View Javadoc
1   /*
2    * Licensed to the Apache Software Foundation (ASF) under one or more
3    * contributor license agreements.  See the NOTICE file distributed with
4    * this work for additional information regarding copyright ownership.
5    * The ASF licenses this file to You under the Apache License, Version 2.0
6    * (the "License"); you may not use this file except in compliance with
7    * the License.  You may obtain a copy of the License at
8    *
9    *      https://www.apache.org/licenses/LICENSE-2.0
10   *
11   * Unless required by applicable law or agreed to in writing, software
12   * distributed under the License is distributed on an "AS IS" BASIS,
13   * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14   * See the License for the specific language governing permissions and
15   * limitations under the License.
16   */
17  
18  /*
19   * This is not the original file distributed by the Apache Software Foundation
20   * It has been modified by the Hipparchus project
21   */
22  
23  package org.hipparchus.linear;
24  
25  import java.util.Arrays;
26  
27  import org.hipparchus.exception.LocalizedCoreFormats;
28  import org.hipparchus.exception.MathIllegalArgumentException;
29  import org.hipparchus.util.FastMath;
30  
31  
32  /**
33   * Class transforming a symmetrical matrix to tridiagonal shape.
34   * <p>A symmetrical m &times; m matrix A can be written as the product of three matrices:
35   * A = Q &times; T &times; Q<sup>T</sup> with Q an orthogonal matrix and T a symmetrical
36   * tridiagonal matrix. Both Q and T are m &times; m matrices.</p>
37   * <p>This implementation only uses the upper part of the matrix, the part below the
38   * diagonal is not accessed at all.</p>
39   * <p>Transformation to tridiagonal shape is often not a goal by itself, but it is
40   * an intermediate step in more general decomposition algorithms like {@link
41   * EigenDecompositionSymmetric eigen decomposition}. This class is therefore intended for internal
42   * use by the library and is not public. As a consequence of this explicitly limited scope,
43   * many methods directly returns references to internal arrays, not copies.</p>
44   */
45  class TriDiagonalTransformer {
46      /** Householder vectors. */
47      private final double[][] householderVectors;
48      /** Main diagonal. */
49      private final double[] main;
50      /** Secondary diagonal. */
51      private final double[] secondary;
52      /** Cached value of Q. */
53      private RealMatrix cachedQ;
54      /** Cached value of Qt. */
55      private RealMatrix cachedQt;
56      /** Cached value of T. */
57      private RealMatrix cachedT;
58  
59      /**
60       * Build the transformation to tridiagonal shape of a symmetrical matrix.
61       * <p>The specified matrix is assumed to be symmetrical without any check.
62       * Only the upper triangular part of the matrix is used.</p>
63       *
64       * @param matrix Symmetrical matrix to transform.
65       * @throws MathIllegalArgumentException if the matrix is not square.
66       */
67      TriDiagonalTransformer(RealMatrix matrix) {
68          if (!matrix.isSquare()) {
69              throw new MathIllegalArgumentException(LocalizedCoreFormats.NON_SQUARE_MATRIX,
70                                                     matrix.getRowDimension(), matrix.getColumnDimension());
71          }
72  
73          final int m = matrix.getRowDimension();
74          householderVectors = matrix.getData();
75          main      = new double[m];
76          secondary = new double[m - 1];
77          cachedQ   = null;
78          cachedQt  = null;
79          cachedT   = null;
80  
81          // transform matrix
82          transform();
83      }
84  
85      /**
86       * Returns the matrix Q of the transform.
87       * <p>Q is an orthogonal matrix, i.e. its transpose is also its inverse.</p>
88       * @return the Q matrix
89       */
90      public RealMatrix getQ() {
91          if (cachedQ == null) {
92              cachedQ = getQT().transpose();
93          }
94          return cachedQ;
95      }
96  
97      /**
98       * Returns the transpose of the matrix Q of the transform.
99       * <p>Q is an orthogonal matrix, i.e. its transpose is also its inverse.</p>
100      * @return the Q matrix
101      */
102     public RealMatrix getQT() {
103         if (cachedQt == null) {
104             final int m = householderVectors.length;
105             double[][] qta = new double[m][m];
106 
107             // build up first part of the matrix by applying Householder transforms
108             for (int k = m - 1; k >= 1; --k) {
109                 final double[] hK = householderVectors[k - 1];
110                 qta[k][k] = 1;
111                 if (hK[k] != 0.0) {
112                     final double inv = 1.0 / (secondary[k - 1] * hK[k]);
113                     double beta = 1.0 / secondary[k - 1];
114                     qta[k][k] = 1 + beta * hK[k];
115                     for (int i = k + 1; i < m; ++i) {
116                         qta[k][i] = beta * hK[i];
117                     }
118                     for (int j = k + 1; j < m; ++j) {
119                         beta = 0;
120                         for (int i = k + 1; i < m; ++i) {
121                             beta += qta[j][i] * hK[i];
122                         }
123                         beta *= inv;
124                         qta[j][k] = beta * hK[k];
125                         for (int i = k + 1; i < m; ++i) {
126                             qta[j][i] += beta * hK[i];
127                         }
128                     }
129                 }
130             }
131             qta[0][0] = 1;
132             cachedQt = MatrixUtils.createRealMatrix(qta);
133         }
134 
135         // return the cached matrix
136         return cachedQt;
137     }
138 
139     /**
140      * Returns the tridiagonal matrix T of the transform.
141      * @return the T matrix
142      */
143     public RealMatrix getT() {
144         if (cachedT == null) {
145             final int m = main.length;
146             double[][] ta = new double[m][m];
147             for (int i = 0; i < m; ++i) {
148                 ta[i][i] = main[i];
149                 if (i > 0) {
150                     ta[i][i - 1] = secondary[i - 1];
151                 }
152                 if (i < main.length - 1) {
153                     ta[i][i + 1] = secondary[i];
154                 }
155             }
156             cachedT = MatrixUtils.createRealMatrix(ta);
157         }
158 
159         // return the cached matrix
160         return cachedT;
161     }
162 
163     /**
164      * Get the Householder vectors of the transform.
165      * <p>Note that since this class is only intended for internal use,
166      * it returns directly a reference to its internal arrays, not a copy.</p>
167      * @return the main diagonal elements of the B matrix
168      */
169     double[][] getHouseholderVectorsRef() {
170         return householderVectors; // NOPMD - returning an internal array is intentional and documented here
171     }
172 
173     /**
174      * Get the main diagonal elements of the matrix T of the transform.
175      * <p>Note that since this class is only intended for internal use,
176      * it returns directly a reference to its internal arrays, not a copy.</p>
177      * @return the main diagonal elements of the T matrix
178      */
179     double[] getMainDiagonalRef() {
180         return main; // NOPMD - returning an internal array is intentional and documented here
181     }
182 
183     /**
184      * Get the secondary diagonal elements of the matrix T of the transform.
185      * <p>Note that since this class is only intended for internal use,
186      * it returns directly a reference to its internal arrays, not a copy.</p>
187      * @return the secondary diagonal elements of the T matrix
188      */
189     double[] getSecondaryDiagonalRef() {
190         return secondary; // NOPMD - returning an internal array is intentional and documented here
191     }
192 
193     /**
194      * Transform original matrix to tridiagonal form.
195      * <p>Transformation is done using Householder transforms.</p>
196      */
197     private void transform() {
198         final int m = householderVectors.length;
199         final double[] z = new double[m];
200         for (int k = 0; k < m - 1; k++) {
201 
202             //zero-out a row and a column simultaneously
203             final double[] hK = householderVectors[k];
204             main[k] = hK[k];
205             double xNormSqr = 0;
206             for (int j = k + 1; j < m; ++j) {
207                 final double c = hK[j];
208                 xNormSqr += c * c;
209             }
210             final double a = (hK[k + 1] > 0) ? -FastMath.sqrt(xNormSqr) : FastMath.sqrt(xNormSqr);
211             secondary[k] = a;
212             if (a != 0.0) {
213                 // apply Householder transform from left and right simultaneously
214 
215                 hK[k + 1] -= a;
216                 final double beta = -1 / (a * hK[k + 1]);
217 
218                 // compute a = beta A v, where v is the Householder vector
219                 // this loop is written in such a way
220                 //   1) only the upper triangular part of the matrix is accessed
221                 //   2) access is cache-friendly for a matrix stored in rows
222                 Arrays.fill(z, k + 1, m, 0);
223                 for (int i = k + 1; i < m; ++i) {
224                     final double[] hI = householderVectors[i];
225                     final double hKI = hK[i];
226                     double zI = hI[i] * hKI;
227                     for (int j = i + 1; j < m; ++j) {
228                         final double hIJ = hI[j];
229                         zI   += hIJ * hK[j];
230                         z[j] += hIJ * hKI;
231                     }
232                     z[i] = beta * (z[i] + zI);
233                 }
234 
235                 // compute gamma = beta vT z / 2
236                 double gamma = 0;
237                 for (int i = k + 1; i < m; ++i) {
238                     gamma += z[i] * hK[i];
239                 }
240                 gamma *= beta / 2;
241 
242                 // compute z = z - gamma v
243                 for (int i = k + 1; i < m; ++i) {
244                     z[i] -= gamma * hK[i];
245                 }
246 
247                 // update matrix: A = A - v zT - z vT
248                 // only the upper triangular part of the matrix is updated
249                 for (int i = k + 1; i < m; ++i) {
250                     final double[] hI = householderVectors[i];
251                     for (int j = i; j < m; ++j) {
252                         hI[j] -= hK[i] * z[j] + z[i] * hK[j];
253                     }
254                 }
255             }
256         }
257         main[m - 1] = householderVectors[m - 1][m - 1];
258     }
259 }