[1] | 1 | SUBROUTINE timestep |
---|
| 2 | |
---|
| 3 | !------------------------------------------------------------------------------! |
---|
| 4 | ! Actual revisions: |
---|
| 5 | ! ----------------- |
---|
[226] | 6 | ! |
---|
[1] | 7 | ! |
---|
| 8 | ! Former revisions: |
---|
| 9 | ! ----------------- |
---|
[3] | 10 | ! $Id: timestep.f90 226 2009-02-02 07:39:34Z letzel $ |
---|
[110] | 11 | ! |
---|
[226] | 12 | ! 222 2009-01-12 16:04:16Z letzel |
---|
| 13 | ! Implementation of a MPI-1 Coupling: replaced myid with target_id |
---|
| 14 | ! Bugfix for nonparallel execution |
---|
| 15 | ! |
---|
[110] | 16 | ! 108 2007-08-24 15:10:38Z letzel |
---|
| 17 | ! modifications to terminate coupled runs |
---|
| 18 | ! |
---|
[3] | 19 | ! RCS Log replace by Id keyword, revision history cleaned up |
---|
| 20 | ! |
---|
[1] | 21 | ! Revision 1.21 2006/02/23 12:59:44 raasch |
---|
| 22 | ! nt_anz renamed current_timestep_number |
---|
| 23 | ! |
---|
| 24 | ! Revision 1.1 1997/08/11 06:26:19 raasch |
---|
| 25 | ! Initial revision |
---|
| 26 | ! |
---|
| 27 | ! |
---|
| 28 | ! Description: |
---|
| 29 | ! ------------ |
---|
| 30 | ! Compute the time step under consideration of the FCL and diffusion criterion. |
---|
| 31 | !------------------------------------------------------------------------------! |
---|
| 32 | |
---|
| 33 | USE arrays_3d |
---|
| 34 | USE control_parameters |
---|
| 35 | USE cpulog |
---|
| 36 | USE grid_variables |
---|
| 37 | USE indices |
---|
| 38 | USE interfaces |
---|
| 39 | USE pegrid |
---|
| 40 | USE statistics |
---|
| 41 | |
---|
| 42 | IMPLICIT NONE |
---|
| 43 | |
---|
| 44 | INTEGER :: i, j, k |
---|
| 45 | |
---|
| 46 | REAL :: div, dt_diff, dt_diff_l, dt_u, dt_v, dt_w, percent_change, & |
---|
| 47 | u_gtrans_l, value, v_gtrans_l |
---|
| 48 | |
---|
| 49 | REAL, DIMENSION(2) :: uv_gtrans, uv_gtrans_l |
---|
| 50 | REAL, DIMENSION(nzb+1:nzt) :: dxyz2_min |
---|
| 51 | |
---|
| 52 | |
---|
| 53 | CALL cpu_log( log_point(12), 'calculate_timestep', 'start' ) |
---|
| 54 | |
---|
| 55 | ! |
---|
| 56 | !-- Determine the maxima of the velocity components. |
---|
| 57 | CALL global_min_max( nzb, nzt+1, nys-1, nyn+1, nxl-1, nxr+1, u, 'abs', & |
---|
| 58 | u_max, u_max_ijk ) |
---|
| 59 | CALL global_min_max( nzb, nzt+1, nys-1, nyn+1, nxl-1, nxr+1, v, 'abs', & |
---|
| 60 | v_max, v_max_ijk ) |
---|
| 61 | CALL global_min_max( nzb, nzt+1, nys-1, nyn+1, nxl-1, nxr+1, w, 'abs', & |
---|
| 62 | w_max, w_max_ijk ) |
---|
| 63 | |
---|
| 64 | ! |
---|
| 65 | !-- In case maxima of the horizontal velocity components have been found at the |
---|
| 66 | !-- bottom boundary (k=nzb), the corresponding maximum at level k=1 is chosen |
---|
| 67 | !-- if the Dirichlet-boundary condition ('mirror') has been set. This is |
---|
| 68 | !-- necessary, because otherwise in case of Galilei-transform a far too large |
---|
| 69 | !-- velocity (having the respective opposite sign) would be used for the time |
---|
| 70 | !-- step determination (almost double the mean flow velocity). |
---|
| 71 | IF ( ibc_uv_b == 0 ) THEN |
---|
| 72 | IF ( u_max_ijk(1) == nzb ) THEN |
---|
| 73 | u_max = -u_max |
---|
| 74 | u_max_ijk(1) = nzb + 1 |
---|
| 75 | ENDIF |
---|
| 76 | IF ( v_max_ijk(1) == nzb ) THEN |
---|
| 77 | v_max = -v_max |
---|
| 78 | v_max_ijk(1) = nzb + 1 |
---|
| 79 | ENDIF |
---|
| 80 | ENDIF |
---|
| 81 | |
---|
| 82 | ! |
---|
| 83 | !-- In case of Galilei-transform not using the geostrophic wind as translation |
---|
| 84 | !-- velocity, compute the volume-averaged horizontal velocity components, which |
---|
| 85 | !-- will then be subtracted from the horizontal wind for the time step and |
---|
| 86 | !-- horizontal advection routines. |
---|
| 87 | IF ( galilei_transformation .AND. .NOT. use_ug_for_galilei_tr ) THEN |
---|
| 88 | IF ( flow_statistics_called ) THEN |
---|
| 89 | ! |
---|
| 90 | !-- Horizontal averages already existent, just need to average them |
---|
| 91 | !-- vertically. |
---|
| 92 | u_gtrans = 0.0 |
---|
| 93 | v_gtrans = 0.0 |
---|
| 94 | DO k = nzb+1, nzt |
---|
| 95 | u_gtrans = u_gtrans + hom(k,1,1,0) |
---|
| 96 | v_gtrans = v_gtrans + hom(k,1,2,0) |
---|
| 97 | ENDDO |
---|
| 98 | u_gtrans = u_gtrans / REAL( nzt - nzb ) |
---|
| 99 | v_gtrans = v_gtrans / REAL( nzt - nzb ) |
---|
| 100 | ELSE |
---|
| 101 | ! |
---|
| 102 | !-- Averaging over the entire model domain. |
---|
| 103 | uv_gtrans_l = 0.0 |
---|
| 104 | DO i = nxl, nxr |
---|
| 105 | DO j = nys, nyn |
---|
| 106 | DO k = nzb+1, nzt |
---|
| 107 | uv_gtrans_l(1) = uv_gtrans_l(1) + u(k,j,i) |
---|
| 108 | uv_gtrans_l(2) = uv_gtrans_l(2) + v(k,j,i) |
---|
| 109 | ENDDO |
---|
| 110 | ENDDO |
---|
| 111 | ENDDO |
---|
| 112 | uv_gtrans_l = uv_gtrans_l / REAL( (nxr-nxl+1)*(nyn-nys+1)*(nzt-nzb) ) |
---|
| 113 | #if defined( __parallel ) |
---|
| 114 | CALL MPI_ALLREDUCE( uv_gtrans_l, uv_gtrans, 2, MPI_REAL, MPI_SUM, & |
---|
| 115 | comm2d, ierr ) |
---|
| 116 | u_gtrans = uv_gtrans(1) / REAL( numprocs ) |
---|
| 117 | v_gtrans = uv_gtrans(2) / REAL( numprocs ) |
---|
| 118 | #else |
---|
| 119 | u_gtrans = uv_gtrans_l(1) |
---|
| 120 | v_gtrans = uv_gtrans_l(2) |
---|
| 121 | #endif |
---|
| 122 | ENDIF |
---|
| 123 | ENDIF |
---|
| 124 | |
---|
| 125 | IF ( .NOT. dt_fixed ) THEN |
---|
| 126 | ! |
---|
| 127 | !-- Variable time step: |
---|
| 128 | ! |
---|
| 129 | !-- For each component, compute the maximum time step according to the |
---|
| 130 | !-- FCL-criterion. |
---|
| 131 | dt_u = dx / ( ABS( u_max - u_gtrans ) + 1.0E-10 ) |
---|
| 132 | dt_v = dy / ( ABS( v_max - v_gtrans ) + 1.0E-10 ) |
---|
| 133 | dt_w = dzu(MAX( 1, w_max_ijk(1) )) / ( ABS( w_max ) + 1.0E-10 ) |
---|
| 134 | |
---|
| 135 | ! |
---|
| 136 | !-- Compute time step according to the diffusion criterion. |
---|
| 137 | !-- First calculate minimum grid spacing which only depends on index k |
---|
| 138 | !-- Note: also at k=nzb+1 a full grid length is being assumed, although |
---|
| 139 | !-- in the Prandtl-layer friction term only dz/2 is used. |
---|
| 140 | !-- Experience from the old model seems to justify this. |
---|
| 141 | dt_diff_l = 999999.0 |
---|
| 142 | |
---|
| 143 | DO k = nzb+1, nzt |
---|
| 144 | dxyz2_min(k) = MIN( dx2, dy2, dzu(k)*dzu(k) ) * 0.125 |
---|
| 145 | ENDDO |
---|
| 146 | |
---|
| 147 | !$OMP PARALLEL private(i,j,k,value) reduction(MIN: dt_diff_l) |
---|
| 148 | !$OMP DO |
---|
| 149 | DO i = nxl, nxr |
---|
| 150 | DO j = nys, nyn |
---|
| 151 | DO k = nzb+1, nzt |
---|
| 152 | value = dxyz2_min(k) / ( MAX( kh(k,j,i), km(k,j,i) ) + 1E-20 ) |
---|
| 153 | |
---|
| 154 | dt_diff_l = MIN( value, dt_diff_l ) |
---|
| 155 | ENDDO |
---|
| 156 | ENDDO |
---|
| 157 | ENDDO |
---|
| 158 | !$OMP END PARALLEL |
---|
| 159 | #if defined( __parallel ) |
---|
| 160 | CALL MPI_ALLREDUCE( dt_diff_l, dt_diff, 1, MPI_REAL, MPI_MIN, comm2d, & |
---|
| 161 | ierr ) |
---|
| 162 | #else |
---|
| 163 | dt_diff = dt_diff_l |
---|
| 164 | #endif |
---|
| 165 | |
---|
| 166 | ! |
---|
| 167 | !-- In case of non-cyclic lateral boundaries, the diffusion time step |
---|
| 168 | !-- may be further restricted by the lateral damping layer (damping only |
---|
| 169 | !-- along x and y) |
---|
| 170 | IF ( bc_lr /= 'cyclic' ) THEN |
---|
| 171 | dt_diff = MIN( dt_diff, 0.125 * dx2 / ( km_damp_max + 1E-20 ) ) |
---|
| 172 | ELSEIF ( bc_ns /= 'cyclic' ) THEN |
---|
| 173 | dt_diff = MIN( dt_diff, 0.125 * dy2 / ( km_damp_max + 1E-20 ) ) |
---|
| 174 | ENDIF |
---|
| 175 | |
---|
| 176 | ! |
---|
| 177 | !-- The time step is the minimum of the 3 components and the diffusion time |
---|
| 178 | !-- step minus a reduction to be on the safe side. Factor 0.5 is necessary |
---|
| 179 | !-- since the leap-frog scheme always progresses by 2 * delta t. |
---|
| 180 | !-- The user has to set the cfl_factor small enough to ensure that the |
---|
| 181 | !-- divergences do not become too large. |
---|
| 182 | !-- The time step must not exceed the maximum allowed value. |
---|
| 183 | IF ( timestep_scheme(1:5) == 'runge' ) THEN |
---|
| 184 | dt_3d = cfl_factor * MIN( dt_diff, dt_u, dt_v, dt_w ) |
---|
| 185 | ELSE |
---|
| 186 | dt_3d = cfl_factor * 0.5 * MIN( dt_diff, dt_u, dt_v, dt_w ) |
---|
| 187 | ENDIF |
---|
| 188 | dt_3d = MIN( dt_3d, dt_max ) |
---|
| 189 | |
---|
| 190 | ! |
---|
| 191 | !-- Remember the restricting time step criterion for later output. |
---|
| 192 | IF ( dt_diff > MIN( dt_u, dt_v, dt_w ) ) THEN |
---|
| 193 | timestep_reason = 'A' |
---|
| 194 | ELSE |
---|
| 195 | timestep_reason = 'D' |
---|
| 196 | ENDIF |
---|
| 197 | |
---|
| 198 | ! |
---|
| 199 | !-- Set flag if the time step becomes too small. |
---|
| 200 | IF ( dt_3d < ( 0.00001 * dt_max ) ) THEN |
---|
| 201 | stop_dt = .TRUE. |
---|
[108] | 202 | |
---|
[1] | 203 | IF ( myid == 0 ) THEN |
---|
| 204 | PRINT*,'+++ time_step: Time step has reached minimum limit.' |
---|
| 205 | PRINT*,' dt = ', dt_3d, ' s Simulation is terminated.' |
---|
| 206 | PRINT*,' old_dt = ', old_dt, ' s' |
---|
| 207 | PRINT*,' dt_u = ', dt_u, ' s' |
---|
| 208 | PRINT*,' dt_v = ', dt_v, ' s' |
---|
| 209 | PRINT*,' dt_w = ', dt_w, ' s' |
---|
| 210 | PRINT*,' dt_diff = ', dt_diff, ' s' |
---|
| 211 | PRINT*,' u_max = ', u_max, ' m/s k=', u_max_ijk(1), & |
---|
| 212 | ' j=', u_max_ijk(2), ' i=', u_max_ijk(3) |
---|
| 213 | PRINT*,' v_max = ', v_max, ' m/s k=', v_max_ijk(1), & |
---|
| 214 | ' j=', v_max_ijk(2), ' i=', v_max_ijk(3) |
---|
| 215 | PRINT*,' w_max = ', w_max, ' m/s k=', w_max_ijk(1), & |
---|
| 216 | ' j=', w_max_ijk(2), ' i=', w_max_ijk(3) |
---|
| 217 | ENDIF |
---|
[108] | 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] | 222 | #if defined( __parallel ) |
---|
[108] | 223 | IF ( coupling_mode /= 'uncoupled' .AND. terminate_coupled == 0 ) THEN |
---|
| 224 | terminate_coupled = 2 |
---|
| 225 | CALL MPI_SENDRECV( & |
---|
[206] | 226 | terminate_coupled, 1, MPI_INTEGER, target_id, 0, & |
---|
| 227 | terminate_coupled_remote, 1, MPI_INTEGER, target_id, 0, & |
---|
[108] | 228 | comm_inter, status, ierr ) |
---|
| 229 | ENDIF |
---|
[222] | 230 | #endif |
---|
[1] | 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 |
---|