By upfolding the frequency-dependent $GW$ quasiparticle equation, we explain the appearance of multiple solutions and unphysical discontinuities in various physical quantities computed within the $GW$ approximation.
By considering the $GW$ self-energy as an effective Hamiltonian, the appearance of these multiple solutions and discontinuities can be directly related to the intruder states problem.
A simple and efficient regularization procedure is proposed to avoid such issues.
Here, for the sake of simplicity, we consider a restricted Hartree-Fock (HF) starting point but the present analysis can be straightforwardly extended to a Kohn-Sham (KS) starting point.
Within the {\GOWO} approximation, in order to obtain the quasiparticle energies and the corresponding satellites, one solve, for each spatial orbital $p$, the following (non-linear) quasiparticle equation
Within the Tamm-Dancoff approximation, the screened two-electron integrals are given by
\begin{equation}
\ERI{pq}{m} = \sum_{ia}\ERI{pq}{ia} X_{ia,m}^\RPA
\end{equation}
where $\Om{m}{\RPA}$ and $\bX{m}{\RPA}$ are respectively the $m$th eigenvalue and eigenvector of the RPA problem in the Tamm-Dancoff approximation, \ie,
As a non-linear equation, Eq.~\eqref{eq:qp_eq} has many solutions $\eps{p,s}{}$ and their corresponding weight is given by the value of the so-called renormalization factor
The non-linear quasiparticle equation \eqref{eq:qp_eq} can be transformed into a larger linear problem via an upfolding process where the 2h1p and 2p1h sectors
The size of this eigenvalue problem is $N =1+ N^\text{2h1p}+ N^\text{2p1h}=1+ O^2 V + O V^2$, and this eigenvalue problem has to be solved for each orbital that one wants to correct.
Because the renormalization factor corresponds to the projection of the vector $\bc{}{(p,s)}$ onto the reference space, the weight of a solution $(p,s)$ is given by the the first coefficient of their corresponding eigenvector $\bc{}{(p,s)}$, \ie,
\begin{equation}
Z_{p,s} = \qty[ c_{1}^{(p,s)} ]^{2}
\end{equation}
It is important to understand that diagonalizing $\bH^{(p)}$ in Eq.~\eqref{eq:Hp} is completely equivalent to solving the quasiparticle equation \eqref{eq:qp_eq}.
The main difference between the two approaches is that, by diagonalizing Eq.~\eqref{eq:Hp}, one has directly access to the eigenvectors associated with each quasiparticle and satellites.
One can see this downfolding process as the construction of an frequency-dependent effective Hamiltonian where the reference (zeroth-order) space is composed by a single determinant of the 1h or 1p sector and the external (first-order) space by all the singly-excited states built from to this reference electronic configuration.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{An illustrative example}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Multiple solution issues in $GW$ appears all the time, especially for orbitals that are far in energy from the Fermi level.
Therefore, such issues are ubiquitous when one wants to compute core ionized states for example.
In order to illustrate the appearance and the origin of these multiple solutions, we consider the hydrogen molecule in the 6-31G basis set which corresponds to a system with 2 electrons and 4 spatial orbitals (one occupied and three virtual).
This example was already considered in our previous work \cite{Veril_2018} but here we provide further insights on the origin of the appearances of these multiple solutions.
Quasiparticle energies (left), correlation part of the self-energy (center) and renormalization factor (right) as functions of the internuclear distance $\RHH$ (in \si{\angstrom}) for various orbitals of \ce{H2} at the {\GOWO}@HF/6-31G level.
}
\end{figure*}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Figure \ref{fig:H2} shows the evolution of the quasiparticle energies at the {\GOWO}@HF/6-31G level as a function on the internuclear distance $\RHH$.
As one can see there are two problematic regions showing obvious discontinuities around $\RHH=\SI{1}{\AA}$ and $\RHH=\SI{1}{\AA}$
This project has received funding from the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation programme (Grant agreement No.~863481).}