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