I want to find a way to dress free Keldysh Green functions with the simplest level broadening. But there seems to be some quite unexpected result.

Let's consider free Keldysh Green functions in textbooks for a single-level with time-translation symmetry \begin{equation}\label{eq:simpleGs} \begin{split} G^{\mathrm{R}(\mathrm{A})}(t) = \mp i \theta(\pm t) e^{- i \varepsilon_0 t} &\longleftrightarrow \frac{1}{\varepsilon - \varepsilon_0 \pm i 0} \\ G^\mathrm{K}(t) = - i (1-2n_F) e^{- i \varepsilon_0 t} &\longleftrightarrow -2\pi i (1-2n_F)\delta(\varepsilon - \varepsilon_0) \\ G^<(t) = i n_F e^{- i \varepsilon_0 t} &\longleftrightarrow 2\pi i n_F \delta(\varepsilon - \varepsilon_0), \end{split} \end{equation} where the right hand side shows the energy space expression and $n_F$ is the Fermi distribution valued at $\varepsilon_0$. My expectation of the simplest possible level broadening is just a constant $\gamma$ as follows \begin{equation} \begin{split} \tilde{G}^{\mathrm{R}(\mathrm{A})}(t) = \mp i \theta(\pm t) e^{- i \varepsilon_0 t} e^{\mp \gamma t} &\longleftrightarrow \frac{1}{\varepsilon - \varepsilon_0 \pm i \gamma} \\ \tilde{G}^\mathrm{K}(t) = - i (1-2n_F) e^{- i \varepsilon_0 t} e^{-\gamma |t|} &\longleftrightarrow -2 i (1-2n_F) \frac{\gamma}{(\varepsilon-\varepsilon_0)^2+\gamma^2} \\ \tilde{G}^<(t) = i n_F e^{- i \varepsilon_0 t} e^{-\gamma |t|} &\longleftrightarrow 2 i n_F \frac{\gamma}{(\varepsilon-\varepsilon_0)^2+\gamma^2}, \end{split}\tag{1}\label{dressedGs} \end{equation} which are consistent with the general relation $$G^< = \frac{1}{2}(G^\mathrm{K}-G^\mathrm{R}+G^\mathrm{A}).\tag{2}\label{G^<}$$ $\tilde{G}^{\mathrm{R}(\mathrm{A})}$ looks very natural as we just replace the infinitesimal imaginary number by a finite linewidth $\gamma$. The other two, $\tilde{G}^\mathrm{K}$ and $\tilde{G}^<$, are not really weird if one recalls the approximated $\delta$-function $\delta(\varepsilon-\varepsilon_0)=\lim_{\gamma\rightarrow0}\frac{1}{\pi}\frac{\gamma}{(\varepsilon-\varepsilon_0)^2+\gamma^2}$.

In order to get this, I tried adding the self-energy $\Sigma^<(\varepsilon)= i \gamma$ and hence $\Sigma^<(t) = \frac{i\gamma}{2\pi} \int d\omega e^{-i\omega(t)} = i\gamma\delta(t)$.
With the general relation $\Sigma^{\mathrm{R}(\mathrm{A})} = \mp \Sigma^<$ valid when we only have $\Sigma^<$, we can apply this to the Dyson equation $$\tilde{G}^{\mathrm{R}(\mathrm{A})}(\varepsilon) = [1 - G^{\mathrm{R}(\mathrm{A})}\Sigma^{\mathrm{R}(\mathrm{A})}]^{-1} G^{\mathrm{R}(\mathrm{A})},$$ which exactly gives our expected $\tilde{G}^{\mathrm{R}(\mathrm{A})}$ in Eq. \eqref{dressedGs}.
Then let's apply the textbook formula [see Eq. (39) in this note] $$\tilde{G}^< = (1+\tilde{G}^\mathrm{R}\Sigma^\mathrm{R}) G^< (1+\Sigma^\mathrm{A}\tilde{G}^\mathrm{A}) + \tilde{G}^\mathrm{R} \Sigma^< \tilde{G}^\mathrm{A}\tag{3}\label{G^<formula}$$ and Eq. \eqref{G^<}. To my surprise, it gives the $\tilde{G}^\mathrm{K}$ and $\tilde{G}^<$ expressions in Eq. \eqref{dressedGs} with $n_F=\frac{1}{2}$, which is quite weird, i.e., the $n_F$ dependence is completely lost and actually $\tilde{G}^\mathrm{K}=0$. (BTW, it's straightforward to see this result. Here the whole first part before the last '+' in Eq. \eqref{G^<formula} vanishes as I've checked, which is indeed what happens for steady states as mentioned below Eq. (39) in that note. Then the remaining part $\tilde{G}^\mathrm{R} \Sigma^< \tilde{G}^\mathrm{A}$ obviously is missing $n_F$.)

Am I doing anything wrong here? Or how should one correctly get the simple level broadening to those Green functions?


1 Answer 1


The simplest case where such a constant broadening arises naturally is a level coupled to a Fermi sea, in a broadband limit (i.e., with the constant density in the Fermi sea), with Hamiltonian: $$ H=\epsilon_0 d^\dagger d + \sum_k\epsilon_k c_k^\dagger c_k + \sum_k\left(Vd^\dagger c_k + V^*c_k^\dagger d\right), $$ The broadening in this case is given by $$\gamma = 2\pi|V|^2\rho,$$ where $$\rho(\epsilon) = \sum_k\delta(\epsilon - \epsilon_k)$$ is the density of states, which is assumed constant (the broad-band limit). The self-energies are then given by $$ \Sigma^{r,a} = \mp\frac{i\Gamma}{2},\\ \Sigma^<(\omega)= i\Gamma n(\omega),\\ \Sigma^>(\omega)= -i\Gamma \left[1-n(\omega)\right], $$ which satisfy the basic relation $$ \Sigma^>-\Sigma^<=\Sigma^r-\Sigma^a. $$

Caution: I am relying on my memory here, so some factors may be incorrect. The problem is solved in some details in Jauho, Meir, and Wingreen's paper, but I am not necessarily following their notations. Another useful reference is the Rammer&Smith's review... everything else is math :)

We separate the Hamilronian into a non-interacting part and the "interaction" (tuneling): $$ H = H_0 + H_T,\\ H_0=\epsilon_0 d^\dagger d + \sum_k\epsilon_k c_k^\dagger c_k,\\ H_T = \sum_k\left(Vd^\dagger c_k + V^*c_k^\dagger d\right). $$ The Keldysh Green's function for the level is: $$ G_{dd}(t,t') = \left\langle T_K\left[Sd(t)d^\dagger(t')\right]\right\rangle, $$ where $T_K$ is ordering along the Keldysh contour, whereas the scattering matrix is given by $$ S=T_K\exp\left[-i\int_{C_K}d\tau H_T(\tau)\right], $$ and all the operators are in the interaction representation (i.e. their evolution is determined by $H_0$).

We can now use Feynmann-Dyson expansion or another method to write down the Feynmann-Dyson equation as: $$ G_{dd}(t,t') = g_{dd}(t,t') + \int_{C_K}d\tau_1d\tau_2g_{dd}(t,\tau_1)\Sigma(\tau_1,\tau_2)G_{dd}(\tau_2,t'),\\ \Sigma(\tau_1,\tau_2)=|V|^2\sum_k g_k(\tau_1,\tau_2). $$ The non-interacting Green's functions are given by $$ g_{dd}(t,t')=\left\langle T_K\left[d(t)d^\dagger(t')\right]\right\rangle,\\ g_{k,k'}(t,t')=\delta_{k,k'}g_k(t,t')=\delta_{k,k'}\left\langle T_K\left[c_k(t)c_k^\dagger(t')\right]\right\rangle. $$ This allows us to write right away: $$ \Sigma^{r,a}(\omega)=|V|^2\sum_{k}\frac{1}{\omega -\epsilon_k\pm i\eta} =\rho|V|^2\int d\epsilon \left[ \mathcal{P}\frac{1}{\omega-\epsilon} \mp\pi\delta(\omega-\epsilon)\right] = \mp\frac{i\Gamma}{2},\\ \Sigma^<(\Omega)=|V|^2\sum_k i2\pi n(\epsilon_k)\delta(\omega-\epsilon_k) = 2\pi i\rho|V|^2 \int d\epsilon n(\epsilon)\delta(\omega-\epsilon) =i\Gamma n(\omega),\\ \Sigma^<(\Omega)=-|V|^2\sum_k i2\pi \left[1-n(\epsilon_k)\right]\delta(\omega-\epsilon_k) = -2\pi i\rho|V|^2 \int d\epsilon \left[1-n(\epsilon)\right]\delta(\omega-\epsilon) =-i\Gamma \left[1-n(\omega)\right]. $$

Using the Langreth algebra and switching to Fourier space the Dyson equation can be written as: $$ G_{dd}^{r,a}(\omega)=g_{dd}^{r,a}(\omega) + g_{dd}^{r,a}(\omega)\Sigma^{r,a}(\omega)G_{dd}^{r,a}(\omega),\\ G_{dd}^{>,<}(\omega)=g_{dd}^{>,<}(\omega) + g_{dd}^{>,<}(\omega)\Sigma^{a}(\omega)G_{dd}^{a}(\omega) + g_{dd}^{r}(\omega)\Sigma^{>,<}(\omega)G_{dd}^{a}(\omega) + g_{dd}^{r}(\omega)\Sigma^{r}(\omega)G_{dd}^{>,<}(\omega). $$ The first equation can be recast as $$ \left\{\left[g_{dd}^{r,a}(\omega)\right]^{-1} - \Sigma^{r,a}(\omega)\right\}G_{dd}^{r,a}(\omega)=\left[G_{dd}^{r,a}(\omega)\right]^{-1}G_{dd}^{r,a}(\omega)=1\\ \Longrightarrow G_{dd}^{r,a}(\omega) = \frac{1}{\left[g_{dd}^{r,a}(\omega)\right]^{-1} - \Sigma^{r,a}(\omega)} = \frac{1}{\omega-\epsilon_d\pm\frac{i\Gamma}{2}} $$ The equation for the greater and the lesser functions can be written as (taking into account that $\left[g_{dd}^{r}(\omega)\right]^{-1}g_{dd}^{>,<}(\omega)=0$): $$ \left\{\left[g_{dd}^{r}(\omega)\right]^{-1} - \Sigma^{r}(\omega)\right\}G_{dd}^{>,<}(\omega) =\left[G_{dd}^{r}(\omega)\right]^{-1}G_{dd}^{>,<}(\omega)=\Sigma^{>,<}(\omega)G_{dd}^{a}(\omega)\\ \Longrightarrow G_{dd}^{>,<}(\omega) = G_{dd}^{r}(\omega)\Sigma^{>,<}(\omega)G_{dd}^{a}(\omega) = \frac{\Sigma^{>,<}(\omega)}{(\omega-\epsilon_d)^2+\frac{\Gamma^2}{4}} $$

  • $\begingroup$ Thank you for your answer. Is there an $\epsilon_k$ missing in the second term of $H$? It seems to give the desired result. But could you please elaborate on how to get these self-energies from $H$? $\endgroup$
    – xiaohuamao
    Jan 6, 2021 at 12:53
  • $\begingroup$ Indeed, I have missed $\epsilon_k$. I will try to sketch the derivation when I have a bit more time. $\endgroup$
    – Roger V.
    Jan 6, 2021 at 14:43
  • $\begingroup$ I have added the derivations. $\endgroup$
    – Roger V.
    Jan 8, 2021 at 10:26
  • $\begingroup$ This is really a nice answer! Only one question left. Your last expression $\frac{\Sigma^{>,<}(\omega)}{(\omega-\epsilon_d)^2+\frac{\Gamma^2}{4}}$ has a complex form transformed to time. To get what I wrote in Eq.(1), one needs $\frac{\Sigma^{>,<}(\epsilon_d)}{(\omega-\epsilon_d)^2+\frac{\Gamma^2}{4}}$ instead. Any way to justify this? $\endgroup$
    – xiaohuamao
    Jan 9, 2021 at 0:33
  • 1
    $\begingroup$ OK, we are on the same page and I don't object to any of the math. So you think the guess is actually wrong for $G^<$ and we just don't have a simple enough real-time form? I presume adding broadening like this is computationally convenient common practice to avoid singular $\delta$-functions or so. This is a bit surprising to me. $\endgroup$
    – xiaohuamao
    Jan 11, 2021 at 12:52

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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