Home

Context Navigation

source: palm/trunk/SOURCE/timestep.f90 @ 987

Last change on this file since 987 was 979, checked in by fricke, 12 years ago
last commit documented
Property svn:keywords set to `Id`
File size: 15.7 KB

Rev	Line
[1]	1	SUBROUTINE timestep
	2
	3	!------------------------------------------------------------------------------!
[258]	4	! Current revisions:
[866]	5	! ------------------
[316]	6	!
[979]	7	!
[1]	8	! Former revisions:
	9	! -----------------
[3]	10	! $Id: timestep.f90 979 2012-08-09 08:50:11Z maronga $
[110]	11	!
[979]	12	! 978 2012-08-09 08:28:32Z fricke
	13	! restriction of the outflow damping layer in the diffusion criterion removed
	14	!
[867]	15	! 866 2012-03-28 06:44:41Z raasch
	16	! bugfix for timestep calculation in case of Galilei transformation,
	17	! special treatment in case of mirror velocity boundary condition removed
	18	!
[708]	19	! 707 2011-03-29 11:39:40Z raasch
	20	! bc_lr/ns replaced by bc_lr/ns_cyc
	21	!
[668]	22	! 667 2010-12-23 12:06:00Z suehring/gryschka
	23	! Exchange of terminate_coupled between ocean and atmosphere via PE0
	24	! Minimum grid spacing dxyz2_min(k) is now calculated using dzw instead of dzu
	25	!
[623]	26	! 622 2010-12-10 08:08:13Z raasch
	27	! optional barriers included in order to speed up collective operations
	28	!
[392]	29	! 343 2009-06-24 12:59:09Z maronga
	30	! Additional timestep criterion in case of simulations with plant canopy
	31	! Output of messages replaced by message handling routine.
	32	!
[226]	33	! 222 2009-01-12 16:04:16Z letzel
	34	! Implementation of a MPI-1 Coupling: replaced myid with target_id
	35	! Bugfix for nonparallel execution
	36	!
[110]	37	! 108 2007-08-24 15:10:38Z letzel
	38	! modifications to terminate coupled runs
	39	!
[3]	40	! RCS Log replace by Id keyword, revision history cleaned up
	41	!
[1]	42	! Revision 1.21 2006/02/23 12:59:44 raasch
	43	! nt_anz renamed current_timestep_number
	44	!
	45	! Revision 1.1 1997/08/11 06:26:19 raasch
	46	! Initial revision
	47	!
	48	!
	49	! Description:
	50	! ------------
	51	! Compute the time step under consideration of the FCL and diffusion criterion.
	52	!------------------------------------------------------------------------------!
	53
	54	USE arrays_3d
	55	USE control_parameters
	56	USE cpulog
	57	USE grid_variables
	58	USE indices
	59	USE interfaces
	60	USE pegrid
	61	USE statistics
	62
	63	IMPLICIT NONE
	64
[866]	65	INTEGER :: i, j, k, u_max_cfl_ijk(3), v_max_cfl_ijk(3)
[1]	66
[318]	67	REAL :: div, dt_diff, dt_diff_l, dt_plant_canopy, &
	68	dt_plant_canopy_l, &
	69	dt_plant_canopy_u, dt_plant_canopy_v, dt_plant_canopy_w, &
	70	dt_u, dt_v, dt_w, lad_max, percent_change, &
[866]	71	u_gtrans_l, u_max_cfl, vabs_max, value, v_gtrans_l, v_max_cfl
[1]	72
	73	REAL, DIMENSION(2) :: uv_gtrans, uv_gtrans_l
	74	REAL, DIMENSION(nzb+1:nzt) :: dxyz2_min
	75
[667]	76
	77
[1]	78	CALL cpu_log( log_point(12), 'calculate_timestep', 'start' )
	79
	80	!
	81	!-- In case of Galilei-transform not using the geostrophic wind as translation
	82	!-- velocity, compute the volume-averaged horizontal velocity components, which
	83	!-- will then be subtracted from the horizontal wind for the time step and
	84	!-- horizontal advection routines.
	85	IF ( galilei_transformation .AND. .NOT. use_ug_for_galilei_tr ) THEN
	86	IF ( flow_statistics_called ) THEN
	87	!
	88	!-- Horizontal averages already existent, just need to average them
	89	!-- vertically.
	90	u_gtrans = 0.0
	91	v_gtrans = 0.0
	92	DO k = nzb+1, nzt
	93	u_gtrans = u_gtrans + hom(k,1,1,0)
	94	v_gtrans = v_gtrans + hom(k,1,2,0)
	95	ENDDO
	96	u_gtrans = u_gtrans / REAL( nzt - nzb )
	97	v_gtrans = v_gtrans / REAL( nzt - nzb )
	98	ELSE
	99	!
	100	!-- Averaging over the entire model domain.
	101	uv_gtrans_l = 0.0
	102	DO i = nxl, nxr
	103	DO j = nys, nyn
	104	DO k = nzb+1, nzt
	105	uv_gtrans_l(1) = uv_gtrans_l(1) + u(k,j,i)
	106	uv_gtrans_l(2) = uv_gtrans_l(2) + v(k,j,i)
	107	ENDDO
	108	ENDDO
	109	ENDDO
	110	uv_gtrans_l = uv_gtrans_l / REAL( (nxr-nxl+1)(nyn-nys+1)(nzt-nzb) )
	111	#if defined( __parallel )
[622]	112	IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr )
[1]	113	CALL MPI_ALLREDUCE( uv_gtrans_l, uv_gtrans, 2, MPI_REAL, MPI_SUM, &
	114	comm2d, ierr )
	115	u_gtrans = uv_gtrans(1) / REAL( numprocs )
	116	v_gtrans = uv_gtrans(2) / REAL( numprocs )
	117	#else
	118	u_gtrans = uv_gtrans_l(1)
	119	v_gtrans = uv_gtrans_l(2)
	120	#endif
	121	ENDIF
	122	ENDIF
	123
[866]	124	!
	125	!-- Determine the maxima of the velocity components.
	126	CALL global_min_max( nzb, nzt+1, nysg, nyng, nxlg, nxrg, u, 'abs', 0.0, &
	127	u_max, u_max_ijk )
	128	CALL global_min_max( nzb, nzt+1, nysg, nyng, nxlg, nxrg, v, 'abs', 0.0, &
	129	v_max, v_max_ijk )
	130	CALL global_min_max( nzb, nzt+1, nysg, nyng, nxlg, nxrg, w, 'abs', 0.0, &
	131	w_max, w_max_ijk )
	132
	133	!
	134	!-- In case of Galilei transformation, the horizontal velocity maxima have
	135	!-- to be calculated from the transformed horizontal velocities
	136	IF ( galilei_transformation ) THEN
	137	CALL global_min_max( nzb, nzt+1, nysg, nyng, nxlg, nxrg, u, 'absoff', &
	138	u_gtrans, u_max_cfl, u_max_cfl_ijk )
	139	CALL global_min_max( nzb, nzt+1, nysg, nyng, nxlg, nxrg, v, 'absoff', &
	140	v_gtrans, v_max_cfl, v_max_cfl_ijk )
	141	ELSE
	142	u_max_cfl = u_max
	143	v_max_cfl = v_max
	144	u_max_cfl_ijk = u_max_ijk
	145	v_max_cfl_ijk = v_max_ijk
	146	ENDIF
	147
	148
[1]	149	IF ( .NOT. dt_fixed ) THEN
	150	!
	151	!-- Variable time step:
	152	!
	153	!-- For each component, compute the maximum time step according to the
[866]	154	!-- CFL-criterion.
	155	dt_u = dx / ( ABS( u_max_cfl ) + 1.0E-10 )
	156	dt_v = dy / ( ABS( v_max_cfl ) + 1.0E-10 )
[1]	157	dt_w = dzu(MAX( 1, w_max_ijk(1) )) / ( ABS( w_max ) + 1.0E-10 )
	158
	159	!
	160	!-- Compute time step according to the diffusion criterion.
	161	!-- First calculate minimum grid spacing which only depends on index k
	162	!-- Note: also at k=nzb+1 a full grid length is being assumed, although
	163	!-- in the Prandtl-layer friction term only dz/2 is used.
	164	!-- Experience from the old model seems to justify this.
	165	dt_diff_l = 999999.0
	166
	167	DO k = nzb+1, nzt
[667]	168	dxyz2_min(k) = MIN( dx2, dy2, dzw(k)dzw(k) ) 0.125
[1]	169	ENDDO
	170
	171	!$OMP PARALLEL private(i,j,k,value) reduction(MIN: dt_diff_l)
	172	!$OMP DO
	173	DO i = nxl, nxr
	174	DO j = nys, nyn
	175	DO k = nzb+1, nzt
	176	value = dxyz2_min(k) / ( MAX( kh(k,j,i), km(k,j,i) ) + 1E-20 )
	177
	178	dt_diff_l = MIN( value, dt_diff_l )
	179	ENDDO
	180	ENDDO
	181	ENDDO
	182	!$OMP END PARALLEL
	183	#if defined( __parallel )
[622]	184	IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr )
[1]	185	CALL MPI_ALLREDUCE( dt_diff_l, dt_diff, 1, MPI_REAL, MPI_MIN, comm2d, &
	186	ierr )
	187	#else
	188	dt_diff = dt_diff_l
	189	#endif
	190
	191	!
[316]	192	!-- Additional timestep criterion with plant canopies:
	193	!-- it is not allowed to extract more than the available momentum
	194	IF ( plant_canopy ) THEN
[318]	195
	196	dt_plant_canopy_l = 0.0
	197	DO i = nxl, nxr
	198	DO j = nys, nyn
	199	DO k = nzb+1, nzt
	200	dt_plant_canopy_u = cdc(k,j,i) * lad_u(k,j,i) * &
	201	SQRT( u(k,j,i)**2 + &
	202	( ( v(k,j,i-1) + &
	203	v(k,j,i) + &
	204	v(k,j+1,i) + &
	205	v(k,j+1,i-1) ) &
	206	/ 4.0 )**2 + &
	207	( ( w(k-1,j,i-1) + &
	208	w(k-1,j,i) + &
	209	w(k,j,i-1) + &
	210	w(k,j,i) ) &
	211	/ 4.0 )**2 )
	212	IF ( dt_plant_canopy_u > dt_plant_canopy_l ) THEN
	213	dt_plant_canopy_l = dt_plant_canopy_u
	214	ENDIF
	215	dt_plant_canopy_v = cdc(k,j,i) * lad_v(k,j,i) * &
	216	SQRT( ( ( u(k,j-1,i) + &
	217	u(k,j-1,i+1) + &
	218	u(k,j,i) + &
	219	u(k,j,i+1) ) &
	220	/ 4.0 )**2 + &
	221	v(k,j,i)**2 + &
	222	( ( w(k-1,j-1,i) + &
	223	w(k-1,j,i) + &
	224	w(k,j-1,i) + &
	225	w(k,j,i) ) &
	226	/ 4.0 )**2 )
	227	IF ( dt_plant_canopy_v > dt_plant_canopy_l ) THEN
	228	dt_plant_canopy_l = dt_plant_canopy_v
	229	ENDIF
	230	dt_plant_canopy_w = cdc(k,j,i) * lad_w(k,j,i) * &
	231	SQRT( ( ( u(k,j,i) + &
	232	u(k,j,i+1) + &
	233	u(k+1,j,i) + &
	234	u(k+1,j,i+1) ) &
	235	/ 4.0 )**2 + &
	236	( ( v(k,j,i) + &
	237	v(k,j+1,i) + &
	238	v(k+1,j,i) + &
	239	v(k+1,j+1,i) ) &
	240	/ 4.0 )**2 + &
	241	w(k,j,i)**2 )
	242	IF ( dt_plant_canopy_w > dt_plant_canopy_l ) THEN
	243	dt_plant_canopy_l = dt_plant_canopy_w
	244	ENDIF
	245	ENDDO
	246	ENDDO
	247	ENDDO
	248
	249	IF ( dt_plant_canopy_l > 0.0 ) THEN
[320]	250	!
	251	!-- Invert dt_plant_canopy_l and apply a security timestep factor 0.1
[318]	252	dt_plant_canopy_l = 0.1 / dt_plant_canopy_l
[320]	253	ELSE
	254	!
	255	!-- In case of inhomogeneous plant canopy, some processors may have no
	256	!-- canopy at all. Then use dt_max as dummy instead.
	257	dt_plant_canopy_l = dt_max
[318]	258	ENDIF
[320]	259
[316]	260	!
[318]	261	!-- Determine the global minumum
	262	#if defined( __parallel )
[622]	263	IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr )
[866]	264	CALL MPI_ALLREDUCE( dt_plant_canopy_l, dt_plant_canopy, 1, MPI_REAL, &
[318]	265	MPI_MIN, comm2d, ierr )
	266	#else
	267	dt_plant_canopy = dt_plant_canopy_l
	268	#endif
[316]	269
	270	ELSE
	271	!
	272	!-- Use dt_diff as dummy value to avoid extra IF branches further below
	273	dt_plant_canopy = dt_diff
	274
	275	ENDIF
	276
	277	!
	278	!-- The time step is the minimum of the 3-4 components and the diffusion time
[1]	279	!-- step minus a reduction to be on the safe side. Factor 0.5 is necessary
	280	!-- since the leap-frog scheme always progresses by 2 * delta t.
	281	!-- The user has to set the cfl_factor small enough to ensure that the
	282	!-- divergences do not become too large.
	283	!-- The time step must not exceed the maximum allowed value.
	284	IF ( timestep_scheme(1:5) == 'runge' ) THEN
[316]	285	dt_3d = cfl_factor * MIN( dt_diff, dt_plant_canopy, dt_u, dt_v, dt_w )
[1]	286	ELSE
[316]	287	dt_3d = cfl_factor * 0.5 * &
	288	MIN( dt_diff, dt_plant_canopy, dt_u, dt_v, dt_w )
[1]	289	ENDIF
	290	dt_3d = MIN( dt_3d, dt_max )
	291
	292	!
	293	!-- Remember the restricting time step criterion for later output.
[316]	294	IF ( MIN( dt_u, dt_v, dt_w ) < MIN( dt_diff, dt_plant_canopy ) ) THEN
[1]	295	timestep_reason = 'A'
[316]	296	ELSEIF ( dt_plant_canopy < dt_diff ) THEN
	297	timestep_reason = 'C'
[1]	298	ELSE
	299	timestep_reason = 'D'
	300	ENDIF
	301
	302	!
	303	!-- Set flag if the time step becomes too small.
	304	IF ( dt_3d < ( 0.00001 * dt_max ) ) THEN
	305	stop_dt = .TRUE.
[108]	306
[320]	307	WRITE( message_string, * ) 'Time step has reached minimum limit.', &
	308	'&dt = ', dt_3d, ' s Simulation is terminated.', &
	309	'&old_dt = ', old_dt, ' s', &
	310	'&dt_u = ', dt_u, ' s', &
	311	'&dt_v = ', dt_v, ' s', &
	312	'&dt_w = ', dt_w, ' s', &
	313	'&dt_diff = ', dt_diff, ' s', &
	314	'&dt_plant_canopy = ', dt_plant_canopy, ' s', &
[866]	315	'&u_max_cfl = ', u_max_cfl, ' m/s k=', u_max_cfl_ijk(1), &
[320]	316	' j=', u_max_ijk(2), ' i=', u_max_ijk(3), &
[866]	317	'&v_max_cfl = ', v_max_cfl, ' m/s k=', v_max_cfl_ijk(1), &
[320]	318	' j=', v_max_ijk(2), ' i=', v_max_ijk(3), &
[866]	319	'&w_max = ', w_max, ' m/s k=', w_max_ijk(1), &
[320]	320	' j=', w_max_ijk(2), ' i=', w_max_ijk(3)
[258]	321	CALL message( 'timestep', 'PA0312', 0, 1, 0, 6, 0 )
[108]	322	!
	323	!-- In case of coupled runs inform the remote model of the termination
	324	!-- and its reason, provided the remote model has not already been
	325	!-- informed of another termination reason (terminate_coupled > 0) before.
[222]	326	#if defined( __parallel )
[108]	327	IF ( coupling_mode /= 'uncoupled' .AND. terminate_coupled == 0 ) THEN
	328	terminate_coupled = 2
[667]	329	IF ( myid == 0 ) THEN
	330	CALL MPI_SENDRECV( &
	331	terminate_coupled, 1, MPI_INTEGER, target_id, 0, &
	332	terminate_coupled_remote, 1, MPI_INTEGER, target_id, 0, &
	333	comm_inter, status, ierr )
	334	ENDIF
	335	CALL MPI_BCAST( terminate_coupled_remote, 1, MPI_INTEGER, 0, comm2d, ierr)
[108]	336	ENDIF
[222]	337	#endif
[1]	338	ENDIF
	339
	340	!
	341	!-- Ensure a smooth value (two significant digits) of the timestep. For
	342	!-- other schemes than Runge-Kutta, the following restrictions appear:
	343	!-- The current timestep is only then changed, if the change relative to
	344	!-- its previous value exceeds +5 % or -2 %. In case of a timestep
	345	!-- reduction, at least 30 iterations have to be performed before a timestep
	346	!-- enlargement is permitted again.
	347	percent_change = dt_3d / old_dt - 1.0
	348	IF ( percent_change > 0.05 .OR. percent_change < -0.02 .OR. &
	349	timestep_scheme(1:5) == 'runge' ) THEN
	350
	351	!
	352	!-- Time step enlargement by no more than 2 %.
	353	IF ( percent_change > 0.0 .AND. simulated_time /= 0.0 .AND. &
	354	timestep_scheme(1:5) /= 'runge' ) THEN
	355	dt_3d = 1.02 * old_dt
	356	ENDIF
	357
	358	!
	359	!-- A relatively smooth value of the time step is ensured by taking
	360	!-- only the first two significant digits.
	361	div = 1000.0
	362	DO WHILE ( dt_3d < div )
	363	div = div / 10.0
	364	ENDDO
	365	dt_3d = NINT( dt_3d * 100.0 / div ) * div / 100.0
	366
	367	!
	368	!-- Now the time step can be adjusted.
	369	IF ( percent_change < 0.0 .OR. timestep_scheme(1:5) == 'runge' ) &
	370	THEN
	371	!
	372	!-- Time step reduction.
	373	old_dt = dt_3d
	374	dt_changed = .TRUE.
	375	ELSE
	376	!
	377	!-- For other timestep schemes , the time step is only enlarged
	378	!-- after at least 30 iterations since the previous time step
	379	!-- change or, of course, after model initialization.
	380	IF ( current_timestep_number >= last_dt_change + 30 .OR. &
	381	simulated_time == 0.0 ) THEN
	382	old_dt = dt_3d
	383	dt_changed = .TRUE.
	384	ELSE
	385	dt_3d = old_dt
	386	dt_changed = .FALSE.
	387	ENDIF
	388
	389	ENDIF
	390	ELSE
	391	!
	392	!-- No time step change since the difference is too small.
	393	dt_3d = old_dt
	394	dt_changed = .FALSE.
	395	ENDIF
	396
	397	IF ( dt_changed ) last_dt_change = current_timestep_number
	398
	399	ENDIF
	400	CALL cpu_log( log_point(12), 'calculate_timestep', 'stop' )
	401
	402	END SUBROUTINE timestep

Note: See TracBrowser for help on using the repository browser.

Download in other formats:

| Impressum | ©Leibniz Universität Hannover |