Rename *ll* and *ul* to ll and ul in poles-in-interval
[maxima.git] / src / series.lisp
blob5e007650526b5b4db8dceb050e2a2c4a0e59f784
1 ;;; -*- Mode: Lisp; Package: Maxima; Syntax: Common-Lisp; Base: 10 -*- ;;;;
2 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
3 ;;; The data in this file contains enhancements. ;;;;;
4 ;;; ;;;;;
5 ;;; Copyright (c) 1984,1987 by William Schelter,University of Texas ;;;;;
6 ;;; All rights reserved ;;;;;
7 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
8 ;;; (c) Copyright 1982 Massachusetts Institute of Technology ;;;
9 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
11 (in-package :maxima)
13 (macsyma-module series)
15 (declare-top (special var *a *m *c *index *gcd*
16 *roots *failures
17 *ratexp *var usexp *schatc-ans* *trigred
18 *form indl *noexpand))
20 (load-macsyma-macros rzmac)
22 ;;******************************************************************************
23 ;; driver stage
24 ;;******************************************************************************
26 (defmfun $powerseries (expr var pt)
27 (when (and (atom var) (not (symbolp var)))
28 (improper-arg-err var '$powerseries))
29 (powerseries expr var pt))
31 ;; An error that can be raised deep in the bowels of POWERSERIES that we'll
32 ;; catch and return a noun form.
33 (define-condition powerseries-expansion-error (error)
34 ((message :initarg :message :initform nil)))
36 (defun powerseries-expansion-error (&optional message &rest arguments)
37 (error 'powerseries-expansion-error
38 :message (when message (cons message arguments))))
40 ;; The top-level routine for $powerseries, which calls this function after
41 ;; spotting invalid arguments.
42 (defun powerseries (expr var pt)
43 (handler-case
44 (if (and (signp e pt) (symbolp var))
45 (seriesexpand* expr)
46 (cond
47 ((signp e pt)
48 (sbstpt expr 'x var var))
49 ((eq pt '$inf)
50 (sbstpt expr (m// 1 'x) var (div* 1 var)))
51 ((eq pt '$minf)
52 (sbstpt expr (m- (m// 1 'x)) var (m- (div* 1 var))))
54 (sbstpt expr (m+ 'x pt) var (simplifya (m- var pt) nil)))))
56 ;; If expansion fails, print a message in verbose mode but then return a
57 ;; noun form.
58 (powerseries-expansion-error (e)
59 (when $verbose
60 (mtell (intl:gettext "Failed to expand ~M with respect to ~M at ~M.~%")
61 expr var pt)
62 (with-slots (message) e
63 (when message
64 (terpri)
65 (finish-output)
66 (apply #'mtell message))))
67 `((%powerseries) ,expr ,var ,pt))))
69 ;; Return a list of the terms in the expression that are used as integration or
70 ;; differentiation variables.
71 (defun intdiff-vars-in-expr (expr)
72 (cond
73 ((atom expr) nil)
75 ;; Arguably, if this is a definite integral, the variable isn't free so we
76 ;; shouldn't return it. But maxima-substitute doesn't know about bound
77 ;; variables and this is a helper function to avoid silly substitutions.
78 ((eq (caar expr) '%integrate) (list (caddr expr)))
80 ((eq (caar expr) '%derivative)
81 (loop for x in (cddr expr) by #'cddr collecting x))
84 (reduce #'union (cdr expr) :key #'intdiff-vars-in-expr))))
86 ;; Series expand EXP after substituting in SEXP for VAR1, the main
87 ;; variable. SEXP should be an expression in 'X and 'X will be replaced by a
88 ;; gensym for the calculation. When the expansion is done, substitute USEXP for
89 ;; the gensym.
90 (defun sbstpt (exp sexp var1 usexp)
91 (let* ((var (gensym))
92 (sub-exp (maxima-substitute (subst var 'x sexp) var1 exp))
93 (expanded (seriesexpand* sub-exp)))
94 ;; If we've ended up with diff(foo, var) then we give up (we can't
95 ;; substitute arbitrary expressions for the differentiation / integration
96 ;; variable).
97 (if (memq var (intdiff-vars-in-expr expanded))
98 (powerseries-expansion-error
99 (intl:gettext "~
100 Couldn't make substitution to evaluate at the given point because the~%~
101 power series expansion contained the expansion variable as an~%~
102 integration / differentiation variable."))
103 (maxima-substitute usexp var expanded))))
105 (defun seriesexpand* (x)
106 (let ((*index (gensumindex)) *n *a *m *c
107 ($cauchysum t) ($ratsimpexpons t)
108 $ratexpand *infsumsimp *ratexp *trigred *noexpand)
109 (meval `(($declare) ,*index $integer))
111 (sp2expand (seriespass1 x))))
113 (defun out-of (e)
114 (let ((e (cond ((and (boundp '*var) *var)
115 (subst (list '(mexpt) *var *gcd*) var e))
116 (t e)))
117 (var (cond ((and (boundp '*var) *var)) (t var))))
118 (cond ((and (boundp 'usexp) usexp)
119 (subst usexp var e))
120 (t e))))
122 (defun show-exp (e)
123 (mtell "~%~%~M~%~%"
124 (list '(mlabel) nil (out-of e))))
126 (defun seriespass1 (e)
127 (let ((w (sratsimp (sp1 e))))
128 (when $verbose
129 (terpri)
130 (finish-output)
131 (mtell (intl:gettext "powerseries: first simplification returned ~%"))
132 (show-exp w))
136 ;;*****************************************************************************
137 ;; pass two expansion phase
138 ;;*****************************************************************************
141 (defun sp2expand (exp)
142 (cond ((or (free exp var) (atom exp)) exp)
143 ((mbagp exp) (cons (car exp) (mapcar #'sp2expand (cdr exp))))
144 ((eq (caar exp) 'mplus) (m+l (mapcar #'sp2expand (cdr exp)))) ;was below 'mtimes test--fixes powerseries(1+x^n,x,0)
145 ((sratp exp var) (ratexp exp))
146 ((eq (caar exp) 'mexpt) (sp2expt exp))
147 ((zl-get (caar exp) 'sp2) (sp2sub (sp2trig exp) (cadr exp)))
148 ((eq (caar exp) 'mtimes) (m*l (mapcar #'sp2expand (cdr exp))))
149 ((eq (caar exp) '%log) (sp2log (cadr exp)))
150 ((eq (caar exp) '%derivative) (sp2diff (cadr exp) (cddr exp)))
151 ((eq (caar exp) '%integrate)
152 (sp2integ (cadr exp) (caddr exp) (cdddr exp)))
153 ((member (caar exp) '(%product %sum) :test #'eq)
154 (list* (car exp) (sp2expand (cadr exp)) (cddr exp)))
155 (t (list '(%sum)
156 (m* (m^ var *index)
157 (m^ (list '(mfactorial) *index) -1)
158 (list '(%at) (list '(%derivative) exp var *index)
159 (list '(mequal) var 0)))
160 *index 0 '$inf))))
162 (defun sp2sub (s exp)
163 (unless (smono exp var)
164 (powerseries-expansion-error
165 (intl:gettext "Can't substitute the value~%~M~%~
166 into another expansion because it isn't a monomial in the~
167 expansion variable.")
168 (out-of exp)))
169 (maxima-substitute exp 'sp2var (simplify s)))
171 (defun ratexp (exp)
172 (let (*gcd*)
173 (if $verbose
174 (mtell (intl:gettext "powerseries: attempt rational function expansion of~%~M")
175 (list '(mlabel) nil exp)))
176 (multiple-value-call #'sratexpnd (numden exp))))
178 (defun sratexpnd (n d)
179 (let ((*schatc-ans* (list nil))
180 (*splist*)
181 ;; A pattern that matches cc*(c*x^m + a)^n
182 (linpat
183 '((mtimes) ((coefftt) (cc not-zero-free var))
184 ((mexpt) ((mplus) ((coeffpt)
185 (w m1 ((mexpt) (x equal var)
186 (m not-zero-free var)))
187 (c freevar))
188 ((coeffpp) (a freevar)))
189 (n not-zero-free var)))))
191 (declare (special *splist*))
192 (cond ((and (not (equal n 1)) (smono n var))
193 (m* n (sratexpnd 1 d)))
194 ((free d var)
195 (cond ((poly? n var)
196 (m// n d))
197 ((m1 n linpat)
198 (m// (srbinexpnd (cdr *schatc-ans*)) d))
200 (powerseries-expansion-error))))
201 ((smonop d var)
202 (cond ((mplusp n)
203 (m+l (mapcar #'(lambda (q) (div* q d)) (cdr n))))
204 (t (m// n d))))
205 ((not (equal 1 (setq *gcd* (ggcd (nconc (exlist n) (exlist d))))))
206 (sratsubst *gcd* n d))
208 ;; Rational expansion theorem for distinct roots. The main point of
209 ;; the POLY? test here is that it guarantees $HIPOW will give the
210 ;; correct degrees (in particular, it only allows known non-negative
211 ;; integer exponents)
212 ((and (poly? n var)
213 (poly? d var)
214 (> ($hipow d var) ($hipow n var))
215 (has-distinct-nonzero-roots-p d var))
216 (expand-distinct-roots n d))
218 ;; The SRBINEXPND call above dealt with expressions of the form
219 ;; cc*(c*x^m+a)^n. Here, we deal with b/(cc*(c*x^m+a)^n). If you
220 ;; explicitly write a polynomial like that, the simplifier will
221 ;; rewrite it as (..)^(-n) before it gets here, but things like
222 ;; 1/sqrt(x+1) won't have been rewritten successfully, so we catch
223 ;; them here.
224 ((and (free n var)
225 (prog2 (setq d (let (($ratfac t))
226 (ratdisrep ($rat (factor d) var))))
227 (m1 d linpat)))
228 ;; We had num/den and LINPAT matched den. We need to replace cc
229 ;; with num/cc and n with -n.
230 (setf (cdr (assoc 'n *schatc-ans*)) (m- (cdr (assoc 'n *schatc-ans*)))
231 (cdr (assoc 'cc *schatc-ans*)) (m// n (cdr (assoc 'cc *schatc-ans*))))
232 (srbinexpnd (cdr *schatc-ans*)))
235 ;; *RATEXP is set by RATEXPAND1, which we call to do the general
236 ;; expansion below. This check makes sure we can't end up recursing
237 ;; infinitely.
238 (when *ratexp
239 (powerseries-expansion-error))
241 ;; Look for a power of var (a zero root) as a term. We already know
242 ;; that d isn't a monomial, so we can only have a zero root if d is
243 ;; a product. We also know that d is not atomic because otherwise
244 ;; we'd have taken the (free d var) clause above.
245 (let ((zero-root
246 (and (eq (caar d) 'mtimes)
247 (find-if (lambda (factor)
248 (or (eq factor var)
249 (and (mexptp factor)
250 (eq (cadr factor) var))))
251 (cdr d)))))
252 (if (not zero-root)
253 (ratexpand1 n d)
254 (let ((other-factors (remove zero-root (cdr d)
255 :test #'eq :count 1)))
256 (m* (sratexpnd n (m*l other-factors))
257 `((mexpt) ,zero-root -1)))))))))
259 ; is a sum with index and bounds from psp2form
260 (defun psp2formp (exp)
261 (and (listp exp)
262 (listp (car exp))
263 (eq (caar exp) '%sum)
264 (eq (caddr exp) *index)
265 (eql (cadddr exp) 0)
266 (eq (cadr (cdddr exp)) '$inf)))
268 ;; A stripped down version of (sumcontract (intosum EXP)). Coalesce occurrences
269 ;; of ((%SUM) <FOO> *INDEX 0 '$INF), which are allowed to start with
270 ;; multiplicative constants. So something like
272 ;; sum(a(i), i, 0, inf) + 2*sum(b(i), i, 0, inf) + c
274 ;; turns into
276 ;; sum(a(i) + 2*b(i), i, 0, inf) + c
278 ;; This is good enough to collect up the multiple infinite sums that you get
279 ;; when expanding a power series by partial fractions. It's (obviously) not
280 ;; clever enough to rename index variables or adjust ranges.
282 (defun psp2foldsum (exp)
283 (when $verbose
284 (mtell (intl:gettext "powerseries: preparing to fold sums~%"))
285 (show-exp exp))
287 (multiple-value-bind (sums others)
288 (partition-by (lambda (e)
289 (or (psp2formp e)
290 (and (mtimesp e)
291 (psp2formp (caddr e)))))
292 (cdr exp))
293 (if (null sums)
294 ;; If there were no sums, just return the original expression.
296 ;; Otherwise contract the sums
297 (let ((contracted
298 (list '(%sum)
299 (m+l (mapcar (lambda (e)
300 (if (eq (caar e) 'mtimes)
301 (m* (cadr e) (cadr (caddr e)))
302 (cadr e)))
303 sums))
304 *index 0 '$inf)))
305 (if (null others)
306 contracted
307 (m+l (cons contracted others)))))))
309 ; solve returns a list: (soln mult soln mult ...)
310 ; distinct-nonzero-roots-p returns true if every
311 ; soln is not nonzero and every mult is 1
312 (defun distinct-nonzero-roots-p (roots)
313 (or (null roots)
314 (and
315 ;; Check all roots have multiplicity one and are nonzero.
316 (loop
317 for (root mult) on roots by #'cddr
318 always (and (eql 1 mult)
319 (eq ($sign `((mabs) ,(caddr root))) '$pos)))
321 ;; Now check that the roots are genuinely different from one another
322 ;; (eg. if I have two symbolic roots, A and B, then I don't need to know
323 ;; the values of A and B, but I do need to know that they aren't equal)
324 (loop
325 for rest on roots by #'cddr
326 always (loop
327 for b in (cddr rest) by #'cddr
328 with a = (car rest)
329 always (eq '$pos
330 ($sign `((mabs) ,(m- (caddr b) (caddr a))))))))))
332 ; returns t if polynomial poly in variable var has all distinct roots
333 (defun has-distinct-nonzero-roots-p (poly var)
334 (let ((*roots nil)
335 (*failures nil))
336 (solve poly var 1)
337 (and (not *failures)
338 (distinct-nonzero-roots-p *roots))))
340 ; Rational Expansion Theorem for Distinct Roots
341 ; Graham, Knuth, Patashnik, "Concrete Math" 2nd ed, p 340
343 ; If R(z) = P(z)/Q(z), where Q(z) = q0*(1-r_1*z)*...*(1-r_l*z) and the
344 ; numbers (r_1 ... r_l) are distinct, and if P(z) is a polynomial of degree less
345 ; than l, then
346 ; [z^n]R(z) = a_1*r_1^n + ... + a_l*r_l^n, where a_k = -r_k*P(1/r_k)/Q'(1/r_k)
347 (defun expand-distinct-roots (num den)
348 (let ((*roots nil)
349 (*failures nil))
350 (solve den var 1)
351 (cond (*failures (error "EXPAND-DISTINCT-ROOTS: failed to solve for roots."))
352 ((distinct-nonzero-roots-p *roots)
353 (psp2form (m+l (mapcar #'(lambda (r)
354 (and $verbose
355 (prog2
356 (mtell (intl:gettext "powerseries: for root:~%"))
357 (show-exp r)
358 (mtell (intl:gettext "powerseries: numerator at root =~%"))
359 (show-exp (maxima-substitute r var num))
360 (mtell (intl:gettext "powerseries: first derivative of denominator at root =~%"))
361 (show-exp (maxima-substitute r var ($diff den var)))))
362 (m* -1
363 (m// 1 r)
364 (maxima-substitute r var num)
365 (m// 1 (maxima-substitute r var ($diff den var)))
366 (m^ (m// 1 r) *index)))
367 (mapcar #'caddr (remove-mult *roots))))
368 *index 0))
369 (t (error "EXPAND-DISTINCT-ROOTS: roots are not distinct.~%")))))
371 ;; Try to expand N/D as a power series in VAR about 0.
373 ;; Can safely assume that D has no zero roots (we remove them in SRATEXPND
374 ;; before calling this function).
375 (defun ratexpand1 (n d)
376 (when $verbose
377 (mtell (intl:gettext
378 "powerseries: attempt partial fraction expansion of "))
379 (show-exp (list '(mquotient) n d))
380 (terpri)
381 (finish-output))
383 ;; *RATEXP serves as a recursion guard: if SRATEXPND is about to call us
384 ;; *recursively, the flag will be set and it gives up instead.
385 (let* ((*ratexp t)
386 (fractions ($partfrac (div* n d) var)))
387 (cond
388 ;; If $PARTFRAC succeeds, it will return a sum of terms. In that case,
389 ;; expand each one (which we assume is going to be easier than what we
390 ;; started with) and try to fold the result into a single power series sum
391 ;; again.
392 ((mplusp fractions)
393 (when $verbose
394 (mtell (intl:gettext "which is ~%"))
395 (show-exp fractions))
396 (psp2foldsum (m+l (mapcar #'ratexp (cdr fractions)))))
398 ;; If that didn't work, maybe it's because the numerator messed things
399 ;; up. If we're really lucky, the numerator is actually a polynomial
400 ;; though. In that case, factor it out and do an expansion of 1/d on its
401 ;; own.
402 ((poly? n var)
403 (when $verbose
404 (mtell (intl:gettext
405 "powerseries: partial fraction expansion failed, ~
406 expanding denominator only.~%")))
407 (m* n (ratexp (m// 1 d))))
409 ;; If n is complicated, there's not really much we can do to make further
410 ;; progress, so give up and return a noun form.
411 (t (powerseries-expansion-error
412 (intl:gettext "Partial fraction expansion failed"))))))
414 (defun sratsubst (gcd num den)
415 (and $verbose
416 (prog2 (mtell (intl:gettext "powerseries: substituting for the occurrences of"))
417 (show-exp (list '(mexpt) var gcd))
418 (mtell (intl:gettext "in"))
419 (show-exp (list '(mquotient) num den))
420 (terpri)
421 (finish-output)))
422 (funcall #'(lambda (var* *var)
423 (setq num (maxima-substitute (m^ var* (m^ gcd -1)) *var num)
424 den (maxima-substitute (m^ var* (m^ gcd -1)) *var den))
425 (maxima-substitute (m^ *var gcd) var*
426 (funcall #'(lambda (var) (sratexpnd num den)) var*)))
427 (gensym) var))
429 (defun ggcd (l)
430 (cond ((null l) 1)
431 ((null (cdr l)) (car l))
432 ((equal 1 (car l)) 1)
433 (t ($gcd (ggcd (cdr l)) (car l)))))
435 (defun exlist (exp)
436 (cond ((null exp) nil)
437 ((atom exp)
438 (and (eq exp var) (ncons 1)))
439 ((and (not (atom (car exp))) (eq (caar exp) 'mplus))
440 (exlist (cdr exp)))
441 ((smono (car exp) var)
442 (cond ((equal *n 0) (exlist (cdr exp)))
443 (t (cons *n (exlist (cdr exp))))))
444 (t (exlist (cdr exp)))))
446 ;; Perform a binomial expansion of (c x^m + a)^n.
448 ;; If n isn't known to be an integer, we'd like to write a sum of terms
450 ;; x^(mk) c^k a^(n-k) / ((n+1) beta(n-k+1, k+1))
452 ;; But we have to be a bit careful: it only works if c, x and a are
453 ;; nonzero. (Otherwise the simplifier quite reasonably simplifies each of the
454 ;; terms to zero, because the only nonzero term has the undefined term 0^0 in
455 ;; it).
457 ;; If we're not sure whether cx is zero, we can just split out the first term
458 ;; from the sum. If a is definitely zero, we can just write down a single
459 ;; term. If we're not sure about a, it's a bit more difficult: if we know that n
460 ;; is a positive integer, the sum is finite (and has at least two terms) and we
461 ;; can just split off the last term. Otherwise, give up.
462 (defun srbinexpnd (*schatc-ans*)
463 (alist-bind (n a m c cc x) *schatc-ans*
464 (m* cc
465 (if (and (integerp n) (minusp n))
466 (srintegexpd (neg n) a m c)
467 (let ((sgn-a ($sign (list '(mabs) a)))
468 (sgn-cx ($sign `((mabs) ((mtimes) ,c ,x))))
469 (sgn-n ($sign n))
470 (general-term
471 (m// (m* (m^ var (m* m *index))
472 (m^ c *index)
473 (m^ a (m- n *index)))
474 (m* (ftake* '%beta (m- n (m1- *index)) (m1+ *index))
475 (m1+ n)))))
476 (cond
477 ((eq sgn-n '$zero) 1)
478 ((eq sgn-cx '$zero) (m^ a n))
479 ((eq sgn-a '$zero) (m* (m^ c n) (m^ x (m* m n))))
481 ((and (eq sgn-a '$pos) (eq sgn-cx '$pos))
482 (if (and ($featurep n '$integer)
483 (memq sgn-n '($pos $pz)))
484 `((%sum) ,general-term ,*index 0 ,n)
485 `((%sum) ,general-term ,*index 0 $inf)))
487 ((eq sgn-a '$pos)
488 (m+ (m^ a n) `((%sum) ,general-term ,*index 1 $inf)))
490 ((and ($featurep n '$integer) (eq sgn-n '$pos))
491 (m+ `((%sum) ,general-term ,*index 0 ,(m1- n))
492 (m* (m^ c n) (m^ x (m* n m)))))
495 (powerseries-expansion-error
496 (intl:gettext
497 "Couldn't expand binomial~%~M~%~
498 as we didn't know which terms were nonzero.")))))))))
500 (defun psp2form (coeff exp bas)
501 (list '(%sum) (m* coeff (m^ var exp)) *index bas '$inf))
503 (defun srintegexpd (n a m c)
504 (and $verbose
505 (prog2 (mtell (intl:gettext "powerseries: apply rule for expressions of ~%"))
506 (show-exp '((mexpt) ((mplus) $a ((mtimes) $c ((mexpt) $var $m)))
507 ((mminus) $n)))
508 (mtell (intl:gettext "powerseries: here we have"))
509 (show-exp (list '(mlist) (list '(mequal) '$n n) (list '(mequal) '$a a)
510 (list '(mequal) '$c c) (list '(mequal) '$m m)))))
511 (cond ((= n 1)
512 (psp2form
513 (m* (m^ a (m* -1 (m+ 1 *index)))
514 (m^ (m* -1 c) *index))
515 (if (equal m 1) *index (m* *index m))
517 ((= 2 n)
518 (psp2form (m* (m+ 1 *index)
519 (m^ a (m* -1 (m+ 2 *index)))
520 (m^ (m* -1 c) *index))
521 (if (equal m 1) *index (m* *index m))
523 (t (psp2form (m* (do ((nn (1- n) (1- nn))
524 (l nil (cons (list '(mplus) *index nn) l)))
525 ((= nn 0)
526 (m*l (cons (m// 1 (factorial (1- n))) l))))
527 (m^ (m* -1 c (m^ a -1)) *index)
528 (m^ a (- n)))
529 (if (equal m 1) *index (m* *index m))
530 0))))
532 (defun sratp (a var)
533 (cond ((atom a) t)
534 ((member (caar a) '(mplus mtimes) :test #'eq) (sandmap (cdr a)))
535 ((eq (caar a) 'mexpt)
536 (cond ((free (cadr a) var) (free (caddr a) var))
537 ((smono a var) t)
538 ((and (free (caddr a) var) (sratp (cadr a) var)))))
539 (t (and (free (mop a) var) (every #'(lambda (s) (free s var)) (margs a))))))
541 (defun sandmap (l) (or (null l) (and (sratp (car l) var) (sandmap (cdr l)))))
543 (defun sp2trig (exp) (subst *index '*index (zl-get (caar exp) 'sp2)))
545 ;; Take an expression, EXPR, and try to write it as a + b*VAR^c. On success,
546 ;; returns (VALUES A B C). On failure, raise a powerseries-expansion-error.
548 ;; One way to do this would be to call $EXPAND and look at the results, but that
549 ;; might be rather slow - if the expression is something like (1+x)^10, expand
550 ;; will take ages and won't help very much. Another approach would be to put it
551 ;; into CRE form, but that would be a mistake if the input was something like
552 ;; (1+x)^100...
554 ;; Instead, we're a bit stupider: we walk through the expression tree. If we see
555 ;; anything other than +,* and ^, we give up. If we find we have more than one
556 ;; different exponent in our terms, we give up.
558 ;; Obviously, there are always examples where this won't work, but $EXPAND will
559 ;; (something like (x+1)^2 - x^2, for example), but I'm assuming that this won't
560 ;; be something we encounter in practice.
561 (defun split-two-term-poly (expr)
562 (cond
563 ((atom expr)
564 (if (eq expr var)
565 (values 0 1 1)
566 (values expr 0 1)))
568 ((freeof var expr)
569 (values expr 0 1))
571 ((eq (caar expr) 'mplus)
572 (let ((a 0) (b) (c))
573 (dolist (arg (cdr expr))
574 (multiple-value-bind (aa bb cc) (split-two-term-poly arg)
575 (setf a (m+ a aa))
576 (unless (eql bb 0)
577 (cond
578 ;; Is this the first nonzero power we've seen?
579 ((not b)
580 (setf b bb c cc))
581 ;; Is this another term with the same power as one we've seen
582 ;; already?
583 ((eql c cc)
584 (setf b (m+ b bb)))
585 ;; Otherwise, give up.
587 (powerseries-expansion-error))))))
588 (values a b c)))
590 ((eq (caar expr) 'mtimes)
591 (let ((prod 1) a b c)
592 (dolist (arg (cdr expr))
593 (multiple-value-bind (aa bb cc) (split-two-term-poly arg)
594 (cond
595 ((eql bb 0)
596 (setf prod (m* prod aa)))
597 ((not a)
598 (setf a aa b bb c cc))
599 ;; This is the second term (a + b*x^c), so we know we'll end up
600 ;; with mixed terms. (We don't try to spot e.g. (1-x)*(1+x))
602 (powerseries-expansion-error)))))
604 (if a
605 (values (m* a prod) (m* b prod) c)
606 (values prod 0 1))))
608 ((eq (caar expr) 'mexpt)
609 ;; We know that EXPR isn't free of VAR. Check that VAR isn't in the
610 ;; exponent.
611 (unless (freeof var (caddr expr))
612 (powerseries-expansion-error))
613 (multiple-value-bind (a b c) (split-two-term-poly (cadr expr))
614 (cond
615 ((eql a 0)
616 (values 0 (m^ b (caddr expr)) (m* c (caddr expr))))
617 ((eql b 0)
618 (values (m^ a (caddr expr)) 0 1))
620 (powerseries-expansion-error)))))
623 (powerseries-expansion-error))))
625 ;; Try to expand log(e) using the series for log(1+x).
627 ;; The basic idea is that if a is nonzero then
629 ;; log(a + b*x^k) = log(a*(1 + b/a*x^k))
630 ;; = log(a) + log(1 + b/a*x^k)
632 ;; and we know a series for that.
633 (defun sp2log (e)
634 (cond
635 ((or (atom e) (free e var))
636 (list '(%log) e))
638 (multiple-value-bind (a b k)
639 (split-two-term-poly (specdisrep e))
640 ;; If a is zero, we can't expand
641 (unless (eq '$positive
642 (asksign (list '(mabs) a)))
643 (powerseries-expansion-error))
645 (let* ((coeff (m* b (m^ a -1)))
646 (coeff-sign ($sign coeff))
647 (negate-coeff-p))
648 ;; If we know that the coefficient is not positive, switch the series
649 ;; around (which gets rid of some ugly minus signs). If we're not sure,
650 ;; but the cofficient "looks negative" (so is something like -7*k),
651 ;; switch it around too.
652 (cond
653 ((member coeff-sign '($neg $nz))
654 (setf negate-coeff-p t
655 coeff (m- coeff)))
656 ((and (not (member coeff-sign '($pos $pz)))
657 (mtimesp coeff)
658 (numberp (cadr coeff))
659 (minusp (cadr coeff)))
660 (setq negate-coeff-p t
661 coeff (simptimes
662 (list* (car coeff) (- (cadr coeff)) (cddr coeff))
663 1 t))))
664 (list '(%sum)
665 (m* -1
666 (m^ (if negate-coeff-p
667 (m* coeff (m^ var k))
668 (m* -1 coeff (m^ var k)))
669 *index)
670 (m^ *index -1))
671 *index 1 '$inf))))))
673 (defun sp2expt (exp)
674 (let ((base (cadr exp))
675 (power (caddr exp)))
676 (cond
677 ;; Do (a^b)^c = a^(bc) when c is a number
678 ((and (numberp power) (mexptp base))
679 (sp2expt (m^ (cadr base)
680 (m* power (caddr base)))))
682 ;; If a positive power which is free of the expansion variable and less
683 ;; than the maximum expansion power then expand the base and do the
684 ;; multiplication explicitly.
685 ((and (free power var)
686 (signp g power)
687 (< power $maxposex))
688 (m*l (dup (sp2expand base) power)))
690 ;; If the base is free of VAR, rewrite as a^b = e^(b log(a)) if necessary
691 ;; and then use the tabulated expansion of exp(x). If POWER is a sum then
692 ;; do the rewrite a^(b+c) = a^b a^c
693 ((free base var)
694 (let ((expansion
695 (subst *index '*index
696 (if (eq base '$%e)
697 (get 'mexpt 'sp2)
698 (subst `((mtimes) ((mlog) ,base) sp2var) 'sp2var
699 (get 'mexpt 'sp2))))))
700 (cond
701 ((not (mplusp power))
702 (sp2sub expansion power))
705 (m*l (mapcar (lambda (summand) (sp2sub expansion summand))
706 (cdr power)))))))
708 (t (powerseries-expansion-error)))))
710 (defun dup (x %n)
711 (if (= %n 1)
712 (ncons x)
713 (cons x (dup x (1- %n)))))
715 (defun sp2diff (exp l)
716 (cond
717 ((free exp var)
718 (sp3form (sp2expand exp) (cons '(%derivative) l)))
720 ;; If the expression isn't free of VAR, we know how to expand the result if
721 ;; the orders of differentiation are all explicit non-negative
722 ;; integers. Otherwise, give up.
723 (let ((indl) (remaining-derivs))
724 (loop
725 for (y order) on l by #'cddr
727 (cond
728 ((not (typep order '(integer 0)))
729 (powerseries-expansion-error))
731 ((not (eq y var))
732 (setf remaining-derivs (list* order y remaining-derivs)))
735 (dotimes (i order)
736 (setf indl nil)
737 (setf exp (sp2diff1 (sp2expand exp) nil nil))))))
739 (if remaining-derivs
740 (sp3form exp `((%derivative)
741 ,@(nreverse remaining-derivs)))
742 exp)))))
744 (defun sp2diff1 (exp ind lol) ;ind is a list of the indices
745 ;lol is a list of the lower limits
746 (cond ((atom exp) (sdiff exp var))
747 ((eq (caar exp) 'mplus)
748 (cons '(mplus)
749 (mapcar #'(lambda (q) (sp2diff1 q ind lol))
750 (cdr exp))))
751 ((eq (caar exp) '%sum)
752 (setq indl (cons (copy-list (cddr exp)) indl))
753 (sp2diff1 (cadr exp)
754 (cons (caddr exp) ind)
755 (cons (cadddr exp) lol)))
756 (t (sp2diff2 exp ind lol))))
758 (defun sp2diff2 (exp ind lol)
759 (let (e fr)
760 (setq e (m2 exp '((mtimes) ((coefftt) (fr freevar))
761 ((coefftt) (e true))))
762 fr (cdr (assoc 'fr e :test #'eq))
763 e (cdr (assoc 'e e :test #'eq)))
764 (sp3reconst
765 (cond ((and (mexptp e) (eq (cadr e) var))
766 (cond ((equal 0 (mbinding (ind lol)
767 (meval (m* fr (caddr e)))))
768 (m* (sp3substp1 ind ind (m* fr (caddr e))) e))
769 ((mgrp 1 (mbinding (ind lol)
770 (simplify (mevalatoms (caddr e)))))
771 (m* fr (caddr e) (m^ (cadr e) (m- (caddr e) 1))))
772 (t (sdiff exp var))))
773 (t (sdiff exp var))))))
775 (defun sp2integ (exp v l)
776 (if (null l)
777 (if (eq var v)
778 (sp2integ1 ($expand (sp2expand exp)))
779 (sp3form (sp2expand exp) (list '(%integrate) v)))
780 (sp2integ2 exp v (car l) (cadr l))))
782 (defun sp2integ1 (exp)
783 (let (pair)
784 (cond ((ratp exp var) (ratint exp var))
785 ((eq (caar exp) 'mplus)
786 (cons '(mplus) (mapcar #'sp2integ1 (cdr exp))))
787 ((and (eq (caar exp) 'mtimes)
788 (prog2 (setq pair (partition exp var 1))
789 (not (equal (car pair) 1))))
790 (mul2* (car pair) (sp2integ1 (cdr pair))))
791 ((and (eq (caar exp) 'mtimes)
792 (prog2 (setq exp ($intosum exp)) nil)))
793 ((or (not (eq (caar exp) '%sum)) (not (isinop (cadr exp) '%sum)))
794 (sinint exp var))
795 (t (let ((indl (ncons (cddr exp))))
796 (sp2integ12 (cadr exp) (ncons (caddr exp)) (ncons (cadddr exp))))))))
798 (defun sp2integ12 (exp ind lol)
799 (cond ((atom exp)
800 (sp3reconst (ratint exp var)))
801 ((eq (caar exp) 'mplus)
802 (sp3reconst
803 (m+l (mapcar #'(lambda (q) (sp2integ13 q ind lol))
804 (cdr exp)))))
805 ((eq (caar exp) '%sum)
806 (setq indl (cons (cddr exp) indl))
807 (sp2integ12 (cadr exp)
808 (cons (caddr exp) ind)
809 (cons (cadddr exp) lol)))
810 (t (sp3reconst (sp2integ13 exp ind lol)))))
812 (defun sp2integ13 (exp ind lol)
813 (let (e fr)
814 (setq e (m2 exp '((mtimes) ((coefftt) (fr freevar))
815 ((coefftt) (e true))))
816 fr (cdr (assoc 'fr e :test #'eq))
817 e (cdr (assoc 'e e :test #'eq)))
818 (cond ((and (mexptp e) (eq (cadr e) var))
819 (cond ((mgrp -1 (mbinding (ind lol)
820 (meval (caddr e))))
821 (m* (sp3substpn ind ind (m* fr (caddr e)) -1) e))
822 (t (sinint exp var))))
823 (t (sinint exp var)))))
825 ;; Substitute NEW for OLD in the expression tree EXPRESSION. This is a bit like
826 ;; MAXIMA-SUBSTITUTE, except it assumes OLD is a symbol. But the important bit
827 ;; is that it is bright enough to avoid substituting for bound variables in
828 ;; %INTEGRATE and %AT forms: even though the symbol might have the same name,
829 ;; it's thought of as a logically different variable.
831 ;; The function returns two values: the new expression and a flag that says
832 ;; whether that expression has changed. If the flag is true on a recursive call,
833 ;; int-diff-substitute resimplifies the result.
834 (defun int-diff-substitute (new old expression)
835 (cond
836 ((eq expression old) (values new t))
837 ((atom expression) (values expression nil))
839 (let ((op (caar expression))
840 (args (cdr expression)))
841 (if (or (and (eq op '%integrate) (eq old (second args)))
842 (and (eq op '%at)
843 (not (atom (second args)))
844 (if (eq (caar (second args)) 'mlist)
845 ;; (second args) looks like ((mlist) ((mequal) ...) ...)
846 (memq old (mapcar #'second (rest (second args))))
847 ;; (second args) looks like ((mequal) ...)
848 (eq old (second (second args))))))
849 (values expression nil)
850 (let* ((some-changed-p nil)
851 (new-args
852 (mapcar (lambda (x)
853 (multiple-value-bind (new-val changed-p)
854 (int-diff-substitute new old x)
855 (when changed-p
856 (setf some-changed-p t))
857 new-val))
858 (cdr expression))))
859 (if (not some-changed-p)
860 (values expression nil)
861 (values
862 (simplifya (cons (list op) new-args) nil) t))))))))
864 ;; Expand a definite integral, integrate(exp, v, lo, hi).
865 (defun sp2integ2 (exp v lo hi)
866 ;; Deal with the case where the integration variable is also the expansion
867 ;; variable. Something's a bit odd if we're doing this, but we assume that the
868 ;; aliasing was accidental and that the (bound) integration variable should be
869 ;; called something else.
870 (when (eq v var)
871 (setq v (gensym)
872 exp (subst v var exp)))
874 (when (boundp '*sp2integ-recursion-guard*)
875 (powerseries-expansion-error
876 (intl:gettext "Recursion when trying to expand the definite integral: ~M")
877 (out-of (symbol-value '*sp2integ-recursion-guard*))))
879 (cond
880 ((not (and (free lo var) (free hi var)))
881 ;; Suppose one or both of the endpoints involves VAR. We'll give up unless
882 ;; they are monomials in VAR (because substituting a non-monomial into a
883 ;; power series is more difficult). If they *are* monomials in VAR, then we
884 ;; try to expand the integral as a power series and find an antiderivative.
885 (let ((lo-exp (sp2expand lo))
886 (hi-exp (sp2expand hi))
887 (*sp2integ-recursion-guard* exp))
888 (declare (special *sp2integ-recursion-guard*))
890 (unless (and (smono lo-exp var) (smono hi-exp var))
891 (powerseries-expansion-error
892 (intl:gettext "Endpoints of definite integral ~M are not monomials in ~
893 the expansion variable.") (out-of exp)))
895 ;; Since this is a formal powerseries calculation, we needn't concern
896 ;; ourselves with radii of convergence here. Just expand the integrand
897 ;; about zero. (Is there something cleverer we could do?)
898 (let ((anti-derivative (sinint ($powerseries exp v 0) v)))
899 ;; If the expansion had failed, we'd have thrown an error to top-level,
900 ;; so we can assume that ANTI-DERIVATIVE is a sum of monomials in
901 ;; V. Substituting in LO-EXP and HI-EXP, we'll get two sums of
902 ;; monomials in VAR. The difference of the two expressions isn't quite
903 ;; of the right form, but hopefully it's close enough for any other
904 ;; code that uses it.
905 (m- (int-diff-substitute hi-exp v anti-derivative)
906 (int-diff-substitute lo-exp v anti-derivative)))))
908 ((free exp var)
909 (list '(%integrate) (subst var v exp) var lo hi))
912 (sp3form (sp2expand exp)
913 (list '(%integrate) v lo hi)))))
915 ;;************************************************************************************
916 ;; phase three miscellaneous garbage and final simplification
917 ;;************************************************************************************
919 (defun sp3reconst (e)
920 (do ((l indl (cdr l)) (e e (list* '(%sum) e (car l))))
921 ((null l) e)))
923 (defun sp3substpn (vars vals exp n)
924 (sp3subst vars (mapcar #'(lambda (q) (add2* q n)) vals) exp))
926 (defun sp3substp1 (vars vals exp) (sp3substpn vars vals exp 1))
928 (defun sp3subst (vars vals exp)
929 (simplify (sublis (mapcar #'cons (cdr vars) (cdr vals))
930 (subst (car vals) (car vars) exp))))
932 (defun sp3form (e *form) (sp3form1 e))
934 (defun sp3form1 (e)
935 (cond ((atom e) (list* (car *form) e (cdr *form)))
936 ((eq (caar e) 'mplus)
937 (cons '(mplus) (mapcar #'sp3form1 (cdr e))))
938 ((eq (caar e) '%sum)
939 (list* '(%sum) (sp3form1 (cadr e)) (cddr e)))
940 (t (list* (car *form) e (cdr *form)))))
942 ;; These are the series expansions for circular functions
944 (defprop %sin
945 ((%sum) ((mtimes)
946 ((mexpt) -1 *index)
947 ((mexpt) ((mfactorial) ((mplus) ((mtimes) 2 *index) 1)) -1)
948 ((mexpt) sp2var ((mplus) ((mtimes) 2 *index) 1)))
949 *index 0 $inf)
950 sp2)
952 (defprop %cos
953 ((%sum) ((mtimes) ((mexpt) -1 *index)
954 ((mexpt) ((mfactorial) ((mtimes) 2 *index)) -1)
955 ((mexpt) sp2var ((mtimes) 2 *index)))
956 *index 0 $inf)
957 sp2)
959 (defprop %tan
960 ((%sum) ((mtimes) ((mexpt) -1 ((mplus) *index -1))
961 ((mexpt) 2 ((mtimes) 2 *index))
962 ((mplus) ((mexpt) 2 ((mtimes) 2 *index)) -1)
963 ((mexpt) ((mfactorial) ((mtimes) 2 *index)) -1)
964 ((%bern) ((mtimes) 2 *index))
965 ((mexpt) sp2var ((mplus) ((mtimes) 2 *index) -1)))
966 *index 0 $inf)
967 sp2)
969 (defprop %csc
970 ((%sum) ((mtimes) 2
971 ((mexpt) -1 ((mplus) *index -1))
972 ((mplus) ((mexpt) 2 ((mplus) ((mtimes) 2 *index) -1)) -1)
973 ((mexpt) ((mfactorial) ((mtimes) 2 *index)) -1)
974 ((%bern) ((mtimes) 2 *index))
975 ((mexpt) sp2var ((mplus) ((mtimes) 2 *index) -1)))
976 *index 0 $inf)
977 sp2)
979 (defprop %cot
980 ((%sum) ((mtimes)
981 ((mexpt) -1 *index)
982 ((mexpt) 2 ((mtimes) 2 *index))
983 ((mexpt) ((mfactorial) ((mtimes) 2 *index)) -1)
984 ((%bern) ((mtimes) 2 *index))
985 ((mexpt) sp2var ((mplus) ((mtimes) 2 *index) -1)))
986 *index 0 $inf)
987 sp2)
989 (defprop %sec
990 ((%sum) ((mtimes) ((mexpt) -1 *index)
991 ((mexpt) ((mfactorial) ((mtimes) 2 *index)) -1)
992 ((%euler) ((mtimes) 2 *index))
993 ((mexpt) sp2var ((mtimes) 2 *index)))
994 *index 0 $inf)
995 sp2)
997 ;; These are the series definitions of exponential functions.
999 (defprop mexpt
1000 ((%sum)
1001 ((mtimes) ((mexpt) ((mfactorial) *index) -1) ((mexpt) sp2var *index))
1002 *index 0 $inf)
1003 sp2)
1005 (defprop %sinh
1006 ((%sum) ((mtimes)
1007 ((mexpt) ((mfactorial) ((mplus) ((mtimes) 2 *index) 1)) -1)
1008 ((mexpt) sp2var ((mplus) ((mtimes) 2 *index) 1)))
1009 *index 0 $inf)
1010 sp2)
1012 (defprop %cosh
1013 ((%sum) ((mtimes)
1014 ((mexpt) ((mfactorial) ((mtimes) 2 *index)) -1)
1015 ((mexpt) sp2var ((mtimes) 2 *index)))
1016 *index 0 $inf)
1017 sp2)
1019 (defprop %tanh
1020 ((%sum)
1021 ((mtimes) ((mexpt) 4 *index)
1022 ((mplus) ((mexpt) 4 *index) -1)
1023 ((%bern) ((mtimes) 2 *index))
1024 ((mexpt) sp2var ((mplus) ((mtimes) 2 *index) -1))
1025 ((mexpt)
1026 ((mfactorial) ((mtimes) 2 *index))
1027 -1))
1028 *index 0 $inf)
1029 sp2)
1031 (defprop %coth
1032 ((%sum)
1033 ((mtimes) ((mexpt) 4 *index)
1034 ((%bern) ((mtimes) 2 *index))
1035 ((mexpt) ((mfactorial) ((mtimes) 2 *index)) -1)
1036 ((mexpt) sp2var ((mplus) ((mtimes) 2 *index) -1)))
1037 *index 0 $inf)
1038 sp2)
1040 (defprop %sech
1041 ((%sum)
1042 ((mtimes) ((%euler) ((mtimes) 2 *index))
1043 ((mexpt) ((mfactorial) ((mtimes) 2 *index)) -1)
1044 ((mexpt) sp2var ((mtimes) 2 *index)))
1045 *index 0 $inf)
1046 sp2)
1048 (defprop %csch
1049 ((%sum)
1050 ((mtimes) -2 ((mplus) ((mexpt) 2 ((mplus) ((mtimes) 2 *index) -1)) -1)
1051 ((mexpt) ((mfactorial) ((mtimes) *index 2)) -1)
1052 ((%bern) ((mtimes) 2 *index))
1053 ((mexpt) sp2var ((mplus) ((mtimes) 2 *index) -1)))
1054 *index 0 $inf)
1055 sp2)
1057 ;;arc trigonometric function expansions
1059 (defprop %asin
1060 ((%sum)
1061 ((mtimes) ((%genfact) ((mplus) ((mtimes) 2 *index) -1) *index 2)
1062 ((mexpt) ((%genfact) ((mtimes) 2 *index) *index 2) -1)
1063 ((mexpt) ((mplus) ((mtimes) 2 *index) 1) -1)
1064 ((mexpt) sp2var ((mplus) ((mtimes) 2 *index) 1)))
1065 *index 0 $inf)
1066 sp2)
1068 (defprop %atan
1069 ((%sum)
1070 ((mtimes) ((mexpt) -1 *index)
1071 ((mexpt) ((mplus) ((mtimes) 2 *index) 1) -1)
1072 ((mexpt) sp2var ((mplus) ((mtimes) 2 *index) 1)))
1073 *index 0 $inf)
1074 sp2)
1076 ;; Lambert W function principal branch
1077 (defprop %lambert_w
1078 ((%sum)
1079 ((mtimes) ((mexpt) -1 *index)
1080 ((mexpt) -1 1)
1081 ((mexpt) *index ( (mplus) *index -1))
1082 ((mexpt) ((mfactorial) *index) -1) ((mexpt) sp2var *index))
1083 *index 1 $inf)
1084 sp2)