Multiparty Quantum : An alternative definition

Asutosh Kumar∗ Harish-Chandra Research Institute, Chhatnag Road, Jhunsi, Allahabad 211 019, India The Institute of Mathematical Sciences, CIT Campus, Taramani, Chennai 600113, India Homi Bhabha National Institute, Anushaktinagar, Mumbai 400094, India

Mutual information is the reciprocal information that is common to or shared by two or more parties. Quantum mutual information for bipartite quantum systems is non-negative, and bears the interpretation of total correlation between the two subsystems. This may, however, no longer be true for three or more party quantum systems. In this paper we propose an alternative definition of multipartite information taking into account the shared information between two and more parties. It is non-negative, observes monotonicity under partial trace as well as completely positive maps, and equals multipartite information measure in literature for pure states. We then define multiparty quantum discord, and give some examples. Interestingly, we observe that quantum discord increases when measurement is performed on a large number of subsystems. Consequently, the symmetric quantum discord, which involves measurement on all parties, reveals the maximal quantumness. This raises question on the interpretation of measured mutual information as classical correlation.

Introduction.– Quantum correlations [1,2] are essen- [9]. It is defined as the amount of work (noise) that is tial ingredients in theory [3]. Vari- required to erase (destroy) the correlations completely. ous quantum correlations, different in nature and types, These properties (non-negativity, interpretation of total find huge applications in quantum information process- correlation) may, however, no longer be true for three ing tasks. Consequently, their characterization and quan- or more party quantum systems. The existing quantum tification is inevitable. Several non-classical correlation version of multipartite information in literature, due to measures have been proposed for bipartite quantum sys- Watanabe [8], is a straightforward generalization of bi- tems, and some of them have been extended to multi- partite QMI partite settings. Nonetheless, quantifying multipartite n quantum correlations in quantum physical systems re- X mains a challenging problem. Recently however, signifi- Ix(A1 : A2 : ··· : An) = S(ρAK ) − S(ρA1A2···An ) cant developments have been made towards this end in k=1

the form of multipartite global (symmetric) quantum dis- = S(ρA1A2···An k ρA1 ⊗ · · · ⊗ ρAn ) cord (GKD) [4], conditional entanglement of multipartite ≥ 0. (2) information (CEMI) [5], quantum correlations relativity (QCR) [6]. In Ref. [7], an operational interpretation of We refer to it as conventional quantum mutual in- GQD was given in terms of the partial state distribution formation (CQMI). It is the sum of the individual protocol. It was also shown that GQD nearly vanishes von Neumann entropies less the joint von Neu- for a multiparty that is approximately lo- mann entropy of a multipartite quantum system, cally recoverable after performing measurements on each ρA1A2···An . It is non-negative, and monotone non- of the subsystems. An important aspect to notice is that increasing under the local discarding of information (i.e., all these developments count on some multipartite infor- Ix(A1X1 : A2X2 : ··· : AnXn) ≥ Ix(A1 : A2 : ··· : An) mation measure (see below) [8]. Multipartite (mutual) for a multiparty quantum state ρA1X1A2X2···AnXn . information, the reciprocal information that is common However, unlike two-party quantum mutual information to or shared by two or more parties, has an authoritative Ix(A1 : A2), it does not have any operational interpreta- stand in the arena. Quantum mutual information (QMI), tion. whose definition is motivated by that of classical mutual arXiv:1504.07176v2 [quant-ph] 28 Jul 2017 information (CMI), is well defined for bipartite quantum In another approach, three-variable CMI [10] is defined systems. QMI of a bipartite quantum state ρAB is de- as fined as

I(ρAB) = S(ρA) + S(ρB) − S(ρAB) K(A : B : C) = K(A : B) − K(A, B|C), (3) = S(ρ k ρ ⊗ ρ ) ≥ 0, (1) AB A B where K(A : B) = H(A) − H(A|B) = H(A) + where S(ρ) = −tr(ρ log ρ) is the H(B) − H(A, B) = H(B) − H(B|A) is two-variable and S(ρ k σ) = tr(ρ log ρ − ρ log σ) is the quantum rela- CMI, K(A, B|C) = H(A|C) + H(B|C) − H(A, B|C) tive entropy. It is non-negative, and bears the interpre- is three-variable conditional mutual information, and tation of total correlation between the two subsystems H(.) is Shannon entropy. Though both K(A : B) and K(A, B|C) are non-negative, the three-variable CMI can be negative. Using the chain rule H(X,Y ) = H(X) + ∗ [email protected] H(Y |X), the following expressions of CMI are equiva- 2 lent: pure states. Moreover, we obtain its lower and upper bounds in terms of CQMI. K1(A : B : C) = [H(A) + H(B) + H(C)] −[H(A, B) + H(A, C) + H(B,C)] + H(A, B, C) (4) The rest of this paper is organised as follows: in the following section we provide an alternative def- K2(A : B : C) = H(A, B) − H(B|A) − H(A|B) inition of multiparty quantum mutual infromation, −H(A|C) − H(B|C) + H(A, B|C) (5) compute its value for some typical states, and prove K3(A : B : C) = [H(A) + H(B) + H(C)] its non-negativity and monotonicity. Then, in the next −[H(A, B) + H(A, C)] + H(A|B,C). (6) section, we discuss multiparty quantum discord and present a few illustrations. Surprisingly, we observe that The above definitions of CMI can be extended to the the symmetric quantum discord reveals the maximal quantum domain. They are obtained by replacing the quantumness. Finally, we conclude. random variables by density matrices and Shannon en- tropies by von Neumann entropies, with appropriate Quantum Mutual Information.– We propose an measurements in the quantum conditional entropies. alternative definition of multiparty quantum mutual Hence, information, via the Venn diagram approach, for an n-party quantum state ρ . As information can I(A : B : C) = [S(A) + S(B) + S(C)] A1A2···An be distributed or stored among m-parties, 2 ≤ m ≤ n, −[S(A, B) + S(A, C) + S(B,C)] + S(A, B, C) (7) our definition takes into account the shared information J1(A : B : C) = S(A, B) − SM(B|A) − SM(A|B) among m-parties, and not only the common information

−SM(A|C) − SM(B|C) + SM(A, B|C) (8) among all parties. This can be understood readily using a Venn diagram. J2(A : B : C) = [S(A) + S(B) + S(C)]

−[S(A, B) + S(A, C)] + SM(A|B,C) (9) where S(X) ≡ S(ρX ), and SM(X|Y ) = S(ρX|MY ) is the quantum obtained after some gener- alized measurement M has been performed on subsystem Y . It is asserted that the above quantum expressions are not equivalent as measurement assumes its role in the quantum conditional entropies. These QMIs have cer- tain drawbacks. First, surprisingly enough I(A : B : C) is identically zero for arbitrary three-party pure quantum states [10] implying that mutual information among the subsystems of three-party pure quantum systems is zero. This is not true in the case of bipartite QMI. Second, I(A : B : C) and other versions of QMI can be negative (a) (b) [10, 11]. How is this negative correlation useful for quan- tum information tasks? Though the existing definition of three-party QMI is argued to reveal the true nature of quantum correlations [10], the fact that QMI, being a measure of correlation, can assume negative value is challenging. This perplexing stance, handicapped with any operational interpretation of multipartite informa- tion, motivated us to propose an alternative definition of multiparty QMI. Our multipartite quantum mutual information for quan- tum state ρA1A2···An assumes the following form X (c) (d) I(A1 : A2 : ··· : An) = S(Xk1 Xk2 ··· Xkn−1 ) − (n − 1)S(A A ··· A ), (10) FIG. 1. (a) Two-variable, (b) three-variable, and (c & d) four- 1 2 n variable, Venn diagrams with possible intersecting regions. While (c) does not represent the true Venn diagram of four- where Xki ∈ {A1,A2, ··· ,An}. It takes into account the shared information among m-parties, 2 ≤ m ≤ n, and variables, (d) does. not only the common information among all parties. This is quite reasonable as information can be dis- tributed or stored among m-parties. We show that it is non-negative, and argue that it, by its very construction, Two-party QMI: From Fig.1(a), we see that the only manifests total correlation. Also it equals CQMI for way the subsystems A and B interact with each other is 3 via region ab, i.e., ab = A ∩ B = A + B − A ∪ B. In language as entropy language, this translates as I(A : B) = S(A) + S(B) − S(AB). This is usual bipartite QMI. I(A : B : C) = S(AB) + S(AC) + S(BC) − 2S(ABC) ⊗2  = S ρABC k ρAB ⊗ ρAC ⊗ ρBC . (11) It guarantees that I(A : B : C) is not identically zero for arbitrary three-party pure quantum systems. Four-party QMI: Fig.1(c) does not represent the true Venn diagram of four variables as pairwise interacting Three-party QMI: From Fig.1(b), the possible ways regions ad and bc are missing. The correct four-variable the subsystems A, B and C interact among each other are Venn diagram is represented in Fig.1(d). The total via region abc, which is common to all three, and regions four-party QMI is then defined as I(A : B : C : D) := ab, ac, bc, which are pairwise common. Taking just the [A ∪ B ∪ C ∪ D − (a + b + c + d)] which, again, after some region abc, i.e., abc = A∩B∩C = A+B+C−(A∪B+A∪ simple algebra, can be expressed as [12] C +B∪C)+A∪B∪C, this “common information” trans- X I(A : B : C : D) = S(X1X2X3) − 3S(ABCD) lates into Ic(A : B : C) := [abc] = [S(A)+S(B)+S(C)]− [S(AB)+S(AC)+S(BC)]+S(ABC). Note that in doing X1,X2,X3 so we have discarded pairwise interactions. However, a   ⊗3 O priori, there is no reason to throw them away. Moreover, = S ρABCD k ρX1X2X3  , (12) they can provide important information when examined {Xi} together. Then “two-party shared information” reads as Is2(A : B : C) := [ab + ac + bc]. Thus, the total three- where Xi ∈ {A, B, C, D}. We list in TableI the values party QMI is the sum of the common information and of common information (Ic) and QMI (I) of some typical the pairwise shared information: I(A : B : C) = Ic(A : states (see also Fig.2). B : C) + Is2(A : B : C) := [A ∪ B ∪ C − (a + b + c)]. After An n-party QMI can be analogously defined (see simple algebra, I(A : B : C) can be expressed in entropy Eq. (13))

I(A1 : A2 : ··· : An) := [A1 ∪ A2 ∪ · · · ∪ An − (a1 + a2 + ··· + an)] n X = S(ρ ) − (n − 1)S(ρ ) Ak A1A2···An k=1 n n ! X O = S ρ k ρ ⊗ ρ  − S ρ k ρ A1A2···An Ak Ak A1A2···An Ak k=1 k=1 n ! O = S ρ⊗n−1 k ρ , (13) A1A2···An Ak k=1

T 1 q where n-party common information is evaluated as entropy, Sq (ρ) = 1−q [tr(ρ ) − 1] [18], and the (smooth) Pn k+1 P Ic(A1 : A2 : ··· : An) = (−1) SAI with min-max entropies [19]. Both Renyi and Tsallis entropies k=1 {AIk } k A ≡ A A ··· A , S ≡ S(ρ ) and S ≡ S(ρ ) = reduce to von Neumann entropy in the limit q → 1. Ik i1 i2 ik X X Ai Ai

S(ρA1···Ai−1Ai+1···An ). Hence, QMI can be re-written as In subsequent theorems, we prove that I(A1 : A2 : ··· : I(A : A : ··· : A ) = Pn S − (n − 1)S . 1 2 n i=1 Ai A1A2···An An) is non-negative, and is monotonically non-increasing An n-party quantum mutual information is the sum of under partial trace and completely positive maps. It (n − 1)-party von Neumann entropies less (n − 1) times equals CQMI for pure states, and give its lower and the joint von Neumann entropy of an n-party quantum upper bounds in terms of CQMI. system. We argue here that multiparty QMI I(A1 : A2 : ··· : An), like I(A1 : A2), by very construct, Theorem 1. I(A1 : A2 : ··· : An) is non-negative. bears the interpretation of total correlation of a mul- Non-negativity of quantum mutual information tiparty quantum system. Further, we can obtain gen- I(A1 : A2 : ··· : An) follows directly from that eralized QMI by replacing the von Neumann entropy, of quantum relative entropy, S(ρ k σ) ≥ 0. Here S(ρ) = −tr(ρ log ρ), with generalized entropies like the we provide an alternative proof. To prove this, R 1 q Renyi entropy, Sq (ρ) = 1−q log[tr(ρ )] [17], the Tsallis we will extensively use a variant of strong sub- 4 additivity relation, SXYZ + SY ≤ SXY + SYZ , SXY ⇒ SX − SXY ≤ SY , we can, respectively, obtain Pn Pn Pn which states that conditioning reduces entropy, i.e., SAi ≤ SA¯ and SA¯ − nSA1A2···An ≤ Pni=1 i=1 i i=1 i SX|YZ ≤ SX|Y . The proof for n-party case follows as i=1 SAi . Therefore, I(A1 : A2 : ··· : An) = [20]: I(A : A : ··· : A ) = S + S + P S −(n−1)S ≥ Pn S −(n− 1 2 n 12···(n−1) 12···(n−2)n Ak1 Ak2 ···Akn−1 A1A2···An i=1 Ai ··· + S23···n − (n − 1)S12···n = S − S − 12···(n−1) 1|23···(n−1)n 1)SA1A2···An = Ix(A1 : A2 : ··· : An) − (n − 2)SA1A2···An . S2|13···(n−1)n − · · · − S(n−1)|12···(n−2)n ≥ S12···(n−1) − Again, I(A : A : ··· : A ) = P S − (n − 1 2 n Ak1 Ak2 ···Akn−1 S1|23···(n−1) − S2|13···(n−1) − · · · − S(n−1)|12···(n−2) = · · · ≥ n  1)S = P S ¯ − nS + S ≤ S − S − S − S = S − S − S ≥ A1A2···An i=1 Ai A1A2···An A1A2···An 123 1|23 2|13 3|12 12 1|23 2|13 Pn S + S = I (A : A : ··· : A ) + S − S − S = S + S − S ≥ 0. Hence, the i=1 Ai A1A2···An x 1 2 n 12 1|2 2|1 1 2 12 2S . Hence, the proof. theorem is proved. A1A2···An   The lower bound being dependent on n is weak. More- over, we find numerically that for an n-party quantum Theorem 2. I(A : A : ··· : A ) equals CQMI for 1 2 n system ρA A ···A (n = 3, 4), we have pure states. 1 2 n X For the pure quantum states, SA1A2···An = 0 and SA¯ = (n) (n) (n) (2) i 0 ≤ I − Ix ≤ Ix − I , (14) SAi . Hence, I(A1 : A2 : ··· : An) = Ix(A1 : A2 : ··· : An) using Eq. (13) and Eq. (2).  where I(k) is the k-party quantum mutual information, and the inequality is saturated for n = 3 (this can be Theorem 3. I(A1 : A2 : ··· : An) observes mono- shown analytically). tonicity under partial trace, and any completely positive map Φ. Theorem 5. I(A1 : A2 : ··· : An) is additive for These are direct consequences of the monotonicity product states. of quantum relative entropy [23] under partial trace, Additivity of quantum mutual information I(A1 : A2 : S(ρ k σ ) ≤ S(ρ k σ ), and any completely A A AX AX ··· : An) follows directly from that of the von Neumann positive map Φ, S (Φ(ρ) k Φ(σ)) ≤ S(ρ k σ).  entropy for product states, S(ρ⊗σ) = S(ρ)+S(σ). Here we consider a four-party state for the illustration (the Theorem 4. Ix − (n − 2)S12···n ≤ I ≤ Ix + 2S12···n. proof is identical irrespective of the number of parties

Using strong subadditivity entropic relation, SX + SY ≤ and the partition across which ρA1A2···An is product).

SXZ + SYZ , and Araki-Lieb inequality, SX − SY ≤ Let ρA1A2A3A4 = ρA1A2 ⊗ ρA3A4 . Then

I(A1 : A2 : A3 : A4) = S(A1A2A3) + S(A1A2A4) + S(A1A3A4) + S(A2A3A4) − 3S(A1A2A3A4)

= S(A1) + S(A2) + S(A3) + S(A4) + 2[S(A1A2) + S(A3A4)] − 3[S(A1A2) + S(A3A4)]

= [S(A1) + S(A2) − S(A1A2)] + [S(A3) + S(A4) − S(A3A4)]

= I(A1 : A2) + I(A3 : A4). (15)

   5 3.0 E ◆▲ E ◆▲ E ◆▲ ●  2.5 ●  ●  ▲ 2.5 ▲ ▲ 4 ◆ ■ ℐc ◆ ■ ℐc ◆ ■ ℐc ▲ ▲ 2.0 ▲ 2.0 ℐ 3 ℐ ▲ ◆ ◆ x ▲ ◆ ◆ ℐx ▲ ◆ ◆ x 1.5 1.5 ▲ ◆ ▲ ℐ ▲ ◆ ▲ ℐ ▲ ◆ 2 ▲ ℐ ▲ ◆ ◆ 1.0 ◆ ● 1.0 ▲ ● ● ▲ ● ● ▲ ◆ ● ● ● ▲ ◆ ● ● ▲ ◆ ● ▲ ◆ ● ● 0.5 ◆ ● ● 0.5 ▲ ◆ ● ● 1 ◆ ● ● ◆▲ ● ▲ ◆ ● ◆▲ ● ▲ ◆▲ ● ◆▲ ◆ ● ◆▲ ● ◆●■▲ ◆ ●■ ●■ ■ ■ p ◆●■▲ ●■ ●■ ■ ■ p ◆●■▲ ●■ ■ ■ p 0.2 0.4■ ■ 0.6■ ■ 0.8■ ■ 1.0 0.2 0.4■ ■ 0.6■ ■ 0.8■ ■ 1.0 0.2 ■ 0.4■ ■ 0.6■ ■ 0.8■ ■ 1.0 (a) (b) (c)

FIG. 2. Plots of logarithmic-negativity EN (A : BC)[21, 22], common information Ic(A : B : C), CQMI Ix(A : B : C), and I QMI I(A : B : C) against the white noise parameter p, of three-party state ρABC = p|ψihψ| + (1 − p) 8 for different |ψis: (a) GHZ state |GHZi = √1 (|000i + |111i)[13], (b) W state |W i = √1 (|001i + |010i + |100i)[14], and (c) totally antisymmetric 2 3 state |ψ i = √1 (|123i − |132i + |231i − |213i + |312i − |321i)[15]. In all the three cases, while I (A : B : C) vanishes at p = 0, 1 as 6 c and is negative at intermediate values, QMI I(A : B : C) vanishes at p = 0 only and is positive for other values. QMI is greater than or equal to Ix(A : B : C). We see that logarithmic-negativity, an entanglement measure, exceeds common information indicating that common information cannot be total correlation. 5

State Ic I pi = trAB[(Ai ⊗ IB)ρ(Ai ⊗ IB)]. I is the identity oper- |GHZ2i 2 2 ator on the Hilbert space of B. Hence, the conditional |GHZ i 0 3 P 3 entropy of ρAB is given by SM(B|A) = i piS(ρB|i). 1 |D3i 0 2.75489 Three-party quantum discord can then be defined, |ψasi 0 4.75489 when measurement is performed on subsystem A, sub- |GHZ4i 2 4 1 system AB and the whole system, as follows |D4i 0.490225 3.24511 2 |D4i 0.490225 4 DA(ρABC ) = I(ρABC ) − max I(ΦA(ρABC )) (16) |C4i -2 4 ΦA DAB(ρABC ) = I(ρABC ) − max I(ΦAB(ρABC )) (17) TABLE I. Values of common information (Ic) and QMI ΦAB (I) of |GHZ i = √1 (|0i⊗n + |1i⊗n)[13], |Dr i = n 2 n 1 P ⊗n−r ⊗r and q n P P[|0i |1i ][14], three-qutrit totally antisym- (r) metric state |ψ i = √1 (|123i − |132i + |231i − |213i + |312i − as 6 DABC (ρABC ) = I(ρABC ) − max I(ΦABC (ρABC ))(18) 1 ΦABC |321i)[15], and four- cluster state |C4i = 2 (|0000i + |0011i + |1100i − |1111i)[16]. While Ic can be negative, I is where I(σXYZ ) = I(X : Y : Z), non-negative. For pure states, I = Ix, from Theorem 2. P ΦA(ρABC ) = ΦA ρABC ΦA ,ΦAB(ρABC ) = P i i i i,j ΦAiBj ρABC ΦAiBj , and ΦABC (ρABC ) = P i,j,k ΦAiBj Ck ρABC ΦAiBj Ck with ΦAi = πi ⊗ I ⊗ I,

ΦAiBj = πi ⊗ πj ⊗ I, and ΦAiBj Ck = πi ⊗ πj ⊗ πk.   Eq. (18) is the symmetric quantum discord or global ◆ ◆ ● A ■ ◆ 1.4 ● A 1.0 ◆■ ■ ◆ quantum discord (GQD) [4]. Similarly, multiparty ◆■ 1.2 ◆ ■ ■ 0.8 ■ AB ◆ 1.0 ■ AB ■ ■ ◆ quantum discord can be defined. ◆ 0.8 ■ 0.6 ◆ ABC ■ ◆ ABC ◆■ ◆■ 0.6 ◆ 0.4 ■ ● ● ● Quantum discord, employing von Neumann entropy, of ◆■ ● ● ● ● 0.4 ◆■ ● ● ● ● ◆■ ● 0.2 ◆■ ● 0.2 ● three-party GHZ state and W state admixed with white ● ◆●■ ● ◆●■ ◆●■ ● p ◆●■ ◆●■ p 0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0 noise is shown in Fig.3. Quite unexpectedly, we find (a) (b) that DA ≤ DAB ≤ DABC , that is, quantumness increases when measurement is performed on a large number of   subsystems. This observation seems to be independent of 1.2 ◆ ◆ ● A ● A the definition of quantum mutual information. The sym- 1.0 ◆ ●■ 1.5 ◆ ■ ■ AB ■ ■ AB 0.8 ◆ ● ◆ ■ metric quantum discord, which requires measurement on ●■ 1.0 ■ 0.6 ◆ ABC ◆■ ◆ ABC ◆ ● ◆ ● ■ ● ●■ ◆■ ● all the parties, reveals the maximal quantumness. This 0.4 ◆●■ 0.5 ◆ ● ◆●■ ◆ ■ ● 0.2 ◆●■ ◆ ●■ ● contradicts the interpretation of measured mutual infor- ◆●■ ◆■ ●■ ◆●■ ◆●■ p ◆●■ ◆●■ ● p 0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0 mation as classical information because measuring more (c) (d) than one subsystem should yield more classical informa- tion and hence less quantum discord. FIG. 3. Plots of quantum discords based on alternative defi- Conclusion.– To sum up, we have proposed an nition of QMI (top panel) and conventional quantum discords alternative definition of quantum mutual information (bottom panel) DA, DAB and DABC against the white noise for multipartite setting. It is non-negative, and obeys parameter p, of three-party states ρABC = p|GHZihGHZ| + I I monotonicity under partial trace and any completely (1 − p) 8 (a & c), and ρABC = p|W ihW | + (1 − p) 8 (b & d). Contrary to our intuition, quantum discord increases when positive map. We argue that it manifests total correla- measurement is performed on larger number of subsystems. tion of a multiparty quantum system. We then employed The higher values of quantum discord suggest that W state this definition of quantum mutual information to define is more robust, as compared to GHZ state, against measure- multiparty quantum discord. Surprisingly, we found ment. that more quantumness can be harnessed by performing measurement on larger number of parties which is quite counter-intuitive. The symmetric quantum discord reveals the maximal quantumness. This suggests that Multiparty Quantum Discord.– In this section, we ex- measured mutual information should not be interpreted tend the definition of bipartite quantum discord [24, 25] as classical correlation. We believe that our work will to multipartite setting. Quantum discord for a bipartite provide further insights in understanding the nature of quantum state ρAB is defined as D(ρAB) = I(ρAB) − non-classical correlations. maxM J(ρAB), where I(ρAB) = I(A : B) = S(A) + S(B) − S(AB) and J(ρAB) = S(B) − SM(B|A). Here A.K. thanks Ujjwal Sen for useful suggestions and illu- measurement is performed on subsystem A with a rank- minating discussions, and acknowledges research fellow- one projection-valued measurement {Ai}, producing the ship from the Department of Atomic Energy, Govern- 1 states ρ = trA[(Ai ⊗IB)ρ(Ai ⊗IB)], with probability ment of India. B|i pi 6

[1] R. Horodecki, P. Horodecki, M. Horodecki, and K. Bells Theorem, Quantum Theory, and Conceptions of the Horodecki, Rev. Mod. Phys. 81, 865 (2009). Universe, edited by M. Kafatos (Kluwer Academic, Dor- [2] K. Modi, A. Brodutch, H. Cable, T. Patrek, and V. Ve- drecht, 1989). dral, Rev. Mod. Phys. 84, 1655 (2012). [14] R. Dicke, Phys. Rev. 93, 99 (1954). [3] M.A. Nielsen and I.L. Chuang, Quantum Computation [15] I. Jex, G. Alber, S. M. Barnett, and A. Delgado, Fortschr. and Quantum Information (Cambridge University Press, Phys. 51, 172 (2003). Cambridge, 2000). [16] H. J. Briegel and R. Raussendorf, Phy. Rev. Lett. 86, [4] M. Piani, P. Horodecki, and R. Horodecki, Phys. Rev. 910 (2001); R. Raussendorf and H. J. Briegel, Phy. Rev. Lett. 100, 090502 (2009); C. C. Rulli and M. S. Sarandy, Lett. 86, 5188 (2001). Phys. Rev. A 84, 042109 (2011). [17] A. Renyi, in Proc. Fourth Berkeley Symp. Math. Stat. [5] D. Yang, M. Horodecki, and Z. D. Wang, Phys. Rev. Prob. 1, 547 (1961). Lett. 101, 140501 (2008). [18] C. Tsallis, J. Stat. Phys. 52, 479 (1988). [6] M. Dugic, M. Arsenijevic, and J. Jeknic-Dugic, Sci [19] R. Renner, Security of quantum key distribution, PhD China-Phys Mech Astron 56(4), 732 (2013). thesis, ETH Zurich, 2005, arXiv:quant-ph/0512258; N. [7] M. M. Wilde, Proceedings of the Royal Society A 471, Dutta and R. Renner, IEEE Transactions on Information 20140941 (2015). Theory 55, 2807 (2009). [8] S. Watanabe, IBM Journal of Research and Development [20] The proof of non-negativity of quantum mutual informa- 4(1), 66 (1960). tion for three-party case goes as: I(A1 : A2 : A3) = S12 + [9] B. Groisman, S. Popescu, and A. Winter, Phys. Rev. A S13 +S23 −2S123 = S12 −S1|23 −S2|13 ≥ S12 −S1|2 −S2|1 = 72, 032317 (2005). S1 − S1|2 = S1 + S2 − S12 ≥ 0. [10] I. Chakrabarty, P. Agrawal, and A. K. Pati, Eur. Phys. [21] G. Vidal and R. F. Werner, Phys. Rev. A 65, 032314 J. D 65, 605 (2011). (2002). [11] M. Horodecki, J. Oppenheim, and A. Winter, Nature [22] M. B. Plenio, Phys. Rev. Lett. 95, 090503 (2005). 436, 673 (2005). [23] M. B. Ruskai, J. Math. Phys. 43, 4358 (2002). [12] This expression for quantum mutual information can also [24] L. Henderson and V. Vedral, J. Phys. A 34, 6899 (2001). be obtained using Fig.1(c) even though it does not rep- [25] H. Ollivier and W.H. Zurek, Phys. Rev. Lett. 88, 017901 resent the true Venn diagram of four variables. (2001). [13] D. M. Greenberger, M. A. Horne, and A. Zeilinger, in