1 |
% $Header: /u/gcmpack/mitgcmdoc/part2/time_stepping.tex,v 1.10 2001/11/06 15:07:32 adcroft Exp $ |
2 |
% $Name: $ |
3 |
|
4 |
The equations of motion integrated by the model involve four |
5 |
prognostic equations for flow, $u$ and $v$, temperature, $\theta$, and |
6 |
salt/moisture, $S$, and three diagnostic equations for vertical flow, |
7 |
$w$, density/buoyancy, $\rho$/$b$, and pressure/geo-potential, |
8 |
$\phi_{hyd}$. In addition, the surface pressure or height may by |
9 |
described by either a prognostic or diagnostic equation and if |
10 |
non-hydrostatics terms are included then a diagnostic equation for |
11 |
non-hydrostatic pressure is also solved. The combination of prognostic |
12 |
and diagnostic equations requires a model algorithm that can march |
13 |
forward prognostic variables while satisfying constraints imposed by |
14 |
diagnostic equations. |
15 |
|
16 |
Since the model comes in several flavors and formulation, it would be |
17 |
confusing to present the model algorithm exactly as written into code |
18 |
along with all the switches and optional terms. Instead, we present |
19 |
the algorithm for each of the basic formulations which are: |
20 |
\begin{enumerate} |
21 |
\item the semi-implicit pressure method for hydrostatic equations |
22 |
with a rigid-lid, variables co-located in time and with |
23 |
Adams-Bashforth time-stepping, \label{it:a} |
24 |
\item as \ref{it:a}. but with an implicit linear free-surface, \label{it:b} |
25 |
\item as \ref{it:a}. or \ref{it:b}. but with variables staggered in time, |
26 |
\label{it:c} |
27 |
\item as \ref{it:a}. or \ref{it:b}. but with non-hydrostatic terms included, |
28 |
\item as \ref{it:b}. or \ref{it:c}. but with non-linear free-surface. |
29 |
\end{enumerate} |
30 |
|
31 |
In all the above configurations it is also possible to substitute the |
32 |
Adams-Bashforth with an alternative time-stepping scheme for terms |
33 |
evaluated explicitly in time. Since the over-arching algorithm is |
34 |
independent of the particular time-stepping scheme chosen we will |
35 |
describe first the over-arching algorithm, known as the pressure |
36 |
method, with a rigid-lid model in section |
37 |
\ref{sect:pressure-method-rigid-lid}. This algorithm is essentially |
38 |
unchanged, apart for some coefficients, when the rigid lid assumption |
39 |
is replaced with a linearized implicit free-surface, described in |
40 |
section \ref{sect:pressure-method-linear-backward}. These two flavors |
41 |
of the pressure-method encompass all formulations of the model as it |
42 |
exists today. The integration of explicit in time terms is out-lined |
43 |
in section \ref{sect:adams-bashforth} and put into the context of the |
44 |
overall algorithm in sections \ref{sect:adams-bashforth-sync} and |
45 |
\ref{sect:adams-bashforth-staggered}. Inclusion of non-hydrostatic |
46 |
terms requires applying the pressure method in three dimensions |
47 |
instead of two and this algorithm modification is described in section |
48 |
\ref{sect:non-hydrostatic}. Finally, the free-surface equation may be |
49 |
treated more exactly, including non-linear terms, and this is |
50 |
described in section \ref{sect:nonlinear-freesurface}. |
51 |
|
52 |
|
53 |
\section{Pressure method with rigid-lid} \label{sect:pressure-method-rigid-lid} |
54 |
|
55 |
\begin{figure} |
56 |
\begin{center} |
57 |
\resizebox{4.0in}{!}{\includegraphics{part2/pressure-method-rigid-lid.eps}} |
58 |
\end{center} |
59 |
\caption{ |
60 |
A schematic of the evolution in time of the pressure method |
61 |
algorithm. A prediction for the flow variables at time level $n+1$ is |
62 |
made based only on the explicit terms, $G^{(n+^1/_2)}$, and denoted |
63 |
$u^*$, $v^*$. Next, a pressure field is found such that $u^{n+1}$, |
64 |
$v^{n+1}$ will be non-divergent. Conceptually, the $*$ quantities |
65 |
exist at time level $n+1$ but they are intermediate and only |
66 |
temporary.} |
67 |
\label{fig:pressure-method-rigid-lid} |
68 |
\end{figure} |
69 |
|
70 |
\begin{figure} |
71 |
\begin{center} \fbox{ \begin{minipage}{4.5in} \begin{tabbing} |
72 |
aaa \= aaa \= aaa \= aaa \= aaa \= aaa \kill |
73 |
FORWARD\_STEP \\ |
74 |
\> DYNAMICS \\ |
75 |
\>\> TIMESTEP \` $u^*$,$v^*$ (\ref{eq:ustar-rigid-lid},\ref{eq:vstar-rigid-lid}) \\ |
76 |
\> SOLVE\_FOR\_PRESSURE \\ |
77 |
\>\> CALC\_DIV\_GHAT \` $H\widehat{u^*}$,$H\widehat{v^*}$ (\ref{eq:elliptic}) \\ |
78 |
\>\> CG2D \` $\eta^{n+1}$ (\ref{eq:elliptic}) \\ |
79 |
\> THE\_CORRECTION\_STEP \\ |
80 |
\>\> CALC\_GRAD\_PHI\_SURF \` $\nabla \eta^{n+1}$ \\ |
81 |
\>\> CORRECTION\_STEP \` $u^{n+1}$,$v^{n+1}$ (\ref{eq:un+1-rigid-lid},\ref{eq:vn+1-rigid-lid}) |
82 |
\end{tabbing} \end{minipage} } \end{center} |
83 |
\caption{Calling tree for the pressure method algorihtm} |
84 |
\label{fig:call-tree-pressure-method} |
85 |
\end{figure} |
86 |
|
87 |
The horizontal momentum and continuity equations for the ocean |
88 |
(\ref{eq:ocean-mom} and \ref{eq:ocean-cont}), or for the atmosphere |
89 |
(\ref{eq:atmos-mom} and \ref{eq:atmos-cont}), can be summarized by: |
90 |
\begin{eqnarray} |
91 |
\partial_t u + g \partial_x \eta & = & G_u \\ |
92 |
\partial_t v + g \partial_y \eta & = & G_v \\ |
93 |
\partial_x u + \partial_y v + \partial_z w & = & 0 |
94 |
\end{eqnarray} |
95 |
where we are adopting the oceanic notation for brevity. All terms in |
96 |
the momentum equations, except for surface pressure gradient, are |
97 |
encapsulated in the $G$ vector. The continuity equation, when |
98 |
integrated over the fluid depth, $H$, and with the rigid-lid/no normal |
99 |
flow boundary conditions applied, becomes: |
100 |
\begin{equation} |
101 |
\partial_x H \widehat{u} + \partial_y H \widehat{v} = 0 |
102 |
\label{eq:rigid-lid-continuity} |
103 |
\end{equation} |
104 |
Here, $H\widehat{u} = \int_H u dz$ is the depth integral of $u$, |
105 |
similarly for $H\widehat{v}$. The rigid-lid approximation sets $w=0$ |
106 |
at the lid so that it does not move but allows a pressure to be |
107 |
exerted on the fluid by the lid. The horizontal momentum equations and |
108 |
vertically integrated continuity equation are be discretized in time |
109 |
and space as follows: |
110 |
\begin{eqnarray} |
111 |
u^{n+1} + \Delta t g \partial_x \eta^{n+1} |
112 |
& = & u^{n} + \Delta t G_u^{(n+1/2)} |
113 |
\label{eq:discrete-time-u} |
114 |
\\ |
115 |
v^{n+1} + \Delta t g \partial_y \eta^{n+1} |
116 |
& = & v^{n} + \Delta t G_v^{(n+1/2)} |
117 |
\label{eq:discrete-time-v} |
118 |
\\ |
119 |
\partial_x H \widehat{u^{n+1}} |
120 |
+ \partial_y H \widehat{v^{n+1}} & = & 0 |
121 |
\label{eq:discrete-time-cont-rigid-lid} |
122 |
\end{eqnarray} |
123 |
As written here, terms on the LHS all involve time level $n+1$ and are |
124 |
referred to as implicit; the implicit backward time stepping scheme is |
125 |
being used. All other terms in the RHS are explicit in time. The |
126 |
thermodynamic quantities are integrated forward in time in parallel |
127 |
with the flow and will be discussed later. For the purposes of |
128 |
describing the pressure method it suffices to say that the hydrostatic |
129 |
pressure gradient is explicit and so can be included in the vector |
130 |
$G$. |
131 |
|
132 |
Substituting the two momentum equations into the depth integrated |
133 |
continuity equation eliminates $u^{n+1}$ and $v^{n+1}$ yielding an |
134 |
elliptic equation for $\eta^{n+1}$. Equations |
135 |
\ref{eq:discrete-time-u}, \ref{eq:discrete-time-v} and |
136 |
\ref{eq:discrete-time-cont-rigid-lid} can then be re-arranged as follows: |
137 |
\begin{eqnarray} |
138 |
u^{*} & = & u^{n} + \Delta t G_u^{(n+1/2)} \label{eq:ustar-rigid-lid} \\ |
139 |
v^{*} & = & v^{n} + \Delta t G_v^{(n+1/2)} \label{eq:vstar-rigid-lid} \\ |
140 |
\partial_x \Delta t g H \partial_x \eta^{n+1} |
141 |
+ \partial_y \Delta t g H \partial_y \eta^{n+1} |
142 |
& = & |
143 |
\partial_x H \widehat{u^{*}} |
144 |
+ \partial_y H \widehat{v^{*}} \label{eq:elliptic} |
145 |
\\ |
146 |
u^{n+1} & = & u^{*} - \Delta t g \partial_x \eta^{n+1} \label{eq:un+1-rigid-lid}\\ |
147 |
v^{n+1} & = & v^{*} - \Delta t g \partial_y \eta^{n+1} \label{eq:vn+1-rigid-lid} |
148 |
\end{eqnarray} |
149 |
Equations \ref{eq:ustar-rigid-lid} to \ref{eq:vn+1-rigid-lid}, solved |
150 |
sequentially, represent the pressure method algorithm used in the |
151 |
model. The essence of the pressure method lies in the fact that any |
152 |
explicit prediction for the flow would lead to a divergence flow field |
153 |
so a pressure field must be found that keeps the flow non-divergent |
154 |
over each step of the integration. The particular location in time of |
155 |
the pressure field is somewhat ambiguous; in |
156 |
Fig.~\ref{fig:pressure-method-rigid-lid} we depicted as co-located |
157 |
with the future flow field (time level $n+1$) but it could equally |
158 |
have been drawn as staggered in time with the flow. |
159 |
|
160 |
The correspondence to the code is as follows: |
161 |
\begin{itemize} |
162 |
\item |
163 |
the prognostic phase, equations \ref{eq:ustar-rigid-lid} and \ref{eq:vstar-rigid-lid}, |
164 |
stepping forward $u^n$ and $v^n$ to $u^{*}$ and $v^{*}$ is coded in |
165 |
{\em TIMESTEP.F} |
166 |
\item |
167 |
the vertical integration, $H \widehat{u^*}$ and $H |
168 |
\widehat{v^*}$, divergence and inversion of the elliptic operator in |
169 |
equation \ref{eq:elliptic} is coded in {\em |
170 |
SOLVE\_FOR\_PRESSURE.F} |
171 |
\item |
172 |
finally, the new flow field at time level $n+1$ given by equations |
173 |
\ref{eq:un+1-rigid-lid} and \ref{eq:vn+1-rigid-lid} is calculated in {\em CORRECTION\_STEP.F}. |
174 |
\end{itemize} |
175 |
The calling tree for these routines is given in |
176 |
Fig.~\ref{fig:call-tree-pressure-method}. |
177 |
|
178 |
|
179 |
|
180 |
\paragraph{Need to discuss implicit viscosity somewhere:} |
181 |
\begin{eqnarray} |
182 |
\frac{1}{\Delta t} u^{n+1} - \partial_z A_v \partial_z u^{n+1} |
183 |
+ g \partial_x \eta^{n+1} & = & \frac{1}{\Delta t} u^{n} + |
184 |
G_u^{(n+1/2)} |
185 |
\\ |
186 |
\frac{1}{\Delta t} v^{n+1} - \partial_z A_v \partial_z v^{n+1} |
187 |
+ g \partial_y \eta^{n+1} & = & \frac{1}{\Delta t} v^{n} + G_v^{(n+1/2)} |
188 |
\end{eqnarray} |
189 |
|
190 |
|
191 |
\section{Pressure method with implicit linear free-surface} |
192 |
\label{sect:pressure-method-linear-backward} |
193 |
|
194 |
The rigid-lid approximation filters out external gravity waves |
195 |
subsequently modifying the dispersion relation of barotropic Rossby |
196 |
waves. The discrete form of the elliptic equation has some zero |
197 |
eigen-values which makes it a potentially tricky or inefficient |
198 |
problem to solve. |
199 |
|
200 |
The rigid-lid approximation can be easily replaced by a linearization |
201 |
of the free-surface equation which can be written: |
202 |
\begin{equation} |
203 |
\partial_t \eta + \partial_x H \widehat{u} + \partial_y H \widehat{v} = P-E+R |
204 |
\label{eq:linear-free-surface=P-E+R} |
205 |
\end{equation} |
206 |
which differs from the depth integrated continuity equation with |
207 |
rigid-lid (\ref{eq:rigid-lid-continuity}) by the time-dependent term |
208 |
and fresh-water source term. |
209 |
|
210 |
Equation \ref{eq:discrete-time-cont-rigid-lid} in the rigid-lid |
211 |
pressure method is then replaced by the time discretization of |
212 |
\ref{eq:linear-free-surface=P-E+R} which is: |
213 |
\begin{equation} |
214 |
\eta^{n+1} |
215 |
+ \Delta t \partial_x H \widehat{u^{n+1}} |
216 |
+ \Delta t \partial_y H \widehat{v^{n+1}} |
217 |
= |
218 |
\eta^{n} |
219 |
+ \Delta t ( P - E + R ) |
220 |
\label{eq:discrete-time-backward-free-surface} |
221 |
\end{equation} |
222 |
where the use of flow at time level $n+1$ makes the method implicit |
223 |
and backward in time. The is the preferred scheme since it still |
224 |
filters the fast unresolved wave motions by damping them. A centered |
225 |
scheme, such as Crank-Nicholson, would alias the energy of the fast |
226 |
modes onto slower modes of motion. |
227 |
|
228 |
As for the rigid-lid pressure method, equations |
229 |
\ref{eq:discrete-time-u}, \ref{eq:discrete-time-v} and |
230 |
\ref{eq:discrete-time-backward-free-surface} can be re-arranged as follows: |
231 |
\begin{eqnarray} |
232 |
u^{*} & = & u^{n} + \Delta t G_u^{(n+1/2)} \label{eq:ustar-backward-free-surface} \\ |
233 |
v^{*} & = & v^{n} + \Delta t G_v^{(n+1/2)} \label{eq:vstar-backward-free-surface} \\ |
234 |
\eta^* & = & \epsilon_{fs} \left( \eta^{n} +P-E+R \right)- \Delta t |
235 |
\partial_x H \widehat{u^{*}} |
236 |
+ \partial_y H \widehat{v^{*}} |
237 |
\\ |
238 |
\partial_x g H \partial_x \eta^{n+1} |
239 |
+ \partial_y g H \partial_y \eta^{n+1} |
240 |
- \frac{\epsilon_{fs} \eta^{n+1}}{\Delta t^2} |
241 |
& = & |
242 |
- \frac{\eta^*}{\Delta t^2} |
243 |
\label{eq:elliptic-backward-free-surface} |
244 |
\\ |
245 |
u^{n+1} & = & u^{*} - \Delta t g \partial_x \eta^{n+1} \label{eq:un+1-backward-free-surface}\\ |
246 |
v^{n+1} & = & v^{*} - \Delta t g \partial_y \eta^{n+1} \label{eq:vn+1-backward-free-surface} |
247 |
\end{eqnarray} |
248 |
Equations~\ref{eq:ustar-backward-free-surface} |
249 |
to~\ref{eq:vn+1-backward-free-surface}, solved sequentially, represent |
250 |
the pressure method algorithm with a backward implicit, linearized |
251 |
free surface. The method is still formerly a pressure method because |
252 |
in the limit of large $\Delta t$ the rigid-lid method is |
253 |
recovered. However, the implicit treatment of the free-surface allows |
254 |
the flow to be divergent and for the surface pressure/elevation to |
255 |
respond on a finite time-scale (as opposed to instantly). To recover |
256 |
the rigid-lid formulation, we introduced a switch-like parameter, |
257 |
$\epsilon_{fs}$, which selects between the free-surface and rigid-lid; |
258 |
$\epsilon_{fs}=1$ allows the free-surface to evolve; $\epsilon_{fs}=0$ |
259 |
imposes the rigid-lid. The evolution in time and location of variables |
260 |
is exactly as it was for the rigid-lid model so that |
261 |
Fig.~\ref{fig:pressure-method-rigid-lid} is still |
262 |
applicable. Similarly, the calling sequence, given in |
263 |
Fig.~\ref{fig:call-tree-pressure-method}, is as for the |
264 |
pressure-method. |
265 |
|
266 |
|
267 |
\section{Explicit time-stepping: Adams-Bashforth} |
268 |
\label{sect:adams-bashforth} |
269 |
|
270 |
In describing the the pressure method above we deferred describing the |
271 |
time discretization of the explicit terms. We have historically used |
272 |
the quasi-second order Adams-Bashforth method for all explicit terms |
273 |
in both the momentum and tracer equations. This is still the default |
274 |
mode of operation but it is now possible to use alternate schemes for |
275 |
tracers (see section \ref{sect:tracer-advection}). |
276 |
|
277 |
\begin{figure} |
278 |
\begin{center} \fbox{ \begin{minipage}{4.5in} \begin{tabbing} |
279 |
aaa \= aaa \= aaa \= aaa \= aaa \= aaa \kill |
280 |
FORWARD\_STEP \\ |
281 |
\> THERMODYNAMICS \\ |
282 |
\>\> CALC\_GT \\ |
283 |
\>\>\> GAD\_CALC\_RHS \` $G_\theta^n = G_\theta( u, \theta^n )$ \\ |
284 |
\>\>\> EXTERNAL\_FORCING \` $G_\theta^n = G_\theta^n + {\cal Q}$ \\ |
285 |
\>\>\> ADAMS\_BASHFORTH2 \` $G_\theta^{(n+1/2)}$ (\ref{eq:adams-bashforth2}) \\ |
286 |
\>\> TIMESTEP\_TRACER \` $\tau^*$ (\ref{eq:taustar}) \\ |
287 |
\>\> IMPLDIFF \` $\tau^{(n+1)}$ (\ref{eq:tau-n+1-implicit}) |
288 |
\end{tabbing} \end{minipage} } \end{center} |
289 |
\caption{ |
290 |
Calling tree for the Adams-Bashforth time-stepping of temperature with |
291 |
implicit diffusion.} |
292 |
\label{fig:call-tree-adams-bashforth} |
293 |
\end{figure} |
294 |
|
295 |
In the previous sections, we summarized an explicit scheme as: |
296 |
\begin{equation} |
297 |
\tau^{*} = \tau^{n} + \Delta t G_\tau^{(n+1/2)} |
298 |
\label{eq:taustar} |
299 |
\end{equation} |
300 |
where $\tau$ could be any prognostic variable ($u$, $v$, $\theta$ or |
301 |
$S$) and $\tau^*$ is an explicit estimate of $\tau^{n+1}$ and would be |
302 |
exact if not for implicit-in-time terms. The parenthesis about $n+1/2$ |
303 |
indicates that the term is explicit and extrapolated forward in time |
304 |
and for this we use the quasi-second order Adams-Bashforth method: |
305 |
\begin{equation} |
306 |
G_\tau^{(n+1/2)} = ( 3/2 + \epsilon_{AB}) G_\tau^n |
307 |
- ( 1/2 + \epsilon_{AB}) G_\tau^{n-1} |
308 |
\label{eq:adams-bashforth2} |
309 |
\end{equation} |
310 |
This is a linear extrapolation, forward in time, to |
311 |
$t=(n+1/2+{\epsilon_{AB}})\Delta t$. An extrapolation to the mid-point |
312 |
in time, $t=(n+1/2)\Delta t$, corresponding to $\epsilon_{AB}=0$, |
313 |
would be second order accurate but is weakly unstable for oscillatory |
314 |
terms. A small but finite value for $\epsilon_{AB}$ stabilizes the |
315 |
method. Strictly speaking, damping terms such as diffusion and |
316 |
dissipation, and fixed terms (forcing), do not need to be inside the |
317 |
Adams-Bashforth extrapolation. However, in the current code, it is |
318 |
simpler to include these terms and this can be justified if the flow |
319 |
and forcing evolves smoothly. Problems can, and do, arise when forcing |
320 |
or motions are high frequency and this corresponds to a reduced |
321 |
stability compared to a simple forward time-stepping of such terms. |
322 |
|
323 |
A stability analysis for an oscillation equation should be given at this point. |
324 |
\marginpar{AJA needs to find his notes on this...} |
325 |
|
326 |
A stability analysis for a relaxation equation should be given at this point. |
327 |
\marginpar{...and for this too.} |
328 |
|
329 |
|
330 |
\section{Implicit time-stepping: backward method} |
331 |
|
332 |
Vertical diffusion and viscosity can be treated implicitly in time |
333 |
using the backward method which is an intrinsic scheme. For tracers, |
334 |
the time discretized equation is: |
335 |
\begin{equation} |
336 |
\tau^{n+1} - \Delta t \partial_r \kappa_v \partial_r \tau^{n+1} = |
337 |
\tau^{n} + \Delta t G_\tau^{(n+1/2)} |
338 |
\label{eq:implicit-diffusion} |
339 |
\end{equation} |
340 |
where $G_\tau^{(n+1/2)}$ is the remaining explicit terms extrapolated |
341 |
using the Adams-Bashforth method as described above. Equation |
342 |
\ref{eq:implicit-diffusion} can be split split into: |
343 |
\begin{eqnarray} |
344 |
\tau^* & = & \tau^{n} + \Delta t G_\tau^{(n+1/2)} |
345 |
\label{eq:taustar-implicit} \\ |
346 |
\tau^{n+1} & = & {\cal L}_\tau^{-1} ( \tau^* ) |
347 |
\label{eq:tau-n+1-implicit} |
348 |
\end{eqnarray} |
349 |
where ${\cal L}_\tau^{-1}$ is the inverse of the operator |
350 |
\begin{equation} |
351 |
{\cal L} = \left[ 1 + \Delta t \partial_r \kappa_v \partial_r \right] |
352 |
\end{equation} |
353 |
Equation \ref{eq:taustar-implicit} looks exactly as \ref{eq:taustar} |
354 |
while \ref{eq:tau-n+1-implicit} involves an operator or matrix |
355 |
inversion. By re-arranging \ref{eq:implicit-diffusion} in this way we |
356 |
have cast the method as an explicit prediction step and an implicit |
357 |
step allowing the latter to be inserted into the over all algorithm |
358 |
with minimal interference. |
359 |
|
360 |
Fig.~\ref{fig:call-tree-adams-bashforth} shows the calling sequence for |
361 |
stepping forward a tracer variable such as temperature. |
362 |
|
363 |
In order to fit within the pressure method, the implicit viscosity |
364 |
must not alter the barotropic flow. In other words, it can on ly |
365 |
redistribute momentum in the vertical. The upshot of this is that |
366 |
although vertical viscosity may be backward implicit and |
367 |
unconditionally stable, no-slip boundary conditions may not be made |
368 |
implicit and are thus cast as a an explicit drag term. |
369 |
|
370 |
\section{Synchronous time-stepping: variables co-located in time} |
371 |
\label{sect:adams-bashforth-sync} |
372 |
|
373 |
\begin{figure} |
374 |
\begin{center} |
375 |
\resizebox{5.0in}{!}{\includegraphics{part2/adams-bashforth-sync.eps}} |
376 |
\end{center} |
377 |
\caption{ |
378 |
A schematic of the explicit Adams-Bashforth and implicit time-stepping |
379 |
phases of the algorithm. All prognostic variables are co-located in |
380 |
time. Explicit tendencies are evaluated at time level $n$ as a |
381 |
function of the state at that time level (dotted arrow). The explicit |
382 |
tendency from the previous time level, $n-1$, is used to extrapolate |
383 |
tendencies to $n+1/2$ (dashed arrow). This extrapolated tendency |
384 |
allows variables to be stably integrated forward-in-time to render an |
385 |
estimate ($*$-variables) at the $n+1$ time level (solid |
386 |
arc-arrow). The operator ${\cal L}$ formed from implicit-in-time terms |
387 |
is solved to yield the state variables at time level $n+1$. } |
388 |
\label{fig:adams-bashforth-sync} |
389 |
\end{figure} |
390 |
|
391 |
\begin{figure} |
392 |
\begin{center} \fbox{ \begin{minipage}{4.5in} \begin{tabbing} |
393 |
aaa \= aaa \= aaa \= aaa \= aaa \= aaa \kill |
394 |
FORWARD\_STEP \\ |
395 |
\> THERMODYNAMICS \\ |
396 |
\>\> CALC\_GT \\ |
397 |
\>\>\> GAD\_CALC\_RHS \` $G_\theta^n = G_\theta( u, \theta^n )$ (\ref{eq:Gt-n-sync})\\ |
398 |
\>\>\> EXTERNAL\_FORCING \` $G_\theta^n = G_\theta^n + {\cal Q}$ \\ |
399 |
\>\>\> ADAMS\_BASHFORTH2 \` $G_\theta^{(n+1/2)}$ (\ref{eq:Gt-n+5-sync}) \\ |
400 |
\>\> TIMESTEP\_TRACER \` $\tau^*$ (\ref{eq:tstar-sync}) \\ |
401 |
\>\> IMPLDIFF \` $\tau^{(n+1)}$ (\ref{eq:t-n+1-sync}) \\ |
402 |
\> DYNAMICS \\ |
403 |
\>\> CALC\_PHI\_HYD \` $\phi_{hyd}^n$ (\ref{eq:phi-hyd-sync}) \\ |
404 |
\>\> MOM\_FLUXFORM or MOM\_VECINV \` $G_{\vec{\bf v}}^n$ (\ref{eq:Gv-n-sync})\\ |
405 |
\>\> TIMESTEP \` $\vec{\bf v}^*$ (\ref{eq:Gv-n+5-sync}, \ref{eq:vstar-sync}) \\ |
406 |
\>\> IMPLDIFF \` $\vec{\bf v}^{**}$ (\ref{eq:vstarstar-sync}) \\ |
407 |
\> SOLVE\_FOR\_PRESSURE \\ |
408 |
\>\> CALC\_DIV\_GHAT \` $\eta^*$ (\ref{eq:nstar-sync}) \\ |
409 |
\>\> CG2D \` $\eta^{n+1}$ (\ref{eq:elliptic-sync}) \\ |
410 |
\> THE\_CORRECTION\_STEP \\ |
411 |
\>\> CALC\_GRAD\_PHI\_SURF \` $\nabla \eta^{n+1}$ \\ |
412 |
\>\> CORRECTION\_STEP \` $u^{n+1}$,$v^{n+1}$ (\ref{eq:v-n+1-sync}) |
413 |
\end{tabbing} \end{minipage} } \end{center} |
414 |
\caption{ |
415 |
Calling tree for the overall synchronous algorithm using |
416 |
Adams-Bashforth time-stepping.} |
417 |
\label{fig:call-tree-adams-bashforth-sync} |
418 |
\end{figure} |
419 |
|
420 |
The Adams-Bashforth extrapolation of explicit tendencies fits neatly |
421 |
into the pressure method algorithm when all state variables are |
422 |
co-located in time. Fig.~\ref{fig:adams-bashforth-sync} illustrates |
423 |
the location of variables in time and the evolution of the algorithm |
424 |
with time. The algorithm can be represented by the sequential solution |
425 |
of the follow equations: |
426 |
\begin{eqnarray} |
427 |
G_{\theta,S}^{n} & = & G_{\theta,S} ( u^n, \theta^n, S^n ) |
428 |
\label{eq:Gt-n-sync} \\ |
429 |
G_{\theta,S}^{(n+1/2)} & = & (3/2+\epsilon_{AB}) G_{\theta,S}^{n}-(1/2+\epsilon_{AB}) G_{\theta,S}^{n-1} |
430 |
\label{eq:Gt-n+5-sync} \\ |
431 |
(\theta^*,S^*) & = & (\theta^{n},S^{n}) + \Delta t G_{\theta,S}^{(n+1/2)} |
432 |
\label{eq:tstar-sync} \\ |
433 |
(\theta^{n+1},S^{n+1}) & = & {\cal L}^{-1}_{\theta,S} (\theta^*,S^*) |
434 |
\label{eq:t-n+1-sync} \\ |
435 |
\phi^n_{hyd} & = & \int b(\theta^n,S^n) dr |
436 |
\label{eq:phi-hyd-sync} \\ |
437 |
\vec{\bf G}_{\vec{\bf v}}^{n} & = & \vec{\bf G}_{\vec{\bf v}} ( \vec{\bf v}^n, \phi^n_{hyd} ) |
438 |
\label{eq:Gv-n-sync} \\ |
439 |
\vec{\bf G}_{\vec{\bf v}}^{(n+1/2)} & = & (3/2 + \epsilon_{AB} ) \vec{\bf G}_{\vec{\bf v}}^{n} - (1/2 + \epsilon_{AB} ) \vec{\bf G}_{\vec{\bf v}}^{n-1} |
440 |
\label{eq:Gv-n+5-sync} \\ |
441 |
\vec{\bf v}^{*} & = & \vec{\bf v}^{n} + \Delta t \vec{\bf G}_{\vec{\bf v}}^{(n+1/2)} |
442 |
\label{eq:vstar-sync} \\ |
443 |
\vec{\bf v}^{**} & = & {\cal L}_{\vec{\bf v}}^{-1} ( \vec{\bf v}^* ) |
444 |
\label{eq:vstarstar-sync} \\ |
445 |
\eta^* & = & \epsilon_{fs} \left( \eta^{n} +P-E+R \right)- \Delta t |
446 |
\nabla \cdot H \widehat{ \vec{\bf v}^{**} } |
447 |
\label{eq:nstar-sync} \\ |
448 |
\nabla \cdot g H \nabla \eta^{n+1} - \frac{\epsilon_{fs} \eta^{n+1}}{\Delta t^2} |
449 |
& = & - \frac{\eta^*}{\Delta t^2} |
450 |
\label{eq:elliptic-sync} \\ |
451 |
\vec{\bf v}^{n+1} & = & \vec{\bf v}^{*} - \Delta t g \nabla \eta^{n+1} |
452 |
\label{eq:v-n+1-sync} |
453 |
\end{eqnarray} |
454 |
Fig.~\ref{fig:adams-bashforth-sync} illustrates the location of |
455 |
variables in time and evolution of the algorithm with time. The |
456 |
Adams-Bashforth extrapolation of the tracer tendencies is illustrated |
457 |
by the dashed arrow, the prediction at $n+1$ is indicated by the |
458 |
solid arc. Inversion of the implicit terms, ${\cal |
459 |
L}^{-1}_{\theta,S}$, then yields the new tracer fields at $n+1$. All |
460 |
these operations are carried out in subroutine {\em THERMODYNAMICS} an |
461 |
subsidiaries, which correspond to equations \ref{eq:Gt-n-sync} to |
462 |
\ref{eq:t-n+1-sync}. |
463 |
Similarly illustrated is the Adams-Bashforth extrapolation of |
464 |
accelerations, stepping forward and solving of implicit viscosity and |
465 |
surface pressure gradient terms, corresponding to equations |
466 |
\ref{eq:Gv-n-sync} to \ref{eq:v-n+1-sync}. |
467 |
These operations are carried out in subroutines {\em DYNAMCIS}, {\em |
468 |
SOLVE\_FOR\_PRESSURE} and {\em THE\_CORRECTION\_STEP}. This, then, |
469 |
represents an entire algorithm for stepping forward the model one |
470 |
time-step. The corresponding calling tree is given in |
471 |
\ref{fig:call-tree-adams-bashforth-sync}. |
472 |
|
473 |
\section{Staggered baroclinic time-stepping} |
474 |
\label{sect:adams-bashforth-staggered} |
475 |
|
476 |
\begin{figure} |
477 |
\begin{center} |
478 |
\resizebox{5.5in}{!}{\includegraphics{part2/adams-bashforth-staggered.eps}} |
479 |
\end{center} |
480 |
\caption{ |
481 |
A schematic of the explicit Adams-Bashforth and implicit time-stepping |
482 |
phases of the algorithm but with staggering in time of thermodynamic |
483 |
variables with the flow. Explicit thermodynamics tendencies are |
484 |
evaluated at time level $n-1/2$ as a function of the thermodynamics |
485 |
state at that time level $n$ and flow at time $n$ (dotted arrow). The |
486 |
explicit tendency from the previous time level, $n-3/2$, is used to |
487 |
extrapolate tendencies to $n$ (dashed arrow). This extrapolated |
488 |
tendency allows thermo-dynamics variables to be stably integrated |
489 |
forward-in-time to render an estimate ($*$-variables) at the $n+1/2$ |
490 |
time level (solid arc-arrow). The implicit-in-time operator ${\cal |
491 |
L_{\theta,S}}$ is solved to yield the thermodynamic variables at time |
492 |
level $n+1/2$. These are then used to calculate the hydrostatic |
493 |
pressure/geo-potential, $\phi_{hyd}$ (vertical arrows). The |
494 |
hydrostatic pressure gradient is evaluated directly an time level |
495 |
$n+1/2$ in stepping forward the flow variables from $n$ to $n+1$ |
496 |
(solid arc-arrow). } |
497 |
\label{fig:adams-bashforth-staggered} |
498 |
\end{figure} |
499 |
|
500 |
For well stratified problems, internal gravity waves may be the |
501 |
limiting process for determining a stable time-step. In the |
502 |
circumstance, it is more efficient to stagger in time the |
503 |
thermodynamic variables with the flow |
504 |
variables. Fig.~\ref{fig:adams-bashforth-staggered} illustrates the |
505 |
staggering and algorithm. The key difference between this and |
506 |
Fig.~\ref{fig:adams-bashforth-sync} is that the new thermodynamics |
507 |
fields are used to compute the hydrostatic pressure at time level |
508 |
$n+1/2$. The essentially allows the gravity wave terms to leap-frog in |
509 |
time giving second order accuracy and more stability. |
510 |
|
511 |
The essential change in the staggered algorithm is the calculation of |
512 |
hydrostatic pressure which, in the context of the synchronous |
513 |
algorithm involves replacing equation \ref{eq:phi-hyd-sync} with |
514 |
\begin{displaymath} |
515 |
\phi_{hyd}^n = \int b(\theta^{n+1},S^{n+1}) dr |
516 |
\end{displaymath} |
517 |
but the pressure gradient must also be taken out of the |
518 |
Adams-Bashforth extrapolation. Also, retaining the integer time-levels, |
519 |
$n$ and $n+1$, does not give a user the sense of where variables are |
520 |
located in time. Instead, we re-write the entire algorithm, |
521 |
\ref{eq:Gt-n-sync} to \ref{eq:v-n+1-sync}, annotating the |
522 |
position in time of variables appropriately: |
523 |
\begin{eqnarray} |
524 |
G_{\theta,S}^{n-1/2} & = & G_{\theta,S} ( u^n, \theta^{n-1/2}, S^{n-1/2} ) |
525 |
\label{eq:Gt-n-staggered} \\ |
526 |
G_{\theta,S}^{(n)} & = & (3/2+\epsilon_{AB}) G_{\theta,S}^{n-1/2}-(1/2+\epsilon_{AB}) G_{\theta,S}^{n-3/2} |
527 |
\label{eq:Gt-n+5-staggered} \\ |
528 |
(\theta^*,S^*) & = & (\theta^{n},S^{n}) + \Delta t G_{\theta,S}^{(n)} |
529 |
\label{eq:tstar-staggered} \\ |
530 |
(\theta^{n+1/2},S^{n+1/2}) & = & {\cal L}^{-1}_{\theta,S} (\theta^*,S^*) |
531 |
\label{eq:t-n+1-staggered} \\ |
532 |
\phi^{n+1/2}_{hyd} & = & \int b(\theta^{n+1/2},S^{n+1/2}) dr |
533 |
\label{eq:phi-hyd-staggered} \\ |
534 |
\vec{\bf G}_{\vec{\bf v}}^{n} & = & \vec{\bf G}_{\vec{\bf v}} ( \vec{\bf v}^n ) |
535 |
\label{eq:Gv-n-staggered} \\ |
536 |
\vec{\bf G}_{\vec{\bf v}}^{(n+1/2)} & = & (3/2 + \epsilon_{AB} ) \vec{\bf G}_{\vec{\bf v}}^{n} - (1/2 + \epsilon_{AB} ) \vec{\bf G}_{\vec{\bf v}}^{n-1} |
537 |
\label{eq:Gv-n+5-staggered} \\ |
538 |
\vec{\bf v}^{*} & = & \vec{\bf v}^{n} + \Delta t \left( \vec{\bf G}_{\vec{\bf v}}^{(n+1/2)} - \nabla \phi_{hyd}^{n+1/2} \right) |
539 |
\label{eq:vstar-staggered} \\ |
540 |
\vec{\bf v}^{**} & = & {\cal L}_{\vec{\bf v}}^{-1} ( \vec{\bf v}^* ) |
541 |
\label{eq:vstarstar-staggered} \\ |
542 |
\eta^* & = & \epsilon_{fs} \left( \eta^{n} +P-E+R \right)- \Delta t |
543 |
\nabla \cdot H \widehat{ \vec{\bf v}^{**} } |
544 |
\label{eq:nstar-staggered} \\ |
545 |
\nabla \cdot g H \nabla \eta^{n+1} - \frac{\epsilon_{fs} \eta^{n+1}}{\Delta t^2} |
546 |
& = & - \frac{\eta^*}{\Delta t^2} |
547 |
\label{eq:elliptic-staggered} \\ |
548 |
\vec{\bf v}^{n+1} & = & \vec{\bf v}^{*} - \Delta t g \nabla \eta^{n+1} |
549 |
\label{eq:v-n+1-staggered} |
550 |
\end{eqnarray} |
551 |
The calling sequence is unchanged from |
552 |
Fig.~\ref{fig:call-tree-adams-bashforth-sync}. The staggered algorithm |
553 |
is activated with the run-time flag {\bf staggerTimeStep=.TRUE.} in |
554 |
{\em PARM01} of {\em data}. |
555 |
|
556 |
The only difficulty with this approach is apparent in equation |
557 |
\ref{eq:Gt-n-staggered} and illustrated by the dotted arrow |
558 |
connecting $u,v^n$ with $G_\theta^{n-1/2}$. The flow used to advect |
559 |
tracers around is not naturally located in time. This could be avoided |
560 |
by applying the Adams-Bashforth extrapolation to the tracer field |
561 |
itself and advection that around but this is not yet available. We're |
562 |
not aware of any detrimental effect of this feature. The difficulty |
563 |
lies mainly in interpretation of what time-level variables and terms |
564 |
correspond to. |
565 |
|
566 |
|
567 |
\section{Non-hydrostatic formulation} |
568 |
\label{sect:non-hydrostatic} |
569 |
|
570 |
[to be written...] |
571 |
|
572 |
|
573 |
|
574 |
|
575 |
\section{Variants on the Free Surface} |
576 |
|
577 |
We now describe the various formulations of the free-surface that |
578 |
include non-linear forms, implicit in time using Crank-Nicholson, |
579 |
explicit and [one day] split-explicit. First, we'll reiterate the |
580 |
underlying algorithm but this time using the notation consistent with |
581 |
the more general vertical coordinate $r$. The elliptic equation for |
582 |
free-surface coordinate (units of $r$), corresponding to |
583 |
\ref{eq:discrete-time-backward-free-surface}, and |
584 |
assuming no non-hydrostatic effects ($\epsilon_{nh} = 0$) is: |
585 |
\begin{eqnarray} |
586 |
\epsilon_{fs} {\eta}^{n+1} - |
587 |
{\bf \nabla}_h \cdot \Delta t^2 (R_o-R_{fixed}) {\bf \nabla}_h b_s |
588 |
{\eta}^{n+1} = {\eta}^* |
589 |
\label{eq-solve2D} |
590 |
\end{eqnarray} |
591 |
where |
592 |
\begin{eqnarray} |
593 |
{\eta}^* = \epsilon_{fs} \: {\eta}^{n} - |
594 |
\Delta t {\bf \nabla}_h \cdot \int_{R_{fixed}}^{R_o} \vec{\bf v}^* dr |
595 |
\: + \: \epsilon_{fw} \Delta_t (P-E)^{n} |
596 |
\label{eq-solve2D_rhs} |
597 |
\end{eqnarray} |
598 |
|
599 |
\fbox{ \begin{minipage}{4.75in} |
600 |
{\em S/R SOLVE\_FOR\_PRESSURE} ({\em solve\_for\_pressure.F}) |
601 |
|
602 |
$u^*$: {\bf GuNm1} ({\em DYNVARS.h}) |
603 |
|
604 |
$v^*$: {\bf GvNm1} ({\em DYNVARS.h}) |
605 |
|
606 |
$\eta^*$: {\bf cg2d\_b} (\em SOLVE\_FOR\_PRESSURE.h) |
607 |
|
608 |
$\eta^{n+1}$: {\bf etaN} (\em DYNVARS.h) |
609 |
|
610 |
\end{minipage} } |
611 |
|
612 |
|
613 |
Once ${\eta}^{n+1}$ has been found, substituting into |
614 |
\ref{eq-tDsC-Hmom} yields $\vec{\bf v}^{n+1}$ if the model is |
615 |
hydrostatic ($\epsilon_{nh}=0$): |
616 |
$$ |
617 |
\vec{\bf v}^{n+1} = \vec{\bf v}^{*} |
618 |
- \Delta t {\bf \nabla}_h b_s {\eta}^{n+1} |
619 |
$$ |
620 |
|
621 |
This is known as the correction step. However, when the model is |
622 |
non-hydrostatic ($\epsilon_{nh}=1$) we need an additional step and an |
623 |
additional equation for $\phi'_{nh}$. This is obtained by substituting |
624 |
\ref{eq-tDsC-Hmom} and \ref{eq-tDsC-Vmom} into |
625 |
\ref{eq-tDsC-cont}: |
626 |
\begin{equation} |
627 |
\left[ {\bf \nabla}_h^2 + \partial_{rr} \right] {\phi'_{nh}}^{n+1} |
628 |
= \frac{1}{\Delta t} \left( |
629 |
{\bf \nabla}_h \cdot \vec{\bf v}^{**} + \partial_r \dot{r}^* \right) |
630 |
\end{equation} |
631 |
where |
632 |
\begin{displaymath} |
633 |
\vec{\bf v}^{**} = \vec{\bf v}^* - \Delta t {\bf \nabla}_h b_s {\eta}^{n+1} |
634 |
\end{displaymath} |
635 |
Note that $\eta^{n+1}$ is also used to update the second RHS term |
636 |
$\partial_r \dot{r}^* $ since |
637 |
the vertical velocity at the surface ($\dot{r}_{surf}$) |
638 |
is evaluated as $(\eta^{n+1} - \eta^n) / \Delta t$. |
639 |
|
640 |
Finally, the horizontal velocities at the new time level are found by: |
641 |
\begin{equation} |
642 |
\vec{\bf v}^{n+1} = \vec{\bf v}^{**} |
643 |
- \epsilon_{nh} \Delta t {\bf \nabla}_h {\phi'_{nh}}^{n+1} |
644 |
\end{equation} |
645 |
and the vertical velocity is found by integrating the continuity |
646 |
equation vertically. Note that, for the convenience of the restart |
647 |
procedure, the vertical integration of the continuity equation has |
648 |
been moved to the beginning of the time step (instead of at the end), |
649 |
without any consequence on the solution. |
650 |
|
651 |
\fbox{ \begin{minipage}{4.75in} |
652 |
{\em S/R CORRECTION\_STEP} ({\em correction\_step.F}) |
653 |
|
654 |
$\eta^{n+1}$: {\bf etaN} (\em DYNVARS.h) |
655 |
|
656 |
$\phi_{nh}^{n+1}$: {\bf phi\_nh} (\em DYNVARS.h) |
657 |
|
658 |
$u^*$: {\bf GuNm1} ({\em DYNVARS.h}) |
659 |
|
660 |
$v^*$: {\bf GvNm1} ({\em DYNVARS.h}) |
661 |
|
662 |
$u^{n+1}$: {\bf uVel} ({\em DYNVARS.h}) |
663 |
|
664 |
$v^{n+1}$: {\bf vVel} ({\em DYNVARS.h}) |
665 |
|
666 |
\end{minipage} } |
667 |
|
668 |
|
669 |
|
670 |
Regarding the implementation of the surface pressure solver, all |
671 |
computation are done within the routine {\it SOLVE\_FOR\_PRESSURE} and |
672 |
its dependent calls. The standard method to solve the 2D elliptic |
673 |
problem (\ref{eq-solve2D}) uses the conjugate gradient method (routine |
674 |
{\it CG2D}); the solver matrix and conjugate gradient operator are |
675 |
only function of the discretized domain and are therefore evaluated |
676 |
separately, before the time iteration loop, within {\it INI\_CG2D}. |
677 |
The computation of the RHS $\eta^*$ is partly done in {\it |
678 |
CALC\_DIV\_GHAT} and in {\it SOLVE\_FOR\_PRESSURE}. |
679 |
|
680 |
The same method is applied for the non hydrostatic part, using a |
681 |
conjugate gradient 3D solver ({\it CG3D}) that is initialized in {\it |
682 |
INI\_CG3D}. The RHS terms of 2D and 3D problems are computed together |
683 |
at the same point in the code. |
684 |
|
685 |
|
686 |
|
687 |
\subsection{Crank-Nickelson barotropic time stepping} |
688 |
|
689 |
The full implicit time stepping described previously is |
690 |
unconditionally stable but damps the fast gravity waves, resulting in |
691 |
a loss of potential energy. The modification presented now allows one |
692 |
to combine an implicit part ($\beta,\gamma$) and an explicit part |
693 |
($1-\beta,1-\gamma$) for the surface pressure gradient ($\beta$) and |
694 |
for the barotropic flow divergence ($\gamma$). |
695 |
\\ |
696 |
For instance, $\beta=\gamma=1$ is the previous fully implicit scheme; |
697 |
$\beta=\gamma=1/2$ is the non damping (energy conserving), unconditionally |
698 |
stable, Crank-Nickelson scheme; $(\beta,\gamma)=(1,0)$ or $=(0,1)$ |
699 |
corresponds to the forward - backward scheme that conserves energy but is |
700 |
only stable for small time steps.\\ |
701 |
In the code, $\beta,\gamma$ are defined as parameters, respectively |
702 |
{\it implicSurfPress}, {\it implicDiv2DFlow}. They are read from |
703 |
the main data file "{\it data}" and are set by default to 1,1. |
704 |
|
705 |
Equations \ref{eq-tDsC-Hmom} and \ref{eq-tDsC-eta} are modified as follows: |
706 |
$$ |
707 |
\frac{ \vec{\bf v}^{n+1} }{ \Delta t } |
708 |
+ {\bf \nabla}_h b_s [ \beta {\eta}^{n+1} + (1-\beta) {\eta}^{n} ] |
709 |
+ \epsilon_{nh} {\bf \nabla}_h {\phi'_{nh}}^{n+1} |
710 |
= \frac{ \vec{\bf v}^* }{ \Delta t } |
711 |
$$ |
712 |
$$ |
713 |
\epsilon_{fs} \frac{ {\eta}^{n+1} - {\eta}^{n} }{ \Delta t} |
714 |
+ {\bf \nabla}_h \cdot \int_{R_{fixed}}^{R_o} |
715 |
[ \gamma \vec{\bf v}^{n+1} + (1-\gamma) \vec{\bf v}^{n}] dr |
716 |
= \epsilon_{fw} (P-E) |
717 |
$$ |
718 |
where: |
719 |
\begin{eqnarray*} |
720 |
\vec{\bf v}^* & = & |
721 |
\vec{\bf v} ^{n} + \Delta t \vec{\bf G}_{\vec{\bf v}} ^{(n+1/2)} |
722 |
+ (\beta-1) \Delta t {\bf \nabla}_h b_s {\eta}^{n} |
723 |
+ \Delta t {\bf \nabla}_h {\phi'_{hyd}}^{(n+1/2)} |
724 |
\\ |
725 |
{\eta}^* & = & |
726 |
\epsilon_{fs} {\eta}^{n} + \epsilon_{fw} \Delta t (P-E) |
727 |
- \Delta t {\bf \nabla}_h \cdot \int_{R_{fixed}}^{R_o} |
728 |
[ \gamma \vec{\bf v}^* + (1-\gamma) \vec{\bf v}^{n}] dr |
729 |
\end{eqnarray*} |
730 |
\\ |
731 |
In the hydrostatic case ($\epsilon_{nh}=0$), allowing us to find |
732 |
${\eta}^{n+1}$, thus: |
733 |
$$ |
734 |
\epsilon_{fs} {\eta}^{n+1} - |
735 |
{\bf \nabla}_h \cdot \beta\gamma \Delta t^2 b_s (R_o - R_{fixed}) |
736 |
{\bf \nabla}_h {\eta}^{n+1} |
737 |
= {\eta}^* |
738 |
$$ |
739 |
and then to compute (correction step): |
740 |
$$ |
741 |
\vec{\bf v}^{n+1} = \vec{\bf v}^{*} |
742 |
- \beta \Delta t {\bf \nabla}_h b_s {\eta}^{n+1} |
743 |
$$ |
744 |
|
745 |
The non-hydrostatic part is solved as described previously. |
746 |
|
747 |
Note that: |
748 |
\begin{enumerate} |
749 |
\item The non-hydrostatic part of the code has not yet been |
750 |
updated, so that this option cannot be used with $(\beta,\gamma) \neq (1,1)$. |
751 |
\item The stability criteria with Crank-Nickelson time stepping |
752 |
for the pure linear gravity wave problem in cartesian coordinates is: |
753 |
\begin{itemize} |
754 |
\item $\beta + \gamma < 1$ : unstable |
755 |
\item $\beta \geq 1/2$ and $ \gamma \geq 1/2$ : stable |
756 |
\item $\beta + \gamma \geq 1$ : stable if |
757 |
$$ |
758 |
c_{max}^2 (\beta - 1/2)(\gamma - 1/2) + 1 \geq 0 |
759 |
$$ |
760 |
$$ |
761 |
\mbox{with }~ |
762 |
%c^2 = 2 g H {\Delta t}^2 |
763 |
%(\frac{1-cos 2 \pi / k}{\Delta x^2} |
764 |
%+\frac{1-cos 2 \pi / l}{\Delta y^2}) |
765 |
%$$ |
766 |
%Practically, the most stringent condition is obtained with $k=l=2$ : |
767 |
%$$ |
768 |
c_{max} = 2 \Delta t \: \sqrt{g H} \: |
769 |
\sqrt{ \frac{1}{\Delta x^2} + \frac{1}{\Delta y^2} } |
770 |
$$ |
771 |
\end{itemize} |
772 |
\end{enumerate} |
773 |
|