/[MITgcm]/manual/s_algorithm/text/tracer.tex
ViewVC logotype

Annotation of /manual/s_algorithm/text/tracer.tex

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


Revision 1.8 - (hide annotations) (download) (as text)
Thu Oct 25 18:36:53 2001 UTC (23 years, 8 months ago) by cnh
Branch: MAIN
Changes since 1.7: +36 -36 lines
File MIME type: application/x-tex
We can't spell

1 cnh 1.8 % $Header: /u/u0/gcmpack/mitgcmdoc/part2/tracer.tex,v 1.7 2001/10/24 23:14:44 cnh Exp $
2 adcroft 1.2 % $Name: $
3 adcroft 1.1
4     \section{Tracer equations}
5 cnh 1.6 \label{sec:tracer_equations}
6 adcroft 1.1
7 adcroft 1.2 The basic discretization used for the tracer equations is the second
8     order piece-wise constant finite volume form of the forced
9 cnh 1.8 advection-diffusion equations. There are many alternatives to second
10 adcroft 1.2 order method for advection and alternative parameterizations for the
11     sub-grid scale processes. The Gent-McWilliams eddy parameterization,
12     KPP mixing scheme and PV flux parameterization are all dealt with in
13     separate sections. The basic discretization of the advection-diffusion
14     part of the tracer equations and the various advection schemes will be
15     described here.
16    
17 adcroft 1.3 \subsection{Time-stepping of tracers: ABII}
18 cnh 1.7 \label{sec:tracer_equations_abII}
19 adcroft 1.3
20     The default advection scheme is the centered second order method which
21     requires a second order or quasi-second order time-stepping scheme to
22     be stable. Historically this has been the quasi-second order
23     Adams-Bashforth method (ABII) and applied to all terms. For an
24     arbitrary tracer, $\tau$, the forced advection-diffusion equation
25     reads:
26     \begin{equation}
27     \partial_t \tau + G_{adv}^\tau = G_{diff}^\tau + G_{forc}^\tau
28     \end{equation}
29     where $G_{adv}^\tau$, $G_{diff}^\tau$ and $G_{forc}^\tau$ are the
30     tendencies due to advection, diffusion and forcing, respectively,
31     namely:
32     \begin{eqnarray}
33     G_{adv}^\tau & = & \partial_x u \tau + \partial_y v \tau + \partial_r w \tau
34     - \tau \nabla \cdot {\bf v} \\
35     G_{diff}^\tau & = & \nabla \cdot {\bf K} \nabla \tau
36     \end{eqnarray}
37     and the forcing can be some arbitrary function of state, time and
38     space.
39    
40     The term, $\tau \nabla \cdot {\bf v}$, is required to retain local
41     conservation in conjunction with the linear implicit free-surface. It
42     only affects the surface layer since the flow is non-divergent
43     everywhere else. This term is therefore referred to as the surface
44     correction term. Global conservation is not possible using the
45     flux-form (as here) and a linearized free-surface
46     (\cite{Griffies00,Campin02}).
47    
48     The continuity equation can be recovered by setting
49     $G_{diff}=G_{forc}=0$ and $\tau=1$.
50    
51 cnh 1.8 The driver routine that calls the routines to calculate tendencies are
52 adcroft 1.3 {\em S/R CALC\_GT} and {\em S/R CALC\_GS} for temperature and salt
53     (moisture), respectively. These in turn call a generic advection
54     diffusion routine {\em S/R GAD\_CALC\_RHS} that is called with the
55 cnh 1.8 flow field and relevant tracer as arguments and returns the collective
56     tendency due to advection and diffusion. Forcing is add subsequently
57     in {\em S/R CALC\_GT} or {\em S/R CALC\_GS} to the same tendency
58 adcroft 1.3 array.
59    
60     \fbox{ \begin{minipage}{4.75in}
61     {\em S/R GAD\_CALC\_RHS} ({\em pkg/generic\_advdiff/gad\_calc\_rhs.F})
62    
63     $\tau$: {\bf tracer} (argument)
64    
65     $G^{(n)}$: {\bf gTracer} (argument)
66    
67     $F_r$: {\bf fVerT} (argument)
68    
69     \end{minipage} }
70    
71 cnh 1.8 The space and time discretization are treated separately (method of
72     lines). Tendencies are calculated at time levels $n$ and $n-1$ and
73 adcroft 1.4 extrapolated to $n+1/2$ using the Adams-Bashforth method:
74 adcroft 1.3 \marginpar{$\epsilon$: {\bf AB\_eps}}
75     \begin{equation}
76 adcroft 1.4 G^{(n+1/2)} =
77 adcroft 1.3 (\frac{3}{2} + \epsilon) G^{(n)} - (\frac{1}{2} + \epsilon) G^{(n-1)}
78     \end{equation}
79     where $G^{(n)} = G_{adv}^\tau + G_{diff}^\tau + G_{src}^\tau$ at time
80 cnh 1.8 step $n$. The tendency at $n-1$ is not re-calculated but rather the
81     tendency at $n$ is stored in a global array for later re-use.
82 adcroft 1.4
83     \fbox{ \begin{minipage}{4.75in}
84     {\em S/R ADAMS\_BASHFORTH2} ({\em model/src/adams\_bashforth2.F})
85    
86     $G^{(n+1/2)}$: {\bf gTracer} (argument on exit)
87    
88     $G^{(n)}$: {\bf gTracer} (argument on entry)
89    
90     $G^{(n-1)}$: {\bf gTrNm1} (argument)
91    
92     $\epsilon$: {\bf ABeps} (PARAMS.h)
93    
94     \end{minipage} }
95    
96 cnh 1.8 The tracers are stepped forward in time using the extrapolated tendency:
97 adcroft 1.4 \begin{equation}
98     \tau^{(n+1)} = \tau^{(n)} + \Delta t G^{(n+1/2)}
99     \end{equation}
100     \marginpar{$\Delta t$: {\bf deltaTtracer}}
101    
102     \fbox{ \begin{minipage}{4.75in}
103     {\em S/R TIMESTEP\_TRACER} ({\em model/src/timestep\_tracer.F})
104    
105     $\tau^{(n+1)}$: {\bf gTracer} (argument on exit)
106    
107     $\tau^{(n)}$: {\bf tracer} (argument on entry)
108    
109     $G^{(n+1/2)}$: {\bf gTracer} (argument)
110    
111     $\Delta t$: {\bf deltaTtracer} (PARAMS.h)
112    
113     \end{minipage} }
114 adcroft 1.3
115     Strictly speaking the ABII scheme should be applied only to the
116 cnh 1.8 advection terms. However, this scheme is only used in conjunction with
117 adcroft 1.3 the standard second, third and fourth order advection
118     schemes. Selection of any other advection scheme disables
119     Adams-Bashforth for tracers so that explicit diffusion and forcing use
120     the forward method.
121    
122    
123    
124    
125 adcroft 1.4 \section{Linear advection schemes}
126 adcroft 1.3
127     \begin{figure}
128     \resizebox{5.5in}{!}{\includegraphics{part2/advect-1d-lo.eps}}
129     \caption{
130     Comparison of 1-D advection schemes. Courant number is 0.05 with 60
131     points and solutions are shown for T=1 (one complete period).
132     a) Shows the upwind biased schemes; first order upwind, DST3,
133     third order upwind and second order upwind.
134     b) Shows the centered schemes; Lax-Wendroff, DST4, centered second order,
135     centered fourth order and finite volume fourth order.
136     c) Shows the second order flux limiters: minmod, Superbee,
137     MC limiter and the van Leer limiter.
138     d) Shows the DST3 method with flux limiters due to Sweby with
139     $\mu=1$, $\mu=c/(1-c)$ and a fourth order DST method with Sweby limiter,
140     $\mu=c/(1-c)$.
141     \label{fig:advect-1d-lo}
142     }
143     \end{figure}
144    
145     \begin{figure}
146     \resizebox{5.5in}{!}{\includegraphics{part2/advect-1d-hi.eps}}
147     \caption{
148     Comparison of 1-D advection schemes. Courant number is 0.89 with 60
149     points and solutions are shown for T=1 (one complete period).
150     a) Shows the upwind biased schemes; first order upwind and DST3.
151     Third order upwind and second order upwind are unstable at this Courant number.
152     b) Shows the centered schemes; Lax-Wendroff, DST4. Centered second order,
153     centered fourth order and finite volume fourth order and unstable at this
154     Courant number.
155     c) Shows the second order flux limiters: minmod, Superbee,
156     MC limiter and the van Leer limiter.
157     d) Shows the DST3 method with flux limiters due to Sweby with
158     $\mu=1$, $\mu=c/(1-c)$ and a fourth order DST method with Sweby limiter,
159     $\mu=c/(1-c)$.
160     \label{fig:advect-1d-hi}
161     }
162     \end{figure}
163    
164     The advection schemes known as centered second order, centered fourth
165     order, first order upwind and upwind biased third order are known as
166     linear advection schemes because the coefficient for interpolation of
167     the advected tracer are linear and a function only of the flow, not
168     the tracer field it self. We discuss these first since they are most
169     commonly used in the field and most familiar.
170    
171 adcroft 1.2 \subsection{Centered second order advection-diffusion}
172    
173     The basic discretization, centered second order, is the default. It is
174 cnh 1.8 designed to be consistent with the continuity equation to facilitate
175 adcroft 1.3 conservation properties analogous to the continuum. However, centered
176 cnh 1.8 second order advection is notoriously noisy and must be used in
177     conjunction with some finite amount of diffusion to produce a sensible
178 adcroft 1.3 solution.
179    
180     The advection operator is discretized:
181 adcroft 1.1 \begin{equation}
182 adcroft 1.3 {\cal A}_c \Delta r_f h_c G_{adv}^\tau =
183     \delta_i F_x + \delta_j F_y + \delta_k F_r
184 adcroft 1.1 \end{equation}
185 adcroft 1.2 where the area integrated fluxes are given by:
186     \begin{eqnarray}
187 adcroft 1.3 F_x & = & U \overline{ \tau }^i \\
188     F_y & = & V \overline{ \tau }^j \\
189     F_r & = & W \overline{ \tau }^k
190 adcroft 1.2 \end{eqnarray}
191 adcroft 1.1 The quantities $U$, $V$ and $W$ are volume fluxes defined:
192     \marginpar{$U$: {\bf uTrans} }
193     \marginpar{$V$: {\bf vTrans} }
194     \marginpar{$W$: {\bf rTrans} }
195     \begin{eqnarray}
196     U & = & \Delta y_g \Delta r_f h_w u \\
197     V & = & \Delta x_g \Delta r_f h_s v \\
198     W & = & {\cal A}_c w
199     \end{eqnarray}
200    
201 adcroft 1.3 For non-divergent flow, this discretization can be shown to conserve
202     the tracer both locally and globally and to globally conserve tracer
203     variance, $\tau^2$. The proof is given in \cite{Adcroft95,Adcroft97}.
204    
205     \fbox{ \begin{minipage}{4.75in}
206     {\em S/R GAD\_C2\_ADV\_X} ({\em gad\_c2\_adv\_x.F})
207    
208     $F_x$: {\bf uT} (argument)
209    
210     $U$: {\bf uTrans} (argument)
211    
212     $\tau$: {\bf tracer} (argument)
213    
214     {\em S/R GAD\_C2\_ADV\_Y} ({\em gad\_c2\_adv\_y.F})
215    
216     $F_y$: {\bf vT} (argument)
217    
218     $V$: {\bf vTrans} (argument)
219    
220     $\tau$: {\bf tracer} (argument)
221    
222     {\em S/R GAD\_C2\_ADV\_R} ({\em gad\_c2\_adv\_r.F})
223    
224     $F_r$: {\bf wT} (argument)
225    
226     $W$: {\bf rTrans} (argument)
227    
228     $\tau$: {\bf tracer} (argument)
229    
230     \end{minipage} }
231    
232    
233     \subsection{Third order upwind bias advection}
234    
235     Upwind biased third order advection offers a relatively good
236     compromise between accuracy and smoothness. It is not a ``positive''
237     scheme meaning false extrema are permitted but the amplitude of such
238     are significantly reduced over the centered second order method.
239    
240     The third order upwind fluxes are discretized:
241     \begin{eqnarray}
242     F_x & = & U \overline{\tau - \frac{1}{6} \delta_{ii} \tau}^i
243     + \frac{1}{2} |U| \delta_i \frac{1}{6} \delta_{ii} \tau \\
244     F_y & = & V \overline{\tau - \frac{1}{6} \delta_{ii} \tau}^j
245     + \frac{1}{2} |V| \delta_j \frac{1}{6} \delta_{jj} \tau \\
246     F_r & = & W \overline{\tau - \frac{1}{6} \delta_{ii} \tau}^k
247     + \frac{1}{2} |W| \delta_k \frac{1}{6} \delta_{kk} \tau
248     \end{eqnarray}
249    
250     At boundaries, $\delta_{\hat{n}} \tau$ is set to zero allowing
251 cnh 1.8 $\delta_{nn}$ to be evaluated. We are currently examine the accuracy
252 adcroft 1.3 of this boundary condition and the effect on the solution.
253    
254 adcroft 1.4 \fbox{ \begin{minipage}{4.75in}
255     {\em S/R GAD\_U3\_ADV\_X} ({\em gad\_u3\_adv\_x.F})
256    
257     $F_x$: {\bf uT} (argument)
258    
259     $U$: {\bf uTrans} (argument)
260    
261     $\tau$: {\bf tracer} (argument)
262    
263     {\em S/R GAD\_U3\_ADV\_Y} ({\em gad\_u3\_adv\_y.F})
264    
265     $F_y$: {\bf vT} (argument)
266    
267     $V$: {\bf vTrans} (argument)
268    
269     $\tau$: {\bf tracer} (argument)
270    
271     {\em S/R GAD\_U3\_ADV\_R} ({\em gad\_u3\_adv\_r.F})
272    
273     $F_r$: {\bf wT} (argument)
274    
275     $W$: {\bf rTrans} (argument)
276    
277     $\tau$: {\bf tracer} (argument)
278    
279     \end{minipage} }
280 adcroft 1.3
281     \subsection{Centered fourth order advection}
282    
283     Centered fourth order advection is formally the most accurate scheme
284     we have implemented and can be used to great effect in high resolution
285 cnh 1.8 simulation where dynamical scales are well resolved. However, the
286     scheme is noisy like the centered second order method and so must be
287 adcroft 1.3 used with some finite amount of diffusion. Bi-harmonic is recommended
288     since it is more scale selective and less likely to diffuse away the
289     well resolved gradient the fourth order scheme worked so hard to
290     create.
291    
292     The centered fourth order fluxes are discretized:
293     \begin{eqnarray}
294     F_x & = & U \overline{\tau - \frac{1}{6} \delta_{ii} \tau}^i \\
295     F_y & = & V \overline{\tau - \frac{1}{6} \delta_{ii} \tau}^j \\
296     F_r & = & W \overline{\tau - \frac{1}{6} \delta_{ii} \tau}^k
297     \end{eqnarray}
298    
299     As for the third order scheme, the best discretization near boundaries
300 cnh 1.8 is under investigation but currently $\delta_i \tau=0$ on a boundary.
301 adcroft 1.3
302 adcroft 1.4 \fbox{ \begin{minipage}{4.75in}
303     {\em S/R GAD\_C4\_ADV\_X} ({\em gad\_c4\_adv\_x.F})
304    
305     $F_x$: {\bf uT} (argument)
306    
307     $U$: {\bf uTrans} (argument)
308    
309     $\tau$: {\bf tracer} (argument)
310    
311     {\em S/R GAD\_C4\_ADV\_Y} ({\em gad\_c4\_adv\_y.F})
312    
313     $F_y$: {\bf vT} (argument)
314    
315     $V$: {\bf vTrans} (argument)
316    
317     $\tau$: {\bf tracer} (argument)
318    
319     {\em S/R GAD\_C4\_ADV\_R} ({\em gad\_c4\_adv\_r.F})
320    
321     $F_r$: {\bf wT} (argument)
322    
323     $W$: {\bf rTrans} (argument)
324    
325     $\tau$: {\bf tracer} (argument)
326    
327     \end{minipage} }
328    
329    
330 adcroft 1.3 \subsection{First order upwind advection}
331    
332     Although the upwind scheme is the underlying scheme for the robust or
333     non-linear methods given later, we haven't actually supplied this
334     method for general use. It would be very diffusive and it is unlikely
335     that it could ever produce more useful results than the positive
336     higher order schemes.
337    
338     Upwind bias is introduced into many schemes using the {\em abs}
339     function and is allows the first order upwind flux to be written:
340     \begin{eqnarray}
341     F_x & = & U \overline{ \tau }^i - \frac{1}{2} |U| \delta_i \tau \\
342     F_y & = & V \overline{ \tau }^j - \frac{1}{2} |V| \delta_j \tau \\
343     F_r & = & W \overline{ \tau }^k - \frac{1}{2} |W| \delta_k \tau
344     \end{eqnarray}
345    
346     If for some reason, the above method is required, then the second
347     order flux limiter scheme described later reduces to the above scheme
348     if the limiter is set to zero.
349    
350    
351     \section{Non-linear advection schemes}
352    
353     Non-linear advection schemes invoke non-linear interpolation and are
354     widely used in computational fluid dynamics (non-linear does not refer
355     to the non-linearity of the advection operator). The flux limited
356     advection schemes belong to the class of finite volume methods which
357     neatly ties into the spatial discretization of the model.
358    
359     When employing the flux limited schemes, first order upwind or
360     direct-space-time method the time-stepping is switched to forward in
361     time.
362    
363     \subsection{Second order flux limiters}
364    
365     The second order flux limiter method can be cast in several ways but
366     is generally expressed in terms of other flux approximations. For
367     example, in terms of a first order upwind flux and second order
368     Lax-Wendroff flux, the limited flux is given as:
369     \begin{equation}
370     F = F_1 + \psi(r) F_{LW}
371     \end{equation}
372     where $\psi(r)$ is the limiter function,
373     \begin{equation}
374     F_1 = u \overline{\tau}^i - \frac{1}{2} |u| \delta_i \tau
375     \end{equation}
376     is the upwind flux,
377     \begin{equation}
378     F_{LW} = F_1 + \frac{|u|}{2} (1-c) \delta_i \tau
379     \end{equation}
380     is the Lax-Wendroff flux and $c = \frac{u \Delta t}{\Delta x}$ is the
381     Courant (CFL) number.
382    
383     The limiter function, $\psi(r)$, takes the slope ratio
384     \begin{eqnarray}
385     r = \frac{ \tau_{i-1} - \tau_{i-2} }{ \tau_{i} - \tau_{i-1} } & \forall & u > 0
386     \\
387     r = \frac{ \tau_{i+1} - \tau_{i} }{ \tau_{i} - \tau_{i-1} } & \forall & u < 0
388     \end{eqnarray}
389     as it's argument. There are many choices of limiter function but we
390     only provide the Superbee limiter \cite{Roe85}:
391     \begin{equation}
392     \psi(r) = \max[0,\min[1,2r],\min[2,r]]
393     \end{equation}
394    
395 adcroft 1.4 \fbox{ \begin{minipage}{4.75in}
396     {\em S/R GAD\_FLUXLIMIT\_ADV\_X} ({\em gad\_fluxlimit\_adv\_x.F})
397    
398     $F_x$: {\bf uT} (argument)
399    
400     $U$: {\bf uTrans} (argument)
401    
402     $\tau$: {\bf tracer} (argument)
403    
404     {\em S/R GAD\_FLUXLIMIT\_ADV\_Y} ({\em gad\_fluxlimit\_adv\_y.F})
405    
406     $F_y$: {\bf vT} (argument)
407    
408     $V$: {\bf vTrans} (argument)
409    
410     $\tau$: {\bf tracer} (argument)
411    
412     {\em S/R GAD\_FLUXLIMIT\_ADV\_R} ({\em gad\_fluxlimit\_adv\_r.F})
413    
414     $F_r$: {\bf wT} (argument)
415    
416     $W$: {\bf rTrans} (argument)
417    
418     $\tau$: {\bf tracer} (argument)
419    
420     \end{minipage} }
421    
422 adcroft 1.3
423     \subsection{Third order direct space time}
424    
425     The direct-space-time method deals with space and time discretization
426 cnh 1.8 together (other methods that treat space and time separately are known
427 adcroft 1.3 collectively as the ``Method of Lines''). The Lax-Wendroff scheme
428     falls into this category; it adds sufficient diffusion to a second
429     order flux that the forward-in-time method is stable. The upwind
430     biased third order DST scheme is:
431     \begin{eqnarray}
432     F = u \left( \tau_{i-1}
433     + d_0 (\tau_{i}-\tau_{i-1}) + d_1 (\tau_{i-1}-\tau_{i-2}) \right)
434     & \forall & u > 0 \\
435     F = u \left( \tau_{i}
436     - d_0 (\tau_{i}-\tau_{i-1}) - d_1 (\tau_{i+1}-\tau_{i}) \right)
437     & \forall & u < 0
438     \end{eqnarray}
439     where
440     \begin{eqnarray}
441     d_1 & = & \frac{1}{6} ( 2 - |c| ) ( 1 - |c| ) \\
442     d_2 & = & \frac{1}{6} ( 1 - |c| ) ( 1 + |c| )
443     \end{eqnarray}
444 cnh 1.8 The coefficients $d_0$ and $d_1$ approach $1/3$ and $1/6$ respectively
445 adcroft 1.3 as the Courant number, $c$, vanishes. In this limit, the conventional
446     third order upwind method is recovered. For finite Courant number, the
447     deviations from the linear method are analogous to the diffusion added
448     to centered second order advection in the Lax-Wendroff scheme.
449    
450     The DST3 method described above must be used in a forward-in-time
451     manner and is stable for $0 \le |c| \le 1$. Although the scheme
452     appears to be forward-in-time, it is in fact second order in time and
453     the accuracy increases with the Courant number! For low Courant
454     number, DST3 produces very similar results (indistinguishable in
455     Fig.~\ref{fig:advect-1d-lo}) to the linear third order method but for
456     large Courant number, where the linear upwind third order method is
457     unstable, the scheme is extremely accurate
458     (Fig.~\ref{fig:advect-1d-hi}) with only minor overshoots.
459    
460 adcroft 1.4 \fbox{ \begin{minipage}{4.75in}
461     {\em S/R GAD\_DST3\_ADV\_X} ({\em gad\_dst3\_adv\_x.F})
462    
463     $F_x$: {\bf uT} (argument)
464    
465     $U$: {\bf uTrans} (argument)
466    
467     $\tau$: {\bf tracer} (argument)
468    
469     {\em S/R GAD\_DST3\_ADV\_Y} ({\em gad\_dst3\_adv\_y.F})
470    
471     $F_y$: {\bf vT} (argument)
472    
473     $V$: {\bf vTrans} (argument)
474    
475     $\tau$: {\bf tracer} (argument)
476    
477     {\em S/R GAD\_DST3\_ADV\_R} ({\em gad\_dst3\_adv\_r.F})
478    
479     $F_r$: {\bf wT} (argument)
480    
481     $W$: {\bf rTrans} (argument)
482    
483     $\tau$: {\bf tracer} (argument)
484    
485     \end{minipage} }
486    
487    
488 adcroft 1.3 \subsection{Third order direct space time with flux limiting}
489    
490     The overshoots in the DST3 method can be controlled with a flux limiter.
491     The limited flux is written:
492     \begin{equation}
493     F =
494     \frac{1}{2}(u+|u|)\left( \tau_{i-1} + \psi(r^+)(\tau_{i} - \tau_{i-1} )\right)
495     +
496     \frac{1}{2}(u-|u|)\left( \tau_{i-1} + \psi(r^-)(\tau_{i} - \tau_{i-1} )\right)
497     \end{equation}
498     where
499     \begin{eqnarray}
500     r^+ & = & \frac{\tau_{i-1} - \tau_{i-2}}{\tau_{i} - \tau_{i-1}} \\
501     r^- & = & \frac{\tau_{i+1} - \tau_{i}}{\tau_{i} - \tau_{i-1}}
502     \end{eqnarray}
503     and the limiter is the Sweby limiter:
504     \begin{equation}
505     \psi(r) = \max[0, \min[\min(1,d_0+d_1r],\frac{1-c}{c}r ]]
506     \end{equation}
507 adcroft 1.1
508 adcroft 1.4 \fbox{ \begin{minipage}{4.75in}
509     {\em S/R GAD\_DST3FL\_ADV\_X} ({\em gad\_dst3\_adv\_x.F})
510    
511     $F_x$: {\bf uT} (argument)
512    
513     $U$: {\bf uTrans} (argument)
514    
515     $\tau$: {\bf tracer} (argument)
516    
517     {\em S/R GAD\_DST3FL\_ADV\_Y} ({\em gad\_dst3\_adv\_y.F})
518    
519     $F_y$: {\bf vT} (argument)
520    
521     $V$: {\bf vTrans} (argument)
522    
523     $\tau$: {\bf tracer} (argument)
524    
525     {\em S/R GAD\_DST3FL\_ADV\_R} ({\em gad\_dst3\_adv\_r.F})
526    
527     $F_r$: {\bf wT} (argument)
528    
529     $W$: {\bf rTrans} (argument)
530    
531     $\tau$: {\bf tracer} (argument)
532    
533     \end{minipage} }
534    
535    
536 adcroft 1.3 \subsection{Multi-dimensional advection}
537 adcroft 1.1
538 adcroft 1.4 \begin{figure}
539     \resizebox{5.5in}{!}{\includegraphics{part2/advect-2d-lo-diag.eps}}
540     \caption{
541     Comparison of advection schemes in two dimensions; diagonal advection
542 cnh 1.8 of a resolved Gaussian feature. Courant number is 0.01 with
543 adcroft 1.4 30$\times$30 points and solutions are shown for T=1/2. White lines
544     indicate zero crossing (ie. the presence of false minima). The left
545     column shows the second order schemes; top) centered second order with
546     Adams-Bashforth, middle) Lax-Wendroff and bottom) Superbee flux
547     limited. The middle column shows the third order schemes; top) upwind
548     biased third order with Adams-Bashforth, middle) third order direct
549     space-time method and bottom) the same with flux limiting. The top
550     right panel shows the centered fourth order scheme with
551     Adams-Bashforth and right middle panel shows a fourth order variant on
552     the DST method. Bottom right panel shows the Superbee flux limiter
553 cnh 1.8 (second order) applied independently in each direction (method of
554 adcroft 1.4 lines).
555     \label{fig:advect-2d-lo-diag}
556     }
557     \end{figure}
558    
559     \begin{figure}
560     \resizebox{5.5in}{!}{\includegraphics{part2/advect-2d-mid-diag.eps}}
561     \caption{
562     Comparison of advection schemes in two dimensions; diagonal advection
563 cnh 1.8 of a resolved Gaussian feature. Courant number is 0.27 with
564 adcroft 1.4 30$\times$30 points and solutions are shown for T=1/2. White lines
565     indicate zero crossing (ie. the presence of false minima). The left
566     column shows the second order schemes; top) centered second order with
567     Adams-Bashforth, middle) Lax-Wendroff and bottom) Superbee flux
568     limited. The middle column shows the third order schemes; top) upwind
569     biased third order with Adams-Bashforth, middle) third order direct
570     space-time method and bottom) the same with flux limiting. The top
571     right panel shows the centered fourth order scheme with
572     Adams-Bashforth and right middle panel shows a fourth order variant on
573     the DST method. Bottom right panel shows the Superbee flux limiter
574 cnh 1.8 (second order) applied independently in each direction (method of
575 adcroft 1.4 lines).
576     \label{fig:advect-2d-mid-diag}
577     }
578     \end{figure}
579    
580     \begin{figure}
581     \resizebox{5.5in}{!}{\includegraphics{part2/advect-2d-hi-diag.eps}}
582     \caption{
583     Comparison of advection schemes in two dimensions; diagonal advection
584 cnh 1.8 of a resolved Gaussian feature. Courant number is 0.47 with
585 adcroft 1.4 30$\times$30 points and solutions are shown for T=1/2. White lines
586     indicate zero crossings and initial maximum values (ie. the presence
587     of false extrema). The left column shows the second order schemes;
588     top) centered second order with Adams-Bashforth, middle) Lax-Wendroff
589     and bottom) Superbee flux limited. The middle column shows the third
590     order schemes; top) upwind biased third order with Adams-Bashforth,
591     middle) third order direct space-time method and bottom) the same with
592     flux limiting. The top right panel shows the centered fourth order
593     scheme with Adams-Bashforth and right middle panel shows a fourth
594     order variant on the DST method. Bottom right panel shows the Superbee
595 cnh 1.8 flux limiter (second order) applied independently in each direction
596 adcroft 1.4 (method of lines).
597     \label{fig:advect-2d-hi-diag}
598     }
599     \end{figure}
600    
601    
602    
603 cnh 1.8 In many of the aforementioned advection schemes the behavior in
604 adcroft 1.3 multiple dimensions is not necessarily as good as the one dimensional
605 cnh 1.8 behavior. For instance, a shape preserving monotonic scheme in one
606 adcroft 1.3 dimension can have severe shape distortion in two dimensions if the
607     two components of horizontal fluxes are treated independently. There
608     is a large body of literature on the subject dealing with this problem
609     and among the fixes are operator and flux splitting methods, corner
610     flux methods and more. We have adopted a variant on the standard
611     splitting methods that allows the flux calculations to be implemented
612     as if in one dimension:
613     \begin{eqnarray}
614     \tau^{n+1/3} & = & \tau^{n}
615     - \Delta t \left( \frac{1}{\Delta x} \delta_i F^x(\tau^{n})
616     + \tau^{n} \frac{1}{\Delta x} \delta_i u \right) \\
617     \tau^{n+2/3} & = & \tau^{n}
618     - \Delta t \left( \frac{1}{\Delta y} \delta_j F^y(\tau^{n+1/3})
619     + \tau^{n} \frac{1}{\Delta y} \delta_i v \right) \\
620     \tau^{n+3/3} & = & \tau^{n}
621     - \Delta t \left( \frac{1}{\Delta r} \delta_k F^x(\tau^{n+2/3})
622     + \tau^{n} \frac{1}{\Delta r} \delta_i w \right)
623     \end{eqnarray}
624 adcroft 1.1
625 adcroft 1.3 In order to incorporate this method into the general model algorithm,
626 cnh 1.8 we compute the effective tendency rather than update the tracer so
627 adcroft 1.3 that other terms such as diffusion are using the $n$ time-level and
628     not the updated $n+3/3$ quantities:
629     \begin{equation}
630     G^{n+1/2}_{adv} = \frac{1}{\Delta t} ( \tau^{n+3/3} - \tau^{n} )
631     \end{equation}
632     So that the over all time-stepping looks likes:
633     \begin{equation}
634     \tau^{n+1} = \tau^{n} + \Delta t \left( G^{n+1/2}_{adv} + G_{diff}(\tau^{n}) + G^{n}_{forcing} \right)
635     \end{equation}
636 adcroft 1.4
637     \fbox{ \begin{minipage}{4.75in}
638     {\em S/R GAD\_ADVECTION} ({\em gad\_advection.F})
639    
640     $\tau$: {\bf Tracer} (argument)
641    
642     $G^{n+1/2}_{adv}$: {\bf Gtracer} (argument)
643    
644     $F_x, F_y, F_r$: {\bf af} (local)
645    
646     $U$: {\bf uTrans} (local)
647    
648     $V$: {\bf vTrans} (local)
649    
650     $W$: {\bf rTrans} (local)
651    
652     \end{minipage} }
653    
654    
655     \section{Comparison of advection schemes}
656    
657     Figs.~\ref{fig:advect-2d-lo-diag}, \ref{fig:advect-2d-mid-diag} and
658     \ref{fig:advect-2d-hi-diag} show solutions to a simple diagonal
659     advection problem using a selection of schemes for low, moderate and
660     high Courant numbers, respectively. The top row shows the linear
661     schemes, integrated with the Adams-Bashforth method. Theses schemes
662     are clearly unstable for the high Courant number and weakly unstable
663     for the moderate Courant number. The presence of false extrema is very
664     apparent for all Courant numbers. The middle row shows solutions
665     obtained with the unlimited but multi-dimensional schemes. These
666     solutions also exhibit false extrema though the pattern now shows
667     symmetry due to the multi-dimensional scheme. Also, the schemes are
668     stable at high Courant number where the linear schemes weren't. The
669     bottom row (left and middle) shows the limited schemes and most
670     obvious is the absence of false extrema. The accuracy and stability of
671     the unlimited non-linear schemes is retained at high Courant number
672 cnh 1.8 but at low Courant number the tendency is to loose amplitude in sharp
673 adcroft 1.4 peaks due to diffusion. The one dimensional tests shown in
674     Figs.~\ref{fig:advect-1d-lo} and \ref{fig:advect-1d-hi} showed this
675 cnh 1.8 phenomenon.
676 adcroft 1.4
677     Finally, the bottom left and right panels use the same advection
678     scheme but the right does not use the mutli-dimensional method. At low
679     Courant number this appears to not matter but for moderate Courant
680 cnh 1.8 number severe distortion of the feature is apparent. Moreover, the
681 adcroft 1.4 stability of the multi-dimensional scheme is determined by the maximum
682     Courant number applied of each dimension while the stability of the
683     method of lines is determined by the sum. Hence, in the high Courant
684     number plot, the scheme is unstable.
685    
686     With many advection schemes implemented in the code two questions
687     arise: ``Which scheme is best?'' and ``Why don't you just offer the
688     best advection scheme?''. Unfortunately, no one advection scheme is
689     ``the best'' for all particular applications and for new applications
690     it is often a matter of trial to determine which is most
691     suitable. Here are some guidelines but these are not the rule;
692     \begin{itemize}
693     \item If you have a coarsely resolved model, using a
694     positive or upwind biased scheme will introduce significant diffusion
695     to the solution and using a centered higher order scheme will
696     introduce more noise. In this case, simplest may be best.
697     \item If you have a high resolution model, using a higher order
698     scheme will give a more accurate solution but scale-selective
699     diffusion might need to be employed. The flux limited methods
700     offer similar accuracy in this regime.
701 cnh 1.8 \item If your solution has shocks or propagating fronts then a
702 adcroft 1.4 flux limited scheme is almost essential.
703     \item If your time-step is limited by advection, the multi-dimensional
704 cnh 1.8 non-linear schemes have the most stability (up to Courant number 1).
705     \item If you need to know how much diffusion/dissipation has occurred you
706 adcroft 1.4 will have a lot of trouble figuring it out with a non-linear method.
707     \item The presence of false extrema is unphysical and this alone is the
708     strongest argument for using a positive scheme.
709     \end{itemize}

  ViewVC Help
Powered by ViewVC 1.1.22