//******************************************************************************
   //
   // 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.potential.constraint;
  
  import ffx.numerics.Constraint;
  import ffx.potential.bonded.Angle;
  import ffx.potential.bonded.Atom;
  import ffx.potential.bonded.Bond;
  import java.util.ArrayList;
  import java.util.Arrays;
  import java.util.HashSet;
  import java.util.List;
  import java.util.Set;
  import java.util.logging.Logger;
  import java.util.stream.IntStream;
  import org.apache.commons.math3.linear.OpenMapRealMatrix;
  import org.apache.commons.math3.linear.QRDecomposition;
  import org.apache.commons.math3.linear.RealMatrix;
  import org.apache.commons.math3.linear.RealVectorPreservingVisitor;
  import org.apache.commons.math3.util.FastMath;
  
  public class CcmaConstraint implements Constraint {
  
    public static final double DEFAULT_CCMA_NONZERO_CUTOFF = 0.01;
    private static final Logger logger = Logger.getLogger(CcmaConstraint.class.getName());
    private static final int DEFAULT_MAX_ITERS = 150;
    // Might be slightly more elegant to have "component constraint" objects.
    // This is how OpenMM does it, though, and arrays are likely more performative.
    private final int[] atoms1;
    private final int[] atoms2;
    private final double[] lengths;
    private final int nConstraints;
    private final int[] uniqueIndices;
    private final int maxIters = DEFAULT_MAX_ITERS;
    private final double elementCutoff = DEFAULT_CCMA_NONZERO_CUTOFF;
    private final double[] reducedMasses;
    private final RealMatrix kInvSparse;
  
    /**
     * Constructs a set of bond length Constraints to be satisfied using the Constaint Constraint
     * Matrix Approximation, a parallelizable stable numeric method.
     *
     * <p>The nonzeroCutoff field specifies how large an element of K-1 needs to be to be kept;
     * smaller elements (indicating weak constraint-constraint couplings) are set to zero to make K-1
     * sparse and thus keep CCMA tractable.
     *
     * <p>Due to leaking-this issues, a thin Factory method is provided to build this.
     *
     * <p>The constrainedAngles will constrain both component bonds and the triangle-closing distance.
     * The constrainedBonds list should not include any bonds included in the constrained angles.
     *
     * @param constrainedBonds Bonds to be constrained, not included in constrainedAngles
     * @param constrainedAngles Angles to be constrained indirectly (by constructing a rigid
     *     triangle).
     * @param allAtoms All Atoms of the system, including unconstrained Atoms.
     * @param masses All masses of the system, including unconstrained atom masses.
     * @param nonzeroCutoff CCMA parameter defining how sparse/dense K-1 should be.
     */
    private CcmaConstraint(
        List<Bond> constrainedBonds,
        List<Angle> constrainedAngles,
        final Atom[] allAtoms,
        final double[] masses,
        double nonzeroCutoff) {
 
     long time = -System.nanoTime();
     int nBonds = constrainedBonds.size();
     int nAngles = constrainedAngles.size();
     assert constrainedAngles.stream()
         .flatMap((Angle a) -> a.getBondList().stream())
         .noneMatch(constrainedBonds::contains);
     nConstraints = nBonds + (3 * nAngles);
     atoms1 = new int[nConstraints];
     atoms2 = new int[nConstraints];
     lengths = new double[nConstraints];
 
     for (int i = 0; i < nBonds; i++) {
       Bond bi = constrainedBonds.get(i);
       atoms1[i] = bi.getAtom(0).getXyzIndex() - 1;
       atoms2[i] = bi.getAtom(1).getXyzIndex() - 1;
       lengths[i] = bi.bondType.distance;
     }
 
     /*logger.info(" Atoms 1: " + Arrays.toString(atoms1));
     logger.info(" Atoms 2: " + Arrays.toString(atoms2));
     logger.info(" Lengths: " + Arrays.toString(lengths));*/
 
     for (int i = 0; i < nAngles; i++) {
       int iAng = nBonds + (3 * i);
       Angle ai = constrainedAngles.get(i);
       Atom center = ai.getCentralAtom();
       Bond b1 = ai.getBond(0);
       Bond b2 = ai.getBond(1);
       Atom at0 = b1.get1_2(center);
       Atom at2 = b2.get1_2(center);
 
       double angVal = ai.angleType.angle[ai.nh];
       double dist1 = b1.bondType.distance;
       double dist2 = b2.bondType.distance;
       double dist3 = SettleConstraint.lawOfCosines(dist1, dist2, angVal);
 
       int index0 = at0.getXyzIndex() - 1;
       int index1 = center.getXyzIndex() - 1;
       int index2 = at2.getXyzIndex() - 1;
 
       atoms1[iAng] = index0;
       atoms2[iAng] = index1;
       lengths[iAng] = dist1;
 
       ++iAng;
       atoms1[iAng] = index1;
       atoms2[iAng] = index2;
       lengths[iAng] = dist2;
 
       ++iAng;
       atoms1[iAng] = index0;
       atoms2[iAng] = index2;
       lengths[iAng] = dist3;
     }
 
     uniqueIndices =
         IntStream.concat(Arrays.stream(atoms1), Arrays.stream(atoms2))
             .sorted()
             .distinct()
             .toArray();
 
     OpenMapRealMatrix kSparse = new OpenMapRealMatrix(nConstraints, nConstraints);
 
     int nAtoms = allAtoms.length;
 
     // Points from an Atom index to the Set of all Constraint indices it's involved in.
     List<Set<Integer>> atomsToConstraints = new ArrayList<>(nAtoms);
     for (int i = 0; i < nAtoms; i++) {
       // Can assume 4 constraints max, since CHONPS is rarely bonded to 5+ atoms.
       atomsToConstraints.add(new HashSet<>(4));
     }
 
     for (int i = 0; i < nConstraints; i++) {
       atomsToConstraints.get(atoms1[i]).add(i);
       atomsToConstraints.get(atoms2[i]).add(i);
     }
 
     logger.info(
         String.format(" Initial CCMA setup: %10.6g sec", 1.0E-9 * (time + System.nanoTime())));
     long subTime = -System.nanoTime();
 
     for (int i = 0; i < nConstraints; i++) {
       int atomi0 = atoms1[i];
       int atomi1 = atoms2[i];
       // DO YOU HAVE ANY IDEA HOW LONG IT TOOK FOR ME TO REALIZE MASSES WERE XYZ-INDEXED?
       double invMassI0 = 1.0 / masses[atomi0 * 3];
       double invMassI1 = 1.0 / masses[atomi1 * 3];
       double sumInv = invMassI0 + invMassI1;
 
       // Indices of all Constraints that involve either atomi0 or atomi1.
       Set<Integer> coupledConstraints = new HashSet<>(atomsToConstraints.get(atomi0));
       coupledConstraints.addAll(atomsToConstraints.get(atomi1));
 
       // Iterate over all coupled Constraints, sharing at least one common Atom with constraint i.
       for (int j : coupledConstraints) {
         // Diagonal element, coupling is obviously 1.0.
         if (i == j) {
           kSparse.setEntry(i, j, 1.0);
           continue;
         }
 
         double scale;
         int atomj0 = atoms1[j];
         int atomj1 = atoms2[j];
         int atoma; // Atom unique to constraint i.
         int atomb; // Atom shared between both constraints.
         int atomc; // Atom unique to constraint j.
 
         if (atomi0 == atomj0) {
           assert atomi1 != atomj1;
           atoma = atomi1;
           atomb = atomi0;
           atomc = atomj1;
           scale = invMassI0 / sumInv;
         } else if (atomi0 == atomj1) {
           assert atomi1 != atomj0;
           atoma = atomi1;
           atomb = atomi0;
           atomc = atomj0;
           scale = invMassI0 / sumInv;
         } else if (atomi1 == atomj0) {
           // Yes IntelliJ, it should always be true. That's why I'm asserting it.
           assert atomi0 != atomj1;
           atoma = atomi0;
           atomb = atomi1;
           atomc = atomj1;
           scale = invMassI1 / sumInv;
         } else if (atomi1 == atomj1) {
           assert atomi0 != atomj0;
           atoma = atomi1;
           atomb = atomi0;
           atomc = atomj1;
           scale = invMassI1 / sumInv;
         } else {
           throw new IllegalArgumentException(
               " Despite by necessity sharing an atom, these constraints don't share an atom.");
         }
 
         // We now have a pair of constraints a-b b-c. Find the a-b-c angle.
 
         // Search for a constraint a-c that closes the triangle.
         boolean foundAngle = false;
         for (int constraintK : atomsToConstraints.get(atoma)) {
           if (atoms1[constraintK] == atomc || atoms2[constraintK] == atomc) {
             double dab = lengths[i];
             double dbc = lengths[j];
             double dac = lengths[constraintK];
 
             // Apply the Law of Cosines to find the angle defined by the a-b b-c a-c lengths.
             double angle = (dab * dab) + (dbc * dbc) - (dac * dac);
             angle /= (2 * dab * dbc);
             // The angle is formally the cosine of its current value, but all we need is that
             // cosine.
             double coupling = scale * angle;
             kSparse.setEntry(i, j, coupling);
             foundAngle = true;
             break;
           }
         }
 
         // Otherwise, look for a force field term. In this case, the approximate matrix K
         // will not quite match the true Jacobian matrix J.
         if (!foundAngle) {
           Atom atA = allAtoms[atoma];
           Atom atB = allAtoms[atomb];
           Atom atC = allAtoms[atomc];
           Angle angleB = atA.getAngle(atB, atC);
           double angVal = angleB.angleType.angle[angleB.nh];
           double coupling = scale * FastMath.cos(FastMath.toRadians(angVal));
           kSparse.setEntry(i, j, coupling);
           foundAngle = true;
         }
 
         if (!foundAngle) {
           logger.severe(
               String.format(" Could not find the angle between coupled constraints %d, %d", i, j));
         }
       }
     }
 
     subTime += System.nanoTime();
     logger.info(
         String.format(" Time to construct K as a sparse matrix: %10.6g sec", 1.0E-9 * subTime));
 
     // K is constructed. Invert it.
     // TODO: Find a fast way to invert sparse matrices, since this is at least O(n^2).
     subTime = -System.nanoTime();
     QRDecomposition qrd = new QRDecomposition(kSparse);
     RealMatrix kInvDense = qrd.getSolver().getInverse();
     subTime += System.nanoTime();
     logger.info(String.format(" Time to invert K: %10.6g sec", 1.0E-9 * subTime));
     subTime = -System.nanoTime();
 
     kInvSparse = new OpenMapRealMatrix(nConstraints, nConstraints);
     IntStream.range(0, nConstraints)
         .
         // parallel(). // I have no idea if OpenMapRealMatrix is thread-safe.
             forEach(
             (int i) -> {
               double[] rowI = kInvDense.getRow(i);
               for (int j = 0; j < nConstraints; j++) {
                 double val = rowI[j];
                 if (Math.abs(val) > elementCutoff) {
                   kInvSparse.setEntry(i, j, val);
                 }
               }
             });
 
     // TODO: Delete this block.
     double[][] dataDense = kInvDense.getData();
     double[][] dataSparse = kInvSparse.getData();
     logger.fine(" Dense array:");
     for (int i = 0; i < nConstraints; i++) {
       logger.fine(Arrays.toString(dataDense[i]));
     }
     logger.fine(" Sparse array:");
     for (int i = 0; i < nConstraints; i++) {
       logger.fine(Arrays.toString(dataSparse[i]));
     }
 
     // TODO: Actually do this.
     reducedMasses = new double[nConstraints];
     for (int i = 0; i < nConstraints; i++) {
       int atI = atoms1[i];
       atI *= 3; // The mass array is XYZ-indexed, not atom-indexed.
       int atJ = atoms2[i];
       atJ *= 3;
       double invMassI = 1.0 / masses[atI];
       double invMassJ = 1.0 / masses[atJ];
       reducedMasses[i] = 0.5 / (invMassI + invMassJ);
     }
   }
 
   /**
    * Constructs a set of bond length Constraints to be satisfied using the Constaint Constraint
    * Matrix Approximation, a parallelizable stable numeric method.
    *
    * <p>The nonzeroCutoff field specifies how large an element of K-1 needs to be to be kept;
    * smaller elements (indicating weak constraint-constraint couplings) are set to zero to make K-1
    * sparse and thus keep CCMA tractable.
    *
    * <p>The constrainedAngles will constrain both component bonds and the triangle-closing distance.
    * The constrainedBonds list should not include any bonds included in the constrained angles.
    *
    * @param constrainedBonds Bonds to be constrained, not included in constrainedAngles
    * @param constrainedAngles Angles to be constrained indirectly (by constructing a rigid
    *     triangle).
    * @param allAtoms All Atoms of the system, including unconstrained Atoms.
    * @param masses All masses of the system, including unconstrained atom masses.
    * @param nonzeroCutoff CCMA parameter defining how sparse/dense K-1 should be.
    * @return Returns a new CcmaConstraint instance.
    */
   public static CcmaConstraint ccmaFactory(
       List<Bond> constrainedBonds,
       List<Angle> constrainedAngles,
       final Atom[] allAtoms,
       final double[] masses,
       double nonzeroCutoff) {
     CcmaConstraint newC =
         new CcmaConstraint(constrainedBonds, constrainedAngles, allAtoms, masses, nonzeroCutoff);
     constrainedBonds.forEach((Bond b) -> b.setConstraint(newC));
     constrainedAngles.forEach((Angle a) -> a.setConstraint(newC));
     return newC;
   }
 
   /** {@inheritDoc} */
   @Override
   public void applyConstraintToStep(double[] xPrior, double[] xNew, double[] masses, double tol) {
     applyConstraints(xPrior, xNew, masses, false, tol);
   }
 
   /** {@inheritDoc} */
   @Override
   public void applyConstraintToVelocities(double[] x, double[] v, double[] masses, double tol) {
     applyConstraints(x, v, masses, true, tol);
   }
 
   /** {@inheritDoc} */
   @Override
   public int[] constrainedAtomIndices() {
     return Arrays.copyOf(uniqueIndices, uniqueIndices.length);
   }
 
   /** {@inheritDoc} */
   @Override
   public boolean constraintSatisfied(double[] x, double tol) {
     return false;
   }
 
   /** {@inheritDoc} */
   @Override
   public boolean constraintSatisfied(double[] x, double[] v, double xTol, double vTol) {
     return false;
   }
 
   /** {@inheritDoc} */
   @Override
   public int getNumDegreesFrozen() {
     return nConstraints;
   }
 
   /**
    * Much of the math for applying to coordinates/velocities is the same. As such, OpenMM just uses a
    * single driver method with a flag to indicate velocities or positions.
    *
    * @param xPrior Either pre-step coordinates (positions) or current coordinates (velocities)
    * @param output Output vector, either constrained coordinates or velocities
    * @param masses Masses of all particles
    * @param constrainV Whether to apply constraints to velocities, or to coordinates.
    * @param tol Numerical tolerance.
    */
   private void applyConstraints(
       double[] xPrior, double[] output, double[] masses, boolean constrainV, double tol) {
     if (xPrior == output) {
       throw new IllegalArgumentException(" xPrior and output must be different arrays!");
     }
     long time = -System.nanoTime();
 
     /*// Distance vector for all constraints. TODO: Flatten to 1D array.
     double[][] r_ij = new double[nConstraints][3];
     // Distance magnitude for all constraints.
     double[] d_ij = new double[nConstraints];
     // Squared distance magnitude for all constraints.
     double[] d_ij2 = new double[nConstraints];
 
     for (int i = 0; i < nConstraints; i++) {
         int atom1 = 3 * atoms1[i];
         int atom2 = 3 * atoms2[i];
         for (int j = 0; j < 3; j++) {
             r_ij[i][j] = xPrior[atom1 + j] - xPrior[atom2 + j];
         }
         d_ij2[i] = VectorMath.dot(r_ij[i], r_ij[i]);
     }
 
     double lowerTol = 1 - 2*tol + tol*tol;
     // TODO: Determine if adding tol+tol to upperTol is a typo in OpenMM. Also why they do it in the first place.
     double upperTol = 1 + 2*tol + tol*tol;
 
     int nConverged = 0;
     double[] constraintDelta = new double[nConstraints];
     double[] tempDelta = new double[nConstraints];
 
     // Main CCMA loop.
     for (int constraintIter = 0; constraintIter <= maxIters; constraintIter++) {
         if (constraintIter >= maxIters) {
             throw new IllegalArgumentException(String.format(" CCMA constraint failed to converge in %d iterations!", maxIters));
         }
         nConverged = 0;
         double rmsd = 0;
         for (int i = 0; i < nConstraints; i++) {
             int atom1 = atoms1[i] * 3;
             int atom2 = atoms2[i] * 3;
 
             // Separation vector I-J at this iteration.
             double[] rp_ij = new double[3];
             for (int j = 0; j < 3; j++) {
                 rp_ij[j] = output[atom1 + j] - output[atom2 + j];
             }
             if (constrainV) {
                 double rrpr = VectorMath.dot(rp_ij, r_ij[i]);
                 constraintDelta[i] = -2 * reducedMasses[i] * rrpr / d_ij2[i];
                 if (Math.abs(constraintDelta[i]) <= tol) {
                     ++nConverged;
                 }
             } else {
                 double rp2 = VectorMath.dot(rp_ij, rp_ij);
                 double dist2 = lengths[i] * lengths[i];
                 double diff = dist2 - rp2;
                 double rrpr = VectorMath.dot(rp_ij, r_ij[i]);
                 constraintDelta[i] = reducedMasses[i] * diff / rrpr;
                 if (rp2 >= lowerTol * dist2 && rp2 <= upperTol * dist2) {
                     ++nConverged;
                 }
             }
         }
 
         // Test if the last iteration satisfied all constraints.
         if (nConverged == nConstraints) {
             break;
         }
 
         if (kInvSparse != null) {
             for (int i = 0; i < nConstraints; i++) {
                 RealVector row = kInvSparse.getRowVector(i);
                 RealVectorPreservingVisitor neo = new MatrixWalker(constraintDelta);
                 double sum = row.walkInOptimizedOrder(neo);
                 tempDelta[i] = sum;
             }
             System.arraycopy(tempDelta, 0, constraintDelta, 0, nConstraints);
         }
         for (int i = 0; i < nConstraints; i++) {
             int atom1 = atoms1[i];
             int atom2 = atoms2[i];
             int at13 = 3 * atom1;
             int at23 = 3 * atom2;
 
             double[] dr = new double[3];
             VectorMath.scalar(r_ij[i], constraintDelta[i], dr);
             for (int j = 0; j < 3; j++) {
                 output[at13 + j] = dr[j] / masses[at13];
                 output[at23 + j] = dr[j] / masses[at23];
             }
         }
         logger.info(String.format(" %d converged at iteration %d", nConverged, constraintIter));
     }
     time += System.nanoTime();
     logger.info(String.format(" Application of CCMA constraint: %10.4g sec", (time * 1.0E-9)));*/
   }
 
   private static class MatrixWalker implements RealVectorPreservingVisitor {
 
     private final double[] constraintDelta; // DO NOT MODIFY.
     private double sum = 0.0;
 
     MatrixWalker(final double[] cDelta) {
       constraintDelta = cDelta;
     }
 
     /** {@inheritDoc} */
     @Override
     public double end() {
       return sum;
     }
 
     /** {@inheritDoc} */
     @Override
     public void start(int dimension, int start, int end) {
     }
 
     /** {@inheritDoc} */
     @Override
     public void visit(int index, double value) {
       sum += (value * constraintDelta[index]);
     }
   }
 }