betaFeedback

WorksPhysics and Astronomy

On equivalence of two formulas of orbital magnetic susceptibility for tight-binding models

Abstract

We prove that two formulas of the orbital magnetic susceptibility, namely, Koshino-Ando's formula and Gómez-Santos and Stauber's formula, are equivalent for the tight-binding models. The difference between these two formulas can be written in a form containing the surface terms arising from the momentum-space integration, and we show that the surface terms are exactly vanishing although the primitive functions are seemingly non-periodic with respect to the translation by the reciprocal lattice vector.

Magnetic-field effects in solids have long drawn interest of condensed matter physicists. Orbital magnetic susceptibility is a fundamental quantity that describes the magnetic-field response of nonmagnetic materials coming from orbital motion of electrons. Even for the noninteracting electrons, the theoretical treatment of magnetic field is challenging and a tremendous amount of effort has been devoted.1–13) Through these works, it has been emphasized that the interband effect plays an essential role in the orbital magnetic susceptibility, and some parts of the interband effects are reinterpreted in terms of modern notions such as topology and quantum geometry. Such an aspect also gains attention from the experimental point of view.14)

The aim of this Short Note is to show the equivalence of two theoretical formulas for the orbital magnetic susceptibility in a detailed manner. One formula is given by Koshino and Ando:8)

χorb(KA)=−e24πℏ2∫−∞∞dϵ f(ϵ)Im⁡[Θ(KA)(ϵ)],(1)\chi_{\mathrm{orb}}^{(\mathrm{KA})}=-\frac{e^2}{4\pi\hbar^2}\int_{-\infty}^{\infty}d\epsilon\,f(\epsilon)\operatorname{Im}[\Theta^{(\mathrm{KA})}(\epsilon)], \tag*{(1)}

with

Θ(KA)(ϵ)=1V∑kTr⁡{vkxGk(ϵ)vkyGk(ϵ)vkxGk(ϵ)vkyGk(ϵ)−2vkxGk(ϵ)vkxGk(ϵ)vkyGk(ϵ)vkyGk(ϵ)−12[Gk(ϵ)vkyGk(ϵ)∂vkx∂kxGk(ϵ)vky+Gk(ϵ)vkxGk(ϵ)∂vky∂kyGk(ϵ)vkx]}.(2)\Theta^{(\mathrm{KA})}(\epsilon)=\frac{1}{V}\sum_{\boldsymbol{k}}\operatorname{Tr}\left\{v_{\boldsymbol{k}}^xG_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^yG_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^xG_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^yG_{\boldsymbol{k}}(\epsilon)-2v_{\boldsymbol{k}}^xG_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^xG_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^yG_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^yG_{\boldsymbol{k}}(\epsilon)-\frac{1}{2}\left[G_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^yG_{\boldsymbol{k}}(\epsilon)\frac{\partial v_{\boldsymbol{k}}^x}{\partial k_x}G_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^y+G_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^xG_{\boldsymbol{k}}(\epsilon)\frac{\partial v_{\boldsymbol{k}}^y}{\partial k_y}G_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^x\right]\right\}. \tag*{(2)}

The other one is given by Gómez-Santos and Stauber.9)

χorb(GS)=−e24πℏ2∫−∞∞dϵ f(ϵ)Im⁡[Θ(GS)(ϵ)],(3)\chi_{\mathrm{orb}}^{(\mathrm{GS})}=-\frac{e^2}{4\pi\hbar^2}\int_{-\infty}^{\infty}d\epsilon\,f(\epsilon)\operatorname{Im}[\Theta^{(\mathrm{GS})}(\epsilon)], \tag*{(3)}

with

Θ(GS)(ϵ)=1V∑kTr⁡[2vkxGk(ϵ)vkyGk(ϵ)vkxGk(ϵ)vkyGk(ϵ)+Gk(ϵ)vkxGk(ϵ)vkyGk(ϵ)vkxGk(ϵ)vkyGk(ϵ)\Theta^{(\mathrm{GS})}(\epsilon)=\frac{1}{V}\sum_{\boldsymbol{k}}\operatorname{Tr}[2v_{\boldsymbol{k}}^xG_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^yG_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^xG_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^yG_{\boldsymbol{k}}(\epsilon)+G_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^xG_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^yG_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^xG_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^yG_{\boldsymbol{k}}(\epsilon)
+(Gk(ϵ)vkxGk(ϵ)vky+Gk(ϵ)vkyGk(ϵ)vkx)Gk(ϵ)∂vky∂kx].(4)\qquad+(G_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^xG_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^y+G_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^yG_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^x)G_{\boldsymbol{k}}(\epsilon)\frac{\partial v_{\boldsymbol{k}}^y}{\partial k_x}]. \tag*{(4)}

Here, ee is the elementary charge, ℏ\hbar is the reduced Planck constant, VV is the volume of the system, and f(ϵ)=1/(eβ(ϵ−μ)+1)f(\epsilon)=1/(e^{\beta(\epsilon-\mu)}+1) is the Fermi-Dirac distribution function. vkμv_{\boldsymbol{k}}^\mu (μ=x,y\mu=x,y) and Gk(ϵ)G_{\boldsymbol{k}}(\epsilon) are the matrices of the velocity and the Green’s function, respectively, whose precise definitions are presented later.

It has been pointed out in Ref.15) that one can show the equivalence of the above two formulas by using the partial integration for the momentum-space integration. However, the equivalence holds only up to the surface term.16) For concreteness and the simplicity of writing, in the following, we consider the case of two-dimensional systems having a square unit cell with the lattice constant being unity, but the argument for the generic tight-binding models (in both two and three dimensions) is essentially the same. In this case, we have

Θ(KA)(ϵ)−Θ(GS)(ϵ)∝∫02πdky[Tr⁡{Gk(ϵ)vkyGk(ϵ)vkxGk(ϵ)vky}]kx=02π+∫02πdkx[Tr⁡{Gk(ϵ)vkxGk(ϵ)vkyGk(ϵ)vkx}]ky=02π.(5)\Theta^{(\mathrm{KA})}(\epsilon)-\Theta^{(\mathrm{GS})}(\epsilon)\propto\int_0^{2\pi}dk_y\left[\operatorname{Tr}\{G_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^yG_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^xG_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^y\}\right]_{k_x=0}^{2\pi} +\int_0^{2\pi}dk_x\left[\operatorname{Tr}\{G_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^xG_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^yG_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^x\}\right]_{k_y=0}^{2\pi}. \tag*{(5)}

To obtain the above, the relation ∂Gk(ϵ)∂kμ=Gk(ϵ)vkμGk(ϵ)\frac{\partial G_{\boldsymbol{k}}(\epsilon)}{\partial k_\mu}=G_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^\mu G_{\boldsymbol{k}}(\epsilon) as well as the cyclic property of the trace have been used. Now, let us clarify the issue we want to discuss in this note. Some of the readers may think that the right-hand side of Eq. (5) is obviously vanishing since the primitive functions, Tr⁡{Gk(ϵ)vkyGk(ϵ)vkxGk(ϵ)vky}\operatorname{Tr}\{G_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^yG_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^xG_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^y\} and Tr⁡{Gk(ϵ)vkxGk(ϵ)vkyGk(ϵ)vkx}\operatorname{Tr}\{G_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^xG_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^yG_{\boldsymbol{k}}(\epsilon)v_{\boldsymbol{k}}^x\}, have periodicity of kμ→kμ+2πk_\mu\to k_\mu+2\pi. However, the periodicity is nontrivial when there are multiple number of sublattices in the unit cell. In such a case, the formula is valid when a Fourier transform is performed with incorporating the sublattice position, as we will elaborate later. Then, at this stage, it is not clear whether Eq. (5) is vanishing. What we shall show in this note is that this periodicity indeed holds even under the sublattice-position-dependent Fourier transformation. This leads to the conclusion that the two formulas are equivalent as far as the tight-binding models are concerned.17)

Let us first manifest the set-up of the present study. We consider the tight-binding Hamiltonian which is generically written as

H=∑R,R′∑a,bhab(R−R′)cR,a†cR′,b,(6)H=\sum_{\boldsymbol{R},\boldsymbol{R}'}\sum_{a,b}h^{ab}(\boldsymbol{R}-\boldsymbol{R}')c_{\boldsymbol{R},a}^{\dagger}c_{\boldsymbol{R}',b}, \tag*{(6)}

where R\boldsymbol{R}, R′\boldsymbol{R}' denote the position of the center of the unit cells, and aa, bb label the internal degrees of freedom including the sublattice degrees of freedom. The translational invariance lies in the fact that hab(R−R′)h^{ab}(\boldsymbol{R}-\boldsymbol{R}') is a function of R−R′\boldsymbol{R}-\boldsymbol{R}'.

We perform a Fourier transform with incorporating the position dependence of each sublattice inside the unit cell as ck,a=1N∑RcR,ae−ik⋅(R+ra)c_{\boldsymbol{k},a}=\frac{1}{\sqrt{N}}\sum_{\boldsymbol{R}}c_{\boldsymbol{R},a}e^{-i\boldsymbol{k}\cdot(\boldsymbol{R}+\boldsymbol{r}_a)} and ck,a†=1N∑RcR,a†eik⋅(R+ra)c_{\boldsymbol{k},a}^{\dagger}=\frac{1}{\sqrt{N}}\sum_{\boldsymbol{R}}c_{\boldsymbol{R},a}^{\dagger}e^{i\boldsymbol{k}\cdot(\boldsymbol{R}+\boldsymbol{r}_a)}, where NN denotes the number of the unit cells and ra\boldsymbol{r}_a denotes the position of the sublattice aa measured from the center of the unit cell. Using this convention, the Hamiltonian of (6) is rewritten as H=∑k∑a,bhkabck,a†ck,bH=\sum_{\boldsymbol{k}}\sum_{a,b}h_{\boldsymbol{k}}^{ab}c_{\boldsymbol{k},a}^{\dagger}c_{\boldsymbol{k},b}, where hkab=∑Rhab(R)e−ik⋅(R+ra−rb)h_{\boldsymbol{k}}^{ab}=\sum_{\boldsymbol{R}}h^{ab}(\boldsymbol{R})e^{-i\boldsymbol{k}\cdot(\boldsymbol{R}+\boldsymbol{r}_a-\boldsymbol{r}_b)}. It is worth noting that the hkabh_{\boldsymbol{k}}^{ab} is not periodic in the momentum space, i.e., hk+Kab≠hkabh_{\boldsymbol{k}+\boldsymbol{K}}^{ab}\ne h_{\boldsymbol{k}}^{ab} with K=2π(nx,ny)\boldsymbol{K}=2\pi(n_x,n_y) (nx,y∈Zn_{x,y}\in\mathbb{Z}) being a reciprocal lattice vector. This is due to the fact that the sublattice positions, ra\boldsymbol{r}_a and rb\boldsymbol{r}_b, are not the integer multiples of the lattice vectors. In the following, we use the notation that hkh_k denotes a M×MM \times M matrix whose abab-element is equal to hkabh_k^{ab}; MM is the number of internal degrees of freedom. Using the matrix representation of the Hamiltonian, one also obtain the (retarded) Green’s function,

Gk(ϵ)=(ϵ+iη−hk)−1G_k(\epsilon) = (\epsilon+ i\eta- h_k)^{-1},

where η\eta corresponds to the damping rate.

The velocity operator in the μ\mu-direction (μ=x,y\mu= x, y) is given as vμ=∑k,a,b[vkμ]abck,a†ck,bv^\mu= \sum_{k,a,b} [v_k^\mu]_{ab} c^\dagger_{k,a} c_{k,b} with

vkμ=∂hk∂kμ.(7)v_k^\mu= \frac{\partial h_k}{\partial k_\mu}. \tag*{(7)}

It is to be emphasized that the definition of the velocity operator in the form of Eq. (7) is valid only when the Fourier transformation incorporates the sublattice position.

Now, let us consider the right-hand side of Eq. (5). It is important to note that the matrix hkh_k can be rewritten as

hk=DkhˉkDk†,(8)h_k = D_k \bar{h}_k D_k^\dagger, \tag*{(8)}

where DkD_k is a unitary diagonal matrix

[Dk]ab=e−ik⋅raδa,b,(9)[D_k]_{ab} = e^{-i k \cdot r_a}\delta_{a,b}, \tag*{(9)}

and [hˉk]ab=∑Rhab(R)e−ik⋅R[\bar{h}_k]^{ab} = \sum_R h^{ab}(R)e^{-i k \cdot R}. Note that hˉk\bar{h}_k satisfies hˉk+K=hˉk\bar{h}_{k+K} = \bar{h}_k. In the following, for clarity, the quantities denoted by ⋅ˉ\bar{\cdot} have a periodicity in the momentum space with respect to the translation by the reciprocal lattice vectors. Note that the Green’s function is also written as

Gk(ϵ)=DkGˉk(ϵ)Dk†,(10)G_k(\epsilon) = D_k \bar{G}_k(\epsilon)D_k^\dagger, \tag*{(10)}

with Gˉk(ϵ)=(ϵ+iη−hˉk)−1\bar{G}_k(\epsilon) = (\epsilon+ i\eta- \bar{h}_k)^{-1}. Further, by substituting Eq. (8) into Eq. (7), we have

vkμ=DkvˉkμDk†+∂Dk∂kμhˉkDk†+Dkhˉk∂Dk†∂kμ,(11)v_k^\mu= D_k\bar{v}_k^\mu D_k^\dagger+ \frac{\partial D_k}{\partial k_\mu}\bar{h}_kD_k^\dagger+ D_k\bar{h}_k\frac{\partial D_k^\dagger}{\partial k_\mu}, \tag*{(11)}

with vˉkμ=∂hˉk∂kμ\bar{v}_k^\mu= \frac{\partial\bar{h}_k}{\partial k_\mu}. In fact, by explicitly performing the kμk_\mu derivative of DkD_k and Dk†D_k^\dagger, we find that Eq. (11) can be rewritten as

vkμ=Dk(vˉkμ+wkμ)Dk†,(12)v_k^\mu= D_k(\bar{v}_k^\mu+ w_k^\mu)D_k^\dagger, \tag*{(12)}

where

wkμ=[Rμ,hˉk],(13)w_k^\mu= [\mathcal{R}_\mu,\bar{h}_k], \tag*{(13)}

with Rμ\mathcal{R}_\mu being the diagonal matrix whose element is [Rμ]ab=−iraμδa,b[\mathcal{R}_\mu]_{ab} = -ir_a^\mu\delta_{a,b}. The key observation is that wk+Kμ=wkμw_{k+K}^\mu= w_k^\mu holds, since Rμ\mathcal{R}_\mu is kk-independent and hˉk\bar{h}_k is periodic in the momentum space. Hence, wkμw_k^\mu is to be written as wˉkμ\bar{w}_k^\mu in our notation. We further define γˉkμ=vˉkμ+wˉkμ\bar{\gamma}_k^\mu= \bar{v}_k^\mu+ \bar{w}_k^\mu, which is periodic in the momentum space. Using γˉkμ\bar{\gamma}_k^\mu, we can rewrite Eq. (12) as

vkμ=DkγˉkμDk†.(14)v_k^\mu= D_k\bar{\gamma}_k^\mu D_k^\dagger. \tag*{(14)}

We now substitute Eqs. (10) and (14) into the right-hand side of Eq. (5). For the integrand of the first term, we have

[Tr⁡{Gk(ϵ)vkxGk(ϵ)vkxGk(ϵ)vkxGk(ϵ)vkx}]kx=02π=[Tr⁡{(DkGˉk(ϵ)Dk†)(DkγˉkxDk†)(DkGˉk(ϵ)Dk†)(DkγˉkxDk†)(DkGˉk(ϵ)Dk†)(DkγˉkxDk†)(DkGˉk(ϵ)Dk†)(DkγˉkxDk†)}]kx=02π=[Tr⁡{Gˉk(ϵ)γˉkxGˉk(ϵ)γˉkxGˉk(ϵ)γˉkxGˉk(ϵ)γˉkx}]kx=02π=0,(15)\begin{aligned} \left[\operatorname{Tr}\{G_k(\epsilon)v_k^xG_k(\epsilon)v_k^xG_k(\epsilon)v_k^xG_k(\epsilon)v_k^x\}\right]_{k_x=0}^{2\pi} \\ &=\left[\operatorname{Tr}\{(D_k\bar{G}_k(\epsilon)D_k^\dagger)(D_k\bar{\gamma}_k^xD_k^\dagger)(D_k\bar{G}_k(\epsilon)D_k^\dagger)(D_k\bar{\gamma}_k^xD_k^\dagger)(D_k\bar{G}_k(\epsilon)D_k^\dagger)(D_k\bar{\gamma}_k^xD_k^\dagger)(D_k\bar{G}_k(\epsilon)D_k^\dagger)(D_k\bar{\gamma}_k^xD_k^\dagger)\}\right]_{k_x=0}^{2\pi} \\ &=\left[\operatorname{Tr}\{\bar{G}_k(\epsilon)\bar{\gamma}_k^x\bar{G}_k(\epsilon)\bar{\gamma}_k^x\bar{G}_k(\epsilon)\bar{\gamma}_k^x\bar{G}_k(\epsilon)\bar{\gamma}_k^x\}\right]_{k_x=0}^{2\pi} \\ &=0, \tag*{(15)} \end{aligned}

since all the entries of the fourth line of Eq. (15) have the periodicity under k→k+Kk \to k+K. The same argument holds for the second term of Eq. (5). Therefore, we find that the right-hand side of Eq. (5) is vanishing, and thus χorb(KA)=χorb(GS)\chi_{\mathrm{orb}}^{(\mathrm{KA})} = \chi_{\mathrm{orb}}^{(\mathrm{GS})} exactly holds for the tight-binding models.

We remark that, using Eqs. (10) and (14), as well as the similar relation for ∂vkμ∂kν\frac{\partial v_k^\mu}{\partial k_\nu}, we can rewrite Θ(KA)/(GS)(ϵ)\Theta^{(\mathrm{KA})/(\mathrm{GS})}(\epsilon) such that all the entries in the trace is periodic in the momentum space. However, it does not mean that these are the sublattice-position independent quantities. Indeed, we can see that wˉkμ\bar{w}_k^\mu is sublattice-position-dependent because it contains Rμ\mathcal{R}_\mu [Eq. (13)]. Note that the importance of the position dependence has been recently emphasized by several literatures,[18, 19, 20, 21, 22] and it indeed affects some physical quantities.

To summarize, we have proven that the Koshino–Ando’s formula and the Gómez-Santos and Stauber’s formula are equivalent for the tight-binding models. The difference between these two formulas is given by the surface term resulting from the partial integration. Importantly, the formal expressions of these formulas are valid only when a Fourier transformation incorporating the sublattice position is used. Then, the Greens’ functions and the velocity operators are not periodic in the momentum space, which makes the treatment of the surface term somewhat delicate. We have shown that the surface terms are exactly vanishing even in this case.

Acknowledgement

We thank Masao Ogata, Hideaki Maebashi, Hiroyasu Matsuura, Soshun Ozaki, and Masaki Kato for fruitful discussions. This work is supported by JSPS KAKENHI Grant No. JP23K03243.

References

  1. [18]18) C. Bena and G. Montambaux: New Journal of Physics 11 (2009) 095003.
  2. [19]19) M. Fruchart, D. Carpentier, and K. Gawedzki: Europhysics Letters 106 (2014) 60002.
  3. [20]20) S. H. Simon and M. S. Rudner: Phys. Rev. B 102 (2020) 165148.
  4. [21]21) J.-N. Fuchs and F. Piéchon: Phys. Rev. B 104 (2021) 235428.
  5. [22]22) J. Cayssol and J. N. Fuchs: Journal of Physics: Materials 4 (2021) 034007.

Paper details

vlnc:W_p8wr34fs
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents