/[ascend]/trunk/models/basemodel.a4l
ViewVC logotype

Contents of /trunk/models/basemodel.a4l

Parent Directory Parent Directory | Revision Log Revision Log


Revision 669 - (show annotations) (download) (as text)
Wed Jun 21 07:00:45 2006 UTC (17 years, 11 months ago) by johnpye
File MIME type: text/x-ascend
File size: 19211 byte(s)
Merged changes from DAE branch (revisions 702 to 819) back into trunk.
This adds the Integration API to the ASCEND solver (in base/generic).
Also provides pre-alpha support for 'IDA' from the SUNDIALS suite, a DAE solver.
Many other minor code clean-ups, including adoption of new 'ASC_NEW' and friends (to replace 'ascmalloc')
Added some very sketchy stuff providing 'DIFF(...)' syntax, although it is anticipated that this will be removed.
1 PROVIDE "basemodel.a4l";
2
3 (*
4 * basemodel.a4l
5 * by Benjamin A. Allan
6 * Part of the ASCEND Library
7 * $Date: 1998/06/17 14:15:10 $
8 * $Revision: 1.3 $
9 * $Author: mthomas $
10 * $Source: /afs/cs.cmu.edu/project/ascend/Repository/models/basemodel.a4l,v $
11 *
12 * This file is part of the ASCEND Modeling Library.
13 *
14 * Copyright (C) 1998 Carnegie Mellon University
15 *
16 * The ASCEND Modeling Library is free software; you can redistribute
17 * it and/or modify it under the terms of the GNU General Public
18 * License as published by the Free Software Foundation; either
19 * version 2 of the License, or (at your option) any later version.
20 *
21 * The ASCEND Modeling Library is distributed in hope that it
22 * will be useful, but WITHOUT ANY WARRANTY; without even the implied
23 * warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
24 * See the GNU General Public License for more details.
25 *
26 * You should have received a copy of the GNU General Public License
27 * along with the program; if not, write to the Free Software
28 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139 USA. Check
29 * the file named COPYING.
30 *)
31
32 (*============================================================================*
33
34 B A S E M O D E L . A 4 L
35 -------------------------
36
37 AUTHOR: Benjamin A. Allan
38
39 DATES: 03/98 - Original Code
40
41 CONTENTS: Basic definitions cmu libraries and standard
42 methods.
43 This file is necessary for all
44 other CMU authored ASCEND models to work in ASCEND IV.
45
46 *============================================================================*)
47
48 MODEL catch_Word_model (* Bill Gates sacrificial goat *);
49 (* This MODEL does nothing.
50 * Normally catch_Word_model just gets parsed and ignored.
51 *
52 * If the user has tried to read a Microsoft Word binary file, Tcl file,
53 * or some other piece of junk as if it were an ASCEND MODEL
54 * source file, then catch_Word_model will die on an unknown
55 * syntax error.
56 * While catch_Word_model is dying the parser returns a good
57 * starting condition.
58 *
59 * Here is the message of recovery when this MODEL fails:
60 Asc-Error: Model definition "catch_Word_model" abandoned due to syntax errors.
61 Asc-Error: Rejected "catch_Word_model" at line basemodel.a4l:62.
62 *)
63 END catch_Word_model;
64
65 (* First define the standard methods, or stand-ins which will tell
66 * us when a standard method has not been written.
67 *)
68 ADD METHODS IN DEFINITION MODEL;
69
70 METHOD ClearAll;
71
72 NOTES 'purpose' SELF {
73 This method finds anything that is a solver_var and changes
74 the .fixed flag on the var to FALSE.
75
76 This method does not change .included flags on relations
77 or return boolean, integer, or symbol variables to a
78 default value.
79 } END NOTES;
80
81 EXTERNAL asc_free_all_variables(SELF);
82 END ClearAll;
83
84 (*
85 * Geniuses make more mistakes than anyone else -- because they
86 * try more things that anyone else. Part (perhaps a very large
87 * part) of what makes a genius different from the rest of
88 * humanity is that they quickly recognize their own mistakes
89 * and move on to try something else before anyone notices
90 * they screwed up! Solving a problem as far and as fast as you
91 * can, then going back to criticize every aspect of the solution
92 * with an eye to improving it is how you usually discover right answers.
93 *
94 * The authors of ASCEND (geniuses or not we'll
95 * leave to our users to decide) have found that it is
96 * best to do things such as writing mathematical MODELs and
97 * writing mathematical modeling software in ways which
98 * makes our mistakes (or your mistakes) very easy to detect.
99 *
100 * Below we describe a methodology (pun intended) which can
101 * help make anyone who can solve a quadratic equation
102 * a mathematical modeling expert. This methodology helps
103 * you to avoid screwing up and to find out about it when you have.
104 *
105 * The ASCEND system will not force you to write standard
106 * methods in your models. :-( METHODs of the sort we advocate
107 * here make your MODELs much easier to use and
108 * much more reliable. They pay off in the short run as well
109 * as the long run. These are _guidelines_, not _laws_: real
110 * genius requires knowing when to color outside the lines. :-)
111 *
112 * If you do not write the standard methods, your MODEL will
113 * inherit the ones given here. The "ClearAll" and "reset"
114 * methods here will work for you if you followed the guidelines.
115 * The other methods contain STOP statements which will warn you
116 * that you have skipped something important, should you accidentally
117 * end up calling one of them.
118 *
119 * The following methods should be redefined by each
120 * reusable library MODEL that REFINES this root MODEL.
121 * Models that do not supply proper versions of these
122 * (and possibly other) methods are very hard to reuse.
123 *
124 * The premise of this method design is that we can
125 * write the _self methods incrementally, building on the
126 * already tested methods of previous MODEL parts we are
127 * reusing. In this way we never have to write a single huge method
128 * that directly checks 100s of variables in a hierarchy.
129 *
130 * The _all methods are methods which simply "top off" the
131 * _self methods. With an _all method, you can treat
132 * just a part of a larger simulation already built
133 * as a self-contained simulation.
134 *
135 *)
136
137 (*
138 * Usually discovery of the information you need to write the methods
139 * proceeds in the order that they appear below:
140 * check, default, specify, bound, scale.
141 *)
142
143 METHOD check_self;
144 NOTES 'purpose' SELF {
145 This method should be written first, though it is run
146 last. Just like they taught you in elementary school,
147 always check your work. Start by defining criteria for a
148 successful solution that will not be included in the
149 equations solved and then computing those in this method.
150 As you develop your MODEL, you should expect to revise the
151 check method from time to time, if you are learning
152 anything about the MODEL. We frequently change our
153 definition of success.
154
155 When a mathematical MODEL is solved, the assumptions that
156 went into writing (deriving) the equations should be
157 checked. Usually there are redundant equations available
158 (more than one way to state the physical MODEL
159 mathematically). These should be used to check the
160 particularly tricky bits of the MODEL.
161
162 Check that the physical or intuitive (qualitative)
163 relationships among variables ch you expect to hold are
164 TRUE, especially if you have not written such relationships
165 in terms of inequalities in the MODEL equations.
166
167 In some models, checking the variable values against
168 absolute physical limits (temperature > 0{K} and
169 temperature < Tcritical for example) may be all that is
170 necessary or possible. Do not check variable values against
171 their .lower_bound or .upper_bound, as any decent algebraic
172 solver or modeling system will do this for you.
173
174 If a check fails, use a STOP or ERROR statement to notify
175 yourself (or you MODEL using customer) that the solution
176 may be bogus.
177
178 Currently only STOP is implemented.
179 STOP raises an error signal and issues an error message;
180 STOP normally also stops further execution of the method
181 and returns control to a higher level, though there are
182 interactive tools to force method execution to continue.
183 STOP does not crash the ASCEND system.
184
185 } END NOTES;
186
187 STOP {Error! Standard method "check_self" called but not written in MODEL.};
188
189 END check_self;
190
191 METHOD check_all;
192
193 NOTES 'purpose' SELF {
194 When solving only a part of a simulation, it is necessary to check
195 the models and variables passed into the part as well as the
196 locally defined parts and variables. This method should check
197 all the received models and variables, then check the local
198 stuff.
199 } END NOTES;
200
201 STOP {Error! Standard method "check_all" called but not written in MODEL.};
202 RUN check_self; (* intentionally _second_ *)
203
204 END check_all;
205
206 METHOD defaults;
207 (*
208 * This is a kluge for interfaces that still think of
209 * 'defaults' as the standard method.
210 *)
211 RUN default_self;
212 STOP {GUI (or somebody) called non-standard method defaults. Call forwarded to default_self before stopping here.};
213 END defaults;
214
215 METHOD on_load;
216 NOTES 'purpose' SELF {
217 This method adds improved ability to perform stuff when a model is first loaded.
218 By default, just 'default_self' will be run (this was the previous behaviour).
219 Any model that has an on_load method can override this behaviour however.
220 Note that this behaviour applies only in the C++/python interface at this stage.
221 } END NOTES;
222 RUN default_self;
223 END on_load;
224
225 METHOD default_self;
226 NOTES 'purpose' SELF {
227 This method should set default values for any variables
228 declared locally (IS_A) to the MODEL. It should run
229 default_self on _all_ the models that are declared locally
230 (with IS_A) in the MODEL also. If the atoms you use to
231 define your variables have a suitable default already, then
232 you do not need to assign them a default in this method.
233
234 This method should not run any methods on MODEL parts that
235 come via WILL_BE in the definition's parameter list. This
236 method also should not change the values of variables that
237 are passed in through the parameter list.
238
239 Sometimes there will be nothing for this method to do.
240 Define it anyway, leaving it empty.
241
242 When a top-level simulation is built by the compiler, this
243 method will be run at the end of compilation by the
244 compiler.
245 } END NOTES;
246
247 STOP {Error! Standard method "default_self" called but not written in MODEL.};
248 END default_self;
249
250 METHOD default_all;
251
252 NOTES 'purpose' SELF {
253 This method assumes that the arguments to the MODEL
254 instance have not been properly initialized, as is
255 frequently the case in one-off modeling efforts. This
256 method should run the default_self method on each of the
257 parts received through the parameter list and should give
258 appropriate default values to any variables received
259 through the parameter list. After these have been done, it
260 should then call default_self to take care of all locally
261 declared default needs.
262 } END NOTES;
263
264 STOP {Error! Standard method "default_all" called but not written in MODEL.};
265 RUN default_self;
266 END default_all;
267
268 METHOD specify;
269 NOTES 'purpose' SELF {
270 * Assuming ClearAll has been run on the MODEL, this method
271 * should get the MODEL to a condition called 'square':
272 * the case where there are as many variables with .fixed == FALSE
273 * as there equations available to compute them.
274 * This is one of the hardest tasks ever invented by mathematicians
275 * if you go about it in the wrong way. We think we know the right way.
276 *
277 * Actually, 'square' is a bit trickier to achieve
278 * than simply counting equations and variables.
279 * Solver, such as QRSlv in ASCEND, may help greatly with the bookkeeping.
280 *
281 The general approach is to:
282
283 (1) Run "specify" for all the parts (both passed in and locally defined)
284 that are not passed on into other parts.
285
286 (2) Fix up (by tweaking .fixed flags on variables) any difficulties
287 that arise when parts compete to calculate the same variable.
288
289 (3) Use the remaining new local variables to take care of any leftover
290 equations among the parts and any new equations written locally.
291
292 At all steps 1-3
293 Pay special attention to indexed variables used in
294 indexed equations; frequently you must fix or free N or
295 N-1 variables of a set sized N, if there are N matching equations.
296 In general, if you think you have specify correctly written, change
297 the sizes of all the sets in your MODEL by one and then by two
298 members. If your specify method still works, you are using sets
299 correctly.
300
301 When writing models that combine parts which do not share
302 very well, or which both try to compute the same variable
303 in different ways, it may even be necessary to write a WHEN
304 statement to selectively TURN OFF the conflicting equations
305 or MODEL fragments. An object or equation USEd in a WHEN
306 statement is turned off by default and becomes a part of
307 the solved MODEL only when the conditions of some CASE
308 which refers to that object are matched.
309
310 The setting of boolean, integer, and symbol variables which
311 are controlling conditions of WHEN and SWITCH statements
312 should be taken care of in the specify method.
313
314 There is no 'one perfect "specify"' for all purposes. This
315 routine should merely define a reasonably useful base
316 configuration of the MODEL.
317
318 Other specify_whatElseYouWant methods can (should) also be
319 written.
320
321 The name of a method is a communication tool. Please use
322 meaningful names as long as necessary to tell what the
323 method does. Avoid cryptic abbreviations and hyper-
324 specialized jargon known only to you and your three friends
325 when you are naming methods; however, do not shy away from
326 technical terms common to the engineering domain in which
327 you are modeling.
328
329 } END NOTES;
330
331 STOP {Error! Standard method "specify" called but not written in MODEL.};
332
333 END specify;
334
335 METHOD reset;
336 NOTES 'purpose' SELF {
337 This method gets the MODEL to some standard starting state,
338 though not necessarily the most useful starting state for a
339 particular application. In Chem. Eng. terms, this method
340 establishes a base case.
341
342 There is no 'one perfect "reset"' for all purposes. This
343 routine should merely define a reasonably useful base
344 configuration of the MODEL.
345
346 Other reset_whatElseYouWant methods can (should) also be
347 written.
348
349 Normally you do not need to write this method: your models
350 will inherit this one unless you override it (redefine it)
351 in your MODEL.
352 }
353 END NOTES;
354
355 RUN ClearAll;
356 RUN specify;
357
358 END reset;
359
360 METHOD bound_self;
361 NOTES 'purpose' SELF {
362 Much of the art of nonlinear physical modeling is in
363 bounding the solution.
364
365 This method should update the bounds on _locally_ defined
366 (IS_A) variables and IS_A defined MODEL parts. Updating
367 bounds requires some care. For example, the bounds on
368 fractions frequently don't need updating.
369
370 A common formula for updating bounds is to define a region
371 around the current value of the variable. A linear region
372 size formula, as an example, would be:
373
374 v.upper_bound := v + boundwidth * v.nominal;
375 v.lower_bound := v - boundwidth * v.nominal;
376
377 Care must be taken that such a formula does not move the
378 bounds (particularly lower bounds) out so far as to allow
379 non-physical solutions. Logarithmic bounding regions are
380 also simple to calculate.
381
382 Here boundwidth IS_A bound_width;
383 boundwidth is a real variable (but not a solver_var) or a
384 value you can use to determine how much "wiggle-room" you
385 want to give a solver. Small powers of 4 and 10 are usually
386 good values of boundwidth.
387
388 Too small a boundwidth can cut off the portion of number
389 space where the solution is found. Too large a bound width
390 can allow solvers to wander for great distances in
391 uninteresting regions of the number space.
392
393 This method should not bound variables passed into the
394 MODEL definition or parts passed into the definition.
395 } END NOTES;
396
397 STOP {Error! Standard method "bound_self" called but not written in MODEL.};
398 END bound_self;
399
400 METHOD bound_all;
401 NOTES 'purpose' SELF {
402 This method should be like bound_self except that it bounds the
403 passed in variables and calls bound_self on the passed in parts.
404 It should then call bound_self.
405 } END NOTES;
406
407 STOP {Error! Standard method "bound_all" called but not written in MODEL.};
408 RUN bound_self;
409 END bound_all;
410
411 METHOD scale_self;
412 NOTES 'purpose' SELF {
413 Most nonlinear (and many linear) models cannot be solved without
414 proper scaling of the variables.
415
416 This method should reset the .nominal value on every real
417 variable in need of scaling. It should then call the
418 scale_self method on all the locally defined (IS_A) parts
419 of the MODEL. 0.0 is the worst possible nominal value. A
420 proper nominal is one such that you expect at the solution
421 the quantity
422
423 abs(variable/(variable.nominal))
424
425 to be around 1 (in the range of [0.1..10] or [0.01..100]).
426
427 Variables (like fractions) bounded such that they cannot be
428 too far away from 1.0 in magnitude probably don't need scaling
429 most of the time if they are also bounded away from 0.0.
430
431 Some solvers, but not all, will attempt to scale the
432 equations and variables by heuristic matrix-based methods.
433 This works, but inconsistently; user-defined scaling is
434 generaly much superior.
435
436 ASCEND makes it easy to do. You scale the variables, which
437 can only be done well by knowing something about where the
438 solution is going to be found (by being an engineer, for
439 example.) Then ASCEND can calculate an appropriate
440 equation-scaling by efficient symbolic methods.
441
442 This method should not change the scaling of models and
443 variables that are received through the parameter list of
444 the MODEL.
445 } END NOTES;
446
447 STOP {Error! Standard method "scale_self" called but not written in MODEL.};
448 END scale_self;
449
450 METHOD scale_all;
451 NOTES 'purpose' SELF {
452 This method should be like scale_self above except that it also
453 should scale the variables and models received through the
454 parameter list. It should then call scale_self to take care of
455 the local variables and models.
456 } END NOTES;
457
458 STOP {Error! Standard method "scale_all" called but not written in MODEL.};
459 RUN scale_self;
460 END scale_all;
461 END METHODS;
462
463 MODEL cmumodel();
464 NOTES
465 'purpose' SELF {
466 This MODEL does nothing except provide a root
467 for a collection of loosely related models.
468 If it happens to reveal a few bugs in the software,
469 and perhaps masks others, well, what me worry?
470 BAA, 8/97.
471 }
472 'methods' SELF {
473 This MODEL also provides a hook to put in candidates for
474 becoming ascBuiltin global methods. Global methods may be
475 overridden by local definitions.
476 BAA, 3/98.
477 }
478 END NOTES;
479
480 END cmumodel;
481
482 MODEL testcmumodel();
483 (*
484 * All CMU test models, of whatever sort should ultimately be
485 * rooted here or be a final refinement of a reusable MODEL.
486 *)
487 METHODS
488 METHOD values;
489 (*
490 * In a final application MODEL, you should record at least one set of
491 * input values (values of the fixed variables and guesses of key
492 * solved-for variables) that leads to a good solution.
493 * Do this so noone need reinvent that set the next time
494 * you use the MODEL or someone picks the MODEL up after you.
495 *)
496 STOP {Error! Standard method "values" called but not written in MODEL.};
497 END values;
498
499 METHOD specify;
500 STOP {Error! Standard method "specify" called but not written in test MODEL.};
501 END specify;
502 (*
503 METHOD ClearAll;
504 EXTERNAL asc_free_all_variables(SELF);
505 END ClearAll;
506
507 METHOD reset;
508 (* This method gets the MODEL to some standard starting state,
509 * though not necessarily the most useful starting state for
510 * a particular application. In Chem. Eng. terms, this method
511 * establishes a base case.
512 * There is no 'one perfect "reset"' for all purposes. This
513 * routine should merely define a reasonably useful base configuration
514 * of the MODEL.
515 * Other reset_whatElseYouWant methods can (should) also be
516 * written.
517 *
518 * Normally you do not need to write this method: your models
519 * will inherit this one unless you override it (redefine it)
520 * in your MODEL.
521 *)
522 RUN ClearAll;
523 RUN specify;
524 END reset;
525 *)
526 END testcmumodel;
527
528 MODEL your_site_models();
529 (* if you create a library to share with the net which is
530 * not just an end application of Carnegie Mellon models,
531 * please create an empy root MODEL such as this and use
532 * it as the origin of your library in the same way that
533 * we use cmumodel as the origin of our libraries.
534 * Thank you.
535 *)
536 END your_site_models;

john.pye@anu.edu.au
ViewVC Help
Powered by ViewVC 1.1.22