18.

024 SPRING OF 2008

DER. DERIVATIVES OF VECTOR FIELDS

Derivatives of vector fields. Derivative theory for vector fields is a straightfor-

ward extension of that for scalar fields. Given f : D ⊂ Rn → Rm a vector field, in
components, f consists of m scalar fields of n variables. That is,
f (x) = (f1 (x), . . . , fm (x)),
n
and each fi : D ⊂ R → R is a scalar field, where i = 1, . . . , m. It seems natural to
define the differentiability of f , thus, componentwise.
We recall that if f : D ⊂ Rn → R is a scalar field and a ∈ int(D) then the
derivative of f at a (if exists) is given by the gradient
∇f (a) = (D1 f (a, D2 f (a), . . . , Dn f (a)).
For our purposes, it is convenient to understand the derivative of f as a row matrix
rather than as a vector. We will write as
Df (a) = [D1 f (a, D2 f (a), . . . , Dn f (a)].
Then, f is differentiable at a (with the total derivative Df (a) if
(*) f (a + h) − f (a) = Df (a)h + khkE(h; a),
and |E(h; a)| → 0 as khk → 0. Here the first term on the right side is matrix
multiplication of Df (a) and the column matrix
 
h1
 .. 
h =  . .
hn
Definition 1. Let f : D ∈ Rm and D ⊂ Rn be open. In scalar form,
f (x) = (f1 (x1 , . . . , xn ), . . . , fm (x1 , . . . , xn )).
We say that f is differentiable at a ∈ D if each of the component functions f1 , . . . , fm
is differentiable at a (in the sense discussed in [Apo, Section 8.11] or above).
Furthermore, by the derivative of f at a we mean the matrix
 
D1 f1 (a) D2 f1 (a) · · · Dn f1 (a)
 D1 f2 (a) D2 f2 (a) · · · Dn f2 a) 
Df (a) =  .
 
.. .. ..
 . . . 
D1 fm (a) D2 fm (a) · · · Dn fm
In other words, Df (a) is the matrix whose ith row is given as the derivative Dfi (a)
of the ith coordinate function of f . It is often called the jacobian matrix∗ of f . An-
other notation for this matrix is
∂(f1 , · · · , fm )
.
∂(x1 , · · · , xn )
∗
It is named after the Prussian mathematician Carl Gustav Jacob Jacobi (1804-1851).
1
2 DERIVATIVES OF VECTOR FIELDS

In [Apo, Section 8.18], the differentiability of a vector field f is given, alterna-

tively, by directly extending the Taylor’s formula (*) in the vector-field setting. Our
definition then is equivalent to that given in [Apo], which is stated below.
Theorem 2. The vector field f is differentiable at a if and only if
f (a + h) − f (a) = Df (a)h + khkE(h; a)
and kE(h; a)k → 0 as khk → 0. Here, f , h and E are written as column matrices.
Proof. We observe that both sides of the equation represent column matrices. Con-
sidering the ith entries of these matrices, we obtain the following equation
fi (a + h) − fi (a) = Dfi )(a) · h + khkEi (h; a).
Now if f is differentiable at a if and only if each fi is, and fi is differentiable at a
if and only if |Ei (h; a)| → 0 as khk → 0. But, |Ei (h)| → 0 as khk → 0 for each i if
and only if kE(h)k → 0 as khk → 0, and we are done. 
The following result is immediate from our definition of differenentiability.
Theorem 3. If f is differentiable at a then f is continuous at a.
The chain rule for derivatives of vector fields. Before considering the chain rule
for vector fields, let us take the chain rule for scalar fields we have already proved
[Apo, Section 8.15] and reformulate it in terms of matrices.
Let f : D ⊂ Rn → R be a scalar field defined in an open ball about a, and let
x : I ⊂ R → Rn be a vector-valued function defined in an open interval about t0 .
Let x(t0 ) = a and x(I) ⊂ D. If f is differentiable at a and if x is differentiable at t0 ,
then we have shown in Theorem 8.8 of [Apo] that f ◦ x is differentiable at t0 and
its derivative is given by
d
f (x(t0 )) = ∇f (x(t0 )) · x0 (t0 ).
dt
We now write this formula in scalar form as
d ∂f dx1 ∂f dxn
(f ◦ x) = + ··· + ;
dt ∂x1 dt ∂xn dt
or, we write it in matrix form as
 dx 
  dt1
d ∂f ∂f  . 
(f ◦ x) = ···  ..  .
dt ∂x1 ∂xn dx n
dt
Recalling the definition of the Jacobian matrix Df the latter formula is recognized
as
d
(f (x(t)) = Df (x(t))Dx(t).
dt
(Note that Df is a row matrix while Dx is by definition a column matrix.) This is
the form of the chain rule which is useful to extend to vector fields.

Let f : Rn → Rp and g : Rp → Rm be vector fields such that the composition

field h = g ◦ f is defined in an open ball about a ∈ Rn . Let f (a) = b ∈ Rp . We
write these fields as
f (x) = f (x1 , . . . , xn ), g(y) = g(y1 , . . . , yp ).
 DERIVATIVES OF VECTOR FIELDS 3

If f and g are differentiable at a and b, respectively, then partial derivatives of

h exist at a by applying the chain rule for each hi (x) = gi (f (x)) as

∂hi ∂gi ∂f1 ∂gi ∂fp

= + ··· + .
∂xj ∂y1 ∂xj ∂yp ∂xj
This is recognized as matrix multiplication
 
Dj f1
[D1 gi D2 gi · · · Dp gi ]  ...  .
 

Dj fp
In other words, its multiplication of the ith row of Dg and the jth column of Df .
Thus, the Jacobian matrix of h is expected to satisfy the matrix equation
Dh(a) = Dg(b)Df (a).
Not exactly. If f and g are differentiable then we know that the partial deriva-
tives of the composite function h exist. But, the mere existence of the partial deriva-
tives of the function hi does not imply the differentiability of hi , and thus, one
needs to provide a separate argument that h is differentiable.
One may avoid giving a separate proof of the differentiability of h by assum-
ing a stronger hypothesis, namely, that both f and g are continuously differentiable.
Then, the formula,
Xp
Dj hi (x) = Dl gi (f (x))Dj fl (x),
l=1
which we have proved, shows that Dj hi is continuous, that is, h is continuously
differentiable. Therefore, h is differentiable.
We summarize these.
Theorem 4. Let f : Rn → Rp and g : Rp → Rm be vector fields defined in open balls
about a ∈ Rn and b ∈ Rp , respectively. Let f (a) = b. If f and g are continuously differ-
entiable on their respective domains, then h(x) = g ◦ f (x) is continuously differentiable
on its domain and
Dh(x) = Dg(f (x) · Df (x).
Most applications of the chain rule we will use in future are in the C 1 -class
setting, and thus this theorem will suffice. But, one may remove the C 1 condition.
Please read the statement and its proof of [Apo, Theorem 8.11].
Example 5 (Polar coordinates). Given a function f (x, y) defined on the (x, y) plane,
we introduce polar coordinates
x = r cos θ and y = r sin θ
and f becomes a function of r and θ as φ(r, θ) = f (r cos θ, r sin θ). Then,
∂φ ∂f ∂x ∂f ∂y ∂f ∂f
= + = cos θ + sin θ
∂r ∂x ∂r ∂y ∂r ∂x ∂y
and
∂φ ∂f ∂f
= −r sin θ + r cos θ.
∂θ ∂x ∂y
4 DERIVATIVES OF VECTOR FIELDS

The second derivative is computed as

∂2φ
   
∂ ∂φ ∂ ∂f ∂f
= = cos θ + sin θ
∂r2 ∂r ∂r ∂r ∂x ∂y
   
∂ ∂f ∂f ∂(cos θ) ∂ ∂f ∂f ∂(sin θ)
= cos θ + + sin θ +
∂r ∂x ∂x ∂r ∂r ∂y ∂y ∂r
           
∂ ∂f ∂ ∂f ∂ ∂f ∂ ∂f
= cos θ + sin θ cos θ + cos θ + sin θ sin θ
∂x ∂x ∂y ∂x ∂x ∂y ∂y ∂y
2 2 2
∂ f ∂ f ∂ f
= + 2 + sin θ cos θ.
∂x2 ∂y ∂x∂y
An analogous formula for ∂ 2 φ/∂θ2 is worked in [Apo, Example 3] and ∂ 2 φ/∂r∂θ
is in [Apo, Section 8.22, Exercise 5].
Derivatives of inverses. Recall that if f (x) is a differentiable real-valued function
of a single variable x, and if f 0 (x) > 0 for x ∈ [a, b] (or f 0 (x) < 0), then f is
strictly increasing (or decreasing), and it has an inverse, say g. Furthermore, g is
differentiable and its derivative satisfies
1
g 0 (f (x)) = 0 .
f (x)
Part of this theory extends to vector fields.
Theorem 6. Let f : Rn → Rn be defined in an open ball of a and f (a) = b. Suppose that
f has an inverse, say g.
If f is differentiable at a and if g is differentiable at b then
Dg(b) = (Df (a))−1 .
In interpretation, the jacobian matrix of the inverse function is just the inverse
of the jacobian matrix.

Proof. We observe that g ◦ f (x) = x. Since f and g are differentiable and so is the
composite function g ◦ f , by the chain rule, it follows that
Dg(b)Df (a) = In ,
which implies the assertion. 

The theorem shows that in order for the differentiable function f to have a dif-
ferentiable inverse, it is necessary that the Jacobian matrix Df have rank n. Roughly
speaking, it is also sufficient for f to have an inverse. More precisely, one has the fol-
lowing Inverse Function Theorem, which is one of the fundamental results in math-
ematical analysis.
Theorem 7 (Inverse function Theorem). If f : D ⊂ Rn → Rn is of C 1 in a neigh-
borhood of a and Df (a) has rank n, then there is an open ball B ∈ Rn about a such that
f : B → C = f (B) is one-to-one and onto. Furthermore, the inverse function g : C → B
of f is continuously differentiable and Dg(f (x)) = (Df (x))−1 .
The proof and further application of the inverse function theorem is learned in
18.100A.
 DERIVATIVES OF VECTOR FIELDS 5

Example 8 (Invertibility of the polar coordinate transformation). Let

f : R2 → R2 , f (r, θ) = (x, y)
be the differentiable transformation from the (x, y)-plane to (r, θ) plane by polar
coordinates:
x = r cos θ and y = r sin θ.
Its Jacobian matrix is computed as
 ∂x ∂x   
∂(x, y) ∂r ∂θ = cos θ −r sin θ .
= ∂y ∂y
∂(r, θ) ∂r ∂θ
sin θ r cos θ
Note that its determinant is r and thus, it is of rank 2 provided that r 6= 0. Ac-
cording to the inverse function theorem,thus, away from the origin (r = 0), the
polar coordinate transformation f is invertible locally, and its inverse is given by
g(x, y) = (r, θ). Moreover,
−1  
∂(r, θ) ∂(x, y) 1 r cos θ r sin θ
= = .
∂(x, y) ∂(r, θ) r − sin θ cos θ
R EFERENCES
[Apo] T. Apostol, Calculus, vol. II, Second edition, Wiley, 1967.

2008
c BY V ERA M IKYOUNG H UR
E-mail address: [email protected]