mirror of https://gitlab.com/QEF/q-e.git
Documentation updated (again)
git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@3174 c92efa57-630b-4861-b058-cf58834340f0
This commit is contained in:
parent
8a72497059
commit
54170ef641
19
Doc/INPUT_PW
19
Doc/INPUT_PW
|
@ -537,14 +537,17 @@ diagonalization
|
|||
'diis' : DIIS-like diagonalization - PRESENTLY DISABLED
|
||||
|
||||
diago_thr_init
|
||||
REAL ( default = 1.D-2 )
|
||||
Convergence threshold for the first iterative diagonalization.
|
||||
For scf calculations, the threshold (ethr) is automatically
|
||||
updated along the self-consistency loop.
|
||||
REAL
|
||||
Convergence threshold for the first iterative diagonalization
|
||||
(the check is on eigenvalue convergence).
|
||||
For scf calculations, the default is 1.D-2 if starting from a
|
||||
superposition of atomic orbitals; 1.D-5 if starting from a
|
||||
charge density. During self consistency the threshold (ethr)
|
||||
is automatically reduced when approaching convergence.
|
||||
For non-scf calculations, this is the threshold used in the
|
||||
iterative diagonalization
|
||||
For 'phonon' calculations, this is ignored: the threshold is
|
||||
set to conv_thr / nelec
|
||||
iterative diagonalization. The default is conv_thr / nelec.
|
||||
For 'phonon' calculations, diago_thr_init is ignored:
|
||||
the threshold is always set to conv_thr / nelec .
|
||||
|
||||
diago_cg_maxiter
|
||||
INTEGER
|
||||
|
@ -567,7 +570,7 @@ diago_full_acc
|
|||
If .TRUE. all the empty states are diagonalized at the same level
|
||||
of accuracy of the occupied ones. Otherwise the empty states are
|
||||
diagonalized using a larger threshold (this should not affect
|
||||
toltal energy, forces, and other ground-state properties).
|
||||
total energy, forces, and other ground-state properties).
|
||||
|
||||
efield REAL ( default = 0.D0 )
|
||||
For finite electric field calculations (lelfield == .TRUE.),
|
||||
|
|
|
@ -2663,35 +2663,38 @@ to perform a calculation with a DFT that differs from what used in
|
|||
PP generation, change the appropriate field in the PP file(s), at
|
||||
your own risk.
|
||||
|
||||
\paragraph{\texttt{pw.x} stops with error in cdiagh or cdiaghg.}
|
||||
\paragraph{\texttt{pw.x} stops with error in cdiaghg or rdiaghg.}
|
||||
|
||||
Possible reasons:
|
||||
Possible reasons for such behavior are not always clear, but they
|
||||
typically fall into one of the following cases:
|
||||
\begin{itemize}
|
||||
\item
|
||||
serious error in data, such as bad atomic positions or bad crystal
|
||||
structure/supercell;
|
||||
\item
|
||||
a bad PP (for instance, with a ghost);
|
||||
a bad PP, typicall with a ghost, but also a US-PP with non-positive
|
||||
charge density, leading to a violation of positiveness of the S
|
||||
matrix appearing in the US-PP formalism;
|
||||
\item
|
||||
a failure of the algorithm performing subspace diagonalization.
|
||||
The LAPACK algorithms used by cdiagh or cdiaghg are very robust
|
||||
The LAPACK algorithms used by cdiaghg/rdiaghg are very robust
|
||||
and extensively tested. Still, it may seldom happen that such
|
||||
algorithms fail. In at least one case the failures was tracked
|
||||
to the non-positiveness of the S matrix appearing in the US-PP
|
||||
formalism. In other cases, the error is found to be non reproducible
|
||||
on different architectures and disappearing if the calculation
|
||||
is repeated with even minimal changes in its parameters.
|
||||
In both cases, the reasons for such behavior are unclear and
|
||||
the only advice is to use conjugate-gradient diagonalization
|
||||
(\texttt{diagonalization='cg'}), a slower but very robust
|
||||
algorithms fail. Try to use conjugate-gradient diagonalization
|
||||
(\texttt{diagonalization='cg'}), a slower but very robust
|
||||
algorithm, and see what happens.
|
||||
\item
|
||||
HP-Compaq alphas with \texttt{cxml} libraries: try to use compiled
|
||||
BLAS and LAPACK (or better, ATLAS) instead of those contained in
|
||||
\texttt{cxml} (just load them before \texttt{cxml}).
|
||||
buggy libraries. Machine-optimized mathematical libraries are
|
||||
very fast but sometimes not so robust from a numerical point
|
||||
of view. Suspicious behavior: you get an error that is not
|
||||
reproducible on other architectures or that disappears if the
|
||||
calculation is repeated with even minimal changes in parameters.
|
||||
One known case: HP-Compaq alphas with \texttt{cxml} libraries.
|
||||
Try to use compiled BLAS and LAPACK (or better, ATLAS) instead of
|
||||
machine-optimized libraries.
|
||||
\end{itemize}
|
||||
|
||||
\paragraph{\texttt{pw.x} crashes with ``floating invalid'' or ``floating divide by zero''.}
|
||||
\paragraph{\texttt{pw.x} crashes with ``floating invalid'' or
|
||||
``floating divide by zero''.}
|
||||
|
||||
If this happens on HP-Compaq True64 Alpha machines with an old
|
||||
version of the compiler: the compiler is most likely buggy.
|
||||
|
|
Loading…
Reference in New Issue