MultiVarCalc MinorCorrectionTo p57 20 01 2014 PDF

Prelims.
Multivariable Calculus
Eamonn Gaffney.
Using extensive material from lecture notes by Richard Earl
and also material from lecture notes by Ruth Baker.
Hilary Term 2013

Recommended Texts
• Erwin Kreyszig, Advanced Engineering Mathematics (Wiley, 8th Edition, 1999).
• D. E. Bourne & P. C. Kendall, Vector Analysis and Cartesian Tensors (Stanley Thornes,
1992).
• H. M. Schey, Div, Grad and Curl and all that (W. W. Norton, Third Edition, 1996).
Further Reading
• Mary L. Boas, Mathematical Models in the Physical Sciences (Wiley 2nd Edition, 1983).
• Gerald B. Folland Advanced Calculus (Prentice Hall, 2002)
Website
https://www.maths.ox.ac.uk/courses/material
2
1. MULTIPLE INTEGRALS IN 2D
1.1 Preliminaries
Definition 1 By a scalar field φ on R3 we shall mean a map φ : R3 → R.
Definition 2 By a vector field F on R3 we shall mean a map F : R3 → R3 .
• We will typically assume that scalar and vector fields are smooth – their partial derivatives
exist with respect to x, y and z to all orders – though for brevity this will not always be
stated.
• Occasionally we may consider more general scalar fields φ : Rn → R and vector fields
F : Rn → Rm .
3
Example 3 The three co-ordinates x, y, z are each scalarp fields on R . The position vector
r = (x, y, z) is a vector field; its magnitude r = |r| = x2 + y 2 + z 2 is a scalar field, though
note it is not smooth at (0, 0, 0) .
We shall also consider scalar and vector fields defined on proper subsets of R3 (or more generally
Rn ). The domains of these fields will usually be open, so that we can define their partial
derivatives.
Definition 4 A set U ⊆ Rn is said to be open if for every x ∈ U there exists ε > 0 such that
" n #1/2
X
n 2
B (x; ε) = {y ∈ R : |y − x| < ε} ⊂ U , where |y − x| = (yi − xi ) ,
i=1
where xi , yi are the ith components of x, y. We refer to B (x; ε) as the open ball of radius
ε about x.
1.1.1 Integrals in one dimension
An informal definition of the integral

Z b
f (x)dx, (1.1)
a
MULTIPLE INTEGRALS IN 2D 3
would be as follows. Suppose f : [a, b] → R is a function. Subdivide [a, b] into m sub-intervals
of equal length δx and let x1 , . . . , xm be points in the respective intervals. On partitioning with
smaller and smaller intervals by taking the limit δx → 0, we have
Z b m
X
f (x)dx = lim f (xr )δx, (1.2)
a δxr →0
r=1
provided the limit exists and is independent of the manner of subdivision. This will always be
the case if f is continuous.
4 MULTIPLE INTEGRALS IN 2D
1.2 Multiple Integrals in Two Dimensions: A Brief Intro-
duction
Figure 1-1 A region R for a double integration, with a square element of area δxδy.
Consider a region of the plane, R, such as depicted above, together with a scalar field ψ(x, y).
• Partition the region into N square elements of equal area δA = δxδy; difficulties with
boundary elements extending outside the region R will disappear below.
• Suppose the scalar field ψ(r) takes the value ψi at the centre of the ith element, i ∈
{1, . . . N }.
• Then, on partitioning with smaller and smaller squares, by taking the limit δA → 0, we
have
N N ZZ ZZ
def
X X
lim ψi δA = lim ψi δxδy = ψdA = ψdxdy. (1.3)
i=1 i=1 R R
noting the region with boundary elements extending outside the region R yields no con-
tribution to the integral in the limit.
• This is an informal definition of a double integral; a rigorous approach would, for

example, verify the limit is independent on the details of the elements used to divide up
MULTIPLE INTEGRALS IN TWO DIMENSIONS: A BRIEF INTRODUCTION 5

R. If ψ is piecewise continuous and R is a suitable region (we only ever consider such
cases) there are no difficulties. In generality, which is not the objective here, complexities
can emerge e.g. in three dimensions, The Banach-Tarski paradox.
RR R
Note If ψ(r) = 1 for all r then the integral R
ψdA = R
dxdy gives the area of R.
Properties of double integrals The following properties
R are inherited from integration with
respect to one variable, i.e. integrals of the form f (x)dx..
• Linearity. Let a, b be constants.

ZZ ZZ ZZ
af (x, y) + bg(x, y)dA = a f (x, y)dA + b g(x, y)dA.
R R R
ZZ ZZ
• Order. If f (x, y) ≥ g(x, y) for all (x, y) ∈ R then f (x, y)dA ≥ g(x, y)dA.
R R
• Domain Splittng. If R = R1 ∪ R2 , with R1 ∩ R2 the empty set, then

ZZ ZZ ZZ
f (x, y)dA = f (x, y)dA + f (x, y)dA.
R R1 R2
Interpretation. We can view the integral as the weight of a plate R in Fig. 1-1, given an
area density ψ(x, y) at (x, y). The mass inside the element with sides δx, and δy is ψi δxδy and
the total mass is given by summing and taking the limit.
Alternatively, we can interpret the integral to give the volume under the surface ψ(x, y)
(provided ψ(x, y) ≥ 0). The height of the cuboid associated with element i is ψi and the area of
its base is δxδy, so the volume of the cuboid is ψi δxδy. The total volume is given by summing
and taking the limit.
Example 5 Take R = [0, 2] × [1, 3] and evaluate
ZZ Z Z
2
I= (x + y ) x y. (1.4)
R
Solution.
We compute (first integrating with respect to x and keeping y constant)
y=3 x=2 y=3 x=2 Z

x2
Z Z Z Z Z
2
I = (x + y ) x y= + y2x y,
y=1 x=0 y=1 2 x=0
y=3 y=3
2y 3
Z Z
4 2
= + 2y − 0 y = 2y + ,
y=1 2 3 y=1
64
= ,
3
or (first integrating with respect to y and keeping x constant)
Z x=2 Z y=3 Z Z
2
I = (x + y ) y x,
x=0 y=1
x=2 y=3 Z Z x=2
y3

Z Z
27 1
= xy + x= 3x + − x+ x,
x=0 3 y=1 x=0 3 3
Z x=2 Z 2 x=2
26 2x 26x
= 2x + x= + ,
x=0 3 2 3 x=0
64
= .
3
Note: We get the same answer regardless of the order we do the integrals here, as one would
expect.
ZZ
Example 6 Let R be the unit square. Determine y cos2 (πxy)dA.
R
Solution. We can represent the domain by x ∈ [0, 1], y ∈ [0, 1], and thus
ZZ ZZ Z 1 Z 1
2 2 2
y cos (πxy)dA = y cos (πxy)dxdy = y cos (πxy)dx dy
0 0
R R
Z 1
1
= {πy + cos(πy) sin(πy)} dy
2π 0
Z 1 Z 1
1 2
= = cos (πxy)dy dx. (1.5)
4 0 0

Note: It is much certainly easier to do the x integral first even though the answer is independent
of the order of integration.
Important: However, in general, we need to be careful with the limits of the integrals for
general domains, and the limits depend on the order of integration, which we illustrate with a
triangle.
Example 7 Calculate the area of the triangle with vertices (0, 0) , (B, 0) and (X, H).
Solution.
We have known, since early school days that the answer is 21 BH, but we shall demonstrate this
here by means of a double integral. The three bounding lines of the triangles are
H H
y = 0, y= x, y= (x − B) .
X X −B
We’ll assume here that 0 < X < B. In order to ”pick up” all of the triangle’s area we need to
let x range from 0 to B and y range appropriately from 0 up to the bounding line above (x, 0) .
As our x and y vary over the triangle we need to pick up an infinitesimal piece of area dx dy
at each point. We can then calculate the triangle’s area as
Z x=X Z y=Hx/X Z x=B Z y=H(x−B)/(X−B)

A = dy dx + dy dx
x=0 y=0 x=X y=0
Z x=X Z x=B
Hx H (x − B)
= dx + dx
x=0 X x=X X −B
X " #B
H x2 H (x − B)2
= +
X 2 0 X −B 2
X
H X2 H (X − B) 2
= −
X 2 X −B 2
HX H (X − B)
= −
2 2
HB
= .
2
Integrating this way, y first then x, we would need to treat X < 0 and X > B as two further
cases. Alternatively, one could also pick up the area by first letting y range from 0 to H and
letting x range over the interior points of the triangle at height y.
Solution. This method is somewhat better as we don’t have to treat the three cases of X < 0,

0 < X < B and B < X separately.
Z y=H Z x=y(X−B)/H+B
A = dx dy
y=0 x=Xy/H
Z y=H
(X − B) y yX
= +B− dy
y=0 H H
Z y=H
By
= B− dy
y=0 H
y=H
y2

= B y−
2H y=0
H2

= B H−
2H
BH
=
2
More generally, we have to choose the limits to reflect the domain:
Example 8 Calculate the area of the disc x2 + y 2 6 a2 .
Solution. Again we know the answer, namely πa2 . If we wish to pick up all of the disc’s
√ area
can let x vary over the range −a to a and, at each x, we need to let y vary from − a2 − x2
we √
to a2 − x2 . So we have
√
Z x=a Z y= a2 −x2
A = √
dy dx
y=− a2 −x2
Zx=−a
x=a √
= 2 a2 − x2 dx
x=−a
Z θ=π/2 p
= 2 a2 − a2 sin2 θ a cos θ dθ [x = a sin θ]
θ=−π/2
Z π/2
2
= a 2 cos2 θ dθ
−π/2
Z π/2
= a2 1 + cos 2θ dθ
−π/2
π/2
2 1
= a θ + sin 2θ
2 −π/2
2
= πa .
In the first example of the triangle using y as the outside variable and x the inside to avoid
making special cases. For this example of the disc though it would have seemed much more
natural to use polar co-ordinates — if we knew how to calculate areas with such!
1.3 Change of Variables. Jacobians.
The Jacobian, or rather its modulus, is a measure of how a general mapping stretches space
locally, near a particular point, even when this stretching effect varies from point to point.
The Jacobian takes its name from the German mathematician Carl Jacobi (1804-1851).
Definition 9 Given two co-ordinates u (x, y) and v (x, y) which depend on variables x and y,
we define the Jacobian
∂ (u, v)
∂ (x, y)
to be the determinant ∂u ∂u
∂x ∂y
det ∂v ∂v .
∂x ∂y
Analogously, in 3D as required below, we define the Jacobian

∂ (u, v, w)
∂ (x, y, z)
to be the determinant ∂u ∂u ∂u
∂x ∂y ∂z

∂v ∂v ∂v

∂x ∂y ∂z
∂w ∂w ∂w
∂x ∂y ∂z
Example 10 Let x = r cos θ and y = r sin θ where r and θ are polar co-ordinates. Then
∂x ∂x
∂ (x, y) ∂r ∂θ
= det ∂y ∂y
∂ (r, θ) ∂r ∂θ

cos θ −r sin θ
= det
sin θ r cos θ
= r cos θ + sin2 θ = r.
2

p
Example 11 In reverse, r = x2 + y 2 and θ = tan−1 (y/x) and
∂r ∂r
∂ (r, θ) ∂x ∂y
= det ∂θ ∂θ
∂ (x, y) ∂x ∂y
√ x √ y
!
= det x2 +y 2 x2 +y 2
−y x
x2 +y 2 x2 +y 2
x2 + y 2
=
(x2 + y 2 )3/2
1 1
= p =
x2 + y 2 r
CHANGE OF VARIABLES. JACOBIANS. 11

Theorem 12 Let f : R → S be a bijection between two regions of R2 , which is differentiable
and has differentiable inverse with
∂ (u, v) ∂ (x, y)
, ,
∂ (x, y) ∂ (u, v)
defined and non-zero everywhere. Further, write (u, v) = f (x, y) and let ψ(x, y) = Ψ(u, v).
Then
ZZ ZZ
∂ (u, v)
Ψ(u, v)du dv = ψ(x, y)
dx dy,
∂ (x, y)
(u,v)∈S (x,y)∈R
ZZ ZZ
∂ (x, y)
ψ(x, y)dx dy = Ψ(u, v) du dv.
∂ (u, v)
(x,y)∈R (u,v)∈S
Proof. (Sketch proof of preceding theorem) It is sufficient to prove the first integral identity.
Divide the region R into N square elements of equal area δxδy as previously with ψi the scalar
field at the centre of the ith element, i ∈ {1, . . . N }.
Consider the mapping of the ith element, which we have is bounded by the co-ordinate lines
x = xi and x = xi + δx and y = yi and y = yi + δy. We have the value of the scalar field at the
mapped element centre is
Ψi = Ψ(f (xi + δx/2, yi + δy/2)) = ψi = ψ(xi + δx/2, yi + δy/2).
Also, given sufficiently small δu, δv the ith element maps to an image region, denoted IMi
below, which is a deformed parallelogram spanned by the vectors
∂f
a = f (xi + δx, yi ) − f (xi , yi ) ≈ (xi , yi ) δx,
∂x
∂f
b = f (xi , yi + δy) − f (xi , yi ) ≈ (xi , yi ) δy,
∂y
and thus of area
∂f ∂f ∂f ∂f
δx ∧ δy = ∧ δx δy.
∂x ∂y ∂x ∂y
Now f = (u, v), so fx = (ux , vx ), fy = (uy , vy ) and thus

∂f ∂f ∂ (u, v)
∂x ∧ ∂y δxδy = |(ux , vx ) ∧ (uy , vy )| δxδy = |(ux vy − uy vx ) k| δxδy = ∂ (x, y) δxδy.

Thus, before taking limits, partitioning S by the images IMi with we have an approximation
for ZZ
Ψ(u, v)du dv
(u,v)∈S
is
N N N
X X ∂ (u, v) X ∂ (u, v)
Ψi Area(IMi ) = Ψi
δxδy = ψi
δxδy.
i=1 i=1
∂ (x, y) i=1
∂ (x, y)
Taking limits gives
ZZ ZZ
∂ (u, v)
Ψ(u, v)du dv = ψ(x, y) dx dy.
∂ (x, y)
(u,v)∈S (x,y)∈R
If the previous motivation for this formula seems somewhat non-rigorous do note that the
chain rule for Jacobians does ensure that it is impossible that we might ever determine two
different answers for a double integral by using different sets of variables.
R∞
Note To evaluate integrals over an infinite domain, e.g. −∞ exp(−x2 )dx we define
Z ∞ Z X
f (x)dx = lim f (x)dx. (1.6)
−∞ X→∞ Y
Y →−∞
Exercise 13 Evaluate ZZ
exp[−(x2 + y 2 )]dA.
R2
R∞
Hence determine −∞
exp[−p2 ]dp.
Solution.
ZZ ZZ
2 2
I = exp[−(x + y )]dA = exp[−(x2 + y 2 )]dxdy
R2 R2
ZZ Z ∞ Z 2π Z ∞
2 ∂ (x, y)
2
= exp[−(r )] drdθ = dθ exp[−(r )]rdr = π 2r exp[−(r2 )]dr
∂ (r, θ) 0 0 0
R2
∞
= π exp[−(r2 )] 0 = π.

R∞
Furthermore let J = −∞ exp[−p2 ]dp. Then
Z ∞ Z ∞ ZZ
2 2 2
J = exp[−p ]dp exp[−q ]dq = exp[−(p2 + q 2 )]dpdq = I.
−∞ −∞
R2
√
Noting J > 0 to give the sign of the root, we thus have J = π.
Example 14 Calculate the area bounded by the curves
2x = 1 − y 2 , 2x = y 2 − 1, 8x = 16 − y 2 , 8x = y 2 − 16.
as shown in the diagram above.
Solution. We see, if we change to (u, v), parabolic coordinates given by

1 2
u − v2 ,

x= y = uv,
2

that the region in question is 1 6 u 6 2, 1 6 v 6 2. The Jacobian is given by
∂x ∂x
∂ (x, y) u −v
∂u
= det ∂y ∂y ∂v = det = u2 + v 2 .
∂ (u, v) ∂u ∂v
v u
Hence the area is given by

u=2 v=2
Z Z
∂ (x, y)
A = ∂ (u, v) dv du

u=1 v=1
Z u=2 Z v=2
u2 + v 2 dv du

=
u=1 v=1
Z u=2 v=2
2 v3
= u v+ du
u=1 3 v=1
Z u=2
2 7
= u + du
u=1 3
3 2
u 7u 7 7 14
= + = + = .
3 3 1 3 3 3
Example 15 Find the area bounded by the Folium of Descartes (see diagram) which has equa-
tion x3 + y 3 = 3xy.
Solution. If we convert to polar co-ordinates this becomes
r3 cos3 θ + r3 sin3 θ = 3r2 sin θ cos θ
which we can rearrange to

3 sin θ cos θ
r= .
cos3 θ + sin3 θ
The curve passes through the origin at θ = 0 and θ = π/2 and hence the area bounded is
π/2
9 sin2 θ cos2 θ
Z
1 3
A= 2 dθ = .
2 0
3
cos3 θ + sin θ 2
Exercise 16 How would you calculate the above integral? The t = tan 12 θ substitution would
be one guaranteed method, as this converts the integrand to a rational function.

2. VOLUME INTEGRATION
2.1 An Informal Definition of The Volume Integral
Consider a scalar field ψ(x, y, z), and a three-dimensional region R.
• Partition R into N cubic elements, of volume δV = δxδyδz and let ψi denote the value
of ψ at the centre of the ith cubic element, i ∈ {1, . . . N }.
• Then we have, on partitioning with smaller and smaller cubes, and taking the limit
δV → 0, we have
N ZZZ ZZZ
def
X
lim ψi δV = ψdV = ψdxdydz (2.1)
i=1 R R
• Once more, this is an informal definition. For the regions R and functions ψ we shall
meet there will never be any issue as to whether the integral might exist or not and we
will usually determine such integrals by calculating three definite integrals separately over
co-ordinates x, y, z (or other more appropriate co-ordinates). Again, the Banach-Tarski
paradox illustrates the complexities that can emerge but these are not the focus.
R R
Note If the scalar field ψ(x, y, z) = 1 then the integral R
ψdV = R
dV gives the volume of
the region R.
2.2 Examples
Example 17 A cone of height h occupies the region
x2 + y 2 6 z 2 , 0 6 z 6 h.
and has density ρ (x, y, z) = (x2 + y 2 ) z at each point. Find the mass of the cone using Cartesian
co-ordinates.
Solution. Our first problem is in dividing up the cone properly and finding the correct limits.
There are many different ways we might begin to take cross-sections of the cones, breaking the
cone into two- and then one-dimensional sections, but poorly chosen co-ordinates, whilst they
VOLUME INTEGRATION 17
could be used in principle to determine the mass, are likely to lead to a very complicated set
of integrals to calculate along the way.
If we use z as our first variable to divide up the cone, then we see the cross-section of the cone
with a plane z = z0 gives a disc x2 + y 2 6 (z0 )2 , z = z0 (at least if 0 6 z0 6 h). We can then,
say, take cross-sections of such discs with the line y = y0 to produce horizontal slithers
x2 6 (z0 )2 − (y0 )2 , y = y0 , z = z0 (at least if |y0 | 6 z0 )
and calculating their contribution to the mass is a simple one-dimensional integral.
So the mass M is given by the triple integral
Z h Z z Z √z2 −y2
M = √ ρ (x, y, z) dx dy dz
z=0 y=−z x=− z 2 −y 2
Z h Z z Z √z2 −y2
x2 + y 2 z dx dy dz.

= √
z=0 y=−z x=− z 2 −y 2
We need to calculate the internal integrals first. We have

Z √z2 −y2 3 √z2 −y2
x
x2 + y 2 z dx = z + y2x

√ 3 √
x=− z 2 −y 2 − z 2 −y 2
!
3/2
(z 2 − y 2 ) p
= 2z + y2 z2 − y2
3
2z 2 p
= z + 2y 2 z2 − y2.
3
The next internal integral is then
Z z
2z 2 p
z + 2y 2 z 2 − y 2 dy.
y=−z 3
It very much seems like a substitution y = z sin t, where −π/2 6 t 6 π/2, will help here. In
which case the integral becomes
Z π/2
2z 2 p
z + 2z 2 sin2 t z 2 − z 2 sin2 t (z cos t dt)
t=−π/2 3
Z π/2
2z 3
1 + 2 sin2 t |z cos t| (z cos t dt)

=
t=−π/2 3
2z 5 π/2
Z
1 + 2 sin2 t cos2 t dt.

=
3 t=−π/2
Now
Z π/2 Z π/2
2 2 2 1 1
cos t + 2 sin t cos t dt = (1 + cos 2t) + sin2 2t dt
t=−π/2 t=−π/2 2 2
Z π/2 π/2
1 1 3t 1 1 3π
= (1 + cos 2t) + (1 − cos 4t) dt = + sin 2t − sin 4t = .
t=−π/2 2 4 4 4 8 −π/2 4
18 VOLUME INTEGRATION
Finally then we have the external integral
h h
2z 5 3π π z6 πh6
Z
× dz = = .
z=0 3 4 2 6 0 12
Solution. (Alternative method with planar polars) The mass M we just calculated could have
been rewritten as
Z h ZZ Z h ZZ
x2 + y 2 z dx dy dz.

M= ρ (x, y, z) dx dy dz =
z=0 z=0
x2 +y 2 6z 2 x2 +y 2 6z 2
We have already met in the MT calculus course how to change from Cartesian co-ordinates x, y
to planar polar co-ordinates r, θ. The change of variable rule is
ZZ ZZ ZZ
φ dA = φ dx dy = φ (r dr dθ)
where the r appears as it is the Jacobian ∂ (x, y) /∂ (r, θ) . Thus
z 2π z z
r4 πz 5
ZZ Z Z Z
2 2 2 3

x +y z dx dy = r z (r dθ dr) = 2πz r dr = 2πz = .
r=0 θ=0 r=0 4 0 2
x2 +y 2 6z 2
Hence, finally
h 6 h
πz 5 πh6
Z
πz
M= dz = = .
z=0 2 12 0 12
This change of variable makes the calculation substantially simpler!
Remark 18 In the above we essentially used cylindrical polar co-ordinates appreciating that
the volume element dV is given by
dV = dx dy dz = r dr dθ dz.
EXAMPLES 19
Example 19 Find the volume of the region R that lies above the paraboloid z = x2 + y 2 and
beneath the plane x + y + z = 1.
Solution.
The x, y co-ordinates of a point (x, y, z) on the top planar surface of R satisfy
1 − x − y 6 x2 + y 2
and so the top planar surface projects vertically down to the set
W = (x, y) ∈ R2 : x + y + x2 + y 2 6 1

( 2 2 )
1 1 3
= (x, y) ∈ R2 : x + + y+ 6
2 2 2
p
which is a disc, centre (−1/2, −1/2) and radius 3/2. Hence we can determine the volume of
the region R as
Z Z Z 1−x−y
V = dz dA
z=x2 +y 2
(x,y)∈W
ZZ
1 − x − y − x2 − y 2 dA

=
(x,y)∈W
ZZ 2 2 !
3 1 1
= − x+ − y+ dA
2 2 2
(x,y)∈W
Natural co-ordinates for parameterising W are r, θ where

1 1 p
x = − + r cos θ, y = − + r sin θ, 0 6 θ < 2π, 0 6 r 6 3/2.
2 2
Hence we have
Z √3/2 Z 2π
3
V = − r2 cos2 θ − r2 sin2 θ (r dθ dr)
r=0 θ=0 2
Z √3/2
3r 3
= 2π − r dr
r=0 2
2 4
√3/2
3r r
= 2π −
4 4 0

9 9
= 2π −
8 16
9π
= .
8
Once more we have used polar coordinates. We therefore should consider what happens
with a general change of variable.
2.3 Changing Co-ordinates in Volume Integrals
Theorem 20 Let f : R → S be a bijection between subsets of R3 which is differentiable and

has differentiable inverse. Further take co-ordinates xi on R and ui on S related by the formula
(u1 , u2 , u3 ) = f (x1 , x2 , x3 ) ,
and let
ψ (x1 , x2 , x3 ) = Ψ (u1 , u2 , u3 )
be scalar fields. Then
ZZZ ZZZ
∂ (u1 , u2 , u3 )
Ψ (u1 , u2 , u3 ) du1 du2 du3 = ψ (x1 , x2 , x3 ) dx1 dx2 dx3 .
∂ (x1 , x2 , x3 )
S R
Proof. (Sketch proof) Partition the region R into N cubic elements and let the ith cubic
element be given by
[x1 , x1 + δx1 ] × [x2 , x2 + δx2 ] × [x3 , x3 + δx3 ]
which has volume δx1 δx2 δx3 . We have the value of the scalar field at the mapped element
centre is
Ψi = Ψ(f (x1 + δx1 /2, x2 + δx2 /2, x3 + δx3 /2)) = ψi = ψ(x1 + δx1 /2, x2 + δx2 /2, x3 + δx3 /2).
CHANGING CO-ORDINATES IN VOLUME INTEGRALS 21

Also, under the map f the ith cubic element maps to a deformed parallelepiped, denoted IMi
below, with sides
∂f
f (x1 + δx1 , x2 , x3 ) − f (x1 , x2 , x3 ) ≈ δx1 ;
∂x1
∂f
f (x1 , x2 + δx2 , x3 ) − f (x1 , x2 , x3 ) ≈ δx2 ;
∂x2
∂f
f (x1 , x2 , x3 + δx3 ) − f (x1 , x2 , x3 ) ≈ δx3 .
∂x3
The volume of a parallelepiped with sides a, b, c is |a · (b ∧ c) | and so, as a first approximation,

the volume of the image of the ith element is

∂f ∂f ∂f ∂ (u1 , u2 , u3 )
∂x1 , ∂x2 , ∂x3 δx1 δx2 δx3 = ∂ (x1 , x2 , x3 ) δx1 δx2 δx3 .

Thus, before taking limits, partitioning S by the images IMi with we have an approximation
for ZZZ
Ψ (u1 , u2 , u3 ) du1 du2 du3
S
is
N N N
X X ∂du1 du2 du3 X ∂ (u1 , u2 , u3 )
Ψi Volume(IMi ) = Ψi δxδy = ψi dx1 dx2 dx3 .
i=1 i=1
∂ (x1 , x2 , x3 ) i=1
∂ (x 1 , x2 , x3 )
Taking limits gives

ZZ ZZ
∂ (u1 , u2 , u3 )
Ψ (u1 , u2 , u3 ) du1 du2 du3 = ψ (x1 , x2 , x3 )
dx1 dx2 dx3 .
∂ (x1 , x2 , x3 )
S (x1 ,x2 ,x3 )∈R
If the previous motivation for this formula seems somewhat non-rigorous do note that the chain
rule for Jacobians does ensure that it is impossible that we might ever determine two different
answers for a volume integral by using different sets of variables.
Example 21 (Cylindrical Polar Co-ordinates)
x = r cos θ, y = r sin θ, z = z.
Then
∂x ∂x ∂x

cos θ −r sin θ 0
∂ (x, y, z) ∂r
∂y
∂θ
∂y
∂z
∂y = sin θ r cos θ 0 = cos θ −r sin θ

= ∂r ∂θ ∂z sin θ r cos θ
= r.
∂ (r, θ, z) ∂z ∂z ∂z

0

∂r ∂θ ∂r
0 1
Example 22 (Spherical Polar Co-ordinates)
x = r sin θ cos φ, y = r sin θ sin φ, z = r cos θ, θ ∈ [0, π], φ ∈ [0, 2π).
Then

∂x ∂x ∂x
∂ (x, y, z) ∂r ∂θ ∂φ
= ∂y ∂y ∂y

∂r ∂θ ∂φ
∂ (r, φ, θ) ∂z
∂r ∂z ∂θ
∂z
∂φ

sin θ cos φ r cos θ cos φ −r sin θ sin φ

= sin θ sin φ r cos θ sin φ r sin θ cos φ
cos θ −r sin θ 0

r cos θ cos φ −r sin θ sin φ sin θ cos φ −r sin θ sin φ
= cos θ + r sin θ
r cos θ sin φ r sin θ cos φ sin θ sin φ r sin θ cos φ
= cos θ r2 sin θ cos θ cos2 φ + sin2 φ + r2 sin3 θ cos2 φ + sin2 φ

= r2 sin θ cos2 θ + sin2 θ

= r2 sin θ ≥ 0 since θ ∈ [0, π].

Thus for spherical polar co-ordinates one effectively has
dV = dxdydz = r2 sin θ dr dθ dφ.
Definition 23 The centre of mass of a body occupying a region R and with density ρ (r) at
the point with position vector r is
ZZZ
1
r̄ = (x̄, ȳ, z̄) = r ρ (r) dV
M
R
where M is the total mass of the body.

Example 24 Find the centre of mass of a uniform hemisphere
(x, y, z) : x2 + y 2 + z 2 < a2 , z > 0 .

Solution. By symmetry, the centre of mass lies at a point (0, 0, z̄) on the z-axis. Say ρ is the
density of the hemisphere, so that its mass is 32 πa3 ρ. Then
Z a Z 2π Z θ=π/2
3
(r cos θ) ρr2 sin θ dθ dφ dr

z̄ = 3
2πa ρ r=0 φ=0 θ=0
Z a Z θ=π/2
3 3
= × 2πρ r dr cos θ sin θ dθ
2πa3 ρ r=0 θ=0
4 a π/2
3 r − cos 2θ
= 3× ×
a 4 0 4 0
4
3 a 1
= 3× ×
a 4 2
3a
= .
8

Example 25 Evaluate the integral ZZZ
x dV
R
√
over the intersection of the unit ball x + y + z 2 6 1 with the half-space x + y > 3 2/5.
2 2
Solution. If we take as an orthonormal basis

1 1
e1 = √ (1, −1, 0) , e2 = (0, 0, 1) , e3 = √ (1, 1, 0) ,
2 2
with corresponding co-ordinates X, Y, Z so that
Xe1 + Y e2 + Ze3 = xi + yj + zk,
√ √
then we have x = (X + Z) / 2 and x + y = Z 2.
So we are left considering the integral
ZZZ
X +Z
√ dV.
2
X 2 +Y 2 +Z 2 61
Z > 3/5
By symmetry ZZZ
X dV = 0,
X 2 +Y 2 +Z 2 61
Z > 3/5
so we are left to calculate

ZZZ Z 2π Z cos−1 (3/5) Z 1
1 1
√ Z dV = √ (r cos θ) r2 sin θ dr dθ dφ
2 2 φ=0 θ=0 r=(3/5) sec θ
X 2 +Y 2 +Z 2 61
Z > 3/5
Z −1
2π cos (3/5) 1
Z
= √ r3 cos θ sin θ dr dθ
2 θ=0 r=(3/5) sec θ
√ Z cos−1 (3/5) r4 1

= π 2 cos θ sin θ dθ
θ=0 4 (3/5) sec θ
√ Z −1
π 2 cos (3/5)

81 4
= 1− sec θ cos θ sin θ dθ
4 θ=0 625
√ Z −1
π 2 cos (3/5)

81 sin θ
= cos θ sin θ − dθ
4 θ=0 625 cos3 θ
√ cos−1 (3/5)
π 2 1 2 81 1
= − cos θ −
4 2 1250 cos2 θ 0
√
π 2 1 9 81 25 1 81
= − × − × + +
4 2 25 1250 9 2 1250
√ √
π 2 128 32π 2
= = .
4 625 625
Example 26 By a suitable rotation of co-ordinates, or otherwise, evaluate
ZZZ
(ax + by + cz)4 dV
taken over the region x2 + y 2 + z 2 6 1, where a, b, c are positive constants.
Solution. Consider instead this integral as

ZZZ
((a, b, c) · r)4 dV.
x2 +y 2 +z 2 61
We can rotate the axes to create new co-ordinates X, Y, Z in such a way that the Z axis points
in the direction of vector (a, b, c) and under this change of co-ordinates the integral becomes
ZZZ √ 4 ZZZ
2
2 2 2
a + b + c e3 · r dV = a2 + b2 + c2 Z 4 dV
X 2 +Y 2 +Z 2 61 X 2 +Y 2 +Z 2 61
where e3 is the unit vector pointing down the Z-axis; notice that we are still integrating over
the unit sphere as we have simply rotated the axes about the origin.
Now, by a change to spherical polar co-ordinates,
ZZZ
2
a2 + b2 + c2 Z 4 dV
X 2 +Y 2 +Z 2 61
Z 1 Z 2π Z π
2 2
2 2
(r cos θ)4 r2 sin θ dθ dφ dr

= a +b +c
r=0 φ=0 θ=0
7
1 π
cos5 θ

2 r 2 2 2

= 2π a + b + c −
7 0 5 0
2 1 2
= 2π a2 + b2 + c2 × ×
7 5
4π 2 2
a + b2 + c 2 .

=
35
Example 27 Let a, b, c ∈ R. Determine

ZZZ
I= cos (ax + by + cz) dx dy dz.
x2 +y 2 +z 2 61
Show that your answer is consistent with the fact the volume of the unit sphere is 4π/3.

Solution. We can rewrite the integral as
ZZZ
I= cos (a · r) dx dy dz
x2 +y 2 +z 2 61
where a = (a, b, c) .
The vector
a (a, b, c)
e3 = =√
|a| a2 + b2 + c2
is of unit length and so we can extend it to an orthonormal basis e1 , e2 , e3 with associated

co-ordinates X, Y, Z. In terms of these co-ordinates
a · r = (0, 0, |a|) · (X, Y, Z) = |a| Z.
The unit sphere x2 +y 2 +z 2 6 1 is still given as X 2 +Y 2 +Z 2 6 1 as e1 , e2 , e3 are an orthonormal

basis and dV = dX dY dZ. So
ZZZ
I= cos (|a| Z) dX dY dZ.
X 2 +Y 2 +Z 2 61
If we now change to spherical polar co-ordinates r, θ, φ associated with X, Y, Z we have

Z 1 Z π Z 2π
I = cos (|a| r cos θ) r2 sin θ dφ dθ dr
r=0 θ=0 φ=0
Z 1 Z π
= 2π cos (|a| r cos θ) r2 sin θ dθ dr
Zr=0
1
θ=0
π
− sin (|a| r cos θ) 2
= 2π r dr
r=0 |a| r 0
Z 1
r
= 2π (2 sin (|a| r)) dr
r=0 |a|
( 1 Z 1 )
4π −r cos (|a| r) cos (|a| r)
= + dr
|a| |a| r=0 r=0 |a|
1
4π −r cos (|a| r) sin (|a| r)
= +
|a| |a| |a|2 r=0

4π − cos |a| sin (|a|)
= +
|a| |a| |a|2
4π
= (sin |a| − |a| cos |a|) .
|a|3
If we wish to determine the volume of the unit sphere we need to set a = b = c = 0. So letting
|a| → 0 we see
! !!
4π 4π |a|3 |a|2
+ O |a|3 + O |a|4

3 (sin |a| − |a| cos |a|) = 3 |a| − − |a| 1 −
|a| |a| 6 2
!
4π |a|3
+ O |a|4

= 3
|a| 3
4π 4π
= + O (|a|) → as |a| → 0.
3 3

3. SURFACE INTEGRALS
3.1 Paramterised Surfaces
We will now be interested in finding integrals over surfaces. To do this, we begin by looking at
some different ways to represent surfaces.
Representation of surfaces
Cartesian representation. In Cartesian coordinates we represent a surface by z = f (x, y),

e.g. the paraboloid z = x2 + y 2 .
Parametric representation. In parametric coordinates we represent a surface by r =

r(u, v) = (x(u, v), y(u, v), z(u, v)) where (u, v) ∈ D ⊂ R2 . So r = r(u, v) maps D in the
(u, v) plane to a surface S in R3 .
Note If z = f (x, y) we can use x and y as parameters: r(x, y) = (x, y, f (x, y)).
Example 28 (Parameterisation of a spherical surface)
SURFACE INTEGRALS 29
To represent the point P = (x, y, z) in spherical polar coordinates (r, θ, φ) we have
x = r sin θ cos φ, (3.1)

y = r sin θ sin φ, (3.2)
z = r cos θ, (3.3)
where θ ∈ [0, π], φ ∈ [0, 2π).

Thus we can parameterise the sphere x2 + y 2 + z 2 = a2 in u = θ, v = φ by
r = r(θ, φ) = (a sin θ cos φ, a sin θ sin φ, a cos θ). (3.4)
Example 29 The quadrics are standard parameterised surfaces:
• Sphere: x2 + y 2 + z 2 = a2 ;
30 SURFACE INTEGRALS
• Ellipsoid: x2 /a2 + y 2 /b2 + z 2 /c2 = 1;
• Hyperboloid of One Sheet: x2 /a2 + y 2 /b2 − z 2 /c2 = 1;
• Hyperboloid of Two Sheets: x2 /a2 − y 2 /b2 − z 2 /c2 = 1;
• Paraboloid: z = x2 + y 2 ;
• Hyperbolic Paraboloid: z = x2 − y 2 ;
• Cone: x2 + y 2 = z 2 .
Definition 30 A smooth parameterised surface is a map r, known as the parametrisa-

tion
r : U → R3 : (u, v) 7→ (x (u, v) , y (u, v) , z (u, v))
from a subset U ⊆ R2 to R3 such that
• x, y, z have continuous partial derivatives with respect to u and v of all orders
• r is a bijection with both r and r−1 being continuous
• at each point the vectors

∂r ∂r
and
∂u ∂v
are linearly independent (i.e. are not scalar multiples of one another). Equivalently
∂r ∂r
∧ 6= 0.
∂u ∂v
We will not be looking to treat this definition with any generality. Rather we shall just look
to parameterise some of the ”standard” surfaces previously described. We define:
Definition 31 Let r : U → R3 be a smooth parameterised surface and let p be a point on the

surface. The plane containing p and which is parallel to the vectors
∂r ∂r
(p) and (p)
∂u ∂v
is called the tangent plane to r (U ) at p. Because these vectors are independent the tangent
plane is well-defined.
Definition 32 Any vector in the direction
∂r ∂r
(p) ∧ (p)
∂u ∂v
is said to be normal to the surface at p. There are two unit normals of length one which
we denote as n and −n.
PARAMTERISED SURFACES 31
3.2 Surface Integrals
Let r : U → R3 be a smooth parameterised surface with
r (u, v) = (x (u, v) , y (u, v) , z (u, v))
and consider the small element of the plane that is bounded by the co-ordinate lines u = u0
and u = u0 + δu and v = v0 and v = v0 + δv. Then r maps this to a small region of the surface
r (U ) and we are interested in calculating the surface area of this small region. Note
∂r
r (u + δu, v) − r (u, v) ≈ (u, v) δu,
∂u
∂r
r (u, v + δv) − r (u, v) ≈ (u, v) δv.
∂v
Recall that the area of a parallelogram with sides a and b is |a ∧ b| . So the element of surface
area we are considering is approximately

∂r ∂r ∂r ∂r
δu ∧ δv = ∧ δu δv.
∂u ∂v ∂u ∂v
Thus, at this point we proceed as with double integrals. Partitioning U by smaller and smaller
elements, of area δA = δuδv, we have in the limit δA → 0
ZZ
X ∂r ∂r def ∂r ∂r
lim ∂u ∧ ∂v δu δv = ∂u ∧ ∂v du dv.

elements U
This gives the surface area of the parameterised surface; as with double integrals the pre-limit
summation can be weighted with a scalar function ψ(r(u, v)) evaluated at the centre of the
elements to yield ZZ
∂r ∂r
ψ(r(u, v)) ∧ du dv.
∂u ∂v
U
Definition 33 We will often write

∂r ∂r
dS =
∧ du dv
∂u ∂v
to denote an infinitesimal part of surface area. We will also write
∂r ∂r
dS = ∧ du dv.
∂u ∂v
This is also commonly written as n dS. Note there is a sign ambiguity in general until one
defines whether the normal is in the direction of
∂r ∂r ∂r ∂r
∧ or ∧ .
∂u ∂v ∂v ∂u
Proposition 34 The surface area of r (U ) is independent of the choice of parametrisation.
Proof. Let Σ = r (U ) = s (W ) be two different parametrisations of a surface X; take u, v as

the co-ordinates on U and p, q as the co-ordinates on W . Let f = (f1 , f2 ) : U → W be the
co-ordinate change map – i.e. for any (u, v) ∈ U we have
r (u, v) = s (f (u, v)) = s (f1 (u, v) , f2 (u, v)) = s(p, q).
Then
∂r ∂s ∂f1 ∂s ∂f2 ∂r ∂s ∂f1 ∂s ∂f2
= + , = + .
∂u ∂p ∂u ∂q ∂u ∂v ∂p ∂v ∂q ∂v
Hence
∂r ∂r ∂s ∂f1 ∂s ∂f2 ∂s ∂f2 ∂s ∂f1
∧ = ∧ + ∧
∂u ∂v ∂p ∂u ∂q ∂v ∂q ∂u ∂p ∂v

∂f1 ∂f2 ∂f1 ∂f2 ∂s ∂s
= − ∧
∂u ∂v ∂v ∂u ∂p ∂q
∂ (p, q) ∂s ∂s
= ∧ .
∂ (u, v) ∂p ∂q
Finally
ZZ ZZ
∂r ∂r ∂ (p, q) ∂s ∂s
∂u ∧ ∂v du dv = ∂ (u, v) ∂p ∧ ∂q du dv

U U
ZZ
∂s ∂s ∂ (p, q)
= ∧
∂p ∂q ∂ (u, v) du dv

U
ZZ
∂s ∂s
= ∂p ∧ ∂q dp dq

W
by the two-dimensional rule for the change of variables in integrals.
General method of evaluation of surface integrals
1. find a suitable parametrisation, e.g. in terms of some u, v;
2. find the range of u, v, i.e. a set D ⊂ R2 ;
3. evaluate ∂r/∂u ∧ ∂r/∂v for u, v ∈ D;
4. substitute into the relevant integral and integrate.
Example 35 Let 0 < a < b. Find the area of the torus obtained by revolving the circle
(x − b)2 + z 2 = a2 in the xz-plane about the z-axis.
Solution. We can parameterise the torus as
r (θ, φ) = ((b + a sin θ) cos φ, (b + a sin θ) sin φ, a cos θ) 0 6 θ, φ 6 2π.
We have
rθ = (a cos θ cos φ, a cos θ sin φ, −a sin θ) , rφ = (− (b + a sin θ) sin φ, (b + a sin θ) cos φ, 0)
and

i j k

rθ ∧ rφ =
a cos θ cos φ a cos θ sin φ −a sin θ

− (b + a sin θ) sin φ (b + a sin θ) cos φ 0

i j k

= a (b + a sin θ) cos θ cos φ cos θ sin φ − sin θ
− sin φ cos φ 0
= a (b + a sin θ) (sin θ cos φ, sin θ sin φ, cos θ) .
The surface area of the torus is
Z 2π Z 2π q
a (b + a sin θ) sin2 θ cos2 φ + sin2 θ sin2 φ + cos2 θ dφ dθ
θ=0 φ=0
Z 2π
= 2πa (b + a sin θ) dθ
θ=0
= 4π 2 ab.
Example 36 Let z = f (x, y) denote the graph of a function f defined on a subset S of the
xy-plane. Show that the graph has surface area
ZZ q
1 + (fx )2 + (fy )2 dx dy.
S
Deduce that a sphere of radius a has surface area 4πa2 .
Solution. We can parameterise the surface as

r (x, y) = (x, y, f (x, y)) (x, y) ∈ S.
Then
i j k

rx ∧ ry = 1 0 fx = (−fx , −fy , 1) .

0 1 fy
Hence the graph has surface area
ZZ ZZ q
|rx ∧ ry | dx dy = 1 + (fx )2 + (fy )2 dx dy.
S S
We can calculate the area of a hemisphere of radius a by setting
p
f (x, y) = a2 − x2 − y 2 x2 + y 2 < a2 .
We then have
−x −y
fx = p , fy = p
a2 − x2 − y 2 a2 − x 2 − y 2
and so the hemisphere’s area is
s
x2 + y 2
ZZ
1+ dx dy
a2 − x 2 − y 2
x2 +y 2 <a2
ZZ
a
= p dx dy
a − x2 − y 2
2
x2 +y 2 <a2
Z a Z 2π
a
= √ r dθ dr
r=0 θ=0 a2 − r 2
h √ ia
2
= 2π −a a − r 2
0
= 2πa2 .
Hence the area of the whole sphere is 4πa2 .
Example 37 Calculate the area of a sphere of radius a using spherical polar co-ordinates.
Solution. We can parameterise the sphere by
r (θ, φ) = (a sin θ cos φ, a sin θ sin φ, a cos θ) 0 6 θ 6 π, 0 6 φ 6 2π.
Then

i j k
∂r ∂r ∂x ∂y ∂z
∧ = ∂θ ∂θ ∂θ
∂θ ∂φ ∂x ∂y ∂z
∂φ ∂φ ∂φ

i j k

= a cos θ cos φ a cos θ sin φ −a sin θ

−a sin θ sin φ a sin θ cos φ 0
= a2 sin2 θ cos φ, sin2 θ sin φ, sin θ cos θ

= a2 sin θ (sin θ cos φ, sin θ sin φ, cos θ) .
Hence

dS = a2 sin θ (− sin θ cos φ, sin θ sin φ, cos θ) dθ dφ = a2 |sin θ| dθ dφ.
Finally
Z π Z 2π
A = a2 |sin θ| dφ dθ
θ=0
Zφ=0
π
= 2πa2 |sin θ| dθ
θ=0
= 4πa2 .
Definition 38 If F and φ are a vector field and scalar field defined on a surface Σ, parame-
terised as r (U ), then we may define the following surface integrals:
ZZ ZZ
∂r ∂r
F · dS = F (r (u, v)) · ∧ du dv;
∂u ∂v
Σ U
ZZ ZZ
∂r ∂r
F dS = F (r (u, v)) ∧ du dv;
∂u ∂v
Σ U
ZZ ZZ
∂r ∂r
φ dS = φ (r (u, v)) ∧ du dv;
∂u ∂v
Σ U
ZZ ZZ
∂r ∂r
φ dS = φ (r (u, v)) ∧ du dv.
∂u ∂v
Σ U
We will most commonly meet integrals of the first type

ZZ
F · dS
Σ
which are known as flux integrals.
Definition 39 The solid angle is the angle an object subtends at a point in three-dimensional
space. More precisely, half-lines from a fixed point (or observer) will either intersect with the
object in question or not; those lines of sight that are blocked by the object represent a subset
of the unit sphere centred on the observer. The solid angle is area of this subset (strictly it
is the area of this subset divided by the unit of length squared to ensure the solid angle is
dimensionless). The unit of solid angle is the steradian. Given that the surface area of a
sphere is 4π (radius)2 then a whole solid angle is 4π steradian.
If Σ is a surface then the solid angle Ω subtended at O by Σ equals, by definition,
er · dS r · dS
ZZ ZZ
Ω= = ,
r2 r3
Σ Σ
where r = |r|.
Example 40 Find the solid angle at the apex of a right pyramid with square base of side 2d
and height h.
Solution. Consider placing the apex of the pyramid at the origin and orientating the axis
of the pyramid along the positive z-axis so that the square base of the pyramid has vertices
(±d, ±d, h) . By symmetry the solid angle at the apex is 4 times the solid angle subtended by
the smaller square with vertices
(0, 0, h) , (0, d, h) , (d, d, h) (0, d, h) .
Hence the solid angle is

d d d d
(x, y, h) · k dx dy
Z Z Z Z
dx dy
Ω=4 = 4h .
y=0 x=0 (x2 + y 2 + h2 )3/2 y=0 x=0 (x2 + y 2 + h2 )3/2
p p
2 2
If we set x = y + h tan t and tan τ = d/ 2 2
y + h then we have
d τ
p
y 2 + h2 sec2 t dt dy
Z Z
Ω = 4h 3/2
y=0 t=0 ((y 2 + h2 ) tan2 t + y 2 + h2 )
d τ
p
y 2 + h2 sec2 t dt dy
Z Z
= 4h 3/2
y=0 t=0 (y 2 + h2 )3/2 (tan2 t + 1)
Z d Z τ
cos t dt dy
= 4h
y=0 t=0 (y 2 + h2 )
Z d
sin τ dy
= 4h
y=0 (y 2 + h2 )
Z d
dy
= 4h p
y=0 (y 2 + h2 ) y 2 + h2 + d2
−1/2
as sin τ = d (y 2 + h2 + d2 ) .
√ √
If we make a similar substitution again, namely y = h2 + d2 tan φ and tan α = d/ h2 + d2
then we see that
α
√
h2 + d2 sec2 φ dφ
Z
Ω = 4h √
φ=0 ((h2 + d2 ) tan2 φ + h2 ) h2 + d2 sec φ
Z α
cos φ dφ
= 4h
(h2 + d2 ) sin2 φ + h2 cos2 φ

φ=0
Z α
cos φ dφ
= 4h
(h + d ) sin2 φ + h2 cos2 φ

2 2
φ=0
Z α
cos φ dφ
= 4h .
φ=0 d sin2 φ + h2
2
√
Our final substitution is u = sin φ so that sin−1 α = d/ h2 + 2d2 and
sin−1 α sin−1 α
d2
Z
du 4h d −1 du −1
Ω = 4h = tan = 4 tan √ .
u=0 d2 u2 + h2 d h h 0 h h2 + 2d2
Example 41 Evaluate the integral ZZ

F ∧ dS
Σ
where in each case Σ is the closed hemispherical surface made up of points (x, y, z) such that
either x2 + y 2 + z 2 = 1 and z > 0, or x2 + y 2 6 1 and z = 0;. orient Σ so that dS is in the
direction of the outward-pointing normal and F = (zx, zy, z 2 ).
Solution. We could proceed to parameterise the two parts of Σ, the upper part of the hemi-
spherical surface Σ1 and the planar disc Σ2 . These can be respectively parameterised as
r1 (θ, φ) = (sin θ cos φ, sin θ sin φ, cos θ) 0 6 φ 6 2π, 0 6 θ 6 π/2;

r2 (r, θ) = (r cos θ, r sin θ, 0) 0 6 θ 6 2π, 0 6 r 6 1.
With some further calculation we would find

∂r1 ∂r1 ∂r1 ∂r1
∧ = sin θ (sin θ cos φ, sin θ sin φ, cos θ) , ∧ = (0, 0, r) .
∂θ ∂φ ∂r ∂θ
In order to get the correct outward-pointing direction we need to set
dS = sin θ (sin θ cos φ, sin θ sin φ, cos θ) dθ dφ, dS = −rk,
respectively.
However if we stop to think a little we can straight away see that ... On Σ1 we have F = zr
and n = r so that F ∧ dS = zr ∧ n dS = zr ∧ r dS = 0 on all of Σ1 , and further as z = 0 on
all of Σ2 then it’s also true that F = 0 on all of Σ2 .
Moral of the story: take some brief time to consider the nature of your function
and the region, what integrals actually need calculating and what co-ordinates are
best.
Remark 42 Two points of notation in relation to sheet 4. In question 4 ∂Σ refers to the

boundary of Σ, which is formed from three curves which can be parameterised
C1 = {(z, 0, z) : 0 6 z 6 2} ,
C2 = {(2 cos t, 2 sin t, 2) : 0 6 t 6 π} ,
C3 = {(z, 0, −z) : 0 6 z 6 2} .
R
Be sure to orient the curves consistently in a loop to find ∂Σ f dr (which is defined only up to
sign).
In question 5 ∂R refers to the boundary of R, which is formed from three surfaces namely
Σ1 = (x, y, z) : x2 + y 2 6 4, z = 0 ,

Σ2 = (x, y, z) : x2 + y 2 = 4, 0 6 z 6 4 ,

Σ3 = (x, y, z) : x2 + y 2 6 4, z = x2 + y 2 .

To ensure equality, when calculating ZZ

F · dS
∂R
then dS should be in the direction of the outward pointing normal. This is in fact a standard
convention; for a surface enclosing a 3d shape, it is standard to use the outward
pointing normal.
4. LINE INTEGRALS & CONSERVATIVE FIELDS
4.1 Curves
Definition 43 By a curve we shall mean a piecewise smooth function γ : I → R3 defined on

an interval I of R. Notice that order on I also gives the curve γ an orientation.
We shall also use the term curve to describe the images of such maps γ. Given such an image
then it will be the image of more than one such map γ and we will talk about parametrisations
γ1 and γ2 of the curve. These parametrisations of the image come in two different possible
orientations.
Definition 44 We say a curve γ : [a, b] → R3 is simple if γ is 1-1, with the one possible
exception that γ (a) = γ (b) may be true; this means that the curve does not cross itself except
possibly by its endpoints meeting.
Definition 45 We say a curve γ : [a, b] → R3 is closed if γ (a) = γ (b) .
Example 46 The line through points p and q can be parameterised as

γ (t) = p + t (q − p) .
When 0 < t < 1 then γ (t) lies between p and q, for t > 1 beyond q and for t < 0 before p.
Example 47 A curve of the form

γ (t) = (a cos t, a sin t, ct) (4.1)
is known as a circular helix.
Example 48 Parameterise the parabola formed by intersecting the plane 3x + 4y + 5z = 1 with

the cone x2 + y 2 = z 2 , z > 0.
Solution.
A general point on the cone can be written as r (θ, z) = (z cos θ, z sin θ, z) where z > 0 and
0 6 θ < 2π. If this point also lies in the plane 3x + 4y + 5z = 1 then
z (3 cos θ + 4 sin θ + 5) = 1
and so we see

cos θ sin θ 1
γ (θ) = , ,
3 cos θ + 4 sin θ + 5 3 cos θ + 4 sin θ + 5 3 cos θ + 4 sin θ + 5
LINE INTEGRALS & CONSERVATIVE FIELDS 41

lies on the parabola. What values should θ range through? We need
0 6= 3 cos θ + 4 sin θ + 5 = 5 cos (θ − α) + 5
where α = tan−1 43 . Hence

1 cos θ sin θ 1
γ (θ) = , , α−π <θ <α+π
5 cos (θ − α) + 1 cos (θ − α) + 1 cos (θ − α) + 1
is a parameterisation.
Aside. Note in the above example, let

1 1 1
e1 = √ (−3, −4, 5), e2 = (−4, 3, 0), e3 = e1 ∧ e2 = √ (3, 4, 5)
50 5 50
be a set of basis vectors. With
X = e1 · γ(θ), Y = e2 · γ(θ))
on finds that
Y 2 /X
is constant for all θ, explicitly confirming we have a parabola in the plane spanned by e1 , e2 .
Example 49 Show that the plane with equation Aa + Bb + Cc = D intersects the unit sphere
x2 + y 2 + z 2 = 1 in a circle if and only if A2 + B 2 + C 2 > D2 . Parameterise the intersection
of x + y + z = 1 with the unit sphere.
Solution. The plane Aa + Bb + Cc = D has normal (A, B, C) and so the point closest to the
origin is the point with position vector λ (A, B, C) which lies on the plane; by substituting this
in we see λ = D/ (A2 + B 2 + C 2 ). This point is within unit distance of the origin if and only if
√
D |D| A2 + B 2 + C 2
1 > 2 (A, B, C) =
A + B2 + C 2 A2 + B 2 + C 2
i.e. if and only if A2 + B 2 + C 2 > D2 .

By this criterion the plane x + y + z = 1 intersects with the unit sphere. The centre of the
circle which makes the intersection is at
1 (1, 1, 1)
(1, 1, 1) = .
12 2
+1 +1 2 3
By Pythagoras’ Theorem the radius r of the circle satisfies
2 r
(1, 1, 1) = 1 =⇒ r = 2 .

r2 +
3 3
42 LINE INTEGRALS & CONSERVATIVE FIELDS

√ √
As e1 = (1, −1, 0) / 2 and e2 = (1, 1, −2) / 6 are two orthonormal vectors parallel to the
plane then every point of the circle can be written in the form γ (t) for 0 6 t < 2π where
r r
(1, 1, 1) 2 2
γ (t) = + e1 cos t + e2 sin t
3 3 3
1 1 1 1 1 1 1 2
= + √ cos t + sin t, − √ cos t + sin t, − sin t .
3 3 3 3 3 3 3 3
4.2 Line Integrals
Definition 50 Let C be a curve in R3 parameterised by γ : [a, b] → R3 , and let F be a vector

field, whose domain includes C. Then we define the line integral of F along C as
Z Z b
F · dr = F (γ (t)) · γ 0 (t) dt,
C a
where · denotes the scalar product.

R
Proposition 51 If oriented the same, the line integral C
f · dr is independent of the choice
of parametrisation.
Proof. Suppose that γ1 : [a1 , b1 ] → R3 and γ2 : [a2 , b2 ] → R3 be two parametrisations of C

with γ1 (a1 ) = γ2 (a2 ) and γ1 (b1 ) = γ2 (b2 ), so that γ1 and γ2 give C the same orientation. Then
γ2 = γ1 ◦ ψ where ψ : [a2 , b2 ] → [a1 , b1 ] associates the γ2 co-ordinates of points on C with their
γ1 co-ordinate.
We now define I : [a1 , b1 ] → R and J : [a2 , b2 ] → R by
Z t Z t
0
I (t) = F (γ1 (s)) · γ1 (s) ds, J (t) = F (γ2 (s)) · γ20 (s) ds.
a1 a2
By the Fundamental Theorem of Calculus
I 0 (t) = F (γ1 (t)) · γ10 (t) , J 0 (t) = F (γ2 (t)) · γ20 (t) .
Further, for ψ(t) a function of t we have, by the chain rule

d
I (ψ (t)) = ψ 0 (t) I 0 (ψ (t))
dt
= ψ 0 (t) F (γ1 (ψ (t))) · γ10 (ψ (t)) .
LINE INTEGRALS 43
Recall
γ2 (t) = γ1 (ψ(t)).
Thus γ20 (t) = ψ 0 (t)γ10 (ψ(t)) and hence
d
I (ψ (t)) = F (γ2 (t)) · (γ10 (ψ (t)) ψ 0 (t))
dt
= F (γ2 (t)) · γ20 (t) = J 0 (t) .
It follows that I (ψ (t)) and J (t) differ by a constant and, as they agree at t = a2 (when they
are both zero), then I (ψ (t)) = J (t) and in particular when t = b2 we have
Z b1 Z b2
F (γ1 (s)) · γ10 (s) ds = F (γ2 (s)) · γ20 (s) ds.
a1 a2
R
Remark 52 If we parameterised C in the reverse orientation then the integral C
F · dr would
give negative what had been previously calculated.
Example 53 Let F = c ∧ r where c is a constant vector and let C be the circular helix
parameterised by
r (t) = (cos t, sin t, t) , 0 6 t 6 2π.
Then

Z Z i2π j k

F · dr = c1
c 2 c 3
· (− sin t, cos t, 1) dt

C 0 cos t sin t t

Z 2π − sin t cos t 1

= c1 c2 c3 dt

0 cos t sin t t
Z 2π
−c2 t sin t + c3 cos2 t + c1 sin t − c2 cos t + c3 sin2 t − c1 t cos t dt

=
Z0 2π
= (−c2 t sin t + c3 − c1 t cos t) dt
0
= 2πc3 − c1 [t sin t + cos t]2π 2π
0 − c2 [−t cos t + sin t]0
= 2π (c2 + c3 ) .

Definition 54 If φ is a scalar field defined on a curve C with parametrisation γ : [a, b] → R3
then we also define the line integral
Z Z b
φ ds = φ (γ (t)) |γ 0 (t)| dt.
C a
If F = (f1 , f2 , f3 ) is a vector field defined on the curve C then we define
Z Z Z Z
F ds = f1 ds, f2 ds, f3 ds .
C C C C
Remark 55 Note that if t is the unit tangent vector field along C, in the same direction as
the parametrisation, then
Z Z
φ ds = (φt) · dr
C C
and so, by inheritance, we see that these line integrals also do not depend on the choice of
parametrisation.
Note further that the two types of integral defined above are also independent of the choice of
orientation of C, as they are independent of the sign of γ 0 (t).
Definition 56 If C is a curve with parametrisation γ : [a, b] → R3 then the arc length of the
curve is
Z Z b
ds = |γ 0 (t)| dt.
C a
LINE INTEGRALS 45
Example 57 Find the arc length of the circular helix r (t) = (cos t, sin t, t) where 0 6 t 6 2π.
Solution. We have
Z 2π Z 2π p
2
Z 2π √ √
s= |(− sin t, cos t, 1)| dt = 2
1 + sin t + cos t dt = 2 dt = 2 2 π.
0 0 0
We could also parameterise the helix ”in reverse” by setting s (t) = r (2π − t) = (cos t, − sin t, 2π − t)
where 0 6 t 6 2π. We would still find
Z 2π Z 2π p
2
Z 2π √ √
s= |(− sin t, − cos t, −1)| dt = 2
1 + sin t + cos t dt = 2 dt = 2 2 π.
0 0 0
Notation 58 We will standardly use the notation
dr = (dx, dy, dz) and ds = |dr| ,
even though this may seem a little non-rigorous and any analysis course would insist on such
differentials only appearing as part of a limit or within an integral. Rest assured that these
differentials can be rigorously defined, though doing so is not a primary concern of this course.
Definition 59 With notation as in the Definition 50, if the vector field F represents a force
on a particle then Z
F · dr (4.2)
C
is the work done by the force in moving the particle along C.
This is a generalization of the formula
Work = Force × Distance
which applies to constant forces acting parallel to the direction of the motion. More generally,
we have
Work = Component of force in direction of travel × Distance
if the force and movement are not parallel. The work integral (4.2) is just an integral of such
infinitesimal contributions of work.
Example 60 Show that the work done by gravity, F = −mgk, in moving a particle along a
straight line from (x1 , y1 , z1 ) to (x2 , y2 , z2 ) equals mg (z1 − z2 ).

Solution. We can parameterise the line segment as
r (t) = (x1 , y1 , z1 ) + t (x2 − x1 , y2 − y1 , z2 − z1 ) , 0 6 t 6 1,
so that
dr = (x2 − x1 , y2 − y1 , z2 − z1 ) dt.
Then
Z Z 1
F · dr = −mgk · (x2 − x1 , y2 − y1 , z2 − z1 ) dt
C 0
Z 1
= −mg (z2 − z1 ) dt
0
= mg (z1 − z2 ) .
In fact it’s the case that the work done by gravity – which is positive if z1 > z2 , i.e. the particle
has dropped – is the loss in gravitational potential energy which, commonly, will have been
converted into kinetic energy. The work done is dependent only on the endpoints (x1 , y1 , z1 )
and (x2 , y2 , z2 ) and is independent of the path taken between these endpoints. This is common
to certain types of field which are known as conservative.
4.3 Conservative Fields
Proposition 61 If F = ∇φ and γ : [a, b] → S is any curve such that γ (a) = p, γ (b) = q then
Z
F (r) · dr = φ (q) − φ (p) .
γ
R
In particular, the integral γ
F (r) · dr depends only on the endpoints of the curve γ.
Proof. If γ (t) = (x (t) , y (t) , z (t)) then
Z Z b
F (r) · dr = F (γ (t)) · γ 0 (t) dt
γ a
Z b
= ∇φ (γ (t)) · γ 0 (t) dt
a
Z b
∂φ dx ∂φ dy ∂φ dz
= + + dt
a ∂x dt ∂y dt ∂z dt
Z b
d
= (φ (γ (t))) dt [by the chain rule]
a dt
= φ (γ (b)) − φ (γ (a))
= φ (q) − φ (p) .
CONSERVATIVE FIELDS 47
Definition 62 Let S be an open subset of R3 . A vector field F : S → R3 is said to be
conservative if there exists a scalar field φ : S → R such that F = ∇φ.
Definition 63 If F = ∇φ then φ is said to be a potential (or scalar potential) for F.
Definition 64 A subset S of R3 is said to be path-connected if for any points p, q ∈ S there

exists a curve γ : [a, b] → S such that γ (a) = p and γ (b) = q.
Corollary 65 If ∇φ = 0 on a path-connected set S then φ is constant. In particular, if φ and

ψ are potentials of the conservative field F, defined on S, then φ and ψ differ by a constant.
Proof. If ∇φ = 0 then for any curve, with endpoints p, q, we have

Z
φ (q) − φ (p) = ∇φ · dr = 0.
γ
Given a fixed point p in S, any q in S is connected to p by a curve, and so φ is constant. If

F = ∇φ = ∇ψ then ∇ (φ − ψ) = 0 and the result follows.
Theorem 66 Let S be an open subset of R3 and let F : S → R3 be a vector field. Then the
following three statements are equivalent.
(i) F is conservative – i.e. F = ∇φ for some scalar field φ : S → R.
(ii) Given any two points p, q ∈ S and curve γ in S, starting at p and ending at q, then
the integral Z
F (r) · dr
γ
is independent of the choice of curve γ.

(iii) For any simple closed curve γ then
Z
F (r) · dr = 0.
γ
Proof.
(i) =⇒ (ii) : This was just proved in Proposition 61 as φ (q) − φ (p) is dependent only on the
endpoints p and q and not on the path taken.
(ii) =⇒ (i) : Let p ∈ S be a fixed point and define for any q ∈ S
Z q
φ (q) = F (r) · dr
p

which, by assumption, is independent of the curve taken and is a function only of q.
As S is open there is an r > 0 small enough that q + ti is in S for 0 6 t 6 r. Then

Z q+ti
φ (q + ti) − φ (q) = F (r) · dr
q
where the curve from q to q + ti can be taken to be a straight line. So

Z s=t Z s=t
φ (q + ti) − φ (q) = F (q + si) · (i ds) = F1 (q + si) ds
s=0 s=0
where F = (F1 , F2 , F3 ). Hence

s=t
φ (q + ti) − φ (q)
Z
∂φ 1
(q) = lim = lim F1 (q + si) ds = F1 (q)
∂x t→0 t t→0 t s=0
by the continuity of F1 at q. Similarly

∂φ ∂φ
(q) = F2 (q) , (q) = F3 (q)
∂y ∂z
and so ∇φ = F as required.
Rq
(ii) =⇒ (iii) : Suppose now that for any two points p and q the line integral p
F (r) · dr is
independent of the curve taken.
Let γ be a simple closed curve in S and let p and q be two distinct points on γ. Then p and
q split γ into two curves γ1 and γ2 , with γ1 running from p to q and γ2 running from q to p.
So, by (ii) and Remark 52,
Z Z Z Z q Z q
F (r) · dr = F (r) · dr + F (r) · dr = F (r) · dr − F (r) · dr = 0
γ γ1 γ2 p p
(iii) =⇒ (ii) : Let p, q ∈ S and let γ1 , γ2 be two curves in S, starting at p and ending at q. If
γ is the curve which follows γ1 and comes back around γ2 , so that we have returned to p then,
by assumption and Remark 52,
Z Z Z Z Z
0 = F (r) · dr = F (r) · dr − F (r) · dr =⇒ F (r) · dr = F (r) · dr
γ γ1 γ2 γ1 γ2
and the line integral of F from p to q is independent of the choice of path taken.
CONSERVATIVE FIELDS 49
5. CONTINUITY AND DIFFERENTIABILITY
Below n ≥ 1 is an integer. Recall that a set U ⊆ Rn is said to be open if for every x ∈ U there
exists δ > 0 such that
" n #1/2
X
B (x; δ) = {y ∈ Rn : |y − x| < δ} ⊂ U , where |y − x| = (yi − xi )2 ,
i=1
where xi , yi are the ith components of x, y and B (x; δ) denotes the open ball of radius δ
about x.
Note B (x; δ),for δ > 0 is also referred to as a open neighbourhood of x.
Definition 67 Define
Bδ0 = B 0 (x; δ) = B (x; δ) \ {x}
as the deleted neighbourhood of {x}.
Recall for a scalar function of a single variable: F (x) tends to l as x tends to a if ∀ > 0 ∃ δ > 0
such that
|F (x) − l| < for all x such that 0 < |x − a| < δ. (5.1)
Definition 68 Limit. Let f : Rn → R , x, a ∈ Rn and l ∈ R. Then we define that the limit

of f (x) as x tends to a is equal to l if and only if
∀ > 0 ∃ δ > 0 such that

|f (x) − l| < ∀x ∈ B 0 (a, δ). (5.2)
We write f (x) → l as x → a or limx)toa f (x) = l.
Example 69 Consider the following two dimensional example. Let

( 2
x y
2 2 (x, y) 6= (0, 0)
f1 (x, y) = x +y . (5.3)
0 (x, y) = (0, 0)
Show that f1 (x, y) → 0 as (x, y) → (0, 0).
We have that 2 2
xy xy
|f1 (x, y) − 0| = 2
− 0=
x2 + y 2 .
(5.4)
x + y2
Given any > 0 we want to find δ > 0 such that |x2 y/(x2 + y 2 )| < whenever
0 < (x2 + y 2 )1/2 < δ.
CONTINUITY AND DIFFERENTIABILITY 51

The form suggests that we use polar coordinates x = r cos θ, y = r sin θ. We then obtain
3
r cos2 θ sin θ

|f1 (x, y) − 0| = − 0 = r cos2 θ sin θ ≤ r < , (5.5)
r2
when r 6= 0 and r = (x2 + y 2 )1/2 < ; so we take δ = .
More generally in two dimensions, if we want to calculate the limit at (a, b) we might like to
take
(x − a) = r cos θ, (y − b) = r sin θ.
Note If a limit exists, its definition requires that f (x) tends to a unique limit value, l, along
any curve C approaching the limit point, a. Hence, if f (x) tends to different limit values along
different curves approaching the limit point, a limit does not exist.
Example 70 Consider (
xy
x2 +y 2
(x, y) 6= (0, 0)
f2 (x, y) = . (5.6)
0 (x, y) = (0, 0)
Show that f (x, y) does not possess a limit as (x, y) → (0, 0).
Approach the limit point (0, 0) on the half-line y = λx, for lambda constant. We have, for
x > 0,
λx2 λ
f2 (x, λx) = 2 = ,
x + λ 2 x2 1 + λ2
which depends on λ in the limit x → 0 and hence the path into the limit point (0, 0). Hence
f (x, y) does not possess a limit as (x, y) → (0, 0)
Definition 71 Continuity Let f : Rn → R , x, a ∈ Rn . A function f (x) which is defined in

some open neighbourhood of a is continuous at a if and only if f (x) → f (a) as x → a.
5.1 Which functions are continuous?
If A, B are constants, f, g are functions that are continuous at a then Af + Bg and f g are
continuous at a. If g(a) 6= 0 then 1/g is continuous at a.
Further, let x1 , a1 WLOG be the first components of x, a respectively. If f (x) = g(x1 ) and g
is continuous at a1 , then f (x) is continuous at a.
Example 72 Thus, for instance, in two dimensions all polynomials in x, y are continuous and
if p(x, y), q(x, y) are polynomials such that q(a, b) 6= 0 then p/q is continuous at (a, b). Similar
comments apply to cos(x + y) = cos x cos y − sin x sin y etc.
52 CONTINUITY AND DIFFERENTIABILITY

Example 73 Reconsider
(
x2 y
x2 +y 2
(x, y) 6= (0, 0)
f1 (x, y) = .
0 (x, y) = (0, 0)
. From the general considerations above, x2 y/(x2 + y 2 ) is continuous, except possibly at (0, 0).
But, as shown already, f1 (x, y) → 0 = f1 (0, 0) as (x, y) → (0, 0). So f1 is also continuous at
(0, 0).
Example 74 Reconsider
(
xy
x2 +y 2
(x, y) 6= (0, 0)
f2 (x, y) = .
0 (x, y) = (0, 0)
This is not continuous at (0, 0), as the limit does not exist. However, f2 (x, 0) = 0 → 0 as
x → 0, so f2 (x, 0) is continuous at x = 0 and similarly f2 (0, y) is continuous at y = 0. Thus,
continuity in the pair (x, y) is stronger than continuity in each variable separately.
5.2 Continuous differentiability
Recall that
∂f f (a + h, b) − f (a, b)
(a, b) = fx (a, b) = lim , (5.7)
∂x h→0 h
and
∂f f (a, b + h) − f (a, b)
(a, b) = fy (a, b) = lim . (5.8)
∂y h→0 h
But the fact that f (x, y) has partial derivatives with respect to x and y does not imply differ-
entiability: in general, it does not even imply continuity!
Example 75 (
x+y if x = 0 or y = 0
f3 (x, y) = . (5.9)
1 otherwise
Note that f3 is not continuous at (0, 0) but it does have partial derivatives with respect to x and
y at (0, 0):
∂f2 ∂f2
(0, 0) = 1, (0, 0) = 1.
∂x ∂y
Definition 76 Continuously differentiable. Let f : Rn → R, x ∈ Rn . A function f (x)
is continuously differentiable in some open set, U ⊂ Rn if and only if its first order partial
derivatives exist and are continuous in U .
We write f ∈ C 1 (D), where the superscript 1 stands for first order derivatives and C denotes
continuous (i.e. f ∈ C(U ) when f is continuous in U ).
CONTINUOUS DIFFERENTIABILITY 53
Note
f continuously differentiable =⇒ f differentiable =⇒ f continuous.
Definition 77 C n functions. Let f : Rn → R, x ∈ Rn . The function f (x) is said to be n

times continuously differentiable on an open subset U ⊂ Rn if any partial derivatives up to and
including any nth partial derivative exist and are continuous. We write f ∈ C n (U ).
Finally, we remark that often one fins the order partial differentiation is performed for second
and higher order partial derivatives does not alter the final result. This is exemplified in the
following theorem:
∂2f ∂2f
Theorem 78 Let f : R2 → R be such that ∂y∂x
and ∂x∂y
exist and are continuous. Then
∂ 2f ∂ 2f
= .
∂y∂x ∂x∂y
Proof. For curiosity/reference only. With x, y, h, k ∈ R define
φ (x, y) = f (x, y + k) − f (x, y) , and ψ (x, y) = f (x + h, y) − f (x, y)
so that
D (x, y) = f (x + h, y + k) − f (x + h, y) − f (x + h, y) + f (x, y)
= φ (x + h, y) − φ (x, y) = ψ (x, y + k) − ψ (x, y) .
By the Mean-Value Theorem, applied twice, there exist θ1 , θ2 ∈ (0, 1) such that
D (x, y) = φ (x + h, y) − φ (x, y) = hφx (x + θ1 h, y)

= h [fx (x + θ1 h, y + k) − fx (x + θ1 h, y)] = hkfxy (x + θ1 h, y + θ2 k) .
Arguing similarly with from the D (x, y) = ψ (x, y + k) − ψ (x, y) expression we know there
exist θ3 , θ4 ∈ (0, 1) such that
D (x, y) = hkfyx (x + θ3 h, y + θ4 k) .
So
fxy (x + θ1 h, y + θ2 k) = fyx (x + θ3 h, y + θ4 k)
Letting h and k tend to 0, and using the continuity of fxy and fyx we see that fxy = fyx as
required.
54 CONTINUITY AND DIFFERENTIABILITY

6. DIV, GRAD AND CURL
Recall:
Definition 79 By a scalar field φ on R3 we shall mean a map φ : R3 → R.
Definition 80 By a vector field F on R3 we shall mean a map F : R3 → R3 . Occasionally

we may consider more general scalar fields φ : Rn → R and vector fields F : Rn → Rm .
We shall also consider scalar and vector fields defined on proper subsets of R3 (or more generally
Rn ). The domains of these fields will usually be open, so that we can define their partial
derivatives.
6.1 Definitions and Properties
Definition 81 (Gradient) Let φ : R3 → R be a scalar field. Then the gradient of φ, written

grad φ, is

∂φ ∂φ ∂φ
grad φ = , , .
∂x ∂y ∂z
Note that grad takes scalar fields to vector fields.
Definition 82 (Divergence) Let F : R3 → R3 be a vector field where
F (x, y, z) = (F1 (x, y, z) , F2 (x, y, z) , F3 (x, y, z)) ,
with respect to a Cartesian coordinate system. Then the divergence of F, written div F, is
∂F1 ∂F2 ∂F3

div F = + + .
∂x ∂y ∂z
Note that div takes vector fields to scalar fields.
Definition 83 (Curl) Let F : R3 → R3 be a vector field where
F (x, y, z) = (F1 (x, y, z) , F2 (x, y, z) , F3 (x, y, z)) .
DIV, GRAD AND CURL 55

with respect to a Cartesian coordinate system. Then the curl of F, written curl F, is

i
∂ ∂j k

∂
∂F 3 ∂F 2 ∂F 1 ∂F 3 ∂F 2 ∂F 1
curl F = ∂x ∂y ∂z = − , − , − .
F1 F2 F3

∂y ∂z ∂z ∂x ∂x ∂y
Note that curl takes vector fields to vector fields.
p
Example 84 Let r = (x, y, z) and r = |r| = x2 + y 2 + z 2 . Then
∂x ∂y ∂z
div r = + + = 3;
∂x ∂y ∂z

i j k
∂ ∂ ∂
curl r = ∂x ∂y ∂z = 0.
x y z
Example 85 With r and r as above, and f a differentiable function on (0, ∞) note

i j k
∂ ∂ ∂

curl (f (r) r) = ∂x

∂y ∂z

f (r) x f (r) y f (r) z

0 ∂r 0 ∂r 0 ∂r 0 ∂r 0 ∂r 0 ∂r
= zf (r) − yf (r) , xf (r) − zf (r) , yf (r) − xf (r)
∂y ∂z ∂z ∂x ∂x ∂y
zy yz xz zx yx xy
= f 0 (r) − , − , − = 0.
r r r r r r
Example 86 Let c = (c1 , c2 , c3 ) be a constant vector. Then
∂ ∂ ∂
div (c ∧ r) = (c2 z − c3 y) + (c3 x − c1 z) + (c1 y − c2 x) = 0;
∂x ∂y ∂z

i j k
∂ ∂ ∂

curl (c ∧ r) = ∂x ∂y ∂z
= (2c1 , 2c2 , 2c3 ) = 2c.

c2 z − c3 y c3 x − c1 z c1 y − c2 x
Definition 87 The differential operator

∂ ∂ ∂
∇= , ,
∂x ∂y ∂z
is called del or nabla.
56 DIV, GRAD AND CURL

Notation 88 By a slight abuse of notation we can write

∂φ ∂φ ∂φ ∂ ∂ ∂
grad φ = , , = , , φ;
∂x ∂y ∂z ∂x ∂y ∂z

∂F1 ∂F2 ∂F3 ∂ ∂ ∂
div F = + + = , , · (F1 , F2 , F3 ) ;
∂x ∂y ∂z ∂x ∂y ∂z

∂F3 ∂F2 ∂F1 ∂F3 ∂F2 ∂F1
curl F = − , − , −
∂y ∂z ∂z ∂x ∂x ∂y

i
∂ ∂j k

∂
∂ ∂ ∂
= ∂x ∂y ∂z =
, , ∧ (F1 , F2 , F3 ) .
F1 F2 F3 ∂x ∂y ∂z
So we often write
grad φ = ∇φ, div F = ∇ · F, curl F = ∇ ∧ F. (6.1)
Remark 89 The notation grad, div, curl and that of (??) are equally common in mathemat-
ical texts, and so these lecture notes and the problem sheets will deliberately make use of both
notations.
Remark 90 For any scalar field φ we have
∂ 2φ ∂ 2φ ∂ 2φ

∂ ∂φ ∂ ∂φ ∂ ∂φ
div (grad φ) = + + = + + 2,
∂x ∂x ∂y ∂y ∂z ∂z ∂x2 ∂y 2 ∂z
which is the Laplacian. As
div grad = ∇ · ∇
then we often write
∂ 2φ ∂ 2φ ∂ 2φ
∇2 φ = + + 2.
∂x2 ∂y 2 ∂z
Also, for vector fields F = (F1 , F2 , F3 ) with respect to a Cartesian coordinate system, it is
common to write
∇2 F = ∇2 F1 , ∇2 F2 , ∇2 F3 .

Thus the Laplacian can both take scalar fields to scalar fields and vector fields to vector fields.
Remark 91 Whilst ∇ looks, and occasionally behaves, like a standard vector it is very unwise
to assume this behaviour of ∇ without proof; most standard vector identities are not true when
they involve ∇. For example, the commutativity of the dot product does not extend:
∇ · F 6= F · ∇.
The LHS is div F. The RHS at first glance doesn’t mean anything, it certainly isn’t a scalar.
Rather

∂ ∂ ∂ ∂ ∂ ∂
F · ∇ = (F1 , F2 , F3 ) · , , = F1 + F2 + F3
∂x ∂y ∂z ∂x ∂y ∂z
which is a differential operator. In fact F · ∇ is standard notation for it and it gives the
directional derivative in the direction of F.
DEFINITIONS AND PROPERTIES 57

In summary, we have the following table:
operator maps:
div = ∇· vector fields to scalar fields
grad = ∇ scalar fields to vector fields
curl = ∇∧ vector fields to vector fields
div grad = ∇2 scalar fields to scalar fields
vector fields to vector fields
F·∇ scalar fields to scalar fields
vector fields to vector fields
The next proposition shows that the formulae for calculating each of div, grad and curl do
not depend on our choice i, j, k of a Cartesian co-ordinate system. Recall that:
Definition 92 The vectors e1 , e2 , e3 are said to be an orthonormal basis of R3 if

1 if i = j,
ei · ej = δij :=
6 j.
0 if i =
So the vectors are all of unit length and mutually perpendicular.

Further it is said to be a right-handed orthonormal basis if e1 ∧ e2 = e3 and left-handed if
e1 ∧ e2 = −e3 . So e1 = i, e2 = j, e3 = k is an example of a right-handed orthonormal basis.
Proposition 93 Let e1 , e2 , e3 be a right-handed orthonormal basis of R3 with associated co-

ordinates X, Y, Z defined by the identity
Xe1 + Y e2 + Ze3 = xi + yj + zk. (6.2)
Then
∂ ∂ ∂ ∂ ∂ ∂
e1 + e2 + e3 =i +j +k .
∂X ∂Y ∂Z ∂x ∂y ∂z

Proof. From (6.2) we have that
x = (e1 · i) X + (e2 · i) Y + (e3 · i) Z,
X = (e1 · i) x + (e1 · j) y + (e1 · k) z,
and four other similar equations for y, z, Y, Z. It follows from the chain rule that
∂ ∂x ∂ ∂y ∂ ∂z ∂ ∂ ∂ ∂
= + + = (e1 · i) + (e1 · j) + (e1 · k)
∂X ∂X ∂x ∂X ∂y ∂X ∂z ∂x ∂y ∂z
and hence
∂ ∂ ∂
e1 + e2 + e3
∂X
∂Y ∂Z
∂ ∂ ∂
= e1 (e1 · i) + (e1 · j) + (e1 · k) + e2 [. . .] + e3 [. . .]
∂x ∂y ∂z
∂ ∂ ∂
= [e1 (e1 · i) + e2 (e2 · i) + e3 (e3 · i)] + [. . .] + [. . .]
∂x ∂y ∂z
∂ ∂ ∂
= i +j +k ;
∂x ∂y ∂z
this last line follows from the orthonormality of the bases as
i = αe1 + βe2 + γe3
leads to α = e1 · i, β = e2 · i, γ = e3 · i, by dotting the equation with e1 , e2 , e3 respectively.
Corollary 94 Let f : R3 → R3 be a vector field given by f = (f1 , f2 , f3 ), let φ : R3 → R be a

scalar field and let e1 , e2 , e3 be a right-handed orthonormal basis of R3 . Suppose that
F1 e1 + F2 e2 + F3 e3 = f1 i + f2 j + f3 k;
Φ (X, Y, Z) = φ (x, y, z) .

That is, F and Φ represent f and φ in the new co-ordinates X, Y, Z. Then
∂Φ ∂Φ ∂Φ
grad φ = e1 + e2 + e3 ;
∂X ∂Y ∂Z
∂F1 ∂F2 ∂F3
div f = + + ;
∂X ∂Y ∂Z

e1 e2 e3
∂ ∂ ∂

curl f = ∂X ∂Y ∂Z
.
F1 F2 F3
That is, we calculate grad, div and curl using the same formulae, irrespective of what right-
handed orthonormal co-ordinate system we are using.
Proof. Sketch Proof The previous proposition showed that the formula for grad does not
change under an orthonormal change of co-ordinates. From Geometry, we have that the formula
for the dot product is similarly invariant as is the formula for the cross product provided the
change is to another right-handed system. Hence the formulae for div = ∇· and curl = ∇∧ do
not change when we change to another right-handed orthonormal system.
Adjustments to the above formulae need to be made for co-ordinate systems that aren’t or-
thonormal – e.g. cylindrical polar co-ordinates or spherical polar co-ordinates. The relevant
formulae below are for reference only; certainly you are not expected to memorize them. The
main point to take away from this is that the grad, div, curl formulae are different for co-
ordinate systems that are not Cartesian.
Definition 95 Cylindrical polar co-ordinates r, θ, z are defined by the identity
r = (x, y, z) = (r cos θ, r sin θ, z) .
If we note
dr = (cos θ, sin θ, 0) dr + (−r sin θ, r cos θ, 0) dθ + (0, 0, 1) dz
then we may write
dr = er dr + reθ dθ + ez dz
where
er = (cos θ, sin θ, 0) , eθ = (− sin θ, cos θ, 0) , ez = (0, 0, 1) .
Note, for any values of r, θ, z, the vectors er , eθ , ez make a right-handed orthonormal basis of
R3 .
Proposition 96 The formulae for grad, div, curl in terms of cylindrical polar co-ordinates of

fields ψ (r, θ, z) and F = Fr er + Fθ eθ + Fz ez are
∂ψ 1 ∂ψ ∂ψ
∇ψ = er + eθ + ez ;
∂r r ∂θ ∂z
1 ∂ 1 ∂Fθ ∂Fz
∇·F = (rFr ) + + ;
r ∂r
r ∂θ ∂z
e re e
1 ∂r ∂ θ ∂z
∇∧F = .
r ∂r ∂θ ∂z
Fr rFθ Fz
Proof. By way of example we shall prove only the first identity here. Recall that Cartesian
co-ordinates are given in terms of cylindrical polar ones by
x = r cos θ, y = r sin θ, z=z
Say φ (x, y, z) = ψ (r, θ, z) . Then
∂ψ ∂φ ∂x ∂φ ∂y ∂φ ∂z ∂φ ∂φ
= + + = cos θ + sin θ,
∂r ∂x ∂r ∂y ∂r ∂z ∂r ∂x ∂y
∂ψ ∂φ ∂x ∂φ ∂y ∂φ ∂z ∂φ ∂φ
= + + = − r sin θ + r cos θ,
∂θ ∂x ∂θ ∂y ∂θ ∂z ∂θ ∂x ∂y
∂ψ ∂φ
= .
∂z ∂z
Hence
∂ψ 1 ∂ψ ∂ψ
er + eθ + ez
∂r
r ∂θ ∂z
∂φ ∂φ 1 ∂φ ∂φ ∂φ
= cos θ + sin θ (cos θ, sin θ, 0) + − r sin θ + r cos θ (− sin θ, cos θ, 0) + k
∂x ∂y r ∂x ∂y ∂z

∂φ 2 ∂φ 2 ∂φ 2 ∂φ 2 ∂φ
= cos θ + sin θ i + sin θ + cos θ j + k
∂x ∂x ∂y ∂y ∂z
= ∇φ.

Definition 97 Spherical polar co-ordinates r, θ, φ are defined by the identity
r = (x, y, z) = (r sin θ cos φ, r sin θ sin φ, r cos θ) .
If we note dr equals
(sin θ cos φ, sin θ sin φ, cos θ) dr + r (cos θ cos φ, cos θ sin φ, − sin θ) dθ + r sin θ (− sin φ, cos φ, 0) dφ
then we may write

dr = er dr + reθ dθ + r sin θeφ dφ
where
er = (sin θ cos φ, sin θ sin φ, cos θ) , eθ = (cos θ cos φ, cos θ sin φ, − sin θ) , eφ = (− sin φ, cos φ, 0) .
Note, for any values of r, θ, φ the vectors er , eθ , eφ , make a right-handed orthonormal basis.
Proposition 98 The formulae for grad, div, curl in terms of spherical polar co-ordinates of
fields ψ (r, φ, θ) and F = Fr er + Fφ eφ + Fθ eθ are
∂ψ 1 ∂ψ 1 ∂ψ
∇ψ = er + eθ + eφ ;
∂r r ∂θ r sin θ ∂φ

1 ∂ 2 ∂ ∂
∇·F = 2 r sin θFr + (r sin θFθ ) + (rFφ ) ;
r sin θ ∂r ∂θ ∂φ

er reθ r sin θeφ
1 ∂ ∂ ∂

∇∧F = 2 ∂r ∂θ ∂φ
.
r sin θ

Fr rFθ r sin θFφ
Those interested in reading up on the more general theory of curvilinear co-ordinates (which
explains the similarities in the previous two propositions) should see Boas (pp.427-435) or
Kreyszig (pp.498-504).

6.2 Identities
Proposition 99 Let F be a vector field on R3 and φ be a scalar field on R3 . Then

curl grad φ = 0; div curl F = 0.
Proof. With the notation ∇φ = (φx , φy , φz ) where subscripts now denote partial differentiation,
we have

i j k
∂ ∂ ∂
curl (∇φ) = ∂x ∂y ∂z = (φzy − φyz , φxz − φzx , φyx − φxy ) = 0.
φx φy φz
Again with subscripts denoting partial differentiation, we also have

∂F3 ∂F2 ∂F1 ∂F3 ∂F2 ∂F1
div (curl F) = div − , − , −
∂y ∂z ∂z ∂x ∂x ∂y
h i h i h i
= (F3 )yx − (F2 )zx + (F1 )zy − (F3 )xy + (F2 )xz − (F1 )yz = 0.
Proposition 100 (Product Rules for div and curl) Let F be a vector field on R3 and φ, ψ be
scalar fields on R3 . Then
div (φF) = ∇φ · F + φ div F;
curl (φF) = φ curl F + ∇φ ∧ F.
Proof. (i) With subscripts denoting partial differentiation we have
div (φF) = (φF1 )x + (φF2 )y + (φF3 )z
= φ (F1 )x + φx F1 + φ (F2 )y + φy F2 + φ (F3 )z + φz F3
h i
= φ (F1 )x + (F2 )y + (F3 )z + (φx , φy , φz ) · (F1 , F2 , F3 )
= φ div F + (∇φ) · F.
(ii) curl (φF) equals

∂ (φF3 ) ∂ (φF2 ) ∂ (φF1 ) ∂ (φF3 ) ∂ (φF2 ) ∂ (φF1 )
− , − , −
∂y ∂z ∂z ∂x ∂x ∂y

∂F3 ∂F2 ∂F1 ∂F3 ∂F2 ∂F1
= φ − , − , − + (φy F3 − φz F2 , φz F1 − φx F3 , φx F2 − φy F1 )
∂y ∂z ∂z ∂x ∂x ∂y

i j k

= φ curl F + φx φy φz
F1 F2 F3
= φ curl F + (∇φ) ∧ F.
IDENTITIES 63
Proposition 101 (Further Identities) For vector fields F, G:
∇ (F · G) = (F · ∇) G + (G · ∇) F + F ∧ (∇ ∧ G) + G ∧ (∇ ∧ F)
∇ · (F ∧ G) = G · (∇ ∧ F) − F · (∇ ∧ G)
∇ ∧ (F ∧ G) = F (∇ · G) − G (∇ · F) + (G · ∇) F − (F · ∇) G
∇ ∧ (∇ ∧ F) = ∇ (∇ · F) − ∇2 F
Remark 102 All vector calculus can be proven by simply grinding out the calculations. But
there are neater ways to write out the formula for div, grad and curl as given below:
X ∂φ X ∂F X ∂F X ∂
∇φ = ei , ∇·F = ei · , ∇∧F = ei ∧ , F·∇ = (F · ei ) ,
i
∂xi i
∂xi i
∂xi i
∂xi
where the dummy variable i ranges over 1, 2, 3 in each case and e1 , e2 , e3 is any right-handed
orthonormal basis. For example
X ∂F ∂F ∂F ∂F
ei ∧ = e1 ∧ + e2 ∧ + e3 ∧
i
∂xi ∂x1 ∂x2 ∂x3

e1 e2 e3 e1 e2 e3 e1 e2 e3

= 1 0 0 + 0 1 0 + 0 0 1
∂F1 ∂F2 ∂F3 ∂F1 ∂F2 ∂F3 ∂F1 ∂F2 ∂F3
∂x1 ∂x1 ∂x1 ∂x2 ∂x2 ∂x2 ∂x3 ∂x3 ∂x3
∂F3 ∂F2 ∂F3 ∂F1 ∂F2 ∂F1
= −e2 + e3 + e1 − e3 − e1 + e2
∂x1 ∂x1 ∂x2 ∂x2 ∂x3 ∂x3
= ∇ ∧ F.
With these formulae the proofs of identities involving div, grad and curl are much shorter; on
the other hand the formulae need to be applied with much more care and attention to detail than
was previously the case.
Proof. The second and third identities appear in the Exercise Sheets. To prove the first
identity recall that u ∧ (v ∧ w) = (u · w) v − (u · v) w and hence
(F · ∇) G + (G · ∇) F + F ∧ (∇ ∧ G) + G ∧ (∇ ∧ F)
! !
X ∂G X ∂F X ∂G X ∂F
= (F · ei ) + (G · ei ) +F∧ ei ∧ +G∧ ei ∧
i
∂x i i
∂x i i
∂x i i
∂xi

X ∂G X ∂F X ∂G X ∂F
= (F · ei ) + (G · ei ) + F ∧ ei ∧ + G ∧ ei ∧
i
∂x i i
∂x i i
∂x i i
∂xi

X ∂G ∂F ∂G ∂G ∂F ∂F
= (F · ei ) + (G · ei ) + F· ei − (F · ei ) + G· ei − (G · ei )
i
∂x i ∂x i ∂x i ∂x i ∂x i ∂xi

X ∂G ∂F
= ei F · +G·
i
∂xi ∂xi
X ∂
= ei (F · G) = ∇ (F · G) .
i
∂x i

To prove the fourth identity recall again that u ∧ (v ∧ w) = (u · w) v − (u · v) w and then
!
X ∂ X ∂F
∇ ∧ (∇ ∧ F) = ei ∧ ej ∧
i
∂xi j
∂xj
∂ 2F
XX
= ei ∧ ej ∧
i j
∂xi ∂xj
∂ 2F ∂ 2F
XX XX
= ei · ej − (ei · ej )
i j
∂xi ∂xj i j
∂xi ∂xj
" #
XX ∂ 2F

X ∂ X ∂F
= ej ei · − δij
j
∂x j i
∂x i i j
∂xi ∂xj
" #
X ∂ X ∂F X ∂ 2F
= ej ei · − 2
= ∇ (∇ · F) − ∇2 F
j
∂x j i
∂x i i
∂x i
IDENTITIES 65
6.3 Physical Interpretation
Example 103 (Fourier’s Law) In an isotropic medium with constant thermal conductivity
k, with temperature T (x, t) at position x and at time t, Fourier’s Law states that
q = −k∇T
where q is the heat flux – that is the amount of energy that flows through a particular surface
per unit area per unit time.
This law contains much of the essence of grad . Recall from Michaelmas Calculus that, for a
given scalar function φ and a unit vector v, the directional derivative of φ in the direction v
is ∇φ · v. In particular, φ increases fastest in the direction of ∇φ and decreases fastest in the
direction −∇φ. So Fourier’s law, loosely put, states that at each point the heat energy moves
towards the coolest nearby points.
How about the physical motivation behind divergence?
Example 104 Consider fluid particles in the xy-plane flowing according to the steady flow
Note how we have a source at (0, 0) from which the fluid emanates. However we have, at every
(x, y) , that
∂x ∂y
div u = + = 2.
∂x ∂y
Suppose a particle starts off at (x0 , y0 ); then the particle’s position (x (t) , y (t)) at time t is
determined by the differential equations
dx dy
= x, = y, x (0) = x0 , y (0) = y0
dt dt

and hence
(x (t) , y (t)) = x0 et , y0 et .

If we consider the blob of fluid of area πε2 which at t = 0 occupies
(x − 1)2 + y 2 < ε2 ,
then where will the particles which make up this blob have moved by time t? Initially the
particles were at x0 = 1 + r cos θ, y = r sin θ where r < ε and they have moved to
(x (t) , y (t)) = et + ret cos θ, ret sin θ

r < ε,
which comprises a disc with centre (et , 0) and area A (t) = πε2 e2t . So throughout the motion we
have had
dA
= (div u) A.
dt
More generally, for a two dimensional flow, it is the case that

1 d
∇·u (x) = lim area (Flow image of B (x, ε) at time t) ,
ε→0 area (Flow image of B (x, ε) at time t) dt
with the analogous result for 3D flows in terms of volume rather than area. In physical terms,
then, the divergence of a vector field is a measure of the flow’s source density, the extent to
which the vector field flow behaves like a source or a sink at a given point.
If you choose to take the second year option in fluid dynamics you will meet:
Example 105 Euler’s Equations for an ideal fluid (Second year course in Fluid Dy-
namics)
Euler’s equations for an incompressible, inviscid fluid state that
∂u ∇p
+ (u · ∇) u = − + g, ∇ · u = 0,
∂t ρ
where u (x, t) is the flow velocity, p is the pressure, ρ is the density and g is acceleration due
to gravity. It is the equation
∇·u=0
which encodes the incompressibility of the fluid.
Such vector calculus are found throughout the applications of mathematics, for instance Maxwell’s
equations of electromagnetism, fluid and solid mechanics and cellular behaviour.
PHYSICAL INTERPRETATION 67
Roughly speaking the curl of a vector field relates to the rotation of the vector field, but very
much in a local sense.
Example 106 (Rotation of a rigid body) Recall from the Michaelmas Dynamics course
that if r = rer then
ṙ = ṙer + rθ̇eθ .
In particular, if a rigid body is rotating with constant angular velocity ω about the z-axis then
the distance of a point r from the z-axis will remain constant and θ̇ = ω so that
ṙ = rωeθ = ωk∧ (rer ) = ωk ∧ r.
We have already seen from Example 86 that
curl ṙ = curl (ωk ∧ r) = 2ωk.

Example 107 It is important to stress though that curl is very subtly a measure of local rota-
tion. Consider the following two examples:
A simple calculation shows that curl u2 = −ck yet there appears to be no sense of global rotation,
whilst in Exercise 5 on Sheet 1 you are asked to show that curl u1 = 0, yet there appears to
be a clear sense of global rotation. Despite particles moving in circles in the flow u1 different
circles are moving around the origin with different angular velocities and with a greater angular
velocity towards the origin; in fact this differential is such that a circular paddle wheel inserted
into the fluid would not spin as it moves around the origin. In contrast, with the shear flow
u2 , whilst the particles are moving in lines, a small paddle wheel inserted into the flow would
spin as the flow for greater y is moving faster. In general, the axis of this spinning is in the
direction of curl .
PHYSICAL INTERPRETATION 69
7. DIVERGENCE AND STOKES’ THEOREMS
7.1 The Divergence Theorem
Theorem 108 (Divergence Theorem – Lagrange 1762, Gauss 1813, Green 1825) Let R be
a region of R3 with a piecewise smooth boundary ∂R, and let F be a differentiable vector field
on R. Then ZZZ ZZ
div F dV = F · dS,
R ∂R
where dS is oriented in the direction of the outward pointing normal from R.
Definition 109 We say that a region R is convex if for any p, q ∈ R the line segment
connecting p and q is contained in R.
We will only prove the Divergence Theorem for convex regions, though the proof extends with
very little extra work to any region that can be decomposed into convex regions.
Proof. (Not examinable.) Let W be a subset of the xy-plane and α, β continuous functions
on W .
Let
R = (x, y, z) ∈ R3 : (x, y) ∈ W, α (x, y) 6 z 6 β (x, y)

and F =F k is a differentiable vector field on R. (That is R is z-convex in the sense that the
intersection of R with any line parallel with the z-axis is an interval.) The Divergence Theorem
in this case reads as ZZZ ZZ
∂F
dV = F k · dS.
∂z
R ∂R
DIVERGENCE AND STOKES’ THEOREMS 71

Note that ∂R is made up of three surfaces:
Σ+ = (x, y, z) ∈ R3 : (x, y) ∈ W, z = β (x, y) ,

Σ = (x, y, z) ∈ R3 : (x, y) ∈ ∂W, α (x, y) 6 z 6 β (x, y) ,

Σ− = (x, y, z) ∈ R3 : (x, y) ∈ W, z = α (x, y) .

Then
ZZZ ZZ Z β(x,y) ZZ
∂F ∂F
dV = dz dA = (F (x, y, β (x, y)) − F (x, y, α (x, y))) dA.
∂z α(x,y) ∂z
R W W
Regarding the boundary integral:

i j k

on Σ+ : dS = 1 0 βx dx dy = (−βx , −βy , 1) dx dy;

0 1 βy

i j k
on Σ− :

dS = − 1 0 αx dx dy = (αx , αy , −1) dx dy;

0 1 αy
on Σ : k · dS = 0.
Hence we have
ZZ ZZ
F k · dS = F (x, y, β (x, y)) dx dy;
+ W
ZΣZ ZZ
F k · dS = − F (x, y, α (x, y)) dx dy;
− W
ZΣZ
F k · dS = 0;
Σ
and we have shown ZZZ ZZ

∂F
dV = F k · dS
∂z
R ∂R
for the given region R. However, any convex body has the property of being in the form of R
when viewed from each of the x, y, z directions. Hence we have shown
ZZZ ZZ
div F dV = F · dS
R ∂R
for any convex body.
72 DIVERGENCE AND STOKES’ THEOREMS

Corollary 110 Let φ be a smooth scalar field defined on a region R ⊆ R3 with a piecewise
smooth boundary. Then ZZZ ZZ
∇φ dV = φ dS.
R ∂R
Proof.
Let c be a constant vector and F = cφ. Then

div (cφ) = c · ∇φ + φ div c = c · ∇φ
as c is constant. By the Divergence Theorem
   
ZZZ ZZ ZZZ ZZ
c · ∇φ dV = cφ · dS, =⇒ c ·  ∇φ dV  = c ·  φdS .
R ∂R R ∂R
As c is an arbitrary vector then

ZZZ ZZ
∇φ dV = φ dS.
R ∂R
Example 111 Verify the Divergence Theorem when R is the finite region bounded by the
paraboloid z = x2 + y 2 and the plane x + y + z = 1, and F = (x, y, z) .
Solution. We already met this region in Example 19. We showed that the top surface of R
projects vertically down to the set
W = (x, y) ∈ R2 : (x + 1/2)2 + (y + 1/2)2 6 3/2

p
which is a disc, centre (−1/2, −1/2) and radius 3/2. Further we calculated the volume of
the region to be 9π/8. As div F = 3 then
ZZZ
27π
div F dV = 3 Vol (R) = .
8
R
The boundary ∂R is made up of two surfaces

Σ1 = (x, y, z) : z > x2 + y 2 , x + y + z = 1 = {(x, y, z) : (x, y) ∈ W, x + y + z = 1} ;

Σ2 = (x, y, z) : z = x2 + y 2 , x + y + z 6 1 = (x, y, z) : (x, y) ∈ W, z = x2 + y 2 .

We can parameterise Σ1 and Σ2 as

r2 (u, v) = u, v, u2 + v 2 .

r1 (u, v) = (u, v, 1 − u − v) ;
THE DIVERGENCE THEOREM 73

Note
i j k

(r1 )u ∧ (r1 )v = 1 0 −1 = (1, 1, 1)

0 1 −1
which is outward-pointing from R, and

i j k

(r2 )u ∧ (r2 )v = 1 0 2u = (−2u, −2v, 1)

0 1 2v
which is inward-pointing and so we will take its negative. So

ZZ ZZ ZZ p 2
F · dS = (u, v, 1 − u − v) · (1, 1, 1) du dv = du dv = π 3/2 = 3π/2.
Σ1 W W
For the second surface

ZZ ZZ
u, v, u2 + v 2 · (2u, 2v, −1) dA

F · dS =
Σ2 W
ZZ
u2 + v 2 dA

=
W
√
Z 3/2 Z 2π
(−1/2 + r cos θ)2 + (−1/2 + r sin θ)2 r dθ dr

=
r=0 θ=0
√
Z 3/2 Z 2π
r2 − r cos θ − r sin θ + 1/2 r dθ dr

=
r=0 θ=0
√
Z 3/2 n
ro
= 2π r3 + dθ dr
r=0 2
√ 3/2
r4 r2

= 2π +
4 4 0
15π
= 2π (9/16 + 3/8) = .
8
Hence, in total, we have
ZZ
3π 15π 12π 15π 27π
F · dS = + = + = .
2 8 8 8 8
∂R
Example 112 The twice continuously differentiable function f (x, y, z) is homogeneous of de-
gree n, so that
f (tx, ty, tz) = tn f (x, y, z) for all t ∈ R. (7.1)

Prove Euler’s Theorem, which states that
r · ∇f = nf.
Let B be the unit ball x2 + y 2 + z 2 < 1 and let ∂B be its boundary. Show that
ZZZ ZZZ
1
f dV = ∇2 f dV,
n (n + 3)
B B
and hence show that

ZZZ
36π
(x − y + z)4 dV = .
35
B
Solution. If we differentiate the identity (7.1) with respect to t then we find
∂f ∂f ∂f
x (tx, ty, tz) + y (tx, ty, tz) + z (tx, ty, tz) = ntn−1 f (x, y, z) .
∂x ∂y ∂z
If we set t = 1 then we arrive at Euler’s Theorem r · ∇f = nf.

By the Divergence Theorem, with B as the unit ball, and ∂B its boundary, then
ZZZ ZZZ ZZ ZZ ZZ
2
∇ f dV = ∇ · ∇f dV = ∇f · n dS = ∇f · r dS = n f dS
B B ∂B ∂B ∂B
as n = r on ∂B. On the other hand
∇ · (f r) = (∇f ) · r + f (∇ · r) = nf + 3f = (n + 3) f.
and so, by the Divergence Theorem again,

ZZZ ZZZ ZZ ZZ
(n + 3) f dV = ∇ · (f r) dV = f r · n dS = f dS
B B ∂B ∂B
as ∂B is the unit sphere. Hence

ZZZ ZZZ
1
f dV = ∇2 f dV,
n (n + 3)
B B
THE DIVERGENCE THEOREM 75

The function (x − y + z)4 is homogeneous of degree 4 in x, y, z and so that
ZZZ
(x − y + z)4 dV
B
ZZZ
1
= ∇2 (x − y + z)4 dV
4×7
Z ZBZ
1
= 36 (x − y + z)2 dV [which is again homogeneous]
4×7
B
ZZZ
36 1
= × ∇2 (x − y + z)2 dV
4×7 2×5
Z ZBZ
36 1
= × 6 dV
4×7 2×5
B
36 6 4π
= × ×
4×7 2×5 3
36π
= .
35
7.2 Applications of The Divergence Theorem
7.2.1 Boundary value problems
The Divergence Theorem also has important implications for certain boundary-value problems.
Corollary 113 (Uniqueness of the Dirichlet Problem) Let R ⊆ R3 be a path-connected
region with piecewise smooth boundary ∂R and let f be a continuous function defined on ∂R.
Suppose that φ1 and φ2 are such that
∇2 φ1 = 0 = ∇2 φ2 in R;
φ1 = f = φ2 on ∂R.
Then φ1 = φ2 in R.
Proof. Let ψ = φ1 − φ2 , so that
∇2 ψ = 0 in R, ψ = 0 on ∂R.
If we consider the function ψ 2 then, by the Divergence Theorem and as ∇2 = ∇ · ∇,

ZZ ZZZ
2
∇2 ψ 2 dV.

∇ ψ · dS =
∂R R
Looking at the LHS,

ZZ ZZ
2

∇ ψ · dS = 2ψ∇ψ · dS = 0 as ψ = 0 on ∂R.
∂R ∂R
On the RHS we have
∇2 ψ 2 = 2ψ∇2 ψ + 2∇ψ · ∇ψ = 2 |∇ψ|2 as ∇2 ψ = 0 in R.

Hence ZZZ
2 |∇ψ|2 dV = 0 =⇒ ∇ψ = 0 in R.
R
As ∇ψ = 0 in R and R is path-connected then ψ is constant throughout R. But ψ = 0 on ∂R

and so by continuity ψ = 0 throughout R.
Corollary 114 (Uniqueness, up to a constant, of the Neumann Problem) Let R ⊆

R3 be a path-connected region with piecewise smooth boundary ∂R and let f be a continuous
function defined on ∂R. Suppose that φ1 and φ2 are such that
∇2 φ1 = 0 = ∇2 φ2 in R;
∂φ1 ∂φ2
= f= on ∂R.
∂n ∂n
Then φ1 − φ2 is constant in R.

∂ψ
∇2 ψ = 0 in R, = ∇ψ · n = 0 on ∂R.
∂n
If we consider the function ψ 2 , then by the Divergence Theorem and as ∇2 = ∇ · ∇,
ZZ ZZZ
2
∇2 ψ 2 dV.

∇ ψ · dS =
∂R R
Looking at the LHS,

ZZ ZZ
2

∇ ψ · dS = 2ψ∇ψ · n dS = 0 as ∇ψ · n = 0 on ∂R.
∂R ∂R
APPLICATIONS OF THE DIVERGENCE THEOREM 77

On the RHS we have
∇2 ψ 2 = 2ψ∇2 ψ + 2∇ψ · ∇ψ = 2 |∇ψ|2 as ∇2 ψ = 0 in R.

Hence ZZZ
2 |∇ψ|2 dV = 0 =⇒ ∇ψ = 0 in R.
R
As ∇ψ = 0 in R then ψ is constant throughout R.
Corollary 115 (Robin (or Mixed) Boundary Problem) Let R ⊆ R3 be a path-connected

region with piecewise smooth boundary ∂R and let f be a continuous function defined on ∂R.
Further let β be a real constant. Suppose that φ1 and φ2 are such that
∇2 φ1 = 0 = ∇2 φ2 in R;
∂φ1 ∂φ2
βφ1 + = f = βφ2 + on ∂R.
∂n ∂n
(i) If β > 0 then φ1 = φ2 in R.
(ii) If β = 0 then φ1 − φ2 is constant in R.
(iii) If β < 0 then the solution need not be unique, even up to a constant.
∂ψ
∇2 ψ = 0 in R, βψ + = βψ + ∇ψ · n = 0 on ∂R.
∂n
If we consider the function ψ 2 then, by the Divergence Theorem and as ∇2 = ∇ · ∇,
ZZ ZZZ
2
∇2 ψ 2 dV.

∇ ψ · dS =
∂R R
Looking at the LHS and noting βψ + ∇ψ · n = 0 on ∂R we have

ZZ ZZ ZZ ZZ
2
ψ 2 dS.

∇ ψ · dS = 2ψ (∇ψ · n) dS = 2ψ (−βψ) dS = −2β
∂R ∂R ∂R ∂R
On the RHS we have as before
∇2 ψ 2 = 2ψ∇2 ψ + 2∇ψ · ∇ψ = 2 |∇ψ|2 as ∇2 ψ = 0 in R.

Hence ZZ ZZZ
−2β 2
ψ dS = 2 |∇ψ|2 dV.
∂R R

If β > 0 then the LHS is non-positive and the RHS is non-negative – consequently both are
zero and ∇ψ = 0 in R and ψ = 0 on ∂R. Hence ψ = 0 throughout R.
If β = 0 then the Robin Boundary Problem is just the Neumann Problem, with which we have
already dealt.
The following example shows that solutions need not be unique if β < 0.
Example 116 Let R be the interior of the cylinder x2 + y 2 = 1. Find two solutions to the
Robin Boundary Problem
∂ψ
∇2 ψ = 0 in R, ψ= on ∂R.
∂n
Solution. Laplace’s equation in cylindrical polars reads
1 ∂ 2ψ ∂ 2ψ

2 1 ∂ ∂ψ
∇ψ= r + 2 2 + 2 = 0.
r ∂r ∂r r ∂θ ∂z
We know for ψ = rn sin nθ, where n is a non-negative integer, that ∇2 ψ = 0 and on the
boundary r = 1 we have
∂ψ ∂ψ
= = nrn−1 sin nθ = n sin nθ; ψ = sin nθ.
∂n ∂r
Hence when n = 1, we see r sin θ satisfies the given boundary problem. Other solutions are 0,
r cos θ and r sin (θ − α) for arbitrary α so we see this solution is far from unique.
7.2.2 Derivation of balance equations in 3D: the heat equation.
Finally, as a last application, we use the Divergence Theorem to rederive the heat equation,
but now in 3D. We need to following lemma first.
Lemma 117 If φ : R3 → R is a continuous scalar field such that

ZZZ
φ dV = 0
R
for all bounded subsets R, then φ ≡ 0.

Proof.
Suppose, for a contradiction that φ (p) 6= 0 for some point p. Without any loss of generality
let’s suppose φ (p) > 0. Let ε = 21 φ (p) > 0. Then, by the continuity of φ at p, there exists
δ > 0 such that
|φ (x) − φ (p)| < ε when |x − p| < δ.
In particular, φ (x) > 12 φ (p) on all of B (p, δ) = {x ∈ R3 : |x − p| < δ} and so

ZZZ
1
φ dV > φ (p) × Vol (B (p, δ)) > 0
2
B(p,δ)
which is a contradiction.
Theorem 118 Let T (x, t) denote the temperature at position x and at time t in an isotropic
medium R with thermal conductivity k, density ρ, specific heat c with heat flow determined by
Fourier’s Law which states that
q = −k∇T
where q is the heat flux. Then T satisfies the heat equation
∂T k 2
= ∇ T.
∂t ρc
Proof. Let S ⊆ R be an arbitrary subset of R. The total heat energy in S equals

ZZZ
ρcT dV.
S
The amount of heat flowing out of S is given by the flux integral

ZZ
q · dS.
∂S
So if no heat is generated within S then the only heat loss or gain is via the boundary ∂S.
Hence we have

ZZZ ZZ
d
ρcT dV = − q · dS
dt
S ∂S
ZZ
= k∇T · dS [by Fourier’s Law]
Z∂S
ZZ
= ∇ · (k∇T ) dV [by the Divergence Theorem]
Z ZSZ
= k ∇2 T dV.
S
So
ZZZ ZZZ
∂T
ρc dV = k ∇2 T dV,
∂t
S S
ZZZ
∂T 2
=⇒ ρc − k ∇ T dV = 0.
∂t
S
As this is true for an arbitrary subset S of R then it follows, from Lemma 117, that
∂T
ρc − k ∇2 T = 0 throughout R.
∂t
7.2.3 Greens’ Theorem in The Plane
Below, recall that a closed set contains its boundary and a simple closed curve is a curve that
has its endpoints meet but otherwise does not cross itself.
Theorem 119 (Green’s Theorem in The Plane (Divergence Theorem Form)) Let D
be a closed bounded region in the (x, y) plane, whose boundary C is a piecewise smooth simple
closed curve, and that p(x, y), q(x, y) have continuous first-order derivatives in D. Then
ZZ Z
∂p ∂q
+ dxdy = (p, q) · nds, (7.2)
D ∂x ∂y C
where n is the outward pointing unit normal to C in the (x, y) plane.
Proof. This is a simple corollary of the divergence theorem and is effectively the divergence
theorem in a plane. Let the vector field F be given by
F = (p(x, y), q(x, y), 0)

and define the three dimensional region R to be
R = (x, y, z) ∈ R3 | (x, y) ∈ D,

z ∈ [0, 1] ,
with boundary ∂R. Noting ∇ · F has no dependence on z, we have
ZZ ZZ Z 1 ZZ Z
∂p ∂q
+ dxdy = ∇ · Fdxdy dz = ∇ · FdV = F · ndS.
D ∂x ∂y D 0 R ∂R
The contribution to the surface integral from the surfaces at z = 0, 1 are zero as the integrand
is zero. For the surface
(x, y, z) ∈ R3 | (x, y) ∈ D, z ∈ [0, 1]

the outward pointing unit normal to R coincides with the outward pointing unit normal to C;
also we can write the surface element dS as
dS = dsdz,
where ds is the arclength element of the curve C since the k direction and the plane of the
region D are perpendicular. Hence
Z Z 1 Z Z 1 Z Z
F · ndS = F · nds dz = (p, q) · nds dz = (p, q) · nds,
∂R 0 C 0 C C
with the final equality from the fact the integrand (p, q) · n has no dependence on z.
Remark 120 By parameterising the curve C in the form (x(s), y(s)), where s is the arclength
of C, increasing on moving anticlockwise around C when viewed from above, one has the unit
tangent, t, and the outward unit normal n are given by

dx dy dy dx
t= , , n= ,− . (7.3)
ds ds ds ds
Important Below, it is important to note the direction around C is anticlockwise when

viewed from above; this is a positive orientation of C, which we will consider in generality
below.
Using the expression for n, we have
Z Z Z
dy dx def
(p, q) · nds = (p, q) · ,− ds = pdy − qdx (7.4)
C C ds ds C

and Greens’ theorem is often written in the equivalent form
ZZ Z
∂p ∂q
+ dxdy = pdy − qdx,
D ∂x ∂y C
where C is positively oriented, i.e. the integration around C is anticlockwise.
Another equivalent form of Green’s theorem in the plane is a precursor for Stokes’ Theorem,
which is our next subject of study.
Theorem 121 (Green’s Theorem in The Plane (Stokes’ Theorem Form)) An equiva-
lent form of Green’s theorem is as follows. Let D be a closed bounded region in the (x, y) plane,
whose boundary C is a piecewise smooth simple closed curve, and that p(x, y), q(x, y) have
continuous first-order derivatives in D. Then
ZZ Z Z
∂q ∂p
− dxdy = (p, q) · dr = pdx + qdy, (7.5)
D ∂x ∂y C C
where C is positively oriented, i.e. the integration around C is anticlockwise.
Proof. Let (p, q) → (q, −p) in Green’s theorem in the plane (divergence form), equation 7.2,
and use equations 7.4. We then have
ZZ Z Z
∂q ∂p dy dx dx dy
− dxdy = (q, −p) · ,− ds = (p, q) · , ds
D ∂x ∂y ds ds ds ds
ZC Z C
= (p, q) · dr = pdx + qdy,

C C
with the direction of integration around C inherited from Green’s theorem in the plane (diver-
gence form), via the relationship (7.3).
7.3 Stokes’ Theorem
Let F = (p(x, y), q(x, y), 0). Then

i j k
∂ ∂ ∂ ∂p ∂q
curl F = ∂x

∂y ∂z = − k,
p (x, y) q (x, y) ∂x ∂y
0
and so
ZZ ZZ ZZ
∂p ∂q
− dA = curl F · (k dA) = curl F · dS.
∂x ∂y
D D D
STOKES’ THEOREM 83
Thus Greens theorem in the plane can be rephrased as
Z ZZ
F · dr = curl F · dS, (7.6)
C D
Here
dS = kdS = ndS, (7.7)
where now, n is the normal to the planar region D in R3 [rather than the in-plane normal to
the boundary C as used in the previous section deducing Green’s theorem in the plane].
The identity 7.6 is purely a rewriting of Green’s Theorem, but it is part of a larger identity
known as Stokes’ Theorem which applies more generally to surfaces in R3 with boundary.
However, before we can properly understand the identity in its fullest generality, we need to
appreciate how to consistently orient a curve C which bounds a surface Σ.
Definition 122 At each point of a smooth surface there are two unit normal vectors ±n. Given
a smooth surface Σ, then by an orientation of Σ we shall mean a continuous choice of unit
normal n for the whole surface.
For the bounding curve, which we shall denote as ∂Σ, there are two possible orientations. Given
a certain orientation of ∂Σ and an oriented tangent vector t along ∂Σ then the vector t ∧ n lies
in the tangent plane of Σ at the boundary point and is normal to the bounding curve ∂Σ. So
t ∧ n points either towards the surface Σ or away from Σ.
We shall always choose to orient a surface Σ and its boundary ∂Σ in such a way that t ∧ n
points away from Σ.
Remark 123 Note that this is consistent in the plane with taking positively oriented curves and
normal k, so that equations (7.6,7.7) are equivalent to Green’s theorem in the plane, equation
(7.5).
In practice it is fairly clear how to consistently orient a surface and its boundary. If one imagines
an observer moving along the boundary ∂Σ (in the direction of the orientation) and upright
(in the sense of the surface normal n) then the surface Σ will be to his/her left.

Example 124 Given the hemisphere Σ = {(x, y, z) : x2 + y 2 + z 2 = 1, z > 0} with boundary
∂Σ = {(x, y, 0) : x2 + y 2 = 1} then we can either orient Σ as
n (x, y, z) = (x, y, z) or − n (x, y, z) = (−x, −y, −z) .
We can parameterise ∂Σ as (cos θ, sin θ, 0). So one choice of unit tangent is
t = (− sin θ, cos θ, 0) .
At (cos θ, sin θ, 0) we have
t ∧ n = (− sin θ, cos θ, 0) ∧ (cos θ, sin θ, 0) = (0, 0, −1)
which points down, away from the hemisphere. The two consistent ways to orient the surface
and boundary are
(n on Σ and t on ∂Σ) or (−n on Σ and − t on ∂Σ) .
Example 125 Not all surfaces have an orientation, such surfaces being called non-orientable.
Examples of non-orientable surfaces are the Möbius strip and the Klein bottle. We shall
only be interested in orientable surfaces though.
Theorem 126 (Stokes’ Theorem c. 1850. The first known appearance of this result though
is in fact in a letter from Kelvin to Stokes.) Let Σ be a smooth oriented surface in R3 whose
boundary is the curve ∂Σ. Let F be a smooth vector field defined on Σ ∪ ∂Σ. Then
Z ZZ
F · dr = curl F · dS. (7.8)
∂Σ
Σ
Remark 127 The following proof applies only to patches of surface – we shall see in later in
Example (134) that Stokes’ Theorem applies generally to more complicated surfaces and their
boundaries. In any event the proof of Stokes’ Theorem is not examinable.
Proof. Suppose that Σ is parameterised as r (u, v) where (u, v) ranges over some S ⊆ R2 , with
∂Σ being the image of ∂S under r. We shall first prove Stokes Theorem for the vector field
F = (F, 0, 0), noting that similar proofs apply for fields (0, F, 0) and (0, 0, F ) and then the more
general (7.8) follows by linearity.
So, if F = (F, 0, 0), then (7.8) reads as
Z ZZ
F dx = (0, Fz , −Fy ) · dS. (7.9)
∂Σ
Σ
Now using the parametrisation r (u, v) we have

ZZ ZZ
(0, Fz , −Fy ) · dS = (0, Fz , −Fy ) · (ru ∧ rv ) du dv
Σ S

ZZ i j k

= (0, Fz , −Fy ) · xu yu zu du dv
xv yv zv
Z SZ
= [Fz (zu xv − xu zv ) − Fy (xu yv − yu xv )] du dv.
S
On the other hand if we apply Green’s Theorem to the LHS of (7.9) we have
Z Z
F dx = F (xu du + xv dv)
∂Σ Z∂S
Z
= [(F xv )u − (F xu )v ] du dv
Z SZ
= [Fu xv + F xvu − Fv xu − F xuv ] du dv
Z SZ
= [Fu xv − Fv xu ] du dv
Z SZ
= [(Fx xu + Fy yu + Fz zu ) xv − (Fx xv + Fy yv + Fz zv ) xu ] du dv
Z SZ
= [Fz (zu xv − xu zv ) − Fy (xu yv − yu xv )] du dv.
S
The result follows.
Remark 128 Stokes’ Theorem and the Divergence Theorem are in fact statements of a more
general theorem, also known as Stokes’ Theorem, which applies in all dimensions. It more
generally reads: Z Z
dω = ω
M ∂M

where M is a (compact) oriented n-dimensional manifold (i.e. the n-dimensional equivalent of
a surface) with boundary ∂M and ω is a differential form of degree n − 1. A proper appreciation
of this result is approximately at fourth year undergraduate level. Note, though, that when n = 1
Stokes’ Theorem reads Z b
f 0 (x) dx = f (b) − f (a) ,
a
when M = [a, b] and ∂M = {a, b} oriented to sum positively the contribution at b and negatively
at a.
Remark 129 For an intuitive appreciation of why Stokes’ Theorem is true, recall that curl F,
when physically interpreted, is a measure of the local spin of the field F. So one might envisage
summing up all the curl over the surface Σ with most contributions to this total spin cancelling
out, in a manner akin to intermeshing cogs. The only contributions that wouldn’t be cancelled
out are those at the boundary, these remaining contributions making up the integral around the
boundary.
Note that Stokes’ Theorem is a scalar identity, that is both sides of the identity are scalar
quantities. There is also a vector identity related to Stokes’ Theorem. But firstly we note:
Lemma 130 If
c · v = c · w for all c ∈ R3
then v = w.
Proof. We have c·(v − w) = 0 for all c and thus for all basis vectors. Thus all the components
of v, w are equal, hence so are the vectors.
Corollary 131 Let Σ be a smooth oriented surface in R3 whose boundary is the curve ∂Σ. Let
ψ be a smooth scalar field defined on Σ ∪ ∂Σ. Then
ZZ Z
∇ψ ∧ dS = − ψ dr.
∂Σ
Σ
Proof. Let c be an arbitrary constant vector and set F = ψc. In this case, and using the curl
product rule, Stokes Theorem reads
ZZ ZZ Z
curl (ψc) · dS = (ψ curl c + ∇ψ ∧ c) · dS = ψc · dr.
∂Σ
Σ Σ
As c is constant then curl c = 0 and we have

 
ZZ ZZ Z
c· −
 ∇ψ ∧ dS =
 ∇ψ ∧ c · dS = c · ψ dr .
∂Σ
Σ Σ
As c is arbitrary then by Lemma 130 we have

ZZ Z
− ∇ψ ∧ dS = ψ dr
∂Σ
Σ
and the result follows.
Example 132 Verify Stokes’ Theorem when F = (y, z, x) on the hemisphere
Σ = (x, y, z) : x2 + y 2 + z 2 = a2 , z > 0 .

Solution. We can parameterise the sphere as
r (θ, φ) = (a sin θ cos φ, a sin θ sin φ, a cos θ) 0 6 θ 6 π, 0 6 φ 6 2π.
Then

i j k
∂r ∂r
= a2 sin θ (sin θ cos φ, sin θ sin φ, cos θ) .
∧ = a cos θ cos φ a cos θ sin φ −a sin θ
∂θ ∂φ
−a sin θ sin φ a sin θ cos φ 0
Now curl F = (−1, −1, −1) and hence

ZZ Z π/2 Z 2π
curl F · dS = (−1, −1, −1) · a2 sin θ (sin θ cos φ, sin θ sin φ, cos θ) dφ dθ
θ=0 φ=0
Σ
Z π/2 Z 2π
2
= −a sin θ (sin θ cos φ + sin θ sin φ + cos θ) dφ dθ
θ=0 φ=0
Z π/2
2
= −2πa sin θ cos θ dθ
θ=0
π/2
sin2 θ

2
= −2πa
2 0
= −πa2 .

On the other hand, once we parameterise and orient ∂Σ as r (θ) = (a cos θ, a sin θ, 0), we have
Z Z 2π
F · dr = (a sin θ, 0, a cos θ) · (−a sin θ, a cos θ, 0) dθ
∂Σ 0
Z 2π
2
= −a sin2 θ dθ
0
= −πa2 .
Example 133 Verify Stokes’ Theorem when F = (2y, 3x, x2 + y 2 + z 2 ) and Σ is the lower half
of the ellipsoid x2 /4 + y 2 /9 + z 2 /27 = 1.
Solution. Here ∂Σ is the ellipse x2 /4+y 2 /9 = 1 in the xy-plane. If ∂Σ is oriented anti-clockwise

then we need to take the upward pointing normal on Σ.
We may parameterise ∂Σ by r (t) = (2 cos t, 3 sin t, 0) and so
Z Z 2π
6 sin t, 6 cos t, 4 cos2 t + 9 sin2 t · (−2 sin t, 3 cos t, 0) dt

F · dr =
∂Σ
Z0 2π
−12 sin2 t + 18 cos2 t dt

=
Z0 2π
= ((−6 − 6 cos 2t) + (9 + 9 cos 2t)) dt
0
= 6π.
On the other hand we can parameterise the lower half of the ellipsoid as
√
r (θ, φ) = 2 sin θ cos φ, 3 sin θ sin φ, 3 3 cos θ .
So

i j √ k

rθ ∧ rφ = 2 cos θ cos φ 3 cos θ sin φ −3 3 sin θ
−2 sin θ sin φ 3 sin θ cos φ 0
√ √
2 2
= 9 3 sin θ cos φ, 6 3 sin θ sin φ, 6 sin θ cos θ ,
but this points downwards, so we will take its negative. Also

i
∂ ∂j k
∂

curl F = ∂x ∂y
= (2y, −2x, 1) .
∂z
2y 3x x2 + y 2 + z 2
Finally, then
ZZ
curl F · dS
Σ
Z π Z 2π √ √
= (6 sin θ sin φ, −4 sin θ cos φ, 1) · −9 3 sin2 θ cos φ, −6 3 sin2 θ sin φ, −6 sin θ cos θ dφ dθ
θ=π/2 φ=0
Zπ Z2π √ √
= −54 3 sin2 θ sin φ cos φ + 24 3 sin3 θ cos φ sin φ − 6 sin θ cos θ dφ dθ
θ=π/2 φ=0
Z π
= −2π 6 sin θ cos θ dθ
θ=π/2
π
3 cos 2θ
= 2π
2 π/2
= 3π (1 + 1) = 6π.
Example 134 Verify Stokes’ Theorem for F = (x2 + 3yz, −2xz, 7y) on Σ, where Σ is that part
of the cylindrical surface x2 + y 2 = a2 between the planes x + y + z = 0 and x + y + z = h,
where h > 0.
Solution. We may parameterise the cylinder as
r (θ, z) = (a cos θ, a sin θ, z)
and orient it with outward pointing normal dS = (cos θ, sin θ, 0) a dθ dz. So

i j k
∂ ∂ ∂

curl F = = (7 + 2x, 3y, −5z) .
∂x ∂y ∂z
x2 + 3yz −2xz 7y
ZZ Z 2π Z h−a cos θ−a sin θ

curl F · dS = (7 + 2a cos θ, 3a sin θ, −5z) · (cos θ, sin θ, 0) a dz dθ
θ=0 z=−a cos θ−a sin θ
Σ
Z 2π Z h−a cos θ−a sin θ
7 cos θ + 2a cos2 θ + 3a sin2 θ a dz dθ

=
θ=0 z=−a cos θ−a sin θ
Z 2π
7 cos θ + 2a cos2 θ + 3a sin2 θ ah dθ

=
θ=0
= (2aπ + 3aπ) ah
= 5πa2 h.

On the other side, we have a disconnected boundary ∂Σ made up of
C1 = {(a cos θ, a sin θ, −a cos θ − a sin θ) : 0 6 θ 6 2π} ;
C2 = {(a cos θ, a sin θ, h − a cos θ − a sin θ) : 0 6 θ 6 2π} .
As we used the outward pointing normal to orient Σ then we need to orient C1 in the direction
of increasing θ but orient C2 in the direction of decreasing θ.
If we write x (θ) = a cos θ, y (θ) = a sin θ, z (θ) = −a cos θ − a sin θ then we have
Z Z 2π
x (θ)2 + 3y (θ) z (θ) , −2x (θ) z (θ) , 7y (θ) · dr (θ)

F · dr =
C 0
Z 1 Z 2π
x (θ)2 + 3y (θ) [z (θ) + h] , −2x (θ) [z (θ) + h] , 7y (θ) · dr (θ) .

F · dr = −
C2 0
Hence
Z Z 2π
F · dr = (−3hy (θ) , 2hx (θ) , 0) · dr (θ)
∂Σ 0
Z 2π
= ah (−3a sin θ, 2a cos θ, 0) · (− sin θ, cos θ, sin θ − cos θ) dθ
0
Z 2π
2
3 sin2 θ + 2 cos2 θ dθ

= ah
0
= 5πa2 h.
Example 135 Let C be a closed curve bounding a smooth surface Σ. Show that
Z ZZ
r ∧ dr = 2 dS.
C
Σ
ZZ
Hence, or otherwise, evaluate dS where Σ is the surface
Σ
x2 y 2 2

2
Σ = (x, y, z) : z = 2 + 2 , x + y < 1 .
a b
[You may assume that ∇ ∧ (u ∧ v) = (v · ∇) u − (∇ · u) v + (∇ · v) u − (u · ∇) v.]
Solution.
If we set F = c ∧ r where c is an arbitrary constant vector, then we have
∇ ∧ (c ∧ r) = (r · ∇) c − (∇ · c) r + (∇ · r) c − (c · ∇) r
= (∇ · r) c − (c · ∇) r [as c is constant]
= 3c − c
= 2c
as

∂ ∂ ∂
(c · ∇) r = c1 + c2 + c3 (x, y, z) = (c1 , c2 , c3 ) = c.
∂x ∂y ∂z
So, by Stokes’ Theorem,
Z ZZ
(c ∧ r) · dr = 2c · dS
C
 Σ 
Z ZZ
=⇒ c· r ∧ dr = c·  2dS .
C
Σ
As c is arbitrary then by Lemma 130
Z ZZ
r ∧ dr = 2 dS. (7.10)
C
Σ
The bounding curve of the given surface Σ is
x2 y 2
z= + 2, x2 + y 2 = 1
a2 b
which can naturally be parameterised as
cos2 θ sin2 θ

r (θ) = cos θ, sin θ, + 2 , 0 6 θ 6 2π.
a2 b

By (7.10) we have (writing c = cos θ, s = sin θ)
1 2π c2 s 2
Z
−1
ZZ
1
dS = c, s, 2 + 2 ∧ −s, c, + 2 2cs dθ
2 0 a b a2 b
Σ
1 2π cs2 (−2cs2 − c3 ) c2 s (−s3 − 2c2 s)
Z
= + , 2 + , 1 dθ
2 0 b2 a2 a b2
1 2π cs2 (−cs2 − c) c2 s (−s − c2 s)
Z
= + , 2 + , 1 dθ
2 0 b2 a2 a b2
3 2π
(−s3 /3 − s) −c3 (c + c3 /3)

1 s
= + , , 2 + ,θ
2 3b2 a2 3a b2 0
1
= [(0, 0, 2π)]
2
= πk.
7.4 Stokes’ Theorem and Conservative Fields
Definition 136 A region R ⊆ R3 is said to be simply connected if every simple closed

curve C can be continuously deformed to a single point. Specifically, if r : [0, 1] → R is a simple
closed curve beginning and ending in p then there exists a map H : [0, 1] × [0, 1] → R such that
H (t, 1) = r (t) and H (t, 0) = p.
Example 137 R3 is simply connected, as is any plane or line. R2 −{0} is not simply connected,
nor is the cylinder x2 + y 2 = a2 , nor a torus. R3 − {0} is simply connected though.
Corollary 138 (Existence of a Potential) Let R be a simply connected region and let F
be a smooth vector field on R for which curl F = 0. Then F is conservative, i.e. there exists a
potential φ on R such that F = ∇φ.
Remark 139 Firstly note that if F is conservative then F = ∇φ and hence curl F = 0. Com-
bining this observation with corollary 138 gives, by use of Theorem 66, that with a restriction
to simply connected regions, curl F = 0 is equivalent to (i), (ii) and (iii):
(i) F is conservative – i.e. F = ∇φ for some scalar field φ : S → R.
(ii) Given any two points p, q ∈ S and curve γ in S, starting at p and ending at q, then
the integral Z
F (r) · dr
C
is independent of the choice of curve C.
STOKES’ THEOREM AND CONSERVATIVE FIELDS 93

(iii) For any simple closed curve C then
Z
F (r) · dr = 0.
C
Proof. Let p be a fixed point in R and define for any q ∈ R,

Z
φ (q) = F · dr
C1
where C1 is a curve in S from p to q. If C2 is a second such curve then there is a simple closed
curve C formed by C1 followed by C2 in reverse orientation. As R is simply connected then C
is the boundary of a surface Σ. By Stokes’ Theorem then
Z ZZ
F · dr = curl F · dS = 0.
C
Σ
This means that Z Z

φ (q) = F · dr = F · dr
C1 C2
and so φ is a well-defined function on R, dependent only on the variable q and not on the
choice of curve from p to q. Further, arguing as in Theorem 66, we can show that ∇φ = F.
But this is not generally equivalent – for example,

y −x
F= , (x, y) 6= (0, 0)
x2 + y 2 x2 + y 2
satisfies curl F = 0 but there is no φ on R2 − {(0, 0)} such that F = ∇φ. See the Exercise
Sheets.

MultiVarCalc MinorCorrectionTo p57 20 01 2014 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

MultiVarCalc MinorCorrectionTo p57 20 01 2014 PDF

Uploaded by

Copyright:

Available Formats

Prelims.

Hilary Term 2013

• Gerald B. Folland Advanced Calculus (Prentice Hall, 2002)

Definition 1 By a scalar field φ on R3 we shall mean a map φ : R3 → R.

Definition 2 By a vector field F on R3 we shall mean a map F : R3 → R3 .

1.1.1 Integrals in one dimension

An informal definition of the integral

• This is an informal definition of a double integral; a rigorous approach would, for

MULTIPLE INTEGRALS IN TWO DIMENSIONS: A BRIEF INTRODUCTION 5

• Linearity. Let a, b be constants.

• Domain Splittng. If R = R1 ∪ R2 , with R1 ∩ R2 the empty set, then

y=3 x=2 y=3 x=2 Z

or (first integrating with respect to y and keeping x constant)

MULTIPLE INTEGRALS IN TWO DIMENSIONS: A BRIEF INTRODUCTION 7

Z x=X Z y=Hx/X Z x=B Z y=H(x−B)/(X−B)

MULTIPLE INTEGRALS IN TWO DIMENSIONS: A BRIEF INTRODUCTION 9

More generally, we have to choose the limits to reflect the domain:

Example 8 Calculate the area of the disc x2 + y 2 6 a2 .

Analogously, in 3D as required below, we define the Jacobian

CHANGE OF VARIABLES. JACOBIANS. 11

Ψi = Ψ(f (xi + δx/2, yi + δy/2)) = ψi = ψ(xi + δx/2, yi + δy/2).

as shown in the diagram above.

Solution. We see, if we change to (u, v), parabolic coordinates given by

CHANGE OF VARIABLES. JACOBIANS. 13

Hence the area is given by

Solution. If we convert to polar co-ordinates this becomes

r3 cos3 θ + r3 sin3 θ = 3r2 sin θ cos θ

which we can rearrange to

CHANGE OF VARIABLES. JACOBIANS. 15

2.1 An Informal Definition of The Volume Integral

Consider a scalar field ψ(x, y, z), and a three-dimensional region R.

Example 17 A cone of height h occupies the region

We need to calculate the internal integrals first. We have

where the r appears as it is the Jacobian ∂ (x, y) /∂ (r, θ) . Thus

Natural co-ordinates for parameterising W are r, θ where

2.3 Changing Co-ordinates in Volume Integrals

Theorem 20 Let f : R → S be a bijection between subsets of R3 which is differentiable and

CHANGING CO-ORDINATES IN VOLUME INTEGRALS 21

The volume of a parallelepiped with sides a, b, c is |a · (b ∧ c) | and so, as a first approximation,

Taking limits gives

Example 21 (Cylindrical Polar Co-ordinates)

= r2 sin θ cos2 θ + sin2 θ

= r2 sin θ ≥ 0 since θ ∈ [0, π].

where M is the total mass of the body.

CHANGING CO-ORDINATES IN VOLUME INTEGRALS 23

Solution. If we take as an orthonormal basis

so we are left to calculate

taken over the region x2 + y 2 + z 2 6 1, where a, b, c are positive constants.

Solution. Consider instead this integral as

Example 27 Let a, b, c ∈ R. Determine

CHANGING CO-ORDINATES IN VOLUME INTEGRALS 25

is of unit length and so we can extend it to an orthonormal basis e1 , e2 , e3 with associated

a · r = (0, 0, |a|) · (X, Y, Z) = |a| Z.

The unit sphere x2 +y 2 +z 2 6 1 is still given as X 2 +Y 2 +Z 2 6 1 as e1 , e2 , e3 are an orthonormal

If we now change to spherical polar co-ordinates r, θ, φ associated with X, Y, Z we have

CHANGING CO-ORDINATES IN VOLUME INTEGRALS 27

3.1 Paramterised Surfaces

Cartesian representation. In Cartesian coordinates we represent a surface by z = f (x, y),

Parametric representation. In parametric coordinates we represent a surface by r =

Example 28 (Parameterisation of a spherical surface)

x = r sin θ cos φ, (3.1)

where θ ∈ [0, π], φ ∈ [0, 2π).

r = r(θ, φ) = (a sin θ cos φ, a sin θ sin φ, a cos θ). (3.4)

y=3 x=2 y=3 x=2 Z