;;; -*- Mode: Lisp; Package: Maxima; Syntax: Common-Lisp; Base: 10 -*- ;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;; The data in this file contains enhancments. ;;;;; ;;; ;;;;; ;;; Copyright (c) 1984,1987 by William Schelter,University of Texas ;;;;; ;;; All rights reserved ;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;; (c) Copyright 1982 Massachusetts Institute of Technology ;;; ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; (in-package :maxima) (macsyma-module float) ;; EXPERIMENTAL BIGFLOAT PACKAGE VERSION 2- USING BINARY MANTISSA ;; AND POWER-OF-2 EXPONENT. ;; EXPONENTS MAY BE BIG NUMBERS NOW (AUG. 1975 --RJF) ;; Modified: July 1979 by CWH to run on the Lisp Machine and to comment ;; the code. ;; August 1980 by CWH to run on Multics and to install ;; new FIXFLOAT. ;; December 1980 by JIM to fix BIGLSH not to pass LSH a second ;; argument with magnitude greater than MACHINE-FIXNUM-PRECISION. ;; Number of bits of precision in a fixnum and in the fields of a flonum for ;; a particular machine. These variables should only be around at eval ;; and compile time. These variables should probably be set up in a prelude ;; file so they can be accessible to all Macsyma files. (eval-when #+gcl (compile load eval) #-gcl (:compile-toplevel :load-toplevel :execute) (defconstant +machine-fixnum-precision+ (integer-length most-positive-fixnum))) ;; External variables (defmvar $float2bf t "If TRUE, no MAXIMA-ERROR message is printed when a floating point number is converted to a bigfloat number.") (defmvar $bftorat nil "Controls the conversion of bigfloat numbers to rational numbers. If FALSE, RATEPSILON will be used to control the conversion (this results in relatively small rational numbers). If TRUE, the rational number generated will accurately represent the bigfloat.") (defmvar $bftrunc t "If TRUE, printing of bigfloat numbers will truncate trailing zeroes. Otherwise, all trailing zeroes are printed.") (defmvar $fpprintprec 0 "Controls the number of significant digits printed for floats. If 0, then full precision is used." fixnum) (defmvar $fpprec 16. "Number of decimal digits of precision to use when creating new bigfloats. One extra decimal digit in actual representation for rounding purposes.") (defmvar bigfloatzero '((bigfloat simp 56.) 0 0) "Bigfloat representation of 0" in-core) (defmvar bigfloatone '((bigfloat simp 56.) #.(expt 2 55.) 1) "Bigfloat representation of 1" in-core) (defmvar bfhalf '((bigfloat simp 56.) #.(expt 2 55.) 0) "Bigfloat representation of 1/2") (defmvar bfmhalf '((bigfloat simp 56.) #.(- (expt 2 55.)) 0) "Bigfloat representation of -1/2") (defmvar bigfloat%e '((bigfloat simp 56.) 48968212118944587. 2) "Bigfloat representation of %E") (defmvar bigfloat%pi '((bigfloat simp 56.) 56593902016227522. 2) "Bigfloat representation of %pi") (defmvar bigfloat%gamma '((bigfloat simp 56.) 41592772053807304. 0) "Bigfloat representation of %gamma") ;; Internal specials ;; Number of bits of precision in the mantissa of newly created bigfloats. ;; FPPREC = ($FPPREC+1)*(Log base 2 of 10) (defvar fpprec) ;; FPROUND uses this to return a second value, i.e. it sets it before ;; returning. This number represents the number of binary digits its input ;; bignum had to be shifted right to be aligned into the mantissa. For ;; example, aligning 1 would mean shifting it FPPREC-1 places left, and ;; aligning 7 would mean shifting FPPREC-3 places left. (defvar *m) ;; *DECFP = T if the computation is being done in decimal radix. NIL implies ;; base 2. Decimal radix is used only during output. (defvar *decfp nil) (defvar max-bfloat-%pi bigfloat%pi) (defvar max-bfloat-%e bigfloat%e) (defvar max-bfloat-%gamma bigfloat%gamma) (declare-top (special *cancelled $float $bfloat $ratprint $ratepsilon $domain $m1pbranch adjust)) ;; Representation of a Bigfloat: ((BIGFLOAT SIMP precision) mantissa exponent) ;; precision -- number of bits of precision in the mantissa. ;; precision = (integer-length mantissa) ;; mantissa -- a signed integer representing a fractional portion computed by ;; fraction = (// mantissa (^ 2 precision)). ;; exponent -- a signed integer representing the scale of the number. ;; The actual number represented is (* fraction (^ 2 exponent)). (defun hipart (x nn) (if (bignump nn) (abs x) (haipart x nn))) (defun fpprec1 (assign-var q) (declare (ignore assign-var)) (if (or (not (fixnump q)) (< q 1)) (merror "Improper value for `fpprec':~%~M" q)) (setq fpprec (+ 2 (integer-length (expt 10. q))) bigfloatone ($bfloat 1) bigfloatzero ($bfloat 0) bfhalf (list (car bigfloatone) (cadr bigfloatone) 0) bfmhalf (list (car bigfloatone) (- (cadr bigfloatone)) 0)) q) ;; FPSCAN is called by lexical scan when a ;; bigfloat is encountered. For example, 12.01B-3 ;; would be the result of (FPSCAN '(/1 /2) '(/0 /1) '(/- /3)) ;; Arguments to FPSCAN are a list of characters to the left of the ;; decimal point, to the right of the decimal point, and in the exponent. (defun fpscan (lft rt exp &aux (*read-base* 10.) (*m 1) (*cancelled 0)) (setq exp (readlist exp)) (bigfloatp (let ((fpprec (+ 4 fpprec (integer-length exp) (floor (1+ (* #.(/ (log 10d0) (log 2d0)) (length lft)))))) $float temp) (setq temp (add (readlist lft) (div (readlist rt) (expt 10. (length rt))))) ($bfloat (cond ((> (abs exp) 1000.) (cons '(mtimes) (list temp (list '(mexpt) 10. exp)))) (t (mul2 temp (power 10. exp)))))))) (defun dim-bigfloat (form result) (let (($lispdisp nil)) (dimension-atom (maknam (fpformat form)) result))) (defun fpformat (l) (if (not (member 'simp (cdar l) :test #'eq)) (setq l (cons (cons (caar l) (cons 'simp (cdar l))) (cdr l)))) (cond ((equal (cadr l) 0) (if (not (equal (caddr l) 0)) (mtell "Warning - an incorrect form for 0.0b0 has been generated.")) (list '|0| '|.| '|0| '|b| '|0|)) (t ;; L IS ALWAYS POSITIVE FP NUMBER (let ((extradigs (floor (1+ (quotient (integer-length (caddr l)) #.(/ (log 10d0) (log 2d0)))))) (*m 1) (*cancelled 0)) (setq l ((lambda (*decfp fpprec of l expon) (setq expon (- (cadr l) of)) (setq l (if (minusp expon) (fpquotient (intofp (car l)) (fpintexpt 2 (- expon) of)) (fptimes* (intofp (car l)) (fpintexpt 2 expon of)))) (incf fpprec (- extradigs)) (list (fpround (car l)) (+ (- extradigs) *m (cadr l)))) t (+ extradigs (decimalsin (- (caddar l) 2))) (caddar l) (cdr l) nil))) (let ((*print-base* 10.) *print-radix* (l1 nil)) (setq l1 (if (not $bftrunc) (explodec (car l)) (do ((l (nreverse (explodec (car l))) (cdr l))) ((not (eq '|0| (car l))) (nreverse l))))) (nconc (ncons (car l1)) (ncons '|.|) (or (and (cdr l1) (cond ((or (zerop $fpprintprec) (not (< $fpprintprec $fpprec)) (null (cddr l1))) (cdr l1)) (t (setq l1 (cdr l1)) (do ((i $fpprintprec (1- i)) (l2)) ((or (< i 2) (null (cdr l1))) (cond ((not $bftrunc) (nreverse l2)) (t (do ((l3 l2 (cdr l3))) ((not (eq '|0| (car l3))) (nreverse l3)))))) (setq l2 (cons (car l1) l2) l1 (cdr l1)))))) (ncons '|0|)) (ncons '|b|) (explodec (1- (cadr l)))))))) (defun bigfloatp (x) (prog nil (cond ((not ($bfloatp x)) (return nil)) ((= fpprec (caddar x)) (return x)) ((> fpprec (caddar x)) (setq x (bcons (list (fpshift (cadr x) (- fpprec (caddar x))) (caddr x))))) (t (setq x (bcons (list (fpround (cadr x)) (+ (caddr x) *m fpprec (- (caddar x)))))))) (return (if (equal (cadr x) 0) (bcons (list 0 0)) x)))) (defun bigfloat2rat (x) (setq x (bigfloatp x)) ((lambda ($float2bf exp y sign) (setq exp (cond ((minusp (cadr x)) (setq sign t y (fpration1 (cons (car x) (fpabs (cdr x))))) (rplaca y (* -1 (car y)))) (t (fpration1 x)))) (cond ($ratprint (princ "`rat' replaced ") (when sign (princ "-")) (princ (maknam (fpformat (cons (car x) (fpabs (cdr x)))))) (princ " by ") (princ (car exp)) (write-char #\/) (princ (cdr exp)) (princ " = ") (setq x ($bfloat (list '(rat simp) (car exp) (cdr exp)))) (when sign (princ "-")) (princ (maknam (fpformat (cons (car x) (fpabs (cdr x)))))) (terpri))) exp) t nil nil nil)) (defun fpration1 (x) ((lambda (fprateps) (or (and (equal x bigfloatzero) (cons 0 1)) (prog (y a) (return (do ((xx x (setq y (invertbigfloat (bcons (fpdifference (cdr xx) (cdr ($bfloat a))))))) (num (setq a (fpentier x)) (+ (* (setq a (fpentier y)) num) onum)) (den 1 (+ (* a den) oden)) (onum 1 num) (oden 0 den)) ((and (not (zerop den)) (not (fpgreaterp (fpabs (fpquotient (fpdifference (cdr x) (fpquotient (cdr ($bfloat num)) (cdr ($bfloat den)))) (cdr x))) fprateps))) (cons num den))))))) (cdr ($bfloat (if $bftorat (list '(rat simp) 1 (exptrl 2 (1- fpprec))) $ratepsilon))))) ;; Convert a floating point number into a bigfloat. (defun floattofp (x) (unless $float2bf (mtell "Warning: Float to bigfloat conversion of ~S~%" x)) (multiple-value-bind (frac exp sign) (integer-decode-float x) ;; Scale frac to the desired number of bits, and adjust the ;; exponent accordingly. (let ((scale (- fpprec (integer-length frac)))) (list (ash (* sign frac) scale) (+ fpprec (- exp scale)))))) ;; Convert a bigfloat into a floating point number. (defmfun fp2flo (l) (let ((precision (caddar l)) (mantissa (cadr l)) (exponent (caddr l)) (fpprec #.machine-mantissa-precision) (*m 0)) ;;Round the mantissa to the number of bits of precision of the machine, ;;and then convert it to a floating point fraction. (setq mantissa (quotient (fpround mantissa) #.(expt 2d0 machine-mantissa-precision))) ;; Multiply the mantissa by the exponent portion. I'm not sure ;; why the exponent computation is so complicated. Using ;; scale-float will prevent possible overflow unless the result ;; really would. (setq precision (errset (scale-float mantissa (+ exponent (- precision) *m #.machine-mantissa-precision)) nil)) (if precision (car precision) (merror "Floating point overflow in converting ~:M to flonum" l)))) ;; New machine-independent version of FIXFLOAT. This may be buggy. - CWH ;; It is buggy! On the PDP10 it dies on (RATIONALIZE -1.16066076E-7) ;; which calls FLOAT on some rather big numbers. ($RATEPSILON is approx. ;; 7.45E-9) - JPG (defun fixfloat (x) (let (($ratepsilon #.(expt 2.0 (- machine-mantissa-precision)))) (maxima-rationalize x))) ;; Takes a flonum arg and returns a rational number corresponding to the flonum ;; in the form of a dotted pair of two integers. Since the denominator will ;; always be a positive power of 2, this number will not always be in lowest ;; terms. (defun bcons (s) `((bigfloat simp ,fpprec) . ,s)) (defmfun $bfloat (x) (let (y) (cond ((bigfloatp x)) ((or (numberp x) (member x '($%e $%pi $%gamma) :test #'eq)) (bcons (intofp x))) ((or (atom x) (member 'array (cdar x) :test #'eq)) (if (eq x '$%phi) ($bfloat '((mtimes simp) ((rat simp) 1 2) ((mplus simp) 1 ((mexpt simp) 5 ((rat simp) 1 2))))) x)) ((eq (caar x) 'mexpt) (if (equal (cadr x) '$%e) (*fpexp (caddr x)) (exptbigfloat ($bfloat (cadr x)) (caddr x)))) ((eq (caar x) 'mncexpt) (list '(mncexpt) ($bfloat (cadr x)) (caddr x))) ((setq y (safe-get (caar x) 'floatprog)) (funcall y (mapcar #'$bfloat (cdr x)))) ((or (trigp (caar x)) (arcp (caar x)) (eq (caar x) '$entier)) (setq y ($bfloat (cadr x))) (if ($bfloatp y) (cond ((eq (caar x) '$entier) ($entier y)) ((arcp (caar x)) (setq y ($bfloat (logarc (caar x) y))) (if (free y '$%i) y (let ($ratprint) (fparcsimp ($rectform y))))) ((member (caar x) '(%cot %sec %csc) :test #'eq) (invertbigfloat ($bfloat (list (ncons (safe-get (caar x) 'recip)) y)))) (t ($bfloat (exponentialize (caar x) y)))) (subst0 (list (ncons (caar x)) y) x))) (t (recur-apply #'$bfloat x))))) (defprop mplus addbigfloat floatprog) (defprop mtimes timesbigfloat floatprog) (defprop %sin sinbigfloat floatprog) (defprop %cos cosbigfloat floatprog) (defprop rat ratbigfloat floatprog) (defprop %atan atanbigfloat floatprog) (defprop %tan tanbigfloat floatprog) (defprop %log logbigfloat floatprog) (defprop mabs mabsbigfloat floatprog) (defmfun addbigfloat (h) (prog (fans tst r nfans) (setq fans (setq tst bigfloatzero) nfans 0) (do ((l h (cdr l))) ((null l)) (cond ((setq r (bigfloatp (car l))) (setq fans (bcons (fpplus (cdr r) (cdr fans))))) (t (setq nfans (list '(mplus) (car l) nfans))))) (return (cond ((equal nfans 0) fans) ((equal fans tst) nfans) (t (simplify (list '(mplus) fans nfans))))))) (defmfun ratbigfloat (l) (bcons (fpquotient (cdar l) (cdadr l)))) (defun decimalsin (x) (do ((i (quotient (* 59. x) 196.) (1+ i))) ;log[10](2)=.301029 (nil) (when (> (integer-length (expt 10. i)) x) (return (1- i))))) (defmfun atanbigfloat (x) (*fpatan (car x) (cdr x))) (defmfun *fpatan (a y) (fpend (let ((fpprec (+ 8. fpprec))) (if (null y) (if ($bfloatp a) (fpatan (cdr ($bfloat a))) (list '(%atan) a)) (fpatan2 (cdr ($bfloat a)) (cdr ($bfloat (car y)))))))) ;; Bigfloat atan (defun fpatan (x) (prog (term x2 ans oans one two tmp) (setq one (intofp 1) two (intofp 2)) (cond ((fpgreaterp (fpabs x) one) ;; |x| > 1. ;; ;; Use A&S 4.4.5: ;; atan(x) + acot(x) = +/- pi/2 (+ for x >= 0, - for x < 0) ;; ;; and A&S 4.4.8 ;; acot(z) = atan(1/z) (setq tmp (fpquotient (fppi) two)) (setq ans (fpdifference tmp (fpatan (fpquotient one x)))) (return (cond ((fpgreaterp ans tmp) (fpdifference ans (fppi))) (t ans)))) ((fpgreaterp (fpabs x) (fpquotient one two)) ;; |x| > 1/2 ;; ;; Use A&S 4.4.42, third formula: ;; ;; atan(z) = z/(1+z^2)*[1 + 2/3*r + (2*4)/(3*5)*r^2 + ...] ;; ;; r = z^2/(1+z^2) (setq tmp (fpquotient x (fpplus (fptimes* x x) one))) (setq x2 (fptimes* x tmp) term (setq ans one)) (do ((n 0 (1+ n))) ((equal ans oans)) (setq term (fptimes* term (fptimes* x2 (fpquotient (intofp (+ 2 (* 2 n))) (intofp (+ (* 2 n) 3)))))) (setq oans ans ans (fpplus term ans))) (setq ans (fptimes* tmp ans))) (t ;; |x| <= 1/2. Use Taylor series (A&S 4.4.42, first ;; formula). (setq ans x x2 (fpminus (fptimes* x x)) term x) (do ((n 3 (+ n 2))) ((equal ans oans)) (setq term (fptimes* term x2)) (setq oans ans ans (fpplus ans (fpquotient term (intofp n))))))) (return ans))) ;; atan(y/x) taking into account the quadrant. (Also equal to ;; arg(x+%i*y).) (defun fpatan2 (y x) (cond ((equal (car x) 0) ;; atan(y/0) = atan(inf), but what sign? (cond ((equal (car y) 0) (merror "atan(0//0) has been generated.")) ((minusp (car y)) ;; We're on the negative imaginary axis, so -pi/2. (fpquotient (fppi) (intofp -2))) (t ;; The positive imaginary axis, so +pi/2 (fpquotient (fppi) (intofp 2))))) ((signp g (car x)) ;; x > 0. atan(y/x) is the correct value. (fpatan (fpquotient y x))) ((signp g (car y)) ;; x < 0, and y > 0. We're in quadrant II, so the angle we ;; want is pi+atan(y/x). (fpplus (fppi) (fpatan (fpquotient y x)))) (t ;; x <= 0 and y <= 0. We're in quadrant III, so the angle we ;; want is atan(y/x)-pi. (fpdifference (fpatan (fpquotient y x)) (fppi))))) (defun tanbigfloat (a) (setq a (car a)) (fpend (let ((fpprec (+ 8. fpprec))) (cond (($bfloatp a) (setq a (cdr ($bfloat a))) (fpquotient (fpsin a t) (fpsin a nil))) (t (list '(%tan) a)))))) ;; Returns a list of a mantissa and an exponent. (defun intofp (l) (cond ((not (atom l)) ($bfloat l)) ((floatp l) (floattofp l)) ((equal 0 l) '(0 0)) ((eq l '$%pi) (fppi)) ((eq l '$%e) (fpe)) ((eq l '$%gamma) (fpgamma)) (t (list (fpround l) (+ *m fpprec))))) ;; It seems to me that this function gets called on an integer ;; and returns the mantissa portion of the mantissa/exponent pair. ;; "STICKY BIT" CALCULATION FIXED 10/14/75 --RJF ;; BASE must not get temporarily bound to NIL by being placed ;; in a PROG list as this will confuse stepping programs. (defun fpround (l &aux (*print-base* 10.) *print-radix*) (prog () (cond ((null *decfp) ;;*M will be positive if the precision of the argument is greater than ;;the current precision being used. (setq *m (- (integer-length l) fpprec)) (when (= *m 0) (setq *cancelled 0) (return l)) ;;FPSHIFT is essentially LSH. (setq adjust (fpshift 1 (1- *m))) (when (minusp l) (setq adjust (- adjust))) (incf l adjust) (setq *m (- (integer-length l) fpprec)) (setq *cancelled (abs *m)) (cond ((zerop (hipart l (- *m))) ;ONLY ZEROES SHIFTED OFF (return (fpshift (fpshift l (- -1 *m)) 1))) ; ROUND TO MAKE EVEN (t (return (fpshift l (- *m)))))) (t (setq *m (- (flatsize (abs l)) fpprec)) (setq adjust (fpshift 1 (1- *m))) (when (minusp l) (setq adjust (- adjust))) (setq adjust (* 5 adjust)) (setq *m (- (flatsize (abs (setq l (+ l adjust)))) fpprec)) (return (fpshift l (- *m))))))) ;; Compute (* L (expt d n)) where D is 2 or 10 depending on ;; *decfp. Throw away an fractional part by truncating to zero. (defun fpshift (l n) (cond ((null *decfp) (cond ((and (minusp n) (minusp l)) ;; Left shift of negative number requires some ;; care. (That is, (truncate l (expt 2 n)), but use ;; shifts instead.) (- (ash (- l) n))) (t (ash l n)))) ((> n 0) (* l (expt 10. n))) ((< n 0.) (quotient l (expt 10. (- n)))) (t l))) ;; Bignum LSH -- N is assumed (and declared above) to be a fixnum. ;; This isn't really LSH, since the sign bit isn't propagated when ;; shifting to the right, i.e. (BIGLSH -100 -3) = -40, whereas ;; (LSH -100 -3) = 777777777770 (on a 36 bit machine). ;; This actually computes (* X (EXPT 2 N)). As of 12/21/80, this function ;; was only called by FPSHIFT. I would like to hear an argument as why this ;; is more efficient than simply writing (* X (EXPT 2 N)). Is the ;; intermediate result created by (EXPT 2 N) the problem? I assume that ;; EXPT tries to LSH when possible. (defun biglsh (x n) (cond ((and (not (bignump x)) (< n #.(- +machine-fixnum-precision+))) 0) ;; Either we are shifting a fixnum to the right, or shifting ;; a fixnum to the left, but not far enough left for it to become ;; a bignum. ((and (not (bignump x)) (or (<= n 0) (< (+ (integer-length x) n) #.+machine-fixnum-precision+))) ;; The form which follows is nearly identical to (ASH X N), however ;; (ASH -100 -20) = -1, whereas (BIGLSH -100 -20) = 0. (if (>= x 0) (ash x n) (- (biglsh (- x) n)))) ;(- x) may be a bignum even is x is a fixnum. ;; If we get here, then either X is a bignum or our answer is ;; going to be a bignum. ((< n 0) (cond ((> (abs n) (integer-length x)) 0) ((> x 0) (hipart x (+ (integer-length x) n))) (t (- (hipart x (+ (integer-length x) n)))))) ((= n 0) x) ;; Isn't this the kind of optimization that compilers are ;; supposed to make? ((< n #.(1- +machine-fixnum-precision+)) (* x (ash 1 n))) (t (* x (expt 2 n))))) ;; exp(x) ;; ;; For negative x, use exp(-x) = 1/exp(x) ;; ;; For x > 0, exp(x) = exp(r+y) = exp(r) * exp(y), where x = r + y and ;; r = floor(x). (defun fpexp (x) (prog (r s) (if (not (signp ge (car x))) (return (fpquotient (fpone) (fpexp (fpabs x))))) (setq r (fpintpart x)) (return (cond ((< r 2) (fpexp1 x)) (t (setq s (fpexp1 (fpdifference x (intofp r)))) (fptimes* s (cdr (bigfloatp ((lambda (fpprec r) (bcons (fpexpt (fpe) r))) ; patch for full precision %E (+ fpprec (integer-length r) -1) r))))))))) ;; exp(x) for small x, using Taylor series. (defun fpexp1 (x) (prog (term ans oans) (setq ans (setq term (fpone))) (do ((n 1 (1+ n))) ((equal ans oans)) (setq term (fpquotient (fptimes* x term) (intofp n))) (setq oans ans) (setq ans (fpplus ans term))) (return ans))) ;; Does one higher precision to round correctly. ;; A and B are each a list of a mantissa and an exponent. (defun fpquotient (a b) (cond ((equal (car b) 0) (merror "`pquotient' by zero")) ((equal (car a) 0) '(0 0)) (t (list (fpround (quotient (fpshift (car a) (+ 3 fpprec)) (car b))) (+ -3 (- (cadr a) (cadr b)) *m))))) (defun fpgreaterp (a b) (fpposp (fpdifference a b))) (defun fplessp (a b) (fpposp (fpdifference b a))) (defun fpposp (x) (> (car x) 0)) (defmfun fpmin (arg1 &rest args) (let ((min arg1)) (mapc #'(lambda (u) (if (fplessp u min) (setq min u))) args) min)) (defmfun fpmax (arg1 &rest args) (let ((max arg1)) (mapc #'(lambda (u) (if (fpgreaterp u max) (setq max u))) args) max)) ;; (FPE) RETURN BIG FLOATING POINT %E. IT RETURNS (CDR BIGFLOAT%E) IF RIGHT ;; PRECISION. IT RETURNS TRUNCATED BIGFLOAT%E IF POSSIBLE, ELSE RECOMPUTES. ;; IN ANY CASE, BIGFLOAT%E IS SET TO LAST USED VALUE. (defun fpe () (cond ((= fpprec (caddar bigfloat%e)) (cdr bigfloat%e)) ((< fpprec (caddar bigfloat%e)) (cdr (setq bigfloat%e (bigfloatp bigfloat%e)))) ((< fpprec (caddar max-bfloat-%e)) (cdr (setq bigfloat%e (bigfloatp max-bfloat-%e)))) (t (cdr (setq max-bfloat-%e (setq bigfloat%e (fpe1))))))) (defun fppi () (cond ((= fpprec (caddar bigfloat%pi)) (cdr bigfloat%pi)) ((< fpprec (caddar bigfloat%pi)) (cdr (setq bigfloat%pi (bigfloatp bigfloat%pi)))) ((< fpprec (caddar max-bfloat-%pi)) (cdr (setq bigfloat%pi (bigfloatp max-bfloat-%pi)))) (t (cdr (setq max-bfloat-%pi (setq bigfloat%pi (fppi1))))))) (defun fpgamma () (cond ((= fpprec (caddar bigfloat%gamma)) (cdr bigfloat%gamma)) ((< fpprec (caddar bigfloat%gamma)) (cdr (setq bigfloat%gamma (bigfloatp bigfloat%gamma)))) ((< fpprec (caddar max-bfloat-%gamma)) (cdr (setq bigfloat%gamma (bigfloatp max-bfloat-%gamma)))) (t (cdr (setq max-bfloat-%gamma (setq bigfloat%gamma (fpgamma1))))))) (defun fpone () (cond (*decfp (intofp 1)) ((= fpprec (caddar bigfloatone)) (cdr bigfloatone)) (t (intofp 1)))) ;;....................................................................................................... ;; ;; ;; (fpe1) returns a bigfloat approximation to E. ;; fpe1 is the bigfloat part of the bfloat(%e) computation ;; (defun fpe1 nil (bcons (list (fpround (compe (+ fpprec 12))) (+ -12 *m)))) ;; ;; compe is the bignum part of the bfloat(%e) computation ;; (compe N)/(2.0^N) is an approximation to E ;; The algorithm is based on the series ;; ;; %e = sum( 1/i! ,i,0,inf ) ;; ;; but sums up 1001 terms to one. ;; (defun compe (prec) (let (s h (n 1) d (k 1001)) (setq h (ash 1 prec)) (setq s h) (do ((i k (+ i k))) ((zerop h)) (setq d (do ((j 1 (1+ j)) (p i)) ((> j (1- k)) (* p n)) (setq p (* p (- i j)))) ) (setq n (do ((j (- k 2) (1- j)) (p 1)) ((< j 0) p) (setq p (1+ (* p (- i j))))) ) (setq h (*quo (* h n) d)) (setq s (+ s h))) s)) ;;................................................................................ Volker van Nek 2007 .. ;; ;;....................................................................................................... ;; ;; ;; (fppi1) returns a bigfloat approximation to PI. ;; fppi1 is the bigfloat part of the bfloat(%pi) computation ;; (defun fppi1 nil (bcons (fpquotient (fprt18231_) (list (fpround (comppi (+ fpprec 12))) (+ -12 *m)) ))) ;; ;; comppi is the bignum part of the bfloat(%pi) computation ;; (comppi N)/(2.0^N) is an approximation to 640320^(3/2)/12 * 1/PI ;; ;; Chudnovsky & Chudnovsky (1987): ;; ;; 640320^(3/2) / (12 * %pi) = ;; ;; sum( (-1)^i*(6*i)!*(545140134*i+13591409) / (i!^3*(3*i)!*640320^(3*i)) ,i,0,inf ) ;; (defun comppi (prec) (let (s h n d) (setq s (ash 13591409 prec)) (setq h (neg (*quo (ash 67047785160 prec) 262537412640768000))) (setq s (+ s h)) (do ((i 2 (1+ i))) ((zerop h)) (setq n (* 12 (- (* 6 i) 5) (- (* 6 i) 4) (- (* 2 i) 1) (- (* 6 i) 1) (+ (* i 545140134) 13591409) )) (setq d (* (- (* 3 i) 2) (expt i 3) (- (* i 545140134) 531548725) 262537412640768000)) (setq h (neg (*quo (* h n) d))) (setq s (+ s h))) s )) ;; ;; fprt18231_ computes sqrt(640320^3/12^2). ;; n[0] n[i+1] = n[i]^2+a*d[i]^2 n[inf] ;; quadratic Heron algorithm: x[0] = ----, , sqrt(a) = ------ ;; d[0] d[i+1] = 2*n[i]*d[i] d[inf] (defun fprt18231_ nil (let (a n d h) (setq a 1823176476672000) (setq n 42698670666) (setq d 1000) (do ((prec 10 (* 2 prec))) ((> prec $fpprec)) (setq h n) (setq n (+ (* n n) (* a d d))) (setq d (* 2 h d)) ) (fpquotient (intofp n) (intofp d)) )) ;;................................................................................ Volker van Nek 2007 .. ;; ;; Compute the main part of the Euler-Mascheroni constant using the ;; Bessel function approach. See ;; http://numbers.computation.free.fr/Constants/Gamma/gamma.html for a ;; description of the algorithm. ;; Roughly, we have ;; ;; %gamma = A(N)/B(N) - log(N) + O(e^(-4*N)) ;; ;; where ;; ;; ;; a*N ;; A(N) = sum (N^2/n!)^2*H(n) ;; n=0 ;; ;; a*N ;; B(N) = sum (N^2/n!)^2 ;; n=0 ;; ;; n ;; H(n) = sum 1/k ;; k=1 ;; ;; with H(0) = 0 ;; ;; and a = 3.591121476668622136649223 where a*(log(a)-1) = 1. ;; ;; This formula can be easily justified by looking at the value ;; K0(2*N)/I0(2*N), where K0 and I0 are the modified Bessel functions. ;; From A&S 9.6.12 and 9.6.13, We see that ;; ;; inf ;; I0(2*N) = sum (N^2/n!)^2 ;; n=0 ;; ;; ;; inf ;; K0(2*N) = -(log(N) + %gamma)*I0(2*N) + sum (N^2/n!)^2*H(n) ;; n=0 ;; ;; So ;; ;; K0(2*N)/I0(2*N) = -log(N) - %gamma + C ;; ;; where ;; ;; C = [sum (N^2/n!)^2*H(n)]/sum (N^2/n!)^2 ;; ;; or ;; ;; For N large, A&S gives ;; ;; I0(2*N) = exp(2*N)/sqrt(4*%pi*N) ;; ;; K0(2*N) = sqrt(%pi/(4*N))*exp(-2*N) ;; ;; So K0(2*N)/I0(2*N) = %pi*exp(-4*N) and ;; ;; O(exp(-4*N)) = -log(N) - %gamma + C ;; ;; or ;; ;; %gamma = C - log(N) + O(exp(-4*N)) ;; ;; And C is approximately A(N)/B(N) if we take enough terms in the ;; sum. ;; (defun comp-bf%gamma (prec) ;; Prec is the number of digits we want. We assume the remainder is ;; really e^(-4*N) and not O(e^(-4*N)). So choose N such that ;; exp(-4*N) is less than the number of digits of precision we want. ;; ;; We also assume don't need a really precise value of beta because ;; our N's are not so big that we need more. (let* ((fpprec prec) (big-n (floor (* 1/4 prec (log 2d0)))) (big-n-sq (intofp (* big-n big-n))) (beta 3.591121476668622136649223d0) (limit (floor (* beta big-n))) (one (fpone)) (term (intofp 1)) (harmonic (intofp 0)) (a-sum (intofp 0)) (b-sum (intofp 1))) (do ((n 1 (1+ n))) ((> n limit)) (let ((bf-n (intofp n))) (setf term (fpquotient (fptimes* term big-n-sq) (fptimes* bf-n bf-n))) (setf harmonic (fpplus harmonic (fpquotient one bf-n))) (setf a-sum (fpplus a-sum (fptimes* term harmonic))) (setf b-sum (fpplus b-sum term)))) (fpplus (fpquotient a-sum b-sum) (fpminus (fplog (intofp big-n)))))) (defun fpgamma1 () ;; Use a few extra bits of precision (bcons (list (fpround (first (comp-bf%gamma (+ fpprec 8)))) 0))) (defun fpdifference (a b) (fpplus a (fpminus b))) (defun fpminus (x) (if (equal (car x) 0) x (list (- (car x)) (cadr x)))) (defun fpplus (a b) (prog (*m exp man sticky) (setq *cancelled 0) (cond ((equal (car a) 0) (return b)) ((equal (car b) 0) (return a))) (setq exp (- (cadr a) (cadr b))) (setq man (cond ((equal exp 0) (setq sticky 0) (fpshift (+ (car a) (car b)) 2)) ((> exp 0) (setq sticky (hipart (car b) (- 1 exp))) (setq sticky (cond ((signp e sticky) 0) ((signp l (car b)) -1) (t 1))) ; COMPUTE STICKY BIT (+ (fpshift (car a) 2) ; MAKE ROOM FOR GUARD DIGIT & STICKY BIT (fpshift (car b) (- 2 exp)))) (t (setq sticky (hipart (car a) (1+ exp))) (setq sticky (cond ((signp e sticky) 0) ((signp l (car a)) -1) (t 1))) (+ (fpshift (car b) 2) (fpshift (car a) (+ 2 exp)))))) (setq man (+ man sticky)) (return (cond ((equal man 0) '(0 0)) (t (setq man (fpround man)) (setq exp (+ -2 *m (max (cadr a) (cadr b)))) (list man exp)))))) (defun fptimes* (a b) (if (or (zerop (car a)) (zerop (car b))) '(0 0) (list (fpround (* (car a) (car b))) (+ *m (cadr a) (cadr b) (- fpprec))))) ;; Don't use the symbol BASE since it is SPECIAL. (defun fpintexpt (int nn fixprec) ;INT is integer (setq fixprec (truncate fixprec (1- (integer-length int)))) ;NN is pos (let ((bas (intofp (expt int (min nn fixprec))))) (if (> nn fixprec) (fptimes* (intofp (expt int (rem nn fixprec))) (fpexpt bas (quotient nn fixprec))) bas))) ;; NN is positive or negative integer (defun fpexpt (p nn) (cond ((zerop nn) (fpone)) ((eql nn 1) p) ((< nn 0) (fpquotient (fpone) (fpexpt p (- nn)))) (t (prog (u) (if (oddp nn) (setq u p) (setq u (fpone))) (do ((ii (quotient nn 2) (quotient ii 2))) ((zerop ii)) (setq p (fptimes* p p)) (when (oddp ii) (setq u (fptimes* u p)))) (return u))))) (defun exptbigfloat (p n) (cond ((equal n 1) p) ((equal n 0) ($bfloat 1)) ((not ($bfloatp p)) (list '(mexpt) p n)) ((equal (cadr p) 0) ($bfloat 0)) ((and (< (cadr p) 0) (ratnump n)) ($bfloat ($expand (list '(mtimes) ($bfloat ((lambda ($domain $m1pbranch) (power -1 n)) '$complex t)) (exptbigfloat (bcons (fpminus (cdr p))) n))))) ((and (< (cadr p) 0) (not (integerp n))) (cond ((or (equal n 0.5) (equal n bfhalf)) (exptbigfloat p '((rat simp) 1 2))) ((or (equal n -0.5) (equal n bfmhalf)) (exptbigfloat p '((rat simp) -1 2))) (($bfloatp (setq n ($bfloat n))) (cond ((equal n ($bfloat (fpentier n))) (exptbigfloat p (fpentier n))) (t ;; for P<0: P^N = (-P)^N*cos(pi*N) + i*(-P)^N*sin(pi*N) (setq p (exptbigfloat (bcons (fpminus (cdr p))) n) n ($bfloat `((mtimes) $%pi ,n))) (add2 ($bfloat `((mtimes) ,p ,(*fpsin n nil))) `((mtimes simp) ,($bfloat `((mtimes) ,p ,(*fpsin n t))) $%i))))) (t (list '(mexpt) p n)))) ((and (ratnump n) (< (caddr n) 10.)) (bcons (fpexpt (fproot p (caddr n)) (cadr n)))) ((not (integerp n)) (setq n ($bfloat n)) (cond ((not ($bfloatp n)) (list '(mexpt) p n)) (t ((lambda (extrabits) (setq p ((lambda (fpprec) (fpexp (fptimes* (cdr (bigfloatp n)) (fplog (cdr (bigfloatp p)))))) (+ extrabits fpprec))) (setq p (list (fpround (car p)) (+ (- extrabits) *m (cadr p)))) (bcons p)) (max 1 (+ (caddr n) (integer-length (caddr p)))))))) ; The number of extra bits required ((< n 0) (invertbigfloat (exptbigfloat p (- n)))) (t (bcons (fpexpt (cdr p) n))))) (defun fproot (a n) ; computes a^(1/n) see Fitch, SIGSAM Bull Nov 74 ;; Special case for a = 0b0. General algorithm loops endlessly in that case. ;; Unlike many or maybe all of the other functions named FP-something, ;; FPROOT assumes it is called with an argument like ;; '((BIGFLOAT ...) FOO BAR) instead of '(FOO BAR). ;; However FPROOT does return something like '(FOO BAR). (if (eq (cadr a) 0) '(0 0) (progn (let* ((ofprec fpprec) (fpprec (+ fpprec 2)) ;assumes a>0 n>=2 (bk (fpexpt (intofp 2) (1+ (quotient (cadr (setq a (cdr (bigfloatp a)))) n))))) (do ((x bk (fpdifference x (setq bk (fpquotient (fpdifference x (fpquotient a (fpexpt x n1))) n)))) (n1 (1- n)) (n (intofp n))) ((or (equal bk '(0 0)) (> (- (cadr x) (cadr bk)) ofprec)) (setq a x)))) (list (fpround (car a)) (+ -2 *m (cadr a)))))) (defun timesbigfloat (h) (prog (fans r nfans) (setq fans (bcons (fpone)) nfans 1) (do ((l h (cdr l))) ((null l)) (if (setq r (bigfloatp (car l))) (setq fans (bcons (fptimes* (cdr r) (cdr fans)))) (setq nfans (list '(mtimes) (car l) nfans)))) (return (if (equal nfans 1) fans (simplify (list '(mtimes) fans nfans)))))) (defun invertbigfloat (a) (if (bigfloatp a) (bcons (fpquotient (fpone) (cdr a))) (simplify (list '(mexpt) a -1)))) (defun *fpexp (a) (fpend (let ((fpprec (+ 8. fpprec))) (if ($bfloatp (setq a ($bfloat a))) (fpexp (cdr a)) (list '(mexpt) '$%e a))))) (defun *fpsin (a fl) (fpend (let ((fpprec (+ 8. fpprec))) (cond (($bfloatp a) (fpsin (cdr ($bfloat a)) fl)) (fl (list '(%sin) a)) (t (list '(%cos) a)))))) (defun fpend (a) (cond ((equal (car a) 0) (bcons a)) ((numberp (car a)) (setq a (list (fpround (car a)) (+ -8. *m (cadr a)))) (bcons a)) (t a))) (defun fparcsimp (e) ; needed for e.g. ASIN(.123567812345678B0) with ;; FPPREC 16, to get rid of the miniscule imaginary ;; part of the a+bi answer. (if (and (mplusp e) (null (cdddr e)) (mtimesp (caddr e)) (null (cdddr (caddr e))) ($bfloatp (cadr (caddr e))) (eq (caddr (caddr e)) '$%i) (< (caddr (cadr (caddr e))) (+ (- fpprec) 2))) (cadr e) e)) (defun sinbigfloat (x) (*fpsin (car x) t)) (defun cosbigfloat (x) (*fpsin (car x) nil)) ;; THIS VERSION OF FPSIN COMPUTES SIN OR COS TO PRECISION FPPREC, ;; BUT CHECKS FOR THE POSSIBILITY OF CATASTROPHIC CANCELLATION DURING ;; ARGUMENT REDUCTION (E.G. SIN(N*%PI+EPSILON)) ;; *FPSINCHECK* WILL CAUSE PRINTOUT OF ADDITIONAL INFO WHEN ;; EXTRA PRECISION IS NEEDED FOR SIN/COS CALCULATION. KNOWN ;; BAD FEATURES: IT IS NOT NECESSARY TO USE EXTRA PRECISION FOR, E.G. ;; SIN(PI/2), WHICH IS NOT NEAR ZERO, BUT EXTRA ;; PRECISION IS USED SINCE IT IS NEEDED FOR COS(PI/2). ;; PRECISION SEEMS TO BE 100% SATSIFACTORY FOR LARGE ARGUMENTS, E.G. ;; SIN(31415926.0B0), BUT LESS SO FOR SIN(3.1415926B0). EXPLANATION ;; NOT KNOWN. (9/12/75 RJF) (defvar *fpsincheck* nil) (defun fpsin (x fl) (prog (piby2 r sign res k *cancelled) (setq sign (cond (fl (signp g (car x))) (t)) x (fpabs x)) (when (equal (car x) 0) (return (if fl (intofp 0) (intofp 1)))) (return (cdr (bigfloatp ((lambda (fpprec xt *cancelled oldprec) (prog (x) loop (setq x (cdr (bigfloatp xt))) (setq piby2 (fpquotient (fppi) (intofp 2))) (setq r (fpintpart (fpquotient x piby2))) (setq x (fpplus x (fptimes* (intofp (- r)) piby2))) (setq k *cancelled) (fpplus x (fpminus piby2)) (setq *cancelled (max k *cancelled)) (when *fpsincheck* (print `(*canc= ,*cancelled fpprec= ,fpprec oldprec= ,oldprec))) (cond ((not (> oldprec (- fpprec *cancelled))) (setq r (rem r 4)) (setq res (cond (fl (cond ((= r 0) (fpsin1 x)) ((= r 1) (fpcos1 x)) ((= r 2) (fpminus (fpsin1 x))) ((= r 3) (fpminus (fpcos1 x))))) (t (cond ((= r 0) (fpcos1 x)) ((= r 1) (fpminus (fpsin1 x))) ((= r 2) (fpminus (fpcos1 x))) ((= r 3) (fpsin1 x)))))) (return (bcons (if sign res (fpminus res))))) (t (incf fpprec *cancelled) (go loop))))) (max fpprec (+ fpprec (cadr x))) (bcons x) 0 fpprec)))))) (defun fpcos1 (x) (fpsincos1 x nil)) ;; Compute SIN or COS in (0,PI/2). FL is T for SIN, NIL for COS. ;; ;; Use Taylor series (defun fpsincos1 (x fl) (prog (ans term oans x2) (setq ans (if fl x (intofp 1)) x2 (fpminus(fptimes* x x))) (setq term ans) (do ((n (if fl 3 2) (+ n 2))) ((equal ans oans)) (setq term (fptimes* term (fpquotient x2 (intofp (* n (1- n)))))) (setq oans ans ans (fpplus ans term))) (return ans))) (defun fpsin1(x) (fpsincos1 x t)) (defun fpabs (x) (if (signp ge (car x)) x (cons (- (car x)) (cdr x)))) (defmfun fpentier (f) (let ((fpprec (caddar f))) (fpintpart (cdr f)))) (defun fpintpart (f) (prog (m) (setq m (- fpprec (cadr f))) (return (if (> m 0) (quotient (car f) (expt 2 m)) (* (car f) (expt 2 (- m))))))) (defun logbigfloat (a) ((lambda (minus) (setq a ((lambda (fpprec) (cond (($bfloatp (car a)) (setq a ($bfloat (car a))) (cond ((zerop (cadr a)) (merror "log(0.0b0) has been generated")) ((minusp (cadr a)) (setq minus t) (fplog (list (- (cadr a)) (caddr a)))) (t (fplog (cdr a))))) (t (list '(%log) (car a))))) (+ 2 fpprec))) (when (numberp (car a)) (setq a (if (zerop (car a)) (list 0 0) (list (fpround (car a)) (+ -2 *m (cadr a))))) (setq a (bcons a))) (if minus (add a (mul '$%i ($bfloat '$%pi))) a)) nil)) ;;; Computes the log of a bigfloat number. ;;; ;;; Uses the series ;;; ;;; log(1+x) = sum((x/(x+2))^(2*n+1)/(2*n+1),n,0,inf); ;;; ;;; ;;; INF x 2 n + 1 ;;; ==== (-----) ;;; \ x + 2 ;;; = 2 > -------------- ;;; / 2 n + 1 ;;; ==== ;;; n = 0 ;;; ;;; ;;; which converges for x > 0. ;;; ;;; Note that FPLOG is given 1+X, not X. ;;; ;;; However, to aid convergence of the series, we scale 1+x until 1/e ;;; < 1+x <= e. ;;; (defun fplog (x) (prog (over two ans oldans term e sum) (unless (> (car x) 0) (merror "Non-positive argument to `fplog'")) (setq e (fpe) over (fpquotient (fpone) e) ans 0) ;; Scale X until 1/e < X <= E. ANS keeps track of how ;; many factors of E were used. Set X to NIL if X is E. (do () (nil) (cond ((equal x e) (setq x nil) (return nil)) ((and (fplessp x e) (fplessp over x)) (return nil)) ((fplessp x over) (setq x (fptimes* x e)) (decf ans)) (t (incf ans) (setq x (fpquotient x e))))) (when (null x) (return (intofp (1+ ans)))) ;; Prepare X for the series. The series is for 1 + x, so ;; get x from our X. TERM is (x/(x+2)). X becomes ;; (x/(x+2))^2. (setq x (fpdifference x (fpone)) ans (intofp ans)) (setq x (fpexpt (setq term (fpquotient x (fpplus x (setq two (intofp 2))))) 2)) ;; Sum the series until the sum (in ANS) doesn't change ;; anymore. (setq sum (intofp 0)) (do ((n 1 (+ n 2))) ((equal sum oldans)) (setq oldans sum) (setq sum (fpplus sum (fpquotient term (intofp n)))) (setq term (fptimes* term x))) (return (fpplus ans (fptimes* two sum))))) (defun mabsbigfloat (l) (prog (r) (setq r (bigfloatp (car l))) (return (if (null r) (list '(mabs) (car l)) (bcons (fpabs (cdr r))))))) ;;;; Bigfloat implementations of special functions. ;;;; ;;; This is still a bit messy. Some functions here take bigfloat ;;; numbers, represented by ((bigfloat) ), but others want ;;; just the FP number, represented by ( ). Likewise, some ;;; return a bigfloat, some return just the FP. ;;; ;;; This needs to be systemized somehow. It isn't helped by the fact ;;; that some of the routines above also do the samething. ;;; ;;; The implementation for the special functions for a complex ;;; argument are mostly taken from W. Kahan, "Branch Cuts for Complex ;;; Elementary Functions or Much Ado About Nothing's Sign Bit", in ;;; Iserles and Powell (eds.) "The State of the Art in Numerical ;;; Analysis", pp 165-211, Clarendon Press, 1987 ;; Compute exp(x) - 1, but do it carefully to preserve precision when ;; |x| is small. X is a FP number, and a FP number is returned. That ;; is, no bigfloat stuff. (defun fpexpm1 (x) ;; What is the right breakpoint here? Is 1 ok? Perhaps 1/e is better? (cond ((fpgreaterp (fpabs x) (fpone)) ;; exp(x) - 1 (fpdifference (fpexp x) (fpone))) (t ;; Use Taylor series for exp(x) - 1 (let ((ans x) (oans nil) (term x)) (do ((n 2 (1+ n))) ((equal ans oans)) (setf term (fpquotient (fptimes* x term) (intofp n))) (setf oans ans) (setf ans (fpplus ans term))) ans)))) ;; log(1+x) for small x. X is FP number, and a FP number is returned. (defun fplog1p (x) ;; Use the same series as given above for fplog. For small x we use ;; the series, otherwise fplog is accurate enough. (cond ((fpgreaterp (fpabs x) (fpone)) (fplog (fpplus x (fpone)))) (t (let* ((sum (intofp 0)) (term (fpquotient x (fpplus x (intofp 2)))) (f (fptimes* term term)) (oldans nil)) (do ((n 1 (+ n 2))) ((equal sum oldans)) (setq oldans sum) (setq sum (fpplus sum (fpquotient term (intofp n)))) (setq term (fptimes* term f))) (fptimes* sum (intofp 2)))))) ;; sinh(x) for real x. X is a bigfloat, and a bigfloat is returned. (defun fpsinh (x) ;; X must be a maxima bigfloat ;; See, for example, Hart et al., Computer Approximations, 6.2.27: ;; ;; sinh(x) = 1/2*(D(x) + D(x)/(1+D(x))) ;; ;; where D(x) = exp(x) - 1. (let ((d (fpexpm1 (cdr (bigfloatp x))))) (bcons (fpquotient (fpplus d (fpquotient d (fpplus d (fpone)))) (intofp 2))))) (defun big-float-sinh (x &optional y) ;; The rectform for sinh for complex args should be numerically ;; accurate, so return nil in that case. (unless y (fpsinh x))) ;; asinh(x) for real x. X is a bigfloat, and a bigfloat is returned. (defun fpasinh (x) ;; asinh(x) = sign(x) * log(|x| + sqrt(1+x*x)) ;; ;; And ;; ;; asinh(x) = x, if 1+x*x = 1 ;; = sign(x) * (log(2) + log(x)), large |x| ;; = sign(x) * log(2*|x| + 1/(|x|+sqrt(1+x*x))), if |x| > 2 ;; = sign(x) * log1p(|x|+x^2/(1+sqrt(1+x*x))), otherwise. ;; ;; But I'm lazy right now and we only implement the last 2 cases. ;; We should implement all cases. (let* ((fp-x (cdr (bigfloatp x))) (absx (fpabs fp-x)) (one (fpone)) (two (intofp 2)) (minus (minusp (car fp-x))) result) ;; We only use two formulas here. |x| <= 2 and |x| > 2. Should ;; we add one for very big x and one for very small x, as given above. (cond ((fpgreaterp absx two) ;; |x| > 2 ;; ;; log(2*|x| + 1/(|x|+sqrt(1+x^2))) (setf result (fplog (fpplus (fptimes* absx two) (fpquotient one (fpplus absx (fproot (bcons (fpplus one (fptimes* absx absx))) 2))))))) (t ;; |x| <= 2 ;; ;; log1p(|x|+x^2/(1+sqrt(1+x^2))) (let ((x*x (fptimes* absx absx))) (setq result (fplog1p (fpplus absx (fpquotient x*x (fpplus one (fproot (bcons (fpplus one x*x)) 2))))))))) (if minus (bcons (fpminus result)) (bcons result)))) (defun complex-asinh (x y) ;; asinh(z) = -%i * asin(%i*z) (multiple-value-bind (u v) (complex-asin (mul -1 y) x) (values v (bcons (fpminus (cdr u)))))) (defun big-float-asinh (x &optional y) (if y (multiple-value-bind (u v) (complex-asinh x y) (add u (mul '$%i v))) (fpasinh x))) (defun fpasin-core (x) ;; asin(x) = atan(x/(sqrt(1-x^2)) ;; = sgn(x)*[%pi/2 - atan(sqrt(1-x^2)/abs(x))] ;; ;; Use the first for 0 <= x < 1/2 and the latter for 1/2 < x <= 1. ;; ;; If |x| > 1, we need to do something else. ;; ;; asin(x) = -%i*log(sqrt(1-x^2)+%i*x) ;; = -%i*log(%i*x + %i*sqrt(x^2-1)) ;; = -%i*[log(|x + sqrt(x^2-1)|) + %i*%pi/2] ;; = %pi/2 - %i*log(|x+sqrt(x^2-1)|) (let ((fp-x (cdr (bigfloatp x)))) (cond ((minusp (car fp-x)) ;; asin(-x) = -asin(x); (mul -1 (fpasin (bcons (fpminus fp-x))))) ((fplessp fp-x (cdr bfhalf)) ;; 0 <= x < 1/2 ;; asin(x) = atan(x/sqrt(1-x^2)) (bcons (fpatan (fpquotient fp-x (fproot (bcons (fptimes* (fpdifference (fpone) fp-x) (fpplus (fpone) fp-x))) 2))))) ((fpgreaterp fp-x (fpone)) ;; x > 1 ;; asin(x) = %pi/2 - %i*log(|x+sqrt(x^2-1)|) ;; ;; Should we try to do something a little fancier with the ;; argument to log and use log1p for better accuracy? (let ((arg (fpplus fp-x (fproot (bcons (fptimes* (fpdifference fp-x (fpone)) (fpplus fp-x (fpone)))) 2)))) (add (div '$%pi 2) (mul -1 '$%i (bcons (fplog arg)))))) (t ;; 1/2 <= x <= 1 ;; asin(x) = %pi/2 - atan(sqrt(1-x^2)/x) (add (div '$%pi 2) (mul -1 (bcons (fpatan (fpquotient (fproot (bcons (fptimes* (fpdifference (fpone) fp-x) (fpplus (fpone) fp-x))) 2) fp-x))))))))) ;; asin(x) for real x. X is a bigfloat, and a maxima number (real or ;; complex) is returned. (defun fpasin (x) ;; asin(x) = atan(x/(sqrt(1-x^2)) ;; = sgn(x)*[%pi/2 - atan(sqrt(1-x^2)/abs(x))] ;; ;; Use the first for 0 <= x < 1/2 and the latter for 1/2 < x <= 1. ;; ;; If |x| > 1, we need to do something else. ;; ;; asin(x) = -%i*log(sqrt(1-x^2)+%i*x) ;; = -%i*log(%i*x + %i*sqrt(x^2-1)) ;; = -%i*[log(|x + sqrt(x^2-1)|) + %i*%pi/2] ;; = %pi/2 - %i*log(|x+sqrt(x^2-1)|) ($bfloat (fpasin-core x))) ;; Square root of a complex number (xx, yy). Both are bigfloats. FP ;; (non-bigfloat) numbers are returned. (defun complex-sqrt (xx yy) (let* ((x (cdr (bigfloatp xx))) (y (cdr (bigfloatp yy))) (rho (fpplus (fptimes* x x) (fptimes* y y)))) (setf rho (fpplus (fpabs x) (fproot (bcons rho) 2))) (setf rho (fpplus rho rho)) (setf rho (fpquotient (fproot (bcons rho) 2) (intofp 2))) (let ((eta rho) (nu y)) (when (fpgreaterp rho (intofp 0)) (setf nu (fpquotient (fpquotient nu rho) (intofp 2))) (when (fplessp x (intofp 0)) (setf eta (fpabs nu)) (setf nu (if (minusp (car y)) (fpminus rho) rho)))) (values eta nu)))) ;; asin(z) for complex z = x + %i*y. X and Y are bigfloats. The real ;; and imaginary parts are returned as bigfloat numbers. (defun complex-asin (x y) (let ((x (cdr (bigfloatp x))) (y (cdr (bigfloatp y)))) (multiple-value-bind (re-sqrt-1-z im-sqrt-1-z) (complex-sqrt (bcons (fpdifference (intofp 1) x)) (bcons (fpminus y))) (multiple-value-bind (re-sqrt-1+z im-sqrt-1+z) (complex-sqrt (bcons (fpplus (intofp 1) x)) (bcons y)) ;; Realpart is atan(x/Re(sqrt(1-z)*sqrt(1+z))) ;; Imagpart is asinh(Im(conj(sqrt(1-z))*sqrt(1+z))) (values (bcons (fpatan (fpquotient x (fpdifference (fptimes* re-sqrt-1-z re-sqrt-1+z) (fptimes* im-sqrt-1-z im-sqrt-1+z))))) (fpasinh (bcons (fpdifference (fptimes* re-sqrt-1-z im-sqrt-1+z) (fptimes* im-sqrt-1-z re-sqrt-1+z))))))))) (defun big-float-asin (x &optional y) (if y (multiple-value-bind (u v) (complex-asin x y) (add u (mul '$%i v))) (fpasin x))) ;; tanh(x) for real x. X is a bigfloat, and a bigfloat is returned. (defun fptanh (x) ;; X is Maxima bigfloat ;; tanh(x) = D(2*x)/(2+D(2*x)) (let* ((two (intofp 2)) (fp (cdr (bigfloatp x))) (d (fpexpm1 (fptimes* fp two)))) (bcons (fpquotient d (fpplus d two))))) ;; tanh(z), z = x + %i*y. X, Y are bigfloats, and a maxima number is ;; returned. (defun complex-tanh (x y) (let* ((tv (cdr (tanbigfloat (list y)))) (beta (fpplus (fpone) (fptimes* tv tv))) (s (cdr (fpsinh x))) (s^2 (fptimes* s s)) (rho (fproot (bcons (fpplus (fpone) s^2)) 2)) (den (fpplus (fpone) (fptimes* beta s^2)))) (values (bcons (fpquotient (fptimes* beta (fptimes* rho s)) den)) (bcons (fpquotient tv den))))) (defun big-float-tanh (x &optional y) (if y (multiple-value-bind (u v) (complex-tanh x y) (add u (mul '$%i v))) (fptanh x))) ;; atanh(x) for real x, |x| <= 1. X is a bigfloat, and a bigfloat is ;; returned. (defun fpatanh (x) ;; atanh(x) = -atanh(-x) ;; = 1/2*log1p(2*x/(1-x)), x >= 0.5 ;; = 1/2*log1p(2*x+2*x*x/(1-x)), x <= 0.5 (let* ((fp-x (cdr (bigfloatp x)))) (cond ((fplessp fp-x (intofp 0)) ;; atanh(x) = -atanh(-x) (mul -1 (fpatanh (bcons (fpminus fp-x))))) ((fpgreaterp fp-x (fpone)) ;; x > 1, so use complex version. (multiple-value-bind (u v) (complex-atanh x (bcons (intofp 0))) (add u (mul '$%i v)))) ((fpgreaterp fp-x (cdr bfhalf)) ;; atanh(x) = 1/2*log1p(2*x/(1-x)) (bcons (fptimes* bfhalf (fplog1p (fpquotient (fptimes* (intofp 2) fp-x) (fpdifference (fpone) fp-x)))))) (t ;; atanh(x) = 1/2*log1p(2*x + 2*x*x/(1-x)) (let ((2x (fptimes* (intofp 2) fp-x))) (bcons (fptimes* (cdr bfhalf) (fplog1p (fpplus 2x (fpquotient (fptimes* 2x fp-x) (fpdifference (fpone) fp-x))))))))))) ;; Stuff which follows is derived from atanh z = (log(1 + z) - log(1 - z))/2 ;; which apparently originates with Kahan's "Much ado" paper. ;; The formulas for eta and nu below can be easily derived from ;; rectform(atanh(x+%i*y)) = ;; ;; 1/4*log(((1+x)^2+y^2)/((1-x)^2+y^2)) + %i/2*(arg(1+x+%i*y)-arg(1-x+%i*(-y))) ;; ;; Expand the argument of log out and divide it out and we get ;; ;; log(((1+x)^2+y^2)/((1-x)^2+y^2)) = log(1+4*x/((1-x)^2+y^2)) ;; ;; When y = 0, Im atanh z = 1/2 (arg(1 + x) - arg(1 - x)) ;; = if x < -1 then %pi/2 else if x > 1 then -%pi/2 else ;; ;; Otherwise, arg(1 - x + %i*(-y)) = - arg(1 - x + %i*y), ;; and Im atanh z = 1/2 (arg(1 + x + %i*y) + arg(1 - x + %i*y)). ;; Since arg(x)+arg(y) = arg(x*y) (almost), we can simplify the ;; imaginary part to ;; ;; arg((1+x+%i*y)*(1-x+%i*y)) = arg((1-x)*(1+x)-y^2+2*y*%i) ;; = atan2(2*y,((1-x)*(1+x)-y^2)) ;; ;; These are the eta and nu forms below. (defun complex-atanh (x y) (let* ((fpx (cdr (bigfloatp x))) (fpy (cdr (bigfloatp y))) (beta (if (minusp (car fpx)) (fpminus (fpone)) (fpone))) (x-lt-minus-1 (mevalp `((mlessp) ,x -1))) (x-gt-plus-1 (mevalp `((mgreaterp) ,x 1))) (y-equals-0 (like y '((bigfloat) 0 0))) (x (fptimes* beta fpx)) (y (fptimes* beta (fpminus fpy))) ;; Kahan has rho = 4/most-positive-float. What should we do ;; here about that? Our big floats don't really have a ;; most-positive float value. (rho (intofp 0)) (t1 (fpplus (fpabs y) rho)) (t1^2 (fptimes* t1 t1)) (1-x (fpdifference (fpone) x)) ;; eta = log(1+4*x/((1-x)^2+y^2))/4 (eta (fpquotient (fplog1p (fpquotient (fptimes* (intofp 4) x) (fpplus (fptimes* 1-x 1-x) t1^2))) (intofp 4))) ;; If y = 0, then Im atanh z = %pi/2 or -%pi/2. ;; Otherwise nu = 1/2*atan2(2*y,(1-x)*(1+x)-y^2) (nu (if y-equals-0 ;; EXTRA FPMINUS HERE TO COUNTERACT FPMINUS IN RETURN VALUE (fpminus (if x-lt-minus-1 (cdr ($bfloat '((mquotient) $%pi 2))) (if x-gt-plus-1 (cdr ($bfloat '((mminus) ((mquotient) $%pi 2)))) (merror "COMPLEX-ATANH: HOW DID I GET HERE?")))) (fptimes* (cdr bfhalf) (fpatan2 (fptimes* (intofp 2) y) (fpdifference (fptimes* 1-x (fpplus (fpone) x)) t1^2)))))) (values (bcons (fptimes* beta eta)) ;; WTF IS FPMINUS DOING HERE ?? (bcons (fpminus (fptimes* beta nu)))))) (defun big-float-atanh (x &optional y) (if y (multiple-value-bind (u v) (complex-atanh x y) (add u (mul '$%i v))) (fpatanh x))) ;; acos(x) for real x. X is a bigfloat, and a maxima number is returned. (defun fpacos (x) ;; acos(x) = %pi/2 - asin(x) ($bfloat (add (div '$%pi 2) (mul -1 (fpasin-core x))))) (defun complex-acos (x y) (let ((x (cdr (bigfloatp x))) (y (cdr (bigfloatp y)))) (multiple-value-bind (re-sqrt-1-z im-sqrt-1-z) (complex-sqrt (bcons (fpdifference (intofp 1) x)) (bcons (fpminus y))) (multiple-value-bind (re-sqrt-1+z im-sqrt-1+z) (complex-sqrt (bcons (fpplus (intofp 1) x)) (bcons y)) (values (bcons (fptimes* (intofp 2) (fpatan (fpquotient re-sqrt-1-z re-sqrt-1+z)))) (fpasinh (bcons (fpdifference (fptimes* re-sqrt-1+z im-sqrt-1-z) (fptimes* im-sqrt-1+z re-sqrt-1-z))))))))) (defun big-float-acos (x &optional y) (if y (multiple-value-bind (u v) (complex-acos x y) (add u (mul '$%i v))) (fpacos x))) (defun complex-log (x y) (let* ((x (cdr (bigfloatp x))) (y (cdr (bigfloatp y))) (t1 (let (($float2bf t)) ;; No warning message, please. (floattofp 1.2d0))) (t2 (intofp 3)) (rho (fpplus (fptimes* x x) (fptimes* y y))) (abs-x (fpabs x)) (abs-y (fpabs y)) (beta (fpmax abs-x abs-y)) (theta (fpmin abs-x abs-y))) (values (if (or (fpgreaterp t1 beta) (fplessp rho t2)) (fpquotient (fplog1p (fpplus (fptimes* (fpdifference beta (fpone)) (fpplus beta (fpone))) (fptimes* theta theta))) (intofp 2)) (fpquotient (fplog rho) (intofp 2))) (fpatan2 y x)))) (defun big-float-log (x &optional y) (if y (multiple-value-bind (u v) (complex-log x y) (add (bcons u) (mul '$%i (bcons v)))) (let ((fp-x (cdr (bigfloatp x)))) (if (fplessp fp-x (intofp 0)) ;; ??? Do we want to return an exact %i*%pi or a float ;; approximation? (add (bcons (fplog (fpminus fp-x))) (mul '$%i (bcons (fppi)))) (bcons (fplog fp-x)))))) (defun big-float-sqrt (x &optional y) (if y (multiple-value-bind (u v) (complex-sqrt x y) (add (bcons u) (mul '$%i (bcons v)))) (let ((fp-x (cdr (bigfloatp x)))) (if (fplessp fp-x (intofp 0)) (mul '$%i (bcons (fproot (bcons (fpminus fp-x)) 2))) (bcons (fproot x 2)))))) (eval-when #+gcl (load) #-gcl (:load-toplevel) (fpprec1 nil $fpprec)) ; Set up user's precision