Skip to content

GitLab

Commit 87ee3bc1 authored by Alberto Garcia's avatar Alberto Garcia Committed by Alberto
Browse files

Clarifications for SO and PEXSI in the manual

parent 68a19bd4
Hide whitespace changes
Inline Side-by-side
Showing with 101 additions and 85 deletions
Docs/siesta.bib
View file @ 87ee3bc1
......@@ -192,7 +192,22 @@ year = {2019}
@article{doi:10.1063/5.0005077,
author = {García,Alberto and Papior,Nick and Akhtar,Arsalan and Artacho,Emilio and Blum,Volker and Bosoni,Emanuele and Brandimarte,Pedro and Brandbyge,Mads and Cerdá,J. I. and Corsetti,Fabiano and Cuadrado,Ramón and Dikan,Vladimir and Ferrer,Jaime and Gale,Julian and García-Fernández,Pablo and García-Suárez,V. M. and García,Sandra and Huhs,Georg and Illera,Sergio and Korytár,Richard and Koval,Peter and Lebedeva,Irina and Lin,Lin and López-Tarifa,Pablo and Mayo,Sara G. and Mohr,Stephan and Ordejón,Pablo and Postnikov,Andrei and Pouillon,Yann and Pruneda,Miguel and Robles,Roberto and Sánchez-Portal,Daniel and Soler,Jose M. and Ullah,Rafi and Yu,Victor Wen-zhe and Junquera,Javier },
author = {Garc\'{\i}a,Alberto and Papior,Nick and Akhtar,Arsalan and
Artacho,Emilio and Blum,Volker and Bosoni,Emanuele
and Brandimarte,Pedro and Brandbyge,Mads and
Cerd\'a,J. I. and Corsetti,Fabiano and
Cuadrado,Ram\'on and Dikan,Vladimir and Ferrer,Jaime
and Gale,Julian and Garc\'{\i}a-Fern\'andez,Pablo
and Garc\'{\i}a-Su\'arez,V. M. and
Garc\'{\i}a,Sandra and Huhs,Georg and Illera,Sergio
and Koryt\'ar,Richard and Koval,Peter and
Lebedeva,Irina and Lin,Lin and L\'opez-Tarifa,Pablo
and Mayo,Sara G. and Mohr,Stephan and
Ordej\'on,Pablo and Postnikov,Andrei and
Pouillon,Yann and Pruneda,Miguel and Robles,Roberto
and S\'anchez-Portal,Daniel and Soler,Jose M. and
Ullah,Rafi and Yu,Victor Wen-zhe and Junquera,Javier
},
title = {Siesta: Recent developments and applications},
journal = {The Journal of Chemical Physics},
volume = {152},
......
Docs/siesta.tex
View file @ 87ee3bc1
......@@ -456,7 +456,7 @@ Extensive review of applications as of summer 2003.
\item
``Improvements on non-equilibrium and transport Green function techniques: The next-generation TranSIESTA'',
Nick Papior, Nicolas Lorente, Thomas Frederiksen, Alberto García and
Nick Papior, Nicolas Lorente, Thomas Frederiksen, Alberto Garc\'{\i}a and
Mads Brandbyge, Computer Physics Communications, \textbf{212}, 8--24 (2017).
Description of the \tsiesta\ method.
......@@ -470,7 +470,7 @@ Extensive review of applications as of summer 2003.
\item
``Siesta: Recent developments and applications'',
Alberto Garc\'\ia, \emph{et al.}, J. Chem. Phys. \textbf{152}, 204108
Alberto Garc\'{\i}a, \emph{et al.}, J. Chem. Phys. \textbf{152}, 204108
(2020).
Extensive review of applications and developments as of 2020.
......@@ -966,17 +966,22 @@ libraries
\item[\href{http://pexsi.org}{PEXSI}]%
\index{PEXSI}%
\index{External library!PEXSI}%
The PEXSI library may be used with \siesta\ for exa-scale
calculations, see Sec.~\ref{SolverPEXSI}. Currently the interface is
implemented (tested) as in PEXSI version 0.8.0, 0.9.0 and 0.9.2. If
newer versions retain the same interface they may also be used.
The PEXSI library may be used with this version of \siesta\ for massively-parallel
calculations, see Sec.~\ref{SolverPEXSI}. Note however that the
PEXSI interface in this version is the original one, corresponding
to PEXSI versions 0.8.X and 0.9.X. In particular, it has been tested
for 0.8.0, 0.9.0 and 0.9.2. It is possible that it might work for
newer versions of the form 0.9.X, but, beginning with version 1.0,
the PEXSI library is no longer compatible with this interface. Newer
versions of \siesta\ (in the Gitlab development site) can use
the current PEXSI library through the ELSI library interface.
To successfully compile \siesta\ with PEXSI support one require the
PEXSI fortran interface. When installing PEXSI copy the
\shell{f\_interface.f90} file to the include directory of
PEXSI such that the module may be found\footnote{Optionally the file
may be copied to the \shell{Obj} directory where the compilation
takes place.} when compiling \siesta.
takes place} when compiling \siesta.
Add these flags to your \file{arch.make} file to enable PEXSI
\begin{shellexample}
......@@ -3930,76 +3935,71 @@ are always stored in the file \sysfile{KP}.
\subsection{Spin--Orbit coupling}
\label{sec:spin-orbit}
\siesta\ includes the posibility to perform fully relativistic
calculations by means of the inclusion in the total Hamiltonian not
\siesta\ includes the possibility to perform fully relativistic
calculations by including in the total Hamiltonian not
only the Darwin and velocity correction terms~(Scalar--Relativistic
calculations), but also the spin-orbit~(SO) contribution. The
implementation is based on the on-site SO approximation, where only
the intra-SO contribution of each atom is taken into account. See
\fdf{Spin} on how to turn on the spin-orbit coupling.
The current implementation in \siesta\ has been implemented by
Dr. Ram\'on Cuadrado based on the original on-site SO formalism and
implementation developed by Prof. Jaime Ferrer, \textit{et al}~(L
calculations), but also the spin-orbit~(SO) contribution.
(See \fdf{Spin} for how to turn on the spin-orbit coupling.)
The SO implementation in this version is based on the ``on-site''
approximation, in which only the matrix elements corresponding to
orbitals on the same atom are taken into account for the SO part of
the Hamiltonian. The implementation has been coded by Dr. Ram\'on
Cuadrado based on the original on-site SO formalism and implementation
developed by Prof. Jaime Ferrer, \textit{et al}~(L
Fern\'andez--Seivane, M Oliveira, S Sanvito, and J Ferrer, Journal of
Physics: Condensed Matter, 2006 vol. 18 pp. 7999; L
Fern\'andez--Seivane and Jaime Ferrer, Phys. Rev. Lett. 99, 2007,
183401).
The inclusion of the SO term in the Hamiltonian~(and in the Density
Matrix) will involve the increase of non-zero elements in their
off-diagonal parts, i.e., for some $\mu\nu$ orbitals,
H$^{\sigma\sigma'}_{\mu\nu}$(DM$^{\sigma\sigma'}_{\mu\nu}$)
[$\sigma,\sigma'$=$\uparrow,\downarrow$] will be $\neq$0. This is
mainly due to the fact that the $\mathbf L\cdot\mathbf S$ operator
will promote the mixing between different spin-up/down components. The
terms responsible of this matrices expansion are the
exchange-correlation potential and the SO. The remaining terms such as
the kinetic energy or Hartree contribution do not depend of the spin
orientations and hence will be only added to the total
Hamiltonian~(and DM) to their diagonal parts.
The current SO formalism enables the possibility of several types of calculations:
\begin{itemize}
%
\item Selfconsistent calculations for gamma point as well as for
bulks~(Not yet implemented for optimizations).
%
\item Magnetic Anisotropy Energy~(MAE) can be easily calculated. From
first principles calculations, MAE is obtained after subtract the
total selfconsistent energy in two different orientations, usually the
total energy associated with easy axis from the hard axis. In \siesta\
it is possible to perform several self-consistent calculations for
different magnetization orientations using the specific block
\fdf{DM.InitSpin} in the fdf file. In doing so one will be able to
include the initial orientation angles of the magnetization for each
atom, as well as an initial value of their net magnetic moments.
%
\item By means of Mulliken analysis, after the self-consistent
procedure, local spin and orbital moments can be calculated by means
of the flags \fdf{WriteMullikenPop} and \fdf{WriteOrbMom}.
\end{itemize}
Note: Due to the small SO energy value contribution to the total
energy, the level of precision requiered to perform a proper fully
relativistic calculation during the selfconsistent process is quite
demanding. The following values must be carefully converged and
checked for each specific system to assure that the results are
accurate enough: \fdf{SCF.H!Tolerance} during the
selfconsistency~(typically <10$^{-5}$eV), \fdf{ElectronicTemperature},
\textbf{k}-point sampling and high values of
\fdf{Mesh!Cutoff}~(specifically for extended solids). In general, one
can say that a good calculation will have high number of k--points,
low \fdf{ElectronicTemperature}, extremely small \fdf{SCF.H!Tolerance}
and high values of \fdf{Mesh!Cutoff}. We encourage the user to test
carefully these options for each system. An additional point to take
into account when the spin--orbit contribution is included is the
mixing scheme employed. You are encouraged to use \fdf{SCF.Mix}
\fdf*{hamiltonian} instead of the density matrix, due to the fact that
the convergence speed increases considerably for the first case. In
addition, the pseudopotentials have to be well generated and tested
for each specific system and they have to be generated in their fully
relativistic form and use the non-linear core corrections.
183401). It should be noted that this approximation, while based on
the physically reasonable idea of the short-range of the SO
interaction, might not be completely appropriate in all cases. Newer
versions of \siesta\ (in the Gitlab development site) implement a
spin--orbit--coupling formalism that does not involve the ``on-site''
approximation and keeps all the SO matrix
elements~\cite{doi:10.1063/5.0005077}. Users might want to use the
latter scheme to check the soundness of the ``on-site'' approximation
for the system of interest. In fact, since the full scheme is not more
expensive than the approximate one, we would recommend the use of the
newer approach, unless there is a concern about employing an as-yet
unreleased version of the code.
The inclusion of the SO term in the Hamiltonian means that the
two components of the spin are coupled, and the calculation is of the
``non-collinear'' type with a doubling of the size of the matrices involved
(as opposed to the ``collinear'' spin case in which two different spin
blocks could be treated independently). Hence a SO calculation using
cube-scaling diagonalization would typically be four-times more
expensive ($2^3/2$) than a collinear-spin one.
Apart from the study of effects of the spin--orbit interaction in the
band structure, a feature enabled by a SO formalism is the computation
of the Magnetic Anisotropy Energy~(MAE): it can be obtained as the
difference in the total selfconsistent energy in two different spin
orientations, usually along the easy axis and the hard axis. In
\siesta\ it is possible to perform calculations for different
magnetization orientations using the block \fdf{DM.InitSpin} in the
fdf file. In doing so one will be able to include the initial
orientation angles of the magnetization for each atom, as well as an
initial value of their net magnetic moments.
Note: Due to the small contribution of the spin--orbit interaction to
the total energy, the level of precision required is quite high. The
following parameters should be carefully checked for each
specific system to assure that the results are converged and accurate enough:
\fdf{SCF.H!Tolerance} during the selfconsistency~(typically
<10$^{-5}$eV), \fdf{ElectronicTemperature}, \textbf{k}-point sampling,
and \fdf{Mesh!Cutoff}~(specifically for extended
solids). In general, one can say that a good calculation will have
a high number of k--points, low \fdf{ElectronicTemperature}, very
small \fdf{SCF.H!Tolerance} and high values of \fdf{Mesh!Cutoff}. We
encourage the user to test carefully these options for each system. An
additional point to take into account when the spin--orbit
contribution is included is the mixing scheme to use. You are
encouraged to use the option to mix the Hamiltonian (\fdf{SCF.Mix} \fdf*{hamiltonian}) instead of the
density matrix to speed up convergence. In addition, the pseudopotentials
have to be well tested for each specific system. They
have to be generated in their fully relativistic form, and should use
non-linear core corrections.
\begin{fdfentry}{Spin!OrbitStrength}[real]<1.0>
......@@ -4062,7 +4062,7 @@ The current version introduces several changes:
value of the changes in the entries of H, but its actual meaning
depends on whether DM or H mixing is in effect: if mixing the DM,
dHmax refers to the change in H(in) with respect to the previous
step; if mixing H, dHmax refers to H(out)-H(in) in the previous(?)
step; if mixing H, dHmax refers to H(out)-H(in) in the current
step.
\item When achieving convergence, the loop might be exited without a
......@@ -6349,10 +6349,6 @@ system sizes) would not benefit from the O(N) or PEXSI solvers.
support in the code is not fool-proof, so this flag should only be
enabled if GPU support is indeed available.
At present, ELPA offers GPU support in the 'elpa-1' solver in
released versions, whereas GPU support for the 'elpa-2' solver has
not been released officially.
\end{fdflogicalF}
\begin{fdflogicalF}{Diag!ParallelOverK}
......@@ -6958,17 +6954,22 @@ large systems has a much lower computational complexity than that
associated with the matrix diagonalization procedure. It is also
highly scalable.
The PEXSI technique can be used to evaluate the electron density, free
The PEXSI technique can be used in this version of \siesta\ to evaluate the electron density, free
energy, atomic forces, density of states and local density of states
without computing any eigenvalue or eigenvector of the Kohn-Sham
Hamiltonian. It can achieve accuracy fully comparable to that obtained
from a matrix diagonalization procedure for general systems, including
metallic systems at low temperature.
The current implementation of the PEXSI solver in \siesta\ makes
use of the full fine-grained-level interface in the PEXSI library
(\url{http://pexsi.org}), and can deal with spin-polarization, but it
is still restricted to $\Gamma$-point calculations.
The current implementation of the PEXSI solver in \siesta\ makes use
of a full fine-grained-level interface to earlier versions (0.8.X and
0.9.X) of the PEXSI library (\url{http://pexsi.org}), and can deal
with spin-polarization, but it is still restricted to $\Gamma$-point
calculations. Newer versions of \siesta\ (in the Gitlab development
site) can use the current PEXSI library through the ELSI library
interface, which offers some more options, although not currently the
density-of-states calculation.
The following is a brief description of the input-file parameters
relevant to the workings of the PEXSI solver. For more background,
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment