Vector MyCoefficientFunction()

ksugahar · February 16, 2024, 10:45am

I am trying to make a vector version of MyCoefficientFunction.

I am refeering to mylittelengsolve.
I find

   // ********************* Necessary **********************************
    virtual double Evaluate(const BaseMappedIntegrationPoint& mip) const override
    {
      // maybe do something with mip and trafo?
      auto pnt = mip.GetPoint();
      return pnt[0] * pnt[1];
    }

    // ******************* Performance improvements **********************

in “myCoefficient.hpp”. This I suppose is a function, but how come it is written in .hcp file? Isn’t it a header file? How do I make it a vector function? I searched in the source files, but I could not find a similar file that I could refer to.

joachim · February 16, 2024, 12:30pm

Hi Kengo,

here are some examples (unfortunately not on the tutorial level as MyLittleNGSolve):

The specialcf.normal(3) ends up as a cl_NormalVectorCF class:

github.com

NGSolve/ngsolve/blob/c4beed395f203157ed84650092bb43cd74370aae/fem/coefficient_geo.cpp#L177


      
          shared_ptr<CoefficientFunction> MakeCoordinateCoefficientFunction (int comp)
          {
            return make_shared<CoordCoefficientFunction> (comp);
          }
          
          
          
          
          
          template <int D>
          class cl_NormalVectorCF : public CoefficientFunctionNoDerivative
          {
          public:
            cl_NormalVectorCF () : CoefficientFunctionNoDerivative(D,false)
            { SetDimensions(Array<int>( { D } )); }
            // virtual int Dimension() const { return D; }
          
            virtual string GetDescription() const override
            {
              return "normal vector";
            }

The inverse of a matrix-valued CF leads to an InverseCoefficientFunction:

github.com

NGSolve/ngsolve/blob/c4beed395f203157ed84650092bb43cd74370aae/fem/coefficient_matrix.cpp#L31


      
          {
            MultMatMatCoefficientFunction :: ~MultMatMatCoefficientFunction() { }
            TransposeCoefficientFunction :: ~TransposeCoefficientFunction() { }
            TraceCoefficientFunction :: ~TraceCoefficientFunction() { }    
            IdentityCoefficientFunction :: ~IdentityCoefficientFunction () {  }
          
          
          
          
            template <int D>
            class InverseCoefficientFunction : public T_CoefficientFunction<InverseCoefficientFunction<D>>
            {
              shared_ptr<CoefficientFunction> c1;
              using BASE = T_CoefficientFunction<InverseCoefficientFunction<D>>;
              using typename BASE::T_DJC;
            public:
              InverseCoefficientFunction() = default;
              InverseCoefficientFunction (shared_ptr<CoefficientFunction> ac1)
                : T_CoefficientFunction<InverseCoefficientFunction>(D*D, ac1->IsComplex()), c1(ac1)
              {
                this->SetDimensions (ngstd::IVec<2> (D,D));

The CF can be tensor-valued (scalar, vector, matrix, or higher order tensor). This is specified by SetDimensions. The Evaluate function writes values linear into a vector (last index is inner dimension)

A difference is whether a CF is a node of the expression tree (like NormalVector), or if it depends on one or serveral input CFs (like InverseCF). Then automatic differentiation for the node shall also be provided.

Joachim