Non-Abelian line graph: A generalized approach to flat bands

Abstract

Line graph (LG) lattices are well known to host flat bands (FBs) with isotropic hop**s in $s$-orbital models. Yet, higher-angular-momentum orbitals with spin-orbit coupling (SOC), which are more common in real materials, lack a general approach for their inclusion in LG to achieve FBs. Here, we introduce a non-Abelian LG theory to construct FBs in realistic systems, which incorporates internal degrees of freedom and goes beyond real-valued isotropic hop**s. The lattice edges and sites in the LG are modified to be associated with arbitrary Hermitian matrices, referred to as the multiple LG. A crucial step is to map the multiple LG Hamiltonian to a tight-binding (TB) model that respects the lattice symmetry through appropriate local non-Abelian transformations in the internal space. We find the general conditions to determine the local transformations. Based on this mechanism, we discuss the realization of $d$-orbital FBs in the Kagome lattice, which may serve as a minimal model for understanding the high-orbital FBs with SOC in Kagome materials. Our approach bridges the known FBs in pure lattice models and the realization in multi-orbital systems.

The quenching of kinetic energy in the flat band (FB) system renders it an ideal platform for investigating diverse correlated phenomena such as ferromagnetism [1, 2], superconductivity and superfluidity [3, 4, 5, 6], Wigner crystal [7] and fractional quantum Hall effect [8, 9, 10, 11, 12, 13, 14, 15]. The destructive interference is a common origin of the FB, which suppress the kinetic energy and leads to the localization of the electrons [16, 17]. In an early seminal work, Mielke identified a special class of lattice known as the line graph (LG) lattice that could naturally lead to destructive interference and therefore produce a FB [1, 2], as guaranteed by the LG theorem [17, 18, 19]. The typical examples are Kagome [20, 16], checkerboard [8, 21] and pyrochlore lattice [22, 23, 16]. Recently, realizing FBs in real systems has attracted significant attention in both theoretical and experimental research, where Kagome stands out as a prominent representative. Notably, both DFT calculations and experiments reveal that Kagome materials have higher-angular-momentum orbitals (high orbitals) near the Fermi surface and the spin-orbit coupling (SOC) is usually presented [24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35], which significantly differ from the isotropic hop**s from $s$-electrons as originally assumed. Meanwhile, FBs are also observed in Moiré systems, arising from the Moiré potential and band folding. Twisted bilayer graphene hosts approximate chiral FBs near charge neutrality [36, 37, 38, 39, 40, 41, 42, 43, 44]. More interestingly, the bands in twisted TMD systems exhibit high-orbital characteristics [45, 46, 47], which is reminiscent of the honeycomb lattice with $p_{x},p_{y}$ orbitals [7, 48]. However, the original LG approach focuses on pure lattice degree of freedom with isotropic hop**s, and its application to systems with internal degrees of freedom (such as orbitals) remains to be studied.

In this work, we develop a non-Abelian LG theory to incorporate internal degrees of freedom. We first generalize the uniform coupling constant (Fig. 1(a)) on each edge to a Hermitian matrix (Fig. 1(b)). The extension is referred to as the multiple LG, which allows the application of the LG theorem to the systems with internal space. When the LG corresponds to a periodic system, the multiple LG serves as the counterpart of the conventional LG to preserve the FB nature. To match the anisotropic hop**s in real materials, we perform local U$(n)$ unitary transformations in the internal space to drive the Hamiltonian beyond the isotropic multiple LG, which gives rise to a non-Abelian LG (Fig. 1(c)). Further, we set up general conditions to distinguish non-Abelian LGs from the tight-binding (TB) models (Fig. 1(d)), where all the realistic lattice symmetries are respected. To demonstrate our theory, we study the $d$-orbital Kagome models, which may provide insights into Kagome materials.

Figure 1: (a) The construction of a LG. Black and green lines belong to the root graph and the LG. The green edges and red loops represent hop**s and on-site energies in the TB model. The “dumbbell” is employed for convenience. (b) The multiple LG. The couplings are generalized to matrices. (c) Schematic depiction of a non-Abelian LG. The triangle corresponds to the gray region of (a). The couplings in different directions are anisotropic but related by “rotations”. (d) Relation between the non-Abelian LG and TB models with lattice symmetry. FBs are present in their intersection.

Multiple Line Graph- Mathematically, a graph consists of a set of vertices and edges, where the edge represents a pairing relation between the vertices [49] (Fig. 1(a)). For a simple graph $X$, we can obtain its LG $L(X)$ by replacing each edge with a vertex, which is placed on the midpoint as agreed. These vertices are adjacent when the corresponding edges in $X$ share a common vertex. $X$ is referred to as the root graph. Fig. 1(a) shows an example of the construction. The LG is related to a TB Hamiltonian of the form $H_{\rm LG}=tA_{L(X)}=t\sum_{i\neq j}c_{i}^{\dagger}c_{j}$. Here, $A_{L(X)}$ is the $\{0,1\}$ off-diagonal adjacency matrix of the LG and $c_{i}\text{ }(c^{\dagger}_{i})$ the annihilation (creation) operator on the $i$-th site and $t$ the hop** constant. Since the edges are identical, there is only a unique hop** parameter $t$ between two connected vertices in the model, and it has to be real to guarantee Hermicity. When the graph is periodic, the spectrum of the LG lattice hosts a macroscopic degenerate subspace, which corresponds to a FB with energy $E_{\text{FB}}=-2t$ in the language of band theory [18, 19]. It is commonly referred to as the LG theorem and the existence of FBs in the LG lattice is fundamentally governed by its specific connectivity. The uniform chemical potential $\mu$ is usually omitted in the LG since it only shifts all the bands trivially. But it is present as a loop (Fig. 1(a)) here for our generalization of the conventional LG to include the internal degree of freedom. Note that the LG Hamiltonian only has two parameters $t$ and $\mu$, which are visualized in the “dumbbell” in Fig. 1(a).

In its essence, we extend the discussion to more general TB models with internal degrees of freedom. In such cases, the coupling between two sites is generalized to a matrix. We propose the Hamiltonian

$$H_{\rm MLG}=T\otimes A_{L(X)}+M\otimes\mathbbm{1}$$ (1)

with $A_{L(X)}$ the $\{0,1\}$ off-diagonal adjacency matrix of the LG, $\mathbbm{1}$ the identity of the same dimensionality with $A_{L(X)}$, $T$ and $M$ the arbitrary Hermitian matrices of dimensionality $n$. The original LG is a specific case where $T=t$ and $M=\mu$. The FB eigenspace of the original LG is spanned by the compact localized states (CLSs) and the non-contractible loop states (NLSs) [16, 50] which are denoted as $\{v_{i}\}$ with $i$ up to the dimension of the FB eigenspace. To inherit the FB character from the LG, we construct the wave function $\psi_{j,i}=u_{j}\otimes v_{i}$ where $u_{j}$ is the $j$-th eigenvector, satisfying $\left(T-\dfrac{1}{2}M\right)u_{j}=\lambda_{j}u_{j}$, in the internal space. Remarkably, it follows that for arbitrary CLS or NLS $v_{i}$, $H\psi_{j,i}=(-2\lambda_{j})\psi_{j,i}$. As a result, the original FB with energy $-2$ has transformed into $n$ FBs, each with an energy of $-2\lambda_{j}$. Their existence is independent of the parameters of the matrices $T$ and $M$, indicating that they originate from the connectivity of the underlying periodic graph. Therefore, we can naturally extend the original LG structure (Fig. 1(a)) to a version with multiple internal degree of freedom (Fig. 1(b)). Specifically, we utilize the Kagome lattice, whose root graph is the honeycomb lattice to illustrate. A single site is now mapped to multiple vertices, which results in a layer structure. We first set $T$ and $M$ to be diagonal for clearance and draw the corresponding graph structure in Fig. 2(a). The blue and green layers represent any two-component degrees of freedom residing on the same site. Each layer features a LG structure, while the weights could be varied in different layers as indicated by $t_{i}$ and $\mu_{i}$. By including off-diagonal elements, we can more generally incorporate inter-layer couplings (magenta and brown arrows in Fig. 1(b)). As long as Hermiticity is maintained, these elements can assume complex values, which could include practical factors such as the SOC or magnetic fields in principle. The distinction from the intra-layer couplings is indicated by the bidirectional arrows. It is reasonable to designate the corresponding graph structure as the multiple LG as the presence of FBs is proven.

Figure 2: (a) The multiple LG on a Kagome lattice with a two-dimensional internal space, where the inter-layer coupling is neglected. The generators of the point group $C_{6v}$ are shown.  (b) Anisotropic hop**s from $d$-orbitals. (c) Isotropic hop**s from $s$-orbitals. (d) Sublattices A,B,C and the inequivalent nearest neighbor hop**s of Kagome.

Physical Interpretation and Non-Abelian Line Graph- We notice that high orbitals in the real material generally lead to an anisotropic hop** Hamiltonian (Fig. 2(b)), which is not compatible with the multiple LG as previously proposed. When the orbitals form a high dimensional irreducible representation (IRREP) of the space group of the real crystal, there is a mismatch between symmetries and the multiple LG, as we demonstrate below. Firstly, the lattice translation symmetry implies that the hop**s associated by the translation operator $\hat{T}$ are equivalent. Further, the point group imposes constraints on the inequivalent hop**s (Fig. 2(b)), denoted as $t^{\alpha\leftarrow\beta}$ with $\alpha,\beta$ the sublattice index. When a symmetry operation $g$ connects the hop** vectors of the inequivalent hop**s, the corresponding hop** matrices only differ by a similarity transformation as

$$t^{\alpha\leftarrow\beta}=D^{\dagger}(g)t^{\gamma\leftarrow\delta}D(g),\quad\mbox{\boldmath$r$}^{\alpha\leftarrow\beta}=g\mbox{\boldmath$r$}^{\gamma\leftarrow\delta}$$ (2)

where $D(g)$ is the high-dimensional IRREP of $g$ in the high orbital basis. In the multiple LG, the symmetry conditions are equivalent to $D^{\dagger}(g)TD(g)=T$ for $g$ is any group element. Therefore, $T$ must be proportional to the identity according to Schur’s lemma. The model reduces to several identical copies of the original LG, which corresponds to a reducible representation as the sum of the one-dimensional representations (Fig. 2(c)). This leads to a contradiction and implies that the multiple LG is not directly applicable for high orbitals. Remarkably, we can remedy the loophole by utilizing the internal degree of freedom. To illustrate that, we first introduce the concept of the non-Abelian LG. The energy spectrum of the multiple LG Hamiltonian (Eq. (1)) should be unaffected if we perform a local transformation $U_{i}(n)$ in the internal space. As a result, the matrix $T$ on the connecting link of the adjacent sites $i$ and $j$ changes as $T\rightarrow U_{i}^{\dagger}TU_{j}:=t^{i\leftarrow j}$ and the on-site matrix $M$ change as $M\rightarrow U^{\dagger}_{i}MU_{i}:=M_{i}$. They are combined and understood as a gauge transformation to guarantee the invariance of the overall Hamiltonian [51, 52] and result in a Hamiltonian with enhanced local $\text{U}(n)$ symmetry

$$H_{\rm NALG}=\sum_{s,s^{\prime}}\left(\sum_{i,j}c_{i,s}^{\dagger}t^{i\leftarrow j}_{ss^{\prime}}c_{j,s^{\prime}}+\sum_{i}c_{i,s}^{\dagger}M_{i}c_{i,s}\right).$$ (3)

The hop**s $t^{i\leftarrow j}$ are non-Abelian matrices since the transformations $U_{i}(n)$ generally make them not commute to each other. Also, they are not necessarily Hermitian since each $t^{i\leftarrow j}$ only represents a one-way hop**. The corresponding graph is referred to as a non-Abelian LG (Fig. 1(c)).

Table 1: The FB condition for the $d$-orbital doublet in terms of the SK integrals and the corresponding local transformation satisfying Eq. (4).

IRREP	FB conditions for SK integrals	Local transformation
$E_{1}(d_{xz}/d_{yz})$	$dd\pi+dd\delta=0$	$U_{A}=D^{2}(\text{C}_{6}),U_{B}=\mathbf{1},U_{C}=D^{-2}(\text{C}_{6})$
$E_{2}(d_{x^{2}-y^{2}}/d_{xy})$	$3dd\sigma+4dd\pi+dd\delta=0$	$U_{A}=D^{-1}(\text{C}_{6}),U_{B}=\mathbf{1},U_{C}=D(\text{C}_{6})$

This novel concept builds a significant connection between the multiple LG and the FB in the real material with internal degree of freedom. A non-Abelian LG can support FBs since there always exist appropriate local transformations that can convert them back to a multiple LG. Meanwhile, the non-commutative hop**s allow it to match the high-dimensional IRREP of the space group. In practice, we can start from a TB Hamiltonian which respects all the lattice symmetry. Since the TB Hamiltonian is periodic, once we find sublattice-dependent local transformations $U_{\alpha,\beta}$ that satisfy two conditions

$$U_{\alpha}t^{\alpha\leftarrow\beta}U_{\beta}^{\dagger}=T\quad\text{and}\quad U_{\alpha}M_{\alpha}U^{\dagger}_{\alpha}=M,$$ (4)

the TB model is determined as a non-Abelian LG, and the presence of FBs is guaranteed. We refer to them as the hop** condition and the onsite condition, respectively. They are general regardless of the specific LG lattice or the orbitals. To better elucidate the point, we still consider the Kagome lattice for convenience. The lattice has three sublattices and six types of nearest neighbor hop**s as shown in Fig. 2(d). The point group of the lattice is $C_{6v}$ with the generators six-fold rotations $\text{C}_{6}$ and mirror M (Fig. 2(a)). Considering the smallest triangle loop in the lattice, we find that the hop** condition in Eq. (4) leads to

$$\tilde{T}_{\alpha}=U^{\dagger}_{\alpha}T^{3}U_{\alpha},\quad\tilde{T}_{\alpha}=\tilde{T}_{\alpha}^{\dagger},$$ (5)

where $\tilde{T}_{\alpha=A,B,C}=\prod_{t\in\triangleright}(t^{\alpha\leftarrow\beta}\cdots t^{\gamma\leftarrow\alpha})$ is the product of the hop** matrices along the loop. All the $\tilde{T}_{\alpha}$ must be Hermitian and share the same eigenvalues, and this is a non-trivial requirement since $t^{\alpha\leftarrow\beta}$ are non-Abelian. Together with the on-site condition in Eq. (4), we offer a direct criterion for determining whether a given TB model is a non-Abelian LG. When the SOC is neglected, we can choose all the wavefunctions to be real and focus on a simplified spinless model. The hop** strength between atomic orbitals is usually captured by the Slater-Koster (SK) integrals [53], which are also real. It is natural to specify the orbitals as the $d$-orbital doublets since they carry the high-dimensional IRREPs $E_{n=1,2}$, and the corresponding representation matrices in Eq. (2) are given by $D(\text{C}_{6})=\cos(n\theta)\sigma_{0}-i\sin(n\theta)\sigma_{y}$ and $D(M)=\sigma_{z}$. To be specific, we consider the $d_{x^{2}-y^{2}}/d_{xy}$ orbitals. Therefore, the hop** matrix along $y$ direction (Fig. 3(a)) is given by

$$t^{A\leftarrow C}=\begin{bmatrix}(3dd\sigma+dd\delta)/4&0\\ 0&dd\pi\end{bmatrix}.$$ (6)

The other two hop**s $t^{C\leftarrow B},t^{B\leftarrow A}$ are related to it via $\text{C}_{6}$ and $\tilde{T}_{\alpha}$ can then be calculated. The algebra of the Pauli matrices offers us a powerful tool to deal with these two-dimensional representations. Detailed calculation [54] reveals that Eq. (5) is satisfied when $\text{Tr}(t^{\alpha\leftarrow\beta})=0$. Equivalently, the FBs exist when $3dd\sigma+4dd\pi+dd\delta=0$. Meanwhile, since $d_{x^{2}-y^{2}}/d_{xy}$ orbitals carry a two-dimensional IRREP, the on-site matrix $M$ is identity. Hence, the on-site condition in Eq. (4) is automatically fulfilled. Notably, $t^{A\leftarrow C}\sim\sigma_{z}$ while $\sigma_{x}$ appears in the other two hop**s with these specific SK integrals. They do not commute and the model is a non-Abelian LG. Interestingly, it is close to the parameters of transition metals [55], which suggests that our proposal may be a candidate to explain the origin of FBs in realistic Kagome materials, considering the orbital components. We set $dd\sigma=4,dd\pi=-3,dd\delta=0$ and $M=0$ for simplicity and plot the band dispersion in Fig. 3(b). We can manifestly identify two sets of $s$-orbital-like Kagome bands [16] in the band structure, which indicates that our non-Abelian LG inherits the conventional LG nature.

Figure 3: (a) Sublattice-dependent rotation that related the multiple LG to the $d$-orbital TB models. The grey line indicates the hop** $t^{A\leftarrow C}$. (b)(c) Band structure and the corresponding multiple LG of the Kagome model with $d_{x^{2}-y^{2}}/d_{xy}$, the non-zero SK integrals are $dd\sigma=4,dd\pi=-3$.  b) without SOC and the inter-orbital couplings; c) with SOC $\lambda=0.5$ and the inter-orbital couplings.

To further illustrate this relation, we explicitly transform the non-Abelian LG back to the multiple LG. It will also provide a more intuitive understanding of the band structures after we distinguish the non-Abelian LG in general TB models. Based on Eq. (5), we further diagonalize $T$ to $\mathcal{T}=\mathcal{U}^{\dagger}T\mathcal{U}$ and obtain $\tilde{T}_{\alpha}=O^{\dagger}_{\alpha}\mathcal{T}^{3}O_{\alpha}$, where $O_{\alpha}=\mathcal{U}U_{\alpha}$. Denoting the eigenvalues of $\tilde{T}_{\alpha}$ as $\text{Eig}(\tilde{T}_{\alpha})=\{\epsilon_{1},\cdots,\epsilon_{n}\}$, it follows that $\mathcal{T}=\text{diag}(\sqrt[3]{\epsilon_{1}},\cdots,\sqrt[3]{\epsilon_{n}})$. In our $d$-orbital case, we have $\text{Eig}(\tilde{T}_{\alpha})=\{(\pm dd\pi)^{3}\}$, leading to $\mathcal{T}=\text{diag}(dd\pi,-dd\pi)$. Therefore, the related multiple LG Hamiltonian is determined as diagonal $T=\mathcal{T}=t\sigma_{z}$ for convenience, where $t=dd\pi$ and $\sigma$ acts on the orbital-space. The corresponding local transformations are listed in Table 1 and illustrated in Fig. 3(a). The resulting multiple LG is shown in Fig. 3(b), where the two layers are interpreted as the rotated orbitals in Fig. 3(a). The blue, green and magenta arrows represent real-valued hop**s, and the yellow and red loops are chemical potentials (set as 0). The Hamiltonian leads to two copies of the original LG with $E_{\text{FB}}=\pm 2t$, and the band structure is symmetric due to the emergent chiral symmetry with the unitary operator $\sigma_{y}$. They are consistent with the bands shown in Fig. 3(b).

On this foundation, we duplicate the model to recover a spinful model and introduce the SOC. The leading term is given by the on-site $\mbox{\boldmath$L$}\cdot\mbox{\boldmath$S$}$ as $\bra{d_{x^{2}-y^{2}},s}\mbox{\boldmath$L$}\cdot\mbox{\boldmath$S$}\ket{d_{xy},s^{\prime}}=\bra{d_{x^{2}-y^{2}}}L_{z}S_{z}\ket{d_{xy}}\delta_{ss^{\prime}}$, which couples the orbitals on the same site. The band structure is plotted in Fig. 3(c). Correspondingly, the related multiple LG Hamiltonian is generalized to

$$T=s_{0}(t\sigma_{z}),\quad M=s_{z}(\lambda\sigma_{y}),$$ (7)

where $s$ acts on the spin-space and $\lambda$ is the SOC strength. We find that $M$ commutes with the local transformations $U_{\alpha}$ [54] so that the on-site condition in Eq. (4) is satisfied. Meanwhile, it provides the coupling between two s-orbital like FBs and open the degeneracies at $E=0$ (Fig. 3(b) and (c)). The coupling due to the SOC is visualized to the brown arrows appear in the multiple LG due to the SOC (Fig. 3(c)). As a result, the bands can no longer be explained as superpositions of the original Kagome bands. Nevertheless, the FBs are still ensured as previously discussed. Using the anti-commutative Clifford algebra, we can solve the FB energies analytically, which are slightly shifted to $\pm\sqrt{4t^{2}+\lambda^{2}}$. Notice that spin is a good quantum number of the model, our analysis holds for each spin component. Note that two spin sectors share the same multiple LG structure but with opposite on-site coupling strength, resulting in a two-fold degenerate band structure. Also, the bands are symmetric due to the chiral symmetry with operator $s_{x}\sigma_{y}$. Similar analyses are also valid for the $d_{xz}/d_{yz}$ doublet, and the main results are listed in Table 1.

Discussion and Conclusion- In conclusion, we have introduced a non-Abelian LG generalized from the original LG, which is adaptable for constructing multiple FBs in practical systems with internal degrees of freedom. It also goes beyond the requirement of real-valued isotropic hop**s. Our proof for the multiple LG can be extended to a plethora of pure lattice models with a FB [56]. As long as the crystal net structure hosts a macroscopic degenerate subspace, we could attach the internal degrees of freedom and the proof still holds. In special cases such as two-dimensional internal space, the energies of multiple FBs are exactly solvable and easily tunable in the model. Though we mainly discussed physical atomic orbitals throughout the work, the internal space of Eq. (1) can stand for any degree of freedom independent of the lattices. It defines an equivalent class $H(T,M)$, and the models that differ only by a gauge transformation result in different non-Abelian LGs but belong to the same class and share an identical energy spectrum. In the case of the Kagome lattice, those equivalent Hamiltonians can be associated with the $d$-orbital doublet via proper local transformations as we discussed, or diatomic Kagome lattice [57, 58, 59], or other more. Although they are equivalent in the gauge sense, the physical realizations are distinct. By leveraging internal degrees of freedom, our work may provide new insight and open an alternative avenue to construct and explore practical FB systems.