Abstract
Topological crystalline states (TCSs) are shortrange entangled states jointly protected by onsite and crystalline symmetries. Here we present a unified scheme for constructing all TCSs, bosonic and fermionic, free and interacting, from realspace building blocks and connectors. Building blocks are lowerdimensional topological states protected by onsite symmetries alone, and connectors are glues that complete the open edges shared by two or multiple building blocks. The resulted assemblies are selected against two physical criteria we call the noopenedge condition and the bubble equivalence. The scheme is then applied to obtaining the full classification of bosonic TCSs protected by several onsite symmetry groups and each of the 17 wallpaper groups in two dimensions and 230 space groups in three dimensions. We claim that our construction scheme can give the complete set of TCSs for bosons and fermions, and prove the boson case analytically using a spectralsequence expansion.
Introduction
Symmetryprotected topological states (SPT)^{1,2,3,4} are gapped states that do not have topological orders^{5,6} (fractional excitations), but cannot be deformed into product states of localized wave functions without either symmetry breaking or gap closing. The constituent particles of SPT can either be bosonic or fermionic. They are probably the most wellunderstood topological states, and the famous examples are AKLTlike states^{7} (bosonic), topological insulators, and topological superconductors^{8,9} (both fermionic). Especially, SPT protected by onsite symmetries (only acting on internal degrees of freedom) have been studied for years, and we now know that bosonic SPT is classified by group cohomology of the symmetry group^{1,2,3,4,10} (with the exception of the socalled “beyondgroupcohomology” states^{11,12,13,14}), and SPT of free fermions is classified by the K theory in the “tenfold way”^{15,16}. The classification of interacting fermions is much harder. Progresses in recent years^{17,18,19,20,21,22,23,24,25,26,27} have provided mathematical frameworks to describe the classification, but the detailed computation is still challenging for general symmetry groups. In contrast to SPT protected by onsite symmetries are crystalline symmetryprotected topological states, or simply topological crystalline states (TCS).
As suggested by name, TCS has its nontrivial topology protected by both onsite and crystalline symmetries. Crystalline symmetries are the symmetry groups of periodic lattices in various dimensions (restricted, for simplicity, to two and three in this paper), and the study of crystalline symmetries as groups has been complete since the end of the last century^{28}. All crystalline symmetries are classified into 17 wallpaper groups into two dimensions (2D) and 230 space groups into three dimensions (3D). Among TCS, those constituted of noninteracting fermions with charge conservation have so far attracted most theoretical and experimental effort. These states are also known as the topological crystalline insulators^{29,30}, the classification and diagnosis of which have only recently been completed^{31,32,33,34,35,36}. Interacting TCS, especially the fermionic one, is far less understood. On one hand, the framework of group cohomology for onsitesymmetry bosonic SPT cannot be directly applied; on the other hand, there is not an obvious way of adapting the K theory, which is key to solving the classification problem of free fermions^{16,37,38}, to the task of classifying interaction fermions. A recent work by Thorngren and Else^{39} provides a mathematical connection between TCS and onsitesymmetry SPT states. Another way to understand TCS is the process of dimensional reduction^{40,41,42}, which constructs TCS by decorating highsymmetry points, lines, and planes with lowerdimensional onsitesymmetry SPT. These dimensionalreduction constructions are easier to compute in practice (because both the dimensionality and symmetry groups are reduced), and offer extra physical insight into the nature of these TCS states. For interacting bosonic and fermionic SPT states, a large class of TCS states has been constructed this way, but the previous works have not been systematically extended to arbitrary symmetry groups, and it has not been shown whether the realspace construction is complete.
We in this paper show that all TCS, bosonic and fermionic, free and interacting, can be built up in real space from two types of elementary ingredients, the building blocks (or simply blocks) and the connectors. Building blocks are finitesize pieces of lowerdimensional SPT that are protected by the respective little symmetry group alone. The little symmetry group includes all symmetry operations, onsite or spatial, that leave each point in the lowerdimensional subspace invariant: they can be considered as enlarged onsite groups by the spatial symmetries that do not change the spatial coordinates on specific subspaces of the lattice. A building block defined on a pdimensional subspace is called a p block. For threedimensional TCS, one considers p = 3, 2, 1, 0 blocks (3block TCSs are simply 3D SPT states protected by the onsite symmetry alone, which are compatible with the crystalline symmetries. This is discussed in more detail in Sec. V of Supplementary Information). To construct a gapped TCS, we arrange p blocks in such a way that the space group of the TCS is preserved, including translation symmetries, pointgroup symmetries, and nonsymmorphic symmetries. The p blocks in general have open boundaries, and, being SPT themselves, gapless boundary states (or more precisely speaking, boundary anomalies) on their (p − 1)dimensional boundaries. Therefore, a symmetric construction with p blocks alone cannot be gapped in the bulk, and in order to build a gapped state, one needs “glue” to close the open edges in the assembly. The glue is the connector, which, technically speaking, is a torsor defined on (p − 1) dimensions. (p − 1) connectors are inserted where multiple p blocks share one (p − 1)dimensional open edge, and should hybridize the gapless states contributed from the joining p blocks so that the edge becomes gapped. When all open edges are completed, that is, when the “noopenedge condition” is met, we obtain assemblies that are (i) symmetric under onsite and spatial symmetries and (ii) gapped. However, this does not mean that the crystal is topologically nontrivial, as we additionally require that it cannot be deformed into a product state. Obviously, this implies that there is at least one building block with p > 0 that is a nontrivial SPT, but this alone is insufficient: there are constructions from nontrivial (p > 0)building blocks that can still be trivialized. We show that the trivialization can be understood as a “bubbling process,” in which constructions that can be canceled by the “bubbles” are excluded, considered trivial. Two TCS are hence topologically equivalent if the decorations can be related by a bubbling process, and this is called the “bubble equivalence”. The space of all TCS is hence the space of symmetric assemblies of building blocks satisfying the noopenedge condition quotient of the bubble equivalence.
One should be aware that both the noopenedge condition and the bubbling equivalence, simple enough in appearance, have their subtleties. While it is obvious that one may use a (p − 1) connector to complete the open edges at the meeting of two or multiple p blocks, after all necessary (p − 1) connectors are added, at the (p − 2) joints where these (p − 1) connectors meet, there may be (p − 2)dimensional open edges. Similarly, while it is natural that bubbles in p + 1 dimensions can be used to annihilate p blocks, there are cases where (p + 2) bubbles, leaving all (p + 1) blocks intact, annihilate p blocks. A third and related subtlety, called the groupextension problem, is about the relations between TCS constructed from p blocks and those constructed from (\({p}^{\prime}< \ \! p<p\) ) blocks. All three subtleties have to do with constructions that have nontrivial connectors or torsors. Torsors are not SPT (but may be understood as fractions of SPT), and their topological properties should be separately considered.
The realspace construction scheme given above allows an automated generation of all inequivalent TCS for arbitrary spatial and onsite symmetry groups in any dimension D, defined in the following steps: (a) make a symmetric arrangement of p blocks, (b) add (p − 1) connectors to complete open edges, (c) use (p + 1) bubbles to “modulo out” trivial constructions, (d) add (p − 2) connectors to complete open edges of (p − 1) connectors, (e) use p + 2 bubbles to “modulo out” trivial constructions, and (f) repeat until the connectors are zerodimensional and the bubbles are ddimensional, where d is the spatial dimension. This process naturally fits the construction of TCS into a spectral sequence, a successive approximation originally designed for computing homology (cohomology) groups of a topological space^{43,44}. We adapt the terminology in mathematics and refer to different orders in this successive approximation as different pages in the spectral sequence. Each page, being worked out from the previous page, can be roughly understood as a certain level of approximation to the exact classification, more accurate than its previous page, and less accurate than the next. Following this observation, we can analytically prove that the realspace construction process as defined above gives exactly the same classification of general bosonic TCS, as derived from the gaugingspatialsymmetry argument^{34}. This proof is presented in Sec. IV of the Supplementary Information. We develop an automated code and generate all bosonic TCS having typical onsite symmetries (such as unitary and antiunitary Z_{n} symmetries), and directly multiply any of the 17 wallpaper groups in 2D and 230 space groups in 3D (see Sec. VII of Supplementary Information). On the fermionic side, the completeness of the construction scheme has been demonstrated for free fermions with charge conservation, timereversal symmetry, and any of the 230 space group symmetries in ref. ^{36}. We also point out that while the general realspace construction still holds, the difficulty in classifying interacting fermionic TCS lies in the nontrivial superposition of states due to fermion statistics, and the lack of a unified bulk and boundary description for fermionic SPT.
Compared with the group cohomology formula in ref. ^{39}, our method for classifying TCS is not only easier to compute, but also more concrete so that for each TCS, we have a realspace, piecewise construction. In particular, it allows us to distinguish TCS made of building blocks in different dimensions, which is an additional structure in the TCS classification. In addition, the paginated structure of the spectral sequence also has physical interpretations: the different levels of approximation can be used to construct variants of the Hasting–Oshikawa–Lieb–Schultz–Mattis (HOLSM) Theorem^{42,45,46,47,48,49,50,51,52,53,54,55}.
Results
Cell decomposition and chain complex
Following ref. ^{36}, for each space group, there is a welldefined cell decomposition process that a ddimensional Euclidean space (R^{d}) is decomposed into the union of p = 0, 1, 2, . . . , ddimensional “cells” with zero overlap. Here a p cell is topologically equivalent to R^{p}, or a pdimensional disk minus its boundary, and we denote the collection of all p cells in a decomposition as Y_{p}. We use Greek letters σ, τ, μ, and γ to denote p = d, d − 1, d − 2, d − 3cells, respectively, and also use σ to denote a general cell with unspecified dimension. We require that any spacegroup operation maps one p cell to another p cell. We also require that the union of all Y_{p} is R^{d} itself. Mathematically, the collection of all cells under these conditions forms a topological space, Y, called a G complex. Furthermore, for any σ ∈ Y_{p}, we consider the subgroup G_{σ} that maps σ to itself. We require that any g ∈ G_{σ} has a pointwise action on σ, which in general includes onsite and crystalline symmetries. In other words, G_{σ} acts as an onsite symmetry group locally on σ. In addition, unlike in ref. ^{36}, the cells are oriented. The orientation is arbitrarily subject to the condition that the orientations of σ and g ⋅ σ are also related by g.
We illustrate the cell decomposition with simple examples. In Fig. 1a, we show the cell decomposition within one unit cell of 2D space group pm, having orthogonal basis vectors with one mirror line mapping x to −x. In Fig. 1b, there is another example of decomposition for 2D space group p2, having a twofold rotation center C_{2}. For this decomposition, we comment that the 1D segment passing the rotation center, instead of being a 1cell by itself, is decomposed into two 1cell (τ_{3} and C_{2}τ_{3}) and one 0cell (μ_{4}). This is in contrast to the example in Fig. 1a, where the segment coincident with the mirror line needs no additional decomposition. The difference is because we require that symmetry group of a p cell, G_{σ}, should act pointwise on σ, and while each point in τ_{3} is invariant under a mirror in Fig. 1a, only μ_{4} is invariant under C_{2}, and τ_{3} and C_{2}τ_{3} are mapped to each other.
After decomposition, we are ready to define the operator ∂ acting on a p cell σ_{p}: ∂σ_{p} gives the (p − 1) cells that are at the boundary at σ, i.e.,
The coefficient 〈σ_{p−1}∣∂σ_{p}〉 = 0, if σ_{p−1} is not part of the boundary of σ_{p}; 〈σ_{p−1}∣∂σ_{p}〉 = ±1, if σ_{p−1} is part of the boundary, and the sign depends on the relative orientations of σ_{p} and σ_{p−1}. For example, in Fig. 1a, the orientation of σ gives, by righthand rule, natural orientations for adjacent 1cell, and if the designated orientation of τ_{i} is parallel (opposite) to it, 〈τ_{i}∣∂σ〉 = +1 (−1): ∂σ = τ_{1} + τ_{3} − t_{y}τ_{1} − τ_{2}. For 1cell τ_{i}, the arrow naturally gives the coefficients 〈μ_{i}∣∂τ_{j}〉: the zero cell at the head and at the tail of the arrow has +1 and −1 coefficients, respectively. For example, ∂τ_{1} = μ_{2} − μ_{1}. The ∂ operation hence defines a chain complex from d cells all the way down to 0cell:
Here, we notice that the homology groups of this chain complex are all trivial, because Y is a decomposition of R^{d}, which is topologically trivial. As a result, the chain complex in Eq. (2) must be an exact sequence, i.e., ∂^{2} = 0, which is important to the proof in Sec. IV of Supplementary Information.
Building blocks and connectors
We start by reviewing topological states we can decorate on each cell. This forms the basis of our realspace construction: each cell σ of dimension p can be decorated with a topological state, which is protected by the local onsite symmetry group G_{σ}. On an isolated cell, such topological states are classified by the pdimensional SPT states protected by G_{σ}, which we denote by Φ^{p}(G_{σ}). We will refer to such decorations as “pdimensional building blocks”, or p blocks for short. However, in our realspace construction, we need to consider a more general case, where σ is part of the boundary of a (p + 1)dimensional cell τ, which is in turn decorated by another topological state. Hence, we need to extend the scope of topological states to also include boundary topological states. In general, we consider a set of representative fixedpoint topological states, which we denote by Ψ^{p}(G).
There are two important operations on Ψ^{p}(G), which are the mathematical foundation for the computation of our realspace constructions. First, we can stack two states ψ_{1,2} in Ψ^{p}(G), and get a new state, which we denote by ψ_{1} ⊞ ψ_{2}. Second, we can compute the coboundary of ψ ∈ Ψ^{p}(G), which should be a state in Ψ^{p+1}(G). The physical meaning of dψ is that ψ can be realized on a pdimensional boundary of dψ.
The structure of such sets Ψ^{p}(G) and the rules of these two operations depend on the nature of the system: being a freefermion, interactingboson, or interactingfermion system. In particular, for bosonic systems, Ψ^{p}(G) is simply the cochain space of G. In the main text, we assume that Ψ^{p}(G), ⊞ and d are known, and use them to compute realspace constructions, while leaving the details to Sec. I of Supplementary Information. In particular, the SPT phases Φ^{p}(G) can be computed from Ψ^{p}(G), and the coboundary maps d: Φ^{p}(G) are given by the subgroup of cochains in Ψ^{p}(G) satisfying dα = 0, quotient of the coboundaries given by the image of the coboundary map d[Ψ^{p−1}(G)].
“Firstpage” (firstorder approximation) candidates for TCS
In general, a ddimensional TCS can be constructed by assembling p blocks (p < d) and gluing them with lowerdimensional connectors, while leaving all higherdimensional cells empty. The state after this construction is termed a pblock topological crystal^{36,41,42}. In this way, all TCSs are organized into pblock topological crystals, where p = 0, …, d: in particular, one can argue that a TCS with trivial (but not necessarily empty) SPT states decorated on d > p cells can be continuously deformed to a topological crystal with all (d > p) cells empty^{36}. The classification and construction of TCS then amounts to enumerating all inequivalent topological crystals for a given symmetry group G. Here, when enumerating all pblock topological crystals, we do not distinguish different choices of lowerdimensional connectors. This is because the difference (the difference between two states A and B is defined as the stacking of A and the inverted state of B, denoted by − B) between two such choices on qcell connectors can be viewed as a qblock topological crystal and described by the qblock classification.
We now consider pblock topological crystals. They are labeled by different choices of p blocks, which are pdimensional SPT states at p cells σ ∈ Y_{p}. Their classification is determined by the local onsite group G_{σ}, and given by Φ^{p}(G_{σ}). For bosonic systems, there is Φ^{p}(G_{σ}) = H^{p+1}[G_{σ}, U(1)]. However, for interacting fermionic systems, the calculation of Φ^{p}(G_{σ}) for arbitrary G_{σ} can be much more complicated^{17,18,19,20,21,22,23,24,25,26,27}.
Naturally, if some p cell σ is decorated with some p block, [α], then symmetry requires that the p cell gσ must be decorated by the p block g ⋅ [α]_{σ} ∈ Φ_{gσ} for g ∈ G. This is possible because of the isomorphism G_{gσ} = gG_{σ}g^{−1} ≃ G_{σ}. In Sec. II of Supplementary Information, we show the explicit definition of g ⋅ [α], and for now, we intuitively understand it as copying the p block at p cell σ to another p cell gσ. Therefore, only the cells in the G orbits of Y_{p}, Y_{p}/G, may have independent p blocks, and the p blocks at all the other cells are determined by symmetry. Physically, σ ∈ Y_{p}/G are the p cells that are not related by any g ∈ G and g ≠ I. Once the p blocks for all cells in Y_{p}/G are specified, we obtain a symmetric assembly denoted by [ψ_{p}], such that
and
Throughout the paper, we use ψ_{p} to denote the state, or wave function, on all p cells, use ∣_{σ} to denote the same wave function restricted to a certain cell σ. We use [ψ] for the phase, the representative state of which is ψ. The collection of all possible symmetric assemblies from p blocks is denoted by \({E}_{p,1}^{p}\)
For reasons to be revealed in later sections, these assemblies are called the firstpage candidates for TCS, which can also be taken as firstorder approximations to TCS. The exact meaning of \({E}_{p,r}^{q}\) will also be introduced and put to use in due course.
Noopenedge condition
Being SPT, [ψ_{σ}] necessarily has gapless modes at τ ∈ ∂σ. In fact, in the G complex, each (p − 1) cell is in the boundary of at least two p cells. (For example, τ_{3} ∈ ∂σ ∩ ∂mσ in Fig. 1a.) Therefore, for a gapped TCS, we require that the gapless modes at σ contributed from all adjacent τ can gap out each other, so that there is no open edge. More precisely speaking, the gapless boundary modes reflect the quantum anomaly of an SPT boundary. The necessary condition for a gapped (p − 1) cell τ is that the anomalies contributed by all adjacent [ψ_{p}∣_{σ}] cancel each other.
Suppose a (p − 1) cell τ is the common edge of n pieces of p cells σ_{1,…,n}, each decorated by a p cell \([{\psi }_{p}{ }_{{\sigma }_{i}}]\in {\Phi }^{p}({G}_{{\sigma }_{i}})\). Then we define the following quantity for τ ∈ Y_{p−1}, given any ψ_{p}:
Physically, [d_{1}ψ_{p}∣_{τ}] is an SPT obtained from stacking [ψ_{p}∣_{σ}] for 〈τ∣∂σ〉 ≠ 0, and is a pSPT with symmetry group G_{τ} ⊇ G_{σ}. The collection of this pSPT attached to (p − 1) cells is denoted by \({E}_{p1,1}^{p}\)
If [d_{1}ψ_{p}∣_{τ}] is a nontrivial SPT with respect to G_{τ}, then on τ, which is the common edge of the stacked layers, there must be gapless modes, meaning that this particular ψ_{p} fails the noopenedge condition. Hence, the nonzero elements in \({E}_{p1,1}^{p}\) are also called anomaly patterns. For a given \([{\psi }_{p}]\in {E}_{p,1}^{p}\), the collection of all [d_{1}ψ_{p}∣_{τ}] is given by \({d}_{1}[{\psi }_{p}]={\oplus }_{\tau \in {Y}_{p1}}[{d}_{1}{\psi }_{p}{ }_{\tau }]\). This defines d_{1} as a linear mapping between \({E}_{p,1}^{p}\) to \({E}_{p1,1}^{p}\). Therefore, the noopenedge condition states that only the kernel of d_{1} may be candidates for TCS, i.e., d_{1}[ψ_{p}] = 0.
On the other hand, if [d_{1}ψ_{p}∣_{τ}] is a trivial pSPT with symmetry group G_{τ}, on τ we can place a “mass term” that gaps out the edge modes contributed by σ_{1,…,n}. The resulted gapped state on the (p − 1) cell τ is called a connector, as in real space it acts as the nexus of the bordering p cells σ_{1,…,n}. Connectors are not SPT in general, but are torsors (see “Methods”). For now, we only need to know that any \([{\psi }_{p}]\in {E}_{p,1}^{p}\) that satisfies the noopenedge condition can be glued by (p − 1) connectors such that any τ ∈ Y_{p−1} is also gapped.
Bubble equivalence
The kernel of \({d}_{1}:{E}_{p,1}^{p}\to {E}_{p1,1}^{p}\) are gapped, symmetric assemblies of p blocks (gapped at any τ ∈ Y_{p−1} by connectors). But two different assemblies in kerd_{1} may be topologically equivalent to each other, and more importantly, some nontrivial (nonvacuum) state in kerd_{1} may even be equivalent to vacuum upon adiabatic deformation. Following ref. ^{36}, every adiabatic deformation is equivalent to the creation (annihilation) of bubbles within some p cell (0 < p ≤ d). A p bubble is a pdimensional disk inside some pcell ν ∈ Y_{p}, the inside of which is vacuum, and the boundary of some (p − 1)SPT protected by G_{ν}. Therefore, it can be considered as some (p − 1)SPT attached to a p cell. It is straightforward to see that creation of bubbles cannot change the topology of the state. Any bubble can shrink to a point and vanish, and, as the inside of any cell in our cell decomposition does not have any spatial symmetry, the shrinking and vanishing process does not break any symmetry.
Therefore, the topology of [ψ_{p}] is unchanged after we decorate, in a Gsymmetric way, the (p + 1) cells with some (p + 1) bubbles. A Gsymmetric assembly of (p + 1) bubbles is denoted \([{\tilde{\psi}}_{p+1}]\) and \([{\tilde{\psi }}_{p+1}]{ }_{\nu }\) is a part of the assembly restricted to some ν ∈ Y_{p+1}. The collection of all (p + 1) bubbles is denoted by \({E}_{p+1,1}^{p}\), defined in a way similar to Eq. (5):
We refer to \([{\tilde{\psi}}_{p+1}]\in {E}_{p,1}^{p+1}\) as a bubbling pattern.
The equivalence relations induced by bubbles in \({E}_{p,1}^{p+1}\) are expressed by another linear map \({\tilde{d}}_{1}\), which has a form similar to Eq. (6),
As shown in the section “Bubbling equivalence on the first page”, \({\tilde{d}}_{1}{\tilde{\psi }}_{p+1}\) gives the state generated by deforming the bubbling pattern \({\tilde{\psi }}_{p+1}\), which should be regarded as a trivial state. These trivial states in \({\rm{img}}{\tilde{d}}_{1}\) give the equivalence relations in \({E}_{p,1}^{p}\): \([{\psi }_{p}] \sim [{\psi }_{p}^{\prime}]\), if and only if \([{\psi }_{p}][{\psi }_{p}^{\prime}]\in {\rm{img}}\ {\tilde{d}}_{1}\).
"Secondpage” candidates for TCS
The noopenedge condition requires that TCS be in the kernel of d_{1} in Eq. (6), and the bubble equivalence states that two TCS related by an image of \({\tilde{d}}_{1}\) are topologically the same. The firstorder approximation to TCS, \({E}_{p,1}^{p}\), is refined by these two conditions into
where we introduce a general version of d_{1} and \({\tilde{d}}_{1}\): \({d}_{q,r}^{p}:{E}_{q,r}^{p}\to {E}_{qr,r}^{p+1r}\). d_{1} and \({\tilde{d}}_{1}\) are the special cases \({d}_{1}={d}_{p1,1}^{p}\), \(\tilde{{d}_{1}}={d}_{p+1,1}^{p}\).
Like those in \({E}_{p,1}^{p}\), states in \({E}_{p,2}^{p}\) are also generated by decorating p cells and are Gsymmetric, but noopenedge condition ensures that all cells in Y_{p−1} are gapped, and the bubble equivalence relation ensures that they cannot be trivialized by (p + 1) bubbles. \({E}_{p,2}^{p}\) is hence the secondpage approximation to the set of TCS made from p blocks (pTCS for short). Here, page is the mathematical terminology referring to the order of approximation in the spectral sequence, and we denote the page number by r in the notation \({E}_{p,r}^{q}\). In further sections, we are to treat higherpage approximations \({E}_{p,3}^{p},{E}_{p,4}^{p}...\), and in the end \({E}_{p,\infty }^{p}\) is the exact collection of pTCS. In Sec. V of Supplementary Information, however, we show that for bosonic systems, if G is a direct product of the onsite and the space symmetry groups G = SG ⊗ G_{0}, then the second page already contains the final answer: \({E}_{p,2}^{p}={E}_{p,\infty }^{p}\), which can be simply summed to produce the full classification
We remark that in this work, the SPT with 0block is considered trivial, hence the lower bound in the summation in Eq. (11).
Further considerations
The discussion presented so far provides an algorithm to enumerate possible topological crystals from lowerdimensional building blocks. In this algorithm, we start by decorating pdimensional cells with SPT states protected by the local symmetry groups. We then compute whether the resulting anomalies cancel on the (p − 1) cells, and whether the state can be trivialized by a trivialization pattern on the (p + 1) cells. However, a careful mathematical analysis in the section “Connectors and secondpage results” reveals that, for a general symmetry group G, such an intuitive algorithm is not complete as the answers in \({E}_{p,2}^{p}\) contain false entries that do not represent valid and nontrivial SPT states.
On one hand, \({E}_{p,2}^{p}\) may contain invalid entries: although we have checked that the building blocks can be glued together in a gapped way along the (p − 1)dimensional edges, it is possible that doing so will always leave gapless modes on the (p − 2)dimensional cells. In fact, two examples will be given in the section “Secondpage noopenedge conditions”. Therefore, to make sure that an SPT building block represents a gapped SPT state, one needs to check that if the decoration can be extended to all lowerdimensional cells without anomaly.
On the other hand, \({E}_{p,2}^{p}\) may contain trivial entries that can be revealed by considering higherdimensional bubbles. It is possible that bubbles on (p + 2)dimensional cells do not create any nontrivial SPT decorations on the (p + 1) cells, but create such decorations on the p cells. Such a pattern of bubbles then indicates that certain decoration on p cells, \([{\psi }_{p}]\in {E}_{p,2}^{p}\) it creates is trivial. In fact, such an example will be given in the section “Application to HOLSM Theorems”. Consequently, to eliminate all trivial entries, one has to consider bubbles on all higherdimensional cells, bounded by d, the spatial dimension.
There is one more subtlety we need to consider. We have generally divided TCS according to the dimensionality of the building blocks. For example, the 3D TCS can be divided into SPTs with plane decorations and line decorations, represented by \({E}_{2,\infty }^{2}\) and \({E}_{1,\infty }^{1}\), respectively. (Point decorations correspond to atomic insulators and do not have boundary states in general, and so are excluded from the set of TCS. Nevertheless, they can also be easily enumerated using our classification scheme.) The complete classification of TCS is then a combination of all topological crystals with all possible 0 < p < d. However, when recombining the classification of topological crystals with different buildingblock dimensions, the result may be a nontrivial group extension of the two respective groups, instead of a direct sum. For example, assume that for some G, \({E}_{2,\infty }^{2}\) and \({E}_{1,\infty }^{1}\) are both \({{\mathbb{Z}}}_{2}\). Naively, the combined classification would be \({{\mathbb{Z}}}_{2}\times {{\mathbb{Z}}}_{2}\). However, the correct result could also be \({{\mathbb{Z}}}_{4}\), which is a nontrivial extension of \({{\mathbb{Z}}}_{2}\) over \({{\mathbb{Z}}}_{2}\). Intuitively, imagine stacking two copies of the nontrivial elements from \({E}_{2,\infty }^{2}\). Since the classification is \({{\mathbb{Z}}}_{2}\), the resulting SPT state is trivial if viewed as a topological crystal with p = 2, i.e., the decoration on the 2cells is trivial. However, the resulting state can have nontrivial decorations on the 1cell, and thus is the nontrivial 1TCS, or, the generator of \({E}_{1,\infty }^{1}\). If this happens, the combined classification is then \({{\mathbb{Z}}}_{4}\), generated by the generator of \({E}_{2,\infty }^{2}\). For example, we consider a free fermion system with the timereversal symmetry and the inversion symmetry. The classifications protected by the timereversal symmetry (the local symmetry) in 3D and 2D are both \({{\mathbb{Z}}}_{2}\). However, the combined classification is \({{\mathbb{Z}}}_{4}\). The stack of two identical 3D centrosymmetric \({{\mathbb{Z}}}_{2}\) topological states is a topological state jointly protected by the timereversal and inversion symmetries, and is equivalent to a construction of 2D \({{\mathbb{Z}}}_{2}\) topological states in certain 2cells^{35,36,56}. In general, stacking two or multiple pTCS may produce trivial decorations on p cells, but nontrivial decorations on \({p}^{\prime}\)cells where \({p}^{\prime}<p\), which is a nontrivial \({p}^{\prime}\)TCS. Finding these relations in general is what we call a groupextension problem. All the three subtleties outlined above involve the connectors on lowerdimensional cells, which are discussed in the section “Connectors and secondpage results”.
Application to HOLSM theorems
Our TCS constructions can be applied to obtain generalized HOLSMtype theorems^{45,46,47,48,49,50,51,52}, and the SPTenforcing HOLSM Theorems^{53,54,55,57}. Using the higherpage derivatives introduced in the section “Connectors and secondpage results”, these HOLSMlike constraints can be summarized as the following in our language: a spatial distribution of projective representations, such as the Kramers doublets in the original HOLSM, is represented by an anomaly pattern \([\tilde{\psi }]\) in \({E}_{0,1}^{1}\). If \([\tilde{\psi }]\) is trivialized through d_{r} by a \([\psi ]\in {E}_{r,r}^{r}\) as \([\tilde{\psi }]={d}_{r}[\psi ]\), then \([\tilde{\psi }]\) can be gapped out by the rblock TCS assembly [ψ]. Otherwise, if \([\tilde{\psi }]\) remains a nontrivial element in \({E}_{0,\infty }^{1}\), then it cannot be gapped out by an SPT state, and will give the consequences of the original LSM Theorem. Therefore, the spectral sequence provides a way to compute these constraints. Detailed explanation of the above statements and examples is presented in the section “Application to HOLSM Theorems”.
Discussion
In this work, we systematically study realspace construction of TCSs. Starting from building blocks made of SPT states protected by the little symmetry group, we construct a TCS by examining the noopenedge conditions, the bubbling equivalences, and solving the groupextension problems. These steps form a framework to compute TCS classification. In particular, for bosonic TCS, we prove that, for any symmetry group, this framework gives exactly the same results as the groupcohomology formula in ref. ^{39}. For the simple cases of G = SG × G_{0}, the computation is greatly simplified: the classification of topological crystals with buildingblock dimension d_{b} = p is given by \({E}_{p,2}^{p}={E}_{p,{\infty} }^{p}\) in Eq. (10).
One advantage of the topologicalcrystal approach is that it allows us to consider topological crystals with different buildingblock dimensions separately. In particular, it allows us to consider a more physical classification of crystalline SPTs, which ignores 0D building blocks. The reason for considering this is that when considering the classification of topological states, we usually identify states that can be smoothly deformed to each other without breaking the symmetries. Included in these smooth deformations are insertion and removal of local nondegenerate degrees of freedom, which in general can carry arbitrary 1D linear representations of the local symmetry group. These degrees of freedom are precisely the content of 0D building blocks. In addition, after removing 0SPTs, all TCSs in our classification have gapless edges. Therefore, the classification ignoring these 0D building blocks is the more physical one to consider, comparing with the fullgroupcohomology classification H^{d+1}[G, U(1)_{PT}]^{39}.
To demonstrate our framework, we develop an automated code to compute bosonic TCS protected by the direct product of typical onsite symmetries and any of the 2D and 3D crystalline symmetry groups. The results are listed in Supplementary Tables I and II.
Our framework also applies to interacting fermions. Indeed, we give several examples to demonstrate this application in “Methods”. The general computation, however, is much more elusive than its bosonic counterpart, due to the complex structure of the space of fixedpoint wave functions of onsite symmetries, denoted by Ψ^{d}(G_{σ}). This is further complicated by the fact that the stacking operation of fermionic wave functions do not necessarily commute, as pointed out in the section “Connectors and secondpage results”. However, the recent progress in the classification of fermionic SPTs protected by onsite symmetries^{17,18,19,20,23,25,27} should allow such computation to be carried out. We will leave this to future works.
In the section “Application to HOLSM Theorems”, we point out that our framework can also be used to study generalized HOLSM Theorems, especially those enforcing nontrivial SPT states. It will be interesting to apply it to look for new SPTenforcing HOLSM Theorems in more general symmetry groups that mix crystalline and onsite symmetries. We will also leave this to future works.
Note added to proof: As this work was being finalized for posting on the arXiv, refs. ^{58,59,60} appeared, which contains some related results.
Methods
Noopenedge conditions on the first page
Here, we review details of noopenedge conditions on the first page.
As we asserted in the section “Noopenedge condition”, the anomaly on a (p − 1) cell τ is computed by directly summing the building blocks on bordering p cells, as in Eq. (6). Several remarks are in order to explain this equation: fFirst, as each p cell has its own local onsite symmetry group \({G}_{{\sigma }_{i}}\), and as the “trivialness” or “nontrivialness” of SPT is only welldefined with respect to some symmetry group, we have to make clear what is the symmetry for the direct sum of the p blocks. The answer is G_{τ}, the local onsite symmetry group of τ, which is the shared edge of σ_{1,…,n}. Physically, τ is a highsymmetry line, and hence has higher symmetry than the bordering highsymmetry planes: \({G}_{{\sigma }_{i}}\subset {G}_{\tau }\). Suppose for a pair of τ and σ satisfying 〈τ∣∂σ〉 ≠ 0, there is g ∈ G_{τ} and g ∉ G_{σ}, then gσ must be another pcell bordering τ, 〈τ∣∂gσ〉 = 〈τ∣∂σ〉. (For better understanding, use Fig. 1a, where \(m\in {G}_{{\tau }_{3}}\) and m ∉ G_{σ}, and 〈τ_{3}∣∂_{σ}〉 = 〈τ_{3}∣∂mσ〉 = 1.) The direct sum [ψ_{p}∣_{σ}] ⊕ [ψ_{p}∣_{gσ}] is hence symmetric under not only under G_{σ} and G_{gσ}, but also under g. Second, we remark that in Eq. (6), if 〈τ∣∂σ〉 = −1, it means that we need to invert the p block at σ, before stacking it. Last, we also note while \([{d}_{1}{\psi }_{\tau }^{p}]\) has the symmetry group of τ, a (p − 1) cell, \([{d}_{1}{\psi }_{\tau }^{p}]\) is still an SPT (trivial or nontrivial) in p dimensions. Or one can say that \([{d}_{1}{\psi }_{\tau }^{p}]\) is a pSPT associated with a (p − 1) cell, protected by the symmetries of the (p − 1) cell.
We use two examples in bosonic systems to show, respectively, that certain p assemblies in \({E}_{p,1}^{p}\) satisfy and do not satisfy the noopenedge condition. The space group is that of a 3D orthogonal lattice having one mirror plane mapping x to −x plotted in Fig. 2a, and the onsite symmetry group is a unitary Z_{4} symmetry. We differentiate the cases where (i) the Z_{4} generator g_{0} commutes with mirror symmetry m: g_{0}m = mg_{0}, and (ii) they do not commute \({g}_{0}m=m{g}_{0}^{1}\). Consider an assembly ψ_{2} generated from decorating σ with a 2block, which is the generator of \({\Phi }^{2}({G}_{{\sigma }_{1}})={Z}_{4}\), [α]. The decorations on the other orbits of Y_{2}/G are set to be vacuum, and within a unit cell, the only decorated 2cells are σ and mσ. For case (i), the mirror operation does not act on the local degrees of freedom, m ⋅ [α] = [α]; but for case (ii), we have m ⋅ [α] = −[α]. We check the assembly against the noopenedge condition at τ between σ and mσ. Following Eq. (6), in case (i) there is
where we have used 〈τ∣∂σ〉 = 〈τ∣∂gσ〉. Since 2[α] ≠ 0, [ψ_{2}] in case (i) fails the noopenedge condition. In case (ii), we have
We conclude that [ψ_{2}] in case (ii) satisfies the noopenedge condition.
Bubbling equivalence on the first page
We now derive the bubbling equivalence in Eq. (9). To show the equivalence relations between assemblies \({E}_{p,1}^{p}\) induced by bubbles in \({E}_{p+1,1}^{p}\), we need to relate \([{\tilde{\psi }}_{p+1}]\) to an element in \({E}_{p,1}^{p}\). We start with enlarging the bubbles, so that a bubble inside ν ∈ Y_{p+1} touches the boundary of ν. The pSPT at the surface of the (p + 1) bubble ν then automatically attaches to all σ ∈ Y_{p} at the boundary of ν. At the same time, we notice that any given σ is the boundary of two or multiple ν ∈ Y_{p+1}, so that the state induced at σ comes from all bordering ν ∈ Y_{p+1}, which is then computed by Eq. (9).
It is important to realize that although \([{\tilde{\psi }}_{p+1}{ }_{\nu }]\in {\Phi }^{p}({G}_{\nu })\) is a pSPT protected by G_{ν}, their sum is a pSPT under a larger group G_{σ} ⊃ G_{ν} (after the orientation alignment is resolved by the coefficients 〈σ∣∂ν〉 in the summation). Therefore, we identify \([{\tilde{d}}_{1}{\tilde{\psi }}_{p+1}{ }_{\sigma }]\) as a pSPT protected by G_{σ}, i.e., \([{\tilde{d}}_{1}{\tilde{\psi }}_{p+1}{ }_{\sigma }]\in {\Phi }^{p}({G}_{\sigma })\subset {E}_{p,1}^{p}\). Eq. (9) maps (p + 1) bubbles to p blocks, establishing a linear map from \({E}_{p+1,1}^{p}\) to \({E}_{p,1}^{p}\), and since all elements in \({E}_{p+1,1}^{p}\) are topologically trivial, the image of the mapping is also trivial. These trivial states in \({\rm{img}}{\tilde{d}}_{1}\) give the equivalence relations in \({E}_{p,1}^{p}\): \([{\psi }_{p}] \sim [{\psi }_{p}^{\prime}]\) if and only if \([{\psi }_{p}][{\psi }_{p}^{\prime}]\in {\rm{img}}\tilde{d}\).
We use the example in Fig. 2b to illustrate the bubble equivalence. It is a fermionic system with charge conservation symmetry, and the lattice is 3D orthogonal with a mirror plane sending x to −x. We consider a certain [ψ_{2}], where some 2cells are decorated with Chern insulators, the Chern numbers of which are shown in Fig. 2b. It can be easily checked that the assembly in Fig. 2b satisfies the noopenedge condition. But in Fig. 2b, we show that the state in \({E}_{2,1}^{2}\) is actually equivalent to a state made from 3bubbles only, that is, a state in \({\rm{img}}{\tilde{d}}_{1}\). There are two 3bubbles in one unit cell. The boundary of the left bubble has Chern number +1, and that of the right boundary has Chern number −1. Therefore this assembly is topologically trivial.
Connectors and secondpage results
In this section, we explain how to solve the subtleties outlined in the section “Further considerations”. A key step in the computations is to determine the concrete content of the connectors decorated on the (d_{b} − 1) cells, which connect the SPT states on neighboring d_{b}cells. Using these connectors, we can compute the secondpage noopenedge conditions and bubbling equivalences, which together give the thirdpage result of the classification. The connector also allows us to solve the groupextension problem arised in the process of combining classifications of TCSs with different d_{b}.
Contents of the connectors: We begin by reviewing the content of connectors. Consider a pblock TCS \([\psi ]\in {E}_{p,2}^{p}\), whose building blocks are pdimensional SPT states decorated on the p cell. The connectors on the (p − 1) cells are then constrained by these building blocks through the bulkboundary relation. Previously, we studied the noopenedge condition in the section “Noopenedge condition,” which ensures the existence of gapped symmetric connectors. However, to further determine the concrete form of these connectors, we need not consider only the SPT phases [ψ_{σ}] decorated on σ ∈ Y_{p}, but also the wave functions representing these phases. Here, we use ψ to denote a wave function representing a TCS phase [ψ]. Just like [ψ] is a collection of local SPT phases, ψ is a collection of local SPT wave functions on all cells: the local SPT phase on σ ∈ Y_{p} is denoted by ψ_{σ} ∈ Ψ^{p}(G_{σ}), where Ψ^{d}(G) denotes the collection of ddimensional Gsymmetric wave functions, as reviewed in Sec. I of Supplementary Information. To form a symmetric wave function, we require that the local wave functions decorated to symmetryrelated cells satisfy the following symmetry condition, similar to the one in Eq. (4):
As ψ is made of pdimensional building blocks, the decorations ψ_{σ} = 0 on cells with dimensionality higher than p. However, the connectors, which are decorations on cells in dimensions lower than p, are in general not vanishing. Collectively, we denote the decorations on d cells by ψ_{d}: ψ_{p} is the building block on p cells, ψ_{p−1} is the connector on (p − 1) cells, etc.
Now consider a (p − 1) cell τ ∈ Y_{p−1}. The connector decorated to τ, which we denote by ψ∣_{τ}, satisfies the following bulkboundary relation:
where ψ_{1} ⊞ ψ_{2} denotes the wave function obtained by stacking the two wave functions ψ_{1} and ψ_{2}. As reviewed in Sec. I of Supplementary Information, for bosonic SPT states whose wave functions are represented by cochains, this stacking is just the normal addition between cochains. However, for fermionic SPT states, this stacking operation is not commutative, ψ_{1} ⊞ ψ_{2} ≠ ψ_{2} ⊞ ψ_{1}, because the statistical signs are associated with reordering of fermionic operators. Because of this subtlety, to unambiguously define the stacking in Eq. (15), one must choose an ordering between the neighboring cells of τ, and such ordering should be compatible with the crystal symmetries.
For simplicity, we introduce an operator ∂ to denote the operation on the righthand side of Eq. (15): ∂ transforms ψ to an anomaly pattern ∂ψ, whose components on each cell are given by
Intuitively, the operator ∂ transfers the wave functions on p cells to their boundary (p − 1) cells, where they are interpreted as boundary anomalies. Using this operator, the relation in Eq. (15) is simplied as
Since such relation exists on every (p − 1) cell, it gives a relation between the p blocks ψ_{p} and the (p − 1) connectors ψ_{p−1}:
For bosonic SPT states, the detailed formula for computing the ∂ operator can be found in Sec. III of Supplementary Information.
Comparing the righthand side of Eq. (15) to Eq. (6), it is easy to check that the SPT phase of (∂ψ)∣_{τ} is precisely (∂ψ)∣_{τ} ~d_{1}[ψ]∣_{τ}. Hence, for a secondpage TCS [ψ] in \({E}_{p,2}^{p}\), the noopenedge condition d_{1}[ψ] = 0 ensures that Eq. (17) has solutions for ψ∣_{τ}, representing possible choices of a connector bridging p cells bordering τ. In the rest of this section, we shall use solutions of this equation to address the problems raised in the section “Further considerations” and obtain a complete classification of TCSs.
a. Bosonic example. We use a simple bosonic example to demonstrate the process of determining the wave functions of connectors. As shown in Sec. V of Supplementary Information, such examples can only be nontrivial when the symmetry group G is not a direct product of SG and G_{0}. In fact, this example involves a magnetic translation symmetry group. This example is adapted from the result of ref. ^{53}. The connection to ref. ^{53} will be revealed in the section “Application to HOLSM Theorems”.
In this example, we consider 2D TCSs protected by the symmetry group \(G={G}^{M}\times {{\mathbb{Z}}}_{2}^{T}\), where \({{\mathbb{Z}}}_{2}^{T}\) is the usual (antiunitary) onsite timereversal symmetry, and G^{M} is a 2D magnetic translation symmetry group. G^{M} has three generators t_{x}, t_{y}, x, representing two translation symmetries and one onsite unitary \({{\mathbb{Z}}}_{2}\) symmetry, respectively. Both t_{x} and t_{y} commutes with x. However, t_{x} and t_{y} do not commute, and instead satisfy
In this case, the onsite symmetry group \({G}_{0}={{\mathbb{Z}}}_{2}^{x}\times {{\mathbb{Z}}}_{2}^{T}\), where \({{\mathbb{Z}}}_{2}^{x}\) denotes the \({{\mathbb{Z}}}_{2}\) group generated by x. The space group is the quotient group \(SG=G/{G}_{0}={{\mathbb{Z}}}^{2}\), generated by the two translation operations. SZD: SG or SG? However, G is not a direct product of G_{0} and SG, due to the nontrivial commutation relation in Eq. (19).
We first decompose the 2D plane \({{\mathbb{R}}}^{2}\) into the Gcomplex Y as outlined in the section “Cell decomposition and chain complex”. Since \(SG={{\mathbb{Z}}}^{2}\) is the simplest wallpaper group, the result of the decomposition is simply a generic oblique lattice, as shown in Fig. 3. There are no pointgroup symmetries anywhere, and the local symmetry group of each cell is just G_{σ} = G_{0}. Therefore, the SPT building blocks are obtained by attaching SPT states in Φ^{p}(G_{0}) = H^{p+1}[G_{0}, U(1)_{T}] to p cells.
In this example, we consider a particular 2block assembly [ψ] in \({E}_{2,1}^{2}\): on each 2cell, we decorate an SPT state represented by the following cocycle [α] ∈ Φ^{2}(G_{0}) = H^{3}[G_{0}, U(1)_{T}], represented by the following 3cocycle:
where the function β is defined as
Here, the \({{\mathbb{Z}}}_{2}\) variables n_{x} and n_{T} are obtained by writing the elements of G_{0} as the following canonical form:
This cocycle represents a nontrivial 2D SPT state protected by both x and T symmetries. The 2blocks of [ψ] are given by [ψ∣_{σ}] = [α].
It is straightforward to check that this element satisfies the cocycle equation on the first page, d_{1}[ψ] = 0, and remains a valid secondpage SPT state in \({E}_{2,2}^{2}\). To see this, we notice that [ψ] decorates the same SPT state on every 2cell. Therefore, on each 1cell, which borders two 2cells, there are two counterpropagating anomalous edge modes, and they cancel each other. Hence, this decoration [ψ] can be gapped out on 1cell.
However, gapping out this 1cell requires nontrivial connectors. In order to compute the connectors, we choose a wave function of 2blocks ψ_{2} representing [ψ]. As reviewed in Sec. I of Supplementary Information, on each 2cell, the wave function is a Gvalued G_{0}invariant 3cococyle \(\tilde{\alpha }\). Without losing generality, we choose the following 3cocycle \(\tilde{\alpha }\):
This equation looks similar to Eq. (20), but the group elements g_{i} take values in G instead of G_{0}, and \({\tilde{n}}_{X}\) is extracted by writing the group elements in the following canonical form:
Here, we emphasize that the \(\tilde{\alpha }\) given in Eq. (23) is not invariant under G actions (it is impossible to find a Ginvariant \(\tilde{a}\) representing this SPT phase, due to the nontrivial structure of G.). In fact, using the commutation relation in Eq. (19), one can show that
which is different from Eq. (23).
If we choose to decorate σ_{1} with \({\tilde{\alpha }}\) and let \({\psi} { }_{{\sigma }_{1}}=\tilde{\alpha }\), the symmetry constraint will fix the decoration on other 2cells. In particular, the decoration on σ_{3}, which is related to σ_{1} by the action of \({t}_{y}^{1}\), is given by \(\psi { }_{{\sigma }_{3}}={t}_{y}^{1}\cdot \tilde{\alpha }\). Using the explicit form of symmetry actions on cochains given in Sec. II of Supplementary Information, we get
Using the result of Eq. (25), we see that the decorations on the two cells are actually different: \(\psi { }_{{\sigma }_{1}}\;\ne\;\psi { }_{{\sigma }_{3}}\). In fact, the two decorations still belong to the same cohomology class in Φ^{2}(G_{0}) = H^{3}[G_{0}, U(1)_{T}]. However, to be compatible with the magnetic translation symmetry, different cochains (of the same cohomology class) have to be decorated to different 2cells. The difference between \(\psi { }_{{\sigma }_{3}}\) and \(\psi { }_{{\sigma }_{1}}\) then implies that one must decorate a nonvanishing 2cochain to its boundary, τ_{13}. Using the explicit form of the cochains, one can derive the explicit form of the cocycle equation \(d\psi { }_{{\tau }_{13}}=\psi { }_{{\sigma }_{3}}\psi { }_{{\sigma }_{3}}\):
We now need to choose an arbitrary solution of this equation. It is easy to check that the following 2cocycle is a choice:
On the other hand, on the 1cell along the x direction, we are allowed to simply choose \(\psi { }_{{\tau }_{12}}=0\), because the cocycle \(\tilde{\alpha }\) is invariant under the action of T_{x} and therefore \(\psi { }_{{\sigma }_{2}}=\psi { }_{{\sigma }_{1}}\). Hence, we have to use nontrivial connectors on some 1cell, because it is impossible to construct a wave function of [α] that is symmetric under all operations in SG.
b. Freefermion examples. Next, we present an example of determining nontrivial connectors in a 2D topological crystaline superconductor. We consider that the 2D system has a wallpaper group p2mm, the generators of which are \({\hat{M}}_{x}\) and \({\hat{M}}_{y}\), a timereversal symmetry, \(\hat{T}\), and a particle–hole symmetry, \(\hat{P}\). The algebra relations of these generators are given by \({\hat{T}}^{2}=1\), \({\hat{P}}^{2}=1\), \({\hat{M}}_{x}^{2}=1\), \({M}_{y}^{2}=1\), \([\hat{T},\hat{P}]=0\), \([\hat{T},{\hat{M}}_{x}]=0\), \(\{\hat{T},{\hat{M}}_{y}\}=0\), \([\hat{P},{\hat{M}}_{x}]=0\), and \([\hat{P},{\hat{M}}_{y}]=0\). Such relations can be realized in a superconductor with significant spin–orbit coupling and an order parameter projectively representing the \({\hat{M}}_{y}\) symmetry. To proceed, we represent these operators as \(\hat{T}=i{s}_{2}K\), \(\hat{P}={\mu }_{1}K\), \({\hat{M}}_{x}=i{\mu }_{3}{s}_{3}\), \({\hat{M}}_{y}={\mu }_{3}{s}_{2}\), where s_{1,2,3} are Pauli matrices representing the spin degree, and μ_{1,2,3} are Pauli matrices representing the “orbital” degree. (The meaning of “orbital” is twisted in BdG Hamiltonian.) The complex structure Y of p2mm is illustrated in Fig. 4a, where the only 2cell in the G orbits Y_{2}/G is σ_{1}, the four 1cell in Y_{1}/G are τ_{i=1,2,3,4}, and the four 0cell in Y_{0}/G are λ_{i=1,2,3,4}.
Now we decorate the 2cell σ_{1} with the following BdG Hamiltonian:
where M > 0. Due to the \({\hat{M}}_{y}\) symmetry, the Hamiltonian decorated on the 2cell σ_{3} should be \({\hat{M}}_{y}{\hat{H}}^{(L)}({k}_{x},{k}_{y}){\hat{M}}_{y}^{1}\), which is identical with Eq. (29). Therefore, we extend the Hamiltonian on σ_{1} (Eq. (29)) to σ_{3}. Apparently, there is no boundary state on τ_{2}. Due to the symmetry \({\hat{M}}_{x}\), the Hamiltonian on M_{x}σ_{1} and C_{2}σ_{1} can be derived as
which is different from Eq. (29). In order to determine the boundary state between H^{(L)} and H^{(R)}, we consider to insert an infinite barrier potential on the edges τ_{1} and M_{y}τ_{1} and the vertex λ_{1}. We assume that the boundary Hamiltonian on τ_{1} is \({\hat{H}}^{({\tau }_{1})}={\mu }_{0}{s}_{3}{k}_{y}\), where the upper block (μ_{3} = 1) is the boundary state of H^{(L)} and the lower block (μ_{3} = −1) is the boundary state of H^{(R)}. The timereversal symmetry acts locally on the two boundaries, and hence can be represented by \({\hat{T}}^{(b)}=i{s}_{2}K\). The particle–hole symmetry also acts locally on the two boundaries, and hence can be represented by \({\hat{P}}^{(b)}=K\). The mirror symmetry M_{x} interchanges the two blocks and hence must be proportional to iμ_{1} or iμ_{2}. M_{x} need to commute with \({H}^{({\tau }_{1})},{\hat{T}}^{(b)},{\hat{P}}^{(b)}\) and square to −1. We find that the only choice is \({\hat{M}}_{x}=i{\mu }_{2}\). We then soften the barrier to introduce coupling between the two blocks. The only symmetryallowed mass term that gaps \({\hat{H}}^{({\tau }_{1})}\) is μ_{2}s_{1}. Hence we model the gapped state on τ_{1} as
Now we study how \({\hat{H}}^{({\tau }_{1})}\) transforms under the M_{y} operation. Using the constraints \(\{{\hat{T}}^{(b)},{\hat{M}}_{y}^{(b)}\}=0\), \([{\hat{P}}^{(b)},{\hat{M}}_{y}^{(b)}]=0\), \(\{{\hat{M}}_{x}^{(b)},{\hat{M}}_{y}^{(b)}\}=0\), \({\hat{M}}_{y}^{(b)2}=1\), \({\hat{M}}_{y}^{(b)}\) can be chosen as μ_{1,3}s_{1,3}. Since M_{y} does not interchange the two blocks, we only consider the two options \({\hat{M}}_{y}^{(b)^{\prime} }={\mu }_{3}{s}_{1}\), \({\hat{M}}_{y}^{(b)^{\prime\prime} }={\mu }_{3}{s}_{3}\). Correspondingly, the boundary Hamiltonian on M_{y}τ_{1} under the two M_{y} representations is given by
respectively. Therefore, either the kinetic term or the mass term will be flipped under M_{y}, leading to a gapless domain wall at λ_{1}. This serves as a nontrivial connector on the 1cell in the vertical direction in Fig. 4a.
In the end, we consider an example of p + ip topological superconductor with nontrivial connector on the 1cell. We consider a p + ip superconductor in the wallpaper group p2. The cell decomposition of p2 is shown in Fig. 4b, where the only 2cell in the G orbits Y_{2}/G is σ_{1}, the three 1cell in Y_{1}/G are τ_{i=1,2,3}, and the four 0cell in Y_{0}/G are λ_{i=1,2,3,4}. We assume the Hamiltonian on σ_{1} as
where M > 0 and the particle–hole symmetry is represented as \(\hat{P}={s}_{x}K\). The Hamiltonian on C_{2}σ_{1} can be generated by acting the C_{2} operation on the above Hamiltonian. We consider the C_{2} operator \({\hat{C}}_{2}={s}_{0}\), it that commutes with \(\hat{P}\) and squares to 1. Thus the Hamiltonian on C_{2}σ_{1} is
In order to study the boundary state on τ_{1} and C_{2}τ_{1}, we consider to insert an infinite barrier potential on τ_{1}, C_{2}τ_{1}, and λ_{1}. Since the \({H}^{({\sigma }_{1})}({\bf{k}})\) and \({H}^{({C}_{2}{\sigma }_{1})}({\bf{k}})\) have the same chirality (rotation does not change chirality), the Majorana chiral modes from the two sides must move in opposite directions. We assume the boundary Hamiltonian on τ_{1} as \({\hat{H}}^{({\tau }_{1})}={k}_{y}{s}_{3}\), where the s_{3} = 1 state comes from the boundary of σ_{1} and the s_{3} = −1 state comes from boundary of C_{2}σ_{1}. Since the particle–hole symmetry acts locally on the two boundaries, we choose its representation as \({\hat{P}}^{(b)}=K\). Now, we soften the barrier to introduce coupling between the two Majorana modes. The only symmetryallowed mass term is s_{2}. Hence, we model the gapped state on τ_{1} as
Now we study how \({\hat{H}}^{({\tau }_{1})}\) transforms under the C_{2} rotation. Since C_{2} interchanges the two boundaries and commutes with \({\hat{P}}^{(b)}\), its representation must be offdiagonal and real, i.e., \({\hat{C}}_{2}^{(b)}={s}_{1}\). The boundary Hamiltonian on C_{2}τ_{1} is hence obtained as
The mass is flipped under the C_{2} rotation, leading to a gapless domain wall at λ_{1}. This serves as a nontrivial connector on the 1cell.
Secondpage noopenedge conditions: For a secondpage TCS \([\psi ]\in {E}_{p,2}^{p}\), the secondpage noopenedge condition demands that all (p − 2) cells can be filled with gapped symmetric connectors. The connectors decorated on (p − 2) cells, collectively denoted by ψ_{p−2}, must satisfy the bulkboundary relation similar to (18)
Hence, the existence of such connectors is determined by the condition that the r.h.s. of Eq. (38) belongs to the trivial SPT phase on each cell, ∂ψ_{p−1} ~ 0.
We introduce a linear map \({d}_{p,2}^{p}:{E}_{p,2}^{p}\to {E}_{p2,2}^{p1}\) to represent this noopenedge condition. As in “Results”, \({d}_{p,2}^{p}\) will be abbrivated to d_{2} if the domain of the map is clear from the context. For each element [ψ] in \({E}_{p,2}^{q}\), we choose a particular wave function ψ_{p} for the building blocks. Then, we choose an arbitrary solution ψ_{p−1} of Eq. (18). The image of d_{2} map is then defined as
Several remarks are in order: first, the domain of the d_{2} maps are the E_{2} modules, because ψ_{p−1} only exists if [ψ] belongs to the E_{2} module, where the cocycle condition d_{1}[ψ] = 0 guarantees the existence of ψ_{p−1}. Second, we explain why the images of d_{2} maps are the E_{2} modules. The meaning of this assertion is twofold: On one hand, d_{2}[ψ] = [∂ψ_{p−1}] satisfies the cocycle condition d_{1}d_{2}[ψ] because ∂^{2} = 0. Therefore, it indeed belongs to E_{2}. On the other hand, the equivalence relation of dψ_{p−2} = ∂ψ_{p−1} ~ 0 should be understood as the one in \({E}_{p2,2}^{p1}\): Instead of requiring the obstruction \([(\partial {\psi }_{p1}){ }_{\sigma }]\) to vanish on every (p − 2)cell σ, we only require that ∂ψ_{p−1} can be trivialized by a bubbling process \(\tilde{\psi }\in {E}_{p1,1}^{q+1}\): \(\partial {\psi }_{p1} \sim {d}_{1}[\tilde{\psi }]=\partial {\tilde{\psi }}_{p}\). This is because when extending ψ to (p − 1)cells, we can choose \({\psi }_{p1}^{\prime}={\psi }_{p1}{\tilde{\psi }}_{p}\) instead of ψ_{p−1} as the connectors. This choice of connectors then satisfies \((\partial {\psi }_{p1}^{\prime}){ }_{\mu } \sim 0\) on every (p − 2)cell.
In summary, the secondpage noopenedge condition, which tests whether [ψ] can be filled with gapped symmetric connectors on (p − 2)cells, is expressed as
a. Bosonic example. We now use this noopenedge condition to examine the example introduced in the section “Bosonic example”. We will see that the d_{2} map is nontrivial in this example. We consider the result of d_{2}[ψ] on the 0cell μ shown in Fig. 3. Recall that using the wavefunction realization ψ we chose in Eq. (23), the connectors ψ_{1} are given as follows: \({\psi }_{1}{ }_{{\tau }_{12}}={\psi }_{1}{ }_{{\tau }_{34}}=0\); \({\psi }_{1}{ }_{{\tau }_{13}}\) is given by Eq. (28). Hence, \({\psi }_{1}{ }_{{\tau }_{24}}\) is constrainted to be \({t}_{x}^{1}\cdot {\psi }_{1}{ }_{{\tau }_{13}}\) and has the following form,
Using these results of ψ_{1}, we can compute d_{2}[ψ] using the definition in Eq. (39),
Here, β is exactly the 2cocycle representing the projective representation of T^{2} = −1, or a Kramers doublet. Hence, in this case, \({d}_{2}:{E}_{2,2}^{3}\to {E}_{0,2}^{2}\) is a nontrivial map that sends \(\hat{\alpha }\) to the nontrivial element \([\tilde{\psi }]\) in \({E}_{0,2}^{2}\):
Here, \([\tilde{\psi }]\) is an anomalous pattern represented by the following decomposition on 0cells: \({[\tilde{\psi }]}_{\mu }=[\beta ]\) on all 0cells in Y. This implies that [ψ] is actually not a 2D TCS because it does not satisfy the secondpage noopenedge condition: it has open edges, actually Kramers doublets, on the 0cells.
Freefermion example: Next, we visit the freefermion example in the section “Freefermion examples”. In this example, we use the connector in Eq. (31) to gap out the 1cells in the y direction in Fig. 4. Furthermore, the symmetry condition in Eq. (14) implies that mass terms on different 1cells are related by symmetries. In particular, using the projected symmetry operation \({\hat{M}}_{y}^{(b)}={s}_{2}\), we see that the mass term must change sign under \({\hat{M}}_{y}^{(b)}\). Therefore the mass terms on the two cells τ_{1} and τ_{3}, related by M_{y}, must have opposite signs. Such two opposite mass terms would left a symmetry protected gapless point at y = 0, i.e., the 0cell λ_{1}. Replacing the mirror symmetries above with mirror symmetries on x = −1/2 and y = 1/2, and following the same analysis, one can easily show that all the 0cells in Y_{0}/G, λ_{1,2,3,4}, are gapless.
a. Secondpage bubbling equivalences. The secondpage bubbling equivalence can be computed using a similar d_{2} map. Each element \([\psi ]\in {E}_{p+2,2}^{p+1}\) represents such a bubbling process, where SPT bubbles are generated on (p + 2)cells. Unlike the bubbling processes studied in the section “Bubble equivalence”, it leaves the (p + 1)blocks intact, and changes the SPT phases on the pcells.
In order to compute the changes to the pcells, we also need to consider additional bubbles generated on lowerdimensional cells. On a \(p^{\prime}\)cell \(\tau \in {Y}_{p^{\prime} }\) (\(p^{\prime} <p+2\)), we can also generate a bubble ψ∣_{τ}, which we will refer to as a \(p^{\prime}\)bubble. Different from the (p + 2)bubbles, the lowerdimensional bubble can have a nontrivial filling: d(ψ∣_{τ}) ≠ 0, meaning that the process not only changes the wave functions on ∂τ by ψ∣_{τ}, but also changes the wave function on τ by d(ψ∣_{τ}). Such a process is allowed because the bubble and its filling satisfy the bulkboundary relation reviewed in Sec. I of Supplementary Information, and together form a gapped symmetric state on τ. In this way, a general bubbling process contains not only pbubbles, denoted by ψ_{p}, but also \(p^{\prime}\)bubbles \({\psi }_{p^{\prime} }\) for all \(p^{\prime} <p\). On \(p^{\prime}\)cells, the total changes made by the bubbling include the \((p^{\prime} +1)\)bubbles and the filling of the \(p^{\prime}\)bubbles:
We now compute the changes to pcells for a bubbling process \([\psi ]\in {E}_{p+2,2}^{p+1}\). First, we choose a wavefunction realization ψ_{p+2} of the (p + 2)bubbles. Since the bubbling process leaves (p + 1)blocks intact, the (p + 1)bubble ψ_{p+1} must satisfy
The existence of solutions of this equation is provided by the fact that \([\psi ]\in {E}_{p+2,2}^{p+1}\) satisfies d_{1}[ψ] = 0. We then choose an arbitrary solution of ψ_{p+1}, and the SPT phases the process generates is dψ_{p} ⊞ ∂ψ_{p+1} ~ [∂ψ_{p+1}]. Therefore, this process can be represented by the following d_{2} map,
Note, that the d_{2} map defined here is different from the one defined in Eq. (39) by a minus sign in the constraint equation. Hence, we can use a generic definition to unify the two d_{2} maps:
We notice that a nontrivial example of d_{2} bubbling process is provided in the section “Application to HOLSM Theorems”.
Taking into account the noopenedge conditions and bubbling equivalences given by the d_{2} map, the thirdpage approximation of the TCS classification is given by the cohomology group of d_{2}:
Higherpage results
This process can be generalized to higher pages.
To consider the rth page noopenedge conditions, we start with an rth page assembly \(\psi \in {E}_{p,r}^{p}\). On previous pages, we have constructed the connectors ψ_{p−1}, ..., ψ_{p−r+1}. The noopenedge condition on the previous page, d_{r−1}[ψ] = 0, guarantees that the equation dψ_{p−r} = ∂ψ_{p−r+1} has solutions. We then pick a solution of ψ_{p−r}, and define \({d}_{r}[\psi ]=[\partial {\psi }_{pr}]\in {E}_{pr,r}^{pr+1}\). The noopenedge condition is given by d_{r}[ψ] = 0.
Similarly, consider an rth page bubbling process \([\psi ]\in {E}_{p+r,r}^{p+r1}\). On previous pages, we have chosen lowerdimensional bubbles ψ_{p+r−2}, ..., ψ_{p+2}, such that the process generates nothing on cells with dimensions higher than p + 2. In order to get a process that also generates nothing on (p + 1)cells, we choose ψ_{p+1} satisfying dψ_{p+1} ⊞ ∂ψ_{p+2} = 0. The existence of solutions is provided by d_{r−1}[ψ] = 0. The resulting trivial TCS is then given by d_{r}[ψ] defined as [∂ψ_{p+1}].
Similar to Eq. (47), we write a unified definition for d_{r}:
Therefore, the classification on the next page is given by the cohomology group of d_{r}:
Iteratively, this process computes a series of pages \({E}_{p,1}^{q},{E}_{p,2}^{q},\ldots\), where \({E}_{p,r}^{p}\) provides a series of finer and finer approximations to the classification of pblock TCSs. Since we are eliminating false and redundant entries on each page, the list of candidate assemblies are getting smaller and smaller, \({E}_{p,1}^{q}\supseteq {E}_{p,2}^{q}\supseteq \cdots \). In the limit of r → ∞, this series of approximations reveals the true answer of the classification problem, which we denote by \({E}_{{p},{\infty} }^{q}\). In particular, \({E}_{{p},{\infty} }^{p}\) classify all pblock TCSs. In fact, this process only takes a finite number of steps to converge to \({E}_{{p},{\infty} }^{q}\), because the d_{r} map reduces the spatial dimension of the cells by r and necessarily becomes trivial once r exceeds the dimension of Y.
Recombing states with different d_{b} through group extension
In our topologicalcrystal constructions, we first divide ddimensional SPTs into TCSs with different buildingblock dimensions d_{b} = 0, 1, …d. We then compute the classification for each pblocks separately. The classification for pblock TCSs is then given by \({E}_{p,{\infty} }^{p}\), which is calculated by a series of cohomologygroup computations. Next, to obtain the full classification of crystaline SPT states, we need to recombine results of d_{b} = 1, 2, …d. We also need to include results of d_{b} = 0 if we want to recover all bosonic SPTs in H^{d+1}[G, U(1)_{T}]. However, as briefly mentioned in the section “Further considerations”, such recombination may not be a simple direct sum but a nontrivial group extension. In this section, we explain how this group extension is computed in general. Sec. V of Supplementary Information will use this method to prove that, for the simple cases G = SG × G_{0}, the group extension is always trivial and one can just take a direct sum.
We begin by recalling that a TCS \({E}_{p,{\infty} }^{p}\) are labeled by different building blocks on pcells, but each state [ψ] also contains connectors on all lowerdimensional cells. The lowerdimensional decorations will affect the results of adding (stacking) two SPTs states if the decoration cancels on higherdimensional cells.
To be more specific, consider an ordern TCS \([\psi ]\in {E}_{p,{\infty} }^{p}\), such that n[ψ] ~0 as in \({E}_{p,{\infty} }^{p}\). This implies that stacking n coplies of [ψ] results in a state \([\tilde{\psi }]=n[\psi ]\) which is trivial if viewed as an element in \({E}_{p,{\infty} }^{p}\). In other words, \([\tilde{\psi }]\) has trivial decorations on all pcells. However, \([\tilde{\psi }]\) may have nontrivial decorations on lowerdimensional cells, and thus should be viewed as a nontrivial topological crystal with a lower buildingblock dimension. To compute this, recall that in the topological crystal [ψ], the subleading terms \({\psi }_{p^{\prime} }\), representing the decoration on \(p^{\prime}\)cells, are obtained in the spectralsequence computation. Using these subleading terms, the decorations on lowerdimensional cells in \(\tilde{\psi }=n\psi\) is computed as
One can then look \({\tilde{\psi }}_{pr}\) up in \({E}_{pr,\infty }^{pr}\) to see whether it is nontrivial. The smallest r such that \({\tilde{\psi }}_{pr}\) is nontrivial then indicates \(\tilde{\psi }=n\psi\) is a nontrivial TCS with (p − r)blocks. When this happens, combining \({E}_{p,{\infty} }^{p}\) and \({E}_{pr,\infty }^{pr}\) then becomes a nontrivial groupextension problem instead of a direct sum.
We notice that the answer of whether n[ψ] is a nontrivial SPT state may depend on the choice of the subleading terms of the generator ψ (on the contrary, the computation of \({E}_{p,r}^{p}\) does not depend on this choice). However, the final result of the groupextension problem is independent of the choice of the generators. In Sec. V of Supplementary Information, we will show that for the simple cases of bosonic TCSs G = SG × G_{0}, if we choose to ignore the 0block TCS, there exists a simple choice of ψ_{p−1} such that the groupextension problem becomes trivial, and a naive direct sum gives the correct classification.
Examples
The algorithm to classify TCSs outlined in above sections can be automated for the bosonic case, using the formulation given in SI. For simplicity, here we only consider the case when the total group is a direct product of space (wallpaper) group and a local symmetry group, i.e., G = SG × G_{0}. In this case, since d_{2} map is trivial, we only need to take care of the firstpage noopenedge condition and the bubble equivalence. By an automated script, we have enumerate the bosonic TCSs with seveal local symmetry groups in all the wallpaper groups and all the space groups. The main results can be found in Supplementary Tables I and II, respectively.
In our results, we find that, although some of the 2D constructions of TCSs are equivalent to decoupled straight lines or straight planes, some of them, however, are beyond such simple layer constructions. Here, we give a bosonic example of a TCS beyond layer constructions, which has a geometric structure similar to an example studied in ref. ^{36}. We consider the space group \(P\bar{4}n2\) with the local symmetry group Z_{2}. The 2cells and 1cells are shown in Fig. 5. The details of the 2cells and 1cells are given in Sec. VI of Supplementary Information. According to refs. ^{3,4,61}, Z_{2} symmetry protects a \({{\mathbb{Z}}}_{2}\) 2D SPT. Thus we can decorate each 2cell with a such a 2D SPT. As discussed in Sec. V of Supplementary Information, the anomalies of these 2D SPTs can cancel each other on the 1cells where they meet. Thus, the noopenedge condition reduces to the constraint that there should be even number of 2D SPT ending at each 1cell^{36}. On the other hand, the bubble equivalence is trivial because on the 2cells the \({{\mathbb{Z}}}_{2}\) bubble is always canceled by its symmetry partner^{36}. Therefore, the classification of bosonic TCS in this case is just given by the allowed decorating configurations. The boundaries of the four inequivalent 2cells are
where T represents the translation group. Since all the 2cells have the same boundary, in order to cancel the anomalies on the 1cell we only need to decorate two of the four 2cells. There are three independent decorations: σ_{1} + σ_{3}, σ_{1} + σ_{2}, σ_{3} + σ_{4}. As shown in Fig. 5b–d, σ_{1} + σ_{2} and σ_{3} + σ_{4} decompose into horizontal and vertical planar layers, whereas σ_{1} + σ_{3} does not. In fact, using similar argument in ref. ^{36}, one can show that σ_{1} + σ_{3} is not equivalent with any layer construction.
Next, we demonstrate the application of our realspace recipe to classify interactingfermion SPTs by revisiting an example discussed in ref. ^{62}. In this example, we consider a 1D superconducting system with a timereversal symmetry T and a mirrorreflection symmetry M that reverses the whole 1D system, and we assume that the fermions transform as spinless fermions under both T and M (i.e. fermions carry linear representations T^{2} = M^{2} = +1). When considering interactingfermion classification, being a superconducting system means there is no U(1) chargeconservation symmetry. To classify topological superconductors in such a system, we divide the 1D system into two 1cells σ_{1,2} and a 0cell τ, as shown in Fig. 6. The 0cell is located at the reflection center, and the two 1cells are mapped to each other by the reflection symmetry M. In 1D, the TCSs we are interested in are given by \({E}_{1,\infty }^{1}\), denoting decorating 1D blocks on 1cells (we are ignoring 0D decorations in \({E}_{0,\infty }^{0}\), as we always do in this paper.)
On the first page, \({E}_{1,1}^{1}\) is given by Eq. (5). Here, Y_{1}/G contains only one Gorbit containing both σ_{1,2}. Choosing σ_{1} to represent this orbit, \({E}_{1,1}^{1}\) is given by \({\Phi }^{1}({G}_{{\sigma }_{1}})\), which contain 1D SPT states protected by the onsite symmetry group \({G}_{{\sigma }_{1}}={{\mathbb{Z}}}_{2}^{T}\). It is wellknown that \({\Phi }^{1}({{\mathbb{Z}}}_{2}^{T})\), representing the phases 1D topological superconductors with T^{2} = + 1, has a \({{\mathbb{Z}}}_{8}\) classification, where the root state is the 1D Kitaev chain^{63,64}. Hence, we have \({E}_{1,1}^{1}={{\mathbb{Z}}}_{2}\).
Next, we compute the firstpage noopenedge condition. (There is no bubbling equivalence in this case, because there is no 2cell.) This condition is described by the firstpage derivative \({d}_{1,1}^{1}:{E}_{1,1}^{1}\to {E}_{0,1}^{1}\). Here, the codomain of this derivative is the space of anomaly patterns on the 0cell τ: \({E}_{0,1}^{1}={\Phi }^{1}({G}_{\tau })\). As the reflection center, \({G}_{\tau }={{\mathbb{Z}}}_{2}^{T}\times {{\mathbb{Z}}}_{2}^{M}\), where M is a unitary onsite symmetry. Correspondingly, the classification of anomaly patterns is given by \({\Phi }^{1}({G}_{\tau })={{\mathbb{Z}}}_{8}\oplus {{\mathbb{Z}}}_{4}\), where the \({{\mathbb{Z}}}_{8}\) and the \({{\mathbb{Z}}}_{4}\) subgroups are generated by the same Kitaev chain and an additional fermionic SPT state with complexfermion decorations, respectively^{65}. The second root state does not play a role in our calculation, so its details will not be discussed here. Representing elements of \({E}_{1,1}^{1}\) and \({E}_{0,1}^{1}\) by a mod8 integer n and a pair of a mod8 integer and a mod4 integer (n_{1}, n_{2}), respectively, the derivative \({d}_{1,1}^{1}\) has the form \({d}_{1,1}^{1}(n)=(2n,0)\). The first component of \({d}_{1,1}^{1}(n)\) is 2n, because σ_{1} and σ_{2} should be decorated by the same state denoted by n as the result of the symmetry condition, and their total contribution to the anomaly on τ is thus 2n. The second component of \({d}_{1,1}^{1}(n)\) can be computed explicitly using the approach in Sec. IC of Supplementary Information, but this result does not affect the classification of \({E}_{1,\infty }^{1}\), and hence will not be discussed here. Using the explicit form of \({d}_{1,1}^{1}\), we see that the noopenedge condition \({d}_{1,1}^{1}(n)=0\) demands that \(2n=0\,\mathrm{mod}\,\,8\), or n = 0, 4. Therefore, the secondpage result is \({E}_{1,2}^{1}=\ker {d}_{1,1}^{1}={{\mathbb{Z}}}_{2}\), generated by n = 4, representing decorating four copies of Kitaev chains on σ_{1,2}. In 1D, \({E}_{1,2}^{1}={E}_{1,\infty }^{1}\) is the final answer. Hence, the topological superconductors in such a 1D system has a \({{\mathbb{Z}}}_{2}\) classificationm, and the root state is the interactionenabled topological superconductor state studied in ref. ^{62}.
The steps described above can be generalized to study interactingfermion SPT states with more complex spacegroup symmetries. Some examples of using similar ideas to study realspace construction of fermionic topological crystalline states can be found in refs. ^{41,62,66}. We shall leave a fully automated implimentation of our realspace recipes to future works.
Application to HOLSM theorems
In this section, we apply our TCS constructions to obtain generalized HOLSMtype theorems. We first review the original HOLSM Theorem and its generalizations to different onsite symmetry groups, and discuss how to understand them using our TCS constructions. We then revisit the first example given in the section “Secondpage noopenedge conditions“, and discuss how to reinterpret it as an SPTenforcing HOLSM Theorem^{53,57}. Last, we will give the general relation between TCS constructions and generalized HOLSM Theorems. For simplicity, we first discuss the 2D examples, then generalize our results to 3D.
The original HOLSM Theorem asserts that in a 2D lattice with translation symmetries and spinrotation symmetries, if there is an odd number of spin\(\frac{1}{2}\) per unit cell, the system cannot have a symmetric gapped unique ground state. This theorem is later generalized to the cases where the spinrotation symmetry and the spin\(\frac{1}{2}\) objects are replaced by an arbitraty onsite symmetry group G_{0} and a nontrivial projective representation of G_{0}, respectively. In this section, we will refer to the original and these generalizations as the "generalized HOLSM Theorems”.
Using our TCS framework, we can view the distribution of projective representations as a nontrivial anomaly pattern \([\tilde{\psi }]\) in the module \({E}_{0,\infty }^{1}\). In our language, the total symmetry group is G = SG × G_{0}, where the space group \(SG={{\mathbb{Z}}}^{2}\). Hence, the Gcomplex Y we construct is the same as the one shown in Fig. 3, with one 0cell per unit cell. The local symmetry group on the 0cell is simply G_{0}. The nontrivial projective representation can be translated to a nontrivial element [β] = H^{2}[G_{0}, U(1)_{T}] = Φ^{1}(G_{0}), which is decorated to the 0cells, as \({[\tilde{\psi }]}_{\mu }=[\beta ]\). In this way, the distribution of a nontrivial projective representation is translated to an anomaly pattern \([\tilde{\psi }]\).
We now argue that, including this \([\tilde{\psi }]\), every nontrivial element in \({E}_{0,\infty }^{1}\) represents an anomaly pattern that cannot be gapped out by a symmetric unique ground state. This is done by reinterpreting the noopenedge conditions we introduced in the section “Connectors and secondpage results”. Consider an element \([\tilde{\psi }]\) in \({E}_{0,r}^{1}\) that is trivialized by the d_{r} map,
In the section “Connectors and secondpage results”, we interpreted this relation as the fact that, the assembly [ψ] does not satisfy the noopenedge condition and cannot be realized as a TCS, because assembling it will result in anomaly patterns specified by \([\tilde{\psi }]\) on the 0cells. However, if the physical Hilbert space already contains an anomaly pattern \([\tilde{\psi }]\) on the 0cells, the assembly − [ψ] can be realized in such physical systems, because the obstruction d_{r}(−[ψ]) = −[ψ] is now canceled by the background anomaly pattern in the Hilbert space. Therefore, Eq. (56) also implies that the anomaly pattern − [ψ] can be gapped out by the TCS assembly [ψ]. In other words, it reveals a UV/IR anomaly matching between the TCS assembly (which can be viewed as the IR limit) and the anomaly pattern (which can be viewed as the UV limit).
A corollary of this reinterpretation is that a nontrivial element in \({E}_{0,\infty }^{1}\) cannot be gapped out by any such TCS assembly, and therefore cannot realize a symmetric gapped unique ground state.
We can also revisit the first example in the section “Secondpage noopenedge conditions” using this alternative interpretation. Recall that Eq. (43) indicates that [ψ] does not represent a valid 2D GSPT state: tiling the 2D plane with the SPT phase [α] will leave one gapless Kramers doublet in each unit cell. However, if we start with a model that has one Kramers doublet per unit cell in the original Hilbert space, this will cancel the anomaly of d_{2}[ψ] and allows the construction of the 2D SPT [α]. The construction of the trivial SPT state in such system, however, becomes impossible, because the anomaly would be left uncanceled. In other words, the existence of a nontrivial anomaly pattern in the Hilbert space requires that a gapped unique ground state must be a nontrivial SPT state. This is precisely the theorem proved in ref. ^{53}, which we will refer to as an SPTenforcing theorem.
In general, we can express these HOLSMlike constraints as the following: In our language, a spatial distribution of projective representations is represented by an anomaly pattern \([\tilde{\psi }]\) in \({E}_{0,1}^{1}\). If \([\tilde{\psi }]\) is trivialized through d_{1} by a \([\psi ]\in {E}_{1,1}^{1}\) as \([\tilde{\psi }]={d}_{1}[\psi ]\), then \([\tilde{\psi }]\) can be gapped out by the 1block TCS assembly [ψ]. If \([\tilde{\psi }]\) is a nontrivial element in \({E}_{0,2}^{1}\) but is trivialized through d_{2} by \(\hat{\alpha }\in {E}_{2,2}^{2}\) as \([\tilde{\psi }]={d}_{2}[\psi ]\), then \([\tilde{\psi }]\) can be gapped out by the 2block TCS assembly [ψ], which must be a strong SPT state (i.e., it is protected solely by G_{0}) because 2cells in a 2D space must have G_{σ} = G_{0}. Furthermore, if \([\tilde{\psi }]\) is a nontrivial element in \({E}_{0,3}^{1}={E}_{0,\infty }^{1}\), then it cannot be gapped out by an SPT state, and will give the consequences of the original LSM Theorem. Therefore, the spectral sequence introduced in the section “Connectors and secondpage results” provides a way to compute these constraints.
These constraints can be further generalized to 3D. There, an anomaly pattern, which is an element in \({E}_{0,1}^{1}\), can be trivialized through d_{1} by a TCS assembly in \({E}_{1,1}^{2}\), through d_{2} by a TCS assembly in \({E}_{2,2}^{2}\), through d_{3} by a TCS assembly in \({E}_{3,3}^{3}\), or cannot be trivialized at all. Among these possiblities, \({E}_{3,3}^{3}\) represents strong 3D SPT states.
The UV/IR anomaly matching condition in Eq. (56), and the resulting SPTenforcing HOLSM Theorems, can be understood by viewing the anomaly pattern as the surface anomaly of a 3D bulk state. We first explain this for the simple cases of the generalized HOLSM Theorem with \(G={G}_{0}\times {{\mathbb{Z}}}^{2}\). Since the projective representation [β] is the edge state of a 1D SPT state, the 3D bulk state can be constructed by decorating one such SPT chain in each 3D unit cell, as shown in Fig. 7a. Here, we argue that the 3D bulk state is closely related to \([\tilde{\psi }]\), and can be constructed mathematical from it. The 3D bulk has the same symmetry G as its surface. Hence, we construct a 3D topological space \(\bar{Y}\) compatible with SG. \(\bar{Y}\) has one 3cell, two 2cells and one 1cell in each unit cell (it has no 0cells). The aforementioned 3D bulk state is represented by an SPT pattern \([\bar{\psi }]\in {\bar{E}}_{1,\infty }^{1}\), constructed by decorating one Haldane chain on each 1cell. Here, the bar on \({\tilde{E}}_{p,r}^{q}\) indicates that it is the spectral sequence constructed for \(\bar{Y}\) instead of Y. It is easy to realize that there is a onetoone correspondence between pcells in Y and (p + 1)cells in \(\bar{Y}\), which can be expressed as an isomorphism \({Y}_{p}\simeq {\bar{Y}}_{p+1}\). We denote the corresponding cells in Y_{p} and \({\bar{Y}}_{p+1}\) as σ and \(\bar{\sigma }\), respectively. Using \(\bar{Y}\), the SPT pattern \([\bar{\psi }]\) can be expressed as decorating one Haldane chain to each 1cell in \(\bar{Y}\). Next, we notice that, mathematically, the anomaly decorated to each σ ∈ Y_{0} and the 1D SPT decorated to each \(\bar{\sigma }\in {\bar{Y}}_{1}\) are represented by the same 2cocycle [β] in H^{2}[G_{0}, U(1)_{T}]. Hence, \([\tilde{\psi }]\) and \([\bar{\psi }]\) can be converted to each other by copying the decoration between corresponding cells. Mathematically, this is described by the fact that the isomorphism \({Y}_{p}\simeq {\bar{Y}}_{p+1}\) naturally induces an isomorphism \({E}_{0,\infty }^{1}\simeq \bar{E}{1}_{1,\infty }\): the isomorphsm map, denoted by \(L:{E}_{0,\infty }^{1}\to {\bar{E}}_{1,\infty }^{1}\), is given by
Using this isomorphism, the relation between the anomaly pattern \([\tilde{\psi }]\) and the corresponding bulk state \([\bar{\psi }]\) is then given by \(\bar{\psi }=L([\tilde{\psi }])\).
The surfacebulk correspondence illustrated above can be generalized to arbitrary symmetry groups and dimensions. We consider a 2D surface and a 3D bulk with the same symmetry group G. Here, the space group SG = G/G_{0} is a 2D wallpaper group instead of a 3D space group. Similar to the previous example, the cellular decomposition for the 3D bulk can be constructed from the one of the 2D surface using the isomorphism \({\bar{Y}}_{p+1}\simeq {Y}_{p}\). Again, this induces an isomorphism between \({E}_{p,r}^{q}\) and \({\bar{E}}_{p+1,r}^{q}\) for arbitrary p, q, and r through the definition in Eq. (57). To understand the physical meaning of this correspondence, we notice that the pcell σ ∈ Y_{p} is an edge of the (p + 1)cell \(\bar{\sigma }\in {\bar{Y}}_{p+1}\). In this way, for \([\tilde{\psi }]\in {E}_{p,r}^{p+1}\), representing an rthpage anomaly pattern on the surface, \(L([\tilde{\psi }])\) represents decorating on \(\bar{\sigma }\) the bulk SPT state \([\tilde{\psi }]{ }_{\sigma }\) that corresponds to the anomaly \([\tilde{\psi }]{ }_{\sigma }\) on its boundary σ. Hence, \(L([\tilde{\psi }])\in {\bar{E}}_{p+1,r}^{p+1}\) is the bulk SPT state corresponding to the surface anomaly pattern \(\hat{\omega }\). Similarly, consider a surface TCS \([\psi ]\in {E}_{p,r}^{p+1}\). \(L([\psi ])\in {\bar{E}}_{p+1,r}^{p+1}\) is a bubbling pattern that generates SPT states [ψ] on the edges of cell \(\bar{\sigma }\), including σ itself. Hence, the 3D bubbling pattern L([ψ]) generates the 2D TCS [ψ] on the surface.
In summary, the isomorphism L defined above allows us to express the correspondence between 2D anomaly patterns and 3D TCSs, and between 2D TCSs and 3D bubbling patterns. It also allows us to convert the 2D noopenedge conditions to 3D bubbling equivalences. Consider a d_{r} map between a 2D assembly and a 2D anomaly pattern, as in Eq. (56). The Lisomorphism maps the r.h.s. to a 3D TCS that can host the anomaly pattern on its 2D surface, and the r.h.s. to a 3D bubbling pattern that generates the 2D assembly on its surface (see Fig. 7b). In this way, the relation (56) also describes a 3D bubbling equivalence,
This dimensionalshifting correspondence is consistent with our understanding that the TCS \(L([\tilde{\psi }])\) is actually trivial because its surface anomaly \([\tilde{\psi }]\) can be realized as a gapped symmetric state.
As an example, we apply this bulkboundary correspondence again to the first example in the section “Secondpage noopenedge conditions“. The Lisomorphism maps the relation in Eq. (43) to the following bubbling relation,
This indicates that \(L([\tilde{\psi }])\) is a 3D TCS trivialized by the secondpage bubbling process L([ψ]). This provides an explicit example illustrating that it is necessary to consider higherdimensional trivialization processes as discussed in the section “Further considerations”, which can be computed using the higherpage d_{r} maps introduced in the section “Secondpage bubbling equivalences”.
Data availability
Data of bosonicSPT classification protected by 2D wallpaper groups and 3D space groups together with various onsite symmetry groups can be found in Sec. VII of SI. Other data that support the findings of this study are available from the corresponding author upon request.
Code availability
All numerical codes in this paper are available upon request to the authors.
References
Gu, Z.C. & Wen, X.G. Tensorentanglementfiltering renormalization approach and symmetryprotected topological order. Phys. Rev. B 80, 155131 (2009).
Chen, X., Liu, Z.X. & Wen, X.G. Twodimensional symmetryprotected topological orders and their protected gapless edge excitations. Phys. Rev. B 84, 235141 (2011).
Chen, X., Gu, Z.C., Liu, Z.X. & Wen, X.G. Symmetryprotected topological orders in interacting bosonic systems. Science 338, 1604–1606 (2012).
Chen, X., Gu, Z.C., Liu, Z.X. & Wen, X.G. Symmetry protected topological orders and the group cohomology of their symmetry group. Phys. Rev. B 87, 155114 (2013).
Wen, X. G. Topological orders in rigid states. Int. J. Mod. Phys. B 04, 239–271 (1990).
Wen, X.G. Topological orders and ChernSimons theory in strongly correlated quantum liquid. Int. J. Mod. Phys. B 05, 1641–1648 (1991).
Affleck, I., Kennedy, T., Lieb, E. H. & Tasaki, H. Rigorous results on valencebond ground states in antiferromagnets. Phys. Rev. Lett. 59, 799–802 (1987).
Hasan, M. Z. & Kane, C. L. Colloquium: topological insulators. Rev. Mod. Phys. 82, 3045–3067 (2010).
Qi, X.L. & Zhang, S.C. Topological insulators and superconductors. Rev. Mod. Phys. 83, 1057–1110 (2011).
Chen, X., Lu, Y.M. & Vishwanath, A. Symmetryprotected topological phases from decorated domain walls. Nat. Commun. 5, 3507 (2014).
Vishwanath, A. & Senthil, T. Physics of threedimensional bosonic topological insulators: surfacedeconfined criticality and quantized magnetoelectric effect. Phys. Rev. X 3, 011016 (2013).
Wang, C. & Senthil, T. Boson topological insulators: a window into highly entangled quantum phases. Phys. Rev. B 87, 235122 (2013).
Burnell, F. J., Chen, X., Fidkowski, L. & Vishwanath, A. Exactly soluble model of a threedimensional symmetryprotected topological phase of bosons with surface topological order. Phys. Rev. B 90, 245122 (2014).
Wen, X.G. Construction of bosonic symmetryprotectedtrivial states and their topological invariants via G × SO(∞) nonlinear σ models. Phys. Rev. B 91, 205101 (2015).
Schnyder, A. P., Ryu, S., Furusaki, A. & Ludwig, A. W. W. Classification of topological insulators and superconductors in three spatial dimensions. Phys. Rev. B 78, 195125 (2008).
Kitaev, A. Periodic table for topological insulators and superconductors. AIP Conf. Proc. 1134, 22–30 (2009).
Gu, Z.C. & Wen, X.G. Symmetryprotected topological orders for interacting fermions: fermionic topological nonlinear σ models and a special group supercohomology theory. Phys. Rev. B 90, 115141 (2014).
Kapustin, A., Thorngren, R., Turzillo, A. & Wang, Z. Fermionic symmetry protected topological phases and cobordisms. J. High. Energy Phys. 2015, 1–21 (2015).
Gaiotto, D. & Kapustin, A. Spin tqfts and fermionic phases of matter. Int. J. Mod. Phys. A 31, 1645044 (2016).
Kapustin, A. & Thorngren, R. Fermionic spt phases in higher dimensions and bosonization. J. High. Energy Phys. 2017, 80 (2017).
Putrov, P., Wang, J. & Yau, S.T. Braiding statistics and link invariants of bosonic/fermionic topological quantum matter in 2+1 and 3+1 dimensions. Ann. Phys. 384, 254 – 287 (2017).
Cheng, M., Tantivasadakarn, N. & Wang, C. Loop braiding statistics and interacting fermionic symmetryprotected topological phases in three dimensions. Phys. Rev. X 8, 011054 (2018).
Wang, Q.R. & Gu, Z.C. Towards a complete classification of symmetryprotected topological phases for interacting fermions in three dimensions and a general group supercohomology theory. Phys. Rev. X 8, 011055 (2018).
Wang, J. et al. Tunneling topological vacua via extended operators: (spin)TQFT spectra and boundary deconfinement in various dimensions. Prog. Theor. Exp. Phys. 2018, 053A01 (2018).
Cheng, M., Bi, Z., You, Y.Z. & Gu, Z.C. Classification of symmetryprotected phases for interacting fermions in two dimensions. Phys. Rev. B 97, 205109 (2018).
Wang, Q.R. & Gu, Z.C. Construction and classification of symmetry protected topological phases in interacting fermion systems. Preprint at http://arxiv.org/abs/1811.00536 (2018).
Lan, T., Zhu, C. & Wen, X.G. Fermion decoration construction of symmetryprotected trivial order for fermion systems with any symmetry and in any dimension. Phys. Rev. B 100, 235141 (2019).
Bradley, C. J. & Cracknell, A. P. The Mathematical Theory of Symmetry in Solids: Representation Theory for Point Groups and Space Groups (Oxford University Press, 2010).
Fu, L. Topological crystalline insulators. Phys. Rev. Lett. 106, 106802 (2011).
Chiu, C.K., Teo, J. C. Y., Schnyder, A. P. & Ryu, S. Classification of topological quantum matter with symmetries. Rev. Mod. Phys. 88, 035005 (2016).
Bradlyn, B. et al. Topological quantum chemistry. Nature 547, 298–305 (2017).
Po, H. C., Vishwanath, A. & Watanabe, H. Symmetrybased indicators of band topology in the 230 space groups. Nat. Commun. 8, 50 (2017).
Kruthoff, J., de Boer, J., van Wezel, J., Kane, C. L. & Slager, R.J. Topological classification of crystalline insulators through band structure combinatorics. Phys. Rev. X 7, 041069 (2017).
Song, Z., Zhang, T., Fang, Z. & Fang, C. Quantitative mappings between symmetry and topology in solids. Nat. Commun. 9, 3530 (2018).
Khalaf, E., Po, H. C., Vishwanath, A. & Watanabe, H. Symmetry indicators and anomalous surface states of topological crystalline insulators. Phys. Rev. X 8, 031070 (2018).
Song, Z., Huang, S.J., Qi, Y., Fang, C. & Hermele, M. Topological states from topological crystals. Sci. Adv. 5, eaax2007 (2019).
Shiozaki, K. & Sato, M. Topology of crystalline insulators and superconductors. Phys. Rev. B 90, 165114 (2014).
Shiozaki, K., Sato, M. & Gomi, K. AtiyahHirzebruch spectral sequence in band topology: general formalism and topological invariants for 230 space groups. Preprint at http://arxiv.org/abs/1802.06694 (2018).
Thorngren, R. & Else, D. V. Gauging spatial symmetries and the classification of topological crystalline phases. Phys. Rev. X 8, 011040 (2018).
Isobe, H. & Fu, L. Theory of interacting topological crystalline insulators. Phys. Rev. B 92, 081304 (2015).
Song, H., Huang, S.J., Fu, L. & Hermele, M. Topological phases protected by point group symmetry. Phys. Rev. X 7, 011020 (2017).
Huang, S.J., Song, H., Huang, Y.P. & Hermele, M. Building crystalline topological phases from lowerdimensional states. Phys. Rev. B 96, 205106 (2017).
McCleary, J. A user’s guide to spectral sequences, 2nd edn. in Cambridge Studies in Advanced Mathematics (Cambridge University Press, 2000).
Brown, K. S. Cohomology of Groups, in Graduate texts in mathematics. (SpringerVerlag, New York, 1982).
Lieb, E., Schultz, T. & Mattis, D. Two soluble models of an antiferromagnetic chain. Ann. Phys. 16, 407–466 (1961).
Oshikawa, M. Commensurability, excitation gap, and topology in quantum manyparticle systems on a periodic lattice. Phys. Rev. Lett. 84, 1535–1538 (2000).
Paramekanti, A. & Vishwanath, A. Extending Luttinger’s theorem to Z_{2} fractionalized phases of matter. Phys. Rev. B 70, 245118 (2004).
Hastings, M. B. Liebschultzmattis in higher dimensions. Phys. Rev. B 69, 104431 (2004).
Hastings, M. B. Sufficient conditions for topological order in insulators. Europhys. Lett. (EPL) 70, 824–830 (2005).
Zaletel, M. P. & Vishwanath, A. Constraints on topological order in mott insulators. Phys. Rev. Lett. 114, 077201 (2015).
Cheng, M., Zaletel, M., Barkeshli, M., Vishwanath, A. & Bonderson, P. Translational symmetry and microscopic constraints on symmetryenriched topological phases: a view from the surface. Phys. Rev. X 6, 041068 (2016).
Qi, Y., Fang, C. & Fu, L. Ground state degeneracy in quantum spin systems protected by crystal symmetries. Preprint at http://arxiv.org/abs/1705.09190 (2017).
Yang, X., Jiang, S., Vishwanath, A. & Ran, Y. Dyonic LiebSchultzMattis theorem and symmetry protected topological phases in decorated dimer models. Phys. Rev. B 98, 125120 (2018).
Wu, J., Ho, T.L. & Lu, Y.M., Symmetryenforced quantum spin hall insulators in πflux models. Preprint at http://arxiv.org/abs/1703.04776 (2017).
Lu, Y.M., Ran, Y. & Oshikawa, M. Fillingenforced constraint on the quantized hall conductivity on a periodic lattice. Ann. Phys. 413, 168060 (2020).
Fang, C. & Fu, L. New classes of topological crystalline insulators having surface rotation anomaly. Sci. Adv. 5, eaat2374 (2019).
Else, D. V. & Thorngren, R. Topological theory of LiebSchultzMattis theorems in quantum spin systems. Phys. Rev. B 101, 224437 (2020).
Rasmussen, A. & Lu, Y.M. Classification and construction of higherorder symmetryprotected topological phases of interacting bosons. Phys. Rev. B 101, 085137 (2020).
Shiozaki, K., Xiong, C. Z. & Gomi, K. Generalized homology and atiyahhirzebruch spectral sequence in crystalline symmetry protected topological phenomena. Preprint at http://arxiv.org/abs/1810.00801 (2018).
Else, D. V. & Thorngren, R. Crystalline topological phases as defect networks. Phys. Rev. B 99, 115116 (2019).
Levin, M. & Gu, Z.C. Braiding statistics approach to symmetryprotected topological phases. Phys. Rev. B 86, 115109 (2012).
Lapa, M. F., Teo, J. C. Y. & Hughes, T. L. Interactionenabled topological crystalline phases. Phys. Rev. B 93, 115131 (2016).
Fidkowski, L. & Kitaev, A. Effects of interactions on the topological classification of free fermion systems. Phys. Rev. B 81, 134509 (2010).
Fidkowski, L. & Kitaev, A. Topological phases of fermions in one dimension. Phys. Rev. B 83, 075103 (2011).
Turzillo, A. & You, M. Fermionic matrix product states and onedimensional shortrange entangled phases with antiunitary symmetries. Phys. Rev. B 99, 035103 (2019).
Cheng, M. & Wang, C. Rotation symmetryprotected topological phases of fermions. Preprint at http://arxiv.org/abs/1810.12308 (2018).
Acknowledgements
Y.Q. is grateful to ZhengCheng Gu, ZhengXin Liu, Shenghai Jiang and Ying Ran for invaluable discussions. Y.Q. also thanks Aspen Center for Physics for hospitality, where part of this work was performed. S.Z.D. and C.F. acknowledge support from Minstry of Science and Technology of China under grant numbers 2016YFA0302400, 2016YFA0300600, from National Science Foundation of China under grant number 11674370, and from Chinese Academy of Sciences under grant number XXH13506202. Y.Q. acknowledges support from Minstry of Science and Technology of China under grant numbers 2015CB921700, and from National Science Foundation of China under grant number 11874115.
Author information
Authors and Affiliations
Contributions
C.F. and Y.Q. conceived the project and developed the theoretical ideas. Y.Q. derived the spectralsequence formulas. Z.S. implemented the algorithm and obtained all classification. All authors contributed to the writing of the paper.
Corresponding authors
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Peer review information Nature Communications thanks Zhengcheng Gu, Zhen Bi and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary information
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Song, Z., Fang, C. & Qi, Y. Realspace recipes for general topological crystalline states. Nat Commun 11, 4197 (2020). https://doi.org/10.1038/s41467020176855
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41467020176855
This article is cited by

Phonons as a platform for nonAbelian braiding and its manifestation in layered silicates
Nature Communications (2022)
Comments
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.