After areas, lengths. The formula is familiar: the length of y=ϕ(x) from a to b is ∫ab1+ϕ′(x)2dx. But Hardy is frank that he cannot prove it at this stage, and the reason is instructive. For area, assuming that regions have areas with the obvious properties was enough to derive everything. For length, the analogous assumption is not enough: even the continuity of the arc length is not obvious. The section therefore does two things: it shows what the formula must be, and it makes clear which extra assumption is used.
The idea
The natural way to measure a curve is with a ruler: mark points on it, join them by chords, and add up the chord lengths. The finer the marking, the closer the polygon hugs the curve, and its length should approach the length of the curve.
But “hugging the curve” is not enough. A staircase of horizontal and vertical steps can lie as close to the curve as we like, and yet its length never changes: it is always the total horizontal run plus the total vertical rise. So which approximations converge to the length is a real question.
ExploreChords versus staircases
Try this.Increase the number of pieces n and watch the polygon length approach the value of the formula. Look at the ratio arc/chord for the worst piece. Then switch on the staircase and increase n again. Try the quarter circle too.
y=41x2,0≤x≤3
xy
the curve
inscribed polygon
Polygon with n=4 chords: length 3.88916. Formula ∫1+ϕ′(x)2dx=23413+log(23+413)=3.89893. Largest ratio arc/chord over the pieces: 1.0054.
What to notice
The chord polygons approach the formula’s value from below (an inscribed polygon is never longer than the arc), and the worst ratio arc/chord tends to 1: this is Hardy’s assumption lim{PP′}/PP′=1. The staircase gets as close to the curve as you like but its length is stuck at (b−a)+∣ϕ(b)−ϕ(a)∣ for a monotonic curve. For the quarter circle it stays at 2 while the arc is 21π≈1.571.
Precise statements
TheoremLength of a curve (assuming lengths exist)Hardy: §149
Let ϕ have a continuous derivative. Suppose that the arc P0P of y=ϕ(x) from x=0 has a length S(x), and that the ratio of a small arc to its chord tends to 1:
limh→0{PP′}/PP′=1, where P′ is the point with abscissa x+h. Then
S′(x)=1+{ϕ′(x)}2,S(x)=∫1+{ϕ′(x)}2dx.
The same reasoning applied to a curve x=ϕ(t), y=ψ(t), or to r=f(θ), gives
dtdS=(dtdx)2+(dtdy)2,dθdS=r2+(dθdr)2.
The polar form is Hardy’s Ex. LIV. 18. These integrals are often not elementary: the arc of an ellipse x=asint, y=bcost leads to a∫1−e2sin2tdt (e the eccentricity), an elliptic integral, which cannot be expressed by the functions at our disposal (Ex. LIV. 16).
Proofs
ProofLength of a curve
Strategy
Write the difference quotient of S as (arc/chord) × (chord/h). The first factor tends to 1 by assumption. The second is computed from Pythagoras and the mean value theorem.
Split the quotient
With P=(x,ϕ(x)) and P′=(x+h,ϕ(x+h)), the increase S(x+h)−S(x) is the length {PP′} of the arc (for h>0; for h<0 swap the roles of P and P′). So
hS(x+h)−S(x)=PP′{PP′}⋅hPP′.
The chord by Pythagoras
With k=ϕ(x+h)−ϕ(x), the chord is PP′=h2+k2=∣h∣1+k2/h2.
Mean value theorem
k=hϕ′(ξ) for some ξ between x and x+h (§126). So PP′/∣h∣=1+ϕ′(ξ)2.
Let h→0
As h→0, ξ→x, and since ϕ′ is continuous, 1+ϕ′(ξ)2→1+ϕ′(x)2. With the assumption {PP′}/PP′→1, the quotient tends to 1+ϕ′(x)2. (For h<0 both the arc and h change sign, and the limit is the same.)
∎
Where is the weak point? It is the assumption that {PP′}/PP′→1, together with the existence of S(x) at all. Both need a definition of length. The modern one is exactly the chord polygon of the widget: the length of a curve is the least upper bound of the lengths of all inscribed polygons, and a curve is rectifiable when this is finite. With that definition, and ϕ′ continuous, one can prove the formula; a continuous curve need not be rectifiable at all (the graph of xsin(1/x) near 0 has infinite length: between consecutive zeros 1/{(k+1)π} and 1/(kπ) it climbs to a height of roughly 1/{(k+21)π} and comes back, and ∑1/k diverges).
Worked examples and exercises
ExerciseAn arc of a parabola●●○Hardy: Examples LIV, 1
Find the length of the arc of y=x2/4a from the origin to x=ξ.
Hint
1+ϕ′(x)2=1+x2/4a2=2a1x2+4a2.
Hint
Use ∫x2+α2dx=21xx2+α2+21α2log{x+x2+α2} with α=2a.
For a=1, ξ=3 this is 4313+log23+13≈3.8989, the value in the widget.
ExerciseA curve with an elementary length●○○Hardy: §149
Find the length of y=x3/2 from x=0 to x=4.
Hint
1+ϕ′(x)2=1+49x.
Solution
∫041+49xdx=[278(1+49x)3/2]04=278(103/2−1)≈9.073, using the linear change of variable. Curves for which 1+ϕ′2 is a perfect square, or simple like this, are the exception rather than the rule.
ExerciseA circle in polar coordinates●○○Hardy: Examples LIV, 18 (i)
Find the area and the perimeter of the circle r=2asinθ, 0≤θ≤π.
Hint
r2+(dr/dθ)2=4a2sin2θ+4a2cos2θ.
Solution
Perimeter: r2+r′2=2a, so the length is 2a⋅π=2πa. Area: 21∫0π4a2sin2θdθ=2a2⋅21π=πa2. This is a circle of radius a through the origin, as it should be.
ExerciseThe cycloid●●○Hardy: §149; the cycloid of Examples LIV, 17
For the cycloid x=a(t+sint), y=a(1+cost), find the length of the arch −π≤t≤π.
Hint
x˙2+y˙2=a2{(1+cost)2+sin2t}=2a2(1+cost). Use 1+cost=2cos221t.
Solution
x˙2+y˙2=2a∣cos21t∣=2acos21t for ∣t∣≤π. An integral is 4asin21t, which increases by 4a−(−4a)=8a. The arch of a cycloid is four times the diameter of the rolling circle, a result of Wren (1658).
ExerciseThe ellipse leads to a new kind of integral●○○Hardy: Examples LIV, 16
Show that the arc of the ellipse x=asint, y=bcost (a>b) from t1 to t2 is the increase of a∫1−e2sin2tdt, where e2=1−b2/a2.
Hint
Compute x˙2+y˙2 and write cos2t=1−sin2t.
Solution
x˙2+y˙2=a2cos2t+b2sin2t=a2−(a2−b2)sin2t=a2(1−e2sin2t). So dS/dt=a1−e2sin2t. For 0<e<1 none of the methods of this chapter applies, and in fact the integral is not an elementary function. This is the origin of the name “elliptic integral”.
Check yourself
Check yourselfTrue or false?
If a sequence of paths gets uniformly closer and closer to a curve, their lengths tend to the length of the curve.
Why
False. The staircases in the widget converge to the curve uniformly, but their lengths are constant, and different from the curve’s length. Length is not continuous under this kind of convergence. That is why length is defined through inscribed polygons, whose vertices lie on the curve.
Check yourselfEnter a number
Compute the length of y=1−x2 from x=0 to x=1 from the formula (three decimals).
Why
1+ϕ′2=1+1−x2x2=1−x21, so the length is arcsin1−arcsin0=21π≈1.571. The integrand is unbounded at x=1; the formula is still right, as a limit.
Check yourselfChoose one
Which ingredient of Hardy’s derivation of S′(x)=1+ϕ′(x)2 is an assumption, not proved in Chapter VI?
Why
The chord length is Pythagoras, the second is the mean value theorem, and the fourth follows from continuity of ϕ′. That arcs have lengths, and that arc and chord are asymptotically equal, are exactly what a definition of length must supply.
Hardy’s language
Indeed it is not possible to proceed further, with any degree of rigour, without a careful analysis of just what is meant by the length of a curve.
— G. H. Hardy, §149
Summary
Summary
Length is measured by inscribed polygons; arbitrary approximating paths (staircases) can have the wrong length.
If arcs have lengths and arc/chord →1, then S′(x)=1+ϕ′(x)2; the proof is Pythagoras plus the mean value theorem.
Parametric form ∫x˙2+y˙2dt; polar form ∫r2+r′2dθ.
Few curves have elementary lengths; the ellipse already leads to a non-elementary (elliptic) integral.