| 10 |
software tool is used that parses and transforms source files |
software tool is used that parses and transforms source files |
| 11 |
according to a set of linguistic and mathematical rules. AD tools are |
according to a set of linguistic and mathematical rules. AD tools are |
| 12 |
like source-to-source translators in that they parse a program code as |
like source-to-source translators in that they parse a program code as |
| 13 |
input and produce a new program code as output. However, unlike a |
input and produce a new program code as output |
| 14 |
|
(we restrict our discussion to source-to-source tools, ignoring |
| 15 |
|
operator-overloading tools). However, unlike a |
| 16 |
pure source-to-source translation, the output program represents a new |
pure source-to-source translation, the output program represents a new |
| 17 |
algorithm, such as the evaluation of the Jacobian, the Hessian, or |
algorithm, such as the evaluation of the Jacobian, the Hessian, or |
| 18 |
higher derivative operators. In principle, a variety of derived |
higher derivative operators. In principle, a variety of derived |
| 23 |
Algorithms in Fortran), developed by Ralf Giering (\cite{gie-kam:98}, |
Algorithms in Fortran), developed by Ralf Giering (\cite{gie-kam:98}, |
| 24 |
\cite{gie:99,gie:00}). The first application of the adjoint of MITgcm |
\cite{gie:99,gie:00}). The first application of the adjoint of MITgcm |
| 25 |
for sensitivity studies has been published by \cite{maro-eta:99}. |
for sensitivity studies has been published by \cite{maro-eta:99}. |
| 26 |
\cite{sta-eta:97,sta-eta:01} use MITgcm and its adjoint for ocean |
\cite{stam-etal:97,stam-etal:02} use MITgcm and its adjoint for ocean |
| 27 |
state estimation studies. In the following we shall refer to TAMC and |
state estimation studies. In the following we shall refer to TAMC and |
| 28 |
TAF synonymously, except were explicitly stated otherwise. |
TAF synonymously, except were explicitly stated otherwise. |
| 29 |
|
|
| 30 |
TAMC exploits the chain rule for computing the first derivative of a |
As of mid-2007 we are also able to generate fairly efficient |
| 31 |
|
adjoint code of the MITgcm using a new, open-source AD tool, |
| 32 |
|
called OpenAD (see \cite{naum-etal:06,utke-etal:08}. |
| 33 |
|
This enables us for the first time to compare adjoint models |
| 34 |
|
generated from different AD tools, providing an additional |
| 35 |
|
accuracy check, complementary to finite-difference gradient checks. |
| 36 |
|
OpenAD and its application to MITgcm is described in detail |
| 37 |
|
in section \ref{sec_ad_openad}. |
| 38 |
|
|
| 39 |
|
The AD tool exploits the chain rule for computing the first derivative of a |
| 40 |
function with respect to a set of input variables. Treating a given |
function with respect to a set of input variables. Treating a given |
| 41 |
forward code as a composition of operations -- each line representing |
forward code as a composition of operations -- each line representing |
| 42 |
a compositional element, the chain rule is rigorously applied to the |
a compositional element, the chain rule is rigorously applied to the |
| 65 |
under consideration, |
under consideration, |
| 66 |
% |
% |
| 67 |
\begin{equation} |
\begin{equation} |
| 68 |
\begin{split} |
\begin{aligned} |
| 69 |
{\cal M} \, : & \, U \,\, \longrightarrow \, V \\ |
{\cal M} \, : & \, U \,\, \longrightarrow \, V \\ |
| 70 |
~ & \, \vec{u} \,\, \longmapsto \, \vec{v} \, = \, |
~ & \, \vec{u} \,\, \longmapsto \, \vec{v} \, = \, |
| 71 |
{\cal M}(\vec{u}) |
{\cal M}(\vec{u}) |
| 72 |
\label{fulloperator} |
\label{fulloperator} |
| 73 |
\end{split} |
\end{aligned} |
| 74 |
\end{equation} |
\end{equation} |
| 75 |
% |
% |
| 76 |
The vectors $ \vec{u} \in U $ and $ v \in V $ may be represented w.r.t. |
The vectors $ \vec{u} \in U $ and $ v \in V $ may be represented w.r.t. |
| 150 |
$\left\langle \,\, , \,\, \right\rangle $ |
$\left\langle \,\, , \,\, \right\rangle $ |
| 151 |
% |
% |
| 152 |
\begin{equation} |
\begin{equation} |
| 153 |
\begin{split} |
\begin{aligned} |
| 154 |
{\cal J} & = \, |
{\cal J} & = \, |
| 155 |
{\cal J} |_{\vec{u}^{(0)}} \, + \, |
{\cal J} |_{\vec{u}^{(0)}} \, + \, |
| 156 |
\left\langle \, \nabla _{u}{\cal J}^T |_{\vec{u}^{(0)}} \, , \, \delta \vec{u} \, \right\rangle |
\left\langle \, \nabla _{u}{\cal J}^T |_{\vec{u}^{(0)}} \, , \, \delta \vec{u} \, \right\rangle |
| 159 |
{\cal J} |_{\vec{v}^{(0)}} \, + \, |
{\cal J} |_{\vec{v}^{(0)}} \, + \, |
| 160 |
\left\langle \, \nabla _{v}{\cal J}^T |_{\vec{v}^{(0)}} \, , \, \delta \vec{v} \, \right\rangle |
\left\langle \, \nabla _{v}{\cal J}^T |_{\vec{v}^{(0)}} \, , \, \delta \vec{v} \, \right\rangle |
| 161 |
\, + \, O(\delta \vec{v}^2) |
\, + \, O(\delta \vec{v}^2) |
| 162 |
\end{split} |
\end{aligned} |
| 163 |
\label{deljidentity} |
\label{deljidentity} |
| 164 |
\end{equation} |
\end{equation} |
| 165 |
% |
% |
| 200 |
invoking the adjoint $ M^{\ast } $ of the tangent linear model $ M $ |
invoking the adjoint $ M^{\ast } $ of the tangent linear model $ M $ |
| 201 |
% |
% |
| 202 |
\begin{equation} |
\begin{equation} |
| 203 |
\begin{split} |
\begin{aligned} |
| 204 |
\nabla _{u}{\cal J}^T |_{\vec{u}} & |
\nabla _{u}{\cal J}^T |_{\vec{u}} & |
| 205 |
= \, M^T |_{\vec{u}} \cdot \nabla _{v}{\cal J}^T |_{\vec{v}} \\ |
= \, M^T |_{\vec{u}} \cdot \nabla _{v}{\cal J}^T |_{\vec{v}} \\ |
| 206 |
~ & = \, M^T |_{\vec{u}} \cdot \delta \vec{v}^{\ast} \\ |
~ & = \, M^T |_{\vec{u}} \cdot \delta \vec{v}^{\ast} \\ |
| 207 |
~ & = \, \delta \vec{u}^{\ast} |
~ & = \, \delta \vec{u}^{\ast} |
| 208 |
\end{split} |
\end{aligned} |
| 209 |
\label{adjoint} |
\label{adjoint} |
| 210 |
\end{equation} |
\end{equation} |
| 211 |
% |
% |
| 253 |
= \nabla_v {\cal J} \cdot \delta \vec{v} $ ) |
= \nabla_v {\cal J} \cdot \delta \vec{v} $ ) |
| 254 |
% |
% |
| 255 |
\begin{equation} |
\begin{equation} |
| 256 |
\begin{split} |
\begin{aligned} |
| 257 |
\nabla_v {\cal J} (M(\delta \vec{u})) & = \, |
\nabla_v {\cal J} (M(\delta \vec{u})) & = \, |
| 258 |
\nabla_v {\cal J} \cdot M_{\Lambda} |
\nabla_v {\cal J} \cdot M_{\Lambda} |
| 259 |
\cdot ...... \cdot M_{\lambda} \cdot ...... \cdot |
\cdot ...... \cdot M_{\lambda} \cdot ...... \cdot |
| 260 |
M_{1} \cdot M_{0} \cdot \delta \vec{u} \\ |
M_{1} \cdot M_{0} \cdot \delta \vec{u} \\ |
| 261 |
~ & = \, \nabla_v {\cal J} \cdot \delta \vec{v} \\ |
~ & = \, \nabla_v {\cal J} \cdot \delta \vec{v} \\ |
| 262 |
\end{split} |
\end{aligned} |
| 263 |
\label{forward} |
\label{forward} |
| 264 |
\end{equation} |
\end{equation} |
| 265 |
% |
% |
| 267 |
% |
% |
| 268 |
\begin{equation} |
\begin{equation} |
| 269 |
\boxed{ |
\boxed{ |
| 270 |
\begin{split} |
\begin{aligned} |
| 271 |
M^T ( \nabla_v {\cal J}^T) & = \, |
M^T ( \nabla_v {\cal J}^T) & = \, |
| 272 |
M_{0}^T \cdot M_{1}^T |
M_{0}^T \cdot M_{1}^T |
| 273 |
\cdot ...... \cdot M_{\lambda}^T \cdot ...... \cdot |
\cdot ...... \cdot M_{\lambda}^T \cdot ...... \cdot |
| 276 |
\cdot ...... \cdot |
\cdot ...... \cdot |
| 277 |
\nabla_{v^{(\lambda)}} {\cal J}^T \\ |
\nabla_{v^{(\lambda)}} {\cal J}^T \\ |
| 278 |
~ & = \, \nabla_u {\cal J}^T |
~ & = \, \nabla_u {\cal J}^T |
| 279 |
\end{split} |
\end{aligned} |
| 280 |
} |
} |
| 281 |
\label{reverse} |
\label{reverse} |
| 282 |
\end{equation} |
\end{equation} |
| 295 |
% |
% |
| 296 |
\begin{equation} |
\begin{equation} |
| 297 |
\boxed{ |
\boxed{ |
| 298 |
\begin{split} |
\begin{aligned} |
| 299 |
\nabla_{v^{(\lambda)}} {\cal J}^T |_{\vec{v}^{(\lambda)}} |
\nabla_{v^{(\lambda)}} {\cal J}^T |_{\vec{v}^{(\lambda)}} |
| 300 |
& = \, |
& = \, |
| 301 |
M_{\lambda}^T |_{\vec{v}^{(\lambda)}} \cdot ...... \cdot |
M_{\lambda}^T |_{\vec{v}^{(\lambda)}} \cdot ...... \cdot |
| 302 |
M_{\Lambda}^T |_{\vec{v}^{(\lambda)}} \cdot \delta \vec{v}^{\ast} \\ |
M_{\Lambda}^T |_{\vec{v}^{(\lambda)}} \cdot \delta \vec{v}^{\ast} \\ |
| 303 |
~ & = \, \delta \vec{v}^{(\lambda) \, \ast} |
~ & = \, \delta \vec{v}^{(\lambda) \, \ast} |
| 304 |
\end{split} |
\end{aligned} |
| 305 |
} |
} |
| 306 |
\end{equation} |
\end{equation} |
| 307 |
% |
% |
| 418 |
$ \delta v^{(\lambda) \, \ast}_{j} = \frac{\partial}{\partial v^{(\lambda)}_{j}} |
$ \delta v^{(\lambda) \, \ast}_{j} = \frac{\partial}{\partial v^{(\lambda)}_{j}} |
| 419 |
{\cal J}^T $, $ j = 1, \ldots , n_{\lambda} $, |
{\cal J}^T $, $ j = 1, \ldots , n_{\lambda} $, |
| 420 |
for intermediate components, yielding |
for intermediate components, yielding |
| 421 |
|
{\small |
| 422 |
\begin{equation} |
\begin{equation} |
| 423 |
\small |
\begin{aligned} |
|
\begin{split} |
|
| 424 |
\left( |
\left( |
| 425 |
\begin{array}{c} |
\begin{array}{c} |
| 426 |
\delta v^{(\lambda) \, \ast}_1 \\ |
\delta v^{(\lambda) \, \ast}_1 \\ |
| 465 |
\delta v^{\ast}_{n} \\ |
\delta v^{\ast}_{n} \\ |
| 466 |
\end{array} |
\end{array} |
| 467 |
\right) |
\right) |
| 468 |
\end{split} |
\end{aligned} |
| 469 |
\end{equation} |
\end{equation} |
| 470 |
|
} |
| 471 |
|
|
| 472 |
Eq. (\ref{forward}) and (\ref{reverse}) are perhaps clearest in |
Eq. (\ref{forward}) and (\ref{reverse}) are perhaps clearest in |
| 473 |
showing the advantage of the reverse over the forward mode |
showing the advantage of the reverse over the forward mode |
| 538 |
Then, $ \nabla_v {\cal J} $ takes the form |
Then, $ \nabla_v {\cal J} $ takes the form |
| 539 |
% |
% |
| 540 |
\begin{equation*} |
\begin{equation*} |
| 541 |
\begin{split} |
\begin{aligned} |
| 542 |
\nabla_v {\cal J}^T & = \, 2 \, \, H \cdot |
\nabla_v {\cal J}^T & = \, 2 \, \, H \cdot |
| 543 |
\left( \, {\cal H}(\vec{v}) - \vec{d} \, \right) \\ |
\left( \, {\cal H}(\vec{v}) - \vec{d} \, \right) \\ |
| 544 |
~ & = \, 2 \sum_{j} \left\{ \sum_k |
~ & = \, 2 \sum_{j} \left\{ \sum_k |
| 545 |
\frac{\partial {\cal H}_k}{\partial v_{j}} |
\frac{\partial {\cal H}_k}{\partial v_{j}} |
| 546 |
\left( {\cal H}_k (\vec{v}) - d_k \right) |
\left( {\cal H}_k (\vec{v}) - d_k \right) |
| 547 |
\right\} \, {\vec{f}_{j}} \\ |
\right\} \, {\vec{f}_{j}} \\ |
| 548 |
\end{split} |
\end{aligned} |
| 549 |
\end{equation*} |
\end{equation*} |
| 550 |
% |
% |
| 551 |
where $H_{kj} = \partial {\cal H}_k / \partial v_{j} $ is the |
where $H_{kj} = \partial {\cal H}_k / \partial v_{j} $ is the |
| 664 |
%\psfrag{v_kn^lev2}{\mathinfigure{v_{k_{n}^{lev2}}}} |
%\psfrag{v_kn^lev2}{\mathinfigure{v_{k_{n}^{lev2}}}} |
| 665 |
%\psfrag{v_k1^lev1}{\mathinfigure{v_{k_{1}^{lev1}}}} |
%\psfrag{v_k1^lev1}{\mathinfigure{v_{k_{1}^{lev1}}}} |
| 666 |
%\psfrag{v_kn^lev1}{\mathinfigure{v_{k_{n}^{lev1}}}} |
%\psfrag{v_kn^lev1}{\mathinfigure{v_{k_{n}^{lev1}}}} |
| 667 |
%\mbox{\epsfig{file=part5/checkpointing.eps, width=0.8\textwidth}} |
%\mbox{\epsfig{file=s_autodiff/figs/checkpointing.eps, width=0.8\textwidth}} |
| 668 |
\resizebox{5.5in}{!}{\includegraphics{part5/checkpointing.eps}} |
\resizebox{5.5in}{!}{\includegraphics{s_autodiff/figs/checkpointing.eps}} |
| 669 |
%\psfull |
%\psfull |
| 670 |
\end{center} |
\end{center} |
| 671 |
\caption{ |
\caption{ |
| 693 |
|
|
| 694 |
In this section we describe in a general fashion |
In this section we describe in a general fashion |
| 695 |
the parts of the code that are relevant for automatic |
the parts of the code that are relevant for automatic |
| 696 |
differentiation using the software tool TAF. |
differentiation using the software tool TAF. |
| 697 |
|
Modifications to use OpenAD are described in \ref{sec_ad_openad}. |
| 698 |
|
|
| 699 |
\input{part5/doc_ad_the_model} |
\input{s_autodiff/text/doc_ad_the_model} |
| 700 |
|
|
| 701 |
The basic flow is depicted in \ref{fig:adthemodel}. |
The basic flow is depicted in \ref{fig:adthemodel}. |
| 702 |
If CPP option \texttt{ALLOW\_AUTODIFF\_TAMC} is defined, |
If CPP option \texttt{ALLOW\_AUTODIFF\_TAMC} is defined, |
| 731 |
{\tt ALLOW\_GRADIENT\_CHECK} is defined. In this case |
{\tt ALLOW\_GRADIENT\_CHECK} is defined. In this case |
| 732 |
the driver routine {\it grdchk\_main} is called after |
the driver routine {\it grdchk\_main} is called after |
| 733 |
the gradient has been computed via the adjoint |
the gradient has been computed via the adjoint |
| 734 |
(cf. Section \ref{section_grdchk}). |
(cf. Section \ref{sec:ad_gradient_check}). |
| 735 |
|
|
| 736 |
%------------------------------------------------------------------ |
%------------------------------------------------------------------ |
| 737 |
|
|
| 741 |
In order to configure AD-related setups the following packages need |
In order to configure AD-related setups the following packages need |
| 742 |
to be enabled: |
to be enabled: |
| 743 |
{\it |
{\it |
| 744 |
\begin{table}[h!] |
\begin{table}[!ht] |
| 745 |
\begin{tabular}{l} |
\begin{tabular}{l} |
| 746 |
autodiff \\ |
autodiff \\ |
| 747 |
ctrl \\ |
ctrl \\ |
| 777 |
|
|
| 778 |
%------------------------------------------------------------------ |
%------------------------------------------------------------------ |
| 779 |
|
|
| 780 |
\subsection{Building the AD code |
\subsection{Building the AD code using TAF |
| 781 |
\label{section_ad_build}} |
\label{section_ad_build}} |
| 782 |
|
|
| 783 |
The build process of an AD code is very similar to building |
The build process of an AD code is very similar to building |
| 785 |
to generate, and on which AD tool is available (TAF or TAMC), |
to generate, and on which AD tool is available (TAF or TAMC), |
| 786 |
the following {\tt make} targets are available: |
the following {\tt make} targets are available: |
| 787 |
|
|
| 788 |
\begin{table}[h!] |
\begin{table}[!ht] |
| 789 |
{\footnotesize |
{\footnotesize |
| 790 |
\begin{tabular}{ccll} |
\begin{tabular}{|ccll|} |
| 791 |
|
\hline |
| 792 |
~ & {\it AD-target} & {\it output} & {\it description} \\ |
~ & {\it AD-target} & {\it output} & {\it description} \\ |
| 793 |
\hline |
\hline |
| 794 |
\hline |
\hline |
| 807 |
~ & ~ & ~ & and compiles all code \\ |
~ & ~ & ~ & and compiles all code \\ |
| 808 |
~ & ~ & ~ & (use of TAF is set as default) \\ |
~ & ~ & ~ & (use of TAF is set as default) \\ |
| 809 |
\hline |
\hline |
|
\hline |
|
| 810 |
\end{tabular} |
\end{tabular} |
| 811 |
} |
} |
| 812 |
\end{table} |
\end{table} |
| 815 |
% |
% |
| 816 |
\begin{itemize} |
\begin{itemize} |
| 817 |
% |
% |
| 818 |
\item [$<$TOOL$>$] |
\item $<$TOOL$>$ |
| 819 |
% |
% |
| 820 |
\begin{itemize} |
\begin{itemize} |
| 821 |
% |
% |
| 824 |
% |
% |
| 825 |
\end{itemize} |
\end{itemize} |
| 826 |
% |
% |
| 827 |
\item [$<$MODE$>$] |
\item $<$MODE$>$ |
| 828 |
% |
% |
| 829 |
\begin{itemize} |
\begin{itemize} |
| 830 |
% |
% |
| 866 |
\item |
\item |
| 867 |
A header file {\tt AD\_CONFIG.h} is generated which contains a CPP option |
A header file {\tt AD\_CONFIG.h} is generated which contains a CPP option |
| 868 |
on which code ought to be generated. Depending on the {\tt make} target, |
on which code ought to be generated. Depending on the {\tt make} target, |
| 869 |
the contents is |
the contents is one of the following: |
| 870 |
\begin{itemize} |
\begin{itemize} |
| 871 |
\item |
\item |
| 872 |
{\tt \#define ALLOW\_ADJOINT\_RUN} |
{\tt \#define ALLOW\_ADJOINT\_RUN} |
| 882 |
and all {\tt .flow} files that are part of the list {\bf AD\_FLOW\_FILES}. |
and all {\tt .flow} files that are part of the list {\bf AD\_FLOW\_FILES}. |
| 883 |
% |
% |
| 884 |
\item |
\item |
| 885 |
The AD tool is invoked with the {\bf <MODE>\_<TOOL>\_FLAGS}. |
The AD tool is invoked with the {\tt <MODE>\_<TOOL>\_FLAGS}. |
| 886 |
The default AD tool flags in {\tt genmake2} can be overrwritten by |
The default AD tool flags in {\tt genmake2} can be overrwritten by |
| 887 |
an {\tt adjoint\_options} file (similar to the platform-specific |
an {\tt adjoint\_options} file (similar to the platform-specific |
| 888 |
{\tt build\_options}, see Section ???. |
{\tt build\_options}, see Section ???. |
| 1015 |
are controlled by the package {\it pkg/ctrl} and will be treated |
are controlled by the package {\it pkg/ctrl} and will be treated |
| 1016 |
in the next section. |
in the next section. |
| 1017 |
|
|
| 1018 |
\input{part5/doc_cost_flow} |
\input{s_autodiff/text/doc_cost_flow} |
| 1019 |
|
|
| 1020 |
\subsubsection{Enabling the package} |
\subsubsection{Enabling the package} |
| 1021 |
|
|
| 1105 |
the chosen cost function contributions. |
the chosen cost function contributions. |
| 1106 |
In the present example ({\bf ALLOW\_COST\_TRACER}), |
In the present example ({\bf ALLOW\_COST\_TRACER}), |
| 1107 |
S/R {\it cost\_tracer} is called. |
S/R {\it cost\_tracer} is called. |
| 1108 |
It accumulates {\bf objf\_tracer} according to eqn. (\ref{???}). |
It accumulates {\bf objf\_tracer} according to eqn. (ref:ask-the-author). |
| 1109 |
% |
% |
| 1110 |
\subsubsection{Finalize all contributions} |
\subsubsection{Finalize all contributions} |
| 1111 |
% |
% |
| 1130 |
\end{equation} |
\end{equation} |
| 1131 |
% |
% |
| 1132 |
The total cost function {\bf fc} will be the |
The total cost function {\bf fc} will be the |
| 1133 |
'dependent' variable in the argument list for TAMC, i.e. |
'dependent' variable in the argument list for TAF, i.e. |
| 1134 |
\begin{verbatim} |
\begin{verbatim} |
| 1135 |
tamc -output 'fc' ... |
taf -output 'fc' ... |
| 1136 |
\end{verbatim} |
\end{verbatim} |
| 1137 |
|
|
| 1138 |
%%%% \end{document} |
%%%% \end{document} |
| 1139 |
|
|
| 1140 |
\input{part5/doc_ad_the_main} |
\input{s_autodiff/text/doc_ad_the_main} |
| 1141 |
|
|
| 1142 |
\subsection{The control variables (independent variables) |
\subsection{The control variables (independent variables) |
| 1143 |
\label{section_ctrl}} |
\label{section_ctrl}} |
| 1157 |
(parameter setting, initialization, perturbation) |
(parameter setting, initialization, perturbation) |
| 1158 |
are controlled by the package {\it pkg/ctrl}. |
are controlled by the package {\it pkg/ctrl}. |
| 1159 |
|
|
| 1160 |
\input{part5/doc_ctrl_flow} |
\input{s_autodiff/text/doc_ctrl_flow} |
| 1161 |
|
|
| 1162 |
\subsubsection{genmake and CPP options} |
\subsubsection{genmake and CPP options} |
| 1163 |
% |
% |
| 1247 |
% |
% |
| 1248 |
The dependency flow for differentiation w.r.t. the controls |
The dependency flow for differentiation w.r.t. the controls |
| 1249 |
starts with adding a perturbation onto the input variable, |
starts with adding a perturbation onto the input variable, |
| 1250 |
thus defining the independent or control variables for TAMC. |
thus defining the independent or control variables for TAF. |
| 1251 |
Three types of controls may be considered: |
Three types of controls may be considered: |
| 1252 |
% |
% |
| 1253 |
\begin{itemize} |
\begin{itemize} |
| 1268 |
a perturbation anomaly is added to the field in S/R |
a perturbation anomaly is added to the field in S/R |
| 1269 |
{\it ctrl\_map\_ini} |
{\it ctrl\_map\_ini} |
| 1270 |
% |
% |
| 1271 |
|
%\begin{eqnarray} |
| 1272 |
\begin{equation} |
\begin{equation} |
| 1273 |
\begin{split} |
\begin{aligned} |
| 1274 |
u & = \, u_{[0]} \, + \, \Delta u \\ |
u & = \, u_{[0]} \, + \, \Delta u \\ |
| 1275 |
{\bf tr1}(...) & = \, {\bf tr1_{ini}}(...) \, + \, {\bf xx\_tr1}(...) |
{\bf tr1}(...) & = \, {\bf tr1_{ini}}(...) \, + \, {\bf xx\_tr1}(...) |
| 1276 |
\label{perturb} |
\label{perturb} |
| 1277 |
\end{split} |
\end{aligned} |
| 1278 |
\end{equation} |
\end{equation} |
| 1279 |
|
%\end{eqnarray} |
| 1280 |
% |
% |
| 1281 |
{\bf xx\_tr1} is a 3-dim. global array |
{\bf xx\_tr1} is a 3-dim. global array |
| 1282 |
holding the perturbation. In the case of a simple |
holding the perturbation. In the case of a simple |
| 1283 |
sensitivity study this array is identical to zero. |
sensitivity study this array is identical to zero. |
| 1284 |
However, it's specification is essential in the context |
However, it's specification is essential in the context |
| 1285 |
of automatic differentiation since TAMC |
of automatic differentiation since TAF |
| 1286 |
treats the corresponding line in the code symbolically |
treats the corresponding line in the code symbolically |
| 1287 |
when determining the differentiation chain and its origin. |
when determining the differentiation chain and its origin. |
| 1288 |
Thus, the variable names are part of the argument list |
Thus, the variable names are part of the argument list |
| 1289 |
when calling TAMC: |
when calling TAF: |
| 1290 |
% |
% |
| 1291 |
\begin{verbatim} |
\begin{verbatim} |
| 1292 |
tamc -input 'xx_tr1 ...' ... |
taf -input 'xx_tr1 ...' ... |
| 1293 |
\end{verbatim} |
\end{verbatim} |
| 1294 |
% |
% |
| 1295 |
Now, as mentioned above, MITgcm avoids maintaining |
Now, as mentioned above, MITgcm avoids maintaining |
| 1296 |
an array for each control variable by reading the |
an array for each control variable by reading the |
| 1297 |
perturbation to a temporary array from file. |
perturbation to a temporary array from file. |
| 1298 |
To ensure the symbolic link to be recognized by TAMC, a scalar |
To ensure the symbolic link to be recognized by TAF, a scalar |
| 1299 |
dummy variable {\bf xx\_tr1\_dummy} is introduced |
dummy variable {\bf xx\_tr1\_dummy} is introduced |
| 1300 |
and an 'active read' routine of the adjoint support |
and an 'active read' routine of the adjoint support |
| 1301 |
package {\it pkg/autodiff} is invoked. |
package {\it pkg/autodiff} is invoked. |
| 1302 |
The read-procedure is tagged with the variable |
The read-procedure is tagged with the variable |
| 1303 |
{\bf xx\_tr1\_dummy} enabling TAMC to recognize the |
{\bf xx\_tr1\_dummy} enabling TAF to recognize the |
| 1304 |
initialization of the perturbation. |
initialization of the perturbation. |
| 1305 |
The modified call of TAMC thus reads |
The modified call of TAF thus reads |
| 1306 |
% |
% |
| 1307 |
\begin{verbatim} |
\begin{verbatim} |
| 1308 |
tamc -input 'xx_tr1_dummy ...' ... |
taf -input 'xx_tr1_dummy ...' ... |
| 1309 |
\end{verbatim} |
\end{verbatim} |
| 1310 |
% |
% |
| 1311 |
and the modified operation to (\ref{perturb}) |
and the modified operation to (\ref{perturb}) |
| 1502 |
$ u_{[k+1]} $ then serves as input for a forward/adjoint run |
$ u_{[k+1]} $ then serves as input for a forward/adjoint run |
| 1503 |
to determine $ {\cal J} $ and $ \nabla _{u}{\cal J} $ at iteration step |
to determine $ {\cal J} $ and $ \nabla _{u}{\cal J} $ at iteration step |
| 1504 |
$ k+1 $. |
$ k+1 $. |
| 1505 |
Tab. \ref{???} sketches the flow between forward/adjoint model |
Tab. ref:ask-the-author sketches the flow between forward/adjoint model |
| 1506 |
and the minimization routine. |
and the minimization routine. |
| 1507 |
|
|
| 1508 |
|
{\scriptsize |
| 1509 |
\begin{eqnarray*} |
\begin{eqnarray*} |
|
\scriptsize |
|
| 1510 |
\begin{array}{ccccc} |
\begin{array}{ccccc} |
| 1511 |
u_{[0]} \,\, , \,\, \Delta u_{[k]} & ~ & ~ & ~ & ~ \\ |
u_{[0]} \,\, , \,\, \Delta u_{[k]} & ~ & ~ & ~ & ~ \\ |
| 1512 |
{\Big\downarrow} |
{\Big\downarrow} |
| 1557 |
~ & ~ & ~ & ~ & \Delta u_{[k+1]} \\ |
~ & ~ & ~ & ~ & \Delta u_{[k+1]} \\ |
| 1558 |
\end{array} |
\end{array} |
| 1559 |
\end{eqnarray*} |
\end{eqnarray*} |
| 1560 |
|
} |
| 1561 |
|
|
| 1562 |
The routines {\it ctrl\_unpack} and {\it ctrl\_pack} provide |
The routines {\it ctrl\_unpack} and {\it ctrl\_pack} provide |
| 1563 |
the link between the model and the minimization routine. |
the link between the model and the minimization routine. |
| 1564 |
As described in Section \ref{???} |
As described in Section ref:ask-the-author |
| 1565 |
the {\it unpack} and {\it pack} routines read and write |
the {\it unpack} and {\it pack} routines read and write |
| 1566 |
control and gradient {\it vectors} which are compressed |
control and gradient {\it vectors} which are compressed |
| 1567 |
to contain only wet points, in addition to the full |
to contain only wet points, in addition to the full |
Parent Directory
| 
Revision Log
| 
Revision Graph
| 
Patch