Quantifying Quantum-Mechanical Processes

The act of describing how a physical process changes a system is the basis for understanding observed phenomena. For quantum-mechanical processes in particular, the affect of processes on quantum states profoundly advances our knowledge of the natural world, from understanding counter-intuitive concepts to the development of wholly quantum-mechanical technology. Here, we show that quantum-mechanical processes can be quantified using a generic classical-process model through which any classical strategies of mimicry can be ruled out. We demonstrate the success of this formalism using fundamental processes postulated in quantum mechanics, the dynamics of open quantum systems, quantum-information processing, the fusion of entangled photon pairs, and the energy transfer in a photosynthetic pigment-protein complex. Since our framework does not depend on any specifics of the states being processed, it reveals a new class of correlations in the hierarchy between entanglement and Einstein-Podolsky-Rosen steering and paves the way for the elaboration of a generic method for quantifying physical processes.


Results
Classical processes. We define a classical process as a set of steps involving the general descriptions of a classical state and its evolution: the initial system can be considered a physical object with properties satisfying the assumption of realism; 20 then, the system evolves according to classical stochastic theory 26 (Fig. 1a). The assumption of realism specifies that the system is in a state described by a set of measurement outcomes, for example, a set of outcomes for N physical properties of interest v ξ ≡ (v 1 , v 2 , ..., v N ). If each physical property has d states, then we have d N possible sets, ... v v , , d we focus on a specific initial state of the kth physical property, e.g., v k = v′ k , then the corresponding final state of the system has the form Indeed, the final states (1) conditioned on different properties and states v′ k can be used to capture the essence of the classical process. We use process tomography (PT), an application of the quantum operations formalism 4 , to systematically exploit these experimentally measurable quantities and then completely characterize the classical process using a positive Hermitian matrix, called the process matrix, We will hereafter use a process matrix to refer to a physical process within the text. In the following, we will illustrate the derivation of a classical process matrix.
Derivation of χ c for classical processes. In order to show explicitly how to apply PT to a classical process to completely characterize its classical features, a classical-process scenario for single two-level systems is given as a concrete example of Eq. (2). Since a classical process treats the initial system as a physical object with properties satisfying the assumption of realism, the system can be considered as a state described by a fixed set v ξ . We assume that the system is described by three properties, say V 1 , V 2 and V 3 , where each one has two possible states. There exist 2 3 = 8 realistic sets underlying the classical object: v ξ (v 1 , v 2 , v 3 ), where v 1 , v 2 , v 3 ∈ {+1, −1} represent the possible measurement outcomes for V 1 , V 2 and V 3 , respectively. The subsequent classical evolution changes the system from v ξ to a final state denoted by μ according to the transition probabilities v Ω µ ξ . Such evolution can always be rephrased as the transition from a specific state set v ξ′ to some final state μ′ with Ω ′ = ′ µ ξ 1 v . Next, by using state tomography, each final state is reconstructed as a density operator ρ μ′ . Then the states under the assumption of realism evolve according to  Figure 1. Quantifying quantum-mechanical processes. (a) Suppose that a physical process is experimentally determined by a process matrix χ expt ; how a system evolves from an arbitrary initial state ρ initial to some final state ρ final is specified by the process matrix χ expt through the mapping χ expt (ρ initial ) = ρ final , which preserves the Hermiticity, trace, and positivity of the system density matrix. The amount of quantumness χ Q of the process, which cannot be described at all by any classical processes χ C , can be characterized and quantified by α (composition), β (robustness), F expt (process fidelity) and S (von Neumann entropy). For instance, for a perfect (worst) experiment on a target quantum process, α, β and F expt will attain their individual maximum (minimum) values whereas S will reach the minimum (maximum) uncertainty of the quantum process. These variables have significant applications to aid in the exploration and evaluation of all physical processes described by the quantum operations formalism, such as (b,c) the dynamics of open quantum systems, (d) the generation of multipartite entanglement, and (e,f) quantum-information processing. (g) This framework shows a new correlation model in the class between genuine multipartite EPR steering and genuine multipartite entanglement, called the χ C -nonclassical correlations.
Scientific REPORTS | 7: 13588 | DOI:10.1038/s41598-017-13604-9 We now consider specific states of physical properties as the input states. If we focus on a state of the third property, say v 3 = v′ 3 , the final state is described as v c v final ( ) 3 3 3 shows the probability of transition from v 3 ′ for all the possible sets v ξ to the final state ρ μ . The transition probabilities therein read for v′ 3 = +1 and v′ 3 = −1, respectively. See Eq. (1). Since P(v′ 3 )P(v ξ |v′ 3 ) = P(v ξ )P(v′ 3 |v ξ ) and P(v′ 3 ) = 1/2 under the assumption of a uniform probability distribution of v k , the final states are written as Similarly, for the other states v′ 1 = ±1, v′ 2 = ±1 under the condition P(v′ 1 ) = P(v′ 2 ) = 1/2, the classical process has the following output states: The essence of PT is that a process of interest is completely characterized by a process matrix. Using the outputs of three complementary observables (e.g. the Pauli matrices I, X, Y, and Z) from the process 4 , it is experimentally feasible to determine the process matrix. A classical process makes these observables decomposable in terms of Eqs (4) and (5): . Then the classical process matrix specifying how states evolve regardless of inputs can be written as the form: where ρ C,00 The above concepts and methods can be extended to multi-level and multipartite physical systems. For instance, a d-level system can be classically described by a fixed set v ξ with d 2 − 1 properties. As illustrated in Eq.
(3), the system evolves according to classical stochastic theory For a given initial state of a specific property, the final state can be written as the same form as Eqs (4) and (5) . Furthermore, the classical process makes d 2 − 1 complementary observables (e.g., the generalized Pauli matrices 27 ) chosen for PT decomposable in terms of the final states v final k ρ ′ | . These observables then can be used to determine χ C of the classical process for the d-level system.
Quantifying quantum-mechanical processes. We now turn to the question of how to quantitatively characterize quantum-mechanical processes. Suppose that a process of interest is created and that its normalized process matrix, χ expt , is derived from experimentally available data using the PT procedure. If the experimental result cannot be described at all by any classical processes, then we say that χ expt is a genuinely quantum process, denoted by χ Q (Fig. 1a). To place this concept into a wider context, we introduce four different approaches for the quantitative characterization of χ Q in χ expt : (A1) Quantum composition: where α denotes the minimum amount of χ Q that can be found in χ expt . The minimum amount of χ Q that can be found in χ expt is obtained by minimizing the following quantity via semi-definite programming (SDP) with MATLAB 28,29 : where χ Q  and  χ C are both unnormalized process matrices. (A2) Process robustness: where β represents the minimum amount of the noise process χ′. The minimum amount of noise process is determined via SDP: Scientific REPORTS | 7: 13588 | DOI:10.1038/s41598-017-13604-9 The first criterion in (12) guarantees that β ≥ 0, and the rest ensures that the noise χ′ and the output states ρ μ are positive semi-definite. For example, when 0 C expt  χ χ − = , χ expt is a genuinely classical process with β = 0. (A3) Fidelity: χ expt is detected as nonclassical close to a target quantum process Q T χ if the process fidelity satisfies that , stated in terms of the average state fidelity 30 . The concept of (13) is to rule out the best classical mimicry to an ideal target quantum process Q T χ . Such capability of classical mimicry is evaluated by performing the following maximization task via SDP: The first constraint in (15) ensures that χ  C is a normalized process matrix satisfying the definitions of the fidelity and a density operator. (A4) Entropy: The process is certified as truly quantum mechanical if C C expt C where S(χ expt(C) ) ≡ − tr(χ expt(C) logχ expt(C) ). Here, S C can be analytically derived from the basic properties of χ C and S. See Methods for the proof.
While the approaches (A1)-(A3) are based on different concepts and points of view, the following three properties of quantum and classical processes reveal close connections between the quantities derived from them: (P1) If an experimental process consists of two sub-processes: χ expt = χ 2 χ 1 , where χ k = α k χ Qk + (1 − α k )χ Ck for k = 1, 2, then both its β β ≡ χ χ are not larger than that of the individual sub-process χ k , i.e., . (P2) Not only does (A3) indicate whether χ expt is reliable with respect to Q T χ , both α and β reflect the reliability of χ expt as well: χ expt is verified as reliable close to , then χ expt is identified as nonclassical and close enough to χ Q T . (P3) Suppose that, given a process χ expt , a classical process χ C is derived from the definition of process robustness (A2). Only when the fidelity of χ C and the target χ Q T may reach the classical upper bound of the process fidelity F C .
The proof of (P1) is based on a condition under which two sub-processes can constitute a quantum-mechanical process. Since a classical process matrix is composed of the matrix elements specified by the classical theories for the initial system and the subsequent state transitions, the product of such matrix and any second process matrix is still a classical one since its matrix elements are linear combinations of those matrix elements of the classical process matrix.
A product of two sub-process matrices is quantum only when both the sub-processes are nonclassical. This implies that only the term χ Q2 χ Q1 with an intensity α 2 α 1 in the expansion of χ 2 χ 1 is not a classical process. Then 2 1 α χ χ determined by SDP can be smaller than or equal to α 2 α 1 , which concludes that Compared with the individual sub-process χ k , the process robustness of χ expt remains or decreases in its intensity β χ χ 2 1 in response to such a non-increasing quantum composition. For (P2), the conditions for a reliable process can be shown by using the reliability criterion, F expt > F C , and the basic definitions of α, β and the process fidelity. These conditions on α and β can be represented in terms of average state fidelities as well.
The last property reminds us that, from the point of view of process robustness, only when the derived χ C is the classical process that has the minimum deviation from Q T χ by an amount of noise β. Therefore the fidelity of χ C and χ Q T may be larger than any other classical similarities and then equal to F C .  Fig. 2a. Moreover, α and β provide a fruitful source of information regarding a measure and the finest detection of non-Markovian dynamics of a system coupled to an environment. If an experimental process is Markovian, χ expt can be arbitrarily divided into sub-process matrices which satisfy the law of decomposition 24,25 χ expt = χ 2 χ 1 . For instance, the property (P1) implies that, with explicit reference to the passage of time for each sub-process, α and β should monotonically decrease with time for a Markovian process. Hence, if we observe an increasing result from χ expt , then the process is verified as non-Markovian. Furthermore, there should be no differences between α and β of the whole process, expt α χ and expt β χ , and those of the process composed of two sub-processes, 2 1 α χ χ and 2 1 β χ χ , for Markovian dynamics. An invalidation of this consistency reveals that the process is non-Markovian. See Fig. 2b.

Examples and applications.
(E3) Fusion of entangled photon pairs. Our framework inherits the far-reaching utility of the quantum operations formalism such that quantum dynamics can be explored by our novel quantification under a wide range of circumstances. The fusion of entangled photon pairs 32 superposes two individual photons in two different spatial modes at a polarizing beam splitter (PBS) and post-selects both outputs in different modes (Fig. 1d) The FMO complex is a seven-site structure used by certain types of bacteria to transfer excitations from a light-harvesting antenna to a reaction centre (Fig. 1c). Figure 3 suggests the first quantifications of nonclassical energy transfer in the FMO complex 33,34 , where several pigments are chosen as a subsystem and single excitation transport is considered therein.
(E5) Quantum computation. We now examine concrete scenarios in which our formalism offers general benchmarks for quantum information. A valid quantum gate is specified by a unitary transformation (α = 1), and an arbitrary quantum gate can be expressed using single qubit and controlled-NOT (CNOT) gates 4 (Fig. 1e). We say that an experiment reliably implements quantum-information processing if χ expt goes beyond the classical descriptions, such as superconducting circuits used for quantum information 5,8 and the quantum gates realized by the IBM quantum computer 35 ; see Table 1.
(E6) Quantum communication. An ideal qubit transmission between two parties acts as an identity unitary transformation on the transmitted qubit, which can be implemented by either sending qubits through an ideal communication channel 36 or using teleportation 37 (Fig. 1f) to move qubits around 6 . For teleportation, both α and β can reflect the qualities of entangled states shared between the sender and the receiver; see Fig. 4a. In particular, our state-fidelity threshold is tighter than the well-known upper bound on the classical teleportation (i.e., F 2/3 0 667 s ,expt = ∼ . 38 ) and guarantees faithful teleportation of the entangled qubits 39 (Fig. 4b). Classical teleportation is a measure-prepare scenario in which the sender measures the unknown input state directly, and then sends the results to the receiver to prepare the output state 38,40 . Such measure-prepare strategy attains its maximum process fidelity F expt = 1/2 at the output state fidelity 2/3 for all arbitrary input states, and therefore is weaker , for q = α, β. As shown in Fig. 1b, we consider a system that is coupled to an environment with a state and depolarized with a rate γ. For example, we have ∼ . when t < 1.1, whereas they certify the dynamics as Markovian. The certifications by the BLP and RHP measures are detailed in ref. 77 Indeed, our method is finer than the BLP and RHP measures for all the settings of γ and p considered therein. than the best classical strategy with F 0 683 C ∼ . and F 0 789 s C , ∼ . found by our method. Alternatively, the criterion S(χ expt ) < 1 restricts the external disturbance to quantum-information processing, which remarkably coincides with the existing result for qubit transmission under coherent attacks [41][42][43] .
Usage and comparison. As illustrated above, (A1)-(A4) can quantify the quantum nature of processes applied to a quantum systems in a wide variety of circumstances. The classification of an experimental process based upon its purpose determines exactly which of the methods (A1-A4) is most useful. For example, compared to (A1) and (A2), for the task-oriented process aiming to experimentally realize quantum-information processing, (A3) can be used to directly evaluate whether χ expt is close to χ Q T and superior to the best mimicry of a classical process. However, for general experiments with the purpose of investigating whether χ expt is a quantum process, such as the energy transfer in FMO complex, (A1) and (A2) offer the advantage in performing two different types of quantitative analysis. The former focuses on the quantum composition of χ expt and concretely determines the maximum proportion of the classical process of χ expt in terms of 1 − α. See Eqs (7) and (8). Whereas, (A2) characterizes how close χ expt is to a classical process in the sense that how large the minimum amount of noise, β, is required to make χ expt classical [Eqs (10) and (11)]. Such a notion helps us understand and appreciate the roles α and β have played in the quantitative analysis. For instance, it is easy to see why an experimental process may possess β which is much smaller α, as shown in Fig. 2a for χ AD at p noise → 1.

Quantum correlations.
With our classical-process model (2) at hand, we can be precise regarding the statement about final states generated by a generic classical process, and uncover new characteristics of quantum states. Let us consider a composite system of N qubits and divide the system into two groups, A and B, consisting of n A and n B qubits, respectively, where n B ≥ 1, and n A + n B = N. An N-qubit state is called χ C -nonclassical iff it cannot be generated by performing any classical processes on each qubit in A: denotes any operation composed of classical processes for each single qubit in A on an initial state ρ κ initial ( ) (Fig. 1g), and κ signifies the bipartition type for A and B. Otherwise, the state is called the χ C -classical state. When considering all the possible partitions of κ, we call a state genuinely multipartite χ C -nonclassical iff it cannot be repre- for all possible bipartitions and probability distributions of p κ . The basic concept behind χ C -classical states can be considered a hybrid of separable-states 44,45 and the local hidden state (LHS) 46-48 models, implying a new property between genuine multipartite entanglement and genuine multipartite Einstein-Podolsky-Rosen (EPR) steering 49,50 , as shown in Methods.
A witness operator that detects genuinely multipartite χ C -nonclassical states that are close to a pure target state T ψ is given by  Characterizing quantum states with process quantifications. Note that the characterization of quantum states can benefit by including a quantum-mechanical process. For example, EPR steering 46-48 can be enlarged by considering that the untrusted party proceeds to perform a quantum-information process, e.g., teleportation (Fig. 4) or one-way quantum computing 52 . Moreover, the model of quantum process explicitly sheds light on the temporal analogue of EPR steering [53][54][55][56] and naturally provides its optimum quantification, which cannot be provided by existing methods 57 .
Let us take the temporal steering for single systems transmitted by a sender, Alice, to a receiver, Bob as an example. The concepts of EPR steering and the LHS model are used for timelike separations between Alice and Bob. For instance, in the temporal version of the LHS model, the joint probability of observing v a by Alice at time t a and v b by Bob at time t b , where t a < t b , is specified by , where σ μ denotes the state of system held by Bob. It is easy to see that this representation of the joint probability can be described through Eq. (1) in the model of classical process, i.e., Our formalism can explain the rationale behind the temporal version of the LHS model and show the result that cannot be provided by existing methods, such as the optimal quantification of temporal steering. The approach introduced in 57 is parallel to the method for quantifying EPR steering. The state of Bob's system conditioned on Alice's result v a t , a can be described by and 0.574, respectively), for single-qubit and two-qubit gate operations, respectively. Using (A4), conditioned on logarithms to base 2, their entropies are all less than the ultimate entropies of classical process S C = N, where N denotes the number of qubits being processed. cos 00 sin 11 is used for teleportation (Fig. 1f). The entanglement of φ θ ( ) measured by concurrence C(θ) = |sin2θ| can be strictly revealed by α and β for the teleportation process. In particular, α exactly coincides with C. (b) Using the relation C ≥ 2F expt − 1 78 , as C F , (yellow region), two such entangled pairs enable teleportation of entanglement of qubits 39 . Compared to the steerable weight for quantifying EPR steering that are maximum for all pure entangled states 79 , both α and β can provide the qualities of entanglement previously shared between the sender and receiver for teleportation.
Scientific REPORTS | 7: 13588 | DOI:10.1038/s41598-017-13604-9 Our method quantifies the optimal temporal steering. One can use α to represent the maximum temporal steering that can be found in a process through single systems. It is easy to see that, after a process χ expt (7), an initial state ρ initial becomes To faithfully show the effects of a process on the system, ρ final is assumed to be pure. Then χ Q (ρ initial ) is still pure to go beyond the description (1). Whereas, by Eqs (1) and (4), χ C (ρ initial ) follows the classical model, which explains the unsteerable state by Compared with the steerable weight, α is optimum for all input states and therefore larger than τ under a given assemblage σ { } v T k with finite elements. See Table 2 for concrete illustrations and comparison.

Discussion
In this work, we clarified and broadened basic ideas behind the distinction between classicality and quantumness, addressing the most basic problem of how to quantitatively characterize physical processes in the quantum world. We showed for the first time that quantum-mechanical processes can be quantified. We revealed that such quantification can have profound implications for the understanding of quantum mechanics, quantum dynamics, and quantum-information processing. Our approach is more general than many existing methods, and much broader in scope than theories based on state analysis. Our formalism is applicable in all physical processes described by the general theory of quantum operations, including but not limited to the fundamental processes postulated in quantum mechanics, the dynamics of open quantum systems, and the task-orientated processes for quantum technology. This far-reaching utility of our framework enables us to explore quantum dynamics under a wide range of circumstances, such as the fusion of entangled photon pairs and the energy transfer in a photosynthetic pigment-protein complex. In addition, our formalism enables quantum states to be characterized in new ways, to uncover new properties of both composite and single systems. Since all of our approaches are experimentally feasible, they can be readily implemented in a wide variety of the present experiments 3,32 , such as the quantum channel simulator 58 and ground-to-satellite teleportation 59 . However, it is important to have a clear appreciation for the limitations of the quantum operations formalism underlying the constructions for our framework, such as the assumption of a system and environment initially in a product state 4,25,60 . Such prior knowledge about the system and environment is therefore required to perform process quantifications.
For future studies and applications of our concept and methods, we anticipate their use in general physical processes, such as superpositions 3 , asymmetries 31 , and randomness 61 . Using modern machine learning techniques 62 could improve the performance and scalability of PT and quantification of complex system processes, such as those found in condensed-matter physics. Furthermore, provided the measurement outcomes are continuous and unbound, it is enlightening to attempt to extend our formalism to encompass the quantifications of nonclassical processes in harmonic systems such as nanomechanical resonators 63 . These essential elements could promote novel recognition and classification of physical processes with a generic process quantifier.  Table 2. Comparison of quantum composition (α) and temporal steerable weight (τ). We consider how temporal steering is demonstrated by single qubits undergoing practical channels in IBM Q 35 and quantified in terms of α and τ 57 . These experimental channels are created with respect to the gates I, H and T. Here, the subscript of τ n indicates that an assemblage σ { } v T k with n elements has been used to determine τ for the experimental states. In determining τ 2 , the eigenstates of X and Y were chosen as the input states of the experimental channels. For n > 2, the eigenstates of X, Y and Z together with 2n − 6 randomly-chosen pure states as the input states were used in calculating τ n . Since our framework is optimal, α is higher than τ n for each experimental channel, irrespective of the type and number of input states chosen to obtain τ.
Projective measurements is an important special case of the measurement postulate where the measurement operators satisfy the conditions of projectors, Quantum transport in the FMO complex. Distinguishing quantum from classical processes for the energy transport in the FMO pigment-protein complex 33,34 is crucial to appreciate the role the nonclassical features play in biological functions 2 . Figure 3 shows that the quantum transport in the FMO complex is identified and quantified on considerable timescales. Here we assume that the FMO system is in the single-excitation state of the form: 66,67 i j , where j in the site basis represents the excitation is shown at site j and E means an empty state in the absence of excitation. The time evolution of the state ρ is described by the Lindblad master equation: The incoherent dynamics is described by the three Lindblad superoperators  diss , sink  , and  deph in (21). The superoperator  diss specifies the dissipative recombination of excitation by where the dephasing rates at each site are 2 1 ps i 1 γ ∼ . − and 9.1 ps −1 for 77 K and 298 K, respectively 69,70 . To quantify the quantum transfer in the FMO system, taking the subsystem composed of the pigments 4, 5 and 6 for example, we implement PT on this subsystem to get the corresponding process matrix χ expt (t). We first use eight properties which correspond to a set of eight complementary observables {V k } where each one has three possible outcomes v k ∈ {+1, 0, −1} to describe such a three-dimensional subsystem. As illustrated at the beginning of the Methods section, a process matrix χ t ( ) expt  can be obtained by analyzing the outputs of the eight complementary observables from the process: denotes the eigenstate of V k corresponding to the eigenvalue ′ v k under the time evolution specified by Eq. (21). It is clear that V k,expt (0) = V k . Note that, since the excitation can transfer between all the seven pigments and eventually leave the subsystem, the process matrix specifies a probability of observing single excitation transport in the subsystem 71 . Here our approaches (A1) and (A2) are applied to quantify the normalized process matrix χ expt expt expt under time evolution, as shown in Fig. 3. With our tool at hand, one can quantitatively investigate how the characteristics of the FMO system change under a variety of external operations or noise processes 72-74 . Criterion for reliable qubit transmission. For the threshold S C = 1 for single two-level systems (d = 2), the classical processes with the minimum entropy S C show that the maximum mutual dependence between the sender and receiver's results of two complementary measurements: ≡ ∑ = I I SR k S R 1 2 k k , is restricted by I SR,C = 1, where I S R k k denotes the mutual information between their results of the kth measurement. Hence I SR > I SR,C indicates that their communication process is reliable. For example, considering a phase damping channel χ PD with noise intensity p noise = 1 which is identified as a classical process, we have the mutual information = I 1 S R C , 1 1 measured in the basis { 0 , 1 } and = I 0 S R C , 2 2 in the basis + − { , } where ( 0 1 ) / 2 ± = ± . When rephrasing I SR in terms of the average state fidelity F s and the error rate D = 1 − F s by S R C SR C s s , , Comparison of entanglement, steering and χ C -nonclassical correlations. We first assume that the measurement outcomes for each qubit correspond to some observable with a set of eigenvalues {v a,k } or {v b,k } for the k th qubit in A and B, respectively. The classical realistic elements v ξ and v Ω µ ξ in a classical process performed on the k th qubit in A prescribe the initial state of qubit with v v a k a k , , = ′ a final state composed of states ρ μ , as shown in Eqs (1) and (2) in the main text. After χ C A ( ) has been done on ρ κ initial ( ) , the corresponding characteristics of states for A and B jointly can be revealed by considering the joint probabilities of obtaining outcomes of the measurements v A = {v a,k |k ∈ n A } and v B = {v b,k |k ∈ n B }: where n A = {1, 2, ..., n A } and n B = {1, 2, ..., n B }. The nonseparability of quantum states (sometimes called entanglement) 44,45 and the EPR steering 46-48 go beyond the predictions of the model of separable states and the local hidden state (LHS) model 46 , respectively. The basic concept behind Eq. (27) can be considered a hybrid of these models. Without loss of generality, we consider the case for two particles (N = 2). Compared to the states of particle A that are determined by shared variables μ such that P(v a ,v b ) = ∑ μ p μ P(v a |μ)P(v b |σ μ ) holds in the LHS model, the output states of χ C involving ρ μ are described by density matrices according to the prescribed realistic elements Ω µ v a k , in the χ C -nonclassical model; see Eq. (27). While these states in the χ C -nonclassical model and those in the separable-state model which predicts that P(v a ,v b ) = ∑ μ p μ P(v a |ρ μ )P(v b |σ μ ), are represented by density operators, A and B do share μ in the latter but A and B do not in the former. For these differences, the χ C -nonclassical correlation is stronger than nonseparability, but EPR steerability can be stronger than or equal to the χ C -nonclassical correlation. Here we illustrate such hierarchy by showing concrete quantum states of multipartite systems with the witness operators . In Eqs (17) and (18), the maximum similarity between ψ T and C ρ χ can be explicitly determined by which is equivalent to finding the best operational strategy for A and B such that a target state after the action on A is closest to the original. As n A = 1 (i.e., n B = N − 1), w C χ is obtained by evaluating the maximum overlap w max ( ) through SDP. For the three-qubit GHZ states 51 , we have w 0 683 C ∼ .
χ which is grater than the maximum value that can be attained for biseparable states 1/2 22 and equal to the threshold for genuinely multipartite EPR steering 50 . When taking W states as the target state, whereas the identified EPR steerability is stronger than the χ C -nonclassical correlation. For example, ∼ . χ w 0 717 C for N = 3 is grater than the threshold of 2/3 for genuine tripartite entanglement 22 but is weaker than the upper bound of (1 2 )/3 0 805 + ∼ . that can be attained by non-genuine tripartite EPR steering 50 .