arXiv:0906.1499v2 [cond-mat.stat-mech] 3 Dec 2009 II ocuin nageeta h are for barrier the as entanglement Conclusion: VIII. I.Etnlmn ln unu computation quantum along Entanglement VII. I.Saigo entanglement of Scaling III. I yaiso Entanglement of Dynamics VI. V te models Other IV. I nepii optto fetnlmn entropy entanglement of computation explicit An II. .Rnraiaino Entanglement of Renormalization V. .Introduction I. .TeX oe 9 11 7 model Lipkin-Meshkov-Glick The D. model XY The A. 4 2 systems One-dimensional A. block a of entropy Entanglement D. Entropy Entanglement A. .Qatmcrut 15 References Acknowledgments simulations classical circuits Quantum A. 12 entanglement block the of evolution Time A. states quantum of Renormalization A. .TeXZmdl9 10 models 3 3 Disordered C. model XXZ The B. 7 law charge Area central and C. theory field Conformal B. State Ground C. model XX B. .Oewyqatmcmuain17 16 computation quantum way One C. computation quantum Adiabatic B. 13 flows RG of Irreversibility B. .Ln ag neatos14 14 entropy interactions block range the Long of evolution C. time for Bounds B. .Pril nageet12 entanglement Particle E. 5 entropy the of Scaling E. .Etnlmn nrp n opizdtriat6 determinant Toeplitz and entropy Entanglement F. entropy et ’srcuaiCnttet el aèi,Universit Matèria, la de Constituents i d’Estructura Dept. ASnmes 36.a 36.d 03.67.Hk 03.65.Ud, 03.67.-a, numbers: PACS computation. quantum a along entanglement trajectories of con group impuriti renormalization the of along presence entanglement on ent of the generally, of and more properties entanglement and, between scaling relation transitions the phase on quantum placed at is Emphasis systems. erve oeo h eetpors ntesuyo nrp o entropy of study the on progress recent the of some review We Contents hr eiwo nageeti unu pnsystems spin quantum in entanglement on review short A .I aor n .Riera A. and Latorre I. J. 6 2 1 8 7 15 12 18 18 18 13 13 td acln,67Daoa,008Breoa Spain Barcelona, 08028 Diagonal, 647 Barcelona, de at h yaia vlto fetnlmn n h fate the and entanglement of evolution dynamical the , hsi aiyudrto sflos e scnie con- a consider system. us Let the of follows. correlation as state nected understood the easily in is distributed This is understandi to entanglement tantamount how is correlations assess quantum correct under these The system of algebraically. the of transition, case phase function the a in a goes or, a as them in exponentially separating spins distance decay the two may between chain obser function spin an correlation typical instance, the as For such able exhibit. they correlations servable hscnetdcreao ol aihietclyfrany for identically vanish state would product correlator connected This eaosta euet rdcso w-on correlators two-point of products to co reduce n-point that instance, produce For relators Hamiltonians) (Gaussian entanglement. particles of free degree system observed the the original in achieve operated the have how Hamiltonian manifest the would in them interactions correla of connected Any n-point functions. or stat tion three- given two-, a consider on could functions We correlations studying by apparent chains. spin in present interaction Hamiltonia the of describe states that ground of shal properties we entanglement particular, the In relevant. focu physically shall are we Yet, that states state. prop quantum entanglement a characterize the that with int erties concerned various thus, using are, ways We different ma actions. in or obtained mechanism be be post-selection effectively may a which using se, created per analyse artificially states to specific sensible of perfectly properties is tanglement It states. to Hamiltonians orltoso esnegMdlwt hs fQuantum of those with the Model compare Heisenberg we a can How of correla- specific correlations dependent. model of is study functions the tion Nonetheless, theorem. Wick’s eae n,a atclrcs,ee h aumdsly a displays theori vacuum field the quantum even in cor structure case, entanglement highly particular non-trivial be a will as system and, interesting related any of state state the ground in the entanglement fr come of only amount can the correlations its consequently, and, erator where s h dao atceetnlmn,teevolution the entanglement, particle of idea the es, unu ytm r liaeycaatrzdb h ob- the by characterized ultimately are systems Quantum ti la httepoet fetnlmn a emade be can entanglement of property the that clear is It from moved has emphasis our point, this at that, Notice eto ralw eas reydsrb the describe briefly also We law. area of cept oyfroedmninlmlipriemodels multi-partite one-dimensional for ropy O i and Ψ nageeti aybd quantum many-body in entanglement f | | O Ψ Ψ O i O | j O = j r prtr tsites at operators are i | .INTRODUCTION I. O Ψ ⊗ j i | | c Ψ ψ ≡ i − htis, That . Ψ | O i | O Ψ i ⊗ Ψ i | Ψ O and | O j tflosthat follows It . j sapoutop- product a is | Ψ j respectively. , study l ment es. on s en- om via (1) er- ng ns v- to e. r- y - - - - 2
Chromodynamics? Each theory brings its own set of local II. AN EXPLICIT COMPUTATION OF ENTANGLEMENT and non-local operatorsthat close an Operator Product Expan- ENTROPY sion. Different theories will carry different sets of operators, so that a naive comparison is hopeless. A wonderful possibil- Let us start our discussion with the study of the behaviour ity to quantify degrees of entanglement for unrelated theories of entanglement at different regimes (critical and non-critical) emerges from the use of Renormalization Group ideas and the of the XX model. As we shall see, entanglement entropy will study of universal properties. For instance, a system may dis- be a good tool to describe the properties of the quantum phase play exponential decays in its correlations functions which is transition which characterize this model [1, 2]. globally controlled by a common correlation length. A model In order to do this, first we need to introduce the Von Neu- with a larger correlation length is expected to present stronger mann entropy as a measure of the bipartite entanglement in long-distance quantum correlations. pure sates. Then, we will study the scaling of the entangle- We may as well try to find a universalunique figure of merit ment entropy for the simple XX model. We then will proceed that would allow for a fair comparison of the entanglement to compute the ground state GS of the system, from which | present in e. g. the ground state of all possible theories. Such we can obtain the spectra of the reduced density matrix ρL for a figure of merit cannot be attached to the correlations proper- the block of L contiguous spins. The knowledge of the eigen- ties of model-dependentoperators since it would not allow for values of ρL will let us determine its entanglement entropy comparison among different theories. The way to overcome SL. Finally, we are going to analyse how the entanglement this problem is to look for an operator which is defined in ev- behaves depending on the critical properties of the model. ery theory. It turns out that there is only one such operator: the stress tensor. To be more precise, we can use the language of conformal field theory which establishes that there always is A. Entanglement Entropy a highest weight operator that we call the Identity. The Iden- tity will bring a tower of descendants, the stress tensor being its first representative. Indeed, the stress tensor is always de- The problem of measuring and quantifying quantum corre- fined in any theory since it corresponds to the operator that lations, or entanglement, in many body quantum systems is a measures the energy content of the system and it is the oper- field of research in its own, that benefits both from condensed ator that couples the system to gravity. Correlators of stress matter and quantum information ideas. Here, we shall only tensor operators are naturally related to entanglement. In par- discuss the Von Neumann entropy as a figure of merit for en- ticular, the coefficient of the two-point stress tensor correlator tanglement. Nevertheless, there are many other measures that in a conformal field theory in two dimensions corresponds to have been largely explored. A detailed explanation of them the central charge of the theory. can be found in several reviews [3, 4, 5, 6, 7, 8, 9, 10, 11]. Our choice for the entropy of entanglement is based on a There is a secondoptionto measure entanglementin a given combination of ideas. Entropy has a clear information the- state with a single measure of entanglement which is closer in ory meaning. It also relates to extensive research in quantum spirit to the ideas of Quantum Information. The basic idea field theory and the physics of black holes. Furthermore, its consists of using the von Neumann (entanglement) entropy of scaling properties are related to the characterization of quan- the reduced density matrix of a sub-part of the system which tum phase transitions as provided by conformal field theory. is analysed. Indeed, the entanglement entropy quantifies the On the other hand, entanglement entropy is not a simple quan- amount of surprise that a sub-part of a system finds when dis- tity to compute, neither a direct observable (though it relates covering it is correlated to the rest of the system. Therefore, to them). entanglement entropy is a bona fide measure of the correla- The von Neumann entropy can be used to measure the en- tions in the system. The advantage of the von Neumann en- tanglement between two parts of the system, that we call A tropy of entanglement is that it can be defined for any system. and B. Let us take a ket ψ belonging to = . We expect its general properties, as the way it scales with the AB A B According to the Schmidt| decomposition, forH anyH pure⊗ bipar- H size of the sub-part of the system we are considering, should tite state we can always find two orthonormal basis ϕ characterized the quantum state in a quite refined way. {| i A} and φj B such that the state ψ AB can be written as It is tantalizing to exhaustively explore the behaviour of the {| } | entropy of entanglement in relevant physical systems. For in- χ stance, will the entropy of entanglement scale differently at a ψ = α ϕ φ , (2) | AB i| i A| i B critical point as compare to a non-critical one? Will scaling i properties depend on the dimensionality of the system. Is dis- order relevant for long-distance correlations? Are there non- where αi can be chosen real and positive and are called local systems where entropy obeys some singular behaviour? Schmidt coefficients, and χ min (dim A, dim B ) is the How does entanglement renormalize? How does entangle- Schmidt number. Note that the≤ Schmidt decompositionH H is just ment evolve dynamically? We can even go further away from the diagonalization of the matrix of coefficients in the original standard dynamical models and question whether entangle- state which is always possible if we can perform two indepen- ment is somehow related to computational complexity prob- dent unitary transformations in A and B. lems, both NP-complete and QMA-complete. We shall now The Von Neumann entropy between these bipartitions is de- briefly review some of these questions. fined as the Shannon entropy of the square of the Schmidt co- 3 efficients, Without loss of generality, we are going to consider that the magnetic field is oriented in the positive Z direction (λ > 0), S = S α2 log α2 . (3) since, if this was notthecase, we could always map thesystem A B ≡− i i i onto an equivalent one with λ > 0 by simply interchanging the spin states up and down. This expression can be written in terms of the reduced den- sity matrices of each part of the system. That is, C. Ground State S S(ρ )= tr (ρ log ρ ) , (4) A ≡ A − A 2 A where Next, we need to compute the ground state GS of the XX Hamiltonian (7). In order to do this, we will follow| two steps: 2 (i) first, we will perform a Jordan-Wigner transformation to ρA = tr B( ψ AB ψ AB)= α φi B ψi B . (5) | | i | | rewrite as a quadratic form of fermionic operators, and i HXX then (ii) we will take profit of the translational invariance of It is easy to see that SA = SB. Thus, the surprise that A the system realizing a Fourier transform which will diagonal- experiences when discovering its correlation to B is identical ize the Hamiltonian. A third step which is needed in the more to the one of B realizing its correlation with A. general XY model, the Bogoliubov transformation, is not nec- The von Neumann entropy verifies the following properties: essary in this particular case. Let us remark that this compu- i) it is invariant under local unitary operations (SA = SB is tation is standard and appears in many text books [13, 14, 15]. a function of the αi’s only); ii) it is continuous (in a certain The Jordan-Wigner transformation maps a spin chain of sense also in the asymptotic limit of infinite copies of the state; interacting qubits onto an equivalent system of interacting see e.g. Ref. [4]); iii) it is additive: S( ψ φ )= S( ψ )+ fermions. This powerful transform is defined by the follow- S( φ ). | ⊗ | | ing relation between the Pauli matrices and the creation and In| our particular case, we are going to use the entanglement annihilation of the fermionic modes entropy to study the quantum correlations of spin chains. We l−1 y will be interested in determine the entanglement between a σx iσ a = σz l − l . (8) block of L contiguous spins and the rest of the chain. Then, l m 2 m=0 if GS represents the ground state of a system of N spins, | ρL = tr N−L( GS GS ) is the reduced density matrix of We, indeed, can check that the fermionic operators al fulfil the block of L contiguous| | spins that we will use in Eq. 4. the canonical commutation relations Finally, let us point out that, in the case the ground state † that we study is translationally invariant, neither ρ nor S a ,am = δlm, al,am =0 . (9) L L { l } { } will depend on the position of the block of spins in the chain. The idea behind the transformation is to identify the state of In this case, it is easy to show that the entropy SL is a concave function respect to L [12] the spin l (0 or 1 in the computational basis) with the occu- pation number of the corresponding fermionic mode. Thus, x y SL−M + SL+M in Eq. 8, the factor (σl iσl )/2 corresponds to the operator SL , (6) − l−1 z ≥ 2 0 1 in the computational basis, and the product m=0 σm generates| | the appropriate sign in order to satisfy the commu- where L =0, , N, and M =0, , min N L,L . { − } tation relations. The Jordan-Wigner transformation casts the XX Hamilto- nian onto B. XX model N−1 N−1 H = a†a + a† a + λ a†a , (10) We shall now present a computation of entanglement en- XX − l l+1 l+1 l l l tropy for the reduced density matrix of the ground state of the l=0 l=0 widely studied XX model [1, 2]. This theory captures the non- which corresponds to a model of free fermions with chemical trivial structure of a quantum phase transition, while remain- potential λ. ing simple enough to carry explicit computations throughout. Now, let us exploit the translational symmetry of the system 1 The XX model consists of a chain of N spin 2 particles with by introducing the Fourier transformed fermionic operators nearest neighbour interactions and an external magnetic field. Its Hamiltonian is given by N−1 1 −i 2π kl b = a e N , (11) k √ l N−1 N−1 N l=0 1 x x y y 1 z HXX = σl σl+1 + σl σl+1 + λ σl , (7) −2 2 where 0 k N 1. As the Fourier transform is a unitary l=0 l=0 ≤ ≤ − transformation, these new bk operators also satisfy the canon- where l labels the N spins, λ is the magnetic field and σl ical commutation relations and, therefore, they are fermionic ( = x,y,z) are the Pauli matrices at site l. operators. 4
The Hamiltonian, written in terms of these bk operators, D. Entanglement entropy of a block displays a diagonal structure The strategy to get the Von Neumann entropy of a block N−1 † of L spins first consists in computing the correlation matrix HXX = Λkbkbk , (12) † aman of the GS in this block. Then, the eigenvalues of this k=0 correlation matrix are related with the eigenvalues of the re- where the energy that penalizes (or favours, depending on the duced density matrix of the block which are required to deter- sign) the occupation of mode k is mine the entanglement entropy. The simple structure of the GS, shown in Eqs. (14) and (15), 2πk makes easy to compute its correlation matrix Λ = λ 2cos . (13) k − N † δpq if Λp < 0 bpbq = . (17) We have assumed that the system satisfied periodic boundary 0 if Λp > 0 conditions. If this was not the case, the Hamiltonian would not be diagonal due to an extra term proportional to 1 . Inthe From now on, we will consider the case in which 2 > λ 0. N † ≥ thermodynamic limit, therefore, this extra term disappears. Notice that if λ > 2, then bpbq = 0 for all p and q. This † We realize that, on one hand, if λ > 2, then Λ 0 k. case is trivial to analyse, since the correlators aman are also k This implies that the ground state of the system is≥ the state∀ null, and the GS is in a product state. annihilated by all bk operators The next step is to go back in the Fourier transform to get the correlation matrix of the an operators b GS =0 k , (14) k k | ∀ 2 c 2π a† a = cos k(m n) . (18) and, therefore, it has 0 energy. m n N N − On the other hand, if 2 > λ 0, the ground state is the k =0 ≥ † state annihilated by the operators bk with Λk > 0 and bm with In the thermodynamic limit, the previous sum becomes an in- Λm < 0, tegral and it can be determined analytically. In this case, the correlation matrix of the block of L fermions in position space b GS =0 if Λ > 0 k| k is, † bm GS =0 if Λm < 0 , (15) | † 1 sin kc(m n) Amn = aman = − , (19) and its energy is simply Λ Λ < 0. InFig.1 andEq. π m n m m ∀ m − (13), we can see that if kc k 0 or N 1 k N kc, where L m,n 0. Notice that, by means of Wick’s the- ≥ ≥ − ≥ ≥ − ≥ ≥ where kc is defined by orem, any operator that acts on the block can be written in terms of the correlation matrix Amn. For instance, N λ kc = arccos , (16) † † † † † † 2π 2 akal aman = akan al am akam al an . (20) − This is due to the fact that the system is Gaussian, and its then Λk < 0, whereas for the rest of cases Λk 0. In Eq. (16), the brackets [] represent the floor function. ≥ eigenstates are determined by the first and second moments of some fermionic operators. The correlation matrix Amn could also be computed using 2 the density matrix of the block ρL, 1.5 Amn = Tr(amanρL) . (21) λ= 1
0.5 We, thus, need to invert the previous equation, that is, to com-
k/N) pute the density matrix ρ from the correlation matrix A . π L mn 0 The matrix Amn is Hermitian, so it can be diagonalized by -0.5 a unitary transformation, 2 cos(2 -1 G = u A u∗ = g†g δ , (22) pq pm mn nq p p pq -1.5 m,n=0 -2 0 kc N-kc N-1 where gp = m upmam. In this case, the density matrix of k the block must also verify † FIG. 1: The two terms of Λk, Eq. (13), are plotted for the particular G = Tr(g g ρ )= ν δ , (23) 2πk mn m n L m mn case λ = 1. We realize that if 2 cos ` N ´ > λ, Λk < 0 while if 2πk which implies that is uncorrelated and it can be written as 2 cos ` N ´ <λ, Λk > 0. ρL ρ = ̺ ̺ , (24) L 1 ⊗ ⊗ L 5 where ̺m is the density matrix corresponding to the m-th Finally, let us mention that the computation of the geomet- † fermionic mode excited by gm. ric entropy of Gaussian systems have been systematized in In order to determine the eigenvalues of the density matrix Refs. [16]. In particular, it is shown that for solvablefermionic † of one mode, let us express the gm, gm and ̺m operators in and bosonic lattice systems, the reduced density matrices can its matrix representation. That is, be determined from the properties of the correlation functions. This subject is reviewed in Ref. [17] in this special issue. 0 0 0 1 g = g† = , (25) m 1 0 m 0 0 and E. Scaling of the entropy
αm βm ̺m = ∗ , (26) It is now easy to compute the entanglement entropy of the β 1 αm m − ground state of the XX model for arbitrary values of the block where αm and βm are the matrix elements of ̺m that we want size L and magnetic field λ. to determine. It is easy to see that βm =0, since g = Tr(g ρ )= β =0 . (27) 3.5 m m L m Moreover, rewriting Eq. 23 in terms of these matrices 3
† 1 0 αm 0 2.5 Tr(gmgmρL)= Tr = νm 0 0 0 1 αm − 2
(28) S(L) we realize that αm = νm. The entanglement entropy between the block and the rest 1.5 of the system is therefore, 1 λ=0 L λ=1.9 0.5 SL = H2 (νl) . (29) 0 50 100 150 200 250 l=1 L where H2(x) = x log x (1 x) log(1 x) denotes the − − − − FIG. 2: Entropy of the reduced density matrix of L spins for the XX binary entropy. → ∞ Summing up, the three steps that we have to follow in order model in the limit N , for two different values of the external magnetic field λ. The maximum entropy is reached when there is to compute the entanglement entropy of the GS of a block of no applied external field (λ = 0). The entropy decreases while the L spins for the XX model are: (i) to determine the correlation magnetic field increases until λ = 2 when the system reaches the matrix A by evaluating Eq. (19) for L m,n 0,(ii) to mn ≥ ≥ ferromagnetic limit and the ground state is a product state in the spin diagonalize this correlation matrix and, with its eigenvalues, basis. (iii) to compute the entanglement entropy according to Eq. (29). In Fig. 2, we show how the entropy of the reduced density Let us emphasize that this method is computationally effi- matrix of a block of L spins varies with L for different values cient, since its computational cost scales polynomially with of the magnetic field. The maximum entropy is reached for 3 the number of spins of the block O(L ), whereas the Hilbert λ = 0. In particular, we recover the result in Ref. [2] and see N space of the problem has dimension 2 . that for λ = 0 the leading scaling of the entropy is perfectly It is also necessary to recall a quite subtle point that we have fitted by a logarithm, skipped along our previous discussion. It turns out that there is no need to perform a final transformationback to spins, that 1 S = log L + a , (31) is, there is no need to invertthe Jordan-Wignertransformation. L 3 2 This is due to the fact that the coefficients of a given state are identical when written in terms of the spin basis or in terms of where a is a constant that was determined analytically in Ref. [18]. the fermionic al operators. More precisely, As we increase the magnetic field, but it is still less than ψ = Ci1,i2,...,in i ,i ,...,i (30) 2, the entropy decreases although it keeps its logarithmic be- | | 1 2 n i ,i ,...,i haviour with L. When λ > 2, the entropy saturates to zero, 1 2 n i1,i2,...,in † i1 † i2 † in since the ground state is already in a product state in the spin = C (a1) (a2) . . . (an) vac . | basis corresponding to the ferromagnetic phase, i i. i1,i2,...,in |↑ The relation between logarithmic scaling and entropy is Thus, the same coefficients appear in the ket, either when writ- confirmed by similar computations in different models. The ten in the initial spin basis, or when expressed as creation op- general result is that entanglement entropy obeys a logarith- erators in the fermionic vacuum, vac . Consequently, the re- mic scaling relation at critical points, that is when the system duced density matrix entropy of entanglement| is identical for is at a phase transition, whereas a saturation of entanglement both expressions. is found away from criticality. This universal logarithmic at 6 critical points must emerge from the basic symmetry that char- Notice that both the correlation matrix Amn, defined in Eq. acterizes phase transitions, namely conformal invariance. We (19), and A˜ are Toeplitz matrices, that is to say, matrices in shall come to this developments in the next section. which each descending diagonal from left to right is constant, Let us mention that the scaling of entanglement entropy was formerly studied in the context of quantum field theory f0 f−1 ... f1−L and black-hole physics. There, the system sits in higher di- . f f . mensions. The entanglement entropy scales following an area A = 1 0 , (35) law that we shall discuss later on. Yet, it is important to note . . . . .. . that one-dimensionalsystems are an exception to the area law. f ...... f Entanglement pervades the system at any distance, not staying L−1 0 just at the point-like borders of a block. 1 sin kcm Summing up, we have seen that the scaling entropy is a where, in this case, fm = π m . good witness for quantum phase transitions. Many other stud- The asymptotic behaviour (when L ) of the deter- ies of different measures of entanglement at quantum phase minant of Toeplitz matrices has been widely→ ∞ studied in many transitions have been presented recently. Let us here mention cases, giving the famous Fisher-Hartwig conjecture (see Refs. that in Refs. [19, 20], quantum phase transitions are charac- [21, 22, 23, 24, 25]). In our particular case, the expression for terized in terms of the overlap (fidelity) function between two the determinant of A˜ was proven in Ref. [23] and, therefore, it ground states obtained for two close values of external param- is a theoreminstead of a conjecture. In this way, we may insert eters. When crossing the critical point a peak of the fidelity is this result in Eq. (34), perform the corresponding complex in- observed. tegral and obtain the asymptotic analytical expression for the entanglement entropy of the XX model. This computation is presented explicitly in Ref. [18] with the final result, F. Entanglement entropy and Toeplitz determinant 1 1 λ 2 ln 2 Before finishing this section, we would like to sketch how SL = lnL+ ln 1 + +Υ1, (36) 3 6 − 2 3 the particular structure of the correlation matrix in Eq. (19) allows us to derive an analytical expression for the scaling law of the entanglement entropy. This result is presented in Ref. where [18]. ∞ In order to obtain an analytical expression for the entangle- e−t 1 cosh(t/2) Υ1 = dt + 2 3 . ment entropy, let us introduce the function − 0 3t t sinh (t/2) − 2 sinh (t/2) (37) DL( ) = det A˜( ) , (32) Indeed, we realize that this analytical expression for the scal- ing of the entanglement entropy is compatible with the nu- merical fit of Eq. (31) and, moreover, it fixes the value of the where A˜( ) IL A, IL is the identity matrix of dimension L, and A is the≡ correlation− matrix defined in Eq. (19). If we independent term. express the matrix A in its diagonal form, we trivially have This procedure is also used to obtain an analytical expres- sion for the entanglement entropy of the XY model in Refs. L [26]. In Ref. [27], the scaling of the Renyi entropy is deter- DL( )= ( νm), (33) mined for the XY model in terms of Klein’s elliptic λ - func- − m=1 tion showing a perfect agreement with the previous results in the particular case in which the Renyi entropy becomes the where ν , with m = 1,...,L, are the eigenvalues of A. m von Neumann entropy. Then, according to Cauchy residue theorem, the entanglement entropy SL can be expressed in terms of an integral in the - complex plane as follows 1 III. SCALING OF ENTANGLEMENT SL = lim lim e(1 + ǫ, )d ln DL( ) ǫ→0+ δ→0+ 2πi c(ǫ,δ) L The logarithmic scaling law that entanglement entropy obeys in the critical regime is a sign of the conformal sym- = H2(νm) , (34) metry of the system. For second order phase transitions, the m=1 correlation length diverges and the system becomes scale in- where c(ǫ,δ) is a closed path that encircles all zeros of DL( ) variant. This scaling symmetry gets enlarged to the conformal and e(1 + ǫ, ) is an arbitrary function that is analytic in the group [28] which, in the case of on-dimensional systems, al- contour c(ǫ,δ) and verifies e(1, νm)= H2(νm) m. lows for a very precise characterization of the operator struc- Thus, if we could obtain an analytical expression∀ for the ture of the underlying theory. The development of confor- DL( ) function, we would be able to get a closed analytical mal field theory is a remarkable achievement that we cannot result for the entanglement entropy. present in this short review [29, 30]. 7
A. One-dimensional systems present in any theory. Let us now see that both ideas merge naturally. For a conformal theory in 1+1 dimensions, the scaling be- In 1+1 dimensions, conformal field theories are classified haviour of the entropy was proven to be logarithmic in Ref. by the representations of the conformal group [29]. The oper- [31]. The general result reads ators of the theory fall into a structure of highest weight oper- ators and its descendants. Each highest weight operator car- c +¯c ries some specific scaling dimensions which dictates those of S log L , (38) L ∼ 6 2 its descendants. The operators close an algebra implemented into the operator product expansion. One operator is particu- where c and c¯ are the so called central charges for the holo- larly important: the energy-momentum tensor T , which is a morphic and anti-holomorphic sectors of the conformal field ν descendant of the identity. It is convenient to introduce holo- theory. These central charges classify conformal field theo- morphic and anti-holomorphic indices defined by the combi- ries and are universal quantities which depend only on basic nations T = T and T¯ = T where z = x0 + ix1 and properties of the system, like effective degrees of freedom of zz z¯z¯ z¯ = x0 ix1. Denoting by 0 the vacuum state, the central the theory, symmetries or spatial dimensions, and they are in- charge of− a conformal field theory| is associated to the coeffi- dependent of the microscopic details of the model. For free cient of the correlator bosons c =1, whereas for free fermions c =1/2. c This result matches perfectly our geometric entropy com- 0 T (z)T (0) 0 = , (42) putation of the critical XX model. In this case, the central | | 2z4 charge c =c ¯ = 1 and the model is seen to belong to the free and the analogous result for c¯ in terms of the correlator boson universality class. 0 T¯(z)T¯(0) 0 . A conformal field theory is characterized The previous result of Eq. (38) was further elaborated and by| its central| charge, the scaling dimensions and the coeffi- extended to finite systems, finite temperature and disjoint re- cients of the operator product expansion. Furthermore, uni- gions in Ref. [31, 32, 33]. For instance, the scaling of the tary theories with c < 1 only exist for discrete values of c entropy for a system with periodic boundary conditions reads and are called minimal models. The lowest lying theory cor- responds to c = 1 and represents the universality class of a c L πℓ 2 S log sin + c′ , (39) free fermion. A ∼ 3 πa L 1 The central charge plays many roles in a conformal field whereas for the open boundary conditions case is theory. It was introduced above as the coefficient of a corre- lator of energy-momentum tensors, which means that it is an c 2L πℓ observable. The central charge also characterizes the response S = log sin +˜c′ . (40) A 6 πa L 1 of a theory to a modification of the background metric where it is defined. Specifically, the scale anomaly associated to the In Ref. [34], the scaling of the entropy of a conformal semi- lack of scale invariance produced by a non-trivial background infinite chain is presented. In Ref. [35], conformal symmetry metric is is further exploited and an analytical computationof the distri- c 0 T 0 = R , (43) bution of eigenvalues of the reduced density matrix of a block | | −12 in a one-dimensional systems is presented. Let us finally mention that the scaling of entanglement have where R is the curvature of the background metric. This been also studied for other entanglement measures by means anomaly can also be seen as the emergence of a non-local ef- of conformal field theory. In particular, in Ref. [36, 37], it is fective action when the field theory modes are integrated out shown that the single copy entanglement scales as in a curved background. Therefore, the central charge which appears as the coeffi- c c π2 cient of the entanglement entropy is naturally related to the E (ρ )= log L + O(1/L) . (41) 1 L 6 − 6 log L stress tensor, which is the operator that is guaranteed to exist in any theory. It is also possible to derive a direct relation be- Note that entropy sub-leading corrections to scaling are sup- tween entropy and the trace of the stress tensor as shown in pressed as 1/L whereas single copy entanglement suffers the original Ref. [31]. from 1/ log L modifications. This makes the numerical ap- proach to the latter more difficult. C. Area law
B. Conformal field theory and central charge The conformal group does not constrain the structure of the Hilbert space in spatial dimensions higher than one as much We stated above that the central charge is a quantity that as it does in one dimension. Actually, the conformal group no characterizes the universality class of a quantum phase tran- longer brings an infinite number of conserved charges (as it sition. We also mentioned in the introduction that a possible does in one dimension) but becomes a finite group. figure of merit for entanglement could be constructed from Nevertheless, a geometric argument establishes the scaling correlation functions made out some operator which is always behaviour of entropy. The basic idea goes as follows. Let us 8 consider a volume of spins (or any local degrees of freedom) logarithmic scaling in one-dimension. The system is more contained in a larger space. For theories with local interac- correlated than what is expected from pure geometrical argu- tions, it is expected that entanglement will be created between ments. In this respect, the leading term in the scaling of the the degrees of freedom that lie outside and inside the surface entropy for fermionic systems was computed analytically as- that encloses the volume we are considering. It follows that suming the Widom conjecture in Ref. [49]. This result was the entropy should naturally scale as an area law even if the checked numerically for two critical fermionic models in Ref. model displays a finite correlation length. [46] finding a good agreement. These arguments were put forward in the study of entan- For other steps into a description of systems with two spa- glement entropy in quantum field theory as a possible source tial dimensions in the framework of conformalfield theory see for black-hole entropy [38, 39, 40]. Furthermore, the rela- Refs. [50, 51]. For a class of critical models in two spatial di- tion between the entropy and the effective action in a curved mensions (including the quantum dimer model), it is found background was developed in Ref. [41]. Let us mention these that S(ρI )=2fs(L/a)+ cg log(L/a)+ O(1), where L is results. For general quantum field theories the entropy is a di- the length of the boundary area, fs is an area law coefficient vergent extensive quantity in more than one spatial dimension that is interpreted as a boundary free energy, and g is a co- obeying an area law efficient that depends on the geometric properties of the par- tition. That is, in addition to a non-universal area law, one L d−1 S c d> 1 , (44) finds a universal logarithmically divergent correction. For a L ∼ 1 ǫ further discussion of steps towards a full theory of entangle- ment entropies in d +1-dimensional conformal field theories, d where L is the size of the volume, ǫ stands for an ultraviolet see Refs. [50, 51]. regulator and the coefficient c1 counts components of the field A particularly interesting issue is the holographic entangle- which is considered. This coefficient is again found in the ment entropy that emerges from the AdS (anti-de-Sitter)/CFT effective action on a gravitational field and, thus, in the trace correspondence. The AdS/CFT correspondence is the conjec- anomaly as a divergent term. A form for the former can be tured equivalence between a quantum gravity theory defined found as on one space, and a quantum field theory without gravity de- ∞ −sm2 fined on the conformal boundary of this space. The entan- e c0 Γ = ds + c R + c sF + c sG + . . . , glement entropy of a region of the boundary in the conformal eff sd/2 s 1 2F 2G s0 field theory is then related with the degrees of freedom of part (45) of the AdS space in the dual gravity side. In Refs. [51, 52], where s acts as a ultraviolet regulator, R is the curvature, F 0 this relation is established explicitly and, in Ref. [53] in this the Weyl tensor and G the Euler density. The main concep- special issue, the recent progress on this topic is presented. tual result to be retained is that entropy measures a very basic counting of degrees of freedom. Note that previous efforts to Let us also mentionthe line of research that deals with topo- logical entropy. Some Hamiltonians produce states such thata make a general c-theorem are all base on c an c , not on 2F 2G combination of geometric entropies exactly cancels the domi- c1. In one spatial dimension, the effective action has a unique structure proportional to the central charge. That is, the cen- nant area law. Then, a topological entropy term characterizes tral charge takes over all manifestations of the trace anomaly, the system [54]. This subject is nowadays a large field of re- at variance with the separate roles that appear in higher di- search that we cannot include in the present review. In this mensions. respect, a review on the scaling of the entanglement entropy It is worth mentioning that computations done in massive of 2D quantum systems in a state with topological order is presented in Ref. [55] in this special issue. theories in any number of dimensions show that S (m = L 0) SL(m = 0) comes out to be ultraviolet finite [42]. Actu- ally,− the ultraviolet cut-off cancels in the computation. This is precisely what is needed to make the RG flow meaningful in IV. OTHER MODELS such a case. This comment hints at the non-trivial issue about observability of the entropy. The standard prejudice is that the We can find in the literature the computation of the scaling leading area law coefficient is not observable since it comes of the entanglement entropy for other spin models. In XY and divided by a necessary ultraviolet cut-off. Yet, if such a co- XXZ models, this logarithmic scaling will confirm the role efficient is also responsible for finite corrections, the situation of the underlying conformal symmetry. We shall also discuss may not be as trivial. that in disordered systems, although the conformal symmetry A review on methods to calculate the entanglement entropy is lost for one particular realization of the disorder, we recover for free fields and some particular examples in two, three and the logarithmic scaling of the entropy with a different central more dimensions are presented in Ref. [43] in this special is- charge of the corresponding homogeneous model, if we do sue. Further explicit computations of area law scaling of en- the average over all the possible realizations of the disorder. tropy in spin and harmonic systems in higher dimensions can We shall also study the scaling of entropy in systems where be found in Refs. [44, 45, 46, 47]. A quite remarkable re- the notion of geometric distance is lost. This is the case of sult found in Ref. [48] is the fact that certain fermionic sys- the Lipkin-Meshkov-Glick model, in which the logarithmic tems may develop logarithmicviolations of the area law, while behaviour of the entropy will be due to the equilibrium of a keeping local interactions. This is somehow analogous of the competition between the long range interactions, that try to in- 9 crease the entanglement, and the symmetries of the problem, known results between the two models and obtain, among oth- that force the ground state to belong to a reduced subspace ers, the additive constant of the entropy of the critical homo- of the Hilbert space. A different case are those systems com- geneous quantum Ising chain and the effective central charge posed of itinerant particles. In particular, we will present the of the random XY chain. scaling of entropy of the Laughlin wave function. Finally, with respect to the particular case of the Ising model, in Ref. [60], the computation of the leading correc- tion to the bipartite entanglement entropy at large sub-system A. TheXYmodel size, in integrable quantum field theories with diagonal scat- tering matrices is presented. This result is used to compute the The XY model is defined as the XX model in Eq. (7), exact value of the saturation in the Ising model and showed it adding an extra parameter γ that determines the degree of to be in good agreement with numerical results. This work is anisotropy of spin-spin interaction in the XY plane. Its Hamil- reviewed in detail in Ref. [61] in this special issue. tonian reads
1 1+γ 1 γ B. TheXXZmodel H = σxσx + − σyσy + λσz , XY −2 2 l l+1 2 l l+1 l l (46) The XXZ model consists of a chain of N spins with near- est neighbour interactions and an external magnetic field. Its where, as in the previous section, l labels the N spins, σ l Hamiltonian is given by, ( = x,y,z) are the Pauli matrices and λ is the transverse magnetic field in the z direction. This notation will be also 1 H = [σxσx + σyσy + ∆σzσz ]+ λσz , followed for the rest of models that are going to be presented. XXZ 2 l l+1 l l+1 l l+1 l Notice that if γ = 0 we recover the XX model, whereas if l (49) γ =1, it becomes the quantum Ising model with a transverse where ∆ is a parameter that controls the anisotropy in the z magnetic field, with Hamiltonian direction. As it happened for the γ parameter of the XY model, the 1 x x z HIsing = σ σ + λσ . (47) ∆ parameter of the XXZ model has two particular interesting −2 l l+1 l l values. If ∆=0, we trivially recover the XX model, and if ∆=1, the system becomes the XXX model that has a fully The XY model was solved in detail in Ref. [2]. In order to isotropic interaction do this, the previousworks on spin chains requiredto solve the XY Hamiltonian were reviewed. In concrete, the XY model 1 H = [σxσx + σyσy + σzσz ]+ λσz . without magnetic field was solved exactly in Ref. [56], the XXX 2 l l+1 l l+1 l l+1 l l spectrum of the XY model with magnetic field was computed (50) in Ref. [57], the correlation function for this model was ob- The XXZ model can be solved analytically by means of tained in Ref. [58], and the entropy SL was computed in Ref. the Bethe Ansatz technique [62]. Bethe Ansatz takes profit [1]. of the two symmetries of the system to find its eigenstates. Later, an extent analytical analysis of the entropy of XY The first symmetry is the rotational symmetry respect to the spin chain was presented in Ref. [26]. In this work, in a sim- z axis. It implies that the z-component of the total spin, ilar way as we have seen previously for the XX model, an z Sz = 1/2 σ , must be conserved and, therefore, the analytical expression for the scaling of the entanglement en- l l Hamiltonian must be diagonal in boxes of constant Sz. The tropy is determined for the XY model by means of Toeplitz other symmetry is the translational invariance, that allow us determinants. to diagonalize these boxes using a kind of generalized Fourier The XY model with γ = 0 is critical for λ = 1. In this transform. Once the ground state is obtained, the correlation case, the entropy of a block scales as functions can be computed in terms of certain determinants (see Ref. [63]). This model is qualitatively different from the 1 XY, since it presents a point of non-analyticity of the ground SXY (L)= log2 L + a(γ), (48) 6 state energy for finite systems. In the XY model, the level crossing between the ground state and the first excited state where a(γ) is a function that only depends on γ. This en- tanglement behaviour corresponds to the scaling dictated bya only occurs in the thermodynamic limit. In this case, instead, the terms of the Hamiltonian σxσx + σyσy , σzσz and conformal theory, Eq. (38), with central charge c =1/2. The l l+1 l l+1 l l+1 z commute and are independent of and , which implies XY model, therefore, falls into the free fermion universality σl ∆ λ class. that there will be an actual level crossing. Both the isotropic case and the anisotropic one for λ = 0 In the non-critical case, that is for λ =1, the entanglement are solved in Refs. [2, 64]. The phases of the system are found entropy saturates to a constant. to be: Let us mention that an exact relationship between the en- tropies of the XY model and the XX model has been found In the XXX model, Eq. (50), there are two limit be- recently [59]. Using this relation it is possible to translate • haviours. On one hand, when λ > 2 the system is | | 10
gapped and it is in a product state in which all spins the previous models) and then perform perturbation theory re- point at the direction of the magnetic field (ferromag- spect to the neighbour couplings. The final result is that two netic phase). On the other hand, for λ = 0 the mag- sites have been eliminated and the Hamiltonian energy scale netization is zero and the system is in a entangled state has been reduced. This process can be iterated until we arrive (anti-ferromagnetic case). In the interval between these at the ground state of the system which is a random singlet two cases 2 > λ 0 the system is gap-less and, there- phase, that is to say, a set of singlets connected randomly and fore, critical. ≥ for arbitrarily long distances. Notice that although this method is not correct when ap- With respect to the anisotropic case with magnetic field • plied to a system with weak disorder, it becomes asymptoti- equal to zero, the system shows a gap-less phase in the cally correct at large distances [69]. 1 ∆ 1 interval. Outside this interval, there is For a particular realization of the disorder, the translational a gap≥ between≥ − the ground and the first excited states. symmetry of the system is broken and, therefore, the confor- These two phases are separated by two phase transitions mal symmetry too. Hence, the scaling of the entanglement in ∆=1 and ∆ = 1. The first one is a Kosterlitz- entropy of this realization of the disorder will not be logarith- Thouless phase transition,− while the second one is of mic, but fluctuating. first order. In Ref. [68], it was shown that although the conformal sym- The scaling of the entanglement entropy is presented in metry is broken, if we take the average over all the realizations Refs. [2, 65]. These numerical results show that the entan- of the disorder the entropy keeps scaling logarithmically with glement entropy behaviour converges to a logarithmic scaling an effective central charged c˜ = c ln 2, where c is the central as the size of the system increases, if the system is critical. On charge for the same model but without disorder. This result the contrary,if the system is not in a critical phase, the entropy has been further checked numerically both for the XX model saturates to a constant value. In particular, in the isotropic in Ref. [71] and for the Heisenberg model in Ref. [72]. 1 model without magnetic field, the entropy scales as In Ref. [71], the disordered XX spin- 2 chain with periodic boundary conditions and positive random spin couplings cho- 1 sen in a flat uniform distribution within the interval [0, 1] was SL log2 L , (51) ∼ 3 studied. The magnetic field was set to zero. It was shown which means that the XXX model with λ = 0 has central that for a block large enough (larger than 20 spins), the en- charge c = 1 and falls into the universality class of a free tropy scales logarithmically according to [68], using around boson. 104 samples for N = 500, 1000, 2000 and 2 104 samples × Finally, let us mention that, recently, analytic expressions for 100 N 400, in order to do the average over all the ≤ ≤ for reduced density matrices, several correlation functions and possible realizations of the disorder. the entanglement entropy of small blocks (up to 6 spins) have The same result was shown for the Heisenberg model in been found for the XXZ model with ∆=1/2 (see Refs. [66] Ref. [72]. In this work, a uniform distribution in the inter- and [67]). val [0, 1] for the couplings between the spins was also chosen. For a system of N = 50 and after averaging the entanglement entropy over 104 different configurations of disorder, the loga- C. Disordered models rithmic scaling of the entropy with an effective central charge c˜ = c ln 2 is recovered. Let us point out that these one di- So far, we have only considered translational invariant sys- mensional systems are particular cases of chains of quantum tems. This symmetry plus the scaling invariance at a critical group (or q-deformed) spins studied in Ref. [73]. It is also in- point produces the conformal symmetry that implies universal teresting to mention that this robustness of the entanglement properties of the scaling of entanglement. Nevertheless, natu- scaling respect to the disorder is not kept for other models like ral systems exhibit a certain amount of disorder due to impuri- the Bose-Hubbard model (see Ref. [74]). ties and imperfections of the system. This disorder breaks the In the case of higher dimensions, the scaling of the entan- translational symmetry and we wonder what happens with the glement entropy in the 2D random Ising model was studied scaling of the entropy taking into account that the conformal in Refs. [75, 76]. In particular, in Ref. [76], the entanglement invariance is lost. entropy of a L L region located in the centre of a square × This question was addressed in Ref. [68]. They computed lattice which is governed by the Hamiltonian analytically the block entropy for the Heisenberg, XX and quantum Ising models with random nearest-neighbour cou- H = J σzσz λ σx , (52) − ij i j − i i plings under the hypothesis of strong disorder by means of the i,j i real space renormalization group technique. This approach was introduced in Ref. [69] and was generalized in Ref. [70]. was computed. The Ising couplings Jij and the transverse This strong disorder hypothesis assumes that if one takes magnetic fields λi take random values given by the uniform the strongest coupling of the chain, its neighbours are much probability distributions in the intervals [0, 1] and [0, λ0] re- weaker than it. Thus, it is possible to diagonalize this spectively. By means of a generalized version for 2 dimen- strongest bond independentlyof the rest of the system, project sions of the real space renormalization group, it was found the system onto the ground state of this subspace (a singlet for that the critical field is at λc =5.37 0.03, and for both criti- 0 ± 11 cal and non-critical λ0 the entropy scaling fulfils the area law: in terms of a basis N/2,M fully symmetric under the per- | 2 S(L) L in the leading term. mutation group and eigenstates of S and Sz. These states Let∼ us mention some disordered spin systems have also N/2,M are called Dicke states. been studied from the fidelity susceptibility point of view in | Notice that the restricted subspace where the ground state Ref. [77]. Finally, it is interesting to point out that, in other must live due to the symmetries of the Hamiltonian will systems, the translational invariance is not broken by means limit the scaling of the entanglement entropy of a block of of random couplings but due to a quantum impurity or a phys- L spins with respect to the remaining NL. As the ground ical boundary. The behaviour of the entanglement entropy in state reduced density matrix is spanned by the set of (L + 1) this kind of systems is reviewed in Ref. [78] in this special Dicke states, the entropy of entanglement obeys the constrain issue. SL,N log2(L + 1) for all L and N, where the upper bound≤ corresponds to the entropy of the maximally mixed state ρL,N = 1l/(L + 1) in the Dicke basis. This argument D. The Lipkin-Meshkov-Glick model implies that entanglement, as measured by the Von Neumann entropy, cannot grow faster than the typical logarithmic scal- The Lipkin-Meshkov-Glick model was proposed in Ref. ing observed in the previous cases. [79]. Unlike the previous models we have considered, where Both the ground state and the entanglement entropy were the spins had short rangeinteractions, in the LMG model, each computed for the LMG model in Ref. [80]. For the isotropic spin interacts with all the spins of the system with the same case (γ = 1) and in the thermodynamic limit (N,L 1), ≫ coupling strength. This system is described by the Hamilto- HLMG is diagonal in the Dicke basis. Then, for λ 1, the ≥ nian entanglement entropy is strictly zero since the ground state is in a fully polarized product state. Instead, if 1 > λ 0, we 1 ≥ H = σxσx + γσyσy λ σz . (53) recover the logarithmic scaling of the entropy, LMG −N i j i j − i i
Although the behaviour of entanglement is very similar to where zj = xj + iyj, j = 1,...,n stands for the position of the XY model, that is to say, it scales logarithmically in the the j-th particle. It describes fractional quantum Hall state at critical point and saturates to a constant in the non-critical a filling fraction ν =1/m, where m is an integer number. phase, the reasons of these scaling laws are different. In the In particular, in Ref. [87], the entanglement entropy of k XY model, entanglement is limited by the facts that interac- particles respect the rest of the system is computed for the tions are local and the system is translationally invariant. At Laughlin wave function with filling fraction one the critical point, the correlation length becomes infinite, the system is conformal symmetric, and the logarithmic scaling n S = log . (59) of the entanglement entropy appears as a manifestation of this n,k 2 k symmetry. Instead, in the LMG model, the long range in- teractions should allow for larger correlations, that is, larger Notice that, in this case, although the state also belongs to entanglement. Nevertheless, the symmetries of the system re- a completely anti-symmetric subspace, the entanglement en- strict the subspace where the GS must belong and, therefore, tropy of half a system grows linearly with the number of par- the scaling law of entanglement. The final result is the same ticles. logarithmic scaling law but which has nothing to do with any In Ref. [88], these ideas are extended, and the particle en- underlying conformal symmetry. tanglement, defined as the entanglement between two subsets Finally, let us mention that other analytical calculations of of particles making up the system, is studied. The general the spectrum of the LMG model both in the thermodynamic structure of particle entanglement in many-fermion ground limit and finite size case have appeared recently [82]. More- states, analogous to the area law for the more usually stud- over, the entanglement entropy for general free bosonic two- ied entanglementbetween spatial regions, are also formulated, mode models is presented in Ref. [83]. In particular, a com- and the basic properties of particle entanglement are uncov- plete classification of the possible scaling behaviours for the ered by considering relatively simple itinerant models. All entanglement entropy in the related collective models as the these ideas are widely reviewed in Ref. [89] in this special LMG, the Dicke model, or the Lieb-Mattis model is obtained. issue.
E. Particle entanglement V. RENORMALIZATION OF ENTANGLEMENT
In a similar way than LMG model, where the notion of A natural question arising within the study of entanglement distance was lost, one can try to compute the entanglement in quantum system is how entanglement evolves along Renor- entropy in systems of moving fermions and bosons. In such malization Group (RG) trajectories. We shall now address this itinerant systems, as the particles are indistinguishable, mov- issue discussing first the RG of quantum states and, then, the ing and partially de-localized, it is not obvious to define the study of particular systems. geometric entropy. What we, indeed, can compute is the von Neumann entropy for any subset of particles for a system of N indistinguishable A. Renormalization of quantum states particles in the state Ψ(r1,...,rn). Notice that, in this case, this von Neumann entropy cannot be interpreted as the num- ber of distillable EPR pairs. Due to the symmetrization, it is It is customary to present RG transformations on Hamil- impossible to associate a label with the particles and perform tonians or observables. In general, a Hamiltonian is de- the appropriate distillation operations. This is a subtle differ- scribed by a set of coupling constants times operators H = i . This set of operators may be infinite, including ence respect to the LMG model. i g i relevant,O marginal and irrelevant operators or, as in the case A particular interesting physical system is the Fractional of renormalizable quantum field theories in the continuum, Quantum Hall Effect (FQHE) [84]. Although a complete un- it may reduce to a finite set of relevant and marginal oper- derstanding of it is still missing, it is commonly believed that ators. Then, upon coarse graining of short-distances and an the interactions between the particles are essentially respon- adequate rescaling, the system is described by a new set of sible for the strange states of matter that the 2D electron gas coupling constants. So to speak, the operator algebra acts asa shows at some particular values of the transverse magnetic basis. The concept of RG trajectory corresponds to analysing field. These states would present a new kind of order called d d observables along the flow = β i , wherethe beta func- topological order and their quasi-particle excitations are nei- dt − i dg dgi ther bosons nor fermions, but anyons, that is to say, quasi- tions correspond to βi dt are related to the change of the particles with any-statistics [85]. In this respect, in 1983, coupling constants as the≡ coarse graining proceeds. Laughlin proposed an Ansatz for the wave function of the Yet, RG transformations can be understood as an action on ground state of the system [86]. This wave function is defined any quantum state, regardless of its relation to any Hamilto- by nian [90]. Coarse graining is independent of any dynamics. This RG procedureon quantumstates is not presented as com- (m) m − N |z |2/2 Ψ (z ,...,z ) (z z ) e Pi=1 i , (58) mon lore since explicit knowledge of e. g. the ground state of L 1 n ∼ i − j i The basic idea to perform RG on states is to produce a 1.5 coarse graining of short-distance degrees of freedom, fol- 1.25 lowed by a clever choice of local basis to retain the long- 1 L distance information which is retained in an optimal way. Let Λ H L 0.75 us take a quantum state ψ0 and determine its RG transformed S ′ 0.5 , ψ0, as follows. We pairwise group the sites in the system and define a coarse-graining transformation for every pair of local 0.25 d-dimensional basis states, e.g. for the sites 2j and 2j +1, 0 0 1 2 3 4 5 as p 2j q 2j+1 = pq j . This transformation yields ψ0 ψ. Λ Then| we| have ψ′ =| U . . . U ψ , where the d2 d2 unitary→ 0 ⊗ ⊗ | × matrix U performs the change of representative in the coarse– FIG. 3: Entropy of entanglement is shown to decrease monotonically grained space. Note that the matrix U is non-local as seen along the RG trajectory that takes the external magnetic field λ away ∗ from the 2j and 2j +1 sites. Some local information is now from its critical value λ = 1. Towards the left the flow takes the washed out, while preserving all the quantum correlations re- system to a GHZ-line state whereas, towards the right, the system is lating the coarse–grained block to other ones. a product state. Operators also get coarse–grained along the above trans- formation. Take for instance an operator acting on one local tensor. This was indeed done in Ref. [92]. Once, the rela- Hilbert space, e.g. O2j . Expectation values must remain un- changed, tion between entropy and the properties of the stress tensor are made apparent. ψ O ψ = ψ′ O′ ψ′ , (60) 0| 2j | 0 0| j | 0 which leads to VI. DYNAMICS OF ENTANGLEMENT ′ † Ol = U(O2j I2j+1)U , (61) ⊗ So far, we have studied the properties of entanglement en- where I is the identity matrix. To complete a RG transforma- tropy of the ground state of the system. Next, we would like to tion we simply need to rescale distances, i.e., to double the analyse how entanglement evolves in time when the system is lattice spacing. prepared in a state that is not an eigenstate of the Hamiltonian. This analysis can be made completely explicit in the case of states which are described as a matrix product state [90]. There, the above transformation amount to a flow on the ma- A. Time evolution of the block entanglement entropy trices that represent the state. In turn, a flow related to the transfer matrix can be computed. Explicit irreversibility of In Ref. [93], the time evolution of the entropy of entan- RG flows and the characterization of critical points followed glement of a block of L spins in a one-dimensional system is from the flow on this transfer matrix. studied. It is considereda system preparedin a pure state ψ , | 0 which corresponds to an eigenstate of H(λ0) with λ0 = λ. Then, for example, at time t = 0, the parameter is suddenly B. Irreversibility of RG flows quenched from λ0 to λ. In general, ψ0 will not be an eigen- state of H(λ), and thus the system| will evolve according to We may as well return to the standard construction of RG the equations of motion given by H(λ). In this work, two transformations on Hamiltonians and perform a detailed study computations are performed: one based on conformal field in some particular case. For instance, we may consider the theory and the other on a particular solvable spin model, the quantum Ising model in a transverse field λ. It is known that Ising model. In the first case, the path integral formulation the parameter λ provides a relevant deformation of the model, and the CFT are used in order to calculate the time evolution ∗ departing form its critical value λ =1. For instance, the de- of the entanglement entropy of a high energy state of the sys- parture that makes λ> 1 get larger and larger corresponds to tem which is not an eigenstate. Then, one has to assume that the increase of the mass of the underlying fermionic descrip- the Hamiltonian is critical in order to make the theory confor- tion. mally invariant. Instead, in the Ising model case, it is possible An analysis of this RG trajectory can be illustrated using to perform calculations starting from a variety of initial states, Fig. 3 ( see Refs. [32, 91]). This result shows that RG tra- considering both critical and non-critical regimes. jectories are monotonically irreversible as dictated by the c- In both calculations, the entanglement entropy increases theorem in 1+1 dimensions. Furthermore, it can be seen that linearly with time t (after transients die away in the lattice the ground state obeys majorization relations. That is, irre- case), up to t∗ = L/2, in units where the maximum propa- versibility is orchestrated at a very refined level, since the gation speed of excitations is taken to be unity. For t t∗, reshuffling of the ground state obeys an exponential set of or- ≫ SL(t) L saturates at an asymptotic value. This behaviour dering relations [44]. can be∼ summarized in the following equation: Irreversibility for the entanglement entropy should then be obtained as a fundamental theorem, equivalent to the c- t t t∗ SL(t) ≤ ∗ . (62) theorem which is usually formulated in terms of the stress ∼ L t t ≥ 14 This behaviour of the entanglement entropy has been dered XY model is studied. In particular, both classical and checked in several lattice models both analytically and nu- quantum correlations are exponentially suppressed outside of merically [72, 94, 95, 96, 97]. In particular, in Ref. [94], the an effective light-cone whose radius grows at most logarith- previous results are provided analytically using Toeplitz ma- mically with time. trix representation and multidimensional phase methods for the XY model and considering large blocks. In Ref. [93], a simple interpretation of this behaviour is B. Bounds for time evolution of the block entropy proposed in terms of quasi-particles excitations emitted from the initial state at t = 0 and freely propagating with velocity All these results are compatible with the rigorous bounds v 1. The idea is that at, t = 0 and at many points of the found in Refs. [99, 100] by means of the Lieb-Robinson ≤ chain, a pair of entangled quasi-particles begin to propagate bound [101] and its generalizations presented in Refs. [102]. in opposite directions at some constant velocity v that we will The Lieb-Robinson bound states that the operator norm of consider 1 for simplicity (see Fig.4). The entanglement be- the commutator of two operators OA and OB that act on dif- tween the block of L spins and the rest of the system at an ferent regions A, B of a spin network with local interactions, arbitrary time is given by the number of pairs that have one hij (t), and in different times verifies quasi-particle in the block while the other is outside. Thus, − L−v|t| the entanglement entropy increases linearly with time until it [OA(t), OB(0)] cNmin OA OB e ξ , (63) saturates when the excitations that started in the middle of the ≤ block arrive at its boundary. where L is the distance between A and B (the number of All the previous results are explained in detail in the Ref. edges in the shortest path connecting A and B), Nmin = [33], where, apart from quantum quenches, a general con- min A , B is the number of spins in the smallest of A formal field theory approach to entanglement entropy is re- and {|B,| while| |} c,v,ξ > 0 are constants depending only on viewed. g = max(i,j)∈E maxt hij (t) and the architecture of the spin lattice. B 2t 2t B Thus, the Lieb Robinson bound, Eq. (63), tells us that the A t norm of the commutator of two operators at different times is exponentially small outside the light-cone given by the veloc- 2t < l ity v that we can understand like the speed of sound. Notice BBl that, by dimensional analysis, this velocity must be propor- A tional to the energy scale g. It is interesting to point out that this result is also valid for the case of fermions or local Hamil- t tonians with exponentially decaying interactions. In Ref. [99], it is shown, using the Lieb-Robinson bound and its generalizations [102], that correlations and informa- tion are propagated at a finite velocity in a spin network with 2t > l nearest-neighbour interactions. This is a non-trivial result since in non-relativistic quantum mechanics there doesn’t ex- ist the notion of a light-cone, i. e. local operations could be FIG. 4: Schematic representation of the dynamics of block entropy. used, in principle, to send information at long distances in ar- Entangled particles are emitted from the region A, they will con- tribute to the block entropy as long as one of the two particles ends bitrary short times. in the region B [from [93]]. Moreover, it is quantified the entanglement entropy that can be generated per unit of time between a block of spins and the rest of the system. In particular, it is found that Let us point out that this increase of the entanglement en- ∗ tropy is unrelated to the second law of thermodynamics. En- SL(t) SL(0) c gPt (64) tanglement entropy can decrease or even oscillate in standard − ≤ time evolution. where c∗ 1.9 is a constant and P is the perimeter of the ≃ Let us also mention that in Ref. [72] the dynamics of entan- block. Finally, let us mention that all these results are com- glement was analysed for disordered systems, i. e. when the plemented in Ref. [100]. couplings between the spins take random values. In particular, the XXZ model with the couplings between the spins follow- ing a uniform distribution in the interval [0, 1] was studied. It C. Long range interactions turns out that, in the presence of disorder, the entanglement entropy does not increase linearly but logarithmically. This The Lieb-Robinson bound is only valid for short range in- logarithmic behaviour does not follow from an extension of teractions. Then, it is interesting to study how does entan- the argument for the clean case assuming a diffusive propa- glement evolve in systems with long range interactions. This gation of the excitations, but it requires some kind of entan- question is addressed in Ref. [103]. glement localization. This behaviour is also observed in Ref. In general, systems with long range interactions are nu- [98] where the propagation of information through the disor- merically intractable since, in them, the entanglement entropy 15 scales with the volume SL L. Nevertheless, in Ref. [103], The study of entanglement along a quantum circuit was ad- the interactions are restricted∼ to Ising-type which allows to dressed in Refs. [108] and [109] by means of majorization study both the static and the dynamical entanglement proper- theory. In these works the introduction of entanglement in ties of the system. Shor’s algorithm and the Grover’s algorithm were analysed It is considered a lattice composed by N spins that interact respectively. according to the Hamiltonian, Let us remind the concept of Majorization relations, which is a more refined measure of ordering of probability distribu- 1 H = f(k,l) (1l σ(k)) (1l σ(l)) , (65) tions than the usual entropy one. We say that a probability dis- 4 − z ⊗ − z k ing: 1. A quantum register is initially prepared on the ground state of a known initial Hamiltonian H0. 2. The system is then made to evolve adiabatically from this Hamiltonian to a new one HP whose ground state codifies the solution to an e.g. NP-complete problem H(s(t)) = (1 s(t))H0 + s(t)HP . (69) FIG. 5: Structure of the quantum circuit performing the exact diago- − nalization of the XY Hamiltonian for 8 sites. The circuit follows the 3. Slow evolution from s(t = 0)= 0 to s(t = T )=1 structure of a Bogoliubov transformation followed by a fast Fourier guarantees that the system will not jump from the in- transform. Three types of gates are involved: type-B (responsible for stantaneous ground state of the system to the first ex- the Bogoliubov transformation and depending on the external mag- netic field λ and the anisotropy parameter γ), type-fSWAP (depicted cited state. as crosses and necessary to implement the anti-commuting proper- Quantum adiabatic computation is proven efficient provided ties of fermions) and type-F (gates associated to the fast Fourier that the minimum gap along the adiabatic evolution is only transform). Some initial gates have been eliminated since they only amount to some reordering of initial qubits [from [111]]. polynomially small in the number of qubits. If this was not the case, the adiabatic computation would require an exponen- tially large time as measure in terms of the number of qubits the system that this proposal may allow is N/2. This is be- in the register. cause the maximum entanglement that can generate a quan- Thus, according to the previous arguments, at some point tum gate that acts on two qubits is 1, that is, from a product of the adiabatic evolution of a hard quantum computation the state to a maximally entangled state (Bell basis). Thus, the system must be highly entangled, in a similar way as it hap- scaling law of the entanglement entropy that this proposal will pened in the previous sections at the quantum phase transi- allow will be tions. This makes us expect some sort of quantum phase tran- sition for a concrete value sc of the Hamiltonian, point that S(N/2) N/2 . (68) would be characterized by a minimum energy gap. ≤ In Ref. [114], adiabatic quantum computation is used to Notice that, as we have seen in the previous sections, the en- solve the NP-Complete Exact Cover problem that is a particu- tanglement entropy of the groundstate in the XY model scales lar case of the 3-SAT problem. It is defined as follows: given only logarithmically. The above circuit, then, can create much the n Boolean variables xi i=1,...n, xi =0, 1 i, where i is more entropy than what is present in the ground state. Yet, the bit index, we define a{clause} C involving the∀ three bits i, j we have also discussed the fact that time evolution does cre- and k by the constraint x +x +x =1. There are only three ate maximum entanglement. This, indeed, is what the above i j k assignments of the set of variables xi, xj , xk that satisfy circuit achieves. This shows that the previous proposal is op- this equation, namely, 1, 0, 0 , 0, 1,{0 and 0,}0, 1 . An in- timal since it carries the minimum possible number of gates stance of the Exact Cover{ problem} { is a} collection{ of} clauses such that maximum entanglement can be created. which involves different groups of three qubits. The problem Let us also add a final example on exact quantum circuits. is to find a string of bits x1, x2 ...,xn which satisfies all In Ref. [112], a quantum circuit that creates the Laughlin state the clauses. { } (Eq. 58) for an arbitrary number of particles (qudits) n in the This problem can be mapped to finding the ground state of case of filling fraction one is presented. The way in which a Hamiltonian HP in the following way [114]: given a clause entanglement grows along the circuit is also related to the C define the Hamiltonian associated to this clause as amount of entanglement that each gate of the circuit can gen- erate. In the case of this Laughlin wave function, the depth of 1 H = (1 + σz)(1 + σz)(1 + σz) (70) the circuit grows linearly with the number of qudits, so linear C 8 i j k z z z z z z entanglement S n can be supported by the circuit. This +(1 σ i )(1 σj )(1 σk)+(1 σi )(1 σj )(1 + σk) is precisely the entanglement∼ that the Laughlin wave function − − − − − +(1 σz)(1 + σz)(1 σz)+(1+ σz)(1 σz)(1 σz) , with filling fraction one requires as shown in Ref. [87]. − i j − k i − j − k The exact circuits we have discussed (XY and Laughlin where σz 0 = 0 , σz 1 = 1 . The quantum states of states) are both able to create linear entanglement entropy. It | | | −| the computational basis that are eigenstates of HC with zero is then impossible they can be simulated classically in an effi- eigenvalue (ground states) are the ones that correspond to the cient way. bit string which satisfies C, whereas the rest of the com- putational states are penalized with an energy equal to one. The problem Hamiltonian is constructed as the sum of all the B. Adiabatic quantum computation Hamiltonians corresponding to all the clauses in the instance, The framework of adiabatic quantum computation (AQC) HP = HC . (71) was introduced in Ref. [113]. The idea of AQC is the follow- C ∈ instance 17 1.4 uct state again. Roughly speaking, the power of the quantum 14 qubits 12 qubits computer respect to the classical one underlies in the paral- 1.2 10 qubits lelism during the computation that the superposition principle allows. 1 Let us make some warning remarks. The numerical simu- lations performed for the Exact Cover problem cannot deter- 0.8 mine the complexity class for the quantum algorithm. It is 0.6 generally believed that quantum computers will no be able to handle NP-complete problems. Yet, the simulation shows that mean entanglement 0.4 the best this quantum algorithm can achieved still requires a huge amount of entanglement in the register. 0.2 The divergence of the entanglement entropy that occurs at the critical point sc have also been observed in Shor’s factor- 0 ing algorithm in Ref. [118], where entropy grows exponen- 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 s tially fast respect the number of qudits. This, again, makes this algorithm hard to simulate classically. FIG. 6: Average over 300 instances of the entanglement entropy be- Notice that in the solution of other problems, the explosion 2 tween two blocks of size n/ as a function of the parameter s control- of the entropy could occur at in such a way that the en- ∼ 0 7 sc =1 ling the adiabatic evolution. A peak appears for sc . . The plot tanglement entropy was monotonically increasing. This sim- also shows the increase of the peak as the number of qubits grows ilar behaviour of the entropy to the quantum phase transition n = 10, 12, 14 [from [93]]. is, therefore, problem dependent. Recently, there has been appeareda new quantumalgorithm The ground state of this Hamiltonian corresponds to the quan- for SAT problemsthat improve the previousresults. It consists tum state whose bit string satisfies all the clauses. of a hybrid procedure that alternates non-adiabatic evolution It is known that Exact Cover is a NP-complete problem, so with adiabatic steps [119]. it cannot be solved in a polynomial number of steps in a clas- sical computer [115]. This makes the Exact Cover problem, particularly interesting, since if we had an algorithm to effi- C. One way quantum computation ciently solve Exact Cover, we could also solve all problems in the much larger NP family [116]. The one-way quantum computation (or measurement based In Ref. [117], the evolution of the entanglement properties QC) is a method to perform quantum computation that con- of the system are studied in order to see the expected sign sists of: (i) first, an entangled resource state is prepared, and of a quantum phase transition. 300 random instances for the (ii) then single qubit measurements are performed on it. It is Exact Cover are generated with only one possible satisfying called "one-way" because the entanglement of the state, which assignment for a small number of qubits. This instances are is the resource of the quantum computation, is destroyed by produced by adding clauses at random until there is exactly the measurements as the computation is being performed. Al- only one satisfying assignment. In order to apply adiabatic though the output of each individual measurement is random, quantum computation the initial Hamiltonian H0 taken is a they are related in such a way that the computation always magnetic field in the x direction succeeds. The idea is that depending on the previousoutcome, one chooses the basis for the next measurements. This implies n di that the measurements cannot be performed at the same time. H = (1 σx) , (72) 0 2 − i This kind of computation was introduced in Ref. [120] i=1 where there was shown that with an initial particular state, where di is the number of clauses in which qubit i appears. called cluster state, any quantum computation could be simu- Then, for each instance, the ground state is computed for sev- lated. Later on, other useful states to perform one-way quan- eral values of s of the Hamiltonian, H(s)=(1 s)H0 +sHP tum computation were found [121, 122, 123, 124, 125] and its corresponding entanglement entropy of− half a chain. The fact that the measurement based quantum computation The mean of the entanglement entropy over these 300 in- is universal is non-obvious, since a quantum computation is stances is performed and plotted respect to the s parameter a unitary process, while a measurement is a random process. for different sizes of the system in Fig. 6. We can observe a The key point is that there are two kinds of qubits in the spin peak of the entropy around the critical value sc 0.7. system: the cluster qubits which will be measured in the pro- We interpret this behaviour of the entanglement∼ entropy as cess of computation and the logical qubits which constitute follows: initially the system is in a product state and its en- the quantum information that is going to be processed. tanglement is zero. Then, the evolution makes the system ex- Although, globally, entanglement is expected to decrease plore different solutions by means of superposition states of along the quantum computation due to the single qubit mea- them, that is, it becomes more and more entangled. Finally, surements, in the set of logical qubits (the register that will be the system throws away the bad solutions, the entanglement read out at the end of the computation), the entanglement may decreases, until the best solution is foundand it rests in a prod- increase. Notice that if the initial state fulfils an area law, the 18 entanglement is enough that the register of the logical qubits and clever ideas are available in the literature. The first idea is as entangled as possible. That is, area law on a 2D state is consists in exploiting the fact that typical interactions are lo- just what is needed to have linear maximal entanglement on a cal. This suggests that entanglement should be created se- register defined on a line in that state. Cluster states are just quentially in space from each local degree of freedom to its enough to handle the expected entanglement in the register. nearest neighbours. Then, a one-dimensional state can be rep- In this respect, it has been recently shown that most quantum resented as a matrix product state which captures such a prin- states are too entangled to be useful in order to perform mea- ciple [127]. In higher dimensions, states can be represented as sured based quantum computation [126]. Projected Entangled Pairs (see Refs. [128]). The second idea to classically represent quantum states as efficiently as pos- sible consists in reconstructing the correlations in the system VIII. CONCLUSION: ENTANGLEMENT AS THE as a renormalization group tree. This goes under the name of BARRIER FOR CLASSICAL SIMULATIONS Multiscale Entanglement Renormalization Ansatz (MERA) is a more sophisticated representation which is specially suited Entanglement is the genuine quantum property that escapes for critical systems. The accuracy of the approximations can classical physics. The Hilbert space structure of a multi- be quantify using the amount of entropy of entanglement that partite quantum system allows for superpositions of exponen- the approximation can accommodate [129]. tially many elements of the basis. Entropy of entanglement is Multi-partite entanglement branches in many others sub- a way to quantify the amount of quantum correlation between jects that escape this short review. Very likely, much more parts of such a multi-partite system. Entanglement entropy work is still needed to get a profound understanding of the is, then, a genuine measure of the global quantumness of the role of entanglement in highly structured quantum systems. state. It serves as a conclusion to recall the deep implications of entanglement entropy in the possibility of producing faithful classical simulations of quantum mechanics. In Ref. [106], Acknowledgments it was proven that efficient simulations are possible for any system where all its Schmidt decompositions in two arbitrary Financial support from QAP (EU), MICINN (Spain), FI parts would carry little entropy. Therefore, entanglement is at program and Grup Consolidat (Generalitat de Catalunya), and the heart of the separation between efficient and non-efficient QOIT Consolider-Ingenio 2010 is acknowledged. We also simulations of quantum mechanics. would like to thank the large number of colleagues with whom What is not fully understood is what is the best strategy we have collaborated and discussed all along these recent to classically account for quantum correlations. Two general years in the never-ending effort to understand entanglement. [1] G. Vidal, J. I. Latorre, E. Rico and A. Kitaev, Phys. Rev. Lett. and Transitions in Transverse Ising Models, Ed. Spinger m.41. 90, 227902 (2003). [16] I. Peschel, M. Kaulke and O. Legeza, Ann. Physik (Leipzig) [2] J. I. Latorre, E. Rico and G. Vidal, Quant. Inf. Comput. 4, 48- 8 153 (1999). M. C. Chung and I. Peschel Phys. Rev. B 64 92 (2004). 064412 (2001). I. Peschel, J. Phys. A: Math. Gen. 36 L205 [3] L. Amico, R. Fazio, A. Ostorloh and V. Vedral, Rev. Mod. (2003). Phys. 80, 517 (2008). [17] I. Peschel and V. Eisler, arXiv:0906.1663 (2009). [4] M. B. Plenio and S. Virmani, Quant. Inf. Comp. 7, 1 (2007). [18] B.-Q. Jin and V. E. Korepin, J. of Stat. Phys. 116, 79 (2004). [5] R. Horodecki, P. Horodecki, M. Horodecki and K. Horodecki, [19] P. Zanardi and N. Paunkovic,´ Phys. Rev. E 74, 031123 (2006). quant-ph/0702225 (2007). P. Zanardi, M. Cozzini and P. Giorda, J. Stat. Mech. L02002 [6] J. Eisert, Entanglement in quantum information theory, (2007). quant-ph/0610253, PhD thesis (2006). [20] H.-Q. Zhou and J.P. Barjaktarevic,´ J. Phys. A: Math. Theor. 41 [7] I. Bengtsson and K. Zyczkowski, Geometry of Quantum States 412001 (2008). H.-Q. Zhou, J.-H. Zhao, and B. Li, J. Phys. A: - An Introduction to Quantum Entanglement Cambridge Uni- Math. Theor. 41 492002 (2008). H.-Q. Zhou, R. Orús, and G. versity Press, Cambridge (2006). Vidal, Phys. Rev. Lett. 100, 080602 (2008). [8] V. Vedral, Rev. Mod. Phys. 74, 197 (2002). [21] T.T. Wu, Phys. Rev. 149, 380, (1966) [9] D. Bruß, J. Math. Phys. 43, 4237 (2002). [22] M.E. Fisher and R.E. Hartwig, Adv. Chem. Phys. 15, 333, [10] W. K. Wooters, Quant. Inf. Comp. 1, 27 (2001). (1968) [11] M. B. Plenio and V. Vedral, Contemp. Phys. 39, 431 (1998). [23] E.L. Basor, Indiana Math. J. 28, 975, (1979) [12] M. Ohya and D. Petz, Quantum entropy and its use (Springer- [24] E.L. Basor and C.A. Tracy, Physica A 177, 167, (1991) Verlag Berlin, 1993). [25] A. Böttcher and B. Silbermann, Analysis of Toeplitz Opera- [13] S. Sachdev, Quantum Phase Transitions, Cambridge Univ. tors, Springer-Verlag, Berlin, 1990 Press (1999). [26] A. R. Its, B.-Q. Jin and V. E. Korepin, Journal Phys. A: Math. [14] P. Christe and M. Henkel, Introduction to conformal invari- Gen. 38, 2975-2990 (2005). F. Franchini, A. R. Its, B.-Q. Jin ance and its applications to critical phenomena, Ed. Springer- and V. E. Korepin, Proceedings of the Third Feynman Work- Verlag m.16. shop (2006). F. Franchini, A. R. Its, B.-Q. Jin and V. E. Kore- [15] B. K. Chakrabarti, A. Dutta and P. Sen, Quantum Ising phases pin, J. Phys. A: Math. Theor. 40 8467-8478 (2007). 19 [27] F. Franchini, A. R. Its and V. E. Korepin, J. Phys. A: Math. [67] B. Nienhuis, M. Campostrini and P. Calabrese, J. Stat. Mech. Theor. 41 025302 (2008). P02063 (2009). [28] A. M. Polyakov, Sov. Phys. JETP Lett. 12, 381 (1970). [68] G. Refael and J. E. Moore, Phys. Rev. Lett. 93, 260602 (2004). [29] Paul Ginsparg, Applied conformal field theory, Les Houches [69] D. S. Fisher, Phys. Rev. B 50, 3799 (1994). Summer School, pp 1-168 (1988). [70] C. Dasgupta and S. K. Ma, Phys. Rev. B 22, 1305 (1980). [30] J. Cardy, Conformal Field Theory and Statistical Mechanics [71] N. Laflorencie, Phys. Rev. B 72, 140408(R) (2005). Les Houches Summer School, cond-mat/0807.3472 (2008). J. [72] G. De Chiara, S. Montangero, P. Calabrese and R. Fazio, J. Cardy, Phase Transitions 11, Academic Press (1987). Stat. Mech. P03001 (2006). [31] C. Holzhey, F. Larsen and F. Wilczek, Nucl.Phys.B 424, 443 [73] N. E. Bonesteel and Kun Yang, Phys. Rev. Lett. 99, 140405 (1994). (2007). [32] P. Calabrese and J. Cardy, J. Stat. Mech. P06002 (2004); P. [74] V. V. França and K. Capelle, Phys. Rev. Lett. 100, 070403 Calabrese, J. Cardy and E. Tonni, arXiv:0905.2069 (2009). (2008). [33] P. Calabrese and J. Cardy, arXiv:0905.4013 (2009). [75] Y.-C. Lin, F. Igloi and H. Rieger Phys. Rev. Lett. 99, 147202 [34] H. Q. Zhou, T. Barthel, J. O. Fjaerestad and U. Schollwoeck, (2007). Entanglement and boundary critical phenomena, Phys. Rev. A [76] R. Yu, H. Saleur and S. Haas, Phys. Rev. B 77, 140402(R) 74, 050305(R) (2006). (2008). [35] P. Calabrese and A. Lefevre, Phys. Rev A 78, 032329 (2008) [77] S. Garnerone, N. T. Jacobson, S. Haas and P. Zanardi, Phys. [36] J. Eisert and M. Cramer, Phys. Rev. A 72, 042112 (2005) Rev. Lett. 102, 057205 (2009). N. T. Jacobson, S. Garnerone, [37] R. Orus, J.I. Latorre, J. Eisert and M. Cramer, Phys. Rev. A S. Haas and P. Zanardi, Phys. Rev. B 79, 184427 (2009). 73, 060303(R) (2006) [78] I. Affleck, N. Laflorencie and E. S. Sorensen, arXiv:0906.1809 [38] M. Srednicki, Phys. Rev. Lett. 71 666 (1993). (2009). [39] T. M. Fiola, J. Preskill, A. Strominger and S. P. Trivedi, Phys. [79] H. J. Lipkin, N. Meshkov and A. J. Glick, Nucl. Phys. 62, 188 Rev. D 50, 3987 (1994). (1965). N. Meshkov, A. J. Glick and H. J. Lipkin, Nucl. Phys. [40] L. Bombelli, R. K. Koul, J. Lee and R. D. Sorkin, Phys. Rev. 62, 199 (1965). A. J. Glick, H. J. Lipkin and N. Meshkov, D 34, 373 (1986). Nucl. Phys. 62, 211 (1965). [41] D. Kabat and M. J. Strassler, Phys. Lett. B 329,46 (1994). [80] J. I. Latorre, R. Orús, E. Rico and J. Vidal, Phys. Rev. A 71, [42] D. Kabat, Nucl. Phys. B 453, 281 (1995). 064101 (2005). [43] H.Casini and M.Huerta, arXiv:0905.2562 (2009). [81] T. Barthel, S.Dusuel and J. Vidal, Phys. Rev. Lett. 97, 220402 [44] A. Riera and J. I. Latorre, Phys. Rev. A 74, 052326 (2006). (2006) [45] M. Cramer, J. Eisert, M. B. Plenio and J. Dreissig, Phys. Rev. [82] P. Ribeiro, J. Vidal and R. Mosseri, Phys. Rev. E 78, 021106 A 73, 12309 (2006). (2008). [46] T. Barthel, M.-C. Chung and U. SchollwÃ˝uck, Phys. Rev. A [83] J. Vidal, S. Dusuel and T. Barthel, J. Stat. Mech. 0701, P015 74, 022329 (2006). (2007). [47] J. Eisert, M. Cramer and M.B. Plenio, quant-ph/0808.3773 [84] D. C. Tsui, H. L. Stormer and A. C. Gossard, Phys. Rev. Lett. (2008). 48 1559 (1982). [48] M. M. Wolf, Phys. Rev. Lett. 96, 010404 (2006). [85] X. G. Wen, Quantum field theory and many-body systems, Ox- [49] D. Gioev and I. Klich, Phys. Rev. Lett. 96, 100503 (2006). ford University Press (2004). [50] E. Fradkin and J. E. Moore, Phys. Rev. Lett. 97, 050404 [86] R. B. Laughlin, Phys. Rev. Lett. 50, 1395 (1983). (2006). [87] S. Iblisdir, J. I. Latorre and R. Orús, Phys. Rev. Lett. 98 [51] S. Ryu and T. Takayanagi, J. High En. Phys. 0608, 045 (2006). 060402 (2007). [52] S. Ryu and T. Takayanagi, Phys. Rev. Lett. 96, 181602 (2006). [88] M. Haque, O. Zozulya and K. Schoutens, Phys. Rev. Lett. 98, [53] T. Nishioka, S. Ryu and T. Takayanagi, arXiv:0905.0932 060401 (2007) O. S. Zozulya, M. Haque and K. Schoutens, (2009). Phys. Rev. A, 78, 042326 (2008); [54] A. Kitaev and J. Preskill, Phys. Rev. Lett. 96, 110404 (2006). [89] M. Haque, O. S. Zozulya and K. Schoutens, arXiv:0905.4024 [55] E. Fradkin, arXiv:0906.1809 (2009). (2009). [56] E. Lieb, T. Schultz and D. Mattis, Annals of Phys. 16 407 [90] F. Verstraete, J.I. Cirac, J.I. Latorre, E. Rico and M.M. Wolf, (1961). Phys. Rev. Lett. 94, 140601 (2005). [57] S. Katsura, Phys. Rev. 127 1508 (1962) [91] J.I. Latorre, C.A. Lutken, E. Rico and G. Vidal, Phys. Rev. A [58] E. Barouch and B. McCoy, Phys. Rev. A 3 786 (1971) 71 034301 (2005). [59] F. Iglói and R. Juhász, Europhys. Lett. 81, 57003 (2008). [92] H. Casini and M. Huerta, Phys. Lett. B 600, 142-150 (2004). [60] J. L. Cardy, O. A. Castro-Alvaredo and B. Doyon, J. Stats. [93] P. Calabrese and J. Cardy, J. Stat. Mech. P04010 (2005). Phys. 130, 129 (2007). [94] M. Fagotti and P. Calabrese, Phys. Rev. A 78, 010306(R) [61] O. A. Castro-Alvaredo and B. Doyon, hep-th/0906.2946 (2008). (2009). [95] L. Cincio, J. Dziarmaga, M. M. Rams and W. H. Zurek, Phys. [62] H. Bethe, Z. Phys. 71, 205 (1931). Rev. A 75, 052321 (2007). [63] N. Bogoliubov, A. Izergin and V. Korepin, Quantum Inverse [96] N. Schuch, M. M. Wolf, K. G. H. Vollbrecht and J. I. Cirac, Scattering Method and Correlation Functions, Cambridge New J. Phys. 10, 033032 (2008). University Press (1993). [97] A. Laeuchli and C. Kollath, J. Stat. Mech. P05018 (2008). [64] M. Takahashi, Thermodynamics of One-dimensional Solvable [98] C. K. Burrell and T. J. Osborne, Phys. Rev. Lett. 99, 167201 Models, Cambridge University Press (1999). (2007). [65] E. Ercolessi, S. Evangelisti and F. Ravanini, arXiv:0905.4000. [99] S. Bravyi, M. B. Hastings and F. Verstraete, Phys. Rev. Lett. [66] J. Sato, M. Shiroishi and M. Takahashi, J. Stat. Mech. 0612, 97, 050401 (2006). P017 (2006). J. Sato and M. Shiroishi, J. Phys. A: Math. Theor. [100] J. Eisert and T. Osborne, Phys. Rev. Lett. 97, 150404 (2006). 40, 8739 (2007). [101] E. H. Lieb and D. W. Robinson, Commun. Math. Phys. 28, 20 251 (1972). mation transmission 9, 265 (1973). [102] M. B. Hastings, Phys. Rev. B 69, 104431 (2004); M.B. Hast- [116] The NP problems are those whose solutions can be verified in ings, Phys. Rev. Lett. 93, 140402 (2004); M.B. Hastings, T. a polynomial time. Koma, Commun.Math.Phys. 265, 781-804 (2006). [117] J. I. Latorre and R. Orús, Phys. Rev. A 69, 062302 (2004). [103] W. Dür, L. Hartmann, M. Hein, M. Lewenstein and H. J. [118] J. I. Latorre and R. Orús, Phys. Rev. A 69, 052308 (2004). Briegel, Phys. Rev. Lett. 94, 097203 (2005). [119] J. J García-Ripoll and M. C. Bañuls, quant-ph/0812.1760 . [104] F. Verstraete and J. I. Cirac, cond-mat/0407066 (2004). [120] R. Raussendorf and H. J. Briegel, Phys. Rev. Lett. 86, 5188 [105] J. I. Latorre, J. Phys. A 40, 6689-6697 (2007). (2001); R. Raussendorf and H. J. Briegel, Quant. Inf. Comp. [106] G. Vidal, Phys. Rev. Lett. 91, 147902 (2003). G. Vidal, Phys. 6, 433 (2002); P. Walther et al., Nature 434, 169 (2005). Rev. Lett. 93, 040502 (2004). [121] D. Gross and J. Eisert, Phys. Rev. Lett. 98, 220503 (2007). [107] M. A. Nielsen, and I. L. Chuang, Quantum Computation and [122] D. Gross, J. Eisert, N. Schuch, and D. Perez-Garcia, Phys. Quantum Information, (Cambridge Univ. Press, Cambridge, Rev. A 76, 052315 (2008). England, 2000). [123] D. Gross and J. Eisert, arXiv:0810.2542 (2008). [108] J. I. Latorre, M. A. Martín-Delgado; Phys. Rev. A66, 022305 [124] M. Van den Nest, W. Dür, A. Miyake, and H. J. Briegel, New (2002). J. Phys. 9, 204 (2007). [109] R. Orús, J. I. Latorre, M. A. Martín-Delgado; Quantum Infor- [125] G.K. Brennen and A. Miyake, Phys. Rev. Lett. 101, 010502 mation Processing, 4, 283-302 (2003). (2008). [110] R. Orús, J. I. Latorre, M. A. Martín-Delgado; European Phys- [126] D. Gross, S. Flammia and J. Eisert, Phys. Rev. Lett. 102, ical Journal D 29, 119-132 (2004). 190501 (2009). [111] F. Vestraete, J. I. Cirac and J. I. Latorre, Phys. Rev. A 79, [127] M. Fannes, B. Nachtergaele, and R. F. Werner, Comm. Math. 032316 (2008). Phys. 144 (1992). D. Pérez-García, F. Verstraete, M.M. Wolf [112] J. I. Latorre, V. Picó and A. Riera, quant-ph/0902.4797 . and J.I. Cirac, Quantum Inf. Comput. 7, 401 (2007). [113] E. Farhi, J. Goldstone, S. Gutmann, J. Lapan, A. Lundgren and [128] F. Verstraete and J. I. Cirac, cond-mat/0407066 (2004). F. Ver- D. Preda, Science 292, 472 (2001). straete, V. Murg and J. I. Cirac, Advances in Physics 57,143 [114] E. Farhi, J. Goldstone, S. Gutmann and M. Sipser (2008). quant-ph/0001106 (2000). [129] L. Tagliacozzo, Thiago. R. de Oliveira, S. Iblisdir and J.I.La- [115] S. A. Cook, in STOC’ 71: Proceedings of the third annual torre, Phys. Rev. B 78, 024410 (2008); F. Pollmann, S. Muk- ACM symposium on Theory of computing (ACM, New York, erjee, A. Turner and J. E. Moore, arXiv:0812.2903 NY, USA, 1971), pp. 151-158. L. A. Levin, Problems of infor-