ARE211, Fall 2004

LECTURE #3: THU, SEP 16, 2004 PRINT DATE: SEPTEMBER 27, 2004 (ANALYSIS3)

C ONTENTS

1. Analysis (cont) 1

1.6. Two Preliminary Results 1

1.7. Cauchy Sequences 2

1. A NALYSIS ( CONT )

1.6. Two Preliminary Results

Theorem: (Theorem A for future ref): If P ⊂ Q then sup(P) ≤ sup(Q) and inf(P) ≥ inf(Q).

Obvious just by looking at a picture: if P is smaller than Q then then every upper bound to Q must also be
an upper bound to P and there may be upper bounds to P that aren’t upper bounds to Q. Doing the proof is
just a matter of saying the obvious in symbols.

Proof: We’ll just prove the first half about sups. Pick P, Q ⊂ R such that P ⊂ Q. if sup(Q) = ∞, there’s
nothing to prove because any possible sup for P has to be less than or equal to infinity. Assume therefore
that sup(Q) = b̄. We’ll consider b > b̄ and prove that b cannot be a least upper bound for P. This will
establish that sup(P) ≤ sup(Q). Well, b̄ ≥ q, for every q ∈ Q. Since P ⊂ Q, b̄ ≥ p, for every p ∈ P. Hence b̄
is an upper bound for P. Since b > b̄, b cannot be a least upper bound for P. 

Theorem: (Theorem B for future ref): Given S ⊂ R, b ∈ R is a least upper bound (supremum) iff b is an
upper bound for S and if for every ε > 0, there exists s ∈ S such that b − s < ε.

Proof: : Need to prove this in both directions, i.e., (a) if b satisfies the requirements of the theorem then it is
a supremum; (b) if it doesn’t satisfy the requirements of the theorem, then it isn’t a supremum.
1
 2 LECTURE #3: THU, SEP 16, 2004 PRINT DATE: SEPTEMBER 27, 2004 (ANALYSIS3)

g replacements S ε

b0 b

F IGURE 1. Proof of (b)

Proof of (a): Assume that for every ε > 0, there exists s ∈ S such that b − s < ε. We need to show that b ≤ b 0
for every upper bound b0 for S. We’ll do this by showing that if b 0 < b then b0 cannot be an upper bound for
S. Pick b0 = b − ε, for some ε > 0. By assumption, ∃s ∈ S s.t. b − s < ε = b − b 0 , i.e., −s < −b0 or s > b0
proving that b0 isn’t an upper bound for S.

Proof of (b): Now assume that one of the requirements of the theorem isn’t satisfied. The first requirement
is that b is an upper bound for S. If this requirement is violated, then trivally b can’t be a supremum; Now
suppose that the ∀ε condition fails, i.e., ∃ε > 0 such that ∀s ∈ S, b − s ≥ ε (see Fig. 1). We’ll show that b
cannot be a least upper bound for S, i.e., we’ll show that there exists b 0 < b such that b0 is an upper bound
for S. Pick b0 = b − ε. Since b − s ≥ ε, for all s ∈ S, then b 0 = b − ε > s, for all s ∈ S. Hence b0 is an upper
bound for S and is also smaller than b. So we’ve proved that b isn’t a least upper bound. 

1.7. Cauchy Sequences

The notion of a Cauchy sequence is another tool that we will need later.

Definition: A sequence {xn }, xn ∈ R is called a Cauchy sequence with respect to a metric d if ∀ε > 0 ∃N ∈ N
such that for n, m ≥ N, d(xn , xm ) < ε.

That is, the sequence bunches up as n progresses. More specifically, if you have the graph of a Cauchy
sequence, you can draw a line above the sequence (i.e., which bounds the sequence from above) and a line
below the sequence (i.e., which bounds the sequence from below) with the property that the two lines get
closer and closer together.

Example: The sequence {xn } defined by xn = (−1)n /n.

Notice the difference between a Cauchy sequence and a convergent sequence: a convergent sequence con-
verges to a point; a Cauchy sequence bunches up, but there may be nothing that it converges to. For example,
consider the sequence 1 − 1/2, ..., 1 − 1/n, ... in the set (0, 1). The sequence satisfies the defn of a Cauchy
sequence, but it is not a convergent sequence. The thing it is trying to converge to just isn’t in the set.

That is, sequences that are trying to converge to something, whether or not the something they are trying to
converge to doesn’t exist, are called Cauchy sequences.

A better example of a Cauchy sequence that isn’t a convergent sequence  is the sequence of continuous
−1 if x ≤ −1/n






functions we defined in a previous lecture, { f 1 , f2 , ... fn ...}, where f n = nx if − 1/n < x < 1/n .




1 if x ≥ 1/n


While we haven’t defined a metric on the space of functions 1, you can see intuitively that the f n ’s are in a
certain sense getting closer and closer together, i.e., their graphs are becoming indistinguishable from each
other. (That is, this sequence is Cauchy with respect to some metrics on the space of continuous functions—
such as the metric-like function defined in footnote 1—and isn’t Cauchy with respect to others.) However,
the sequence doesn’t have a limit in the space of continuous functions: if f were a limit function, it would
have to have the property that f (x) = 1, for all x > 0 and f (x) = −1 for all x < 0. But what happens at
zero?? Clearly the function can’t be continuous. This is a typical example of a sequence of objects that all
belong to the same set (i.e., the set of continuous functions), in which the elements get closer and closer
together, but don’t converge to anything in that original set.

The real number system has the following very special property:

Theorem: Every sequence in R that is Cauchy w.r.t. some metric converges with respect to that metric to a
number x ∈ R.

Why do we care? One reason is that in many cases in economics, we want to prove that some scalar, vector,
or function exists. E.g., proving the existence of equilibrium prices for an economy: a price could be a scalar,
a vector or in some cases a function. Often, the easiest way to proceed is to construct a Cauchy sequence
of, say, equilibrium prices of some sequence of nearby economies, and then argue that that the limit of this
sequence is the equilibrium we need. But if the limit isn’t the same kind of object as the members of the

1 In class I proposed the following candidate for a metric (don’t know if it satisfies allt he axioms of a metric, however, but for present purposes
it will do: Given f : → and g : → , define δ( f , g) = `( f , g) × h( f , g), where `( f , g) is the “length” of the set of points X ⊂ on which
f and g differ and h( f , g) is max{| f (x) − g(x)| : x ∈ X}.
4 LECTURE #3: THU, SEP 16, 2004 PRINT DATE: SEPTEMBER 27, 2004 (ANALYSIS3)

Cauchy sequence, then we can’t pursue this route. E.g., our equilibrium price function might be required to
be a continuous function of, say, some signal; but the limit of a Cauchy sequence of continuous functions
need not be a continuous function, so in this case, the Cauchy approach won’t work.

The above theorem is another way of saying that the real number system doesn’t have any holes in it.
Obvious as it is, the proof involves some work. Essentially the proof formalizes the picture above: draw line
above and below the graph of the sequence, so that you aren’t leaving much “air” in between. We’ll argue
that the lines have to get closer and closer together. Trick is to make this idea precise.

Proof: We’ll prove this theorem for the metric defined by d(x, y) = |x − y|, though it’s true for any metric. We
first need to show that every Cauchy sequence is bounded both above and below. Set ε = 1; ∃N ∈ N such that
for n, m ≥ N, d(xn , xm ) < 1. The set {x1 , ...xN , xN+1 } is a finite set and has a maximum. Call it x̄. (Question:
why wouldn’t the proof below be correct if we kept everything the same, except we defined x̄ to be the
maximum of the set {x1 , ...xN }?) Clearly, x̄ + 1 is an upper bound for the entire sequence: let m = N + 1, and
note that for all n > N, since d(xm , xn ) = |xn − xm | < 1, we have that for all n > N, xn < xN+1 + 1 ≤ x̄ + 1.
(Note that this is where we use the specific functional form for the metric.) Similarly, the set is bounded
below.

Now for each m ∈ N, let Am = {xm , xm+1 , ...,} and let am = sup(Am ). From Theorem A, the sequence
{am }∞
m=1 is a nonincreasing sequence. Moreover the sequence is bounded below (previous paragraph).

Hence the sequence {am }∞

We’ll show {xn } also converges to a. Fix ε > 0.

• pick N1 sufficiently large that ∀n, m ≥ N1 , d(xn , xm ) < ε/3 (using the fact that the sequence is
Cauchy.)
• pick N2 sufficiently large that ∀n ≥ N2 , d(an , a) < ε/3. (using the fact that the sequence of sup’s
converges to a).
• let N = max{N1 , N2 }
• Note that for n, m > N both of the following are true:
– ∀n, m ≥ N, d(xn , xm ) < ε/3
– ∀n ≥ N, d(an , a) < ε/3.

2 Recall from last lecture the The Axiom of Completeness: Every non-decreasing sequence {x } in that is bounded above converges to a point
n
x ∈ in the Pythagorian metric.
 ARE211, Fall 2004 5

That is, all of the points beyond the N’th in the sequence are bunched together and the supremum of
all of these points is close to a. To tie the proof together we just have to relate all of these points in
the tail of the sequence to the supremum of these points and we are done.
• pick K > N such that d(aN , xK ) < ε/3: we can do this for the following reason: a N is the sup of
AN = {xN , xN+1 , ...,}; by Theorem B, there’s some point in A N which is within ε/3 of aN ; Let K
denote the index of this point.

Now apply the triangle inequality which just formalizes the ideas that: the points in the sequence are
bunched; the suprema are close to a; the top end the sequence is close to the supremum. Formally, for
arbitrary n > N, we have

d(xn , a) ≤ d(xn , xK ) + d(xK , aN ) + d(aN , a)

≤ ε/3 + ε/3 + ε/3 = ε 

Question: What part of the proof invokes the special property of the real line. That is, the theorem isn’t true
for Cauchy Sequences in general.

Answer: We utilize the fact that every nonincreasing, bounded sequence in R converges to a point in R. In
this case the nonincreasing sequence is the sequence of sups.