16 boost::shared_ptr<BaseFunctionCtx> ctx_ptr) {
21 int nb_gauss_pts = pts.size2();
28 "Wrong dimension of pts, should be at least 3 rows with coordinates");
35 data.
dataOnEntities[MBVERTEX][0].getN(base).resize(nb_gauss_pts, 2,
39 &pts(0, 0), nb_gauss_pts);
45 (
unsigned int)nb_gauss_pts)
47 "Base functions or nodes has wrong number of integration points "
51 data.
dataOnEntities[MBVERTEX][0].getDiffN(base).resize(nb_gauss_pts, 2 * 1,
53 for (
auto gg = 0; gg != nb_gauss_pts; ++gg)
102 const int nb_gauss_pts = pts.size2();
105 auto &ptr = data.getBBAlphaIndicesSharedPtr(
field_name);
119 auto &ptr = data.getBBDiffNSharedPtr(
field_name);
126 auto &get_diff_n = get_diff_base(data.
dataOnEntities[MBVERTEX][0]);
127 get_n.resize(nb_gauss_pts, 2,
false);
128 get_diff_n.resize(nb_gauss_pts, 2,
false);
133 (
unsigned int)nb_gauss_pts)
135 "Base functions or nodes has wrong number of integration points "
141 vertex_alpha.resize(2, 2,
false);
142 vertex_alpha(0, 0) = data.
dataOnEntities[MBVERTEX][0].getBBNodeOrder()[0];
143 vertex_alpha(0, 1) = 0;
144 vertex_alpha(1, 0) = 0;
145 vertex_alpha(1, 1) = data.
dataOnEntities[MBVERTEX][0].getBBNodeOrder()[1];
148 1, nb_gauss_pts, vertex_alpha.size1(), &vertex_alpha(0, 0), &
lambda(0, 0),
150 std::array<double, 2> f = {
151 boost::math::factorial<double>(
153 boost::math::factorial<double>(
156 for (
int g = 0;
g != nb_gauss_pts; ++
g)
157 for (
int n = 0;
n != 2; ++
n) {
159 get_diff_n(
g,
n) *= f[
n];
165 "Wrong size ent of ent data");
172 get_n.resize(nb_gauss_pts, nb_dofs,
false);
173 get_diff_n.resize(nb_gauss_pts, nb_dofs,
false);
177 edge_alpha.resize(nb_dofs, 2);
180 order, nb_gauss_pts, edge_alpha.size1(), &edge_alpha(0, 0),
185 get_n.resize(nb_gauss_pts, 0,
false);
186 get_diff_n.resize(nb_gauss_pts, 0,
false);
197 PetscErrorCode (*base_polynomials)(
int p,
double s,
double *diff_s,
double *
L,
198 double *
diffL,
const int dim) =
201 int nb_gauss_pts = pts.size2();
203 const int side_number = 0;
220 for (
int gg = 0; gg != nb_gauss_pts; gg++) {
222 double s = 2 * pts(0, gg) - 1;
230 &*
L.data().begin(), &*
diffL.data().begin(), 1);
232 for (
unsigned int pp = 0;
262 int nb_gauss_pts = pts.size2();
264 const int side_number = 0;
275 double diff_shape[] = {-1, 1};
279 for (
int gg = 0; gg != nb_gauss_pts; gg++) {
280 const double mu0 = 1.0 - pts(0, gg);
281 const double mu1 = pts(0, gg);
287 order, &*shape.data().begin(), diff_shape,
288 &*data.
dataOnEntities[MBEDGE][side_number].getN(base).data().begin(),
321 PetscErrorCode (*base_polynomials)(
int p,
double s,
double *diff_s,
double *
L,
322 double *
diffL,
const int dim) =
328 int nb_gauss_pts = pts.size2();
330 constexpr int side_number = 0;
339 &*data.
dataOnEntities[MBEDGE][side_number].getN(base).data().begin();
341 &*data.
dataOnEntities[MBEDGE][side_number].getDiffN(base).data().begin();
348 for (
int gg = 0; gg != nb_gauss_pts; gg++) {
349 double mu = 2 * pts(0, gg) - 1;
353 fun_n[qd_shift +
n] =
l[
n];
354 diff_fun_n[qd_shift +
n] = diff_l[
n];
369 "Make no sense, unless problem is 2d (2d not implemented yet)");
379 PetscErrorCode (*base_polynomials)(
int p,
double s,
double *diff_s,
double *
L,
380 double *
diffL,
const int dim) =
383 int nb_gauss_pts = pts.size2();
393 data.
dataOnEntities[MBEDGE][0].getN(base).resize(nb_gauss_pts, 3 * nb_dofs,
395 data.
dataOnEntities[MBEDGE][0].getDiffN(base).resize(nb_gauss_pts, 0,
401 &*data.
dataOnEntities[MBEDGE][0].getN(base).data().begin(),
nullptr,
402 nb_gauss_pts, base_polynomials);
405 data.
dataOnEntities[MBEDGE][0].getN(base).resize(nb_gauss_pts, 0,
false);
406 data.
dataOnEntities[MBEDGE][0].getDiffN(base).resize(nb_gauss_pts, 0,
420 int nb_gauss_pts = pts.size2();
425 "No data structure to store base functions");
431 data.
dataOnEntities[MBEDGE][0].getN(base).resize(nb_gauss_pts, 3 * nb_dofs,
433 data.
dataOnEntities[MBEDGE][0].getDiffN(base).resize(nb_gauss_pts, 0,
444 data.
dataOnEntities[MBEDGE][0].getN(base).resize(nb_gauss_pts, 0,
false);
445 data.
dataOnEntities[MBEDGE][0].getDiffN(base).resize(nb_gauss_pts, 0,
FieldApproximationBase
approximation base
@ AINSWORTH_LEGENDRE_BASE
Ainsworth Cole (Legendre) approx. base nme:nme847.
@ AINSWORTH_BERNSTEIN_BEZIER_BASE
#define MoFEMFunctionReturnHot(a)
Last executable line of each PETSc function used for error handling. Replaces return()
@ L2
field with C-1 continuity
@ HCURL
field with continuous tangents
@ HDIV
field with continuous normal traction
#define MoFEMFunctionBegin
First executable line of each MoFEM function, used for error handling. Final line of MoFEM functions ...
@ MOFEM_DATA_INCONSISTENCY
static const char *const ApproximationBaseNames[]
#define MoFEMFunctionReturn(a)
Last executable line of each PETSc function used for error handling. Replaces return()
#define CHKERR
Inline error check.
#define MoFEMFunctionBeginHot
First executable line of each MoFEM function, used for error handling. Final line of MoFEM functions ...
PetscErrorCode Lobatto_polynomials(int p, double s, double *diff_s, double *L, double *diffL, const int dim)
Calculate Lobatto base functions fuentes2015353.
#define NBEDGE_DEMKOWICZ_HCURL(P)
#define NBEDGE_H1(P)
Number of base function on edge for H1 space.
#define NBEDGE_L2(P)
Number of base functions on edge from L2 space.
#define NBEDGE_AINSWORTH_HCURL(P)
const double v
phase velocity of light in medium (cm/ns)
const double n
refractive index of diffusive medium
FTensor::Index< 'l', 3 > l
MoFEMErrorCode H1_BubbleShapeFunctions_ONSEGMENT(int p, double *N, double *diffN, double *bubbleN, double *diff_bubbleN, int nb_integration_pts)
PetscErrorCode MoFEMErrorCode
MoFEM/PETSc error code.
UBlasMatrix< int > MatrixInt
UBlasMatrix< double > MatrixDouble
implementation of Data Operators for Forces and Sources
MoFEMErrorCode Hcurl_Demkowicz_EdgeBaseFunctions_MBEDGE(int sense, int p, double *n, double *diff_n, double *phi, double *diff_phi, int nb_integration_pts)
Edge based H-curl base functions on edge.
MoFEMErrorCode Hcurl_Ainsworth_EdgeBaseFunctions_MBTET_ON_EDGE(int sense, int p, double *N, double *diffN, double *edgeN, double *diff_edgeN, int nb_integration_pts, PetscErrorCode(*base_polynomials)(int p, double s, double *diff_s, double *L, double *diffL, const int dim))
Edge based H-curl base functions on edge.
constexpr auto field_name
static MoFEMErrorCode baseFunctionsEdge(const int N, const int gdim, const int n_alpha, const int *alpha, const double *lambda, const double *grad_lambda, double *base, double *grad_base)
static MoFEMErrorCode generateIndicesEdgeEdge(const int N, int *alpha)
Calculate base functions on tetrahedral.
MoFEMErrorCode getValueH1AinsworthBase(MatrixDouble &pts)
MoFEMErrorCode getValue(MatrixDouble &pts, boost::shared_ptr< BaseFunctionCtx > ctx_ptr)
MoFEMErrorCode getValueH1DemkowiczBase(MatrixDouble &pts)
MoFEMErrorCode getValueHdiv(MatrixDouble &pts)
MoFEMErrorCode getValueL2AinsworthBase(MatrixDouble &pts)
MoFEMErrorCode getValueHcurl(MatrixDouble &pts)
MoFEMErrorCode query_interface(boost::typeindex::type_index type_index, UnknownInterface **iface) const
MoFEMErrorCode getValueH1BernsteinBezierBase(MatrixDouble &pts)
MoFEMErrorCode getValueL2(MatrixDouble &pts)
MoFEMErrorCode getValueHcurlDemkowiczBase(MatrixDouble &pts)
EntPolynomialBaseCtx * cTx
MoFEMErrorCode getValueHcurlAinsworthBase(MatrixDouble &pts)
MoFEMErrorCode getValueH1(MatrixDouble &pts)
MoFEMErrorCode getValueL2DemkowiczBase(MatrixDouble &pts)
Class used to pass element data to calculate base functions on tet,triangle,edge.
const FieldApproximationBase copyNodeBase
const std::string fieldName
PetscErrorCode(* basePolynomialsType0)(int p, double s, double *diff_s, double *L, double *diffL, const int dim)
const FieldApproximationBase bAse
data structure for finite element entity
std::array< std::bitset< LASTSPACE >, MBMAXTYPE > spacesOnEntities
spaces on entity types
std::array< boost::ptr_vector< EntData >, MBMAXTYPE > dataOnEntities
base class for all interface classes
MoFEMErrorCode getInterface(IFACE *&iface) const
Get interface reference to pointer of interface.