You are on page 1of 4

Intro to Cauchy Sequences

Sample Assignment: Cauchy Sequences


by Maxim Beketov

In this passage we’ll introduce what’s called Cauchy sequences and will see why these are
important and helpful when studying sequence convergence. There is evidence that such
sequences were first introduced by Bolzano (1781 – 1848), who, yet lacking a rigorous
definition of real numbers, tried to prove their convergence. A correct proof was then provided
by Cauchy (1789 – 1857) (hence the name), who presumed the principle of nested intervals to
obviously hold true, which was later proved by Cantor (1845 – 1918).

Definition: A sequence {𝑠𝑛 } is called a Cauchy sequence, if for any 𝜀 > 0 there exists a number
𝑁, such that for any 𝑚, 𝑛 > 𝑁 (so 𝑠𝑚 and 𝑠𝑛 are coming later in the sequence than 𝑠𝑁 ) holds

|𝑠𝑚 − 𝑠𝑛 | < 𝜀

That is, for an arbitrarily small distance 𝜀, there is a certain moment after which any two
elements of the sequence are less apart than that 𝜀.

Fig. 1: For a given 𝜀, after a certain number 𝑁, the elements


are never more apart than 𝜀.

Cauchy sequences are important because the following statement is true:

Theorem: A number sequence converges iff (if and only if) it is a Cauchy sequence.

Let us first try to apply it to a couple examples, and then attempt to prove it (the way Cauchy
himself did).

1 1 1
Example 1: Consider a sequence 𝑠𝑛 = 1 + 2 + 3 + ⋯ + 𝑛 . Does it converge?

Solution: Since our theorem works in both ways (iff), if 𝑠𝑛 is not a Cauchy sequence – then it
does not converge. So the negation of Cauchy’s condition is:
There exists such an 𝜀 > 0 such that for any 𝑁, there are such two elements 𝑠𝑚 , 𝑠𝑛 coming
after 𝑁 (𝑚, 𝑛 > 𝑁) that |𝑠𝑚 − 𝑠𝑛 | ≥ 𝜀.
Though it seems that each next addition to 𝑠𝑛 is getting arbitrarily small, one can construct such
a counterexample – consider 𝑠𝑛 and 𝑠2𝑛 , then for any 𝑛

1 1 1
|𝑠2𝑛 − 𝑠𝑛 | = + + ⋯+
𝑛+1 𝑛+2 𝑛+𝑛
1 1
where each term is larger or equal to 𝑛+𝑛 = 2𝑛, and there are 𝑛 terms in total, so

1 1
|𝑠2𝑛 − 𝑠𝑛 | ≥ 𝑛 ⋅ =
2𝑛 2

the sequence is not a Cauchy sequence, and thus does not converge.

Example 2: “Longer bit strings”. Consider the following sequence of binary fractions:

𝑠1 = 0, 𝑠2 = 0. 𝑏1 , 𝑠3 = 0. 𝑏1 𝑏2 , …

where each next element is obtained by adding a bit – either 0 or 1 to the previous one (after
the binary point). Does this sequence converge somewhere?

Solution: Let us show that no matter the bits’ values, such a sequence would always converge!
(The limit itself depends on the values, of course).
Consider two elements, 𝑠𝑛 and 𝑠𝑚 coming after a certain 𝑠𝑁 and let’s estimate their difference
(let also 𝑚 > 𝑛)

𝑏𝑛+1 𝑏𝑚 1 1
|𝑠𝑚 − 𝑠𝑛 | = | 𝑛+1
+ ⋯ + 𝑚 | ≤ 𝑛+1 + ⋯ + 𝑚
2 2 2 2
1
which is a sum of (𝑚 − 𝑛) terms of a geometric with 2−(𝑛+1) being its first term and 2 – its
common ratio, so one obtains

1 𝑛+1 1 𝑚+1
(2) − (2) 1
|𝑠𝑚 − 𝑠𝑛 | ≤ <
1 2𝑛
1−2

1 𝑛
where the latter is obtained by taking (2) out as a common multiple in the numerator.
Thus for any given 𝜀 > 0 one can find such 𝑁 that for 𝑚 > 𝑛 > 𝑁

1 1
|𝑠𝑚 − 𝑠𝑛 | < 𝑛
< 𝑁<𝜀
2 2

and so 𝑠𝑛 is a Cauchy sequence.

Example 3: “The sign-changing sequence”. Using Cauchy’s argument, show that 𝑠𝑛 = (−1)𝑛
does not converge anywhere.

Solution: Indeed, for any two consecutive elements

|𝑠𝑛 − 𝑠𝑛+1 | = 2

so for a given 0 < 𝜀 < 2 it’s impossible to find such a moment after which any two elements of
the sequence will differ by less than 𝜀.
One can attempt to get a bit deeper and try to prove the above theorem. To do so, the
following fundamental property of the real line, ℝ, (the one that Cauchy presumed obvious in
his proof) is very helpful.

Lemma: Nested intervals principle


Consider a sequence of intervals (sets containing all such numbers 𝑥 that 𝑎 ≤ 𝑥 ≤ 𝑏),

[𝑎1 , 𝑏1 ] ⊃ [𝑎2 , 𝑏2 ] ⊃ ⋯ ⊃ [𝑎𝑛 , 𝑏𝑛 ] ⊃ ⋯

such that each next interval is a subset of the previous one.

Fig. 2, Three nested intervals on the real line

The lemma then is that there exists a point (at least one) that belongs to all the intervals of the
sequence.
If, additionally, the intervals’ lengths (𝑙𝑛 = 𝑏𝑛 − 𝑎𝑛 ) tend to zero as 𝑛 → ∞

lim 𝑙𝑛 = 0
𝑛→∞

then there is only one such common point.

Notice, however, that this does not work with open intervals, (𝑎𝑛 , 𝑏𝑛 ) – such that don’t contain
the boundary points. Can you think of such a counterexample?

1
Answer: For example, a sequence of (0, 𝑛) such open intervals has no common point.

Proof of the theorem:


The theorem’s statement holds in both ways (iff). As usual, proving the ‘if’ part (also called
‘necessary condition’) is a bit easier.

If a sequence converges (has a limit, denote it lim 𝑠𝑛 = 𝑆), then it’s a Cauchy sequence.
𝑛→∞

𝜀
Proof: Given 𝜀 > 0, let’s find such a number 𝑁 so that for any 𝑛 > 𝑁 holds |𝑠𝑛 − 𝑆| < 2 (this is
the definition of a sequence having a limit). Then for 𝑚, 𝑛 > 𝑁

𝜀 𝜀
|𝑠𝑚 − 𝑠𝑛 | < |𝑠𝑚 − 𝑆| + |𝑠𝑛 − 𝑆| < + =𝜀
2 2

(first step is triangle inequality).


Now let’s prove that if {𝑠𝑛 } is a Cauchy sequence – then it converges.

Provided with 𝜀 > 0, let’s find such 𝑁 that for 𝑚, 𝑘 ≥ 𝑁 holds

𝜀
|𝑠𝑚 − 𝑠𝑘 | <
3

Having fixed 𝑚 = 𝑁 it follows that for any 𝑘 > 𝑁

𝜀 𝜀
𝑠𝑁 − < 𝑠𝑘 < 𝑠𝑁 + (1)
3 3

so {𝑠𝑛 } is a bounded sequence (its terms never leaving a certain interval after a certain
moment).
Denote 𝑎𝑛 = inf𝑘≥𝑛 𝑠𝑘 and 𝑏𝑛 = sup𝑘≥𝑛 𝑠𝑘 – the infimum and supremum of the sequence
“tail”, starting at arbitrary 𝑛.
It is clear that

𝑎𝑛 ≤ 𝑎𝑛+1 ≤ 𝑏𝑛+1 ≤ 𝑏𝑛

these are not getting further apart as the tail is getting “shorter”. That means, a sequence of
intervals [𝑎𝑛 , 𝑏𝑛 ] is nested and thus has at least one common point, denote it 𝑆.
This means that for any 𝑛

𝑎𝑛 ≤ 𝑆 ≤ 𝑏𝑛

which together with obvious 𝑎𝑛 ≤ 𝑠𝑘 ≤ 𝑏𝑛 , where 𝑘 ≥ 𝑛, gives

|𝑠𝑘 − 𝑆| ≤ 𝑏𝑛 − 𝑎𝑛 (2)

But it follows from (1) that for any 𝑛 > 𝑁

𝜀 𝜀
𝑠𝑁 − ≤ 𝑎𝑛 ≤ 𝑏𝑛 ≤ 𝑠𝑁 +
3 3
2𝜀
and thus 𝑏𝑛 − 𝑎𝑛 ≤ < 𝜀, so from (2) now follows that for any 𝑘 ≥ 𝑁
3

|𝑠𝑘 − 𝑆| < 𝜀

so 𝑠𝑛 converges and 𝑆 is its limit. Q.E.D.

Materials used:
1. Zorich, Vladimir A., and R. Cooke. "Mathematical Analysis I. 2004."
2. Figures made in Geogebra software, geogebra.org

You might also like