// ******************************************************************************
   //
   // Title:       Force Field X.
   // Description: Force Field X - Software for Molecular Biophysics.
   // Copyright:   Copyright (c) Michael J. Schnieders 2001-2025.
   //
   // This file is part of Force Field X.
   //
   // Force Field X is free software; you can redistribute it and/or modify it
  // under the terms of the GNU General Public License version 3 as published by
  // the Free Software Foundation.
  //
  // Force Field X 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 General Public License for more
  // details.
  //
  // You should have received a copy of the GNU General Public License along with
  // Force Field X; if not, write to the Free Software Foundation, Inc., 59 Temple
  // Place, Suite 330, Boston, MA 02111-1307 USA
  //
  // Linking this library statically or dynamically with other modules is making a
  // combined work based on this library. Thus, the terms and conditions of the
  // GNU General Public License cover the whole combination.
  //
  // As a special exception, the copyright holders of this library give you
  // permission to link this library with independent modules to produce an
  // executable, regardless of the license terms of these independent modules, and
  // to copy and distribute the resulting executable under terms of your choice,
  // provided that you also meet, for each linked independent module, the terms
  // and conditions of the license of that module. An independent module is a
  // module which is not derived from or based on this library. If you modify this
  // library, you may extend this exception to your version of the library, but
  // you are not obligated to do so. If you do not wish to do so, delete this
  // exception statement from your version.
  //
  // ******************************************************************************
  package ffx.algorithms.thermodynamics;
  
  import edu.rit.mp.DoubleBuf;
  import edu.rit.pj.Comm;
  import ffx.algorithms.AlgorithmListener;
  import ffx.algorithms.cli.DynamicsOptions;
  import ffx.algorithms.cli.LambdaParticleOptions;
  import ffx.algorithms.dynamics.Barostat;
  import ffx.algorithms.dynamics.integrators.Stochastic;
  import ffx.algorithms.optimize.Minimize;
  import ffx.crystal.Crystal;
  import ffx.crystal.CrystalPotential;
  import ffx.numerics.Potential;
  import ffx.numerics.integrate.DataSet;
  import ffx.numerics.integrate.DoublesDataSet;
  import ffx.numerics.integrate.Integrate1DNumeric;
  import ffx.numerics.integrate.Integrate1DNumeric.IntegrationType;
  import ffx.potential.MolecularAssembly;
  import ffx.potential.SystemState;
  import ffx.potential.Utilities;
  import ffx.potential.bonded.LambdaInterface;
  import ffx.potential.parsers.PDBFilter;
  import ffx.potential.parsers.SystemFilter;
  import ffx.potential.parsers.XYZFilter;
  import org.apache.commons.configuration2.CompositeConfiguration;
  import org.apache.commons.io.FilenameUtils;
  
  import javax.annotation.Nullable;
  import java.io.File;
  import java.io.PrintWriter;
  import java.util.ArrayList;
  import java.util.Arrays;
  import java.util.List;
  import java.util.Optional;
  import java.util.logging.Level;
  import java.util.logging.Logger;
  
  import static ffx.numerics.integrate.Integrate1DNumeric.IntegrationType.SIMPSONS;
  import static ffx.utilities.Constants.R;
  import static java.lang.String.format;
  import static java.lang.System.arraycopy;
  import static java.util.Arrays.fill;
  import static org.apache.commons.math3.util.FastMath.PI;
  import static org.apache.commons.math3.util.FastMath.abs;
  import static org.apache.commons.math3.util.FastMath.asin;
  import static org.apache.commons.math3.util.FastMath.exp;
  import static org.apache.commons.math3.util.FastMath.floor;
  import static org.apache.commons.math3.util.FastMath.sin;
  import static org.apache.commons.math3.util.FastMath.sqrt;
  
  /**
   * An implementation of the Orthogonal Space Tempering algorithm.
   *
   * <p>This only partially implements the LambdaInterface, since it does not return 2nd lambda
   * derivatives. The 2nd derivatives of the bias require 3rd derivatives of the underlying Hamiltonian
   * (not these derivatives are not needed for OST MD).
   *
   * @author Michael J. Schnieders, James Dama, Wei Yang and Pengyu Ren
   * @since 1.0
   */
  public class OrthogonalSpaceTempering implements CrystalPotential, LambdaInterface {
  
   private static final Logger logger = Logger.getLogger(OrthogonalSpaceTempering.class.getName());
   /**
    * The potential energy of the system.
    */
   protected final CrystalPotential potential;
   /**
    * Reference to the Barostat in use; if present this must be turned off during optimization.
    */
   protected final Barostat barostat;
   /**
    * The AlgorithmListener is called each time a count is added.
    */
   protected final AlgorithmListener algorithmListener;
   /**
    * Print detailed energy information.
    */
   protected final boolean print = false;
   /**
    * Number of variables.
    */
   protected final int nVariables;
   /**
    * A potential energy that implements the LambdaInterface.
    */
   private final LambdaInterface lambdaInterface;
   /**
    * List of additional Histograms this OST can switch to.
    */
   private final List<Histogram> allHistograms = new ArrayList<>();
   /**
    * Parameters to control saving local optimizations.
    */
   private final OptimizationParameters optimizationParameters;
   /**
    * Properties.
    */
   private final CompositeConfiguration properties;
   /**
    * The MolecularAssembly being simulated.
    */
   protected MolecularAssembly molecularAssembly;
   /**
    * Are FAST varying energy terms being computed, SLOW varying energy terms, or BOTH. OST is not
    * active when only FAST varying energy terms are being propagated.
    */
   protected Potential.STATE state = Potential.STATE.BOTH;
   /**
    * Force Field Potential Energy (i.e. with no bias terms added).
    */
   protected double forceFieldEnergy;
   /**
    * Contains counts for the OST bias.
    */
   private Histogram histogram;
   /**
    * Index of the current Histogram.
    */
   private int histogramIndex;
   /**
    * Flag to indicate that the Lambda particle should be propagated.
    */
   private boolean propagateLambda = true;
   /**
    * Mixed second partial derivative with respect to coordinates and lambda.
    */
   private final double[] dUdXdL;
 
   /**
    * Gradient array needed when the OST Energy method is called.
    */
   private final double[] tempGradient;
 
   /**
    * Partial derivative of the force field energy with respect to lambda.
    */
   private double dForceFieldEnergydL;
   /**
    * Magnitude of the 2D orthogonal space bias G(L,dE/dL).
    */
   private double gLdEdL = 0.0;
   /**
    * OST Bias energy.
    */
   private double biasEnergy;
   /**
    * Total system energy.
    */
   private double totalEnergy;
   /**
    * Total partial derivative of the potential (U) being sampled with respect to lambda.
    */
   private double dUdLambda;
   /**
    * Second partial derivative of the potential being sampled with respect to lambda.
    */
   private double d2UdL2;
   /**
    * If true, values of (lambda, dU/dL) that have not been observed are rejected.
    */
   private boolean hardWallConstraint = false;
 
   private final DynamicsOptions dynamicsOptions;
 
   private final LambdaParticleOptions lambdaParticleOptions;
 
   /**
    * OST Constructor.
    *
    * @param lambdaInterface       defines Lambda and dU/dL.
    * @param potential             defines the Potential energy.
    * @param histogramData         contains histogram restart data.
    * @param lambdaData            contains lambda restart data.
    * @param properties            defines System properties.
    * @param dynamicsOptions       defines molecular dynamics parameters.
    * @param lambdaParticleOptions defines lambda particle parameters.
    * @param algorithmListener     the AlgorithmListener to be notified of progress.
    */
   public OrthogonalSpaceTempering(LambdaInterface lambdaInterface, CrystalPotential potential,
                                   HistogramData histogramData, LambdaData lambdaData, CompositeConfiguration properties,
                                   DynamicsOptions dynamicsOptions, LambdaParticleOptions lambdaParticleOptions,
                                   AlgorithmListener algorithmListener) {
     this.lambdaInterface = lambdaInterface;
     this.potential = potential;
     this.properties = properties;
     this.dynamicsOptions = dynamicsOptions;
     this.lambdaParticleOptions = lambdaParticleOptions;
     this.algorithmListener = algorithmListener;
     nVariables = potential.getNumberOfVariables();
 
     if (potential instanceof Barostat) {
       barostat = (Barostat) potential;
     } else {
       barostat = null;
     }
 
     dUdXdL = new double[nVariables];
     tempGradient = new double[nVariables];
 
     // Init the Histogram.
     histogram = new Histogram(properties, histogramData, lambdaData);
     histogramIndex = 0;
     allHistograms.add(histogram);
 
     // Configure optimization parameters.
     optimizationParameters = new OptimizationParameters(properties);
   }
 
   /**
    * Add an alternate Histogram this OST can use.
    *
    * @param histogramData Settings to use for the new Histogram.
    */
   public void addHistogram(HistogramData histogramData, LambdaData lambdaData) {
     Histogram newHisto = new Histogram(properties, histogramData, lambdaData);
     histogramData.asynchronous = this.histogram.hd.asynchronous;
     allHistograms.add(newHisto);
   }
 
   /**
    * {@inheritDoc}
    */
   @Override
   public boolean dEdLZeroAtEnds() {
     return false;
   }
 
   /**
    * {@inheritDoc}
    */
   @Override
   public boolean destroy() {
     // Shut down the CountReceiveThread.
     histogram.destroy();
     return potential.destroy();
   }
 
   /**
    * Compute the force field + bias energy.
    */
   public double energy(double[] x) {
 
     // OST is propagated with the slowly varying terms.
     if (state == Potential.STATE.FAST) {
       forceFieldEnergy = potential.energy(x);
       return forceFieldEnergy;
     }
 
     // Stochastic dynamics pre-force propagation.
     if (propagateLambda) {
       histogram.stochasticPreForce();
     }
 
     // We have to compute the energy and gradient, since we need dU/dL to be computed.
     fill(tempGradient, 0.0);
     forceFieldEnergy = potential.energyAndGradient(x, tempGradient);
 
     dForceFieldEnergydL = lambdaInterface.getdEdL();
     d2UdL2 = lambdaInterface.getd2EdL2();
     int lambdaBin = histogram.indexForLambda();
     dUdLambda = dForceFieldEnergydL;
 
     gLdEdL = 0.0;
     double bias1D;
     // Update the free energy difference.
     histogram.updateFreeEnergyDifference(false, false);
     if (histogram.hd.metaDynamics) {
       bias1D = histogram.energyAndGradientMeta(true);
     } else {
       // Calculate recursion kernel G(L, F_L) and its derivatives with respect to L and F_L.
       if (histogram.hd.biasMag > 0.0) {
         double[] chainRule = new double[2];
         gLdEdL = histogram.energyAndGradient2D(dUdLambda, chainRule);
         double dGdLambda = chainRule[0];
         double dGdFLambda = chainRule[1];
         dUdLambda += dGdLambda + dGdFLambda * d2UdL2;
       }
 
       // Compute the energy and gradient for the recursion worker at F(L) using interpolation.
       bias1D = histogram.energyAndGradient1D(true);
     }
 
     // The total bias energy is the sum of the 1D and 2D terms.
     biasEnergy = bias1D + gLdEdL;
 
     if (print) {
       logger.info(format(" Bias Energy        %16.8f", biasEnergy));
       logger.info(format(" %s %16.8f  (Kcal/mole)", "OST Potential    ", forceFieldEnergy + biasEnergy));
     }
 
     if (propagateLambda) {
       histogram.stochasticPostForce();
 
       histogram.ld.stepsTaken++;
       long energyCount = histogram.ld.stepsTaken;
 
       // Log the current Lambda state.
       int printFrequency = dynamicsOptions.getReportFrequency(100);
       if (energyCount % printFrequency == 0) {
         double dBdL = dUdLambda - dForceFieldEnergydL;
         double lambda = histogram.ld.lambda;
         int lambdaBins = histogram.hd.getLambdaBins();
         if (lambdaBins < 1000) {
           logger.info(format(" L=%6.4f (%3d) F_LU=%10.4f F_LB=%10.4f F_L=%10.4f V_L=%10.4f", lambda,
               lambdaBin, dForceFieldEnergydL, dBdL, dUdLambda, histogram.ld.thetaVelocity));
         } else {
           logger.info(format(" L=%6.4f (%4d) F_LU=%10.4f F_LB=%10.4f F_L=%10.4f V_L=%10.4f", lambda,
               lambdaBin, dForceFieldEnergydL, dBdL, dUdLambda, histogram.ld.thetaVelocity));
         }
       }
 
       // Metadynamics grid counts (every 'countInterval' steps).
       if (energyCount % histogram.hd.countInterval == 0) {
         histogram.addBias(dForceFieldEnergydL);
 
         // Locally optimize the current state.
         if (optimizationParameters.doOptimization) {
           optimizationParameters.optimize(forceFieldEnergy, x, tempGradient);
         }
 
         if (algorithmListener != null) {
           algorithmListener.algorithmUpdate(molecularAssembly);
         }
       }
     }
 
     totalEnergy = forceFieldEnergy + biasEnergy;
 
     return totalEnergy;
   }
 
   /**
    * {@inheritDoc}
    */
   @Override
   public double energyAndGradient(double[] x, double[] gradient) {
 
     // Stochastic dynamics pre-force propagation.
     if (state != STATE.FAST && propagateLambda) {
       histogram.stochasticPreForce();
     }
 
     // Compute the force field energy and gradient.
     forceFieldEnergy = potential.energyAndGradient(x, gradient);
 
     // OST is propagated with the slowly varying terms.
     if (state == STATE.FAST) {
       return forceFieldEnergy;
     }
 
     // dU/dL is the partial derivative of the force field energy with respect to lambda.
     dForceFieldEnergydL = lambdaInterface.getdEdL();
     dUdLambda = dForceFieldEnergydL;
     d2UdL2 = lambdaInterface.getd2EdL2();
 
     gLdEdL = 0.0;
     double bias1D;
     // Update the free energy difference.
     histogram.updateFreeEnergyDifference(false, false);
     if (histogram.hd.metaDynamics) {
       bias1D = histogram.energyAndGradientMeta(true);
     } else {
       if (histogram.hd.biasMag > 0) {
         double[] chainRule = new double[2];
         gLdEdL = histogram.energyAndGradient2D(dUdLambda, chainRule);
         double dGdLambda = chainRule[0];
         double dGdFLambda = chainRule[1];
 
         // Lambda gradient due to recursion kernel G(L, F_L).
         dUdLambda += dGdLambda + dGdFLambda * d2UdL2;
 
         // Cartesian coordinate gradient due to recursion kernel G(L, F_L).
         fill(dUdXdL, 0.0);
         lambdaInterface.getdEdXdL(dUdXdL);
         for (int i = 0; i < nVariables; i++) {
           gradient[i] += dGdFLambda * dUdXdL[i];
         }
       }
 
       // Compute the energy and gradient for the recursion worker at F(L) using interpolation.
       bias1D = histogram.energyAndGradient1D(true);
     }
 
     // The total bias is the sum of 1D and 2D terms.
     biasEnergy = bias1D + gLdEdL;
 
     if (print) {
       logger.info(format(" %s %16.8f", "Bias Energy       ", biasEnergy));
       logger.info(format(" %s %16.8f  %s", "OST Potential    ", forceFieldEnergy + biasEnergy, "(Kcal/mole)"));
     }
 
     if (propagateLambda) {
       histogram.stochasticPostForce();
 
       histogram.ld.stepsTaken++;
       long energyCount = histogram.ld.stepsTaken;
 
       // Log the current Lambda state.
       int printFrequency = dynamicsOptions.getReportFrequency(100);
       if (energyCount % printFrequency == 0) {
         double dBdL = dUdLambda - dForceFieldEnergydL;
         int lambdaBin = histogram.indexForLambda();
         double lambda = histogram.ld.lambda;
         int lambdaBins = histogram.hd.getLambdaBins();
         if (lambdaBins < 1000) {
           logger.info(format(" L=%6.4f (%3d) F_LU=%10.4f F_LB=%10.4f F_L=%10.4f V_L=%10.4f", lambda,
               lambdaBin, dForceFieldEnergydL, dBdL, dUdLambda, histogram.ld.thetaVelocity));
         } else {
           logger.info(format(" L=%6.4f (%4d) F_LU=%10.4f F_LB=%10.4f F_L=%10.4f V_L=%10.4f", lambda,
               lambdaBin, dForceFieldEnergydL, dBdL, dUdLambda, histogram.ld.thetaVelocity));
         }
       }
 
       // Metadynamics grid counts (every 'countInterval' steps).
       if (energyCount % histogram.hd.countInterval == 0) {
         histogram.addBias(dForceFieldEnergydL);
 
         // Locally optimize the current state.
         if (optimizationParameters.doOptimization) {
           optimizationParameters.optimize(forceFieldEnergy, x, gradient);
         }
 
         if (algorithmListener != null) {
           algorithmListener.algorithmUpdate(molecularAssembly);
         }
       }
     }
 
     totalEnergy = forceFieldEnergy + biasEnergy;
 
     return totalEnergy;
   }
 
   /**
    * {@inheritDoc}
    */
   @Override
   public double[] getAcceleration(double[] acceleration) {
     return potential.getAcceleration(acceleration);
   }
 
   public Histogram[] getAllHistograms() {
     int nHisto = allHistograms.size();
     Histogram[] ret = new Histogram[nHisto];
     ret = allHistograms.toArray(ret);
     return ret;
   }
 
   /**
    * {@inheritDoc}
    */
   @Override
   public double[] getCoordinates(double[] doubles) {
     return potential.getCoordinates(doubles);
   }
 
   /**
    * {@inheritDoc}
    */
   @Override
   public void setCoordinates(double[] doubles) {
     potential.setCoordinates(doubles);
   }
 
   /**
    * {@inheritDoc}
    */
   @Override
   public Crystal getCrystal() {
     return potential.getCrystal();
   }
 
   /**
    * {@inheritDoc}
    */
   @Override
   public void setCrystal(Crystal crystal) {
     potential.setCrystal(crystal);
   }
 
   /**
    * Returns the number of energy evaluations performed by this OST, including those picked up in the
    * lambda file.
    *
    * @return Number of energy steps taken by this walker.
    */
   public long getEnergyCount() {
     return histogram.ld.stepsTaken;
   }
 
   /**
    * {@inheritDoc}
    */
   @Override
   public Potential.STATE getEnergyTermState() {
     return state;
   }
 
   /**
    * {@inheritDoc}
    */
   @Override
   public void setEnergyTermState(Potential.STATE state) {
     this.state = state;
     potential.setEnergyTermState(state);
   }
 
   /**
    * Getter for the field <code>forceFieldEnergy</code>.
    *
    * @return a double.
    */
   public double getForceFieldEnergy() {
     return forceFieldEnergy;
   }
 
   /**
    * Return the current 2D Histogram of counts.
    *
    * @return the Histogram.
    */
   public Histogram getHistogram() {
     return histogram;
   }
 
   /**
    * Getter for the field <code>lambda</code>.
    *
    * @return a double.
    */
   public double getLambda() {
     return histogram.ld.lambda;
   }
 
   /**
    * Setter for the field <code>lambda</code>.
    *
    * @param lambda a double.
    */
   public void setLambda(double lambda) {
     if (histogram != null) {
       lambda = histogram.mapLambda(lambda);
     } else {
       logger.warning(" OrthogonalSpaceTempering.setLambda was called before histogram constructed!");
       logger.info(Utilities.stackTraceToString(new RuntimeException()));
     }
     lambdaInterface.setLambda(lambda);
     histogram.ld.lambda = lambda;
     histogram.ld.theta = asin(sqrt(lambda));
   }
 
   /**
    * {@inheritDoc}
    */
   @Override
   public double[] getMass() {
     return potential.getMass();
   }
 
   /**
    * {@inheritDoc}
    */
   @Override
   public int getNumberOfVariables() {
     return potential.getNumberOfVariables();
   }
 
   /**
    * Return the OST optimization information.
    *
    * @return The OST optimization parameters.
    */
   public OptimizationParameters getOptimizationParameters() {
     return optimizationParameters;
   }
 
   /**
    * getPotentialEnergy.
    *
    * @return a {@link ffx.numerics.Potential} object.
    */
   public Potential getPotentialEnergy() {
     return potential;
   }
 
   /**
    * {@inheritDoc}
    */
   @Override
   public double[] getPreviousAcceleration(double[] previousAcceleration) {
     return potential.getPreviousAcceleration(previousAcceleration);
   }
 
   /**
    * {@inheritDoc}
    */
   @Override
   public double[] getScaling() {
     return potential.getScaling();
   }
 
   /**
    * {@inheritDoc}
    */
   @Override
   public void setScaling(double[] scaling) {
     potential.setScaling(scaling);
   }
 
   /**
    * {@inheritDoc}
    */
   @Override
   public double getTotalEnergy() {
     return totalEnergy;
   }
 
   /**
    * getTotaldEdLambda.
    *
    * @return a double.
    */
   public double getTotaldEdLambda() {
     return dUdLambda;
   }
 
   /**
    * {@inheritDoc}
    *
    * <p>Return a reference to each variables type.
    */
   @Override
   public Potential.VARIABLE_TYPE[] getVariableTypes() {
     return potential.getVariableTypes();
   }
 
   /**
    * {@inheritDoc}
    */
   @Override
   public double[] getVelocity(double[] velocity) {
     return potential.getVelocity(velocity);
   }
 
   /**
    * {@inheritDoc}
    */
   @Override
   public double getd2EdL2() {
     throw new UnsupportedOperationException(
         " Second derivatives of the bias are not implemented, as they require third derivatives of the potential.");
   }
 
   /**
    * {@inheritDoc}
    */
   @Override
   public double getdEdL() {
     return getTotaldEdLambda();
   }
 
   /**
    * {@inheritDoc}
    */
   @Override
   public void getdEdXdL(double[] gradient) {
     throw new UnsupportedOperationException(
         " Second derivatives of the bias are not implemented, as they require third derivatives of the potential.");
   }
 
   // TODO: Delete method when debugging of RepexOST is done.
   public void logOutputFiles(int index) {
     logger.info(format(" OST: Lambda file %s, histogram %s", histogram.ld.getLambdaFile(),
         allHistograms.get(index).hd.getHistogramFile()));
   }
 
   /**
    * {@inheritDoc}
    */
   @Override
   public void setAcceleration(double[] acceleration) {
     potential.setAcceleration(acceleration);
   }
 
   /**
    * If this flag is true, (lambda, dU/dL) Monte Carlo samples that have no weight in the Histogram
    * are rejected.
    *
    * @param hardWallConstraint If true, MC samples outside the current range are rejected.
    */
   public void setHardWallConstraint(boolean hardWallConstraint) {
     this.hardWallConstraint = hardWallConstraint;
   }
 
   public void setMolecularAssembly(MolecularAssembly molecularAssembly) {
     this.molecularAssembly = molecularAssembly;
   }
 
   /**
    * {@inheritDoc}
    */
   @Override
   public void setPreviousAcceleration(double[] previousAcceleration) {
     potential.setPreviousAcceleration(previousAcceleration);
   }
 
   /**
    * Indicate if the Lambda extended system particle should be propagated using Langevin dynamics.
    *
    * @param propagateLambda If true, Lambda will be propagated using Langevin dynamics.
    */
   public void setPropagateLambda(boolean propagateLambda) {
     this.propagateLambda = propagateLambda;
   }
 
   /**
    * If true, the Lambda extended system particle is propagated using Langevin dynamics.
    */
   public boolean getPropagateLambda() {
     return propagateLambda;
   }
 
   /**
    * {@inheritDoc}
    */
   @Override
   public void setVelocity(double[] velocity) {
     potential.setVelocity(velocity);
   }
 
   /**
    * Switch to an alternate Histogram.
    *
    * @param index Index of the Histogram to use.
    */
   public void switchHistogram(int index) {
     histogramIndex = index;
     histogram = allHistograms.get(histogramIndex);
     logger.info(" OST switching to histogram " + histogramIndex);
   }
 
   @Override
   public void writeAdditionalRestartInfo(boolean recursive) {
     histogram.writeRestart();
     if (recursive) {
       potential.writeAdditionalRestartInfo(recursive);
     }
   }
 
   double getLambdaWriteOut() {
     return histogram.hd.resetHistogramAtLambda;
   }
 
   /**
    * getForceFielddEdL.
    *
    * @return a double.
    */
   double getForceFielddEdL() {
     return dForceFieldEnergydL;
   }
 
   /**
    * Getter for the field <code>biasEnergy</code>.
    *
    * @return a double.
    */
   double getBiasEnergy() {
     return biasEnergy;
   }
 
   /**
    * If the dUdLHardWall flag is set to true, this method will return false if the (lambda, dU/dL)
    * sample is has not been seen.
    *
    * @param lambda The proposed lambda value.
    * @param dUdL   The proposed dU/dL value.
    * @return Returns false only if the dUdLHardWall flag is true, and the (lambda, dU/dL) sample has
    * not been seen.
    */
   boolean insideHardWallConstraint(double lambda, double dUdL) {
     if (hardWallConstraint) {
       double weight = histogram.getRecursionKernelValue(lambda, dUdL);
       return weight > 0.0;
     }
     return true;
   }
 
   /**
    * Parameters for running local optimizations during OST sampling.
    */
   public class OptimizationParameters {
 
     /**
      * Flag to turn on OST optimization.
      *
      * <p>The default doOptimization = false.
      */
     private boolean doOptimization = false;
     /**
      * Reset unit cell parameters, molecular orientation and translation.
      */
     private final boolean doUnitCellReset;
     /**
      * OST optimization only runs if Lambda is greater than the lambdaCutoff.
      *
      * <p>The default lambdaCutoff = 0.8.
      */
     private final double lambdaCutoff;
     /**
      * The lowest energy found via optimizations.
      *
      * <p>The optimumEnergy is initially set to Double.MAX_VALUE.
      */
     private double optimumEnergy = Double.MAX_VALUE;
     /**
      * The OST optimization frequency
      *
      * <p>The default is once every 10,000 steps.
      */
     private final int frequency;
     /**
      * The OST optimization convergence criteria.
      *
      * <p>The default eps = 0.1.
      */
     private final double eps;
     /**
      * The OST tolerance when checking for equal energy after coordinate reversion.
      *
      * <p>The default is 1.0e-8 kcal/mol.
      */
     private final double tolerance;
     /**
      * The OST optimization energy window.
      *
      * <p>The default is 4.0 kcal/mol, which is convenient for small organic crystals.
      */
     private final double energyWindow;
     /**
      * Holds the lowest potential energy coordinates.
      */
     private double[] optimumCoords;
     /**
      * File instance used for saving optimized structures.
      */
     private File optimizationFile;
     /**
      * SystemFilter used to save optimized structures.
      */
     private SystemFilter optimizationFilter;
 
     /**
      * Empty constructor.
      */
     OptimizationParameters(CompositeConfiguration properties) {
       energyWindow = properties.getDouble("ost-opt-energy-window", 10.0);
       eps = properties.getDouble("ost-opt-eps", 0.1);
       tolerance = properties.getDouble("ost-opt-tolerance", 1.0e-8);
       frequency = properties.getInt("ost-opt-frequency", 10000);
       lambdaCutoff = properties.getDouble("ost-opt-lambda-cutoff", 0.8);
       doUnitCellReset = properties.getBoolean("ost-opt-unitcell-reset", false);
     }
 
     private void log() {
       logger.info("\n Optimization Parameters");
       logger.info(format("  Energy Window:                  %6.3f (kcal/mol)", energyWindow));
       logger.info(format("  EPS:                            %6.4f RMS (kcal/mol/Å)", eps));
       logger.info(format("  Tolerance:                      %6.4f (kcal/mol)", tolerance));
       logger.info(format("  Frequency:                      %6d (steps)", frequency));
       logger.info(format("  Lambda Cutoff:                  %6.4f", lambdaCutoff));
       logger.info(format("  Unit Cell Reset:                %6B", doUnitCellReset));
       logger.info(format("  File:                           %s", optimizationFile.getName()));
     }
 
     /**
      * The coordinates of the lowest energy structure found during optimization.
      *
      * @return the coordinates of the lowest energy structure found during optimization.
      */
     public double[] getOptimumCoordinates() {
       if (optimumEnergy < Double.MAX_VALUE) {
         return optimumCoords;
       } else {
         logger.info(
             "Lambda optimization cutoff was not reached. Try increasing the number of timesteps.");
         return null;
       }
     }
 
     /**
      * getOptimumEnergy.
      *
      * @return a double.
      */
     public double getOptimumEnergy() {
       if (optimumEnergy == Double.MAX_VALUE) {
         logger.info(
             "Lambda optimization cutoff was not reached. Try increasing the number of timesteps.");
       }
       return optimumEnergy;
     }
 
     /**
      * Run a local optimization.
      *
      * @param e        Current energy.
      * @param x        Current atomic coordinates.
      * @param gradient Work array for collecting the gradient.
      */
     public void optimize(double e, double[] x, @Nullable double[] gradient) {
 
       // Return if the optimization flag is not set, or if lambda is not beyond the cutoff.
       double lambda = histogram.ld.lambda;
       long stepsTaken = histogram.ld.stepsTaken;
       if (doOptimization && lambda > lambdaCutoff && stepsTaken % frequency == 0) {
         if (gradient == null) {
           gradient = new double[x.length];
         }
       } else {
         return;
       }
 
       logger.info(format("\n OST Minimization (Step %d)", stepsTaken));
 
       // Set the underlying Potential's Lambda value to 1.0.
       lambdaInterface.setLambda(1.0);
 
       // Use all energy terms.
       potential.setEnergyTermState(Potential.STATE.BOTH);
 
       // Turn off the Barostat.
       boolean origBaroActive = true;
       if (barostat != null) {
         origBaroActive = barostat.isActive();
         barostat.setActive(false);
       }
 
       // Optimize the system.
       try {
         double startingEnergy = potential.energy(x);
         Minimize minimize = new Minimize(molecularAssembly, potential, algorithmListener);
         minimize.minimize(eps);
         // Collect the minimum energy.
         double minEnergy = potential.getTotalEnergy();
         // Check for a new minimum within an energy window of the lowest energy structure found.
         if (minEnergy < optimumEnergy + energyWindow) {
           if (minEnergy < optimumEnergy) {
             optimumEnergy = minEnergy;
           }
           int n = potential.getNumberOfVariables();
           optimumCoords = new double[n];
           optimumCoords = potential.getCoordinates(optimumCoords);
           double mass = molecularAssembly.getMass();
           Crystal crystal = molecularAssembly.getCrystal();
           double density = crystal.getDensity(mass);
           optimizationFilter.writeFile(optimizationFile, true);
           Crystal uc = crystal.getUnitCell();
           logger.info(format(" Minimum: %12.6f %s (%12.6f g/cc) optimized from %12.6f at step %d.",
               minEnergy, uc.toShortString(), density, startingEnergy, stepsTaken));
         }
       } catch (Exception ex) {
         String message = ex.getMessage();
         logger.info(
             format(" Exception minimizing coordinates at lambda=%8.6f\n %s.", lambda, message));
         logger.info(" Sampling will continue.");
       }
 
       // Set the underlying Potential's Lambda value back to current lambda value.
       lambdaInterface.setLambda(lambda);
 
       // Reset the Potential State
       potential.setEnergyTermState(state);
 
       // Reset the Barostat
       if (barostat != null) {
         barostat.setActive(origBaroActive);
       }
 
       if (doUnitCellReset) {
         logger.info("\n Resetting Unit Cell");
         double mass = molecularAssembly.getMass();
         double density = molecularAssembly.getCrystal().getDensity(mass);
         molecularAssembly.applyRandomDensity(density);
         molecularAssembly.applyRandomSymOp(0.0);
         lambda = 0.0;
         lambdaInterface.setLambda(lambda);
       } else {
         // Revert to the coordinates and gradient prior to optimization.
         double eCheck = potential.energyAndGradient(x, gradient);
 
         if (abs(eCheck - e) > tolerance) {
           logger.warning(
               format(" Optimization could not revert coordinates %16.8f vs. %16.8f.", e, eCheck));
         }
       }
     }
 
     /**
      * setOptimization.
      *
      * @param doOptimization    a boolean.
      * @param molecularAssembly a {@link ffx.potential.MolecularAssembly} object.
      */
     public void setOptimization(boolean doOptimization, MolecularAssembly molecularAssembly) {
       this.doOptimization = doOptimization;
       OrthogonalSpaceTempering.this.molecularAssembly = molecularAssembly;
       File file = molecularAssembly.getFile();
       String fileName = FilenameUtils.removeExtension(file.getAbsolutePath());
       String ext = FilenameUtils.getExtension(file.getAbsolutePath());
       if (optimizationFilter == null) {
         if (ext.toUpperCase().contains("XYZ")) {
           optimizationFile = new File(fileName + "_opt.arc");
           optimizationFilter = new XYZFilter(optimizationFile, molecularAssembly,
               molecularAssembly.getForceField(), molecularAssembly.getProperties());
         } else {
           optimizationFile = new File(fileName + "_opt.pdb");
           PDBFilter pdbFilter = new PDBFilter(optimizationFile, molecularAssembly,
               molecularAssembly.getForceField(), molecularAssembly.getProperties());
           int models = pdbFilter.countNumModels();
           pdbFilter.setModelNumbering(models);
           optimizationFilter = pdbFilter;
         }
       }
 
       if (this.doOptimization) {
         log();
       }
     }
   }
 
   /**
    * Store and operate on a 2D Histogram of (Lambda, dU/dL) observations to produce an OST bias.
    *
    * @author Michael J. Schnieders
    * @since 1.0
    */
   public class Histogram {
 
     /**
      * If a 2D bias is in use (i.e. lambda-bias-cutoff > 0), do not normalize bias height.
      */
     private static final double TWO_D_NORMALIZATION = 1.0;
     /**
      * If a 1D bias is in use (i.e. lambda-bias-cutoff == 0), normalize bias height to deposit the same
      * volume of bias.
      */
     private static final double ONE_D_NORMALIZATION = Math.sqrt(2.0 * Math.PI);
     /**
      * If the real bias magnitude is 0 (for no 2D bias), temporarily set biasMag to this value to calculate the ensemble average dU/dL.
      *
      * <p>Any value that does not overflow/underflow double precision summations of the count matrix
      * will give identical results. For example, values of 1.0e-20, 1.0 and 1.0e20 were tested and
      * found to give identical results.
      *
      * <p>Thus, a value of 1.0 is good choice.
      */
     private static final double PSEUDO_BIAS_MAGNITUDE = 1.0;
     /**
      * Parallel Java world communicator.
      */
     protected final Comm world;
     /**
      * Rank of this process.
      */
     protected final int rank;
     /**
      * 1D PMF with respect to lambda F(L).
      */
     final double[] ensembleAveragedUdL;
     /**
      * This deltaT is used to determine the tempering weight as described below for the
      * temperingWeight variable.
      *
      * <p>deltaT = temperingFactor * kB * T.
      */
     private final double deltaT;
     /**
      * The integration algorithm used for thermodynamic integration.
      */
     private final IntegrationType integrationType;
     /**
      * Send OST counts asynchronously.
      */
     private final SendAsynchronous sendAsynchronous;
     /**
      * Send OST counts synchronously.
      */
     private final SendSynchronous sendSynchronous;
     /**
      * The Dama et al. transition-tempering weight. The initial temperingWeight = 1.0,
      * and more generally temperingWeight = exp(-biasHeight/deltaT)
      */
     private double temperingWeight = 1.0;
     /**
      * If the recursion kernel becomes too large or too small for some combinations of (Lambda,
      * dU/dL), then its statistical weight = exp(kernel * beta) cannot be represented by a double
      * value.
      */
     private double[] kernelValues;
     /**
      * Most recent lambda values for each Walker.
      */
     private double lastReceivedLambda;
     /**
      * Most recent dU/dL value for each walker.
      */
     private double lastReceiveddUdL;
     private final boolean spreadBias;
     /**
      * The HistogramData object contains the parameters for the Histogram that are saved to disk.
      */
     final HistogramData hd;
     /**
      * The LambdaData object contains the parameters for the Lambda particle that are saved to disk.
      */
     final LambdaData ld;
     /**
      * The Stochastic integrator used to propagate the lambda particle.
      */
     private final Stochastic stochastic;
     /**
      * The state of the lambda particle.
      */
     private final SystemState lambdaState;
     /**
      * Number of hills when the last free energy estimate was made.
      */
     private int currentNumberOfHills = 0;
     /**
      * The current free energy difference.
      */
     private double currentFreeEnergyDifference = 0.0;
 
     /**
      * Histogram constructor.
      *
      * @param properties    a CompositeConfiguration used to configure the Histogram.
      * @param histogramData An object containing the values this Histogram will be set to.
      */
     Histogram(CompositeConfiguration properties, HistogramData histogramData, LambdaData lambdaData) {
       hd = histogramData;
       ld = lambdaData;
 
       /*
       double gaussNormalization;
       if (hd.lambdaBiasCutoff == 0 && !hd.wasHistogramRead()) {
         gaussNormalization = ONE_D_NORMALIZATION;
         logger.info(format(" Bias magnitude multiplied by a factor of %.4f "
             + "sqrt(2*pi) to match 1D Gaussian volume to 2D Gaussian volume.", gaussNormalization));
       } else {
         gaussNormalization = TWO_D_NORMALIZATION;
       }
       biasMag = settings.getBiasMag() * gaussNormalization;
        */
 
       deltaT = hd.temperingFactor * R * dynamicsOptions.getTemperature();
 
       // Allocate space to compute the <dU/dL>
       ensembleAveragedUdL = new double[hd.getLambdaBins()];
 
       // Allocate space to regularize kernel values.
       kernelValues = new double[hd.getDUDLBins()];
 
       String propString = properties.getString("ost-integrationType", "SIMPSONS");
       IntegrationType testType;
       try {
         testType = IntegrationType.valueOf(propString.toUpperCase());
       } catch (Exception ex) {
         logger.warning(format(" Invalid argument %s to ost-integrationType; resetting to SIMPSONS", propString));
         testType = SIMPSONS;
       }
       integrationType = testType;
 
       spreadBias = properties.getBoolean("ost-spread-bias", false);
 
       /*
        Set up the multi-walker communication variables for Parallel Java
        communication between nodes.
       */
       world = Comm.world();
       int numProc = world.size();
       rank = world.rank();
       if (hd.asynchronous) {
         // Use asynchronous communication.
         sendAsynchronous = new SendAsynchronous(this);
         sendAsynchronous.start();
         sendSynchronous = null;
       } else {
         Histogram[] histograms = new Histogram[numProc];
         int[] rankToHistogramMap = new int[numProc];
         for (int i = 0; i < numProc; i++) {
           histograms[i] = this;
           rankToHistogramMap[i] = 0;
         }
         sendSynchronous = new SendSynchronous(histograms, rankToHistogramMap);
         sendAsynchronous = null;
       }
 
       lastReceivedLambda = ld.lambda;
       if (hd.discreteLambda) {
         lastReceivedLambda = mapLambda(lastReceivedLambda);
         logger.info(format(" Discrete lambda: initializing lambda to nearest bin %.5f", lastReceivedLambda));
         ld.lambda = lastReceivedLambda;
         ld.theta = asin(sqrt(lastReceivedLambda));
         lambdaInterface.setLambda(lastReceivedLambda);
         lambdaState = null;
         stochastic = null;
       } else {
         // Configure the Stochastic integrator.
         lambdaState = new SystemState(1);
         stochastic = new Stochastic(lambdaParticleOptions.getLambdaFriction(), lambdaState);
         stochastic.setTemperature(dynamicsOptions.getTemperature());
         stochastic.setTimeStep(dynamicsOptions.getDtPsec());
         double[] mass = lambdaState.getMass();
         double[] thetaPosition = lambdaState.x();
         double[] thetaVelocity = lambdaState.v();
         double[] thetaAccel = lambdaState.a();
         mass[0] = lambdaParticleOptions.getLambdaMass();
         thetaPosition[0] = ld.theta;
         thetaVelocity[0] = ld.thetaVelocity;
         thetaAccel[0] = ld.thetaAcceleration;
       }
 
       lastReceiveddUdL = getdEdL();
 
       if (hd.wasHistogramRead()) {
         updateFreeEnergyDifference(true, false);
       }
     }
 
     public void disableResetStatistics() {
       hd.resetHistogram = false;
     }
 
     /**
      * evaluateTotalBias.
      *
      * @param bias If false, return the negative of the Total OST bias.
      * @return A StringBuffer The total OST Bias.
      */
     public StringBuffer evaluateTotalOSTBias(boolean bias) {
       StringBuffer sb = new StringBuffer();
       double[] chainRule = new double[2];
       for (int dUdLBin = 0; dUdLBin < hd.getDUDLBins(); dUdLBin++) {
         double currentdUdL = dUdLforBin(dUdLBin);
         sb.append(format(" %16.8f", currentdUdL));
         for (int lambdaBin = 0; lambdaBin < hd.getLambdaBins(); lambdaBin++) {
           double currentLambda = lambdaForIndex(lambdaBin);
           double bias1D = energyAndGradient1D(currentLambda, false);
           double bias2D = energyAndGradient2D(currentLambda, currentdUdL, chainRule, hd.biasMag);
           double totalBias = bias1D + bias2D;
           // If the bias flag is false, turn the total bias into the PMF.
           if (!bias) {
             totalBias = -totalBias;
           }
           sb.append(format(" %16.8f", totalBias));
         }
         sb.append("\n");
       }
       return sb;
     }
 
     /**
      * evaluate2DOSTBias.
      *
      * @param bias If false, return the negative of the 2D OST bias.
      * @return A StringBuffer with the 2D OST bias.
      */
     public StringBuffer evaluate2DOSTBias(boolean bias) {
       StringBuffer sb = new StringBuffer();
       double[] chainRule = new double[2];
       for (int dUdLBin = 0; dUdLBin < hd.getDUDLBins(); dUdLBin++) {
         double currentdUdL = dUdLforBin(dUdLBin);
         sb.append(format(" %16.8f", currentdUdL));
         for (int lambdaBin = 0; lambdaBin < hd.getLambdaBins(); lambdaBin++) {
           double currentLambda = lambdaForIndex(lambdaBin);
           double bias2D = energyAndGradient2D(currentLambda, currentdUdL, chainRule, hd.biasMag);
           if (!bias) {
             bias2D = -bias2D;
           }
           sb.append(format(" %16.8f", bias2D));
         }
         sb.append("\n");
       }
       return sb;
     }
 
     /**
      * For MPI parallel jobs, returns true if the walkers are independent (i.e. contribute to only
      * their own histogram).
      *
      * @return True if the walkers are independent.
      */
     public boolean getIndependentWalkers() {
       return hd.independentWalkers;
     }
 
     public double getLambdaResetValue() {
       return hd.resetHistogramAtLambda;
     }
 
     /**
      * For MPI parallel jobs, return the rank of this process.
      *
      * @return The rank of this process.
      */
     public int getRank() {
       return rank;
     }
 
     public boolean getResetStatistics() {
       return hd.resetHistogram;
     }
 
     /**
      * Return the SynchronousSend associated with this Histogram, if any.
      *
      * @return The SynchronousSend, if any.
      */
     public Optional<SendSynchronous> getSynchronousSend() {
       return Optional.ofNullable(sendSynchronous);
     }
 
     public void setLastReceiveddUdL(double lastReceiveddUdL) {
       this.lastReceiveddUdL = lastReceiveddUdL;
     }
 
     public double updateFreeEnergyDifference(boolean print, boolean save) {
       if (hd.metaDynamics) {
         return updateMetaDynamicsFreeEnergyDifference(print, save);
       } else {
         return updateOSTFreeEnergyDifference(print, save);
       }
     }
 
     /**
      * Update the free energy estimate for Meta Dynamics.
      *
      * @param print Whether to write the histogram to screen.
      * @param save  Whether to write the histogram to disk.
      * @return Free energy (via integration of ensemble-average dU/dL)
      */
     public double updateMetaDynamicsFreeEnergyDifference(boolean print, boolean save) {
       synchronized (this) {
         if (!print && currentNumberOfHills == hd.counts) {
           return currentFreeEnergyDifference;
         }
         double freeEnergyDifferenceMeta = 0.0;
         double freeEnergyDifferenceTI = 0.0;
 
         double minFL = Double.MAX_VALUE;
 
         int lambdaBins = hd.getLambdaBins();
         double[] metaFreeEnergy = new double[lambdaBins];
 
         // Total histogram weight.
         double totalWeight = 0;
         StringBuilder stringBuilder = new StringBuilder();
 
         // Loop over lambda bins, computing <dU/dL> for each bin.
         for (int lambdaBin = 0; lambdaBin < lambdaBins; lambdaBin++) {
           int firstdUdLBin = firstdUdLBin(lambdaBin);
           int lastdUdLBin = lastdUdLBin(lambdaBin);
           double lambdaCount = 0;
           // The dUdL range sampled for lambda.
           double mindUdLForLambda = 0.0;
           double maxdUdLforLambda = 0.0;
           double maxBias = 0;
           if (firstdUdLBin == -1 || lastdUdLBin == -1) {
             ensembleAveragedUdL[lambdaBin] = 0.0;
             minFL = 0.0;
           } else {
             double ensembleAverageFLambda = 0.0;
             double partitionFunction = 0.0;
 
             // Evaluate and regularize all kernel values for this value of lambda.
             double offset = evaluateKernelForLambda(lambdaBin, firstdUdLBin, lastdUdLBin);
 
             for (int dUdLBin = firstdUdLBin; dUdLBin <= lastdUdLBin; dUdLBin++) {
               double kernel = kernelValues[dUdLBin];
 
               if (kernel - offset > maxBias) {
                 maxBias = kernel - offset;
               }
 
               // The weight is just the kernel value for no 2D bias.
               partitionFunction += kernel;
 
               double currentdUdL = dUdLforBin(dUdLBin);
               ensembleAverageFLambda += currentdUdL * kernel;
               lambdaCount += getRecursionKernelValue(lambdaBin, dUdLBin);
             }
             if (minFL > maxBias) {
               minFL = maxBias;
             }
             ensembleAveragedUdL[lambdaBin] = (partitionFunction == 0) ? 0 : ensembleAverageFLambda / partitionFunction;
             mindUdLForLambda = hd.dUdLMinimum + firstdUdLBin * hd.dUdLBinWidth;
             maxdUdLforLambda = hd.dUdLMinimum + (lastdUdLBin + 1) * hd.dUdLBinWidth;
           }
 
           double deltaFreeEnergy = ensembleAveragedUdL[lambdaBin] * deltaForLambdaBin(lambdaBin);
           freeEnergyDifferenceTI += deltaFreeEnergy;
           totalWeight += lambdaCount;
 
 
           double lambdaBinWidth = hd.lambdaBinWidth;
           double llL = lambdaBin * lambdaBinWidth - hd.lambdaBinWidth_2;
           double ulL = llL + lambdaBinWidth;
           if (llL < 0.0) {
             llL = 0.0;
           }
           if (ulL > 1.0) {
             ulL = 1.0;
           }
 
           double midLambda = llL;
           if (!hd.discreteLambda) {
             midLambda = (llL + ulL) / 2.0;
           }
           double bias1D = energyAndGradient1D(midLambda, false);
           metaFreeEnergy[lambdaBin] = energyAndGradientMeta(midLambda, false);
           freeEnergyDifferenceMeta = -(metaFreeEnergy[lambdaBin] - metaFreeEnergy[0]);
 
           if (print || save) {
             stringBuilder.append(format(" %6.2e %7.5f %7.1f %7.1f %9.2f %9.2f %9.2f %9.2f %9.2f\n", lambdaCount,
                 midLambda, mindUdLForLambda, maxdUdLforLambda, ensembleAveragedUdL[lambdaBin],
                 bias1D, freeEnergyDifferenceTI, metaFreeEnergy[lambdaBin], freeEnergyDifferenceMeta));
           }
         }
 
         double temperingOffset = hd.getTemperingOffset();
         double temperEnergy = (minFL > temperingOffset) ? temperingOffset - minFL : 0;
         temperingWeight = exp(temperEnergy / deltaT);
 
         if (print) {
           logger.info("  Weight   Lambda      dU/dL Bins   <dU/dL>      g(L) dG_OST(L)  Meta(L) dG_Meta(L)");
           logger.info(stringBuilder.toString());
           logger.info(" Histogram Evaluation");
           logger.info(format("  Free Energy Difference: %12.4f kcal/mol", freeEnergyDifferenceMeta));
           logger.info(format("  Number of Hills:        %12d", hd.counts));
           logger.info(format("  Total Bias Added:       %12.4f kcal/mol", totalWeight * hd.biasMag));
           logger.info(format("  Minimum Bias:           %12.4f kcal/mol", minFL));
           logger.info(format("  Tempering Percentage:   %12.4f %%\n", temperingWeight * 100));
         }
 
         if (save) {
           String modelFilename = molecularAssembly.getFile().getAbsolutePath();
           File saveDir = new File(FilenameUtils.getFullPath(modelFilename));
           String dirName = saveDir + File.separator;
           String fileName = dirName + "histogram.txt";
           try {
             logger.info(" Writing " + fileName);
             PrintWriter printWriter = new PrintWriter(fileName);
             printWriter.write(stringBuilder.toString());
             printWriter.close();
           } catch (Exception e) {
             logger.info(format(" Failed to write %s.", fileName));
           }
         }
 
         currentNumberOfHills = hd.counts;
         currentFreeEnergyDifference = freeEnergyDifferenceMeta;
         return currentFreeEnergyDifference;
       }
     }
 
     /**
      * Eqs. 7 and 8 from the 2012 Crystal Thermodynamics paper.
      *
      * @param print Whether to write the histogram to screen.
      * @param save  Whether to write the histogram to disk.
      * @return Free energy (via integration of ensemble-average dU/dL)
      */
     public double updateOSTFreeEnergyDifference(boolean print, boolean save) {
       synchronized (this) {
         if (!print && currentNumberOfHills == hd.counts) {
           return currentFreeEnergyDifference;
         }
 
         double freeEnergyDifferenceOST = 0.0;
         double minFL = Double.MAX_VALUE;
 
         // If the bias magnitude is zero, computing <dU/dL> from
         // counts will not be correct. Assign a temporary non-zero bias magnitude.
         boolean biasMagZero = hd.biasMag <= 0.0;
 
         // Total histogram weight.
         double totalWeight = 0;
         double beta = 1.0 / (R * dynamicsOptions.getTemperature());
         StringBuilder stringBuilder = new StringBuilder();
 
         // Loop over lambda bins, computing <dU/dL> for each bin.
         for (int lambdaBin = 0; lambdaBin < hd.lambdaBins; lambdaBin++) {
           int firstdUdLBin = firstdUdLBin(lambdaBin);
           int lastdUdLBin = lastdUdLBin(lambdaBin);
           double lambdaCount = 0;
           // The FL range sampled for lambda bin [iL*dL .. (iL+1)*dL]
           double mindUdLforLambda = 0.0;
           double maxdUdLforLambda = 0.0;
           double maxBias = 0;
           if (firstdUdLBin == -1 || lastdUdLBin == -1) {
             ensembleAveragedUdL[lambdaBin] = 0.0;
             minFL = 0.0;
           } else {
             double ensembleAverage = 0.0;
             double partitionFunction = 0.0;
 
             // Evaluate and regularize all kernel values for this value of lambda.
             double offset = evaluateKernelForLambda(lambdaBin, firstdUdLBin, lastdUdLBin);
 
             for (int dUdLBin = firstdUdLBin; dUdLBin <= lastdUdLBin; dUdLBin++) {
               double kernel = kernelValues[dUdLBin];
 
               if (kernel - offset > maxBias) {
                 maxBias = kernel - offset;
               }
 
               double weight;
               if (biasMagZero) {
                 // The weight is just the kernel value for no 2D bias.
                 weight = kernel;
               } else {
                 // The Boltzmann weight based on temperature and the 2D bias height.
                 weight = exp(kernel * beta);
               }
               partitionFunction += weight;
 
               double currentdUdL = dUdLforBin(dUdLBin);
               ensembleAverage += currentdUdL * weight;
               lambdaCount += getRecursionKernelValue(lambdaBin, dUdLBin);
             }
             if (minFL > maxBias) {
               minFL = maxBias;
             }
             ensembleAveragedUdL[lambdaBin] = (partitionFunction == 0) ? 0 : ensembleAverage / partitionFunction;
             mindUdLforLambda = hd.dUdLMinimum + firstdUdLBin * hd.dUdLBinWidth;
             maxdUdLforLambda = hd.dUdLMinimum + (lastdUdLBin + 1) * hd.dUdLBinWidth;
           }
 
           double deltaFreeEnergy = ensembleAveragedUdL[lambdaBin] * deltaForLambdaBin(lambdaBin);
           freeEnergyDifferenceOST += deltaFreeEnergy;
           totalWeight += lambdaCount;
 
           if (print || save) {
             double llL = lambdaBin * hd.lambdaBinWidth - hd.lambdaBinWidth_2;
             double ulL = llL + hd.lambdaBinWidth;
             if (llL < 0.0) {
               llL = 0.0;
             }
             if (ulL > 1.0) {
               ulL = 1.0;
             }
 
             double midLambda = llL;
             if (!hd.discreteLambda) {
               midLambda = (llL + ulL) / 2.0;
             }
             double bias1D = energyAndGradient1D(midLambda, false);
 
             double bias2D = 0.0;
             if (!biasMagZero) {
               bias2D = computeBiasEnergy(midLambda, ensembleAveragedUdL[lambdaBin]) - bias1D;
             }
 
             stringBuilder.append(
                 format(" %6.2e %7.5f %7.1f %7.1f %8.2f %8.2f %8.2f %8.2f %8.2f   %8.2f\n", lambdaCount,
                     midLambda, mindUdLforLambda, maxdUdLforLambda, ensembleAveragedUdL[lambdaBin],
                     bias1D, bias2D, bias1D + bias2D, freeEnergyDifferenceOST, bias1D + bias2D + freeEnergyDifferenceOST));
           }
         }
 
         if (!biasMagZero) {
           double temperingOffset = hd.getTemperingOffset();
           double temperEnergy = (minFL > temperingOffset) ? temperingOffset - minFL : 0;
           temperingWeight = exp(temperEnergy / deltaT);
         }
 
         freeEnergyDifferenceOST = integrateNumeric(ensembleAveragedUdL, integrationType);
 
         if (print) {
           logger.info("  Weight   Lambda      dU/dL Bins  <dU/dL>    g(L)  f(L,<dU/dL>) Bias    dG(L) Bias+dG(L)");
           logger.info(stringBuilder.toString());
           logger.info(" Histogram Evaluation");
           logger.info(format("  Free Energy Difference: %12.4f kcal/mol", freeEnergyDifferenceOST));
           logger.info(format("  Number of Hills:        %12d", hd.counts));
           logger.info(format("  Total Bias Added:       %12.4f kcal/mol", totalWeight * hd.biasMag));
           logger.info(format("  Minimum Bias:           %12.4f kcal/mol", minFL));
           logger.info(format("  Tempering Percentage:   %12.4f %%\n", temperingWeight * 100));
         }
 
         if (save) {
           String modelFilename = molecularAssembly.getFile().getAbsolutePath();
           File saveDir = new File(FilenameUtils.getFullPath(modelFilename));
           String dirName = saveDir + File.separator;
           String fileName = dirName + "histogram.txt";
           try {
             logger.info("\n Writing " + fileName);
             PrintWriter printWriter = new PrintWriter(fileName);
             printWriter.write(stringBuilder.toString());
             printWriter.close();
           } catch (Exception e) {
             logger.info(format(" Failed to write %s.", fileName));
           }
         }
 
         currentNumberOfHills = hd.counts;
         currentFreeEnergyDifference = freeEnergyDifferenceOST;
         return currentFreeEnergyDifference;
       }
     }
 
     /**
      * For thermodynamic integration, return the integration width for the given Lambda lambdaBin.
      *
      * @param lambdaBin The lambda lambdaBin.
      * @return The integration width.
      */
     private double deltaForLambdaBin(int lambdaBin) {
       if (!hd.discreteLambda && (lambdaBin == 0 || lambdaBin == hd.lambdaBins - 1)) {
         // The first and last lambda bins are half size for continuous lambda.
         return hd.lambdaBinWidth_2;
       } else if (hd.discreteLambda && lambdaBin == 0) {
         // The free energy change to move from L=0 to L=0 is zero.
         return 0.0;
       }
       // All other cases.
       return hd.lambdaBinWidth;
     }
 
     /**
      * Returns the index of the first dU/dL bin with counts, or -1 if there are no counts for the
      * given lambda bin.
      *
      * @param lambdaBin Lambda bin to consider.
      * @return Index of the first dUdL bin with counts.
      */
     private int firstdUdLBin(int lambdaBin) {
       // Synchronize use of the recursion kernel.
       synchronized (this) {
         // Find the smallest FL bin that has counts.
         for (int jFL = 0; jFL < hd.dUdLBins; jFL++) {
           double count = hd.zHistogram[lambdaBin][jFL];
           if (count > 0) {
             return jFL;
           }
         }
         return -1;
       }
     }
 
     /**
      * Returns the index of the last dU/dL bin with counts, or -1 if there are no counts for the
      * given lambda bin.
      *
      * @param lambdaBin Lambda bin to consider.
      * @return Index of the last dUdL bin with counts.
      */
     private int lastdUdLBin(int lambdaBin) {
       // Synchronize use of the recursion kernel.
       synchronized (this) {
         // Find the largest FL bin that has counts.
         for (int jFL = hd.dUdLBins - 1; jFL >= 0; jFL--) {
           double count = hd.zHistogram[lambdaBin][jFL];
           if (count > 0) {
             return jFL;
           }
         }
         return -1;
       }
     }
 
     /**
      * Write histogram and lambda restart files.
      * <p>
      * For MPI parallel jobs, only rank 0 writes a histogram restart file (unless the "independentWrite" flag is set).
      */
     void writeRestart() {
       if (rank == 0 || hd.independentWrite) {
         updateFreeEnergyDifference(true, false);
         try {
           hd.writeHistogram();
           logger.info(format(" Wrote histogram restart to: %s.", hd.getHistogramFileName()));
         } catch (Exception ex) {
           String message = format(" Exception writing histogram restart file %s.", hd.getHistogramFileName());
           logger.log(Level.INFO, Utilities.stackTraceToString(ex));
           logger.log(Level.SEVERE, message, ex);
         }
       }
 
       // All ranks write a lambda restart file.
       try {
         ld.writeLambdaData();
         logger.info(format(" Wrote lambda restart to:    %s.", ld.getLambdaFileName()));
       } catch (Exception ex) {
         String message = format(" Exception writing lambda restart file %s.", ld.getLambdaFileName());
         logger.log(Level.INFO, Utilities.stackTraceToString(ex));
         logger.log(Level.SEVERE, message, ex);
       }
     }
 
     private double mapLambda(double lambda) {
       if (hd.discreteLambda) {
         return hd.lambdaLadder[indexForDiscreteLambda(lambda)];
       } else {
         return lambda;
       }
     }
 
     /**
      * For continuous lambda, the returned value is the lambda bin. For discrete lambda, the returned
      * value is the discrete lambda index.
      *
      * @param lambda a double.
      * @return a int.
      */
     int indexForLambda(double lambda) {
       if (hd.discreteLambda) {
         return indexForDiscreteLambda(lambda);
       } else {
         return indexForContinuousLambda(lambda);
       }
     }
 
     int indexForLambda() {
       return indexForLambda(ld.lambda);
     }
 
     private double lambdaForIndex(int bin) {
       if (hd.discreteLambda) {
         return hd.lambdaLadder[bin];
       } else {
         return bin * hd.lambdaBinWidth;
       }
     }
 
     private int indexForContinuousLambda(double lambda) {
       int lambdaBin = (int) floor((lambda - hd.minLambda) / hd.lambdaBinWidth);
       if (lambdaBin < 0) {
         lambdaBin = 0;
       }
       if (lambdaBin >= hd.lambdaBins) {
         lambdaBin = hd.lambdaBins - 1;
       }
       return lambdaBin;
     }
 
     private int indexForDiscreteLambda(double lambda) {
       assert hd.discreteLambda && hd.lambdaLadder != null && hd.lambdaLadder.length > 0;
 
       int initialGuess = indexForContinuousLambda(lambda);
       double minErr = Double.MAX_VALUE;
       int minErrBin = -1;
       for (int i = -1; i < 2; i++) {
         int guessBin = i + initialGuess;
         if (guessBin < 0 || guessBin >= hd.lambdaBins) {
           continue;
         }
         double guessLam = hd.lambdaLadder[guessBin];
         double guessErr = Math.abs(guessLam - lambda);
         if (guessErr < minErr) {
           minErr = guessErr;
           minErrBin = guessBin;
         }
       }
 
       assert minErr < 1.0E-6;
       return minErrBin;
     }
 
     /**
      * Find the bin for the supplied dEdLambda.
      * <p>
      * If the supplied dEdL is outside the range of the count matrix, then -1 is returned.
      *
      * @param dUdL a double.
      * @return The dUdL bin.
      */
     int binFordUdL(double dUdL) {
 
       // No counts below mindUdL.
       if (dUdL < hd.dUdLMinimum) {
         return -1;
       }
 
       // No counts above the maxdUdL.
       if (dUdL > hd.dUdLMaximum) {
         return -1;
       }
 
       int bin = (int) floor((dUdL - hd.dUdLMinimum) / hd.dUdLBinWidth);
 
       if (bin == hd.dUdLBins) {
         bin = hd.dUdLBins - 1;
       }
 
       if (bin < 0) {
         bin = 0;
       }
 
       return bin;
     }
 
     /**
      * Return the value of a recursion kernel bin. Mirror conditions are applied to the lambda bin.
      * If the dUdLBin is outside the range of the count matrix, zero is returned.
      *
      * @param lambdaBin The lambda bin.
      * @param dUdLBin   The dU/dL bin.
      * @return The value of the bin.
      */
     double getRecursionKernelValue(int lambdaBin, int dUdLBin) {
       // Synchronize use of the recursion kernel.
       synchronized (this) {
         // Apply lambda mirror condition.
         lambdaBin = lambdaMirror(lambdaBin);
 
         // For dUdL outside the count matrix the weight is 0.
         if (dUdLBin < 0 || dUdLBin >= hd.dUdLBins) {
           return 0.0;
         }
 
         return hd.zHistogram[lambdaBin][dUdLBin];
       }
     }
 
     /**
      * Return the value of a recursion kernel bin. Mirror conditions are applied to the lambda bin.
      * If the dUdL is outside the range of the count matrix, zero is returned.
      *
      * @param lambda the lambda value.
      * @param dUdL   the dU/dL value.
      * @return The value of the Histogram.
      */
     double getRecursionKernelValue(double lambda, double dUdL) {
       return getRecursionKernelValue(indexForLambda(lambda), binFordUdL(dUdL));
     }
 
     /**
      * Add to the value of a recursion kernel bin.
      *
      * @param currentLambda The lambda bin.
      * @param currentdUdL   The dU/dL bin.
      * @param value         The value of the bin.
      */
     void addToRecursionKernelValue(double currentLambda, double currentdUdL, double value) {
       synchronized (this) {
         if (spreadBias) {
           // Expand the recursion kernel if necessary.
           int dUdLbiasCutoff = hd.dUdLBiasCutoff;
           checkRecursionKernelSize(dUdLforBin(binFordUdL(currentdUdL) - dUdLbiasCutoff));
           checkRecursionKernelSize(dUdLforBin(binFordUdL(currentdUdL) + dUdLbiasCutoff));
 
           int currentLambdaBin = indexForLambda(currentLambda);
           int currentdUdLBin = binFordUdL(currentdUdL);
 
           // Variances are only used when dividing by twice their value.
           double invLs2 = 0.5 / hd.lambdaVariance;
           double invFLs2 = 0.5 / hd.dUdLVariance;
 
           // Compute the normalization.
           double normalize = 0.0;
           int lambdaBiasCutoff = hd.lambdaBiasCutoff;
           for (int iL = -lambdaBiasCutoff; iL <= lambdaBiasCutoff; iL++) {
             int lambdaBin = currentLambdaBin + iL;
             double deltaL = currentLambda - lambdaBin * hd.lambdaBinWidth;
             double deltaL2 = deltaL * deltaL;
             // Pre-compute the lambda bias.
             double L2exp = exp(-deltaL2 * invLs2);
             for (int jFL = -dUdLbiasCutoff; jFL <= dUdLbiasCutoff; jFL++) {
               int dUdLBin = currentdUdLBin + jFL;
               double deltaFL = currentdUdL - dUdLforBin(dUdLBin);
               double deltaFL2 = deltaFL * deltaFL;
               normalize += L2exp * exp(-deltaFL2 * invFLs2);
             }
           }
 
           for (int iL = -lambdaBiasCutoff; iL <= lambdaBiasCutoff; iL++) {
             int lambdaBin = currentLambdaBin + iL;
             double deltaL = currentLambda - lambdaBin * hd.lambdaBinWidth;
             double deltaL2 = deltaL * deltaL;
             // Pre-compute the lambda bias.
             double L2exp = exp(-deltaL2 * invLs2);
             lambdaBin = lambdaMirror(lambdaBin);
             for (int jFL = -dUdLbiasCutoff; jFL <= dUdLbiasCutoff; jFL++) {
               int dUdLBin = currentdUdLBin + jFL;
               double deltaFL = currentdUdL - dUdLforBin(dUdLBin);
               double deltaFL2 = deltaFL * deltaFL;
               double weight = value / normalize * L2exp * exp(-deltaFL2 * invFLs2);
               hd.zHistogram[lambdaBin][dUdLBin] += weight;
             }
           }
           hd.counts++;
         } else {
           // Check the recursion kernel size.
           checkRecursionKernelSize(currentdUdL);
           int lambdaBin = indexForLambda(currentLambda);
           int dUdLBin = binFordUdL(currentdUdL);
           try {
             hd.zHistogram[lambdaBin][dUdLBin] += value;
             hd.counts++;
           } catch (IndexOutOfBoundsException e) {
             logger.info(format(" Histogram dimensions %d %d", hd.lambdaBins, hd.dUdLBins));
             logger.info(format(
                 " Count skipped in addToRecursionKernelValue due to an index out of bounds exception.\n L=%10.8f (%d), dU/dL=%10.8f (%d) and count=%10.8f",
                 currentLambda, lambdaBin, currentdUdL, dUdLBin, value));
 
           }
         }
       }
     }
 
     /**
      * Allocate memory for the recursion kernel.
      */
     void allocateRecursionKernel() {
       // Synchronize updates of the recursion kernel.
       synchronized (this) {
         hd.zHistogram = new double[hd.lambdaBins][hd.dUdLBins];
         kernelValues = new double[hd.dUdLBins];
       }
     }
 
     /**
      * Integrates dUdL over lambda using more sophisticated techniques than midpoint rectangular
      * integration.
      *
      * <p>The ends (from 0 to dL and 1-dL to 1) are integrated with trapezoids for continuous
      * lambda.
      *
      * @param dUdLs dUdL at the midpoint of each bin.
      * @param type  Integration type to use.
      * @return Current delta-G estimate.
      */
     private double integrateNumeric(double[] dUdLs, IntegrationType type) {
       double val;
       if (hd.discreteLambda) {
         // Integrate between the first bin and the last bin.
         double[] lams = Integrate1DNumeric.generateXPoints(0.0, 1.0, hd.lambdaBins, false);
         DataSet dSet = new DoublesDataSet(lams, dUdLs, false);
         val = Integrate1DNumeric.integrateData(dSet, Integrate1DNumeric.IntegrationSide.LEFT, type);
       } else {
         // Integrate between the second bin midpoint and the second-to-last bin midpoint.
         double[] midLams = Integrate1DNumeric.generateXPoints(hd.lambdaBinWidth, 1.0 - hd.lambdaBinWidth,
             (hd.lambdaBins - 2), false);
         double[] midVals = Arrays.copyOfRange(dUdLs, 1, (hd.lambdaBins - 1));
         DataSet dSet = new DoublesDataSet(midLams, midVals, false);
 
         val = Integrate1DNumeric.integrateData(dSet, Integrate1DNumeric.IntegrationSide.LEFT, type);
 
         // Everything after this is just adding in the endpoint contributions.
 
         double dL_4 = hd.lambdaBinWidth_2 * 0.5;
 
         // Initially, assume dU/dL is exactly 0 at the endpoints. This is sometimes a true
         // assumption.
         double val0 = 0;
         double val1 = 0;
 
         // If we cannot guarantee that dUdL is exactly 0 at the endpoints, interpolate.
         if (!lambdaInterface.dEdLZeroAtEnds()) {
           double recipSlopeLen = 1.0 / (hd.lambdaBinWidth * 0.75);
 
           double slope = dUdLs[0] - dUdLs[1];
           slope *= recipSlopeLen;
           val0 = dUdLs[0] + (slope * dL_4);
 
           slope = dUdLs[hd.lambdaBins - 1] - dUdLs[hd.lambdaBins - 2];
           slope *= recipSlopeLen;
           val1 = dUdLs[hd.lambdaBins - 1] + (slope * dL_4);
           logger.fine(format(" Inferred dU/dL values at 0 and 1: %10.5g , %10.5g", val0, val1));
         }
 
         // Integrate trapezoids from 0 to the second bin midpoint, and from second-to-last bin
         // midpoint to 1.
         val += trapezoid(0, dL_4, val0, dUdLs[0]);
         val += trapezoid(dL_4, hd.lambdaBinWidth, dUdLs[0], dUdLs[1]);
         val += trapezoid(1.0 - hd.lambdaBinWidth, 1.0 - dL_4, dUdLs[hd.lambdaBins - 2],
             dUdLs[hd.lambdaBins - 1]);
         val += trapezoid(1.0 - dL_4, 1.0, dUdLs[hd.lambdaBins - 1], val1);
       }
 
       return val;
     }
 
     /**
      * Integrates a trapezoid.
      *
      * @param x0  First x point
      * @param x1  Second x point
      * @param fx0 First f(x) point
      * @param fx1 Second f(x) point
      * @return The area under a trapezoid.
      */
     private double trapezoid(double x0, double x1, double fx0, double fx1) {
       double val = 0.5 * (fx0 + fx1);
       val *= (x1 - x0);
       return val;
     }
 
     /**
      * Evaluate the kernel across dU/dL values at given value of lambda. The largest kernel value V
      * is used to define an offset (-V), which is added to all to kernel values. Then, the largest
      * kernel value is zero, and will result in a statistical weight of 1.
      *
      * <p>The offset avoids the recursion kernel becoming too large for some combinations of
      * (Lambda, dU/dL). This can result in its statistical weight = exp(kernel * beta) exceeding the
      * maximum representable double value.
      *
      * @param lambda Value of Lambda to evaluate the kernel for.
      * @param llFL   Lower FLambda bin.
      * @param ulFL   Upper FLambda bin.
      * @return the applied offset.
      */
     private double evaluateKernelForLambda(int lambda, int llFL, int ulFL) {
       double maxKernel = Double.MIN_VALUE;
 
       double gaussianBiasMagnitude = hd.biasMag;
       if (hd.biasMag <= 0.0) {
         gaussianBiasMagnitude = PSEUDO_BIAS_MAGNITUDE;
       }
 
       for (int jFL = llFL; jFL <= ulFL; jFL++) {
         double value = evaluateKernel(lambda, jFL, gaussianBiasMagnitude);
         kernelValues[jFL] = value;
         if (value > maxKernel) {
           maxKernel = value;
         }
       }
 
       // Only offset the kernel values for use with the 2D OST bias.
       double offset = 0.0;
       if (hd.biasMag > 0.0 && !hd.metaDynamics) {
         offset = -maxKernel;
         for (int jFL = llFL; jFL <= ulFL; jFL++) {
           kernelValues[jFL] += offset;
         }
       }
       return offset;
     }
 
     /**
      * Mirror the lambda bin if its less < 0 or greater than the last bin.
      *
      * @param bin Lambda bin to mirror.
      * @return The mirrored lambda bin.
      */
     private int lambdaMirror(int bin) {
       if (bin < 0) {
         return -bin;
       }
       int lastBin = hd.lambdaBins - 1;
       if (bin > lastBin) {
         // Number of bins past the last bin.
         bin -= lastBin;
         // Return Mirror bin
         return lastBin - bin;
       }
       // No mirror condition.
       return bin;
     }
 
     /**
      * For continuous lambda, the width of the first and last bins is dLambda_2, so the mirror
      * condition is to double their counts.
      *
      * @param bin Current lambda bin.
      * @return The mirror factor (either 1.0 or 2.0).
      */
     private double mirrorFactor(int bin) {
       if (hd.discreteLambda) {
         return 1.0;
       }
       if (bin == 0 || bin == hd.lambdaBins - 1) {
         return 2.0;
       }
       return 1.0;
     }
 
     /**
      * Compute the value of dU/dL for the given Histogram bin.
      *
      * @param dUdLBin The bin index in the dU/dL dimension.
      * @return The value of dU/dL at the center of the bin.
      */
     private double dUdLforBin(int dUdLBin) {
       return hd.dUdLMinimum + dUdLBin * hd.dUdLBinWidth + hd.dUdLBinWidth_2;
     }
 
     /**
      * Evaluate the bias at [currentLambdaBin, cF_lambda]
      */
     private double evaluateKernel(int currentLambdaBin, int currentdUdLBin, double gaussianBiasMagnitude) {
       // Compute the value of L and FL for the center of the current bin.
       double currentLambda = currentLambdaBin * hd.lambdaBinWidth;
       double currentdUdL = dUdLforBin(currentdUdLBin);
 
       // Variances are only used when dividing by twice their value.
       double invLs2 = 0.5 / hd.lambdaVariance;
       double invFLs2 = 0.5 / hd.dUdLVariance;
 
       double sum = 0.0;
       int lambdaBiasCutoff = hd.lambdaBiasCutoff;
       for (int iL = -lambdaBiasCutoff; iL <= lambdaBiasCutoff; iL++) {
         int lambdaBin = currentLambdaBin + iL;
         double deltaL = currentLambda - lambdaBin * hd.lambdaBinWidth;
         double deltaL2 = deltaL * deltaL;
 
         // Pre-compute the lambda bias.
         double L2exp = exp(-deltaL2 * invLs2);
 
         // Mirror condition for Lambda counts.
         lambdaBin = lambdaMirror(lambdaBin);
         double mirrorFactor = mirrorFactor(lambdaBin);
 
         int dUdLbiasCutoff = hd.dUdLBiasCutoff;
         for (int jFL = -dUdLbiasCutoff; jFL <= dUdLbiasCutoff; jFL++) {
           int dUdLBin = currentdUdLBin + jFL;
           double weight = mirrorFactor * getRecursionKernelValue(lambdaBin, dUdLBin);
           if (weight <= 0.0) {
             continue;
           }
           double deltaFL = currentdUdL - dUdLforBin(dUdLBin);
           double deltaFL2 = deltaFL * deltaFL;
           double e = weight * L2exp * exp(-deltaFL2 * invFLs2);
           sum += e;
         }
       }
       return gaussianBiasMagnitude * sum;
     }
 
     /**
      * Compute the total Bias energy at (currentLambda, currentdUdL).
      *
      * @param currentLambda The value of lambda.
      * @param currentdUdL   The value of dU/dL.
      * @return The bias energy.
      */
     double computeBiasEnergy(double currentLambda, double currentdUdL) {
       synchronized (this) {
         int currentLambdaBin = indexForLambda(currentLambda);
         int currentdUdLBin = binFordUdL(currentdUdL);
 
         double bias1D;
         double bias2D = 0.0;
 
         if (!hd.metaDynamics) {
           if (hd.biasMag > 0.0) {
             int lambdaBiasCutoff = hd.lambdaBiasCutoff;
             for (int iL = -lambdaBiasCutoff; iL <= lambdaBiasCutoff; iL++) {
 
               int lambdaBin = currentLambdaBin + iL;
               double deltaL = currentLambda - (lambdaBin * hd.lambdaBinWidth);
               double deltaL2 = deltaL * deltaL;
               double expL2 = exp(-deltaL2 / (2.0 * hd.lambdaVariance));
 
               // Mirror conditions for recursion kernel counts.
               lambdaBin = lambdaMirror(lambdaBin);
               double mirrorFactor = mirrorFactor(lambdaBin);
 
               int dUdLbiasCutoff = hd.dUdLBiasCutoff;
               for (int iFL = -dUdLbiasCutoff; iFL <= dUdLbiasCutoff; iFL++) {
                 int dUdLBin = currentdUdLBin + iFL;
 
                 double weight = mirrorFactor * getRecursionKernelValue(lambdaBin, dUdLBin);
                 if (weight <= 0.0) {
                   continue;
                 }
 
                 double deltaFL = currentdUdL - dUdLforBin(dUdLBin);
                 double deltaFL2 = deltaFL * deltaFL;
                 double bias = weight * hd.biasMag * expL2 * exp(-deltaFL2 / (2.0 * hd.dUdLVariance));
                 bias2D += bias;
               }
             }
           }
           // Compute the energy for the recursion worker at F(L) using interpolation.
           bias1D = energyAndGradient1D(currentLambda, false);
         } else {
           bias1D = energyAndGradientMeta(currentLambda, false);
         }
 
         // Return the total bias.
         return bias1D + bias2D;
       }
     }
 
     /**
      * Compute the total Bias energy at (currentLambda, currentdUdL) and its gradient.
      *
      * @param currentdUdLambda The value of dU/dL.
      * @param chainRule        The chain rule terms for the bias.
      * @return The bias energy.
      */
     double energyAndGradient2D(double currentdUdLambda, double[] chainRule) {
       return energyAndGradient2D(ld.lambda, currentdUdLambda, chainRule, hd.biasMag);
     }
 
     /**
      * Compute the total Bias energy at (currentLambda, currentdUdL) and its gradient.
      *
      * @param currentLambda         The value of lambda.
      * @param currentdUdLambda      The value of dU/dL.
      * @param chainRule             The chain rule terms for the bias.
      * @param gaussianBiasMagnitude The magnitude of the Gaussian bias.
      * @return The bias energy.
      */
     double energyAndGradient2D(double currentLambda, double currentdUdLambda, double[] chainRule,
                                double gaussianBiasMagnitude) {
 
       if (gaussianBiasMagnitude <= 0.0) {
         chainRule[0] = 0.0;
         chainRule[1] = 0.0;
         return 0;
       }
 
       // Calculate recursion kernel G(L, F_L) and its derivatives with respect to L and F_L.
       double gLdEdL = 0.0;
       double dGdLambda = 0.0;
       double dGdFLambda = 0.0;
       int currentLambdaBin = indexForLambda(currentLambda);
       int currentdUdLBin = binFordUdL(currentdUdLambda);
 
       int lambdaBiasCutoff = hd.lambdaBiasCutoff;
       for (int iL = -lambdaBiasCutoff; iL <= lambdaBiasCutoff; iL++) {
         int lambdaBin = currentLambdaBin + iL;
 
         // Pass in the bin offset to the weight instance.
         double deltaL = currentLambda - (lambdaBin * hd.lambdaBinWidth);
         double deltaL2 = deltaL * deltaL;
         double expL2 = exp(-deltaL2 / (2.0 * hd.lambdaVariance));
 
         // Mirror conditions for recursion kernel counts.
         double mirrorFactor = mirrorFactor(lambdaBin);
         int dUdLbiasCutoff = hd.dUdLBiasCutoff;
         for (int iFL = -dUdLbiasCutoff; iFL <= dUdLbiasCutoff; iFL++) {
           int dUdLBin = currentdUdLBin + iFL;
 
           double weight = mirrorFactor * getRecursionKernelValue(lambdaBin, dUdLBin);
           if (weight <= 0.0) {
             continue;
           }
 
           double deltaFL = currentdUdLambda - dUdLforBin(dUdLBin);
           double deltaFL2 = deltaFL * deltaFL;
           double bias = weight * gaussianBiasMagnitude * expL2 * exp(-deltaFL2 / (2.0 * hd.dUdLVariance));
           gLdEdL += bias;
           dGdLambda -= deltaL / hd.lambdaVariance * bias;
           dGdFLambda -= deltaFL / hd.dUdLVariance * bias;
         }
       }
 
       chainRule[0] = dGdLambda;
       chainRule[1] = dGdFLambda;
       return gLdEdL;
     }
 
 
     /**
      * This calculates the 1D OST bias and its derivative with respect to Lambda.
      * <p>
      * See Equation 8 in <a href="http://doi.org/10.1021/ct300035u">
      * The Structure, Thermodynamics, and Solubility of Organic Crystals from Simulation with a
      * Polarizable Force Field</a>.
      * <p>
      *
      * @param gradient If true, compute the gradient.
      * @return a double.
      */
     private double energyAndGradient1D(boolean gradient) {
       return energyAndGradient1D(ld.lambda, gradient);
     }
 
     /**
      * This calculates the 1D OST bias and its derivative with respect to Lambda.
      * <p>
      * See Equation 8 in <a href="http://doi.org/10.1021/ct300035u">
      * The Structure, Thermodynamics, and Solubility of Organic Crystals from Simulation with a
      * Polarizable Force Field</a>.
      * <p>
      *
      * @param currentLambda The current value of lambda.
      * @param gradient      If true, compute the gradient.
      * @return a double.
      */
     private double energyAndGradient1D(double currentLambda, boolean gradient) {
       synchronized (this) {
         double biasEnergy = 0.0;
         for (int iL0 = 0; iL0 < hd.lambdaBins - 1; iL0++) {
           int iL1 = iL0 + 1;
 
           // Find bin centers and values for interpolation / extrapolation points.
           double L0 = iL0 * hd.lambdaBinWidth;
           double L1 = L0 + hd.lambdaBinWidth;
           double FL0 = ensembleAveragedUdL[iL0];
           double FL1 = ensembleAveragedUdL[iL1];
           double deltaFL = FL1 - FL0;
           /*
            If the lambda is less than or equal to the upper limit, this is
            the final interval. Set the upper limit to L, compute the partial
            derivative and break.
           */
           boolean done = false;
           if (currentLambda <= L1) {
             done = true;
             L1 = currentLambda;
           }
 
           // Upper limit - lower limit of the integral of the extrapolation / interpolation.
           biasEnergy += (FL0 * L1 + deltaFL * L1 * (0.5 * L1 - L0) / hd.lambdaBinWidth);
           biasEnergy -= (FL0 * L0 + deltaFL * L0 * (-0.5 * L0) / hd.lambdaBinWidth);
           if (done) {
             // Compute the gradient d F(L) / dL at L.
             if (gradient) {
               dUdLambda -= FL0 + (L1 - L0) * deltaFL / hd.lambdaBinWidth;
             }
             break;
           }
         }
         return -biasEnergy;
       }
     }
 
     /**
      * This calculates the 1D Metadynamics bias and its derivative with respect to Lambda.
      *
      * @param gradient If true, compute the gradient.
      * @return a double.
      */
     private double energyAndGradientMeta(boolean gradient) {
       return energyAndGradientMeta(ld.lambda, gradient);
     }
 
     /**
      * This calculates the 1D Metadynamics bias and its derivative with respect to Lambda.
      *
      * @param currentLambda The current value of lambda.
      * @param gradient      If true, compute the gradient.
      * @return a double.
      */
     private double energyAndGradientMeta(double currentLambda, boolean gradient) {
       // Zero out the metadynamics bias energy.
       double biasEnergy = 0.0;
 
       // Current lambda bin.
       int currentBin = indexForLambda(currentLambda);
 
       // Loop over bins within the cutoff.
       int lambdaBiasCutoff = hd.lambdaBiasCutoff;
       for (int iL = -lambdaBiasCutoff; iL <= lambdaBiasCutoff; iL++) {
         int lambdaBin = currentBin + iL;
         double deltaL = currentLambda - (lambdaBin * hd.lambdaBinWidth);
         double deltaL2 = deltaL * deltaL;
         // Mirror conditions for the lambda bin and count magnitude.
         lambdaBin = lambdaMirror(lambdaBin);
         double mirrorFactor = mirrorFactor(lambdaBin);
         double weight = mirrorFactor * hd.biasMag * countsForLambda(lambdaBin);
         if (weight > 0) {
           double e = weight * exp(-deltaL2 / (2.0 * hd.lambdaVariance));
           biasEnergy += e;
           // Add the dMeta/dL contribution.
           if (gradient) {
             dUdLambda -= deltaL / hd.lambdaVariance * e;
           }
         }
       }
 
       return biasEnergy;
     }
 
     /**
      * Marginalize over dU/dL counts for the given lambda bin.
      *
      * @param lambdaBin Lambda bin to marginalize for.
      * @return Total number of counts.
      */
     private double countsForLambda(int lambdaBin) {
       synchronized (this) {
         double count = 0.0;
         for (int i = 0; i < hd.dUdLBins; i++) {
           count += hd.zHistogram[lambdaBin][i];
         }
         return count;
       }
     }
 
     /**
      * If necessary, allocate more space.
      */
     void checkRecursionKernelSize(double currentdUdL) {
       // Synchronize updates of the recursion kernel.
       synchronized (this) {
         if (currentdUdL > hd.dUdLMaximum) {
           logger.info(format(" Current F_lambda %8.2f > maximum histogram size %8.2f.", currentdUdL, hd.dUdLMaximum));
 
           double origDeltaG = updateFreeEnergyDifference(false, false);
 
           int newdUdLBins = hd.dUdLBins;
           while (hd.dUdLMinimum + newdUdLBins * hd.dUdLBinWidth < currentdUdL) {
             newdUdLBins += 100;
           }
           double[][] newRecursionKernel = new double[hd.lambdaBins][newdUdLBins];
 
           // We have added bins above the indices of the current counts just copy them into the new array.
           for (int i = 0; i < hd.lambdaBins; i++) {
             arraycopy(hd.zHistogram[i], 0, newRecursionKernel[i], 0, hd.dUdLBins);
           }
 
           hd.zHistogram = newRecursionKernel;
           hd.dUdLBins = newdUdLBins;
           kernelValues = new double[hd.dUdLBins];
           hd.dUdLMaximum = hd.dUdLMinimum + hd.dUdLBinWidth * hd.dUdLBins;
           logger.info(format(" New histogram %8.2f to %8.2f with %d bins.\n", hd.dUdLMinimum, hd.dUdLMaximum, hd.dUdLBins));
 
           double newFreeEnergy = updateFreeEnergyDifference(true, false);
           assert (origDeltaG == newFreeEnergy);
         }
         if (currentdUdL < hd.dUdLMinimum) {
           logger.info(format(" Current F_lambda %8.2f < minimum histogram size %8.2f.", currentdUdL, hd.dUdLMinimum));
 
           double origDeltaG = updateFreeEnergyDifference(false, false);
 
           int offset = 100;
           while (currentdUdL < hd.dUdLMinimum - offset * hd.dUdLBinWidth) {
             offset += 100;
           }
           int newdUdLBins = hd.dUdLBins + offset;
           double[][] newRecursionKernel = new double[hd.lambdaBins][newdUdLBins];
 
           // We have added bins below the current counts,
           // so their indices must be increased by: offset = newFLBins - FLBins
           for (int i = 0; i < hd.lambdaBins; i++) {
             arraycopy(hd.zHistogram[i], 0, newRecursionKernel[i], offset, hd.dUdLBins);
           }
 
           hd.zHistogram = newRecursionKernel;
           hd.dUdLMinimum = hd.dUdLMinimum - offset * hd.dUdLBinWidth;
           hd.dUdLBins = newdUdLBins;
           kernelValues = new double[hd.dUdLBins];
           logger.info(format(" New histogram %8.2f to %8.2f with %d bins.\n", hd.dUdLMinimum, hd.dUdLMaximum, hd.dUdLBins));
 
           double newFreeEnergy = updateFreeEnergyDifference(true, false);
           assert (origDeltaG == newFreeEnergy);
         }
       }
     }
 
     /**
      * Add a Gaussian hill to the Histogram at (lambda, dUdL).
      *
      * @param dUdL The value of dU/dL.
      */
     void addBias(double dUdL) {
       // Communicate adding the bias to all walkers.
       if (hd.asynchronous) {
         sendAsynchronous.send(ld.lambda, dUdL, temperingWeight);
       } else {
         sendSynchronous.send(ld.lambda, dUdL, temperingWeight);
       }
     }
 
     /**
      * Pre-force portion of the stochastic integrator.
      */
     private void stochasticPreForce() {
       // Update theta.
       double[] thetaPosition = lambdaState.x();
       thetaPosition[0] = ld.theta;
       stochastic.preForce(OrthogonalSpaceTempering.this);
       ld.theta = thetaPosition[0];
 
       // Maintain theta in the interval -PI to PI.
       if (ld.theta > PI) {
         ld.theta -= 2.0 * PI;
       } else if (ld.theta <= -PI) {
         ld.theta += 2.0 * PI;
       }
 
       // Update lambda as sin(theta)^2.
       double sinTheta = sin(ld.theta);
       ld.lambda = sinTheta * sinTheta;
       lambdaInterface.setLambda(ld.lambda);
     }
 
     /**
      * Post-force portion of the stochastic integrator.
      */
     private void stochasticPostForce() {
       double[] thetaPosition = lambdaState.x();
       double[] gradient = lambdaState.gradient();
       // Update theta and dU/dtheta.
       gradient[0] = dUdLambda * sin(2 * ld.theta);
       thetaPosition[0] = ld.theta;
       stochastic.postForce(gradient);
 
       // Load the new theta, velocity and acceleration.
       double[] thetaVelocity = lambdaState.v();
       double[] thetaAcceleration = lambdaState.a();
       ld.theta = thetaPosition[0];
       ld.thetaVelocity = thetaVelocity[0];
       ld.thetaAcceleration = thetaAcceleration[0];
 
       // Maintain theta in the interval -PI to PI.
       if (ld.theta > PI) {
         ld.theta -= 2.0 * PI;
       } else if (ld.theta <= -PI) {
         ld.theta += 2.0 * PI;
       }
 
       // Update lambda as sin(theta)^2.
       double sinTheta = sin(ld.theta);
       ld.lambda = sinTheta * sinTheta;
       lambdaInterface.setLambda(ld.lambda);
     }
 
     /**
      * Gets the last lambda value received by this Histogram. This can be out-of-date w.r.t. the
      * OST's current lambda!
      *
      * @return Lambda value of the last bias added to this Histogram.
      */
     double getLastReceivedLambda() {
       return lastReceivedLambda;
     }
 
     public void setLastReceivedLambda(double lastReceivedLambda) {
       this.lastReceivedLambda = lastReceivedLambda;
     }
 
     /**
      * Gets the last dU/dL value received by this Histogram. This can be out-of-date w.r.t. the OST's
      * current dU/dL!
      *
      * @return dU/dL value of the last bias added to this Histogram.
      */
     double getLastReceivedDUDL() {
       return lastReceiveddUdL;
     }
 
     void destroy() {
       if (sendAsynchronous != null && sendAsynchronous.isAlive()) {
         double[] killMessage = new double[]{Double.NaN, Double.NaN, Double.NaN, Double.NaN};
         DoubleBuf killBuf = DoubleBuf.buffer(killMessage);
         try {
           logger.fine(" Sending the termination message.");
           world.send(rank, killBuf);
           logger.fine(" Termination message was sent successfully.");
           logger.fine(format(" Receive thread alive %b status %s", sendAsynchronous.isAlive(),
               sendAsynchronous.getState()));
         } catch (Exception ex) {
           String message = format(" Asynchronous Multiwalker OST termination signal "
               + "failed to be sent for process %d.", rank);
           logger.log(Level.SEVERE, message, ex);
         }
       } else {
         logger.fine(
             " CountReceiveThread was either not initialized, or is not alive. This is the case for the Histogram script.");
       }
     }
 
   }
 
 }