Introduction
Luttinger’s theorem [1 ] is a fundamental concept in modern theories of interacting electrons. Initially derived as a straightforward consequence of self-consistent many-body perturbation theory [1 ] , its significance has evolved over time, with a growing emphasis on its topological nature [2 , 3 , 4 ] , thereby transcending its perturbative origins. Indeed, it is related to one of the simplest topological quantities that can be constructed from the fully interacting single-particle Green’s function. It also has a profound connection to another celebrated topological response, the quantum Hall conductivity, through the Streda formula, as recently pointed out by Gavensky, Sachdev and Goldman [5 ] .
At zero temperature, T = 0 T = 0 T = 0 , where the discrete Matsubara frequency ϵ n = ( 2 n + 1 ) π T \epsilon_n = (2n+1)\pi T ϵ n = ( 2 n + 1 ) π T becomes a continuous variable ϵ ∈ [ − ∞ , ∞ ] \epsilon\in[-\infty, \infty] ϵ ∈ [ − ∞ , ∞ ] , the Luttinger count N L N_{\mathrm{L}} N L is defined by
N L : = ∑ k ∫ − ∞ ∞ d ϵ 2 π e i ϵ 0 + ∂ ∂ i ϵ Tr ( ln G ( i ϵ , k ) − 1 ) , (1) N_{\mathrm{L}} := \sum_{\mathbf{k}} \int_{-\infty}^{\infty} \frac{\mathrm{d}\epsilon}{2\pi} e^{i\epsilon0^{+}} \frac{\partial}{\partial i\epsilon} \operatorname{Tr}\left(\ln G(i\epsilon,\mathbf{k})^{-1}\right), \tag*{(1)} N L := k ∑ ∫ − ∞ ∞ 2 π d ϵ e i ϵ 0 + ∂ i ϵ ∂ Tr ( ln G ( i ϵ , k ) − 1 ) , (1)
where G ( i ϵ , k ) G(i\epsilon,\mathbf{k}) G ( i ϵ , k ) is the Green’s function matrix represented in a generic basis of M M M single-particle Bloch wavefunctions, with k \mathbf{k} k the lattice momentum. Luttinger’s theorem asserts that the electron number N N N at T = 0 T = 0 T = 0 satisfies N = N L N = N_{\mathrm{L}} N = N L .
According to the Streda formula [6 , 7 ] , the Hall conductivity at T = 0 T = 0 T = 0 in units where ℏ = 1 \hbar= 1 ℏ = 1 and, for instance, in two dimensions, can be obtained through
σ H = − e c V ∂ N ∂ B : = e 2 2 π ν H , (2) \sigma_{\mathrm{H}} = -\frac{ec}{V}\frac{\partial N}{\partial B} := \frac{e^{2}}{2\pi}\nu_{\mathrm{H}}, \tag*{(2)} σ H = − V ec ∂ B ∂ N := 2 π e 2 ν H , (2)
where V V V is the surface of the sample, B B B a uniform magnetic field perpendicular to it, − e < 0 -e < 0 − e < 0 the electron charge, and ν H \nu_{\mathrm{H}} ν H is the dimensionless Hall number. It follows that, if we define the finite-field counterpart of the Luttinger count as
σ H L : = − e c V ∂ N L ∂ B : = e 2 2 π ν H L , (3) \sigma_{\mathrm{H L}} := -\frac{ec}{V}\frac{\partial N_{\mathrm{L}}}{\partial B} := \frac{e^2}{2\pi}\nu_{\mathrm{H L}}, \tag*{(3)} σ HL := − V ec ∂ B ∂ N L := 2 π e 2 ν HL , (3)
then ν H = ν H L \nu_{\mathrm{H}} = \nu_{\mathrm{H L}} ν H = ν HL if Luttinger’s theorem (1) holds. It has been demonstrated [5 , 8 ] that ν H L \nu_{\mathrm{H L}} ν HL coincides with the Ishikawa-Matsuyama-Volovik [9 , 10 , 11 ] expression for the Hall number
ν H L = i π V ∑ k ∫ d ϵ 2 π ϵ μ ν Tr ( G ( i ϵ , k ) ⋅ ∂ μ G ( i ϵ , k ) − 1 ⋅ ∂ i ϵ G ( i ϵ , k ) ⋅ ∂ ν G ( i ϵ , k ) − 1 ) , (4) \nu_{\mathrm{H L}} = \frac{i\pi}{V}\sum_{\mathbf{k}}\int\frac{d\epsilon}{2\pi}\epsilon_{\mu\nu}\operatorname{Tr}\left(G(i\epsilon,\mathbf{k})\cdot\partial_{\mu}G(i\epsilon,\mathbf{k})^{-1}\cdot\partial_{i\epsilon}G(i\epsilon,\mathbf{k})\cdot\partial_{\nu}G(i\epsilon,\mathbf{k})^{-1}\right), \tag*{(4)} ν HL = V iπ k ∑ ∫ 2 π d ϵ ϵ μν Tr ( G ( i ϵ , k ) ⋅ ∂ μ G ( i ϵ , k ) − 1 ⋅ ∂ i ϵ G ( i ϵ , k ) ⋅ ∂ ν G ( i ϵ , k ) − 1 ) , (4)
where ϵ μ ν \epsilon_{\mu\nu} ϵ μν is the antisymmetric tensor and ∂ μ : = ∂ k μ \partial_{\mu} := \partial_{k_{\mu}} ∂ μ := ∂ k μ . The dimensionless quantity ν H L \nu_{\mathrm{H L}} ν HL is another Green’s function topological invariant, specifically, the winding number of the map ( ϵ , k ) → G ( i ϵ , k ) ∈ GL ( M , C ) (\epsilon,\mathbf{k}) \to G(i\epsilon,\mathbf{k}) \in\operatorname{GL}(M,\mathbb{C}) ( ϵ , k ) → G ( i ϵ , k ) ∈ GL ( M , C ) . It follows that ν H = ν H L \nu_{\mathrm{H}} = \nu_{\mathrm{H L}} ν H = ν HL serves as yet another assertion of Luttinger’s theorem.
Although N L N_{\mathrm{L}} N L and, consequently, ν H L \nu_{\mathrm{H L}} ν HL are legitimate topological invariants [12 ] , there is compelling evidence [13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 , 22 ] suggesting that in correlated metals and insulators Luttinger’s theorem can be violated, thereby implying that N ≠ N L N \ne N_{\mathrm{L}} N = N L and ν H ≠ ν H L \nu_{\mathrm{H}} \ne\nu_{\mathrm{H L}} ν H = ν HL . A particularly transparent example is provided by fractional Chern insulators, where the physical ν H \nu_{\mathrm{H}} ν H is fractional, whereas ν H L \nu_{\mathrm{H L}} ν HL remains integer-valued [23 ] .
As mentioned at the beginning, the Luttinger theorem was originally proved [1 ] exploiting the skeleton-diagram expansion [24 ] , i.e. assuming the validity of the self-consistent (or bold ) many-body perturbation theory. Hence, the most natural explanation for possible violations of the Luttinger theorem is that its skeleton-diagram derivation [1 ] ceases to hold when the skeleton-diagram expansion breaks down [13 , 25 , 26 , 27 ] .
The breakdown of perturbation theory, which underlies relevant physical features (such as the electronic localization preceding Mott-Hubbard metal-insulator transitions [28 , 29 , 30 , 31 ] , adjacent phase-separation instabilities [32 , 33 ] or poles in the self-energy [15 , 20 , 21 , 22 ] ), is directly linked to important formal consequences in the many-body theory (such as divergencies of the two-particle irreducible vertices [25 , 34 , 35 , 36 , 37 ] , crossings of solutions of the Luttinger-Ward functional [26 , 38 , 39 , 27 , 40 ] , and misleading convergences of the iterative diagrammatic algorithms [41 , 42 , 43 ] ).
At the same time, we must emphasize that the Luttinger theorem can also be derived fully non-perturbatively [44 ] . Therefore, a breakdown of the skeleton-diagram expansion is not sufficient, per se, to account for the violation of the theorem (for the sake of clarity, throughout this paper we make a distinction between the skeleton-diagram expansion and the skeleton-diagram formalism [45 ] ).
In this respect, we note that, if one excludes the spurious/unphysical solutions [26 , 38 , 27 ] of the (multivalued) LW functional, all dynamical properties of the interacting systems, which are dependent on one frequency (such as, e.g., the one-particle Green’s function or the Kubo response functions) are well behaved for any nonzero frequency (ϵ ≠ 0 \epsilon\ne0 ϵ = 0 ) [46 ] , since the only physically meaningful discontinuities in these quantities can occur at zero Matsubara frequency. There, the imaginary frequency axis crosses the real axis, where the branch cut of the physical Green’s (or Kubo response) function lies. For instance, Tr ( G ( i 0 − , k ) − Tr ( G ( i 0 + , k ) \operatorname{Tr}(G(i0^{-},\mathbf{k}) - \operatorname{Tr}(G(i0^{+},\mathbf{k}) Tr ( G ( i 0 − , k ) − Tr ( G ( i 0 + , k ) coincides with 2 π i 2\pi i 2 π i times the momentum-dependent single-particle density of states at the chemical potential.
Building upon this insight, we demonstrate that the occurrence Luttinger’s theorem violations is exclusively associated with the above mentioned discontinuities at ϵ = 0 \epsilon= 0 ϵ = 0 . Consequently these discontinuities must be handled with utmost care under a fixed branch prescription. Indeed, as we show below, they generate a nontrivial global contribution in the Green’s function space associated with the Luttinger-Ward functional.
The Luttinger theorem
Since Luttinger’s theorem does not rely on any symmetry but particle conservation, we find it convenient in this section not to specify the basis of the single-particle wavefunctions ϕ α ( r ) \phi_\alpha(\mathbf{r}) ϕ α ( r ) , α = 1 , … , M \alpha= 1,\ldots,M α = 1 , … , M that includes the spin. The thermal Green’s function is therefore an M × M M \times M M × M complex matrix satisfying G ( i ϵ n ) = G ( − i ϵ n ) † G(i\epsilon_n) = G(-i\epsilon_n)^\dagger G ( i ϵ n ) = G ( − i ϵ n ) † , with matrix elements G α β ( i ϵ n ) G_{\alpha\beta}(i\epsilon_n) G α β ( i ϵ n ) and ϵ n = ( 2 n + 1 ) π T \epsilon_n = (2n+1)\pi T ϵ n = ( 2 n + 1 ) π T the Matsubara frequency. If Φ [ G ] \Phi[G] Φ [ G ] is the Luttinger-Ward (LW) functional, then the self-energy Σ ( i ϵ n ) = Σ ( − i ϵ n ) † \Sigma(i\epsilon_n) = \Sigma(-i\epsilon_n)^\dagger Σ ( i ϵ n ) = Σ ( − i ϵ n ) † has matrix elements defined through
Σ α β ( i ϵ n ) = δ Φ [ G ] δ G β α ( i ϵ n ) . (5) \Sigma_{\alpha\beta}(i\epsilon_n) = \frac{\delta\Phi[G]}{\delta G_{\beta\alpha}(i\epsilon_n)} . \tag*{(5)} Σ α β ( i ϵ n ) = δ G β α ( i ϵ n ) δ Φ [ G ] . (5)
Self-energy and Green’s function are related to each other by the Dyson equation
G ( i ϵ n ) − 1 = i ϵ n − H 0 − Σ ( i ϵ n ) , (6) G(i\epsilon_n)^{-1} = i\epsilon_n - H_0 - \Sigma(i\epsilon_n), \tag*{(6)} G ( i ϵ n ) − 1 = i ϵ n − H 0 − Σ ( i ϵ n ) , (6)
where H 0 H_0 H 0 is the non-interacting Hamiltonian represented in the chosen basis and which includes the chemical potential term. Since the number of electrons N N N is a conserved quantity, the LW functional is invariant under a shift of the Matsubara frequencies of all internal Green’s functions by the same bosonic frequency ω m = 2 m π T \omega_m = 2m\pi T ω m = 2 mπ T . This invariance property can be expressed mathematically as:
Φ [ G ( i ϵ n ) + Δ G ( i ϵ n ; i ω m ) ] − Φ [ G ( i ϵ n ) ] ≡ 0 , ∀ ω m = 2 m π T , (7) \Phi[G(i\epsilon_n) + \Delta G(i\epsilon_n;i\omega_m)] - \Phi[G(i\epsilon_n)] \equiv0,\quad\forall\omega_m = 2m\pi T, \tag*{(7)} Φ [ G ( i ϵ n ) + Δ G ( i ϵ n ; i ω m )] − Φ [ G ( i ϵ n )] ≡ 0 , ∀ ω m = 2 mπ T , (7)
where
Δ G ( i ϵ n ; i ω m ) : = G ( i ϵ n + i ω m ) − G ( i ϵ n ) . (8) \Delta G(i\epsilon_n;i\omega_m) := G(i\epsilon_n + i\omega_m) - G(i\epsilon_n). \tag*{(8)} Δ G ( i ϵ n ; i ω m ) := G ( i ϵ n + i ω m ) − G ( i ϵ n ) . (8)
For T → 0 T \to0 T → 0 also ω m → 0 \omega_m \to0 ω m → 0 at fixed m m m , which formally enables us to rewrite (7) using (5) as
0 ≡ lim T → 0 Φ [ G ( i ϵ n ) + Δ G ( i ϵ n ; i ω m ) ] − Φ [ G ( i ϵ n ) ] i ω m 0 \equiv\lim_{T\to0} \frac{\Phi[G(i\epsilon_n) + \Delta G(i\epsilon_n;i\omega_m)] - \Phi[G(i\epsilon_n)]}{i\omega_m} 0 ≡ T → 0 lim i ω m Φ [ G ( i ϵ n ) + Δ G ( i ϵ n ; i ω m )] − Φ [ G ( i ϵ n )]
≡ ? ∫ − ∞ ∞ d ϵ 2 π e i ϵ 0 + Tr ( Σ ( i ϵ ) ∂ G ( i ϵ ) ∂ i ϵ ) = ∫ − ∞ ∞ d ϵ 2 π Tr ( Σ ( i ϵ ) ∂ G ( i ϵ ) ∂ i ϵ ) = − ∫ − ∞ ∞ d ϵ 2 π Tr ( G ( i ϵ ) ∂ Σ ( i ϵ ) ∂ i ϵ ) . (9) \begin{aligned}
\mathrel{\stackrel{?}{\equiv}} \int_{-\infty}^{\infty}\frac{d\epsilon}{2\pi}e^{i\epsilon0^+}\operatorname{Tr}\left(\Sigma(i\epsilon)\frac{\partial G(i\epsilon)}{\partial i\epsilon}\right) \\
&= \int_{-\infty}^{\infty}\frac{d\epsilon}{2\pi}\operatorname{Tr}\left(\Sigma(i\epsilon)\frac{\partial G(i\epsilon)}{\partial i\epsilon}\right)
= -\int_{-\infty}^{\infty}\frac{d\epsilon}{2\pi}\operatorname{Tr}\left(G(i\epsilon)\frac{\partial\Sigma(i\epsilon)}{\partial i\epsilon}\right). \tag*{(9)}
\end{aligned} ≡ ? ∫ − ∞ ∞ 2 π d ϵ e i ϵ 0 + Tr ( Σ ( i ϵ ) ∂ i ϵ ∂ G ( i ϵ ) ) = ∫ − ∞ ∞ 2 π d ϵ Tr ( Σ ( i ϵ ) ∂ i ϵ ∂ G ( i ϵ ) ) = − ∫ − ∞ ∞ 2 π d ϵ Tr ( G ( i ϵ ) ∂ i ϵ ∂ Σ ( i ϵ ) ) . (9)
We have marked the second equality with a question mark because, as we show in § 2.3 , this superficially innocuous application of the functional chain rule is precisely where the conventional proof of Luttinger’s theorem fails. We adopt this convention throughout, to flag every subsequent occurrence of the same step. Moreover, in the last two equations of (9) we disregard the regularizing factor e i ϵ 0 + e^{i\epsilon0^+} e i ϵ 0 + and perform integration by parts. This is permissible because the integrand converges absolutely at infinity and does not exhibit any discontinuities at ϵ = 0 \epsilon= 0 ϵ = 0 in all relevant cases, such as correlated insulators and metals.
The electron number N N N can be obtained through
N = T ∑ n e i ϵ n 0 + Tr ( G ( i ϵ n ) ) , (10) N = T \sum_n e^{i\epsilon_n 0^+} \operatorname{Tr}\left(G(i\epsilon_n)\right), \tag*{(10)} N = T n ∑ e i ϵ n 0 + Tr ( G ( i ϵ n ) ) , (10)
where the regularizing exponential factor is required since G ( i ϵ n ) ∼ 1 / i ϵ n G(i\epsilon_n) \sim1/i\epsilon_n G ( i ϵ n ) ∼ 1/ i ϵ n for large ∣ ϵ n ∣ |\epsilon_n| ∣ ϵ n ∣ . For T → 0 T \to0 T → 0 , N N N becomes an integer, the sum over the Matsubara frequency an integral, so that (10) can be written through (6) as [1 ]
N = ∫ − ∞ ∞ d ϵ 2 π e i ϵ 0 + Tr ( G ( i ϵ ) ) = ∫ − ∞ ∞ d ϵ 2 π e i ϵ 0 + ∂ ∂ i ϵ Tr ( ln G ( i ϵ ) − 1 ) + ∫ − ∞ ∞ d ϵ 2 π e i ϵ 0 + Tr ( G ( i ϵ ) ∂ Σ ( i ϵ ) ∂ i ϵ ) : = N L + I L , (11) \begin{aligned}
N &= \int_{-\infty}^{\infty} \frac{\mathrm{d}\epsilon}{2\pi} e^{i\epsilon0^+} \operatorname{Tr}\left(G(i\epsilon)\right) \\
&= \int_{-\infty}^{\infty} \frac{\mathrm{d}\epsilon}{2\pi} e^{i\epsilon0^+} \frac{\partial}{\partial i\epsilon} \operatorname{Tr}\left(\ln G(i\epsilon)^{-1}\right) + \int_{-\infty}^{\infty} \frac{\mathrm{d}\epsilon}{2\pi} e^{i\epsilon0^+} \operatorname{Tr}\left(G(i\epsilon)\frac{\partial\Sigma(i\epsilon)}{\partial i\epsilon}\right) \\
&:= N_{\mathrm{L}} + I_{\mathrm{L}}, \tag*{(11)}
\end{aligned} N = ∫ − ∞ ∞ 2 π d ϵ e i ϵ 0 + Tr ( G ( i ϵ ) ) = ∫ − ∞ ∞ 2 π d ϵ e i ϵ 0 + ∂ i ϵ ∂ Tr ( ln G ( i ϵ ) − 1 ) + ∫ − ∞ ∞ 2 π d ϵ e i ϵ 0 + Tr ( G ( i ϵ ) ∂ i ϵ ∂ Σ ( i ϵ ) ) := N L + I L , (11)
which serves as a definition of the Luttinger integral I L I_{\mathrm{L}} I L , and N L N_{\mathrm{L}} N L is the Luttinger count defined in Eq. (1) . As discussed above, we can safely ignore the regularizing exponential factor in I L I_{\mathrm{L}} I L . Comparing (11) with (9) one must conclude that the Luttinger integral vanishes identically, thus that N ≡ N L N \equiv N_{\mathrm{L}} N ≡ N L at T = 0 T = 0 T = 0 . This is precisely the statement of Luttinger’s theorem [1 ] . Except for the step marked with a question mark, the derivation is free of any apparent flaw and extends beyond the validity limits of perturbation theory, as discussed by Potthoff in [44 ] . Given the several known violations of Luttinger’s theorem, it must be precisely that step which fails, as we show in the following.
A more in-depth examination
Let us begin our analysis by exploring more in detail the properties of the Green’s function G ( i ϵ ) G(i\epsilon) G ( i ϵ ) and of the Luttinger count. We write
Σ ( i ϵ ) = Σ ( − i ϵ ) † = Σ 1 ( ϵ ) + i Σ 2 ( ϵ ) , \Sigma(i\epsilon) = \Sigma(-i\epsilon)^\dagger= \Sigma_1(\epsilon) + i\Sigma_2(\epsilon), Σ ( i ϵ ) = Σ ( − i ϵ ) † = Σ 1 ( ϵ ) + i Σ 2 ( ϵ ) ,
where Σ 1 ( ϵ ) = Σ 1 ( ϵ ) † = Σ 1 ( − ϵ ) \Sigma_1(\epsilon) = \Sigma_1(\epsilon)^\dagger= \Sigma_1(-\epsilon) Σ 1 ( ϵ ) = Σ 1 ( ϵ ) † = Σ 1 ( − ϵ ) , and Σ 2 ( ϵ ) = Σ 2 ( ϵ ) † = − Σ 2 ( − ϵ ) \Sigma_2(\epsilon) = \Sigma_2(\epsilon)^\dagger= -\Sigma_2(-\epsilon) Σ 2 ( ϵ ) = Σ 2 ( ϵ ) † = − Σ 2 ( − ϵ ) is negative definite for ϵ > 0 \epsilon> 0 ϵ > 0 and positive otherwise. We define the quasiparticle residue matrix [8 ] as
Z ( ϵ ) = Z ( − ϵ ) = ( 1 − Σ 2 ( ϵ ) ϵ ) − 1 , (12) Z(\epsilon) = Z(-\epsilon) = \left(1 - \frac{\Sigma_2(\epsilon)}{\epsilon}\right)^{-1}, \tag*{(12)} Z ( ϵ ) = Z ( − ϵ ) = ( 1 − ϵ Σ 2 ( ϵ ) ) − 1 , (12)
which is semi-positive with eigenvalues ∈ [ 0 , 1 ] \in[0,1] ∈ [ 0 , 1 ] . Through (12) we can write
G ( i ϵ ) = Z ( ϵ ) 1 i ϵ − H ∗ ( ϵ ) Z ( ϵ ) : = Z ( ϵ ) G ∗ ( i ϵ ) Z ( ϵ ) , (13) G(i\epsilon) = \sqrt{Z(\epsilon)} \frac{1}{i\epsilon- H_*(\epsilon)} \sqrt{Z(\epsilon)} := \sqrt{Z(\epsilon)}G_*(i\epsilon)\sqrt{Z(\epsilon)}, \tag*{(13)} G ( i ϵ ) = Z ( ϵ ) i ϵ − H ∗ ( ϵ ) 1 Z ( ϵ ) := Z ( ϵ ) G ∗ ( i ϵ ) Z ( ϵ ) , (13)
where
H ∗ ( ϵ ) = H ∗ ( ϵ ) † = H ∗ ( − ϵ ) = Z ( ϵ ) ( H 0 + Σ 1 ( ϵ ) ) Z ( ϵ ) , (14) H_*(\epsilon) = H_*(\epsilon)^\dagger= H_*(-\epsilon) = \sqrt{Z(\epsilon)}\left(H_0 + \Sigma_1(\epsilon)\right)\sqrt{Z(\epsilon)}, \tag*{(14)} H ∗ ( ϵ ) = H ∗ ( ϵ ) † = H ∗ ( − ϵ ) = Z ( ϵ ) ( H 0 + Σ 1 ( ϵ ) ) Z ( ϵ ) , (14)
is the quasiparticle Hamiltonian with eigenvalues ϵ ℓ ( ϵ ) = ϵ ℓ ( − ϵ ) \epsilon_{\ell}(\epsilon)=\epsilon_{\ell}(-\epsilon) ϵ ℓ ( ϵ ) = ϵ ℓ ( − ϵ ) and eigenstates ∣ u ℓ ( ϵ ) ⟩ \lvert u_{\ell}(\epsilon)\rangle ∣ u ℓ ( ϵ )⟩ , ℓ = 1 , … , M \ell=1,\ldots,M ℓ = 1 , … , M , which depend parametrically on ϵ \epsilon ϵ . We denote ϵ ℓ ( ϵ = 0 ) = ϵ ℓ \epsilon_{\ell}(\epsilon=0)=\epsilon_{\ell} ϵ ℓ ( ϵ = 0 ) = ϵ ℓ as the actual quasiparticle energies. Since Z ( ϵ ) = Z ( − ϵ ) ≃ 1 − O ( 1 / ϵ 2 ) Z(\epsilon)=Z(-\epsilon)\simeq1-O(1/\epsilon^{2}) Z ( ϵ ) = Z ( − ϵ ) ≃ 1 − O ( 1/ ϵ 2 ) for large ∣ ϵ ∣ \lvert\epsilon\rvert ∣ ϵ ∣ , one readily finds that
N L = ∫ − ∞ ∞ d ϵ 2 π e i ϵ 0 + ∂ ∂ i ϵ Tr ( ln G ( i ϵ ) − 1 ) = ∫ − ∞ ∞ d ϵ 2 π e i ϵ 0 + ∂ ∂ i ϵ Tr ( ln G ∗ ( i ϵ ) − 1 ) = ∑ ℓ θ ( − ϵ ℓ ) = : ∑ ℓ n L ℓ . (15) \begin{aligned}
N_{L}=\int_{-\infty}^{\infty}\frac{\mathrm{d}\epsilon}{2\pi}e^{i\epsilon0^{+}}\frac{\partial}{\partial i\epsilon}\operatorname{Tr}\left(\ln G(i\epsilon)^{-1}\right)=\int_{-\infty}^{\infty}\frac{\mathrm{d}\epsilon}{2\pi}e^{i\epsilon0^{+}}\frac{\partial}{\partial i\epsilon}\operatorname{Tr}\left(\ln G_{*}(i\epsilon)^{-1}\right) \\
&=\sum_{\ell}\theta(-\epsilon_{\ell})=:\sum_{\ell}n_{L\ell}. \tag*{(15)}
\end{aligned} N L = ∫ − ∞ ∞ 2 π d ϵ e i ϵ 0 + ∂ i ϵ ∂ Tr ( ln G ( i ϵ ) − 1 ) = ∫ − ∞ ∞ 2 π d ϵ e i ϵ 0 + ∂ i ϵ ∂ Tr ( ln G ∗ ( i ϵ ) − 1 ) = ℓ ∑ θ ( − ϵ ℓ ) =: ℓ ∑ n L ℓ . (15)
which also clarifies the term quasiparticles that we adopted, and where the Luttinger occupation number n L ℓ n_{L\ell} n L ℓ is either 1 or 0.
We represent the matrix G ( i ϵ ) G(i\epsilon) G ( i ϵ ) in the basis of the eigenstates ∣ u ℓ ( 0 ) ⟩ \lvert u_{\ell}(0)\rangle ∣ u ℓ ( 0 )⟩ of H ∗ ( ϵ = 0 ) H_{*}(\epsilon=0) H ∗ ( ϵ = 0 ) . For each eigenstate, we associate the annihilation operator c ℓ c_{\ell} c ℓ and the creation operator c ℓ † c_{\ell}^{\dagger} c ℓ † . In this basis, the diagonal elements G ℓ ℓ ( i ϵ ) G_{\ell\ell}(i\epsilon) G ℓℓ ( i ϵ ) of the Green’s function G ( i ϵ ) G(i\epsilon) G ( i ϵ ) have the following spectral decomposition at T = 0 T=0 T = 0 :
G ℓ ℓ ( i ϵ ) = ∑ n > 0 { ∣ ⟨ n ∣ c ℓ † ∣ 0 ⟩ ∣ 2 i ϵ − E n + E 0 + ∣ ⟨ n ∣ c ℓ ∣ 0 ⟩ ∣ 2 i ϵ + E n − E 0 } = : ∑ n > 0 { R ℓ n + i ϵ − ω n + R ℓ n − i ϵ + ω n } . (16) G_{\ell\ell}(i\epsilon)=\sum_{n>0}\left\{\frac{\lvert\langle n\vert c_{\ell}^{\dagger}\vert0\rangle\rvert^{2}}{i\epsilon-E_{n}+E_{0}}+\frac{\lvert\langle n\vert c_{\ell}\vert0\rangle\rvert^{2}}{i\epsilon+E_{n}-E_{0}}\right\}=:\sum_{n>0}\left\{\frac{R_{\ell n}^{+}}{i\epsilon-\omega_{n}}+\frac{R_{\ell n}^{-}}{i\epsilon+\omega_{n}}\right\}. \tag*{(16)} G ℓℓ ( i ϵ ) = n > 0 ∑ { i ϵ − E n + E 0 ∣⟨ n ∣ c ℓ † ∣0 ⟩ ∣ 2 + i ϵ + E n − E 0 ∣⟨ n ∣ c ℓ ∣0 ⟩ ∣ 2 } =: n > 0 ∑ { i ϵ − ω n R ℓ n + + i ϵ + ω n R ℓ n − } . (16)
where ∣ n ⟩ \lvert n\rangle ∣ n ⟩ are the many-body eigenstates with eigenvalues E n E_{n} E n , n ≥ 0 n\geq0 n ≥ 0 , with ∣ 0 ⟩ \lvert0\rangle ∣ 0 ⟩ the ground state, assumed non-degenerate, and ω n = E n − E 0 > 0 \omega_{n}=E_{n}-E_{0}>0 ω n = E n − E 0 > 0 for n > 0 n>0 n > 0 . By definition,
∑ n > 0 R ℓ n + = ⟨ 0 ∣ c ℓ c ℓ † ∣ 0 ⟩ : = 1 − n ℓ ∈ [ 0 , 1 ] , ∑ n > 0 R ℓ n − = ⟨ 0 ∣ c ℓ † c ℓ ∣ 0 ⟩ : = n ℓ ∈ [ 0 , 1 ] . (17) \sum_{n>0}R_{\ell n}^{+}=\langle0\vert c_{\ell}c_{\ell}^{\dagger}\vert0\rangle:=1-n_{\ell}\in[0,1],\qquad\sum_{n>0}R_{\ell n}^{-}=\langle0\vert c_{\ell}^{\dagger}c_{\ell}\vert0\rangle:=n_{\ell}\in[0,1]. \tag*{(17)} n > 0 ∑ R ℓ n + = ⟨ 0∣ c ℓ c ℓ † ∣0 ⟩ := 1 − n ℓ ∈ [ 0 , 1 ] , n > 0 ∑ R ℓ n − = ⟨ 0∣ c ℓ † c ℓ ∣0 ⟩ := n ℓ ∈ [ 0 , 1 ] . (17)
For the sake of simplicity, hereafter we perform our analysis under the following conditions: (i) we consider an insulating state ; (ii) we assume, as mentioned above, that the ground state of such insulator is non-degenerate ; (iii) we discard sets of measure zero in the Hamiltonian parameter space where the two-particle irreducible vertex function is indeterminate [25 , 34 , 35 , 36 , 37 ] , and, finally (iv) we assume that G ( i ϵ ) G(i\epsilon) G ( i ϵ ) at ϵ = 0 \epsilon=0 ϵ = 0 , which is Hermitian, has no vanishing eigenvalue .
In practice, (i) implies that, even in the thermodynamic limit, ω n ≥ Δ \omega_{n}\geq\Delta ω n ≥ Δ , with Δ > 0 \Delta>0 Δ > 0 being the single-particle gap. This way, we put aside the additional complications of Fermi-liquid corrections discussed, for instance, in Ref. [47 ] , which are, anyway, not fully relevant for the analysis of the quantization of the Hall conductivity mentioned in the Introduction. At the same time, (ii) rules out possible spurious effects on I L I_{L} I L directly linked to the special situation of a degenerate many-body ground state. Assumption (iii) excludes pathological Hamiltonian parameter combinations where the skeleton diagrammatic formalism is locally ill defined, despite being well-defined anywhere else away from such isolated points. Assumption (iv) makes our scenario analogous to the model studied in Ref. [14 ] , where the violation of Luttinger’s theorem cannot be attributed to a singular behavior of Σ ( i ϵ ) \Sigma(i\epsilon) Σ ( i ϵ ) as ϵ → 0 \epsilon\to0 ϵ → 0 . This renders its violation all the more non-trivial.
In the chosen basis of single-particle wavefunctions ∣ u ℓ ( 0 ) ⟩ \lvert u_{\ell}(0)\rangle ∣ u ℓ ( 0 )⟩ , we can thus write the Luttinger integral (11) simply as
I L = N − N L = ∑ ℓ ( n ℓ − n L ℓ ) = ∑ ℓ ( n ℓ − θ ( − ϵ ℓ ) ) . (18) I_{L}=N-N_{L}=\sum_{\ell}\left(n_{\ell}-n_{L\ell}\right)=\sum_{\ell}\left(n_{\ell}-\theta(-\epsilon_{\ell})\right). \tag*{(18)} I L = N − N L = ℓ ∑ ( n ℓ − n L ℓ ) = ℓ ∑ ( n ℓ − θ ( − ϵ ℓ ) ) . (18)
The equation above already explains how and why Luttinger’s theorem can be violated. While n L ℓ n_{L\ell} n L ℓ is always either 0 or 1, the actual n ℓ n_{\ell} n ℓ can be any number in the range [ 0 , 1 ] [0,1] [ 0 , 1 ] . This deviation occurs when part of the spectral weight is transferred across the chemical potential [48 ] . In a weakly correlated insulator, different deviations compensate exactly to yield I L = 0 I_L=0 I L = 0 ; thus, Luttinger’s theorem is valid. However, when correlation is strong, single bands split into lower and upper Hubbard bands, with the lower band potentially accommodating any fractional number of electrons. In this case, some of the quasiparticle bands lose their connections with poles of the physical single-particle spectrum [8 ] , and the perfect compensation stops, leading to a finite I L I_L I L and the violation of Luttinger’s theorem. In particular, under varying the Hamiltonian parameters without closing the single-particle gap nor crossing a transition, the electron number N N N remains constant. In this case, the only way I L I_L I L in (18) can change is that one or more eigenvalues ϵ ℓ \epsilon_\ell ϵ ℓ of H ∗ ( 0 ) H_*(0) H ∗ ( 0 ) cross zero. Since the physical single-particle gap remains open, such crossing corresponds to det ( Z ( 0 ) ) = 0 = det ( G ( 0 ) ) \det(Z(0))=0=\det(G(0)) det ( Z ( 0 )) = 0 = det ( G ( 0 )) , see Eq. (14) , i.e. to poles of the self-energy at ϵ = 0 \epsilon=0 ϵ = 0 .
However, the above observations explain why I L I_L I L can become non-zero, but do not explain where the theorem’s proof fails. To address this, under the assumption of a fixed branch cut of the logarithm, i.e. Im ln z = Im ln ( x + i y ) = Arg ( z ) ∈ [ − π , π ] \operatorname{Im}\ln z=\operatorname{Im}\ln(x+iy)=\operatorname{Arg}(z)\in[-\pi,\pi] Im ln z = Im ln ( x + i y ) = Arg ( z ) ∈ [ − π , π ] , we rewrite (16) as
G ℓ ℓ ( i ϵ ) = ∂ ∂ i ϵ ∑ n > 0 { R ℓ n + ln ( i ϵ − ω n ) + R ℓ n − ln ( i ϵ + ω n ) } = ∂ ∂ i ϵ ln ( ∏ n > 0 ( i ϵ − ω n ) R ℓ n + ( i ϵ + ω n ) R ℓ n − ) : = ∂ ln S ℓ ( i ϵ ) ∂ i ϵ , (19) \begin{aligned}
G_{\ell\ell}(i\epsilon) &= \frac{\partial}{\partial i\epsilon}\sum_{n>0}\left\{R_{\ell n}^{+}\ln(i\epsilon-\omega_n)+R_{\ell n}^{-}\ln(i\epsilon+\omega_n)\right\}\\
&= \frac{\partial}{\partial i\epsilon}\ln\left(\prod_{n>0}(i\epsilon-\omega_n)^{R_{\ell n}^{+}}(i\epsilon+\omega_n)^{R_{\ell n}^{-}}\right):=\frac{\partial\ln S_\ell(i\epsilon)}{\partial i\epsilon}, \tag*{(19)}
\end{aligned} G ℓℓ ( i ϵ ) = ∂ i ϵ ∂ n > 0 ∑ { R ℓ n + ln ( i ϵ − ω n ) + R ℓ n − ln ( i ϵ + ω n ) } = ∂ i ϵ ∂ ln ( n > 0 ∏ ( i ϵ − ω n ) R ℓ n + ( i ϵ + ω n ) R ℓ n − ) := ∂ i ϵ ∂ ln S ℓ ( i ϵ ) , (19)
where we consider S ℓ ( i ϵ ) S_\ell(i\epsilon) S ℓ ( i ϵ ) the elements of a diagonal matrix S ( i ϵ ) S(i\epsilon) S ( i ϵ ) . Writing the sum of the logarithms as the logarithm of the product is consistent with the fixed branch cut assumption. Indeed, since 0 ≤ R ℓ n ± ≤ 1 0\le R_{\ell n}^{\pm}\le1 0 ≤ R ℓ n ± ≤ 1 and, through (17) , ∑ n > 0 ( R ℓ n + + R ℓ n − ) = 1 \sum_{n>0}(R_{\ell n}^{+}+R_{\ell n}^{-})=1 ∑ n > 0 ( R ℓ n + + R ℓ n − ) = 1 , then
\operatorname{Im}\ln S_\ell(i\epsilon)=-\operatorname{Im}\ln S_\ell(-i\epsilon)
=\sum_{n>0}\left\{R_{\ell n}^{+}\operatorname{Arg}(i\epsilon-\omega_n)+R_{\ell n}^{-}\operatorname{Arg}(i\epsilon+\omega_n)\right\}\in\begin{cases}[0,\pi] & \epsilon>0,\\[-\pi,0] & \epsilon<0.\end{cases} \tag*{(20)}
which ensures that the sum of the imaginary part of the logarithms does not fall outside the interval [ − π , π ] [-\pi,\pi] [ − π , π ] . Specifically,
Im ln S ℓ ( ± i 0 ) = ± π ( 1 − n ℓ ) , Im ln S ℓ ( ± i ∞ ) = ± π 2 . \operatorname{Im}\ln S_\ell(\pm i0)=\pm\pi(1-n_\ell),\qquad\operatorname{Im}\ln S_\ell(\pm i\infty)=\pm\frac{\pi}{2}. Im ln S ℓ ( ± i 0 ) = ± π ( 1 − n ℓ ) , Im ln S ℓ ( ± i ∞ ) = ± 2 π .
By the definition (11) of the Luttinger integral we can conclude that
I L = N − N L = ∫ − ∞ ∞ d ϵ 2 π e i ϵ 0 + ∂ ∂ i ϵ { Tr ( ln S ( i ϵ ) ) − Tr ( ln G ( i ϵ ) − 1 ) } = ∫ − ∞ ∞ d ϵ 2 π ∂ ∂ i ϵ { Tr ( ln S ( i ϵ ) ) − Tr ( ln G ∗ ( i ϵ ) − 1 ) } , (21) \begin{aligned}
I_L=N-N_L&=\int_{-\infty}^{\infty}\frac{d\epsilon}{2\pi}e^{i\epsilon0^+}\frac{\partial}{\partial i\epsilon}\left\{\operatorname{Tr}\left(\ln S(i\epsilon)\right)-\operatorname{Tr}\left(\ln G(i\epsilon)^{-1}\right)\right\}\\
&=\int_{-\infty}^{\infty}\frac{d\epsilon}{2\pi}\frac{\partial}{\partial i\epsilon}\left\{\operatorname{Tr}\left(\ln S(i\epsilon)\right)-\operatorname{Tr}\left(\ln G_*(i\epsilon)^{-1}\right)\right\}, \tag*{(21)}
\end{aligned} I L = N − N L = ∫ − ∞ ∞ 2 π d ϵ e i ϵ 0 + ∂ i ϵ ∂ { Tr ( ln S ( i ϵ ) ) − Tr ( ln G ( i ϵ ) − 1 ) } = ∫ − ∞ ∞ 2 π d ϵ ∂ i ϵ ∂ { Tr ( ln S ( i ϵ ) ) − Tr ( ln G ∗ ( i ϵ ) − 1 ) } , (21)
where we have ignored the regularizing exponential term since the integrand vanishes fast at ϵ = ± ∞ \epsilon=\pm\infty ϵ = ± ∞ . In essence, I L I_L I L is a pure boundary term, and this is the primary reason why the conventional proof of Luttinger’s theorem is flawed.
The arguments of both S ( i ϵ ) S(i\epsilon) S ( i ϵ ) and G ∗ ( i ϵ ) − 1 G_*(i\epsilon)^{-1} G ∗ ( i ϵ ) − 1 in (21) are Hermitian matrices, odd in ϵ \epsilon ϵ and with eigenvalues ∈ [ 0 , π ] \in[0,\pi] ∈ [ 0 , π ] for ϵ ≥ 0 \epsilon\ge0 ϵ ≥ 0 . Therefore, the difference of the arguments is also Hermitian and odd in ϵ \epsilon ϵ . Moreover, according to Weyl’s inequalities, its eigenvalues remain strictly within ∈ [ − π , π ] \in[-\pi,\pi] ∈ [ − π , π ] for both positive and negative ϵ \epsilon ϵ , they can cross a branch cut exclusively at ϵ = 0 \epsilon=0 ϵ = 0 and vanish for ϵ → ± ∞ \epsilon\to\pm\infty ϵ → ± ∞ .
It follows that, given the prescription of a fixed branch cut, we are allowed to define
ln X ( i ϵ ) : = ln S ( i ϵ ) − ln G ∗ ( i ϵ ) − 1 , (22) \ln X(i\epsilon) := \ln S(i\epsilon)-\ln G_{*}(i\epsilon)^{-1}, \tag*{(22)} ln X ( i ϵ ) := ln S ( i ϵ ) − ln G ∗ ( i ϵ ) − 1 , (22)
thus
I L = ∫ − ∞ ∞ d ϵ 2 π ∂ ∂ i ϵ Tr ( ln X ( i ϵ ) ) = − 1 2 π { Tr ( Arg ( X ( i 0 + ) ) ) − Tr ( Arg ( X ( i 0 − ) ) ) } . (23) I_{L}=\int_{-\infty}^{\infty}\frac{\mathrm{d}\epsilon}{2\pi}\frac{\partial}{\partial i\epsilon}\operatorname{Tr}\bigl(\ln X(i\epsilon)\bigr)=-\frac{1}{2\pi}\left\{\operatorname{Tr}\bigl(\operatorname{Arg}(X(i0^{+}))\bigr)-\operatorname{Tr}\bigl(\operatorname{Arg}(X(i0^{-}))\bigr)\right\}. \tag*{(23)} I L = ∫ − ∞ ∞ 2 π d ϵ ∂ i ϵ ∂ Tr ( ln X ( i ϵ ) ) = − 2 π 1 { Tr ( Arg ( X ( i 0 + )) ) − Tr ( Arg ( X ( i 0 − )) ) } . (23)
Since ln S ( i ϵ ) \ln S(i\epsilon) ln S ( i ϵ ) and ln G ∗ ( i ϵ ) − 1 \ln G_{*}(i\epsilon)^{-1} ln G ∗ ( i ϵ ) − 1 are diagonal in the same basis at ϵ = 0 \epsilon=0 ϵ = 0 , the eigenvalues ϕ ℓ ( ϵ ) = − ϕ ℓ ( − ϵ ) \phi_{\ell}(\epsilon)=-\phi_{\ell}(-\epsilon) ϕ ℓ ( ϵ ) = − ϕ ℓ ( − ϵ ) of Arg ( X ( i ϵ ) ) \operatorname{Arg}(X(i\epsilon)) Arg ( X ( i ϵ )) at ϵ = 0 \epsilon=0 ϵ = 0 are simply
ϕ ℓ ( 0 + ) = − ϕ ℓ ( 0 − ) = − π ( n ℓ − n L ℓ ) : = − π ζ ℓ , (24) \phi_{\ell}(0^{+})=-\phi_{\ell}(0^{-})=-\pi\left(n_{\ell}-n_{L\ell}\right):=-\pi\zeta_{\ell}, \tag*{(24)} ϕ ℓ ( 0 + ) = − ϕ ℓ ( 0 − ) = − π ( n ℓ − n L ℓ ) := − π ζ ℓ , (24)
where ζ ℓ \zeta_{\ell} ζ ℓ is n ℓ ∈ [ 0 , 1 ] n_{\ell}\in[0,1] n ℓ ∈ [ 0 , 1 ] if n L ℓ = 0 n_{L\ell}=0 n L ℓ = 0 and n ℓ − 1 ∈ [ − 1 , 0 ] n_{\ell}-1\in[-1,0] n ℓ − 1 ∈ [ − 1 , 0 ] if n L ℓ = 1 n_{L\ell}=1 n L ℓ = 1 , thus ζ ℓ ∈ [ − 1 , 1 ] \zeta_{\ell}\in[-1,1] ζ ℓ ∈ [ − 1 , 1 ] is generally not quantized. However,
I L = − 1 π Tr ( Arg ( X ( i 0 + ) ) ) = − 1 π ∑ ℓ ϕ ℓ ( 0 + ) = ∑ ℓ ζ ℓ ≡ N − N L : = Z ∈ Z (25) I_{L}=-\frac{1}{\pi}\operatorname{Tr}\bigl(\operatorname{Arg}(X(i0^{+}))\bigr)=-\frac{1}{\pi}\sum_{\ell}\phi_{\ell}(0^{+})=\sum_{\ell}\zeta_{\ell}\equiv N-N_{L}:=Z\in\mathbb{Z} \tag*{(25)} I L = − π 1 Tr ( Arg ( X ( i 0 + )) ) = − π 1 ℓ ∑ ϕ ℓ ( 0 + ) = ℓ ∑ ζ ℓ ≡ N − N L := Z ∈ Z (25)
is instead quantized in integer values since both N N N and N L N_{L} N L are integers.
Connections with the Luttinger-Ward functional
We can delve deeper into the previous results to finally pinpoint where the proof of Luttinger’s theorem encounters a flaw. We note that the expression (23) of I L I_{L} I L , together with the properties of the LW functional Φ [ G ] \Phi[G] Φ [ G ] , also implies that
I L = ∫ − ∞ ∞ d ϵ 2 π Tr ( G ( i ϵ ) ∂ Σ ( i ϵ ) ∂ i ϵ ) = − ∫ − ∞ ∞ d ϵ 2 π Tr ( δ Φ [ G ] δ G ( i ϵ ) ∂ G ( i ϵ ) ∂ i ϵ ) ≡ ∫ − ∞ ∞ d ϵ 2 π ∂ ∂ i ϵ Tr ( ln X ( i ϵ ) ) = ∫ − ∞ ∞ d ϵ 2 π ∂ ∂ ϵ Im Tr ( ln X ( i ϵ ) ) , (26) \begin{aligned}
I_{L}&=\int_{-\infty}^{\infty}\frac{\mathrm{d}\epsilon}{2\pi}\operatorname{Tr}\left(G(i\epsilon)\frac{\partial\Sigma(i\epsilon)}{\partial i\epsilon}\right)=-\int_{-\infty}^{\infty}\frac{\mathrm{d}\epsilon}{2\pi}\operatorname{Tr}\left(\frac{\delta\Phi[G]}{\delta G(i\epsilon)}\frac{\partial G(i\epsilon)}{\partial i\epsilon}\right)\\
&\equiv\int_{-\infty}^{\infty}\frac{\mathrm{d}\epsilon}{2\pi}\frac{\partial}{\partial i\epsilon}\operatorname{Tr}\bigl(\ln X(i\epsilon)\bigr)=\int_{-\infty}^{\infty}\frac{\mathrm{d}\epsilon}{2\pi}\frac{\partial}{\partial\epsilon}\operatorname{Im}\operatorname{Tr}\bigl(\ln X(i\epsilon)\bigr), \tag*{(26)}
\end{aligned} I L = ∫ − ∞ ∞ 2 π d ϵ Tr ( G ( i ϵ ) ∂ i ϵ ∂ Σ ( i ϵ ) ) = − ∫ − ∞ ∞ 2 π d ϵ Tr ( δ G ( i ϵ ) δ Φ [ G ] ∂ i ϵ ∂ G ( i ϵ ) ) ≡ ∫ − ∞ ∞ 2 π d ϵ ∂ i ϵ ∂ Tr ( ln X ( i ϵ ) ) = ∫ − ∞ ∞ 2 π d ϵ ∂ ϵ ∂ Im Tr ( ln X ( i ϵ ) ) , (26)
where, as we earlier mentioned, the regularizing exponential can be safely ignored in all expressions. It is then quite plausible that the Luttinger-Ward functional Φ [ G ] \Phi[G] Φ [ G ] entails terms like − Φ L [ X ] -\Phi_{L}[X] − Φ L [ X ] , where, at T = 0 T=0 T = 0 ,
Φ L [ X ] : = ∫ − ∞ ∞ d ϵ 2 π Tr ( ln X ( i ϵ ) ) = ∫ − ∞ ∞ d ϵ 2 π Re Tr ( ln X ( i ϵ ) ) , (27) \Phi_{L}[X]:=\int_{-\infty}^{\infty}\frac{\mathrm{d}\epsilon}{2\pi}\operatorname{Tr}\bigl(\ln X(i\epsilon)\bigr)=\int_{-\infty}^{\infty}\frac{\mathrm{d}\epsilon}{2\pi}\operatorname{Re}\operatorname{Tr}\bigl(\ln X(i\epsilon)\bigr), \tag*{(27)} Φ L [ X ] := ∫ − ∞ ∞ 2 π d ϵ Tr ( ln X ( i ϵ ) ) = ∫ − ∞ ∞ 2 π d ϵ Re Tr ( ln X ( i ϵ ) ) , (27)
which is the only term that contributes to I L I_{L} I L . In other words, Φ L [ X ] \Phi_{L}[X] Φ L [ X ] , which we shall regard as a functional of X ( i ϵ ) X(i\epsilon) X ( i ϵ ) that is in turn a functional of G ( i ϵ ) G(i\epsilon) G ( i ϵ ) , determines whether Luttinger’s theorem is valid or violated. We further observe that I L I_{L} I L in (26) actually measures the monodromy of Im Tr ( ln X ( i ϵ ) ) = − Im Tr ( ln X ( − i ϵ ) ) \operatorname{Im}\operatorname{Tr}\bigl(\ln X(i\epsilon)\bigr)=-\operatorname{Im}\operatorname{Tr}\bigl(\ln X(-i\epsilon)\bigr) Im Tr ( ln X ( i ϵ ) ) = − Im Tr ( ln X ( − i ϵ ) ) . This is not the monodromy of Φ L [ X ] \Phi_{L}[X] Φ L [ X ] in (27), which is only contributed by Re Tr ( ln X ( i ϵ ) ) = Re Tr ( ln X ( − i ϵ ) ) \operatorname{Re}\operatorname{Tr}\bigl(\ln X(i\epsilon)\bigr)=\operatorname{Re}\operatorname{Tr}\bigl(\ln X(-i\epsilon)\bigr) Re Tr ( ln X ( i ϵ ) ) = Re Tr ( ln X ( − i ϵ ) ) . By the definition (22), X ( i ϵ ) X(i\epsilon) X ( i ϵ ) has the correct asymptotic behavior to make the integral in (27) absolutely convergent1 , which guarantees the continuous translational invariance
∫ − ∞ ∞ d ϵ 2 π Tr ( ln X ( i ϵ + i ω ) ) − ∫ − ∞ ∞ d ϵ 2 π Tr ( ln X ( i ϵ ) ) = 0. (28) \int_{-\infty}^{\infty} \frac{d\epsilon}{2\pi}\operatorname{Tr}\left(\ln X(i\epsilon+i\omega)\right)-\int_{-\infty}^{\infty}\frac{d\epsilon}{2\pi}\operatorname{Tr}\left(\ln X(i\epsilon)\right)=0. \tag*{(28)} ∫ − ∞ ∞ 2 π d ϵ Tr ( ln X ( i ϵ + iω ) ) − ∫ − ∞ ∞ 2 π d ϵ Tr ( ln X ( i ϵ ) ) = 0. (28)
Paradox of the functional chain rule and flaw in theorem's proof
Since frequency translational invariance (28) holds,
∂ Φ L [ X ( i ϵ + i ω ) ] ∂ i ω ≡ 0. \frac{\partial\Phi_L[X(i\epsilon+i\omega)]}{\partial i\omega}\equiv0. ∂ iω ∂ Φ L [ X ( i ϵ + iω )] ≡ 0.
Computing the naïve functional derivative at ω = 0 \omega=0 ω = 0 leads to
0 ≡ ∂ Φ L [ X ( i ϵ + i ω ) ] ∂ i ω ∣ ω = 0 = ? ∫ − ∞ ∞ d ϵ 2 π ∂ ∂ i ϵ Tr ( ln X ( i ϵ ) ) ≡ I L , (29) 0\equiv\left.\frac{\partial\Phi_L[X(i\epsilon+i\omega)]}{\partial i\omega}\right|_{\omega=0}\overset{?}{=}\int_{-\infty}^{\infty}\frac{d\epsilon}{2\pi}\frac{\partial}{\partial i\epsilon}\operatorname{Tr}\left(\ln X(i\epsilon)\right)\equiv I_L, \tag*{(29)} 0 ≡ ∂ iω ∂ Φ L [ X ( i ϵ + iω )] ω = 0 = ? ∫ − ∞ ∞ 2 π d ϵ ∂ i ϵ ∂ Tr ( ln X ( i ϵ ) ) ≡ I L , (29)
which is again Luttinger’s proof of the theorem [1 ] . However, (29) contradicts the correct result (18). The paradox is solved by noticing that the Im Tr ( ln X ( i ϵ + i ω ) ) \operatorname{Im}\operatorname{Tr}(\ln X(i\epsilon+i\omega)) Im Tr ( ln X ( i ϵ + iω )) has a jump − 2 π Z -2\pi Z − 2 π Z at ϵ = − ω \epsilon=-\omega ϵ = − ω , which implies that
∂ Φ L [ X ( i ϵ + i ω ) ] ∂ i ω = ∂ ∂ i ω { ∫ − ∞ − ω − 0 d ϵ 2 π Tr ( ln X ( i ϵ + i ω ) ) + ∫ − ω + 0 ∞ d ϵ 2 π Tr ( ln X ( i ϵ + i ω ) ) } = 1 2 π [ Im Tr ( ln X ( i 0 + ) ) − Im Tr ( ln X ( − i 0 + ) ) ] + ∫ − ∞ ∞ d ϵ 2 π ∂ ∂ i ϵ Tr ( ln X ( i ϵ + i ω ) ) ⟶ ω → 0 1 2 π [ Im Tr ( ln X ( i 0 + ) ) − Im Tr ( ln X ( − i 0 + ) ) ] + I L = 0 , (30) \begin{aligned}
\frac{\partial\Phi_L[X(i\epsilon+i\omega)]}{\partial i\omega}
={}&\frac{\partial}{\partial i\omega}\left\{\int_{-\infty}^{-\omega-0}\frac{d\epsilon}{2\pi}\operatorname{Tr}\left(\ln X(i\epsilon+i\omega)\right)\right.\\
&\left.+\int_{-\omega+0}^{\infty}\frac{d\epsilon}{2\pi}\operatorname{Tr}\left(\ln X(i\epsilon+i\omega)\right)\right\}\\
={}&\frac{1}{2\pi}\left[\operatorname{Im}\operatorname{Tr}\left(\ln X(i0^+)\right)-\operatorname{Im}\operatorname{Tr}\left(\ln X(-i0^+)\right)\right]\\
&+\int_{-\infty}^{\infty}\frac{d\epsilon}{2\pi}\frac{\partial}{\partial i\epsilon}\operatorname{Tr}\left(\ln X(i\epsilon+i\omega)\right)\\
\underset{\omega\to0}{\longrightarrow}{}&\frac{1}{2\pi}\left[\operatorname{Im}\operatorname{Tr}\left(\ln X(i0^+)\right)-\operatorname{Im}\operatorname{Tr}\left(\ln X(-i0^+)\right)\right]+I_L=0, \tag*{(30)}
\end{aligned} ∂ iω ∂ Φ L [ X ( i ϵ + iω )] = = ω → 0 ⟶ ∂ iω ∂ { ∫ − ∞ − ω − 0 2 π d ϵ Tr ( ln X ( i ϵ + iω ) ) + ∫ − ω + 0 ∞ 2 π d ϵ Tr ( ln X ( i ϵ + iω ) ) } 2 π 1 [ Im Tr ( ln X ( i 0 + ) ) − Im Tr ( ln X ( − i 0 + ) ) ] + ∫ − ∞ ∞ 2 π d ϵ ∂ i ϵ ∂ Tr ( ln X ( i ϵ + iω ) ) 2 π 1 [ Im Tr ( ln X ( i 0 + ) ) − Im Tr ( ln X ( − i 0 + ) ) ] + I L = 0 , (30)
in perfect agreement with Eq. (28) . This result is equivalent to assuming that
∂ Φ L [ X ( i ϵ + i ω ) ] ∂ i ω ∣ ω = 0 ≡ ∫ − ∞ ∞ d ϵ 2 π { Tr [ X ( i ϵ ) − 1 ∂ i ϵ X ( i ϵ ) ] r e g − 2 π Z δ ( ϵ ) } . (31) \left.\frac{\partial\Phi_L[X(i\epsilon+i\omega)]}{\partial i\omega}\right|_{\omega=0}\equiv\int_{-\infty}^{\infty}\frac{d\epsilon}{2\pi}\left\{\operatorname{Tr}\left[X(i\epsilon)^{-1}\partial_{i\epsilon}X(i\epsilon)\right]_{\mathrm{reg}}-2\pi Z\delta(\epsilon)\right\}. \tag*{(31)} ∂ iω ∂ Φ L [ X ( i ϵ + iω )] ω = 0 ≡ ∫ − ∞ ∞ 2 π d ϵ { Tr [ X ( i ϵ ) − 1 ∂ i ϵ X ( i ϵ ) ] reg − 2 π Z δ ( ϵ ) } . (31)
The regular piece integrates exactly to Z Z Z , which is flawlessly canceled by the δ \delta δ -function weight. This is the very element that was missing in Luttinger’s proof of the theorem [1 ] . As we demonstrated, the Luttinger integral is a pure boundary term, so the existence of the δ \delta δ -function anomaly cannot be ruled out a priori. In fact, it is essential to restore frequency translational invariance whenever I L ≠ 0 I_L\ne0 I L = 0 .
G ( i ϵ ) ≃ G ∗ ( i ϵ ) ≃ 1 i ϵ − H H F ≃ 1 i ϵ + ( i ϵ ) − 2 H H F + O ( ϵ − 3 ) , G(i\epsilon)\simeq G_*(i\epsilon)\simeq\frac{1}{i\epsilon-H_{\mathrm{HF}}}\simeq\frac{1}{i\epsilon}+(i\epsilon)^{-2}H_{\mathrm{HF}}+O(\epsilon^{-3}), G ( i ϵ ) ≃ G ∗ ( i ϵ ) ≃ i ϵ − H HF 1 ≃ i ϵ 1 + ( i ϵ ) − 2 H HF + O ( ϵ − 3 ) ,
where H H F ≡ H ∗ ( ϵ → ± ∞ ) H_{\mathrm{HF}}\equiv H_*(\epsilon\to\pm\infty) H HF ≡ H ∗ ( ϵ → ± ∞ ) is the Hartree-Fock Hamiltonian. Therefore,
ln S ℓ ( i ϵ ) − [ ln G ∗ ( i ϵ ) − 1 ] ℓ ℓ ≃ ( ln i ϵ − 1 i ϵ [ H H F ] ℓ ℓ ) − [ ln ( i ϵ − H H F ) ] ℓ ℓ ≃ O ( ϵ − 2 ) . \ln S_\ell(i\epsilon)-\left[\ln G_*(i\epsilon)^{-1}\right]_{\ell\ell}\simeq\left(\ln i\epsilon-\frac{1}{i\epsilon}[H_{\mathrm{HF}}]_{\ell\ell}\right)-\left[\ln(i\epsilon-H_{\mathrm{HF}})\right]_{\ell\ell}\simeq O(\epsilon^{-2}). ln S ℓ ( i ϵ ) − [ ln G ∗ ( i ϵ ) − 1 ] ℓℓ ≃ ( ln i ϵ − i ϵ 1 [ H HF ] ℓℓ ) − [ ln ( i ϵ − H HF ) ] ℓℓ ≃ O ( ϵ − 2 ) .
At T ≠ 0 T \ne0 T = 0 , the dangerous part of the Luttinger-Ward functional is
− Φ L [ X ( i ϵ n ) ] = − T ∑ n Tr ( ln X ( i ϵ n ) ) . -\Phi_L[X(i\epsilon_n)] = -T \sum_n \operatorname{Tr}\left(\ln X(i\epsilon_n)\right). − Φ L [ X ( i ϵ n )] = − T n ∑ Tr ( ln X ( i ϵ n ) ) .
Translational invariance is trivially satisfied:
Φ L [ X ( i ϵ n + i ω m ) ] − Φ L [ X ( i ϵ n ) ] ≡ 0 , ω m = 2 m π T . \Phi_L[X(i\epsilon_n+i\omega_m)]-\Phi_L[X(i\epsilon_n)] \equiv0,\qquad\omega_m=2m\pi T. Φ L [ X ( i ϵ n + i ω m )] − Φ L [ X ( i ϵ n )] ≡ 0 , ω m = 2 mπ T .
For T → 0 T \to0 T → 0 , thus ω m → 0 \omega_m \to0 ω m → 0 , it is tempting to make a Taylor expansion in
Δ X ( i ϵ n ; i ω m ) = X ( i ϵ n + i ω m ) − X ( i ϵ n ) . (32) \Delta X(i\epsilon_n;i\omega_m)=X(i\epsilon_n+i\omega_m)-X(i\epsilon_n). \tag*{(32)} Δ X ( i ϵ n ; i ω m ) = X ( i ϵ n + i ω m ) − X ( i ϵ n ) . (32)
Naïvely, one would argue that Δ X → 0 \Delta X \to0 Δ X → 0 for T → 0 T \to0 T → 0 and conclude that
0 ≡ lim T → 0 Φ L [ X ( i ϵ n + i ω m ) ] − Φ L [ X ( i ϵ n ) ] i ω m 0 \equiv\lim_{T\to0}\frac{\Phi_L[X(i\epsilon_n+i\omega_m)]-\Phi_L[X(i\epsilon_n)]}{i\omega_m} 0 ≡ T → 0 lim i ω m Φ L [ X ( i ϵ n + i ω m )] − Φ L [ X ( i ϵ n )]
≡ ? lim T → 0 T i ω m ∑ n ∑ k = 1 ∞ ( − 1 ) k − 1 k Tr { ( X ( i ϵ n ) − 1 ⋅ Δ X ( i ϵ n ; i ω m ) ) k } (33) \mathrel{\stackrel{?}{\equiv}} \lim_{T\to0}\frac{T}{i\omega_m}\sum_n\sum_{k=1}^{\infty}\frac{(-1)^{k-1}}{k}\operatorname{Tr}\left\{\left(X(i\epsilon_n)^{-1}\cdot\Delta X(i\epsilon_n;i\omega_m)\right)^k\right\} \tag*{(33)} ≡ ? T → 0 lim i ω m T n ∑ k = 1 ∑ ∞ k ( − 1 ) k − 1 Tr { ( X ( i ϵ n ) − 1 ⋅ Δ X ( i ϵ n ; i ω m ) ) k } (33)
≡ ? lim T → 0 T ∑ n Tr ( X ( i ϵ n ) − 1 ⋅ Δ X ( i ϵ n ; i ω m ) i ω m ) , \mathrel{\stackrel{?}{\equiv}} \lim_{T\to0}T\sum_n\operatorname{Tr}\left(X(i\epsilon_n)^{-1}\cdot\frac{\Delta X(i\epsilon_n;i\omega_m)}{i\omega_m}\right), ≡ ? T → 0 lim T n ∑ Tr ( X ( i ϵ n ) − 1 ⋅ i ω m Δ X ( i ϵ n ; i ω m ) ) ,
which recovers only the regular part in (31) and thus violates frequency-translational invariance. There are several critical steps in (33). First, the Taylor expansion in (33) derives formally from
Tr ( ln X ( i ϵ n + i ω m ) ) = Tr { ln [ ( 1 + Δ X ( i ϵ n ; i ω m ) ⋅ X ( i ϵ n ) − 1 ) ⋅ X ( i ϵ n ) ] } = Tr { ln ( 1 + Δ X ( i ϵ n ; i ω m ) ⋅ X ( i ϵ n ) − 1 ) } + Tr ( ln X ( i ϵ n ) ) , \begin{aligned}
\operatorname{Tr}\left(\ln X(i\epsilon_n+i\omega_m)\right)=\operatorname{Tr}\left\{\ln\left[\left(1+\Delta X(i\epsilon_n;i\omega_m)\cdot X(i\epsilon_n)^{-1}\right)\cdot X(i\epsilon_n)\right]\right\} \\
&=\operatorname{Tr}\left\{\ln\left(1+\Delta X(i\epsilon_n;i\omega_m)\cdot X(i\epsilon_n)^{-1}\right)\right\}+\operatorname{Tr}\left(\ln X(i\epsilon_n)\right),
\end{aligned} Tr ( ln X ( i ϵ n + i ω m ) ) = Tr { ln [ ( 1 + Δ X ( i ϵ n ; i ω m ) ⋅ X ( i ϵ n ) − 1 ) ⋅ X ( i ϵ n ) ] } = Tr { ln ( 1 + Δ X ( i ϵ n ; i ω m ) ⋅ X ( i ϵ n ) − 1 ) } + Tr ( ln X ( i ϵ n ) ) ,
which might not hold under the prescription of a fixed branch cut. In addition, the further expansion in powers of Δ X / X \Delta X/X Δ X / X assumes that Δ X ( i ϵ n ; i ω m ) → 0 \Delta X(i\epsilon_n;i\omega_m)\to0 Δ X ( i ϵ n ; i ω m ) → 0 for T → 0 T\to0 T → 0 , which may not always occur, in particular when ϵ n ( ϵ n + ω m ) < 0 \epsilon_n(\epsilon_n+\omega_m)<0 ϵ n ( ϵ n + ω m ) < 0 .
In reality,
0 ≡ Φ L [ X ( i ϵ n + i ω m ) ] − Φ L [ X ( i ϵ n ) ] i ω m = T i ω m ∑ n { Tr ( ln X ( i ϵ n + i ω m ) ) − Tr ( ln X ( i ϵ n ) ) } = T i ω m ∑ n θ ( − ϵ n ( ϵ n + ω m ) ) { Tr ( ln X ( i ϵ n + i ω m ) ) − Tr ( ln X ( i ϵ n ) ) } + T i ω m ∑ n θ ( ϵ n ( ϵ n + ω m ) ) { Tr ( ln X ( i ϵ n + i ω m ) ) − Tr ( ln X ( i ϵ n ) ) } . (34) \begin{aligned}
0\equiv\frac{\Phi_L[X(i\epsilon_n+i\omega_m)]-\Phi_L[X(i\epsilon_n)]}{i\omega_m} \\
&=\frac{T}{i\omega_m}\sum_n\left\{\operatorname{Tr}\left(\ln X(i\epsilon_n+i\omega_m)\right)-\operatorname{Tr}\left(\ln X(i\epsilon_n)\right)\right\} \\
&=\frac{T}{i\omega_m}\sum_n\theta\left(-\epsilon_n(\epsilon_n+\omega_m)\right)\left\{\operatorname{Tr}\left(\ln X(i\epsilon_n+i\omega_m)\right)-\operatorname{Tr}\left(\ln X(i\epsilon_n)\right)\right\} \\
&+\frac{T}{i\omega_m}\sum_n\theta\left(\epsilon_n(\epsilon_n+\omega_m)\right)\left\{\operatorname{Tr}\left(\ln X(i\epsilon_n+i\omega_m)\right)-\operatorname{Tr}\left(\ln X(i\epsilon_n)\right)\right\}. \tag*{(34)}
\end{aligned} 0 ≡ i ω m Φ L [ X ( i ϵ n + i ω m )] − Φ L [ X ( i ϵ n )] = i ω m T n ∑ { Tr ( ln X ( i ϵ n + i ω m ) ) − Tr ( ln X ( i ϵ n ) ) } = i ω m T n ∑ θ ( − ϵ n ( ϵ n + ω m ) ) { Tr ( ln X ( i ϵ n + i ω m ) ) − Tr ( ln X ( i ϵ n ) ) } + i ω m T n ∑ θ ( ϵ n ( ϵ n + ω m ) ) { Tr ( ln X ( i ϵ n + i ω m ) ) − Tr ( ln X ( i ϵ n ) ) } . (34)
The first term in (34) can be written, assuming ω m > 0 \omega_m>0 ω m > 0 and ϵ n = ( 2 n + 1 ) π T \epsilon_n=(2n+1)\pi T ϵ n = ( 2 n + 1 ) π T , as
1 2 π i m ∑ n = 0 m − 1 { Tr ( ln X ( i ϵ n ) ) − Tr ( ln X ( − i ϵ n ) ) } , \frac{1}{2\pi i m}\sum_{n=0}^{m-1}\left\{\operatorname{Tr}\left(\ln X(i\epsilon_n)\right)-\operatorname{Tr}\left(\ln X(-i\epsilon_n)\right)\right\}, 2 π im 1 n = 0 ∑ m − 1 { Tr ( ln X ( i ϵ n ) ) − Tr ( ln X ( − i ϵ n ) ) } ,
which, for T → 0 T\to0 T → 0 at fixed m m m , yields the anomaly term in (31). The second term in (34) can be safely expanded in Δ X \Delta X Δ X and gives the regular term in (31).
Final remarks
Now, we recall that X X X is itself a functional of G G G ; therefore, the fact that
I L = ∫ − ∞ ∞ d ϵ 2 π Tr ( G ( i ϵ ) ∂ Σ ( i ϵ ) ∂ i ϵ ) = Z ≠ 0 , I_L=\int_{-\infty}^{\infty}\frac{\mathrm{d}\epsilon}{2\pi}\operatorname{Tr}\left(G(i\epsilon)\frac{\partial\Sigma(i\epsilon)}{\partial i\epsilon}\right)=Z\ne0, I L = ∫ − ∞ ∞ 2 π d ϵ Tr ( G ( i ϵ ) ∂ i ϵ ∂ Σ ( i ϵ ) ) = Z = 0 ,
does not imply that the Luttinger-Ward functional loses its invariance under a shift of the internal frequency, which is merely a consequence of the particle number conservation, or that the skeleton-diagram formalism [45 ] breaks down when correlations are strong. It only means that the formal Taylor expansion is incorrect unless one adds explicitly an anomaly term, i.e., a δ \delta δ -function in ϵ \epsilon ϵ , see (31) , which enforces the invariance.
The conjecture that the integrand in I L I_L I L contains a term which is the derivative of Im Tr ln X ( i ϵ ) \operatorname{Im}\operatorname{Tr}\ln X(i\epsilon) Im Tr ln X ( i ϵ ) , with X ( i ϵ ) X(i\epsilon) X ( i ϵ ) in (22) , suggests that the regular part of such derivative behaves as the difference between two Lorentzian functions. Consider, as an illustrative example, the function
F ( ϵ ) = γ 1 ϵ 2 + γ 1 2 − γ 2 ϵ 2 + γ 2 2 , F(\epsilon)=\frac{\gamma_1}{\epsilon^2+\gamma_1^2}-\frac{\gamma_2}{\epsilon^2+\gamma_2^2}, F ( ϵ ) = ϵ 2 + γ 1 2 γ 1 − ϵ 2 + γ 2 2 γ 2 ,
whose integral vanishes only if γ 1 γ 2 > 0 \gamma_1\gamma_2>0 γ 1 γ 2 > 0 , corresponding to the regime of validity of Luttinger’s theorem. Once γ 1 γ 2 < 0 \gamma_1\gamma_2<0 γ 1 γ 2 < 0 , Luttinger’s theorem is violated even though F ( ϵ ) F(\epsilon) F ( ϵ ) does not exhibit any singularity. In other words, it is impossible to detect any indication of Luttinger’s theorem violation from the local behavior of F ( ϵ ) F(\epsilon) F ( ϵ ) or, equivalently, of G ( i ϵ ) ∂ i ϵ Σ ( i ϵ ) G(i\epsilon)\partial_{i\epsilon}\Sigma(i\epsilon) G ( i ϵ ) ∂ i ϵ Σ ( i ϵ ) , unless right at the point when the theorem ceases to be valid, which is when either γ 1 \gamma_1 γ 1 or γ 2 \gamma_2 γ 2 vanishes.
We also remark that Eq. (26) does not imply that G ( i ϵ ) ∂ i ϵ Σ ( i ϵ ) G(i\epsilon)\partial_{i\epsilon}\Sigma(i\epsilon) G ( i ϵ ) ∂ i ϵ Σ ( i ϵ ) is directly the derivative of a logarithm. Instead, it suggests that several diagonal elements conspire to reconstruct the derivative of a logarithm.
We conclude by briefly discussing Refs. [2 ] and [21 ] , which also claim that I L I_L I L is a boundary term. The authors of Ref. [2 ] assume that I L = N − N L I_L=N-N_L I L = N − N L is equivalent to N 0 − N L N_0-N_L N 0 − N L , where N 0 N_0 N 0 is the electron number in the non-interacting ground state. However, the assumption N = N 0 N=N_0 N = N 0 is generally not true at a chemical potential common to the interacting and non-interacting systems. By contrast, in an earlier work by one of the authors [21 ] it was concluded that I L I_L I L is a boundary term from the conjecture that the limit T → 0 T\to0 T → 0 in (9) is singular. Specifically, Ref. [21 ] takes for granted that Eq. (7) implies
0 ≡ lim T → 0 Φ [ G ( i ϵ n ) + Δ G ( i ϵ n ; i ω m ) ] − Φ [ G ( i ϵ n ) ] i ω m 0\equiv\lim_{T\to0}\frac{\Phi[G(i\epsilon_n)+\Delta G(i\epsilon_n;i\omega_m)]-\Phi[G(i\epsilon_n)]}{i\omega_m} 0 ≡ T → 0 lim i ω m Φ [ G ( i ϵ n ) + Δ G ( i ϵ n ; i ω m )] − Φ [ G ( i ϵ n )]
≡ ? lim T → 0 T ∑ n Tr ( Σ ( i ϵ n ) Δ G ( i ϵ n ; i ω m ) i ω m ) (35) \overset{?}{\equiv}\lim_{T\to0}T\sum_n\operatorname{Tr}\left(\Sigma(i\epsilon_n)\frac{\Delta G(i\epsilon_n;i\omega_m)}{i\omega_m}\right) \tag*{(35)} ≡ ? T → 0 lim T n ∑ Tr ( Σ ( i ϵ n ) i ω m Δ G ( i ϵ n ; i ω m ) ) (35)
i.e. that the functional chain rule applies at T ≠ 0 T\ne0 T = 0 , but argues that the limit T → 0 T\to0 T → 0 does not correspond to − I L -I_L − I L . As we discussed in Sec. 2.4, this is only one piece of the puzzle, specifically the fact that Δ X \Delta X Δ X in (32) might not vanish for T → 0 T\to0 T → 0 , thus invalidating the Taylor expansion. The other piece of the puzzle is that the functional chain rule by itself may miss an anomaly that is crucial to enforce frequency-translation invariance. We also mention that a pathology in the T → 0 T\to0 T → 0 limit was originally proposed by Potthoff [44 ] as an explanation of Luttinger’s theorem violation. Our proof that I L I_L I L is a boundary term is instead exact and does not rely on any a priori assumption.
Luttinger's theorem and the Streda formula
In this section we elaborate on the interesting results by Gavensky, Sachdev and Goldman [5 ] and extend them further to elucidate what the violation of Luttinger’s theorem actually implies in a topological correlated insulator.
Since translational symmetry is important, we assume here, without loss of generality, a basis of M M M topologically trivial Bloch wavefunctions ϕ ℓ k ( r ) \phi_{\ell\mathbf{k}}(\mathbf{r}) ϕ ℓ k ( r ) (ℓ = 1 , … , M \ell= 1,\ldots,M ℓ = 1 , … , M ). Therefore, the non-trivial topology is entirely encoded in the Green’s function G ( i ϵ , k ) G(i\epsilon,\mathbf{k}) G ( i ϵ , k ) , an M × M M \times M M × M complex matrix for each momentum k \mathbf{k} k , and in the functional derivatives of the LW functional.
An important role hereafter will be played by the irreducible vertex in the particle-hole channel with transferred imaginary frequency i ω i\omega iω and momentum q \mathbf{q} q . This is defined through
δ 2 Φ [ G ] δ G m ℓ ( i ϵ , k ; i ϵ + i ω , k + q ) δ G m ′ ℓ ′ ( i ϵ ′ + i ω , k ′ + q ; i ϵ ′ , k ′ ) = Γ 0 ( i ϵ + i ω , k + q , ℓ ; i ϵ ′ , k ′ , ℓ ′ ∣ i ϵ ′ + i ω , k ′ + q , m ′ ; i ϵ , k , m ) , (36) \frac{\delta^{2}\Phi[G]}{\delta G_{m\ell}(i\epsilon,\mathbf{k};i\epsilon+i\omega,\mathbf{k}+\mathbf{q})\delta G_{m'\ell'}(i\epsilon'+i\omega,\mathbf{k}'+\mathbf{q};i\epsilon',\mathbf{k}')} = \Gamma_{0}(i\epsilon+i\omega,\mathbf{k}+\mathbf{q},\ell;i\epsilon',\mathbf{k}',\ell'|i\epsilon'+i\omega,\mathbf{k}'+\mathbf{q},m';i\epsilon,\mathbf{k},m), \tag*{(36)} δ G m ℓ ( i ϵ , k ; i ϵ + iω , k + q ) δ G m ′ ℓ ′ ( i ϵ ′ + iω , k ′ + q ; i ϵ ′ , k ′ ) δ 2 Φ [ G ] = Γ 0 ( i ϵ + iω , k + q , ℓ ; i ϵ ′ , k ′ , ℓ ′ ∣ i ϵ ′ + iω , k ′ + q , m ′ ; i ϵ , k , m ) , (36)
where ℓ \ell ℓ and ℓ ′ \ell' ℓ ′ identify the incoming legs, while m m m and m ′ m' m ′ the outgoing ones. The vertex satisfies the following identities:
Γ 0 ( i ϵ + i ω , k + q , ℓ ; i ϵ ′ , k ′ , ℓ ′ ∣ i ϵ ′ + i ω , k ′ + q , m ′ ; i ϵ , k , m ) = Γ 0 ( i ϵ ′ , k ′ , ℓ ′ ; i ϵ + i ω , k + q , ℓ ∣ i ϵ , k , m ; i ϵ ′ + i ω , k ′ + q , m ′ ) , Γ 0 ( i ϵ + i ω , k + q , ℓ ; i ϵ ′ , k ′ , ℓ ′ ∣ i ϵ ′ + i ω , k ′ + q , m ′ ; i ϵ , k , m ) ∗ = Γ 0 ( − i ϵ , k , m ; − i ϵ ′ − i ω , k ′ + q , m ′ ∣ − i ϵ ′ , k ′ , ℓ ′ ; − i ϵ − i ω , k + q , ℓ ) . (37) \begin{aligned}
\Gamma_{0}(i\epsilon+i\omega,\mathbf{k}+\mathbf{q},\ell;i\epsilon',\mathbf{k}',\ell'|i\epsilon'+i\omega,\mathbf{k}'+\mathbf{q},m';i\epsilon,\mathbf{k},m)
&= \Gamma_{0}(i\epsilon',\mathbf{k}',\ell';i\epsilon+i\omega,\mathbf{k}+\mathbf{q},\ell|i\epsilon,\mathbf{k},m;i\epsilon'+i\omega,\mathbf{k}'+\mathbf{q},m'),\\
\Gamma_{0}(i\epsilon+i\omega,\mathbf{k}+\mathbf{q},\ell;i\epsilon',\mathbf{k}',\ell'|i\epsilon'+i\omega,\mathbf{k}'+\mathbf{q},m';i\epsilon,\mathbf{k},m)^{*}
&= \Gamma_{0}(-i\epsilon,\mathbf{k},m;-i\epsilon'-i\omega,\mathbf{k}'+\mathbf{q},m'|-i\epsilon',\mathbf{k}',\ell';-i\epsilon-i\omega,\mathbf{k}+\mathbf{q},\ell). \tag*{(37)}
\end{aligned} Γ 0 ( i ϵ + iω , k + q , ℓ ; i ϵ ′ , k ′ , ℓ ′ ∣ i ϵ ′ + iω , k ′ + q , m ′ ; i ϵ , k , m ) Γ 0 ( i ϵ + iω , k + q , ℓ ; i ϵ ′ , k ′ , ℓ ′ ∣ i ϵ ′ + iω , k ′ + q , m ′ ; i ϵ , k , m ) ∗ = Γ 0 ( i ϵ ′ , k ′ , ℓ ′ ; i ϵ + iω , k + q , ℓ ∣ i ϵ , k , m ; i ϵ ′ + iω , k ′ + q , m ′ ) , = Γ 0 ( − i ϵ , k , m ; − i ϵ ′ − iω , k ′ + q , m ′ ∣ − i ϵ ′ , k ′ , ℓ ′ ; − i ϵ − iω , k + q , ℓ ) . (37)
Moreover, if the system is invariant under time-reversal symmetry T \mathcal{T} T , such that T ( ϕ ℓ k ( r ) ) = ϕ ℓ ˉ − k ( r ) \mathcal{T}(\phi_{\ell\mathbf{k}}(\mathbf{r}))=\phi_{\bar{\ell}-\mathbf{k}}(\mathbf{r}) T ( ϕ ℓ k ( r )) = ϕ ℓ ˉ − k ( r ) , then
Γ 0 ( i ϵ + i ω , k + q , ℓ ; i ϵ ′ , k ′ , ℓ ′ ∣ i ϵ ′ + i ω , k ′ + q , m ′ ; i ϵ , k , m ) ≡ Γ 0 ( − i ϵ − i ω , − k − q , ℓ ˉ ; − i ϵ ′ , − k ′ , ℓ ˉ ′ ∣ − i ϵ ′ − i ω , − k ′ − q , m ˉ ′ ; − i ϵ , − k , m ˉ ) ∗ . (38) \begin{aligned}
\Gamma_{0}(i\epsilon+i\omega,\mathbf{k}+\mathbf{q},\ell;i\epsilon',\mathbf{k}',\ell'|i\epsilon'+i\omega,\mathbf{k}'+\mathbf{q},m';i\epsilon,\mathbf{k},m)
\equiv\Gamma_{0}(-i\epsilon-i\omega,-\mathbf{k}-\mathbf{q},\bar{\ell};-i\epsilon',-\mathbf{k}',\bar{\ell}'|-i\epsilon'-i\omega,-\mathbf{k}'-\mathbf{q},\bar{m}';-i\epsilon,-\mathbf{k},\bar{m})^{*}. \tag*{(38)}
\end{aligned} Γ 0 ( i ϵ + iω , k + q , ℓ ; i ϵ ′ , k ′ , ℓ ′ ∣ i ϵ ′ + iω , k ′ + q , m ′ ; i ϵ , k , m ) ≡ Γ 0 ( − i ϵ − iω , − k − q , ℓ ˉ ; − i ϵ ′ , − k ′ , ℓ ˉ ′ ∣ − i ϵ ′ − iω , − k ′ − q , m ˉ ′ ; − i ϵ , − k , m ˉ ) ∗ . (38)
as well as
G n m ( i ϵ , k ) ≡ G n ˉ m ˉ ( − i ϵ , − k ) ∗ . (39) G_{nm}(i\epsilon,\mathbf{k}) \equiv G_{\bar{n}\bar{m}}(-i\epsilon,-\mathbf{k})^{*}. \tag*{(39)} G nm ( i ϵ , k ) ≡ G n ˉ m ˉ ( − i ϵ , − k ) ∗ . (39)
To simplify the notation, we find it convenient to introduce two four-index tensors, Γ 0 S ( i ϵ , k ; i ϵ ′ , k ′ ∣ i ω , q ) \Gamma_{0}^{S}(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}'|i\omega,\mathbf{q}) Γ 0 S ( i ϵ , k ; i ϵ ′ , k ′ ∣ iω , q ) and Γ 0 A ( i ϵ , k ; i ϵ ′ , k ′ ∣ i ω , q ) \Gamma_{0}^{A}(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}'|i\omega,\mathbf{q}) Γ 0 A ( i ϵ , k ; i ϵ ′ , k ′ ∣ iω , q ) with the following tensor elements
Γ 0 ∣ ℓ m , ℓ ′ m ′ S ( i ϵ , k ; i ϵ ′ , k ′ ∣ i ω , q ) ≔ 1 2 ( Γ 0 ( i ϵ + i ω , k + q , ℓ ; i ϵ ′ , k ′ , ℓ ′ ∣ i ϵ ′ + i ω , k ′ + q , m ′ ; i ϵ , k , m ) + Γ 0 ( i ϵ ′ + i ω , k ′ + q , ℓ ′ ; i ϵ , k , ℓ ∣ i ϵ + i ω , k + q , m ; i ϵ ′ , k ′ , m ′ ) ) , Γ 0 ∣ ℓ m , ℓ ′ m ′ A ( i ϵ , k ; i ϵ ′ , k ′ ∣ i ω , q ) ≔ 1 2 ( Γ 0 ( i ϵ + i ω , k + q , ℓ ; i ϵ ′ , k ′ , ℓ ′ ∣ i ϵ ′ + i ω , k ′ + q , m ′ ; i ϵ , k , m ) − Γ 0 ( i ϵ ′ + i ω , k ′ + q , ℓ ′ ; i ϵ , k , ℓ ∣ i ϵ + i ω , k + q , m ; i ϵ ′ , k ′ , m ′ ) ) . (40) \begin{aligned}
\Gamma_{0|\ell m,\ell' m'}^{S}(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}'|i\omega,\mathbf{q})
&\coloneqq\frac{1}{2}\bigl(\Gamma_{0}(i\epsilon+i\omega,\mathbf{k}+\mathbf{q},\ell;i\epsilon',\mathbf{k}',\ell'|i\epsilon'+i\omega,\mathbf{k}'+\mathbf{q},m';i\epsilon,\mathbf{k},m)\\
&\qquad+\Gamma_{0}(i\epsilon'+i\omega,\mathbf{k}'+\mathbf{q},\ell';i\epsilon,\mathbf{k},\ell|i\epsilon+i\omega,\mathbf{k}+\mathbf{q},m;i\epsilon',\mathbf{k}',m')\bigr),\\
\Gamma_{0|\ell m,\ell' m'}^{A}(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}'|i\omega,\mathbf{q})
&\coloneqq\frac{1}{2}\bigl(\Gamma_{0}(i\epsilon+i\omega,\mathbf{k}+\mathbf{q},\ell;i\epsilon',\mathbf{k}',\ell'|i\epsilon'+i\omega,\mathbf{k}'+\mathbf{q},m';i\epsilon,\mathbf{k},m)\\
&\qquad-\Gamma_{0}(i\epsilon'+i\omega,\mathbf{k}'+\mathbf{q},\ell';i\epsilon,\mathbf{k},\ell|i\epsilon+i\omega,\mathbf{k}+\mathbf{q},m;i\epsilon',\mathbf{k}',m')\bigr). \tag*{(40)}
\end{aligned} Γ 0∣ ℓ m , ℓ ′ m ′ S ( i ϵ , k ; i ϵ ′ , k ′ ∣ iω , q ) Γ 0∣ ℓ m , ℓ ′ m ′ A ( i ϵ , k ; i ϵ ′ , k ′ ∣ iω , q ) : = 2 1 ( Γ 0 ( i ϵ + iω , k + q , ℓ ; i ϵ ′ , k ′ , ℓ ′ ∣ i ϵ ′ + iω , k ′ + q , m ′ ; i ϵ , k , m ) + Γ 0 ( i ϵ ′ + iω , k ′ + q , ℓ ′ ; i ϵ , k , ℓ ∣ i ϵ + iω , k + q , m ; i ϵ ′ , k ′ , m ′ ) ) , : = 2 1 ( Γ 0 ( i ϵ + iω , k + q , ℓ ; i ϵ ′ , k ′ , ℓ ′ ∣ i ϵ ′ + iω , k ′ + q , m ′ ; i ϵ , k , m ) − Γ 0 ( i ϵ ′ + iω , k ′ + q , ℓ ′ ; i ϵ , k , ℓ ∣ i ϵ + iω , k + q , m ; i ϵ ′ , k ′ , m ′ ) ) . (40)
where S S S and A A A stand for symmetric and antisymmetric properties under
( i ϵ , k ) ↔ ( i ϵ ′ , k ′ ) ∧ ( ℓ , m ) ↔ ( ℓ ′ , m ′ ) . (i\epsilon,\mathbf{k}) \leftrightarrow(i\epsilon',\mathbf{k}') \wedge(\ell,m) \leftrightarrow(\ell',m'). ( i ϵ , k ) ↔ ( i ϵ ′ , k ′ ) ∧ ( ℓ , m ) ↔ ( ℓ ′ , m ′ ) .
In addition, we indicate the matrix product by ⋅ \cdot ⋅ and the tensor contraction by ⋅ ¨ \ddot{\cdot} ⋅ ¨ , meaning that
Γ 0 ⋅ ¨ M = ∑ ℓ ′ m ′ Γ 0 ∣ ℓ m , ℓ ′ m ′ M m ′ ℓ ′ \Gamma_0 \ddot{\cdot} M = \sum_{\ell'm'} \Gamma_{0|\ell m,\ell' m'} M_{m'\ell'} Γ 0 ⋅ ¨ M = ℓ ′ m ′ ∑ Γ 0∣ ℓ m , ℓ ′ m ′ M m ′ ℓ ′
is a matrix. Moreover, whenever we need to denote symbolically the tensor contraction and the integral over internal frequencies and momenta, we use the symbol ⊙ \odot ⊙ . We also adopt the following conventions:
Γ 0 S / A ( i ϵ , k ; i ϵ ′ , k ′ ∣ 0 , 0 ) : = Γ 0 S / A ( i ϵ , k ; i ϵ ′ , k ′ ) , \Gamma_0^{S/A}(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}'|0,0) := \Gamma_0^{S/A}(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}'), Γ 0 S / A ( i ϵ , k ; i ϵ ′ , k ′ ∣0 , 0 ) := Γ 0 S / A ( i ϵ , k ; i ϵ ′ , k ′ ) ,
∂ i ω Γ 0 S / A ( i ϵ , k ; i ϵ ′ , k ′ ∣ i ω , q ) ∣ ω = q = 0 : = Γ ˙ 0 S / A ( i ϵ , k ; i ϵ ′ , k ′ ) , \left.\partial_{i\omega}\Gamma_0^{S/A}(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}'|i\omega,\mathbf{q})\right|_{\omega=\mathbf{q}=0} := \dot{\Gamma}_0^{S/A}(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}'), ∂ iω Γ 0 S / A ( i ϵ , k ; i ϵ ′ , k ′ ∣ iω , q ) ω = q = 0 := Γ ˙ 0 S / A ( i ϵ , k ; i ϵ ′ , k ′ ) ,
∂ q μ Γ 0 S / A ( i ϵ , k ; i ϵ ′ , k ′ ∣ i ω , q ) ∣ ω = q = 0 : = Γ 0 μ S / A ( i ϵ , k ; i ϵ ′ , k ′ ) . \left.\partial_{q_\mu}\Gamma_0^{S/A}(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}'|i\omega,\mathbf{q})\right|_{\omega=\mathbf{q}=0} := \Gamma_{0\mu}^{S/A}(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}'). ∂ q μ Γ 0 S / A ( i ϵ , k ; i ϵ ′ , k ′ ∣ iω , q ) ω = q = 0 := Γ 0 μ S / A ( i ϵ , k ; i ϵ ′ , k ′ ) .
We emphasize that, although Γ 0 A ( i ϵ , k ; i ϵ ′ , k ′ ) \Gamma_0^A(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}') Γ 0 A ( i ϵ , k ; i ϵ ′ , k ′ ) trivially vanishes, its derivatives Γ ˙ 0 A ( i ϵ , k ; i ϵ ′ , k ′ ) \dot{\Gamma}_0^A(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}') Γ ˙ 0 A ( i ϵ , k ; i ϵ ′ , k ′ ) and Γ 0 μ A ( i ϵ , k ; i ϵ ′ , k ′ ) \Gamma_{0\mu}^A(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}') Γ 0 μ A ( i ϵ , k ; i ϵ ′ , k ′ ) have no reason to vanish either.
With the above notation, the representation within the skeleton-diagram formalism of the Hall conductance in two dimensions reads [8 ]
σ H = σ H L + i e 2 2 V 2 ∑ k k ′ ∫ d ϵ 2 π d ϵ ′ 2 π ϵ μ ν Tr ( ∂ μ G ( i ϵ , k ) ⋅ Γ ˙ 0 A ( i ϵ , k ; i ϵ ′ , k ′ ) ⋅ ¨ ∂ ν G ( i ϵ ′ , k ′ ) ) ≡ − e c V ∂ N L ∂ B − e c V ∂ I L ∂ B = e 2 2 π ν H L + e 2 2 π ν H L I , (41) \begin{aligned}
\sigma_{\mathrm{H}}=\sigma_{\mathrm{H}L}+\frac{ie^2}{2V^2}\sum_{\mathbf{k}\mathbf{k}'}\int\frac{d\epsilon}{2\pi}\frac{d\epsilon'}{2\pi}\epsilon_{\mu\nu}\operatorname{Tr}\left(\partial_\mu G(i\epsilon,\mathbf{k})\cdot\dot{\Gamma}_0^A(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}')\ddot{\cdot}\partial_\nu G(i\epsilon',\mathbf{k}')\right) \\
&\equiv-\frac{ec}{V}\frac{\partial N_{\mathrm{L}}}{\partial B}-\frac{ec}{V}\frac{\partial I_{\mathrm{L}}}{\partial B}=\frac{e^2}{2\pi}\nu_{\mathrm{H}L}+\frac{e^2}{2\pi}\nu_{\mathrm{H}L}^{I}, \tag*{(41)}
\end{aligned} σ H = σ H L + 2 V 2 i e 2 k k ′ ∑ ∫ 2 π d ϵ 2 π d ϵ ′ ϵ μν Tr ( ∂ μ G ( i ϵ , k ) ⋅ Γ ˙ 0 A ( i ϵ , k ; i ϵ ′ , k ′ ) ⋅ ¨ ∂ ν G ( i ϵ ′ , k ′ ) ) ≡ − V ec ∂ B ∂ N L − V ec ∂ B ∂ I L = 2 π e 2 ν H L + 2 π e 2 ν H L I , (41)
where ν H L \nu_{\mathrm{H}L} ν H L has been defined in (4) and ν H L I \nu_{\mathrm{H}L}^{I} ν H L I quantifies the violation of Luttinger’s theorem within the Streda formula [5 ] . One can readily verify that σ H \sigma_{\mathrm{H}} σ H indeed vanishes if the system is time-reversal invariant, namely when Eqs. (39) and (38) are satisfied. The equivalence between ν H L \nu_{\mathrm{H}L} ν H L and the Ishikawa-Matsuyama-Volovik topological invariant has been discussed in [8 ] and explicitly proven in [5 ] . In what follows, we demonstrate that − e c ∂ B I L / V -ec\,\partial_B I_{\mathrm{L}}/V − ec ∂ B I L / V yields exactly the integral in (41) . Although this relation may appear straightforward, its explicit verification is highly instructive, as it confirms that the skeleton-diagram formalism remains fully consistent with the Streda formula, a strict thermodynamic identity.
The Luttinger integral in a magnetic field
Let us now consider a correlated two-dimensional insulator in a weak perpendicular magnetic field B B B . Following [5 ] , and upon introducing the Peierls phase factor
ϕ ( R 1 , R 2 ) = − e c ∫ R 2 R 1 d r ⋅ A ( r ) , (42) \phi(\mathbf{R}_1,\mathbf{R}_2)=-\frac{e}{c}\int_{\mathbf{R}_2}^{\mathbf{R}_1}d\mathbf{r}\cdot\mathbf{A}(\mathbf{r}), \tag*{(42)} ϕ ( R 1 , R 2 ) = − c e ∫ R 2 R 1 d r ⋅ A ( r ) , (42)
the non-interacting Hamiltonian can be written as
H ( R 1 , R 2 ) = e i ϕ ( R 1 , R 2 ) H 0 ( R 1 , R 2 ) , H(\mathbf{R}_1,\mathbf{R}_2)=e^{i\phi(\mathbf{R}_1,\mathbf{R}_2)}H_0(\mathbf{R}_1,\mathbf{R}_2), H ( R 1 , R 2 ) = e i ϕ ( R 1 , R 2 ) H 0 ( R 1 , R 2 ) ,
where H 0 ( R 1 , R 2 ) H_0(\mathbf{R}_1,\mathbf{R}_2) H 0 ( R 1 , R 2 ) is its expression in absence of B B B and R i \mathbf{R}_i R i are lattice sites. Correspondingly, one can write [5 ]
G ( i ϵ ; R 1 , R 2 ) = e i ϕ ( R 1 , R 2 ) G B ( i ϵ ; R 1 , R 2 ) , Σ ( i ϵ ; R 1 , R 2 ) = e i ϕ ( R 1 , R 2 ) Σ B ( i ϵ ; R 1 , R 2 ) , G(i\epsilon;\mathbf{R}_1,\mathbf{R}_2)=e^{i\phi(\mathbf{R}_1,\mathbf{R}_2)}G_B(i\epsilon;\mathbf{R}_1,\mathbf{R}_2),\quad\Sigma(i\epsilon;\mathbf{R}_1,\mathbf{R}_2)=e^{i\phi(\mathbf{R}_1,\mathbf{R}_2)}\Sigma_B(i\epsilon;\mathbf{R}_1,\mathbf{R}_2), G ( i ϵ ; R 1 , R 2 ) = e i ϕ ( R 1 , R 2 ) G B ( i ϵ ; R 1 , R 2 ) , Σ ( i ϵ ; R 1 , R 2 ) = e i ϕ ( R 1 , R 2 ) Σ B ( i ϵ ; R 1 , R 2 ) ,
where the functions with the B B B subscript are gauge invariant, as well as
G B ( i ϵ ; R 1 , R 2 ) = G ( 0 ) ( i ϵ ; R 1 , R 2 ) + B G ( 1 ) ( i ϵ ; R 1 , R 2 ) , (43) G_B(i\epsilon;\mathbf{R}_1,\mathbf{R}_2)=G^{(0)}(i\epsilon;\mathbf{R}_1,\mathbf{R}_2)+B\,G^{(1)}(i\epsilon;\mathbf{R}_1,\mathbf{R}_2), \tag*{(43)} G B ( i ϵ ; R 1 , R 2 ) = G ( 0 ) ( i ϵ ; R 1 , R 2 ) + B G ( 1 ) ( i ϵ ; R 1 , R 2 ) , (43)
and similarly for the self-energy. The first term on the right hand side of (43) is the Green’s function at zero field while the second is its first order correction. It was found in [5 ] that the Fourier transform of G ( 1 ) G^{(1)} G ( 1 ) satisfies
G ( 1 ) ( i ϵ , k ) = i e 2 c ϵ μ ν G ( 0 ) ( i ϵ , k ) ⋅ ∂ μ G ( 0 ) ( i ϵ , k ) − 1 ⋅ ∂ ν G ( 0 ) ( i ϵ , k ) + G ( 0 ) ( i ϵ , k ) ⋅ Σ ( 1 ) ( i ϵ , k ) ⋅ G ( 0 ) ( i ϵ , k ) . (44) G^{(1)}(i\epsilon,\mathbf{k})=i\frac{e}{2c}\epsilon_{\mu\nu}G^{(0)}(i\epsilon,\mathbf{k})\cdot\partial_\mu G^{(0)}(i\epsilon,\mathbf{k})^{-1}\cdot\partial_\nu G^{(0)}(i\epsilon,\mathbf{k})+G^{(0)}(i\epsilon,\mathbf{k})\cdot\Sigma^{(1)}(i\epsilon,\mathbf{k})\cdot G^{(0)}(i\epsilon,\mathbf{k}). \tag*{(44)} G ( 1 ) ( i ϵ , k ) = i 2 c e ϵ μν G ( 0 ) ( i ϵ , k ) ⋅ ∂ μ G ( 0 ) ( i ϵ , k ) − 1 ⋅ ∂ ν G ( 0 ) ( i ϵ , k ) + G ( 0 ) ( i ϵ , k ) ⋅ Σ ( 1 ) ( i ϵ , k ) ⋅ G ( 0 ) ( i ϵ , k ) . (44)
Magnetic field corrections to the self-energy
Now, we extend the results of [5 ] . The Luttinger-Ward functional Φ [ G ] \Phi[G] Φ [ G ] does not depend explicitly on B B B , which enters only through G G G , so that Σ [ G ] = δ Φ / δ G \Sigma[G]=\delta\Phi/\delta G Σ [ G ] = δ Φ/ δ G is the same functional at any field. In short notation,
Σ B = e − i ϕ ( B ) Σ [ e i ϕ ( B ) G B ] , (45) \Sigma_B=e^{-i\phi(B)}\Sigma\left[e^{i\phi(B)}G_B\right], \tag*{(45)} Σ B = e − i ϕ ( B ) Σ [ e i ϕ ( B ) G B ] , (45)
so that, if we write ϕ ( B ) = B ϕ ( 1 ) \phi(B)=B\phi^{(1)} ϕ ( B ) = B ϕ ( 1 ) , ignore for the moment the frequency, and expand (45) to first order in B B B , the chain rule yields Σ ( 1 ) = Σ d i r e c t ( 1 ) + Σ e x c h a n g e ( 1 ) \Sigma^{(1)}=\Sigma_{\mathrm{direct}}^{(1)}+\Sigma_{\mathrm{exchange}}^{(1)} Σ ( 1 ) = Σ direct ( 1 ) + Σ exchange ( 1 ) , where Σ d i r e c t ( 1 ) = Γ 0 ⋅ G ( 1 ) \Sigma_{\mathrm{direct}}^{(1)}=\Gamma_0\mathbin{\cdot}G^{(1)} Σ direct ( 1 ) = Γ 0 ⋅ G ( 1 ) , while the contribution of the Peierls phases reads
Σ e x c h a n g e ( 1 ) ( R 1 , R 2 ) = i ∑ R 3 , R 4 Γ 0 ( R 1 , R 2 ; R 3 , R 4 ) ⋅ ¨ ϕ ( 1 ) ( R 4 , R 3 ) G ( 0 ) ( R 4 , R 3 ) − i ϕ ( 1 ) ( R 1 , R 2 ) Σ ( 0 ) ( R 1 , R 2 ) . (46) \Sigma_{\mathrm{exchange}}^{(1)}(\mathbf{R}_1,\mathbf{R}_2)=i\sum_{\mathbf{R}_3,\mathbf{R}_4}\Gamma_0(\mathbf{R}_1,\mathbf{R}_2;\mathbf{R}_3,\mathbf{R}_4)\mathbin{\ddot{\cdot}}\phi^{(1)}(\mathbf{R}_4,\mathbf{R}_3)G^{(0)}(\mathbf{R}_4,\mathbf{R}_3)-i\phi^{(1)}(\mathbf{R}_1,\mathbf{R}_2)\Sigma^{(0)}(\mathbf{R}_1,\mathbf{R}_2). \tag*{(46)} Σ exchange ( 1 ) ( R 1 , R 2 ) = i R 3 , R 4 ∑ Γ 0 ( R 1 , R 2 ; R 3 , R 4 ) ⋅ ¨ ϕ ( 1 ) ( R 4 , R 3 ) G ( 0 ) ( R 4 , R 3 ) − i ϕ ( 1 ) ( R 1 , R 2 ) Σ ( 0 ) ( R 1 , R 2 ) . (46)
Equation (46) can be evaluated in any gauge, because the Luttinger-Ward functional is gauge invariant 2 . In particular, in the symmetric gauge, A ( r ) = 1 2 B × ( r − r 0 ) \mathbf{A}(\mathbf{r})=\frac{1}{2}\mathbf{B}\times(\mathbf{r}-\mathbf{r}_0) A ( r ) = 2 1 B × ( r − r 0 ) ,
ϕ ( 1 ) ( R a , R b ) = e 2 c ( R a − r 0 ) × ( R b − r 0 ) , \phi^{(1)}(\mathbf{R}_a,\mathbf{R}_b)=\frac{e}{2c}(\mathbf{R}_a-\mathbf{r}_0)\times(\mathbf{R}_b-\mathbf{r}_0), ϕ ( 1 ) ( R a , R b ) = 2 c e ( R a − r 0 ) × ( R b − r 0 ) ,
and vanishes when r 0 \mathbf{r}_0 r 0 lies on the line through R a \mathbf{R}_a R a and R b \mathbf{R}_b R b . We therefore center the gauge at the midpoint of the external pair, r 0 = ( R 1 + R 2 ) / 2 \mathbf{r}_0=(\mathbf{R}_1+\mathbf{R}_2)/2 r 0 = ( R 1 + R 2 ) /2 , so that ϕ ( 1 ) ( R 1 , R 2 ) = 0 \phi^{(1)}(\mathbf{R}_1,\mathbf{R}_2)=0 ϕ ( 1 ) ( R 1 , R 2 ) = 0 , only the first term in (46) survives and reads
Σ e x c h a n g e ( 1 ) ( R 1 , R 2 ) = i e 4 c ∑ R 3 , R 4 ( R 3 + R 4 − R 1 − R 2 ) × ( R 3 − R 4 ) Γ 0 ( R 1 , R 2 ; R 3 , R 4 ) ⋅ ¨ G ( 0 ) ( R 4 , R 3 ) . (47) \Sigma_{\mathrm{exchange}}^{(1)}(\mathbf{R}_1,\mathbf{R}_2)=i\frac{e}{4c}\sum_{\mathbf{R}_3,\mathbf{R}_4}(\mathbf{R}_3+\mathbf{R}_4-\mathbf{R}_1-\mathbf{R}_2)\times(\mathbf{R}_3-\mathbf{R}_4)\Gamma_0(\mathbf{R}_1,\mathbf{R}_2;\mathbf{R}_3,\mathbf{R}_4)\mathbin{\ddot{\cdot}}G^{(0)}(\mathbf{R}_4,\mathbf{R}_3). \tag*{(47)} Σ exchange ( 1 ) ( R 1 , R 2 ) = i 4 c e R 3 , R 4 ∑ ( R 3 + R 4 − R 1 − R 2 ) × ( R 3 − R 4 ) Γ 0 ( R 1 , R 2 ; R 3 , R 4 ) ⋅ ¨ G ( 0 ) ( R 4 , R 3 ) . (47)
We can now take the Fourier transform and, after some lengthy but straightforward calculations, obtain
Σ e x c h a n g e ( 1 ) ( k ) = − i e 2 c 1 V ∑ k ′ ϵ μ ν Γ μ A ( k ; k ′ ) ⋅ ¨ ∂ ν G ( 0 ) ( k ′ ) . (48) \Sigma_{\mathrm{exchange}}^{(1)}(\mathbf{k})=-\frac{ie}{2c}\frac{1}{V}\sum_{\mathbf{k}'}\epsilon_{\mu\nu}\Gamma_\mu^A(\mathbf{k};\mathbf{k}')\mathbin{\ddot{\cdot}}\partial_\nu G^{(0)}(\mathbf{k}'). \tag*{(48)} Σ exchange ( 1 ) ( k ) = − 2 c i e V 1 k ′ ∑ ϵ μν Γ μ A ( k ; k ′ ) ⋅ ¨ ∂ ν G ( 0 ) ( k ′ ) . (48)
( R 1 − R 2 ) Σ ( 0 ) ( R 1 , R 2 ) = ∑ R 3 , R 4 Γ 0 ( R 1 , R 2 ; R 3 , R 4 ) ⋅ ¨ ( R 4 − R 3 ) G ( 0 ) ( R 4 , R 3 ) , (\mathbf{R}_1-\mathbf{R}_2)\Sigma^{(0)}(\mathbf{R}_1,\mathbf{R}_2)=\sum_{\mathbf{R}_3,\mathbf{R}_4}\Gamma_0(\mathbf{R}_1,\mathbf{R}_2;\mathbf{R}_3,\mathbf{R}_4)\mathbin{\ddot{\cdot}}(\mathbf{R}_4-\mathbf{R}_3)G^{(0)}(\mathbf{R}_4,\mathbf{R}_3), ( R 1 − R 2 ) Σ ( 0 ) ( R 1 , R 2 ) = R 3 , R 4 ∑ Γ 0 ( R 1 , R 2 ; R 3 , R 4 ) ⋅ ¨ ( R 4 − R 3 ) G ( 0 ) ( R 4 , R 3 ) ,
which can be employed to demonstrate that, whatever gauge choice we make, we always end up in Eq. (47) .
Collecting all results, we thus find that
Σ ( 1 ) ( k ) = 1 V ∑ k ′ Γ 0 ( k ; k ′ ) ⋅ G ( 1 ) ( k ′ ) − i e 2 c 1 V ∑ k ′ ϵ μ ν Γ 0 μ A ( k ; k ′ ) ⋅ ∂ ν G ( 0 ) ( k ′ ) . (49) \Sigma^{(1)}(\mathbf{k}) = \frac{1}{V}\sum_{\mathbf{k}'} \Gamma_0(\mathbf{k};\mathbf{k}') \mathbin{\cdot} G^{(1)}(\mathbf{k}') - \frac{ie}{2c}\frac{1}{V}\sum_{\mathbf{k}'} \epsilon_{\mu\nu}\Gamma^A_{0\mu}(\mathbf{k};\mathbf{k}') \mathbin{\cdot} \partial_\nu G^{(0)}(\mathbf{k}'). \tag*{(49)} Σ ( 1 ) ( k ) = V 1 k ′ ∑ Γ 0 ( k ; k ′ ) ⋅ G ( 1 ) ( k ′ ) − 2 c i e V 1 k ′ ∑ ϵ μν Γ 0 μ A ( k ; k ′ ) ⋅ ∂ ν G ( 0 ) ( k ′ ) . (49)
We rewrite Eq. (44) as
G ( 1 ) ( k ) = i e 2 c ϵ μ ν G ( 0 ) ( k ) ⋅ ∂ μ G ( 0 ) ( k ) − 1 ⋅ ∂ ν G ( 0 ) ( k ) + G ( 0 ) ( k ) ⋅ Σ ( 1 ) ( k ) ⋅ G ( 0 ) ( k ) G^{(1)}(\mathbf{k}) = i\frac{e}{2c}\epsilon_{\mu\nu}G^{(0)}(\mathbf{k}) \mathbin{\cdot} \partial_\mu G^{(0)}(\mathbf{k})^{-1} \mathbin{\cdot} \partial_\nu G^{(0)}(\mathbf{k}) + G^{(0)}(\mathbf{k}) \mathbin{\cdot} \Sigma^{(1)}(\mathbf{k}) \mathbin{\cdot} G^{(0)}(\mathbf{k}) G ( 1 ) ( k ) = i 2 c e ϵ μν G ( 0 ) ( k ) ⋅ ∂ μ G ( 0 ) ( k ) − 1 ⋅ ∂ ν G ( 0 ) ( k ) + G ( 0 ) ( k ) ⋅ Σ ( 1 ) ( k ) ⋅ G ( 0 ) ( k )
: = Δ G ( k ) + G ( 0 ) ( k ) ⋅ Σ ( 1 ) ( k ) ⋅ G ( 0 ) ( k ) , (50) := \Delta G(\mathbf{k}) + G^{(0)}(\mathbf{k}) \mathbin{\cdot} \Sigma^{(1)}(\mathbf{k}) \mathbin{\cdot} G^{(0)}(\mathbf{k}), \tag*{(50)} := Δ G ( k ) + G ( 0 ) ( k ) ⋅ Σ ( 1 ) ( k ) ⋅ G ( 0 ) ( k ) , (50)
where Δ G ( k ) \Delta G(\mathbf{k}) Δ G ( k ) looks like the simple variation of G G G in presence of B B B . It follows that, setting G ( 0 ) ≡ G G^{(0)} \equiv G G ( 0 ) ≡ G ,
Σ ( 1 ) ( k ) = Σ d i r e c t ( 1 ) ( k ) + Σ e x c h a n g e ( 1 ) ( k ) = 1 V ∑ k ′ Γ 0 ( k ; k ′ ) ⋅ G ( 1 ) ( k ′ ) − i e 2 c 1 V ∑ k ′ ϵ μ ν Γ 0 μ A ( k ; k ′ ) ⋅ ∂ ν G ( k ′ ) = 1 V ∑ k ′ Γ 0 ( k ; k ′ ) ⋅ Δ G ( k ′ ) + 1 V ∑ k ′ Γ 0 ( k ; k ′ ) ⋅ G ( k ′ ) ⋅ Σ ( 1 ) ( k ′ ) ⋅ G ( k ′ ) + Σ e x c h a n g e ( 1 ) ( k ) , \begin{aligned}
\Sigma^{(1)}(\mathbf{k}) &= \Sigma^{(1)}_{\mathrm{direct}}(\mathbf{k}) + \Sigma^{(1)}_{\mathrm{exchange}}(\mathbf{k}) \\
&= \frac{1}{V}\sum_{\mathbf{k}'}\Gamma_0(\mathbf{k};\mathbf{k}') \mathbin{\cdot} G^{(1)}(\mathbf{k}') - \frac{ie}{2c}\frac{1}{V}\sum_{\mathbf{k}'}\epsilon_{\mu\nu}\Gamma^A_{0\mu}(\mathbf{k};\mathbf{k}') \mathbin{\cdot} \partial_\nu G(\mathbf{k}') \\
&= \frac{1}{V}\sum_{\mathbf{k}'}\Gamma_0(\mathbf{k};\mathbf{k}') \mathbin{\cdot} \Delta G(\mathbf{k}') + \frac{1}{V}\sum_{\mathbf{k}'}\Gamma_0(\mathbf{k};\mathbf{k}') \mathbin{\cdot} G(\mathbf{k}') \mathbin{\cdot} \Sigma^{(1)}(\mathbf{k}') \mathbin{\cdot} G(\mathbf{k}') \\
&\qquad+ \Sigma^{(1)}_{\mathrm{exchange}}(\mathbf{k}),
\end{aligned} Σ ( 1 ) ( k ) = Σ direct ( 1 ) ( k ) + Σ exchange ( 1 ) ( k ) = V 1 k ′ ∑ Γ 0 ( k ; k ′ ) ⋅ G ( 1 ) ( k ′ ) − 2 c i e V 1 k ′ ∑ ϵ μν Γ 0 μ A ( k ; k ′ ) ⋅ ∂ ν G ( k ′ ) = V 1 k ′ ∑ Γ 0 ( k ; k ′ ) ⋅ Δ G ( k ′ ) + V 1 k ′ ∑ Γ 0 ( k ; k ′ ) ⋅ G ( k ′ ) ⋅ Σ ( 1 ) ( k ′ ) ⋅ G ( k ′ ) + Σ exchange ( 1 ) ( k ) ,
which we shortly write as
( I − Γ 0 ⊙ R ) ⊙ Σ ( 1 ) = Γ 0 ⊙ Δ G + Σ e x c h a n g e ( 1 ) , \left(\mathbb{I}-\Gamma_0\mathbin{\odot}R\right)\mathbin{\odot}\Sigma^{(1)}=\Gamma_0\mathbin{\odot}\Delta G+\Sigma^{(1)}_{\mathrm{exchange}}, ( I − Γ 0 ⊙ R ) ⊙ Σ ( 1 ) = Γ 0 ⊙ Δ G + Σ exchange ( 1 ) ,
where R = G ⊗ G R=G\otimes G R = G ⊗ G is a tensor. Since, if Γ \Gamma Γ is the reducible vertex, the following equations hold:
( I − Γ 0 ⊙ R ) − 1 ⊙ Γ 0 ≡ Γ , \left(\mathbb{I}-\Gamma_0\mathbin{\odot}R\right)^{-1}\mathbin{\odot}\Gamma_0\equiv\Gamma, ( I − Γ 0 ⊙ R ) − 1 ⊙ Γ 0 ≡ Γ ,
( I − Γ 0 ⊙ R ) − 1 − I = ( I − Γ 0 ⊙ R ) − 1 ⊙ { I − ( I − Γ 0 ⊙ R ) } = Γ ⊙ R , \left(\mathbb{I}-\Gamma_0\mathbin{\odot}R\right)^{-1}-\mathbb{I}=\left(\mathbb{I}-\Gamma_0\mathbin{\odot}R\right)^{-1}\mathbin{\odot}\left\{\mathbb{I}-\left(\mathbb{I}-\Gamma_0\mathbin{\odot}R\right)\right\}=\Gamma\mathbin{\odot}R, ( I − Γ 0 ⊙ R ) − 1 − I = ( I − Γ 0 ⊙ R ) − 1 ⊙ { I − ( I − Γ 0 ⊙ R ) } = Γ ⊙ R ,
we find that
Σ ( 1 ) = Σ d i r e c t ( 1 ) + Σ e x c h a n g e ( 1 ) = Γ ⊙ Δ G + ( I − Γ 0 ⊙ R ) − 1 ⊙ Σ e x c h a n g e ( 1 ) , \Sigma^{(1)}=\Sigma^{(1)}_{\mathrm{direct}}+\Sigma^{(1)}_{\mathrm{exchange}}=\Gamma\mathbin{\odot}\Delta G+\left(\mathbb{I}-\Gamma_0\mathbin{\odot}R\right)^{-1}\mathbin{\odot}\Sigma^{(1)}_{\mathrm{exchange}}, Σ ( 1 ) = Σ direct ( 1 ) + Σ exchange ( 1 ) = Γ ⊙ Δ G + ( I − Γ 0 ⊙ R ) − 1 ⊙ Σ exchange ( 1 ) ,
thus
Σ d i r e c t ( 1 ) = Γ ⊙ Δ G + Γ ⊙ R ⊙ Σ e x c h a n g e ( 1 ) . \Sigma^{(1)}_{\mathrm{direct}}=\Gamma\mathbin{\odot}\Delta G+\Gamma\mathbin{\odot}R\mathbin{\odot}\Sigma^{(1)}_{\mathrm{exchange}}. Σ direct ( 1 ) = Γ ⊙ Δ G + Γ ⊙ R ⊙ Σ exchange ( 1 ) .
On the other hand,
G ( 1 ) = Δ G + G ⋅ Σ d i r e c t ( 1 ) ⋅ G + G ⋅ Σ e x c h a n g e ( 1 ) ⋅ G = Δ G + G ⋅ Γ ⊙ Δ G ⋅ G + G ⋅ Γ ⊙ R ⊙ Σ e x c h a n g e ( 1 ) ⋅ G + G ⋅ Σ e x c h a n g e ( 1 ) ⋅ G ≡ Δ G + G ⋅ Σ e x c h a n g e ( 1 ) ⋅ G + G ⋅ Γ ⊙ ( Δ G + G ⋅ Σ e x c h a n g e ( 1 ) ⋅ G ) ⋅ G . \begin{aligned}
G^{(1)} &= \Delta G+G\mathbin{\cdot}\Sigma^{(1)}_{\mathrm{direct}}\mathbin{\cdot}G+G\mathbin{\cdot}\Sigma^{(1)}_{\mathrm{exchange}}\mathbin{\cdot}G \\
&= \Delta G+G\mathbin{\cdot}\Gamma\mathbin{\odot}\Delta G\mathbin{\cdot}G+G\mathbin{\cdot}\Gamma\mathbin{\odot}R\mathbin{\odot}\Sigma^{(1)}_{\mathrm{exchange}}\mathbin{\cdot}G+G\mathbin{\cdot}\Sigma^{(1)}_{\mathrm{exchange}}\mathbin{\cdot}G \\
&\equiv\Delta G+G\mathbin{\cdot}\Sigma^{(1)}_{\mathrm{exchange}}\mathbin{\cdot}G+G\mathbin{\cdot}\Gamma\mathbin{\odot}\left(\Delta G+G\mathbin{\cdot}\Sigma^{(1)}_{\mathrm{exchange}}\mathbin{\cdot}G\right)\mathbin{\cdot}G.
\end{aligned} G ( 1 ) = Δ G + G ⋅ Σ direct ( 1 ) ⋅ G + G ⋅ Σ exchange ( 1 ) ⋅ G = Δ G + G ⋅ Γ ⊙ Δ G ⋅ G + G ⋅ Γ ⊙ R ⊙ Σ exchange ( 1 ) ⋅ G + G ⋅ Σ exchange ( 1 ) ⋅ G ≡ Δ G + G ⋅ Σ exchange ( 1 ) ⋅ G + G ⋅ Γ ⊙ ( Δ G + G ⋅ Σ exchange ( 1 ) ⋅ G ) ⋅ G .
This looks like an equation for a three-leg vertex with the bare value
Λ B 0 ( i ϵ , k ) : = − i π ϵ μ ν { ∂ μ G ( i ϵ , k ) − 1 ⋅ G ( i ϵ , k ) ⋅ ∂ ν G ( i ϵ , k ) − 1 + 1 V ∑ k ′ ∫ d ϵ ′ 2 π Γ 0 μ A ( i ϵ , k ; i ϵ ′ , k ′ ) ⋅ ∂ ν G ( i ϵ ′ , k ′ ) } , (51) \Lambda^0_B(i\epsilon,\mathbf{k}) := -i\pi\epsilon_{\mu\nu}\left\{\partial_\mu G(i\epsilon,\mathbf{k})^{-1}\mathbin{\cdot}G(i\epsilon,\mathbf{k})\mathbin{\cdot}\partial_\nu G(i\epsilon,\mathbf{k})^{-1}
+\frac{1}{V}\sum_{\mathbf{k}'}\int\frac{d\epsilon'}{2\pi}\Gamma^A_{0\mu}(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}')\mathbin{\cdot}\partial_\nu G(i\epsilon',\mathbf{k}')\right\}, \tag*{(51)} Λ B 0 ( i ϵ , k ) := − iπ ϵ μν { ∂ μ G ( i ϵ , k ) − 1 ⋅ G ( i ϵ , k ) ⋅ ∂ ν G ( i ϵ , k ) − 1 + V 1 k ′ ∑ ∫ 2 π d ϵ ′ Γ 0 μ A ( i ϵ , k ; i ϵ ′ , k ′ ) ⋅ ∂ ν G ( i ϵ ′ , k ′ ) } , (51)
the fully interacting one
Λ B ( i ϵ , k ) = Λ B 0 ( i ϵ , k ) + 1 V ∑ k ′ ∫ d ϵ ′ 2 π Γ ( i ϵ , k ; i ϵ ′ , k ′ ) ⋅ G ( i ϵ ′ , k ′ ) ⋅ Λ B 0 ( i ϵ ′ , k ′ ) ⋅ G ( i ϵ ′ , k ′ ) , (52) \Lambda_B(i\epsilon,\mathbf{k})=\Lambda_B^0(i\epsilon,\mathbf{k})+\frac{1}{V}\sum_{\mathbf{k}'}\int\frac{d\epsilon'}{2\pi}\,\Gamma(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}')\mathbin{\vcenter{\text{\(\cdot\)}}}G(i\epsilon',\mathbf{k}')\mathbin{\vcenter{\text{\(\cdot\)}}}\Lambda_B^0(i\epsilon',\mathbf{k}')\mathbin{\vcenter{\text{\(\cdot\)}}}G(i\epsilon',\mathbf{k}') \, , \tag*{(52)} Λ B ( i ϵ , k ) = Λ B 0 ( i ϵ , k ) + V 1 k ′ ∑ ∫ 2 π d ϵ ′ Γ ( i ϵ , k ; i ϵ ′ , k ′ ) ⋅ G ( i ϵ ′ , k ′ ) ⋅ Λ B 0 ( i ϵ ′ , k ′ ) ⋅ G ( i ϵ ′ , k ′ ) , (52)
so that
G ( 1 ) ( i ϵ , k ) = e 2 π c G ( i ϵ , k ) ⋅ Λ B ( i ϵ , k ) ⋅ G ( i ϵ , k ) . (53) G^{(1)}(i\epsilon,\mathbf{k})=\frac{e}{2\pi c}G(i\epsilon,\mathbf{k})\mathbin{\vcenter{\text{\(\cdot\)}}}\Lambda_B(i\epsilon,\mathbf{k})\mathbin{\vcenter{\text{\(\cdot\)}}}G(i\epsilon,\mathbf{k}) \, . \tag*{(53)} G ( 1 ) ( i ϵ , k ) = 2 π c e G ( i ϵ , k ) ⋅ Λ B ( i ϵ , k ) ⋅ G ( i ϵ , k ) . (53)
If we assume, for simplicity, that Tr ( … ) \operatorname{Tr}(\ldots) Tr ( … ) includes the sum over k \mathbf{k} k and the integral over ϵ \epsilon ϵ times the regularizing factor, and note that
1 = ∂ i ϵ G ( i ϵ , k ) − 1 + ∂ i ϵ Σ ( i ϵ , k ) ≔ G ˙ − 1 + Σ ˙ , 1=\partial_{i\epsilon}G(i\epsilon,\mathbf{k})^{-1}+\partial_{i\epsilon}\Sigma(i\epsilon,\mathbf{k})\coloneqq\dot{G}^{-1}+\dot{\Sigma}, 1 = ∂ i ϵ G ( i ϵ , k ) − 1 + ∂ i ϵ Σ ( i ϵ , k ) : = G ˙ − 1 + Σ ˙ ,
which is precisely the trick used to derive Luttinger’s theorem, then, using Eq. (98) ,
Tr ( G ( 1 ) ) = e 2 π c Tr ( G ⋅ G ⋅ Λ B ) = e 2 π c Tr ( G ⋅ G ˙ − 1 ⋅ G ⋅ Λ B ) + e 2 π c Tr ( G ⋅ Σ ˙ ⋅ G ⋅ Λ B ) = − e 2 π c Tr ( G ˙ ⋅ Λ B ) + e 2 π c Tr ( G ⋅ ( Γ 0 ⋅ G ˙ ) ⋅ G ⋅ Λ B ) = − e 2 π c Tr ( G ˙ ⋅ Λ B 0 ) − e 2 π c Tr ( G ˙ ⋅ Γ 0 ⋅ R ⋅ Λ B 0 ) + e 2 π c Tr ( G ˙ ⋅ Γ 0 ⋅ R ⋅ Λ B 0 ) + e 2 π c Tr ( G ˙ ⋅ Γ 0 ⋅ R ⋅ Γ ⋅ R ⋅ Λ B 0 ) = − e 2 π c Tr ( G ˙ ⋅ Λ B 0 ) , (54) \begin{aligned}
\operatorname{Tr}\left(G^{(1)}\right)&=\frac{e}{2\pi c}\operatorname{Tr}\left(G\cdot G\cdot\Lambda_B\right)\\
&=\frac{e}{2\pi c}\operatorname{Tr}\left(G\cdot\dot{G}^{-1}\cdot G\cdot\Lambda_B\right)+\frac{e}{2\pi c}\operatorname{Tr}\left(G\cdot\dot{\Sigma}\cdot G\cdot\Lambda_B\right)\\
&=-\frac{e}{2\pi c}\operatorname{Tr}\left(\dot{G}\cdot\Lambda_B\right)+\frac{e}{2\pi c}\operatorname{Tr}\left(G\cdot\left(\Gamma_0\cdot\dot{G}\right)\cdot G\cdot\Lambda_B\right)\\
&=-\frac{e}{2\pi c}\operatorname{Tr}\left(\dot{G}\cdot\Lambda_B^0\right)-\frac{e}{2\pi c}\operatorname{Tr}\left(\dot{G}\cdot\Gamma_0\cdot R\cdot\Lambda_B^0\right)\\
&\quad+\frac{e}{2\pi c}\operatorname{Tr}\left(\dot{G}\cdot\Gamma_0\cdot R\cdot\Lambda_B^0\right)+\frac{e}{2\pi c}\operatorname{Tr}\left(\dot{G}\cdot\Gamma_0\cdot R\cdot\Gamma\cdot R\cdot\Lambda_B^0\right)\\
&=-\frac{e}{2\pi c}\operatorname{Tr}\left(\dot{G}\cdot\Lambda_B^0\right), \tag*{(54)}
\end{aligned} Tr ( G ( 1 ) ) = 2 π c e Tr ( G ⋅ G ⋅ Λ B ) = 2 π c e Tr ( G ⋅ G ˙ − 1 ⋅ G ⋅ Λ B ) + 2 π c e Tr ( G ⋅ Σ ˙ ⋅ G ⋅ Λ B ) = − 2 π c e Tr ( G ˙ ⋅ Λ B ) + 2 π c e Tr ( G ⋅ ( Γ 0 ⋅ G ˙ ) ⋅ G ⋅ Λ B ) = − 2 π c e Tr ( G ˙ ⋅ Λ B 0 ) − 2 π c e Tr ( G ˙ ⋅ Γ 0 ⋅ R ⋅ Λ B 0 ) + 2 π c e Tr ( G ˙ ⋅ Γ 0 ⋅ R ⋅ Λ B 0 ) + 2 π c e Tr ( G ˙ ⋅ Γ 0 ⋅ R ⋅ Γ ⋅ R ⋅ Λ B 0 ) = − 2 π c e Tr ( G ˙ ⋅ Λ B 0 ) , (54)
since − Γ -\Gamma\mathbin{\vcenter{\text{}}} − Γ \cdotR + Γ 0 R+\Gamma_0\mathbin{\vcenter{\text{}}} R + Γ 0 \cdotR + Γ 0 R+\Gamma_0\mathbin{\vcenter{\text{}}} R + Γ 0 \cdotR R\mathbin{\vcenter{\text{}}} R \cdotΓ \Gamma\mathbin{\vcenter{\text{}}} Γ \cdotR = 0 R=0 R = 0 . More explicitly, and upon defining N ( B ) : = N ( 0 ) + B N ( 1 ) N(B):=N^{(0)}+BN^{(1)} N ( B ) := N ( 0 ) + B N ( 1 ) , N L ( B ) : = N L ( 0 ) + B N L ( 1 ) N_L(B):=N_L^{(0)}+BN_L^{(1)} N L ( B ) := N L ( 0 ) + B N L ( 1 ) and I L ( B ) : = I L ( 0 ) + B I L ( 1 ) I_L(B):=I_L^{(0)}+BI_L^{(1)} I L ( B ) := I L ( 0 ) + B I L ( 1 ) ,
Tr ( G ( 1 ) ) = N ( 1 ) = − e 2 π c Tr ( G ˙ ⋅ Λ B 0 ) = i e 2 c ∑ k ∫ d ϵ 2 π ϵ μ ν Tr ( ∂ i ϵ G ( i ϵ , k ) ⋅ ∂ μ G ( i ϵ , k ) − 1 ⋅ G ( i ϵ , k ) ⋅ ∂ ν G ( i ϵ , k ) − 1 ) + i e 2 c 1 V ∑ k k ′ ∫ d ϵ 2 π d ϵ ′ 2 π ϵ μ ν Tr ( ∂ i ϵ G ( i ϵ , k ) ⋅ Γ μ A ( i ϵ , k ; i ϵ ′ , k ′ ) ⋅ ∂ ν G ( i ϵ ′ , k ′ ) ) = N L ( 1 ) + i e 2 c 1 V ∑ k k ′ ∫ d ϵ 2 π d ϵ ′ 2 π ϵ μ ν Tr ( ∂ i ϵ G ( i ϵ , k ) ⋅ Γ μ A ( i ϵ , k ; i ϵ ′ , k ′ ) ⋅ ∂ ν G ( i ϵ ′ , k ′ ) ) ≡ N L ( 1 ) + I L ( 1 ) , \begin{aligned}
\operatorname{Tr}\left(G^{(1)}\right)&=N^{(1)}=-\frac{e}{2\pi c}\operatorname{Tr}\left(\dot{G}\cdot\Lambda_B^0\right)\\
&=\frac{ie}{2c}\sum_{\mathbf{k}}\int\frac{d\epsilon}{2\pi}\,\epsilon_{\mu\nu}\operatorname{Tr}\left(\partial_{i\epsilon}G(i\epsilon,\mathbf{k})\cdot\partial_\mu G(i\epsilon,\mathbf{k})^{-1}\cdot G(i\epsilon,\mathbf{k})\cdot\partial_\nu G(i\epsilon,\mathbf{k})^{-1}\right)\\
&\quad+\frac{ie}{2c}\frac{1}{V}\sum_{\mathbf{k}\mathbf{k}'}\int\frac{d\epsilon}{2\pi}\frac{d\epsilon'}{2\pi}\,\epsilon_{\mu\nu}\operatorname{Tr}\left(\partial_{i\epsilon}G(i\epsilon,\mathbf{k})\cdot\Gamma_\mu^A(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}')\cdot\partial_\nu G(i\epsilon',\mathbf{k}')\right)\\
&=N_L^{(1)}+\frac{ie}{2c}\frac{1}{V}\sum_{\mathbf{k}\mathbf{k}'}\int\frac{d\epsilon}{2\pi}\frac{d\epsilon'}{2\pi}\,\epsilon_{\mu\nu}\operatorname{Tr}\left(\partial_{i\epsilon}G(i\epsilon,\mathbf{k})\cdot\Gamma_\mu^A(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}')\cdot\partial_\nu G(i\epsilon',\mathbf{k}')\right)\\
&\equiv N_L^{(1)}+I_L^{(1)},
\end{aligned} Tr ( G ( 1 ) ) = N ( 1 ) = − 2 π c e Tr ( G ˙ ⋅ Λ B 0 ) = 2 c i e k ∑ ∫ 2 π d ϵ ϵ μν Tr ( ∂ i ϵ G ( i ϵ , k ) ⋅ ∂ μ G ( i ϵ , k ) − 1 ⋅ G ( i ϵ , k ) ⋅ ∂ ν G ( i ϵ , k ) − 1 ) + 2 c i e V 1 k k ′ ∑ ∫ 2 π d ϵ 2 π d ϵ ′ ϵ μν Tr ( ∂ i ϵ G ( i ϵ , k ) ⋅ Γ μ A ( i ϵ , k ; i ϵ ′ , k ′ ) ⋅ ∂ ν G ( i ϵ ′ , k ′ ) ) = N L ( 1 ) + 2 c i e V 1 k k ′ ∑ ∫ 2 π d ϵ 2 π d ϵ ′ ϵ μν Tr ( ∂ i ϵ G ( i ϵ , k ) ⋅ Γ μ A ( i ϵ , k ; i ϵ ′ , k ′ ) ⋅ ∂ ν G ( i ϵ ′ , k ′ ) ) ≡ N L ( 1 ) + I L ( 1 ) ,
where one readily demonstrates that
I L ( 1 ) ≡ Tr ( G ( 1 ) ⋅ Σ ˙ ( 0 ) ) + Tr ( G ( 0 ) ⋅ Σ ˙ ( 1 ) ) . (55) I_L^{(1)}\equiv\operatorname{Tr}\left(G^{(1)}\mathbin{\vcenter{\text{\(\cdot\)}}}\dot{\Sigma}^{(0)}\right)+\operatorname{Tr}\left(G^{(0)}\mathbin{\vcenter{\text{\(\cdot\)}}}\dot{\Sigma}^{(1)}\right). \tag*{(55)} I L ( 1 ) ≡ Tr ( G ( 1 ) ⋅ Σ ˙ ( 0 ) ) + Tr ( G ( 0 ) ⋅ Σ ˙ ( 1 ) ) . (55)
Using the Ward-Takahashi identity derived in Appendix A, see Eq. (99) , we find that
1 V ∑ k k ′ ∫ d ϵ 2 π d ϵ ′ 2 π ϵ μ ν Tr ( ∂ μ G ( i ϵ , k ) ⋅ Γ 0 A ( i ϵ , k ; i ϵ ′ , k ′ ) ⋅ ∂ ν G ( i ϵ ′ , k ′ ) ) = − 1 V ∑ k k ′ ∫ d ϵ 2 π d ϵ ′ 2 π ϵ μ ν Tr ( ∂ i ϵ G ( i ϵ , k ) ⋅ Γ μ A ( i ϵ , k ; i ϵ ′ , k ′ ) ⋅ ∂ ν G ( i ϵ ′ , k ′ ) ) , (56) \begin{aligned}
\frac{1}{V}\sum_{\mathbf{k}\mathbf{k}'}\int\frac{d\epsilon}{2\pi}\frac{d\epsilon'}{2\pi}\,\epsilon_{\mu\nu}\operatorname{Tr}\left(\partial_\mu G(i\epsilon,\mathbf{k})\mathbin{\vcenter{\text{\(\cdot\)}}}\Gamma_0^A(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}')\mathbin{\vcenter{\text{\(\cdot\)}}}\partial_\nu G(i\epsilon',\mathbf{k}')\right)\\
=-\frac{1}{V}\sum_{\mathbf{k}\mathbf{k}'}\int\frac{d\epsilon}{2\pi}\frac{d\epsilon'}{2\pi}\,\epsilon_{\mu\nu}\operatorname{Tr}\left(\partial_{i\epsilon}G(i\epsilon,\mathbf{k})\mathbin{\vcenter{\text{\(\cdot\)}}}\Gamma_\mu^A(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}')\mathbin{\vcenter{\text{\(\cdot\)}}}\partial_\nu G(i\epsilon',\mathbf{k}')\right), \tag*{(56)}
\end{aligned} V 1 k k ′ ∑ ∫ 2 π d ϵ 2 π d ϵ ′ ϵ μν Tr ( ∂ μ G ( i ϵ , k ) ⋅ Γ 0 A ( i ϵ , k ; i ϵ ′ , k ′ ) ⋅ ∂ ν G ( i ϵ ′ , k ′ ) ) = − V 1 k k ′ ∑ ∫ 2 π d ϵ 2 π d ϵ ′ ϵ μν Tr ( ∂ i ϵ G ( i ϵ , k ) ⋅ Γ μ A ( i ϵ , k ; i ϵ ′ , k ′ ) ⋅ ∂ ν G ( i ϵ ′ , k ′ ) ) , (56)
and thereby conclude that
− I L ( 1 ) = − i e 2 c 1 V ∑ k k ′ ∫ d ϵ 2 π d ϵ ′ 2 π ϵ μ ν Tr ( ∂ i ϵ G ( i ϵ , k ) ⋅ Γ 0 μ A ( i ϵ , k ; i ϵ ′ , k ′ ) ⋅ ∂ ν G ( i ϵ ′ , k ′ ) ) = i e 2 c 1 V ∑ k k ′ ∫ d ϵ 2 π d ϵ ′ 2 π ϵ μ ν Tr ( ∂ μ G ( i ϵ , k ) ⋅ Γ ˙ 0 A ( i ϵ , k ; i ϵ ′ , k ′ ) ⋅ ∂ ν G ( i ϵ ′ , k ′ ) ) , (57) \begin{aligned}
-I_L^{(1)} &= -\frac{ie}{2c}\frac{1}{V}\sum_{\mathbf{k}\mathbf{k}'}\int\frac{d\epsilon}{2\pi}\frac{d\epsilon'}{2\pi}\,\epsilon_{\mu\nu}\operatorname{Tr}\left(\partial_{i\epsilon}G(i\epsilon,\mathbf{k})\cdot\Gamma^A_{0\mu}(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}')\cdot\partial_\nu G(i\epsilon',\mathbf{k}')\right) \\
&= \frac{ie}{2c}\frac{1}{V}\sum_{\mathbf{k}\mathbf{k}'}\int\frac{d\epsilon}{2\pi}\frac{d\epsilon'}{2\pi}\,\epsilon_{\mu\nu}\operatorname{Tr}\left(\partial_\mu G(i\epsilon,\mathbf{k})\cdot\dot{\Gamma}^A_0(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}')\cdot\partial_\nu G(i\epsilon',\mathbf{k}')\right), \tag*{(57)}
\end{aligned} − I L ( 1 ) = − 2 c i e V 1 k k ′ ∑ ∫ 2 π d ϵ 2 π d ϵ ′ ϵ μν Tr ( ∂ i ϵ G ( i ϵ , k ) ⋅ Γ 0 μ A ( i ϵ , k ; i ϵ ′ , k ′ ) ⋅ ∂ ν G ( i ϵ ′ , k ′ ) ) = 2 c i e V 1 k k ′ ∑ ∫ 2 π d ϵ 2 π d ϵ ′ ϵ μν Tr ( ∂ μ G ( i ϵ , k ) ⋅ Γ ˙ 0 A ( i ϵ , k ; i ϵ ′ , k ′ ) ⋅ ∂ ν G ( i ϵ ′ , k ′ ) ) , (57)
which is precisely the correction to the Ishikawa-Matsuyama-Volovik expression for the Hall conductance in (41) .
We introduce a three-component index vector μ , ν , ρ ∈ { 0 , 1 , 2 } \mu,\nu,\rho\in\{0,1,2\} μ , ν , ρ ∈ { 0 , 1 , 2 } , where ∂ 0 : = ∂ i ϵ \partial_0 := \partial_{i\epsilon} ∂ 0 := ∂ i ϵ , ∂ 1 : = ∂ k x \partial_1 := \partial_{k_x} ∂ 1 := ∂ k x , and ∂ 2 : = ∂ k y \partial_2 := \partial_{k_y} ∂ 2 := ∂ k y , as well as
γ 0 μ = 0 A ( i ϵ , k ; i ϵ ′ , k ′ ) : = Γ ˙ 0 A ( i ϵ , k ; i ϵ ′ , k ′ ) , \gamma^A_{0\mu=0}(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}') := \dot{\Gamma}^A_0(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}'), γ 0 μ = 0 A ( i ϵ , k ; i ϵ ′ , k ′ ) := Γ ˙ 0 A ( i ϵ , k ; i ϵ ′ , k ′ ) ,
γ 0 μ = 1 , 2 A ( i ϵ , k ; i ϵ ′ , k ′ ) : = Γ 0 μ = 1 , 2 A ( i ϵ , k ; i ϵ ′ , k ′ ) , \gamma^A_{0\mu=1,2}(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}') := \Gamma^A_{0\mu=1,2}(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}'), γ 0 μ = 1 , 2 A ( i ϵ , k ; i ϵ ′ , k ′ ) := Γ 0 μ = 1 , 2 A ( i ϵ , k ; i ϵ ′ , k ′ ) ,
which enables rewriting the first-order magnetic correction (41) compactly as
ν H L I = − i π 3 1 V 2 ∑ k k ′ ∫ d ϵ 2 π d ϵ ′ 2 π ϵ μ ν ρ Tr ( ∂ μ G ( i ϵ , k ) ⋅ γ 0 ν A ( i ϵ , k ; i ϵ ′ , k ′ ) ⋅ ∂ ρ G ( i ϵ ′ , k ′ ) ) , (58) \nu^I_{HL}=-i\frac{\pi}{3}\frac{1}{V^2}\sum_{\mathbf{k}\mathbf{k}'}\int\frac{d\epsilon}{2\pi}\frac{d\epsilon'}{2\pi}\,\epsilon_{\mu\nu\rho}\operatorname{Tr}\left(\partial_\mu G(i\epsilon,\mathbf{k})\cdot\gamma^A_{0\nu}(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}')\cdot\partial_\rho G(i\epsilon',\mathbf{k}')\right), \tag*{(58)} ν H L I = − i 3 π V 2 1 k k ′ ∑ ∫ 2 π d ϵ 2 π d ϵ ′ ϵ μν ρ Tr ( ∂ μ G ( i ϵ , k ) ⋅ γ 0 ν A ( i ϵ , k ; i ϵ ′ , k ′ ) ⋅ ∂ ρ G ( i ϵ ′ , k ′ ) ) , (58)
where ϵ μ ν ρ \epsilon_{\mu\nu\rho} ϵ μν ρ is the completely antisymmetric Levi-Civita tensor in three dimensions. Equation (58) represents the exact interaction-induced correction to the Ishikawa-Matsuyama-Volovik invariant ν H L I \nu^I_{HL} ν H L I when strong correlations in a topological insulator drive a violation of Luttinger’s theorem. Unlike ν H L \nu_{HL} ν H L , which is quantized in integer values, the interaction correction ν H L I \nu^I_{HL} ν H L I can also acquire proper fractional values when the actual Hall number is fractional. Not surprisingly, ν H L I \nu^I_{HL} ν H L I is invariant under any variation of the Green’s function that leaves the system an insulator without Luttinger surfaces, as we demonstrate in Appendix § 9 .
The Luttinger Integral as a novel topological invariant
The inherent consistency between the skeleton-diagram formalism and the thermodynamic Streda formula raises a fundamental question: what precise mathematical conditions dictate the vanishing or non-vanishing of I L I_L I L ?
To this end, we decompose the single-particle Green’s function and the self-energy as in Sec. 2.1,
G ( i ϵ , k ) = G 1 ( ϵ , k ) + i G 2 ( ϵ , k ) , Σ ( i ϵ , k ) = Σ 1 ( ϵ , k ) + i Σ 2 ( ϵ , k ) , (59) G(i\epsilon,\mathbf{k})=G_1(\epsilon,\mathbf{k})+iG_2(\epsilon,\mathbf{k}),\qquad\Sigma(i\epsilon,\mathbf{k})=\Sigma_1(\epsilon,\mathbf{k})+i\Sigma_2(\epsilon,\mathbf{k}), \tag*{(59)} G ( i ϵ , k ) = G 1 ( ϵ , k ) + i G 2 ( ϵ , k ) , Σ ( i ϵ , k ) = Σ 1 ( ϵ , k ) + i Σ 2 ( ϵ , k ) , (59)
where G 1 ( ϵ , k ) G_1(\epsilon,\mathbf{k}) G 1 ( ϵ , k ) and Σ 1 ( ϵ , k ) \Sigma_1(\epsilon,\mathbf{k}) Σ 1 ( ϵ , k ) are hermitian and even in ϵ \epsilon ϵ , while G 2 ( ϵ , k ) G_2(\epsilon,\mathbf{k}) G 2 ( ϵ , k ) and Σ 2 ( ϵ , k ) \Sigma_2(\epsilon,\mathbf{k}) Σ 2 ( ϵ , k ) are also hermitian but odd in ϵ \epsilon ϵ and negative for ϵ > 0 \epsilon>0 ϵ > 0 . Absorbing the momentum summation into the trace definition, which also includes the sum over spin and Bloch waves, the zero-field Luttinger integral can be written as:
I L = ∫ − ∞ ∞ d ϵ 2 π Tr ( G ( i ϵ ) ⋅ ∂ i ϵ Σ ( i ϵ ) ) = − ∫ − ∞ ∞ d ϵ 2 π Tr ( Σ ( i ϵ ) ⋅ ∂ i ϵ G ( i ϵ ) ) = − ∫ − ∞ ∞ d ϵ 2 π { Tr ( Σ 1 ( ϵ ) ⋅ ∂ ϵ G 2 ( ϵ ) ) + Tr ( Σ 2 ( ϵ ) ⋅ ∂ ϵ G 1 ( ϵ ) ) } . (60) \begin{aligned}
I_L &= \int_{-\infty}^{\infty}\frac{d\epsilon}{2\pi}\operatorname{Tr}\left(G(i\epsilon)\cdot\partial_{i\epsilon}\Sigma(i\epsilon)\right)=-\int_{-\infty}^{\infty}\frac{d\epsilon}{2\pi}\operatorname{Tr}\left(\Sigma(i\epsilon)\cdot\partial_{i\epsilon}G(i\epsilon)\right) \\
&=-\int_{-\infty}^{\infty}\frac{d\epsilon}{2\pi}\left\{\operatorname{Tr}\left(\Sigma_1(\epsilon)\cdot\partial_\epsilon G_2(\epsilon)\right)+\operatorname{Tr}\left(\Sigma_2(\epsilon)\cdot\partial_\epsilon G_1(\epsilon)\right)\right\}. \tag*{(60)}
\end{aligned} I L = ∫ − ∞ ∞ 2 π d ϵ Tr ( G ( i ϵ ) ⋅ ∂ i ϵ Σ ( i ϵ ) ) = − ∫ − ∞ ∞ 2 π d ϵ Tr ( Σ ( i ϵ ) ⋅ ∂ i ϵ G ( i ϵ ) ) = − ∫ − ∞ ∞ 2 π d ϵ { Tr ( Σ 1 ( ϵ ) ⋅ ∂ ϵ G 2 ( ϵ ) ) + Tr ( Σ 2 ( ϵ ) ⋅ ∂ ϵ G 1 ( ϵ ) ) } . (60)
If we consider physical variations of the Green’s function δ G ( i ϵ ) = δ G 1 ( ϵ ) + i δ G 2 ( ϵ ) \delta G(i\epsilon)=\delta G_{1}(\epsilon)+i\delta G_{2}(\epsilon) δ G ( i ϵ ) = δ G 1 ( ϵ ) + i δ G 2 ( ϵ ) , where δ G 1 ( ϵ ) \delta G_{1}(\epsilon) δ G 1 ( ϵ ) and δ G 2 ( ϵ ) \delta G_{2}(\epsilon) δ G 2 ( ϵ ) inherit the symmetry properties of G 1 ( ϵ ) G_{1}(\epsilon) G 1 ( ϵ ) and G 2 ( ϵ ) G_{2}(\epsilon) G 2 ( ϵ ) (i.e., even and odd under ϵ → − ϵ \epsilon\to-\epsilon ϵ → − ϵ , respectively), we find:
δ Φ [ G ] = ∫ − ∞ ∞ d ϵ 2 π Tr ( Σ ( i ϵ ) ⋅ δ G ( i ϵ ) ) = ∫ − ∞ ∞ d ϵ 2 π { Tr ( Σ 1 ( ϵ ) ⋅ δ G 1 ( ϵ ) ) − Tr ( Σ 2 ( ϵ ) ⋅ δ G 2 ( ϵ ) ) } + i ∫ − ∞ ∞ d ϵ 2 π { Tr ( Σ 1 ( ϵ ) ⋅ δ G 2 ( ϵ ) ) + Tr ( Σ 2 ( ϵ ) ⋅ δ G 1 ( ϵ ) ) } = ∫ − ∞ ∞ d ϵ 2 π { Tr ( Σ 1 ( ϵ ) ⋅ δ G 1 ( ϵ ) ) − Tr ( Σ 2 ( ϵ ) ⋅ δ G 2 ( ϵ ) ) } . \begin{aligned}
\delta\Phi[G]&=\int_{-\infty}^{\infty}\frac{d\epsilon}{2\pi}\operatorname{Tr}\left(\Sigma(i\epsilon)\cdot\delta G(i\epsilon)\right)\\
&=\int_{-\infty}^{\infty}\frac{d\epsilon}{2\pi}\left\{\operatorname{Tr}\left(\Sigma_{1}(\epsilon)\cdot\delta G_{1}(\epsilon)\right)-\operatorname{Tr}\left(\Sigma_{2}(\epsilon)\cdot\delta G_{2}(\epsilon)\right)\right\}\\
&\quad+i\int_{-\infty}^{\infty}\frac{d\epsilon}{2\pi}\left\{\operatorname{Tr}\left(\Sigma_{1}(\epsilon)\cdot\delta G_{2}(\epsilon)\right)+\operatorname{Tr}\left(\Sigma_{2}(\epsilon)\cdot\delta G_{1}(\epsilon)\right)\right\}\\
&=\int_{-\infty}^{\infty}\frac{d\epsilon}{2\pi}\left\{\operatorname{Tr}\left(\Sigma_{1}(\epsilon)\cdot\delta G_{1}(\epsilon)\right)-\operatorname{Tr}\left(\Sigma_{2}(\epsilon)\cdot\delta G_{2}(\epsilon)\right)\right\}.
\end{aligned} δ Φ [ G ] = ∫ − ∞ ∞ 2 π d ϵ Tr ( Σ ( i ϵ ) ⋅ δ G ( i ϵ ) ) = ∫ − ∞ ∞ 2 π d ϵ { Tr ( Σ 1 ( ϵ ) ⋅ δ G 1 ( ϵ ) ) − Tr ( Σ 2 ( ϵ ) ⋅ δ G 2 ( ϵ ) ) } + i ∫ − ∞ ∞ 2 π d ϵ { Tr ( Σ 1 ( ϵ ) ⋅ δ G 2 ( ϵ ) ) + Tr ( Σ 2 ( ϵ ) ⋅ δ G 1 ( ϵ ) ) } = ∫ − ∞ ∞ 2 π d ϵ { Tr ( Σ 1 ( ϵ ) ⋅ δ G 1 ( ϵ ) ) − Tr ( Σ 2 ( ϵ ) ⋅ δ G 2 ( ϵ ) ) } .
This structure implies the exact functional derivative relations:
δ Φ [ G ] δ G 1 ( ϵ ) ≡ Σ 1 ( ϵ ) , δ Φ [ G ] δ G 2 ( ϵ ) ≡ − Σ 2 ( ϵ ) . (61) \frac{\delta\Phi[G]}{\delta G_{1}(\epsilon)}\equiv\Sigma_{1}(\epsilon),\qquad\frac{\delta\Phi[G]}{\delta G_{2}(\epsilon)}\equiv-\Sigma_{2}(\epsilon). \tag*{(61)} δ G 1 ( ϵ ) δ Φ [ G ] ≡ Σ 1 ( ϵ ) , δ G 2 ( ϵ ) δ Φ [ G ] ≡ − Σ 2 ( ϵ ) . (61)
By substituting (61) into (60), the Luttinger integral can be compactly recast as:
I L = − ∫ − ∞ ∞ d ϵ 2 π ( Tr ( δ Φ [ G ] δ G 1 ( ϵ ) ⋅ ∂ ϵ G 2 ( ϵ ) ) − Tr ( δ Φ [ G ] δ G 2 ( ϵ ) ⋅ ∂ ϵ G 1 ( ϵ ) ) ) = ∫ ∞ − ∞ d ϵ 2 π ( Tr ( δ Φ [ G ] δ G 1 ( ϵ ) ⋅ ∂ ϵ G 2 ( ϵ ) ) − Tr ( δ Φ [ G ] δ G 2 ( ϵ ) ⋅ ∂ ϵ G 1 ( ϵ ) ) ) = 1 2 π ∮ γ { Tr ( δ Φ [ G ] δ G 1 ⋅ d G 2 ) − Tr ( δ Φ [ G ] δ G 2 ⋅ d G 1 ) } . (62) \begin{aligned}
I_{L}&=-\int_{-\infty}^{\infty}\frac{d\epsilon}{2\pi}\left(\operatorname{Tr}\left(\frac{\delta\Phi[G]}{\delta G_{1}(\epsilon)}\cdot\partial_{\epsilon}G_{2}(\epsilon)\right)-\operatorname{Tr}\left(\frac{\delta\Phi[G]}{\delta G_{2}(\epsilon)}\cdot\partial_{\epsilon}G_{1}(\epsilon)\right)\right)\\
&=\int_{\infty}^{-\infty}\frac{d\epsilon}{2\pi}\left(\operatorname{Tr}\left(\frac{\delta\Phi[G]}{\delta G_{1}(\epsilon)}\cdot\partial_{\epsilon}G_{2}(\epsilon)\right)-\operatorname{Tr}\left(\frac{\delta\Phi[G]}{\delta G_{2}(\epsilon)}\cdot\partial_{\epsilon}G_{1}(\epsilon)\right)\right)\\
&=\frac{1}{2\pi}\oint_{\gamma}\left\{\operatorname{Tr}\left(\frac{\delta\Phi[G]}{\delta G_{1}}\cdot dG_{2}\right)-\operatorname{Tr}\left(\frac{\delta\Phi[G]}{\delta G_{2}}\cdot dG_{1}\right)\right\}. \tag*{(62)}
\end{aligned} I L = − ∫ − ∞ ∞ 2 π d ϵ ( Tr ( δ G 1 ( ϵ ) δ Φ [ G ] ⋅ ∂ ϵ G 2 ( ϵ ) ) − Tr ( δ G 2 ( ϵ ) δ Φ [ G ] ⋅ ∂ ϵ G 1 ( ϵ ) ) ) = ∫ ∞ − ∞ 2 π d ϵ ( Tr ( δ G 1 ( ϵ ) δ Φ [ G ] ⋅ ∂ ϵ G 2 ( ϵ ) ) − Tr ( δ G 2 ( ϵ ) δ Φ [ G ] ⋅ ∂ ϵ G 1 ( ϵ ) ) ) = 2 π 1 ∮ γ { Tr ( δ G 1 δ Φ [ G ] ⋅ d G 2 ) − Tr ( δ G 2 δ Φ [ G ] ⋅ d G 1 ) } . (62)
This equation demonstrates that I L I_{L} I L is the line integral of a functional one-form derived from Φ [ G 1 , G 2 ] \Phi[G_{1},G_{2}] Φ [ G 1 , G 2 ] , specifically the conjugate differential form d c Φ [ G 1 , G 2 ] d^{c}\Phi[G_{1},G_{2}] d c Φ [ G 1 , G 2 ] , over a closed path γ = γ < + γ > \gamma=\gamma_{<}+\gamma_{>} γ = γ < + γ > . The initial trajectory γ < \gamma_{<} γ < begins at ( G 1 , G 2 ) = ( 0 , 0 − ) (G_{1},G_{2})=(0,0^{-}) ( G 1 , G 2 ) = ( 0 , 0 − ) , which corresponds to ϵ = ∞ \epsilon=\infty ϵ = ∞ . It traverses entirely within the lower half-plane ( G 2 < 0 ) (G_{2}<0) ( G 2 < 0 ) and concludes at ( G 1 , G 2 ) = ( G 1 ( 0 ) , 0 ) (G_{1},G_{2})=(G_{1}(0),0) ( G 1 , G 2 ) = ( G 1 ( 0 ) , 0 ) when ϵ = 0 + \epsilon=0^{+} ϵ = 0 + , consistent with our assumption of an insulator lacking Luttinger surfaces. The second half of the path, γ > \gamma_{>} γ > , is obtained by reflecting the first half: G 1 → G 1 G_{1}\to G_{1} G 1 → G 1 and G 2 → − G 2 G_{2}\to-G_{2} G 2 → − G 2 . It starts at ϵ = 0 − \epsilon=0^{-} ϵ = 0 − where, once again, ( G 1 , G 2 ) = ( G 1 ( 0 ) , 0 ) (G_{1},G_{2})=(G_{1}(0),0) ( G 1 , G 2 ) = ( G 1 ( 0 ) , 0 ) , and terminates at ( G 1 , G 2 ) = ( 0 , 0 + ) (G_{1},G_{2})=(0,0^{+}) ( G 1 , G 2 ) = ( 0 , 0 + ) , implying ϵ = − ∞ \epsilon=-\infty ϵ = − ∞ . Because physical boundary conditions require G ( i ∞ ) = G ( − i ∞ ) = 0 G(i\infty)=G(-i\infty)=0 G ( i ∞ ) = G ( − i ∞ ) = 0 , the Matsubara trajectory from ϵ = ∞ \epsilon=\infty ϵ = ∞ to ϵ = − ∞ \epsilon=-\infty ϵ = − ∞ natively forms a closed loop γ \gamma γ in the space of Green’s functions, see Figure 1 . Moreover, since γ > \gamma_{>} γ > is the mirror image of γ < \gamma_{<} γ < , then
Figure 1. Sketch of a generic path traversed by G = G 1 + i G 2 G = G_1 + i G_2 G = G 1 + i G 2 along the Matsubara trajectory from ϵ = + ∞ \epsilon= +\infty ϵ = + ∞ to ϵ = − ∞ \epsilon= -\infty ϵ = − ∞ in the case of a correlated insulator. Even though G 1 G_1 G 1 and G 2 G_2 G 2 are matrices that reside in a multidimensional space, for simplicity we project the path onto two dimensions as if they were scalars. The trajectory is composed of two symmetric segments: γ < \gamma_{<} γ < , in red, which is restricted to the lower half-plane (G 2 < 0 G_2 < 0 G 2 < 0 ), and its reflected path γ > \gamma_{>} γ > (G 2 → − G 2 G_2 \to-G_2 G 2 → − G 2 ), in blue. The crucial point along γ \gamma γ is the reflection point (indicated in green), which corresponds to Matsubara frequency ϵ = 0 \epsilon= 0 ϵ = 0 . This point marks where the physical trajectory intersects a branch cut, explicitly generating the non-vanishing I L I_L I L whenever Luttinger’s theorem is violated. Specifically, I L I_L I L can change only if one or more eigenvalues of G 1 ( 0 ) G_1(0) G 1 ( 0 ) , which is Hermitian, vanish. In the simplified two-dimensional projection, this occurs when the reflection point reaches the origin ( G 1 , G 2 ) = ( 0 , 0 ) (G_1, G_2) = (0, 0) ( G 1 , G 2 ) = ( 0 , 0 ) (indicated by the dotted path in the figure) and subsequently moves away, carrying the topological defect with it.
δ Φ [ G ] δ G 1 → γ < ↔ γ > δ Φ [ G ] δ G 1 , δ Φ [ G ] δ G 2 → γ < ↔ γ > − δ Φ [ G ] δ G 2 , \frac{\delta\Phi[G]}{\delta G_{1}}\xrightarrow{\gamma_{<}\leftrightarrow\gamma_{>}}\frac{\delta\Phi[G]}{\delta G_{1}},\qquad
\frac{\delta\Phi[G]}{\delta G_{2}}\xrightarrow{\gamma_{<}\leftrightarrow\gamma_{>}}-\frac{\delta\Phi[G]}{\delta G_{2}}, δ G 1 δ Φ [ G ] γ < ↔ γ > δ G 1 δ Φ [ G ] , δ G 2 δ Φ [ G ] γ < ↔ γ > − δ G 2 δ Φ [ G ] ,
d G 1 → γ < ↔ γ > − d G 1 , d G 2 → γ < ↔ γ > d G 2 , dG_{1}\xrightarrow{\gamma_{<}\leftrightarrow\gamma_{>}}-dG_{1},\qquad dG_{2}\xrightarrow{\gamma_{<}\leftrightarrow\gamma_{>}}dG_{2}, d G 1 γ < ↔ γ > − d G 1 , d G 2 γ < ↔ γ > d G 2 ,
which implies that
1 2 π ∫ γ > Tr ( d c Φ [ G 1 , G 2 ] ) = 1 2 π ∫ γ < Tr ( d c Φ [ G 1 , G 2 ] ) . (63) \frac{1}{2\pi}\int_{\gamma_{>}}\operatorname{Tr}\left(d^{c}\Phi[G_{1},G_{2}]\right)=\frac{1}{2\pi}\int_{\gamma_{<}}\operatorname{Tr}\left(d^{c}\Phi[G_{1},G_{2}]\right). \tag*{(63)} 2 π 1 ∫ γ > Tr ( d c Φ [ G 1 , G 2 ] ) = 2 π 1 ∫ γ < Tr ( d c Φ [ G 1 , G 2 ] ) . (63)
namely, that γ > \gamma_{>} γ > and γ < \gamma_{<} γ < add together constructively, in contrast to Tr ( d Φ [ G 1 , G 2 ] ) \operatorname{Tr}\left(d\Phi[G_{1},G_{2}]\right) Tr ( d Φ [ G 1 , G 2 ] ) where they cancel each other. To prove that I L I_{L} I L in Eq. (62) is indeed a topological invariant, we evaluate how it changes under an adiabatic variation δ G ( i ϵ ) \delta G(i\epsilon) δ G ( i ϵ ) of G ( i ϵ ) G(i\epsilon) G ( i ϵ ) that vanishes at ϵ = ± ∞ \epsilon= \pm\infty ϵ = ± ∞ , and such that neither δ G ( i ϵ ) \delta G(i\epsilon) δ G ( i ϵ ) nor the corresponding δ Σ ( i ϵ ) \delta\Sigma(i\epsilon) δ Σ ( i ϵ ) are singular at ϵ = 0 \epsilon= 0 ϵ = 0 . In other words, the variation does not transform the insulator into a metal, nor create a Luttinger surface. Since the functional derivative of Σ [ G ] \Sigma[G] Σ [ G ] is the irreducible vertex Γ 0 \Gamma_0 Γ 0 and coincides with Γ 0 S \Gamma_0^{S} Γ 0 S at zero transferred frequency and momentum, we readily find, using Eq. (98) , that
δ I L = ∫ − ∞ ∞ d ϵ 2 π Tr ( δ G ( i ϵ ) ⋅ ∂ i ϵ Σ ( i ϵ ) ) − ∫ − ∞ ∞ d ϵ 2 π Tr ( ∂ i ϵ G ( i ϵ ) ⋅ δ Σ ( i ϵ ) ) = ∫ − ∞ ∞ d ϵ 2 π d ϵ ′ 2 π Tr ( δ G ( i ϵ ) ⋅ Γ 0 S ( i ϵ , i ϵ ′ ) ⋅ ∂ i ϵ ′ G ( i ϵ ′ ) ) − ∫ − ∞ ∞ d ϵ 2 π d ϵ ′ 2 π Tr ( ∂ i ϵ G ( i ϵ ) ⋅ Γ 0 S ( i ϵ , i ϵ ′ ) ⋅ δ G ( i ϵ ′ ) ) = 0. (64) \begin{aligned}
\delta I_L &= \int_{-\infty}^{\infty}\frac{d\epsilon}{2\pi}\operatorname{Tr}\left(\delta G(i\epsilon)\cdot\partial_{i\epsilon}\Sigma(i\epsilon)\right)-\int_{-\infty}^{\infty}\frac{d\epsilon}{2\pi}\operatorname{Tr}\left(\partial_{i\epsilon}G(i\epsilon)\cdot\delta\Sigma(i\epsilon)\right) \\
&= \int_{-\infty}^{\infty}\frac{d\epsilon}{2\pi}\frac{d\epsilon'}{2\pi}\operatorname{Tr}\left(\delta G(i\epsilon)\cdot\Gamma_0^{S}(i\epsilon,i\epsilon')\cdot\partial_{i\epsilon'}G(i\epsilon')\right) \\
&\quad-\int_{-\infty}^{\infty}\frac{d\epsilon}{2\pi}\frac{d\epsilon'}{2\pi}\operatorname{Tr}\left(\partial_{i\epsilon}G(i\epsilon)\cdot\Gamma_0^{S}(i\epsilon,i\epsilon')\cdot\delta G(i\epsilon')\right)=0. \tag*{(64)}
\end{aligned} δ I L = ∫ − ∞ ∞ 2 π d ϵ Tr ( δ G ( i ϵ ) ⋅ ∂ i ϵ Σ ( i ϵ ) ) − ∫ − ∞ ∞ 2 π d ϵ Tr ( ∂ i ϵ G ( i ϵ ) ⋅ δ Σ ( i ϵ ) ) = ∫ − ∞ ∞ 2 π d ϵ 2 π d ϵ ′ Tr ( δ G ( i ϵ ) ⋅ Γ 0 S ( i ϵ , i ϵ ′ ) ⋅ ∂ i ϵ ′ G ( i ϵ ′ ) ) − ∫ − ∞ ∞ 2 π d ϵ 2 π d ϵ ′ Tr ( ∂ i ϵ G ( i ϵ ) ⋅ Γ 0 S ( i ϵ , i ϵ ′ ) ⋅ δ G ( i ϵ ′ ) ) = 0. (64)
Therefore, I L I_L I L in (62) is invariant under any smooth deformation of the path γ \gamma γ . Moreover, in light of the relation connecting d c Φ [ G 1 , G 2 ] d^{c}\Phi[G_1,G_2] d c Φ [ G 1 , G 2 ] to ∂ i ϵ Tr ( ln X ( i ϵ ) ) \partial_{i\epsilon}\operatorname{Tr}\left(\ln X(i\epsilon)\right) ∂ i ϵ Tr ( ln X ( i ϵ ) ) in (26) , all the monodromy captured by I L I_L I L is concentrated exactly at the reflection point, the only point where, as discussed in § 1 , a physically meaningful discontinuity occurs.
Since I L I_L I L is a topological invariant quantized to integer values, it vanishes identically across regions of the Hamiltonian parameter space smoothly connected to the non-interacting limit, where I L I_L I L trivially vanishes since Σ ( i ϵ ) ≡ 0 \Sigma(i\epsilon) \equiv0 Σ ( i ϵ ) ≡ 0 . Upon the breakdown of (self-consistent) perturbation theory, the reflection point of the Matsubara trajectory cannot acquire a nontrivial anomaly unless a specific condition is met. In the multidimensional space where the matrices G 1 G_1 G 1 and G 2 G_2 G 2 reside, we can describe the coordinates of the reflection point ( G 1 ( 0 ) , 0 ) (G_1(0),0) ( G 1 ( 0 ) , 0 ) using the basis in which the Hermitian matrix G 1 ( 0 ) G_1(0) G 1 ( 0 ) is diagonal with eigenvalues ξ ℓ \xi_\ell ξ ℓ . In this representation, the reflection point is defined by the coordinates:
( G 1 ( 0 ) , 0 ) = ( ( ξ 1 , 0 ) , ( ξ 2 , 0 ) , … , ( ξ ℓ , 0 ) , … ) . (65) (G_1(0),0)=\bigl((\xi_1,0),(\xi_2,0),\ldots,(\xi_\ell,0),\ldots\bigr). \tag*{(65)} ( G 1 ( 0 ) , 0 ) = ( ( ξ 1 , 0 ) , ( ξ 2 , 0 ) , … , ( ξ ℓ , 0 ) , … ) . (65)
The monodromy encoded in I L I_L I L can only change if one or more of these tuples reach ( ξ ℓ , 0 ) = ( 0 , 0 ) (\xi_\ell,0)=(0,0) ( ξ ℓ , 0 ) = ( 0 , 0 ) and subsequently move away from the origin, carrying the topological defect with them. This transition occurs when G 2 ( 0 ) ≡ 0 G_2(0)\equiv0 G 2 ( 0 ) ≡ 0 and det ( G 1 ( 0 ) ) = 0 \det(G_1(0))=0 det ( G 1 ( 0 )) = 0 , which corresponds to a divergence of the self-energy at ϵ = 0 \epsilon=0 ϵ = 0 . In a periodic system, this divergence signals the existence of a so-called Luttinger surface [46 ] , which marks the location of the zeros [49 , 50 , 51 ] of det ( G ( 0 ) ) = det ( G 1 ( 0 ) ) \det(G(0))=\det(G_1(0)) det ( G ( 0 )) = det ( G 1 ( 0 )) in the Brillouin zone. As long as the Luttinger surface exists, Eq. (64) does not hold, implying that I L I_L I L is not topologically protected and can vary continuously. However, when the Luttinger surface disappears and the defect is acquired, both G ( i ϵ ) G(i\epsilon) G ( i ϵ ) and Σ ( i ϵ ) \Sigma(i\epsilon) Σ ( i ϵ ) may regain analyticity everywhere, including at ϵ = 0 \epsilon=0 ϵ = 0 . Nevertheless, the Luttinger integral remains non-zero and quantized. This observation is consistent with the example discussed at the end of § 2.5 , and points out that only when I L I_L I L changes can one find signals in the local-in-ϵ \epsilon ϵ behavior of Σ ( i ϵ ) \Sigma(i\epsilon) Σ ( i ϵ ) . Once I L I_L I L has changed, the anomaly becomes a global property of the LW functional unrelated to any infrared singularity.
A few relevant examples
Here we briefly examine a few simple examples that implement the ideas from the prior section 4.
Two-orbital toy model
When the interaction is turned on, at first it modifies the non-interacting ground state by transferring pairs of electrons from occupied to unoccupied single-particle levels. To mimic this effect, we consider a very simple model that consists of just two orbitals, 1 and 2, described by the Hamiltonian
H = ∑ σ = ↑ , ↓ ( ϵ 1 c 1 σ † c 1 σ + ϵ 2 c 2 σ † c 2 σ ) − J ( c 1 ↑ † c 1 ↓ † c 2 ↓ c 2 ↑ + c 2 ↑ † c 2 ↓ † c 1 ↓ c 1 ↑ ) . (66) H=\sum_{\sigma=\uparrow,\downarrow}\left(\epsilon_1 c_{1\sigma}^{\dagger}c_{1\sigma}+\epsilon_2 c_{2\sigma}^{\dagger}c_{2\sigma}\right)-J\left(c_{1\uparrow}^{\dagger}c_{1\downarrow}^{\dagger}c_{2\downarrow}c_{2\uparrow}+c_{2\uparrow}^{\dagger}c_{2\downarrow}^{\dagger}c_{1\downarrow}c_{1\uparrow}\right). \tag*{(66)} H = σ =↑ , ↓ ∑ ( ϵ 1 c 1 σ † c 1 σ + ϵ 2 c 2 σ † c 2 σ ) − J ( c 1 ↑ † c 1 ↓ † c 2 ↓ c 2 ↑ + c 2 ↑ † c 2 ↓ † c 1 ↓ c 1 ↑ ) . (66)
Without loss of generality, we assume that ϵ 1 < ϵ 2 \epsilon_1<\epsilon_2 ϵ 1 < ϵ 2 . The lowest energy state with two electrons has energy
E 0 = ϵ 1 + ϵ 2 − ( ϵ 2 − ϵ 1 ) 2 + J 2 : = ϵ 1 + ϵ 2 − Δ , (67) E_0=\epsilon_1+\epsilon_2-\sqrt{(\epsilon_2-\epsilon_1)^2+J^2}:=\epsilon_1+\epsilon_2-\Delta, \tag*{(67)} E 0 = ϵ 1 + ϵ 2 − ( ϵ 2 − ϵ 1 ) 2 + J 2 := ϵ 1 + ϵ 2 − Δ , (67)
and reads
∣ ψ 0 ⟩ = cos θ c 1 ↑ † c 1 ↓ † ∣ 0 ⟩ + sin θ c 2 ↑ † c 2 ↓ † ∣ 0 ⟩ , (68) |\psi_0\rangle=\cos\theta\,c_{1\uparrow}^{\dagger}c_{1\downarrow}^{\dagger}|0\rangle+\sin\theta\,c_{2\uparrow}^{\dagger}c_{2\downarrow}^{\dagger}|0\rangle, \tag*{(68)} ∣ ψ 0 ⟩ = cos θ c 1 ↑ † c 1 ↓ † ∣0 ⟩ + sin θ c 2 ↑ † c 2 ↓ † ∣0 ⟩ , (68)
where here and in the next section ∣ 0 ⟩ |0\rangle ∣0 ⟩ denotes the vacuum state, and
cos 2 θ = ϵ 2 − ϵ 1 Δ , sin 2 θ = J Δ . \cos2\theta=\frac{\epsilon_2-\epsilon_1}{\Delta},\qquad\sin2\theta=\frac{J}{\Delta}. cos 2 θ = Δ ϵ 2 − ϵ 1 , sin 2 θ = Δ J .
We further assume that ∣ ψ 0 ⟩ |\psi_0\rangle ∣ ψ 0 ⟩ is the ground state, which implies the following inequality
Δ > max ( − ϵ 1 , ϵ 2 , − ϵ 1 − ϵ 2 , ϵ 1 + ϵ 2 ) . (69) \Delta> \max\left(-\epsilon_1,\epsilon_2,-\epsilon_1-\epsilon_2,\epsilon_1+\epsilon_2\right). \tag*{(69)} Δ > max ( − ϵ 1 , ϵ 2 , − ϵ 1 − ϵ 2 , ϵ 1 + ϵ 2 ) . (69)
The Green’s function is diagonal in the orbital index, is spin independent, and its matrix elements read:
G 11 ( i ϵ ) = i ϵ − 2 ϵ 2 + ϵ 1 ( i ϵ − ϵ 2 + Δ ) ( i ϵ − ϵ 2 − Δ ) , G 22 ( i ϵ ) = i ϵ − 2 ϵ 1 + ϵ 2 ( i ϵ − ϵ 1 + Δ ) ( i ϵ − ϵ 1 − Δ ) , (70) G_{11}(i\epsilon)=\frac{i\epsilon-2\epsilon_2+\epsilon_1}{(i\epsilon-\epsilon_2+\Delta)(i\epsilon-\epsilon_2-\Delta)},\quad G_{22}(i\epsilon)=\frac{i\epsilon-2\epsilon_1+\epsilon_2}{(i\epsilon-\epsilon_1+\Delta)(i\epsilon-\epsilon_1-\Delta)}, \tag*{(70)} G 11 ( i ϵ ) = ( i ϵ − ϵ 2 + Δ ) ( i ϵ − ϵ 2 − Δ ) i ϵ − 2 ϵ 2 + ϵ 1 , G 22 ( i ϵ ) = ( i ϵ − ϵ 1 + Δ ) ( i ϵ − ϵ 1 − Δ ) i ϵ − 2 ϵ 1 + ϵ 2 , (70)
and correspond to a diagonal self-energy with spin-independent elements
Σ 11 ( i ϵ ) = J 2 i ϵ − 2 ϵ 2 + ϵ 1 , Σ 22 ( i ϵ ) = J 2 i ϵ − 2 ϵ 1 + ϵ 2 . (71) \Sigma_{11}(i\epsilon)=\frac{J^2}{i\epsilon-2\epsilon_2+\epsilon_1},\quad\Sigma_{22}(i\epsilon)=\frac{J^2}{i\epsilon-2\epsilon_1+\epsilon_2}. \tag*{(71)} Σ 11 ( i ϵ ) = i ϵ − 2 ϵ 2 + ϵ 1 J 2 , Σ 22 ( i ϵ ) = i ϵ − 2 ϵ 1 + ϵ 2 J 2 . (71)
It is easy to show that
I L = ∑ σ = ↑ , ↓ ∑ i = 1 2 ∫ − ∞ ∞ d ϵ 2 π G i i ( i ϵ ) ∂ Σ i i ( i ϵ ) ∂ i ϵ ≡ ∫ − ∞ ∞ d ϵ 2 π ∂ Tr ( ln X ( i ϵ ) ) ∂ i ϵ , (72) I_L=\sum_{\sigma=\uparrow,\downarrow}\sum_{i=1}^{2}\int_{-\infty}^{\infty}\frac{d\epsilon}{2\pi}\,G_{ii}(i\epsilon)\frac{\partial\Sigma_{ii}(i\epsilon)}{\partial i\epsilon}\equiv\int_{-\infty}^{\infty}\frac{d\epsilon}{2\pi}\frac{\partial\operatorname{Tr}\left(\ln X(i\epsilon)\right)}{\partial i\epsilon}, \tag*{(72)} I L = σ =↑ , ↓ ∑ i = 1 ∑ 2 ∫ − ∞ ∞ 2 π d ϵ G ii ( i ϵ ) ∂ i ϵ ∂ Σ ii ( i ϵ ) ≡ ∫ − ∞ ∞ 2 π d ϵ ∂ i ϵ ∂ Tr ( ln X ( i ϵ ) ) , (72)
where one readily finds that the matrix X ( i ϵ ) X(i\epsilon) X ( i ϵ ) is also diagonal with elements
X 1 σ ( i ϵ ) = ( i ϵ − 2 ϵ 2 + ϵ 1 ) ( i ϵ − ϵ 2 + Δ ) ( ϵ 2 − ϵ 1 − Δ ) / 2 Δ ( i ϵ − ϵ 2 − Δ ) ( ϵ 1 − ϵ 2 − Δ ) / 2 Δ , X_{1\sigma}(i\epsilon)=(i\epsilon-2\epsilon_2+\epsilon_1)(i\epsilon-\epsilon_2+\Delta)^{(\epsilon_2-\epsilon_1-\Delta)/2\Delta}(i\epsilon-\epsilon_2-\Delta)^{(\epsilon_1-\epsilon_2-\Delta)/2\Delta}, X 1 σ ( i ϵ ) = ( i ϵ − 2 ϵ 2 + ϵ 1 ) ( i ϵ − ϵ 2 + Δ ) ( ϵ 2 − ϵ 1 − Δ ) /2Δ ( i ϵ − ϵ 2 − Δ ) ( ϵ 1 − ϵ 2 − Δ ) /2Δ ,
X 2 σ ( i ϵ ) = ( i ϵ − 2 ϵ 1 + ϵ 2 ) ( i ϵ − ϵ 1 + Δ ) ( ϵ 1 − ϵ 2 − Δ ) / 2 Δ ( i ϵ − ϵ 1 − Δ ) ( ϵ 2 − ϵ 1 − Δ ) / 2 Δ . (73) X_{2\sigma}(i\epsilon)=(i\epsilon-2\epsilon_1+\epsilon_2)(i\epsilon-\epsilon_1+\Delta)^{(\epsilon_1-\epsilon_2-\Delta)/2\Delta}(i\epsilon-\epsilon_1-\Delta)^{(\epsilon_2-\epsilon_1-\Delta)/2\Delta}. \tag*{(73)} X 2 σ ( i ϵ ) = ( i ϵ − 2 ϵ 1 + ϵ 2 ) ( i ϵ − ϵ 1 + Δ ) ( ϵ 1 − ϵ 2 − Δ ) /2Δ ( i ϵ − ϵ 1 − Δ ) ( ϵ 2 − ϵ 1 − Δ ) /2Δ . (73)
The constraint (69) implies
Δ − ϵ 1 > Δ − ϵ 2 > 0 , Δ + ϵ 2 > Δ + ϵ 1 > 0 , \Delta-\epsilon_1>\Delta-\epsilon_2>0,\quad\Delta+\epsilon_2>\Delta+\epsilon_1>0, Δ − ϵ 1 > Δ − ϵ 2 > 0 , Δ + ϵ 2 > Δ + ϵ 1 > 0 ,
so that (72) becomes simply
I L = 2 ( 1 − θ ( 2 ϵ 2 − ϵ 1 ) − θ ( 2 ϵ 1 − ϵ 2 ) ) . (74) I_L = 2\left(1-\theta(2\epsilon_2-\epsilon_1)-\theta(2\epsilon_1-\epsilon_2)\right). \tag*{(74)} I L = 2 ( 1 − θ ( 2 ϵ 2 − ϵ 1 ) − θ ( 2 ϵ 1 − ϵ 2 ) ) . (74)
To simplify the graphical representation of I L I_L I L , we write ϵ 1 = − E − μ \epsilon_1=-E-\mu ϵ 1 = − E − μ , ϵ 2 = E − μ \epsilon_2=E-\mu ϵ 2 = E − μ with 0 < E ≡ 1 0<E\equiv1 0 < E ≡ 1 the unit of energy. It follows that 2 ϵ 2 − ϵ 1 ≡ 3 − μ 2\epsilon_2-\epsilon_1\equiv3-\mu 2 ϵ 2 − ϵ 1 ≡ 3 − μ , 2 ϵ 1 − ϵ 2 ≡ − 3 − μ 2\epsilon_1-\epsilon_2\equiv-3-\mu 2 ϵ 1 − ϵ 2 ≡ − 3 − μ , thus
I L = 2 ( 1 − θ ( 3 − μ ) − θ ( − 3 − μ ) ) , (75) I_L = 2\left(1-\theta(3-\mu)-\theta(-3-\mu)\right), \tag*{(75)} I L = 2 ( 1 − θ ( 3 − μ ) − θ ( − 3 − μ ) ) , (75)
with the constraint
Δ = 4 + J 2 > max ( 1 + μ , 1 − μ , 2 μ , − 2 μ ) . (76) \Delta=\sqrt{4+J^2}>\max(1+\mu,1-\mu,2\mu,-2\mu). \tag*{(76)} Δ = 4 + J 2 > max ( 1 + μ , 1 − μ , 2 μ , − 2 μ ) . (76)
In Figure 2 we plot I L I_L I L in the region where the ground state has two electrons, i.e. when the inequality (76) is valid. We observe that Luttinger’s theorem may hold even when the non-interacting and interacting ground states are not adiabatically connected.
Figure 2. Luttinger integral corresponding to the Hamiltonian (66) with ϵ 1 = − E − μ \epsilon_1=-E-\mu ϵ 1 = − E − μ and ϵ 2 = E − μ \epsilon_2=E-\mu ϵ 2 = E − μ (E > 0 E>0 E > 0 ) in the μ / E \mu/E μ / E vs. J / E J/E J / E parameter space where the ground state has two electrons, N = 2 N=2 N = 2 (N ≠ 2 N\ne2 N = 2 in the gray regions). The horizontal lines μ = 3 E \mu=3E μ = 3 E and μ = − 3 E \mu=-3E μ = − 3 E correspond to the zeros of G 11 ( 0 ) G_{11}(0) G 11 ( 0 ) and G 22 ( 0 ) G_{22}(0) G 22 ( 0 ) , respectively.
Hubbard dimer
We consider now a Hubbard dimer with Hamiltonian
H = − t ∑ σ ( c 1 σ † c 2 σ + c 2 σ † c 1 σ ) − μ ∑ i = 1 2 ∑ σ n i σ + U ∑ i = 1 2 n i ↑ n i ↓ , (77) H=-t\sum_{\sigma}\left(c^\dagger_{1\sigma}c_{2\sigma}+c^\dagger_{2\sigma}c_{1\sigma}\right)-\mu\sum_{i=1}^{2}\sum_{\sigma}n_{i\sigma}+U\sum_{i=1}^{2}n_{i\uparrow}n_{i\downarrow}, \tag*{(77)} H = − t σ ∑ ( c 1 σ † c 2 σ + c 2 σ † c 1 σ ) − μ i = 1 ∑ 2 σ ∑ n iσ + U i = 1 ∑ 2 n i ↑ n i ↓ , (77)
with both t t t and U U U positive. The Hamiltonian (77) is invariant under 1 ↔ 2 1\leftrightarrow2 1 ↔ 2 , which implies that the bonding b b b and antibonding a a a combinations
c b σ = 1 2 ( c 1 σ + c 2 σ ) , c a σ = 1 2 ( c 1 σ − c 2 σ ) , (78) c_{b\sigma}=\frac{1}{\sqrt{2}}\left(c_{1\sigma}+c_{2\sigma}\right),\qquad c_{a\sigma}=\frac{1}{\sqrt{2}}\left(c_{1\sigma}-c_{2\sigma}\right), \tag*{(78)} c bσ = 2 1 ( c 1 σ + c 2 σ ) , c aσ = 2 1 ( c 1 σ − c 2 σ ) , (78)
are irreducible representations of the symmetry group. The lowest energy state with two electrons is
∣ ψ 0 ⟩ = cos θ c b ↑ † c b ↓ † ∣ 0 ⟩ − sin θ c a ↑ † c a ↓ † ∣ 0 ⟩ , (79) |\psi_0\rangle=\cos\theta\,c^\dagger_{b\uparrow}c^\dagger_{b\downarrow}|0\rangle-\sin\theta\,c^\dagger_{a\uparrow}c^\dagger_{a\downarrow}|0\rangle, \tag*{(79)} ∣ ψ 0 ⟩ = cos θ c b ↑ † c b ↓ † ∣0 ⟩ − sin θ c a ↑ † c a ↓ † ∣0 ⟩ , (79)
with energy
E 0 = − 2 μ + U 2 − ( U 2 ) 2 + 4 t 2 = − 2 μ + U 2 − Δ , (80) E_0=-2\mu+\frac{U}{2}-\sqrt{\left(\frac{U}{2}\right)^2+4t^2}=-2\mu+\frac{U}{2}-\Delta, \tag*{(80)} E 0 = − 2 μ + 2 U − ( 2 U ) 2 + 4 t 2 = − 2 μ + 2 U − Δ , (80)
and where
cos 2 θ = 2 t Δ , sin 2 θ = U 2 Δ . \cos2\theta=\frac{2t}{\Delta},\qquad\sin2\theta=\frac{U}{2\Delta}. cos 2 θ = Δ 2 t , sin 2 θ = 2Δ U .
The state ∣ ψ 0 ⟩ |\psi_0\rangle ∣ ψ 0 ⟩ in (79) is the actual ground state if
μ min ( t , U ) ≤ μ ≤ μ max ( t , U ) , (81) \mu_{\min}(t,U)\leq\mu\leq\mu_{\max}(t,U), \tag*{(81)} μ m i n ( t , U ) ≤ μ ≤ μ m a x ( t , U ) , (81)
where
μ min ( t , U ) = U 2 + t − Δ , μ max ( t , U ) = U 2 − t + Δ . \mu_{\min}(t,U)=\frac{U}{2}+t-\Delta,\qquad\mu_{\max}(t,U)=\frac{U}{2}-t+\Delta. μ m i n ( t , U ) = 2 U + t − Δ , μ m a x ( t , U ) = 2 U − t + Δ.
In particular, the non-interacting and interacting ground states are adiabatically connected if μ min ( t , U ) ≤ μ ≤ μ max ( t , 0 ) = t \mu_{\min}(t,U)\leq\mu\leq\mu_{\max}(t,0)=t μ m i n ( t , U ) ≤ μ ≤ μ m a x ( t , 0 ) = t . Assuming (81), the Green’s function, which is diagonal in the bonding-antibonding basis (78) , has elements
G b ( i ϵ ) = sin 2 θ i ϵ + μ − 2 t − μ max ( t , U ) + cos 2 θ i ϵ + μ − μ min ( t , U ) = i ϵ + μ − 3 t − U / 2 ( i ϵ + μ − 2 t − μ max ( t , U ) ) ( i ϵ + μ − μ min ( t , U ) ) , G a ( i ϵ ) = cos 2 θ i ϵ + μ − μ max ( t , U ) + sin 2 θ i ϵ + μ + 2 t − μ min ( t , U ) = i ϵ + μ + 3 t − U / 2 ( i ϵ + μ − μ max ( t , U ) ) ( i ϵ + μ + 2 t − μ min ( t , U ) ) . (82) \begin{aligned}
G_b(i\epsilon) &= \frac{\sin^2\theta}{i\epsilon+\mu-2t-\mu_{\max}(t,U)}+\frac{\cos^2\theta}{i\epsilon+\mu-\mu_{\min}(t,U)} \\
&= \frac{i\epsilon+\mu-3t-U/2}{(i\epsilon+\mu-2t-\mu_{\max}(t,U))(i\epsilon+\mu-\mu_{\min}(t,U))}, \\
G_a(i\epsilon) &= \frac{\cos^2\theta}{i\epsilon+\mu-\mu_{\max}(t,U)}+\frac{\sin^2\theta}{i\epsilon+\mu+2t-\mu_{\min}(t,U)} \\
&= \frac{i\epsilon+\mu+3t-U/2}{(i\epsilon+\mu-\mu_{\max}(t,U))(i\epsilon+\mu+2t-\mu_{\min}(t,U))}. \tag*{(82)}
\end{aligned} G b ( i ϵ ) G a ( i ϵ ) = i ϵ + μ − 2 t − μ m a x ( t , U ) sin 2 θ + i ϵ + μ − μ m i n ( t , U ) cos 2 θ = ( i ϵ + μ − 2 t − μ m a x ( t , U )) ( i ϵ + μ − μ m i n ( t , U )) i ϵ + μ − 3 t − U /2 , = i ϵ + μ − μ m a x ( t , U ) cos 2 θ + i ϵ + μ + 2 t − μ m i n ( t , U ) sin 2 θ = ( i ϵ + μ − μ m a x ( t , U )) ( i ϵ + μ + 2 t − μ m i n ( t , U )) i ϵ + μ + 3 t − U /2 . (82)
We observe that G b ( 0 ) > 0 G_b(0)>0 G b ( 0 ) > 0 if μ < 3 t + U / 2 \mu<3t+U/2 μ < 3 t + U /2 , otherwise G b ( 0 ) ≤ 0 G_b(0)\leq0 G b ( 0 ) ≤ 0 . Similarly, G a ( 0 ) < 0 G_a(0)<0 G a ( 0 ) < 0 if μ > − 3 t + U / 2 \mu>-3t+U/2 μ > − 3 t + U /2 , otherwise G a ( 0 ) ≥ 0 G_a(0)\geq0 G a ( 0 ) ≥ 0 . In Figure 3 we show the value of the Luttinger integral corresponding to the Hamiltonian (77) in the parameter space where the ground state has two electrons, i.e. when Eq. (81) holds. We observe that if μ min ( t , U ) ≤ μ ≤ t \mu_{\min}(t,U)\leq\mu\leq t μ m i n ( t , U ) ≤ μ ≤ t , i.e. when the non-interacting and interacting ground states are adiabatically connected, Luttinger’s theorem can be still violated, the blue region with I L = − 2 I_L=-2 I L = − 2 where μ min ( t , U ) ≤ μ ≤ − 3 t + U / 2 \mu_{\min}(t,U)\leq\mu\leq-3t+U/2 μ m i n ( t , U ) ≤ μ ≤ − 3 t + U /2 .
Figure 3. Luttinger integral corresponding to the Hamiltonian (77) in the μ \mu μ vs. U U U parameter space where the ground state has two electrons. The dashed and dotted lines correspond to the zeros of G b ( 0 ) G_b(0) G b ( 0 ) and G a ( 0 ) G_a(0) G a ( 0 ) , respectively.
Hubbard plaquette pierced by a magnetic flux
The next example we consider is that of a single-orbital Hubbard model on a plaquette pierced by a magnetic flux ϕ \phi ϕ and with nearest and next-nearest neighbor hopping, see Figure 4 . This is a very simple model where we can investigate the validity of Luttinger’s theorem with broken time-reversal symmetry. The Hamiltonian reads
Figure 4. Tight-binding model with nearest, − t -t − t , and next-nearest, − t 0 -t_0 − t 0 , neighbor hopping on a plaquette pierced by a magnetic flux ϕ \phi ϕ .
H = H 0 − μ ∑ i = 0 3 ∑ σ = ↑ , ↓ n i σ + U ∑ i = 0 3 n i ↑ n i ↓ , (83) H = H_0 - \mu\sum_{i=0}^{3}\sum_{\sigma=\uparrow,\downarrow} n_{i\sigma} + U \sum_{i=0}^{3} n_{i\uparrow}n_{i\downarrow}, \tag*{(83)} H = H 0 − μ i = 0 ∑ 3 σ =↑ , ↓ ∑ n iσ + U i = 0 ∑ 3 n i ↑ n i ↓ , (83)
where n i σ = c i σ † c i σ n_{i\sigma} = c_{i\sigma}^{\dagger}c_{i\sigma} n iσ = c iσ † c iσ is the occupation number of spin σ = ↑ , ↓ \sigma=\uparrow,\downarrow σ =↑ , ↓ electrons on site i = 0 , … , 3 i=0,\ldots,3 i = 0 , … , 3 , and the non-interacting Hamiltonian is, see Figure 4 ,
H 0 = − t ∑ σ ∑ i = 0 3 ( e i ϕ / 4 c i σ † c i + 1 σ + e − i ϕ / 4 c i + 1 σ † c i σ ) − t 0 ∑ σ ( c 2 σ † c 0 σ + c 0 σ † c 2 σ + c 1 σ † c 3 σ + c 3 σ † c 1 σ ) , (84) H_0 = -t\sum_{\sigma}\sum_{i=0}^{3}\left(e^{i\phi/4}c_{i\sigma}^{\dagger}c_{i+1\sigma}+e^{-i\phi/4}c_{i+1\sigma}^{\dagger}c_{i\sigma}\right)
-t_0\sum_{\sigma}\left(c_{2\sigma}^{\dagger}c_{0\sigma}+c_{0\sigma}^{\dagger}c_{2\sigma}+c_{1\sigma}^{\dagger}c_{3\sigma}+c_{3\sigma}^{\dagger}c_{1\sigma}\right), \tag*{(84)} H 0 = − t σ ∑ i = 0 ∑ 3 ( e i ϕ /4 c iσ † c i + 1 σ + e − i ϕ /4 c i + 1 σ † c iσ ) − t 0 σ ∑ ( c 2 σ † c 0 σ + c 0 σ † c 2 σ + c 1 σ † c 3 σ + c 3 σ † c 1 σ ) , (84)
where c 4 σ ≡ c 0 σ c_{4\sigma}\equiv c_{0\sigma} c 4 σ ≡ c 0 σ . At finite flux, the Hamiltonian (83) is only invariant under the point group C 4 C_4 C 4 that has four one-dimensional irreducible representations (irreps); two real ones, A A A and B B B , respectively, even and odd under C 4 C_4 C 4 , and two complex ones, E 1 E_1 E 1 and E 2 E_2 E 2 . The eigenvalues of H 0 H_0 H 0 can be readily obtained:
ϵ A ( ϕ ) = − t 0 − 2 t cos ϕ 4 , ϵ B ( ϕ ) = − t 0 + 2 t cos ϕ 4 , ϵ E 1 ( ϕ ) = t 0 − 2 t sin ϕ 4 , ϵ E 2 ( ϕ ) = t 0 + 2 t sin ϕ 4 , (85) \begin{aligned}
\epsilon_A(\phi) &= -t_0-2t\cos\frac{\phi}{4}, \qquad&\epsilon_B(\phi) &= -t_0+2t\cos\frac{\phi}{4},\\
\epsilon_{E_1}(\phi) &= t_0-2t\sin\frac{\phi}{4}, \qquad&\epsilon_{E_2}(\phi) &= t_0+2t\sin\frac{\phi}{4}, \tag*{(85)}
\end{aligned} ϵ A ( ϕ ) ϵ E 1 ( ϕ ) = − t 0 − 2 t cos 4 ϕ , = t 0 − 2 t sin 4 ϕ , ϵ B ( ϕ ) ϵ E 2 ( ϕ ) = − t 0 + 2 t cos 4 ϕ , = t 0 + 2 t sin 4 ϕ , (85)
where the states are labeled as the corresponding irreps, and are periodic in ϕ \phi ϕ with period 4 π 4\pi 4 π . Their evolution with ϕ \phi ϕ is non-trivial because of several level crossings, see Figure 5 . Hereafter, we set t = 1 t=1 t = 1 as the unit of energy, and assume 0 ≤ t 0 ≤ t = 1 0\leq t_0\leq t=1 0 ≤ t 0 ≤ t = 1 .
Figure 5. Non-interacting eigenvalues, ϵ n ( ϕ ) \epsilon_n(\phi) ϵ n ( ϕ ) , n = A , B , E 1 , E 2 n = A, B, E_1, E_2 n = A , B , E 1 , E 2 , as function of ϕ \phi ϕ . The values of μ \mu μ such that the non-interacting ground state has N = 4 N = 4 N = 4 electrons at flux ϕ \phi ϕ are marked by a yellow region.
The lowest energy eigenstate with N = 4 N=4 N = 4 electrons is non-degenerate and resides in the spin-singlet subspace of B B B symmetry for any ϕ \phi ϕ at U > 0 U>0 U > 0 , and only for ϕ ≠ 0 , 2 π , 4 π \phi\neq0,2\pi,4\pi ϕ = 0 , 2 π , 4 π at U = 0 U=0 U = 0 . In Figure 6 we show Luttinger integral as function of ϕ \phi ϕ and μ \mu μ at fixed t 0 = 0.1 t t_0=0.1t t 0 = 0.1 t and U = 8 t U=8t U = 8 t .
Figure 6. Value of the Luttinger integral I L I_{\mathrm{L}} I L of the plaquette pierced by a magnetic flux as function of ϕ \phi ϕ and μ \mu μ at fixed t 0 = 0.1 t t_0=0.1t t 0 = 0.1 t and U = 8 t U=8t U = 8 t where the ground state has four electrons. The black lines correspond to the zeros of the Green’s function at frequency zero. The region where the ground state is adiabatically connected to the non-interacting ground state is colored in yellow.
The region where non-interacting and interacting ground states are adiabatically connected is marked in yellow, see also the inset. It is interesting to note that the shape of the region where I L = 0 I_L = 0 I L = 0 is similar to the region where the non-interacting ground state has four electrons (yellow region in Figure 5 ). To further illustrate the role of adiabatic continuity, Figure 7 shows the U U U vs. μ \mu μ phase diagram at two fixed values of the flux and t 0 = 0.1 t t_0 = 0.1t t 0 = 0.1 t . The shaded yellow area marks the region in which the interacting ground state is adiabatically connected to the non-interacting. For ϕ = 2 π / 3 \phi= 2\pi/3 ϕ = 2 π /3 [panel (a)] this region lies entirely within I L = 0 I_L = 0 I L = 0 , so Luttinger’s theorem holds wherever the ground state is adiabatically connected to the non-interacting one. For ϕ = 0.9 π \phi= 0.9\pi ϕ = 0.9 π [panel (b)], instead, the shaded region extends into the area with I L ≠ 0 I_L \ne0 I L = 0 .
Figure 7. Value of the Luttinger integral I L I_L I L of the plaquette pierced by a magnetic flux as function of U U U and μ \mu μ at fixed t 0 = 0.1 t t_0 = 0.1t t 0 = 0.1 t and ϕ = 2 π / 3 \phi= 2\pi/3 ϕ = 2 π /3 [panel (a)] and ϕ = 0.9 π \phi= 0.9\pi ϕ = 0.9 π [panel (b)] where the ground state has four electrons. The black lines correspond to the zeros of the Green’s function at frequency zero. The yellow region corresponds to the case in which non-interacting and interacting ground states are adiabatically connected.
Conclusions
In this work, we have provided a non-perturbative resolution of the long-standing apparent paradox surrounding Luttinger’s theorem violations in strongly correlated systems. By re-examining the properties of the Luttinger-Ward functional Φ [ G ] \Phi[G] Φ [ G ] under imaginary frequency translations, we demonstrated that reported violations of Luttinger’s theorem imply neither that Φ [ G ] \Phi[G] Φ [ G ] loses its invariance under frequency shifts nor that the skeleton diagram representation of physical quantities breaks down. In fact, we have shown that this representation, regardless of potential failures of the skeleton diagrammatic expansion , remains consistent with the thermodynamic Streda formula.
Specifically, considering the simple case of a correlated insulator, we find that the Green’s function G ( i ϵ , k ) = G ( − i ϵ , k ) † = G 1 ( ϵ , k ) + i G 2 ( ϵ , k ) G(i\epsilon,\mathbf{k}) = G(-i\epsilon,\mathbf{k})^\dagger= G_1(\epsilon,\mathbf{k}) + iG_2(\epsilon,\mathbf{k}) G ( i ϵ , k ) = G ( − i ϵ , k ) † = G 1 ( ϵ , k ) + i G 2 ( ϵ , k ) along the Matsubara trajectory from ϵ = + ∞ \epsilon= +\infty ϵ = + ∞ to ϵ = − ∞ \epsilon= -\infty ϵ = − ∞ traverses a closed path γ \gamma γ in the ( G 1 , G 2 ) (G_1,G_2) ( G 1 , G 2 ) space, as the one sketched in Figure 1 . One can readily demonstrate that, if L L L is the total number of available single-particle states (L = 2 M V L = 2MV L = 2 M V in a periodic model with V V V unit cells and M M M Bloch waves, with the factor 2 coming from the spin) then the Luttinger count (1) can be rewritten as3
N L = 1 2 π i ∮ γ d Tr ( ln G ) + L 2 , (86) N_{\mathrm{L}}=\frac{1}{2\pi i}\oint_{\gamma}\mathrm{d}\operatorname{Tr}(\ln G)+\frac{L}{2}, \tag*{(86)} N L = 2 π i 1 ∮ γ d Tr ( ln G ) + 2 L , (86)
where the trace includes summation over momentum, Bloch waves and spin. We have shown that the Luttinger integral I L I_{\mathrm{L}} I L is the circulation of the conjugate differential form d c Φ [ G 1 , G 2 ] \mathrm{d}^{c}\Phi[G_{1},G_{2}] d c Φ [ G 1 , G 2 ]
∑ ℓ = 1 L ⟨ c ℓ c ℓ † ⟩ = ∑ ℓ = 1 L ( 1 − ⟨ c ℓ † c ℓ ⟩ ) = L − N , \sum_{\ell=1}^{L}\langle c_{\ell}c_{\ell}^{\dagger}\rangle=\sum_{\ell=1}^{L}(1-\langle c_{\ell}^{\dagger}c_{\ell}\rangle)=L-N, ℓ = 1 ∑ L ⟨ c ℓ c ℓ † ⟩ = ℓ = 1 ∑ L ( 1 − ⟨ c ℓ † c ℓ ⟩) = L − N ,
similarly, if
N L = − ∫ − ∞ ∞ d ϵ 2 π e i ϵ 0 + ∂ ∂ i ϵ Tr ( ln G ( i ϵ ) ) = − ∫ − ∞ ∞ d ϵ 2 π e i ϵ 0 + ∂ ∂ i ϵ Tr ( ln G ∗ ( i ϵ ) ) , N_{\mathrm{L}}=-\int_{-\infty}^{\infty}\frac{\mathrm{d}\epsilon}{2\pi}e^{i\epsilon0^{+}}\frac{\partial}{\partial i\epsilon}\operatorname{Tr}\left(\ln G(i\epsilon)\right)=-\int_{-\infty}^{\infty}\frac{\mathrm{d}\epsilon}{2\pi}e^{i\epsilon0^{+}}\frac{\partial}{\partial i\epsilon}\operatorname{Tr}\left(\ln G_{*}(i\epsilon)\right), N L = − ∫ − ∞ ∞ 2 π d ϵ e i ϵ 0 + ∂ i ϵ ∂ Tr ( ln G ( i ϵ ) ) = − ∫ − ∞ ∞ 2 π d ϵ e i ϵ 0 + ∂ i ϵ ∂ Tr ( ln G ∗ ( i ϵ ) ) ,
then
N L − L = − ∫ − ∞ ∞ d ϵ 2 π e − i ϵ 0 + ∂ ∂ i ϵ Tr ( ln G ( i ϵ ) ) , N_{\mathrm{L}}-L=-\int_{-\infty}^{\infty}\frac{\mathrm{d}\epsilon}{2\pi}e^{-i\epsilon0^{+}}\frac{\partial}{\partial i\epsilon}\operatorname{Tr}\left(\ln G(i\epsilon)\right), N L − L = − ∫ − ∞ ∞ 2 π d ϵ e − i ϵ 0 + ∂ i ϵ ∂ Tr ( ln G ( i ϵ ) ) ,
which implies that
N L = − ∫ − ∞ ∞ d ϵ 2 π cos ( ϵ 0 + ) ∂ ∂ i ϵ Tr ( ln G ( i ϵ ) ) + L 2 = ∫ − ∞ ∞ d ϵ 2 π i ∂ ∂ ϵ Tr ( ln G ( i ϵ ) ) + L 2 , N_{\mathrm{L}}=-\int_{-\infty}^{\infty}\frac{\mathrm{d}\epsilon}{2\pi}\cos(\epsilon0^{+})\frac{\partial}{\partial i\epsilon}\operatorname{Tr}\left(\ln G(i\epsilon)\right)+\frac{L}{2}=\int_{-\infty}^{\infty}\frac{\mathrm{d}\epsilon}{2\pi i}\frac{\partial}{\partial\epsilon}\operatorname{Tr}\left(\ln G(i\epsilon)\right)+\frac{L}{2}, N L = − ∫ − ∞ ∞ 2 π d ϵ cos ( ϵ 0 + ) ∂ i ϵ ∂ Tr ( ln G ( i ϵ ) ) + 2 L = ∫ − ∞ ∞ 2 π i d ϵ ∂ ϵ ∂ Tr ( ln G ( i ϵ ) ) + 2 L ,
where we safely drop the cosine term, thus (86) . of the Luttinger-Ward functional along the same closed path, namely, via Eq. (62) :
I L = 1 2 π ∮ γ { Tr ( δ Φ [ G ] δ G 1 ⋅ d G 2 ) − Tr ( δ Φ [ G ] δ G 2 ⋅ d G 1 ) } . (87) I_L = \frac{1}{2\pi}\oint_{\gamma}\left\{\operatorname{Tr}\left(\frac{\delta\Phi[G]}{\delta G_1}\cdot dG_2\right)-\operatorname{Tr}\left(\frac{\delta\Phi[G]}{\delta G_2}\cdot dG_1\right)\right\}. \tag*{(87)} I L = 2 π 1 ∮ γ { Tr ( δ G 1 δ Φ [ G ] ⋅ d G 2 ) − Tr ( δ G 2 δ Φ [ G ] ⋅ d G 1 ) } . (87)
Also I L I_L I L is quantized in integer values and remains strictly invariant under any smooth deformation of path γ \gamma γ . It vanishes when the correlated insulator can be smoothly connected to a non-interacting band insulator, implying that Luttinger’s theorem is certainly valid within a finite parameter region adjacent to the non-interacting case. We remark that, see the examples discussed in Sec. 5, a smooth connection, as well as the associated validity of the theorem, does not reduce strictly to the fact that the interacting ground state can be adiabatically continued into the non-interacting one.
The Luttinger integral I L I_L I L can become non-zero, and Luttinger’s theorem violated, only if one or more eigenvalues of G 1 ( 0 ) G_1(0) G 1 ( 0 ) , which is Hermitian, vanish. This signals a divergence of the self-energy at ϵ = 0 \epsilon= 0 ϵ = 0 . Once the topological charge has changed, the self-energy can become analytic again, including at ϵ = 0 \epsilon= 0 ϵ = 0 . Because this condition of a diverging self-energy, the violation of the Luttinger theorem has been often associated to a breakdown of the skeleton expansion. Indeed, it appears that the conventional derivation [24 ] of the theorem, and thus the result I L = 0 I_L = 0 I L = 0 , would hold until the skeleton expansions converges (uniformly, or at least in a point-wise Lebesgue’s dominated convergence fashion 4 ) to the physical solutions.
Correspondingly, the Ishikawa-Matsuyama-Volovik topological invariant in two dimensions can be written, using the same notations as in § 3 , Eq. (58) , as
ν H L = − 2 π c e V ∂ N L ∂ B = − i π 3 1 V ∑ k ∫ d ϵ 2 π ϵ μ ν ρ Tr ( G ( i ϵ , k ) ⋅ ∂ μ G ( i ϵ , k ) − 1 ⋅ ∂ ν G ( i ϵ , k ) ⋅ ∂ ρ G ( i ϵ , k ) − 1 ) , (88) \nu_{\mathrm{HL}} = -\frac{2\pi c}{eV}\frac{\partial N_L}{\partial B}
= -i\frac{\pi}{3}\frac{1}{V}\sum_{\mathbf{k}}\int\frac{d\epsilon}{2\pi}\,\epsilon_{\mu\nu\rho}\operatorname{Tr}\left(G(i\epsilon,\mathbf{k})\cdot\partial_\mu G(i\epsilon,\mathbf{k})^{-1}\cdot\partial_\nu G(i\epsilon,\mathbf{k})\cdot\partial_\rho G(i\epsilon,\mathbf{k})^{-1}\right), \tag*{(88)} ν HL = − e V 2 π c ∂ B ∂ N L = − i 3 π V 1 k ∑ ∫ 2 π d ϵ ϵ μν ρ Tr ( G ( i ϵ , k ) ⋅ ∂ μ G ( i ϵ , k ) − 1 ⋅ ∂ ν G ( i ϵ , k ) ⋅ ∂ ρ G ( i ϵ , k ) − 1 ) , (88)
and represents the winding number of the map ( ϵ , k ) ↦ G ( i ϵ , k ) (\epsilon,\mathbf{k})\mapsto G(i\epsilon,\mathbf{k}) ( ϵ , k ) ↦ G ( i ϵ , k ) , which is quantized in integer values. However, the physical Hall number is actually ν H = ν H L + ν H L I \nu_{\mathrm{H}}=\nu_{\mathrm{HL}}+\nu_{\mathrm{HL}}^{I} ν H = ν HL + ν HL I , where the interaction correction reads, through Eq. (58) ,
ν H L I = − 2 π c e V ∂ I L ∂ B = − i π 3 1 V 2 ∑ k k ′ ∫ d ϵ 2 π d ϵ ′ 2 π ϵ μ ν ρ Tr ( ∂ μ G ( i ϵ , k ) ⋅ γ ν A ( i ϵ , k ; i ϵ ′ , k ′ ) ⋅ ∂ ρ G ( i ϵ ′ , k ′ ) ) . (89) \nu_{\mathrm{HL}}^{I}=-\frac{2\pi c}{eV}\frac{\partial I_L}{\partial B}
= -i\frac{\pi}{3}\frac{1}{V^2}\sum_{\mathbf{k}\mathbf{k}'}\int\frac{d\epsilon}{2\pi}\frac{d\epsilon'}{2\pi}\,\epsilon_{\mu\nu\rho}\operatorname{Tr}\left(\partial_\mu G(i\epsilon,\mathbf{k})\cdot\gamma_{\nu}^{A}(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}')\cdot\partial_\rho G(i\epsilon',\mathbf{k}')\right). \tag*{(89)} ν HL I = − e V 2 π c ∂ B ∂ I L = − i 3 π V 2 1 k k ′ ∑ ∫ 2 π d ϵ 2 π d ϵ ′ ϵ μν ρ Tr ( ∂ μ G ( i ϵ , k ) ⋅ γ ν A ( i ϵ , k ; i ϵ ′ , k ′ ) ⋅ ∂ ρ G ( i ϵ ′ , k ′ ) ) . (89)
Here, γ ν A \gamma_{\nu}^{A} γ ν A are the derivatives of the component Γ A \Gamma^{A} Γ A of the irreducible vertex antisymmetric under the exchange of the incoming particle-hole pair with the outgoing one, see Eq. (40) , taken with respect to the transferred frequency (∂ 0 ≡ ∂ i ω \partial_0\equiv\partial_{i\omega} ∂ 0 ≡ ∂ iω ) or the transferred momentum (∂ 1 ≡ ∂ q x \partial_1\equiv\partial_{q_x} ∂ 1 ≡ ∂ q x and ∂ 2 ≡ ∂ q y \partial_2\equiv\partial_{q_y} ∂ 2 ≡ ∂ q y ), evaluated at ω = q = 0 \omega=q=0 ω = q = 0 . Since γ ν A \gamma_{\nu}^{A} γ ν A are very peculiar higher order functional derivatives of Φ [ G ] \Phi[G] Φ [ G ] , we could not find a simple interpretation of (90) comparable to the ones we found for Eqs. (86) , (87) and (88) . Nonetheless, we can certainly state that ν H L I \nu_{\mathrm{HL}}^{I} ν HL I is also invariant under any non-singular variation of the Green’s function and that, when Luttinger’s theorem is violated and ν H ≠ ν H L \nu_{\mathrm{H}}\ne\nu_{\mathrm{HL}} ν H = ν HL , it can assume integer or fractional values. All these insights open up promising avenues for exploring how such anomalies manifest in other topological invariants governing strongly interacting quantum matter.
Acknowledgements
We are extremely grateful to Ivan Pasqua, Frank Ebel, and Evgeny Kozik for helpful discussions and comments.4
∫ − ∞ ∞ d ϵ 2 π Tr ( Σ [ n ] ( i ϵ ) ∂ G ( i ϵ ) ∂ i ϵ ) = 0 , ∀ n , (90) \int_{-\infty}^{\infty}\frac{d\epsilon}{2\pi}\operatorname{Tr}\left(\Sigma^{[n]}(i\epsilon)\frac{\partial G(i\epsilon)}{\partial i\epsilon}\right)=0,\qquad\forall n, \tag*{(90)} ∫ − ∞ ∞ 2 π d ϵ Tr ( Σ [ n ] ( i ϵ ) ∂ i ϵ ∂ G ( i ϵ ) ) = 0 , ∀ n , (90)
where Σ [ n ] ( i ϵ ) \Sigma^{[n]}(i\epsilon) Σ [ n ] ( i ϵ ) collects all skeleton diagrams of order n n n , evaluated at the exact G G G . Being linear in Σ \Sigma Σ , (88) holds for every partial sum S N ( i ϵ ) : = ∑ n ≤ N Σ [ n ] ( i ϵ ) S_N(i\epsilon):=\sum_{n\le N}\Sigma^{[n]}(i\epsilon) S N ( i ϵ ) := ∑ n ≤ N Σ [ n ] ( i ϵ ) . Whether it survives the limit N → ∞ N\to\infty N → ∞ is therefore a question of exchanging this limit with the frequency integral. By dominated convergence the exchange is justified, and thus I L = − lim N → ∞ ∫ d ϵ 2 π Tr ( S N ( i ϵ ) ∂ i ϵ G ) = 0 I_L=-\lim_{N\to\infty}\int\frac{d\epsilon}{2\pi}\operatorname{Tr}(S_N(i\epsilon)\partial_{i\epsilon}G)=0 I L = − lim N → ∞ ∫ 2 π d ϵ Tr ( S N ( i ϵ ) ∂ i ϵ G ) = 0 , if S N ( i ϵ ) S_N(i\epsilon) S N ( i ϵ ) converges to the physical Σ ( i ϵ ) \Sigma(i\epsilon) Σ ( i ϵ ) and there is an integrable function F ( ϵ ) F(\epsilon) F ( ϵ ) such that ∣ Tr ( S N ( i ϵ ) ∂ i ϵ G ( i ϵ ) ) ∣ ≤ F ( ϵ ) \left|\operatorname{Tr}(S_N(i\epsilon)\partial_{i\epsilon}G(i\epsilon))\right|\le F(\epsilon) ∣ Tr ( S N ( i ϵ ) ∂ i ϵ G ( i ϵ )) ∣ ≤ F ( ϵ ) for all N N N . Conversely, wherever I L ≠ 0 I_L\ne0 I L = 0 the skeleton series of Σ \Sigma Σ at the exact G G G either diverges, is not dominated, or converges to an unphysical self-energy [26 , 38 , 27 , 41 ] . Since the examples of Sec. 5 show I L ≠ 0 I_L\ne0 I L = 0 in regions adiabatically connected to the non-interacting ground state, the skeleton expansion can break down already there.
Funding information HE and AT acknowledge support from Austrian Science Fund (FWF), through projects with Grant DOI: Grant-DOI 10.55776/I5487 and Grant-DOI 10.55776/KIN2563725.
Ward-Takahashi identity
Since the identity (56) may not be familiar to everyone, we derive it in this Appendix. The density operator in a basis of Bloch waves e i k ⋅ r u n k ( r ) e^{i\mathbf{k}\cdot\mathbf{r}}u_{nk}(\mathbf{r}) e i k ⋅ r u nk ( r ) is
ρ ( q ) = ∑ n m k σ ρ 0 n m ( k , k + q ) c n k σ † c m k + q σ . \rho(\mathbf{q}) = \sum_{nmk\sigma} \rho_{0\,nm}(\mathbf{k},\mathbf{k}+\mathbf{q})c^\dagger_{nk\sigma}c_{mk+q\sigma}. ρ ( q ) = nmk σ ∑ ρ 0 nm ( k , k + q ) c nk σ † c mk + q σ .
where
ρ 0 n m ( k , k + q ) = ∫ d r u n k ( r ) ∗ u m k + q ( r ) = ρ 0 m n ( k + q , k ) ∗ . \rho_{0\,nm}(\mathbf{k},\mathbf{k}+\mathbf{q}) = \int d\mathbf{r}\,u_{nk}(\mathbf{r})^*u_{mk+q}(\mathbf{r}) = \rho_{0\,mn}(\mathbf{k}+\mathbf{q},\mathbf{k})^*. ρ 0 nm ( k , k + q ) = ∫ d r u nk ( r ) ∗ u mk + q ( r ) = ρ 0 mn ( k + q , k ) ∗ .
For q = 0 \mathbf{q}=0 q = 0 , ρ 0 n m ( k , k ) = δ n m \rho_{0\,nm}(\mathbf{k},\mathbf{k})=\delta_{nm} ρ 0 nm ( k , k ) = δ nm . Up to second order in q \mathbf{q} q , the matrix ρ 0 ( k , k + q ) = ρ 0 ( k + q , k ) † \rho_0(\mathbf{k},\mathbf{k}+\mathbf{q})=\rho_0(\mathbf{k}+\mathbf{q},\mathbf{k})^\dagger ρ 0 ( k , k + q ) = ρ 0 ( k + q , k ) † with elements ρ 0 n m ( k , k + q ) \rho_{0\,nm}(\mathbf{k},\mathbf{k}+\mathbf{q}) ρ 0 nm ( k , k + q ) reads
ρ 0 ( k , k + q ) = 1 − i q ⋅ A ( k ) + 1 2 ∑ μ ν q μ q ν K μ ν ( k ) , (91) \rho_0(\mathbf{k},\mathbf{k}+\mathbf{q}) = 1-i\mathbf{q}\cdot\mathbf{A}(\mathbf{k})+\frac{1}{2}\sum_{\mu\nu}q_\mu q_\nu K^{\mu\nu}(\mathbf{k}), \tag*{(91)} ρ 0 ( k , k + q ) = 1 − i q ⋅ A ( k ) + 2 1 μν ∑ q μ q ν K μν ( k ) , (91)
where A ( k ) \mathbf{A}(\mathbf{k}) A ( k ) is the Berry connection with elements A n m ( k ) = i ⟨ u n k ∣ ∇ k u m k ⟩ A_{nm}(\mathbf{k})=i\langle u_{nk}\mid\nabla_k u_{mk}\rangle A nm ( k ) = i ⟨ u nk ∣ ∇ k u mk ⟩ , while
K μ ν ( k ) = − g μ ν ( k ) − i 2 ( ∂ μ A ν ( k ) + ∂ ν A μ ( k ) ) (92) K^{\mu\nu}(\mathbf{k})=-g^{\mu\nu}(\mathbf{k})-\frac{i}{2}\left(\partial_\mu A^\nu(\mathbf{k})+\partial_\nu A^\mu(\mathbf{k})\right) \tag*{(92)} K μν ( k ) = − g μν ( k ) − 2 i ( ∂ μ A ν ( k ) + ∂ ν A μ ( k ) ) (92)
with g μ ν ( k ) g^{\mu\nu}(\mathbf{k}) g μν ( k ) the hermitian quantum metric with elements
g n m μ ν ( k ) = 1 2 ( ⟨ ∂ k μ u n k ∣ ∂ k ν u m k ⟩ + ⟨ ∂ k ν u n k ∣ ∂ k μ u m k ⟩ ) . g_{nm}^{\mu\nu}(\mathbf{k})=\frac{1}{2}\left(\langle\partial_{k_\mu}u_{nk}\mid\partial_{k_\nu}u_{mk}\rangle+\langle\partial_{k_\nu}u_{nk}\mid\partial_{k_\mu}u_{mk}\rangle\right). g nm μν ( k ) = 2 1 ( ⟨ ∂ k μ u nk ∣ ∂ k ν u mk ⟩ + ⟨ ∂ k ν u nk ∣ ∂ k μ u mk ⟩ ) .
For convenience, we rewrite ρ 0 ( k , k + q ) \rho_0(\mathbf{k},\mathbf{k}+\mathbf{q}) ρ 0 ( k , k + q ) in terms of two hermitian matrices as
ρ 0 ( k , k + q ) = ρ 0 ( k , q ) − i ρ 1 ( k , q ) , ρ 0 ( k + q , k ) = ρ 0 ( k , q ) + i ρ 1 ( k , q ) , (93) \rho_0(\mathbf{k},\mathbf{k}+\mathbf{q})=\rho_0(\mathbf{k},\mathbf{q})-i\rho_1(\mathbf{k},\mathbf{q}),\qquad\rho_0(\mathbf{k}+\mathbf{q},\mathbf{k})=\rho_0(\mathbf{k},\mathbf{q})+i\rho_1(\mathbf{k},\mathbf{q}), \tag*{(93)} ρ 0 ( k , k + q ) = ρ 0 ( k , q ) − i ρ 1 ( k , q ) , ρ 0 ( k + q , k ) = ρ 0 ( k , q ) + i ρ 1 ( k , q ) , (93)
where ρ 1 ( k , q ) \rho_1(\mathbf{k},\mathbf{q}) ρ 1 ( k , q ) contains the Berry connection and vanishes at q = 0 \mathbf{q}=0 q = 0 .
The Ward-Takahashi identity for the charge can be written in matrix notation, see Sec. 4.10 in [52 ] , either as
ρ 0 ( k , k + q ) ⋅ Σ ( i ϵ n + i ω m , k + q ) − Σ ( i ϵ n , k ) ⋅ ρ 0 ( k , k + q ) = 1 V ∑ k ′ T ∑ n ′ Γ 0 ( i ϵ n , k ; i ϵ n ′ + i ω m , k ′ + q ∣ i ϵ n ′ , k ′ ; i ϵ n + i ω m , k + q ) ⋅ ( ρ 0 ( k ′ , k ′ + q ) ⋅ G ( i ϵ n ′ + i ω m , k ′ + q ) − G ( i ϵ n ′ , k ′ ) ⋅ ρ 0 ( k ′ , k ′ + q ) ) , (94) \begin{aligned}
\rho_0(\mathbf{k},\mathbf{k}+\mathbf{q})\cdot\Sigma(i\epsilon_n+i\omega_m,\mathbf{k}+\mathbf{q})-\Sigma(i\epsilon_n,\mathbf{k})\cdot\rho_0(\mathbf{k},\mathbf{k}+\mathbf{q}) \\
=\frac{1}{V}\sum_{\mathbf{k}'}T\sum_{n'}\Gamma_0(i\epsilon_n,\mathbf{k};i\epsilon_{n'}+i\omega_m,\mathbf{k}'+\mathbf{q}\mid i\epsilon_{n'},\mathbf{k}';i\epsilon_n+i\omega_m,\mathbf{k}+\mathbf{q}) \\
\cdot\left(\rho_0(\mathbf{k}',\mathbf{k}'+\mathbf{q})\cdot G(i\epsilon_{n'}+i\omega_m,\mathbf{k}'+\mathbf{q})-G(i\epsilon_{n'},\mathbf{k}')\cdot\rho_0(\mathbf{k}',\mathbf{k}'+\mathbf{q})\right), \tag*{(94)}
\end{aligned} ρ 0 ( k , k + q ) ⋅ Σ ( i ϵ n + i ω m , k + q ) − Σ ( i ϵ n , k ) ⋅ ρ 0 ( k , k + q ) = V 1 k ′ ∑ T n ′ ∑ Γ 0 ( i ϵ n , k ; i ϵ n ′ + i ω m , k ′ + q ∣ i ϵ n ′ , k ′ ; i ϵ n + i ω m , k + q ) ⋅ ( ρ 0 ( k ′ , k ′ + q ) ⋅ G ( i ϵ n ′ + i ω m , k ′ + q ) − G ( i ϵ n ′ , k ′ ) ⋅ ρ 0 ( k ′ , k ′ + q ) ) , (94)
where ϵ n = ( 2 n + 1 ) π T \epsilon_n=(2n+1)\pi T ϵ n = ( 2 n + 1 ) π T and ϵ n ′ = ( 2 n ′ + 1 ) π T \epsilon_{n'}=(2n'+1)\pi T ϵ n ′ = ( 2 n ′ + 1 ) π T are fermionic Matsubara frequencies, while ω m = 2 m π T \omega_m=2m\pi T ω m = 2 mπ T a bosonic one, or as
Σ ( i ϵ n + i ω m , k + q ) ⋅ ρ 0 ( k + q , k ) − ρ 0 ( k + q , k ) ⋅ Σ ( i ϵ n , k ) = 1 V ∑ k ′ T ∑ n ′ Γ 0 ( i ϵ n + i ω m , k + q ; i ϵ n ′ , k ′ ∣ i ϵ n ′ + i ω m , k ′ + q ; i ϵ n , k ) ⋅ ( G ( i ϵ n ′ + i ω m , k ′ + q ) ⋅ ρ 0 ( k ′ + q , k ′ ) − ρ 0 ( k ′ + q , k ′ ) ⋅ G ( i ϵ n ′ , k ′ ) ) . (95) \begin{aligned}
\Sigma(i\epsilon_n+i\omega_m,\mathbf{k}+\mathbf{q})\cdot\rho_0(\mathbf{k}+\mathbf{q},\mathbf{k})-\rho_0(\mathbf{k}+\mathbf{q},\mathbf{k})\cdot\Sigma(i\epsilon_n,\mathbf{k}) \\
=\frac{1}{V}\sum_{\mathbf{k}'}T\sum_{n'}\Gamma_0(i\epsilon_n+i\omega_m,\mathbf{k}+\mathbf{q};i\epsilon_{n'},\mathbf{k}'\mid i\epsilon_{n'}+i\omega_m,\mathbf{k}'+\mathbf{q};i\epsilon_n,\mathbf{k}) \\
\cdot\left(G(i\epsilon_{n'}+i\omega_m,\mathbf{k}'+\mathbf{q})\cdot\rho_0(\mathbf{k}'+\mathbf{q},\mathbf{k}')-\rho_0(\mathbf{k}'+\mathbf{q},\mathbf{k}')\cdot G(i\epsilon_{n'},\mathbf{k}')\right). \tag*{(95)}
\end{aligned} Σ ( i ϵ n + i ω m , k + q ) ⋅ ρ 0 ( k + q , k ) − ρ 0 ( k + q , k ) ⋅ Σ ( i ϵ n , k ) = V 1 k ′ ∑ T n ′ ∑ Γ 0 ( i ϵ n + i ω m , k + q ; i ϵ n ′ , k ′ ∣ i ϵ n ′ + i ω m , k ′ + q ; i ϵ n , k ) ⋅ ( G ( i ϵ n ′ + i ω m , k ′ + q ) ⋅ ρ 0 ( k ′ + q , k ′ ) − ρ 0 ( k ′ + q , k ′ ) ⋅ G ( i ϵ n ′ , k ′ ) ) . (95)
where ρ 0 ( k + q , k ) = ρ 0 ( k , k + q ) † \rho_{0}(\mathbf{k}+\mathbf{q},\mathbf{k})=\rho_{0}(\mathbf{k},\mathbf{k}+\mathbf{q})^{\dagger} ρ 0 ( k + q , k ) = ρ 0 ( k , k + q ) † .
If we take half the sum of (94) and (95) and use the conventions in § 3 , we obtain
1 2 { Σ ( i ϵ n + i ω m , k + q ) , ρ 0 ( k , q ) } − 1 2 { Σ ( i ϵ n , k ) , ρ 0 ( k , q ) } + i 2 [ Σ ( i ϵ n + i ω m , k + q ) , ρ 1 ( k , q ) ] + i 2 [ Σ ( i ϵ n , k ) , ρ 1 ( k , q ) ] = 1 2 V ∑ k ′ T ∑ n ′ Γ 0 S ( i ϵ n , k ; i ϵ n ′ , k ′ ∣ i ω m , q ) ¨ × ( { G ( i ϵ n ′ + i ω m , k ′ + q ) , ρ 0 ( k ′ , q ) } − { G ( i ϵ n ′ , k ′ ) , ρ 0 ( k ′ , q ) } + i [ G ( i ϵ n ′ + i ω m , k ′ + q ) , ρ 1 ( k ′ , q ) ] + i [ G ( i ϵ n ′ , k ′ ) , ρ 1 ( k ′ , q ) ] ) + 1 2 V ∑ k ′ T ∑ n ′ Γ 0 A ( i ϵ n , k ; i ϵ n ′ , k ′ ∣ i ω m , q ) ¨ × ( [ G ( i ϵ n ′ + i ω m , k ′ + q ) , ρ 0 ( k ′ , q ) ] + [ G ( i ϵ n ′ , k ′ ) , ρ 0 ( k ′ , q ) ] + i { G ( i ϵ n ′ + i ω m , k ′ + q ) , ρ 1 ( k ′ , q ) } − i { G ( i ϵ n ′ , k ′ ) , ρ 1 ( k ′ , q ) } ) . (96) \begin{aligned}
&\frac{1}{2}\left\{\Sigma(i\epsilon_n+i\omega_m,\mathbf{k}+\mathbf{q}),\rho_0(\mathbf{k},\mathbf{q})\right\}
-\frac{1}{2}\left\{\Sigma(i\epsilon_n,\mathbf{k}),\rho_0(\mathbf{k},\mathbf{q})\right\}\\
&\quad+\frac{i}{2}\left[\Sigma(i\epsilon_n+i\omega_m,\mathbf{k}+\mathbf{q}),\rho_1(\mathbf{k},\mathbf{q})\right]
+\frac{i}{2}\left[\Sigma(i\epsilon_n,\mathbf{k}),\rho_1(\mathbf{k},\mathbf{q})\right]\\
&=\frac{1}{2V}\sum_{\mathbf{k}'}T\sum_{n'}\Gamma_0^S(i\epsilon_n,\mathbf{k};i\epsilon_{n'},\mathbf{k}'|i\omega_m,\mathbf{q})\ddot{\ }\\
&\quad\times\left(\left\{G(i\epsilon_{n'}+i\omega_m,\mathbf{k}'+\mathbf{q}),\rho_0(\mathbf{k}',\mathbf{q})\right\}
-\left\{G(i\epsilon_{n'},\mathbf{k}'),\rho_0(\mathbf{k}',\mathbf{q})\right\}\right.\\
&\qquad\left.+i\left[G(i\epsilon_{n'}+i\omega_m,\mathbf{k}'+\mathbf{q}),\rho_1(\mathbf{k}',\mathbf{q})\right]
+i\left[G(i\epsilon_{n'},\mathbf{k}'),\rho_1(\mathbf{k}',\mathbf{q})\right]\right)\\
&\quad+\frac{1}{2V}\sum_{\mathbf{k}'}T\sum_{n'}\Gamma_0^A(i\epsilon_n,\mathbf{k};i\epsilon_{n'},\mathbf{k}'|i\omega_m,\mathbf{q})\ddot{\ }\\
&\quad\times\left(\left[G(i\epsilon_{n'}+i\omega_m,\mathbf{k}'+\mathbf{q}),\rho_0(\mathbf{k}',\mathbf{q})\right]
+\left[G(i\epsilon_{n'},\mathbf{k}'),\rho_0(\mathbf{k}',\mathbf{q})\right]\right.\\
&\qquad\left.+i\left\{G(i\epsilon_{n'}+i\omega_m,\mathbf{k}'+\mathbf{q}),\rho_1(\mathbf{k}',\mathbf{q})\right\}
-i\left\{G(i\epsilon_{n'},\mathbf{k}'),\rho_1(\mathbf{k}',\mathbf{q})\right\}\right). \tag*{(96)}
\end{aligned} 2 1 { Σ ( i ϵ n + i ω m , k + q ) , ρ 0 ( k , q ) } − 2 1 { Σ ( i ϵ n , k ) , ρ 0 ( k , q ) } + 2 i [ Σ ( i ϵ n + i ω m , k + q ) , ρ 1 ( k , q ) ] + 2 i [ Σ ( i ϵ n , k ) , ρ 1 ( k , q ) ] = 2 V 1 k ′ ∑ T n ′ ∑ Γ 0 S ( i ϵ n , k ; i ϵ n ′ , k ′ ∣ i ω m , q ) ¨ × ( { G ( i ϵ n ′ + i ω m , k ′ + q ) , ρ 0 ( k ′ , q ) } − { G ( i ϵ n ′ , k ′ ) , ρ 0 ( k ′ , q ) } + i [ G ( i ϵ n ′ + i ω m , k ′ + q ) , ρ 1 ( k ′ , q ) ] + i [ G ( i ϵ n ′ , k ′ ) , ρ 1 ( k ′ , q ) ] ) + 2 V 1 k ′ ∑ T n ′ ∑ Γ 0 A ( i ϵ n , k ; i ϵ n ′ , k ′ ∣ i ω m , q ) ¨ × ( [ G ( i ϵ n ′ + i ω m , k ′ + q ) , ρ 0 ( k ′ , q ) ] + [ G ( i ϵ n ′ , k ′ ) , ρ 0 ( k ′ , q ) ] + i { G ( i ϵ n ′ + i ω m , k ′ + q ) , ρ 1 ( k ′ , q ) } − i { G ( i ϵ n ′ , k ′ ) , ρ 1 ( k ′ , q ) } ) . (96)
Similarly, the half difference of (95) and (94) reads
1 2 [ Σ ( i ϵ n + i ω m , k + q ) , ρ 0 ( k , q ) ] + 1 2 [ Σ ( i ϵ n , k ) , ρ 0 ( k , q ) ] + i 2 { Σ ( i ϵ n + i ω m , k + q ) , ρ 1 ( k , q ) } − i 2 { Σ ( i ϵ n , k ) , ρ 1 ( k , q ) } = 1 2 V ∑ k ′ T ∑ n ′ Γ 0 A ( i ϵ n , k ; i ϵ n ′ , k ′ ∣ i ω m , q ) ¨ × ( { G ( i ϵ n ′ + i ω m , k ′ + q ) , ρ 0 ( k ′ , q ) } − { G ( i ϵ n ′ , k ′ ) , ρ 0 ( k ′ , q ) } + i [ G ( i ϵ n ′ + i ω m , k ′ + q ) , ρ 1 ( k ′ , q ) ] + i [ G ( i ϵ n ′ , k ′ ) , ρ 1 ( k ′ , q ) ] ) + 1 2 V ∑ k ′ T ∑ n ′ Γ 0 S ( i ϵ n , k ; i ϵ n ′ , k ′ ∣ i ω m , q ) ¨ × ( [ G ( i ϵ n ′ + i ω m , k ′ + q ) , ρ 0 ( k ′ , q ) ] + [ G ( i ϵ n ′ , k ′ ) , ρ 0 ( k ′ , q ) ] + i { G ( i ϵ n ′ + i ω m , k ′ + q ) , ρ 1 ( k ′ , q ) } − i { G ( i ϵ n ′ , k ′ ) , ρ 1 ( k ′ , q ) } ) . (97) \begin{aligned}
&\frac{1}{2}\left[\Sigma(i\epsilon_n+i\omega_m,\mathbf{k}+\mathbf{q}),\rho_0(\mathbf{k},\mathbf{q})\right]
+\frac{1}{2}\left[\Sigma(i\epsilon_n,\mathbf{k}),\rho_0(\mathbf{k},\mathbf{q})\right]\\
&\quad+\frac{i}{2}\left\{\Sigma(i\epsilon_n+i\omega_m,\mathbf{k}+\mathbf{q}),\rho_1(\mathbf{k},\mathbf{q})\right\}
-\frac{i}{2}\left\{\Sigma(i\epsilon_n,\mathbf{k}),\rho_1(\mathbf{k},\mathbf{q})\right\}\\
&=\frac{1}{2V}\sum_{\mathbf{k}'}T\sum_{n'}\Gamma_0^A(i\epsilon_n,\mathbf{k};i\epsilon_{n'},\mathbf{k}'|i\omega_m,\mathbf{q})\ddot{\ }\\
&\quad\times\left(\left\{G(i\epsilon_{n'}+i\omega_m,\mathbf{k}'+\mathbf{q}),\rho_0(\mathbf{k}',\mathbf{q})\right\}
-\left\{G(i\epsilon_{n'},\mathbf{k}'),\rho_0(\mathbf{k}',\mathbf{q})\right\}\right.\\
&\qquad\left.+i\left[G(i\epsilon_{n'}+i\omega_m,\mathbf{k}'+\mathbf{q}),\rho_1(\mathbf{k}',\mathbf{q})\right]
+i\left[G(i\epsilon_{n'},\mathbf{k}'),\rho_1(\mathbf{k}',\mathbf{q})\right]\right)\\
&\quad+\frac{1}{2V}\sum_{\mathbf{k}'}T\sum_{n'}\Gamma_0^S(i\epsilon_n,\mathbf{k};i\epsilon_{n'},\mathbf{k}'|i\omega_m,\mathbf{q})\ddot{\ }\\
&\quad\times\left(\left[G(i\epsilon_{n'}+i\omega_m,\mathbf{k}'+\mathbf{q}),\rho_0(\mathbf{k}',\mathbf{q})\right]
+\left[G(i\epsilon_{n'},\mathbf{k}'),\rho_0(\mathbf{k}',\mathbf{q})\right]\right.\\
&\qquad\left.+i\left\{G(i\epsilon_{n'}+i\omega_m,\mathbf{k}'+\mathbf{q}),\rho_1(\mathbf{k}',\mathbf{q})\right\}
-i\left\{G(i\epsilon_{n'},\mathbf{k}'),\rho_1(\mathbf{k}',\mathbf{q})\right\}\right). \tag*{(97)}
\end{aligned} 2 1 [ Σ ( i ϵ n + i ω m , k + q ) , ρ 0 ( k , q ) ] + 2 1 [ Σ ( i ϵ n , k ) , ρ 0 ( k , q ) ] + 2 i { Σ ( i ϵ n + i ω m , k + q ) , ρ 1 ( k , q ) } − 2 i { Σ ( i ϵ n , k ) , ρ 1 ( k , q ) } = 2 V 1 k ′ ∑ T n ′ ∑ Γ 0 A ( i ϵ n , k ; i ϵ n ′ , k ′ ∣ i ω m , q ) ¨ × ( { G ( i ϵ n ′ + i ω m , k ′ + q ) , ρ 0 ( k ′ , q ) } − { G ( i ϵ n ′ , k ′ ) , ρ 0 ( k ′ , q ) } + i [ G ( i ϵ n ′ + i ω m , k ′ + q ) , ρ 1 ( k ′ , q ) ] + i [ G ( i ϵ n ′ , k ′ ) , ρ 1 ( k ′ , q ) ] ) + 2 V 1 k ′ ∑ T n ′ ∑ Γ 0 S ( i ϵ n , k ; i ϵ n ′ , k ′ ∣ i ω m , q ) ¨ × ( [ G ( i ϵ n ′ + i ω m , k ′ + q ) , ρ 0 ( k ′ , q ) ] + [ G ( i ϵ n ′ , k ′ ) , ρ 0 ( k ′ , q ) ] + i { G ( i ϵ n ′ + i ω m , k ′ + q ) , ρ 1 ( k ′ , q ) } − i { G ( i ϵ n ′ , k ′ ) , ρ 1 ( k ′ , q ) } ) . (97)
One can now take the limit T → 0 T\to0 T → 0 , so that ϵ n → ϵ \epsilon_n\to\epsilon ϵ n → ϵ , ϵ n ′ → ϵ ′ \epsilon_{n'}\to\epsilon' ϵ n ′ → ϵ ′ as well as ω m → ω \omega_m\to\omega ω m → ω all become continuous variables, and the sum over ϵ n ′ \epsilon_{n'} ϵ n ′ an integral. Subsequently, we differentiate both sides of the equations (96) and (97) with respect to q \mathbf{q} q and ω \omega ω , evaluated at ω = q = 0 \omega=\mathbf{q}=0 ω = q = 0 . This process results in a series of identities that relate the derivatives of the self-energy, Green’s function, and irreducible vertices.
The results in Sec. 3 were obtained using a topologically trivial basis where the Berry connection is zero throughout the Brillouin zone, including its boundaries. Therefore, we derive some of those identities in the case where A ( k ) = 0 \mathbf{A}(\mathbf{k})=0 A ( k ) = 0 . The simplest ones are obtained by taking the first derivatives of (96) with respect to ω \omega ω or q \mathbf{q} q , which leads to
∂ i ϵ Σ ( i ϵ , k ) = 1 V ∑ k ′ ∫ d ϵ ′ 2 π Γ 0 S ( i ϵ , k ; i ϵ ′ , k ′ ) : ∂ i ϵ ′ G ( i ϵ ′ , k ′ ) , ∂ μ Σ ( i ϵ , k ) = 1 V ∑ k ′ ∫ d ϵ ′ 2 π Γ 0 S ( i ϵ , k ; i ϵ ′ , k ′ ) : ∂ μ G ( i ϵ ′ , k ′ ) , (98) \begin{aligned}
\partial_{i\epsilon}\Sigma(i\epsilon,\mathbf{k}) &= \frac{1}{V}\sum_{\mathbf{k}'}\int\frac{d\epsilon'}{2\pi}\,\Gamma^{S}_{0}(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}')\mathbin{\vcenter{\text{\(:\)}}}\partial_{i\epsilon'}G(i\epsilon',\mathbf{k}'),\\
\partial_{\mu}\Sigma(i\epsilon,\mathbf{k}) &= \frac{1}{V}\sum_{\mathbf{k}'}\int\frac{d\epsilon'}{2\pi}\,\Gamma^{S}_{0}(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}')\mathbin{\vcenter{\text{\(:\)}}}\partial_{\mu}G(i\epsilon',\mathbf{k}'), \tag*{(98)}
\end{aligned} ∂ i ϵ Σ ( i ϵ , k ) ∂ μ Σ ( i ϵ , k ) = V 1 k ′ ∑ ∫ 2 π d ϵ ′ Γ 0 S ( i ϵ , k ; i ϵ ′ , k ′ ) : ∂ i ϵ ′ G ( i ϵ ′ , k ′ ) , = V 1 k ′ ∑ ∫ 2 π d ϵ ′ Γ 0 S ( i ϵ , k ; i ϵ ′ , k ′ ) : ∂ μ G ( i ϵ ′ , k ′ ) , (98)
where ∂ μ Σ ( i ϵ , k ) = ∂ k μ Σ ( i ϵ , k ) \partial_{\mu}\Sigma(i\epsilon,\mathbf{k})=\partial_{k_{\mu}}\Sigma(i\epsilon,\mathbf{k}) ∂ μ Σ ( i ϵ , k ) = ∂ k μ Σ ( i ϵ , k ) and ∂ μ G ( i ϵ ′ , k ′ ) = ∂ k μ ′ G ( i ϵ ′ , k ′ ) \partial_{\mu}G(i\epsilon',\mathbf{k}')=\partial_{k'_{\mu}}G(i\epsilon',\mathbf{k}') ∂ μ G ( i ϵ ′ , k ′ ) = ∂ k μ ′ G ( i ϵ ′ , k ′ ) . Equation (56) is instead obtained by the mixed partial derivative of (97) with respect to ω \omega ω and q μ q_{\mu} q μ , which yields
1 V ∑ k ′ ∫ d ϵ ′ 2 π Γ 0 μ A ( i ϵ , k ; i ϵ ′ , k ′ ) : ∂ μ G ( i ϵ ′ , k ′ ) = − 1 V ∑ k ′ ∫ d ϵ ′ 2 π Γ 0 μ A ( i ϵ , k ; i ϵ ′ , k ′ ) : ∂ i ϵ ′ G ( i ϵ ′ , k ′ ) , (99) \begin{aligned}
\frac{1}{V}\sum_{\mathbf{k}'}\int\frac{d\epsilon'}{2\pi}\,\Gamma^{A}_{0\mu}(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}')\mathbin{\vcenter{\text{\(:\)}}}\partial_{\mu}G(i\epsilon',\mathbf{k}')
&=-\frac{1}{V}\sum_{\mathbf{k}'}\int\frac{d\epsilon'}{2\pi}\,\Gamma^{A}_{0\mu}(i\epsilon,\mathbf{k};i\epsilon',\mathbf{k}')\mathbin{\vcenter{\text{\(:\)}}}\partial_{i\epsilon'}G(i\epsilon',\mathbf{k}'), \tag*{(99)}
\end{aligned} V 1 k ′ ∑ ∫ 2 π d ϵ ′ Γ 0 μ A ( i ϵ , k ; i ϵ ′ , k ′ ) : ∂ μ G ( i ϵ ′ , k ′ ) = − V 1 k ′ ∑ ∫ 2 π d ϵ ′ Γ 0 μ A ( i ϵ , k ; i ϵ ′ , k ′ ) : ∂ i ϵ ′ G ( i ϵ ′ , k ′ ) , (99)
from which (56) directly follows.
Invariance property of ν H L I \nu^{I}_{HL} ν H L I
In this appendix we show that also
ν H L I = − i π 3 V 2 ∑ k 1 k 2 ∫ d ϵ 1 2 π d ϵ 2 2 π ϵ μ ν ρ Tr ( ∂ μ G ( i ϵ 1 , k 1 ) ⋅ γ 0 ν A ( i ϵ 1 , k 1 ; i ϵ 2 , k 2 ) ⋅ ∂ ρ G ( i ϵ 2 , k 2 ) ) , (100) \nu^{I}_{HL}=-i\frac{\pi}{3V^{2}}\sum_{\mathbf{k}_{1}\mathbf{k}_{2}}\int\frac{d\epsilon_{1}}{2\pi}\frac{d\epsilon_{2}}{2\pi}\,\epsilon_{\mu\nu\rho}\operatorname{Tr}\left(\partial_{\mu}G(i\epsilon_{1},\mathbf{k}_{1})\mathbin{\cdot}\gamma^{A}_{0\nu}(i\epsilon_{1},\mathbf{k}_{1};i\epsilon_{2},\mathbf{k}_{2})\mathbin{\cdot}\partial_{\rho}G(i\epsilon_{2},\mathbf{k}_{2})\right), \tag*{(100)} ν H L I = − i 3 V 2 π k 1 k 2 ∑ ∫ 2 π d ϵ 1 2 π d ϵ 2 ϵ μν ρ Tr ( ∂ μ G ( i ϵ 1 , k 1 ) ⋅ γ 0 ν A ( i ϵ 1 , k 1 ; i ϵ 2 , k 2 ) ⋅ ∂ ρ G ( i ϵ 2 , k 2 ) ) , (100)
is invariant under any non-singular variation of the Green’s function. To simplify the notation, we define x i : = ( i ϵ i , k i ) \mathbf{x}_{i}:=(i\epsilon_{i},\mathbf{k}_{i}) x i := ( i ϵ i , k i ) ,
∫ d x i ⋯ : = 1 V ∑ k i ∫ d ϵ i 2 π ⋯ , \int d\mathbf{x}_{i}\,\cdots:=\frac{1}{V}\sum_{\mathbf{k}_{i}}\int\frac{d\epsilon_{i}}{2\pi}\,\cdots, ∫ d x i ⋯ := V 1 k i ∑ ∫ 2 π d ϵ i ⋯ ,
and further assume that the trace contracts the orbital indices attached to the same x i \mathbf{x}_{i} x i , so that
ν H L I = − i π 3 ∫ d x 1 d x 2 ϵ μ ν ρ Tr ( ∂ μ G ( x 1 ) ⋅ γ 0 ν A ( x 1 ; x 2 ) ⋅ ∂ ρ G ( x 2 ) ) . (101) \nu^{I}_{HL}=-i\frac{\pi}{3}\int d\mathbf{x}_{1}\,d\mathbf{x}_{2}\,\epsilon_{\mu\nu\rho}\operatorname{Tr}\left(\partial_{\mu}G(\mathbf{x}_{1})\mathbin{\cdot}\gamma^{A}_{0\nu}(\mathbf{x}_{1};\mathbf{x}_{2})\mathbin{\cdot}\partial_{\rho}G(\mathbf{x}_{2})\right). \tag*{(101)} ν H L I = − i 3 π ∫ d x 1 d x 2 ϵ μν ρ Tr ( ∂ μ G ( x 1 ) ⋅ γ 0 ν A ( x 1 ; x 2 ) ⋅ ∂ ρ G ( x 2 ) ) . (101)
We now take a variation and, exploiting the properties of the fully antisymmetric tensor and assuming that integration by parts does not yield any boundary term, we find
δ ν H L I = i 2 π 3 ∫ d x 1 d x 2 ϵ μ ν ρ Tr ( δ G ( x 1 ) ∂ x 1 μ γ 0 ν A ( x 1 ; x 2 ) ∂ x 2 ρ G ( x 2 ) ) − i π 3 ∫ d x 1 d x 2 ϵ μ ν ρ Tr ( ∂ x 1 μ G ( x 1 ) δ γ 0 ν A ( x 1 ; x 2 ) ∂ x 2 ρ G ( x 2 ) ) = i 2 π 3 ∫ d x 1 d x 2 ϵ μ ν ρ Tr ( δ G ( x 1 ) ∂ μ γ 0 ν A ( x 1 ; x 2 ) ∂ ρ G ( x 2 ) ) − i π 3 ∫ d x 1 d x 2 ϵ μ ν ρ Tr ( ∂ μ G ( x 1 ) δ γ 0 ν A ( x 1 ; x 2 ) ∂ ρ G ( x 2 ) ) , \begin{aligned}
\delta\nu^{I}_{HL}
={}&i\frac{2\pi}{3}\int d\mathbf{x}_{1}\,d\mathbf{x}_{2}\,\epsilon_{\mu\nu\rho}\operatorname{Tr}\left(\delta G(\mathbf{x}_{1})\partial_{x_{1\mu}}\gamma^{A}_{0\nu}(\mathbf{x}_{1};\mathbf{x}_{2})\partial_{x_{2\rho}}G(\mathbf{x}_{2})\right)\\
&-i\frac{\pi}{3}\int d\mathbf{x}_{1}\,d\mathbf{x}_{2}\,\epsilon_{\mu\nu\rho}\operatorname{Tr}\left(\partial_{x_{1\mu}}G(\mathbf{x}_{1})\delta\gamma^{A}_{0\nu}(\mathbf{x}_{1};\mathbf{x}_{2})\partial_{x_{2\rho}}G(\mathbf{x}_{2})\right)\\
={}&i\frac{2\pi}{3}\int d\mathbf{x}_{1}\,d\mathbf{x}_{2}\,\epsilon_{\mu\nu\rho}\operatorname{Tr}\left(\delta G(\mathbf{x}_{1})\partial_{\mu}\gamma^{A}_{0\nu}(\mathbf{x}_{1};\mathbf{x}_{2})\partial_{\rho}G(\mathbf{x}_{2})\right)\\
&-i\frac{\pi}{3}\int d\mathbf{x}_{1}\,d\mathbf{x}_{2}\,\epsilon_{\mu\nu\rho}\operatorname{Tr}\left(\partial_{\mu}G(\mathbf{x}_{1})\delta\gamma^{A}_{0\nu}(\mathbf{x}_{1};\mathbf{x}_{2})\partial_{\rho}G(\mathbf{x}_{2})\right),
\end{aligned} δ ν H L I = = i 3 2 π ∫ d x 1 d x 2 ϵ μν ρ Tr ( δ G ( x 1 ) ∂ x 1 μ γ 0 ν A ( x 1 ; x 2 ) ∂ x 2 ρ G ( x 2 ) ) − i 3 π ∫ d x 1 d x 2 ϵ μν ρ Tr ( ∂ x 1 μ G ( x 1 ) δ γ 0 ν A ( x 1 ; x 2 ) ∂ x 2 ρ G ( x 2 ) ) i 3 2 π ∫ d x 1 d x 2 ϵ μν ρ Tr ( δ G ( x 1 ) ∂ μ γ 0 ν A ( x 1 ; x 2 ) ∂ ρ G ( x 2 ) ) − i 3 π ∫ d x 1 d x 2 ϵ μν ρ Tr ( ∂ μ G ( x 1 ) δ γ 0 ν A ( x 1 ; x 2 ) ∂ ρ G ( x 2 ) ) ,
having defined
∂ μ γ 0 ν A ( x 1 ; x 2 ) : = ∑ i = 1 2 ∂ x i μ γ 0 ν A ( x 1 ; x 2 ) . \partial_{\mu}\gamma^{A}_{0\nu}(x_{1};x_{2}) := \sum_{i=1}^{2}\partial_{x_{i\mu}}\gamma^{A}_{0\nu}(x_{1};x_{2}) . ∂ μ γ 0 ν A ( x 1 ; x 2 ) := i = 1 ∑ 2 ∂ x i μ γ 0 ν A ( x 1 ; x 2 ) .
We can write
∂ μ γ 0 ν A ( x 1 ; x 2 ) = ∫ d x 3 S ν ( x 1 , x 2 ; x 3 ) ∂ μ G ( x 3 ) , \partial_{\mu}\gamma^{A}_{0\nu}(x_{1};x_{2}) = \int dx_{3}\,S_{\nu}(x_{1},x_{2};x_{3})\,\partial_{\mu}G(x_{3}), ∂ μ γ 0 ν A ( x 1 ; x 2 ) = ∫ d x 3 S ν ( x 1 , x 2 ; x 3 ) ∂ μ G ( x 3 ) ,
δ γ 0 ν A ( x 1 ; x 2 ) = ∫ d x 3 S ν ( x 1 , x 2 ; x 3 ) δ G ( x 3 ) , (102) \delta\gamma^{A}_{0\nu}(x_{1};x_{2}) = \int dx_{3}\,S_{\nu}(x_{1},x_{2};x_{3})\,\delta G(x_{3}), \tag*{(102)} δ γ 0 ν A ( x 1 ; x 2 ) = ∫ d x 3 S ν ( x 1 , x 2 ; x 3 ) δ G ( x 3 ) , (102)
where S ν ( x 1 , x 2 ; x 3 ) = − S ν ( x 2 , x 1 ; x 3 ) : = δ γ 0 ν A ( x 1 ; x 2 ) / δ G ( x 3 ) S_{\nu}(x_{1},x_{2};x_{3})=-S_{\nu}(x_{2},x_{1};x_{3}):=\delta\gamma^{A}_{0\nu}(x_{1};x_{2})/\delta G(x_{3}) S ν ( x 1 , x 2 ; x 3 ) = − S ν ( x 2 , x 1 ; x 3 ) := δ γ 0 ν A ( x 1 ; x 2 ) / δ G ( x 3 ) is a six-leg vertex. Since internal lines do not enjoy the same crossing relation γ 0 ν A ( x 1 ; x 2 ) = − γ 0 ν A ( x 2 ; x 1 ) \gamma^{A}_{0\nu}(x_{1};x_{2})=-\gamma^{A}_{0\nu}(x_{2};x_{1}) γ 0 ν A ( x 1 ; x 2 ) = − γ 0 ν A ( x 2 ; x 1 ) of the external legs, the six-leg vertices S ν ( x 1 , x 2 ; x 3 ) S_{\nu}(x_{1},x_{2};x_{3}) S ν ( x 1 , x 2 ; x 3 ) , S ν ( x 3 , x 2 ; x 1 ) S_{\nu}(x_{3},x_{2};x_{1}) S ν ( x 3 , x 2 ; x 1 ) and S ν ( x 1 , x 3 ; x 2 ) S_{\nu}(x_{1},x_{3};x_{2}) S ν ( x 1 , x 3 ; x 2 ) cannot be mapped onto each other by any simple sign change. However, one can show that S ν ( x 1 , x 2 ; x 3 ) S_{\nu}(x_{1},x_{2};x_{3}) S ν ( x 1 , x 2 ; x 3 ) satisfies the cyclic condition:
S ν ( x 1 , x 2 ; x 3 ) + S ν ( x 2 , x 3 ; x 1 ) + S ν ( x 3 , x 1 ; x 2 ) = 0. (103) S_{\nu}(x_{1},x_{2};x_{3})+S_{\nu}(x_{2},x_{3};x_{1})+S_{\nu}(x_{3},x_{1};x_{2})=0 . \tag*{(103)} S ν ( x 1 , x 2 ; x 3 ) + S ν ( x 2 , x 3 ; x 1 ) + S ν ( x 3 , x 1 ; x 2 ) = 0. (103)
Indeed, since the irreducible vertex with transferred frequency and momentum y = ( i ω , q ) y=(i\omega,\mathbf{q}) y = ( iω , q ) is defined as
Γ 0 ( x 1 + y , x 2 ∣ x 2 + y , x 1 ) = δ 2 Φ [ G ] δ G ( x 1 ; x 1 + y ) δ G ( x 2 + y ; x 2 ) , (104) \Gamma_{0}(x_{1}+y,x_{2}\vert x_{2}+y,x_{1})=
\frac{\delta^{2}\Phi[G]}{\delta G(x_{1};x_{1}+y)\,\delta G(x_{2}+y;x_{2})}, \tag*{(104)} Γ 0 ( x 1 + y , x 2 ∣ x 2 + y , x 1 ) = δ G ( x 1 ; x 1 + y ) δ G ( x 2 + y ; x 2 ) δ 2 Φ [ G ] , (104)
then
S ν ( x 1 , x 2 ; x 3 ) = 1 2 ∂ y ν ( δ 3 Φ [ G ] δ G ( x 1 ; x 1 + y ) δ G ( x 2 + y ; x 2 ) δ G ( x 3 ; x 3 ) − δ 3 Φ [ G ] δ G ( x 1 + y ; x 1 ) δ G ( x 2 ; x 2 + y ) δ G ( x 3 ; x 3 ) ) ∣ y = 0 . (105) S_{\nu}(x_{1},x_{2};x_{3})=\frac{1}{2}\partial_{y_{\nu}}\left(
\frac{\delta^{3}\Phi[G]}{\delta G(x_{1};x_{1}+y)\,\delta G(x_{2}+y;x_{2})\,\delta G(x_{3};x_{3})}
-
\frac{\delta^{3}\Phi[G]}{\delta G(x_{1}+y;x_{1})\,\delta G(x_{2};x_{2}+y)\,\delta G(x_{3};x_{3})}
\right)\bigg|_{y=0}. \tag*{(105)} S ν ( x 1 , x 2 ; x 3 ) = 2 1 ∂ y ν ( δ G ( x 1 ; x 1 + y ) δ G ( x 2 + y ; x 2 ) δ G ( x 3 ; x 3 ) δ 3 Φ [ G ] − δ G ( x 1 + y ; x 1 ) δ G ( x 2 ; x 2 + y ) δ G ( x 3 ; x 3 ) δ 3 Φ [ G ] ) y = 0 . (105)
We label each pair ( a i ; b i ) (a_i;b_i) ( a i ; b i ) in δ G ( a i ; b i ) \delta G(a_i;b_i) δ G ( a i ; b i ) , i = 1 , 2 , 3 i=1,2,3 i = 1 , 2 , 3 , by its mean ( a i + b i ) / 2 (a_i+b_i)/2 ( a i + b i ) /2 and its transfer a i − b i = y s i a_i-b_i=y s_i a i − b i = y s i , s i = 0 , ± 1 s_i=0,\pm1 s i = 0 , ± 1 . The y y y -derivative in (105) is then the sum of the derivatives with respect to the y y y -dependent means and to the y y y -dependent transfers. The two terms in (105) have the same means, ( x 1 + y / 2 , x 2 + y / 2 , x 3 ) (x_1+y/2,x_2+y/2,x_3) ( x 1 + y /2 , x 2 + y /2 , x 3 ) , so the contributions of the shifts of the means cancel; on the contrary, they have opposite transfers, ( s 1 , s 2 , s 3 ) = ± ( − 1 , 1 , 0 ) (s_1,s_2,s_3)=\pm(-1,1,0) ( s 1 , s 2 , s 3 ) = ± ( − 1 , 1 , 0 ) , so those contributions add. Hence S ν ( x 1 , x 2 ; x 3 ) S_{\nu}(x_1,x_2;x_3) S ν ( x 1 , x 2 ; x 3 ) is the derivative of Φ ( 3 ) \Phi^{(3)} Φ ( 3 ) at zero transfer along the direction ( s 1 , s 2 , s 3 ) = ( − 1 , 1 , 0 ) (s_1,s_2,s_3)=(-1,1,0) ( s 1 , s 2 , s 3 ) = ( − 1 , 1 , 0 ) in transfer space, the components referring to the pairs attached to x 1 x_1 x 1 , x 2 x_2 x 2 and x 3 x_3 x 3 , in this order. Since Φ ( 3 ) \Phi^{(3)} Φ ( 3 ) , being a third functional derivative, is symmetric under permutations of its three pairs, the same ordering can be used for S ν ( x 2 , x 3 ; x 1 ) S_{\nu}(x_2,x_3;x_1) S ν ( x 2 , x 3 ; x 1 ) and S ν ( x 3 , x 1 ; x 2 ) S_{\nu}(x_3,x_1;x_2) S ν ( x 3 , x 1 ; x 2 ) , which are then the derivatives of the same function along ( 0 , − 1 , 1 ) (0,-1,1) ( 0 , − 1 , 1 ) and ( 1 , 0 , − 1 ) (1,0,-1) ( 1 , 0 , − 1 ) . The three directions add up to zero, i.e. there is no shift at linear order in y y y , hence the identity (103) .
Through (102) we can compactly rewrite δ ν H L I \delta\nu^{I}_{HL} δ ν H L I as
δ ν H L I = i π 3 ∫ ∏ i = 1 3 d x i ∑ μ Tr ( δ G ( x 1 ) I μ ( x 2 , x 3 ) Q μ ( x 1 , x 2 , x 3 ) ) , (106) \delta\nu^{I}_{HL}=i\frac{\pi}{3}\int\prod_{i=1}^{3}dx_i\,\sum_{\mu}\operatorname{Tr}\left(\delta G(x_1)\,I_{\mu}(x_2,x_3)\,Q_{\mu}(x_1,x_2,x_3)\right), \tag*{(106)} δ ν H L I = i 3 π ∫ i = 1 ∏ 3 d x i μ ∑ Tr ( δ G ( x 1 ) I μ ( x 2 , x 3 ) Q μ ( x 1 , x 2 , x 3 ) ) , (106)
having defined
I μ ( x 2 , x 3 ) : = ∑ ν ρ ϵ μ ν ρ ∂ ν G ( x 2 ) ∂ ρ G ( x 3 ) , (107) I_{\mu}(x_{2},x_{3}) := \sum_{\nu\rho} \epsilon_{\mu\nu\rho}\,\partial_{\nu}G(x_{2})\,\partial_{\rho}G(x_{3}), \tag*{(107)} I μ ( x 2 , x 3 ) := ν ρ ∑ ϵ μν ρ ∂ ν G ( x 2 ) ∂ ρ G ( x 3 ) , (107)
Q μ ( x 1 , x 2 , x 3 ) : = 2 S μ ( x 1 , x 2 ; x 3 ) − S μ ( x 3 , x 2 ; x 1 ) = 2 S μ ( x 1 , x 2 ; x 3 ) + S μ ( x 2 , x 3 ; x 1 ) . Q_{\mu}(x_{1},x_{2},x_{3}) := 2S_{\mu}(x_{1},x_{2};x_{3}) - S_{\mu}(x_{3},x_{2};x_{1}) = 2S_{\mu}(x_{1},x_{2};x_{3}) + S_{\mu}(x_{2},x_{3};x_{1}). Q μ ( x 1 , x 2 , x 3 ) := 2 S μ ( x 1 , x 2 ; x 3 ) − S μ ( x 3 , x 2 ; x 1 ) = 2 S μ ( x 1 , x 2 ; x 3 ) + S μ ( x 2 , x 3 ; x 1 ) .
Since I μ ( x 2 , x 3 ) = − I μ ( x 3 , x 2 ) I_{\mu}(x_{2},x_{3}) = -I_{\mu}(x_{3},x_{2}) I μ ( x 2 , x 3 ) = − I μ ( x 3 , x 2 ) , we can replace Q μ ( x 1 , x 2 , x 3 ) Q_{\mu}(x_{1},x_{2},x_{3}) Q μ ( x 1 , x 2 , x 3 ) with
Q μ ( x 1 , x 2 , x 3 ) → 1 2 ( Q μ ( x 1 , x 2 , x 3 ) − Q μ ( x 1 , x 3 , x 2 ) ) = 1 2 ( 2 S μ ( x 1 , x 2 ; x 3 ) + S μ ( x 2 , x 3 ; x 1 ) − 2 S μ ( x 1 , x 3 ; x 2 ) − S μ ( x 3 , x 2 ; x 1 ) ) = S μ ( x 1 , x 2 ; x 3 ) + S μ ( x 2 , x 3 ; x 1 ) + S μ ( x 3 , x 1 ; x 2 ) ≡ 0 , (108) \begin{aligned}
Q_{\mu}(x_{1},x_{2},x_{3}) \to\frac{1}{2}\left(Q_{\mu}(x_{1},x_{2},x_{3}) - Q_{\mu}(x_{1},x_{3},x_{2})\right) \\
&= \frac{1}{2}\left(2S_{\mu}(x_{1},x_{2};x_{3}) + S_{\mu}(x_{2},x_{3};x_{1}) - 2S_{\mu}(x_{1},x_{3};x_{2}) - S_{\mu}(x_{3},x_{2};x_{1})\right) \\
&= S_{\mu}(x_{1},x_{2};x_{3}) + S_{\mu}(x_{2},x_{3};x_{1}) + S_{\mu}(x_{3},x_{1};x_{2}) \equiv0, \tag*{(108)}
\end{aligned} Q μ ( x 1 , x 2 , x 3 ) → 2 1 ( Q μ ( x 1 , x 2 , x 3 ) − Q μ ( x 1 , x 3 , x 2 ) ) = 2 1 ( 2 S μ ( x 1 , x 2 ; x 3 ) + S μ ( x 2 , x 3 ; x 1 ) − 2 S μ ( x 1 , x 3 ; x 2 ) − S μ ( x 3 , x 2 ; x 1 ) ) = S μ ( x 1 , x 2 ; x 3 ) + S μ ( x 2 , x 3 ; x 1 ) + S μ ( x 3 , x 1 ; x 2 ) ≡ 0 , (108)
which implies that (106) vanishes, thus δ ν I L H = 0 \delta_{\nu}I^{H}_{L} = 0 δ ν I L H = 0 for any variation δ G \delta G δ G such that the system remains an insulator without a Luttinger surface. The proof ultimately relies on the condition that the Luttinger-Ward functional is three times differentiable, including finite-transfer off-diagonal directions.
References