! #include "copyright.h" #include "../include/assert.fh" #include "../include/dprec.fh" !+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ !+ Main routine for Amber's traditional minimization methods !----------------------------------------------------------------------- ! --- RUNMIN --- !----------------------------------------------------------------------- ! Various combinations of steepest descent and conjugate gradient ! optimizations. subroutine runmin(xx,ix,ih,ipairs,x,fg,w,ib,jb,conp, & winv,igrp,skips,ene,carrms, qsetup) use fastwt use constants, only : zero, one, TEN_TO_MINUS5, TEN_TO_MINUS6 use qmmm_module, only : qmmm_nml, qmmm_struct, qmmm_mpi, qm2_struct use poisson_boltzmann, only: outwat, oution use bintraj, only: end_binary_frame use file_io_dat #if defined( MPI ) use evb_data, only: evb_frc #endif /* MPI */ #ifdef MPI /* SOFT CORE */ use softcore, only: extra_atoms, sc_ener, sc_ener_tmp, sc_dvdl, sc_tot_dvdl, & sc_tot_dvdl_partner, ifsc, sc_mix_sum, sc_print_energies,& sc_dvdl_ee, sc_tot_dvdl_ee, sc_tot_dvdl_partner_ee, & ti_ene_cnt, nsoftcore, nsoftcore_partner, nmixed, nsc #endif use state implicit none #ifdef MPI # include "parallel.h" # ifdef MPI_DOUBLE_PRECISION # undef MPI_DOUBLE_PRECISION # endif include 'mpif.h' integer ierr #ifdef CRAY_PVP # define MPI_DOUBLE_PRECISION MPI_REAL8 #endif integer ist(MPI_STATUS_SIZE), partner #endif #include "md.h" #include "box.h" #include "memory.h" #include "nmr.h" #include "extra.h" #include "ew_cntrl.h" #include "../../AmberTools/src/pbsa/pb_md.h" ! ------ passed in variables -------------------- _REAL_ xx(*) integer ix(*), ipairs(*) character(len=4) ih(*) _REAL_ x(*),fg(*),w(*) integer ib(*),jb(*) _REAL_ conp(*),winv(*) integer igrp(*) logical skips(*) type(state_rec) :: ene _REAL_ carrms logical :: qsetup ! ------ External Functions ----------------- _REAL_ ddot ! ------ local variables -------------------- logical skip,newstr,steep,belly _REAL_ betax, ddspln, dfpr, dxst, dxsth _REAL_ f, fch, finit, fmin, fnq, fold, gamma, gamden _REAL_ ginit, gmin, gnew, gspln, gsqrd, sbound, step _REAL_ stepch, stmin, sum, work type(state_rec) :: ecopy _REAL_ shkh(1) _REAL_ rms,fndfp,swork,fdmax integer i,n,nr,nrx,nct,ndfp,nstcyc,nitp,ier integer mstcyc,linmin,iterrs integer irsdx,irsdg,iginit,ixopt,igopt,iterc,n_force_calls integer iterfm,nfopt,nfbeg,iretry integer l_frcti integer imes logical itdump,ixdump logical loutfm logical qspatial integer ti_nrp, ti_nrp_partner integer ti_nct(2), ti_nct_partner(2) integer ti_nct_tot integer ti_ib, ti_jb _REAL_ ti_pot_ene, ti_pot_ene_partner _REAL_ ti_sum, ti_sum_partner _REAL_, dimension(:), allocatable :: frcti integer maxlin,mxfcon,kstcyc ! comments on parameters are guesses by SRB Sep 2003 parameter ( maxlin = 10 ) ! maximum number of line searches ? parameter ( mxfcon = 4 ) ! maximum force con ? parameter ( kstcyc = 4 ) ! number of starting cycles ? _REAL_ crits, dxstm, dfpred parameter ( dxstm = TEN_TO_MINUS5 ) ! ? parameter ( crits = TEN_TO_MINUS6 ) ! ? parameter ( dfpred = ONE ) ! ? in kcal/mol logical :: do_list_update=.false. logical :: lout ! Zero the state type as done in runmd() ene = null_state_rec ! ----- EVALUATE SOME CONSTANTS ----- if (imin /= 5 .and. master) call amopen(7,mdinfo,'U','F','W') fmin = 0.0d0 nr = nrp n = 3*nr belly = ibelly > 0 ier = 0 nct = 0 if (ntc == 2) nct = nbonh if (ntc == 3) nct = nbonh + nbona ndfp = n-nct if(belly) ndfp = 3*natbel-nct ntnb = 1 fndfp = ndfp fnq = sqrt(fndfp) #ifdef MPI ! correct fnq so that it is the same on both nodes if (icfe .ne. 0 .and. ifsc .ne. 0) then if( master ) then partner = ieor(masterrank,1) ti_nrp = nmixed + nsoftcore + nsoftcore_partner ti_nct(:) = 0 if (ntc >= 2) then do i = 1,nbonh ti_ib = ix(iibh + i)/3 + 1 ti_jb = ix(ijbh + i)/3 + 1 if (nsc(ti_ib) .eq. 0 .and. nsc(ti_jb) .eq. 0) then ti_nct(1) = ti_nct(1) + 1 else ti_nct(2) = ti_nct(2) + 1 end if end do end if call mpi_sendrecv( ti_nct, 2, MPI_INTEGER, partner, 5, & ti_nct_partner, 2, MPI_INTEGER, partner, 5, & commmaster, ist, ierr ) ti_nct_tot = ti_nct(1) + ti_nct(2) + ti_nct_partner(2) ndfp = 3*ti_nrp-ti_nct_tot fndfp = ndfp fnq = sqrt(fndfp) end if ! if( numtasks>0 ) then ! call mpi_bcast(fnq,1,MPI_DOUBLE_PRECISION,0,commsander,ierr) ! end if end if #endif rms = 0.0d0 skip = .false. newstr = .false. ! determine the number of steepest descent steps steep = .false. nstcyc = 0 mstcyc = kstcyc if(ntmin == 2) mstcyc = maxcyc if(ntmin == 1) mstcyc = ncyc if(ntmin > 0) steep = .true. ! Ben Roberts: Enable writing to a trajectory file if ! requested. ! Number of atoms to write to the trajectory ! If NTWPRT.NE.0, only print the atoms up to this value nrx = n if (ntwprt > 0) nrx = ntwprt*3 ! Trajectory format loutfm = (ioutfm <= 0) #ifdef MPI if( icfe /= 0 ) then allocate( frcti( n+3*extra_atoms ), stat = ier ) REQUIRE( ier == 0 ) end if #endif fold = 0.0d0 dxst = dx0 linmin = 0 gmin = 1.d0 if (iscale > 0) n = n + iscale ! ----- PARTITION THE WORKING ARRAY ----- irsdx = n irsdg = irsdx+n iginit = irsdg+n ixopt = iginit+n igopt = ixopt+n ! ----- SET SOME PARAMETERS TO BEGIN THE CALCULATION ----- iterc = 0 n_force_calls = 0 iterfm = iterc ! ----- LET THE INITIAL SEARCH DIRECTION BE MINUS THE GRADIENT ! VECTOR. ITERRS GIVES THE ITERATION NUMBER OF THE MOST ! RECENT RESTART , BUT IS SET TO ZERO WHEN STEEPEST DESCENT ! DIRECTION IS USED ----- !==================================================================== ! (Here is the beginning of a big loop:) 20 continue !==================================================================== ! ----- GATHER THE SUBMOLECULES INTO THE BOX ----- n_force_calls = n_force_calls + 1 if (mod(n_force_calls,nsnb) == 0) ntnb = 1 if(ntnb == 1 .and. n_force_calls > 1) steep = .true. !==================================================================== ! ----- CALCULATE THE FORCE AND ENERGY ----- ! ----- APPLY SHAKE TO CONSTRAIN BONDS IF NECESSARY ----- !==================================================================== if(ntc /= 1) then fg(1:n) = x(1:n) nitp = 0 qspatial=.false. call shake(nr,nbonh,nbona,0,ib,jb,igrp,winv,conp,skips, & fg,x,nitp,belly,ix(iifstwt),ix(noshake), & shkh,qspatial) call quick3(fg,x,ix(iifstwr),natom,nres,ix(i02)) if(nitp <= 0) then ! shake failed ier = 135 goto 290 end if end if ! reset pb-related flags if(master)then if ( igb == 10 .or. ipb /= 0 ) then !if ( mod(n_force_calls,npbgrid) == 0 .and. n_force_calls /= maxcyc ) pbgrid = .true. !if ( mod(n_force_calls,ntpr) == 0 .or. n_force_calls == maxcyc ) pbprint = .true. !if ( mod(n_force_calls,nsnbr) == 0 .and. n_force_calls /= maxcyc ) ntnbr = 1 !if ( mod(n_force_calls,nsnba) == 0 .and. n_force_calls /= maxcyc ) ntnba = 1 if ( mod(n_force_calls,npbgrid) == 0 ) pbgrid = .true. if ( mod(n_force_calls,ntpr) == 0 ) pbprint = .true. if ( mod(n_force_calls,nsnbr) == 0 ) ntnbr = 1 if ( mod(n_force_calls,nsnba) == 0 ) ntnba = 1 npbstep = n_force_calls !write(6,*) 'inside runmin', npbgrid, ntpr, nsnbr, nsnba !write(6,*) 'inside runmin', n_force_calls, pbgrid, pbprint, ntnbr, ntnba end if endif iprint = 0 if (n_force_calls == maxcyc .or. n_force_calls == 1) iprint=1 lout = .false. if (mod(n_force_calls,ntpr) == 0 .or. n_force_calls == 1) lout = .true. ! Ben Roberts: Switches to enable writing of restart files ! and trajectory coordinates for this particular step. Will ! set the flags to "false" unless the number of steps is right. ! Also requires that ntwr and ntwx are non-zero. ! Restart file ixdump = .false. if (ntwr /= 0 .and. mod(n_force_calls,ntwr) == 0) ixdump = .true. ! Trajectory ! DRR - Dont write traj during post-processing. itdump = .false. if (ntwx /= 0 .and. mod(n_force_calls,ntwx) == 0 .and. imin /= 5) itdump = .true. irespa = n_force_calls call force(xx,ix,ih,ipairs,x,fg,ene,ene%vir, & xx(l96), xx(l97), xx(l98),xx(l99),qsetup, do_list_update,n_force_calls) #ifdef MPI ! If softcore potentials are used, collect their dvdl contribution from the nodes ! this is disabled as long as multi-CPU TI minimizations dont work if ( ifsc /= 0 ) then !call mpi_reduce(sc_dvdl, sc_tot_dvdl, 1, MPI_DOUBLE_PRECISION, MPI_SUM, 0, commsander, ierr) !call mpi_reduce(sc_dvdl_ee, sc_tot_dvdl_ee, 1, MPI_DOUBLE_PRECISION, MPI_SUM, 0, commsander, ierr) sc_tot_dvdl = sc_dvdl sc_dvdl=0.0d0 ! zero for next step sc_tot_dvdl_ee = sc_dvdl_ee sc_dvdl_ee=0.0d0 ! zero for next step !call mpi_reduce(sc_ener, sc_ener_tmp, ti_ene_cnt, MPI_DOUBLE_PRECISION, MPI_SUM, 0, commsander, ierr) !sc_ener(1:ti_ene_cnt) = sc_ener_tmp(1:ti_ene_cnt) end if if ( icfe /= 0 )then ! ---free energies using thermodynamic integration (icfe /= 0) ! --- first, send the forces and energy to your partner: if( master ) then partner = ieor(masterrank,1) call mpi_sendrecv( fg, n, MPI_DOUBLE_PRECISION, partner, 5, & frcti, n+3*extra_atoms, MPI_DOUBLE_PRECISION, partner, 5, & commmaster, ist, ierr ) call mpi_sendrecv( ene, state_rec_len, MPI_DOUBLE_PRECISION, partner, 5, & ecopy, state_rec_len, MPI_DOUBLE_PRECISION, partner, 5, & commmaster, ist, ierr) ! exchange sc-dvdl contributions between masters call mpi_sendrecv( sc_tot_dvdl, 1, MPI_DOUBLE_PRECISION, partner, 5, & sc_tot_dvdl_partner, 1, MPI_DOUBLE_PRECISION, partner, 5, & commmaster, ist, ierr ) call mpi_sendrecv( sc_tot_dvdl_ee, 1, MPI_DOUBLE_PRECISION, partner, 5, & sc_tot_dvdl_partner_ee, 1, MPI_DOUBLE_PRECISION, partner, 5, & commmaster, ist, ierr ) if( masterrank==0 ) then call mix_frcti(frcti,ecopy,fg,ene,n,clambda,klambda) else call mix_frcti(fg,ene,frcti,ecopy,n,clambda,klambda) end if ti_pot_ene = sc_ener(12) call mpi_sendrecv( ti_pot_ene, 1, MPI_DOUBLE_PRECISION, partner, 5, & ti_pot_ene_partner, 1, MPI_DOUBLE_PRECISION, partner, 5, & commmaster, ist, ierr ) end if ! if( numtasks>0 ) then ! call mpi_bcast(fg,n,MPI_DOUBLE_PRECISION,0,commsander,ierr) ! call mpi_bcast(ene,51,MPI_DOUBLE_PRECISION,0,commsander,ierr) ! end if end if #endif !f = ene(23) f = ene%pot%tot #if defined( MPI ) if( ievb /= 0 ) f = evb_frc%evb_nrg #endif /* MPI */ ntnb = 0 sum = ddot(n,fg,1,fg,1) #ifdef MPI /* SOFT CORE */ if (ifsc == 1) then ! This stabilizes softcore minimizations by including the energy of ! the decoupled system. This is needed since the forces for the decoupled ! system are non-zero, so the corresponding energy should be included here. f = f + ti_pot_ene + ti_pot_ene_partner ! The forces have already been scaled by lambda, so rather than ! repeating the scaling with the sum we add the forces for the ! system including the decoupled partner if (master) then ti_sum = 0.d0 do i = 1, nr if (nsc(i) .ne. 0) then ti_sum = ti_sum + fg(3*(i-1)+1) * fg(3*(i-1)+1) ti_sum = ti_sum + fg(3*(i-1)+2) * fg(3*(i-1)+2) ti_sum = ti_sum + fg(3*(i-1)+3) * fg(3*(i-1)+3) end if end do call mpi_sendrecv( ti_sum, 1, MPI_DOUBLE_PRECISION, partner, 5, & ti_sum_partner, 1, MPI_DOUBLE_PRECISION, partner, 5, & commmaster, ist, ierr ) sum = sum + ti_sum_partner end if ! Now only called to handle round off error call sc_mix_sum(sum) ! if( numtasks>0 ) then ! call mpi_bcast(sum,1,MPI_DOUBLE_PRECISION,0,commsander,ierr) ! end if end if #endif if ((ifcap == 2 .or. ifcap == 5) .and. n_force_calls == 1) then ! HG added this to account for waters not being considered if ifcap == 2,5 fnq = fnq * fnq fnq = fnq - 3 * (outwat + oution) fnq = sqrt(fnq) end if rms = sqrt(sum)/fnq ! ----- PRINT THE INTERMEDIATE RESULTS ----- ! x+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++x ! | Output EVB data | ! x+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++x #ifdef MPI if( ievb /= 0 ) call out_evb ( n_force_calls) #endif ! DAN ROE: modified so that during traj post-proc. only final results ! are printed. if (lout .and. imin /= 5) then call report_min_progress( n_force_calls, rms, fg, & ene, ih(m04), xx(l15) ) ! ih(m04) = atom names, xx(l15) = charges #ifdef MPI /* SOFT CORE */ if (ifsc /= 0) call sc_print_energies(6, sc_ener) if (ifsc /= 0) call sc_print_energies(7, sc_ener) #endif !--- Print QM/MM Mulliken Charges if needed --- if (qmmm_nml%ifqnt) then if (qmmm_nml%printcharges .and. qmmm_mpi%commqmmm_master) then call qm2_print_charges(n_force_calls,qmmm_nml%dftb_chg,qmmm_struct%nquant_nlink, & qm2_struct%scf_mchg,qmmm_struct%iqm_atomic_numbers) end if if (qmmm_nml%printdipole /= 0) then call qmmm_dipole(x,xx(Lmass),ix(i02),ih(m02),nres) end if end if !------ end if !==================================================================== ! ----- DO SOME STEEPEST STEPS BEFORE ENTERING THE CONJUGATE ! GRADIENT METHOD ----- !==================================================================== if (steep) then nstcyc = nstcyc+1 if (nstcyc <= mstcyc) then if(dxst <= crits) dxst = dxstm dxst = dxst/2.0d0 if(f < fold) dxst = dxst*2.4d0 dxsth = dxst/sqrt(sum) if(nstcyc <= 1 .or. f <= fmin) then fmin = f nfopt = n_force_calls w(ixopt+1:ixopt+n) = x(1:n) w(igopt+1:igopt+n) = -fg(1:n) end if ! ----- CHECK FOR CONVERGENCE ----- if (rms <= drms) then goto 300 end if if (n_force_calls >= maxcyc) then ier = 131 goto 290 end if fold = f x(1:n) = x(1:n)+dxsth*fg(1:n) ! Ben Roberts: Write the restart file and/or the trajectory ! frame. if (master) then ! Write a restart file if appropriate if (ixdump) call minrit(n_force_calls,nrp,ntxo,x) ! Write to the trajectory if appropriate if (itdump) then call corpac(x,1,nrx,MDCRD_UNIT,loutfm) if (ioutfm > 0) call end_binary_frame(MDCRD_UNIT) end if end if goto 20 else ! (arrive here when finished with this ! set of steepest descent cycles) steep = .false. newstr = .true. nstcyc = 0 mstcyc = kstcyc end if end if !==================================================================== ! ----- START OF CONJUGATE GRADIENT STEPS ----- !==================================================================== fg(1:n) = -fg(1:n) if (.not. newstr .and. n_force_calls > 1) goto 82 70 continue w(1:n) = -fg(1:n) iterrs = 0 if(newstr) iterc = 0 if(iterc > 0) goto 140 82 continue gnew = ddot(n,w,1,fg,1) ! ----- STORE THE VALUES OF X, F AND G, IF THEY ARE THE BEST THAT ! HAVE BEEN CALCULATED SO FAR. TEST FOR CONVERGENCE ---- if (newstr .or. n_force_calls == 1) then ! artificially set fch less than zero to simplify the code. ! fmin will be properly initialized in the nested if statement below fch = -ONE else fch = f - fmin end if if (fch <= ZERO) then if (fch < ZERO .or. gnew/gmin >= -ONE) then fmin = f gsqrd = sum nfopt = n_force_calls w(ixopt+1:ixopt+n) = x(1:n) w(igopt+1:igopt+n) = fg(1:n) end if if (rms <= drms) then goto 300 end if end if ! ----- TEST IF THE VALUE OF MAXCYC ALLOWS ANOTHER CALL OF FUNCT --- if (n_force_calls >= maxcyc) then ier = 131 goto 290 end if if (.not.newstr .and. n_force_calls > 1) goto 180 ! ------ This section is executed at the beginning of a conjugate ! gradient set of minimization steps. ! ----- SET DFPR TO DX0*GSQRD. DFPR IS THE REDUCTION IN THE FUNCTION ! VALUE. STMIN IS USUALLY THE STEP-LENGTH OF THE MOST RECENT ! LINE SEARCH THAT GIVES THE LEAST VALUE OF F ----- ! --- dac change, 10/91: return to original idea of trying to ! go downhill by the absolute amount, DFPRED (which defaults ! to 1 kcal/mol, see data statement above). This can eliminate ! very bad initial conjugate gradient steps. dfpr = dfpred stmin = dfpred/gsqrd newstr = .false. !==================================================================== ! ----- Begin the main conjugate gradient iteration ----- !==================================================================== 140 iterc = iterc+1 finit = f ginit = 0.0d0 w(iginit+1:iginit+n) = fg(1:n) ginit = ddot(n,w,1,fg,1) if(ginit >= 0.0d0) goto 260 gmin = ginit sbound = -1.0d0 nfbeg = n_force_calls iretry = -1 stepch = min(stmin,abs(dfpr/ginit)) stmin = dxstm 160 step = stmin+stepch dxst = step swork = 0.0d0 do i=1,n x(i) = w(ixopt+i)+stepch*w(i) swork = max(swork,abs(x(i)-w(ixopt+i))) end do ! Ben Roberts: Write the restart file and/or the trajectory frame ! if appropriate. if(swork > 0.0d0) then if (master) then ! Write a restart file if appropriate if (ixdump) call minrit(n_force_calls,nrp,ntxo,x) ! Write to the trajectory if appropriate if (itdump) then call corpac(x,1,nrx,MDCRD_UNIT,loutfm) if (ioutfm > 0) call end_binary_frame(MDCRD_UNIT) end if end if goto 20 end if ! "work = swork" may not be needed - wont hurt. -gls work = swork ! ----- TERMINATE THE LINE SEARCH IF STEPCH IS EFFECTIVELY ZERO --- if(n_force_calls > nfbeg+1 .or. abs(gmin/ginit) > 0.2d0) then if (master) write(6,370) steep = .true. linmin = linmin+1 end if goto 270 180 work = (fch+fch)/stepch-gnew-gmin ddspln = (gnew-gmin)/stepch if (n_force_calls > nfopt) then sbound = step else if(gmin*gnew <= 0.0d0) sbound = stmin stmin = step gmin = gnew stepch = -stepch end if if(fch /= 0.0d0) ddspln = ddspln+(work+work)/stepch ! ----- TEST FOR CONVERGENCE OF THE LINE SEARCH, BUT FORCE ATLEAST ! TWO STEPS TO BE TAKEN IN ORDER NOT TO LOSE QUADRATIC ! TERMINATION ----- if(gmin == 0.0d0) goto 270 if(n_force_calls <= nfbeg+1) goto 200 if(abs(gmin/ginit) <= 0.2d0) goto 270 ! ----- APPLY THE TEST THAT DEPENDS ON THE PARAMETER MAXLIN ----- 190 if(n_force_calls < nfopt+maxlin) goto 200 ! ----- POSSIBLE NON BONDED UPDATE. MAKE A RESTART ----- if (master) write(6,370) steep = .true. linmin = linmin+1 goto 270 200 stepch = 0.5d0*(sbound-stmin) if(sbound < -0.5d0) stepch = 9.0d0*stmin gspln = gmin+stepch*ddspln if(gmin*gspln < 0.0d0) stepch = stepch*gmin/(gmin-gspln) goto 160 ! ----- CALCULATE THE VALUE OF BETAX IN THE NEW DIRECTION ----- 210 sum = ddot(n,fg,1,w(iginit+1),1) betax = (gsqrd-sum)/(gmin-ginit) ! ----- TEST THAT THE NEW SEARCH DIRECTION CAN BE MADE DOWNHILL. ! IF NOT THEN TRY TO IMPROVE THE ACCURACY OF THE LINE ! SEARCH ----- if(abs(betax*gmin) <= 0.2d0*gsqrd) goto 220 iretry = iretry+1 if(iretry <= 0) goto 190 220 if (f < finit) iterfm = iterc if (iterc >= iterfm+mxfcon) then if (master) write(6,370) steep = .true. linmin = linmin+1 goto 270 end if dfpr = stmin*ginit ! ----- BRANCH IF A RESTART PROCEDURE IS REQUIRED DUE TO THE ! ITERATION NUMBER OR DUE TO THE SCALAR PRODUCT OF ! CONSECUTIVE GRADIENTS ----- if(iretry > 0) goto 70 if(iterrs == 0) goto 240 if(iterc-iterrs >= n) goto 240 if(abs(sum) >= 0.2d0*gsqrd) goto 240 ! ----- CALCULATE GAMMA IN THE NEW SEARCH DIRECTION. GAMDEN IS ! SET BY THE RESTART PROCEDURE ----- gamma = ddot(n,fg,1,w(irsdg+1),1) sum = ddot(n,fg,1,w(irsdx+1),1) gamma = gamma/gamden ! ----- RESTART IF THE NEW SEARCH DIRECTION IS NOT SUFFICIENTLY ! DOWNHILL ---- if(abs(betax*gmin+gamma*sum) >= 0.2d0*gsqrd) goto 240 ! ----- CALCULATE THE NEW SEARCH DIRECTION ----- w(1:n) = -fg(1:n)+betax*w(1:n)+gamma*w(irsdx+1:irsdx+n) ! --- cycle back for more conjugate gradient steps: goto 140 ! ----- APPLY THE RESTART PROCEDURE ----- 240 gamden = gmin-ginit do i=1,n w(irsdx+i) = w(i) w(irsdg+i) = fg(i)-w(iginit+i) w(i) = -fg(i)+betax*w(i) end do iterrs = iterc goto 140 ! ----- SET IER TO INDICATE THAT THE SEARCH DIRECTION IS UPHILL --- 260 continue steep = .true. if (master) write(6,370) linmin = linmin+1 ! ----- ENSURE THAT F, X AND G ARE OPTIMAL ----- 270 continue if (n_force_calls /= nfopt) then f = fmin x(1:n) = w(ixopt+1:ixopt+n) fg(1:n) = w(igopt+1:igopt+n) end if if (linmin > 4) then ier = 133 goto 290 end if ! Ben Roberts: Write the restart file and/or the trajectory frame ! if appropriate. if (steep) then if (master) then ! Write a restart file if appropriate if (ixdump) call minrit(n_force_calls,nrp,ntxo,x) ! Write to the trajectory if appropriate if (itdump) then call corpac(x,1,nrx,MDCRD_UNIT,loutfm) if (ioutfm > 0) call end_binary_frame(MDCRD_UNIT) end if end if goto 20 end if if (ier == 0) goto 210 290 continue ! The unconverged minimization terminates if (master) then select case ( ier ) case ( 131 ) write(6,'(//,a)') ' Maximum number of minimization cycles reached.' case ( 133 ) write(6,'(/5x,a)') '***** REPEATED LINMIN FAILURE *****' write(6,'(/5x,a)') '***** SEE http://ambermd.org/Questions/linmin.html FOR MORE INFO *****' case ( 135 ) write(6,'(/5x,a)') '***** ERROR: SHAKE FAILURE *****' case default ! invalid ier ASSERT( .false. ) end select end if 300 continue ! The converged minimization terminates ! do a final force call with iprint=1 to get proper nmr restaint printout: iprint=1 call force(xx,ix,ih,ipairs,x,fg,ene,ene%vir, & xx(l96), xx(l97), xx(l98),xx(l99),qsetup, do_list_update,n_force_calls) #ifdef MPI if ( icfe /= 0 )then sc_tot_dvdl = sc_dvdl sc_dvdl=0.0d0 ! zero for next step sc_tot_dvdl_ee = sc_dvdl_ee sc_dvdl_ee=0.0d0 ! zero for next step ! ---free energies using thermodynamic integration (icfe /= 0) ! --- first, send the forces and energy to your partner: if( master ) then partner = ieor(masterrank,1) call mpi_sendrecv( fg, n, MPI_DOUBLE_PRECISION, partner, 5, & frcti, n+3*extra_atoms, MPI_DOUBLE_PRECISION, partner, 5, & commmaster, ist, ierr ) call mpi_sendrecv( ene, state_rec_len, MPI_DOUBLE_PRECISION, partner, 5, & ecopy, state_rec_len, MPI_DOUBLE_PRECISION, partner, 5, & commmaster, ist, ierr) ! exchange sc-dvdl contributions between masters call mpi_sendrecv( sc_tot_dvdl, 1, MPI_DOUBLE_PRECISION, partner, 5, & sc_tot_dvdl_partner, 1, MPI_DOUBLE_PRECISION, partner, 5, & commmaster, ist, ierr ) call mpi_sendrecv( sc_tot_dvdl_ee, 1, MPI_DOUBLE_PRECISION, partner, 5, & sc_tot_dvdl_partner_ee, 1, MPI_DOUBLE_PRECISION, partner, 5, & commmaster, ist, ierr ) if( masterrank==0 ) then call mix_frcti(frcti,ecopy,fg,ene,n,clambda,klambda) else call mix_frcti(fg,ene,frcti,ecopy,n,clambda,klambda) end if ti_pot_ene = sc_ener(12) call mpi_sendrecv( ti_pot_ene, 1, MPI_DOUBLE_PRECISION, partner, 5, & ti_pot_ene_partner, 1, MPI_DOUBLE_PRECISION, partner, 5, & commmaster, ist, ierr ) end if end if #endif ! ----- WRITE THE FINAL RESULTS ----- ! Ben Roberts:Write restart file here, instead of (as before) ! in sander.f, so n_force_calls may be provided. if (master) then ! Write a restart file if appropriate call minrit(n_force_calls,nrp,ntxo,x) ! Write to the trajectory if appropriate if (itdump) then call corpac(x,1,nrx,MDCRD_UNIT,loutfm) if (ioutfm > 0) call end_binary_frame(MDCRD_UNIT) end if end if call report_min_results( n_force_calls, rms, x, & fg, ene, ih(m04), xx, ix, ih ) ! ih(m04) = atom names carrms = rms #ifdef MPI /* SOFT CORE */ if (ifsc /= 0) call sc_print_energies(6, sc_ener) if (ifsc /= 0) call sc_print_energies(7, sc_ener) #endif #ifdef MPI if( ievb /= 0 ) then call evb_dealloc endif #endif if( icfe /= 0 ) then deallocate( frcti, stat=ier ) REQUIRE( ier == 0 ) end if return 370 format(/4x,' ... RESTARTED DUE TO LINMIN FAILURE ...') end subroutine runmin