src/gromacs/awh/histogramsize.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2015,2016,2017,2018,2019, 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 HistogramSize 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 "histogramsize.h"
  48
  49 #include <cstdio>
  50 #include <cstdlib>
  51 #include <cstring>
  52
  53 #include <algorithm>
  54
  55 #include "gromacs/mdtypes/awh_history.h"
  56 #include "gromacs/mdtypes/awh_params.h"
  57 #include "gromacs/utility/gmxassert.h"
  58 #include "gromacs/utility/stringutil.h"
  59
  60 #include "biasparams.h"
  61 #include "pointstate.h"
  62
  63 namespace gmx
  64 {
  65
  66 HistogramSize::HistogramSize(const AwhBiasParams &awhBiasParams,
  67                              double               histogramSizeInitial) :
  68     numUpdates_(0),
  69     histogramSize_(histogramSizeInitial),
  70     inInitialStage_(awhBiasParams.eGrowth == eawhgrowthEXP_LINEAR),
  71     equilibrateHistogram_(awhBiasParams.equilibrateHistogram),
  72     logScaledSampleWeight_(0),
  73     maxLogScaledSampleWeight_(0),
  74     havePrintedAboutCovering_(false)
  75 {
  76 }
  77
  78 double HistogramSize::newHistogramSizeInitialStage(const BiasParams &params,
  79                                                    double            t,
  80                                                    bool              detectedCovering,
  81                                                    ArrayRef<double>  weightsumCovering,
  82                                                    FILE             *fplog)
  83 {
  84     /* The histogram size is kept constant until the sampling region has been covered
  85        and the the current sample weight is large enough and the histogram is ready. */
  86     if (!detectedCovering ||
  87         (logScaledSampleWeight_ < maxLogScaledSampleWeight_) ||
  88         equilibrateHistogram_)
  89     {
  90         return histogramSize_;
  91     }
  92
  93     /* Reset the covering weight histogram. If we got this far we are either entering a
  94        new covering stage with a new covering histogram or exiting the initial stage
  95        altogether. */
  96     std::fill(weightsumCovering.begin(), weightsumCovering.end(), 0);
  97
  98     /*  The current sample weigth is now the maximum. */
  99     double prevMaxLogScaledSampleWeight = maxLogScaledSampleWeight_;
 100     maxLogScaledSampleWeight_ = logScaledSampleWeight_;
 101
 102     /* Increase the histogram size by a constant scale factor if we can, i.e. if the sample weight
 103        resulting from such a scaling is still larger than the previous maximum sample weight
 104        (ensuring that the sample weights at the end of each covering stage are monotonically
 105        increasing). If we cannot, exit the initial stage without changing the histogram size. */
 106
 107     /* The scale factor. The value is not very critical but should obviously be > 1 (or the exit
 108        will happen very late) and probably < 5 or so (or there will be no initial stage). */
 109     static const double growthFactor = 3;
 110
 111     /* The scale factor is in most cases very close to the histogram growth factor. */
 112     double scaleFactor      = growthFactor/(1. + params.updateWeight*params.localWeightScaling/histogramSize_);
 113
 114     bool   exitInitialStage = (logScaledSampleWeight_ - std::log(scaleFactor) <= prevMaxLogScaledSampleWeight);
 115     double newHistogramSize = exitInitialStage ? histogramSize_ : histogramSize_*growthFactor;
 116
 117     /* Update the AWH bias about the exit. */
 118     inInitialStage_ = !exitInitialStage;
 119
 120     /* Print information about coverings and if there was an exit. */
 121     if (fplog != nullptr)
 122     {
 123         std::string prefix = gmx::formatString("\nawh%d:", params.biasIndex + 1);
 124         fprintf(fplog, "%s covering at t = %g ps. Decreased the update size.\n", prefix.c_str(), t);
 125
 126         if (exitInitialStage)
 127         {
 128             fprintf(fplog, "%s out of the initial stage at t = %g.\n", prefix.c_str(), t);
 129             /* It would be nice to have a way of estimating a minimum time until exit but it
 130                is difficult because the exit time is determined by how long it takes to cover
 131                relative to the time it takes to "regaining" enough sample weight. The latter
 132                is easy to calculate, but how the former depends on the histogram size
 133                is not known. */
 134         }
 135         fflush(fplog);
 136     }
 137     return newHistogramSize;
 138 }
 139
 140 namespace
 141 {
 142
 143 /*! \brief
 144  * Checks if the histogram has equilibrated to the target distribution.
 145  *
 146  * The histogram is considered equilibrated if, for a minimum fraction of
 147  * the target region, the relative error of the sampled weight relative
 148  * to the target is less than a tolerance value.
 149  *
 150  * \param[in] pointStates  The state of the bias points.
 151  * \returns true if the histogram is equilibrated.
 152  */
 153 bool histogramIsEquilibrated(const std::vector<PointState> &pointStates)
 154 {
 155     /* Get the total weight of the total weight histogram; needed for normalization. */
 156     double totalWeight     = 0;
 157     int    numTargetPoints = 0;
 158     for (auto &pointState : pointStates)
 159     {
 160         if (!pointState.inTargetRegion())
 161         {
 162             continue;
 163         }
 164         totalWeight += pointState.weightSumTot();
 165         numTargetPoints++;
 166     }
 167     GMX_RELEASE_ASSERT(totalWeight > 0, "No samples when normalizing AWH histogram.");
 168     double              inverseTotalWeight   = 1./totalWeight;
 169
 170     /* Points with target weight below a certain cutoff are ignored. */
 171     static const double minTargetCutoff  = 0.05;
 172     double              minTargetWeight  = 1./numTargetPoints*minTargetCutoff;
 173
 174     /* Points with error less than this tolerance pass the check.*/
 175     static const double errorTolerance   = 0.2;
 176
 177     /* Sum up weight of points that do or don't pass the check. */
 178     double equilibratedWeight    = 0;
 179     double notEquilibratedWeight = 0;
 180     for (auto &pointState : pointStates)
 181     {
 182         double targetWeight  = pointState.target();
 183         double sampledWeight = pointState.weightSumTot()*inverseTotalWeight;
 184
 185         /* Ignore these points. */
 186         if (!pointState.inTargetRegion() || targetWeight < minTargetWeight)
 187         {
 188             continue;
 189         }
 190
 191         if (std::abs(sampledWeight/targetWeight - 1) > errorTolerance)
 192         {
 193             notEquilibratedWeight += targetWeight;
 194         }
 195         else
 196         {
 197             equilibratedWeight += targetWeight;
 198         }
 199     }
 200
 201     /* It is enough if sampling in at least a fraction of the target region follows the target
 202        distribution. Boundaries will in general fail and this should be ignored (to some extent). */
 203     static const double minFraction = 0.8;
 204
 205     return equilibratedWeight/(equilibratedWeight + notEquilibratedWeight) > minFraction;;
 206 }
 207
 208 }   // namespace
 209
 210 double HistogramSize::newHistogramSize(const BiasParams              &params,
 211                                        double                         t,
 212                                        bool                           covered,
 213                                        const std::vector<PointState> &pointStates,
 214                                        ArrayRef<double>               weightsumCovering,
 215                                        FILE                          *fplog)
 216 {
 217     double newHistogramSize;
 218     if (inInitialStage_)
 219     {
 220         /* Only bother with checking equilibration if we have covered already. */
 221         if (equilibrateHistogram_ && covered)
 222         {
 223             /* The histogram is equilibrated at most once. */
 224             equilibrateHistogram_ = !histogramIsEquilibrated(pointStates);
 225
 226             if (fplog != nullptr)
 227             {
 228                 std::string prefix = gmx::formatString("\nawh%d:", params.biasIndex + 1);
 229                 if (!equilibrateHistogram_)
 230                 {
 231                     fprintf(fplog, "%s equilibrated histogram at t = %g ps.\n", prefix.c_str(), t);
 232                 }
 233                 else if (!havePrintedAboutCovering_)
 234                 {
 235                     fprintf(fplog, "%s covered but histogram not equilibrated at t = %g ps.\n", prefix.c_str(), t);
 236                     havePrintedAboutCovering_ = true; /* Just print once. */
 237                 }
 238             }
 239         }
 240
 241         /* In the initial stage, the histogram grows dynamically as a function of the number of coverings. */
 242         newHistogramSize = newHistogramSizeInitialStage(params, t, covered, weightsumCovering, fplog);
 243     }
 244     else
 245     {
 246         /* If not in the initial stage, the histogram grows at a linear, possibly scaled down, rate. */
 247         newHistogramSize = histogramSize_ + params.updateWeight*params.localWeightScaling;
 248     }
 249
 250     return newHistogramSize;
 251 }
 252
 253 void HistogramSize::setHistogramSize(double histogramSize,
 254                                      double weightHistogramScalingFactor)
 255 {
 256     GMX_ASSERT(histogramSize > 0, "The histogram should not be empty");
 257     GMX_ASSERT(weightHistogramScalingFactor > 0, "The histogram scaling factor should be positive");
 258
 259     histogramSize_ = histogramSize;
 260
 261     /* The weight of new samples relative to previous ones change
 262      * when the histogram is rescaled. We keep the log since this number
 263      * can become very large.
 264      */
 265     logScaledSampleWeight_ -= std::log(weightHistogramScalingFactor);
 266 };
 267
 268 void HistogramSize::restoreFromHistory(const AwhBiasStateHistory &stateHistory)
 269 {
 270     numUpdates_               = stateHistory.numUpdates;
 271     histogramSize_            = stateHistory.histSize;
 272     inInitialStage_           = stateHistory.in_initial;
 273     equilibrateHistogram_     = stateHistory.equilibrateHistogram;
 274     logScaledSampleWeight_    = stateHistory.logScaledSampleWeight;
 275     maxLogScaledSampleWeight_ = stateHistory.maxLogScaledSampleWeight;
 276     havePrintedAboutCovering_ = false;
 277 }
 278
 279 void HistogramSize::storeState(AwhBiasStateHistory *stateHistory) const
 280 {
 281     stateHistory->numUpdates               = numUpdates_;
 282     stateHistory->histSize                 = histogramSize_;
 283     stateHistory->in_initial               = inInitialStage_;
 284     stateHistory->equilibrateHistogram     = equilibrateHistogram_;
 285     stateHistory->logScaledSampleWeight    = logScaledSampleWeight_;
 286     stateHistory->maxLogScaledSampleWeight = maxLogScaledSampleWeight_;
 287     /* We'll print again about covering when restoring the state */
 288 }
 289
 290 } // namespace gmx