703 lines
33 KiB
TeX
703 lines
33 KiB
TeX
\documentclass[aip,jcp,reprint,noshowkeys,superscriptaddress]{revtex4-2}
|
|
\usepackage{graphicx,dcolumn,bm,xcolor,microtype,multirow,amscd,amsmath,amssymb,amsfonts,physics,longtable,wrapfig,txfonts}
|
|
\usepackage[version=4]{mhchem}
|
|
|
|
\usepackage[utf8]{inputenc}
|
|
\usepackage[T1]{fontenc}
|
|
\usepackage{txfonts}
|
|
|
|
\usepackage[
|
|
colorlinks=true,
|
|
citecolor=blue,
|
|
breaklinks=true
|
|
]{hyperref}
|
|
\urlstyle{same}
|
|
|
|
\begin{document}
|
|
|
|
\title{A Tale of Three Dynamical Kernels}
|
|
|
|
\author{Pierre-Fran\c{c}ois \surname{Loos}}
|
|
\email{loos@irsamc.ups-tlse.fr}
|
|
\affiliation{\LCPQ}
|
|
|
|
\begin{abstract}
|
|
We discuss the physical properties and accuracy of three distinct dynamical (\ie, frequency-dependent) kernels for the computation of optical excitations within linear response theory:
|
|
i) an \textit{a priori} built kernel inspired by the dressed time-dependent density-functional theory (TDDFT) kernel proposed by Maitra and coworkers [\href{https://doi.org/10.1063/1.1651060}{J.~Chem.~Phys.~120, 5932 (2004)}],
|
|
ii) the dynamical kernel stemming from the Bethe-Salpeter equation (BSE) formalism derived originally by Strinati [\href{https://doi.org/10.1007/BF02725962}{Riv.~Nuovo Cimento 11, 1--86 (1988)}], and
|
|
iii) the second-order BSE kernel derived by Yang and coworkers [\href{https://doi.org/10.1063/1.4824907}{J.~Chem.~Phys.~139, 154109 (2013)}].
|
|
In particular, using a simple two-level model, we analyze, for each kernel, the appearance of spurious excitations, as first evidenced by Romaniello and collaborators [\href{https://doi.org/10.1063/1.3065669}{J.~Chem.~Phys.~130, 044108 (2009)}], due to the approximate nature of the kernels.
|
|
%\\
|
|
%\bigskip
|
|
%\begin{center}
|
|
% \boxed{\includegraphics[width=0.5\linewidth]{TOC}}
|
|
%\end{center}
|
|
%\bigskip
|
|
\end{abstract}
|
|
|
|
\maketitle
|
|
|
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
|
\section{Linear response theory}
|
|
\label{sec:LR}
|
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
|
Linear response theory is a powerful approach that allows to directly access the optical excitations $\omega_s$ of a given electronic system (such as a molecule) and their corresponding oscillator strengths $f_s$ [extracted from their eigenvectors $\T{(\bX_s \bY_s)}$] via the response of the system to a weak electromagnetic field. \cite{Casida_1995}
|
|
From a practical point of view, these quantities are obtained by solving non-linear, frequency-dependent Casida-like equations in the space of single excitations and de-excitations \cite{Casida_1995}
|
|
\begin{equation} \label{eq:LR}
|
|
\begin{pmatrix}
|
|
\bR^{\sigma}(\omega_s) & \bC^{\sigma}(\omega_s)
|
|
\\
|
|
-\bC^{\sigma}(-\omega_s) & -\bR^{\sigma}(-\omega_s)
|
|
\end{pmatrix}
|
|
\cdot
|
|
\begin{pmatrix}
|
|
\bX_s^{\sigma}
|
|
\\
|
|
\bY_s^{\sigma}
|
|
\end{pmatrix}
|
|
=
|
|
\omega_s
|
|
\begin{pmatrix}
|
|
\bX_s^{\sigma}
|
|
\\
|
|
\bY_s^{\sigma}
|
|
\end{pmatrix}
|
|
\end{equation}
|
|
where the explicit expressions of the resonant and coupling blocks, $\bR^{\sigma}(\omega)$ and $\bC^{\sigma}(\omega)$, depend on the spin manifold ($\sigma =$ $\updw$ for singlets and $\sigma =$ $\upup$ for triplets) and the level of approximation that one employs.
|
|
Neglecting the coupling block [\ie, $\bC^{\sigma}(\omega) = 0$] between the resonant and anti-resonants parts, $\bR(\omega)$ and $-\bR(-\omega)$, is known as the Tamm-Dancoff approximation (TDA).
|
|
The non-linear eigenvalue problem defined in Eq.~\eqref{eq:LR} has particle-hole symmetry which means that it is invariant via the transformation $\omega \to -\omega$.
|
|
Therefore, without loss of generality, we will restrict our analysis to positive frequencies.
|
|
The central point here is that, thanks to their non-linear nature stemming from their frequency dependence, dynamical kernels potentially generate more than just single excitations.
|
|
|
|
In the one-electron basis of (real) spatial orbitals $\lbrace \MO{p} \rbrace$, the elements of the matrices defined in Eq.~\eqref{eq:LR} have the following generic forms: \cite{Dreuw_2005}
|
|
\begin{subequations}
|
|
\begin{gather}
|
|
R_{ia,jb}^{\sigma}(\omega) = (\e{a} - \e{i}) \delta_{ij} \delta_{ab} + f_{ia,jb}^{\Hxc,\sigma}(\omega)
|
|
\\
|
|
C_{ia,jb}^{\sigma}(\omega) = f_{ia,bj}^{\Hxc,\sigma}(\omega)
|
|
\end{gather}
|
|
\end{subequations}
|
|
where $\delta_{pq}$ is the Kronecker delta, $\e{p}$ is the one-electron energy associated with $\MO{p}$, and
|
|
\begin{equation} \label{eq:kernel}
|
|
f_{ia,jb}^{\Hxc,\sigma}(\omega)
|
|
= \iint \MO{i}(\br) \MO{a}(\br) f^{\Hxc,\sigma}(\omega) \MO{j}(\br') \MO{b}(\br') d\br d\br'
|
|
\end{equation}
|
|
In Eq.~\eqref{eq:kernel},
|
|
\begin{equation} \label{eq:kernel-Hxc}
|
|
f^{\Hxc,\sigma}(\omega) = f^{\Hx,\sigma} + f^{\co,\sigma}(\omega)
|
|
\end{equation}
|
|
is the (spin-resolved) Hartree-exchange-correlation (Hxc) dynamical kernel.
|
|
Here, $i$ and $j$ are occupied orbitals, $a$ and $b$ are unoccupied orbitals, and $p$ and $q$ indicate arbitrary orbitals.
|
|
As readily seen from Eq.~\eqref{eq:kernel-Hxc}, only the correlation (c) part of the kernel is frequency dependent and, in a wave function context, the static Hartree-exchange (Hx) matrix elements read
|
|
\begin{equation}
|
|
f_{ia,jb}^{\Hx,\sigma} = \sigma \ERI{ia}{jb} - \ERI{ib}{ja}
|
|
\end{equation}
|
|
where $\sigma = 1 $ or $0$ for singlet and triplet excited states (respectively), and
|
|
\begin{equation}
|
|
\ERI{ia}{jb} = \iint \MO{i}(\br) \MO{a}(\br) \frac{1}{\abs{\br - \br'}} \MO{j}(\br') \MO{b}(\br') d\br d\br'
|
|
\end{equation}
|
|
are the usual (bare) two-electron integrals.
|
|
In the case of a spin-independent kernel, we will drop the superscript $\sigma$.
|
|
Unless otherwise stated, atomic units are used and we assume real quantities throughout this manuscript.
|
|
|
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
|
\section{The concept of dynamical quantities}
|
|
\label{sec:dyn}
|
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%s
|
|
As a chemist, it is maybe difficult to understand the concept of dynamical properties, the motivation behind their introduction, and their actual usefulness.
|
|
Here, we will try to give a pedagogical example showing the importance of dynamical quantities and their main purposes. \cite{Romaniello_2009b,Sangalli_2011,ReiningBook}
|
|
To do so, let us consider the usual chemical scenario where one wants to get the optical excitations of a given system.
|
|
In most cases, this can be done by solving a set of linear equations of the form
|
|
\begin{equation}
|
|
\label{eq:lin_sys}
|
|
\bA \cdot \bc = \omega \, \bc
|
|
\end{equation}
|
|
where $\omega$ is one of the optical excitation energies of interest and $\bc$ its transition vector .
|
|
If we assume that the operator $\bA$ has a matrix representation of size $N \times N$, this \textit{linear} set of equations yields $N$ excitation energies.
|
|
However, in practice, $N$ might be (very) large (\eg, equal to the total number of single and double excitations generated from a reference Slater determinant), and it might therefore be practically useful to recast this system as two smaller coupled systems, such that
|
|
\begin{equation}
|
|
\label{eq:lin_sys_split}
|
|
\begin{pmatrix}
|
|
\bA_1 & \T{\bb} \\
|
|
\bb & \bA_2 \\
|
|
\end{pmatrix}
|
|
\cdot
|
|
\begin{pmatrix}
|
|
\bc_1 \\
|
|
\bc_2 \\
|
|
\end{pmatrix}
|
|
= \omega
|
|
\begin{pmatrix}
|
|
\bc_1 \\
|
|
\bc_2 \\
|
|
\end{pmatrix}
|
|
\end{equation}
|
|
where the blocks $\bA_1$ and $\bA_2$, of sizes $N_1 \times N_1$ and $N_2 \times N_2$ (with $N_1 + N_2 = N$), can be associated with, for example, the single and double excitations of the system.
|
|
Note that this \textit{exact} decomposition does not alter, in any case, the values of the excitation energies, not their eigenvectors.
|
|
|
|
Solving separately each row of the system \eqref{eq:lin_sys_split} and assuming that $\omega \bI - \bA_2$ is invertible yields
|
|
\begin{subequations}
|
|
\begin{gather}
|
|
\label{eq:row1}
|
|
\bA_1 \cdot \bc_1 + \T{\bb} \cdot \bc_2 = \omega \, \bc_1
|
|
\\
|
|
\label{eq:row2}
|
|
\bc_2 = (\omega \, \bI - \bA_2)^{-1} \cdot \bb \cdot \bc_1
|
|
\end{gather}
|
|
\end{subequations}
|
|
Substituting Eq.~\eqref{eq:row2} into Eq.~\eqref{eq:row1} yields the following effective \textit{non-linear}, frequency-dependent operator
|
|
\begin{equation}
|
|
\label{eq:non_lin_sys}
|
|
\Tilde{\bA}_1(\omega) \cdot \bc_1 = \omega \, \bc_1
|
|
\end{equation}
|
|
with
|
|
\begin{equation}
|
|
\Tilde{\bA}_1(\omega) = \bA_1 + \T{\bb} \cdot (\omega \, \bI - \bA_2)^{-1} \cdot \bb
|
|
\end{equation}
|
|
which has, by construction, exactly the same solutions as the linear system \eqref{eq:lin_sys} but a smaller dimension.
|
|
For example, an operator $\Tilde{\bA}_1(\omega)$ built in the single-excitation basis can potentially provide excitation energies for double excitations thanks to its frequency-dependent nature, the information from the double excitations being ``folded'' into $\Tilde{\bA}_1(\omega)$ via Eq.~\eqref{eq:row2}. \cite{ReiningBook}
|
|
|
|
How have we been able to reduce the dimension of the problem while keeping the same number of solutions?
|
|
To do so, we have transformed a linear operator $\bA$ into a non-linear operator $\Tilde{\bA}_1(\omega)$ by making it frequency dependent.
|
|
In other words, we have sacrificed the linearity of the system in order to obtain a new, non-linear systems of equations of smaller dimension.
|
|
This procedure converting degrees of freedom into frequency or energy dependence is very general and can be applied in various contexts. \cite{Garniron_2018,QP2}
|
|
Thanks to its non-linearity, Eq.~\eqref{eq:non_lin_sys} can produce more solutions than its actual dimension.
|
|
However, because there is no free lunch, this non-linear system is obviously harder to solve than its corresponding linear analog given by Eq.~\eqref{eq:lin_sys}.
|
|
Nonetheless, approximations can be now applied to Eq.~\eqref{eq:non_lin_sys} in order to solve it efficiently.
|
|
For example, assuming that $\bA_2$ is a diagonal matrix is of common practice (see, for example, Ref.~\onlinecite{Garniron_2018} and references therein).
|
|
|
|
Another of these approximations is the so-called \textit{static} approximation, where one sets the frequency to a particular value.
|
|
For example, as commonly done within the Bethe-Salpeter equation (BSE) formalism of many-body perturbation theory (MBPT), \cite{Strinati_1988} $\Tilde{\bA}_1(\omega) = \Tilde{\bA}_1 \equiv \Tilde{\bA}_1(\omega = 0)$.
|
|
In such a way, the operator $\Tilde{\bA}_1$ is made linear again by removing its frequency-dependent nature.
|
|
A similar example in the context of time-dependent density-functional theory (TDDFT) \cite{Runge_1984} is provided by the ubiquitous adiabatic approximation, \cite{Tozer_2000} which neglects all memory effects by making static the exchange-correlation (xc) kernel (\ie, frequency-independent). \cite{Maitra_2016}
|
|
These approximations come with a heavy price as the number of solutions provided by the system of equations \eqref{eq:non_lin_sys} has now been reduced from $N$ to $N_1$.
|
|
Coming back to our example, in the static (or adiabatic) approximation, the operator $\Tilde{\bA}_1$ built in the single-excitation basis cannot provide double excitations anymore, and the $N_1$ excitation energies are associated with single excitations.
|
|
All additional solutions associated with higher excitations have been forever lost.
|
|
In the next section, we illustrate these concepts and the various tricks that can be used to recover some of these dynamical effects starting from the static eigenproblem.
|
|
|
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
|
\section{Dynamical kernels}
|
|
\label{sec:kernel}
|
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
|
|
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
|
\subsection{Exact Hamiltonian}
|
|
\label{sec:exact}
|
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
|
|
|
Let us consider a two-level quantum system made of two orbitals \cite{Romaniello_2009b} in its singlet ground state (\ie, the lowest orbital is doubly occupied).
|
|
We will label these two orbitals, $\MO{v}$ and $\MO{c}$, as valence ($v$) and conduction ($c$) orbitals with respective one-electron Hartree-Fock (HF) energies $\e{v}$ and $\e{c}$.
|
|
In a more quantum chemical language, these correspond to the HOMO and LUMO orbitals (respectively).
|
|
The ground state $\ket{0}$ has a one-electron configuration $\ket{v\bar{v}}$, while the doubly-excited state $\ket{D}$ has a configuration $\ket{c\bar{c}}$.
|
|
There is then only one single excitation possible which corresponds to the transition $v \to c$ with different spin-flip configurations.
|
|
As usual, this can produce a singlet singly-excited state $\ket{S} = (\ket{v\bar{c}} + \ket{c\bar{v}})/\sqrt{2}$, and a triplet singly-excited state $\ket{T} = (\ket{v\bar{c}} - \ket{c\bar{v}})/\sqrt{2}$. \cite{SzaboBook}
|
|
|
|
For the singlet manifold, the exact Hamiltonian in the basis of the (spin-adapted) configuration state functions reads
|
|
\begin{equation} \label{eq:H-exact}
|
|
\bH^{\updw} =
|
|
\begin{pmatrix}
|
|
\mel{0}{\hH}{0} & \mel{0}{\hH}{S} & \mel{0}{\hH}{D} \\
|
|
\mel{S}{\hH}{0} & \mel{S}{\hH}{S} & \mel{S}{\hH}{D} \\
|
|
\mel{D}{\hH}{0} & \mel{D}{\hH}{S} & \mel{D}{\hH}{D} \\
|
|
\end{pmatrix}
|
|
\end{equation}
|
|
with
|
|
\begin{subequations}
|
|
\begin{align}
|
|
\mel{0}{\hH}{0} & = 2\e{v} - \ERI{vv}{vv} = \EHF
|
|
\\
|
|
\mel{S}{\hH - \EHF}{S} & = \Delta\e{} + \ERI{vc}{cv} - \ERI{vv}{cc}
|
|
\\
|
|
\begin{split}
|
|
\mel{D}{\hH - \EHF}{D}
|
|
& = 2\Delta\e{} + \ERI{vv}{vv} + \ERI{cc}{cc}
|
|
\\
|
|
& + 2\ERI{vc}{cv} - 4\ERI{vv}{cc}
|
|
\end{split}
|
|
\\
|
|
\mel{0}{\hH}{S} & = 0
|
|
\\
|
|
\mel{S}{\hH}{D} & = \sqrt{2}[\ERI{vc}{cc} - \ERI{cv}{vv}]
|
|
\\
|
|
\mel{0}{\hH}{D} & = \ERI{vc}{cv}
|
|
\end{align}
|
|
\end{subequations}
|
|
and $\Delta\e{} = \e{c} - \e{v}$.
|
|
The energy of the only triplet state is simply $\mel{T}{\hH}{T} = \EHF + \Delta\e{} - \ERI{vv}{cc}$.
|
|
|
|
For the sake of illustration, we will use the same numerical example throughout this study, and consider the singlet ground state of the \ce{He} atom in Pople's 6-31G basis set.
|
|
This system contains two orbitals and the numerical values of the various quantities defined above are
|
|
\begin{subequations}
|
|
\begin{align}
|
|
\e{v} & = -0.914\,127
|
|
&
|
|
\e{c} & = + 1.399\,859
|
|
\\
|
|
\ERI{vv}{vv} & = 1.026\,907
|
|
&
|
|
\ERI{cc}{cc} & = 0.766\,363
|
|
\\
|
|
\ERI{vv}{cc} & = 0.858\,133
|
|
&
|
|
\ERI{vc}{cv} & = 0.227\,670
|
|
\\
|
|
\ERI{vv}{vc} & = 0.316\,490
|
|
&
|
|
\ERI{vc}{cc} & = 0.255\,554
|
|
\end{align}
|
|
\end{subequations}
|
|
This yields the following exact singlet and triplet excitation energies
|
|
\begin{align} \label{eq:exact}
|
|
\omega_{1}^{\updw} & = 1.92145
|
|
&
|
|
\omega_{3}^{\updw} & = 3.47880
|
|
&
|
|
\omega_{1}^{\upup} & = 1.47085
|
|
\end{align}
|
|
where $\omega_{1}^{\updw}$ and $\omega_{3}^{\updw}$ are the singlet single and double excitations (respectively), and $\omega_{1}^{\upup}$ is the triplet single excitation.
|
|
We are going to use these as reference for the remaining of this study.
|
|
|
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
|
\subsection{Maitra's dynamical kernel}
|
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
|
The kernel proposed by Maitra and coworkers \cite{Maitra_2004,Cave_2004} in the context of dressed TDDFT (D-TDDFT) corresponds to an \textit{ad hoc} many-body theory correction to TDDFT.
|
|
More specifically, D-TDDFT adds to the static kernel a frequency-dependent part by reverse-engineering the exact Hamiltonian: a single and double excitations, assumed to be strongly coupled, are isolated from among the spectrum and added manually to the static kernel.
|
|
The very same idea was taking further by Huix-Rotllant, Casida and coworkers, \cite{Huix-Rotllant_2011} and tested on a large set of molecules.
|
|
Here, we start instead from a HF reference.
|
|
The static problem corresponds then to the TDHF Hamiltonian, while in the TDA, it reduces to CIS.
|
|
|
|
For the two-level model, the reverse-engineering process of the exact Hamiltonian \eqref{eq:H-exact} yields
|
|
\begin{equation} \label{eq:f-Maitra}
|
|
f_M^{\co,\updw}(\omega) = \frac{\abs*{\mel{S}{\hH}{D}}^2}{\omega - (\mel{D}{\hH}{D} - \mel{0}{\hH}{0}) }
|
|
\end{equation}
|
|
while $f_M^{\co,\upup}(\omega) = 0$.
|
|
The expression \eqref{eq:f-Maitra} can be easily obtained by folding the double excitation onto the single excitation, as explained in Sec.~\ref{sec:dyn}.
|
|
It is clear that one must know \textit{a priori} the structure of the Hamiltonian to construct such dynamical kernel, and this obviously hampers its applicability to realistic photochemical systems where it is sometimes hard to get a clear picture of the interplay between excited states. \cite{Boggio-Pasqua_2007}
|
|
|
|
For the two-level model, the non-linear equations defined in Eq.~\eqref{eq:LR} provides the following effective Hamiltonian
|
|
\begin{equation} \label{eq:H-M}
|
|
\bH_\text{D-TDHF}^{\sigma}(\omega) =
|
|
\begin{pmatrix}
|
|
R_M^{\sigma}(\omega) & C_M^{\sigma}(\omega)
|
|
\\
|
|
-C_M^{\sigma}(-\omega) & -R_M^{\sigma}(-\omega)
|
|
\end{pmatrix}
|
|
\end{equation}
|
|
with
|
|
\begin{subequations}
|
|
\begin{gather}
|
|
\label{eq:R_M}
|
|
R_M^{\sigma}(\omega) = \Delta\e{} + 2 \sigma \ERI{vc}{vc} - \ERI{vc}{vc} + f_M^{\co,\sigma}(\omega)
|
|
\\
|
|
\label{eq:C_M}
|
|
C_M^{\sigma}(\omega) = 2 \sigma \ERI{vc}{cv} - \ERI{vv}{cc} + f_M^{\co,\sigma}(\omega)
|
|
\end{gather}
|
|
\end{subequations}
|
|
yielding the excitation energies reported in Table \ref{tab:Maitra} when diagonalized.
|
|
The TDHF Hamiltonian is obtained from Eq.~\eqref{eq:H-M} by setting $f_M^{\co,\sigma}(\omega) = 0$ in Eqs.~\eqref{eq:R_M} and \eqref{eq:C_M}.
|
|
In Fig.~\ref{fig:Maitra}, we plot $\det[\bH(\omega) - \omega \bI]$ as a function of $\omega$ for both the singlet (black and gray) and triplet (orange) manifolds.
|
|
The roots of $\det[\bH(\omega) - \omega \bI]$ indicate the excitation energies.
|
|
Because, there is nothing to dress for the triplet state, only the static TDHF excitation energy is reported.
|
|
|
|
%%% TABLE I %%%
|
|
\begin{table}
|
|
\caption{Singlet and triplet excitation energies at various levels of theory.
|
|
\label{tab:Maitra}
|
|
}
|
|
\begin{ruledtabular}
|
|
\begin{tabular}{|c|cccc|c|}
|
|
Singlets & CIS & TDHF & D-CIS & D-TDHF & Exact \\
|
|
\hline
|
|
$\omega_1^{\updw}$ & 1.91119 & 1.89758 & 1.90636 & 1.89314 & 1.92145 \\
|
|
$\omega_3^{\updw}$ & & & & 3.44865 & 3.47880 \\
|
|
\hline
|
|
Triplets & & & & & Exact \\
|
|
\hline
|
|
$\omega_1^{\upup}$ & 1.45585 & 1.43794 & 1.45585 & 1.43794 & 1.47085 \\
|
|
\end{tabular}
|
|
\end{ruledtabular}
|
|
\end{table}
|
|
%%% %%% %%% %%%
|
|
|
|
%%% FIGURE 1 %%%
|
|
\begin{figure}
|
|
\includegraphics[width=\linewidth]{Maitra}
|
|
\caption{
|
|
$\det[\bH(\omega) - \omega \bI]$ as a function of $\omega$ for both the singlet (black and gray) and triplet (orange) manifolds.
|
|
The static TDHF Hamiltonian (dashed) and dynamic D-TDHF Hamiltonian (solid) are considered.
|
|
\label{fig:Maitra}
|
|
}
|
|
\end{figure}
|
|
%%% %%% %%% %%%
|
|
|
|
Although not particularly accurate for the single excitations, Maitra's dynamical kernel allows to access the double excitation with good accuracy and provides exactly the right number of solutions (two singlets and one triplet).
|
|
Note that the correlation kernel is known to work best in the weak correlation regime (which is the case here) where the true excitations have a clear single and double excitation character, \cite{Loos_2019,Loos_2020d} but it is not intended to explore strongly correlated systems. \cite{Carrascal_2018}
|
|
Its accuracy for the single excitations could be certainly improved in a DFT context.
|
|
However, this is not the point of the present investigation.
|
|
Table \ref{tab:Maitra} also reports the slightly-improves (thanks to error compensation) CIS and D-CIS excitation energies.
|
|
Graphically, the curves obtained for CIS and D-CIS are extremely similar to the ones of TDHF and D-TDHF depicted in Fig.~\ref{fig:Maitra}.
|
|
|
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
|
\subsection{Dynamical BSE kernel}
|
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
|
|
|
\titou{Titou will add a bit more background information about the BSE dynamical kernel.}
|
|
|
|
Within the so-called $GW$ approximation of MBPT, \cite{Golze_2019} one can easily compute the quasiparticle energies associated with the valence and conduction orbitals.
|
|
Assuming that the dynamically-screened Coulomb potential $W$ has been calculated at the random-phase approximation (RPA) level and within the TDA, the expression of the $\GW$ quasiparticle energy is
|
|
\begin{equation}
|
|
\e{p}^{\GW} = \e{p} + Z_{p}^{\GW} \SigGW{p}(\e{p})
|
|
\end{equation}
|
|
where $p = v$ or $c$,
|
|
\begin{equation}
|
|
\label{eq:SigC}
|
|
\SigGW{p}(\omega) = \frac{2 \ERI{pv}{vc}^2}{\omega - \e{v} + \Omega} + \frac{2 \ERI{pc}{cv}^2}{\omega - \e{c} - \Omega}
|
|
\end{equation}
|
|
is the correlation part of the self-energy $\Sig{}$, and
|
|
\begin{equation}
|
|
Z_{p}^{\GW} = \qty( 1 - \left. \pdv{\SigGW{p}(\omega)}{\omega} \right|_{\omega = \e{p}} )^{-1}
|
|
\end{equation}
|
|
is the renormalization factor.
|
|
In Eq.~\eqref{eq:SigC}, $\Omega = \Delta\eGW{} + 2 \ERI{vc}{vc}$ is the sole (singlet) RPA excitation energy of the system, with $\Delta\eGW{} = \eGW{c} - \eGW{v}$.
|
|
Numerically, we get
|
|
\begin{align}
|
|
\Omega & = 2.769\,327
|
|
&
|
|
\eGW{v} & = -0.863\,700
|
|
&
|
|
\eGW{c} & = +1.373\,640
|
|
\end{align}
|
|
|
|
One can now build the dynamical BSE (dBSE) Hamiltonian \cite{Strinati_1988}
|
|
\begin{equation} \label{eq:HBSE}
|
|
\bH_{\dBSE}^{\sigma}(\omega) =
|
|
\begin{pmatrix}
|
|
R_{\dBSE}^{\sigma}(\omega) & C_{\dBSE}^{\sigma}(\omega)
|
|
\\
|
|
-C_{\dBSE}^{\sigma}(-\omega) & -R_{\dBSE}^{\sigma}(-\omega)
|
|
\end{pmatrix}
|
|
\end{equation}
|
|
with
|
|
\begin{subequations}
|
|
\begin{gather}
|
|
R_{\dBSE}^{\sigma}(\omega) = \Delta\eGW{} + 2 \sigma \ERI{vc}{cv} - \ERI{vv}{cc} - W^{\co}_R(\omega)
|
|
\\
|
|
C_{\dBSE}^{\sigma}(\omega) = 2 \sigma \ERI{vc}{cv} - \ERI{vc}{cv} - W^{\co}_C(\omega)
|
|
\end{gather}
|
|
\end{subequations}
|
|
and
|
|
\begin{subequations}
|
|
\begin{gather}
|
|
W^{\co}_R(\omega) = \frac{4 \ERI{vv}{vc} \ERI{vc}{cc}}{\omega - \Omega - \Delta\eGW{}}
|
|
\\
|
|
W^{\co}_C(\omega) = \ERI{vc}{cv} + \frac{4 \ERI{vc}{cv}^2}{\omega - \Omega}
|
|
\end{gather}
|
|
\end{subequations}
|
|
are the elements of the correlation part of the dynamically-screened Coulomb potential for the resonant and coupling blocks of the dBSE Hamiltonian.
|
|
Note that, in this case, the correlation kernel is spin blind.
|
|
|
|
It can be easily shown that solving the equation
|
|
\begin{equation}
|
|
\det[\bH_{\dBSE}^{\sigma}(\omega) - \omega \bI] = 0
|
|
\end{equation}
|
|
yields 3 solutions per spin manifold (see Fig.~\ref{fig:dBSE}).
|
|
Their numerical values are reported in Table \ref{tab:dBSE} alongside other variants discussed below.
|
|
As mentioned in Ref.~\onlinecite{Romaniello_2009b}, spurious solutions appears due to the approximate nature of the dBSE kernel.
|
|
Indeed, diagonalizing the exact Hamiltonian produces only two singlet solutions corresponding to the singly- and doubly-excited states, and one triplet state (see Sec.~\ref{sec:exact}).
|
|
Therefore, there is one spurious solution for the singlet manifold ($\omega_{2}^{\dBSE,\updw}$) and two spurious solutions for the triplet manifold ($\omega_{2}^{\dBSE,\upup}$ and $\omega_{3}^{\dBSE,\upup}$).
|
|
It is worth mentioning that, around $\omega = \omega_1^{\dBSE,\sigma}$, the slope of the curves depicted in Fig.~\ref{fig:dBSE} is small, while the two other solutions, $\omega_2^{\dBSE,\sigma}$ and $\omega_3^{\dBSE,\sigma}$, stem from poles and consequently the slope is very large around these frequency values.
|
|
This evidences that dBSE reproduces qualitatively well the singlet and triplet single excitations, but quite badly the double excitation which is off by more than 1 hartree.
|
|
|
|
%%% TABLE I %%%
|
|
\begin{table*}
|
|
\caption{BSE singlet and triplet excitation energies at various levels of theory.
|
|
\label{tab:dBSE}
|
|
}
|
|
\begin{ruledtabular}
|
|
\begin{tabular}{|c|ccccccc|c|}
|
|
Singlets & BSE & BSE1 & BSE1(dTDA) & dBSE & BSE(TDA) & BSE1(TDA) & dBSE(TDA) & Exact \\
|
|
\hline
|
|
$\omega_1^{\updw}$ & 1.92778 & 1.90022 & 1.91554 & 1.90527 & 1.95137 & 1.94004 & 1.94005 & 1.92145 \\
|
|
$\omega_2^{\updw}$ & & & & 2.78377 & & & & \\
|
|
$\omega_3^{\updw}$ & & & & 4.90134 & & & 4.90117 & 3.47880 \\
|
|
\hline
|
|
Triplets & BSE & BSE1 & BSE1(dTDA) & dBSE & BSE(TDA) & BSE1(TDA) & dBSE(TDA) & Exact \\
|
|
\hline
|
|
$\omega_1^{\upup}$ & 1.48821 & 1.46860 & 1.46260 & 1.46636 & 1.49603 & 1.47070 & 1.47070 & 1.47085 \\
|
|
$\omega_2^{\upup}$ & & & & 2.76178 & & & & \\
|
|
$\omega_3^{\upup}$ & & & & 4.91545 & & & 4.91517 & \\
|
|
\end{tabular}
|
|
\end{ruledtabular}
|
|
\end{table*}
|
|
%%% %%% %%% %%%
|
|
|
|
%%% FIGURE 2 %%%
|
|
\begin{figure}
|
|
\includegraphics[width=\linewidth]{dBSE}
|
|
\caption{
|
|
$\det[\bH^{\dBSE}(\omega) - \omega \bI]$ as a function of $\omega$ for both the singlet (red) and triplet (blue) manifolds.
|
|
\label{fig:dBSE}
|
|
}
|
|
\end{figure}
|
|
%%% %%% %%% %%%
|
|
|
|
Within the usual static approximation, the BSE Hamiltonian is
|
|
\begin{equation}
|
|
\bH_{\BSE}^{\sigma} =
|
|
\begin{pmatrix}
|
|
R_{\BSE}^{\sigma} & C_{\BSE}^{\sigma}
|
|
\\
|
|
-C_{\BSE}^{\sigma} & -R_{\BSE}^{\sigma}
|
|
\end{pmatrix}
|
|
\end{equation}
|
|
with
|
|
\begin{subequations}
|
|
\begin{gather}
|
|
R_{\BSE}^{\sigma} = \Delta\eGW{} + 2 \sigma \ERI{vc}{vc} - W_R(\omega = \Delta\eGW{})
|
|
\\
|
|
C_{\BSE}^{\sigma} = 2 \sigma \ERI{vc}{vc} - W_C(\omega = 0)
|
|
\end{gather}
|
|
\end{subequations}
|
|
In the static approximation, only one solution per spin manifold is obtained by diagonalizing $\bH_{\BSE}^{\sigma}$ (see Fig.~\ref{fig:dBSE} and Table \ref{tab:dBSE}).
|
|
Therefore, the static BSE Hamiltonian does not produce spurious excitations but misses the (singlet) double excitation, and it shows that the physical single excitation stemming from the dBSE Hamiltonian is the lowest one for each spin manifold, \ie, $\omega_1^{\dBSE,\updw}$ and $\omega_1^{\dBSE,\upup}$.
|
|
|
|
Enforcing the TDA, which corresponds to neglecting the coupling term between the resonant and anti-resonant part of the dBSE Hamiltonian, allows to remove some of these spurious excitations.
|
|
There is thus only one spurious excitation in the triplet manifold ($\omega_{3}^{\BSE,\upup}$), the two solutions of the singlet manifold corresponding to the single and double excitations.
|
|
Figure \ref{fig:dBSE-TDA} shows the same curves as Fig.~\ref{fig:dBSE} but in the TDA.
|
|
|
|
%%% FIGURE 2 %%%
|
|
\begin{figure}
|
|
\includegraphics[width=\linewidth]{dBSE-TDA}
|
|
\caption{
|
|
$\det[\bH^{\TDAdBSE}(\omega) - \omega \bI]$ as a function of $\omega$ for both the singlet (red) and triplet (blue) manifolds within the TDA.
|
|
\label{fig:dBSE-TDA}
|
|
}
|
|
\end{figure}
|
|
%%% %%% %%% %%%
|
|
|
|
|
|
Another way to access dynamical effects while staying in the static framework is to use perturbation theory.
|
|
To do so, one must decompose the dBSE Hamiltonian into a (zeroth-order) static part and a dynamical perturbation, such that
|
|
\begin{equation}
|
|
\bH_{\dBSE}^{\sigma}(\omega) = \underbrace{\bH_{\BSE}^{\sigma}}_{\bH^{(0)}} + \underbrace{\qty[ \bH_{\dBSE}^{\sigma}(\omega) - \bH_{\BSE}^{\sigma} ]}_{\bH^{(1)}}
|
|
\end{equation}
|
|
Thanks to (renormalized) first-order perturbation theory, one gets
|
|
\begin{equation}
|
|
\begin{split}
|
|
\omega_{1}^{\BSE1,\sigma}
|
|
& = \omega_{1}^{\BSE,\sigma}
|
|
\\
|
|
& + Z_{1}
|
|
\T{\begin{pmatrix}
|
|
X_1 \\ Y_1
|
|
\end{pmatrix}
|
|
}
|
|
\cdot \qty[ \bH^{\dBSE}(\omega = \omega_{1}^{\BSE,\sigma}) - \bH^{\BSE} ] \cdot
|
|
\begin{pmatrix}
|
|
X_1 \\ Y_1
|
|
\end{pmatrix}
|
|
\end{split}
|
|
\end{equation}
|
|
where
|
|
\begin{equation}
|
|
\bH^{\BSE}
|
|
\cdot
|
|
\begin{pmatrix}
|
|
X_1 \\ Y_1
|
|
\end{pmatrix}
|
|
= \omega_{1}^{\BSE,\sigma}
|
|
\begin{pmatrix}
|
|
X_1 \\ Y_1
|
|
\end{pmatrix}
|
|
\end{equation}
|
|
are the eigenvectors of $\bH^{\BSE}$, and
|
|
\begin{equation}
|
|
Z_{1} = \qty{ 1 -
|
|
\T{
|
|
\begin{pmatrix}
|
|
X_1 \\ Y_1
|
|
\end{pmatrix}
|
|
}
|
|
\cdot \left. \pdv{\bH^{\dBSE}(\omega)}{\omega} \right|_{\omega = \omega_{1}^{\BSE,\sigma}} \cdot
|
|
\begin{pmatrix}
|
|
X_1 \\ Y_1
|
|
\end{pmatrix}
|
|
}^{-1}
|
|
\end{equation}
|
|
This corresponds to a dynamical correction to the static excitations, and the TDA can be applied to the dynamical correction, a scheme we label as dTDA in the following.
|
|
|
|
The perturbatively-corrected values are also reported in Table \ref{tab:dBSE}, which shows that this scheme is very efficient at reproducing the dynamical value.
|
|
Note that, although the BSE1(dTDA) value is further from the dBSE value than BSE1, it is quite close to the exact excitation energy.
|
|
|
|
|
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
|
\subsection{Second-order BSE kernel}
|
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
|
|
|
Here, we follow a different strategy and compute the dynamical second-order BSE kernel as illustrated by Yang and collaborators \cite{Zhang_2013}, and Rebolini and Toulouse \cite{Rebolini_2016}.
|
|
|
|
First, let us compute the second-order quasiparticle energies, which reads
|
|
\begin{equation}
|
|
\eGF{p} = \e{p} + Z_{p}^{\GF} \SigGF{p}(\e{p})
|
|
\end{equation}
|
|
where the second-order self-energy is
|
|
\begin{equation}
|
|
\label{eq:SigGF}
|
|
\SigGF{p}(\omega) = \frac{2 \ERI{pv}{vc}^2}{\omega - \e{v} + \e{c} - \e{v}} + \frac{2 \ERI{pc}{cv}^2}{\omega - \e{c} - (\e{c} - \e{v})}
|
|
\end{equation}
|
|
and
|
|
\begin{equation}
|
|
Z_{p}^{\GF} = \qty( 1 - \left. \pdv{\SigGF{p}(\omega)}{\omega} \right|_{\omega = \e{p}} )^{-1}
|
|
\end{equation}
|
|
This expression can be easily obtained in the present case by the substitution $\Omega \to \e{c} - \e{v}$ which transforms the $GW$ self-energy into its GF2 analog.
|
|
|
|
The static Hamiltonian for this theory is just the usual RPAx (or TDHF) Hamiltonian, \ie,
|
|
\begin{equation}
|
|
\bH^{\RPAx} =
|
|
\begin{pmatrix}
|
|
A^{\stat} & B^{\stat}
|
|
\\
|
|
-B^{\stat} & -A^{\stat}
|
|
\end{pmatrix}
|
|
\end{equation}
|
|
with
|
|
\begin{subequations}
|
|
\begin{gather}
|
|
A^{\stat} = \Delta\eGF{} + 2 \sigma \ERI{vc}{vc} - \ERI{vv}{cc}
|
|
\\
|
|
B^{\stat} = 2 \sigma \ERI{vc}{vc} - \ERI{vc}{cv}
|
|
\end{gather}
|
|
\end{subequations}
|
|
The dynamical part of the kernel for BSE2 (that we will call dRPAx for notational consistency) is a bit ugly but it simplifies greatly in the case of the present model to yield
|
|
\begin{equation}
|
|
\bH^{\dRPAx} = \bH^{\RPAx} +
|
|
\begin{pmatrix}
|
|
A(\omega) & B
|
|
\\
|
|
-B & -A(-\omega)
|
|
\end{pmatrix}
|
|
\end{equation}
|
|
with
|
|
\begin{subequations}
|
|
\begin{gather}
|
|
A^{\updw}(\omega) = - \frac{4 \ERI{cv}{vv} \ERI{vc}{cc} - \ERI{vc}{cc}^2 - \ERI{cv}{vv}^2 }{\omega - 2 \Delta\eGF{}}
|
|
\\
|
|
B^{\updw} = - \frac{4 \ERI{vc}{cv}^2 - \ERI{cc}{cc} \ERI{vc}{cv} - \ERI{vv}{vv} \ERI{vc}{cv} }{2 \Delta\eGF{}}
|
|
\end{gather}
|
|
\end{subequations}
|
|
and
|
|
\begin{subequations}
|
|
\begin{gather}
|
|
A^{\upup}(\omega) = - \frac{ \ERI{vc}{cc}^2 + \ERI{cv}{vv}^2 }{\omega - 2 \Delta\eGF{}}
|
|
\\
|
|
B^{\upup} = - \frac{\ERI{cc}{cc} \ERI{vc}{cv} + \ERI{vv}{vv} \ERI{vc}{cv} }{2 \Delta\eGF{}}
|
|
\end{gather}
|
|
\end{subequations}
|
|
Note that the coupling blocks $B$ are frequency independent, as they should.
|
|
This has an important consequence as this lack of frequency dependence removes one of the spurious pole.
|
|
The singlet manifold has then the right number of excitations.
|
|
However, one spurious triplet excitation remains (see Fig.~\ref{fig:dBSE2}).
|
|
Numerical results for the two-level model are reported in Table \ref{tab:RPAx} with the usual approximations and perturbative treatments.
|
|
In the case of dRPAx, the perturbative partitioning is simply
|
|
\begin{equation}
|
|
\bH^{\dRPAx}(\omega) = \underbrace{\bH^{\RPAx}}_{\bH^{(0)}} + \underbrace{\qty[ \bH^{\dRPAx}(\omega) - \bH^{\RPAx} ]}_{\bH^{(1)}}
|
|
\end{equation}
|
|
This might not be the smartest way of decomposing the Hamiltonian though but it seems to work fine.
|
|
|
|
%%% TABLE II %%%
|
|
\begin{table*}
|
|
\caption{RPAx singlet and triplet excitation energies at various levels of theory.
|
|
\label{tab:RPAx}
|
|
}
|
|
\begin{ruledtabular}
|
|
\begin{tabular}{|c|ccccccc|c|}
|
|
Singlets & RPAx & RPAx1 & RPAx1(dTDA) & dRPAx & RPAx(TDA) & RPAx1(TDA) & dRPAx(TDA) & Exact \\
|
|
\hline
|
|
$\omega_1$ & 1.84903 & 1.90940 & 1.90950 & 1.90362 & 1.86299 & 1.92356 & 1.92359 & 1.92145 \\
|
|
$\omega_2$ & & & & & & & & \\
|
|
$\omega_3$ & & & & 4.47124 & & & 4.47097 & 3.47880 \\
|
|
\hline
|
|
Triplets & RPAx & RPAx1 & RPAx1(dTDA) & dRPAx & RPAx(TDA) & RPAx1(TDA) & dRPAx(TDA) & Exact \\
|
|
\hline
|
|
$\omega_1$ & 1.38912 & 1.44285 & 1.44304 & 1.42564 & 1.40765 & 1.46154 & 1.46155 & 1.47085 \\
|
|
$\omega_2$ & & & & & & & & \\
|
|
$\omega_3$ & & & & 4.47797 & & & 4.47767 & \\
|
|
\end{tabular}
|
|
\end{ruledtabular}
|
|
\end{table*}
|
|
%%% %%% %%% %%%
|
|
|
|
%%% FIGURE 3 %%%
|
|
\begin{figure}
|
|
\includegraphics[width=\linewidth]{dBSE2}
|
|
\caption{
|
|
$\det[\bH^{\dRPAx}(\omega) - \omega \bI]$ as a function of $\omega$ for both the singlet (red) and triplet (blue) manifolds.
|
|
\label{fig:dBSE2}
|
|
}
|
|
\end{figure}
|
|
%%% %%% %%% %%%
|
|
|
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
|
\subsection{Sangalli's kernel}
|
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
|
\titou{This section is experimental...}
|
|
In Ref.~\onlinecite{Sangalli_2011}, Sangalli proposed a dynamical kernel (based on the second RPA) without (he claims) spurious excitations thanks to the design of a number-conserving approach which correctly describes particle indistinguishability and Pauli exclusion principle.
|
|
We will first start by writing down explicitly this kernel as it is given in obscure physicist notations.
|
|
|
|
The dynamical BSE Hamiltonian with Sangalli's kernel is
|
|
\begin{equation}
|
|
\bH^\text{NC}(\omega) =
|
|
\begin{pmatrix}
|
|
H(\omega) & K(\omega)
|
|
\\
|
|
-K(-\omega) & -H(-\omega)
|
|
\end{pmatrix}
|
|
\end{equation}
|
|
with
|
|
\begin{subequations}
|
|
\begin{gather}
|
|
H_{ia,jb}(\omega) = \delta_{ij} \delta_{ab} (\eGW{a} - \eGW{i}) + \Xi_{ia,jb} (\omega)
|
|
\\
|
|
K_{ia,jb}(\omega) = \Xi_{ia,bj} (\omega)
|
|
\end{gather}
|
|
\end{subequations}
|
|
and
|
|
\begin{subequations}
|
|
\begin{gather}
|
|
\Xi_{ia,jb} (\omega) = \sum_{m \neq n} \frac{ C_{ia,mn} C_{jb,mn} }{\omega - ( \omega_{m} + \omega_{n})}
|
|
\\
|
|
C_{ia,mn} = \frac{1}{2} \sum_{jb,kc} \qty{ \qty[ \ERI{ij}{kc} \delta_{ab} + \ERI{kc}{ab} \delta_{ij} ] \qty[ R_{m,jc} R_{n,kb}
|
|
+ R_{m,kb} R_{n,jc} ] }
|
|
\end{gather}
|
|
\end{subequations}
|
|
where $R_{m,ia}$ are the elements of the RPA eigenvectors.
|
|
|
|
For the two-level model, Sangalli's kernel reads
|
|
\begin{align}
|
|
H(\omega) & = \Delta\eGW{} + \Xi_H (\omega)
|
|
\\
|
|
K(\omega) & = \Xi_K (\omega)
|
|
\end{align}
|
|
|
|
\begin{gather}
|
|
\Xi_H (\omega) = 2 \frac{ [\ERI{vv}{vc} + \ERI{vc}{cc}]^2 }{\omega - 2\omega_1}
|
|
\\
|
|
\Xi_K (\omega) = 0
|
|
\end{gather}
|
|
|
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
|
\section{Take-home messages}
|
|
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
|
What have we learnt here?
|
|
|
|
%%%%%%%%%%%%%%%%%%%%%%%%
|
|
\acknowledgements{
|
|
%%%%%%%%%%%%%%%%%%%%%%%%
|
|
The author thanks the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation programme (Grant agreement No.~863481) for financial support.
|
|
|
|
|
|
% BIBLIOGRAPHY
|
|
\bibliography{../BSEdyn}
|
|
|
|
\end{document}
|