1 | SUBROUTINE timestep |
---|
2 | |
---|
3 | !------------------------------------------------------------------------------! |
---|
4 | ! Current revisions: |
---|
5 | ! ----------------- |
---|
6 | ! Output of messages replaced by message handling routine. |
---|
7 | ! |
---|
8 | ! |
---|
9 | ! Former revisions: |
---|
10 | ! ----------------- |
---|
11 | ! $Id: timestep.f90 258 2009-03-13 12:36:03Z raasch $ |
---|
12 | ! |
---|
13 | ! 222 2009-01-12 16:04:16Z letzel |
---|
14 | ! Implementation of a MPI-1 Coupling: replaced myid with target_id |
---|
15 | ! Bugfix for nonparallel execution |
---|
16 | ! |
---|
17 | ! 108 2007-08-24 15:10:38Z letzel |
---|
18 | ! modifications to terminate coupled runs |
---|
19 | ! |
---|
20 | ! RCS Log replace by Id keyword, revision history cleaned up |
---|
21 | ! |
---|
22 | ! Revision 1.21 2006/02/23 12:59:44 raasch |
---|
23 | ! nt_anz renamed current_timestep_number |
---|
24 | ! |
---|
25 | ! Revision 1.1 1997/08/11 06:26:19 raasch |
---|
26 | ! Initial revision |
---|
27 | ! |
---|
28 | ! |
---|
29 | ! Description: |
---|
30 | ! ------------ |
---|
31 | ! Compute the time step under consideration of the FCL and diffusion criterion. |
---|
32 | !------------------------------------------------------------------------------! |
---|
33 | |
---|
34 | USE arrays_3d |
---|
35 | USE control_parameters |
---|
36 | USE cpulog |
---|
37 | USE grid_variables |
---|
38 | USE indices |
---|
39 | USE interfaces |
---|
40 | USE pegrid |
---|
41 | USE statistics |
---|
42 | |
---|
43 | IMPLICIT NONE |
---|
44 | |
---|
45 | INTEGER :: i, j, k |
---|
46 | |
---|
47 | REAL :: div, dt_diff, dt_diff_l, dt_u, dt_v, dt_w, percent_change, & |
---|
48 | u_gtrans_l, value, v_gtrans_l |
---|
49 | |
---|
50 | REAL, DIMENSION(2) :: uv_gtrans, uv_gtrans_l |
---|
51 | REAL, DIMENSION(nzb+1:nzt) :: dxyz2_min |
---|
52 | |
---|
53 | |
---|
54 | CALL cpu_log( log_point(12), 'calculate_timestep', 'start' ) |
---|
55 | |
---|
56 | ! |
---|
57 | !-- Determine the maxima of the velocity components. |
---|
58 | CALL global_min_max( nzb, nzt+1, nys-1, nyn+1, nxl-1, nxr+1, u, 'abs', & |
---|
59 | u_max, u_max_ijk ) |
---|
60 | CALL global_min_max( nzb, nzt+1, nys-1, nyn+1, nxl-1, nxr+1, v, 'abs', & |
---|
61 | v_max, v_max_ijk ) |
---|
62 | CALL global_min_max( nzb, nzt+1, nys-1, nyn+1, nxl-1, nxr+1, w, 'abs', & |
---|
63 | w_max, w_max_ijk ) |
---|
64 | |
---|
65 | ! |
---|
66 | !-- In case maxima of the horizontal velocity components have been found at the |
---|
67 | !-- bottom boundary (k=nzb), the corresponding maximum at level k=1 is chosen |
---|
68 | !-- if the Dirichlet-boundary condition ('mirror') has been set. This is |
---|
69 | !-- necessary, because otherwise in case of Galilei-transform a far too large |
---|
70 | !-- velocity (having the respective opposite sign) would be used for the time |
---|
71 | !-- step determination (almost double the mean flow velocity). |
---|
72 | IF ( ibc_uv_b == 0 ) THEN |
---|
73 | IF ( u_max_ijk(1) == nzb ) THEN |
---|
74 | u_max = -u_max |
---|
75 | u_max_ijk(1) = nzb + 1 |
---|
76 | ENDIF |
---|
77 | IF ( v_max_ijk(1) == nzb ) THEN |
---|
78 | v_max = -v_max |
---|
79 | v_max_ijk(1) = nzb + 1 |
---|
80 | ENDIF |
---|
81 | ENDIF |
---|
82 | |
---|
83 | ! |
---|
84 | !-- In case of Galilei-transform not using the geostrophic wind as translation |
---|
85 | !-- velocity, compute the volume-averaged horizontal velocity components, which |
---|
86 | !-- will then be subtracted from the horizontal wind for the time step and |
---|
87 | !-- horizontal advection routines. |
---|
88 | IF ( galilei_transformation .AND. .NOT. use_ug_for_galilei_tr ) THEN |
---|
89 | IF ( flow_statistics_called ) THEN |
---|
90 | ! |
---|
91 | !-- Horizontal averages already existent, just need to average them |
---|
92 | !-- vertically. |
---|
93 | u_gtrans = 0.0 |
---|
94 | v_gtrans = 0.0 |
---|
95 | DO k = nzb+1, nzt |
---|
96 | u_gtrans = u_gtrans + hom(k,1,1,0) |
---|
97 | v_gtrans = v_gtrans + hom(k,1,2,0) |
---|
98 | ENDDO |
---|
99 | u_gtrans = u_gtrans / REAL( nzt - nzb ) |
---|
100 | v_gtrans = v_gtrans / REAL( nzt - nzb ) |
---|
101 | ELSE |
---|
102 | ! |
---|
103 | !-- Averaging over the entire model domain. |
---|
104 | uv_gtrans_l = 0.0 |
---|
105 | DO i = nxl, nxr |
---|
106 | DO j = nys, nyn |
---|
107 | DO k = nzb+1, nzt |
---|
108 | uv_gtrans_l(1) = uv_gtrans_l(1) + u(k,j,i) |
---|
109 | uv_gtrans_l(2) = uv_gtrans_l(2) + v(k,j,i) |
---|
110 | ENDDO |
---|
111 | ENDDO |
---|
112 | ENDDO |
---|
113 | uv_gtrans_l = uv_gtrans_l / REAL( (nxr-nxl+1)*(nyn-nys+1)*(nzt-nzb) ) |
---|
114 | #if defined( __parallel ) |
---|
115 | CALL MPI_ALLREDUCE( uv_gtrans_l, uv_gtrans, 2, MPI_REAL, MPI_SUM, & |
---|
116 | comm2d, ierr ) |
---|
117 | u_gtrans = uv_gtrans(1) / REAL( numprocs ) |
---|
118 | v_gtrans = uv_gtrans(2) / REAL( numprocs ) |
---|
119 | #else |
---|
120 | u_gtrans = uv_gtrans_l(1) |
---|
121 | v_gtrans = uv_gtrans_l(2) |
---|
122 | #endif |
---|
123 | ENDIF |
---|
124 | ENDIF |
---|
125 | |
---|
126 | IF ( .NOT. dt_fixed ) THEN |
---|
127 | ! |
---|
128 | !-- Variable time step: |
---|
129 | ! |
---|
130 | !-- For each component, compute the maximum time step according to the |
---|
131 | !-- FCL-criterion. |
---|
132 | dt_u = dx / ( ABS( u_max - u_gtrans ) + 1.0E-10 ) |
---|
133 | dt_v = dy / ( ABS( v_max - v_gtrans ) + 1.0E-10 ) |
---|
134 | dt_w = dzu(MAX( 1, w_max_ijk(1) )) / ( ABS( w_max ) + 1.0E-10 ) |
---|
135 | |
---|
136 | ! |
---|
137 | !-- Compute time step according to the diffusion criterion. |
---|
138 | !-- First calculate minimum grid spacing which only depends on index k |
---|
139 | !-- Note: also at k=nzb+1 a full grid length is being assumed, although |
---|
140 | !-- in the Prandtl-layer friction term only dz/2 is used. |
---|
141 | !-- Experience from the old model seems to justify this. |
---|
142 | dt_diff_l = 999999.0 |
---|
143 | |
---|
144 | DO k = nzb+1, nzt |
---|
145 | dxyz2_min(k) = MIN( dx2, dy2, dzu(k)*dzu(k) ) * 0.125 |
---|
146 | ENDDO |
---|
147 | |
---|
148 | !$OMP PARALLEL private(i,j,k,value) reduction(MIN: dt_diff_l) |
---|
149 | !$OMP DO |
---|
150 | DO i = nxl, nxr |
---|
151 | DO j = nys, nyn |
---|
152 | DO k = nzb+1, nzt |
---|
153 | value = dxyz2_min(k) / ( MAX( kh(k,j,i), km(k,j,i) ) + 1E-20 ) |
---|
154 | |
---|
155 | dt_diff_l = MIN( value, dt_diff_l ) |
---|
156 | ENDDO |
---|
157 | ENDDO |
---|
158 | ENDDO |
---|
159 | !$OMP END PARALLEL |
---|
160 | #if defined( __parallel ) |
---|
161 | CALL MPI_ALLREDUCE( dt_diff_l, dt_diff, 1, MPI_REAL, MPI_MIN, comm2d, & |
---|
162 | ierr ) |
---|
163 | #else |
---|
164 | dt_diff = dt_diff_l |
---|
165 | #endif |
---|
166 | |
---|
167 | ! |
---|
168 | !-- In case of non-cyclic lateral boundaries, the diffusion time step |
---|
169 | !-- may be further restricted by the lateral damping layer (damping only |
---|
170 | !-- along x and y) |
---|
171 | IF ( bc_lr /= 'cyclic' ) THEN |
---|
172 | dt_diff = MIN( dt_diff, 0.125 * dx2 / ( km_damp_max + 1E-20 ) ) |
---|
173 | ELSEIF ( bc_ns /= 'cyclic' ) THEN |
---|
174 | dt_diff = MIN( dt_diff, 0.125 * dy2 / ( km_damp_max + 1E-20 ) ) |
---|
175 | ENDIF |
---|
176 | |
---|
177 | ! |
---|
178 | !-- The time step is the minimum of the 3 components and the diffusion time |
---|
179 | !-- step minus a reduction to be on the safe side. Factor 0.5 is necessary |
---|
180 | !-- since the leap-frog scheme always progresses by 2 * delta t. |
---|
181 | !-- The user has to set the cfl_factor small enough to ensure that the |
---|
182 | !-- divergences do not become too large. |
---|
183 | !-- The time step must not exceed the maximum allowed value. |
---|
184 | IF ( timestep_scheme(1:5) == 'runge' ) THEN |
---|
185 | dt_3d = cfl_factor * MIN( dt_diff, dt_u, dt_v, dt_w ) |
---|
186 | ELSE |
---|
187 | dt_3d = cfl_factor * 0.5 * MIN( dt_diff, dt_u, dt_v, dt_w ) |
---|
188 | ENDIF |
---|
189 | dt_3d = MIN( dt_3d, dt_max ) |
---|
190 | |
---|
191 | ! |
---|
192 | !-- Remember the restricting time step criterion for later output. |
---|
193 | IF ( dt_diff > MIN( dt_u, dt_v, dt_w ) ) THEN |
---|
194 | timestep_reason = 'A' |
---|
195 | ELSE |
---|
196 | timestep_reason = 'D' |
---|
197 | ENDIF |
---|
198 | |
---|
199 | ! |
---|
200 | !-- Set flag if the time step becomes too small. |
---|
201 | IF ( dt_3d < ( 0.00001 * dt_max ) ) THEN |
---|
202 | stop_dt = .TRUE. |
---|
203 | |
---|
204 | WRITE( message_string, * ) 'Time step has reached minimum limit.', & |
---|
205 | '&dt = ', dt_3d, ' s Simulation is terminated.', & |
---|
206 | '&old_dt = ', old_dt, ' s', & |
---|
207 | '&dt_u = ', dt_u, ' s', & |
---|
208 | '&dt_v = ', dt_v, ' s', & |
---|
209 | '&dt_w = ', dt_w, ' s', & |
---|
210 | '&dt_diff = ', dt_diff, ' s', & |
---|
211 | '&u_max = ', u_max, ' m/s k=', u_max_ijk(1), & |
---|
212 | ' j=', u_max_ijk(2), ' i=', u_max_ijk(3), & |
---|
213 | '&v_max = ', v_max, ' m/s k=', v_max_ijk(1), & |
---|
214 | ' j=', v_max_ijk(2), ' i=', v_max_ijk(3), & |
---|
215 | '&w_max = ', w_max, ' m/s k=', w_max_ijk(1), & |
---|
216 | ' j=', w_max_ijk(2), ' i=', w_max_ijk(3) |
---|
217 | CALL message( 'timestep', 'PA0312', 0, 1, 0, 6, 0 ) |
---|
218 | ! |
---|
219 | !-- In case of coupled runs inform the remote model of the termination |
---|
220 | !-- and its reason, provided the remote model has not already been |
---|
221 | !-- informed of another termination reason (terminate_coupled > 0) before. |
---|
222 | #if defined( __parallel ) |
---|
223 | IF ( coupling_mode /= 'uncoupled' .AND. terminate_coupled == 0 ) THEN |
---|
224 | terminate_coupled = 2 |
---|
225 | CALL MPI_SENDRECV( & |
---|
226 | terminate_coupled, 1, MPI_INTEGER, target_id, 0, & |
---|
227 | terminate_coupled_remote, 1, MPI_INTEGER, target_id, 0, & |
---|
228 | comm_inter, status, ierr ) |
---|
229 | ENDIF |
---|
230 | #endif |
---|
231 | ENDIF |
---|
232 | |
---|
233 | ! |
---|
234 | !-- Ensure a smooth value (two significant digits) of the timestep. For |
---|
235 | !-- other schemes than Runge-Kutta, the following restrictions appear: |
---|
236 | !-- The current timestep is only then changed, if the change relative to |
---|
237 | !-- its previous value exceeds +5 % or -2 %. In case of a timestep |
---|
238 | !-- reduction, at least 30 iterations have to be performed before a timestep |
---|
239 | !-- enlargement is permitted again. |
---|
240 | percent_change = dt_3d / old_dt - 1.0 |
---|
241 | IF ( percent_change > 0.05 .OR. percent_change < -0.02 .OR. & |
---|
242 | timestep_scheme(1:5) == 'runge' ) THEN |
---|
243 | |
---|
244 | ! |
---|
245 | !-- Time step enlargement by no more than 2 %. |
---|
246 | IF ( percent_change > 0.0 .AND. simulated_time /= 0.0 .AND. & |
---|
247 | timestep_scheme(1:5) /= 'runge' ) THEN |
---|
248 | dt_3d = 1.02 * old_dt |
---|
249 | ENDIF |
---|
250 | |
---|
251 | ! |
---|
252 | !-- A relatively smooth value of the time step is ensured by taking |
---|
253 | !-- only the first two significant digits. |
---|
254 | div = 1000.0 |
---|
255 | DO WHILE ( dt_3d < div ) |
---|
256 | div = div / 10.0 |
---|
257 | ENDDO |
---|
258 | dt_3d = NINT( dt_3d * 100.0 / div ) * div / 100.0 |
---|
259 | |
---|
260 | ! |
---|
261 | !-- Now the time step can be adjusted. |
---|
262 | IF ( percent_change < 0.0 .OR. timestep_scheme(1:5) == 'runge' ) & |
---|
263 | THEN |
---|
264 | ! |
---|
265 | !-- Time step reduction. |
---|
266 | old_dt = dt_3d |
---|
267 | dt_changed = .TRUE. |
---|
268 | ELSE |
---|
269 | ! |
---|
270 | !-- For other timestep schemes , the time step is only enlarged |
---|
271 | !-- after at least 30 iterations since the previous time step |
---|
272 | !-- change or, of course, after model initialization. |
---|
273 | IF ( current_timestep_number >= last_dt_change + 30 .OR. & |
---|
274 | simulated_time == 0.0 ) THEN |
---|
275 | old_dt = dt_3d |
---|
276 | dt_changed = .TRUE. |
---|
277 | ELSE |
---|
278 | dt_3d = old_dt |
---|
279 | dt_changed = .FALSE. |
---|
280 | ENDIF |
---|
281 | |
---|
282 | ENDIF |
---|
283 | ELSE |
---|
284 | ! |
---|
285 | !-- No time step change since the difference is too small. |
---|
286 | dt_3d = old_dt |
---|
287 | dt_changed = .FALSE. |
---|
288 | ENDIF |
---|
289 | |
---|
290 | IF ( dt_changed ) last_dt_change = current_timestep_number |
---|
291 | |
---|
292 | ENDIF |
---|
293 | |
---|
294 | CALL cpu_log( log_point(12), 'calculate_timestep', 'stop' ) |
---|
295 | |
---|
296 | |
---|
297 | END SUBROUTINE timestep |
---|