For example, let's consider a quadratic fermionic Hamiltonian on a 2D lattice with translation symmetry, and assume that the Fourier transformed Hamiltonian is described by a $2\times2$ Hermitian matrix $H(\mathbf{k})=a(\mathbf{k})\sigma_x+b(\mathbf{k})\sigma_y+c(\mathbf{k})\sigma_z $ and has a finite energy gap, then the Chern number $N$ can be determined.

If $H(-\mathbf{k})=H(\mathbf{k})$ holds for all $\mathbf{k}\in BZ$, then the Chern number $N$ is always an even number, am I right? This seems to be true from the geometrical interpretation of Chern number as a winding number covering a unit sphere, but I have not yet found a rigorous mathematical proof.

Remark: The necessary condition finite energy gap ($\Leftrightarrow$ The map $(a(\mathbf{k}),b(\mathbf{k}),c(\mathbf{k}))/\sqrt{a(\mathbf{k})^2+b(\mathbf{k})^2+c(\mathbf{k})^2}$ from BZ(2D torus) to the unit sphere is well defined) is to ensure that the Chern number/winding number is well defined.


I just found a relative rigorous argument supporting my conjecture:

The Chern number $N=\frac{1}{2\pi}\int _{BZ}b(\mathbf{k})$, where $b(\mathbf{k})$ is the Berry curvature. Since $H(-\mathbf{k})=H(\mathbf{k})$, it's easy to show that $b(-\mathbf{k})=b(\mathbf{k})$, accordingly, we can divide the $BZ$ into two halfs called $BZ_1$ and $BZ_2$, therefore, $N=\frac{1}{2\pi}\int _{BZ_1}b(\mathbf{k})+\frac{1}{2\pi}\int _{BZ_2}b(\mathbf{k})=2\times \frac{1}{2\pi}\int _{BZ_1}b(\mathbf{k})$. Now there are two ways to show that $\frac{1}{2\pi}\int _{BZ_1}b(\mathbf{k})$ is an integer,

(1)Due to the relation $b(-\mathbf{k})=b(\mathbf{k})$ and periodic structure of the $BZ$, the half Brillouin zone $BZ_1$ is topologically equivalent to a sphere, and the 'flux' through a sphere(closed surface) $\frac{1}{2\pi}\int _{BZ_1}b(\mathbf{k})$ must be quantized as an integer;

(2)Since $H(-\mathbf{k})=H(\mathbf{k})$, the eigenfunction $\psi(\mathbf{k})$ is also even(i.e. $\psi(-\mathbf{k})=\psi(\mathbf{k})$), then the Berry connection $\mathbf{a(k)} \propto \left \langle \psi(\mathbf{k})\mid \bigtriangledown_{\mathbf{k}} \psi(\mathbf{k})\right \rangle$ is odd, thus, it's easy to show that $\int _{BZ_1}b(\mathbf{k})=\oint _ {\partial BZ_1}\mathbf{a(k)}\cdot d\mathbf{k}=0$(where $\partial BZ_1$ is the boundary of $BZ_1$), however, to be consistent, the number '0'(Berry phase) here should be understood as $2\pi\times integer$.

Remark: The key point in our argument (1) is that the points $\mathbf{k}$ and $-\mathbf{k}$ are equivalent, and hence the half $BZ$ is topologically equivalent to a sphere which is a closed surface.

  • $\begingroup$ See the Supplemental Material in arXiv:1604.04781 $\endgroup$
    – Kai Li
    Apr 25 '16 at 7:36

I agree with your argument, but I thought I would just rephrase the same logic in a slightly different way, similar to how one would prove it in an algebraic topology course. (I would have done this as a comment, but it's a bit too big for that.)

Basically the Chern number measures the topology of the map $\newcommand{\ra}[1]{\kern-1.5ex\xrightarrow{ \ \ #1\ \ }\phantom{}\kern-1.5ex}T^2 \;\ra{n} \; \;S^2$, where $n(\mathbf k) = (a(\mathbf k),b(\mathbf k),c(\mathbf k))$. More exactly the map $n$ induces a map of the homology groups $\newcommand{\ra}[1]{\kern-1.5ex\xrightarrow{ \ \ #1\ \ }\phantom{}\kern-1.5ex}H_2(T^2) \; \;\ra{n} \; \;H_2(S^2)$ and the Chern number is given by $n ([T^2]) = C_1 *[S^2]$, where $[X]$ is the generator of the group $H_2(X)$.

Now because of the property of $n$, we can write down the following commutative diagram:

$\newcommand{\ra}[1]{\kern-1.5ex\xrightarrow{ \ \ #1\ \ }\phantom{}\kern-1.5ex} \begin{array}{ccccc} T^2 && \ra{\qquad\quad n \qquad\quad} & &S^2 \\ &\searrow_g&&\nearrow_h& \\ && (T^2/\sim) \;\simeq S^2& \end{array}$

where $g$ is the map that identifies the points $\mathbf k$ and $-\mathbf k$ on the torus. As you point out, this quotient space is homotopically equivalent to $S^2$. The above also implies a commutative diagram for the homology groups, such that $n([T^2]) = h \circ g ([T^2])$, but it is clear that $g ([T^2]) = 2 [S^2]$. (One can justify that rigorously using the fact that every point in $S^2$ has two pre-images in $T^2$.) Hence we have proven that $C_1 [S^2] = 2 \; h([S^2])$, i.e. $C_1$ is even.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.