Every power xn has an antiderivative that is again a power, xn+1/(n+1), with one exception: n=−1. No polynomial, rational function or familiar algebraic expression has derivative 1/x. Hardy’s response in §204 is to define a new function by the property we want, namely as the area under y=1/t, and then to discover everything about it from that definition alone. This lesson sets up the definition and proves the first properties; the rest of the chapter grows out of it.
The idea
Mathematics often gains new functions the way it gained new numbers. When x2=2 has no rational solution, we enlarge the number system. When ∫dx/x cannot be expressed in known functions, we enlarge the stock of functions: we give a name to the integral itself and study it.
So for x>0 we let logx be the area under the curve y=1/t between t=1 and t=x, counted positive when x>1 and negative when x<1 (because the integral then runs backwards). Everything we want to know about logx must now be read off this picture.
ExploreSweep out the area under 1/t
Try this.Drag the purple point x from 1 to the right, then to the left of 1 towards 0.15. Watch the lower graph, which records the area as a function of x.
F(x)=∫axt1dt, a=1
tf(t)
xF(x)
f(t)
counts +
counts −
F(x)
x (drag)
F(2.5)=∫12.5t1dt=0.91629. Slope of F at x: F′(x)=f(x)=0.4.
What to notice
The area grows as x increases, but ever more slowly, because the strip being added gets lower. To the left of 1 the area counts as negative and plunges as x→0, because 1/t shoots up there. The slope of the lower graph at x is always the height 1/x of the upper one.
Four facts are visible in the picture, and each will become a theorem.
log1=0. There is no area between 1 and 1.
logx increases, and its slope at x is 1/x: steep for small x, gentle for large x.
It tends to +∞, however slowly: the area under 1/t out to infinity is infinite, even though the curve itself tends to 0.
It tends to −∞ as x→0, because the area between x and 1 becomes unbounded.
Why is the lower limit 1? Any positive number would do, but 1 gives log1=0, and we shall see in the next lesson that this is exactly what makes logxy=logx+logy hold without an extra constant.
Precise statements
DefinitionThe logarithmHardy: §205
For x>0,
logx=∫1xtdt.
When 0<x<1 the integral is taken in the usual sense ∫1x=−∫x1. The definition is restricted to x>0: an interval of integration containing t=0 would contain the infinity of 1/t, and the integral would have no meaning.
TheoremFirst properties of logxHardy: §205
log1=0.
logx is continuous and strictly increasing for x>0, with derivative dxdlogx=x1.
logx→+∞ as x→∞.
log(1/x)=−logx. Consequently logx<0 for 0<x<1, and logx→−∞ as x→+0.
Part 2 is the fundamental theorem of calculus (Chapter VII): 1/t is continuous on every interval [a,b] with a>0, so its integral is differentiable with derivative 1/x; a positive derivative makes the function strictly increasing, and a differentiable function is continuous.
ExploreTwo rectangles trap the logarithm
Try this.With x=2, compare the shaded area with the inner and outer rectangles over [1,x]. Then move x below 1.
logx=∫1xtdt
t1/t
y = 1/t
area = log x
inner rectangle
outer rectangle
xx−1=0.5≤logx=0.6931≤x−1=1. The curved area lies between the two rectangles over [1, x].
What to notice
On [1,x] the curve 1/t lies between the heights 1/x and 1, so logx lies between (x−1)/x and x−1. For x<1 the same two expressions still trap logx (now all three are negative). At x=2 this already says 21≤log2≤1.
The rectangles give our first quantitative facts about log:
Equivalently, with 1+h in place of x: 1+hh<log(1+h)<h for h>−1, h=0.
Proofs
Part 3 does not follow from the curve tending to 0. (The area under 1/t2 out to infinity is finite.) Hardy refers back to the divergence of ∫1∞dt/t in §181; here is a self-contained version of the same idea, using areas over the intervals [1,2],[2,4],[4,8],….
Prooflogx→+∞
Strategy
Cut [1,2n] into the n pieces [2k−1,2k]. Each piece contributes at least 21 to the area, so log2n≥n/2. Since log increases, it eventually exceeds any bound.
Each doubling interval contributes at least 21
On [2k−1,2k] we have 1/t≥1/2k, and the interval has length 2k−1. So ∫2k−12kdt/t≥2k−1/2k=21.
Add up: log2n≥n/2
The integral over [1,2n] is the sum of the integrals over the n pieces, each at least 21.
Given Δ, take x0=2n with n>2Δ
For x≥x0 we have logx≥logx0≥n/2>Δ, since log is increasing. This is the definition of logx→∞.
∎
Notice that the bound log2n≥n/2 is very weak growth: to make the area exceed 50 this argument needs x≥2100. The logarithm tends to infinity, but slowly. The third lesson makes “slowly” precise.
ExploreThe same area on the other side of 1
Try this.Move x and compare the two shaded areas: one between 1 and x, the other between 1 and 1/x.
logx=∫1xtdt
t1/t
y = 1/t
between 1 and x
between 1 and 1/x
Area between 1 and x=3: 1.099. Area between 1 and 1/x=0.333: 1.099. Equal, but on opposite sides of 1, so log(1/x)=−logx=−1.099.
What to notice
They always have exactly the same size, though one is tall and thin and the other long and low. Being on opposite sides of 1, they count with opposite signs: log(1/x)=−logx.
Prooflog(1/x)=−logx, and the behaviour near 0
Strategy
Substitute t=1/u in the integral defining logx. The substitution turns the area between 1 and x into the area between 1 and 1/x, with a change of sign.
Substitute t=1/u
Then dt=−du/u2 and 1/t=u, so tdt=−udu. When t runs from 1 to x, u runs from 1 to 1/x.
Compute
logx=∫1xtdt=−∫11/xudu=−logx1.
Sign for 0<x<1
If 0<x<1 then 1/x>1, so log(1/x)>0 (an integral of a positive function over [1,1/x]), hence logx<0.
Limit at 0
As x→+0, 1/x→+∞, so log(1/x)→+∞ by part 3, and logx=−log(1/x)→−∞.
∎
ProofThe basic inequality
Strategy
Compare 1/t with its smallest and largest values on the interval of integration, and integrate.
Case x>1
For 1<t<x we have 1/x<1/t<1. Integrating over [1,x], an interval of length x−1, gives (x−1)/x<logx<x−1. The inequalities are strict because the functions differ on a whole interval.
Case 0<x<1
Now logx=−∫x1dt/t, and for x<t<1 we have 1<1/t<1/x. Integrating over [x,1], of length 1−x: 1−x<−logx<(1−x)/x. Multiplying by −1 reverses the inequalities: (x−1)/x<logx<x−1, the same statement.
∎
Is logx really new?
Hardy does not prove that logx is not an algebraic or trigonometric function (that is hard), but he notes an easy part: it is not a rational function. The reason is worth seeing, because it shows why1/x is special.
Prooflogx is not a rational function (Examples XLII, 4)
Strategy
Differentiating a rational function can only produce poles of order at least 2. But 1/x has a simple pole at 0.
Suppose logx=P(x)/Q(x) for x>0
Here P, Q are polynomials without a common factor. If Q is constant, then P′(x)=1/x, so xP′(x)=1 for all x>0; two polynomials that agree for all x>0 agree everywhere, and x=0 gives 0=1. So Q has a root a (perhaps complex), of some multiplicity k≥1.
Near a, the derivative has a pole of order k+1≥2
Write Q(x)=(x−a)kS(x) with S(a)=0, and g=P/S, so g(a)=0. Then
(QP)′=(x−a)kg′(x)−(x−a)k+1kg(x),
and the second term has a pole of order k+1 at a that the first cannot cancel.
Contradiction
The identity (P/Q)′=1/x holds for all x>0, hence (both sides being rational) for all x except poles. But 1/x has no pole of order 2 or more anywhere. So no such P/Q exists.
∎
Worked examples and exercises
Exercise●○○Hardy: Examples LXXXIII, 1
Prove from the definition that
(a) 1+xx<log(1+x)<x for x>0, and
(b) x<−log(1−x)<1−xx for 0<x<1.
Hint
For (a), log(1+x)=∫11+xdt/t, and on that interval 1/t lies between 1/(1+x) and 1.
Hint
For (b), −log(1−x)=log1−x1=∫1−x1tdt.
Solution
(a) For 1<t<1+x, 1+x1<t1<1. Integrating over an interval of length x gives the result.
(b) By part 4 of the theorem, −log(1−x)=∫1−x1dt/t. For 1−x<t<1 we have 1<1/t<1/(1−x), and the interval has length x. So x<−log(1−x)<x/(1−x).
Both are the basic inequality with 1+x or 1−x in place of x.
Exercise●○○Hardy: Examples LXXXIII, 3
Show that x−1logx→1 as x→1, that is, ylog(1+y)→1 as y→0.
Hint
Divide the basic inequality by x−1, minding the sign of x−1.
Solution
For x>1, dividing (x−1)/x<logx<x−1 by x−1>0 gives 1/x<x−1logx<1. For 0<x<1, dividing by x−1<0 reverses the signs: 1<x−1logx<1/x. In both cases the ratio lies between 1 and 1/x, and 1/x→1 as x→1. So the ratio tends to 1. (This is also the statement that the derivative of log at 1 is 1/1=1.)
Exercise●●○Hardy: Examples LXXXIII, 2 (i)
Prove that x−21x2<log(1+x) for x>0.
Hint
Rectangles are too crude. Compare 1+u1 with 1−u for u>0 instead, and integrate from 0 to x.
Solution
Substituting t=1+u, log(1+x)=∫0x1+udu. For u>0, (1+u)(1−u)=1−u2<1, so 1−u<1+u1. Integrating over [0,x]: x−21x2<log(1+x).
(Alternatively, f(x)=log(1+x)−x+21x2 has f(0)=0 and f′(x)=1+x1−1+x=1+xx2>0.) Together with (a) above: x−21x2<log(1+x)<x, the first two terms of a series we shall meet in §220.
Exercise●●○Hardy: Examples LXXXIII, 2 (iii)
Prove that 4(x−1)−2logx<2xlogx<x2−1 for x>1.
Hint
Move everything to one side and call it f(x). What are f(1) and f′(1)?
Hint
Both f(1)=0 and f′(1)=0, so look at the sign of f′′(x) for x>1.
Solution
Right-hand inequality. Let g(x)=x2−1−2xlogx. Then g(1)=0, g′(x)=2x−2logx−2, so g′(1)=0, and g′′(x)=2−2/x>0 for x>1. So g′ is strictly increasing on [1,∞), hence g′(x)>g′(1)=0 for x>1; then g is strictly increasing there, and g(x)>g(1)=0.
Left-hand inequality. Let f(x)=2xlogx+2logx−4(x−1). Then f(1)=0, f′(x)=2logx+2+2/x−4, so f′(1)=0, and f′′(x)=2/x−2/x2=2(x−1)/x2>0 for x>1. The same two steps give f(x)>0 for x>1.
Everything used about log here is log1=0 and (logx)′=1/x: exactly what the definition gives.
Check yourself
Check yourselfChoose one
Using only the two rectangles over [1,2], what can you say about log2?
Why
On [1,2] the curve 1/t lies between heights 21 and 1, over a width of 1, so 21<log2<1. (In fact log2=0.6931….) The basic inequality (x−1)/x<logx<x−1 holds for every positive x=1.
Check yourselfTrue or false?
Since the slope 1/x of logx tends to 0, the function logx is bounded above.
Why
False. A slope tending to 0 means slower and slower growth, not bounded growth. Each doubling of x adds at least 21 to logx (in fact exactly log2, as the next lesson shows), so logx→∞. Compare x, whose slope also tends to 0.
Check yourselfChoose one
Why does Hardy define logx only for x>0?
Why
The definition is the integral ∫1xdt/t. For x≤0 the range from 1 to x passes through 0, where the integrand is not defined and not bounded, and the integral (even as an infinite integral) diverges. The sign of 1/t for t<0 would be no obstacle by itself. Negative and complex logarithms are taken up in Chapter X.
Check yourselfEnter a number
Given log2=0.6931 (to 4 places), what is log21?
Why
log(1/x)=−logx, so log21=−0.6931. In the picture, the area between 21 and 1 equals the area between 1 and 2, and counts negatively.
Hardy’s language
We have, therefore, not indeed a strict proof that log x is a new function—that we do not profess to give—but a reasonable presumption that it is.
— G. H. Hardy, §204
Summary
Summary
logx=∫1xdt/t for x>0: a new function, defined by the property (logx)′=1/x that no rational function has.
log1=0; log is continuous and strictly increasing; logx→+∞ as x→∞ (each doubling adds at least 21) and logx→−∞ as x→+0.
The substitution t=1/u gives log(1/x)=−logx: equal areas on the two sides of 1.
Rectangles give xx−1<logx<x−1 for x=1, and hence x−1logx→1 as x→1.