UNIVERSITY OF BABYLON

COLLEGE OF ENGINEERING
DEPARTMENT OF CIVIL ENGINEERING

Mathematics Report

B
Lengths of Plane Curves
For a general curve in a two-dimensional plane it is not clear exactly how to measure its
length. In everyday physical situations one can place a string on top of the curve, and then
measure the length of the string when it is straightened out, noting that the length of the
string is the same whether it is wound up or not. Unfortunately, we have no means of run-
ning a string over an arbtirary curve, one that we might not even be able to sketch. Instead,
we need to use the notion of approximation, and use a limit to make the approximation as
accurate as we would like. The simplest means of approximating a curve is using straight
line segments. As we increase the number of segments, they begin to hang closer and closer
to the curve, and in the limit that the number of segments approaches infinity, we find the
exact length of the curve.

Let L denote the length of a curve C, and correspondingly let ds represent the length of
a segment of infinitessimal length of the curve. If we sum up the lengths of all of these
infinitessimal pieces, then we will find the entire length of the curve. This is conveniently
expressed in integral notation as Z
L= ds
C
which can be interpreted as saying the length of a curve is given the sum of the length
of infinitessimal segments of the curve. Although this notation conveniently represents the
problem, it does not tell us how to find its solution. What we need to do is represent the
differential ds in terms of the variables x and y. If we draw a line segment between two
points of the curve, the length of the segment will be the hypotenuse of a triangle with sides
∆x and ∆y which represent the function’s change in the x and y coordinates. If we let ∆s
denote the length of the hypotenuse, using the Pythagorean theorem we find
(∆s)2 = (∆x)2 + (∆y)2
or p
∆s = (∆x)2 + (∆y)2
Next we note that as the lengths of ∆x and ∆y tend to 0, the length of the hypotenuse of
this triangle becomes an increasingly accurate estimate of the length of the change of the
curve. Thus, when we consider infinitessimal changes in x and y, we find that
p
ds = dx2 + dy 2
which gives us a means of representing ds in terms of known quantities. However, we are
still not yet finished. We now need a means of representing the differentials dx and dy. The
way we represent these quantities depends on the form the curve is represented in. If we
have a differentiable function y(x), then we can write
dy
dy = dx
dx
and r r
 dy 2  dy 2
ds = dx2 + dx2 = 1+ dx
dx dx
In order to find the length of the curve we would simply integrate over the appropriate
limits in x, so if we want to find the length of the function as x varies from a to b, we would
calculate Z Z br  dy 2
L= ds = 1+ dx
C a dx
Similarly, if we had a differentiable function x(y) then we would represent

dx
dx = dy
dy
and s s
 dx 2  dx 2
ds = dy 2 + dy 2 = 1+ dy
dy dy
Then if we wanted to find the length as y varied from c to d, we would simply calculate
Z Z ds  dx 2
L= ds = 1+ dy
C c dy

Finally, it is worth noting that the number of curves we can represent as functions y(x)
or x(y) is rather limited. By looking to parametrized curves we could greatly increase the
number of curves for which we can calculate the length.

Consider the following examples.

Example 1 Find the length of the curve y = x3/2 from 0 ≤ x ≤ 2.

Solution Here we have y as a function of x, so we want to represent ds in terms of x. We
find that
dy 3
= x1/2
dx 2
so  dy 2 9
= x
dx 4
Now evaluating the appropriate integral
Z 2r
9
L= 1 + xdx
0 4
which we solve through substitution, letting
9
u=1+ x
4
and
9 4du
du = dx or = dx
4 9
and so u(0) = 1 and u(2) = 1 + 92 = 11
2
. Thus,
Z 2r
4 11/2 √
Z
9 8 3/2 11/2 8 h 11 3/2 i
L= 1 + xdx = udu = u  = · − 1 ≈ 3.5255
0 4 9 1 27 1 27 2

Example 2 Find the length of y = (x/2)2/3 from x = 0 to x = 2.

Solution When we find
dy 2  x −1/3  1  1  2 1/3
= · =
dx 3 2 2 3 x
we notice that this derivative does not exist for x = 0. In fact, this function grows without
bound as x → 0, which means that it is not Riemann integrable (only bounded functions are
Riemann integrable). Thus, we cannot directly calculate the length of this curve. However,
can alternative represent x as a function of y, and find
 x 2/3
y =
2
3/2 x
y =
2
x = 2y 3/2

which we have previously seen is differentiable. When x = 0 we have y = 0, and when

x = 2 we have y = 1, so our limits of integration will be for y from 0 to 1. Calculating the
derivative
dx 3
=2 y 1/2 = 3y 1/2
dy 2
and  dx 2
= 9y
dy
Thus, Z 1 p 1 2 1
3/2  2 √
L= 1 + 9ydy = · (1 + 9y)  = (10 10 − 1) ≈ 2.27
0 9 3 0 27
Example 3 Find the length of x = (y 3 /6) + 1/(2y) from y = 2 to y = 3.
Solution We begin by finding the derivative

dx y2 1
= − 2
dy 2 2y
and  dx 2 1
= (y 4 − 2 + y −4 )
dy 4
Thus,
3
r Z 3r
1 3p 2
Z Z
1 4 −4
1 4 −4
L = 1 + (y − 2 + y )dy = (y + 2 + y )dy = (y + y −2 )2 dy
2 4 2 4 2 2
1 3 2 h y 3 1 3 1 h 27 1   8 1 i 13
Z
1
= (y + y −2 )dy = −  = − − − =

2 2 2 3 y 2 2 3 3 3 2 4
The key to solving the above problem was the fact that
 dx 2
1+
dy
was a perfect square, so it canceled out with the square root. However, most curves do not
work out so nicely. Unfortunately, we cannot evaluate the vast majority of integrals that
arise in finding arc length, and most of them cannot be evaluated by hand. The following is
an example using an integral we cannot calculate.

Example 4 Find the circumference of the circle x2 + y 2 = a2 .

Solution This curve is not given by a single function, but two functions. In order to find
the circumference we could calculate the length of the cuves given by each function, but
alternatively we can simply exploit the symmetry of the two functions - they both have the
same length. If we find the length of one of the curves, we simply double it to find the
circumference of the circle. Here we have the choice of calculating the length as an integral
in x or y. First, let us write √
y = a2 − x2
so
dy 2x
= −√
dx a2 − x2
and  dy 2 4x2
=
dx a2 − x2
Finally, r
a
4x2
Z
L= 1+ dx
−a a2 − x2
which we do not how to evaluate, but we know the result is L = π.
1.3. CURVATURE AND PLANE CURVES 7

1.3 Curvature and Plane Curves

We want to be able to associate to a curve a function that measures how much the curve
bends at each point.
Let α : (a, b) → R2 be a curve parameterized by arclength. Now, in the Euclidean plane
any three non-collinear points lie on a unique circle, centered at the orthocenter of the
triangle defined by the three points.
For s ∈ (a, b) choose s1 , s2 , and s3 near s so that α(s1 ), α(s2 ), and α(s3 ) are non-
collinear. This is possible as long as α is not linear near α(s). Let C = C(s1 , s2 , s3 )
be the center of the circle through α(s1 ), α(s2 ), and α(s3 ). The radius of this circle is
approximately |α(s) − C|. A better function to consider is the square of the radius:

ρ(s) = (α(s) − C) · (α(s) − C).

Since α is smooth, so is ρ. Now, α(s1 ), α(s2 ), and α(s3 ) lie on the circle so ρ(s1 ) =
ρ(s2 ) = ρ(s3 ). By Rolle’s Theorem there are points t1 ∈ (s1 , s2 ) and t2 ∈ (s2 , s3 ) so that
ρ0 (t1 ) = ρ0 (t2 ) = 0. Then, using Rolle’s Theorem again on these points, there is a point
u ∈ (t1 , t2 ) so that ρ00 (u) = 0. Using Leibnitz’ Rule we have ρ0 (s) = 2α0 (s) · (α(s) − C) and

ρ00 (s) = 2[α00 (s) · (α(s) − C) + α0 (s) · α0 (s)].

Since ρ00 (u) = 0, we get

α00 (u) · (α(u) − C) = −α0 (s) · α(s) = −1.

Now, as s1 , s2 , and s3 get closer to s, then the center of the circles will converge to a
value Cα (s). Then t1 and t2 go to s, so ρ0 (s) = 0 which forces α0 (s) · (α(s) − Cα (s)) = 0.
Furthermore, α00 (s) · (α(s) − Cα (s)) = −1.
This says that the circle centered at Cα (s) with radius α(s) − Cα (s) shares the point
α(s) with the curve α. Furthermore, from the above the tangent to the circle at α(s) is a
multiple of α0 (s). Thus, this circle, called the osculating circle, is tangent to the curve at
α(s). The point Cα (s) is called the center of curvature of α at s, and the curve given by
the function Cα (s) is called the curve of centers of curvature.

Definition 1.5 The (unsigned) plane curvature of α at s is the reciprocal of the radius
of the osculating circle:
1
κ± (s) = |.
|α(s) − Cα s

Theorem 1.4 κ± (s) = |α00 (s)|.

Proof: Since α0 (s) · α0 (s) = 1, differentiating gives α0 (s) · α00 (s) = 0. This means that
α00 (s) is perpendicular to α0 (s). Since we have seen that α(s) − Cα (s) is also perpendicular
to α0 (s), there exists a k ∈ R so that

α(s) − Cα (s) = kα00 (s).

8

Figure 1.1: Tractrix curve

From above we have

−1 = α00 (s) · (α(s) − Cα (s))

= α00 (s) · kα00 (s)
= k|α00 (s)|2 .

Thus,
1 1
|α(s) − Cα (s)| = |k| |α00 (s)| = |α00 (s)| = .
|α00 (s)|2 |α00 (s)|

We rarely can symbolically represent a curve as parameterized by arclength. Quite

often, a different parameterization is more reasonable. To find the curvature, though,
would require that we parameterize by arclength and then differentiate. There is an easier
way.

Theorem 1.5 The plane curvature of a regular plane curve σ(t) = (x(t), y(t)) is given by
 x00 y 0 − y 00 x0 
 
κ± (t) =  0 2
 .
((x ) + (y 0 )2 )3/2 

1.3.1 Tractrix
Describe the curve followed by a weight being dragged on the end of a fixed straight length
and the other end moves along a fixed straight line. The tractrix is the curve characterized
by the condition that the length of the segment of the tangent line to the curve from the
curve to the y-axis is constant. It has the following equation for a given constant a:
p
a + a2 − y 2 p
x = a ln( ) − a2 − y 2 .
y
and has graph shown in Figure 1.1.
Let the curve begin at (a, 0) on the x-axis. Now, we can see that
√
y0 dy a2 − x2
= = . (1.1)
x0 dx x
1.3. CURVATURE AND PLANE CURVES 9

Square both sides of the equation and simplify

 a 2
(x0 )2 + (y 0 )2 = (x0 )2 . (1.2)
x
Now, if we differentiate the first equation (1.1), we get

x0 y 00 − x00 y 0 −a2 x0
= √ (1.3)
(x0 )2 x 2 z 2 − x2
a2 (x0 )3
x00 y 0 − x0 y 00 = √ . (1.4)
x2 a2 − x2
Thus,
 −a2 x0 3
   
x   x 
κ± (x, y) =  √ 3 0 3
= √ . (1.5)
x2 z 2 − x2 a (x )   a a2 − x2 
Of course, we can integrate Equation 1.1 to get
Z √ 2
a − x2
y(x) = dx (1.6)
x
A change of variables of the form x = a sin(t) gives:

σ(t) = (a sin(t), a ln(tan(t/2)) + a cos(t)),

 
 tan(t) 
which gives the plane curvature as κ± (t) =  .
a 
 a 2
Also, to parameterize the tractrix by arclength, we need (x0 )2 +(y 0 )2 = 1, thus (x0 )2 =
x
1, which gives x0 = ± a1 x. Let’s take a = 1 and consider just the case x0 = x. Then, x(s) = es
from which it follows that
√
dy 1 − x2 dx p
= = 1 − e2s
ds p x ds
y(x) = 1 − e2s − arccosh(e−s ).

This requires that 0 ≤ e2 s ≤ 1. Take the curve traced out in the opposite direction by
replacing s by −s. The parameterization is now:
p
σ(s) = (e−s , 1 − e−2s − arccosh(es )), s ≥ 0.

For a = 1 we have the plane curvature:

e−s
κ± (s) = |σ 00 (s)| = √ .
1 − e−2s
Let α : (a, b) → R2 be a curve. The reverse curve is α̂ : (a, b) → R2 is given by
α̂(t) = α(b − t). We wish to distinguish between these two curves.

Definition 1.6 Let e1 , e2 denote the standard basis vectors in R2 . An ordered pair of
vectors [u, v], u, v ∈ R2 is said to be in standard orientation if the matrix representing
the transformation from [u, v] to [e1 , e2 ] has a positive determinant.
10

If α(s) is a regular curve parameterized by arclength, then the unit tangent vector is
T(s) = α0 (s). Let N(s) denote the unique unit vector perpendicular to T(s) with standard
orientation [T(s), N(s)]. N(s) is the unit normal vector to α at s. Since T(s) is a unit
vector, we see that T(s) · T0 (s) = 0. Thus, α00 (s) = T0 (s) must be a multiple of N(s).
Definition 1.7 The directed curvature κ(s) of a unit-speed curve α is given by the
identity
α00 (s) = κ(s)N(s).
Note that since N(s) is a unit vector, we see that |κ(s)| = |α(s)| = κ± (s).
Theorem 1.6 (Fundamental Theorem for Plane Curves) Given any continuous func-
tion κ : (a, b) → R, there is a curve σ : (a, b) → R2 , which is parameterized by arclength,
such that κ(s) is the directed curvature of σ at s for all s ∈ (a, b). Furthermore, any other
curve σ̄ : (a, b) → R2 satisfying these conditions differs from σ by a rotation followed by a
translation.
The proof of this is a very neat, simple proof which uses differential equations.
Proof: From the theorem, we have a function f : (a, b) → R2 written as f (s) = (f1 (s), f2 (s))
satisfying the following system of differential equations:
(f10 (s), f20 (s)) = κ(s)(−f2 (s), f1 (s)),
subject to f (c) = u and |u| = 1
Note that if f is a solution to this differential equation, then it is a unit-speed curve
because
d 2
(f (s) + f22 (s)) = 2f1 (s)f10 (s) + 2f2 (s)f20 (s)
ds 1
= 2(f1 (s), f2 (s)) · (f10 (s), f20 (s))
= 2κ(s)(f1 (s), f2 (s)) · (−f2 (s), f1 (s)) = 0
Thus, |f (s)| is a constant and since |f (c)| = 1, |f (s)| = 1 for all s ∈ (a, b).
Lemma 1.1 If g(t) is a continuous (n × n)-matrix-valued function on an interval, then
there exist solutions, F : (a, b) → Rn , to the differential equation F 0 (t) = g(t)F (t).
Applying this lemma, we have a function g(s) given by
 
0 −κ(s)
g(s) =
κ(s) 0
The equation T0 (s) = κ(s)N(s) becomes T0 (s) = g(s)T(s). Thus, the above lemma gives us
the function T(s) for the curve σ(s) with the correct curvature. To find the curve σ(s) we
only need to integrate T(s). We can choose σ(c) to be any point in R2 and we can choose u
to be any unit vector in R2 . Changing u at σ(c) involves a rotation. That rotation passes
through the differential equation so that another solution would appear as T(s) = ρθ T(s),
where ρθ is a rotation matrix. A translation resets the point σ(c) to be any point in R2 .
Thus a second solution σ(s) must satisfy
σ(s) = ρθ σ(s) + ω0 .
This proves the theorem.