\documentclass[10pt]{book} \usepackage{graphicx} \usepackage{amsmath} %% pour mettre du texte en mode math \usepackage{tcolorbox} \newcommand\uom{\underline{\omega}} \newcommand\Oms{{\Omega}_s} \newcommand\hOms{\frac{{\Omega}_s}{2}} \begin{document} \chapter{Strinati for the big kids} We wish to solve the self-consistent Bethe-Salpeter equation: \begin{equation*} L(1,2;1',2') = L_0(1,2;1',2') + \int d3456 \; L_0(1,4;1',3) \Xi[3,5;4,6] L(6,2;5,2') \end{equation*} where e.g. $1=(x_1,t_1)$ is a space/spin and time position and: \begin{align*} iL(1,2;1',2') &= -G_2(1,2 ; 1',2') + G(1,1')G(2,2') \\ iL_0(1,2;1',2') &= G(1,2')G(2,1') \\ iG(1,2') &= \langle N | T {\hat \psi}(1) {\hat \psi}^{\dagger}(2') | N \rangle \\ i^2 G_2(1,2 ; 1',2') &= \langle N | T \Big[ {\hat \psi}(1) {\hat \psi}(2) {\hat \psi}^{\dagger}(2') {\hat \psi}^{\dagger}(1') \Big] | N \rangle \end{align*} For optical properties, with instantaneous creation/destruction of excitations, we are interested in the collapse of the time variables starting with imposing $t_2'=t_2^{+}$ in the right-hand side $L(1,2;1',2')$ and left hand-side $L(6,2;5,2')$ response functions. Since $t_2'=t_2^{+}=(t_2 + 0^+)$ indicates that the (2,2') operators cannot be separated, we see that the ordering operator selects two possible channels depending on the times ordering, namely: \begin{align*} i^2 G_2^{I}(1,2 ; 1',2') &= \theta( t_{m}^{11'} -t_2 )\langle N | T [{\hat \psi}(1) {\hat \psi}^{\dagger}(1') ] T [ {\hat \psi}(2) {\hat \psi}^{\dagger}(2') ] | N \rangle \\ i^2 G_2^{II}(1,2 ; 1',2') &= \theta( t_2 - t_{M}^{11'} )\langle N | T [ {\hat \psi}(2) {\hat \psi}^{\dagger}(2') ] T [{\hat \psi}(1) {\hat \psi}^{\dagger}(1') ] | N \rangle \end{align*} with $t_{M}^{11'} = \max(t_1,t_{1'}) = | \tau_{11'} |/2 + t^{11'}$ and $t_{m}^{11'} = \min(t_1,t_{1'}) = - | \tau_{11'} |/2 + t^{11'}$ where $\tau_{11'} = t_1 - t_{1'}$ and $t^{11'} = (t_1 + t_{1'})/2$. Introducing the complete set of eigenstates of the N-electron systems $\lbrace | N,s \rangle \rbrace$, where s index the eigenstates and $| N,s=0 \rangle = | N \rangle$ the ground-state, one obtains for $G^{I}$ : \begin{align*} i^2 G_2^{I}(1,2 ; 1',2') &= \theta( t_{m}^{11'} -t_2 ) \sum_{s>0} \langle N | T {\hat \psi}(1) {\hat \psi}^{\dagger}(1') | N,s \rangle \langle N,s | T {\hat \psi}(2) {\hat \psi}^{\dagger}(2') | N \rangle \\ & + \theta( t_{m}^{11'} -t_2 ) \frac{ G(1,1') }{ i } \frac{ G(2,2') }{ i } \\ \end{align*} where in the right-hand side of the first line the sum $(s>0)$ avoids the (s=0) ground-state contribution that builds the 1-body Green's functions product in the 2nd line. This yields a first channel for $L(1,2;1',2')$ namely: $$ iL^{I}(1,2;1',2') = \theta( t_{m}^{11'} -t_2 ) \sum_{s>0} \langle N | T {\hat \psi}(1) {\hat \psi}^{\dagger}(1') | N,s \rangle \langle N,s | T {\hat \psi}(2) {\hat \psi}^{\dagger}(2') | N \rangle $$ Switching from the Heisenberg representation (time-dependent operators) to the Schr{\"o}dinger one for the field operators, with e.g. $$ {\hat \psi}(2) = {\hat \psi}(x_2 t_2) = e^{ i H t_2} {\hat \psi}(x_2) e^{ -i H t_2} $$ one obtains \begin{align*} \langle N,s | T {\hat \psi}(2) {\hat \psi}^{\dagger}(2') | N \rangle &= - \langle N,s | e^{i H t_2^+} {\hat \psi}^{\dagger}(x_2') e^{i H (t_2 - t_2^+) } {\hat \psi}(x_2) e^{-i H t_2} | N \rangle \\ &= - \langle N,s | {\hat \psi}^{\dagger}(x_{2'}) {\hat \psi}(x_2) | N \rangle \times e^{i \Omega_s t_2 } \\ & = {\tilde \chi}_s(x_2,x_{2'}) e^{i \Omega_s t_2 } \end{align*} with $\;{\tilde \chi}_s(x_2,x_{2'}) = \langle N,s | {\hat \psi}(x_2) {\hat \psi}^{\dagger}(x_{2'}) | N \rangle$. The energy $\Omega_s = E_s^N - E_0^N$ is the s-th neutral excitation energy of the system, namely the energy we are looking for. The (-) sign in the right-hand side comes from the commutation of the creation and destruction operators with $t_2^+ > t_2$. A similar calculation for the second channel [$t_2 > \max(t_1,t_{1'})$] leads to a frequency dependence in $e^{ -i \Omega_s t_2 }$. Putting everything together, one obtains : \begin{align*} iL (1,2;1',2') = & \; \theta( t_{m}^{11'} -t_2 ) \sum_{s>0} \langle N | T {\hat \psi}(1) {\hat \psi}^{\dagger}(1') | N,s \rangle {\tilde \chi}_s(x_2,x_{2'}) \times e^{i \Omega_s t_2 } \\ - & \; \theta( t_2 - t_{M}^{11'} ) \sum_{s>0} \langle N,s | T {\hat \psi}(1) {\hat \psi}^{\dagger}(1') | N \rangle {\chi}_s(x_2,x_{2'}) \times e^{ -i \Omega_s t_2 } \end{align*} with $\;{\chi}_s(x_2,x_{2'}) = \langle N | {\hat \psi}(x_2) {\hat \psi}^{\dagger}(x_{2'}) | N,s \rangle$. \\ Performing the very same analysis for $L(6,2;5,2')$ and picking up a specific $e^{ +i \Omega_s t_2 }$ contribution, dividing by $ \langle N,s | T {\hat \psi}(2) {\hat \psi}^{\dagger}(2') | N \rangle $ both sides of the Bethe-Salpeter equation leads to: \\ \begin{tcolorbox}[ width=\textwidth, colback={white}] \begin{align} \langle N & | T {\hat \psi}(1) {\hat \psi}^{\dagger}(1') | N,s \rangle \theta( t_{m}^{11'} -t_2 ) = \\ &= \int d3456 \; L_0(1,4;1',3) \Xi[3,5;4,6] \langle N | T {\hat \psi}(6) {\hat \psi}^{\dagger}(5) | N,s \rangle \theta( t_{m}^{56} -t_2 ) \nonumber \end{align} \end{tcolorbox} \noindent where the $L_0(1,2;1',2')$ does not contribute if \textbf{ we select some $\Omega_s$ frequency below the quasiparticle gap of the system}. This is the common situation in molecular systems where due to excitonic effects, namely electron-hole interaction, the lowest optical excitation energies are lower than the photoemission quasiparticle gap. $ L_0(1,2;1',2')$ has indeed a lowest excitation energy at the energy gap $E_g$ and cannot participate to the $e^{i \Omega_s t_2}$ response of the system if $\Omega_s < E_g$. \\ This is equation (11.3) of Strinati with e.g. $$ \langle N | T {\hat \psi}(1) {\hat \psi}^{\dagger}(1') | N,s \rangle = \chi_s(x_1,x_{1'}; \tau_1) e^{-i \Omega_s t^1}$$ (see equation G.3a of Appendix G) but keeping the Heaviside fonctions for a better definition of time integrals. \\ \noindent \textbf{Spectral representation of $G(t_1-t_3)G(t_4-t_1^+)$ } \\ We now consider the $t_1$ time. Using the same algebra as above, one obtains $$ \langle N | T {\hat \psi}(1) {\hat \psi}^{\dagger}(1') | N,s \rangle = \langle N | T {\hat \psi}(x_1) {\hat \psi}^{\dagger}(x_{1'}) | N,s \rangle \times e^{- i \Oms t_1 } $$ imposing now $t_1' = t_1^+$. To extract the $e^{-i \Omega_s t_1 }$ behavior in the right hand-side of the BSE equation, one must find the Fourier components with respect to the $t_1$ variable of $L_0(1,4;1',3) = -i G(1,3) G(4,1')$ by multiplying by $ \int dt_1 e^{i \omega_1 t_1}$ both sides of the BSE equation, \textcolor{red}{ with $( \omega_1 \rightarrow \Oms + i\delta )$ where we introduce the small infinitesimal $\delta = 0^+$ for defining properly the $t_1$ time integral on the left-hand side: } $$ \int dt_1 e^{i \omega_1 t_1} e^{ - i \Oms t_1 } \theta( t_1 -t_2 ) = e^{-\delta t_2} \int d \tau e^{- \delta \tau } \theta( \tau ) = ( \delta^{-1} ) $$ This yields an annoying but useful factor that will canceled out in the following. Expressing again the field operators in their Schr\"{o}dinger representation, one see that the 1-body Green's functions depend on the time-differences $\tau_{13} = t_1 - t_3$ and $\tau_{41} = t_4 - t_1$, respectively, with e.g. $$ \langle N | \psi(1) \psi^{\dagger}(3) | N \rangle = \langle N | \psi(x_1) e^{-i H \tau_{13}} \psi^{\dagger}(x_3) | N \rangle e^{+i E_N^0 \tau_{13}} $$ Taking by convention the time difference as that of the destruction operator minus that of the creation analog and introducing our notations for the Fourier representation : \begin{equation*} G( \tau_{13} ) = {1 \over 2\pi} \int d \omega \; G(\omega) e^{ - i \omega \tau_{13} } \;\;\;\;\; G(\tau_{41} ) = {1 \over 2\pi} \int d \omega' \; G(\omega') e^{ - i {\omega}' \tau_{41} } \end{equation*} one obtains (keeping only time/frequency variables for clarity) $$ i L_0(1,4;1',3) = \int { d\omega \over 2\pi } { d{\omega}' \over 2\pi } \; G(\omega) G({\omega}') e^{ -i \omega \tau_{13} } e^{ - i {\omega}' \tau_{41} } $$ The Fourier components of ($iL_0$) with respect to the ($t_1$) time are: \begin{align*} [iL_0]( \omega_1 ) &= \int dt_1 e^{ i \omega_1 t_1 } \int { d\omega \over 2\pi } { d{\omega}' \over 2\pi } \; G(\omega ) G( {\omega}' ) e^{- i \omega (t_1 - t_3) } e^{ -i {\omega}' (t_4 - t_1) } \\ &= {1 \over (2\pi)^2 } \int d \omega d{\omega}' \; G(\omega ) G( {\omega}' ) 2\pi \delta( -\omega + \omega_1 + {\omega}' ) e^{ i \omega t_3 } e^{ - i {\omega}' t_4 } \\ &= {1 \over 2\pi } \int d \omega \; G(\omega ) G( {\omega} - \omega_1 ) e^{ i \omega t_3 } e^{ -i ( \omega - \omega_1) t_4 } \\ &= { 1 \over 2\pi } \int d \omega \; G(\omega - {\omega_1 \over 2} ) G( {\omega} + {\omega_1 \over 2} ) e^{ i \omega \tau_{34} } e^{ i \omega_1 t^{34} } \end{align*} where in the last line we made the change of variable $( \omega \rightarrow \omega + {\omega_1} / 2 ) $ and introduced the average time $t^{34} = (t_3 + t_4) /2$. We now make the identification $( \omega_1 = {\Oms} + i\delta )$ as done to properly pick-up the left-hand side $e^{-i \Oms t_1}$ contribution and obtain : \begin{align*} [iL_0 ]( \Omega_s) = { 1 \over 2\pi } \int d \omega \; G(\omega - \hOms ) G( {\omega} + \hOms ) e^{ i \omega \tau_{34} } e^{ i \Oms^{+} t^{34} } \end{align*} with $( \Oms^{+} = \Oms + i\delta )$ keeping the infinitesimal $\delta$ where it will be needed. Using the Lehman representation of the 1-body Green's functions, e.g. $$ G(x_1,x_3 ; \omega) = \sum_n \frac{ \phi_n(x_1) \phi_n^*(x_3) } { \omega - \varepsilon_n + i \eta \text{sgn}(\varepsilon_n - \mu) } $$ where we reintroduced the (space, spin) variables, one obtains by picking the residues associated with the poles of each Green's function: \begin{align*} \int d \omega & \; G(x_1,x_3; \omega - \hOms ) G(x_4,x_{1'}; \omega + \hOms ) e^{ i \omega \tau_{34} } = \\ & 2 {i} \pi \theta( \tau_{34} ) \sum_{nj} \frac{ \phi_n(x_1) \phi_n^*(x_3) \phi_j(x_4) \phi_j^*(x_{1'}) } { \varepsilon_j + \Omega_s - \varepsilon_n + i \eta \times \text{sgn}(\varepsilon_n - \mu) } e^{i (\varepsilon_j + {\Omega_s \over 2} ) \tau_{34} } \\ + & 2 {i} \pi \theta( \tau_{34} ) \sum_{jn} \frac{ \phi_j(x_1) \phi_j^*(x_3) \phi_n(x_4) \phi_n^*(x_{1'}) } { \varepsilon_j - \Omega_s - \varepsilon_n + i \eta \times \text{sgn}(\varepsilon_n - \mu) } e^{i (\varepsilon_j - {\Omega_s \over 2} ) \tau_{34} } \\ - & 2 {i} \pi \theta(- \tau_{34} ) \sum_{nb} \frac{ \phi_n(x_1) \phi_n^*(x_3) \phi_b(x_4) \phi_b^*(x_{1'}) } { \varepsilon_b + \Omega_s - \varepsilon_n + i \eta \times \text{sgn}(\varepsilon_n - \mu) } e^{i (\varepsilon_b + {\Omega_s \over 2} ) \tau_{34} } \\ - & 2 {i} \pi \theta(- \tau_{34} ) \sum_{bn} \frac{ \phi_b(x_1) \phi_b^*(x_3) \phi_n(x_4) \phi_n^*(x_{1'}) } { \varepsilon_b - \Omega_s - \varepsilon_n + i \eta \times \text{sgn}(\varepsilon_n - \mu) } e^{i (\varepsilon_b - {\Omega_s \over 2} ) \tau_{34} } \end{align*} where for $\tau_{34} > 0$ (resp. $\tau_{34} < 0$) the contour is close in the upper (resp. lower) half complex plane. We adopted chemist notations with (i,j) and (a,b) indexing occupied and virtual states, respectively. Projecting now on $ \phi_a^*(x_1) \phi_i(x_{1'}) \; $ one obtains by orthonormalization : \begin{align*} \int dx_1 dx_{1'} \; \phi_a^*(x_1) \phi_i(x_{1'}) \int d \omega & \; G(x_1,x_3; \omega + {\Omega_s \over 2} ) G(x_4,x_{1'}; \omega - {\Omega_s \over 2} ) e^{ i \omega \tau_{34} } = \\ 2 & {i} \pi \theta( \tau_{34} ) \frac{ \phi_a^*(x_3) \phi_i(x_4) } { \varepsilon_i + \Omega_s - \varepsilon_a + i \eta } e^{i (\varepsilon_i + {\Omega_s \over 2} ) \tau_{34} } \\ -2 & {i} \pi \theta( - \tau_{34} ) \frac{ \phi_a^*(x_3) \phi_i(x_4) } { \varepsilon_a - \Omega_s - \varepsilon_i - i \eta } e^{i (\varepsilon_a - {\Omega_s \over 2} ) \tau_{34} } \end{align*} After multiplication by $ ( \varepsilon_a - \varepsilon_i - \Omega_s -i \eta )$ (and setting $\eta \rightarrow 0$), the Bethe-Salpeter equation becomes: \\ \begin{tcolorbox}[ width=\textwidth, colback={white}] \begin{align*} ( \varepsilon_a - \varepsilon_i & - \Omega_s ) \int dx_1 \phi_a^*(x_1) \phi_i(x_{1'}) \langle N | T {\hat \psi}(x_1) {\hat \psi}^{\dagger}(x_{1'}) | N,s \rangle ( \delta^{-1} ) = \nonumber \\ & - \int d34 \; \phi_a^*(x_3) \phi_i(x_4) \left[ \theta( \tau_{34} ) e^{i (\varepsilon_i + {\Omega_s \over 2} ) \tau_{34} } + \theta( - \tau_{34} ) e^{i (\varepsilon_a - {\Omega_s \over 2} ) \tau_{34} } \right] \\ & \times e^{ i \Omega_s^{+} t^{34} } \int d56 \; \Xi[3,6;4,5] \langle N | T {\hat \psi}(5) {\hat \psi}^{\dagger}(6) | N,s \rangle \times \theta( t^{56}_m - t_2 ) \end{align*} \end{tcolorbox} \vskip .3cm \noindent This is equation (11.5) by Strinati with the same $$ \theta( \tau ) e^{i (\varepsilon_i + {\Omega_s \over 2} ) \tau } + \theta( - \tau ) e^{ i (\varepsilon_a - {\Omega_s \over 2} ) \tau } = e^{i {\Omega_s \over 2} \tau } \Big( \theta( \tau ) e^{i \varepsilon_i \tau } + \theta( - \tau ) e^{i (\varepsilon_a - \Omega_s )\tau } \Big) $$ time-structure where $\tau = \tau_{34} = t_3 - t_4 $. \\ \noindent{\textbf{The $GW$ approximation. } Adopting the $GW$ approximation, and neglecting the $(\partial W / \partial G)$ of higher order in $W$, one obtains: \begin{equation} \Xi[3,6;4,5] = v(3,5) \delta(3,4) \delta(5,6) - W(3^+,4) \delta(3,5) \delta(4,6) \end{equation} so that the Bethe-Salpeter equation becomes: \begin{align*} ( \varepsilon_a - \varepsilon_i & - \Omega_s ) \int dx_1 \phi_a^*(x_1) \phi_i(x_{1'}) \langle N | T {\hat \psi}(x_1) {\hat \psi}^{\dagger}(x_{1'}) | N,s \rangle ( \delta^{-1} ) = \nonumber \\ & - \int d35 \; \phi_a^*(x_3) \phi_i(x_3) e^{ i \Omega_s^{+} t_3 } v(3,5) \langle N | T [{\hat \psi}(5) {\hat \psi}^{\dagger}(5) ] | N,s \rangle \times \theta( t_5 - t_2 ) \\ & + \int d34 \; \phi_a^*(x_3) \phi_i(x_4) \left[ \theta( \tau_{34} ) e^{i (\varepsilon_i + {\Omega_s \over 2} ) \tau_{34} } + \theta( - \tau_{34} ) e^{i (\varepsilon_a - {\Omega_s \over 2} ) \tau_{34} } \right] \times \\ & \hskip 1cm \times e^{ i \Omega_s^{+} t^{34} } W(3^+,4) \langle N | T [{\hat \psi}(3) {\hat \psi}^{\dagger}(4) ] | N,s \rangle \times \theta( t^{34}_m - t_2 ) \end{align*} \\ \noindent\textbf{Spectral representation of $\langle N | T {\hat \psi}(3) {\hat \psi}^{\dagger}(4) | N,s \rangle$ } \\ \noindent Starting with: \begin{align*} \langle N | T {\hat \psi}(3) {\hat \psi}^{\dagger}(4) | N,s \rangle &= \theta( \tau_{34} ) \langle N | {\hat \psi}(3) {\hat \psi}^{\dagger}(4) | N,s \rangle \\ &- \theta( -\tau_{34} ) \langle N | {\hat \psi}^{\dagger}(4) {\hat \psi}(3) | N,s \rangle \end{align*} and decomposing the static (Schr\"{o}dinger) field operators over the Hartree-Fock molecular orbitals creation/destruction operators: $$ {\hat \psi}^{\dagger}(x_4) = \sum_n \phi_n^*(x_4) {\hat a}_n^{\dagger} \;\;\; \text{ and} \;\;\; {\hat \psi}(x_3) = \sum_m \phi_m(x_3) {\hat a}_m $$ one obtains \begin{align*} \langle N | T {\hat \psi}(3) {\hat \psi}^{\dagger}(4) | N,s \rangle &= \sum_{mn} \phi_m(x_3) \phi_n^*(x_4) \times \\ \times & \Big( \theta( \tau_{34} ) \langle N | {\hat a}_m e^{- i H \tau_{34} } {\hat a}^{\dagger}_n | N,s \rangle e^{-i E_s^N t_4} e^{+ i E_0^N t_3} \\ &- \theta( -\tau_{34} ) \langle N | {\hat a_n}^{\dagger} e^{ i H \tau_{34} } {\hat a}_m | N,s \rangle e^{ - i E_s^N t_3} e^{ + i E_0^N t_4} \Big) \end{align*} Taking the conjugate of the $ {\hat a}_m e^{- i H \tau_{34} } $ and $ {\hat a_n}^{\dagger} e^{ i H \tau_{34} } $ operators to act on the ground-state $| N \rangle$ state, one obtains: $$ e^{ i H \tau } {\hat a}_m^{\dagger} | N \rangle = e^{ i (E_0^N + \varepsilon_m ) \tau } {\hat a} _m^{\dagger} | N \rangle \;\;\;\;\; \text{and} \;\;\;\; e^{ - i H \tau } {\hat a}_n | N \rangle = e^{ - i ( E_0^N - \varepsilon_n ) \tau } {\hat a}_n | N \rangle $$ This properties hold in the case of Hartree-Fock formalism (Koopman's theorem) but holds more generally if $\lbrace \varepsilon_m , \varepsilon_n \rbrace$ are quasiparticle energies, namely true electronic removal or addition energies such as in the $GW$ formalism. Forming the corresponding bra, one obtains : \begin{align*} \langle N | T {\hat \psi}(3) & {\hat \psi}^{\dagger}(4) | N,s \rangle = - \sum_{mn} \phi_m(x_3) \phi_n^*(x_4) \langle N | {\hat a}_n^{\dagger} {\hat a}_m | N,s \rangle \times \\ \times & \Big( \theta( \tau_{34} ) e^{- i \varepsilon_m \tau_{34} } e^{ -i \Oms t_4 } + \theta( -\tau_{34} ) e^{ - i \varepsilon_n \tau_{34} } e^{ - i \Oms t_3 } \Big) \end{align*} With $t_3 = \tau_{34}/2 + t^{34}$ and $t_4 = -\tau_{34}/2 + t^{34}$, one finally obtains : \begin{align*} \langle N | T {\hat \psi}(3) {\hat \psi}^{\dagger}(4) & | N,s \rangle = - \Big( e^{ -i \Omega_s t^{34} } \Big) \sum_{mn} \phi_m(x_3) \phi_n^*(x_4) \langle N | {\hat a}_n^{\dagger} {\hat a}_m | N,s \rangle \times \nonumber \\ \times & \Big( \theta( \tau_{34} ) e^{- i ( \varepsilon_m - \hOms ) \tau_{34} } + \theta( -\tau_{34} ) e^{ - i ( \varepsilon_n + \hOms) \tau_{34} } \Big) \end{align*} We must now discuss the (m,n) indexes. The most natural channel for ${\hat a}_m^{\dagger} {\hat a}_n | N \rangle$ to project on an excited state occurs for (n,m) indexing (occupied,virtual) orbitals, respectively, in order to promote an electron from the occupied to the empty orbitals. We will label $A_{jb}^{s} = \langle N | {\hat a}_j^{\dagger} {\hat a}_b | N,s \rangle$ these large amplitude projections. However, in the case of a generic correlated system, the ground-state does not project exclusively on the mono-determinant built from the occupied 1-body orbitals $\lbrace \phi_i \rbrace$, but contains also contributions from singly, doubly, etc. excited determinants in the full $\lbrace \phi_n \rbrace$ space. As such, the $\langle N | {\hat a}_b^{\dagger} {\hat a}_j | N,s \rangle$ weights are small in general but not strictly zero. We will label $B_{jb}^{s}$ these small amplitudes. Neglecting these $B_{jb}^{s}$ coefficients leads to the Tamm-Dancoff approximation (TDA). We start with the TDA in the following for simplicity, adding the "small B" contribution in a second step. \\ \noindent In the TDA, namely with $B_{jb}^{s} = 0$, one obtains : \begin{align*} \langle N | T {\hat \psi}(3) & {\hat \psi}^{\dagger}(4) | N,s \rangle \cong - \Big( e^{ -i \Omega_s t^{34} } \Big) \sum_{jb} \phi_b(x_3) \phi_j^*(x_4) A_{jb}^{s} \times \nonumber \\ \times & \Big( \theta( \tau_{34} ) e^{- i ( \varepsilon_b - \hOms ) \tau_{34} } + \theta( -\tau_{34} ) e^{ - i ( \varepsilon_j + \hOms) \tau_{34} } \Big) \end{align*} As a result, the integral associated with the screened Coulomb potential reads: \begin{align*} \int d34 \; \phi_a^*(x_3) & \phi_i(x_4) \left[ \theta( \tau_{34} ) e^{i (\varepsilon_i + {\Omega_s \over 2} ) \tau_{34} } + \theta( - \tau_{34} ) e^{i (\varepsilon_a - {\Omega_s \over 2} ) \tau_{34} } \right] \times \\ & \times e^{ i \Omega_s^{+} t^{34} } W(3^+,4) \langle N | T [{\hat \psi}(3) {\hat \psi}^{\dagger}(4) ] | N,s \rangle \times \theta( t^{34}_m - t_2 ) \\ = - \int d34 \; \phi_a^*(x_3) & \phi_i(x_4) \theta( t^{34}_m - t_2 ) W(3^+,4) \sum_{jb} \phi_b(x_3) \phi_j^*(x_4) A_{jb}^{s} e^{ - \delta t^{34} } \times \\ & \times \left[ \theta( \tau_{34} ) e^{i ( \varepsilon_i - \varepsilon_b + \Oms ) \tau_{34} } + \theta( - \tau_{34} ) e^{i (\varepsilon_a - \varepsilon_j - \Oms ) \tau_{34} } \right] \\ = - \sum_{jb} A_{jb}^{s} & \int d\tau_{34} dt^{34} e^{ - \delta t^{34} } \theta( t^{34} - | \tau_{34} |/2 - t_2 ) W_{ij,ab}( \tau_{34}^{+} ) \times \\ & \times \left[ \theta( \tau_{34} ) e^{i ( \varepsilon_i - \varepsilon_b + \Oms ) \tau_{34} } + \theta( - \tau_{34} ) e^{i (\varepsilon_a - \varepsilon_j - \Oms ) \tau_{34} } \right] \end{align*} where we have explicited $t^{34}_m$ and made a change of variable $(t_3,t_4) \rightarrow (\tau_{34}, t^{34})$ with unit wronskian, defining the screened Coulomb matrix elements \begin{align*} W_{ij,ab}( \tau_{34}^{+} )&= \int d{\bf r}_3 d{\bf r}_4 \; \phi_i(x_4) \phi_j^*(x_4) W(3^+,4) \phi_a^*(x_3) \phi_b(x_3) \;\;\;\; \end{align*} with $(\tau_{34}^{+} = t_3^{+} - t_4)$ and adopting Mulliken notations where we group together the index associated with orbitals taken at the same space position (and putting complexe conjugate orbitals as inner indexes). The $t^{34}$ time integral yields again $( \delta^{-1} )$ as done previously for the $t_1$-time integral, leaving only the dependence on the time difference $\tau_{34} = t_3 - t_4$. We can understand the origin of this problematic $( \delta^{-1} )$ factor : the screened- Coulomb potential contribution only depends on the time difference $\tau_{34}$ so that the time variable $t^{34}$ becomes problematic. To deal now with the bare Coulomb contribution, we need $\langle N | T {\hat \psi}(5) {\hat \psi}^{\dagger}(5) | N,s \rangle$ that in the TDA reads : $$ \langle N | T {\hat \psi}(5) {\hat \psi}^{\dagger}(5) | N,s \rangle \cong - \Big( e^{ -i \Omega_s t_5 } \Big) \sum_{jb} \phi_b(x_5) \phi_j^*(x_5) A_{jb}^{s} $$ leading to the following contribution : \begin{align*} - \int d35 \; \phi_a^*(x_3) & \phi_i(x_3) e^{ i \Omega_s^{+} t_3 } v(3,5) \langle N | T [{\hat \psi}(5) {\hat \psi}^{\dagger}(5) ] | N,s \rangle \times \theta( t_5 - t_2 ) \\ & = \sum_{jb} A_{jb}^{s} v_{ia,jb} \int dt_3 dt_5 \delta( t_3 - t_5) e^{ i \Omega_s^{+} t_3 } e^{ -i \Omega_s t_5 }\theta( t_5 - t_2 ) \end{align*} with $v(3,5) = v( {\bf r}_3 - {\bf r}_5 ) \delta( t_3 - t_5)$ a static interaction and : $$ v_{ia,jb} = \int d{\bf r}_3 d{\bf r}_5 \; \phi_i(x_3) \phi_a^*(x_3) v( {\bf r}_3 - {\bf r}_5 ) \phi_j^*(x_5) \phi_b(x_5) $$ The time integral reads now $\int dt_3 e^{- \delta t_3} \theta(t_3 - t_2 )$ yielding $( \delta^{-1} )$ again. This factor in front of all terms can now be factorized out. \\ \noindent Finally, for the right-hand side of the Bethe-Salpeter equation, we develop: \begin{equation*} \langle N | {\hat \psi}^{\dagger}(x_{1'}) {\hat \psi}(x_1) | N,s \rangle = \sum_{jb} \phi_b(x_1) \phi_j^*(x_1) A_{jb}^{s} \end{equation*} so that by orthonormalization: \begin{equation*} \int dx_1 dx_{1'} \; \phi_a^*(x_1) \phi_i(x_{1'}) \langle N | {\hat \psi}^{\dagger}(x_{1'}) {\hat \psi}(x_1) | N,s \rangle = \sum_{jb} A_{jb}^{s} \delta_{ij} \delta_{ab} = A_{ia}^{s} \end{equation*} Note that the projection on $\phi_a^*(x_1) \phi_i(x_{1'})$ will always cancel the small $B_{jb}^{s}$ contribution so that this last expression is always valid. We obtain thus a much simplified reformulation of the dynamical TDA Bethe-Salpeter equation: \\ \begin{tcolorbox}[ width=\textwidth, colback={white}] \begin{align*} ( \varepsilon_a - \varepsilon_i - \Omega_s ) & A_{ia}^{s} + \sum_{jb} A_{jb}^{s} \times v_{ai,bj} \stackrel{\text{TDA}}{=} \sum_{jb} A_{jb}^{s} \times \int d\tau \; W_{ij,ab}( {\tau}^{+} ) \times \\ & \times \left[ \theta( \tau ) e^{i (\varepsilon_i - \varepsilon_b+ \Oms ) \tau } + \theta( - \tau ) e^{i (\varepsilon_a - \varepsilon_j - \Oms ) \tau } \right] \end{align*} \end{tcolorbox} \noindent where we dropped the index of ${\tau}_{34}$. Taking finally the Fourier representation of $W$, the time integration reads: \begin{align*} \widetilde{W}_{ij,ab}(\Oms) =& \int d\tau \; W_{ij,ab}( \tau^{+} ) \times \; \left[ \theta( \tau ) e^{i (\varepsilon_i - \varepsilon_b+ \Oms ) \tau } + \theta( - \tau ) e^{i (\varepsilon_a - \varepsilon_j - \Oms ) \tau } \right] \\ =& \int { d\omega \over 2\pi } W_{ij,ab}( \omega ) \int d\tau \; e^{-i \omega ( \tau + 0^+) } \left[ \theta( \tau ) e^{i (\varepsilon_i - \varepsilon_b+ \Oms ) \tau } + \theta( - \tau ) e^{i (\varepsilon_a - \varepsilon_j - \Oms ) \tau } \right] \\ =& { i \over 2\pi} \int d\omega e^{-i \omega 0^{+}} \left[ \frac{ W_{ij,ab}( \omega ) }{ ( \Oms - \omega ) - ( \varepsilon_b - \varepsilon_i) + i\eta } + \frac{ W_{ij,ab}( \omega ) }{ ( \Oms + \omega) - ( \varepsilon_a - \varepsilon_j ) + i\eta } \right] \end{align*} where we have introduced again a small damping $( \Oms \rightarrow \Oms + i\eta)$ of the excitation for convergency. This leads to equation (11.8) by Strinati: \\ \begin{tcolorbox}[ width=\textwidth, colback={white}] \begin{align*} ( \varepsilon_a - \varepsilon_i - \Omega_s ) & A_{ia}^{s} + \sum_{jb} A_{jb}^{s} \times v_{ai,bj} \stackrel{\text{TDA}}{=} { i \over 2 \pi} \sum_{jb} A_{jb}^{s} \int d\omega e^{-i \omega 0^+ } W_{ij,ab} \times \\ \hskip 1cm &\times \left[ \frac{1}{( \Oms - \omega) - (\varepsilon_b - \varepsilon_i) +i \eta } + \frac{1}{ (\Oms + \omega) - (\varepsilon_a - \varepsilon_j) + i\eta } \right] \end{align*} \end{tcolorbox} \vskip .3cm \noindent or to a form similar to the static limit : \\ \begin{tcolorbox}[ width=\textwidth, colback={white}] \begin{align} ( \varepsilon_a - \varepsilon_i - \Omega_s ) A_{ia}^{s} + \sum_{jb} \Big( v_{ai,bj} - \widetilde{W}_{ij,ab}(\Oms) \Big) A_{jb}^{s} \stackrel{\text{TDA}}{=} 0 \end{align} \end{tcolorbox} \vskip .3cm \noindent but with an effective dynamical screened Coulomb potential (see Pina eq. 24): \\ \begin{tcolorbox}[ width=\textwidth, colback={white}] \begin{align} \widetilde{W}_{ij,ab}(\Oms) &= { i \over 2 \pi} \int d\omega \; e^{-i \omega 0^+ } W_{ij,ab} \times \\ \hskip 1cm &\times \left[ \frac{1}{ (\Oms - \omega) - (\varepsilon_b - \varepsilon_i) +i \eta } + \frac{1}{ (\Oms + \omega) - (\varepsilon_a - \varepsilon_j) + i\eta } \right] \nonumber \end{align} \end{tcolorbox} \vskip .3cm \noindent Taking now the spectral representation for the (time-ordered) $W_{ij,ab}( \omega )$, namely: \begin{align*} W_{ij,ab}(\omega) &= (ij|ab) + 2 \sum_m^{OV} [ij|m] [ab|m] \times \\ & \times \Big( \frac{1}{ \omega-\Omega_m^{RPA} + i\eta } - \frac{1}{ \omega + \Omega_m^{RPA} - i\eta } \Big) \end{align*} the frequency integral becomes, closing in the lower half plane thanks to the $e^{-i\omega 0^+}$ factor. \textcolor{red}{(This is the only place where the (+) in $3^+$ makes a difference as for the $GW$ operator ... check carefully origin ...)}: \begin{align*} \int d\omega e^{-i \omega 0^{+}} & \Big( \frac{ W_{ij,ab}( \omega ) }{ ( \Oms - \omega ) - ( \varepsilon_b - \varepsilon_i) + i\eta } + \frac{ W_{ij,ab}( \omega ) }{ ( \Oms + \omega) - ( \varepsilon_a - \varepsilon_j ) + i\eta } \Big)= \\ = & (-2i \pi) W_{ij,ab}( - \Oms + (\varepsilon_a - \varepsilon_j) ) + (-2i \pi) \times 2 \sum_m^{OV} [ij|m] [ab|m] \times \\ & \times \Big( \frac{ 1 }{ ( \Oms - \Omega_m^{RPA} ) - ( \varepsilon_b - \varepsilon_i) + i\eta } + \frac{ 1 }{ ( \Oms + \Omega_m^{RPA} ) - ( \varepsilon_a - \varepsilon_j ) + i\eta } \Big) \end{align*} that is : \begin{align*} \widetilde{W}_{ij,ab}( \Oms ) = & (ij|ab) + 2 \sum_m^{OV} [ij|m] [ab|m] \times \\ \times & \Big( \frac{ 1}{ -\Oms + (\varepsilon_a - \varepsilon_j) -\Omega_m^{RPA} + i\eta } - \frac{ 1}{ -\Oms + (\varepsilon_a - \varepsilon_j) + \Omega_m^{RPA} - i\eta } \\ + & \frac{ 1 }{ ( \Oms - \Omega_m^{RPA} ) - ( \varepsilon_b - \varepsilon_i) + i\eta } + \frac{ 1 }{ ( \Oms + \Omega_m^{RPA} ) - ( \varepsilon_a - \varepsilon_j ) + i\eta } \Big) \end{align*} \\ or \begin{align*} \widetilde{W}_{ij,ab}( \Oms ) = & (ij|ab) + 2 \sum_m^{OV} [ij|m] [ab|m] \times \\ \times & \Big( \frac{ 1 }{ ( \Oms - \Omega_m^{RPA} ) - ( \varepsilon_b - \varepsilon_i) + i\eta } + \frac{ 1}{ ( \Oms - \Omega_m^{RPA} ) - (\varepsilon_a - \varepsilon_j) + i\eta } \\ + & \frac{ 1 }{ ( \Oms + \Omega_m^{RPA} ) - ( \varepsilon_a - \varepsilon_j ) + i\eta } - \frac{ 1}{ ( \Oms +\Omega_m^{RPA} ) - (\varepsilon_a - \varepsilon_j) - i\eta } \Big) \end{align*} \\ Keeping only \textbf{ the two resonant contributions : } (ACTUALLY the other ones in the last line cancel when $\eta \rightarrow 0$ !! check ...) one obtains Pina's equation (27) \\ \begin{tcolorbox}[ width=\textwidth, colback={white}] \begin{align} \widetilde{W}_{ij,ab}&( \Oms ) = (ij|ab) + 2 \sum_m^{OV} [ij|m] [ab|m] \times \\ \times & \left[ \frac{ 1 }{ ( \Oms - \Omega_m^{RPA} ) - ( \varepsilon_b - \varepsilon_i) + i\eta } + \frac{ 1}{ ( \Oms - \Omega_m^{RPA} ) - (\varepsilon_a - \varepsilon_j) + i\eta } \right] \nonumber \end{align} \end{tcolorbox} \noindent This dynamical $\widetilde{W}( \Oms )$ kernel can be rewritten: \begin{align} \widetilde{W}_{ij,ab}( \Oms ) &= (ij|ab) + 2 \sum_m^{OV} [ij|m] [ab|m] \times \\ & \times \left[ \frac{ 1 }{ \Omega_{ib}^{s} - \Omega_m^{RPA} + i\eta } + \frac{ 1}{ \Omega_{ja}^{s} - \Omega_m^{RPA} + i\eta } \right] \nonumber \end{align} with e.g. $ \Omega_{ib}^{s} = \Oms - ( \varepsilon_b - \varepsilon_i) $. This formulation is similar to the spectral representation of $W_{ij,ab}(\omega)$ but with $( \omega \rightarrow \Omega_{ib}^{s} )$ and $( \Omega \rightarrow \Omega_{ja}^{s} )$. \vskip 1cm \noindent {\textbf {Non-resonant contributions.}} We now add the $B_{jb}^{s}$ small contributions to the $\langle N | T {\hat \psi}(3) {\hat \psi}^{\dagger}(4) | N,s \rangle $ weight: \begin{align*} \langle N | T {\hat \psi}(3) {\hat \psi}^{\dagger}(4) | N,s \rangle^{\text{corr}} &= - \Big( e^{ -i \Omega_s t^{34} } \Big) \sum_{bj} \phi_j(x_3) \phi_b^*(x_4) B_{jb}^{s} \times \nonumber \\ \times & \Big( \theta( \tau_{34} ) e^{- i ( \varepsilon_j - \hOms ) \tau_{34} } + \theta( -\tau_{34} ) e^{ - i ( \varepsilon_b + \hOms) \tau_{34} } \Big) \end{align*} where ``corr" stands for correction. As a result, the integral associated with the screened Coulomb potential must be corrected by: \begin{align*} \int d34 \; \phi_a^*(x_3) & \phi_i(x_4) \left[ \theta( \tau_{34} ) e^{i (\varepsilon_i + {\Omega_s \over 2} ) \tau_{34} } + \theta( - \tau_{34} ) e^{i (\varepsilon_a - {\Omega_s \over 2} ) \tau_{34} } \right] \times \\ & \times e^{ i \Omega_s^{+} t^{34} } W(3^+,4) \langle N | T [{\hat \psi}(3) {\hat \psi}^{\dagger}(4) ] | N,s \rangle \times \theta( t^{34}_m - t_2 ) \\ = - \int d34 \; \phi_a^*(x_3) & \phi_i(x_4) \theta( t^{34}_m - t_2 ) W(3^+,4) \sum_{bj} \phi_j(x_3) \phi_b^*(x_4) B_{jb}^{s} e^{ - \delta t^{34} } \times \\ & \times \left[ \theta( \tau_{34} ) e^{i ( \varepsilon_i - \varepsilon_j + \Oms ) \tau_{34} } + \theta( - \tau_{34} ) e^{i (\varepsilon_a - \varepsilon_b - \Oms ) \tau_{34} } \right] \\ = - ({ \delta}^{-1} ) \sum_{jb} B_{jb}^{s} & \int d\tau \; W_{ib,aj}( {\tau}^{+} ) \times \\ & \times \left[ \theta( \tau ) e^{i ( \varepsilon_i - \varepsilon_j + \Oms ) \tau } + \theta( - \tau ) e^{i (\varepsilon_a - \varepsilon_b - \Oms ) \tau } \right] \end{align*} The time integration yields now: \begin{align*} \widetilde{W}_{ib,aj}(\Oms) =& \int d\tau \; W_{ib,aj}( \tau^{+} ) \times \; \left[ \theta( \tau ) e^{i (\varepsilon_i - \varepsilon_j+ \Oms ) \tau } + \theta( - \tau ) e^{i (\varepsilon_a - \varepsilon_b - \Oms ) \tau } \right] \\ =& \int { d\omega \over 2\pi } W_{ib,aj}( \omega ) \int d\tau \; e^{-i \omega ( \tau + 0^+) } \left[ \theta( \tau ) e^{i (\varepsilon_i - \varepsilon_j+ \Oms ) \tau } + \theta( - \tau ) e^{i (\varepsilon_a - \varepsilon_b - \Oms ) \tau } \right] \\ =& { i \over 2\pi} \int d\omega e^{-i \omega 0^{+}} \left[ \frac{ W_{ib,aj}( \omega ) }{ ( \Oms - \omega ) - ( \varepsilon_j - \varepsilon_i) + i\eta } + \frac{ W_{ib,aj}( \omega ) }{ ( \Oms + \omega) - ( \varepsilon_a - \varepsilon_b ) + i\eta } \right] \end{align*} \\ Plugging now the spectral representation of $W_{ib,aj}( \omega )$: \begin{align*} \widetilde{W}_{ib,aj}( \Oms ) = & (ib|aj) + 2 \sum_m^{OV} [ib|m] [aj|m] \times \\ \times & \Big( \frac{ 1 }{ ( \Oms - \Omega_m^{RPA} ) - ( \varepsilon_j - \varepsilon_i) + i\eta } + \frac{ 1}{ ( \Oms - \Omega_m^{RPA} ) - (\varepsilon_a - \varepsilon_b) + i\eta } \\ + & \frac{ 1 }{ ( \Oms + \Omega_m^{RPA} ) - ( \varepsilon_a - \varepsilon_b ) + i\eta } - \frac{ 1}{ ( \Oms +\Omega_m^{RPA} ) - (\varepsilon_a - \varepsilon_b) - i\eta } \Big) \end{align*} where again the last line contribution cancel. \\ \noindent The correction to the bare Coulomb contribution stems from the correction to $\langle N | T {\hat \psi}(5) {\hat \psi}^{\dagger}(5) | N,s \rangle$, namely : $$ \langle N | T {\hat \psi}(5) {\hat \psi}^{\dagger}(5) | N,s \rangle^{\text{corr}} = - \Big( e^{ -i \Omega_s t_5 } \Big) \sum_{bj} \phi_j(x_5) \phi_b^*(x_5) B_{jb}^{s} $$ leading to the following correction : \begin{align*} - \int d35 \; \phi_a^*(x_3) & \phi_i(x_3) e^{ i \Omega_s^{+} t_3 } v(3,5) \langle N | T [{\hat \psi}(5) {\hat \psi}^{\dagger}(5) ] | N,s \rangle \times \theta( t_5 - t_2 ) \\ & = \sum_{bj} B_{bj}^{s} v_{ia,bj} ( \delta^{-1} ) \end{align*} As discussed above, there is no correction to the diagonal part of the BSE Hamiltonian so that one obtains beyond TDA so that: \\ \begin{tcolorbox}[ width=\textwidth, colback={white}] \begin{align} ( \varepsilon_a - \varepsilon_i - \Omega_s ) A_{ia}^{s} +& \sum_{jb} \Big( v_{ai,bj} - \widetilde{W}_{ij,ab}(\Oms) \Big) A_{jb}^{s} \nonumber \\ + & \sum_{bj}\Big( v_{ai,jb} - \widetilde{W}_{ib,aj}(\Oms) \Big) B_{jb}^{s} = 0 \end{align} \end{tcolorbox} \noindent with: \begin{align*} \widetilde{W}_{ib,aj}( \Oms ) = & (ib|aj) + 2 \sum_m^{OV} [ib|m] [aj|m] \times \\ \times & \left[ \frac{ 1 }{ ( \Oms - \Omega_m^{RPA} ) - ( \varepsilon_j - \varepsilon_i) + i\eta } + \frac{ 1}{ ( \Oms - \Omega_m^{RPA} ) - (\varepsilon_a - \varepsilon_b) + i\eta } \right] \end{align*} %%%%%%%%%% \end{document}