blob: a3b4d3196e2d0ede28dbcecc4535a9b449193fb0 (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
|
\subsection{The \texttt{LNSR:} module}
The {\tt LNSR:} module performs a single iteration of the line search algorithm, used as inner iteration of steepest descent,
conjugate gradient or quasi-Newton optimization techniques.\cite{recipie} The Armijo line search algorithm is used.\cite{armijo}
The {\tt LNSR:} module is generally used inside an inner loop, as depicted in Fig~\ref{fig:fig_lnsr}.
\vskip 0.2cm
\begin{figure}[h!]
\begin{center}
\includegraphics[scale=0.85]{Figures/lnsr.eps}
\caption{Line search iterations.}\label{fig:fig_lnsr}
\end{center}
\end{figure}
The calling specifications are:
\begin{DataStructure}{Structure \moc{LNSR:}}
\dusa{OPTIM} \moc{:=} \moc{LNSR:} \dusa{OPTIM} \moc{::} \dstr{lnsr\_data}
\end{DataStructure}
\noindent where
\begin{ListeDeDescription}{mmmmmmmm}
\item[\dusa{OPTIM}] \texttt{character*12} name of the \dds{optimize} object ({\tt L\_OPTIMIZE} signature) containing the
optimization informations. Object \dusa{OPTIM} must appear on both LHS and RHS to be able to update the previous values.
\item[\dstr{lnsr\_data}] structure containing the data to the module \texttt{LNSR:} (see Sect.~\ref{sect:lnsr_data}).
\end{ListeDeDescription}
\vskip 0.2cm
\goodbreak
\subsubsection{Data input for module \texttt{LNSR:}}\label{sect:lnsr_data}
\begin{DataStructure}{Structure \moc{lnsr\_data}}
$[$ \moc{EDIT} \dusa{iprint} $]$ \\
$[~\{$ \moc{MAXIMIZE} $|$ \moc{MINIMIZE} $\}~]$ \\
$[$ \moc{OUT-STEP-LIM} \dusa{sr} $]$ \\
$[$ \moc{OUT-STEP-EPS} \dusa{$\epsilon_{ext}$} $]~[$ \moc{INN-STEP-EPS} \dusa{$\epsilon_{inn}$} $]$ \\
\moc{OUT-ITER-MAX} \dusa{maxE} \\
$[$ \moc{OUT-RESTART} \dusa{nstart} $]$ \\
$[~\{$ \moc{SD} $|$ \moc{CG} $|$ \moc{BFGS} $|$ \moc{LBFGS} $[$ \moc{hist\_nr} $]~|$ \moc{NEWT} $\}~]$ \\
$[$ \moc{INN-CONV-TST} {\tt >>} \dusa{$l_{convI}$} {\tt <<} \moc{OUT-CONV-TST} {\tt >>} \dusa{$l_{convE}$} {\tt <<} $]$ \\
;
\end{DataStructure}
\noindent where
\begin{ListeDeDescription}{mmmmmmmm}
\item[\moc{EDIT}] keyword used to set \dusa{iprint}.
\item[\dusa{iprint}] index used to control the printing in module.
\item[\moc{MAXIMIZE}] keyword used to specify that the optimization problem will be a maximization.
\item[\moc{MINIMIZE}] keyword used to specify that the optimization problem will be a minimization (default).
\item[\moc{OUT-STEP-LIM}] keyword used to set or reset the maximum stepsize for the line search (default value is \dusa{sr} $=1.0$
or the value recovered from \dusa{OPTIM}).
\item[\dusa{sr}] initial radius of the stepsize (real or double precision).
\item[\moc{OUT-STEP-EPS}] keyword used to set the tolerance of outer iteration convergence inside module {\tt LNSR:} (default value
is $1.0 \times 10^{-4}$).
\item[\dusa{$\epsilon_{ext}$}] tolerance value (real or double precision).
\item[\moc{INN-STEP-EPS}] keyword used to set the tolerance used for the line search algorithm (default value
is $1.0 \times 10^{-4}$).
\item[\dusa{$\epsilon_{inn}$}] tolerance value (real or double precision).
\item[\moc{OUT-ITER-MAX}] keyword used to set the maximum number of external iterations.
\item[\dusa{maxE}] maximum number of external iterations.
\item[\moc{OUT-RESTART}] keyword used to set the external iteration restart cycle. A new recursion cycle
is restarted from the conjugate gradient method. By default, no restart is performed.
\item[\dusa{nstart}] number of iterations in one external iteration restart cycle.
\item[\moc{SD}] steepest descent method (default option).
\item[\moc{CG}] conjugate gradient method.
\item[\moc{BFGS}] Broyden-Fletcher-Goldfarb-Shanno method.\cite{recipie}
\item[\moc{LBFGS}] Memory limited Broyden-Fletcher-Goldfarb-Shanno method.\cite{nocedal}
\item[\dusa{hist\_nr}] number of corrections used in the memory limited Broyden-Fletcher-Goldfarb-Shanno method (default value
is 10).
\item[\moc{NEWT}] Newton method for unconstrained optimization.\cite{sph1981}
\item[\moc{INN-CONV-TST}] keyword used to determine if the line search convergence has been reached.
\item[\dusa{$l_{convI}$}] $=1$ means that line search convergence has been reached; $=0$ otherwise.
\item[\moc{OUT-CONV-TST}] keyword used to determine if the external convergence has been reached.
\item[\dusa{$l_{convI}$}] $=1$ means that external convergence has been reached; $=0$ otherwise.
\end{ListeDeDescription}
\clearpage
|
