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     GMX_RELEASE_ASSERT(!(params_.convolveForce && grid_.hasLambdaAxis()),
 124                        "When using AWH to sample an FEP lambda dimension the AWH potential cannot "
 125                        "be convolved.");
 126
 127     state_.setCoordValue(grid_, coordValue);
 128
 129     std::vector<double, AlignedAllocator<double>>& probWeightNeighbor = alignedTempWorkSpace_;
 130
 131     /* If the convolved force is needed or this is a sampling step,
 132      * the bias in the current neighborhood needs to be up-to-date
 133      * and the probablity weights need to be calculated.
 134      */
 135     const bool        isSampleCoordStep = params_.isSampleCoordStep(step);
 136     const bool        moveUmbrella      = (isSampleCoordStep || step == 0);
 137     double            convolvedBias     = 0;
 138     const CoordState& coordState        = state_.coordState();
 139
 140     if (params_.convolveForce || moveUmbrella || isSampleCoordStep)
 141     {
 142         if (params_.skipUpdates())
 143         {
 144             state_.doSkippedUpdatesInNeighborhood(params_, grid_);
 145         }
 146         convolvedBias = state_.updateProbabilityWeightsAndConvolvedBias(
 147                 dimParams_, grid_, moveUmbrella ? neighborLambdaEnergies : ArrayRef<const double>{},
 148                 &probWeightNeighbor);
 149
 150         if (isSampleCoordStep)
 151         {
 152             updateForceCorrelationGrid(probWeightNeighbor, neighborLambdaDhdl, t);
 153
 154             state_.sampleCoordAndPmf(dimParams_, grid_, probWeightNeighbor, convolvedBias);
 155         }
 156     }
 157
 158     /* Set the bias force and get the potential contribution from this bias.
 159      * The potential jump occurs at different times depending on how
 160      * the force is applied (and how the potential is normalized).
 161      * For the convolved force it happens when the bias is updated,
 162      * for the umbrella when the umbrella is moved.
 163      */
 164     *potentialJump = 0;
 165     double potential;
 166     if (params_.convolveForce)
 167     {
 168         state_.calcConvolvedForce(dimParams_, grid_, probWeightNeighbor,
 169                                   moveUmbrella ? neighborLambdaDhdl : ArrayRef<const double>{},
 170                                   tempForce_, biasForce_);
 171
 172         potential = -convolvedBias * params_.invBeta;
 173     }
 174     else
 175     {
 176         /* Umbrella force */
 177         GMX_RELEASE_ASSERT(state_.points()[coordState.umbrellaGridpoint()].inTargetRegion(),
 178                            "AWH bias grid point for the umbrella reference value is outside of the "
 179                            "target region.");
 180         potential = state_.calcUmbrellaForceAndPotential(
 181                 dimParams_, grid_, coordState.umbrellaGridpoint(),
 182                 moveUmbrella ? neighborLambdaDhdl : ArrayRef<const double>{}, biasForce_);
 183
 184         /* Moving the umbrella results in a force correction and
 185          * a new potential. The umbrella center is sampled as often as
 186          * the coordinate so we know the probability weights needed
 187          * for moving the umbrella are up-to-date.
 188          */
 189         if (moveUmbrella)
 190         {
 191             double newPotential =
 192                     state_.moveUmbrella(dimParams_, grid_, probWeightNeighbor, neighborLambdaDhdl,
 193                                         biasForce_, step, seed, params_.biasIndex);
 194             *potentialJump = newPotential - potential;
 195         }
 196     }
 197
 198     /* Update the free energy estimates and bias and other history dependent method parameters */
 199     if (params_.isUpdateFreeEnergyStep(step))
 200     {
 201         state_.updateFreeEnergyAndAddSamplesToHistogram(dimParams_, grid_, params_, commRecord, ms,
 202                                                         t, step, fplog, &updateList_);
 203
 204         if (params_.convolveForce)
 205         {
 206             /* The update results in a potential jump, so we need the new convolved potential. */
 207             double newPotential = -calcConvolvedBias(coordState.coordValue()) * params_.invBeta;
 208             *potentialJump      = newPotential - potential;
 209         }
 210     }
 211
 212     /* Return the potential. */
 213     *awhPotential = potential;
 214
 215     /* Check the sampled histograms and potentially warn user if something is suspicious */
 216     warnForHistogramAnomalies(t, step, fplog);
 217
 218     return biasForce_;
 219 }
 220
 221 /*! \brief
 222  * Count the total number of samples / sample weight over all grid points.
 223  *
 224  * \param[in] pointState  The state of the points in a bias.
 225  * \returns the total sample count.
 226  */
 227 static int64_t countSamples(const std::vector<PointState>& pointState)
 228 {
 229     double numSamples = 0;
 230     for (const PointState& point : pointState)
 231     {
 232         numSamples += point.weightSumTot();
 233     }
 234
 235     return gmx::roundToInt64(numSamples);
 236 }
 237
 238 /*! \brief
 239  * Check if the state (loaded from checkpoint) and the run are consistent.
 240  *
 241  * When the state and the run setup are inconsistent, an exception is thrown.
 242  *
 243  * \param[in] params  The parameters of the bias.
 244  * \param[in] state   The state of the bias.
 245  */
 246 static void ensureStateAndRunConsistency(const BiasParams& params, const BiasState& state)
 247 {
 248     int64_t numSamples = countSamples(state.points());
 249     int64_t numUpdatesFromSamples =
 250             numSamples / (params.numSamplesUpdateFreeEnergy_ * params.numSharedUpdate);
 251     int64_t numUpdatesExpected = state.histogramSize().numUpdates();
 252     if (numUpdatesFromSamples != numUpdatesExpected)
 253     {
 254         std::string mesg = gmx::formatString(
 255                 "The number of AWH updates in the checkpoint file (%" PRId64
 256                 ") does not match the total number of AWH samples divided by the number of samples "
 257                 "per update for %d sharing AWH bias(es) (%" PRId64 "/%d=%" PRId64 ")",
 258                 numUpdatesExpected, params.numSharedUpdate, numSamples,
 259                 params.numSamplesUpdateFreeEnergy_ * params.numSharedUpdate, numUpdatesFromSamples);
 260         mesg += " Maybe you changed AWH parameters.";
 261         /* Unfortunately we currently do not store the number of simulations
 262          * sharing the bias or the state to checkpoint. But we can hint at
 263          * a mismatch in the number of sharing simulations.
 264          */
 265         if (numUpdatesFromSamples % state.histogramSize().numUpdates() == 0)
 266         {
 267             mesg += gmx::formatString(
 268                     " Or the run you continued from used %" PRId64
 269                     " sharing simulations, whereas you now specified %d sharing simulations.",
 270                     numUpdatesFromSamples / state.histogramSize().numUpdates(), params.numSharedUpdate);
 271         }
 272         GMX_THROW(InvalidInputError(mesg));
 273     }
 274 }
 275
 276 void Bias::restoreStateFromHistory(const AwhBiasHistory* biasHistory, const t_commrec* cr)
 277 {
 278     GMX_RELEASE_ASSERT(thisRankDoesIO_ == MASTER(cr),
 279                        "The master rank should do I/O, the other ranks should not");
 280
 281     if (MASTER(cr))
 282     {
 283         GMX_RELEASE_ASSERT(biasHistory != nullptr,
 284                            "On the master rank we need a valid history object to restore from");
 285         state_.restoreFromHistory(*biasHistory, grid_);
 286
 287         /* Ensure that the state is consistent with our current run setup,
 288          * since the user can have changed AWH parameters or the number
 289          * of simulations sharing the bias.
 290          */
 291         ensureStateAndRunConsistency(params_, state_);
 292
 293         if (forceCorrelationGrid_ != nullptr)
 294         {
 295             forceCorrelationGrid_->restoreStateFromHistory(biasHistory->forceCorrelationGrid);
 296         }
 297     }
 298
 299     if (PAR(cr))
 300     {
 301         state_.broadcast(cr);
 302     }
 303 }
 304
 305 void Bias::initHistoryFromState(AwhBiasHistory* biasHistory) const
 306 {
 307     GMX_RELEASE_ASSERT(biasHistory != nullptr, "Need a valid biasHistory");
 308
 309     state_.initHistoryFromState(biasHistory);
 310
 311     if (forceCorrelationGrid_ != nullptr)
 312     {
 313         biasHistory->forceCorrelationGrid = initCorrelationGridHistoryFromState(forceCorrelationGrid());
 314     }
 315 }
 316
 317 void Bias::updateHistory(AwhBiasHistory* biasHistory) const
 318 {
 319     GMX_RELEASE_ASSERT(biasHistory != nullptr, "Need a valid biasHistory");
 320
 321     state_.updateHistory(biasHistory, grid_);
 322
 323     if (forceCorrelationGrid_ != nullptr)
 324     {
 325         updateCorrelationGridHistory(&biasHistory->forceCorrelationGrid, forceCorrelationGrid());
 326     }
 327 }
 328
 329 Bias::Bias(int                            biasIndexInCollection,
 330            const AwhParams&               awhParams,
 331            const AwhBiasParams&           awhBiasParams,
 332            const std::vector<DimParams>&  dimParamsInit,
 333            double                         beta,
 334            double                         mdTimeStep,
 335            int                            numSharingSimulations,
 336            const std::string&             biasInitFilename,
 337            ThisRankWillDoIO               thisRankWillDoIO,
 338            BiasParams::DisableUpdateSkips disableUpdateSkips) :
 339     dimParams_(dimParamsInit),
 340     grid_(dimParamsInit, awhBiasParams.dimParams),
 341     params_(awhParams,
 342             awhBiasParams,
 343             dimParams_,
 344             beta,
 345             mdTimeStep,
 346             disableUpdateSkips,
 347             numSharingSimulations,
 348             grid_.axis(),
 349             biasIndexInCollection),
 350     state_(awhBiasParams, params_.initialHistogramSize, dimParams_, grid_),
 351     thisRankDoesIO_(thisRankWillDoIO == ThisRankWillDoIO::Yes),
 352     biasForce_(ndim()),
 353
 354     tempForce_(ndim()),
 355     numWarningsIssued_(0)
 356 {
 357     /* For a global update updateList covers all points, so reserve that */
 358     updateList_.reserve(grid_.numPoints());
 359
 360     state_.initGridPointState(awhBiasParams, dimParams_, grid_, params_, biasInitFilename, awhParams.numBias);
 361
 362     if (thisRankDoesIO_)
 363     {
 364         /* Set up the force correlation object. */
 365
 366         /* We let the correlation init function set its parameters
 367          * to something useful for now.
 368          */
 369         double blockLength = 0;
 370         /* Construct the force correlation object. */
 371         forceCorrelationGrid_ = std::make_unique<CorrelationGrid>(
 372                 state_.points().size(), ndim(), blockLength,
 373                 CorrelationGrid::BlockLengthMeasure::Time, awhParams.nstSampleCoord * mdTimeStep);
 374
 375         writer_ = std::make_unique<BiasWriter>(*this);
 376     }
 377 }
 378
 379 void Bias::printInitializationToLog(FILE* fplog) const
 380 {
 381     if (fplog != nullptr && forceCorrelationGrid_ != nullptr)
 382     {
 383         std::string prefix = gmx::formatString("\nawh%d:", params_.biasIndex + 1);
 384
 385         fprintf(fplog,
 386                 "%s initial force correlation block length = %g %s"
 387                 "%s force correlation number of blocks = %d",
 388                 prefix.c_str(), forceCorrelationGrid().getBlockLength(),
 389                 forceCorrelationGrid().blockLengthMeasure == CorrelationGrid::BlockLengthMeasure::Weight
 390                         ? ""
 391                         : "ps",
 392                 prefix.c_str(), forceCorrelationGrid().getNumBlocks());
 393     }
 394 }
 395
 396 void Bias::updateForceCorrelationGrid(gmx::ArrayRef<const double> probWeightNeighbor,
 397                                       ArrayRef<const double>      neighborLambdaDhdl,
 398                                       double                      t)
 399 {
 400     if (forceCorrelationGrid_ == nullptr)
 401     {
 402         return;
 403     }
 404
 405     const std::vector<int>& neighbor = grid_.point(state_.coordState().gridpointIndex()).neighbor;
 406
 407     gmx::ArrayRef<double> forceFromNeighbor = tempForce_;
 408     for (size_t n = 0; n < neighbor.size(); n++)
 409     {
 410         double weightNeighbor = probWeightNeighbor[n];
 411         int    indexNeighbor  = neighbor[n];
 412
 413         /* Add the force data of this neighbor point. Note: the sum of these forces is the convolved force.
 414
 415            We actually add the force normalized by beta which has the units of 1/length. This means that the
 416            resulting correlation time integral is directly in units of friction time/length^2 which is really what
 417            we're interested in. */
 418         state_.calcUmbrellaForceAndPotential(dimParams_, grid_, indexNeighbor, neighborLambdaDhdl,
 419                                              forceFromNeighbor);
 420
 421         /* Note: we might want to give a whole list of data to add instead and have this loop in the data adding function */
 422         forceCorrelationGrid_->addData(indexNeighbor, weightNeighbor, forceFromNeighbor, t);
 423     }
 424 }
 425
 426 /* Return the number of data blocks that have been prepared for writing. */
 427 int Bias::numEnergySubblocksToWrite() const
 428 {
 429     GMX_RELEASE_ASSERT(writer_ != nullptr, "Should only request data from an initialized writer");
 430
 431     return writer_->numBlocks();
 432 }
 433
 434 /* Write bias data blocks to energy subblocks. */
 435 int Bias::writeToEnergySubblocks(t_enxsubblock* subblock) const
 436 {
 437     GMX_RELEASE_ASSERT(writer_ != nullptr, "Should only request data from an initialized writer");
 438
 439     return writer_->writeToEnergySubblocks(*this, subblock);
 440 }
 441
 442 bool Bias::isSampleCoordStep(const int64_t step) const
 443 {
 444     return params_.isSampleCoordStep(step);
 445 }
 446
 447
 448 } // namespace gmx