Proof of bulk-edge correspondence for band topology by Toeplitz algebra

Zixian Zhou and Liang-Liang Wan∗

Shenzhen Key Laboratory of Ultraintense Laser and Advanced Material Technology,
Center for Intense Laser Application Technology, and College of Engineering Physics,
Shenzhen Technology University, Shenzhen 518118, China

We rigorously yet concisely prove the bulk-edge correspondence for general d-dimensional (dD)
topological insulators in complex Altland-Zirnbauer classes, which states that the bulk topological
number equals to the edge-mode index. Specifically, an essential formula is discovered that links
the quantity expressed by Toeplitz algebra, i.e., hopping terms on the lattice with an edge, to the
arXiv:2410.19539v1 [cond-mat.mes-hall] 25 Oct 2024

Fourier series on the bulk Brillouin zone. We then apply it to chiral models and utilize exterior
differential calculations, instead of the sophisticated K -theory, to show that the winding number of
bulk system equals to the Fredholm index of 1D edge Hamiltonian, or to the sum of edge winding
numbers for higher odd dimensions. Moreover, this result is inherited to the even-dimensional Chern
insulators as each of them can be mapped to an odd-dimensional chiral model. It is revealed that
the Chern number of bulk system is identical to the spectral flow of 2D edge Hamiltonian, or to the
negative sum of edge Chern numbers for higher even dimensions. Our methods and conclusions are
friendly to physicists and could be easily extended to other physical scenarios.

∗ [email protected]
 2

I. INTRODUCTION

Topology plays an increasingly important role in characterizing states of matter, especially the lattice band sys-
tems. Apart from the well known fermionic topological insulators and superconductors in Altland-Zirnbauer (AZ)
classification [1–3], topological bands are also found and classified in quadratic-bosonic [4–6] and even non-Hermitian
systems [7, 8]. In all these systems, there is an iconic phenomenon called bulk-edge correspondence (BEC) [9, 10],
that is, when the lattice has an edge, edge states would occur within the band gap of the bulk system and exhibit a
topological behavior related to the bulk topology but irrespective to other physical details. This striking phenomenon
not only implicates a profound index theorem in mathematical physics, but also associates with many practical ef-
fects like quantized Hall conductivity [11] and fractional statistics of Majorana edge modes [12]. Therefore, a clear
understanding of BEC becomes a fundamental problem.
Although the BEC in free-fermion systems has been studied by twisted equivariant K -theory [13, 14], it is too
sophisticated for physicists and hard to extend to other physical systems. Besides, physical community has also paid
many efforts to achieve concise proofs to various degrees of rigor and generality. The physical approaches include
Green’s functions [15], Levinson’s theorem in scattering theory [16] and Atiyah-Singer index theorem [17]. Nevertheless,
all these prior works have made approximations and have somewhat limitations. For instance, the approach of Green’s
function discards higher order expansion terms which are not infinitesimal; the approach of long-wave limit reduces
the lattice Hamiltonian to a differential operator on a continuum system, which alters the essence of lattice. Thus,
the key problem still exists: Is there a rigorous yet concise approach to prove the BEC for fermionic band systems,
which can be easily transferred to other physical scenarios like bosonic and non-Hermitian systems?
This work answers yes to the problem. We use Toeplitz algebra and exterior differentiation instead of the sophis-
ticated K -theory to rigorously yet concisely prove the BEC for general dimensional topological insulators in classes
A and AIII according to the AZ classification [3]. Specifically, the tight-binding models on the lattice with an edge
are described by a polynomial of Toeplitz matrix that represents the hopping term of particles on the half infinite
lattice. Via replacing the Toeplitz matrices by plane wave functions, the edge Hamiltonian (i.e., the Hamiltonian on
the lattice with an edge) simply reduces to the Bloch Hamiltonian of the bulk. This is known as a homomorphism
between the Toeplitz algebra generated by Toeplitz matrices and Fourier series. Neither the K -theory approach uti-
lizing exact sequences nor the former physical approaches avoiding Toeplitz algebra, here we discover a simple and
powerful formula telling that the trace of commutator of Toeplitz algebra equals to an integral of Fourier series. This
implies that a proper quantity formulated by edge Hamiltonian can be converted to that by bulk one. Therefore,
our strategy seeks to define edge-mode indices explicitly by edge Hamiltonians and also write the bulk topological
numbers by bulk Hamiltonians. With this task done, the equivalence of bulk and edge topological quantities can be
easily proved by algebraic and calculus computations. In this way, a clear understanding of the BEC is provided for
fermionic insulators, which is transferable to other physical systems.
Based on the above strategy, we firstly investigate the odd dimensional chiral models in class AIII, for their bulk
topology is characterized by a winding number which already has an explicit expression by bulk Hamiltonian. Besides,
the topology of edge modes is characterized by a Fredholm index in one dimension which is equivalent to the partition
function of (square) edge Hamiltonian in zero temperature. In higher odd dimensions, we successfully define an edge
winding number which can still be formulated by the edge Hamiltonian. Thus, our strategy is implemented. By
applying our formula, we are able to prove that the bulk winding number equals to the edge-mode index, that is, to
prove the BEC.
Next, we turn to prove the BEC for the even dimensional Chern insulators in class A whose bulk topology is
characterized by a Chern number formulated by Berry curvature rather than Hamiltonian itself. Fortunately, each
d-dimensional (dD) Chern insulator can be mapped to a (d + 1)D chiral model by a simple construction. It turns out
that the Chern number of the Chern insulator is identical to the winding number of the constructed chiral model up
to a sign. The edge-mode index of the Chern insulator is also induced by that of the chiral model. Therefore, the
BEC of Chern insulators is inherited from the former result of chiral models. We thus concisely prove the BEC for
all the complex AZ classes in any dimension.
Our article is organized as follows. In Sec. II, the edge and bulk Hamiltonians are given. In Sec. III, the properties
of Toeplitz algebra and the formula are revealed. In Sec. IV, the BEC for odd dimensional chiral models is proved.
Subsequently, in Sec. V, the BEC for even dimensional Chern insulators is proved. In Sec. VI, conclusions are made.

II. INSULATORS WITH AND WITHOUT BOUNDARY

We study the dD fermionic insulator described by a tight-binding model on an infinite crystal lattice with N internal
T
degrees of freedom (e.g., spins) in each unit cell. By denoting ĉr = (ĉr1 , · · · , ĉrN ) the column vector of fermionic
 3

annihilation operators at lattice site r = (r1 , . . . , rd ), we write down the many-body Hamiltonian
X
Ĥ = ĉ†r hr−r′ ĉr′ , (1)
r,r ′

where hr−r′ ∈ MN (C) is an N × N matrix and describes the hopping of particle from site r ′ to r. These hopping
terms are finite in physical models.
We firstly consider the half infinite lattice structured by N × Zd−1 which has an edge (r1 ∈ N) along the r1 -direction
andhas noP boundaries along the remaining dimensions (r̃ = (r2 , · · · , rd ) ∈ Zd−1 ). After the Fourier transformations
ĉr1 k̃ = r̃ ĉr e−ik̃·r̃ with crystal momenta k̃ = (k2 , . . . , kd ) ∈ T d−1 (ki ∈ [0, 2π)), the degrees of freedom k̃ are
decoupled in the Hamiltonian
ˆ
dk2 · · · dkd X †      
Ĥ = d−1
ĉ r k̃ hr−r ′ k̃ ĉr′ k̃ (2)
(2π) r,r ′ ∈N

 
−ik̃·r̃
. Here the associated Brillouin zone (BZ) is a (d − 1)D torus T d−1 . Now the trans-
P
with hr1 k̃ = r̃ hr e
 
formed Hamiltonian matrix Hrr′ = hr−r′ k̃ describes an effective 1D tight-binding Hamiltonian model on lattice
N, parameterized by k̃. With the help of Toeplitz matrix
 
0
1 0 
T = 1 0  , Ti,j = δi−1,j , (3)
 
 
.. ..
. .

this Hamiltonian H = (Hrr′ ) is expressed as

  X  
r X  
H k̃ = h−r k̃ ⊗ T † + hr k̃ ⊗ T r . (4)
r≥0 r>0

Here T and T † describe the nearest-neighbor hopping of particles towards the infinity and the edge side, respectively.
They satisfy identity

T †T = I (5)
 
with identity operator I but T T † ̸= I. In the follows, we are interested in the bound eigenstates of H k̃ which
 
are known as the edge modes. Hence we restrict H k̃ to the Hilbert space CN ⊗ ℓ2 (N) with the scattering states
 
excluded. Then H k̃ becomes an effective Hamiltonian for the edge degrees of freedom, dubbed as edge Hamiltonian.
 
Since the hopping terms are finite, H k̃ is a bounded self-adjoint operator and is smooth regarding with k̃.
We next consider the lattice structured by Zd without boundaries, which is equivalent to the periodic boundary
condition with infinite crystal volume. In this case, full-dimension Fourier transformations ĉ (k) = r ĉr e−ik·r can
P
be made with k = (k1 , . . . , kd ) ∈ T d (ki ∈ [0, 2π)), and all the degrees of freedom are decoupled in the Hamiltonian
ˆ
dk1 · · · dkd †
Ĥ = d
ĉ (k) h (k) ĉ (k) , (6)
(2π)

in which the bulk Hamiltonian (i.e., Bloch Hamiltonian) is given by

X
h (k) = hr e−ik·r . (7)
r

The Hermiticity of h (k) stems from the many-body Hamiltonian Ĥ. Here we assume that the bands are gapped at
zero energy, i.e., det h (k) ̸= 0 for all k. This restriction is natural for the fermionic insulators and paves the road to
define topological invariants for bulk Hamiltonians.
 4

III. TOEPLITZ ALGEBRA

The tight-binding model on the half infinite lattice, represented by Toeplitz matrices, does not have the complete
crystal translation symmetry. As a result, these degrees of freedom cannot be decoupled by Fourier transformation,
and many quantities like eigen-energies are difficult to compute. Nevertheless, there is still a homomorphism relation
between Toeplitz matrices and Fourier series. Through this relation, some special quantities like topological invariant
can be cleverly computed.
To be specific, T generates a C ∗ -algebra (by addition, multiplication, C-number multiplication, and Hermitian
conjugation) called Toeplitz algebra T [18]. Each element X ∈ T, as an operator on the Hilbert space ℓ2 (N), is
required to be bounded. Due to the identity Eq. (5), each X ∈ T can be uniquely expanded as
X
n
X= xmn T m T † , xmn ∈ C. (8)
m,n∈N

On the other hand, the bounded functions on the 1D BZ S 1 also form a C ∗ -algebra L∞ S 1 = −ink

P
1 ∗ ∞
n∈Z
1

fn e in
which {fn } ∈ ℓ (Z) are Fourier coefficients. There is a C -algebra homomorphism φ : T → L S given by the
replacement T 7→ e−ik , i.e.,
 
X n X
xmn T m T †  = xmn e−imk eink .


φ (9)
m,n∈N m,n∈N

One can readily check the following homomorphism axioms (for X, Y ∈ T and µ, ν ∈ C)
φ (µX + νY ) = µφ (X) + νφ (Y ) , (10)
φ (XY ) = φ (X) φ (Y ) , (11)
†
φ X † = [φ (X)] ,


(12)
†
and see [φ (T )] φ (T ) = 1 consistent
with the identity Eq. (5). After extending the algebraic homomorphism to
φ : MN (C) ⊗ T → MN (C) ⊗ L∞ S 1 and let k = k1 , one immediately finds that φ maps the edge Hamiltonian Eq.
(4) to the bulk one Eq. (7), i.e.,
  
φ H k̃ = h (k) . (13)

Based on this homomorphism, we put forward an important formula that links the quantity expressed by Toeplitz
matrices to that by Fourier series: For X, Y ∈ MN (C) ⊗ T, there is
ˆ
i
Tr [X, Y ] = tr φ (X) dφ (Y ) , (14)
2π S1

in which Tr traces over the Hilbert

space and tr traces over finite dimensional space. This implies that Tr [X, Y ] is
finite if φ (X) , φ (Y ) ∈ C 1 S 1 .
The proof is quite simple. Due to the bi-linearity, we just need to verify the basis of T satisfying Eq. (14), i.e.,
ˆ
h
s
n i i
Tr T t T † , T m T † = e−i(t−s)k de−i(m−n)k . (15)
2π S 1
By straightforward calculations, we know that its left-handed side equals to (m − n) δ−t+s,m−n which coincides with
its right-handed side. Thus, the proof has been finished. Note that Tr (XY ) diverges in general case such that the
cyclic property of trace does not hold and Tr [X, Y ] could be finite. This formula allows us to establish the BEC in
a concise approach without the sophisticated K -theory. More explicitly, if the left-handed side of Eq. (14) depicts
a topological index expressed by the edge Hamiltonian, the right-handed side corresponds to a topological invariant
expressed by the bulk one.  
It is worth mentioning that, to define the topological index by H k̃ , one would utilize the inverse operator of
Toeplitz algebra. In Appendix A we show X −1 ∈ MN (C) ⊗ T if X ∈ MN (C) ⊗ T and φ (X) ∈ MN (C) are both
invertible. According to Eq. (11), we obtain the property
−1
φ X −1 = [φ (X)] .


(16)
 
Remarkably, an invertible φ (X) does not ensure X invertible. Thus, the edge Hamiltonian H k̃ could be singular
despite the invertible bulk Hamiltonian h (k) which obeys the gapped condition det h (k) ̸= 0.
 5

IV. BEC FOR ODD DIMENSIONAL CHIRAL MODELS

Now let us consider the (2n + 1)D (n ≥ 0) chiral model which belongs to class AIII in the AZ classification [3]. In
this case, the bulk topology is characterized by a winding number with an explicit expression. After defining the edge-
mode index by the edge Hamiltonian, we will prove the equivalence of these two integers without any approximation.
The bulk Hamiltonian with chiral symmetry respects a constraint

Sh (k) S −1 = −h (k) , (17)

with chiral matrix S = IN/2 ⊕ −IN/2 (Im denotes the identity matrix in Mm (C) and N is an even number) such
that it must take the off-diagonal form
 
0 q (k)
h (k) = . (18)
q † (k) 0

Due to finite hopping terms, q (k) is a smooth matrix-valued function. The gapped condition requires det q (k) ̸= 0
such that q is a smooth mapping from T 2n+1 to the general linear group GLN/2 (C) ≃ U (N/2) (≃ denoting the
homotopy equivalence). One can define the degree of mapping as follows
ˆ

2n+1
deg = Cn tr q −1 dq ∈ Z, (19)
T 2n+1

with the coefficient

 n+1
n! i
Cn = − . (20)
(2n + 1)! 2π

Here d is the exterior differentiation and the power of q −1 dq refers to the power of exterior product. The degree of
mapping is also called as the winding number.
Like the bulk Hamiltonian, the edge Hamiltonian also takes the off-diagonal form
  
  Q k̃
H k̃ =    , (21)
Q† k̃
    
with homomorphism φ Q k̃ = q (k). Despite the invertible bulk Hamiltonian, the edge Hamiltonian H k̃ could
be singular and thus have zero modes that are nontrivial bound states ψ satisfying Hψ = 0. These zero modes would
exhibit a topological property reflected by an edge-mode index. This index has distinct definitions in one and higher
dimensions, and hence we discuss them separately and prove the BEC respectively.

A. 1D case

Due to the chirality, the zero-mode space is decomposed to a negative-chirality subspace and a positive-chirality
one, i.e., ker H = ker Q ⊕ ker Q† (ker denotes the kernel). And due to invertible φ (Q) = q, Q is a Fredholm operator
with finite dimensional ker Q and ker Q† [19]. Therefore, the following Fredholm index is well defined

ind = dim (ker Q) − dim ker Q† ∈ Z

(22)

and it serves as the edge-mode index for 1D chiral models. In a physical sense, its explicit expression is given by a
couple of partition functions in zero temperature [20]
 † †

ind = Tr lim e−βQ Q
− e−βQQ . (23)
β→+∞

⊥
The strict deduction resorts to the fact that Fredholm operator Q restricted to (ker Q) (⊥ the orthogonal complement)
†
has a lower bound such that e−βQ Q → 0, and the same for Q† .
 6

Now we prove the BEC that says the bulk winding number equaling to the Fredholm index, i.e., deg = ind. This
result is known as the Toeplitz index theorem [21]. Firstly, we recast the edge-mode index Eq. (23) as a trace of
commutator
" ∞ #
X (−β)m
m−1 †
†
ind = Tr lim Q Q Q ,Q . (24)
β→+∞
m=1
m!

According to the formula Eq. (14) and properties Eqs. (10–12), it reduces to
ˆ ∞ m
i X (−β)
m−1 †
ind = lim tr q† q q dq (25)
2π β→+∞ S 1 m=1 m!
ˆ †
i e−βq q − IN/2 †
= lim tr q dq, (26)
2π β→+∞ S 1 q† q
where φ (Q) = q (k) has been used in the above. The gapped condition det q ̸= 0 guarantees the positive definiteness
†
of q † q such that limβ→+∞ e−βq q = 0. Therefore, we arrive at
ˆ
1
ind = trq −1 dq = deg . (27)
2πi S 1
Thus, the BEC for the 1D case has been quickly proved.

B. (2n + 1)D cases for n ≥ 1

The occurrence of zero modes in higher dimensions becomes a complicated issue. First of all, we exclude a trivial
case that H(k̃) has zero mode in the entire BZ T 2n . In this trivial case, the unique topological index is the Fredholm
index
ˆ
† 1 ∂q
trq −1


dim (ker Q) − dim ker Q = dk1 , (28)
2πi S 1 ∂k1

and the BEC simply reduces to the 1D case. In fact, the Fredholm index preserves as k̃ varies, for the right-handed
side of the above equation is a homotopy invariant under the gap condition det q ̸= 0. Hence, a nonzero Fredholm
index necessarily results in a global zero mode. In the nontrivial case that we investigate, there

is no global zero mode
such that the Fredholm index must be vanishing, which indicates dim (ker Q) = dim ker Q† for all k̃.
Since H(k̃) is smooth regarding with k̃, each of its zero mode forms a smooth vector field ψ(k̃) whose domain is
a smooth submanifold or a single point in BZ T 2n . The smooth submanifold should not have an edge and is hence
closed. Moreover, the dimensions of these closed submanifolds must be lower than 2n, for a (2n)D submanifold in
T 2n (but not equaling T 2n ) necessarily has a boundary and is not closed. Due to probable essential degeneracy,
some zero modes may share the same domain Di . Different zero-mode domains may overlap with each other and
cause accidental degeneracy,S nevertheless, we can treat them separately. Eventually, the entire domain of zero modes
D = {p̃| ker H (p̃) ̸= 0} = i Di consists of several lower-dimensional closed submanifolds (or discrete points) Di (see
Fig. 1), with each Di corresponding to a fixed degree of degeneracy.
Now we adopt a non-singular hypothesis: the derivative of zero-mode spectra along the normal directions of Di
is always nonzero. This implies the edge-mode spectra crossing zero energy directly along normal directions such
that the zero mode is substantial. In physical language, the zero mode has a linear dispersion with finite group
velocity. Physical systems (i.e., topological insulators) usually meet with this condition. Evidently, this hypothesis
automatically ensures the absence of global zero mode and hence a vanishing Fredholm index.
Then, we are able to define the edge-mode
S index for (2n + 1)D chiral models (n ≥ 1). To be specific, we construct
a (2n)D neighborhood B (Di ; ρ) = p̃∈Di B (p̃; ρ) for each Di , in which B (p̃; ρ) denotes the ball with center p̃ and
radius ρ → 0 (see Fig. 1). On the boundary ∂B (Di ; ρ), we define a finite-size matrix
  1  
gi k̃ = Ψ†+ Q k̃ Ψ− , (29)
ρ
in which Ψ+ = (ψ1 , . . . , ψs ) and Ψ− = (ψs+1 , . . . , ψ2s ) is the orthonormal basis of ker Q† (p̃) and ker Q (p̃) (their
identical dimensions stemmed from the vanishing Fredholm index), respectively. Here p̃ ∈ Di is the nearest zero-
mode point to k̃. According to the perturbation theory, the derivative of zero-mode spectra along the normal direction
 7

Figure 1. Zero-mode domain of the edge Hamiltonian for (2n + 1)D chiral models (n ≥ 1). The domain of zero modes (red) in
BZ T 2n consists of smooth closed submanifolds (or discrete points) Di contained in a neighborhood B (Di ; ρ) (blue), with ρ the
distance between the boundary of neighborhood ∂B (Di ; ρ) (green) and Di . Here ∂B (Di ; ρ) is a (2n − 1)D closed manifold.

k̃ − p̃ is given by the eigenvalues of following matrix

    
 †  H k̃ − H (p̃)
Ψ+
gi k̃
lim † Ψ+ Ψ− = lim  †   . (30)
ρ→0 Ψ− ρ ρ→0 gi k̃
   
The non-singular hypothesis ensures det gi k̃ ̸= 0. Therefore, gi k̃ is a smooth mapping from the (2n − 1)D closed
manifold ∂B (Di ; ρ) to GLs (C) ≃ U (s), with degree of mapping
ˆ

2n−1
degi = Cn−1 tr gi−1 dgi ∈ Z. (31)
∂B(Di ;ρ)

Note that degi is invariant under gauge transformations Ψ± 7→ Ψ± U± (U± ∈ U (s)). We call degi as the edge winding
number. Finally, the overall edge-mode index is defined as a sum
X
ind = degi . (32)
i

With the edge-mode index defined, we would like to express it explicitly by the edge Hamiltonian. For simplicity,
we suppose no overlap between those Di . In Appendix B we show
   
Ψ− lim gi−1 k̃ Ψ†+ = lim ρQ−1 k̃ , (33)
ρ→0 ρ→0

by which the edge-mode index can be recast as

ˆ
X
2n−1
ind = Cn−1 Tr lim Q−1 dQ . (34)
ρ→0 ∂B(Di ;ρ)
i

It is worthy to emphasize that Tr and lim cannot interchange, for only the limit of operator is finite-rank and hence
trace-class. Then we apply Stokes’ formula and achieve
ˆ

2n−1
ind = −Cn−1 Tr lim d Q−1 dQ . (35)
ρ→0
S
T 2n \ B(Di ;ρ)
ˆ i

2n
= Cn−1 Tr Q−1 dQ , (36)
T 2n \D

Here Q−1 (k̃) exists for k̃ ∈ T 2n \D, where H(k̃) has no zero mode, for ker Q = 0 means Q injective and ker Q† = 0
means Q surjective. Actually, more careful analysis shows that Eq. (36) holds even if those Di overlap with each
other (see Fig. 2). Remarkably, the integrand belongs to MN (C) ⊗ T.
 8

´
2n
Figure 2. Treatment of Eq. (36) for overlapping Di . In this case, the Stokes’ formula gives Cn−1 Tr T 2n \D Q−1 dQ =
P ´ −1

2n−1 −1

2n−1 −1

2n−1
i Cn−1 Tr Xi Q dQ . Due to Eq. (33), Tr Q dQ = tr gi dgi for the most area of Xi . Besides, Xi
P ´ −1

2n−1
covers the most area of ∂B (Di ; ρ). Therefore, under the limit ρ → 0, i Cn−1 Tr Xi Q dQ could be replaced by
P ´ −1

2n−1 P
i C n−1 ∂B(D ;ρ)
tr gi dg i = i deg i = ind, which means that Eq. (36) still holds.
i

Now we are ready to prove the BEC that states deg = ind. We still recast the index Eq. (36) as a trace of
commutator (commutator of Toeplitz algebra is trace-class) and apply the formula Eq. (14), yielding
ˆ 2n+1
X 1 h −1
2n−1 −1 i
ind = Cn−1 Tr Q dQ , Q ∂i Q ∧ dki (37)
T 2n \D i=2 2
ˆ ˆ
i h
2n−1 i
trφ Q−1 dQ ∧ ∂1 φ Q−1 dQ .


= Cn−1 dk1 (38)
4π S1 T 2n \D

Then we apply properties Eqs. (10–12) and Eq. (16), reducing the above equation to
ˆ
i
2n−1
dk1 ∧ tr q −1 dq ∧ ∂1 q −1 dq ,


ind = Cn−1 (39)
4π T 2n+1

where φ (Q) = q has been substituted into the equation. Here the integration area S 1 × T 2n \D has been replaced
by S 1 × T 2n = T 2n+1 because D has a vanishing (2n)D volume and q −1 (k) no longer has singularity in the whole BZ
T 2n+1 (due to gapped condition det q (k) ̸= 0).
To see the final result, we continue the calculation and obtain
ˆ
i
2n−1
dk1 ∧ tr q −1 dq ∧ −q −1 ∂1 q · q −1 dq


ind = Cn−1
4π T 2n+1
h
2n−1 −1 i
+d dk1 ∧ tr q −1 dq q · ∂1 q . (40)
h
2n−1 −1 i
Here identity d q −1 dq q = 0 has been used. The second term is vanishing since it integrates an exact form
over a closed manifold. Finally, the edge-mode index reduces to
ˆ
i
2n
ind = Cn−1 trq −1 dk1 ∂1 q ∧ q −1 dq (41)
4π
ˆ
T 2n+1

i 1
2n+1
= Cn−1 tr q −1 dq (42)
4π 2n + 1 T 2n+1
ˆ

2n+1
= Cn tr q −1 dq = deg . (43)
T 2n+1

It is seen that the BEC for (2n + 1)D cases has been proved. It is worthy to mention that the conclusion still holds
if parameter k̃ belongs to other closed manifold.
At last, we completely state our conclusions as follows.  
q
Proposition 1: Suppose a (2n + 1)D chiral model h (k) = with a smooth q (k) satisfying the gapped
q†
 9
 
Q
condition det q (k) ̸= 0. And its corresponding edge Hamiltonian H(k̃) = satisfies the non-singular
Q†
hypothesis (the normal derivative of zero-mode spectra at the zero-mode domain is always nonzero) if n ≥ 1. Then
ˆ
1
trq −1 dq = dim (ker Q) − dim ker Q† , n = 0,


(44)
2πi S 1

 n+1 ˆ
n! i
2n+1 X
− tr q −1 dq = degi , n ≥ 1, (45)
(2n + 1)! 2π T 2n+1 i

in which degi is defined in Eq. (31).

V. BEC FOR EVEN DIMENSIONAL CHERN INSULATORS

In this section we consider the (2n)D (n ≥ 1) Chern insulators without any internal symmetry, thus belonging
to class A in the AZ classification. In this case, the bulk topology of Hamiltonians is characterized by the Chern
number of the valence (negative-energy) bands which is formulated by Berry curvature (see Appendix C). To apply
our formula, we need to explicitly write the Chern number by bulk Hamiltonian. This could be done by a simple
construction.
Actually, every (2n)D Chern insulator Hamiltonian h (k) can be mapped to a (2n + 1)D chiral model in form of
Eq. (18) via construction

q (k, ω) = h (k) − iωIN (46)

with ω ∈ R ∪ {∞} ≃ S 1 . The positive definiteness of q † q guarantees the gapped condition det q (k, ω) ̸= 0 such that
the degree of mapping q is well defined, i.e.,
ˆ

2n+1
deg = Cn tr q −1 dq . (47)
T 2n ×S 1

In Appendix C, we show that the Chern number Ch of the original Hamiltonian is given by
n
Ch = (−1) deg . (48)

In this way, the Chern number has an explicit

 expression regarding with the bulk Hamiltonian h (k).
Similarly, the edge Hamiltonian H k̃ can also be mapped to the chiral model Eq. (21) with
   
Q k̃, ω = H k̃ − iωI, (49)

and the definition of the edge-mode index for the Chern insulator can be induced by that of the chiral model. Like the
previous treatment, we still adopt the non-singular hypothesis that the spectra crosses zero energy substantially, which
automatically ensures the absence of global zero mode. Then the zero-mode domain D = {p̃| ker H (p̃) ̸= 0} ⊂ T 2n−1
consists of several lower-dimensional closed manifolds (or discrete points) Di , with each Di corresponding to a fixed
degree of degeneracy dim (ker H).
Correspondingly, the zero-mode domain of Q(k̃, ω) consists of Di′ = (Di ,0) (abbreviation of {(p̃, 0) |p̃ ∈ Di }).
Each Di′ has a (2n)D neighborhood B ′ (Di′ ; ρ) = (B (Di ; ρ cos θ) , ρ sin θ) |θ ∈ − π2 , π2

(see Fig. 3). Each point on
∂B ′ (Di′ ; ρ) can be expressed as (p̃ + (k̃ − p̃) cos θ, ρ sin θ) with a certain k̃ ∈ ∂B (Di ; ρ) and the nearest zero-mode
point p̃ ∈ Di . Then the finite-size matrix gi defined by Eq. (29) reads (up to the main order of ρ)

1 † h  i
gi = Ψ k̃ − p̃ cos θ · ∇H (p̃) Ψ − iIs sin θ (50)
ρ
1  
= Ψ† H k̃ Ψ cos θ − iIs sin θ, (51)
ρ
 10

Figure 3. Zero-mode domains of the edge Hamiltonian for (2n)D Chern insulators and the constructed (2n + 1)D chiral
models. The zero-mode domain of H(k̃) in BZ T 2n−1 consists of smooth closed manifolds Di (red) contained in a (2n − 1)D
neighborhood B (Di ; ρ) (dark blue). Correspondingly the zero-mode domain of Q(k̃, ω) = H(k̃) − iωI in T 2n−1 × S 1 consists of
Di′ = (Di , 0) contained in a (2n)D neighborhood B ′ (Di′ ; ρ) (light blue). Here ∂B (Di ; ρ) and ∂B ′ (Di ; ρ) are a (2n − 2)D and
a (2n − 1)D closed manifold, respectively.

in which Ψ = (ψ1 , · · · , ψs ) is the orthonormal basis of ker H (p̃). Therefore, the degree of mapping g : ∂B ′ (Di′ ; ρ) →
GLs (C) ≃ U (s) is given by
ˆ

2n−1
degi = Cn−1 tr gi−1 dgi (52)
∂B ′ (Di′ ;ρ)
ˆ  2n−1
1
= Cn−1 tr d (Gi − iνIs ) , (53)
∂B(Di ;ρ)×R Gi − iνIs

with ν = tan θ ∈ R, and the reduced Hamiltonian on ∂B (Di ; ρ) is defined by

  1  
Gi k̃ = Ψ† H k̃ Ψ. (54)
ρ

The non-singular hypothesis implies det Gi (k̃) ̸= 0, i.e., the reduced Hamiltonian is still gapped.
Now the edge-mode index of the Chern insulator can be induced. For n ≥ 2, we compare Eq. (53) with Eqs.
(47–48) and conclude that
n−1
degi = (−1) Chi , n ≥ 2,

in which Chi is the Chern number for the “valence bands” of reduced Hamiltonian Gi (k̃) over (2n − 2)D closed
manifold ∂B (Di ; ρ), dubbed as edge Chern number. For n = 1, the parameter k̃ is 1D and the zero-mode domain Di
must be a single point p̃. Then ∂B (Di ; ρ) = {p̃ + ρ} − {p̃ − ρ} ≃ S 0 ({p̃ + ρ} and − {p̃ − ρ} refer to 0-chain whose
coefficients reflect the orientations) such that degi reduces to

ˆ
 p̃+ρ
+∞
1 −idν
degi = tr     (55)
2πi −∞ Gi k̃ − iνI
k̃=p̃−ρ
= −tr sgn [Gi (p̃ + ρ) − Gi (p̃ − ρ)] (56)
= −tr sgnG′i (p̃) . (57)

Here tr sgnG′i (p̃) (the prime denotes the derivative) is the spectral flow of H (k2 ) at k2 = p̃, which counts
P the sign
flipping of edge-mode spectra when k 2 passes through p̃. The negative sum of edge Chern number − i Chi or the
total spectral flow i tr sgnG′i serves as the edge-mode index.
P
Finally, we recall the BEC for the chiral models and achieve
n
X
(−1) Ch = degi , (58)
i
 11

which can be recast as

X
Ch = − Chi , n ≥ 2, (59)
i
X
Ch = tr sgnG′i , n = 1. (60)
i

These two equations reflect the BEC of Chern insulator. Remarkably, the last equation reproduces the well known
conclusion of the 2D Chern insulator, that its Chern number counts the overall sign flipping of the edge-mode energy.
The complete conclusions are stated as follows.
Proposition 2: Suppose a (2n)D Chern insulator described by a smooth Hamiltonian h (k) that satisfies gapped
condition det h (k) ̸= 0. And its corresponding edge Hamiltonian H(k̃) satisfies the non-singular hypothesis. Then
one can define several (2n − 2)D gapped Hamiltonians Gi (k̃) near the zero-mode domains according to Eq. (54) such
that
X
Ch = tr sgnG′i , n = 1, (61)
i
X
Ch = − Chi , n ≥ 2, (62)
i

in which Ch and Chi is the Chern number of the valence bands for h (k) and Gi (k̃), respectively.

VI. CONCLUSIONS

In conclusion, the BEC for the complex AZ classes in any dimension has been completely proved. The key step of
the proof is the formula of Toeplitz algebra that links the quantity expressed by edge Hamiltonian to that by bulk one.
It enables us to rigorously prove the BEC by calculus computations after the bulk and edge topological quantities are
explicitly expressed by Hamiltonians.
For odd-dimensional chiral models, the bulk winding number already has an explicit expression. In the 1D case,
the edge-mode index is the Fredholm index of the edge Hamiltonian. With the help of our formula, the BEC, that the
winding number equals to the Fredhom index, has been quickly proved. In higher dimensional cases, the edge-mode
index has been defined as the sum of edge winding numbers, and it has also been proved to be identical to the bulk
winding number.
For even-dimensional Chern insulators, after they are mapped to the odd-dimensional chiral models, the bulk Chern
number and edge-mode index are both induced by the topological quantities of the constructed chiral model. Hence,
the BEC of the Chern insulators are inherited from that of the chiral models. It turns out that the bulk Chern number
equals to the spectral flow (for 2D case) or the negative sum of edge Chern numbers (for higher-even dimensions).
Our work provides a clear understanding to the BEC without the sophisticated K -theory and can probably be
extended to other physical scenarios like the BEC for Z2 topological insulators in real AZ classes or the BEC for
topological Bogoliubov excitation of bosons.

ACKNOWLEDGEMENTS

We thank Prof. Zhi-Fang Xu for his helpful discussions. This work is supported by the Natural Science Foundation of
Top Talent of SZTU (GDRC202202, GDRC202312), Guangdong Provincial Quantum Science Strategic Initiative (No.
GDZX2305006), and Characteristic Innovation Project of Guangdong Provincial Universities (No. 2024KTSCX053).

Appendix A: Inverse Operator of Toeplitz algebra

Here we prove X −1 ∈ MN (C) ⊗ T if X ∈ MN (C) ⊗ T and φ (X) ∈ MN (C) are both invertible. (Actually, an
invertible X necessarily leads to an invertible φ (X), however, it costs additional proof and we do not need such strong
conclusion here.)
n
m
T † with xmn ∈ MN (C), can be
P
Firstly, any element in MN (C) ⊗ T, expressed by X = m,n∈N xmn ⊗ T
decomposed as X = Tφ(X) + K. Here the Toeplitz operator Tf with continuous symbol f (k) = l∈Z fl e−ilk (fl ∈
P
 12

MN (C)) is defined as
X X
l
Tf = fl T l + f−l T † (A1)
l≥0 l>0

with property φ (Tf ) = f . And the operator

∞ X h i ∞ X h i
† n
X X
m
l
l n
xm,m+l ⊗ T m T † − 1 T †


K= xn+l,n ⊗ T T T −I + (A2)
l=0 n l=1 m

n
m
l
is a limit of finite-rank matrix and hence compact, for T l T n T † − I and T m T † − 1 T † are both finite-
   
rank matrices. Evidently, φ (K) = 0. This decomposition leads to a corollary: X ∈ MN (C) ⊗ T is compact if and
only if φ (X) = 0.
Next, the invertible X and φ (X) allow us to construct operator
O = X −1 − T[φ(X)]−1 . (A3)

Since XO = I − XT[φ(X)]−1 ∈ MN (C) ⊗ T, we are able to compute

 
φ (XO) = φ I − XT[φ(X)]−1 = 0, (A4)

which implies XO being compact. On the other hand, since X is a bounded invertible operator, its inverse X −1 is
also bounded. It turns out that O is compact due to the fact that a bounded operator multiplying a compact operator
produces a compact one, i.e., X −1 · XO = O.
In general, the compact operator O can be expanded by finite-rank matrices, i.e.,
X
O= omn ⊗ Jmn , omn ∈ MN (C) , (A5)
m,n∈N+

where Jmn ∈ M∞ (C) is the matrix whose entry at row m and column n is 1, with all the other entries vanishing.
Each Jmn can be expressed by Toeplitz algebra, i.e.,

m−1
m
Jmm = T m−1 T † − Tm T† ∈ T, m ∈ N+ , (A6)
l
Jm,m+l = Jmm T † ∈ T, l ∈ N+ ,


(A7)
Jm+l,m = T l Jmm ∈ T, l ∈ N+ . (A8)
Therefore, O ∈ MN (C) ⊗ T.
Eventually, the bounded operator X −1 = T[φ(X)]−1 + O has been expressed by MN (C) ⊗ T, i.e., X −1 ∈ MN (C) ⊗ T.

Appendix B: Q−1 near zero-mode point

 
We compute Q−1 k̃ in the limit ρ = k̃ − p̃ → 0 in which p̃ is a zero-mode point. Our main idea is to solve
h   i h   i
linear equation Q (p̃) ψ + Q k̃ − Q (p̃) ψ = ϕ by perturbation theory where Q k̃ − Q (p̃) ∝ ρ is infinitesimal.
The solution of ψ would be an infinite serieswhose
 convergence would be verified by functional analysis. The main
order term of ψ simply gives the limit of Q−1 k̃ . We denote the orthonormal basis of ker Q† (p̃) and ker Q (p̃) as Ψ+
and Ψ− , respectively. Note that ker Q† (p̃) and ker Q (p̃) share the same dimension s due to the vanishing Fredholm
index.  
We firstly prove that Q−1 k̃ is bounded as its domain restricted to imQ (p̃). For a vector ϕ0 ∈ imQ (p̃) =
⊥
ker Q† (p̃) (im denoting image and ⊥ denoting orthogonal complement), there always exists a vector ξ0 to make

⊥
Q (p̃) ξ0 = ϕ0 and hence Ψ†+ ϕ0 = 0. The Fredholm operator Q (p̃) restricted to [ker Q (p̃)] has a lower bound L such
−1
that ∥Q (p̃) ξ0 ∥ > L ∥ξ0 ∥ or, equivalently, ∥ξ0 ∥ < L ∥ϕ0 ∥. Based on ξ0 , we construct a new vector
 
Q k̃ − Q (p̃)
ϕ1 = − (ξ0 + Ψ− v0 ) (B1)
ρ
 13

with
 1  
v0 = −g −1 k̃ Ψ†+ Q k̃ ξ0 ∈ Cs (B2)
ρ
⊥
such that ϕ1 ∈ ker Q† (p̃) = imQ (p̃), i.e.,


1  
Ψ†+ ϕ1 = − Ψ†+ Q k̃ (ξ0 + Ψ− v0 ) = 0. (B3)
ρ
   
Here ρ1 Ψ†+ Q k̃ Ψ− = g k̃ has been used (the subscript of g has been omitted). All the operators above are bounded,
therefore, there exists a bound M0 such that ∥Ψ− v0 ∥ < M0 ∥ξ0 ∥ and a bound M1 such that ∥ϕ1 ∥ < M1 ∥ξ0 ∥.
Now we recast Eq. (B1) as
   
Q−1 k̃ ϕ0 = (ξ0 + Ψ− v0 ) + ρQ−1 k̃ ϕ1 , (B4)

and find

∥ξ0 + Ψ− v0 ∥ < L−1 (1 + M0 ) ∥ϕ0 ∥ , (B5)

∥ϕ1 ∥ < L−1 M1 ∥ϕ0 ∥ . (B6)

Due to ϕ1 ∈ imQ (p̃) again, we can continuously repeat the above treatment and obtain the expansion
 
Q−1 k̃ ϕ0 = (ξ0 + Ψ− v0 ) + ρ (ψ1 + Ψ− v1 ) + · · · , (B7)

with
 
Q−1 k̃ ϕ0 < L−1 (1 + M0 ) ∥ϕ0 ∥ + L−1 (1 + M0 ) · ρL−1 M1 ∥ϕ0 ∥ + · · · (B8)
−1
L (1 + M0 )
= ∥ϕ0 ∥ . (B9)
1 − ρL−1 M1
 
Thus Q−1 k̃ restricted to imQ (p̃) is bounded, which indicates
 
lim ρQ−1 k̃ ϕ0 = 0, ∀ϕ0 ∈ imQ (p̃) . (B10)
ρ→0

 
⊥ ⊥
We next calculate Q−1 k̃ restricted to [imQ (p̃)] . A vector ϕ ∈ [imQ (p̃)] = ker Q† (p̃) can always be expanded
as ϕ = Ψ+ v with v ∈ Cs . And we construct

1    
ϕ0 = ϕ − Q k̃ Ψ− g −1 k̃ v (B11)
ρ
 
which satisfies Ψ†+ ϕ0 = 0, i.e., ϕ0 ∈ imQ (p̃). From the previous result, we know limρ→0 ρQ−1 k̃ ϕ0 = 0. Therefore,
we obtain
     
lim ρQ−1 k̃ ϕ = lim ρQ−1 k̃ ϕ0 + lim Ψ− g −1 k̃ v (B12)
ρ→0 ρ→0 ρ→0
 
= Ψ− lim g −1 k̃ Ψ†+ ϕ. (B13)
ρ→0

At last, we combine Eqs. (B10) and (B13) together, and arrive at

   
lim ρQ−1 k̃ = Ψ− lim g −1 k̃ Ψ†+ . (B14)
ρ→0 ρ→0
 14

Appendix C: Expressions of Chern number

The Chern number of a (2n)D Chern insulator is defined as follows. Firstly, we make local unitary diagonalization
for the bulk Hamiltonian
U † (k) h (k) U (k) = E (k) , (C1)
where E (k) is the diagonal matrix of eigen-energies and U (k) is unitary. Let E = E+ ⊕ E− in which E± corresponds
to the conductive (positive) and valence (negative) band energies, respectively. Simultaneously, the unitary matrix
takes the form of U = (u+ , u− ) in which u± (k) is the orthonormal basis of conductive and valence bands, respectively.
Next, we consider the valence bands as a vector bundle and define its Berry connection (generally non-Abelian) as
A (k) = u†− (k) du− (k) . (C2)
It induces the Berry curvature 2-form
F = dA + A2 = −u†− du+ ∧ u†+ du− , (C3)

2
in which identity d U −1 dU = − U −1 dU has been used. Finally, the Chern number is defined by

ˆ  n
1 i
Ch = tr F (C4)
n! T 2n 2π
 n ˆ n
1 i 
= tr −du+ · u†+ ∧ du− · u†− . (C5)
n! 2π T 2n

Now we show that the Chern number is related to the winding number of the constructed chiral model given by Eq.
(47). We note that the winding number is invariant under homotopy qt (k) (t ∈ [0, 1]) that keeps the gap condition
det qt (k) ̸= 0, for
ˆ
δ δ −1

2n
deg = (2n + 1) Cn tr qt dqt ∧ qt−1 dqt (C6)
δt 2n+1 δt
ˆT  
δ
2n
= (2n + 1) Cn dtr qt−1 qt · qt−1 dqt (C7)
T 2n+1 δt
= 0, (C8)
2n
where identity d qt−1 dqt = 0 has been used. This allows us to utilize the flattened Hamiltonian h1 = U (sgnE) U †


obtained from the homotopy
− t
ht (k) = h2 (k) 2 h (k) , t ∈ [0, 1]

(C9)
to calculate Eq. (47), leading to
ˆ
 2n
 
1  −1 −idω
deg = (2n + 1) Cn tr U dU, sgnE ∧ . (C10)
T 2n+1 sgnE − iωIN sgnE − iωIN
By substituting U = (u+ , u− ) into it, we reduce it to
ˆ ˆ  n  n 1
4 † †
deg = −iCn (2n + 1) dω tr u + du − ∧ u− du+
T 2n 1 + ω2 1 − iω
 n 
4  n 1
+ tr u†− du+ ∧ u†+ du− . (C11)
1 + ω2 −1 − iω
After integrating over ω, we finally achieve
 n ˆ n
1 i 
deg = tr du+ · u†+ ∧ du− · u†− (C12)
n! 2π T 2n
n
= (−1) Ch. (C13)

[1] M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82, 3045 (2010).
 15

[2] X.-L. Qi and S.-C. Zhang, Rev. Mod. Phys. 83, 1057 (2011).
[3] C.-K. Chiu, J. C. Y. Teo, A. P. Schnyder, and S. Ryu, Rev. Mod. Phys. 88, 035005 (2016).
[4] F. Lu and Y.-M. Lu, “Magnon band topology in spin-orbital coupled magnets: classification and application to α-rucl3 ,”
(2018), arXiv:1807.05232 [cond-mat.str-el].
[5] Q.-R. Xu, V. P. Flynn, A. Alase, E. Cobanera, L. Viola, and G. Ortiz, Phys. Rev. B 102, 125127 (2020).
[6] Z. Zhou, L.-L. Wan, and Z.-F. Xu, J. Phys. A: Math. Theor. 53, 425203 (2020).
[7] Z. Gong, Y. Ashida, K. Kawabata, K. Takasan, S. Higashikawa, and M. Ueda, Phys. Rev. X 8, 031079 (2018).
[8] K. Kawabata, K. Shiozaki, M. Ueda, and M. Sato, Phys. Rev. X 9, 041015 (2019).
[9] R. B. Laughlin, Phys. Rev. B 23, 5632 (1981).
[10] Y. Hatsugai, Phys. Rev. Lett. 71, 3697 (1993).
[11] D. J. Thouless, M. Kohmoto, M. P. Nightingale, and M. den Nijs, Phys. Rev. Lett. 49, 405 (1982).
[12] C. Nayak, S. H. Simon, A. Stern, M. Freedman, and S. Das Sarma, Rev. Mod. Phys. 80, 1083 (2008).
[13] J. Kellendonk, T. Richter, and H. Schulz-Baldes, Rev. Math. Phys. 14, 87 (2002).
[14] E. Prodan and H. Schulz-Blades, Bulk and Boundary Invariants for Complex Topological Insulators (Springer Cham, 2016).
[15] A. M. Essin and V. Gurarie, Phys. Rev. B 84, 125132 (2011).
[16] G. M. Graf and M. Porta, Commun. Math. Phys. 324, 851 (2013).
[17] T. Fukui, K. Shiozaki, T. Fujiwara, and S. Fujimoto, J. Phys. Soc. Jpn. 81, 114602 (2012).
[18] W. Arveson, A Short Course on Spectral Theory (Springer New York, NY, 2001).
[19] Atkinson’s theorem tells that Q is a Fredholm operator if and only if ∃P making P Q − I and QP − I compact. This can
be realized by constructing a P satisfying φ (P ) = q −1 such that φ (P Q − I) = φ (QP − I) = 0, while Appendix A reveals
that φ (K) = 0 implies the compactness of K.
[20] Z. Zhou, X. Duan, and K.-J. Sun, J. Phys. A: Math. Theor. 51, 315201 (2018).
[21] S. M. G. A. Böttcher, Spectral Properties of Banded Toeplitz Matrices (SIAM, Philadelphia, 2005).