Mercurial > octave-nkf
comparison scripts/ode/private/integrate_adaptive.m @ 20568:fcb792acab9b
Moving ode45, odeset, odeget, and levenshtein from odepkg to core.
* libinterp/corefcn/levenshtein.cc: move function from odepkg into core
* libinterp/corefcn/module.mk: include levenshtein.cc
* scripts/ode: move ode45, odeset, odeget, and all dependencies
from odepkg into core
* scripts/module.mk: include them
* doc/interpreter/diffeq.txi: add documentation for ode45,
odeset, odeget
* NEWS: announce functions included with this changeset
* scripts/help/__unimplemented__.m: removed new functions
author | jcorno <jacopo.corno@gmail.com> |
---|---|
date | Thu, 24 Sep 2015 12:58:46 +0200 |
parents | |
children | 6256f6e366ac |
comparison
equal
deleted
inserted
replaced
20567:2480bbcd1333 | 20568:fcb792acab9b |
---|---|
1 ## Copyright (C) 2013, Roberto Porcu' <roberto.porcu@polimi.it> | |
2 ## OdePkg - A package for solving ordinary differential equations and more | |
3 ## | |
4 ## This program is free software; you can redistribute it and/or modify | |
5 ## it under the terms of the GNU General Public License as published by | |
6 ## the Free Software Foundation; either version 2 of the License, or | |
7 ## (at your option) any later version. | |
8 ## | |
9 ## This program is distributed in the hope that it will be useful, | |
10 ## but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 ## MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 ## GNU General Public License for more details. | |
13 ## | |
14 ## You should have received a copy of the GNU General Public License | |
15 ## along with this program; If not, see <http://www.gnu.org/licenses/>. | |
16 | |
17 | |
18 ## -*- texinfo -*- | |
19 ## @deftypefn {Command} {[@var{t}, @var{y}] =} | |
20 ## integrate_adaptive (@var{@@stepper}, @var{order}, @var{@@fun}, @var{tspan}, | |
21 ## @var{x0}, @var{options}) | |
22 ## | |
23 ## This function file can be called by a ODE solver function in order to | |
24 ## integrate the set of ODEs on the interval @var{[t0,t1]} with an | |
25 ## adaptive timestep. | |
26 ## | |
27 ## This function must be called with two output arguments: @var{t} and @var{y}. | |
28 ## Variable @var{t} is a column vector and contains the time stamps, instead | |
29 ## @var{y} is a matrix in which each column refers to a different unknown | |
30 ## of the problem and the rows number is the same of @var{t} rows number so | |
31 ## that each row of @var{y} contains the values of all unknowns at the time | |
32 ## value contained in the corresponding row in @var{t}. | |
33 ## | |
34 ## The first input argument must be a function_handle or an inline function | |
35 ## representing the stepper, that is the function responsible for step-by-step | |
36 ## integration. This function discriminates one method from the others. | |
37 ## | |
38 ## The second input argument is the order of the stepper. It is needed | |
39 ## to compute the adaptive timesteps. | |
40 ## | |
41 ## The third input argument is a function_handle or an inline function that | |
42 ## defines the set of ODE: | |
43 ## @ifhtml | |
44 ## @example | |
45 ## @math{y' = f(t,y)} | |
46 ## @end example | |
47 ## @end ifhtml | |
48 ## @ifnothtml | |
49 ## @math{y' = f(t,y)}. | |
50 ## @end ifnothtml | |
51 ## | |
52 ## The fourth input argument is the time vector which defines integration | |
53 ## interval, that is @var{[tspan(1),tspan(end)]} and all the intermediate | |
54 ## elements are taken as times at which the solution is required. | |
55 ## | |
56 ## The fifth argument represents the initial conditions for the ODEs and the | |
57 ## last input argument contains some options that may be needed for the stepper. | |
58 ## | |
59 ## @end deftypefn | |
60 ## | |
61 ## @seealso{integrate_const, integrate_n_steps} | |
62 | |
63 function solution = integrate_adaptive (stepper, order, func, tspan, x0, options) | |
64 | |
65 solution = struct; | |
66 | |
67 ## first values for time and solution | |
68 t = tspan(1); | |
69 x = x0(:); | |
70 | |
71 ## get first initial timestep | |
72 dt = odeget (options, "InitialStep", | |
73 starting_stepsize (order, func, t, x, options.AbsTol, | |
74 options.RelTol, options.vnormcontrol), | |
75 "fast_not_empty"); | |
76 vdirection = odeget (options, "vdirection", [], "fast"); | |
77 if (sign (dt) != vdirection) | |
78 dt = -dt; | |
79 endif | |
80 dt = vdirection * min (abs (dt), options.MaxStep); | |
81 | |
82 ## set parameters | |
83 k = length (tspan); | |
84 counter = 2; | |
85 comp = 0.0; | |
86 tk = tspan(1); | |
87 options.comp = comp; | |
88 | |
89 ## factor multiplying the stepsize guess | |
90 facmin = 0.8; | |
91 fac = 0.38^(1/(order+1)); ## formula taken from Hairer | |
92 t_caught = false; | |
93 | |
94 | |
95 ## Initialize the OutputFcn | |
96 if (options.vhaveoutputfunction) | |
97 if (options.vhaveoutputselection) | |
98 solution.vretout = x(options.OutputSel,end); | |
99 else | |
100 solution.vretout = x; | |
101 endif | |
102 feval (options.OutputFcn, tspan, solution.vretout, | |
103 "init", options.vfunarguments{:}); | |
104 endif | |
105 | |
106 ## Initialize the EventFcn | |
107 if (options.vhaveeventfunction) | |
108 odepkg_event_handle (options.Events, t(end), x, | |
109 "init", options.vfunarguments{:}); | |
110 endif | |
111 | |
112 solution.vcntloop = 2; | |
113 solution.vcntcycles = 1; | |
114 vcntiter = 0; | |
115 solution.vunhandledtermination = true; | |
116 solution.vcntsave = 2; | |
117 | |
118 z = t; | |
119 u = x; | |
120 | |
121 k_vals = feval (func, t , x, options.vfunarguments{:}); | |
122 | |
123 while (counter <= k) | |
124 facmax = 1.5; | |
125 | |
126 ## compute integration step from t to t+dt | |
127 [s, y, y_est, k_vals] = stepper (func, z(end), u(:,end), | |
128 dt, options, k_vals); | |
129 | |
130 if (options.vhavenonnegative) | |
131 x(options.NonNegative,end) = abs (x(options.NonNegative,end)); | |
132 y(options.NonNegative,end) = abs (y(options.NonNegative,end)); | |
133 y_est(options.NonNegative,end) = abs (y_est(options.NonNegative,end)); | |
134 endif | |
135 | |
136 if (options.vhaveoutputfunction && options.vhaverefine) | |
137 vSaveVUForRefine = u(:,end); | |
138 endif | |
139 | |
140 err = AbsRel_Norm (y(:,end), u(:,end), options.AbsTol, options.RelTol, | |
141 options.vnormcontrol, y_est(:,end)); | |
142 | |
143 ## solution accepted only if the error is less or equal to 1.0 | |
144 if (err <= 1) | |
145 | |
146 [tk, comp] = kahan (tk, comp, dt); | |
147 options.comp = comp; | |
148 s(end) = tk; | |
149 | |
150 ## values on this interval for time and solution | |
151 z = [z(end);s]; | |
152 u = [u(:,end),y]; | |
153 | |
154 ## if next tspan value is caught, update counter | |
155 if ((z(end) == tspan(counter)) | |
156 || (abs (z(end) - tspan(counter)) / | |
157 (max (abs (z(end)), abs (tspan(counter)))) < 8*eps) ) | |
158 counter++; | |
159 | |
160 ## if there is an element in time vector at which the solution is required | |
161 ## the program must compute this solution before going on with next steps | |
162 elseif (vdirection * z(end) > vdirection * tspan(counter)) | |
163 ## initialize counter for the following cycle | |
164 i = 2; | |
165 while (i <= length (z)) | |
166 | |
167 ## if next tspan value is caught, update counter | |
168 if ((counter <= k) | |
169 && ((z(i) == tspan(counter)) | |
170 || (abs (z(i) - tspan(counter)) / | |
171 (max (abs (z(i)), abs (tspan(counter)))) < 8*eps)) ) | |
172 counter++; | |
173 endif | |
174 ## else, loop until there are requested values inside this subinterval | |
175 while ((counter <= k) | |
176 && (vdirection * z(i) > vdirection * tspan(counter))) | |
177 ## choose interpolation scheme according to order of the solver | |
178 switch order | |
179 case 1 | |
180 u_interp = linear_interpolation ([z(i-1) z(i)], | |
181 [u(:,i-1) u(:,i)], | |
182 tspan(counter)); | |
183 case 2 | |
184 if (! isempty (k_vals)) | |
185 der = k_vals(:,1); | |
186 else | |
187 der = feval (func, z(i-1) , u(:,i-1), | |
188 options.vfunarguments{:}); | |
189 endif | |
190 u_interp = quadratic_interpolation ([z(i-1) z(i)], | |
191 [u(:,i-1) u(:,i)], | |
192 der, tspan(counter)); | |
193 case 3 | |
194 u_interp = ... | |
195 hermite_cubic_interpolation ([z(i-1) z(i)], | |
196 [u(:,i-1) u(:,i)], | |
197 [k_vals(:,1) k_vals(:,end)], | |
198 tspan(counter)); | |
199 case 4 | |
200 ## if ode45 is used without local extrapolation this function | |
201 ## doesn't require a new function evaluation. | |
202 u_interp = dorpri_interpolation ([z(i-1) z(i)], | |
203 [u(:,i-1) u(:,i)], | |
204 k_vals, tspan(counter)); | |
205 case 5 | |
206 ## ode45 with Dormand-Prince scheme: | |
207 ## 4th order approximation of y in t+dt/2 as proposed by | |
208 ## Shampine in Lawrence, Shampine, "Some Practical | |
209 ## Runge-Kutta Formulas", 1986. | |
210 u_half = u(:,i-1) ... | |
211 + 1/2*dt*((6025192743/30085553152) * k_vals(:,1) | |
212 + (51252292925/65400821598) * k_vals(:,3) | |
213 - (2691868925/45128329728) * k_vals(:,4) | |
214 + (187940372067/1594534317056) * k_vals(:,5) | |
215 - (1776094331/19743644256) * k_vals(:,6) | |
216 + (11237099/235043384) * k_vals(:,7)); | |
217 u_interp = ... | |
218 hermite_quartic_interpolation ([z(i-1) z(i)], | |
219 [u(:,i-1) u_half u(:,i)], | |
220 [k_vals(:,1) k_vals(:,end)], | |
221 tspan(counter)); | |
222 | |
223 ## it is also possible to do a new function evaluation and use | |
224 ## the quintic hermite interpolator | |
225 ## f_half = feval (func, t+1/2*dt, u_half, | |
226 ## options.vfunarguments{:}); | |
227 ## u_interp = | |
228 ## hermite_quintic_interpolation ([z(i-1) z(i)], | |
229 ## [u(:,i-1) u_half u(:,i)], | |
230 ## [k_vals(:,1) f_half k_vals(:,end)], | |
231 ## tspan(counter)); | |
232 otherwise | |
233 warning ("High order interpolation not yet implemented: ", | |
234 "using cubic iterpolation instead"); | |
235 der(:,1) = feval (func, z(i-1) , u(:,i-1), | |
236 options.vfunarguments{:}); | |
237 der(:,2) = feval (func, z(i) , u(:,i), | |
238 options.vfunarguments{:}); | |
239 u_interp = ... | |
240 hermite_cubic_interpolation ([z(i-1) z(i)], | |
241 [u(:,i-1) u(:,i)], | |
242 der, tspan(counter)); | |
243 endswitch | |
244 | |
245 ## add the interpolated value of the solution | |
246 u = [u(:,1:i-1), u_interp, u(:,i:end)]; | |
247 | |
248 ## add the time requested | |
249 z = [z(1:i-1);tspan(counter);z(i:end)]; | |
250 | |
251 ## update counters | |
252 counter++; | |
253 i++; | |
254 endwhile | |
255 | |
256 ## if new time requested is not out of this interval | |
257 if ((counter <= k) | |
258 && (vdirection * z(end) > vdirection * tspan(counter))) | |
259 ## update the counter | |
260 i++; | |
261 else | |
262 ## stop the cycle and go on with the next iteration | |
263 i = length (z) + 1; | |
264 endif | |
265 | |
266 endwhile | |
267 endif | |
268 | |
269 if (mod (solution.vcntloop-1, options.OutputSave) == 0) | |
270 x = [x,u(:,2:end)]; | |
271 t = [t;z(2:end)]; | |
272 solution.vcntsave = solution.vcntsave + 1; | |
273 endif | |
274 solution.vcntloop = solution.vcntloop + 1; | |
275 vcntiter = 0; | |
276 | |
277 ## Call plot only if a valid result has been found, therefore this | |
278 ## code fragment has moved here. Stop integration if plot function | |
279 ## returns false | |
280 if (options.vhaveoutputfunction) | |
281 for vcnt = 0:options.Refine # Approximation between told and t | |
282 if (options.vhaverefine) # Do interpolation | |
283 vapproxtime = (vcnt + 1) / (options.Refine + 2); | |
284 vapproxvals = (1 - vapproxtime) * vSaveVUForRefine ... | |
285 + (vapproxtime) * y(:,end); | |
286 vapproxtime = s(end) + vapproxtime * dt; | |
287 else | |
288 vapproxvals = x(:,end); | |
289 vapproxtime = t(end); | |
290 endif | |
291 if (options.vhaveoutputselection) | |
292 vapproxvals = vapproxvals(options.OutputSel); | |
293 endif | |
294 vpltret = feval (options.OutputFcn, vapproxtime, | |
295 vapproxvals, [], options.vfunarguments{:}); | |
296 if (vpltret) # Leave refinement loop | |
297 break | |
298 endif | |
299 endfor | |
300 if (vpltret) # Leave main loop | |
301 solution.vunhandledtermination = false; | |
302 break | |
303 endif | |
304 endif | |
305 | |
306 ## Call event only if a valid result has been found, therefore this | |
307 ## code fragment has moved here. Stop integration if veventbreak is | |
308 ## true | |
309 if (options.vhaveeventfunction) | |
310 solution.vevent = odepkg_event_handle (options.Events, t(end), | |
311 x(:,end), [], options.vfunarguments{:}); | |
312 if (! isempty (solution.vevent{1}) | |
313 && solution.vevent{1} == 1) | |
314 t(solution.vcntloop-1,:) = solution.vevent{3}(end,:); | |
315 x(:,solution.vcntloop-1) = solution.vevent{4}(end,:)'; | |
316 solution.vunhandledtermination = false; | |
317 break | |
318 endif | |
319 endif | |
320 | |
321 else | |
322 facmax = 1.0; | |
323 endif | |
324 | |
325 ## Compute next timestep, formula taken from Hairer | |
326 err += eps; # adding an eps to avoid divisions by zero | |
327 dt = dt * min (facmax, max (facmin, | |
328 fac * (1 / err)^(1 / (order + 1)))); | |
329 dt = vdirection * min (abs (dt), options.MaxStep); | |
330 | |
331 ## Update counters that count the number of iteration cycles | |
332 solution.vcntcycles = solution.vcntcycles + 1; # Needed for cost statistics | |
333 vcntiter = vcntiter + 1; # Needed to find iteration problems | |
334 | |
335 ## Stop solving because in the last 1000 steps no successful valid | |
336 ## value has been found | |
337 if (vcntiter >= 5000) | |
338 error (["Solving has not been successful. The iterative", | |
339 " integration loop exited at time t = %f before endpoint at", | |
340 " tend = %f was reached. This happened because the iterative", | |
341 " integration loop does not find a valid solution at this time", | |
342 " stamp. Try to reduce the value of ''InitialStep'' and/or", | |
343 " ''MaxStep'' with the command ''odeset''.\n"], | |
344 s(end), tspan(end)); | |
345 endif | |
346 | |
347 ## if this is the last iteration, save the length of last interval | |
348 if (counter > k) | |
349 j = length (z); | |
350 endif | |
351 endwhile | |
352 | |
353 ## Check if integration of the ode has been successful | |
354 if (vdirection * z(end) < vdirection * tspan(end)) | |
355 if (solution.vunhandledtermination == true) | |
356 error ("OdePkg:InvalidArgument", | |
357 ["Solving has not been successful. The iterative", | |
358 " integration loop exited at time t = %f", | |
359 " before endpoint at tend = %f was reached. This may", | |
360 " happen if the stepsize grows smaller than defined in", | |
361 " vminstepsize. Try to reduce the value of ''InitialStep''", | |
362 " and/or ''MaxStep'' with the command ''odeset''.\n"], | |
363 z(end), tspan(end)); | |
364 else | |
365 warning ("OdePkg:InvalidArgument", | |
366 ["Solver has been stopped by a call of ''break'' in the main", | |
367 " iteration loop at time t = %f before endpoint at tend = %f ", | |
368 " was reached. This may happen because the @odeplot function", | |
369 " returned ''true'' or the @event function returned", | |
370 " ''true''.\n"], | |
371 z(end), tspan(end)); | |
372 endif | |
373 endif | |
374 | |
375 ## Compute how many values are out of time inerval | |
376 d = vdirection * t((end-(j-1)):end) > vdirection * tspan(end)*ones (j, 1); | |
377 f = sum (d); | |
378 | |
379 ## Remove not-requested values of time and solution | |
380 solution.t = t(1:end-f); | |
381 solution.x = x(:,1:end-f)'; | |
382 | |
383 endfunction | |
384 | |
385 ## Local Variables: *** | |
386 ## mode: octave *** | |
387 ## End: *** |