quadrature_gauss_3D.C

Go to the documentation of this file.

// The libMesh Finite Element Library.
// Copyright (C) 2002-2018 Benjamin S. Kirk, John W. Peterson, Roy H. Stogner

// This library is free software; you can redistribute it and/or
// modify it under the terms of the GNU Lesser General Public
// License as published by the Free Software Foundation; either
// version 2.1 of the License, or (at your option) any later version.

// This library is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
// Lesser General Public License for more details.

// You should have received a copy of the GNU Lesser General Public
// License along with this library; if not, write to the Free Software
// Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA



// Local includes
#include "libmesh/quadrature_gauss.h"
#include "libmesh/quadrature_conical.h"
#include "libmesh/quadrature_gm.h"

namespace libMesh
{

void QGauss::init_3D(const ElemType type_in,
                     unsigned int p)
{
#if LIBMESH_DIM == 3

  //-----------------------------------------------------------------------
  // 3D quadrature rules
  switch (type_in)
    {
      //---------------------------------------------
      // Hex quadrature rules
    case HEX8:
    case HEX20:
    case HEX27:
      {
        // We compute the 3D quadrature rule as a tensor
        // product of the 1D quadrature rule.
        QGauss q1D(1,_order);
        q1D.init(EDGE2,p);

        tensor_product_hex( q1D );

        return;
      }



      //---------------------------------------------
      // Tetrahedral quadrature rules
    case TET4:
    case TET10:
      {
        switch(_order + 2*p)
          {
            // Taken from pg. 222 of "The finite element method," vol. 1
            // ed. 5 by Zienkiewicz & Taylor
          case CONSTANT:
          case FIRST:
            {
              // Exact for linears
              _points.resize(1);
              _weights.resize(1);


              _points[0](0) = .25;
              _points[0](1) = .25;
              _points[0](2) = .25;

              _weights[0] = Real(1)/6;

              return;
            }
          case SECOND:
            {
              // Exact for quadratics
              _points.resize(4);
              _weights.resize(4);


              const Real a = .585410196624969;
              const Real b = .138196601125011;

              _points[0](0) = a;
              _points[0](1) = b;
              _points[0](2) = b;

              _points[1](0) = b;
              _points[1](1) = a;
              _points[1](2) = b;

              _points[2](0) = b;
              _points[2](1) = b;
              _points[2](2) = a;

              _points[3](0) = b;
              _points[3](1) = b;
              _points[3](2) = b;



              _weights[0] = Real(1)/24;
              _weights[1] = _weights[0];
              _weights[2] = _weights[0];
              _weights[3] = _weights[0];

              return;
            }



            // Can be found in the class notes
            // http://www.cs.rpi.edu/~flaherje/FEM/fem6.ps
            // by Flaherty.
            //
            // Caution: this rule has a negative weight and may be
            // unsuitable for some problems.
            // Exact for cubics.
            //
            // Note: Keast (see ref. elsewhere in this file) also gives
            // a third-order rule with positive weights, but it contains points
            // on the ref. elt. boundary, making it less suitable for FEM calculations.
          case THIRD:
            {
              if (allow_rules_with_negative_weights)
                {
                  _points.resize(5);
                  _weights.resize(5);


                  _points[0](0) = .25;
                  _points[0](1) = .25;
                  _points[0](2) = .25;

                  _points[1](0) = .5;
                  _points[1](1) = Real(1)/6;
                  _points[1](2) = Real(1)/6;

                  _points[2](0) = Real(1)/6;
                  _points[2](1) = .5;
                  _points[2](2) = Real(1)/6;

                  _points[3](0) = Real(1)/6;
                  _points[3](1) = Real(1)/6;
                  _points[3](2) = .5;

                  _points[4](0) = Real(1)/6;
                  _points[4](1) = Real(1)/6;
                  _points[4](2) = Real(1)/6;


                  _weights[0] = Real(-2)/15;
                  _weights[1] = .075;
                  _weights[2] = _weights[1];
                  _weights[3] = _weights[1];
                  _weights[4] = _weights[1];

                  return;
                } // end if (allow_rules_with_negative_weights)
              else
                {
                  // If a rule with positive weights is required, the 2x2x2 conical
                  // product rule is third-order accurate and has less points than
                  // the next-available positive-weight rule at FIFTH order.
                  QConical conical_rule(3, _order);
                  conical_rule.init(type_in, p);

                  // Swap points and weights with the about-to-be destroyed rule.
                  _points.swap (conical_rule.get_points() );
                  _weights.swap(conical_rule.get_weights());

                  return;
                }
              // Note: if !allow_rules_with_negative_weights, fall through to next case.
            }



            // Originally a Keast rule,
            //    Patrick Keast,
            //    Moderate Degree Tetrahedral Quadrature Formulas,
            //    Computer Methods in Applied Mechanics and Engineering,
            //    Volume 55, Number 3, May 1986, pages 339-348.
            //
            // Can also be found the class notes
            // http://www.cs.rpi.edu/~flaherje/FEM/fem6.ps
            // by Flaherty.
            //
            // Caution: this rule has a negative weight and may be
            // unsuitable for some problems.
          case FOURTH:
            {
              if (allow_rules_with_negative_weights)
                {
                  _points.resize(11);
                  _weights.resize(11);

                  // The raw data for the quadrature rule.
                  const Real rule_data[3][4] = {
                    {0.250000000000000000e+00,                         0.,                            0.,  -0.131555555555555556e-01},  // 1
                    {0.785714285714285714e+00,   0.714285714285714285e-01,                            0.,   0.762222222222222222e-02},  // 4
                    {0.399403576166799219e+00,                         0.,      0.100596423833200785e+00,   0.248888888888888889e-01}   // 6
                  };


                  // Now call the keast routine to generate _points and _weights
                  keast_rule(rule_data, 3);

                  return;
                } // end if (allow_rules_with_negative_weights)
              // Note: if !allow_rules_with_negative_weights, fall through to next case.
            }

            libmesh_fallthrough();


            // Walkington's fifth-order 14-point rule from
            // "Quadrature on Simplices of Arbitrary Dimension"
            //
            // We originally had a Keast rule here, but this rule had
            // more points than an equivalent rule by Walkington and
            // also contained points on the boundary of the ref. elt,
            // making it less suitable for FEM calculations.
          case FIFTH:
            {
              _points.resize(14);
              _weights.resize(14);

              // permutations of these points and suitably-modified versions of
              // these points are the quadrature point locations
              const Real a[3] = {0.31088591926330060980,    // a1 from the paper
                                 0.092735250310891226402,   // a2 from the paper
                                 0.045503704125649649492};  // a3 from the paper

              // weights.  a[] and wt[] are the only floating-point inputs required
              // for this rule.
              const Real wt[3] = {0.018781320953002641800,    // w1 from the paper
                                  0.012248840519393658257,    // w2 from the paper
                                  0.0070910034628469110730};  // w3 from the paper

              // The first two sets of 4 points are formed in a similar manner
              for (unsigned int i=0; i<2; ++i)
                {
                  // Where we will insert values into _points and _weights
                  const unsigned int offset=4*i;

                  // Stuff points and weights values into their arrays
                  const Real b = 1. - 3.*a[i];

                  // Here are the permutations.  Order of these is not important,
                  // all have the same weight
                  _points[offset + 0] = Point(a[i], a[i], a[i]);
                  _points[offset + 1] = Point(a[i],    b, a[i]);
                  _points[offset + 2] = Point(   b, a[i], a[i]);
                  _points[offset + 3] = Point(a[i], a[i],    b);

                  // These 4 points all have the same weights
                  for (unsigned int j=0; j<4; ++j)
                    _weights[offset + j] = wt[i];
                } // end for


              {
                // The third set contains 6 points and is formed a little differently
                const unsigned int offset = 8;
                const Real b = 0.5*(1. - 2.*a[2]);

                // Here are the permutations.  Order of these is not important,
                // all have the same weight
                _points[offset + 0] = Point(b   ,    b, a[2]);
                _points[offset + 1] = Point(b   , a[2], a[2]);
                _points[offset + 2] = Point(a[2], a[2],    b);
                _points[offset + 3] = Point(a[2],    b, a[2]);
                _points[offset + 4] = Point(   b, a[2],    b);
                _points[offset + 5] = Point(a[2],    b,    b);

                // These 6 points all have the same weights
                for (unsigned int j=0; j<6; ++j)
                  _weights[offset + j] = wt[2];
              }


              // Original rule by Keast, unsuitable because it has points on the
              // reference element boundary!
              //       _points.resize(15);
              //       _weights.resize(15);

              //       _points[0](0) = 0.25;
              //       _points[0](1) = 0.25;
              //       _points[0](2) = 0.25;

              //       {
              // const Real a = 0.;
              // const Real b = Real(1)/3;

              // _points[1](0) = a;
              // _points[1](1) = b;
              // _points[1](2) = b;

              // _points[2](0) = b;
              // _points[2](1) = a;
              // _points[2](2) = b;

              // _points[3](0) = b;
              // _points[3](1) = b;
              // _points[3](2) = a;

              // _points[4](0) = b;
              // _points[4](1) = b;
              // _points[4](2) = b;
              //       }
              //       {
              // const Real a = Real(8)/11;
              // const Real b = Real(1)/11;

              // _points[5](0) = a;
              // _points[5](1) = b;
              // _points[5](2) = b;

              // _points[6](0) = b;
              // _points[6](1) = a;
              // _points[6](2) = b;

              // _points[7](0) = b;
              // _points[7](1) = b;
              // _points[7](2) = a;

              // _points[8](0) = b;
              // _points[8](1) = b;
              // _points[8](2) = b;
              //       }
              //       {
              // const Real a = 0.066550153573664;
              // const Real b = 0.433449846426336;

              // _points[9](0) = b;
              // _points[9](1) = a;
              // _points[9](2) = a;

              // _points[10](0) = a;
              // _points[10](1) = a;
              // _points[10](2) = b;

              // _points[11](0) = a;
              // _points[11](1) = b;
              // _points[11](2) = b;

              // _points[12](0) = b;
              // _points[12](1) = a;
              // _points[12](2) = b;

              // _points[13](0) = b;
              // _points[13](1) = b;
              // _points[13](2) = a;

              // _points[14](0) = a;
              // _points[14](1) = b;
              // _points[14](2) = a;
              //       }

              //       _weights[0]  = 0.030283678097089;
              //       _weights[1]  = 0.006026785714286;
              //       _weights[2]  = _weights[1];
              //       _weights[3]  = _weights[1];
              //       _weights[4]  = _weights[1];
              //       _weights[5]  = 0.011645249086029;
              //       _weights[6]  = _weights[5];
              //       _weights[7]  = _weights[5];
              //       _weights[8]  = _weights[5];
              //       _weights[9]  = 0.010949141561386;
              //       _weights[10] = _weights[9];
              //       _weights[11] = _weights[9];
              //       _weights[12] = _weights[9];
              //       _weights[13] = _weights[9];
              //       _weights[14] = _weights[9];

              return;
            }




            // This rule is originally from Keast:
            //    Patrick Keast,
            //    Moderate Degree Tetrahedral Quadrature Formulas,
            //    Computer Methods in Applied Mechanics and Engineering,
            //    Volume 55, Number 3, May 1986, pages 339-348.
            //
            // It is accurate on 6th-degree polynomials and has 24 points
            // vs. 64 for the comparable conical product rule.
            //
            // Values copied 24th June 2008 from:
            // http://people.scs.fsu.edu/~burkardt/f_src/keast/keast.f90
          case SIXTH:
            {
              _points.resize (24);
              _weights.resize(24);

              // The raw data for the quadrature rule.
              const Real rule_data[4][4] = {
                {0.356191386222544953e+00 , 0.214602871259151684e+00 ,                       0., 0.00665379170969464506e+00}, // 4
                {0.877978124396165982e+00 , 0.0406739585346113397e+00,                       0., 0.00167953517588677620e+00}, // 4
                {0.0329863295731730594e+00, 0.322337890142275646e+00 ,                       0., 0.00922619692394239843e+00}, // 4
                {0.0636610018750175299e+00, 0.269672331458315867e+00 , 0.603005664791649076e+00, 0.00803571428571428248e+00}  // 12
              };


              // Now call the keast routine to generate _points and _weights
              keast_rule(rule_data, 4);

              return;
            }


            // Keast's 31 point, 7th-order rule contains points on the reference
            // element boundary, so we've decided not to include it here.
            //
            // Keast's 8th-order rule has 45 points.  and a negative
            // weight, so if you've explicitly disallowed such rules
            // you will fall through to the conical product rules
            // below.
          case SEVENTH:
          case EIGHTH:
            {
              if (allow_rules_with_negative_weights)
                {
                  _points.resize (45);
                  _weights.resize(45);

                  // The raw data for the quadrature rule.
                  const Real rule_data[7][4] = {
                    {0.250000000000000000e+00,                        0.,                        0.,   -0.393270066412926145e-01},  // 1
                    {0.617587190300082967e+00,  0.127470936566639015e+00,                        0.,    0.408131605934270525e-02},  // 4
                    {0.903763508822103123e+00,  0.320788303926322960e-01,                        0.,    0.658086773304341943e-03},  // 4
                    {0.497770956432810185e-01,                        0.,  0.450222904356718978e+00,    0.438425882512284693e-02},  // 6
                    {0.183730447398549945e+00,                        0.,  0.316269552601450060e+00,    0.138300638425098166e-01},  // 6
                    {0.231901089397150906e+00,  0.229177878448171174e-01,  0.513280033360881072e+00,    0.424043742468372453e-02},  // 12
                    {0.379700484718286102e-01,  0.730313427807538396e+00,  0.193746475248804382e+00,    0.223873973961420164e-02}   // 12
                  };


                  // Now call the keast routine to generate _points and _weights
                  keast_rule(rule_data, 7);

                  return;
                } // end if (allow_rules_with_negative_weights)
              // Note: if !allow_rules_with_negative_weights, fall through to next case.
            }

            libmesh_fallthrough();


            // Fall back on Grundmann-Moller or Conical Product rules at high orders.
          default:
            {
              if ((allow_rules_with_negative_weights) && (_order + 2*p < 34))
                {
                  // The Grundmann-Moller rules are defined to arbitrary order and
                  // can have significantly fewer evaluation points than conical product
                  // rules.  If you allow rules with negative weights, the GM rules
                  // will be more efficient for degree up to 33 (for degree 34 and
                  // higher, CP is more efficient!) but may be more susceptible
                  // to round-off error.  Safest is to disallow rules with negative
                  // weights, but this decision should be made on a case-by-case basis.
                  QGrundmann_Moller gm_rule(3, _order);
                  gm_rule.init(type_in, p);

                  // Swap points and weights with the about-to-be destroyed rule.
                  _points.swap (gm_rule.get_points() );
                  _weights.swap(gm_rule.get_weights());

                  return;
                }

              else
                {
                  // The following quadrature rules are generated as
                  // conical products.  These tend to be non-optimal
                  // (use too many points, cluster points in certain
                  // regions of the domain) but they are quite easy to
                  // automatically generate using a 1D Gauss rule on
                  // [0,1] and two 1D Jacobi-Gauss rules on [0,1].
                  QConical conical_rule(3, _order);
                  conical_rule.init(type_in, p);

                  // Swap points and weights with the about-to-be destroyed rule.
                  _points.swap (conical_rule.get_points() );
                  _weights.swap(conical_rule.get_weights());

                  return;
                }
            }
          }
      } // end case TET4,TET10



      //---------------------------------------------
      // Prism quadrature rules
    case PRISM6:
    case PRISM15:
    case PRISM18:
      {
        // We compute the 3D quadrature rule as a tensor
        // product of the 1D quadrature rule and a 2D
        // triangle quadrature rule

        QGauss q1D(1,_order);
        QGauss q2D(2,_order);

        // Initialize
        q1D.init(EDGE2,p);
        q2D.init(TRI3,p);

        tensor_product_prism(q1D, q2D);

        return;
      }



      //---------------------------------------------
      // Pyramid
    case PYRAMID5:
    case PYRAMID13:
    case PYRAMID14:
      {
        // We compute the Pyramid rule as a conical product of a
        // Jacobi rule with alpha==2 on the interval [0,1] two 1D
        // Gauss rules with suitably modified points.  The idea comes
        // from: Stroud, A.H. "Approximate Calculation of Multiple
        // Integrals."
        QConical conical_rule(3, _order);
        conical_rule.init(type_in, p);

        // Swap points and weights with the about-to-be destroyed rule.
        _points.swap (conical_rule.get_points() );
        _weights.swap(conical_rule.get_weights());

        return;

      }



      //---------------------------------------------
      // Unsupported type
    default:
      libmesh_error_msg("ERROR: Unsupported type: " << type_in);
    }
#endif
}

} // namespace libMesh