Topic 5

Vector Operators: Grad, Div and Curl

In the rst lecture of the second part of this course we move more to consider
properties of elds. We introduce three eld operators which reveal interesting
collective eld properties, viz.
the gradient of a scalar eld,
the divergence of a vector eld, and
the curl of a vector eld.
There are two points to get over about each:
The mechanics of taking the grad, div or curl, for which you will need to
brush up your multivariate calculus.
The underlying physical meaning that is, why they are worth bothering
about.
In Lecture 6 we will look at combining these vector operators.
5.1 The gradient of a scalar eld
Recall the discussion of temperature distribution throughout a room in the
overview, where we wondered how a scalar would vary as we moved off in
an arbitrary direction. Here we nd out how.
If U(x, y, z) is a scalar eld, ie a scalar function of position r = [x, y, z] in 3
dimensions, then its gradient at any point is dened in Cartesian co-ordinates
1
5/2
by
gradU =
U
x
+
U
y
+
U
z

k .
It is usual to dene the vector operator which is called del or nabla
=

x
+

y
+

k

z
.
Then
gradU U .
Note immediately that U is a vector eld!
Without thinking too carefully about it, we can see that the gradient of a scalar
eld tends to point in the direction of greatest change of the eld. Later we will
be more precise.
Worked examples of gradient evaluation
1. U = x
2
U =
_

x
+

y
+

z

k
_
x
2
= 2x .
2. U = r
2
r
2
= x
2
+ y
2
+ z
2
U =
_

x
+

y
+

z

k
_
(x
2
+ y
2
+ z
2
)
= 2x + 2y + 2z

k = 2 r .
3. U = c r, where c is constant.
U =
_

x
+

y
+

k

z
_
(c
1
x + c
2
y + c
3
z) = c
1
+c
2
+c
3

k = c .
5/3
4. U = f(r), where r =
_
(x
2
+ y
2
+ z
2
)
U is a function of r alone so df/dr exists. As U = f(x, y, z) also,
f
x
=
df
dr
r
x
f
y
=
df
dr
r
y
f
z
=
df
dr
r
z
.
U =
f
x
+
f
y
+
f
z

k =
df
dr
_
r
x
+
r
y
+
r
z

k
_
But r =
_
x
2
+ y
2
+ z
2
, so r/x = x/r and similarly for y, z.
U =
df
dr
_
x + y + z

k
r
_
=
df
dr
_
r
r
_
.
5.2 The signicance of grad
If our current position is r in some scalar eld U (Fig. 5.1), and we move an
innitesimal distance dr, we know that the change in U is
dU =
U
x
dx +
U
y
dy +
U
z
dz .
But we know that dr = (dx + dy +

kdz) and U = (U/x + U/y +

kU/z), so that the change in U is also given by the scalar product

dU = U dr .
Now divide both sides by ds
dU
ds
= U
dr
ds
.
But remember that |dr| = ds, so dr/ds is a unit vector in the direction of dr.
This result can be paraphrased as:
gradU has the property that the rate of change of U wrt distance
in a particular direction (

d) is the projection of gradU onto that

direction (or the component of gradU in that direction).
The quantity dU/ds is called a directional derivative, but note that in general it
has a different value for each direction, and so has no meaning until you specify
the direction.
5/4
gradU
r
U(r)
U(r + dr)
dr
Figure 5.1: The directional derivative
We could also say that
At any point P, gradU points in the direction of greatest change of
U at P, and has magnitude equal to the rate of change of U wrt
distance in that direction.
4
2
0
2
4
4
2
0
2
4
0
0.02
0.04
0.06
0.08
0.1
Another nice property emerges if we think of a surface of constant U that is
the locus (x, y, z) for
U(x, y, z) = constant .
If we move a tiny amount within that iso-U surface, there is no change in U, so
dU/ds = 0. So for any dr/ds in the surface
U
dr
ds
= 0 .
But dr/ds is a tangent to the surface, so this result shows that
gradU is everywhere NORMAL to a surface of constant U.
5/5
Surface of constant U
These are called Level Surfaces
Surface of constant U
gradU
5.3 The divergence of a vector eld
The divergence computes a scalar quantity from a vector eld by differentiation.
If a(x, y, z) is a vector function of position in 3 dimensions, that is a = a
1
+
a
2
+ a
3

k, then its divergence at any point is dened in Cartesian co-ordinates

by
diva =
a
1
x
+
a
2
y
+
a
3
z
We can write this in a simplied notation using a scalar product with the
vector differential operator:
diva =
_

x
+

y
+

k

z
_
a = a
Notice that the divergence of a vector eld is a scalar eld.
5/6
Examples of divergence evaluation
a diva
1) x 1
2) r(= x + y + z

k) 3
3) r/r
3
0
4) rc, for c constant (r c)/r
We work through example 3).
The x component of r/r
3
is x.(x
2
+ y
2
+ z
2
)
3/2
, and we need to nd /x of it.

x
x.(x
2
+ y
2
+ z
2
)
3/2
= 1.(x
2
+ y
2
+ z
2
)
3/2
+ x
3
2
(x
2
+ y
2
+ z
2
)
5/2
.2x
= r
3
_
1 3x
2
r
2
_
.
The terms in y and z are similar, so that
div(r/r
3
) = r
3
_
3 3(x
2
+ y
2
+ z
2
)r
2
_
= r
3
(3 3)
= 0
5.4 The signicance of div
Consider a typical vector eld, water ow, and denote it by a(r). This vector has
magnitude equal to the mass of water crossing a unit area perpendicular to the
direction of a per unit time.
Now take an innitesimal volume element dV and gure out the balance of the
ow of a in and out of dV .
To be specic, consider the volume element dV = dxdydz in Cartesian co-
ordinates, and think rst about the face of area dxdz perpendicular to the y
axis and facing outwards in the negative y direction. (That is, the one with
surface area dS = dxdz .)
The component of the vector a normal to this face is a = a
y
, and is pointing
inwards, and so the its contribution to the OUTWARD ux from this surface is
a dS = a
y
(y)dzdx ,
where a
y
(y) means that a
y
is a function of y. (By the way, ux here denotes
mass per unit time.)
5/7
dS =
dS =
j
z
x
y
dz
dx
dy
j dxdz +dxdz
Figure 5.2: Elemental volume for calculating divergence.
A similar contribution, but of opposite sign, will arise from the opposite face,
but we must remember that we have moved along y by an amount dy, so that
this OUTWARD amount is
a
y
(y + dy)dzdx =
_
a
y
+
a
y
y
dy
_
dxdz
The total outward amount from these two faces is
a
y
y
dydxdz =
a
y
y
dV
Summing the other faces gives a total outward ux of
_
a
x
x
+
a
y
y
+
a
z
z
_
dV = a dV
So we see that
The divergence of a vector eld represents the ux generation per unit
volume at each point of the eld. (Divergence because it is an efux not
an inux.)
Interestingly we also saw that the total efux from the innitesimal volume was
equal to the ux integrated over the surface of the volume.
5/8
(NB: The above does not constitute a rigorous proof of the assertion because we
have not proved that the quantity calculated is independent of the co-ordinate
system used, but it will sufce for our purposes.)
5.5 The Laplacian: div(gradU) of a scalar eld
Recall that gradU of any scalar eld U is a vector eld. Recall also that we
can compute the divergence of any vector eld. So we can certainly compute
div(gradU), even if we dont know what it means yet.
Here is where the operator starts to be really handy.
(U) =
_

x
+

y
+

k

z
_

__

x
+

y
+

k

z
_
U
_
=
__

x
+

y
+

k

z
_

x
+

y
+

k

z
__
U
=
_

2
x
2
+

2
y
2
+

2
z
2
_
U
=
_

2
U
x
2
+

2
U
y
2
+

2
U
z
2
_
This last expression occurs frequently in engineering science (you will meet it
next in solving Laplaces Equation in partial differential equations). For this
reason, the operator
2
is called the Laplacian

2
U =
_

2
x
2
+

2
y
2
+

2
z
2
_
U
Laplaces equation itself is

2
U = 0
5/9
Examples of
2
U evaluation
U
2
U
1) r
2
(= x
2
+ y
2
+ z
2
) 6
2) xy
2
z
3
2xz
3
+ 6xy
2
z
3) 1/r 0
Lets prove example (3) (which is particularly signicant can you guess why?).
1/r = (x
2
+ y
2
+ z
2
)
1/2

x
(x
2
+ y
2
+ z
2
)
1/2
=

x
x.(x
2
+ y
2
+ z
2
)
3/2
= (x
2
+ y
2
+ z
2
)
3/2
+ 3x.x.(x
2
+ y
2
+ z
2
)
5/2
= (1/r
3
)(1 + 3x
2
/r
2
)
Adding up similar terms for y and z

2
1
r
=
1
r
3
_
3 + 3
(x
2
+ y
2
+ x
2
)
r
2
_
= 0
5.6 The curl of a vector eld
So far we have seen the operator applied to a scalar eld U; and dotted
with a vector eld a.
We are now overwhelmed by an irrestible temptation to
cross it with a vector eld a
This gives the curl of a vector eld
a curl(a)
We can follow the pseudo-determinant recipe for vector products, so that
a =

z
a
x
a
y
a
z

(remember it this way)

=
_
a
z
y

a
y
z
_
+
_
a
x
z

a
z
y
_
+
_
a
y
x

a
x
y
_

k
5/10
Examples of curl evaluation
a a
1) y + x 2

k
2) x
2
y
2

k 2x
2
y 2xy
2

5.7 The signcance of curl
Perhaps the rst example gives a clue. The eld a = y + x is sketched in
Figure 5.3(a). (It is the eld you would calculate as the velocity eld of an
object rotating with = [0, 0, 1].) This eld has a curl of 2

k, which is in the r-h

screw sense out of the page. You can also see that a eld like this must give a
nite value to the line integral around the complete loop
_
C
a dr.
y
x
y
x
x
x
y+dy
y
(y+dy) a
(
x
+
d
x
)
a
x+dx
dy
(
x
)
a
y
dx
a (y)
y
x
y
(a) (b)
Figure 5.3: (a) A rough sketch of the vector eld y +x . (b) An element in which to calculate
curl.
In fact curl is closely related to the line integral around a loop.
The circulation of a vector a round any closed curve C is dened to be
_
C
a dr
and the curl of the vector eld a represents the vorticity, or circulation
per unit area, of the eld.
5/11
Our proof uses the small rectangular element dx by dy shown in Figure 5.3(b).
Consider the circulation round the perimeter of a rectangular element.
The elds in the x direction at the bottom and top are
a
x
(y) and a
x
(y + dy) = a
x
(y) +
a
x
y
dy,
where a
x
(y) denotes a
x
is a function of y, and the elds in the y direction at the
left and right are
a
y
(x) and a
y
(x + dx) = a
y
(x) +
a
y
x
dx
Starting at the bottom and working round in the anticlockwise sense, the four
contributions to the circulation dC are therefore as follows, where the minus
signs take account of the path being opposed to the eld:
dC = +[a
x
(y) dx] + [a
y
(x + dx) dy] [a
x
(y + dy) dx] [a
y
(x) dy]
= +[a
x
(y) dx] +
__
a
y
(x) +
a
y
x
dx
_
dy
_

__
a
x
(y) +
a
x
y
dy
_
dx
_
[a
y
(x) dy]
=
_
a
y
x

a
x
y
_
dx dy
= (a) dS
where dS = dxdy

k.
NB: Again, this is not a completely rigorous proof as we have not shown that
the result is independent of the co-ordinate system used.
5.8 Some denitions involving div, curl and grad
A vector eld with zero divergence is said to be solenoidal.
A vector eld with zero curl is said to be irrotational.
A scalar eld with zero gradient is said to be, er, constant.
Revised Oct 2005