src/gromacs/applied_forces/awh/bias.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2015,2016,2017,2018,2019,2020,2021, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35
  36 /*! \internal \file
  37  * \brief
  38  * Implements the Bias class.
  39  *
  40  * \author Viveca Lindahl
  41  * \author Berk Hess <hess@kth.se>
  42  * \ingroup module_awh
  43  */
  44
  45 #include "gmxpre.h"
  46
  47 #include "bias.h"
  48
  49 #include <cassert>
  50 #include <cinttypes>
  51 #include <cmath>
  52 #include <cstdio>
  53 #include <cstdlib>
  54 #include <cstring>
  55
  56 #include <algorithm>
  57 #include <memory>
  58
  59 #include "gromacs/fileio/gmxfio.h"
  60 #include "gromacs/gmxlib/network.h"
  61 #include "gromacs/math/functions.h"
  62 #include "gromacs/math/utilities.h"
  63 #include "gromacs/mdtypes/awh_history.h"
  64 #include "gromacs/mdtypes/awh_params.h"
  65 #include "gromacs/mdtypes/commrec.h"
  66 #include "gromacs/utility/exceptions.h"
  67 #include "gromacs/utility/gmxassert.h"
  68 #include "gromacs/utility/stringutil.h"
  69
  70 #include "correlationgrid.h"
  71 #include "correlationhistory.h"
  72 #include "pointstate.h"
  73
  74 namespace gmx
  75 {
  76
  77 void Bias::warnForHistogramAnomalies(double t, int64_t step, FILE* fplog)
  78 {
  79     const int maxNumWarningsInCheck = 1;  /* The maximum number of warnings to print per check */
  80     const int maxNumWarningsInRun   = 10; /* The maximum number of warnings to print in a run */
  81
  82     if (fplog == nullptr || numWarningsIssued_ >= maxNumWarningsInRun || state_.inInitialStage()
  83         || !params_.isCheckHistogramForAnomaliesStep(step))
  84     {
  85         return;
  86     }
  87
  88     numWarningsIssued_ +=
  89             state_.warnForHistogramAnomalies(grid_, biasIndex(), t, fplog, maxNumWarningsInCheck);
  90
  91     if (numWarningsIssued_ >= maxNumWarningsInRun)
  92     {
  93         fprintf(fplog, "\nawh%d: suppressing future AWH warnings.\n", biasIndex() + 1);
  94     }
  95 }
  96
  97 void Bias::doSkippedUpdatesForAllPoints()
  98 {
  99     if (params_.skipUpdates())
 100     {
 101         state_.doSkippedUpdatesForAllPoints(params_);
 102     }
 103 }
 104
 105 gmx::ArrayRef<const double> Bias::calcForceAndUpdateBias(const awh_dvec         coordValue,
 106                                                          ArrayRef<const double> neighborLambdaEnergies,
 107                                                          ArrayRef<const double> neighborLambdaDhdl,
 108                                                          double*                awhPotential,
 109                                                          double*                potentialJump,
 110                                                          const t_commrec*       commRecord,
 111                                                          const gmx_multisim_t*  ms,
 112                                                          double                 t,
 113                                                          int64_t                step,
 114                                                          int64_t                seed,
 115                                                          FILE*                  fplog)
 116 {
 117     if (step < 0)
 118     {
 119         GMX_THROW(InvalidInputError(
 120                 "The step number is negative which is not supported by the AWH code."));
 121     }
 122
 123     state_.setCoordValue(grid_, coordValue);
 124
 125     std::vector<double, AlignedAllocator<double>>& probWeightNeighbor = alignedTempWorkSpace_;
 126
 127     /* If the convolved force is needed or this is a sampling step,
 128      * the bias in the current neighborhood needs to be up-to-date
 129      * and the probablity weights need to be calculated.
 130      */
 131     const bool        isSampleCoordStep = params_.isSampleCoordStep(step);
 132     const bool        moveUmbrella      = (isSampleCoordStep || step == 0);
 133     double            convolvedBias     = 0;
 134     const CoordState& coordState        = state_.coordState();
 135
 136     if (params_.convolveForce || moveUmbrella || isSampleCoordStep)
 137     {
 138         if (params_.skipUpdates())
 139         {
 140             state_.doSkippedUpdatesInNeighborhood(params_, grid_);
 141         }
 142         convolvedBias = state_.updateProbabilityWeightsAndConvolvedBias(
 143                 dimParams_, grid_, moveUmbrella ? neighborLambdaEnergies : ArrayRef<const double>{}, &probWeightNeighbor);
 144
 145         if (isSampleCoordStep)
 146         {
 147             updateForceCorrelationGrid(probWeightNeighbor, neighborLambdaDhdl, t);
 148
 149             state_.sampleCoordAndPmf(dimParams_, grid_, probWeightNeighbor, convolvedBias);
 150         }
 151         /* Set the umbrella grid point (for the lambda axis) to the
 152          * current grid point. */
 153         if (params_.convolveForce && grid_.hasLambdaAxis())
 154         {
 155             state_.setUmbrellaGridpointToGridpoint();
 156         }
 157     }
 158
 159     /* Set the bias force and get the potential contribution from this bias.
 160      * The potential jump occurs at different times depending on how
 161      * the force is applied (and how the potential is normalized).
 162      * For the convolved force it happens when the bias is updated,
 163      * for the umbrella when the umbrella is moved.
 164      */
 165     *potentialJump = 0;
 166     double potential;
 167     if (params_.convolveForce)
 168     {
 169         state_.calcConvolvedForce(dimParams_,
 170                                   grid_,
 171                                   probWeightNeighbor,
 172                                   moveUmbrella ? neighborLambdaDhdl : ArrayRef<const double>{},
 173                                   tempForce_,
 174                                   biasForce_);
 175
 176         potential = -convolvedBias * params_.invBeta;
 177     }
 178     else
 179     {
 180         /* Umbrella force */
 181         GMX_RELEASE_ASSERT(state_.points()[coordState.umbrellaGridpoint()].inTargetRegion(),
 182                            "AWH bias grid point for the umbrella reference value is outside of the "
 183                            "target region.");
 184         potential = state_.calcUmbrellaForceAndPotential(
 185                 dimParams_,
 186                 grid_,
 187                 coordState.umbrellaGridpoint(),
 188                 moveUmbrella ? neighborLambdaDhdl : ArrayRef<const double>{},
 189                 biasForce_);
 190
 191         /* Moving the umbrella results in a force correction and
 192          * a new potential. The umbrella center is sampled as often as
 193          * the coordinate so we know the probability weights needed
 194          * for moving the umbrella are up-to-date.
 195          */
 196         if (moveUmbrella)
 197         {
 198             const bool onlySampleUmbrellaGridpoint = false;
 199             double     newPotential                = state_.moveUmbrella(dimParams_,
 200                                                       grid_,
 201                                                       probWeightNeighbor,
 202                                                       neighborLambdaDhdl,
 203                                                       biasForce_,
 204                                                       step,
 205                                                       seed,
 206                                                       params_.biasIndex,
 207                                                       onlySampleUmbrellaGridpoint);
 208             *potentialJump                         = newPotential - potential;
 209         }
 210     }
 211
 212     /* Update the free energy estimates and bias and other history dependent method parameters */
 213     if (params_.isUpdateFreeEnergyStep(step))
 214     {
 215         state_.updateFreeEnergyAndAddSamplesToHistogram(
 216                 dimParams_, grid_, params_, commRecord, ms, t, step, fplog, &updateList_);
 217
 218         if (params_.convolveForce)
 219         {
 220             /* The update results in a potential jump, so we need the new convolved potential. */
 221             double newPotential = -calcConvolvedBias(coordState.coordValue()) * params_.invBeta;
 222             *potentialJump      = newPotential - potential;
 223         }
 224     }
 225     /* If there is a lambda axis it is still controlled using an umbrella even if the force
 226      * is convolved in the other dimensions. */
 227     if (moveUmbrella && params_.convolveForce && grid_.hasLambdaAxis())
 228     {
 229         const bool onlySampleUmbrellaGridpoint = true;
 230         state_.moveUmbrella(dimParams_,
 231                             grid_,
 232                             probWeightNeighbor,
 233                             neighborLambdaDhdl,
 234                             biasForce_,
 235                             step,
 236                             seed,
 237                             params_.biasIndex,
 238                             onlySampleUmbrellaGridpoint);
 239     }
 240
 241     /* Return the potential. */
 242     *awhPotential = potential;
 243
 244     /* Check the sampled histograms and potentially warn user if something is suspicious */
 245     warnForHistogramAnomalies(t, step, fplog);
 246
 247     return biasForce_;
 248 }
 249
 250 /*! \brief
 251  * Count the total number of samples / sample weight over all grid points.
 252  *
 253  * \param[in] pointState  The state of the points in a bias.
 254  * \returns the total sample count.
 255  */
 256 static int64_t countSamples(ArrayRef<const PointState> pointState)
 257 {
 258     double numSamples = 0;
 259     for (const PointState& point : pointState)
 260     {
 261         numSamples += point.weightSumTot();
 262     }
 263
 264     return gmx::roundToInt64(numSamples);
 265 }
 266
 267 /*! \brief
 268  * Check if the state (loaded from checkpoint) and the run are consistent.
 269  *
 270  * When the state and the run setup are inconsistent, an exception is thrown.
 271  *
 272  * \param[in] params  The parameters of the bias.
 273  * \param[in] state   The state of the bias.
 274  */
 275 static void ensureStateAndRunConsistency(const BiasParams& params, const BiasState& state)
 276 {
 277     int64_t numSamples = countSamples(state.points());
 278     int64_t numUpdatesFromSamples =
 279             numSamples / (params.numSamplesUpdateFreeEnergy_ * params.numSharedUpdate);
 280     int64_t numUpdatesExpected = state.histogramSize().numUpdates();
 281     if (numUpdatesFromSamples != numUpdatesExpected)
 282     {
 283         std::string mesg = gmx::formatString(
 284                 "The number of AWH updates in the checkpoint file (%" PRId64
 285                 ") does not match the total number of AWH samples divided by the number of samples "
 286                 "per update for %d sharing AWH bias(es) (%" PRId64 "/%d=%" PRId64 ")",
 287                 numUpdatesExpected,
 288                 params.numSharedUpdate,
 289                 numSamples,
 290                 params.numSamplesUpdateFreeEnergy_ * params.numSharedUpdate,
 291                 numUpdatesFromSamples);
 292         mesg += " Maybe you changed AWH parameters.";
 293         /* Unfortunately we currently do not store the number of simulations
 294          * sharing the bias or the state to checkpoint. But we can hint at
 295          * a mismatch in the number of sharing simulations.
 296          */
 297         if (numUpdatesFromSamples % state.histogramSize().numUpdates() == 0)
 298         {
 299             mesg += gmx::formatString(
 300                     " Or the run you continued from used %" PRId64
 301                     " sharing simulations, whereas you now specified %d sharing simulations.",
 302                     numUpdatesFromSamples / state.histogramSize().numUpdates(),
 303                     params.numSharedUpdate);
 304         }
 305         GMX_THROW(InvalidInputError(mesg));
 306     }
 307 }
 308
 309 void Bias::restoreStateFromHistory(const AwhBiasHistory* biasHistory, const t_commrec* cr)
 310 {
 311     GMX_RELEASE_ASSERT(thisRankDoesIO_ == MASTER(cr),
 312                        "The master rank should do I/O, the other ranks should not");
 313
 314     if (MASTER(cr))
 315     {
 316         GMX_RELEASE_ASSERT(biasHistory != nullptr,
 317                            "On the master rank we need a valid history object to restore from");
 318         state_.restoreFromHistory(*biasHistory, grid_);
 319
 320         /* Ensure that the state is consistent with our current run setup,
 321          * since the user can have changed AWH parameters or the number
 322          * of simulations sharing the bias.
 323          */
 324         ensureStateAndRunConsistency(params_, state_);
 325
 326         if (forceCorrelationGrid_ != nullptr)
 327         {
 328             forceCorrelationGrid_->restoreStateFromHistory(biasHistory->forceCorrelationGrid);
 329         }
 330     }
 331
 332     if (PAR(cr))
 333     {
 334         state_.broadcast(cr);
 335     }
 336 }
 337
 338 void Bias::initHistoryFromState(AwhBiasHistory* biasHistory) const
 339 {
 340     GMX_RELEASE_ASSERT(biasHistory != nullptr, "Need a valid biasHistory");
 341
 342     state_.initHistoryFromState(biasHistory);
 343
 344     if (forceCorrelationGrid_ != nullptr)
 345     {
 346         biasHistory->forceCorrelationGrid = initCorrelationGridHistoryFromState(forceCorrelationGrid());
 347     }
 348 }
 349
 350 void Bias::updateHistory(AwhBiasHistory* biasHistory) const
 351 {
 352     GMX_RELEASE_ASSERT(biasHistory != nullptr, "Need a valid biasHistory");
 353
 354     state_.updateHistory(biasHistory, grid_);
 355
 356     if (forceCorrelationGrid_ != nullptr)
 357     {
 358         updateCorrelationGridHistory(&biasHistory->forceCorrelationGrid, forceCorrelationGrid());
 359     }
 360 }
 361
 362 Bias::Bias(int                            biasIndexInCollection,
 363            const AwhParams&               awhParams,
 364            const AwhBiasParams&           awhBiasParams,
 365            ArrayRef<const DimParams>      dimParamsInit,
 366            double                         beta,
 367            double                         mdTimeStep,
 368            int                            numSharingSimulations,
 369            const std::string&             biasInitFilename,
 370            ThisRankWillDoIO               thisRankWillDoIO,
 371            BiasParams::DisableUpdateSkips disableUpdateSkips) :
 372     dimParams_(dimParamsInit.begin(), dimParamsInit.end()),
 373     grid_(dimParamsInit, awhBiasParams.dimParams),
 374     params_(awhParams,
 375             awhBiasParams,
 376             dimParams_,
 377             beta,
 378             mdTimeStep,
 379             disableUpdateSkips,
 380             numSharingSimulations,
 381             grid_.axis(),
 382             biasIndexInCollection),
 383     state_(awhBiasParams, params_.initialHistogramSize, dimParams_, grid_),
 384     thisRankDoesIO_(thisRankWillDoIO == ThisRankWillDoIO::Yes),
 385     biasForce_(ndim()),
 386
 387     tempForce_(ndim()),
 388     numWarningsIssued_(0)
 389 {
 390     /* For a global update updateList covers all points, so reserve that */
 391     updateList_.reserve(grid_.numPoints());
 392
 393     state_.initGridPointState(awhBiasParams, dimParams_, grid_, params_, biasInitFilename, awhParams.numBias);
 394
 395     if (thisRankDoesIO_)
 396     {
 397         /* Set up the force correlation object. */
 398
 399         /* We let the correlation init function set its parameters
 400          * to something useful for now.
 401          */
 402         double blockLength = 0;
 403         /* Construct the force correlation object. */
 404         forceCorrelationGrid_ =
 405                 std::make_unique<CorrelationGrid>(state_.points().size(),
 406                                                   ndim(),
 407                                                   blockLength,
 408                                                   CorrelationGrid::BlockLengthMeasure::Time,
 409                                                   awhParams.nstSampleCoord * mdTimeStep);
 410
 411         writer_ = std::make_unique<BiasWriter>(*this);
 412     }
 413 }
 414
 415 void Bias::printInitializationToLog(FILE* fplog) const
 416 {
 417     if (fplog != nullptr && forceCorrelationGrid_ != nullptr)
 418     {
 419         std::string prefix = gmx::formatString("\nawh%d:", params_.biasIndex + 1);
 420
 421         fprintf(fplog,
 422                 "%s initial force correlation block length = %g %s"
 423                 "%s force correlation number of blocks = %d",
 424                 prefix.c_str(),
 425                 forceCorrelationGrid().getBlockLength(),
 426                 forceCorrelationGrid().blockLengthMeasure == CorrelationGrid::BlockLengthMeasure::Weight
 427                         ? ""
 428                         : "ps",
 429                 prefix.c_str(),
 430                 forceCorrelationGrid().getNumBlocks());
 431     }
 432 }
 433
 434 void Bias::updateForceCorrelationGrid(gmx::ArrayRef<const double> probWeightNeighbor,
 435                                       ArrayRef<const double>      neighborLambdaDhdl,
 436                                       double                      t)
 437 {
 438     if (forceCorrelationGrid_ == nullptr)
 439     {
 440         return;
 441     }
 442
 443     const std::vector<int>& neighbor = grid_.point(state_.coordState().gridpointIndex()).neighbor;
 444
 445     gmx::ArrayRef<double> forceFromNeighbor = tempForce_;
 446     for (size_t n = 0; n < neighbor.size(); n++)
 447     {
 448         double weightNeighbor = probWeightNeighbor[n];
 449         int    indexNeighbor  = neighbor[n];
 450
 451         /* Add the force data of this neighbor point. Note: the sum of these forces is the convolved force.
 452
 453            We actually add the force normalized by beta which has the units of 1/length. This means that the
 454            resulting correlation time integral is directly in units of friction time/length^2 which is really what
 455            we're interested in. */
 456         state_.calcUmbrellaForceAndPotential(
 457                 dimParams_, grid_, indexNeighbor, neighborLambdaDhdl, forceFromNeighbor);
 458
 459         /* Note: we might want to give a whole list of data to add instead and have this loop in the data adding function */
 460         forceCorrelationGrid_->addData(indexNeighbor, weightNeighbor, forceFromNeighbor, t);
 461     }
 462 }
 463
 464 /* Return the number of data blocks that have been prepared for writing. */
 465 int Bias::numEnergySubblocksToWrite() const
 466 {
 467     GMX_RELEASE_ASSERT(writer_ != nullptr, "Should only request data from an initialized writer");
 468
 469     return writer_->numBlocks();
 470 }
 471
 472 /* Write bias data blocks to energy subblocks. */
 473 int Bias::writeToEnergySubblocks(t_enxsubblock* subblock) const
 474 {
 475     GMX_RELEASE_ASSERT(writer_ != nullptr, "Should only request data from an initialized writer");
 476
 477     return writer_->writeToEnergySubblocks(*this, subblock);
 478 }
 479
 480 bool Bias::isSampleCoordStep(const int64_t step) const
 481 {
 482     return params_.isSampleCoordStep(step);
 483 }
 484
 485
 486 } // namespace gmx