2.4 Convergence: the Ξ΅\varepsilon-NN definition of a limit

Earlier, we considered the sequence (0,0.3,0.33,0.333,…)(0,0.3,0.33,0.333,\dots) coming from the decimal expansion of 1/31/3 and informally observed that the terms get β€˜closer and closer’ to 1/31/3. Our goal here is to make the idea of β€˜closer and closer’ precise. To do this, we introduce the concept of a limit. This turns to one of the most important definitions not only in this course, but in the whole mathematics degree!

Example 2.22 (Informal).

Before giving the formal definition of a limit, let’s try to motivate some of the key ideas by considering another (simpler) example. Recall the graph of the terms of the sequence (1/n)nβˆˆβ„•(1/n)_{n\in\mathbb{N}} from FigureΒ 2.1, which is also reproduced in FigureΒ 2.6. The graph has a horizontal asymptote around the line y=0y=0, which reflects the fact that the terms 1/n1/n get closer and closer to 0 as nn gets large. Intuitively, we’d like to say the limit of the sequence is 0, but what do we really mean by this?

Let’s try to be a bit more concrete. After the 100100th term, we will always have 0≀1/n<0.010\leq 1/n<0.01, so the terms are then within 0.010.01 of the limit 0. After the 10001000th term we will always have 0≀1/n<0.0010\leq 1/n<0.001, so the terms are then within 0.0010.001 of the limit 0. In fact, if we prescribe any measure of β€˜closeness’ Ξ΅>0\varepsilon>0 (which could be 0.010.01, 0.0010.001, or even 0.000000000000010.00000000000001), then after the ⌈1/Ξ΅βŒ‰\lceil 1/\varepsilon\rceil-term,22 2 Here ⌈xβŒ‰:=min⁑{kβˆˆβ„€:kβ‰₯x}\lceil x\rceil:=\min\{k\in\mathbb{Z}:k\geq x\} denotes the ceiling function; in other words, it is the number formed by rounding xx up to the nearest integer. we will have 0≀1/n<Ξ΅0\leq 1/n<\varepsilon: see FigureΒ 2.6. Thus,

(2.1) (2.1) For any measure of closeness, the terms of the sequence are eventuallyclose to the value of the limit.\begin{gathered}\text{{For any measure of closeness, the terms of the sequence% are eventually}}\\ \text{{close to the value of the limit.}}\end{gathered}
Figure 2.6: An illustration of the Ξ΅\varepsilon-NN definition of a limit for the sequence (1/n)nβˆˆβ„•(1/n)_{n\in\mathbb{N}}. Given Ξ΅>0\varepsilon>0, we can find some Nβˆˆβ„•N\in\mathbb{N} such that all terms of the sequence which come after the NNth term lie within Ξ΅\varepsilon of 0.

The purpose of the Ξ΅\varepsilon-NN definition of a limit is to make the informal idea (2.1) precise.

Definition 2.23 (Ξ΅\varepsilon-NN definition of a limit).
  1. 1
    ​

    A sequence (an)nβˆˆβ„•(a_{n})_{n\in\mathbb{N}} is said to converge to a real number aβˆˆβ„a\in\mathbb{R} if for all Ξ΅>0\varepsilon>0 there exists some Nβˆˆβ„•N\in\mathbb{N} (which in general depends on Ξ΅\varepsilon) such that

    (2.2) (2.2) |anβˆ’a|<Ξ΅for allΒ nβˆˆβ„•Β withΒ n>N.|a_{n}-a|<\varepsilon\quad\text{\text@underline{for all} $n\in\mathbb{N}$ with% $n>N$.}
  2. 2
    ​

    We write anβ†’aa_{n}\to a as nβ†’βˆžn\to\infty to denote that (an)nβˆˆβ„•(a_{n})_{n\in\mathbb{N}} converges to aa. The number aa is called the limit of the sequence (an)nβˆˆβ„•(a_{n})_{n\in\mathbb{N}} and is often written as limnβ†’βˆžan\displaystyle\lim_{n\to\infty}a_{n}.

  3. 3
    ​

    We say a sequence (an)nβˆˆβ„•(a_{n})_{n\in\mathbb{N}} converges if there exists some aβˆˆβ„a\in\mathbb{R} such that (an)nβˆˆβ„•(a_{n})_{n\in\mathbb{N}} converges to aa.

  4. 4
    ​

    A sequence that does not converge is said to diverge.

Warning 2.24.

The symbol ∞\infty appearing in β€˜nβ†’βˆžn\to\infty’ and β€˜limnβ†’βˆžan\lim_{n\to\infty}a_{n}’ is merely part of the notation! In particular, here we do not define β€˜βˆž\infty’ to be a number or any other mathematical object: it is just a symbol used to articulate the concept. Since ∞\infty is not a number, it does not make sense to manipulate it algebraically, so in this course you should never write or try to work with expressions such as 1+∞1+\infty, 2Γ—βˆž2\times\infty or 1/∞1/\infty.

This is a subtle definition, which takes work to master. One source of difficulty is the number of quantifiers: we have three! As always, the order in which the quantifiers appear is very important: we explore this in detail in ExerciseΒ 2.36 below.

Let’s compare the rigorous definition from DefinitionΒ 2.23 with our informal idea of a limit from (2.1). As hinted earlier, Ξ΅\varepsilon plays the role of the β€˜measure of closeness’. The parameter NN plays the role of β€˜eventually’ (that is, we only consider terms after some specified threshold). Finally, |anβˆ’a||a_{n}-a| tells us how close the term ana_{n} is from the target limit aa.

Informal idea (2.1) Ξ΅\varepsilon-NN definition (DefinitionΒ 2.23)
1
For any measure of closeness,
For all Ξ΅>0\varepsilon>0
2
the sequence is eventually
there exists some Nβˆˆβ„•N\in\mathbb{N}
3
close to the value of the limit.
such that |anβˆ’a|<Ξ΅|a_{n}-a|<\varepsilon for all nβˆˆβ„•n\in\mathbb{N} with n>Nn>N.

We can also interpret DefinitionΒ 2.23 in terms of our visual aids. Intuitively, the sequence (an)nβˆˆβ„•(a_{n})_{n\in\mathbb{N}} converges to aa as nβ†’βˆžn\to\infty if all the points on the graph eventually lie in a thin corridor around aa: see FigureΒ 2.7. We can make the corridor as thin as we like, provided we consider terms far enough along the sequence.

Figure 2.7: An illustration of the Ξ΅\varepsilon-NN definition of a limit, for a sequence converging to aa. Beyond the threshold NN, the terms of the sequence lie in a horizontal β€˜Ξ΅\varepsilon-corridor’ around the line y=ay=a.

To get to grips with these definitions, we begin by very slowly walking through a simple example.

Example 2.25.

Consider the sequence (1/n)nβˆˆβ„•(1/n)_{n\in\mathbb{N}} which was informally discussed in ExampleΒ 2.22. We expect that 1/nβ†’01/n\to 0 as nβ†’βˆžn\to\infty, but how do we prove 1/nβ†’01/n\to 0 as nβ†’βˆžn\to\infty working from the Ξ΅\varepsilon-NN definition?

Walk through. Before presenting the formal proof, we’ll walk through the thought processes involved. For this, we slowly and carefully consider the definition, one quantifier at a time. The first quantifier in DefinitionΒ 2.23 is for all Ξ΅>0\varepsilon>0. We therefore begin by fixing some arbitrary Ξ΅>0\varepsilon>0 to work with (that is, we prescribe a measure of closeness). A typical way to start a proof is:

1

Let Ξ΅>0\varepsilon>0 be given.

Now we have fixed Ξ΅>0\varepsilon>0, we move on to the next quantifier. We want to show there exists some Nβˆˆβ„•N\in\mathbb{N} such that (2.2) holds with an:=1/na_{n}:=1/n and a:=0a:=0. This is typically where we need to do the real work, but here most of what we need was already discussed in ExampleΒ 2.22. Let’s begin by writing out (2.2) explicitly for our choice of sequence and target limit. In particular, |anβˆ’a|=|1/nβˆ’0|=1/n|a_{n}-a|=|1/n-0|=1/n, and so (2.2) becomes

1n<Ξ΅for allΒ n>N.\frac{1}{n}<\varepsilon\qquad\text{for all $n>N$.}

In other words, we want to find sufficiently large Nβˆˆβ„•N\in\mathbb{N} such that all terms beyond the NNth term are tiny: they are smaller than the prescribed Ξ΅\varepsilon. How can we do this? Well, if we take N:=1/Ξ΅N:=1/\varepsilon, then whenever n>Nn>N we have 1/n<1/N=Ξ΅1/n<1/N=\varepsilon, so this choice would work. One small issue is that DefinitionΒ 2.23 asks that NN should be a natural number, and there is no guarantee that the choice 1/Ξ΅1/\varepsilon has this property.33 3 The condition that NN is a natural number is not an essential part of the definition but more a matter of convention. However, it is a commonly used convention, so we shall stick with it. However, this is easily remedied by taking N:=⌈1/Ξ΅βŒ‰N:=\lceil 1/\varepsilon\rceil, as we did in ExampleΒ 2.22.

Now we’ve worked out what our choice of NN should be, we can write the next line of the proof:

2

Choose N:=⌈1/Ξ΅βŒ‰βˆˆβ„•N:=\lceil 1/\varepsilon\rceil\in\mathbb{N}.

The next step is to verify that this choice of NN works, in the sense that the desired condition (2.2) holds for this choice. However, we chose NN precisely with this property in mind, so we can move swiftly on to the next line of the proof:

3

If nβˆˆβ„•n\in\mathbb{N} with n>Nn>N, then |1/nβˆ’0|=1/n<1/N≀Ρ|1/n-0|=1/n<1/N\leq\varepsilon.

Finally, it’s helpful to conclude our proof with a wrap up of what we’ve shown:

Thus, by the Ξ΅\varepsilon-NN definition of a limit, 1/nβ†’01/n\to 0 as nβ†’βˆžn\to\infty.

This concludes our walk through of this example. Now we can put the pieces together to come up with our formal proof.

Proof.

Let Ξ΅>0\varepsilon>0 be given and choose N:=⌈1/Ξ΅βŒ‰βˆˆβ„•N:=\lceil 1/\varepsilon\rceil\in\mathbb{N}. If nβˆˆβ„•n\in\mathbb{N} with n>Nn>N, then

|1/nβˆ’0|=1/n<1/N≀Ρ.|1/n-0|=1/n<1/N\leq\varepsilon.

Thus, by the Ξ΅\varepsilon-NN definition of a limit, 1/nβ†’01/n\to 0 as nβ†’βˆžn\to\infty. ∎

In the above example, given Ξ΅>0\varepsilon>0, it is a straightforward matter to find a suitable choice of NN to verify the Ξ΅\varepsilon-NN definition of a limit. This was because we had |anβˆ’a|=1/n|a_{n}-a|=1/n, which led us to see that N=⌈1/Ξ΅βŒ‰N=\lceil 1/\varepsilon\rceil was a good choice. However, in most cases, finding a suitable choice of NN is more challenging: typically, one needs to do some work at the start of the proof to find a suitable bound for |anβˆ’a||a_{n}-a| in terms of nn. The following example shows this in action.

Example 2.26.

We claim that 5⁒n2n2+3⁒nβ†’5\displaystyle\frac{5n^{2}}{n^{2}+3n}\to 5 as nβ†’βˆžn\to\infty.

For large values of nn, the term 3⁒n3n on the denominator is much, much smaller than the term n2n^{2}. Therefore, intuitively, n2+3⁒nn^{2}+3n is relatively close to n2n^{2} for large values of nn and, since 5⁒n2n2=5\frac{5n^{2}}{n^{2}}=5, we expect 5⁒n2n2+3⁒nβ†’5\frac{5n^{2}}{n^{2}+3n}\to 5 as nβ†’βˆžn\to\infty. We shall make this intuition precise using the Ξ΅\varepsilon-NN definition.

Proof.

For nβˆˆβ„•n\in\mathbb{N}, note that

(2.3) (2.3) |5⁒n2n2+3⁒nβˆ’5|=|5⁒n2βˆ’5⁒(n2+3⁒n)n2+3⁒n|=15⁒nn2+3⁒n<15⁒nn2=15n.\displaystyle\left|\frac{5n^{2}}{n^{2}+3n}-5\right|=\left|\frac{5n^{2}-5(n^{2}% +3n)}{n^{2}+3n}\right|=\frac{15n}{n^{2}+3n}<\frac{15n}{n^{2}}=\frac{15}{n}.

Let Ξ΅>0\varepsilon>0 be given and choose N:=⌈15/Ξ΅βŒ‰βˆˆβ„•N:=\lceil 15/\varepsilon\rceil\in\mathbb{N}. If nβˆˆβ„•n\in\mathbb{N} with n>Nn>N, then

|5⁒n2n2+3⁒nβˆ’5|<15n<15N≀Ρ.\Big{|}\frac{5n^{2}}{n^{2}+3n}-5\Big{|}<\frac{15}{n}<\frac{15}{N}\leq\varepsilon.

Thus, by the Ξ΅\varepsilon-NN definition of a limit, 5⁒n2n2+3⁒nβ†’5\frac{5n^{2}}{n^{2}+3n}\to 5 as nβ†’βˆžn\to\infty. ∎

Remarks 2.27.

We further discuss some of the features of ExampleΒ 2.25 and ExampleΒ 2.26.

  1. 1.
    ​

    Note how the form of the proofs in ExampleΒ 2.25 and ExampleΒ 2.26 align with the Ξ΅\varepsilon-NN definition of a limit from DefinitionΒ 2.23. For instance, taking ExampleΒ 2.26, we can see each part of the Ξ΅\varepsilon-NN definition has a corresponding part of the proof:

    Ξ΅\varepsilon-NN definition (DefinitionΒ 2.23) Proof from ExampleΒ 2.25
    For all Ξ΅>0\varepsilon>0 Let Ξ΅>0\varepsilon>0 be given
    there exists some Nβˆˆβ„•N\in\mathbb{N} choose N:=⌈15/Ξ΅βŒ‰βˆˆβ„•N:=\lceil 15/\varepsilon\rceil\in\mathbb{N}
    such that |anβˆ’a|<Ξ΅|a_{n}-a|<\varepsilon for all nβˆˆβ„•n\in\mathbb{N} with n>Nn>N.
    If nβˆˆβ„•n\in\mathbb{N} with n>Nn>N, then […]
    |5⁒n2n2+3⁒nβˆ’5|<Ξ΅\Big{|}\frac{5n^{2}}{n^{2}+3n}-5\Big{|}<\varepsilon.

    This way, the proof clearly verifies that the definition holds. Your Ξ΅\varepsilon-NN proofs should also mirror the definition in this way.

  2. 2.
    ​

    If you were asked in, say, an assignment to show 1/nβ†’01/n\to 0 or 5⁒n2n2+3⁒nβ†’5\frac{5n^{2}}{n^{2}+3n}\to 5 as nβ†’βˆžn\to\infty, then all that would be required would be the formal proof from either ExampleΒ 2.25 or ExampleΒ 2.26. The walk through and intuition and are included above to illustrate the thought processes which allow one to arrive at the proof. As with all proofs, you may need to do some separate β€œrough work” to figure out the right approach, before you write up the final, formal, proof.

  3. 3.
    ​

    We did something smart in (2.3). When working with the Ξ΅\varepsilon-NN definition, we are often only interested in inequalities rather than precise identities. We used this flexibility to pass from the complicated expression 15⁒nn2+3⁒n\tfrac{15n}{n^{2}+3n} to the simpler expression 15n\frac{15}{n} in (2.3), which made matters much more straightforward when it came to finding a suitable value of NN. It’s a good idea to carry out such simplifications wherever possible!

  4. 4.
    ​

    To verify the Ξ΅\varepsilon-NN definition, we just need to show there exists some NN and there are always many choices. For instance, we used N:=⌈15/Ξ΅βŒ‰N:=\lceil 15/\varepsilon\rceil in ExampleΒ 2.26, but there is nothing stopping us from using a larger value such as N:=⌈15/Ξ΅βŒ‰+100N:=\lceil 15/\varepsilon\rceil+100 or N:=⌈1000/Ξ΅βŒ‰N:=\lceil 1000/\varepsilon\rceil.

Exercise 2.28.

Arguing from the Ξ΅\varepsilon-NN definition of a limit, show that each of the following sequences converges and find the limit.

  1. (i)
    ​

    (3⁒n2⁒n+5)nβˆˆβ„•\displaystyle\Big{(}\frac{3n}{2n+5}\Big{)}_{n\in\mathbb{N}};

  2. (ii)
    ​

    (nn2+1)nβˆˆβ„•\displaystyle\Big{(}\frac{n}{n^{2}+1}\Big{)}_{n\in\mathbb{N}}.

A video walk-through of a solution to part (ii) can be found here.

Exercise 2.29.

Let aβˆˆβ„a\in\mathbb{R} and consider the constant sequence an:=aa_{n}:=a for all nβˆˆβ„•n\in\mathbb{N}. Thus, (an)nβˆˆβ„•=(a,a,a,β‹―)(a_{n})_{n\in\mathbb{N}}=(a,a,a,\cdots) is a rather boring sequence where all the terms have the same value. Arguing from the Ξ΅\varepsilon-NN definition of a limit, show that anβ†’aa_{n}\to a as nβ†’βˆžn\to\infty.

The following inequality from IMU is often useful when working with limits.

Lemma 2.30 (Bernoulli inequality).

For all Οβˆˆβ„\rho\in\mathbb{R} with ρβ‰₯0\rho\geq 0 and all nβˆˆβ„•n\in\mathbb{N}, we have

(1+ρ)nβ‰₯1+n⁒ρ.(1+\rho)^{n}\geq 1+n\rho.
Proof.

See IMU or Worksheet 1. ∎

Exercise 2.31.

This important exercise establishes some basic properties of geometric sequences. For 0<ρ<10<\rho<1 note that

1βˆ’Ο=1βˆ’Ο21+ρ.1-\rho=\frac{1-\rho^{2}}{1+\rho}.

Let 0<r<10<r<1. Using the above identity for ρ:=1βˆ’r\rho:=1-r and the Bernoulli inequality from LemmaΒ 2.30, show that rnβ†’0r^{n}\to 0 as nβ†’βˆžn\to\infty.

The Bernoulli inequality can also be used to prove the following useful limit identities.

Lemma 2.32 (Examples of limits).

The following limit identities hold:

  1. 1
    ​

    For all r>0r>0, we have r1/nβ†’1r^{1/n}\to 1 as nβ†’βˆžn\to\infty.

  2. 2
    ​

    n1/nβ†’1n^{1/n}\to 1 as nβ†’βˆžn\to\infty.

Proof.

See Worksheet 3 and also Exercise 2.73 below. ∎

Exercise 2.33.

Consider the sequence (an)nβˆˆβ„•=(0,0.9,0.99,0.999,…)(a_{n})_{n\in\mathbb{N}}=(0,0.9,0.99,0.999,\dots). Observe that an=1βˆ’10βˆ’(nβˆ’1)a_{n}=1-10^{-(n-1)} for nβˆˆβ„•n\in\mathbb{N} and use this to show anβ†’1a_{n}\to 1 as nβ†’βˆžn\to\infty. This is the rigorous interpretation of the decimal expansion 0.999⁒…=10.999\ldots=1.44 4 We shall study decimal expansions more systematically in the next section.

Example 2.34.

We claim that n+1βˆ’nβ†’0\sqrt{n+1}-\sqrt{n}\to 0 as nβ†’βˆžn\to\infty.

Intuitively, we expect n+1βˆ’nβ†’0\sqrt{n+1}-\sqrt{n}\to 0 because the graph of the square root function gets β€˜flatter and flatter’ as nn gets large: see FigureΒ 2.8. Thus, for large value of nn there is very little difference between n+1\sqrt{n+1} and n\sqrt{n}. We shall make this intuition precise using the Ξ΅\varepsilon-NN definition.

In the proof, similarly to ExerciseΒ 2.31, we use the factorisation x2βˆ’y2=(xβˆ’y)⁒(x+y)x^{2}-y^{2}=(x-y)(x+y) for difference of squares to rewrite

xβˆ’y=x2βˆ’y2x+yforΒ x=n+1Β andΒ y=n.x-y=\frac{x^{2}-y^{2}}{x+y}\qquad\text{for $x=\sqrt{n+1}$ and $y=\sqrt{n}$.}

It seems like this should give a more complicated expression, but in this case it becomes simpler because for x=n+1x=\sqrt{n+1} and y=ny=\sqrt{n} we have x2βˆ’y2=1x^{2}-y^{2}=1. This is a bit of a trick, so don’t worry if you didn’t spot it – at least now you know!

Proof.

For nβˆˆβ„•n\in\mathbb{N}, note that

(n+1βˆ’n)⁒(n+1+n)=(n+1)2βˆ’(n)2=(n+1)βˆ’n=1.(\sqrt{n+1}-\sqrt{n})(\sqrt{n+1}+\sqrt{n})=(\sqrt{n+1})^{2}-(\sqrt{n})^{2}=(n+% 1)-n=1.

Thus,

n+1βˆ’nβˆ’0=1n+1+n<1n.\sqrt{n+1}-\sqrt{n}-0=\frac{1}{\sqrt{n+1}+\sqrt{n}}<\frac{1}{\sqrt{n}}.

Let Ξ΅>0\varepsilon>0 be given and choose N:=⌈1/Ξ΅2βŒ‰βˆˆβ„•N:=\lceil 1/\varepsilon^{2}\rceil\in\mathbb{N}. If nβˆˆβ„•n\in\mathbb{N} with n>Nn>N, then

|n+1βˆ’nβˆ’0|=1n+1+n<1n<1N≀Ρ.|\sqrt{n+1}-\sqrt{n}-0|=\frac{1}{\sqrt{n+1}+\sqrt{n}}<\frac{1}{\sqrt{n}}<\frac% {1}{\sqrt{N}}\leq\varepsilon.

Thus, by the Ξ΅\varepsilon-NN definition of a limit, n+1βˆ’nβ†’0\sqrt{n+1}-\sqrt{n}\to 0 as nβ†’βˆžn\to\infty. ∎

Figure 2.8: The graph of x\sqrt{x} gets β€˜flatter and flatter’ as xx increases. Thus, for large values of nn, the difference n+1βˆ’n\sqrt{n+1}-\sqrt{n} is very small.
Exercise 2.35.

Show that (n2+5)1/2βˆ’nβ†’0(n^{2}+5)^{1/2}-n\to 0 as nβ†’βˆžn\to\infty.

As always, the order and form of the quantifiers in the Ξ΅\varepsilon-NN definition of a limit is very important!

Exercise 2.36 (Quantifiers).

Let (an)nβˆˆβ„•(a_{n})_{n\in\mathbb{N}} be a sequence and aβˆˆβ„a\in\mathbb{R}. Match each statement in Group A with an equivalent statement from Group B.

Group A:

  1. (i)
    ​

    For all Ξ΅>0\varepsilon>0 there exists some Nβˆˆβ„•N\in\mathbb{N} such that |anβˆ’a|<Ξ΅|a_{n}-a|<\varepsilon for all n>Nn>N.

  2. (ii)
    ​

    For all Ξ΅>0\varepsilon>0, for all Nβˆˆβ„•N\in\mathbb{N}, we have |anβˆ’a|<Ξ΅|a_{n}-a|<\varepsilon for all nβ‰₯Nn\geq N.

  3. (iii)
    ​

    There exists some Ξ΅>0\varepsilon>0 such that for all Nβˆˆβ„•N\in\mathbb{N} we have |anβˆ’a|<Ξ΅|a_{n}-a|<\varepsilon for all nβ‰₯Nn\geq N.

  4. (iv)
    ​

    There exists some Ξ΅>0\varepsilon>0 and there exists some Nβˆˆβ„•N\in\mathbb{N} such that |anβˆ’a|<Ξ΅|a_{n}-a|<\varepsilon for all n>Nn>N.

  5. (v)
    ​

    For all Nβˆˆβ„•N\in\mathbb{N} there exists some Ξ΅>0\varepsilon>0 such that |anβˆ’a|<Ξ΅|a_{n}-a|<\varepsilon for all nβ‰₯Nn\geq N.

  6. (vi)
    ​

    There exists some Nβˆˆβ„•N\in\mathbb{N} such that for all Ξ΅>0\varepsilon>0 we have |anβˆ’a|<Ξ΅|a_{n}-a|<\varepsilon for all n>Nn>N.

Group B:

  1. (a)
    ​

    anβ†’aa_{n}\to a as nβ†’βˆžn\to\infty.

  2. (b)
    ​

    (an)nβˆˆβ„•(a_{n})_{n\in\mathbb{N}} is bounded.

  3. (c)
    ​

    (an)nβˆˆβ„•(a_{n})_{n\in\mathbb{N}} is constant with an=aa_{n}=a for all nβˆˆβ„•n\in\mathbb{N}.

  4. (d)
    ​

    The sequence (an)nβˆˆβ„•(a_{n})_{n\in\mathbb{N}} is eventually constant: in particular, there exists some Nβˆˆβ„•N\in\mathbb{N} such that a=aN+1=aN+2=aN+3=β‹―a=a_{N+1}=a_{N+2}=a_{N+3}=\cdots.

There is one loose end we need to deal with. In DefinitionΒ 2.23 we talked about the (definite article) limit of a sequence, which implicitly assumes limits are unique. This requires proof!

Lemma 2.37 (Uniqueness of limits).

If (an)nβˆˆβ„•(a_{n})_{n\in\mathbb{N}} is a convergent sequence of real numbers and bb, cβˆˆβ„c\in\mathbb{R} are both limits of (an)nβˆˆβ„•(a_{n})_{n\in\mathbb{N}}, then b=cb=c.

Proof.

By the Ξ΅\varepsilon-NN definition of a limit applied to both bb and cc, given Ξ΅>0\varepsilon>0, there exists N1N_{1} and N2βˆˆβ„•N_{2}\in\mathbb{N} such that

|anβˆ’b|<Ξ΅/2for allΒ n>N1and|anβˆ’c|<Ξ΅/2for allΒ n>N2.|a_{n}-b|<\varepsilon/2\quad\text{for all $n>N_{1}$}\qquad\text{and}\qquad|a_{% n}-c|<\varepsilon/2\quad\text{for all $n>N_{2}$.}

In particular, if we take nβˆˆβ„•n\in\mathbb{N} with n>max⁑{N1,N2}n>\max\{N_{1},N_{2}\}, then, by the triangle inequality,

|bβˆ’c|≀|bβˆ’an|+|anβˆ’c|<Ξ΅/2+Ξ΅/2=Ξ΅.|b-c|\leq|b-a_{n}|+|a_{n}-c|<\varepsilon/2+\varepsilon/2=\varepsilon.

Since the above inequality holds for all Ξ΅>0\varepsilon>0, we must have |bβˆ’c|=0|b-c|=0 and so b=cb=c, as required. ∎

The proof of LemmaΒ 2.37 involves an β€˜abstract’ Ξ΅\varepsilon-NN argument. Previously, we have focused on verifying the Ξ΅\varepsilon-NN definition for explicit sequences, such as (1/n)nβˆˆβ„•(1/n)_{n\in\mathbb{N}} or (n+1βˆ’n)nβˆˆβ„•\big{(}\sqrt{n+1}-\sqrt{n}\big{)}_{n\in\mathbb{N}}. In LemmaΒ 2.37, the goal is quite different. Here we are told in advance that some β€˜abstract’ sequence (an)nβˆˆβ„•(a_{n})_{n\in\mathbb{N}} converges; in particular, we know the Ξ΅\varepsilon-NN definition holds for this sequence. The proof works by using the Ξ΅\varepsilon-NN definition to derive some new information (in this case, that the limit of the sequence is unique).

The ability to argue abstractly with the Ξ΅\varepsilon-NN definition makes it very powerful. It allows us to prove general statements about sequences, rather than studying them on a case-by-case basis. In this way, we can develop a general theory of convergent sequences. We shall explore this theory in the following sections.