Professional Documents
Culture Documents
Later we will need precise language to discuss the notion of one real number being
‘‘close to’’ another. If a is a given real number, then saying that a real number x is ‘‘close
to’’ a should mean that the distance jx aj between them is ‘‘small.’’ A context in which
this idea can be discussed is provided by the terminology of neighborhoods, which we
now define.
2.2.7 Definition Let a 2 R and e > 0. Then the e-neighborhood of a is the set
V e ðaÞ :¼ fx 2 R : jx aj < eg.
2.2.8 Theorem Let a 2 R. If x belongs to the neighborhood V e ðaÞ for every e > 0, then
x ¼ a.
Proof. If a particular x satisfies jx aj < e for every e > 0, then it follows from 2.1.9 that
jx aj ¼ 0, and hence x ¼ a. Q.E.D.
2.2.9 Examples (a) Let U :¼ fx : 0 < x < 1g. If a 2 U, then let e be the smaller of the
two numbers a and 1 a. Then it is an exercise to show that V e ðaÞ is contained in U. Thus
each element of U has some e-neighborhood of it contained in U.
(b) If I :¼ fx : 0 x 1g, then for any e > 0, the e-neighborhood V e ð0Þ of 0 contains
points not in I, and so V e ð0Þ is not contained in I. For example, the number xe :¼ e=2 is in
V e ð0Þ but not in I.
(c) If jx aj < e and jy bj < e, then the Triangle Inequality implies that
1. If a; b 2 R pand
ffiffiffiffiffi b 6¼ 0, show that:
(a) jaj ¼ a2 ; (b) ja=bj ¼ jaj=jbj:
2. If a; b 2 R, show that ja þ bj ¼ jaj þ jbj if and only if ab 0.
3. If x; y; z 2 R and x z, show that x y z if and only if jx yj þ jy zj ¼ jx zj. Interpret
this geometrically.
C05 12/08/2010 14:19:45 Page 135
bounded. In the next theorem, however, we show that continuous functions on a certain
type of interval are necessarily bounded.
5.3.2 Boundedness Theoremy Let I :¼ ½a; b be a closed bounded interval and let
f : I ! R be continuous on I. Then f is bounded on I.
Proof. Suppose that f is not bounded on I. Then, for any n 2 N there is a number xn 2 I
such that j f ðxn Þj > n. Since I is bounded, the sequence X :¼ ðxn Þ is bounded. Therefore,
the Bolzano-Weierstrass Theorem 3.4.8 implies that there is a subsequence X 0 ¼ ðxnr Þ of X
that converges to a number x. Since I is closed and the elements of X 0 belong to I, it follows
from Theorem 3.2.6 that x 2 I. Then f is continuous at x, so that ð f ðxnr ÞÞ converges to f ðxÞ.
We then conclude from Theorem 3.2.2 that the convergent sequence ðf ðxnr ÞÞ must be
bounded. But this is a contradiction since
j f ðxnr Þj > nr r for r 2 N:
Therefore the supposition that the continuous function f is not bounded on the closed
bounded interval I leads to a contradiction. Q.E.D.
5.3.3 Definition Let A R and let f : A ! R. We say that f has an absolute maximum
on A if there is a point x 2 A such that
f ðx Þ f ðxÞ for all x 2 A:
We say that f has an absolute minimum on A if there is a point x 2 A such that
f ðx Þ f ðxÞ for all x 2 A:
We say that x is an absolute maximum point for f on A, and that x is an absolute
minimum point for f on A, if they exist.
We note that a continuous function on a set A does not necessarily have an absolute
maximum or an absolute minimum on the set. For example, f ðxÞ :¼ 1=x has neither an
absolute maximum nor an absolute minimum on the set A :¼ ð0; 1Þ. (See Figure 5.3.1.)
There can be no absolute maximum for f on A since f is not bounded above on A, and there
is no point at which f attains the value 0 ¼ inff f ðxÞ : x 2 Ag. The same function has
y
This theorem, as well as 5.3.4, is true for an arbitrary closed bounded set. For these developments, see Sections
11.2 and 11.3.
C05 12/08/2010 14:19:45 Page 136
neither an absolute maximum nor an absolute minimum when it is restricted to the set
(0, 1), while it has both an absolute maximum and an absolute minimum when it is
restricted to the set [1, 2]. In addition, f ðxÞ ¼ 1=x has an absolute maximum but no
absolute minimum when restricted to the set [1, 1), but no absolute maximum and no
absolute minimum when restricted to the set (1, 1).
It is readily seen that if a function has an absolute maximum point, then this point is
not necessarily uniquely determined. For example, the function gðxÞ :¼ x2 defined for
x 2 A :¼ ½1; þ1 has the two points x ¼
1 giving the absolute maximum on A, and the
single point x ¼ 0 yielding its absolute minimum on A. (See Figure 5.3.2.) To pick an
extreme example, the constant function hðxÞ :¼ 1 for x 2 R is such that every point of R is
both an absolute maximum and an absolute minimum point for h.
Proof. Consider the nonempty set f ðIÞ :¼ ff ðxÞ : x 2 Ig of values of f on I. In Theorem 5.3.2
it was established that f (I ) is a bounded subset of R. Let s :¼ sup f ðIÞ and s :¼ inf f ðI Þ.
We claim that there exist points x and x in I such that s ¼ f ðx Þ and s ¼ f ðx Þ. We will
establish the existence of the point x , leaving the proof of the existence of x to the reader.
Since s ¼ sup f ðIÞ, if n 2 N, then the number s 1=n is not an upper bound of the
set f (I). Consequently there exists a number xn 2 I such that
1
ð1Þ s < f ðxn Þ s for all n 2 N:
n
Since I is bounded, the sequence X :¼ ðxn Þ is bounded. Therefore, by the Bolzano-
Weierstrass Theorem 3.4.8, there is a subsequence X 0 ¼ ðxnr Þ of X that converges to some
number x . Since the elements of X0 belong to I ¼ ½a; b, it follows from Theorem 3.2.6 that
x 2 I. Therefore f is continuous at x so that limð f ðxnr ÞÞ ¼ f ðx Þ. Since it follows from
(1) that
1
s < f ð xn r Þ s for all r 2 N;
nr
we conclude from the Squeeze Theorem 3.2.7 that limð f ðxnr ÞÞ ¼ s . Therefore we have
f ðx Þ ¼ limð f ðxnr ÞÞ ¼ s ¼ sup f ðIÞ:
We conclude that x is an absolute maximum point of f on I. Q.E.D.
C05 12/08/2010 14:19:48 Page 137
The next result is the theoretical basis for locating roots of a continuous function by
means of sign changes of the function. The proof also provides an algorithm, known as the
Bisection Method, for the calculation of roots to a specified degree of accuracy and can be
readily programmed for a computer. It is a standard tool for finding solutions of equations
of the form f (x) ¼ 0, where f is a continuous function. An alternative proof of the theorem
is indicated in Exercise 5.3.11.
Proof. We assume that f ðaÞ < 0 < f ðbÞ. We will generate a sequence of intervals by
successive bisections. Let I 1 :¼ ½a1 ; b1 , where a1 :¼ a; b1 :¼ b, and let p1 be the midpoint
p1 :¼ 12 ða1 þ b1 Þ. If f ðp1 Þ ¼ 0, we take c :¼ p1 and we are done. If f ðp1 Þ 6¼ 0, then either
f ðp1 Þ > 0 or f ðp1 Þ < 0. If f ðp1 Þ > 0, then we set a2 :¼ a1 ; b2 :¼ p1 , while if f ðp1 Þ < 0,
then we set a2 :¼ p1 ; b2 :¼ b1 . In either case, we let I 2 :¼ ½a2 ; b2 ; then we have I 2 I 1 and
f ða2 Þ < 0; f ðb2 Þ > 0.
We continue the bisection process. Suppose that the intervals I 1 ; I 2 ; . . . ; I k have
been obtained by successive bisection in the same manner. Then we have f ðak Þ < 0 and
f ðbk Þ > 0, and we set pk :¼ 12 ðak þ bk Þ. If f ðpk Þ ¼ 0, we take c :¼ pk and we are done.
If f ðpk Þ > 0, we set akþ1 :¼ ak ; bkþ1 :¼ pk , while if f ðpk Þ < 0, we set akþ1 :¼
pk ; bkþ1 :¼ bk . In either case, we let I kþ1 :¼ ½akþ1 ; bkþ1 ; then I kþ1 I k and
f ðakþ1 Þ < 0; f ðbkþ1 Þ > 0.
If the process terminates by locating a point pn such that f ðpn Þ ¼ 0, then we are done.
If the process does not terminate, then we obtain a nested sequence of closed bounded
intervals I n :¼ ½an ; bn such that for every n 2 N we have
f ðan Þ < 0 and f ðbn Þ > 0:
Furthermore, since the intervals are obtained by repeated bisection, the length of In is
equal to bn an ¼ ðb aÞ=2n1 . It follows from the Nested Intervals Property 2.5.2
that there exists a point c that belongs to In for all n 2 N. Since an c bn for all
n 2 N and limðbn an Þ ¼ 0, it follows that limðan Þ ¼ c ¼ limðbn Þ. Since f is continuous
at c, we have
lim ð f ðan ÞÞ ¼ f ðcÞ ¼ lim ð f ðbn ÞÞ:
The fact that f ðan Þ < 0 for all n 2 N implies that f ðcÞ ¼ lim ð f ðan ÞÞ 0. Also, the fact
that f ðbn Þ > 0 for all n 2 N implies that f ðcÞ ¼ lim ð f ðbn ÞÞ 0. Thus, we conclude that
f ðcÞ ¼ 0. Consequently, c is a root of f. Q.E.D.
The following example illustrates how the Bisection Method for finding roots is
applied in a systematic fashion.
5.3.6 Example The equation f ðxÞ ¼ xex 2 ¼ 0 has a root c in the interval [0, 1],
because f is continuous on this interval and f ð0Þ ¼ 2 < 0 and f ð1Þ ¼ e 2 > 0. Using a
calculator we construct the following table, where the sign of f ( pn) determines the interval
at the next step. The far right column is an upper bound on the error when pn is used to
approximate the root c, because we have
Bolzano’s Theorem
The next result is a generalization of the Location of Roots Theorem. It assures us that a
continuous function on an interval takes on (at least once) any number that lies between
two of its values.
Proof. Suppose that a < b and let gðxÞ :¼ f ðxÞ k; then gðaÞ < 0 < gðbÞ. By the Location
of Roots Theorem 5.3.5 there exists a point c with a < c < b such that 0 ¼ gðcÞ ¼ f ðcÞ k.
Therefore f ðcÞ ¼ k.
If b < a, let hðxÞ :¼ k f ðxÞ so that hðbÞ < 0 < hðaÞ. Therefore there exists a point c
with b < c < a such that 0 ¼ hðcÞ ¼ k f ðcÞ, whence f ðcÞ ¼ k. Q.E.D.
Proof. It follows from the Maximum-Minimum Theorem 5.3.4 that there are points c
and c in I such that
inf f ðIÞ ¼ f ðc Þ k f ðc Þ ¼ sup f ðIÞ:
The next theorem summarizes the main results of this section. It states that the image
of a closed bounded interval under a continuous function is also a closed bounded interval.
The endpoints of the image interval are the absolute minimum and absolute maximum
values of the function, and the statement that all values between the absolute minimum
and the absolute maximum values belong to the image is a way of describing Bolzano’s
Intermediate Value Theorem.