This page documents the use of time-propagation TDDFT in :ref:`LCAO
mode <lcao>`. The implementation is described in [#Kuisma2015]_.
...
...
@@ -54,9 +54,9 @@ With the improved Hamiltonian, have functions are again propagated from t to t+d
This procedure is repeated using time step of 5-40as and for 500-2000 times to
obtain time evolution of electrons.
========================================
Time-propagation TDDFT with LCAO : Usage
========================================
=====
Usage
=====
Create LCAOTDDFT object like a GPAW calculator::
...
...
@@ -72,7 +72,7 @@ Important points are:
* Completely unoccupied bands should be left out of the calculation, since they are not needed.
* Convergence of density should be few orders of magnitude more accurate than in ground state calculations.
* Convergence of poisson solver should be at least 1e-14, but 1e-20 does not hurt (this is the quadratic error).
* One should use multipole corrected poisson solver in any TDDFT run. See PoissonSolver documentation about flags.
* One should use multipole corrected poisson solver in any TDDFT run. For more advanced poisson solvers, see :ref:`advancedpoisson`.
Perform a regular ground state calculation, the get the ground state wave functions::
...
...
@@ -158,111 +158,6 @@ LCAO-TDDFT is parallelized using ScaLAPACK. It runs without scalapack, but in th
* It is necessary that N*M equals to total number of cores and max(N,M)*64 < nbands. 64 can be changed to 16 if necessary.
* Apart from parallelization of linear algrebra, normal domain and band parallelizations can be used. As in LCAO-mode usually, use band parallelization to reduce memory consumption.
PoissonSolver
=============
The ``PoissonSolver`` with default parameters uses zero boundary conditions on
the cell boundaries. This becomes a problem in systems involving large dipole
moment, for example (due to, e.g., plasmonic charge oscillation on a
nanoparticle). The potential due to the dipole is long-ranged and, thus, the
converged potential requires large vacuum sizes.
However, in LCAO approach large vacuum size is often unnecessary. Thus, to avoid
using large vacuum sizes but get converged potential, one can use two approaches
or their combination: 1) use multipole moment corrections or 2) solve Poisson
equation on a extended grid. These two approaches are implemented in
``ExtendedPoissonSolver``. Also regular PoissonSolver in GPAW has the option
remove_moment.
In any nano-particle plasmonics calculation, it is necessary to use multipole
correction. Without corrections more than 10Å of vacuum is required for converged
results.
Multipole moment corrections
----------------------------
The boundary conditions can be improved by adding multipole moment corrections to
the density so that the corresponding multipoles of the density vanish. The
potential of these corrections is added to the obtained potential. For a
description of the method, see [#Castro2003]_.
This can be accomplished by following solver::
from gpaw.poisson_extended import ExtendedPoissonSolver
poissonsolver = ExtendedPoissonSolver(eps=eps,
moment_corrections=4)
This corrects the 4 first multipole moments, i.e., `s`, `p_x`, `p_y`, and `p_z` type multipoles. The range of
multipoles can be changed by changing ``moment_corrections`` parameter. For example, ``moment_correction=9`` includes in addition to the previous multipoles, also `d_{xx}`, `d_{xy}`,
`d_{yy}`, `d_{yz}`, and `d_{zz}` type multipoles.
This setting suffices usually for spherical-like metallic nanoparticles, but more
complex geometries require inclusion of very high multipoles or, alternatively, a
multicenter multipole approach. For this, consider the advanced syntax of the
moment_corrections. The previous code snippet is equivalent to::
from gpaw.poisson_extended import ExtendedPoissonSolver