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    *      http://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  package org.apache.commons.math4.legacy.ode.nonstiff;
19  
20  import org.apache.commons.math4.legacy.exception.DimensionMismatchException;
21  import org.apache.commons.math4.legacy.exception.MaxCountExceededException;
22  import org.apache.commons.math4.legacy.exception.NoBracketingException;
23  import org.apache.commons.math4.legacy.exception.NumberIsTooSmallException;
24  import org.apache.commons.math4.legacy.ode.ExpandableStatefulODE;
25  import org.apache.commons.math4.core.jdkmath.JdkMath;
26  
27  /**
28   * This class implements the common part of all embedded Runge-Kutta
29   * integrators for Ordinary Differential Equations.
30   *
31   * <p>These methods are embedded explicit Runge-Kutta methods with two
32   * sets of coefficients allowing to estimate the error, their Butcher
33   * arrays are as follows :
34   * <pre>
35   *    0  |
36   *   c2  | a21
37   *   c3  | a31  a32
38   *   ... |        ...
39   *   cs  | as1  as2  ...  ass-1
40   *       |--------------------------
41   *       |  b1   b2  ...   bs-1  bs
42   *       |  b'1  b'2 ...   b's-1 b's
43   * </pre>
44   *
45   * <p>In fact, we rather use the array defined by ej = bj - b'j to
46   * compute directly the error rather than computing two estimates and
47   * then comparing them.</p>
48   *
49   * <p>Some methods are qualified as <i>fsal</i> (first same as last)
50   * methods. This means the last evaluation of the derivatives in one
51   * step is the same as the first in the next step. Then, this
52   * evaluation can be reused from one step to the next one and the cost
53   * of such a method is really s-1 evaluations despite the method still
54   * has s stages. This behaviour is true only for successful steps, if
55   * the step is rejected after the error estimation phase, no
56   * evaluation is saved. For an <i>fsal</i> method, we have cs = 1 and
57   * asi = bi for all i.</p>
58   *
59   * @since 1.2
60   */
61  
62  public abstract class EmbeddedRungeKuttaIntegrator
63    extends AdaptiveStepsizeIntegrator {
64  
65      /** Indicator for <i>fsal</i> methods. */
66      private final boolean fsal;
67  
68      /** Time steps from Butcher array (without the first zero). */
69      private final double[] c;
70  
71      /** Internal weights from Butcher array (without the first empty row). */
72      private final double[][] a;
73  
74      /** External weights for the high order method from Butcher array. */
75      private final double[] b;
76  
77      /** Prototype of the step interpolator. */
78      private final RungeKuttaStepInterpolator prototype;
79  
80      /** Stepsize control exponent. */
81      private final double exp;
82  
83      /** Safety factor for stepsize control. */
84      private double safety;
85  
86      /** Minimal reduction factor for stepsize control. */
87      private double minReduction;
88  
89      /** Maximal growth factor for stepsize control. */
90      private double maxGrowth;
91  
92    /** Build a Runge-Kutta integrator with the given Butcher array.
93     * @param name name of the method
94     * @param fsal indicate that the method is an <i>fsal</i>
95     * @param c time steps from Butcher array (without the first zero)
96     * @param a internal weights from Butcher array (without the first empty row)
97     * @param b propagation weights for the high order method from Butcher array
98     * @param prototype prototype of the step interpolator to use
99     * @param minStep minimal step (sign is irrelevant, regardless of
100    * integration direction, forward or backward), the last step can
101    * be smaller than this
102    * @param maxStep maximal step (sign is irrelevant, regardless of
103    * integration direction, forward or backward), the last step can
104    * be smaller than this
105    * @param scalAbsoluteTolerance allowed absolute error
106    * @param scalRelativeTolerance allowed relative error
107    */
108   protected EmbeddedRungeKuttaIntegrator(final String name, final boolean fsal,
109                                          final double[] c, final double[][] a, final double[] b,
110                                          final RungeKuttaStepInterpolator prototype,
111                                          final double minStep, final double maxStep,
112                                          final double scalAbsoluteTolerance,
113                                          final double scalRelativeTolerance) {
114 
115     super(name, minStep, maxStep, scalAbsoluteTolerance, scalRelativeTolerance);
116 
117     this.fsal      = fsal;
118     this.c         = c;
119     this.a         = a;
120     this.b         = b;
121     this.prototype = prototype;
122 
123     exp = -1.0 / getOrder();
124 
125     // set the default values of the algorithm control parameters
126     setSafety(0.9);
127     setMinReduction(0.2);
128     setMaxGrowth(10.0);
129   }
130 
131   /** Build a Runge-Kutta integrator with the given Butcher array.
132    * @param name name of the method
133    * @param fsal indicate that the method is an <i>fsal</i>
134    * @param c time steps from Butcher array (without the first zero)
135    * @param a internal weights from Butcher array (without the first empty row)
136    * @param b propagation weights for the high order method from Butcher array
137    * @param prototype prototype of the step interpolator to use
138    * @param minStep minimal step (must be positive even for backward
139    * integration), the last step can be smaller than this
140    * @param maxStep maximal step (must be positive even for backward
141    * integration)
142    * @param vecAbsoluteTolerance allowed absolute error
143    * @param vecRelativeTolerance allowed relative error
144    */
145   protected EmbeddedRungeKuttaIntegrator(final String name, final boolean fsal,
146                                          final double[] c, final double[][] a, final double[] b,
147                                          final RungeKuttaStepInterpolator prototype,
148                                          final double   minStep, final double maxStep,
149                                          final double[] vecAbsoluteTolerance,
150                                          final double[] vecRelativeTolerance) {
151 
152     super(name, minStep, maxStep, vecAbsoluteTolerance, vecRelativeTolerance);
153 
154     this.fsal      = fsal;
155     this.c         = c;
156     this.a         = a;
157     this.b         = b;
158     this.prototype = prototype;
159 
160     exp = -1.0 / getOrder();
161 
162     // set the default values of the algorithm control parameters
163     setSafety(0.9);
164     setMinReduction(0.2);
165     setMaxGrowth(10.0);
166   }
167 
168   /** Get the order of the method.
169    * @return order of the method
170    */
171   public abstract int getOrder();
172 
173   /** Get the safety factor for stepsize control.
174    * @return safety factor
175    */
176   public double getSafety() {
177     return safety;
178   }
179 
180   /** Set the safety factor for stepsize control.
181    * @param safety safety factor
182    */
183   public void setSafety(final double safety) {
184     this.safety = safety;
185   }
186 
187   /** {@inheritDoc} */
188   @Override
189   public void integrate(final ExpandableStatefulODE equations, final double t)
190       throws NumberIsTooSmallException, DimensionMismatchException,
191              MaxCountExceededException, NoBracketingException {
192 
193     sanityChecks(equations, t);
194     setEquations(equations);
195     final boolean forward = t > equations.getTime();
196 
197     // create some internal working arrays
198     final double[] y0  = equations.getCompleteState();
199     final double[] y = y0.clone();
200     final int stages = c.length + 1;
201     final double[][] yDotK = new double[stages][y.length];
202     final double[] yTmp    = y0.clone();
203     final double[] yDotTmp = new double[y.length];
204 
205     // set up an interpolator sharing the integrator arrays
206     final RungeKuttaStepInterpolator interpolator = (RungeKuttaStepInterpolator) prototype.copy();
207     interpolator.reinitialize(this, yTmp, yDotK, forward,
208                               equations.getPrimaryMapper(), equations.getSecondaryMappers());
209     interpolator.storeTime(equations.getTime());
210 
211     // set up integration control objects
212     stepStart         = equations.getTime();
213     double  hNew      = 0;
214     boolean firstTime = true;
215     initIntegration(equations.getTime(), y0, t);
216 
217     // main integration loop
218     isLastStep = false;
219     do {
220 
221       interpolator.shift();
222 
223       // iterate over step size, ensuring local normalized error is smaller than 1
224       double error = 10;
225       while (error >= 1.0) {
226 
227         if (firstTime || !fsal) {
228           // first stage
229           computeDerivatives(stepStart, y, yDotK[0]);
230         }
231 
232         if (firstTime) {
233           final double[] scale = new double[mainSetDimension];
234           if (vecAbsoluteTolerance == null) {
235               for (int i = 0; i < scale.length; ++i) {
236                 scale[i] = scalAbsoluteTolerance + scalRelativeTolerance * JdkMath.abs(y[i]);
237               }
238           } else {
239               for (int i = 0; i < scale.length; ++i) {
240                 scale[i] = vecAbsoluteTolerance[i] + vecRelativeTolerance[i] * JdkMath.abs(y[i]);
241               }
242           }
243           hNew = initializeStep(forward, getOrder(), scale,
244                                 stepStart, y, yDotK[0], yTmp, yDotK[1]);
245           firstTime = false;
246         }
247 
248         stepSize = hNew;
249         if (forward) {
250             if (stepStart + stepSize >= t) {
251                 stepSize = t - stepStart;
252             }
253         } else {
254             if (stepStart + stepSize <= t) {
255                 stepSize = t - stepStart;
256             }
257         }
258 
259         // next stages
260         for (int k = 1; k < stages; ++k) {
261 
262           for (int j = 0; j < y0.length; ++j) {
263             double sum = a[k-1][0] * yDotK[0][j];
264             for (int l = 1; l < k; ++l) {
265               sum += a[k-1][l] * yDotK[l][j];
266             }
267             yTmp[j] = y[j] + stepSize * sum;
268           }
269 
270           computeDerivatives(stepStart + c[k-1] * stepSize, yTmp, yDotK[k]);
271         }
272 
273         // estimate the state at the end of the step
274         for (int j = 0; j < y0.length; ++j) {
275           double sum    = b[0] * yDotK[0][j];
276           for (int l = 1; l < stages; ++l) {
277             sum    += b[l] * yDotK[l][j];
278           }
279           yTmp[j] = y[j] + stepSize * sum;
280         }
281 
282         // estimate the error at the end of the step
283         error = estimateError(yDotK, y, yTmp, stepSize);
284         if (error >= 1.0) {
285           // reject the step and attempt to reduce error by stepsize control
286           final double factor =
287               JdkMath.min(maxGrowth,
288                            JdkMath.max(minReduction, safety * JdkMath.pow(error, exp)));
289           hNew = filterStep(stepSize * factor, forward, false);
290         }
291       }
292 
293       // local error is small enough: accept the step, trigger events and step handlers
294       interpolator.storeTime(stepStart + stepSize);
295       System.arraycopy(yTmp, 0, y, 0, y0.length);
296       System.arraycopy(yDotK[stages - 1], 0, yDotTmp, 0, y0.length);
297       stepStart = acceptStep(interpolator, y, yDotTmp, t);
298       System.arraycopy(y, 0, yTmp, 0, y.length);
299 
300       if (!isLastStep) {
301 
302           // prepare next step
303           interpolator.storeTime(stepStart);
304 
305           if (fsal) {
306               // save the last evaluation for the next step
307               System.arraycopy(yDotTmp, 0, yDotK[0], 0, y0.length);
308           }
309 
310           // stepsize control for next step
311           final double factor =
312               JdkMath.min(maxGrowth, JdkMath.max(minReduction, safety * JdkMath.pow(error, exp)));
313           final double  scaledH    = stepSize * factor;
314           final double  nextT      = stepStart + scaledH;
315           final boolean nextIsLast = forward ? (nextT >= t) : (nextT <= t);
316           hNew = filterStep(scaledH, forward, nextIsLast);
317 
318           final double  filteredNextT      = stepStart + hNew;
319           final boolean filteredNextIsLast = forward ? (filteredNextT >= t) : (filteredNextT <= t);
320           if (filteredNextIsLast) {
321               hNew = t - stepStart;
322           }
323       }
324     } while (!isLastStep);
325 
326     // dispatch results
327     equations.setTime(stepStart);
328     equations.setCompleteState(y);
329 
330     resetInternalState();
331   }
332 
333   /** Get the minimal reduction factor for stepsize control.
334    * @return minimal reduction factor
335    */
336   public double getMinReduction() {
337     return minReduction;
338   }
339 
340   /** Set the minimal reduction factor for stepsize control.
341    * @param minReduction minimal reduction factor
342    */
343   public void setMinReduction(final double minReduction) {
344     this.minReduction = minReduction;
345   }
346 
347   /** Get the maximal growth factor for stepsize control.
348    * @return maximal growth factor
349    */
350   public double getMaxGrowth() {
351     return maxGrowth;
352   }
353 
354   /** Set the maximal growth factor for stepsize control.
355    * @param maxGrowth maximal growth factor
356    */
357   public void setMaxGrowth(final double maxGrowth) {
358     this.maxGrowth = maxGrowth;
359   }
360 
361   /** Compute the error ratio.
362    * @param yDotK derivatives computed during the first stages
363    * @param y0 estimate of the step at the start of the step
364    * @param y1 estimate of the step at the end of the step
365    * @param h  current step
366    * @return error ratio, greater than 1 if step should be rejected
367    */
368   protected abstract double estimateError(double[][] yDotK,
369                                           double[] y0, double[] y1,
370                                           double h);
371 }