This section works through several proofs that use the definition of the limit directly. With some study and practice you should be able to write proofs like these yourself.
Each example has a discussion, which is the scratch work you would do on paper, followed by the formal proof you would hand in.
Intuitionthe shape of every ε–N proof
Every proof that has the same skeleton:
Let . Let . Then implies , so .
The scratch work is a search for what to put in the blank, and it usually runs backward from the goal. The formal proof runs forward and only has to check that the chosen works. A reader should never have to see how you found it.
Example 1.
Prove .
Discussion.
We must take an arbitrary and produce an , depending on , such that implies .
So the formal proof should begin “Let ” and end with “Hence implies .” In between, it must name and check that it works.
To find , we work backward from what we want, as you might with a trigonometric identity, making sure every step is reversible. We want .
Multiplying by and dividing by , this is , or . This suggests .
Formal Proof
Let and let . Then implies , which implies and hence .
Thus implies . This proves .
∎
Example 2.
Prove .
Discussion.
For each we must decide how big must be to guarantee . Combining the fractions, we want
These steps are reversible, so we put . Any larger would also work.
Formal Proof
Let and let . Then implies , hence , hence ,
hence , and hence . This proves the claim.
∎
This is the that the ε–N widget in §7 uses for this sequence.
Example 3.
Prove .
Discussion.
We must find how large has to be to make
For we can drop the absolute values. This time it would be very hard to solve for . But we do not need the least, only some, so we simplify with estimates.
The idea is that is at most a constant times for large . To get such a bound we make the numerator bigger and the denominator smaller.
Since , it suffices to get . Next, once , which holds for .
So for it suffices to get , i.e. , i.e. .
Formal Proof
Let and let . Then implies , hence , hence .
Since , we also have , and . So implies
∎
Intuitionhow to estimate
You want , but is too messy to solve for . Find a simpler with for all large , and make instead. To make a fraction bigger, increase the numerator or decrease the denominator, never the reverse.
Each simplification usually needs to be large enough, like above. Collect those requirements and build them into with a .
Common pitfall
When shrinking a denominator, keep it positive. is useful only because both sides are positive once .
Backward scratch work is safe only if each step can be reversed. If one cannot, check the forward direction separately in the formal proof.
Every assumption about has to be justified by . If your estimate needs , the formal proof needs .
As Example 3 shows, direct proofs of even simple limits can get messy. With the limit theorems of §9 we will just write
Assume and look for a contradiction. Whatever is, at least one of and is at distance at least from . So cannot hold for all large .
Formal Proof
Assume for some . Taking in the definition of limit, there is such that implies .
Using an even and an odd , we get and . By the Triangle Inequality 3.7,
This absurdity shows the assumption was wrong, so does not converge.
∎
Intuitionproving divergence
To show , negate Definition 7.1: find one such that, however large is, some has . Example 4 does this with for every candidate at once. A single bad tolerance is enough.
Example 5.
Let be a sequence of nonnegative real numbers with . Note (Exercise 8.9(a)). Prove .
Discussion.
This time we cannot expect a formula for in terms of , because the problem is so general. But we can show some exists.
The trick is to break with algebra training and “irrationalize the denominator”:
Since we can make the numerator small. If the denominator is at least , so ,
and we choose so that for . Such an exists because the definition of limit can be applied to just as well as to .
If the denominator can be small too, but then a direct argument works (Exercise 8.3).
Formal Proof
Case I: . Let . Since , there is such that implies . Then implies
Let be a convergent sequence with for all and . Prove .
Discussion.
“Most” terms are close to and hence not close to . More precisely, most are within of , so most satisfy . The finitely many remaining terms are nonzero, so they cannot drag the infimum down to .
Formal Proof
Let . Since , there is such that implies . Then
since otherwise the Triangle Inequality would give , which is absurd.
Put . Then and, by (1), for all . So .
∎
Reading alongside Abbott
Abbott’s Section 2.2 works through similar examples and stresses the same two stages: rough work to find , then a clean forward proof that starts with “Let ”.
This is the classic slip in turning scratch work into a proof.
Spot the error: the limit of 1/n
One line of this argument is wrong. Press the button on the line you think is the mistake.
Claim: . Let .
Suppose that .This assumes the conclusion. The inequality |1/n − 0| < ε is what we must show for n > N, so it cannot be a starting point. Scratch work may begin there to discover N, but the proof must begin with n > N and end with the inequality.
Imitate Example 4. (a) The values (at ) and (at ) recur forever, so take . (b) A convergent sequence is bounded (Theorem 9.1), or argue directly. (c) and recur forever; take .
Solution
For the term is and for it is ; both values occur for arbitrarily large . Suppose the limit were . Taking , for all large we’d have , and choosing large with and large with gives , a contradiction.
is unbounded, but convergent sequences are bounded (Theorem 9.1). So does not converge.
For the term is , and for it is ; both recur forever. If , take : large and with these values give , a contradiction.