1.1 Basis vectors
The usual basis vectors are denoted by i, j, k and are as the following picture describes.
6
¡
¡ª

k
j
i
The vectors, i, j, k, are ﬁxed. If v is a vector, there are unique scalars called components such that v =v
1
i+v
2
j+v
3
k
. This is what we mean when we say i, j, k is a basis.
Now suppose e
1
, e
2
, e
3
are three vectors which satisfy
e
1
×e
2
· e
3
6= 0.
Recall this means the volume of the box spanned by the three vectors is not zero.
£
£
££°
»
»:
@
@R
@
@
»
»
@
@»
»
£
£
££
»
»
£
£
££ e
1
e
3
e
2
Suppose e
1
, e
2
, e
3
are as just described. Does it follow that they form a basis? We show this is the case. Thus
there are unique scalars, v
1
, v
2
, and v
3
such that
v =v
i
e
i
.
This is the content of the following theorem.
Theorem 1 If e
1
, e
2
, e
3
are three vectors, then they form a basis if and only if
e
1
×e
2
· e
3
6= 0.
Proof: Suppose ﬁrst the above condition holds. Let i
1
≡ i, i
2
≡ j, i
3
≡ k and suppose v =u
j
i
j
. Therefore,
u
1
, u
2
, u
3
are the components of the vector v with respect to the usual basis vectors. Also let
e
i
= a
j
i
i
j
thus writing each e
i
in terms of the vectors i
j
. Then from the deﬁnition of the box product in terms of the usual
basis vectors, we see
0 6= e
1
×e
2
· e
3
= det
³
a
j
i
´
≡ det
a
1
1
a
2
1
a
3
1
a
1
2
a
2
2
a
3
2
a
1
3
a
2
3
a
3
3
= det
a
1
1
a
1
2
a
1
3
a
2
1
a
2
2
a
2
3
a
3
1
a
3
2
a
3
3
(1)
and we want to show there exists a unique solution to
v =u
j
i
j
= v
i
e
i
= a
j
i
v
i
i
j
. (2)
In other words, we want a unique solution to the system of equations
u
j
= a
j
i
v
i
, j = 1, 2, 3.
1
This is in the form
u =Av (3)
where A is matrix which, by 1, has non zero determinant.
If the box product is equal to zero, then the system which needs to be solved is of the form 3 where det (A) = 0.
Therefore, A has an eigenvector, v =(v
1
, v
2
, v
3
)
T
which corresponds to the eigen value λ = 0. Therefore, 2 shows 0
has more than one set of components with respect to the vectors e
1
, e
2
, e
3
and this set of vectors is therefore not a
basis.
This gives a simple geometric condition which determines whether a list of three vectors forms a basis in R
3
. One
simply takes the box product. If the box product is not equal to zero, then the vectors form a basis. If not, the list
of three vectors does not form a basis. This condition generalizes to R
n
as follows. If e
i
= a
j
i
i
j
, then {e
i
}
n
i=1
forms
a basis if and only if det
³
a
j
i
´
6= 0.
These vectors may or may not be orthonormal. In any case, it is convenient to deﬁne something called the dual
basis.
Deﬁnition 2 Let {e
i
}
n
i=1
form a basis for R
n
. Then
©
e
i
ª
n
i=1
is called the dual basis if
e
i
· e
j
= δ
i
j
≡
½
1 if i = j
0 if i 6= j
. (4)
Theorem 3 If {e
i
}
n
i=1
is a basis then
©
e
i
ª
n
i=1
is also a basis provided 4 holds.
Proof: Suppose
v =v
i
e
i
. (5)
Then taking the dot product of both sides of 5 with e
j
,yields
v
j
= v · e
j
. (6)
Thus there is at most one choice of scalars v
j
such that v =v
j
e
j
and it is given by 6.
¡
v −v · e
j
e
j
¢
· e
k
= 0
and so, since {e
i
}
n
i=1
is a basis,
¡
v −v · e
j
e
j
¢
· w = 0
for all vectors, w. It follows v −v · e
j
e
j
= 0 and this shows
©
e
i
ª
n
i=1
is a basis.
In the above argument we obtained formulas for the components of a vector v, v
i
, with respect to the dual basis,
found to be v
j
= v · e
j
. In the same way, we may ﬁnd the components of a vector with respect to the basis {e
i
}
n
i=1
.
Let v be any vector and let
v =v
j
e
j
. (7)
Then using 4 and taking the dot product of both sides of 7 with e
i
we see v
i
= e
i
· v.
Does there exist a dual basis and is it uniquely determined?
Theorem 4 If {e
i
}
n
i=1
is a basis for R
n
, then there exists a unique dual basis,
©
e
j
ª
n
j=1
satisfying
e
j
· e
i
= δ
j
i
.
2
Proof: First we show the dual basis is unique. Suppose
©
f
j
ª
n
j=1
is another set of vectors which satisﬁes f
j
·e
i
= δ
j
i
.
Then
f
j
= f
j
· e
i
e
i
= δ
j
i
e
i
= e
j
.
Note that from the deﬁnition, the dual basis to {i
j
}
n
j=1
is just i
j
= i
j
. Letting
e
i
= a
j
i
i
j
where the vectors, {i
j
}
n
j=1
are the standard basis vectors, it follows that since the e
i
form a basis that the matrix
whose ij th entry is a
j
i
is an invertible matrix. Letting
e
k
= b
k
r
i
r
,
we need to choose b
k
r
such that
b
k
r
i
r
· a
j
i
i
j
= b
k
r
a
j
i
δ
r
j
= b
k
j
a
j
i
= δ
k
i
.
But this is nothing more than the matrix equation for B which is of the form
AB = I
where A =
³
a
j
i
´
and has an inverse. There exists a unique solution to this equation given by B = A
−1
and this
proves the existence of the dual basis.
Summarizing what has been shown so far, we know that {e
i
}
n
i=1
is a basis for R
n
if and only if when e
i
= a
j
i
i
j
,
det
³
a
j
i
´
6= 0. (8)
If {e
i
}
n
i=1
is a basis, then there exists a unique dual basis,
©
e
j
ª
n
j=1
satisfying
e
j
· e
i
= δ
j
i
, (9)
and that if v is any vector,
v =v
j
e
j
, v =v
j
e
j
. (10)
The components of v which have the index on the top are called the contravariant components of the vector while
the components which have the index on the bottom are called the covariant components. In general v
i
6= v
j
! We
also have formulae for these components in terms of the dot product.
v
j
= v · e
j
, v
j
= v · e
j
. (11)
We deﬁne g
ij
≡ e
i
· e
j
and g
ij
≡ e
i
· e
j
. The next theorem describes the process of raising or lowering an index.
Theorem 5 The following hold.
g
ij
e
j
= e
i
, g
ij
e
j
= e
i
, (12)
g
ij
v
j
= v
i
, g
ij
v
j
= v
i
, (13)
g
ij
g
jk
= δ
i
k
, (14)
det (g
ij
) > 0, det
¡
g
ij
¢
> 0. (15)
3
Proof: First,
e
i
= e
i
· e
j
e
j
= g
ij
e
j
by 10 and 11. Similarly, by 10 and 11,
e
i
= e
i
· e
j
e
j
= g
ij
e
j
.
This veriﬁes 12. To verify 13,
v
i
= e
i
· v =g
ij
e
j
· v =g
ij
v
j
.
The proof of the remaining formula in 13 is similar.
To verify 14,
g
ij
g
jk
= e
i
· e
j
e
j
· e
k
=
¡¡
e
i
· e
j
¢
e
j
¢
· e
k
= e
i
· e
k
= δ
i
k
.
This shows the two determinants in 15 are non zero because the two matrices are inverses of each other. It only
remains to verify that one of these is greater than zero. Letting e
i
= a
j
i
i
j
= b
i
j
i
j
, we see that since i
j
= i
j
, a
j
i
= b
i
j
.
Therefore,
e
i
· e
j
= a
r
i
i
r
· b
j
k
i
k
= a
r
i
b
j
k
δ
k
r
= a
k
i
b
j
k
= a
k
i
a
k
j
.
It follows that for G the matrix whose ij th entry is e
i
· e
j
, G = AA
T
where the ik th entry of A is a
k
i
. Therefore,
det (G) = det (A) det
¡
A
T
¢
= det (A)
2
> 0. It follows from 14 that if H is the matrix whose ij th entry is g
ij
, then
GH = I and so H = G
−1
and
det (G) det
¡
G
−1
¢
= det
¡
g
ij
¢
det (G) = 1.
Therefore, det
¡
G
−1
¢
> 0 also. This proves the theorem.
Deﬁnition 6 The matrix (g
ij
) = G is called the metric tensor.
1.2 Exercises
1. Let e
1
= i +j, e
2
= i −j, e
3
= j +k. Find e
1
, e
2
, e
3
, (g
ij
) ,
¡
g
ij
¢
. If v = i+2j +k, ﬁnd v
i
and v
j
, the con
travariant and covariant components of the vector.
2. Let e
1
= 2i +j, e
2
= i −2j, e
3
= k. Find e
1
, e
2
, e
3
, (g
ij
) ,
¡
g
ij
¢
. If v =2i−2j +k, ﬁnd v
i
and v
j
, the contravari
ant and covariant components of the vector.
3. Suppose e
1
, e
2
, e
3
have the property that e
i
· e
j
= 0 whenever i 6= j. Show the same is true of the dual basis
and that in fact, e
i
is a multiple of e
i
.
4. Let e
1
, · · ·, e
3
be a basis for R
n
and let v =v
i
e
i
= v
i
e
i
, w =w
j
e
j
= w
j
e
j
be two vectors. Show
v · w =g
ij
v
i
w
j
= g
ij
v
i
w
j
.
5. Show if {e
i
}
3
i=1
is a basis in R
3
e
1
=
e
2
×e
3
e
2
×e
3
· e
1
, e
2
=
e
1
×e
2
e
1
×e
3
· e
2
, e
3
=
e
1
×e
2
e
1
×e
2
· e
3
.
4
6. Let {e
i
}
n
i=1
be a basis and deﬁne
e
∗
i
≡
e
i
e
i

, e
∗i
≡ e
i
e
i
 .
Show e
∗i
· e
∗
j
= δ
i
j
.
7. If v is a vector, v
∗
i
and v
∗i
, are deﬁned by
v ≡v
∗
i
e
∗i
≡ v
∗i
e
∗
i
.
These are called the physical components of v. Show
v
∗
i
=
v
i
e
i

, v
∗i
= v
i
e
i
 ( No summation on i ).
1.3 Curvilinear Coordinates
With the algebraic preparation of the last section, we are ready to consider curvilinear coordinates. Let D ⊆ R
n
be
an open set and let M :D →R
n
satisfy
M is C
2
, (16)
M is one to one. (17)
Letting x ∈D, we can write
M(x) = M
k
(x) i
k
where, as usual, i
k
are the standard basis vectors for R
n
, i
k
being the vector in R
n
which has a one in the kth
coordinate and a 0 in every other spot. For a ﬁxed x ∈ D, we can consider the curves,
t →M(x+ti
k
)
for t ∈ I, some open interval containing 0. Then for the point x,we let
e
k
≡
∂M
∂x
k
(x) ≡
d
dt
(M(x+ti
k
)) 
t=0
.
We will denote this vector as e
k
(x) to emphasize its dependence on x. The following picture illustrates the situation
in R
3
.
¡
¡
¡µ
¢
¢¸
H
H
H
HY
e
1
e
2
e
3
t →M(x
1
0
, x
2
0
, t)
t →M(t, x
2
0
, x
3
0
)
t →M(x
1
0
, t, x
3
0
)
We want {e
k
}
n
k=1
to be a basis. Thus we need
det
µ
∂M
i
∂x
k
¶
6= 0. (18)
5
Let
y
i
= M
i
(x) i = 1, · · ·, n (19)
so that the y
i
are the usual coordinates with respect to the usual basis vectors {i
k
}
n
k=1
of the point M(x) . Letting
x ≡
¡
x
1
, · · ·, x
n
¢
, it follows from the inverse function theorem of advanced calculus that M(D) is open, and that
18, 16, and 17 imply the equations 19 deﬁne each x
i
as a C
2
function of y ≡
¡
y
1
, · · ·, y
n
¢
T
. Thus, abusing notation
slightly, the equations 19 are equivalent to
x
i
= x
i
(y) , i = 1, · · ·, n
where x
i
is a C
2
function. Thus
∇x
k
(y) =
∂x
k
(y)
∂y
j
i
j
.
Then
∇x
k
(y) · e
j
=
∂x
k
∂y
s
i
s
·
∂y
r
∂x
j
i
r
=
∂x
k
∂y
s
∂y
s
∂x
j
= δ
k
j
by the chain rule. Therefore, the dual basis is given by
e
k
(x) = ∇x
k
(y) . (20)
Notice that it might be hard or even impossible to solve algebraically for x
i
in terms of the y
j
. Thus the straight
forward approach to ﬁnding e
k
by 20 might be impossible. Also, this approach leads to an expression in terms of the
y coordinates rather than the desired x coordinates. Therefore, it is expedient to use another method to obtain these
vectors. The vectors, e
k
(x) may always be found by using formula 12 and the result is in terms of the curvilinear
coordinates, x. We illustrate in the following example.
Example 7 D ≡ (0, ∞) × (0, π) × (0, 2π) and
y
1
y
2
y
3
=
x
1
sin
¡
x
2
¢
cos
¡
x
3
¢
x
1
sin
¡
x
2
¢
sin
¡
x
3
¢
x
1
cos
¡
x
2
¢
(We usually write this as
x
y
z
=
ρ sin(φ) cos (θ)
ρ sin(φ) sin(θ)
ρ cos (φ)
where (ρ, φ, θ) are the spherical coordinates. We are calling them x
1
, x
2
, and x
3
to preserve the notation just dis
cussed.) Thus
e
1
(x) = sin
¡
x
2
¢
cos
¡
x
3
¢
i
1
+sin
¡
x
2
¢
sin
¡
x
3
¢
i
2
+cos
¡
x
2
¢
i
3
,
e
2
(x) = x
1
cos
¡
x
2
¢
cos
¡
x
3
¢
i
1
+x
1
cos
¡
x
2
¢
sin
¡
x
3
¢
i
2
−x
1
sin
¡
x
2
¢
i
3
,
e
3
(x) = −x
1
sin
¡
x
2
¢
sin
¡
x
3
¢
i
1
+x
1
sin
¡
x
2
¢
cos
¡
x
3
¢
i
2
+0i
3
.
6
It follows the metric tensor is
G =
1 0 0
0
¡
x
1
¢
2
0
0 0
¡
x
1
¢
2
sin
2
¡
x
2
¢
= (g
ij
) = (e
i
· e
j
) . (21)
Therefore, by Theorem 5
G
−1
=
¡
g
ij
¢
=
¡
e
i
, e
j
¢
=
1 0 0
0
¡
x
1
¢
−2
0
0 0
¡
x
1
¢
−2
sin
−2
¡
x
2
¢
.
To obtain the dual basis, use Theorem 5 to write
e
1
= g
1j
e
j
= e
1
e
2
= g
2j
e
j
=
¡
x
1
¢
−2
e
2
e
3
= g
3j
e
j
=
¡
x
1
¢
−2
sin
−2
¡
x
2
¢
e
3
.
It is natural to ask if we can get a transformation M such that
∂M
∂x
1
= i = i
1
,
∂M
∂x
2
= j = i
2
,
∂M
∂x
3
= k = i
3
. (22)
The answer is that we can. Let
M
¡
x
1
, x
2
, x
3
¢
≡ x
1
i+x
2
j+x
3
k.
Then 22 holds for this transformation.
1.4 Exercises
1. Let
y
1
y
2
y
3
=
x
1
+ 2x
2
x
2
+ x
3
x
1
−2x
2
where the y
i
are the rectangular coordinates of the point. Find e
i
, e
i
, i = 1, 2, 3, and ﬁnd (g
ij
) (x) and
¡
g
ij
(x)
¢
.
2. Let y = y(x,t) where t signiﬁes time and x ∈U ⊆ R
m
for U an open set, while y ∈ R
n
and suppose x is a
function of t. Physically, this corresponds to an object moving over a surface in R
n
which may be changing
as a function of t. The point y = y (x(t) , t) is the point in R
n
corresponding to t. For example, consider the
pendulum
D
D
D
D
D D
• m
l
θ
in which n = 2, l is ﬁxed and y
1
= l sinθ, y
2
= l − l cos θ. Thus, in this simple example, m = 1. If l were
changing in a known way with respect to t, then this would be of the form y = y(x,t) . The kinetic energy is
7
deﬁned as
T ≡
1
2
m˙ y · ˙ y (∗)
where the dot on the top signiﬁes diﬀerentiation with respect to t. Show
∂T
∂ ˙ x
k
= m˙ y·
∂y
∂x
k
.
Hint:First show
˙ y =
∂y
∂x
j
˙ x
j
+
∂y
∂t
(∗∗)
and so
∂ ˙ y
∂ ˙ x
j
=
∂y
∂x
j
.
3. ↑ Show
d
dt
µ
∂T
∂ ˙ x
k
¶
= m¨ y·
∂y
∂x
k
+ m˙ y·
∂
2
y
∂x
k
∂x
r
˙ x
r
+ m˙ y ·
∂
2
y
∂t∂x
k
.
4. ↑ Show
∂T
∂x
k
= m˙ y·
µ
∂
2
y
∂x
r
∂x
k
˙ x
r
+
∂
2
y
∂t∂x
k
¶
.
Hint: Use ∗ and ∗ ∗ .
5. ↑ Now show from Newton’s second law ( mass times acceleration equals force ) that for F the force,
d
dt
µ
∂T
∂ ˙ x
k
¶
−
∂T
∂x
k
= m¨ y·
∂y
∂x
k
= F·
∂y
∂x
k
. (∗ ∗ ∗)
6. ↑ In the example of the simple pendulum above,
y =
µ
l sinθ
l −l cos θ
¶
= l sinθi+(l −l cos θ) j.
Use ∗ ∗ ∗ to ﬁnd a diﬀerential equation which describes the vibrations of the pendulum in terms of θ. First
write the kinetic energy and then consider the force acting on the mass which is
−mgj.
7. The above problem is fairly easy to do without the formalism developed. Now consider the case where
x =(ρ, θ, φ) , spherical coordinates, and write diﬀerential equations for ρ, θ, and φ to describe the motion
of an object in terms of these coordinates given a force, F.
8. Suppose the pendulum is not assumed to vibrate in a plane. Let it be suspended at the origin and consider
spherical coordinates. Find diﬀerential equations for θ and φ.
8
9. If there are many masses, m
α
, α = 1, · ··, R, the kinetic energy is the sum of the kinetic energies of the individual
masses. Thus,
T ≡
1
2
R
X
α=1
m
α
 ˙ y
α

2
.
Generalize the above problems to show that, assuming
y
α
= y
α
(x,t) ,
d
dt
µ
∂T
∂ ˙ x
k
¶
−
∂T
∂x
k
=
R
X
α=1
F
α
·
∂y
α
∂x
k
where F
α
is the force acting on m
α
.
10. Discuss the equivalence of these formulae with Newton’s second law, force equals mass times acceleration.
What is gained from the above so called Lagrangian formalism?
11. The double pendulum has two masses instead of only one.
D
D
D
D
D D
• m
1
l
1
θ
D
D
D
D
D D
• m
2
l
2
φ
Write diﬀerential equations for θ and φ to describe the motion of the double pendulum.
1.5 Transformation of coordinates.
The scalars
©
x
i
ª
are called cuvilinear coordinates. Note they can be used to identify a point in R
n
and x =
¡
x
1
, · · ·, x
n
¢
is a point in R
n
. The basis vectors associated with this particular set of curvilinear coordinates at a point identiﬁed
by x are denoted by e
i
(x) and the dual basis vectors at this point are denoted by e
j
(x). What if other curvilinear
coordinates are used? How do we write e
k
(x) in terms of the vectors, e
j
(z) where z is some other type of curvilinear
coordinates? We consider this topic next.
Consider the following picture in which U is an open set in R
n
, D, and
b
D are open sets in R
n
, and M, N are C
2
mappings which are one to one from D and
b
D respectively. We will suppose that a point in U is identiﬁed by the
curvilinear coordinates x in D and z in
b
D.
 ¾
U
D
b
D
M N
(x
1
, x
2
, x
3
) (z
1
, z
2
, z
3
)
Thus M(x) = N(z) . Now by the chain rule,
e
i
(z) ≡
∂N
∂z
i
=
∂M
∂x
j
∂x
j
∂z
i
=
∂x
j
∂z
i
e
j
(x) . (23)
9
We deﬁne the covariant and contravariant coordinates for the various curvilinear coordinates in the obvious way.
Thus,
v =v
i
(x) e
i
(x) =v
i
(x) e
i
(x) = v
j
(z) e
j
(z) = v
j
(z) e
j
(z) .
Then we have the following theorem about transforming the vectors and coordinates.
Theorem 8 The following transformation rules hold for pairs of curvilinear coordinates.
v
i
(z) =
∂x
j
∂z
i
v
j
(x) , v
i
(z) =
∂z
i
∂x
j
v
j
(x) , (24)
e
i
(z) =
∂x
j
∂z
i
e
j
(x) , e
i
(z) =
∂z
i
∂x
j
e
j
(x) , (25)
g
ij
(z) =
∂x
r
∂z
i
∂x
s
∂z
j
g
rs
(x) , g
ij
(z) =
∂z
i
∂x
r
∂z
j
∂x
s
g
rs
(x) . (26)
Proof: We already have shown the ﬁrst part of 25 in 23. Then, from 23,
e
i
(z) = e
i
(z) · e
j
(x) e
j
(x) = e
i
(z) ·
∂z
k
∂x
j
e
k
(z) e
j
(x)
= δ
i
k
∂z
k
∂x
j
e
j
(x) =
∂z
i
∂x
j
e
j
(x)
and this proves the second part of 25. Now to show 24,
v
i
(z) = v · e
i
(z) = v·
∂x
j
∂z
i
e
j
(x) =
∂x
j
∂z
i
v
j
(x)
and
v
i
(z) = v · e
i
(z) = v·
∂z
i
∂x
j
e
j
(x) =
∂z
i
∂x
j
v
j
(x) .
To verify 26,
g
ij
(z) = e
i
(z) · e
j
(z) = e
r
(x)
∂x
r
∂z
i
· e
s
(x)
∂x
s
∂z
j
= g
rs
(x)
∂x
r
∂z
i
∂x
s
∂z
j
.
This proves the theorem.
We will denote by y the curvilinear coordinates with the property that
e
k
(y) = i
k
= e
k
(y) .
1.6 Diﬀerentiation and Christoﬀel Symbols
Let F :U →R
n
be diﬀerentiable. We call F a vector ﬁeld and it is used to model force, velocity, acceleration, or any
other vector quantity which may change from point to point in U. Then
∂F(x)
∂x
j
is a vector and so there exist scalars, F
i
,j
(x) and F
i,j
(x) such that
∂F(x)
∂x
j
= F
i
,j
(x) e
i
(x) = F
i,j
(x) e
j
(x) . (27)
We will see how these scalars transform when the coordinates are changed.
10
Theorem 9 If x and z are curvilinear coordinates,
F
r
,s
(x) = F
i
,j
(z)
∂x
r
∂z
i
∂z
j
∂x
s
, F
r,s
(x)
∂x
r
∂z
i
∂x
s
∂z
j
= F
i,j
(z) . (28)
Proof:
F
r
,s
(x) e
r
(x) ≡
∂F(x)
∂x
s
=
∂F(z)
∂z
j
∂z
j
∂x
s
≡
F
i
,j
(z) e
i
(z)
∂z
j
∂x
s
= F
i
,j
(z)
∂x
r
∂z
i
∂z
j
∂x
s
e
r
(x)
which shows the ﬁrst formula of 27. To show the other formula,
F
i,j
(z) e
i
(z) ≡
∂F(z)
∂z
j
=
∂F(x)
∂x
s
∂x
s
∂z
j
≡
F
r,s
(x) e
r
(x)
∂x
s
∂z
j
= F
r,s
(x)
∂x
r
∂z
i
∂x
s
∂z
j
e
i
(z) ,
and this shows the second formula for transforming these scalars.
Now F(x) = F
i
(x) e
i
(x) and so by the product rule,
∂F
∂x
j
=
∂F
i
∂x
j
e
i
(x) + F
i
(x)
∂e
i
(x)
∂x
j
.
Now
∂ei(x)
∂x
j
is a vector and so there exist scalars,
©
k
ij
ª
such that
∂e
i
(x)
∂x
j
=
½
k
ij
¾
e
k
(x) .
Therefore,
∂F
∂x
j
=
∂F
k
∂x
j
e
k
(x) + F
i
(x)
½
k
ij
¾
e
k
(x)
which shows
F
k
,j
(x) =
∂F
k
∂x
j
+ F
i
(x)
½
k
ij
¾
.
This is sometimes called the covariant derivative.
These scalars are called the Christoﬀel symbols of the second kind. The next theorem is devoted to properties of
these Christoﬀel symbols. Before stating the theorem, we recall that the mapping, M which deﬁnes the curvilinear
coordinates is C
2
. The reason for this is that we want to be able to assert the mixed partial derivatives are equal.
Theorem 10 The Christoﬀel symbols of the second kind satisfy the following
∂e
i
(x)
∂x
j
=
½
k
ij
¾
e
k
(x) , (29)
∂e
i
(x)
∂x
j
= −
½
i
kj
¾
e
k
(x) , (30)
½
k
ij
¾
=
½
k
ji
¾
, (31)
½
m
ik
¾
=
g
jm
2
·
∂g
ij
∂x
k
+
∂g
kj
∂x
i
−
∂g
ik
∂x
j
¸
. (32)
11
Proof: Formula 29 is the deﬁnition of the Christoﬀel symbols. We verify 30 next. To do so, note
e
i
(x) · e
k
(x) = δ
i
k
.
Then from the product rule,
∂e
i
(x)
∂x
j
· e
k
(x) +e
i
(x) ·
∂e
k
(x)
∂x
j
= 0.
Now from the deﬁnition,
∂e
i
(x)
∂x
j
· e
k
(x) = −e
i
(x) ·
½
r
kj
¾
e
r
(x) = −
½
i
kj
¾
.
But also,
∂e
i
(x)
∂x
j
=
∂e
i
(x)
∂x
j
· e
k
(x) e
k
(x) = −
½
i
kj
¾
e
k
(x) .
This veriﬁes 30.
Letting
∂M(x)
∂x
j
= e
j
(x) , it follows from equality of mixed partial derivatives,
½
k
ij
¾
e
k
(x) =
∂e
i
∂x
j
≡
∂
2
M
∂x
j
∂x
i
=
∂
2
M
∂x
i
∂x
j
=
∂e
j
∂x
i
=
½
k
ji
¾
e
k
(x) ,
which shows 31. It remains to show 32.
∂g
ij
∂x
k
=
∂e
i
∂x
k
· e
j
+e
i
·
∂e
j
∂x
k
=
½
r
ik
¾
e
r
· e
j
+e
i
· e
r
½
r
jk
¾
.
Therefore,
∂g
ij
∂x
k
=
½
r
ik
¾
g
rj
+
½
r
jk
¾
g
ri
. (33)
Switching i and k while remembering 31 yields
∂g
kj
∂x
i
=
½
r
ik
¾
g
rj
+
½
r
ji
¾
g
rk
. (34)
Now switching j and k in 33, we obtain,
∂g
ik
∂x
j
=
½
r
ij
¾
g
rk
+
½
r
jk
¾
g
ri
. (35)
Adding 33 to 34 and subtracting 35 yields
∂g
ij
∂x
k
+
∂g
kj
∂x
i
−
∂g
ik
∂x
j
= 2
½
r
ik
¾
g
rj
.
Now multiplying both sides by g
jm
and using the fact shown earlier in Theorem 5 that
g
rj
g
jm
= δ
m
r
,
we obtain
2
½
m
ik
¾
= g
jm
µ
∂g
ij
∂x
k
+
∂g
kj
∂x
i
−
∂g
ik
∂x
j
¶
which proves 32.
This is a very interesting formula because it shows the Christoﬀel symbols are completely determined by the
metric tensor and its derivatives.
12
1.7 Gradients and divergence
In this section we express the gradient and the divergence of a vector ﬁeld in general curvilinear coordinates. As
before, y will denote the standard coordinates with respect to the usual basis vectors. Thus
N(y) ≡ y
k
i
k
, e
k
(y) = i
k
= e
k
(y) .
Let φ : U →R be a diﬀerentiable scalar function, sometimes called a “scalar ﬁeld” in this subject. We write φ(x)
to denote the value of φ at the point whose coordinates are x. In general, we follow this practice for any ﬁeld, vector
or scalar. Thus F(x) is the value of a vector ﬁeld at the point of U determined by the coordinates x. If we are using
the standard coordinates, we know what we mean by the gradient of φ. It is given by the following formula.
∇φ(y) =
∂φ(y)
∂y
k
e
k
(y) .
Therefore, using the chain rule, if the coordinates of the point of U are given as x,
∇φ(x) = ∇φ(y)
=
∂φ(x)
∂x
r
∂x
r
∂y
k
∂y
k
∂x
s
e
s
(x) =
∂φ(x)
∂x
r
δ
r
s
e
s
(x) =
∂φ(x)
∂x
r
e
r
(x) .
This shows the covariant components of ∇φ(x) are
(∇φ(x))
r
=
∂φ(x)
∂x
r
. (36)
To ﬁnd the contravariant components, we “raise the index” in the usual way. Thus
(∇φ(x))
r
= g
rk
(x) (∇φ(x))
k
= g
rk
(x)
∂φ(x)
∂x
k
. (37)
What about the divergence of a vector ﬁeld? The divergence of a vector ﬁeld, F deﬁned on U is a scalar ﬁeld,
div (F) which we know from calculus to be
∂F
k
∂y
k
(y) = F
k
,k
(y)
in terms of the usual coordinates y. The reason the above equation holds in this case is that e
k
(y) is a constant and
so the Christoﬀel symbols are zero. We want an expression for the divergence in an arbitrary coordinate system.
From Theorem 9,
F
i
,j
(y) = F
r
,s
(x)
∂x
s
∂y
j
∂y
i
∂x
r
=
µ
∂F
r
(x)
∂x
s
+ F
k
(x)
½
r
ks
¾
(x)
¶
∂x
s
∂y
j
∂y
i
∂x
r
.
Letting j = i yields
div (F) =
µ
∂F
r
(x)
∂x
s
+ F
k
(x)
½
r
ks
¾
(x)
¶
∂x
s
∂y
i
∂y
i
∂x
r
=
µ
∂F
r
(x)
∂x
s
+ F
k
(x)
½
r
ks
¾
(x)
¶
δ
s
r
=
µ
∂F
r
(x)
∂x
r
+ F
k
(x)
½
r
kr
¾
(x)
¶
. (38)
13
We will simplify
©
r
kr
ª
using the description of it in Theorem 10. Thus, from this theorem,
½
r
rk
¾
=
g
jr
2
·
∂g
rj
∂x
k
+
∂g
kj
∂x
r
−
∂g
rk
∂x
j
¸
Now consider
g
jr
2
times the last two terms in [·] . Relabeling the indices r and j in the second term implies
g
jr
2
∂g
kj
∂x
r
−
g
jr
2
∂g
rk
∂x
j
=
g
jr
2
∂g
kj
∂x
r
−
g
rj
2
∂g
jk
∂x
r
= 0.
Therefore,
½
r
rk
¾
=
g
jr
2
∂g
rj
∂x
k
. (39)
Now recall g ≡ det (g
ij
) = det (G) > 0 from Theorem 5. Also from the formula for the inverse of a matrix and this
theorem,
g
jr
= A
rj
(det G)
−1
= A
jr
(det G)
−1
where A
rj
is the rjth cofactor of the matrix (g
ij
) . Also recall that
g =
n
X
r=1
g
rj
A
rj
no sum on j.
Therefore, g is a function of the variables {g
rj
} and
∂g
∂g
rj
= A
rj
.
From 39,
½
r
rk
¾
=
g
jr
2
∂g
rj
∂x
k
=
1
2g
∂g
rj
∂x
k
A
jr
=
1
2g
∂g
∂g
rj
∂g
rj
∂x
k
=
1
2g
∂g
∂x
k
and so from 38,
div (F) =
∂F
k
(x)
∂x
k
+
+F
k
(x)
1
2g (x)
∂g (x)
∂x
k
=
1
p
g (x)
∂
∂x
i
³
F
i
(x)
p
g (x)
´
. (40)
This is our formula for the divergence of a vector ﬁeld in general curvilinear coordinates.
The Laplacian of a scalar ﬁeld is nothing more than the divergence of the gradient. In symbols,
∆φ ≡ ∇· ∇φ
From 40 and 37 it follows
∆φ(x) =
1
p
g (x)
∂
∂x
i
µ
g
ik
(x)
∂φ(x)
∂x
k
p
g (x)
¶
. (41)
We summarize the conclusions of this section in the following theorem.
14
Theorem 11 The following formulas hold for the gradient, divergence and Laplacian in general curvilinear coordi
nates.
(∇φ(x))
r
=
∂φ(x)
∂x
r
, (42)
(∇φ(x))
r
= g
rk
(x)
∂φ(x)
∂x
k
, (43)
div (F) =
1
p
g (x)
∂
∂x
i
³
F
i
(x)
p
g (x)
´
, (44)
∆φ(x) =
1
p
g (x)
∂
∂x
i
µ
g
ik
(x)
∂φ(x)
∂x
k
p
g (x)
¶
. (45)
1.8 Exercises
1. Let y
1
= x
1
+ 2x
2
, y
2
= x
2
+ 3x
3
, y
3
= x
1
+ x
3
. Let
F(x) = x
1
e
1
(x) + x
2
e
2
(x) +
¡
x
3
¢
2
e (x) .
Find div (F) (x) .
2. For the coordinates of the preceding problem, and φ a scalar ﬁeld, ﬁnd
(∇φ(x))
3
in terms of the partial derivatives of φ taken with respect to the variables x
i
.
3. Let y
1
= 7x
1
+ 2x
2
, y
2
= x
2
+ 3x
3
, y
3
= x
1
+ x
3
. Let φ be a scalar ﬁeld. Find ∇
2
φ(x) .
4. Derive ∇
2
u in cylindrical coordinates, r, θ, z, where u is a scalar ﬁeld on R
3
.
x = r cos θ, y = r sinθ, z = z.
5. ↑ Find all solutions to ∇
2
u = 0 which depend only on r where r ≡
p
x
2
+ y
2
.
6. Let u be a scalar ﬁeld on R
3
. Find all solutions to ∇
2
u = 0 which depend only on
ρ ≡
p
x
2
+ y
2
+ z
2
.
7. The temperature, u, in a solid satisﬁes ∇
2
u = 0 after a long time. Suppose in a long pipe of inner radius 9 and
outer radius 10 the exterior surface is held at 100
◦
while the inner surface is held at 200
◦
ﬁnd the temperature
in the solid part of the pipe.
8. Show
½
l
ij
¾
=
∂e
i
∂x
j
· e
l
.
Find the Christoﬀel symbols of the second kind for spherical coordinates in which x
1
= φ, x
2
= θ, and x
3
= ρ.
Do the same for cylindrical coordinates letting x
1
= r, x
2
= θ, x
3
= z.
15
9. Show velocity can be expressed as v =v
i
(x) e
i
(x) , where
v
i
(x) =
∂r
i
∂x
j
dx
j
dt
−r
p
(x)
½
p
ik
¾
dx
k
dt
and r
i
(x) are the covariant components of the displacement vector,
r =r
i
(x) e
i
(x) .
10. ↑ Using problem 8 and 9, show the covariant components of velocity in spherical coordinates are
v
1
= ρ
2
dφ
dt
, v
2
= ρ
2
sin
2
(φ)
dθ
dt
, v
3
=
dρ
dt
.
Hint: First observe that if r is the position vector from the origin, then r =ρe
3
so r
1
= 0 = r
2
, and r
3
= ρ.
Now use 9.
1.9 Curl and cross products
In this section we consider the curl and cross product in general curvilinear coordinates in R
3
. We will always assume
that for x a set of curvilinear coordinates,
det
µ
∂y
i
∂x
j
¶
> 0 (46)
Where the y
i
are the usual coordinates in which e
k
(y) = i
k
.
Theorem 12 Let 46 hold. Then
det
µ
∂y
i
∂x
j
¶
=
p
g (x) (47)
and
det
µ
∂x
i
∂y
j
¶
=
1
p
g (x)
. (48)
Proof:
e
i
(x) =
∂y
k
∂x
i
i
k
and so
g
ij
(x) =
∂y
k
∂x
i
i
k
·
∂y
l
∂x
j
i
l
=
∂y
k
∂x
i
∂y
k
∂x
j
.
Therefore, g = det (g
ij
(x)) =
³
det
³
∂y
k
∂x
i
´´
2
. By 46,
√
g = det
³
∂y
k
∂x
i
´
as claimed. Now
∂y
k
∂x
i
∂x
i
∂y
r
= δ
k
r
and so
det
µ
∂x
i
∂y
r
¶
=
1
p
g (x)
.
16
This proves the theorem.
To get the curl and cross product in curvilinear coordinates, let ²
ijk
be the usual permutation symbol. Thus,
²
123
= 1
and when any two indices in ²
ijk
are switched, the sign changes. Thus
²
132
= −1, ²
312
= 1, etc.
Now deﬁne
ε
ijk
(x) ≡ ²
ijk
1
p
g (x)
.
Then for x and z satisfying 46,
ε
ijk
(x)
∂z
r
∂x
i
∂z
s
∂x
j
∂z
t
∂x
k
= ²
ijk
det
µ
∂x
p
∂y
q
¶
∂z
r
∂x
i
∂z
s
∂x
j
∂z
t
∂x
k
= ²
rst
det
µ
∂x
p
∂y
q
¶
det
µ
∂z
i
∂x
k
¶
= ²
rst
det (MN)
where N is the matrix whose pq th entry is
∂x
p
∂y
q
and M is the matrix whose ik th entry is
∂z
i
∂x
k
. Therefore, from the
deﬁnition of matrix multiplication and the chain rule, this equals
= ²
rst
det
µ
∂z
i
∂y
p
¶
≡ ε
rst
(z)
from the above discussion.
Now ε
ijk
(y) = ²
ijk
and for a vector ﬁeld, F,
curl (F) ≡ ε
ijk
(y) F
k,j
(y) e
i
(y) .
Therefore, since we know how everything transforms assuming 46, it is routine to write this in terms of x.
curl (F) = ε
rst
(x)
∂y
i
∂x
r
∂y
j
∂x
s
∂y
k
∂x
t
F
p,q
(x)
∂x
p
∂y
k
∂x
q
∂y
j
e
m
(x)
∂x
m
∂y
i
= ε
rst
(x) δ
m
r
δ
q
s
δ
p
t
F
p,q
(x) e
m
(x) = ε
mqp
(x) F
p,q
(x) e
m
(x) . (49)
More simpliﬁcation is possible. Recalling the deﬁnition of F
p,q
(x) ,
∂F
∂x
q
≡ F
p,q
(x) e
p
(x) =
∂
∂x
q
[F
p
(x) e
p
(x)]
=
∂F
p
(x)
∂x
q
e
p
(x) + F
p
(x)
∂e
p
∂x
q
=
∂F
p
(x)
∂x
q
e
p
(x) −F
r
(x)
½
r
pq
¾
e
p
(x)
by Theorem 10. Therefore,
F
p,q
(x) =
∂F
p
(x)
∂x
q
−F
r
(x)
½
r
pq
¾
17
and so
curl (F) = ε
mqp
(x)
∂F
p
(x)
∂x
q
e
m
(x) −ε
mqp
(x) F
r
(x)
½
r
pq
¾
e
m
(x) .
However, because
©
r
pq
ª
=
©
r
qp
ª
, the second term in this expression equals 0. To see this,
ε
mqp
(x)
½
r
pq
¾
= ε
mpq
(x)
½
r
qp
¾
= −ε
mqp
(x)
½
r
pq
¾
.
Therefore, by 49,
curl (F) = ε
mqp
(x)
∂F
p
(x)
∂x
q
e
m
(x) . (50)
What about the cross product of two vector ﬁelds? Let F and G be two vector ﬁelds. Then in terms of standard
coordinates, y,
F×G =ε
ijk
(y) F
j
(y) G
k
(y) e
i
(y)
= ε
rst
(x)
∂y
i
∂x
r
∂y
j
∂x
s
∂y
k
∂x
t
F
p
(x)
∂x
p
∂y
j
G
q
(x)
∂x
q
∂y
k
e
l
(x)
∂x
l
∂y
i
= ε
rst
(x) δ
p
s
δ
q
t
δ
l
r
F
p
(x) G
q
(x) e
l
(x) = ε
lpq
(x) F
p
(x) G
q
(x) e
l
(x) . (51)
We summarize these results in the following theorem.
Theorem 13 Suppose x is a system of curvilinear coordinates in R
3
such that
det
µ
∂y
i
∂x
j
¶
> 0.
Let
ε
ijk
(x) ≡ ²
ijk
1
p
g (x)
.
Then the following formulas for curl and cross product hold in this system of coordinates.
curl (F) = ε
mqp
(x)
∂F
p
(x)
∂x
q
e
m
(x) ,
and
F×G =ε
lpq
(x) F
p
(x) G
q
(x) e
l
(x) .
18