Number Systems, Chapter 4 defined what means in an
ordered field (Number Systems: From to , Definition 4.3.2)
and proved the limit laws. It also gave a method for
proving a limit from the definition (Number Systems: From to , Example (Method: proving a limit from the definition)).
The method bounds the distance by for , by enlarging
numerators and shrinking denominators, and then takes
. This section does not teach that method
again. It adds two ways to find the bound when the
expression is less tame: keep the dominant term of a
quotient, and bound a power from below by one term of
the binomial theorem. Analysis on the Real Line
uses all three.
Throughout, sequences are sequences in , a complete ordered
field (Number Systems: From to , Theorem 5.5.2),
and stands for
, as in
Number Systems. Where a formula such as needs , the sequence
starts at .
Such a sequence is the shift by one of a sequence
indexed from , and
a shift changes no limit (Number Systems: From to , Proposition 4.3.2
(b)).
The method,
recalled
The method of Number Systems: From to , Example (Method: proving a limit from the definition)
has four steps.
- (1) Bring the
distance to one fraction.
- (2) Bound it, for
every , by
a constant
over .
- (3) Choose a
threshold
for that bound, and let .
- (4) Write the proof
forwards: first, then
, then an
arbitrary .
In two
remarks simplify the work. First, step (3) always
succeeds. is
complete, hence Archimedean (Number Systems: From to , Theorem 6.1.2).
So Number Systems: From to , Proposition 4.2.3
applies to it.
Second, a bound for serves as well as
. Since
is Archimedean,
there is an integer . Then , because reciprocals reverse the
order of positive numbers (Number Systems: From to , Proposition 3.2.2
(b)). So for every , because a
larger denominator gives a smaller quotient (Number Systems: From to , Proposition 4.1.2
(c)).
Example
(A first estimate in )
Show that in .
Solution. The numbers (1)–(4) are
the four steps of the method.
For ,
and , so the distance is itself.
Let .
Then . Also , so . A larger
denominator gives a smaller quotient, so .
Multiplying these two inequalities between non-negative
numbers, So and .
Let . Choose an integer
and let . It depends on
only.
Let .
Then , so
the bound of (2) applies, and since : where the last step uses . Hence .
Both bounds in step (2) needed a range. The bound
fails
at . The bound
holds from
on. Here the ranges were found at sight. When the terms
do not simplify so readily, two techniques find them:
keeping the dominant term, and bounding a power from
below.
Keeping the
dominant term
Take . For large the terms and swamp the others, and
the quotient should be close to . To bound a quotient
from above we need its numerator from above and its
denominator from below. Enlarging a numerator is easy.
Shrinking a denominator that contains a subtraction is
where care is needed, because is negative for
small .
Example
(Method: keep the dominant term)
How it goes. In a quotient of sums,
keep the term that grows fastest and absorb the others
into it. From some on, a lower-order
term is at most a fixed fraction of the dominant term.
For example, once , so there. Then the quotient is bounded
by a quotient of single powers. This is the bounding
step of the method of Number Systems: From to , Example (Method: proving a limit from the definition),
with the
largest of the ranges used.
When to use it. Rational expressions
in , possibly with
.
Common mistakes. Subtracting in the
denominator without stating a range. For instance,
is false for .
Example
(A limit that is not zero)
Show that .
Solution.
No natural number has square , so every is defined. For , , and
so the distance is .
Keep the dominant terms, on top and below. Let . Then , so
. And
, so . Inverting the second, which reverses it, and
multiplying by the first,
Let . Choose an integer and let . It depends on
only.
For :
. Hence .
The candidate limit is the ratio of the
leading coefficients. Reading off the dominant terms is
how the limit is guessed, and the method is how the
guess is proved. The range came from two requirements, and , and
the larger of their thresholds serves both.
Quick check. For which is ?
Solution. The inequality says , that
is, . For it holds with equality. For the factor is positive, so it
holds exactly when , that is, . So the inequality holds for and for every , and fails for
. A
proof that uses takes .
Powers
Every bound so far came from dropping or absorbing
terms of a sum. A power has only one term,
and there is nothing to drop. Bernoulli’s inequality and
the binomial theorem turn it into a sum, and a sum of
non-negative terms is at least any one of its terms.
The first case is the geometric sequence. For the terms
do not shrink,
and for they should tend to . Take first. Then
, and Bernoulli’s inequality gives . So for every , a bound of the
form with . The proposition
below does the same for every with , by
writing
as .
Proposition
1.4.1 (Geometric Sequences Tend to Zero)
Let
with .
- (a) If and , then and for every integer .
- (b) .
Idea (Scratch work). The goal, from
Number Systems: From to , Definition 4.3.2,
is for every beyond some . Solving for would need a logarithm,
which we do not have yet. So we follow the bounding step
of the method and look for a simpler bound of the form
.
The trouble is that is a small
number. The tools we have, Bernoulli’s inequality and
the binomial theorem, bound powers of numbers
above from
below. So we turn small into large: we
write as
the reciprocal of a number above , so that . Now is small as
soon as
is large, and a lower bound is
enough. Bernoulli’s inequality supplies one at once:
.
|
below
every for large |
| , so |
enough: bounded below by |
| Sets and Functions, Theorem 3.3.7:
|
: the bounding step of the method
with ; the gap is closed |
The proof follows these rows from top to bottom. It
then finishes with the choice of the threshold and the
forward proof, using the bound .
Proof.
(a) Let . Then ,
since the absolute value of a non-zero number is
positive. Reciprocals reverse the order of positive
numbers, so , that is, . By the choice of , , so
and . Since , Bernoulli’s inequality (Sets and Functions, Theorem 3.3.7)
gives For , as a product of positive numbers. So ,
and taking reciprocals reverses the inequality: .
(b) If , then
for every ,
so for every and
every . So
works
for every . Let and , and let be as in (a). Then
. Since is Archimedean, Number Systems: From to , Proposition 4.2.3
gives with for
every . This
depends on and on , through . Let . Then , and by
induction on ,
since . By (a), This proves .
Part (a) is the useful half. It gives an explicit
bound, valid for every , from which a threshold can be read off for
any tolerance.
Solution. Here ,
so and for . Now exactly when . So works: for , . The bound is wasteful, since already
. That costs nothing: the definition asks
for some threshold, not the least.
The same steps handle a power in a denominator under
a numerator that is a power of : write the base as
, keep one
term of the expansion, and invert.
Example
(Method: bound a power from below)
How it goes. To make a quotient
small when its denominator contains with :
When to use it. Quotients . Powers with , through
.
The -th roots of
Section 1.5.
Common mistakes. One mistake is to
choose too small:
then the bound does not tend to . Another is to use the
binomial bound with .
In Example (A lower bound for powers of two)
the base was , and
gave for
. The same
steps work for every base . The numerator is , to the first power, so
.
Proposition
1.4.2 (Powers Beat n)
Let and
.
Then for every integer , and .
Proof. Let . Since , . By Corollary 1.2.4
(b) with and
, and
since , Since , , so . Multiplying the inequalities
and , between non-negative numbers, gives . Hence . Inverting, which reverses the inequality,
and multiplying by , For the limit, , so the
distance to is
for
, with . Let
,
choose an integer , and let . It depends on and . For , . So .
With , that
is, Bernoulli’s inequality, the same steps give only
. That bound does not tend to . The choice of is the whole
difference.
A comparison with a power can also run the other way:
a sequence that is hard to handle directly is bounded by
a geometric one, and Proposition 1.4.1
finishes.
Example
(Factorials beat powers of two)
Show that for
every integer , and hence that .
Solution. The first
inequality. We show for , by induction on . For : . Suppose
.
Also ,
and both inequalities are between non-negative numbers,
so we may multiply them: Here , as a power of a positive number, and . So
inverting reverses the inequality, and .
The second inequality. By Proposition 1.4.1
(a) with , , we have
for . Since
,
The limit. So for : a bound with . Given , take
an integer . It depends on only. Then
, and for
, . Hence .
Warning. is not . The definition of a limit asks for
the strict inequality. A threshold chosen to make the
bound at most can miss it.
For , the choice gives, at , the distance
,
which is not . Ask for , strictly.
↺Second readingWhen
the one-term bound fails
Change one hypothesis of Corollary 1.2.4
(b): keep and
allow . At
, the claim would
read , that is, , which is false.
What breaks is the step that keeps one term of a sum.
That step needs every term non-negative, and with the terms
alternate in sign. Bernoulli’s inequality survives for
,
because its proof is an induction, not a dropped sum. So
for the method first passes to with
, as the
proof of Proposition 1.4.1
did. It never expands directly.
Exercises
A. Check your
understanding
Exercise
(A1)
- (a) Write the
negation of “” with all its quantifiers.
- (b) In Example (A first estimate in ),
on what does
depend? Could it depend on ?
- (c) Name the method
used: for .
- (d) True or false:
if for every , then .
B. Practice
Exercise
(B1: Say which method, then complete it)
Show that . Start: write . By Corollary 1.2.4
(b) with ,
for . Say which
method this is, then complete it.
Exercise
(B2: Put the proof in order)
The claim is . Put the
lines below in an order that proves it, and say which
line does not belong.
- (i) Let .
- (ii) For , .
- (iii) Choose with and
.
- (iv) Choose .
- (v) Hence .
- (vi) for every .
Solution. An admissible order is
(i), (iii), (vi), (ii), (v). Line (vi) uses nothing
chosen, so it may also come first: (vi), (i), (iii),
(ii), (v). Line (ii) must come after (iii), since it
uses . It must
also come after (vi), since it uses the formula for the
distance. In (ii), holds because a larger denominator gives a
smaller quotient, and follows from .
The line that does not belong is (iv): it makes depend on , but must be fixed before
is named, and
ranges over all
numbers .
Exercise
(B3: Prove or disprove)
Prove each true statement and refute each false one.
In parts (a), (b), (c) and (e), denotes the -th term of the sequence
in question.
- (a) .
- (b) .
- (c) .
- (d) For every , every
integer has .
- (e) .
Solution.
(a)
True. For , , so . Given , take
an integer . Then , and for , .
(b)
False. For , , so . Take . For every , the index
has . This is the negation
of A1 (a).
(c)
True. , since and . Given
, an
integer works, as in (a).
(d)
False. Take and .
Then ,
which is not . The non-strict
does not give the strict .
(e)
True. For , . Also , by induction on from . So Every factor with lies in
, so their
product lies in . The last factor, , is kept. Given
,
take an integer and . For , .
C. Going deeper
Exercise
(C1: Chapter check: a rational limit)
Prove from the definition that .
Solution. Let . The
goal is an with
for every .
The distance. For every , so .
The bound. Let . Then , and ,
so . Multiplying, .
The threshold. Choose an integer .
Then , and
depends on only. For
, , so where the middle step holds because and the last
because . Hence .
What a complete answer needs. (10
points)
- 2: fixes arbitrary and states the goal.
- 3: computes exactly.
- 3: bounds it by (or ) on a stated
range ,
with each inequality justified.
- 2: chooses
before , depending
only on , and ends the
chain with a strict .
Slips that lose the step. Letting
depend on . Bounding in the wrong
direction, that is, shrinking a numerator or enlarging a
denominator.
Exercise
(C2: Squares over powers)
Let and
.
Show that . Start: by Corollary 1.2.4
(b) with ,
for . Say which method this is, then
complete it.
Solution (Hint 1). Bound and below by multiples
of , for .
Solution (Hint 2). For , ,
since this rearranges to . Find a
similar bound for , then multiply.
Solution. The method:
bound the power from below, as in Example (Method: bound a power from below).
The start wrote and kept one term, with , one more than the
power of on top.
What remains is to bound below and finish the limit.
The lower bound. Let . Then ,
since this rearranges to . Also
, since this rearranges to . Both
sides of both inequalities are non-negative. So we may
multiply them, with as the third factor, and get . Hence
The limit. Inverting, which reverses the
inequality, and multiplying by , we get for . Let , take
an integer and . It depends on and . For , Hence .
What a complete answer needs. (6
points)
- 1: names the method (bounding a power from
below).
- 2: linear lower bounds for and on a stated,
justified range .
- 1: their product, with non-negativity checked, and
the inversion, with positivity stated.
- 2: the choice of , and the strict chain
ending in .
Slips that lose the step. Using
or without the range . Multiplying
inequalities without checking that the factors are
non-negative.