Update rtestint problem 237 to match the actual result.
[maxima.git] / src / trigi.lisp
blob7b62804d1e5e5d7d5920467430cbb7b3717bb18b
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 trigi)
15 (load-macsyma-macros mrgmac)
17 ;;; Arithmetic utilities.
19 (defun sqrt1-x^2 (x)
20 (power (sub 1 (power x 2)) 1//2))
22 (defun sqrt1+x^2 (x)
23 (power (add 1 (power x 2)) 1//2))
25 (defun sqrtx^2-1 (x)
26 (power (add (power x 2) -1) 1//2))
28 (defun sq-sumsq (x y)
29 (power (add (power x 2) (power y 2)) 1//2))
31 (defun trigp (func)
32 (member func '(%sin %cos %tan %csc %sec %cot %sinh %cosh %tanh %csch %sech %coth)
33 :test #'eq))
35 (defun arcp (func)
36 (member func '(%asin %acos %atan %acsc %asec %acot %asinh %acosh %atanh %acsch %asech %acoth)
37 :test #'eq))
39 ;;; The trigonometric functions distribute of lists, matrices and equations.
41 (dolist (x '(%sin %cos %tan %cot %csc %sec
42 %sinh %cosh %tanh %coth %csch %sech
43 %asin %acos %atan %acot %acsc %asec
44 %asinh %acosh %atanh %acoth %acsch %asech))
45 (setf (get x 'distribute_over) '(mlist $matrix mequal)))
47 (defun domain-error (x f)
48 (merror (intl:gettext "~A: argument ~:M isn't in the domain of ~A.") f (complexify x) f))
50 ;; Some Lisp implementations goof up branch cuts for ASIN, ACOS, and/or ATANH.
51 ;; Here are definitions which have the right branch cuts
52 ;; (assuming LOG, PHASE, and SQRT have the right branch cuts).
53 ;; Don't bother trying to sort out which implementations get it right or wrong;
54 ;; we'll make all implementations use these functions.
56 ;; Apply formula from CLHS if X falls on a branch cut.
57 ;; Otherwise punt to CL:ASIN.
58 (defun maxima-branch-asin (x)
59 ;; Test for (IMAGPART X) is EQUAL because signed zero is EQUAL to zero.
60 (if (and (> (abs (realpart x)) 1.0) (equal (imagpart x) 0.0))
61 ;; The formula from CLHS is asin(x) = -%i*log(%i*x+sqrt(1-x^2)).
62 ;; This has problems with overflow for large x.
64 ;; Let's rewrite it, where abs(x)>1
66 ;; asin(x) = -%i*log(%i*x+abs(x)*sqrt(1-1/x^2))
67 ;; = -%i*log(%i*x*(1+abs(x)/x*sqrt(1-1/x^2)))
68 ;; = -%i*[log(abs(x)*abs(1+abs(x)/x*sqrt(1-1/x^2)))
69 ;; + %i*arg(%i*x*(1+abs(x)/x*sqrt(1-1/x^2)))]
70 ;; = -%i*[log(abs(x)*(1+abs(x)/x*sqrt(1-1/x^2)))
71 ;; + %i*%pi/2*sign(x)]
72 ;; = %pi/2*sign(x) - %i*[log(abs(x)*(1+abs(x)/x*sqrt(1-1/x^2))]
74 ;; Now, look at log part. If x > 0, we have
76 ;; log(x*(1+sqrt(1-1/x^2)))
78 ;; which is just fine. For x < 0, we have
80 ;; log(abs(x)*(1-sqrt(1-1/x^2))).
82 ;; But
83 ;; 1-sqrt(1-1/x^2) = (1-sqrt(1-1/x^2))*(1+sqrt(1-1/x^2))/(1+sqrt(1-1/x^2))
84 ;; = (1-(1-1/x^2))/(1+sqrt(1-1/x^2))
85 ;; = 1/x^2/(1+sqrt(1-1/x^2))
87 ;; So
89 ;; log(abs(x)*(1-sqrt(1-1/x^2)))
90 ;; = log(abs(x)/x^2/(1+sqrt(1-1/x^2)))
91 ;; = -log(x^2/abs(x)*(1+sqrt(1-1/x^2))
92 ;; = -log(abs(x)*(1+sqrt(1-1/x^2)))
94 ;; Thus, for x < 0,
96 ;; asin(x) = -%pi/2+%i*log(abs(x)*(1+sqrt(1-1/x^2)))
97 ;; = -asin(-x)
99 ;; If we had an accurate f(x) = log(1+x) function, we should
100 ;; probably evaluate log(1+sqrt(1-1/x^2)) via f(x) instead of
101 ;; log. One other accuracy change is to evaluate sqrt(1-1/x^2)
102 ;; as sqrt(1-1/x)*sqrt(1+1/x), because 1/x^2 won't underflow as
103 ;; soon as 1/x.
104 (let* ((absx (abs x))
105 (recip (/ absx))
106 (result (complex (/ #.(float pi) 2)
107 (- (log (* absx
108 (1+ (* (sqrt (+ 1 recip))
109 (sqrt (- 1 recip))))))))))
110 (if (minusp x)
111 (- result)
112 result))
113 (cl:asin x)))
115 ;; Apply formula from CLHS if X falls on a branch cut.
116 ;; Otherwise punt to CL:ACOS.
117 (defun maxima-branch-acos (x)
118 ; Test for (IMAGPART X) is EQUAL because signed zero is EQUAL to zero.
119 (if (and (> (abs (realpart x)) 1.0) (equal (imagpart x) 0.0))
120 (- #.(/ (float pi) 2) (maxima-branch-asin x))
121 (cl:acos x)))
123 (defun maxima-branch-acot (x)
124 ;; Allow 0.0 in domain of acot, otherwise use atan(1/x)
125 (if (and (equal (realpart x) 0.0) (equal (imagpart x) 0.0))
126 #.(/ (float pi) 2)
127 (cl:atan (/ 1 x))))
129 ;; Apply formula from CLHS if X falls on a branch cut.
130 ;; Otherwise punt to CL:ATANH.
131 (defun maxima-branch-atanh (x)
132 ; Test for (IMAGPART X) is EQUAL because signed zero is EQUAL to zero.
133 (if (and (> (abs (realpart x)) 1.0) (equal (imagpart x) 0.0))
134 (/ (- (cl:log (+ 1 x)) (cl:log (- 1 x))) 2)
135 (cl:atanh x)))
137 ;; Fill the hash table.
138 (macrolet ((frob (mfun dfun) `(setf (gethash ',mfun *flonum-op*) ,dfun)))
139 (frob mplus #'+)
140 (frob mtimes #'*)
141 (frob mquotient #'/)
142 (frob mminus #'-)
144 (frob %cos #'cl:cos)
145 (frob %sin #'cl:sin)
146 (frob %tan #'cl:tan)
148 (frob %sec #'(lambda (x)
149 (let ((y (ignore-errors (/ 1 (cl:cos x)))))
150 (if y y (domain-error x 'sec)))))
152 (frob %csc #'(lambda (x)
153 (let ((y (ignore-errors (/ 1 (cl:sin x)))))
154 (if y y (domain-error x 'csc)))))
156 (frob %cot #'(lambda (x)
157 (let ((y (ignore-errors (/ 1 (cl:tan x)))))
158 (if y y (domain-error x 'cot)))))
160 (frob %acos #'maxima-branch-acos)
161 (frob %asin #'maxima-branch-asin)
163 (frob %atan #'cl:atan)
165 (frob %asec #'(lambda (x)
166 (let ((y (ignore-errors (maxima-branch-acos (/ 1 x)))))
167 (if y y (domain-error x 'asec)))))
169 (frob %acsc #'(lambda (x)
170 (let ((y (ignore-errors (maxima-branch-asin (/ 1 x)))))
171 (if y y (domain-error x 'acsc)))))
173 (frob %acot #'(lambda (x)
174 (let ((y (ignore-errors (maxima-branch-acot x))))
175 (if y y (domain-error x 'acot)))))
177 (frob %cosh #'cl:cosh)
178 (frob %sinh #'cl:sinh)
179 (frob %tanh #'cl:tanh)
181 (frob %sech #'(lambda (x)
182 (flet ((sech (x)
183 ;; For large x > 0, cosh(x) ~= exp(x)/2.
184 ;; Hence, sech(x) ~= 2*exp(-x). And since
185 ;; cosh(x) is even, we only need to deal
186 ;; with |x|. Note also that if |x| >=
187 ;; sqrt(most-positive-double-float),
188 ;; exp(-x) is basically zero, so we can use
189 ;; a threshold of sqrt(most-positive).
191 ;; Several Lisp's can not compute acosh()
192 ;; for very large values, e.g.
193 ;; (acosh most-positive-double-float)
194 ;; Therefore use the numerical value
195 ;; 710.4758600739439d0 = (acosh most-positive-double-float)
196 ;; instead of computing the value.
197 ;; The most-positive-double-float is standardized (IEEE 754).
198 (if (and (floatp x)
199 (>= (abs x) 710.4758600739439d0))
200 (* 2 (exp (- (abs x))))
201 (/ (cl:cosh x)))))
202 (let ((y (ignore-errors (sech x))))
203 (if y y (domain-error x 'sech))))))
205 (frob %csch #'(lambda (x)
206 (flet ((csch (x)
207 ;; For large x > 0, sinh(x) ~= exp(x)/2.
208 ;; Hence csch(x) = 2*exp(-x). Since
209 ;; sinh(x) is odd, we also have csch(x) =
210 ;; -2*exp(x) when x < 0 and |x| is large.
212 ;; Several Lisp's can not compute asinh()
213 ;; for very large values, e.g.
214 ;; (asinh most-positive-double-float)
215 ;; Therefore use the numerical value
216 ;; 710.4758600739439d0 = (asinh most-positive-double-float)
217 ;; instead of computing the value.
218 ;; The most-positive-double-float is standardized (IEEE 754).
219 (if (and (floatp x)
220 (>= (abs x) 710.4758600739439d0))
221 (float-sign x (* 2 (exp (- (abs x)))))
222 (/ (cl:sinh x)))))
223 (let ((y (ignore-errors (csch x))))
224 (if y y (domain-error x 'csch))))))
226 (frob %coth #'(lambda (x)
227 (let ((y (ignore-errors (/ 1 (cl:tanh x)))))
228 (if y y (domain-error x 'coth)))))
230 (frob %acosh #'cl:acosh)
231 (frob %asinh #'cl:asinh)
233 (frob %atanh #'maxima-branch-atanh)
235 (frob %asech #'(lambda (x)
236 (let ((y (ignore-errors (cl:acosh (/ 1 x)))))
237 (if y y (domain-error x 'asech)))))
239 (frob %acsch
240 #'(lambda (x)
241 (flet ((acsch (x)
242 ;; logarc(acsch(x)) = log(1/x+sqrt(1/x^2+1)).
243 ;; Assume x > 0. Then we can rewrite this as
244 ;; log((1+sqrt(1+x^2))/x). If we choose x such
245 ;; that 1+x^2 = 1, then this simplifies to
246 ;; log(2/x). However for very small x, 2/x can
247 ;; overflow, so use log(2)-log(x).
249 ;; 1+x^2 = 1 when x^2 = double-float-epsilon. So
250 ;; x = sqrt(double-float-epsilon). We'd really
251 ;; like to use
252 ;; least-positive-normalized-double-float, but
253 ;; some lisps like clisp don't have denormals.
254 ;; In that case, use sqrt(double-float-epsilon).
255 (let ((absx (abs x)))
256 (cond ((and (floatp x)
257 (< absx
258 #-clisp
259 least-positive-normalized-double-float
260 #+clisp
261 (sqrt double-float-epsilon)))
262 (float-sign x (- (log 2d0) (log (abs x)))))
264 (cl:asinh (/ x)))))))
265 (let ((y (ignore-errors (acsch x))))
266 (if y y (domain-error x 'acsch))))))
268 (frob %acoth #'(lambda (x)
269 (let ((y (ignore-errors (maxima-branch-atanh (/ 1 x)))))
270 (if y y (domain-error x 'acoth)))))
272 (frob mabs #'cl:abs)
273 (frob %exp #'cl:exp)
274 (frob mexpt #'cl:expt)
275 (frob %sqrt #'cl:sqrt)
276 (frob %log #'(lambda (x)
277 (let ((y (ignore-errors (cl:log x))))
278 (if y y (domain-error x 'log)))))
280 (frob %plog #'(lambda (x)
281 (let ((y (ignore-errors (cl:log x))))
282 (if y y (domain-error x 'log)))))
284 (frob $conjugate #'cl:conjugate)
285 (frob $floor #'cl:ffloor)
286 (frob $ceiling #'cl:fceiling)
287 (frob $realpart #'cl:realpart)
288 (frob $imagpart #'cl:imagpart)
289 (frob $max #'cl:max)
290 (frob $min #'cl:min)
291 (frob %signum #'cl:signum)
292 (frob %atan2 #'cl:atan))
294 (macrolet ((frob (mfun dfun) `(setf (gethash ',mfun *big-float-op*) ,dfun)))
295 ;; All big-float implementation functions MUST support a required x
296 ;; arg and an optional y arg for the real and imaginary parts. The
297 ;; imaginary part does not have to be given.
298 (frob %asin #'big-float-asin)
299 (frob %sinh #'big-float-sinh)
300 (frob %asinh #'big-float-asinh)
301 (frob %tanh #'big-float-tanh)
302 (frob %atanh #'big-float-atanh)
303 (frob %acos 'big-float-acos)
304 (frob %log 'big-float-log)
305 (frob %sqrt 'big-float-sqrt))
307 ;; Here is a general scheme for defining and applying reflection rules. A
308 ;; reflection rule is something like f(-x) --> f(x), or f(-x) --> %pi - f(x).
310 ;; We define functions for the two most common reflection rules; these
311 ;; are the odd function rule (f(-x) --> -f(x)) and the even function rule
312 ;; (f(-x) --> f(x)). A reflection rule takes two arguments (the operator and
313 ;; the operand).
315 (defun odd-function-reflect (op x)
316 (neg (take (list op) (neg x))))
318 (defun even-function-reflect (op x)
319 (take (list op) (neg x)))
321 ;; Put the reflection rule on the property list of the exponential-like
322 ;; functions.
324 (setf (get '%cos 'reflection-rule) 'even-function-reflect)
325 (setf (get '%sin 'reflection-rule) 'odd-function-reflect)
326 (setf (get '%tan 'reflection-rule) 'odd-function-reflect)
327 (setf (get '%sec 'reflection-rule) 'even-function-reflect)
328 (setf (get '%csc 'reflection-rule) 'odd-function-reflect)
329 (setf (get '%cot 'reflection-rule) 'odd-function-reflect)
331 ;; See A&S 4.4.14--4.4.19
333 (setf (get '%acos 'reflection-rule) #'(lambda (op x) (sub '$%pi (take (list op) (neg x)))))
334 (setf (get '%asin 'reflection-rule) 'odd-function-reflect)
335 (setf (get '%atan 'reflection-rule) 'odd-function-reflect)
336 (setf (get '%asec 'reflection-rule) #'(lambda (op x) (sub '$%pi (take (list op) (neg x)))))
337 (setf (get '%acsc 'reflection-rule) 'odd-function-reflect)
338 (setf (get '%acot 'reflection-rule) 'odd-function-reflect)
340 (setf (get '%cosh 'reflection-rule) 'even-function-reflect)
341 (setf (get '%sinh 'reflection-rule) 'odd-function-reflect)
342 (setf (get '%tanh 'reflection-rule) 'odd-function-reflect)
343 (setf (get '%sech 'reflection-rule) 'even-function-reflect)
344 (setf (get '%csch 'reflection-rule) 'odd-function-reflect)
345 (setf (get '%coth 'reflection-rule) 'odd-function-reflect)
347 (setf (get '%asinh 'reflection-rule) 'odd-function-reflect)
348 (setf (get '%atanh 'reflection-rule) 'odd-function-reflect)
349 (setf (get '%asech 'reflection-rule) 'even-function-reflect)
350 (setf (get '%acsch 'reflection-rule) 'odd-function-reflect)
351 (setf (get '%acoth 'reflection-rule) 'odd-function-reflect)
353 ;; When b is nil, do not apply the reflection rule. For trigonometric like
354 ;; functions, b is $trigsign. This function uses 'great' to decide when to
355 ;; apply the rule. Another possibility is to apply the rule when (mminusp* x)
356 ;; evaluates to true. Maxima <= 5.9.3 uses this scheme; with this method, we have
357 ;; assume(z < 0), cos(z) --> cos(-z). I (Barton Willis) think this goofy.
359 ;; The function 'great' is non-transitive. I don't think this bug will cause
360 ;; trouble for this function. If there is an expression such that both
361 ;; (great (neg x) x) and (great x (neg x)) evaluate to true, this function
362 ;; could cause an infinite loop. I could protect against this possibility with
363 ;; (and b f (great (neg x) x) (not (great x (neg x))).
365 (defun apply-reflection-simp (op x &optional (b t))
366 (let ((f (get op 'reflection-rule)))
367 (if (and b f (great (neg x) x)) (funcall f op x) nil)))
369 (defun taylorize (op x)
370 (if ($taylorp x)
371 (mfuncall '$apply '$taylor `((mlist) ((,op) ,($ratdisrep x)) ,@(cdr ($taylorinfo x)))) nil))
373 (defun float-or-rational-p (x)
374 (or (floatp x) ($ratnump x)))
376 (defun bigfloat-or-number-p (x)
377 (or ($bfloatp x) (numberp x) ($ratnump x)))
379 ;; When z is a Maxima complex float or when 'numer' is true and z is a
380 ;; Maxima complex number, evaluate (op z) by applying the mapping from
381 ;; the Maxima operator 'op' to the operator in the hash table
382 ;; '*flonum-op*'. When z isn't a Maxima complex number, return
383 ;; nil.
385 (defun flonum-eval (op z)
386 (let ((op (gethash op *flonum-op*)))
387 (when op
388 (multiple-value-bind (bool R I)
389 (complex-number-p z #'float-or-rational-p)
390 (when (and bool (or $numer (floatp R) (floatp I)))
391 (setq R ($float R))
392 (setq I ($float I))
393 (complexify (funcall op (if (zerop I) R (complex R I)))))))))
395 ;; For now, big float evaluation of trig-like functions for complex
396 ;; big floats uses rectform. I suspect that for some functions (not
397 ;; all of them) rectform generates expressions that are poorly suited
398 ;; for numerical evaluation. For better accuracy, these functions
399 ;; (maybe acosh, for one) may need to be special cased. If they are
400 ;; special-cased, the *big-float-op* hash table contains the special
401 ;; cases.
403 (defun big-float-eval (op z)
404 (when (complex-number-p z 'bigfloat-or-number-p)
405 (let ((x ($realpart z))
406 (y ($imagpart z))
407 (bop (gethash op *big-float-op*)))
408 ;; If bop is non-NIL, we want to try that first. If bop
409 ;; declines (by returning NIL), we silently give up and use the
410 ;; rectform version.
411 (cond ((and ($bfloatp x) (like 0 y))
412 (or (and bop (funcall bop x))
413 ($bfloat `((,op simp) ,x))))
414 ((or ($bfloatp x) ($bfloatp y))
415 (or (and bop (funcall bop ($bfloat x) ($bfloat y)))
416 (let ((z (add ($bfloat x) (mul '$%i ($bfloat y)))))
417 (setq z ($rectform `((,op simp) ,z)))
418 ($bfloat z))))))))
420 ;; For complex big float evaluation, it's important to check the
421 ;; simp flag -- otherwise Maxima can get stuck in an infinite loop:
422 ;; asin(1.23b0 + %i * 4.56b0) ---> (simp-%asin ((%asin) ...) -->
423 ;; (big-float-eval ((%asin) ...) --> (risplit ((%asin simp) ...) -->
424 ;; (simp-%asin ((%asin simp) ...). If the simp flag is ignored, we've
425 ;; got trouble.
427 (def-simplifier sin (y)
428 (let (z)
429 (cond ((flonum-eval (mop form) y))
430 ((and (not (member 'simp (car form))) (big-float-eval (mop form) y)))
431 ((taylorize (mop form) (second form)))
432 ((and $%piargs (cond ((zerop1 y) 0)
433 ((has-const-or-int-term y '$%pi) (%piargs-sin/cos y)))))
434 ((and $%iargs (multiplep y '$%i)) (mul '$%i (ftake* '%sinh (coeff y '$%i 1))))
435 ((and $triginverses (not (atom y))
436 (cond ((eq '%asin (setq z (caar y))) (cadr y))
437 ((eq '%acos z) (sqrt1-x^2 (cadr y)))
438 ((eq '%atan z) (div (cadr y) (sqrt1+x^2 (cadr y))))
439 ((eq '%acot z) (div 1 (sqrt1+x^2 (cadr y))))
440 ((eq '%asec z) (div (sqrtx^2-1 (cadr y)) (cadr y)))
441 ((eq '%acsc z) (div 1 (cadr y)))
442 ((eq '%atan2 z) (div (cadr y) (sq-sumsq (cadr y) (caddr y)))))))
443 ((and $trigexpand (trigexpand '%sin y)))
444 ($exponentialize (exponentialize '%sin y))
445 ((and $halfangles (halfangle '%sin y)))
446 ((apply-reflection-simp (mop form) y $trigsign))
447 ;((and $trigsign (mminusp* y)) (neg (ftake* '%sin (neg y))))
448 (t (give-up)))))
450 (def-simplifier cos (y)
451 (let (z)
452 (cond ((flonum-eval (mop form) y))
453 ((and (not (member 'simp (car form))) (big-float-eval (mop form) y)))
454 ((taylorize (mop form) (second form)))
455 ((and $%piargs (cond ((zerop1 y) 1)
456 ((has-const-or-int-term y '$%pi)
457 (%piargs-sin/cos (add %pi//2 y))))))
458 ((and $%iargs (multiplep y '$%i)) (ftake* '%cosh (coeff y '$%i 1)))
459 ((and $triginverses (not (atom y))
460 (cond ((eq '%acos (setq z (caar y))) (cadr y))
461 ((eq '%asin z) (sqrt1-x^2 (cadr y)))
462 ((eq '%atan z) (div 1 (sqrt1+x^2 (cadr y))))
463 ((eq '%acot z) (div (cadr y) (sqrt1+x^2 (cadr y))))
464 ((eq '%asec z) (div 1 (cadr y)))
465 ((eq '%acsc z) (div (sqrtx^2-1 (cadr y)) (cadr y)))
466 ((eq '%atan2 z) (div (caddr y) (sq-sumsq (cadr y) (caddr y)))))))
467 ((and $trigexpand (trigexpand '%cos y)))
468 ($exponentialize (exponentialize '%cos y))
469 ((and $halfangles (halfangle '%cos y)))
470 ((apply-reflection-simp (mop form) y $trigsign))
471 ;((and $trigsign (mminusp* y)) (ftake* '%cos (neg y)))
472 (t (give-up)))))
474 (defun %piargs-sin/cos (x)
475 (let ($float coeff ratcoeff zl-rem)
476 (setq ratcoeff (get-const-or-int-terms x '$%pi)
477 coeff (linearize ratcoeff)
478 zl-rem (get-not-const-or-int-terms x '$%pi))
479 (cond ((zerop1 zl-rem) (%piargs coeff ratcoeff))
480 ((not (mevenp (car coeff))) nil)
481 ((equal 0 (setq x (mmod (cdr coeff) 2))) (ftake* '%sin zl-rem))
482 ((equal 1 x) (neg (ftake* '%sin zl-rem)))
483 ((alike1 1//2 x) (ftake* '%cos zl-rem))
484 ((alike1 '((rat) 3 2) x) (neg (ftake* '%cos zl-rem))))))
487 (defun filter-sum (pred form simp-flag)
488 "Takes form to be a sum and a sum of the summands for which pred is
489 true. Passes simp-flag through to addn if there is more than one
490 term in the sum."
491 (if (mplusp form)
492 (addn (mapcan
493 #'(lambda (term)
494 (when (funcall pred term) (list term))) (cdr form))
495 simp-flag)
496 (if (funcall pred form) form 0)))
498 ;; collect terms of form A*var where A is a constant or integer.
499 ;; returns sum of all such A.
500 ;; does not expand form, so does not find constant term in (x+1)*var.
501 ;; thus we cannot simplify sin(2*%pi*(1+x)) => sin(2*%pi*x) unless
502 ;; the user calls expand. this could be extended to look a little
503 ;; more deeply into the expression, but we don't want to call expand
504 ;; in the core simplifier for reasons of speed and predictability.
505 (defun get-const-or-int-terms (form var)
506 (coeff
507 (filter-sum (lambda (term)
508 (let ((coeff (coeff term var 1)))
509 (and (not (zerop1 coeff))
510 (or ($constantp coeff)
511 (maxima-integerp coeff)))))
512 form
514 var 1))
516 ;; collect terms skipped by get-const-or-int-terms
517 (defun get-not-const-or-int-terms (form var)
518 (filter-sum (lambda (term)
519 (let ((coeff (coeff term var 1)))
520 (not (and (not (zerop1 coeff))
521 (or ($constantp coeff)
522 (maxima-integerp coeff))))))
523 form
526 (defun has-const-or-int-term (form var)
527 "Tests whether form has at least some term of the form a*var where a
528 is constant or integer"
529 (not (zerop1 (get-const-or-int-terms form var))))
531 (def-simplifier tan (y)
532 (let (z)
533 (cond ((flonum-eval (mop form) y))
534 ((and (not (member 'simp (car form))) (big-float-eval (mop form) y)))
535 ((taylorize (mop form) (second form)))
536 ((and $%piargs (cond ((zerop1 y) 0)
537 ((has-const-or-int-term y '$%pi) (%piargs-tan/cot y)))))
538 ((and $%iargs (multiplep y '$%i)) (mul '$%i (ftake* '%tanh (coeff y '$%i 1))))
539 ((and $triginverses (not (atom y))
540 (cond ((eq '%atan (setq z (caar y))) (cadr y))
541 ((eq '%asin z) (div (cadr y) (sqrt1-x^2 (cadr y))))
542 ((eq '%acos z) (div (sqrt1-x^2 (cadr y)) (cadr y)))
543 ((eq '%acot z) (div 1 (cadr y)))
544 ((eq '%asec z) (sqrtx^2-1 (cadr y)))
545 ((eq '%acsc z) (div 1 (sqrtx^2-1 (cadr y))))
546 ((eq '%atan2 z) (div (cadr y) (caddr y))))))
547 ((and $trigexpand (trigexpand '%tan y)))
548 ($exponentialize (exponentialize '%tan y))
549 ((and $halfangles (halfangle '%tan y)))
550 ((apply-reflection-simp (mop form) y $trigsign))
551 ;((and $trigsign (mminusp* y)) (neg (ftake* '%tan (neg y))))
552 (t (give-up)))))
554 (def-simplifier cot (y)
555 (let (z)
556 (cond ((flonum-eval (mop form) y))
557 ((and (not (member 'simp (car form))) (big-float-eval (mop form) y)))
558 ((taylorize (mop form) (second form)))
559 ((and $%piargs (cond ((zerop1 y) (domain-error y 'cot))
560 ((and (has-const-or-int-term y '$%pi)
561 (setq z (%piargs-tan/cot (add %pi//2 y))))
562 (neg z)))))
563 ((and $%iargs (multiplep y '$%i)) (mul -1 '$%i (ftake* '%coth (coeff y '$%i 1))))
564 ((and $triginverses (not (atom y))
565 (cond ((eq '%acot (setq z (caar y))) (cadr y))
566 ((eq '%asin z) (div (sqrt1-x^2 (cadr y)) (cadr y)))
567 ((eq '%acos z) (div (cadr y) (sqrt1-x^2 (cadr y))))
568 ((eq '%atan z) (div 1 (cadr y)))
569 ((eq '%asec z) (div 1 (sqrtx^2-1 (cadr y))))
570 ((eq '%acsc z) (sqrtx^2-1 (cadr y)))
571 ((eq '%atan2 z) (div (caddr y) (cadr y))))))
572 ((and $trigexpand (trigexpand '%cot y)))
573 ($exponentialize (exponentialize '%cot y))
574 ((and $halfangles (halfangle '%cot y)))
575 ((apply-reflection-simp (mop form) y $trigsign))
576 ;((and $trigsign (mminusp* y)) (neg (ftake* '%cot (neg y))))
577 (t (give-up)))))
579 (defun %piargs-tan/cot (x)
580 "If x is of the form tan(u) where u has a nonzero constant linear
581 term in %pi, then %piargs-tan/cot returns a simplified version of x
582 without this constant term."
583 ;; Set coeff to be the coefficient of $%pi collecting terms with no
584 ;; other atoms, so given %pi(x+1/2), coeff = 1/2. Let zl-rem be the
585 ;; remainder (TODO: computing zl-rem could probably be prettier.)
586 (let* ((nice-terms (get-const-or-int-terms x '$%pi))
587 (coeff (linearize nice-terms))
588 (zl-rem (get-not-const-or-int-terms x '$%pi))
589 (sin-of-coeff-pi)
590 (cos-of-coeff-pi))
591 (cond
592 ;; sin-of-coeff-pi and cos-of-coeff-pi are only non-nil if they
593 ;; are constants that %piargs-offset could compute, and we just
594 ;; checked that cos-of-coeff-pi was nonzero. Thus we can just
595 ;; return their quotient.
596 ((and (zerop1 zl-rem)
597 (setq sin-of-coeff-pi
598 (%piargs coeff nil)))
599 (setq cos-of-coeff-pi
600 (%piargs (cons (car coeff)
601 (rplus 1//2 (cdr coeff))) nil))
602 (cond ((zerop1 sin-of-coeff-pi)
603 0) ;; tan(integer*%pi)
604 ((zerop1 cos-of-coeff-pi)
605 (merror (intl:gettext "tan: ~M isn't in the domain of tan.") x))
606 (cos-of-coeff-pi
607 (div sin-of-coeff-pi cos-of-coeff-pi))))
609 ;; This expression sets x to the coeff of %pi (mod 1) as a side
610 ;; effect and then, if this is zero, returns tan of the
611 ;; rest, because tan has periodicity %pi.
612 ((zerop1 (setq x (mmod (cdr coeff) 1)))
613 (ftake* '%tan zl-rem))
615 ;; Similarly, if x = 1/2 then return -cot(x).
616 ((alike1 1//2 x)
617 (neg (ftake* '%cot zl-rem))))))
619 (def-simplifier csc (y)
620 (let (z)
621 (cond ((flonum-eval (mop form) y))
622 ((and (not (member 'simp (car form))) (big-float-eval (mop form) y)))
623 ((taylorize (mop form) (second form)))
624 ((and $%piargs (cond ((zerop1 y) (domain-error y 'csc))
625 ((has-const-or-int-term y '$%pi) (%piargs-csc/sec y)))))
626 ((and $%iargs (multiplep y '$%i)) (mul -1 '$%i (ftake* '%csch (coeff y '$%i 1))))
627 ((and $triginverses (not (atom y))
628 (cond ((eq '%acsc (setq z (caar y))) (cadr y))
629 ((eq '%asin z) (div 1 (cadr y)))
630 ((eq '%acos z) (div 1 (sqrt1-x^2 (cadr y))))
631 ((eq '%atan z) (div (sqrt1+x^2 (cadr y)) (cadr y)))
632 ((eq '%acot z) (sqrt1+x^2 (cadr y)))
633 ((eq '%asec z) (div (cadr y) (sqrtx^2-1 (cadr y))))
634 ((eq '%atan2 z) (div (sq-sumsq (cadr y) (caddr y)) (cadr y))))))
635 ((and $trigexpand (trigexpand '%csc y)))
636 ($exponentialize (exponentialize '%csc y))
637 ((and $halfangles (halfangle '%csc y)))
638 ((apply-reflection-simp (mop form) y $trigsign))
639 ;((and $trigsign (mminusp* y)) (neg (ftake* '%csc (neg y))))
641 (t (give-up)))))
643 (def-simplifier sec (y)
644 (let (z)
645 (cond ((flonum-eval (mop form) y))
646 ((and (not (member 'simp (car form))) (big-float-eval (mop form) y)))
647 ((taylorize (mop form) (second form)))
648 ((and $%piargs (cond ((zerop1 y) 1)
649 ((has-const-or-int-term y '$%pi) (%piargs-csc/sec (add %pi//2 y))))))
650 ((and $%iargs (multiplep y '$%i)) (ftake* '%sech (coeff y '$%i 1)))
651 ((and $triginverses (not (atom y))
652 (cond ((eq '%asec (setq z (caar y))) (cadr y))
653 ((eq '%asin z) (div 1 (sqrt1-x^2 (cadr y))))
654 ((eq '%acos z) (div 1 (cadr y)))
655 ((eq '%atan z) (sqrt1+x^2 (cadr y)))
656 ((eq '%acot z) (div (sqrt1+x^2 (cadr y)) (cadr y)))
657 ((eq '%acsc z) (div (cadr y) (sqrtx^2-1 (cadr y))))
658 ((eq '%atan2 z) (div (sq-sumsq (cadr y) (caddr y)) (caddr y))))))
659 ((and $trigexpand (trigexpand '%sec y)))
660 ($exponentialize (exponentialize '%sec y))
661 ((and $halfangles (halfangle '%sec y)))
662 ((apply-reflection-simp (mop form) y $trigsign))
663 ;((and $trigsign (mminusp* y)) (ftake* '%sec (neg y)))
665 (t (give-up)))))
667 (defun %piargs-csc/sec (x)
668 (prog ($float coeff ratcoeff zl-rem)
669 (setq ratcoeff (get-const-or-int-terms x '$%pi)
670 coeff (linearize ratcoeff)
671 zl-rem (get-not-const-or-int-terms x '$%pi))
672 (return (cond ((and (zerop1 zl-rem) (setq zl-rem (%piargs coeff nil))) (div 1 zl-rem))
673 ((not (mevenp (car coeff))) nil)
674 ((equal 0 (setq x (mmod (cdr coeff) 2))) (ftake* '%csc zl-rem))
675 ((equal 1 x) (neg (ftake* '%csc zl-rem)))
676 ((alike1 1//2 x) (ftake* '%sec zl-rem))
677 ((alike1 '((rat) 3 2) x) (neg (ftake* '%sec zl-rem)))))))
679 (def-simplifier atan (y)
680 (cond ((flonum-eval (mop form) y))
681 ((and (not (member 'simp (car form))) (big-float-eval (mop form) y)))
682 ((taylorize (mop form) (second form)))
683 ;; Simplification for special values
684 ((zerop1 y) y)
685 ((or (eq y '$inf) (alike1 y '((mtimes) -1 $minf)))
686 (div '$%pi 2))
687 ((or (eq y '$minf) (alike1 y '((mtimes) -1 $inf)))
688 (div '$%pi -2))
689 ((and $%piargs
690 ;; Recognize more special values
691 (cond ((equal 1 y) (div '$%pi 4))
692 ((equal -1 y) (div '$%pi -4))
693 ;; sqrt(3)
694 ((alike1 y '((mexpt) 3 ((rat) 1 2)))
695 (div '$%pi 3))
696 ;; -sqrt(3)
697 ((alike1 y '((mtimes) -1 ((mexpt) 3 ((rat) 1 2))))
698 (div '$%pi -3))
699 ;; 1/sqrt(3)
700 ((alike1 y '((mexpt) 3 ((rat) -1 2)))
701 (div '$%pi 6))
702 ;; -1/sqrt(3)
703 ((alike1 y '((mtimes) -1 ((mexpt) 3 ((rat) -1 2))))
704 (div '$%pi -6))
705 ((alike1 y '((mplus) -1 ((mexpt) 2 ((rat) 1 2))))
706 (div '$%pi 8))
707 ((alike1 y '((mplus) 1 ((mexpt) 2 ((rat) 1 2))))
708 (mul 3 (div '$%pi 8))))))
709 ((and $%iargs (multiplep y '$%i))
710 ;; atan(%i*y) -> %i*atanh(y)
711 (mul '$%i (take '(%atanh) (coeff y '$%i 1))))
712 ((and (not (atom y)) (member (caar y) '(%cot %tan))
713 (if ($constantp (cadr y))
714 (let ((y-val (mfuncall '$mod
715 (if (eq (caar y) '%tan)
716 (cadr y)
717 (sub %pi//2 (cadr y)))
718 '$%pi)))
719 (cond ((eq (mlsp y-val %pi//2) t) y-val)
720 ((eq (mlsp y-val '$%pi) t) (sub y-val '$%pi)))))))
721 ((and (eq $triginverses '$all) (not (atom y))
722 (if (eq (caar y) '%tan) (cadr y))))
723 ((and (eq $triginverses t) (not (atom y)) (eq (caar y) '%tan)
724 ;; Check if y in [-%pi/2, %pi/2]
725 (if (and (member (csign (sub (cadr y) %pi//2)) '($nz $neg) :test #'eq)
726 (member (csign (add (cadr y) %pi//2)) '($pz $pos) :test #'eq))
727 (cadr y))))
728 ($logarc (logarc '%atan y))
729 ((apply-reflection-simp (mop form) y $trigsign))
730 (t (give-up))))
732 (defun %piargs (x ratcoeff)
733 (let (offset-result)
734 (cond ((and (integerp (car x)) (integerp (cdr x))) 0)
735 ((not (mevenp (car x)))
736 (cond ((null ratcoeff) nil)
737 ((and (integerp (car x))
738 (setq offset-result (%piargs-offset (cdr x))))
739 (mul (power -1 (sub ratcoeff (cdr x)))
740 offset-result))))
741 ((%piargs-offset (mmod (cdr x) 2))))))
743 ; simplifies sin(%pi * x) where x is between 0 and 1
744 ; returns nil if can't simplify
745 (defun %piargs-offset (x)
746 (cond ((or (alike1 '((rat) 1 6) x) (alike1 '((rat) 5 6) x)) 1//2)
747 ((or (alike1 '((rat) 1 4) x) (alike1 '((rat) 3 4) x)) (div (power 2 1//2) 2))
748 ((or (alike1 '((rat) 1 3) x) (alike1 '((rat) 2 3) x)) (div (power 3 1//2) 2))
749 ((alike1 1//2 x) 1)
750 ((or (alike1 '((rat) 7 6) x) (alike1 '((rat) 11 6) x)) -1//2)
751 ((or (alike1 '((rat) 4 3) x) (alike1 '((rat) 5 3) x)) (div (power 3 1//2) -2))
752 ((or (alike1 '((rat) 5 4) x) (alike1 '((rat) 7 4) x)) (mul -1//2 (power 2 1//2)))
753 ((alike1 '((rat) 3 2) x) -1)))
755 ;; identifies integer part of form
756 ;; returns (X . Y) if form can be written as X*some_integer + Y
757 ;; returns nil otherwise
758 (defun linearize (form)
759 (cond ((integerp form) (cons 0 form))
760 ((numberp form) nil)
761 ((atom form)
762 (let (dum)
763 (cond ((setq dum (evod form))
764 (if (eq '$even dum) '(2 . 0) '(2 . 1)))
765 ((maxima-integerp form) '(1 . 0)))))
766 ((eq 'rat (caar form)) (cons 0 form))
767 ((eq 'mplus (caar form)) (lin-mplus form))
768 ((eq 'mtimes (caar form)) (lin-mtimes form))
769 ((eq 'mexpt (caar form)) (lin-mexpt form))))
771 (defun lin-mplus (form)
772 (do ((tl (cdr form) (cdr tl)) (dummy) (coeff 0) (zl-rem 0))
773 ((null tl) (cons coeff (mmod zl-rem coeff)))
774 (setq dummy (linearize (car tl)))
775 (if (null dummy) (return nil)
776 (setq coeff (rgcd (car dummy) coeff) zl-rem (rplus (cdr dummy) zl-rem)))))
778 (defun lin-mtimes (form)
779 (do ((fl (cdr form) (cdr fl)) (dummy) (coeff 0) (zl-rem 1))
780 ((null fl) (cons coeff (mmod zl-rem coeff)))
781 (setq dummy (linearize (car fl)))
782 (cond ((null dummy) (return nil))
783 (t (setq coeff (rgcd (rtimes coeff (car dummy))
784 (rgcd (rtimes coeff (cdr dummy)) (rtimes zl-rem (car dummy))))
785 zl-rem (rtimes (cdr dummy) zl-rem))))))
787 (defun lin-mexpt (form)
788 (prog (dummy)
789 (cond ((and (integerp (caddr form)) (not (minusp (caddr form)))
790 (not (null (setq dummy (linearize (cadr form))))))
791 (return (cons (car dummy) (mmod (cdr dummy) (caddr form))))))))
793 (defun rgcd (x y)
794 (cond ((integerp x)
795 (cond ((integerp y) (gcd x y))
796 (t (list '(rat) (gcd x (cadr y)) (caddr y)))))
797 ((integerp y) (list '(rat) (gcd (cadr x) y) (caddr x)))
798 (t (list '(rat) (gcd (cadr x) (cadr y)) (lcm (caddr x) (caddr y))))))
800 (defun maxima-reduce (x y)
801 (prog (gcd)
802 (setq gcd (gcd x y) x (truncate x gcd) y (truncate y gcd))
803 (if (minusp y) (setq x (- x) y (- y)))
804 (return (if (eql y 1) x (list '(rat simp) x y)))))
806 ;; The following four functions are generated in code by TRANSL. - JPG 2/1/81
808 (defun rplus (x y) (addk x y))
810 (defun rdifference (x y) (addk x (timesk -1 y)))
812 (defun rtimes (x y) (timesk x y))
814 (defun rremainder (x y)
815 (cond ((equal 0 y) (dbz-err))
816 ((integerp x)
817 (cond ((integerp y) (maxima-reduce x y))
818 (t (maxima-reduce (* x (caddr y)) (cadr y)))))
819 ((integerp y) (maxima-reduce (cadr x) (* (caddr x) y)))
820 (t (maxima-reduce (* (cadr x) (caddr y)) (* (caddr x) (cadr y))))))
822 (defmfun $exponentialize (exp)
823 (let ($demoivre)
824 (cond ((atom exp) exp)
825 ((trigp (caar exp))
826 (exponentialize (caar exp) ($exponentialize (cadr exp))))
827 (t (recur-apply #'$exponentialize exp)))))
829 (defun exponentialize (op arg)
830 (cond ((eq '%sin op)
831 (div (sub (power '$%e (mul '$%i arg)) (power '$%e (mul -1 '$%i arg)))
832 (mul 2 '$%i)))
833 ((eq '%cos op)
834 (div (add (power '$%e (mul '$%i arg)) (power '$%e (mul -1 '$%i arg))) 2))
835 ((eq '%tan op)
836 (div (sub (power '$%e (mul '$%i arg)) (power '$%e (mul -1 '$%i arg)))
837 (mul '$%i (add (power '$%e (mul '$%i arg))
838 (power '$%e (mul -1 '$%i arg))))))
839 ((eq '%cot op)
840 (div (mul '$%i (add (power '$%e (mul '$%i arg))
841 (power '$%e (mul -1 '$%i arg))))
842 (sub (power '$%e (mul '$%i arg)) (power '$%e (mul -1 '$%i arg)))))
843 ((eq '%csc op)
844 (div (mul 2 '$%i)
845 (sub (power '$%e (mul '$%i arg)) (power '$%e (mul -1 '$%i arg)))))
846 ((eq '%sec op)
847 (div 2 (add (power '$%e (mul '$%i arg)) (power '$%e (mul -1 '$%i arg)))))
848 ((eq '%sinh op)
849 (div (sub (power '$%e arg) (power '$%e (neg arg))) 2))
850 ((eq '%cosh op)
851 (div (add (power '$%e arg) (power '$%e (mul -1 arg))) 2))
852 ((eq '%tanh op)
853 (div (sub (power '$%e arg) (power '$%e (neg arg)))
854 (add (power '$%e arg) (power '$%e (mul -1 arg)))))
855 ((eq '%coth op)
856 (div (add (power '$%e arg) (power '$%e (mul -1 arg)))
857 (sub (power '$%e arg) (power '$%e (neg arg)))))
858 ((eq '%csch op)
859 (div 2 (sub (power '$%e arg) (power '$%e (neg arg)))))
860 ((eq '%sech op)
861 (div 2 (add (power '$%e arg) (power '$%e (mul -1 arg)))))))
863 (defun coefficient (exp var pow)
864 (coeff exp var pow))
866 (defun mmod (x mod)
867 (cond ((and (integerp x) (integerp mod))
868 (if (minusp (if (zerop mod) x (setq x (- x (* mod (truncate x mod))))))
869 (+ x mod)
871 ((and ($ratnump x) ($ratnump mod))
872 (let
873 ((d (lcm ($denom x) ($denom mod))))
874 (setq x (mul* d x))
875 (setq mod (mul* d mod))
876 (div (mod x mod) d)))
877 (t nil)))
879 (defun multiplep (exp var)
880 (and (not (zerop1 exp)) (zerop1 (sub exp (mul var (coeff exp var 1))))))
882 (defun linearp (exp var)
883 (and (setq exp (islinear exp var)) (not (equal (car exp) 0))))
885 (defun mminusp (x)
886 (= -1 (signum1 x)))
888 (defun mminusp* (x)
889 (let (sign)
890 (setq sign (csign x))
891 (or (member sign '($neg $nz) :test #'eq)
892 (and (mminusp x) (not (member sign '($pos $pz) :test #'eq))))))
894 ;; This should give more information somehow.
896 (defun dbz-err ()
897 (cond ((not errorsw) (merror (intl:gettext "Division by zero attempted.")))
898 (t (throw 'errorsw t))))
900 (defun dbz-err1 (func)
901 (cond ((not errorsw) (merror (intl:gettext "~A: division by zero attempted.") func))
902 (t (throw 'errorsw t))))