| 1 | MODULE solsor_tam |
|---|
| 2 | !!====================================================================== |
|---|
| 3 | !! *** MODULE solsor_tam *** |
|---|
| 4 | !! NEMOVAR : Tangent linear .... |
|---|
| 5 | !!====================================================================== |
|---|
| 6 | |
|---|
| 7 | !!---------------------------------------------------------------------- |
|---|
| 8 | !! sol_sor_tan : Red-Black Successive Over-Relaxation solver TL |
|---|
| 9 | !!---------------------------------------------------------------------- |
|---|
| 10 | !! * Modules used |
|---|
| 11 | USE par_kind , ONLY: & ! Precision variables |
|---|
| 12 | & wp ! Variable kinds |
|---|
| 13 | USE par_oce , ONLY: & |
|---|
| 14 | & jpi, & |
|---|
| 15 | & jpj, & |
|---|
| 16 | & jpk, & |
|---|
| 17 | & jpr2di, & |
|---|
| 18 | & jpr2dj, & |
|---|
| 19 | & jpiglo |
|---|
| 20 | USE in_out_manager, ONLY: & ! I/O manager |
|---|
| 21 | & nit000, lwp |
|---|
| 22 | USE sol_oce , ONLY: & ! solver variables |
|---|
| 23 | & gcdmat, & |
|---|
| 24 | & rnorme, & |
|---|
| 25 | & eps, & |
|---|
| 26 | & epsr, & |
|---|
| 27 | & nmax, & |
|---|
| 28 | & nmin, & |
|---|
| 29 | & ncut, & |
|---|
| 30 | & nmod, & |
|---|
| 31 | & res, & |
|---|
| 32 | & resmax, & |
|---|
| 33 | & gcp, & |
|---|
| 34 | & sor, & |
|---|
| 35 | & nsol_arp, & |
|---|
| 36 | & c_solver_pt, & |
|---|
| 37 | & niter |
|---|
| 38 | USE lib_mpp , ONLY: & ! distributed memory computing |
|---|
| 39 | & lk_mpp, & |
|---|
| 40 | & mpp_max |
|---|
| 41 | USE lbclnk , ONLY: & ! ocean lateral boundary conditions (or mpp link) |
|---|
| 42 | & lbc_lnk_e |
|---|
| 43 | USE gridsum , ONLY: & ! Global sums |
|---|
| 44 | & global_sum |
|---|
| 45 | USE lbclnk_tam ! ocean lateral boundary conditions (or mpp link) |
|---|
| 46 | USE sol_oce_tam , ONLY: & ! solver variables (assimilation) |
|---|
| 47 | & nitsor, & |
|---|
| 48 | & gcx_tl, & |
|---|
| 49 | & gcb_tl, & |
|---|
| 50 | & gcr_tl, & |
|---|
| 51 | & gcx_ad, & |
|---|
| 52 | & gcb_ad |
|---|
| 53 | USE dom_oce , ONLY: & ! ocean space and time domain |
|---|
| 54 | & mig, & |
|---|
| 55 | & mjg, & |
|---|
| 56 | & nimpp, & |
|---|
| 57 | & njmpp, & |
|---|
| 58 | & e2t, & |
|---|
| 59 | & e1t, & |
|---|
| 60 | & tmask, & |
|---|
| 61 | & nlci, & |
|---|
| 62 | & nlcj, & |
|---|
| 63 | & nldi, & |
|---|
| 64 | & nldj, & |
|---|
| 65 | & nlei, & |
|---|
| 66 | & nlej |
|---|
| 67 | USE gridrandom , ONLY: & ! Random Gaussian noise on grids |
|---|
| 68 | & grid_random |
|---|
| 69 | USE dotprodfld , ONLY: & ! Computes dot product for 3D and 2D fields |
|---|
| 70 | & dot_product |
|---|
| 71 | USE tstool_tam , ONLY: & |
|---|
| 72 | & prntst_adj, & |
|---|
| 73 | & stdgc, & |
|---|
| 74 | & prntst_tlm |
|---|
| 75 | |
|---|
| 76 | |
|---|
| 77 | IMPLICIT NONE |
|---|
| 78 | PRIVATE |
|---|
| 79 | |
|---|
| 80 | !! * Routine accessibility |
|---|
| 81 | PUBLIC sol_sor_adj ! |
|---|
| 82 | PUBLIC sol_sor_tan ! |
|---|
| 83 | PUBLIC sol_sor_adj_tst ! called by tamtst.F90 |
|---|
| 84 | #if defined key_tst_tlm |
|---|
| 85 | PUBLIC sol_sor_tlm_tst ! called by tamtst.F90 |
|---|
| 86 | #endif |
|---|
| 87 | |
|---|
| 88 | CONTAINS |
|---|
| 89 | |
|---|
| 90 | SUBROUTINE sol_sor_tan( kt, kindic ) |
|---|
| 91 | !!---------------------------------------------------------------------- |
|---|
| 92 | !! *** ROUTINE sol_sor_tan : TL of sol_sor *** |
|---|
| 93 | !! |
|---|
| 94 | !! ** Purpose : Solve the ellipic equation for the barotropic stream |
|---|
| 95 | !! function system (lk_dynspg_rl=T) or the transport divergence |
|---|
| 96 | !! system (lk_dynspg_flt=T) using a red-black successive-over- |
|---|
| 97 | !! relaxation method. |
|---|
| 98 | !! In the former case, the barotropic stream function trend has a |
|---|
| 99 | !! zero boundary condition along all coastlines (i.e. continent |
|---|
| 100 | !! as well as islands) while in the latter the boundary condition |
|---|
| 101 | !! specification is not required. |
|---|
| 102 | !! This routine provides a MPI optimization to the existing solsor |
|---|
| 103 | !! by reducing the number of call to lbc. |
|---|
| 104 | !! |
|---|
| 105 | !! ** Method : Successive-over-relaxation method using the red-black |
|---|
| 106 | !! technique. The former technique used was not compatible with |
|---|
| 107 | !! the north-fold boundary condition used in orca configurations. |
|---|
| 108 | !! Compared to the classical sol_sor, this routine provides a |
|---|
| 109 | !! mpp optimization by reducing the number of calls to lnc_lnk |
|---|
| 110 | !! The solution is computed on a larger area and the boudary |
|---|
| 111 | !! conditions only when the inside domain is reached. |
|---|
| 112 | !! |
|---|
| 113 | !! References : |
|---|
| 114 | !! Madec et al. 1988, Ocean Modelling, issue 78, 1-6. |
|---|
| 115 | !! Beare and Stevens 1997 Ann. Geophysicae 15, 1369-1377 |
|---|
| 116 | !! |
|---|
| 117 | !! History of the direct routine: |
|---|
| 118 | !! ! 90-10 (G. Madec) Original code |
|---|
| 119 | !! ! 91-11 (G. Madec) |
|---|
| 120 | !! 7.1 ! 93-04 (G. Madec) time filter |
|---|
| 121 | !! ! 96-05 (G. Madec) merge sor and pcg formulations |
|---|
| 122 | !! 9.0 ! 03-04 (C. Deltel, G. Madec) Red-Black SOR in free form |
|---|
| 123 | !! 9.0 ! 05-09 (R. Benshila, G. Madec) MPI optimization |
|---|
| 124 | !! History of the T&A routine: |
|---|
| 125 | !! ! 96-11 (A. Weaver) correction to preconditioning |
|---|
| 126 | !! 8.2 ! 03-02 (C. Deltel) OPAVAR tangent-linear version |
|---|
| 127 | !! 9.0 ! 07-09 (K. Mogensen) tangent of the 03-04 version |
|---|
| 128 | !! 9.0 ! 09-02 (A. Vidard) tangent of the 05-09 version |
|---|
| 129 | !!---------------------------------------------------------------------- |
|---|
| 130 | !! * Arguments |
|---|
| 131 | INTEGER, INTENT( in ) :: kt ! Current timestep. |
|---|
| 132 | INTEGER, INTENT( inout ) :: kindic ! solver indicator, < 0 if the conver- |
|---|
| 133 | ! ! gence is not reached: the model is |
|---|
| 134 | ! ! stopped in step |
|---|
| 135 | ! ! set to zero before the call of solsor |
|---|
| 136 | !! * Local declarations |
|---|
| 137 | INTEGER :: ji, jj, jn ! dummy loop indices |
|---|
| 138 | INTEGER :: ishift, icount, istp |
|---|
| 139 | REAL(wp) :: ztmp, zres, zres2 |
|---|
| 140 | |
|---|
| 141 | INTEGER :: ijmppodd, ijmppeven |
|---|
| 142 | INTEGER :: ijpr2d |
|---|
| 143 | !!---------------------------------------------------------------------- |
|---|
| 144 | |
|---|
| 145 | ijmppeven = MOD(nimpp+njmpp+jpr2di+jpr2dj ,2) |
|---|
| 146 | ijmppodd = MOD(nimpp+njmpp+jpr2di+jpr2dj+1,2) |
|---|
| 147 | ijpr2d = MAX(jpr2di,jpr2dj) |
|---|
| 148 | icount = 0 |
|---|
| 149 | ! ! ============== |
|---|
| 150 | DO jn = 1, nmax ! Iterative loop |
|---|
| 151 | ! ! ============== |
|---|
| 152 | |
|---|
| 153 | ! applied the lateral boundary conditions |
|---|
| 154 | IF( MOD(icount,ijpr2d+1) == 0 ) CALL lbc_lnk_e( gcx_tl, c_solver_pt, 1.0_wp ) |
|---|
| 155 | |
|---|
| 156 | ! Residuals |
|---|
| 157 | ! --------- |
|---|
| 158 | |
|---|
| 159 | ! Guess black update |
|---|
| 160 | DO jj = 2-jpr2dj, nlcj - 1 + jpr2dj |
|---|
| 161 | ishift = MOD( jj-ijmppodd-jpr2dj, 2 ) |
|---|
| 162 | DO ji = 2-jpr2di+ishift, nlci - 1 + jpr2di, 2 |
|---|
| 163 | ztmp = gcb_tl(ji ,jj ) & |
|---|
| 164 | & - gcp(ji,jj,1) * gcx_tl(ji ,jj-1) & |
|---|
| 165 | & - gcp(ji,jj,2) * gcx_tl(ji-1,jj ) & |
|---|
| 166 | & - gcp(ji,jj,3) * gcx_tl(ji+1,jj ) & |
|---|
| 167 | & - gcp(ji,jj,4) * gcx_tl(ji ,jj+1) |
|---|
| 168 | ! Estimate of the residual |
|---|
| 169 | zres = ztmp - gcx_tl(ji,jj) |
|---|
| 170 | gcr_tl(ji,jj) = zres * gcdmat(ji,jj) * zres |
|---|
| 171 | ! Guess update |
|---|
| 172 | gcx_tl(ji,jj) = sor * ztmp + ( 1.0_wp - sor ) * gcx_tl(ji,jj) |
|---|
| 173 | END DO |
|---|
| 174 | END DO |
|---|
| 175 | icount = icount + 1 |
|---|
| 176 | |
|---|
| 177 | ! applied the lateral boundary conditions |
|---|
| 178 | IF( MOD(icount,ijpr2d+1) == 0 ) CALL lbc_lnk_e( gcx_tl, c_solver_pt, 1.0_wp ) |
|---|
| 179 | |
|---|
| 180 | ! Guess red update |
|---|
| 181 | DO jj = 2-jpr2dj, nlcj-1+jpr2dj |
|---|
| 182 | ishift = MOD( jj-ijmppeven-jpr2dj, 2 ) |
|---|
| 183 | DO ji = 2-jpr2di+ishift, nlci-1+jpr2di, 2 |
|---|
| 184 | ztmp = gcb_tl(ji ,jj ) & |
|---|
| 185 | & - gcp(ji,jj,1) * gcx_tl(ji ,jj-1) & |
|---|
| 186 | & - gcp(ji,jj,2) * gcx_tl(ji-1,jj ) & |
|---|
| 187 | & - gcp(ji,jj,3) * gcx_tl(ji+1,jj ) & |
|---|
| 188 | & - gcp(ji,jj,4) * gcx_tl(ji ,jj+1) |
|---|
| 189 | ! Estimate of the residual |
|---|
| 190 | zres = ztmp - gcx_tl(ji,jj) |
|---|
| 191 | gcr_tl(ji,jj) = zres * gcdmat(ji,jj) * zres |
|---|
| 192 | ! Guess update |
|---|
| 193 | gcx_tl(ji,jj) = sor * ztmp + ( 1.0_wp - sor ) * gcx_tl(ji,jj) |
|---|
| 194 | END DO |
|---|
| 195 | END DO |
|---|
| 196 | icount = icount + 1 |
|---|
| 197 | |
|---|
| 198 | ! test of convergence |
|---|
| 199 | IF ( (jn > nmin .AND. MOD( jn-nmin, nmod ) == 0) .OR. jn==nmax) THEN |
|---|
| 200 | |
|---|
| 201 | SELECT CASE ( nsol_arp ) |
|---|
| 202 | CASE ( 0 ) |
|---|
| 203 | ! absolute precision (maximum value of the residual) |
|---|
| 204 | zres2 = MAXVAL( gcr_tl(2:nlci - 1,2:nlcj - 1) ) |
|---|
| 205 | IF( lk_mpp ) CALL mpp_max( zres2 ) ! max over the global domain |
|---|
| 206 | ! test of convergence |
|---|
| 207 | IF( zres2 < resmax .OR. jn == nmax ) THEN |
|---|
| 208 | res = SQRT( zres2 ) |
|---|
| 209 | niter = jn |
|---|
| 210 | ncut = 999 |
|---|
| 211 | ! Store number of iterations for adjoint computation |
|---|
| 212 | istp = kt - nit000 + 1 |
|---|
| 213 | nitsor(istp) = niter |
|---|
| 214 | ENDIF |
|---|
| 215 | CASE ( 1 ) ! relative precision |
|---|
| 216 | ! rnorme = SUM( gcr_tl(2:nlci - 1,2:nlcj - 1) ) |
|---|
| 217 | ! IF( lk_mpp ) CALL mpp_sum( rnorme ) ! sum over global domain |
|---|
| 218 | rnorme = global_sum(gcr_tl) |
|---|
| 219 | ! test of convergence |
|---|
| 220 | IF( rnorme < epsr .OR. jn == nmax ) THEN |
|---|
| 221 | res = SQRT( rnorme ) |
|---|
| 222 | niter = jn |
|---|
| 223 | ncut = 999 |
|---|
| 224 | ! Store number of iterations for adjoint computation |
|---|
| 225 | istp = kt - nit000 + 1 |
|---|
| 226 | nitsor(istp) = niter |
|---|
| 227 | ENDIF |
|---|
| 228 | END SELECT |
|---|
| 229 | |
|---|
| 230 | !**** |
|---|
| 231 | ! IF(lwp)WRITE(numsol,9300) jn, res, sqrt( epsr ) / eps |
|---|
| 232 | !9300 FORMAT(' niter :',i6,' res :',e20.10,' b :',e20.10) |
|---|
| 233 | !**** |
|---|
| 234 | |
|---|
| 235 | ENDIF |
|---|
| 236 | ! indicator of non-convergence or explosion |
|---|
| 237 | IF( jn == nmax ) nitsor(istp) = jn |
|---|
| 238 | IF( jn == nmax .OR. SQRT(epsr)/eps > 1.e+20 ) kindic = -2 |
|---|
| 239 | IF( ncut == 999 ) GOTO 999 |
|---|
| 240 | |
|---|
| 241 | ! ! ===================== |
|---|
| 242 | END DO ! END of iterative loop |
|---|
| 243 | ! ! ===================== |
|---|
| 244 | |
|---|
| 245 | 999 CONTINUE |
|---|
| 246 | |
|---|
| 247 | ! Output in gcx_tl |
|---|
| 248 | ! ---------------- |
|---|
| 249 | |
|---|
| 250 | CALL lbc_lnk_e( gcx_tl, c_solver_pt, 1.0_wp ) ! Lateral BCs |
|---|
| 251 | |
|---|
| 252 | END SUBROUTINE sol_sor_tan |
|---|
| 253 | SUBROUTINE sol_sor_adj( kt, kindic ) |
|---|
| 254 | !!---------------------------------------------------------------------- |
|---|
| 255 | !! *** ROUTINE sol_sor_adj : adjoint of sol_sor *** |
|---|
| 256 | !! |
|---|
| 257 | !! ** Purpose : Solve the ellipic equation for the barotropic stream |
|---|
| 258 | !! function system (lk_dynspg_rl=T) or the transport divergence |
|---|
| 259 | !! system (lk_dynspg_flt=T) using a red-black successive-over- |
|---|
| 260 | !! relaxation method. |
|---|
| 261 | !! In the former case, the barotropic stream function trend has a |
|---|
| 262 | !! zero boundary condition along all coastlines (i.e. continent |
|---|
| 263 | !! as well as islands) while in the latter the boundary condition |
|---|
| 264 | !! specification is not required. |
|---|
| 265 | !! This routine provides a MPI optimization to the existing solsor |
|---|
| 266 | !! by reducing the number of call to lbc. |
|---|
| 267 | !! |
|---|
| 268 | !! ** Method : Successive-over-relaxation method using the red-black |
|---|
| 269 | !! technique. The former technique used was not compatible with |
|---|
| 270 | !! the north-fold boundary condition used in orca configurations. |
|---|
| 271 | !! Compared to the classical sol_sor, this routine provides a |
|---|
| 272 | !! mpp optimization by reducing the number of calls to lbc_lnk |
|---|
| 273 | !! The solution is computed on a larger area and the boudary |
|---|
| 274 | !! conditions only when the inside domain is reached. |
|---|
| 275 | !! ** Comments on adjoint routine : |
|---|
| 276 | !! When the step in a tangent-linear DO loop is an arbitrary |
|---|
| 277 | !! integer then care must be taken in computing the lower bound |
|---|
| 278 | !! of the adjoint DO loop; i.e., |
|---|
| 279 | !! |
|---|
| 280 | !! If the tangent-linear DO loop is: low_tl, up_tl, step |
|---|
| 281 | !! |
|---|
| 282 | !! then the adjoint DO loop is: low_ad, up_ad, -step |
|---|
| 283 | !! |
|---|
| 284 | !! where low_ad = low_tl + step * INT( ( up_tl - low_tl ) / step ) |
|---|
| 285 | !! up_ad = low_tl |
|---|
| 286 | !! |
|---|
| 287 | !! NB. If step = 1 then low_ad = up_tl |
|---|
| 288 | !! |
|---|
| 289 | !! References : |
|---|
| 290 | !! Madec et al. 1988, Ocean Modelling, issue 78, 1-6. |
|---|
| 291 | !! Beare and Stevens 1997 Ann. Geophysicae 15, 1369-1377 |
|---|
| 292 | !! |
|---|
| 293 | !! History of the direct routine: |
|---|
| 294 | !! ! 90-10 (G. Madec) Original code |
|---|
| 295 | !! ! 91-11 (G. Madec) |
|---|
| 296 | !! 7.1 ! 93-04 (G. Madec) time filter |
|---|
| 297 | !! ! 96-05 (G. Madec) merge sor and pcg formulations |
|---|
| 298 | !! 9.0 ! 03-04 (C. Deltel, G. Madec) Red-Black SOR in free form |
|---|
| 299 | !! 9.0 ! 05-09 (R. Benshila, G. Madec) MPI optimization |
|---|
| 300 | !! History of the T&A routine: |
|---|
| 301 | !! ! 96-11 (A. Weaver) correction to preconditioning |
|---|
| 302 | !! 8.2 ! 03-02 (C. Deltel) OPAVAR tangent-linear version |
|---|
| 303 | !! 9.0 ! 07-09 (K. Mogensen, A. Weaver) adjoint of the 03-04 version |
|---|
| 304 | !! 9.0 ! 09-02 (A. Vidard) adjoint of the 05-09 version |
|---|
| 305 | !!---------------------------------------------------------------------- |
|---|
| 306 | !! * Arguments |
|---|
| 307 | INTEGER, INTENT( in ) :: kt ! Current timestep. |
|---|
| 308 | INTEGER, INTENT( inout ) :: kindic ! solver indicator, < 0 if the conver- |
|---|
| 309 | ! ! gence is not reached: the model is |
|---|
| 310 | ! ! stopped in step |
|---|
| 311 | ! ! set to zero before the call of solsor |
|---|
| 312 | !! * Local declarations |
|---|
| 313 | INTEGER :: ji, jj, jn ! dummy loop indices |
|---|
| 314 | INTEGER :: ishift, icount, istp, iter, ilower |
|---|
| 315 | REAL(wp) :: ztmp, zres, zres2 |
|---|
| 316 | |
|---|
| 317 | INTEGER :: ijmppodd, ijmppeven |
|---|
| 318 | INTEGER :: ijpr2d |
|---|
| 319 | !!---------------------------------------------------------------------- |
|---|
| 320 | |
|---|
| 321 | ijmppeven = MOD(nimpp+njmpp+jpr2di+jpr2dj,2) |
|---|
| 322 | ijmppodd = MOD(nimpp+njmpp+jpr2di+jpr2dj+1,2) |
|---|
| 323 | ijpr2d = MAX(jpr2di,jpr2dj) |
|---|
| 324 | |
|---|
| 325 | ! Fixed number of iterations |
|---|
| 326 | istp = kt - nit000 + 1 |
|---|
| 327 | iter = nitsor(istp) |
|---|
| 328 | icount = iter * 2 |
|---|
| 329 | ! Output in gcx_ad |
|---|
| 330 | ! ---------------- |
|---|
| 331 | |
|---|
| 332 | CALL lbc_lnk_e_adj( gcx_ad, c_solver_pt, 1.0_wp ) ! Lateral BCs |
|---|
| 333 | |
|---|
| 334 | ! ! ============== |
|---|
| 335 | DO jn = iter, 1, -1 ! Iterative loop |
|---|
| 336 | ! ! ============== |
|---|
| 337 | ! Guess red update |
|---|
| 338 | DO jj = nlcj-1+jpr2dj, 2-jpr2dj, -1 |
|---|
| 339 | ishift = MOD( jj-ijmppeven-jpr2dj, 2 ) |
|---|
| 340 | ilower = 2-jpr2dj+ishift + 2 * INT( ( ( nlci-1+jpr2dj )-( 2-jpr2dj+ishift ) ) / 2 ) |
|---|
| 341 | DO ji = ilower, 2-jpr2dj+ishift, -2 |
|---|
| 342 | gcb_ad(ji ,jj ) = gcb_ad(ji ,jj ) + sor * gcx_ad(ji,jj) |
|---|
| 343 | gcx_ad(ji ,jj-1) = gcx_ad(ji ,jj-1) - sor * gcx_ad(ji,jj) & |
|---|
| 344 | & * gcp(ji,jj,1) |
|---|
| 345 | gcx_ad(ji-1,jj ) = gcx_ad(ji-1,jj ) - sor * gcx_ad(ji,jj) & |
|---|
| 346 | & * gcp(ji,jj,2) |
|---|
| 347 | gcx_ad(ji+1,jj ) = gcx_ad(ji+1,jj ) - sor * gcx_ad(ji,jj) & |
|---|
| 348 | & * gcp(ji,jj,3) |
|---|
| 349 | gcx_ad(ji ,jj+1) = gcx_ad(ji ,jj+1) - sor * gcx_ad(ji,jj) & |
|---|
| 350 | & * gcp(ji,jj,4) |
|---|
| 351 | gcx_ad(ji ,jj ) = gcx_ad(ji ,jj ) * ( 1.0_wp - sor ) |
|---|
| 352 | END DO |
|---|
| 353 | END DO |
|---|
| 354 | icount = icount - 1 |
|---|
| 355 | ! applied the lateral boundary conditions |
|---|
| 356 | IF( MOD(icount,ijpr2d+1) == 0 ) CALL lbc_lnk_e_adj( gcx_ad, c_solver_pt, 1.0_wp ) ! Lateral BCs |
|---|
| 357 | |
|---|
| 358 | ! Residus |
|---|
| 359 | ! ------- |
|---|
| 360 | |
|---|
| 361 | ! Guess black update |
|---|
| 362 | DO jj = nlcj - 1+jpr2dj, 2-jpr2dj, -1 |
|---|
| 363 | ishift = MOD( jj-ijmppodd-jpr2dj, 2 ) |
|---|
| 364 | ilower = 2-jpr2dj+ishift + 2 * INT( ( ( nlci-1+jpr2dj )-( 2-jpr2dj+ishift ) ) / 2 ) |
|---|
| 365 | DO ji = ilower, 2-jpr2dj+ishift, -2 |
|---|
| 366 | gcb_ad(ji ,jj ) = gcb_ad(ji ,jj ) + sor * gcx_ad(ji,jj) |
|---|
| 367 | gcx_ad(ji ,jj-1) = gcx_ad(ji ,jj-1) - sor * gcx_ad(ji,jj) & |
|---|
| 368 | & * gcp(ji,jj,1) |
|---|
| 369 | gcx_ad(ji-1,jj ) = gcx_ad(ji-1,jj ) - sor * gcx_ad(ji,jj) & |
|---|
| 370 | & * gcp(ji,jj,2) |
|---|
| 371 | gcx_ad(ji+1,jj ) = gcx_ad(ji+1,jj ) - sor * gcx_ad(ji,jj) & |
|---|
| 372 | & * gcp(ji,jj,3) |
|---|
| 373 | gcx_ad(ji ,jj+1) = gcx_ad(ji ,jj+1) - sor * gcx_ad(ji,jj) & |
|---|
| 374 | & * gcp(ji,jj,4) |
|---|
| 375 | gcx_ad(ji ,jj ) = gcx_ad(ji ,jj ) * ( 1.0_wp - sor ) |
|---|
| 376 | END DO |
|---|
| 377 | END DO |
|---|
| 378 | |
|---|
| 379 | icount = icount - 1 |
|---|
| 380 | ! applied the lateral boundary conditions |
|---|
| 381 | IF( MOD(icount,ijpr2d+1) == 0 ) CALL lbc_lnk_e_adj( gcx_ad, c_solver_pt, 1.0_wp ) ! Lateral BCs |
|---|
| 382 | |
|---|
| 383 | ! ! ===================== |
|---|
| 384 | END DO ! END of iterative loop |
|---|
| 385 | ! ! ===================== |
|---|
| 386 | |
|---|
| 387 | END SUBROUTINE sol_sor_adj |
|---|
| 388 | SUBROUTINE sol_sor_adj_tst( kumadt ) |
|---|
| 389 | !!----------------------------------------------------------------------- |
|---|
| 390 | !! |
|---|
| 391 | !! *** ROUTINE example_adj_tst *** |
|---|
| 392 | !! |
|---|
| 393 | !! ** Purpose : Test the adjoint routine. |
|---|
| 394 | !! |
|---|
| 395 | !! ** Method : Verify the scalar product |
|---|
| 396 | !! |
|---|
| 397 | !! ( L dx )^T W dy = dx^T L^T W dy |
|---|
| 398 | !! |
|---|
| 399 | !! where L = tangent routine |
|---|
| 400 | !! L^T = adjoint routine |
|---|
| 401 | !! W = diagonal matrix of scale factors |
|---|
| 402 | !! dx = input perturbation (random field) |
|---|
| 403 | !! dy = L dx |
|---|
| 404 | !! |
|---|
| 405 | !! |
|---|
| 406 | !! History : |
|---|
| 407 | !! ! 08-08 (A. Vidard) |
|---|
| 408 | !!----------------------------------------------------------------------- |
|---|
| 409 | !! * Modules used |
|---|
| 410 | |
|---|
| 411 | !! * Arguments |
|---|
| 412 | INTEGER, INTENT(IN) :: & |
|---|
| 413 | & kumadt ! Output unit |
|---|
| 414 | |
|---|
| 415 | !! * Local declarations |
|---|
| 416 | INTEGER :: & |
|---|
| 417 | & ji, & ! dummy loop indices |
|---|
| 418 | & jj, & |
|---|
| 419 | & jk, & |
|---|
| 420 | & kindic,& ! flags fo solver convergence |
|---|
| 421 | & kmod, & ! frequency of test for the SOR solver |
|---|
| 422 | & kt ! number of iteration |
|---|
| 423 | INTEGER, DIMENSION(jpi,jpj) :: & |
|---|
| 424 | & iseed_2d ! 2D seed for the random number generator |
|---|
| 425 | REAL(KIND=wp) :: & |
|---|
| 426 | & zsp1, & ! scalar product involving the tangent routine |
|---|
| 427 | & zsp2 ! scalar product involving the adjoint routine |
|---|
| 428 | REAL(KIND=wp), DIMENSION(:,:), ALLOCATABLE :: & |
|---|
| 429 | & zgcb_tlin , & ! Tangent input |
|---|
| 430 | & zgcx_tlin , & ! Tangent input |
|---|
| 431 | & zgcx_tlout, & ! Tangent output |
|---|
| 432 | & zgcx_adin , & ! Adjoint input |
|---|
| 433 | & zgcb_adout, & ! Adjoint output |
|---|
| 434 | & zgcx_adout, & ! Adjoint output |
|---|
| 435 | & zr ! 3D random field |
|---|
| 436 | CHARACTER(LEN=14) :: cl_name |
|---|
| 437 | ! Allocate memory |
|---|
| 438 | |
|---|
| 439 | |
|---|
| 440 | ALLOCATE( & |
|---|
| 441 | & zgcb_tlin( jpi,jpj), & |
|---|
| 442 | & zgcx_tlin( jpi,jpj), & |
|---|
| 443 | & zgcx_tlout(jpi,jpj), & |
|---|
| 444 | & zgcx_adin( jpi,jpj), & |
|---|
| 445 | & zgcx_adout(jpi,jpj), & |
|---|
| 446 | & zgcb_adout(jpi,jpj), & |
|---|
| 447 | & zr( jpi,jpj) & |
|---|
| 448 | & ) |
|---|
| 449 | !================================================================== |
|---|
| 450 | ! 1) dx = ( un_tl, vn_tl, hdivn_tl ) and |
|---|
| 451 | ! dy = ( hdivb_tl, hdivn_tl ) |
|---|
| 452 | !================================================================== |
|---|
| 453 | |
|---|
| 454 | !-------------------------------------------------------------------- |
|---|
| 455 | ! Reset the tangent and adjoint variables |
|---|
| 456 | !-------------------------------------------------------------------- |
|---|
| 457 | zgcb_tlin( :,:) = 0.0_wp |
|---|
| 458 | zgcx_tlin( :,:) = 0.0_wp |
|---|
| 459 | zgcx_tlout(:,:) = 0.0_wp |
|---|
| 460 | zgcx_adin( :,:) = 0.0_wp |
|---|
| 461 | zgcx_adout(:,:) = 0.0_wp |
|---|
| 462 | zgcb_adout(:,:) = 0.0_wp |
|---|
| 463 | zr( :,:) = 0.0_wp |
|---|
| 464 | !-------------------------------------------------------------------- |
|---|
| 465 | ! Initialize the tangent input with random noise: dx |
|---|
| 466 | !-------------------------------------------------------------------- |
|---|
| 467 | |
|---|
| 468 | kt=nit000 |
|---|
| 469 | kindic=0 |
|---|
| 470 | ! kmod = nmod ! store frequency of test for the SOR solver |
|---|
| 471 | ! nmod = 1 ! force frequency to one (remove adj_tst dependancy to nn_nmin) |
|---|
| 472 | |
|---|
| 473 | DO jj = 1, jpj |
|---|
| 474 | DO ji = 1, jpi |
|---|
| 475 | iseed_2d(ji,jj) = - ( 596035 + & |
|---|
| 476 | & mig(ji) + ( mjg(jj) - 1 ) * jpiglo ) |
|---|
| 477 | END DO |
|---|
| 478 | END DO |
|---|
| 479 | CALL grid_random( iseed_2d, zr, c_solver_pt, 0.0_wp, stdgc ) |
|---|
| 480 | DO jj = nldj, nlej |
|---|
| 481 | DO ji = nldi, nlei |
|---|
| 482 | zgcb_tlin(ji,jj) = zr(ji,jj) |
|---|
| 483 | END DO |
|---|
| 484 | END DO |
|---|
| 485 | DO jj = 1, jpj |
|---|
| 486 | DO ji = 1, jpi |
|---|
| 487 | iseed_2d(ji,jj) = - ( 264792 + & |
|---|
| 488 | & mig(ji) + ( mjg(jj) - 1 ) * jpiglo ) |
|---|
| 489 | END DO |
|---|
| 490 | END DO |
|---|
| 491 | CALL grid_random( iseed_2d, zr, c_solver_pt, 0.0_wp, stdgc ) |
|---|
| 492 | DO jj = nldj, nlej |
|---|
| 493 | DO ji = nldi, nlei |
|---|
| 494 | zgcx_tlin(ji,jj) = zr(ji,jj) |
|---|
| 495 | END DO |
|---|
| 496 | END DO |
|---|
| 497 | ncut = 1 ! reinitilize the solver convergence flag |
|---|
| 498 | gcr_tl(:,:) = 0.0_wp |
|---|
| 499 | gcb_tl(:,:) = zgcb_tlin(:,:) |
|---|
| 500 | gcx_tl(:,:) = zgcx_tlin(:,:) |
|---|
| 501 | CALL sol_sor_tan(kt, kindic) |
|---|
| 502 | zgcx_tlout(:,:) = gcx_tl(:,:) |
|---|
| 503 | |
|---|
| 504 | |
|---|
| 505 | !-------------------------------------------------------------------- |
|---|
| 506 | ! Initialize the adjoint variables: dy^* = W dy |
|---|
| 507 | !-------------------------------------------------------------------- |
|---|
| 508 | |
|---|
| 509 | DO jj = nldj, nlej |
|---|
| 510 | DO ji = nldi, nlei |
|---|
| 511 | zgcx_adin(ji,jj) = zgcx_tlout(ji,jj) & |
|---|
| 512 | & * e1t(ji,jj) * e2t(ji,jj) & |
|---|
| 513 | & * tmask(ji,jj,1) |
|---|
| 514 | END DO |
|---|
| 515 | END DO |
|---|
| 516 | !-------------------------------------------------------------------- |
|---|
| 517 | ! Compute the scalar product: ( L dx )^T W dy |
|---|
| 518 | !-------------------------------------------------------------------- |
|---|
| 519 | |
|---|
| 520 | zsp1 = DOT_PRODUCT( zgcx_tlout, zgcx_adin ) |
|---|
| 521 | |
|---|
| 522 | !-------------------------------------------------------------------- |
|---|
| 523 | ! Call the adjoint routine: dx^* = L^T dy^* |
|---|
| 524 | !-------------------------------------------------------------------- |
|---|
| 525 | gcb_ad(:,:) = 0.0_wp |
|---|
| 526 | gcx_ad(:,:) = zgcx_adin(:,:) |
|---|
| 527 | CALL sol_sor_adj(kt, kindic) |
|---|
| 528 | zgcx_adout(:,:) = gcx_ad(:,:) |
|---|
| 529 | zgcb_adout(:,:) = gcb_ad(:,:) |
|---|
| 530 | |
|---|
| 531 | |
|---|
| 532 | zsp2 = DOT_PRODUCT( zgcx_tlin, zgcx_adout ) & |
|---|
| 533 | & + DOT_PRODUCT( zgcb_tlin, zgcb_adout ) |
|---|
| 534 | |
|---|
| 535 | ! 14 char:'12345678901234' |
|---|
| 536 | cl_name = 'sol_sor_adj ' |
|---|
| 537 | CALL prntst_adj( cl_name, kumadt, zsp1, zsp2 ) |
|---|
| 538 | |
|---|
| 539 | ! nmod = kmod ! restore initial frequency of test for the SOR solver |
|---|
| 540 | |
|---|
| 541 | DEALLOCATE( & |
|---|
| 542 | & zgcb_tlin, & |
|---|
| 543 | & zgcx_tlin, & |
|---|
| 544 | & zgcx_tlout, & |
|---|
| 545 | & zgcx_adin, & |
|---|
| 546 | & zgcx_adout, & |
|---|
| 547 | & zgcb_adout, & |
|---|
| 548 | & zr & |
|---|
| 549 | & ) |
|---|
| 550 | |
|---|
| 551 | |
|---|
| 552 | |
|---|
| 553 | END SUBROUTINE sol_sor_adj_tst |
|---|
| 554 | #if defined key_tst_tlm |
|---|
| 555 | SUBROUTINE sol_sor_tlm_tst( kumadt ) |
|---|
| 556 | !!----------------------------------------------------------------------- |
|---|
| 557 | !! |
|---|
| 558 | !! *** ROUTINE example_adj_tst *** |
|---|
| 559 | !! |
|---|
| 560 | !! ** Purpose : Test the tangent routine. |
|---|
| 561 | !! |
|---|
| 562 | !! ** Method : Verify the tangent with Taylor expansion |
|---|
| 563 | !! |
|---|
| 564 | !! M(x+hdx) = M(x) + L(hdx) + O(h^2) |
|---|
| 565 | !! |
|---|
| 566 | !! where L = tangent routine |
|---|
| 567 | !! M = direct routine |
|---|
| 568 | !! dx = input perturbation (random field) |
|---|
| 569 | !! h = ration on perturbation |
|---|
| 570 | !! |
|---|
| 571 | !! In the tangent test we verify that: |
|---|
| 572 | !! M(x+h*dx) - M(x) |
|---|
| 573 | !! g(h) = ------------------ ---> 1 as h ---> 0 |
|---|
| 574 | !! L(h*dx) |
|---|
| 575 | !! and |
|---|
| 576 | !! g(h) - 1 |
|---|
| 577 | !! f(h) = ---------- ---> k (costant) as h ---> 0 |
|---|
| 578 | !! p |
|---|
| 579 | !! |
|---|
| 580 | !! History : |
|---|
| 581 | !! ! 10-02 (A. Vigilant) |
|---|
| 582 | !!----------------------------------------------------------------------- |
|---|
| 583 | #if defined key_tam |
|---|
| 584 | !! * Modules used |
|---|
| 585 | USE solsor ! Red-Black Successive Over-Relaxation solver |
|---|
| 586 | USE tamtrj ! writing out state trajectory |
|---|
| 587 | USE par_tlm, ONLY: & |
|---|
| 588 | & tlm_bch, & |
|---|
| 589 | & cur_loop, & |
|---|
| 590 | & h_ratio |
|---|
| 591 | USE istate_mod |
|---|
| 592 | USE wzvmod ! vertical velocity |
|---|
| 593 | USE gridrandom, ONLY: & |
|---|
| 594 | & grid_rd_sd |
|---|
| 595 | USE trj_tam |
|---|
| 596 | USE sol_oce , ONLY: & ! ocean dynamics and tracers variables |
|---|
| 597 | & gcb, gcx, ncut |
|---|
| 598 | USE oce , ONLY: & ! |
|---|
| 599 | & ua, ub, un |
|---|
| 600 | USE opatam_tst_ini, ONLY: & |
|---|
| 601 | & tlm_namrd |
|---|
| 602 | USE tamctl, ONLY: & ! Control parameters |
|---|
| 603 | & numtan, numtan_sc |
|---|
| 604 | !! * Arguments |
|---|
| 605 | INTEGER, INTENT(IN) :: & |
|---|
| 606 | & kumadt ! Output unit |
|---|
| 607 | |
|---|
| 608 | !! * Local declarations |
|---|
| 609 | INTEGER :: & |
|---|
| 610 | & ji, & ! dummy loop indices |
|---|
| 611 | & jj, & |
|---|
| 612 | & jk, & |
|---|
| 613 | & kindic,& ! flags fo solver convergence |
|---|
| 614 | & kt ! number of iteration |
|---|
| 615 | INTEGER, DIMENSION(jpi,jpj) :: & |
|---|
| 616 | & iseed_2d ! 2D seed for the random number generator |
|---|
| 617 | REAL(KIND=wp) :: & |
|---|
| 618 | & zsp1, zsp2, zsp3, & ! scalar product |
|---|
| 619 | & zsp_gcb, zsp_gcx, & |
|---|
| 620 | & zsp, & |
|---|
| 621 | & gamma, & |
|---|
| 622 | & zgsp1, & |
|---|
| 623 | & zgsp2, & |
|---|
| 624 | & zgsp3, & |
|---|
| 625 | & zgsp4, & |
|---|
| 626 | & zgsp5, & |
|---|
| 627 | & zgsp6, & |
|---|
| 628 | & zgsp7 |
|---|
| 629 | REAL(KIND=wp), DIMENSION(:,:), ALLOCATABLE :: & |
|---|
| 630 | & zgcb_tlin , & ! Tangent input |
|---|
| 631 | & zgcx_tlin , & ! Tangent input |
|---|
| 632 | & zgcb_out , & ! Direct output |
|---|
| 633 | & zgcx_out , & ! Direct output |
|---|
| 634 | & zgcb_wop , & ! Direct output without perturbation |
|---|
| 635 | & zgcx_wop , & ! Direct output without perturbation |
|---|
| 636 | & zr ! 3D random field |
|---|
| 637 | CHARACTER(LEN=14) :: cl_name |
|---|
| 638 | CHARACTER (LEN=128) :: file_out, file_wop, file_xdx |
|---|
| 639 | CHARACTER (LEN=90) :: FMT |
|---|
| 640 | REAL(KIND=wp), DIMENSION(100):: & |
|---|
| 641 | & zscgcb,zscgcx, & |
|---|
| 642 | & zscerrgcb, zscerrgcx |
|---|
| 643 | INTEGER, DIMENSION(100):: & |
|---|
| 644 | & iiposgcb, ijposgcb, & |
|---|
| 645 | & iiposgcx, ijposgcx |
|---|
| 646 | INTEGER:: & |
|---|
| 647 | & ii, & |
|---|
| 648 | & isamp=40, & |
|---|
| 649 | & jsamp=40, & |
|---|
| 650 | & numsctlm |
|---|
| 651 | REAL(KIND=wp), DIMENSION(jpi,jpj) :: & |
|---|
| 652 | & zerrgcb, zerrgcx |
|---|
| 653 | |
|---|
| 654 | ! Allocate memory |
|---|
| 655 | |
|---|
| 656 | ALLOCATE( & |
|---|
| 657 | & zgcb_tlin( jpi,jpj), & |
|---|
| 658 | & zgcx_tlin( jpi,jpj), & |
|---|
| 659 | & zgcb_out ( jpi,jpj), & |
|---|
| 660 | & zgcx_out ( jpi,jpj), & |
|---|
| 661 | & zgcb_wop ( jpi,jpj), & |
|---|
| 662 | & zgcx_wop ( jpi,jpj), & |
|---|
| 663 | & zr( jpi,jpj) & |
|---|
| 664 | & ) |
|---|
| 665 | !================================================================== |
|---|
| 666 | ! 1) dx = ( un_tl, vn_tl, hdivn_tl ) and |
|---|
| 667 | ! dy = ( hdivb_tl, hdivn_tl ) |
|---|
| 668 | !================================================================== |
|---|
| 669 | |
|---|
| 670 | !-------------------------------------------------------------------- |
|---|
| 671 | ! Reset the tangent and adjoint variables |
|---|
| 672 | !-------------------------------------------------------------------- |
|---|
| 673 | zgcb_tlin( :,:) = 0.0_wp |
|---|
| 674 | zgcx_tlin( :,:) = 0.0_wp |
|---|
| 675 | zgcb_out ( :,:) = 0.0_wp |
|---|
| 676 | zgcx_out ( :,:) = 0.0_wp |
|---|
| 677 | zgcb_wop ( :,:) = 0.0_wp |
|---|
| 678 | zgcx_wop ( :,:) = 0.0_wp |
|---|
| 679 | zr( :,:) = 0.0_wp |
|---|
| 680 | |
|---|
| 681 | !-------------------------------------------------------------------- |
|---|
| 682 | ! Initialize the tangent input with random noise: dx |
|---|
| 683 | !-------------------------------------------------------------------- |
|---|
| 684 | |
|---|
| 685 | !-------------------------------------------------------------------- |
|---|
| 686 | ! Output filename Xn=F(X0) |
|---|
| 687 | !-------------------------------------------------------------------- |
|---|
| 688 | CALL tlm_namrd |
|---|
| 689 | gamma = h_ratio |
|---|
| 690 | file_wop='trj_wop_solsor' |
|---|
| 691 | file_xdx='trj_xdx_solsor' |
|---|
| 692 | !-------------------------------------------------------------------- |
|---|
| 693 | ! Initialize the tangent input with random noise: dx |
|---|
| 694 | !-------------------------------------------------------------------- |
|---|
| 695 | IF (( cur_loop .NE. 0) .OR. ( gamma .NE. 0.0_wp) )THEN |
|---|
| 696 | CALL grid_rd_sd( 596035, zr, c_solver_pt, 0.0_wp, stdgc) |
|---|
| 697 | DO jj = nldj, nlej |
|---|
| 698 | DO ji = nldi, nlei |
|---|
| 699 | zgcb_tlin(ji,jj) = zr(ji,jj) |
|---|
| 700 | END DO |
|---|
| 701 | END DO |
|---|
| 702 | CALL grid_rd_sd( 264792, zr, c_solver_pt, 0.0_wp, stdgc) |
|---|
| 703 | DO jj = nldj, nlej |
|---|
| 704 | DO ji = nldi, nlei |
|---|
| 705 | zgcx_tlin(ji,jj) = zr(ji,jj) |
|---|
| 706 | END DO |
|---|
| 707 | END DO |
|---|
| 708 | ENDIF |
|---|
| 709 | |
|---|
| 710 | !-------------------------------------------------------------------- |
|---|
| 711 | ! Complete Init for Direct |
|---|
| 712 | !------------------------------------------------------------------- |
|---|
| 713 | CALL istate_p |
|---|
| 714 | |
|---|
| 715 | ! *** initialize the reference trajectory |
|---|
| 716 | ! ------------ |
|---|
| 717 | |
|---|
| 718 | ! gcx (:,:) = ( ua(:,:,1) + ub(:,:,1) ) / 10.0_wp |
|---|
| 719 | ! gcb (:,:) = ( ua(:,:,3) + ub(:,:,3) ) / 10.0_wp |
|---|
| 720 | CALL trj_rea( nit000-1, 1 ) |
|---|
| 721 | CALL trj_rea( nit000, 1 ) |
|---|
| 722 | gcx (:,:) = un(:,:,1) / 10.0_wp |
|---|
| 723 | gcb (:,:) = un(:,:,3) / 10.0_wp |
|---|
| 724 | |
|---|
| 725 | IF (( cur_loop .NE. 0) .OR. ( gamma .NE. 0.0_wp) )THEN |
|---|
| 726 | zgcb_tlin(:,:) = gamma * zgcb_tlin(:,:) |
|---|
| 727 | gcb(:,:) = gcb(:,:) + zgcb_tlin(:,:) |
|---|
| 728 | |
|---|
| 729 | zgcx_tlin(:,:) = gamma * zgcx_tlin(:,:) |
|---|
| 730 | gcx(:,:) = gcx(:,:) + zgcx_tlin(:,:) |
|---|
| 731 | ENDIF |
|---|
| 732 | |
|---|
| 733 | !-------------------------------------------------------------------- |
|---|
| 734 | ! Compute the direct model F(X0,t=n) = Xn |
|---|
| 735 | !-------------------------------------------------------------------- |
|---|
| 736 | kindic=0 |
|---|
| 737 | ncut=1 |
|---|
| 738 | IF ( tlm_bch /= 2 ) CALL sol_sor(kindic) |
|---|
| 739 | IF ( tlm_bch == 0 ) CALL trj_wri_spl(file_wop) |
|---|
| 740 | IF ( tlm_bch == 1 ) CALL trj_wri_spl(file_xdx) |
|---|
| 741 | !-------------------------------------------------------------------- |
|---|
| 742 | ! Compute the Tangent |
|---|
| 743 | !-------------------------------------------------------------------- |
|---|
| 744 | IF ( tlm_bch == 2 ) THEN |
|---|
| 745 | !-------------------------------------------------------------------- |
|---|
| 746 | ! Initialize the tangent variables: dy^* = W dy |
|---|
| 747 | !-------------------------------------------------------------------- |
|---|
| 748 | gcr_tl(:,:) = 0.0_wp |
|---|
| 749 | gcb_tl (:,:) = zgcb_tlin (:,:) |
|---|
| 750 | gcx_tl (:,:) = zgcx_tlin (:,:) |
|---|
| 751 | |
|---|
| 752 | !----------------------------------------------------------------------- |
|---|
| 753 | ! Initialization of the dynamics and tracer fields for the tangent |
|---|
| 754 | !----------------------------------------------------------------------- |
|---|
| 755 | ncut=1 !reset indicator of solver convergence |
|---|
| 756 | CALL sol_sor_tan(nit000, kindic) |
|---|
| 757 | !-------------------------------------------------------------------- |
|---|
| 758 | ! Compute the scalar product: ( L(t0,tn) gamma dx0 ) ) |
|---|
| 759 | !-------------------------------------------------------------------- |
|---|
| 760 | |
|---|
| 761 | zsp_gcx = DOT_PRODUCT( gcx_tl, gcx_tl ) |
|---|
| 762 | zsp2 = zsp_gcx |
|---|
| 763 | !-------------------------------------------------------------------- |
|---|
| 764 | ! Storing data |
|---|
| 765 | !-------------------------------------------------------------------- |
|---|
| 766 | CALL trj_rd_spl(file_wop) |
|---|
| 767 | zgcx_wop (:,:) = gcx (:,:) |
|---|
| 768 | CALL trj_rd_spl(file_xdx) |
|---|
| 769 | zgcx_out (:,:) = gcx (:,:) |
|---|
| 770 | !-------------------------------------------------------------------- |
|---|
| 771 | ! Compute the Linearization Error |
|---|
| 772 | ! Nn = M( X0+gamma.dX0, t0,tn) - M(X0, t0,tn) |
|---|
| 773 | ! and |
|---|
| 774 | ! Compute the Linearization Error |
|---|
| 775 | ! En = Nn -TL(gamma.dX0, t0,tn) |
|---|
| 776 | !-------------------------------------------------------------------- |
|---|
| 777 | ! Warning: Here we re-use local variables z()_out and z()_wop |
|---|
| 778 | ii=0 |
|---|
| 779 | DO jj = 1, jpj |
|---|
| 780 | DO ji = 1, jpi |
|---|
| 781 | zgcx_out (ji,jj) = zgcx_out (ji,jj) - zgcx_wop (ji,jj) |
|---|
| 782 | zgcx_wop (ji,jj) = zgcx_out (ji,jj) - gcx_tl (ji,jj) |
|---|
| 783 | IF ( gcx_tl(ji,jj) .NE. 0.0_wp ) zerrgcx(ji,jj) = zgcx_out(ji,jj)/gcx_tl(ji,jj) |
|---|
| 784 | IF( (MOD(ji, isamp) .EQ. 0) .AND. & |
|---|
| 785 | & (MOD(jj, jsamp) .EQ. 0) ) THEN |
|---|
| 786 | ii = ii+1 |
|---|
| 787 | iiposgcx(ii) = ji |
|---|
| 788 | ijposgcx(ii) = jj |
|---|
| 789 | IF ( INT(tmask(ji,jj,1)) .NE. 0) THEN |
|---|
| 790 | zscgcx (ii) = zgcx_wop(ji,jj) |
|---|
| 791 | zscerrgcx (ii) = ( zerrgcx(ji,jj) - 1.0_wp ) / gamma |
|---|
| 792 | ENDIF |
|---|
| 793 | ENDIF |
|---|
| 794 | END DO |
|---|
| 795 | END DO |
|---|
| 796 | |
|---|
| 797 | zsp_gcx = DOT_PRODUCT( zgcx_out, zgcx_out ) |
|---|
| 798 | |
|---|
| 799 | zsp1 = zsp_gcx |
|---|
| 800 | |
|---|
| 801 | zsp_gcx = DOT_PRODUCT( zgcx_wop, zgcx_wop ) |
|---|
| 802 | |
|---|
| 803 | zsp3 = zsp_gcx |
|---|
| 804 | !-------------------------------------------------------------------- |
|---|
| 805 | ! Print the linearization error En - norme 2 |
|---|
| 806 | !-------------------------------------------------------------------- |
|---|
| 807 | ! 14 char:'12345678901234' |
|---|
| 808 | cl_name = 'sol_sor: En ' |
|---|
| 809 | zsp = SQRT(zsp3) |
|---|
| 810 | zgsp5 = zsp |
|---|
| 811 | CALL prntst_tlm( cl_name, kumadt, zsp, h_ratio ) |
|---|
| 812 | |
|---|
| 813 | !-------------------------------------------------------------------- |
|---|
| 814 | ! Compute TLM norm2 |
|---|
| 815 | !-------------------------------------------------------------------- |
|---|
| 816 | zsp = SQRT(zsp2) |
|---|
| 817 | |
|---|
| 818 | zgsp4 = zsp |
|---|
| 819 | cl_name = 'sol_sor: Ln2 ' |
|---|
| 820 | CALL prntst_tlm( cl_name, kumadt, zsp, h_ratio ) |
|---|
| 821 | |
|---|
| 822 | !-------------------------------------------------------------------- |
|---|
| 823 | ! Print the linearization error Nn - norme 2 |
|---|
| 824 | !-------------------------------------------------------------------- |
|---|
| 825 | zsp = SQRT(zsp1) |
|---|
| 826 | |
|---|
| 827 | cl_name = 'solsor:Mhdx-Mx' |
|---|
| 828 | CALL prntst_tlm( cl_name, kumadt, zsp, h_ratio ) |
|---|
| 829 | |
|---|
| 830 | zgsp3 = SQRT( zsp3/zsp2 ) |
|---|
| 831 | zgsp7 = zgsp3/gamma |
|---|
| 832 | zgsp1 = zsp |
|---|
| 833 | zgsp2 = zgsp1 / zgsp4 |
|---|
| 834 | zgsp6 = (zgsp2 - 1.0_wp)/gamma |
|---|
| 835 | |
|---|
| 836 | FMT = "(A8,2X,I4.4,2X,E6.1,2X,E20.13,2X,E20.13,2X,E20.13,2X,E20.13,2X,E20.13,2X,E20.13,2X,E20.13)" |
|---|
| 837 | WRITE(numtan,FMT) 'solsor ', cur_loop, h_ratio, zgsp1, zgsp2, zgsp3, zgsp4, zgsp5, zgsp6, zgsp7 |
|---|
| 838 | !-------------------------------------------------------------------- |
|---|
| 839 | ! Unitary calculus |
|---|
| 840 | !-------------------------------------------------------------------- |
|---|
| 841 | FMT = "(A8,2X,A8,2X,I4.4,2X,E6.1,2X,I4.4,2X,I4.4,2X,I4.4,2X,E20.13,1X)" |
|---|
| 842 | cl_name ='sol_sor ' |
|---|
| 843 | IF(lwp) THEN |
|---|
| 844 | DO ii=1, 100, 1 |
|---|
| 845 | IF ( zscgcx(ii) .NE. 0.0_wp ) WRITE(numtan_sc,FMT) cl_name, 'zscgcx ', & |
|---|
| 846 | & cur_loop, h_ratio, ii, iiposgcx(ii), ijposgcx(ii), zscgcx(ii) |
|---|
| 847 | ENDDO |
|---|
| 848 | DO ii=1, 100, 1 |
|---|
| 849 | IF ( zscerrgcx(ii) .NE. 0.0_wp ) WRITE(numtan_sc,FMT) cl_name, 'zscerrgcx ', & |
|---|
| 850 | & cur_loop, h_ratio, ii, iiposgcx(ii), ijposgcx(ii), zscerrgcx(ii) |
|---|
| 851 | ENDDO |
|---|
| 852 | ! write separator |
|---|
| 853 | WRITE(numtan_sc,"(A4)") '====' |
|---|
| 854 | ENDIF |
|---|
| 855 | |
|---|
| 856 | ENDIF |
|---|
| 857 | |
|---|
| 858 | DEALLOCATE( & |
|---|
| 859 | & zgcb_tlin, & |
|---|
| 860 | & zgcx_tlin, & |
|---|
| 861 | & zgcb_out , & |
|---|
| 862 | & zgcx_out , & |
|---|
| 863 | & zgcb_wop , & |
|---|
| 864 | & zgcx_wop , & |
|---|
| 865 | & zr & |
|---|
| 866 | & ) |
|---|
| 867 | #else |
|---|
| 868 | !! * Arguments |
|---|
| 869 | INTEGER, INTENT(IN) :: & |
|---|
| 870 | & kumadt ! Output unit |
|---|
| 871 | ! dummy routine |
|---|
| 872 | #endif |
|---|
| 873 | END SUBROUTINE sol_sor_tlm_tst |
|---|
| 874 | #endif |
|---|
| 875 | END MODULE solsor_tam |
|---|
