This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
0. 0
Lemma 2.19 Let 0 E H(D) be such that 0(x + iy) = 01(x, y) + i02 (X, y) for every x, y E R. Set
JO(x,y):=det(j Tl. Then J,(x, y) > 0 for every x, y E R.
DEGREE THEORY IN FINITE-DIMENSIONAL SPACES
44
Ch. 2
Proof Since (zo) _ li .
O(zo + h) - 4(zo)
taking h = t E R (resp. h = it, t E R), we obtain (zo) =
(zo) = VOL + i
(resp.
- i L) which yield the Cauchy-Riemann equations 1901
amz
ax = ay
and
801
aO2
ay = ay
We conclude that
J2>0.
J-0
Theorem 2.20 Let D C C be a connected, bounded, open set such that 8D is a C' closed curve. Let 4 : D -+ C be a holomorphic function and let p it O(aD). Then
d(c, D, p) = w(5, D, p) =
tai
f
D
O() z) p
dz.
Proof Suppose, first, that 0 is constant, O(z) = c for every z E D. For every p E C \ O(8D) we have p E C \ O(D) and so d(m, D, p) = 0. Also, w(q, D, p) _ °-: = 0 and so 2 i f8D a-p d(O, D, p) = w(O, D, p) -
Now let 0 be a nonconstant holomorphic function in D. If p then d(O, D, p) = 0 and z : n is a holomorphic function on D. Therefore,
feDO:zpdz=0and d(O, D, p) = w(O, D, p)
Assume that p E O(D). Since O(z) is nonconstant, 0-'{p} must be finite (see Remark 2.16), say
¢-'{p} = {z1,...,zk}. By Lemma 2.18, for each j = I,-, k, there are R, > 0,
3
: B(zJ, R,) -+ C
such that
cps E H(B(zj, Rj)), B(z., Rj) CC D, : B(zi, R,) -+ vj (B (z.,, R,)) is a bijection, $,(z) - p = (z - z))m gj(z), for every z E B(z,, Rj), cps
DECREE AND WINDING NUMBER
§2.5
45
(,p.1)m = (z - zj)m gj(z) for every z E B(z7, Rj),
for some gj E H(B(zj, Rj)). If R := z min{Rl,... , Rk}, then w (0, D , p ) =
1J
(z)
.
27rt
1
k
(Z )
O(z)
OD
-P
dz =
27ri J8B(z,R)
j_1
O(z) - P
d z.
We now show that
¢ (z) dz
1
27ri f 8B(z,,R) O(z) - P
d z = mj.
Indeed,
O(Z) = P+ (z - zi),gi(z) and so z,)":,-1[mjg1(z)
4 (z) = (z -
+ (z - zj)9g(z))
This implies that (z) dz = 8B(z;,R) O(z) - P
mi +
J
Z - z0
9j (z) dz = 27rimj;
fm.,,,R) 9j (z)
therefore, k
w(O,D,P)=Emj. j=1
mj. It remains to show that d(o,D,p) Let C be the connected component of C \ 0(8D) containing p. It is obvious that C is an open set, hence there is 60 > 0 such that p+6 E C for every 161 < 60 and we have d(f, D, p) = d(¢, D, p + 6) for every 161 < 6o. The equation
/(z) = p + 6, z E B(zj, R) is equivalent to
p + (V(z))m' = p + 6, z E B(zj, R), which, in turn, is equivalent to (w(z))m' = Iblesxte,
= e2w'B and 9 E (0, If. As (pj is injective on B(z,, R), the equation (wj(z))'", = I6le2xie has exactly m j distinct solutions in B(z,,R) for every where
161 < 61, for some 60 > 61 > 0. Let z,',..., z,'j be these solutions. We have
46
DEGREE THEORY IN FINITE-DIMENSIONAL SPACES
Ch. 2
0' (z") = mj(' ,(z,))m3-lpj(z") 0 0.
Using the definition of the degree for C' mappings by Lemma 2.19, we have k
mj
k
sgn(J,(z )) _ E mu
d(O, D, p + b) = E
j=1
j=1 l=1
and we conclude that d(0, D, p) = w(0, D, p) -
0 2.6 Exercises Exercise 2.1 Let D = (-1,1) x (-1,1) and let 0 E C(D)2 be defined by 0(x1, x2) _ (max{Ixi I, Ix2I}, 0), for (xl, x2) E D.
Find d(0, D, p) at p = (0, 0).
Solution 2.1. First we note that p = (0, 0) ¢ 0(8D) and so d(0, D, p) is well defined. Let IP E C1(D)2 be defined by ?(xl,x2) = (1,0), for(xl,x2) E D.
We have cI8D = 01 aD and p 0 t/'(D). By Theorem 2.1 we have d(tb, D, p) = 0 and by Theorem 2.4 we deduce that d(0, D, p) = 0.
Exercise 2.2 Let D = (-1,1) x (-1,1) and let 0 E C(D)N be defined by O(xl, x2) = (x2 -
X31
, x2), (xl, x2) E D.
Find d(0, D, p) at p = (0, 0).
Solution 2.2. We have 0(x1, x2) = (0,0) if and only if (x1, x2) = (0,0)Therefore, (0,0) ¢ 0(8D) and so d(0, D, p) is well defined. Also, J4(xl, x2) _
-3xi and so (0,0) E 0(Z0). Setting q = (-11,0), then q and (0,0) belong to the same connected component of R2 \ 0(8D). Moreover, 0(x1, x2) = q if and only if (X 1, x2) _ (2', 0). Since q 0 0(8D), it is easy to see that d(0, D, q) _ Ese'_'(o) sgn(J.O(x))
= -1; thus
d(0, D, p) = -1.
Exercise 2.3 Let a, b E R be such that a < b and set D = (a, b). Assume that 0 E C(D) and p ¢ {0(a), 0(b)}. Prove that d(0, D, p) E (-1,0, 11.
EXERCISES
§2.6
47
Solution 2.3. We first note that p ¢ 0(8D) and so d(0, D, p) is well defined. Set
0(b) ?P(x)
- a(a) (x - a) + 0(a).
If 0(b) = O(a), then 0 is a constant and d(i,1, D, p) = 0. Assume that 0(b) 96 0(a). We obtain Ob _Q ° P E [0( a ) , 0(b)] sgn d(O,D,P) = 0
p V 10(a), c(b)]
and Since 0I8D ='I8D, by Theorem 2.4 we deduce that d(d, D, p) = d(t, D, p) and so d(O,D,p) E {-1,0,1}.
Exercise 2.4 [Poincard-Bohl] Let D C RN be an open, bounded set, 0, ) E C(D)N and p [O(x),Vi(x)] := {aO(x) + (1 - a)10(x) : a E [0,1]} for every x E 8D. Prove that d(0, D, p) = d(ib, D, p). Solution 2.4. It is obvious that d(¢, D, p) and d(o, D, p) exist. Set
H(x, t) := tO(x) + (1 - t)0(x), x E D, t E [0,1].
H is a C homotopy between tai and 0 and p E H(8D, t) for every t E [0, 1]. Therefore, Theorem 2.3 implies that and so d(O, D, p) = d(+L, D, p).
t), D, p) does not depend on t E [0,1]
3
SOME APPLICATIONS OF THE DEGREE THEORY TO TOPOLOGY 3.1 The Brouwer Fixed Point Theorem In this section we give two versions of the Brouwer Fixed Point Theorem.
Definition 3.1 A set E C RN is said to be a convex set if for every x, y E E and for every A E (0, 1), Ax + (1 - A)y E E.
Definition 3.2 Let E be a convex set such that 0 E int E. We define the gauge of E (or Minkowski function) , PE, by
xERN.
PE(x):=inf{t>0 : t EE
Remark 3.3 Since 0 E int E, there is n > 0 such that B(0, 77) C E. As i E E it follows that {t > 0 : E E} 0 0 and so pE(x) is well for every t > defined.
Lemma 3.4 Let E C RN be a convex set containing 0 in its interior. Let PE be the gauge of E. Then
(1) XEE
pE(x) 0 and for every x E RN; (3) there exists m > 0 such that pE(x) 5 mIx12 for all x E RN; (4) PE (X + y) 5 PE (x) + PE (Y) for every x, y E RN; (5) PE is continuous in RN; (6) if E is bounded, then pE(x) > 0 for every x 96 0 and os = E E.
Proof (1) Clearly, if x = i E E, then inf{t > 0 : e E E} 5 1, i.e. pE(x) < 1. (2) Given A > 0,
PE(AX) =inf{t>O : = inf r As
fx
EE}
EE, s > 0 }
:
JJJ
=Ainf{s>0 = APE(x) 48
:
x EE}
THE BROUWER FIXED POINT THEOREM
§3.1
49
(3) Since 0 E E, there exists 77 > 0 such that B(0, q) C E. For every t _> ""
we obtain that i E E and so PEW < ' (4) Let x, y E RN. There exist two sequences {an }, 10n) C (0, oo) such that
PE (X) = Jim an,
x -an EE
and
PE(Y) =n-.+oo lim Qn, LEE. Since E is a convex set, we deduce that x an an + On an thus Qn+
On q qy E E; an + Yn Fin
E E and so an + i3n > PE (X + y). Passing to the limit we have PE(x) + PE(Y) >- PE(X + Y)-
(5) Let x, h E RN. By (3) and (4) we have PE(x + h) - pE(x) 5 PE(h) < Alhl2
PE(x + h) < PE(x) + PE(h)
and
-PE(x + h) + PE(x) 5 \lhl2-
PE(x) 5 PE(x + h) + pE(-h) Therefore,
IpE(x + h) - pE(x)I 5 A h12
and so PE is continuous in RN.
(6) Assume that E is bounded and let R > 0 be such that E C B(0, R). If i E E, then 41
I
R > 0 if x 0 0.
Also, ° E E for every t > 0 implies that PE(0) = 0. Finally, considering a sequence {a,, } C (0,oo) such that pE(x) = lim an
and -& E E, taking the limit as n goes to infinity, we conclude that x PE(x)
E E.
13
50 SOME APPLICATIONS OF THE DEGREE THEORY TO TOPOLOGY Ch. 3
Proposition 3.5 Let E C RN be a compact, convex set such that 0 E int E. There exists a homeomorphism a : RN : RN such that
a(E) = B, where B is the unit ball of RN with respect to the norm I
Proof
- 12
R
a(x) :=
I x x#0 0
x=0,
vsyy y0 /3(y)
10
y=0.
O we have ao/3(y) = a(1E y) = PEW 2 p_ y = y and ao/3(0) = 0. Thus ao,Q(y) = y for every y E RN and, vsimilarly,y f3oa(x) = x for every x E RN. Therefore, a : RN RN is a bijection. From Lemma 3.4 (5) it follows that a is continuous at every x 96 0. Moreover, For every y
Ia(x) - a(0)I2 = IPIE12)xl2 = PE(x); therefore lim a(x) = 0 and a is continuous at 0. Hence, a is continuous on RN. X-0 Similarly, #: RN -+ RN is continuous.
Next, we prove that a(E) = B. By Lemma 3.4 (1), x E E implies that Ia(x)I2 = PE(x) < 1, therefore a(E) C B(0,1). Conversely, let y E B(0,1). Then y = a(/3(y)) and pE(/3(y)) = Iyi2 5 1 and by Lemma 3.4 (6) we have WOW) E E. Since E is a convex set containing 0, we obtain (1 - PE o ft)) 0 + PE 0)3(Y)
0(y) E E, PE 00(y)
i.e. /3(y) E E. Thus y E a(E) and so 6(0,1) C a(E).
Remark 3.6 Using the gauge function it can be shown that if Cl c RN is a smooth domain (e.g. strongly Lipschitz) and if CN(0) = CN (B(0,1)), then there exists a Lipschitz map v : fl --+ B such that (i) v(fl) = B; (ii) det Vv = 1 CN a.e. x E 11; (iii) v E C'(U;) for some finite partition {U,}i=1.....M of Cl into smooth domains (e.g. strongly Lipschitz).
For a proof of this result, we refer the reader to Fonseca and Parry (1987). The next two results are known as Brouwer's Fixed Point Theorem, the first one being the most commonly stated, while the latter has wider applications.
§3.1
THE BROUWER FIXED POINT THEOREM
51
Theorem 3.7 [First Version of the Brouwer Fixed Point Theorem] Let D C RN be an open, bounded set such that D is homeomorphic to the closed unit ball B. Let 0 E C(D)N be such that ¢(D) C D. Then 0 has a fixed point in D. Proof Let a : D - B be a homeomorphism. Setting 1G := a o o o a-1, we show that t/i : B - B has a fixed point. Clearly, 0 is continuous and either there exists
x E 8B such that O(x) = x, in which case' admits a fixed point, or O(x) 96 x for every x E 8B. Define
H(x, t) := x - tO(x) for x E B, t E [0,1]. Then 0 ¢ H(8B, t) for every t E [1, 0] and so, by Theorem 2.3 (2), d(H(., t), B, 0) does not depend on t. This yields
d(I -0, B, 0) = =
B, 0) 0), B, 0)
= d(I, B, 0) = 1.
By Theorem 2.7 the equation (I - t/)(x) = 0 admits a solution in B, i.e. lk has a fixed point x E k Setting y = a''(x), we have y E D and ¢(y) = Y.
0
Corollary 3.8 [Second Version of Brouwer Fixed Point Theorem) Let K C RN be a compact, convex set such that the interior of K is not empty. Let ¢ E C(K)N be such that O(K) C K. Then 0 admits a fixed point.
Proof Let xo E int K and define T : RN RN by T(x) = x - xo. Then T(K) is a compact, convex set such that 0 E int T(K). By Proposition 3.5 there exists a homeomorphism a : T(K) B and we define
tp:=To0oT-1. Then t[i : T(K) - T(K) is continuous and by Theorem 3.7 1b has a fixed point
y E T(K). Let x E K be such that y = T(x). It is clear that x is a fixed point of 0.
We give some applications of the Brouwer Fixed Point Theorem.
Proposition 3.9 Assume that N is odd, 0 E D, and 0 E C(D)N is such that 0 0 q5(8D). Then there exist y E 8D, \ 0 0, such that ¢(y) = ay. Proof If for every y E 8D, A E R, we have 0(y) 4 Ay, then set H(x, t) := tx + (1 - t)O(x), x E D, t E [0,1]
52 SOME APPLICATIONS OF THE DEGREE THEORY TO TOPOLOGY Ch. 3 and
K(x, t) := -tx + (1 - t)O(x), x E D, t E [0,1]. We have 0 ¢ H(OD, t) for every t E [0, 1] and 0 it K(8D, t) for every t E [0,1]. Therefore, by Theorem 2.3, d(H(0, t), D, 0) and t), D, 0) are independent of t and so, as H(., 0) = 0),
1 = d(I, D, 0) = d(-1, D, 0) _ (-1)N = -1, which yields a contradiction. We conclude that there exist y E 8D and A E R 0 such that 0(y) = Ay and, since 0 E O(8D), we must have A 96 0.
Theorem 3.10 [Perron-Frobenius] Let A = (ate) be an N x N matrix such that ai, > 0 for all i, j. Then there exist A > 0, x # 0, such that x; > 0 for every i and Ax = Ax.
0
Proof See Exercice 3.1.
Remark 3.11 In Proposition 3.9 the condition `N is odd' is essential. Indeed, let N = 2 and define 0 : B -+ R2 in polar co-ordinates by 4(r, 9) _ (r, 0 + r). The equation (cos(9 + 1), sin(B + 1)) = A(cos 0, sin 0)
has no solution.
The next application of Proposition 3.9 provides periodic solutions for Lipschitz flows. We will need some standard results on ordinary differential equations, which we will state without proving them.
Theorem 3.12 Let f : B(0, r) x R -+ RN be a locally Lipschitz function. Consider the initial value problem
fx= f(x,t) j x(to) = xo, when to E R, xo E B(0, r). Then there exists a function x : U . RN, (t; to, xo) r+ x(t, to, xo),
when U := ((t, to, xo) (to, xo) E R x B(0, r), t E 1(to, xo) ), such that (i) for each (to, xo) E R x B(0, r), x(t; to, xo) is continuously differentiable with respect to t, and it satisfies (3.12) in the open interval I(to, xo); (ii) if y is a solution of (3.12) defined in (a, p), then (a, p) C 1 (to, xo) and y(t) = x(t, to, xo) for every t E (a, p); (lli) if I(to, xo) = (r, a), then either a = r or lim Ix(t, to, xo) I = r; t-V (iv) for every (to,xo) E R x B(0,r), them exists a 6 > 0 such that Iso - toI + :
Iyo - xoI < 6 implies that 1(to, xo) C 1(so, yo).
§3.1
THE BROUWER FIXED POINT THEOREM
53
Definition 3.13 Assume that there exist T > 0, b : R - R, such that = f (0, t), 0(t + T) = \O(t),
for every t E R and for some constant A 0 0. Then 0 is called a Floquet solution
of (3.12). Moreover, if A = 1, 0 is said to be periodic of the first kind and if A 9k 1, 0, 0 is said to be periodic of the second kind.
The task ahead will be to prove that if f (x, ) is periodic for every x E RN and if Ixo12 is small enough, then there exists a Floquet solution for the initial value problem
{ ¢ = f(o,t) 0(0) =
xo.
Theorem 3.14 Let f : RN x R
RN be a locally Lipschitz function such that t) is positively homogeneous of degree one (i.e. f (ryx, t) = -yf(x, t) for every ry > 0 and for every (x, t) E RN x R). Assume that f (x, ) is periodic, of period T > 0, for every x E RN. Then there exists a 6 > 0 such that for every 0 < a < 6 there exists yQ E B(0, a) such that
f
1 x = f(x,t)
ll x(0) =
y,,,
admits a Floquet solution. Futhermore, if N is odd, then ya can be taken on 8B(0, a). Proof First we suppose that N is odd. As f is continuous and positively homogeneous of degree one, we must have f (O, t) = 0 for every t E R. Therefore, by Theorem 3.12 (ii), x(t) - 0 for all t E R is the unique solution of the equation
i = f (x, t) X(0) =
0.
Let us denote by x(t, 0, c) the solution of the equation f i =f(x,t)
l x(0) =
(E)
c.
By Theorem 3.12 (iv) there exists 6 > 0 such that Ic12 < 6 implies that the equation (E) admits a unique solution defined in R. Fix 0 < a < 6 and define F(c) := x(T, 0, c).
By Theorem 3.12 F is continuous and by the uniqueness of the solutions of (E),
F(c) = 0 if and only if c = 0. Therefore, 0 it F(8B(0, a)) and, by Proposition 3.9, there are C. E 8B(0, a), A > 0, such that T(co) = Ac.. For t E R set 0(t) := Ax(t, 0, cQ)
54 SOME APPLICATIONS OF THE DEGREE THEORY TO TOPOLOGY Ch. 3 and
+/i(t) := x(t +T,0,c0). It suffices to prove that 0(t) =1G(t) for every t E R. Indeed, fi(t)
0, c0 )
= Af(x(t,0,C0,t)
=f
o, C.), t)
= f(Ot),t) and 0(0) = Axx(O,0,c0) = .ca. Also,
*(t) = z(t + T, 0, c o = f(x(t + T, 0,c0),t)
= f (x(t + T, 0, c,), t + T) = f (,/,(t), t + T) = f (+G(t), t) and
0(0) = x(T, O, c0) = Tca, = Ac0.
By the uniqueness of the solution of (E), we conclude that 4(t) = V1 (t) for every
tER.
Finally, if N is even we define g : RN+1 _, RN+l by 9(x, xN+1, t) :_ (f (x, t), xN+1) Since N + 1 is odd, there exists (y, yN+1) E R1+1, I (y, YN+ 1) 12 = a, for which s 1N+1 x(O)
( XN+1(0))
= _ 9(x, xN+1, t) y = (VN+1)
admits a Floquet solution. We conclude that (E) admits a Floquet solution for some initial value y, (y12 < a.
3.2 Odd mappings The aim of this section is to show that the degree of a continuous, odd function RN is an odd number (see Theorem 3.23). As a consequence of this 0 : fl result, if 0: B(0,1) C RN -, RN is an odd function then the equation O(x) = 0 admits a solution in B(0,1). Indeed, since d(O, B(0,1), 0) is odd, we obtain d(m, B(0,1), 0) 34 0
and using Theorem 2.1 we conclude that the equation m(x) = 0 admits a solution.
ODD MAPPINGS
§3.2
55
Definition 3.15 The set D C RN is said to be symmetric if for every x E D we have -x E D. The mapping : D -' RM is said to be an odd mapping if O(x) = -0(-x) for every x E D. Before stating the main result of this section, the Odd Mapping Theorem (Theorem 3.23), following the scheme of Schwartz (1969) we first make some remarks on the Tietze Extension Theorem (see Theorem 1.15) and we prove some lemmas which will be used in the sequel.
Remark 3.16 We may deduce from Tietze's Extension Theorem (Theorem 1.15) that if K, L C RN are two compact sets such that K C L and if f : K RM is a continuous function, then there is a continuous extension g : L -+ RN
of f such that If IK = I9IL = sup{Ig;IL : i =1, ... , M}.
This is Exercise 3.2.
Proposition 3.17 Let M > N be two integers, let K C RN be a compact set, and let 0 E Cl(K)M. Then 4(K) has measure zero in RM.
Proof We identify K with K' C RM and
: K -, RM with
: K' - RM,
where
K':={(x,0,...,0)EKxRM-N} and +)(x,0,...,0)
O(x)-
Since J ,(y) = 0 for every y E K', by Sard's Lemma 1.4 we have G"'(O(K)) = GM(tP(K')) = 0.
Theorem 3.18 Let K C L C RN be two compact, nonempty sets, let M > N be two integer numbers, and let 0 E C(K)M be a function such that 0 is nowhere zero. Then m can be extended to a mapping X E C(L)M which is nowhere zero.
Proof Let c := min{I0(x)I : x E K} > 0 and choose 0 < e < z. Claim 1. We claim that there exists IP E C'(L)M such that t/, is nowhere zero and 10(x) - O(x)I < e for every x E K. Indeed, by Tietze's Extension Theorem (see Theorem 1.15), there exists 01 E C(L)M such that 011K = 0-
Let 02 E Cl(L)M be such that I02(x) - 01(x)I < 2 for every x E L. By Proposition 3.17,
56 SOME APPLICATIONS OF THE DEGREE THEORY TO TOPOLOGY Ch. 3
GM(02(L)) = 0
and so there exists p ¢ 02(L) such that IpI < 2. Let
02(x) - p
O(x)
Clearly, ip is nowhere zero, 0 E Cl(L)M, and, for every x E K, I+G(x) - 02(x)I + I02(x) - m(x)1 < 2 + 2 = e.
Claim 2. There exists 4D E C(L)M such that I'(x) - O(x)I < e for every x E K for every x E L. Indeed, letz
and I4D(x)I >
1
rl(t)
2t
t>
t
2, then
t(x) = t/,(x) and so I1(x)I = k&(x)I > Z. Therefore, I4'(x)I > z for every x E L. Also, for every x E K, we have
10(0
10(0 - I0(x) - +Gx)I > c - 2 =
c 2'
which implies that fi(x) _ O(x) and so I -OW - O(x)1
0. This implies that m : = m __
61
which yields a contradiction. Therefore, H(x, t) i4 0 for every x E 8D and for every t E [0, 11, and we conclude that ,
d(O, D, 0) = d(t,i, D, 0).
Due to the first step, in the sequel we assume, in addition, that 0 is an odd mapping. Second step. We prove that we can assume, without loss of generality, that 0(x)
x in a neighbourhood of 0. Let e > 0 be such that U := B(0, e) C D and set
R'' by
DI := D \ U. We define 01 : U U OD
x
1fi(x)
xEU x E 8D.
Then ¢1 is a continuous function,
8D1 =ODu8U =ODu8B(0,e) and 01 is an odd mapping. Let 42 := 01IaD, .
As ¢2 is nowhere zero, by Lemma 3.21 there exists a continuous extension of 02,
03 Dl - RN, such that 0 is odd and nowhere zero on D1 n RN-1. Define 03(x)
04(x) := Ix
x E D1 x E U.
We show that 04 is well defined. Indeed, if x E U n Dl = U n D, then x E 8U and 03(x) _ 42(x) = x. It is also easy to show that 04 is an odd, continuous mapping. For every x E 8D, we have x E 8D1 and 04(x) = 03(x) = 02(x) = 01(x) = 0(x)
Therefore, d(04, D, 0) = d(0, D, 0), 0 ¢ 04(OD) and m z 96 m
for every
x E OD.
Third step. We show that d(03i D1,0) is an even number. Let K := D1 n RN-1. Then 0 11 03(K) and it is easy to see that K is a compact set included in D1. By the excision property of the degree (see Theorem 2.7 (2)) we have d(03, Dl, 0) = d(03, D1 \ K, 0) = d(03, Dl , uDi, 0),
where Di := D1 n R+ and Di := D1 n R. By the decomposition property of the degree (see Theorem 2.7 (1)) we deduce that
62 SOME APPLICATIONS OF THE DEGREE THEORY TO TOPOLOGY Ch. 3 d(03, D1, 0) = d(03, Dl , 0) + d(03, Di , 0)
and, using the invariance of the degree under a C' change of variables (see Theorem 1.20), we have that
d(43, Di , 0) = d((-I) o
o (-I), -I (Di ), -1(0)) = d(03, Di , 0).
Therefore, d(, Dl, 0) = 2d(¢3i Di , 0). Fourth step. We conclude that d(04, D, 0) is an odd number. Since 018D = 04 18D,
04laintu = I, 04ID1 = 031D1, and D \ U = Dl, we have by the excision and decomposition properties of the degree (see Theorem 2.7), d(O, D, 0) = d(04i D, 0) = d(04, D \ 8B(0, e), 0) = d(q54, int U U (D \ U), 0) = d(04,int U,0) + d(¢4i D \ U,0)
= d(¢4, int U U (D \ U), 0) = 1 + d(04, D \ U, 0) = 1 + d(cb3, D1, 0)
= 1 + 2d(03, Di , 0).
Theorem 3.25 [The Borsuk Fixed Point Theorem] Let D C RN be an open, bounded, symmetric set containing 0 and let 0 : 8D - RM be a continuous function, M < N. Then there exists x E 8D such that O(x) = m(-x).
Proof We identify RM with {x = (xl, ... , xN) E RN : xN+1 = ... xN = 0}. Define >G(x) := m(x) - 0(-x) for x E 8D. We must show that there exists x E OD
such that O(x) = 0. Assume, on the contrary, that tb(x) 96 0 for every x E OD. Since 10 : OD RM is a continuous, odd, nowhere zero mapping and
OW # V,(-x) 10W I
10(-x)1
for every x E 8D,
we obtain by Theorem 3.23 that d(t,, D, 0) is an odd number. By Theorem 2.3 there exists co > 0 such that, for every 0 < e < co, d(t,b, D, pE) = d(tl,, D, 0),
where P, = (0,...,U, e).
Therefore, d(o, D, pE) 96 0 and, by Theorem 2.1, pE E O(D) C RM for every 0 < e < co, which contradicts p. ¢ RM.
ODD MAPPINGS
§3.2
63
The following corollary is also known as the Borsuk-Ulam Theorem.
Corollary 3.26 [Borsuk-Ulam Theorem) Let SN C RN+1 be the N-sphere and let ¢ : SN -. RN be a continuous mapping. Then there exists x E SN such that 0(-x) = O(x). Proof The result follows immediately from Theorem 3.25, where we set D:= B(0,1) C RN+1 0
Corollary 3.27 Let SN be covered by N closed subsets A,,-, AN. Then some At must contain a pair of antipodal points, i.e. there exists a set A. and there
existsxEA, such that -x E At. Proof For x E SN we define f,(x) := p(x, A1) and
f(x) :_ (fl(x)...., fN(x)), and we recall that p(x, y) = max{ Ixj - yj I : j = 1, ... , N}. Clearly, the function f : SN - RN is continuous and by the Borsuk-Ulam Theorem (Corollary 3.26) there exists l; E SN such that f (l;) = f (-l;). Since {As}i cover SN, there exists
some i E {1, ... , N} such that l; E At and so f,(1;) = 0 = f,(-l;). Since At is
0
closed, we conclude that l;, -£ E At.
Theorem 3.28 [The Ham-Sandwich Theorem] Let Xl,... , XN C RN be N bounded, measurable sets. Then there exists an (N-1)-hyperplane Y C RN that bisects each of the sets Xl,... , XN.
Proof Without loss of generality, we assume that GN(X,) > 0,
i =1,...,N.
(3.1)
Recall that an (N-1)-hyperplane llb,Q C RN has the form 11b.a
l = {(x1,...,XN) E RN : xlbl +... +ZNbN = a}
where b E RN, b 0, and a E R. We identify RN with {(yl, ... , yN+l) E RN+1 yN+1 = 0}. Let c = (0, ... , 0,1) E RN} 1 and define f : SN RN by
f :_(f1,...,fN), fi(x):_.CN({yEXi : (y-c)-x>0})We show that f, is continuous. Fix x ESN and let {x,.} C RN be a sequence such that x,. -s- x when n tends to infinity. By Lebesgue's Dominated Convergence Theorem and since X, is a bounded set, we have X(yEX, :
X{yEx.: (y-c)-x>O}
64 SOME APPLICATIONS OF THE DEGREE THEORY TO TOPOLOGY Ch. 3
almost everywhere and so fi(x,.) -- f,(x). By the Borsuk-Ulam Theorem there exists a E SN such that
f(a) = f(-a), i.e. for every i = 1, ... , N,
CN({yEXi :
(3.2)
We claim that there exists j E { 1, ... , N} such that a. 0 0. Assume, on the contrary, that a1 = ... = aN = 0. Since a ESN we have aN+1 = ±1 and (3.2) implies that
CN({yEX, :
yN+1-1>0})=,CN({yEX;
YN+1 - 1 < 0))) .
As Xi C {y E RN+I : YN+1 = 0}, we obtain CN(X1) = 0, contradicting (3.1). Now (3.2) is equivalent to
CN({(yl,...,yN)EX; :
a y = a1y1 +... + aNYN. Setting Y:= {(y1,...,YN) we conclude that Y bisects each of X,.... , XN.
3.3 The Jordan Separation Theorem In R2 the Jordan Separation Theorem asserts that if C is a Jordan curve, then R2 \ C is a union of two connected, disjoint, open sets D1 and D2 such that D1 is bounded and D2 is unbounded. In this section we present a generalization of this theorem in RN. We recall that K, L C RN are said to be homeomorphic if there exists a bijection h : K - L such that h and h-1 are both continuous.
Theorem 3.29 [Jordan Separation Theorem] Let K, L C RN be two compact sets such that K and L are homeomorphic. Then either K` and L` have the same finite number of connected components or both have countably infinitely many connected components.
Proof Let {Ai : i E I) be the family of connected components of K`. The sets Di are mutually disjoint and, since K is closed, each Ai is open. Therefore, there are at most countably infinitely many Di, since each of them contains a point with rational co-ordinates. We may write I C N and assume without loss
of generality that if i E I and i > 2, then i - 1 E I. As K is a bounded set, exactly one of the connected components is unbounded, say Oo. By the same argument, let {D,: r E R} be the set of connected components of L` such that
R C N, and R has the property that r E R and r > 2 implies that r- 1 E R.
§3.3
THE JORDAN SEPARATION THEOREM
65
Let Do be the unique, unbounded, connected component. Let h : K -+ L be a homeomorphism. By the Tietze Extension Theorem (Theorem 1.15) there exist RN, and a continuous extension of a continuous extension of h, 0 : RN h-1,10: RN RN. We show that, for every i, j > 1, J R1
bij = E d(4, Di, Dk)d(iP, D,, ,&j), k=1 III
bij = L d(O, Ak, Di)d(,G, Dj, Ok) k=1
and conclude the theorem by means of an elementary argument of linear algebra. Here 6,, denotes the Kronecker symbol, equal to zero if i 54 j and one if i 54 j. Fix j E N and let {Gjj:I E Al be the set of connected components of (O(80,))c,
where A C N and A has the property that l E A and l> 2 implies l- 1 E A. Since 8,&j C K and is bounded, then O(80j) is a compact set and so exactly one of the Gi is unbounded, say Go. We have
K` = UjEI,, L` = U.ERDr, (4(&))C = UIELG. Claim 1. For every r E N, there exists I E N such that Dr C Gl,.
Indeed, fixing r E N, we observe that 80j C K implies that O(80j) C O(K) = L and so Lc C ¢(80j)c. Therefore, Dr C L° C o(8e,j)c and, since Dr is connected, there exists I E N such that Dr C G, . The collection {Dr} may be relabelled {D1 k} in such a way that
Uo =Do UDo.1UDo,2U...CGo ,
U1 =Di.1UDi.2u...CGil. Claim 2. d(tli, G) j', p) = F,+- d(t&, Di.k, p) for p E Di and I > 1. First we prove that d(ip, G31, p) is well defined. It suffices to prove that 8G)I C
L. Let x E 8G{ and assume that x it L. We obtain that x E A. for a suitable r and so, since Dr C Gi(r), we conclude that x E Gl(r). Thus Gi f1G;(r) 0 0 and so I = 1(r). Therefore, we have x E 8Gi f1 GJ,, which yields a contradiction. Hence 8G! C L and
0(8Gi) C O(L) = K. Finally, P .E Di implies that p ¢ K, which, in turn, implies that p ¢ b,(8G'i) and so d(>1, G21, p) is well defined.
Since 8DI k = 8D,. for a suitable r and LC = urEKD,., we deduce that MY, A; C L. Hence d(O, D%k, p) is well defined.
66 SOME APPLICATIONS OF THE DEGREE THEORY TO TOPOLOGY Ch. 3
Consider the compact set M := Gi \Ui and fix x E M. Then either x E 8GI' or x E GJ, \U'; therefore x ¢ Dr for every r E N. Thus X E L and t/i(x) E t6(L) = K.
Moreover, if p E Di, then p ¢ t/i(M) and by the excision property and the decomposition property of the domain (see Theorem 2.7) we obtain that +oo
d(t,&,GI,p) =d(t,b,G)l \M,p) =d(i,b,Uf,p) = Ed(V,,V1.k,p)k=1 +oo
Claim 3. We claim that bij = 1: d(1G, Dk, 0,)d(cb, 03, Dk) for i, j > 1. k=1
Let us recall that d(tk, Dk, ii) = d(cb, Ak, p) for all p E Ai and these are well
defined since Ai C K° = (cb(L))`. Fixing p E 0, and using the multiplication theorem, we have +00
d(1,b,G{,p)d(tu,03,G1)
d(tGo0,0j,p) _ l=1
and the summation is finite. Since D; k C Gi , we deduce that d(o, A., Cl) _ d(q5, Aj, D1 k) for every k, which, together with claim 2, yields +oo
d(tfio0,Oj,p) _ Ed(tI,,Gi,p)d(0,03,Gi) t= t
+00 +00
_
d(O, Dj',k, p)d(,0, A), D11,01=1
k=1
Recall that d(t/', 03, D03k) = d(o, Oj, Go) = 0 because Go is unbounded and so +00 +00 d(t/i o O, Aj, p)
_ E E d(0, D1,k1 &(0, 03, Pf",k) 1=0 k=1 +oo
_ E d(1G, Dr, p)d(O, I j, Dr), r=1
since {Dr : r > 1} _ {Df.k : l > 0, k > 1}. The last equality can be written as +00
d(tb o 0, 03, Di) = E d(O, Dr, Di)d(0, A3, Dr)r=1
Using the fact that 803 c K and tG o O(x) = x for every x E K, we have d(tG o 0, Aj, Di) = d(I, Aj, A,) = bij;
THE JORDAN SEPARATION THEOREM
§3.3 thus
67
too
bij _
fsi
d(i, Dr, ti)d(ct, ij, Dr).
(3.3)
Using a similar argument we have that +oo
bi, = E d(?,b, D1, Or)d(t, Or, Dj).
r-i
(3.4)
Claim 4. {Dr : r > 1,r E R} and {0j
1, j E I} are in bijection. We divide the proof of claim 4 in three cases.
First case. R < +oo and I < +oo. Let A E RRx I be defined by ar, := d(q5, Ar, D,)
and let B E RIxR be defined by
br, := d(', Dr, 0,). We have ABERRxRand
(AB)ij _
aikbkj = k-1
d(O, Di, Dk)d(tb, Dk, Ai) = bij k=1
hence AB is the identity matrix of RRxR and so it has rank R. We observe that
R = rank (AB) < rank A < min{R, I} < I. Similarly, BA E RI x I is the identity matrix of R"',
I = rank (BA) < rank B < min{R,1} S R. and we conclude that R = I. Second case. R = +oo and I = +oo. In this case, it is obvious that {D, : r E R} and {Di : i E I} are in bijection. Third case. R = +oo and I < +oo. Let
X = R[x]
be the set of real polynomials with the basis 1, x, x2, x3, .... Set
Y := span {1,x,x2,...,x1} and define two linear applications f : Y - X and g : X -+ Y by
68 SOME APPLICATIONS OF THE DEGREE THEORY TO TOPOLOGY Ch. 3 00
f(x') := 1: ar,2r,j = 1,...,I r=1
and
9(x1) :_00>ar,xr, j = 1,...,00. r=1
Since are = 0, except for finitely many r E N, j E {1, ..., I} , f and g are well defined. By (3.3) and (3.4), we obtain that
fo9=Ix and
9of =Iy, where IX stands for the identity matrix of X and Iy stands for the identity matrix of X. Hence
+oo = rank (fog) < rank f < I, which yields a contradiction.
Fourth case. R < +oo and I = +oo. The proof of this case is identical to the proof of the third case. We give an application of the Jordan Separation Theorem.
Theorem 3.30 (Invariance of Domain) Let D C RN be an open set and let 0: D - RN be a continuous, injective function. Then 0(D) is an open set.
Proof We observe that for every p E D there exists r(p) > 0 such that B(p,r(p)) CC D. Thus 0(D) =pEo O(B(p, r(p)))
To show that 0(D) is an open set, it suffices to show that Q(B(p, r(p))) is an open set for every p E D. Fix p E D, and set B := B(p, r(p)). Since 8B is a compact
set and 0 is injective, we obtain that 0: 8B - ¢(8B) is a homeomorphism and, by the Jordan Separation Theorem, (O(8B))C has two connected components, 01 open and bounded and A2 open and unbounded. We claim that 0(B) = 01. Indeed, as B is a compact set and 0 is injective, we have that 101g : B -. 0(B) is a homeomorphism and so, by the Jordan Separation Theorem, we conclude that (0(B))` is an unbounded, connected set.
Considering, in addition, the fact that (0(B))° C (O(8B))c, we deduce that either (O(B))C C Al or (003))c C O2. Since 01 is bounded and (0(B))° is not, we must have (O(B))` C 02.
§3.3
THE JORDAN SEPARATION THEOREM
69
This implies that Al U ¢(aB) = A2 CO(B) U 0(8B) and so, as 46 is an injection, we obtain that
01 C O(B).
(3.5)
Similarly, O(B) is a connected, bounded set, O(B) C (O(8B))`, and we conclude
that ¢(B) C 01 or 4(B) C 02, which, together with (3.5), yields
O(B) = 01.
0
We conclude that 0(B) is an open set.
Corollary 3.31 Let M < N be two integer numbers. Then there is no injective mapping 0: RN _ RM. Proof Assume that there exists an injective mapping 0 : RN - RM and define
tji:RN -RNby
'+)(x) :_ (0(2), a)
X E RN ,
where a = (0,. .. , 0) E RN-M. Then 0 is injective and, by Theorem 3.30, O(RN) = RM x {a} is an open set of RN. This yields a contradiction.
O
Corollary 3.32 Let N < M be two integer numbers and let 0 : RN -, RM be an injective mapping. Then (O(RN))` is dense in RM.
The proof of this result uses Baire's Theorem. We recall its statement and refer the reader to Eisenberg (1974) for a proof.
Theorem 3.33 [Baire's Theorem] Let X be a complete, metric space and let {U;, i E N} be a collection of open, dense subsets of X. Then l$ENUi is dense in X.
Proof of Corollary 3.32. We first prove that (O(Ik ))` is dense in RM, where Ik := [-k, k].
Assume, on the contrary, that there exist a E RM and r > 0 such that B(a, r) C O(I'). Setting K := 0-1(B(a,r)), we observe that K is a compact set and so, since 0 is injective,
9:K-B(a,r) is a homeomorphism, where 9:= 01K-
Then
70 SOME APPLICATIONS OF THE DEGREE THEORY TO TOPOLOGY Ch. 3
g- 1:B(a,r)CRM_KCRN is a homeomorphism, contradicting the Borsuk-Ulam Theorem (see Corollary 3.26).
Next we show that (O(RN))` is dense in RM. IN, we obtain Since RN = UkENIk (O(RN))c -kEN (O(Ik ))c.
We observe that IN are compact sets, 0 is continuous, and so O(IN) are compact sets. Hence (0(IF))` are open sets, each (0(IN))` is dense in Of, and thus by Baire's Theorem nkeN(0(Ik ))` is dense in 0
P.
Remark 3.34 Peano's Theorem states that there exists a continuous surjection f : [0, 1] -+ [0, 11 x [0, 1]. We refer the reader to Eisenberg (1974, pp.367-370) or Kuratowsky (1966, pp. 150-151). Actually, it is possible to prove that, for every N, M > 1, there exists a continuous surjection 0 : RN - RM (see Exercise 3.3).
Theorem 3.35 Let D C RN be an open, bounded set and let ¢ E C(D)N be an injective mapping. Then for every p E 0(D) we have
d(0,D,p) = ±1. Proof By Theorem 3.30, 0(D) is an open set and O-' : 0(D) D is a homeomorphism. Let p E 0(D). Then there exists r > 0 such that B := B(p,r) CC 0(D) and we set
A:= D \ 0-' (8B). 0(D) and We want to apply the Multiplication Theorem to 0 o O-' : B : j E J} be the countable family of the connected 0-1 : B .- D. Let {Aj components of A. We have
pit 0o0-'(8B)=8B and also,
p g O(OD).
Therefore, by the Multiplication Theorem (Theorem 2.10), +00
1 =d(I,B,p)
=d(0o0-1,B,p)
=
E,d(0,Di,p)d(0-1, B,0:).
(3.6)
i=1
Also, by the Jordan Separation Theorem, 0-'(8B)° has two open connected components D1 and D2 such that Dl is bounded and D2 is unbounded.
EXERCISES
§3.4
71
We claim that there exists io E J such that D1 = 0,0. Indeed, 0 = D \ /-1(8B) C 4' 1(8B)` and, by an argument similar to that of the proof of Theorem 3.30, we have that
D1 = 0-'(B) C D \ 0-'(8B) = A= UjEjAj Therefore, D1 C 0;o for some it, E J and as
A,,cicr-1(8B)`cD1uD2, we have 0,o C D1 and we conclude that Aio = D1. Hence A j C D2 for every i 31 io and so
d(O-1, B, 0;) = d(O-1, B, D2) = 0 for every i 0 it,, which, together with (3.6), yields 1 = d(O, Ob,p)d(O-1, B, Ob)
and so d(4, Dio, p) = ±1.
(3.7)
Now it suffices to show that d(O, i{o, p) = d(O, D, p). Let
K:=D\A,o. Since d(O, A,o, p) # 0, by Theorem 2.1 we deduce that p E O(A,o) and, as 0is injective, we deduce that
pV Using the excision property of the degree (see Theorem 2.7), we have d(O, O 0, p) = d(O, D \ K, p) = d(O, D, p),
which, together with (3.7), yields d(o, D, p) = t1.
0 3.4
Exercises
Exercise 3.1 [Perron-Frobenius Theorem] Let A = (ai j) be an N x N matrix such that a;, > 0 for all i, j. Prove that there exist A > 0, z 96 0 such that x, > 0 for every i and Ax = Ax.
72 SOME APPLICATIONS OF THE DEGREE THEORY TO TOPOLOGY Ch. 3
Solution 3.1.Let D:_Ix ERN:xi>O,i=1,...,N, 1NIxi=1}.If Ax = 0 for some x E D, then it suffices to set A = 0. Assume that Ax i4 0 for every x E D. Then Nj(Ax)j > a for everyx E D and for some a > 0, and the function f : D R defined by f (x) :=
EN1(Ax)i
is continuous in D. It is clear that N
E fi(x) i=1 and A(x) > 0,
for every x E D. Hence,
f (D) C D. By the Brouwer Fixed Point Theorem (see Corollary 3.8), there exists xo E D
such that f(xo) = xo. Setting \:= EN 1(Axo)i we obtain that Axo = Axo.
A generalization of this theorem can be found in Varga (1962).
Exercise 3.2 Let K, L C RN be two compact sets such that K C L. Assume that f : K -+ RM is a continuous function. Prove that there exists a continuous function 9 : L - RM which coincides with f on K such that IfIK = 191L = suP{I9jIL : i = 1, ... , M}. Solution 3.2. Since K is a compact set and f : K - RM is continuous, we deduce that f is bounded. Let I fIIK = sup{I fi(x)I : x E K). We recall that IfIK = sup{Ifil : i = 1, ... , M}. By Tietze's Extension Theorem, for each i = 1,... , m we obtain the existence of a continuous function gi : RN -+ R such
that inf g, = inf fi. 9i I K = fi, sup 9i = sup fi, and xEL xEK xEL
xEK
We have 191M>_191K=IfIK.Conversely 191L=max{I9i3L: i=1,...,M}= I9io IL for some io E {1,. .. , M} and, as L is a compact set, we obtain that I9io IL = I9io (.t) I for some z E L. There exist a, b E K such that fio(a) = minXEKfio(x) :5 gio(x) !5 m
fio(x) = f.0(b)
and either I9b(2)I 0, there exists a continuous surjection fk : [2k - 2,2k - 1] - [-k, k] x [-k, k]. Define gk [2k - 1, 2k] - R2 by gk(2k - 1 + t) := (1 - t) fk(2k - 1) + t fk+1(2k), t E [0, 1].
We observe that gk is continuous and defining 0: R - R2 as
0(x) :=
f, (0)
x < 0
fk(x)
x E [2k - 2,2k - 11 x E [2k - 1,2k),
9k(x)
then ¢ : R - R2 is a continuous surjection. We easily deduce that, for every N, M > 1, there exists a continuous surjection 0 : RN -+ RM.
Exercise 3.4 Let f : SN
SN be a continuous function such that f (x) 34 -x for every x E SN. Show that, if N is even, then f has a fixed point on SN. Solution 3.4. By Tietze's Extension Theorem (Theorem 1.15) there exists a continuous function F : RN+1 - RN+1 such that
F(x) = f (x) 0 -x for every x ESN. By Proposition 3.9, and as N + 1 is odd and F(SN) C SN, we conclude that there exist A > 0 and xo E SN such that F(xo) = Axo
f (xo) = Axo.
Since IxoI = 1 = If (xo)I and f (xo) 36 -xo, we must have f(xo) = xo.
4
MEASURE THEORY AND SOBOLEV SPACES In this chapter we recall some properties of Sobolev functions and measure theory. In order to measure the sets on which a Sobolev function is continuous, we introduce the notion of Hausdorff measures and p-capacities. For a detailed description of these topics, we refer the reader to Evans and Gariepy (1992) and Ziemer (1989). The Hausdorff measures were first introduced by Caratheodory (1914). He developed only the Hausdorff linear measure in RN and indicated how the k-dimensional measure could be defined for k E N. Later, motivated by the study of the dimension of the Cantor ternary set, Hausdorff (1919) developed the theory of the k-dimensional measure.
4.1 Review of measure theory We recall some definitions and well-known results used in measure theory, such as the Riesz Representation Theorem, the Radon-Nikodym Differentiation Theorem and Vitali's Covering Theorem. For more details and for the proofs of the results presented in this section, we refer the reader to Evans and Gariepy (1992).
Definition 4.1 (i) We say that I C RN is a closed N-interval if there exist a, < b;, i, ... , N, such that I = [a1, bi) x ... x [aN, bN].
We set m(I) :_ (b1 - al) ... (bN - aN). (ii) We define the Lebesgue outer measure GN(E) of a set E C RN by 00
GN(E)
inf { E m(Ik) : E 1k=1
CkU
Ik, Ik closed N-interval }
.
JJJ
Definition 4.2 Let X C RN be a nonempty set, let P(X) be the collection of the subsets of X, and let µ : P(X) -+ [0, oo).
(i) u is said to be an outer measure if 00
µ(A,)
µ(o) = 0, p(,U A.) O for all r > 0
p(B(x,r)) = Ofor somer > 0
and
Dµv(x)
lim inf
v(B(x, r))
r-O+ p(B(x, r))
p(B(x, r)) > 0 for all r > 0
u(B(x,r)) = Ofor somer > 0.
+00
Definition 4.12 Let p and v be two Radon measures on RN and let x E RN. If 5m-v(x) = D,v(x) < +oo, then we say that v is differentiable with respect to p and we write Dµv(x) = Dµv(x) = Dµv(x).
_
D,v(x) is called the derivative of v with respect to p.
Definition 4.13 Let p be a measure on RN. A function f : RN [-oo + oo) is said to be p-measurable if f -1((-oo, a)) is p-measurable for every a E R. When p is the Lebesgue measure, a p-measurable function is said to be measurable.
Definition 4.14 Let p be a measure on RN, let 1 < p:5 +oo, and let f : RN [-oo + oo) be a p -measurable function. We say that f E LP(RN, p) if /RN if I9 du < +oo for p < +oo,
and if p = +oo there exists M E R such that
If(x)I <M for ua.e.xERN. The following is also known as the Radon-Nikodym Theorem.
Theorem 4.15 Let p and v be two Radon measures on RN. Then for p almost every x E RN, we have that DP(x) exists and Dµv(x) < oo. lrthennore, Dµv is p-measurable.
REVIEW OF MEASURE THEORY
§4.1
77
Definition 4.16 Let p and v be two measures on RN. (i) We say that v is absolutely continuous with respect to µ, and we write
if p(A) = 0 implies v(A) = 0 for every A C RN. (ii) We say that v and u are mutually singular, and we write
vlp, if there exists a Bonel set B C RN such that
v(B) = p(RN \ B).
Theorem 4.17 [Differentiation Theorem for Radon measures) Let p and N. Then v = va + v where v be two Radon measures on RN va « µ, v3 .1. {A. Moreover,
D,,v = D,,va, D,,v, = 0 p a.e. and
v(A) = j D,,vdp+v,(A) for every Bonel set A C RN.
Theorem 4.18 [Lebesgue-Besicovitch Differentiation Theorem] Let p be a Radon measure on RN, 1 < p < oo, and let f E L ;c(RN, µ). Then /1 l µ(B(x, iO e))
B(s,c)
If (y) - f (x)IP dµ(y) = 0
(4.1)
for p a.e. x E RN. Definition 4.19 A point x for which (4.1) holds is called a p-Lebesgue point of f with respect to p. We say that a 1-Lebesgue point off with respect to GN is a Lebesgue point of f.
Theorem 4.20 [R.iesz Representation Theorem] Let L : Cc(RN)M - R be a linear functional such that
sup{L(f) : f E CC(RN)M, IfI2 < 1,spt(f) C K} < 00 for each compact set K C RN. Then there exists a Radon measure p on RN and a p measurable function o : RN RM such that (i) Io(x)I2 = 1 for p a.e. x E RN, (ii) L(f) = f R. f o dp, for all f E CC(RN)M.
Ch. 4
MEASURE THEORY AND SOBOLEV SPACES
78
4.2 Hausdorff measures The Hausdorff measure H'(E) of a set E C R2 was introduced for the purpose of generalizing the notion of length in the case where E is a smooth curve. As a first approximation, we define 00
inf { diam(A;) : E C$EN A, } 1.-
.
JJ
It is easy to see that if N = 2 and if E:= {(t, sin 1) : t E (0,1] }, then A(E) < oo, while the length of E is infinite. This is due to the fact that in the definition of A(E) the sets A; are not forced to follow the geometry of E. In view of this, Hk(E) will be defined as a limit of outer measures H6 (E) which will follow the geometry of E.
Definition 4.21 Let E C RN, 0 < 6 < oo. We define
H
inf
0 a(s) (diarnCs)s 2
E c u C;diam C, < 6 iEN
JJJ
where a(s) := r +I and r(s) := f o'
e-=x'-1 dx (0
< s < oo) is the usual
Gamma function.
R.emark 4.22 If 61 < 62, then H62 (E) < H6, (E) and so lim6_o H6 (E) _ sup6>0 H68 (E) exists.
Definition 4.23 Let E C RN and 0 < s < oo. We define the s-dimensional Hausdorff measure on RN by
H'(E) = aim H6(E).
Theorem 4.24 For each 0 < s < oo, H' is a Borel measure. The proof of this result can be found in Evans and Gariepy (1992).
Theorem 4.25 (i) Ho is the counting measure. (ii) H1 = H6 =41 on R for every b > 0. (iii) HI (E) = 0 for all E C RN and for all s > N. (iv) Hs(AE) = A'H'(E) for all E C RN and for all A > 0. (v) HN(L(E)) = H'(E) for all E C RN and for all affine isometry L : RN R
.
HAUSDORFF MEASURES
§4.2
79
Proof (i) We observe that a(0) = 1. Therefore, H°({a}) = 1 for every a E RN and
H6({al,...,ak}) > k for every mutually distinct a1,. .. , ak E RN and for every positive b such that 0 < 6 < I min{Iai - a.I : i 0 j}. Hence,
H°({a1i...,ak)) = k, i.e. H° is the counting measure. (ii) Fix 6 > O and E C R. We have 00
£'(E) = inf
la, - biI : E C,EN [ai,bi]} =1 00
diam(C*) : E
= inf
CiEN
ci}
00
U Ci, diam Ci < 6 } l -1 diam (Ci) : E C iEN
< inf {
JJ
= Hb''(E) On the other hand, setting Ik := [6k, 6(k + 1)], k E Z, for every C C R we observe that diam (C n Ik) < 6 and too
E diam (C n Ik) < diem (C). k=-oo
Therefore, we have 00
G'(E) = inf
diam (Ci) : E CiEN `i=1
C,I
00
> inf
diem (Ci n Ik) If, i=1 k=-oo
EC iEN Ci I
> H6(E) and this concludes (ii).
(iii) Let Q = (0,1)N and let m > 1 be an integer. For k = (kl,... , kN) E K:=
{0,...,m-1)Nset Qk :=
Lk-m', klm 1J
x ... x [kN,kN+1]. ,n
MEASURE THEORY AND SOBOLEV SPACES
80
Ch. 4
We observe that Q
kEK
Qk and diem (Qk) _
rN-
and so la
H (Q) < >2 a(s) kE K
( \Z/
= a(s)mN-'
Letting m - +oo we deduce that
H'(Q) = 0, which yields
H'(RN) = 0. (iv) Using the property that diam (AC) = A diam (C),it is easy to verify that A'H'(C) = H'(AC) for every A > 0 and for every C C RN. (v) This follows from the fact that diam (L(C)) = diam (C) for every affine isometry L : RN - RN and for every C C RN. 0
Remark 4.26 It turns out that the equality HN(E) = GN(E)
holds for every E C RN. The proof is not trivial and it uses the isodiamettic inequality
G'(E):5 a(N)
(diam E )'
which is valid for every E C RN. We remark that E does not have to be contained in a ball of diameter diam E.
Lemma 4.27 Let E C RN, 0 < 6 < oo, and let 0 < s < oo be such that H6 (E) = 0. Then
H'(E) = 0. Proof Ifs = 0, H6 (E) = 0 implies that E = 0 and so H°(E) = 0. Now assume that s > 0 and fix e > 0. There exist sets {C,},EN such that E C U,ENC$, diam C; < 6 for every i E N and 00
diam C. 2
< J
HAUSDORFF MEASURES
§4.2
We observe that a(s) (d'
81
)' < e for every i = 1, ... , oo and so
_: l(e).
diam Ci < 2 Cads)/ Hence,
Hi'(,) (E) < e
and letting e go to zero we deduce that H'(E) = 0. Lemma 4.28 Let E C RN and let 0 < s < t < oo be two real numbers.
(i) If H'(E) < +oo, then Ht(E) = 0. (ii) If Ht(E) > 0, then H'(E) = +oo. Proof Assume that H'(E) < +oo and fix b > 0. There exist sets {C,}IEN such that E C UieNC1, diam C; < b and 00
/diam Ci 2
i=1 Since
H6(E)
00a(t)
(diam C1)t 00
a/ s(8) 2'-t 1
s while Ht(E) = +oo for every t < s. The dimension Hdim(E) is not necessarily an integer and may be any number in [0, NJ.
Proposition 4.31 Let f : RN - RM be a Lipschitz function, E C RN, and let 0:5 s < oo. Then
H'(f(E)) < K'H'(E), where K := sup { 4 _y
y, x, y E RN }
82
MEASURE THEORY AND SOBOLEV SPACES
Ch. 4
Proof Without loss of generality, assume that H'(E) < +oo and fix b, e > 0. There exist sets {C,};EN such that E C U;ENC;, diam C, < 6 for every i E N, and
00a(s) gal
(diem C;
\
2
J
k, diam B < 2 diam B B' and 9 is a countable family since any collection of disjoint open sets must be countable. Corollary 4.35 Let.F be a fine cover of A by closed balls, i.e. inf {diam B: B E .F, x E B} = 0 for every x E A, and suppose, further, that sup{diem B : B E F} < +oo. There exists a countable family Q of disjoint balls in.F such that for each finite subset {B1, ... , CF we have B') CBE9\{s U
Proof By Vitali's Covering Theorem there exists a countable family g of disjoint balls in F such that
A CBE B
CU
B.
Fixing B1, ... , B E .F, if x ¢ A\(.U B;), then, as B, are dosed and .F is a fine =1 cover of A, there exists B E F with x E B, B n Bi = 0 for i = 1, ... , n. From the claim in the proof of Vitali's Theorem, it follows that there exists B' E Q with
B'nB00;hence
BCB'.
Proof of Proposition 4.33. It is clear that to prove (4.2) it suffices to show that H'(At) = 0
Ch. 4
MEASURE THEORY AND SOBOLEV SPACES
84
for every t > 0, where
At:= xERN\E:limsup H'(B(zr r) n E) >t 1 r-o+
Fix e > 0. As H' LE is a Radon measure, we may find a compact set K C E such that H'(E\K) < E. Thus At C (RN\K) and, for fixed 6 > 0, we consider
.F:=
r) n E) {.(xr) : B(x, r) C (R N \K), O < r < 6, H'(B(z, a(s)r'
>tI
Without loss of generality, we may assume that F j4 0; otherwise At = 0 and so
H'(At) = 0. Using Vitali's Covering Theorem we write At
CiU
B,
where {B(xi,r,}°=1 C F is a disjoint family of closed balls. Then 00
Hio6(At) 5 Eo(s)(5r,)' i=1
00
J:H'(Bi n E)
tI
Since H' LE is a Radon measure, there exists an open set U containing Be such
that
HAUSDORFF MEASURES
§4.2
85
H'(U n E):5 Hs(Bt) + e and we define, for fixed 6 > 0,
.F:= 11 B(x, r) : B(x, r) C U, 0 < r < b,
H'(B(x, r) n E) > t a(s)rs
By Corollary 4.35, for each m E N there exists a countable disjoint family of balls {Bi } 1 in F such that m
Bt CsUI
B,
00
s=m+1 Bi ,
where Bi = B(x,, r,). Then m
00
H106(Bt) < Eo(s)ri + > a(s)(5ri)' i=m+1
i=1
t
M
00
tEH'(BinE)+ 5' i=1
H'(BinE) i=m+1
< H'(U n E) + L H' (i=m+1 B n E)
.
Letting m -+ +oo we deduce that H1'106 (Bt)
1. 0 Remark 4.36 Under the hypotheses of Proposition 4.33, it can be shown that lim sup
c-0
H"
e) n E) > or(s)es
1
- 2s
for H' a.e. x0 E E. It is worth noting that it is possible to have
86
Ch. 4
MEASURE THEORY AND SOBOLEV SPACES
H"(B(xo, e) fl E) 01
Then H8 (A,) = 0.
Proof Without loss of generality, we may assume that f E L' (RN), as it suffices
to prove the result for f 'k, O k E CO(RN),
0 < 1Ik < 1, V)k(x) = 1, if
x E B(0, k). By the Lebesgue-Besicovitch Differentiation Theorem (see Theorem 4.18),
lim
GN (B(x, r))
JB(x,r) I f (y) I dy =I f (x)
for £Na.e.xERNandso lim 1
r-0 r"
f
If(y)I dy = 0.
B(x,r)
Thus CN(A,) = 0. Fix e,6,a > 0, and as f E L1(RN) choose q > 0 such that JuIf(x)Idxe}. JJJ
Since A; C A, we have GN(A") = 0
and so we may find an open set U D A, such that GN(U) < ti. We set
OVERVIEW OF SOBOLEV SPACES
§4.3
87
F:= B(x,r): xEA;,0 1) }
.
1
Remark 4.54 (i) If K C RN is a compact set, then, using mollification of the characteristic function Xk, one has Capp(K) = inf 1JR N
I
Vflpdt : f E
XK1.
(ii) It is clear that if E C F C RN, then Capp(E) < Capp(F). The following Sobolev-Niremberg-Gagliardo generalization will be used to prove that Capp is an outer measure.
Lemma 4.55 Let 1 < p < N. Then there exists C = C(N, p) such that /
{LN
l dx
1
l 1/p
/PIf(x)I"
C(N,p) If N Ivf(x)'I dx
for every f E KP.
Proof We start by constructing a sequence Wn E CCO°(RN), 0 < Pn < 1, converges increasingly to 1 for a.e. x E RN, and ft) If(x)I'*
dx}
1-p/p'
If
IVVn(x)Ip(j)' dx} N
and p (y) = N. Thus, p/p' N
If(z )VWn(x)I'
dxnj
and, due to the integrability of I f Ip*, we conclude the result.
Theorem 4.56 Let 1:5 p < N. Then Capp is an outer measure on R''. Proof Let {Ek, k E N} be a sequence of subsets of R1' and
E:= U Ek. kEN
We want to show that
0
94
MEASURE THEORY AND SOBOLEV SPACES
Ch. 4
00
Capp(E) k=1
Capp(Ek)
and so, we may assume, without loss of generality, that Cap,(Ek) < +oo. kEN
Fix e > 0. For each k E N, we choose a function fk E KP such that Ek c int{fk > 1} and
JRN
IVfk(x)I2 dx < Cap,(Ek) +
2k
.
Let g := supkEN fk. We observe that E C int{g > 1} and, by Lemma 4.55,
RN
g(x)p dx = JRN su p fk (x) dx 00
0, N-1 < p < N and let E C B(0,1) C RN. Assume that for each r E I there exists a unique x,. E 8B(0, r) such that xr E E. Then Capp(E) > 0.
Proof Let f : RN -+ R be defined by f (x) := Ix12. We observe that f is a Lipschitz mapping such that If (X) - f (0I2 0.
0 Next we show how the notion of p-capacity plays an important role in the study of the continuity property for Sobolev functions.
Definition 4.60 Let f : RN -+ R be a measurable function. f is said to be p-quasicontinuous if for every e > 0 there exists an open set V C RN such that f IRN \y is continuous and Capp(V) < e.
Theorem 4.61 Suppose that f E W 1p (RN) and 1 < p < N. (i) Then exists a Bowl set E C RN such that Capp(E) = 0 and
r-o LN(B(x, r)) JB(:,r) f (y) dy = f*(x) exists for every x E RN \ E.
(ii) In addition, r
N(B(x,r))
for every x E RN \ E.
IB(z,r) 11(y) - f*(x)1p dy = 0
Ch. 4
MEASURE THEORY AND SOBOLEV SPACES
98
(iii) f' is p-quasicontinuous. The proof of this result uses the following lemma.
Lemma 4.62 Let l < p < N, f E Kp, let e > 0, and
E:=SxERN: 11
CN
((I
))
B(=.r)
f(y)dy>eforsome r>0}. JJJ
Then
Capp(E)
_
ji))
J(.) f(y)dy > E a.e. in B,')
and, setting
h:=
i = 1, ... , k, j E N},
we have h > 0. We claim that h E KP. Indeed, h E L1(RN) because, by (4.8), k
JRN iJsup[h(x)]'dx
o0
i=1 j=1 k
JRN [h(x))adx ao
CzE
IVf(x)ID dx
j=1 fB,(.
e a.e. in E and as E is an open set we deduce that
V(f)
Capp(E)!5J RN
5 p{
x)IP
(
dx
I
f
IVf(x)IPdx+
RN
f IOh(x)IPdx R
which, together with (4.9) and (4.12), yields CapP(E) N - p and so
Hdim(E) < N - p.
(4.11)
On the other hand, (4.11) does not necessarily imply that HN-P(E) < +oo, which, in turn, yields Capp(E) = 0. (4.12)
by Proposition 4.58. Therefore, (4.12) may be a stronger statement than (4.11).
(ii) Recently, Maly sharpened Theorem 4.61 by showing that the quasicontinuous representative f' of f E W1,N(el;RN) is approximately Holder continuous, except on a set of Hausdorff dimension zero, i.e. if S is the set of all points of S1 at which f' is approximately Holder continuous, then Hdim(St\S) = 0.
Actually, in view of (i), Maly's precise statement asserts that for every e > 0, p < N, there exists an open set G C RN such that Capp(G) < e and f In\a is locally Holder continuous. Here f' is said to be approximately Holder continuous at xo if for every oY E (0,11 there exists a set M such that the Lebesgue density of M at xo is 1, i.e. £ N(M n B(xo, f)) = 1 lim 0+0; CN(B(xo,f))
Ch. 4
MEASURE THEORY AND SOBOLEV SPACES
102 and
I f (y) - f (x) I < +00.
lim sup
I y - xo la
Y-xo,YEM
Proof of Theorem 4.61 Let f E WI,P(R°),1 < p < N. (i) We want to show that there exists E C RN, Capp(E) = 0 such that rl 0* GN (B(x, r)) J B(x.r)
f (=J) dfJ - f *(x)
for every x E RN \E. We define
A:=I xE RN:limsup 1 r-.e+ rn
JB(x,r)
IVf(y)Indy>0}.
By Proposition 4.37, HIV-P(A) = 0 and so, by Theorem 4.57 (v), Cap9(A) = 0. By Poincare's Inequality we have e
li oGN (
IB(x.r) I f (y) - GN(B(x r)) fB(.,r) f (z)dz
(x,r))
B
Nrf
< C lim
r-.o+ r
P */p
B(=,r)
I Vf(y) I9
dy 1} n'/P
1
C r-.O lim {l r
IP*
N-P BJ(x,r) I Vf (y) IP dy
I
= 0
(4.13)
for every x ¢ A. Due to the density of smooth functions in WI-P, for each i E N we choose fi E WI,P(RN) n COO(RN) such that
JN
I Vf (x) - Vfi(x) In dx < 2(P1
and we set
B; := {XERN.
1
CN(B(x,r))
8(x,r)
If(y)-fi(y)I dy>
for some r > 0 } .
By Lemma 4.62 and also because, due to Theorem 4.44, if f E WI,P(RN),
then 111 K, we have
P-CAPACITY
g4.4
Cap (Bi) < C
103
I V f(x) - Vfi(x) I P dx JRN
C
2(P+I)i'
Capp(Bi)S2
(4.14)
Now 1
CN(B(x, r))
J (y,r) f(y)dy-fi(x)I If (y) - GN(B(x, f (z) dz r')) JB(x,r)
< CN(B(x, r) JB(z,r)
dy
I f (z) - fi(z) I dz
1
+ CN(B(x,r) B(zr) 1
+,CN(B(x, r)) B(x,r)
I fi(z) - fi(x) I dz.
As fi is smooth, the last integral in the above inequality converges to zero as r 0+, and so, due to (4.13), we conclude that lim sup
Lz,r)f(y)dy-fi(x) I
C N(B(x,r'))
for every x ¢ (A U B,). Set
Ek:=Au (.k Bj I.
/
Then, by Theorem 4.56 and (4.14), we have +"0
Capp(Ek) < Capp(A) + ECapp(Bj) j=k +00
< CEOI j=k
and if x E RN\Ek, i, j > k, then, by (4.15),
2'
(4.15)
Ch. 4
MEASURE THEORY AND SOBOLEV SPACES
104
If.(x) - fi(x)I S lim sup r-.0+
+ lim sup
1
f (y) dy - f: (x)
1
f (y) dy - f, (x)
GN(B(x,r)) L(z,r)
GN(B(x,r)) L(x,r)
1 1 Zt+23
We conclude that f,
g in LO°(RN\Ek) and g is a continuous function.
Also, lim su r-O+
x - GN( B (x, r)) J (x,r) f (y) dy 9() 1
S 1g(X) - A(x)1 + limsup f:(x - £N(B(x, r))
JB(z,r) f (y) dy
S Ig(x) - f:(x)I + 2, , where we have used (4.15). Thus, 9(x) = rll.0+
CN(B(x, r))
J (x.r) f (y) dy
f' (x) for every x E RN\Ek. Setting E:= lkENEk, then, by Theorem 4.57,
Capp(E) < lim Capp(Ek) k+00
=0
and f' (x) exists for every x E RN\E. (ii) We claim that lim
1
r0+ GN(B(x,r)) 1B(x,r)
I f (y) - f' (x) I p dy = 0.
Indeed, for every x it E, by definition of f' we have
r11.0 J CN(B(x, r)) JB(x,r) I f (y)
- f (x) Ip dy f(z)dzlp.
< r* LN(B(x,r)) fB (x,r)
If (y)
1
- GN(B(x,r)) JB(x,r)
dy
P-CAPACITY
§4.4
1
1
+ CN(B(x,r)) lira
1
B(x,r)
CN(B(x,r)) fB(x,r)
r- O+ ,CN(B(x,r)) fB(x,r)
105
f(z)dz - f*(x)
f(z)dz - f'(x)
= 0.
(iii) It remains to prove that f' is p-quasicontinuous. Fix e > 0 and choose k large enough so that CapP(Ek) < 2. By Theorem 4.57 (i) there exists an open set U D Ek such that CapP(U) < e. Then
fi - f' in L°°(RN\U) and so f' I(RN\U) is continuous. O
There are other notions of capacity, such as Bessel capacity, linear capacity, etc., and the interested reader may find a detailed description in Hayman and Kennedy (1976), Havin and Maz'ya (1972), Stein (1970), and Ziemer (1989). With the help of these concepts one can prove the following.
Theorem 4.65 Let f E Ll(RN) be a function with compact support, spt f C
KCCR".Let
k Ix - yl0
when a E [0, 1), and let Kl C RN be a compact set such that w(x) = +oo on Kl. Then CapN_a_E(Kl) = 0 for all o > e > 0. The proof of this result may be found in Hayman and Kennedy (1976), where it is stated more generally for measures of bounded variation in place of £N.
5
PROPERTIES OF THE DEGREE FOR SOBOLEV FUNCTIONS In this chapter we present a generalization of Sard's Lemma for Sobolev functions and the following change of variables formulae. Let D C RN be open and
bounded and assume that either p > N or N - 1 < p< N and J4(x) > 0 CN a.e. x E D. Then =
ID
JRN
v(y)N(O,D,y)dy
for every v E LOO(RN) and for every 0 E WI,P(D)N, where the multiplicity function of ¢ at y E RN with respect to D is defined by
N(0,D,y):=0{xED: t5(x)=y}. A second change of variables formula is
ID
O(x)J(x) dx = j v(y)d(q, D, y) N
for every v E LOO(RN) and for every 0 E W l.P(D)N, where d(o, D, y) stands for the topological degree of ¢ with respect to D at the point y.
As an immediate consequence of the latter change of variables formula, if a sequence {0n} C WI,P(D)N converges to 0 uniformly, then the sequence {d(on, D, y)} C Z converges to d(o, D, y) and so, under some additional assumptions, UD D v o 0.(x)J#,(x) dx} converges to fD v o O(x)J,(x) dx. Given an open set D C RN, we will use the following notation: if 1 < p:5 +oo,
f E LP(D), then IIftIp(D) := fD IfIPdx, if p < 00 and
11f1100(D) := inf {M > 0 : I f (x)I < M CN a.e. x E D}
If 0 E LP(D)M, then N
I I0I IP(D) :=
EJ
and if -0 E WI-P(D)M, then 106
D
I0.IP
,
if p = oo.
§5.1
RESULTS OF WEAKLY DIFFERENTIAL MAPPINGS
107
N
I1011i,p(D) := 11011p (D) +
(I
is=1 8xj
I lp(D)p
5.1 Results of weakly differential mappings In this section we present some results relating the notion of topological degree to weak differentiability conditions. As usual, we denote by B the unit ball centred at 0 of RN with respect to the norm I.12
Lemma 5.1 Let F E C(B)N and let L be a linear mapping of RN into RN. Then
CN(F(P)) < CN(L(B)) + O(e),
where P:= {x ED : IF(x) - L(x)I < e}.
Proof Fix e > 0. If det (L) = 0, then L maps B into a hyperplane P of dimension N - 1. Let 1:= max{IL(x)I : x E B}.
Then F(P) lies inside a parallelepiped which has (N-1) sides of length less th 21 + e and the Nth side of length less than 2e. Hence,
CN(F(P)) < (21 + e)N-12e = CN (L(B)) + 0(e)
(5.1)
because CN (L(B)) = 0. If det (L) A 0, then L is a homeomorphism of RN onto RN and L-1(F(P)) C {x E RN : p(x,B) 5 e11L-1I1},
where we recall that p(x, B) = inf{Ix - ti norm of L-1 in C(8B). Thus, CN(L_1(F(P)))
:
t E B} and 11L-I II stands for the
0 and, given z E (0,1), we choose mo E N such
that Em < 6(1 - z),
for every m>mo. Claim 1. We claim that
RESULTS OF WEAKLY DIFFERENTIAL MAPPINGS
§5.1
117
IYi-Y2I2>6(1-z)hm for every yi E L(8B(xo, h,,,)) and for every y2 E L(B(xo, zhm)). Indeed,
yi = V0(xo)(zi - xo) + s(xo) for some z1 such that 1z1 - xoI = h,,, and
y2 = V (xo)(z2 - xo) + O(xo) for some z2 such/ that Iz2 - xoI < zh,,,. We have Iy1-Y2I2 = IVO(xo)(zi-z2)I > 6121 -z2I
6(Izi-xoI-Ixo-z2I) > 6(1-z)hm.
Let
H(x, t) :_ (1 - t)o(x) + tL(x), x E B(xo, hm), t E [0, 1]. Claim 2. H(8B(xo, hm), t) C RN \L(B(xo, zhm)) for every m > mo(z) and every t E [0, 1].
Assume, on the contrary, that for some t E [0, 11, z2 E B(xo, zhm) and for
some z1 E 8B(xo,hm), we have H(zi,t) = L(z2), i.e. (1 - t)4(zi) + tL(zi) _ L(z2). Then
(1 - t)('(zl) - L(zi)) = L(z2)
- L(zi)
and by Claim 1 we have b(1 - z)hm < IL(z2)
-
L(zl)I2 :5
I0(zl) - L(zl)I2
Using (5.24) and the fact that zi = xo + hma for some a E 8B(0,1), we deduce that 6(1 - z)hm < Em, which yields a contradiction. Hence,
H(8B(xo, hm), t) C RN \ L(B(xo, zhm))
for every t E 10,11 and every m > mo(z). Thus t), B(xo, hm), y) is well defined for every y E L(B(xo, zhm)). Using the fact that L is a bijection and H is a C homotopy between 0 and L, for every y E L(B(xo, zhm)) we have d(0, B(xo, hm), y) = d(L, B(xo, hm), y) = s9n JL(xo) # 0
and so we obtain (5.22). Furthermore, by Theorem 2.1 we obtain that y E O(B(xo, hm)) thus
PROPERTIES OF THE DEGREE FOR SOBOLEV FUNCTIONS
118
Ch. 5
L(B(xo, zhm)) C qS(B(Xo, hm))
for every z E (0, 1) and every m > mo(z), which, together with (5.25), yields GN(cb(B(xo,hm))) m +oo GN(B(xo, hm))
CN(L(B(xo,zhm)))
liminf
= zNIJ#(xo)I
C'(B(xo, hm))
for every z E (0, 1). Letting z go to 1 we conclude (5.23).
Theorem 5.11 Let D C RN be an open, bounded set and let 0 E C(D)N be a mapping which has a weak differential almost everywhere and has the N-property.
Assume that Jj E Lf (D). Then for every measurable set E C D we have JE
I Jm(x)I dx = j N(4), E, y) dy. N
Proof Recall that, by Theorem 5.5, N(¢, E, ) is measurable. Also, by Lemma 5.8 it suffices to show that ( I Jo(x)I dx I J#(x)I
Since
,CN(4)(B(x,
hm))) < in N(4), B(x, hm), y) dy = a
we have lim M-+00 oo
h.))
G(B(x, hm)) , IJm()I
hm)),
WEAKLY MONOTONE FUNCTIONS
§5.2
119
By the Differentiation Theorem for Radon measures, we deduce that 4)(x) > I JJ(x) I
almost everywhere in D and so JRN N(O, E, y) dy = L 4'(x)dx ? L IJm(x)I dx.
5.2 Weakly monotone functions In this section we study regularity properties for functions 0 E
sat-
isfying J,(x) > 0 Vv a.e. x E D, where D C RN is an open set. Following Gold'sthein and Vodopyanov (1977), we show that if p = N, then 0 is continuous and monotonic (see Theorems 5.14 and 5.17 ). The definition of monotone functions was introduced by Lebesgue (1907) and, heuristically, a function is said to be monotone if it satisfies certain weak maximum and minimum principles. The following class of functions was introduced by Ball (1978) in his fundamental work on nonlinear elasticity and it will be used at length here, ;,g(D) := f4) E WI.P(D)N : adj(V4)) E
L9(D)N,N,
Jm > 0 a.e. X E D}
where adj(V4)) is the matrix of the cofactors of V4), p > 1, and q >
.
9verak
(1988) proved that if p > N - 1, then the mappings of ..4 (D) are continuous except perhaps on a set of p-capacity zero and this set is empty if p = N (see Theorem 5.17 and Remark 5.18). Here, to obtain $verak's (1988) results we follow Manfredi's (1994) approach.
Definition 5.12 Let D C RN be an open set, 1:5 p < +oo and let f E Wl (D). We say that f is weakly monotone if for every Cl cc D open, bounded, connected set and for every pair of constants m < M such that
(m - f)+ E we have that
W0,P(Q)N
and (f - M)+ E W01.P(1)N,
m0 t+:-ft - 1o t0
-t t Pp l and let 0 E WI.P(D)N. Assume that Jm > 041v a.e. x E D and adj(V) E LQ(D)NxN.
Then for each i = 1, ... , N the ith component 0, is weakly monotone. In particular, if o E W1,N(D)N and if J,0 > 0 GN a.e. x E D, then 0 is continuous.
Proof Fix i = 1, ... , N, let f? CC D be an open, bounded, connected set and let m < M be a pair of constants such that (m - 0i)+ E Wo,P(c2)N and (0i - M)+ E Wo,P(n)N
Set g
(m - 0i)+ and let {or: r E N} C C°°(D) be a sequence such that
or -.0 in
WI.P(1)N.
By Exercise 1.3 and Theorem 4.42, we have N
E
k
(5.26)
(adj(V0r))ik = 0
and by Theorem 4.42 we obtain Jm dx
Jflr1(.<m}
N V'
dx = Jn(#;
E a"' (adJ(VOr))ik dx
J r kv
N r r limo
dx
Jk=1
= r llm o
r
N
/nk=1 F9
(adj(omr))ik dx
= 0.
Since JJ(x) > 0 LN a.e. X E 91, we deduce that GN({Oi < m}) = 0 and so
4,(x) > m a.e. X E ft
WEAKLY MONOTONE FUNCTIONS
§5.2
121
Using a similar argument we obtain that 0,(x) < M a.e. x E Q, thus ¢; is weakly monotone.
Recall that if X C RN is a Coo paracompact manifold of dimension r E N, then X can be covered by finitely many manifolds X; C X such that for each i there exists a C°D diffeomorphism u, : (0, 1)r - X. We say that f E W1'P(X) if and only if f ou; E W1-P((0, 1)1). Recall also that, given any ball B(xo, r) C RN, 8B(xo, r) is a C°O paracompact manifold of dimension N - 1.
Theorem 5.15 Let N-1 < p:5 N and let f E W'.P(8B(xo,r)) be a continuous function on the sphere 8B(xo, r). Then, (diam(f (8B(xo, r))))P < C(N,
p)r°+1-N
J8B(xo.r))
IVf(x)Ii dH N-',
where C(N, p) is a constant depending only on N and p.
Proof Since p > diam(8B(xo, r)) = N -1, by the Sobolev Imbedding Theorem expressed on 8B(xo, r), we have (diam(f (B(xo, r))))P < C(r, p, N) J
B(zo,r)
I Vf (x)IdHN1.
Using a rescaling argument, we conclude that
C(r,p, N) =
rP+1-NC(N,p)
In the sequel, given f E WI-P(D) for some open set D C RN, we denote by f' the p-quasicontinuous representative of f of Theorem 4.61 and by LP(8B(xo, r)) the trace operator for every r such that T,. : W 1-P(B(xo, R)) B(xo,r) CC D (see Theorem 4.50).
Theorem 5.16 Let D C RN be an open, bounded set, 1 < p < +00, f E W'.P(D). Then (i) f* 18B(..,,-) E W' P(8B(xo, r)) for H' almost every r such that B(xo, r) cc D;
(ii) f'(x) = Tr(x) for HN-1 almost every x E 8B(xo,r), for every r such that B(xo,r) CC D;
(iii) if p > N - 1 and if B(xo, h) cc D, then f admits a representative f E W l.P(B(xo, h)) such that f $8B(=o,r) is continuous on 8B(xo, r) for H1 almost every r E (0, h).
Proof We assume, without loss of generality, that f' - f.
122
PROPERTIES OF THE DEGREE FOR SOBOLEV FUNCTIONS
Ch. 5
(i) Let xo E D and let h > 0 be such that B(xo, h) CC D. Since f E W1'P(D) there exists a sequence { fn : n E N) C C' (D) such that
L Ifn -fIdx 0 be such that that B(xo, h) CC D. By assertions (i) and (ii) and
the Sobolev Imbedding Theorem on 88(xo, r), there exists J C (0, h) such that £1(J) = 0 and Tr(f) has a continuous representative gr on 8B(xo, r) for every r E (0, h) \ J. Define f by f (x)
{
g;(x) f *(x)
if lx - xol = r E (0, h) \ J if lx -xoI = r E J.
We observe that
f(x) = f'(x) for GN almost every x E B(xo, h) and f has the property required in assertion (iii).
0
Theorem 5.17 Let D C RN be an open set, N - 1 < p < N, and let f E W"""(D) be a weakly monotone function. Then f admits a representative (still denoted by f) such that f I ors : D \ S -. R is continuous for some set S C D such that Capp_.(S) = 0
for every 0 < a < p. If p = N, then S = 0 and f is continuous in D. In particular, if f E W1.N(D)N and if Jj > 0 GN a.e. in D, then f is continuous in D. The argument we use in the proof below was introduced by Gold'sthein and Reshetnyak (1990) in the case where p > N, and later extended by Manfredi (1994) to the case where p > N - 1. In fact, in the theorem above we can choose the set S such that its Hausdorff dimension is equal to N - p.
Proof Let xo E D, let 0 < h < 2 be such that B(xo, h) cc D, and assume, without loss of generality, that
f =f, where f is the representative of f given by Theorem 5.16 (iii). Let J C (0, h) be such that V (J) = 0, f I8B(=,,r) is continuous and f I8B(:o,r) E W 1"(8B(xo, r))
for every r E (0, h) \ J. Set
PROPERTIES OF THE DEGREE FOR SOBOLEV FUNCTIONS
124
Ch. 5
mr(xo) := inf{f(x) : x c- 8B(xo,r)} and
Me(xo) := sup( f (x) : x E 9B(xo, r)},
for every r E (0, h). Claim 1. mr(xo) < f(x) < Me(xo) for GN almost every x E B(xo,r) and for
every rE (0,h)\J Indeed, for every r E (0, h) \ J
Tr(mr(xo) - f)+ = 0 and
Tr(f - Mr(xo))+ = 0 and since f is weakly monotone on D, we deduce that mr(xo) < f(x) 5 Mr (x0)
for GN almost every x E B(xo, r) and for every r E (0, h) \ J. Define eSSOSCB(xo r) = esS SUp{ f (x) : x E B(xo, r)} - ess inf { f (x) : x E B(xo, r)}
c(r, xo ) and
d(r)
h
Jr
tP-Ndt
JBzo,r)
IVf I°dHN-1. a
Let C(N,p) be the constant in Theorem 5.15. Claim 2. (c(r, xo))P log < C(N, p)d(r) for every r E (0, h). Indeed, by TheoremT 5.15 and the fact that f = f, (Me(xo) - mt(xo))P
0 be such that
rC(N, P) ( IL loges
I Df (y) I i dy
1B(zo.26) Iy -xI2
"
0 such that < f (x)
£''(B(x, r)) B(z,r)
f (y) dy
N . Also, by Theorem
4.57, Hdtm(S) < N - p implies that Capp_a(S) = 0 for all 0 < a < p and we recover Theorem 4.57.
The following result is due to Reshetnyak (1989).
Corollary 5.19 Let D C RN be an open set, let K be a compact set, and let V be an open, bounded set such that K C V CC D. Let 0 E Wia (D)N be such that J0 (x) > 0 VV a.e. X E D. Then there exists a constant 6 >0050 such that I0(xl) - O(x2)I2 < C(N)M*6(Ixl - x212) for every X1, X2 E K such that Ixl - X212 < 6, where
M:=
fV
I Vv(x)I N dx, 6(t) :_
(
\lo 2
l/*
lim 0(t) = 0
e_Q+
and C(N) is the constant of Theorem 5.15.
Proof Fix i E {1, ... , N}and set f := 0;. Let d:= dist(K, RN \ V) and
6:=max{t: 0 0. For t > 0 set
Et:={XED:u(x)OandEg={xED\(AUM) :u(x) N. This was extended later by
Marcus and Mizel (1973) to functions 0 E W-P(D)', p > N, and N:5 m, and it was shown that, for every v : R' - R measurable and for every measurable set E C D, JE
v o O(x)JN(4(x)) dCN(x) = f(E) v(y)N(O, E, y) d.Cn(y) m
whenever one of the two sides is meaningful. Here we use the notation (JN(.O(x))12
=
N P 1: J:(F'i(x))2
where F is the matrix of the N x N minors of V (x) and P :=
m
As it turns out, Theorem 5.23 will generalize this result for m = N, since, by Theorems 5.21 and 5.28, if 0 E Wl,P(D)N and p > N, then is weakly differentiable and has the N-property.
Proof of Theorem 5.23 Without loss of generality, we may assume that v(y) > 0
for almost every y E RN. We divide the proof into three cases. Case 1. Assume that v = XF for some measurable set F C RN. By Proposition 5.22, 0-'(F) is a measurable set and so
0-'(F)nE=0-1(F\A)nE is measurable. By Theorem 5.11,
f-I(F)nE
I J.(x) I dx =
f
f voq5(x)IJ,0(x)I dx = E
N(O, 0-'(F) n E, y) dy,
RN
f
RN
N(O,E,y)v(y)dy.
(5.41)
It follows that (5.40) holds if v is a linear combination of a finite number of characteristic functions of measurable, bounded sets E C D. Case 2. Assume that E C D. Let {va : n E N} be a nondecreasing, nonnegative sequence of simple, measurable functions and let C C RN be a set of measure zero such that lim vn(y) n+oo
= v(y)
134
PROPERTIES OF THE DEGREE FOR SOBOLEV FUNCTIONS
Ch. 5
for every y E RN \ C. Since E C D and Jm E L' (D), by Theorem 5.11 we deduce that
N(4, E, y) < +oo
for almost every y E RNand
vnN(4, E, ) ""' vN(4, E, ) in L1(RN). By Proposition 5.22,
,CN(4-'(C) n (E \ A)) = 0,
where A:= {x E D : J#(x) = 01, and as limn+,,. v 04(x) = v o 4(x) for every x E E \ 4' 1(C), we obtain lim V. O 4(x)IJm(x)I = v o 4(x)I Jm(x)I
n +00
for almost every x E E\A. On the other hand, by Case 1 for all n E N,
E\A
vn o 4(x)I JJ(x) I dx = j N(4, E \ A, y)vn(y) dy N
and since {vn o 4(x)I JJ(x)I } and {N(4, E, y)vn(y)} are nonnegative and nondecreasing, by the Beppo-Levi Theorem (or the Lebesgue Monotone Convergence Theorem), we deduce that
E\A
v o 4(x)I JJ(x)I dx =
JRN
N(4, E \ A, y), dy,
which, together with the fact that JJ = 0 a.e. in A, and using Lemma 5.8, implies
Lvo,1dx = f\A
v
E
=
r
N (4, E\A, y) dy N
f N(4,E,y) dy. RN
ase 3. General case, E C D and V E LOO(RN). Case
There exists a sequence {Kk : k E N} of compact sets such that
Kk C Kk+1 C E and GN(E \ Kk) < k,
CHANGE OF VARIABLES VIA THE DEGREE
§5.4
135
for every k E N. Since
JRN N(, Kk, y) dy
IK,,
for every k E N, passing to the limit in k and using the Beppo-Levi Theorem, we deduce that
Lvo1JI
= f N(O, E, y) dy Ht
0
5.4 Change of variables via the degree Proposition 5.25 Let D C RN be an open, bounded set and let 0 E C(D)N be a mapping which has a weak differential almost everywhere and has the Nproperty. Assume that Jm E Lloc(D). Then, for every bounded, open set G CC D satisfying GN(8G) = 0, we have (i) E L1(RN); (ii) fG J.(x) dx = ,fRN d(-O, G, y) dy.
Proof Fix G CC D. Given v E L°°(RN), set
1: No (v, G, y) =
v(x)
if N(0, G, y) < +oo
xEm-' (y)nG
if N(O, G, y) _ +oo.
+00
Claim 1. No (v, D, y) = No (w, D, y) EN a.e. y E RN and for every w E L°°(D) such that v(x) = w(x) CA( a.e. x E RN. Indeed, let
A :_ {y E RN : N(0, G, y) = +oo}. By Theorem 5.11 we obtain GN(A) = 0.
Let B C D be such that
CN(B) = 0 and v(x) = w(x) for every x E D \ B. Since 0 has the N-property, GN(4(B)) = 0, and for every
yERN\(4'(B)UA)wehave
E xE0-'(y)nG
v(x) _
w(x), xE0-'(y)nG
thus No (v, G, y) = No (w, G, y)
for almost every y E RN. Claim 2. E LI(RN).
136
PROPERTIES OF THE DEGREE FOR SOBOLEV FUNCTIONS
Ch. 5
Let {vn : n E N} be a nondecreasing sequence of simple, measurable functions which converges pointwise to v, and let
k
vn :=
[anXE.. i=1
As
k No(vn, G, y) _ E anN(O, En, y) i=1
by Theorem 5.5, Nm(vn, G, ) is measurable. Furthermore, as {N'j(vn, n E N} is a nondecreasing sequence which converges pointwise to No (v, G, ), we conclude
that N,(v,G, ) is measurable. This, together with the fact that
E
LI(RN) (see Theorem 5.11) and the fact that N4 (v, G,
N(O, G, -) sup 1V(x)1' xED
yields
No(v, G, ) E L1(RN),
proving Claim 2. Now, by Theorem 5.11, we obtain
jv(x)IJ(x)ldx =
JRN
No(v,G, y) dy
and so by the Beppo-Levi Theorem we deduce that
Jv(x)I.4(x)Idx =
jN
No (v, G, y) dy.
(5.42)
In the sequel, we set v(x) = sgn (J,(x)) and (5.42) reduces to
14 J,(x) dx = JgN N4,(v, G, y) dy. Claim 3. We claim that N4,(v, G, y) = d(O, G, y) for almost every y E RN. Indeed, let
El :_ {x E D :.0 does not have a weak differential at x},
E2 :_ {x E D \ E, : Js(x) = 0},
(5.43)
§5.4
CHANGE OF VARIABLES VIA THE DEGREE
137
and
C:= A u 4,(E1) u 4,(E2) U 4,(8G).
Since GN(E1) = GN(8G) = 0 and 0 has the N-property, we deduce that 'CN(.O(El)) = £N(4,(8G)) = 0.
Since G C D is a compact set, by Theorem 5.6 we obtain ,CN(O(E2))
=0
and so
,CN(C)=0.
Let y E RN \ C. If 4-1(y) n G = 0, then 4-1(y) n G = 0 and by the excision property of the degree (see Theorem 2.7), we have d(4,, G, y) = 0 = N,(v, G, y).
If 4-1(y) n G = {a1, ... , ak }, by Lemma 5.10, we deduce that there exists r > 0 such that B(al, r), ... , B(ak, r) CC G are mutually disjoint and d(4,, B(ai, r), y) = sgn(J.(ai)) (i = 1, ... , k).
(5.44)
By the decomposition and the excision property of the degree (see Theorem 2.7) we obtain k
d(4,,B(ai,r),y) = No(v,G,y)
d(4,,G,y) _ i=1 and so
d(4,, G, y) = No (v, G, y)
for almost every y E RN. This, together with Claim 2, yields E L1(RN)
and using (5.43) and (5.45) we conclude that
fJ(x)dx = JRN d(4,, G, y) dy Remark 5.26 Let D, 0, and G be as in Proposition 5.25.
(5.45)
PROPERTIES OF THE DEGREE FOR SOBOLEV FUNCTIONS
138
Ch. 5
(i) (5.44) yields d(O,G,y) _
sgn(JJ(x)) =Em-'
for almost every y E RN and so Id(0, G, y) 15 N(O, G, y)
for almost every y E RN.
(ii) We observe that if we assume in the proof that CN(o(8G)) = 0, without assuming that CI(8G) = 0, the conclusion of Proposition 5.25 still holds.
Next, we state and prove a change of variables formula using the degree function.
Theorem 5.27 Let D C RN be an open set, let 0 E C(D)N be such that 0 has the N-property, m has a weak differential almost everywhere, and JO E L' (D). Assume that G C D is a bounded, open set and that v E LOO (RN ). Then (i) y v(y)d(O, G, y) is integrable, (ii) v o 0 Jm is locally integrable, (iii) fc v o c(x)J,(x) dx = f R, v(y)d(c5, G, y) dy.
Proof By Remark 5.26, Id(¢, G, y)I < N(¢, G, y) CN a.e. y E R" and, together with Theorem 5.23, we conclude (ii). By Proposition 5.25, (i) and (iii) hold in the case where v is a characteristic function. Let v E LOO(RN) and we may assume without loss of generality that v > 041v a.e. Let {vn : n E N} be a nondecreasing sequence of simple functions, bounded in LOO(RN), and converging pointwise to
v. Let A C RN be such that CN(A) = 0 and lira vn(y) = v(y) n+oo
for every y E RN \ A. Setting
B:={xED: Jo(x)=0}, we observe that
m-'(A) = (4-'(A) n B) U (0-'(A) \ B) and by Theorem 5.11 we obtain
CN(4-'(A) \ B) = 0. Clearly,
lm vn o 4(x)Jm(x) = v o ¢(x)J©(x) n-oo
CHANGE OF VARIABLES VIA THE DEGREE
§5.4
139
for every x E (c-'(A) n B) U (D \ 4-'(A)), i.e. {v o OJo} converges to v o OJo almost everywhere. Since {vn} is a sequence bounded in LOO(RN) and Jo E we have vn o O(x)JO -y v o O(x)Jo in L' (G)
(5.46)
and, by Proposition 5.25, v. d(O, G,
v d(O, G, ) in L' (RN).
(5.47)
Suppose, first, that v = Xv, where V C D is an open set such that LN(8V) = 0. Set
F:=Gn0-'(V)nG.
Then F is an open set, F CC D, and GN(5(aF)) N be two positive integers, p > N, let D C RN be an open bounded set, and let 0 E W1,n(D)M. Assume that 0 coincides with its continuous representative. Then, for every A C RN measurable set, we have A
HN(O(A))
N
C(N, p),C (A)" VA I V0(x)IDdCN(x)l P
where a = 1 - v and C(N, p) is a constant depending only on N and p.
Proof We may assume, without loss of generality, that A is an open set and let {Q3 : j E N} be a partition of A into half-open cubes. Using the Sobolev Imbedding Theorem for W1"p (see Theorem 4.45), we obtain for each of these cubes the estimate P
ma I0i(x) - iI
C'(N,P)r1-1 [IQ ,V 1(x)Ip dGN(x)J
,
where ; is the mean value of 0; over Q, r is the edge length of Q, and C'(N,p) is a constant depending only on N and p. Therefore, ¢(Qj) is contained in the mM) and having edges of length cube centred at I.1 = 2C'(N,p)r
Vmj(x)II dGN(x)1 LJ Q
I
J
and so O(Qj) is contained in the ball B. of diameter P 1, = 2IRC'(N,p)r1 p [I IV0 (x)l 'dGN(x)I
.
Q
Since I, tends to 0 when r. tends to zero, we obtain
f
00
HN(O(A)) 5 C(N,P),r, =1
If
IVO(x)I°dGN(x)I 3
J
§5.5
1=
CHANGE OF VARIABLES FOR SOBOLEV FUNCTIONS 1_.1 v
00
< C(N, p)
Lv
oo
[iJ
[r]N
j=1
j=1
141
C(N,p)LN(A)1
VA
v
IVO(x)IP dGN(x)
IDm(x)I°dGN(x)P
.
Remark 5.29 If 0 E W1,P(D)N, p > N, then by the Sobolev Imbedding Theorem (Theorem 4.45) and by Theorems 5.21 and 5.27 we conclude that ¢ is continuous, it is weakly differentiable almost everywhere in D, and it has the N-property. The following versions of Theorems 5.23 and 5.27 are due to Marcus and Mizel (1973).
Theorem 5.30 Let D C RN be an open, bounded set, let p > N, let 0 E W1,P(D)N be equal to its continuous representative, and let v E LOO(RN). For every measurable set E C D, we have
Lv o (x)I J4(x)I dx = f v(y)N(4, E, y) dy.
(5.48)
R
Proof By The Sobolev Imbedding Theorem 0 is continuous (0 E CO,o(D)N, where a = 1 - ). By Theorem 5.28, ¢ has the N-property and by Theorem 5.21, Q has a weak v differential almost everywhere. The result now follows from 0 Theorem 5.23.
As mentioned in Remark 5.24, Marcus and Mizel (1973) proved Theorem
5.30 for 0EWl.P(D)m,m>N. Theorem 5.31 Let D C RN be an open, bounded set, let p > N, and let yh E be equal to its continuous representative. If v E LOO(RN), then, for every open set G C D such that GN(8G) = 0, we have
fvo(x)J(x)dx = JRN v(y)d(e, G, y) dy P roof The result follows from Remark 5.29 and Theorem 5.27.
0
We will focus now on functions in W1,N. Due to Corollary 5.19, in what follows we identify 0 E W1.N(D)N, J,,(x) > 0 a.e. x E D, with its continuous representative.
Theorem 5.32 Let D C RN be an open, bounded set and let 0 E be a mapping such that J#(x) > 0 GN a.e. x E D. Then 0 has the N- and the N-1-properties.
PROPERTIES OF THE DEGREE FOR SOBOLEV FUNCTIONS
142
Ch. 5
Proof (i) We start by showing that 0 has the N-property. Let a E D, R. > 0, be such that QR. = (al
- Ra, al + Ra) x ... x (aN - Ra, aN + Ra) CC D.
Using the argument of the proof of Theorem 5.16, we observe that for each
i E {1,...,N} there exists I; C (-R.,R.) such that L1(I;) = 0 and 0IPr(a) E W1,N(Pr'(a))N (1C(P, (a))N
for every r E (-Ra, Ra) \ Ii, where Ra, a,-1 + Ra) x {a, + r} PT(a) = (a1 - Ra, a1 + Ra) x ... x x (a:+l - Ra, ai+l + Ra,) x ... x (aN - Ra, aN + Ra ).
By Theorem 5.28 we obtain HN-1(((P,'.(a))) < +oo
for every r E (-Ra, Ra) \ I, and so, by Lemma 4.28, and Remark 4.26, we deduce that
GN(0(P'(a))) = 0
(5.49)
for every r E (- R., Ra) \ 1;. Let 9 : RN -. R be a symmetric mollifier (see Definition 1.16), such that 0 < 9(x) for every x E RN,spt(9) C B(0,1), and fore > 0 define
0,(x):= -N 9 ('),X E RN, e > 0, and q,, := 9. * 0. Then 0n E COO(D),
0n _.0 in W1'N(D)N
and, by Corollary 5.19,
4 -4 0 uniformly in QR.. By (5.49) there exists I C (0, Ra) such that L1(I) = 0 and
LN(0(0Qr(a))) = 0
(5.50)
for every r E (0, R.) \ I. Fix r E (0, R0) \ I. Claim 1. For every b E ,(Qr(a)) \ 4,(8Qr(a)) we have d(4,, Qr(a), b) Let b E RN \4(8Qr(a)) and 0 < e < dist(b, ¢(8Qr(a))). Since 4,n converges uniformly to 0 in QR., for n large enough, and by Proposition 1.7 we have d(O,, Qr(a), b) = JRN 0,(on(x) - b)Jm. (x) dx.
CHANGE OF VARIABLES FOR SOBOLEV FUNCTIONS
§5.5
143
Letting n -' +oo and by Theorem 2.3 we deduce that d(4,Qr(a),b) = JRN e(O(x) - b)Jo(x) dx,
and since J41(x) > 0,CN a.e. x E D and d(o, Q, (a), b) is an integer number, by (5.50) we deduce that d(O, Qr (a), b) > 1
for L' almost every b E RN. Claim 2. Xn converges to X GN almost everywhere, where Xn is the characteristic function of 4'n(Qr(a)) and X is the characteristic function of O(Qr(a)) Indeed, let b E RN \ O(8Qr(a)). If b ¢ ¢(Qr(a)) there exists n(x, e) E N such that I0n(x) - O(x)12 < dist(b, 4(Qr(a)) for all x E Q,-(a)
for every n > n(x, e). Therefore, b ¢ On(Qr(a)) for every n > n(x, e) and so
lim Xn(b) = X(b) n-.+oo If b E O(Qr(a))\4(8Qr(a)), by the result of Claim 1 we have d(o, Qr(a), b) >
1 and by Theorem 2.1, X(b) = I. Using the fact that d(¢, Qr(a), b) _ d(On, Qr(a), b) for n large enough, and again by Theorem 2.1, we deduce that Xn(b) = 1 and so lim Xn(b) = X(b) n+oo
Recalling that £N(i)(8Qr(a))) = 0 (see (5.50)), we obtain that Xn converges to X GN almost everywhere. Claim S. GN(O(Q,.(a))) < fQ,(a) J,6(x) dx for every r E (0, Ra).
If r E (0, Ra) \ I, using Fatou's Lemma, Theorem 5.31, and the result of Claim 2, we obtain GNWQr(a))) =
x(x) dx JRN
< lim inf in X. (x) dx n-.+00
< lim inf
N
J
n-+oo Q,.(a)
Jo (x) dx
Jj (x) dx.
(5.51)
Q,4 i)
The result for arbitrary r E (0, Ra) follows from (5.51) and the Lebesgue Monotone Convergence Theorem, where it suffices to consider an increasing
144
PROPERTIES OF THE DEGREE FOR SOBOLEV FUNCTIONS
sequence rk E (0, R.) \ I such that rk Claim 4. 0 has the N-property.
Ch. 5
r.
Let A C D be such that LN(A) = 0. We may assume without loss of generality that A Cc- D. Fix e > 0. Since Jo E L1(D) there exists an open
set 0 cc D cointaining A such that
Jo
Jo(x) dx < e
and there exists a collection {QI: j E N} of half-open cubes such that {Qj: j E N} is a partition of 0 and
O=u Qj iEN (see Rudin 1966, Section 2.19). By (5.51) we have
GN(q (A)) < CN(o(C)) 00
j:CN(0(Qj)) i=1
f Jo(x) dx v
= f JJ(x) dx < e. It suffices to let a -* 0+.
(ii) We show now that q5 has the N'1-property. In fact, by part (i) and by Theorems 5.21 and 5.11, we have
J J0 (x) dx =
JN
N(O, E, y) dy
(5.52)
for every E C D. Thus, setting E = /-1(F) where GN(F) = 0, we obtain
J -'(F) J0 (x) dx = =
JRN
f
N(O, 0-1(F'), y) dy
N(O,
4-1(F), y) dy
=0 and as J4,(x) > 0 for CN a.e. X E D, we conclude that GN(¢-1(F)) = 0.
0
§5.5
CHANGE OF VARIABLES FOR SOBOLEV FUNCTIONS
145
Remark 5.33 It can be shown that if 0 E WI,N(D)N is continuous, open and discrete (i.e. 4' 1 {y} is finite, for all y E RN), then 0 satisfies the N-property. For details we refer the reader to Martio and Ziemer (1992).
The two following theorems were proved by Gold'sthein and Reshetnyak (1990), where, as usual, due to Corollary 5.19 we assume that 4, coincides with its continuous representative.
Theorem 5.34 Let D C RN be an open, bounded set and let 4, E WI,N(D)N be a mapping such that Jm > 0 GN a.e. x E D. If V E L°°(RN) then, for every measurable set E C D, we have JE
v ° 0(x)jJm(x)j dx = j v(y)N(4,, E, y) dy ^'
Proof By Theorem 5.21 0 has a weak differential almost everywhere and, by Theorem 5.32, d has the N-property. The result now follows from Theorem 5.23.
0 Theorem 5.35 Let D C RN be an open, bounded set and let 0 E W1,N(D)N be a mapping such that J# > 0 GN a.e. x E D. If v E L°°(RN), then, for every open set G C D such that GN(8G) = 0, we have JC
v o 4,(x)Jm(x) dx = j v(y)d(4,, G, y) dy.
(5.53)
N
Proof By Corollary 5.19 and Theorem 5.21, 0 is continuous and has a weak differential almost everywhere and, by Theorem 5.32, 0 has the N-property. It suffices to apply Theorem 5.27. Recently, $verak (1988) and Miiller et al. (1994), generalized Theorems 5.34 and 5.35 to a class of Sobolev mappings not necessarily continuous but under some additional regularity hypotheses on the boundary. We recall that the notion of Lipschitz boundary was introduced in Definition 4.40. Although the space Ap,q(D) has already been defined in Section 5.2, for convenience we include its description below.
Definition 5.36 Let D C RN be an open set with Lipschitz boundary, let p > N - 1, q > f-iN-1. Then Ap,a(D) := JOE Wl,P(D)N : adj(VO) E LQ(D)NXN} and
4,q (D) = {¢ E Ap,q(D) : 4(x) > 0 GN a.e. x E D}.
We recall that if p > N - 1 mappings in 4.,(D) are continuous outside sets of Hausdorff dimension less than or equal to N - p (see Remark 5.18). Also, if 0 E A,,a(D), then Jm E LI(D). In fact, from the identity
PROPERTIES OF THE DEGREE FOR SOBOLEV FUNCTIONS
146
Ch. 5
J,(x)IN = V (x) adj (O(x)) we obtain I J.(x)I N= I JJ(x)I Idetadj(0(x))I S I JJ(x)I Iadj(0(x))IN
and so
E L'(D).
I Jo(x)I S I adj (-O(x))I'°
We use A to denote the exterior product (or wedge product) between vectors and we recall that the mapping
N-1) E
(RN)N-1
,
.
t1 A... AtN-1
is multilinear, alternate and, if {el, ... , eN } is the canonical basis of RN, then e1A...Ae;-1Ae:+1A...AeN=(-1)N-le,,
i=1,...,N.
It can easily be verified that if F is a N x N matrix, then ladj F) = sup {I(Fe;) A ... A (F1;N-1)I : I& A ... A tN-1I N - 1, and yo E RN \'(8D). By Theorem 1.12 and (5.55), d(0, 8D, yo) will be well defined and (5.55) will hold for ¢ if we show that we may find a sequence Tn E C°O(D)N such that %P,,18D -, 0I8D uniformly and W,a - ¢ in W1,p(8D)N. To construct the approximating sequence {'P, : n E NJ, using a partition of unity and a change of coordinates we may assume that there exist a, fl > 0 and a Lipschitz function a such that
a(0) = 0, and
loll N(N - 1), then w o v(x) = x for every x E D, v o w(y) = y for every y E B(yo, r), v is a local homeomorphism, and v is an open mapping
on 0 \ L for some set L C 12 of measure zero. In particular, if N = 2, then N(N -1) = N = 2 and v is a local homeomorphism at xo.
Remark 6.4 (i) By Theorem 5.32, if v E W 1,N(f2)N with det Vv(x) > 0 for GN a.e. X E fl,
then v satisfies the N- and the N'-properties. Also, by Theorem 5.32, if v E W 1.p(f2)N, p> N, then v satisfies the N-property. (ii) A function v E W1.N(O)N is said to be a mapping of bounded distortion if lVv(x)Iz < K(det Vv(x)) for 0 a.e. x E fl and for some constant K. It is well known that every mapping of bounded distortion v E W1.N(f2)N is a homeomorphism locally at almost every point xo E f2 (see Reshetnyak 1989, Theorem 6.6, p. 187). Moreover, mappings of bounded distortion are open mappings or are constant in ft (see Reshetnyak 1989, Theorem 6.4, p. 184). More generally, defining the dilatation K of v as et
lvtz)l ec m N
det Ve() W
1
det VO(x) = 0,
K(x) :=
0
v is said to have finite dilatation if 1 < K(x) < +oo for GN a.e. X E fl. Gold'sthein and Vodopyanov (1977) proved that a mapping v of finite dilatation is continuous and its components vi are weakly monotone. Recently, Heinonen and Koskela (1993) and Manfredi and Villamor (in press) proved that if V E W 1.N (f2)N has finite dilatation, if v is quasi-light (i.e. v'1({y}) is compact for all y E RN), and if K E LN-1+, for some e > 0,
then v is open and discrete (i.e. v-'({y}) is finite for all yin RN). In particular, by Remark 5.33 v satisfies the N-property. (iii) Note that, even if v E C1(St)N is such that det Vv(x) > y > 0 for all x E QZ, we cannot expect global invertibility of v without any regularity assumptions on the trace of v. As an example, consider f2 :_ {(x, y) E R2 : 1 < x2 + y2 < 2},
v(x, y) :_ (x2 - y2, 2xy).
For every (x, y) E fl we have det Vv(x, y) = 4(x2 + y2) > 4, although v(x, y) = v(-x, -y) (see also Ball 1981). (iv) Under the assumptions of Theorem 6.1, we cannot expect v to be locally invertible everywhere. The following example is provided by Ball (1981): let N > 3 and consider the cylinder
fl ={xERN :0 0 a.e. x E il. Therefore, to
prove Theorem 6.1 one cannot expect to approximate the function v by a sequence of locally invertible, smooth functions v,,. (vi) Note that for every bounded, open set fl C RN , there exist a measurable set E C fl of nonzero measure and a homeomorphism v E W1,O°(ci)N such that det Vv(x) = 0 for every x E E. The following example is provided by Martio and Ziemer (1992, Remarks 3.7). Let E C [0,1] be a Cantor set of positive one-dimensional measure 0 < 1 - a < 1 and write
vl(x) =v1(x1,...,XN) = f i x--- (t) dt 0
where XEC is the characteristic function of the complement of E in [0,1]. Setting v(x) := (vl (x), x2, ... , xN), it follows that v E W1 -00 (ci) N, v is a homeomorphism of [0, 1]N onto [0, a] x [0,11N-1, and det Vv(x) = 0 a.e. X E
Ex
[0,1]N-1.
(vii) Due to the previous remark, the assumption det Vv(x) 4 0 a.e. x E S2 in Corollary 6.2 is essential.
Lemma 6.5 Let 5t C RN be an open set, let v E W1,N((j)N be such that det Vv(x) > 0 a.e. x e St. Then for every xo E SZ such that v is differentiable at xo (in the classical sense) and det Vv(xo) > 0 there is Ro =- Ro(xo) such that for every 0 < R < Ro the following holds:
N(v, B(xo, R), y) = 1 for almost every y E CR, d(v, B(xo, R), y) =1 for every y E CR, d(v, B, y) = 1 for every y E v(B) \ v(BB),
(6.4) (6.5) (6.6)
for every nonempty, open set B C v 1(CR) n B(xo, R) such that CN(8B) = 0, where CR is the connected component of RN \ v(8B(xo, R)) containing yo v(xo) and N(v, E, y) is the cardinality of the set {x E E : v(x) = y}.
LOCAL INVERTIBILITY IN W'.^'
§6.1
153
Proof By Theorem 5.21, v is differentiable at almost every point x E 11. Fix xo E SZ such that v is differentiable at x0 and detVv(xo) > 0. By Lemma 5.9 there is Ro > 0 such that B(xo, Ro) cc SZ and d(v, B(xo, R), yo) = 1 for every 0 < R < Ro and (6.5) follows from Theorem 2.3. Since det Vv(x) > 0 a.e. x E SZ, Theorems 5.23, 5.27, along with (6.5), yield (6.4). Finally, we prove (6.6).
As v satisfies the N-property (see Remark 6.4) LN(v(cB(xo, Ro))) = 0 and GN(v(OB)) = 0. Since B is a nonempty open set, by Theorem 5.32 we have that GN(v(B)) 96 0 and so GN(v(B) \v(OB)) > 0. Let y E v(B) \v(OB) and let C be the connected component of RN\v(OB) containing y. As GN(v(8B(xo, R0))) = 0 and since d(v, B, -) is a constant on C, we may assume without loss of generality that y ¢ v(8B(xo, Ro)). Let p. E C°D(RN) be such that
00 2 < P" (y), for all yEB(0,2) spilt p, C B(0, e), for all e > 0
JRNp,(y)dy=1 for all e>0.
(6.7)
Since y E v(B) there exists x E B such that y = v(x) and by Theorem 2.3 we have aim
J p, (v(z) - y)det Vv(z) dz = d(v, B, y)
(6.8)
B
and using the continuity of v at x, we deduce that for every e > 0 there exists 6 > 0 such that Iv(z) - y12 < 1 for every z E B(x, 6). Recalling that det Vv(z) > 0 a.e. Z E B(x, 6), by (6.7) and (6.8) we obtain d(v, B, y) > 0.
(6.9)
Finally as the degree d(v, , y) is a nondecreasing function of the set, using (6.5) and the fact that B C v'1(CR) n B(xo, R) we obtain d(v, B, y) < d(v, B(xo, R), y) = 1
(6.10)
which, together with (6.9) and the fact that the degree is an integer number, yields (6.6).
Lemma 6.6 Let SZ, v, Ro, xo, be as in Lemma 6.5, (6.4), (6.5). Let C& be the connected component of RN \ v(aB(xo, R0)) containing yo := v(xo). Then for every r > 0 such that B(yo, r) CC C&,
v(O) = B(yo,r), v(aO) C av(O) = aB(yo, r), where 0:= v-1(B(yo, r)) n B(xo, Ro) CC B(xo, Ro).
154
LOCAL INVERTIBILITY OF SOBOLEV FUNCTIONS
Ch. 6
Proof It is clear that v(O) C B(yo, r). Conversely, if y E B(yo,r), by (6.5) d(v, B(xo, Ro), y) = 1 and so by Theorem 2.1, there exists x E B(xo, Ro) such that y = v(x), implying y E v(O). Let x E 80 and let {an} C 0, {bn} C B(xo, Ro) \ 0 be such that
lim an = n-.+oo lim bn = x.
n-+oo
We have v(an) E v(O) and, as v(O) = v(v-I(B(yo, r))) = B(yo, r), we observe that v(bn) ¢ v(O) = B(yo, r). Using the continuity of v at x, we have
v(x) =n-+oo lim v(an) = n-+oo lim v(bn) 0
and this gives x E 8v(O).
Lemma 6.7 Let v E W1'N(S2)N, detVv(x) > 0 a.e. x E S1 and let xo E D be such that v(x) 0 v(xo) for every x E B(xo, Ro) \ {xo}. For every 0 < R < Ro there exists r > 0 such that v-' (B(yo, r)) n B(xo, R) CC B(xo, R).
Proof Define d(6) := sup{fix - x012: x E B(xo, R), Iv(x) - v(xo)12 < 6}. Since v(x) 9& v(xo) for every x E B(xo, R) \ (xo) and v is uniformly continuous on B(xo, R), we have lmd(b)=0.
Now take r > 0 such that d(r) < a . We have v-' (B(yo, r)) n B(xo, R) C B ( xo, 2) CC B(xo, R). 0
Proof of Theorem 6.1 Let fly be the set of points xo E f2 such that v is completely differentiable at xo and detVv(xo) > 0. By Theorem 5.21 we have GN(cl \ fI) = 0. In the sequel, we fix xo E ff, we set yo = v(xo), and we show that v is locally invertible at xo. By Lemmas 5.9 and 6.5 there exists Ro > 0 such that B(xo, Ro) CC f2, N(v, B(xo, Ro), y) = 1 a.e. y E C&,
(6.11)
where Cp is the connected component of RN \ v(8B(xo, Ro)) containing yo, with N(v, B(xo, Ro), yo) = 1. By Lemma 6.7 we deduce that there exists r > 0 such that v-1(B(yo, r)) n B(xo, Ro) CC B(xo, Ro) (6.12) and
LOCAL INVERTIBILITY IN W'-N
§6.1
B(yo,r) CC Cp.
155
(6.13)
Setting D := v-1(B(yo,r)) fl B(xo,Ro), by (6.12) and (6.13) we have D C v-1(C,,) fl B(xo, Ra) and, by Lemma 6.6, v(D) = B(yo, r), v(OD) C 09v(D) = OB(yo, r).
(6.14)
By the N-'-property of v (see Theorem 5.32) and (6.14), we have CN(8D) = 0 which, together with (6.6), yields
d(v, D, y) =1 for ally E v(D) \ v(8D).
(6.15)
Using the definition of D, the fact that D C B(xo, Ro), (6.11), and (6.13), we obtain N(v, D, y) = 1 a.e. y E v(D). (6.16) Let E := {y E v(D) = B(yo,r) : N(v, D, y) 54 1}, so that CN(E) = 0. We define the candidate for the local inverse function, w, by
w(y) := x if y E v(D) \ E, and v(x) = y, x E D, w(y) := x if y E E, v(x) = y,
(6.17)
(6.18)
x E D being chosen by the axiom of choice. Claiml. w E L°O(B(yo, r))N. We have w(y) E D C 11 for every y E v(D) and so w is uniformly bounded in v(D). To prove that w is Lebesgue measurable, we fix a E R and show that
the set A:= {y E v(D) : wi(y) > a} is measurable, i = 1, ... , N. Clearly, A = Al U A2, where
Al :_ {y E v(D) \ E : wi(y) > a}, A2 :_ {y E E : wi(y) > a}. Since CN(A2) = 0 we deduce that A2 is measurable. Using the fact that the restriction of v to v-1(v(D) \ E) is one-to-one, one can see that
Al = {v(x) : x E v-r'(v(D) \ N), xi > a} _ (v(D) \ E) fl I U V ({x E B(xo, Ro) : a + n 5 xi < a + n + 11) J . Using the fact that for every n E N, {x E B(xo, Ro) : a + n < xi < a+n+1} is a compact set, v is a continuous function, and v(D) \ E is measurable we conclude that Al is measurable. Claim 2.
V 0 w(y) = y for every y E v(D) = B(yo, r),
(6.19)
156
LOCAL INVERTIBILITY OF SOBOLEV FUNCTIONS
W o v(x) = x for every x E D\ v-1(E).
Ch. 6 (6.20)
These follow immediately from (6.17) and (6.18). We notice that, due to Theorem 5.32, GN(v-1(N)) = 0.
Claim 3. f o w is measurable for every f : D -+ R measurable. Since every Lebesgue measurable set is a union of a Borel measurable set and a set of measure zero, to show that f o w is measurable, by Claim 1 it suffices to show that w-1(R) is measurable for every R C D such that C' (R) = 0. Indeed, by (6.19), w-1(R) C v(R),
and by the N-property of v we obtain that LN(w-1(R)) = 0. Thus w-1(R) is measurable. Let g : v(D) = B(yo, r)
R be defined by g(y) __ Iadjov(w(y))I2
detVv(w(y))
Claim 4. g E L'(v(D)). By Claim 3 g is measurable and by Theorem 5.23, Claim 2, and (6.16) we obtain
J N(v, D, y) Ig(y)I dy = N
f
D) lg(y)I dy
= L Ig o v(x)Idet vv(x) dx = fIadiVv(x)I2dx. Therefore, g E L1(v(D)). Claim 5. w E W1,1(v(D))`v and Vw(y) _ (a vv w y )T To prove this claim, we fix 0 E Cp (v(D)) and set K := spt0. We show that
(adJVv(w(Y)))'
f (D)wa(y)°j(y)dy=-f
a(y)dy,
(D)
detVv(w(y))
where Aa denotes the component of the jth row and the ath column of a matrix
A. Set 6 := dist(K,Ov(D)) > 0. Using the uniform continuity of v on D C B(xo, Ro) we choose e > 0 such that
Iv(x)-v(x')I2 N - 1, if v is continuous, adjVv E L7T(S2), and detVv(x) > 0 a.e. in S2, then there is local invertibility a.e. in 0, i.e. for a.e. xo E S2 there exists r > 0 such that VI B(xo,r) is almost everywhere injective with the inverse w E BV,c(vIv(B(xo,,)) )N and there
exists a set E C v(B(xo, r)) such that
E is an open set of v(B(xo, r)), GN(v(B(xo, r) \ E)) = 0, in E
W1.1(E)N,
v o w(y) = y a.e. y E v(B(xo, r)), w o v(x) = x a.e. X E B(xo, r).
By Theorem 5.21 v is weakly differentiable a.e. in S2 and by Lemma 5.10
d(v, B(zo, r), v(xo)) = I
for some r > 0. Let Co be the connected component of RN \ v(aB(xo, r)) which contains v(xo). Then d(v, B(xo, r), y) = 1 (6.24)
for every y e Co and so, if we choose 0 < r' < r such that B(xo, r') C B(xo, r) n v- I (CO),
then, by (6.24) and since det Vv > 0 a.e., repeating the argument used in (6.7)(6.10) we have d(v, B(xo, r'), y) < 1
for every y E RN \ v(8B(zo, r')). It suffices now to use Tang's results (1988, (1.3)-(1.5), (2.26), and Theorem 3.7 (i)). Note, however, that in Tang (1988) it
is assumed that adjVv E Lr, r > I and if N - 1 < q < N, then >p as was no te d by As it t urns ou t , T ang' s resu lts (1988 ) still h old for r Muller et al. (1994, Theorem 5.3).
Proof of Corollary 6.2. (a) We have V E W1'N(f 1)N,
det Vv(x) > 0 a.e. x E SZ1
and
v E W"N(02)N, det Vv(x) < 0 a.e. X E 522. It suffices to apply Theorem 6.1 to v in S21 and to Rov in 522i where Ro is a constant rotation with det Ro = -1.
LOCAL INVERTIBILITY IN W1,N
§6.1
159
(b) Now we assume that v E W1'Q(1l)N, q > N, det Vv(x) # 0 a.e. x E S1 and for almost every xo E Sl, v is locally almost injective in a neighbourhood of xo, in the sense that there is an open set D =_ D(xo) CC S2 and there is a function w : v(D) -i D such that
wov(x)=xa.e.xED. By a corollary of Vitali's Covering Theorem (see Corollary 4.35) there is a countable family of nonempty, open, mutually disjoint balls {B i : i E N) and there is a sequence of functions wi : v(Bi) St such that Bi C S and ,CN(S1\
iEN Bi) = 0,
w o v(x) = x a.e. x c Bi.
(6.25)
The task ahead will be to partition B, into three subsets B; , B?, and Ni such that B,, Bi are two open sets and Ni is a set of measure zero, det Vv(x) > 0 a.e. x E Bi ,
detVv(x) < 0 a.e. X E B. Using the fact that v E W1,9(Bi)N, q > N, by (6.25) and Theorem 5.21 we deduce that there is a set of measure zero A, C B, such that v is differentiable at every x E Bi \ A,, w o v(x) = x for every x E Bi \ Ai, 0 for every x E Bi \ Ai.
(6.26)
det Vv(x)
Let {Ci } be the countable collection of the (open) connected components of RN \ v(8Bi). By Theorem 5.32 we have ,CN(v-1(v(8B, U A,))) = 0.
We claim the following.
Claim 1. d(v, Bi, v(x)) = sgndetVv(x) for every x E Bi \ v-1(v(8B, U A,)). Fix x E Bi \ v-1(v(8Bi U A,)). As v is differentiable at x and det Vv(x) 54 0, by Lemma 5.9 there exists ro > 0 such that for every 0 < r < ro we have d(v, B(x, r), v(x)) = sgn det Vv(x).
On the other hand, setting K := Bi \ B(x, ro), K is a compact set included in Bi and by (6.26) v(x) ¢ v(K) because v(x) ¢ v(A,). By the excision property of the degree (see Theorem 2.7) we obtain d(v, Bi, v(x)) = d(v, B(x, ro), v(x)) = sgn det Vv(x).
160
LOCAL INVERTIBILITY OF SOBOLEV FUNCTIONS
Ch. 6
Claim 2. sgn det Vv(x) = sgn det Vv(x') for every x, x' E v'' (C') \ v-' (v(8Bi U A,))-
Assume that x, x' E v-' (G) \ v-' (v(8B, U Ai)). Using Claim 1 and the fact that the degree d(v, Bi, -) is constant on each C', we obtain that sgn det Vv(x) _ sgn det Vv(x'). Now we conclude the proof of (b). Let I {j E N : detVv(x) > 0a.e.x E
v''(Ci)) and J = {j E N : detVv(x) < 0 a.e. x E v-'(C')}. Set
B, :=U v-'(C')nB,, B,2 :=U v-'(C')nBi, 3EJ JE!
and
Ni:=Bi\(B, UBt)Then Bi := B, U B? U Ni and by setting D1 := U,B; , f12 := U,B,2 and E 52 WI U f12), we obtain that CN (E) = 0 and 521, S22 have the required properties.
0 Proof of Corollary 6.3 To obtain that w E W1'71 (v(D), D) we take s = xg- in Theorem 6.1. If q > N(N - 1), then w E W' N, detVw(y)
= det Ov(w(y)) >
0 a.e. y E v(D)
and so, by Corollary 5.19, we deduce that w is continuous. Hence v and w are homeomorphisms and v is an open mapping in 52' for some fY C 1 open, where
LN(52\52')=0. 6.2 Energy functionals involving variation of the domain The variational treatment of crystals with defects leads to the study of functionals of the type
E(u,v) := J W(Vu(x)(Dv(x))-')dx, n
where 52 C RN is a reference domain, W is the strain energy density, u is the elastic deformation, and v represents the slip (rearrangement) or plastic deformation, with det (Vv(x)) = 1 a.e. x E 52. The underlying kinematical model for slightly defective crystals was introduced by Davini (1986) and later developed by Davini and Parry (1989). As it turns out, matrices of the form
Du(x)(Ov(x))-' represent lattice matrices of defect-preserving deformations (neutral deformations), and, taking the viewpoint that equilibria correspond to a variational principle, Fonseca and Parry (1987) studied the structure of some kinds of generalized related variational minimizer (Young measure solutions) for the energy problems were also investigated in Fonseca and Parry 1987).
§6.2 ENERGY FUNCTIONALS INVOLVING VARIATION OF THE DOMAIN 161
The lower semicontinuity and relaxation properties of E(-, ) were addressed only under additional material symmetry assumptions on W. The existence and regularity properties for minimizers of E(-, -) were obtained by Dacorogna and Fonseca (1992). Following this work, we stress the fact that the direct meth-
ods of the calculus of variations fail to apply to this problem, as sequential is not sufficient to guarantee the existence of minimizers. Indeed, if W(F) = IFIz, it was shown by Dacorogna and Fonseca (1992) that there are no minimizers in {(u, v) E W l-00 x W1,001 u(x) _ x on i1, det (Vv(x)) = 1 a.e.} if 0 < r < N = 2, while for r > N existence is weak lower semicontinuity of
obtained for smooth (u, v) (Dacorogna and Fonseca 1992, Theorem 2.3). It is clear that if is a minimizing sequence and if IVu4,(Vvn)'1j2 is bounded in L1, then
Vun(Vvn)-1
L in L', unto = uo, det (Vvn) = 1 a.e.
and so, if some type of lower semicontinuity prevails, then
L It would remain to show that L will still have the same structure, precisely
L = Vu(Vv)-1 where uI - = uo, det (Vv) = 1 a.e. X E Q. Using the div-curl lemma it follows
that if un -- u in W 1.00 w - *and v -- v in W1,00 w - *, then Vun(Vvn)-1
Vu(Vv)-1 in L°O w-*.
Note that (6.27) is always satisfied if W is a convex function. On the other hand, formally, as det (Vv) = 1 a.e. and setting w = u(v'1), the energy becomes
40)
W(Vw(y)) dy,
which is now an energy functional involving variations of the domain. Hence, under this new formulation, quasiconvexity seems to be more appropriate than convexity (see Acerbi and Fusco 1983, Ball 1978, and Dacorogna 1987). Suppose that W is a quasiconvex function, i.e.
W(F)
0, r > s > 1, p > 1, q > N, p + NQ 11 = .1 (W > 0 if r = s = 1). If un - u in W"p(fI)N, vn v in W1.q(SZ)N, and if det (Vvn) = 1 a.e. in Sl, then
L
W(Vu(Ov)-I)dx < liminf J W(Vun(Vvn)-I)dx - n-++oo ft
Before proving Theorem 6.9 we make some remarks.
Remark 6.10 (i) It is clear that if u E WI.p, v E W1,9 and det Vv = 1 a.e., then Du(Vv)'1 E
L''. (ii)
If r > 1, then s < r is a necessary condition, as the counterexample by Murat and Tartar shows (see Ball and Murat 1984). Here r = s = 2 = NJ I = (0,1)2, W(F) =r det (F), u,, u inr HI(Sl), vn(x) = x, and det Du 2: liminf Jn
detVun. fnn
(iii) The growth condition cannot be dropped even if W is polyconvex and nonnegative. Precisely, if the relation between p, q, r, and s does not hold, the conclusion of Theorem 6.9 may be false. Indeed, using the example by
Mali (1993) with q = +oo, p < N - 1, W (F) = det F, N = r = s, we may u in W",P, u(x) _- x with vn(x) __ x, and find u,, f Idet (Vu) I > liminf fo Idet (Vun)1.
Moreover, the growth condition prescribed in Theorem 6.9 is the wellknown growth condition ensuring weak lower semicontinuity of E(u, id) in WI.p (see Acerbi and Fusco 1983 and Dacorogna 1987).
(iv) It is natural to ask whether or not these results can be extended to the < q < N, since, due to Miiller's result (1990b) and if we assume that det Vv = I a.e., then Det Vv = det Vv a.e. in IL case
§6.2 ENERGY FUNCTIONALS INVOLVING VARIATION OF THE DOMAIN 163
(v) Having obtained lower semicontinuity of the energy in Theorem 6.9, the question now amounts to showing that one can find a minimizing sequence {Dun(Vvn)-1} where {un} is bounded in W1.p and {vn} is bounded in W 1.O°. Actually, one only needs to show that there exists a sequence {f} C W 1"°°(Sl, S2) such that vn o fn is bounded in W 1-°° and
detVfn(x) = 1 =1 fn(x)
a.e. X E 11
xEOS2,
= V(un o fn) (V(vn o fn))-1. Due to the examples provided by Dacorogna and Fonseca (1992), we know that this may not be possible since the infimum of E may be zero, which prevents the existence of a minimizing sequence bounded in W 1.p x W l .Q. since
Vun(Vvn)-1
As is usual in variational problems for which existence of minimizers is not guaranteed (such as variational problems for material that changes phase or, as in our case, for slightly defective materials), rather then studying the macroscopic limit of we focus our attention on the properties of the minimizing sequences. The following may contribute to a better understanding of why the boundedness of {Vun(Vvn)-1} may not entail the boundedness of {Vun} and {Vvn}. Using Theorem 6.9 we show that we can construct a minimizing sequence of the form {VuE(Vvv)-1} with IIot4 llp = 0(E;), IIVVEIIq = 0(-a), for any a,0 > 0. Consider the `perturbed' family of variational problems EE(u, v) := J W(Vu(Vv)'1) dx + e°pIIVu1IIP + E09IIVvEIIq,
where ulan = uo, det Vv = 1 a.e., and Zw'M fn v(x) dx = 0. Using the direct method of the Calculus of Variations, Poincare's Inequality and Theorem 6.9, it follows immediately that there exists (ui,vv) E W11p X W' such that EE(u(,vC) = inf {EE(u,v) : (u,v) E W1'1' x W1,1, detVv = 1 a.e.}
.
Then, given an admissible pair (u, v), E(u,v) = lim EE(u,v)
E-o+
> lim sup EE
v,)
E-»o+
> lim sup E(uE, v.), E-.o+
> inf E. Using the same reasoning with lim E(u1, vE) and taking the infimum in (u, v) we conclude that inf E = lim E(uE, vE) E-.o+
LOCAL INVERTIBILITY OF SOBOLEV FUNCTIONS
164
Ch. 6
and
Iloucllp =
Ilovcll4 = 0(E ).
The following two lemmas will be useful in proving Theorem 6.9.
Lemma 6.11 Let W, 0 be two open sets of RN such that SY CC 12, let q > N, and let v, vn E W1.9(11)N be such that det Vv(x) = det Vvn(x) = 1 a.e. X E 12. Assume that vn v in W1,9(12)N. Then then exists a subsequence of {vn} (not nlabelled) such that for almost every xo E 12' then exist open sets D, Dn C S2'
containing xo and them exist no E N, ro = r(xo) > 0, w : B(yo, ro) -+ D, wn : B(yo,ro)
Dn, with yo = v(xo) such that for n > no,
wn o vn(x) = x a.e. X E Dn, vn o wn(y) = y for every y E B(yo, ro) and vn(Dn) = B(yo, ro),
w o v(x) = x a.e. x E D and v(xo) 96 v(x) for x E D, x 0 xo, v o w(y) = y for every y E B(yo, ro) and v(D) = B(yo, ro),
wn,w E W''. Proof Using Corollary 5.19 and the Ascoli-Arzela Theorem we obtain that, up to a subsequence, vn converges to v uniformly in 12'. By Lemma 6.5 for almost every xo E SY, there is RD > 0 such that B(xo,Ro) CC W, N(v, B(xo, Ro), y) = 1 for almost every y E C8, d(v, B(xo, Ro), y) = 1 for every y E Cp, d(v, B, y) = 1 for every y E B \ v(8B),
for every nonempty open set B C v-1(Cp) n B(xo, Ro) such that GN(v(8B)) _ 0, where Cp is the connected component of RN \v(8B(xo, Ro)) containing yo v(xo). Since v is differentiable at xo and det Vv(xo) 96 0 we may assume without loss of generality that N(v, B(xo, -o), yo) = 1. Fix 0 < e < d(yo, v(8B(xo, Ro))) and choose no E N such that Ilvn - v11 < e. Set
A , :_ {y E CR.: dist(y, v(8B(xo, Ro))) > e}. It is obvious that A, is a nonempty open set. Claim 1. d(vn, B(xo, Ro), y) exists and is equal to 1 for every y E A. and every
n>no. By Theorem 2.3, together with the fact that d(v, B(xo, Ro), y) = 1 for every
yECa,wehave d(vn, B(xo, Ro), y) = 1
(6.28)
for every y E A, and every n > no. By Lemma 6.7 there is 0 < ro < Ro such that
§6.2 ENERGY FUNCTIONALS INVOLVING VARIATION OF THE DOMAIN 165
B(yo, ro) cC AE and v-' (B(yo, ro)) n B(xo, Ro) cc B(xo, Ro).
(6.29)
Claim 2. We claim that B(yo, ro) CC Cn
(6.30)
where CL is the connected component of RN \ vn(8B(xo, Ro)) containing yo. We first prove that AE C RN \vn(8B(xo, R0)). Assume, on the contrary, that there is y E AE flvn(8B(xo, Ra)) and choose x E 8B(xo, R0) such that y = vn(x). We would then have Ivn(x) - v(x)I2 = J Y - v(x)I2 > e > Ilvn - vii, which yields a contradiction. Fix r' > ro such that B(yo,r') C AE. We have that B(yo,r') is a connected set included in RN \ vn(8B(xo, Ro)) and containing yo. We deduce
that B(yo, r') C C and B(yo, ro) CC C. Set
D:=v-'(B(yo, ro)) fl B(xo, Ro) CC 52', Dn :=vn'(B(yo, ro)) fl B(xo, Ro) CC l'.
By (6.28), (6.29), (6.30), and using arguments similar to those in the proof of Theorem 6.1, together with Corollary 6.3, we deduce that for n > no there exists
wn : B(yo,ro) - Dn and w : B(yo,ro) -+ D such that wn, w E W'' 7r (B(yo, ro)) N, and
wn o vn(x) = x a.e. X E Dn, vn o Wn(y) = y a.e. y E B(yo,ro),
w o v(x) = x a.e. X E D and v(xo) y6 v(x) for x E D, x 34 x0, v o w(y) = y a.e. y E B(yo, ro). Finally, by Lemma 6.6, we conclude that vn(Dn) = v(D) = B(yo, ro).
Remark 6.12 (i) It can easily be seen by the preceding proof that if the conclusion of Lemma
6.11 holds for r = r(xo) > 0, then it also holds for 0 < r' < r. Thus, as v is continuous on D, v(x) ,jA v(xo) for x E D and x # xo, we deduce that
lim max{Ix - xo12: x E D, v(x) E B(yo, r)} = 0. r-.0
(ii) It is possible to show that limn+oo GN(DODn) = 0. First, we prove that lim GN(D \ Dn) = 0. n-+oo Let FE := B(yoi ro -e) and O, := v-' (FE) fl D. We claim that for each fixed e there exists no = no(e) E N such that n a no implies OE C D. Indeed,
LOCAL INVERTIBILITY OF SOBOLEV FUNCTIONS
166
Ch. 6
since {vn} converges to v uniformly, there exists no - no(e) E N such that IIv - vnI1 < 2 for every n > no. If X E Oe, we obtain Ivn(x) - Yo12 S Iv(x) - Yo12 + 1v(x) - vn(x)I2 < ro
and so x E D. As UEO, = D and the sequence {O,} is nonincreasing, we have lim GN(D \ Oc)
0,
which, together with the fact that CN(D \ D,,) < GN(D \ O,) for n > no, yields limn+oo, N(D\Dn) = 0. Next, we prove that limn. +oo £N(Dn \ For
n > no. We have
{xE B(xo,Ro):r- 20 for every u,v E E, we have 2 2
Iu,VI>Iu-VI2
(7.17)
with equality if and only if u E OB or v E OR If N = 2 we define 7r1
Gl(u,v) :_ 1 1091U - vI2, G2(u, v) _ loglu, vl,
u0v,
and if N > 3 let Gl(u, v) G2(u, v) :=
-1 2(N - 2)wNlu - v12 -1
2(N - 2)wNlu, VIN -2'
u u
v' v,
where WN = LN(B). Define
G:=G1 -G2. For a E B and r > 0, we set E(a, r) = B \ B(a, r) rt B and S(a, r) = 8B(a, and we define
U:={(u,v):u,vE RN, ug6v}.
FIRST APPLICATION OF THE DEGREE THEORY TO PDES
§7.5
195
Lemma 7.37 Let F E C1(U) and let L > 0 be such that IF(u,v)12
0 such that (see Exercise 7.2) IG1(u,v)12
0 such that If(u) - f(v)12 < KIu - v12 for every u, V E B. Thus we can still apply Lemma 7.37 to eC u'9 If (u) - f (v)) to obtain e
If (u) - f (v)) du av; v fB'(u'v) 8G(u, v) 8 f (v) f 82G(u, v) du If (u) - f (v)) du - f 8v; 8v2 B B 8v,8vj
=
f BIG( v) If (u) - f (v)) du avick
If ,>(v) .
We deduce that 82x(v)
8v^
- JB
82G(u, v) If(u)8v;avi
b f(v))du+f(v), N
hence 02'(u,') avifty is continuous and (7.22) together with (7.23) yield Oz(v) = f(v)
for every vEB. Step 4. We prove that IIxIIz 0, and f E C(D x (0, +oo)). If f (x, 0) > 0, then (7.24) is said to be a positone problem, and there is an extensive literature on this subject (see Amann 1976, Ambrosetti and Hess 1980, and Hess and Kato 1980). We say that (7.24) is semi-positone if
f (x, 0) < 0 for all x E D.
(7.25)
Work on this class of problems can be found in Castro, Garner, and Shivaji (1993), and Smoller and Wasserman (1987), among others. For related applications of the degree theory to boundary value problems, we refer to Rabinowitz (1971, 1975).
We will show that a rescaling argument and the theory of Leray-Schauder topological degree will enable us to obtain positive solutions for the semi-positone problem (7.24) for certain values of A, provided f (x, ) grows linearly at infinity.
Precisely, we assume that there exists m > 0 such that 1(u) lira u-+oo U = m,
(7.26)
and we define
m where Al is the first eigenvalue of -A with zero boundary conditions. Let c1 be the corresponding eigenfunction, with v1 > 0 in D and IIVIIL2(D) = 1. As in the
DEGREE IN INFINITE-DIMENSIONAL SPACES
200
Ch. 7
previous section, we denote by F the Green operator for -A with zero Dirichlet boundary conditions; F : C(D) -i C(D) is a continuous, compact operator such that v = F(g) if and only if
Ov = g(v) V
=0
xED
xEOD'
In the case where f = f (u), we will show that A,,. is a bifurcation from infinity, i.e. there exist A,, -. \,,. and un E C(D) such that IIuIIn -, +oo and un - anF(f(Iunl)) = 0. Setting wn := unllunll-', elliptic theory and (7.26) yield that, up to a subsequence,
wn -' to in Cl"°(D) for some 0 < a < 1, where llwll = 1 and
Ow = ailwl 1W
=0
xED
xEBD'
(7.27)
By the maximum principle to > 0 and as to i4 0, we must have to = -fpl, for some -y > 0. Thus un > 0 in D for n large. Ambrosetti, Arcoya, and Buffoni (1994) used this argument to prove the theorem below, where
a(x) := limes (f (x, u) - mu) , A(x) := lira sup (f (x, u) - mu) . u-+oo
Theorem 7.40 If f E C(D x [0, +oo)) satisfies (7.25), (7.26), then them exists e > 0 such that (7.24) has positive solutions u E W2,p(D) n W02 2(D), for all p > 1, provided either (i) a > 0 (possibly +oo) in D and ,\ E [am - e, ate[ or (ii) A < 0 (possibly -co) in D and X E )A00, aoc + e). In the remainder of this section, following Ambrosetti, Arcoya, and Buffoni (1994), we prove that ,\oo is a bifurcation from infinity.
(7.28)
We remark that u is a positive solution of (7.24) if and only if u is a positive solution of u) = 0, where
t(,\, u) := u - \F(f(Jul)). We rescale 0 as
§7.6
SECOND APPLICATION OF THE DEGREE THEORY TO PDES IIzII 4 (A, T7
to
)
z
201
0
z=0.
Due to (7.26) the function W is continuous. It is clear that A is a bifurcation from infinity for (7.24) if and only if it is a bifurcation from the trivial solution
z=0forT=0.
Lemma 7.41 For every A < A there exists r = r- (A) > 0 such that 4i(µ, u) # 0 for all p E [0, AJ, I Jul I > r.
Proof Suppose, on the contrary, that there exist A' < A,,, {A :n E N), {u,,: n E N}, such that
An - A', IIunll - +oo, un = AnF(f(lunl)). Setting wn := unllunll-1, elliptic theory and (7.26) yield that, up to a subsequence,
win C" (D)
wn
for some 0 < a < 1, where IIwII = 1 and Aw = A'mlwl W
=0
XED
xEBD'
(7.29)
By virtue of the maximum principle, w > 0 and so w = rywl, for some ry > 0. Since IIwII = 1, and in particular w 34 0, we must have A'm = Al, which is in O contradiction with the hypothesis A' < A,,. The proof of the following lemma can be found in Ambrosetti and Hess (1980).
Lemma 7.42 If A > A there exists r = r+ (A) > 0 such that 4;(A, u) 96 t 0, l lul l > r.
Setting An := A - n and A,+, := A. + we choose an increasing sequence {rn} +oo and rn > max{r (A;), r+(A,+,) }. If A < A- , by Lemma such that rn 7.41 and Theorem 7.10 d (W(A, ), B (0,
n)
0 I = d (W(0, ), B = d (I, B (0, 1.
\
(O, !) '0)
')'0) (7.30)
On the other hand, by Lemma 7.42 and Theorem 7.10 and for all t E [0, 1), d
B
(0, n) , 0)
= d (W(A,+ ) - tcoi, B (01 n) +O)
202
DEGREE IN INFINITE-DIMENSIONAL SPACES
Ch. 7
We claim that
To this end, and by virtue of Theorem 7.8, it suffices to prove that the equation
T(A ,0 has no solutions on B (0, B (0, such that n)
n).
Indeed, suppose that if there is a function zn E T(,\n , zn) _ W1'
Setting un := IIznII-2zn, we have IIunII ? rn and
un - anF(f(Iunl)) =
(7.33)
Due to (7.26), we may assume that If(Iul)I 0 such that for every u, v E B, u 34 v, we have L
101(U, V)12
IG2 (u, v)i2 < - Iu FU_ vI2 '2'
IVuG1(u,v)I2 ` Iu-V1 N_1, 2
IvuG1(u,v)I2 < Iu
vI2 ,
IVuG2(u,v)I2