/[MITgcm]/manual/s_algorithm/text/time_stepping.tex

Diff of /manual/s_algorithm/text/time_stepping.tex

Parent Directory | Revision Log | View Revision Graph Revision Graph | View Patch Patch

-revision 1.25 by jmc,
Wed Jun 28 16:55:53 2006 UTC
+revision 1.31 by jmc,
Mon May  9 13:45:05 2011 UTC
 Line 10 
 describe the spatial discretization. The
  terms are described first, afterwards the schemes that apply to
  passive and dynamically active tracers are described.
- \input{part2/notation}
+ \input{s_algorithm/text/notation}
  \section{Time-stepping}
+ \label{sec:time_stepping}
  \begin{rawhtml}
  <!-- CMIREDIR:time-stepping: -->
  \end{rawhtml}
-Line 50 
 evaluated explicitly in time. Since the
+Line 51 
 evaluated explicitly in time. Since the
  independent of the particular time-stepping scheme chosen we will
  describe first the over-arching algorithm, known as the pressure
  method, with a rigid-lid model in section
- \ref{sect:pressure-method-rigid-lid}. This algorithm is essentially
+ \ref{sec:pressure-method-rigid-lid}. This algorithm is essentially
  unchanged, apart for some coefficients, when the rigid lid assumption
  is replaced with a linearized implicit free-surface, described in
- section \ref{sect:pressure-method-linear-backward}. These two flavors
+ section \ref{sec:pressure-method-linear-backward}. These two flavors
  of the pressure-method encompass all formulations of the model as it
  exists today. The integration of explicit in time terms is out-lined
- in section \ref{sect:adams-bashforth} and put into the context of the
+ in section \ref{sec:adams-bashforth} and put into the context of the
- overall algorithm in sections \ref{sect:adams-bashforth-sync} and
+ overall algorithm in sections \ref{sec:adams-bashforth-sync} and
- \ref{sect:adams-bashforth-staggered}. Inclusion of non-hydrostatic
+ \ref{sec:adams-bashforth-staggered}. Inclusion of non-hydrostatic
  terms requires applying the pressure method in three dimensions
  instead of two and this algorithm modification is described in section
- \ref{sect:non-hydrostatic}. Finally, the free-surface equation may be
+ \ref{sec:non-hydrostatic}. Finally, the free-surface equation may be
  treated more exactly, including non-linear terms, and this is
- described in section \ref{sect:nonlinear-freesurface}.
+ described in section \ref{sec:nonlinear-freesurface}.
  \section{Pressure method with rigid-lid}
- \label{sect:pressure-method-rigid-lid}
+ \label{sec:pressure-method-rigid-lid}
  \begin{rawhtml}
  <!-- CMIREDIR:pressure_method_rigid_lid: -->
  \end{rawhtml}
  \begin{figure}
  \begin{center}
- \resizebox{4.0in}{!}{\includegraphics{part2/pressure-method-rigid-lid.eps}}
+ \resizebox{4.0in}{!}{\includegraphics{s_algorithm/figs/pressure-method-rigid-lid.eps}}
  \end{center}
  \caption{
  A schematic of the evolution in time of the pressure method
-Line 198 
 The calling tree for these routines is g
+Line 199 
 The calling tree for these routines is g
  Fig.~\ref{fig:call-tree-pressure-method}.
+ %\paragraph{Need to discuss implicit viscosity somewhere:}
- \paragraph{Need to discuss implicit viscosity somewhere:}
+ In general, the horizontal momentum time-stepping can contain some terms
+ that are treated implicitly in time,
+ such as the vertical viscosity when using the backward time-stepping scheme
+ (\varlink{implicitViscosity}{implicitViscosity} {\it =.TRUE.}).
+ The method used to solve those implicit terms is provided in
+ section \ref{sec:implicit-backward-stepping}, and modifies
+ equations \ref{eq:discrete-time-u} and \ref{eq:discrete-time-v} to
+ give:
  \begin{eqnarray}
- \frac{1}{\Delta t} u^{n+1} - \partial_z A_v \partial_z u^{n+1}
+ u^{n+1} - \Delta t \partial_z A_v \partial_z u^{n+1}
- + g \partial_x \eta^{n+1} & = & \frac{1}{\Delta t} u^{n} +
+ + \Delta t g \partial_x \eta^{n+1} & = & u^{n} + \Delta t G_u^{(n+1/2)}
- G_u^{(n+1/2)}
  \\
- \frac{1}{\Delta t} v^{n+1} - \partial_z A_v \partial_z v^{n+1}
+ v^{n+1} - \Delta t \partial_z A_v \partial_z v^{n+1}
- + g \partial_y \eta^{n+1} & = & \frac{1}{\Delta t} v^{n} + G_v^{(n+1/2)}
+ + \Delta t g \partial_y \eta^{n+1} & = & v^{n} + \Delta t G_v^{(n+1/2)}
  \end{eqnarray}
  \section{Pressure method with implicit linear free-surface}
- \label{sect:pressure-method-linear-backward}
+ \label{sec:pressure-method-linear-backward}
  \begin{rawhtml}
  <!-- CMIREDIR:pressure_method_linear_backward: -->
  \end{rawhtml}
-Line 245 
 pressure method is then replaced by the
+Line 252 
 pressure method is then replaced by the
  \label{eq:discrete-time-backward-free-surface}
  \end{equation}
  where the use of flow at time level $n+1$ makes the method implicit
- and backward in time. The is the preferred scheme since it still
+ and backward in time. This is the preferred scheme since it still
  filters the fast unresolved wave motions by damping them. A centered
- scheme, such as Crank-Nicholson, would alias the energy of the fast
+ scheme, such as Crank-Nicholson (see section \ref{sec:freesurf-CrankNick}),
- modes onto slower modes of motion.
+ would alias the energy of the fast modes onto slower modes of motion.
  As for the rigid-lid pressure method, equations
  \ref{eq:discrete-time-u}, \ref{eq:discrete-time-v} and
-Line 256 
 As for the rigid-lid pressure method, eq
+Line 263 
 As for the rigid-lid pressure method, eq
  \begin{eqnarray}
  u^{*} & = & u^{n} + \Delta t G_u^{(n+1/2)} \label{eq:ustar-backward-free-surface} \\
  v^{*} & = & v^{n} + \Delta t G_v^{(n+1/2)} \label{eq:vstar-backward-free-surface} \\
- \eta^* & = & \epsilon_{fs} \left( \eta^{n} + \Delta t (P-E) \right)- \Delta t
+ \eta^* & = & \epsilon_{fs} \left( \eta^{n} + \Delta t (P-E) \right)
-   \partial_x H \widehat{u^{*}}
+          - \Delta t \left( \partial_x H \widehat{u^{*}}
- + \partial_y H \widehat{v^{*}}
+                          + \partial_y H \widehat{v^{*}} \right)
  \\
    \partial_x g H \partial_x \eta^{n+1}
  & + & \partial_y g H \partial_y \eta^{n+1}
-Line 279 
 recovered. However, the implicit treatme
+Line 286 
 recovered. However, the implicit treatme
  the flow to be divergent and for the surface pressure/elevation to
  respond on a finite time-scale (as opposed to instantly). To recover
  the rigid-lid formulation, we introduced a switch-like parameter,
- $\epsilon_{fs}$, which selects between the free-surface and rigid-lid;
+ $\epsilon_{fs}$ (\varlink{freesurfFac}{freesurfFac}),
+ which selects between the free-surface and rigid-lid;
  $\epsilon_{fs}=1$ allows the free-surface to evolve; $\epsilon_{fs}=0$
  imposes the rigid-lid. The evolution in time and location of variables
  is exactly as it was for the rigid-lid model so that
-Line 290 
 pressure-method.
+Line 298 
 pressure-method.
  \section{Explicit time-stepping: Adams-Bashforth}
- \label{sect:adams-bashforth}
+ \label{sec:adams-bashforth}
  \begin{rawhtml}
  <!-- CMIREDIR:adams_bashforth: -->
  \end{rawhtml}
-Line 300 
 time discretization of the explicit term
+Line 308 
 time discretization of the explicit term
  the quasi-second order Adams-Bashforth method for all explicit terms
  in both the momentum and tracer equations. This is still the default
  mode of operation but it is now possible to use alternate schemes for
- tracers (see section \ref{sect:tracer-advection}).
+ tracers (see section \ref{sec:tracer-advection}).
  \begin{figure}
  \begin{center} \fbox{ \begin{minipage}{4.5in} \begin{tabbing}
-Line 317 
 FORWARD\_STEP \\
+Line 325 
 FORWARD\_STEP \\
  \end{tabbing} \end{minipage} } \end{center}
  \caption{
  Calling tree for the Adams-Bashforth time-stepping of temperature with
- implicit diffusion.}
+ implicit diffusion.
+   (\filelink{THERMODYNAMICS}{model-src-thermodynamics.F},
+    \filelink{ADAMS\_BASHFORTH2}{model-src-adams_bashforth2.F})}
  \label{fig:call-tree-adams-bashforth}
  \end{figure}
-Line 348 
 simpler to include these terms and this
+Line 358 
 simpler to include these terms and this
  and forcing evolves smoothly. Problems can, and do, arise when forcing
  or motions are high frequency and this corresponds to a reduced
  stability compared to a simple forward time-stepping of such terms.
- The model offers the possibility to leave the forcing term outside the
+ The model offers the possibility to leave the tracer and momentum
- Adams-Bashforth extrapolation, by turning off the logical flag
+ forcing terms and the dissipation terms outside the
- {\bf forcing\_In\_AB } (parameter file {\em data}, namelist {\em PARM01},
+ Adams-Bashforth extrapolation, by turning off the logical flags
- default value = True).
+ \varlink{forcing\_In\_AB}{forcing_In_AB}
+ (parameter file {\em data}, namelist {\em PARM01}, default value = True).
+ (\varlink{tracForcingOutAB}{tracForcingOutAB}, default=0,
+ \varlink{momForcingOutAB}{momForcingOutAB}, default=0)
+ and \varlink{momDissip\_In\_AB}{momDissip_In_AB}
+ (parameter file {\em data}, namelist {\em PARM01}, default value = True).
+ respectively.
  A stability analysis for an oscillation equation should be given at this point.
  \marginpar{AJA needs to find his notes on this...}
-Line 359 
 A stability analysis for an oscillation
+Line 375 
 A stability analysis for an oscillation
  A stability analysis for a relaxation equation should be given at this point.
  \marginpar{...and for this too.}
+ \begin{figure}
+ \begin{center}
+ \resizebox{5.5in}{!}{\includegraphics{s_algorithm/figs/oscil+damp_AB2.eps}}
+ \end{center}
+ \caption{
+ Oscillatory and damping response of
+ quasi-second order Adams-Bashforth scheme for different values
+ of the $\epsilon_{AB}$ parameter (0., 0.1, 0.25, from top to bottom)
+ The analytical solution (in black), the physical mode (in blue)
+ and the numerical mode (in red) are represented with a CFL
+ step of 0.1.
+ The left column represents the oscillatory response
+ on the complex plane for CFL ranging from 0.1 up to 0.9.
+ The right column represents the damping response amplitude
+ (y-axis) function of the CFL (x-axis).
+ }
+ \label{fig:adams-bashforth-respons}
+ \end{figure}
  \section{Implicit time-stepping: backward method}
+ \label{sec:implicit-backward-stepping}
  \begin{rawhtml}
  <!-- CMIREDIR:implicit_time-stepping_backward: -->
  \end{rawhtml}
-Line 388 
 using the Adams-Bashforth method as desc
+Line 425 
 using the Adams-Bashforth method as desc
  \end{eqnarray}
  where ${\cal L}_\tau^{-1}$ is the inverse of the operator
  \begin{equation}
- {\cal L} = \left[ 1 + \Delta t \partial_r \kappa_v \partial_r \right]
+ {\cal L}_\tau = \left[ 1 + \Delta t \partial_r \kappa_v \partial_r \right]
  \end{equation}
  Equation \ref{eq:taustar-implicit} looks exactly as \ref{eq:taustar}
  while \ref{eq:tau-n+1-implicit} involves an operator or matrix
-Line 408 
 unconditionally stable, no-slip boundary
+Line 445 
 unconditionally stable, no-slip boundary
  implicit and are thus cast as a an explicit drag term.
  \section{Synchronous time-stepping: variables co-located in time}
- \label{sect:adams-bashforth-sync}
+ \label{sec:adams-bashforth-sync}
  \begin{rawhtml}
  <!-- CMIREDIR:adams_bashforth_sync: -->
  \end{rawhtml}
  \begin{figure}
  \begin{center}
- \resizebox{5.0in}{!}{\includegraphics{part2/adams-bashforth-sync.eps}}
+ \resizebox{5.0in}{!}{\includegraphics{s_algorithm/figs/adams-bashforth-sync.eps}}
  \end{center}
  \caption{
  A schematic of the explicit Adams-Bashforth and implicit time-stepping
-Line 505 
 G_{\theta,S}^{(n+1/2)} & = & (3/2+\epsil
+Line 542 
 G_{\theta,S}^{(n+1/2)} & = & (3/2+\epsil
  \nabla \cdot g H \nabla \eta^{n+1} & - & \frac{\epsilon_{fs} \eta^{n+1}}{\Delta t^2}
  ~ = ~ - \frac{\eta^*}{\Delta t^2}
  \label{eq:elliptic-sync} \\
- \vec{\bf v}^{n+1} & = & \vec{\bf v}^{*} - \Delta t g \nabla \eta^{n+1}
+ \vec{\bf v}^{n+1} & = & \vec{\bf v}^{**} - \Delta t g \nabla \eta^{n+1}
  \label{eq:v-n+1-sync}
  \end{eqnarray}
  Fig.~\ref{fig:adams-bashforth-sync} illustrates the location of
-Line 528 
 time-step. The corresponding calling tre
+Line 565 
 time-step. The corresponding calling tre
  \ref{fig:call-tree-adams-bashforth-sync}.
  \section{Staggered baroclinic time-stepping}
- \label{sect:adams-bashforth-staggered}
+ \label{sec:adams-bashforth-staggered}
  \begin{rawhtml}
  <!-- CMIREDIR:adams_bashforth_staggered: -->
  \end{rawhtml}
  \begin{figure}
  \begin{center}
- \resizebox{5.5in}{!}{\includegraphics{part2/adams-bashforth-staggered.eps}}
+ \resizebox{5.5in}{!}{\includegraphics{s_algorithm/figs/adams-bashforth-staggered.eps}}
  \end{center}
  \caption{
  A schematic of the explicit Adams-Bashforth and implicit time-stepping
-Line 576 
 thermodynamics solver is delayed from ha
+Line 613 
 thermodynamics solver is delayed from ha
  allowing the use of the most recent velocities to compute
  the advection terms. Once the thermodynamics fields are
  updated, the hydrostatic pressure is computed
- to step forwrad the dynamics.
+ to step forward the dynamics.
  Note that the pressure gradient must also be taken out of the
  Adams-Bashforth extrapolation. Also, retaining the integer time-levels,
  $n$ and $n+1$, does not give a user the sense of where variables are
-Line 600 
 position in time of variables appropriat
+Line 637 
 position in time of variables appropriat
  \nabla \cdot g H \nabla \eta^{n+1/2} & - & \frac{\epsilon_{fs} \eta^{n+1/2}}{\Delta t^2}
  ~ = ~ - \frac{\eta^*}{\Delta t^2}
  \label{eq:elliptic-staggered} \\
- \vec{\bf v}^{n+1/2} & = & \vec{\bf v}^{*} - \Delta t g \nabla \eta^{n+1/2}
+ \vec{\bf v}^{n+1/2} & = & \vec{\bf v}^{**} - \Delta t g \nabla \eta^{n+1/2}
  \label{eq:v-n+1-staggered} \\
  G_{\theta,S}^{n} & = & G_{\theta,S} ( u^{n+1/2}, \theta^{n}, S^{n} )
  \label{eq:Gt-n-staggered} \\
-Line 673 
 time-level variables and terms correspon
+Line 710 
 time-level variables and terms correspon
  \section{Non-hydrostatic formulation}
- \label{sect:non-hydrostatic}
+ \label{sec:non-hydrostatic}
  \begin{rawhtml}
  <!-- CMIREDIR:non-hydrostatic_formulation: -->
  \end{rawhtml}
-Line 681 
 time-level variables and terms correspon
+Line 718 
 time-level variables and terms correspon
  The non-hydrostatic formulation re-introduces the full vertical
  momentum equation and requires the solution of a 3-D elliptic
  equations for non-hydrostatic pressure perturbation. We still
- intergrate vertically for the hydrostatic pressure and solve a 2-D
+ integrate vertically for the hydrostatic pressure and solve a 2-D
  elliptic equation for the surface pressure/elevation for this reduces
  the amount of work needed to solve for the non-hydrostatic pressure.
-Line 722 
 Substituting into the depth integrated c
+Line 759 
 Substituting into the depth integrated c
  \partial_x H \partial_x \left( g \eta^{n+1} + \widehat{\phi}_{nh}^{n+1} \right)
  +
  \partial_y H \partial_y \left( g \eta^{n+1} + \widehat{\phi}_{nh}^{n+1} \right)
-  - \frac{\epsilon_{fs}\eta^*}{\Delta t^2}
+  - \frac{\epsilon_{fs}\eta^{n+1}}{\Delta t^2}
  = - \frac{\eta^*}{\Delta t^2}
  \end{equation}
  which is approximated by equation
-Line 730 
 which is approximated by equation
+Line 767 
 which is approximated by equation
  $\phi_{nh}^{n+1}$ is not yet known and ii) $\nabla \widehat{\phi}_{nh}
  << g \nabla \eta$. If \ref{eq:elliptic-backward-free-surface} is
  solved accurately then the implication is that $\widehat{\phi}_{nh}
- \approx 0$ so that thet non-hydrostatic pressure field does not drive
+ \approx 0$ so that the non-hydrostatic pressure field does not drive
  barotropic motion.
  The flow must satisfy non-divergence
-Line 750 
 u^{*} & = & u^{n} + \Delta t G_u^{(n+1/2
+Line 787 
 u^{*} & = & u^{n} + \Delta t G_u^{(n+1/2
  v^{*} & = & v^{n} + \Delta t G_v^{(n+1/2)} \label{eq:vstar-nh} \\
  w^{*} & = & w^{n} + \Delta t G_w^{(n+1/2)} \label{eq:wstar-nh} \\
  \eta^* ~ = ~ \epsilon_{fs} \left( \eta^{n} + \Delta t (P-E) \right)
- & - & \Delta t
+ & - & \Delta t \left( \partial_x H \widehat{u^{*}}
-   \partial_x H \widehat{u^{*}}
+                     + \partial_y H \widehat{v^{*}} \right)
- + \partial_y H \widehat{v^{*}}
  \\
    \partial_x g H \partial_x \eta^{n+1}
  + \partial_y g H \partial_y \eta^{n+1}
-Line 765 
 u^{**} & = & u^{*} - \Delta t g \partial
+Line 801 
 u^{**} & = & u^{*} - \Delta t g \partial
  v^{**} & = & v^{*} - \Delta t g \partial_y \eta^{n+1} \label{eq:vnx-nh}\\
  \partial_{xx} \phi_{nh}^{n+1} + \partial_{yy} \phi_{nh}^{n+1} +
  \partial_{rr} \phi_{nh}^{n+1} & = &
- \partial_x u^{**} + \partial_y v^{**} + \partial_r w^{*} \\
+ \partial_x u^{**} + \partial_y v^{**} + \partial_r w^{*}  \label{eq:phi-nh}\\
  u^{n+1} & = & u^{**} - \Delta t \partial_x \phi_{nh}^{n+1} \label{eq:un+1-nh}\\
  v^{n+1} & = & v^{**} - \Delta t \partial_y \phi_{nh}^{n+1} \label{eq:vn+1-nh}\\
  \partial_r w^{n+1} & = & - \partial_x u^{n+1} - \partial_y v^{n+1}
-Line 776 
 $w^{n+1}$.
+Line 812 
 $w^{n+1}$.
  \section{Variants on the Free Surface}
- \label{sect:free-surface}
+ \label{sec:free-surface}
  We now describe the various formulations of the free-surface that
  include non-linear forms, implicit in time using Crank-Nicholson,
-Line 889 
 at the same point in the code.
+Line 925 
 at the same point in the code.
  \subsection{Crank-Nickelson barotropic time stepping}
- \label{sect:freesurf-CrankNick}
+ \label{sec:freesurf-CrankNick}
  The full implicit time stepping described previously is
  unconditionally stable but damps the fast gravity waves, resulting in
-Line 913 
 Equations \ref{eq:ustar-backward-free-su
+Line 949 
 Equations \ref{eq:ustar-backward-free-su
  \frac{ \vec{\bf v}^{n+1} }{ \Delta t }
  + {\bf \nabla}_h b_s [ \beta {\eta}^{n+1} + (1-\beta) {\eta}^{n} ]
  + \epsilon_{nh} {\bf \nabla}_h {\phi'_{nh}}^{n+1}
-  = \frac{ \vec{\bf v}^* }{ \Delta t }
+  = \frac{ \vec{\bf v}^{n} }{ \Delta t }
+  + \vec{\bf G}_{\vec{\bf v}} ^{(n+1/2)}
+  + {\bf \nabla}_h {\phi'_{hyd}}^{(n+1/2)}
  \end{eqnarray*}
  \begin{eqnarray}
  \epsilon_{fs} \frac{ {\eta}^{n+1} - {\eta}^{n} }{ \Delta t}
-Line 922 
 Equations \ref{eq:ustar-backward-free-su
+Line 960 
 Equations \ref{eq:ustar-backward-free-su
  = \epsilon_{fw} (P-E)
  \label{eq:eta-n+1-CrankNick}
  \end{eqnarray}
- where:
+ We set
  \begin{eqnarray*}
  \vec{\bf v}^* & = &
  \vec{\bf v} ^{n} + \Delta t \vec{\bf G}_{\vec{\bf v}} ^{(n+1/2)}

 Legend:



Removed from v.1.25
 


changed lines


 
Added in v.1.31
 Legend:



Removed from v.1.25
 


changed lines


 
Added in v.1.31
-Removed from v.1.25
+Added in v.1.31

	ViewVC Help
Powered by ViewVC 1.1.22