[9102] | 1 | ///
|
---|
| 2 | /// This file is part of ILNumerics Community Edition.
|
---|
| 3 | ///
|
---|
| 4 | /// ILNumerics Community Edition - high performance computing for applications.
|
---|
| 5 | /// Copyright (C) 2006 - 2012 Haymo Kutschbach, http://ilnumerics.net
|
---|
| 6 | ///
|
---|
| 7 | /// ILNumerics Community Edition is free software: you can redistribute it and/or modify
|
---|
| 8 | /// it under the terms of the GNU General Public License version 3 as published by
|
---|
| 9 | /// the Free Software Foundation.
|
---|
| 10 | ///
|
---|
| 11 | /// ILNumerics Community Edition is distributed in the hope that it will be useful,
|
---|
| 12 | /// but WITHOUT ANY WARRANTY; without even the implied warranty of
|
---|
| 13 | /// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
---|
| 14 | /// GNU General Public License for more details.
|
---|
| 15 | ///
|
---|
| 16 | /// You should have received a copy of the GNU General Public License
|
---|
| 17 | /// along with ILNumerics Community Edition. See the file License.txt in the root
|
---|
| 18 | /// of your distribution package. If not, see <http://www.gnu.org/licenses/>.
|
---|
| 19 | ///
|
---|
| 20 | /// In addition this software uses the following components and/or licenses:
|
---|
| 21 | ///
|
---|
| 22 | /// =================================================================================
|
---|
| 23 | /// The Open Toolkit Library License
|
---|
| 24 | ///
|
---|
| 25 | /// Copyright (c) 2006 - 2009 the Open Toolkit library.
|
---|
| 26 | ///
|
---|
| 27 | /// Permission is hereby granted, free of charge, to any person obtaining a copy
|
---|
| 28 | /// of this software and associated documentation files (the "Software"), to deal
|
---|
| 29 | /// in the Software without restriction, including without limitation the rights to
|
---|
| 30 | /// use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of
|
---|
| 31 | /// the Software, and to permit persons to whom the Software is furnished to do
|
---|
| 32 | /// so, subject to the following conditions:
|
---|
| 33 | ///
|
---|
| 34 | /// The above copyright notice and this permission notice shall be included in all
|
---|
| 35 | /// copies or substantial portions of the Software.
|
---|
| 36 | ///
|
---|
| 37 | /// =================================================================================
|
---|
| 38 | ///
|
---|
| 39 |
|
---|
| 40 | using System;
|
---|
| 41 | using System.Collections.Generic;
|
---|
| 42 | using System.Text;
|
---|
| 43 | using ILNumerics.Storage;
|
---|
| 44 | using ILNumerics.Misc;
|
---|
| 45 | using ILNumerics.Exceptions;
|
---|
| 46 |
|
---|
| 47 |
|
---|
| 48 |
|
---|
| 49 | namespace ILNumerics {
|
---|
| 50 | public partial class ILMath {
|
---|
| 51 |
|
---|
| 52 | |
---|
| 53 |
|
---|
| 54 | /// <summary>
|
---|
| 55 | /// Solve linear equation system
|
---|
| 56 | /// </summary>
|
---|
| 57 | /// <param name="A">Matrix A. Size [n x q]</param>
|
---|
| 58 | /// <param name="B">Right hand side B. Size [n x m]</param>
|
---|
| 59 | /// <returns>Solution x solving the equation system: multiply(A, x) = B. Size [n x m]</returns>
|
---|
| 60 | /// <remarks><para>Depending on the structure and properties of A, the equation system will be solved in different ways:
|
---|
| 61 | /// <list type="bullet">
|
---|
| 62 | /// <item>If A is square (q == n) and an <b>upper or lower triangular</b> matrix, the system will directly be solved via backward- or forward substitution. Therefore the Lapack function ?trtrs will be used, whenever the memory layout of A is suitable. This may be the case even for reference ILArray's!
|
---|
| 63 | /// <example><code><![CDATA[ILArray<double> A = ILMath.randn(4); // construct 4 x 4 matrix
|
---|
| 64 | /// A = A.T; // A is a reference array now! The transpose is fast and does not consume much memory
|
---|
| 65 | /// // now construct a right side and solve the equations:
|
---|
| 66 | /// ILArray<double> B = new ILArray<double> (1.0,2.0,3.0).T;
|
---|
| 67 | /// ILMath.linsolve(A,B); // ... will be carried out via Lapack, even for all arrays involved being reference arrays! ]]></code></example></item>
|
---|
| 68 | /// <item><para>if A is square and symmetric or hermitian, A will be decomposed into a triangular equation system using cholesky factorization and Lapack. The system is than solved using performant Lapack routines.</para>
|
---|
| 69 | /// <para>If during the cholesky factorization A was found to be <b>not positive definite</b> - the cholesky factorization is canceled. </para></item>
|
---|
| 70 | /// <item>otherwise, if A is square only, it will be decomposed into upper and lower triangular martices using LU decomposition and Lapack. The triangular system is than solved using performant Lapack routines.</item>
|
---|
| 71 | /// <item>otherwise, if A is of size [q x n] and q != n, the system is solved using QR decomposition. A may be rank deficient. The solution is computed by use of the Lapack routine '?gelsy' and may be a reference array.</item>
|
---|
| 72 | /// </list></para>
|
---|
| 73 | /// <para>Compatibility with Matlab<sup>(R)</sup>: If A is square, the algorithm used follows the same logic as Matlab up to Rel 14, with the exception of Hessenberg matrices wich are solved via LU factorization here. The un-squared case is handled differently. A direct Lapack driver function (?gelsy) is used here. Therefore the solutions might differ! However, the solution will of course fullfill the equation A * x = B without round off errrors. </para>
|
---|
| 74 | /// <para>For specifiying the rank of A in the unsquare case (q != n), <see cref="ILNumerics.ILMath.eps"/> is used.</para></remarks>
|
---|
| 75 | public static ILRetArray< double > linsolve(ILInArray< double > A, ILInArray< double > B) {
|
---|
| 76 | if (object.Equals(A,null) || object.Equals(B,null))
|
---|
| 77 | throw new ILArgumentException("parameter must not be null!");
|
---|
| 78 | using (ILScope.Enter(A, B)) {
|
---|
| 79 | MatrixProperties props = MatrixProperties.None;
|
---|
| 80 | if (A.Size[0] == A.Size[1]) {
|
---|
| 81 | props |= MatrixProperties.Square;
|
---|
| 82 | if (ILMath.istriup(A)) {
|
---|
| 83 | props |= MatrixProperties.UpperTriangular;
|
---|
| 84 | return linsolve(A, B, ref props);
|
---|
| 85 | }
|
---|
| 86 | if (ILMath.istrilow(A)) {
|
---|
| 87 | props |= MatrixProperties.LowerTriangular;
|
---|
| 88 | return linsolve(A, B, ref props);
|
---|
| 89 | }
|
---|
| 90 | if (ILMath.ishermitian(A)) {
|
---|
| 91 | // give cholesky a try
|
---|
| 92 | props |= MatrixProperties.Hermitian;
|
---|
| 93 | props |= MatrixProperties.PositivDefinite;
|
---|
| 94 | ILArray< double> ret = linsolve(A, B, ref props);
|
---|
| 95 | if (!object.Equals(ret, null)) {
|
---|
| 96 | return ret;
|
---|
| 97 | } else {
|
---|
| 98 | props ^= MatrixProperties.PositivDefinite;
|
---|
| 99 | }
|
---|
| 100 | }
|
---|
| 101 | }
|
---|
| 102 | return linsolve(A, B, ref props);
|
---|
| 103 | }
|
---|
| 104 | }
|
---|
| 105 |
|
---|
| 106 | /// <summary>
|
---|
| 107 | /// Solve linear equation system
|
---|
| 108 | /// </summary>
|
---|
| 109 | /// <param name="A">Matrix A. Size [n x q]</param>
|
---|
| 110 | /// <param name="B">Right hand side B. Size [n x m]</param>
|
---|
| 111 | /// <param name="props">Matrix properties. If defined, no checks are made for the structure of A. If the
|
---|
| 112 | /// matrix A was found to be (close to or) singular, the 'MatrixProperties.Singular' flag in props will be set.
|
---|
| 113 | /// This flag should be tested on return, in order to verify the reliability of the solution.</param>
|
---|
| 114 | /// <returns>The solution x solving multiply(A,x) = B. Size [n x m]</returns>
|
---|
| 115 | /// <remarks><para>Depending on the <paramref name="props"/> parameter the equation system will be solved
|
---|
| 116 | /// differently for special structures of A:
|
---|
| 117 | /// <list type="bullet">
|
---|
| 118 | /// <item>If A is square (q == n) and an <b>upper or lower triangular</b> matrix, the system will directly
|
---|
| 119 | /// be solved via backward- or forward substitution. Therefore the Lapack function ?trtrs will be used,
|
---|
| 120 | /// whenever the memory layout of A is suitable. This may be the case even for reference ILArray's!
|
---|
| 121 | /// <example><code><![CDATA[ILArray<double> A = ILMath.randn(4); // construct 4 x 4 matrix
|
---|
| 122 | /// A = A.T; // A is a reference array now! The transpose is fast and does not consume much memory
|
---|
| 123 | /// // now construct a right side and solve the equations:
|
---|
| 124 | /// ILArray<double> B = new ILArray<double> (1.0,2.0,3.0).T;
|
---|
| 125 | /// ILMath.linsolve(A,B); // ... will be carried out via Lapack, even for all arrays involved being reference arrays! ]]></code></example></item>
|
---|
| 126 | /// <item><para>If A is square and symmetric or hermitian, A will be decomposed into a triangular equation
|
---|
| 127 | /// system using cholesky factorization and Lapack. The system is than solved using performant Lapack routines.</para>
|
---|
| 128 | /// <para>If during the cholesky factorization A was found to be <b>not positive definite</b> - the
|
---|
| 129 | /// corresponding flag in props will be cleaned and <c>null</c> will be returned.</para></item>
|
---|
| 130 | /// <item>Otherwise if A is square only, it will be decomposed into upper and lower triangular matrices
|
---|
| 131 | /// using LU decomposition and Lapack. The triangular system is than solved using performant Lapack routines.</item>
|
---|
| 132 | /// <item>Otherwise, if A is of size [q x n] and q != n, the system is solved using QR decomposition.
|
---|
| 133 | /// A may be rank deficient. The solution is computed by use of the Lapack routine '?gelsy' and may be a
|
---|
| 134 | /// reference array.</item>
|
---|
| 135 | /// </list></para>
|
---|
| 136 | /// <para>Compatibility with Matlab<sup>(R)</sup>: If A is square, the algorithm used follows the same
|
---|
| 137 | /// logic as Matlab up to Rel 14, with the exception of Hessenberg matrices wich are solved via LU
|
---|
| 138 | /// factorization here. The un-squared case is handled differently. A direct Lapack driver function
|
---|
| 139 | /// (?gelsy) is used here. Therefore the solutions might differ! However, the solution will of course
|
---|
| 140 | /// fullfill the equation A * x = B without round off errrors. </para>
|
---|
| 141 | /// <para>For specifiying the rank of A in the unsquare case (q != n),
|
---|
| 142 | /// <see cref="ILNumerics.ILMath.eps"/> is used.</para></remarks>
|
---|
| 143 | public static ILRetArray< double > linsolve(ILInArray< double > A, ILInArray< double > B, ref MatrixProperties props) {
|
---|
| 144 | if (object.Equals(A,null))
|
---|
| 145 | throw new ILArgumentException("input argument A must not be null!");
|
---|
| 146 | if (object.Equals(B,null))
|
---|
| 147 | throw new ILArgumentException("input argument B must not be null!");
|
---|
| 148 | using (ILScope.Enter(A, B)) {
|
---|
| 149 | if (A.IsEmpty || B.IsEmpty)
|
---|
| 150 | return empty< double>(A.Size);
|
---|
| 151 | if (A.Size[0] != B.Size[0])
|
---|
| 152 | throw new ILArgumentException("number of rows for matrix A must match number of rows for RHS!");
|
---|
| 153 | int info = 0, m = A.Size[0];
|
---|
| 154 | ILArray< double> ret = empty<double>(ILSize.Empty00);
|
---|
| 155 | if (m == A.Size[1]) {
|
---|
| 156 | props |= MatrixProperties.Square;
|
---|
| 157 | if ((props & MatrixProperties.LowerTriangular) != 0) {
|
---|
| 158 | ret.a = solveLowerTriangularSystem(A, B, ref info);
|
---|
| 159 | if (info > 0)
|
---|
| 160 | props |= MatrixProperties.Singular;
|
---|
| 161 | return ret;
|
---|
| 162 | }
|
---|
| 163 | if ((props & MatrixProperties.UpperTriangular) != 0) {
|
---|
| 164 | ret.a = solveUpperTriangularSystem(A, B, ref info);
|
---|
| 165 | if (info > 0)
|
---|
| 166 | props |= MatrixProperties.Singular;
|
---|
| 167 | return ret;
|
---|
| 168 | }
|
---|
| 169 | if ((props & MatrixProperties.Hermitian) != 0) {
|
---|
| 170 | ILDenseStorage< double> cholFact = A.Storage.copyUpperTriangle(m);
|
---|
| 171 | /*!HC:lapack_*potrf*/
|
---|
| 172 | Lapack.dpotrf('U', m, cholFact.GetArrayForWrite(), m, ref info);
|
---|
| 173 | if (info > 0) {
|
---|
| 174 | props ^= MatrixProperties.Hermitian;
|
---|
| 175 | cholFact.Dispose();
|
---|
| 176 | return null;
|
---|
| 177 | } else {
|
---|
| 178 | // solve
|
---|
| 179 | ret.a = B.C;
|
---|
| 180 | /*!HC:lapack_*potrs*/
|
---|
| 181 | Lapack.dpotrs('U', m, B.Size[1], cholFact.GetArrayForWrite(), m, ret.GetArrayForWrite(), m, ref info);
|
---|
| 182 | cholFact.Dispose();
|
---|
| 183 | return ret;
|
---|
| 184 | }
|
---|
| 185 | } else {
|
---|
| 186 | // attempt complete (expensive) LU factorization
|
---|
| 187 | ILArray< double> L = A.C;
|
---|
| 188 | int[] pivInd = ILMemoryPool.Pool.New<int>(m);
|
---|
| 189 | /*!HC:lapack_*getrf*/
|
---|
| 190 | Lapack.dgetrf(m, m, L.GetArrayForWrite(), m, pivInd, ref info);
|
---|
| 191 | if (info > 0)
|
---|
| 192 | props |= MatrixProperties.Singular;
|
---|
| 193 | ret.a = B.C;
|
---|
| 194 | /*!HC:lapack_*getrs*/
|
---|
| 195 | Lapack.dgetrs('N', m, B.Size[1], L.GetArrayForWrite(), m, pivInd, ret.GetArrayForWrite(), m, ref info);
|
---|
| 196 | if (info < 0)
|
---|
| 197 | throw new ILArgumentException("failed to solve via lapack dgetrs");
|
---|
| 198 | return ret;
|
---|
| 199 | }
|
---|
| 200 | } else {
|
---|
| 201 | // under- / overdetermined system
|
---|
| 202 | int n = A.Size[1], rank = 0, minMN = (m < n) ? m : n, maxMN = (m > n) ? m : n;
|
---|
| 203 | int nrhs = B.Size[1];
|
---|
| 204 | if (B.Size[0] != m)
|
---|
| 205 | throw new ILArgumentException("right hand side matrix B must match input A!");
|
---|
| 206 | ILArray</*!HCinArr1*/ double> tmpA = A.C;
|
---|
| 207 | if (m < n) {
|
---|
| 208 | ret.a = zeros< double>(n, nrhs);
|
---|
| 209 | ret[r(0, m - 1), full] = B.C;
|
---|
| 210 | } else {
|
---|
| 211 | ret.a = B.C;
|
---|
| 212 | }
|
---|
| 213 | int[] JPVT = new int[n];
|
---|
| 214 | /*!HC:Lapack.?gelsy*/
|
---|
| 215 | Lapack.dgelsy(m, n, B.Size[1], tmpA.GetArrayForWrite(), m, ret.GetArrayForWrite(),
|
---|
| 216 | maxMN, JPVT, ILMath.MachineParameterDouble.eps,
|
---|
| 217 | ref rank, ref info);
|
---|
| 218 | if (n < m) {
|
---|
| 219 | ret.a = ret[r(0, n - 1), full];
|
---|
| 220 | }
|
---|
| 221 | if (rank < minMN)
|
---|
| 222 | props |= MatrixProperties.RankDeficient;
|
---|
| 223 | return ret;
|
---|
| 224 | }
|
---|
| 225 | }
|
---|
| 226 | }
|
---|
| 227 |
|
---|
| 228 | /// <summary>
|
---|
| 229 | /// Solve system of linear equations A*x = b, with A being a upper triangular matrix
|
---|
| 230 | /// </summary>
|
---|
| 231 | /// <param name="A">Input matrix of size [n x n], must be upper triangular. No check is made for that!</param>
|
---|
| 232 | /// <param name="B">Solution vector or matrix. Size [n x m]</param>
|
---|
| 233 | /// <param name="singularityDetect">[Output] This value gives the row of A, where a singularity has been detected (if any). If A is not singular, this will be a negative value.</param>
|
---|
| 234 | /// <returns>Solution x solving A * x = b.</returns>
|
---|
| 235 | /// <remarks><para>The solution will be determined via backward substitution</para>
|
---|
| 236 | /// <para>Make sure, A and b are of correct size, since no checks are made for that!</para>
|
---|
| 237 | /// <para>This function is used by ILMath.linsolve. There should be rare need for you to call this function directly.</para>
|
---|
| 238 | /// <para>Elements of A below the main diagonal will not be accessed.</para>
|
---|
| 239 | /// <para>If A has been found to be singular, the array returned will contain NaN values for corresponding elements!</para></remarks>
|
---|
| 240 | internal static ILRetArray< double > solveUpperTriangularSystem (ILInArray< double > A, ILInArray< double > B, ref int singularityDetect) {
|
---|
| 241 | System.Diagnostics.Debug.Assert(B.Size[1] >= 0);
|
---|
| 242 | System.Diagnostics.Debug.Assert(B.Size[0] == A.Size[1]);
|
---|
| 243 | System.Diagnostics.Debug.Assert(A.Size[0] == A.Size[1]);
|
---|
| 244 | using (ILScope.Enter(A, B)) {
|
---|
| 245 | singularityDetect = -1;
|
---|
| 246 | int n = A.Size[0];
|
---|
| 247 | int m = B.Size[1];
|
---|
| 248 | int info = 0;
|
---|
| 249 | ILArray< double> ret = B.C;
|
---|
| 250 |
|
---|
| 251 | double[] retArr = ret.GetArrayForWrite();
|
---|
| 252 | // solve using Lapack
|
---|
| 253 | unsafe {
|
---|
| 254 | fixed ( double* ptrA = A.GetArrayForRead())
|
---|
| 255 | fixed ( double* ptrB = ret.GetArrayForWrite()) {
|
---|
| 256 | /*!HC:lapack.?trtrs*/
|
---|
| 257 | Lapack.dtrtrs('U', 'N', 'N', A.Size[0], B.Size[1], (IntPtr)ptrA, A.Size[0], (IntPtr)ptrB, B.Size[0], ref info);
|
---|
| 258 | }
|
---|
| 259 | }
|
---|
| 260 | if (info < 0)
|
---|
| 261 | throw new ILArgumentException("error inside Lapack function ?trtrs for argument: " + (-info));
|
---|
| 262 | if (info > 0) {
|
---|
| 263 | singularityDetect = info - 1;
|
---|
| 264 | for (m = 0; m < ret.Size[1]; m++) {
|
---|
| 265 | info = m * n + singularityDetect;
|
---|
| 266 | for (int i = singularityDetect; i < n; i++) {
|
---|
| 267 | retArr[info++] = double.NaN;
|
---|
| 268 | }
|
---|
| 269 | }
|
---|
| 270 | } else {
|
---|
| 271 | singularityDetect = -1;
|
---|
| 272 | }
|
---|
| 273 | return ret;
|
---|
| 274 | }
|
---|
| 275 | }
|
---|
| 276 |
|
---|
| 277 | /// <summary>
|
---|
| 278 | /// Solve system of linear equations A*x = b, with A being a lower triangular matrix
|
---|
| 279 | /// </summary>
|
---|
| 280 | /// <param name="A">Input matrix of size [n x n], must be lower triangular. No check is made for that!</param>
|
---|
| 281 | /// <param name="B">Solution vector. Size [n x 1]</param>
|
---|
| 282 | /// <param name="singularityDetect">[Output] This value gives the row of A, where a singularity has been detected (if any). If A is not singular, this will be a negative value.</param>
|
---|
| 283 | /// <returns>Solution x solving A * x = b.</returns>
|
---|
| 284 | /// <remarks><para>The solution will be determined via forward substitution</para>
|
---|
| 285 | /// <para>Make sure, A and b are of correct size, since no checks are made for that!</para>
|
---|
| 286 | /// <para>This function is used by ILMath.linsolve. There should be rare need for you to call this function directly.</para>
|
---|
| 287 | /// <para>Elements of A above the main diagonal will not be accessed.</para>
|
---|
| 288 | /// <para>If A has been found to be singular, the array returned will contain NaN values for corresponding elements!</para></remarks>
|
---|
| 289 | internal static ILRetArray< double> solveLowerTriangularSystem( ILInArray< double> A, ILInArray< double> B, ref int singularityDetect ) {
|
---|
| 290 | System.Diagnostics.Debug.Assert(B.Size[1] >= 0);
|
---|
| 291 | System.Diagnostics.Debug.Assert(B.Size[0] == A.Size[1]);
|
---|
| 292 | System.Diagnostics.Debug.Assert(A.Size[0] == A.Size[1]);
|
---|
| 293 | using (ILScope.Enter(A, B)) {
|
---|
| 294 | singularityDetect = -1;
|
---|
| 295 | int n = A.Size[0];
|
---|
| 296 | int m = B.Size[1];
|
---|
| 297 | int info = 0;
|
---|
| 298 | ILArray< double> ret = B.C;
|
---|
| 299 |
|
---|
| 300 | double[] retArr = ret.GetArrayForWrite();
|
---|
| 301 | // solve using Lapack
|
---|
| 302 | unsafe {
|
---|
| 303 | fixed ( double* ptrA = A.GetArrayForRead())
|
---|
| 304 | fixed ( double* ptrB = ret.GetArrayForWrite()) {
|
---|
| 305 | /*!HC:lapack.?trtrs*/
|
---|
| 306 | Lapack.dtrtrs('L', 'N', 'N', A.Size[0], B.Size[1], (IntPtr)ptrA, A.Size[0], (IntPtr)ptrB, B.Size[0], ref info);
|
---|
| 307 | }
|
---|
| 308 | }
|
---|
| 309 | if (info < 0)
|
---|
| 310 | throw new ILArgumentException("error inside Lapack function ?trtrs for argument: " + (-info));
|
---|
| 311 | if (info > 0) {
|
---|
| 312 | singularityDetect = info - 1;
|
---|
| 313 | for (m = 0; m < ret.Size[1]; m++) {
|
---|
| 314 | info = m * n + singularityDetect;
|
---|
| 315 | for (int i = singularityDetect; i < n; i++) {
|
---|
| 316 | retArr[info++] = double.NaN;
|
---|
| 317 | }
|
---|
| 318 | }
|
---|
| 319 | } else {
|
---|
| 320 | singularityDetect = -1;
|
---|
| 321 | }
|
---|
| 322 | return ret;
|
---|
| 323 | }
|
---|
| 324 | }
|
---|
| 325 |
|
---|
| 326 | |
---|
| 327 | #region HYCALPER AUTO GENERATED CODE
|
---|
| 328 | |
---|
| 329 |
|
---|
| 330 | /// <summary>
|
---|
| 331 | /// Solve linear equation system
|
---|
| 332 | /// </summary>
|
---|
| 333 | /// <param name="A">Matrix A. Size [n x q]</param>
|
---|
| 334 | /// <param name="B">Right hand side B. Size [n x m]</param>
|
---|
| 335 | /// <returns>Solution x solving the equation system: multiply(A, x) = B. Size [n x m]</returns>
|
---|
| 336 | /// <remarks><para>Depending on the structure and properties of A, the equation system will be solved in different ways:
|
---|
| 337 | /// <list type="bullet">
|
---|
| 338 | /// <item>If A is square (q == n) and an <b>upper or lower triangular</b> matrix, the system will directly be solved via backward- or forward substitution. Therefore the Lapack function ?trtrs will be used, whenever the memory layout of A is suitable. This may be the case even for reference ILArray's!
|
---|
| 339 | /// <example><code><![CDATA[ILArray<double> A = ILMath.randn(4); // construct 4 x 4 matrix
|
---|
| 340 | /// A = A.T; // A is a reference array now! The transpose is fast and does not consume much memory
|
---|
| 341 | /// // now construct a right side and solve the equations:
|
---|
| 342 | /// ILArray<double> B = new ILArray<double> (1.0,2.0,3.0).T;
|
---|
| 343 | /// ILMath.linsolve(A,B); // ... will be carried out via Lapack, even for all arrays involved being reference arrays! ]]></code></example></item>
|
---|
| 344 | /// <item><para>if A is square and symmetric or hermitian, A will be decomposed into a triangular equation system using cholesky factorization and Lapack. The system is than solved using performant Lapack routines.</para>
|
---|
| 345 | /// <para>If during the cholesky factorization A was found to be <b>not positive definite</b> - the cholesky factorization is canceled. </para></item>
|
---|
| 346 | /// <item>otherwise, if A is square only, it will be decomposed into upper and lower triangular martices using LU decomposition and Lapack. The triangular system is than solved using performant Lapack routines.</item>
|
---|
| 347 | /// <item>otherwise, if A is of size [q x n] and q != n, the system is solved using QR decomposition. A may be rank deficient. The solution is computed by use of the Lapack routine '?gelsy' and may be a reference array.</item>
|
---|
| 348 | /// </list></para>
|
---|
| 349 | /// <para>Compatibility with Matlab<sup>(R)</sup>: If A is square, the algorithm used follows the same logic as Matlab up to Rel 14, with the exception of Hessenberg matrices wich are solved via LU factorization here. The un-squared case is handled differently. A direct Lapack driver function (?gelsy) is used here. Therefore the solutions might differ! However, the solution will of course fullfill the equation A * x = B without round off errrors. </para>
|
---|
| 350 | /// <para>For specifiying the rank of A in the unsquare case (q != n), <see cref="ILNumerics.ILMath.eps"/> is used.</para></remarks>
|
---|
| 351 | public static ILRetArray< float > linsolve(ILInArray< float > A, ILInArray< float > B) {
|
---|
| 352 | if (object.Equals(A,null) || object.Equals(B,null))
|
---|
| 353 | throw new ILArgumentException("parameter must not be null!");
|
---|
| 354 | using (ILScope.Enter(A, B)) {
|
---|
| 355 | MatrixProperties props = MatrixProperties.None;
|
---|
| 356 | if (A.Size[0] == A.Size[1]) {
|
---|
| 357 | props |= MatrixProperties.Square;
|
---|
| 358 | if (ILMath.istriup(A)) {
|
---|
| 359 | props |= MatrixProperties.UpperTriangular;
|
---|
| 360 | return linsolve(A, B, ref props);
|
---|
| 361 | }
|
---|
| 362 | if (ILMath.istrilow(A)) {
|
---|
| 363 | props |= MatrixProperties.LowerTriangular;
|
---|
| 364 | return linsolve(A, B, ref props);
|
---|
| 365 | }
|
---|
| 366 | if (ILMath.ishermitian(A)) {
|
---|
| 367 | // give cholesky a try
|
---|
| 368 | props |= MatrixProperties.Hermitian;
|
---|
| 369 | props |= MatrixProperties.PositivDefinite;
|
---|
| 370 | ILArray< float> ret = linsolve(A, B, ref props);
|
---|
| 371 | if (!object.Equals(ret, null)) {
|
---|
| 372 | return ret;
|
---|
| 373 | } else {
|
---|
| 374 | props ^= MatrixProperties.PositivDefinite;
|
---|
| 375 | }
|
---|
| 376 | }
|
---|
| 377 | }
|
---|
| 378 | return linsolve(A, B, ref props);
|
---|
| 379 | }
|
---|
| 380 | }
|
---|
| 381 |
|
---|
| 382 | /// <summary>
|
---|
| 383 | /// Solve linear equation system
|
---|
| 384 | /// </summary>
|
---|
| 385 | /// <param name="A">Matrix A. Size [n x q]</param>
|
---|
| 386 | /// <param name="B">Right hand side B. Size [n x m]</param>
|
---|
| 387 | /// <param name="props">Matrix properties. If defined, no checks are made for the structure of A. If the
|
---|
| 388 | /// matrix A was found to be (close to or) singular, the 'MatrixProperties.Singular' flag in props will be set.
|
---|
| 389 | /// This flag should be tested on return, in order to verify the reliability of the solution.</param>
|
---|
| 390 | /// <returns>The solution x solving multiply(A,x) = B. Size [n x m]</returns>
|
---|
| 391 | /// <remarks><para>Depending on the <paramref name="props"/> parameter the equation system will be solved
|
---|
| 392 | /// differently for special structures of A:
|
---|
| 393 | /// <list type="bullet">
|
---|
| 394 | /// <item>If A is square (q == n) and an <b>upper or lower triangular</b> matrix, the system will directly
|
---|
| 395 | /// be solved via backward- or forward substitution. Therefore the Lapack function ?trtrs will be used,
|
---|
| 396 | /// whenever the memory layout of A is suitable. This may be the case even for reference ILArray's!
|
---|
| 397 | /// <example><code><![CDATA[ILArray<double> A = ILMath.randn(4); // construct 4 x 4 matrix
|
---|
| 398 | /// A = A.T; // A is a reference array now! The transpose is fast and does not consume much memory
|
---|
| 399 | /// // now construct a right side and solve the equations:
|
---|
| 400 | /// ILArray<double> B = new ILArray<double> (1.0,2.0,3.0).T;
|
---|
| 401 | /// ILMath.linsolve(A,B); // ... will be carried out via Lapack, even for all arrays involved being reference arrays! ]]></code></example></item>
|
---|
| 402 | /// <item><para>If A is square and symmetric or hermitian, A will be decomposed into a triangular equation
|
---|
| 403 | /// system using cholesky factorization and Lapack. The system is than solved using performant Lapack routines.</para>
|
---|
| 404 | /// <para>If during the cholesky factorization A was found to be <b>not positive definite</b> - the
|
---|
| 405 | /// corresponding flag in props will be cleaned and <c>null</c> will be returned.</para></item>
|
---|
| 406 | /// <item>Otherwise if A is square only, it will be decomposed into upper and lower triangular matrices
|
---|
| 407 | /// using LU decomposition and Lapack. The triangular system is than solved using performant Lapack routines.</item>
|
---|
| 408 | /// <item>Otherwise, if A is of size [q x n] and q != n, the system is solved using QR decomposition.
|
---|
| 409 | /// A may be rank deficient. The solution is computed by use of the Lapack routine '?gelsy' and may be a
|
---|
| 410 | /// reference array.</item>
|
---|
| 411 | /// </list></para>
|
---|
| 412 | /// <para>Compatibility with Matlab<sup>(R)</sup>: If A is square, the algorithm used follows the same
|
---|
| 413 | /// logic as Matlab up to Rel 14, with the exception of Hessenberg matrices wich are solved via LU
|
---|
| 414 | /// factorization here. The un-squared case is handled differently. A direct Lapack driver function
|
---|
| 415 | /// (?gelsy) is used here. Therefore the solutions might differ! However, the solution will of course
|
---|
| 416 | /// fullfill the equation A * x = B without round off errrors. </para>
|
---|
| 417 | /// <para>For specifiying the rank of A in the unsquare case (q != n),
|
---|
| 418 | /// <see cref="ILNumerics.ILMath.eps"/> is used.</para></remarks>
|
---|
| 419 | public static ILRetArray< float > linsolve(ILInArray< float > A, ILInArray< float > B, ref MatrixProperties props) {
|
---|
| 420 | if (object.Equals(A,null))
|
---|
| 421 | throw new ILArgumentException("input argument A must not be null!");
|
---|
| 422 | if (object.Equals(B,null))
|
---|
| 423 | throw new ILArgumentException("input argument B must not be null!");
|
---|
| 424 | using (ILScope.Enter(A, B)) {
|
---|
| 425 | if (A.IsEmpty || B.IsEmpty)
|
---|
| 426 | return empty< float>(A.Size);
|
---|
| 427 | if (A.Size[0] != B.Size[0])
|
---|
| 428 | throw new ILArgumentException("number of rows for matrix A must match number of rows for RHS!");
|
---|
| 429 | int info = 0, m = A.Size[0];
|
---|
| 430 | ILArray< float> ret = empty<float>(ILSize.Empty00);
|
---|
| 431 | if (m == A.Size[1]) {
|
---|
| 432 | props |= MatrixProperties.Square;
|
---|
| 433 | if ((props & MatrixProperties.LowerTriangular) != 0) {
|
---|
| 434 | ret.a = solveLowerTriangularSystem(A, B, ref info);
|
---|
| 435 | if (info > 0)
|
---|
| 436 | props |= MatrixProperties.Singular;
|
---|
| 437 | return ret;
|
---|
| 438 | }
|
---|
| 439 | if ((props & MatrixProperties.UpperTriangular) != 0) {
|
---|
| 440 | ret.a = solveUpperTriangularSystem(A, B, ref info);
|
---|
| 441 | if (info > 0)
|
---|
| 442 | props |= MatrixProperties.Singular;
|
---|
| 443 | return ret;
|
---|
| 444 | }
|
---|
| 445 | if ((props & MatrixProperties.Hermitian) != 0) {
|
---|
| 446 | ILDenseStorage< float> cholFact = A.Storage.copyUpperTriangle(m);
|
---|
| 447 |
|
---|
| 448 | Lapack.spotrf('U', m, cholFact.GetArrayForWrite(), m, ref info);
|
---|
| 449 | if (info > 0) {
|
---|
| 450 | props ^= MatrixProperties.Hermitian;
|
---|
| 451 | cholFact.Dispose();
|
---|
| 452 | return null;
|
---|
| 453 | } else {
|
---|
| 454 | // solve
|
---|
| 455 | ret.a = B.C;
|
---|
| 456 |
|
---|
| 457 | Lapack.spotrs('U', m, B.Size[1], cholFact.GetArrayForWrite(), m, ret.GetArrayForWrite(), m, ref info);
|
---|
| 458 | cholFact.Dispose();
|
---|
| 459 | return ret;
|
---|
| 460 | }
|
---|
| 461 | } else {
|
---|
| 462 | // attempt complete (expensive) LU factorization
|
---|
| 463 | ILArray< float> L = A.C;
|
---|
| 464 | int[] pivInd = ILMemoryPool.Pool.New<int>(m);
|
---|
| 465 |
|
---|
| 466 | Lapack.sgetrf(m, m, L.GetArrayForWrite(), m, pivInd, ref info);
|
---|
| 467 | if (info > 0)
|
---|
| 468 | props |= MatrixProperties.Singular;
|
---|
| 469 | ret.a = B.C;
|
---|
| 470 |
|
---|
| 471 | Lapack.sgetrs('N', m, B.Size[1], L.GetArrayForWrite(), m, pivInd, ret.GetArrayForWrite(), m, ref info);
|
---|
| 472 | if (info < 0)
|
---|
| 473 | throw new ILArgumentException("failed to solve via lapack dgetrs");
|
---|
| 474 | return ret;
|
---|
| 475 | }
|
---|
| 476 | } else {
|
---|
| 477 | // under- / overdetermined system
|
---|
| 478 | int n = A.Size[1], rank = 0, minMN = (m < n) ? m : n, maxMN = (m > n) ? m : n;
|
---|
| 479 | int nrhs = B.Size[1];
|
---|
| 480 | if (B.Size[0] != m)
|
---|
| 481 | throw new ILArgumentException("right hand side matrix B must match input A!");
|
---|
| 482 | ILArray< float> tmpA = A.C;
|
---|
| 483 | if (m < n) {
|
---|
| 484 | ret.a = zeros< float>(n, nrhs);
|
---|
| 485 | ret[r(0, m - 1), full] = B.C;
|
---|
| 486 | } else {
|
---|
| 487 | ret.a = B.C;
|
---|
| 488 | }
|
---|
| 489 | int[] JPVT = new int[n];
|
---|
| 490 |
|
---|
| 491 | Lapack.sgelsy(m, n, B.Size[1], tmpA.GetArrayForWrite(), m, ret.GetArrayForWrite(),
|
---|
| 492 | maxMN, JPVT, ILMath.MachineParameterSingle.eps,
|
---|
| 493 | ref rank, ref info);
|
---|
| 494 | if (n < m) {
|
---|
| 495 | ret.a = ret[r(0, n - 1), full];
|
---|
| 496 | }
|
---|
| 497 | if (rank < minMN)
|
---|
| 498 | props |= MatrixProperties.RankDeficient;
|
---|
| 499 | return ret;
|
---|
| 500 | }
|
---|
| 501 | }
|
---|
| 502 | }
|
---|
| 503 |
|
---|
| 504 | /// <summary>
|
---|
| 505 | /// Solve system of linear equations A*x = b, with A being a upper triangular matrix
|
---|
| 506 | /// </summary>
|
---|
| 507 | /// <param name="A">Input matrix of size [n x n], must be upper triangular. No check is made for that!</param>
|
---|
| 508 | /// <param name="B">Solution vector or matrix. Size [n x m]</param>
|
---|
| 509 | /// <param name="singularityDetect">[Output] This value gives the row of A, where a singularity has been detected (if any). If A is not singular, this will be a negative value.</param>
|
---|
| 510 | /// <returns>Solution x solving A * x = b.</returns>
|
---|
| 511 | /// <remarks><para>The solution will be determined via backward substitution</para>
|
---|
| 512 | /// <para>Make sure, A and b are of correct size, since no checks are made for that!</para>
|
---|
| 513 | /// <para>This function is used by ILMath.linsolve. There should be rare need for you to call this function directly.</para>
|
---|
| 514 | /// <para>Elements of A below the main diagonal will not be accessed.</para>
|
---|
| 515 | /// <para>If A has been found to be singular, the array returned will contain NaN values for corresponding elements!</para></remarks>
|
---|
| 516 | internal static ILRetArray< float > solveUpperTriangularSystem (ILInArray< float > A, ILInArray< float > B, ref int singularityDetect) {
|
---|
| 517 | System.Diagnostics.Debug.Assert(B.Size[1] >= 0);
|
---|
| 518 | System.Diagnostics.Debug.Assert(B.Size[0] == A.Size[1]);
|
---|
| 519 | System.Diagnostics.Debug.Assert(A.Size[0] == A.Size[1]);
|
---|
| 520 | using (ILScope.Enter(A, B)) {
|
---|
| 521 | singularityDetect = -1;
|
---|
| 522 | int n = A.Size[0];
|
---|
| 523 | int m = B.Size[1];
|
---|
| 524 | int info = 0;
|
---|
| 525 | ILArray< float> ret = B.C;
|
---|
| 526 |
|
---|
| 527 | float[] retArr = ret.GetArrayForWrite();
|
---|
| 528 | // solve using Lapack
|
---|
| 529 | unsafe {
|
---|
| 530 | fixed ( float* ptrA = A.GetArrayForRead())
|
---|
| 531 | fixed ( float* ptrB = ret.GetArrayForWrite()) {
|
---|
| 532 |
|
---|
| 533 | Lapack.strtrs('U', 'N', 'N', A.Size[0], B.Size[1], (IntPtr)ptrA, A.Size[0], (IntPtr)ptrB, B.Size[0], ref info);
|
---|
| 534 | }
|
---|
| 535 | }
|
---|
| 536 | if (info < 0)
|
---|
| 537 | throw new ILArgumentException("error inside Lapack function ?trtrs for argument: " + (-info));
|
---|
| 538 | if (info > 0) {
|
---|
| 539 | singularityDetect = info - 1;
|
---|
| 540 | for (m = 0; m < ret.Size[1]; m++) {
|
---|
| 541 | info = m * n + singularityDetect;
|
---|
| 542 | for (int i = singularityDetect; i < n; i++) {
|
---|
| 543 | retArr[info++] = float.NaN;
|
---|
| 544 | }
|
---|
| 545 | }
|
---|
| 546 | } else {
|
---|
| 547 | singularityDetect = -1;
|
---|
| 548 | }
|
---|
| 549 | return ret;
|
---|
| 550 | }
|
---|
| 551 | }
|
---|
| 552 |
|
---|
| 553 | /// <summary>
|
---|
| 554 | /// Solve system of linear equations A*x = b, with A being a lower triangular matrix
|
---|
| 555 | /// </summary>
|
---|
| 556 | /// <param name="A">Input matrix of size [n x n], must be lower triangular. No check is made for that!</param>
|
---|
| 557 | /// <param name="B">Solution vector. Size [n x 1]</param>
|
---|
| 558 | /// <param name="singularityDetect">[Output] This value gives the row of A, where a singularity has been detected (if any). If A is not singular, this will be a negative value.</param>
|
---|
| 559 | /// <returns>Solution x solving A * x = b.</returns>
|
---|
| 560 | /// <remarks><para>The solution will be determined via forward substitution</para>
|
---|
| 561 | /// <para>Make sure, A and b are of correct size, since no checks are made for that!</para>
|
---|
| 562 | /// <para>This function is used by ILMath.linsolve. There should be rare need for you to call this function directly.</para>
|
---|
| 563 | /// <para>Elements of A above the main diagonal will not be accessed.</para>
|
---|
| 564 | /// <para>If A has been found to be singular, the array returned will contain NaN values for corresponding elements!</para></remarks>
|
---|
| 565 | internal static ILRetArray< float> solveLowerTriangularSystem( ILInArray< float> A, ILInArray< float> B, ref int singularityDetect ) {
|
---|
| 566 | System.Diagnostics.Debug.Assert(B.Size[1] >= 0);
|
---|
| 567 | System.Diagnostics.Debug.Assert(B.Size[0] == A.Size[1]);
|
---|
| 568 | System.Diagnostics.Debug.Assert(A.Size[0] == A.Size[1]);
|
---|
| 569 | using (ILScope.Enter(A, B)) {
|
---|
| 570 | singularityDetect = -1;
|
---|
| 571 | int n = A.Size[0];
|
---|
| 572 | int m = B.Size[1];
|
---|
| 573 | int info = 0;
|
---|
| 574 | ILArray< float> ret = B.C;
|
---|
| 575 |
|
---|
| 576 | float[] retArr = ret.GetArrayForWrite();
|
---|
| 577 | // solve using Lapack
|
---|
| 578 | unsafe {
|
---|
| 579 | fixed ( float* ptrA = A.GetArrayForRead())
|
---|
| 580 | fixed ( float* ptrB = ret.GetArrayForWrite()) {
|
---|
| 581 |
|
---|
| 582 | Lapack.strtrs('L', 'N', 'N', A.Size[0], B.Size[1], (IntPtr)ptrA, A.Size[0], (IntPtr)ptrB, B.Size[0], ref info);
|
---|
| 583 | }
|
---|
| 584 | }
|
---|
| 585 | if (info < 0)
|
---|
| 586 | throw new ILArgumentException("error inside Lapack function ?trtrs for argument: " + (-info));
|
---|
| 587 | if (info > 0) {
|
---|
| 588 | singularityDetect = info - 1;
|
---|
| 589 | for (m = 0; m < ret.Size[1]; m++) {
|
---|
| 590 | info = m * n + singularityDetect;
|
---|
| 591 | for (int i = singularityDetect; i < n; i++) {
|
---|
| 592 | retArr[info++] = float.NaN;
|
---|
| 593 | }
|
---|
| 594 | }
|
---|
| 595 | } else {
|
---|
| 596 | singularityDetect = -1;
|
---|
| 597 | }
|
---|
| 598 | return ret;
|
---|
| 599 | }
|
---|
| 600 | }
|
---|
| 601 |
|
---|
| 602 |
|
---|
| 603 | /// <summary>
|
---|
| 604 | /// Solve linear equation system
|
---|
| 605 | /// </summary>
|
---|
| 606 | /// <param name="A">Matrix A. Size [n x q]</param>
|
---|
| 607 | /// <param name="B">Right hand side B. Size [n x m]</param>
|
---|
| 608 | /// <returns>Solution x solving the equation system: multiply(A, x) = B. Size [n x m]</returns>
|
---|
| 609 | /// <remarks><para>Depending on the structure and properties of A, the equation system will be solved in different ways:
|
---|
| 610 | /// <list type="bullet">
|
---|
| 611 | /// <item>If A is square (q == n) and an <b>upper or lower triangular</b> matrix, the system will directly be solved via backward- or forward substitution. Therefore the Lapack function ?trtrs will be used, whenever the memory layout of A is suitable. This may be the case even for reference ILArray's!
|
---|
| 612 | /// <example><code><![CDATA[ILArray<double> A = ILMath.randn(4); // construct 4 x 4 matrix
|
---|
| 613 | /// A = A.T; // A is a reference array now! The transpose is fast and does not consume much memory
|
---|
| 614 | /// // now construct a right side and solve the equations:
|
---|
| 615 | /// ILArray<double> B = new ILArray<double> (1.0,2.0,3.0).T;
|
---|
| 616 | /// ILMath.linsolve(A,B); // ... will be carried out via Lapack, even for all arrays involved being reference arrays! ]]></code></example></item>
|
---|
| 617 | /// <item><para>if A is square and symmetric or hermitian, A will be decomposed into a triangular equation system using cholesky factorization and Lapack. The system is than solved using performant Lapack routines.</para>
|
---|
| 618 | /// <para>If during the cholesky factorization A was found to be <b>not positive definite</b> - the cholesky factorization is canceled. </para></item>
|
---|
| 619 | /// <item>otherwise, if A is square only, it will be decomposed into upper and lower triangular martices using LU decomposition and Lapack. The triangular system is than solved using performant Lapack routines.</item>
|
---|
| 620 | /// <item>otherwise, if A is of size [q x n] and q != n, the system is solved using QR decomposition. A may be rank deficient. The solution is computed by use of the Lapack routine '?gelsy' and may be a reference array.</item>
|
---|
| 621 | /// </list></para>
|
---|
| 622 | /// <para>Compatibility with Matlab<sup>(R)</sup>: If A is square, the algorithm used follows the same logic as Matlab up to Rel 14, with the exception of Hessenberg matrices wich are solved via LU factorization here. The un-squared case is handled differently. A direct Lapack driver function (?gelsy) is used here. Therefore the solutions might differ! However, the solution will of course fullfill the equation A * x = B without round off errrors. </para>
|
---|
| 623 | /// <para>For specifiying the rank of A in the unsquare case (q != n), <see cref="ILNumerics.ILMath.eps"/> is used.</para></remarks>
|
---|
| 624 | public static ILRetArray< fcomplex > linsolve(ILInArray< fcomplex > A, ILInArray< fcomplex > B) {
|
---|
| 625 | if (object.Equals(A,null) || object.Equals(B,null))
|
---|
| 626 | throw new ILArgumentException("parameter must not be null!");
|
---|
| 627 | using (ILScope.Enter(A, B)) {
|
---|
| 628 | MatrixProperties props = MatrixProperties.None;
|
---|
| 629 | if (A.Size[0] == A.Size[1]) {
|
---|
| 630 | props |= MatrixProperties.Square;
|
---|
| 631 | if (ILMath.istriup(A)) {
|
---|
| 632 | props |= MatrixProperties.UpperTriangular;
|
---|
| 633 | return linsolve(A, B, ref props);
|
---|
| 634 | }
|
---|
| 635 | if (ILMath.istrilow(A)) {
|
---|
| 636 | props |= MatrixProperties.LowerTriangular;
|
---|
| 637 | return linsolve(A, B, ref props);
|
---|
| 638 | }
|
---|
| 639 | if (ILMath.ishermitian(A)) {
|
---|
| 640 | // give cholesky a try
|
---|
| 641 | props |= MatrixProperties.Hermitian;
|
---|
| 642 | props |= MatrixProperties.PositivDefinite;
|
---|
| 643 | ILArray< fcomplex> ret = linsolve(A, B, ref props);
|
---|
| 644 | if (!object.Equals(ret, null)) {
|
---|
| 645 | return ret;
|
---|
| 646 | } else {
|
---|
| 647 | props ^= MatrixProperties.PositivDefinite;
|
---|
| 648 | }
|
---|
| 649 | }
|
---|
| 650 | }
|
---|
| 651 | return linsolve(A, B, ref props);
|
---|
| 652 | }
|
---|
| 653 | }
|
---|
| 654 |
|
---|
| 655 | /// <summary>
|
---|
| 656 | /// Solve linear equation system
|
---|
| 657 | /// </summary>
|
---|
| 658 | /// <param name="A">Matrix A. Size [n x q]</param>
|
---|
| 659 | /// <param name="B">Right hand side B. Size [n x m]</param>
|
---|
| 660 | /// <param name="props">Matrix properties. If defined, no checks are made for the structure of A. If the
|
---|
| 661 | /// matrix A was found to be (close to or) singular, the 'MatrixProperties.Singular' flag in props will be set.
|
---|
| 662 | /// This flag should be tested on return, in order to verify the reliability of the solution.</param>
|
---|
| 663 | /// <returns>The solution x solving multiply(A,x) = B. Size [n x m]</returns>
|
---|
| 664 | /// <remarks><para>Depending on the <paramref name="props"/> parameter the equation system will be solved
|
---|
| 665 | /// differently for special structures of A:
|
---|
| 666 | /// <list type="bullet">
|
---|
| 667 | /// <item>If A is square (q == n) and an <b>upper or lower triangular</b> matrix, the system will directly
|
---|
| 668 | /// be solved via backward- or forward substitution. Therefore the Lapack function ?trtrs will be used,
|
---|
| 669 | /// whenever the memory layout of A is suitable. This may be the case even for reference ILArray's!
|
---|
| 670 | /// <example><code><![CDATA[ILArray<double> A = ILMath.randn(4); // construct 4 x 4 matrix
|
---|
| 671 | /// A = A.T; // A is a reference array now! The transpose is fast and does not consume much memory
|
---|
| 672 | /// // now construct a right side and solve the equations:
|
---|
| 673 | /// ILArray<double> B = new ILArray<double> (1.0,2.0,3.0).T;
|
---|
| 674 | /// ILMath.linsolve(A,B); // ... will be carried out via Lapack, even for all arrays involved being reference arrays! ]]></code></example></item>
|
---|
| 675 | /// <item><para>If A is square and symmetric or hermitian, A will be decomposed into a triangular equation
|
---|
| 676 | /// system using cholesky factorization and Lapack. The system is than solved using performant Lapack routines.</para>
|
---|
| 677 | /// <para>If during the cholesky factorization A was found to be <b>not positive definite</b> - the
|
---|
| 678 | /// corresponding flag in props will be cleaned and <c>null</c> will be returned.</para></item>
|
---|
| 679 | /// <item>Otherwise if A is square only, it will be decomposed into upper and lower triangular matrices
|
---|
| 680 | /// using LU decomposition and Lapack. The triangular system is than solved using performant Lapack routines.</item>
|
---|
| 681 | /// <item>Otherwise, if A is of size [q x n] and q != n, the system is solved using QR decomposition.
|
---|
| 682 | /// A may be rank deficient. The solution is computed by use of the Lapack routine '?gelsy' and may be a
|
---|
| 683 | /// reference array.</item>
|
---|
| 684 | /// </list></para>
|
---|
| 685 | /// <para>Compatibility with Matlab<sup>(R)</sup>: If A is square, the algorithm used follows the same
|
---|
| 686 | /// logic as Matlab up to Rel 14, with the exception of Hessenberg matrices wich are solved via LU
|
---|
| 687 | /// factorization here. The un-squared case is handled differently. A direct Lapack driver function
|
---|
| 688 | /// (?gelsy) is used here. Therefore the solutions might differ! However, the solution will of course
|
---|
| 689 | /// fullfill the equation A * x = B without round off errrors. </para>
|
---|
| 690 | /// <para>For specifiying the rank of A in the unsquare case (q != n),
|
---|
| 691 | /// <see cref="ILNumerics.ILMath.eps"/> is used.</para></remarks>
|
---|
| 692 | public static ILRetArray< fcomplex > linsolve(ILInArray< fcomplex > A, ILInArray< fcomplex > B, ref MatrixProperties props) {
|
---|
| 693 | if (object.Equals(A,null))
|
---|
| 694 | throw new ILArgumentException("input argument A must not be null!");
|
---|
| 695 | if (object.Equals(B,null))
|
---|
| 696 | throw new ILArgumentException("input argument B must not be null!");
|
---|
| 697 | using (ILScope.Enter(A, B)) {
|
---|
| 698 | if (A.IsEmpty || B.IsEmpty)
|
---|
| 699 | return empty< fcomplex>(A.Size);
|
---|
| 700 | if (A.Size[0] != B.Size[0])
|
---|
| 701 | throw new ILArgumentException("number of rows for matrix A must match number of rows for RHS!");
|
---|
| 702 | int info = 0, m = A.Size[0];
|
---|
| 703 | ILArray< fcomplex> ret = empty<fcomplex>(ILSize.Empty00);
|
---|
| 704 | if (m == A.Size[1]) {
|
---|
| 705 | props |= MatrixProperties.Square;
|
---|
| 706 | if ((props & MatrixProperties.LowerTriangular) != 0) {
|
---|
| 707 | ret.a = solveLowerTriangularSystem(A, B, ref info);
|
---|
| 708 | if (info > 0)
|
---|
| 709 | props |= MatrixProperties.Singular;
|
---|
| 710 | return ret;
|
---|
| 711 | }
|
---|
| 712 | if ((props & MatrixProperties.UpperTriangular) != 0) {
|
---|
| 713 | ret.a = solveUpperTriangularSystem(A, B, ref info);
|
---|
| 714 | if (info > 0)
|
---|
| 715 | props |= MatrixProperties.Singular;
|
---|
| 716 | return ret;
|
---|
| 717 | }
|
---|
| 718 | if ((props & MatrixProperties.Hermitian) != 0) {
|
---|
| 719 | ILDenseStorage< fcomplex> cholFact = A.Storage.copyUpperTriangle(m);
|
---|
| 720 |
|
---|
| 721 | Lapack.cpotrf('U', m, cholFact.GetArrayForWrite(), m, ref info);
|
---|
| 722 | if (info > 0) {
|
---|
| 723 | props ^= MatrixProperties.Hermitian;
|
---|
| 724 | cholFact.Dispose();
|
---|
| 725 | return null;
|
---|
| 726 | } else {
|
---|
| 727 | // solve
|
---|
| 728 | ret.a = B.C;
|
---|
| 729 |
|
---|
| 730 | Lapack.cpotrs('U', m, B.Size[1], cholFact.GetArrayForWrite(), m, ret.GetArrayForWrite(), m, ref info);
|
---|
| 731 | cholFact.Dispose();
|
---|
| 732 | return ret;
|
---|
| 733 | }
|
---|
| 734 | } else {
|
---|
| 735 | // attempt complete (expensive) LU factorization
|
---|
| 736 | ILArray< fcomplex> L = A.C;
|
---|
| 737 | int[] pivInd = ILMemoryPool.Pool.New<int>(m);
|
---|
| 738 |
|
---|
| 739 | Lapack.cgetrf(m, m, L.GetArrayForWrite(), m, pivInd, ref info);
|
---|
| 740 | if (info > 0)
|
---|
| 741 | props |= MatrixProperties.Singular;
|
---|
| 742 | ret.a = B.C;
|
---|
| 743 |
|
---|
| 744 | Lapack.cgetrs('N', m, B.Size[1], L.GetArrayForWrite(), m, pivInd, ret.GetArrayForWrite(), m, ref info);
|
---|
| 745 | if (info < 0)
|
---|
| 746 | throw new ILArgumentException("failed to solve via lapack dgetrs");
|
---|
| 747 | return ret;
|
---|
| 748 | }
|
---|
| 749 | } else {
|
---|
| 750 | // under- / overdetermined system
|
---|
| 751 | int n = A.Size[1], rank = 0, minMN = (m < n) ? m : n, maxMN = (m > n) ? m : n;
|
---|
| 752 | int nrhs = B.Size[1];
|
---|
| 753 | if (B.Size[0] != m)
|
---|
| 754 | throw new ILArgumentException("right hand side matrix B must match input A!");
|
---|
| 755 | ILArray< fcomplex> tmpA = A.C;
|
---|
| 756 | if (m < n) {
|
---|
| 757 | ret.a = zeros< fcomplex>(n, nrhs);
|
---|
| 758 | ret[r(0, m - 1), full] = B.C;
|
---|
| 759 | } else {
|
---|
| 760 | ret.a = B.C;
|
---|
| 761 | }
|
---|
| 762 | int[] JPVT = new int[n];
|
---|
| 763 |
|
---|
| 764 | Lapack.cgelsy(m, n, B.Size[1], tmpA.GetArrayForWrite(), m, ret.GetArrayForWrite(),
|
---|
| 765 | maxMN, JPVT, ILMath.MachineParameterSingle.eps,
|
---|
| 766 | ref rank, ref info);
|
---|
| 767 | if (n < m) {
|
---|
| 768 | ret.a = ret[r(0, n - 1), full];
|
---|
| 769 | }
|
---|
| 770 | if (rank < minMN)
|
---|
| 771 | props |= MatrixProperties.RankDeficient;
|
---|
| 772 | return ret;
|
---|
| 773 | }
|
---|
| 774 | }
|
---|
| 775 | }
|
---|
| 776 |
|
---|
| 777 | /// <summary>
|
---|
| 778 | /// Solve system of linear equations A*x = b, with A being a upper triangular matrix
|
---|
| 779 | /// </summary>
|
---|
| 780 | /// <param name="A">Input matrix of size [n x n], must be upper triangular. No check is made for that!</param>
|
---|
| 781 | /// <param name="B">Solution vector or matrix. Size [n x m]</param>
|
---|
| 782 | /// <param name="singularityDetect">[Output] This value gives the row of A, where a singularity has been detected (if any). If A is not singular, this will be a negative value.</param>
|
---|
| 783 | /// <returns>Solution x solving A * x = b.</returns>
|
---|
| 784 | /// <remarks><para>The solution will be determined via backward substitution</para>
|
---|
| 785 | /// <para>Make sure, A and b are of correct size, since no checks are made for that!</para>
|
---|
| 786 | /// <para>This function is used by ILMath.linsolve. There should be rare need for you to call this function directly.</para>
|
---|
| 787 | /// <para>Elements of A below the main diagonal will not be accessed.</para>
|
---|
| 788 | /// <para>If A has been found to be singular, the array returned will contain NaN values for corresponding elements!</para></remarks>
|
---|
| 789 | internal static ILRetArray< fcomplex > solveUpperTriangularSystem (ILInArray< fcomplex > A, ILInArray< fcomplex > B, ref int singularityDetect) {
|
---|
| 790 | System.Diagnostics.Debug.Assert(B.Size[1] >= 0);
|
---|
| 791 | System.Diagnostics.Debug.Assert(B.Size[0] == A.Size[1]);
|
---|
| 792 | System.Diagnostics.Debug.Assert(A.Size[0] == A.Size[1]);
|
---|
| 793 | using (ILScope.Enter(A, B)) {
|
---|
| 794 | singularityDetect = -1;
|
---|
| 795 | int n = A.Size[0];
|
---|
| 796 | int m = B.Size[1];
|
---|
| 797 | int info = 0;
|
---|
| 798 | ILArray< fcomplex> ret = B.C;
|
---|
| 799 |
|
---|
| 800 | fcomplex[] retArr = ret.GetArrayForWrite();
|
---|
| 801 | // solve using Lapack
|
---|
| 802 | unsafe {
|
---|
| 803 | fixed ( fcomplex* ptrA = A.GetArrayForRead())
|
---|
| 804 | fixed ( fcomplex* ptrB = ret.GetArrayForWrite()) {
|
---|
| 805 |
|
---|
| 806 | Lapack.ctrtrs('U', 'N', 'N', A.Size[0], B.Size[1], (IntPtr)ptrA, A.Size[0], (IntPtr)ptrB, B.Size[0], ref info);
|
---|
| 807 | }
|
---|
| 808 | }
|
---|
| 809 | if (info < 0)
|
---|
| 810 | throw new ILArgumentException("error inside Lapack function ?trtrs for argument: " + (-info));
|
---|
| 811 | if (info > 0) {
|
---|
| 812 | singularityDetect = info - 1;
|
---|
| 813 | for (m = 0; m < ret.Size[1]; m++) {
|
---|
| 814 | info = m * n + singularityDetect;
|
---|
| 815 | for (int i = singularityDetect; i < n; i++) {
|
---|
| 816 | retArr[info++] = new fcomplex(float.NaN,float.NaN);
|
---|
| 817 | }
|
---|
| 818 | }
|
---|
| 819 | } else {
|
---|
| 820 | singularityDetect = -1;
|
---|
| 821 | }
|
---|
| 822 | return ret;
|
---|
| 823 | }
|
---|
| 824 | }
|
---|
| 825 |
|
---|
| 826 | /// <summary>
|
---|
| 827 | /// Solve system of linear equations A*x = b, with A being a lower triangular matrix
|
---|
| 828 | /// </summary>
|
---|
| 829 | /// <param name="A">Input matrix of size [n x n], must be lower triangular. No check is made for that!</param>
|
---|
| 830 | /// <param name="B">Solution vector. Size [n x 1]</param>
|
---|
| 831 | /// <param name="singularityDetect">[Output] This value gives the row of A, where a singularity has been detected (if any). If A is not singular, this will be a negative value.</param>
|
---|
| 832 | /// <returns>Solution x solving A * x = b.</returns>
|
---|
| 833 | /// <remarks><para>The solution will be determined via forward substitution</para>
|
---|
| 834 | /// <para>Make sure, A and b are of correct size, since no checks are made for that!</para>
|
---|
| 835 | /// <para>This function is used by ILMath.linsolve. There should be rare need for you to call this function directly.</para>
|
---|
| 836 | /// <para>Elements of A above the main diagonal will not be accessed.</para>
|
---|
| 837 | /// <para>If A has been found to be singular, the array returned will contain NaN values for corresponding elements!</para></remarks>
|
---|
| 838 | internal static ILRetArray< fcomplex> solveLowerTriangularSystem( ILInArray< fcomplex> A, ILInArray< fcomplex> B, ref int singularityDetect ) {
|
---|
| 839 | System.Diagnostics.Debug.Assert(B.Size[1] >= 0);
|
---|
| 840 | System.Diagnostics.Debug.Assert(B.Size[0] == A.Size[1]);
|
---|
| 841 | System.Diagnostics.Debug.Assert(A.Size[0] == A.Size[1]);
|
---|
| 842 | using (ILScope.Enter(A, B)) {
|
---|
| 843 | singularityDetect = -1;
|
---|
| 844 | int n = A.Size[0];
|
---|
| 845 | int m = B.Size[1];
|
---|
| 846 | int info = 0;
|
---|
| 847 | ILArray< fcomplex> ret = B.C;
|
---|
| 848 |
|
---|
| 849 | fcomplex[] retArr = ret.GetArrayForWrite();
|
---|
| 850 | // solve using Lapack
|
---|
| 851 | unsafe {
|
---|
| 852 | fixed ( fcomplex* ptrA = A.GetArrayForRead())
|
---|
| 853 | fixed ( fcomplex* ptrB = ret.GetArrayForWrite()) {
|
---|
| 854 |
|
---|
| 855 | Lapack.ctrtrs('L', 'N', 'N', A.Size[0], B.Size[1], (IntPtr)ptrA, A.Size[0], (IntPtr)ptrB, B.Size[0], ref info);
|
---|
| 856 | }
|
---|
| 857 | }
|
---|
| 858 | if (info < 0)
|
---|
| 859 | throw new ILArgumentException("error inside Lapack function ?trtrs for argument: " + (-info));
|
---|
| 860 | if (info > 0) {
|
---|
| 861 | singularityDetect = info - 1;
|
---|
| 862 | for (m = 0; m < ret.Size[1]; m++) {
|
---|
| 863 | info = m * n + singularityDetect;
|
---|
| 864 | for (int i = singularityDetect; i < n; i++) {
|
---|
| 865 | retArr[info++] = new fcomplex(float.NaN,float.NaN);
|
---|
| 866 | }
|
---|
| 867 | }
|
---|
| 868 | } else {
|
---|
| 869 | singularityDetect = -1;
|
---|
| 870 | }
|
---|
| 871 | return ret;
|
---|
| 872 | }
|
---|
| 873 | }
|
---|
| 874 |
|
---|
| 875 |
|
---|
| 876 | /// <summary>
|
---|
| 877 | /// Solve linear equation system
|
---|
| 878 | /// </summary>
|
---|
| 879 | /// <param name="A">Matrix A. Size [n x q]</param>
|
---|
| 880 | /// <param name="B">Right hand side B. Size [n x m]</param>
|
---|
| 881 | /// <returns>Solution x solving the equation system: multiply(A, x) = B. Size [n x m]</returns>
|
---|
| 882 | /// <remarks><para>Depending on the structure and properties of A, the equation system will be solved in different ways:
|
---|
| 883 | /// <list type="bullet">
|
---|
| 884 | /// <item>If A is square (q == n) and an <b>upper or lower triangular</b> matrix, the system will directly be solved via backward- or forward substitution. Therefore the Lapack function ?trtrs will be used, whenever the memory layout of A is suitable. This may be the case even for reference ILArray's!
|
---|
| 885 | /// <example><code><![CDATA[ILArray<double> A = ILMath.randn(4); // construct 4 x 4 matrix
|
---|
| 886 | /// A = A.T; // A is a reference array now! The transpose is fast and does not consume much memory
|
---|
| 887 | /// // now construct a right side and solve the equations:
|
---|
| 888 | /// ILArray<double> B = new ILArray<double> (1.0,2.0,3.0).T;
|
---|
| 889 | /// ILMath.linsolve(A,B); // ... will be carried out via Lapack, even for all arrays involved being reference arrays! ]]></code></example></item>
|
---|
| 890 | /// <item><para>if A is square and symmetric or hermitian, A will be decomposed into a triangular equation system using cholesky factorization and Lapack. The system is than solved using performant Lapack routines.</para>
|
---|
| 891 | /// <para>If during the cholesky factorization A was found to be <b>not positive definite</b> - the cholesky factorization is canceled. </para></item>
|
---|
| 892 | /// <item>otherwise, if A is square only, it will be decomposed into upper and lower triangular martices using LU decomposition and Lapack. The triangular system is than solved using performant Lapack routines.</item>
|
---|
| 893 | /// <item>otherwise, if A is of size [q x n] and q != n, the system is solved using QR decomposition. A may be rank deficient. The solution is computed by use of the Lapack routine '?gelsy' and may be a reference array.</item>
|
---|
| 894 | /// </list></para>
|
---|
| 895 | /// <para>Compatibility with Matlab<sup>(R)</sup>: If A is square, the algorithm used follows the same logic as Matlab up to Rel 14, with the exception of Hessenberg matrices wich are solved via LU factorization here. The un-squared case is handled differently. A direct Lapack driver function (?gelsy) is used here. Therefore the solutions might differ! However, the solution will of course fullfill the equation A * x = B without round off errrors. </para>
|
---|
| 896 | /// <para>For specifiying the rank of A in the unsquare case (q != n), <see cref="ILNumerics.ILMath.eps"/> is used.</para></remarks>
|
---|
| 897 | public static ILRetArray< complex > linsolve(ILInArray< complex > A, ILInArray< complex > B) {
|
---|
| 898 | if (object.Equals(A,null) || object.Equals(B,null))
|
---|
| 899 | throw new ILArgumentException("parameter must not be null!");
|
---|
| 900 | using (ILScope.Enter(A, B)) {
|
---|
| 901 | MatrixProperties props = MatrixProperties.None;
|
---|
| 902 | if (A.Size[0] == A.Size[1]) {
|
---|
| 903 | props |= MatrixProperties.Square;
|
---|
| 904 | if (ILMath.istriup(A)) {
|
---|
| 905 | props |= MatrixProperties.UpperTriangular;
|
---|
| 906 | return linsolve(A, B, ref props);
|
---|
| 907 | }
|
---|
| 908 | if (ILMath.istrilow(A)) {
|
---|
| 909 | props |= MatrixProperties.LowerTriangular;
|
---|
| 910 | return linsolve(A, B, ref props);
|
---|
| 911 | }
|
---|
| 912 | if (ILMath.ishermitian(A)) {
|
---|
| 913 | // give cholesky a try
|
---|
| 914 | props |= MatrixProperties.Hermitian;
|
---|
| 915 | props |= MatrixProperties.PositivDefinite;
|
---|
| 916 | ILArray< complex> ret = linsolve(A, B, ref props);
|
---|
| 917 | if (!object.Equals(ret, null)) {
|
---|
| 918 | return ret;
|
---|
| 919 | } else {
|
---|
| 920 | props ^= MatrixProperties.PositivDefinite;
|
---|
| 921 | }
|
---|
| 922 | }
|
---|
| 923 | }
|
---|
| 924 | return linsolve(A, B, ref props);
|
---|
| 925 | }
|
---|
| 926 | }
|
---|
| 927 |
|
---|
| 928 | /// <summary>
|
---|
| 929 | /// Solve linear equation system
|
---|
| 930 | /// </summary>
|
---|
| 931 | /// <param name="A">Matrix A. Size [n x q]</param>
|
---|
| 932 | /// <param name="B">Right hand side B. Size [n x m]</param>
|
---|
| 933 | /// <param name="props">Matrix properties. If defined, no checks are made for the structure of A. If the
|
---|
| 934 | /// matrix A was found to be (close to or) singular, the 'MatrixProperties.Singular' flag in props will be set.
|
---|
| 935 | /// This flag should be tested on return, in order to verify the reliability of the solution.</param>
|
---|
| 936 | /// <returns>The solution x solving multiply(A,x) = B. Size [n x m]</returns>
|
---|
| 937 | /// <remarks><para>Depending on the <paramref name="props"/> parameter the equation system will be solved
|
---|
| 938 | /// differently for special structures of A:
|
---|
| 939 | /// <list type="bullet">
|
---|
| 940 | /// <item>If A is square (q == n) and an <b>upper or lower triangular</b> matrix, the system will directly
|
---|
| 941 | /// be solved via backward- or forward substitution. Therefore the Lapack function ?trtrs will be used,
|
---|
| 942 | /// whenever the memory layout of A is suitable. This may be the case even for reference ILArray's!
|
---|
| 943 | /// <example><code><![CDATA[ILArray<double> A = ILMath.randn(4); // construct 4 x 4 matrix
|
---|
| 944 | /// A = A.T; // A is a reference array now! The transpose is fast and does not consume much memory
|
---|
| 945 | /// // now construct a right side and solve the equations:
|
---|
| 946 | /// ILArray<double> B = new ILArray<double> (1.0,2.0,3.0).T;
|
---|
| 947 | /// ILMath.linsolve(A,B); // ... will be carried out via Lapack, even for all arrays involved being reference arrays! ]]></code></example></item>
|
---|
| 948 | /// <item><para>If A is square and symmetric or hermitian, A will be decomposed into a triangular equation
|
---|
| 949 | /// system using cholesky factorization and Lapack. The system is than solved using performant Lapack routines.</para>
|
---|
| 950 | /// <para>If during the cholesky factorization A was found to be <b>not positive definite</b> - the
|
---|
| 951 | /// corresponding flag in props will be cleaned and <c>null</c> will be returned.</para></item>
|
---|
| 952 | /// <item>Otherwise if A is square only, it will be decomposed into upper and lower triangular matrices
|
---|
| 953 | /// using LU decomposition and Lapack. The triangular system is than solved using performant Lapack routines.</item>
|
---|
| 954 | /// <item>Otherwise, if A is of size [q x n] and q != n, the system is solved using QR decomposition.
|
---|
| 955 | /// A may be rank deficient. The solution is computed by use of the Lapack routine '?gelsy' and may be a
|
---|
| 956 | /// reference array.</item>
|
---|
| 957 | /// </list></para>
|
---|
| 958 | /// <para>Compatibility with Matlab<sup>(R)</sup>: If A is square, the algorithm used follows the same
|
---|
| 959 | /// logic as Matlab up to Rel 14, with the exception of Hessenberg matrices wich are solved via LU
|
---|
| 960 | /// factorization here. The un-squared case is handled differently. A direct Lapack driver function
|
---|
| 961 | /// (?gelsy) is used here. Therefore the solutions might differ! However, the solution will of course
|
---|
| 962 | /// fullfill the equation A * x = B without round off errrors. </para>
|
---|
| 963 | /// <para>For specifiying the rank of A in the unsquare case (q != n),
|
---|
| 964 | /// <see cref="ILNumerics.ILMath.eps"/> is used.</para></remarks>
|
---|
| 965 | public static ILRetArray< complex > linsolve(ILInArray< complex > A, ILInArray< complex > B, ref MatrixProperties props) {
|
---|
| 966 | if (object.Equals(A,null))
|
---|
| 967 | throw new ILArgumentException("input argument A must not be null!");
|
---|
| 968 | if (object.Equals(B,null))
|
---|
| 969 | throw new ILArgumentException("input argument B must not be null!");
|
---|
| 970 | using (ILScope.Enter(A, B)) {
|
---|
| 971 | if (A.IsEmpty || B.IsEmpty)
|
---|
| 972 | return empty< complex>(A.Size);
|
---|
| 973 | if (A.Size[0] != B.Size[0])
|
---|
| 974 | throw new ILArgumentException("number of rows for matrix A must match number of rows for RHS!");
|
---|
| 975 | int info = 0, m = A.Size[0];
|
---|
| 976 | ILArray< complex> ret = empty<complex>(ILSize.Empty00);
|
---|
| 977 | if (m == A.Size[1]) {
|
---|
| 978 | props |= MatrixProperties.Square;
|
---|
| 979 | if ((props & MatrixProperties.LowerTriangular) != 0) {
|
---|
| 980 | ret.a = solveLowerTriangularSystem(A, B, ref info);
|
---|
| 981 | if (info > 0)
|
---|
| 982 | props |= MatrixProperties.Singular;
|
---|
| 983 | return ret;
|
---|
| 984 | }
|
---|
| 985 | if ((props & MatrixProperties.UpperTriangular) != 0) {
|
---|
| 986 | ret.a = solveUpperTriangularSystem(A, B, ref info);
|
---|
| 987 | if (info > 0)
|
---|
| 988 | props |= MatrixProperties.Singular;
|
---|
| 989 | return ret;
|
---|
| 990 | }
|
---|
| 991 | if ((props & MatrixProperties.Hermitian) != 0) {
|
---|
| 992 | ILDenseStorage< complex> cholFact = A.Storage.copyUpperTriangle(m);
|
---|
| 993 |
|
---|
| 994 | Lapack.zpotrf('U', m, cholFact.GetArrayForWrite(), m, ref info);
|
---|
| 995 | if (info > 0) {
|
---|
| 996 | props ^= MatrixProperties.Hermitian;
|
---|
| 997 | cholFact.Dispose();
|
---|
| 998 | return null;
|
---|
| 999 | } else {
|
---|
| 1000 | // solve
|
---|
| 1001 | ret.a = B.C;
|
---|
| 1002 |
|
---|
| 1003 | Lapack.zpotrs('U', m, B.Size[1], cholFact.GetArrayForWrite(), m, ret.GetArrayForWrite(), m, ref info);
|
---|
| 1004 | cholFact.Dispose();
|
---|
| 1005 | return ret;
|
---|
| 1006 | }
|
---|
| 1007 | } else {
|
---|
| 1008 | // attempt complete (expensive) LU factorization
|
---|
| 1009 | ILArray< complex> L = A.C;
|
---|
| 1010 | int[] pivInd = ILMemoryPool.Pool.New<int>(m);
|
---|
| 1011 |
|
---|
| 1012 | Lapack.zgetrf(m, m, L.GetArrayForWrite(), m, pivInd, ref info);
|
---|
| 1013 | if (info > 0)
|
---|
| 1014 | props |= MatrixProperties.Singular;
|
---|
| 1015 | ret.a = B.C;
|
---|
| 1016 |
|
---|
| 1017 | Lapack.zgetrs('N', m, B.Size[1], L.GetArrayForWrite(), m, pivInd, ret.GetArrayForWrite(), m, ref info);
|
---|
| 1018 | if (info < 0)
|
---|
| 1019 | throw new ILArgumentException("failed to solve via lapack dgetrs");
|
---|
| 1020 | return ret;
|
---|
| 1021 | }
|
---|
| 1022 | } else {
|
---|
| 1023 | // under- / overdetermined system
|
---|
| 1024 | int n = A.Size[1], rank = 0, minMN = (m < n) ? m : n, maxMN = (m > n) ? m : n;
|
---|
| 1025 | int nrhs = B.Size[1];
|
---|
| 1026 | if (B.Size[0] != m)
|
---|
| 1027 | throw new ILArgumentException("right hand side matrix B must match input A!");
|
---|
| 1028 | ILArray< complex> tmpA = A.C;
|
---|
| 1029 | if (m < n) {
|
---|
| 1030 | ret.a = zeros< complex>(n, nrhs);
|
---|
| 1031 | ret[r(0, m - 1), full] = B.C;
|
---|
| 1032 | } else {
|
---|
| 1033 | ret.a = B.C;
|
---|
| 1034 | }
|
---|
| 1035 | int[] JPVT = new int[n];
|
---|
| 1036 |
|
---|
| 1037 | Lapack.zgelsy(m, n, B.Size[1], tmpA.GetArrayForWrite(), m, ret.GetArrayForWrite(),
|
---|
| 1038 | maxMN, JPVT, ILMath.MachineParameterDouble.eps,
|
---|
| 1039 | ref rank, ref info);
|
---|
| 1040 | if (n < m) {
|
---|
| 1041 | ret.a = ret[r(0, n - 1), full];
|
---|
| 1042 | }
|
---|
| 1043 | if (rank < minMN)
|
---|
| 1044 | props |= MatrixProperties.RankDeficient;
|
---|
| 1045 | return ret;
|
---|
| 1046 | }
|
---|
| 1047 | }
|
---|
| 1048 | }
|
---|
| 1049 |
|
---|
| 1050 | /// <summary>
|
---|
| 1051 | /// Solve system of linear equations A*x = b, with A being a upper triangular matrix
|
---|
| 1052 | /// </summary>
|
---|
| 1053 | /// <param name="A">Input matrix of size [n x n], must be upper triangular. No check is made for that!</param>
|
---|
| 1054 | /// <param name="B">Solution vector or matrix. Size [n x m]</param>
|
---|
| 1055 | /// <param name="singularityDetect">[Output] This value gives the row of A, where a singularity has been detected (if any). If A is not singular, this will be a negative value.</param>
|
---|
| 1056 | /// <returns>Solution x solving A * x = b.</returns>
|
---|
| 1057 | /// <remarks><para>The solution will be determined via backward substitution</para>
|
---|
| 1058 | /// <para>Make sure, A and b are of correct size, since no checks are made for that!</para>
|
---|
| 1059 | /// <para>This function is used by ILMath.linsolve. There should be rare need for you to call this function directly.</para>
|
---|
| 1060 | /// <para>Elements of A below the main diagonal will not be accessed.</para>
|
---|
| 1061 | /// <para>If A has been found to be singular, the array returned will contain NaN values for corresponding elements!</para></remarks>
|
---|
| 1062 | internal static ILRetArray< complex > solveUpperTriangularSystem (ILInArray< complex > A, ILInArray< complex > B, ref int singularityDetect) {
|
---|
| 1063 | System.Diagnostics.Debug.Assert(B.Size[1] >= 0);
|
---|
| 1064 | System.Diagnostics.Debug.Assert(B.Size[0] == A.Size[1]);
|
---|
| 1065 | System.Diagnostics.Debug.Assert(A.Size[0] == A.Size[1]);
|
---|
| 1066 | using (ILScope.Enter(A, B)) {
|
---|
| 1067 | singularityDetect = -1;
|
---|
| 1068 | int n = A.Size[0];
|
---|
| 1069 | int m = B.Size[1];
|
---|
| 1070 | int info = 0;
|
---|
| 1071 | ILArray< complex> ret = B.C;
|
---|
| 1072 |
|
---|
| 1073 | complex[] retArr = ret.GetArrayForWrite();
|
---|
| 1074 | // solve using Lapack
|
---|
| 1075 | unsafe {
|
---|
| 1076 | fixed ( complex* ptrA = A.GetArrayForRead())
|
---|
| 1077 | fixed ( complex* ptrB = ret.GetArrayForWrite()) {
|
---|
| 1078 |
|
---|
| 1079 | Lapack.ztrtrs('U', 'N', 'N', A.Size[0], B.Size[1], (IntPtr)ptrA, A.Size[0], (IntPtr)ptrB, B.Size[0], ref info);
|
---|
| 1080 | }
|
---|
| 1081 | }
|
---|
| 1082 | if (info < 0)
|
---|
| 1083 | throw new ILArgumentException("error inside Lapack function ?trtrs for argument: " + (-info));
|
---|
| 1084 | if (info > 0) {
|
---|
| 1085 | singularityDetect = info - 1;
|
---|
| 1086 | for (m = 0; m < ret.Size[1]; m++) {
|
---|
| 1087 | info = m * n + singularityDetect;
|
---|
| 1088 | for (int i = singularityDetect; i < n; i++) {
|
---|
| 1089 | retArr[info++] = new complex(double.NaN,double.NaN);
|
---|
| 1090 | }
|
---|
| 1091 | }
|
---|
| 1092 | } else {
|
---|
| 1093 | singularityDetect = -1;
|
---|
| 1094 | }
|
---|
| 1095 | return ret;
|
---|
| 1096 | }
|
---|
| 1097 | }
|
---|
| 1098 |
|
---|
| 1099 | /// <summary>
|
---|
| 1100 | /// Solve system of linear equations A*x = b, with A being a lower triangular matrix
|
---|
| 1101 | /// </summary>
|
---|
| 1102 | /// <param name="A">Input matrix of size [n x n], must be lower triangular. No check is made for that!</param>
|
---|
| 1103 | /// <param name="B">Solution vector. Size [n x 1]</param>
|
---|
| 1104 | /// <param name="singularityDetect">[Output] This value gives the row of A, where a singularity has been detected (if any). If A is not singular, this will be a negative value.</param>
|
---|
| 1105 | /// <returns>Solution x solving A * x = b.</returns>
|
---|
| 1106 | /// <remarks><para>The solution will be determined via forward substitution</para>
|
---|
| 1107 | /// <para>Make sure, A and b are of correct size, since no checks are made for that!</para>
|
---|
| 1108 | /// <para>This function is used by ILMath.linsolve. There should be rare need for you to call this function directly.</para>
|
---|
| 1109 | /// <para>Elements of A above the main diagonal will not be accessed.</para>
|
---|
| 1110 | /// <para>If A has been found to be singular, the array returned will contain NaN values for corresponding elements!</para></remarks>
|
---|
| 1111 | internal static ILRetArray< complex> solveLowerTriangularSystem( ILInArray< complex> A, ILInArray< complex> B, ref int singularityDetect ) {
|
---|
| 1112 | System.Diagnostics.Debug.Assert(B.Size[1] >= 0);
|
---|
| 1113 | System.Diagnostics.Debug.Assert(B.Size[0] == A.Size[1]);
|
---|
| 1114 | System.Diagnostics.Debug.Assert(A.Size[0] == A.Size[1]);
|
---|
| 1115 | using (ILScope.Enter(A, B)) {
|
---|
| 1116 | singularityDetect = -1;
|
---|
| 1117 | int n = A.Size[0];
|
---|
| 1118 | int m = B.Size[1];
|
---|
| 1119 | int info = 0;
|
---|
| 1120 | ILArray< complex> ret = B.C;
|
---|
| 1121 |
|
---|
| 1122 | complex[] retArr = ret.GetArrayForWrite();
|
---|
| 1123 | // solve using Lapack
|
---|
| 1124 | unsafe {
|
---|
| 1125 | fixed ( complex* ptrA = A.GetArrayForRead())
|
---|
| 1126 | fixed ( complex* ptrB = ret.GetArrayForWrite()) {
|
---|
| 1127 |
|
---|
| 1128 | Lapack.ztrtrs('L', 'N', 'N', A.Size[0], B.Size[1], (IntPtr)ptrA, A.Size[0], (IntPtr)ptrB, B.Size[0], ref info);
|
---|
| 1129 | }
|
---|
| 1130 | }
|
---|
| 1131 | if (info < 0)
|
---|
| 1132 | throw new ILArgumentException("error inside Lapack function ?trtrs for argument: " + (-info));
|
---|
| 1133 | if (info > 0) {
|
---|
| 1134 | singularityDetect = info - 1;
|
---|
| 1135 | for (m = 0; m < ret.Size[1]; m++) {
|
---|
| 1136 | info = m * n + singularityDetect;
|
---|
| 1137 | for (int i = singularityDetect; i < n; i++) {
|
---|
| 1138 | retArr[info++] = new complex(double.NaN,double.NaN);
|
---|
| 1139 | }
|
---|
| 1140 | }
|
---|
| 1141 | } else {
|
---|
| 1142 | singularityDetect = -1;
|
---|
| 1143 | }
|
---|
| 1144 | return ret;
|
---|
| 1145 | }
|
---|
| 1146 | }
|
---|
| 1147 |
|
---|
| 1148 |
|
---|
| 1149 | #endregion HYCALPER AUTO GENERATED CODE
|
---|
| 1150 |
|
---|
| 1151 | }
|
---|
| 1152 | }
|
---|