xref: /openbsd-src/lib/libm/man/exp.3 (revision db3296cf5c1dd9058ceecc3a29fe4aaa0bd26000)
1.\"	$OpenBSD: exp.3,v 1.16 2003/06/02 20:18:41 millert Exp $
2.\"
3.\" Copyright (c) 1985, 1991 Regents of the University of California.
4.\" All rights reserved.
5.\"
6.\" Redistribution and use in source and binary forms, with or without
7.\" modification, are permitted provided that the following conditions
8.\" are met:
9.\" 1. Redistributions of source code must retain the above copyright
10.\"    notice, this list of conditions and the following disclaimer.
11.\" 2. Redistributions in binary form must reproduce the above copyright
12.\"    notice, this list of conditions and the following disclaimer in the
13.\"    documentation and/or other materials provided with the distribution.
14.\" 3. Neither the name of the University nor the names of its contributors
15.\"    may be used to endorse or promote products derived from this software
16.\"    without specific prior written permission.
17.\"
18.\" THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
19.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21.\" ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
22.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28.\" SUCH DAMAGE.
29.\"
30.\"     from: @(#)exp.3	6.12 (Berkeley) 7/31/91
31.\"
32.Dd July 31, 1991
33.Dt EXP 3
34.Os
35.Sh NAME
36.Nm exp ,
37.Nm expf ,
38.Nm expm1 ,
39.Nm expm1f ,
40.Nm log ,
41.Nm logf ,
42.Nm log10 ,
43.Nm log10f ,
44.Nm log1p ,
45.Nm log1pf ,
46.Nm pow ,
47.Nm powf
48.Nd exponential, logarithm, power functions
49.Sh SYNOPSIS
50.Fd #include <math.h>
51.Ft double
52.Fn exp "double x"
53.Ft float
54.Fn expf "float x"
55.Ft double
56.Fn expm1 "double x"
57.Ft float
58.Fn expm1f "float x"
59.Ft double
60.Fn log "double x"
61.Ft float
62.Fn logf "float x"
63.Ft double
64.Fn log10 "double x"
65.Ft float
66.Fn log10f "float x"
67.Ft double
68.Fn log1p "double x"
69.Ft float
70.Fn log1pf "float x"
71.Ft double
72.Fn pow "double x" "double y"
73.Ft float
74.Fn powf "float x" "float y"
75.Sh DESCRIPTION
76The
77.Fn exp
78function computes the exponential value of the given argument
79.Fa x .
80The
81.Fn expf
82function is a single precision version of
83.Fn exp .
84.Pp
85The
86.Fn expm1
87function computes the value exp(x)\-1 accurately even for tiny argument
88.Fa x .
89The
90.Fn expm1f
91function is a single precision version of
92.Fn expm1 .
93.Pp
94The
95.Fn log
96function computes the value of the natural logarithm of argument
97.Fa x .
98The
99.Fn logf
100function is a single precision version of
101.Fn log .
102.Pp
103The
104.Fn log10
105function computes the value of the logarithm of argument
106.Fa x
107to base 10.
108The
109.Fn log10f
110function is a single precision version of
111.Fn log10 .
112.Pp
113The
114.Fn log1p
115function computes
116the value of log(1+x) accurately even for tiny argument
117.Fa x .
118The
119.Fn log1pf
120function is a single precision version of
121.Fn log1p .
122.Pp
123The
124.Fn pow
125function computes the value of
126.Ar x
127to the exponent
128.Ar y .
129The
130.Fn powf
131function is a single precision version of
132.Fn pow .
133.Sh RETURN VALUES
134These functions will return the appropriate computation unless an error
135occurs or an argument is out of range.
136The functions
137.Fn exp ,
138.Fn expm1
139and
140.Fn pow
141detect if the computed value will overflow,
142set the global variable
143.Va errno
144to
145.Er ERANGE
146and cause a reserved operand fault on a
147.Tn VAX
148or
149.Tn Tahoe .
150The function
151.Fn pow x y
152checks to see if
153.Fa x
154< 0 and
155.Fa y
156is not an integer, in the event this is true,
157the global variable
158.Va errno
159is set to
160.Er EDOM
161and on the
162.Tn VAX
163and
164.Tn Tahoe
165generate a reserved operand fault.
166On a
167.Tn VAX
168and
169.Tn Tahoe ,
170.Va errno
171is set to
172.Er EDOM
173and the reserved operand is returned
174by log unless
175.Fa x
176> 0, by
177.Fn log1p
178unless
179.Fa x
180> \-1.
181.Sh ERRORS (due to Roundoff etc.)
182exp(x), log(x), expm1(x) and log1p(x) are accurate to within
183an
184.Em ulp ,
185and log10(x) to within about 2
186.Em ulps ;
187an
188.Em ulp
189is one
190.Em Unit
191in the
192.Em Last
193.Em Place .
194The error in
195.Fn pow x y
196is below about 2
197.Em ulps
198when its
199magnitude is moderate, but increases as
200.Fn pow x y
201approaches
202the over/underflow thresholds until almost as many bits could be
203lost as are occupied by the floating\-point format's exponent
204field; that is 8 bits for
205.Tn "VAX D"
206and 11 bits for IEEE 754 Double.
207No such drastic loss has been exposed by testing; the worst
208errors observed have been below 20
209.Em ulps
210for
211.Tn "VAX D" ,
212300
213.Em ulps
214for
215.Tn IEEE
216754 Double.
217Moderate values of
218.Fn pow
219are accurate enough that
220.Fn pow integer integer
221is exact until it is bigger than 2**56 on a
222.Tn VAX ,
2232**53 for
224.Tn IEEE
225754.
226.Sh NOTES
227The functions exp(x)\-1 and log(1+x) are called
228expm1 and logp1 in
229.Tn BASIC
230on the Hewlett\-Packard
231.Tn HP Ns \-71B
232and
233.Tn APPLE
234Macintosh,
235.Tn EXP1
236and
237.Tn LN1
238in Pascal, exp1 and log1 in C
239on
240.Tn APPLE
241Macintoshes, where they have been provided to make
242sure financial calculations of ((1+x)**n\-1)/x, namely
243expm1(n\(**log1p(x))/x, will be accurate when x is tiny.
244They also provide accurate inverse hyperbolic functions.
245.Pp
246The function
247.Fn pow x 0
248returns x**0 = 1 for all x including x = 0,
249.if n \
250Infinity
251.if t \
252\(if
253(not found on a
254.Tn VAX ) ,
255and
256.Em NaN
257(the reserved
258operand on a
259.Tn VAX ) .
260Previous implementations of pow may
261have defined x**0 to be undefined in some or all of these cases.
262Here are reasons for returning x**0 = 1 always:
263.Bl -enum -width indent
264.It
265Any program that already tests whether x is zero (or
266infinite or \*(Na) before computing x**0 cannot care
267whether 0**0 = 1 or not.
268Any program that depends upon 0**0 to be invalid is dubious anyway since that
269expression's meaning and, if invalid, its consequences
270vary from one computer system to another.
271.It
272Some Algebra texts (e.g., Sigler's) define x**0 = 1 for
273all x, including x = 0.
274This is compatible with the convention that accepts a[0]
275as the value of polynomial
276.Bd -literal -offset indent
277p(x) = a[0]\(**x**0 + a[1]\(**x**1 + a[2]\(**x**2 +...+ a[n]\(**x**n
278.Ed
279.Pp
280at x = 0 rather than reject a[0]\(**0**0 as invalid.
281.It
282Analysts will accept 0**0 = 1 despite that x**y can
283approach anything or nothing as x and y approach 0
284independently.
285The reason for setting 0**0 = 1 anyway is this:
286.Bd -filled -offset indent
287If x(z) and y(z) are
288.Em any
289functions analytic (expandable
290in power series) in z around z = 0, and if there
291x(0) = y(0) = 0, then x(z)**y(z) \(-> 1 as z \(-> 0.
292.Ed
293.It
294If 0**0 = 1, then
295.if n \
296infinity**0 = 1/0**0 = 1 too; and
297.if t \
298\(if**0 = 1/0**0 = 1 too; and
299then \*(Na**0 = 1 too because x**0 = 1 for all finite
300and infinite x, i.e., independently of x.
301.El
302.Sh SEE ALSO
303.Xr infnan 3 ,
304.Xr math 3
305.Sh HISTORY
306A
307.Fn exp ,
308.Fn log
309and
310.Fn pow
311functions
312appeared in
313.At v6 .
314A
315.Fn log10
316function
317appeared in
318.At v7 .
319The
320.Fn log1p
321and
322.Fn expm1
323functions appeared in
324.Bx 4.3 .
325