3785
3785
calculations by means of the inclusion in the total Hamiltonian not
3786
3786
only the Darwin and velocity correction terms~(Scalar--Relativistic
3787
3787
calculations), but also the spin--orbit~(SO) contribution. There are
3788
two approximations regarding the SO formalism: on--site and off--site.
3789
Within the on--site only the intra--atomic SO contribution is taken
3790
into account whilst when using the off--site additional neigboring
3788
two approaches regarding the SO formalism: on--site and off--site.
3789
Within the on--site approximation only the intra--atomic SO contribution is taken
3790
into account. In the off--site scheme additional neighboring
3791
3791
interactions are also included in the SO term. By default, the off--site SO
3792
3792
formalism is switched on, being necessary to change the \fdf{Spin} flag
3793
3793
in the input file if the on--site approximation wants to be used. See
3794
3794
\fdf{Spin} on how to handle the spin--orbit coupling.
3796
The on--site implementation in \siesta\ has been performed by
3796
The on--site spin-orbit scheme in this version of \siesta\ has been implemented by
3797
3797
Dr. Ram\'on Cuadrado based on the original on--site SO formalism and
3798
implementation developed by Prof. Jaime Ferrer, \textit{et al}~(L
3798
implementation developed by Prof. Jaime Ferrer and his collaborators \textit{et al}~(L
3799
3799
Fern\'andez--Seivane, M Oliveira, S Sanvito, and J Ferrer, Journal of
3800
3800
Physics: Condensed Matter, {\bf 18}, 7999 (2006); L Fern\'andez--Seivane
3801
3801
and Jaime Ferrer, Phys. Rev. Lett. {\bf 99}, 183401 (2007)).
3803
The off--site implementation in \siesta\ has been performed by
3803
The off--site scheme has been implemented by
3804
3804
Dr. Ram\'on Cuadrado and Dr. Jorge I. Cerd\'a based on their initial
3805
3805
work~(R. Cuadrado and J. I. Cerd\'a ``Fully relativistic pseudopotential
3806
3806
formalism under an atomic orbital basis: spin-orbit splittings and
3810
3810
J. Klemmer and R. W. Chantrell, Applied Physics Letters, {\bf 108},
3811
3811
123102 (2016)).
3813
The inclusion of the SO term in the Hamiltonian~(and in the Density
3814
Matrix) will involve the increase of non--zero elements in their
3813
The inclusion of the SO term in the Hamiltonian (and in the Density
3814
Matrix) causes an increase in the number of non--zero elements in their
3815
3815
off--diagonal parts, i.e., for some $(\mu,\nu)$ pair of basis
3816
3816
orbitals, H$^{\sigma\sigma'}_{\mu\nu}$~(DM$^{\sigma\sigma'}_{\mu\nu}$)
3817
3817
[$\sigma,\sigma'$=$\uparrow,\downarrow$] will be $\neq$0. This is
3820
3820
In addition, these H$^{\sigma\sigma'}_{\mu\nu}$~(and
3821
3821
DM$^{\sigma\sigma'}_{\mu\nu}$) elements will be complex, in contrast
3822
3822
with typical polarized/non--polarized calculations where these
3823
matrices are purely real. Due to this, we encourage that previously to
3824
start with a full spin-orbit calculation it has to take special
3825
attention to the memory needed to perform it, based mainly in the size
3826
of the physical system and hence in the number of orbital
3827
involved. Note that for a non--SO calculation the memory will be
3828
around thirty percent lower than with SO.
3823
matrices are purely real. Since the spin-up and spin-down manifolds
3824
are essentially mixed, the solver has to deal with matrices whose
3825
dimensions are twice as large as for the collinear (unmixed) spin
3826
problem. Due to this, we advise to take special
3827
attention to the memory needed to perform a spin-orbit calculation.
3830
3830
Unless explicitly advised the following type of calculation can be carried out
3831
3831
regardless of whether on--site or off--site approximation is employed:
3834
3834
\item Selfconsistent calculations for gamma point as well as for
3837
\item Structure optimizations (only supported by the off--site SO
3837
\item Structure optimizations %% only supported by the off--site SO
3838
%% formalism *** Why ?
3840
\item LDA+U calculations~(See Sect.\ref{sec:lda+u} for further info).
3840
%%% *** Incompatible... \item LDA+U calculations~(See Sect.\ref{sec:lda+u} for further info).
3842
3842
\item Magnetic Anisotropy Energy~(MAE) can be easily
3843
calculated. From first principles it is obtained after subtract
3843
calculated. From first principles it is obtained after subtracting
3844
3844
the total selfconsistent energy calculated for two different
3845
3845
magnetic orientations. In \siesta\ it is possible to perform
3846
calculations when different initial magnetic ordering is necessary
3846
calculations with different initial magnetic orderings
3847
3847
by means of the use of the block \fdf{DM.InitSpin} in the fdf
3848
3848
file. In doing so one will be able to include the initial
3849
3849
orientation angles of the magnetization for each atom, as well as
3858
Note: Due to the small SO energy value contribution to the total
3858
Note: Due to the small SO contribution to the total
3859
3859
energy, the level of precision required to perform a proper fully
3860
3860
relativistic calculation during the selfconsistent process is quite
3861
3861
demanding. The following values must be carefully converged and
3863
3863
accurate enough: \fdf{SCF.H!Tolerance} during the
3864
3864
selfconsistency~(typically between 10$^{-3}$eV -- 10$^{-4}$eV),
3865
3865
\fdf{ElectronicTemperature},
3866
\textbf{k--point} sampling and high values of
3866
\textbf{k}--point sampling and high values of
3867
3867
\fdf{MeshCutoff}~(specifically for extended solids). In general, one
3868
can say that a good calculation will have high number of \textbf{k--points},
3868
can say that a good calculation will have high number of \textbf{k}--points,
3869
3869
low \fdf{ElectronicTemperature}, extremely small \fdf{SCF.H!Tolerance}
3870
3870
and high values of \fdf{MeshCutoff}. We encourage the user to test
3871
3871
carefully these options for each system. An additional point to take
3872
3872
into account is the mixing scheme employed. You are encouraged to use
3873
3873
\fdf{SCF.Mix} \fdf*{hamiltonian} (currently is set up by default)
3874
instead of the density matrix, since it speeds up the convergence.
3874
instead of density matrix mixing, since it speeds up the convergence.
3875
3875
The pseudopotentials have to be properly generated and tested for each
3876
specific system and they have to be in their fully relativistic form
3876
specific system and they have to be in their fully relativistic form,
3877
3877
together with the non--linear core corrections. Finally it is worth to
3878
mention that the selfconsistent convergence for some non--high symmetric
3878
mention that the selfconsistent convergence for some non--highly symmetric
3879
3879
magnetizations directions with respect to the physical symmetry axis
3880
could be coumbersome, however
3880
could still be difficult.
3882
3882
\begin{fdfentry}{Spin!OrbitStrength}[real]<1.0>
3884
3884
It allows to vary the strength of the
3885
3885
spin--orbit interaction from zero to any positive value. It can be
3886
used for both the on--site and off-site SOC flavors.
3886
used for both the on--site and off-site SOC flavors, but only for
3887
debugging and testing purposes, as the only physical value is 1.0.
3902
3903
\end{fdflogicalF}
3904
For the off--site SO approximation some mandatory flags have to set
3905
up to \fdffalse\ in the fdf file:
3907
\begin{fdflogicalT}{PAO!OldStylePolOrbs}
3909
By default is set up to \fdftrue, however it has to set up to \fdffalse.
3913
\begin{fdflogicalT}{DM.MixSCF1}
3915
By default is set up to \fdftrue, however it has to set up to \fdffalse.
3919
\begin{fdflogicalT}{Restricted.Radial.Grid}
3921
By default is set up to \fdftrue, however it has to set up to \fdffalse.
3905
% *** The following items should not be relevant.
3907
%%For the off--site SO approximation some mandatory flags have to set
3908
%%up to \fdffalse\ in the fdf file:
3910
%%\begin{fdflogicalT}{PAO!OldStylePolOrbs}
3912
%% By default is set up to \fdftrue, however it has to set up to \fdffalse.
3916
%%\begin{fdflogicalT}{DM.MixSCF1}
3918
%% By default is set up to \fdftrue, however it has to set up to \fdffalse.
3922
%%\begin{fdflogicalT}{Restricted.Radial.Grid}
3924
%% By default is set up to \fdftrue, however it has to set up to \fdffalse.
6309
6312
\begin{fdfentry}{OccupationFunction}[string]<FD>
6311
6314
String variable to select the function that determines the
6312
occupation of the electronic states. Two options are available:
6315
occupation of the electronic states. These options are available:
6313
6316
\begin{fdfoptions}
6315
6318
The usual Fermi-Dirac occupation function is used.
6318
6321
The occupation function proposed by Methfessel and
6319
6322
Paxton (Phys. Rev. B, \textbf{40}, 3616 (1989)), is used.
6325
The occupation function proposed by Marzari, Vanderbilt et. al
6326
(PRL, \textbf{82}, 16 (1999)), is used, this is commonly referred
6327
to as \emph{cold smearing}.
6321
6329
\end{fdfoptions}
6322
The smearing of the electronic occupations is done, in both cases,
6330
The smearing of the electronic occupations is done, in all cases,
6323
6331
using an energy width defined by the \fdf{ElectronicTemperature}
6324
6332
variable. Note that, while in the case of Fermi-Dirac, the
6325
6333
occupations correspond to the physical ones if the electronic
6329
6337
integration of the physical quantities at a lower cost. In
6330
6338
particular, the Methfessel-Paxton scheme has the advantage that,
6331
6339
even for quite large smearing temperatures, the obtained energy is
6332
very close to the physical energy at $T=0$. Also, it allows a much
6340
very close to the physical energy at $T=0$. Also, it allows a much
6333
6341
faster convergence with respect to $k$-points, specially for
6334
6342
metals. Finally, the convergence to selfconsistency is very much
6335
6343
improved (allowing the use of larger mixing coefficients).
6337
For the Methfessel-Paxton case, one can use relatively large values
6338
for the \fdf{ElectronicTemperature} parameter. How large depends
6339
on the specific system. A guide can be found in the article by
6340
J. Kresse and J. Furthm\"uller, Comp. Mat. Sci. \textbf{6}, 15
6345
For the Methfessel-Paxton case, and similarly for cold smearing, one
6346
can use relatively large values for the \fdf{ElectronicTemperature}
6347
parameter. How large depends on the specific system. A guide can be
6348
found in the article by J. Kresse and J. Furthm\"uller,
6349
Comp. Mat. Sci. \textbf{6}, 15 (1996).
6343
6351
If Methfessel-Paxton smearing is used, the order of the
6344
6352
corresponding Hermite polynomial expansion must also be chosen (see
6372
6380
\begin{fdfentry}{ElectronicTemperature}[temperature/energy]<$300\,\mathrm{K}$>
6374
Temperature for Fermi-Dirac or Methfessel-Paxton
6375
distribution. Useful specially for metals, and to accelerate
6376
selfconsistency in some cases.
6382
Temperature for occupation function. Useful specially for metals,
6383
and to accelerate selfconsistency in some cases.