857
877
The COOP curves are orbital, energy and $k$-resolved and they may
858
878
thus result in very large output files.
860
\note \fdf{TBT.DOS!Gf} has to be \fdftrue\ for this to be
861
calculated, otherwise a warning will be issued.
865
882
\end{fdflogicalF}
925
942
This generates \sysfile{TEIG\_<1>\_<2>} and
926
\sysfile{AVTEIG\_<1>\_<2>}, however, only for the non-equivalent
927
electrode combinations (\tbtrans\ intrinsically assumes
928
time-reversal symmetry).
943
\sysfile{AVTEIG\_<1>\_<2>}, possibly \sysfile{CEIG\_<1>} and
944
\sysfile{AVCEIG\_<1>}. The former is for two different electrodes
945
$i\neq j$, while the latter is for electrode $i=j$.
930
947
\note if you specify a number of eigenvalues above the available
931
948
number of eigenvalues, \tbtrans\ will automatically truncate it to a
932
949
reasonable number.
951
\note The transmission eigenvalues for $N>2$ systems is not fully
952
understood and the transmission eigenvalues calculated in \tbtrans\ is
953
done by diagonalizing this sub-matrix:
955
\mathbf G \boldsymbol \Gamma_i \mathbf G^\dagger \boldsymbol \Gamma_j.
936
960
\begin{fdflogicalF}{TBT.T!All}
999
1023
where we have left out the pre-factor ($e/\hbar$)
1000
1024
intentionally. \sisl\ may be used to analyze the orbital currents
1001
and enables easy transformation of orbital currents to
1002
bond currents and activity currents.
1025
and enables easy transformation of orbital currents to bond currents
1026
and activity currents\cite{Papior2017}.
1004
1028
\note this requires \tbtrans\ to be compiled with NetCDF-4 support,
1005
1029
see Sec.~\ref{sec:arch-make}.
1835
1859
flag controls the partitioning for the device region.
1862
\begin{fdflogicalF}{TBT!Analyze}
1864
As the pivoting algorithm \emph{highly} influences the performance
1865
and throughput of the transport calculation it is crucial to select
1866
the best performing algorithm available. This option tells \tbtrans\
1867
to analyze the pivoting table for nearly all the implemented
1868
algorithms and print-out information about them.
1870
\note we advice users to \emph{always} run an analyzation step prior
1871
to actual calculation and select the \emph{best} BTD format. This
1872
analyzing step is very fast and can be performed on small
1873
work-station computers, even on systems of $\gg10,000$ orbitals.
1875
To run the analyzing step you may do:
1876
\begin{shellexample}
1877
tbtrans -fdf TBT.Analyze RUN.fdf > analyze.out
1879
note that there is little gain on using MPI and it should complete
1880
within a few minutes, no matter the number of orbitals.
1882
Choosing the best one may be difficult. Generally one should choose
1883
the pivoting scheme that uses the least amount of memory. However,
1884
one should also choose the method with the largest block-size being
1885
as small as possible. As an example:
1886
\begin{shellexample}
1887
TBT.BTD.Pivot.Device atom+GPS
1890
[ 2984, 2776, 192, 192, 1639, 4050, 105 ]
1891
BTD matrix block size [max] / [average]: 4050 / 1705.429
1892
BTD matrix elements in % of full matrix: 47.88707 %
1894
TBT.BTD.Pivot.Device atom+GGPS
1897
[ 2880, 2916, 174, 174, 2884, 2910 ]
1898
BTD matrix block size [max] / [average]: 2916 / 1989.667
1899
BTD matrix elements in % of full matrix: 48.62867 %
1902
Although the GPS method uses the least amount of memory, the GGPS
1903
will likely perform better as the largest block in GPS is $4050$
1904
vs. $2916$ for the GGPS method.
1838
1909
\begin{fdfentry}{TBT.BTD!Pivot.Device}[string]<atom-\nonvalue{largest
1839
1910
overlapping electrode}>
1935
\begin{fdflogicalF}{TBT!Analyze}
1937
As the pivoting algorithm \emph{highly} influences the performance
1938
and throughput of the transport calculation it is crucial to select
1939
the best performing algorithm available. This option tells \tbtrans\
1940
to analyze the pivoting table for nearly all the implemented
1941
algorithms and print-out information about them.
1943
\note we advice users to \emph{always} run an analyzation step prior
1944
to actual calculation and select the \emph{best} BTD format. This
1945
analyzing step is very fast and may be performed on small
1946
work-station computers, even on systems of $\gg10,000$ orbitals.
1948
To run the analyzing step you may do:
1949
\begin{shellexample}
1950
tbtrans -fdf TBT.Analyze RUN.fdf > analyze.out
1952
note that there is little gain on using MPI and it should complete
1953
within a few minutes, no matter the number of orbitals.
1955
Choosing the best one may be difficult. Generally one should choose
1956
the pivoting scheme that uses the least amount of memory. However,
1957
one should also choose the method with the largest block-size being
1958
as small as possible. As an example:
1959
\begin{shellexample}
1960
TBT.BTD.Pivot.Device atom+GPS
1963
[ 2984, 2776, 192, 192, 1639, 4050, 105 ]
1964
BTD matrix block size [max] / [average]: 4050 / 1705.429
1965
BTD matrix elements in % of full matrix: 47.88707 %
1967
TBT.BTD.Pivot.Device atom+GGPS
1970
[ 2880, 2916, 174, 174, 2884, 2910 ]
1971
BTD matrix block size [max] / [average]: 2916 / 1989.667
1972
BTD matrix elements in % of full matrix: 48.62867 %
1975
Although the GPS method uses the least amount of memory, the GGPS
1976
will likely perform better as the largest block in GPS is $4050$
1977
vs. $2916$ for the GGPS method.
1982
2006
\begin{fdflogicalF}{TBT.BTD!Pivot.Graphviz}
1984
2008
Create Graphviz\footnote{\url{www.graphviz.org}} compatible input