2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
5 * Copyright (c) 2001-2004, The GROMACS development team.
6 * Copyright (c) 2011,2012,2013,2014,2015,2016, by the GROMACS development team, led by
7 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
8 * and including many others, as listed in the AUTHORS file in the
9 * top-level source directory and at http://www.gromacs.org.
11 * GROMACS is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU Lesser General Public License
13 * as published by the Free Software Foundation; either version 2.1
14 * of the License, or (at your option) any later version.
16 * GROMACS is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * Lesser General Public License for more details.
21 * You should have received a copy of the GNU Lesser General Public
22 * License along with GROMACS; if not, see
23 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
24 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
26 * If you want to redistribute modifications to GROMACS, please
27 * consider that scientific software is very special. Version
28 * control is crucial - bugs must be traceable. We will be happy to
29 * consider code for inclusion in the official distribution, but
30 * derived work must not be called official GROMACS. Details are found
31 * in the README & COPYING files - if they are missing, get the
32 * official version at http://www.gromacs.org.
34 * To help us fund GROMACS development, we humbly ask that you cite
35 * the research papers on the package. Check out http://www.gromacs.org.
48 #include "gromacs/domdec/domdec.h"
49 #include "gromacs/gmxlib/network.h"
50 #include "gromacs/math/units.h"
51 #include "gromacs/math/vec.h"
52 #include "gromacs/mdlib/main.h"
53 #include "gromacs/mdtypes/commrec.h"
54 #include "gromacs/mdtypes/inputrec.h"
55 #include "gromacs/mdtypes/md_enums.h"
56 #include "gromacs/random/threefry.h"
57 #include "gromacs/random/uniformintdistribution.h"
58 #include "gromacs/random/uniformrealdistribution.h"
59 #include "gromacs/utility/fatalerror.h"
60 #include "gromacs/utility/pleasecite.h"
61 #include "gromacs/utility/smalloc.h"
64 #define PROBABILITYCUTOFF 100
65 /* we don't bother evaluating if events are more rare than exp(-100) = 3.7x10^-44 */
67 //! Rank in the multisimulaiton
68 #define MSRANK(ms, nodeid) (nodeid)
71 ereTEMP, ereLAMBDA, ereENDSINGLE, ereTL, ereNR
73 const char *erename[ereNR] = { "temperature", "lambda", "end_single_marker", "temperature and lambda"};
74 /* end_single_marker merely notes the end of single variable replica exchange. All types higher than
75 it are multiple replica exchange methods */
76 /* Eventually, should add 'pressure', 'temperature and pressure', 'lambda_and_pressure', 'temperature_lambda_pressure'?;
77 Let's wait until we feel better about the pressure control methods giving exact ensembles. Right now, we assume constant pressure */
79 typedef struct gmx_repl_ex
81 int repl; /* replica ID */
82 int nrepl; /* total number of replica */
83 real temp; /* temperature */
84 int type; /* replica exchange type from ere enum */
85 real **q; /* quantity, e.g. temperature or lambda; first index is ere, second index is replica ID */
86 gmx_bool bNPT; /* use constant pressure and temperature */
87 real *pres; /* replica pressures */
88 int *ind; /* replica indices */
89 int *allswaps; /* used for keeping track of all the replica swaps */
90 int nst; /* replica exchange interval (number of steps) */
91 int nex; /* number of exchanges per interval */
92 int seed; /* random seed */
93 int nattempt[2]; /* number of even and odd replica change attempts */
94 real *prob_sum; /* sum of probabilities */
95 int **nmoves; /* number of moves between replicas i and j */
96 int *nexchange; /* i-th element of the array is the number of exchanges between replica i-1 and i */
98 /* these are helper arrays for replica exchange; allocated here so they
99 don't have to be allocated each time */
107 /* helper arrays to hold the quantities that are exchanged */
116 static gmx_bool repl_quantity(const gmx_multisim_t *ms,
117 struct gmx_repl_ex *re, int ere, real q)
123 snew(qall, ms->nsim);
125 gmx_sum_sim(ms->nsim, qall, ms);
128 for (s = 1; s < ms->nsim; s++)
130 if (qall[s] != qall[0])
138 /* Set the replica exchange type and quantities */
141 snew(re->q[ere], re->nrepl);
142 for (s = 0; s < ms->nsim; s++)
144 re->q[ere][s] = qall[s];
151 gmx_repl_ex_t init_replica_exchange(FILE *fplog,
152 const gmx_multisim_t *ms,
153 const t_state *state,
154 const t_inputrec *ir,
155 int nst, int nex, int init_seed)
159 struct gmx_repl_ex *re;
161 gmx_bool bLambda = FALSE;
163 fprintf(fplog, "\nInitializing Replica Exchange\n");
165 if (ms == NULL || ms->nsim == 1)
167 gmx_fatal(FARGS, "Nothing to exchange with only one replica, maybe you forgot to set the -multi option of mdrun?");
169 if (!EI_DYNAMICS(ir->eI))
171 gmx_fatal(FARGS, "Replica exchange is only supported by dynamical simulations");
172 /* Note that PAR(cr) is defined by cr->nnodes > 1, which is
173 * distinct from MULTISIM(cr). A multi-simulation only runs
174 * with real MPI parallelism, but this does not imply PAR(cr)
177 * Since we are using a dynamical integrator, the only
178 * decomposition is DD, so PAR(cr) and DOMAINDECOMP(cr) are
179 * synonymous. The only way for cr->nnodes > 1 to be true is
180 * if we are using DD. */
186 re->nrepl = ms->nsim;
187 snew(re->q, ereENDSINGLE);
189 fprintf(fplog, "Repl There are %d replicas:\n", re->nrepl);
191 check_multi_int(fplog, ms, state->natoms, "the number of atoms", FALSE);
192 check_multi_int(fplog, ms, ir->eI, "the integrator", FALSE);
193 check_multi_int64(fplog, ms, ir->init_step+ir->nsteps, "init_step+nsteps", FALSE);
194 check_multi_int64(fplog, ms, (ir->init_step+nst-1)/nst,
195 "first exchange step: init_step/-replex", FALSE);
196 check_multi_int(fplog, ms, ir->etc, "the temperature coupling", FALSE);
197 check_multi_int(fplog, ms, ir->opts.ngtc,
198 "the number of temperature coupling groups", FALSE);
199 check_multi_int(fplog, ms, ir->epc, "the pressure coupling", FALSE);
200 check_multi_int(fplog, ms, ir->efep, "free energy", FALSE);
201 check_multi_int(fplog, ms, ir->fepvals->n_lambda, "number of lambda states", FALSE);
203 re->temp = ir->opts.ref_t[0];
204 for (i = 1; (i < ir->opts.ngtc); i++)
206 if (ir->opts.ref_t[i] != re->temp)
208 fprintf(fplog, "\nWARNING: The temperatures of the different temperature coupling groups are not identical\n\n");
209 fprintf(stderr, "\nWARNING: The temperatures of the different temperature coupling groups are not identical\n\n");
214 bTemp = repl_quantity(ms, re, ereTEMP, re->temp);
215 if (ir->efep != efepNO)
217 bLambda = repl_quantity(ms, re, ereLAMBDA, (real)ir->fepvals->init_fep_state);
219 if (re->type == -1) /* nothing was assigned */
221 gmx_fatal(FARGS, "The properties of the %d systems are all the same, there is nothing to exchange", re->nrepl);
223 if (bLambda && bTemp)
230 please_cite(fplog, "Sugita1999a");
231 if (ir->epc != epcNO)
234 fprintf(fplog, "Repl Using Constant Pressure REMD.\n");
235 please_cite(fplog, "Okabe2001a");
237 if (ir->etc == etcBERENDSEN)
239 gmx_fatal(FARGS, "REMD with the %s thermostat does not produce correct potential energy distributions, consider using the %s thermostat instead",
240 ETCOUPLTYPE(ir->etc), ETCOUPLTYPE(etcVRESCALE));
245 if (ir->fepvals->delta_lambda != 0) /* check this? */
247 gmx_fatal(FARGS, "delta_lambda is not zero");
252 snew(re->pres, re->nrepl);
253 if (ir->epct == epctSURFACETENSION)
255 pres = ir->ref_p[ZZ][ZZ];
261 for (i = 0; i < DIM; i++)
263 if (ir->compress[i][i] != 0)
265 pres += ir->ref_p[i][i];
271 re->pres[re->repl] = pres;
272 gmx_sum_sim(re->nrepl, re->pres, ms);
275 /* Make an index for increasing replica order */
276 /* only makes sense if one or the other is varying, not both!
277 if both are varying, we trust the order the person gave. */
278 snew(re->ind, re->nrepl);
279 for (i = 0; i < re->nrepl; i++)
284 if (re->type < ereENDSINGLE)
287 for (i = 0; i < re->nrepl; i++)
289 for (j = i+1; j < re->nrepl; j++)
291 if (re->q[re->type][re->ind[j]] < re->q[re->type][re->ind[i]])
293 /* Unordered replicas are supposed to work, but there
294 * is still an issues somewhere.
295 * Note that at this point still re->ind[i]=i.
297 gmx_fatal(FARGS, "Replicas with indices %d < %d have %ss %g > %g, please order your replicas on increasing %s",
300 re->q[re->type][i], re->q[re->type][j],
304 re->ind[i] = re->ind[j];
307 else if (re->q[re->type][re->ind[j]] == re->q[re->type][re->ind[i]])
309 gmx_fatal(FARGS, "Two replicas have identical %ss", erename[re->type]);
315 /* keep track of all the swaps, starting with the initial placement. */
316 snew(re->allswaps, re->nrepl);
317 for (i = 0; i < re->nrepl; i++)
319 re->allswaps[i] = re->ind[i];
325 fprintf(fplog, "\nReplica exchange in temperature\n");
326 for (i = 0; i < re->nrepl; i++)
328 fprintf(fplog, " %5.1f", re->q[re->type][re->ind[i]]);
330 fprintf(fplog, "\n");
333 fprintf(fplog, "\nReplica exchange in lambda\n");
334 for (i = 0; i < re->nrepl; i++)
336 fprintf(fplog, " %3d", (int)re->q[re->type][re->ind[i]]);
338 fprintf(fplog, "\n");
341 fprintf(fplog, "\nReplica exchange in temperature and lambda state\n");
342 for (i = 0; i < re->nrepl; i++)
344 fprintf(fplog, " %5.1f", re->q[ereTEMP][re->ind[i]]);
346 fprintf(fplog, "\n");
347 for (i = 0; i < re->nrepl; i++)
349 fprintf(fplog, " %5d", (int)re->q[ereLAMBDA][re->ind[i]]);
351 fprintf(fplog, "\n");
354 gmx_incons("Unknown replica exchange quantity");
358 fprintf(fplog, "\nRepl p");
359 for (i = 0; i < re->nrepl; i++)
361 fprintf(fplog, " %5.2f", re->pres[re->ind[i]]);
364 for (i = 0; i < re->nrepl; i++)
366 if ((i > 0) && (re->pres[re->ind[i]] < re->pres[re->ind[i-1]]))
368 fprintf(fplog, "\nWARNING: The reference pressures decrease with increasing temperatures\n\n");
369 fprintf(stderr, "\nWARNING: The reference pressures decrease with increasing temperatures\n\n");
378 re->seed = static_cast<int>(gmx::makeRandomSeed());
384 gmx_sumi_sim(1, &(re->seed), ms);
388 re->seed = init_seed;
390 fprintf(fplog, "\nReplica exchange interval: %d\n", re->nst);
391 fprintf(fplog, "\nReplica random seed: %d\n", re->seed);
396 snew(re->prob_sum, re->nrepl);
397 snew(re->nexchange, re->nrepl);
398 snew(re->nmoves, re->nrepl);
399 for (i = 0; i < re->nrepl; i++)
401 snew(re->nmoves[i], re->nrepl);
403 fprintf(fplog, "Replica exchange information below: ex and x = exchange, pr = probability\n");
405 /* generate space for the helper functions so we don't have to snew each time */
407 snew(re->destinations, re->nrepl);
408 snew(re->incycle, re->nrepl);
409 snew(re->tmpswap, re->nrepl);
410 snew(re->cyclic, re->nrepl);
411 snew(re->order, re->nrepl);
412 for (i = 0; i < re->nrepl; i++)
414 snew(re->cyclic[i], re->nrepl+1);
415 snew(re->order[i], re->nrepl);
417 /* allocate space for the functions storing the data for the replicas */
418 /* not all of these arrays needed in all cases, but they don't take
419 up much space, since the max size is nrepl**2 */
420 snew(re->prob, re->nrepl);
421 snew(re->bEx, re->nrepl);
422 snew(re->beta, re->nrepl);
423 snew(re->Vol, re->nrepl);
424 snew(re->Epot, re->nrepl);
425 snew(re->de, re->nrepl);
426 for (i = 0; i < re->nrepl; i++)
428 snew(re->de[i], re->nrepl);
434 static void exchange_reals(const gmx_multisim_t gmx_unused *ms, int gmx_unused b, real *v, int n)
444 MPI_Sendrecv(v, n*sizeof(real),MPI_BYTE,MSRANK(ms,b),0,
445 buf,n*sizeof(real),MPI_BYTE,MSRANK(ms,b),0,
446 ms->mpi_comm_masters,MPI_STATUS_IGNORE);
451 MPI_Isend(v, n*sizeof(real), MPI_BYTE, MSRANK(ms, b), 0,
452 ms->mpi_comm_masters, &mpi_req);
453 MPI_Recv(buf, n*sizeof(real), MPI_BYTE, MSRANK(ms, b), 0,
454 ms->mpi_comm_masters, MPI_STATUS_IGNORE);
455 MPI_Wait(&mpi_req, MPI_STATUS_IGNORE);
458 for (i = 0; i < n; i++)
467 static void exchange_doubles(const gmx_multisim_t gmx_unused *ms, int gmx_unused b, double *v, int n)
477 MPI_Sendrecv(v, n*sizeof(double),MPI_BYTE,MSRANK(ms,b),0,
478 buf,n*sizeof(double),MPI_BYTE,MSRANK(ms,b),0,
479 ms->mpi_comm_masters,MPI_STATUS_IGNORE);
484 MPI_Isend(v, n*sizeof(double), MPI_BYTE, MSRANK(ms, b), 0,
485 ms->mpi_comm_masters, &mpi_req);
486 MPI_Recv(buf, n*sizeof(double), MPI_BYTE, MSRANK(ms, b), 0,
487 ms->mpi_comm_masters, MPI_STATUS_IGNORE);
488 MPI_Wait(&mpi_req, MPI_STATUS_IGNORE);
491 for (i = 0; i < n; i++)
499 static void exchange_rvecs(const gmx_multisim_t gmx_unused *ms, int gmx_unused b, rvec *v, int n)
509 MPI_Sendrecv(v[0], n*sizeof(rvec),MPI_BYTE,MSRANK(ms,b),0,
510 buf[0],n*sizeof(rvec),MPI_BYTE,MSRANK(ms,b),0,
511 ms->mpi_comm_masters,MPI_STATUS_IGNORE);
516 MPI_Isend(v[0], n*sizeof(rvec), MPI_BYTE, MSRANK(ms, b), 0,
517 ms->mpi_comm_masters, &mpi_req);
518 MPI_Recv(buf[0], n*sizeof(rvec), MPI_BYTE, MSRANK(ms, b), 0,
519 ms->mpi_comm_masters, MPI_STATUS_IGNORE);
520 MPI_Wait(&mpi_req, MPI_STATUS_IGNORE);
523 for (i = 0; i < n; i++)
525 copy_rvec(buf[i], v[i]);
531 static void exchange_state(const gmx_multisim_t *ms, int b, t_state *state)
533 /* When t_state changes, this code should be updated. */
535 ngtc = state->ngtc * state->nhchainlength;
536 nnhpres = state->nnhpres* state->nhchainlength;
537 exchange_rvecs(ms, b, state->box, DIM);
538 exchange_rvecs(ms, b, state->box_rel, DIM);
539 exchange_rvecs(ms, b, state->boxv, DIM);
540 exchange_reals(ms, b, &(state->veta), 1);
541 exchange_reals(ms, b, &(state->vol0), 1);
542 exchange_rvecs(ms, b, state->svir_prev, DIM);
543 exchange_rvecs(ms, b, state->fvir_prev, DIM);
544 exchange_rvecs(ms, b, state->pres_prev, DIM);
545 exchange_doubles(ms, b, state->nosehoover_xi.data(), ngtc);
546 exchange_doubles(ms, b, state->nosehoover_vxi.data(), ngtc);
547 exchange_doubles(ms, b, state->nhpres_xi.data(), nnhpres);
548 exchange_doubles(ms, b, state->nhpres_vxi.data(), nnhpres);
549 exchange_doubles(ms, b, state->therm_integral.data(), state->ngtc);
550 exchange_rvecs(ms, b, as_rvec_array(state->x.data()), state->natoms);
551 exchange_rvecs(ms, b, as_rvec_array(state->v.data()), state->natoms);
554 static void copy_state_serial(const t_state *src, t_state *dest)
558 /* Currently the local state is always a pointer to the global
559 * in serial, so we should never end up here.
560 * TODO: Implement a (trivial) t_state copy once converted to C++.
562 GMX_RELEASE_ASSERT(false, "State copying is currently not implemented in replica exchange");
566 static void scale_velocities(t_state *state, real fac)
570 if (as_rvec_array(state->v.data()))
572 for (i = 0; i < state->natoms; i++)
574 svmul(fac, state->v[i], state->v[i]);
579 static void print_transition_matrix(FILE *fplog, int n, int **nmoves, int *nattempt)
584 ntot = nattempt[0] + nattempt[1];
585 fprintf(fplog, "\n");
586 fprintf(fplog, "Repl");
587 for (i = 0; i < n; i++)
589 fprintf(fplog, " "); /* put the title closer to the center */
591 fprintf(fplog, "Empirical Transition Matrix\n");
593 fprintf(fplog, "Repl");
594 for (i = 0; i < n; i++)
596 fprintf(fplog, "%8d", (i+1));
598 fprintf(fplog, "\n");
600 for (i = 0; i < n; i++)
602 fprintf(fplog, "Repl");
603 for (j = 0; j < n; j++)
606 if (nmoves[i][j] > 0)
608 Tprint = nmoves[i][j]/(2.0*ntot);
610 fprintf(fplog, "%8.4f", Tprint);
612 fprintf(fplog, "%3d\n", i);
616 static void print_ind(FILE *fplog, const char *leg, int n, int *ind, gmx_bool *bEx)
620 fprintf(fplog, "Repl %2s %2d", leg, ind[0]);
621 for (i = 1; i < n; i++)
623 fprintf(fplog, " %c %2d", (bEx != 0 && bEx[i]) ? 'x' : ' ', ind[i]);
625 fprintf(fplog, "\n");
628 static void print_allswitchind(FILE *fplog, int n, int *pind, int *allswaps, int *tmpswap)
632 for (i = 0; i < n; i++)
634 tmpswap[i] = allswaps[i];
636 for (i = 0; i < n; i++)
638 allswaps[i] = tmpswap[pind[i]];
641 fprintf(fplog, "\nAccepted Exchanges: ");
642 for (i = 0; i < n; i++)
644 fprintf(fplog, "%d ", pind[i]);
646 fprintf(fplog, "\n");
648 /* the "Order After Exchange" is the state label corresponding to the configuration that
649 started in state listed in order, i.e.
654 configuration starting in simulation 3 is now in simulation 0,
655 configuration starting in simulation 0 is now in simulation 1,
656 configuration starting in simulation 1 is now in simulation 2,
657 configuration starting in simulation 2 is now in simulation 3
659 fprintf(fplog, "Order After Exchange: ");
660 for (i = 0; i < n; i++)
662 fprintf(fplog, "%d ", allswaps[i]);
664 fprintf(fplog, "\n\n");
667 static void print_prob(FILE *fplog, const char *leg, int n, real *prob)
672 fprintf(fplog, "Repl %2s ", leg);
673 for (i = 1; i < n; i++)
677 sprintf(buf, "%4.2f", prob[i]);
678 fprintf(fplog, " %3s", buf[0] == '1' ? "1.0" : buf+1);
685 fprintf(fplog, "\n");
688 static void print_count(FILE *fplog, const char *leg, int n, int *count)
692 fprintf(fplog, "Repl %2s ", leg);
693 for (i = 1; i < n; i++)
695 fprintf(fplog, " %4d", count[i]);
697 fprintf(fplog, "\n");
700 static real calc_delta(FILE *fplog, gmx_bool bPrint, struct gmx_repl_ex *re, int a, int b, int ap, int bp)
703 real ediff, dpV, delta = 0;
704 real *Epot = re->Epot;
707 real *beta = re->beta;
709 /* Two cases; we are permuted and not. In all cases, setting ap = a and bp = b will reduce
710 to the non permuted case */
716 * Okabe et. al. Chem. Phys. Lett. 335 (2001) 435-439
718 ediff = Epot[b] - Epot[a];
719 delta = -(beta[bp] - beta[ap])*ediff;
722 /* two cases: when we are permuted, and not. */
724 ediff = E_new - E_old
725 = [H_b(x_a) + H_a(x_b)] - [H_b(x_b) + H_a(x_a)]
726 = [H_b(x_a) - H_a(x_a)] + [H_a(x_b) - H_b(x_b)]
727 = de[b][a] + de[a][b] */
730 ediff = E_new - E_old
731 = [H_bp(x_a) + H_ap(x_b)] - [H_bp(x_b) + H_ap(x_a)]
732 = [H_bp(x_a) - H_ap(x_a)] + [H_ap(x_b) - H_bp(x_b)]
733 = [H_bp(x_a) - H_a(x_a) + H_a(x_a) - H_ap(x_a)] + [H_ap(x_b) - H_b(x_b) + H_b(x_b) - H_bp(x_b)]
734 = [H_bp(x_a) - H_a(x_a)] - [H_ap(x_a) - H_a(x_a)] + [H_ap(x_b) - H_b(x_b)] - H_bp(x_b) - H_b(x_b)]
735 = (de[bp][a] - de[ap][a]) + (de[ap][b] - de[bp][b]) */
736 /* but, in the current code implementation, we flip configurations, not indices . . .
737 So let's examine that.
738 = [H_b(x_ap) - H_a(x_a)] - [H_a(x_ap) - H_a(x_a)] + [H_a(x_bp) - H_b(x_b)] - H_b(x_bp) - H_b(x_b)]
739 = [H_b(x_ap) - H_a(x_ap)] + [H_a(x_bp) - H_b(x_pb)]
740 = (de[b][ap] - de[a][ap]) + (de[a][bp] - de[b][bp]
741 So, if we exchange b<=> bp and a<=> ap, we return to the same result.
742 So the simple solution is to flip the
743 position of perturbed and original indices in the tests.
746 ediff = (de[bp][a] - de[ap][a]) + (de[ap][b] - de[bp][b]);
747 delta = ediff*beta[a]; /* assume all same temperature in this case */
751 /* delta = reduced E_new - reduced E_old
752 = [beta_b H_b(x_a) + beta_a H_a(x_b)] - [beta_b H_b(x_b) + beta_a H_a(x_a)]
753 = [beta_b H_b(x_a) - beta_a H_a(x_a)] + [beta_a H_a(x_b) - beta_b H_b(x_b)]
754 = [beta_b dH_b(x_a) + beta_b H_a(x_a) - beta_a H_a(x_a)] +
755 [beta_a dH_a(x_b) + beta_a H_b(x_b) - beta_b H_b(x_b)]
756 = [beta_b dH_b(x_a) + [beta_a dH_a(x_b) +
757 beta_b (H_a(x_a) - H_b(x_b)]) - beta_a (H_a(x_a) - H_b(x_b))
758 = beta_b dH_b(x_a) + beta_a dH_a(x_b) - (beta_b - beta_a)(H_b(x_b) - H_a(x_a) */
759 /* delta = beta[b]*de[b][a] + beta[a]*de[a][b] - (beta[b] - beta[a])*(Epot[b] - Epot[a]; */
760 /* permuted (big breath!) */
761 /* delta = reduced E_new - reduced E_old
762 = [beta_bp H_bp(x_a) + beta_ap H_ap(x_b)] - [beta_bp H_bp(x_b) + beta_ap H_ap(x_a)]
763 = [beta_bp H_bp(x_a) - beta_ap H_ap(x_a)] + [beta_ap H_ap(x_b) - beta_bp H_bp(x_b)]
764 = [beta_bp H_bp(x_a) - beta_ap H_ap(x_a)] + [beta_ap H_ap(x_b) - beta_bp H_bp(x_b)]
765 - beta_pb H_a(x_a) + beta_ap H_a(x_a) + beta_pb H_a(x_a) - beta_ap H_a(x_a)
766 - beta_ap H_b(x_b) + beta_bp H_b(x_b) + beta_ap H_b(x_b) - beta_bp H_b(x_b)
767 = [(beta_bp H_bp(x_a) - beta_bp H_a(x_a)) - (beta_ap H_ap(x_a) - beta_ap H_a(x_a))] +
768 [(beta_ap H_ap(x_b) - beta_ap H_b(x_b)) - (beta_bp H_bp(x_b) - beta_bp H_b(x_b))]
769 + beta_pb H_a(x_a) - beta_ap H_a(x_a) + beta_ap H_b(x_b) - beta_bp H_b(x_b)
770 = [beta_bp (H_bp(x_a) - H_a(x_a)) - beta_ap (H_ap(x_a) - H_a(x_a))] +
771 [beta_ap (H_ap(x_b) - H_b(x_b)) - beta_bp (H_bp(x_b) - H_b(x_b))]
772 + beta_pb (H_a(x_a) - H_b(x_b)) - beta_ap (H_a(x_a) - H_b(x_b))
773 = ([beta_bp de[bp][a] - beta_ap de[ap][a]) + beta_ap de[ap][b] - beta_bp de[bp][b])
774 + (beta_pb-beta_ap)(H_a(x_a) - H_b(x_b)) */
775 delta = beta[bp]*(de[bp][a] - de[bp][b]) + beta[ap]*(de[ap][b] - de[ap][a]) - (beta[bp]-beta[ap])*(Epot[b]-Epot[a]);
778 gmx_incons("Unknown replica exchange quantity");
782 fprintf(fplog, "Repl %d <-> %d dE_term = %10.3e (kT)\n", a, b, delta);
786 /* revist the calculation for 5.0. Might be some improvements. */
787 dpV = (beta[ap]*re->pres[ap]-beta[bp]*re->pres[bp])*(Vol[b]-Vol[a])/PRESFAC;
790 fprintf(fplog, " dpV = %10.3e d = %10.3e\n", dpV, delta + dpV);
798 test_for_replica_exchange(FILE *fplog,
799 const gmx_multisim_t *ms,
800 struct gmx_repl_ex *re,
801 gmx_enerdata_t *enerd,
806 int m, i, j, a, b, ap, bp, i0, i1, tmp;
808 gmx_bool bPrint, bMultiEx;
809 gmx_bool *bEx = re->bEx;
810 real *prob = re->prob;
811 int *pind = re->destinations; /* permuted index */
812 gmx_bool bEpot = FALSE;
813 gmx_bool bDLambda = FALSE;
814 gmx_bool bVol = FALSE;
815 gmx::ThreeFry2x64<64> rng(re->seed, gmx::RandomDomain::ReplicaExchange);
816 gmx::UniformRealDistribution<real> uniformRealDist;
817 gmx::UniformIntDistribution<int> uniformNreplDist(0, re->nrepl-1);
819 bMultiEx = (re->nex > 1); /* multiple exchanges at each state */
820 fprintf(fplog, "Replica exchange at step %" GMX_PRId64 " time %.5f\n", step, time);
824 for (i = 0; i < re->nrepl; i++)
829 re->Vol[re->repl] = vol;
831 if ((re->type == ereTEMP || re->type == ereTL))
833 for (i = 0; i < re->nrepl; i++)
838 re->Epot[re->repl] = enerd->term[F_EPOT];
839 /* temperatures of different states*/
840 for (i = 0; i < re->nrepl; i++)
842 re->beta[i] = 1.0/(re->q[ereTEMP][i]*BOLTZ);
847 for (i = 0; i < re->nrepl; i++)
849 re->beta[i] = 1.0/(re->temp*BOLTZ); /* we have a single temperature */
852 if (re->type == ereLAMBDA || re->type == ereTL)
855 /* lambda differences. */
856 /* de[i][j] is the energy of the jth simulation in the ith Hamiltonian
857 minus the energy of the jth simulation in the jth Hamiltonian */
858 for (i = 0; i < re->nrepl; i++)
860 for (j = 0; j < re->nrepl; j++)
865 for (i = 0; i < re->nrepl; i++)
867 re->de[i][re->repl] = (enerd->enerpart_lambda[(int)re->q[ereLAMBDA][i]+1]-enerd->enerpart_lambda[0]);
871 /* now actually do the communication */
874 gmx_sum_sim(re->nrepl, re->Vol, ms);
878 gmx_sum_sim(re->nrepl, re->Epot, ms);
882 for (i = 0; i < re->nrepl; i++)
884 gmx_sum_sim(re->nrepl, re->de[i], ms);
888 /* make a duplicate set of indices for shuffling */
889 for (i = 0; i < re->nrepl; i++)
891 pind[i] = re->ind[i];
894 rng.restart( step, 0 );
898 /* multiple random switch exchange */
902 for (i = 0; i < re->nex + nself; i++)
904 // For now this is superfluous, but just in case we ever add more
905 // calls in different branches it is safer to always reset the distribution.
906 uniformNreplDist.reset();
908 /* randomly select a pair */
909 /* in theory, could reduce this by identifying only which switches had a nonneglibible
910 probability of occurring (log p > -100) and only operate on those switches */
911 /* find out which state it is from, and what label that state currently has. Likely
912 more work that useful. */
913 i0 = uniformNreplDist(rng);
914 i1 = uniformNreplDist(rng);
918 continue; /* self-exchange, back up and do it again */
921 a = re->ind[i0]; /* what are the indices of these states? */
926 bPrint = FALSE; /* too noisy */
927 /* calculate the energy difference */
928 /* if the code changes to flip the STATES, rather than the configurations,
929 use the commented version of the code */
930 /* delta = calc_delta(fplog,bPrint,re,a,b,ap,bp); */
931 delta = calc_delta(fplog, bPrint, re, ap, bp, a, b);
933 /* we actually only use the first space in the prob and bEx array,
934 since there are actually many switches between pairs. */
944 if (delta > PROBABILITYCUTOFF)
950 prob[0] = exp(-delta);
952 // roll a number to determine if accepted. For now it is superfluous to
953 // reset, but just in case we ever add more calls in different branches
954 // it is safer to always reset the distribution.
955 uniformRealDist.reset();
956 bEx[0] = uniformRealDist(rng) < prob[0];
958 re->prob_sum[0] += prob[0];
962 /* swap the states */
968 re->nattempt[0]++; /* keep track of total permutation trials here */
969 print_allswitchind(fplog, re->nrepl, pind, re->allswaps, re->tmpswap);
973 /* standard nearest neighbor replica exchange */
975 m = (step / re->nst) % 2;
976 for (i = 1; i < re->nrepl; i++)
981 bPrint = (re->repl == a || re->repl == b);
984 delta = calc_delta(fplog, bPrint, re, a, b, a, b);
993 if (delta > PROBABILITYCUTOFF)
999 prob[i] = exp(-delta);
1001 // roll a number to determine if accepted. For now it is superfluous to
1002 // reset, but just in case we ever add more calls in different branches
1003 // it is safer to always reset the distribution.
1004 uniformRealDist.reset();
1005 bEx[i] = uniformRealDist(rng) < prob[i];
1007 re->prob_sum[i] += prob[i];
1011 /* swap these two */
1013 pind[i-1] = pind[i];
1015 re->nexchange[i]++; /* statistics for back compatibility */
1024 /* print some statistics */
1025 print_ind(fplog, "ex", re->nrepl, re->ind, bEx);
1026 print_prob(fplog, "pr", re->nrepl, prob);
1027 fprintf(fplog, "\n");
1031 /* record which moves were made and accepted */
1032 for (i = 0; i < re->nrepl; i++)
1034 re->nmoves[re->ind[i]][pind[i]] += 1;
1035 re->nmoves[pind[i]][re->ind[i]] += 1;
1037 fflush(fplog); /* make sure we can see what the last exchange was */
1041 cyclic_decomposition(const int *destinations,
1050 for (i = 0; i < nrepl; i++)
1054 for (i = 0; i < nrepl; i++) /* one cycle for each replica */
1065 for (j = 0; j < nrepl; j++) /* potentially all cycles are part, but we will break first */
1067 p = destinations[p]; /* start permuting */
1075 break; /* we've reached the original element, the cycle is complete, and we marked the end. */
1079 cyclic[i][c] = p; /* each permutation gives a new member of the cycle */
1085 *nswap = maxlen - 1;
1089 for (i = 0; i < nrepl; i++)
1091 fprintf(debug, "Cycle %d:", i);
1092 for (j = 0; j < nrepl; j++)
1094 if (cyclic[i][j] < 0)
1098 fprintf(debug, "%2d", cyclic[i][j]);
1100 fprintf(debug, "\n");
1107 compute_exchange_order(int **cyclic,
1114 for (j = 0; j < maxswap; j++)
1116 for (i = 0; i < nrepl; i++)
1118 if (cyclic[i][j+1] >= 0)
1120 order[cyclic[i][j+1]][j] = cyclic[i][j];
1121 order[cyclic[i][j]][j] = cyclic[i][j+1];
1124 for (i = 0; i < nrepl; i++)
1126 if (order[i][j] < 0)
1128 order[i][j] = i; /* if it's not exchanging, it should stay this round*/
1135 fprintf(debug, "Replica Exchange Order\n");
1136 for (i = 0; i < nrepl; i++)
1138 fprintf(debug, "Replica %d:", i);
1139 for (j = 0; j < maxswap; j++)
1141 if (order[i][j] < 0)
1145 fprintf(debug, "%2d", order[i][j]);
1147 fprintf(debug, "\n");
1154 prepare_to_do_exchange(struct gmx_repl_ex *re,
1155 const int replica_id,
1157 gmx_bool *bThisReplicaExchanged)
1160 /* Hold the cyclic decomposition of the (multiple) replica
1162 gmx_bool bAnyReplicaExchanged = FALSE;
1163 *bThisReplicaExchanged = FALSE;
1165 for (i = 0; i < re->nrepl; i++)
1167 if (re->destinations[i] != re->ind[i])
1169 /* only mark as exchanged if the index has been shuffled */
1170 bAnyReplicaExchanged = TRUE;
1174 if (bAnyReplicaExchanged)
1176 /* reinitialize the placeholder arrays */
1177 for (i = 0; i < re->nrepl; i++)
1179 for (j = 0; j < re->nrepl; j++)
1181 re->cyclic[i][j] = -1;
1182 re->order[i][j] = -1;
1186 /* Identify the cyclic decomposition of the permutation (very
1187 * fast if neighbor replica exchange). */
1188 cyclic_decomposition(re->destinations, re->cyclic, re->incycle, re->nrepl, maxswap);
1190 /* Now translate the decomposition into a replica exchange
1191 * order at each step. */
1192 compute_exchange_order(re->cyclic, re->order, re->nrepl, *maxswap);
1194 /* Did this replica do any exchange at any point? */
1195 for (j = 0; j < *maxswap; j++)
1197 if (replica_id != re->order[replica_id][j])
1199 *bThisReplicaExchanged = TRUE;
1206 gmx_bool replica_exchange(FILE *fplog, const t_commrec *cr, struct gmx_repl_ex *re,
1207 t_state *state, gmx_enerdata_t *enerd,
1208 t_state *state_local, gmx_int64_t step, real time)
1212 int exchange_partner;
1214 /* Number of rounds of exchanges needed to deal with any multiple
1216 /* Where each replica ends up after the exchange attempt(s). */
1217 /* The order in which multiple exchanges will occur. */
1218 gmx_bool bThisReplicaExchanged = FALSE;
1222 replica_id = re->repl;
1223 test_for_replica_exchange(fplog, cr->ms, re, enerd, det(state_local->box), step, time);
1224 prepare_to_do_exchange(re, replica_id, &maxswap, &bThisReplicaExchanged);
1226 /* Do intra-simulation broadcast so all processors belonging to
1227 * each simulation know whether they need to participate in
1228 * collecting the state. Otherwise, they might as well get on with
1229 * the next thing to do. */
1230 if (DOMAINDECOMP(cr))
1233 MPI_Bcast(&bThisReplicaExchanged, sizeof(gmx_bool), MPI_BYTE, MASTERRANK(cr),
1234 cr->mpi_comm_mygroup);
1238 if (bThisReplicaExchanged)
1240 /* Exchange the states */
1241 /* Collect the global state on the master node */
1242 if (DOMAINDECOMP(cr))
1244 dd_collect_state(cr->dd, state_local, state);
1248 copy_state_serial(state_local, state);
1253 /* There will be only one swap cycle with standard replica
1254 * exchange, but there may be multiple swap cycles if we
1255 * allow multiple swaps. */
1257 for (j = 0; j < maxswap; j++)
1259 exchange_partner = re->order[replica_id][j];
1261 if (exchange_partner != replica_id)
1263 /* Exchange the global states between the master nodes */
1266 fprintf(debug, "Exchanging %d with %d\n", replica_id, exchange_partner);
1268 exchange_state(cr->ms, exchange_partner, state);
1271 /* For temperature-type replica exchange, we need to scale
1272 * the velocities. */
1273 if (re->type == ereTEMP || re->type == ereTL)
1275 scale_velocities(state, sqrt(re->q[ereTEMP][replica_id]/re->q[ereTEMP][re->destinations[replica_id]]));
1280 /* With domain decomposition the global state is distributed later */
1281 if (!DOMAINDECOMP(cr))
1283 /* Copy the global state to the local state data structure */
1284 copy_state_serial(state, state_local);
1288 return bThisReplicaExchanged;
1291 void print_replica_exchange_statistics(FILE *fplog, struct gmx_repl_ex *re)
1295 fprintf(fplog, "\nReplica exchange statistics\n");
1299 fprintf(fplog, "Repl %d attempts, %d odd, %d even\n",
1300 re->nattempt[0]+re->nattempt[1], re->nattempt[1], re->nattempt[0]);
1302 fprintf(fplog, "Repl average probabilities:\n");
1303 for (i = 1; i < re->nrepl; i++)
1305 if (re->nattempt[i%2] == 0)
1311 re->prob[i] = re->prob_sum[i]/re->nattempt[i%2];
1314 print_ind(fplog, "", re->nrepl, re->ind, NULL);
1315 print_prob(fplog, "", re->nrepl, re->prob);
1317 fprintf(fplog, "Repl number of exchanges:\n");
1318 print_ind(fplog, "", re->nrepl, re->ind, NULL);
1319 print_count(fplog, "", re->nrepl, re->nexchange);
1321 fprintf(fplog, "Repl average number of exchanges:\n");
1322 for (i = 1; i < re->nrepl; i++)
1324 if (re->nattempt[i%2] == 0)
1330 re->prob[i] = ((real)re->nexchange[i])/re->nattempt[i%2];
1333 print_ind(fplog, "", re->nrepl, re->ind, NULL);
1334 print_prob(fplog, "", re->nrepl, re->prob);
1336 fprintf(fplog, "\n");
1338 /* print the transition matrix */
1339 print_transition_matrix(fplog, re->nrepl, re->nmoves, re->nattempt);