Math 641 Lecture #36

¶7.22,7.23,7.24
Differentiable Transformations
Definitions (7.22). For an open V ⊂ Rk a map T : V → Rk is differentiable at x ∈ V
if there exists a linear operator A on Rk such that
|T (x + h) − T (x) − Ah|
lim = 0.
h→0 |h|
When T is differentiable at x, the linear operator A is unique, and we write T 0 (x) = A,
and call the linear operator T 0 (x) the derivative or differential of T at x.
When T is differentiable at x, the difference T (x + h) − T (x) is approximated by the
linear function T 0 (x)h.
For a linear operator A : Rk → Rk , Theorem 2.20(e) shows that there is a nonnegative
number ∆(A) such that
m(A(E)) = ∆(A)m(E)
for all Lebesgue measurable subsets E of Rk .
Since the map x → Ax is differentiable where A0 (x) = A, and since every differentiable
transformation can be locally approximated by a constant plus a linear transformation
(a.k.a. an affine map), we might conjecture that
m(T (E))
≈ ∆(T 0 (x))
m(E)
for suitable sets E that are close to x.
This we will prove shortly.
Recall that ∆(A) = | det(A)|.
When T is differentiable at x, the determinant of T 0 (x) is called the Jacobian of T at
x, and is denoted by JT (x).
Thus
∆(T 0 (x)) = |JT (x)|.
To prove the above mentioned conjecture, we will need an elementary result from Topol-
ogy, which we state without proof.
Brouwer’s Fixed Point Theorem. A continuous map f : B(0, 1) → B(0, 1) has a
fixed point, i.e., there exists x ∈ B(0, 1) such that f (x) = x.
Lemma (7.23). Let S = {x : |x| = 1} be the unit sphere in Rk that is the boundary of
the open unit ball B = B(0, 1). If F : B → Rk is continuous, 0 <  < 1, and

|F (x) − x| <  for all x ∈ S,

then F (B) ⊃ B(0, 1 − ).

Proof. By way of contradiction, assume there is a point a ∈ B(0, 1 − ) that is not in
F (B).
The condition |F (x) − x| <  for all x ∈ S implies that |F (x)| > 1 −  for x ∈ S:
 
 |x| − |F (x)|  ≤ |x − F (x)| < ,
⇒ 1 − |F (x)| = |x| − |F (x)| < ,
⇒ |F (x)| > 1 − .

Thus a is not in F (S), and since a 6∈ F (B), we have a 6= F (x) for every x ∈ B.
We can define a continuous map G : B → B by

a − F (x)
G(x) = .
|a − F (x)|

With · the standard scalar product on Rk , an x ∈ S satisfies x · x = 1, so that

x · (a − F (x)) = x · a + x · (x − F (x)) − 1 < |a| +  − 1 < 0.

[Here we used the Cauchy-Schwarz inequality, |x · y| ≤ |x| |y|, twice; once with y = a and
then with y = x − F (x).]
This shows that x · G(x) < 0 for x ∈ S, so that x 6= G(x) for all x ∈ S.
On the other hand, if x ∈ B, then x 6= G(x) because |G(x)| = 1 while |x| < 1.
Thus we have a continuous map G : B → B without a fixed point, contradicting
Brouwer’s Fixed Point Theorem. 
Theorem (7.24). Suppose V is open in Rk and T : V → Rk is continuous. If T is
differentiable at a point x ∈ V , then


m T (B(x, r))
= ∆ T 0 (x) .


lim
r→0 m(B(x, r)

Remark. The set T (B(x, r)) is Lebesgue measurable; it is σ-compact because B(x, r)
is σ-compact and T is continuous.
Proof. WLOG we assume that x = 0 and T (x) = 0, and set A = T 0 (0).
Case 1: A is one-to-one (i.e., A is invertible). The function

F (x) = A−1 T (x), x ∈ V

satisfies F 0 (0) = A−1 T 0 (0) = A−1 A = I (the identity operator).

Since T (x) = AF (x), we have for every ball B that F (B) is Lebesgue measurable, and
so by Theorem 2.20(e),

m(T (B)) = m(A(F (B))) = ∆(A)m(F (B)).

To get the result, it suffices to show that

m(F (B(0, r))

lim = 1;
r→0 m(B(0, r)

this shows that

m(T (B(0, r))) m(F (B(0, r)))
lim = ∆(A) lim = ∆(A).
r→0 m(B(0, r)) r→0 m(B(0, r))

Since F (0) = 0 and F 0 (0) = I, the differentiability of F at 0 implies for each  > 0 there
is a δ > 0 such that
|F (0 + x) − F (0) − Ix|
< , 0 < |x| < δ,
|x|

that is,
|F (x) − x| ≤ |x| for 0 < |x| < δ.
When 0 < r < δ, we claim the following inclusions hold:

B(0, (1 − )r) ⊂ F (B(0, r)) ⊂ B(0, (1 + )r).

The first inclusion follows from Lemma 7.23 applied to B(0, r) in place of B(0, 1) because
|F (x) − x| < r for all x with |x| = r.
The second inclusion follows since |F (x) − x| < |x| implies |F (x)| < (1 + )|x|:
 
|F (x)| − |x| ≤ |F (x)| − |x| ≤ |F (x) − x| < |x|.

Since m(B(0, η)) = A(k)η k , the inclusions imply that

m(F (B(0, r))

(1 − )k ≤ ≤ (1 + )k .
m(B(0, r)

Consequently, as  is arbitrary, we have

m(F (B(0, r))

lim = 1.
r→0 m(B(0, r)

Case 2: A is not one-to-one (i.e., A is not invertible). Here A, as a linear operator on

Rk , maps into a lower dimensional subspace of Rk , i.e., into a set of measure 0.
Thus for  > 0 there is η > 0 such that m(Eη ) <  where Eη is the set of points in Rk
whose distance from A(B(0, 1)) is less than η.
Since T (0) = 0 and T 0 (0) = A, the differentiability of T at 0 implies for each η > 0 there
is a δ > 0 such that
|T (0 + x) − T (0) − Ax|
<η for 0 < |x| < δ,
|x|
that is,
|T (x) − Ax| < η|x| for 0 < |x| < δ.
Thus, if r < δ, then T (B(0, r)) lies in the set E that consists of points whose distance
from A(B(0, r)) is less than ηr.
For y ∈ E, we have

dist(r−1 y, A(B(0, 1))) = dist(r−1 y, r−1 A(B(0, r)))

= r−1 dist(y, A(B(0, r)))
< r−1 (ηr) = η,

so that r−1 y is within a distance of η of A(B(0, 1)), i.e., r−1 y ∈ Eη , or y ∈ rEη .

Thus E ⊂ rEη so that m(E) ≤ m(rEη ) = rk m(Eη ) < rk .
[We have used Theorem 2.20 (e) where the map z → rz is a linear operator on Rk with
determinant rk .]
Since T (B(0, r)) ⊂ E when 0 < r < δ, we have

m T (B(0, r)) < rk , 0 < r < δ.

Since rk = m(B(0, r))/m(B(0, 1)), we get

m(T (B(0, r))) rk rk 

< = k = .
m(B(0, r)) m(B(0, r)) r m(B(0, 1)) m(B(0, 1))

The arbitrariness of  implies that

m(T (B(0, r)))

lim = 0.
r→0 m(B(0, r))

This completes the proof.