src/gromacs/applied_forces/awh/tests/bias_fep_lambda_state.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 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 #include "gmxpre.h"
  36
  37 #include <cmath>
  38
  39 #include <memory>
  40 #include <tuple>
  41 #include <vector>
  42
  43 #include <gmock/gmock.h>
  44 #include <gtest/gtest.h>
  45
  46 #include "gromacs/applied_forces/awh/bias.h"
  47 #include "gromacs/applied_forces/awh/correlationgrid.h"
  48 #include "gromacs/applied_forces/awh/pointstate.h"
  49 #include "gromacs/mdtypes/awh_params.h"
  50 #include "gromacs/utility/stringutil.h"
  51
  52 #include "gromacs/applied_forces/awh/tests/awh_setup.h"
  53 #include "testutils/refdata.h"
  54 #include "testutils/testasserts.h"
  55
  56 namespace gmx
  57 {
  58
  59 namespace test
  60 {
  61
  62 //! The number of lambda states to use in the tests.
  63 static constexpr int c_numLambdaStates = 16;
  64
  65
  66 //! Convenience typedef: growth type enum, potential type enum, disable update skips
  67 typedef std::tuple<AwhHistogramGrowthType, AwhPotentialType, BiasParams::DisableUpdateSkips> BiasTestParameters;
  68
  69 /*! \brief Test fixture for testing Bias updates
  70  */
  71 class BiasFepLambdaStateTest : public ::testing::TestWithParam<BiasTestParameters>
  72 {
  73 private:
  74     //! Storage for test parameters.
  75     std::unique_ptr<AwhTestParameters> params_;
  76
  77 public:
  78     //! Random seed for AWH MC sampling
  79     int64_t seed_;
  80
  81     //! The awh Bias
  82     std::unique_ptr<Bias> bias_;
  83
  84     BiasFepLambdaStateTest()
  85     {
  86         /* We test all combinations of:
  87          *   eawhgrowth:
  88          *     eawhgrowthLINEAR:     final, normal update phase
  89          *     ewahgrowthEXP_LINEAR: intial phase, updated size is constant
  90          *   eawhpotential (test both, but for the FEP lambda state dimension MC will in practice be used,
  91          *                  except that eawhpotentialCONVOLVED also gives a potential output):
  92          *     eawhpotentialUMBRELLA:  MC on lambda state
  93          *     eawhpotentialCONVOLVED: MD on a convolved potential landscape (falling back to MC on lambda state)
  94          *   disableUpdateSkips (should not affect the results):
  95          *     BiasParams::DisableUpdateSkips::yes: update the point state for every sample
  96          *     BiasParams::DisableUpdateSkips::no:  update the point state at an interval > 1 sample
  97          *
  98          * Note: It would be nice to explicitly check that eawhpotential
  99          *       and disableUpdateSkips do not affect the point state.
 100          *       But the reference data will also ensure this.
 101          */
 102         AwhHistogramGrowthType         eawhgrowth;
 103         AwhPotentialType               eawhpotential;
 104         BiasParams::DisableUpdateSkips disableUpdateSkips;
 105         std::tie(eawhgrowth, eawhpotential, disableUpdateSkips) = GetParam();
 106
 107         /* Set up a basic AWH setup with a single, 1D bias with parameters
 108          * such that we can measure the effects of different parameters.
 109          */
 110         constexpr AwhCoordinateProviderType coordinateProvider = AwhCoordinateProviderType::FreeEnergyLambda;
 111         constexpr int                       coordIndex = 0;
 112         constexpr double                    origin     = 0;
 113         constexpr double                    end        = c_numLambdaStates - 1;
 114         constexpr double                    period     = 0;
 115         // Correction for removal of GaussianGeometryFactor/2 in histogram size
 116         constexpr double diffusion = 1e-4 / (0.12927243028700 * 2);
 117         const auto       awhDimBuffer =
 118                 awhDimParamSerialized(coordinateProvider, coordIndex, origin, end, period, diffusion);
 119         const auto awhDimArrayRef = gmx::arrayRefFromArray(&awhDimBuffer, 1);
 120         params_                   = std::make_unique<AwhTestParameters>(getAwhTestParameters(
 121                 eawhgrowth, eawhpotential, awhDimArrayRef, false, 0.4, true, 1.0, c_numLambdaStates));
 122
 123         seed_ = params_->awhParams.seed();
 124
 125         double mdTimeStep = 0.1;
 126
 127         bias_ = std::make_unique<Bias>(-1,
 128                                        params_->awhParams,
 129                                        params_->awhParams.awhBiasParams()[0],
 130                                        params_->dimParams,
 131                                        params_->beta,
 132                                        mdTimeStep,
 133                                        1,
 134                                        "",
 135                                        Bias::ThisRankWillDoIO::No,
 136                                        disableUpdateSkips);
 137     }
 138 };
 139
 140 TEST_P(BiasFepLambdaStateTest, ForcesBiasPmf)
 141 {
 142     gmx::test::TestReferenceData    data;
 143     gmx::test::TestReferenceChecker checker(data.rootChecker());
 144
 145     Bias& bias = *bias_;
 146
 147     /* Make strings with the properties we expect to be different in the tests.
 148      * These also helps to interpret the reference data.
 149      */
 150     std::vector<std::string> props;
 151     props.push_back(formatString("stage:           %s", bias.state().inInitialStage() ? "initial" : "final"));
 152     props.push_back(formatString("convolve forces: %s", bias.params().convolveForce ? "yes" : "no"));
 153     props.push_back(formatString("skip updates:    %s", bias.params().skipUpdates() ? "yes" : "no"));
 154
 155     SCOPED_TRACE(gmx::formatString("%s, %s, %s", props[0].c_str(), props[1].c_str(), props[2].c_str()));
 156
 157     std::vector<double> force, pot;
 158
 159     double potentialJump = 0;
 160     double mdTimeStep    = 0.1;
 161     int    nSteps        = 501;
 162
 163     /* Some energies to use as base values (to which some noise is added later on). */
 164     std::vector<double> neighborLambdaEnergies(c_numLambdaStates);
 165     std::vector<double> neighborLambdaDhdl(c_numLambdaStates);
 166     const double        magnitude = 12.0;
 167     for (int i = 0; i < c_numLambdaStates; i++)
 168     {
 169         neighborLambdaEnergies[i] = magnitude * std::sin(i * 0.1);
 170         neighborLambdaDhdl[i]     = magnitude * std::cos(i * 0.1);
 171     }
 172
 173     for (int step = 0; step < nSteps; step++)
 174     {
 175         int      umbrellaGridpointIndex = bias.state().coordState().umbrellaGridpoint();
 176         awh_dvec coordValue = { bias.getGridCoordValue(umbrellaGridpointIndex)[0], 0, 0, 0 };
 177         double   potential  = 0;
 178         gmx::ArrayRef<const double> biasForce = bias.calcForceAndUpdateBias(coordValue,
 179                                                                             neighborLambdaEnergies,
 180                                                                             neighborLambdaDhdl,
 181                                                                             &potential,
 182                                                                             &potentialJump,
 183                                                                             nullptr,
 184                                                                             nullptr,
 185                                                                             step * mdTimeStep,
 186                                                                             step,
 187                                                                             seed_,
 188                                                                             nullptr);
 189
 190         force.push_back(biasForce[0]);
 191         pot.push_back(potential);
 192     }
 193
 194     /* When skipping updates, ensure all skipped updates are performed here.
 195      * This should result in the same bias state as at output in a normal run.
 196      */
 197     if (bias.params().skipUpdates())
 198     {
 199         bias.doSkippedUpdatesForAllPoints();
 200     }
 201
 202     std::vector<double> pointBias, logPmfsum;
 203     for (const auto& point : bias.state().points())
 204     {
 205         pointBias.push_back(point.bias());
 206         logPmfsum.push_back(point.logPmfSum());
 207     }
 208
 209     constexpr int ulpTol = 10;
 210
 211     checker.checkSequence(props.begin(), props.end(), "Properties");
 212     checker.setDefaultTolerance(absoluteTolerance(magnitude * GMX_DOUBLE_EPS * ulpTol));
 213     checker.checkSequence(force.begin(), force.end(), "Force");
 214     checker.checkSequence(pot.begin(), pot.end(), "Potential");
 215     checker.setDefaultTolerance(relativeToleranceAsUlp(1.0, ulpTol));
 216     checker.checkSequence(pointBias.begin(), pointBias.end(), "PointBias");
 217     checker.checkSequence(logPmfsum.begin(), logPmfsum.end(), "PointLogPmfsum");
 218 }
 219
 220 /* Scan initial/final phase, MC/convolved force and update skip (not) allowed
 221  * Both the convolving and skipping should not affect the bias and PMF.
 222  * It would be nice if the test would explicitly check for this.
 223  * Currently this is tested through identical reference data.
 224  */
 225 INSTANTIATE_TEST_CASE_P(WithParameters,
 226                         BiasFepLambdaStateTest,
 227                         ::testing::Combine(::testing::Values(AwhHistogramGrowthType::Linear,
 228                                                              AwhHistogramGrowthType::ExponentialLinear),
 229                                            ::testing::Values(AwhPotentialType::Umbrella,
 230                                                              AwhPotentialType::Convolved),
 231                                            ::testing::Values(BiasParams::DisableUpdateSkips::yes,
 232                                                              BiasParams::DisableUpdateSkips::no)));
 233
 234 // Test that we detect coverings and exit the initial stage at the correct step
 235 TEST(BiasFepLambdaStateTest, DetectsCovering)
 236 {
 237     constexpr AwhCoordinateProviderType coordinateProvider = AwhCoordinateProviderType::FreeEnergyLambda;
 238     constexpr int                       coordIndex         = 0;
 239     constexpr double                    origin             = 0;
 240     constexpr double                    end                = c_numLambdaStates - 1;
 241     constexpr double                    period             = 0;
 242     constexpr double                    diffusion          = 1e-4 / (0.12927243028700 * 2);
 243     auto                                awhDimBuffer =
 244             awhDimParamSerialized(coordinateProvider, coordIndex, origin, end, period, diffusion);
 245     auto                    awhDimArrayRef = gmx::arrayRefFromArray(&awhDimBuffer, 1);
 246     const AwhTestParameters params(getAwhTestParameters(AwhHistogramGrowthType::ExponentialLinear,
 247                                                         AwhPotentialType::Convolved,
 248                                                         awhDimArrayRef,
 249                                                         false,
 250                                                         0.4,
 251                                                         true,
 252                                                         1.0,
 253                                                         c_numLambdaStates));
 254
 255     const double mdTimeStep = 0.1;
 256
 257     Bias bias(-1,
 258               params.awhParams,
 259               params.awhParams.awhBiasParams()[0],
 260               params.dimParams,
 261               params.beta,
 262               mdTimeStep,
 263               1,
 264               "",
 265               Bias::ThisRankWillDoIO::No);
 266
 267     const int64_t exitStepRef = 320;
 268
 269     bool inInitialStage = bias.state().inInitialStage();
 270
 271     /* Some energies to use as base values (to which some noise is added later on). */
 272     std::vector<double> neighborLambdaEnergies(c_numLambdaStates);
 273     std::vector<double> neighborLambdaDhdl(c_numLambdaStates);
 274     const double        magnitude = 12.0;
 275     for (int i = 0; i < c_numLambdaStates; i++)
 276     {
 277         neighborLambdaEnergies[i] = magnitude * std::sin(i * 0.1);
 278         neighborLambdaDhdl[i]     = magnitude * std::cos(i * 0.1);
 279     }
 280
 281     int64_t step;
 282     /* Normally this loop exits at exitStepRef, but we extend with failure */
 283     for (step = 0; step <= 2 * exitStepRef; step++)
 284     {
 285         int      umbrellaGridpointIndex = bias.state().coordState().umbrellaGridpoint();
 286         awh_dvec coordValue = { bias.getGridCoordValue(umbrellaGridpointIndex)[0], 0, 0, 0 };
 287
 288         double potential     = 0;
 289         double potentialJump = 0;
 290         bias.calcForceAndUpdateBias(coordValue,
 291                                     neighborLambdaEnergies,
 292                                     neighborLambdaDhdl,
 293                                     &potential,
 294                                     &potentialJump,
 295                                     nullptr,
 296                                     nullptr,
 297                                     step,
 298                                     step,
 299                                     params.awhParams.seed(),
 300                                     nullptr);
 301
 302         inInitialStage = bias.state().inInitialStage();
 303         if (!inInitialStage)
 304         {
 305             break;
 306         }
 307     }
 308
 309     EXPECT_EQ(false, inInitialStage);
 310     if (!inInitialStage)
 311     {
 312         EXPECT_EQ(exitStepRef, step);
 313     }
 314 }
 315
 316 } // namespace test
 317 } // namespace gmx