### Eigenvalue equation for interacting Green's functions

93
views
0
4 months ago by
Studying the articles "Topological Hamiltonian as an exact tool for topological invariants" (<https://arxiv.org/abs/1207.7341>) and "Simplified Topological Invariants for Interacting Insulators" (<https://arxiv.org/abs/1201.6431>) I stumbled upon the following eigenvalue equation

\begin{align}
G^{-1}(k,i\omega) |\alpha(k,i\omega)> &= \mu_\alpha(k,i\omega) |\alpha(k,i\omega)>
\end{align}

where

\begin{align}
G^{-1}(k,i\omega) &= i\omega - H_0(k) - \Sigma(k,i\omega)
\end{align}

is the full Green's function of an interacting problem evaluated at the Matsubara frequencies $\omega_n = (2n+1)\pi/\beta$ (Probably a similar formular also holds for real frequencies, then just replace $i\omega \rightarrow \omega$). $H_0(k)$ is the Hamiltonian matrix of the noninteracting problem and $\Sigma(k,i\omega)$ the $\omega$ dependent self-energy that appears in Dyson's equation $G = G_0 + G_0 \Sigma G$. On a mathematical level, $G^{-1}(k,i\omega) \in\$GL(N,C) is an invertible complex $N\times N$ matrix, and $\{|\alpha(k,i\omega)>\}$ a set of orthogonal eigenvectors of that matrix with eigenvalues $\mu_\alpha(k,i\omega)$.

The eigenvalue equation can also be written as

\begin{align}
\left(i\omega - H_0(k) - \Sigma(k,i\omega)\right) |\alpha(k,i\omega)> &= \mu_\alpha(k,i\omega) |\alpha(k,i\omega)>\\
\left(H_0(k) + \Sigma(k,i\omega)\right) |\alpha(k,i\omega)> &= -(\mu_\alpha(k,i\omega)+i\omega) |\alpha(k,i\omega)>
\end{align}

where the last line looks pretty similar to the time independent Schroedinger equation of a noninteracting system

\begin{align}
H_0(k) |\alpha(k)> &= \epsilon(k)|\alpha(k)>
\end{align}

with the big difference that the first equation depends on $\omega$ and the second does not. I want to understand that equation and have the following questions

- What is the meaning of an $\omega$-dependent eigenstate?
- What is the meaning of an $\omega$-dependent (energy?) eigenvalue?

0
4 months ago by
I do not understand the meaning of an expression like
$\Big( i\omega - H_0(k) - \Sigma(k,i\omega) \Big) | \alpha(k, i\omega) \rangle$

AFAIU, $H_0(k)$ and $\Sigma(k,i\omega)$ are not operators, but scalars, unless you define precisely what you mean.

“On a mathematical level, $G^{-1}(k,i\omega) \in$ GL(N,C) is an invertible complex $N \times N$ matrix...” is quite vague for me. That's perhaps the reason you arrive at incomprehensible objects.
In these equations, $H_0(k)$, $\Sigma(k,i\omega)$ as well as $G^{-1}(k,i\omega)$ are all matrices acting on the vectors $|\alpha(k,i\omega)>$.
written 4 months ago by LidoDiCamaiore
It's better to start from a very simple problem; like a non-interacting problem or a two-level mixing (or hybridization of two levels). Then try to represent or interpret the Green's function as a matrix. You might then get an idea.

For me, it is totally unclear what is meant by that matrix representation.
written 4 months ago by AlQuemist
I agree with the Boss; the statement

$\Big( i\omega - H_0(k) - \Sigma(k,i\omega) \Big) | \alpha(k, i\omega) \rangle$

is quite ambiguous. If it is indeed, as you say, a multiplication of matrices with vectors, then this implies that the operators and states have been projected to a specific basis, and in doing this one sees that there will be integrals involved (due to inserting a complete set of states). These integrals are missing in the above equation.
written 4 months ago by Master Dragon
In my case I have $H=H_{MF}+H_{fl}$. The quadratic part can be written in matrix form as

\begin{align}
H_{MF} &= \sum_{k}
\begin{pmatrix} c_k^\dagger \\ f_k^\dagger \end{pmatrix}
\underbrace{\begin{pmatrix} \xi_k^c\cdot 1 & b^* V_0\cdot\Phi(\vec{k}) \\ b V_0\cdot\Phi(\vec{k}) & \xi_k^f\cdot 1
\end{pmatrix}}_{=H_0(k)}
\begin{pmatrix} c_k \\ f_k \end{pmatrix}
\end{align}

The self energy $\Sigma$, on the other hand, is obtained by perturbation theory and there are plenty of sums and integrals involved. But in the mentioned eigenvalue equation, I don't care how to calculate it in practice, I just act as if I knew it. It does not really matter.

I have the suspicion, however, that the equation does not have much physical meaning in the first place, but is used as a mathematical trick in the mentioned papers. Did you have a look at them? They are indeed interesting papers.
written 4 months ago by LidoDiCamaiore
You seem to be mixing apples and coconuts ;)
The ingredients of the Hamiltonian are operators on the Fock space.
The self-energy is a ‘simple’ scalar-valued function.

At any case, you cannot represent the quartic (or higher-order parts) of a Hamiltonian with matrices. The reason is fundamental: Matrices represent linear transformations (operators) while such interactions are nonlinear operators.
written 4 months ago by AlQuemist
First: you can gather the functions, that the self-energy is, in a matrix if you have different particle types.
Second: I do not represent the Hamiltonian by matrices, but the Green's function. And that comes straight from Dyson's equation
written 4 months ago by LidoDiCamaiore
Just make clear what you want to achieve with such a “matrix-method”. Perhaps, then we can find a proper way to do that.
written 4 months ago by AlQuemist
Aaaand... I think the problem is solved by now, you can read everything in my thesis later ;)
written 4 months ago by LidoDiCamaiore
Well then; good luck!
But that remains still unclear.
written 4 months ago by AlQuemist
0
4 months ago by
Actually, notice that taking this statement literally,

“On a mathematical level, $G^{-1}(k,i\omega) \in$ GL(N,C) is an invertible complex $N \times N$ matrix...”

implies that any function, $f(\mathbf{x}, y) \in GL(N,C)$! This is absurd.