DUNE PDELab (2.7)

convectiondiffusiondg.hh
1// -*- tab-width: 2; indent-tabs-mode: nil -*-
2#ifndef DUNE_PDELAB_LOCALOPERATOR_CONVECTIONDIFFUSIONDG_HH
3#define DUNE_PDELAB_LOCALOPERATOR_CONVECTIONDIFFUSIONDG_HH
4
7
8#include<dune/geometry/referenceelements.hh>
9
10#include<dune/pdelab/common/quadraturerules.hh>
11#include<dune/pdelab/common/function.hh>
12#include<dune/pdelab/localoperator/pattern.hh>
13#include<dune/pdelab/localoperator/flags.hh>
14#include<dune/pdelab/localoperator/idefault.hh>
15#include<dune/pdelab/localoperator/defaultimp.hh>
16#include<dune/pdelab/finiteelement/localbasiscache.hh>
17
18#include"convectiondiffusionparameter.hh"
19
26namespace Dune {
27 namespace PDELab {
28
29 struct ConvectionDiffusionDGMethod
30 {
31 enum Type { NIPG, SIPG, IIPG };
32 };
33
34 struct ConvectionDiffusionDGWeights
35 {
36 enum Type { weightsOn, weightsOff };
37 };
38
54 template<typename T, typename FiniteElementMap>
59 public Dune::PDELab::InstationaryLocalOperatorDefaultMethods<typename T::Traits::RangeFieldType>
60 {
61 enum { dim = T::Traits::GridViewType::dimension };
62
63 using Real = typename T::Traits::RangeFieldType;
64 using BCType = typename ConvectionDiffusionBoundaryConditions::Type;
65
66 public:
67 // pattern assembly flags
68 enum { doPatternVolume = true };
69 enum { doPatternSkeleton = true };
70
71 // residual assembly flags
72 enum { doAlphaVolume = true };
73 enum { doAlphaSkeleton = true };
74 enum { doAlphaBoundary = true };
75 enum { doLambdaVolume = true };
76
86 ConvectionDiffusionDGMethod::Type method_=ConvectionDiffusionDGMethod::SIPG,
87 ConvectionDiffusionDGWeights::Type weights_=ConvectionDiffusionDGWeights::weightsOn,
88 Real alpha_=1.0,
89 int intorderadd_=0
90 )
91 : param(param_)
92 , method(method_)
93 , weights(weights_)
94 , alpha(alpha_)
95 , intorderadd(intorderadd_)
96 , quadrature_factor(2)
97 , cache(20)
98 {
99 theta = 1.0;
100 if (method==ConvectionDiffusionDGMethod::SIPG) theta = -1.0;
101 if (method==ConvectionDiffusionDGMethod::IIPG) theta = 0.0;
102 }
103
104 // volume integral depending on test and ansatz functions
105 template<typename EG, typename LFSU, typename X, typename LFSV, typename R>
106 void alpha_volume (const EG& eg, const LFSU& lfsu, const X& x, const LFSV& lfsv, R& r) const
107 {
108 // define types
109 using RF = typename LFSU::Traits::FiniteElementType::
110 Traits::LocalBasisType::Traits::RangeFieldType;
111 using size_type = typename LFSU::Traits::SizeType;
112
113 // dimensions
114 const int dim = EG::Entity::dimension;
115 const int order = std::max(lfsu.finiteElement().localBasis().order(),
116 lfsv.finiteElement().localBasis().order());
117
118 // Get cell
119 const auto& cell = eg.entity();
120
121 // Get geometry
122 auto geo = eg.geometry();
123
124 // evaluate diffusion tensor at cell center, assume it is constant over elements
125 auto ref_el = referenceElement(geo);
126 auto localcenter = ref_el.position(0,0);
127 auto A = param.A(cell,localcenter);
128
129 // Initialize vectors outside for loop
130 std::vector<Dune::FieldVector<RF,dim> > gradphi(lfsu.size());
131 std::vector<Dune::FieldVector<RF,dim> > gradpsi(lfsv.size());
132 Dune::FieldVector<RF,dim> gradu(0.0);
133 Dune::FieldVector<RF,dim> Agradu(0.0);
134
135 // Transformation matrix
136 typename EG::Geometry::JacobianInverseTransposed jac;
137
138 // loop over quadrature points
139 auto intorder = intorderadd + quadrature_factor * order;
140 for (const auto& ip : quadratureRule(geo,intorder))
141 {
142 // update all variables dependent on A if A is not cell-wise constant
143 if (!Impl::permeabilityIsConstantPerCell<T>(param))
144 {
145 A = param.A(cell, ip.position());
146 }
147
148 // evaluate basis functions
149 auto& phi = cache[order].evaluateFunction(ip.position(),lfsu.finiteElement().localBasis());
150 auto& psi = cache[order].evaluateFunction(ip.position(),lfsv.finiteElement().localBasis());
151
152 // evaluate u
153 RF u=0.0;
154 for (size_type i=0; i<lfsu.size(); i++)
155 u += x(lfsu,i)*phi[i];
156
157 // evaluate gradient of basis functions
158 auto& js = cache[order].evaluateJacobian(ip.position(),lfsu.finiteElement().localBasis());
159 auto& js_v = cache[order].evaluateJacobian(ip.position(),lfsv.finiteElement().localBasis());
160
161 // transform gradients of shape functions to real element
162 jac = geo.jacobianInverseTransposed(ip.position());
163 for (size_type i=0; i<lfsu.size(); i++)
164 jac.mv(js[i][0],gradphi[i]);
165
166 for (size_type i=0; i<lfsv.size(); i++)
167 jac.mv(js_v[i][0],gradpsi[i]);
168
169 // compute gradient of u
170 gradu = 0.0;
171 for (size_type i=0; i<lfsu.size(); i++)
172 gradu.axpy(x(lfsu,i),gradphi[i]);
173
174 // compute A * gradient of u
175 A.mv(gradu,Agradu);
176
177 // evaluate velocity field
178 auto b = param.b(cell,ip.position());
179
180 // evaluate reaction term
181 auto c = param.c(cell,ip.position());
182
183 // integrate (A grad u - bu)*grad phi_i + a*u*phi_i
184 RF factor = ip.weight() * geo.integrationElement(ip.position());
185 for (size_type i=0; i<lfsv.size(); i++)
186 r.accumulate(lfsv,i,( Agradu*gradpsi[i] - u*(b*gradpsi[i]) + c*u*psi[i] )*factor);
187 }
188 }
189
190 // apply jacobian of volume term
191 template<typename EG, typename LFSU, typename X, typename LFSV, typename Y>
192 void jacobian_apply_volume (const EG& eg, const LFSU& lfsu, const X& x, const LFSV& lfsv, Y& y) const
193 {
194 alpha_volume(eg,lfsu,x,lfsv,y);
195 }
196
197 // jacobian of volume term
198 template<typename EG, typename LFSU, typename X, typename LFSV, typename M>
199 void jacobian_volume (const EG& eg, const LFSU& lfsu, const X& x, const LFSV& lfsv,
200 M& mat) const
201 {
202 // define types
203 using RF = typename LFSU::Traits::FiniteElementType::
204 Traits::LocalBasisType::Traits::RangeFieldType;
205 using size_type = typename LFSU::Traits::SizeType;
206
207 // dimensions
208 const int dim = EG::Entity::dimension;
209 const int order = std::max(lfsu.finiteElement().localBasis().order(),
210 lfsv.finiteElement().localBasis().order());
211
212 // Get cell
213 const auto& cell = eg.entity();
214
215 // Get geometry
216 auto geo = eg.geometry();
217
218 // evaluate diffusion tensor at cell center, assume it is constant over elements
219 auto ref_el = referenceElement(geo);
220 auto localcenter = ref_el.position(0,0);
221 auto A = param.A(cell,localcenter);
222
223 // Initialize vectors outside for loop
224 std::vector<Dune::FieldVector<RF,dim> > gradphi(lfsu.size());
225 std::vector<Dune::FieldVector<RF,dim> > Agradphi(lfsu.size());
226
227 // Transformation matrix
228 typename EG::Geometry::JacobianInverseTransposed jac;
229
230 // loop over quadrature points
231 auto intorder = intorderadd + quadrature_factor * order;
232 for (const auto& ip : quadratureRule(geo,intorder))
233 {
234 // update all variables dependent on A if A is not cell-wise constant
235 if (!Impl::permeabilityIsConstantPerCell<T>(param))
236 {
237 A = param.A(cell, ip.position());
238 }
239
240 // evaluate basis functions
241 auto& phi = cache[order].evaluateFunction(ip.position(),lfsu.finiteElement().localBasis());
242
243 // evaluate gradient of basis functions
244 auto& js = cache[order].evaluateJacobian(ip.position(),lfsu.finiteElement().localBasis());
245
246 // transform gradients of shape functions to real element
247 jac = geo.jacobianInverseTransposed(ip.position());
248 for (size_type i=0; i<lfsu.size(); i++)
249 {
250 jac.mv(js[i][0],gradphi[i]);
251 A.mv(gradphi[i],Agradphi[i]);
252 }
253
254 // evaluate velocity field
255 auto b = param.b(cell,ip.position());
256
257 // evaluate reaction term
258 auto c = param.c(cell,ip.position());
259
260 // integrate (A grad u - bu)*grad phi_i + a*u*phi_i
261 auto factor = ip.weight() * geo.integrationElement(ip.position());
262 for (size_type j=0; j<lfsu.size(); j++)
263 for (size_type i=0; i<lfsu.size(); i++)
264 mat.accumulate(lfsu,i,lfsu,j,( Agradphi[j]*gradphi[i] - phi[j]*(b*gradphi[i]) + c*phi[j]*phi[i] )*factor);
265 }
266 }
267
268 // skeleton integral depending on test and ansatz functions
269 // each face is only visited ONCE!
270 template<typename IG, typename LFSU, typename X, typename LFSV, typename R>
271 void alpha_skeleton (const IG& ig,
272 const LFSU& lfsu_s, const X& x_s, const LFSV& lfsv_s,
273 const LFSU& lfsu_n, const X& x_n, const LFSV& lfsv_n,
274 R& r_s, R& r_n) const
275 {
276 // define types
277 using RF = typename LFSV::Traits::FiniteElementType::
278 Traits::LocalBasisType::Traits::RangeFieldType;
279 using size_type = typename LFSV::Traits::SizeType;
280
281 // dimensions
282 const int dim = IG::Entity::dimension;
283 const int order = std::max(
284 std::max(lfsu_s.finiteElement().localBasis().order(),
285 lfsu_n.finiteElement().localBasis().order()),
286 std::max(lfsv_s.finiteElement().localBasis().order(),
287 lfsv_n.finiteElement().localBasis().order())
288 );
289
290 // References to inside and outside cells
291 const auto& cell_inside = ig.inside();
292 const auto& cell_outside = ig.outside();
293
294 // Get geometries
295 auto geo = ig.geometry();
296 auto geo_inside = cell_inside.geometry();
297 auto geo_outside = cell_outside.geometry();
298
299 // Get geometry of intersection in local coordinates of cell_inside and cell_outside
300 auto geo_in_inside = ig.geometryInInside();
301 auto geo_in_outside = ig.geometryInOutside();
302
303 // evaluate permeability tensors
304 auto ref_el_inside = referenceElement(geo_inside);
305 auto ref_el_outside = referenceElement(geo_outside);
306 auto local_inside = ref_el_inside.position(0,0);
307 auto local_outside = ref_el_outside.position(0,0);
308 auto A_s = param.A(cell_inside,local_inside);
309 auto A_n = param.A(cell_outside,local_outside);
310
311 // face diameter for anisotropic meshes taken from Paul Houston et al.
312 // this formula ensures coercivity of the bilinear form
313 auto h_F = std::min(geo_inside.volume(),geo_outside.volume())/geo.volume();
314
315 // tensor times normal
316 auto n_F = ig.centerUnitOuterNormal();
318 A_s.mv(n_F,An_F_s);
320 A_n.mv(n_F,An_F_n);
321
322 // compute weights
323 RF omega_s;
324 RF omega_n;
325 RF harmonic_average(0.0);
326 if (weights==ConvectionDiffusionDGWeights::weightsOn)
327 {
328 RF delta_s = (An_F_s*n_F);
329 RF delta_n = (An_F_n*n_F);
330 omega_s = delta_n/(delta_s+delta_n+1e-20);
331 omega_n = delta_s/(delta_s+delta_n+1e-20);
332 harmonic_average = 2.0*delta_s*delta_n/(delta_s+delta_n+1e-20);
333 }
334 else
335 {
336 omega_s = omega_n = 0.5;
337 harmonic_average = 1.0;
338 }
339
340 // get polynomial degree
341 auto order_s = lfsu_s.finiteElement().localBasis().order();
342 auto order_n = lfsu_n.finiteElement().localBasis().order();
343 auto degree = std::max( order_s, order_n );
344
345 // penalty factor
346 auto penalty_factor = (alpha/h_F) * harmonic_average * degree*(degree+dim-1);
347
348 // Initialize vectors outside for loop
349 std::vector<Dune::FieldVector<RF,dim> > tgradphi_s(lfsu_s.size());
350 std::vector<Dune::FieldVector<RF,dim> > tgradpsi_s(lfsv_s.size());
351 std::vector<Dune::FieldVector<RF,dim> > tgradphi_n(lfsu_n.size());
352 std::vector<Dune::FieldVector<RF,dim> > tgradpsi_n(lfsv_n.size());
353 Dune::FieldVector<RF,dim> gradu_s(0.0);
354 Dune::FieldVector<RF,dim> gradu_n(0.0);
355
356 // Transformation matrix
357 typename IG::Entity::Geometry::JacobianInverseTransposed jac;
358
359 // loop over quadrature points
360 auto intorder = intorderadd+quadrature_factor*order;
361 for (const auto& ip : quadratureRule(geo,intorder))
362 {
363 // exact normal
364 auto n_F_local = ig.unitOuterNormal(ip.position());
365
366 // update all variables dependent on A if A is not cell-wise constant
367 if (!Impl::permeabilityIsConstantPerCell<T>(param))
368 {
369 A_s = param.A(cell_inside,geo_in_inside.global(ip.position()));
370 A_n = param.A(cell_outside,geo_in_outside.global(ip.position()));
371 A_s.mv(n_F_local,An_F_s);
372 A_n.mv(n_F_local,An_F_n);
373 if (weights==ConvectionDiffusionDGWeights::weightsOn)
374 {
375 RF delta_s = (An_F_s*n_F_local);
376 RF delta_n = (An_F_n*n_F_local);
377 omega_s = delta_n/(delta_s+delta_n+1e-20);
378 omega_n = delta_s/(delta_s+delta_n+1e-20);
379 harmonic_average = 2.0*delta_s*delta_n/(delta_s+delta_n+1e-20);
380 penalty_factor = (alpha/h_F) * harmonic_average * degree*(degree+dim-1);
381 }
382 }
383
384
385 // position of quadrature point in local coordinates of elements
386 auto iplocal_s = geo_in_inside.global(ip.position());
387 auto iplocal_n = geo_in_outside.global(ip.position());
388
389 // evaluate basis functions
390 auto& phi_s = cache[order_s].evaluateFunction(iplocal_s,lfsu_s.finiteElement().localBasis());
391 auto& phi_n = cache[order_n].evaluateFunction(iplocal_n,lfsu_n.finiteElement().localBasis());
392 auto& psi_s = cache[order_s].evaluateFunction(iplocal_s,lfsv_s.finiteElement().localBasis());
393 auto& psi_n = cache[order_n].evaluateFunction(iplocal_n,lfsv_n.finiteElement().localBasis());
394
395 // evaluate u
396 RF u_s=0.0;
397 for (size_type i=0; i<lfsu_s.size(); i++)
398 u_s += x_s(lfsu_s,i)*phi_s[i];
399 RF u_n=0.0;
400 for (size_type i=0; i<lfsu_n.size(); i++)
401 u_n += x_n(lfsu_n,i)*phi_n[i];
402
403 // evaluate gradient of basis functions
404 auto& gradphi_s = cache[order_s].evaluateJacobian(iplocal_s,lfsu_s.finiteElement().localBasis());
405 auto& gradphi_n = cache[order_n].evaluateJacobian(iplocal_n,lfsu_n.finiteElement().localBasis());
406 auto& gradpsi_s = cache[order_s].evaluateJacobian(iplocal_s,lfsv_s.finiteElement().localBasis());
407 auto& gradpsi_n = cache[order_n].evaluateJacobian(iplocal_n,lfsv_n.finiteElement().localBasis());
408
409 // transform gradients of shape functions to real element
410 jac = geo_inside.jacobianInverseTransposed(iplocal_s);
411 for (size_type i=0; i<lfsu_s.size(); i++) jac.mv(gradphi_s[i][0],tgradphi_s[i]);
412 for (size_type i=0; i<lfsv_s.size(); i++) jac.mv(gradpsi_s[i][0],tgradpsi_s[i]);
413 jac = geo_outside.jacobianInverseTransposed(iplocal_n);
414 for (size_type i=0; i<lfsu_n.size(); i++) jac.mv(gradphi_n[i][0],tgradphi_n[i]);
415 for (size_type i=0; i<lfsv_n.size(); i++) jac.mv(gradpsi_n[i][0],tgradpsi_n[i]);
416
417 // compute gradient of u
418 gradu_s = 0.0;
419 for (size_type i=0; i<lfsu_s.size(); i++)
420 gradu_s.axpy(x_s(lfsu_s,i),tgradphi_s[i]);
421 gradu_n = 0.0;
422 for (size_type i=0; i<lfsu_n.size(); i++)
423 gradu_n.axpy(x_n(lfsu_n,i),tgradphi_n[i]);
424
425 // evaluate velocity field and upwinding, assume H(div) velocity field => may choose any side
426 auto b = param.b(cell_inside,iplocal_s);
427 auto normalflux = b*n_F_local;
428 RF omegaup_s, omegaup_n;
429 if (normalflux>=0.0)
430 {
431 omegaup_s = 1.0;
432 omegaup_n = 0.0;
433 }
434 else
435 {
436 omegaup_s = 0.0;
437 omegaup_n = 1.0;
438 }
439
440 // integration factor
441 auto factor = ip.weight()*geo.integrationElement(ip.position());
442
443 // convection term
444 auto term1 = (omegaup_s*u_s + omegaup_n*u_n) * normalflux *factor;
445 for (size_type i=0; i<lfsv_s.size(); i++)
446 r_s.accumulate(lfsv_s,i,term1 * psi_s[i]);
447 for (size_type i=0; i<lfsv_n.size(); i++)
448 r_n.accumulate(lfsv_n,i,-term1 * psi_n[i]);
449
450 // diffusion term
451 auto term2 = -(omega_s*(An_F_s*gradu_s) + omega_n*(An_F_n*gradu_n)) * factor;
452 for (size_type i=0; i<lfsv_s.size(); i++)
453 r_s.accumulate(lfsv_s,i,term2 * psi_s[i]);
454 for (size_type i=0; i<lfsv_n.size(); i++)
455 r_n.accumulate(lfsv_n,i,-term2 * psi_n[i]);
456
457 // (non-)symmetric IP term
458 auto term3 = (u_s-u_n) * factor;
459 for (size_type i=0; i<lfsv_s.size(); i++)
460 r_s.accumulate(lfsv_s,i,term3 * theta * omega_s * (An_F_s*tgradpsi_s[i]));
461 for (size_type i=0; i<lfsv_n.size(); i++)
462 r_n.accumulate(lfsv_n,i,term3 * theta * omega_n * (An_F_n*tgradpsi_n[i]));
463
464 // standard IP term integral
465 auto term4 = penalty_factor * (u_s-u_n) * factor;
466 for (size_type i=0; i<lfsv_s.size(); i++)
467 r_s.accumulate(lfsv_s,i,term4 * psi_s[i]);
468 for (size_type i=0; i<lfsv_n.size(); i++)
469 r_n.accumulate(lfsv_n,i,-term4 * psi_n[i]);
470 }
471 }
472
473 // apply jacobian of skeleton term
474 template<typename IG, typename LFSU, typename X, typename LFSV, typename Y>
475 void jacobian_apply_skeleton (const IG& ig,
476 const LFSU& lfsu_s, const X& x_s, const LFSV& lfsv_s,
477 const LFSU& lfsu_n, const X& x_n, const LFSV& lfsv_n,
478 Y& y_s, Y& y_n) const
479 {
480 alpha_skeleton(ig,lfsu_s,x_s,lfsv_s,lfsu_n,x_n,lfsv_n,y_s,y_n);
481 }
482
483 template<typename IG, typename LFSU, typename X, typename LFSV, typename M>
484 void jacobian_skeleton (const IG& ig,
485 const LFSU& lfsu_s, const X& x_s, const LFSV& lfsv_s,
486 const LFSU& lfsu_n, const X& x_n, const LFSV& lfsv_n,
487 M& mat_ss, M& mat_sn,
488 M& mat_ns, M& mat_nn) const
489 {
490 // define types
491 using RF = typename LFSV::Traits::FiniteElementType::
492 Traits::LocalBasisType::Traits::RangeFieldType;
493 using size_type = typename LFSV::Traits::SizeType;
494
495 // dimensions
496 const int dim = IG::Entity::dimension;
497 const int order = std::max(
498 std::max(lfsu_s.finiteElement().localBasis().order(),
499 lfsu_n.finiteElement().localBasis().order()),
500 std::max(lfsv_s.finiteElement().localBasis().order(),
501 lfsv_n.finiteElement().localBasis().order())
502 );
503
504 // References to inside and outside cells
505 const auto& cell_inside = ig.inside();
506 const auto& cell_outside = ig.outside();
507
508 // Get geometries
509 auto geo = ig.geometry();
510 auto geo_inside = cell_inside.geometry();
511 auto geo_outside = cell_outside.geometry();
512
513 // Get geometry of intersection in local coordinates of cell_inside and cell_outside
514 auto geo_in_inside = ig.geometryInInside();
515 auto geo_in_outside = ig.geometryInOutside();
516
517 // evaluate permeability tensors
518 auto ref_el_inside = referenceElement(geo_inside);
519 auto ref_el_outside = referenceElement(geo_outside);
520 auto local_inside = ref_el_inside.position(0,0);
521 auto local_outside = ref_el_outside.position(0,0);
522 auto A_s = param.A(cell_inside,local_inside);
523 auto A_n = param.A(cell_outside,local_outside);
524
525 // face diameter for anisotropic meshes taken from Paul Houston et al.
526 // this formula ensures coercivity of the bilinear form
527 auto h_F = std::min(geo_inside.volume(),geo_outside.volume())/geo.volume();
528
529 // tensor times normal
530 auto n_F = ig.centerUnitOuterNormal();
532 A_s.mv(n_F,An_F_s);
534 A_n.mv(n_F,An_F_n);
535
536 // compute weights
537 RF omega_s;
538 RF omega_n;
539 RF harmonic_average(0.0);
540 if (weights==ConvectionDiffusionDGWeights::weightsOn)
541 {
542 RF delta_s = (An_F_s*n_F);
543 RF delta_n = (An_F_n*n_F);
544 omega_s = delta_n/(delta_s+delta_n+1e-20);
545 omega_n = delta_s/(delta_s+delta_n+1e-20);
546 harmonic_average = 2.0*delta_s*delta_n/(delta_s+delta_n+1e-20);
547 }
548 else
549 {
550 omega_s = omega_n = 0.5;
551 harmonic_average = 1.0;
552 }
553
554 // get polynomial degree
555 auto order_s = lfsu_s.finiteElement().localBasis().order();
556 auto order_n = lfsu_n.finiteElement().localBasis().order();
557 auto degree = std::max( order_s, order_n );
558
559 // penalty factor
560 auto penalty_factor = (alpha/h_F) * harmonic_average * degree*(degree+dim-1);
561
562 // Initialize vectors outside for loop
563 std::vector<Dune::FieldVector<RF,dim> > tgradphi_s(lfsu_s.size());
564 std::vector<Dune::FieldVector<RF,dim> > tgradphi_n(lfsu_n.size());
565
566 // Transformation matrix
567 typename IG::Entity::Geometry::JacobianInverseTransposed jac;
568
569 // loop over quadrature points
570 const int intorder = intorderadd+quadrature_factor*order;
571 for (const auto& ip : quadratureRule(geo,intorder))
572 {
573 // exact normal
574 auto n_F_local = ig.unitOuterNormal(ip.position());
575
576 // update all variables dependent on A if A is not cell-wise constant
577 if (!Impl::permeabilityIsConstantPerCell<T>(param))
578 {
579 A_s = param.A(cell_inside,geo_in_inside.global(ip.position()));
580 A_n = param.A(cell_outside,geo_in_outside.global(ip.position()));
581 A_s.mv(n_F_local,An_F_s);
582 A_n.mv(n_F_local,An_F_n);
583 if (weights==ConvectionDiffusionDGWeights::weightsOn)
584 {
585 RF delta_s = (An_F_s*n_F_local);
586 RF delta_n = (An_F_n*n_F_local);
587 omega_s = delta_n/(delta_s+delta_n+1e-20);
588 omega_n = delta_s/(delta_s+delta_n+1e-20);
589 harmonic_average = 2.0*delta_s*delta_n/(delta_s+delta_n+1e-20);
590 penalty_factor = (alpha/h_F) * harmonic_average * degree*(degree+dim-1);
591 }
592 }
593
594 // position of quadrature point in local coordinates of elements
595 auto iplocal_s = geo_in_inside.global(ip.position());
596 auto iplocal_n = geo_in_outside.global(ip.position());
597
598 // evaluate basis functions
599 auto& phi_s = cache[order_s].evaluateFunction(iplocal_s,lfsu_s.finiteElement().localBasis());
600 auto& phi_n = cache[order_n].evaluateFunction(iplocal_n,lfsu_n.finiteElement().localBasis());
601
602 // evaluate gradient of basis functions
603 auto& gradphi_s = cache[order_s].evaluateJacobian(iplocal_s,lfsu_s.finiteElement().localBasis());
604 auto& gradphi_n = cache[order_n].evaluateJacobian(iplocal_n,lfsu_n.finiteElement().localBasis());
605
606 // transform gradients of shape functions to real element
607 jac = geo_inside.jacobianInverseTransposed(iplocal_s);
608 for (size_type i=0; i<lfsu_s.size(); i++) jac.mv(gradphi_s[i][0],tgradphi_s[i]);
609 jac = geo_outside.jacobianInverseTransposed(iplocal_n);
610 for (size_type i=0; i<lfsu_n.size(); i++) jac.mv(gradphi_n[i][0],tgradphi_n[i]);
611
612 // evaluate velocity field and upwinding, assume H(div) velocity field => may choose any side
613 auto b = param.b(cell_inside,iplocal_s);
614 auto normalflux = b*n_F_local;
615 RF omegaup_s, omegaup_n;
616 if (normalflux>=0.0)
617 {
618 omegaup_s = 1.0;
619 omegaup_n = 0.0;
620 }
621 else
622 {
623 omegaup_s = 0.0;
624 omegaup_n = 1.0;
625 }
626
627 // integration factor
628 auto factor = ip.weight() * geo.integrationElement(ip.position());
629 auto ipfactor = penalty_factor * factor;
630
631 // do all terms in the order: I convection, II diffusion, III consistency, IV ip
632 for (size_type j=0; j<lfsu_s.size(); j++) {
633 auto temp1 = -(An_F_s*tgradphi_s[j])*omega_s*factor;
634 for (size_type i=0; i<lfsu_s.size(); i++) {
635 mat_ss.accumulate(lfsu_s,i,lfsu_s,j,omegaup_s * phi_s[j] * normalflux *factor * phi_s[i]);
636 mat_ss.accumulate(lfsu_s,i,lfsu_s,j,temp1 * phi_s[i]);
637 mat_ss.accumulate(lfsu_s,i,lfsu_s,j,phi_s[j] * factor * theta * omega_s * (An_F_s*tgradphi_s[i]));
638 mat_ss.accumulate(lfsu_s,i,lfsu_s,j,phi_s[j] * ipfactor * phi_s[i]);
639 }
640 }
641 for (size_type j=0; j<lfsu_n.size(); j++) {
642 auto temp1 = -(An_F_n*tgradphi_n[j])*omega_n*factor;
643 for (size_type i=0; i<lfsu_s.size(); i++) {
644 mat_sn.accumulate(lfsu_s,i,lfsu_n,j,omegaup_n * phi_n[j] * normalflux *factor * phi_s[i]);
645 mat_sn.accumulate(lfsu_s,i,lfsu_n,j,temp1 * phi_s[i]);
646 mat_sn.accumulate(lfsu_s,i,lfsu_n,j,-phi_n[j] * factor * theta * omega_s * (An_F_s*tgradphi_s[i]));
647 mat_sn.accumulate(lfsu_s,i,lfsu_n,j,-phi_n[j] * ipfactor * phi_s[i]);
648 }
649 }
650 for (size_type j=0; j<lfsu_s.size(); j++) {
651 auto temp1 = -(An_F_s*tgradphi_s[j])*omega_s*factor;
652 for (size_type i=0; i<lfsu_n.size(); i++) {
653 mat_ns.accumulate(lfsu_n,i,lfsu_s,j,-omegaup_s * phi_s[j] * normalflux *factor * phi_n[i]);
654 mat_ns.accumulate(lfsu_n,i,lfsu_s,j,-temp1 * phi_n[i]);
655 mat_ns.accumulate(lfsu_n,i,lfsu_s,j,phi_s[j] * factor * theta * omega_n * (An_F_n*tgradphi_n[i]));
656 mat_ns.accumulate(lfsu_n,i,lfsu_s,j,-phi_s[j] * ipfactor * phi_n[i]);
657 }
658 }
659 for (size_type j=0; j<lfsu_n.size(); j++) {
660 auto temp1 = -(An_F_n*tgradphi_n[j])*omega_n*factor;
661 for (size_type i=0; i<lfsu_n.size(); i++) {
662 mat_nn.accumulate(lfsu_n,i,lfsu_n,j,-omegaup_n * phi_n[j] * normalflux *factor * phi_n[i]);
663 mat_nn.accumulate(lfsu_n,i,lfsu_n,j,-temp1 * phi_n[i]);
664 mat_nn.accumulate(lfsu_n,i,lfsu_n,j,-phi_n[j] * factor * theta * omega_n * (An_F_n*tgradphi_n[i]));
665 mat_nn.accumulate(lfsu_n,i,lfsu_n,j,phi_n[j] * ipfactor * phi_n[i]);
666 }
667 }
668 }
669 }
670
671 // Helper function that can be used to accumulate the alhpa_boundary term
672 // (if jacobian_apply is set to false) or the jacobian_apply_boundary (if
673 // jacobian_apply is set to true)
674 //
675 // A different way of solving this would be to implement all non u
676 // dependent parts in lambda_boundary and have only the u dependent parts
677 // in alpha_volume. Then jacobian_apply_bonudary could just call
678 // alpha_volume.
679 //
680 // It was done in this way for two reasons:
681 // - Easier to verify the implementation and compare to skeleton methods
682 // - More efficient
683 template<typename IG, typename LFSU, typename X, typename LFSV, typename R>
684 void residual_boundary_integral (const IG& ig,
685 const LFSU& lfsu_s, const X& x_s, const LFSV& lfsv_s,
686 R& r_s, bool jacobian_apply=false) const
687 {
688 // define types
689 using RF = typename LFSV::Traits::FiniteElementType::
690 Traits::LocalBasisType::Traits::RangeFieldType;
691 using size_type = typename LFSV::Traits::SizeType;
692
693 // dimensions
694 const int dim = IG::Entity::dimension;
695 const int order = std::max(
696 lfsu_s.finiteElement().localBasis().order(),
697 lfsv_s.finiteElement().localBasis().order()
698 );
699
700 // References to the inside cell
701 const auto& cell_inside = ig.inside();
702
703 // Get geometries
704 auto geo = ig.geometry();
705 auto geo_inside = cell_inside.geometry();
706
707 // Get geometry of intersection in local coordinates of cell_inside
708 auto geo_in_inside = ig.geometryInInside();
709
710 // evaluate permeability tensors
711 auto ref_el_inside = referenceElement(geo_inside);
712 auto local_inside = ref_el_inside.position(0,0);
713 auto A_s = param.A(cell_inside,local_inside);
714
715 // face diameter for anisotropic meshes taken from Paul Houston et al.
716 // this formula ensures coercivity of the bilinear form
717 auto h_F = geo_inside.volume()/geo.volume();
718
719 // compute weights
720 auto n_F = ig.centerUnitOuterNormal();
722 A_s.mv(n_F,An_F_s);
723 RF harmonic_average;
724 if (weights==ConvectionDiffusionDGWeights::weightsOn)
725 harmonic_average = An_F_s*n_F;
726 else
727 harmonic_average = 1.0;
728
729 // get polynomial degree
730 auto order_s = lfsu_s.finiteElement().localBasis().order();
731 auto degree = order_s;
732
733 // penalty factor
734 auto penalty_factor = (alpha/h_F) * harmonic_average * degree*(degree+dim-1);
735
736 // Initialize vectors outside for loop
737 std::vector<Dune::FieldVector<RF,dim> > tgradphi_s(lfsu_s.size());
738 std::vector<Dune::FieldVector<RF,dim> > tgradpsi_s(lfsv_s.size());
739 Dune::FieldVector<RF,dim> gradu_s(0.0);
740
741 // Transformation matrix
742 typename IG::Entity::Geometry::JacobianInverseTransposed jac;
743
744 // loop over quadrature points
745 auto intorder = intorderadd+quadrature_factor*order;
746 for (const auto& ip : quadratureRule(geo,intorder))
747 {
748 // local normal
749 auto n_F_local = ig.unitOuterNormal(ip.position());
750
751 // update all variables dependent on A if A is not cell-wise constant
752 if (!Impl::permeabilityIsConstantPerCell<T>(param))
753 {
754 A_s = param.A(cell_inside,geo_in_inside.global(ip.position()));
755 A_s.mv(n_F_local,An_F_s);
756 if (weights==ConvectionDiffusionDGWeights::weightsOn)
757 {
758 harmonic_average = An_F_s*n_F_local;
759 penalty_factor = (alpha/h_F) * harmonic_average * degree*(degree+dim-1);
760 }
761 }
762
763 auto bctype = param.bctype(ig.intersection(),ip.position());
764
765 if (bctype == ConvectionDiffusionBoundaryConditions::None)
766 continue;
767
768 // position of quadrature point in local coordinates of elements
769 auto iplocal_s = geo_in_inside.global(ip.position());
770
771 // evaluate basis functions
772 auto& phi_s = cache[order_s].evaluateFunction(iplocal_s,lfsu_s.finiteElement().localBasis());
773 auto& psi_s = cache[order_s].evaluateFunction(iplocal_s,lfsv_s.finiteElement().localBasis());
774
775 // integration factor
776 RF factor = ip.weight() * geo.integrationElement(ip.position());
777
778 if (bctype == ConvectionDiffusionBoundaryConditions::Neumann)
779 {
780 if (not jacobian_apply){
781 // evaluate flux boundary condition
782 auto j = param.j(ig.intersection(),ip.position());
783
784 // integrate
785 for (size_type i=0; i<lfsv_s.size(); i++)
786 r_s.accumulate(lfsv_s,i,j * psi_s[i] * factor);
787 }
788 continue;
789 }
790
791 // evaluate u
792 RF u_s=0.0;
793 for (size_type i=0; i<lfsu_s.size(); i++)
794 u_s += x_s(lfsu_s,i)*phi_s[i];
795
796 // evaluate velocity field and upwinding, assume H(div) velocity field => choose any side
797 auto b = param.b(cell_inside,iplocal_s);
798 auto normalflux = b*n_F_local;
799
800 if (bctype == ConvectionDiffusionBoundaryConditions::Outflow)
801 {
802 if (normalflux<-1e-30)
804 "Outflow boundary condition on inflow! [b("
805 << geo.global(ip.position()) << ") = "
806 << b << ")");
807
808 // convection term
809 auto term1 = u_s * normalflux *factor;
810 for (size_type i=0; i<lfsv_s.size(); i++)
811 r_s.accumulate(lfsv_s,i,term1 * psi_s[i]);
812
813 if (not jacobian_apply){
814 // evaluate flux boundary condition
815 auto o = param.o(ig.intersection(),ip.position());
816
817 // integrate
818 for (size_type i=0; i<lfsv_s.size(); i++)
819 r_s.accumulate(lfsv_s,i,o * psi_s[i] * factor);
820 }
821 continue;
822 }
823
824 // evaluate gradient of basis functions
825 assert (bctype == ConvectionDiffusionBoundaryConditions::Dirichlet);
826 auto& gradphi_s = cache[order_s].evaluateJacobian(iplocal_s,lfsu_s.finiteElement().localBasis());
827 auto& gradpsi_s = cache[order_s].evaluateJacobian(iplocal_s,lfsv_s.finiteElement().localBasis());
828
829 // transform gradients of shape functions to real element
830 jac = geo_inside.jacobianInverseTransposed(iplocal_s);
831 for (size_type i=0; i<lfsu_s.size(); i++) jac.mv(gradphi_s[i][0],tgradphi_s[i]);
832 for (size_type i=0; i<lfsv_s.size(); i++) jac.mv(gradpsi_s[i][0],tgradpsi_s[i]);
833
834 // compute gradient of u
835 gradu_s = 0.0;
836 for (size_type i=0; i<lfsu_s.size(); i++)
837 gradu_s.axpy(x_s(lfsu_s,i),tgradphi_s[i]);
838
839 // evaluate Dirichlet boundary condition
840 auto g = param.g(cell_inside,iplocal_s);
841
842 if (jacobian_apply){
843 g = 0.0;
844 }
845
846 // upwind
847 RF omegaup_s, omegaup_n;
848 if (normalflux>=0.0)
849 {
850 omegaup_s = 1.0;
851 omegaup_n = 0.0;
852 }
853 else
854 {
855 omegaup_s = 0.0;
856 omegaup_n = 1.0;
857 }
858
859 // convection term
860 auto term1 = (omegaup_s*u_s + omegaup_n*g) * normalflux *factor;
861 for (size_type i=0; i<lfsv_s.size(); i++)
862 r_s.accumulate(lfsv_s,i,term1 * psi_s[i]);
863
864 // diffusion term
865 auto term2 = (An_F_s*gradu_s) * factor;
866 for (size_type i=0; i<lfsv_s.size(); i++)
867 r_s.accumulate(lfsv_s,i,-term2 * psi_s[i]);
868
869 // (non-)symmetric IP term
870 auto term3 = (u_s-g) * factor;
871 for (size_type i=0; i<lfsv_s.size(); i++)
872 r_s.accumulate(lfsv_s,i,term3 * theta * (An_F_s*tgradpsi_s[i]));
873
874 // standard IP term
875 auto term4 = penalty_factor * (u_s-g) * factor;
876 for (size_type i=0; i<lfsv_s.size(); i++)
877 r_s.accumulate(lfsv_s,i,term4 * psi_s[i]);
878 }
879 }
880
881 // boundary integral depending on test and ansatz functions
882 // We put the Dirchlet evaluation also in the alpha term to save some geometry evaluations
883 template<typename IG, typename LFSU, typename X, typename LFSV, typename R>
884 void alpha_boundary (const IG& ig,
885 const LFSU& lfsu_s, const X& x_s, const LFSV& lfsv_s,
886 R& r_s) const
887 {
888 residual_boundary_integral(ig, lfsu_s, x_s, lfsv_s, r_s);
889 }
890
891 // apply jacobian of boundary term
892 template<typename IG, typename LFSU, typename X, typename LFSV, typename Y>
893 void jacobian_apply_boundary (const IG& ig,
894 const LFSU& lfsu_s, const X& x_s, const LFSV& lfsv_s,
895 Y& y_s) const
896 {
897 // Call helper function and tell that we want to do jacobian apply
898 residual_boundary_integral(ig, lfsu_s, x_s, lfsv_s, y_s, true);
899 }
900
901 template<typename IG, typename LFSU, typename X, typename LFSV, typename M>
902 void jacobian_boundary (const IG& ig,
903 const LFSU& lfsu_s, const X& x_s, const LFSV& lfsv_s,
904 M& mat_ss) const
905 {
906 // define types
907 using RF = typename LFSV::Traits::FiniteElementType::
908 Traits::LocalBasisType::Traits::RangeFieldType;
909 using size_type = typename LFSV::Traits::SizeType;
910
911 // dimensions
912 const int dim = IG::Entity::dimension;
913 const int order = std::max(
914 lfsu_s.finiteElement().localBasis().order(),
915 lfsv_s.finiteElement().localBasis().order()
916 );
917
918 // References to the inside cell
919 const auto& cell_inside = ig.inside();
920
921 // Get geometries
922 auto geo = ig.geometry();
923 auto geo_inside = cell_inside.geometry();
924
925 // Get geometry of intersection in local coordinates of cell_inside
926 auto geo_in_inside = ig.geometryInInside();
927
928 // evaluate permeability tensors
929 auto ref_el_inside = referenceElement(geo_inside);
930 auto local_inside = ref_el_inside.position(0,0);
931 auto A_s = param.A(cell_inside,local_inside);
932
933 // face diameter for anisotropic meshes taken from Paul Houston et al.
934 // this formula ensures coercivity of the bilinear form
935 auto h_F = geo_inside.volume()/geo.volume();
936
937 // compute weights
938 auto n_F = ig.centerUnitOuterNormal();
940 A_s.mv(n_F,An_F_s);
941 RF harmonic_average;
942 if (weights==ConvectionDiffusionDGWeights::weightsOn)
943 harmonic_average = An_F_s*n_F;
944 else
945 harmonic_average = 1.0;
946
947 // get polynomial degree
948 auto order_s = lfsu_s.finiteElement().localBasis().order();
949 auto degree = order_s;
950
951 // penalty factor
952 auto penalty_factor = (alpha/h_F) * harmonic_average * degree*(degree+dim-1);
953
954 // Initialize vectors outside for loop
955 std::vector<Dune::FieldVector<RF,dim> > tgradphi_s(lfsu_s.size());
956
957 // Transformation matrix
958 typename IG::Entity::Geometry::JacobianInverseTransposed jac;
959
960 // loop over quadrature points
961 auto intorder = intorderadd+quadrature_factor*order;
962 for (const auto& ip : quadratureRule(geo,intorder))
963 {
964 // local normal
965 auto n_F_local = ig.unitOuterNormal(ip.position());
966
967 // update all variables dependent on A if A is not cell-wise constant
968 if (!Impl::permeabilityIsConstantPerCell<T>(param))
969 {
970 A_s = param.A(cell_inside,geo_in_inside.global(ip.position()));
971 A_s.mv(n_F_local,An_F_s);
972 if (weights==ConvectionDiffusionDGWeights::weightsOn)
973 {
974 harmonic_average = An_F_s*n_F_local;
975 penalty_factor = (alpha/h_F) * harmonic_average * degree*(degree+dim-1);
976 }
977 }
978
979 auto bctype = param.bctype(ig.intersection(),ip.position());
980
981 if (bctype == ConvectionDiffusionBoundaryConditions::None ||
982 bctype == ConvectionDiffusionBoundaryConditions::Neumann)
983 continue;
984
985 // position of quadrature point in local coordinates of elements
986 auto iplocal_s = geo_in_inside.global(ip.position());
987
988 // evaluate basis functions
989 auto& phi_s = cache[order_s].evaluateFunction(iplocal_s,lfsu_s.finiteElement().localBasis());
990
991 // integration factor
992 auto factor = ip.weight() * geo.integrationElement(ip.position());
993
994 // evaluate velocity field and upwinding, assume H(div) velocity field => choose any side
995 auto b = param.b(cell_inside,iplocal_s);
996 auto normalflux = b*n_F_local;
997
998 if (bctype == ConvectionDiffusionBoundaryConditions::Outflow)
999 {
1000 if (normalflux<-1e-30)
1002 "Outflow boundary condition on inflow! [b("
1003 << geo.global(ip.position()) << ") = "
1004 << b << ")" << n_F_local << " " << normalflux);
1005
1006 // convection term
1007 for (size_type j=0; j<lfsu_s.size(); j++)
1008 for (size_type i=0; i<lfsu_s.size(); i++)
1009 mat_ss.accumulate(lfsu_s,i,lfsu_s,j,phi_s[j] * normalflux * factor * phi_s[i]);
1010
1011 continue;
1012 }
1013
1014 // evaluate gradient of basis functions
1015 auto& gradphi_s = cache[order_s].evaluateJacobian(iplocal_s,lfsu_s.finiteElement().localBasis());
1016
1017 // transform gradients of shape functions to real element
1018 jac = geo_inside.jacobianInverseTransposed(iplocal_s);
1019 for (size_type i=0; i<lfsu_s.size(); i++) jac.mv(gradphi_s[i][0],tgradphi_s[i]);
1020
1021 // upwind
1022 RF omegaup_s = normalflux>=0.0 ? 1.0 : 0.0;
1023
1024 // convection term
1025 for (size_type j=0; j<lfsu_s.size(); j++)
1026 for (size_type i=0; i<lfsu_s.size(); i++)
1027 mat_ss.accumulate(lfsu_s,i,lfsu_s,j,omegaup_s * phi_s[j] * normalflux * factor * phi_s[i]);
1028
1029 // diffusion term
1030 for (size_type j=0; j<lfsu_s.size(); j++)
1031 for (size_type i=0; i<lfsu_s.size(); i++)
1032 mat_ss.accumulate(lfsu_s,i,lfsu_s,j,-(An_F_s*tgradphi_s[j]) * factor * phi_s[i]);
1033
1034 // (non-)symmetric IP term
1035 for (size_type j=0; j<lfsu_s.size(); j++)
1036 for (size_type i=0; i<lfsu_s.size(); i++)
1037 mat_ss.accumulate(lfsu_s,i,lfsu_s,j,phi_s[j] * factor * theta * (An_F_s*tgradphi_s[i]));
1038
1039 // standard IP term
1040 for (size_type j=0; j<lfsu_s.size(); j++)
1041 for (size_type i=0; i<lfsu_s.size(); i++)
1042 mat_ss.accumulate(lfsu_s,i,lfsu_s,j,penalty_factor * phi_s[j] * phi_s[i] * factor);
1043 }
1044 }
1045
1046 // volume integral depending only on test functions
1047 template<typename EG, typename LFSV, typename R>
1048 void lambda_volume (const EG& eg, const LFSV& lfsv, R& r) const
1049 {
1050 // define types
1051 using size_type = typename LFSV::Traits::SizeType;
1052
1053 // Get cell
1054 const auto& cell = eg.entity();
1055
1056 // get geometries
1057 auto geo = eg.geometry();
1058
1059 // loop over quadrature points
1060 auto order = lfsv.finiteElement().localBasis().order();
1061 auto intorder = intorderadd + 2 * order;
1062 for (const auto& ip : quadratureRule(geo,intorder))
1063 {
1064 // evaluate shape functions
1065 auto& phi = cache[order].evaluateFunction(ip.position(),lfsv.finiteElement().localBasis());
1066
1067 // evaluate right hand side parameter function
1068 auto f = param.f(cell,ip.position());
1069
1070 // integrate f
1071 auto factor = ip.weight() * geo.integrationElement(ip.position());
1072 for (size_type i=0; i<lfsv.size(); i++)
1073 r.accumulate(lfsv,i,-f*phi[i]*factor);
1074 }
1075 }
1076
1078 void setTime (Real t)
1079 {
1081 param.setTime(t);
1082 }
1083
1084 private:
1085 T& param; // two phase parameter class
1086 ConvectionDiffusionDGMethod::Type method;
1087 ConvectionDiffusionDGWeights::Type weights;
1088 Real alpha, beta;
1089 int intorderadd;
1090 int quadrature_factor;
1091 Real theta;
1092
1093 using LocalBasisType = typename FiniteElementMap::Traits::FiniteElementType::Traits::LocalBasisType;
1095
1096 // In theory it is possible that one and the same local operator is
1097 // called first with a finite element of one type and later with a
1098 // finite element of another type. Since finite elements of different
1099 // type will usually produce different results for the same local
1100 // coordinate they cannot share a cache. Here we use a vector of caches
1101 // to allow for different orders of the shape functions, which should be
1102 // enough to support p-adaptivity. (Another likely candidate would be
1103 // differing geometry types, i.e. hybrid meshes.)
1104
1105 std::vector<Cache> cache;
1106
1107 template<class GEO>
1108 void element_size (const GEO& geo, typename GEO::ctype& hmin, typename GEO::ctype hmax) const
1109 {
1110 using DF = typename GEO::ctype;
1111 hmin = 1.0E100;
1112 hmax = -1.0E00;
1113 const int dim = GEO::coorddimension;
1114 if (dim==1)
1115 {
1116 Dune::FieldVector<DF,dim> x = geo.corner(0);
1117 x -= geo.corner(1);
1118 hmin = hmax = x.two_norm();
1119 return;
1120 }
1121 else
1122 {
1123 Dune::GeometryType gt = geo.type();
1124 for (int i=0; i<Dune::ReferenceElements<DF,dim>::general(gt).size(dim-1); i++)
1125 {
1126 Dune::FieldVector<DF,dim> x = geo.corner(Dune::ReferenceElements<DF,dim>::general(gt).subEntity(i,dim-1,0,dim));
1127 x -= geo.corner(Dune::ReferenceElements<DF,dim>::general(gt).subEntity(i,dim-1,1,dim));
1128 hmin = std::min(hmin,x.two_norm());
1129 hmax = std::max(hmax,x.two_norm());
1130 }
1131 return;
1132 }
1133 }
1134 };
1135 }
1136}
1137#endif // DUNE_PDELAB_LOCALOPERATOR_CONVECTIONDIFFUSIONDG_HH
FieldTraits< value_type >::real_type two_norm() const
two norm sqrt(sum over squared values of entries)
Definition: densevector.hh:642
Base class for Dune-Exceptions.
Definition: exceptions.hh:94
vector space out of a tensor product of fields.
Definition: fvector.hh:96
Unique label for each type of entities that can occur in DUNE grids.
Definition: type.hh:280
Definition: convectiondiffusiondg.hh:60
void setTime(Real t)
set time in parameter class
Definition: convectiondiffusiondg.hh:1078
ConvectionDiffusionDG(T &param_, ConvectionDiffusionDGMethod::Type method_=ConvectionDiffusionDGMethod::SIPG, ConvectionDiffusionDGWeights::Type weights_=ConvectionDiffusionDGWeights::weightsOn, Real alpha_=1.0, int intorderadd_=0)
constructor: pass parameter object and define DG-method
Definition: convectiondiffusiondg.hh:85
sparsity pattern generator
Definition: pattern.hh:30
sparsity pattern generator
Definition: pattern.hh:14
Default class for additional methods in instationary local operators.
Definition: idefault.hh:90
void setTime(R t_)
set time for subsequent evaluation
Definition: idefault.hh:104
store values of basis functions and gradients in a cache
Definition: localbasiscache.hh:18
Default flags for all local operators.
Definition: flags.hh:19
A few common exception classes.
Implements a vector constructed from a given type representing a field and a compile-time given size.
#define DUNE_THROW(E, m)
Definition: exceptions.hh:216
bool gt(const T &first, const T &second, typename EpsilonType< T >::Type epsilon)
test if first greater than second
Definition: float_cmp.cc:156
unspecified value type referenceElement(T &&... t)
Returns a reference element for the objects t....
std::size_t degree(const Node &node)
Returns the degree of node as run time information.
Definition: nodeinterface.hh:71
auto min(ADLTag< 0 >, const V &v1, const V &v2)
implements binary Simd::min()
Definition: defaults.hh:87
auto max(ADLTag< 0 >, const V &v1, const V &v2)
implements binary Simd::max()
Definition: defaults.hh:79
Dune namespace.
Definition: alignedallocator.hh:14
Class providing access to the singletons of the reference elements.
Definition: referenceelements.hh:168
Creative Commons License   |  Legal Statements / Impressum  |  Hosted by TU Dresden  |  generated with Hugo v0.111.3 (Jul 15, 22:36, 2024)