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, 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 "testutils/refdata.h"
  53 #include "testutils/testasserts.h"
  54
  55 namespace gmx
  56 {
  57
  58 namespace test
  59 {
  60
  61 //! The number of lambda states to use in the tests.
  62 const int numLambdaStates = 16;
  63
  64 /*! \internal \brief
  65  * Struct that gathers all input for setting up and using a Bias
  66  */
  67 struct AwhFepLambdaStateTestParameters
  68 {
  69     AwhFepLambdaStateTestParameters() = default;
  70     //! Move constructor
  71     AwhFepLambdaStateTestParameters(AwhFepLambdaStateTestParameters&& o) noexcept :
  72         beta(o.beta),
  73         awhDimParams(o.awhDimParams),
  74         awhBiasParams(o.awhBiasParams),
  75         awhParams(o.awhParams),
  76         dimParams(std::move(o.dimParams))
  77     {
  78         awhBiasParams.dimParams = &awhDimParams;
  79         awhParams.awhBiasParams = &awhBiasParams;
  80     }
  81     double beta; //!< 1/(kB*T)
  82
  83     AwhDimParams  awhDimParams;  //!< Dimension parameters pointed to by \p awhBiasParams
  84     AwhBiasParams awhBiasParams; //!< Bias parameters pointed to by \[ awhParams
  85     AwhParams     awhParams;     //!< AWH parameters, this is the struct to actually use
  86
  87     std::vector<DimParams> dimParams; //!< Dimension parameters for setting up Bias
  88 };
  89
  90 //! Helper function to set up the C-style AWH parameters for the test
  91 static AwhFepLambdaStateTestParameters getAwhFepLambdaTestParameters(int eawhgrowth, int eawhpotential)
  92 {
  93     AwhFepLambdaStateTestParameters params;
  94
  95     params.beta = 0.4;
  96
  97     AwhDimParams& awhDimParams = params.awhDimParams;
  98
  99     awhDimParams.period = 0;
 100     // Correction for removal of GaussianGeometryFactor/2 in histogram size
 101     awhDimParams.diffusion      = 1e-4 / (0.12927243028700 * 2);
 102     awhDimParams.origin         = 0;
 103     awhDimParams.end            = numLambdaStates - 1;
 104     awhDimParams.coordValueInit = awhDimParams.origin;
 105     awhDimParams.coverDiameter  = 0;
 106     awhDimParams.eCoordProvider = eawhcoordproviderFREE_ENERGY_LAMBDA;
 107
 108     AwhBiasParams& awhBiasParams = params.awhBiasParams;
 109
 110     awhBiasParams.ndim                 = 1;
 111     awhBiasParams.dimParams            = &awhDimParams;
 112     awhBiasParams.eTarget              = eawhtargetCONSTANT;
 113     awhBiasParams.targetBetaScaling    = 0;
 114     awhBiasParams.targetCutoff         = 0;
 115     awhBiasParams.eGrowth              = eawhgrowth;
 116     awhBiasParams.bUserData            = FALSE;
 117     awhBiasParams.errorInitial         = 1.0 / params.beta;
 118     awhBiasParams.shareGroup           = 0;
 119     awhBiasParams.equilibrateHistogram = FALSE;
 120
 121     int64_t seed = 93471803;
 122
 123     params.dimParams.push_back(DimParams::fepLambdaDimParams(numLambdaStates, params.beta));
 124
 125     AwhParams& awhParams = params.awhParams;
 126
 127     awhParams.numBias                    = 1;
 128     awhParams.awhBiasParams              = &awhBiasParams;
 129     awhParams.seed                       = seed;
 130     awhParams.nstOut                     = 0;
 131     awhParams.nstSampleCoord             = 1;
 132     awhParams.numSamplesUpdateFreeEnergy = 10;
 133     awhParams.ePotential                 = eawhpotential;
 134     awhParams.shareBiasMultisim          = FALSE;
 135
 136     return params;
 137 }
 138
 139 //! Convenience typedef: growth type enum, potential type enum, disable update skips
 140 typedef std::tuple<int, int, BiasParams::DisableUpdateSkips> BiasTestParameters;
 141
 142 /*! \brief Test fixture for testing Bias updates
 143  */
 144 class BiasFepLambdaStateTest : public ::testing::TestWithParam<BiasTestParameters>
 145 {
 146 public:
 147     //! Random seed for AWH MC sampling
 148     int64_t seed_;
 149
 150     //! The awh Bias
 151     std::unique_ptr<Bias> bias_;
 152
 153     BiasFepLambdaStateTest()
 154     {
 155         /* We test all combinations of:
 156          *   eawhgrowth:
 157          *     eawhgrowthLINEAR:     final, normal update phase
 158          *     ewahgrowthEXP_LINEAR: intial phase, updated size is constant
 159          *   eawhpotential (test both, but for the FEP lambda state dimension MC will in practice be used,
 160          *                  except that eawhpotentialCONVOLVED also gives a potential output):
 161          *     eawhpotentialUMBRELLA:  MC on lambda state
 162          *     eawhpotentialCONVOLVED: MD on a convolved potential landscape (falling back to MC on lambda state)
 163          *   disableUpdateSkips (should not affect the results):
 164          *     BiasParams::DisableUpdateSkips::yes: update the point state for every sample
 165          *     BiasParams::DisableUpdateSkips::no:  update the point state at an interval > 1 sample
 166          *
 167          * Note: It would be nice to explicitly check that eawhpotential
 168          *       and disableUpdateSkips do not affect the point state.
 169          *       But the reference data will also ensure this.
 170          */
 171         int                            eawhgrowth;
 172         int                            eawhpotential;
 173         BiasParams::DisableUpdateSkips disableUpdateSkips;
 174         std::tie(eawhgrowth, eawhpotential, disableUpdateSkips) = GetParam();
 175
 176         /* Set up a basic AWH setup with a single, 1D bias with parameters
 177          * such that we can measure the effects of different parameters.
 178          */
 179         const AwhFepLambdaStateTestParameters params =
 180                 getAwhFepLambdaTestParameters(eawhgrowth, eawhpotential);
 181
 182         seed_ = params.awhParams.seed;
 183
 184         double mdTimeStep = 0.1;
 185
 186         bias_ = std::make_unique<Bias>(-1,
 187                                        params.awhParams,
 188                                        params.awhBiasParams,
 189                                        params.dimParams,
 190                                        params.beta,
 191                                        mdTimeStep,
 192                                        1,
 193                                        "",
 194                                        Bias::ThisRankWillDoIO::No,
 195                                        disableUpdateSkips);
 196     }
 197 };
 198
 199 TEST_P(BiasFepLambdaStateTest, ForcesBiasPmf)
 200 {
 201     gmx::test::TestReferenceData    data;
 202     gmx::test::TestReferenceChecker checker(data.rootChecker());
 203
 204     Bias& bias = *bias_;
 205
 206     /* Make strings with the properties we expect to be different in the tests.
 207      * These also helps to interpret the reference data.
 208      */
 209     std::vector<std::string> props;
 210     props.push_back(formatString("stage:           %s", bias.state().inInitialStage() ? "initial" : "final"));
 211     props.push_back(formatString("convolve forces: %s", bias.params().convolveForce ? "yes" : "no"));
 212     props.push_back(formatString("skip updates:    %s", bias.params().skipUpdates() ? "yes" : "no"));
 213
 214     SCOPED_TRACE(gmx::formatString("%s, %s, %s", props[0].c_str(), props[1].c_str(), props[2].c_str()));
 215
 216     std::vector<double> force, pot;
 217
 218     double potentialJump = 0;
 219     double mdTimeStep    = 0.1;
 220     int    nSteps        = 501;
 221
 222     /* Some energies to use as base values (to which some noise is added later on). */
 223     std::vector<double> neighborLambdaEnergies(numLambdaStates);
 224     std::vector<double> neighborLambdaDhdl(numLambdaStates);
 225     const double        magnitude = 12.0;
 226     for (int i = 0; i < numLambdaStates; i++)
 227     {
 228         neighborLambdaEnergies[i] = magnitude * std::sin(i * 0.1);
 229         neighborLambdaDhdl[i]     = magnitude * std::cos(i * 0.1);
 230     }
 231
 232     for (int step = 0; step < nSteps; step++)
 233     {
 234         int      umbrellaGridpointIndex = bias.state().coordState().umbrellaGridpoint();
 235         awh_dvec coordValue = { bias.getGridCoordValue(umbrellaGridpointIndex)[0], 0, 0, 0 };
 236         double   potential  = 0;
 237         gmx::ArrayRef<const double> biasForce = bias.calcForceAndUpdateBias(coordValue,
 238                                                                             neighborLambdaEnergies,
 239                                                                             neighborLambdaDhdl,
 240                                                                             &potential,
 241                                                                             &potentialJump,
 242                                                                             nullptr,
 243                                                                             nullptr,
 244                                                                             step * mdTimeStep,
 245                                                                             step,
 246                                                                             seed_,
 247                                                                             nullptr);
 248
 249         force.push_back(biasForce[0]);
 250         pot.push_back(potential);
 251     }
 252
 253     /* When skipping updates, ensure all skipped updates are performed here.
 254      * This should result in the same bias state as at output in a normal run.
 255      */
 256     if (bias.params().skipUpdates())
 257     {
 258         bias.doSkippedUpdatesForAllPoints();
 259     }
 260
 261     std::vector<double> pointBias, logPmfsum;
 262     for (auto& point : bias.state().points())
 263     {
 264         pointBias.push_back(point.bias());
 265         logPmfsum.push_back(point.logPmfSum());
 266     }
 267
 268     constexpr int ulpTol = 10;
 269
 270     checker.checkSequence(props.begin(), props.end(), "Properties");
 271     checker.setDefaultTolerance(absoluteTolerance(magnitude * GMX_DOUBLE_EPS * ulpTol));
 272     checker.checkSequence(force.begin(), force.end(), "Force");
 273     checker.checkSequence(pot.begin(), pot.end(), "Potential");
 274     checker.setDefaultTolerance(relativeToleranceAsUlp(1.0, ulpTol));
 275     checker.checkSequence(pointBias.begin(), pointBias.end(), "PointBias");
 276     checker.checkSequence(logPmfsum.begin(), logPmfsum.end(), "PointLogPmfsum");
 277 }
 278
 279 /* Scan initial/final phase, MC/convolved force and update skip (not) allowed
 280  * Both the convolving and skipping should not affect the bias and PMF.
 281  * It would be nice if the test would explicitly check for this.
 282  * Currently this is tested through identical reference data.
 283  */
 284 INSTANTIATE_TEST_CASE_P(WithParameters,
 285                         BiasFepLambdaStateTest,
 286                         ::testing::Combine(::testing::Values(eawhgrowthLINEAR, eawhgrowthEXP_LINEAR),
 287                                            ::testing::Values(eawhpotentialUMBRELLA, eawhpotentialCONVOLVED),
 288                                            ::testing::Values(BiasParams::DisableUpdateSkips::yes,
 289                                                              BiasParams::DisableUpdateSkips::no)));
 290
 291 // Test that we detect coverings and exit the initial stage at the correct step
 292 TEST(BiasFepLambdaStateTest, DetectsCovering)
 293 {
 294     const AwhFepLambdaStateTestParameters params =
 295             getAwhFepLambdaTestParameters(eawhgrowthEXP_LINEAR, eawhpotentialCONVOLVED);
 296
 297     const double mdTimeStep = 0.1;
 298
 299     Bias bias(-1,
 300               params.awhParams,
 301               params.awhBiasParams,
 302               params.dimParams,
 303               params.beta,
 304               mdTimeStep,
 305               1,
 306               "",
 307               Bias::ThisRankWillDoIO::No);
 308
 309     const int64_t exitStepRef = 320;
 310
 311     bool inInitialStage = bias.state().inInitialStage();
 312
 313     /* Some energies to use as base values (to which some noise is added later on). */
 314     std::vector<double> neighborLambdaEnergies(numLambdaStates);
 315     std::vector<double> neighborLambdaDhdl(numLambdaStates);
 316     const double        magnitude = 12.0;
 317     for (int i = 0; i < numLambdaStates; i++)
 318     {
 319         neighborLambdaEnergies[i] = magnitude * std::sin(i * 0.1);
 320         neighborLambdaDhdl[i]     = magnitude * std::cos(i * 0.1);
 321     }
 322
 323     int64_t step;
 324     /* Normally this loop exits at exitStepRef, but we extend with failure */
 325     for (step = 0; step <= 2 * exitStepRef; step++)
 326     {
 327         int      umbrellaGridpointIndex = bias.state().coordState().umbrellaGridpoint();
 328         awh_dvec coordValue = { bias.getGridCoordValue(umbrellaGridpointIndex)[0], 0, 0, 0 };
 329
 330         double potential     = 0;
 331         double potentialJump = 0;
 332         bias.calcForceAndUpdateBias(coordValue,
 333                                     neighborLambdaEnergies,
 334                                     neighborLambdaDhdl,
 335                                     &potential,
 336                                     &potentialJump,
 337                                     nullptr,
 338                                     nullptr,
 339                                     step,
 340                                     step,
 341                                     params.awhParams.seed,
 342                                     nullptr);
 343
 344         inInitialStage = bias.state().inInitialStage();
 345         if (!inInitialStage)
 346         {
 347             break;
 348         }
 349     }
 350
 351     EXPECT_EQ(false, inInitialStage);
 352     if (!inInitialStage)
 353     {
 354         EXPECT_EQ(exitStepRef, step);
 355     }
 356 }
 357
 358 } // namespace test
 359 } // namespace gmx