The Axiom of Completeness for R

(8/7/23)

Alex Nita

Contents
1 Axiom of Completeness 2

2 Monotone Convergence 4

3 Nested Interval Property 5

4 Archimedean Property 5

5 Bolzano-Weierstrass Theorem 5

6 Cauchy Criterion 5

7 Intermediate Value Theorem 7

8 Equivalent Characterizations of the Completeness of R 8

1
1 Axiom of Completeness

Definition 1 A collection A ⊆ R of real numbers is said to be bounded above by

a real number M , the upper bound, if every number x in the collection A satisfies
x ≤ M . Let us write
def
Au = {all upper bounds of A}
and let us write
def
sup A = min Au
for the supremum or least upper bound, which is by definition the minimum, or
smallest of the upper bounds, if it exists. ■

Axiom 1 The Axiom of Completeness (AoC) postulates the existence of

min Au in R whenever Au ̸= ∅ and calls it the supremum, sup A. ■

Remark 1 We can form the analogous statement for greatest lower bound, but the
existence of this turns out to be deducible from the existence of the least upper bound,
so it doesn’t need a separate axiom. ■

Definition 2 A collection A ⊆ R of real numbers is said to be bounded below by

a real number M , the lower bound, if every number x in the collection A satisfies
M ≤ x. Let us write
def
Aℓ = {all lower bounds of A}
and let us write
def
inf A = max Aℓ
for the infimum or greatest lower bound, which is by definition the maximum, or
biggest of the upper bounds, if it exists. ■

Proposition 2 (Existence of the Infimum) AoC implies the existence of inf A

for any nonempty set A ⊆ R satisfying Aℓ ̸= ∅. In fact,

inf A = − sup(−A)

def
where −A = {−a ∈ R | a ∈ A}.

Proof: Since Aℓ ̸= ∅, ∃ℓ ∈ Aℓ satisfying ℓ ≤ a for all a ∈ A. Therefore −a ≤ −ℓ

for all −a ∈ −A, which shows that (−A)u ̸= ∅. By AoC, s = sup(−A) = min(−A)u
exists in R, so we just need to show that −s = inf A = max Aℓ : take any ℓ ∈ Aℓ , then
−ℓ ∈ (−A)u , so s ≤ −ℓ, which means ℓ ≤ −s, and indeed −s = max Aℓ = inf A. ■

2
Proposition 3 (Characterization of sup A) Let A ̸= ∅ and Au ̸= ∅. For
any s ∈ Au we have


s = sup A ⇐⇒ ∀ε > 0, ∃a ∈ A, s − ε < a

Proof: (i) s = sup A = min Au =⇒ ∀ε > 0, s−ε ∈ / Au =⇒ (∃a ∈ A, s−ε < a).


(ii) If s ∈ Au and ∀ε > 0, ∃a ∈ A, s−ε < a , then, to see that s = min Au , consider
any other b ∈ R. We wish to prove that b ∈ Au =⇒ s ≤ b, by the contrapositive: If
def
/ Au (because ∃a ∈ A,
s ̸≤ b, that is, if s > b, then ε = s − b > 0 satisfies s − ε = b ∈
s − ε < a). ■

Proposition 4 (Characterization of inf A) Let A ̸= ∅ and Aℓ ̸= ∅. For

any t ∈ Aℓ we have


t = inf A ⇐⇒ ∀ε > 0, ∃a ∈ A, a < t + ε

Proof: (i) t = inf A = max Aℓ =⇒ ∀ε > 0, t+ε ∈ / Aℓ =⇒ (∃a ∈ A, a < t+ε).


(ii) If t ∈ Aℓ and ∀ε > 0, ∃a ∈ A, a < t + ε , then, to see that t = max Aℓ , consider
any other b ∈ R. We wish to prove that b ∈ Aℓ =⇒ b ≤ t, by the contrapositive: If
def
/ Aℓ (because ∃a ∈ A,
b ̸≤ t, that is, if b > t, then ε = b − t > 0 satisfies t + ε = b ∈
a < t + ε). ■

Proposition 5 (Maximum vs Supremum) Let A ̸= ∅ and Au ̸= ∅. If

max A exists, then
max A = sup A

Proof: By definition, if m = max A exists, then m ∈ A ∩ Au . Therefore, s =

sup A = min Au ≤ m, but since m ∈ A, we also have m ≤ s, which together imply
m = s (by Axiom 2.3 of R). ■

Proposition 6 (Uniqueness of the Supemum and Infimum) For any sub-

set A ⊆ R, the real numbers sup A and inf A are unique, if they exist.

Proof: Suppose s = sup A = min Au exists, and that t = sup A = min Au , too.
Then s = t because the minimum, when it exists, is unique: s ≤ t (because s =
min Au and t ∈ Au ) and s ≥ t (because t = min Au and s ∈ Au ) together imply s = t
(by Axiom 2.3 of R), . Similarly, the maximum is unique: if ℓ, k = inf A = max Aℓ ,
then ℓ ≥ k and ℓ ≤ k together imply ℓ = k (by Axiom 2.3 of R). ■

3
2 Monotone Convergence

Definition 3 A convergent sequence (an )n∈N of real numbers, with limit a ∈ R,

satisfies

∀ε > 0, ∃N ∈ N, n ≥ N =⇒ |an − a| < ε ■

Definition 4 A bounded sequence (an )n∈N in R satisfies

∃M > 0, ∀n ∈ N, |an | ≤ M ■

Definition 5 A sequence (an )n∈N in R is called increasing if

an ≤ an+1 , ∀n ∈ N

and called decreasing if

an ≥ an+1 , ∀n ∈ N
If a sequence is either increasing or decreasing it is called monotone. ■

Axiom 2 The Monotone Convergence Theorem (MCT), as an axiom of

R, postulates the existence of a limit for every bounded monotone sequence in
R. ■

Definition 6 A subsequence (ank )k∈N of a real sequence (an )n∈N is a new sequence
derived from the original sequence by selecting a subset {ank |k ∈ N} of its terms while
maintaining their original relative order, that is by indexing them by an increasing
sequence n1 < n2 < n3 < · · · . ■

Lemma 7 Every sequence of real numbers has a monotonic subsequence.

Proof: Let (an )n∈N be a sequence in R, and call the m-th term dominant, or
a peak term, if for all n ≥ m we have an ≤ am . Either there are infinitely
many dominant terms (example: an = 1/n) or there are finitely many (example:
a1 = 1, an = 1 − 1/n for n ≥ 2, has one dominant term). If there are infinitely
many, let (ank )k∈N be the subsequence consisting solely of dominant terms: it is
clearly monotonic decreasing. If there are finitely many, let ar denote the last
dominant term and choose n1 > r. Then, for all N ≥ n1 , aN is not dominant, so
∃n2 ≥ N such that an2 > an1 . Repeat the process inductively: we have a monotonic
increasing subsequence (ank )k∈N . ■

4
3 Nested Interval Property

Axiom 3 The Nested Interval Property (NIP), taken as an axiom of R,

postulates that every nested sequence of closed intervals of real numbers has
nonempty intersection,
)
(1) I1 ⊇ I2 ⊇ I3 ⊇ · · · \
=⇒ In ̸= ∅ ■
(2) In = [an , bn ] ⊆ R n∈N

4 Archimedean Property

Axiom 4 The Archimedean property (AP) of R postulates, for any real

number x ∈ R, the existence of a larger natural number n ∈ N,

∀x ∈ R, ∃n ∈ N, x < n ■

5 Bolzano-Weierstrass Theorem
Axiom 5 The Bolzano-Weierstrass Theorem (BW), taken as an axiom
of R, postulates that every bounded sequence (an )n∈N of real numbers has a
convergent subsequence. ■

6 Cauchy Criterion

Definition 7 A Cauchy sequence (an )n∈N of real numbers satisfies

∀ε > 0, ∃N ∈ N, n, m ≥ N =⇒ |an − am | < ε ■

Axiom 6 The Cauchy Criterion (CC) for R postulates the logical equiva-
lence, for a real sequence, of being Cauchy and being convergent,

(an )n∈N is convergent in R ⇐⇒ (an )n∈N is Cauchy in R ■

5
Regardless of which of the axioms we take as our starting point, we have half of the
CC equivalence as a mere consequence of our definitions of convergent sequences and
Cauchy sequences:

Proposition 8 Every convergent sequence is a Cauchy sequence.

Proof: Let (an )n∈N be a convergent sequence, with limit lim an = a in R. Then,
n→∞
∀ε > 0, ∃N ∈ N, (n ≥ N =⇒ |an − a| < 2ε ). If n, m ≥ N , then

|an − am | = |an −a + a−am |

≤ |an − a| + |a − am |
ε ε
< + = ε
2 2
which shows that (an )n∈N is Cauchy. ■

Proposition 9 Every Cauchy sequence is bounded.

Proof: Let (an )n∈N be a Cauchy sequence. Then, ∀ε > 0, ∃N ∈ N, (m, n ≥ N =⇒

|an − am | < ε). Let’s take a concrete numerical value, say ε = 1, then

|an − aN | < 1

for all n ≥ N , for the appropriate N , which expands into the double inequality

−1 < an − aN < 1 ⇐⇒ −|aN | − 1 ≤ aN − 1 < an < aN + 1 ≤ |aN | + 1

⇐⇒ |an | < |aN | + 1

Thus, the tail of the sequence is bounded by |aN | + 1. The first N − 1 terms, too,
are bounded, by M = max{|a1 |, . . . , |aN −1 |}, so the whole sequence is bounded by
def
K = max{M, |aN | + 1}. ■

Proposition 10 If a Cauchy sequence (an )n∈N has a convergent subsequence

(ank )k∈N with lim ank = a, then the sequence (an )n∈N is convergent and lim an =
k→∞ n→∞
a.

Proof: Let (an )n∈N be a Cauchy sequence and suppose lim ank = a. Then,
k→∞
ε

• ∀ε > 0, ∃N ∈ N, m, n ≥ N =⇒ |an − am | <
2
ε

• ∀ε > 0, ∃M ∈ N, k ≥ M =⇒ |ank − a| <
2
def
For any ε > 0, then, ∃K = max{N, M },

n, nk ≥ K =⇒ |an − a| = |an −ank + ank −a|

≤ |an − ank | + |ank − a|
ε ε
< +
2 2
= ε

We conclude that an → a. ■

6
7 Intermediate Value Theorem

Definition 8 Let A ⊆ R. We call B ⊆ A (relatively) open in A if B = O ∩ A for

some O ∈ TR , and we call B ⊆ A (relatively) closed in A if B = C ∩ A for some
C ∈ CR . Let us denote the set of (relatively) open subsets of A by TA and the set of
(relatively) closed subsets of A by CA . ■

Let f : A → B, where A, B ⊆ R.

Definition 9 (Continuity) We say that f is continuous at a point a ∈ A, and

write
lim f (x) = f (a)
x→a

if

∀ε > 0, ∃δ > 0, |x − a| < δ =⇒ |f (x) − f (a)| < ε

We say f is continuous on a set A if it is continuous at all a ∈ A. Let

def
C(A) = {f : A → R | f is continuous on A}

denote the set of all continuous functions on A. ■

Theorem 11 (Convergence Criterion for Continuous Functions)

Let A ⊆ R, and let f : A → R. For any a ∈ A,

lim f (x) = f (a) ⇐⇒ lim f (an ) = f (a) = f ( lim an )

x→a n→∞ n→∞

for all sequences (an )n∈N in A with an → a.

Proof: See Theorem 12, Lecture 10. ■

Axiom 7 The Intermediate Value Theorem (IVT), taken as an axiom of

R, postulates the existence of all intermediate y-values, between f (a) and f (b),
for any continuous function f on a closed and bounded interval [a, b],

∃c ∈ [a, b], f (c) = y

whether f (a) ≤ y ≤ f (b) or f (a) ≥ y ≥ f (b).

7
8 Equivalent Characterizations of the Completeness of
R
Theorem 12 (Completeness of R) The following are logically equivalent as
axioms of R,
(1) AoC ( =⇒ AP)
(2) MCT ( =⇒ AP)
(3) NIP + AP
(4) BW ( =⇒ AP)
(5) CC + AP
(6) IVT

Remark 13 I have tried to indicate above the fact that AoC, MCT and BW each
implies AP, whereas NIP and CC do not imply AP, but require it as an ancillary
assumption. See the counterexamples following the proof. ■
Proof :
(i) (1) =⇒ (2), or AoC =⇒ MCT (Theorem 2.4.2 in Abbott): Suppose AoC
(∅ ̸= A ⊆ R with Au ̸= ∅ =⇒ s = sup A = min Au exists) and let us deduce
MCT (every bounded monotone sequence of real numbers converges).

def
If (an )n∈N is an increasing bounded sequence in R, then A = {an | n ∈ N}
is nonempty and bounded, and we claim that
def
lim an = s = sup A
n→∞


By Proposition 3 we know ∀ε, ∃aN ∈ A, s − ε < aN , which means that


∃N ∈ N, n ≥ N =⇒ aN ≤ an since (an )n∈N is increasing


=⇒ |an − s| = s − an ≤ s − aN < ε

and this shows that lim an = s exists in R.

n→∞
def
If (an )n∈N is decreasing and bounded, then A = {an | n ∈ N} is nonempty
and bounded, and we claim that
def
lim an = t = inf A
n→∞


By Proposition 4 we know ∀ε, ∃aN ∈ A, aN < t + ε , which means that


∃N ∈ N, n ≥ N =⇒ an ≤ aN since (an )n∈N is decreasing


=⇒ |an − t| = an − t ≤ aN − t < ε

and this shows that lim an = t exists in R. ■

n→∞

8
(ii) (1) =⇒ (3), or AoC =⇒ NIP + AP (Theorems 1.4.1-1.4.2 in Abbott):
Suppose AoC (∅ ̸= A ⊆ R with Au ̸= ∅ =⇒ s = sup A = min Au exists) and
let us deduce NIP (every nested sequence of closed intervals of real numbers has
nonempty intersection) and AP (∀x ∈ R, ∃n ∈ N, x < n, and ∀x > 0, ∃n ∈
N, 0 < n1 < x).

We first prove AoC implies NIP: Consider a sequence of nested intervals

I1 ⊇ I2 ⊇ · · · , where each InT= [an , bn ] is a closed subinterval of R, and let
n
us show that AoC implies n=1 In ̸= ∅. The nestedness of the intervals
can be expressed as a sequence of inequalities:

a1 ≤ a2 ≤ · · · ≤ an · · · ≤ bn ≤ · · · ≤ b2 ≤ b1

From this we conclude that the sets

def
A = {an | n ∈ N}
def
B = {bn | n ∈ N}

are nonempty and bounded (by M = max{|a1 |, |b1 |}), and moreover B ⊆
Au and A ⊆ B ℓ . Since we assumed AoC, we have assumed that

sup A = min Au
inf B = max B ℓ

exist, and these must satisfy

an ≤ sup A ≤ inf B ≤ bn

for all n ∈ N. (The middle inequality is shown as follows: sup A =

min Au ≤ b for all b ∈ B since B ⊆ Au , so sup A ∈ B ℓ , and therefore
T∞
sup A ≤ max B ℓ = inf B.) We conclude that ∃ sup A, inf B ∈ n=1 In ,
which is therefore nonempty. This proves NIP.
Next, we show that AoC implies AP: The key feature of N is that it pos-
def
sesses a successor function, according to the Peano axioms, s(n) = n + 1
for all n ∈ N. If N were bounded above, that is if Nu ̸= ∅, then AoC would
guarantee the existence of α = sup N = min Nu , which would consequently
satisfy
∀n ∈ N, n ≤ α (8.1)
But then take ε = 1 and recall Proposition 3, which says that

∃n ∈ N, α − 1 < n

We would then be forced into a contradiction of (8.1):

∃s(n) = n + 1 ∈ N, α < n + 1 (8.2)

Conclusion: Nu = ∅, so ∀x ∈ R, ∃n ∈ N, x < n.
For the second part of the AP, let x > 0, and use the first part of AP to
find n ∈ N with x1 < n, which can be rewritten n1 < x. ■

9
(iii) (1) =⇒ (4), or AoC =⇒ BW, is Exercise 2.5.9 in Abbott: Suppose AoC
(∅ ̸= A ⊆ R with Au ̸= ∅ =⇒ s = sup A = min Au exists) and let us deduce
BW (every bounded sequence has a convergent subsequence).

Suppose AoC and consider a sequence (an )n∈N in R bounded by M > 0.

Then all terms an lie in the interval [−M, M ]. Define
def
A = {x ∈ R | x < an for infinitely many an }

and note that −M ∈ A, so A is nonempty, while M ∈ Au , so Au is

nonempty. AoC says ∃s = sup A = min Au , and we claim that there exists
a subsequence (ank )k∈N converging to s: By Proposition 3, s = sup A iff
(∀ε > 0, ∃x ∈ A, s − ε < x), which here means

s − ε < x < infinitely many an

For this same ε > 0, s + ε ∈

/ A (otherwise s would bound it, s + ε ≤ s, an
impossibility), meaning only finitely many an > s + ε. We conclude that

infinitely many an lie in [s − ε, s + ε]

Let us now take εk = k1 for each k ∈ N, and let us form Ik = [s − k1 , s + k1 ].

Choose ank ∈ Ik from among the infinitely many terms lying in it, and
observe that
2
|ank − s| ≤
k
which we could make less than any ε > 0 by using AP to find k > 2ε . This
shows that ank → s. ■

(iv) (1) =⇒ (4) =⇒ (5), or AoC =⇒ BW =⇒ CC: Suppose AoC (∅ ̸=
A ⊆ R with Au ̸= ∅ =⇒ s = sup A = min Au exists) and let us deduce CC (a
sequence of real numbers converges iff it is Cauchy).

We already saw (Proposition 8) that, directly from the definitions, a con-

vergent sequence must be Cauchy. We must therefore prove that, if we
assume AoC, then every Cauchy sequence in R must converge. There are
several routes toward this result, we indicate one. Let (an )n∈N be a Cauchy
sequence in R. Then by Proposition 9 it is bounded, say by M > 0. We
can now repeat verbatim the preceding proof that AoC (together with its
consequent AP) implies that bounded sequences must have a convergent
subsequence, or we can merely cite BW (which AoC implies) and say that
(an )n∈N has a convergent subsequence (ank )k∈N , say with limit a in R. This
recalls Proposition 10, which says that (an )n∈N itself converges to a. ■

10
(v) (2) =⇒ AP and (1), or MCT =⇒ AP + AoC, is Exercise 2.4.4 in Abbott:
Suppose MCT (every bounded monotone sequence of real numbers converges),
and let us deduce AP (∀x ∈ R, ∃n ∈ N, x < n, and ∀x > 0, ∃n ∈ N, 0 < n1 < x)
and AoC (∅ ̸= A ⊆ R with Au ̸= ∅ =⇒ s = sup A = min Au exists).

We first prove MCT implies AP: Let x ∈ R. If x ≤ 0, then any n ∈ N with

n ≥ 1 will do, so suppose x > 0. Now, if x ∈ Nu , then ∀n ∈ N, n ≤ x,
so we can consider the sequence an = n, which is increasing and bounded.
Our assumption, the MCT, then says that this sequence converges, say to
L ∈ R, meaning


∀ε > 0, ∃N ∈ N, n ≥ N =⇒ |n − L| < ε

Since this is true for all ε, let us take a specific numerical value, say ε = 1.
The inequality |n − L| < 1 expands into the double inequality
note
z }| {
−1 < n − L < 1 ⇐⇒ L − 1 < n < L + 1

This is true for all n ≥ N for the appropriate N ∈ N, so it is true in

particular for n + 2 > n ≥ N :

L − 1 < n + 2 < L + 1 =⇒ L − 3 < n < L − 1

| {z }
note

We have reached the contradiction L − 1 < n < L − 1, so we conclude that

/ Nu , and instead that ∃n ∈ N, n > x.
x∈

We next prove MCT+AP implies AoC: Assume MCT and let ∅ ̸= A ⊆ R

with Au ̸= ∅. We will demonstrate that sup A = min Au exists in R.
Choose

a1 ∈ A
b1 ∈ Au

def
If a1 = b1 , then s = a1 = max A, which therefore must equal min Au =
sup A by Proposition 5, and we are done. Otherwise, if a1 < b1 , take the
midpoint between them,
def a1 + b1
c1 =
2
and consider the two cases:
(a) Case 1: [c1 ∩ b1 ) ∩ A ̸= ∅: In this case choose a2 ∈ [c1 , b1 ) ∩ A and
let b2 = b1 .
(b) Case 2: [c1 ∩ b1 ) ∩ A = ∅: In this case, we let a2 = a1 and choose
b2 ∈ [c1 , b1 ) ⊆ Au .
Either way, we have

a1 ≤ a2 ∈ A
b2 ≤ b1 ∈ Au

11
Use the above procedure to define an algorithm which produces an+1 ∈
A and bn+1 ∈ Au from an ∈ A and bn ∈ Au , and notice that
(an )n∈N is increasing and (bn )n∈N is decreasing, and both are bounded
(by max{|a1 |, |b1 |}). MCT says the sequences converge in R:

lim an = a
n→∞
lim bn = b
n→∞

Moreover, since bn − cn = cn − an = 21 (bn − an ) for all n ∈ N, we have

1
bn+1 − an+1 ≤ (bn − an )
2
1
≤ (bn−1 − an−1 )
22
..
.
1
≤ (b1 − a1 )
2n
Since 21n < 1
n for all n ∈ N, for every ε we can use the AP to find an N ∈ N
for which
1 1 ε
n ≥ N =⇒ ≤ <
n N 3(b1 − a1 )
We conclude that
ε

∀ε > 0, ∃N ∈ N, n ≥ N =⇒ |bn+1 − an+1 | <
3
If, for this same ε, we find M ∈ N so large that for n ≥ max{N, M } we
have |b − bn+1 | < 3ε and |b − bn+1 | < 3ε , we finally reach:

|b − a| = |b − bn+1 + bn+1 − an+1 + an+1 − a|

≤ |b − bn+1 | + |bn+1 − an+1 | + |an+1 − a|
ε ε ε
< + +
3 3 3
= ε

Since this is true for all ε > 0, we have that a = b. We claim that this
common value is the supremum of A,
def
s = sup A = a = b

To see this, use Proposition 3: Let ε > 0, and use the above results to get
that ∃N ∈ N, (n ≥ N =⇒ a − an < ε), which implies that a − ε < an .
Since an ∈ A, this shows that a = b = s = sup A. ■

Remark 14 Concerning our algorithm, we should include a line in the

inductive step: if an = bn for some n ∈ N in the procedure, we should
stop the algorithm, because an = bn ∈ A ∩ Au =⇒ an = bn = max A =
min Au = sup A. ■

12
(vi) (2) =⇒ (3), or MCT =⇒ NIP+ AP, is Exercise 2.4.4 in Abbott: Suppose
MCT (every bounded monotone sequence of real numbers converges), and let
us deduce NIP (every nested sequence of closed intervals of real numbers has
nonempty intersection) and AP (∀x ∈ R, ∃n ∈ N, x < n, and ∀x > 0, ∃n ∈
N, 0 < n1 < x).

We first prove MCT implies NIP: Consider a sequence of nested intervals

I1 ⊇ I2 ⊇ · · · , where each InT= [an , bn ] is a closed subinterval of R, and let
n
us show that MCT implies n=1 In ̸= ∅. The nestedness of the intervals
can be expressed as a sequence of inequalities:

a1 ≤ a2 ≤ · · · ≤ an · · · ≤ bn ≤ · · · ≤ b2 ≤ b1

From this we conclude that (an )n∈N is increasing, (bn )n∈N is decreasing,
and both are bounded (by M = max{|a1 |, |b1 |}). MCT applies to give us
the existence of the limits

lim an = a
n→∞
lim an = b
n→∞

in R. The Order Limit Laws say a ≤ b, and this in combination with the
facts an ≤ a and bn ≥ b give

an ≤ a ≤ b ≤ bn
T∞
for all n ∈ N. Therefore, ∃a, b ∈ n=1 In , and the intersection is nonempty.
We next prove MCT implies AP: Let x ∈ R, and suppose by way of con-
tradiction that x ∈ Nu , i.e. n ≤ x for all n ∈ N. Then the increasing
sequence an = n is bounded by x, so MCT applies to give the existence of
a limit, lim n = L:
n→∞

∀ε > 0, ∃N ∈ N, (n ≥ N =⇒ |n − L| < ε)

Expanded to a double inequality, this says L − ε < n < L + ε for all

n ≥ N and all ε > 0. Taking ε = 1, this says all n ≥ N lie in the interval
(L − 1, L + 1) of length 2, an impossibility (since, e.g., n and n + 3 are a
distance of 3 apart yet lie in an interval of length 2). ■

13
(vii) (2) =⇒ (4), or MCT =⇒ BW, is Exercise 2.5.8 in Abbott: Suppose
MCT (every bounded monotone sequence of real numbers converges), and let
us deduce BW (every bounded sequence has a convergent subsequence).

By Lemma 7 we know that every real sequence (an )n∈N (bounded or und-
bounded) has a monotonic subsequence (ank )k∈N . If (an )n∈N is bounded,
then so is (ank )k∈N , and the MCT assumption applies: (ank )k∈N converges
to a limit a ∈ R. ■

(viii) (2) =⇒ (5), or MCT =⇒ CC + AP: Suppose MCT (every bounded monotone
sequence of real numbers converges), and let us deduce CC (a sequence of real
numbers converges iff it is Cauchy). [We have already shown, two proofs above,
that MCT implies AP.]

Let (an )n∈N be a Cauchy sequence in R. By Proposition 9 we know it is

bounded, while by Lemma 7 we know it possesses a monotone subsequence
(ank )k∈N . Our MCT assumption says (ank )k∈N converges, say to a in R.
By Proposition 10 the original sequence itself must also converge to a. ■

(ix) (3) =⇒ (1), or NIP + AP =⇒ AoC, is Exercise 2.5.4 in Abbott: Suppose

NIP (every nested sequence of closed intervals of real numbers has nonempty
intersection) and AP (∀x ∈ R, ∃n ∈ N, x < n, and ∀x > 0, ∃n ∈ N, 0 < n1 < x)
and let us deduce AoC (∅ ̸= A ⊆ R with Au ̸= ∅ =⇒ s = sup A = min Au
exists).

def
Let ∅ ̸= A ⊆ R with Au ̸= ∅, and let us show that s = sup A exists in R.
Since we are assuming NIP, we describe an algorithm which constructs a
nested sequence of closed intervals from A and Au as follows: choose a1 ∈ A
def
and b1 ∈ Au , which automatically satisfy a1 ≤ b1 , and let I1 = [a1 , b1 ].
u u
If a1 = b1 ∈ A ∩ A , then s = a1 = b1 = max A = min A = sup A
exists in R by Proposition 5, and we are done. Otherwise, if a1 < b1 ,
def a1 + b1
take the midpoint c1 = of I1 and see whether c1 ∈ A or in Au –if
2
c1 ∈ A∩A , then again c = max A = min Au = sup A exists by Proposition
u

5 and we are done. Otherwise, if c1 ∈ Au − A, then let a2 = a1 and

b2 = c1 . If c1 ∈ A − Au , then let a2 = c1 and b2 = b1 . Either way, we form
def
I2 = [a2 , b2 ]. Repeat inductively, stopping the algorithm if at any step
an = bn , since then we have found our supremum of A. If the algorithm
runs to infinity, then we have constructed a countable collection of nested
closed intervals
I1 ⊇ I2 ⊇ · · ·
T∞
The intersection n=1 In is therefore nonempty, by NIP.

14
 To see that this intersection consists of only one point, s, we invoke AP:
Since
1 1
bn+1 − an+1 = (bn − an ) = · · · = n (b1 − a1 )
2 2
and since 21n < n1 for all n ∈ N, we have that ∀ε > 0, ∃N ∈ N, (n ≥ N =⇒
1 ε
n < b1 −a1 . This shows that

lim |bn+1 − an+1 | = 0

n→∞

and
T∞ this in turn shows that the intersection consists Tof∞only one point,
n=1 In = {s} (if there were at least two points a, b ∈ n=1 In , say a < b,
then we could take ε = b−a 2 and find an interval In of smaller length than
the distance from a to b, an impossibility).
T∞
Finally, to see that s = sup A, note that s ∈ n=1 In means an ≤ s ≤ bn
for all n ∈ N, which means s ∈ Au . But for any b ∈ Au we also have
s ≤ b, for if b < s, then we could take ε = s−b 2 and reach the following
contradiction: for this ε, there is an n ∈ N such that
s−b
|an − s| ≤ |an − bn | < ε =
2
or b−s
2 < an − s < s−b 2 . The first inequality implies b <
b+s
2 < an ,
contradicting the fact that b ∈ Au . ■

(x) (3) =⇒ (2), or NIP + AP =⇒ MCT: Suppose NIP (every nested sequence
of closed intervals of real numbers has nonempty intersection) and AP (∀x ∈
R, ∃n ∈ N, x < n, and ∀x > 0, ∃n ∈ N, 0 < n1 < x) and let us deduce MCT
(every bounded monotone sequence of real numbers converges).

Let (an )n∈N be a bounded monotone sequence of real numbers. A direct

proof of its convergence would be a repetition of NIP + AP =⇒ AoC =⇒
def
MCT proofs, for we would let A = {an ∈ R | n ∈ N}, which would satisfy
A ̸= ∅, A ̸= ∅ and A ̸= ∅, since an ∈ A, M ∈ Au , −M ∈ Aℓ .
u ℓ

If an is increasing, then by letting a1 = a1 , b1 = M and c1 = a1 +b2

1
and
u
checking whether c1 ∈ A or A , then letting a2 = c1 in the first case,
b2 = c1 in the second case, and repeating, we would get nested inter-
vals In = [an , bn ] whose intersection would be nonempty and in fact, by
AP, consisting of a single point s, namely s = sup A. (See the last proof
above.) Once we established the existence of s = sup A, we would prove
that limn→∞ an = s as in the proof of AoC =⇒ MCT, by using Propo-
sition 3. If an is decreasing, we would similarly establish the existence of
t = inf A and apply Proposition 4 to show limn→∞ an = t.
In short, we can just cite the fact that NIP + AP imply AoC, which in
turn implies MCT, but these two implications can be made explicit upon
demand. ■

15
(xi) (3) =⇒ (4), or NIP + AP =⇒ BW: Suppose NIP (every nested sequence
of closed intervals of real numbers has nonempty intersection) and AP (∀x ∈
R, ∃n ∈ N, x < n, and ∀x > 0, ∃n ∈ N, 0 < n1 < x) and let us deduce BW
(every bounded sequence of real numbers has a convergent subsequence).

Let (an )n∈N be a bounded sequence with |an | ≤ M for all n ∈ N, and let us
show that some subsequence (ank )k∈N converges. The inequality |an | ≤ M
expands into the double inequality

−M ≤ an ≤ M

for all n ∈ N. Let a0 = −M , b0 = M , and I0 = [a0 , b0 ]. Divide I0

into two halves [−M, 0] and [0, M ] and choose for I1 = [a1 , b1 ] that which
contains infinitely many terms an of the sequence (if both contain infinitely
many terms, choose either). Then choose an1 ∈ I1 from among the infinite
choices. Next, bisect I1 and choose for I2 the half which contains infinitely
many terms an , and pick an2 ∈ I2 from among the infinitely many choices.
Repeat inductively. Since

I1 ⊇ I2 ⊇ · · · =⇒ an1 ≤ an2 ≤ · · ·

we have both a sequence of nested closed intervals and an increasing sub-

T∞
sequence (ank )k∈N . By our NIP assumption, n=1 In ̸= ∅, while by our
M 1
AP assumption, |bn − an | < 2n−1 < n−1 can be made less than any ε > 0
1
by finding N ∈ N with N > ε + 1 and taking n ≥ N . Together, these show
T∞
that n=1 In consists of a single point a, the limit of the subsequence, since
|an − a| ≤ |bn − an | < ε for large enough n. ■

(xii) (3) =⇒ (5), or NIP + AP =⇒ CC: Suppose NIP (every nested sequence
of closed intervals of real numbers has nonempty intersection) and AP (∀x ∈
R, ∃n ∈ N, x < n, and ∀x > 0, ∃n ∈ N, 0 < n1 < x) and let us deduce CC (a
sequence of real numbers converges iff it is Cauchy).

That a convergent sequence is Cauchy was proved in Proposition 8, so

suppose (an )n∈N is Cauchy and let us show that NIP + AP imply its
convergence. First, Proposition 9 implies the sequence is bounded: ∀n ∈
N, ∃M > 0, (n ≥ N =⇒ |an | ≤ M ). Let I1 = [−M, M ]. Then, by
taking εk = k1 , the Cauchy property says that ∃Nk ∈ N, (n ≥ N =⇒
|an − aNk | < k1 ), which means that

1 1
n ≥ Nk =⇒ aNk − < an < aNk +
k k
Define Ik = [aNk − k1 , aNk + k1 ] ∩ Ik−1 recursively, and obtain a nested
T∞
sequence of closed intervals I1 ⊇ I2 ⊇ · · · . The NIP says n=1 In ̸= ∅,
while AP says the intersection consists of a single point a (because if Ik =
[αk , βk ], then |αk − βk | < k1 , which shows the lengths of the intervals Ik
tend to 0). Since |an − a| ≤ |αk − βk | < k1 , we have an → a. ■

16
(xiii) (4) =⇒ (1), or BW =⇒ AP + AoC: Suppose BW (every bounded sequence
of real numbers has a convergent subsequence) and AP (∀x ∈ R, ∃n ∈ N, x < n,
and ∀x > 0, ∃n ∈ N, 0 < n1 < x) and let us deduce AoC (∅ ̸= A ⊆ R with
Au ̸= ∅ =⇒ s = sup A = min Au exists).

Let ∅ ̸= A ⊆ R and suppose Au ̸= ∅. We use the same algorithm as in the

proof of (5) =⇒ (1) below, which constructs Cauchy sequences (an )n∈N in
A and (bn )n∈N in Au , satisfying

a1 ≤ · · · an ≤ an+1 ≤ bn+1 ≤ bn ≤ · · · ≤ b1

for all n ∈ N:
(1) Choose a1 ∈ A ̸= ∅ and b1 ∈ Au ̸= ∅, which automatically satisfy
a1 ≤ b1 .
(a) If a1 = b1 ∈ A ∩ Au , then s = a1 = b1 = max A = min Au =
sup A exists in R by Proposition 5, and we are done.
def a1 +b1
(b) If a1 < b1 , take the midpoint c1 = 2 between a1 and b1
and see whether c1 ∈ A or in Au .
(2) (a) If c1 ∈ A ∩ Au , then again c = max A = min Au = sup A exists
by Proposition 5 and we are done.
(b) If c1 ∈ Au − A, then let a2 = a1 and b2 = c1 . If c1 ∈ A − Au ,
then let a2 = c1 and b2 = b1 .
(3) Repeat inductively, stopping the algorithm if at any step an = bn ,
since then we have found our supremum of A.
If the algorithm runs to infinity, then we have constructed two Cauchy
sequences, (an )n∈N in A and (bn )n∈N in Au , the first increasing, the second
decreasing: let ε > 0 and use AP to find N ∈ N such that
1 1 1 ε
n≥N =⇒ < ≤ <
2n n N 4(b1 − a1 )
b1 − a1 ε
=⇒ |an+1 − bn+1 | < <
2n 4
Therefore, for all m = n + k ≥ n ≥ N , we have

|am − an | = |an+k −bn+k + bn+k −an |

≤ |an+k − bn+k | + |bn+k − an |
≤ |an+k − bn+k | + |bn − an |
b1 − a 1 b1 − a1
< + n−1
2n+k 2
b1 − a1
≤
2n−2
< ε
b1 −a1
Similarly, |bm − bn | < 2n−2 < ε for all m = n + k ≥ n ≥ N . The sequences
are Cauchy.
Cauchy sequences are bounded (Proposition 9), so our BW assumption
says they have convergent subsequences, ank → a and bnk → b.

17
 By Proposition 10, if a subsequence of a Cauchy sequences converges to a
(or b in the second case), then the original sequence converges to a, too,
an → a (and bn → b).
Lastly, by the order limit laws, we know a ≤ b. We claim that in fact
a = b = sup A. By a slight modification of the application of AP, we can
show that |an − bn | < b12−a
n
1
< ε, which shows limn→∞ |an − bn | = 0. Since
an ≤ a ≤ b ≤ bn , the limit laws say

0 ≤ |a − b| ≤ lim |an − bn | = 0
n→∞

which means |a − b| = 0, and therefore a = b. To see that a = b = sup A,

use Proposition 3: Let ε > 0, and use the definition of convergence to get
that ∃N ∈ N, (n ≥ N =⇒ a − an < ε), which implies that a − ε < an .
Since an ∈ A, this shows that a = b = s = sup A. ■

(xiv) (4) =⇒ (2), or BW =⇒ MCT, is Exercise 2.6.7(a) in Abbott: Suppose BW

(every bounded sequence of real numbers has a convergent subsequence) and let
us deduce AP and MCT (every bounded monotone sequence of real numbers
converges).

We first show that BW implies AP, which we will need to show BW im-
plies MCT: If x ≤ 0, then any n ∈ N will do to show n > x, so suppose
def
x > 0. Suppose AP fails to hold, suppose ∃x ∈ Nu , then let an = n and
note that |an | ≤ x for all n ∈ N, so our sequence is bounded. BW says a
subsequence ank = nk (which is still increasing) converges, say to a. Thus,

∀ε > 0, ∃N ∈ N, |ank − a| < ε

Taking a concrete numerical value, say ε = 1, we conclude that

−1 < nk − a < 1 =⇒ a − 1 < nk < a + 1

for all nk ∈ N. Now, nk+2 ≥ nk + 2 yet also satisfies this:

note
z }| {
−1 < nk+2 − a < 1 =⇒ a − 1 < nk+2 < a + 1

and also
a + 1 = 2 + (a − 1) < 2 + nk ≤ nk+2
| {z }
note

/ Nu but instead Nu = ∅,
which is a contradiction. We conclude that x ∈
and therefore ∃n ∈ N, n > x.

18
We now show that BW implies MCT: Let (an )n∈N be a bounded mono-
tonic sequence of real numbers. First of all, this sequence is Cauchy: We
assumed an ≤ M for all n ∈ N, but if (an )n∈N is not Cauchy, then

∃ε > 0, ∀N ∈ N, (∃n, m ≥ N, |an − am | ≥ ε)

def
Let N1 = N . Since an is increasing, we may suppose WOLOG n1 > m1 ≥
N1 , so that an1 ≥ am1 and therefore

an1 − am1 ≥ ε

Next, choose N2 > n1 , and find n2 > m2 ≥ N2 satisfying

an2 − am2 ≥ ε

Repeat inductively, and observe that we obtain a sequence of natural num-

bers
m1 < n1 < m2 < n2 · · ·
whose corresponding sequential terms form intervals Ik = (amk , ank ) of
length ≥ ε, ∀k ∈ N. But by AP, ∃k ∈ N so large that
M − am1
k>
ε
and for this k we have

ank − am1 ≥ k lengths of the intervals Ik

≥ kε
M − am1
> ·ε
ε
= M − am1

which shows ank > M , a contradiction. We conclude that (an )n∈N is

Cauchy. The decreasing case is handled similarly, but using −M instead,
and finding −M ≤ ank < M .
Secondly, since (an )n∈N is assumed bounded, our BW assumption guaran-
tees the existence of a convergent subsequence, ank → a. Proposition 10
then tells us that an → a, too, which shows that BW implies MCT. ■

19
 (xv) (4) =⇒ (3), or BW =⇒ NIP + AP: Suppose BW (every bounded sequence
of real numbers has a convergent subsequence) and let us deduce NIP (every
nested sequence of closed intervals of real numbers has nonempty intersection)
and AP (∀x ∈ R, ∃n ∈ N, x < n, and ∀x > 0, ∃n ∈ N, 0 < n1 < x).

Consider a nested sequence of closed intervals I1 ⊇ I2 ⊇ · · · , where In =

T
[an , bn ], and let us use CC + AP to show that n∈N In ̸= ∅. Then a1 ≤
a2 ≤ · · · ≤ an ≤ · · · ≤ bn ≤ · · · ≤ b2 ≤ b1 , so (an )n∈N is increasing
and bounded, (bn )n∈N is decreasing and bounded. Exactly as in the proof
of (4) =⇒ (2), we conclude that (an )n∈N and (bn )n∈N are Cauchy, and
therefore converge, an → a and bn → b, with a ≤ b by the order limit laws.
T
Since an ≤ a ≤ b ≤ bn for all n, we conclude that a, b ∈ n∈N —possibly
a = b—which is therefore nonempty. That BW implies AP was proved in
the course of showing (4) =⇒ (2). ■

(xvi) (4) =⇒ (5), or BW =⇒ CC: Suppose BW (every bounded sequence of real

numbers has a convergent subsequence) and let us deduce CC (a sequence of
real numbers converges iff it is Cauchy).

Any convergent sequence is Cauchy (Proposition 8), so we need only show

that a Cauchy sequence converges. Suppose (an )n∈N is a Cauchy sequence
in R. By Proposition 9 it is bounded, so by our BW assumption it has
a convergent subsequence, ank → a. Proposition 10 then guarantees that
an → a, too. ■

(xvii) (5) =⇒ (1), or CC + AP =⇒ AoC, is Theorem 2.6.4 in Abbott: Suppose

CC (a sequence of real numbers converges iff it is Cauchy) and AP (∀x ∈ R, ∃n ∈
N, x < n, and ∀x > 0, ∃n ∈ N, 0 < n1 < x) and let us deduce AoC (∅ ̸= A ⊆ R
with Au ̸= ∅ =⇒ s = sup A = min Au exists).

Let ∅ ̸= A ⊆ R and let Au ̸= ∅. We use the same algorithm as in the

proof of (4) =⇒ (1) above, which constructs Cauchy sequences (an )n∈N in
A and (bn )n∈N in Au , satisfying

a1 ≤ · · · an ≤ an+1 ≤ bn+1 ≤ bn ≤ · · · ≤ b1

for all n ∈ N. We will not repeat the details here. By our CC assumption,
these Cauchy sequences must converge, say an → a, bn → b. By the order
limit laws, a ≤ b. We claim that in fact a = b = sup A. An application of
AP as in the proof of (4) =⇒ (1) shows that |an − bn | < b12−a
n
1
< ε, which
shows limn→∞ |an − bn | = 0. Since an ≤ a ≤ b ≤ bn , the limit laws say
0 ≤ |a − b| ≤ limn→∞ |an − bn | = 0, which means |a − b| = 0, and therefore
a = b. To see that a = b = sup A, use Proposition 3: Let ε > 0, and use the
definition of convergence to get that ∃N ∈ N, (n ≥ N =⇒ a − an < ε),
which implies that a − ε < an . Since an ∈ A, this shows that a = b = s =
sup A. ■

20
(xviii) (5) =⇒ (2), or CC + AP =⇒ MCT: Suppose CC (a sequence of real
numbers converges iff it is Cauchy) and AP (∀x ∈ R, ∃n ∈ N, x < n, and
∀x > 0, ∃n ∈ N, 0 < n1 < x) and let us deduce MCT (every bounded monotone
sequence of real numbers converges).

Let (an )n∈N be a bounded (say by M ) monotone sequence in R. The same

proof as in (4) =⇒ (2) shows that (an )n∈N is Cauchy. Our CC assumption
will imply that it converges, say to a ∈ R. ■

(xix) (5) =⇒ (3), or CC + AP =⇒ NIP + AP: Suppose CC (a sequence of real

numbers converges iff it is Cauchy) and AP (∀x ∈ R, ∃n ∈ N, x < n, and
∀x > 0, ∃n ∈ N, 0 < n1 < x) and let us deduce NIP (every nested sequence of
closed intervals of real numbers has nonempty intersection), naturally + AP.

Consider a nested sequence of closed intervals I1 ⊇ I2 ⊇ · · · , where In =

T
[an , bn ], and let us use CC + AP to show that n∈N In ̸= ∅. Then a1 ≤
a2 ≤ · · · ≤ an ≤ · · · ≤ bn ≤ · · · ≤ b2 ≤ b1 , so (an )n∈N is increasing and
bounded, (bn )n∈N is decreasing and bounded. As in the proof of (5) =⇒
(2), or CC + AP implies MCT, we conclude that (an )n∈N and (bn )n∈N
are Cauchy, and therefore converge, an → a and bn → b, with a ≤ b by
the order limit laws. Since an ≤ a ≤ b ≤ bn for all n, we conclude that
T
a, b ∈ n∈N —possibly a = b—which is therefore nonempty. ■

(xx) (5) =⇒ (4), or CC + AP =⇒ BW, is Exercise 2.6.7(b) in Abbott: Suppose

CC (a sequence of real numbers converges iff it is Cauchy) and AP (∀x ∈
R, ∃n ∈ N, x < n, and ∀x > 0, ∃n ∈ N, 0 < n1 < x) and let us deduce BW
(every bounded sequence of real numbers has a convergent subsequence).

Let (an )n∈N be a bounded sequence of real numbers. By Lemma 7, (an )n∈N
has a (necessarily bounded) monotonic subsequence (ank )k∈N , whether in-
creasing or decreasing. We could now repeat the proof of (5) =⇒ (2), or
CC + AP implies MCT to show that this subsequence is Cauchy, and hence
converges by our CC + AP assumptions, or we can merely cite the fact
that CC + AP implies MCT which implies that ank → a. ■

21