ViewVC Help
View File | Revision Log | Show Annotations | View Changeset | Root Listing
root/group/trunk/OOPSE-4/src/UseTheForce/doForces.F90
(Generate patch)

Comparing trunk/OOPSE-4/src/UseTheForce/doForces.F90 (file contents):
Revision 1930 by gezelter, Wed Jan 12 22:41:40 2005 UTC vs.
Revision 3126 by gezelter, Fri Apr 6 21:53:43 2007 UTC

# Line 45 | Line 45
45  
46   !! @author Charles F. Vardeman II
47   !! @author Matthew Meineke
48 < !! @version $Id: doForces.F90,v 1.9 2005-01-12 22:40:37 gezelter Exp $, $Date: 2005-01-12 22:40:37 $, $Name: not supported by cvs2svn $, $Revision: 1.9 $
48 > !! @version $Id: doForces.F90,v 1.85 2007-04-06 21:53:41 gezelter Exp $, $Date: 2007-04-06 21:53:41 $, $Name: not supported by cvs2svn $, $Revision: 1.85 $
49  
50  
51   module doForces
# Line 57 | Line 57 | module doForces
57    use neighborLists  
58    use lj
59    use sticky
60 <  use dipole_dipole
61 <  use charge_charge
62 <  use reaction_field
63 <  use gb_pair
60 >  use electrostatic_module
61 >  use gayberne
62    use shapes
63    use vector_class
64    use eam
65 +  use suttonchen
66    use status
67   #ifdef IS_MPI
68    use mpiSimulation
# Line 73 | Line 72 | module doForces
72    PRIVATE
73  
74   #define __FORTRAN90
75 < #include "UseTheForce/fSwitchingFunction.h"
75 > #include "UseTheForce/fCutoffPolicy.h"
76 > #include "UseTheForce/DarkSide/fInteractionMap.h"
77 > #include "UseTheForce/DarkSide/fElectrostaticSummationMethod.h"
78  
79    INTEGER, PARAMETER:: PREPAIR_LOOP = 1
80    INTEGER, PARAMETER:: PAIR_LOOP    = 2
81  
81  logical, save :: haveRlist = .false.
82    logical, save :: haveNeighborList = .false.
83    logical, save :: haveSIMvariables = .false.
84  logical, save :: havePropertyMap = .false.
84    logical, save :: haveSaneForceField = .false.
85 <  
85 >  logical, save :: haveInteractionHash = .false.
86 >  logical, save :: haveGtypeCutoffMap = .false.
87 >  logical, save :: haveDefaultCutoffs = .false.
88 >  logical, save :: haveSkinThickness = .false.
89 >  logical, save :: haveElectrostaticSummationMethod = .false.
90 >  logical, save :: haveCutoffPolicy = .false.
91 >  logical, save :: VisitCutoffsAfterComputing = .false.
92 >  logical, save :: do_box_dipole = .false.
93 >
94    logical, save :: FF_uses_DirectionalAtoms
95 <  logical, save :: FF_uses_LennardJones
89 <  logical, save :: FF_uses_Electrostatic
90 <  logical, save :: FF_uses_charges
91 <  logical, save :: FF_uses_dipoles
92 <  logical, save :: FF_uses_sticky
95 >  logical, save :: FF_uses_Dipoles
96    logical, save :: FF_uses_GayBerne
97    logical, save :: FF_uses_EAM
98 <  logical, save :: FF_uses_Shapes
99 <  logical, save :: FF_uses_FLARB
100 <  logical, save :: FF_uses_RF
98 >  logical, save :: FF_uses_SC
99 >  logical, save :: FF_uses_MEAM
100 >
101  
102    logical, save :: SIM_uses_DirectionalAtoms
100  logical, save :: SIM_uses_LennardJones
101  logical, save :: SIM_uses_Electrostatics
102  logical, save :: SIM_uses_Charges
103  logical, save :: SIM_uses_Dipoles
104  logical, save :: SIM_uses_Sticky
105  logical, save :: SIM_uses_GayBerne
103    logical, save :: SIM_uses_EAM
104 <  logical, save :: SIM_uses_Shapes
105 <  logical, save :: SIM_uses_FLARB
109 <  logical, save :: SIM_uses_RF
104 >  logical, save :: SIM_uses_SC
105 >  logical, save :: SIM_uses_MEAM
106    logical, save :: SIM_requires_postpair_calc
107    logical, save :: SIM_requires_prepair_calc
108    logical, save :: SIM_uses_PBC
109 <  logical, save :: SIM_uses_molecular_cutoffs
109 >  logical, save :: SIM_uses_AtomicVirial
110  
111 <  real(kind=dp), save :: rlist, rlistsq
111 >  integer, save :: electrostaticSummationMethod
112 >  integer, save :: cutoffPolicy = TRADITIONAL_CUTOFF_POLICY
113  
114 +  real(kind=dp), save :: defaultRcut, defaultRsw, largestRcut
115 +  real(kind=dp), save :: skinThickness
116 +  logical, save :: defaultDoShift
117 +
118    public :: init_FF
119 +  public :: setCutoffs
120 +  public :: cWasLame
121 +  public :: setElectrostaticMethod
122 +  public :: setBoxDipole
123 +  public :: getBoxDipole
124 +  public :: setCutoffPolicy
125 +  public :: setSkinThickness
126    public :: do_force_loop
119  public :: setRlistDF
127  
128   #ifdef PROFILE
129    public :: getforcetime
# Line 124 | Line 131 | module doForces
131    real :: forceTimeInitial, forceTimeFinal
132    integer :: nLoops
133   #endif
134 +  
135 +  !! Variables for cutoff mapping and interaction mapping
136 +  ! Bit hash to determine pair-pair interactions.
137 +  integer, dimension(:,:), allocatable :: InteractionHash
138 +  real(kind=dp), dimension(:), allocatable :: atypeMaxCutoff
139 +  real(kind=dp), dimension(:), allocatable, target :: groupMaxCutoffRow
140 +  real(kind=dp), dimension(:), pointer :: groupMaxCutoffCol
141  
142 <  type :: Properties
143 <     logical :: is_Directional   = .false.
130 <     logical :: is_LennardJones  = .false.
131 <     logical :: is_Electrostatic = .false.
132 <     logical :: is_Charge        = .false.
133 <     logical :: is_Dipole        = .false.
134 <     logical :: is_Sticky        = .false.
135 <     logical :: is_GayBerne      = .false.
136 <     logical :: is_EAM           = .false.
137 <     logical :: is_Shape         = .false.
138 <     logical :: is_FLARB         = .false.
139 <  end type Properties
142 >  integer, dimension(:), allocatable, target :: groupToGtypeRow
143 >  integer, dimension(:), pointer :: groupToGtypeCol => null()
144  
145 <  type(Properties), dimension(:),allocatable :: PropertyMap
145 >  real(kind=dp), dimension(:), allocatable,target :: gtypeMaxCutoffRow
146 >  real(kind=dp), dimension(:), pointer :: gtypeMaxCutoffCol
147 >  type ::gtypeCutoffs
148 >     real(kind=dp) :: rcut
149 >     real(kind=dp) :: rcutsq
150 >     real(kind=dp) :: rlistsq
151 >  end type gtypeCutoffs
152 >  type(gtypeCutoffs), dimension(:,:), allocatable :: gtypeCutoffMap
153  
154 +  real(kind=dp), dimension(3) :: boxDipole
155 +
156   contains
157  
158 <  subroutine setRlistDF( this_rlist )
146 <    
147 <    real(kind=dp) :: this_rlist
148 <
149 <    rlist = this_rlist
150 <    rlistsq = rlist * rlist
151 <    
152 <    haveRlist = .true.
153 <
154 <  end subroutine setRlistDF    
155 <
156 <  subroutine createPropertyMap(status)
158 >  subroutine createInteractionHash()
159      integer :: nAtypes
158    integer :: status
160      integer :: i
161 <    logical :: thisProperty
162 <    real (kind=DP) :: thisDPproperty
161 >    integer :: j
162 >    integer :: iHash
163 >    !! Test Types
164 >    logical :: i_is_LJ
165 >    logical :: i_is_Elect
166 >    logical :: i_is_Sticky
167 >    logical :: i_is_StickyP
168 >    logical :: i_is_GB
169 >    logical :: i_is_EAM
170 >    logical :: i_is_Shape
171 >    logical :: i_is_SC
172 >    logical :: i_is_MEAM
173 >    logical :: j_is_LJ
174 >    logical :: j_is_Elect
175 >    logical :: j_is_Sticky
176 >    logical :: j_is_StickyP
177 >    logical :: j_is_GB
178 >    logical :: j_is_EAM
179 >    logical :: j_is_Shape
180 >    logical :: j_is_SC
181 >    logical :: j_is_MEAM
182 >    real(kind=dp) :: myRcut
183  
184 <    status = 0
185 <
184 >    if (.not. associated(atypes)) then
185 >       call handleError("doForces", "atypes was not present before call of createInteractionHash!")
186 >       return
187 >    endif
188 >    
189      nAtypes = getSize(atypes)
190 <
190 >    
191      if (nAtypes == 0) then
192 <       status = -1
192 >       call handleError("doForces", "nAtypes was zero during call of createInteractionHash!")
193         return
194      end if
195 <        
196 <    if (.not. allocated(PropertyMap)) then
197 <       allocate(PropertyMap(nAtypes))
195 >
196 >    if (.not. allocated(InteractionHash)) then
197 >       allocate(InteractionHash(nAtypes,nAtypes))
198 >    else
199 >       deallocate(InteractionHash)
200 >       allocate(InteractionHash(nAtypes,nAtypes))
201      endif
202  
203 +    if (.not. allocated(atypeMaxCutoff)) then
204 +       allocate(atypeMaxCutoff(nAtypes))
205 +    else
206 +       deallocate(atypeMaxCutoff)
207 +       allocate(atypeMaxCutoff(nAtypes))
208 +    endif
209 +        
210      do i = 1, nAtypes
211 <       call getElementProperty(atypes, i, "is_Directional", thisProperty)
212 <       PropertyMap(i)%is_Directional = thisProperty
211 >       call getElementProperty(atypes, i, "is_LennardJones", i_is_LJ)
212 >       call getElementProperty(atypes, i, "is_Electrostatic", i_is_Elect)
213 >       call getElementProperty(atypes, i, "is_Sticky", i_is_Sticky)
214 >       call getElementProperty(atypes, i, "is_StickyPower", i_is_StickyP)
215 >       call getElementProperty(atypes, i, "is_GayBerne", i_is_GB)
216 >       call getElementProperty(atypes, i, "is_EAM", i_is_EAM)
217 >       call getElementProperty(atypes, i, "is_Shape", i_is_Shape)
218 >       call getElementProperty(atypes, i, "is_SC", i_is_SC)
219 >       call getElementProperty(atypes, i, "is_MEAM", i_is_MEAM)
220  
221 <       call getElementProperty(atypes, i, "is_LennardJones", thisProperty)
181 <       PropertyMap(i)%is_LennardJones = thisProperty
182 <      
183 <       call getElementProperty(atypes, i, "is_Electrostatic", thisProperty)
184 <       PropertyMap(i)%is_Electrostatic = thisProperty
221 >       do j = i, nAtypes
222  
223 <       call getElementProperty(atypes, i, "is_Charge", thisProperty)
224 <       PropertyMap(i)%is_Charge = thisProperty
188 <      
189 <       call getElementProperty(atypes, i, "is_Dipole", thisProperty)
190 <       PropertyMap(i)%is_Dipole = thisProperty
223 >          iHash = 0
224 >          myRcut = 0.0_dp
225  
226 <       call getElementProperty(atypes, i, "is_Sticky", thisProperty)
227 <       PropertyMap(i)%is_Sticky = thisProperty
226 >          call getElementProperty(atypes, j, "is_LennardJones", j_is_LJ)
227 >          call getElementProperty(atypes, j, "is_Electrostatic", j_is_Elect)
228 >          call getElementProperty(atypes, j, "is_Sticky", j_is_Sticky)
229 >          call getElementProperty(atypes, j, "is_StickyPower", j_is_StickyP)
230 >          call getElementProperty(atypes, j, "is_GayBerne", j_is_GB)
231 >          call getElementProperty(atypes, j, "is_EAM", j_is_EAM)
232 >          call getElementProperty(atypes, j, "is_Shape", j_is_Shape)
233 >          call getElementProperty(atypes, j, "is_SC", j_is_SC)
234 >          call getElementProperty(atypes, j, "is_MEAM", j_is_MEAM)
235  
236 <       call getElementProperty(atypes, i, "is_GayBerne", thisProperty)
237 <       PropertyMap(i)%is_GayBerne = thisProperty
236 >          if (i_is_LJ .and. j_is_LJ) then
237 >             iHash = ior(iHash, LJ_PAIR)            
238 >          endif
239 >          
240 >          if (i_is_Elect .and. j_is_Elect) then
241 >             iHash = ior(iHash, ELECTROSTATIC_PAIR)
242 >          endif
243 >          
244 >          if (i_is_Sticky .and. j_is_Sticky) then
245 >             iHash = ior(iHash, STICKY_PAIR)
246 >          endif
247  
248 <       call getElementProperty(atypes, i, "is_EAM", thisProperty)
249 <       PropertyMap(i)%is_EAM = thisProperty
248 >          if (i_is_StickyP .and. j_is_StickyP) then
249 >             iHash = ior(iHash, STICKYPOWER_PAIR)
250 >          endif
251  
252 <       call getElementProperty(atypes, i, "is_Shape", thisProperty)
253 <       PropertyMap(i)%is_Shape = thisProperty
252 >          if (i_is_EAM .and. j_is_EAM) then
253 >             iHash = ior(iHash, EAM_PAIR)
254 >          endif
255  
256 <       call getElementProperty(atypes, i, "is_FLARB", thisProperty)
257 <       PropertyMap(i)%is_FLARB = thisProperty
258 <    end do
256 >          if (i_is_SC .and. j_is_SC) then
257 >             iHash = ior(iHash, SC_PAIR)
258 >          endif
259  
260 <    havePropertyMap = .true.
260 >          if (i_is_GB .and. j_is_GB) iHash = ior(iHash, GAYBERNE_PAIR)
261 >          if (i_is_GB .and. j_is_LJ) iHash = ior(iHash, GAYBERNE_LJ)
262 >          if (i_is_LJ .and. j_is_GB) iHash = ior(iHash, GAYBERNE_LJ)
263  
264 <  end subroutine createPropertyMap
264 >          if (i_is_Shape .and. j_is_Shape) iHash = ior(iHash, SHAPE_PAIR)
265 >          if (i_is_Shape .and. j_is_LJ) iHash = ior(iHash, SHAPE_LJ)
266 >          if (i_is_LJ .and. j_is_Shape) iHash = ior(iHash, SHAPE_LJ)
267  
212  subroutine setSimVariables()
213    SIM_uses_DirectionalAtoms = SimUsesDirectionalAtoms()
214    SIM_uses_LennardJones = SimUsesLennardJones()
215    SIM_uses_Electrostatics = SimUsesElectrostatics()
216    SIM_uses_Charges = SimUsesCharges()
217    SIM_uses_Dipoles = SimUsesDipoles()
218    SIM_uses_Sticky = SimUsesSticky()
219    SIM_uses_GayBerne = SimUsesGayBerne()
220    SIM_uses_EAM = SimUsesEAM()
221    SIM_uses_Shapes = SimUsesShapes()
222    SIM_uses_FLARB = SimUsesFLARB()
223    SIM_uses_RF = SimUsesRF()
224    SIM_requires_postpair_calc = SimRequiresPostpairCalc()
225    SIM_requires_prepair_calc = SimRequiresPrepairCalc()
226    SIM_uses_PBC = SimUsesPBC()
268  
269 <    haveSIMvariables = .true.
269 >          InteractionHash(i,j) = iHash
270 >          InteractionHash(j,i) = iHash
271  
272 <    return
231 <  end subroutine setSimVariables
272 >       end do
273  
274 <  subroutine doReadyCheck(error)
234 <    integer, intent(out) :: error
274 >    end do
275  
276 <    integer :: myStatus
276 >    haveInteractionHash = .true.
277 >  end subroutine createInteractionHash
278  
279 <    error = 0
239 <    
240 <    if (.not. havePropertyMap) then
279 >  subroutine createGtypeCutoffMap()
280  
281 <       myStatus = 0
282 <
283 <       call createPropertyMap(myStatus)
284 <
285 <       if (myStatus .ne. 0) then
286 <          write(default_error, *) 'createPropertyMap failed in doForces!'
287 <          error = -1
288 <          return
281 >    logical :: i_is_LJ
282 >    logical :: i_is_Elect
283 >    logical :: i_is_Sticky
284 >    logical :: i_is_StickyP
285 >    logical :: i_is_GB
286 >    logical :: i_is_EAM
287 >    logical :: i_is_Shape
288 >    logical :: i_is_SC
289 >    logical :: GtypeFound
290 >
291 >    integer :: myStatus, nAtypes,  i, j, istart, iend, jstart, jend
292 >    integer :: n_in_i, me_i, ia, g, atom1, ja, n_in_j,me_j
293 >    integer :: nGroupsInRow
294 >    integer :: nGroupsInCol
295 >    integer :: nGroupTypesRow,nGroupTypesCol
296 >    real(kind=dp):: thisSigma, bigSigma, thisRcut, tradRcut, tol
297 >    real(kind=dp) :: biggestAtypeCutoff
298 >
299 >    if (.not. haveInteractionHash) then
300 >       call createInteractionHash()      
301 >    endif
302 > #ifdef IS_MPI
303 >    nGroupsInRow = getNgroupsInRow(plan_group_row)
304 >    nGroupsInCol = getNgroupsInCol(plan_group_col)
305 > #endif
306 >    nAtypes = getSize(atypes)
307 > ! Set all of the initial cutoffs to zero.
308 >    atypeMaxCutoff = 0.0_dp
309 >    do i = 1, nAtypes
310 >       if (SimHasAtype(i)) then    
311 >          call getElementProperty(atypes, i, "is_LennardJones", i_is_LJ)
312 >          call getElementProperty(atypes, i, "is_Electrostatic", i_is_Elect)
313 >          call getElementProperty(atypes, i, "is_Sticky", i_is_Sticky)
314 >          call getElementProperty(atypes, i, "is_StickyPower", i_is_StickyP)
315 >          call getElementProperty(atypes, i, "is_GayBerne", i_is_GB)
316 >          call getElementProperty(atypes, i, "is_EAM", i_is_EAM)
317 >          call getElementProperty(atypes, i, "is_Shape", i_is_Shape)
318 >          call getElementProperty(atypes, i, "is_SC", i_is_SC)
319 >
320 >          if (haveDefaultCutoffs) then
321 >             atypeMaxCutoff(i) = defaultRcut
322 >          else
323 >             if (i_is_LJ) then          
324 >                thisRcut = getSigma(i) * 2.5_dp
325 >                if (thisRCut .gt. atypeMaxCutoff(i)) atypeMaxCutoff(i) = thisRCut
326 >             endif
327 >             if (i_is_Elect) then
328 >                thisRcut = defaultRcut
329 >                if (thisRCut .gt. atypeMaxCutoff(i)) atypeMaxCutoff(i) = thisRCut
330 >             endif
331 >             if (i_is_Sticky) then
332 >                thisRcut = getStickyCut(i)
333 >                if (thisRCut .gt. atypeMaxCutoff(i)) atypeMaxCutoff(i) = thisRCut
334 >             endif
335 >             if (i_is_StickyP) then
336 >                thisRcut = getStickyPowerCut(i)
337 >                if (thisRCut .gt. atypeMaxCutoff(i)) atypeMaxCutoff(i) = thisRCut
338 >             endif
339 >             if (i_is_GB) then
340 >                thisRcut = getGayBerneCut(i)
341 >                if (thisRCut .gt. atypeMaxCutoff(i)) atypeMaxCutoff(i) = thisRCut
342 >             endif
343 >             if (i_is_EAM) then
344 >                thisRcut = getEAMCut(i)
345 >                if (thisRCut .gt. atypeMaxCutoff(i)) atypeMaxCutoff(i) = thisRCut
346 >             endif
347 >             if (i_is_Shape) then
348 >                thisRcut = getShapeCut(i)
349 >                if (thisRCut .gt. atypeMaxCutoff(i)) atypeMaxCutoff(i) = thisRCut
350 >             endif
351 >             if (i_is_SC) then
352 >                thisRcut = getSCCut(i)
353 >                if (thisRCut .gt. atypeMaxCutoff(i)) atypeMaxCutoff(i) = thisRCut
354 >             endif
355 >          endif
356 >                    
357 >          if (atypeMaxCutoff(i).gt.biggestAtypeCutoff) then
358 >             biggestAtypeCutoff = atypeMaxCutoff(i)
359 >          endif
360 >
361         endif
362 +    enddo
363 +    
364 +    istart = 1
365 +    jstart = 1
366 + #ifdef IS_MPI
367 +    iend = nGroupsInRow
368 +    jend = nGroupsInCol
369 + #else
370 +    iend = nGroups
371 +    jend = nGroups
372 + #endif
373 +    
374 +    !! allocate the groupToGtype and gtypeMaxCutoff here.
375 +    if(.not.allocated(groupToGtypeRow)) then
376 +     !  allocate(groupToGtype(iend))
377 +       allocate(groupToGtypeRow(iend))
378 +    else
379 +       deallocate(groupToGtypeRow)
380 +       allocate(groupToGtypeRow(iend))
381 +    endif
382 +    if(.not.allocated(groupMaxCutoffRow)) then
383 +       allocate(groupMaxCutoffRow(iend))
384 +    else
385 +       deallocate(groupMaxCutoffRow)
386 +       allocate(groupMaxCutoffRow(iend))
387 +    end if
388 +
389 +    if(.not.allocated(gtypeMaxCutoffRow)) then
390 +       allocate(gtypeMaxCutoffRow(iend))
391 +    else
392 +       deallocate(gtypeMaxCutoffRow)
393 +       allocate(gtypeMaxCutoffRow(iend))
394 +    endif
395 +
396 +
397 + #ifdef IS_MPI
398 +       ! We only allocate new storage if we are in MPI because Ncol /= Nrow
399 +    if(.not.associated(groupToGtypeCol)) then
400 +       allocate(groupToGtypeCol(jend))
401 +    else
402 +       deallocate(groupToGtypeCol)
403 +       allocate(groupToGtypeCol(jend))
404 +    end if
405 +
406 +    if(.not.associated(groupMaxCutoffCol)) then
407 +       allocate(groupMaxCutoffCol(jend))
408 +    else
409 +       deallocate(groupMaxCutoffCol)
410 +       allocate(groupMaxCutoffCol(jend))
411 +    end if
412 +    if(.not.associated(gtypeMaxCutoffCol)) then
413 +       allocate(gtypeMaxCutoffCol(jend))
414 +    else
415 +       deallocate(gtypeMaxCutoffCol)      
416 +       allocate(gtypeMaxCutoffCol(jend))
417 +    end if
418 +
419 +       groupMaxCutoffCol = 0.0_dp
420 +       gtypeMaxCutoffCol = 0.0_dp
421 +
422 + #endif
423 +       groupMaxCutoffRow = 0.0_dp
424 +       gtypeMaxCutoffRow = 0.0_dp
425 +
426 +
427 +    !! first we do a single loop over the cutoff groups to find the
428 +    !! largest cutoff for any atypes present in this group.  We also
429 +    !! create gtypes at this point.
430 +    
431 +    tol = 1.0e-6_dp
432 +    nGroupTypesRow = 0
433 +    nGroupTypesCol = 0
434 +    do i = istart, iend      
435 +       n_in_i = groupStartRow(i+1) - groupStartRow(i)
436 +       groupMaxCutoffRow(i) = 0.0_dp
437 +       do ia = groupStartRow(i), groupStartRow(i+1)-1
438 +          atom1 = groupListRow(ia)
439 + #ifdef IS_MPI
440 +          me_i = atid_row(atom1)
441 + #else
442 +          me_i = atid(atom1)
443 + #endif          
444 +          if (atypeMaxCutoff(me_i).gt.groupMaxCutoffRow(i)) then
445 +             groupMaxCutoffRow(i)=atypeMaxCutoff(me_i)
446 +          endif          
447 +       enddo
448 +       if (nGroupTypesRow.eq.0) then
449 +          nGroupTypesRow = nGroupTypesRow + 1
450 +          gtypeMaxCutoffRow(nGroupTypesRow) = groupMaxCutoffRow(i)
451 +          groupToGtypeRow(i) = nGroupTypesRow
452 +       else
453 +          GtypeFound = .false.
454 +          do g = 1, nGroupTypesRow
455 +             if ( abs(groupMaxCutoffRow(i) - gtypeMaxCutoffRow(g)).lt.tol) then
456 +                groupToGtypeRow(i) = g
457 +                GtypeFound = .true.
458 +             endif
459 +          enddo
460 +          if (.not.GtypeFound) then            
461 +             nGroupTypesRow = nGroupTypesRow + 1
462 +             gtypeMaxCutoffRow(nGroupTypesRow) = groupMaxCutoffRow(i)
463 +             groupToGtypeRow(i) = nGroupTypesRow
464 +          endif
465 +       endif
466 +    enddo    
467 +
468 + #ifdef IS_MPI
469 +    do j = jstart, jend      
470 +       n_in_j = groupStartCol(j+1) - groupStartCol(j)
471 +       groupMaxCutoffCol(j) = 0.0_dp
472 +       do ja = groupStartCol(j), groupStartCol(j+1)-1
473 +          atom1 = groupListCol(ja)
474 +
475 +          me_j = atid_col(atom1)
476 +
477 +          if (atypeMaxCutoff(me_j).gt.groupMaxCutoffCol(j)) then
478 +             groupMaxCutoffCol(j)=atypeMaxCutoff(me_j)
479 +          endif          
480 +       enddo
481 +
482 +       if (nGroupTypesCol.eq.0) then
483 +          nGroupTypesCol = nGroupTypesCol + 1
484 +          gtypeMaxCutoffCol(nGroupTypesCol) = groupMaxCutoffCol(j)
485 +          groupToGtypeCol(j) = nGroupTypesCol
486 +       else
487 +          GtypeFound = .false.
488 +          do g = 1, nGroupTypesCol
489 +             if ( abs(groupMaxCutoffCol(j) - gtypeMaxCutoffCol(g)).lt.tol) then
490 +                groupToGtypeCol(j) = g
491 +                GtypeFound = .true.
492 +             endif
493 +          enddo
494 +          if (.not.GtypeFound) then            
495 +             nGroupTypesCol = nGroupTypesCol + 1
496 +             gtypeMaxCutoffCol(nGroupTypesCol) = groupMaxCutoffCol(j)
497 +             groupToGtypeCol(j) = nGroupTypesCol
498 +          endif
499 +       endif
500 +    enddo    
501 +
502 + #else
503 + ! Set pointers to information we just found
504 +    nGroupTypesCol = nGroupTypesRow
505 +    groupToGtypeCol => groupToGtypeRow
506 +    gtypeMaxCutoffCol => gtypeMaxCutoffRow
507 +    groupMaxCutoffCol => groupMaxCutoffRow
508 + #endif
509 +
510 +    !! allocate the gtypeCutoffMap here.
511 +    allocate(gtypeCutoffMap(nGroupTypesRow,nGroupTypesCol))
512 +    !! then we do a double loop over all the group TYPES to find the cutoff
513 +    !! map between groups of two types
514 +    tradRcut = max(maxval(gtypeMaxCutoffRow),maxval(gtypeMaxCutoffCol))
515 +
516 +    do i = 1, nGroupTypesRow      
517 +       do j = 1, nGroupTypesCol
518 +      
519 +          select case(cutoffPolicy)
520 +          case(TRADITIONAL_CUTOFF_POLICY)
521 +             thisRcut = tradRcut
522 +          case(MIX_CUTOFF_POLICY)
523 +             thisRcut = 0.5_dp * (gtypeMaxCutoffRow(i) + gtypeMaxCutoffCol(j))
524 +          case(MAX_CUTOFF_POLICY)
525 +             thisRcut = max(gtypeMaxCutoffRow(i), gtypeMaxCutoffCol(j))
526 +          case default
527 +             call handleError("createGtypeCutoffMap", "Unknown Cutoff Policy")
528 +             return
529 +          end select
530 +          gtypeCutoffMap(i,j)%rcut = thisRcut
531 +          
532 +          if (thisRcut.gt.largestRcut) largestRcut = thisRcut
533 +
534 +          gtypeCutoffMap(i,j)%rcutsq = thisRcut*thisRcut
535 +
536 +          if (.not.haveSkinThickness) then
537 +             skinThickness = 1.0_dp
538 +          endif
539 +
540 +          gtypeCutoffMap(i,j)%rlistsq = (thisRcut + skinThickness)**2
541 +
542 +          ! sanity check
543 +
544 +          if (haveDefaultCutoffs) then
545 +             if (abs(gtypeCutoffMap(i,j)%rcut - defaultRcut).gt.0.0001) then
546 +                call handleError("createGtypeCutoffMap", "user-specified rCut does not match computed group Cutoff")
547 +             endif
548 +          endif
549 +       enddo
550 +    enddo
551 +
552 +    if(allocated(gtypeMaxCutoffRow)) deallocate(gtypeMaxCutoffRow)
553 +    if(allocated(groupMaxCutoffRow)) deallocate(groupMaxCutoffRow)
554 +    if(allocated(atypeMaxCutoff)) deallocate(atypeMaxCutoff)
555 + #ifdef IS_MPI
556 +    if(associated(groupMaxCutoffCol)) deallocate(groupMaxCutoffCol)
557 +    if(associated(gtypeMaxCutoffCol)) deallocate(gtypeMaxCutoffCol)
558 + #endif
559 +    groupMaxCutoffCol => null()
560 +    gtypeMaxCutoffCol => null()
561 +    
562 +    haveGtypeCutoffMap = .true.
563 +   end subroutine createGtypeCutoffMap
564 +
565 +   subroutine setCutoffs(defRcut, defRsw)
566 +
567 +     real(kind=dp),intent(in) :: defRcut, defRsw
568 +     character(len = statusMsgSize) :: errMsg
569 +     integer :: localError
570 +
571 +     defaultRcut = defRcut
572 +     defaultRsw = defRsw
573 +    
574 +     defaultDoShift = .false.
575 +     if (abs(defaultRcut-defaultRsw) .lt. 0.0001) then
576 +        
577 +        write(errMsg, *) &
578 +             'cutoffRadius and switchingRadius are set to the same', newline &
579 +             // tab, 'value.  OOPSE will use shifted ', newline &
580 +             // tab, 'potentials instead of switching functions.'
581 +        
582 +        call handleInfo("setCutoffs", errMsg)
583 +        
584 +        defaultDoShift = .true.
585 +        
586 +     endif
587 +    
588 +     localError = 0
589 +     call setLJDefaultCutoff( defaultRcut, defaultDoShift )
590 +     call setElectrostaticCutoffRadius( defaultRcut, defaultRsw )
591 +     call setCutoffEAM( defaultRcut )
592 +     call setCutoffSC( defaultRcut )
593 +     call set_switch(defaultRsw, defaultRcut)
594 +     call setHmatDangerousRcutValue(defaultRcut)
595 +        
596 +     haveDefaultCutoffs = .true.
597 +     haveGtypeCutoffMap = .false.
598 +
599 +   end subroutine setCutoffs
600 +
601 +   subroutine cWasLame()
602 +    
603 +     VisitCutoffsAfterComputing = .true.
604 +     return
605 +    
606 +   end subroutine cWasLame
607 +  
608 +   subroutine setCutoffPolicy(cutPolicy)
609 +    
610 +     integer, intent(in) :: cutPolicy
611 +    
612 +     cutoffPolicy = cutPolicy
613 +     haveCutoffPolicy = .true.
614 +     haveGtypeCutoffMap = .false.
615 +    
616 +   end subroutine setCutoffPolicy
617 +    
618 +   subroutine setBoxDipole()
619 +
620 +     do_box_dipole = .true.
621 +    
622 +   end subroutine setBoxDipole
623 +
624 +   subroutine getBoxDipole( box_dipole )
625 +
626 +     real(kind=dp), intent(inout), dimension(3) :: box_dipole
627 +
628 +     box_dipole = boxDipole
629 +
630 +   end subroutine getBoxDipole
631 +
632 +   subroutine setElectrostaticMethod( thisESM )
633 +
634 +     integer, intent(in) :: thisESM
635 +
636 +     electrostaticSummationMethod = thisESM
637 +     haveElectrostaticSummationMethod = .true.
638 +    
639 +   end subroutine setElectrostaticMethod
640 +
641 +   subroutine setSkinThickness( thisSkin )
642 +    
643 +     real(kind=dp), intent(in) :: thisSkin
644 +    
645 +     skinThickness = thisSkin
646 +     haveSkinThickness = .true.    
647 +     haveGtypeCutoffMap = .false.
648 +    
649 +   end subroutine setSkinThickness
650 +      
651 +   subroutine setSimVariables()
652 +     SIM_uses_DirectionalAtoms = SimUsesDirectionalAtoms()
653 +     SIM_uses_EAM = SimUsesEAM()
654 +     SIM_requires_postpair_calc = SimRequiresPostpairCalc()
655 +     SIM_requires_prepair_calc = SimRequiresPrepairCalc()
656 +     SIM_uses_PBC = SimUsesPBC()
657 +     SIM_uses_SC = SimUsesSC()
658 +     SIM_uses_AtomicVirial = SimUsesAtomicVirial()
659 +
660 +     haveSIMvariables = .true.
661 +    
662 +     return
663 +   end subroutine setSimVariables
664 +
665 +  subroutine doReadyCheck(error)
666 +    integer, intent(out) :: error
667 +    integer :: myStatus
668 +
669 +    error = 0
670 +
671 +    if (.not. haveInteractionHash) then      
672 +       call createInteractionHash()      
673      endif
674  
675 <    if (.not. haveSIMvariables) then
676 <       call setSimVariables()
675 >    if (.not. haveGtypeCutoffMap) then        
676 >       call createGtypeCutoffMap()      
677      endif
678  
679 <    if (.not. haveRlist) then
680 <       write(default_error, *) 'rList has not been set in doForces!'
681 <       error = -1
682 <       return
679 >    if (VisitCutoffsAfterComputing) then
680 >       call set_switch(largestRcut, largestRcut)      
681 >       call setHmatDangerousRcutValue(largestRcut)
682 >       call setCutoffEAM(largestRcut)
683 >       call setCutoffSC(largestRcut)
684 >       VisitCutoffsAfterComputing = .false.
685      endif
686  
687 +    if (.not. haveSIMvariables) then
688 +       call setSimVariables()
689 +    endif
690 +
691      if (.not. haveNeighborList) then
692         write(default_error, *) 'neighbor list has not been initialized in doForces!'
693         error = -1
694         return
695      end if
696 <
696 >    
697      if (.not. haveSaneForceField) then
698         write(default_error, *) 'Force Field is not sane in doForces!'
699         error = -1
700         return
701      end if
702 <
702 >    
703   #ifdef IS_MPI
704      if (.not. isMPISimSet()) then
705         write(default_error,*) "ERROR: mpiSimulation has not been initialized!"
# Line 281 | Line 709 | contains
709   #endif
710      return
711    end subroutine doReadyCheck
284    
712  
286  subroutine init_FF(use_RF_c, thisStat)
713  
714 <    logical, intent(in) :: use_RF_c
714 >  subroutine init_FF(thisStat)
715  
716      integer, intent(out) :: thisStat  
717      integer :: my_status, nMatches
718      integer, pointer :: MatchList(:) => null()
293    real(kind=dp) :: rcut, rrf, rt, dielect
719  
720      !! assume things are copacetic, unless they aren't
721      thisStat = 0
722  
298    !! Fortran's version of a cast:
299    FF_uses_RF = use_RF_c
300    
723      !! init_FF is called *after* all of the atom types have been
724      !! defined in atype_module using the new_atype subroutine.
725      !!
726      !! this will scan through the known atypes and figure out what
727      !! interactions are used by the force field.    
728 <  
728 >
729      FF_uses_DirectionalAtoms = .false.
308    FF_uses_LennardJones = .false.
309    FF_uses_Electrostatic = .false.
310    FF_uses_Charges = .false.    
730      FF_uses_Dipoles = .false.
312    FF_uses_Sticky = .false.
731      FF_uses_GayBerne = .false.
732      FF_uses_EAM = .false.
733 <    FF_uses_Shapes = .false.
734 <    FF_uses_FLARB = .false.
317 <    
733 >    FF_uses_SC = .false.
734 >
735      call getMatchingElementList(atypes, "is_Directional", .true., &
736           nMatches, MatchList)
737      if (nMatches .gt. 0) FF_uses_DirectionalAtoms = .true.
738  
322    call getMatchingElementList(atypes, "is_LennardJones", .true., &
323         nMatches, MatchList)
324    if (nMatches .gt. 0) FF_uses_LennardJones = .true.
325    
326    call getMatchingElementList(atypes, "is_Electrostatic", .true., &
327         nMatches, MatchList)
328    if (nMatches .gt. 0) then
329       FF_uses_Electrostatic = .true.
330    endif
331
332    call getMatchingElementList(atypes, "is_Charge", .true., &
333         nMatches, MatchList)
334    if (nMatches .gt. 0) then
335       FF_uses_charges = .true.  
336       FF_uses_electrostatic = .true.
337    endif
338    
739      call getMatchingElementList(atypes, "is_Dipole", .true., &
740           nMatches, MatchList)
741 <    if (nMatches .gt. 0) then
342 <       FF_uses_dipoles = .true.
343 <       FF_uses_electrostatic = .true.
344 <       FF_uses_DirectionalAtoms = .true.
345 <    endif
741 >    if (nMatches .gt. 0) FF_uses_Dipoles = .true.
742      
347    call getMatchingElementList(atypes, "is_Sticky", .true., nMatches, &
348         MatchList)
349    if (nMatches .gt. 0) then
350       FF_uses_Sticky = .true.
351       FF_uses_DirectionalAtoms = .true.
352    endif
353    
743      call getMatchingElementList(atypes, "is_GayBerne", .true., &
744           nMatches, MatchList)
745 <    if (nMatches .gt. 0) then
746 <       FF_uses_GayBerne = .true.
358 <       FF_uses_DirectionalAtoms = .true.
359 <    endif
360 <    
745 >    if (nMatches .gt. 0) FF_uses_GayBerne = .true.
746 >
747      call getMatchingElementList(atypes, "is_EAM", .true., nMatches, MatchList)
748      if (nMatches .gt. 0) FF_uses_EAM = .true.
363    
364    call getMatchingElementList(atypes, "is_Shape", .true., &
365         nMatches, MatchList)
366    if (nMatches .gt. 0) then
367       FF_uses_Shapes = .true.
368       FF_uses_DirectionalAtoms = .true.
369    endif
749  
750 <    call getMatchingElementList(atypes, "is_FLARB", .true., &
751 <         nMatches, MatchList)
373 <    if (nMatches .gt. 0) FF_uses_FLARB = .true.
750 >    call getMatchingElementList(atypes, "is_SC", .true., nMatches, MatchList)
751 >    if (nMatches .gt. 0) FF_uses_SC = .true.
752  
753 <    !! Assume sanity (for the sake of argument)
753 >
754      haveSaneForceField = .true.
377    
378    !! check to make sure the FF_uses_RF setting makes sense
379    
380    if (FF_uses_dipoles) then
381       if (FF_uses_RF) then
382          dielect = getDielect()
383          call initialize_rf(dielect)
384       endif
385    else
386       if (FF_uses_RF) then          
387          write(default_error,*) 'Using Reaction Field with no dipoles?  Huh?'
388          thisStat = -1
389          haveSaneForceField = .false.
390          return
391       endif
392    endif
755  
394    !sticky module does not contain check_sticky_FF anymore
395    !if (FF_uses_sticky) then
396    !   call check_sticky_FF(my_status)
397    !   if (my_status /= 0) then
398    !      thisStat = -1
399    !      haveSaneForceField = .false.
400    !      return
401    !   end if
402    !endif
403
756      if (FF_uses_EAM) then
757 <         call init_EAM_FF(my_status)
757 >       call init_EAM_FF(my_status)
758         if (my_status /= 0) then
759            write(default_error, *) "init_EAM_FF returned a bad status"
760            thisStat = -1
# Line 411 | Line 763 | contains
763         end if
764      endif
765  
414    if (FF_uses_GayBerne) then
415       call check_gb_pair_FF(my_status)
416       if (my_status .ne. 0) then
417          thisStat = -1
418          haveSaneForceField = .false.
419          return
420       endif
421    endif
422
423    if (FF_uses_GayBerne .and. FF_uses_LennardJones) then
424    endif
425    
766      if (.not. haveNeighborList) then
767         !! Create neighbor lists
768         call expandNeighborList(nLocal, my_status)
# Line 432 | Line 772 | contains
772            return
773         endif
774         haveNeighborList = .true.
775 <    endif    
776 <    
775 >    endif
776 >
777    end subroutine init_FF
438  
778  
779 +
780    !! Does force loop over i,j pairs. Calls do_pair to calculates forces.
781    !------------------------------------------------------------->
782    subroutine do_force_loop(q, q_group, A, eFrame, f, t, tau, pot, &
# Line 456 | Line 796 | contains
796  
797      !! Stress Tensor
798      real( kind = dp), dimension(9) :: tau  
799 <    real ( kind = dp ) :: pot
799 >    real ( kind = dp ),dimension(LR_POT_TYPES) :: pot
800      logical ( kind = 2) :: do_pot_c, do_stress_c
801      logical :: do_pot
802      logical :: do_stress
803      logical :: in_switching_region
804   #ifdef IS_MPI
805 <    real( kind = DP ) :: pot_local
805 >    real( kind = DP ), dimension(LR_POT_TYPES) :: pot_local
806      integer :: nAtomsInRow
807      integer :: nAtomsInCol
808      integer :: nprocs
# Line 475 | Line 815 | contains
815      integer :: istart, iend
816      integer :: ia, jb, atom1, atom2
817      integer :: nlist
818 <    real( kind = DP ) :: ratmsq, rgrpsq, rgrp, vpair, vij
818 >    real( kind = DP ) :: ratmsq, rgrpsq, rgrp, rag, vpair, vij
819      real( kind = DP ) :: sw, dswdr, swderiv, mf
820 <    real(kind=dp),dimension(3) :: d_atm, d_grp, fpair, fij
821 <    real(kind=dp) :: rfpot, mu_i, virial
820 >    real( kind = DP ) :: rVal
821 >    real(kind=dp),dimension(3) :: d_atm, d_grp, fpair, fij, fg, dag
822 >    real(kind=dp) :: rfpot, mu_i
823 >    real(kind=dp):: rCut
824      integer :: me_i, me_j, n_in_i, n_in_j
825      logical :: is_dp_i
826      integer :: neighborListSize
# Line 486 | Line 828 | contains
828      integer :: localError
829      integer :: propPack_i, propPack_j
830      integer :: loopStart, loopEnd, loop
831 +    integer :: iHash
832 +    integer :: i1
833  
834 <    real(kind=dp) :: listSkin = 1.0  
835 <    
834 >    !! the variables for the box dipole moment
835 > #ifdef IS_MPI
836 >    integer :: pChgCount_local
837 >    integer :: nChgCount_local
838 >    real(kind=dp) :: pChg_local
839 >    real(kind=dp) :: nChg_local
840 >    real(kind=dp), dimension(3) :: pChgPos_local
841 >    real(kind=dp), dimension(3) :: nChgPos_local
842 >    real(kind=dp), dimension(3) :: dipVec_local
843 > #endif
844 >    integer :: pChgCount
845 >    integer :: nChgCount
846 >    real(kind=dp) :: pChg
847 >    real(kind=dp) :: nChg
848 >    real(kind=dp) :: chg_value
849 >    real(kind=dp), dimension(3) :: pChgPos
850 >    real(kind=dp), dimension(3) :: nChgPos
851 >    real(kind=dp), dimension(3) :: dipVec
852 >    real(kind=dp), dimension(3) :: chgVec
853 >
854 >    !! initialize box dipole variables
855 >    if (do_box_dipole) then
856 > #ifdef IS_MPI
857 >       pChg_local = 0.0_dp
858 >       nChg_local = 0.0_dp
859 >       pChgCount_local = 0
860 >       nChgCount_local = 0
861 >       do i=1, 3
862 >          pChgPos_local = 0.0_dp
863 >          nChgPos_local = 0.0_dp
864 >          dipVec_local = 0.0_dp
865 >       enddo
866 > #endif
867 >       pChg = 0.0_dp
868 >       nChg = 0.0_dp
869 >       pChgCount = 0
870 >       nChgCount = 0
871 >       chg_value = 0.0_dp
872 >      
873 >       do i=1, 3
874 >          pChgPos(i) = 0.0_dp
875 >          nChgPos(i) = 0.0_dp
876 >          dipVec(i) = 0.0_dp
877 >          chgVec(i) = 0.0_dp
878 >          boxDipole(i) = 0.0_dp
879 >       enddo
880 >    endif
881 >
882      !! initialize local variables  
883 <    
883 >
884   #ifdef IS_MPI
885      pot_local = 0.0_dp
886      nAtomsInRow   = getNatomsInRow(plan_atom_row)
# Line 500 | Line 890 | contains
890   #else
891      natoms = nlocal
892   #endif
893 <    
893 >
894      call doReadyCheck(localError)
895      if ( localError .ne. 0 ) then
896         call handleError("do_force_loop", "Not Initialized")
# Line 508 | Line 898 | contains
898         return
899      end if
900      call zero_work_arrays()
901 <        
901 >
902      do_pot = do_pot_c
903      do_stress = do_stress_c
904 <    
904 >
905      ! Gather all information needed by all force loops:
906 <    
906 >
907   #ifdef IS_MPI    
908 <    
908 >
909      call gather(q, q_Row, plan_atom_row_3d)
910      call gather(q, q_Col, plan_atom_col_3d)
911  
912      call gather(q_group, q_group_Row, plan_group_row_3d)
913      call gather(q_group, q_group_Col, plan_group_col_3d)
914 <        
914 >
915      if (FF_UsesDirectionalAtoms() .and. SIM_uses_DirectionalAtoms) then
916         call gather(eFrame, eFrame_Row, plan_atom_row_rotation)
917         call gather(eFrame, eFrame_Col, plan_atom_col_rotation)
918 <      
918 >
919         call gather(A, A_Row, plan_atom_row_rotation)
920         call gather(A, A_Col, plan_atom_col_rotation)
921      endif
922 <    
922 >
923   #endif
924 <    
924 >
925      !! Begin force loop timing:
926   #ifdef PROFILE
927      call cpu_time(forceTimeInitial)
928      nloops = nloops + 1
929   #endif
930 <    
930 >
931      loopEnd = PAIR_LOOP
932      if (FF_RequiresPrepairCalc() .and. SIM_requires_prepair_calc) then
933         loopStart = PREPAIR_LOOP
# Line 551 | Line 941 | contains
941         ! (but only on the first time through):
942         if (loop .eq. loopStart) then
943   #ifdef IS_MPI
944 <          call checkNeighborList(nGroupsInRow, q_group_row, listSkin, &
945 <             update_nlist)
944 >          call checkNeighborList(nGroupsInRow, q_group_row, skinThickness, &
945 >               update_nlist)
946   #else
947 <          call checkNeighborList(nGroups, q_group, listSkin, &
948 <             update_nlist)
947 >          call checkNeighborList(nGroups, q_group, skinThickness, &
948 >               update_nlist)
949   #endif
950         endif
951 <      
951 >
952         if (update_nlist) then
953            !! save current configuration and construct neighbor list
954   #ifdef IS_MPI
# Line 569 | Line 959 | contains
959            neighborListSize = size(list)
960            nlist = 0
961         endif
962 <      
962 >
963         istart = 1
964   #ifdef IS_MPI
965         iend = nGroupsInRow
# Line 579 | Line 969 | contains
969         outer: do i = istart, iend
970  
971            if (update_nlist) point(i) = nlist + 1
972 <          
972 >
973            n_in_i = groupStartRow(i+1) - groupStartRow(i)
974 <          
974 >
975            if (update_nlist) then
976   #ifdef IS_MPI
977               jstart = 1
# Line 596 | Line 986 | contains
986               ! make sure group i has neighbors
987               if (jstart .gt. jend) cycle outer
988            endif
989 <          
989 >
990            do jnab = jstart, jend
991               if (update_nlist) then
992                  j = jnab
# Line 605 | Line 995 | contains
995               endif
996  
997   #ifdef IS_MPI
998 +             me_j = atid_col(j)
999               call get_interatomic_vector(q_group_Row(:,i), &
1000                    q_group_Col(:,j), d_grp, rgrpsq)
1001   #else
1002 +             me_j = atid(j)
1003               call get_interatomic_vector(q_group(:,i), &
1004                    q_group(:,j), d_grp, rgrpsq)
1005 < #endif
1005 > #endif      
1006  
1007 <             if (rgrpsq < rlistsq) then
1007 >             if (rgrpsq < gtypeCutoffMap(groupToGtypeRow(i),groupToGtypeCol(j))%rListsq) then
1008                  if (update_nlist) then
1009                     nlist = nlist + 1
1010 <                  
1010 >
1011                     if (nlist > neighborListSize) then
1012   #ifdef IS_MPI                
1013                        call expandNeighborList(nGroupsInRow, listerror)
# Line 629 | Line 1021 | contains
1021                        end if
1022                        neighborListSize = size(list)
1023                     endif
1024 <                  
1024 >
1025                     list(nlist) = j
1026                  endif
1027 <                
1028 <                if (loop .eq. PAIR_LOOP) then
1029 <                   vij = 0.0d0
1030 <                   fij(1:3) = 0.0d0
1031 <                endif
1032 <                
1033 <                call get_switch(rgrpsq, sw, dswdr, rgrp, group_switch, &
1034 <                     in_switching_region)
1035 <                
1036 <                n_in_j = groupStartCol(j+1) - groupStartCol(j)
645 <                
646 <                do ia = groupStartRow(i), groupStartRow(i+1)-1
1027 >                
1028 >                if (rgrpsq < gtypeCutoffMap(groupToGtypeRow(i),groupToGtypeCol(j))%rCutsq) then
1029 >
1030 >                   rCut = gtypeCutoffMap(groupToGtypeRow(i),groupToGtypeCol(j))%rCut
1031 >                   if (loop .eq. PAIR_LOOP) then
1032 >                      vij = 0.0_dp
1033 >                      fij(1) = 0.0_dp
1034 >                      fij(2) = 0.0_dp
1035 >                      fij(3) = 0.0_dp
1036 >                   endif
1037                    
1038 <                   atom1 = groupListRow(ia)
1038 >                   call get_switch(rgrpsq, sw, dswdr,rgrp, in_switching_region)
1039                    
1040 <                   inner: do jb = groupStartCol(j), groupStartCol(j+1)-1
1040 >                   n_in_j = groupStartCol(j+1) - groupStartCol(j)
1041 >                  
1042 >                   do ia = groupStartRow(i), groupStartRow(i+1)-1
1043                        
1044 <                      atom2 = groupListCol(jb)
1044 >                      atom1 = groupListRow(ia)
1045                        
1046 <                      if (skipThisPair(atom1, atom2)) cycle inner
1047 <
1048 <                      if ((n_in_i .eq. 1).and.(n_in_j .eq. 1)) then
1049 <                         d_atm(1:3) = d_grp(1:3)
1050 <                         ratmsq = rgrpsq
1051 <                      else
1046 >                      inner: do jb = groupStartCol(j), groupStartCol(j+1)-1
1047 >                        
1048 >                         atom2 = groupListCol(jb)
1049 >                        
1050 >                         if (skipThisPair(atom1, atom2))  cycle inner
1051 >                        
1052 >                         if ((n_in_i .eq. 1).and.(n_in_j .eq. 1)) then
1053 >                            d_atm(1) = d_grp(1)
1054 >                            d_atm(2) = d_grp(2)
1055 >                            d_atm(3) = d_grp(3)
1056 >                            ratmsq = rgrpsq
1057 >                         else
1058   #ifdef IS_MPI
1059 <                         call get_interatomic_vector(q_Row(:,atom1), &
1060 <                              q_Col(:,atom2), d_atm, ratmsq)
1059 >                            call get_interatomic_vector(q_Row(:,atom1), &
1060 >                                 q_Col(:,atom2), d_atm, ratmsq)
1061   #else
1062 <                         call get_interatomic_vector(q(:,atom1), &
1063 <                              q(:,atom2), d_atm, ratmsq)
1062 >                            call get_interatomic_vector(q(:,atom1), &
1063 >                                 q(:,atom2), d_atm, ratmsq)
1064   #endif
1065 <                      endif
1066 <
1067 <                      if (loop .eq. PREPAIR_LOOP) then
1065 >                         endif
1066 >                        
1067 >                         if (loop .eq. PREPAIR_LOOP) then
1068   #ifdef IS_MPI                      
1069 <                         call do_prepair(atom1, atom2, ratmsq, d_atm, sw, &
1070 <                              rgrpsq, d_grp, do_pot, do_stress, &
1071 <                              eFrame, A, f, t, pot_local)
1069 >                            call do_prepair(atom1, atom2, ratmsq, d_atm, sw, &
1070 >                                 rgrpsq, d_grp, rCut, do_pot, do_stress, &
1071 >                                 eFrame, A, f, t, pot_local)
1072   #else
1073 <                         call do_prepair(atom1, atom2, ratmsq, d_atm, sw, &
1074 <                              rgrpsq, d_grp, do_pot, do_stress, &
1075 <                              eFrame, A, f, t, pot)
1073 >                            call do_prepair(atom1, atom2, ratmsq, d_atm, sw, &
1074 >                                 rgrpsq, d_grp, rCut, do_pot, do_stress, &
1075 >                                 eFrame, A, f, t, pot)
1076   #endif                                              
1077 <                      else
1077 >                         else
1078   #ifdef IS_MPI                      
1079 <                         call do_pair(atom1, atom2, ratmsq, d_atm, sw, &
1080 <                              do_pot, &
1081 <                              eFrame, A, f, t, pot_local, vpair, fpair)
1079 >                            call do_pair(atom1, atom2, ratmsq, d_atm, sw, &
1080 >                                 do_pot, eFrame, A, f, t, pot_local, vpair, &
1081 >                                 fpair, d_grp, rgrp, rCut)
1082   #else
1083 <                         call do_pair(atom1, atom2, ratmsq, d_atm, sw, &
1084 <                              do_pot,  &
1085 <                              eFrame, A, f, t, pot, vpair, fpair)
1083 >                            call do_pair(atom1, atom2, ratmsq, d_atm, sw, &
1084 >                                 do_pot, eFrame, A, f, t, pot, vpair, fpair, &
1085 >                                 d_grp, rgrp, rCut)
1086   #endif
1087 +                            vij = vij + vpair
1088 +                            fij(1) = fij(1) + fpair(1)
1089 +                            fij(2) = fij(2) + fpair(2)
1090 +                            fij(3) = fij(3) + fpair(3)
1091 +                            if (do_stress.and.SIM_uses_AtomicVirial) then
1092 +                               call add_stress_tensor(d_atm, fpair, tau)
1093 +                            endif
1094 +                         endif
1095 +                      enddo inner
1096 +                   enddo
1097  
1098 <                         vij = vij + vpair
1099 <                         fij(1:3) = fij(1:3) + fpair(1:3)
1100 <                      endif
1101 <                   enddo inner
1102 <                enddo
1103 <                
1104 <                if (loop .eq. PAIR_LOOP) then
1105 <                   if (in_switching_region) then
1106 <                      swderiv = vij*dswdr/rgrp
1107 <                      fij(1) = fij(1) + swderiv*d_grp(1)
1108 <                      fij(2) = fij(2) + swderiv*d_grp(2)
1109 <                      fij(3) = fij(3) + swderiv*d_grp(3)
1110 <                      
1111 <                      do ia=groupStartRow(i), groupStartRow(i+1)-1
1112 <                         atom1=groupListRow(ia)
1113 <                         mf = mfactRow(atom1)
1114 < #ifdef IS_MPI
1115 <                         f_Row(1,atom1) = f_Row(1,atom1) + swderiv*d_grp(1)*mf
1116 <                         f_Row(2,atom1) = f_Row(2,atom1) + swderiv*d_grp(2)*mf
1117 <                         f_Row(3,atom1) = f_Row(3,atom1) + swderiv*d_grp(3)*mf
1118 < #else
711 <                         f(1,atom1) = f(1,atom1) + swderiv*d_grp(1)*mf
712 <                         f(2,atom1) = f(2,atom1) + swderiv*d_grp(2)*mf
713 <                         f(3,atom1) = f(3,atom1) + swderiv*d_grp(3)*mf
1098 >                   if (loop .eq. PAIR_LOOP) then
1099 >                      if (in_switching_region) then
1100 >                         swderiv = vij*dswdr/rgrp
1101 >                         fij(1) = fij(1) + swderiv*d_grp(1)
1102 >                         fij(2) = fij(2) + swderiv*d_grp(2)
1103 >                         fij(3) = fij(3) + swderiv*d_grp(3)
1104 >                        
1105 >                         do ia=groupStartRow(i), groupStartRow(i+1)-1
1106 >                            atom1=groupListRow(ia)
1107 >                            mf = mfactRow(atom1)
1108 >                            ! fg is the force on atom ia due to cutoff group's
1109 >                            ! presence in switching region
1110 >                            fg = swderiv*d_grp*mf
1111 > #ifdef IS_MPI
1112 >                            f_Row(1,atom1) = f_Row(1,atom1) + fg(1)
1113 >                            f_Row(2,atom1) = f_Row(2,atom1) + fg(2)
1114 >                            f_Row(3,atom1) = f_Row(3,atom1) + fg(3)
1115 > #else
1116 >                            f(1,atom1) = f(1,atom1) + fg(1)
1117 >                            f(2,atom1) = f(2,atom1) + fg(2)
1118 >                            f(3,atom1) = f(3,atom1) + fg(3)
1119   #endif
1120 <                      enddo
1121 <                      
1122 <                      do jb=groupStartCol(j), groupStartCol(j+1)-1
718 <                         atom2=groupListCol(jb)
719 <                         mf = mfactCol(atom2)
1120 >                            if (do_stress.and.SIM_uses_AtomicVirial) then
1121 >                               ! find the distance between the atom and the center of
1122 >                               ! the cutoff group:
1123   #ifdef IS_MPI
1124 <                         f_Col(1,atom2) = f_Col(1,atom2) - swderiv*d_grp(1)*mf
1125 <                         f_Col(2,atom2) = f_Col(2,atom2) - swderiv*d_grp(2)*mf
723 <                         f_Col(3,atom2) = f_Col(3,atom2) - swderiv*d_grp(3)*mf
1124 >                               call get_interatomic_vector(q_Row(:,atom1), &
1125 >                                    q_group_Row(:,i), dag, rag)
1126   #else
1127 <                         f(1,atom2) = f(1,atom2) - swderiv*d_grp(1)*mf
1128 <                         f(2,atom2) = f(2,atom2) - swderiv*d_grp(2)*mf
727 <                         f(3,atom2) = f(3,atom2) - swderiv*d_grp(3)*mf
1127 >                               call get_interatomic_vector(q(:,atom1), &
1128 >                                    q_group(:,i), dag, rag)
1129   #endif
1130 <                      enddo
1130 >                               call add_stress_tensor(dag,fg,tau)
1131 >                            endif
1132 >                         enddo
1133 >                        
1134 >                         do jb=groupStartCol(j), groupStartCol(j+1)-1
1135 >                            atom2=groupListCol(jb)
1136 >                            mf = mfactCol(atom2)
1137 >                            ! fg is the force on atom jb due to cutoff group's
1138 >                            ! presence in switching region
1139 >                            fg = -swderiv*d_grp*mf
1140 > #ifdef IS_MPI
1141 >                            f_Col(1,atom2) = f_Col(1,atom2) + fg(1)
1142 >                            f_Col(2,atom2) = f_Col(2,atom2) + fg(2)
1143 >                            f_Col(3,atom2) = f_Col(3,atom2) + fg(3)
1144 > #else
1145 >                            f(1,atom2) = f(1,atom2) + fg(1)
1146 >                            f(2,atom2) = f(2,atom2) + fg(2)
1147 >                            f(3,atom2) = f(3,atom2) + fg(3)
1148 > #endif
1149 >                            if (do_stress.and.SIM_uses_AtomicVirial) then
1150 >                               ! find the distance between the atom and the center of
1151 >                               ! the cutoff group:
1152 > #ifdef IS_MPI
1153 >                               call get_interatomic_vector(q_Col(:,atom2), &
1154 >                                    q_group_Col(:,j), dag, rag)
1155 > #else
1156 >                               call get_interatomic_vector(q(:,atom2), &
1157 >                                    q_group(:,j), dag, rag)
1158 > #endif
1159 >                               call add_stress_tensor(dag,fg,tau)                              
1160 >                            endif
1161 >                            
1162 >                         enddo
1163 >                      endif
1164 >                      if (do_stress.and.(.not.SIM_uses_AtomicVirial)) then
1165 >                         call add_stress_tensor(d_grp, fij, tau)
1166 >                      endif
1167                     endif
731                  
732                   if (do_stress) call add_stress_tensor(d_grp, fij)
1168                  endif
1169 <             end if
1169 >             endif
1170            enddo
1171 +          
1172         enddo outer
1173 <      
1173 >
1174         if (update_nlist) then
1175   #ifdef IS_MPI
1176            point(nGroupsInRow + 1) = nlist + 1
# Line 748 | Line 1184 | contains
1184               update_nlist = .false.                              
1185            endif
1186         endif
1187 <            
1187 >
1188         if (loop .eq. PREPAIR_LOOP) then
1189            call do_preforce(nlocal, pot)
1190         endif
1191 <      
1191 >
1192      enddo
1193 <    
1193 >
1194      !! Do timing
1195   #ifdef PROFILE
1196      call cpu_time(forceTimeFinal)
1197      forceTime = forceTime + forceTimeFinal - forceTimeInitial
1198   #endif    
1199 <    
1199 >
1200   #ifdef IS_MPI
1201      !!distribute forces
1202 <    
1202 >
1203      f_temp = 0.0_dp
1204      call scatter(f_Row,f_temp,plan_atom_row_3d)
1205      do i = 1,nlocal
1206         f(1:3,i) = f(1:3,i) + f_temp(1:3,i)
1207      end do
1208 <    
1208 >
1209      f_temp = 0.0_dp
1210      call scatter(f_Col,f_temp,plan_atom_col_3d)
1211      do i = 1,nlocal
1212         f(1:3,i) = f(1:3,i) + f_temp(1:3,i)
1213      end do
1214 <    
1214 >
1215      if (FF_UsesDirectionalAtoms() .and. SIM_uses_DirectionalAtoms) then
1216         t_temp = 0.0_dp
1217         call scatter(t_Row,t_temp,plan_atom_row_3d)
# Line 784 | Line 1220 | contains
1220         end do
1221         t_temp = 0.0_dp
1222         call scatter(t_Col,t_temp,plan_atom_col_3d)
1223 <      
1223 >
1224         do i = 1,nlocal
1225            t(1:3,i) = t(1:3,i) + t_temp(1:3,i)
1226         end do
1227      endif
1228 <    
1228 >
1229      if (do_pot) then
1230         ! scatter/gather pot_row into the members of my column
1231 <       call scatter(pot_Row, pot_Temp, plan_atom_row)
1232 <      
1231 >       do i = 1,LR_POT_TYPES
1232 >          call scatter(pot_Row(i,:), pot_Temp(i,:), plan_atom_row)
1233 >       end do
1234         ! scatter/gather pot_local into all other procs
1235         ! add resultant to get total pot
1236         do i = 1, nlocal
1237 <          pot_local = pot_local + pot_Temp(i)
1237 >          pot_local(1:LR_POT_TYPES) = pot_local(1:LR_POT_TYPES) &
1238 >               + pot_Temp(1:LR_POT_TYPES,i)
1239         enddo
1240 <      
1240 >
1241         pot_Temp = 0.0_DP
1242 <      
1243 <       call scatter(pot_Col, pot_Temp, plan_atom_col)
1242 >       do i = 1,LR_POT_TYPES
1243 >          call scatter(pot_Col(i,:), pot_Temp(i,:), plan_atom_col)
1244 >       end do
1245         do i = 1, nlocal
1246 <          pot_local = pot_local + pot_Temp(i)
1246 >          pot_local(1:LR_POT_TYPES) = pot_local(1:LR_POT_TYPES)&
1247 >               + pot_Temp(1:LR_POT_TYPES,i)
1248         enddo
1249 <      
1249 >
1250      endif
1251   #endif
1252 <    
1253 <    if (FF_RequiresPostpairCalc() .and. SIM_requires_postpair_calc) then
1254 <      
815 <       if (FF_uses_RF .and. SIM_uses_RF) then
1252 >
1253 >    if (SIM_requires_postpair_calc) then
1254 >       do i = 1, nlocal            
1255            
1256 +          ! we loop only over the local atoms, so we don't need row and column
1257 +          ! lookups for the types
1258 +          
1259 +          me_i = atid(i)
1260 +          
1261 +          ! is the atom electrostatic?  See if it would have an
1262 +          ! electrostatic interaction with itself
1263 +          iHash = InteractionHash(me_i,me_i)
1264 +
1265 +          if ( iand(iHash, ELECTROSTATIC_PAIR).ne.0 ) then
1266   #ifdef IS_MPI
1267 <          call scatter(rf_Row,rf,plan_atom_row_3d)
1268 <          call scatter(rf_Col,rf_Temp,plan_atom_col_3d)
1269 <          do i = 1,nlocal
1270 <             rf(1:3,i) = rf(1:3,i) + rf_Temp(1:3,i)
1271 <          end do
1267 >             call self_self(i, eFrame, pot_local(ELECTROSTATIC_POT), &
1268 >                  t, do_pot)
1269 > #else
1270 >             call self_self(i, eFrame, pot(ELECTROSTATIC_POT), &
1271 >                  t, do_pot)
1272   #endif
1273 +          endif
1274 +  
1275            
1276 <          do i = 1, nLocal
1276 >          if (electrostaticSummationMethod.eq.REACTION_FIELD) then
1277              
1278 <             rfpot = 0.0_DP
1278 >             ! loop over the excludes to accumulate RF stuff we've
1279 >             ! left out of the normal pair loop
1280 >            
1281 >             do i1 = 1, nSkipsForAtom(i)
1282 >                j = skipsForAtom(i, i1)
1283 >                
1284 >                ! prevent overcounting of the skips
1285 >                if (i.lt.j) then
1286 >                   call get_interatomic_vector(q(:,i), q(:,j), d_atm, ratmsq)
1287 >                   rVal = sqrt(ratmsq)
1288 >                   call get_switch(ratmsq, sw, dswdr, rVal,in_switching_region)
1289   #ifdef IS_MPI
1290 <             me_i = atid_row(i)
1290 >                   call rf_self_excludes(i, j, sw, eFrame, d_atm, rVal, &
1291 >                        vpair, pot_local(ELECTROSTATIC_POT), f, t, do_pot)
1292   #else
1293 <             me_i = atid(i)
1293 >                   call rf_self_excludes(i, j, sw, eFrame, d_atm, rVal, &
1294 >                        vpair, pot(ELECTROSTATIC_POT), f, t, do_pot)
1295   #endif
1296 <            
1297 <             if (PropertyMap(me_i)%is_Dipole) then
1298 <                
1299 <                mu_i = getDipoleMoment(me_i)
1300 <                
838 <                !! The reaction field needs to include a self contribution
839 <                !! to the field:
840 <                call accumulate_self_rf(i, mu_i, eFrame)
841 <                !! Get the reaction field contribution to the
842 <                !! potential and torques:
843 <                call reaction_field_final(i, mu_i, eFrame, rfpot, t, do_pot)
1296 >                endif
1297 >             enddo
1298 >          endif
1299 >
1300 >          if (do_box_dipole) then
1301   #ifdef IS_MPI
1302 <                pot_local = pot_local + rfpot
1302 >             call accumulate_box_dipole(i, eFrame, q(:,i), pChg_local, &
1303 >                  nChg_local, pChgPos_local, nChgPos_local, dipVec_local, &
1304 >                  pChgCount_local, nChgCount_local)
1305   #else
1306 <                pot = pot + rfpot
1307 <      
1306 >             call accumulate_box_dipole(i, eFrame, q(:,i), pChg, nChg, &
1307 >                  pChgPos, nChgPos, dipVec, pChgCount, nChgCount)
1308   #endif
1309 <             endif            
1310 <          enddo
852 <       endif
1309 >          endif
1310 >       enddo
1311      endif
1312 <    
855 <    
1312 >
1313   #ifdef IS_MPI
857    
1314      if (do_pot) then
1315 <       pot = pot + pot_local
1316 <       !! we assume the c code will do the allreduce to get the total potential
1317 <       !! we could do it right here if we needed to...
1315 > #ifdef SINGLE_PRECISION
1316 >       call mpi_allreduce(pot_local, pot, LR_POT_TYPES,mpi_real,mpi_sum, &
1317 >            mpi_comm_world,mpi_err)            
1318 > #else
1319 >       call mpi_allreduce(pot_local, pot, LR_POT_TYPES,mpi_double_precision, &
1320 >            mpi_sum, mpi_comm_world,mpi_err)            
1321 > #endif
1322      endif
1323 <    
1324 <    if (do_stress) then
1325 <       call mpi_allreduce(tau_Temp, tau, 9,mpi_double_precision,mpi_sum, &
1326 <            mpi_comm_world,mpi_err)
1327 <       call mpi_allreduce(virial_Temp, virial,1,mpi_double_precision,mpi_sum, &
1328 <            mpi_comm_world,mpi_err)
1329 <    endif
1330 <    
1323 >        
1324 >    if (do_box_dipole) then
1325 >
1326 > #ifdef SINGLE_PRECISION
1327 >       call mpi_allreduce(pChg_local, pChg, 1, mpi_real, mpi_sum, &
1328 >            mpi_comm_world, mpi_err)
1329 >       call mpi_allreduce(nChg_local, nChg, 1, mpi_real, mpi_sum, &
1330 >            mpi_comm_world, mpi_err)
1331 >       call mpi_allreduce(pChgCount_local, pChgCount, 1, mpi_integer, mpi_sum,&
1332 >            mpi_comm_world, mpi_err)
1333 >       call mpi_allreduce(nChgCount_local, nChgCount, 1, mpi_integer, mpi_sum,&
1334 >            mpi_comm_world, mpi_err)
1335 >       call mpi_allreduce(pChgPos_local, pChgPos, 3, mpi_real, mpi_sum, &
1336 >            mpi_comm_world, mpi_err)
1337 >       call mpi_allreduce(nChgPos_local, nChgPos, 3, mpi_real, mpi_sum, &
1338 >            mpi_comm_world, mpi_err)
1339 >       call mpi_allreduce(dipVec_local, dipVec, 3, mpi_real, mpi_sum, &
1340 >            mpi_comm_world, mpi_err)
1341   #else
1342 <    
1343 <    if (do_stress) then
1344 <       tau = tau_Temp
1345 <       virial = virial_Temp
1342 >       call mpi_allreduce(pChg_local, pChg, 1, mpi_double_precision, mpi_sum, &
1343 >            mpi_comm_world, mpi_err)
1344 >       call mpi_allreduce(nChg_local, nChg, 1, mpi_double_precision, mpi_sum, &
1345 >            mpi_comm_world, mpi_err)
1346 >       call mpi_allreduce(pChgCount_local, pChgCount, 1, mpi_integer,&
1347 >            mpi_sum, mpi_comm_world, mpi_err)
1348 >       call mpi_allreduce(nChgCount_local, nChgCount, 1, mpi_integer,&
1349 >            mpi_sum, mpi_comm_world, mpi_err)
1350 >       call mpi_allreduce(pChgPos_local, pChgPos, 3, mpi_double_precision, &
1351 >            mpi_sum, mpi_comm_world, mpi_err)
1352 >       call mpi_allreduce(nChgPos_local, nChgPos, 3, mpi_double_precision, &
1353 >            mpi_sum, mpi_comm_world, mpi_err)
1354 >       call mpi_allreduce(dipVec_local, dipVec, 3, mpi_double_precision, &
1355 >            mpi_sum, mpi_comm_world, mpi_err)
1356 > #endif
1357 >
1358      endif
1359      
1360   #endif
1361 +
1362 +    if (do_box_dipole) then
1363 +       ! first load the accumulated dipole moment (if dipoles were present)
1364 +       boxDipole(1) = dipVec(1)
1365 +       boxDipole(2) = dipVec(2)
1366 +       boxDipole(3) = dipVec(3)
1367 +
1368 +       ! now include the dipole moment due to charges
1369 +       ! use the lesser of the positive and negative charge totals
1370 +       if (nChg .le. pChg) then
1371 +          chg_value = nChg
1372 +       else
1373 +          chg_value = pChg
1374 +       endif
1375        
1376 +       ! find the average positions
1377 +       if (pChgCount .gt. 0 .and. nChgCount .gt. 0) then
1378 +          pChgPos = pChgPos / pChgCount
1379 +          nChgPos = nChgPos / nChgCount
1380 +       endif
1381 +
1382 +       ! dipole is from the negative to the positive (physics notation)
1383 +       chgVec(1) = pChgPos(1) - nChgPos(1)
1384 +       chgVec(2) = pChgPos(2) - nChgPos(2)
1385 +       chgVec(3) = pChgPos(3) - nChgPos(3)
1386 +
1387 +       boxDipole(1) = boxDipole(1) + chgVec(1) * chg_value
1388 +       boxDipole(2) = boxDipole(2) + chgVec(2) * chg_value
1389 +       boxDipole(3) = boxDipole(3) + chgVec(3) * chg_value
1390 +
1391 +    endif
1392 +
1393    end subroutine do_force_loop
1394 <  
1394 >
1395    subroutine do_pair(i, j, rijsq, d, sw, do_pot, &
1396 <       eFrame, A, f, t, pot, vpair, fpair)
1396 >       eFrame, A, f, t, pot, vpair, fpair, d_grp, r_grp, rCut)
1397  
1398 <    real( kind = dp ) :: pot, vpair, sw
1398 >    real( kind = dp ) :: vpair, sw
1399 >    real( kind = dp ), dimension(LR_POT_TYPES) :: pot
1400      real( kind = dp ), dimension(3) :: fpair
1401      real( kind = dp ), dimension(nLocal)   :: mfact
1402      real( kind = dp ), dimension(9,nLocal) :: eFrame
# Line 893 | Line 1407 | contains
1407      logical, intent(inout) :: do_pot
1408      integer, intent(in) :: i, j
1409      real ( kind = dp ), intent(inout) :: rijsq
1410 <    real ( kind = dp )                :: r
1410 >    real ( kind = dp ), intent(inout) :: r_grp
1411      real ( kind = dp ), intent(inout) :: d(3)
1412 +    real ( kind = dp ), intent(inout) :: d_grp(3)
1413 +    real ( kind = dp ), intent(inout) :: rCut
1414 +    real ( kind = dp ) :: r
1415 +    real ( kind = dp ) :: a_k, b_k, c_k, d_k, dx
1416      integer :: me_i, me_j
1417 +    integer :: k
1418  
1419 +    integer :: iHash
1420 +
1421      r = sqrt(rijsq)
1422 <    vpair = 0.0d0
1423 <    fpair(1:3) = 0.0d0
1422 >    
1423 >    vpair = 0.0_dp
1424 >    fpair(1:3) = 0.0_dp
1425  
1426   #ifdef IS_MPI
1427      me_i = atid_row(i)
# Line 909 | Line 1431 | contains
1431      me_j = atid(j)
1432   #endif
1433  
1434 < !    write(*,*) i, j, me_i, me_j
1434 >    iHash = InteractionHash(me_i, me_j)
1435      
1436 <    if (FF_uses_LennardJones .and. SIM_uses_LennardJones) then
1437 <      
1438 <       if ( PropertyMap(me_i)%is_LennardJones .and. &
917 <            PropertyMap(me_j)%is_LennardJones ) then
918 <          call do_lj_pair(i, j, d, r, rijsq, sw, vpair, fpair, pot, f, do_pot)
919 <       endif
920 <      
1436 >    if ( iand(iHash, LJ_PAIR).ne.0 ) then
1437 >       call do_lj_pair(i, j, d, r, rijsq, rcut, sw, vpair, fpair, &
1438 >            pot(VDW_POT), f, do_pot)
1439      endif
1440      
1441 <    if (FF_uses_charges .and. SIM_uses_charges) then
1442 <      
1443 <       if (PropertyMap(me_i)%is_Charge .and. PropertyMap(me_j)%is_Charge) then
926 <          call do_charge_pair(i, j, d, r, rijsq, sw, vpair, fpair, &
927 <               pot, f, do_pot)
928 <       endif
929 <      
1441 >    if ( iand(iHash, ELECTROSTATIC_PAIR).ne.0 ) then
1442 >       call doElectrostaticPair(i, j, d, r, rijsq, rcut, sw, vpair, fpair, &
1443 >            pot(ELECTROSTATIC_POT), eFrame, f, t, do_pot)
1444      endif
1445      
1446 <    if (FF_uses_dipoles .and. SIM_uses_dipoles) then
1447 <      
1448 <       if ( PropertyMap(me_i)%is_Dipole .and. PropertyMap(me_j)%is_Dipole) then
1449 <          call do_dipole_pair(i, j, d, r, rijsq, sw, vpair, fpair, &
1450 <               pot, eFrame, f, t, do_pot)
1451 <          if (FF_uses_RF .and. SIM_uses_RF) then
1452 <             call accumulate_rf(i, j, r, eFrame, sw)
1453 <             call rf_correct_forces(i, j, d, r, eFrame, sw, f, fpair)
1454 <          endif
1455 <       endif
1446 >    if ( iand(iHash, STICKY_PAIR).ne.0 ) then
1447 >       call do_sticky_pair(i, j, d, r, rijsq, sw, vpair, fpair, &
1448 >            pot(HB_POT), A, f, t, do_pot)
1449 >    endif
1450 >    
1451 >    if ( iand(iHash, STICKYPOWER_PAIR).ne.0 ) then
1452 >       call do_sticky_power_pair(i, j, d, r, rijsq, sw, vpair, fpair, &
1453 >            pot(HB_POT), A, f, t, do_pot)
1454 >    endif
1455 >    
1456 >    if ( iand(iHash, GAYBERNE_PAIR).ne.0 ) then
1457 >       call do_gb_pair(i, j, d, r, rijsq, sw, vpair, fpair, &
1458 >            pot(VDW_POT), A, f, t, do_pot)
1459 >    endif
1460 >    
1461 >    if ( iand(iHash, GAYBERNE_LJ).ne.0 ) then
1462 >       call do_gb_pair(i, j, d, r, rijsq, sw, vpair, fpair, &
1463 >            pot(VDW_POT), A, f, t, do_pot)
1464 >    endif
1465 >    
1466 >    if ( iand(iHash, EAM_PAIR).ne.0 ) then      
1467 >       call do_eam_pair(i, j, d, r, rijsq, sw, vpair, fpair, &
1468 >            pot(METALLIC_POT), f, do_pot)
1469 >    endif
1470 >    
1471 >    if ( iand(iHash, SHAPE_PAIR).ne.0 ) then      
1472 >       call do_shape_pair(i, j, d, r, rijsq, sw, vpair, fpair, &
1473 >            pot(VDW_POT), A, f, t, do_pot)
1474 >    endif
1475 >    
1476 >    if ( iand(iHash, SHAPE_LJ).ne.0 ) then      
1477 >       call do_shape_pair(i, j, d, r, rijsq, sw, vpair, fpair, &
1478 >            pot(VDW_POT), A, f, t, do_pot)
1479 >    endif
1480  
1481 +    if ( iand(iHash, SC_PAIR).ne.0 ) then      
1482 +       call do_SC_pair(i, j, d, r, rijsq, rcut, sw, vpair, fpair, &
1483 +            pot(METALLIC_POT), f, do_pot)
1484      endif
1485 +    
1486 +  end subroutine do_pair
1487  
1488 <    if (FF_uses_Sticky .and. SIM_uses_sticky) then
1488 >  subroutine do_prepair(i, j, rijsq, d, sw, rcijsq, dc, rCut, &
1489 >       do_pot, do_stress, eFrame, A, f, t, pot)
1490  
1491 <       if ( PropertyMap(me_i)%is_Sticky .and. PropertyMap(me_j)%is_Sticky) then
1492 <          call do_sticky_pair(i, j, d, r, rijsq, sw, vpair, fpair, &
1493 <               pot, A, f, t, do_pot)
1494 <       endif
1495 <      
1496 <    endif
1491 >    real( kind = dp ) :: sw
1492 >    real( kind = dp ), dimension(LR_POT_TYPES) :: pot
1493 >    real( kind = dp ), dimension(9,nLocal) :: eFrame
1494 >    real (kind=dp), dimension(9,nLocal) :: A
1495 >    real (kind=dp), dimension(3,nLocal) :: f
1496 >    real (kind=dp), dimension(3,nLocal) :: t
1497  
1498 +    logical, intent(inout) :: do_pot, do_stress
1499 +    integer, intent(in) :: i, j
1500 +    real ( kind = dp ), intent(inout)    :: rijsq, rcijsq, rCut
1501 +    real ( kind = dp )                :: r, rc
1502 +    real ( kind = dp ), intent(inout) :: d(3), dc(3)
1503  
1504 <    if (FF_uses_GayBerne .and. SIM_uses_GayBerne) then
1505 <      
1506 <       if ( PropertyMap(me_i)%is_GayBerne .and. &
958 <            PropertyMap(me_j)%is_GayBerne) then
959 <          call do_gb_pair(i, j, d, r, rijsq, sw, vpair, fpair, &
960 <               pot, A, f, t, do_pot)
961 <       endif
962 <      
963 <    endif
1504 >    integer :: me_i, me_j, iHash
1505 >
1506 >    r = sqrt(rijsq)
1507      
1508 <    if (FF_uses_EAM .and. SIM_uses_EAM) then
1509 <      
1510 <       if ( PropertyMap(me_i)%is_EAM .and. PropertyMap(me_j)%is_EAM) then
1511 <          call do_eam_pair(i, j, d, r, rijsq, sw, vpair, fpair, pot, f, &
1512 <               do_pot)
1513 <       endif
1514 <      
972 <    endif
1508 > #ifdef IS_MPI  
1509 >    me_i = atid_row(i)
1510 >    me_j = atid_col(j)  
1511 > #else  
1512 >    me_i = atid(i)
1513 >    me_j = atid(j)  
1514 > #endif
1515  
1516 +    iHash = InteractionHash(me_i, me_j)
1517  
1518 < !    write(*,*) PropertyMap(me_i)%is_Shape,PropertyMap(me_j)%is_Shape
1518 >    if ( iand(iHash, EAM_PAIR).ne.0 ) then      
1519 >            call calc_EAM_prepair_rho(i, j, d, r, rijsq)
1520 >    endif
1521  
1522 <    if (FF_uses_Shapes .and. SIM_uses_Shapes) then
1523 <       if ( PropertyMap(me_i)%is_Shape .and. &
979 <            PropertyMap(me_j)%is_Shape ) then
980 <          call do_shape_pair(i, j, d, r, rijsq, sw, vpair, fpair, &
981 <               pot, A, f, t, do_pot)
982 <       endif
983 <      
1522 >    if ( iand(iHash, SC_PAIR).ne.0 ) then      
1523 >            call calc_SC_prepair_rho(i, j, d, r, rijsq, rcut )
1524      endif
1525      
1526 <  end subroutine do_pair
1526 >  end subroutine do_prepair
1527  
988  subroutine do_prepair(i, j, rijsq, d, sw, rcijsq, dc, &
989       do_pot, do_stress, eFrame, A, f, t, pot)
1528  
1529 <   real( kind = dp ) :: pot, sw
1530 <   real( kind = dp ), dimension(9,nLocal) :: eFrame
1531 <   real (kind=dp), dimension(9,nLocal) :: A
994 <   real (kind=dp), dimension(3,nLocal) :: f
995 <   real (kind=dp), dimension(3,nLocal) :: t
996 <  
997 <   logical, intent(inout) :: do_pot, do_stress
998 <   integer, intent(in) :: i, j
999 <   real ( kind = dp ), intent(inout)    :: rijsq, rcijsq
1000 <   real ( kind = dp )                :: r, rc
1001 <   real ( kind = dp ), intent(inout) :: d(3), dc(3)
1002 <  
1003 <   logical :: is_EAM_i, is_EAM_j
1004 <  
1005 <   integer :: me_i, me_j
1006 <  
1529 >  subroutine do_preforce(nlocal,pot)
1530 >    integer :: nlocal
1531 >    real( kind = dp ),dimension(LR_POT_TYPES) :: pot
1532  
1533 <    r = sqrt(rijsq)
1534 <    if (SIM_uses_molecular_cutoffs) then
1010 <       rc = sqrt(rcijsq)
1011 <    else
1012 <       rc = r
1533 >    if (FF_uses_EAM .and. SIM_uses_EAM) then
1534 >       call calc_EAM_preforce_Frho(nlocal,pot(METALLIC_POT))
1535      endif
1536 <  
1536 >    if (FF_uses_SC .and. SIM_uses_SC) then
1537 >       call calc_SC_preforce_Frho(nlocal,pot(METALLIC_POT))
1538 >    endif
1539 >  end subroutine do_preforce
1540  
1541 < #ifdef IS_MPI  
1542 <   me_i = atid_row(i)
1543 <   me_j = atid_col(j)  
1544 < #else  
1545 <   me_i = atid(i)
1546 <   me_j = atid(j)  
1547 < #endif
1548 <  
1549 <   if (FF_uses_EAM .and. SIM_uses_EAM) then
1550 <      
1551 <      if (PropertyMap(me_i)%is_EAM .and. PropertyMap(me_j)%is_EAM) &
1552 <           call calc_EAM_prepair_rho(i, j, d, r, rijsq )
1553 <      
1554 <   endif
1555 <  
1556 < end subroutine do_prepair
1557 <
1558 <
1559 < subroutine do_preforce(nlocal,pot)
1560 <   integer :: nlocal
1561 <   real( kind = dp ) :: pot
1562 <  
1563 <   if (FF_uses_EAM .and. SIM_uses_EAM) then
1564 <      call calc_EAM_preforce_Frho(nlocal,pot)
1565 <   endif
1566 <  
1567 <  
1568 < end subroutine do_preforce
1569 <
1570 <
1571 < subroutine get_interatomic_vector(q_i, q_j, d, r_sq)
1572 <  
1573 <   real (kind = dp), dimension(3) :: q_i
1574 <   real (kind = dp), dimension(3) :: q_j
1575 <   real ( kind = dp ), intent(out) :: r_sq
1576 <   real( kind = dp ) :: d(3), scaled(3)
1577 <   integer i
1578 <  
1579 <   d(1:3) = q_j(1:3) - q_i(1:3)
1580 <  
1581 <   ! Wrap back into periodic box if necessary
1582 <   if ( SIM_uses_PBC ) then
1583 <      
1584 <      if( .not.boxIsOrthorhombic ) then
1585 <         ! calc the scaled coordinates.
1586 <        
1587 <         scaled = matmul(HmatInv, d)
1588 <        
1589 <         ! wrap the scaled coordinates
1590 <        
1591 <         scaled = scaled  - anint(scaled)
1592 <        
1593 <        
1594 <         ! calc the wrapped real coordinates from the wrapped scaled
1595 <         ! coordinates
1596 <        
1597 <         d = matmul(Hmat,scaled)
1598 <        
1599 <      else
1600 <         ! calc the scaled coordinates.
1601 <        
1602 <         do i = 1, 3
1603 <            scaled(i) = d(i) * HmatInv(i,i)
1604 <            
1605 <            ! wrap the scaled coordinates
1606 <            
1607 <            scaled(i) = scaled(i) - anint(scaled(i))
1083 <            
1084 <            ! calc the wrapped real coordinates from the wrapped scaled
1085 <            ! coordinates
1086 <            
1087 <            d(i) = scaled(i)*Hmat(i,i)
1088 <         enddo
1089 <      endif
1090 <      
1091 <   endif
1092 <  
1093 <   r_sq = dot_product(d,d)
1094 <  
1095 < end subroutine get_interatomic_vector
1096 <
1097 < subroutine zero_work_arrays()
1098 <  
1541 >
1542 >  subroutine get_interatomic_vector(q_i, q_j, d, r_sq)
1543 >
1544 >    real (kind = dp), dimension(3) :: q_i
1545 >    real (kind = dp), dimension(3) :: q_j
1546 >    real ( kind = dp ), intent(out) :: r_sq
1547 >    real( kind = dp ) :: d(3), scaled(3)
1548 >    integer i
1549 >
1550 >    d(1) = q_j(1) - q_i(1)
1551 >    d(2) = q_j(2) - q_i(2)
1552 >    d(3) = q_j(3) - q_i(3)
1553 >
1554 >    ! Wrap back into periodic box if necessary
1555 >    if ( SIM_uses_PBC ) then
1556 >
1557 >       if( .not.boxIsOrthorhombic ) then
1558 >          ! calc the scaled coordinates.
1559 >          ! scaled = matmul(HmatInv, d)
1560 >
1561 >          scaled(1) = HmatInv(1,1)*d(1) + HmatInv(1,2)*d(2) + HmatInv(1,3)*d(3)
1562 >          scaled(2) = HmatInv(2,1)*d(1) + HmatInv(2,2)*d(2) + HmatInv(2,3)*d(3)
1563 >          scaled(3) = HmatInv(3,1)*d(1) + HmatInv(3,2)*d(2) + HmatInv(3,3)*d(3)
1564 >          
1565 >          ! wrap the scaled coordinates
1566 >
1567 >          scaled(1) = scaled(1) - anint(scaled(1), kind=dp)
1568 >          scaled(2) = scaled(2) - anint(scaled(2), kind=dp)
1569 >          scaled(3) = scaled(3) - anint(scaled(3), kind=dp)
1570 >
1571 >          ! calc the wrapped real coordinates from the wrapped scaled
1572 >          ! coordinates
1573 >          ! d = matmul(Hmat,scaled)
1574 >          d(1)= Hmat(1,1)*scaled(1) + Hmat(1,2)*scaled(2) + Hmat(1,3)*scaled(3)
1575 >          d(2)= Hmat(2,1)*scaled(1) + Hmat(2,2)*scaled(2) + Hmat(2,3)*scaled(3)
1576 >          d(3)= Hmat(3,1)*scaled(1) + Hmat(3,2)*scaled(2) + Hmat(3,3)*scaled(3)
1577 >
1578 >       else
1579 >          ! calc the scaled coordinates.
1580 >
1581 >          scaled(1) = d(1) * HmatInv(1,1)
1582 >          scaled(2) = d(2) * HmatInv(2,2)
1583 >          scaled(3) = d(3) * HmatInv(3,3)
1584 >          
1585 >          ! wrap the scaled coordinates
1586 >          
1587 >          scaled(1) = scaled(1) - anint(scaled(1), kind=dp)
1588 >          scaled(2) = scaled(2) - anint(scaled(2), kind=dp)
1589 >          scaled(3) = scaled(3) - anint(scaled(3), kind=dp)
1590 >
1591 >          ! calc the wrapped real coordinates from the wrapped scaled
1592 >          ! coordinates
1593 >
1594 >          d(1) = scaled(1)*Hmat(1,1)
1595 >          d(2) = scaled(2)*Hmat(2,2)
1596 >          d(3) = scaled(3)*Hmat(3,3)
1597 >
1598 >       endif
1599 >
1600 >    endif
1601 >
1602 >    r_sq = d(1)*d(1) + d(2)*d(2) + d(3)*d(3)
1603 >
1604 >  end subroutine get_interatomic_vector
1605 >
1606 >  subroutine zero_work_arrays()
1607 >
1608   #ifdef IS_MPI
1100  
1101   q_Row = 0.0_dp
1102   q_Col = 0.0_dp
1609  
1610 <   q_group_Row = 0.0_dp
1611 <   q_group_Col = 0.0_dp  
1612 <  
1613 <   eFrame_Row = 0.0_dp
1614 <   eFrame_Col = 0.0_dp
1615 <  
1616 <   A_Row = 0.0_dp
1617 <   A_Col = 0.0_dp
1618 <  
1619 <   f_Row = 0.0_dp
1620 <   f_Col = 0.0_dp
1621 <   f_Temp = 0.0_dp
1622 <  
1623 <   t_Row = 0.0_dp
1624 <   t_Col = 0.0_dp
1625 <   t_Temp = 0.0_dp
1626 <  
1627 <   pot_Row = 0.0_dp
1628 <   pot_Col = 0.0_dp
1629 <   pot_Temp = 0.0_dp
1630 <  
1631 <   rf_Row = 0.0_dp
1632 <   rf_Col = 0.0_dp
1633 <   rf_Temp = 0.0_dp
1128 <  
1610 >    q_Row = 0.0_dp
1611 >    q_Col = 0.0_dp
1612 >
1613 >    q_group_Row = 0.0_dp
1614 >    q_group_Col = 0.0_dp  
1615 >
1616 >    eFrame_Row = 0.0_dp
1617 >    eFrame_Col = 0.0_dp
1618 >
1619 >    A_Row = 0.0_dp
1620 >    A_Col = 0.0_dp
1621 >
1622 >    f_Row = 0.0_dp
1623 >    f_Col = 0.0_dp
1624 >    f_Temp = 0.0_dp
1625 >
1626 >    t_Row = 0.0_dp
1627 >    t_Col = 0.0_dp
1628 >    t_Temp = 0.0_dp
1629 >
1630 >    pot_Row = 0.0_dp
1631 >    pot_Col = 0.0_dp
1632 >    pot_Temp = 0.0_dp
1633 >
1634   #endif
1635 <
1636 <   if (FF_uses_EAM .and. SIM_uses_EAM) then
1637 <      call clean_EAM()
1638 <   endif
1639 <  
1640 <   rf = 0.0_dp
1641 <   tau_Temp = 0.0_dp
1642 <   virial_Temp = 0.0_dp
1643 < end subroutine zero_work_arrays
1644 <
1645 < function skipThisPair(atom1, atom2) result(skip_it)
1646 <   integer, intent(in) :: atom1
1647 <   integer, intent(in), optional :: atom2
1648 <   logical :: skip_it
1649 <   integer :: unique_id_1, unique_id_2
1650 <   integer :: me_i,me_j
1651 <   integer :: i
1652 <  
1653 <   skip_it = .false.
1654 <  
1655 <   !! there are a number of reasons to skip a pair or a particle
1656 <   !! mostly we do this to exclude atoms who are involved in short
1657 <   !! range interactions (bonds, bends, torsions), but we also need
1153 <   !! to exclude some overcounted interactions that result from
1154 <   !! the parallel decomposition
1155 <  
1635 >
1636 >    if (FF_uses_EAM .and. SIM_uses_EAM) then
1637 >       call clean_EAM()
1638 >    endif
1639 >
1640 >  end subroutine zero_work_arrays
1641 >
1642 >  function skipThisPair(atom1, atom2) result(skip_it)
1643 >    integer, intent(in) :: atom1
1644 >    integer, intent(in), optional :: atom2
1645 >    logical :: skip_it
1646 >    integer :: unique_id_1, unique_id_2
1647 >    integer :: me_i,me_j
1648 >    integer :: i
1649 >
1650 >    skip_it = .false.
1651 >
1652 >    !! there are a number of reasons to skip a pair or a particle
1653 >    !! mostly we do this to exclude atoms who are involved in short
1654 >    !! range interactions (bonds, bends, torsions), but we also need
1655 >    !! to exclude some overcounted interactions that result from
1656 >    !! the parallel decomposition
1657 >
1658   #ifdef IS_MPI
1659 <   !! in MPI, we have to look up the unique IDs for each atom
1660 <   unique_id_1 = AtomRowToGlobal(atom1)
1659 >    !! in MPI, we have to look up the unique IDs for each atom
1660 >    unique_id_1 = AtomRowToGlobal(atom1)
1661   #else
1662 <   !! in the normal loop, the atom numbers are unique
1663 <   unique_id_1 = atom1
1662 >    !! in the normal loop, the atom numbers are unique
1663 >    unique_id_1 = atom1
1664   #endif
1665 <  
1666 <   !! We were called with only one atom, so just check the global exclude
1667 <   !! list for this atom
1668 <   if (.not. present(atom2)) then
1669 <      do i = 1, nExcludes_global
1670 <         if (excludesGlobal(i) == unique_id_1) then
1671 <            skip_it = .true.
1672 <            return
1673 <         end if
1674 <      end do
1675 <      return
1676 <   end if
1677 <  
1665 >
1666 >    !! We were called with only one atom, so just check the global exclude
1667 >    !! list for this atom
1668 >    if (.not. present(atom2)) then
1669 >       do i = 1, nExcludes_global
1670 >          if (excludesGlobal(i) == unique_id_1) then
1671 >             skip_it = .true.
1672 >             return
1673 >          end if
1674 >       end do
1675 >       return
1676 >    end if
1677 >
1678   #ifdef IS_MPI
1679 <   unique_id_2 = AtomColToGlobal(atom2)
1679 >    unique_id_2 = AtomColToGlobal(atom2)
1680   #else
1681 <   unique_id_2 = atom2
1681 >    unique_id_2 = atom2
1682   #endif
1683 <  
1683 >
1684   #ifdef IS_MPI
1685 <   !! this situation should only arise in MPI simulations
1686 <   if (unique_id_1 == unique_id_2) then
1687 <      skip_it = .true.
1688 <      return
1689 <   end if
1690 <  
1691 <   !! this prevents us from doing the pair on multiple processors
1692 <   if (unique_id_1 < unique_id_2) then
1693 <      if (mod(unique_id_1 + unique_id_2,2) == 0) then
1694 <         skip_it = .true.
1695 <         return
1696 <      endif
1697 <   else                
1698 <      if (mod(unique_id_1 + unique_id_2,2) == 1) then
1699 <         skip_it = .true.
1700 <         return
1701 <      endif
1702 <   endif
1685 >    !! this situation should only arise in MPI simulations
1686 >    if (unique_id_1 == unique_id_2) then
1687 >       skip_it = .true.
1688 >       return
1689 >    end if
1690 >
1691 >    !! this prevents us from doing the pair on multiple processors
1692 >    if (unique_id_1 < unique_id_2) then
1693 >       if (mod(unique_id_1 + unique_id_2,2) == 0) then
1694 >          skip_it = .true.
1695 >          return
1696 >       endif
1697 >    else                
1698 >       if (mod(unique_id_1 + unique_id_2,2) == 1) then
1699 >          skip_it = .true.
1700 >          return
1701 >       endif
1702 >    endif
1703   #endif
1704 <  
1705 <   !! the rest of these situations can happen in all simulations:
1706 <   do i = 1, nExcludes_global      
1707 <      if ((excludesGlobal(i) == unique_id_1) .or. &
1708 <           (excludesGlobal(i) == unique_id_2)) then
1709 <         skip_it = .true.
1710 <         return
1711 <      endif
1712 <   enddo
1713 <  
1714 <   do i = 1, nSkipsForAtom(atom1)
1715 <      if (skipsForAtom(atom1, i) .eq. unique_id_2) then
1716 <         skip_it = .true.
1717 <         return
1718 <      endif
1719 <   end do
1720 <  
1721 <   return
1722 < end function skipThisPair
1723 <
1724 < function FF_UsesDirectionalAtoms() result(doesit)
1725 <   logical :: doesit
1726 <   doesit = FF_uses_DirectionalAtoms .or. FF_uses_Dipoles .or. &
1727 <        FF_uses_Sticky .or. FF_uses_GayBerne .or. FF_uses_Shapes
1728 < end function FF_UsesDirectionalAtoms
1729 <
1730 < function FF_RequiresPrepairCalc() result(doesit)
1731 <   logical :: doesit
1732 <   doesit = FF_uses_EAM
1733 < end function FF_RequiresPrepairCalc
1734 <
1233 < function FF_RequiresPostpairCalc() result(doesit)
1234 <   logical :: doesit
1235 <   doesit = FF_uses_RF
1236 < end function FF_RequiresPostpairCalc
1237 <
1704 >
1705 >    !! the rest of these situations can happen in all simulations:
1706 >    do i = 1, nExcludes_global      
1707 >       if ((excludesGlobal(i) == unique_id_1) .or. &
1708 >            (excludesGlobal(i) == unique_id_2)) then
1709 >          skip_it = .true.
1710 >          return
1711 >       endif
1712 >    enddo
1713 >
1714 >    do i = 1, nSkipsForAtom(atom1)
1715 >       if (skipsForAtom(atom1, i) .eq. unique_id_2) then
1716 >          skip_it = .true.
1717 >          return
1718 >       endif
1719 >    end do
1720 >
1721 >    return
1722 >  end function skipThisPair
1723 >
1724 >  function FF_UsesDirectionalAtoms() result(doesit)
1725 >    logical :: doesit
1726 >    doesit = FF_uses_DirectionalAtoms
1727 >  end function FF_UsesDirectionalAtoms
1728 >
1729 >  function FF_RequiresPrepairCalc() result(doesit)
1730 >    logical :: doesit
1731 >    doesit = FF_uses_EAM .or. FF_uses_SC &
1732 >         .or. FF_uses_MEAM
1733 >  end function FF_RequiresPrepairCalc
1734 >
1735   #ifdef PROFILE
1736 < function getforcetime() result(totalforcetime)
1737 <   real(kind=dp) :: totalforcetime
1738 <   totalforcetime = forcetime
1739 < end function getforcetime
1736 >  function getforcetime() result(totalforcetime)
1737 >    real(kind=dp) :: totalforcetime
1738 >    totalforcetime = forcetime
1739 >  end function getforcetime
1740   #endif
1244
1245 !! This cleans componets of force arrays belonging only to fortran
1741  
1742 < subroutine add_stress_tensor(dpair, fpair)
1248 <  
1249 <   real( kind = dp ), dimension(3), intent(in) :: dpair, fpair
1250 <  
1251 <   ! because the d vector is the rj - ri vector, and
1252 <   ! because fx, fy, fz are the force on atom i, we need a
1253 <   ! negative sign here:  
1254 <  
1255 <   tau_Temp(1) = tau_Temp(1) - dpair(1) * fpair(1)
1256 <   tau_Temp(2) = tau_Temp(2) - dpair(1) * fpair(2)
1257 <   tau_Temp(3) = tau_Temp(3) - dpair(1) * fpair(3)
1258 <   tau_Temp(4) = tau_Temp(4) - dpair(2) * fpair(1)
1259 <   tau_Temp(5) = tau_Temp(5) - dpair(2) * fpair(2)
1260 <   tau_Temp(6) = tau_Temp(6) - dpair(2) * fpair(3)
1261 <   tau_Temp(7) = tau_Temp(7) - dpair(3) * fpair(1)
1262 <   tau_Temp(8) = tau_Temp(8) - dpair(3) * fpair(2)
1263 <   tau_Temp(9) = tau_Temp(9) - dpair(3) * fpair(3)
1264 <  
1265 <   virial_Temp = virial_Temp + &
1266 <        (tau_Temp(1) + tau_Temp(5) + tau_Temp(9))
1267 <  
1268 < end subroutine add_stress_tensor
1269 <
1270 < end module doForces
1742 >  !! This cleans componets of force arrays belonging only to fortran
1743  
1744 < !! Interfaces for C programs to module....
1744 >  subroutine add_stress_tensor(dpair, fpair, tau)
1745  
1746 < subroutine initFortranFF(use_RF_c, thisStat)
1747 <    use doForces, ONLY: init_FF
1276 <    logical, intent(in) :: use_RF_c
1746 >    real( kind = dp ), dimension(3), intent(in) :: dpair, fpair
1747 >    real( kind = dp ), dimension(9), intent(inout) :: tau
1748  
1749 <    integer, intent(out) :: thisStat  
1750 <    call init_FF(use_RF_c, thisStat)
1749 >    ! because the d vector is the rj - ri vector, and
1750 >    ! because fx, fy, fz are the force on atom i, we need a
1751 >    ! negative sign here:  
1752  
1753 < end subroutine initFortranFF
1753 >    tau(1) = tau(1) - dpair(1) * fpair(1)
1754 >    tau(2) = tau(2) - dpair(1) * fpair(2)
1755 >    tau(3) = tau(3) - dpair(1) * fpair(3)
1756 >    tau(4) = tau(4) - dpair(2) * fpair(1)
1757 >    tau(5) = tau(5) - dpair(2) * fpair(2)
1758 >    tau(6) = tau(6) - dpair(2) * fpair(3)
1759 >    tau(7) = tau(7) - dpair(3) * fpair(1)
1760 >    tau(8) = tau(8) - dpair(3) * fpair(2)
1761 >    tau(9) = tau(9) - dpair(3) * fpair(3)
1762  
1763 <  subroutine doForceloop(q, q_group, A, eFrame, f, t, tau, pot, &
1284 <       do_pot_c, do_stress_c, error)
1285 <      
1286 <       use definitions, ONLY: dp
1287 <       use simulation
1288 <       use doForces, ONLY: do_force_loop
1289 <    !! Position array provided by C, dimensioned by getNlocal
1290 <    real ( kind = dp ), dimension(3, nLocal) :: q
1291 <    !! molecular center-of-mass position array
1292 <    real ( kind = dp ), dimension(3, nGroups) :: q_group
1293 <    !! Rotation Matrix for each long range particle in simulation.
1294 <    real( kind = dp), dimension(9, nLocal) :: A    
1295 <    !! Unit vectors for dipoles (lab frame)
1296 <    real( kind = dp ), dimension(9,nLocal) :: eFrame
1297 <    !! Force array provided by C, dimensioned by getNlocal
1298 <    real ( kind = dp ), dimension(3,nLocal) :: f
1299 <    !! Torsion array provided by C, dimensioned by getNlocal
1300 <    real( kind = dp ), dimension(3,nLocal) :: t    
1763 >  end subroutine add_stress_tensor
1764  
1765 <    !! Stress Tensor
1303 <    real( kind = dp), dimension(9) :: tau  
1304 <    real ( kind = dp ) :: pot
1305 <    logical ( kind = 2) :: do_pot_c, do_stress_c
1306 <    integer :: error
1307 <    
1308 <    call do_force_loop(q, q_group, A, eFrame, f, t, tau, pot, &
1309 <       do_pot_c, do_stress_c, error)
1310 <      
1311 < end subroutine doForceloop
1765 > end module doForces

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines