![Arxiv:1506.08857V1 [Quant-Ph] 29 Jun 2015](http://data.docslib.org/img/05b519f346e381ff2db204019c119401-1.webp)
Absolutely Maximally Entangled states, combinatorial designs and multi-unitary matrices
Dardo Goyeneche National Quantum Information Center of Gda´nsk,81-824 Sopot, Poland and Faculty of Applied Physics and Mathematics, Technical University of Gda´nsk,80-233 Gda´nsk,Poland
Daniel Alsina Dept. Estructura i Constituents de la Mat`eria,Universitat de Barcelona, Spain.
Jos´e I. Latorre Dept. Estructura i Constituents de la Mat`eria,Universitat de Barcelona, Spain. and Center for Theoretical Physics, MIT, USA
Arnau Riera ICFO-Institut de Ciencies Fotoniques, Castelldefels (Barcelona), Spain
Karol Zyczkowski˙ Institute of Physics, Jagiellonian University, Krak´ow,Poland and Center for Theoretical Physics, Polish Academy of Sciences, Warsaw, Poland (Dated: June 29, 2015) Absolutely Maximally Entangled (AME) states are those multipartite quantum states that carry absolute maximum entanglement in all possible partitions. AME states are known to play a relevant role in multipartite teleportation, in quantum secret sharing and they provide the basis novel tensor networks related to holography. We present alternative constructions of AME states and show their link with combinatorial designs. We also analyze a key property of AME, namely their relation to tensors that can be understood as unitary transformations in every of its bi-partitions. We call this property multi-unitarity.
I. INTRODUCTION entropy S(ρ) = −Tr(ρ log ρ) , (1)
A complete characterization, classification and it is possible to show [1, 2] that the average entropy quantification of entanglement for quantum states re- of the reduced state σ = TrN/2|ψihψ| to N/2 qubits mains an unfinished long-term goal in Quantum Infor- reads: N mation theory. Nevertheless, a large number of rele- S(σ) = log 2 − c + (subleading terms), (2) vant results related to entanglement are well-known by 2 now. It is, for instance, known that generic condensed- where the constant c depends on the choice of the en- matter physical systems characterized by homoge- semble used to generate random states [3, 4]. For con- neous nearest neighbor interactions carry an amount venience we will consider natural logarithm along the of entropy bounded by an area law. This fact opens work. In the case the average is defined with respect arXiv:1506.08857v1 [quant-ph] 29 Jun 2015 the possibility of applying tensor network techniques to the unitarily invariant Haar measure on the space to describe the ground state of relevant physical sys- of unit vectors of size 2N the constant reads c = 1/2 tems, including such phenomena as quantum phase [1]. Although random states are almost maximally en- transitions. tangled, their entropy differs from the maximal value by a negative constant. One may then investigate However, there are many situations where entangle- when truly maximally entangled states appear, what ment entropy is maximal or nearly maximal, that is, properties they have and what they are useful for. A it scales as the volume of the quantum system. This relevant motivation for this study is the role that ab- is the case of generic time evolution of local Hamil- solute maximal entanglement may play in the context tonians, of evolution of a quantum computer address- of holography, as we shall discuss later on. ing a quantum Merlin–Arthur (QMA) problem, and The aim of the present paper is to extend the rela- of random states. The latter example is useful for our tion of absolutely maximally entangled (AME) states discussion, since it is known that the average entan- to a different branch of mathematics, so-called combi- glement of a random pure state |ψi of N qubits is close natorial designs. On one hand, new families of maxi- to maximal. Indeed, making use of the von Neumann mally entangled multipartite states are presented. On 2 the other hand, we demonstrate a direct link between quantum states that carry maximum entropy in all multipartite entanglement and combinatorial struc- their bi-partitions. It is a remarkable fact that the tures, including mutually orthogonal Latin squares existence of such states is not at all trivial and deepens and Latin cubes, and symmetric sudoku designs. Fur- into several branches of mathematics. Let us be more thermore, we introduce the concept of multiunitary precise and define an AME(N,d) state |ψi ∈ H, made matrices, which exists in power prime dimensions, with N qudits of local dimension d, H = (Cd)⊗N as a show their usage in constructing AME states and state such that all its reduced density matrices in any d ⊗ N present a small catalog of such matrices in small di- subspace A = (C ) 2 , H = A ⊗ A¯, carry maximal mensions. entropy This work is organized as follows. In Section II we present some important aspects of entanglement N S(ρA) = log d ∀A . (3) theory, AME states and its link with holography. In 2 Section III we review the current state of the art of This is tantamount to asking the reduced density ma- AME states for multipartite systems. In Section IV trices to k qudits to be proportional to the identity we consider the particular case of AME states hav- ing minimal support and its relation to classical codes 1 N ρ = I k ∀k ≤ . (4) and orthogonal arrays of index unity. A link between k dk d 2 AME states, mutually orthogonal Latin squares and Latin cubes and hypercubes is explored in Section V. Let us note the fact that a k–uniform state is also In Section VI we introduce the concept of multiu- k0–uniform for any 0 < k0 < k. nitary matrices and demonstrate that they are one- There is an obstruction for a state to reach maxi- to-one connected with AME states. Additionally, we mal entanglement in all partitions due to the concept provide some constructions of multi-unitary matrices of monogamy of entanglement [15, 16]. Every local de- and its associated AME states. In Section VII we re- gree of freedom that tries to get maximally entangled sume and discuss the most important results obtained with another one is, then, forced to disentangle from in our work and conclude. In Appendix A we dis- any third party. Therefore, entanglement can be seen cuss the non-existence of AME states of four qubits as a resource to be shared with other parties. If two and present the most entangled known states. In Ap- local degrees of freedom get largely entangled among pendix B we detailedly explain the simplest case of themselves, then they are less able to be entangled 2-unitary matrices of order D = d2, which are asso- with the rest of the system. But this rule is not al- ciated to AME states of 4 qudits. some examples of ways fulfilled. There are cases where the values of the multiunitary matrices are presented. In Appendix C local dimensions d and the total number of qudits N we establish a relationship between AME states and are such that AME states exist. For a given N, there a special class of Sudoku. Finally, in Appendix D we is always a large enough d for which there exists an present a mini-catalog of multi-unitary matrices ex- AME state [13]. However, the lowest value of d such isting in low dimensions. that an AME state exists is not known in general. Let us mention that AME states are useful and nec- essary to accomplish certain classes of multipartite protocols. In particular, in Ref. [13], it was shown II. AME STATES: DEFINITION AND BASIC PROPERTIES that AME states are needed to implement two dif- ferent categories of protocols. First, they are needed to achieve perfect multipartite teleportation. Second, A. Definition they provide the resource needed for quantum secret sharing. These connections hint at further relations A lot of attention was recently paid to identify en- between AME states and different branches of Mathe- tangled states of N–party systems, such that tracing matics. For instance, AME states are related to Reed- out arbitrary N − k subsystems, the remaining k sub- Solomon codes [17]; Also, AME states (and k–uniform systems are maximally mixed [5–10]. Such states are states in general) are deeply linked to error correction often called k–uniform [11, 12] and by construction codes [11]. the integer number k cannot exceed N/2. In this pa- There is yet another surprising connection between per, we shall focus on the extremal case, k = bN/2c AME states and holography [18, 19]. It can be seen (we put the floor function bc to include cases of N even that AME states provide the basis for a tensor network and odd) , and analyze properties of states called ab- structure that distributes entanglement in a most ef- solutely maximally entangled (AME) – see [13, 14]. ficient and isotropic way. This tensor network can Also, such states were previously known as perfect be proven to deliver holographic codes, that may be maximally multipartite entangled states [29]. useful as quantum memories and as microscopic mod- The definition of AME states corresponds to those els for quantum gravity. A key property for these 3 new developments is related to the properties of multi- in the system is far less than the maximum allowed. unitarity that will be explored in Sect. VI B. Degrees of freedom in the bulk will not carry maximal correlations, neither the von Neumann entropy of any sub-part of the system will scale as its volume. B. Local Unitary equivalence To gain insight in quantum holography, it is natural to investigate the bulk/boundary correspondence of Entanglement is invariant under choices of local ba- the operator content of the theory. On the other hand, sis. It is then natural to introduce the concept of Local Quantum Information brings a new point of view on Unitary (LU) equivalence among AME states. Two this issue, since it focuses on the properties of states quantum states |Φi and |Ψi are called LU-equivalent rather than on the dynamics that generates it. In if there exist N local unitary matrices, U1,...UN such this novel context, we may ask what is the structure that of quantum states that display holographic properties. That is, we aim at finding which is the detailed entan- |Φi = U1 ⊗ U2 ⊗ ... ⊗ UN |Ψi . (5) glement scaffolding that guarantees that information flows from the boundary to the bulk of a system in a If |Ψi is a maximally entangled state any other state perfect way. LU-equivalent to it is also maximally entangled. In A concept separate from holography turns out to this respect, we will define AME(N, d) as the set of be very useful to address the analysis of holography all AME states in the Hilbert space H(N, d) and de- from this new Quantum Information perspective, that note their elements by a Greek letter, e. g. |Ω4,3i ∈ is Tensor Networks of the kind of matrix product AME(4, 3) is an AME of four qutrits. states (MPS), projected entangled pair states (PEPS) The LU transformations introduce equivalence and multi-scale entanglement renormalization ansatz classes of states. A question arises naturally about (MERA) Indeed, Tensor Networks provide a frame to which state should be chosen as the representative of analyze how correlations get distributed in quantum the class, that will be denoted as canonical form of states, and thus to understand holography at the level an AME state. It is possible to argue in two differ- of quantum states. Each connection among ancillary ent directions. On one hand, we may consider that a indices quantifies the amount of entanglement which natural representative may carry all the elements of links parts of the system. Holography must necessar- the computational basis. It would then be necessary ily relay on some very peculiar entanglement struc- to establish theorems and a criterion to fix the coeffi- ture. cients. On the other hand, an alternative possibility A first attempt to understand the basic property is to choose the element of the class with a minimal behind holography of quantum states was presented support on the computational basis. Results in both in [18]. There, it was proposed to create a quan- directions are presented in Sec. IV. tum state on a triangular lattice based on a tensor It is not known in general how many different LU network that uses as ancillary states an absolutely classes there are in the set AME(N, d) for every N and maximally entangled state. To be precise, the state d. This question can be tackled by the construction |Ωi ∈ AME(4, 3) (see Eq.(11) was defined on tetra- of LU-invariants. A few examples are at hand for few hedrons, in such a way that the vertices in its basis qubits. For three qubits, it is known how to obtain connect the tensor network and the tip of the tetra- a canonical form of any state using LU and that all hedron corresponds to a physical index. states are classified by 5 invariants [20], only one of In a subsequent work [19], another construction was them is genuinely multipartite, the tangle. For four based on the 5-qubit |Υ5,2i ∈ AME(5, 2) state 8. qubits, there are ways to construct a canonical form Again, the fact that the internal construction of the and to find the hyperdeterminant, as well. Yet, it state is based on isometries is at the heart of the holo- is unknown how to proceed to larger local dimensions graphic property. and number of parties. It is arguable that the subset of There are two obvious observations on the surpris- AME states is characterized by several LU-invariants, ing relation between AME states and holography. The probably related to distinct physical tasks. In such a first is related to the natural link between AME states case, there would be different AME states not related and error correction codes. It is arguable, then, that to each other by LU. the essence of holography is error correction, which limits the amount of information in the system. The second is that the very property responsible for holog- C. AME and holography raphy is multi-unitarity, which is analyzed in depth in section VI. It is further arguable that multi-unitarity Quantum holography amounts to the fact that the is the building block of symmetries, since the sense information content of a quantum system is that of of direction is lost and can be defined at will. Those its boundary. It follows that the information present ideas deserve a much deeper analysis. 4
D. Related definitions 2. Maximally Multipartite Entangled states
In Ref. [7], Maximally Multipartite Entangled (MME) states are introduced as those states that 1. Maximally Entangled sets maximize the average entanglement (measured in terms of purity) where the average is taken over all the balanced bipartitions i. e. |A| = bN/2c. More In the context of convertibility of states via Local specifically, the MME states are defined as the mini- Operations and Classical Communication (LOCC), mizers of the potential of multipartite entanglement multipartite entanglement is significantly different N X from the bipartite one. While in the bipartite case, π = π , (6) ME bN/2c A there is a single maximally entangled state (up to lo- |A|=bN/2c cal unitaries) that can be transformed into any other state by LOCC (and cannot be obtained from any 2 where πA = Tr(ρA) is the purity of the partition other), in the multipartite scenario this is no longer A. Note that the above potential is bounded by bN/2c true. In Ref. [21], the notion of the Maximally Entan- 1/d ≤ πME ≤ 1 and its lower bound is only gled (ME) set of N-partite states is introduced as the saturated by AME states. set of states from which any state outside of it can be By minimizing the multipartite entanglement po- obtained via LOCC from one of the states within the tential, explicit examples of AME states of 5 and 6 set and no state in the set can be obtained from any qubits are presented in Ref. [7]. It is remarkable that other state via LOCC. Note that this notion of max- even for a relatively small number of qubits (N ≥ 7), imal entanglement is strictly weaker than the AME, such minimization problem has a landscape of the pa- in the sense that most of (or all) states in the ME set rameter space with a large number of local minima, will not be an AME state, but any AME state will be what implies a very slow convergence. The reason for in its corresponding ME set. In Ref. [21], the ME set that is frustration. The condition that purity satu- is characterized for the case of three and four qubits. rates its minimum can be satisfied for some but not It is interesting to point out that, unlike the 3-qubit for all the bipartitions (see [8] for details). In this case, deterministic LOCC transformations are almost respect, in [24], the minimization of the multipartite never possible among fully entangled 4-partite states. entanglement potential is mapped into a classical sta- As a consequence of this, while the ME set is of mea- tistical mechanics problem. The multipartite entan- sure zero for 3-qubit states, almost all states are in glement potential is seen there as a Hamiltonian which the 4-qubit ME set. This suggests the following pic- is minimized by simulated annealing techniques. ture; given a fixed local dimension and for an increas- ing number of parties, the AME states become more an more rare at the same time that more and more III. EXAMPLES OF AME STATES states need to be included in the ME set. In other words, while maximally entangled states defined from A. Qubit AMEs an operational point of view become typical when the number of parties increases, AME states are exotic. Let us consider states made out of qubits, that is the dimension of the local Hilbert space is d = 2. The issue of inter-convertibility between quantum The simplest cases of AME states are any of the Bell states has been addressed under a larger set of op- states. There is a unique partition of two qubits and erations than the LOCC: the Stochastic Local Oper- it is possible to entangle both parties maximally. It ations and Classical Communication (SLOCC). The is easy to argue that there is a unique quantity that SLOCC identify states that can be interconverted by describes the amount of entanglement in the system. LOCC in a non-deterministic way, but with a non- This can be chosen to be the first eigenvalue of the zero probability of success. In this respect, a sys- Schmidt decomposition of the state, or some other tematic classification of multipartite entanglement in quantity derived from it as the von Neumann entropy. terms of equivalence classes of states under SLOCC is All two particle states, whatever their local dimension, presented in Ref. [51]. In particular, it is shown that can be entangled maximally. Those states are of no the SLOCC equivalence class of multipartite states is interest for our present discussion which is genuinely characterized by ratios of homogeneous polynomials centered in multipartite entanglement. that are invariant under local action of the special In the case of 3 qubits the well-known GHZ state linear group. This work generalizes for an arbitrary [25] is an AME, but it is known that there is no 4- number of partitions and finite local dimension the qubit AME [6]. The amount of degrees of freedom in complete classification made for 4-qubits in Ref. [22]. the definition of the state is insufficient to fulfill all the 5 constraints coming from the requirement of maximum where all qutrit indices are computed mod(3). It is entanglement. easy to verify that all the reduced density matrices to 1 For 5 and 6 qubits, there are AME states. In par- two qutrits are equal to ρ = 9 I9, so that this state ticular, a 5-qubit state |Υ5,2i ∈ AME(5, 2) can be carries entropy S = 2 log 3 for every one of its bi- defined by the coefficients of the superposition of ba- partitions. sis states that form it: The state |Ω4,3i can be viewed as a map of a two- qutrit product basis into a second one. That is 25−1 1 X (Υ) |Υ5,2i = ci |ii, (7) X X 25/2 |Ω4,3i = |uii|vii = |uiiUij|uji (12) i=0 i=0,8 i,j=0,8 where we used the usual shorthand notation for the el- ements in the computational basis and the coefficients where {|uii} and {|vii} are product basis for two have the same modulus and signs given by [7] qutrits, and |vii = Uij|uji. The matrix Uij is not only unitary (as it must as a consequence of multiu- c(Υ) = {1, 1, 1, 1, 1, −1, −1, 1, 1, −1, −1, nitarity) but also a bijective map between the sets of words of length 2 over an alphabet Z = {0, 1, 2}. 1, 1, 1, 1, 1, 1, 1, −1, −1, 1, −1, 1, 3 In other words, the entries of Uij are 0 with a single −1, −1, 1, −1, 1, −1, −1, 1, 1}. (8) 1 per row/column. This property remains the same whatever partition is analysed, though the unitary will A state AME(5,2) found in [10] was proven to be use- vary. ful for a number of multipartite tasks in Ref. [26]. It A second feature of the state |Ω4,3i is that the Ham- can also be found as the superposition of a perfect er- ming distance between any pair of elements in the ror correcting code as presented in Ref. [27] and also state is three, DH = 3. As all the sequences in |Ω4,3i from orthogonal arrays [12]. differ in 3 elements, any single qutrit error can be cor- For the sake of completeness, let us also provide an rected. Hence, maximal entanglement is related to absolutely maximally entangled 6-qubit state error correction codes. Both of these properties are related to the fact that |Ω4,3i is an AME of minimal 26−1 1 X (Ξ) support, a concept we are going to explore in Sect. |Ξ6,2i = √ 6 ci |ii, (9) 2 IV. i=0 It is also natural to expect AME states to accom- with modate easily to some magic-square like relations. A simple example goes as follows. Write the coefficients (Ξ) c = {−1, −1, −1, +1, −1, 1, 1, 1, of |Ω4,3i as a matrix of row i and column i, giving the −1, −1, −1, 1, 1, −1, −1, −1, composed value a of the remaining two qutrits from 0 to 8, that is a = 3(i + j)mod(3) + (i + 2j)mod(3). −1, −1, 1, −1, −1, 1, −1, −1, The square reads 1, 1, −1, 1, −1, 1, −1, −1, −1, 1, −1, −1, −1, −1, 1, −1, 0 5 7 4 6 2 1, −1, 1, 1, −1, −1, 1, −1, 8 1 3 1, −1, −1, −1, 1, 1, 1, −1, 1, −1, −1, −1, −1, −1, −1, 1}. (10) where all rows and columns add up to 12. The same properties are maintain if we interchange the indices The case of 7 qubits remains unsolved but numerical in the state. These kind of combinatorial designs are evidence hints at the impossibility of finding such an going to be explored in sect. V. AME state [28]. For 8 qubits or more, there are no From an experimental point of view, |Ω4,3i can be AME states [11]. created defining a quantum circuit that generates it. Such a circuit makes use of the following two gates:
B. A central example: AME(4,3) 1 Fourier F3|0i = √ (|0i + |1i + |2i) , 3 The first non-trivial example of AME for larger local C3 − adder UC−adder|ii|ji = |ii|(i + j)mod3i. dimensions corresponds to a state made of 4 qutrits, (13) AME(4,3). Its explicit construction is This gate C generalizes the CNOT gate for qubits 1 X 3 |Ω i = |ii|ji|i + ji|i + 2ji , (11) and it is represented in the circuit using the usual 4,3 3 i,j,=0,1,2 symbol for CNOT with the subscript 3. The |Ω4,3i 6
the case of a certain bipartition j of the set Sn. The (n) real coefficients cj are free as long as they satisfy Eq. (15) and the normalization condition. With this expressions all AMEs of a small number of qubits are classified. It is also worth noting that AME states with√ maximal support and uniform amplitudes N rk = 1/ d ∀k are possible to define (by just setting all cj to 0 in Eq. (16)) [29].
FIG. 1: Quantum circuit required to generate the state IV. AME STATES OF MINIMAL SUPPORT AME Ω4,3 (4 qutrits) based on the Fourier gate F3 and AND CLASSICAL CODES control-adders gate C3 − adder . In Ref. [30], a subclass of AME(N, d) states is shown to be constructed by means of classical maximum- state can be constructed as a sequence of a Fourier F 3 distance separable (MDS) codes. In this section we and C -adders following the circuit depicted in Fig.1 3 show that such a subclass corresponds to the set of acting on the initial state |0000i of 4 qutrits. AME states of minimal support and exploit these ideas to get conditions for their existence. C. General expression for AME states
The states AME(5,2) and AME(6,2) have the maxi- A. Support of AME states mal number of terms whereas AME(4,3) has the mini- mal possible number. Therefore the following question From the explicit examples we have presented in arises: is there a general expression for AME(N, d) Sec. III, AME states appear to need a different having the maximal number of dN terms? Facchi par- numbers of elements to be written. For instance, tially solved this issue for qubits in [29]. It can be AME(4,3) is made of the superposition of 9 states, all shown that a state expressed as weighted with the same coefficient. Yet, AME(6,2) as written using the 64 basis states with coefficients 2n−1 X either 1 or -1. |Ψi = zk|ki, zk = rkξk, (14) Let us define the support of a state |ψi as the num- k=0 ber of non-zero coefficients when |ψi is written in the where the |ki span the whole computational basis and computational basis. The support of a class is defined r and ξ are respectively the modulus and the phase as the support of the state inside the class with mini- k k mal support. Note that the support of a class defines of the complex coefficients zk, is an AME if it satisfies the following equations: in turn another equivalence class. Two states are sup- port equivalent if they belong to a LU classes with equal support. X In this sense, it is interesting to point out that the rl L mrl0 L mξl L mξl0 L m = 0, (15) m state AME(6,2), defined in Eq. (9) with the maximal support of 26 states, is LU equivalent to a state of sup- where l, l0 and m stand for both parts of a certain port 16 by applying some Hadamard gates on its basis. balanced bipartition (of bN/2c particles). It can be proven that 8 is the minimal support that The general form of the squared modulus of the such state could have but it is not attainable. Also in coefficients will be: [10] a state AME(5,2) is built using 8 elements, while the theoretical minimum would be 4. We may won- der why the naive minimum possible number of 2bN/2c 2 −N X X (n) Y (j,n) rk = 2 + cj (2kh − 1), (16) elements is not always attained. The answer to this N j∈Sn 1≤h≤n 2
dbN/2c 1 X d ≥ bN/2c + 1 . (22) |Ψi = r eiθk |x [A]i|x [A¯]i . (19) dbN/2c k k k k=1 We can prove it as follows: let us try to construct an AME state by building an MDS code. Due to the In order for |Ψi to be an AME, the reduced density relabeling freedom, the first d + 1 words of the code matrix of any bipartition A = {a1, . . . , abN/2c} needs can be chosen as to be the completely mixed state. It is easy to see that this has the following implications: 0 ...... 00 0 ...... 0, 0 ...... 01 1 ...... 1, d + 1 1. The modulus rk = 1 for all k. . . . code . . . words 2. The phases θk are arbitrary. 0 ... 0(d − 1) (d − 1) ... (d − 1), 3. For any balanced bipartition A, with |A| = 0 ...... 10 xd[bN/2c + 1] . . . xd[N], bN/2c, two words xi, xj ∈ XΨ have subwords where the letters xd[i], for bN/2c+1 ≤ i ≤ N, are still xi[A] = xj[A] if and only if i = j. Equivalently, bN/2c unknown and every word is written in two subwords XΨ[A] = Zd . of lengths bN/2c and b(N + 1)/2c respectively. Condition 3 implies that any pair of different words Note that: (i) none of the unknown letters xd[i] x , x ∈ X have Hamming distance can be 0 in order for the word xd to have Hamming i j Ψ distance bN/2c + 1 with the first word 0 ... 00 ... 0.
dH (xi, xj) ≥ bN/2c + 1, (20) (ii) None of the unknown letters xd[i] can be repeated in order for xd to have Hamming distance bN/2c + 1 where the Hamming distance between two codewords with the other d − 1 words. Therefore, if b(N + 1)/2c is defined as the number of positions in which they variables, xd[i], must take b(N +1)/2c different values, 8 and all of them must be different from 0, it is necessary shown that a family of valid generators is given by to extend the alphabet, forcing that d ≥ bN/2c + 1. Interestingly, Eq.(22) forbids the existence of 1 1 ... 1 0 g0 g1 . . . gd−1 0 AME(N,2) states having minimal support for N > G = , (24) ...... 3. However, this does not represent a proof that k k k AME(4,2) do not exist. Also, this inequality is sat- g0 g1 . . . gd−1 1 urated for the states AME(4,3) and AME(6,4). The where d ∈ Prime, N = d + 1, and k = N/2. existence of the cases AME(8,5) and AME(8,6) is still The case of AME(4,3) can be re-obtained using open whereas the states AME(8,7) and AME(8,8) are known. 1 1 1 0 G = (25) 0 1 2 1
Another concrete example corresponds to g0 = 1, C. A less trivial example: AME(6,4) g1 = 1,..., g6 = 6 that corresponds to AME(8,7), with N = 8 d = 7-dits. Then a total of 74 codewords are An application of the above connection between obtained that differ by a minimum Hamming distance AME states and MDS codes is the construction of dH = 5. AME states by exploring the set of all words and se- As the above construction can only be accomplished lecting those which differ in at least bN/2c + 1 ele- for d being prime and N = d+1, it is interesting to ad- ments. For the case of the AME(6,4), such a search dress the question of whether given some AME(N, d) gives a state with a equal superposition of the follow- it is possible to construct another AME(N 0, d0). In ing entries: this context, the following lemma can be useful which allows to construct an AME(N 0, d) for any N 0 < N: { 000000, 001111, 002222, 003333, 010123, 011032, if there exists an AME state with minimal support in H(N, d) with N being even, then there exist other 012301, 013210, 020231, 021320, 022013, 023102, AME states with minimal support in the Hilbert 030312, 031203, 032130, 033021, 100132, 101023, spaces H(N 0, d) for any N 0 ≤ N. 102310, 103201, 110011, 111100, 112233, 113322, To prove this last statement, we will consider sepa- 120303, 121212, 122121, 123030, 130220, 131331, rately the transitions N → N − 1 and N − 1 → N − 2. Transition N → N − 1: The existence of an AME 132002, 133113, 200213, 201302, 202031, 203120, state with minimal support implies the existence of a 210330, 211221, 212112, 213003, 220022, 221133, code of dN/2 words of length N with Hamming dis- 222200, 223311, 230101, 231010, 232323, 233232, tance dH ≥ N/2 + 1. Let us order the words in the 300321, 301230, 302103, 303012, 310202, 311313, code in increasing order and take the subset of the first dN/2−1 words which start with 0. Note that by 312020, 313131, 320110, 321001, 322332, 323223, suppressing such a 0, we get a code of dN/2−1 words 330033, 331122, 332211, 333300} . (23) of length N − 1 with Hammings distance N/2 + 1, forming an AME. Such an AME state carries the minimum possible sup- Transition N −1 → N −2: From the previous steps, port. we are left with a code of dN/2−1 words of length N − 1 and Hamming distance N/2 + 1. Note that by suppressing an arbitrary letter from all the words of the code, one is left with a set of d(N−2)/2 words of D. Construction of AMEs with minimal support length N − 2 with Hamming distance N/2 + 1 − 1 = (N − 2)/2 + 1, which is a MDS code. Finding AME states by exploring the set of all By iterating the previous procedure, we obtain words is highly inefficient and becomes in practice un- MDS codes (and AME states of minimal support) feasible from a relatively small number of parties. In for any 2 ≤ N 0 ≤ N. Equivalently, MDS codes this context, the Reed-Solomon codes [17] can be a are constructed from considering orthogonal arrays of useful tool to produce systematic construction of MDS strength one [12]. codes and equivalently AME states. Let us review here the particular case of d prime and N = d + 1. Let us refer to the elements of the E. Non-minimal support AMEs and perfect quantum error correcting codes superposition in the quantum states as words xi and the word of a half-partition as ui. The code words are obtained using the action of a generator G, xi = ui ·G. AME states are deeply related to classical error cor- The problem is then reduced to fixing G. It can be rection codes and compression [11, 27, 32]. This is 9 somewhat intuitive since maximal entropy is related been recently found [12]. Furthermore, they are a to maximally mixed subsets. The measure of any local fundamental tool in optimal design of experiments degree of freedom delivers an output which is com- [36, 37]. The existence of some combinatorial de- pletely random. This is, in turn, the basic element signs can be extremely difficult to prove. For example to correct errors. Hence, a relation between the el- the existence of Hadamard matrices in every dimen- ements superposed to form an AME state and error sion multiple of four (i.e., the Hadamard conjecture) is corrections codes is expected. open since 1893 [38] and it represents one of the most Le us illustrate the connection of an AME(5,2) state important open problems in Combinatorics. with the well-known 5-qubit code [27]. It is easy to A. Relation to mutually orthogonal Latin see that by applying some Hadamard gates on local squares qubits, the AME(5,2) state, defined through the 32 coefficient given in Eq. 8, is LU-equivalent to a state Let us consider the explicit expression of the with fewer non-zero coefficients. Actually, a represen- AME(4,3) state defined in Eq.(11): tative of the same AME(5,2) class is found to have only 8 coefficients. That state corresponds to a super- 1 position of the two logical states in the error correcting |Ω4,3i = (|0000i + |0112i + |0221i 3 codes found in [27] +|1011i + |1120i + |1202i 1 +|2022i + |2101i + |2210i) . (27) |Ω5,2i = √ (|0iL + |1iL) , (26) 2 Here, the third and fourth symbols appearing in every where the logical qubits are defined as term of the state can be arranged into 2 Greco-Latin squares of size three: 1 |0iL = (|00000i + |00011i + |01100i − |01111i) 2 Aα Bγ Cβ A♠ K♣ Q♦ 1 |1iL = (|11010i + |11001i + |10110i − |10101i) Bβ Cα Aγ = K♦ Q♠ A♣ , (28) 2 Cγ Aβ Bα Q♣ A♦ K♠ Note the fact that the coefficients carry both plus and minus signs as is the case in the non-minimal support where every symbol of the sets {A, α}, {B, β} and AMEs. {C, γ}, is associated to 0, 1 and 2, respectively. Hence a pair of symbols may represent a card of a given rank and suit. Note that the first two digits in Eq.(27) may V. COMBINATORIAL DESIGNS AND be interpreted as addresses determining the position of k-UNIFORMITY a symbol in the square. Furthermore, by considering the following four Mutually Orthogonal Latin Squares Combinatorial designs are arrangements of elements (MOLS) [39] of size five satisfying some specific properties so called balanced [33]. Such elements are restricted to a finite set, typi- 0000 4321 3142 2413 1234 cally considered as subsets of integer numbers. Some 1111 0432 4203 3024 2340 remarkable examples are block designs, t-designs, or- 2222 1043 0314 4130 3401 , (29) thogonal Latin squares and orthogonal arrays (see 3333 2104 1420 0241 4012 [34] and references therein). Combinatorial designs 4444 3210 2031 1302 0123 have important applications in quantum physics [35]. Indeed, a connection between genuinely multipartite we define a 2-uniform state of 6 subsystems with five maximally entangled states and orthogonal arrays has levels each:
1 |Φ5i = 5 (|000000i + |104321i + |203142i + |302413i + |401234i + |011111i + |110432i + |214203i + |313024i + |412340i + |022222i + |121043i + |220314i + |324130i + |423401i + |033333i + |132104i + |231420i + |330241i + |434012i + |044444i + |143210i + |242031i + |341302i + |440123i). (30) 10
A state locally equivalent to |Φ5i has been previously d = 10 expression (33) describes a 2-uniform state of found in [12] but its connection to MOLS is firstly 4 subsystems with 10 levels each. given here. By considering the standard construction In general, the problem of constructing N −2 MOLS of maximal sets of d − 1 MOLS of prime size d we can of size d is equivalent to construct a 2-uniform state generalize the above construction for quantum states of N qudits of d levels having d2 positive terms. Note of a prime number of levels d and N = d + 1 parties that d2 is the minimal number of terms that can have as follows a 2-uniform state of qudits of d level systems.
d−1 d−1 1 X O |Φ i = |i, ji |i + jmi. (31) d+1,d d B. AME states and hypercubes i,j=0 m=1
It is well known that a maximal set of d − 1 MOLS In the previous section we considered maximal sets of size d exist for every prime power d = pm [40]. of MOLS to construct 2-uniform states of qudits. This means that the above general expression can be However, this construction is not useful to find AME extended to the case of prime power level systems. For states for d > 4. The aim of this section is to instance, the maximal set of 3 MOLS of order d = 4 consider combinatorial arrangements for constructing can be represented by a color figure AME states in such cases. The main result is in- spired in a generalization of the AME(4,3) state |Ωi A♠ K♦ Q♥ J♣ given in Eq.(27) and the AME state of 6 ququarts J♦ Q♠ K♣ A♥ . (32) presented in Eq. (23). In [12] it was shown that this Q♣ J♥ A♦ K♠ K♥ A♣ J♠ Q♦
This design determines an AME(5,4) given by
|Ω5,4i = |00000i + |10312i + |20231i + |30123i + |01111i + |11203i + |21320i + |31032i + |02222i + |12130i + |22013i + |32301i + |03333i + |13021i + |23102i + |33210i, FIG. 2: Three mutually orthogonal Latin cubes of dimen- where every symbol of the sets {A, ♠, blue}, sion 3 and size 4. This arrangement allows us to generate {J, ♦, orange}, {Q, ♣, green} and {K, ♥, red}, is as- a state AME(6,4) of 6 ququarts. Each of the 12 planes sociated to 0, 1, 2 and 3, respectively, while the first (4 horizontal, 4 vertical and 4 oblique) contains a set of 3 two digits of every term label the position of a symbol MOLS of size 4. in the pattern. In the above expression a normaliza- tion factor is required. Note that this state, or a state state can be derived from the irredundant orthogonal equivalent with respect to local unitary transforma- array IrOA(64, 6, 4, 3). Furthermore, this orthogonal tions, arises from the Red-Solomon code of length five array can be interpreted as a set of three mutually [41]. orthogonal Latin cubes (see Fig.2). Also note that Furthermore, the construction can be extended to the AME(4,3) state |Ωi arises from an IrOA(9,4,3,2) any dimension d in the following way: (See Eq.(B1) in [12]). Thus, if k mutually orthog- onal hypercubes of dimension k having k + 1 sym- d−1 N (d) 1 X O bols exist they are one to one connected with an |Φ i = |i, ji |λ [i, j]i, (33) k d+1,d d m IrOA((k + 1) , 2k, k + 1, k) and, therefore, it would i,j=0 m=1 produce an AME(2k,k+1) state. This family of states would saturate the bound where N (d) denotes the maximal number of MOLS of size d. Here λm[ij] denotes the entries of the m-th N d ≥ + 1, (34) Latin square, so the above expression can be consid- 2 ered as a direct generalization of Eq.(2). It is worth to add that for dimensions d ≥ 12 not equal to a already defined in Eq.(22) Note that for k = 1 one ob-√ prime power number only lower bounds for the func- tains the standard Bell state, |Φ1i = (|01i + |10i)/ 2 tion N (d) are known [39]. The problem is solved only (1 Latin square of size 2 with 2 symbols). Taking for smaller dimensions, as N (6) = 1 in agreement with k = 2 and using 2 MOLS of size 3 and 3 symbols we unsolvability of the famous Euler problem of 36 offi- arrive to the AME(4,3) state of |Ωi Eq.(11) which, cers, while N (10) = 2 – see [40] where explicit form from this point of view, can be considered as a gen- of a pair of MOLS of size ten is derived. Thus for eralization of the Bell state. Furthermore, the state 11
(23) corresponding to k = 3 also belongs to this fam- in this case, to check unitarity for the three cases ap- ily and it is associated to three mutually orthogonal pearing in the first column of the right side of Eq.(38). Latin cubes of dimension 3 and size 4. It is interesting That is, taking combined indices in the original tensor, to check, whether there exist other states with k ≥ 5 (1) (2) (3) belonging to this family. tijkl = U(ij)(kl) = U(ik)(jl) = U(ij)(kl). (39)
absolute maximal entanglement is achieved if the ma- VI. AME STATES AND MULTI-UNITARY trices U (1), U (2) and U (3) correspond to different MATRICES changes of bases, that is unitary matrices. We refer to such particular kind of unitary matrices A. Unitary matrices and bipartite systems as multi-unitary. See Appendix B for further expla- nations about reorderings of elements in matrices of Let us illustrate the connection between unitary square size D ≥ 4. matrices and AME states for the simplest case of 2 qubits. Let us assume that the state of the system is given by C. General multi-unitarity 1 |φi = √ (U0,0|00i + U0,1|01i + U1,0|10i + U1,1|11i), Let us consider a more general case of pure states 2 of N subsystems with d levels each. That is, (35) where ρ = 1 UU †, ρ = 1 (U T )(U T )† and T denotes d−1 A 2 B 2 X transposition. Any unitary matrix U of size 2 repre- |φi = ts0,...,sN−1 |s0, . . . , sN−1i. (40) sents a Bell–like state. Furthermore, the Pauli set of s0,...,sN−1=0 four unitary matrices U = {I, σx, σy, σz}, orthogonal in the sense of the Hilbert-Schmidt product, defines Let us assume here that the number of subsystems 2k the maximally entangled Bell basis in H2 ⊗ H2. is even, N = 2k, so there exist M = k possible splittings of the system into two parts of the same size. B. Multi-unitary for AME(4,3) A necessary condition for |φi to be an AME state is that the tensor t with 2k indices, reshaped into a k We shall consider again the AME(4,3) state of four square matrix of size d forms a unitary matrix U. This is so, as the reduction associated to the first k qutrits |Ω4,3i, and represent its coefficients by a four– † index tensor qudits, given by ρk = UU , should be proportional to the identity. To arrive at an AME(2k,k) state, simi- X |Ω4,3i = tijkl |ijkli. (36) lar conditions have to hold for all M different square i,j,k,l=0,1,2 matrices obtained from the tensor t by all possible ways of reshaping its entries into a square matrix. where the entries of the tensor t can be expressed as This observation provides a clear motivation to in- product of the Kronecker delta functions troduce the notion of multi-unitarity: A square ma- trix A of order dk (k ≥ 2), acting on a composed 1 t = δ δ . (37) Hilbert space H⊗k and represented in a product basis ijkl 9 k,i+j l,i+2j d by An1,...,nk := hn1, . . . , nk|A|ν1, . . . , νki is k–unitary ν1,...,νk Here, the addition operations are modulo 3. As dis- if it is unitary for M = 2k reorderings of its entries cussed in the previous section, all non-zero coefficients k 4 corresponding to all possible choices of k indices out are equal. The tensor tijkl consists of 3 = 81 ele- of 2k. ments which can be reshaped to form a square matrix 4 In this way, we can establish the following one-to- of order 9. Note that there exist altogether 2 = 6 one connections: different ways of choosing a bi-partition of the indices and forming a matrix Uµ,ν . That is, AME(2, d) ≡ unitary of order d, (i + 2j, k + 2l), (k + 2l, i + 2j) for bipartite systems of 2 qudits having d levels each (µ, ν) = (i + 2k, j + 2l), (j + 2l, i + 2k) . (38) and, in general: (i + 2l, j + 2k), (j + 2k, i + 2l) AME(2k, d) ≡ k-unitary of order dk, The non-trivial property of an AME(4,3) tensor is that these six matrices are unitary. As transposition for multipartite systems of N = 2k qudits having d of a unitary matrix remains unitary, it is sufficient, levels each. By construction, 1–unitarity reduces to 12 standard unitarity. Any k–unitary matrix with k > 1 U T2 = P erm(0, 5, 7, 1, 3, 8, 2, 4, 6) and U R = is called multi-unitary. It is well-known that unitarity P erm(0, 2, 1, 4, 3, 5, 8, 7, 6), where T2 and R mean par- of matrices is invariant under multiplication. Multi- tial transposition and reshuffling (see appendix B for unitarity imposes more restrictions on a given matrix further details). Here, P erm denotes a permutation U than unitarity. Therefore, the product of two multi- matrix. A mini-catalogue of all the multi-unitary ma- unitary matrices in general is not multi-unitary. For trices defined in this work can be found in Appendix instance, the matrix O8 (see Eq.(41) below) is hermi- D. 2 tian and 3-unitary, but O8 = I is only 1-unitary, as it represents a 6–qubit quantum state equivalent to GHZ. D. AME and Hadamard matrices Similarly, the case of the AME(4,3) state |Ωi re- duces to analyzing the properties of the tensor tijkl in For six qubits, the AME(6,2) state |Ξ6,2i of Eq.(10) Eq. (36) and verify the multi-unitarity of U. Indeed, having a maximal number of terms arises from graph for this state we have U = P erm(0, 5, 7, 4, 6, 2, 8, 1, 3), states [42]. Let us write it explicitly
1 |Ξ i = ( −|000000i − |000001i − |000010i + |000011i − |000100i + |000101i + |000110i + |000111i 6,2 8 −|001000i − |001001i − |001010i + |001011i + |001100i − |001101i − |001110i − |001111i −|010000i − |010001i + |010010i − |010011i − |010100i + |010101i − |010110i − |010111i +|011000i + |011001i − |011010i + |011011i − |011100i + |011101i − |011110i − |011111i −|100000i + |100001i − |100010i − |100011i − |100100i − |100101i + |100110i − |100111i +|101000i − |101001i + |101010i + |101011i − |101100i − |101101i + |101110i − |101111i +|110000i − |110001i − |110010i − |110011i + |110100i + |110101i + |110110i − |110111i +|111000i − |111001i − |111010i − |111011i − |111100i − |111101i − |111110i + |111111i) .
This state leads to the following orthogonal matrix of trix. It represents a maximally entangled state with order D = 23 = 8 which is 3–unitary: the maximal number of terms having all entries of the same amplitude. For example, the AME states aris- −1 −1 −1 1 −1 1 1 1 ing from coding theory [11, 17] and graph states [42] −1 −1 −1 1 1 −1 −1 −1 are of this form. We recall that k-uniform states of −1 −1 1 −1 −1 1 −1 −1 N qudits with d levels having the minimum number 1 1 1 −1 1 −1 1 −1 −1 of terms (dk) are closely related to linear MDS codes O = √ . 8 −1 1 −1 −1 −1 −1 1 −1 8 (See Section 4.3 of [34]) and also to orthogonal arrays 1 −1 1 1 −1 −1 1 −1 of index unity [12]. This reasoning implies that any 1 −1 −1 −1 1 1 1 −1 pure state of n subsystems with d levels each having 1 −1 −1 −1 −1 −1 −1 1 at least one reduction to n/2 qudits maximally mixed (41) can have all its entries of the form ±d−n/2 if d is even. Note that the entries of |Ξ6,2i are given by the con- catenation of the rows of O8, up to normalization. This matrix is symmetric and equivalent up to enphas- ing and permutations [43] to the symmetric Hadamard ⊗3 matrix H8 = H2 . ⊗3 E. Further constructions of AME states Note that O8 is 3-unitary but H2 is not, so permu- tation or enphasing of a unitary matrix can spoil its multi-unitarity. Moreover, from the concatenation of Note that the state AME(4,3) |Ω4,3i is not equiv- ⊗3 the rows of H2 we only generate a 1-uniform state, alent with respect to local unitaries to a real state ⊗3 −2 which means that H2 is only 1-unitary (i.e., unitary). having all its entries of the form ±3 . This is a We conjecture that for any AME state one can consequence of the fact that a real Hadamard matrix choose suitable local unitary operations such that it of size 9 does not exist. However, the state |Ω4,3i is is related to a multi-unitary complex Hadamard ma- equivalent under local unitary operations to the state 13
(See Eq.(3) in [44]): define the discrete Weyl-Heisenberg group. This ap- proach can be easily generalized to any prime d > 2. 2 1 X Indeed, for d prime every reordering of indices lead us |Ω0 i = ωj(i−k)+l(i+k)|i, j, k, li, (42) 4,3 9 to the same matrix up to permutation of columns and i,j,k,l=0 rows, and therefore it remains unitary. This shows where ω = e2πi/3. This state is associated to the fol- a construction of AME(4,d) working for any prime lowing 2-unitary complex Hadamard matrix: number of levels d. Moreover, it is likely that this construction can be generalized for prime powers d by 1 1 1 1 w w2 1 w2 w considering the theory of Galois fields. Observe that 2 2 1 1 1 w 1 w w 1 w the above construction is essentially different from the 2 2 construction of AME states used in coding theory. In- 1 1 1 w w 1 w w 1 1 w w2 1 w2 w 1 1 1 deed, the tensor products of N displacement operators 1 2 2 2 2 2 U = w w 1 1 w w w w w . bases of size d, i.e. the set {Dp1,p1 ⊗ · · · ⊗ DpN ,pN }, P 1 2 1 2 3 w2 1 w 1 w2 w w w w produce codes and states AME(N,d) [11]. 1 w2 w 1 1 1 1 w w2 w2 w 1 w w w 1 w w2 2 2 2 2 2 w 1 w w w w 1 w w VII. CONCLUSION (43) Interestingly, every integer power (U )m is a complex P In this work we have analyzed some new proper- Hadamard matrix for m 6= 4 (Mod 4) and (U )8 = I. P ties of Absolutely Maximally Entangled (AME) states Moreover, UP is equivalent to the tensor product of † in multipartite systems. First of all, we have re- Fourier matrices F3 ⊗ F3 , that is viewed and extended several ways of constructing them. Then, we have explored their relation to U = DF ⊗ F †PD, (44) P 3 3 the field of combinatorial designs. For instance, a where D = Diag(1, 1, 1, 1, ω, ω2, 1, ω2, ω) is a diagonal state AME(4,3) consisting of four maximally entan- unitary matrix, while P is a permutation matrix which gled qutrits is linked to the set of two mutually orthog- changes the order of the columns from {1,..., 9} to onal Latin squares of order 3, while a state AME(6,4) {1, 4, 7, 2, 5, 8, 3, 6, 9}. made of 6 ququarts is related to the set of three mu- In order to construct a 2-unitary matrix one has to tually orthogonal Latin cubes of order 4. take a unitary U such that its partially transpose U T2 A profound relation between AME states and ten- and the reshuffled matrix U R are unitary – see Ap- sors that display the property of multi-unitarity has pendix B. In the case of a matrix U of size D = 32 been found: AME states made out of an even num- this implies that the set of nine 3 × 3 unitary matrices ber N of degrees of freedom are equivalent to multi- appearing in the 3 × 3 blocks of Eq.(43) define an or- unitary matrices (i.e., matrices being unitary after N thogonal basis for the Hilbert-Schmidt product. It is M = N/2 rearrangements of its entries). This re- thus possible to obtain AME states by considering or- markable property may be at the core of the use of thogonal bases of unitary operators. For instance, one AME states in holography [18, 19]. can construct the |Ωi state from the following matrix: Furthermore, making use of a state AME(2k,d) con- sisting of N = 2k parties, one can construct a k − 1- 1 0 0 1 0 0 1 0 0 uniform state state of 2k − 1 parties by removing a 0 1 0 0 ω 0 0 ω2 0 single subsystem. 0 0 1 0 0 ω2 0 0 ω We have proven the existence of states AME(4,d) 2 0 0 1 0 0 ω 0 0 ω for every prime d > 2. Note that this also provides the 1 0 √ 1 0 0 ω2 0 0 ω 0 0 existence of states AME(3,d) for every prime d > 2. UP = . (45) 3 0 1 0 0 1 0 0 1 0 AME states remain mainly unexplored. Let us 2 bring a number of open problems that deserve to be 0 1 0 0 ω 0 0 ω 0 solved. 0 0 1 0 0 1 0 0 1 2 1 0 0 ω 0 0 ω 0 0 • 7 qubits. We applied here the orthogonal basis defined by the Is there an AME state of 7 qubits? displacement operators of size d = 3: • Classification of AME and LU invariants. p1p2 p1 p2 It is unclear whether there is a clear cut classi- Dp1,p2 = τ X Z , (46) fication of AME, which is related to LU invari- 2 πi/d 2πi/d where p = (p1, p2) ∈ Zd, τ = −e , ω = e , ants. The fact that some AME states carry dif- X|ki = |k + 1i and Z|ki = ωk|ki. These operators ferent minimal supports or that some AME are 14
right away related to Reed-Solomon codes hints sufficient to analyze the purity of reduced density ma- a some unknown structure among AME states. trices. The purity Trρ2, serves as a measure of the degree of mixedness of the density matrix ρ, but also • Non minimal support. as a measure of entanglement of the initially pure state Examples of AME states with non-minimal sup- reduced to ρ by a partial trace. port are only explicitly known for 5 and 6 qubits. The same argument works for multipartite systems. It would be natural to find examples for higher Let us denote the qubits by A, B, C and D. The purity local dimensions. 2 of the bipartition AB of a state |ψi is given by Tr(ρAB) • Computation of invariants. with ρAB = TrCD|ψihψ| being the reduced density LU invariants grow exponentially with the size matrix of AB. The theoretical minimum for this quan- of the number of parties. An example of them tity is 1/N; so 1/4 in the case of four qubits. However, is the hyperdeterminant, which has only been one cannot attain this minimum for the three biparti- computed up to 4 qubits. There are no compu- tions at the same time, as it was proven analytically tations of hyperdeterminants of 4 qutrits. Do in Ref. [6]. The best one can have is 1/3 in all bipar- AME states carry maximum values for some LU titions. One state that accomplishes this was given in invariant? From the general theory of hyperde- that same paper: terminants, we know that the rank of the tensor defining a four-qutrit state has rank 1269, which 1 |HSi = √ [|0011i + |1100i + ω|0101i+ seems out of reach for any practical computa- 6 tion. ω|1010i + ω2 (|0110i + |1001i) , • Bell inequalities. (A1) Bell inequalities related to AME(4,3) are un- 2iπ known. where w = exp( 3 ). This state has also maximum entropy of entanglement. Another state that has min- The small corner of the Hilbert space formed by imum purity in all bipartitions is: AME states may well be more complex than expected. 1 |HDi = √ (|0001i + |0010i + |0100i Acknowledgements. It is a pleasure to thank 6 J. Bielawski, M. Grassl and S. Pascazio for useful re- √ marks. DG is thankful to Pawel Horodecki for the +|1000i + 2|1111i). (A2) hospitality during his stay in Sopot, where this project was initiated. JIL acknowledges financial support This state was found by the authors and by [46] to by FIS2013-41757-P and DA the APIF grant from also carry maximum hyperdeterminant, an extension Universitat de Barcelona. AR thanks support from of the concept of determinant to higher dimensions the Beatriu de Pin´osfellowship (BP-DGR 2013) and [47], and an interesting entanglement measure that is EU IP SIQS. This work was supported by the ERC intrinsically multipartite. An equivalent state already Advanced Grant QOLAPS coordinated by Ryszard appeared in [5] as an example of a ”symmetric maxi- Horodecki. mally entangled state” of 4 qubits. Let us note that AME states are defined through their entanglement properties and thus can be trans- Appendix A: Four-qubit entangled states formed into any equivalent state under local unitaries. Gour and Wallach [48] found |Li and |Mi states For completeness we will present in this appendix while searching, respectively, for the states that max- a discussion of maximally entangled states of four imize the average Tsallis α-entropy of entanglement qubits. To see that no AME(4,2) states exist it is for α > 2 and for 0 < α < 2: 15
1 |Li =√ (((1 + ω)(|0000i + |1111i) + (1 − ω)(|0011i + |1100i) 12 +ω2 (|0101i + |0110i + |1001i + |1010i) (A3) 1 i 1 i 1 |Mi =√ + √ (|0000i + |1111i) + − √ (|0011i + |1100i) + 2 2 12 2 12 1 √ (|0101i + |1010i) , (A4) 3
where ω = e2πi/3. Remarkably, states |Li and |Mi can a vector of length 4 and putting into the reordered be transformed by SLOCC into the states |HDi and matrix XR. That is, |HSi respectively. They are called SLOCC equivalent. In fact Verstraete et al. gave a classification of all X11 X21 X13 X23 pure 4-qubit states in 9 SLOCC inequivalent classes T X12 X22 X14 X24 [22]. The most important class is called the generic X 2 := , X X X X class and is presented by Gour and Wallach [48] in the 31 42 33 43 following compact form: X24 X42 X34 X44 X11 X12 X21 X22 + + − − + + G ≡ {z0|φ i|φ i + z1|φ i|φ i + z2|ψ i|ψ i R X13 X14 X23 X24 − − X := . (B1) +z3|ψ i|ψ i | z0, z1, z2, z3 ∈ C}, (A5) X31 X32 X41 X42 √ ± ± X33 X34 X43 X44 where√|φ i = (|00i ± |11i/ 2 and |ψ i = (|01i ± |10i/ 2 are the Bell states. Here, colored exchanged entries are set in boldface The method to transform a state from the compu- (colors in online version). The other 3 matrices are tational basis into the Bell basis (which means ob- transpositions of U, U T2 and U R. Therefore, a matrix taining the values of the zi) is explained in detail U of size 4 is multi-unitary if U, U T2 and U R are in Verstraete’s paper. Any 2 states that have the unitary. The same restrictions hold for matrices of same 4 parameters z0, z1, z2 and z3 in this basis are size d2. SLOCC-equivalent. The SLOCC-equivalence between It is possible to demonstrate that 2-unitary matrices the states |Li and |HDi (as well as between the states of size D = 4 do not exist [6]. The smallest 2-unitary |Mi and |HSi) is proven by showing that they have matrix exists for order D = 9 – see Eq.(43). precisely the same zi coefficients.
Appendix C: Symmetric sudoku, generalized Appendix B: Partial transposition, reshuffling AME(4,3) states and 2–unitary permutations and 2-unitarity According to the usual sudoku rules all digits in 2 For any matrix X of a square order D = d repre- a single row/column of the matrix or block of size sented in a product basis, Xmn := hm, n|X|µ, νi, one µ,ν three are different. Let us distinguish other sets of T2 nine elements: location, which contain all digits from defines [49] its partial transposition, Xmn = Xmν , and µν µn the same place in each block (e.g. nice centers of the R reshuffling, Xmn = Xmµ. To get a better feeling of µν nν blocks - set in red in the matrix below); broken rows these particular reorderings of elements we consider containing three rows of length three occurring in the a matrix X of order four. Let us now switch to the same position of three blocks (blue example in the standard, two–index notation, and write its elements matrix) and an analogous notion of broken column. 4 as Xij with i, j = 1, 2, 3, 4. Here, we have 2 = 6 Note that the standard operation of matrix trans- different reorderings of 2 indices out of 4. Two of pose, XT , exchanges columns of the matrix with its these reorderings are particularly interesting: (a) the rows. The partial transposition XT2 exchanges rows partially transposed matrix XT2 is equivalent to the with broken columns and columns with broken rows. matrix with all four blocks of size two transposed, and Furthermore, reshuffling operation, XR interchanges (b) the reshuffled matrix XR is obtained by taking lex- blocks with rows and columns with locations – see icographically each 2 × 2 block of X, reshaping it into (B1), 16
The following matrix shows an example of a sym- unity [39]. Here, we present a resume of all the multi- metric sudoku pattern analyzed in [50]: all digits in unitary matrices presented in this work and some ad- each column, each row, each block, each location (red ditional ones: in online version), each broken row (in blue), and each broken column (boldface) are different. 3 (i) D = 2 = 8. The orthogonal matrix O8 defined in (41) is 3–unitary. This matrix is equivalent with respect to permutations and enphasing to 8 16 2 4 9 5 7 3 the three–qubit Hadamard matrix H = H⊗3 8 2 3 57 6 81 9 24 (which is 1–unitary). It is not possible to con- 4 9 2 7 3 5 1 6 8 struct a 3–unitary permutation matrix of size 735 1 6 8 4 9 2 D = 8 [12]. S9 := 249 573 816 . (C1) (ii) D = 32 = 9. There exist 2–unitary permutation 6 8 1 9 2 4 3 5 7 matrices related to symmetric sudoku designs – 924 3 5 7 6 8 1 168 492 735 see Appendix C. Also permutation matrices de- fined at the end of Section VI B and the complex 5 7 3 8 1 6 2 4 9 Hadamard matrix (43) are 2-unitary. The latter one is equivalent to the tensor product of two Fourier matrices F ⊗ F , but this product is Consider now a matrix P1 of size nine with all en- 3 3 tries equal to zero besides nine elements placed in the 1–unitary only. positions of digit ’1’ of the matrix (C1) equal to unity. (iii) D = 24 = 42 = 16. In this case As all nine digits in each location of S are different 9 there are no 4–unitary matrices [11]. There it is clear that this is a legitimate permutation matrix exist a 2–unitary permutation matrix given of size 9. As the address of each non-zero element can by P erm(4, 3, 13, 10, 14, 9, 7, 0, 11, 12, 2, 5, 1, 6, 8, 15) be interpreted as a pair of two ternary digits, it repre- created from OA(16,4,4,2), which was generated sents a Graeco-Latin square (28) and determines the using the Gendex Module NOA [60]. AME(4,3) state |Ωi (27). Thus permutation matrix P1 is 2–unitary, as its (iv) D = p2 for a prime p. Two-unitary matri- partial transpose and reshuffling remain unitary. The ces of this size exist, but they are neither com- same property holds also for other permutation ma- plex Hadamard matrices (d4 non-zero entries) trices Pm with m = 2,..., 9, obtained by placing nine nor permutations (d non-zero entries). Indeed ones in the positions occupied by digits ’m’ in pattern they have p2 non-zero entries which correspond (C1). The property of 2-unitarity is preserved if we to powers of the main root of the unity e2πi/d. enphase a permutation matrix by multiplying it by Construction: Let us denote (j, k) to the j-th 0 a diagonal unitary matrix D, and take Pm = DPm. row block of size p and the k-th column block Thus it is fair to say that the symmetric sudoku ma- of size p of U (note that U has d2 blocks of size trix (C1) encodes nine families of enphased 2–unitary p). Each block (j, k) has to be filled with the dis- permutation matrices or, families of AME(4,3) states. placement operator Dj,k of size d. Displacement operators are defined in Eq.(46). The explicit U for d = 3 is provided in Eq.(45). Appendix D: Mini catalogue of multi–unitary matrices As we can see, for some dimensions there are no 2–unitary permutation matrices. Indeed, a 2–unitary Multi–unitary matrices are defined for orders equal matrix of size d(d+1)/2 exists if and only if a projec- to powers of integers. As there are no 2-unitary matri- tive plane of odd order d exist (see Theorem 8.43 in ces of size 4 the smallest interesting cases are D = 8, 9 [34]). In particular, they exist for every prime (and and D = 16. In general, multi-unitary permutation odd prime power) d. For example, the existence of matrices are one-to-one connected [12] to combinato- AME(4,3) states relies on the existence of the projec- rial arrangements called orthogonal arrays of index tive plane of order 3.
[1] D. N. Page, Average entropy of a subsystem, Phys. Cambridge University Press (2006). Rev. Lett. 71, 1291 (1993). [3] S. Sen, Average Entropy of a Quantum Subsystem, [2] I. Bengtsson and K. Zyczkowski,˙ Geometry of quan- Phys. Rev. Lett. 77, 1 (1996). tum states: an introduction to quantum entanglement, [4]K. Zyczkowski˙ and H.-J. Sommers, Induced measures 17
in the space of mixed quantum states, J. Phys. A 34, be entangled in two different ways. Phys. Rev. A 62, 7111-7125 (2001). 062314 (2000). [5] N. Gisin and H. Bechmann–Pasquinucci, Bell inequal- [26] S. Muralidharan and P. K. Panigrahi, Perfect telepor- ity, Bell states and maximally entangled states for n tation, quantum-state sharing, and superdense coding qubits?, Phys. Lett. A 246, 1 (1998). through a genuinely entangled five-qubit state, Phys. [6] A. Higuchi and A. Sudbery, How entangled can two Rev. A 77, 032321 (2008). couples get?, Phys. Lett. A 272, 213 (2000). [27] R. Laflamme, C. Miquel, J. P. Paz and W. H. Zurek, [7] P. Facchi, G. Florio, G. Parisi, and S. Pascazio, Max- Perfect Quantum Error Correcting Code, Phys. Rev. imally multipartite entangled states, Phys. Rev. A 77, Lett. 77, 198 (1996). 060304 (2008). [28] A. Borras et al., Multi-Qubit Systems: Highly Entan- [8] P. Facchi, G. Florio, U. Marzolino, G. Parisi, S. Pas- gled States and Entanglement Distribution J. Phys. A: cazio, Multipartite entanglement and frustration, NJP Math. Theor. 40 (2007) 13407 12, 025015 (2010). [29] P. Facchi, Multipartite entanglement in qubit systems, [9] L. Arnaud and N. Cerf, Exploring pure quantum states Rend. Lincei Mat. Appl. 20, 25-67 (2009). with maximally mixed reductions, Phys. Rev. A 87, [30] W. Helwig and W. Cui, Absolutely Maximally 012319 (2013). Entangled States: Existence and Applications [10] I. D. K. Brown, S . Stepney, A . Sudbery, and S . arXiv:1306.2536 (2013). L. Braunstein, Search for highly entangled multi-qubit [31] R. Singleton, Maximum distance q -nary codes, Infor- states, J. Phys. A: Math. Gen. 38, 1119 (2005). mation Theory, IEEE Transactions 10, 116 (1964). [11] A. J. Scott, Multipartite entanglement, quantum- [32] E. M. Rains, Quantum Shadow Enumerators, IEEE error-correcting codes, and entangling power of quan- transactions on information theory, 45, 2361 (1999). tum evolutions, Phys. Rev. A 69, 052330 (2004). [33] D. Stinson, Combinatorial Designs: Constructions [12] D. Goyeneche and K. Zyczkowski,˙ Genuinely multi- and Analysis, (Springer-Verlag, New York, 2004) partite entangled states and orthogonal arrays, Phys. [34] A. S. Hedayat, N. J. A. Sloane, and J. Stufken, Or- Rev. A 90, 022316 (2014). thogonal Arrays: Theory and Applications, (Springer- [13] W. Helwig, W. Cui, J. I. Latorre, A. Riera, and H.- Verlag, New York, 1999). K. Lo, Absolute maximal entanglement and quantum [35] G. Zauner, Quantendesigns: Grundz¨uge einer secret sharing, Phys. Rev. A 86, 052335 (2012). nichtkommutativen Designtheorie, Ph. D. thesis, Uni- [14] W. Helwig, Absolutely Maximally Entangled Qudit versit¨atWien (1999). Graph States, arXiv:1306.2879 [quant-ph] (2013). [36] D. Raghavarao, Constructions and Combinatorial [15] V. Coffman, J. Kundu, and W. K. Wootters, Dis- Problems in Design of Experiments (Dover, New tributed Entanglement, Phys. Rev. A 61, 052306 York, 1988). (2000). [37] A. Atkinson, A. Donev, R. Tobias, Optimum Experi- [16] B. M. Terhal (2003), Is Entanglement Monogamous?, mental Designs, with SAS. (Oxford University Press, IBM J. of Res. Dev. 48, No.1, pp. 71– 78 (2004). Oxford, 2007) [17] I. Reed and G. Solomon, Polynomial codes over cer- [38] J. Hadamard, R´esolutiond’une question relative aux tain finite fields, J. of the Soc. for Ind. and App. Math. d´eterminants, Bulletin des Sciences Math´ematiques 8, 300 (1960). 17, 240, (1893). [18] J. I. Latorre and G. Sierra, Holographic codes, [39] C. J. Colbourn, and J. H. Dinitz (Eds.) Handbook of arXiv:1502.06618 (2015). Combinatorial Designs, Second Edition, (CRC Press, [19] F. Pastawski, B. Yoshida, D. Harlow and J. Boca Raton FL, 2006). Preskill, Holographic quantum error-correcting codes: [40] D. R. Stinson, Combinatorial Designs: Construction Toy models for the bulk/boundary correspondence, and Analysis, (Springer-Verlag, New York, 2004). arXiv:1503.06237 (2015). [41] http://neilsloane.com/oadir/ [20] A. Ac´ın,A. Andrianov, L. Costa, E. Jan´e,J. I. La- [42] M. Hein, J. Eisert, H.J. Briegel, Multi-party entangle- torre, R. Tarrach, Generalized Schmidt decomposition ment in graph states, Phys. Rev. A 69, 062311 (2004) and classification of three-quantum-bit states, Phys. [43] W. Tadej and K. Zyczkowski,˙ A concise guide to com- Rev. Lett. 85, 1560, (2000). plex Hadamard matrices, Open Syst. Inf. Dyn. 13, [21] J. I. de Vicente, C. Spee and B. Kraus, The max- 133-177 (2006). imally entangled set of multipartite quantum states, [44] M. Gaeta, A. Klimov, J. Lawrence, Maximally Entan- Phys. Rev. Lett. 111, 110502 (2013). gled States of Four Nonbinary Particles, Phys. Rev. A [22] F. Verstraete, J. Dehaene, B. De Moor and H. Ver- 91, 012332 (2015). schelde, Four qubits can be entangled in nine different [45] A. Ac´ın,T. Durt, N. Gisin, and J. I. Latorre, Quan- ways, Phys. Rev. A 65, 052112 (2002). tum non-locality in two three-level systems, Phys. Rev. [23] P. Dit¸˘a,Some results on the parametrization of com- A 65, 052325 (2002). plex Hadamard matrices, J. Phys. A: Math. Gen. 37, [46] A. Osterloh and J. Siewert, Entanglement monotones 5355 (2004). and maximally entangled states in multipartite qubit [24] P. Facchi, G. Florio, U. Marzolino, G. Parisi, and S. systems, Int. J. Quant. Inf. 4, 531 (2006) Pascazio Classical Statistical Mechanics Approach to [47] I. M. Gelfand, M. M. Kapranov and A. V. Zelevin- Multipartite Entanglement J. Phys. A: Math. Theor. sky, Discriminants, Resultants and Multidimensional 43, 225303 (2010). Determinants, Ed. Birkhauser (1994). [25] W. D¨ur,G. Vidal and J. I. Cirac, Three qubits can [48] G. Gour and N. R. Wallach, All maximally entangled 18
four-qubit states, J. Math. Phys. 51, 112201 (2010). imally entangled symmetric state in terms of the geo- [49]K. Zyczkowski˙ and I. Bengtsson, On duality between metric measure, New J. Phys. 12, 073025 (2010). quantum states and quantum maps, Open Syst. Inf. [55] D. J. H. Markham, Entanglement and symmetry Dyn. 11, 3-42 (2004). in permutation-symmetric states, Phys. Rev. A 83, [50] R. A. Bailey, P. J. Cameron and R. Connelly, Sudoku, 042332 (2011). gerechte designs, resolutions, affine space, spreads, [56] J. Luque and J. Thibon, Polynomial invariants of four reguli, and Hamming codes, Am. Math. Monthly 115, qubits, Phys. Rev. A 67, 042303 (2003). 383-404 (2008). [57] S. Turgut, Y. Gul and N. K. Pak, Deterministic [51] G. Gour, and N. R. Wallach, Classification of Mul- Transformations of Multipartite Entangled States with tipartite Entanglement of All Finite Dimensionality, Tensor Rank 2, Phys. Rev. A 81, 012317 (2010). Phys. Rev. Lett. 111, 060502 (2013). [58] S. Kintas a S. Turgut, Transformations of W-Type [52] A. Klappenecker, M. Rotteler, Beyond Stabilizer Entangled States, J. Math. Phys. 51, 092202 (2010). Codes I: Nice Error Bases, IEEE Trans. Inform. The- [59] R. F. Werner, All teleportation and dense coding ory, 48, 8, 2392-2395 (2002). schemes, J. Phys. A 34, 7081 (2001). [53] J. Martin, O. Giraud, P. A. Braun, D. Braun, and T. [60] See webpage on orthogonal arrays Bastin, Multiqubit symmetric states with high geomet- http://designcomputing.net/gendex/noa ric entanglement, Phys. Rev. A 81, 062347 (2010). [54] M. Aulbach, D. Markham, and M. Murao, The max-