| 1 |
// ------------------------------------------------------------------------ |
| 2 |
// |
| 3 |
// SPDX-License-Identifier: LGPL-2.1-or-later |
| 4 |
// Copyright (C) 1999 - 2025 by the deal.II authors |
| 5 |
// |
| 6 |
// This file is part of the deal.II library. |
| 7 |
// |
| 8 |
// Part of the source code is dual licensed under Apache-2.0 WITH |
| 9 |
// LLVM-exception OR LGPL-2.1-or-later. Detailed license information |
| 10 |
// governing the source code and code contributions can be found in |
| 11 |
// LICENSE.md and CONTRIBUTING.md at the top level directory of deal.II. |
| 12 |
// |
| 13 |
// ------------------------------------------------------------------------ |
| 14 |
|
| 15 |
#ifndef dealii_sparse_vanka_h |
| 16 |
#define dealii_sparse_vanka_h |
| 17 |
|
| 18 |
#include <deal.II/base/config.h> |
| 19 |
|
| 20 |
#include <deal.II/base/multithread_info.h> |
| 21 |
#include <deal.II/base/observer_pointer.h> |
| 22 |
#include <deal.II/base/types.h> |
| 23 |
|
| 24 |
#include <vector> |
| 25 |
|
| 26 |
DEAL_II_NAMESPACE_OPEN |
| 27 |
|
| 28 |
// Forward declarations |
| 29 |
#ifndef DOXYGEN |
| 30 |
template <typename number> |
| 31 |
class FullMatrix; |
| 32 |
template <typename number> |
| 33 |
class SparseMatrix; |
| 34 |
template <typename number> |
| 35 |
class Vector; |
| 36 |
|
| 37 |
template <typename number> |
| 38 |
class SparseVanka; |
| 39 |
template <typename number> |
| 40 |
class SparseBlockVanka; |
| 41 |
#endif |
| 42 |
|
| 43 |
/** |
| 44 |
* @addtogroup Preconditioners |
| 45 |
* @{ |
| 46 |
*/ |
| 47 |
|
| 48 |
/** |
| 49 |
* Point-wise Vanka preconditioning. This class does Vanka preconditioning on |
| 50 |
* a point-wise base. Vanka preconditioners are used for saddle point problems |
| 51 |
* like Stokes' problem or problems arising in optimization where Lagrange |
| 52 |
* multipliers occur and the Newton method matrix has a zero block. With these |
| 53 |
* matrices the application of Jacobi or Gauss-Seidel methods is impossible, |
| 54 |
* because some diagonal elements are zero in the rows of the Lagrange |
| 55 |
* multiplier. The approach of Vanka is to solve a small (usually indefinite) |
| 56 |
* system of equations for each Langrange multiplier variable (we will also |
| 57 |
* call the pressure in Stokes' equation a Langrange multiplier since it can |
| 58 |
* be interpreted as such). |
| 59 |
* |
| 60 |
* Objects of this class are constructed by passing a vector of indices of the |
| 61 |
* degrees of freedom of the Lagrange multiplier. In the actual |
| 62 |
* preconditioning method, these rows are traversed in the order in which the |
| 63 |
* appear in the matrix. Since this is a Gauß-Seidel like procedure, remember |
| 64 |
* to have a good ordering in advance (for transport dominated problems, |
| 65 |
* Cuthill-McKee algorithms are a good means for this, if points on the inflow |
| 66 |
* boundary are chosen as starting points for the renumbering). |
| 67 |
* |
| 68 |
* For each selected degree of freedom, a local system of equations is built |
| 69 |
* by the degree of freedom itself and all other values coupling immediately, |
| 70 |
* i.e. the set of degrees of freedom considered for the local system of |
| 71 |
* equations for degree of freedom @p i is @p i itself and all @p j such that |
| 72 |
* the element <tt>(i,j)</tt> is a nonzero entry in the sparse matrix under |
| 73 |
* consideration. The elements <tt>(j,i)</tt> are not considered. We now pick |
| 74 |
* all matrix entries from rows and columns out of the set of degrees of |
| 75 |
* freedom just described out of the global matrix and put it into a local |
| 76 |
* matrix, which is subsequently inverted. This system may be of different |
| 77 |
* size for each degree of freedom, depending for example on the local |
| 78 |
* neighborhood of the respective node on a computational grid. |
| 79 |
* |
| 80 |
* The right hand side is built up in the same way, i.e. by copying all |
| 81 |
* entries that coupled with the one under present consideration, but it is |
| 82 |
* augmented by all degrees of freedom coupling with the degrees from the set |
| 83 |
* described above (i.e. the DoFs coupling second order to the present one). |
| 84 |
* The reason for this is, that the local problems to be solved shall have |
| 85 |
* Dirichlet boundary conditions on the second order coupling DoFs, so we have |
| 86 |
* to take them into account but eliminate them before actually solving; this |
| 87 |
* elimination is done by the modification of the right hand side, and in the |
| 88 |
* end these degrees of freedom do not occur in the matrix and solution vector |
| 89 |
* any more at all. |
| 90 |
* |
| 91 |
* This local system is solved and the values are updated into the destination |
| 92 |
* vector. |
| 93 |
* |
| 94 |
* Remark: the Vanka method is a non-symmetric preconditioning method. |
| 95 |
* |
| 96 |
* |
| 97 |
* <h3>Example of Use</h3> This little example is taken from a program doing |
| 98 |
* parameter optimization. The Lagrange multiplier is the third component of |
| 99 |
* the finite element used. The system is solved by the GMRES method. |
| 100 |
* @code |
| 101 |
* // tag the Lagrange multiplier variable |
| 102 |
* vector<bool> signature(3); |
| 103 |
* signature[0] = signature[1] = false; |
| 104 |
* signature[2] = true; |
| 105 |
* |
| 106 |
* // tag all dofs belonging to the Lagrange multiplier |
| 107 |
* vector<bool> selected_dofs (dof.n_dofs(), false); |
| 108 |
* DoFTools::extract_dofs(dof, signature, p_select); |
| 109 |
* // create the Vanka object |
| 110 |
* SparseVanka<double> vanka (global_matrix, selected_dofs); |
| 111 |
* |
| 112 |
* // create the solver |
| 113 |
* SolverGMRES<> gmres(control,memory,504); |
| 114 |
* |
| 115 |
* // solve |
| 116 |
* gmres.solve (global_matrix, solution, right_hand_side, |
| 117 |
* vanka); |
| 118 |
* @endcode |
| 119 |
* |
| 120 |
* |
| 121 |
* <h4>Implementor's remark</h4> At present, the local matrices are built up |
| 122 |
* such that the degree of freedom associated with the local Lagrange |
| 123 |
* multiplier is the first one. Thus, usually the upper left entry in the |
| 124 |
* local matrix is zero. It is not clear to me (W.B.) whether this might pose |
| 125 |
* some problems in the inversion of the local matrices. Maybe someone would |
| 126 |
* like to check this. |
| 127 |
* |
| 128 |
* @note Instantiations for this template are provided for <tt>@<float@> and |
| 129 |
* @<double@></tt>; others can be generated in application programs (see the |
| 130 |
* section on |
| 131 |
* @ref Instantiations |
| 132 |
* in the manual). |
| 133 |
*/ |
| 134 |
template <typename number> |
| 135 |
class SparseVanka |
| 136 |
{ |
| 137 |
public: |
| 138 |
/** |
| 139 |
* Declare type for container size. |
| 140 |
*/ |
| 141 |
using size_type = types::global_dof_index; |
| 142 |
|
| 143 |
/** |
| 144 |
* Constructor. Does nothing. |
| 145 |
* |
| 146 |
* Call the initialize() function before using this object as preconditioner |
| 147 |
* (vmult()). |
| 148 |
*/ |
| 149 |
SparseVanka(); |
| 150 |
|
| 151 |
/** |
| 152 |
* Constructor. Gets the matrix for preconditioning and a bit vector with |
| 153 |
* entries @p true for all rows to be updated. A reference to this vector |
| 154 |
* will be stored, so it must persist longer than the Vanka object. The same |
| 155 |
* is true for the matrix. |
| 156 |
* |
| 157 |
* The matrix @p M which is passed here may or may not be the same matrix |
| 158 |
* for which this object shall act as preconditioner. In particular, it is |
| 159 |
* conceivable that the preconditioner is build up for one matrix once, but |
| 160 |
* is used for subsequent steps in a nonlinear process as well, where the |
| 161 |
* matrix changes in each step slightly. |
| 162 |
*/ |
| 163 |
SparseVanka(const SparseMatrix<number> &M, const std::vector<bool> &selected); |
| 164 |
|
| 165 |
/** |
| 166 |
* Destructor. Delete all allocated matrices. |
| 167 |
*/ |
| 168 |
~SparseVanka(); |
| 169 |
|
| 170 |
/** |
| 171 |
* Parameters for SparseVanka. |
| 172 |
*/ |
| 173 |
class AdditionalData |
| 174 |
{ |
| 175 |
public: |
| 176 |
/** |
| 177 |
* Constructor. For the parameters' description, see below. |
| 178 |
*/ |
| 179 |
explicit AdditionalData(const std::vector<bool> &selected); |
| 180 |
|
| 181 |
/** |
| 182 |
* Indices of those degrees of freedom that we shall work on. |
| 183 |
*/ |
| 184 |
const std::vector<bool> &selected; |
| 185 |
}; |
| 186 |
|
| 187 |
|
| 188 |
/** |
| 189 |
* If the default constructor is used then this function needs to be called |
| 190 |
* before an object of this class is used as preconditioner. |
| 191 |
* |
| 192 |
* For more detail about possible parameters, see the class documentation |
| 193 |
* and the documentation of the SparseVanka::AdditionalData class. |
| 194 |
* |
| 195 |
* After this function is called the preconditioner is ready to be used |
| 196 |
* (using the <code>vmult</code> function of derived classes). |
| 197 |
*/ |
| 198 |
void |
| 199 |
initialize(const SparseMatrix<number> &M, |
| 200 |
const AdditionalData &additional_data); |
| 201 |
|
| 202 |
/** |
| 203 |
* Do the preconditioning. This function takes the residual in @p src and |
| 204 |
* returns the resulting update vector in @p dst. |
| 205 |
*/ |
| 206 |
template <typename number2> |
| 207 |
void |
| 208 |
vmult(Vector<number2> &dst, const Vector<number2> &src) const; |
| 209 |
|
| 210 |
/** |
| 211 |
* Apply transpose preconditioner. This function takes the residual in @p |
| 212 |
* src and returns the resulting update vector in @p dst. |
| 213 |
*/ |
| 214 |
template <typename number2> |
| 215 |
void |
| 216 |
Tvmult(Vector<number2> &dst, const Vector<number2> &src) const; |
| 217 |
|
| 218 |
/** |
| 219 |
* Clear all memory. |
| 220 |
*/ |
| 221 |
void |
| 222 |
clear(); |
| 223 |
|
| 224 |
/** |
| 225 |
* Return the dimension of the codomain (or range) space. Note that the |
| 226 |
* matrix is of dimension $m \times n$. |
| 227 |
* |
| 228 |
* @note This function should only be called if the preconditioner has been |
| 229 |
* initialized. |
| 230 |
*/ |
| 231 |
size_type |
| 232 |
m() const; |
| 233 |
|
| 234 |
/** |
| 235 |
* Return the dimension of the domain space. Note that the matrix is of |
| 236 |
* dimension $m \times n$. |
| 237 |
* |
| 238 |
* @note This function should only be called if the preconditioner has been |
| 239 |
* initialized. |
| 240 |
*/ |
| 241 |
size_type |
| 242 |
n() const; |
| 243 |
|
| 244 |
protected: |
| 245 |
/** |
| 246 |
* Apply the inverses corresponding to those degrees of freedom that have a |
| 247 |
* @p true value in @p dof_mask, to the @p src vector and move the result |
| 248 |
* into @p dst. The transpose of the inverse is applied instead if |
| 249 |
* @p transpose equals true. Actually, only values for allowed indices are |
| 250 |
* written to @p dst, so the application of this function only does what is |
| 251 |
* announced in the general documentation if the given mask sets all values |
| 252 |
* to zero. |
| 253 |
* |
| 254 |
* The reason for providing the mask anyway is that in derived classes we |
| 255 |
* may want to apply the preconditioner to parts of the matrix only, in |
| 256 |
* order to parallelize the application. Then, it is important to only write |
| 257 |
* to some slices of @p dst, in order to eliminate the dependencies of |
| 258 |
* threads of each other. |
| 259 |
* |
| 260 |
* If a null pointer is passed instead of a pointer to the @p dof_mask (as |
| 261 |
* is the default value), then it is assumed that we shall work on all |
| 262 |
* degrees of freedom. This is then equivalent to calling the function with |
| 263 |
* a <tt>vector<bool>(n_dofs,true)</tt>. |
| 264 |
* |
| 265 |
* The @p vmult of this class of course calls this function with a null |
| 266 |
* pointer |
| 267 |
*/ |
| 268 |
|
| 269 |
template <typename number2> |
| 270 |
void |
| 271 |
apply_preconditioner(Vector<number2> &dst, |
| 272 |
const Vector<number2> &src, |
| 273 |
const bool transpose = false, |
| 274 |
const std::vector<bool> *const dof_mask = nullptr) const; |
| 275 |
|
| 276 |
/** |
| 277 |
* Determine an estimate for the memory consumption (in bytes) of this |
| 278 |
* object. |
| 279 |
*/ |
| 280 |
std::size_t |
| 281 |
memory_consumption() const; |
| 282 |
|
| 283 |
private: |
| 284 |
/** |
| 285 |
* Pointer to the matrix. |
| 286 |
*/ |
| 287 |
ObserverPointer<const SparseMatrix<number>, SparseVanka<number>> matrix; |
| 288 |
|
| 289 |
/** |
| 290 |
* Indices of those degrees of freedom that we shall work on. |
| 291 |
*/ |
| 292 |
const std::vector<bool> *selected; |
| 293 |
|
| 294 |
/** |
| 295 |
* Array of inverse matrices, one for each degree of freedom. Only those |
| 296 |
* elements will be used that are tagged in @p selected. |
| 297 |
*/ |
| 298 |
mutable std::vector<ObserverPointer<FullMatrix<float>, SparseVanka<number>>> |
| 299 |
inverses; |
| 300 |
|
| 301 |
/** |
| 302 |
* The dimension of the range space. |
| 303 |
*/ |
| 304 |
size_type _m; |
| 305 |
|
| 306 |
/** |
| 307 |
* The dimension of the domain space. |
| 308 |
*/ |
| 309 |
size_type _n; |
| 310 |
|
| 311 |
/** |
| 312 |
* Compute the inverses of all selected diagonal elements. |
| 313 |
*/ |
| 314 |
void |
| 315 |
compute_inverses(); |
| 316 |
|
| 317 |
/** |
| 318 |
* Compute the inverses at positions in the range <tt>[begin,end)</tt>. In |
| 319 |
* non-multithreaded mode, <tt>compute_inverses()</tt> calls this function |
| 320 |
* with the whole range, but in multithreaded mode, several copies of this |
| 321 |
* function are spawned. |
| 322 |
*/ |
| 323 |
void |
| 324 |
compute_inverses(const size_type begin, const size_type end); |
| 325 |
|
| 326 |
/** |
| 327 |
* Compute the inverse of the block located at position @p row. Since the |
| 328 |
* vector is used quite often, it is generated only once in the caller of |
| 329 |
* this function and passed to this function which first clears it. Reusing |
| 330 |
* the vector makes the process significantly faster than in the case where |
| 331 |
* this function re-creates it each time. |
| 332 |
*/ |
| 333 |
void |
| 334 |
compute_inverse(const size_type row, std::vector<size_type> &local_indices); |
| 335 |
|
| 336 |
// Make the derived class a friend. This seems silly, but is actually |
| 337 |
// necessary, since derived classes can only access non-public members |
| 338 |
// through their @p this pointer, but not access these members as member |
| 339 |
// functions of other objects of the type of this base class (i.e. like |
| 340 |
// <tt>x.f()</tt>, where @p x is an object of the base class, and @p f one |
| 341 |
// of it's non-public member functions). |
| 342 |
// |
| 343 |
// Now, in the case of the @p SparseBlockVanka class, we would like to take |
| 344 |
// the address of a function of the base class in order to call it through |
| 345 |
// the multithreading framework, so the derived class has to be a friend. |
| 346 |
template <typename T> |
| 347 |
friend class SparseBlockVanka; |
| 348 |
}; |
| 349 |
|
| 350 |
|
| 351 |
|
| 352 |
/** |
| 353 |
* Block version of the sparse Vanka preconditioner. This class divides the |
| 354 |
* matrix into blocks and works on the diagonal blocks only, which of course |
| 355 |
* reduces the efficiency as preconditioner, but is perfectly parallelizable. |
| 356 |
* The constructor takes a parameter into how many blocks the matrix shall be |
| 357 |
* subdivided and then lets the underlying class do the work. Division of the |
| 358 |
* matrix is done in several ways which are described in detail below. |
| 359 |
* |
| 360 |
* This class is probably useless if you don't have a multiprocessor system, |
| 361 |
* since then the amount of work per preconditioning step is the same as for |
| 362 |
* the @p SparseVanka class, but preconditioning properties are worse. On the |
| 363 |
* other hand, if you have a multiprocessor system, the worse preconditioning |
| 364 |
* quality (leading to more iterations of the linear solver) usually is well |
| 365 |
* balanced by the increased speed of application due to the parallelization, |
| 366 |
* leading to an overall decrease in elapsed wall-time for solving your linear |
| 367 |
* system. It should be noted that the quality as preconditioner reduces with |
| 368 |
* growing number of blocks, so there may be an optimal value (in terms of |
| 369 |
* wall-time per linear solve) for the number of blocks. |
| 370 |
* |
| 371 |
* To facilitate writing portable code, if the number of blocks into which the |
| 372 |
* matrix is to be subdivided, is set to one, then this class acts just like |
| 373 |
* the @p SparseVanka class. You may therefore want to set the number of |
| 374 |
* blocks equal to the number of processors you have. |
| 375 |
* |
| 376 |
* Note that the parallelization is done if <tt>deal.II</tt> was configured |
| 377 |
* for multithread use and that the number of threads which is spawned equals |
| 378 |
* the number of blocks. This is reasonable since you will not want to set the |
| 379 |
* number of blocks unnecessarily large, since, as mentioned, this reduces the |
| 380 |
* preconditioning properties. |
| 381 |
* |
| 382 |
* |
| 383 |
* <h3>Splitting the matrix into blocks</h3> |
| 384 |
* |
| 385 |
* Splitting the matrix into blocks is always done in a way such that the |
| 386 |
* blocks are not necessarily of equal size, but such that the number of |
| 387 |
* selected degrees of freedom for which a local system is to be solved is |
| 388 |
* equal between blocks. The reason for this strategy to subdivision is |
| 389 |
* load-balancing for multithreading. There are several possibilities to |
| 390 |
* actually split the matrix into blocks, which are selected by the flag @p |
| 391 |
* blocking_strategy that is passed to the constructor. By a block, we will in |
| 392 |
* the sequel denote a list of indices of degrees of freedom; the algorithm |
| 393 |
* will work on each block separately, i.e. the solutions of the local systems |
| 394 |
* corresponding to a degree of freedom of one block will only be used to |
| 395 |
* update the degrees of freedom belonging to the same block, but never to |
| 396 |
* update degrees of freedom of other blocks. A block can be a consecutive |
| 397 |
* list of indices, as in the first alternative below, or a nonconsecutive |
| 398 |
* list of indices. Of course, we assume that the intersection of each two |
| 399 |
* blocks is empty and that the union of all blocks equals the interval |
| 400 |
* <tt>[0,N)</tt>, where @p N is the number of degrees of freedom of the |
| 401 |
* system of equations. |
| 402 |
* |
| 403 |
* <ul> |
| 404 |
* <li> @p index_intervals: Here, we chose the blocks to be intervals |
| 405 |
* <tt>[a_i,a_{i+1</tt>)}, i.e. consecutive degrees of freedom are usually |
| 406 |
* also within the same block. This is a reasonable strategy, if the degrees |
| 407 |
* of freedom have, for example, be re-numbered using the Cuthill-McKee |
| 408 |
* algorithm, in which spatially neighboring degrees of freedom have |
| 409 |
* neighboring indices. In that case, coupling in the matrix is usually |
| 410 |
* restricted to the vicinity of the diagonal as well, and we can simply cut |
| 411 |
* the matrix into blocks. |
| 412 |
* |
| 413 |
* The bounds of the intervals, i.e. the @p a_i above, are chosen such that |
| 414 |
* the number of degrees of freedom on which we shall work (i.e. usually the |
| 415 |
* degrees of freedom corresponding to Lagrange multipliers) is about the same |
| 416 |
* in each block; this does not mean, however, that the sizes of the blocks |
| 417 |
* are equal, since the blocks also comprise the other degrees of freedom for |
| 418 |
* which no local system is solved. In the extreme case, consider that all |
| 419 |
* Lagrange multipliers are sorted to the end of the range of DoF indices, |
| 420 |
* then the first block would be very large, since it comprises all other DoFs |
| 421 |
* and some Lagrange multipliers, while all other blocks are rather small and |
| 422 |
* comprise only Langrange multipliers. This strategy therefore does not only |
| 423 |
* depend on the order in which the Lagrange DoFs are sorted, but also on the |
| 424 |
* order in which the other DoFs are sorted. It is therefore necessary to note |
| 425 |
* that this almost renders the capability as preconditioner useless if the |
| 426 |
* degrees of freedom are numbered by component, i.e. all Lagrange multipliers |
| 427 |
* en bloc. |
| 428 |
* |
| 429 |
* <li> @p adaptive: This strategy is a bit more clever in cases where the |
| 430 |
* Langrange DoFs are clustered, as in the example above. It works as follows: |
| 431 |
* it first groups the Lagrange DoFs into blocks, using the same strategy as |
| 432 |
* above. However, instead of grouping the other DoFs into the blocks of |
| 433 |
* Lagrange DoFs with nearest DoF index, it decides for each non-Lagrange DoF |
| 434 |
* to put it into the block of Lagrange DoFs which write to this non-Lagrange |
| 435 |
* DoF most often. This makes it possible to even sort the Lagrange DoFs to |
| 436 |
* the end and still associate spatially neighboring non-Lagrange DoFs to the |
| 437 |
* same blocks where the respective Lagrange DoFs are, since they couple to |
| 438 |
* each other while spatially distant DoFs don't couple. |
| 439 |
* |
| 440 |
* The additional computational effort to sorting the non-Lagrange DoFs is not |
| 441 |
* very large compared with the inversion of the local systems and applying |
| 442 |
* the preconditioner, so this strategy might be reasonable if you want to |
| 443 |
* sort your degrees of freedom by component. If the degrees of freedom are |
| 444 |
* not sorted by component, the results of the both strategies outlined above |
| 445 |
* does not differ much. However, unlike the first strategy, the performance |
| 446 |
* of the second strategy does not deteriorate if the DoFs are renumbered by |
| 447 |
* component. |
| 448 |
* </ul> |
| 449 |
* |
| 450 |
* |
| 451 |
* <h3>Typical results</h3> |
| 452 |
* |
| 453 |
* As a prototypical test case, we use a nonlinear problem from optimization, |
| 454 |
* which leads to a series of saddle point problems, each of which is solved |
| 455 |
* using GMRES with Vanka as preconditioner. The equation had approx. 850 |
| 456 |
* degrees of freedom. With the non-blocked version @p SparseVanka (or @p |
| 457 |
* SparseBlockVanka with <tt>n_blocks==1</tt>), the following numbers of |
| 458 |
* iterations is needed to solver the linear system in each nonlinear step: |
| 459 |
* @verbatim |
| 460 |
* 101 68 64 53 35 21 |
| 461 |
* @endverbatim |
| 462 |
* |
| 463 |
* With four blocks, we need the following numbers of iterations |
| 464 |
* @verbatim |
| 465 |
* 124 88 83 66 44 28 |
| 466 |
* @endverbatim |
| 467 |
* As can be seen, more iterations are needed. However, in terms of computing |
| 468 |
* time, the first version needs 72 seconds wall time (and 79 seconds CPU |
| 469 |
* time, which is more than wall time since some other parts of the program |
| 470 |
* were parallelized as well), while the second version needed 53 second wall |
| 471 |
* time (and 110 seconds CPU time) on a four processor machine. The total time |
| 472 |
* is in both cases dominated by the linear solvers. In this case, it is |
| 473 |
* therefore worth while using the blocked version of the preconditioner if |
| 474 |
* wall time is more important than CPU time. |
| 475 |
* |
| 476 |
* The results with the block version above were obtained with the first |
| 477 |
* blocking strategy and the degrees of freedom were not numbered by |
| 478 |
* component. Using the second strategy does not much change the numbers of |
| 479 |
* iterations (at most by one in each step) and they also do not change when |
| 480 |
* the degrees of freedom are sorted by component, while the first strategy |
| 481 |
* significantly deteriorated. |
| 482 |
*/ |
| 483 |
template <typename number> |
| 484 |
class SparseBlockVanka : public SparseVanka<number> |
| 485 |
{ |
| 486 |
public: |
| 487 |
/** |
| 488 |
* Declare type for container size. |
| 489 |
*/ |
| 490 |
using size_type = types::global_dof_index; |
| 491 |
|
| 492 |
/** |
| 493 |
* Enumeration of the different methods by which the DoFs are distributed to |
| 494 |
* the blocks on which we are to work. |
| 495 |
*/ |
| 496 |
enum BlockingStrategy |
| 497 |
{ |
| 498 |
/** |
| 499 |
* Block by index intervals. |
| 500 |
*/ |
| 501 |
index_intervals, |
| 502 |
/** |
| 503 |
* Block with an adaptive strategy. |
| 504 |
*/ |
| 505 |
adaptive |
| 506 |
}; |
| 507 |
|
| 508 |
/** |
| 509 |
* Constructor. Pass all arguments except for @p n_blocks to the base class. |
| 510 |
*/ |
| 511 |
SparseBlockVanka(const SparseMatrix<number> &M, |
| 512 |
const std::vector<bool> &selected, |
| 513 |
const unsigned int n_blocks, |
| 514 |
const BlockingStrategy blocking_strategy); |
| 515 |
|
| 516 |
/** |
| 517 |
* Apply the preconditioner. |
| 518 |
*/ |
| 519 |
template <typename number2> |
| 520 |
void |
| 521 |
vmult(Vector<number2> &dst, const Vector<number2> &src) const; |
| 522 |
|
| 523 |
|
| 524 |
/** |
| 525 |
* Apply the transpose preconditioner. |
| 526 |
*/ |
| 527 |
template <typename number2> |
| 528 |
void |
| 529 |
Tvmult(Vector<number2> &dst, const Vector<number2> &src) const; |
| 530 |
|
| 531 |
/** |
| 532 |
* Determine an estimate for the memory consumption (in bytes) of this |
| 533 |
* object. |
| 534 |
*/ |
| 535 |
std::size_t |
| 536 |
memory_consumption() const; |
| 537 |
|
| 538 |
private: |
| 539 |
/** |
| 540 |
* Store the number of blocks. |
| 541 |
*/ |
| 542 |
const unsigned int n_blocks; |
| 543 |
|
| 544 |
/** |
| 545 |
* In this field, we precompute for each block which degrees of freedom |
| 546 |
* belong to it. Thus, if <tt>dof_masks[i][j]==true</tt>, then DoF @p j |
| 547 |
* belongs to block @p i. Of course, no other <tt>dof_masks[l][j]</tt> may |
| 548 |
* be @p true for <tt>l!=i</tt>. This computation is done in the |
| 549 |
* constructor, to avoid recomputing each time the preconditioner is called. |
| 550 |
*/ |
| 551 |
std::vector<std::vector<bool>> dof_masks; |
| 552 |
|
| 553 |
/** |
| 554 |
* Compute the contents of the field @p dof_masks. This function is called |
| 555 |
* from the constructor. |
| 556 |
*/ |
| 557 |
void |
| 558 |
compute_dof_masks(const SparseMatrix<number> &M, |
| 559 |
const std::vector<bool> &selected, |
| 560 |
const BlockingStrategy blocking_strategy); |
| 561 |
}; |
| 562 |
|
| 563 |
/** @} */ |
| 564 |
/* ---------------------------------- Inline functions ------------------- */ |
| 565 |
|
| 566 |
#ifndef DOXYGEN |
| 567 |
|
| 568 |
template <typename number> |
| 569 |
inline typename SparseVanka<number>::size_type |
| 570 |
SparseVanka<number>::m() const |
| 571 |
{ |
| 572 |
Assert(_m != 0, ExcNotInitialized()); |
| 573 |
return _m; |
| 574 |
} |
| 575 |
|
| 576 |
template <typename number> |
| 577 |
inline typename SparseVanka<number>::size_type |
| 578 |
SparseVanka<number>::n() const |
| 579 |
{ |
| 580 |
Assert(_n != 0, ExcNotInitialized()); |
| 581 |
return _n; |
| 582 |
} |
| 583 |
|
| 584 |
#endif // DOXYGEN |
| 585 |
|
| 586 |
DEAL_II_NAMESPACE_CLOSE |
| 587 |
|
| 588 |
#endif |