Entanglement and entropy in multipartite systems: a useful approach

Quantum entanglement and quantum entropy are crucial concepts in the study of multipartite quantum systems. In this work we show how the notion of concurrence vector, re-expressed in a particularly useful form, provides new insights and computational tools for the analysis of both. In particular, using this approach for a general multipartite pure state, one can easily prove known relations in an easy way and to build up new relations between the concurrences associated with the different bipartitions. The approach is also useful to derive sufficient conditions for genuine entanglement in generic multipartite systems that are computable in polynomial time. From an entropy-of-entanglement perspective, the approach is powerful to prove properties of the Tsallis-$2$ entropy, such as the subadditivity, and to derive new ones, e.g. a modified version of the strong subadditivity which is always fulfilled; thanks to the purification theorem these results hold for any multipartite state, whether pure or mixed.


Introduction
Entanglement is a key notion in quantum mechanics and in the highly developing field of quantum information.For bipartite systems, H = H A ⊗ H B , there are rather simply criteria to detect and quantify the entanglement of pure states.In particular, if the state is entangled the reduced density matrix, ρ A = tr B ρ, corresponds to a mixed state, with tr ρ 2 A < 1. Accordingly, the value of the concurrence, C 2 ∝ 1 − tr ρ 2  A or the von Neumann entropy, S = − tr ρ A ln ρ A are standard measures of entanglement.
Things get more involved for multipartite Hilbert spaces, In that case there are 2 N −1 − 1 inequivalent bipartitions.If the system is entangled with respect to all of them, it is called a genuine multipartite entangled state.Apart from its theoretical importance, genuine entanglement is relevant for applications such as quantum cryptography [1], quantum computation [2], quantum teleportation [3], high-precision metrology [4], spin chains [5] and even biological systems [6].The structure of genuine entanglement is quite intricate and its complexity grows exponentially with the number of subsystems.Actually, the problem of classifying and quantifying the genuine entanglement of general multipartite systems is still open (see [7][8][9][10][11][12][13][14] for relevant work in this subject).
In this work we explore the power of the "concurrence vector" notion [15][16][17][18] to extract powerful information about the entanglement structure of a general multipartite pure state.In section 2 we review the concept of concurrence vector and introduce a simple expression which turns out to be quite useful to establish connections between the entanglement of different bipartitions (section 3).In section 4 we show that this formalism allows a simple derivation and an intuitive, geometrical, understanding of several triangular inequalities presented in the literature.Also, it allows to obtain new inequalities and generalize results that were only proven for a system of three qubits (the complete proof of the latter is given in the Appendix).In section 5 we explore the concurrence from an entropy-of-entanglement perspective, showing that the concurrence vector framework allows to easily prove the subadditivity condition for the Tsallis-2 entropy (directly related to the concurrence), as well as the violation of the strong subadditivity condition.In this sense we propose a modified version of the strong subadditivity condition which is always fulfilled, and derive other new relations.Thanks to the purification theorem these results are completely general for multipartite states, whether pure or mixed.In section 6 we exploit the concurrence vector approach to formulate very simple sufficient conditions for genuine entanglement, which are computable in polynomial time.Finally, the conclusions are presented in section 7.

I| Û I
, as well as alternative equivalent definitions, can be found in the literature.)The important point is that the state is biseparable if and only if C 2

The concurrence vector
Consider again the previous bipartition Clearly, the state (2) is separable as long as it can be written as so that Then, if we view a I Û I as a matrix of coefficients, for a separable state it has rank=1, which means that all its 2 × 2 minors are vanishing.Denoting by the minor corresponding to the rows I 1 , I 2 and columns Û I 1 , Û I 2 , the state is biseparable iff Here ) run over all possible values of I ( Û I).The latter expression is actually equivalent to the concurrence definition (3).To check this, notice that 1 4 where the last expression is just the sum of the principal minors of the reduced density matrix aa † = ρ A .On the other hand, by Cauchy-Binet theorem, 1 2 where λ I are the eigenvalues of ρ A and we have used λ I = 1.In summary, the concurrence definition (3) is equivalent to where the "concurrence vector" is the ordered list of all the minors of the matrix of coefficients a I Û I : Note that, denoting D = dim H, the concurrence vector has length D 2 , since a I Û I has D entries.The previous notion of concurrence vector has been already considered in the literature [15][16][17][18].

A useful expression
Let us express the concurrence vector (11) in a more useful way.First define the (dimension D 2 ) vector ⃗ A, with components for all possible values of the indices.Notice that A i This is simply a certain re-ordering of the components of ⃗ A. Note that P i are Hermitian linear operators satisfying P2 i = 1 and [P i , P j ] = 0. Let us now consider an "elementary bipartition", i.e. one between one of the initial parties and the rest, say H 1 ⊗ (H 2 ⊗ H 3 ⊗ • • •).Then the concurrence vector (11) (with Similarly, the concurrence vector (11) associated to the bipartition These expressions are straightforwardly extended for other bipartitions.In general, for a bipartition I| Û I, with I = i, j, ..., m (or any other subset of indices) the concurrence vector reads Mathematically, the elementary permutations, P i , P j , . . .are the generators of the group of all permutations of indices, {P I }, which is a commutative Z N 2 group.Note that A is obviously symmetric under the interchange of all indices.So, the group contains 2 N −1 inequivalent permutations, corresponding to the possible bipartitions of H.This includes the trivial bipartition H = H A ⊗ ∅, corresponding to the identity of the group.So, there are in fact 2 N −1 − 1 non-trivial permutations and bipartitions.For each one, the associated concurrence vector is given by (16).Note also that (1 − P I ) are "projectors", satisfying (1 Next we will exploit the notion of concurrence vector and its latter expression (16) to derive some direct consequences.

Connection between the entanglements of different bipartitions
The idea of concurrence vector allows to derive the entanglement of any of the 2 N −1 − 1 different bipartitions from the entanglement of the N elementary bipartitions.In other words, we can use the concurrence vectors of the latter as building blocks to construct any other concurrence.
For example, the concurrence of two parts with respect to the rest, say or equivalently More generically, the concurrence vector, ⃗ C I| Û I , of an arbitrary bipartition, say I = {i, j, ...., m, n}, can be expressed as This shows that the information contained in the N elementary concurrence vectors is enough to built any other concurrence vector, and thus the corresponding concurrence, C 2

Triangular inequalities
Consider a tripartition of the Hilbert space {i, j, Ù ij}.The corresponding concurrence vectors are related by Eq.( 17), which can be written as As these three vectors form a triangle, their lengths satisfy the triangular inequality.Since (and permutations of the three terms).This triangular inequality has been shown in previous literature, e.g. in refs [13,14].With the use of concurrence vectors it is a quite trivial result since the involved vectors form a triangle (with a permutation of components in one of them).This provides a transparent geometrical interpretation of the triangular inequality.We can also show very easily the (stronger) inequality for the squared of the concurrences [28], (and permutations).To see this notice that where the last term is negative semidefinite: Clearly, relations (21,22) hold for any tripartition of the Hilbert space, i.e. the indices i, j can be replaced by collective indices I, J when where I △ J = I ∪ J \ I ∩ J is the set of indices in either I or J, but not in both.Hence, analogous expressions to Eqs. (23,24) can be obtained, and in consequence the triangular inequalities (21,22) generalize to non-disjoint subsystems as On the other hand, in ref. [29] it was shown that, if the system consists of three qubits, the inequality ( 22) becomes an equality only when either C i| Û i or C j| Û j are vanishing; otherwise the " ≤ " sign in (22) becomes strict inequality " < ".In Appendix A we show that this interesting result holds for any tripartition, independently of the dimension of the subsystems.As a consequence, the area of the "concurrence triangle" built up with the squared concurrences of a tripartite system is a consistent measure of genuine entanglement.This entanglement measure was advocated in ref. [14,29] for systems of three qubits, but this result shows that it can be extended to arbitrary dimension of the subsystems.Besides, the same property holds for non-disjoint tripartitions, i.e.Eq.( 27) becomes an strict equality iff either C I| Û I or C J| Û J are vanishing.Finally, using the general expression (19) one can go beyond tripartitions and derive other inequalities.An obvious one comes from the fact that the vectors involved in (19) form a (nonnecessarily flat) polygon, so they satisfy the polygonal inequality (i.e. the extension of the triangular inequality to polygons).Since the permutations do not modify the norm of the various vectors involved, we conclude that as well as (and permutations).
A straightforward way to get new inequalities among concurrences is to consider the inequality Since (1±P ) = (1±P ) 2 /2, the l.h.s. of this equation is positive semidefinite, and it can be expressed as a combination of concurrences.

Subadditivity and strong subadditivity
From an entropy-of-entanglement perspective, the concurrence can be identified (up to a normalization factor) with the so-called Tsallis-2 (or "linear") entropy [30-32] where we have kept the notation previously introduced.Some of the well-known properties of the von Neumann entropy, S VN (ρ) = − tr (ρ log ρ), are also shared by S 2 (ρ) [33].In particular, given two subsystems, H A ⊗ H B , the S 2 −entropy satisfies the subadditivity conditions [34,35]: The previous second inequality actually corresponds to the positivity of an information-theory quantity known as mutual information, associated with the total amount of correlations in ρ AB [36] I (A : Now, these results stem trivially from the triangular inequalities (21,22) by simply considering a tripartition of the Hilbert space, H = H A ⊗ H B ⊗ H ÃB , and identifying the indices i, j with the subsystems H A ⊗ H B .We stress that the subadditivity condition holds for an arbitrary density matrix ρ AB .In this context, H ÃB can be considered as an extension of the Hilbert space H A ⊗ H B , so that the global state becomes pure, which is always possible thanks to the purification theorem.
In addition, we observe the following interesting property.Since the triangular inequality ( 22) becomes an equality iff either C i| Û i or C j| Û j are vanishing (see discussion in section 4 and a proof in Appendix A), it follows that the subadditivity condition ( 33) is saturated iff the entropy of one of the two subsystems is vanishing: Further relations concerning the Tsallis-2 entropy have been tested and, among them, special attention has been paid to the strong subadditivity [37].For a generic entropy, S(ρ), strong subadditivity reads where the last expression indicates that correlations are non-decreasing under the extension of one of the subsystems.It is well known that von Neumann entropy, S VN (ρ), satisfies the strong subadditivity condition [37].In contrast, generically the Tsallis-2 entropy, S 2 (ρ), does not satisfy it, as it was proven in ref. [38] by constructing several counterexamples.We can easily check this result using the concurrence-vector approach.Consider a Hilbert space partitioned as H = H A ⊗ H B ⊗ H C ⊗ H ȂBC , i.e. {i, j, k, îjk}.In terms of concurrences, condition (35) for S 2 (ρ) would read However, the l.h.s. of this relation is exactly the quantity: which is not negative semidefinite, showing that strong subadditivity (36) can be violated for particular choices of the state ⃗ A. E.g. for (1 This inequality is strict iff, moreover, C 2 j| Û j , C 2 k| Û k ̸ = 0 (see discussion after Eq.( 27) and Appendix A), thus leading to the violation of the strong subadditivity.An example of this setup is a state On the other hand, the quantity (37) becomes obviously negative semidefinite for so in this case the strong subadditivity is fulfilled.Actually, this instance is equivalent to the "ordinary" subadditivity (32) for the tripartition 37) vanishes, so the strong subadditivity holds as an equality.
To finish this section, let us construct a modified version of the strong subadditivity which does always hold for the Tsallis-2 entropy (31).Expanding the inequality (see Eq.( 30)) we get where the term within brackets is positive semidefinite, see Eq.( 22), and represents the departure from the strong-subadditivity condition (36).In terms of the Tsallis-2 entropy this softened version of strong subadditivity reads where the positivity (semi)definiteness of the term in brackets is equivalent to the ordinary subadditivity condition (32).In terms of mutual information this relation reads I (A : B) ≤ I (A : BC) + I (A : C).Since we can exchange B ↔ C, it can be expressed as On the other hand, we can yet construct an alternative subadditivity using the general relation (27).By considering a Hilbert space partitioned as H = H A ⊗ H B ⊗ H C ⊗ H ȂBC , and identifying I = {i, j}, J = {j, k}, we get In terms of entropies: which is a kind of triangular inequality for the S 2 −entropies.Analogously to the ordinary subadditivity, the fact that relation (43) becomes an equality iff either Concerning the mutual information, further inequalities can also be derived.In particular, applying Eq. ( 30) to the combination (1 − P i )(1 − P j )(1 − P k ) we obtain a non-negative condition over the tripartite information [39] S 2 (ρ where I (A : B : C) = I (A : B) + I (A : C) − I (A : BC).The sign of this quantity for the von Neumann entropy has been explored in the literature [39].For the case of holographic theories it has been proven to be non-positive (Monogamy of Mutual Information) [40], i.e. exactly the opposite behavior as the one found above for the Tsallis-2 entropy.The same relation in a similar context was obtained in refs.[41,42].

Sufficient conditions for genuine entanglement
Let us now use the idea of concurrence vector and its expression (16) to derive new sufficient conditions for genuine entanglement, which are computable in polynomial time.
For the sake of the clarity of the discussion, let us denote P 1 , P 2 , . . .P N the elementary permutations associated with the indices i, j, k, . . ., corresponding to the Hilbert spaces which is a linear combination of all the concurrences: where #I is the cardinality of I.The subscript N indicates that the (1 − P N ) factor has been left outside the product in Eq.( 47).In this way we avoid the duplication of concurrences, since ( A are equal and correspond to the same bipartition.Now suppose that there exists a permutation P Σ = P σ 1 P σ 2 • • • P σm with m odd, such that In other words, suppose that the state is separable along the bipartition Σ| Û Σ.Without loss of generality, we can suppose σ 1 , σ 2 , • • • σ m ≤ N − 1 (otherwise, consider the equivalent permutation P Û Σ instead of P Σ ).Then, Eq.(47) vanishes since the terms of the sum can be paired as: In consequence, if ⃗ V N ̸ = 0, this is a sufficient condition to exclude any separable "odd" bipartition.It is remarkable that this check of the ∼ 2 N −2 odd bipartitions is carried out by N − 1 simple operations on ⃗ A. The previous condition though says nothing about the possibility of (1 − P Σ ) ⃗ A = 0 where In a multipartite system, states with genuine entanglement are by definition those that are entangled under any possible bipartition.Using the approach described here for a general multipartite pure state, it is easy to obtain equalities that relate the concurrence vectors of different bipartitions, and thus about the genuine entanglement of the system.On the one hand, this provides interesting geometrical insights, e.g. the well-known triangular inequality for concurrences arises from the trivial fact that the corresponding concurrence vectors form a triangle.On the other hand, it allows to prove known relations in an easy way and to build up new ones.An example of the latter is the fact that the triangular inequality C 2 becomes an equality iff C A| Ù A = 0 or C B| Ù B = 0 ; a result that was proven in ref. [29] only for the three-qubit case.In consequence, the area of the triangle built up with the squared concurrences is a sound measurement of genuine entanglement for generic tripartite systems, as was proposed in refs.[14,29] for the three-qubit case.The present concurrence-vector approach is also useful to derive sufficient conditions for entanglement in generic multipartite systems that are computable in polynomial time.
From an entropy-of-entanglement perspective, the concurrence of a bipartite system in a pure state can be identified (up to a normalization factor) with the so-called Tsallis-2 entropy, S 2 (ρ A ), in one of the two subsystems.This allows to use the present approach to easily prove relations such as the subadditivity of S 2 .Likewise, it is a useful tool to prove new relations, e.g. a modified version of the strong subadditivity which is always fulfilled.Thanks to the purification theorem all these results hold for any multipartite state, whether pure or mixed.
All this shows that the concurrence vector approach is a useful tool for the study of the entanglement and the entropy of multipartite systems.
Eq.( 57) represents (3 3 ) 2 quadratic equations, 54 of which are non-equivalent.The complexity of Eq.(57) grows geometrically with the dimension of the Hilbert spaces of the three subsystems.However, the result for three qutrits is all we need to recursively extend statement to any dimension, as shown below.

Extension to arbitrary dimension
The above result for three qutrits can be re-stated in the following way.Given, by hypothesis, Eq. (57), if (1−P j ) ⃗ A ̸ = 0 for some component(s) of ⃗ A, then necessarily (1−P i ) ⃗ A = 0 for all components.(Of course the same holds for i ↔ j.) It is convenient at this point to define the restriction of the vector ⃗ A to a subset of components, i.e. to the subspace spanned by a subset of indices.E.g.
Note that the action of P i , P j or P k on this restricted vector keeps it inside the same subspace.In this notation, the previous result for qutrits reads as follows.Given a state such that (1 (for some component(s)), then To extend this result to higher dimensions, let us consider the multipartite Hilbert space with dim H n = d n ≥ 4. Our hypothesis is again Eq.(57) and Eq.( 63) in some subspace corresponding to the indices {(i 1 i 2 ) (j 1 j 2 ) (k 1 k 2 )}, i.e. some subspace of three qubits.Let us consider the addition of a fourth set of indices {i 4 j 4 k 4 }.Our goal is to show that Since we have shown that we can go from (63) to Eq.(64) for any new indices i 3 , j 3 , k 3 , it follows that Eq.( 66) is satisfied by many subsets of components, namely those corresponding to extend the three qubits of Eq.(63) to any three qutrits: ( Hence, the only components to check are those (or vice versa) and/or j = j 3 , j ′ = j 4 (or vice versa) and/or k = k 3 , k ′ = k 4 (or vice versa).Let A pqr;p ′ q ′ r ′ = (a pqr ) (a p ′ q ′ r ′ ) be one of such components.We have to show that (1 From Eq.( 63), there must be some component of ⃗ A the argument goes the same for any other component).Then the following chain of equalities follow from Eq.( 67): (a i 1 j 1 k 1 ) (a i 2 j 2 k 2 ) (a pqr ) (a p ′ q ′ r ′ ) (69) = (a pj 1 k 1 ) (a p ′ j 2 k 2 ) (a i 1 qr ) (a i 2 p ′ r ′ ) = (a i 2 j 1 k 1 ) (a i 1 j 2 k 2 ) (a p ′ qr ) (a pq ′ r ′ ) = (a i 1 j 1 k 1 ) (a i 2 j 2 k 2 ) (a p ′ qr ) (a pq ′ r ′ ) .
Cancelling the common factor in the first and latter expression, which by hypothesis is different from zero, we obtain (a pqr ) (a p ′ q ′ r ′ ) = (a p ′ qr ) (a pq ′ r ′ ) (70) as desired.Consequently, Eq.(66) holds.
In a similar fashion the result is extended to arbitrary dimensions of the three subsystems.

Extension to non-disjoint systems
When considering non-disjoint systems we have to deal with collective indices, I, J such that I ∩ J ̸ = ∅.Then, as was shown in section 4, the triangular inequality for the square concurrences reads where I △ J = I ∪ J \ I ∩ J is the set of indices in either I or J, but not in both.Following a procedure analogous to the one in previous subsections, it is straightforward to show that this relation becomes an equality iff either C I| Û I or C J| Û J are vanishing.More precisely, one starts by considering a tetrapartition {i, j, k, îjk}, identifying I = {i, k}, J = {j, k}.Then Eq.( 73 As for the disjoint case the strategy is to show "by brute force" (using Singular) that this is indeed the case for 4 qutrits, and then extend the result in a recursive way to higher dimensions.The procedure is completely analogous, but rather long and tedious, so we prefer to spare the reader.

From
Eqs.(26) the equality in this relation (74) is equivalent to the condition−2 ⃗ A † (1 − P i P k )(1 − P j P k ) ⃗ A = 0 ⇐⇒ (1 − P i P k )(1 − P j P k ) ⃗ A = 0 .(75)So our goal is to prove the following implication(1 − P i P k )(1 − P j P k ) ⃗ A = 0 =⇒ (1 − P i P k ) ⃗ A = 0 or (1 − P j P k ) ⃗ A = 0 .(76) 1 j 1 k 1 ...; i 2 j 2 k 2 ... are simply the coefficients of the state |ψ⟩ ⊗ |ψ⟩ ∈ H ⊗ H.If all the coefficients are real the entries of ⃗ A coincide with those of the density matrix, but this is not the general case.Under a permutation of the i−index, say P i , associated with the Hilbert space H 1 , ⃗ A changes as 1