this is for holding javascript data
Xavier Andrade edited Casida, Tamm-Dancoff, and excited-state forces.tex
over 9 years ago
Commit id: b0a9dea7492b4d99f5fa8815a31d2b51399c2324
deletions | additions
diff --git a/Casida, Tamm-Dancoff, and excited-state forces.tex b/Casida, Tamm-Dancoff, and excited-state forces.tex
index e14bb4d..793118a 100644
--- a/Casida, Tamm-Dancoff, and excited-state forces.tex
+++ b/Casida, Tamm-Dancoff, and excited-state forces.tex
...
\section{Linear response in the electron-hole basis}
An alternate approach to linear response is not to solve for the response function but rather for its
poles (the poles, the excitation energies
$\omega_k$) $\omega_k$, and electric dipole matrix elements $\vec{d}_k$. The polarizability is given by
\begin{align}
\alpha_{ij} \left( \omega \right) = \sum_k \left[ \frac{\left( \hat{i} \cdot \vec{d}_k \right)^{*} \left( \hat{j} \cdot \vec{d}_k \right)}{\omega_k - \omega - i \delta}
+ \frac{\left( \hat{i} \cdot \vec{d}_k \right)^{*} \left( \hat{j} \cdot \vec{d}_k \right)}{\omega_k + \omega + i \delta} \right]
...
% cite Petersilka
in which only the diagonal elements of the matrix are considered.
The eigenvectors are simply the KS transitions,
as like in
the RPA
(so case, as are the the dipole matrix
elements are the same as in RPA), elements,
and the positive eigenvalues are $\omega_{cv} = \epsilon_c - \epsilon_v + A_{cv}$.
This can be a reasonable approximation when there is little mixing between KS transitions,
but generally fails when there are degenerate or nearly degenerate transitions.
...
\left< \varphi_{c'} \varphi_{v'} \right| v \left| \varphi_c \varphi_v \right>
= \int \varphi_{c'} \left( r \right) \varphi_{v'} \left( r \right) P \left[ \varphi_c \varphi_v \right] dr
\end{align}
Our basic parallelization strategy for computation of the matrix elements is by domains, as
for Octopus discussed in
general, section~\ref{sec:parallelization},
but we
can add an additional level of
parallelization here over occupied-unoccupied pairs. We distribute the columns of the matrix, and do not distribute
the rows, to avoid duplication of Poisson solves. We can reduce the number of matrix elements to be computed by
almost half using the Hermitian nature of the matrix, \textit{i.e.} $M_{cv,c'v'} = M_{c'v',cv}^{*}$. If there are $N$