Chapter IV · Limits of functions of a positive integral variable

“For large values of n”

§§53–57 35 min

What it means for a property to hold for all sufficiently large n, how this differs from holding for infinitely many n, and how it gives meaning to “n tends to infinity”.

Why this matters

The definition of a limit in the next lesson is one sentence long, but it is built from a phrase that does all the work: “for sufficiently large values of nn”. §§53–57 builds that phrase carefully. It sorts every property of a sequence into three cases, explains why “large” by itself means nothing, gives the only meaning that ”nn tends to ∞\infty” will ever have, and turns ”1/n1/n is small when nn is large” into a precise statement that the definition of a limit then copies.

The idea

Take a sequence ϕ(n)\phi(n) and a property PP that each value may or may not have, such as "ϕ(n)<0.001\phi(n) < 0.001" or ”ϕ(n)\phi(n) is odd”. Go through n=1,2,3,…n = 1, 2, 3, \dots and mark each nn as having PP or not. Because there are infinitely many nn, at least one of the two classes is infinite (last lesson), and exactly three things can happen.

  • (a) Only finitely many nn fail to have PP (perhaps none).
  • (b) Only finitely many nn have PP (perhaps none).
  • (c) Neither: infinitely many nn have PP and infinitely many do not.

Deciding which case you are in is a skill, and the first few terms are often misleading. Hardy’s eleven examples are all in the explorer.

Explore Classify Hardy’s eleven examples

Try this. For each example, look at the window starting at n=1n = 1, then the windows further out, and only then choose (a), (b) or (c). Pay special attention to (vii), (x) and (xi).

(i) ϕ(n)=n\phi(n) = n, property P:ϕ(n) is a perfect squareP: \phi(n) \text{ is a perfect square}
12345678910111213141516171819202122232425262728293031323334353637383940
  • φ(n) has P
  • φ(n) does not have P
Look at n from
Your verdict
6 of the 40 values of n shown have the property. Decide: (a) all but finitely many nn have PP; (b) only finitely many nn have PP; (c) neither.
What to notice

In (vii) the first 4040 values look like case (c), half good and half bad, but beyond n=2000n = 2000 every value is good: case (a). The exceptions need not be the first values of nn, but in case (a) there is always a last one. In (i) and (v) no window, however far out, is all one colour: case (c).

In case (a) the exceptions form a finite set of positive integers, so they have a largest member nNn_N, and every n>nNn > n_N has the property. Hardy says that ϕ(n)\phi(n) has the property for large values of nn, or for all sufficiently large nn. “Large” here is relative: n>1000n > 1000 is large enough for 1/n<0.0011/n < 0.001 but not for 1/n<0.00011/n < 0.0001. As Hardy remarks, six goals are a large score in football and six runs are not in cricket. In mathematics “large” always means large enough for the purpose in hand.

Hardy then offers a second picture (§55). Let nn run through 1,2,3,…1, 2, 3, \dots, one value per second, say. It grows beyond every bound, and we say nn tends to infinity, n→∞n \to \infty. If ϕ(n)\phi(n) has a property for large nn, then as nn tends to infinity it eventually has the property and keeps it. So “what does ϕ(n)\phi(n) do for large nn?” and “how does ϕ(n)\phi(n) behave as n→∞n \to \infty?” are the same question. The picture of time is only a convenience; nothing actually moves.

Precise statements

Definition For large values of nn Hardy: §54

A sequence ϕ(n)\phi(n) has a property PP for large values of nn (or for all sufficiently large nn) if there is a number n0n_0 such that ϕ(n)\phi(n) has PP for every n≥n0n \ge n_0.

Theorem Finitely many exceptions means: from some point on Hardy: §54

ϕ(n)\phi(n) has PP for large values of nn if and only if case (a) holds, that is, the nn for which ϕ(n)\phi(n) lacks PP form a finite class.

Two further facts are used constantly from now on, usually without comment.

Theorem Two properties at once

If ϕ(n)\phi(n) has PP for large nn and has QQ for large nn, then it has both PP and QQ for large nn.

Theorem The negation of “for large nn”

ϕ(n)\phi(n) does not have PP for large values of nn if and only if ϕ(n)\phi(n) lacks PP for infinitely many nn.

Now the two statements of §56 and §57, which the next lesson turns into the general definitions.

Example 1/n1/n is small when nn is large Hardy: §56

However small the positive number δ\delta is, 1/n<δ1/n < \delta for sufficiently large values of nn. That is: for every δ>0\delta > 0 there is an n0(δ)n_0(\delta) with 1/n<δ1/n < \delta for all n≥n0(δ)n \ge n_0(\delta). We write lim⁡n→∞1/n=0\lim_{n\to\infty} 1/n = 0, or 1/n→01/n \to 0.

Example n2n^2 is large when nn is large Hardy: §57

However large the number Δ\Delta is, n2>Δn^2 > \Delta for sufficiently large values of nn. We write n2→+∞n^2 \to +\infty. Similarly −n2→−∞-n^2 \to -\infty.

Proofs

Proof Finitely many exceptions means: from some point on
Strategy

A finite set of positive integers has a largest member (or is empty); start just after it. Conversely, if everything from n0n_0 on is good, the bad nn are among 1,…,n0−11, \dots, n_0 - 1.

Case (a) gives an n0n_0

Let EE be the set of nn for which ϕ(n)\phi(n) lacks PP, and suppose EE is finite. If EE is empty take n0=1n_0 = 1. Otherwise EE has a largest member nNn_N (previous lesson); take n0=nN+1n_0 = n_N + 1. Every n≥n0n \ge n_0 is larger than every member of EE, so it is not in EE: ϕ(n)\phi(n) has PP.

An n0n_0 gives case (a)

If ϕ(n)\phi(n) has PP for all n≥n0n \ge n_0, every member of EE is one of 1,2,…,n0−11, 2, \dots, n_0 - 1. So EE has fewer than n0n_0 members and is finite.

Proof Two properties at once
Strategy

Each property comes with its own threshold. Go past both.

Name the thresholds

There is n1n_1 with PP for all n≥n1n \ge n_1, and n2n_2 with QQ for all n≥n2n \ge n_2.

Take n0=max⁡(n1,n2)n_0 = \max(n_1, n_2)

If n≥n0n \ge n_0 then n≥n1n \ge n_1, so PP holds, and n≥n2n \ge n_2, so QQ holds.

The same works for any finite number of properties

Take the largest of finitely many thresholds. (It fails for infinitely many properties: "n>kn > k" holds for large nn for each kk, but no nn has all of them.)

Proof The negation of “for large nn”
Strategy

Combine the theorem above with the fact that a set of positive integers is finite if and only if it has no members beyond some point.

Translate

By the first theorem, ”PP for large nn” means that the set EE of nn lacking PP is finite.

Negate

So “not PP for large nn” means that EE is infinite, that is, ϕ(n)\phi(n) lacks PP for infinitely many nn. Equivalently: however large n0n_0 is, there is some n≥n0n \ge n_0 at which PP fails.

The last theorem is the reason that case (c) matters. “Infinitely many nn” is much weaker than “all large nn”: the even numbers are infinitely many, but they do not include all large numbers. Much confusion about limits comes from mixing up the two.

Hardy’s “imaginary opponent” from §56 makes the statement about 1/n1/n concrete. The opponent names smaller and smaller numbers δ\delta; you must reply each time with a point beyond which 1/n<δ1/n < \delta. Play the opponent below.

Explore Hardy’s opponent

Try this. The opponent names δ=0.1\delta = 0.1, then 0.010.01, then 0.0010.001. For each, find an n0n_0 from which every point is inside the band, and claim it.

ϕ(n)=1n\phi(n) = \frac{1}{n} , candidate limit l=0l = 0Round 1 of 3
Opponent: “I name δ=0.1\delta = 0.1. Find me an n0n_0 so that every term from n0n_0 on is within δ\delta of ll.”
nφ(n)
  • band l ± δ
  • l
  • n₀ (drag)
  • n ≥ n₀, good
  • n ≥ n₀, bad
What to notice

You always win, and your answer is always about 1/δ1/\delta. That is the content of “1/n1/n is small for large nn”: not one statement but a whole family, one for each δ\delta, each with its own n0(δ)n_0(\delta).

Proof 1/n<δ1/n < \delta for large nn, and n2>Δn^2 > \Delta for large nn
Strategy

Solve each inequality for nn, then take n0n_0 to be the first integer beyond the solution.

Given δ>0\delta > 0, take n0=[1/δ]+1n_0 = [1/\delta] + 1

Then n0>1/δn_0 > 1/\delta, so for n≥n0n \ge n_0 we have n>1/δn > 1/\delta and 1/n<δ1/n < \delta. With δ=0.001\delta = 0.001 this gives n0=1001n_0 = 1001, matching example (viii).

Given Δ\Delta, take n0=[Δ]+1n_0 = [\sqrt{\Delta}] + 1 when Δ>0\Delta > 0

Then n≥n0n \ge n_0 gives n>Δn > \sqrt{\Delta}, so n2>Δn^2 > \Delta. If Δ≤0\Delta \le 0, n0=1n_0 = 1 already works, since n2≥1>Δn^2 \ge 1 > \Delta.

For −n2-n^2

−n2<−Δ-n^2 < -\Delta exactly when n2>Δn^2 > \Delta, so the same n0n_0 shows that −n2-n^2 is below any −Δ-\Delta for large nn.

Worked examples and exercises

Exercise Exceptions that are not the first values ●○○ Hardy: §53, Example (vii)

Let ϕ(n)=1000{1+(−1)n}/n\phi(n) = 1000\{1 + (-1)^n\}/n. Find exactly the nn for which ϕ(n)≥1\phi(n) \ge 1, and the smallest n0n_0 such that ϕ(n)<1\phi(n) < 1 for all n≥n0n \ge n_0.

Solution

For odd nn, 1+(−1)n=01 + (-1)^n = 0 and ϕ(n)=0<1\phi(n) = 0 < 1. For even nn, ϕ(n)=2000/n\phi(n) = 2000/n, which is at least 11 exactly when n≤2000n \le 2000. So the exceptions are 2,4,6,…,20002, 4, 6, \dots, 2000: a finite set whose largest member is 20002000. The smallest n0n_0 is 20012001. The first exception is n=2n = 2, not n=1n = 1, which is Hardy’s point: exceptions need not come first.

Exercise How large is large? ●○○ Hardy: §53, Example (xi)

Let ϕ(n)=(n−1)/(n+1)\phi(n) = (n-1)/(n+1). Find the least n0n_0 with 1−ϕ(n)<0.00011 - \phi(n) < 0.0001 for all n≥n0n \ge n_0. What is the least n0n_0 if 0.00010.0001 is replaced by a general δ>0\delta > 0?

Hint
1−ϕ(n)=2/(n+1)1 - \phi(n) = 2/(n+1).

Solution

1−n−1n+1=2n+11 - \dfrac{n-1}{n+1} = \dfrac{2}{n+1}, which decreases as nn grows. It is less than 0.00010.0001 exactly when n+1>20000n + 1 > 20000, that is, n≥20000n \ge 20000. So n0=20000n_0 = 20000. In general 2/(n+1)<δ2/(n+1) < \delta exactly when n>2/δ−1n > 2/\delta - 1, so (for δ<2\delta < 2) the least n0n_0 is the least integer greater than 2/δ−12/\delta - 1, which is [2/δ][2/\delta] whether or not 2/δ2/\delta is itself an integer. Check: δ=0.0001\delta = 0.0001 gives [20000]=20000[20000] = 20000. So n0(δ)n_0(\delta) grows like 2/δ2/\delta.

Exercise Primes beat their index ●○○ Hardy: §53, Example (iv)

Let pnp_n be the nn-th prime. Show that pn>np_n > n for every nn, so that case (a) holds with no exceptions at all.

Solution

p1=2>1p_1 = 2 > 1. The primes form a strictly increasing sequence of integers, so pn+1≥pn+1p_{n+1} \ge p_n + 1. If pn≥n+1p_n \ge n + 1, then pn+1≥n+2p_{n+1} \ge n + 2. By induction pn≥n+1>np_n \ge n + 1 > n for every nn.

Exercise Large for every purpose? ●●○ Hardy: §§54–55

Show that there is no single n0n_0 such that 1/n<δ1/n < \delta for all n≥n0n \ge n_0 and every δ>0\delta > 0. What does this say about the phrase ”1/n1/n is small for large nn”?

Hint
Suppose such an n0n_0 exists and choose δ=1/n0\delta = 1/n_0.

Solution

If such an n0n_0 existed, take δ=1/n0\delta = 1/n_0 and n=n0n = n_0: we would need 1/n0<1/n01/n_0 < 1/n_0, which is false. So the threshold must depend on δ\delta, and ”1/n1/n is small for large nn” cannot mean “there is one point beyond which 1/n1/n is small for every purpose”. It means: for each δ>0\delta > 0, 1/n<δ1/n < \delta for large nn. The order “first δ\delta, then n0n_0” is forced. This is exactly why Hardy writes n0(δ)n_0(\delta).

Exercise Tending to infinity, or merely unbounded? ●●○ Hardy: §57

Let ϕ(n)=n\phi(n) = n for even nn and ϕ(n)=1\phi(n) = 1 for odd nn. Show that for every Δ\Delta, ϕ(n)>Δ\phi(n) > \Delta for infinitely many nn, but that ϕ(n)>Δ\phi(n) > \Delta does not hold for large nn when Δ≥1\Delta \ge 1. Which of the cases (a), (b), (c) is "ϕ(n)>1\phi(n) > 1" in?

Solution

For every Δ\Delta, all even n>Δn > \Delta give ϕ(n)=n>Δ\phi(n) = n > \Delta; there are infinitely many. But when Δ≥1\Delta \ge 1, every odd nn gives ϕ(n)=1≤Δ\phi(n) = 1 \le \Delta, so the property fails for infinitely many nn, and by the negation theorem it does not hold for large nn. The property "ϕ(n)>1\phi(n) > 1" holds exactly for the even n≥2n \ge 2 and fails for the odd nn: case (c). So ϕ(n)\phi(n) does not tend to +∞+\infty in the sense of §57: being larger than Δ\Delta infinitely often is not enough.

Check yourself

Check yourself Choose one

”ϕ(n)\phi(n) has property PP for large values of nn” means:

Why

“Large” has no fixed size: it means beyond some threshold n0n_0, which depends on PP. “Infinitely many” is weaker: the property ”nn is even” holds for infinitely many nn but not for large nn. A single large nn proves nothing about the others.

Check yourself True or false?

If ϕ(n)\phi(n) does not have PP for large nn, then it lacks PP for infinitely many nn.

Why

True: this is the negation theorem. If only finitely many nn lacked PP, the largest of them plus one would serve as n0n_0, and PP would hold for large nn.

Check yourself Choose one

For ϕ(n)=1−(−1)n/n\phi(n) = 1 - (-1)^n/n and the property ϕ(n)<1\phi(n) < 1, which case holds?

Why

ϕ(n)<1\phi(n) < 1 exactly when (−1)n/n>0(-1)^n/n > 0, that is, for even nn. Both the even and the odd numbers are infinite classes, so neither (a) nor (b) holds.

Check yourself True or false?

For a very large nn it is correct to write n=∞n = \infty.

Why

False. There is no number ∞\infty, and "n=∞n = \infty" means nothing. The symbol ∞\infty has a meaning only inside phrases that have been defined as a whole, such as "n→∞n \to \infty" or "n2→+∞n^2 \to +\infty".

Hardy’s language

The reader cannot impress upon himself too strongly that when we say that n ‘tends to ∞’ we mean simply that n is supposed to assume a series of values which increase beyond all limit.

— G. H. Hardy, §55

Summary

Summary
  • For any property PP of ϕ(n)\phi(n), exactly one of three cases holds: (a) all but finitely many nn have PP; (b) only finitely many do; (c) neither.
  • Case (a) is the same as ”PP holds for large nn”: there is an n0n_0 with PP for every n≥n0n \ge n_0. The exceptions need not be the first nn, but there is a last one.
  • Two (or finitely many) properties that each hold for large nn hold together for large nn: take the larger threshold.
  • “Not for large nn” means “fails for infinitely many nn”. Infinitely many is not the same as all large.
  • "n→∞n \to \infty" means only that nn increases beyond every bound; ∞\infty alone means nothing. "1/n→01/n \to 0" means: for every δ>0\delta > 0, 1/n<δ1/n < \delta for large nn.