Documente Academic
Documente Profesional
Documente Cultură
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
Mathematical Association of America is collaborating with JSTOR to digitize, preserve and extend access to
The American Mathematical Monthly.
http://www.jstor.org
Early Transcendentals
StevenH. Weintraub
Several current calculus texts have '4earlytranscendental"versions, in which the
exponential and logarithm functions are introduced early in the text. These
functions are usuallyjustified by various 4'hand-waving"
arguments.The point of
this article is to show how they may be introducedrigorously.
We proceed in two steps. We begin with the basic existence theorem (BET):
There is a differentiablefunction y = f(x),defined for x 2 0, satisfying
f(o)= 1, f'(X)=f(X)
(8)
derive that there is a unique function f(x),
f(O) = 1,
f'(x) =f(x)
(8)
for allx 2 O.
For the remainderof this section we assumethe BET. We proveit in the next
section, as Corollary2.16.
Note that the BET does not tell us that there is such a functionf(x), muchless
a uniquesuch function, defined for all real x. Our first objectiveis to derive this
from the BET. Actually,we prove a slightlymore general result in Theorem 1.4.
Lemma1.2. If f(x) is a functionsatisfying
theBET, thenf(x) > Ofor allx 2 O.
1997]
EARLY TRANSCENDENTALS
623
x < O.
d
d
(1/ft -x)) = ( -f '( -x)/f(
dgC
(e(x)) = d;,C
= f '( -x) /f ( _X)2
* (
-x)2)
f ( -x) /f ( _X)2
-1)
1/f ( -x) = ef x) .
For the same reasons, the derivativefrom the right at x= O exists and has the
value 1, and the derivativefrom the left at x = O exists and has value 1, so e'(O)
exists and has value 1. Thus e'(x) existsfor all x, e'(O)= 1, and e'(x) = e(x) for all
R
x, so e(x) satisfies(* ) for all x E IFR.
g
crandb. Thereexistsa uniquedifferenttablefi>nction
Theorem1.4. Ftxrealnumbers
satisfying
g(O) = b,
forallx
(t *)
g'(x) = seg(x)
R.
EARLY TRANSCENDENTALS
[August-September
d) e)
d)
It limx
For
follows
any
+0O
n,
immediately
exp(x)
limx=+OOexp(x)/xn
+ oo,
from
limx
c) _0O
that
= oo.
exp(x)
limx +OOexp(x)
= O.
= +oo. By the uniqueness
Now that we have uniquefunctions with our desired properties,we can give
them names.
Defilnition1.5. For a fixed real numberog,let expX(x)be the unique differentiable
functionsatisfying
exp (O) = 1,
[FR.
We abbreviateexp1(x)by exp(x).
Corollary1.6. Foranyfiuced
realnumber(x,expX(x)= exp(ogx).
[FR.
Proof:
expX(x + w)
exp(w)
Corollary1.9. Foranyinteger
n andallx E , expX(nx)= exp(x)n.
1997]
EARLY TRANSCENDENTALS
625
exp:
Defilnition
exponential
[Ft (0, 1.12.
function
oo). Theexp:
natural
1R(0,
logarithm
oo) hasfunction
a continuous
ln: (0,co)
inverse.
1Ris the inverse of
Proof:
Inductionand Theorem1.8.
ex = exptx)
( 1.11)
the usual laws of expo-
( 1.15)
ax = expln(d2)(
x) = exp( xln( a) ) .
( 1.16)
dX( aX)=
(expln(a)(
x) ) = ln( a)expln(a)(x) = aXlnta) .
amAn,
( 1.17)
EARLY TRANSCENDENTALS
[August-September
anX,
R.
n =O
If, for some value of x, this series convergesto some value S, we say that the sum
of the series is S - S(x). In this way we define a function,and we write
(2.1)
anx .
S( x) - ,
n =O
an
2 0 for all n 2 O.
E(X)=1+X+
X3
X4
2! + 3! + 4! +nn
>
limnOOan+1/an
Proof: Let x > Obe arbitrary.Choose N so that an > Oand an+l/an < 1/(x + 1)
O.Let r = x/(x + 1) < 1. For any
=
for n 2 N; this is possible as limn<>an+l/an
n 2 N we have
a
Xn =
eX
n )(x
n-l
) *X (x
=
{aNx
r
r
so S(x) is eventuallygeometricallydominated.
<
1997]
aNx
N+l )aNXN
)r,
EARLY TRANSCENDENTALS
627
= anXn
n=O+ anx
.
n=O
-
E .
anXO
n=O < 8/2 + 8/2 = g.
an =
be a peacefulseries.Thenthisseriesconverges
for
Emn=O
anXn
oo
oo
N- 1
anx
anX
n=O
n=O
oo
N-1
x,
anX
n=N
anxn
Crn.
n=N
n=O
Considerthis second series. Its first summandis a finite summand;denote its sum
by S1. Its second summand is a geometric series with sum CrN/(l
- r); in
particular,this sum is an upper bound of the partialsums. Each term in the first
series is less than or equal to the correspondingterm in the second series, so the
partial sums of the first series are bounded from above by Sl + CrN/(l
- r).
Thus, this series converges.
2
Lemma2.9. Let S(x) bepeaceful. ThenS(x) is an increasingfunction
on [O,oo).
Proof:IfX2>XlS
co
S(x2) - S(xl) = ,
co
anx2 - ,
co
anxl =
an(X2 -X1 ) 2 0,
O<
cc
e anX<
n =N+ 1
Cc
E
n =N+ 1
anXl= S(xl) -
E
n =0
anXl<
s/2.
O <
S(x)
S(xo)
co
anX
n=O
oo
<
x,
anx
n =O
cc
n=N+l
628
anxO
n=O
N
-
x,
oo
N
anxo
n =0
anX
n =0
n =N+ 1
n=O
n=O
EARLY TRANSCENDENTALS
N
anx
anXO
n =0
[August-September
2x
3x2
4X3
=1+x+
Lemma2.13. Let S(x)
*n
X3
2! + 3! +n
=E(x).
= En=O
anXn
bepeaceful.ThenS(x) is alsopeaceful.
+ l)an+lXn
< (n + l)Bqn+l/x.
But for n 2 N,
(
) ( n-1
) ( n-2
and, since
n+ 1
n
n
n-1
<
<
*<
<
N+ 1
N '
we have
n+1<
{N+
, N
lAn-N+l
SO
{/N+
bnxn< ttl N
lil-N
\
aN+
NBq/xJq l N
lin
J
1997]
= S(x).
EARLY TRANSCENDENTALS
629
Proof: We have to show that for everyxO2 0 and everye > 0, there exists a 8 > O
such that
S(x)-S(x0)
S(xO)
<
x-xO
{ n=O
, a
X -Xo
1
n=O
anXO
co
co
x,
xn
an(xn
-Xo
xn+l
x,
Xo n=l
an+l
Xo+
-x
n=O
co
(n
co
< E
l)an+lXO
n=O
an+l
Yn
<
n=O
,
n=O
(n
+ 1)an+lxn,
.e.,
(
<
S(x)
S(
<
S(x)S
and so
S(x)-S(x0)
S(xO)
<
S(x)-S(xo)
<
x-xO
(where
x
the
last
inequality
is
true
by
our
choice
of
b).
similar
argument
< xO .
holds
if
0
x2
X3
2! + 3! +X
forx20
EARLY TRANSCENDENTALS
[August-September
X3
L(x) =x + 2 + 3 + *2X.
It is easy to check that L(x) is 1-peaceful.Then L(x) = l + X + X2 + X3 + ***,
and we recognizethis as a geometricseries with sum 1/(1 - x) for 0 < x < 1, i.e.,
L(x) = 1/(1 - x) for 0 < x < 1. Thus we see that L'(x) = L(x) = l/(1 - x) fo1*
0 < x < 1. On the other hand, if f(x) = -ln(1 - x), then, by Proposition1.13 and
the chain rule, f'(x) = 1/(1 - x) for 0 < x < 1. We have two functionswith the
same derivative,so they must differ by a constant (a consequence of the mean
value theorem). But L(0) = 0 = f(0), so we conclude Lf x) = f ( x) = -ln( 1 - x)
for 0 < x < 1.
This gives us a method of computing natural logarithms:For 0 < y < 1, set
x= 1 -y (so y= 1 -x) and then ln(y)= -L(x)= -L(1 -y). We know
ln(1) = 0. For y > 1, set x = (y - 1)/y (so 1 - x = 1/y) and then ln(y) =
-ln(l/y) = L(x) = L((y-1)/y).
Departmentof lkvltathematics
LouisianaState University
BatonRouge,LA 70803-4918
weintr@math.lsu.edu
1997]
EARLY TRANSCENDENTALS
631