src/gromacs/gmxlib/nonbonded/nb_free_energy.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
   5  * Copyright (c) 2001-2004, The GROMACS development team.
   6  * Copyright (c) 2013,2014, by the GROMACS development team, led by
   7  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   8  * and including many others, as listed in the AUTHORS file in the
   9  * top-level source directory and at http://www.gromacs.org.
  10  *
  11  * GROMACS is free software; you can redistribute it and/or
  12  * modify it under the terms of the GNU Lesser General Public License
  13  * as published by the Free Software Foundation; either version 2.1
  14  * of the License, or (at your option) any later version.
  15  *
  16  * GROMACS is distributed in the hope that it will be useful,
  17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  * Lesser General Public License for more details.
  20  *
  21  * You should have received a copy of the GNU Lesser General Public
  22  * License along with GROMACS; if not, see
  23  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  24  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  25  *
  26  * If you want to redistribute modifications to GROMACS, please
  27  * consider that scientific software is very special. Version
  28  * control is crucial - bugs must be traceable. We will be happy to
  29  * consider code for inclusion in the official distribution, but
  30  * derived work must not be called official GROMACS. Details are found
  31  * in the README & COPYING files - if they are missing, get the
  32  * official version at http://www.gromacs.org.
  33  *
  34  * To help us fund GROMACS development, we humbly ask that you cite
  35  * the research papers on the package. Check out http://www.gromacs.org.
  36  */
  37 #ifdef HAVE_CONFIG_H
  38 #include <config.h>
  39 #endif
  40
  41 #include <math.h>
  42
  43 #include "gromacs/math/vec.h"
  44 #include "typedefs.h"
  45 #include "nonbonded.h"
  46 #include "nb_kernel.h"
  47 #include "nrnb.h"
  48 #include "macros.h"
  49 #include "nb_free_energy.h"
  50
  51 #include "gromacs/utility/fatalerror.h"
  52
  53 void
  54 gmx_nb_free_energy_kernel(const t_nblist * gmx_restrict    nlist,
  55                           rvec * gmx_restrict              xx,
  56                           rvec * gmx_restrict              ff,
  57                           t_forcerec * gmx_restrict        fr,
  58                           const t_mdatoms * gmx_restrict   mdatoms,
  59                           nb_kernel_data_t * gmx_restrict  kernel_data,
  60                           t_nrnb * gmx_restrict            nrnb)
  61 {
  62
  63 #define  STATE_A  0
  64 #define  STATE_B  1
  65 #define  NSTATES  2
  66     int           i, j, n, ii, is3, ii3, k, nj0, nj1, jnr, j3, ggid;
  67     real          shX, shY, shZ;
  68     real          Fscal, FscalC[NSTATES], FscalV[NSTATES], tx, ty, tz;
  69     real          Vcoul[NSTATES], Vvdw[NSTATES];
  70     real          rinv6, r, rt, rtC, rtV;
  71     real          iqA, iqB;
  72     real          qq[NSTATES], vctot, krsq;
  73     int           ntiA, ntiB, tj[NSTATES];
  74     real          Vvdw6, Vvdw12, vvtot;
  75     real          ix, iy, iz, fix, fiy, fiz;
  76     real          dx, dy, dz, rsq, rinv;
  77     real          c6[NSTATES], c12[NSTATES], c6grid;
  78     real          LFC[NSTATES], LFV[NSTATES], DLF[NSTATES];
  79     double        dvdl_coul, dvdl_vdw;
  80     real          lfac_coul[NSTATES], dlfac_coul[NSTATES], lfac_vdw[NSTATES], dlfac_vdw[NSTATES];
  81     real          sigma6[NSTATES], alpha_vdw_eff, alpha_coul_eff, sigma2_def, sigma2_min;
  82     real          rp, rpm2, rC, rV, rinvC, rpinvC, rinvV, rpinvV;
  83     real          sigma2[NSTATES], sigma_pow[NSTATES], sigma_powm2[NSTATES], rs, rs2;
  84     int           do_tab, tab_elemsize;
  85     int           n0, n1C, n1V, nnn;
  86     real          Y, F, G, H, Fp, Geps, Heps2, epsC, eps2C, epsV, eps2V, VV, FF;
  87     int           icoul, ivdw;
  88     int           nri;
  89     const int *   iinr;
  90     const int *   jindex;
  91     const int *   jjnr;
  92     const int *   shift;
  93     const int *   gid;
  94     const int *   typeA;
  95     const int *   typeB;
  96     int           ntype;
  97     const real *  shiftvec;
  98     real          dvdl_part;
  99     real *        fshift;
 100     real          tabscale = 0;
 101     const real *  VFtab    = NULL;
 102     const real *  x;
 103     real *        f;
 104     real          facel, krf, crf;
 105     const real *  chargeA;
 106     const real *  chargeB;
 107     real          sigma6_min, sigma6_def, lam_power, sc_power, sc_r_power;
 108     real          alpha_coul, alpha_vdw, lambda_coul, lambda_vdw, ewc_lj;
 109     real          ewcljrsq, ewclj, ewclj2, exponent, poly, vvdw_disp, vvdw_rep, sh_lj_ewald;
 110     real          ewclj6;
 111     const real *  nbfp, *nbfp_grid;
 112     real *        dvdl;
 113     real *        Vv;
 114     real *        Vc;
 115     gmx_bool      bDoForces, bDoShiftForces, bDoPotential;
 116     real          rcoulomb, rvdw, sh_invrc6;
 117     gmx_bool      bExactElecCutoff, bExactVdwCutoff, bExactCutoffAll;
 118     gmx_bool      bEwald, bEwaldLJ;
 119     real          rcutoff_max2;
 120     const real *  tab_ewald_F_lj;
 121     const real *  tab_ewald_V_lj;
 122     real          d, d2, sw, dsw, rinvcorr;
 123     real          elec_swV3, elec_swV4, elec_swV5, elec_swF2, elec_swF3, elec_swF4;
 124     real          vdw_swV3, vdw_swV4, vdw_swV5, vdw_swF2, vdw_swF3, vdw_swF4;
 125     gmx_bool      bConvertEwaldToCoulomb, bConvertLJEwaldToLJ6;
 126     gmx_bool      bComputeVdwInteraction, bComputeElecInteraction;
 127     const real *  ewtab;
 128     int           ewitab;
 129     real          ewrt, eweps, ewtabscale, ewtabhalfspace, sh_ewald;
 130
 131     sh_ewald            = fr->ic->sh_ewald;
 132     ewtab               = fr->ic->tabq_coul_FDV0;
 133     ewtabscale          = fr->ic->tabq_scale;
 134     ewtabhalfspace      = 0.5/ewtabscale;
 135     tab_ewald_F_lj      = fr->ic->tabq_vdw_F;
 136     tab_ewald_V_lj      = fr->ic->tabq_vdw_V;
 137
 138     x                   = xx[0];
 139     f                   = ff[0];
 140
 141     fshift              = fr->fshift[0];
 142
 143     nri                 = nlist->nri;
 144     iinr                = nlist->iinr;
 145     jindex              = nlist->jindex;
 146     jjnr                = nlist->jjnr;
 147     icoul               = nlist->ielec;
 148     ivdw                = nlist->ivdw;
 149     shift               = nlist->shift;
 150     gid                 = nlist->gid;
 151
 152     shiftvec            = fr->shift_vec[0];
 153     chargeA             = mdatoms->chargeA;
 154     chargeB             = mdatoms->chargeB;
 155     facel               = fr->epsfac;
 156     krf                 = fr->k_rf;
 157     crf                 = fr->c_rf;
 158     ewc_lj              = fr->ewaldcoeff_lj;
 159     Vc                  = kernel_data->energygrp_elec;
 160     typeA               = mdatoms->typeA;
 161     typeB               = mdatoms->typeB;
 162     ntype               = fr->ntype;
 163     nbfp                = fr->nbfp;
 164     nbfp_grid           = fr->ljpme_c6grid;
 165     Vv                  = kernel_data->energygrp_vdw;
 166     lambda_coul         = kernel_data->lambda[efptCOUL];
 167     lambda_vdw          = kernel_data->lambda[efptVDW];
 168     dvdl                = kernel_data->dvdl;
 169     alpha_coul          = fr->sc_alphacoul;
 170     alpha_vdw           = fr->sc_alphavdw;
 171     lam_power           = fr->sc_power;
 172     sc_r_power          = fr->sc_r_power;
 173     sigma6_def          = fr->sc_sigma6_def;
 174     sigma6_min          = fr->sc_sigma6_min;
 175     bDoForces           = kernel_data->flags & GMX_NONBONDED_DO_FORCE;
 176     bDoShiftForces      = kernel_data->flags & GMX_NONBONDED_DO_SHIFTFORCE;
 177     bDoPotential        = kernel_data->flags & GMX_NONBONDED_DO_POTENTIAL;
 178
 179     rcoulomb            = fr->rcoulomb;
 180     rvdw                = fr->rvdw;
 181     sh_invrc6           = fr->ic->sh_invrc6;
 182     sh_lj_ewald         = fr->ic->sh_lj_ewald;
 183     ewclj               = fr->ewaldcoeff_lj;
 184     ewclj2              = ewclj*ewclj;
 185     ewclj6              = ewclj2*ewclj2*ewclj2;
 186
 187     if (fr->coulomb_modifier == eintmodPOTSWITCH)
 188     {
 189         d               = fr->rcoulomb-fr->rcoulomb_switch;
 190         elec_swV3       = -10.0/(d*d*d);
 191         elec_swV4       =  15.0/(d*d*d*d);
 192         elec_swV5       =  -6.0/(d*d*d*d*d);
 193         elec_swF2       = -30.0/(d*d*d);
 194         elec_swF3       =  60.0/(d*d*d*d);
 195         elec_swF4       = -30.0/(d*d*d*d*d);
 196     }
 197     else
 198     {
 199         /* Avoid warnings from stupid compilers (looking at you, Clang!) */
 200         elec_swV3 = elec_swV4 = elec_swV5 = elec_swF2 = elec_swF3 = elec_swF4 = 0.0;
 201     }
 202
 203     if (fr->vdw_modifier == eintmodPOTSWITCH)
 204     {
 205         d               = fr->rvdw-fr->rvdw_switch;
 206         vdw_swV3        = -10.0/(d*d*d);
 207         vdw_swV4        =  15.0/(d*d*d*d);
 208         vdw_swV5        =  -6.0/(d*d*d*d*d);
 209         vdw_swF2        = -30.0/(d*d*d);
 210         vdw_swF3        =  60.0/(d*d*d*d);
 211         vdw_swF4        = -30.0/(d*d*d*d*d);
 212     }
 213     else
 214     {
 215         /* Avoid warnings from stupid compilers (looking at you, Clang!) */
 216         vdw_swV3 = vdw_swV4 = vdw_swV5 = vdw_swF2 = vdw_swF3 = vdw_swF4 = 0.0;
 217     }
 218
 219     if (fr->cutoff_scheme == ecutsVERLET)
 220     {
 221         const interaction_const_t *ic;
 222
 223         ic = fr->ic;
 224         if (EVDW_PME(ic->vdwtype))
 225         {
 226             ivdw         = GMX_NBKERNEL_VDW_LJEWALD;
 227         }
 228         else
 229         {
 230             ivdw         = GMX_NBKERNEL_VDW_LENNARDJONES;
 231         }
 232
 233         if (ic->eeltype == eelCUT || EEL_RF(ic->eeltype))
 234         {
 235             icoul        = GMX_NBKERNEL_ELEC_REACTIONFIELD;
 236         }
 237         else if (EEL_PME_EWALD(ic->eeltype))
 238         {
 239             icoul        = GMX_NBKERNEL_ELEC_EWALD;
 240         }
 241         else
 242         {
 243             gmx_incons("Unsupported eeltype with Verlet and free-energy");
 244         }
 245
 246         bExactElecCutoff = TRUE;
 247         bExactVdwCutoff  = TRUE;
 248     }
 249     else
 250     {
 251         bExactElecCutoff = (fr->coulomb_modifier != eintmodNONE) || fr->eeltype == eelRF_ZERO;
 252         bExactVdwCutoff  = (fr->vdw_modifier != eintmodNONE);
 253     }
 254
 255     bExactCutoffAll = (bExactElecCutoff && bExactVdwCutoff);
 256     rcutoff_max2    = max(fr->rcoulomb, fr->rvdw);
 257     rcutoff_max2    = rcutoff_max2*rcutoff_max2;
 258
 259     bEwald          = (icoul == GMX_NBKERNEL_ELEC_EWALD);
 260     bEwaldLJ        = (ivdw == GMX_NBKERNEL_VDW_LJEWALD);
 261
 262     /* For Ewald/PME interactions we cannot easily apply the soft-core component to
 263      * reciprocal space. When we use vanilla (not switch/shift) Ewald interactions, we
 264      * can apply the small trick of subtracting the _reciprocal_ space contribution
 265      * in this kernel, and instead apply the free energy interaction to the 1/r
 266      * (standard coulomb) interaction.
 267      *
 268      * However, we cannot use this approach for switch-modified since we would then
 269      * effectively end up evaluating a significantly different interaction here compared to the
 270      * normal (non-free-energy) kernels, either by applying a cutoff at a different
 271      * position than what the user requested, or by switching different
 272      * things (1/r rather than short-range Ewald). For these settings, we just
 273      * use the traditional short-range Ewald interaction in that case.
 274      */
 275     bConvertEwaldToCoulomb = (bEwald && (fr->coulomb_modifier != eintmodPOTSWITCH));
 276     /* For now the below will always be true (since LJ-PME only works with Shift in Gromacs-5.0),
 277      * but writing it this way means we stay in sync with coulomb, and it avoids future bugs.
 278      */
 279     bConvertLJEwaldToLJ6   = (bEwaldLJ && (fr->vdw_modifier   != eintmodPOTSWITCH));
 280
 281     /* We currently don't implement exclusion correction, needed with the Verlet cut-off scheme, without conversion */
 282     if (fr->cutoff_scheme == ecutsVERLET &&
 283         ((bEwald   && !bConvertEwaldToCoulomb) ||
 284          (bEwaldLJ && !bConvertLJEwaldToLJ6)))
 285     {
 286         gmx_incons("Unimplemented non-bonded setup");
 287     }
 288
 289     /* fix compiler warnings */
 290     nj1   = 0;
 291     n1C   = n1V   = 0;
 292     epsC  = epsV  = 0;
 293     eps2C = eps2V = 0;
 294
 295     dvdl_coul  = 0;
 296     dvdl_vdw   = 0;
 297
 298     /* Lambda factor for state A, 1-lambda*/
 299     LFC[STATE_A] = 1.0 - lambda_coul;
 300     LFV[STATE_A] = 1.0 - lambda_vdw;
 301
 302     /* Lambda factor for state B, lambda*/
 303     LFC[STATE_B] = lambda_coul;
 304     LFV[STATE_B] = lambda_vdw;
 305
 306     /*derivative of the lambda factor for state A and B */
 307     DLF[STATE_A] = -1;
 308     DLF[STATE_B] = 1;
 309
 310     for (i = 0; i < NSTATES; i++)
 311     {
 312         lfac_coul[i]  = (lam_power == 2 ? (1-LFC[i])*(1-LFC[i]) : (1-LFC[i]));
 313         dlfac_coul[i] = DLF[i]*lam_power/sc_r_power*(lam_power == 2 ? (1-LFC[i]) : 1);
 314         lfac_vdw[i]   = (lam_power == 2 ? (1-LFV[i])*(1-LFV[i]) : (1-LFV[i]));
 315         dlfac_vdw[i]  = DLF[i]*lam_power/sc_r_power*(lam_power == 2 ? (1-LFV[i]) : 1);
 316     }
 317     /* precalculate */
 318     sigma2_def = pow(sigma6_def, 1.0/3.0);
 319     sigma2_min = pow(sigma6_min, 1.0/3.0);
 320
 321     /* Ewald (not PME) table is special (icoul==enbcoulFEWALD) */
 322
 323     do_tab = (icoul == GMX_NBKERNEL_ELEC_CUBICSPLINETABLE ||
 324               ivdw == GMX_NBKERNEL_VDW_CUBICSPLINETABLE);
 325     if (do_tab)
 326     {
 327         tabscale         = kernel_data->table_elec_vdw->scale;
 328         VFtab            = kernel_data->table_elec_vdw->data;
 329         /* we always use the combined table here */
 330         tab_elemsize     = 12;
 331     }
 332
 333     for (n = 0; (n < nri); n++)
 334     {
 335         int npair_within_cutoff;
 336
 337         npair_within_cutoff = 0;
 338
 339         is3              = 3*shift[n];
 340         shX              = shiftvec[is3];
 341         shY              = shiftvec[is3+1];
 342         shZ              = shiftvec[is3+2];
 343         nj0              = jindex[n];
 344         nj1              = jindex[n+1];
 345         ii               = iinr[n];
 346         ii3              = 3*ii;
 347         ix               = shX + x[ii3+0];
 348         iy               = shY + x[ii3+1];
 349         iz               = shZ + x[ii3+2];
 350         iqA              = facel*chargeA[ii];
 351         iqB              = facel*chargeB[ii];
 352         ntiA             = 2*ntype*typeA[ii];
 353         ntiB             = 2*ntype*typeB[ii];
 354         vctot            = 0;
 355         vvtot            = 0;
 356         fix              = 0;
 357         fiy              = 0;
 358         fiz              = 0;
 359
 360         for (k = nj0; (k < nj1); k++)
 361         {
 362             jnr              = jjnr[k];
 363             j3               = 3*jnr;
 364             dx               = ix - x[j3];
 365             dy               = iy - x[j3+1];
 366             dz               = iz - x[j3+2];
 367             rsq              = dx*dx + dy*dy + dz*dz;
 368
 369             if (bExactCutoffAll && rsq >= rcutoff_max2)
 370             {
 371                 /* We save significant time by skipping all code below.
 372                  * Note that with soft-core interactions, the actual cut-off
 373                  * check might be different. But since the soft-core distance
 374                  * is always larger than r, checking on r here is safe.
 375                  */
 376                 continue;
 377             }
 378             npair_within_cutoff++;
 379
 380             if (rsq > 0)
 381             {
 382                 rinv         = gmx_invsqrt(rsq);
 383                 r            = rsq*rinv;
 384             }
 385             else
 386             {
 387                 /* The force at r=0 is zero, because of symmetry.
 388                  * But note that the potential is in general non-zero,
 389                  * since the soft-cored r will be non-zero.
 390                  */
 391                 rinv         = 0;
 392                 r            = 0;
 393             }
 394
 395             if (sc_r_power == 6.0)
 396             {
 397                 rpm2             = rsq*rsq;  /* r4 */
 398                 rp               = rpm2*rsq; /* r6 */
 399             }
 400             else if (sc_r_power == 48.0)
 401             {
 402                 rp               = rsq*rsq*rsq; /* r6 */
 403                 rp               = rp*rp;       /* r12 */
 404                 rp               = rp*rp;       /* r24 */
 405                 rp               = rp*rp;       /* r48 */
 406                 rpm2             = rp/rsq;      /* r46 */
 407             }
 408             else
 409             {
 410                 rp             = pow(r, sc_r_power);  /* not currently supported as input, but can handle it */
 411                 rpm2           = rp/rsq;
 412             }
 413
 414             Fscal = 0;
 415
 416             qq[STATE_A]      = iqA*chargeA[jnr];
 417             qq[STATE_B]      = iqB*chargeB[jnr];
 418
 419             tj[STATE_A]      = ntiA+2*typeA[jnr];
 420             tj[STATE_B]      = ntiB+2*typeB[jnr];
 421
 422             if (nlist->excl_fep == NULL || nlist->excl_fep[k])
 423             {
 424                 c6[STATE_A]      = nbfp[tj[STATE_A]];
 425                 c6[STATE_B]      = nbfp[tj[STATE_B]];
 426
 427                 for (i = 0; i < NSTATES; i++)
 428                 {
 429                     c12[i]             = nbfp[tj[i]+1];
 430                     if ((c6[i] > 0) && (c12[i] > 0))
 431                     {
 432                         /* c12 is stored scaled with 12.0 and c6 is scaled with 6.0 - correct for this */
 433                         sigma6[i]       = 0.5*c12[i]/c6[i];
 434                         sigma2[i]       = pow(sigma6[i], 1.0/3.0);
 435                         /* should be able to get rid of this ^^^ internal pow call eventually.  Will require agreement on
 436                            what data to store externally.  Can't be fixed without larger scale changes, so not 4.6 */
 437                         if (sigma6[i] < sigma6_min)   /* for disappearing coul and vdw with soft core at the same time */
 438                         {
 439                             sigma6[i] = sigma6_min;
 440                             sigma2[i] = sigma2_min;
 441                         }
 442                     }
 443                     else
 444                     {
 445                         sigma6[i]       = sigma6_def;
 446                         sigma2[i]       = sigma2_def;
 447                     }
 448                     if (sc_r_power == 6.0)
 449                     {
 450                         sigma_pow[i]    = sigma6[i];
 451                         sigma_powm2[i]  = sigma6[i]/sigma2[i];
 452                     }
 453                     else if (sc_r_power == 48.0)
 454                     {
 455                         sigma_pow[i]    = sigma6[i]*sigma6[i];       /* sigma^12 */
 456                         sigma_pow[i]    = sigma_pow[i]*sigma_pow[i]; /* sigma^24 */
 457                         sigma_pow[i]    = sigma_pow[i]*sigma_pow[i]; /* sigma^48 */
 458                         sigma_powm2[i]  = sigma_pow[i]/sigma2[i];
 459                     }
 460                     else
 461                     {    /* not really supported as input, but in here for testing the general case*/
 462                         sigma_pow[i]    = pow(sigma2[i], sc_r_power/2);
 463                         sigma_powm2[i]  = sigma_pow[i]/(sigma2[i]);
 464                     }
 465                 }
 466
 467                 /* only use softcore if one of the states has a zero endstate - softcore is for avoiding infinities!*/
 468                 if ((c12[STATE_A] > 0) && (c12[STATE_B] > 0))
 469                 {
 470                     alpha_vdw_eff    = 0;
 471                     alpha_coul_eff   = 0;
 472                 }
 473                 else
 474                 {
 475                     alpha_vdw_eff    = alpha_vdw;
 476                     alpha_coul_eff   = alpha_coul;
 477                 }
 478
 479                 for (i = 0; i < NSTATES; i++)
 480                 {
 481                     FscalC[i]    = 0;
 482                     FscalV[i]    = 0;
 483                     Vcoul[i]     = 0;
 484                     Vvdw[i]      = 0;
 485
 486                     /* Only spend time on A or B state if it is non-zero */
 487                     if ( (qq[i] != 0) || (c6[i] != 0) || (c12[i] != 0) )
 488                     {
 489                         /* this section has to be inside the loop because of the dependence on sigma_pow */
 490                         rpinvC         = 1.0/(alpha_coul_eff*lfac_coul[i]*sigma_pow[i]+rp);
 491                         rinvC          = pow(rpinvC, 1.0/sc_r_power);
 492                         rC             = 1.0/rinvC;
 493
 494                         rpinvV         = 1.0/(alpha_vdw_eff*lfac_vdw[i]*sigma_pow[i]+rp);
 495                         rinvV          = pow(rpinvV, 1.0/sc_r_power);
 496                         rV             = 1.0/rinvV;
 497
 498                         if (do_tab)
 499                         {
 500                             rtC        = rC*tabscale;
 501                             n0         = rtC;
 502                             epsC       = rtC-n0;
 503                             eps2C      = epsC*epsC;
 504                             n1C        = tab_elemsize*n0;
 505
 506                             rtV        = rV*tabscale;
 507                             n0         = rtV;
 508                             epsV       = rtV-n0;
 509                             eps2V      = epsV*epsV;
 510                             n1V        = tab_elemsize*n0;
 511                         }
 512
 513                         /* Only process the coulomb interactions if we have charges,
 514                          * and if we either include all entries in the list (no cutoff
 515                          * used in the kernel), or if we are within the cutoff.
 516                          */
 517                         bComputeElecInteraction = !bExactElecCutoff ||
 518                             ( bConvertEwaldToCoulomb && r < rcoulomb) ||
 519                             (!bConvertEwaldToCoulomb && rC < rcoulomb);
 520
 521                         if ( (qq[i] != 0) && bComputeElecInteraction)
 522                         {
 523                             switch (icoul)
 524                             {
 525                                 case GMX_NBKERNEL_ELEC_COULOMB:
 526                                     /* simple cutoff */
 527                                     Vcoul[i]   = qq[i]*rinvC;
 528                                     FscalC[i]  = Vcoul[i];
 529                                     /* The shift for the Coulomb potential is stored in
 530                                      * the RF parameter c_rf, which is 0 without shift.
 531                                      */
 532                                     Vcoul[i]  -= qq[i]*fr->ic->c_rf;
 533                                     break;
 534
 535                                 case GMX_NBKERNEL_ELEC_REACTIONFIELD:
 536                                     /* reaction-field */
 537                                     Vcoul[i]   = qq[i]*(rinvC + krf*rC*rC-crf);
 538                                     FscalC[i]  = qq[i]*(rinvC - 2.0*krf*rC*rC);
 539                                     break;
 540
 541                                 case GMX_NBKERNEL_ELEC_CUBICSPLINETABLE:
 542                                     /* non-Ewald tabulated coulomb */
 543                                     nnn        = n1C;
 544                                     Y          = VFtab[nnn];
 545                                     F          = VFtab[nnn+1];
 546                                     Geps       = epsC*VFtab[nnn+2];
 547                                     Heps2      = eps2C*VFtab[nnn+3];
 548                                     Fp         = F+Geps+Heps2;
 549                                     VV         = Y+epsC*Fp;
 550                                     FF         = Fp+Geps+2.0*Heps2;
 551                                     Vcoul[i]   = qq[i]*VV;
 552                                     FscalC[i]  = -qq[i]*tabscale*FF*rC;
 553                                     break;
 554
 555                                 case GMX_NBKERNEL_ELEC_GENERALIZEDBORN:
 556                                     gmx_fatal(FARGS, "Free energy and GB not implemented.\n");
 557                                     break;
 558
 559                                 case GMX_NBKERNEL_ELEC_EWALD:
 560                                     if (bConvertEwaldToCoulomb)
 561                                     {
 562                                         /* Ewald FEP is done only on the 1/r part */
 563                                         Vcoul[i]   = qq[i]*(rinvC-sh_ewald);
 564                                         FscalC[i]  = qq[i]*rinvC;
 565                                     }
 566                                     else
 567                                     {
 568                                         ewrt      = rC*ewtabscale;
 569                                         ewitab    = (int) ewrt;
 570                                         eweps     = ewrt-ewitab;
 571                                         ewitab    = 4*ewitab;
 572                                         FscalC[i] = ewtab[ewitab]+eweps*ewtab[ewitab+1];
 573                                         rinvcorr  = rinvC-sh_ewald;
 574                                         Vcoul[i]  = qq[i]*(rinvcorr-(ewtab[ewitab+2]-ewtabhalfspace*eweps*(ewtab[ewitab]+FscalC[i])));
 575                                         FscalC[i] = qq[i]*(rinvC-rC*FscalC[i]);
 576                                     }
 577                                     break;
 578
 579                                 case GMX_NBKERNEL_ELEC_NONE:
 580                                     FscalC[i]  = 0.0;
 581                                     Vcoul[i]   = 0.0;
 582                                     break;
 583
 584                                 default:
 585                                     gmx_incons("Invalid icoul in free energy kernel");
 586                                     break;
 587                             }
 588
 589                             if (fr->coulomb_modifier == eintmodPOTSWITCH)
 590                             {
 591                                 d                = rC-fr->rcoulomb_switch;
 592                                 d                = (d > 0.0) ? d : 0.0;
 593                                 d2               = d*d;
 594                                 sw               = 1.0+d2*d*(elec_swV3+d*(elec_swV4+d*elec_swV5));
 595                                 dsw              = d2*(elec_swF2+d*(elec_swF3+d*elec_swF4));
 596
 597                                 FscalC[i]        = FscalC[i]*sw - rC*Vcoul[i]*dsw;
 598                                 Vcoul[i]        *= sw;
 599
 600                                 FscalC[i]        = (rC < rcoulomb) ? FscalC[i] : 0.0;
 601                                 Vcoul[i]         = (rC < rcoulomb) ? Vcoul[i] : 0.0;
 602                             }
 603                         }
 604
 605                         /* Only process the VDW interactions if we have
 606                          * some non-zero parameters, and if we either
 607                          * include all entries in the list (no cutoff used
 608                          * in the kernel), or if we are within the cutoff.
 609                          */
 610                         bComputeVdwInteraction = !bExactVdwCutoff ||
 611                             ( bConvertLJEwaldToLJ6 && r < rvdw) ||
 612                             (!bConvertLJEwaldToLJ6 && rV < rvdw);
 613                         if ((c6[i] != 0 || c12[i] != 0) && bComputeVdwInteraction)
 614                         {
 615                             switch (ivdw)
 616                             {
 617                                 case GMX_NBKERNEL_VDW_LENNARDJONES:
 618                                     /* cutoff LJ */
 619                                     if (sc_r_power == 6.0)
 620                                     {
 621                                         rinv6            = rpinvV;
 622                                     }
 623                                     else
 624                                     {
 625                                         rinv6            = rinvV*rinvV;
 626                                         rinv6            = rinv6*rinv6*rinv6;
 627                                     }
 628                                     Vvdw6            = c6[i]*rinv6;
 629                                     Vvdw12           = c12[i]*rinv6*rinv6;
 630
 631                                     Vvdw[i]          = ( (Vvdw12 - c12[i]*sh_invrc6*sh_invrc6)*(1.0/12.0)
 632                                                          - (Vvdw6 - c6[i]*sh_invrc6)*(1.0/6.0));
 633                                     FscalV[i]        = Vvdw12 - Vvdw6;
 634                                     break;
 635
 636                                 case GMX_NBKERNEL_VDW_BUCKINGHAM:
 637                                     gmx_fatal(FARGS, "Buckingham free energy not supported.");
 638                                     break;
 639
 640                                 case GMX_NBKERNEL_VDW_CUBICSPLINETABLE:
 641                                     /* Table LJ */
 642                                     nnn = n1V+4;
 643                                     /* dispersion */
 644                                     Y          = VFtab[nnn];
 645                                     F          = VFtab[nnn+1];
 646                                     Geps       = epsV*VFtab[nnn+2];
 647                                     Heps2      = eps2V*VFtab[nnn+3];
 648                                     Fp         = F+Geps+Heps2;
 649                                     VV         = Y+epsV*Fp;
 650                                     FF         = Fp+Geps+2.0*Heps2;
 651                                     Vvdw[i]   += c6[i]*VV;
 652                                     FscalV[i] -= c6[i]*tabscale*FF*rV;
 653
 654                                     /* repulsion */
 655                                     Y          = VFtab[nnn+4];
 656                                     F          = VFtab[nnn+5];
 657                                     Geps       = epsV*VFtab[nnn+6];
 658                                     Heps2      = eps2V*VFtab[nnn+7];
 659                                     Fp         = F+Geps+Heps2;
 660                                     VV         = Y+epsV*Fp;
 661                                     FF         = Fp+Geps+2.0*Heps2;
 662                                     Vvdw[i]   += c12[i]*VV;
 663                                     FscalV[i] -= c12[i]*tabscale*FF*rV;
 664                                     break;
 665
 666                                 case GMX_NBKERNEL_VDW_LJEWALD:
 667                                     if (sc_r_power == 6.0)
 668                                     {
 669                                         rinv6            = rpinvV;
 670                                     }
 671                                     else
 672                                     {
 673                                         rinv6            = rinvV*rinvV;
 674                                         rinv6            = rinv6*rinv6*rinv6;
 675                                     }
 676                                     c6grid           = nbfp_grid[tj[i]];
 677
 678                                     if (bConvertLJEwaldToLJ6)
 679                                     {
 680                                         /* cutoff LJ */
 681                                         Vvdw6            = c6[i]*rinv6;
 682                                         Vvdw12           = c12[i]*rinv6*rinv6;
 683
 684                                         Vvdw[i]          = ( (Vvdw12 - c12[i]*sh_invrc6*sh_invrc6)*(1.0/12.0)
 685                                                              - (Vvdw6 - c6[i]*sh_invrc6 - c6grid*sh_lj_ewald)*(1.0/6.0));
 686                                         FscalV[i]        = Vvdw12 - Vvdw6;
 687                                     }
 688                                     else
 689                                     {
 690                                         /* Normal LJ-PME */
 691                                         ewcljrsq         = ewclj2*rV*rV;
 692                                         exponent         = exp(-ewcljrsq);
 693                                         poly             = exponent*(1.0 + ewcljrsq + ewcljrsq*ewcljrsq*0.5);
 694                                         vvdw_disp        = (c6[i]-c6grid*(1.0-poly))*rinv6;
 695                                         vvdw_rep         = c12[i]*rinv6*rinv6;
 696                                         FscalV[i]        = vvdw_rep - vvdw_disp - c6grid*(1.0/6.0)*exponent*ewclj6;
 697                                         Vvdw[i]          = (vvdw_rep - c12[i]*sh_invrc6*sh_invrc6)/12.0 - (vvdw_disp - c6[i]*sh_invrc6 - c6grid*sh_lj_ewald)/6.0;
 698                                     }
 699                                     break;
 700
 701                                 case GMX_NBKERNEL_VDW_NONE:
 702                                     Vvdw[i]    = 0.0;
 703                                     FscalV[i]  = 0.0;
 704                                     break;
 705
 706                                 default:
 707                                     gmx_incons("Invalid ivdw in free energy kernel");
 708                                     break;
 709                             }
 710
 711                             if (fr->vdw_modifier == eintmodPOTSWITCH)
 712                             {
 713                                 d                = rV-fr->rvdw_switch;
 714                                 d                = (d > 0.0) ? d : 0.0;
 715                                 d2               = d*d;
 716                                 sw               = 1.0+d2*d*(vdw_swV3+d*(vdw_swV4+d*vdw_swV5));
 717                                 dsw              = d2*(vdw_swF2+d*(vdw_swF3+d*vdw_swF4));
 718
 719                                 FscalV[i]        = FscalV[i]*sw - rV*Vvdw[i]*dsw;
 720                                 Vvdw[i]         *= sw;
 721
 722                                 FscalV[i]  = (rV < rvdw) ? FscalV[i] : 0.0;
 723                                 Vvdw[i]    = (rV < rvdw) ? Vvdw[i] : 0.0;
 724                             }
 725                         }
 726
 727                         /* FscalC (and FscalV) now contain: dV/drC * rC
 728                          * Now we multiply by rC^-p, so it will be: dV/drC * rC^1-p
 729                          * Further down we first multiply by r^p-2 and then by
 730                          * the vector r, which in total gives: dV/drC * (r/rC)^1-p
 731                          */
 732                         FscalC[i] *= rpinvC;
 733                         FscalV[i] *= rpinvV;
 734                     }
 735                 }
 736
 737                 /* Assemble A and B states */
 738                 for (i = 0; i < NSTATES; i++)
 739                 {
 740                     vctot         += LFC[i]*Vcoul[i];
 741                     vvtot         += LFV[i]*Vvdw[i];
 742
 743                     Fscal         += LFC[i]*FscalC[i]*rpm2;
 744                     Fscal         += LFV[i]*FscalV[i]*rpm2;
 745
 746                     dvdl_coul     += Vcoul[i]*DLF[i] + LFC[i]*alpha_coul_eff*dlfac_coul[i]*FscalC[i]*sigma_pow[i];
 747                     dvdl_vdw      += Vvdw[i]*DLF[i] + LFV[i]*alpha_vdw_eff*dlfac_vdw[i]*FscalV[i]*sigma_pow[i];
 748                 }
 749             }
 750             else if (icoul == GMX_NBKERNEL_ELEC_REACTIONFIELD)
 751             {
 752                 /* For excluded pairs, which are only in this pair list when
 753                  * using the Verlet scheme, we don't use soft-core.
 754                  * The group scheme also doesn't soft-core for these.
 755                  * As there is no singularity, there is no need for soft-core.
 756                  */
 757                 VV = krf*rsq - crf;
 758                 FF = -2.0*krf;
 759
 760                 if (ii == jnr)
 761                 {
 762                     VV *= 0.5;
 763                 }
 764
 765                 for (i = 0; i < NSTATES; i++)
 766                 {
 767                     vctot      += LFC[i]*qq[i]*VV;
 768                     Fscal      += LFC[i]*qq[i]*FF;
 769                     dvdl_coul  += DLF[i]*qq[i]*VV;
 770                 }
 771             }
 772
 773             if (bConvertEwaldToCoulomb && ( !bExactElecCutoff || r < rcoulomb ) )
 774             {
 775                 /* See comment in the preamble. When using Ewald interactions
 776                  * (unless we use a switch modifier) we subtract the reciprocal-space
 777                  * Ewald component here which made it possible to apply the free
 778                  * energy interaction to 1/r (vanilla coulomb short-range part)
 779                  * above. This gets us closer to the ideal case of applying
 780                  * the softcore to the entire electrostatic interaction,
 781                  * including the reciprocal-space component.
 782                  */
 783                 real v_lr, f_lr;
 784
 785                 ewrt      = r*ewtabscale;
 786                 ewitab    = (int) ewrt;
 787                 eweps     = ewrt-ewitab;
 788                 ewitab    = 4*ewitab;
 789                 f_lr      = ewtab[ewitab]+eweps*ewtab[ewitab+1];
 790                 v_lr      = (ewtab[ewitab+2]-ewtabhalfspace*eweps*(ewtab[ewitab]+f_lr));
 791                 f_lr     *= rinv;
 792
 793                 /* Note that any possible Ewald shift has already been applied in
 794                  * the normal interaction part above.
 795                  */
 796
 797                 if (ii == jnr)
 798                 {
 799                     /* If we get here, the i particle (ii) has itself (jnr)
 800                      * in its neighborlist. This can only happen with the Verlet
 801                      * scheme, and corresponds to a self-interaction that will
 802                      * occur twice. Scale it down by 50% to only include it once.
 803                      */
 804                     v_lr *= 0.5;
 805                 }
 806
 807                 for (i = 0; i < NSTATES; i++)
 808                 {
 809                     vctot      -= LFC[i]*qq[i]*v_lr;
 810                     Fscal      -= LFC[i]*qq[i]*f_lr;
 811                     dvdl_coul  -= (DLF[i]*qq[i])*v_lr;
 812                 }
 813             }
 814
 815             if (bConvertLJEwaldToLJ6 && (!bExactVdwCutoff || r < rvdw))
 816             {
 817                 /* See comment in the preamble. When using LJ-Ewald interactions
 818                  * (unless we use a switch modifier) we subtract the reciprocal-space
 819                  * Ewald component here which made it possible to apply the free
 820                  * energy interaction to r^-6 (vanilla LJ6 short-range part)
 821                  * above. This gets us closer to the ideal case of applying
 822                  * the softcore to the entire VdW interaction,
 823                  * including the reciprocal-space component.
 824                  */
 825                 /* We could also use the analytical form here
 826                  * iso a table, but that can cause issues for
 827                  * r close to 0 for non-interacting pairs.
 828                  */
 829                 real rs, frac, f_lr;
 830                 int  ri;
 831
 832                 rs     = rsq*rinv*ewtabscale;
 833                 ri     = (int)rs;
 834                 frac   = rs - ri;
 835                 f_lr   = (1 - frac)*tab_ewald_F_lj[ri] + frac*tab_ewald_F_lj[ri+1];
 836                 /* TODO: Currently the Ewald LJ table does not contain
 837                  * the factor 1/6, we should add this.
 838                  */
 839                 FF     = f_lr*rinv/6.0;
 840                 VV     = (tab_ewald_V_lj[ri] - ewtabhalfspace*frac*(tab_ewald_F_lj[ri] + f_lr))/6.0;
 841
 842                 if (ii == jnr)
 843                 {
 844                     /* If we get here, the i particle (ii) has itself (jnr)
 845                      * in its neighborlist. This can only happen with the Verlet
 846                      * scheme, and corresponds to a self-interaction that will
 847                      * occur twice. Scale it down by 50% to only include it once.
 848                      */
 849                     VV *= 0.5;
 850                 }
 851
 852                 for (i = 0; i < NSTATES; i++)
 853                 {
 854                     c6grid      = nbfp_grid[tj[i]];
 855                     vvtot      += LFV[i]*c6grid*VV;
 856                     Fscal      += LFV[i]*c6grid*FF;
 857                     dvdl_vdw   += (DLF[i]*c6grid)*VV;
 858                 }
 859             }
 860
 861             if (bDoForces)
 862             {
 863                 tx         = Fscal*dx;
 864                 ty         = Fscal*dy;
 865                 tz         = Fscal*dz;
 866                 fix        = fix + tx;
 867                 fiy        = fiy + ty;
 868                 fiz        = fiz + tz;
 869                 /* OpenMP atomics are expensive, but this kernels is also
 870                  * expensive, so we can take this hit, instead of using
 871                  * thread-local output buffers and extra reduction.
 872                  */
 873 #pragma omp atomic
 874                 f[j3]     -= tx;
 875 #pragma omp atomic
 876                 f[j3+1]   -= ty;
 877 #pragma omp atomic
 878                 f[j3+2]   -= tz;
 879             }
 880         }
 881
 882         /* The atomics below are expensive with many OpenMP threads.
 883          * Here unperturbed i-particles will usually only have a few
 884          * (perturbed) j-particles in the list. Thus with a buffered list
 885          * we can skip a significant number of i-reductions with a check.
 886          */
 887         if (npair_within_cutoff > 0)
 888         {
 889             if (bDoForces)
 890             {
 891 #pragma omp atomic
 892                 f[ii3]        += fix;
 893 #pragma omp atomic
 894                 f[ii3+1]      += fiy;
 895 #pragma omp atomic
 896                 f[ii3+2]      += fiz;
 897             }
 898             if (bDoShiftForces)
 899             {
 900 #pragma omp atomic
 901                 fshift[is3]   += fix;
 902 #pragma omp atomic
 903                 fshift[is3+1] += fiy;
 904 #pragma omp atomic
 905                 fshift[is3+2] += fiz;
 906             }
 907             if (bDoPotential)
 908             {
 909                 ggid               = gid[n];
 910 #pragma omp atomic
 911                 Vc[ggid]          += vctot;
 912 #pragma omp atomic
 913                 Vv[ggid]          += vvtot;
 914             }
 915         }
 916     }
 917
 918 #pragma omp atomic
 919     dvdl[efptCOUL]     += dvdl_coul;
 920  #pragma omp atomic
 921     dvdl[efptVDW]      += dvdl_vdw;
 922
 923     /* Estimate flops, average for free energy stuff:
 924      * 12  flops per outer iteration
 925      * 150 flops per inner iteration
 926      */
 927 #pragma omp atomic
 928     inc_nrnb(nrnb, eNR_NBKERNEL_FREE_ENERGY, nlist->nri*12 + nlist->jindex[n]*150);
 929 }
 930
 931 real
 932 nb_free_energy_evaluate_single(real r2, real sc_r_power, real alpha_coul, real alpha_vdw,
 933                                real tabscale, real *vftab,
 934                                real qqA, real c6A, real c12A, real qqB, real c6B, real c12B,
 935                                real LFC[2], real LFV[2], real DLF[2],
 936                                real lfac_coul[2], real lfac_vdw[2], real dlfac_coul[2], real dlfac_vdw[2],
 937                                real sigma6_def, real sigma6_min, real sigma2_def, real sigma2_min,
 938                                real *velectot, real *vvdwtot, real *dvdl)
 939 {
 940     real       r, rp, rpm2, rtab, eps, eps2, Y, F, Geps, Heps2, Fp, VV, FF, fscal;
 941     real       qq[2], c6[2], c12[2], sigma6[2], sigma2[2], sigma_pow[2], sigma_powm2[2];
 942     real       alpha_coul_eff, alpha_vdw_eff, dvdl_coul, dvdl_vdw;
 943     real       rpinv, r_coul, r_vdw, velecsum, vvdwsum;
 944     real       fscal_vdw[2], fscal_elec[2];
 945     real       velec[2], vvdw[2];
 946     int        i, ntab;
 947
 948     qq[0]    = qqA;
 949     qq[1]    = qqB;
 950     c6[0]    = c6A;
 951     c6[1]    = c6B;
 952     c12[0]   = c12A;
 953     c12[1]   = c12B;
 954
 955     if (sc_r_power == 6.0)
 956     {
 957         rpm2             = r2*r2;   /* r4 */
 958         rp               = rpm2*r2; /* r6 */
 959     }
 960     else if (sc_r_power == 48.0)
 961     {
 962         rp               = r2*r2*r2; /* r6 */
 963         rp               = rp*rp;    /* r12 */
 964         rp               = rp*rp;    /* r24 */
 965         rp               = rp*rp;    /* r48 */
 966         rpm2             = rp/r2;    /* r46 */
 967     }
 968     else
 969     {
 970         rp             = pow(r2, 0.5*sc_r_power);  /* not currently supported as input, but can handle it */
 971         rpm2           = rp/r2;
 972     }
 973
 974     /* Loop over state A(0) and B(1) */
 975     for (i = 0; i < 2; i++)
 976     {
 977         if ((c6[i] > 0) && (c12[i] > 0))
 978         {
 979             /* The c6 & c12 coefficients now contain the constants 6.0 and 12.0, respectively.
 980              * Correct for this by multiplying with (1/12.0)/(1/6.0)=6.0/12.0=0.5.
 981              */
 982             sigma6[i]       = 0.5*c12[i]/c6[i];
 983             sigma2[i]       = pow(0.5*c12[i]/c6[i], 1.0/3.0);
 984             /* should be able to get rid of this ^^^ internal pow call eventually.  Will require agreement on
 985                what data to store externally.  Can't be fixed without larger scale changes, so not 5.0 */
 986             if (sigma6[i] < sigma6_min)   /* for disappearing coul and vdw with soft core at the same time */
 987             {
 988                 sigma6[i] = sigma6_min;
 989                 sigma2[i] = sigma2_min;
 990             }
 991         }
 992         else
 993         {
 994             sigma6[i]       = sigma6_def;
 995             sigma2[i]       = sigma2_def;
 996         }
 997         if (sc_r_power == 6.0)
 998         {
 999             sigma_pow[i]    = sigma6[i];
1000             sigma_powm2[i]  = sigma6[i]/sigma2[i];
1001         }
1002         else if (sc_r_power == 48.0)
1003         {
1004             sigma_pow[i]    = sigma6[i]*sigma6[i];       /* sigma^12 */
1005             sigma_pow[i]    = sigma_pow[i]*sigma_pow[i]; /* sigma^24 */
1006             sigma_pow[i]    = sigma_pow[i]*sigma_pow[i]; /* sigma^48 */
1007             sigma_powm2[i]  = sigma_pow[i]/sigma2[i];
1008         }
1009         else
1010         {    /* not really supported as input, but in here for testing the general case*/
1011             sigma_pow[i]    = pow(sigma2[i], sc_r_power/2);
1012             sigma_powm2[i]  = sigma_pow[i]/(sigma2[i]);
1013         }
1014     }
1015
1016     /* only use softcore if one of the states has a zero endstate - softcore is for avoiding infinities!*/
1017     if ((c12[0] > 0) && (c12[1] > 0))
1018     {
1019         alpha_vdw_eff    = 0;
1020         alpha_coul_eff   = 0;
1021     }
1022     else
1023     {
1024         alpha_vdw_eff    = alpha_vdw;
1025         alpha_coul_eff   = alpha_coul;
1026     }
1027
1028     /* Loop over A and B states again */
1029     for (i = 0; i < 2; i++)
1030     {
1031         fscal_elec[i] = 0;
1032         fscal_vdw[i]  = 0;
1033         velec[i]      = 0;
1034         vvdw[i]       = 0;
1035
1036         /* Only spend time on A or B state if it is non-zero */
1037         if ( (qq[i] != 0) || (c6[i] != 0) || (c12[i] != 0) )
1038         {
1039             /* Coulomb */
1040             rpinv            = 1.0/(alpha_coul_eff*lfac_coul[i]*sigma_pow[i]+rp);
1041             r_coul           = pow(rpinv, -1.0/sc_r_power);
1042
1043             /* Electrostatics table lookup data */
1044             rtab             = r_coul*tabscale;
1045             ntab             = rtab;
1046             eps              = rtab-ntab;
1047             eps2             = eps*eps;
1048             ntab             = 12*ntab;
1049             /* Electrostatics */
1050             Y                = vftab[ntab];
1051             F                = vftab[ntab+1];
1052             Geps             = eps*vftab[ntab+2];
1053             Heps2            = eps2*vftab[ntab+3];
1054             Fp               = F+Geps+Heps2;
1055             VV               = Y+eps*Fp;
1056             FF               = Fp+Geps+2.0*Heps2;
1057             velec[i]         = qq[i]*VV;
1058             fscal_elec[i]    = -qq[i]*FF*r_coul*rpinv*tabscale;
1059
1060             /* Vdw */
1061             rpinv            = 1.0/(alpha_vdw_eff*lfac_vdw[i]*sigma_pow[i]+rp);
1062             r_vdw            = pow(rpinv, -1.0/sc_r_power);
1063             /* Vdw table lookup data */
1064             rtab             = r_vdw*tabscale;
1065             ntab             = rtab;
1066             eps              = rtab-ntab;
1067             eps2             = eps*eps;
1068             ntab             = 12*ntab;
1069             /* Dispersion */
1070             Y                = vftab[ntab+4];
1071             F                = vftab[ntab+5];
1072             Geps             = eps*vftab[ntab+6];
1073             Heps2            = eps2*vftab[ntab+7];
1074             Fp               = F+Geps+Heps2;
1075             VV               = Y+eps*Fp;
1076             FF               = Fp+Geps+2.0*Heps2;
1077             vvdw[i]          = c6[i]*VV;
1078             fscal_vdw[i]     = -c6[i]*FF;
1079
1080             /* Repulsion */
1081             Y                = vftab[ntab+8];
1082             F                = vftab[ntab+9];
1083             Geps             = eps*vftab[ntab+10];
1084             Heps2            = eps2*vftab[ntab+11];
1085             Fp               = F+Geps+Heps2;
1086             VV               = Y+eps*Fp;
1087             FF               = Fp+Geps+2.0*Heps2;
1088             vvdw[i]         += c12[i]*VV;
1089             fscal_vdw[i]    -= c12[i]*FF;
1090             fscal_vdw[i]    *= r_vdw*rpinv*tabscale;
1091         }
1092     }
1093     /* Now we have velec[i], vvdw[i], and fscal[i] for both states */
1094     /* Assemble A and B states */
1095     velecsum  = 0;
1096     vvdwsum   = 0;
1097     dvdl_coul = 0;
1098     dvdl_vdw  = 0;
1099     fscal     = 0;
1100     for (i = 0; i < 2; i++)
1101     {
1102         velecsum      += LFC[i]*velec[i];
1103         vvdwsum       += LFV[i]*vvdw[i];
1104
1105         fscal         += (LFC[i]*fscal_elec[i]+LFV[i]*fscal_vdw[i])*rpm2;
1106
1107         dvdl_coul     += velec[i]*DLF[i] + LFC[i]*alpha_coul_eff*dlfac_coul[i]*fscal_elec[i]*sigma_pow[i];
1108         dvdl_vdw      += vvdw[i]*DLF[i] + LFV[i]*alpha_vdw_eff*dlfac_vdw[i]*fscal_vdw[i]*sigma_pow[i];
1109     }
1110
1111     dvdl[efptCOUL]     += dvdl_coul;
1112     dvdl[efptVDW]      += dvdl_vdw;
1113
1114     *velectot           = velecsum;
1115     *vvdwtot            = vvdwsum;
1116
1117     return fscal;
1118 }