Introduce SimulatorBuilder

[gromacs.git] / src / gromacs / awh / tests / bias.cpp

/*
 * This file is part of the GROMACS molecular simulation package.
 *
 * Copyright (c) 2017,2018,2019, by the GROMACS development team, led by
 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
 * and including many others, as listed in the AUTHORS file in the
 * top-level source directory and at http://www.gromacs.org.
 *
 * GROMACS is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public License
 * as published by the Free Software Foundation; either version 2.1
 * of the License, or (at your option) any later version.
 *
 * GROMACS is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with GROMACS; if not, see
 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
 *
 * If you want to redistribute modifications to GROMACS, please
 * consider that scientific software is very special. Version
 * control is crucial - bugs must be traceable. We will be happy to
 * consider code for inclusion in the official distribution, but
 * derived work must not be called official GROMACS. Details are found
 * in the README & COPYING files - if they are missing, get the
 * official version at http://www.gromacs.org.
 *
 * To help us fund GROMACS development, we humbly ask that you cite
 * the research papers on the package. Check out http://www.gromacs.org.
 */
#include "gmxpre.h"

#include "gromacs/awh/bias.h"

#include <cmath>

#include <memory>
#include <tuple>
#include <vector>

#include <gmock/gmock.h>
#include <gtest/gtest.h>

#include "gromacs/awh/correlationgrid.h"
#include "gromacs/awh/pointstate.h"
#include "gromacs/mdtypes/awh_params.h"
#include "gromacs/utility/stringutil.h"

#include "testutils/refdata.h"
#include "testutils/testasserts.h"

namespace gmx
{

namespace test
{

/*! \internal \brief
 * Struct that gathers all input for setting up and using a Bias
 */
struct AwhTestParameters
{
    AwhTestParameters() = default;
    //!Move constructor
    AwhTestParameters(AwhTestParameters &&o) noexcept : beta(o.beta), awhDimParams(o.awhDimParams),
                                                        awhBiasParams(o.awhBiasParams), awhParams(o.awhParams),
                                                        dimParams(std::move(o.dimParams))
    {
        awhBiasParams.dimParams = &awhDimParams;
        awhParams.awhBiasParams = &awhBiasParams;
    }
    double                 beta;          //!< 1/(kB*T)

    AwhDimParams           awhDimParams;  //!< Dimension parameters pointed to by \p awhBiasParams
    AwhBiasParams          awhBiasParams; //!< Bias parameters pointed to by \[ awhParams
    AwhParams              awhParams;     //!< AWH parameters, this is the struct to actually use

    std::vector<DimParams> dimParams;     //!< Dimension parameters for setting up Bias
};

//! Helper function to set up the C-style AWH parameters for the test
static AwhTestParameters getAwhTestParameters(int eawhgrowth,
                                              int eawhpotential)
{
    AwhTestParameters params;

    params.beta = 0.4;

    AwhDimParams &awhDimParams = params.awhDimParams;

    awhDimParams.period           = 0;
    awhDimParams.diffusion        = 0.1;
    awhDimParams.origin           = 0.5;
    awhDimParams.end              = 1.5;
    awhDimParams.coordValueInit   = awhDimParams.origin;
    awhDimParams.coverDiameter    = 0;

    AwhBiasParams &awhBiasParams = params.awhBiasParams;

    awhBiasParams.ndim                 = 1;
    awhBiasParams.dimParams            = &awhDimParams;
    awhBiasParams.eTarget              = eawhtargetCONSTANT;
    awhBiasParams.targetBetaScaling    = 0;
    awhBiasParams.targetCutoff         = 0;
    awhBiasParams.eGrowth              = eawhgrowth;
    awhBiasParams.bUserData            = FALSE;
    awhBiasParams.errorInitial         = 0.5/params.beta;
    awhBiasParams.shareGroup           = 0;
    awhBiasParams.equilibrateHistogram = FALSE;

    double                 convFactor  = 1;
    double                 k           = 1000;
    int64_t                seed        = 93471803;

    params.dimParams.emplace_back(convFactor, k, params.beta);

    AwhParams             &awhParams = params.awhParams;

    awhParams.numBias                    = 1;
    awhParams.awhBiasParams              = &awhBiasParams;
    awhParams.seed                       = seed;
    awhParams.nstOut                     = 0;
    awhParams.nstSampleCoord             = 1;
    awhParams.numSamplesUpdateFreeEnergy = 10;
    awhParams.ePotential                 = eawhpotential;
    awhParams.shareBiasMultisim          = FALSE;

    return params;
}

//! Database of 21 test coordinates that represent a trajectory */
const double g_coords[] = {
    0.62,
    0.70,
    0.68,
    0.80,
    0.93,
    0.87,
    1.16,
    1.14,
    0.95,
    0.89,
    0.91,
    0.86,
    0.88,
    0.79,
    0.75,
    0.82,
    0.74,
    0.70,
    0.68,
    0.71,
    0.73
};

//! Convenience typedef: growth type enum, potential type enum, disable update skips
typedef std::tuple<int, int, BiasParams::DisableUpdateSkips> BiasTestParameters;

/*! \brief Test fixture for testing Bias updates
 */
class BiasTest : public ::testing::TestWithParam<BiasTestParameters>
{
    public:
        //! Random seed for AWH MC sampling
        int64_t           seed_;

        //! Coordinates representing a trajectory in time
        std::vector<double>   coordinates_;
        //! The awh Bias
        std::unique_ptr<Bias> bias_;

        BiasTest() :
            coordinates_(std::begin(g_coords), std::end(g_coords))
        {
            /* We test all combinations of:
             *   eawhgrowth:
             *     eawhgrowthLINEAR:     final, normal update phase
             *     ewahgrowthEXP_LINEAR: intial phase, updated size is constant
             *   eawhpotential (should only affect the force output):
             *     eawhpotentialUMBRELLA:  MC on lambda (umbrella potential location)
             *     eawhpotentialCONVOLVED: MD on a convolved potential landscape
             *   disableUpdateSkips (should not affect the results):
             *     BiasParams::DisableUpdateSkips::yes: update the point state for every sample
             *     BiasParams::DisableUpdateSkips::no:  update the point state at an interval > 1 sample
             *
             * Note: It would be nice to explicitly check that eawhpotential
             *       and disableUpdateSkips do not affect the point state.
             *       But the reference data will also ensure this.
             */
            int eawhgrowth;
            int eawhpotential;
            BiasParams::DisableUpdateSkips disableUpdateSkips;
            std::tie(eawhgrowth, eawhpotential, disableUpdateSkips) = GetParam();

            /* Set up a basic AWH setup with a single, 1D bias with parameters
             * such that we can measure the effects of different parameters.
             * The idea is to, among other things, have part of the interval
             * not covered by samples.
             */
            const AwhTestParameters params = getAwhTestParameters(eawhgrowth, eawhpotential);

            seed_ = params.awhParams.seed;

            double mdTimeStep = 0.1;

            int    numSamples = coordinates_.size() - 1; // No sample taken at step 0
            GMX_RELEASE_ASSERT(numSamples % params.awhParams.numSamplesUpdateFreeEnergy == 0, "This test is intended to reproduce the situation when the might need to write output during a normal AWH run, therefore the number of samples should be a multiple of the free-energy update interval (but the test should also runs fine without this condition).");

            bias_ = std::make_unique<Bias>(-1, params.awhParams, params.awhBiasParams, params.dimParams, params.beta, mdTimeStep, 1, "", Bias::ThisRankWillDoIO::No, disableUpdateSkips);
        }
};

TEST_P(BiasTest, ForcesBiasPmf)
{
    gmx::test::TestReferenceData     data;
    gmx::test::TestReferenceChecker  checker(data.rootChecker());

    Bias                            &bias = *bias_;

    /* Make strings with the properties we expect to be different in the tests.
     * These also helps to interpret the reference data.
     */
    std::vector<std::string>         props;
    props.push_back(formatString("stage:           %s", bias.state().inInitialStage() ? "initial" : "final"));
    props.push_back(formatString("convolve forces: %s", bias.params().convolveForce ? "yes" : "no"));
    props.push_back(formatString("skip updates:    %s", bias.params().skipUpdates() ? "yes" : "no"));

    SCOPED_TRACE(gmx::formatString("%s, %s, %s", props[0].c_str(), props[1].c_str(), props[2].c_str()));

    std::vector<double> force, pot, potJump;

    double              coordMaxValue = 0;
    double              potentialJump = 0;
    int64_t             step          = 0;
    for (auto &coord : coordinates_)
    {
        coordMaxValue = std::max(coordMaxValue, std::abs(coord));

        awh_dvec                    coordValue = { coord, 0, 0, 0 };
        double                      potential  = 0;
        gmx::ArrayRef<const double> biasForce  =
            bias.calcForceAndUpdateBias(coordValue,
                                        &potential, &potentialJump,
                                        nullptr, nullptr, step, step, seed_,
                                        nullptr);

        force.push_back(biasForce[0]);
        pot.push_back(potential);
        potJump.push_back(potentialJump);

        step++;
    }

    /* When skipping updates, ensure all skipped updates are performed here.
     * This should result in the same bias state as at output in a normal run.
     */
    if (bias.params().skipUpdates())
    {
        bias.doSkippedUpdatesForAllPoints();
    }

    std::vector<double> pointBias, logPmfsum;
    for (auto &point : bias.state().points())
    {
        pointBias.push_back(point.bias());
        logPmfsum.push_back(point.logPmfSum());
    }

    /* The umbrella force is computed from the coordinate deviation.
     * In taking this deviation we lose a lot of precision, so we should
     * compare against k*max(coord) instead of the instantaneous force.
     */
    const double  kCoordMax = bias.dimParams()[0].k*coordMaxValue;

    constexpr int ulpTol    = 10;

    checker.checkSequence(props.begin(),     props.end(),     "Properties");
    checker.setDefaultTolerance(absoluteTolerance(kCoordMax*GMX_DOUBLE_EPS*ulpTol));
    checker.checkSequence(force.begin(),     force.end(),     "Force");
    checker.checkSequence(pot.begin(),       pot.end(),       "Potential");
    checker.checkSequence(potJump.begin(),   potJump.end(),   "PotentialJump");
    checker.setDefaultTolerance(relativeToleranceAsUlp(1.0, ulpTol));
    checker.checkSequence(pointBias.begin(), pointBias.end(), "PointBias");
    checker.checkSequence(logPmfsum.begin(), logPmfsum.end(), "PointLogPmfsum");
}

/* Scan initial/final phase, MC/convolved force and update skip (not) allowed
 * Both the convolving and skipping should not affect the bias and PMF.
 * It would be nice if the test would explicitly check for this.
 * Currently this is tested through identical reference data.
 */
INSTANTIATE_TEST_CASE_P(WithParameters, BiasTest,
                            ::testing::Combine(
                                    ::testing::Values(eawhgrowthLINEAR, eawhgrowthEXP_LINEAR),
                                    ::testing::Values(eawhpotentialUMBRELLA, eawhpotentialCONVOLVED),
                                    ::testing::Values(BiasParams::DisableUpdateSkips::yes, BiasParams::DisableUpdateSkips::no)));

// Test that we detect coverings and exit the initial stage at the correct step
TEST(BiasTest, DetectsCovering)
{
    const AwhTestParameters params       = getAwhTestParameters(eawhgrowthEXP_LINEAR, eawhpotentialCONVOLVED);
    const AwhDimParams     &awhDimParams = params.awhParams.awhBiasParams[0].dimParams[0];

    const double            mdTimeStep   = 0.1;

    Bias                    bias(-1, params.awhParams, params.awhBiasParams, params.dimParams, params.beta, mdTimeStep, 1, "", Bias::ThisRankWillDoIO::No);

    /* We use a trajectory of the sum of two sines to cover the reaction
     * coordinate range in a semi-realistic way. The period is 4*pi=12.57.
     * We get out of the initial stage after 4 coverings at step 300.
     */
    const int64_t     exitStepRef = 300;
    const double      midPoint    = 0.5*(awhDimParams.end + awhDimParams.origin);
    const double      halfWidth   = 0.5*(awhDimParams.end - awhDimParams.origin);

    bool              inInitialStage = bias.state().inInitialStage();
    /* Normally this loop exits at exitStepRef, but we extend with failure */
    int64_t           step;
    for (step = 0; step <= 2*exitStepRef; step++)
    {
        double   t     = step*mdTimeStep;
        double   coord = midPoint + halfWidth*(0.5*std::sin(t) + 0.55*std::sin(1.5*t));

        awh_dvec coordValue    = { coord, 0, 0, 0 };
        double   potential     = 0;
        double   potentialJump = 0;
        bias.calcForceAndUpdateBias(coordValue,
                                    &potential, &potentialJump,
                                    nullptr, nullptr,
                                    step, step, params.awhParams.seed, nullptr);

        inInitialStage = bias.state().inInitialStage();
        if (!inInitialStage)
        {
            break;
        }
    }

    EXPECT_EQ(false, inInitialStage);
    if (!inInitialStage)
    {
        EXPECT_EQ(exitStepRef, step);
    }
}

}  // namespace test
}  // namespace gmx