Arxiv:1601.03499V1 [Quant-Ph] 14 Jan 2016 Networks [23]

Non-Hermitian tight-binding network engineering

Stefano Longhi Dipartimento di Fisica, Politecnico di Milano and Istituto di Fotonica e Nanotecnologie del Consiglio Nazionale delle Ricerche, Piazza L. da Vinci 32, I-20133 Milano, Italy

We suggest a simple method to engineer a tight-binding quantum network based on proper coupling to an auxiliary non-Hermitian cluster. In particular, it is shown that eﬀective complex non-Hermitian hopping rates can be realized with only complex on-site energies in the network. Three applications of the Hamiltonian engineering method are presented: the synthesis of a nearly transparent defect in an Hermitian linear lattice; the realization of the Fano-Anderson model with complex coupling; and the synthesis of a PT -symmetric tight-binding lattice with a bound state in the continuum.

PACS numbers: 03.65.-w, 11.30.Er, 72.20.Ee, 42.82.Et

I. INTRODUCTION self-sustained emission in semi-infinite non-Hermitian systems at the exceptional point [29], optical simulation of PT -symmetric quantum field theories in the ghost Hamiltonian engineering is a powerful technique regime [30, 31], invisible defects in tight-binding lattices to control classical and quantum phenomena with [32], non-Hermitian bound states in the continuum [33], important applications in many areas of physics such and Bloch oscillations with trajectories in complex plane as quantum control [1–4], quantum state transfer and [34]. Previous proposals to implement complex hopping quantum information processing [5–10], quantum simu- amplitudes are based on fast temporal modulations of lation [11–13], and topological phases of matter [14–19]. complex on-site energies [31, 32], however such methods In quantum systems described by a tight-binding are rather challenging in practice and, as a matter of Hamiltonian, quantum engineering is usually aimed at fact, to date there is not any experimental demonstration tailoring and controlling hopping rates and site energies, of non-Hermitian complex couplings in tight-binding using either static or dynamic methods. For example, networks. special tailoring of the hopping rates in a linear tight- binding chain allows one to realize perfect state transfer In this work we suggest a simple method to engineer between distant sites in the chain [6, 9, 20], whereas hopping amplitudes and site energies in a tight-binding external time-dependent perturbations represent a network, which simply involves Hermitian couplings and rich and versatile resource to realize synthetic gauge no synthetic gauge fields. The method is based on proper fields, thus achieving topological phases in systems that coupling of the main tight-binding network to an auxil- are topologically trivial in equilibrium [15–19]. The iary non-Hermitian cluster. In particular, it is shown that recent growing interest in non-Hermitian quantum and effective complex (non-Hermitian) hopping rates can be classical systems [21], especially in those possessing PT realized with only static on-site complex potentials in the symmetry [22], has motivated the extension of quantum network, i.e. avoiding fast modulation and thus greatly control methods and Hamiltonian engineering into the simplifying its practical implementation. Three applica- non-Hermitian realm [23–27], with ramifications and tions of the tight-binding network engineering method important applications to e.g. PT -symmetric integrated are presented: the synthesis of a nearly invisible defect photonic devices [28]. The ability to tailor complex on- in an Hermitian tight-binding linear lattice; the realiza- site potentials and non-Hermitian hopping amplitudes is tion of the Fano-Anderson model with complex coupling; a key task in the engineering of non-Hermitian quantum and the synthesis of a PT -symmetric tight-binding lattice with a bound state in the continuum. arXiv:1601.03499v1 [quant-ph] 14 Jan 2016 networks [23]. While the engineering of complex on-site potentials is a rather feasible task, the realization of complex hopping amplitudes remains a rather challenging issue. For example, in optics non-Hermitian II. NETWORK ENGINEERING METHOD tight-binding networks with complex on-site potentials are readily implemented by evanescent coupling of light Let us consider a rather general tight-binding network modes trapped in optical waveguides or resonators with S, which is constructed topologically by N sites |niS and optical gain and loss in them, while the realization the various connections between them. As a simplified of controllable non-Hermitian coupling constants is model, it captures the essential features of many discrete a much less trivial task. However, complex hopping classical and quantum systems [23]. To engineer the hop- amplitudes play an important role for the observation of ping rates and site potentials of the network S, we con- a wide variety of phenomena that have been disclosed sider an auxiliary cluster A, with M sites |αiA, which is in recent works [24, 29–34]. These include incoherent coupled to the main network S [Fig.1(a)]. As a limiting control of non-Hermitian Bose-Hubbard dimers [24], case, the network S can comprise an infinite number of 2 sites, for example it can describe an infinitely-extended To obtain the dynamical behavior of the system solely, one-dimensional tight-binding lattice, side coupled to the one might try to proceed by elimination of the auxiliary auxiliary cluster A. The tight-binding Hamiltonian of the amplitudes aα(t) from Eqs.(6) and (7), thus obtaining full system S+A is given by coupled integro-differential equations for the system amplitudes c (t) (see Appendix A). However, such a proce- Hˆ = Hˆ + Hˆ + Hˆ (1) n S A I dure turns out to be useful in defining an effective energy- ˆ where independent Hamiltonian Heff for the system S when- M ever the auxiliary system A is an almost continuum of X (S) X (A) states (i.e. M → ∞) and the S-A coupling is weak. In- HˆS = H |niShm|S , HÂ = H |αiAhβ|A n,m α,β deed, this is the usual way to describe metastability of n,m α,β=1 (2) Markovian open quantum systems (see, for instance, [35– are the Hamiltonians of the main (S) and auxiliary (A) 37]). Here, however, we typically consider a finite (and networks, respectively, and possibly small) number of auxiliary sites M, a typical infinitely-extended system S (N → ∞), and do not nec- M essarily require the S-A coupling to be weak. In such a ˆ X X HI = (ρα,n|αiAhn|S +ρ ñ,α|niShα|A) (3) case, the reduction procedure of open quantum systems α=1 n and derivation of an effective Hamiltonian can be applied describes their interaction. In the above equations, the under certain conditions solely, which are discussed in the roman and greek indices run over the sites of main net- Appendix A. For our purposes, we follow here a different work S and auxiliary cluster A, respectively, the ma- strategy. Let us look for an eigenstate of Hˆ with energy trix H(S) describes on-site potentials (diagonal elements E, which can be either a bound state or a scattered state (S) when N = ∞. Assuming the dependence ∼ exp(−iEt) Hn,n) and hopping amplitudes (off-diagonal elements (S) for the amplitudes in Eqs.(6) and (7), one obtains Hn,m, n 6= m) among the various sites of the main sys- (A) tem S, and the M × M matrix H is the analogous Ec = H(S)c +ρ ã (8) matrix for the auxiliary system. The two matrices ρ and (A) ρ˜, entering in Eq.(3), describe the interaction between Ea = H a + ρc, (9) the sites of S and A. We assume that the hopping rates T where c = (..., c−1, c0, c1, c2, ...) and a = among the different sites in both main (S) and auxiliary T (a1, a2, a3, ..., aM ) are the vectors of S and A site (A) networks are Hermitian and that there are not gauge amplitudes. After elimination of the amplitudes a, one fields that introduce Peierls’ phases in the hopping ampli- obtains tudes. Such an assumption implies that the non-diagonal elements of the matrices H(S) and H(A), and all the ele- Ec = Heff (E)c, (10) ments of the matrices ρ andρ ˜, are real, with ρα,n =ρ ñ,α, (S) (S) (A) (A) where we have set Hn,m = Hm,n and Hα,β = Hβ,α i.e.

T (S) (S) T (A) (A) T (S) (A) −1 ρ˜ = ρ , H = H , H = H . (4) Heff (E) ≡ H +ρ ˜(E − H ) ρ. (11)

(S) (A) However, site potentials Hn,n and Hα,α , in either or both Equation (11) shows that the effect of the auxiliary clus- the main and auxiliary networks, are allowed to be com- ter A is to renormalize the hopping amplitudes and site plex. Note that, if the auxiliary cluster is made of purely potentials of the network S by adding, to the Hamilto- (A) (S) dissipative sites, i.e. the imaginary parts of Hα,α are nian H , a generally energy-dependent term [the second either zero or negative, the eigenvalues of H(A) have neg- term on the right hand side of Eq.(11)]. Such an addi- ative (or vanishing) imaginary parts, and secularly grow- tional term is analogous to the so-called ’optical poten- ing terms of the auxiliary site amplitudes are avoided in tial’ found in the effective Hamiltonian description of de- the weak coupling regime ρ → 0. After expanding the caying open quantum systems using the Feshbach’s pro- state vector of the full system as jection operator method [36]. Interestingly, the optical potential term generally M X X makes the off-diagonal elements of Heff complex, i.e. the |ψ(t)i = cn(t)|niS + aα(t)|αiA, (5) effective hopping amplitudes are non-Hermitian in spite n α=1 all hopping amplitudes in the network and auxiliary clus- from the Schrödingerequation i∂t|ψ(t)i = Hˆ |ψi one ob- ter are Hermitian. In particular, one can readily shown tains that: M (i) If the cluster A is Hermitian, i.e. on-site potentials dcn X (S) X (A) i = H c + ρ˜ a (6) Hα,α are real, then Heff is real and symmetric. In this dt n,m m n,α α m α=1 case the effect of A is to renormalize the hopping rates M of sites in S because of additional tunneling paths intro- daα X (A) X i = H a + ρ c . (7) duced by the the auxiliary sites, however they remain dt α,β β α,n n β=1 n Hermitian. A typical example is provided by indirect 3

(a) (A) S (b) on-site potentials Hα,α of the auxiliary sites are (in mod- ... ulus) much larger than the energy |E|, the inverse matrix -1 n m (A) −1 >S 1 S 2 0>S 0>S (E −H ) entering in the ’optical potential’ is weakly -2 > >S >S dependent on the energy E, and thus one can approxi- ... 0 >S mately set ... (S) (A) −1 2 M H 'H − ρ˜(H ) ρ. (13) >A A >A eff α 1 >A 0>A In this way the dependence of the eﬀective Hamiltonian Heff on energy is removed. In particular, by further FIG. 1. (Color online) (a) Schematic of a tight-binding net- (A) (A) (A) −1 work S coupled to an auxiliary cluster A made of M sites. (b) assuming |Hα,α| |Hβ,γ | (β 6= γ), (H ) is diagonal (A) Control of hopping amplitude between sites |n0iS and |m0iS with elements 1/Hα,α, so that taking into account that T via an auxiliary site |α0iA. ρ˜ = ρ one has

M X ρα,nρα,m (H ) 'H(S) − . (14) (second-order) tunneling, which is described in Sec.III.A. eff n,m n,m (A) (A) α=1 Hα,α (ii) If the on-site potentials Hα,α in the auxiliary sites are complex, for any real energy E the matrix Heff (E) Equation (14) enables, in principle, to engineer the ef- is symmetric but not Hermitian, i.e. one has fective matrix elements (Heff )n,m in a rather flexible and independent way. For example, to engineer the hop- (Heff )n,m = (Heff )m,n (12) ping amplitude between two prescribed sites n = n0 and m = m0 of the network S, we can consider an auxiliary but (Heff )n,m is generally complex. This is one of the site, say |α0iA, which is the only site of A coupled to most important result of the analysis and shows that |n0iS and |m0iS [Fig.1(b)]. From Eq.(14) one then ob- complex on-site potentials in the auxiliary sites result in tains an effective non-Hermitian hopping amplitudes among ρ ρ sites in the network S. (H ) 'H(S) − α0,n0 α0,m0 . (15) eff n0,m0 n0,m0 (A) It should be noted that the above equivalence holds for Hα0,α0 a prescribed energy E, and that Eq.(10) is actually an im- Note that, while the hopping amplitudes (H(S)) , plicit eigenvalue equation since the effective Hamiltonian n0,m0 (A) Heff (E) depends on energy E via the ’optical potential’ ρα0,n0 and ρα0,m0 are real, the on-site potential Hα0,α0 (A) term [the second term on the right hand side of Eq.(11)]. is complex, so that by a judicious choice of (H )α0,α0

For weak S-A coupling, i.e. for ρ → 0, an iterative proce- and ρα0,n0 ρα0,m0 a desired non-Hermitian complex hop- dure can be used to solve Eq.(10), while in certain special ping amplitude (Heff )n0,m0 can be realized. Note that, cases some bound states can be determined in a closed as opposed to the weak-coupling limit described in the form without resorting to any approximation (see, for ex- Appendix A, in such a procedure there is no restriction ample, the model discussed in Sec.III.C). However, there on the magnitude of the S-A coupling ρ, so that the cor- are at least two important cases where the problem is rection to the hopping rate provided by the second term amenable of analytical results, without requiring small on the right hand side of Eq.(15) is not necessarily small. interaction limit. (i) Linear tight-binding homogeneous lattices. Let us sup- pose that the network S is an infinitely-extended one- III. APPLICATIONS dimensional tight-binding lattice with (asymptotically) homogeneous nearest-neighbor hopping rate κ and uni- The rather general procedure of network engineering (S) form site potentials, i.e. Hn,m → κ(δn,m+1 + δn,m−1) presented in Sec.II is exemplified by considering three ap- as n, m → ±∞. Since the auxiliary cluster A cou- plications to some important physical problems, namely ples only with a few sites in S with finite index n, the the synthesis of a nearly invisible defect in an Hermi- scattering states of S are asymptotically plane waves, tian homogeneous lattice, the realization of the Fano- cn ∼ exp(±iqn) as n → ±∞, and their energy is known Anderson model with non-Hermitian coupling, and the and given by E = 2κ cos(q), where −π ≤ q < π is the synthesis of a PT -symmetric tight-binding lattice with a Bloch wave number. Hence, for a fixed value of the wave bound state in the continuum. number q, the effective Hamiltonian Heff is known and scattering states , including reflection/transmission coefficients, can be readily determined by standard methods. A. Nearly-invisible defect in an Hermitian This approach can be applied to the determination of tight-binding lattice bound states as well, looking at the poles of the spectral transmission. An example is discussed in Sec.III.A. The possibility of synthesizing transparent defects in (ii) Large on-site potentials of auxiliary cluster. If the tight-binding lattices has received an increasing interest 4

1 ping rate θ between sites |0iS and |1iS; σ is the potential (a) U >A A at sites |0iS and |1iS; ω is the hopping rate between the ω ω ... σ σ ... auxiliary site |1iA and the two sites |0iS and |1iS; and κ κθ κ κ U is the potential of site |1iA [see Fig.2(a)]. According -2 -1 0 1 2 3 >S >S >S >S S >S >S to Eq.(11), elimination of the auxiliary site yields the following eﬀective Hamiltonian for the lattice S (b) σ‘ σ’ ...... 2 1 κ κ κ κ (S) ω X θ’ (Heff )n,m = Hn,m + δn,lδm,k (19) -2 S -1 S 0 S 1 >S 2 >S 3 >S E − U > > > S k,l=0 (c) 1 4 which basically describes the modiﬁed linear lattice de- 3 2 picted in Fig.2(b). As it can be seen, the role of the 0.5 1 side-coupled auxiliary site |0iA is to modify the potentials and hopping rate between sites |0iS and |1iS to the

0 effective values transmittance transmittance −2 −1 0 1 2 2 σ0 = σ + ω2/(E − U) (20) 1 θ0 = θ + ω2/(E − U). (21) 1 2 3 0 0 4 In particular, note that the effective hopping rate θ is phase (rad) phase (rad) −2 −1 0 1 2 given by the interference of two terms: direct tunneling energy E/ κ between sites |0iS and |1iS with hopping amplitude θ, and indirect (second-order) tunneling via the auxiliary FIG. 2. (Color online) (a) Schematic of a linear tight-binding site with energy-dependent hopping amplitude ω2/(E − lattice S side-coupled to an auxiliary site A. (b) Equivalent U). lattice after elimination of the auxiliary site, with energy- 0 0 The spectral transmission and reflection of the result- dependent renormalized hopping rate θ and site potential σ , ing lattice of Fig.1(b), as well as bound states, can be defined by Eqs.(20) and (21) given in the text. (c) Behavior calculated by standard methods, with the results of the of the spectral transmittance |t(E)|2 and phase of t(E) versus energy E = 2κ cos(q) for parameter values θ/κ = 0.2, σ/κ = analysis given below. However, the conditions for a −0.8 and for: U/κ = −5, ω/κ = 2 (curve 1); U/κ = −10, nearly-invisible defect can be readily established from an √ inspection of Eqs.(20) and (21) without any detailed cal- ω/κ = 2 2 (curve√ 2); U/κ = −20, ω/κ = 4 (curve 3); U/κ = −40, ω/κ = 4 2 (curve 4). culation. In fact, as discussed for the general case in Sec.III, in the large |U| limit the effective Hamiltonian turns out to be independent of the energy E, the latter in the past recent years [32, 38–40], with the experimen- being bounded in the interval (−2κ, 2κ) for scattering tal demonstration of reflectionless potentials in arrays states. Hence for large |U| one can assume σ0 ' σ−ω2/U of evanescently-coupled optical waveguides with tailored and θ0 ' θ − ω2/U. Interestingly, with the choice coupling constants [39]. In optics, reflectionless defects 2 sustaining propagative bound states offer the possibility σ = θ − κ , ω = U(θ − κ) (22) to realize transparent optical intersections in photonic one has σ0 ' 0 and ω0 ' κ, i.e. the effective lattice in circuits [40]. Transparent defect modes are generally syn- Fig.1(b) is homogeneous and thus invisible. Note that, thesized by inverse scattering or supersymmetric meth- for θ < κ, invisibility is obtained for U < 0 and σ < 0. ods [32, 38], which require a careful control of hopping The onset of invisibility can be checked by exact calcula- amplitudes over several lattice sites. Here it is shown that tion of the spectral transmission and reflection coefficient a nearly invisible defect mode can be simply realized in for the lattice of Fig.1(b) following a standard procedure. an otherwise homogeneous tight-binding linear lattice ex- Let us look for a scattered state solution to the eigenvalue ploiting the hopping rate engineering method discussed equation (10) of the form in the previous section. Let us consider the linear tight- binding lattice S shown in Fig.2(a), which is side coupled exp(−iqn) + r(q) exp(iqn) n ≤ 0 c = (23) to one auxiliary site A (M = 1). For such a system we n t(q) exp(−iqn) n ≥ 1 have (H(S)) = κ(δ + δ ) + σ(δ δ + δ δ ) where q is the Bloch wave number, t(q) and r(q) are the n,m n,m+1 n,m−1 n,0 m,0 n,1 m,1 spectral transmission and reflection coefficients, respec- + (θ − κ)(δn,0δm,1 + δn,1δm,0) (16) tively, and E = 2κ cos(q) is the energy. The expressions H(A) = U (17) of t and r can be determined by writing coupled equations for amplitudes at sites |0iS and |1iS, i.e. ρ1,n = ω(δn,0 + δn,1). (18) 0 0 In the above equations, κ is the hopping rate between Ec0 = κc−1 + θ c1 + σ c0 (24) 0 0 adjacent sites in the main lattice S, with a defective hop- Ec1 = κc2 + θ c0 + σ c1. (25) 5

After substitution of the Ansatz (23) into Eqs.(24) and (25), coupled equations for r and t are obtained, which g=- iG can be solved for t yielding (a) (b) tight-binding continuum σ 0 2iκθ sin(q) exp(iq) 0 ... t(q) = . (26) > g=- iG κ κ κ κ κ [κ exp(iq) − σ0 + θ0][κ exp(iq) − σ0 − θ0] imaginary coupling continuum The dependence of the corrected hopping rate θ0 and site (c) (d) potential σ0 on energy E is determined by Eqs.(20) and G PT broken phase 2 0 1 S (21). Substitution of Eqs.(20) and (21) into Eq.(26) fi- II >S >S 2 >S κ κ κ ... nally yields 1 σ θ σ κ κ I 1 ω ω 2 A 2 0 U 2iκ[θ(E − U) + ω ] sin(q) exp(iq) 1 >A t(q) = . 2κ σ [κ exp(iq) − σ + θ]{(E − U)[κ exp(iq) − σ − θ] − 2ω2} (e) (27) 6 A typical behavior of t(E) (modulus and phase) for in- 4 ) t creasing values of |U|/κ and for θ = κ/5 are shown in ( P 2 Fig.2(c). The site potential σ and hopping rate ω are 0 chosen to satisfy the invisibility condition Eq.(22). Note 0 5 10 15 20 25 30 that, according to the theoretical prediction, a near in- normalized time κt visible defect over the entire tight-binding energy band is realized at increasing values of |U|/κ. FIG. 3. (Color online) (a) Schematic of a bound state cou- Bound states sustained by the lattice at the defective re- pled coupled to a continuum of states. When the energy of gion can be determined by looking at the poles of t(q) the bund state is not embedded in the continuous spectrum, in the complex q plane, with Im(q) < 0. Assuming that for small coupling g there is not a decay while the energy of the bound state is renormalized. The ghost regime of the Lee the conditions (22) are satisfied, after setting y = exp(iq) model is realized when the coupling g to the continuum is the condition t(q) = ∞ leads to the following algebraic imaginary (g = −iG). (b) Tight-binding lattice realization of (cubic) equation for y the Lee model in the ghost regime. The continuum of states is given by the Bloch modes of a semi-infinite homogeneous U 2θ U 2θ y3 + 1 − − y2 + 1 + y + 1 − = 0. (28) lattice. The end of the semi-lattice is attached by complex κ κ κ κ coupling g to the localized state |0i. (c) Phase diagram of the Lee model. Below the curve 2 the system is in the unbroken with the constraint |y| > 1. Assuming U < 0 and PT phase, with one bound state in region I (the ”physical” 0 ≤ θ < κ, such an equation admits of one acceptable particle state) and two bound states in region II (the ”physi- solution, corresponding to the existence of one bound cal” particle state plus the ”ghost”). Above curve 2 the PT state, for U < −2θ. Hence, in the near transparency symmetry is broken. Analytic equations of curves 1 and 2 p p 2 regime (i.e. for |U|/κ large), there is always one bound are G = κ −2 + (σ/κ) and G = κ −1 + (σ/2κ) , respec- state, corresponding in the physical lattice of FIg.1(a) to tively. (d) Implementation of the non-Hermitian (imaginary) coupling by an auxiliary site |1iA. (d) Numerically-computed high localization in the auxiliary site |1iA. evolution of the occupation probability P (t) as obtained from the exact Lee Hamiltonian (solid curve) and from the synthesized lattice (dashed curve). Parameter values are given in B. Fano-Anderson model with complex coupling the text.

The Fano-Anderson model [41, 42], also referred to as the Friedrichs-Lee model in quantum field theory [43], and quantum amplifiers [46, 47]. Let us discuss here the is ubiquitous in different areas of physics and describes Lee model with complex coupling, in which the Hamil- quite generally the coupling of a bund state to a con- tonian is not Hermitian but PT symmetric [30]. A sim- tinuum [Fig.3(a)]. A paradigmatic example of the Fano- ple tight-binding lattice realization of the Lee model in Anderson model, which is ofter encountered in the theory the ghost regime is shown in Fg.3(b) [31]. It consists of coherent transport in mesoscopic condensed-matter of a semi-infinite homogeneous tight-binding chain with systems and in integrated photonic systems, is provided (Hermitian) hopping amplitude κ between adjacent sites by the coupling of a localized state to an infinite or semi- and connected to the end to a node with on-site poten- infinite tight-binding lattice, i.e. to a continuous band tial σ > 2κ and with complex coupling g = −iG . The of Bloch modes [44, 45]. Hermitian coupling (hopping Hamiltonian of the tight-binding Lee model of Fig.3(b) amplitude g real) is the gold standard in such mod- reads ∞ els, however complexification of the coupling constant (g X imaginary) is of some interest in certain quantum mod- Hˆ = κ (|nihn + 1| + |n + 1ihn|) + els, such as the Lee model in the so-called ghost regime n=1 [30] or in the theory of the inverted quantum oscillators + σ|0ih0| + g(|0ih1| + |1ih0|) (29) 6 where g = −iG is the imaging coupling of the localized tive energy-dependent Hamiltonian for S is found state at site |0i with the semi-infinite tight-binding lattice. The phase space diagram of the Lee Hamiltonian 2 1 (S) ω X (29) is depicted in Fig.3(c) [31]. In the unbroken PT (Heff )n,m = H + δn,lδm,k. (33) n,m E − U phase, the lattice can sustain either one or two bound k,l=0 states. For a small coupling G [domain I in Fig.3(c)], the system shows a single bound state with energy slightly Since we wish to simulate the Lee Hamiltonian at energies modified from the unperturbed value σ and given by in proximity of the physical particle-V state, an approx- imate energy-independent Hamiltonian can be obtained 2 from Eq.(33) by letting E = E , where E is an energy σ/2 + p(σ/2)2 − G2 − 1 + (1 + G2)2 0 0 close to either the particle or ghost state energies E1,2. E1 = . (30) (1 + G2) σ/2 + p(σ/2)2 − G2 − 1 Taking for example E = E2, the effective Hamiltonian (33) reduces to the Lee Hamiltonian (29) with complex coupling provided that the site potentials U, σ and σ In the framework of the Lee model, such a state repre- 1 2 are tuned at the values sents the ”physical” particle state of the V fermion with renormalized mass [30]. However, as G is increased, in ω2 U = + E (34) addition to the ”physical” particle state a new bound θ + iG 2 state appears at the energy σ1 = σ + θ + iG (35) 2 σ = θ + iG. (36) σ/2 − p(σ/2)2 − G2 − 1 + (1 + G2)2 2 E2 = . (31) For example, let us consider the Lee Hamiltonian for pa- (1 + G2) σ/2 − p(σ/2)2 − G2 − 1 rameter values G/κ = 1.05 and σ/κ = 3, i.e. inside the domain II of Fig.3(b) and corresponding to the existence which is called a ”ghost” [domain II in Fig.3(c)]. As of two bound states (the physical V state and the ghost discussed in Ref.[31], the appearance of a ghost state in state). To implement such an Hamiltonian, we assume addition to the physical V-particle state can be detected ω/κ = 7, θ/κ = 0.2 and tune the values of U, σ1 and by monitoring the temporal evolution of the occupation σ2 according to Eqs.(34-36), namely σ1/κ = 3.2 + 1.05i, probability P (t) = |c (t)|2, when the system is initially 0 σ2/κ = 0.2 + 1.05i and U/κ ' 11 − 45i. To check the prepared in the bare V state, i.e. for cn(0) = δn,0: the fidelity of the synthesized Hamiltonian, in Fig.3(e) we existence of the ghost state is visualized as an undamped compare the numerically-computed evolution of the oc- oscillatory behavior of P (t) that arises from the interfer- 2 cupation probability P (t) = |c0(t)| with the initial con- ence of the physical and ghost states. dition cn(0) = δn,0, as obtained by the exact Lee Hamil- The most challenging issue toward an experimental im- tonian with complex coupling [Eq.(29)] and by the syn- plementation of the tight-binding lattice relies on the re- thesized effective Hamiltonian [Eq.(33)]. Note that the alization of the non-Hermitian coupling g = −iG of site oscillatory behavior of the occupation probability, aris- |0i with the semi-array. A few proposals have been pre- ing form the interference of the physical particle state viously suggest, based on a fast temporal modulation of and the ghost state, is satisfactorily reproduced by the complex on-site energies [31, 32], however such methods effective Hamiltonian. are rather challenging in practice and, as a matter of fact, to date there are not any experimental demonstration of non-Hermitian complex couplings in tight-binding networks. A simple way to synthesize a complex hopping, C. Bound states in the continuum in a PT -symmetric tight-binding lattice and thus the Lee Hamiltonian (29) in the ghost regime, is shown in Fig.3(d). The two terminating sites in the main lattice S are connected to an auxiliary site A with Bound states in the continuum (BIC) are quite anoma- complex potential U. The Hamiltonian of S is given by lous bound states with energy embedded into the continuous spectrum of scattered states. In simple terms, they (S) (H )n,m = κ(δn,m+1 + δn,m−1) + σ1δn,0δm,0 + σ2δn,1δm,1 can be viewed as resonances of zero width. Originally predicted by Von Neumann and Wigner in certain slowly- + (θ − κ)(δn,0δm,1 + δn,1δm,0) (32) decaying oscillating potentials [48], they have been later (n, m = 0, 1, 2, ....), where: κ is the hopping rate between found to arise from quite different mechanisms. In exper- adjacent sites, with a defective hopping rate θ between iments, BICs have been predicted and observed in a wide sites |0iS and |1iS; and σ1, σ2 are the potentials at sites range of physical systems, such as condensed-matter, |0iS and |1iS, respectively. The auxiliary site, with com- electromagnetic, optical, acoustical and hydrodynamic plex potential U, is connected to sites |0iS and |1iS via a systems [44, 49–53]. In particular, classical and quan- hopping amplitude ω. Following a similar procedure than tum tight-binding networks provide a fertile platform to the one discussed in the previous example (Sec.III.A), af- tailor the energy spectrum and to synthesize BIC modes ter elimination of the auxiliary site A the following effec- [44, 53, 54]. While most of previous studies on BIC states 7 have been limited to considering Hermitian systems, re- (a) cent works have extended the idea of BIC modes to PT - S symmetric non-Hermitian photonic networks [33, 55, 56], -3 -2 -1 0 1 >S >S >S -iU >S iU >S 2 >S 3 >S where they can appear either below or above the sym- ... κ ω ω κ ... -2 κ ω ω κ 3 metry breaking threshold. While BIC states above the -1 1 2 2 symmetry breaking threshold are quite common and are >A >A similar to bound states outside the continuum because they have complex energies [55], BIC modes in the un- A broken PT phase are less common and their synthesis (b) requires special lattice engineering [33]. In particular, a -3 -2 -1 0 1 2 non-Hermitian PT -symmetric lattice sustaining one BIC >S >S >S >S >S >S 3 >S mode below the symmetry breaking threshold can be ... κ κ θ θ κ κ ... -2 -1 σ 0 σ 1 σ 2 3 synthesized following the proposal of Ref.[33], however -1 0 1 complex hopping rates are required. The tight-binding (c) (d) Hamiltonian of the lattice is given by [33] 300

2κ 2 BOC R ∞ X κ Hˆ = (κn+1|n + 1ihn| + κn|nihn − 1|) (37) 200 n=−∞ 0 100 1 BIC with inhomogeneous hopping amplitudes given by spectrum κ 2 BOC 2 BOC p -2 κ

κ participationratio n (n + 1)/(n − 1) n even , n 6= 0 κ0 = −ig 2 BOC 0 = p 0 200 400 0 200 400 κ (n − 2)/n n odd , n 6= 1 κ1 = ig (38) mode number mode number where g > 0 is a real-valued parameter. Note that, FIG. 4. (Color online) (a) Schematic of the PT -symmetric since κn/κ → 1 as n → ±∞, the lattice is asymptotically homogeneous. It also satisfies the PT symmetry tight-binding lattice, with Hermitian hopping amplitudes and ∗ imaginary potentials ±iU at the auxiliary sites |1iA and |2iA, requirement κ−n = κn+1. The non-Hermitian nature of that admits of a bound state in the continuum. The hop- the lattice arises from the imaginary value of the hopping rates κn (n 6= 0, 1) are defined by Eq.(37) given in the ping amplitudes κ0 and κ1. The energy spectrum of text. (b) Equivalent tight-binding lattice model obtained ˆ H is real-valued for g ≤ gth = κ, i.e. PT symmetry after elimination of the two auxiliary sites. The energy- breaking occurs at gth = κ. In the unbroken PT phase dependent hopping amplitudes θ0,1 and site potentials σ0,1,2 2 2 (g < gth), the energy spectrum comprises, in addition to are given by θ0(E) = ω /(E + iU), θ1(E) = ω /(E − iU), 2 2 2 2 the continuous spectrum (−2κ, 2κ) of scattered states of σ−1(E) = ω /(E + iU), σ0(E) = 2Eω /(E + U ), and 2 the asymptotic homogeneous lattice, one BIC mode with σ1(E) = ω /(E−iU). (c) Numerically-computed energy spectrum of the tight-binding lattice of Fig.4(a) for ω/κ = 1 and algebraic localization at the energy E0 = 0, given by [33] U/κ = 0.4. The lattice comprises N = 403 sites. Eigenmodes  0 n odd are ordered for increasing values of eigenenergy. The two ar-  rows in the figure highlight the existence of four bound states cn = κ/g n = 0 (39) n+1 with energies in the gap, two above and the other two below  n √i n even, n 6= 0 |n| n2−1 the tight-binding energy band. (d) Numerically-computed participation ratio R of the lattice eigenmodes. Localized Here we suggest a simpler PT -symmetric tight-binding modes, corresponding to low values of R, are highlighted by lattice, which does not require complex hopping rates the arrows in the figure. The central eigenmode, with energy and that admits of the same BIC state. It comprises a E0 = 0 in the middle of the allowed band, corresponds to the tight-binding network S and two auxiliary sites |1iA and BIC state. |2iA in the geometrical setting of Fig.4(a). For such a system we can write κ δ + κ δ both n, m 6= 0 sites, according to Eq.(11) the following effective energy- H(S) = n n,m+1 n+1 n,m−1 n,m 0 either n, m = 0 dependent Hamiltonian is obtained (40) (H ) = H(S) eff n,m n,m (A) −iU 0 H = (41) ω2 0 iU + (δ δ + δ δ + δ δ ) E + iU n,−1 m,−1 n,−1 m,0 n,0 m,−1 ω2 ρα,n = ωδα,1(δn,−1 + δn,0) + ωδα,2(δn,0 + δn,1) (42) + (δ δ + δ δ + δ δ ) E − iU n,0 m,1 n,1 m,0 n,1 m,1 2 where κn (with n 6= 0, 1) are defined by Eq.(38), and ω, 2Eω + δ δ (43) U are real parameters. After elimination of the auxiliary E2 + U 2 n,0 m,0 8 which is illustrated in the scheme of Fig.4(b). Note that, Appendix A: Effective Hamiltonian description in for E = E0 = 0, the effective Hamiltonian (43) is equiv- the weak coupling limit alent to the Hamiltonian (37) with g = ω2/U, except for additional energy potentials ∓iω2/U at the odd sites In this Appendix we briefly discuss the possibility n = ±1 [Fig.4(b)]. Since the BIC mode of the Hamil- to derive an energy-independent effective Hamiltonian tonian (37) does not occupy odd sites of the lattice [see Hêff for the tight-binding network S after elimination Eq.(39)], it follows that the Hamiltonian (43), i.e. the of the degree of freedoms of the auxiliary system A. lattice depicted in Fig.4(a), has one BIC state at energy To this aim, let us introduce the vectors of amplitudes T T E0 = 0 as well. It should be noted that, owing to the c(t) = (...c−1, c0, c1, c2, ...) and a(t) = (a1, a2, a3, ....) dependence of Heff on the energy E, its energy spec- and write the coupled differential equations (6,7) in the trum is not equivalent to the one of the Hamiltonian (37), compact form even thought they admit the same BIC mode at energy dc E0 = 0. In particular, the lattices of Fig.4(a) sustains i = H(S)c + ρT a (A1) additional bound states in the gap, i.e., bound states out- dt da side the continuum (BOC), while the original Hamilto- i = H(A)a + ρc. (A2) nian (37) does not. We numerically computed the energy dt spectrum of the lattice of Fig.4(a) for ω/κ = 1 and for Equation (A2) can be formally integrated, yielding increasing values of the on-site potential U/κ, assuming typically N = 403 lattice sites with reflective boundary Z t conditions [57]. The spectrum turns out to be real (un- a(t) = −i dξ exp[iH(A)(ξ − t)]ρc(ξ). (A3) broken PT phase) for U/κ <∼ 0.46. As an example, 0 Fig.4(c) shows the numerically-computed spectrum for In writing Eq.(A3) we assumed that the auxiliary sites U/κ = 0.4. The degree of localization of the eigenstate are not excited at initial time, i.e. a(0) = 0. Such a cn(E) with energy E is measured by the participation condition is not necessary when the auxiliary cluster A P 2 2 P 4 ratio R(E), given by R(E) = ( n |cn| ) /( n |cn| ). is dissipative, i.e. all eigenvalues of H(A) have negative For localized modes, R ∼ 1 while for extended states imaginary part. Substitution of Eq.(A3) into Eq.(A1) R ∼ N. The distribution of R(E) for the N = 403 lat- yields the following integro-differential equation for c(t) tice eigenmodes is shown in Fig. 4(d). The figure clearly indicates the existence of one BIC state with algebraic dc Z t i = H(S)c − i dξρT exp[iH(A)(ξ − t)]ρc(ξ). (A4) localization at energy E0 = 0, together with four BOCs dt with exponential decay tails and with energies outside 0 the lattice band. The outer BOC states have an energy The standard derivation of an effective Hamiltonian from E1,2 ' ±2.202κ and E3,4 ' ±2.142κ. the integro-differential equation (A4) within Markovian approximation generally requires that [35] (i) the auxiliary cluster A has a continuous spectrum, i.e. M → ∞, and (ii) the S-A coupling is week, i.e. ρ → 0 (Weisskopf- Wigner approximation). For a finite number M of sites in A, such a reduction can not be generally accomplished. There is, however, a special case where it can be done and IV. CONCLUSIONS that deserves to be briefly mentioned, although it has some narrow application for the purpose of network en- In this paper we have suggested a simple method to gineering. Let us assume that (i) S is Hermitian, so they the eigenvalues of H(S) are real; (ii) A is non-Hermitian engineer a tight-binding quantum network by judicious (A) coupling to an auxiliary cluster. Remarkably, the tech- and dissipative, with all eigenvalues of H with nega- nique allows one to implement effective non-Hermitian tive imaginary part; (iii) S-A coupling is weak, i.e. ρ → 0. hopping rates with only complex on-site energies and In this case, it is worth considering the dynamics in the Hermitian couplings in both the network and auxiliary interaction picture. After setting cluster, avoiding the use of external time-dependent con- c(t) = exp(−iH(S)t)c˜(t), (A5) trol fields. As compared to other engineering methods, such as those based on inverse scattering, supersymme- form Eqs.(A4) and (A5) one obtains try or external time-dependent control fields, the method turns out to be rather simple and flexible for a practical d˜c i = −i exp(iH(S)t) × (A6) implementation. We have discussed three applications of dt the method to timely problems: the synthesis of a nearly Z t T (A) (S) transparent defect in an Hermitian linear lattice; the re- × dτρ exp(−iH τ)ρ exp[iH (τ − t)]˜c(t − τ). alization of the Fano-Anderson model with complex cou- 0 pling; and the synthesis of a PT -symmetric tight-binding In the ρ → 0 limit, c˜(t) varies slowly on time, and lattice with a bound state in the continuum. c˜(t − τ) under the sign of integral on the right hand 9 side of Eq.(A6) can be calculated at τ = 0, since where we have set exp(−iH(A)τ) → 0 and exp(iH(S)τ) remains limited at τ → ∞. After extending the upper integral limit on the right hand side of Eq.(A6) to ∞, the integro- differential equation(A6) simplifies into the following differential equation H = H(S) + Φ. (A10) d˜c eff i = exp(iH(S)t)Φ exp(−iH(S)t)˜c(t) (A7) dt where we have set Z ∞ Φ ≡ −i dτρT exp(−iH(A)τ)ρ exp(iH(S)τ) (A8) 0 Therefore, coupling with the auxiliary dissipative cluster A yields a correction to the the tight-binding Hamilto- In terms of the original amplitude c(t), using Eqs.(A5) nian of the network S, given by the term Φ defined by and (A7) one finally obtains Eq.(A8). However, since the above derivation holds in dc the weak coupling approximation, the correction Φ to i = H c(t) (A9) H(S) is generally a small one. dt eff

[1] D. D’Alessandro, Introduction to Quantum Control and [17] A. Gomez-Leon and G. Platero, Phys. Rev. Lett. 110, Dynamics (Chapman and Hall/CRC, London, 2007) 200403 (2013). [2] H. Rabitz, New J. Phys. 11, 105030 (2009). [18] T. Tanamoto, K. Ono, Yu-Xi Liu, and F. Nori, Sci. Rep. [3] P. Doria, T. Calarco, and S. Montangero, Phys. Rev. 5, 10076 (2015). Lett. 106, 190501 (2011). [19] N. Goldman and J. Dalibard, Phys. Rev. X 4, 031027 [4] E. Torrontegui, S. Martinez-Garaot, and J.G. Muga, (2014). Phys. Rev. A 89, 043408 (2014). [20] G.M. Nikolopoulos, D. Petrosyan, and P.J. Lambropou- [5] S. Bose, Phys. Rev. Lett. 91, 207901 (2003); S. Bose, los, J. Phys.: Condens. Matter 16, 4991 (2004); M. Bel- Contemp. Phys. 48, 13 (2007). lec, M. Georgios, M. Nikolopoulos, and S. Tzortzakis, [6] M. Christandl, N. Datta, A. Ekert, and A. J. Landahl, Opt. Lett. 37, 4504 (2012). Phys. Rev. Lett. 92, 187902 (2004). [21] N. Moiseyev, Non-Hermitian Quantum Mechanics (Cam- [7] A.D. Greentree, S.J. Devitt, and L.C.L. Hollenberg, bridge University Press, Cambridge, 2011) Phys. Rev. A 73, 032319 (2006). [22] C. Bender, Rep. Prog. Phys. 70, 947 (2007). [8] P. Rebentrost, I. Serban, T. Schulte-Herbrüggen,and F. [23] L. Jin and Z. Song, Phys. Rev. A 84, 042116 (2011). K. Wilhelm, Phys. Rev. Lett. 102, 090401 (2009). [24] H. Zhong, W. Hai, G. Lu, and Z. Li, Phys. Rev. A 84, [9] A. Kay, Int. J. Quantum Inf. 8, 641 (2010). 013410 (2011). [10] G.M. Nikolopoulos and I. Jex, Quantum State Transfer [25] X.Z. Zhang and Z. Song, Ann. Phys. 339, 109 (2013); X. and Network Engineering (Springer, Berlin, 2014). Lian, H. Zhong , Q. Xie, X. Zhou, Y. Wu, and W. Liao, [11] D. Jaksch and P. Zoller, New J. Phys. 5, 56 (2003); F. Eur. Phys. J. D 68, 189 (2014); S. Lin, X. Z. Zhang, and Dreisow, A. Szameit, M. Heinrich, T. Pertsch, S. Nolte, Z. Song Phys. Rev. A 92, 012117 (2015); J. Gong and A. Tünnermann,and S. Longhi, Phys. Rev. Lett. 101, Q.-H. Wang Phys. Rev. A 91, 042135 (2015); T.E. Lee 143602 (2008); S. Longhi, Phys. Rev. Lett. 101, 193902 and Y.N. Joglekar, Phys. Rev. A 92, 042103 (2015). (2008); M. Aidelsburger, M. Atala, S. Nascimbene, S. [26] B.T. Torosov, G. Della Valle, and S. Longhi, Phys. Rev. Trotzky, Y.A.Chen, and I. Bloch, Phys. Rev. Lett. 107 , A 87, 052502 (2013); G. Della Valle and S. Longhi, Phys. 255301 (2011); M. Aidelsburger, M. Atala, M. Lohse, J. Rev. A 87, 022119 (2013); X. Luo, J. Huang, H. Zhong, T. Barreiro, B. Paredes, and I. Bloch, Phys. Rev. Lett. X. Qin, Q. Xie, Y.S. Kivshar, and C. Lee, Phys. Rev. 111, 185301 (2013). Lett. 110, 243902 (2013); S. Longhi and G. Della Valle, [12] M. Müller,S. Diehl, G. Pupillo, and P. Zoller, Adv. Phys. Rev. A 87, 052116 (2013); S. Ibanez and J. G. Atom. Mol. Opt. Phys. 61,1 (2012). Muga, Phys. Rev. A 89, 033403 (2014); J. D’Ambroise, [13] A. Ajoy and P. Cappellaro, Phys. Rev. Lett. 110, 220503 B.A. Malomed, and P.G. Kevrekidis, Chaos 24, 023136 (2013); D. Hayes, S.T. Flammia, and M.J. Biercuk, New (2014); C. Yuce, Eur. Phys. J. D 69, 184 (2015). J. Phys. 16, 083027 (2014). [27] M. Znojil, J. Phys.: Conf. Ser. 624, 012011 (2015). [14] T.D. Stanescu, V. Galitski, J.Y. Vaishnav, C.W. Clark, [28] S. Zhang, Z. Ye, Y. Wang, Y. Park, G. Bartal, M. Mre- and S. Das Sarma, Phys. Rev. A 79, 053639 (2009). jen, X. Yin, and X. Zhang, Phys. Rev. Lett. 109, 193902 [15] Y.J. Lin, R.L. Compton, K. Jimenez-Garcia, J.V. Porto, (2012); P. Ginzburg, F.J. Rodrguez-Fortuno, A. Mar- and I.B. Spielman, Nature 462, 628 (2009). tinez, and A.V. Zayats, Nano Lett. 12, 6309 (2012); G. [16] N.H. Lindner, G. Refael, V. Galitski, Nature Phys. 7, 495 Castaldi, S. Savoia, V. Galdi, A. Alú,and N. Engheta, (2011). Phys. Rev. Lett. 110, 173901 (2013); C.M. Gentry and M.A. Popovic, Opt. Lett. 39, 4136 (2014); L. Feng, X. 10

Zhu, S. Yang, H. Zhu, P. Zhang, X. Yin, Y. Wang, and X. [48] J. von Neumann and E. Wigner, Z. Phys. 30, 465 (1929). Zhang, Opt. Express 22, 1760 (2014); H. Hodaei, M.-A. [49] W. Koch, J. Sound Vib. 88, 233 (1983); M.D. Groves, Miri, M. Heinrich, D.N. Christodoulides, and M. Kha- Math. Method. Appl. Sci. 21, 479 (1998); E. Davies and javikhan, Science 346, 975 (2014). L. Parnovski, Q. J. Mech. Appl. Math. 51, 477 (1998); [29] X. Z. Zhang, L. Jin, and Z. Song, Phys. Rev. A 87, S. Hein, W. Koch, and L. Nannen, J. Fluid Mech. 692, 042118 (2013) 257 (2012). [30] C.M. Bender, S.F. Brandt, J.-H. Chen, and Q. Wang, [50] M. Callan, C.M. Linton, and D.V. Evans, J. Fluid Mech. Phys. Rev. D 71, 025014 (2005). 229, 51 (1991); P.J. Cobelli, V. Pagneux, A. Maurel, and [31] S. Longhi and G. Della Valle, Phys. Rev. A 85, 012112 P. Petitjeans, EPL 88, 20006 (2009). (2012). [51] J.U. Nöckel, Phys. Rev. B 46, 15348 (1992); E.N. Bul- [32] S. Longhi, Phys. Rev. A 82, 032111 (2010). gakov, P. Exner, K.N. Pichugin, A.F. Sadreev, Phys. Rev. [33] S. Longhi, Opt. Lett. 39, 1697 (2014); S. Longhi and G. B 66, 155109 (2002); S. Longhi, J. Opt. Soc. Am. B Della Valle, Phys. Rev. A 89, 052132 (2014). 11, 1098 (1994); L.S. Cederbaum, R.S. Friedman, V.M. [34] S. Longhi, Phys. Rev. A 92, 042116 (2015). Ryaboy, and N. Moiseyev, Phys. Rev. Lett. 90, 013001 [35] L. Fonda, G. C. Ghirardi, and A. Rimini, Rep. Prog. (2003); M.L. Ladron de Guevara, F. Claro, P.A. Orel- Phys. 41, 587 (1978); P.L. Knight, M.A. Lauder, and lana, Phys. Rev. B 67, 195335 (2003); I. Rotter, A.F. B.J. Dalton, Phys. Rep. 190, 1 (1990); H. Nakazato, M. Sadreev, Phys. Rev. E 71, 046204 (2005); M.L. Ladron Namiki, and S. Pascazio, Int. J. Mod. Phys. B 10, 247 de Guevara, and P.A. Orellana, Phys. Rev. B 73, 205303 (1996); S. Longhi, J. Mod. Opt. 56, 729 (2009). (2006); W.-J. Gong, X.-Y. Sui, Y. Wang, G.-D. Yu, and [36] M.M. Sternheim and J.F. Walker, Phys. Rev. C 6, 114 X.-H. Chen, Nanoscale Res. Lett. 8, 330 (2013). (1972). [52] D.C. Marinica, A.G. Borisov, and S.V. Shabanov, Phys. [37] I. Rotter, Rep. Prog. Phys. 54, 635 (1991); V.V. Sokolov Rev. Lett. 100, 183902 (2008); E.N. Bulgakov and A.F. and V.G. Zelevinsky, Ann. Phys. 216, 323 (1992); G.G. Sadreev, Phys. Rev. B 78, 075105 (2008); M. I. Molina, Giusteri, F. Mattiotti, and G.L. Celardo, Phys. Rev. B A. E. Miroshnichenko, and Y. S. Kivshar, Phys. Rev. 91, 094301 (2015). Lett. 108, 070401 (2012); C. W. Hsu, B. Zhen, J. Lee, [38] A.A. Sukhorukov, Opt. Lett. 35, 989 (2010); S. Longhi S.-L. Chua, S. G. Johnson, J. D. Joannopoulos, and M. and G. Della Valle, Phys. Rev. B 84, 193105 (2011); Soljacic, Nature 499, 188 (2013); Y. Yang, C. Peng, Y. U.A. Khawaja and A.A. Sukhorukov, Opt. Lett. 40, 2719 Liang, Z. Li, and S. Noda, Phys. Rev. Lett. 113, 037401 (2015). (2014); B. Zhen, C.W. Hsu, L. Lu, A.D. Stone, and M. [39] A. Szameit, F. Dreisow, M. Heinrich, S. Nolte, and A.A. Soljacic, Phys. Rev. Lett. 113, 257401 (2014). Sukhorukov, Phys. Rev. Lett. 106, 193903 (2011). [53] F. Dreisow, A. Szameit, M. Heinrich, R. Keil, S. Nolte, [40] S. Longhi, Opt. Lett. 40, 463 (2015). A. Tünnermann,and S. Longhi, Opt. Lett. 34, 2405 [41] U. Fano, Phys. Rev. 124, 1866 (1961); P.W. Anderson, (2009); Y. Plotnik, O. Peleg, F. Dreisow, M. Heinrich, S. Phys. Rev. 164, 41 (1961). Nolte, A. Szameit, and M. Segev, Phys. Rev. Lett. 107, [42] G.D. Mahan, Many-Particle Physics (New York, Plenum 183901 (2011); G. Corrielli, G. Della Valle, A. Crespi, R. Press, 1990), pp. 272-285; M. Cini, Topics and Meth- Osellame, and S. Longhi, Phys. Rev. Lett. 111, 220403 ods in Condensed-Matter Theory (Springer, Heidelberg, (2013); S. Weimann, Y. Xu, R. Keil, A. E. Mirosh- 2007), Chap. 5, pp. 81-89. nichenko, A. Tunnermann, S. Nolte, A. A. Sukhorukov, [43] K.O. Friedrichs, Commun. Pure Appl. Math. 1, 361 A. Szameit, and Y. S. Kivshar, Phys. Rev. Lett. 111, (1948); T.D. Lee, Phys. Rev. 95, 1329 (1954); I. Pri- 240403 (2013). gogine, Phys. Rep. 219, 93 (1992). [54] K.-K. Voo and C.S. Chu, Phys.Rev. B 74, 155306 (2006); [44] P. A. Orellana, F. Dominguez-Adame, I. Gomez, and M. S. SreeRanjani. A.K. Kapoor, and P.K. Panigrahi, AIP L. Ladron de Guevara, Phys. Rev. B 67, 085321 (2003); Conf. Proc. 864, 236 (2006); J. M. Zhang, D. Braak, P. Orellana and F. Dominguez-Adame, Phys. Stat. Sol. and M. Kollar, Phys. Rev. Lett. 109, 116405 (2012); S. A 203, 1178 (2006); A.V. Malyshev, P.A.Orellana, and Longhi and G. Della Valle, Sci. Rep. 3, 2219 (2013); J. F. Dominguez-Adame, Phys. Rev. B 74, 033308 (2006); Mur-Petit and R.A. Molina, Phys. Rev. B 90, 035434 I. Rotter and A. F. Sadreev, Phys. Rev. E 69, 066201 (2014); C.-L. Zou, J.-M. Cui, F.-W. Sun, X. Xiong, X.- (2004); H. Nakamura, N. Hatano, S. Garmon, and T. B. Zou, Z.-F. Han, and G.-C. Guo, Laser & Photon. Rev. Petrosky, Phys. Rev. Lett. 99, 210404 (2007); S. Garmon, 9, 114 (2015). H. Nakamura, N. Hatano, and T. Petrosky, Phys. Rev. [55] A. Regensburger, M.-A. Miri, C. Bersch, J. Näger,G. On- B 80, 115318 (2009); S. Longhi,Eur. Phys. J. B 57, 45 ishchukov, D. N. Christodoulides, and U. Peschel, Phys. (2007); S. Longhi, Phys. Rev. B 75, 184306 (2007); R. Rev. Lett. 110, 223902 (2013). Farchioni, G.Grosso, and G.P. Parravicini, Eur. Phys. J. [56] Y.N. Joglekar, D.D. Scott, and A. Saxena, Phys. Rev. A B 84, 227 (2011). 90, 032108 (2014); M.I. Molina and Y.S. Kivshar, Studies [45] A.E. Miroshnichenko, S. Flach, and Y.S. Kivshar, Rev. Apple. Math. 133 337 (2014); S. Garmon, M. Gianfreda, Mod. Phys. 82, 2257 (2010). and N. Hatano, Phys. Rev. A 92, 022125 (2015). [46] W.H. Louisell, Radiation and Noise in Quantum Elec- [57] We checked that, for a sufficiently large number of sites tronics (Mc-Graw-Hill, New York, 1964), sec. 7.6. N, lattice truncation does not introduce boundary ef- [47] R.J. Glauber, Annals of the New York Academy of Sci- fects (such as surface states) and the energy spectrum is ences 480, 336 (1986). almost insensitive to the precise value of N.