Manipulation of edge states in microwave artificial graphene

Edge states are one important ingredient to understand transport properties of graphene nanoribbons. We study experimentally the existence and the internal structure of edge states under uniaxial strain of the three main edges: zigzag, bearded, and armchair. The experiments are performed on artificial microwave graphene flakes, where the wavefunctions are obtained by direct imaging. We show that uniaxial strain can be used to manipulate the edge states: a single parameter controls their existence and their spatial extension into the ribbon. By combining tight-binding approach and topological arguments, we provide an accurate description of our experimental findings. A new type of zero-energy state appearing at the intersection of two edges, namely the corner state, is also observed and discussed.


Introduction
Edge states play an essential role in condensed matter physics for both fundamental aspects and electronic transport applications. Recently, due to their immunity against disorder or impurities (absence of backscattering), topologically protected edge states have raised interest in quantum spin Hall systems [1,2,3] as well as in topological insulators [4,5]. Edge states have been first predicted in graphene ribbons [6,7,8] and later observed along "zigzag" edges [9,10]. Although not strictly speaking topologically protected, edge states in graphene possess a topological origin coined by the Zak phase and remain robust against weak chiral symmetry perturbations [11,12].
These peculiar features are due to the multicomponent (spinorial) structure of the wave function. While for scalar wave functions (like for free particles in a box), hard wall boundary conditions impose the vanishing of the wave function at the edges, for a two-component wave function only one component has to vanish, leaving the possibility of a finite amplitude for the other component. In graphene, the two components are simply the amplitudes of the wave function on each of the two atoms of the bipartite honeycomb lattice.
The richness of edge-state physics is not limited to condensed matter. Any finite system characterized by multi-component wavefunctions can constitute a good candidate. Over the past few years, pertinent realizations of artificial graphene have emerged in various contexts such as 2D electron gases in molecular assembly [13] or in nanopatterned semiconductor [14], ultracold atoms in optical lattice [15], polaritons in semiconductor microcavity [16] as well as light and microwaves in photonic crystals [17,18,19,20] (see [21] for a recent review). The main advantages of these analogue systems reside in the high tunability and control regarding their lattice properties. The synthetic honeycomb lattices offer the possibility to investigate phenomena that are hardly reachable in genuine graphene, and more particularly those appearing at the edges. For instance, edge states at bearded terminations (not stable in real graphene) have been first observed in photonic lattices [22]. As pointed out in [23,24], the manipulation and control of edge states may lead to promising photonic applications.
Moreover, graphene band structure, and consequently transport properties, can be engineered via lattice strain [25,26,27,28,29]. While not observable in the realm of genuine graphene, bandgap opening and associated topological phase transition, in uniaxial strain honeycomb structures, as well as pseudo-magnetic field creation via inhomogeneous strain, have been observed in various artificial systems [13,15,17,20,30].
In this paper, we propose an experimental manipulation of edge state properties by controlling an uniaxial strain. We use a photonic implementation of honeycomb lattice in the microwave regime [20,31]. Fig. 1(a) shows a realization of an artificial graphene ribbon exhibiting the three usual types of edges, namely zigzag, bearded and armchair. The sites of the lattice are occupied by dielectric microwave resonators with a cylindrical shape. The resonance frequency of an isolated resonator ν 0 is around 6.65 GHz and corresponds to the on-site energy of atoms in a tight-binding model. The dielectric cylinders are coupled by an evanescent magnetic field, so that the wave propagation between the resonators is well described by a tight-binding-like hopping term. The coupling strength t between two resonators depends on their separation d and varies from t = 0.05 GHz to t = 0.3 GHz when d varies from 15 to 11 mm. Via a reflection measurement, one has access, at each site, to the local density of states and to the wavefunction intensity associated to each eigenfrequency. The density of states (DOS) is obtained by averaging the local density of states over all resonator positions. The experimental setup and the tight-binding description of the microwave artificial graphene are detailed in [31]. The paper is organized as follows. In Section 2, we first focus on zigzag and bearded boundary geometries. We show experimentally how uniaxial strain acts as a switch between zigzag and bearded edge states. Based on a tight-binding analysis, a diagram of existence of edge states is theoretically proposed. We recall in Section 3 the topological origin of the three types of considered edge states, namely zigzag, bearded and armchair. A geometrical analysis in the k-space allows to predict the presence of edge states and their evolution under strain. Section 4 is dedicated to armchair geometries. A quantitative experimental and theoretical analysis is done. The existence of a new type of states, appearing at the intersection of two type of edges, namely corner states, is eventually discussed.

Zigzag and bearded edges in honeycomb lattice under uniaxial strain
The lattice presented in Fig. 1(a) exhibits three different edges: Armchair, zigzag and bearded. We will consider ribbons uniaxially strained along one lattice axis. The strain only changes one of the three hopping parameters which will be denoted by t . We introduce the anisotropy parameter β = t /t. Armchair edges are along the strain axis and consequently will not support any edge state whatever the anisotropy as will be discussed in Sec. 3. Fig. 1(b) shows a typical DOS measured in an unstrained ribbon, i.e. β = 1. The Dirac frequency ν D defines the frequency origin [31]. The peak observed at the origin corresponds to "zero-energy" modes, we will call them "zeromodes" in the following. Experimentally, we can extract the intensity distributions by means of reflection measurements (see [31] for details). To go beyond this qualitative discussion, we propose in the following part a theoretical description by means of a tight-binding model. Although both zigzag and bearded edges are built experimentally on the same ribbon sample, we will consider theoretically two independent semi-infinite lattices. We demonstrated in a previous work the validity of the TB model to describe our artificial graphene realization [31]. Here, to address the issue of edge states, we will restrict the TB model to first nearest neighbor couplings. Let us first consider a zigzag edge [ Fig. 3(a)]. The A−B unit cell, appropriated to describe such an edge, is represented by the dashed box. The edge is built by translating the dimer with vectors m a 2 (m integer). According to the Bloch theorem, an additional phase e imk is acquired after m translations. Here, k ≡ √ 3k a, where a is the lattice spacing and k is the 1D wave vector pertaining to the edge. The 1D Brillouin zone is defined by k ∈ −π/ √ 3a, π/ √ 3a corresponding to k ∈ [−π, π]. For a given m, the bulk sites correspond to translations of the dimer with vectors −n a 1 (n positive integer). The presence of the edge breaks the translation symmetry and the Bloch theorem do not apply anymore. We label the site positions with the index n from n = 0 (edge) to n = N → ∞ (bulk) [see Fig. 3 Taking into account only nearest-neighbor couplings, each sublattice gives a tightbinding recurrence equation which defines the zero-modes [8]: i t i A i = 0 and i t i B i = 0, where i counts the three nearest-neighbors of a given B-site and Asite respectively, t i the corresponding coupling strength, and A i and B i denoting the amplitudes at the corresponding sites. In Fig. 3(a), the missing column of sites nearest to the edge belongs to the A sublattice: A −1 ≡ 0. Thus, for zigzag edge, the tight-binding recurrence implies that all A-sites are identically null. For the B-sites, amplitudes of zero-modes fulfill the following condition: The intensity reads: which defines a localisation length ξ zz (k, β) plotted with a red colorscale in Fig. 3(f). The zigzag edge states exist when: Conversely, for the bearded edges, sketched in Fig. 3(b), all B-sites are identically null and one obtains for non-vanishing amplitude A: and From Eqs. (3) and (6), one obtains the existence diagram of edge states depicted in Fig. 3(c). The red and green color scales give respectively the spatial extension of zigzag and bearded edge states extracted from Eqs. (2) and (5). For β = 1, the diagram shows that zigzag zero-modes occupy 1/3 of the 1D Brillouin zone whereas the proportion is 2/3 for the bearded states as well-known [6,7,8,12]. Thus, the TB prediction for the ratio between number of bearded and zigzag states is 2. The diagram also shows that the zero-modes are more localized along zigzag edges than along bearded edges. The square dots in the Fig. 2 (bottom row) represent normalized measured intensities integrated over vertical lines of resonators. Green and red zones are used as a guide for the eyes and indicate the bulk extension of the bearded and zigzag zeromodes respectively. The total intensity in each zone is proportional to the number of corresponding zero-modes [31]. For β = 1, the measured ratio is 2.3 in close agreement with the expected value and the bearded edge states have indeed a larger extension. For β = 0.4, we observe a majority of strongly localized bearded edge states and only a few extended zigzag zero-modes. For β = 1.5, both edge states are equally present, the zigzag ones being slightly more localized. Finally, for β = 2.5, only zigzag zero-modes exist (the intensity along the bearded edge is strictly zero) with a larger extension compare to the previous case. These observations are in good accordance with the features depicted in the diagram of Fig 3(c) and demonstrate that the anisotropy parameter allows for an accurate manipulation of edges states.
In the next section, we propose a topological argument, first presented in [11,12], to address the existence of zero-modes, not only for bearded and zigzag but also for armchair edges.

Zak phase
A simple geometrical way to describe the existence of edge states is to relate their existence to a topological quantity which is a 1D winding number called the Zak phase [32]. The spectrum of an infinite ribbon of finite width containing M dimers A−B consists in 2M one-dimensional bands j (k ). Therefore for each wave vector k , there are 2M states. Among these 2M states, two zero-modes may be localized along the edges. We recall below the relation between their existence and the Zak phase [11,12].
where the function f (k) describes the coupling between atoms of one sublattice with the three nearest neighbors belonging to the other sublattice (see for instance ref. [33]). The wavefunctions have the form where φ(k) = arg[f (k)] and ± corresponds to positive and negative energies. The winding of the relative phase φ(k) in the reciprocal space has quite interesting properties. First of all, as seen on Fig. 4, around each Dirac point, the phase rotates by ±2π. The circulation of φ(k) along a surrounding closed path is quantized: the Berry phase is defined by 1/2 dk ∇ k φ(k) = ±π [34]. Our purpose here is to stress that the winding of the φ(k) phase in reciprocal space carries an additional information related to the existence of edge states. Let us consider a ribbon geometry and define the directions parallel ( ) and perpendicular (⊥) to the ribbon length. For a ribbon of width containing M dimers, the perpendicular wavevector of a bulk state k ⊥ is quantized by the boundary conditions. For a two-component wavefunction, the quantization condition reads k ⊥ (M +1)a−φ(k) = κπ with κ = 1, · · · , M [12,35]. This equation has M or M −1 solutions depending on the winding of the phase under variation of k ⊥ . The missing solution corresponds to an edge state. More precisely one has the correspondance where the Zak phase Z(k ) is the phase accumulated in the first 1D Brillouin zone (BZ) along the k ⊥ direction: Therefore existence of edge states may be read immediately from a plot of the phase φ(k , k ⊥ ) as plotted in Fig. 4. For a given k , if the rotation of the phase is 2π, there is an edge state, if the total rotation is 0, there is no edge state. The function φ(k) depends on the boundary conditions (BC). For a given type of edge, the ribbon is constructed by the translations of an elementary dimer [dotted box in Fig. 3(a, b) and in Fig. 7(a)], so that the writing of the bulk Hamiltonian has a form which depends on the BC [11,12]. Let us consider first the zigzag BC. The function f (k) reads in this case where the two elementary vectors a 1 and a 2 are shown on Fig. 3(a). Similarly, the functions f (k) for the bearded [ Fig. 3 Fig. 7(a)] cases read respectively Fig. 4 shows φ(k) for the different BC and for various deformations characterized by the parameter β. It is seen that when the distortion increases, the density of edge states increases in the zigzag case, while it decreases in the bearded case as indicated by the red and green zone respectively. In the armchair case, while there are no edge states in the undistorted lattice, as is well-known [6,7,8], they appear in the presence of a distortion which is not parallel to the edge. If the distortion is along the edge, the function f (k) reads f ac (k) = β +e ik·a 1 +e ik·a 2 instead of (12). The corresponding plot of φ(k , k ⊥ ) is shown on Fig. 5 and we see that Z(k ) = 0 everywhere, as all lines of phase jumps are parallel to k ⊥ , confirming the absence of edge states in this case. A thorough experimental investigation of armchair zero-modes is presented in the following section.

Study of armchair edges in honeycomb lattice under uniaxial strain
In previous work, we addressed the issue of topological phase transition in strained artificial graphene and incidentally observed armchair edge states [20]. Here, to perform a quantitative study, one consider ribbons with only armchair edges, the anisotropy axis being along their widths. Fig. 6 shows the zero-modes for β ranging from 0.4 to 3.5. It is worth noting that no zero-modes appear on armchair edges along the anisotropy axis whatever the value of β. Moreover, for the case β = 1, the ribbon does not support any armchair edge states. These features were already observed with an hexagonal flake in Ref. [20] and are in accordance with the Zak phase analysis developed in Sec. 3. For β = 1, zero-modes along the two oblique edges are clearly present. Two main features have to be stressed: (i ) Edge states live only on one of the two triangular sub-lattices. The excited sub-lattice depends whether β is less or larger than 1, as seen in Fig. 6(a) and Fig. 6(c-f) respectively. The switch between the two sub-lattices is clearly visible in the insets. (ii ) The zero-mode localization length along the edges depends on the anisotropy and, for β > 1, decreases with β.
To understand the observations (i ) and (ii ), we propose a tight-binding analysis comparable to what has been reported in Sec. 2. Let us consider a semi-infinite honeycomb lattice with armchair edge [ Fig. 7(a)]. The appropriate A − B unit cell is represented by the dashed box. Now, the armchair edge is built by translating the dimer with vectors m (a 1 + a 2 ) (m integer). Here, k ≡ 3k a, where a is the lattice spacing and k is the 1D wave vector pertaining to the armchair edge. The 1D Brillouin zone is defined by k ∈ −π 3a , π 3a corresponding to k ∈ [−π, π]. The bulk lattice is built by translating the dimer with vectors n a 2 (n positive integer). The tight-binding recurrence equations for zero-modes read [see Fig. 7

(a)]:
A n+2 + A n+1 + βA n e ik = 0, According to Eq. (13), for β > 1, the amplitude increases with n so that A-sites cannot support any edge states, we have A n ≡ 0, ∀n. Inversely, for β < 1, Eq. (14) implies B n ≡ 0. Here, we have provided a simple explanation of the switching mechanism experimentally observed and described above. For β > 1, with B 0 = 1 as initial condition and B 1 = −e ik /β to ensure B i<0 ≡ 0, the intensities on B-sites are given by: with For |r + | < |r − | < 1, one obtain decaying solutions with : The diagram of existence of armchair edge states, defined by condition (18), is plotted in Fig. 7(b). The blue color scale gives the transverse spatial extension of the armchair zero-modes extracted from Eq. (15). First, as pointed out in Section 3 and observed in Fig. 6(b) armchair edges along the parallel axis do not support any edge states for β = 1. Then, we see that their existence is not related to the topological transition observed in Ref. [20]: zero-modes appear as soon as β = 1. In the case β = 0.4, A-sites support zero-modes occupying a limited range of k, as given by Eq. (18), almost half of them being well localized (dark blue colors in the diagram). For 1 < β < β c = 2, the  edge states supported by the B-sites are still belonging to a finite range of k and are mainly delocalized (light blue colors in the diagram), as observed in the Fig. 6(b) and (c). For β > 2, zero-modes run over the full 1D Brillouin zone and their localization lengths decrease with β. To be more quantitative, we plotted in Fig. 8 the intensities of B n (resp. A n ), for β > 1 (resp. β < 1) and n ranging from 0 to 5. Fig. 8(a) shows the experimental profiles extracted from Fig. 6 by integrating over vertical lines of sites. Fig. 8(b) depicts the intensities calculated from Eqs. (15 -17) with an integration of k over all possible values defined by Eq. (18). For β = 2.5 (green diamonds) and β = 3.5 (red triangles), the agreement is very good. For β = 0.4 (blue circles) and β = 1.8 (orange squares), the difference between measured and calculated profiles is more important. To explain these discrepancies, it is worth mentioning that for β > β c zero-modes actually appear in the bandgap and are therefore clearly isolated in the DOS. When β < β c , bulk states are very close to the Dirac point and spoil the zero-mode intensity extraction. Moreover, the calculations are performed with semi-infinite lattices whereas experiments use finite ribbons. Consequently, in the experiments, we have to take into account a discrete sampling of the 1D Brillouin zone, the precise k-values depending on the boundary conditions. Thus, the discrete sum of such zero-modes having different localization lengths can lead to a larger overall extension compared to the continuous case. Here again, β stands as control parameter allowing notably to switch edge states from one sublattice to the other.
To conclude this section, we would like to raise the existence of a new type of zero-modes in finite size systems located at the intersection of two edges, namely corner states. The comparison between Fig. 6(a) and 6(f) clearly underlines that the distribution of the zero-mode along the parallel edge is homogeneous for β = 3.5 (i.e. all edge B-sites are illuminated) whereas the intensity decreases from the top-right corner for β = 0.4. Once more, a tight-binding analysis for zero-modes allows to obtain non-homogeneous edge states. In Fig. 9(a), we build for β < 1 the amplitude of a  Fig. 7(b)], this k = 0 decaying solution dominates the overall intensity distribution. On the contrary, for β = 3.5, such decaying solutions may also exist for k = 0 but are compensated by uniform solutions associated with larger k-values, thus leading to an homogeneous overall intensity distribution. For β > 2, a different type of corner states appear at the intersection between armchair and zigzag edges for β > 2 as can be seen in Fig. 2(d). Such corner states, which will be the subject of further studies, present strong similarities with zero-modes associated to localized impurities in anisotropic graphene where the hierarchy of their amplitudes follows the rules of a Pascal triangle [36] [see Fig. 9(b)].

Conclusion
We have studied edge states in artificial microwave graphene ribbons under uniaxial strain. By directly imaging the zero-modes, we have shown how the anisotropy parameter allows to create and manipulate edge states. Based on a tight-binding analysis and supported by topological arguments (Zak phase), we have drawn diagrams of existence and localization length variation for zigzag, bearded and armchair edge states in remarkable agreement with experiments. We also discuss the existence of a new type of zero-modes at the intersection of two type of edges, namely the corner states whose study is left for a future work.