Chapter VI · Derivatives and integrals
The derivative
The tangent as the limit of chords, the derivative as the limit of a difference quotient, and the ways in which a continuous function can fail to have one.
Why this matters
A continuous curve has no breaks, but it may still have corners, cusps or endless wiggles. The curves of elementary geometry are smoother than that: at each point they have a definite direction. In §§111–113 Hardy turns “direction” into a limit, the derivative, and then shows that it is a genuinely new property: every differentiable function is continuous, but not every continuous function is differentiable. Every later section of the chapter is built on this one definition.
The idea
Fix a point on the graph of , and a second point on the same graph. The chord rises by over a run of , so its slope is
The school definition of the tangent at is “the limiting position of the chord as moves up to ”. If that limiting line exists and is not vertical, its slope must be the limit of as . Notice what kind of limit this is. The quotient is a function of that is not defined at (it would read ), so we are asking exactly the question of §93: what happens to as approaches , from either side, without reaching it? Note also that may be negative, with to the left of . The run and the rise then both carry signs, and the slope is still their quotient.
Try this. With at , press Halve h several times and watch the two chord slopes in the read-out. Then drag the zoom slider up to about .
- y = f(x)
- chord to a + h
- chord to a − h
- tangent
- the point a (drag)
to the left:
What to notice
The slopes approach from both sides, the right-hand chord from above () and the left-hand chord from below (). Under strong zoom the parabola becomes indistinguishable from its tangent line: near a point where a derivative exists, the graph is nearly straight, and is the slope of that straight line.
So for each where the limit exists we get a number , the slope of the tangent there. Letting vary gives a new function, . Hardy calls it the derivative or differential coefficient of .
Try this. Drag the purple point slowly along the whole graph of from left to right. The lower plot records the slope of the tangent at every place you have visited.
- y = φ(x)
- tangent
- φ′(x): the slopes you have visited
What to notice
The trace you paint is the graph of . It is zero exactly where the tangent is horizontal (the tops and bottoms of the sine wave), largest where the sine curve crosses the axis going up, and negative wherever the curve is falling. Use Reveal to check.
Precise statements
Hardy reached the derivative through geometry, but he stresses (§112, remark 3) that the definition itself mentions no curve at all.
Let be defined in an interval containing in its interior. If the limit
exists (as a finite number), it is called the derivative of at , and is said to be differentiable at . The limit must be the same whether through positive or through negative values.
The two one-sided limits, as and as , are the slopes of the right and left half-tangents. A derivative exists exactly when both one-sided limits exist and are equal. If both exist but are different, the curve has a corner (Hardy’s Fig. 35).
Two other readings of the same limit are worth keeping in mind. In dynamics, if is the position of a particle at time , then is its average velocity over the time from to , and is its velocity at the instant . In geometry, is the slope of the tangent, where is the angle the tangent makes with the -axis. The tangent at is then the line .
If exists, then is continuous at .
The converse is false, and §112 and §113 are a catalogue of the ways it fails. The widget below plots the quotient itself, as a function of , so that you can watch its limit at exist or fail to exist.
Try this. Go through the menu. For each function, drag the zoom slider all the way down (so that ranges over a tiny interval around ) and decide: does approach one value from both sides?
- q(h), the chord slope
- its limit φ′(a)
What to notice
For and the graph of flattens onto the dashed line: a derivative. For the two sides approach different values (), a corner. For the quotient is , which keeps oscillating at every scale. For it oscillates too, but inside the band , so it does tend to . For it grows without bound: a vertical tangent.
The three failures in that gallery are different in kind.
- A corner: both one-sided limits exist but differ. Example: at , with slopes and . The curve has two directions at the point.
- No direction at all: the quotient oscillates. Example (§112): for , . This is continuous at , since , but has no limit.
- A vertical tangent: the quotient tends to or . The chord does have a limiting position, but the tangent is perpendicular to the -axis and its slope is not a number. Hardy’s Fig. 36 shows four versions, given by , , and at the origin. In the last two the curve exists on one side only and splits into two branches , so can approach from one side only. That alone rules out a derivative.
Hardy mentions, without proof, that there are even continuous functions that have a derivative at no point at all (Weierstrass found the first, in 1872). So continuity is a much weaker property than differentiability.
If is not continuous at (in particular, if is not defined there), then has no derivative at . The converse fails: can be continuous at and still have no derivative there.
This is just the contrapositive of the previous theorem, together with the examples above. For instance and have no derivative at , and , the integral part of , has none at any integer.
Proofs
Write the change in as (chord slope) (run). As the first factor tends to the finite number and the second to , so the product tends to .
Rewrite the increment
For , . This identity is the whole trick: it is legitimate because .
Take limits of the two factors
As the first factor tends to by hypothesis, and the second factor tends to . By the theorem on the limit of a product (§92), the product tends to .
Conclude
So as , which is exactly continuity at . Notice where finiteness was used. If (a vertical tangent), the product “” tells us nothing, and indeed the function need not be continuous then. Try at , where .
Continuity comes from squeezing between and . For the derivative, compute exactly: it is , and we exhibit two sequences of tending to along which it takes different constant values.
Continuity at
For , . Given , take : if then .
The quotient
For , .
Two sequences with different limits
Along we have ; along we have . Both sequences tend to through positive values. If had a limit as , then along every such sequence would tend to , so and at once. Hence there is no limit, even from the right alone; by symmetry ( is odd in , since is odd, so ) none from the left either.
Worked examples and exercises
Show from the definition that a constant has derivative , and that has derivative . What do these say about the graphs?
Hint
Solution
For , for every , so the limit is . For , for every , so . Geometrically: every chord of a straight line is the line itself, so the tangent is the line, and its slope is (zero for a horizontal line). Here is constant, so no real limiting process is needed.
Show that the derivative of is when is a positive integer. Then find the derivative of () by a special device.
Hint
Hint
Solution
By the binomial theorem,
This is a polynomial in with finitely many terms; every term after the first contains a factor and tends to . So the limit is .
For the binomial expansion is no longer a finite sum, which is why Hardy asks for a device. Rationalise:
using the continuity of at . This agrees with for . The general rational case is proved in §119.
Prove that the derivative of is , and that of is .
Hint
Hint
Solution
By the identity in the hint,
As the first factor tends to , because cosine is continuous, and the second tends to . So the derivative is . Similarly, gives
Write down the tangent and the normal at a point of the parabola . Show that if , , the tangent is .
Hint
Solution
Here and . The tangent at is . When the normal is ; at the vertex the tangent is and the normal is .
With and the slope is , so the tangent is , that is, . Multiplying by gives , or .
Where does have a derivative, and what is it? Find a formula built from for a function with exactly the same derivative.
Hint
Solution
If for an integer , then for all small enough (say ), so for those and . At an integer , is discontinuous (it jumps from to ), so by the theorem of §113 there is no derivative. So is at every non-integer and undefined at every integer.
The function (Hardy’s example) is wherever and is undefined at the integers. Its derivative is therefore at every non-integer and undefined at the integers: exactly the same as . Two quite different functions can have the same derivative, a point that returns in §127.
For each of the curves and at the origin, describe how behaves as , and explain why neither gives a derivative at .
Hint
Solution
For , and , which is positive and tends to as from either side. The tangent is the -axis. There is no derivative, because the limit is not a finite number. (The function is continuous at , so this is a failure of a new kind, not caused by a discontinuity.)
For there are two functions, and , both defined only for . Only can be considered: gives and gives . No derivative exists, both because the quotient is unbounded and because can approach from one side only.
Check yourself
If is continuous at , then is differentiable at .
False. is continuous at but has a corner there; is continuous at and its chord slopes oscillate; is continuous at and has a vertical tangent. The true implication goes the other way: differentiable implies continuous.
For at , which statement is correct?
is for and for . A derivative needs a single limit from both sides. A lowest point does not force unless the derivative exists (that is Theorem C of §123). Averaging the two slopes is what the symmetric quotient does, which is why that quotient is not the definition. And is perfectly continuous.
Let for and . What is ?
, and . So even though oscillates: the oscillations are squeezed. Compare , where the quotient is not squeezed at all.
A particle is at position at time . Which quantity is its velocity at ?
Velocity at an instant is the limit of average velocities over shorter and shorter intervals; . The second option is the average velocity over one second, and the first is the average from time , assuming the particle started at .
Hardy’s language
The geometry of curves is merely one of many departments of mathematics in which the idea of a derivative finds an application.
Summary
- : the limit of chord slopes, taken from both sides, which must be a finite number.
- Geometrically is the slope of the tangent; dynamically it is an instantaneous velocity. Under magnification a differentiable graph looks straight.
- Differentiable implies continuous; continuous does not imply differentiable.
- Three ways to fail at a point of continuity: a corner (one-sided limits differ), oscillation (), a vertical tangent ().
- is itself a function, defined wherever the limit exists.