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.ode.nonstiff;
24  
25  import java.util.Arrays;
26  import java.util.HashMap;
27  import java.util.Map;
28  
29  import org.hipparchus.Field;
30  import org.hipparchus.CalculusFieldElement;
31  import org.hipparchus.linear.Array2DRowFieldMatrix;
32  import org.hipparchus.linear.ArrayFieldVector;
33  import org.hipparchus.linear.FieldDecompositionSolver;
34  import org.hipparchus.linear.FieldLUDecomposition;
35  import org.hipparchus.linear.FieldMatrix;
36  import org.hipparchus.util.MathArrays;
37  
38  /** Transformer to Nordsieck vectors for Adams integrators.
39   * <p>This class is used by {@link AdamsBashforthIntegrator Adams-Bashforth} and
40   * {@link AdamsMoultonIntegrator Adams-Moulton} integrators to convert between
41   * classical representation with several previous first derivatives and Nordsieck
42   * representation with higher order scaled derivatives.</p>
43   *
44   * <p>We define scaled derivatives s<sub>i</sub>(n) at step n as:
45   * \[
46   *   \left\{\begin{align}
47   *   s_1(n) &amp;= h y'_n \text{ for first derivative}\\
48   *   s_2(n) &amp;= \frac{h^2}{2} y_n'' \text{ for second derivative}\\
49   *   s_3(n) &amp;= \frac{h^3}{6} y_n''' \text{ for third derivative}\\
50   *   &amp;\cdots\\
51   *   s_k(n) &amp;= \frac{h^k}{k!} y_n^{(k)} \text{ for } k^\mathrm{th} \text{ derivative}
52   *   \end{align}\right.
53   * \]</p>
54   *
55   * <p>With the previous definition, the classical representation of multistep methods
56   * uses first derivatives only, i.e. it handles y<sub>n</sub>, s<sub>1</sub>(n) and
57   * q<sub>n</sub> where q<sub>n</sub> is defined as:
58   * \[
59   *   q_n = [ s_1(n-1) s_1(n-2) \ldots s_1(n-(k-1)) ]^T
60   * \]
61   * (we omit the k index in the notation for clarity).</p>
62   *
63   * <p>Another possible representation uses the Nordsieck vector with
64   * higher degrees scaled derivatives all taken at the same step, i.e it handles y<sub>n</sub>,
65   * s<sub>1</sub>(n) and r<sub>n</sub>) where r<sub>n</sub> is defined as:
66   * \[
67   * r_n = [ s_2(n), s_3(n) \ldots s_k(n) ]^T
68   * \]
69   * (here again we omit the k index in the notation for clarity)
70   * </p>
71   *
72   * <p>Taylor series formulas show that for any index offset i, s<sub>1</sub>(n-i) can be
73   * computed from s<sub>1</sub>(n), s<sub>2</sub>(n) ... s<sub>k</sub>(n), the formula being exact
74   * for degree k polynomials.
75   * \[
76   * s_1(n-i) = s_1(n) + \sum_{j\gt 0} (j+1) (-i)^j s_{j+1}(n)
77   * \]
78   * The previous formula can be used with several values for i to compute the transform between
79   * classical representation and Nordsieck vector at step end. The transform between r<sub>n</sub>
80   * and q<sub>n</sub> resulting from the Taylor series formulas above is:
81   * \[
82   * q_n = s_1(n) u + P r_n
83   * \]
84   * where u is the [ 1 1 ... 1 ]<sup>T</sup> vector and P is the (k-1)&times;(k-1) matrix built
85   * with the \((j+1) (-i)^j\) terms with i being the row number starting from 1 and j being
86   * the column number starting from 1:
87   * \[
88   *   P=\begin{bmatrix}
89   *   -2  &amp;  3 &amp;   -4 &amp;    5 &amp; \ldots \\
90   *   -4  &amp; 12 &amp;  -32 &amp;   80 &amp; \ldots \\
91   *   -6  &amp; 27 &amp; -108 &amp;  405 &amp; \ldots \\
92   *   -8  &amp; 48 &amp; -256 &amp; 1280 &amp; \ldots \\
93   *       &amp;    &amp;  \ldots\\
94   *    \end{bmatrix}
95   * \]
96   *
97   * <p>Changing -i into +i in the formula above can be used to compute a similar transform between
98   * classical representation and Nordsieck vector at step start. The resulting matrix is simply
99   * the absolute value of matrix P.</p>
100  *
101  * <p>For {@link AdamsBashforthIntegrator Adams-Bashforth} method, the Nordsieck vector
102  * at step n+1 is computed from the Nordsieck vector at step n as follows:
103  * <ul>
104  *   <li>y<sub>n+1</sub> = y<sub>n</sub> + s<sub>1</sub>(n) + u<sup>T</sup> r<sub>n</sub></li>
105  *   <li>s<sub>1</sub>(n+1) = h f(t<sub>n+1</sub>, y<sub>n+1</sub>)</li>
106  *   <li>r<sub>n+1</sub> = (s<sub>1</sub>(n) - s<sub>1</sub>(n+1)) P<sup>-1</sup> u + P<sup>-1</sup> A P r<sub>n</sub></li>
107  * </ul>
108  * <p>where A is a rows shifting matrix (the lower left part is an identity matrix):</p>
109  * <pre>
110  *        [ 0 0   ...  0 0 | 0 ]
111  *        [ ---------------+---]
112  *        [ 1 0   ...  0 0 | 0 ]
113  *    A = [ 0 1   ...  0 0 | 0 ]
114  *        [       ...      | 0 ]
115  *        [ 0 0   ...  1 0 | 0 ]
116  *        [ 0 0   ...  0 1 | 0 ]
117  * </pre>
118  *
119  * <p>For {@link AdamsMoultonIntegrator Adams-Moulton} method, the predicted Nordsieck vector
120  * at step n+1 is computed from the Nordsieck vector at step n as follows:
121  * <ul>
122  *   <li>Y<sub>n+1</sub> = y<sub>n</sub> + s<sub>1</sub>(n) + u<sup>T</sup> r<sub>n</sub></li>
123  *   <li>S<sub>1</sub>(n+1) = h f(t<sub>n+1</sub>, Y<sub>n+1</sub>)</li>
124  *   <li>R<sub>n+1</sub> = (s<sub>1</sub>(n) - s<sub>1</sub>(n+1)) P<sup>-1</sup> u + P<sup>-1</sup> A P r<sub>n</sub></li>
125  * </ul>
126  * From this predicted vector, the corrected vector is computed as follows:
127  * <ul>
128  *   <li>y<sub>n+1</sub> = y<sub>n</sub> + S<sub>1</sub>(n+1) + [ -1 +1 -1 +1 ... &plusmn;1 ] r<sub>n+1</sub></li>
129  *   <li>s<sub>1</sub>(n+1) = h f(t<sub>n+1</sub>, y<sub>n+1</sub>)</li>
130  *   <li>r<sub>n+1</sub> = R<sub>n+1</sub> + (s<sub>1</sub>(n+1) - S<sub>1</sub>(n+1)) P<sup>-1</sup> u</li>
131  * </ul>
132  * <p>where the upper case Y<sub>n+1</sub>, S<sub>1</sub>(n+1) and R<sub>n+1</sub> represent the
133  * predicted states whereas the lower case y<sub>n+1</sub>, s<sub>n+1</sub> and r<sub>n+1</sub>
134  * represent the corrected states.</p>
135  *
136  * <p>We observe that both methods use similar update formulas. In both cases a P<sup>-1</sup>u
137  * vector and a P<sup>-1</sup> A P matrix are used that do not depend on the state,
138  * they only depend on k. This class handles these transformations.</p>
139  *
140  * @param <T> the type of the field elements
141  */
142 public class AdamsNordsieckFieldTransformer<T extends CalculusFieldElement<T>> {
143 
144     /** Cache for already computed coefficients. */
145     private static final Map<Integer,
146                          Map<Field<? extends CalculusFieldElement<?>>,
147                                    AdamsNordsieckFieldTransformer<? extends CalculusFieldElement<?>>>> CACHE = new HashMap<>();
148 
149     /** Field to which the time and state vector elements belong. */
150     private final Field<T> field;
151 
152     /** Update matrix for the higher order derivatives h<sup>2</sup>/2 y'', h<sup>3</sup>/6 y''' ... */
153     private final Array2DRowFieldMatrix<T> update;
154 
155     /** Update coefficients of the higher order derivatives wrt y'. */
156     private final T[] c1;
157 
158     /** Simple constructor.
159      * @param field field to which the time and state vector elements belong
160      * @param n number of steps of the multistep method
161      * (excluding the one being computed)
162      */
163     private AdamsNordsieckFieldTransformer(final Field<T> field, final int n) {
164 
165         this.field = field;
166         final int rows = n - 1;
167 
168         // compute coefficients
169         FieldMatrix<T> bigP = buildP(rows);
170         FieldDecompositionSolver<T> pSolver =
171             new FieldLUDecomposition<T>(bigP).getSolver();
172 
173         T[] u = MathArrays.buildArray(field, rows);
174         Arrays.fill(u, field.getOne());
175         c1 = pSolver.solve(new ArrayFieldVector<T>(u, false)).toArray();
176 
177         // update coefficients are computed by combining transform from
178         // Nordsieck to multistep, then shifting rows to represent step advance
179         // then applying inverse transform
180         T[][] shiftedP = bigP.getData();
181         for (int i = shiftedP.length - 1; i > 0; --i) {
182             // shift rows
183             shiftedP[i] = shiftedP[i - 1];
184         }
185         shiftedP[0] = MathArrays.buildArray(field, rows);
186         Arrays.fill(shiftedP[0], field.getZero());
187         update = new Array2DRowFieldMatrix<>(pSolver.solve(new Array2DRowFieldMatrix<T>(shiftedP, false)).getData());
188 
189     }
190 
191     /** Get the Nordsieck transformer for a given field and number of steps.
192      * @param field field to which the time and state vector elements belong
193      * @param nSteps number of steps of the multistep method
194      * (excluding the one being computed)
195      * @return Nordsieck transformer for the specified field and number of steps
196      * @param <T> the type of the field elements
197      */
198     public static <T extends CalculusFieldElement<T>> AdamsNordsieckFieldTransformer<T> // NOPMD - PMD false positive
199     getInstance(final Field<T> field, final int nSteps) {
200         synchronized(CACHE) {
201             Map<Field<? extends CalculusFieldElement<?>>,
202                       AdamsNordsieckFieldTransformer<? extends CalculusFieldElement<?>>> map = CACHE.get(nSteps);
203             if (map == null) {
204                 map = new HashMap<>();
205                 CACHE.put(nSteps, map);
206             }
207             @SuppressWarnings("unchecked")
208             AdamsNordsieckFieldTransformer<T> t = (AdamsNordsieckFieldTransformer<T>) map.get(field);
209             if (t == null) {
210                 t = new AdamsNordsieckFieldTransformer<>(field, nSteps);
211                 map.put(field, t);
212             }
213             return t;
214 
215         }
216     }
217 
218     /** Build the P matrix.
219      * <p>The P matrix general terms are shifted \((j+1) (-i)^j\) terms
220      * with i being the row number starting from 1 and j being the column
221      * number starting from 1:
222      * <pre>
223      *        [  -2   3   -4    5  ... ]
224      *        [  -4  12  -32   80  ... ]
225      *   P =  [  -6  27 -108  405  ... ]
226      *        [  -8  48 -256 1280  ... ]
227      *        [          ...           ]
228      * </pre></p>
229      * @param rows number of rows of the matrix
230      * @return P matrix
231      */
232     private FieldMatrix<T> buildP(final int rows) {
233 
234         final T[][] pData = MathArrays.buildArray(field, rows, rows);
235 
236         for (int i = 1; i <= pData.length; ++i) {
237             // build the P matrix elements from Taylor series formulas
238             final T[] pI = pData[i - 1];
239             final int factor = -i;
240             T aj = field.getZero().add(factor);
241             for (int j = 1; j <= pI.length; ++j) {
242                 pI[j - 1] = aj.multiply(j + 1);
243                 aj = aj.multiply(factor);
244             }
245         }
246 
247         return new Array2DRowFieldMatrix<T>(pData, false);
248 
249     }
250 
251     /** Initialize the high order scaled derivatives at step start.
252      * @param h step size to use for scaling
253      * @param t first steps times
254      * @param y first steps states
255      * @param yDot first steps derivatives
256      * @return Nordieck vector at start of first step (h<sup>2</sup>/2 y''<sub>n</sub>,
257      * h<sup>3</sup>/6 y'''<sub>n</sub> ... h<sup>k</sup>/k! y<sup>(k)</sup><sub>n</sub>)
258      */
259 
260     public Array2DRowFieldMatrix<T> initializeHighOrderDerivatives(final T h, final T[] t,
261                                                                    final T[][] y,
262                                                                    final T[][] yDot) {
263 
264         // using Taylor series with di = ti - t0, we get:
265         //  y(ti)  - y(t0)  - di y'(t0) =   di^2 / h^2 s2 + ... +   di^k     / h^k sk + O(h^k)
266         //  y'(ti) - y'(t0)             = 2 di   / h^2 s2 + ... + k di^(k-1) / h^k sk + O(h^(k-1))
267         // we write these relations for i = 1 to i= 1+n/2 as a set of n + 2 linear
268         // equations depending on the Nordsieck vector [s2 ... sk rk], so s2 to sk correspond
269         // to the appropriately truncated Taylor expansion, and rk is the Taylor remainder.
270         // The goal is to have s2 to sk as accurate as possible considering the fact the sum is
271         // truncated and we don't want the error terms to be included in s2 ... sk, so we need
272         // to solve also for the remainder
273         final T[][] a     = MathArrays.buildArray(field, c1.length + 1, c1.length + 1);
274         final T[][] b     = MathArrays.buildArray(field, c1.length + 1, y[0].length);
275         final T[]   y0    = y[0];
276         final T[]   yDot0 = yDot[0];
277         for (int i = 1; i < y.length; ++i) {
278 
279             final T di    = t[i].subtract(t[0]);
280             final T ratio = di.divide(h);
281             T dikM1Ohk    = h.reciprocal();
282 
283             // linear coefficients of equations
284             // y(ti) - y(t0) - di y'(t0) and y'(ti) - y'(t0)
285             final T[] aI    = a[2 * i - 2];
286             final T[] aDotI = (2 * i - 1) < a.length ? a[2 * i - 1] : null;
287             for (int j = 0; j < aI.length; ++j) {
288                 dikM1Ohk = dikM1Ohk.multiply(ratio);
289                 aI[j]    = di.multiply(dikM1Ohk);
290                 if (aDotI != null) {
291                     aDotI[j]  = dikM1Ohk.multiply(j + 2);
292                 }
293             }
294 
295             // expected value of the previous equations
296             final T[] yI    = y[i];
297             final T[] yDotI = yDot[i];
298             final T[] bI    = b[2 * i - 2];
299             final T[] bDotI = (2 * i - 1) < b.length ? b[2 * i - 1] : null;
300             for (int j = 0; j < yI.length; ++j) {
301                 bI[j]    = yI[j].subtract(y0[j]).subtract(di.multiply(yDot0[j]));
302                 if (bDotI != null) {
303                     bDotI[j] = yDotI[j].subtract(yDot0[j]);
304                 }
305             }
306 
307         }
308 
309         // solve the linear system to get the best estimate of the Nordsieck vector [s2 ... sk],
310         // with the additional terms s(k+1) and c grabbing the parts after the truncated Taylor expansion
311         final FieldLUDecomposition<T> decomposition = new FieldLUDecomposition<>(new Array2DRowFieldMatrix<T>(a, false));
312         final FieldMatrix<T> x = decomposition.getSolver().solve(new Array2DRowFieldMatrix<T>(b, false));
313 
314         // extract just the Nordsieck vector [s2 ... sk]
315         final Array2DRowFieldMatrix<T> truncatedX =
316                         new Array2DRowFieldMatrix<>(field, x.getRowDimension() - 1, x.getColumnDimension());
317         for (int i = 0; i < truncatedX.getRowDimension(); ++i) {
318             for (int j = 0; j < truncatedX.getColumnDimension(); ++j) {
319                 truncatedX.setEntry(i, j, x.getEntry(i, j));
320             }
321         }
322         return truncatedX;
323 
324     }
325 
326     /** Update the high order scaled derivatives for Adams integrators (phase 1).
327      * <p>The complete update of high order derivatives has a form similar to:
328      * \[
329      * r_{n+1} = (s_1(n) - s_1(n+1)) P^{-1} u + P^{-1} A P r_n
330      * \]
331      * this method computes the P<sup>-1</sup> A P r<sub>n</sub> part.</p>
332      * @param highOrder high order scaled derivatives
333      * (h<sup>2</sup>/2 y'', ... h<sup>k</sup>/k! y(k))
334      * @return updated high order derivatives
335      * @see #updateHighOrderDerivativesPhase2(CalculusFieldElement[], CalculusFieldElement[], Array2DRowFieldMatrix)
336      */
337     public Array2DRowFieldMatrix<T> updateHighOrderDerivativesPhase1(final Array2DRowFieldMatrix<T> highOrder) {
338         return update.multiply(highOrder);
339     }
340 
341     /** Update the high order scaled derivatives Adams integrators (phase 2).
342      * <p>The complete update of high order derivatives has a form similar to:
343      * \[
344      * r_{n+1} = (s_1(n) - s_1(n+1)) P^{-1} u + P^{-1} A P r_n
345      * \]
346      * this method computes the (s<sub>1</sub>(n) - s<sub>1</sub>(n+1)) P<sup>-1</sup> u part.</p>
347      * <p>Phase 1 of the update must already have been performed.</p>
348      * @param start first order scaled derivatives at step start
349      * @param end first order scaled derivatives at step end
350      * @param highOrder high order scaled derivatives, will be modified
351      * (h<sup>2</sup>/2 y'', ... h<sup>k</sup>/k! y(k))
352      * @see #updateHighOrderDerivativesPhase1(Array2DRowFieldMatrix)
353      */
354     public void updateHighOrderDerivativesPhase2(final T[] start,
355                                                  final T[] end,
356                                                  final Array2DRowFieldMatrix<T> highOrder) {
357         final T[][] data = highOrder.getDataRef();
358         for (int i = 0; i < data.length; ++i) {
359             final T[] dataI = data[i];
360             final T c1I = c1[i];
361             for (int j = 0; j < dataI.length; ++j) {
362                 dataI[j] = dataI[j].add(c1I.multiply(start[j].subtract(end[j])));
363             }
364         }
365     }
366 
367 }