Update instructions in containers.rst

[gromacs.git] / src / gromacs / tables / quadraticsplinetable.cpp

/*
 * This file is part of the GROMACS molecular simulation package.
 *
 * Copyright (c) 2016,2017,2018,2019,2020, 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.
 */

/*! \internal \file
 * \brief
 * Implements classes for quadratic spline table functions
 *
 * \author Erik Lindahl <erik.lindahl@gmail.com>
 * \ingroup module_tables
 */
#include "gmxpre.h"

#include "quadraticsplinetable.h"

#include <cmath>

#include <algorithm>
#include <functional>
#include <initializer_list>
#include <utility>
#include <vector>

#include "gromacs/tables/tableinput.h"
#include "gromacs/utility/alignedallocator.h"
#include "gromacs/utility/arrayref.h"
#include "gromacs/utility/exceptions.h"
#include "gromacs/utility/real.h"

#include "splineutil.h"

namespace gmx
{

namespace
{

/*! \brief Construct the data for a single quadratic table from analytical functions
 *
 * \param[in]  function             Analytical function
 * \param[in]  derivative           Analytical derivative
 * \param[in]  range                Upper/lower limit of region to tabulate
 * \param[in]  spacing              Distance between table points
 * \param[out] functionTableData    Output table with function data
 * \param[out] derivativeTableData  OUtput table with (adjusted) derivative data
 */
void fillSingleQuadraticSplineTableData(const std::function<double(double)>& function,
                                        const std::function<double(double)>& derivative,
                                        const std::pair<real, real>&         range,
                                        double                               spacing,
                                        std::vector<real>*                   functionTableData,
                                        std::vector<real>*                   derivativeTableData)
{
    std::size_t endIndex = static_cast<std::size_t>(range.second / spacing + 2);

    functionTableData->resize(endIndex);
    derivativeTableData->resize(endIndex);

    double      maxMagnitude      = 0.0001 * GMX_REAL_MAX;
    bool        functionIsInRange = true;
    std::size_t lastIndexInRange  = endIndex - 1;

    for (int i = endIndex - 1; i >= 0; i--)
    {
        double x = i * spacing;
        double tmpFunctionValue;
        double tmpDerivativeValue;

        if (range.first > 0 && i == 0)
        {
            // Avoid x==0 if it is not in the range, since it can lead to
            // singularities even if the value for i==1 was within or required magnitude
            functionIsInRange = false;
        }

        if (functionIsInRange)
        {
            tmpFunctionValue = function(x);

            // Calculate third derivative term (2nd derivative of the derivative)
            // Make sure we stay in range. In practice this means we use one-sided
            // interpolation at the interval endpoints (indentical to an offset for 3-point formula)
            const double h = std::pow(GMX_DOUBLE_EPS, 0.25);
            double       y = std::min(std::max(x, range.first + h), range.second - h);
            double       thirdDerivativeValue =
                    (derivative(y + h) - 2.0 * derivative(y) + derivative(y - h)) / (h * h);

            tmpDerivativeValue = derivative(x) - spacing * spacing * thirdDerivativeValue / 12.0;

            if (std::abs(tmpFunctionValue) > maxMagnitude || std::abs(tmpDerivativeValue) > maxMagnitude)
            {
                functionIsInRange = false; // Once this happens, it never resets to true again
            }
        }

        if (functionIsInRange)
        {
            (*functionTableData)[i]   = tmpFunctionValue;
            (*derivativeTableData)[i] = tmpDerivativeValue;
            lastIndexInRange--;
        }
        else
        {
            // Once the function or derivative (more likely) has reached very large values,
            // we simply make a linear function from the last in-range value of the derivative.
            double lastIndexFunction   = (*functionTableData)[lastIndexInRange];
            double lastIndexDerivative = (*derivativeTableData)[lastIndexInRange];
            (*functionTableData)[i] =
                    lastIndexFunction + lastIndexDerivative * (i - lastIndexInRange) * spacing;
            (*derivativeTableData)[i] = lastIndexDerivative;
        }
    }
}


/*! \brief Construct the data for a single quadratic table from vector data
 *
 * \param[in]  function             Input vector with function data
 * \param[in]  derivative           Input vector with derivative data
 * \param[in]  inputSpacing         Distance between points in input vectors
 * \param[in]  range                Upper/lower limit of region to tabulate
 * \param[in]  spacing              Distance between table points
 * \param[out] functionTableData    Output table with function data
 * \param[out] derivativeTableData  OUtput table with (adjusted) derivative data
 */
void fillSingleQuadraticSplineTableData(ArrayRef<const double>       function,
                                        ArrayRef<const double>       derivative,
                                        double                       inputSpacing,
                                        const std::pair<real, real>& range,
                                        double                       spacing,
                                        std::vector<real>*           functionTableData,
                                        std::vector<real>*           derivativeTableData)
{
    std::size_t endIndex = static_cast<std::size_t>(range.second / spacing + 2);

    functionTableData->resize(endIndex);
    derivativeTableData->resize(endIndex);

    std::vector<double> thirdDerivative(internal::vectorSecondDerivative(derivative, inputSpacing));

    double maxMagnitude      = 0.0001 * GMX_REAL_MAX;
    bool   functionIsInRange = true;
    int    lastIndexInRange  = static_cast<int>(endIndex) - 1;

    for (int i = lastIndexInRange; i >= 0; i--)
    {
        double x = i * spacing;
        double tmpFunctionValue;
        double tmpDerivativeValue;

        if (range.first > 0 && i == 0)
        {
            // Avoid x==0 if it is not in the range, since it can lead to
            // singularities even if the value for i==1 was within or required magnitude
            functionIsInRange = false;
        }

        if (functionIsInRange)
        {
            // Step 1: Interpolate the function value at x from input table.
            double inputXTab  = x / inputSpacing;
            int    inputIndex = static_cast<std::size_t>(inputXTab);
            double inputEps   = inputXTab - inputIndex;

            // Linear interpolation of input derivative and third derivative
            double thirdDerivativeValue = (1.0 - inputEps) * thirdDerivative[inputIndex]
                                          + inputEps * thirdDerivative[inputIndex + 1];
            double derivativeValue =
                    (1.0 - inputEps) * derivative[inputIndex] + inputEps * derivative[inputIndex + 1];

            // Quadratic interpolation for function value
            tmpFunctionValue = function[inputIndex]
                               + 0.5 * (derivative[inputIndex] + derivativeValue) * inputEps * inputSpacing;
            tmpDerivativeValue = derivativeValue - spacing * spacing * thirdDerivativeValue / 12.0;

            if (std::abs(tmpFunctionValue) > maxMagnitude || std::abs(tmpDerivativeValue) > maxMagnitude)
            {
                functionIsInRange = false; // Once this happens, it never resets to true again
            }
        }

        if (functionIsInRange)
        {
            (*functionTableData)[i]   = tmpFunctionValue;
            (*derivativeTableData)[i] = tmpDerivativeValue;
            lastIndexInRange--;
        }
        else
        {
            // Once the function or derivative (more likely) has reached very large values,
            // we simply make a linear function from the last in-range value of the derivative.
            GMX_ASSERT(lastIndexInRange >= 0, "Array index is unexpectedly negative.");
            double lastIndexFunction   = (*functionTableData)[lastIndexInRange];
            double lastIndexDerivative = (*derivativeTableData)[lastIndexInRange];
            (*functionTableData)[i] =
                    lastIndexFunction + lastIndexDerivative * (i - lastIndexInRange) * spacing;
            (*derivativeTableData)[i] = lastIndexDerivative;
        }
    }
}

/*! \brief Create merged DDFZ vector from function & derivative data
 *
 *  \param functionTableData     Function values
 *  \param derivativeTableData   Derivative values. We have already subtracted the
 *                               small third derivative component when calling this
 *                               function, but in practice it is just an arbitrary
 *                               vector here.
 *  \param ddfzTableData         Vector four times longer, filled with
 *                               the derivative, the difference to the next derivative
 *                               point, the function value, and zero.
 *
 *  \throws If the vector lengths do not match.
 */
void fillDdfzTableData(const std::vector<real>& functionTableData,
                       const std::vector<real>& derivativeTableData,
                       std::vector<real>*       ddfzTableData)
{
    GMX_ASSERT(functionTableData.size() == derivativeTableData.size(),
               "Mismatching vector lengths");

    std::size_t points = functionTableData.size();

    ddfzTableData->resize(4 * points);

    for (std::size_t i = 0; i < points; i++)
    {
        (*ddfzTableData)[4 * i] = derivativeTableData[i];

        double nextDerivative = (i < functionTableData.size() - 1) ? derivativeTableData[i + 1] : 0.0;

        (*ddfzTableData)[4 * i + 1] = nextDerivative - derivativeTableData[i];
        (*ddfzTableData)[4 * i + 2] = functionTableData[i];
        (*ddfzTableData)[4 * i + 3] = 0.0;
    }
}

} // namespace


const real QuadraticSplineTable::defaultTolerance = 10.0 * GMX_FLOAT_EPS;


QuadraticSplineTable::QuadraticSplineTable(std::initializer_list<AnalyticalSplineTableInput> analyticalInputList,
                                           const std::pair<real, real>&                      range,
                                           real tolerance) :
    numFuncInTable_(analyticalInputList.size()),
    range_(range)
{
    // Sanity check on input values
    if (range_.first < 0.0 || (range_.second - range_.first) < 0.001)
    {
        GMX_THROW(InvalidInputError(
                "Range to tabulate cannot include negative values and must span at least 0.001"));
    }

    if (tolerance < GMX_REAL_EPS)
    {
        GMX_THROW(ToleranceError("Table tolerance cannot be smaller than GMX_REAL_EPS"));
    }

    double minQuotient = GMX_REAL_MAX;

    // loop over all functions to find smallest spacing
    for (const auto& thisFuncInput : analyticalInputList)
    {
        try
        {
            internal::throwUnlessDerivativeIsConsistentWithFunction(
                    thisFuncInput.function, thisFuncInput.derivative, range_);
        }
        catch (gmx::GromacsException& ex)
        {
            ex.prependContext("Error generating quadratic spline table for function '"
                              + thisFuncInput.desc + "'");
            throw;
        }
        // Calculate the required table spacing h. The error we make with linear interpolation
        // of the derivative will be described by the third-derivative correction term.
        // This means we can compute the required spacing as h = sqrt(12*tolerance*min(f'/f''')),
        // where f'/f''' is the first and third derivative of the function, respectively.

        double thisMinQuotient = internal::findSmallestQuotientOfFunctionAndSecondDerivative(
                thisFuncInput.derivative, range_);

        minQuotient = std::min(minQuotient, thisMinQuotient);
    }

    double spacing = std::sqrt(12.0 * tolerance * minQuotient);

    halfSpacing_ = 0.5 * spacing;
    tableScale_  = 1.0 / spacing;

    if (range_.second * tableScale_ > 1e6)
    {
        GMX_THROW(
                ToleranceError("Over a million points would be required for table; decrease range "
                               "or increase tolerance"));
    }

    // Loop over all tables again.
    // Here we create the actual table for each function, and then
    // combine them into a multiplexed table function.
    std::size_t funcIndex = 0;

    for (const auto& thisFuncInput : analyticalInputList)
    {
        try
        {
            std::vector<real> tmpFuncTableData;
            std::vector<real> tmpDerTableData;
            std::vector<real> tmpDdfzTableData;

            fillSingleQuadraticSplineTableData(thisFuncInput.function, thisFuncInput.derivative,
                                               range_, spacing, &tmpFuncTableData, &tmpDerTableData);

            fillDdfzTableData(tmpFuncTableData, tmpDerTableData, &tmpDdfzTableData);

            internal::fillMultiplexedTableData(tmpDerTableData, &derivativeMultiTableData_, 1,
                                               numFuncInTable_, funcIndex);

            internal::fillMultiplexedTableData(tmpDdfzTableData, &ddfzMultiTableData_, 4,
                                               numFuncInTable_, funcIndex);

            funcIndex++;
        }
        catch (gmx::GromacsException& ex)
        {
            ex.prependContext("Error generating quadratic spline table for function '"
                              + thisFuncInput.desc + "'");
            throw;
        }
    }
}


QuadraticSplineTable::QuadraticSplineTable(std::initializer_list<NumericalSplineTableInput> numericalInputList,
                                           const std::pair<real, real>&                     range,
                                           real tolerance) :
    numFuncInTable_(numericalInputList.size()),
    range_(range)
{
    // Sanity check on input values
    if (range.first < 0.0 || (range.second - range.first) < 0.001)
    {
        GMX_THROW(InvalidInputError(
                "Range to tabulate cannot include negative values and must span at least 0.001"));
    }

    if (tolerance < GMX_REAL_EPS)
    {
        GMX_THROW(ToleranceError("Table tolerance cannot be smaller than GMX_REAL_EPS"));
    }

    double minQuotient = GMX_REAL_MAX;

    // loop over all functions to find smallest spacing
    for (auto thisFuncInput : numericalInputList)
    {
        try
        {
            // We do not yet know what the margin is, but we need to test that we at least cover
            // the requested range before starting to calculate derivatives
            if (thisFuncInput.function.size() < range_.second / thisFuncInput.spacing + 1)
            {
                GMX_THROW(
                        InconsistentInputError("Table input vectors must cover requested range, "
                                               "and a margin beyond the upper endpoint"));
            }

            if (thisFuncInput.function.size() != thisFuncInput.derivative.size())
            {
                GMX_THROW(InconsistentInputError(
                        "Function and derivative vectors have different lengths"));
            }

            internal::throwUnlessDerivativeIsConsistentWithFunction(
                    thisFuncInput.function, thisFuncInput.derivative, thisFuncInput.spacing, range_);
        }
        catch (gmx::GromacsException& ex)
        {
            ex.prependContext("Error generating quadratic spline table for function '"
                              + thisFuncInput.desc + "'");
            throw;
        }
        // Calculate the required table spacing h. The error we make with linear interpolation
        // of the derivative will be described by the third-derivative correction term.
        // This means we can compute the required spacing as h = sqrt(12*tolerance*min(f'/f''')),
        // where f'/f''' is the first and third derivative of the function, respectively.
        // Since we already have an analytical form of the derivative, we reduce the numerical
        // errors by calculating the quotient of the function and second derivative of the
        // input-derivative-analytical function instead.

        double thisMinQuotient = internal::findSmallestQuotientOfFunctionAndSecondDerivative(
                thisFuncInput.derivative, thisFuncInput.spacing, range_);

        minQuotient = std::min(minQuotient, thisMinQuotient);
    }

    double spacing = std::sqrt(12.0 * tolerance * minQuotient);

    halfSpacing_ = 0.5 * spacing;
    tableScale_  = 1.0 / spacing;

    if (range_.second * tableScale_ > 1e6)
    {
        GMX_THROW(
                ToleranceError("Requested tolerance would require over a million points in table"));
    }

    // Loop over all tables again.
    // Here we create the actual table for each function, and then
    // combine them into a multiplexed table function.
    std::size_t funcIndex = 0;

    for (auto thisFuncInput : numericalInputList)
    {
        try
        {
            if (spacing < thisFuncInput.spacing)
            {
                GMX_THROW(
                        ToleranceError("Input vector spacing cannot achieve tolerance requested"));
            }

            std::vector<real> tmpFuncTableData;
            std::vector<real> tmpDerTableData;
            std::vector<real> tmpDdfzTableData;

            fillSingleQuadraticSplineTableData(thisFuncInput.function, thisFuncInput.derivative,
                                               thisFuncInput.spacing, range, spacing,
                                               &tmpFuncTableData, &tmpDerTableData);

            fillDdfzTableData(tmpFuncTableData, tmpDerTableData, &tmpDdfzTableData);

            internal::fillMultiplexedTableData(tmpDerTableData, &derivativeMultiTableData_, 1,
                                               numFuncInTable_, funcIndex);

            internal::fillMultiplexedTableData(tmpDdfzTableData, &ddfzMultiTableData_, 4,
                                               numFuncInTable_, funcIndex);

            funcIndex++;
        }
        catch (gmx::GromacsException& ex)
        {
            ex.prependContext("Error generating quadratic spline table for function '"
                              + thisFuncInput.desc + "'");
            throw;
        }
    }
}

} // namespace gmx