Chapter IV was about sequences: functions whose variable n moves in jumps 1,2,3,…. Most functions of interest, such as 1/x, sinx or x, are defined for every real x. Before we can talk about continuity or derivatives we need to know what “ϕ(x) tends to l” means when x runs through all real values. In §§90–92 Hardy does this for x→±∞. The definitions hardly change, but one consequence does, and it is worth seeing clearly.
The idea
Picture a point moving along the number line to the right, passing through every point on its way rather than hopping from integer to integer. Hardy says that x tends to infinity continuously. The question is the same as before: does ϕ(x) settle down near some number l once x is large?
The challenge is also the same. Someone names a tolerance δ>0; we must answer with a point x0 such that everyx≥x0 gives ∣ϕ(x)−l∣<δ. The only difference from Chapter IV is the word every: it now covers all the real numbers beyond x0, not just the integers.
ExploreAnswer the challenge for a continuous variable
Try this.Start with 1/x. Make δ small and drag x0 (the purple handle) until the verdict turns green. Then choose sinxπ, switch on “Only integer x”, and find an x0. Now switch the toggle off.
ϕ(x)=x1, claim ϕ(x)→0 as x→∞
xφ(x)
y = φ(x)
band l ± δ for x ≥ x₀
x₀ (drag)
worst escape
This x₀ fails. At x=2 the distance from 0 is 0.5, not less than δ. Move x₀ to the right.
What to notice
For 1/x the right x0 is just over 1/δ, exactly as for 1/n. For sinxπ the integer points are all 0, so the sequence ϕ(n) tends to 0 with x0=1; but between the integers the graph swings up to 1 and down to −1 forever, and no x0 works for the function. Knowing ϕ(n) tells you very little about ϕ(x).
So the passage from n to x is not just a change of letter. A function of x has many more values to control, and each sequence ϕ(1),ϕ(2),… is only a selection from them. If ϕ(x)→l then certainly ϕ(n)→l, but not conversely.
Precise statements
Hardy restates the three definitions of Chapter IV word for word, with x in place of n.
DefinitionLimit as x→∞Hardy: Definition 1, §90
The function ϕ(x)tends to the limit l as x→∞ if, for every positive δ, however small, there is a number x0(δ) such that
∣ϕ(x)−l∣<δfor all x≥x0(δ).
We write limx→∞ϕ(x)=l, or ϕ(x)→l.
DefinitionTending to ∞Hardy: Definition 2, §90
ϕ(x)→∞ as x→∞ if, for every number Δ, however large, there is an x0(Δ) such that ϕ(x)>Δ for all x≥x0(Δ). The definition of ϕ(x)→−∞ is the same with ϕ(x)<−Δ.
DefinitionOscillationHardy: Definition 3, §90
If ϕ(x) neither tends to a limit nor to ∞ nor to −∞, it oscillates as x→∞. It oscillates finitely if there are constants K and x0 with ∣ϕ(x)∣<K for all x≥x0, and infinitely otherwise.
Notice the x0 in the last definition. For sequences it made no difference whether we asked for ∣ϕ(n)∣<K for all n or only for large n, because only finitely many terms come before n0. Here there are infinitely many x below x0, and they can be unbounded: 1/x is huge near 0, yet it tends to 0 as x→∞. Only large x count.
Tending to −∞ (§91). Put x=−y. As x→−∞, y→∞, so “ϕ(x)→l as x→−∞” simply means “ϕ(−y)→l as y→∞”. In words: for every δ>0 there is an x0 with ∣ϕ(x)−l∣<δ for all x≤x0.
The theorems carry over (§92). The theorems of §§63–68 on sums, products and quotients hold for functions of x with the same proofs: if ϕ(x)→l and ψ(x)→l′ then ϕ+ψ→l+l′, ϕψ→ll′, and ϕ/ψ→l/l′ provided l′=0. So does the theorem on monotone sequences, once we say what “monotone” means for a function.
DefinitionIncreasing steadilyHardy: §92
ϕ(x)increases steadily if ϕ(x2)≥ϕ(x1) whenever x2>x1, and increases strictly if ϕ(x2)>ϕ(x1) whenever x2>x1. Often this holds only from some point on: for x2>x1≥x0.
TheoremMonotone functions have limits at infinityHardy: §92 (from §69)
If ϕ(x) increases steadily for x≥x0, then either ϕ(x) is bounded above for x≥x0, and then ϕ(x) tends to a limit l (its upper bound there), or it is not, and then ϕ(x)→∞.
Proofs
ProofMonotone functions have limits at infinity
Strategy
The limit, if it exists, should be the least number that ϕ(x) never exceeds. So take l to be the upper bound of the values ϕ(x), x≥x0. Since l−δ is not an upper bound, some value ϕ(x1) exceeds it, and monotonicity keeps every later value above it too.
Bounded case: let l be the upper bound
If ϕ(x)≤K for all x≥x0, the set of values ϕ(x) (x≥x0) has an upper bound l (§80): no value exceeds l, and every number less than l is exceeded by some value.
Given δ>0, find x1 with ϕ(x1)>l−δ
Because l−δ<l, the number l−δ is exceeded by at least one value. Call the place x1.
Every x≥x1 works
If x≥x1 then ϕ(x)≥ϕ(x1)>l−δ because ϕ increases steadily, and ϕ(x)≤l because l is an upper bound. So ∣ϕ(x)−l∣<δ. Thus x1 serves as x0(δ), and ϕ(x)→l.
Unbounded case
If no K bounds ϕ above, then for any Δ some ϕ(x1)>Δ, and then ϕ(x)≥ϕ(x1)>Δ for every x≥x1. That is ϕ(x)→∞.
∎
The relation between ϕ(x) and ϕ(n) is a one-line proof, but its failure in the other direction is the real lesson of the widget.
ProofIf ϕ(x)→l then ϕ(n)→l; the converse is false
Strategy
An x0 that works for all real x≥x0 works in particular for the integers. For the converse, one example is enough.
Given δ, take x0(δ) for the function
Then ∣ϕ(x)−l∣<δ for all real x≥x0, so in particular for all integers n≥x0. Any integer n0≥x0 serves for the sequence.
The converse fails for sinxπ
sinnπ=0 for every integer n, so ϕ(n)→0. But sinxπ=1 at every x=2k+21 and −1 at every x=2k−21. With δ=21, every x0 has such points beyond it, at distance 1 from any candidate limit l in one direction or the other, so no l works. Since ∣sinxπ∣≤1, the function oscillates finitely.
∎
Worked examples and exercises
Exercise●○○Hardy: Examples XXXIV, 1
Describe the behaviour as x→∞ of [x], x−[x] and [x]+x−[x].
Hint
Sketch each graph on [0,4]. Where are the values at the integers, and where in between?
Hint
For the third, note that [x]≤[x]+x−[x]<[x]+1.
Solution
[x]→∞. Given Δ, take x0=max(Δ+1,0). For x≥x0, [x]>x−1≥Δ.
x−[x] oscillates finitely. Its values lie in [0,1), so it is bounded. It equals 0 at every integer and equals 21 at every n+21, so with δ=41 no l is within δ of both 0 and 21, and no x0 works. Note that the sequence n−[n]=0 does tend to 0.
[x]+x−[x]→∞, because it is at least [x], which tends to ∞. (In fact it is continuous and strictly increasing: on [n,n+1) it runs from n up towards n+1.)
Exercise●●○Hardy: Examples XXXIV, 2
How do these behave as x→∞?
xsinxπ,xsinxπ,(xsinxπ)2,acos2xπ+bsin2xπ.Hint
Bound ∣sinxπ∣≤1 for the first. For the others, evaluate at integers and at half-integers.
Solution
First.∣sinxπ/x∣≤1/x<δ when x>1/δ, so it tends to 0.
Second. At x=2k+21 it equals 2k+21, which is arbitrarily large; at x=2k−21 it equals −(2k−21). It is unbounded in both directions for large x, so it neither tends to a limit nor to ±∞: it oscillates infinitely.
Third. It is 0 at every integer and (k+21)2 at x=k+21. It is unbounded, so it does not tend to a limit, and the zeros at the integers stop it tending to ∞. It oscillates infinitely, although it is never negative.
Fourth. At integers it equals a, at half-integers b, and always lies between them. If a=b it is the constant a and tends to a. If a=b it oscillates finitely.
Exercise●●○Hardy: Examples XXXIV, 4
Suppose ϕ(x)→l with l=0. Show that ϕ(x)cosxπ oscillates finitely. What happens if instead ϕ(x)→∞?
Hint
Look at the values at even and at odd integers.
Solution
Take δ=∣l∣: for x≥x0 we have ∣ϕ(x)∣<2∣l∣, so ∣ϕ(x)cosxπ∣<2∣l∣ there; the function is bounded for large x. At even integers it equals ϕ(2k)→l, at odd integers −ϕ(2k+1)→−l. If it tended to some limit L, both of these sequences would tend to L, forcing l=L=−l, so l=0, which is excluded. A bounded function cannot tend to ±∞, so it oscillates finitely.
If ϕ(x)→∞, the values at even integers tend to +∞ and those at odd integers to −∞; the function is unbounded in both directions and oscillates infinitely. Its graph is a wave trapped between y=ϕ(x) and y=−ϕ(x).
ExerciseWhich of these increase steadily?●●○Hardy: §92
Decide whether each function increases steadily, at least for large x: x2−x, x+sinx, x+2sinx, [x], [x]+sinx.
Hint
For x+sinx, write sinx2−sinx1=2cos21(x1+x2)sin21(x2−x1) and use ∣sint∣<∣t∣ for t=0.
Hint
For x+2sinx, compare the values at 2kπ+21π and 2kπ+π.
Solution
x2−x: yes for x≥21, since x22−x2−(x12−x1)=(x2−x1)(x2+x1−1)>0 when x2>x1≥21. It is strictly increasing there.
x+sinx: yes, strictly, everywhere. For x2>x1 put t=21(x2−x1)>0. Then ∣sinx2−sinx1∣≤2∣sint∣<2t=x2−x1, so (x2+sinx2)−(x1+sinx1)>0.
x+2sinx: no. At x=2kπ+21π the value is 2kπ+21π+2≈2kπ+3.57; at the larger x=2kπ+π it is 2kπ+π≈2kπ+3.14. This happens for every k, so the function never settles into increasing.
[x]: yes, steadily but not strictly: it is constant on each [n,n+1).
[x]+sinx: no. Each interval [2kπ+21π,2kπ+23π] has length π>1, so it contains a stretch of length at least 21 free of integers; there [x] is constant while sinx strictly decreases.
All five tend to ∞: monotonicity is not needed for that.
Check yourself
Check yourselfTrue or false?
If ϕ(n)→l as n→∞ through the integers, then ϕ(x)→l as x→∞.
Why
False. The integers are only a selection of the values of x. sinxπ and x−[x] are 0 at every integer but oscillate between the integers. The true statement goes the other way: ϕ(x)→l implies ϕ(n)→l.
Check yourselfChoose one
Which function oscillates finitely as x→∞?
Why
sinxπ+1/x is bounded for x≥1 (it lies between −1 and 2) and has no limit, because it is near 1 at x=2k+21 and near −1 at x=2k−21. xsinxπ is unbounded, so it oscillates infinitely. x−sinx≥x−1→∞. 1/x tends to 0.
Check yourselfTrue or false?
1/x (for x>0) is unbounded, yet it is bounded “as x→∞” in the sense of Definition 3.
Why
True. Definition 3 asks only for ∣ϕ(x)∣<K when x≥x0; take x0=1, K=2. What happens near 0 is irrelevant to the behaviour as x→∞.
Check yourselfChoose one
ϕ(x) increases steadily for all x and ϕ(x)<3 for every x. What can you conclude?
Why
A steadily increasing function bounded above tends to its upper bound, which is some l≤3, not necessarily 3: ϕ(x)=1−1/(1+x2) for x≥0, extended by 0 for x<0, increases steadily, stays below 3 and tends to 1. Monotonicity rules out oscillation, and strictness is not needed.
Hardy’s language
The only difference between the ‘tending of n to ∞’ discussed in the last chapter, and this ‘tending of x to ∞’, is that x assumes all values as it tends to ∞
— G. H. Hardy, §90
Summary
Summary
ϕ(x)→l as x→∞: for every δ>0 there is x0 with ∣ϕ(x)−l∣<δ for all realx≥x0. Likewise for ±∞ and for oscillation.
x→−∞ reduces to x→∞ by the substitution x=−y.
ϕ(x)→l implies ϕ(n)→l, but not conversely: sinxπ and x−[x] vanish at the integers and oscillate between them.
The algebra of limits and the monotone-limit theorem carry over unchanged: a steadily increasing function tends to its upper bound, or to ∞ if it is unbounded.