Professional Documents
Culture Documents
Msii Lectures 2012-05-04 Signed
Msii Lectures 2012-05-04 Signed
Msii Lectures 2012-05-04 Signed
L := x 2 RN : x = x0 + tv; t 2 R ;
In view of the previous remark, one would be tempted to say that if the
directional derivatives at x0 exist and are …nite in every direction, then f is
continuous at x0 . This is false in general, as the following example shows.
Example 3 Let
1 if y = x2 ; x 6= 0;
f (x; y) :=
0 otherwise.
Given a direction v = (v1 ; v2 ), the line L through 0 in the direction v intersects
the parabola y = x2 only in 0 and in at most one point. Hence, if t is very
small,
f (0 + tv1 ; 0 + tv2 ) = 0:
It follows that
1
Exercise 4 Let
(
x2 y
x4 +y 2 if (x; y) 6= (0; 0) ;
f (x; y) :=
0 if (x; y) = (0; 0) :
T (x + y) = T (x) + T (y)
f (x) f (x0 ) T (x x0 )
lim = 0:
x!x0 kx x0 k
provided the limit exists. The function T , if it exists, is called the di¤erential of
f at x0 and is denoted df (x0 ) or dfx0 .
2
Example 8 Let V = ff : [ 1; 1] ! R : f is di¤ erentiable in [ 1; 1]g. Note
that V is a normed space with the norm kf k := maxx2[ 1;1] jf (x)j. Consider
the linear function T : V ! R de…ned by
T (f ) := f 0 (0) :
R (x)
lim = 0:
x!x0 kx x0 k
Then
f (x) f (x0 ) = T (x x0 ) + R (x) = b (x x0 ) + R (x) :
Hence, by Cauchy’s inequality for x 2 E, x 6= x0 ,
3
Theorem 10 Let E RN , let f : E ! R be di¤ erentiable at some point
x0 2 E . Then
(i) all the directional derivatives of f at x0 exist and
@f
(x0 ) = dfx0 (v) ;
@v
(ii) for every direction v,
X @fN
@f
(x0 ) = (x0 ) vi : (1)
@v i=1
@xi
Remark 11 If in the previous theorem x0 is not an interior point but for some
direction v 2 RN , the point x0 is an accumulation point of the set E \ L, where
L is the line through x0 in the direction v, then as in the …rst part of the proof
@f
we can show that there exists the directional derivative @v (x0 ) and
@f
(x0 ) = dfx0 (v) :
@v
4
If all the partial derivatives of f at x0 exist, the vector
@f @f
(x0 ) ; : : : ; (x0 ) 2 RN
@x1 @xN
is called the gradient of f at x0 and is denoted by rf (x0 ) or grad f (x0 ) or
Df (x0 ). Note that part (ii) of the previous theorem shows that
XN
@f
dfx0 (v) = T (v) = rf (x0 ) v = (x0 ) vi : (2)
i=1
@xi
Exercise 13 Let
(
x2 y
x2 +y 4 if (x; y) 6= (0; 0) ;
f (x; y) :=
0 if (x; y) = (0; 0) :
Find all directional derivatives of f at 0. Study the continuity and the di¤ eren-
tiability of f at 0.
Example 14 Let
x if y = x2 ; x 6= 0;
f (x; y) :=
0 otherwise.
5
It follows that
@f f (0 + tv1 ; 0 + tv2 ) f (0; 0) 0 0
(0; 0) = lim = lim = 0:
@v t!0 t t!0 t
Thus formula (1) holds. Moreover, f is continuous in 0, since jf (x; y)j jxj !
0 as (x; y) ! (0; 0). We claim that f is not di¤ erentiable at (0; 0). To negate
di¤ erentiability, in view of the previous theorem and since (0; 0) is an interior
point, we need to prove that the quotient
f (x; y) f (0; 0) rf (0; 0) (x; y)
q
2 2
(x 0) + (y 0)
f x; x2 f (0; 0) rf (0; 0) x; x2 x 0 0 x; x2
q =q
2 2 2 2
(x 0) + (x2 0) (x 0) + (x2 0)
x x
=p = p 9 0:
x2 + x4 jxj 1 + x2
Exercise 15 Let f : E ! R be Lipschitz and let x0 2 E .
@f
(i) Assume that all the directional derivatives of f at x0 exist and that @v (x0 ) =
PN @f
i=1 @xi (x0 ) vi for every direction v. Prove that f is di¤ erentiable at x0 .
(ii) Assume that all the partial derivatives of f at x0 exist, that the directional
@f
derivatives @v (x0 ) exist for all v 2 S, where S is dense in the unit sphere,
@f PN @f
and that @v (x0 ) = i=1 @x i
(x0 ) vi for every direction v 2 S. Prove that
f is di¤ erentiable at x0 .
6
where z1 := ( 1 x1 + (1 1 ) y1 ; x2 ; : : : ; xN ) for some 1 2 (0; 1). Note that
kz1 x0 k kx x0 k :
Similarly, for i = 2; : : : ; N ,
@f
f (y1 ; : : : ; yi 1 ; xi ; : : : ; xN ) f (y1 ; : : : ; yi 1 ; yi ; : : : ; xN ) = (zi ) (xi yi ) ;
@xi
where zi := (y1 ; : : : ; yi 1 ; i xi + (1 i ) yi ; xi+1 ; : : : ; xN ) for some i 2 (0; 1)
and
kzi x0 k kx x0 k :
Write
Then
XN
jf (x) f (x0 ) rf (x0 ) (x x0 )j @f @f jxi yi j
(zi ) (x0 ) :
kx x0 k i=1
@xi @xi kx x0 k
jxi yi j
Since kx x0 k 1, we have that
XN
jf (x) f (x0 ) rf (x0 ) (x x0 )j @f @f
0 (zi ) (x0 ) : (4)
kx x0 k i=1
@xi @xi
XN
jf (x) f (x0 ) rf (x0 ) (x x0 )j @f @f
0 (zi ) (x0 ) ! 0;
kx x0 k i=1
@xi @xi
7
Proof. Without loss of generality, we may assume that i = N . Reasoning as
before we have that
f (x) f (x0 ) rf (x0 ) (x x0 )
N
X1 @f @f
= (zi ) (x0 ) (xi yi )
i=1
@xi @xi
f (y1 ; : : : ; yN 1 ; xN ) f (y1 ; : : : ; yN ) @f
+ (x0 ) (xN yN ) :
xN yN @xN
Hence, as before,
N
X1
jf (x) f (x0 ) rf (x0 ) (x x0 )j @f @f
0 (zi ) (x0 ) (5)
kx x0 k i=1
@xi @xi
f (y1 ; : : : ; yN 1 ; yN + (xN yN )) f (y1 ; : : : ; yN ) @f
+ (x0 ) :
xN yN @xN
Since t := xN yN ! 0 as x ! x0 , we have that
f (y1 ; : : : ; yN 1 ; yN + (xN yN )) f (y1 ; : : : ; yN )
xN yN
f (y1 ; : : : ; yN 1 ; yN + t) f (y1 ; : : : ; yN ) @f
= ! (x0 ) ;
t @xN
and so the right-hand side of (5) goes to zero as x ! x0 .
The next exercise shows that the previous conditions are su¢ cient but not
necessary for di¤erentiability.
Example 18 Let
1
x2 + y 2 sin x+y if (x; y) 6= (0; 0) or x + y 6= 0;
f (x; y) :=
0 otherwise.
Let’s prove that f is di¤ erentiable at (0; 0) but the partial derivatives are not
continuous at (0; 0). We have
1
f (x; 0) f (0; 0) x2 + 0 sin x+0 0 1
= = x sin !0
x 0 x 0 x
@f @f
as x ! 0, since sin x1 is bounded. Hence, @x (0; 0) = 0. Similarly, @y (0; 0) = 0.
Hence,
8 2 2
f (x; 0) f (0; 0) @f (0; 0) (x 0) @f
(0; 0) (y 0) < (x +y
p 2) 2x+y
sin 1
if x + y 6= 0;
q @x @y
= x +y
2 2 : 0 otherwise.
(x 0) + (y 0)
p 1
x2 + y 2 sin x+y if x + y 6= 0;
=
0 otherwise.
!0
8
1
p
as (x; y) ! (0; 0) since sin x+y is bounded and x2 + y 2 ! 0. Hence, f is
di¤ erentiable at (0; 0).
On the other hand, if x + y 6= 0,
!
@f 1 1 1
(x; y) = (2x + 0) sin + x2 + y 2 cos 2 ;
@x x+y x+y (x + y)
!
@f 1 1 1
(x; y) = (0 + 2y) sin + x2 + y 2 cos 2 :
@y x+y x+y (x + y)
Taking x = y gives
@f 1 1
(x; x) = 2x sin cos ;
@x 2x 2x
@f 1 1
(x; x) = 2x sin cos
@y 2x 2x
E U [ V; E \ U 6= ;; E \ V 6= ;; E \ U \ V = ;:
E A [ B; E \ A 6= ;; E \ B 6= ;; E \ A \ B = ;;
9
Theorem 21 A set C R is connected if and only if it is convex.
Proof. Exercise.
We now introduce another notion of connectedness, which is simpler to verify.
Proof. We claim that E is connected. If not, then there exist two nonempty
open sets U; V RN such that
E U [ V; E \ U 6= ;; E \ V 6= ;; E \ U \ V = ;:
which is a contradiction.
The next example and exercise show that in RN a connected set may fail to
be pathwise connected, unless the set is open.
E1 = (x; y) 2 R2 : x = 0; 1 y 1 ;
1
E2 = (x; y) 2 R2 : x > 0; y = sin ;
x
E = E1 [ E2 :
for some ci ; di 2 RN .
10
Theorem 28 Let O RN be open and connected. Then O is pathwise con-
nected.
and
We claim that U and V are open. To see this, let x 2 U . Since O is open, there
exists B (x; r) O. Let’s show that B (x; r) U . Indeed, let y 2 B (x; r).
Since x 2 U , there exists a polygonal path ' : [a; b] ! RN such that ' (a) = x0 ,
' (b) = x and ' ([a; b]) O. De…ne the polygonal path : [a; b + 1] ! RN as
follows
' (t) if t 2 [a; b] ;
(t) :=
x + (t b) y if t 2 [b; b + 1] :
Then (a) = x0 , (b + 1) = y. Since ([a; b + 1]) = ' ([a; b]) [ ([b; b + 1])
and ([b; b + 1]) is the segment joining x with y, which is contained in B (x; r),
we have that ([a; b + 1]) is contained in O. Hence, B (x; r) U . This shows
that U is open.
To prove that V is open, let x 2 V . Since O is open, there exists B (x; r) O.
Let’s show that B (x; r) V . Indeed, let y 2 B (x; r). Assume by contradiction
that y 2 U . Then there exists a polygonal path ' : [a; b] ! RN such that ' (a) =
x0 , ' (b) = y and ' ([a; b]) O. De…ne the polygonal path : [a; b + 1] ! RN
as follows
' (t) if t 2 [a; b] ;
(t) :=
y + (t b) x if t 2 [b; b + 1] :
Then (a) = x0 , (b + 1) = x and ([a; b + 1]) is contained in O, which shows
that x 2 U , which is a contradiction. Hence, B (x; r) V . This shows that V
is open.
Hence, U and V are open and by their de…nition they are disjoint and
O = U [ V . Moreover, U is nonempty, since x0 2 U . Thus, V must be
empty, otherwise O would be disconnected. Hence, O = U . It follows that
U is pathwise connected. Indeed, if x; y 2 O then there exist two polygonal
paths ' : [a; b] ! RN and : [c; d] ! RN such that ' (a) = x0 , ' (b) = x and
' ([a; b]) O, and (c) = x0 , (d) = y and ([c; d]) O. By “gluing” these
two polygonal paths appropriately, we can construct a polygonal path joining x
and y with range contained in O.
11
Friday, January 27, 2012
Next we show that if a set is not connected, we can decompose it uniquely
into a disjoint union of maximal connected subsets.
Proof. We claim that E is connected. If not, then there exist two nonempty
open sets U; V RN such that
E U [ V; E \ U 6= ;; E \ V 6= ;; E \ U \ V = ;:
12
The proof makes use of the mean value theorem.
@f
f (x) f (y) = ( x + (1 ) y) kx yk : (6)
@v
Lemma 37 Under the hypotheses of the previous theorem, the function g (t) :=
f (tx + (1 t) y), t 2 [0; 1], is di¤ erentiable for all t 2 (0; 1), with
@f
g 0 (t) = (tx + (1 t) y) kx yk :
@v
Proof. Fix t0 2 (0; 1) and consider
Proof of the Mean Value Theorem. Consider the function g (t) := f (tx + (1 t) y),
t 2 [0; 1]. Since compositions of continuous functions is continuous, we have that
g is continuous. By the previous lemma, we are in a position to apply the mean
value theorem to the function g to …nd 2 [0; 1] such that
dg
g (1) g (0) = ( ) (1 0) ;
dt
that is,
@f
f (x) f (y) = ( x + (1 ) y) kx yk :
@v
13
Remark 38 If f is de…ned on a larger domain E and all points of S except at
most x and y are interior points of E, then by (1), we have that
XN
@f @f (xi yi )
( x + (1 ) y) = ( x + (1 ) y) ;
@v i=i
@xi kx yk
XN
@f
f (x) f (y) = ( x + (1 ) y) (xi yi ) :
i=i
@xi
XN
@f
f (x) f (y) = ( x + (1 ) y) (xi yi ) = 0:
i=i
@xi
A := fx 2 U : f (x) = cg :
We claim that A is open. Indeed, if x 2 A, then by the previous step there exists
B (x; r) U such that f is constant in B (x; r). Hence, f = c in B (x; r), which
implies that B (x; r) A. Thus every point of A is an interior point and so A
is open. Moreover, A is nonempty since x0 2 A. Next consider the set
1
A1 = fx 2 U : f (x) 6= cg = f (( 1; c) [ (c; 1)) :
U = A [ A1 ;
where A and A1 are open and disjoint. Since A is nonempty and U is connected,
necessarily A1 must be empty, since otherwise we would negate the fact that
U is connected. Hence, U = A, that is, f (x) = c for all x 2 U .
Monday, January 30, 2012
14
2 Higher Order Derivatives
Let E RN , let f : E ! R and let x0 2 E. Let i 2 f1; : : : ; N g and assume
@f
that there exists the partial derivatives @x i
(x) for all x 2 E. If j 2 f1; : : : ; N g
and x0 is an accumulation point of E \ L, where L is the line through x0 in
@f
the direction ej , then we can consider the partial derivative of the function @x i
with respect to xj , that is,
@ @f @2f
=: :
@xj @xi @xj @xi
Example 39 Let
y 2 arctan xy if y 6= 0;
f (x; y) :=
0 if y = 0:
If y 6= 0, then
@f @ x 1 @ x
(x; y) = y 2 arctan = y2 2
@x @x y x @x y
1+ y
y3
= ;
x2 + y2
and
@f @ x x 1 @ x
(x; y) = y 2 arctan = 2y arctan + y2 2
@y @y y y x @y y
1+ y
x xy 2
= 2y arctan ;
y x2 + y 2
@f f (x0 + t; 0) f (x0 ; 0) 0 0
(x0 ; 0) = lim = lim = 0;
@x t!0 t t!0 t
@f f (x0 ; 0 + t) f (x0 ; 0) t arctan xt0
2
0
(x0 ; 0) = lim = lim
@y t!0 t t!0 t
x0
= lim t arctan = 0;
t!0 t
where we have used the fact that arctan xt0 is bounded and t ! 0. Thus,
( (
y3 2
@f 2 2 if y =
6 0; @f 2y arctan xy x2xy
+y 2 if y 6= 0;
(x; y) = x +y (x; y) =
@x 0 if y = 0; @y 0 if y = 0:
15
@2f
To …nd @y@x (0; 0), we calculate
@f @f
@2f @ @f @x (0; 0 + t) @x (0; 0)
(0; 0) = (0; 0) = lim
@y@x @y @x t!0 t
t3
0+t2 0
= lim = lim 1 = 1;
t!0 t t!0
while
@f @f
@2f @ @f @y (0 + t; 0) @y (0; 0)
(0; 0) = (0; 0) = lim
@x@y @x @y t!0 t
0 0
= lim = lim 0 = 0:
t!0 t t!0
@2f @2f
Hence, @x@y (0; 0) 6= @y@x (0; 0).
Exercise 40 Let
(
x3 y xy 3
x2 +y 2 if (x; y) 6= (0; 0) ;
f (x; y) :=
0 if (x; y) = (0; 0) :
@2f @2f
Prove that @x@y (0; 0) 6= @y@x (0; 0).
@2f @2f
(x0 ) = (x0 ) :
@xi @xj @xj @xi
Proof. Let ` = lim(s;t)!(0;0) A (s; t). Then for every " > 0 there exists =
((0; 0) ; ") > 0 such that
jA (s; t) `j "
q
2 2
for all (s; t) 2 (( r; r) n f0g) (( r; r) n f0g), with js 0j + jt 0j .
16
Fix s 2 2; 2 n f0g. Then for all t 2 2; 2 n f0g,
jA (s; t) `j "
and so letting t ! 0 in the previous inequality (and using the fact that the limit
limt!0 A (s; t) exists), we get
for all s 2 2 ; 2 nf0g. But this implies that there exists lims!0 limt!0 A (s; t) =
`.
Proof of Theorem 41. Step 1: Assume that N = 2. Let jtj ; jsj < pr2 . Then
the points (x0 + s; y0 ), (x0 + s; y0 + t), and (x0 ; y0 + t) belong to B ((x0 ; y0 ) ; r).
De…ne
f (x0 + s; y0 + t) f (x0 + s; y0 ) f (x0 ; y0 + t) + f (x0 ; y0 )
A (s; t) := ;
st
g (x) := f (x; y0 + t) f (x; y0 ) :
@2f
h (b) h (a) = h0 (c) (b a) = ( t ; c) (b a)
@y@x
for some c between a and b. Taking b = t and a = 0, we get
@f @f @2f
( t ; y0 + t) ( t ; y0 ) = ( t; t) t
@x @x @y@x
where t is between y0 and y0 + t. Hence,
@2f @2f
A (s; t) = ( t; t) ! (x0 ; y0 ) ;
@y@x @y@x
where we have used the fact that ( ; ) ! (x0 ; y0 ) as (s; t) ! (0; 0) together
@2f
with the continuity of @y@x at (x0 ; y0 ). Note that this shows that there exists
the limit
@2f
lim A (s; t) = (x0 ; y0 ) :
(s;t)!(0;0) @y@x
17
On the other hand, for all s 6= 0,
@2f
(x0 ; y0 ) = lim A (s; t) = lim lim A (s; t)
@y@x (s;t)!(0;0) s!0 t!0
@f @f
@y (x0 + s; y0 ) @y (x0 ; y0 ) @2f
= lim = (x0 ; y0 ) :
s!0 s @x@y
@ @j j
:= ;
@x @x1 1 @xNN
@0f
where x = (x1 ; : : : ; xN ). If = 0, we set @x0 := f .
@ (2;1;0) @3
= :
@ (x; y; z)
(2;1;0) @x2 @y
! := 1! N !; x := x1 1 xNN :
If = 0, we set x0 := 1.
Using this notation, we can extend the binomial theorem.
18
Theorem 44 (Multinomial Theorem) Let x = (x1 ; : : : ; xN ) 2 RN and let
n 2 N. Then
n
X n!
(x1 + + xN ) = x :
!
m ulti-in dex, j j=n
Proof. Exercise.
Given an open set U RN , for every nonnegative integer m 2 N0 , we
m
denote by C (U ) the space of all functions that are continuous together with
T
1
their partial derivatives up to order m. We set C 1 (U ) := C m (U ).
m=0
where
Rm (x)
lim m = 0:
x!x0 kx x0 k
with for all t 2 [0; 1]. By repeated applications of Lemma 37 and Remark 38,
we get that
d(n) g n
(t) = (h r) f (x0 + th)
dtn
n
for all n = 1; : : : ; m, where (h r) means that we apply the operator
@ @
h r = h1 + + hN
@x1 @xN
19
n times to f . By the multinomial theorem, and the fact that for functions in
C m partial derivatives commute,
n
n @ @
(h r) = h1 + + hN
@x1 @xN
X n! @
= h ;
! @x
multi-index, j j=n
and so
d(n) g X n! @ f
(t) = h (x0 + th) :
dtn ! @x
multi-index, j j=n
and set
X 1 @ f @ f
Rm (x) := (x x0 ) (x0 + c (x x0 )) (x0 ) :
! @x @x
j j=m
Fix " > 0 and …nd = (x0 ; ") > 0 such that
@ f @ f
(x) (x0 ) "
@x @x
for all j j = m and all x 2 U with kx x0 k . Then, since (x x0 )
m
kx x0 k , we have that
jRm (x)j X 1 @ f @ f
m (x0 + c (x x0 )) (x0 )
kx x0 k ! @x @x
j j=m
X 1
" ;
!
j j=m
20
Rm (x)
which shows that limx!x0 kx x0 km = 0.
Example 46 Let’s calculate the limit
y
(1 + x) 1
lim p :
(x;y)!(0;0) x2 + y 2
By substituting we get 00 . Consider the function
y y
f (x; y) = (1 + x) 1 = elog(1+x) 1 = ey log(1+x) 1;
which is de…ned in the set U := (x; y) 2 R2 : 1 + x > 0 . The function f is of
class C 1 . Let’s use Taylor’s formula of order m = 1 at (0; 0),
@f @f p
f (x; y) = f (0; 0) + (0; 0) (x 0) + (0; 0) (y 0) + o x2 + y 2 :
@x @y
We have
@f @ 1
(x; y) = ey log(1+x) 1 = ey log(1+x) y ;
@x @x 1+x
@f @
(x; y) = ey log(1+x) 1 = ey log(1+x) y log (1 + x) ;
@y @y
and so p
f (x; y) = 0 + 0 (x 0) + 0 (y 0) + o x2 + y 2 ;
which means that
p
(1 + x)
y
1 o x2 + y 2
lim p = lim p = 0:
(x;y)!(0;0) x2 + y 2 (x;y)!(0;0) x2 + y 2
Note that if we had to calculate the limit
y
(1 + x) 1
lim 2 2
;
(x;y)!(0;0) x +y
then we would need Taylor’s formula of order m = 2 at (0; 0),
@f @f
f (x; y) = f (0; 0) + (0; 0) (x 0) + (0; 0) (y 0)
@x @y
1 @2f 2 1 @2f
+ (0; 0) (x 0) + (0; 0) (x 0) (y 0)
(2; 0)! @x2 (1; 1)! @x@y
1 @2f 2
+ (0; 0) (y 0) + o x2 + y 2 :
(0; 2)! @y 2
Another simpler method would be to use the Taylor’s formulas for et and log (1 + s).
Exercise 47 Calculate the limit
1 cos [(x 2) y]
lim h i:
(x;y)!(2;0) 2
log 1 + (x 2) + y 2
21
3 Local Minima and Maxima
De…nition 48 Let f : E ! R, where E RN , and let x0 2 E. We say that
(i) f attains a local minimum at x0 if there exists r > 0 such that f (x)
f (x0 ) for all x 2 E \ B (x0 ; r),
(ii) f attains a global minimum at x0 if f (x) f (x0 ) for all x 2 E,
(iii) f attains a local maximum at x0 if there exists r > 0 such that f (x)
f (x0 ) for all x 2 E \ B (x0 ; r),
(iv) f attains a global maximum at x0 if f (x) f (x0 ) for all x 2 E.
Theorem 49 Let f : E ! R, where E RN . Assume that f attains a local
minimum (or maximum) at some point x0 2 E. If there exists a direction v
and > 0 such that the set
fx0 + tv : t 2 ( ; )g E (9)
@f @f
and if there exists @v(x0 ), then necessarily, (x0 ) = 0. In particular, if x0 is
@v
an interior point of E and f is di¤ erentiable at x0 , then all partial derivatives
and directional derivatives of f at x0 are zero.
Proof. Exercise.
Remark 50 In view of the previous theorem, when looking for local minima
and maxima, we have to search among the following:
Interior points at which f is di¤ erentiable and rf (x) = 0, these are
called critical points. Note that if rf (x0 ) = 0, the function f may not
attain a local minimum or maximum at x0 . Indeed, consider the function
f (x) = x3 . Then f 0 (0) = 0, but f is strictly increasing, and so f does
not attain a local minimum or maximum at 0.
Interior points at which f is not di¤ erentiable. The function f (x) = jxj
attains a global minimum at x = 0, but f is not di¤ erentiable at x = 0.
Boundary points.
To …nd su¢ cient conditions for a critical point to be a point of local minimum
or local maximu, we study the second order derivatives of f .
De…nition 51 Let f : E ! R, where E RN , and let x0 2 E. The Hessian
matrix of f at x0 is the N N matrix
0 @2f @2f
1
@x2
(x0 ) @xN @x1 (x0 )
B . 1 .. C
Hf (x0 ) : = B
@ .. .
C
A
@2f @2f
@x1 @xN (x0 ) @x2N
(x0 )
N
@2f
= (x0 ) ;
@xi @xj i;j=1
whenever it is de…ned.
22
Remark 52 If the hypotheses of Schwartz’s theorem are satis…ed for all i; j =
1; : : : ; N , then
@2f @2f
(x0 ) = (x0 ) ;
@xi @xj @xj @xi
which means that the Hessian matrix Hf (x0 ) is symmetric.
p (t) = tN + aN 1t
N 1
+ aN 2t
N 2
+ + a1 t + a0 ; t 2 R;
where the coe¢ cients ai are real for every i = 0; : : : ; N 1, assume that all roots
of p are real. Then
(i) all roots of p are positive if and only if the coe¢ cients alternate sign, that
is, aN 1 < 0, aN 2 > 0, aN 3 < 0, etc.
(ii) all roots of p are negative if and only ai > 0 for every i = 0; : : : ; N 1.
(x y) = ( x) y = (Hx) y = x (Hy) = x ( y) = (x y) ;
and so
( ) (x y) = 0:
Since 6= , it follows that x y = 0.
Step 2: We prove the existence of one real eigenvalue 1. Consider the function
PN PN
(Hx) x j=1 i=1 Hij xi xj
f (x) := 2 = 2
kxk x1 + + x2N
23
(Hx) x
de…ned for all x 2 RN , x 6= 0. The expression kxk2
is called the Rayleigh
quotient. Note that f is of class C 1 and that
PN PN PN PN
@f x21 + + x2N j=1 i=1 Hij ( i;k xj + xi j;k ) j=1 i=1 Hij xi xj 2xk
(x) = 2
@xk (x21 + + x2N )
PN PN PN PN
x21 + + x2N j=1 Hkj xj + i=1 Hik xi j=1 i=1 Hij xi xj 2xk
= 2
(x21 + + x2N )
(10)
PN PN PN
2 x21 + + x2N j=1 Hkj xj j=1 i=1 Hij xi xj 2xk
= 2
(x21 + + x2N )
(Hx)k (Hx) x
=2 2 2 4 xk
kxk kxk
for all k = 1; : : : ; N , where in the third equality we have used the fact that H
is symmetric.
Since f is a continuous function and @B (0; 1) is compact, by the Weierstrass
theorem, there exists
max f (x) = f (x1 ) :
x2@B(0;1)
x
Hence, f (x) f (x1 ) for all x 2 @B (0; 1). If now x 2 RN , x 6= 0, then kxk
belongs to @B (0; 1), and so
x
f (x) = f f (x1 ) ;
kxk
where we have used the fact that f (tx) = f (x) for t > 0. This shows that x1
is a point of absolute maximum for f in RN n f0g. Since x1 is an interior point
and f is of class C 1 , it follows from Theorem 49 that x1 is a critical point of
f , that is,
rf (x1 ) = 0
for all k = 1; : : : ; N . It follows by (10) and the fact that kx1 k = 1 that
Hx1 ((Hx1 ) x1 ) x1 = 0;
where
1 := (Hx1 ) x1 :
This shows that 1 is a real eigenvalue of H and x1 is the corresponding eigen-
vector. Note that
24
Step 3: Let 1 n N 1 and assume by induction that real eigenvalues
1 ,. . . , n of H have been found with corresponding orthonormal eigenvectors
x1 ,. . . , xn . Note that the eigenvalues may not be distinct, but xi xj = i;j for
all i; j = 1; : : : ; n. Consider the set
Y := x 2 RN : x xi = 0 for all i = 1; : : : ; n :
x
f (x) = f f (xn+1 ) ;
kxk
where we have used the fact that f (tx) = f (x) for t > 0. In particular, if v 2 Y
is a direction, the for every t 2 R,
which shows that the function g (t) := f (xn+1 + tv) has a maximum at t = 0.
Hence,
@f
0 = g 0 (0) = (xn+1 ) :
@v
Using Theorems 16 and 10, we get that
rf (xn+1 ) v = 0
for all vectors v 2 Y of norm one, and, in turn, for all v 2 Y . By (10) and the
fact that kxn+1 k = 1 it follows that
where we have used the fact that xn+1 2 Y . Thus the vector rf (xn+1 ) =
2 (Hxn+1 ((Hxn+1 ) xn+1 ) xn+1 ) belongs to Y and so it is orthogonal to itself
by (11). It follows that
so that
Hxn+1 = n+1 xn+1 ;
25
where n+1 := (Hxn+1 ) xn+1 . This proves that n+1 is an eigenvalue and
xn+1 a corresponding eigenvector. Note that n+1 may coincide with some of
the previous eigenvalues, but since xn+1 belongs to Y , xn+1 is distinct from all
the previous eigenvectors x1 ,. . . , xn . Moreover,
(tIN H) xi = (t i ) xi ;
det (tIN H) = (t 1) (t N) :
26
Proof. (i) Assume that Hf (x0 ) is positive de…nite. Then by Remark ??,
N X
X N
@2f 2
(x0 ) vi vj m kvk (12)
j=1 i=1
@xj @xi
where we have used the fact that x0 is a critical point and where
R2 (x)
lim 2 = 0:
x!x0 kx x0 k
m
Using the de…nition of limit with " = 2, we can …nd > 0 such that
R2 (x) m
2 2
kx x0 k
Since g (t) = f (x0 + tv) f (x0 ) = g (0) for all t 2 ( r; r), g attains a local
minimum at t = 0. By two applications of Lemma 37 and Remark 38, we have
that for t 2 ( r; r),
XN
@f
g 0 (t) = (x0 + tv) vi ;
i=1
@xi
N X
X N
@2f
g 00 (t) = (x0 + tv) vi vj :
j=1 i=1
@xj @xi
27
Hence, g 2 C 2 ( r; r), and so, since g attains a local minimum at t = 0,
N X
X N
@2f
0 g 00 (0) = (x0 ) vi vj :
j=1 i=1
@xj @xi
Since this is true for every direction v, it follows that Hf (x0 ) is positive semi-
de…nite.
Hence, (0; 0) is the only critical point. Let’s …nd the Hessian matrix at these
points. Note that the function is of class C 1 , so we can apply Schwartz’s theo-
rem. We have
@2f @
(x; y) = (2x ) = 2;
@x2 @x
@2f @
(x; y) = 4y 3 = 12y 2 ;
@y 2 @y
@2f @
(x; y) = (2x) = 0:
@y@x @y
Hence,
2 0
Hf (0; 0) = ;
0 0
so that
1 0
0 = det Hf (0; 0)
0 1
2 0
= det =( 2) ( 0) 0
0 0
=( 2) ;
28
and so the roots are = 2 or = 0. Hence, at (0; 0) we cannot have a local
maximum. But it could be a local minimum. However, taking
f (0; y) = y4 ;
which has a strict maximum at y = 0. This shows that f does not admit a local
minimum or a local maximum at (0; 0).
4 Vector-Valued Functions
Since we will work with the di¤erent euclidean spaces, RM and RN , when needed
and to avoid confusion, we will denote by k kM and k kN the norms in RM and
RN , respectively. We recall that a function T : RN ! RM is linear if
T (x + y) = T (x) + T (y)
for all x; y 2 RN and
T (sx) = sT (x)
N
for all s 2 R and x 2 R .
De…nition 59 Let E RN , let f : E ! RM , and let x0 2 E be an accumu-
lation point of E. The function f is di¤erentiable at x0 if there exists a linear
function T : RN ! RM (depending on f and x0 ) such that
f (x) f (x0 ) T (x x0 )
lim = 0: (13)
x!x0 kx x0 kN
provided the limit exists. The function T , if it exists, is called the di¤erential of
f at x0 and is denoted df (x0 ) or dfx0 .
Theorem 60 Let E RN , let f : E ! RM , and let x0 2 E be an accumulation
point of E. Then f = (f1 ; : : : ; fM ) is di¤ erentiable at x0 if and only if all
its components fj , j = 1; : : : ; M , are di¤ erentiable at x0 . Moreover, dfx0 =
d (f1 )x0 ; : : : ; d (fM )x0 .
Proof. The proof relies on Exercise ?? and is left as an exercise.
We study the di¤erentiability of composite functions.
Theorem 61 (Chain Rule) Let F RM , E RN , let g : F ! E, g =
(g1 ; : : : ; gN ), and let f : E ! R. Assume that at some point y0 2 F there exist
the directional derivatives @g @gN
@v (y0 ), . . . , @v (y0 ) for some direction v and f is
1
di¤ erentiable at the point g (y0 ). Then the composite function f g admits a
directional derivative at y0 in the direction v and
X N
@ (f g) @gi
(y0 ) = dfg(y0 ) (ei ) (y0 ) :
@v i=1
@v
29
Proof. Set x0 := g (y0 ). Since f is di¤erentiable at x0 , we can write
R (x)
lim = 0: (14)
x!x0 kx x0 kN
Note that
R (x0 ) = 0:
Take x = g (y0 + tv). Then
and so
(f g) (y0 + tv) (f g) (y0 ) f (g (y0 + tv)) f (g (y0 ))
= (15)
t t
g (y0 + tv) g (y0 ) R (g (y0 + tv))
=b + :
t t
We claim that
R (g (y0 + tv))
lim = 0:
t!0 t
If for some t, we have g (y0 + tv) = g (y0 ), then
R (g (y0 + tv))
lim = 0; (18)
t!0 kg (y0 + tv) g (y0 )kN
where the limit is taken over all t such that g (y0 + tv) 6= g (y0 ). On the other
hand, since g(y0 +tv)
t
g(y0 ) @g
! @v (y0 ), and jtj
t is bounded by one, we have that
jtj g(y0 +tv) g(y0 )
t t is bounded, which, together with (17) and (14), implies
N
that
R (g (y0 + tv))
lim =0
t!0t
where the limit is taken over all t such that g (y0 + tv) 6= g (y0 ) (here we are
using the fact that if a function is bounded and another function goes to zero,
30
then their product goes to zero). Together with (16), this implies that the claim
holds.
Using the claim, it follows that
(f g) (y0 + tv) (f g) (y0 ) g (y0 + tv) g (y0 ) R (g (y0 + tv))
=b +
t t t
@g
!b (y0 ) + 0;
@v
which proves the …rst part of the statement.
The second part of the statement is left as an exercise.
Remark 62 If x0 2 E , then
X @f N
@ (f g) @gi
(y0 ) = (g (y0 )) (y0 )
@v i=1
@xi @v
@g
= rf (g (y0 )) (y0 ) :
@v
Exercise 63 Prove the second part of the theorem.
Then
@f @f
(x1 ; x2 ) = 1x2 0; (x1 ; x2 ) = x1 1 0;
@x1 @x2
@g1 @g1
(y1 ; y2 ) = 1y2 ey1 ; (y1 ; y2 ) = y1 1 0;
@y1 @y2
@g2 @g2
(y1 ; y2 ) = 1 sin (y1 y2 ) + y1 cos (y1 y2 ) (1y2 ) ; (y1 ; y2 ) = y1 cos (y1 y2 ) (y1 1) :
@y1 @y2
First method: Consider the composition
while
@ (f g) @f @g1 @f @g2
(y1 ; y2 ) = (g1 (y1 ; y2 ) ; g2 (y1 ; y2 )) (y1 ; y2 ) + (g1 (y1 ; y2 ) ; g2 (y1 ; y2 )) (y1 ; y2 )
@y2 @x1 @y2 @x2 @y2
= y1 sin (y1 y2 ) (y1 1 0) + (y1 y2 ey1 ) (y1 cos (y1 y2 ) (y1 1)) :
31
Second method: Write the explicit formula for (f g), that is,
(f g) (y1 ; y2 ) = f (g1 (y1 ; y2 ) ; g2 (y1 ; y2 ))
= (y1 y2 ey1 ) (y1 sin (y1 y2 )) 1:
Then
@ (f g) @
(y1 ; y2 ) = [(y1 y2 ey1 ) (y1 sin (y1 y2 )) 1]
@y1 @y1
= (1y2 ey1 ) (y1 sin (y1 y2 )) + (y1 y2 ey1 ) (1 sin (y1 y2 ) + y1 cos (y1 y2 ) (1y2 )) 0
and
@ (f g) @
(y1 ; y2 ) = [(y1 y2 ey1 ) (y1 sin (y1 y2 )) 1]
@y2 @y2
= (y1 1 0) (y1 sin (y1 y2 )) + (y1 y2 ey1 ) (y1 cos (y1 y2 ) (y1 1)) 0:
Next we de…ne the Jacobian of a vectorial function f : E ! RM .
De…nition 65 Given a set E RN and a function f : E ! RM , the Jacobian
matrix of f = (f1 ; : : : ; fM ) at some point x0 2 E, whenever it exists, is the
M N matrix
0 1 0 @f1 @f1 1
rf1 (x0 ) @x1 (x0 ) @xN (x0 )
B .. C B .. .. C
Jf (x0 ) := @ . A=@ . . A:
@fM @fM
rfM (x0 ) @x1 (x0 ) @xN (x0 )
It is also denoted
@ (f1 ; : : : ; fM )
(x0 ) :
@ (x1 ; : : : ; xN )
When M = N , Jf (x0 ) is an N N square matrix and its determinant is called
the Jacobian determinant of f at x0 . Thus,
@fj
det Jf (x0 ) = det (x0 ) :
@xi i;j=1;:::;N
Hence,
dfx0 (v) = d (f1 )x0 (v) ; : : : ; d (fM )x0 (v)
= Jf (x0 ) v:
32
As a corollary of Theorem 61, we have the following result.
f (x; y) = 0:
We want to solve for y, that is, we are interested in …nding a function y = g (x)
such that
f (x; g (x)) = 0:
We will see under which conditions we can do this. The result is going to be
local.
In what follows given x 2 RN and y 2 RM and f (x; y), we write
0 @f1 @f1 1
@x1 (x; y) @xN (x; y)
@f B C
(x; y) := @ ... ..
. A
@x @f @fM
M
@x1 (x; y) @xN (x; y)
and 0 1
@f1 @f1
@y1 (x; y) @yM (x; y)
@f B . .. C
(x; y) := B
@ .. .
C:
A
@y @f @fM
M
@y1 (x; y) @yM (x; y)
@f
f (a; b) = 0 and det (a; b) 6= 0:
@y
33
Then there exists BN (a; r0 ) RN and BM (b; r1 ) RM such that BN (a; r0 )
BM (b; r1 ) U and for every x 2 BN (a; r0 ) there exists a unique y 2 BM (b; r1 )
such that f (x; y) = 0. Moreover the function
g : BN (a; r0 ) ! BM (b; r1 )
x 7! y
R := [a r; a + r] [b r; b + r] U
and
@f
(x; y) > 0 for all (x; y) 2 R:
@y
Consider the function h (y) := f (a; y), y 2 [b r; b + r]. Since
@f
h0 (y) = (a; y) > 0 for all y 2 [b r; b + r] ;
@y
we have that h is strictly increasing. Using the fact that h (b) = f (a; b) = 0, it
follows that
Consider the function k1 (x) := f (x; b r), x 2 [a r; a + r]. Since k1 (a) < 0
and k1 is continuous at a, there exists 0 < 1 < r such that
34
we have that k is strictly increasing. Using the fact that k (b r) = f (x; b r) <
0 and k (b + r) = f (x; b + r) > 0, it follows that there exists a unique y 2
(b r; b + r) (depending on x) such that 0 = k (y) = f (x; y).
Thus, we have shown that for every x 2 (a ; a + ) there exists a unique
y 2 (b r; b + r) depending on x such that f (x; y) = 0. We de…ne g (x) := y.
Step 2: Continuity of g. Fix x0 2 (a ; a + ). Note that b r < g (x0 ) <
b + r. Let " > 0 be so small that
Consider the function j1 (x) := f (x; g (x0 ) "), x 2 (a ; a + ). Since j1 (x0 ) <
0 and j1 is continuous at x0 , there exists 0 < 1 < such that
35
Hence,
@f
g (x) g (x0 ) @x ( x + (1 ) x0 ; g (x) + (1 ) g (x0 ))
= @f
x x0 @y ( x + (1 ) x0 ; g (x) + (1 ) g (x0 ))
@f @f
letting x ! x0 and using the continuity of g and of @x and of @y , we get
@f
g (x) g (x0 ) @x (x0 ; g (x0 ))
lim = @f
x!x0 x x0 @y (x0 ; g (x0 ))
and so
@f
(x0 ; g (x0 ))
g 0 (x0 ) = @x
@f
:
@y (x0 ; g (x0 ))
Since the right-hand side is continuous, it follows that g 0 is continuous. Thus g
is of class C 1 .
@f
The next examples show that when det @y (a; b) = 0, then anything can
happen.
@f
Example 70 In all these examples N = M = 1 and @y (x0 ; y0 ) = 0.
f (x; y) := (xy 1) x2 + y 2 :
@f
Then f (0; 0) = 0, @y (0; 0) = 0 but
0 if x = 0;
g (x) = 1
x if x =
6 0;
which is discontinuous.
36
Example 71 Consider the function
f (x; y) := xey y:
xeg(x) g (x) = 0:
1e0 + 0 g 0 (0) = 0;
e0 g 0 (0) + g 0 (0) e0 + 0 + 0 g 00 (0) = 0:
37
and so
@f cos 1 2 1 sin 1 2
det 1; 1; = det = 1 6= 0:
@ (y; z) 2 sin 1 2 1 cos 1 2
Taking x = 1 and using the fact that g1 (1) = 1 and g2 (1) = 2, we obtain
det Jf (a) 6= 0:
Then there exists B (a; r0 ) U such that f (B (a; r0 )) is open, the function
f : B (a; r0 ) ! f (B (a; r0 ))
1
is invertible and f 2 C m (f (B (a; r0 ))). Moreover,
1 1
Jf 1 (y) = Jf f (y) :
38
Proof. We apply the implicit function theorem to the function h : U RN !
RN de…ned by
h (x; y) := f (x) y:
Let b = f (a). Then h (a; b) = 0 and
@h
det (a; b) = det Jf (a) 6= 0:
@x
Hence, by the implicit function theorem there exists B (a; r0 ) RN and B (b; r1 )
RN such that B (a; r0 ) B (b; r1 ) U RN and a function g : B (b; r1 ) !
m
B (a; r0 ) of class C such that h (g (y) ; y) = 0 for all y 2 B (b; r1 ), that is,
f (g (y)) = y
1
for all y 2 B (b; r1 ). This implies that g = f . Moreover,
1
@g @h @h
(y) = (g (y) ; y) (g (y) ; y) ;
@yk @x @yk
that is,
@f 1 1 1
(y) = Jf f (y) ek :
@yk
The next exercise shows that di¤erentiability is not enough for the inverse
function theorem.
0 if x = 0;
f1 (x; y) =
x + 2x2 sin x1 if x =
6 0;
f2 (x; y) = y:
Prove that f = (f1 ; f2 ) is di¤ erentiable in (0; 0) and Jf (0; 0) = 1. Prove that f
is not one-to-one in any neighborhood of (0; 0).
The next exercise shows that the existence of a local inverse at every point
does not imply the existence of a global inverse.
Prove that det Jf (x; y) 6= 0 for all (x; y) 2 R2 but that f is not injective.
39
Theorem 76 (Inverse Function) Let U RN be open, let f : U ! RN , and
@f
let a 2 U . Assume that f is continuous and that for all x 2 U there exist @xji ,
i; j = 1; : : : ; N , and that they are continuous at a. If
det Jf (a) 6= 0;
then there exist r0 > 0 and r1 > 0 such that the function
f : B (a; r0 ) ! f (B (a; r0 ))
Proof. Step 1: Assume that a = 0, that f (0) =0, and that Jf (0) = IN , the
identity matrix. Write
f (x) = x + h (x) :
@hj
Then h(0) =0 and Jh (0) = 0N . Since @xi are continuous at 0, there exists
r0 > 0 such that B (0; r0 ) U and
1
kJh (x)k < for all x 2 B (0; r0 ):
2
By the mean value theorem (applied how?) for all x1 ; x2 2 B (0; r0 ) and j =
1; : : : ; N ,
XN
@hj
jhj (x1 ) hj (x2 )j kx1 x2 k (zi;j ) ;
i=1
@xi
kh (x1 ) h (x2 )k
v
u !2
uX N XN
u @hj 1
kx1 x2 k t (zi;j ) < kx1 x2 k :
j=1 i=1
@xi 2
hy (x) := y h (x) :
1 1 1
khy (x)k kh (x)k + kyk kxk + kyk r0 + r0 :
2 2 2
40
Thus, we can apply the Banach …xed point theorem to hy : B (0; r0 ) ! B (0; r0 )
to conclude that hy has a unique …xed point z 2 B (0; r0 ), that is,
y h (z) = hy (z) = z:
In turn,
f (z) = z + h (z) = y:
Hence, we proved that for every y 2 B 0; 12 r0 there exists a unique z 2
B (0; r0 ) such that f (z) = y. This means that we can de…ne f 1 : B 0; 12 r0 !
B (0; r0 ).
To prove that f 1 is continuous, let y1 ; y1 2 B 0; 12 r0 and de…ne x1 :=
f (y1 ) and x2 := f 1 (y1 ). Then
1
x1 +h (x1 ) = y1 ;
x2 +h (x2 ) = y2 ;
1
and since kh (x1 ) h (x2 )k 2 kx1 x2 k, we have
1
kx1 x2 k ky1 y2 k + kh (x1 ) h (x2 )k ky1 y2 k + kx1 x2 k ;
2
and so
1
kx1 x2 k ky1 y2 k ;
2
which shows that
1 1
f (y1 ) f (y2 ) 2 ky1 y2 k :
41
Friday, February 17, 2012
Given two nonempty sets X; Y , a (binary) relation is a subset R X Y .
Usually, we associate a symbol to it, say , so that x y means that (x; y) 2 R.
A partial ordering on a nonempty set is a relation R X X, denoted ,
such that
42
where c 2 R has to be chosen appropriately. We claim that
f (x + ty0 ) L1 (x) L1 (x) f (x sy0 )
inf sup ; (19)
x2X1 ; t>0 t x2X1 ; s>0 s
or, equivalently,
f (x1 + ty0 ) L1 (x1 ) L1 (x2 ) f (x2 sy0 )
t s
for all x1 , x2 2 X1 and for all s, t > 0. This inequality may be rewritten as
which shows that the numbers in (19) are real and thus we can choose a real
number c 2 R with
f (x + ty0 ) L1 (x) L1 (x) f (x sy0 )
inf c sup :
x2X1 ; t>0 t x2X1 ; s>0 s
f (x + ty0 ) ^ (x + ty0 )
L1 (x) + tc = L
^
for all x 2 X1 and all t 2 R. Hence, we have extended L1 as a linear function L
^
to Y in such a way that L f in Y .
Step 2: Now we use Zorn’s lemma. We consider all extensions L; ~ X~ of
~ is a subspace of X containing X1 , L
(L1 ; X1 ), where X ~ : X~ ! R is linear,
~ ~ ~
L f in X, and L = L1 in X1 .
43
We introduce a partial order. We say that
(L0 ; X 0 ) 4 (L00 ; X 00 )
if X 0 X 00 and L00 = L0 in X 0 . Given a chain f(L ; X )g, de…ne
[
X1 := X
@f
kf (x) f (y)kY sup (w) kx ykX :
w2S @v Y
44
The proof is left as an exercise.
Monday, February 20, 2012
The next exercise shows that the existence of a local inverse at every point
does not imply the existence of a global inverse.
Prove that det Jf (x; y) 6= 0 for all (x; y) 2 R2 but that f is not injective.
6 Lagrange Multipliers
In Section 3 (see Theorem 56) we have seen how to …nd points of local minima
and maxima of a function f : E ! R in the interior E of E. Now we are ready
to …nd points of local minima and maxima of a function f : E ! R on the
boundary @E of E. We assume that the boundary of E has a special form, that
is, it is given by a set of the form
x 2 RN : g (x) = 0 :
F := fx 2 U : g (x) = 0g :
Let x0 2 F and assume that f attains a constrained local minimum (or maxi-
mum) at x0 . If Jg (x0 ) has maximum rank M , then there exist 1 , . . . , M 2 R
such that
rf (x0 ) = 1 rg1 (x0 ) + + M rgM (x0 ) :
45
Since Jg (x0 ) has maximum rank M , there exists a M M submatrix which
has determinant di¤erent from zero. By relabeling the coordinates, if necessary,
we may assume that x = (z; y) 2 RN M RM , x0 = (a; b) and
@g
det (a; b) 6= 0:
@y
0 = Jg (x0 ) Jk (a)
0 @g1 @g1 1 0 @k1 @k1
1
@x1 (x0 ) @xN (x0 ) @z1 (a) @zN M (a)
B .. .. CBB .. .. C
C:
=@ . . A@ . . A
@gM @gM @kN @kN
@x1 (x0 ) @xN (x0 ) @z1 (a) @zN M (a)
which implies that the vectors rgi (x0 ), i = 1; : : : ; M , belong to the kernel of
the linear transformation T : RN ! RM de…ned by
T
T (x) := (Jk (a)) x; x 2 RN :
Hence,
V := span frg1 (x0 ) ; : : : ; rgM (x0 )g ker T:
46
T
But dim V = rank Jg (x0 ) = M = N rank (Jk (a)) = dim ker T. Hence,
V = ker T:
Example 86 We want to …nd points of local minima and maxima of the func-
tion f (x; y; z) := x y + 2z over the set
E = (x; y; z) 2 R3 : x2 + y 2 + 2z 2 2 :
@E = (x; y; z) 2 R3 : x2 + y 2 + 2z 2 = 2 :
In this case the maximum rank is one, so it is enough to check that (2x; 2y; 4z) 6=
(0; 0; 0) at points in the constraint @E. But (2x; 2y; 4z) = (0; 0; 0) only at the
origin and this point does not belong to the constraint.
47
To …nd the Lagrange multipliers, let’s consider the function
F (x; y; z) := f (x; y; z) + g (x; y; z) :
Let’s …nd the critical points of F which are on @E. We have
8 @F @
>
> @x (x; y; z) = @x x y + 2z + x2 + y 2 + 2z 2 2 = 1 + 2x = 0;
< @F @
y + 2z + x2 + y 2 + 2z 2 2 = 1 + 2y = 0;
@y (x; y; z) = @y x
@F @
> @z (x; y; z) = @z x y + 2z + x2 + y 2 + 2z 2 2 = 2 + 4z = 0;
>
:
g (x; y; z) = x2 + y 2 + 2z 2 2 = 0:
If = 0, we have no solution, while if 6= 0, we get
8
>
> x = 21 ;
<
y = 21 ;
>
> z = 21 ;
: 2
x + y 2 + 2z 2 2 = 0:
1 2 1 2 1 2 2
Substituting gives 2 + 2 +2 2 2 = 0, that is, 2 1 = 0,
p p p
p1 . 2 2 2
and so = 2
Thus, the only two possible points are 2 ; 2 ; 2 and
p p p
2 2 2
2 ; 2 ; 2
. Now, since E is closed and bounded (why?), it is compact.
Since f is continuous, it follows by the Weierstrass theorempthatp there
p
exist
2 2 2
the minimum and the maximum of f over E. Hence, f 2 ; 2 ; 2 =
p p p p p p p p p
2 2 2 2 2 2 2 2 2
2 2 2 2 is the minimum and f 2 ; 2 ; 2 = 2 + 2 +2 2 is
the maximum.
Exercise 87 Given the set
E = (x; y; z) 2 R3 : x2 xy + y 2 z 2 = 1; x2 + y 2 = 1 ;
…nd the points of E having minimal distance from the origin.
7 Curves
Let I R be an interval and let ' : I ! RN be a function. As the parameter
t traverses I, ' (t) traverses a curve in RN . Rather than calling ' a curve, it
is better to regard any vector function g obtained from ' by a suitable change
of parameter as representing the same curve as '. Thus, one should de…ne a
curve as an equivalence class of equivalent parametric representations.
De…nition 88 Given two intervals I; J R and two functions ' : I ! RN
N
and : J ! R , we say that they are equivalent if there exists a continuous,
bijective function h : I ! J such that
' (t) = (h (t))
for all t 2 I. We write ' and we call ' and parametric representations
and the function h a parameter change.
48
1
Note that in view of a theorem in the Real Analysis I notes, h : J ! I is
also continuous.
is not equivalent to the previous two, since the …rst curve is simple and the
second no. Note that the range is the same, the unit circle.
49
To draw the range of the curve in the xy plane, note that x increasing means
that we are moving to the right, x decreasing to the left, y increasing means that
we are moving upwards, y decreasing downwards.
The next result shows that the class of continuous curves is somehow too
large for our intuitive idea of curve. Indeed, we construct a continuous curve
that …lls the unit square in R2 . The …rst example of this type was given by
Peano in 1890.
Proof. For every n 2 N divide the interval [0; 1] into 4n closed intervals Ik;n ,
2
k = 1; : : : ; 4n , of length 41n and the square [0; 1] into 4n closed squares Qk;n ,
n 1
k = 1; : : : ; 4 , of side length 2n . Construct a bijective correspondence between
the 4n intervals Ik;n and the 4n squares Qk;n in such a way that (see Figure ??)
By relabeling the squares, if necessary, we will assume that the square Qk;n
corresponds to the interval Ik;n . Let
F := fIk;n : n 2 N; k = 1; : : : ; 4n g ;
G := fQk;n : n 2 N; k = 1; : : : ; 4n g :
50
di¤erent intervals Ik;n , k = 1; : : : ; 4n , for all n m. Hence, it belongs to two
di¤erent sequences fJj g, Jj0 F. Since the squares Rj and Rj0 , corresponding
0
to Jj and Jj , respectively, are adjacent by property (i), it follows that
1
\ 1
\
Rj = Rj0 :
j=1 j=1
2
Thus, to every t 2 [0; 1] there corresponds a unique (x; y) 2 [0; 1] that we
denote u (t).
2
Since every (x; y) 2 [0; 1] belongs to one, two, three, or four sequences
fRj g G, there exists one, two, three, or four t 2 [0; 1] such that ' (t) = (x; y).
2
Hence, ' ([0; 1]) = [0; 1] .
To prove that ' is continuous, write ' (t) = (x (t) ; y (t)), t 2 [0; 1]. By
conditions (i) and (ii) we have that
1 1
jx (t1 ) x (t2 )j 2 ; jy (t1 ) y (t2 )j 2
2n 2n
1
for all t1 ; t2 2 [0; 1], with jt1 t2 j 4n . This proves the uniform continuity of
'.
In view of the previous result, to recover the intuitive idea of a curve, we
restrict the class of continuous curves to those with …nite length.
In what follows, given an interval I R, a partition of I is a …nite set
P := ft0 ; : : : ; tn g I, where
L ( ) := VarI ':
51
In view of the previous exercise, the de…nition makes sense, that is, the length
of the curve does not depend on the particular representation of the curve.
Friday, February 24, 2012
Peano’s example shows that continuous curves in general are not recti…able,
that is, they may have in…nite length. Next we show that C 1 or piecewise C 1
curves are recti…able.
Note that at each ti , i = 1; : : : ; n 1, the function ' has a right and a left
derivative, but they might be di¤erent.
are both C 1 functions. Note that the left derivative at 0 is '0 (0) = (1; 1)
while the right derivative is '0+ (0) = (1; 1). Hence, the curve has a corner at
the point ' (0) = (0; 0).
52
Rd
Hence, c
f (t) dt is a vector in RN .
Proof. Since products and sums of Riemann integrable functions are Riemann
integrable, it follows that the function
is Riemann integrable over [c; d]. Using the fact that the square root is a con-
tinuous function, we have that
p
kf (t)k = h (t)
The opposite inequality does not hold in general, but we have the following
lemma.
Proof. We have
Z d Z d
f (t) dt = (d c) f (t0 ) + (f (t) f (t0 )) dt;
c c
53
and so
Z d Z d
f (t) dt (d c) kf (t0 )k f (t) f (t0 ) dt
c c
Z d Z d
kf (t0 )k dt kf (t) f (t0 )k dt:
c c
On the other hand, kf (t0 )k kf (t)k kf (t) f (t0 )k, and so, upon integration,
Z d Z d Z d
kf (t0 )k dt kf (t)k dt kf (t) f (t0 )k dt:
c c c
To prove the other inequality …x " > 0. Since '0 is uniformly continuous, there
exists > 0 such that k'0 (t) '0 (s)k " for all t; ys 2 [a; b] with jt sj .
Consider a partition P := ft0 ; : : : ; tn g with ti ti 1 . Then by Lemma 104
in [ti 1 ; ti ],
Z ti
k' (ti ) ' (ti 1 )k = '0 (t) dt
ti 1
Z ti Z ti
k'0 (t)k dt 2 k'0 (t) '0 (ti 1 )k dt
ti 1 ti 1
Z ti Z ti
k'0 (t)k dt 2 " dt;
ti 1 ti 1
and so
n
X n Z
X ti Z ti
0
L( ) k' (ti ) ' (ti 1 )k k' (t)k dt 2 " dt
i=1 i=1 ti 1 ti 1
Z b
= k'0 (t)k dt 2" (b a) :
a
54
Remark 105 The previous theorem continues to hold for piecewise C 1 curves.
Exercise 106 Let f : [0; 1] ! R be the Cantor function and consider the curve
of parametric representation
Prove that L ( ) = 2.
Remark 107 Note that the derivative of the Cantor function f exists and is
zero except on a set of Lebesgue measure zero, hence (here we are using the
Lebesgue integral)
Z 1 Z 1
p
k'0 (t)k dt = 1 + 0 dt = 1 < L ( ) = Var ' = 2:
0 0
Rb
It turns out that the class of curves for which a k'0 (t)k dt = L ( ) (again,
using the Lebesgue integral instead of the Riemann integral) is given by ab-
solutely continuous curves.
X̀
k' (bk ) ' (ak )k " (23)
k=1
Remark 109 Note that since ` is arbitrary, we can also take ` = 1, namely,
replace …nite sums by series.
(i) Prove that f belongs to AC (I) if and only if for every " > 0 there exists
> 0 such that
X̀
(f (bk ) f (ak )) "
k=1
55
(ii) Assume that for every " > 0 there exists > 0 such that
X̀
(f (bk ) f (ak )) "
k=1
Part (ii) of the previous exercise shows that in De…nition 108 we cannot
remove the condition that the intervals (ak ; bk ) are pairwise disjoint.
By taking ` = 1 in De…nition 108, it follows that an absolutely continuous
function f : I ! R is uniformly continuous. The next exercise shows that the
converse is false.
(i) Prove that f may be extended uniquely to I in such a way that the extended
function is still uniformly continuous.
56
(iii) Prove that there exist A; B > 0 such that for all x 2 I,
jf (x)j A + B jxj :
b a
where we have used the fact that n 2.
Monday, February 27, 2012
Solutions, …rst midterm :(
Wednesday, February 29, 2012
Given a curve , with parametric representation ' : I ! RN , where I R
is an interval. If ' is di¤erentiable at a point t0 2 I and '0 (t0 ) 6= 0, the straight
line parametrized by
(t) := ' (t0 ) + '0 (t0 ) (t t0 ) ; t 2 R;
is called a tangent line to at the point ' (t0 ), and the vector '0 (t0 ) is called
a tangent vector to at the point ' (t0 ). Note that if is a simple arc, then
the tangent line at a point y of the range of , if it exists, is unique, while
if the curve is self-intersecting at some point y 2 , then there could be more
than one tangent line at y.
Given a function f : I ! R, where I R is an interval, the graph of f is
the range of a curve parametrized by
' (t) = (t; f (t)) :
If f is di¤erentiable at some point t0 2 I, then the tangent vector is given by
'0 (t0 ) = (1; f 0 (t0 )) :
57
De…nition 117 Given a recti…able curve , we say that is parametrized by
arclength if it admits a parametrization : [0; L ( )] ! RN such that is
0
di¤ erentiable for all but …nitely many 2 [0; L ( )] with ( ) = 1 for all but
…nitely many .
The name arclength comes from the fact that if is piecewise C 1 then for
every 1 < 2 , by Theorem 101 we have that the length of the portion of the
curve parametrized by : [ 1 ; 2 ] ! RN is given by
Z 2 Z 2
0
( ) d = 1d = 2 1:
1 1
for all but …nitely many . Since ' is piecewise C 1 and the left and right
derivatives of ' are always di¤erent from zero, it follows from (24) and the
continuity of s 1 that s 1 is piecewise C 1 . Thus, the function : [0; L ( )] !
RN , de…ned by
( ) := ' s 1 ( ) ; 2 [0; L ( )] ; (25)
58
is equivalent to '. Moreover, by the chain rule and (24), for all but …nitely
many ,
0 d '0 s 1 ( )
( ) = '0 s 1 ( ) s 1 ( )=
d k'0 (s 1 ( ))k
0
and so ( ) = 1.
Example 120 Consider the curve with parametric representations ' : [0; 2 ] !
R3 , given by
' (t) := (R cos t; R sin t; 2t) ; t 2 [0; 2 ] :
Since
'0 (t) = ( R sin t; R cot st; 2) 6= (0; 0; 0) ;
the curve is regular. Let’s parametrize it using arclength. Set
Z t Z tp
s (t) := 0
k' (r)k dr = R2 sin2 t + R2 cos2 t + 4 dr
0 0
p
= R2 + 4t:
p
In particular, s (2 ) = 2 R 2
p + 4 is the length of the curve. The inverse of s
is t (s) := pRs2 +4 , s 2 0; 2 R2 + 4 . Hence,
s s 2s
(s) := ' (t (s)) = R cos p ; R sin p ;p ;
R2 +4 R2 +4 R2 + 4
p
where s 2 0; 2 R2 + 4 , is the parametrization by arclength.
8 Curve Integrals
In this section we de…ne the integral of a function along a curve.
whenever the function t 2 [a; b] 7! f (' (t)) k'0 (t)k is Riemann integrable.
R
Note that f ds is always de…ned if f is continuous.
Next we show that the curve integral does not depend on the particular
parametric representation.
59
Proposition 122 Let be a piecewise C 1 curve and let ' : [a; b] ! RN and
: [c; d] ! RN be two parametric representations. Given a continuous function
f : E ! R, where E contains the range of ,
Z b Z d
0
f (' (t)) k'0 (t)k dt = f ( ( )) ( ) d :
a c
Proof. Exercise
Z Z
(ii) if f g, then f ds g ds,
Z Z
(iii) f ds jf j ds L ( ) max jf j, where is the range of ,
60
De…nition 127 An oriented curve is an equivalence class of parametric rep-
resentations with the same orientation.
Note that any curve gives rise to two oriented curves. Indeed, it is enough
to …x a parametric representation ' : I ! RN and considering the equivalence
class + of parametric representations with the same orientation of ' and the
equivalence class of parametric representations with the opposite orientation
of '.
Let E RN and let g : E ! RN be a continuous function. Given a piecewise
C oriented curve with parametric representation ' : [a; b] ! RN such that
1
Proof. De…ne p (t) := f (' (t)) and observe that by Theorem 61, p is piecewise
C 1 with
XN
@f
p0 (t) = (' (t)) '0i (t)
i=1
@xi
61
(i) g is a conservative vector …eld,
(ii) for every x, y 2 U and for every two piecewise C 1 oriented curves 1 and
N
2 with parametric representations '1 : [a; b] ! R and '2 : [c; d] ! RN ,
respectively, such that '1 (b) = '2 (d) = x, '1 (a) = '2 (c) = y, and
'1 ([a; b]) ; '2 ([c; d]) U ,
Z Z
g= g:
1 2
(iii) for every piecewise C 1 closed oriented curve with range contained in U ,
Z
g = 0:
Conversely assume that (ii) holds. We need to …nd a scalar potential for g. Fix
a point x0 2 U and for every x 2 U de…ne
Z
f (x) := g;
62
where in the last equality we have used the change of variable s = (t b) h. It
follows by the mean value theorem that
Z h
f (x + hei ) f (x) 1
= gi (x + sei ) ds = gi (x + sh ei ) ;
h h 0
f (x + hei ) f (x)
lim = lim gi (x + sh ei ) = gi (x) ;
h!0 h h!0
y x
g (x; y) := ; :
x2 + y 2 x2 + y 2
Note that
@ y x2 + y 2
= 2;
@y x2 + y 2 (x2 + y 2 )
@ x x2 + y 2
= 2;
@x x2 + y2 (x2 + y 2 )
63
and so g is irrotational. However, if we consider the closed curve of parametric
representation ' (t) = (cos t; sin t), t 2 [0; 2 ], we have that
Z Z 2
sin t 0 cos t 0
g= (cos t) + (sin t) dt
0 cos2 t + sin2 t cos2 t + sin2 t
Z 2
= sin2 t + cos2 t dt = 2 6= 0;
0
The problem here is the fact that the domain has a hole.
@g
and @xi is continuous in U .
Proof. Exercise.
We now turn to the proof of Theorem 135.
Proof of Theorem 135. For every x 2 U de…ne
Z
f (x) := g;
64
By the previous lemma, we have that
0 1
Z 1 XN
@f @ @
(x) = gj (x0 + t (x x0 )) (xj x0j )A dt
@xi 0 @x i j=1
0 1
Z 1 X N
@ @gj
= (x0 + t (x x0 )) t (xj x0j ) + gi (x0 + t (x x0 )) 1A dt
0 j=1
@x i
0 1
Z 1 X N
@ @gi
= (x0 + t (x x0 )) t (xj x0j ) + gi (x0 + t (x x0 )) 1A dt;
0 j=1
@x j
where we have used the fact that g is an irrotational vector …eld. De…ne
By Theorem 61,
XN
@gi
h0 (t) = (x0 + t (x x0 )) t (xj x0j ) + gi (x0 + t (x x0 )) :
j=1
@xj
h (t; 0) = ' (t) for all t 2 [a; b] ; h (t; 1) = (t) for all t 2 [a; b] ;
h (a; s) = h (b; s) for all s 2 [0; 1] :
65
Example 139 A star-shaped set is simply connected. Indeed, let E RN be
star-shaped with respect to some point x0 2 E and consider a continuous closed
curve with parametric representation ' : [a; b] ! RN such that ' ([a; b]) E.
Then the function
h (t; s) := s' (t) + (1 s) x0
is an homotopy between and the point x0 .
Remark 140 It can be shown that R2 n a point is not simply connected, while
R3 n a point is. On the other hand, R3 n a line is not simply connected.
By increasing the number n and by counting some of the balls more than once,
we can construct a partition P of [a; b],
such that ' ([ti 1 ; ti ]) Bi for all i = 1; : : : ; n. Since each ball is convex, by
Poincaré lemma g : Bi ! RN is conservative, and so there exists a function
fi : Bi ! R of class C 1 such that rfi = g in Bi . Let i be the curve with
parametric representation ' : [ti 1 ; ti ] ! RN . Assuming for a moment that is
piecewise C 1 , it follows by the fundamental theorem of calculus for curves that
Z n Z
X n Z
X n
X
g= g= rfi = (fi (' (ti )) fi (' (ti 1 ))) :
i=1 i i=1 i i=1
If instead is onlyR continuous, we could use the right-hand side of the previous
equality to de…ne g, that is, we would de…ne
Z n
X
g := (fi (' (ti )) fi (' (ti 1 ))) : (26)
i=1
66
Proof. Step 1: Let’s start by showing that (26) does not change if we add a
point to the partition. Let i 2 f1; : : : ; ng and …x c 2 (ti 1 ; ti ). Consider the
partition P 0 := P [ fcg. Since ' (c) 2 Bi , we have that
(fi (' (ti )) fi (' (c)))+(fi (' (c)) fi (' (ti 1 ))) = (fi (' (ti )) fi (' (ti 1 ))) :
R
Hence, the number g does not change if we add c to the partition P . More
R
generally, if we consider any re…nement of P , the number g does not change.
Step 2: Assume now that Q is another partition with corresponding balls B10 ,
0
. . . , Bm and with function hj : Bj0 ! R of class C 1 such that rhj = g in Bj0 .
Consider a partition P 0 = fs1 ; : : : ; s` g, which is a common re…nement of both
P and Q. For every k = 1; : : : ; `, we have that [sk 1 ; sk ] is contained in some
interval [ti 1 ; ti ] for some i = 1; : : : ; n and so ' ([sk 1 ; sk ]) Bi . Similarly,
' ([sk 1 ; sk ]) Bj0 for some j = 1; : : : ; m. Note that the set Bi \ Bj0 and on
this set we have that rfi = g = rhj . Hence, r (fi hj ) = 0 in Bi \ Bj0 and
since this set is convex (and in particular connected), it follows from Theorem
?? that fi (x) hj (x) = cij for all x 2 Bi \ Bj0 . Since ' ([sk 1 ; sk ]) Bi \ Bj0 ,
it follows that
fi (' (sk )) fi (' (sk 1 )) = (hj (' (sk )) + cij ) (hj (' (sk 1 ))
+ cij ) = hj (' (sk )) hj (' (sk 1 )) :
R
Thus, the two sums have the same elements, which shows that g as de…ned
in (26) does not depend on the particular partition P .
Remark 142 Note that if the curve is closed and ' ([a; b]) is contained in one
ball B U , then by Poincaré lemma there exists a function f : B ! R of class
C 1 such that rf = g in B, and so by considering the partition t0 := a < t1 := b,
we get Z
g = f (' (b)) f (' (a)) = 0:
Next we prove that the integral of an irrotational vector …eld over homotopic
curves does not change.
67
Proof. In view of Theorem 130, it su¢ ces to show that
Z
g=0
for every piecewise C 1 closed oriented curve with range contained in U . Fix
such a curve 1 . Since U is simply connected, we have that 2 is homotopic
to a point of U , that is, to a curve 2 with constant parametric representation.
By the previous theorem Z Z
g= g;
1 2
R
but since 2 has a constant parametric representation, we have that g = 0,
2
and the proof is completed.
Wednesday, March 07, 2012
To prove Theorem 130 we begin with a useful lemma.
Proof. If one of the U is RN , then any > 0 will do. Thus, assume that
none of the U is the entire space. Since K is compact, there exist a …nite
subfamily of fU g that still covers K, say, U1 ; : : : ; Un . Consider the closed set
Ci := RN n Ui and de…ne
n
1X
f (x) := dist (x; Ci ) :
n i=1
Let := minx2K f (x) and not that > 0 by what we proved before. Consider
a set E K with diameter less than , and let x0 2 E. Then
n n
1X 1X
f (x0 ) = dist (x0 ; Ci ) dist (x0 ; Cm ) = dist (x0 ; Cm ) ;
n i=1 n i=1
where dist (x0 ; Cm ) := max fdist (x0 ; Ci ) : i = 1; : : : ; ng. Since dist (x0 ; Cm )
, it follows that B (x0 ; ) Um . But E has diameter less than , and so
E B (x0 ; ) Um and the proof is completed.
We now turn to the proof of Theorem 143.
68
Proof of Theorem 143.. Let ' : [a; b] ! RN and : [a; b] ! RN be paramet-
ric representations of 1 and 2 . Since 1 and 2 are homotopic in U , there ex-
ists a continuous function h : [a; b] [0; 1] ! RN such that h ([a; b] [0; 1]) U ,
h (t; 0) = ' (t) for all t 2 [a; b] ; h (t; 1) = (t) for all t 2 [a; b] :
Since U is open, for every x 2 h ([a; b] [0; 1]) we may …nd B (x; rx ) U . Since
[a; b] [0; 1] is compact and h is continuous, the set h ([a; b] [0; 1]) is compact.
The family of balls fB (x; rx ) : x 2 h ([a; b] [0; 1])g covers the compact set
h ([a; b] [0; 1]). Thus, there exists a …nite number of balls B (x1 ; r1 ), . . . ,
B (xn ; rn ) such that
n
[
h ([a; b] [0; 1]) B (xi ; ri ) :
i=1
Since the function h is continuous and B (xi ; ri ) is open, we have that h 1 (B (xi ; ri ))
is relatively open in [a; b] [0; 1], that is, it is given by the intersection of an
open set Wi with [a; b] [0; 1]. Since the sets W1 ; : : : ; Wn cover the compact set
[a; b] [0; 1], by the Lebesgue’s number lemma there exists > 0 such that any
subset of [a; b] [0; 1] with diameter less than is contained in some of the Wi .
Consider a partition a = t0 < t1 < < t` = b of [a; b] and a partition
0 = s0 < s1 < < sm = 1 of [0; 1] such that each rectangle Rj;k := [tj 1 ; tj ]
[sk 1 ; sk ] has diameter less than . By the Lebesgue number lemma, there
exists i 2 f1; : : : ; ng such that Rj;k h 1 (B (xi ; ri )). Hence, h (@Rj;k )
B (xi ; ri ). If we move counterclockwise along the boundary of Rj;k starting
from the vertex (tj 1 ; sk 1 ), we have a closed curve. By composing a parametric
representation of this curve with h we obtain a closed continuous curve j;k with
range h (@Rj;k ) B (xi ; ri ). It follows by Remark 142 that
Z
g = 0:
j;k
Hence,
m Z
X̀ X
g = 0:
j=1 k=1 j;k
Now each j;k can be decomposed into four curves corresponding to the four
sides of the rectangle Rj;k . Moreover, if one of the side of Rj;k is in the interior of
[a; b] [0; 1] (except for its endpoints), then the corresponding curve will appear
twice with opposite orientation. Hence, after cancelling out all the interior
curves, we are left with four curves corresponding to the boundary of [a; b] [0; 1],
precisely, Z Z Z Z
g g+ g g = 0;
1 3 4 2
where 3 and 3 are the curves parametrized by h (a; s) and h (b; s), s 2 [0; 1],
respectively. But since h (a; s) = h (b; s) for all s 2 [0; 1], it follows that
Z Z
g g = 0;
1 2
69
and the proof is completed.
We have,
@
(yex ) = 1ex ;
@y
@ x
(e cos y) = ex 0;
@x
and so g is irrotational. Since R2 is convex, it is starshaped with respect to
every point. Hence, by the previous theorem, g is conservative. To …nd a scalar
potential, note that
@f @f
(x; y) = yex ; (x; y) = ex cos y
@x @y
and so
Z x Z x
@f
f (x; y) f (0; y) = (s; y) ds = yes ds
0 @x 0
= y (ex 1) :
Hence,
f (x; y) = f (0; y) + y (ex 1) :
Di¤ erentiating with respect to y, we get
@f @f
ex cos y = (x; y) = (0; y) + ex 1
@y @y
and so
@f
(0; y) = cos y + 1:
@y
Hence,
Z y Z y
@f
f (0; y) f (0; 0) = (0; t) dt = ( cos t + 1) dt
0 @y 0
=y sin y
and so
70
Given an open set U RN and a di¤erentiable function g : U ! RN , a
di¤erentiable function h : U ! R is called an integrating factor for g if the
function
hg = (hg1 ; : : : ; hgN )
is irrotational.
x
g (x; y) := 2 ln (xy) + 1; ;
y
Remark 149 Conservative …elds are useful is solving di¤ erential equations.
Consider the di¤ erential equation
g1 (x; y (x))
y 0 (x) = ;
g2 (x; y (x))
where
g (x; y) = (g1 (x; y) ; g2 (x; y))
is of class C 1 . If g conservative, then there exists a function f : U ! R such
that rf = g, so that
@f
@x (x; y (x))
y 0 (x) = @f ;
@y (x; y (x))
71
Consider the function
h (x) := f (x; y (x)) :
By the chain rule, we have that
@f @f
h0 (x) = (x; y (x)) + (x; y (x)) y 0 (x) = 0:
@x @y
Hence, h is constant in each connected domain of its domain. Assume that h is
de…ned in some interval I, then
for all x 2 I. Thus, we have solved implicitly our di¤ erential equation.
If g is not irrotational, but it admits an integrating factor h 6= 0, then
Theorem 150 Let U R2 be an open bounded connected set. Then the fol-
lowing are equivalent:
4. R2 n U is connected.
R
5. g = 0 for every irrotational vector …eld g : U ! R2 of class C 1 and for
every continuous closed oriented curve with range contained in U .
72
Proof. Let’s prove that (1) implies (2). Assume that there exists an invertible
function : U ! B ((0; 0) ; 1), which is continuous together with its inverse and
consider a continuous closed curve, with parametric representation ' : [a; b] !
R2 such that ' ([a; b]) U . De…ne the function h : [a; b] [0; 1] ! R2 by
1
h (t; s) = (s (' (t))) :
2r cos 2r sin r2 1
(r cos ; r sin ) := ; ; ; (1) := (0; 0; 1) :
r2 + 1 r2 + 1 r2 + 1
Next we de…ne the winding number of a closed curve around a point not on
its range. Let x0 2 R2 and let be a continuous oriented curve with range
not containing x0 and parametric representation ' : [a; b] ! R2 . Then for every
t 2 [a; b], the vector ' (t) x0 is di¤erent from zero and so we can write it in
polar coordinates, namely,
' (t) x0
= (cos (t) ; sin (t)) :
k' (t) x0 k
Now for each t, there are in…nitely many possible (t), which di¤er by multiples
of 2 . We want to prove that it is possible to chose (t) in such a way that
is a continuous function.
73
Theorem 152 Let x0 2 R2 and let be a continuous oriented curve with
range not containing x0 and parametric representation ' : [a; b] ! R2 , and let
0 be such that
' (a) x0
= (cos 0 ; sin 0 ) :
k' (a) x0 k
Then there exists a unique continuous function : [a; b] ! R such that (a) = 0
and
' (t) x0
= (cos (t) ; sin (t)) (27)
k' (t) x0 k
for all t 2 [a; b]. Moreover, the number (b) (a) does not depend on the
particular choice of 0 and on the parametric representation '.
Proof. Step 1: Assume …rst that (' (t) x0 ) (' (a) x0 ) > 0 for all t 2 [a; b].
Let
u := (cos 0 ; sin 0 ) ; v := ( sin 0 ; cos 0 )
and observe that fu; vg is an orthonormal basis in R2 . De…ne the function
' (t) x0
f (t) := :
k' (t) x0 k
Then f (t) u > 0 for all t 2 [a; b] and so we may de…ne the continuous function
: [a; b] ! R given by
f (t) v
(t) := 0 + arctan :
f (t) u
Since u v = 0,
u v
(a) = 0 + arctan
= 0:
u u
It remains to show that (27) holds. Since j (t) 0j < 2, we have that
cos ( (t) 0 ) > 0 for all t 2 [a; b], and so
1 1
cos ( (t) 0) =p 2
=r
1 + tan ( (t) 2
0) f (t) v
1+ f (t) u
f (t) u f (t) u
=q = = f (t) u;
2 2 kf (t)k
(f (t) u) + (f (t) v)
where we have used the facts that fu; vg is an orthonormal basis in R2 , that
f (t) u > 0, and that kf (t)k = 1. In turn,
74
Hence, since fu; vg is an orthonormal basis in R2 ,
Since ' is continuous on the compact set [a; b], it is uniformly continuous, and
so there exists = ("0 ) > 0 such that
"0
k' (t) ' (s)k
2
for all s; t 2 [a; b] with js tj . Consider a partition
(' (t) x0 ) (' (ti 1) x0 ) = (' (t) ' (ti 1) + ' (ti 1) x0 ) (' (ti 1) x0 )
2
= (' (t) ' (ti 1 )) (' (ti 1) x0 ) + k' (ti 1) x0 k
2
k' (t) ' (ti 1 )k k' (ti 1 ) x0 k + k' (ti 1 ) x0 k
"0
+ k' (ti 1 ) x0 k k' (ti 1 ) x0 k > 0:
2
Hence, we can apply Step 1 in the interval [t0 ; t1 ] to construct a continuous func-
tion : [t0 ; ti ] ! R such that (t0 ) = 0 and (27) holds for all t 2 [t0 ; t1 ]. Induc-
tively, assuming that we have constructed a continuous function : [t0 ; ti 1 ] !
R such that (t0 ) = 0 , we set i 1 := (ti 1 ) and apply Step 1 in the in-
terval [ti 1 ; ti ] to construct a continuous function : [ti 1 ; ti ] ! R such that
(ti 1 ) = i 1 and (27) holds for all t 2 [ti 1 ; ti ]. This gives the desired function
.
Step 3: We prove that is unique. Assume that there exist two continuous
function 1 : [a; b] ! R and 2 : [a; b] ! R such that 1 (a) = 2 (a) = 0 and
(27) holds (with 1 and 2 in place of ) for all t 2 [a; b].
Since (cos 1 (t) ; sin 1 (t)) = (cos 2 (t) ; sin 2 (t)) for all t 2 [a; b], we have
that for every t 2 [a; b] there exists kt 2 Z such that 1 (t) 2 (t) = 2 kt .
But since 1 and 2 are continuous, we have that kt has to be the same for all
t. Hence, 1 (t) 2 (t) = 2 k0 for all t 2 [a; b] and for some k0 2 Z. But,
1 (a) = 2 (a) = 0 and so k0 = 0.
,
75
Step 4: We show that the number (b) (a) does not depend on the particular
choice of 0 . Let 01 and 02 be such that
' (a) x0
= (cos 01 ; sin 01 ) = (cos 02 ; sin 02 )
k' (a) x0 k
and let 1and 2 be the corresponding functions constructed in Step 2. Then
01 02 = 2 k for some k 2 Z, and so the function
3 (t) := 1 (t) 01 + 02
This shows that the function h is admissible for ' and so by uniqueness,
(b) (a) = (h (b)) (h (a)) = (d) (c) ;
where we have used the fact that h is increasing. This completes the proof.
Remark 153 Note that if the parametrization ' is of class C 1 or piecewise C 1 ,
then the function is of class C 1 .
In particular, if is a continuous closed oriented curve with range not con-
taining x0 and with parametric representation ' : [a; b] ! R2 , then
' (a) x0 ' (b) x0
(cos (a) ; sin (a)) = = = (cos (b) ; sin (b)) ;
k' (a) x0 k k' (b) x0 k
and so the number (b) (a) is a multiple of 2 . We de…ne the winding number
of around x0 to be the integer
(b) (a)
ind (x0 ) = : (28)
2
76
Note that ind (x0 ) gives the total number of times that curve travels coun-
terclockwise around the point x0 . Let’s …nd an explicit formula for ind (x0 ) in
the case in which the curve is piecewise C 1 .
Theorem 154 Let x0 = (x0 ; y0 ) 2 R2 and let be a piecewise C 1 closed
oriented curve with range not containing x0 . Then
Z
1
ind (x0 ) = g;
2
77
Since ' : [a; b] ! R2 is piecewise C 1 , there exists a partition
is continuous. Writing,
n Z
1 X ti
ind (x) = g (t; x; y) dt;
2 i=1 ti 1
we have that the function ind is continuous in R2 n . Since it takes only integer
values, it follows that it must be constant on each connected component of the
open set R2 n .
It remains to show that it is zero in the unbounded connected component
of R2 n . Since ' : [a; b] ! R2 is piecewise C 1 , we can …nd M > 0 such that
k' (t)k M for all t 2 [a; b], and k'0 (t)k M for all but …nitely many t 2 [a; b].
Hence, for kxk > R > M , we have that
and so
(M + kxk) M + (M + kxk) M 4
jg (t; x; y)j 2 <
(kxk M) b a
provided R is su¢ ciently large. It follows that for kxk > R,
1
jind (x)j ;
2
and since ind takes only integer values, ind (x) = 0.
Friday, March 23, 2012
Another important application of Theorem 154 is the following.
Theorem 156 Let U R2 be an open set and let 1 and 2 be two continuous,
closed, oriented curves that are homotopic in U . Then
ind 1
(x) = ind 2
(x)
78
Proof. Fix x0 2 R2 n U and let 1 and 2 be as in the statement.RSince the R the
vector …eld (29) is irrotational, it follows by Theorem 143, that g= g,
1 2
and so ind 1 (x0 ) = ind 2 (x0 ). On the other hand, if U is simply connected,
then every continuous closed oriented curve 1 is homotopic to a point.
R But
for a curve 2 with constant parametric representation we have that g = 0,
2
and so by the …rst part of the theorem, ind 1 (x0 ) = 0.
Remark 157 The previous theorem proves the implication (2)=)(3) in Theo-
rem 150.
To prove that (3)=)(4) in Theorem 150 we need the following result. Given
n closed continuous oriented curves 1 , . . . , N , the family := f 1 ; : : : ; N g
is called a cycle. The range of is given by the union of the ranges of 1 ,
. . . , N . Given a point x 2 R2 not contained in the range of , we de…ne the
winding number of around x to be the integer
n
X
ind (x) := ind k
(x) :
k=1
1 if x 2 K;
ind (x) =
0 if x 2 R2 n U:
Proof. Exercise.
We prove that (3)=)(4) in Theorem 150.
Proof of Theorem 150, continued. Assume that condition (3) in Theorem
150 is satis…ed but that R2 n U is not connected. Since R2 n U is closed, its
connected components are also closed. Hence, we can …nd two disjoint nonempty
closed sets C and K such that
R2 n U = C [ K:
Moreover, since U is bounded, we may assume that there is only one unbounded
connected component contained in C, so that K is compact.
Let V := R2 n C. Then V is open and contains K. By the previous theorem
there exists a cycle with range contained in V n K such that
1 if x 2 K;
ind (x) =
0 if x 2 R2 n V:
79
range. By Theorem 155, the function ind is constant on connected components
of R2 n . But R2 nU R2 n and each connected component of the smaller set is
contained in a connected component of the larger set. Hence, the function ind
is constant on connected components of R2 n U . Since R2 n U is connected and
unbounded, it follows that it must be contained in the unbounded connected
component of R2 n , hence by Theorem 155, ind (x) = 0 for all x 2 R2 n U .
This proves that (3) holds.
Exercise 159 Prove that when U is open and connected (but not necessarily
bounded) then condition (3) in Theorem 150 is equivalent to the fact that S2 n U
is connected.
9 Integration
The next theorem is very important in exercises. It allows to calculate triple,
double, etc.. integrals by integrating one variable at a time.
80
we have
m X̀
X
L (f; P) L (f; P 0 ) = measN Si measM Tj inf inf f (x; y)
x2Si y2Tj
i=1 j=1
m
X X̀
measN Si inf measM Tj inf f (x; y)
x2Si y2Tj
i=1 j=1
Xm Z Z Z
measN Si inf f (x; y) dy f (x; y) dy dx
x2Si T S T
i=1
Z Z m
X Z
f (x; y) dy dx measN Si sup f (x; y) dy
S T i=1 x2Si T
m
X X̀
measN Si sup measM Tj sup f (x; y)
i=1 x2Si j=1 y2Tj
m X̀
X
measN Si measM Tj sup sup f (x; y)
i=1 j=1 x2Si y2Tj
= U (f; P 0 ) U (f; Q) ;
81
Remark 161 Note that if in Theorem 160 we remove the hypothesis that for
every x 2 S, the function y 2 T 7! f (x; y) is Riemann integrable, we can still
R R
prove that the functions x 2 S 7! T f (x; y) dy and x 2 S 7! T f (x; y) dy are
Riemann integrable with
Z Z Z !
f (x; y) d (x; y) = f (x; y) dy dx
S T S T
Z Z
= f (x; y) dy dx:
S T
The proof is similar to the one of Theorem 160 and we leave it as an exercise.
In turn, !
Z Z Z
f (x; y) dy f (x; y) dy dx = 0:
S T T
R R
Since T f (x; y) dy T
f (x; y) dy 0, it follows from Exercise ?? that there
exists a set E S of Lebesgue measure zero such that
Z Z
f (x; y) dy f (x; y) dy = 0
T T
The following example shows that without assuming that f is Riemann in-
tegrable, the previous theorem fails.
Let E be the set of all (x; y) 2 [0; 1] [0; 1] for which there exist p 2 prime and
m n
m; n 2 N such that (x; y) = p; p . Using the density of the rationals and of
the irrationals, it can be shown that both E and ([0; 1] [0; 1]) n E are dense in
[0; 1] [0; 1]. Hence, the set of discontinuity points of f is [0; 1] [0; 1]. Thus,
f is not Riemann integrable. On the other hand, if we …x x 2 [0; 1] and we
consider the function f (x; ), then we have the following two cases. If x = m p
for some p 2 prime and some m 2 N, then
( n o
1 if y 2 p1 ; : : : ; p p 1 ; pp ;
f (x; y) =
0 otherwise.
82
Thus f (x; ) is only discontinuous at a …nite number of points and so it is
Riemann integrable in [0; 1] with
Z 1
f (x; y) dy = 0:
0
In the second case, x cannot be written in the form mp for some p 2 prime
and some m 2 N. In this case f (x; y) = 0 for all y 2 [0; 1] and so again
R1
0
f (x; y) dy = 0, which shows that
Z 1 Z 1 Z 1
f (x; y) dy dy = 0 dy = 0
0 0 0
and similarly, Z Z Z
1 1 1
f (x; y) dx dx =
0 dx = 0:
0 0 0
R
So the iterated integrals exist and are equal, but the integral [0;1] [0;1]
f (x; y) d (x; y)
does not exist.
The following example shows that the fact that f : S T ! R be Riemann
integrable does not imply that for every x 2 S, the function y 2 T 7! f (x; y)
is Riemann integrable.
Example 163 Let f : [0; 1] [0; 1] ! R be de…ned by
1 if y 2 [0; 1] \ Q and x = 21 ;
f (x; y) :=
0 otherwise.
The function f is discontinuous only on the segment 12 [0; 1], which has
Lebesgue measure zero. Hence, f is Riemann integrable in [0; 1] [0; 1]. On
the other hand, if we …x x = 12 and we consider the function g (y) = f 12 ; y ,
y 2 [0; 1], we have that g is discontinuous at every y 2 [0; 1], and so g is not
Riemann integrable in [0; 1].
R
Exercise 164 Calculate [0;1] [0;1] f (x; y) d (x; y) in two di¤ erent ways, where
f (x; y) := x sin (x + y) :
10 Peano–Jordan Measure
Given a bounded set E RN , let R be a rectangle containing E. We say that
E is Peano–Jordan measurable if the function E is Riemann integrable over R.
In this case the Peano–Jordan measure of E is given by
Z
meas E := E (x) dx:
R
83
Exercise 165 Prove that the previous de…nition does not depend on the choice
of the rectangle R containing E.
Remark 166 Note that a rectangle is Peano–Jordan measurable and its Peano–
Jordan measure coincides with its elementary measure (see Remark ??).
meas E meas F:
n
X
meas E meas Ei :
i=1
Remark 173 In view of the previous exercise and of Exercise 169, it follows
that every pluri-rectangle is Peano–Jordan measurable.
84
Theorem 174 Given a bounded set E RN ,
Z
measi E = E (x) dx; (31)
R
Z
measo E = E (x) dx: (32)
R
Proof. Exercise.
This implies that E has Lebesgue measure zero. However, the opposite is not
true. For example the set E := [0; 1] \ Q has Lebesgue measure zero (since it
is countable), but it is not Peano Jordan measurable and its outer measure is
actually one. Indeed, its characteristic function is the Dirichlet function.
85
Note that P2 n P1 is still a pluri-rectangle (exercise) and since
E P2 ; P1 E ;
we have that
@E = E n E P2 n P1 :
Hence,
Let P1 be the pluri-rectangle given by the union of all the rectangles Ri that
are contained in E, so that P1 E. Let P2 := P [ P1 . We claim that
E P2 :
86
Theorem 178 Let R RN be a rectangle and let f : R ! [0; 1) be a bounded
function. Then f is Riemann integrable over R if and only if the set
Proof. Step 1: Assume that f is Riemann integrable over R. Given " > 0, by
Theorem ?? there exists a partition P " of R such that
Ti := Ri 0; inf f ; Ui := Ri 0; sup f
Ri Ri
and
n
[ n
[
P1 := Ti ; P2 := Ui :
i=1 i=1
It follows from Theorem 174 that the set Sf is Peano–Jordan measurable with
Z
measN +1 Sf = f (x) dx:
R
87
is Riemann integrable. Hence, by Theorem 160, the function
Z Z
x 2 R 7! Sf (x; y) dy = [0;f (x)] (y) dy
[0;M ] [0;M ]
Z f (x)
= 1 dy = f (x)
0
Remark 179 With a similar proof one can show that if R RN is a rectangle
and f : R ! R is a bounded function, with f (x) c for all x 2 R. Then f is
Riemann integrable over R if and only if the set
Gr f := f(x; y) 2 R R : y = f (x)g
@ (E n F ) @E [ @F
88
and let f : E ! R be a bounded continuous function. Then f is Riemann
integrable over E and
Z Z Z !
(x)
f (x; y) d (x; y) = f (x; y) dy dx:
E R (x)
f (x; y) if (x; y) 2 E;
g (x; y) :=
0 if (x; y) 2 (R [a; b]) n E:
We need to show that g is Riemann integrable over R [a; b]. Hence, we need
to look at the set of discontinuity points of g. Since g is continuous in E, we
have that the discontinuity points of g are on the boundary of E. Thus, we
need to show that @E has Lebesgue measure zero in RN +1 . We will show more,
namely that it has Peano–Jordan measure zero in RN +1 . Let c 2 R be such
that (x) c for all x 2 R. Then
Remark 183 If and are continuous, then the set of discontinuity points of
g is given by the union of the graphs of and , but if and are discontinuous,
then the set of discontinuity points of g is larger (why?).
89
Given a rectangle R RN , a Peano–Jordan measurable set E R, and a
Riemann integrable function f : R ! R, we de…ne the integral of f over E as
Z Z
f (x) dx := E (x) f (x) dx:
E R
Note that this integral is well-de…ned since the product of Riemann integrable
functions is still Riemann integrable.
(iii) Prove that if E is connected and has positive measure, then there exists
c 2 E such that Z
1
f (x) dx = f (c) :
meas E E
where
E := (x; y) 2 R2 : y x; x2 + y 2 1; x 0; y 0 :
We can rewrite E as follows,
( p )
2 p
E= (x; y) 2 R2 : 0 y ;y x 1 y2
2
and since the function f (x; y) := x (1p y) is bounded and continuous in E and
the functions (y) := y and (y) := 1 y 2 are continuous, we can apply the
previous corollary to conclude that
0 p 1
ZZ Z p22 Z p1 y2 ! Z p22 2 x= 1 y2
x
x (1 y) dxdy = x (1 y) dx dy = (1 y) @ A dy
E 0 y 0 2 x=y
p
Z 2
2 1 y2 y2
= (1 y) dy
0 2 2
1p 1
= 2 :
6 16
90
Exercise 186 Calculate the integral
ZZZ
(x + z) dxdydz;
E
where
E := (x; y; z) 2 R3 : x + y + z 1; x 0; y 0; z 0 :
11 Improper Integrals
In this section we study integrability for functions that are either unbounded
or are de…ned on unbounded sets. If f : [a; 1) ! R is a continuous function,
we have seen that the improper or generalized Riemann integral of f over R is
given by Z Z
1 r
f (x) dx := lim f (x) dx;
a r!1 a
provided the limit exists. If now f : E ! R is continuous, where, say
E := (x; y) 2 R2 : x 0; y 0
we would be tempted to follow a similar procedure, that is, to de…ne
ZZ ZZ
f (x; y) dxdy := lim f (x; y) dxdy;
E r!1 [0;r] [0;r]
or ZZ ZZ
f (x; y) dxdy := lim f (x; y) dxdy:
E r!1 E\B((0;0);r)
Unfortunately, in general, it could happen than one limit exist but the other
does not.
Example 187 Buck. Consider the function
f (x; y) := sin x2 + y 2
de…ned in the set
E := (x; y) 2 R2 : x 0; y 0 :
By Theorem 160,
ZZ ZZ
sin x2 + y 2 dxdy = sin x2 cos y 2 + cos x2 sin y 2 dxdy
[0;r] [0;r] [0;r] [0;r]
Z r Z r
= sin x2 cos y 2 + cos x2 sin y 2 dx dy
0 0
Z r Z r
2
=2 sin x dx cos y 2 dy
0 0
Z 1 Z 1
!2 sin x2 dx cos y 2 dy =
0 0 4
91
as r ! 1 (Fresnel integrals). On the other hand, using polar coordinates,
ZZ Z 2
Z r
2 2 2
sin x + y dxdy = sin d d
E\B((0;0);r) 0 0
=r
1 2
= cos
2 2 =0
1 1
= cos r2
2 2 2
for every exhausting sequence fEn g of E such that f is Riemann integrable over
each En . The number
R ` is called the improper Riemann integral of f over E
and is denoted E f (x) dx.
Remark 189 Note that we do not test the existence of the limit along every
exhausting sequence but only over those exhausting sequences fEn g for which f
is Riemann integrable over each En .
92
Proof. Exercise.
The de…nition of improper Riemann integral is not practical, since one needs
to check all exhausting sequences of E. The next result shows that for nonneg-
ative functions (and similarly for nonpositive) it is enough to consider only one
exhausting sequence.
Theorem 191 Let E RN and let f : E ! [0; 1). If there exists one ex-
hausting sequence fEn g such that f is Riemann integrable over each En and the
limit Z
lim f (x) dx
n!1 En
where in the last inequality we have used the fact that f 0.n Letting k !o 1
R
and using again the fact that f 0 (so that the sequence Fk
f (x) dx is
increasing, we obtain
Z Z
lim f (x) dx lim f (x) dx:
k!1 Fk n!1 En
We calculate ZZ
x2 y 2
e dxdy:
R2
2 2
Consider the exhausting sequence fB ((0; 0) ; n)g. Since f (x; y) = e x y is
continuous and bounded over each ball, it is Riemann integrable over each ball.
93
Using polar coordinates, we have
ZZ Z n Z 2 Z n
2 2
r2 r2
e x y dxdy = e r d dr = 2 e r dr
B((0;0);n) 0 0 0
r=n
1 r2 1 n2 1
=2 e =2 e + :
2 r=0 2 2
Hence,
ZZ ZZ
x2 y 2 x2 y 2
e dxdy = lim e dxdy
R2 n!1 B((0;0);n)
1 n2 1
= lim 2 e + = :
n!1 2 2
and so
ZZ ZZ
x2 y 2 x2 y 2
= e dxdy = lim e dxdy
R2 n!1 [ n;n] [ n;n]
Z n 2 Z 1 2
t2 t2
= lim e dt = e dt ;
n!1 n 1
(i) If
jf (x)j g (x)
R
and g is Riemann integrable in the improper sense over E with E g (x) dx <
1, then f and jf j are Riemann integrable in the improper sense over E
and Z Z Z
f (x) dx jf (x)j dx g (x) dx:
E E E
94
(ii) If
f (x) g (x) 0
R
and g is Riemann integrable in the improper sense over E with E g (x) dx =
1, then f is Riemann integrable in the improper sense over E and
Z
f (x) dx = 1:
E
Proof. (i) Let fEn g be an exhausting sequence of nE such that g ois Riemann
R
integrable over each En . Since jf j 0, the sequence En jf (x)j dx is increas-
ing, and so there exists the limit
Z
lim jf (x)j dx = ` 2 [0; 1] :
n!1 En
R R
On the other hand, since En jf (x)j dx En
g (x) dx by Proposition ??, it
follows that
Z Z Z
lim jf (x)j dx lim g (x) dx = g (x) dx < 1:
n!1 En n!1 En E
Letting n ! 1, we have that the right-hand side has a limit, and so there exists
Z Z Z
lim f (x) dx = lim jf (x)j dx lim (jf (x)j f (x)) dx
n!1 E n!1 E n!1 E
n
Z n
Z n
95
This shows that for every exhausting sequence fEn g of E such that f is Riemann
integrable over each En , there exists the limit
Z Z Z
lim f (x) dx = jf (x)j dx (jf (x)j f (x)) dx:
n!1 En E E
Since the limit is the same for every exhausting sequence fEn g, we have that f
is Riemann integrable in the improper sense over E.
(ii) Let fEn g be an exhausting sequence of E such that g is Riemann inte-
grable over each En . By hypothesis f is Riemann integrable over each En , and
so by Proposition ??, it follows that
Z Z
f (x) dx g (x) dx:
En En
Letting n ! 1, we get
Z Z Z
lim f (x) dx lim g (x) dx = g (x) dx = 1:
n!1 En n!1 En E
It follows from
R Theorem 191 that f is Riemann integrable in the improper sense
over E with E f (x) dx = 1.
Prove that f is Riemann integrable in the improper sense over E and that the
improper Riemann integral is …nite.
x2 y2
f (x; y) := 2
(x2 + y 2 )
de…ned in the open square Q of vertices (0; 0), (0; 1), (1; 1), and (1; 0). Prove
that f is not Riemann integrable in the improper sense over Q.
96
The next result shows that if f is Riemann integrable in the improper sense
over E with …nite improper Riemann integral, then the same is true for jf j.
This is in sharp contrast with the 1-dimensional case. Hence, the de…nition of
improper integral in RN is not an extension of the one given for N = 1.
Let G be given by the union of all rectangles Ri such that inf x2Ri F (x) f + (x) >
0. Then G is a pluri-rectangle, and so it is Peano–Jordan measurable. Moreover,
if x 2 G, then x 2 Ri for some i. In turn, F (x) f + (x) > 0, which implies that
x 2 F . Thus, G F . Moreover, f (x) = f + (x) for all x 2 G. Hence,
Z Z
f (x) dx = f + (x) dx
G G
Z
s 1
L F f +; P > = jf (x)j dx:
3 3 F
97
for every Peano–Jordan measurable set F E such that f is Riemann integrable
over F . Assume by contradiction that this is not the case and let fEn g be
an exhausting sequence of E such that f is Riemann integrable over each En .
De…ne G1 := E1 . Since f is Riemann integrable
R over G1 , we have that jf j is also
Riemann integrable over G1 , so that G1 jf (x)j dx < 1. By the contradiction
hypothesis, there exists a Peano–Jordan measurable set F1 E such that f is
Riemann integrable over F1 and
Z Z
f (x) dx 1+ jf (x)j dx;
F1 G1
R
since otherwise we could take L := 1 + G1 jf (x)j dx. Inductively, for every
n 2 we can …nd a Peano–Jordan measurable set Fn E such that f is
Riemann integrable over Fn and
Z Z
f (x) dx n+ jf (x)j dx;
Fn Gn
where Gn := En [ F1 [ [ Fn 1 .
Note that the sequence Hn := Fn [ Gn is an admissible exhausting sequence
of E. Moreover, using the fact that Hn n Fn Gn , we have that
Z Z Z
f (x) dx = f (x) dx + f (x) dx
Hn Fn Hn nFn
Z Z
f (x) dx f (x) dx
Fn Hn nFn
Z Z
f (x) dx jf (x)j dx
Fn Hn nFn
Z Z
f (x) dx jf (x)j dx n;
Fn Gn
98
Since jf j 0, letting n ! 1 and using Theorem 191, we get that
Z
jf (x)j dx 3L:
E
Hence,
Z 1 Z n n
X
sin x sin x 2
dx = lim dx lim = 1:
x n!1 x n!1 k
k=2
99
We have
Z Z (2n 1) X 2n Z (2k+1)
sin x sin x sin x
dx = dx + dx
En x x 2k x
k=n
Z (2n 1) 2n
X Z (2k+1)
sin x 1
dx + sin x dx
x (2k + 1) 2k
k=n
Z (2n 1) X 2n
sin x 2
= dx +
x (2k + 1)
k=n
Z (2n 1) X 2n
sin x 2
dx +
x (4n + 1)
k=n
Z (2n 1)
sin x 2n + 2
= dx + :
x +4 n
which shows that f cannot be Riemann integrable in the improper sense accord-
ing to De…nition 188, since along two exhausting sequences we obtain di¤ erent
limits.
12 Change of Variables
We show that Lipschitz transformations map sets of measure zero into sets of
measure zero.
Proof. Exercise.
Next we show that Peano–Jordan measurability is preserved under C 1 trans-
formations.
Proof. Exercise.
100
Remark 201 In particular, it g is the parametrization of a curve, then N = 1
so that by the previous theorem it follows that the range of a curve has measure
zero for all M > 1. Hence, sets of the type
E1 := (x; y) 2 R2 : 4x2 + 9y 2 16 ;
2 2 2
E2 := (x; y) 2 R : x + y 1 ;
Proof. Exercise.
Using the previous theorems, we can show the change of variables theorem.
101
where
F := (x; y) 2 R2 : 0 < x < y < 2x; 1 < xy < 2 :
Consider the change of variables u = xy and v = xy . Let’s solve for x and y.
p p
We have x = uv and y = uv. De…ne g : U ! R2 as follows
r
u p
g (u; v) := ; uv ;
v
where
U := (u; v) 2 R2 : u > 0; v > 0 :
Then g is one-to-one and its inverse is given by g 1 (x; y) := xy; xy . (Note
that in general it is much simpler to check that g is one-to-one than …nding the
inverse). Moreover,
@g1 @g1
@u (u; v) @v (u; v)
det Jg (u; v) = det @g2 @g2
@u (u; v) @v (u; v)
p
u
!
p1 3=2 1
2p uv p2v
= det = :
pv pu 2v
2 u 2 v
Set E = (1; 2) (1; 2) and note g (E) = F . By the change of variables theorem,
ZZ ZZ
(x + y) dxdy = (g1 (u; v) + g2 (u; v)) jdet Jg (u; v)j dudv
F E
ZZ r
u p 1
= + uv dudv
E v 2v
ZZ p p
1 u u
= 3=2
+ dudv:
2 E v v
It follows that
ZZ Z Z p p Z p v=2
!
2 2 2 p
1 u u 1 u
(x + y) dxdy = + dv du = 2 + u log v du
F 2 1 1 v 3=2 v 2 1 v 1=2 v=1
Z 2 p
1 p 1
= u ln 2 du = (ln 2) 2 2 1 :
2 1 3
102
for all i = 1; : : : ; N , with i 6= m, and all x 2 U , so that
@gm
det Jg (x) = (x) : (34)
@xm
A function g : U ! RN is called a ‡ip if there exist m; n 2 f1; : : : ; N g such that
for all i = 1; : : : ; N , with i 6= m; n, and all x 2 U ,
gi (x) = xi ;
while
gm (x) = xn ; gn (x) = xm :
g = g1 gn in B (x0 ; r)
0
Proof. Assume that g 0 (the case g 0 0 is similar). Since [a; b] is compact
and g : [a; b] ! R is continuous, g 0 is uniformly continuous, and so given " > 0,
0
103
Moreover, since g ([xi 1 ; xi ]) = Ji ,
Hence,
n
X n
X
(yi yi 1 ) sup f (y) (xi xi 1) sup f (g (x)) sup g 0 (z) :
i=1 y2Ji i=1 x2[xi 1 ;xi ] z2[xi 1 ;xi ]
(37)
Given x; z 2 [xi 1 ; xi ], we have that
where we have used (35), and where M 0 is a bound for jf j. It follows that
and so
n
X
(xi xi 1) sup f (g (x)) sup g 0 (z)
i=1 x2[xi 1 ;xi ] z2[xi 1 ;xi ]
n
X n
X
(xi xi 1) sup f (g (x)) g 0 (x) + (xi xi 1 ) "M (39)
i=1 x2[xi 1 ;xi ] i=1
= U ((f g) g 0 ; P" ) + "M (b a) U ((f 0
g) g ; P ) + "M (b a) :
To prove the opposite inequality, consider the function f1 (x) := f (g (x)) g 0 (x),
let g ([a; b]) = [c; d] and apply what we just proved to the nonnegative bounded
104
1
function f1 , using g in place of g, to get
Z b Z Z d
0 1 1 0
f (g (x)) g (x) dx = f1 (x) dx f1 g (y) g (y) dy
a g 1 ([c;d]) c
Z Z
1 1 0
= f g g (y) g0 g 1
(y) g (y) dy = f (y) dy;
g([a;b]) g([a;b])
the section
Cx0 = fxN 2 R : (x0 ; xN ) 2 Cg
is closed. Indeed, R n Cx0 = fxN 2 R : (x0 ; xN ) 2= Cg = RN n C x0 , which is
open.
We now turn to the proof of Theorem 204.
Proof of Theorem 204. Step 1: Let’s prove that the theorem holds if U is
open and bounded and g is elementary and bounded. Without loss of generality,
we may assume that the only component that is not the identity is the last one,
so that for all i = 1; : : : ; N 1 and all x 2 U ,
gi (x) = xi :
f (y) if y 2 R;
h (y) := (40)
0 if y 2 R n g (E) :
Let
R = I1 IN :
105
Write y = (y0 ; yN ) 2 RN 1 R and let R = R0 IN , where R0 := I1 IN 1.
By Theorem 160 and Remark 161,
Z Z Z
h (y) dy = h (y0 ; yN ) dyN dy0 : (42)
R R0 IN
Then [
Ky0 Vy0 = Jn IN ;
n
S`
and so by compactness there exists ` 2 N such that Ky0 n=1 Jn IN .
Using the fact that h (y0 ; yN ) = 0 for all yN 2
= (g (E))y0 Ky0 , we have that
(see Proposition 242 in the Real Analysis I notes)
Z X̀ Z
0
h (y ; yN ) dyN = h (y0 ; yN ) dyN : (43)
IN n=1 Jn
X̀ Z X̀ Z @gN 0
h (y0 ; yN ) dyN = h (y0 ; gN (y0 ; xN ))
(y ; xN ) dxN :
n=1 Jn
1
n=1 k (Jn )
@xN
(44)
Since gi (y0 ; xN ) = yi for i 6= N , k 1 (Jn ) Uy0 , and h (y0 ; gN (y0 ; xN )) = 0
for xN 2
= Ey 0 Uy0 , if we consider an interval R1 := R0 J containing U and
de…ne (
@gN
f (g (x)) @x (x) if x 2 E;
h1 (x) := N
0 if y 2 R1 n E;
then by (40) and Proposition 242 in the Real Analysis I notes
X̀ Z @gN 0 X̀ Z
0 0
h (y ; gN (y ; xN )) (y ; xN ) dxN = h1 (y0 ; xN ) dxN
n=1 k 1 (J )
n
@xN n=1 k
1 (J
n)
(45)
Z
= h1 (y0 ; xN ) dxN :
J
106
Hence, by (41)-(45), Theorem 160 and Remark 161,
Z Z Z
f (y) dy = h1 (y0 ; xN ) dxN dy0
g(E) R0 J
Z
= h1 (x) dx
R1
Z
= f (g (x)) jdet Jg (x)j dx:
E
where i = 1; 2.
Consider the function g := g1 g2 : U2 ! RN , let E RN be a Peano–
Jordan measurable set with E U2 and let f : g (E) ! R be Riemann inte-
grable. Then Z Z
f (y) dy = f (y) dy;
g(E) g1 (g2 (E))
and so, by applying the change of variable y = g1 (z) (with E1 := g2 (E) and
f1 := f ), we get
Z Z
f (y) dy = f (g1 (z)) jdet Jg1 (z)j dz:
g1 (g2 (E)) g2 (E)
On the other hand, applying the change of variable z = g2 (x) (with f2 (z) :=
f (g1 (z)) jdet Jg1 (z)j and E2 := E), we get
Z Z
f (g1 (z)) jdet Jg1 (z)j dz = f (g1 (g2 (x))) jdet Jg1 (g2 (x))j jdet Jg2 (x)j dx
g2 (E) E
Z
= f (g (x)) jdet Jg (x)j dx;
E
107
by Corollary 67, so that
Then
Z Z m Z
X
f (g (x)) jdet Jg (x)j dx = h2 (x) dx = h2 (x) dx
E P j=1 Rj
X Z
= f (g (x)) jdet Jg (x)j dx;
j such that Rj \E6=; Rj \E
where in the last R equality we have used the fact that if Rj does not intersect
the set E, then Rj h2 (x) dx = 0.
Fix j such that Rj \ E 6= ;. Then by (46), there exists i 2 f1; : : : ; mg such
that Rj intersects B xi ; 12 rxi . On the other hand, the diameter of Rj is less
1
than 2 rxi , and so Rj is contained in B (xi ; rxi ).
By construction g restricted to B (xi ; rxi ) is given by the union of …nitely
many elementary di¤eomorphisms and ‡ips. Hence, by applying Steps 1, 2, and
3 (with U replaced by B (xi ; rxi ) and with E replaced by Rj \ E), we conclude
that Z Z
f (g (x)) jdet Jg (x)j dx = f (y) dy;
Rj \E g(Rj \E)
108
where we have used the fact that, since g is one-to-one, the sets g (Rj \ E) are
disjoint, so that
0 1
[ [
g (Rj \ E) = g @ (Rj \ E)A
j such that Rj \E6=; j such that Rj \E6=;
0 1
m
[
= g@ (Rj \ E)A = g (P \ E) = g (E) :
j=1
13 Spherical Coordinates in RN
The most important applications of the previous theorem is spherical coor-
dinates in RN . We introduce them by induction. For N = 2 we use polar
coordinates. More precisely, we de…ne
Note that partial derivatives of any order exist in R2 and are continuous. The
Jacobian of g is given by
@g1 @g1
@r (r; ) @ (r; )
det Jg (r; ) = det @g2 @g2
@r (r; ) @ (r; )
cos r sin
= det = r cos2 + r sin2 = r;
sin r cos
109
for some large R > 0. Moreover g (F ) is a bounded subset of the x-axis. Hence,
meas (F ) = meas (g (F )) = 0.
where
F := (x; y) 2 R2 : 1 < x2 + y 2 < 4; x > 0; y > 0 :
1 x1
1 := cos ; 0< 1 < :
r
Then x1 = r cos 1 . Let ( ; 2 ; : : : ; N 1) be the spherical coordinates of x0 =
(x2 ; : : : ; xN ). Then
q p q
2
= kx0 kN 1 = kxk x21 = r2 r2 cos2 1 =r sin2 1 = r jsin 1j = r sin 1:
x1 = r cos 1;
x2 = r sin 1 cos 2;
..
.
xN 1 = r sin 1 sin 2 sin N 2 cos N 1;
xN = r sin 1 sin 2 sin N 2 sin N 1:
jdet Jg (r; 1; : : : ; N 1 )j = rN 1
sinN 2
1 sinN 3
2 sin N 2:
110
Then g : U ! V is invertible and the inverse is given by
r = kxk ;
x1
1 = arccot p ;
x22 + + x2N
..
.
xN 2
N 2 = arccot q ;
x2N 1+ x2N xN 1
xN
N 1 = 2 arccot q :
x2N 2
1 + xN xN 1
N
meas (B (0; R)) = RN = meas (B (0; 1)) RN : (47)
N
To …nd meas (B (0; 1)), we have the following:
111
We now integrate into two di¤erent ways.
Z Z Z !
1
y y
e dxdy = e dx dy
D 0 B (0;y 1=2 )
Z Z !
1
y
= e 1 dx dy
0 B (0;y 1=2 )
Z 1
y
= e measN B 0; y 1=2 dy
0
Z 1 Z 1
y N=2 y N=2
= e y measN B (0; 1) dy = measN B (0; 1) e y dy
0 0
= (1 + N=2) measN B (0; 1) :
where Z
t2
I := e dt:
R
This shows that
(1 + N=2) measN B (0; 1) = I N :
Now for N = 2, we know that meas2 B (0; 1) = , while (1 + 2=2) = (2) = 1.
Hence,
= I 2;
1=2
which shows that I = . This completes the proof.
112
1
coordinates, we have that for all n > R,
Z Z R Z R
1 rN 1
a dx = N dr = N rN a 1
dr
En kx x0 k 1=n ra 1=n
8
< [log r]r=R if a = N;
= N h N ar=1=n
ir=R
: rN a if a 6= N
r=1=n
log R + log n if a = N;
= N 1 N a 1
N a R nN a if a =
6 N:
113
Proof. Since f is continuous and jf (x)j g (x) := kx Cx0 ka , we have that f
is bounded whenever g is bounded. Hence, f is Riemann integrable over every
subset F E over which g is Riemann integrable. Thus, we are in a position
to apply the Theorem 193.
Note that if in part (i) the set E does not contain a punctured ball, then the
integral could be …nite.
Prove that f is Riemann integrable in the improper sense over E and that the
improper Riemann integral is …nite.
114
14 Di¤erential Surfaces
De…nition 214 Given 1 k < N , a nonempty set M RN is called a k-
dimensional di¤erential surface or manifold if for every x0 2 M there exist an
open set U containing x0 and a di¤ erentiable function ' : W ! RN , where
W Rk is an open set such that
Remark 215 The di¤ erence between curves and surfaces, it’s that a surface is
a set of RN , while a curve is an equivalence class of functions. Also for surfaces
we do not allow self-intersections. Also, a curve can be described by only one
parametrization, while a surface usually needs more than one.
Remark 216 Note that Theorem 200 implies that if a bounded k-dimensional
di¤ erential surface is Peano–Jordan measurable, then its measure must be zero.
M := (x; y) 2 R2 : x2 y2 = 1 :
Note that the set M is not the range of a curve, because it is not connected. To
cover M we need at least two local charts, precisely, we can take the open sets
115
Note that both ' and are of class C 1 (the argument inside the square roots
t p t
is never zero). Moreover, '0 (t) = p1+t2
; 1 and 0 (t) := 1+t2
; 1 , and
1
so the rank of J' (t) and of J (t) is one. Finally, ' : M \ V ! R and
1
: M \ W ! R are given by
1 1
' (x; y) = y; (x; y) = y;
Gr h := f(y; t) 2 V R : t = h (y)g :
A chart is given by the function ' : V ! Rk+1 given by ' (y) := (y; h (y)).
Then,
Ik
J' (y) = ;
rf (y)
which has rank N . Note that ' is one-to-one and that ' (V ) = Gr h. Hence,
there exists ' 1 : Gr f ! V . Moreover, ' 1 is continuous, since the projection
: Rk+1 ! Rk
(y; t) 7! y
M \ U = fx 2 U : g (x) = 0g
116
where a > 0 and r > 0. De…ne M := ' ((0; 2 ) (0; 2 )) R3 .
First method. Let’s prove that M is a 2-dimensional surface. Note that ' is
of class C 1 and
0 1
r sin u cos v (r cos u + a) sin v
J' (u; v) = @ r sin u sin v (r cos u + a) cos v A :
r cos u 0
Observe that if r cos u + a = 0, then
0 1
r sin u cos v 0
J' (u; v) = @ r sin u sin v 0 A
r cos u 0
and so J' (u; v) cannot have rank 2. Thus, for M to be a 2-dimensional surface
we need to assume that r cos u + a 6= 0 for all u 2 (0; 2 ). In what follows we
assume that
a > r:
r sin u cos v (r cos u + a) sin v
The submatrix has determinant r sin u (r cos u + a).
r sin u sin v (r cos u + a) cos v
Then r cos u + a > 0. If sin u 6= 0, then the determinant is di¤ erent from zero
and so so J' (u; v) has rank 2. On the other hand, when sin u = 0, that is, when
u = , then 0 1
0 ( r + a) sin v
J' ( ; v) = @ 0 ( r + a) cos v A :
r 0
For any v 2 (0; 2 ), either cos v 6= 0 or sin v 6= 0 (or both) and so either the
0 ( r + a) cos v 0 ( r + a) sin v
submatrix or have determi-
r 0 r 0
nant di¤ erent from zero. Thus, we have shown that when a > r, J' (u; v) has
rank 2 for all (u; v) 2 (0; 2 ) (0; 2 ).
To see that ' is one-to-one, consider (x; y; z) 2 M . We want to …nd a
unique pair (u; v) such that ' (u; v) = (x; y; z). Note that to …nd u we can use
sin u = zr . The problem is that there are two values of u, one in 0; 2 [ 32 ; 2
and one in 2 ; 32 . If x2 + y 2 a2 , then
2 2 2
(r cos u + a) cos2 v + (r cos u + a) sin2 v = (r cos u + a) a2 ;
3
that is cos u (r cos u + 2a) 0, so that cos u 0, that is, 2 u 2 . So in
this case this determines uniquely u in terms of (x; y; z). On the other hand, if
x2 + y 2 > a2 , then cos u > 0, and so either 0 < u < 2 or 32 u < 2 . Again,
this determines uniquely u in terms of (x; y; z). In turn,
x y
cos v = ; sin v = ;
(r cos u + a) (r cos u + a)
which determine uniquely v. Thus, ' is one-to-one. We leave as an exercise to
check that ' 1 is continuous.
117
Second method: Let’s write down M explicitly.
p 2
M= (x; y; z) 2 R3 : z 2 = r2 x2 + y 2 a :
p 2
Note that r2 x2 + y 2 a 0, that is
p
r x2 + y 2 a :
p p
Since a > r, if x2 +py 2 a 0, we would get r a x2 + y 2 , which is a
contradiction. Hence, x2 + y 2 > a. Consider the function
p 2
g (x; y; z) := z 2 + x2 + y 2 a r2 :
Then
Jg (x; y; z) = rg (x; y; z)
!
p 2x p 2y
= 2 x2 + y 2 a p ;2 x2 + y 2 a p ; 2z :
2 x2 + y 2 2 x2 + y 2
p @g @g
Since x2 + y 2 > a, x and y cannot be both zero, so @x (x; y; z) or @y (x; y; z)
are di¤ erent from zero. Thus, Jg (x; y; z) has maximum rank 1. This shows that
M is a 2-dimensional surface of class C 1 .
Note that M is obtained by taking the circle of radius r and center (0; a; 0),
namely,
2
(y a) + z 2 = r2
and revolving it about the z.
The Klein bottle is obtained by starting from a rectangle of R2 , re‡ecting
one of its sides across its center and then performing the identi…cation.
Exercise 221 (Klein Bottle) Consider the function ' : (0; 2 ) (0; 2 ) ! R4
given by
u u
' (u; v) := (r cos v + a) cos u; (r cos v + a) sin u; r sin v cos ; r sin v sin :
2 2
Then Klein bottle is the set M := ' ((0; 2 ) (0; 2 )) R4 . Prove that M is a
2-dimensional surface of class C 1 .
Example 222 Consider the unit sphere in RN , N 2,
M := x 2 RN : kxk = 1 :
2
Given x0 2 M consider the function g (x) := kxk 1 (here N k = 1 and so
k = N 1). Then Jg (x) = rg (x) = 2x. By taking a ball B (x0 ; r) so small that
it does not intersect the origin, we have that 2x 6= 0 for all x 2 M \ B (x0 ; r)
and so M is a (N 1)-dimensional surface of class C 1 .
118
Example 223 Given an open set U RN and a function f : U ! R of class
1
C , for every c 2 R consider the level set
Bc := fx 2 U : f (x) = cg :
M := fx 2 U : f (x) = cg n fx 2 U : rf (x) = 0g :
Then 0 1
@'1 @'1
@y1 (y0 ) @yk (y0 )
B . .. C
det Jf (y0 ) = det B
@ .. .
C 6= 0;
A
@'k @'k
@y1 (y0 ) @y1 (y0 )
and so by the inverse function theorem there exists r > 0 such that Bk (y0 ; r)
W , f (Bk (y0 ; r)) is open, and f : Bk (y0 ; r) ! f (Bk (y0 ; r)) is invertible, with
inverse f 1 : f (Bk (y0 ; r)) ! Bk (y0 ; r) of class C m . Let z := (x1 ; : : : ; xk ),
then we have shown that we can write y as a function of z, y = f 1 (z). Let
w := (xk+1 ; : : : ; xN ) so that x = (z; w).
119
Since ' is a homeomorphism, the set ' (Bk (y0 ; r)) is relatively open in M ,
that is, it can be written as
' (Bk (y0 ; r)) = M \ U1
N
for some open set U1 R . Then
M \ U1 = f' (y) : y 2 Bk (y0 ; r)g :
1 1
= z; 'k+1 f (z) ; : : : ; 'N f (z) : z 2 U1 :
This shows that M \ U1 is given by the graph of the function z 2 U1 7!
'k+1 f 1 (z) ; : : : ; 'N f 1 (z) . Consider the function g : U1 ! RN k of
class C m de…ned by
1 1
g (x) := xk+1 'k+1 f (x1 ; : : : ; xk ) ; : : : ; xN 'N f (x1 ; : : : ; xk ) :
Then
M \ U1 = fx 2 U1 : g (x) = 0g :
Moreover, Jg (x) contains the submatrix IN k, since for i; j k + 1,
@gi @ 1
(x) = xi 'i f (x1 ; : : : ; xk ) = i;j 0:
@xj @xj
Hence, Jg (x) has maximum rank N k for all x 2 U1 .
Step 2: We prove that (ii) implies (i). Exercise.
15 Surface Integrals
To de…ne the integral of a function over a surface, we use local charts. Given
1 k < N we de…ne
N;k := 2 Nk0 : 1 1 < 2 < < k N :
m
Given a k-dimensional surface M of class C , m 2 N, consider a local chart
' : V ! M , let E ' (V ) be such that ' 1 (E) is Peano–Jordan measurable
and let f : E ! R be a bounded function. The surface integral of f over E is
de…ned as
v
Z Z u X 2
u @ (' 1 ; : : : ; ' k )
k
f dH := f (' (y)) t det (y) dy; (50)
E ' 1 (E) @ (y1 ; : : : ; yk )
2 N;k
s
2
1
P @ (' 1 ;:::;' k )
provided the function y 2 ' (E) 7! f (' (y)) 2 N;k det @(y1 ;:::;yk ) (y)
R
is Riemann integrable. Note that E f dHk exists if f is bounded and continu-
ous. In particular, if f = 1, the number
v
Z Z u X 2
u @ (' 1 ; : : : ; ' k )
Hk (E) := 1d = t det (y) dy
E ' 1 (E) @ (y1 ; : : : ; yk )
2 N;k
120
Remark 224 To understand the expression under the square root, observe that
the Jacobian matrix of ',
0 1
@'1 @'1
@y1 (y) @yk (y)
B .. .. C
J' (y) = B
@ . .
C
A
@'N @'N
@y1 (y) @yk (y)
M := (x; y; z; w) 2 R4 : x2 + y 2 = 1; z 2 + w2 = 1; x; z > 0 :
so that 0 1
sin u 0
B cos u 0 C
B
J' (u; v) = @ C:
0 sin v A
0 cos v
cos u 0
Since cos u; cos v > 0, the submatrix has determinant cos u cos v <
0 cos v
0, and so J' (u; v) has rank 2. Moreover ' is one-to-one with continuous inverse
and of class C 1 . Hence,
X @ (' 1 ; ' 2 )
2
sin u 0
2
sin u 0
2
det (u; v) = det + det
@ (u; v) cos u 0 0 sin v
2 4;2
2 2
sin u 0 cos u 0
+ det + det
0 cos v 0 sin v
2 2
cos u 0 0 sin v
+ det + det
0 cos v 0 sin v
= 0 + sin2 u sin2 v + sin2 u cos2 v + cos2 u sin2 v + cos2 u cos2 v + 0
= 1:
121
R
Let’s now calculate the surface integral M
x2 + z 2 dH2 . We have
Z Z 2 Z 2 p
x2 + z 2 dH2 = cos2 u + cos2 v 1 dudv
M 2 2
Z 2
Z 2
= cos2 u du + cos2 v dv = 2
:
2 2
16 Divergence Theorem
De…nition 227 An open and bounded set U RN is regular if there exist a
function g 2 C 1 RN , with rg (x) 6= 0 for all x 2 @U , such that
U = x 2 RN : g (x) < 0 ;
@U = x 2 RN : g (x) = 0 :
Since rg (x) 6= 0 for all x 2 @U , it follows that @U is an (N 1)-dimensional
surface of class C 1 .
De…nition 228 Given an open, bounded, regular set U RN , a point x0 2 @U ,
the tangent space to @U at x0 is given by
Tx0 = x 2 RN : rg (x0 ) x = 0 :
A vector 2 RN n f0g is called a normal vector to @U at x0 if it is orthogonal
to all vectors in Tx0 , that is, x = 0 for all x 2 Tx0 . A unit normal vector
to @U at x0 is called a unit outward normal to U at x0 if there exists > 0
such that x0 t 2 U and x0 + t 2 RN n U for all 0 < t < .
Exercise 229 Prove that if U RN is an open, bounded, regular set, and g is
as in De…nition 227, then for every x0 2 @U , the unit outward normal to U at
x0 is the unit vector
rg (x0 )
(x0 ) = :
krg (x0 )k
Hence, : @U ! RN is a continuous function.
We are ready to prove the divergence theorem.
Theorem 230 (Divergence Theorem) Let U RN be an open, bounded,
N
regular set and let f : U ! R be such that f is bounded and continuous in
U and there exist the partial derivatives of f in RN at all x 2 U and they are
continuous and bounded. Then
Z Z
div f (x) dx = f (x) (x) dHN 1 (x) ;
U @U
122
where
XN
@fi
div f := :
i=1
@xi
Proof. Step 1: We …rst prove that the divergence theorem holds when U is
replaced by a rectangle
R := (a1 ; b1 ) (aN ; bN ) :
Write R0 := (a2 ; b2 ) (aN ; bN ) and let x0 := (x2 ; : : : ; xN ). Then by
Theorem 160,
Z Z Z b1 !
@f1 @f1
(x) dx = (x1 ; x0 ) dx1 dx0
R @x1 R0 a1 @x1
Z
= (f1 (b1 ; x0 ) f1 (a1 ; x0 )) dx0
R0
Z Z
0 0
= f (b1 ; x ) e1 dx + f (a1 ; x0 ) ( e1 ) dx0
R0 R0
Z Z
= f (x) dHN 1 (x) + f (x) dHN 1 (x) ;
fb1 g R0 fa1 g R0
where in the third equality we have used the fundamental theorem of calculus
applied to the function of one variable x1 2 [a1 ; b1 ] 7! f1 (x1 ; x0 ) with x0 …xed,
and in the last equality
R @fi we have used formula (??). Repeating this argument for
all the integrals R @x i
(x) dx and then summing the resulting identities gives
the desired result.
Step 2: Next we prove that the divergence theorem holds when U is replaced
by a set of the form
V := (x1 ; x0 ) 2 R RN 1
: h (x0 ) < x1 < b1 ; x0 2 R0 ;
where R0 := (a2 ; b2 ) (aN ; bN ) and h : RN 1 ! R is a function of class C 1
and f = 0 in an open neighborhood of @V n graph h. The change of variables
y1 := x1 h (x0 ) ; y0 := x0
maps the set V into the set
W := (y1 ; y0 ) 2 R RN 1
: 0 < y1 < b1 h (y0 ) ; x0 2 R0 :
Let R := (0; b1 ) R0 and consider the function
f (y1 + h (y0 ) ; y0 ) if y 2 W;
g (y) :=
0 if y 2 R n W:
123
By Theorem 61, if y 2 W ,
@f1
div g (y) = (y1 + h (y0 ) ; y0 )
@x1
XN
@fi @h 0 @fi
+ (y1 + h (y0 ) ; y0 ) (y ) + (y1 + h (y0 ) ; y0 )
i=2
@x 1 @yi @xi
XN
@fi @h 0
= div f (y1 + h (y0 ) ; y0 ) + (y1 + h (y0 ) ; y0 ) (y ) :
i=2
@x1 @yi
124
and so
Z Z N Z
X @h 0
div f (x) dx = f1 (h (x0 ) ; x0 ) dx0 + fi (h (x0 ) ; x0 ) (x ) dx0
V R0 i=2 R0 @xi
Z N Z
X @h 0
= f1 (h (x0 ) ; x0 ) dx0 + fi (h (x0 ) ; x0 ) (x ) dx0
R0 i=2 R0 @xi
q
Z 2
1 + krh (x0 )kN
0 0 0 1
= f (h (x ) ; x ) ( 1; rh (x )) q dx0
R0 2
1 + krh (x0 )kN 1
Z
= f (x) (x) dHN 1 (x) ;
@V
where in the last equality we have used formula (??). This concludes the proof
in this case.
Note that a similar proof works when the role of the variable x1 is played
by another variable xi , so that
V := x 2 RN 1
: h (xi ) < xi < bi ; xi 2 R0 ; (51)
V := x 2 RN 1
: ai < xi < h (xi ) ; xi 2 R0 ; (52)
1 1
Q (x0 ; rx0 ) = a rx0 ; a + rx0 QN 1 (a; rx0 ) (a ; a + ) BN 1 (b; r)
2 2
and by the previous discussion U \ Q (x0 ; rx0 ) has the form (51).
125
Step 4: If x 2 @U , there exists an open cube Q (x; rx ) centered at x and of
side-length rx such that U \ Q (x; rx ) is either of the form (51) or (52). On the
other hand, if x 2 U , which is open, then there exists Q (x; rx ) U .
For every x 2 U construct a nonnegative function 'x 2 C 1 RN such that
'x > 0 in Q x; 21 rx and 'x = 0 outsideQ (x; rx ) (exercise). Since U is closed
and bounded, it is compact. Since the family of open sets Q x; 12 rx x2U
cover U , there exist a …nite number Q x1 ; 12 rx1 , . . . , Q xn ; 21 rxn that still
cover U . Note that Q (xk ; rxk ) is either contained in U or U \ Q (xk ; rxk ) is of
the form (51) or (52) and that
n
X
' := 'k > 0 in U ;
k=1
since Q x1 ; 12 rx1 , . . . , Q xn ; 12 rxn that still cover U and 'k > 0 in Wk . De…ne
(
'k (x)
'(x) if 'k (x) > 0;
k (x) :=
0 otherwise.
Pn
Then k 2 C 1 RN , k = 0 outside Q (xk ; rxk ) and k=1 k = 1 in U . The
family f k g is called a partition of unity subordinated to the family of open sets
Q (x1 ; rx1 ), . . . , Q (xn ; rxn ). Hence,
n
! n
@ @ X X @ k
0= (1) = k =
@xi @xi @xi
k=1 k=1
where we have used the fact thatPn k f is zero outside Q (xk ; rxk ). Summing (53)
over k and using the fact that k=1 k = 1 in U , we have
Z Z n
! n Z
X X
div f dx = div kf dx = div ( k f ) dx
U U k=1 k=1 U
n Z
X Z n
X Z
= kf dHN 1
= kf dHN 1
= f dHN 1
;
k=1 @U @U k=1 @U
126
R
Remark 231 In physics @U f (x) (x) dHN 1 (x) represents the outward ‡ux
of a vector …eld f across the boundary of a region U .
Friday, April 27, 2012
If E RN and f : E ! RN is di¤erentiable, then f is called a divergence-free
…eld or solenoidal …eld if
div f = 0:
Thus for a smooth solenoidal …eld, the outward ‡ux across the boundary of a
regular set U is zero.
Corollary 232 The theorem continues to hold if U RN is open, bounded, and
its boundary consists of two sets E1 and E2 , where E1 is a closed set contained
in the …nite union of compact surfaces of class C 1 and dimension less than
N 1, while for every x0 2 E2 there exist a ball B (x0 ; r) and a function
g 2 C 1 (B (x0 ; r)) such that with rg (x) 6= 0 for all x 2 B (x0 ; r) \ @U , such
that
B (x0 ; r) \ U = fx 2 B (x0 ; r) : g (x) < 0g ;
B (x0 ; r) n U = fx 2 B (x0 ; r) : g (x) > 0g ;
B (x0 ; r) \ @U = fx 2 B (x0 ; r) : g (x) = 0g :
Note that the radius of the ball and the function g depend on x0 .
Example 233 Let’s calculate the outward ‡ux of the function
f (x; y; z) := (0; yz; x)
across the boundary of the region
U := (x; y; z) 2 R3 : x2 + y 2 < z 2 ; x2 + y 2 + z 2 < 2y; z > 0 :
Note that U is not a regular open set (why?) but it satis…es the hypotheses of
the previous corollary (why?).
We have
@ @ @
div f (x; y; z) = (0) + (yz) + (x) = 0 + 1z + 0;
@x @y @z
and so by the divergence theorem
Z Z
f dH2 = z dxdydz:
@U U
127
Hence, by changing variables
Z Z Z Z p ! !
sin 2r sin r2
z dxdydz = z dz r dr d
U 0 0 r
Z Z p
z= 2r sin r2
!
sin
z2
= r dr d
0 0 2 z=r
Z Z !
sin
2 3
= r sin r dr d
0 0
Z r=sin Z
r3 r4 1 1
= sin d = sin4 sin4 d
0 3 4 r=0 0 3 4
Z
1 1
= sin4 d = :
0 12 32
f (x) g (x) if j = i;
fj (x) :=
0 if j 6= i:
Then
XN
@fj @ (f g) @g @f
div f = = =f +g ;
j=1
@xj @x i @xi @xi
and so
Z Z Z Z
@g @f N 1
f +g dx = div f dx = f dH = fg i dHN 1
:
U @xi @xi U @U @U
128
R2 . Then the hypotheses of Corollary 232 are satis…ed (why?). Given t 2 [a; b],
the vector '0 (t) is tangent to the curve at the point ' (t) 2 @U . Hence, if
' (t) = (x (t) ; y (t)), the outer normal to @U at the point ' (t) is either
depending on the orientation of the curve. We say that has a positive orien-
tation for @U if as we traverse the curve starting from t = a we …nd U on our
left. In this case, the outward normal at the point ' (t) is is given by
(y 0 (t) ; x0 (t))
(' (t)) = q :
2 2
(x0 (t)) + (y 0 (t))
129
Taking f (x; y) = x in (54), we get the second area’s formula
Z Z b Z
meas U = 1 dxdy = x (t) y 0 (t) dt = x dy;
U a @U
Example 236 Let’s …nd the area of the region U enclosed by the curve of para-
metric representation
x ( ) = (1 cos ) cos ;
2 [0; 2 ] :
y ( ) = (1 cos ) sin ;
To see that it is simple, note that for 0 < < 2 , (1 cos ) > 0. The curve
starts from the origin and goes around clockwise only once. Note that
y 0 (0) = y 0 (2 ) = 2 + 1 + 1 = 0; y0 ( ) = 2 1 + 1;
1 1 1
y0 = 2 cos2 + cos + 1 = 1;
3 3 3
5 5 5
y0 = 2 cos2 + cos + 1 = 1:
3 3 3
Hence, the curve is not regular, but since these are only a …nite number of bad
points, we can apply Corollary 232. Using the third area formula, we get
Z
1 2
meas U = (1 cos ) sin [sin cos (1 cos ) sin ] d
2 0
Z
1 2
+ (1 cos ) cos [sin sin + (1 cos ) cos ] d
2 0
Z Z
1 2 2 1 2 2 3
= (1 cos ) cos2 + sin2 d = (1 cos ) d = :
2 0 2 0 2
Monday, April 30, 2012
130
Proposition 237 Given two open sets A; D RN , with D bounded and dist (A; D)
3d > 0 for some d > 0, there exists a function f 2 C 1 RN such that,
0 f 1, f (x) = 0 for all x 2 A, f (x) = 1 for all x 2 D, and
C
krf (x)k
d
for all x 2 RN , where C > 0 depends only on N .
De…ne Z
1 x y
f (x) := g W (y) dy; x 2 RN :
dN RN d
Let’s prove that f has the right properties. Since g = 0 in RN n B (0; 1), we
have that g x d y = 0 for all x d y > 1, and so
Z
1 x y
f (x) = N g dy:
d B(x;d)\W d
x y
where we have made the change of variables d = z.
131
To prove that f is of class C 1 , let ei , i = 1; : : : ; N , be an element of the
canonical basis of RN and for every h 2 R, with 0 < jhj , consider
Z
f (x + hei ) f (x) 1 1 @g x y
W (y) dy
h dN RN d @xi d
2 3
Z g x+hei y
g x y
1 4 d d 1 @g x y 5
= N W (y) dy
d RN h d @xi d
Z Z !
1 1 h @g x y + tei @g x y
= N +1 dt W (y) dy
d RN h 0 @xi d @xi d
Z Z !
1 1 h @g x y + tei @g x y
= N +1 W (y) dy dt;
d h 0 B(x;d+ d) @xi d @xi d
where we have used Fubini’s theorem and the fact that supp g B (0; 1).
Since g 2 Cc1 RN its partial derivatives are uniformly continuous. Hence
for every > 0 there exists = ( ; x; ) > 0 such that
@g @g
(z) (w)
@xi @xi
for all z; y 2 B (x; 1 + ), with jz wj . Then for 0 < jhj min f ; g we
have
Z
f (x + hei ) f (x) 1 @g x y
W (y) dy ;
h dN +1 RN @xi d
which shows that
Z
@f 1 @g x y
(x) = N +1 W (y) dy:
@xi d RN @xi d
132
Proof. The proof is very similar to the one of Theorem 199 and is left as an
exercise.
Using the previous theorem we can show that if two k-dimensional surfaces
intersect “transversally”, then their intersection has k-Peano–Jordan measure
zero.
Using again the fact that partial derivatives of g are uniformly continuous, we
can prove that the partial derivatives of f are continuous, so that f is of class
C 1.
133
and so, using the fact that f = 0 in D,
Z Z Z n
X n Z
X
div f dx = div f dx = div ( kf ) dx = div ( kf ) dx
U U nD U nD k=1 k=1 U nD
n Z
X Z n
X Z
N 1 N 1
= kf dH = kf dH = f dHN 1
;
k=1 @U nD @U k=1 @U
1
An := x 2 RN : dist (x; E1 ) < ;
2n
3
Dn := x 2 RN : dist (x; E1 ) > n :
2
1
Then dist (An ; Dn ) 2n . By the previous proposition there exists 'n 2
1 N
C R such that 'n (x) = 0 for all x 2 An , 'n (x) = 1 for all x 2 Dn ,
and
kr'n (x)k C2n
for all x 2 RN , where C > 0 depends only on N .
Then the function 'n f satis…es all the hypotheses of Step 1 and so
Z Z
div ('n f ) dx= 'n f dHN 1 :
U @U
Now
N
X N
X XN
@ ('n fi ) @'n @fi
div ('n f ) = = fi + 'n
i=1
@xi i=1
@xi i=1
@xi
and so Z Z Z
div ('n f ) dx = rf r'n dx + 'n div f dx:
U U U
Since 'n (x) = 1 for all x 2 Dn , 0 'n 1, and the partial derivatives of f are
bounded,
Z Z Z
1 div f dx 'n div f dx = (1 'n ) div f dx
U U U
Z
= (1 'n ) div f dx
U nDn
3
M measo x 2 U : dist (x; E1 ) !0
2n
134
as n ! 1 by Lemma 243. Moreover, since r'n (x) = 0 if x 2 An [ Dn , and
kr'n (x)k C2n otherwise,
Z Z
rf r'n dx = rf r'n dx
U fy2U : 21n dist(y;E0 ) 23n g
3
CM 2n measo x 2 U : dist (x; E1 ) !0
2n
as n ! 1 again by Lemma 243.
Finally, since 'n (x) = 1 for all x 2 Dn and f is bounded,
Z Z Z
1f dHN 1 'n f dHN 1 = (1 'n ) f dHN 1
@U @U @U
Z
M (1 'n ) dHN 1 M HN 1
(@U n Dn ) :
@U
Using improper surface integrals, which we have note done..., one can show that
HN 1 (@U n Dn ) ! 0.
Although it was not needed in the proof, it actually turn out the sets Dn
are Peano–Jordan measurable.
Friday, May 04, 2012
Lemma 243 Let K RN be a compact set with (N 1)-Peano–Jordan mea-
sure zero and for every r > 0 consider the set
Ar := x 2 RN : dist (x; K) < r :
Then
1
lim+ measo (Ar ) = 0:
r!0 r
Proof. Since K has (N 1)-Peano–Jordan measure zero for every " > 0 there
exists r" > 0 with the property that for every 0 <Sr r" there exists a …nite
family fQ (xn;r ; r)gn of open cubes such that K n Q (xn;r ; r) and
X p N 1 X N (N 1)=2
N 1
Nr = (diamQ (xn;r ; r)) ":
n n
2N
S
If x 2 Ar , then there is y 2 K such that kx yk < r. Since, K n Q (xn;r ; r),
there is a cube Q (xn;r ; S r) such that x 2 Q (xn;r ; r). It follows that y 2
Q (xn;r ; 2r) and so Ar n Q (xn;r ; 2r). In turn,
1 1X 1X N
measo (Ar ) meas (Q (xn;r ; 2r)) = (2r)
r r n r n
n
X
2N p N 1
= Nr ":
N (N 1)=2
n
135
17 Stokes Theorem
We recall that given two vectors a = (a1 ; a2 ; a3 ) and b = (b1 ; b2 ; b3 ) of R3 , their
cross-product or vector-product is de…ned by
a b := (a2 b3 a3 b2 ; a3 b1 a1 b3 ; a1 b2 a2 b1 )
0 1
e1 e2 e3
= det @ a1 a2 a3 A :
b1 b2 b3
We will present Stokes’theorem in a very special setting. The general case relies
on the notion of oriented manifolds with boundaries. Let W R2 be an open
3 2
set and let ' : W ! R be a one-to-one function of class C such that J' (u; v)
has maximum rank 2 for all (u; v) 2 W . Consider a bounded regular open set
V R2 with V W and and assume that its boundary @V is the range of a
closed, simple, regular curve with parametric representation : [a; b] ! R2 .
Let M := ' (V ). Note that M is a 2-dimensional di¤erential surface. We
will call the boundary of M to be the set
@ M := ' (@V ) :
136
Theorem 244 (Stokes’Theorem) Let U R3 be an open set and let f :
3 1
U ! R be a function of class C . Let M and be as above with M [@ M U .
Then Z Z
2
curl f dH = f:
M
ZZ @' @'
@f1 @f3 @'3 @'1 @'1 @'3 @u @v
+ ' ' dudv
V @z @x @u @v @u @v @' @'
@u @v
ZZ @' @'
@f2 @f1 @'1 @'2 @'2 @'2 @u @v
+ ' ' dudv:
V @x @y @u @v @u @v @' @'
@u @v
After some tedious calculations, we obtain that the right-hand side of the pre-
vious equality reduces to
ZZ
@h @g
dudv;
V @u @v
where
@'1 @'2 @'3
g (u; v) := f1 (' (u; v)) (u; v) + f2 (' (u; v)) (u; v) + f3 (' (u; v)) (u; v) ;
@u @u @u
@'1 @'2 @'3
h (u; v) := f1 (' (u; v)) (u; v) + f2 (' (u; v)) (u; v) + f3 (' (u; v)) (u; v) :
@v @v @v
137
Indeed, by the chain rule,
@h @g @f1 @'1 @'1 @f1 @'2 @'1 @f1 @'3 @'1
= ' + ' + '
@u @v @x @u @v @y @u @v @z @u @v
@f2 @'1 @'2 @f2 @'2 @'2 @f2 @'3 @'2
+ ' + ' + '
@x @u @v @y @u @v @z @u @v
@f3 @'1 @'3 @f3 @'2 @'3 @f3 @'3 @'3
+ ' + ' + '
@x @u @v @y @u @v @z @u @v
@f1 @'1 @'1 @f1 @'2 @'1 @f1 @'3 @'1
' ' '
@x @v @u @y @v @u @z @v @u
@f2 @'1 @'2 @f2 @'2 @'2 @f2 @'3 @'2
' ' '
@x @v @u @y @v @u @z @v @u
@f3 @'1 @'3 @f3 @'2 @'3 @f3 @'3 @'3
' ' '
@x @v @u @y @v @u @z @v @u
@f1 @'2 @'1 @'2 @'1 @f1 @'3 @'1 @'3 @'1
= ' + '
@y @u @v @v @u @z @u @v @v @u
@f2 @'1 @'2 @'1 @'2 @f2 @'3 @'2 @'3 @'2
+ ' + '
@x @u @v @v @u @z @u @v @v @u
@f3 @'1 @'3 @'1 @'3 @f3 @'2 @'3 @'2 @'3
+ ' + ' :
@x @u @v @v @u @y @u @v @v @u
138
where V := (u; v) 2 R2 : u2 + v 2 < 1 , let’s calculate
Z
curl f dH2 ;
M
where
f (x; y; z) := x2 z; y; yz ; (x; y; z) 2 R3 :
Note that ' is de…ned in R2 so we can take W = R2 . To see that ' is one-to-
one, we …nd the inverse, we have x = u v, y = u, z = u2 + v 2 and so, u = z,
v = u x = z x. Thus,
1
' (x; y; z) = (z; z x) ;
1 1
The submatrix has determinant di¤ erent from zero, and so J' (u; v)
1 0
has always maximal rank two. Thus, we are in a position to apply Stokes’
theorem. The boundary of V is the unit circle. Thus a parametric representation
of is : [0; 2 ] ! R2 , where (t) := (cos t; sin t). In turn, a parametric
representation for is given by := ' : [0; 2 ] ! R3 , where
In turn,
0
(t) = ( sin t cos t; sin t; 0) :
Hence, by Stokes’ theorem
Z Z
curl f dH2 = f
M
Z 2 h i
2
= (cos t sin t) 1 ( sin t cos t) + (cos t) ( sin t) + (cos t) (1) 0 dt
0
Z 2
= cos2 t sin2 t (cos t sin t) + cos t sin t dt
0
Z 2
= 1 2 sin2 t cos t 1 + 2 cos2 t sin t + cos t sin t dt
0
t=2
2 2 1
= sin3 t + cos t + cos3 t + sin2 t
sin t = 0:
3 3 2 t=0
R
Exercise 246 In the previous example calculate M curl f dH2 directly, with-
out using Stokes’s theorem.
139
Remark 247 Stokes’ theorem continues to hold for oriented manifolds with
boundaries. Roughly speaking, a 2-dimensional manifold M in R3 is orientable
if it is possible to de…ne a normal as a function of (x; y; z) in a continuous way.
The Klein bottle or the Möbius strip are a typical examples of manifolds that not
orientable. Indeed, the normal as a function of (x; y; z) is discontinuous (but it
is continuous if you look at it as a function of (u; v)). If you think of the normal
as a person walking on the manifold, after going around once, that person would
…nd himself/herself upside down, which means that there is a discontinuity.
140
use (50) for the surface integral of f hi over the set E \ 'i (Wi ). Hence, we can
de…ne the k-dimensional surface integral of f over E to be
Z Xn Z
k
f dH := f hi dHk ; (57)
E i=1 E\'i (Wi )
R
whenever all the surface integrals E\'i (Wi ) f hi dHk exist in the sense of (50).
Note that this is always the case if f is bounded and continuous.
We need to show that this de…nition does not depend on the choice of the
partition of unity. Let j : Vj ! M \ Dj , j = 1; : : : ; m, be other local charts
covering M , with Vj Rk and Dj RN open and let pj 2 C 1 RN , j =
1; : : : ; m, be a partition
Pn of unity subordinated to the family of open sets D1 ,
. . . , Dm . Since i=1 hi = 1 in M , for every j = 1; : : : ; m and every x 2 M ,
n
X
f (x) pj (x) = f (x) pj (x) hi (x) ;
i=1
and so using the surface integral de…ned in (50) for the local chart j (note that
f pj is zero outside Dj ), we have
Z Z Xn Xn Z
f pj dHk = f pj hi dHk = f pj hi dHk
E\ j (Vj ) E\ j (Vj ) i=1 i=1 E\ j (Vj )
n Z
X
= f pj hi dHk :
i=1 E\ j (Vj )\'i (Wi )
Since the right-hand sides of (58) and (59) are equal, it follows that
Xm Z Xn Z
k
f pj dH = f hi dHk ;
j=1 E\ j (Vj ) i=1 E\'i (Wi )
which shows that the de…nition (57) does not depend on the particular choice
of the local charts and partition of unity.
141