1

CURVATURE

E. L. Lady

The curvature of a curve is, roughly speaking, the rate at which that curve is turning.
Since the tangent line or the velocity vector shows the direction of the curve, this means
that the curvature is, roughly, the rate at which the tangent line or velocity vector is
turning.
There are two refinements needed for this definition. First, the rate at which the
tangent line of a curve is turning will depend on how fast one is moving along the curve.
But curvature should be a geometric property of the curve and not be changed by the
way one moves along it. Thus we define curvature to be the absolute value of the rate at
which the tangent line is turning when one moves along the curve at a speed of one unit
per second.
At first, remembering the determination in Calculus I of whether a curve is curving
upwards or downwards (“concave up or concave down”) it may seem that curvature
should be a signed quantity. However a little thought shows that this would be
undesirable. If one looks at a circle, for instance, the top is concave down and the bottom
is concave up, but clearly one wants the curvature of a circle to be positive all the way
round. Negative curvature simply doesn’t make sense for curves.

The second problem with defining curvature to be the rate at which the tangent line is
turning is that one has to figure out what this means.

The Curvature of a Graph in the Plane.

In the plane, the situation is clear. If ϕ is the angle between the tangent line and the
x-axis, then one defines the curvature to be

dϕ
κ=| |,
ds
where s is arc length. (I. e. s measures distance as one travels along the curve.)
dϕ ds dϕ ds
By the chain rule, = . Now write ν = . Then
ds dx dx dx
dϕ dϕ
| | = | | / ν.
ds dx
dϕ
Thus to find the curvature, it suffices to find and to find ν .
dx
 2
Z
Clearly s = ν dx . It one remembers the formula for arc-length, one can then
anticipate that
s
 2
dy
ν= 1+ .
dx

(Below, we will derive this formula.) The intuitive significance of ν is that it is the speed
at which a point travels along the curve
Z when its x-coordinate increases at a rate of one
unit/second. (Thus the formula s = ν dx says that in order to compute distance, one
integrates speed.)
Since ϕ is the angle between the direction in which the point on the curve is moving
and the direction of the x-axis (i. e. horizontal), one can see that ν = sec ϕ. Since tan ϕ
is the slope of the curve, i. e. tan ϕ = dy/dx , we get

 2
2 2 2 dy
ν = sec ϕ = 1 + tan ϕ = 1 + ,
dx

which is essentially the formula for ν anticipated above.

(An alternative explanation is to derive the formula ν = sec ϕ by starting with the
standard formula for arc length,
s
Z p Z  2
dy
s= dx2 + dy 2 = 1+ dx
dx

to see that s
 2 q
ds dy
ν= = 1+ = 1 + tan2 ϕ = sec ϕ.
dx dx

(Note that ϕ is by definition an acute angle, so sec ϕ ≥ 0. ))

The formula for the curvature of the graph of a function in the plane is now easy to
obtain. Since ϕ is the angle of the tangent line, one knows that tan ϕ is the slope the
curve at a given point, i. e.
dy
tan ϕ(x) = .
dx
Differentiating with respect to x yields (by the chain rule)

dϕ d2 y
sec2 ϕ = 2,
dx dx
 3

and so
dϕ y 00 y 00 y 00
= = =  2 ,
dx sec2 ϕ ν2 dy
1+
dx
and so
dϕ |y 00 | 1 |y 00 | |y 00 |
κ= = 2 = 3 = .
ds ν ν ν [1 + (y 0 )2 ]3/2

Curves in Parametric Form in the Plane. The formula for the curvature of a curve
in the plane described parametrically can easily be derived from the case just considered.
But it is more enlightening to start from scratch, since the principles thus derived can
then be adapted to the case of curves in three-space.
Given a curve r(t) , we will write as usual v(t) = r0 (t) . If we think of r(t) as being a
moving point, then v is the velocity vector, and the direction of the curve is the same as
the direction of v .
We let ν(t) = ||v(t)|| and write T(t) = v(t)/ ν(t) . Thus T(t) is a unit vector with the
same direction as v , and is usually called the unit tangent vector. The calculation of
the curvature depends on the following fact:

Theorem. If ϕ is the angle between v and the positive real axis, then d T/ dt is
orthogonal to T and
dT dϕ
= .
dt dt
Therefore if κ(t) is the curvature, then
1 dT
κ(t) = .
ν dt

proof: If we move T(t) to the origin, then since it is a unit vector, it becomes the
dT
radius vector for a point moving in a circle with radius 1. is the the velocity vector
dt
for this moving point, and thus is tangent to that circle, hence is perpendicular to T .
dT
Furthermore, is the speed at which T moves around that circle. Since the circle
dt
has radius 1, the angle ϕ of T is also the distance measured along the circumference, and
dϕ
since speed is the derivative of distance, this speed is thus .
dt
Now almost by definition,
1 dϕ 1 dT
k(t) = = . X
ν dt ν dt
 4

Second Proof For a less “conceptual” proof, notice that since T(t) is a unit vector with
an angle of ϕ to the polar axis, simple trigonometry yields

T(t) = cos ϕ(t) i + sin ϕ(t) j.

It follows from the chain rule that

dT
= ϕ0 (t)(− sin ϕ i + cos ϕ j ),
dt

and this vector is clearly orthogonal to T and has magnitude |ϕ0 (t)| , as claimed. X

This formula is in practice rather unwieldly. It requires one to differentiate T = v/ν ,

and ν is generally given by a rather messy square root. The following example illustrates
the difficulty.

Example. Consider the ellipse

r(t) = a cos t i + b sin t j .

We get
v = r0 = −a sin t i + b cos t j ,
and so
v −a sin t i + b cos t j
T= =p .
ν a2 sin2 t + b2 cos2 t
Calculating d T/ dt is quite ugly, namely

dT
= 12 (a2 sin2 t + b2 cos2 t)−3/2 (2a2 sin t cos t − 2b2 cos t sin t)(−a sin t i + b cos t j )
dt
+ (a2 sin2 t + b2 cos2 t)−1/2 (−a cos t i − b sin t j)

(a2 − b2 ) sin 2t (−a sin t i + b cos t j ) − 2ν 2 (a cos t i + b sin t j )

= ,
2ν 3
p
where ν = a2 sin2 t + b2 cos2 t .
The thought of finding the magnitude of this vector is likely to intimidate even the
most diligent adherent to the doctrine of brute force, although the fact that the two main
summands in the numerator of this fraction are orthogonal to each other simplifies things
somewhat. Even so, it takes considerable work to find the curvature in simplified form,
ab
which turns out to be k(t) = .
ν(t)3
 5

Parametric Curves in General. The purpose of deriving the above formula

1 dT
k(t) =
ν dt
for a curve in the plane given parametrically is not that it is the most practical way of
determining curvature in that case. In fact, we will be able to derive a much simpler
formula in that case. But this formula will depend on some insights that apply just as
well to the more difficult problem of determining curvature of curves in three-dimensional
or even higher dimensional space.
In 3-space, we cannot define curvature as ν −1 dϕ/dt , because a direction in 3-space
cannot be described by an angle. To define ϕ as the angle between the tangent line to the
curve and the z-axis, for instance, or the xy-plane will not work. This is clear when one
considers the fact that if ϕ is defined in this way, then ϕ is identically 0 for a horizontal
circle (or indeed any horizontal curve), but clearly one does not want the curvature to
be 0 in this case.
For a curve in 3-space, we still want the curvature to be equal to the rate at which the
tangent to the curve turns as one moves along the curve at a speed of one, but we need to
find a way of defining what this means. What is needed is to take the angle between the
tangent vectors to the curve at two nearby points r(t) and r(t + h) and then divide by
the distance between these two points. The exact curvature will then be the limit of this
ratio as h approaches 0.
More practically, we can write
1 αt (h)
k(t) = lim ,
ν h→0 t
where αt (h) is the angle between the two vectors r(t) and r(t + h) .
Two points to note here about αt (h) . First of all, since vectors do not have any
fixed location, it does make sense to talk about the angle between these two vectors. For
practical purposes, this means that we move the vectors r0 (t) and r0 (t + h) to a common
location and then measure the angle between them.
Secondly, we will consider this angle to always be positive. In fact, whereas in
the plane we consider angles to be positive if measured clockwise and negative if
counter-clockwise, in 3-space there is no consistent way of assigning a sign to an angle.
If we imagine an airplane flying along the curve, and imagine that the pilot has a way
of keeping track of the direction he is flying in in space (or, more likely, this determination
could be made on the ground by a control tower), then the pilot can determine his
curvature by measuring the angle between his directions at two nearby points, and then
dividing by the distance traveled (i. e. dividing by νh, where h is the elapsed time).
Needless to say, this “airplane method” is not very useful to a calculus student.
 6

The key to determining the curvature of a curve given in parametric form is to notice
that the acceleration vector v0 (t) is the sum of two orthogonal components, one of which
shows how fast the speed is changing, and the other shows how fast the direction of the
curve is turning.
More generally, the following is true for any vector function.

v(t)
Let v(t) be any vector function of t (time), let u(t) = (the unit
||v(t)||
vector with the same direction as v ), and let v0 (t) denote the derivative of v
with respect to t . Then v0 can be written as the sum of two components, one
of which has the same direction as v (or the opposite direction) and shows
the rate at which the magnitude of v is increasing or decreasing. The other
component is orthogonal to v and pointing in the direction that v is turning,
and its magnitude is the product of ||v|| and the rate at which v is turning.

So for instance if v(2) = 5 cm and v is decreasing at a rate of .2 cm/sec and

turning at a rate of π/8 radians/sec when t = 2 , then v0 (t) is the sum of the vector
−.2u plus a vector in the direction towards which which v is turning with a magnitude
of 5π/8 cm/sec.
The above theorem is simply the product rule applied to the equation
v
v(t) = ||v|| = ||v(t)|| u(t) ,
||v||

which yields
d ||v||
v0 (t) = u + ||v|| u0 (t) .
dt
(Applied to the case when v is the velocity vector for a curve, this becomes the formula

dν dT
a(t) = v0 (t) = T+ν
dt dt
giving the the tangential and normal components of the acceleration vector.) Clearly
d ||v||
u is a vector parallel to v which shows how fast ||v|| is increasing or decreasing.
dt
So the only thing we need to see is that the magnitude of u0 (t) is the rate at which v(t)
is turning, which is, of course, the same as the rate at which u(t) is turning.
In the plane, we derive this by noticing (as was done above for the case u(t) = T(t) )
that if the unit vector u(t) is moved to the origin then it becomes the position vector for
 7

a point moving in a circle with radius one. If we write, as previously, ϕ for the angle u
makes with the x-axis, then u(t) = cos ϕ i + sin ϕ j and it is easy to see that ||du/ dt||
is the speed at which this point moves around that circle, and that this is the same as
|dϕ/ dt| .
If we look at the reasoning involved more carefully, we see that the fact that the tip
of the unit tangent vector u(t) moves in a circle is not crucial. In general, let αt (h) be
the angle between u(t) and u(t + h) . If we draw a circular arc from the tip of u(t) to
u(t + h) (where we have located both these vectors at the origin, and also make the
origin the center of this circular arc) then because the radius of this arc has length 1
(because u(t) is a unit vector), the length of the arc will be equal to the angle between
these two vectors, i. e. to αt (h) . But when h is very small, so that these two vectors
are very close to each other, then this arc is very close to a straight line, so that it’s
length αt (h) is almost the same as the distance between the tips of the two vectors,
i. e. to ||u(t + h) − u(t)|| . Thus we get

αt (h) ||u(t + h) − u(t)|| du

lim = lim = ,
h→0 h h→0 h dt

du
which just says that is the rate at which u(t) is turning (and therefore also the
dt
rate at which v(t) is turning, since they both have the same direction).

Returning to the case where v(t) is the velocity vector for a curve, the corresponding
unit vector is T(t) , and we see that ||T0 (t)|| is the rate at which the curve is turning,
which, as we have seen, is ν(t)k(t) , where ν is the speed and k is the curvature. Thus we
recover the formula established previously for the case of a plane curve:

1 dT
k(t) = .
ν dt

As the example of the ellipse in the plane showed, it is usually not practical to
compute dT/ dt directly. A less painful approach is to use the formula derived above for
the tangential and normal components of the acceleration vector. Recall that by using the
product rule to differentiate the formula v = νT , we derived

a = v 0 = ν 0 T + ν T0 ,

where ν 0 T is parallel to v , and T0 is orthogonal to T (and thus orthogonal to v ).

Furthermore, as just seen, ||T0 || = νk, where k(t) is the curvature.
 8

Let
T0 T0
N= = ,
||T0 || νk(t)
using the formula
1
k(t) = ||T0 || ,
ν
as shown above. Then N is a unit vector perpendicular to the direction of the curve at
the location r(t) . N is called the unit normal to the curve at r(t) .
We now have
a(t) = v0 (t) = ν 0 (t)T(t) + ν 2 (t)k(t)N.
This expresses the acceleration vector as the sum of a vector in the direction of the curve
(the tangential component) with magnitude ν 0 (t) and a component orthogonal to the
curve (the normal component) having magnitude ν 2 (t)k(t). (The normal component is
proportional to the so-called “centrifugal force” that someone riding along the curve will
feel. The formula above shows that doubling one’s speed will quadruple this centrifugal
force — an important consideration for riders of motorcycles.)
Now since N · T = 0 and N · N = 1 , from the equation above we get that

N · a = ν 0 N · T + ν 2 k N · N = ν 2 k,

yielding the following formula for curvature:

N·a
k(t) = .
ν2

This finally yields a reasonable formula for a curve in the plane given parametrically.

Theorem. If r(t) = x(t) i + y(t) j is a plane curve, the the curvature is given by

|y 0 x00 − x0 y 00 |
k(t) = ,
ν3
p
where ν(t) = x0 (t)2 + y 0 (t)2 .

proof: In the plane, it is easy to find the unit normal N simply from the fact that
N ⊥ v and ||N|| = 1 , since in the plane there are only two vectors with length 1
perpendicular to any given vector. Now v = x0 (t) + y 0 (t) and one sees immediately that

v ⊥ y 0 i − x0 j
 9
p
and ||y 0 i − x0 j|| = (y 0 )2 + (x0 )2 = ν . Therefore
y 0 i − x0 j
N=± ,
ν
and
(y 0 i − x0 j) · (x00 i + y 00 j)
N·a = ±
ν
|y 0 x00 − x0 y 00 |
= .
ν
(We have seen above that N · a ≥ 0 , and of course ν ≥ 0 . Therefore the absolute value.)
Finally we get
N·a |y 0 x00 − x0 y 00 |
k(t) = = . X
ν2 ν3

Example. Consider again the ellipse

r(t) = a cos t i + b sin t j.
We get
x0 = −a sin t y 0 = b cos t
x00 = −a cos t y 00 = −b sin t
so that
|y 0 x00 − x0 y 00 | | − ab cos2 t − ab sin2 t| ab
k(t) = 3
= 3
= 3.
ν ν ν

Now in 3-space, we can’t find N so cheaply. But in 3-space, we have the cross-product
available.
Consider again the equation a(t) = v0 (t) = ν 0 (t) T(t) + ν 2 (t)k(t) N. We want a
way to isolate the second summand on the right hand side. To do this, note that since N
and T are orthogonal unit vectors, the cross product N × T has length 1. On the other
hand, the cross product of any vector with itself is always 0. Thus we get
T × a = ν0 T × T + ν2k T × N = ν2k T × N
and
||v × a|| = ν ||T × a||
= ν 3 k ||T × N|| = ν 3 k.

This yields the following theorem:

Theorem. The curvature of a curve r(t) in three-space is given by the formula

||v(t) × a(t)||
k(t) = .
ν3