Académique Documents
Professionnel Documents
Culture Documents
Multivariable Calculus
Eamonn Gaffney.
Using extensive material from lecture notes by Richard Earl
and also material from lecture notes by Ruth Baker.
• 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).
Website
https://www.maths.ox.ac.uk/courses/material
2
1. MULTIPLE INTEGRALS IN 2D
1.1 Preliminaries
• 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.
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.
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
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.
6 MULTIPLE INTEGRALS IN 2D
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)
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
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
8 MULTIPLE INTEGRALS IN 2D
at each point. We can then calculate the triangle’s area as
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,
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!
10 MULTIPLE INTEGRALS IN 2D
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
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
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
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)
12 MULTIPLE INTEGRALS IN 2D
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.
14 MULTIPLE INTEGRALS IN 2D
Example 15 Find the area bounded by the Folium of Descartes (see diagram) which has equa-
tion x3 + y 3 = 3xy.
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.
• 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
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
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)
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
20 VOLUME INTEGRATION
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.
(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).
∂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
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 )
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.
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
22 VOLUME INTEGRATION
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
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
By symmetry ZZZ
X dV = 0,
X 2 +Y 2 +Z 2 61
Z > 3/5
24 VOLUME INTEGRATION
Example 26 By a suitable rotation of co-ordinates, or otherwise, evaluate
ZZZ
(ax + by + cz)4 dV
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
Show that your answer is consistent with the fact the volume of the unit sphere is 4π/3.
where a = (a, b, c) .
The vector
a (a, b, c)
e3 = =√
|a| a2 + b2 + c2
26 VOLUME INTEGRATION
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
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
Note If z = f (x, y) we can use x and y as parameters: r(x, y) = (x, y, f (x, y)).
SURFACE INTEGRALS 29
To represent the point P = (x, y, z) in spherical polar coordinates (r, θ, φ) we have
• Sphere: x2 + y 2 + z 2 = a2 ;
30 SURFACE INTEGRALS
• Ellipsoid: 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 .
∂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:
∂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.
∂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
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
32 SURFACE INTEGRALS
Proposition 34 The surface area of r (U ) is independent of the choice of parametrisation.
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
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.
SURFACE INTEGRALS 33
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
34 SURFACE INTEGRALS
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 .
Example 37 Calculate the area of a sphere of radius a using spherical polar co-ordinates.
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
Hence
dS = a2 sin θ (− sin θ cos φ, sin θ sin φ, cos θ) dθ dφ = a2 |sin θ| dθ dφ.
SURFACE INTEGRALS 35
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
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|.
36 SURFACE INTEGRALS
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
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
SURFACE INTEGRALS 37
−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
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
38 SURFACE INTEGRALS
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.
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 .
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.
SURFACE INTEGRALS 39
40 SURFACE INTEGRALS
4. LINE INTEGRALS & CONSERVATIVE FIELDS
4.1 Curves
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.
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
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 0 (t) = F (γ1 (t)) · γ10 (t) , J 0 (t) = F (γ2 (t)) · γ20 (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 ) .
Z Z b
φ ds = φ (γ (t)) |γ 0 (t)| dt.
C a
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
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
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 ).
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.
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.
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
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
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
50 LINE INTEGRALS & CONSERVATIVE FIELDS
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)
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
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)
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.
. 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.
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
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
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
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.
Recall:
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.
with respect to a Cartesian coordinate system. Then the divergence of F, written div F, is
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
∂ ∂ ∂
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
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.
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:
Then
∂ ∂ ∂ ∂ ∂ ∂
e1 + e2 + e3 =i +j +k .
∂X ∂Y ∂Z ∂x ∂y ∂z
∂ ∂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.
∂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.
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
∂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
∂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.
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
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).
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.
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
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
(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
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
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
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
70 DIV, GRAD AND CURL
7. DIVERGENCE AND STOKES’ THEOREMS
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
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
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
Proof.
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
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)
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
∂f ∂f ∂f
x (tx, ty, tz) + y (tx, ty, tz) + z (tx, ty, tz) = ntn−1 f (x, y, z) .
∂x ∂y ∂z
∇ · (f r) = (∇f ) · r + f (∇ · r) = nf + 3f = (n + 3) f.
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 = ∇ · ∇,
Hence ZZZ
2 |∇ψ|2 dV = 0 =⇒ ∇ψ = 0 in R.
R
∇2 φ1 = 0 = ∇2 φ2 in R;
∂φ1 ∂φ2
= f= on ∂R.
∂n ∂n
Then φ1 − φ2 is constant in R.
Hence ZZZ
2 |∇ψ|2 dV = 0 =⇒ ∇ψ = 0 in R.
R
∇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
Hence ZZ ZZZ
−2β 2
ψ dS = 2 |∇ψ|2 dV.
∂R R
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
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.
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.
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.
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
∂T k 2
= ∇ T.
∂t ρc
So if no heat is generated within S then the only heat loss or gain is via the boundary ∂S.
Hence we have
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
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
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
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
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
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
with the direction of integration around C inherited from Green’s theorem in the plane (diver-
gence form), via the relationship (7.3).
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.
t = (− sin θ, cos θ, 0) .
which points down, away from the hemisphere. The two consistent ways to orient the surface
and boundary are
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.
STOKES’ THEOREM 85
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
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
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
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
STOKES’ THEOREM 87
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
Σ = (x, y, z) : x2 + y 2 + z 2 = a2 , z > 0 .
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
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.
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 ,
STOKES’ THEOREM 89
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.
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.]
STOKES’ THEOREM 91
Solution.
∇ ∧ (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
Z ZZ
(c ∧ r) · dr = 2c · dS
C
Σ
Z ZZ
=⇒ c· r ∧ dr = c· 2dS .
C
Σ
Z ZZ
r ∧ dr = 2 dS. (7.10)
C
Σ
x2 y 2
z= + 2, x2 + y 2 = 1
a2 b
cos2 θ sin2 θ
r (θ) = cos θ, sin θ, + 2 , 0 6 θ 6 2π.
a2 b
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.
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.
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
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.
satisfies curl F = 0 but there is no φ on R2 − {(0, 0)} such that F = ∇φ. See the Exercise
Sheets.