 042118. G. J. Pryde (2015), Nat. Commun. 6, 5886. Heinosaari, T., T. Miyadera, and M. Ziman (2016), J. Phys. Kogias, I., A. R. Lee, S. Ragy, and G. Adesso (2015a), Phys. A: Math. Theor. 49, 123001. Rev. Lett. 114, 060403. Heinosaari, T., and M. Wolf (2010), J. Math. Phys. 51, Kogias, I., P. Skrzypczyk, D. Cavalcanti, A. Ac´ın, and 092201. G. Adesso (2015b), Phys. Rev. Lett. 115, 210401. Hillery, M., V. Buˇzek, and A. Berthiaume (1999), Phys. Rev. Kogias, I., Y. Xiang, Q. He, and G. Adesso (2017), Phys. A 59, 1829. Rev. A 95, 012315. Hirsch, F., M. Quintino, and N. Brunner (2018), Phys. Rev. Kriváchy, T., F. Fröwis, and N. Brunner (2018), Phys. Rev. A 97, 012129. A 98, 062111. Hirsch, F., M. T. Quintino, J. Bowles, and N. Brunner (2013), Ku, H.-Y., S.-L. Chen, C. Budroni, A. Miranowicz, Y.-N. Phys. Rev. Lett. 111, 160402. Chen, and F. Nori (2018a), Phys. Rev. A 97, 022338. Hirsch, F., M. T. Quintino, J. Bowles, T. Vértesi, and Ku, H.-Y., S.-L. Chen, N. Lambert, Y.-N. Chen, and F. Nori N. Brunner (2016a), New J. Phys 18, 113019. (2018b), Phys. Rev. A 98, 022104. Hirsch, F., M. T. Quintino, T. Vértesi,M. Navascués, and Kunjwal, R., C. Heunen, and T. Fritz (2014), Phys. Rev. A N. Brunner (2017), Quantum 1, 3. 89, 052126. Hirsch, F., M. T. Quintino, T. Vértesi,M. F. Pusey, and Lami, L., C. Hirche, G. Adesso, and A. Winter (2016), Phys. N. Brunner (2016b), Phys. Rev. Lett. 117, 190402. Rev. Lett. 117, 220502. Hoban, M. J., and A. B. Sainz (2018), New J. Phys. 20, Law, Y. Z., L. P. Thinh, J.-D. Bancal, and V. Scarani (2014), 053048. J. Phys. A: Math. Theor. 47, 424028. Hofmann, H. F., and S. Takeuchi (2003), Phys. Rev. A 68, Leinaas, J. M., J. Myrheim, and E. Ovrum (2006), Phys. 032103. Rev. A 74, 012313. Horodecki, M., and P. Horodecki (1999), Phys. Rev. A 59, Li, C.-M., K. Chen, Y.-N. Chen, Q. Zhang, Y.-A. Chen, and 4206. J.-W. Pan (2015), Phys. Rev. Lett. 115, 010402. Horodecki, M., P. Horodecki, and R. Horodecki (1996), Phys. Ma, X., and N. Lütkenhaus (2012), Quantum Inf. Comput. Lett. A 223, 1. 12, 203. Horodecki, M., P. Horodecki, and R. Horodecki (1998), Phys. Maassen, H., and J. B. M. Uffink (1988), Phys. Rev. Lett. Rev. Lett. 80, 5239. 60, 1103. Horodecki, R., P. Horodecki, M. Horodecki, and K. Horodecki Maccone, L., and A. K. Pati (2014), Phys. Rev. Lett. 113, (2009), Rev. Mod. Phys. 81, 865. 260401. Hsieh, C.-Y., Y.-C. Liang, and R.-K. Lee (2016), Phys. Rev. Marciniak, M., A. Rutkowski, Z. Yin, M. Horodecki, and A 94, 062120. R. Horodecki (2015), Phys. Rev. Lett. 115, 170401. Huang, Y. (2012), Phys. Rev. A 86, 024101. McCloskey, R., A. Ferraro, and M. Paternostro (2017), Phys. Hughston, L. P., R. Jozsa, and W. K. Wootters (1993), Phys. Rev. A 95, 012320. Lett. A 183, 14. Midgley, S. L. W., A. J. Ferris, and M. K. Olsen (2010), Jaynes, E. T. (1957), Phys. Rev. 108, 171. Phys. Rev. A 81, 022101. Jeon, I., and H. Jeong (2019), Phys. Rev. A 99, 012318. Miller, C. A., R. Colbeck, and Y. Shi (2018), J. Math. Phys. Jevtic, S., M. J. W. Hall, M. R. Anderson, M. Zwierz, and 59, 022103. H. M. Wiseman (2015), J. Opt. Soc. Am. B 32, A40. Milne, A., J. Sania, D. Jennings, H. Wiseman, and 42

T. Rudolph (2014), New J. Phys. 16, 083017. Quintino, M. T., T. Vértesi, D. Cavalcanti, R. Augusiak, Milne, A., J. Sania, D. Jennings, H. Wiseman, and M. Demianowicz, A. Ac´ın, and N. Brunner (2015), Phys. T. Rudolph (2015), New J. Phys. 17, 019501. Rev. A 92, 032107. Miranowicz, A., M. Piani, P. Horodecki, and R. Horodecki Reeb, D., D. Reitzner, and M. Wolf (2013), J. Phys. A: Math. (2009), Phys. Rev. A 80, 052303. Theor. 46, 462002. Moroder, T., J.-D. Bancal, Y.-C. Liang, M. Hofmann, and Reid, M. D. (1989), Phys. Rev. A 40, 913. O. Gühne(2013), Phys. Rev. Lett. 111, 030501. Reid, M. D. (2013), Phys. Rev. A 88, 062108. Moroder, T., and O. Gittsovich (2012), Phys. Rev. A 85, Reid, M. D., P. D. Drummond, W. P. Bowen, E. G. Cav- 032301. alcanti, P. K. Lam, H. A. Bachor, U. L. Andersen, and Moroder, T., O. Gittsovich, M. Huber, and O. Gühne(2014), G. Leuchs (2009), Rev. Mod. Phys. 81, 1727. Phys. Rev. Lett. 113, 050404. Riccardi, A., C. Macchiavello, and L. Maccone (2018), Phys. Moroder, T., O. Gittsovich, M. Huber, R. Uola, and Rev. A 97, 052307. O. Gühne(2016), Phys. Rev. Lett. 116, 090403. Rudolph, O. (2005), Quant. Inf. Proc. 4, 219. Moroder, T., M. Keyl, and N. Lütkenhaus (2008), J. Phys. Rutkowski, A., A. Buraczewski, P. Horodecki, and A: Math. Theor. 41, 275302. M. Stobińska (2017), Phys. Rev. Lett. 118, 020402. Navascués,M., S. Pironio, and A. Ac´ın(2007), Phys. Rev. Sainz, A. B., L. Aolita, M. Piani, M. J. Hoban, and Lett. 98, 010401. P. Skrzypczyk (2018), New J. Phys. 20, 083040. Navascués,M., S. Pironio, and A. Ac´ın(2008), New J. Phys. Sainz, A. B., N. Brunner, D. Cavalcanti, P. Skrzypczyk, and 10, 073013. T. Vértesi(2015), Phys. Rev. Lett. 115, 190403. Nguyen, H. C., and O. Gühne(2019), arXiv:1909.03963. Saunders, D. J., S. J. Jones, H. M. Wiseman, and G. J. Pryde Nguyen, H. C., and O. Gühne(2020), arXiv:2001.03514. (2010), Nat. Phys. 6, 845. Nguyen, H. C., A. Milne, T. Vu, and S. Jevtic (2018), J. Scarani, V., H. Bechmann-Pasquinucci, N. J. Cerf, M. Duˇsek, Phys. A: Math. Theor. 51, 355302. N. Lütkenhaus, and M. Peev (2009), Rev. Mod. Phys. 81, Nguyen, H. C., H.-V. Nguyen, and O. Gühne(2019), Phys. 1301. Rev. Lett. 122, 240401. Schneeloch, J., C. J. Broadbent, S. P. Walborn, E. G. Caval- Nguyen, H. C., and T. Vu (2016a), Europhys. Lett. 115, canti, and J. C. Howell (2013), Phys. Rev. A 87, 062103. 10003. Schrödinger,E. (1935), Proc. Cambridge Philos. Soc. 31, 555. Nguyen, H. C., and T. Vu (2016b), Phys. Rev. A 94, 012114. Schrödinger,E. (1936), Proc. Cambridge Philos. Soc. 32, 446. Olsen, M. K. (2013), Phys. Rev. A 88, 051802(R). Schwonnek, R., L. Dammeier, and R. F. Werner (2017), Phys. Oszmaniec, M., L. Guerini, P. Wittek, and A. Ac´ın(2017), Rev. Lett. 119, 170404. Phys. Rev. Lett. 119, 190501. Sedlák,M., D. Reitzner, G. Chiribella, and M. Ziman (2016), Palazuelos, C. (2012), Phys. Rev. Lett. 109, 190401. Phys. Rev. A 93, 052323. Pankowski,L., M. Piani, M. Horodecki, and P. Horodecki Shchukin, E., and W. Vogel (2005), Phys. Rev. Lett. 95, (2010), IEEE Transactions on Information Theory 56, 230502. 4085. Shi, M., F. Jiang, C. Sun, and J. Du (2011), New J. Phys. Passaro, E., D. Cavalcanti, P. Skrzypczyk, and A. Ac´ın 13, 073016. (2015), New J. Phys. 17, 113010. Shi, M., C. Sun, F. Jiang, X. Yan, and J. Du (2012), Phys. Pellonpää,J.-P. (2014), J. Phys. A: Math. Theor. 47, 052002. Rev. A 85, 064104. Peres, A. (1996a), Phys. Rev. A 54, 2685. Shor, P. W., J. A. Smolin, and A. V. Thapliyal (2003), Phys. Peres, A. (1996b), Phys. Rev. Lett. 77, 1413. Rev. Lett. 90, 107901. Peres, A. (1999), Found. of Phys. 29, 589. Skrzypczyk, P., and D. Cavalcanti (2018), Phys. Rev. Lett. Piani, M. (2015), J. Opt. Soc. Am. B 32, A1. 120, 260401. Piani, M., and J. Watrous (2009), Phys. Rev. Lett. 102, Skrzypczyk, P., M. Navascués, and D. Cavalcanti (2014), 250501. Phys. Rev. Lett. 112, 180404. Piani, M., and J. Watrous (2015), Phys. Rev. Lett. 114, Skrzypczyk, P., I. Supić,ˇ and D. Cavalcanti (2019), Phys. 060404. Rev. Lett. 122, 130403. Pirandola, S., J. Eisert, C. Weedbrook, A. Furusawa, and Smith, D. H., G. Gillett, M. P. de Almeida, C. Branciard, S. L. Braunstein (2015), Nat. Photonics 9, 641. A. Fedrizzi, T. J. Weinhold, A. Lita, B. Calkins, T. Gerrits, Popescu, S. (1995), Phys. Rev. Lett. 74, 2619. H. M. Wiseman, S. W. Nam, and A. G. White (2012), Nat. Popescu, S., and D. Rohrlich (1994), Found. Phys. 24, 379. Commun. 3, 625. Primas, H. (1983), In: Moderne Naturphilosophie, edited by Spekkens, R. W. (2007), Phys. Rev. A 75, 032110. B. Kanitscheider, Königshausen + Neumann, Würzburg, Stano, P., D. Reitzner, and T. Heinosaari (2008), Phys. Rev. p. 243-260. A 78, 012315. Pusey, M. F. (2013), Phys. Rev. A 88, 032313. Sun, K., X.-J. Ye, Y. Xiao, X.-Y. Xu, Y.-C. Wu, J.-S. Xu, Qin, Z., X. Deng, C. Tian, M. Wang, X. Su, C. Xie, and J.-L. Chen, C.-F. Li, and G.-C. Guo (2018), npj Quantum K. Peng (2017), Phys. Rev. A 95, 052114. Information 4, 12. Quan, Q., H. Zhu, H. Fan, and W.-L. Yang (2017), Phys. Sun, K., X.-J. Ye, J.-S. Xu, X.-Y. Xu, J.-S. Tang, Y.-C. Wu, Rev. A 95, 062111. J.-L. Chen, C.-F. Li, and G.-C. Guo (2016), Phys. Rev. Quan, Q., H. Zhu, S.-Y. Liu, S.-M. Fei, H. Fan, and W.-L. Lett. 116, 160404. Yang (2016), Sci. Rep. 6, 22025. Tavakoli, A., and R. Uola (2019), arXiv:1905.03614. Quintino, M. T., N. Brunner, and M. Huber (2016), Phys. Tischler, N., F. Ghafari, T. J. Baker, S. Slussarenko, R. B. Rev. A 94, 062123. Patel, M. M. Weston, S. Wollmann, L. K. Shalm, V. B. Quintino, M. T., T. Vértesi, and N. Brunner (2014), Phys. Verma, S. W. Nam, H. C. Nguyen, H. M. Wiseman, and Rev. Lett. 113, 160402. G. J. Pryde (2018), Phys. Rev. Lett. 121, 100401. 43

Tóth,G. (2005), Phys. Rev. A 71, 010301. C.-J. Zhang, and L. Ye (2019), arXiv:1909.09327. Tóth,G., and O. Gühne(2005), Phys. Rev. A 72, 022340. Yu, B.-C., Z.-A. Jia, Y.-C. Wu, and G.-C. Guo (2018a), Phys. Uffink, J., and M. Seevinck (2008), Phys. Lett. A 372, 1205. Rev. A 98, 052345. Uola, R., C. Budroni, O. Gühne, and J.-P. Pellonpää(2015), Yu, B.-C., Z.-A. Jia, Y.-C. Wu, and G.-C. Guo (2018b), Phys. Phys. Rev. Lett. 115, 230402. Rev. A 97, 012130. Uola, R., T. Bullock, T. Kraft, J.-P. Pellonpää,and N. Brun- Yu, S., and N. Liu (2005), Phys. Rev. Lett. 95, 150504. ner (2019a), arXiv:1909.10484. Yu, S., N. Liu, L. Li, and C. Oh (2010), Phys. Rev. A 81, Uola, R., T. Kraft, J. Shang, X.-D. Yu, and O. Gühne 062116. (2019b), Phys. Rev. Lett. 122, 130404. Yu, S., and C. H. Oh (2017), Phys. Rev. A 95, 032111. Uola, R., F. Lever, O. Gühne, and J.-P. Pellonpää(2018), Zhang, C., S. Cheng, L. Li, Q.-Y. Liang, B.-H. Liu, Y.-F. Phys. Rev. A 97, 032301. Huang, C.-F. Li, G.-C. Guo, M. J. Hall, H. M. Wiseman, Uola, R., K. Luoma, T. Moroder, and T. Heinosaari (2016), and G. J. Pryde (2019), Phys. Rev. Lett. 122, 070402. Phys. Rev. A 94, 022109. Zhang, C.-J., Y.-S. Zhang, S. Zhang, and G.-C. Guo (2008), Uola, R., T. Moroder, and O. Gühne(2014), Phys. Rev. Lett. Phys. Rev. A 77, 060301(R). 113, 160403. Zhang, F.-L., and Y.-Y. Zhang (2019), Phys. Rev. A 99, Uola, R., G. Vitagliano, and C. Budroni (2019), Phys. Rev. 062314. A 100, 042117. Zhen, Y.-Z., Y.-L. Zheng, W.-F. Cao, L. Li, Z.-B. Chen, N.-L. Vallone, G. (2013), Phys. Rev. A 87, 020101(R). Liu, and K. Chen (2016), Phys. Rev. A 93, 012108. Verstraete, F. (2002), Ph.D. Thesis, Katholieke Universiteit Zhou, C., P. Xu, W.-S. Bao, Y. Wang, Y. Zhang, M.-S. Jiang, Leuven. and H.-W. Li (2017), Opt. Express 25, 16971. Verstraete, F., J. Dehaene, and B. De Moor (2003), Phys. Zhu, H., M. Hayashi, and L. Chen (2016), Phys. Rev. Lett. Rev. A 68, 012103. 116, 070403. Vértesi,T., and E. Bene (2010), Phys. Rev. A 82, 062115. Zukowski,˙ M., A. Zeillinger, M. A. Horne, and H. Weinfurter Vértesi,T., and N. Brunner (2014), Nat. Commun. 5, 5297. (1998), Acta Phys. Pol. A 93, 187. von Meyenn, K. (2011), Eine Entdeckung von ganz außerordentlicher Tragweite. SchrödingersBriefwechsel zur Wellenmechanik und zum Katzenparadoxon, Springer. Vujiˇcić,M., and F. Herbut (1988), J. Phys. A: Math. Gen. 21, 2931. Walborn, S. P., A. Salles, R. M. Gomes, F. Toscano, and P. H. Souto Ribeiro (2011), Phys. Rev. Lett. 106, 130402. Walk, N., S. Hosseini, J. Geng, O. Thearle, J. Y. Haw, S. Arm- strong, S. M. Assad, J. Janousek, T. C. Ralph, T. Symul, H. M. Wiseman, and P. K. Lam (2016), Optica 3, 634. Wang, Y., W.-S. Bao, H.-w. Li, C. Zhou, and Y. Li (2013), Phys. Rev. A 88, 052322. Wehner, S., and A. Winter (2010), New J. Phys. 12, 025009. Werner, R. F. (1984), In: Reduction in Science, edited by W. Balzer, D.A. Pearce, and H.-J. Schmidt Reidel, Dordrecht, p. 419-442 (1984). Werner, R. F. (1989), Phys. Rev. A 40, 4277. Werner, R. F. (2014), J. Phys. A: Math. Theor. 47, 424008. Weston, M. M., S. Slussarenko, H. M. Chrzanowski, S. Woll- mann, L. K. Shalm, V. B. Verma, M. S. Allman, S. W. Nam, and G. J. Pryde (2018), Science Advances 4, e1701230. Wiseman, H. M., S. J. Jones, and A. C. Doherty (2007), Phys. Rev. Lett. 98, 140402. Wittmann, B., S. Ramelow, F. Steinlechner, N. K. Langford, N. Brunner, H. M. Wiseman, R. Ursin, and A. Zeilinger (2012), New J. Phys. 14, 053030. Wolf, M., D. Perez-Garcia, and C. Fernandez (2009), Phys. Rev. Lett. 103, 230402. Wollmann, S., R. Uola, and A. C. S. Costa (2019), arXiv:1909.04001. Wollmann, S., N. Walk, A. J. Bennet, H. M. Wiseman, and G. J. Pryde (2016), Phys. Rev. Lett. 116, 160403. Wu, S., S. Yu, and K. Mølmer (2009), Phys. Rev. A 79, 022104. Xiang, Y., I. Kogias, G. Adesso, and Q. He (2017), Phys. Rev. A 95, 010101(R). Xiao, Y., X.-J. Ye, K. Sun, J.-S. Xu, C.-F. Li, and G.-C. Guo (2017), Phys. Rev. Lett. 118, 140404. Yang, H., Z.-Y. Ding, D. Wang, H. Yuan, X.-K. Song, J. Yang,