1
Chapter 1 REAL DIFFERENTIABLE MANIFOLDS WITH CORNERS In this chapter we introduce real differentiable manifolds with corners, which include, in particular, manifolds with simple boundary and without boundary. Our manifolds will be modelled over real Banach spaces. Moreover we introduce the morphisms between manifolds, or in other words, the maps of class p. But since our manifolds may have corners, we need to establish before some results about Differential Calculus over open sets of quadrants of Banach spaces. In Global Analysis for instance in Morse Theory, we often have infinite dimensional manifolds with boundary or corners. For this reason we introduce the manifolds with such a generality. 1.1. The Differential of Maps over Open Sets of Quadrants of Banach Spaces.
In this paragraph we develop some results about differential Calculus over open sets of quadrants of Banach spaces, in order to introduce differentiable manifolds with corners. The method of exposition will be similar to the classical one in ordinary Differential Calculus on open sets of Banach spaces. But in our case there is an additional difficulty in the points of the boundary, whose resolution involves Whitney#s extension theorem. First we recall some definitions and elementary results about real Banach spaces and we introduce the notations that will be used in the book. Let E be a real linear space and 1 !:Ed a map. The map I is a norm if: llv11~0 for all VEE; llvll=O if and only if v=O; IIrvII=IrI llvl for all VEE, reR; IIv+uIIsIIvII+IIun, for all U,VEE. If E is a real linear space and 1 1 is a norm in El we say that (E,1 1 ) is a normed real linear space.
Chapter 1
2
Let ( E l 1 1) be a normed real linear space. Then it is easy to see that the map :ExE-+R, defined by d (VrU)=iV-Uur dll IIi is a metric in E. The topology of the metric d will be the Y U topology considered in E. Moreover we can see that the operations in E, addition and product by a real number, are continuous with the topology of E and hence E is a topological linear space.
H
is Let (El 1 1) be a normed real linear space. Then d Ill invariant by translations in E, that is, d III (u+z,v+z)=d, ,(u,v), and BE(v)=v+BE(0), Bi(v)=v+B;(O) for all veE and all &ER, &>O where BE (v) Be (0) are open balls and BZ(v) , Bi(0) are closed balls with the metric d I
n u-
If the metric space (E,d, I) is complete, we say that (E,II II), or E, is a real Banach space. Sometimes, if no confusion can occur, we shall use 1 1 to denote different norms. Let (E,! 1) be a real Banach space and F a closed linear subspace of E. Then (F,II is a real Banach space whose topology is the topology induced by E in F. Let (Eli I) be a real Banach space and F a closed linear subspace of E. Then it can be shown that ( E / F , [ 1,) , where 1 [v]Il=inf. ( UV-U///UEF) , is a real Banach space called quotient space of E by F. The topology of E/F is the quotient topology of E.
Let E be a real linear space and <,>:EXEM a map. We say that <,> is an inner product on E if the following conditions are satisfied: <,> is bilinear, symmetric over ExE;
>O for all veE , v+O
.
If E is a real linear space and <,> an inner product on E, it can be easily shown that: a) For inequality)
all
U,VEE,
JI2s,
(Cauchy-Schwarz
b) The map 1 [ : E a r given by Ivl=+J, is a norm in E, called the norm associated to the inner product <,>.
Real Differentiable M a n i f o l d s with Corners
3
From the condition a) , it follows that the map C,>:ExE+R is continuous with the natural topology induced by [v[=+J . Let E be a real linear space and c , > an inner product on E. We say that (El<,>) is a real Hilbert space, if the norm 1 associated to the inner product <,> satisfies that (El[ 1) is a real Banach space. Let (E,<,>) be a real Hilbert space and F a closed linear subspace of E. Then (F,<,>IFxF)is a real Hilbert space whose topology is the topology induced by E in F. For all ndN it is easy to see that Rn, with the usual structure of real linear space and with the product <,>:R nxR n+R n given by <(xl, ...,xn),(yl,...,yn)~=~x.y is a real Hilbert space 1 it I
.
. . . .+,./-I/
with associated norm 1 ( x ~ , ,xn) topology the usual topology.
..+xi
2
and
associated
If n=1, =xy and IIxII=IxI. Let E,F be real Banach spaces. We set L(E,F)=(h:EaF/h is a linear map) and Y(E,F)=(h:E+F/h is a continuous linear map}. If E is finite dimensional, then Y(E,F)=L(E,F)
.
In order to know when a map is a linear continuous map, it is important the following lemma: Let E,F be real Banach spaces and aeL(E,F). Then the following statements are equivalent. 1) a is continuous, 2) a is continuous at the point OEE, 3) the set (~a(v)[/v€E,lv~Sl)is bounded, 4) There exists a>O such that (a(V)lSallvU for all vcE. Let
E,F be real Banach spaces. We consider the map given by IIaII=Sup( IIa(v) (/WE, Ilvllsl). Then it can be shown that (Y(E,F), 1 11) is a real Banach space.
11 11 :Y(E,F)+R
Another important result that we shall often use is the following:
Chapter 1
4
Let E,F be real Banach spaces and u r f ( E , F ) . Then we have: 1) ua(v) Isl[aulvi f o r all vcE. 2) lul=inf(kzO/la(v) Ilsk[vi for all VEE),
3)
/laU=Sup(/ a ( v )II/vrE, iv[=l)=Sup {J-!!f-/vsE-(O}
is another real Banach space and & ! t ( F , G ) ,
then ~ B o u l b p i . ~ a [ .
A l s o , it can be shown that an element u belonging to I ( E , F ) is an open map if a is onto. In particular a is a homeomorphism if and only if a is a continuous linear isomorphism.
Let I 1(11/1 112 be two norms in an real linear space E. We say that these norms are equivalent if the associated topologies coincide. are two n o m s in If E is a real linear space and 1 [ l r l l E, the following statements are equivalent: a) 1 I1and i2 are equivalent, b) There exist A>O and B>O such that for all y6E.
AUY~~S~~Y~~SBIYJ~
If E is a finite dimensional real linear space and 11 i l l 11 112 are any two norms in E, then it can be shown that these norms are equivalent. If (E,1 1 ) is a normed finite dimensional real linear space and {el,...,en) is a basis of E l then the map vt+(xl, xn), n where v= 1 x.e is a linear homeomorphism from E onto Rn. i=l 1 i'
...,
...
If Ep,...,E, are real Banach spaces, the map I (:E1x xEn+R ixnII) is a norm in the real defined by 1 (x,!. . ,xn)I/=Max{ixli, linear space E1x.. .xEn and (E1x.. .xEn,1 1) is a real Banach space, whose topology is the product topology of the spaces Ell... ,En'
.
. ..,
It can be shown that the maps $
I (XI,. - .'Xn) iil=uxlu+. . .+II~,II and
111[
n (Xll..
/2:~1~...~E n4 given by
I I ~ 2= ~..+II~,II I ~2, I I +.
are also norms in E1x.. .xEn and they are equivalent to the norm, defined above.
Real Differentiable Hanifolds with Corners
5
.
If Ell.. ,En,F are real Banach spaces, the set (f:Elx.. .xEn-+F/f is an n-linear map) will be denoted by n L (E1,...,En;F) and the set (f:Elx xEn+F/f is a continuous n n-linear map) will be denoted by I (Ell...lEn;F). If El= En=El
...
...=
the sets defined above will be also denoted by Ln(E;F) L"(E;F) respectively. We will often use the continuous multilinear maps.
following
and
characterizations of
.
Let u be an element of Ln ( E l l . . ,E,;F). Then the following statements are equivalent: a) a is continuous, b) a is continuous at 0, c) the set ( (la(xll... ,xn)p[xlll=l,. I ixn#sl) is bounded, for d) There exists k>O such that ia(x ll...,xn)~~k~x,~...~xn~, all (x,,.. .,xn )EE,x xEn.
..
...
If Ell...,En,F are real Banach spaces, then the map n 1 11:L (Ell... ,En:F)+R given by lall=Sup( Ila(xll.. .,xn)II/uxllldll.. . ,IIxnlllsl) is a norm in the real n n linear space L (E,,...,En;F) and (I (El,...,En;F),I 1) is a real Banach space. Moreover We have that
.
II~II=SUP( iia(xlI.. .,Xn)ii/iixln=l,.. I lxnn=l)= =Inf(k~O/~~a(xl,. ..,xn)IlskllxlII.. Hxnll for all x1"E1,.. ,xneEn). Hence if u is an element of Ln(E1,.. ,En;F) it can be shown that IIu(xl,.. .,xn)IIcllull IIxlll.. . IIxnII for all element (xll.. ,xn) of E1x. . .xEn.
.
.
.
.
If E1,...,En are finite dimensional real Banach spaces, then n L (E,,...,E n ;F)=2n (E,,...,En;F).
.
Let Ell.. ,En,F be real Banach spaces and ,En;F) the map defined by p:L(El,L(E2,.. ,L(EnlF). .))+Yn(E1,.. p ( u ) (x,,.. .,x,)=(. .. (a(xl)) (x2) .) (x,). Then cp is a linear homeomorphism and 1 p (a) 1 = 1 u 1 .
.
.
..
.
From the properties of the nonns of linear continuous maps that we have seen above it follows that:
Chapter 1
6
a) The map o : I ( E , F ) x k ? ( F , G ) + J e ( E , G ) given by o ( c X l 8 ) = 8 o 0 r iS a bilinear continuous map. b) The map e:Exl(E,F)+F given by e(v,a)=a(v) is a bilinear continuous map. c) The map e:E 1x...xE nxi?n (El ,En ;F)+F given by e(v ll...lvn' a)=a(v lI...,vn) is an (n+l)-linear continuous map.
,...
If E,F are real Banach spaces and n is an element of IN, then the set (f:En+F/f is an n-linear symstric map) will be denoted by L:(E;F) and the set (f:En+F/f is an n-linear continuous symmetric map) will be denoted by Y:(E;F). It can be shown that Y:(E;F) is a closed linear subspace of Lt(E;F) and hence (k?:(E;F) I 1 1) is a real Banach space. If E is a topological real linear space and V ( 0 ) is the system of neighbourhoods of 0 in El then U=(UVcExE/Vd(O)], where Uv={(x,y)~ExE/x-y~V), is a uniformity in E whose associated topology coincides with the topology of E. We shall say that a topological real linear space, El is complete if the associated uniform space (E,U) is complete. topological real linear space, E l is called Banachable if it is complete and there exists a norm, 1 11, in E whose associated topology coincides with the topology of E; of course (E,1 11) is a real Banach space. A
Definition 1.1.1 Let E be a real Banach space and A=(hll independent system of elements of I(E,R).
...,An)
a
linearly
..
Then the set (x~E/k~(x)tO,. ,hn(x)~O) is called A-quadrant of order n of E and is denoted by Ei or by Ei if A=(h). Obviously, ve have E
=
+ if [y Eii and EA=E
A=@.
1=1
If E=!R2 and A=pl where pl(xly)=x, then E~={(x,y)€R2/xrO). If E=R2
and
A= ( p1 ,p2 ) ,
where
p1 ( x , y ) =x
and
p2 (xI y)=y ,
then
Real Differentiable Manifolds vith Corners
7
+ 2 EA=((x,y)~R/ x ~ O , y ~ O ) . If E is a real Banach space and A=(Al,...,An) is a linearly independent system of elements of Z(E,R), then the closed linear n 0 0 or by EA if A=(A). subspace (1 ker(Ai) of E is denoted by EA 1=l
o n o
0
Obviously, EA=() Eai and EA=E if A=#. 1=1 Definition 1.1.2 Let E be a real Banach space and F a closed linear subspace of E. A linear subspace H of E is said to be a topological supplement of F if E=F+H, FnH=(O) and H is closed. In such case ve shall write E=FsTH and it can be shovn that the map 8:FxH+E given by B(x,y)=x+y is a linear homeomorphism. We shall often use the following facts: a) If two closed linear subspaces of a Banach space have the same topological supplement, then they are linearly isomorphic. If two closed linear subspaces of a Banach space have the same codimension and this is finite, then they are linearly isomorphic. b) Every finite dimensional linear subspace of a real Banach space, admits a topological supplement. c) Every closed linear subspace of a real Banach space whose codimension is finite, admits a topological supplement. d) Every closed linear subspace, F, of a real Hilbert space, I (HI<,>), admits the subspace F =(uEH/=O for all veF) as a topological supplement. We shall also use the following result. e) Let E be a real Banach space and F a linear subspace of E. Then the following statements are equivalent. i) F admits a topological supplement in E. ii) There exists a linear continuous map, p:E+E, pop=p and Ker(p)=F.
such that
a
Chapter 1
iii) There exists a linear continuous map, q:E+E, such that q o q = q and irn(q)=F. L e m m a 1.1.3
Let E be a real linear space. Then ve have: a) Let (hl,...,X n ) be a set of linear maps from E to W. Then {Al,...lhn) is a linearly independent system if and only if there exist XI,. .. ,xnrE such that X . (x.)=aij, where B is the ij 1 7 Kronecker index, b) Let (yll...,yn) be a set of elements of E. Then {yl,...lyn) is a linearly independent system if and only if there exist linear maps ul,...,~nfrom E to R such that u.(y.)=8 1 J ij* using t h i s lemma it can be shown that: Proposition 1.1.4
a) Let E be a real Banach space and A=(hl,...,Xn) a linearly independent system of elements of 2(E,W). Then there exist linearly independent elements of E, xl,. . . ,xn, such that 0 o L(xl ,xn)' vhere L ( x ~ ~ . . . , x ~is ) the Ai(x.)=aij and E=EA 7 T linear subspace of E generated by xl,~..,xn.Moreover the map a:E AoxR n+E, defined by a(xo,rl,.. .,rn)=xO+rlxl+. +rnx n is a linear homeomorphism such that
,...
..
+
+
Finally, EA is homeomorphic to EA for all AEA. b)
Let
M=(qll...,qm)
be a real Banach space and h=(A l,... ,An), linearly independent systems of elements of Y ( E , R ) E
+ + Then ve have: 1) Eh=% 0 0 such that Eh=En. , 2 ) n-rn, 3 ) there exists a bijective mag, ~:(1,2,~..~n)+{1,2,...,n), such that for every + satisfying Ai=riqT(i). k { l , . . . , n) there exists r.ER 1 c)
Let
E
be
a
real
Banach
space
and
A=(Al,
... ,An),
M={TI~,...,~~ linearly ) independent systems of elements of Y ( E , R ) .
Real Differentiable Manifolds vith Corners Then there exists a linear homeomorphism, u:E+E, + 0 0 )=EM and u(E )=EM.n u(E + A A
9
such that
Now we introduce the differential of a map defined on an open set of a quadrant of a Banach space. Proposition 1.1.5 Let E,F be real Banach spaces, A=(Al,. . .,An} a linearly + independent system of elements of Z(E,IR), U an open set of EA, f:U+F a map and x a point of U. Then if u,v are elements of Z(E,F) such that
ve have that u=v. Proof. Let z be a point of Ei with z+O and & > O . Since + lim = O f there exists 6>0 such that B6(x)nEAcU and Y-X Y+X for all yeU satisfying O
(u-v) ( 2 ) p c 1 z n .
Now let t be a point of E. By proposition + t=z+r1x1+...+ rnxn' where zeEA. Then Il(u-v) (t)ll=O and U=V.O
1.1.4
Such a map ucY(E,F) as in the preceding proposition, is called a tangent to f at the point x. Then we have proved that two tangents to f at the same point x, must be equal. Now we can give the following definition: Definition 1.1.6 Let E,F be real Banach spaces, A=(Al, ..., An} a linearly independent system of elements of Z(E,IR), U an open set of E + hf f:U+F a map and x a point of U. Then if there exists an element ucY(E,F) such that
Chapter 1
10
differentiable at x, u is called differential of f at x and u is denoted by Df(x). If f is differentiable at every XEU, f is said to be differentiable on U.
+
obviously if A=#, EA=E and the preceding notions coincide with the similar notions given in the case of open sets of Banach spaces. The notions of differentiability and differential of a map at a point, are independent of the equivalent norms considered in 1.1.5 and 1.1.6. Proposition 1.1.7 Let E,F be real Banach spaces, A=(hl,. . . ,An) a linearly + independent system of elements of L(E,R) , U an open set of EAl f:U+F a map and x a point of U. If f is differentiable at x, then f i s continuous at x.
Proposition 1.1.8 With the same hypothesis of the preceding proposition, if f + is differentiable at x and v (resp.-v) belongs to El\, we have f (x+tv)-f (x) that resp. t Df (x)(v)=lim+ t-ro Df (x)(v)=lim- f (x+tv)- f ( x ) . In particular if A=#, it happens
I
that
t+O
t
Real Differentiable Manifolds with Corners
11
-f (x) for all vcE. Df (x) (v)=lim f (x+tv) t t+o
+ Proof. If vcEA, __ then
f (X+tV)-f (X)-Df (X) (tV) +lim+ Df(x) (tv) t t t t+o t+o t+o =Df (x)(v) (The same for the case -vcEi) .o Theorem 1.1.9. Let E,F,G be real Banach spaces, h=(AID...DAn) a linearly M=(T)~,...,T)~) a independent system of elements of Z ( E , W ) , linearly independent system of elements of Z(F,R), U an open set + + ,, V an open set of FN, f:U+V a map, g:V+G a map and x a of E point of U. Then if f is differentiable at x and g is differentiable at f(x), it happens that gof is differentiable at x and D(g0f) (x)=Dg(f(x))oDf(x)
such that for all ycU with 0
1I f(Y) -f(x) -Df(X) (Y-X) . *
Proposition 1.1.10 Let a,b be real numbers with a
Chapter 1
12
Suppose that h and g are differentiable at every point of (a,b) and that URh(t)U'Dg(t)(l) for all tc(a,b). Then
Ih(b)-h(a) ('g(b)-Q(a)
-0
Proposition 1.1.11 (Mean Value Theorem) Let E , F be real Banach spaces A=(A1,. #,An) a linearly + independent s y s t e m of elements of Y(E,R), U an open set of EA and f:U+F a map. Let x,y be elements o f U w i t h x*y such that [x,y]cU, is continuous, f is differentiable at every point of ZXIYI (xly) and Sup( UDf ((1-t)x+ty) 1 te(0,l) )=k is a real number. Then I
f'
gf(Y)-f(x) I%kUY-X(
-
Proof. We consider the map h:[O,l]+F defined by h (t)=f ( (1-t)) x+ty) From the hypothesis and the proposition l.ls9, we deduce that the map h is continuous ~n C O , l J , differentiable at every point of (0,l) and f o r all ts(O,l) Dh(t) (s)=Df((l-t)x+ty) ((y-x)s).
.
Let g:[O,l]-+R be the map defined by g(s)=k(y-xus. Then, (1) for all tc(O,l) and from the proposition IDh(t) ~ck~~y-x~~=Dg(t) 1.1,IO it h o l d s that Ilf(y)-f(x) [sk[y-xi.u Corullary 1.1.12
Proof. Let g:U+F be the map defined by
g(z)=f(z)-Pf(xg) ( 2 ) * Then it suffices to apply the above preposition to the map
g.e
The p times differentiable maps and the differential of order p will be introduced by induction. L e t E,F be real Banach spaces, A={X1,..,,X,) a linearly independent system of elements of L'(E,R), U an open set of Eh+ and f:U+F a map.
Real Differentiable Manifolds with Corners
13
Suppose that f is differentiable at every point of U. Then we have the map Df :U+Z(E,F) which assigns Df (x)&(E,F) to every point x of U. If Df is differentiable at XEU, we say that f is twice differentiable at x. Obviously D(Df) (x) belongs to Y(E,Y(E,F)) 2 A s we have said, the spaces Y(E,Y(E,F)) and Y (E;F)=(v:ExE+F/v is bilinear and continuous) are linearly homeomorphic by the map a1 (u)(r,s)=u (r)(s). Then the element a1 (D(Df(x)) ) is denoted by
.
D2 f(x) and it is called second differential of f at x. If f is twice differentiable at every point of U, we say that f is twice differentiable on U and the map 2 2 2 XEUMD f(x)& (E;F) is denoted by D f. By induction we define the p times differentiable maps on U and the differential of order p of f at every point of U for every PEN, using that I (E,Yn-l (E,F)) is linearly homeomorphic to In(E,F)
.
ProDosition 1.1.13 Let E,F be real Banach spaces h=(A l,...,hn) a linearly + independent system of elements of Z(E,R), U an open set of EA and f:U+F a map. Suppose that f is p-1 times differentiable on U and p times is a p-linear, differentiable at x, where p”2. Then #f(x) continuous and symmetric map. Proof. By the definitions that we have given, #f(x) is a p-linear continuous map. To see that DPf(x) is a symmetric map, we first suppose that p=2. Let 6 > 0 be such that for all pair (u,v) of elements of + Bg(0)r\EA and for all tE[0,2) it holds that x+tu+v, and x+tu belong to U, and Df is bounded in the intervals [x+v,x+u+v] and [XIx+u] . Let (u,v) be a pair of elements of Bg(0)nEh+ and the map :[0,2)+F defined by g (t)=f(x+tu+v)-f(x+tu). By 1.1.12
g(UIV)
(UIV)
Chapter 1
14
Since f is twice differentiable at x, if E>O there exists 6'>0 with 6'<6 such that for all pair (u,v) of elements of f Bgr(0)nEA and for all te[0,2) we have: UDf (x+tu+v)-Bf(x)-B(Df) (X) (ttl+v)1Sltu+Vl~s.(2puu+Rvy) and IDf(x+tu)-Df (x)-D(Df) ( x ) (tu)Is2clu[. Then for all pair (u,v) of + elements of Bg,(0)nEA and all tc[0,2) we have that: 2 [ (Df (x+tu+v)-Df ( x ) 'D9(urv) (t)(1)-D f (x)(v,u) -D(Df) (x)(tu+v)J (u)-[ (Df (x+tu)-Df(x)-D(Df) ( X ) (tu)] (u)us 2 x ) 1s 5 4 4 ~ 1+1w n v u ) : ng(u,v) (l)-g(u,v) ( o ) - D ~ ~ ((vru) SSUPI uD9(u,v) (t)-Dg(urv)( 0 ) W/te(orl) )+4+1 ([uU+[vp SSUP( IID9(u,v) (t)(l)-Dg(urv)( 0 ) ( l ) + ~ ~ f ( (vru)-D2f(x) x) (v,u)I/ te(o,i) ) + ~ & ~ u ~ ~ ~ ( ~ u ~ + ~ v [ ) and ~ ~ ~ E ~ u ~ ( ~ u ~ h v , u) -g(v, u) ( o ) - D ~ ~ ( (urv) x ) is12.1~~ (uun+iivn); moreover
-
CJ (u,v)
-qu ,v) (0)= q v ,u) (1)-g (v,u) (0)-
Then
for
all
pair
(u,v)
of
elements
+
of
Bg,(0)nEA,
[D2f(x) (v,u)-D2f (X)(U,V)1.512&(iUi+[V1)2. Since D2f(x) is bilinear, we have uD2f(x) (v,u)-D2f(x) (u,v) 1 = a 2 ~ ( l u [ + l v ~ )for ~ all + Then D2f(x) (v,u)=D2 f(x) (u,v) for pair (u,v) of elements of EA. + all pair (u,v) of elements of E A r because of c is arbitrary. Then from 1.1.4,
If pr2 and (v,, the
4
. ..,vp)
is an element of Ep'2r
9 (V3r.--rvP):U+F
map
( V 3 r * * * #PV1
it follows that D2 f(x) is symmetric.
..
( y ) = # - 2 f ( y ) ( v ~ , . tvp).
differentiable
at
X
.
and
Then g 2
we consider
defined
by
( V 3 r - * - iPv1
is twice
qv3,. ..,v ) (XI P
(v1rv2)=
=Dpf(x) (vl,v2,v3,.. ,vP) for all pair (vlrv2) of elements of E, (1.1.16).
By step p=2 we have that DPf(X)(VlrV2rV3r ...rV ) = P =Dpf (x)(v2,v1,v3, ,v for all pair (vl,v2) , of elements of E.
...
P
Real Differentiable Manifolds vith Corners
,...,vP)=#f(x)
Therefore DPf(x) (v1,v2,v3 for all (vl, v )EEP.
... P
15
,...,vP)
(v2,v1,v3
Suppose that the result is true for p-1. Then for all t )cEP'l, for all permutation -c of (2 p) and for all (t2 P yeU we have that # ' f (y)(t,, ,tp)=#-'f (y)(tt(2), ,tt(p) ) If (t2, t )sEP'l we consider the map g :U+F defined P (t2,...#tP) Then is ,tP) (y)(t,, by qt,, ,tp)(y)=#-'f g(t,,***,tP1
,..., ..., ...
,...,
...
...
differentiable at x and Dg
...
.
(t2,..'#tP1 (x)(t,)=#f(x)
.
--
(tl,t21
for all tlcE (1.1.16).
I
tP)
Finally for all permutation -c of ( 2 , ...,p) and all we have that #f(x) (tl,t2, tp)= (t1,t2,. * ,t W P , P =#f(x) (tl,tr(2)I * * 't7;(P))=#f(x) (t,,tl,t3, ,tp)= ) and the proof is finished.o ,tt =#f (x)( tt ( ,tl,tT ( ,
.
,
-
...,
...
Definition 1.1.14 Let E,F be real Banach spaces, A=(A l,...,An) a linearly + and independent system of elements of f(E,R), U an open set of EA f:U+F a map. a) We say that f is of class
0
on U, if f is continuous on
U. b) If PEN, we say that f is of class p on U, if f is p times is continuous. differentiable on U and the map #f:U+f:(E,F) c) We say that f is of class m on U, if f is of class p on U for all peH. Proposition 1.1.15
Let E,F be real Banach spaces, U an open set of a quadrant + EA of E, f:U+F a map and peN. Then the folloving statements are equivalent: a) f is of class p on U, b) f is differentiable on U and the map Df:U+f(E,F) class p-1.0
is of
Chapter 1
16
Obviously, from 1.1.7, if f is of class p it happens that f is of class q where Oiqip. Examples 1.1.16 a) Any linear continuous map, a, is of class m on every open set of every quadrant and Da(x)=a and #a(x)=O for all pr2. b) Let E,F,G be real Banach spaces, a:ExF+G a continuous
bilinear map and U an open set of a quadrant of ExF. Then 0: is of class m on U. For all (x,y)~Uand for all (u,v)cExF, we have that
...
c) Let El ,...,En,G be real Banach spaces, B:Elx xEn+G a continuous n-linear map and U an open set of a quadrant of E1x.. xEn. Then 6 is of class m on U and for all (xl,. ,xn)dJ we have that
..
.
Let E,F be real Banach spaces, ncN and f:E+F a continuous map such that there is a symmetric n-linear map u:En-rF with f (x)=u(x, n) ,x) Then f is of class 03 and
...
.
Bk€ ( x , ) = n ( n - l ) , . , ( n - ( k - l ) ) u ( x 0 ’ ” ”n-k) ’”0’---~~)-
d) Let E be a real Banach space. Then it happens that: is~) 1) the set G L ( E ) = ( k : E = + E / k is a linear h o ~ e u ~ o ~ ~ kis an open set of X ( E , E ) (See 6.1.8.V). 2 ) the map t:GL(E)=GL(E), defined by T(u)=u-~, is of class m and for all uO6GL(E) , Dt(u,) is the linear homeomorphism, Dz(u,) :X(E,E)+Y?(E,E) , defined by Dt(uo) ( v ) = - u ~ ~ o0v o-u - ~ Proposition 1.1.17 Let E,F,G be real Banach spaces, U an open set Q€ a quadrant of E, V an open set of a quadrant of F, f:U+V a map of class p and g:V+G a m a p of c l a s s p, v h e r e peNu(m). Then gof is a map of class p on u.
Proof. It follows from 1.1.9, 1.1.15 and 1.1.16.0
Real Differentiable Hanif ol ds with Corners
17
ProPosition 1.1.18 a) Let f be a map of class p on U, where U is an open set of a quadrant of a real Banach space. Then if V is an open set of U, and hence of the quadrant, the restriction f is of class p. IV b) Let f:U+F be a map, where U is an open set of a quadrant of a real Banach space E and F is a real Banach space. Suppose that U = ( J Ui, where Ui is an open set of U (and hence of the ie1 quadrant) for all k1. If f is of class p for all i d , then f IUi is of class p on U. + c) Let V be an open set of EAcE and f:V4 a map of class p, + is a quadrant of the real Banach space E and G is a real where EA Banach space. Suppose that F is a closed linear subspace of E and VnF is an open set of a quadrant of F. Then f lVnF is of class p on VnF and D f I VnF (x)=Df(x) for all xEVnF. IF
[
]
Proof. a) It follows from the equality f,,=foj, where j:V4U is the inclusion. b) It follows from the fact that for xEUinU j' c) It follows-from 1.1.17.0 Other definitions of map of class p, not used in this book except for special cases, are the following: Let E,F be real Banach spaces, A a finite linearly + independent system of elements of !E(E,R), U an open set of EA, f:U+F a map and pcNU(m). Then: A) We say that f is of class p on U, if for all element XEU there exists an open neighbourhood Vx of x on E and there exists a map fx:Vx+F of class p (with the meaning of the ordinary Differential Calculus) such that fx and f coincide on VxnU. B) We say that f is of class p on U, if there exists an open set G of E and there exists a map T:G+F of class p (with the + meaning of the ordinary Differential Calculus) such that GnEA=U and f,f coincide on U.
Chapter 1
18 Of
course we have that Bf*A)+1.1.14.
Later, it will be shewn that if E: is a real Banach space ~ d ~ ~ t tdifferentiable i n ~ partitions of unity sf claes p , then A)trB). It will also be seen, via Whitney's extension theorem, that A)oB)e1.1.14 if E is a euclidean space. Finally, it w i l l be Seen that if pcpl the A ) e S . 1 . 1 4 (2.1.35). 1.2. Differentiable Manifolds with Corners L e t X be a set and g~Qlv(=). We say that (~'~,(E,A)) its a
chart an X if it holds the fellowing: U is a s\lbsst s f X; E is a real Banach space; A is a finite linearly independent System of =#= elements of Z(E,R); p:U+Eh is an injective map and p ( U ) is an 4= open set of EA. If ( U , p , (E,h)) is a chart QPL X, the set U will be called the domain of the chart, the map cp the morphism of the chart and the space E the medel of the chart. be charts en X. We say that they are compatible of class p if: cp(UnU') and p'(UnU') are open Let ( U , p , ( E , A ) ) ,
~f;
(U',p',(E',A')) =#=
E ' ~ ,
respectively, and the maps and ppr -1 :p, (UnU')+cp(UnU~) are of class p (and hence homeomorphisms).
sets
ef
ptp-':p(UnUt)+pf
and
I
(UnLJl)
Charts on X is called an atlas of class p on X, if the domains of the charts of B cover X and any two of them are compatible of class p . A Set, 8 , of
Proposition 1.2.1 Let X be a set and p E P i u ( m ) . Then the preceding binary relation, 2, is a equivalence relation Over the atlases of class p on X. ~t is clear that p is a reflexive and symmetric relation. prove the transitivety, let d=( (Ui,pil(Ei,Ai))/ieI)r ( V j , ' t j l (E.,A.))/j€J} and G=( (Wk,7k, (EkIAk))/keK) be atlases of
res sf.
To 91=(
3
3
Real Differentiable Manifolds with Corners class p on X such that 14% and 8%. atlas of class p on X.
19
We must prove that &Us
is an
It is clear that A& is a set of charts of X whose domains cover X. If is1 and kEK we must prove that the corresponding charts are compatible of class p. In fact we have that pi(uinwk)=(J 9 . q-1 . [ q . (UinV.)n9 (v nW )] jcJ 1 3 3 3 j j k and r (U.nW )=(I r q-'[qj(UinV.)nq (V nW ) ] jcJ k j I j j k and y
is an open set of E?
it follows that pi(UinWk)
+ k (UinWk) is an open set of E kAk
.
LAi
and
Finally from 1.1.17 and 1.1.18 it follows that the maps rkpi and pia r-1 k are of class p.0 -1
Definition 1.2.2 Let X be a set and psNu{m). If d is an atlas of class p on X, the equivalent class, [A], is called differentiable structure of class p on X and the pair (X,[A]) is called differentiable manifold of class p or a CP-manifold, usually only denoted by X. If (X,[d]) is a differentiable manifold of class p, we shall say that 8 is an atlas of (X,[A]) if 8e[I4] and we shall say that c=(U,p, ( E , A ) ) is a chart of (X,[A]) if there exists 84[A] such that ce8. If c=(U,p,(E,A)) is a chart of (XI[&]) and xsU, we shall say that c is a chart at x. Remarks about the preceding definitions: 1) Any two charts of a manifold of class p are compatible of
class p. 2) Let (X,[la])
be a differentiable manifold of class p and c=(U,pf ( E , A ) ) a chart on the set X. Then c is a chart of (X,[B]) if and only if there exists an atlas 8 of (X,[I4]) such that c is compatible of class p with c f , for all ~'€8.
Chapter 1
20
Let d , 8 be atlases of class p on a set X. Then d , B are equivalent of class p if and only if for all xrx there exist 6=(u,cpI ( E ’ A ) ) and C ’ = = ( V , * , ( P , M ) ) , charts of dl and B respectively, such that xdJnV and c,c‘ are compatible of class p. 3)
Proof. If 4 , B are equivalent of class p , it is obvious that the preceding condition is satisfied. Now suppose that the condition is satisfied and let (U,pI (E,A ) ) , (V,Q , (F,M) ) be charts of S? and % ~ ~ s ~ e ~ t i vsuch ely that VnU*$. For all xeUnV let (Uxrpx,( E x , h x ) ) (V,,@,, (Fx,Mx)) be charts of B and B respectively compatible of class p and with I
and hence p(UnV) is an open set sf El;, f open set of FM.
4)
Let
(U,pl(E,A)),
Analogously, @(UnV) i s an
a Inanifold of class p r l and xeX. If (U‘,@’,(E’,A’)) are charts of X with XEUAU’, then E (X,[d])
be
and E’ are linearly homeomorphic and dim(E) is called dimension of X at x and denoted by d i m x X . If dimXX is finite for all XEX, we shall say that X is locally of finite dimension. 5) Our definition of a differentiable manifold of class p ,
makes sense for the case p=O. In this case the corresponding manifold will be called topological manifold. Let X be a topological manifold and XEX. Then if there exists a chart of X at x, ( U , p , ( E , A ) ) , such that dimE
Real Differentiable Manifolds with Corners
21
not linearly homeomorphic. In fact, from Torunzick's theorem, if we take a non Hilbertable Banach space, E, there is a Hilbert space, H, such that, E and H are homeomorphic. The differentiable manifolds are endowed with topology induced by the differential structure.
a
natural
Proposition 1.2.3 Let X a differentiable manifold of class p. Then the set B=(UcX/U is a domain of a chart of X) is a basis of a topology on X.
Proof. Obviously X=W(U/UEB). Let (U,cp,( E , A ) ) and (U8,cp', (E',A')) be charts of X such that UnU'+4- Then (UnU',c~lu~utr (E,A)) and (UnU',(~'~u~utr(E',A')) are charts of X. Indeed, let (V,*, (F,M)) be a chart of X. It is easy to see that cp (unu'nv) =cp(p' ( c p (unu' ncpn (u'Av) and *(UnU'nV)=*cp'-'(cp' (UAU')A(P' (U'AV)) and hence (p(UnU'nV) and + + 9(UnU'nV) are open sets of EA and FM respectively. From 1.1.17 and 1.1.18 it follows that *cp-' and cpq." are maps of class p on cp(UnU'nV) and *(UnU'nV) respectively. Then, from the preceding remark 2 ) , (UnU',cpIuAu,,(E,h)) is a chart of X. In the same way (UAU',~~'~~~~,,(E~,A')) is a chart of X. Thus we have proved that UAU'EB.
.
If (X,[ A ] ) is a differentiable manifold, topology will be denoted by T
[dl
For topological proposition.
-
manifolds
(p=O),
we
the associated
have
a
similar
Remarks 1) If two topological manifolds, over the same set X, have the same associated topology, then these topological manifolds coincide.
Chapter 1
22
2 ) There exist different differentiable structures, Over a
same set, whose topologies are the same. Proposition 1.2.4 Let X be a differentiable manifold of class p and T the associated topology. Then if ( U , p , ( E , A ) ) is a chart of X, it follows that cp:U+cp(U) is a homeomarphism, where U is endowed with the topology T and cp(U) vith the topology induced by E. IU
Proof. It follows from the fact that, if A is an open set of -1 P ( W f ( c p ~ A L P , c p - l ( A l t ( E , A ) ) , i s a chart of x.0 The next proposition gives u s a criterion
tsl
knslw when a
topology, T I , on a manifold x is the assaciated top~logy.
Prorrosition 1.2.5 Let X be a differentiable manifold of class p, T €ha associated topology and T’ a topology on X. Then the following statements are equivalent: a) T=T’. b ) there is an atlas S-{(U.,p 1 i’ (Ei,hi))/irI) of X, such that
Proof. If two topuiugies agree on every element
slf
a carmion open
covering, then they must be equal.
This assertion and 1.2.4 g i v e the equiva1ence.a Boundary of a Differentiable Manifold Definition 1 . 2 . 6
Let E be a real Banach space, A=(Al,...fAn) a linearly + independent system of elements of L(E,R) and U an open set of EA.
Real Differentiable Manifolds with Corners
23
+ ve call A-index of x, and we denote by indA(x) or a) If xeEA only by ind(x), the cardinal of the set (i/hi(x)=O). b) The set (xeU/indA(x)rl) will be called A-boundary of U and denoted by aAU or by aU. c) The set (xdJ/ind,(x)=O) will be called A-interior of U and denoted by intAU or by intU. Example 1.2.7
+ E=R2 , A=(p1,p2) and U=EAnBr(0). Then ind((o,o))=2, r -))=l, ind( (f,;))=O. aU=( (x,y)~U/x*y=O)and ind( ( 2 1 0 ) ) = 1 1 ind( (0'2 intU=( (x,y)eU/x-y>O). Let r
Our next proposition shows some properties of the boundary and interior of open sets of quadrants. Proposition 1.2.8 Let E be a real Banach space, A=(hl,...,hn) a linearly + independent system of elements of Z(E,R) and U an open set of EA. Then we have: a) U-aAU=intAU=UnintAEA + b) U= ( aAU)uintAU and ( aAU)nintAU=@ c) If A=@, aAU=@ and intAU=U d) intAU is the topological interior of U in E. Hence intAU is an open set in E. e) If aAU##, then A## and intAU*@.
Proof. a),b),c) and d) are trivially proved. To see e) we note that, by c) , A*@ and by 1.1.4 there exist linearly independent elements of E l xl,. ,xn' such that 0 h1. (7~ . ) = 6 and ~ ~ E=EA8T L(xl, ,xn}.
...
..
a:E AOxR n--+E Since the map a(xIrl,...,rn)=x+r1x1+...+r nxn is a linear o + u(0))")=EA + and a(EAx(R 0 + ) n )=intAEA + , it a(EAx(R
given by homeomorphism, follows that
Chapter 1
24
Our purpose in the sequel is to prove that the boundary is preserved by diffeomorphisms. Definition 1.2.9. U a n o p e n set o f a q u a d r a n t o f E, V a n o p e n set o f a q u a d r a n t o f F and f:U+V a map. W e s a y t h a t f is a d i f f e o m o r p h i s m o f c l a s s p i f f i s a b i j e c t i v e map and f-’:V+u, f:u+v a r e maps o f c l a s s p. L e t E,F be r e a l Banach s p a c e s ,
Note that if XEU and f is a diffeomorphism of class prl, then Df(x):E+F is a linear homeomorphism by 1.1.9. Lema 1.2.10 L e t E be a r e a l Banach s p a c e ,
U a n o p e n set o f E, F a real
Banach s p a c e , A a n element of Z(F,R) w i t h h+O, f:U+F a map of 0 c l a s s pz1 w i t h f(U)cF: and xcu. S u p p o s e t h a t f(x)EFA. Then 0 Df(x) (E)cFA.
Proof. Let veE. Then from 1.1.8 we have that . If & > O , there is 6&>0 such that Df (x)(v)=lim f(x+tv)-f(x) t t+O x+tvcU and Df (x)(v)- f ( ~ + t ~ ) - ~ ( ~
I
I
&
& (x+tv)-f and by vt the element Ut . Thus we have Df(x) (v) t & 0 t&vf=tDf (x)(v)-f (x+tv)+ f (x) and since -f ( X)€FA th (Df(x)(v) =A ( f (x+tv)-f (x)) 20. Then it follows that:
-
-WE)
lst) If
O
&
0 is arbitrary, h(Df(x) (v))=O and Df(x) (v)eFA.o
+ then Df(x)(EA)cFA. + + Note that if in the preceding Lemma UcEA,
Real Differentiable Manif ol ds vith Corners
25
Lemma 1.2.11
...,
Let E be a real Banach space, A=(A,, An) a linearly independent system of elements of Z(E,W) and Jc(l,...,n)=I. We 0 denote by EJ the closed linear subspace of E, [I EA Then we IEJ j have that:
.
a) AJ=(A ilEJ/icI-J) is a linearly independent system of elements of Je(E,,W). LI+ b) (E ) + =EAnEJ. ‘J Proof. a) From 1.1.4 there exist linearly independent elements of 0 E, xl,...,xn such that Ai(x.)=Gij and E=EA BTL{xl, x,). 3
...,
C ri[AilEJ]=O. Then it follows that ieI-J for all j e 1 since Ai(x.)=O if jeJ and id-J and C riAi(x.)=O icI-J 3 3 X.EE if jeJ. We conclude that C r.A.=O and ri=O for all 3 J ieI-J 1 1 icI-J. Suppose
b) Is easily pr0ved.o
Theorem 1.2.12 (Boundary Invariance)
...,
An) a linearly Let E,F be real Banach spaces, A=(A1, M=(ql,. ,qm) a independent system of elements of I(E,R), linearly independent system of elements of P(F,R), U an open set + + of EA, V an open set of FM and f:U+V a diffeomorphism of class pzl. Then ve have that:
a) b) follovs c)
For all XEU, ind,(x)=in%(f(x)). aAU*# if and only if aMV*#. that A*# and M+#. f(intAu)=intMV, f 8 u =aMV.
(.I
..
Furthermore, if aMU*# it
d) flintAu:intAU+int MV is a diffeomorphism of class p.
Chapter 1
26
Proof. a) If XEU, k=indA (x) and kI=in%(f(x)) , then, exactly k vanish at x and exactly k' elements elements of A , Ail'. . . l A ik , vanish at f ( x ) . We consider J=(il,.. ,ik)l of M, qil,.. . , q ikI and hJ=(h /i€I-J]. ilEJ
.
From 1.2.11, hJ is a linearly independent system of elements + of !f(EJ,R) and H=(EJ)iJ=EAnEJ. Hence UnH is an open set of H and, using 1.2.8, it follows that int (UnH) is an open set of EJ AJ which contains the point x. We denoted by Vx the open set inth (UnH) . J x + On the other hand f :V +VcFM is a map of class p and, by 1 VX Lemma 1.2.10,
Hence
and
k' codimDf (x)(EJ ) =codim (EJ)=krcodim (I -J=1
=k8. Similarly, using f",
b) and c) follow from a) and d) follows from c).o natural question emerges. Is the preceding theorem true for homeomorphisms (p=O)?. The answer is no, because for any real Hilbert space of infinite dimension, El and any A d ( E , R ) - ( O ) , the + + o sets EA and EA-EA are homeomorphic. (See Convex Bodies and P e r i o d i c Homeomorphisms i n Hilbert S p a c e s , V.J. Klee, T.A.M.S. , V. 7 4 , 1953, pag. 31). A
But if the dimension is finite and p=O we have the following result:
.
"Let m,nckN, A=( A .., AP ) a linearly independent system of 1' elements of 2(Rn,R), M=(ql,. , g ) a linearly independent system
..
Real Differentiable Manifolds vith Corners
27
of elements of Je(RmlR), U an open set of (R"): with U+#, V an open set of (Rm)L with V*# and f:U+V a homeomorphism. Then a) m=n, b) aAU*@ if and only if aMV+#. Furthermore, if a,U+# we have A*# and M+#, c) f(int,U)=in%V and f(aAU)=aMV, d) :intAU+int V is a homeomorphism". f M I int,,u The proof is an easy consequence of the following result: *#LetB be a subset of Rp and f : M p a map such that f:B+f(B) is a homeomorphism with the euclidean topology. Then f(int(B))=int(f(B)) and f(Yr(B)nB)=3r(f(B))nf(B)88. (See, Topology, J. Dugundgi, pag. 359, Allyn and Bacon). Note, also, that the index of a point is not, in general, preserved by homeomorphisms. After 1.2.12, we can define the index of a point of a differentiable manifold. Proposition 1.2.13 Let X be a differentiable manifold of class p ~ land xcX. Then, if (U,p,(E,A)),(V,P,(F,M)) are charts of X such that xeUnV, ve have ind,(p(x))=in%(P(x)).
Proof. It follows from the fact that 9p-l is a diffeomorphism of class p.n Definition 1.2.14 Let X be a differentiable manifold of class phl and XEX. The A-index of p(x), where (U,p,( E , A ) ) is a chart of X vith XEU, will be called index of x and will be denoted by ind(x).
Chapter 1
28
Proposition 1.2.15 Let X be a differentiable manifold of class p and XEX. Then there is a chart (U,p,(E,h)) of X such that XEU and p(x)=O and hence ind(x) =card(A) , (Such a chart will be called centered chart at the point x). Proof. Let ( V , t k , (F,M)) be a chart of X such that xeV. We assume that M=(~,....,?I,,), vi (U(x))=O oir(q(x))=O and vi(U(x))*0 1
....,
for all i~(lr...,n)-(i1,...,ir). 1)
If
(ilr...ri~)={lr.~.rn~l take
the
chart
2) If (il,...,i~)#(1,...,n), take an open set W of U ( V ) such
that *(x)EW and WnFo =@ for all ie(lr...,n)-(il,...,ir} "i
and the where
Definition 1.2.16 Let X be a differentiable manifold of class pzl. a) F o r all ktrlu(O), the set (xeXlind(x)zk) is called k-boundary of X and denoted by akX. The set alX is also called boundary of X and denoted by ax. It is clear that a0X=X. the set (xeX/ind(x)=k} is denoted by BkX. The set BOX is called interior of X and denoted by IntX. b) For all kENw(O),
If X is a locally finite dimensional topological manifold and XEX, we say that x is a point of the boundary of X if there is a chart of X, (U,p,(E,A)) such that xeU and indA(p(x))>O. In this case, by 1.1.4.a), there is a chart (VIUr(Fl{~)))such that xeV and ind (q(x))=l, that is qU(x)=O. ("1
Real Differentiable Manifolds vith Corners
29
Proposition 1.2.17
Let X be a differentiable manifold of class p and kcN. Then ve have that: is a closed set of X. b) Int(X) is a dense open set of X. a ) 8%
Proof. It is an easy consequence of 1.2.15.0 Proposition 1.2.18
Let X be a differentiable manifold of class p and keMu(0). Then there exists a unique differentiable structure of class p on BkX such that for a l l xeBkX and a l l charts (U,p, (E,A)) of X with 0 xeU and cp(x)=O (1.2.15), the triplet (UnBkX,p,UnB X,EA) is a k chart of that structure. Furthermore BkX has not boundary, that is B(BkX)=#, and the topology of the manifold BkX is the topology induced by X. Proof. If xeBkX and (U,cp, ( E , A ) ) is a chart of X such that xeU and 0 cp(x)=O, we have p(UnBkX)=cp(U)nEA. From this equality it is easy to deduce the assertions in the statement.0 Corollary 1.2.19
Let X be a differentiable manifold of class p. Then ve have: There is a unique differentiable structure of class p on Int(X) such that for a l l xeInt(X) and all charts (U,p,(E,A)) of X 0 with xeU and p(x)=O (hence A=#, UcInt(X) and EA=E) the triplet (U,cp,E) is a chart of IntX. Furthermore Int(X) has not boundary and its topology is the topology induced by X. a)
b) If a 2X=#, there is a unique differentiable structure of class p on 8X such that for a l l xe8X and a l l charts (U,p,(E,h)) of X with xeU and cp(x)=O (hence A=(h)), it happens that
Chapter 1
30 0
(Urtax,u,IunaxrEA)is a chart of boundary and its topology is X.0
ax.
Furthermore the topology
ax
has not induced by
Examples of Differentiable Manifolds A ) Let E be a real Banach space and A a finite linearly + is endowed with independent system of elements of Z ( E , ! R ) . Then EA a differentiable structure of class 0 0 , defined by the atlas of + class m ( ( E A r 1 +,(E,h)) ) . It is clear that the topology of this
+ manifold is the topology induced by E . Furthermore aEi=aAEA and + + IntEA=IntAEA. In particular if A=@, E is a differentiable manifold of class m, without boundary, defined by the atlas of class {(ErlE,(Er@)))* Note that if E is a Banachable space and 11 11, 1 Ill are compatible norms in this space, then the manifolds of class m , ( E , ! 11) and (E,II 1), are the same. B) Let X be a differentiable manifold of class p and G an open set of X. Then there is a unique differentiable structure of class p on G such that any chart of X, whose domain is included on G, is a chart of that structure. Furthermore akG=(a kX)nG and BkG=(BkX)nG for all keNu(0).
us consider n&i, Sn={xeR n+1/llx~~=l) and for all ., .. + + n) , (U:,pT,R n ) , + n/x.>O), ie(1,. . .,n+l), (Ui,pi,R where Ui=(xeS 1 + u;=(x€s n/Xi<0), and rPi(X)=(X,i * * * rxit rXn+l) ~i(y)=(ylr ...,yi,.."y,+1)' Then it is easily proved that + + n A=( (ui,pi," )/i=l,. . .,n+l)v( (u;,p;,d')/i=l,. ..,n+l) is an atlas C)
Let
A
---
of class m on Sn. Furthermore the topology of the manifold (S", [ A ] ) is the usual topology of Sn and asn=@.
It is clear that class m. Let ie( 1,. . ,n), D)
.
us
S
0
=(-1,l) is a differentiable manifold of
consider
neN,
+ + n (Vit*ir(" 'Pi
Real
Differentiable
+ V.=(xeD 1 n/xi>O) + qi(x1r *
J1-(xl+ 2
-
-
*
Manifolds
with Corners
VT=(xeD 1 n/xi<0)
tXn)=(xlr
*
-
*
rXi-1,
...+x.+...+Xn)-Xi'Xi+l'...'Xn)' "21 2
*i(X1r***rX n )=(X1r---rXi,lr "2 2 +x 1.+. +xn)-xi,X~+~,...,Xn ) and pi(xl,
2 -J 1- (xl+. ..
31
..
- -
...'xn )=xi.
Then it is easily proved that + + n ( ~ " , - p ~)/i=lr.. ) ,n)u B=( (Vi' qir (R ,pi))/i=lr.. 'n)u( (virqir n u((Bl(0),j,R ) ) is an atlas of class m on Dn. Furthermore the topology of the manifold (Dn,[B]) is the usual topology of Dn, aD =S n-1 ,a 2Dn=# and IntDn=B1(0). n
.
.
Finally the differentiable structure of aDn, constructed in n ' l 1.2.19. b) coincides with the differentiable structure of S constructed in the preceding example C). E)
O=vl, ...'v
Let 2"
us consider nsH, the vertices of Cn.
Cn={xeRn/Osxlslr..
. O=xnsl)
and
..
For all ie(l,2,. '2 n ) we consider ( U i r p i r (Rn,h)) where h = { ~ ~ ~ . . . ~ p , )Ui ~ is the set (xsCn/x does not belong to Hi) where Hi is the union of the closed faces which are not adjacent to vi and pi:U.+U1 is the bijective map defined by: 1 1 n V 1 vi x1 n ,vi+ (-1) i xn) , where pi ( xl, xn)= (vi+(-1) 1 n v 1= (v. ' ,Vi) 1
...
.
... . .. .
.
n n ) is Then it can be shown that dl=((Uirpi,(R ,A))/i=lr...'2 an atlas of class m on Cn. Furthermore the topology of the manifold (Cn,[dl]) is the usual topology of Cnr aCn is the union of the closed faces of Cnr anCn= ( V ~ ~ . . . ~ V ) and arcn=# for all r>n. 2" F) Grassmann Manifolds Let E be a real Banach space and G(E)=(HcE/H is a closed linear subspace of E that admits a topological supplement in E}. For
The set ( ( F , G ) s G ( E ) x G ( E ) / E = F @ , G ) will be denoted by I ( E ) . every element (F,G)eI(E), the linear homeomorphism,
Chapter 1
32
(x,y)eFxG~x+yeE,will be denoted by 8(F,G). For every element GeG(E) , the set (HeG(E)/E=HeTG) is not empty and will be denoted by UG. For all (F,G)eI(E), the map (p :UG+P(F,G) is defined as (F,G) where uH is given by the diagram G-F
I
U
H
t
I -p2
where j is the inclusion and p2(x,y)=y for all (x,y).
-' (H) for
Now, it can be shown that graph(cp (F,G)(H)) =8 (F,G) (F,G ) €1 (E) and HeUG.
$I=(
In the sequel we will prove that (UG,Y'(F,G),I(F,G))/(F,G)eI(E)) is an atlas of class m on G(E).
1) For every (F,G)eI(E), (UG,cp ,Je(F,G)) is a chart of (FIG) G(E). Indeed, if H,H'eUG with H#H', by the preceding formula for the graph we have cp ( F , G) (H') and hence cp is an +V (F,G) (F,G) injective map. If ueJe(F,G), it follows that graph(u) is a subspace of FxG and H=B(F,G) (graph(u) ) is a subspace of E. Furthermore E=HeTG and hence H is UG. Finally cp (H)=u and cp is a surjective (F,G) (F,G)
closed linear closed linear an element of map.
2) It is clear that the domains of the charts of rB cover G(E)
*
3 ) Let (F,G),(H,K) be elements of I ( E ) . Then we have the following cases:
G=K. In this case, there is a&(H,K) such that (u)=uo(p,e(F,G)-lk)-a for all ueZ(F,G), where k is I P w , ~O'"(F,G) ) the inclusion of H in E and p1 is the projection of FxG on F. -1 Therefore and the charts is of class m cp (H,K) (F,G) ('GP cp (F,G) ,Je(F,G)), (UK,cp(H,K),P(H,K)) are compatible of class m. 3)
-1
Ocp
Real Differentiable Manifolds with Corners
33
3)2 F=H. In this case we consider the linear continuous
map, p:G+F, defined by the diagram G
'
> F
Then the map, a:G+E, defined by a(x)=p(x)+x is a linear homeomorphism of G over K and if LEUG we have that IgLEUK if and only if lF-,90(p (L):F+F is a linear homeomorphism". Therefore (F,G) (p(F,G) (UGnUK)=(ud(F,G)/lF-f30u:F+F is a linear homeomorphism) and (U nu ) is an open set of by 1.1.16.d) it follows that (p (FrG) G K Je(F,G). In the same way (p (U nu ) is an open set of P(H,K). W,K) G K For
all
UE(P(F,G) (UGnUK) we
have
and au ( l,-pu) -l) -1 Then, (u)=au (lF-Bu) (p(H,K)(p(F,G) -1 P(H,K)P(F,G) :(P(F,G) ('GnUK)3(H,K) ('GnUK) -1 analogously cp is of class m. (F,G) (H,K) =8 (F, K) (graph
8 (F,G) (graph (u)) =
that
.
-1
is
by
therefore 1.1.16,
Of
and
3) F+H and G*K. Let L be an element of UGnUK. By 3) 1, -1 we have that (p :Z(H,K)+I(L,K) is a diffeomorphism of (L,K) (H,K) (U nu ) is an open set of class m. By 3)2, we have that (p (L,K) G K Je(L,K). Therefore (p (U nu ) is an open set of P(H,K) and (H,K) G K analogously we have that p (U nu ) is an open set of Je(F,G). (F,G) G K OP
Finaly
it is easy are of class P ( ~ , o~ P)( ~ , ~ )
-'
to m.
prove
that
-1
and
(p(~, o( ~p )( ~ , ~ )
The differentiable manifold of class m, (G(E),[d]), will be called Grassmannian of E. It is clear that BG(E)=#. G) Real Projective Spaces Let E be a real Banach space and (G(E),[ I s ] ) the Grassmannian manifold of E, constructed in F). Then the set G,(E)=(FEG(E)/dimF=l) is open and closed in (G(E), [ a ] ) . Indeed, if FcG1(E), there is GEG(E) such that E=FsTG. It is clear that
Chapter 1
34
F€UGcG1(E). Therefore G1(E) is an open set of G(E). F€G(E)-G1(E) and G is an element of G(E) such that E=FeTG, we have that FdJGcG(E)-G1(E). Then G1(E) is a closed set of G(E)
If
.
By the example B), there is a differentiable structure of class m on G1(E) , induced by the Grassmann manifold (G(E) ,[ d ] ) . The manifold associated to E.
G1(E)
will
be
called
projective
space
It can be shown that for all kcP(u(O), Gk(E)=(FeG(E)/dimF=k) and Gk (E)=(FEG(E)/codim(F)=k) are open and closed sets of (G(E) , [ S ] 1 . Then, by the example B) , the sets Gk(E) and Gk (E) are These differentiable differentiable manifolds of class m. structures are induced by the manifold (G(E) ,[ A ] ) . To end this example, if FeG(E) we have that the set G(F,E)=(HeG(E)/H is linearly homeomorphic to F) is an open and closed set of (G(E),[AJ). Then, by example B), G(F,E) is a differentiable manifold of class m called Grassmann F-manifold of E.
H) Let E be a real Banach space and F a closed linear subspace of E which admits topological supplement in E. Then the set S(F,E)=(ueZ(F,E)/u is an injective map and im(u) admits topological supplement in E) is an open set of Z(F,E) (See 3.2.9) and, by example B), S(F,E) is a differentiable manifold of class m, called Stiefel F-manifold of E. If we consider the equivalent relation on S(F,E),
-, given
by : "u-u'athere is vcGL(F) such that UV=U'*~, then we have that the map, a:S(F,E)/-+G(F,E) , defined by a([u])=im(u) is a bijective map. 1.3. Differentiable Maps.
First we introduce the weakened differentiable manifolds of a given manifold.
Real Differentiable Manifolds vith Corners
35
Proposition 1.3.1 Let (x,[d]) be a differentiable manifold of class p and qePI such that q
b) If S1,B2 are atlases of the manifold (X,[d]), it occurs that the atlases of class q on X, B1 and S2, are equivalent of class q and so [SlIq=[B2lq. c) The topologies of (X,[d]) and (X,[d] ) coincide. q d) For every XEX the index in the manifold (X,[d]), same that the index, in the manifold (X,[d] ) . q
Therefore ak (X,[d])=a k (X,[d] ) and Bk(X,[d])=Bk(X,[d] q
a l l kcNu( 0) .u
is the
)
9
for
The differentiable manifold (X,[d] ) will be called weakened q differentiable manifold of (X,[d]). The statements a),b) and c) of the preceding proposition are also true if q=O, in which case the corresponding weakened differentiable manifold is a topological manifold. Later, in 10.2.3, it will be proved that if (X,[d]) is a 1 satisfying certain differentiable manifold of class hypothesis, then there is a differentiable structure of class m on X, [El, such that (X,[E],)=(X,[d]). However, Kervaire has proved that there is a topological manifold, modelled over do, which does not admit a differentiable structure whose associated topology coincides with the topology of the given topological manifold. ( A manifold vhich does not admit any differentiable structure, 1960).
36
Chapter 1
Recently, S . Donaldson has proved that there are topological 4 manifolds, modelled over R , which do not admit compatible differentiable structures. ( A n application of Gauge Theory to four dimensional topology, 1983). Definition 1.3.2 Let X,Xf be differentiable manifolds of class prl and f:X+X' a map. We say that f is a map of class p or a Cp-map, if for every xeX there is a chart (U,cp, ( E , h ) ) of X at x and there is a chart ( V f 9 , ( F , M ) )of Xf at f(x) such that f(U)cV and the map 9fp-l :cp (U)+9 (V) is a map of c l a s s p.
This definition extends naturally to the case p=O, that is, to topological manifolds. Note that if (X,[S]) and ( X ' , [ S ' ] ) are differentiable manifolds of class p and f:X+X' is a map of class p, then for all qeW{O) with q
q'
In the category of topological manifolds a Co-map is nothing but a continuous map. From this fact we conclude that every $'-map is a continuous map. Also, it is easy to prove that if f:X+X' is a map of class p, (U,cp, ( E , h ) ) is a chart of X and ( V , 9 , ( F , M ) ) is a chart of X' such that f(U)cV, then +fcp-':p(U)+9(V) is a map of class p.
Finally if f:U+F is a map, where U is an open set of a quadrant of a real Banach space, E, and F is a real Banach space, the notion of map of class p given via 1.1.14 coincides with the notion of map of class p defined above, considering U and F as differentiable manifolds with the usual differentiable structures.
Real Differentiable Manifolds with Corners
37
Proposition 1.3.3
a) If X is a differentiable manifold identity map, 1X :X-+X, is a map of class p.
of class p,
the
b) The finite composition of Cp-maps is a Cp-map. c) Let X be a CP-manifold and U an open set of X. Then the inclusion map, j:U+X, is a map of class p. d) If f:X+Y is a map of class p and U is an open set of X, then f. :U+Y is a map of class p. IU e) If X,Y are differentiable manifolds of class p, f:X+Y is a map and 41 is an open covering of X such that f u:U+Y is a map I of class p for all UE'U, then f:X+Y is a map of class p.0 Definition 1.3.4
Let X,X' be differentiable manifolds of class p and f:X+X' a map. Then f is called a diffeomorphism of class p if f is a bijective map and f ,f-l are maps of class p. This definition extends naturally to the case p=O, that is, to the case of topological manifolds and then the notions of diffeomorphism of class 0 and homeomorphism coincide. Obviously, if f is a diffeomorphism of class p, then f is a diffeomorphism of class q for all q&W(O) such that q
Chapter 1
38
(W, [ s f ' ] ) .
Indeed, it suffices to take
sf=( ( R , l R , R ) ) ,
A'=( ( R , q , R ) ) ,
3
where p(t)=t3 and f(t)=fi. In order to prove that the diffeomorphisms of class pzl preserve the index and hence the boundary we need the following lemma whose proof is immediate: Lemma 1.3.5 Let X be a differentiable manifold of class p, U an open set of X, E a real Banach space, A a finite linearly independent system of elements of X(E,R) and (p:U+E a map such that p ( U ) is an + open set of EA. Then the folloving statements are equivalent: a) ( U , p , ( E , A ) ) is a chart of the manifold X. b) " p U + p ( U ) is a diffeomorphism of class p, where U and q ( U ) have the differentiable structure described in the example B) of the preceding paragraph. o Theorem 1.3.6 Let X,Xf be differentiable manifolds of class ptl and f:X+X' diffeomorphism of class p. Then we have that: a ) ind(x)=ind(f(x)) for all XEX, b) f(a kX ) = a kX' and f(BkX)=BkX' for
a
all kdNu(0). Proof. a) Let XEX and (U,p , ( E , A ) ) a chart of X such that xeU and cp(x)=O. Then by the preceding Lemma (f(U),qf-', ( E , A ) ) is a chart X' such that f(x)Ef(U) and pf-l(f(x))=O. Hence of ind(x) =card ( A ) =ind( f ( x ) )
.
b) Follows from a).o
Real Differentiable Manifolds with Corners
39
Proposition 1.3.7 Let X,X' be differentiable manifolds of class p and f:X+X' a diffeomorphism of class p. Then, for all k d " ( 0 ) , f I BkX:BkX+BkX' is a diffeomorphism of c l a s s p, vhere BkX and BkX' are the manifolds described in Proposition 1.2.18. In particular, if a2X=#, f is a diffeomorphism of class p of aX=BIX onto aX'=BIX'.
Proof. Let xcBkX and consider a chart (U,p,(E,A)) of X such that 0 XCU and p(x)=O. Then (UnBkX,pIUnB EA) is a chart of BkX. k
of
1( ' (
Since f is a diffeomorphism, (f(U),q=pf-', (E,A)) is a chart f (XIEf (U) and pf-l (f(x)) =o. Hence X'I and 0 nBkX' I' 1 f (U)r\BkX#I EA) is a chart of BkX'. Then it is clear
that 9f(p I UnBkX)
f (UnBkX)=f (U)nBkX' and the 0 0 is the identity map. -1:p(U)nEA+p(U)nEA
map Therefore
f lBkX and [flBkX]-l are maps of class p.0 If X and X' are locally finite dimensional topological manifolds and f:X+Xt is a homeomorphism then, by Brouwer's theorem, we have f (ax)=ax'
.
The main purpose of Differential Topology is to study the properties of the differentiable manifolds which are preserved by diffeomorphisms. The idea of differentiable manifold can already be seen in Riemann's work about analytical prolongation (1860). But the first abstract definition of differentiable manifold was given by Hennann Weyl and it can be seen in his book on Riemann Surfaces (1912). The works of H. Whitney, developed after 1936, gave a great impulse to the study of the differentiable manifolds, but it is in 1956, with Milnor's work, when Differential Topology reachs special importance.
Chapter 1
40
In these works Milnor constructs several differentiable 7 structures of class m over the sphere, S , whose associated topologies are the usual topology of S7 and any two of them are not diffeomorphic (see [E-K]). It is well known that if X is a 1, 2 or 3 dimensional differentiable manifold which has certain topological properties, then any two differentiable structures over X whose associated topologies are homeomorphic, must be diffeomorphic (see [LA] and EK-SI)
*
Examples of Differentiable Maps
A) Let [0,1] be the differentiable manifold of class a of Example E) of differentiable manifolds and let S1 be the differentiable manifold of class m of Example C). Then, the map f: [O,l].+SL defined by 2nit f (t)=(cos2ntrsen2nt)=e is a map of class m. B) Let E be a real Banach space. Consider the real Banach space, E =2 (E,R ) , and the Grassmannians G (E) and G (E‘)
.
I
For all FEG(E) we denote by F the set {AeE’/A(F)=(O)). It I can be shown that F is an element of G(E’). (See the result about projectors that follows t o 1.1.2.) I
.
I
On the other hand if E=FoTG one sees that E’=F eTG Also, I. I I if E=Fe G and u is an element of P(F,G) , then the map u .F 4 T I defined by u (A)=hupl6-l is a continuous linear map, where p 1 :FxG+F is the projection map and 8:Fx-E is the map given by 6 (x,y)=x+y. Let us see that the map a:G(E)+G(E’), an injective map of class m.
defined by a(F)=F
.
I
,
is
Indeed, let F be an element of G(E) Then there exists GEG(E) such that E=Fs G and hence (F,G)€I(E). Therefore T (UG,p(F,G),Y(F,G)) is a chart of G(E) such that FdJG.
Real Differentiable Manif ol ds vi th Corners
Since
I
E’=F
Q
I
TG
,
it
follows
I
that
I
(F ,G )EI(E,)
41 and
I I
(U *,p I I ,f(F ,G ) ) is a chart of G(E’). G (F ,G 1 On the other hand u(UG)cU I and we have the commutative G diagram U
B
f(F,G)
. Since p
I
where B(u)=-u a map of class
I u,
,f(F
is a map of class
m,
I
I
IG 1
it follows that u is
m.
In order to see that a is an injective map, it suffices to apply Hahn-Banach theorem [E.V.T., 1.39,Bl. C) Let E be a real Banach space and keNu(0). Consider the real Banach space E’=f(E,R) and the manifolds Gk(E) , Gk(E’) constructed in Example G). Then the map u of class m constructed above, sends the :Gk (E)-Gk(E’) manifold Gk (E) onto the manifold Gk(E’) and u
IG~(E)
is a diffeomorphism of class m. In order to see that, note that if h=(hll...,Ak) is a linearly independent system of elements of E’=I(E,R) then (E;)I=L(A,, ,Ak).
...
1.4. Topological Properties of the Differentiable Manifolds In this paragraph the differentiable manifolds will be studied from the point of view of General Topology: countability, separability, and covering axioms, connectedness, dimension, etc.
Chapter 1
42
Proposition 1.4.1 Every differentiable manifold satisfies the first axiom of countability, [M-0-PI VI.1.21.0 real Banach space that is not separable is a differentiable manifold which does not satisfy the second axiom of countability [M-0-PI VI.2.11. A
Proposition 1.4.2 Let X be a differentiable manifold. Then the following statements are equivalent: a) X satisfies the second axiom of countability. b) X satisfies the Lindelof property [M-0-PI VI.3.11 and has an atlas vhose charts are modelled over separable real Banach spaces. c) X has a countable atlas vhose charts are modelled over separable real Banach spaces.0 Let us see an example of differentiable manifold of class m which is separable but does not satisfy the second axiom of countability: Let X be the set (Rx(O))u((O)x(R+u(O))). Then consider + c0=(WX(O),~~,R),where cp 0 (r,O)=r and c = ( UtllptlR) for all telR I with Ut=(Rx(0)-( (0,O) ) ) u ( (0,t)) and cpt(rlO)=r, cpt(Olt)=O.
+
It is easy to see that 8=(co)u(ctlteR ) is an atlas of class m over X. Moreover the associated topology does not satisfy the second axiom of countability, it is separable and it does not satisfy the Lindelof property. Finally this manifold does not satisfy the Hausdorff axiom and satisfies the T1 axiom. (i.e. its points are closed). In fact any differentiable manifold satisfies the T1 axiom.
Real Differentiable Manifolds with Corners
43
Proposition 1.4.3
Let X be a differentiable or topological manifold. Then the following statements are equivalent: a) The associated topology satisfies the Hausdorff axiom. ) / the ~ E I manifold ) b) There is an atlas s ~ = { ( U ~ , ( ~ ~ , ( E ~ , A ~ )of X such that for all (i,j)eIxI, the graph of the map (p.(pT1:(pi(UinU.)+(p.(UinU.) is a closed set of (pi(Ui)x(p. (U.).o 3
3 1
3
3
3
3
ProPosition 1.4.4
Let X be a differentiable or topological manifold. Then: a) X is a locally connected space. Therefore every connected component of X is open and closed, ([M-0-P, V.5, pag. 24]), b) X is a locally path-connected space. Therefore X is a connected space if and only if X is a path-connected space ([M-0-P, V.5, p. 90]).0 Proposition 1.4.5
Let X be a differentiable manifold, C a connected component of X and X,ZEC. Then dimxX=dimzX. Therefore, if X is a connected space, dimxX is constant when xcX. Proof Since
---
C
is
an
open
set
in
X,
there
are
(U1,c1)I I (UnlPn) of X such that U 1nu 2+$,U2nU3*$, .,Un-lnUn+$, zcU, and UicC for all i e ( 1 , ([M-0-P, V.5, pg. 231). It follows dimxX=dimzX.o Theorem 1.4.6
..
charts XEUl I
...,n),
(Riesz's Theorem, Concerning Differentiable Manifolds)
Let X be a differentiable manifold. Then the folloving statements are equivalent: a) X is a locally compact space, b) X
Chapter 1
44
is locally of finite dimension. Proof
.
a) + b) Let x be an element of X and (U,cp, ( E , A ) ) a chart of X with XEU. Since X is a locally compact space, there is a locally compact open ball of E. Therefore E is a locally compact space and, by Riesz's theorem, E is a finite dimensional Banach space, [D. pg. 1093 and dimxX is finite. b) + a) Let x be an element of X and (U,cp,( E , A ) ) a chart of X with XEU. Then, by the hypothesis, we have that E is a finite + dimensional Banach space and therefore E , E A and p(U) are locally compact spaces. Then U is also a locally compact space.0 It is well known that every locally compact Hausdorff topological space is a Baire space, (Baire's theorem, [M-0-P, V.2, pg. 2401). From this result we deduce that every Hausdorff locally finite dimensional differentiable manifold is a Baire space. In fact we shall see, by a direct proof, that every differentiable manifold is a Baire space. Proposition 1.4.7 Let X be a differentiable or topological manifold. Then X is a Baire space. Proof. Let x be an element of X and (U,cp,(E,A)) a chart of X with + + XEU. Since EA is a closed set of E, we have that EA is a complete + metric space. Then EA is a Baire space and therefore cp(U) is a Baire space. Then it follows that U is a Baire space, thus X is locally a space of Baire and hence X is a Baire space ([M-0-PI V.2, pg. 240 and 241]).0 It is well known the result of General Topology that every Hausdorff locally compact topological space satisfies the Tychonoff axiom [M-0-P, V.2, pg. 2311. By this and the Riesz's theorem every Hausdorff locally finite dimensional differentiable manifold satisfies the Tychonoff axiom. This last affirmation is
Real Differentiable Manif ol ds vith Corners
45
not true for arbitrary Hausdorff differentiable manifolds. In [M.O.l] there is an example of a Hausdorff connected differentiable manifold X of class m, such that ax=#, X is not regular and X admits an atlas whose charts are modelled over an infinite dimensional separable real Hilbert space. Proposition 1.4.8
Let X be a regular differentiable manifold. Then for every xcX and every chart (U,p,(E,A)) of X at x there is an open set V of X, vith xrVcU and such that a set YcV is closed in X if and only if p(Y) is closed in E. Proof Since Ei is a closed set in E it follows that there is an + open set A of EA such that p(x)rAczcp(U) , where is the closure of A in E. Because of the regularity of X I there is an open set V of X, such that x ~ V c k p - ~ ( A ) c p - ~ ( ~ ) cNOW, U . it is easily checked that V satisfies the conditions of the statement.0
x
The next theorem concerns metric manifolds. Theorem 1.4.9.
lSmirnovl
Let X be a differentiable or topological Hausdorff manifold. Then the folloving statements are equivalent: a) X is a metrizable space, b) X is a paracompact space. This theorem
is a particular
case of
Smirnov's
theorem
( [M-0-PI V.4, pg. 2451).
Theorem 1.4.10
Let X be a differentiable or topological metrizable manifold. Then there exists a metric d on X such that the topology associated to d is the topology associated to the manifold X and the space (X,d) is a complete metric space.
Chapter 1
46
Let x be an element of X and (U,p,(E,A)) a chart of X with + xeU. Let V be a closed neighbourhood of cp(x) in EA such that Vccp (U) Then VX=cp-'(V) is a completely metrizable neighbourhood of x and by [M-0-P, V.4, XI.2.353 , the manifold X is completely metrizable.
.
ProDosition 1.4.11 Let X be a connected differentiable or topological manifold which satisfies the T3 axiom and whose charts are modelled over separable real Banach spaces. Then the following statements are equivalent: a ) X is a metrizable space, b ) X satisfies the second axiom of countability, c ) X is a Lindelof space, d ) X is a para compa ct space
.
Pro0f a)
9
d) is a particular case of 1.4.9.
a) * b). By Sierpinski*s theorem, X is a separable space ([M-0-P, V.2, VI.4.22 3 ) . Therefore X satisfies the second axiom of countability ([M-0-P, V.2, VI.4.151). b) + c) Is a well known result of General Topology. c) + d). It follows from a theorem of E. Michael about paracompact spaces ([M-0-P, V.4, pg. 9 ) ) . o Note that Sierpinski's theorem, mentioned above, allows us to conclude that: "Every connected metrizable differentiable or topological manifold whose charts are modelled over separable real Banach spaces, is a separable spacell. Finally, note that in the preceding proposition we cannot change the hypothesis 1aT311 by the hypothesis 'lHausdorf f" , since if we take the open set YuD, where D is a countable dense set of Ker(h) in the manifold (X,[d]) described in [M-0,1], we have that
Real Differentiable Manifolds with Corners
47
YuD, endowed with the structure induced by [ & I , is a Hausdorff connected differentiable manifold of class m whose charts are modelled over separable real Hilbert spaces and this manifold satisfies the second countable axiom but does not satisfy the T 3 axiom (hence it is not a metrizable space). Corollary 1.4.12 Let X be a differentiable or topological manifold that satisfies the Tg axiom and whose charts are modelled over separable real Banach spaces. Then the folloving statements are equivalent: a ) X is a metrizable differentiable manifold, b) all connected components of X fulfil the second countability axiom, c) all connected components of X are Lindelof spaces, d ) the manifold X is a paracompact space. Proof. Every connected component of X is open and closed. Therefore X is the topological sum of its connected components. On the other hand every connected component of X is a differentiable manifold that fulfils the T3 axiom and whose charts are modelled over separable real Banach space. Having in mind the properties of the topological sum about metrizability and paracompactness, the result follows from 1.4.11.0 Topological Dimension in Differentiable Manifolds Let X be a set and U=(Ui)icI a family of subsets of X with We consider the set H=(neHu( O)/there are different elements ill..., in+l of I such that Ui n.. .nui +#). Then Sup(H) 1 n+ 1 is an element of (O)uHu(+oo) and it is called order of 2L (ord(41)). If 41=(Ui)icI is a family of subsets of X with Ui=@ for all ieI, we say that the order of 41 is -1. ( i/Ui+#)t#.
Let X be a topological space. Consider the set F=( ncHu( -1 ,0 ,+-)/every finite open covering of X has an open
48
Chapter 1
refinement V , such that ord(V)sn). Then the minimum of F is an element of the set (-1,O,+m)ulN that will be called covering dimension of X and will be denoted by dim(x). It is clear that dimX=-1 if and only if X=#. Also, it can be easily proved that the covering dimension is a topological invariant. We have the following results concerning covering dimension: A)
Let X be a topological space and C closed set of X. Then
dim(C)sdim(X) (see [PE] p. 1141).
B) Let X be a normal topological space and (Ci)iEm a closed covering of X such that dim(Ci)sn for all ifm, where ndNu(-l,O,+m). Then dim(X)an (see [PE] p. 1251). C) For all nEMu(O), dim(Rn)=n. (see [PE, p. 1271).
It follows from A), B), C), that all open or closed balls of W n have covering dimension n. Theorem 1.4.13 Let X be a second countable Hausdorff differentiable manifold. Suppose that dimxX=m for all xfX. Then dim(X)=m. Proof. From 1.4.12, it follows that X is a Hausdorff paracompact space and hence X is a normal space. From 1.4.8 we have that for all XEX there is a chart (UXIpXI(EXIAX)) of X at x such that a subset Y of Ux is closed in X if and only if px(Y) is closed in Ex. Since X is a regular space, for all XEX there is an open neighbourhood Vx of x in X such that vxcUx. As U=(Vx)xfx is an open covering of X and X is a second countable space, there is ( V ) cp1 such that ( J V =X. Moreover xn ncm ncm *n
Real Differentiable Manifolds vith Corners
49
(v ) is a closed set in E and dim9 (v )sm xn xn xn xn xn for all nd" Therefore dimvx im for all new and dimXm. n
we have that cp
On
the
other hand if dim(X)sm-l we would have that )im-l. But this is a contradiction dim(v ) i m - l and dimcpx xn n n because the closed balls of Rm have covering dimension equal to
(vx
m.o
Lemma 1.4.14 (Milnor) Let X be a regular paracompact topological space (in particular X is a Hausdorff paracompact topological space) such that dim(X)=m, vhere m is an element of N w ( 0 ) . Then for every open covering, U=(Ui/ie1), of X there are families VII...,Vm+l of open sets of X such that Y=V1w...vVm+l is a locally finite refinement of U and for all iE(1 ,...,m+l) the elements of Vi are pairvise disjoint ([PI p . 7). Theorem 1.4.15 Let X be a second countable Hausforff differentiable manifold such that dimxX=m for all XEX and a 2X=#. Then dim(X)=m and the manifold X has an atlas vith exactly m+l charts. Proof. By 1.4.13 we have that dim(X)=m. By Lemma 1.4.14 there is * * * a countable atlas V =V 1v... wYm+ 1 of X such that for every ic(1,. .,m+l) the domains of the charts of V: are pairwise disjoint
.
.
2
*
Since a X=# and Y1 is a countable family we can suppose that * for any two charts, (U,cp) and (V,q) of V1, cp(U)n*(V)=#. The same * * remark holds for the families V2,...,Vm+1.
*
Then V1 induces a chart of the manifold X whose domain is * the union of the domains of the charts of Vl, and analogously for * * the families V2,...,lfm+l. Then these m+l charts are an atlas of the manifold X.0
Chapter 1
50
1.5. Differentiable Partitions of Unity The differentiable partitions of unity are an important tool to construct differentiable maps, sections of fibered spaces, Riemannian metrics, etc. First we reduce the study to the local study. Definition 1.5.1 Let X be a differentiable manifold of class p and Q=(Qi/i€I) a family of functions of class p of X into R . We say that Q is a partition of unity of class p in X if the following statements hold: p.1) Iki(x)tO for all icI and for all xcx p.2) The family (Supp(Qi)/icI) is a locally finite family, vhere Supp(Qi)=(xeX/Qi(x)fO). Note that for all xeX, the set (ieI/Oi(x)*o) is a finite set. p.3) For all XEX, 1 9i(x)=l. ie1
Definition 1.5.2 Let X be a differentiable manifold of class p, U=(Ui/iEI) an open covering of X and e=(Qi/ieI) a partition of unity of class p in X. We say that O is subordinated to U if Supp(Qi)cUi for all ieI.
Definition 1.5.3 Let X be a differentiable manifold of class p. We say that X admits partitions of unity of class p if for all open covering 21 of X there is a partition of unity Q of class p in X subordinated to 21.
Definition 1.5.4 Let E be a real Banach space. We say that E satisfies the
Real Differentiable Manifolds vith Corners
51
Urysohn condition of class p if for every pair of closed sets A and B of E such that A+$, B+$ and AnB=$, there is a function f:Ea[O,l] of class p such that f ( A ) = ( O ) and f(B)=(l).
Proposition 1.5.5 Let X be class p vhose satisfy the partitions of
a paracompact Hausdorff differentiable manifold of charts are modelled over real Banach spaces which Urysohn condition of class p. Then X admits unity of class p.
Proof. Let U=(Ui/icI) be an open covering of X. Since X is a paracompact Hausdorff topological space, X is a T4 space, that is, X is a normal space and its points are closed. Then, by 1.4.8, for all xeX there is a chart (Vx,(px,(Ex,hx)) of X at x such that: 1) Vx is included in some element Ui of U , 2 ) a X
subset Y of Vx is closed in X if and only if (px(Y) is closed in + Ex, 3 ) if Vxf(x), then (px(Vx)f(E Since X is a paracompact space, there is a locally finite open refinement, V=(V./jcJ), of (Vx/xeX). Since X is a T4 space, 3 there is a contraction W=(W./jcJ) of V such that W.+O for all jeJ 3 3 (that is, W is an open covering of X such that for all jEJ v.cV 1. j and Wj+#). By the same argument, there is a contraction 8=(G./jcJ) of W such that G.+$ for all j€J. 3
3
For all jcJ there is X.EX such that V.cV since V is a 3 3 Xj' refinement of (Vx/xcX). Let j be an element of J and suppose, first, that Vx +(xj). j (E ) + j' ' x j C.+$ and 3
. and therefore (px. (Ej) and 3 3 3 7 3 x j (W.)=C. are closed sets of Ex such that p(, (Fj)f$, j 3 3 j j (G ) nC =# . Then, by the hypothesis, there is a map of G .cE.cw. cV cV
Then -(px (p
xiJ
class p, f.:Ex --t[O,l] such that f.((px (E.))=(l)and f.(C.)=(O). 7 3 I j I j 3 Since X=(X-v.)uVx , it is clear that the function g :X+[O,l] I j j
Chapter 1
52
defined by j
9j(XI’
if x N x j is a function of class p because of g.(X-g.)=(O). 3
If Vx =(x.) we take the function 3 j 0 if x+x j 9j( X I = {I if x=xj and in this case we also have Supp(g.)cV.cV 3
Since
2/
is a locally finite family
which never vanishes in X. Therefore (h.=g 1 i/
3
3
Moreover
.
CU xj ix j
1 g is a map of class p j€J j
c
/icJ) is a partition of unity of class 4j j€J p in X subordinated to V . Since V is an open refinement of ‘U, there is a map a:J+I such that V.cU for all jeJ. For all icI 3 a(j) let r k1. = C h. if a-’(i)t$ and let qi be the constant map zero -1 jea (i) J if a-’(i)=@. Then it is easy to see that (qi/iGI) is a partition of unity of class p in X subordinated to 21.0 Next we shall see that the Urysohn condition of class p, in real Banach spaces, is equivalent to the existence of partitions of unity of class p. Lemma 1.5.6 a ) The function f:W+[O,l] defined by 0
is a function of c l a s s
if tso
W.
b) For all 6>0, the function gs:R+[O,l] defined by
Real Differentiable Manifolds with Corners
53
is a function of class m. Note that g6(t)=0 for a l l ts0, g6(t)=l for a l l tr6 and O0, the function h6:R+[0,1] defined by h6 (t)=ga(-t)
is a function of class m. Note that h6(t)=l for a l l ts-6, h6(t)=0 for all trO and O0, the function k6:R+[0,1] defined by k6 (t)=g6(t+1+6)*g6(-t+1+6)
is a function of class m. Note that k6(t)=0 for a l l t with ltlr1+6, k6(t)=l for a l l t with It151 and O
Proposition 1.5.7
Let E be a real Banach space and peHu(m). Then the following statements are equivalent: a ) E verifies the Urysohn condition of class p. b) E admits partitions of unity of class p. c) The family H=(Supp,(f)/f:E-+[O,l] is a function of class p) includes a u-locally finite basis of the topology of E, where
Supp, ( f)=( xeE/f (x)+0)
.
Proof. a)*b). First, note that E is a paracompact Hausdorff differentiable manifold of class m. Then, by 1.5.5, E admits partitions of unity of class p. c)+a). Let A , B be disjoint closed non-empty sets of E. By the hypothesis, there exists B , basis of the topology of E such that B= Bnl where Bn is a locally finite family whose elements ndN
Chapter 1
54
belong to H for all nd4. For every ndi, let us consider G. Vn=(G~Bn/GcX-A), Un= (I G and Vn= GfUn GEV~
Un=(GeBn/GcX-B),
For every neM and every GEUn' let fz:E-+[O,l] be a function of class p such that Suppo(fz)=G. Since Un is a locally finite n is a function of class p. family, then Fn= 1 fG:E-+[O,+m) GeUn Moreover Suppo (Fn)=Un. Let us consider fn=foTn, where f is the function defined in a) of Lemma 1.5.6. It is clear that fn:E-+[O,l] is a function of class p and Suppo(fn)=Un. If Un=$, then fn:E+[O,l] would be C o . In that case, obviously, Un=9=SuPP0 ( fn) * Analogously, for every n 4 , one takes a function gn:E+[O,l] of class p such that Suppo(g,)=Vn. Since B is a basis of the topology of E we have W=(Un/ndN)u(Vn/ndN) is an open covering of E. Let us consider U*=U 1 1 and Vi=V 1' For every ncP(-(1) we consider and
Now we have:
*
*
*
1) W =(Un/n&i)u(Vn/ndN) is an open covering of E. Indeed, if
*
*
XEE and no=min(n&i/xrUnuVn), it is clear that X E U ~ ~ U V ~ ~ . 2) W* is a locally finite family. Indeed, if xeE and no=nin(ndN/xdJnuVn) , then we have f (x)> O or gn (x)>O. Therefore 0 no 1 1O. or gn (x)>,> Then there is m>n such that f (x)>,>O
0
VX=f-l ((:,1]) "0
if f
"0
(x)>,1 and Vx=g-l(($,l]) "0
neighbourhoods of x such that VxnU:=~
"0
and
if g O (x)>1 are open
* VXnV =# n
"0
for all n>m.
Real Differentiable Manifolds with Corners
*
55
*
For all ncH, Un and Vn are open supports of functions of class p from E to [0,1]. Indeed, for all ic(1,2,...,n-l) it holds 1 -1 -1 -1 1 -1 -1 ((0,111 and gi ([O,E))=gi ((Ol11), where f~l([o,E))=fi rk:[O,+rn)+lR is a function of class m such that OsP(x)sl for xeO, *(x)=O for ~ 21 % and * ( x ) > O for x~[O,l/n). Then, for all -1 1 -1 1 i~(l,2,...,n-l), fi ( [ O I E ) ) and gi ( [ O I E ) ) are open supports of functions of class p from E to [o,l]. Since Un and Vn are also open supports of functions of class p from E to [O,l] and the support of a product of functions is the intersection of the supports of the factors, the statement 3 ) is proved. 3)
*
*
Let us
*
*
consider U=[) Un and V=\) Vn and fU,gV:E+[O,l] ncH ncH functions of class p such that Suppo(fU)=U and Suppo(gv)=V, where fU'gV have been obtained in the same way that fnlgn above. It is clear that AcUcX-B, BcVcX-A and h=f :E+[0,1] is U/fu+gv a function of class p such that h(A)=(l) and h(B)=(O). b)+c) . For every nEH, Un=(Bl,n(x)/x~~) is an open covering of E. Then there is a partition of unity of class p in El 9=(Q:/xcE) , subordinated to 21,. It is clear that gn'(Suppo(?!~)/xcE) is a locally finite open Therefore B = ( ) Bn is a u-locally finite basis refinement of 21,. new of the topology of E.o There are real Banach spaces which do not admit differentiable partitions of unity. But, as it is well known, every real Banach space admits continuous partitions of unity. Example 1.5.8 Let p be an element of IN and let Lp be the set { ( ~ ~ ) ~ ~ a r / x ~ c R 02
VneH and
C
Ixnlp
Chapter 1
56
is a norm in Lp and (tp,1 i) is a separable real Banach space. In this space we have: "If f:tp-H? is a function of class qzp, p odd, and Suppo(f) is bounded, then f=Ott. (See R. Bonic - J. Frampton; Differentiable functions on certain Banach spaces, B.A.M.S., 71, 1965, pp. 393-395). Therefore, by 1.5.7, if p,q are elements of PI such that p is odd and qzp, then Lp does not admit partitions of unity of class
q. In particular t l does not admit differentiable partitions of unity. ProDosition 1.5.9 L e t (H,<,>) be a r e a l Hilbert s p a c e . Then we h a v e t h a t : a ) E v e r y o p e n b a l l o f H i s the o p e n s u p p o r t of a n element of (f:H+[O,l]/f is of class m}, b ) If H is a s e p a r a b l e s p a c e , t h e n H v e r i f i e s t h e U r y s o h n c o n d i t i o n of c l a s s m.
Proof x be an element of H and e>O. We consider the h:H+[O,l] , defined function of class M I is the function defined in by:h(y)=L1,l(G ) where L a) Let
1/2
&
.
e) of 1.5.6. Clearly Suppo (h)=Be (x)
b) Since H is a separable space, the topology of H has a countable basis whose elements are open balls. Then the result follows from a) and 1.5.7.0
If we omit the condition of separability in the statement b) of the preceding proposition, then the Hilbert space (HI<,>) verifies, again, the Urysohn condition of class m , but the proof becomes more complicated. We develop this proof in the sequel. Lemma 1.5.10 Let
(C0(A),I
A
be
lo),
a
set,
A*#.
t
Consider
the
real
Banach
space
w h e r e C O ( A ) = f/f:A-H? i s a map s u c h that t h e set
Real Differentiable Manifolds with Corners (acA/lf(a) I > ):1
is finite for a l l ndN
57
and llfUo=maX.( If(a) I/aEA)
for all feCO (A). Let
So
be the set
{a:Co(A)+[O,l]/a
is a function of class
a
such that for a l l y€CO(A) there exist n EN, an open neighbourhood Vy of y in Co(A),al,
...,an EA and
Y
n
a function of class
Y which verify that for all feVY,a(f)=cp(f(al),
a,
cp:iaY4
1
...,f(anY ) ) .
Then
has a cr-locally finite basis vhose the topology of (CO(A), 1 1), elements are open supports of functions of So. In particular (CO(A),I 1), verifies the Urysohn condition of class a. Proof. First we shall see that every open ball of CO(A) is an open support of a function of So. Of course it suffices to prove that every open ball of CO(A) centered at 0 is an open support of a function of So. Consider &>O and let (p:R+R be a function of class a such that Os(p(t)Sl for all t m , (p(t)=O for all ts-&, p(t)=O for all the and (p(t)=l for all tE[Then the function a:Co(A)4?
:,$I.
n
defined by or(g)= (p(g(a)) is a Ca-function. Moreover acSO and aEA SuPPO(a)=B&(0) * The topology of CO(A) has a cr-locally finite basis whose elements are open balls. Indeed, for all nEH, let Hn be the set {hn:(l,2,
...,n)+A/hn
is an injective map
} and
{[(qlI...,qn),q]~~nxQ/lqll>q>O,..., Iqnl>q>o}.
let K,
or every hnE~,,
consider the map Th :Rn+CO(A) defined by n 0 if atimhn T (XI,.. ‘Xn) (a)= hn xi if a=hn(i)
-
I
It is clear that Th is a linear map. n
be the set we
Chapter 1
58
If
eKn)],
then it follows that:
a) B is a basis of the topology of CO(A). Indeed, let g be (O)/ncLN) we can suppose an element of CO(A) and c>O. Since B>(B l/n g+O and Ilgllo>c. Clearly the set (Ig(a) (/acA)u(O) is a countably compact set. Let q be an element of for
all
I
1
( 0 , ~- { ) Ig(a) j/aeA) ncl such that q>r
re{ Ig(a) I/aeA n ( ~ , c ) , let
(al,...,an)
..
be
the
set
(aEA/ I g(a) IZE)*I#I and let (ql,. ,qn) be an element of 9" such that lqil>q and lqi-g(ai)l
,...,
...,
,...,
..
..
b) 33 is a o-locally finite family:
of course it suffices to prove that for all ncLN and for all ((q,,
- - - IS,) rq)EKn,
finite family.
iTh (q1,* n
- - rqn)+Bq(0)/hnEHn)
is a
locally
and Let g be an element of CO(A), &= -Zinf{lqll-q,...,lqnI'g) 1 F the finite set (aeA/lg(a) I > & ) . Then for all yEg+B&(O) and f o r all aeA-F we have ly(a)1<2&. For all hncHn, all zeTh (q,, ,qn)+Bq(0) and all aeim(hn) we also have that n lz(a))<2&. Therefore it is clear that g+Bc(0) can only intersect elements of (Th (ql,...,qn)+B (0)/hncHn) which verify im(hn)cF. n q Hence g+Bc(0) only intersects a finite number of elements of
.. .
Real Differentiable Manifolds vith Corners
59
Proposition 1.5.11
Let E be a real Banach space and p~Hu(m). Then the folloving statements are equivalent: a ) E admits partitions of unity of class p. b) There are a set A and a map u:E+CO(A) such that: 1) u:E+u(E) is an horneomorfism, 2 ) for every aEA, the function
ua:Em defined by ua(x)=u(x)(a), is a function of class p. Proof a)+b). By 1.5.7, there exists a a-locally finite basis, '3, of the topology of E whose elements are open supports of functions of class p from E into [0,1]. where Sn is a locally finite family for all nEH nEH, and for every Be8 there exists a function fB:E+[O,l] of class p such that Suppo(fB)=B. Moreover we can suppose '3n&m--$ for all m,ncH with m+n. Then
%I=() Bn,
1 Let u:E+CO(B) be the function defined by u(x) (B)= nfB(x) if 1 is a function of class p. BE%?,. Then for every BEB uB= nfB:E4R Therefore 2) of the statement b) is proved.
I) Let x be an element of E and 0 0 . Then there is nOEH such 1 < E and since '3 u... that 1 VBnOis a locally finite family there "0 is an open neighbourhood Vx of x in E such that F=(BEB~V...VB /mvX+#) is a finite set. "0 1 1 For every BEF we denote by GB the set (fB(x)- -,f,(x)+ -)A "0 "0 n[o,l] and by VB the open set fil(GB). It is clear that VB is an open neighbourhood of x in E and that WX=Vxn( (1 VB) is a BE F 2 neighbourhood of x such that Ilu(x)-u(y) 1
"0
Therefore we have proved that the map u:E+CO(B) is continuous.
Chapter 1
60
11) u is an injective map. Indeed, if x,y are elements of E
such that xfy, then there is BEBncB such that ycB and xeB. Hence 1 1 u(x) (B)= ,fB(x)=O and u(y) (B)= flfB(y)>O. 111) If xe&Bn the map u-l:u(E)+E proved.
1 and ye^, then ~u(x)-u(y)~,zr;fB(x). That is, is continuous and 1) of the statement b) is
b)+a). By Lemma 1.5.10,
has a cr-locally finite basis, Then, by the hypothesis, V=u (8) is a o-locally finite basis of the topology of E. If VEV, then there are BEB and fBeSo such that V=u -1 (B) and Suppo(fB)=B. Therefore V=Suppo(fBou). Since fB€Sof from 2) we deduce that fBou:E+[O,l] is a function of class p. Finally, the statement a) follows from 1.5.7.0 CO(A)
IB, whose elements are open supports of functions of S o . -1
Proposition 1.5.12 Let (H,<,>) be a real Hilbert space, (Li/icI) an orthonormal basis of
(HI<,>), L,(I)=
f:I+IR/(i~I/f(i)*O) is a countable set
1 /f(i)12= 1 f(i)*g(i). Then: a ) (l2(1),<,>) is ic I i d a real Hilbert space, b) there is a linear isomorphism from (H,<,>) over (L2(I),<,>) which is an isometry.
and
Proof a) It is a well known result on Hilbert spaces. b) Consider the map (p :H4, (I) defined by (p ( y)= () ieI. Then it is easy to see that (p is a linear isomorphism whose inverse map, q , is given by q(y)= C f(i).Ci. ic1
Finally we have <(p(x),(p(y)>= (See [B,E.V.T., 21-23]) . O
C.V.
,
p.
Real Differentiable Manifolds with Corners
61
ProDosition 1.5.13 Every real Hilbert condition of class m.
space,
(HI<,>), fulfils
the Urysohn
Proof. By 1.5.12, there is a set I such that H and L2(I) are isomorphic and isometric. Let L be an element that u:L2 (I)+Co((L)uI) the map given by
does
not
belong
to
I
and
It can be shown that u fulfils 1) and 2 ) of b) of Proposition 1.5.11 and hence L2(I) and (HI<,>)fulfil the Urysohn condition of class m . 0 Corollary 1.5.14 Let X be a Hausdorff differentiable manifold of class p, whose charts are modelled over real Hilbert spaces. Then X admits partitions of unity of class p if and only if X is a paracompact space. o
We note that a Hausdorff topological manifold, X, admits continuous partitions of unity if and only if X is a paracompact space. To end the paragraph we shall see an application of the differentiable partitions of unity: proof that A)oB) , where A) and B) are the definitions of maps of class p given in the paragraph 1. We have already said that B)+A). But it is also true that if E admits partitions of unity of class p, then A)+B). First, by the hypothesis, for every XEU there are an open neighbourhood Vx of x in E, and a map fx:Vx+F of class p (with the meaning of the ordinary Differential Calculus) such that f, and f coincide on VxnU.
Chapter 1
62
{I Vx. Then it is clear that A is an open X€U set of E and UcA. Therefore A admits partitions of unity of class p. Then there exists a partition of unity of class p in A, ~=(cpX/x&J),subordinated to (Vx/xdJ). Of course g= C cpxfx:A+F is xeu a map of class p (with the meaning of the ordinary Differential Calculus) and g IU=f Let A be the set
-
On the other hand there exists an open set G' of E such that Then to prove B) it suffices to take G=AnG' and ?=g IG'
+ U=EAnG'.
1.6. Tangent Space of a Manifold at a Point
To define the derivative or the linear tangent map of a differentiable map at a point we need to define first the tangent space of a manifold at that point. The tangent space of a manifold at a point is generalization of the tangent plane to a surface at a point.
a
Proriosition 1.6.1 Let X be a differentiable manifold of c l a s s p (peNu(m)) and aeX. We denote by Ca(X) the class ((c,v)/c=(U,p,(E,A)) is a chart of X at a and WE). We consider the binary relation, -, on Ca(X) defined by:
(c,v)- (c'lv8)*D(cp'P-1)
(cp(a) 1 (v)=vl
This binary relation is an equivalence relation on Ca(X). quotient set Ca(X)/- will be denoted by Ta(X).
Proof
The
Real Differentiable Manifolds with Corners
63
Pronosition 1.6.2
Let X be a differentiable manifold of class p and aaX. Then: a a) For every chart c=(U,p,(E,h)) of X at a, the map Oc:E+Ta(X) defined by Oz(v)=-((c,v)) is a bijective map. (The class of equivalence, -((c,v)), will be also denoted by [(c,v)], or [(c,~)].) b) There is a unique structure of topological real linear space on Ta(X) such that for every chart c=(U,p,(E,A)) of X with aaU, OE:E+T a (X) is a linear homeomorphism. This structure is Banachable (see [M.O.P., VIII.1.581). c) If c=(U,p,(E,h)) and c~=(U',p~,(Et,h')) are charts of X at the point a, then
The real Banachable space, Ta(X), will be called tangent space of X at a and the elements of Ta(X) will be called tangent vectors of X at a. Proposition 1.6.3
Let X and X' be differentiable manifolds of class p, f:X+Xt a map of class p and aeX. Then there is a unique continuous linear map, Taf:Ta(X)-+Tf (a)( X t ) such that for every chart c=(U,p, ( E , h ) ) of X at a and every chart c'=(U',p', (E',A')) of X' at f(a) it holds -1 Taf=O:!a)D(p'fp-l) (p(a)) (Oz]
.
The map Taf will be called tangent linear map to f at the point a. Proof. The result follows from (1.1.9).0 Proposition 1.6.4
a) Let X be a differentiable manifold of class p and aaX. Then TalX=l T,(X)
-
b) Let f:X+Xt, g:X'+X" T,(qof)=T,
(,)Waf.
be maps of class p and aeX. Then
Chapter 1
64
c ) Let f:X+Xt be a diffeomorphism of class p and aeX. Then Taf is a linear homeomorphism. d) Let X be a differentiable manifold of class p, U an open set of X, j:U+X the inclusion map and a N . Then Taj:Ta(U)+Ta(X) is a linear homeomorphism.
a
Note that if c=(V,p, ( E , A ) ) is a chart of U (hence c is a l s o chart of X) such that ad3 and vcE, then
~ a ~ ~ ~ ~ ~ l ~ ~ l u f = l ~ ~ l ~ ~ I x . ~ Proposition 1.6.5
Let X,X' be differentiable manifolds of class p and f:X+X' a map of class p. Then f is locally constant if and only if Txf=O for all XEX. Proof. It is clear that if f is locally constant, then Txf=O for all XEX. Let x be an element of X. By the hypothesis, there are a chart ( U , p , ( E , A ) ) of X I and a chart (U',p',(Et,A')) of X', such that XEU, T f=O for all YEU, p ( U ) is a convex set and f(U)cU'. Y Then, by 1.1.11, for all ycU we have that f(y)=f(x).o Definition 1.6.6 (Rank of a Differentiable Map)
Let X and X' be differentiable manifolds of class p and f:X+Xt a map of class p. We consider the map rf:Xa defined by: dim(imTxf) if this dimension is finite (rf)(x)=rxf= +m, on the other case The map rf will be called rank map of f and the integer rxf vill be called rank of f at x.
Lemma 1.6.7
Let E be a real Banach space and (vl,..., vn) a linearly
Real Differentiable Manifolds with Corners
65
independent system of elements of E. Then there is &>O such that for all (x,,.. ,xn)~Bc(vl)x.. .xBc(vn) , the system (x,,.. ,xn) is linearly independent.
.
.
..
Proof. For all k(l, .,n) we Hi=L(vlr.. ,iir.. .,vn ) . Therefore is( 1,.. ,n).
.
.
have that vieHi, where ci=D(virHi)>O for all
and let xi be an element
Let c be the element min
..
.
of BE (vi) for every is( 1,. ,n). Suppose that rlxl+. .+rnxn= O . If Ir I=max(lrll, ...,I rnl)+O we have in IJ i' 0-1 li0+l rn --- rio vi +l-. rvn) Is c q v i -(- y r1v l vio-l io 0 l0 0 iO
..-
...
a contradiction. Then rl=
...=rn=O.o
Proposition 1.6.8 Let X,X' be differentiable manifolds of class p and f:X+X' map of class p. Then the map rf:X* is lower semicontinuous.
a
Proof. Let x be an element of X and SGR such that s
...
... .
By Lemma 1.6.7 there is &>O such that for all xi~B&(vi), i=lr...,n, (x;,...,~;) is a linearly independent system. On the other hand *(y,v)=D((p'f(p-l) (y)(v) is such that vl,.. ,vnEE
.
the map *:(p(U)xE+E' defined by a continuous map and there are D((p'f(p-') ((p(x)) (vi)=vi, i=1,. ,n.
Therefore there is an open neighbourhood V(p(x) of
..
(p(x)
in p ( U )
Chapter 1
66
such that D(cp'f(p-') (y)(vi)eBC(vi) for all ~EV'(~) and ie(1,. ,n). Finally, rzfrn>s for all z ~ p - ~ ( V ~ ( ~ ) ) = V ~ . o
..
all
Tangent Bundle Manifold Let X be a differentiable manifold of class p. We denote by the set 1 Tx(X)=( (x,v)/xeX,v~T~(X) ) and by tx the map xex rX:TX-+X, 7;X (x,v)=x.
TX
For every chart c=(U,p,(E,h)) of X, we consider the map -1
-1
c :tx (U)+ExE defined by (pc((x,v))=(p(x) , [ U z ] (v)) and the triplet dC=(T,~(U),(~~,(EXE,AO~,)). Then dc is a chart of TX. Indeed, pc is injective and pc(til(U))=p(U)xE is an open set of + the quadrant (ExE)+ =E xE. A"P1 A
(p
If c=(U,(p,(E,h)) and c'=(U',p', (E',A')) are charts of X, then the charts dc and dc, of TX, are compatible of class p-1. -1 -1 and Indeed, 'PC(tX (Writ, (U,))=v(UnUt)xE -1(U)ntX -1 (Up))=p' (UnU')xE' are open sets of ( E ~+E ) ~ ~ ~ ~ a n d p,, (tx
+ (E'xE')A/opi respectively. On the other hand, are maps of class p-1, since -1 -1 (y,z)=(Cp'(P
(Pcr(Pc
'Pc'P,r
-1
-1
(pcr(pc
and
-1 pep,,
(Y), D ( C p P ) (Y)(2)1 and
( Y ~ l z ~ ) = ( w ~ - l (,D(cpcp' Y~l
-1
1 (Y')
Therefore, if (ci/ieI) is an atlas of (d /ieI) is an atlas of class p-1 of TX. In i' unique structure of differentiable manifold (topological manifold, if p=l) such that for dc is a chart of this structure.
(2,)
1.
class p of X, then this way there is a of class p-1 on TX every chart c of X,
This manifold will be called tangent bundle manifold of X. Proposition 1.6.9 Let X be a differentiable manifold of c l a s s p. Consider the tangent bundle manifold TX of X. Then: 1) rX:TX+X is a map of class p-1, 2) if p22, for all (x,v)€TX we have
Real Differentiable Manifolds with Corners
67
-1 (a kX) Therefore and ind ( (x,v)) =ind (x). ak (TX)=rX -1 Bk(TX)=tX (Bk(X)) for all keHu(O), 3 ) for all xaX, the topology of the Banachable space Tx(X) and the topology of Tx(X), induced by the manifold TX, coincide.0 Proposition 1.6.10 Let f:X+X' defined by
be a map of class p. Then the map Tf:TX+TX' Tf(x,v)=(f(x) ,Txf(v) 1
is a map of class p-1.
Proof. Consider (x,v)eTX, a chart c=(U,p, (E,h)) of X at x and a chart c'=(U',p',(E',h')) of X' at f(x) such that f(U)cU'. Then dc=(-cX -1(U),pc, (ExE,hop ) ) is a chart of TX with (x,v)a~i~(U) , and 1 -1 dc,=(-cx, (U') ,pC,,(E'xE',h'op;) ) is a chart of TX' such that Tf (ti1(U)) c t -1 X , (U')
.
Since pc,oTfop~l:p(U)xE+p* (U')xE' sends the element (Y,Z)E(P(U)XE to the e1ement (y)(z))ap'(U')xE', it follows that Tf is a (p'fp-'(y) ,D(p'fp-') map of class p-1.0 Proposition 1.6.11 Let X,X',X" be differentiable manifolds of class p, f:X+X' a map of class p and g:X'+Xff a map of class p. Then we have: a) T(lX)=lTX , b) T(gof)=T(g)oT(f), c) If f is a diffeomorphism of class p, then T(f) is a diffeomorphism of class p-1.0 Next we shall describe the tangent space of a manifold at a point x, using tangent vectors to the curves passing through this point x. Let X be a differentiable manifold of class p and xaX. A curve of class r on X with origin x, O=r=p, is a map u:J+X of class r, where J=[O,a) or J=(b,O] or J=(c,d) with Oe(c,d), such that u(O)=x.
Chapter 1
68
If a is a curve of class r on X (lsrsp) with origin x defined on J=[O,a), then the element of Tx(X) defined by T aOo (1), where co=( [ 0,a) ,i , (R, lR) ) , will be called tangent O
co
vector to a at the point 0 and will be denoted by A ( 0 ) . If f3 is a curve of class r on X (lsrsp) with origin x defined on J=(b,O], then the element of Tx(X) defined by T BOO (1), where co=( (b,01 ,i, (R, -lR)) , will be called tangent O
co
vector to f3 at the point 0 and it will be denoted by b ( 0 ) . If J=(c,d), b(0) is defined analogously.
x,
We note that i f c=(U,p,(E,h)) is a chart of X at the point 0 a(t)-pa(0) then i(0)=ToaoOc0 (l)=OgD(pa) (0)(1)=Og t+O lim, t
pa) X ,(O) where
0'
cO
(respectively,
B ( o ) = o ~ lim-pf3(t)-v'(o)) t
t+o (respectively,
:R+TO([O,a))
U z : E+TxX are the natural isomorphisms.
Oo
cO
,
see 1.1.8,
:R-+TO((b,O]))
and
If v is a tangent vector of X at x given by a curve a:[O,a)+X (or a:(b,O]+X) of class 1 on X with origin x i.e. i(O)=v, then we shall say that v is an inner (respectively outer) tangent vector at x. The set of the inner tangent vectors at x will be denoted by (TxX)i and the set of the outer tangent vectors at x w i l l be denoted by (T,X)O.
Proposition 1.6.12 Let X be a differentiable manifold of class p and XEX. Then Y e have: (TxX)i=- (TxX)O f TxX=L( (TxX)i) and TxX=L( (TxX)O )
.
Proof. Let c=(U,p,(E,h)) be a chart of X at x such that p(x)=O and p ( U ) is a convex set. Suppose that h=(A1,...,An). By Proposition elements vl,.. 0 1vn E=EAeT L(vl,...,vn).
.
1.1.4,
of
there are linearly independent E such that Ai(v.)=aij and 3
Real Differentiable Manifolds with Corners
...
69
.
0 Then we have that TxX=Og ( Ea) eTL( 0 : (v,) , ,OcX (v,) ) On the other hand for all k ( 1 , n) there is aicR such that ai>O and aiviq(U). Therefore the map ai: [ O,ai)+X defined by ai(t)=p-'(tvi) is a curve of class p on X with origin x such that X ii(0)=Oc(vi). Then we have that Og(vi) is an inner tangent vector at x for all ie(1, n). Analogously we have that Og(v) is an 0 inner tangent vector at x for all xeEA. Then TxX=L((TXX)') and i analogously ( T ~ x )=- ( T ~ xO). 0
...,
...,
In fact we have proved the following result: "Let X be a differentiable manifold of class p, x an element of X and c=(U,cp, (E,h)) a chart of X such that XEU and p(x)=O. Then it happens that 0: (Ei)c (TxX) In the next proposition we shall see that the equality is also true. Proposition 1.6.13 Under
the
hypothesis
of
the
preceding
result,
.
Proof. Let w be an element of (TxX)i Then there exists a map of class 1, a:[O,a)+U, such that a(O)=x and &(O)=w. Therefore + p a : [ O,a)+p(U)cEA is a map of class 1 and w=&(O)=O:((pa) ' (0) is an x ( EA) + element of Oc
.
Lemma 1.6.14 Let g:U+V be a diffeomorphism of class pz1 such that OEU and + g(O)=O, where U (resp. V) is an open set of a quadrant EA (resp.
+
FM). Suppose that a:[O,a)+U is a map of class 1 such that a ( O ) = O + Then we have that (ga) (0)eaFM. + and a' (0)&EA.
Proof. From 1.2.12
it follows that g(intAU)=intMV and g(aAU)=aMV.
Let #3: [ O,b)+U be a map of class 1 such that #3 ( O ) = O ,
im@caAU and
Chapter 1
70
p'(O)=a'(o). Then it follows that (ga)'(O)=(gp)'(O). + hand (qp) ' ( 0 ) & F M and the result is proved.=
On the other
ProDosition 1.6.15
Let X be a differentiable manifold of class p, x an element of X and c=(U,p, (E,h)), c'=(U',(p', (E',A')) charts of X such that xeUnU' and cp (x)=cp' (x)=o. Then ve have that 0X (intEh)=Oc, + x (intEi,)c(TxX)i. + The elements of Oz(intEi) vill be C called strictly inner tangent vectors at x and the elements of + vill be called strictly outer tangent vectors at x. -Ox(intEA) C
+ Proof. Let v be an element of intEh. Then there exists a map of X (v) Therefore class 1, a: [ 0 ,a)+UnU' such that a (0)=x and i ( 0 ) =OC : [O,a)+cp' (UnU') is a map of p a ( O ) = O , (9a)' (O)=v and p=p,p-'(pa)
.
class 1 with @(O)=O
and @ ' ( o ) = [ O z , ]
-1
Oz(v). X
Then, from Lemma
+
X
1.6.14, it follows that 6' (O)EintEi+ and Oc(v)~Oc,(intEi,) .o
Note that if X is a differentiable manifold of class p and x is an element of intX, then Tx(X) =(TxX) i=(TxX) and all vectors of Tx(X) are both strictly inner and strictly outer. Proposition 1.6.16
Let X be a differentiable manifold of class p, x an element of X , c=(U,p,(E,h)) a chart of X vith XEU and veTx(X). Then: I) The folloving statements are equivalent: a) v is an
element of (TxX)i, b) if XEA and Xp(x)=O, then A(O:]
-1
(v)rO.
IT) The folloving statements are equivalent: 1) v is a
strictly inner tangent vector at x, 2) if hch vith hp(x)=O,
Proof. I) a)+b).
Since VE(T~X)~, there is a curve a:[O,a)+U
class 1 with origin at x such that &(O)=v.
of
On the other hand
Real Differentiable Manifolds vith Corners -1
] : O [
71
0
(v)=(pu)' (O)=lim+ pa(t)-pu(o) t and p(x)eEA. tao
Therefore A
.
($1
-1
(v)20.
.
b)+a) Let Al be the set (AeA/Ap (0)=0) If A1=#, then xeintX Then, by the hypothesis, and v€(T.*X) i Suppose that Al+#. -1 2L + (V)E (1 EA and Ap(x)>O for all A E A - A ~ . Therefore there w=(Og) A€&
.
I
exists a>O such that p(x)+awE
(1
A€A-Al
Ei.
If we take a8O
small enough, then we have the map u : [O,a')ap(U) defined by a(t)=p(x)+tw. Finally, B=p-la:[O,a')+U is a map of class 1, X i B(O)=x, b(O)=O Cw=v and hence ve(TxX)
.
11) 1)+2). Consider A1=(A~A/Ap(x)=O). Then there is an open 0 set A of X such that xcAcU and p(A)nEA=# for all AcA-hl. It is clear that c'=(A,tx ] : O [
and -1
(E,hl)) is a chart of X at P(X)o(pIA' + Therefore [O:,)-'(v)~intE~ and
t -p(x)op(x)=O. -1
(v)=(O:,]
1
( v ) , that is, A
(v)>O for all AEA1.
2)+1). Consider A1=(A~A/Ap(x)=O). If Al=#,
then v is a strictly
inner tangent vector at x. Suppose that A1+#. that w=(O:)
-1
(v)e()
AEhl
that p(x)+awe
(1
Then, from the hypothesis, it follows
+ intEA.
On the other hand there is a>O such
+
intEA. If we take a'
AEA-A~
then there is a map a(t)=p-l(p(x)+tw) and
of it
class holds
a'>O
small enough,
1, u : [O,at)aU, defined by cf(O)=x, i(0)=Oc(w)=v X and
a((0,a'))cintX. If we take the chart c'=(A,t- p(x) "(PIA,(E,hl))I + then v=&(O)=O:,(w) and, since wcintEA , the vector v is a strictly inner tangent vector at x.o
1
Chapter 1
72 Proposition 1.6.17 Let
i
f:X+X'
be
Txf((TxX) )c(Tf(x)x')
i
-
a
map
of
class
p
and
Then
xcX.
Proof. We consider charts c=(U,p,(E,A)) and c'=(U',p',(E',A')) of X and X' respectively, such that x~U,p(x)=O, f(U)cU' and pp'f (x)=0. Then , ) c O E $ x ) (EiT)=(Tf(XIXI) i Txf ( (TxX)i)=Oz$x)D(p'fp-l) (0)(0:) -1 (oc(E;) x . . (1.2.10) .o
If X is a differentiable manifold of class p, we denote by (TX)i the subset (TxX)i of TX, and we denote by (TX),i the xcx subset C (TxX)i of TX, where (TxX)i is the set of strictly inner xcx tangent vectors at x. Analogous notation will be used for the outer tangent vectors and for the strictly outer tangent vectors. Example be the differentiable manifold of class 0 0 , Let X W+w( 0).Consider the chart of X I co=(X,i f(R, 1)) and the chart of ) , where i :TX-mxR is the map defined by cO
It
is
clear
Cm-diffeomorphic
that to
i
the
cO
(TX)=XxR
manifold
-1 ] ((R+xR)w((O)x(R+w(O)))) cO
(TX)i=[i
(TX):=[ico)
-1
((R+xR)u(
and
therefore
XxR= (RxR)
+
P1
.
TX
is
Moreover and
(0)xR+ ) .
Finally (TX)i is not a locally closed set in TX, that is, it is not an intersection of an open set and a closed set.