This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Symmetry-protected flatband conditions for Hamiltonians with local symmetry

Jung-Wan Ryu Center for Theoretical Physics of Complex Systems, Institute for Basic Science (IBS), Daejeon 34126, Republic of Korea Basic Science Program, Korea University of Science and Technology (UST), Daejeon 34113, Republic of Korea    Alexei Andreanov [email protected] Center for Theoretical Physics of Complex Systems, Institute for Basic Science (IBS), Daejeon 34126, Republic of Korea Basic Science Program, Korea University of Science and Technology (UST), Daejeon 34113, Republic of Korea    Hee Chul Park [email protected] Department of Physics, Pukyong National University, Busan 48513, Republic of Korea    Jae-Ho Han [email protected] Center for Theoretical Physics of Complex Systems, Institute for Basic Science (IBS), Daejeon 34126, Republic of Korea
Abstract

We derive symmetry-based conditions for tight-binding Hamiltonians with flatbands to have compact localized eigenstates occupying a single unit cell. The conditions are based on unitary operators commuting with the Hamiltonian and are associated with local symmetries that guarantee compact localized states and a flatband. We illustrate the conditions for compact localized states and flatbands with simple Hamiltonians with given symmetries. We also apply the results to general cases such as a Hamiltonian with long-range hopping and a higher-dimensional Hamiltonian.

I Introduction

Hamiltonians with discrete translational symmetry can have dispersionless bands known as flatbands, with energy spectra E(𝒌)E(\bm{k}) independent of momentum 𝒌\bm{k} [1, 2, 3, 4, 5]. The flatness of a band implies zero group velocity, infinite effective mass of electrons, and suppressed electron and wave transport. The origin of flatbands is destructive interference due to a fine-tuning of the hopping or lattice symmetry. An important property of flatbands in short-range Hamiltonians is compact localized states (CLSs)—flatband eigenstates that are perfectly localized on a finite number of lattice sites. This is in contrast to Anderson localization where eigenstates are localized exponentially over the entire lattice [6]. Since the first report of a flatband in a dice lattice [7], a variety of flatband models have been identified, e.g., Lieb [8, 9, 10, 11, 12, 13], kagome [14, 15, 16, 17, 18, 19, 20], and honeycomb [21, 22, 23, 24] lattices. Despite their fine-tuned character and strong sensitivity to perturbations, flatbands have been realized in multiple experiments in different settings: superconducting networks [25, 26], photonic flatbands [27, 28, 29, 30, 31], optical lattices for cold atoms [32, 33, 34, 35, 36], and engineered atomic lattices [37, 38, 39].

Symmetry, one of the fundamental principles of physics, allows us to predict certain properties of a system without solving the often complicated underlying equations. In quantum mechanics, symmetry is associated with an operator that commutes with a Hamiltonian. A symmetry is global if the respective operator is independent of lattice sites, and a symmetry is local if the operator is dependent on lattice sites. It has been discovered that certain classes of local symmetries can indeed be systematically linked to CLSs and flatbands [3, 40, 41]. Although it is well known that flatbands and CLSs result from destructive interference caused by fine-tuning [42] or by specific symmetries [7, 8], a relation between fine-tuning and symmetry has not been fully established yet [43, 44]. In this work, we derive the exact conditions for CLSs occupying a single unit cell and the corresponding flatbands from the global and local symmetries of the system. We then propose a method to design lattice Hamiltonians with flatbands in terms of given symmetries and corresponding unitary operators. As a result, we demonstrate that if a Hamiltonian possesses a local symmetry for which the associated unitary operators are also operators of a global symmetry of the Hamiltonian, such Hamiltonian should have at least one compact localized state and corresponding flatband.

This paper is structured as follows. In Section II, we derive the conditions for CLSs and corresponding flatbands in a Hamiltonian with discrete translational symmetry in terms of both global and local symmetries. In Section III, we illustrate the method with several simple examples, including already-known flatband models. Section IV introduces generalizations of our method to longer-range hopping and higher dimensions. In Section V, we summarize and discuss our results.

II Flatbands generated by symmetries

Consider a one-dimensional (1D) tight-binding model with nearest-neighbor unit cell hopping. The Hamiltonian in the second quantized form is given by

\displaystyle\mathcal{H} =\displaystyle= i=1Nα,β=1m(c^iαH0;αβc^iβ+c^i+1αH1;αβc^iβ\displaystyle\sum_{i=1}^{N}\sum_{\alpha,\beta=1}^{m}\Big{(}\hat{c}_{i\alpha}^{\dagger}H_{0;\alpha\beta}\hat{c}_{i\beta}+\hat{c}_{i+1\alpha}^{\dagger}H_{1;\alpha\beta}\hat{c}_{i\beta}
+c^iαH1;βαc^i+1β)\displaystyle~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}+\hat{c}_{i\alpha}^{\dagger}H_{1;\beta\alpha}^{*}\hat{c}_{i+1\beta}\Big{)} (1)
=\displaystyle= i(c^iH0c^i+c^i+1H1c^i+c^iH1c^i+1),\displaystyle\sum_{i}\Big{(}\hat{c}_{i}^{\dagger}H_{0}\hat{c}_{i}+\hat{c}_{i+1}^{\dagger}H_{1}\hat{c}_{i}+\hat{c}_{i}^{\dagger}H_{1}^{\dagger}\hat{c}_{i+1}\Big{)},

where α=1,2,,m\alpha=1,2,\cdots,m labels the basis in the unit cell, c^iα\hat{c}_{i\alpha} (c^iα\hat{c}_{i\alpha}^{\dagger}) is an annihilation (creation) operator at unit cell ii and basis site α\alpha, and NN is the total number of cells. H0;αβH_{0;\alpha\beta} are on-site energies (α=β\alpha=\beta) and hopping constants (αβ\alpha\neq\beta) within the unit cell, and H1;αβH_{1;\alpha\beta} are inter-cell hoppings. The matrix notation with respect to the basis α\alpha is introduced in the second equation. Since we are interested in single-particle models, the statistics of c^iα\hat{c}_{i\alpha} (fermionic or bosonic) is irrelevant.

Let us assume that a given matrix H0H_{0} has a nontrivial finite symmetry group 𝒢\mathcal{G} of order larger than one, e.g., with more than one element. Then all matrices of a representation UiU_{i} of group 𝒢\mathcal{G} commute with the intra-cell Hamiltonian

[Ui(g),H0]=0,Ui(g)Ui(g)=1,g𝒢.\displaystyle[U_{i}(g),H_{0}]=0,\ \ \ U_{i}(g)^{\dagger}U_{i}(g)=1,\ \ \ \forall g\in\mathcal{G}. (2)

Here the subscript ii indicates the lattice site. Our goal is to systematically find H1H_{1} consistent with the symmetries 𝒢\mathcal{G}, such that the full Hamiltonian, Eq. (1), has flatbands. This consistency condition has important implications when the group 𝒢\mathcal{G} has more than one element. The symmetry operation on the total system can be represented as a product of these operators, Utot(g)=i=1NUi(g)U_{\mathrm{tot}}(g)=\otimes_{i=1}^{N}U_{i}(g). For a global symmetry, the operator acting on site ii Ui(g)=U(g)U_{i}(g)=U(g) is the same for all sites. Requesting that the Hamiltonian \mathcal{H} is invariant under global symmetry, one obtains the condition

[U(g),H1]=0,\displaystyle[U(g),H_{1}]=0, (3)

in addition to the condition in Eq. (2).

Refer to caption
Figure 1: (a) Three-band lattice model with discrete translational symmetry. H0H_{0} and H1H_{1} encode intra- and inter-cell hopping, respectively, and (l)(l) labels the unit cells. Schematic presentation of (b) global and (c) local symmetry transformations. σm\sigma_{m} and σm\sigma_{m^{\prime}} are unitary operators commuting with the total Hamiltonian, Eq. (1). For example, σm\sigma_{m} represents a 120120^{\circ} discrete rotation in the counterclockwise direction, and σm\sigma_{m^{\prime}} is a reflection with respect to a vertical axis.

Now, consider two different symmetries of H0H_{0}, g1g_{1} and g2g_{2}, both elements of 𝒢\mathcal{G}. Requesting that the total Hamiltonian is invariant under g1g_{1} and g2g_{2} globally, we have the following commutation relations,

[σ1,2,H0]=0,[σ1,2,H1]=0,\displaystyle[\sigma_{1,2},H_{0}]=0,\ \ \ [\sigma_{1,2},H_{1}]=0, (4)

where σ1,2=U(g1,2)\sigma_{1,2}=U(g_{1,2}). However, the presence of two distinct elements allows to construct more complex symmetries. In particular, to impose a CLS, we consider a kink operation, namely a symmetry Utot=i=1NσiU_{\mathrm{tot}}=\otimes_{i=1}^{N}\sigma_{i} acting as σ1\sigma_{1} for sites ili\leq l and σ2\sigma_{2} for il+1i\geq l+1, as shown in Fig. 1. We now require that the total Hamiltonian is invariant under the kink operation. Then at the location of the kink, we have the condition

σ2H1σ1=H1.\displaystyle\sigma_{2}^{\dagger}H_{1}\sigma_{1}=H_{1}. (5)

To show that this condition implies the existence of a CLS, we rewrite the condition as

H1(1σ21σ1)=0.\displaystyle H_{1}(1-\sigma_{2}^{-1}\sigma_{1})=0. (6)

Acting with this operator on eigenstates ϕn\phi_{n} of H0H_{0}, we find

H1ϕn=0,ϕn(1σ21σ1)ϕn.\displaystyle H_{1}\phi_{n}^{\prime}=0,\ \ \ \phi_{n}^{\prime}\equiv(1-\sigma_{2}^{-1}\sigma_{1})\phi_{n}. (7)

If ϕn=0\phi_{n}^{\prime}=0 for all nn, then (1σ21σ1)=0(1-\sigma_{2}^{-1}\sigma_{1})=0 or σ1=σ2\sigma_{1}=\sigma_{2}, due to the completeness of the eigenstates {ϕn}\{\phi_{n}\}. This contradicts the assumption of two different symmetries g1g_{1} and g2g_{2}. There is therefore at least one nontrivial state ϕn\phi_{n}^{\prime} that is a zero mode of H1H_{1} and that is also an eigenstate of H0H_{0}, since σ1,2\sigma_{1,2} commute with H0H_{0} [Eq. (4)]. Accordingly, the state ϕn\phi_{n}^{\prime} is an eigenstate of the total Hamiltonian with H1ϕn=0H_{1}\phi_{n}^{\prime}=0, and so it is a CLS.

We note that the operator σi\sigma_{i} is determined up to phase; see Eq. (2). The number of allowed phases is finite for a finite symmetry group. Consider an element gg of order NgN_{g},

gNg=e,Ng,\displaystyle g^{N_{g}}=e,\ \ \ N_{g}\in\mathbb{N}, (8)

where ee is the identity element. Let U(g)U(g) be the operator corresponding to gg. Then the operator eiθU(g)e^{i\theta}U(g), θ\theta\in\mathbb{R} also satisfies Eq. (2). However, the condition gNg=eg^{N_{g}}=e restricts the values of θ\theta, eiNgθ=1e^{iN_{g}\theta}=1, or θn=2πnNg\theta_{n}=2\pi\frac{n}{N_{g}}, n=0,1,,Ng1n=0,1,\cdots,N_{g}-1. Thus there can be NN operators

U(g),eiθ1U(g),eiθ2U(g),,eiθNg1U(g)\displaystyle U(g),\ e^{i\theta_{1}}U(g),\ e^{i\theta_{2}}U(g),\cdots,e^{i\theta_{N_{g}-1}}U(g) (9)

corresponding to the element gg. Due to the unitarity and homeomorphic property of U(g)U(g), the phase eiθe^{i\theta} is the same as the eigenvalues of σgU(g)\sigma_{g}\equiv U(g). Therefore, we can rewrite the flatband condition in Eq. (5) as

λ2iσ1H1=λ1jσ2H1,\displaystyle\lambda_{2}^{i}\sigma_{1}H_{1}=\lambda_{1}^{j}\sigma_{2}H_{1}, (10)

where λgi\lambda_{g}^{i} are eigenvalues of σg\sigma_{g} with i=1,2,3,,Ngi=1,2,3,\cdots,N_{g}. This is the condition that is mainly used in the following sections. From Eqs. (4) and (10), one can obtain the total Hamiltonian \mathcal{H} with flatbands.

III Obtaining H1H_{1} from given H0H_{0}

We illustrate the generic method outlined above by considering a given intra-cell hopping matrix H0H_{0}. First, we determine the symmetry group 𝒢\mathcal{G} of H0H_{0}, e.g., the unitary operators commuting with H0H_{0}. Then we identify inter-cell hopping matrices H1H_{1} from the given H0H_{0} and the associated unitary operators using the flatband condition derived in the previous section, Eq. (10). We note that setting H1=0H_{1}=0 (or equivalent) is also a solution; however, this corresponds to the trivial case of disconnected unit cells and we do not consider such cases in the derivations below.

III.1 2×22\times 2 Hamiltonian

We start with the simplest possible setting in which flatbands can appear: a 2×22\times 2 Hamiltonian with two bands,

=H0+eikH1+eikH1,\displaystyle\mathcal{H}=H_{0}+e^{-ik}H_{1}+e^{ik}H_{1}^{\dagger}, (11)

with H0H_{0} having parity symmetry

H0=(0110),H1=(t11t12t21t22).\displaystyle H_{0}=\begin{pmatrix}0&1\\ 1&0\end{pmatrix},H_{1}=\begin{pmatrix}t_{11}&t_{12}\\ t_{21}&t_{22}\end{pmatrix}. (12)

Our purpose is to find H1H_{1} that satisfies both the global and local symmetry constraints of the given H0H_{0}, and therefore \mathcal{H} has at least one dispersionless energy band. First, we identify the symmetries of H0H_{0}: there are two unitary operators,

σ0=(1001),σ1=(0110),\displaystyle\sigma_{0}=\begin{pmatrix}1&0\\ 0&1\end{pmatrix},\ \sigma_{1}=\begin{pmatrix}0&1\\ 1&0\end{pmatrix},\ (13)

commuting with H0H_{0}. We note that an identity matrix σ0\sigma_{0} is always a unitary operator commuting with any Hamiltonian.

Given these symmetry generators, the flatband conditions for H1H_{1}[Eqs. (4) and (10)] are given by

[σ0,H1]=0,[σ1,H1]=0,σ1H1=±σ0H1=±H1.\displaystyle\left[\sigma_{0},H_{1}\right]=0,\left[\sigma_{1},H_{1}\right]=0,\sigma_{1}H_{1}=\pm\sigma_{0}H_{1}=\pm H_{1}. (14)

These conditions can also be obtained considering three symmetry operators, σ0\sigma_{0}, σ1\sigma_{1}, and σ1-\sigma_{1} instead of the two symmetry operators with phases. Resolving the above flatband constraints with respect to tijt_{ij}, we find the following hopping matrix:

H1=(t11±t11±t11t11).\displaystyle H_{1}=\begin{pmatrix}t_{11}&\pm t_{11}\\ \pm t_{11}&t_{11}\end{pmatrix}. (15)

This corresponds to the well-known case of a cross-stitch lattice with a single flatband [3] for t110t_{11}\neq 0 and all elements of the same sign.

The other symmetry groups for the two-band case are studied systematically in Appendix B.

III.2 3×33\times 3 Hamiltonian

As the next example, consider a 3×33\times 3 Hamiltonian with symmetric all-to-all hopping within the unit cell,

=H0+eikH1+eikH1,\displaystyle\mathcal{H}=H_{0}+e^{-ik}H_{1}+e^{ik}H_{1}^{\dagger}, (16)

where we choose

H0=(011101110),H1=(t11t12t13t21t22t23t31t32t33).\displaystyle H_{0}=\begin{pmatrix}0&1&1\\ 1&0&1\\ 1&1&0\end{pmatrix},H_{1}=\begin{pmatrix}t_{11}&t_{12}&t_{13}\\ t_{21}&t_{22}&t_{23}\\ t_{31}&t_{32}&t_{33}\end{pmatrix}. (17)

The symmetry group 𝒢\mathcal{G} of H0H_{0} consists of three symmetry operators,

σ0=(100010001),σ1=(100001010),σ2=(010001100).\displaystyle\sigma_{0}=\begin{pmatrix}1&0&0\\ 0&1&0\\ 0&0&1\end{pmatrix},\sigma_{1}=\begin{pmatrix}1&0&0\\ 0&0&1\\ 0&1&0\end{pmatrix},\sigma_{2}=\begin{pmatrix}0&1&0\\ 0&0&1\\ 1&0&0\end{pmatrix}. (18)

The flatband condition from Eq. (10) in this case is more complicated compared to the case of the 2×22\times 2 Hamiltonian because there are three distinct unitary operators in the symmetry group that commute with H0H_{0}. Conditions can be formulated for four different combinations of unitary operators: (σ0\sigma_{0}, σ1\sigma_{1}), (σ0\sigma_{0}, σ2\sigma_{2}), (σ1\sigma_{1}, σ2\sigma_{2}), and (σ0\sigma_{0}, σ1\sigma_{1}, σ2\sigma_{2}). Consequently, H0H_{0} satisfies the following relations:

[σ0,H0]\displaystyle\left[\sigma_{0},H_{0}\right] =0,[σ1,H0]=0for(σ0,σ1)\displaystyle=0,\left[\sigma_{1},H_{0}\right]=0~{}~{}\mathrm{for}~{}~{}(\sigma_{0},\sigma_{1}) (19)
[σ0,H0]\displaystyle\left[\sigma_{0},H_{0}\right] =0,[σ2,H0]=0for(σ0,σ2)\displaystyle=0,\left[\sigma_{2},H_{0}\right]=0~{}~{}\mathrm{for}~{}~{}(\sigma_{0},\sigma_{2}) (20)
[σ1,H0]\displaystyle\left[\sigma_{1},H_{0}\right] =0,[σ2,H0]=0for(σ1,σ2)\displaystyle=0,\left[\sigma_{2},H_{0}\right]=0~{}~{}\mathrm{for}~{}~{}(\sigma_{1},\sigma_{2}) (21)
[σ0,H0]\displaystyle\left[\sigma_{0},H_{0}\right] =0,[σ1,H0]=0,[σ2,H0]=0for(σ0,σ1,σ2).\displaystyle=0,\left[\sigma_{1},H_{0}\right]=0,\left[\sigma_{2},H_{0}\right]=0~{}\mathrm{for}~{}(\sigma_{0},\sigma_{1},\sigma_{2}). (22)

The flatband conditions for H1H_{1} are

[σ0,H1]\displaystyle\left[\sigma_{0},H_{1}\right] =0,[σ1,H1]=0,σ0H1=±σ1H1for(σ0,σ1)\displaystyle=0,\left[\sigma_{1},H_{1}\right]=0,\sigma_{0}H_{1}=\pm\sigma_{1}H_{1}~{}~{}~{}~{}~{}\mathrm{for}~{}~{}(\sigma_{0},\sigma_{1}) (23)
[σ0,H1]\displaystyle\left[\sigma_{0},H_{1}\right] =0,[σ2,H1]=0,σ0H1=ei2jπ/3σ2H1(j=0,1,2)for(σ0,σ2)\displaystyle=0,\left[\sigma_{2},H_{1}\right]=0,\sigma_{0}H_{1}=e^{-i2j\pi/3}\sigma_{2}H_{1}~{}(j=0,1,2)~{}~{}~{}~{}~{}\mathrm{for}~{}~{}(\sigma_{0},\sigma_{2}) (24)
[σ1,H1]\displaystyle\left[\sigma_{1},H_{1}\right] =0,[σ2,H1]=0,σ1H1=±ei2jπ/3σ2H1(j=0,1,2)for(σ1,σ2)\displaystyle=0,\left[\sigma_{2},H_{1}\right]=0,\sigma_{1}H_{1}=\pm e^{-i2j\pi/3}\sigma_{2}H_{1}~{}(j=0,1,2)~{}~{}~{}~{}~{}\mathrm{for}~{}~{}(\sigma_{1},\sigma_{2}) (25)
[σ0,H1]\displaystyle\left[\sigma_{0},H_{1}\right] =0,[σ1,H1]=0,[σ2,H1]=0,\displaystyle=0,\left[\sigma_{1},H_{1}\right]=0,\left[\sigma_{2},H_{1}\right]=0, (26)
σ0H1\displaystyle\sigma_{0}H_{1} =±σ1H1,σ0H1=ei2jπ/3σ2H1,σ1H1=±ei2jπ/3σ2H1(j=0,1,2)for(σ0,σ1,σ2).\displaystyle=\pm\sigma_{1}H_{1},\sigma_{0}H_{1}=e^{-i2j\pi/3}\sigma_{2}H_{1},\sigma_{1}H_{1}=\pm e^{-i2j\pi/3}\sigma_{2}H_{1}~{}(j=0,1,2)~{}~{}~{}~{}~{}\mathrm{for}~{}~{}(\sigma_{0},\sigma_{1},\sigma_{2}). (27)

Resolving the above conditions with respect to the hopping matrix H1H_{1}, parameterized as follows,

H1\displaystyle H_{1} =(t11t12t13t21t22t23t31t32t33),\displaystyle=\begin{pmatrix}t_{11}&t_{12}&t_{13}\\ t_{21}&t_{22}&t_{23}\\ t_{31}&t_{32}&t_{33}\end{pmatrix}, (28)

we find the following solutions,

H1\displaystyle H_{1} =(t11t12t12t21t22t22t21t22t22)or(0000t22t220t22t22),for(σ0,σ1)\displaystyle=\begin{pmatrix}t_{11}&t_{12}&t_{12}\\ t_{21}&t_{22}&t_{22}\\ t_{21}&t_{22}&t_{22}\end{pmatrix}\mathrm{or}\begin{pmatrix}0&0&0\\ 0&t_{22}&-t_{22}\\ 0&-t_{22}&t_{22}\end{pmatrix},~{}~{}~{}~{}~{}\mathrm{for}~{}~{}(\sigma_{0},\sigma_{1}) (29)
H1\displaystyle H_{1} =(t11t11t11t11t11t11t11t11t11)or(t11ei2π/3t11e±i2π/3t11e±i2π/3t11t11ei2π/3t11ei2π/3t11e±i2π/3t11t11)\displaystyle=\begin{pmatrix}t_{11}&t_{11}&t_{11}\\ t_{11}&t_{11}&t_{11}\\ t_{11}&t_{11}&t_{11}\end{pmatrix}\mathrm{or}\begin{pmatrix}t_{11}&e^{\mp i2\pi/3}t_{11}&e^{\pm i2\pi/3}t_{11}\\ e^{\pm i2\pi/3}t_{11}&t_{11}&e^{\mp i2\pi/3}t_{11}\\ e^{\mp i2\pi/3}t_{11}&e^{\pm i2\pi/3}t_{11}&t_{11}\end{pmatrix} (30)

for the remaining cases. For the choice of two unitary operators (σ0\sigma_{0}, σ1\sigma_{1}), with the hopping t12=1t_{12}=1 (or t21=1t_{21}=1) and all other hoppings set to zero for the first H1H_{1} in the above, the Hamiltonian corresponds to a diamond chain with a vertical link that has a tunable flatband [3]. If we consider special cases in which all matrix elements of H1H_{1} are equal to each other in Eq. (30), there is always only one dispersive band (see Appendix A).

IV Generalizations

Similarly as in the previous section, one can extend the analysis to the case of an arbitrary number of bands. Our approach can be extended to longer-range hopping and higher dimensions, as we demonstrate below.

IV.1 Hamiltonian with long-range hopping

We consider a 1D Hamiltonian with long-range hopping,

H=H0+j=1L(eikjHj+eikjHj),\displaystyle H=H_{0}+\sum_{j=1}^{L}\left(e^{-ikj}H_{j}+e^{ikj}H_{j}^{\dagger}\right), (31)

where LL is the longest hopping range. In the example case of three bands, the hopping matrices can be parameterized as

H0=(ϵ11ϵ12ϵ13ϵ21ϵ22ϵ23ϵ31ϵ32ϵ33),Hj=(tj11tj12tj13tj21tj22tj23tj31tj32tj33).\displaystyle H_{0}=\begin{pmatrix}\epsilon_{11}&\epsilon_{12}&\epsilon_{13}\\ \epsilon_{21}&\epsilon_{22}&\epsilon_{23}\\ \epsilon_{31}&\epsilon_{32}&\epsilon_{33}\end{pmatrix},H_{j}=\begin{pmatrix}t_{j11}&t_{j12}&t_{j13}\\ t_{j21}&t_{j22}&t_{j23}\\ t_{j31}&t_{j32}&t_{j33}\end{pmatrix}. (32)

We impose the following two symmetries on H0H_{0}, expressed as unitary operators,

σ0=(100010001),σ1=(100001010).\displaystyle\sigma_{0}=\begin{pmatrix}1&0&0\\ 0&1&0\\ 0&0&1\end{pmatrix},\sigma_{1}=\begin{pmatrix}1&0&0\\ 0&0&1\\ 0&1&0\end{pmatrix}. (33)

Then H0H_{0} satisfies the relation

[σ0,H0]=[σ1,H0]=0,\displaystyle\left[\sigma_{0},H_{0}\right]=\left[\sigma_{1},H_{0}\right]=0, (34)

and the flatband conditions for H1H_{1} read

[σ0,Hj]=0,[σ1,Hj]=0,σ1Hj=±σ0Hj=±Hj.\displaystyle\left[\sigma_{0},H_{j}\right]=0,\left[\sigma_{1},H_{j}\right]=0,\sigma_{1}H_{j}=\pm\sigma_{0}H_{j}=\pm H_{j}. (35)

From these constraints, it follows that

H0\displaystyle H_{0} =(ϵ11ϵ12ϵ12ϵ21ϵ22ϵ23ϵ21ϵ23ϵ22)\displaystyle=\begin{pmatrix}\epsilon_{11}&\epsilon_{12}&\epsilon_{12}\\ \epsilon_{21}&\epsilon_{22}&\epsilon_{23}\\ \epsilon_{21}&\epsilon_{23}&\epsilon_{22}\end{pmatrix} (36)

and

Hj\displaystyle H_{j} =(tj11tj12tj12tj21tj22tj22tj21tj22tj22)or(0000tj22tj220tj22tj22).\displaystyle=\begin{pmatrix}t_{j11}&t_{j12}&t_{j12}\\ t_{j21}&t_{j22}&t_{j22}\\ t_{j21}&t_{j22}&t_{j22}\end{pmatrix}\mathrm{or}\begin{pmatrix}0&0&0\\ 0&t_{j22}&-t_{j22}\\ 0&-t_{j22}&t_{j22}\end{pmatrix}. (37)

There is one flatband with EFB=ϵ22ϵ23E_{\mathrm{FB}}=\epsilon_{22}-\epsilon_{23} for the first choice of HjH_{j}, or two flatbands with EFB=(ϵ11+ϵ22+ϵ238ϵ12ϵ21+(ϵ11+ϵ22+ϵ23)2)/2E_{\mathrm{FB}}=(\epsilon_{11}+\epsilon_{22}+\epsilon_{23}\mp\sqrt{8\epsilon_{12}\epsilon_{21}+(-\epsilon_{11}+\epsilon_{22}+\epsilon_{23})^{2}})/2 for the second choice of HjH_{j}. An example of a 1D flatband Hamiltonian with next-nearest hopping terms derived using our method is presented in Appendix C.

IV.2 2D Hamiltonian

A two-dimensional (2D) generalization of the 1D Hamiltonian in Eq. (11) is given by

H=H0+eikxH1+eikxH1+eikyH2+eikyH2.\displaystyle H=H_{0}+e^{-ik_{x}}H_{1}+e^{ik_{x}}H_{1}^{\dagger}+e^{-ik_{y}}H_{2}+e^{ik_{y}}H_{2}^{\dagger}. (38)

In the example case of three bands, the hopping matrices can be parameterized as

H0=(ϵ11ϵ12ϵ13ϵ21ϵ22ϵ23ϵ31ϵ32ϵ33),\displaystyle H_{0}=\begin{pmatrix}\epsilon_{11}&\epsilon_{12}&\epsilon_{13}\\ \epsilon_{21}&\epsilon_{22}&\epsilon_{23}\\ \epsilon_{31}&\epsilon_{32}&\epsilon_{33}\end{pmatrix}, (39)
H1=(t11t12t13t21t22t23t31t32t33),H2=(s11s12s13s21s22s23s31s32s33).\displaystyle H_{1}=\begin{pmatrix}t_{11}&t_{12}&t_{13}\\ t_{21}&t_{22}&t_{23}\\ t_{31}&t_{32}&t_{33}\end{pmatrix},H_{2}=\begin{pmatrix}s_{11}&s_{12}&s_{13}\\ s_{21}&s_{22}&s_{23}\\ s_{31}&s_{32}&s_{33}\end{pmatrix}.

We impose the following two symmetries on H0H_{0}, expressed as unitary operators,

σ0=(100010001),σ1=(100001010),\displaystyle\sigma_{0}=\begin{pmatrix}1&0&0\\ 0&1&0\\ 0&0&1\end{pmatrix},\sigma_{1}=\begin{pmatrix}1&0&0\\ 0&0&1\\ 0&1&0\end{pmatrix}, (40)

and coinciding with the same symmetries imposed in the previous example. Then H0H_{0} satisfies the relation

[σ0,H0]=[σ1,H0]=0\displaystyle\left[\sigma_{0},H_{0}\right]=\left[\sigma_{1},H_{0}\right]=0 (41)

and the flatband conditions [Eq. (10)] for H1H_{1} read

[σ0,H1]=0,[σ1,H1]=0,σ1H1=±σ0H1=±H1,\displaystyle\left[\sigma_{0},H_{1}\right]=0,\left[\sigma_{1},H_{1}\right]=0,\sigma_{1}H_{1}=\pm\sigma_{0}H_{1}=\pm H_{1}, (42)
[σ0,H2]=0,[σ1,H2]=0,σ1H2=±σ0H2=±H2.\displaystyle\left[\sigma_{0},H_{2}\right]=0,\left[\sigma_{1},H_{2}\right]=0,\sigma_{1}H_{2}=\pm\sigma_{0}H_{2}=\pm H_{2}. (43)

From these constraints, H0H_{0} follows as

H0\displaystyle H_{0} =(ϵ11ϵ12ϵ12ϵ21ϵ22ϵ23ϵ21ϵ23ϵ22).\displaystyle=\begin{pmatrix}\epsilon_{11}&\epsilon_{12}&\epsilon_{12}\\ \epsilon_{21}&\epsilon_{22}&\epsilon_{23}\\ \epsilon_{21}&\epsilon_{23}&\epsilon_{22}\end{pmatrix}. (44)

There are two distinct solutions for H1,2H_{1,2}. The first one gives

H1=(t11t12t12t21t22t22t21t22t22),H2=(s11s12s12s21s22s22s21s22s22),\displaystyle H_{1}=\begin{pmatrix}t_{11}&t_{12}&t_{12}\\ t_{21}&t_{22}&t_{22}\\ t_{21}&t_{22}&t_{22}\end{pmatrix},H_{2}=\begin{pmatrix}s_{11}&s_{12}&s_{12}\\ s_{21}&s_{22}&s_{22}\\ s_{21}&s_{22}&s_{22}\end{pmatrix}, (45)

having one flatband with EFB=ϵ22ϵ23E_{\mathrm{FB}}=\epsilon_{22}-\epsilon_{23}. The second solution for H1,2H_{1,2} reads

H1=(0000t22t220t22t22),H2=(0000s22s220s22s22),\displaystyle H_{1}=\begin{pmatrix}0&0&0\\ 0&t_{22}&-t_{22}\\ 0&-t_{22}&t_{22}\end{pmatrix},H_{2}=\begin{pmatrix}0&0&0\\ 0&s_{22}&-s_{22}\\ 0&-s_{22}&s_{22}\end{pmatrix}, (46)

having two flatbands with EFB=(ϵ11+ϵ22+ϵ238ϵ12ϵ21+(ϵ11+ϵ22+ϵ23)2)/2E_{\mathrm{FB}}=(\epsilon_{11}+\epsilon_{22}+\epsilon_{23}\mp\sqrt{8\epsilon_{12}\epsilon_{21}+(-\epsilon_{11}+\epsilon_{22}+\epsilon_{23})^{2}})/2. We note that these flatbands are the same as those in the previous 1D long-range Hamiltonian case.

An example of a combined 2D Hamiltonian with a cross-stitch chain along the x-axis and a tunable diamond chain along the y-axis is presented in Appendix D.

V Summary

We derived the conditions for lattice Hamiltonians with flatbands to have compact localized eigenstates that localize perfectly in a single unit cell. The conditions are based on unitary operators commuting with the Hamiltonian and are associated with local symmetries that guarantee compact localized states and flatbands. Beyond flatbands in lattice models, we can also apply our results to perturbed Hamiltonians where some internal states are not affected by additional perturbations (see Appendix E). We expect that the conditions derived here can be extended to design extraordinary states robust against local perturbations or environmental changes in a variety of coupled systems.

Acknowledgments

The authors thank Emil Yuzbashyan for helpful discussions. We acknowledge financial support from the Institute for Basic Science in the Republic of Korea through the project IBS-R024-D1.

Appendix A Uniform H1H_{1}

The n×nn\times n Hamiltonian for special cases with uniform H1H_{1} is

=H0+eikH1+eikH1,\displaystyle\mathcal{H}=H_{0}+e^{-ik}H_{1}+e^{ik}H_{1}^{\dagger}, (47)

where H0H_{0} is the n×nn\times n Hamiltonian with zero on-site energies and where all connected intra-cell hopping strengths are the same, i.e.,

H0=(011101110),H1=(ttttttttt).\displaystyle H_{0}=\begin{pmatrix}\ddots&\vdots&\vdots&\vdots&\reflectbox{$\ddots$}\\ \cdots&0&1&1&\cdots\\ \cdots&1&0&1&\cdots\\ \cdots&1&1&0&\cdots\\ \reflectbox{$\ddots$}&\vdots&\vdots&\vdots&\ddots\end{pmatrix},H_{1}=\begin{pmatrix}\ddots&\vdots&\vdots&\vdots&\reflectbox{$\ddots$}\\ \cdots&t&t&t&\cdots\\ \cdots&t&t&t&\cdots\\ \cdots&t&t&t&\cdots\\ \reflectbox{$\ddots$}&\vdots&\vdots&\vdots&\ddots\end{pmatrix}. (48)

The 2×22\times 2 case is the cross-stitch lattice in Section III.1. Similar to the 2×22\times 2 case, there is a trivial solution H1=0H_{1}=0, which we dismiss. For the nontrivial case, the Hamiltonian has nn bands given by

ED\displaystyle E_{\mathrm{D}} =(n1)+2ntcosk,\displaystyle=(n-1)+2nt\cos{k}, (49)
EFB\displaystyle E_{\mathrm{FB}} =1,\displaystyle=-1, (50)

where EDE_{\mathrm{D}} and EFBE_{\mathrm{FB}} represent dispersive and flat band energies, respectively. The number of dispersive and flat bands is 11 and n1n-1, respectively. These special cases are also covered by our flatband conditions.

Appendix B Generic 2×22\times 2 Hamiltonian: Deriving H0H_{0} and H1H_{1} from symmetries

In the 2×22\times 2 Hamiltonian case above, we only considered parity symmetric cases with two unitary operators, σ0\sigma_{0} and σx\sigma_{x}, commuting with H0H_{0}. A generic 2×22\times 2 Hermitian Hamiltonian can be written in terms of three Pauli matrices σx\sigma_{x}, σy\sigma_{y}, σz\sigma_{z} and the identity matrix σ0\sigma_{0}. Accordingly, the possible combinations of symmetries are as follows: (σ0,σx)(\sigma_{0},\sigma_{x}), (σ0,σy)(\sigma_{0},\sigma_{y}), (σ0,σz)(\sigma_{0},\sigma_{z}), (σx,σy)(\sigma_{x},\sigma_{y}), (σx,σz)(\sigma_{x},\sigma_{z}), (σy,σz)(\sigma_{y},\sigma_{z}), (σ0,σx,σy)(\sigma_{0},\sigma_{x},\sigma_{y}), (σ0,σx,σz)(\sigma_{0},\sigma_{x},\sigma_{z}), (σ0,σy,σz)(\sigma_{0},\sigma_{y},\sigma_{z}), (σx,σy,σz)(\sigma_{x},\sigma_{y},\sigma_{z}), and (σ0,σx,σy,σz)(\sigma_{0},\sigma_{x},\sigma_{y},\sigma_{z}). The 2×22\times 2 case we obtained in Section III.1 corresponds to the first combination, (σ0,σx)(\sigma_{0},\sigma_{x}). Below we consider all other possible combinations of symmetries. The three combinations (σ0,σx)(\sigma_{0},\sigma_{x}), (σ0,σy)(\sigma_{0},\sigma_{y}), and (σ0,σz)(\sigma_{0},\sigma_{z}) give similar conditions:

[σ0,x,H0]=0,[σ0,x,H1]=0for(σ0,σx)\displaystyle\left[\sigma_{0,x},H_{0}\right]=0,\left[\sigma_{0,x},H_{1}\right]=0~{}~{}~{}~{}~{}\mathrm{for}~{}~{}(\sigma_{0},\sigma_{x}) (51)
[σ0,y,H0]=0,[σ0,y,H1]=0for(σ0,σy)\displaystyle\left[\sigma_{0,y},H_{0}\right]=0,\left[\sigma_{0,y},H_{1}\right]=0~{}~{}~{}~{}~{}\mathrm{for}~{}~{}(\sigma_{0},\sigma_{y}) (52)
[σ0,z,H0]=0,[σ0,z,H1]=0for(σ0,σz)\displaystyle\left[\sigma_{0,z},H_{0}\right]=0,\left[\sigma_{0,z},H_{1}\right]=0~{}~{}~{}~{}~{}\mathrm{for}~{}~{}(\sigma_{0},\sigma_{z}) (53)

from Eq. (4) and

σ0H1=±σxH1for(σ0,σx)\displaystyle\sigma_{0}H_{1}=\pm\sigma_{x}H_{1}~{}~{}~{}~{}~{}\mathrm{for}~{}~{}(\sigma_{0},\sigma_{x}) (54)
σ0H1=±σyH1for(σ0,σy)\displaystyle\sigma_{0}H_{1}=\pm\sigma_{y}H_{1}~{}~{}~{}~{}~{}\mathrm{for}~{}~{}(\sigma_{0},\sigma_{y}) (55)
σ0H1=±σzH1for(σ0,σz)\displaystyle\sigma_{0}H_{1}=\pm\sigma_{z}H_{1}~{}~{}~{}~{}~{}\mathrm{for}~{}~{}(\sigma_{0},\sigma_{z}) (56)

from Eq. (10). Parameterizing H0H_{0} and H1H_{1} as

H0=(ϵ11ϵ12ϵ21ϵ22),H1=(t11t12t21t22)\displaystyle H_{0}=\begin{pmatrix}\epsilon_{11}&\epsilon_{12}\\ \epsilon_{21}&\epsilon_{22}\end{pmatrix},H_{1}=\begin{pmatrix}t_{11}&t_{12}\\ t_{21}&t_{22}\end{pmatrix} (57)

and imposing the flatband conditions Eqs. (51)–(56), we find the following solutions

H0\displaystyle H_{0} =(ϵ11ϵ12ϵ12ϵ11),H1=(t11±t11±t11t11),\displaystyle=\begin{pmatrix}\epsilon_{11}&\epsilon_{12}\\ \epsilon_{12}&\epsilon_{11}\end{pmatrix},H_{1}=\begin{pmatrix}t_{11}&\pm t_{11}\\ \pm t_{11}&t_{11}\end{pmatrix}, (58)
H0\displaystyle H_{0} =(ϵ11iϵ12iϵ12ϵ11),H1=(t11it11±it11t11),\displaystyle=\begin{pmatrix}\epsilon_{11}&i\epsilon_{12}\\ -i\epsilon_{12}&\epsilon_{11}\end{pmatrix},H_{1}=\begin{pmatrix}t_{11}&\mp it_{11}\\ \pm it_{11}&t_{11}\end{pmatrix}, (59)
H0\displaystyle H_{0} =(ϵ1100ϵ22),H1=(t11000)or(000t22),\displaystyle=\begin{pmatrix}\epsilon_{11}&0\\ 0&\epsilon_{22}\end{pmatrix},H_{1}=\begin{pmatrix}t_{11}&0\\ 0&0\end{pmatrix}\mathrm{or}\begin{pmatrix}0&0\\ 0&t_{22}\end{pmatrix}, (60)

for the choices of symmetry operators (σ0,σx)(\sigma_{0},\sigma_{x}), (σ0,σy)(\sigma_{0},\sigma_{y}), and (σ0,σz)(\sigma_{0},\sigma_{z}), respectively. The corresponding flatband energies are ϵ11ϵ12\epsilon_{11}\mp\epsilon_{12}, ϵ11ϵ12\epsilon_{11}\mp\epsilon_{12}, and ϵ22\epsilon_{22} (or ϵ11\epsilon_{11}), respectively. Interestingly, the Hamiltonian of Eq. (60) can be obtained by unitary transformations of the Hamiltonian of Eq. (58), i.e., detangling flatbands [3].

Considering the three combinations of (σx,σy)(\sigma_{x},\sigma_{y}), (σx,σz)(\sigma_{x},\sigma_{z}), and (σy,σz)(\sigma_{y},\sigma_{z}) and using the same methods as above to resolve the hopping matrices, we obtain H0=ϵ11σ0H_{0}=\epsilon_{11}\sigma_{0} and tij=0(i,j=1,2)t_{ij}=0~{}(i,j=1,2) in all cases. These are trivial cases of flatbands with no inter-cell hopping. As a consequence, the remaining combinations with three Pauli operators also give rise to trivial flatbands because these combinations always contain at least two Pauli matrices, similar to the case just discussed. As a result, Eqs. (58)–(60) are 2×22\times 2 Hermitian Hamiltonians with flatbands including trivial cases.

Refer to caption
Figure 2: (a) 2D lattice model combining a cross-stitch lattice with a tunable diamond lattice. (b) Energy bands of the 2D lattice model. (c) 1D lattice model with different nearest and next-nearest hoppings. (d) Energy bands of the lattice model of (c).

Appendix C Hamiltonian with further neighbor hopping

In Section IV.1 we considered an example of a 1D flatband Hamiltonian with next-nearest hopping terms,

H=H0+eikH1+eikH1+e2ikH2+e2ikH2.\displaystyle H=H_{0}+e^{-ik}H_{1}+e^{ik}H_{1}^{\dagger}+e^{-2ik}H_{2}+e^{2ik}H_{2}^{\dagger}. (61)

Imposing the symmetries σ0\sigma_{0} and σ1\sigma_{1} in Eq. (33) makes H0H_{0}, H1H_{1}, and H2H_{2} take the forms as in Eqs. (36) and (37). An example is a tunable diamond lattice with additional next-nearest hopping terms, which are described by the matrices

H0=(011101110),H1=(000011011),H2=(000100100).\displaystyle H_{0}=\begin{pmatrix}0&1&1\\ 1&0&1\\ 1&1&0\end{pmatrix},H_{1}=\begin{pmatrix}0&0&0\\ 0&1&1\\ 0&1&1\end{pmatrix},H_{2}=\begin{pmatrix}0&0&0\\ 1&0&0\\ 1&0&0\end{pmatrix}. (62)

Figure 2 (a) and (b) show this 1D lattice with different nearest and next-nearest hoppings and energy bands, among which one band is flat.

Appendix D Combined 2D Hamiltonians

In Section IV.2 we considered an example of a 2D flatband Hamiltonian,

H=H0+eikxH1+eikxH1+eikyH2+eikyH2.\displaystyle H=H_{0}+e^{-ik_{x}}H_{1}+e^{ik_{x}}H_{1}^{\dagger}+e^{-ik_{y}}H_{2}+e^{ik_{y}}H_{2}^{\dagger}. (63)

Imposing symmetries σ0\sigma_{0} and σ1\sigma_{1} in Eq. (40), H0H_{0}, H1H_{1}, and H2H_{2} take the forms in Eqs. (44), (45), and (46), respectively. As an example, a cross-stitch lattice with a tunable diamond lattice can be constructed from the same matrices as in Eq. (62). Figure 2 (c) and (d) show this 2D lattice model and energy bands, among which one band is flat.

Appendix E Hamiltonian with unperturbed internal states

Beyond flatbands in translationally lattice systems, our method can also be applied to a perturbed Hamiltonian,

H=H0+δHp,\displaystyle H=H_{0}+\delta H_{p}, (64)

where HpH_{p} describes the perturbation and δ\delta is the perturbation strength. If we obtain HpH_{p} satisfying our conditions from a given H0H_{0}, some eigenstates of HH are not affected by the perturbation as δ\delta increases. For example, we consider the perturbed Hamiltonian with

H0=(011101110),Hp=(011100100).\displaystyle H_{0}=\begin{pmatrix}0&1&1\\ 1&0&1\\ 1&1&0\end{pmatrix},H_{p}=\begin{pmatrix}0&1&1\\ 1&0&0\\ 1&0&0\end{pmatrix}. (65)

The eigenvalues are 1-1 and (1±9+8δ(2+δ))/2(1\pm\sqrt{9+8\delta(2+\delta)})/2. One of the eigenvalues is irrespective of the perturbation strength δ\delta since HpH_{p} satisfies the condition of Eq. (10).

References