| 1 |
// ------------------------------------------------------------------------ |
| 2 |
// |
| 3 |
// SPDX-License-Identifier: LGPL-2.1-or-later |
| 4 |
// Copyright (C) 1999 - 2024 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_mg_base_h |
| 16 |
#define dealii_mg_base_h |
| 17 |
|
| 18 |
/* |
| 19 |
* This file contains some abstract base classes |
| 20 |
* used by Multigrid. |
| 21 |
*/ |
| 22 |
|
| 23 |
#include <deal.II/base/config.h> |
| 24 |
|
| 25 |
#include <deal.II/base/enable_observer_pointer.h> |
| 26 |
#include <deal.II/base/observer_pointer.h> |
| 27 |
|
| 28 |
#include <deal.II/lac/vector.h> |
| 29 |
|
| 30 |
|
| 31 |
DEAL_II_NAMESPACE_OPEN |
| 32 |
|
| 33 |
/** |
| 34 |
* @addtogroup mg |
| 35 |
* @{ |
| 36 |
*/ |
| 37 |
|
| 38 |
/** |
| 39 |
* Multilevel matrix base. This class sets up the interface needed by |
| 40 |
* multilevel algorithms. It has no relation to the actual matrix type and |
| 41 |
* takes the vector class as only template argument. |
| 42 |
* |
| 43 |
* Usually, the derived class mg::Matrix, which operates on an MGLevelObject |
| 44 |
* of matrices, will be sufficient for applications. |
| 45 |
*/ |
| 46 |
template <typename VectorType> |
| 47 |
class MGMatrixBase : public EnableObserverPointer |
| 48 |
{ |
| 49 |
public: |
| 50 |
/* |
| 51 |
* Virtual destructor. |
| 52 |
*/ |
| 53 |
virtual ~MGMatrixBase() override = default; |
| 54 |
|
| 55 |
/** |
| 56 |
* Matrix-vector-multiplication on a certain level. |
| 57 |
*/ |
| 58 |
virtual void |
| 59 |
vmult(const unsigned int level, |
| 60 |
VectorType &dst, |
| 61 |
const VectorType &src) const = 0; |
| 62 |
|
| 63 |
/** |
| 64 |
* Adding matrix-vector-multiplication on a certain level. |
| 65 |
*/ |
| 66 |
virtual void |
| 67 |
vmult_add(const unsigned int level, |
| 68 |
VectorType &dst, |
| 69 |
const VectorType &src) const = 0; |
| 70 |
|
| 71 |
/** |
| 72 |
* Transpose matrix-vector-multiplication on a certain level. |
| 73 |
*/ |
| 74 |
virtual void |
| 75 |
Tvmult(const unsigned int level, |
| 76 |
VectorType &dst, |
| 77 |
const VectorType &src) const = 0; |
| 78 |
|
| 79 |
/** |
| 80 |
* Adding transpose matrix-vector-multiplication on a certain level. |
| 81 |
*/ |
| 82 |
virtual void |
| 83 |
Tvmult_add(const unsigned int level, |
| 84 |
VectorType &dst, |
| 85 |
const VectorType &src) const = 0; |
| 86 |
|
| 87 |
/** |
| 88 |
* Return the minimal level for which matrices are stored. |
| 89 |
*/ |
| 90 |
virtual unsigned int |
| 91 |
get_minlevel() const = 0; |
| 92 |
|
| 93 |
/** |
| 94 |
* Return the minimal level for which matrices are stored. |
| 95 |
*/ |
| 96 |
virtual unsigned int |
| 97 |
get_maxlevel() const = 0; |
| 98 |
}; |
| 99 |
|
| 100 |
|
| 101 |
/** |
| 102 |
* Base class for coarse grid solvers. This defines the virtual parenthesis |
| 103 |
* operator, being the interface used by multigrid methods. Any implementation |
| 104 |
* will be done by derived classes. |
| 105 |
*/ |
| 106 |
template <typename VectorType> |
| 107 |
class MGCoarseGridBase : public EnableObserverPointer |
| 108 |
{ |
| 109 |
public: |
| 110 |
/** |
| 111 |
* Virtual destructor. |
| 112 |
*/ |
| 113 |
virtual ~MGCoarseGridBase() override = default; |
| 114 |
|
| 115 |
/** |
| 116 |
* Solution operator. |
| 117 |
*/ |
| 118 |
virtual void |
| 119 |
operator()(const unsigned int level, |
| 120 |
VectorType &dst, |
| 121 |
const VectorType &src) const = 0; |
| 122 |
}; |
| 123 |
|
| 124 |
|
| 125 |
/** |
| 126 |
* Base class used to declare the operations needed by a concrete class |
| 127 |
* implementing prolongation and restriction of vectors in the multigrid |
| 128 |
* context. This class is abstract and has no implementation of these |
| 129 |
* operations. |
| 130 |
* |
| 131 |
* There are several derived classes, reflecting the fact that vector types |
| 132 |
* and numbering of the fine-grid discretization and of the multi-level |
| 133 |
* implementation are independent. |
| 134 |
* |
| 135 |
* If you use multigrid for a single PDE or for your complete system of |
| 136 |
* equations, you will use MGTransferPrebuilt together with Multigrid. The |
| 137 |
* vector types used on the fine grid as well as for the multilevel operations |
| 138 |
* may be Vector or BlockVector. In both cases, MGTransferPrebuilt will |
| 139 |
* operate on all components of the solution. |
| 140 |
* |
| 141 |
* @note For the following, it is important to realize the difference between |
| 142 |
* a solution |
| 143 |
* @ref GlossComponent "component" |
| 144 |
* and a solution |
| 145 |
* @ref GlossBlock "block". |
| 146 |
* The distinction only applies if vector valued elements are used, but is |
| 147 |
* quite important then. This is reflected in the fact that it is not possible |
| 148 |
* right now to use transfer classes based on MGTransferComponentBase for |
| 149 |
* genuine vector valued elements, but descendants of MGTransferBlockBase |
| 150 |
* would have to be applied. In the following text, we will use the term |
| 151 |
* <em>block</em>, but remark that it might refer to components as well. |
| 152 |
* |
| 153 |
* @todo update the following documentation, since it does not reflect the |
| 154 |
* latest changes in structure. |
| 155 |
* |
| 156 |
* For mixed systems, it may be required to do multigrid only for a single |
| 157 |
* component or for some components. The classes MGTransferSelect and |
| 158 |
* MGTransferBlock handle these cases. |
| 159 |
* |
| 160 |
* MGTransferSelect is used if you use multigrid (on Vector objects) for a |
| 161 |
* single component, possibly grouped using <tt>mg_target_component</tt>. |
| 162 |
* |
| 163 |
* The class MGTransferBlock handles the case where your multigrid method |
| 164 |
* operates on BlockVector objects. These can contain all or a consecutive set |
| 165 |
* of the blocks of the complete system. Since most smoothers cannot operate |
| 166 |
* on block structures, it is not clear whether this case is really useful. |
| 167 |
* Therefore, a tested implementation of this case will be supplied when |
| 168 |
* needed. |
| 169 |
*/ |
| 170 |
template <typename VectorType> |
| 171 |
class MGTransferBase : public EnableObserverPointer |
| 172 |
{ |
| 173 |
public: |
| 174 |
/** |
| 175 |
* Destructor. Does nothing here, but needs to be declared virtual anyway. |
| 176 |
*/ |
| 177 |
virtual ~MGTransferBase() override = default; |
| 178 |
|
| 179 |
/** |
| 180 |
* Prolongate a vector from level <tt>to_level-1</tt> to level |
| 181 |
* <tt>to_level</tt>. The previous content of <tt>dst</tt> is overwritten. |
| 182 |
* |
| 183 |
* @arg src is a vector with as many elements as there are degrees of |
| 184 |
* freedom on the coarser level involved. |
| 185 |
* |
| 186 |
* @arg dst has as many elements as there are degrees of freedom on the |
| 187 |
* finer level. |
| 188 |
*/ |
| 189 |
virtual void |
| 190 |
prolongate(const unsigned int to_level, |
| 191 |
VectorType &dst, |
| 192 |
const VectorType &src) const = 0; |
| 193 |
|
| 194 |
/** |
| 195 |
* Prolongate a vector from level <tt>to_level-1</tt> to level |
| 196 |
* <tt>to_level</tt>, summing into the previous content of <tt>dst</tt>. |
| 197 |
* |
| 198 |
* @arg src is a vector with as many elements as there are degrees of |
| 199 |
* freedom on the coarser level involved. |
| 200 |
* |
| 201 |
* @arg dst has as many elements as there are degrees of freedom on the |
| 202 |
* finer level. |
| 203 |
*/ |
| 204 |
virtual void |
| 205 |
prolongate_and_add(const unsigned int to_level, |
| 206 |
VectorType &dst, |
| 207 |
const VectorType &src) const; |
| 208 |
|
| 209 |
/** |
| 210 |
* Restrict a vector from level <tt>from_level</tt> to level |
| 211 |
* <tt>from_level-1</tt> and add this restriction to <tt>dst</tt>. If the |
| 212 |
* region covered by cells on level <tt>from_level</tt> is smaller than that |
| 213 |
* of level <tt>from_level-1</tt> (local refinement), then some degrees of |
| 214 |
* freedom in <tt>dst</tt> are active and will not be altered. For the other |
| 215 |
* degrees of freedom, the result of the restriction is added. |
| 216 |
* |
| 217 |
* @arg src is a vector with as many elements as there are degrees of |
| 218 |
* freedom on the finer level |
| 219 |
* |
| 220 |
* @arg dst has as many elements as there are degrees of freedom on the |
| 221 |
* coarser level. |
| 222 |
*/ |
| 223 |
virtual void |
| 224 |
restrict_and_add(const unsigned int from_level, |
| 225 |
VectorType &dst, |
| 226 |
const VectorType &src) const = 0; |
| 227 |
}; |
| 228 |
|
| 229 |
|
| 230 |
|
| 231 |
/** |
| 232 |
* Base class for multigrid smoothers. Does nothing but defining the interface |
| 233 |
* used by multigrid methods. |
| 234 |
* |
| 235 |
* The smoother interface provides two methods, a smooth() method and an |
| 236 |
* apply() method. The multigrid preconditioner interfaces distinguish between |
| 237 |
* the two for efficiency reasons: Upon entry to the preconditioner operations, |
| 238 |
* the vector @p u needs to be set to zero and smoothing starts by a simple |
| 239 |
* application of the smoother on the @p rhs vector. This method is provided by |
| 240 |
* the apply() method of this class. It is the same as first setting @p u to |
| 241 |
* zero and then calling smooth(), but for many classes the separate apply() |
| 242 |
* interface is more efficient because it can skip one matrix-vector product. |
| 243 |
* |
| 244 |
* In the multigrid preconditioner interfaces, the apply() method is used for |
| 245 |
* the pre-smoothing operation because the previous content in the solution |
| 246 |
* vector needs to be overwritten for a new incoming residual. On the other |
| 247 |
* hand, all subsequent operations need to smooth the content already present |
| 248 |
* in the vector @p u given the right hand side, which is done by smooth(). |
| 249 |
*/ |
| 250 |
template <typename VectorType> |
| 251 |
class MGSmootherBase : public EnableObserverPointer |
| 252 |
{ |
| 253 |
public: |
| 254 |
/** |
| 255 |
* Virtual destructor. |
| 256 |
*/ |
| 257 |
virtual ~MGSmootherBase() override = default; |
| 258 |
|
| 259 |
/** |
| 260 |
* Release matrices. |
| 261 |
*/ |
| 262 |
virtual void |
| 263 |
clear() = 0; |
| 264 |
|
| 265 |
/** |
| 266 |
* Smoothing function that smooths the content in @p u given the right hand |
| 267 |
* side vector @p rhs. This is the function used in multigrid methods. |
| 268 |
*/ |
| 269 |
virtual void |
| 270 |
smooth(const unsigned int level, |
| 271 |
VectorType &u, |
| 272 |
const VectorType &rhs) const = 0; |
| 273 |
|
| 274 |
/** |
| 275 |
* As opposed to the smooth() function, this function applies the action of |
| 276 |
* the smoothing, overwriting the previous content in the vector u. This |
| 277 |
* function must be equivalent to the following code |
| 278 |
* @code |
| 279 |
* u = 0; |
| 280 |
* smooth(level, u, rhs); |
| 281 |
* @endcode |
| 282 |
* but can usually be implemented more efficiently than the former. If a |
| 283 |
* particular smoother does not override the apply() method, the default |
| 284 |
* implementation as described here is used. |
| 285 |
* |
| 286 |
* In the multigrid preconditioner interfaces, the apply() method is used for |
| 287 |
* the pre-smoothing operation because the previous content in the solution |
| 288 |
* vector needs to be overwritten for a new incoming residual. On the other |
| 289 |
* hand, all subsequent operations need to smooth the content already present |
| 290 |
* in the vector @p u given the right hand side, which is done by smooth(). |
| 291 |
*/ |
| 292 |
virtual void |
| 293 |
apply(const unsigned int level, VectorType &u, const VectorType &rhs) const; |
| 294 |
}; |
| 295 |
|
| 296 |
/** @} */ |
| 297 |
|
| 298 |
DEAL_II_NAMESPACE_CLOSE |
| 299 |
|
| 300 |
#endif |