Chapter IV · Limits of functions of a positive integral variable

Functions of a positive integral variable

§50 20 min

A sequence is a function whose field of definition is the positive integers, and its graph is a row of dots.

Why this matters

Chapter II studied functions y=ϕ(x)y = \phi(x) of a real variable. Chapter IV studies something that looks narrower: functions defined only when the variable is 1,2,3,…1, 2, 3, \dots. These are the sequences of modern books, and they are where the idea of a limit is easiest to see and to make precise. §50 sets the scene by asking a simple question about any function: for which values of the variable is it defined at all?

The idea

When we write a formula in xx we rarely say for which xx it makes sense. Usually that is harmless, but Hardy’s list of examples in §50 shows how different the answers can be. The function xx makes sense for every real number. The function x\sqrt{x} only for x≥0x \ge 0. “The denominator of xx” only when xx is rational: π\pi has no denominator. “The xx-th prime number” only when xx is a positive integer: there is no 2122\tfrac12-th prime. Hardy’s last example, the height of convict number xx in Dartmoor prison, is defined only for x=1,2,…,Nx = 1, 2, \dots, N, where NN is the number of prisoners: a finite field.

The set of values for which a function is defined is its field of definition. Seeing the field is easier than describing it, so try the explorer.

Explore Same letter, different fields

Try this. Choose “(iii) the denominator of x” and raise the largest denominator drawn from 1 to 10. Then switch on “keep only x = 1, 2, 3, …”. Repeat with (ii) and (vii).

y=q when x=p/q in lowest termsy = q \text{ when } x = p/q \text{ in lowest terms}
xy
  • values on the whole field
  • field of definition (on the x-axis)
  • values at x = 1, 2, 3, …
Field of definition: the rational numbers only. Only fractions with denominator at most 6 are drawn; the field itself has points in every interval, however short, yet misses every irrational number.
What to notice

The denominator function lives on the rationals: its dots never fill a curve, however many you draw. Throwing away everything except the positive integers leaves a row of dots, and for the denominator that row is constant, because every integer has denominator 1. For the x-th prime nothing is thrown away: its field was the positive integers all along.

Hardy then fixes attention on functions like the xx-th prime, defined exactly at the positive integers. He offers two ways of looking at such a function.

  • As a function of the real variable xx whose definition fails at every xx that is not a positive integer.
  • As a function of a new variable nn that only ever takes the values 1,2,3,…1, 2, 3, \dots. Then it is defined everywhere it is asked to be, and we write y=ϕ(n)y = \phi(n).

The second view is the one used for the rest of the chapter. Its graph (in the sense of Chapter II) is not a curve but an infinite row of isolated points (1,ϕ(1)),(2,ϕ(2)),(3,ϕ(3)),…(1, \phi(1)), (2, \phi(2)), (3, \phi(3)), \dots, going off to the right for ever.

Precise statements

Definition Field of definition

The field of definition of a function is the set of values of its variable for which the function is defined.

Definition Function of a positive integral variable Hardy: §50

A function of the positive integral variable nn is a rule ϕ\phi that assigns a real number ϕ(n)\phi(n) to each n=1,2,3,…n = 1, 2, 3, \dots. Its values, listed in order, form the sequence

ϕ(1), ϕ(2), ϕ(3), …, ϕ(n), …\phi(1),\ \phi(2),\ \phi(3),\ \dots,\ \phi(n),\ \dots

The link with functions of xx goes both ways, and Hardy states both directions.

Theorem Restriction and extension Hardy: §50

(1) Every function of xx whose field contains all the positive integers gives a function of nn, by ignoring every other value of xx.

(2) Every function of nn arises in this way from infinitely many different functions of xx defined for all real xx.

Proofs

Part (1) needs no proof beyond the words: from y=x2y = x^2 we get y=n2y = n^2 by simply not looking at non-integral xx. Part (2) is more interesting, because it says that the values at the integers never determine the function of xx.

Proof Every function of nn comes from infinitely many functions of xx
Strategy

Build one function of xx that agrees with ϕ(n)\phi(n) at every positive integer, then spoil it between the integers in infinitely many different ways that leave the integers untouched.

One extension: join the dots

For x≥1x \ge 1 let n=[x]n = [x] (the integral part of xx) and set F(x)=ϕ(n)+(x−n){ϕ(n+1)−ϕ(n)}F(x) = \phi(n) + (x - n)\{\phi(n+1) - \phi(n)\}, the straight line from (n,ϕ(n))(n, \phi(n)) to (n+1,ϕ(n+1))(n+1, \phi(n+1)). For x<1x < 1 set F(x)=ϕ(1)F(x) = \phi(1). When x=nx = n is a positive integer the bracket is multiplied by 00, so F(n)=ϕ(n)F(n) = \phi(n).

A function that vanishes at every integer

The function sin⁡πx\sin \pi x is zero exactly when xx is an integer, and it is not identically zero (it equals 11 at x=12x = \tfrac12).

Infinitely many extensions

For each real number cc put Fc(x)=F(x)+csin⁡πxF_c(x) = F(x) + c\sin \pi x. Then Fc(n)=ϕ(n)+0=ϕ(n)F_c(n) = \phi(n) + 0 = \phi(n) for every positive integer nn, so each FcF_c is an extension of ϕ\phi.

They are all different

If c≠c′c \neq c' then Fc(12)−Fc′(12)=(c−c′)sin⁡π2=c−c′≠0F_c(\tfrac12) - F_{c'}(\tfrac12) = (c - c')\sin\tfrac{\pi}{2} = c - c' \neq 0. So different values of cc give different functions, and there are infinitely many values of cc.

The freedom in step 3 is the subject of the next section, §51: when there are so many functions of xx through the same dots, which one deserves to be called the natural one?

Worked examples and exercises

Exercise Fields of definition ●○○ Hardy: §50

Find the field of definition of each of Hardy’s functions (i)–(vii): (i) xx; (ii) x\sqrt{x}; (iii) the denominator of xx; (iv) the square root of the product of the numerator and the denominator of xx; (v) the largest prime factor of xx; (vi) the product of x\sqrt{x} and the largest prime factor of xx; (vii) the xx-th prime.

Hint
For each one, ask what xx must be for the words to make sense: a denominator needs a fraction, a prime factor needs an integer, a square root of a real number needs something ≥0\ge 0.

Solution

(i) all real xx. (ii) x≥0x \ge 0 (Hardy says “positive”, treating 0=0\sqrt{0} = 0 as a borderline case). (iii) all rational xx, with x=p/qx = p/q written in lowest terms and q>0q > 0. (iv) the positive rationals (and 0=0/10 = 0/1, with the same borderline remark as in (ii)): for x=p/q>0x = p/q > 0 the product pqpq is positive, while for negative xx it is negative and has no real square root. (v) the integers that have a prime factor, that is, every integer except 00 and ±1\pm 1; Hardy says loosely “all integral values”. (vi) both conditions at once: x≥0x \ge 0 and xx an integer with a prime factor, so x=2,3,4,…x = 2, 3, 4, \dots, essentially the positive integers. (vii) the positive integers.

Exercise What restriction can lose ●○○ Hardy: §50

Restrict each function to the positive integers: ϕ(x)=sin⁡πx\phi(x) = \sin \pi x, ψ(x)=x−[x]\psi(x) = x - [x] and χ(x)=\chi(x) = the denominator of xx. What sequences do you get, and what does this tell you about recovering a function of xx from its function of nn?

Hint
Evaluate each at x=1,2,3x = 1, 2, 3.

Solution

All three give the constant sequences 0,0,0,…0, 0, 0, \dots (for sin⁡nπ\sin n\pi and n−[n]n - [n]) and 1,1,1,…1, 1, 1, \dots (for the denominator). Three very different functions of xx (a smooth wave, a sawtooth, and a function defined only on the rationals) become trivial after restriction. Restriction throws away almost all the information about a function of xx, which is why part (2) of the theorem has infinitely many answers.

Exercise A function of n with no formula ●●○ Hardy: §50

Define u(1)=u(2)=1u(1) = u(2) = 1 and u(n+2)=u(n+1)+u(n)u(n+2) = u(n+1) + u(n) for n≥1n \ge 1. Explain why this defines a function of the positive integral variable nn, although no formula in xx was given. Then find an extension to all real x≥1x \ge 1.

Hint
For the first part: knowing u(1),…,u(k)u(1), \dots, u(k) determines u(k+1)u(k+1). For the second, the proof above works for any function of nn.

Solution

The rule fixes u(1)u(1) and u(2)u(2), and each later value is fixed by the two before it. So every positive integer receives exactly one value (formally, by induction on nn): 1,1,2,3,5,8,13,…1, 1, 2, 3, 5, 8, 13, \dots That is all a function of nn requires. Any rule in words or by recursion counts, just as Hardy’s “xx-th prime” does.

An extension: join consecutive dots by straight lines, F(x)=u(n)+(x−n){u(n+1)−u(n)}F(x) = u(n) + (x - n)\{u(n+1) - u(n)\} with n=[x]n = [x]. Adding csin⁡πxc \sin \pi x gives infinitely many others. (A famous smooth one is Binet’s formula {τx−cos⁡(πx) τ−x}/5\{\tau^x - \cos(\pi x)\,\tau^{-x}\}/\sqrt{5} with τ=12(1+5)\tau = \tfrac12(1 + \sqrt 5), which you can check agrees with u(n)u(n) at n=1,2n = 1, 2 and satisfies the same recursion.)

Exercise A finite field ●○○ Hardy: §50

Hardy’s function (viii), the height of convict xx, is defined only for x=1,2,…,Nx = 1, 2, \dots, N. Explain why such a function is of no use in Chapter IV, whose question is how ϕ(n)\phi(n) behaves “as nn tends to infinity”.

Solution

Its field is finite: there is a last value x=Nx = N and nothing beyond it. Every question in this chapter is about the values ϕ(n)\phi(n) for large nn, that is, for nn beyond any given point. A function with a finite field has no such values. The essential property of the field 1,2,3,…1, 2, 3, \dots, which Hardy singles out in §53, is that it is infinite.

Check yourself

Check yourself Choose one

What is the field of definition of “the denominator of xx”?

Why

A denominator exists only for a number written as a fraction p/qp/q, that is, for a rational number (in lowest terms, with q>0q > 0). An irrational number such as 2\sqrt 2 has no denominator. The positive integers are part of the field (with denominator 11), but not all of it.

Check yourself True or false?

If two functions of xx agree at every positive integer, they are the same function.

Why

False. 00 and sin⁡πx\sin \pi x agree at every integer and nowhere else between consecutive integers. The values at 1,2,3,…1, 2, 3, \dots say nothing about what happens in between.

Check yourself Choose one

What does the graph of a function of nn look like?

Why

There is exactly one point (n,ϕ(n))(n, \phi(n)) for each n=1,2,3,…n = 1, 2, 3, \dots, and nothing in between, so the points are isolated. There are infinitely many of them because there are infinitely many positive integers. Joining them with a curve, as some plots do, is a visual aid, not part of the function.

Hardy’s language

Summary

Summary
  • The field of definition (domain) of a function is part of the function; the same formula can live on the reals, the rationals or the integers.
  • A function of the positive integral variable nn, a sequence, is defined exactly at n=1,2,3,…n = 1, 2, 3, \dots; its graph is an infinite row of isolated dots.
  • Restricting a function of xx to the positive integers gives a function of nn, but loses all information between the integers.
  • Conversely, every function of nn extends to infinitely many functions of xx, for example by adding csin⁡πxc\sin\pi x.