Every convergence proof so far began with the limit
in hand: to show we estimated , and the
limit laws of Section 4.4 compute a new limit only from
limits already known. That is no help for the sequence
of Chapter 2 that closes in on the gap at (Example (A Rational Sequence Aimed at the Gap)),
whose terms visibly settle down while no
rational is available to be their limit. This section
names a property that can be checked from the terms
alone: eventually they are all close to each
other. We prove the facts about it that Chapter
5 needs: convergent sequences have it, sequences with it
are bounded and closed under sums and products, and,
when such a sequence does not tend to , its terms eventually
keep one sign and stay away from .
Throughout, is
an ordered field (Definition 2.3.2),
sequences are indexed by (Definition 4.3.1),
and
ranges over . We use the
absolute-value rules of Proposition 4.1.3
and the halving rules of Proposition 4.1.2
without restating them, and write for in .
Terms that
agree with one another
A limit relates a sequence to a point of ; boundedness (Definition 4.3.4)
is a property of the sequence alone. We want a second
property of the sequence alone, saying “the terms settle
down” without saying where.
In the
Newton terms satisfy , and the later terms are closer
still to one another (Section 4.6 proves it). The way to
say “settle down” without a target is to compare terms
with terms: however small a tolerance is named, from
some index on, any two terms differ by less than it.
The terms eventually agree with one another to
within any tolerance.
Definition
4.5.1 (Cauchy Sequence)
A sequence in
is a Cauchy sequence in (or
Cauchy) if for every
there exists such that for all with and . In symbols:
In words: the first quantifier says the tolerance is
arbitrary and is named first; the second says a
threshold exists, and since it comes after , it may
depend on and on
nothing else; the last two say that
every pair of indices beyond the
threshold, near or far apart, gives terms within of each
other. Compared with Definition 4.3.2,
the point has
been replaced by a second index .
The tolerances are the elements of , so the field
is part of the definition. For , the tolerances
are rational, and “Cauchy in ” is a statement about
rationals only, as Chapter 5 needs. Since (Proposition 4.1.3
(b)), the roles of and are symmetric, and
there is nothing else to check for the definition to
make sense.
Negation. Moving inward quantifier by
quantifier (Logic and Proof, Proposition 6.2.3,
Logic and Proof, Proposition 6.2.2)
and reading as by (O1), “ is not Cauchy”
reads One tolerance is beaten beyond every
threshold.
Examples.
A constant sequence. Let for every . Then for all , so works for every
. This
degenerate case matters: Chapter 5 represents each
rational by its
constant sequence.
The reciprocals in . Let in . Given , Corollary 2.5.4
gives with . Let
. Then
and
, so
and (Proposition 2.5.2).
Two elements
of satisfy
and , so by Proposition 4.1.3
(d). Hence . The threshold
depends on only: for
, works.
Sequences with a limit. By the first result
of the next subsection, the convergent sequences of
Section 4.3 (Example (First Limits))
are all examples. The definition earns its keep on
sequences with no limit available, which Section 4.6
supplies.
Non-example by a minimal change.
Change the constant sequence at every odd index: let
for even
and for odd . The sequence is still
bounded, by , and
every two terms still differ by at most . What fails is “for
every ”:
take . Whatever is offered, the indices
and are both , one even and one
odd, and their terms differ by . So the
negation above holds with , , .
Why this definition. Each clause is
there for a reason.
- Every pair, not only neighbors. Asking only
that for is too weak: the
example at the end of this section has consecutive
differences tending to and is not Cauchy.
- The threshold after the tolerance. With the
quantifiers swapped, “”, the condition forces for every , hence for all (the argument
of Theorem 4.3.4):
only sequences that are eventually constant would
qualify.
- Tolerances from the field. Changing the set
of tolerances changes the notion: in the non-Archimedean
field of
Section 4.2 there are more tolerances than in , among them , and Exercise C2 of
Section 4.6 shows a sequence of rationals that passes
every rational test and fails the test .
Warning. The definition asks for
one that serves
every pair . A proof that
checks only some pairs, say each against , proves nothing: in
the non-example above, for every
, so those pairs
pass every test, and the sequence still fails.
Payoff. The tail-change principle
extends at once. If for every and is Cauchy, then for
and its threshold for , every
give . So is Cauchy, with
thresholds , as in Proposition 4.3.2.
Convergent
sequences are Cauchy, and Cauchy sequences are
bounded
The first result says the new property is no stronger
than convergence: it is a necessary condition, one we
can test without knowing the limit.
Proposition
4.5.1 (Convergent Sequences Are Cauchy)
If in
, then is a Cauchy
sequence in .
Idea. Two terms that are each
within of are within of each
other: insert the point between them with the
triangle inequality, and spend half the tolerance on
each side.
Proof. Let . Then by Proposition 4.1.2
(b), so, since , there is with for every ; depends only on . Let . By the
three-point triangle inequality (Proposition 4.1.3
(e)) with in the
middle, and by (Proposition 4.1.3
(b)), where the strict inequality adds two strict
inequalities (Proposition 2.3.5
(b)) and the equality is Proposition 4.1.2
(b). This proves that is Cauchy.
The converse is the question of the chapter. In it fails (Section
4.6); Chapter 5 builds a field in which it holds.
Quick check. Is the sequence Cauchy in ? Answer without
estimating .
The second result is the Cauchy version of Proposition 4.3.5.
There the tail was trapped near the limit ; here no limit is
available, so we trap the tail near one of its own
terms.
Proposition
4.5.2 (Cauchy Sequences Are Bounded)
Every Cauchy sequence in is bounded.
Idea. Apply the definition once,
with the tolerance . It gives an index
beyond which
every term is within of the term , a base point we can
name. The finitely many terms before are handled by a finite
maximum, as for convergent sequences.
Proof. Let be Cauchy in . Since (Proposition 2.3.5
(e)), there is with for all . For , taking and using the
triangle inequality (Proposition 4.1.3
(e), (b)), Let a finite maximum in the totally ordered set
(Sets, Relations, and Cardinality, Proposition 2.3.8,
whose proof works in any totally ordered set; for the set is and
).
Then , so . For , because is in the set;
for , . So for every , and is bounded (Definition 4.3.4).
The bound
depends on the whole sequence, not on any tolerance.
That is what makes it usable in the next proof: it can
be fixed before is named.
Sums and
products of Cauchy sequences
Chapter 5 will add and multiply real numbers by
adding and multiplying Cauchy sequences of rationals
term by term. For that, sums and products of Cauchy
sequences must again be Cauchy. The limit laws (Theorem 4.4.3)
do not apply, since they need limits; the proofs below
copy their estimates with a second index in place of the
limit.
Theorem
4.5.3 (Sums and Products of Cauchy
Sequences)
Let and
be Cauchy
sequences in , and
let .
- (a) Every constant
sequence and every null sequence in is Cauchy.
- (b) The sequences
, , and are
Cauchy.
- (c) The sequence
is
Cauchy.
Idea.
- (a) is Proposition 4.5.1.
For (b), split : give each of
and the tolerance and take
the larger threshold. For (c), the difference splits
as . The factors and do not get small, but
they are bounded (Proposition 4.5.2).
So bound, then shrink: replace each
factor we cannot control by a bound fixed in advance,
then make the other factor small enough to absorb the
bound.
Proof.
(a) A constant
sequence
converges to , and
a null sequence converges to (Example (First Limits),
Definition 4.3.3).
Both are Cauchy by Proposition 4.5.1.
(b)
Sums. Let , so (Proposition 4.1.2
(b)). Since
is Cauchy, there is with for all ; since is Cauchy, there is
with for all . Let , which depends only on . For , we have (Proposition 2.3.2),
so the triangle inequality (Proposition 4.1.3
(e)) gives where the strict step adds two strict
inequalities (Proposition 2.3.5
(b)) and the equality is Proposition 4.1.2
(b).
Multiples. If , then is the constant sequence (Proposition 2.3.2
(b)), Cauchy by (a). If , then (Proposition 4.1.3
(a)), so (Proposition 2.5.2,
(O4)); choose
(depending on and ) with for all . For such
, by distributivity (F9) and Proposition 2.3.2
(c), then Proposition 4.1.3
(b) and Proposition 2.3.5
(c); the last equality is (F5)–(F8).
Negatives and differences. Since (Proposition 2.3.2
(c)), is the
multiple with . Then is the sum of two Cauchy
sequences.
- (c) By Proposition 4.5.2,
fix with and for every
; they
depend on the two sequences only. Now let . Then and
lie in
(Proposition 4.1.2
(b), (O4), Proposition 2.5.2
(a)), so there are with for all , and with for all . Let , depending on , , , and let . Expanding the
right side shows By the triangle inequality and
(Proposition 4.1.3
(e), (b)), then Proposition 4.1.2
(d) applied to and to ,
where the strict step multiplies by and (Proposition 2.3.5
(c)) and adds (Proposition 2.3.5
(b)); the equality uses (Lemma 4.1.1
(a), (F5)–(F8)), likewise for , and Proposition 4.1.2
(b). So
is Cauchy.
The order of choices in (c) is the whole proof: and are chosen before , from the
sequences alone, so that is one
fixed tolerance. Exercise C3 shows what goes wrong when
the bound is skipped.
The sign of
a Cauchy sequence that is not null
A null sequence is the Cauchy analogue of . Chapter 5 must decide,
for each Cauchy sequence of rationals that is not null,
whether it represents a positive or a negative number,
and it needs an inverse for it. Both rest on one fact:
such a sequence eventually keeps one sign, at a definite
distance from .
Theorem
4.5.4 (The Sign of a Non-Null Cauchy
Sequence)
Let be a
Cauchy sequence in that is not null. Then
there are and such that either The two alternatives do not both hold.
Try it first (ten minutes): write “ is not null” in
symbols. Which does it hand
you, and what does the Cauchy condition do with
it?
Idea. “Not null” is the negation of
Definition 4.3.3:
The obvious attempt reads this as “
for all large ”,
and that is not what it says: it promises, beyond each
threshold, one term at least from , and between such terms
others might come near or change sign. The
Cauchy condition is what forbids that wandering. So
fix the bad and
keep it for the rest of the proof. Ask the Cauchy
condition for a threshold with tolerance , and only
then ask the negation for one term , beyond , with . Every later term is within of , so it has the
sign of and
size at least . Before
reading on: where is used?
Proof. Since is not null, the
negation of Definition 4.3.3
holds. Moving
inward by Logic and Proof, Proposition 6.2.3
(a), (b), Logic and Proof, Proposition 6.2.1
(a) and Logic and Proof, Proposition 6.2.2,
and reading as
by (O1), it says: there is
such that for every some has . Fix such an .
By Proposition 4.1.2
(b), . Since is Cauchy, there is
(depending only on the sequence, through ) with Applying the negation with gives
with
and
. By Definition 4.1.1,
either and , or and ,
that is, (Proposition 2.3.5
(a)).
Let .
Since both and
are , , that is, (Proposition 4.1.3
(d)).
Case 1: . Adding to the left
inequality (Proposition 2.3.5
(b)), , where the last equality is Proposition 4.1.2
(b).
Case 2: . Adding to the right
inequality, .
Let and ; both depend only on . In Case 1, for every ; in Case 2, for every
.
Finally, both alternatives cannot hold, even with
different constants and thresholds: if for and for , with ,
then gives , against
(O1). This proves the theorem.
The answer to the prompt: is used once,
to apply the Cauchy estimate to the pair . Had we taken
from the
negation first and afterwards, might lie below , and the term would say nothing
about the tail.
Both hypotheses are needed; the witnesses are in
.
- Not null. The sequence is Cauchy
(the examples after Definition 4.5.1)
and null. For any and any , Corollary 2.5.4
gives an index with , so neither alternative holds.
- Cauchy. The sequence is not null,
since for every . Its terms alternate
between and , so neither
alternative holds for any and : and have opposite
signs.
Sign
stability and reciprocals
Two smaller results complete the toolkit: the sign
found by Theorem 4.5.4
survives a null change of the sequence, and a sequence
that stays away from can be inverted.
Proposition
4.5.5 (Sign Stability under Null
Differences)
Let and
be sequences
in such that
is
null. If , , and for every , then there is with for every
.
Neither sequence needs to be Cauchy here. In Chapter
5, two Cauchy sequences with null difference represent
the same real number, and this proposition is what shows
that “represented by a sequence eventually for some ” does not depend
on the representative chosen.
Proposition
4.5.6 (Reciprocals of Cauchy Sequences)
Let be a
Cauchy sequence in , and let and be such that
for
every .
Define Then is a Cauchy
sequence in , and
for
every .
Idea. For , , so is defined. The
difference of reciprocals is ;
its numerator can be made small, and the denominator is
at least in
absolute value. So , and the
tolerance for gives the tolerance
for
.
Proof. For , ,
so (Proposition 4.1.3
(a)), and is defined, with . For , Lemma 4.1.1
(c) gives By Proposition 4.1.3
(b), . From
and , Proposition 4.1.2
(d) gives , and then Proposition 4.1.2
(c) gives . As (Proposition 2.5.2
(a)), Proposition 4.1.2
(d) with gives Now let . Then by (O4), so there is
(depending on
and
) with for all . Let . For , multiplying by (Proposition 2.5.2
(a), Proposition 2.3.5
(c)), So
is Cauchy. The values for play no part; any values would do, by the
tail-change remark after Definition 4.5.1.
The hypothesis “” cannot be weakened to “”. In , is Cauchy
and never , but
is
not bounded (for each , Theorem 2.5.3
gives an with
), so
it is not Cauchy, by Proposition 4.5.2.
Combined with Theorem 4.5.4,
which supplies
and for every
Cauchy sequence that is not null, this proposition is
how Chapter 5 inverts a non-zero real number.
Small steps
do not make a Cauchy sequence
The commonest wrong proof that a sequence is Cauchy
looks only at neighboring terms.
Warning. “If , then
is Cauchy” is
false. The next example gives a
sequence of rationals whose steps shrink to and whose terms do not
settle down at all.
Example
(Shrinking Steps, Terms That Do Not Settle)
Here is an argument for the false claim in the
warning.
“Let . Since ,
choose with for every . Let . Then
is the
sum of the consecutive differences for , and
each of them is smaller than in absolute
value. So , and is Cauchy.”
In , let
and
for . Show that , that
for every , and that is not Cauchy. Find
the line of the argument that fails.
Solution. The sequence. By
Sets, Relations, and Cardinality, Theorem 2.3.1
(read on as Theorem 1.5.1,
by Corollary 1.7.3
(b)), with ,
starting value
and , there is exactly one sequence in with and . Its first terms are .
Shrinking steps. , and is null in (Example (First Limits)).
Long stretches. Fix . We show, by Logic and Proof, Axiom 4.2.1
on (through Corollary 1.7.3
(b)), that Base : both sides are
. Step: assume it
for . Since , we have
, so (Proposition 2.5.2
(b)). With , Proposition 4.1.2
(d) gives . Hence, by the induction hypothesis,
transitivity (Proposition 2.3.5
(i)) and adding to both sides (Proposition 2.3.5
(b)), Taking gives . (Directly: each of the
terms
is at least .)
Not Cauchy. We prove the negation displayed
after Definition 4.5.1
with . Let , and put and . Then
, , and . So no threshold works for the tolerance
.
The failing line. “Each of them is smaller
than …
so .” A sum of terms, each of
absolute value below , is only
below (triangle inequality, Proposition 4.1.3
(e), applied times). The number of summands is not fixed: is chosen
after , and the definition
allows it to be as far beyond as we like. To repair
the line one would need each step below , a
tolerance that depends on and ; no single is promised for it. For
, with and , the steps are each below
, and
their sum is at least .
Section 4.6 goes further: its Exercise B2 shows for
every , so
even grows
without bound.
Exercises
A. Check your
understanding
Exercise
(A1)
These questions are about Theorem 4.5.4
and its proof.
- (a) Write “ is not null” in
symbols, with every quantifier and its domain.
- (b) The proof asks
the Cauchy condition for the tolerance . Why not
itself?
- (c) Where is the
inequality used?
- (d) Run the proof in
on with
:
check that this witnesses
“not null”, give the least that the Cauchy
condition with tolerance allows, choose , and state the and the proof
produces.
- (e) State the idea
of the proof in two sentences.
Solution.
(a) : one
tolerance and, beyond every threshold, some term at
least that far from .
(b) With
tolerance , Case 1 would
give only :
every later term is positive, but no positive with comes out, and
the terms might still approach . Any tolerance strictly
smaller than leaves a
positive margin; is the
simplest.
(c) In the line
“since both and
are , ”: the Cauchy estimate holds only
for pairs of indices beyond . It is the reason
is chosen first
and the negation is applied at .
(d) Here , which is at and at least for every (as ). So beyond
any the index
has :
witnesses “not null”. Next, . For , both reciprocals lie in , so their
difference is below , as in the examples
after Definition 4.5.1.
For the
pair , gives . So the least is . Take : , ,
Case 2. The proof gives for , and it states
the result as , : for every
. ( is not allowed:
it is below ,
and .)
(e) “Not null”
provides one tolerance and, beyond
any threshold, a term at least from . Taking the threshold
from the Cauchy condition with first,
every later term stays within of that one
term, so it keeps its sign and stays at least from .
B. Practice
Exercise
(B1)
Determine which of the following sequences in are Cauchy. Justify
your answers, and for the first one work directly from
Definition 4.5.1.
- (a) .
- (b) .
- (c) .
Solution.
(a) Cauchy. For
every ,
, so . Let , and use Corollary 2.5.4
to choose with . For
, both
and lie in , since (Proposition 2.5.2
(b)). Two elements of differ by less than (as in the examples
after the definition), so .
(b) Not Cauchy.
Take . For any , the indices and have and , so .
(c) Not Cauchy.
If it were, it would be bounded (Proposition 4.5.2):
for
some and every . By Theorem 2.5.3
there is
with , a contradiction. (Archimedes is used
here; in the non-Archimedean field of Section 4.2 the
sequence is bounded, by , and the boundedness
argument says nothing.)
Exercise
(B2)
Let and
be Cauchy
sequences in .
Prove directly from Definition 4.5.1
that is
Cauchy, naming the rule behind each step.
Solution. Let . By Proposition 4.1.2
(b), . Since is Cauchy, choose
with for all ; since is Cauchy, choose
with for all . Let (a maximum of two natural numbers; it
depends only on ). Let ; then and . By the ring
rules (Proposition 2.3.2),
By the triangle inequality (Proposition 4.1.3
(e)) and (Proposition 4.1.3
(b)), adding strict inequalities by Proposition 2.3.5
(b) and using Proposition 4.1.2
(b) for the equality. Since was
arbitrary, is Cauchy.
Exercise
(B3)
Let in .
Show that is
Cauchy and not null, and find and
as in Theorem 4.5.4.
Then run the proof of the theorem with and the
least admissible : which and does it produce?
Solution. Cauchy. is the difference
of the constant sequence and the Cauchy sequence
(examples after Definition 4.5.1),
so it is Cauchy by Theorem 4.5.3
(a), (b).
Not null, and , directly. For
every , ,
so . Then
for
every , so witnesses
“not null”, and ,
satisfy the first alternative. (Any smaller positive
, for instance
, works with
as well.)
The proof run. With , the
Cauchy condition is asked for tolerance . Here ; for both reciprocals lie in , so the
difference is below , while fails (, : ). So . Take : , Case 1. The proof gives , : for every .
C. Going deeper
Exercise
(C1)
Let be a
Cauchy sequence in and a sequence in such that is null.
Prove that is
Cauchy.
Here is the beginning of a proof. Name the move,
then complete it.
“Let . For all , the
triangle inequality (Proposition 4.1.3
(e)), applied twice, gives …”
Solution. The move is
split , now
into three parts, one for each term on the right: two
come from the null difference, one from the Cauchy
condition.
Completion. Let ; then (Proposition 4.2.1
(b)), so (Proposition 2.5.2
(a), (O4)), and : by (F5)–(F9), with . Since is null,
choose with
for every . Since is Cauchy, choose
with for all . Let , which depends only on . For , using (Proposition 4.1.3
(b)), adding three strict inequalities (Proposition 2.3.5
(b)). So is
Cauchy.
The split , , , which
needs only halving, works as well.
Exercise
(C2: Exit check: the sign theorem)
Let be a
Cauchy sequence in an ordered field that is not null. Prove
that there are and such that either for every , or for every
.
Solution. “Null” means . Its
negation (Logic and Proof, Proposition 6.2.3,
Logic and Proof, Proposition 6.2.1
(a), Logic and Proof, Proposition 6.2.2,
then (O1)) is Fix such an ; it stays
fixed from here on.
Apply the Cauchy condition with the tolerance (Proposition 4.1.2
(b)): there is
with for all . Apply the
negation at :
there is with . By the definition of ,
or .
Let .
As ,
(Proposition 4.1.3
(d)). If , then . If ,
then . With
and , the first case gives and the second
, for
every .
Marking scheme. (10 points) - 2:
negates “null” correctly, with the quantifiers in order,
and fixes one . - 2: applies
the Cauchy condition with a fixed tolerance such that (for example )
to get . - 2:
obtains
with from the negation applied
at . - 3: derives the
one-signed bound (with ) for every , in both sign
cases. - 1: names
and and
concludes. Fatal errors (0 for the step): chosen before , or from the negation
at a threshold unrelated to ; “
for all large ”
asserted from “not null” alone. A complete proof by
contradiction (assume neither alternative holds for any
, , and refute it at ,
) maps onto
the same 2 + 2 + 2 + 3 + 1.
Exercise
(C3: Find the error)
Here is an argument that the product of two Cauchy
sequences ,
in is Cauchy.
“For all ,
. Given , choose such that and for all . Then for
, .”
Is the conclusion true? Does the argument prove it?
If not, name the line that fails, say why, and repair
it.
Solution. The conclusion is
true: it is Theorem 4.5.3
(c).
The argument does not prove it. The failing
line is “choose
such that … for all ”. The
tolerances
and
depend on and
, which are
introduced after . The definition of
Cauchy supplies a threshold for one fixed tolerance
, named before the threshold; here the
threshold is asked to serve a tolerance that changes
with the very indices it controls. This is a witness
depending on variables introduced later, entry (N2) of
Logic and Proof, Example (What is not allowed).
A counterexample to the step. The step gets
one for all the
tolerances
from their being positive. In take the Cauchy
sequence , the never-zero (but unbounded) , and . For any
, and give which is not smaller: no works. Inside the
exercise is
Cauchy, hence bounded, but the argument never uses that;
the repair does.
Repair. Fix bounds first: by Proposition 4.5.2,
choose with and for every
, before is named.
Then the fixed tolerances (when ) and give
thresholds ,
, and works, as in the proof of Theorem 4.5.3
(c).