This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
=, we get (24)
f (a I ) d7 In =
x
J x
QA *,3.
2.3 Sobolev Spaces of Forms Let X be an n-dimensional compact submanifold of u : X ---f R which are 7-ln L X-measurable and with
J Iulsd7-InLX
s>1
I[Yn+N. The class of functions
2.3 Sobolev Spaces of Forms
537
is denoted by LS (X) . LS (X) is clearly a Banach space, and for s = 2, L2(X) is an Hilbert space with the obvious scalar product. If (Vi, (pi), Vi : V -> X is
a system of charts for X, we define the class Wl,'(X), s > 1 as the class of u E LS(X) such that u o Wi E Wl,s(V), and we set II u ll wl,s (X) := 11 U 111s(X) + E f I D(u O Wi) IS dx.
(1)
Vi
It is readily seen that the definition of Wl,s(X) is independent of the system of charts, while the norm does depend. However different system of charts give rise to equivalent norms on W""s(X). An intrinsic norm on Wl,s(X) can be defined in terms of approximate tangential derivatives DX to X : one easily verifies that IIUIILS(X)+1
1
1/S
f IDXulsdHnLX)
is a norm equivalent to (1). Properties of the standard Sobolev spaces on open sets of Euclidean spaces are immediately transferred to analogous properties of W1"2(X), we in particular state (i) W"2(X) is a Hilbert space with the scalar product /
(u I V)Wi,2(X) := (u I v)L2(X) +
f(DXU I DXV)TX dfn LX.
LS(X) is compact (ii) The inclusion Wl,s(X) (iii) C°°(X) is dense in W""s(X).
A differential p-form w on X is said to be measurable if <w(y),v(y)> is 7-ln-measurable for any smooth vector field v, equivalently if for any local chart cp2 : Vi -> X, cp#w is L'-measurable, i.e., all coefficients of cp#w are measurable.
Definition 1. Let s > 1. The space of measurable p -forms on X such that
f IwlsdfnLXAPT,,X = <7r#w,7r#77>APRn+N
a scalar product on L22 (X) is given by (w,71)L;
f <w(y), 77(y) >APT,,X /Ln =
V w,7) E A°TyX,
/
<7r#W, rr#17 >npRn+N dH-.
538
5. Cartesian Currents in Riemannian Manifolds
Definition 2. The class of p -forms w in L4(X) such that cp#w have coefficients in W1'1 (V,1[2) for every coordinate chart (V,cp) are denoted by WI,"(X). (X) can be defined by choosing a coordinate system of charts A norm on (Vi, fpi) and setting II W II W;'s(X)
II W II LP(X)
+
#wI WP's(Vi).
Though the definition of WP' (X) is independent of the coordinate system, the norm does depend on it, but one easily verifies that different coordinate systems produce equivalent norms. Again properties of WP"°s(X) are inferred from properties of standard Sobolev spaces; we state
(i) W l ,' (X) is a Hilbert space.
L'(X) is compact. (ii) The immersion WP ,3 (X) (iii) Smooth forms are dense inWP(X). Finally we remark that similarly one can introduce the spaces Wp'3(X) and more generally WW ,r (X ), but we shall not dwell any further on this point.
2.4 Harmonic Forms Let X be a smooth compact n-dimensional submanifold of 1[in+k possibly with boundary. We assume that X is a CO° submanifold even if the claims on this subsection hold for Cl,l submanifolds with trivial changes. Let d be the differential operator on forms.
d : C°°(X,APTX) -+ C°°(X,AP+ITX). Definition 1. The codifferential operator
b : C' (X, APTX) ---> C' (X, AP-'TX) is the (formal) adjoint of d, i.e., is the operator defined for all a E CO° (X, APTX)
and p E C,- (X \ 9X, AP-'TX) by (Sa I )3)L2(X,AP-1TX) :=-(a I d,3)L2(X,AP-1TX).
Clearly b2 = 0, and S extends to all
WP,' forms.
Proposition 1. 8 : C°°(X,APTX) -+ C°°(X,AP-1TX) is a differential operator. If x1, ... , x'n are local coordinates and G = (gij) the metric tensor of X, then for w = (1)
aI =P wa
(6w). =
dxa we have Sw = E ,,I=P(Sw). dx-I where
E E E g7ao,(i,
Ial=p-1 i¢a 1QI=p
(a+2)wR)-
2.4 Harmonic Forms
539
Proof. Let 7) E C°° (X \ 8X, AP-'TX) be a (p - 1)-form with support in coordinate chart. We have
-(w,dr1)APTTx = - = gQ7wa(di1)7 IQI=I7I=P
( IQI=I'YI=P
1
\ a+i=ry
I$I=P Ial=P-1 i0«
Integration by parts on X then yields
-(4' dr7)L2 = - f (w, drl)f dx
-
a)Di(f9p(«+i)«)
0(2,
Igdx,
Ii3I=p i s
«1=p=1
Ial=pY_iOa o-(i, a)Dii.e. (1).
consequently g'Y«(Sw)ry =
Proposition 2. If X is oriented and * is the Hodge operator on X then S = (_1)n(p+1) * d * Proof. For a E COD (X, APTX) and /3 E C°° (X \ 8X, AP-1TX) we have
d/3A * a = d(/3A * a) = d(/3A * a)
-
(-1)P-1,3Ad
*a
(-1)p-1(-1)[n-(n-P+1)](n-p+l)/3n * *d * a
= d(/3A * a) - (-1)n(p+1)OA * *d * a. Therefore, by Stokes theorem and (24) in Sec. 5.2.2, we get
(Sa1/3) _ -(aId3)=-J d/3A*a (-1)n(p+l) f /3A * *d * a =
(-1)n.(p+l)(*d
* a 1)3).
Definition 2. The Laplace-Beltrami operator on Coo (X, APTX) is the operator
a=dS+Sd : COO(X,APTX) a form w is called harmonic if Zw = 0. The following claims follow at once
C°°(X,APTX);
540
5. Cartesian Currents in Riemannian Manifolds
Proposition 3. We have (i) A is formally self-adjoint, i.e., (La 1,3) _ (a Ai3) for a, 3 E Cc-(X \ 3X, APTX). (ii) If w E CC°(X \8X,A2TX) then Aw = 0 ifdw = bw = 0, in fact (w dw) _ J8wJ2
+ dwI2 (iii) A is well defined onW2'2(X,APTX). Definition 3. The Dirichlet integral associated to a p-form is D(w) := f(dwI2 + Ibw12) d7-ln
x 1 The Laplace-Beltrami operator on functions. First we compute b on 1-forms. In a chart U, if cp E C° ° (U) and w = > wi dxi we have
f f -
(6w I 0) =
(w, dcc) dy = -
f
-gUwi 28x
U
Dj(V,-gg2'wi)Wf dx
U
where (gij) is the metric tensor on X, gig Sw =
Dj
(gij)-1, and g := det (gij). Hence (-,fg-g2'wi)
and for any function f E C°°(X) A f = (bd + d8) f = bdf =
D; Voij -Lf
axi
Also in local coordinates, i.e., for f E C' (X) with compact support in a coordinate chart we have
D(f) = f Idf12dW2 = f X
8f of dx. /g2j 8xi axi
Notice that, by the maximum principle, any harmonic function on a compact and boundaryless manifold must be constant.
0 The Laplace-Beltrami operator on p-forms in IRn. Given w = E wadxa 1c1=p
we have dw
and, from (1)
=
E wa>z o-(2, a) dxa+i lal=p i a
2.4 Harmonic Forms
Sw = (-1)n(P+1) * d * w =
(2)
Q(i )w(j+i i
1
I01=p-1
541
J dxO.
i j3
We now compute the Laplace-Beltrami operator. Of course
d5w = E(dSw)a dxa, a
and, according to the above
(dbw)a =
/
o(i, ) (dw)a,i =
T Q(i, a)o,(j, Q)wa+j,ji
= E E o(i, a - i)o(j, a - i)wa-i+j,ij ; iE« jva-i now we split the sum in j as
>
joa-i
_ j=i
j$a
to get
o(i, a - i)o(i, a - i)wa,ii
(dbw)a = iEa
o(i, a - i)a(j, a - i)wa+j-i,ij.
+
(3)
iEa
j0a Also 8dw = E(Sdw)a dxa, and (Sdw)a =
o(j, a)(dw)a+j>j =
E a(j, a)o(i, )3)w0,ij j 0a (j+i=a+j
i 0a joaiEa+j
Similarly we split the sum in i as the sum of terms in which i E a and terms in which i = j to get (Sdw)« _
(4)
o(j, a)o(j, a)wa,jj j4za
o(j, a)o(i, a+ j - i)wa+j-i,ij
+ iEa
Summing (3) and (4) we then get (zlw)a = (dbw + Sdw)a = Lwa
+
[o (i, a - i)o(j, a - i) + a(j, a)o(i, a + j - i)]wa+j-i,ij iEa
ja
542
5. Cartesian Currents in Riemannian Manifolds
As fori<j (5)
u(i, a - i) = Q(i, a - i + j)
o-(j, a, - i) = -o-(j, a),
while for i > j (6)
o(i,a-i)+j)
o(j, a-i)=o(j,a),
we see that the term in square brackets in (5) vanish, therefore we conclude
L1w = E (Qwa) dxa
(7)
lal=p
i.e., a p-form in the flat manifold R n is harmonic if and only if its coefficients are harmonic functions. Obviously 1dwj2
+
lawl2
=
IDwal2 + L(w) lal=p
L(w) being a non-zero quadratic form of the derivatives of wa. Since the Euler equations of the two integrals Do(w, U) :=
fI&2 + 82) dx and
Dwa U
2
,
lal=p
as we have seen agree with Awfa = 0, we infer that (8)
Do(w, U) =
JU
IDwa12 dx
V WE W01,2 (U)
Moreover L(w) is a null Lagrangian; in particular fu L(w) dx depends only on
the values of w (wa) at the boundary 8U of U and vanishes if w E C' (U), compare e.g. Giaquinta and Hildebrandt [275]. Of course one can compute bw, L,w, I dw 12 + I dw 2 explicitly in terms of the metric tensor and of its derivative, but, since we do not need it in full generality, we postpone this topic to Sec. 5.2.7 and here we confine ourselves to the following remark. From (1) one sees that (dw).y has the form (9)
(8w).y(x) = api(G)wQ,i + bp(G, DG)wo
where aQi(G) and bp(G, DG) are smooth functions which pointwise depend re-
spectively on the metric tensor and on the metric tensor and its derivatives. Indeed api(G(x)) are Lipschitz functions and bp(G(x), DG(x)) are L°°-functions on X if the metric tensor G is Lipschitz. Moreover we notice that, in case G is the Euclidean metric,
2.5 Hodge and Hodge-Kodaira-Morrey Theorems
543
(8w)7 = E E o (i,,Q)DjwQ+i IQI=p-1 i0Q
consequently b Q (Id , 0) = 0. On account of the previous considerations it is then easy seen that in local coordinates the integrand of the Dirichlet integral has the form (10)
(IdwI2+ 5wI2)/dx = A«Qw«,iwQ,.7 +Bapiwa,iw,+C«QWaWQ
A" (G(x)), BaRi(x) = BaQi(G(x), DG(x)), CCQ(x) _ Cap(G(x),DG(x)) and
where
Aaa(Id)
BaQi(Id, 0) _ C«Q(Id, 0) = 0.
2.5 Representing Cohomology Classes by Harmonic Forms: Hodge and Hodge-Kodaira-Morrey Theorems We are now ready to prove the celebrated theorem of Hodge which states that every cohomology class of a compact smooth manifold without boundary contains exactly one harmonic form. More generally we shall prove a decomposition theorem for forms, Hodge-Kodaira-Morrey theorem, from which Hodge result follows. The proof, based on variational methods and more precisely on Dirichlet principle, shows the entire theory of linear elliptic partial differential equations at work.
In this subsection X is a compact, possibly non orientable, boundaryless Riemannian manifold of dimension n that is thought without loss in generality as isometrically embedded in some RN. We assume that X is a CO° manifold even if one can work on C""l manifolds with trivial changes. One of the key ingredients of the variational proof we are going to present is the following simple but important lemma
Lemma 1 (Gaffney). There is a constant C such that for each point y E X there exist a ball B(0, p) C ]R' and a chart (U,,, cp), cp : B(0, p) -+ Uy C X, cp(0) = y, such that
D(w) ?
2
f
I2
Diwa2 dx - C11 w L2(U,)
B(O,p) 2'a
whenever cp#w =
wa dxa has support in U,,.
Proof. Fix a point y E X and choose coordinates so that the metric tensor at y is the identity, gij(0) = bij. Then using the representation (10) in Sec. 5.2.4 for D(w) we infer D(w)
>
f A'Q(0)w«,iwQ,j dx - J[A(0) - f IBaQi(x)I w«,iIIwQldx - J
ICaQIIWal
dx IwQldx.
5. Cartesian Currents in Riemannian Manifolds
544
As
B(0, p)) = L(O,P)
if A,a(0)wa,tiwg,j dx =
IDwl2 dx, i,a
by (8) in Sec. 5.2.4 the result follows at once choosing p sufficiently small so that A'3 (0)1 be small. El A%3Q
Choosing a finite cover of X by charts for which Lemma 1 holds we readily infer that (i) D(w) +cll w lli2(x) is an equivalent norm on W1"2(X, APTX) for a suitable
constant c depending on the covering, and in fact there is a positive constant cl and a non- negative constant c2 such that (1)
2
2
D(w) >- cl 11 w IIw 1,2 - 0211 w IIL2
and, being ll w II L2(x) continuous with respect to the weak convergence in W1,2 by Rellich theorem, (ii)
D(w) is lower semicontinuous with respect to the weak convergence in W1"2 (X, APTX).
Definition 1. We say that a p-form w E W1,2 (X, JIPTX) is harmonic if dw = bw = 0. The linear space of harmonic p forms is denoted by HP.
Theorem 1. HP is finite dimensional. Proof. Otherwise we could find a sequence {wk} C HP which is orthonormal in L2. As ll wk IIL2 = 1 and D(wk) = 0 a subsequence will converge strongly in L2, which is impossible since wk is orthonormal in L2.
Theorem 2. There exists a positive constant c such that (2)
D(w) > cl w I12 2(x,APTx)
for any w E W1,2 (X, APTX) which is orthogonal to HP. Proof. Otherwise we find a sequence Wk E W1,2 (X, APTX) n HPl such that 1) (LOO < kll wk liL2.
If 77k := wk/I1 wk IIL2, we have 11 77k IIL2 = 1, D(77k) < 1/k. Passing to a subse71 weakly in W1'2, Ink - rl strongly in L2, hence 1171 IIL2 = 1, by quence 77k
semicontinuity 71 E HP' and D(77) = 0, i.e., 77 E HP fl HP' or in other words 71 = 0: a contradiction.
2.5 Hodge and Hodge-Kodaira-Morrey Theorems
545
Given a p-form f E L2 (X, APTX) we are now interested in finding a weak solution of the equation
aw = f, that is a p-form w E W1,2 (X, APTX) such that (3)
(
drl)L2 + (8w l'rl)L2 = (f 7)LZ
V 77E W1,2(X, APTX).
We have
Theorem 3. Equation (3) is solvable if and only if f belongs to the orthogonal in L2 (X, APTX) of HP, denoted by HP. Moreover there exists a unique solution wo in W1"2(X,APTX) n HP1 and the linear transformation from f to wo is a bounded linear transformation from L2 (X, APTX) into W I,2 (X, APTX). Proof. Clearly if w satisfies (3) then f is orthogonal to HP. Conversely assume
f E L2 (X, APTX) n HP1 and consider the integral
F(w) := D(w) - 2(f I U;) L2 defined on W1"2(X,APTX) n HP1. Trivially F is lower semicontinuous with W1,2, W1,2 respect to the weak convergence in n HP1 is weakly closed and, by Theorem 2,
F(W)>CIIwIIL2-21If
IL2IIWIIL2? 2IIWIIL2
- III IL2.
Therefore, taking also into account (1) we immediately infer that there exists a minimizer wo E W1"2 (X, APTX) n HP. From the minimality of wo we now infer (4)
0 = dtF(wo + t() _ (dwo I d() + (Swo 15()
- (f I
for all C E Wl°2(X, APTX) n HP' and hence for any E W1"2 (X, APTX) since any such C can be written uniquely in the form ( = h+(o where Co is orthogonal to HP and h E HP. From (4) we also get D(wo) < 11f IIL21IwoIIL2
which together with (1) and (2) proves the boundedness of the map f -+ wo. Finally, if w1 is another solution of (3) we have for (:= wo - wl I<12 + Ij(I2 = 0 i.e. wo - wl E HP, and if w1 also belongs to HP1, wo - wl = 0.
Equation (3) is the Euler equation of the integral D(w) - 2(f I w)L2
therefore on account of the representation (10) in Sec. 5.2.4, we see that in local
coordinates the coefficients of any solution of (3) are in fact solutions of an elliptic system. The regularity theory for elliptic systems then applies, and in particular we can infer
546
5. Cartesian Currents in Riemannian Manifolds
Theorem 4. Let X be of class C1'1 and let wo solves (3). We have (i) If f E L2 (X, APTX), then wo E W2'2(X, APTX).
If f E LP(X,APTX), then wo E W2'P(X,APTX), p > 1. If f E C°'N'(X, APTX), then wo E C2,4 (X, APTX).
Higher regularity follows, if we assume higher regularity of X and f. For instance, if X and f are of class C°° or analytic, then wo is C°° or analytic. (ii) In particular we see that w is weakly harmonic ifw is smooth and harmonic.
For each p set now
Imd = {da Im5 = {8a H = {a E
a E W1'2(X, AP-'TX) } a E Wi'2(X, AP+iTX)} Wi'2(X,AP-iTX) da = 8a = 0}. I
Then we have
Theorem 5 (Hodge-Kodaira-Morrey). The linear subspaces Im d, Im 6, H of L2 (X, APTX) are mutually orthogonal and closed. Moreover
L2 (X, APTX) = H ®Im 8 ®Imd, i.e., any w E L2(X, APTX) decomposes uniquely as
w=h+8a+d,Q
(5)
with a E Wi'2(X, AP+iTX), /3 E W12 (X, AP-'TX) and h E H, and (6)
IIhIIL2+ aIIw1.2+II/0IIw1,2
and in fact, if X is of class CO°, h E C°°, and for all k (7)
II a IIW1,k + IIQ IIW1,k
Proof. Let w E L2(X, APTX). Denote by h its projection on H which is finite dimensional. Then w - h E H -L and by Theorem 3 and Theorem 4 there is wo E W1"2 (X, APTX) such that
. wo=w-h. Writing a := dwo, /3 = 8w°, (5) and (6) follow at once. Let us prove now that the three subspaces Im 5, Imd and H are mutually orthogonal and closed, from which one easily infer that the decomposition (5) is unique. Let w E H, for a E Wi'2(X,AP-iTX) we have
(dalw)=-(a,bw)=0 and for 0 E W1'2 (X, AP+iTX)
2.5 Hodge and Hodge-Kodaira-Morrey Theorems
547
(5aIw)=-(aIdw)=0; while we also have
(da 15/3) = -(d2a /3) = 0 provided a belongs to W2'2. If a belongs to W1"2 we choose a sequence ak E W1,2
such that ak -> a strongly in W1,2 and we get (da 15p)
lim (dak I SO) = 0.
k
00
Finally let us prove that Imb is closed. Let wk := b77k, 17k E W1,2 (X, AP+'TX), and assume that wk -> w in L2. By (5) we have wk = hk + Sak + dQk,
the uniqueness of the decomposition yields Sak = Wk, moreover 11 ak I I W 1.2 <- C11 wk I
I L2 < cont.
Therefore we infer passing to a subsequence that
ak -k a weakly in W"2,
ak -> a strongly in L2,
and a E W"2. Consequently ba = w, which proves that Im S is closed. Similarly one proves that Im d is closed.
Before stating Hodge theorem, let us recall the definition of the De Rham cohomology groups . A smooth p-form a is called closed if da = 0, exact if there
exists 77 with d77 = a. Because of d o d = 0, exact forms are always closed. Two closed forms a,,3 are called cohomologous if a - /j is exact. This property determines an equivalence relation on the space of closed p-forms and the set of equivalence classes is a vector space over R, called the p-th de Rham cohomology group and denoted by HHR(X)
Theorem 6 (Hodge). In every cohomology class in HHR(X) there exists a unique harmonic p-form. Proof. Let wo be a closed differential form representing a cohomology class. All forms homologous to wo have the form
w=wo+d3; as wo E Im b1, we can write by the decomposition theorem wo = ho + d/30
where ho is harmonic, therefore every w cohomologous to wo has the form
w=ho+d(,3+,do).
548
5. Cartesian Currents in Riemannian Manifolds
It is worth noticing again that if w = h + d/3 is the decomposition of a closed form w, h being harmonic, we also have, assuming X of class C°° II 0Ilwk+l,r < Ck11 W Ilwk.r
J1 Q I1Ck+1,µ < CkI W 11C-,µ.
Later we shall be interested in closed forms in the product manifold f2 x X, where Q is an open set of an oriented Riemannian manifold. The product structure in ,(2 x X induces a canonical splitting of the exterior differential operator
din,(2xXas
d=d,+dy
with respect to local coordinates (x, y), x E ,(2 and y E X and the splitting of DP (f2 x X) as direct sum min(p,n)
Dp(,(2xX)_
®
Dp,k(,RxX)
k=max(O,p-n)
where Dp,k (,R x X) denotes the class of p-forms in the product ( x X with exactly k-differentials in y. A simple consequence of the decomposition theorem is the following result which will be used later
Proposition 1. Let w be a compactly supported p-form in 0 x X which has exactly f-differentials in y, w E Dp,e(.(2 x X), and is dy-closed, i.e., dyw = 0. Then w can written as s
w = 1 cps(x)no" (y) + dyrl (x, y) s=1
where i is a compactly supported (p - 1)-form in .(2 x X with at most £ - 1
[as], Qi harmonic, form a basis of
differentials in y, cps E VP-e, and the cohomology group HHR(X).
Proof. Choose an open covering {Ui} of U so that on each Ui there exists a
basis of simple tangent (n - P)-vector fields eay,i. Denote by i the dual form of ea,i and by {iii} a decomposition of unity associated to the covering {Ui}. On each Ui i jw can be seen as ,.Ijiw
77anwa
where w E CO° (Ui, V (X) ). Thinking of x as a parameter, Theorem 5 applied to yields s
(x) _
cpa's (x)os + dy?7a s=1
with cpa's E C:°(UU) and 77i E C°O(Uj,DP-1(X)). The result then follows summing on i and a.
2.6 Relative Cohomology: Hodge-Morrey Decomposition
549
2.6 Relative Cohomology: Hodge-Morrey Decomposition Let X be a smooth compact, possibly non orientable, Riemannian manifold of dimension n with smooth boundary. We think that X is isometrically embedded in an Euclidean space R' +N and we also assume that X is a C°° manifold with a C°° boundary aX even if our presentation extends to C1,1 manifolds with Lipschitz boundary with only trivial changes. For x E aX, let v(x) be the inward normal vector to aX at X, i.e., the unit vector v(x) E TTX orthogonal to TxX at x and pointing inside X. We infer the existence of special charts for X near each point x0 E 9X. For p > 0 set B+(0, p) := {x = ( x , , . .. , x.) I xI < p, x,, > o} F (O, P) :_ {x E R'-' (x', 0) E B(0, p), x' :_ (x1, ... I
I
Each point x0 E aX has then a chart cp : B+(0, po) --f X such that cp(0) = xo, : T(0, po) --> aX is a local chart for aX, and, for x' E F(0, po), Dgo(x', 0)e,,, = v(co(x', 0)) is the inward unit normal vector field to aX. We refer to these charts as to admissible charts. .
Tangential and normal traces of a form. Let us define the tangential and normal parts to an of a form. Denote by t the restriction map
r : C°° (X, APTX) --+ C°° (aX, APTX ) which associate to each w E C°° (X, APTX) its trace rw E C°O (aX, APTX), rw(x) = w(x), x E aX. As aX is a submanifold of X, the Riemannian structure on X yields a canonical decomposition of TxX in two orthogonal components TxX = Tx5X EB l v(x).
Such a decomposition then extends to p-vectors, and we can decompose every p -vector l; E APTXX in a unique way as
6 = i + 62nv,
61 E APTxax, S2 E AP-1TxaX.
Thus, give a form w E C°° (X, APTX) we can define the tangential and the normal parts of the trace of w, respectively tw and nw, as the functions in C°° (aX, APTX) defined by (1)
_
for
= 61 + 6 Av(x). Obviously we have rw = t w + n w and
(2)
IrWI2 =
ItwI2
+ InwI2
Also, if i : aX --> X is the immersion and it : X fl UEo -* aX where U60 is a neighborhood of aX is a smooth right inverse of i, then
550
5. Cartesian Currents in Riemannian Manifolds
tw(x) _ 7#ti#w(x).
(3)
In local coordinates (x', ... , X1, t) around xo E 8X we can write w E Dr (X) as w _ E1a1=P wao (x, t) dxa + E1a1=p-1 w,,,6 dxaAdt, then
t w(x, t) _ E wao (x, 0) dxa,
nw (x, t) =
wao (x, 0) dx' Adt. 1a1=P-1
IaI=P
We recall now that the restriction at the boundary f -> flan of smooth functions f defined on a Lipschitz domain .f2 extends to a continuous linear
operator from W',8(0) to LS (8Q) for s > 1. Based on that it is then easy to see using local coordinates that the trace operator r : C°° (X, APTX ) C°° (aX, APTX) extends to a continuous linear operator, the trace operator
r : Wl,'(X) -> L'(aX)
for anys>1. From (2) we then infer that the tangential and normal operators extend to linear continuous operators
t,n : WW'S(X) -* LP(X)
for all s > 1. In an oriented manifold, the associated Hodge operator * commutes tangential and normal operators. We have
Proposition 1. Let X be an oriented manifold, let * be the Hodge operator on X, and let S be the formal adjoint of d. Then we have (i) Forw E DP(X) *nw = t (*w),
*t w = n (*w),
t (dw) = d(t w),
n (bw) = 8(n w).
(ii) For wEDP (X),rlED4(X),q
which is also orientation preserving. In such a system, if w = wao dxaAdt as above, one easily computes
wao (x, t) dxa +
o(c , a)wao(x, 0) dxandt = n (*w)
*t W = Ia1=P
*nw =
Q(c , a)
O (x, 0) dxa = t (*w).
lal=P-1
Moreover, from (3) we infer
t (dw) = lr#i#dw = d(7r#i#w) = d(tw).
2.6 Relative Cohomology: Hodge-Morrey Decomposition
551
Equality nSw = Snw then follows as 6 = (-1)P("-P) * d*. Finally, again from (3) we infer t (WA * 77) = 7r*i# (wn * 77) = 7r#i#wAlr#i# * 77 = t wAt (*77) = t wA * n77.
Recall that if X is an oriented manifold, then one fixes an orientation on OX by choosing aX (x) as the unit (n - 1)-vector field orienting TXaX in such a way
that (x) = X(x) We then set for y E C°°(0X, A'-'TX) v(x)A
'Y
ax
f
for X E aX.
> dH"'
<'Y,
ax
Proposition 2. Suppose that X is oriented and let * be the associated Hodge operator. We have (i) If w E W,1,'11(X ), then fx dw = fax i#w (ii) If W E Wp,2(X) and 77 E W112 (X), then (
= fax t w.
I 77)Lp}1(X) + (w I S7I)LP(X) _ f twA * n?]. ax
(iii) If W E WP2'2 (X) and C E W 2 (X ), then (
d()L2P}1 (X) + (6w 16C)Lp_1(x)
= -(nw I OLP(y) + f {t(A *n &-,) + t Swn * n (}. ax
Proof. Since x"' LX-a.eT#W(x) =rW(x)IAPT ox and i#W(x) =tWIAPT,aX, (i) follows from a similar claim for smooth w, on account of the continuity of the trace and of the tangential part operators. To prove (ii) we observe that wA * 77 belongs to Wn' 11 (X), and d(wA * 77) = dwA * 77 + (-1)Pwnd(*77) = dwn * 77 + wA * 877.
Integrating we then get on account of (i) (dw 177)Lz + (w I S77)L2 =
d(wA * 77) = ax
f ax
J x
dwA * 77 + J wA * 677
x i# (WA * 77) = f i#wni# ex
* 77.
552
5. Cartesian Currents in Riemannian Manifolds
On the other hand i#w = t w I A,TaX, hence i#wAi# (*7)) = t wAt (*77) I APTX = twA * bn 77I ApTaX .
Finally (iv) follows at once integrating by parts, (zlw I () = (ddw I ()+(8dw I 11
We now observe that Proposition 2 extends to non orientable manifolds. In fact as any non orientable manifold X is nevertheless orientable in the small, the scalar function
(x) := ,
x E 8X
is well defined if either a E C°° (X, APTX) or 0 E Coo (X, AP+'TX) are supported in a coordinate patch U, and is independent on the local orientation on U. Therefore < a, /3 > is a well defined function in CO° (8X) even if X is non orientable.
Proposition 3. We have Wp,2 then t (dw) = d(tw), n (bw) = b(nw). (i) If W E then (ii) If w E WP(X) and 77 E
f
(dw I fl)LP+,(X)+(w I Si)L2(X) =
dHn-1
ax
(iii) If W E Wp'2(X) and r) E WP+2(X), then (dw I b?7)L2P+, (X)
=0
provided either t w = 0 or n7? = 0. (iv) If w c Wp'2(X) and 77 E Wp"2(X), then (dw I dr))LP+,(X) + (8w I drl)Lp_,(X)
_ -(nw I
77)Lp(x) +
J ax
{+ } d7-l'i-1.
Proof. (i) follows from (i) of Proposition 1 as the claim is local. (ii) and (iv) of Proposition 2 proves respectively (ii) and (iv) if we assume moreover that rl is supported in a local chart. The general cases of (ii) and (iv) then follows using a partition of unity. Finally if tw = 0 and 77j ri in W1"2, the r)j being smooth, we have by (ii)
(dwI drlj)_(dwI arli)+(wI
6277j)=
Jd?i'-1=0. ax
(iii) then follows letting j tend to infinity. Similarly one proceeds if nr) = 0 by approximating w.
2.6 Relative Cohomology: Hodge-Morrey Decomposition
553
The Hodge-Morrey decomposition theorem. A key step toward a decomposition theorem similarly to the boundaryless case is a Gaffney type lemma at boundary points. We have
Lemma 1 (Gaffney). For each xo E aX there are constants c, p > 0 and an admissible chart cp : B+(0, p) --p U C X, (0) = xo, such that
f
D(w, U) >
(4)
E JDiwa12 dx - cII w IIL2(U)
B+(O,p)
i,a
whenever cp#W = E wa dxa is compactly supported in B(0, p) and we have either
tw = 0 or nw = 0 on 6X. Proof. Proceeding as in the proof of Gaffney lemma in the interior, we can bound from below the Dirichlet integral D(w, U) by the flat Dirichlet's integral Do(cc#w, B+(0 p)) getting f D(w, U) > lDiwa12 dx - c1l w 11L2(U) p)) 2
J
B+(O,p) i'a
The coercivity inequality (4) then follows at once from the following lemma.
Lemma 2. We have
Do(w, B+(0 p)) = f
IDiwal2 dx,
i'a B+(o,p)
if w = E wa dx' is compactly supported in B (0, p) and moreover either t w = 0
ornw=0 onFp. Proof. From Proposition 2 (iv) we have (5)
Do (w, B+(0, p)) + f (dW I w) = B+(O,p)
r
J
{f wA * n dw + t bwn * n w}.
rv
On the other hand in Rn the Laplacian on forms is just the Laplacian on coefficients, i.e.,
LXW = E / wa dxa, a
consequently, integrating by parts and using Gauss-Green formula
f B+(o,p)
we infer
Digi dx =
f aB+(o,p)
E(-1)x_1gi
i
dxi
V g E C'(B+(O, p)),
5. Cartesian Currents in Riemannian Manifolds
554
(6)
(aw j w) dx
r B+(O,p)
IDw12
I
+
WaWa,n dxlA...
(-1)n-1
B+(O,p)
/ndxn-1.
1p
From (5) and (6) we then get (7)
Do(w,
B+(0, P)) -
IDw12 dx
J
B+ (O,p)
= f{twA *n dw+tbwn *nw+ (-1)n r, wawa n dx1A...
Adxn-1}.
a
rp
We now claim that the right hand side of (7) vanishes if either tw = 0 or nw = 0
on I''. In fact, if tw = 0, then, in I'p we have w = nw =
wa dxa, consequently a3n
t 8w
EE
o (i, a - i)Wa i
dxa-'=(-1)P-1 I
a-i3Jn a3n i
Hence t bwn * nW =
a3n
` 1` wQwa,nu(,a, ) dxa-nAdxQ.
(-1)P-1
3 n, we have dxc
(_i)P-1
twA * nw
wan dxa-n.
a3n
Adxa 54 0 if and only if a = 0, therefore
wawa na(a, a)o(a - n, a) dx1A... a3n
(-1)P-l+n-P
ndxn-1
wawa,n dx1A ... a3n
=
(-1)n-1
E WaWa,n dX1 A ...
ndxn-1
a3n which proves the claim in this case.
Ifnw=0, then in f wehavew=tw=
wadxa, a on
n6w =
E 1: a i, a)wa,i dxa+2 = (-1)P a+i3n aln
Wa,n dxa+n
aon
and
*ndw = (-1)P
Wa,nQ(U + n, a - n) aRfn
Therefore
d?-n.
ndxn-1
2.6 Relative Cohomology: Hodge-Morrey Decomposition
555
t wA * n dw = (-1)P E E wpwa,no-(a + n, a - n) dxandx«-n. )33(n
a
n, we have dxQAdxa-n = 0 unless a
hence
twn * ndw
_ (-1)P E wawa na(a,
a)(-1)n-p-10'(a,
a - n) dx1A... Adxn-1
aon
_
(-1)n-1
EWawa,n
dxlA...
Adxn-1
a5Jn
which concludes the proof.
Gaffney's lemmas in the interior and at the boundary now yield at once
Theorem 1. Let X be a compact Riemannian manifold with boundary. There exist constants cl > 0, c2 > 0 such that D(w) > cl II w IIwp.2(X) - c2II W II2P(X)
for every w E WP,2(X) satisfying one of the boundary conditions
nw = 0 on 8X,
tw = 0 on 8X.
Set now
WPT(X) W12
(X)
HP (X) HPN (X)
{wEWP'2(X) I tw=0} :_ {w E Wp'2(X) I nw = 0} {WEWP"T(X) I dw = 0, dw=0} {W E WP, N (X)
I dw = 0, Sw = 0}.
Then, as in the boundaryless case, one proves
Theorem 2. HT and HN are finite dimensional vector spaces. Theorem 3. There exists a positive constant c such that D(w) > CII W II L2(X) for any w E WP'N(X) which is orthogonal in L2 to HN and for any w E WP'T(X) which is orthogonal in L2 to H. Given a form f E L2 (X), we can now seek a weak solution of the problem
J '6w = f in X (8)
nw=0 onBX
that is we seek a form w E WP,'N (X) such that
5. Cartesian Currents in R.iemannian Manifolds
556
(
(9)
dC)L2 + (SW I SC)L2 = (f I ()
or a weak solution of
V
C E WP,'N(X)
J aw= f inX
(10)
tw=0
on &X
equivalently w E WP 'T (X) such that (11)
(dw I d()L2 + (SW SC)L2 = (f C)LZ
V C E WP'T(X).
As in the boundaryless case, an easy consequence of Theorem 1 and Theorem 3 is the following
Theorem 4 (Hodge-Morrey). We have (i) Problem (8), equivalently equation (9), is solvable if and only if f is orthogonal in LP to HN . Moreover there is a unique solution wo E WP'N (X) n HN
of equation (9), and the linear transformation from f to w° is a bounded linear operator from LP (X) n HN into WP'N(X) n HN . (ii) Problem (10), equivalently equation (11), is solvable if and only if f is orthog-
onal in LP to H. Moreover there is a unique solution w° E WPT(X) n HT of equation (11), and the linear transformation from f to w° is a bounded linear operator from LP(X) n HT into WP'T(X) n HT . Problems (8) and (10) are easily seen to be elliptic boundary value problems, where one has zero Dirichlet data for suitable coefficients of w and zero Neumann
data for the others. Elliptic regularity theory then allows to complement the existence theorem in Theorem 4 with the following Regularity Theorem
Theorem 5. Let w in WP'N (X) or in W' 2 (X) be respectively a solution of (9) or of (11). Then we have (i) w E WP2'2(X) if f E LP(X). (ii) w E WP's(X) n CO, ` (X, APTX), a = 1 - P, if f E L; (X), s > n. (iii) w E C2,- (X, APTX) if f E C°Ia(X, APTX). (iv) Higher regularity of w follows from higher regularity of f, for instance w E C°° or w is analytic if f E C°° or is analytic. (v) In particular harmonic normal forms, i.e., w E HN, or harmonic tangential forms, w E HT, are C°° or analytic up to the boundary if X and aX are of class C°° or analytic. Define now
ImdN := {da ImSN := {Sa then we can state - ---------
a E WP'21,N(X)} ImdT aE
W1,2
N(X)} ImST
{da I a E WP'21,T(X)} {Sa I a E
---------------------------
- - - - ----_
2.6 Relative Cohomology: Hodge-Morrey Decomposition
557
Theorem 6 (Hodge-Morrey decomposition). We have (i) The spaces HN, Im dN and IM SN are mutually orthogonal and closed in
LP(X), and one has the decomposition
L2r(X) = HN ®ImdN ®ImSN, that is, every f E L2 (X) decomposes uniquely as
f =h+da+Sf
(12)
where h E HN, a E Wp 1,N(X), /3 E WP+1,N(X) and IIhIJW,,2+IIaIIw1.2+II13 IIWI,2 <_ cII f IILZ'
(ii)
The spaces HT, Im dT and Im ST are mutually orthogonal and closed in L2r (X), moreover
L22(X) = HT ®ImdT ®ImST, i.e., every f E LP2 (X) uniquely decomposes as
f =h+da+SQ
(13)
where h E HT, a E W l T(X), 3 E
and
IIhIIW1,2 +IIa1W1.2+IIQIIW1,2
Moreover in both cases h E C°° (X, APTX) and also a and Q are of class Cup to the boundary if f is of class C°° up to the boundary. Finally, SQ = 0 if
df =0. Proof. The proof follows the same path of the analogous proof of the decomposition theorem in the boundaryless case: one has only to take into account the formulas of integration by parts in Proposition 6 in Sec. 5.1.3. For the reader's convenience, we only prove (12). Let h be the projection (in L22) of f into HN and let wo E Wp'N (X) n HN be the unique solution of (14)
(dw I d()L2 + (SW I SS)L2 = (f I ()L2
V
E Wp'N(X )
which exists by Theorem 4. By the regularity theorem wo actually belongs to Wp'2, therefore (iv) of Proposition 3 yields (15)
(SW +J{
(dw I dC)L2
+
rr (QW I S)L2
+} dl-C-1
ax
for all C E WP'2(X). From (14), (15) we infer by taking C E C°°(X, APTX) that
5. Cartesian Currents in Riemannian Manifolds
558
Awo=f - h i.e.,
f = h + da + 60 if a := dwo, /3 := Swo E WP+1 (X) . As n a = n Swo = Sn wo = 0, we see that a E Wp'21,N(X). From (14), (15) we also infer where a E WW_21(X ),
J ax
< t C, n dwo > dx" = 0
V C E Wp,'N (X ),
and, being t( arbitrary on X, that n dwo = 0, hence no = it dwo = 0, i.e. 1,2
E WP+1,N(X)-
Remark 1. It is worth remarking that, since Hodge operator transforms pharmonic forms into (n - p)-harmonic forms and
*nw=t(*w)
*tw=n(*w),
actually on oriented manifolds Theorem 4 (ii) follows from Theorem 4 (i) and Theorem 6 (ii) follows from Theorem 6 (i). Remark 2. One might consider other boundary value problems for the LaplaceBeltrami operator and consequently infer other decomposition theorems. However we shall not pursue this topic any further. For k = 0, ... , n, define
Zk(X, aX) :_ {w E Co(X, AkTX) I i#w = 0, dw = 0} to be the vector space of relatively closed k-forms, i.e. the space of closed k-forms which vanish on aQ, and
Bk(X,OX) :_ {w = da I a E CO(X,.AkTX), i#a = 0, or w = 0 if k = 0}. the vector space of relatively exact k-forms. The relative de Rham cohomology groups are then defined by
HHR(X, aX) := Zk(X, aX)1Bk(X, aX). An immediate consequence of Theorem 6 is
Corollary 1. We have (i) Each cohomology class [w] E HHR(X) contains exactly one harmonic form h
with n h = 0. In particular HHR (X) is finite dimensional. (ii) Each relative cohomology class [w] E HHR(X, 9X) contains exactly one harmonic form h with t h = 0. In particular HHR (X, OX) is finite dimensional.
2.7 Weitzenbock Formula
559
2.7 Weitzenbock Formula In this section we derive a more intrinsic expression for the Laplace Beltrami operator on a Riemannian manifold X in terms of the covariant derivatives and of the curvature of X. Denote by r(TX) the smooth vector fields on X and if p : E -* X is a fiber bundle on X, denote by r(E) the space of sections over E. A connection V on E is an operator
denoted by Vxt; for X E V(TX) and
E r(E) which satisfies the following
properties
(i) V X t; is tensorial in X i.e.
Vx+Y = OX6+VYe,
OfXe = fvXe
for x, y E r(TX ), 6 E r(E), f E C°° (X ). (ii) Vx6 is linear in iii) Vxl; satisfies the following chains rule
fVxe
Vx(fe) =
(1)
for X E r(TX), f E C°°(X), 6 E r(E). Notice that (i) implies that Vxl; at a point x E M depends only on the tangent field X at point x and moreover that V can be also seen as an operator
V : r(E) -* r(E) ® r(A'TX). Moreover, (iii) implies that Vxe(x) depends only on t= near x. Given a connection V on a fiber bundle E one defines a connection V * on the dual bundle E* by the formula
w> _
(2)
for e E r(E), w E r(E*), X E r(TX) and also on all exterior algebra of E. In fact sections
E r(11pE) can be represented locally as
6=
S°(x)eai
(x)A...
Ae%(x)
aEI(p,n)
where a c Coo (X) and eat, ... , eaP are smooth sections in E. Consequently, by setting (3)
Vx f (x) :_,
(4)
Vx(t 1A... Abp)
P
i=1
f E C°°(X)
S1^ ... Si-1^OX6ti^Ci+1A ... Al p
560
5. Cartesian Currents in Riemannian Manifolds
E N(E) we define an operator
for
v: N(APE) ®T(TX) --> T(ApE)
which is easily seen to be a connection. Starting with V* on N(E*) a similar construction produces a connection on APE*
APE*
P
Vx(w'A... AW) :_ Ew1A... Awz-1AVxw'Awx+'A... nwP,
(5)
i=1
Proposition 1. Let 6 E N(APE), o E T(AgE), w E N(APE), 77 E N(APE). We have (6) (7)
OxeAo + eAV x o Vx(wAr7) =Vxwnr7+WAVxr7
(8)
<w, Vxe> += , X >.
Vx
Also, if p < q
Vx(e-J 77) _ CJ Vx?7+Vxl;J 17.
(9)
Proof. Let e1i... , e, be a local basis of sections near xo and fl, ... , fn be the dual basis in N(E*) so that < fi (x), ej (x) > = Jij
near x0.
If a E I (p, n), 0 E I (q, n), from (4) we infer Vx (e,,,Aep) = Vxe,,,Ae,3 + e,AVxeo hence (6) follows. Analogously (7) follows from (5). To prove (8) notice that
V*fa(ea) = f«(Vxeo) = 0 if a
,3
while for a = Q P
Oxfa(eo)+f, (Vxe«)
{Oxfck i(eai +fai(Vxe«i)} i=1
,X>=0. Let us prove (9). Using (6) and (8), for o E T(Aq_pE)
V_ 77)(o,) = -eJ 77(V xo) + < d<77, one >, X > = -77(OxoAe) +
X>
= -77(Vx(o-ne))+,X>+
_ C-J V*77(o) + Vxe-J 77(C)
2.7 Weitzenbock Formula
561
A basic result on Riemannian geometry is that there is only one connection V on TX, the Levi-Civita connection , which satisfies the following additional requirements (1) V is metric, i.e.
(VxYIZ)+(YIVxZ) _VX,Y,ZET(TX) (10)
(2) V is torsion free, i. e. for X, Y E T (T, X) (11)
T(x, y) =VxY-VyX - [X, Y] - 0
being [X, Y] the Lie bracket of X and Y. In this case, d
[X, Y] := X3
aXj
ei - Yj axi e2
where X := Xiei, Y := Yiei.
One can compute V in terms of the metric on T (TX ), and one has
(VxxlZ) = 2 {x(X Z) - Z(XJX) + x(XIZ) (12)
-(XI[X,Z])+(ZI[X,Y])+(XI[Z,X])}.
In local coordinates, denoting by Gij the metric tensor of X and by e1, ... , en a local basis for I'(TX), we have n
(13)
>rkj ek
Deyej
k=1
where {T kj} are the Christoffel symbols of the connection in the given basis; (12) is then equivalent to (14)
T = 2Gk1{Gil,j + Gj(,i -
Gij,1}.
The torsion free condition (11) implies
T =rk
di,j,k.
Proposition 2. Let X be a Riemannian manifold. Denote by V the Levi-Civita connection in X and also its extension to the algebra of p-vector fields and forms defined on an open set U c X and let dx...... dxn be the dual frame of p-forms. Then dX AVej W in U. (15) dW If moreover ej are orthonormal at xo E U then at xo (16)
dW = >ejJ Ve, w.
j
5. Cartesian Currents in Riemannian Manifolds
562
Proof. The idea is to prove the invariance of both sides in (15) and (16) with respect to changes of coordinates and then proves (15) and (16) in a suitable frame.
Let us prove (15). First we note the the expression
E dxj AVejw
j is independent of the choice of local frame In fact if fl,... , f., is another frame field, we have f j = A ek for some A E COO (U, ]R). For the corresponding dual frame f 1, ... , f' we then have f j = Bkdxk with B := A-1. Consequently
f3AOfjw =
j
8kdxkAVe=w = dxiAV,,w.
w= j,k,i
i,k
i
Since d is also independent on the frame field, it suffices to check (15) for a special frame field. We select around x0, normal coordinates for which we have Gij(xo) = 5ij, Gij,l (xo) = 0 V i, j,1. Hence for the coordinate frame fields ei,... , en, we have
Veiej(xo) = f (xo)ek(xo) = 0 from (13) (14). Thus, if w = codxa, a E I(p, n), f E C°O(X), we obtain at xo
EdxiAvej(cpdxa) _
dxa = d(c'dxa).
dxjA 9
j
To prove (16) we observe that the right hand side is independent of the chosen of
the normal frame at x0. In fact if fl, ... , fn is another orthonormal frame field, f j = A3 ek, A3 being an orthogonal matrix, A A3 = 5ik. Then we have
eiJ V.
Ve;w =
fi-J V f,w =
j
j,k,i
i
Again we chose normal coordinates at xo. The coordinate orthogonal at x0 and V ej = 0. Consequently
ejj Vet (pdxa) =
(17)
j
en) frame is
a e J dxa.
8xj
On the other hand from (1) in Sec. 5.2.4, taking into account that Gij(xo) _ Gti(xo) = V, Gij,a(xo) = 0 we have at x0 1:(5(Wdxa))7 dxl =
6(Wdxa)
(18)
_
a (i,'y) y+i=a
&p
(xi
aacpxe
dxa = iEa
o(i, a) dxa-i
2.7 Weitzenbock Formula
563
and the claim follows comparing (17) (18) since
ifi0a
0
ej i dxa = iS
o-(i, a)
dxa-2
if i E a. 71
In order to express the Laplace-Beltrami operator in terms of covariant derivatives, we introduce the second covariant derivative defined for w E F(APX) and X, Y E TX by
OXYw :_ VX(VyW) -Vvxyw and the curvature tensor associated with a connection (19)
(20)
R(X, Y) :_ VxVy -VyVX : APTX -4 APTX. is tensorial in X, Y, i.e. that OXY is linear in X
It is easy to check that and Y and for all xo E Y
oXY(W)(xo)
depends only on the values X (xo ), Y(xo) of the vector fields X and Y at xo. In fact, for f, g E C°°, X, Y smooth vector fields,
VfX(V9yw) = fVyw+fgVX(VYw) and
VVfx(9Y)W = V fX+f90xY = f Vyw + fgVvxyw so
v2fX,YW = f(x)g(x)VXYW. We need also the following algebraic lemma.
(21)
Lemma 1. Let e1,. .. , en E Al V, dx', ... , dx" the dual basis in A'V, w E APV. Then (22)
> {eiJ dxjAw +dxjA(eiJ w)} = 0
(23)
eiJ dxinw + dx1A(eiJ w) = w
ij
Proof. Assume first i
V i = 1, ... , n.
j if j E a, if j a, i 0 a 0 -i if j a, i E a. o, (j, a) o, (i, a + j - i) dxa+j
10
eiJ (dxjndxa) =
ifi a,
10 dx3A(eiJdxa)
ifiEa, jEa
0
v(i, a - i)o,(j, a - i)
dxa+j-z
if j
a, i E a.
5. Cartesian Currents in Riemannian Manifolds
564
a + j - i) = -o (i, a - i)o-(j, a - i), compare (5) in Sec. 5.2.4, (6) As v(j, in Sec. 5.2.4, we see that eiJ (dx3Adxa) + dx3A(eii dxa) = 0
`d a E I(p, n), consequently (22) is proved. To prove (23) it suffices to compute
ifiEa,
(0
eii dx"Adxa = {
dxa
if i
dxineiJ dxa =
a,
r dx' ifiEa
Sl
0
if i
,
c
,
thus
eiJ dx2Aw + dxiheiJ w = w.
Theorem 1 (Wietzenbock formula). Let X be a Riemannian manifold and let V be the Levi-Civita connection on forms. If el, ... , en is a local frame on X which is orthonormal at xo and dxl,... , dxn the dual frame on T*X, then
Qw = Deieiw + dxzA(ejl R(ei, ej)w)
(24)
Proof. By definition, Aw is intrinsic to the Riemannian manifold and depends on the Riemannian structure on X. Also we check that both terms in the right hand side of (24) are intrinsic. In fact, if fl, . . . , f, is another orthonormal frame field, then fj = A ek, A (xo) being orthogonal. Thus (21) implies Vfifiw = 1: AkAt Vekehw =
Dehehw. h
j,h,k
i
Also if fj is the dual basis of fj, fj = Bkdxk, B
R(fi, fj) _
A-1
AkA; R(ek, eh) h,k
fj i R(fi, fj) i
hkj,s
AkehJ R(ek, eh)
AsAi Aj esJ R(ek, eh) k,h
Therefore
ftin(fjJ R(fi, f3)) _
BsAkdxsn(ehJ R(ek, eh)) h,k,s,i
dxkn(ehJ R(ek, eh)). h,k
Consequently it suffices to prove (24) for a special frame. Choose normal coordinates around xo, and dual coordinate frames e1, ... , en, and dxl,... , dxn. At xo we have Veiej(xo) = 0, therefore Vei(dxj)(xo) = 0, Deieyw(xo) = VeiVeiw(xo) and R(ei,ej) =Ve;,Vej - VejVei
2.8 Poincare and Poincare-Lefschetz Dualities in Cohomology
565
Consequently, using Proposition 2 at xo (25)
dxin0ei(ejJ Vejw)
dSw =
dxin(ejJ DeyVejw)
i,j
i,j
and also using Lemma 1
Sdw = EejJ Vej(dx'AVeiw) i, j
E ejJ (dx'AVejVeiw)
- 1: dxin(ejJ Dej Veiw) i=j
i,j n
(26)
+ejJ (dxinvejVei.w)+dx2A(eiJ DeiVeiw)} i=1
E dx2A(ej-i Vej Ve, w) +
Dei Vei
i, j
Adding (25) and (26) we get the claim.
11
Remark 1. If p = 0, i.e. if w = f E C°°(X) then R(ei,ej)f = fR(ei,ej)1 = 0 because of the tensorial property of R, consequently for functions
If =Eoeieyf. 2.8 Poincare and Poincare-Lefschetz Dualities in Cohomology Compact manifolds without boundary. Let X be a compact manifold without boundary. Recalling Hodge's theorem Theorem 6 in Sec. 5.2.5 we first notice that Hodge operator * induces naturally an isomorphism between the homology classes HHR (X) and HHR' (X ), in fact since
*w=*,Aw we see that *w is harmonic if w is harmonic, hence the map * : w ---* *w defines an isomorphism
* : HaR(X) --> HHRP(X). However such an isomorphism depends not only on the orientation of X, but on the metric. An algebraic isomorphism independent from the metric between HHR (X) and the dual of H(X) is obtained as follows. Given two closed forms w E W1'2 (X, APTX) and 77 E W"2 (X, An-PTX) it is readily seen that
<w,77> :=
xJ
wnr7
does not change if we replace w and 77 by cohomologous forms, hence < , > defines a bilinear map over HHR (X) x H(X)
5. Cartesian Currents in Riemannian Manifolds
566
Poinc<, >: HaR(X) x HHRP(X) --> J R. We have
Theorem 1 (Poincare duality). The bilinear form Poinc<, > is non degenerate, that is for every non zero cohomology class [w] there is a non zero class [77] such that Poinc< [w], [77] > 54 0,
consequently HHR(X) is isomorphic to the dual (HaRP(X))* of HdRP(X).
Proof. For each non zero cohomology class that we may assume represented by an harmonic form w we see that *w represents a cohomology class in H.R P (X ) for which
J
wA*w=(w,w) #0.
The second part of the claim follows at once from Lemma 1 below as de Rham's cohomology classes are finite dimensional vector spaces.
Lemma 1. Let E, F be Hilbert spaces and let a(.,) : E x F -+ R be a bilinear map satisfying (i) Ia(u,v)I0 such that a(uo, v) > co I v I V v E F
(iii) a(u,v)=0VvEF implies u = 0 Then F ^ E*. Proof. Consider the linear map T : F -> E* defined by TT(u) := a(u, v). By (i) E- < k v IIF; also, by (ii)
T is continuous and T
IIVIIF <
IITV IIE
11uo
It follows that T has closed range, is invertible, and T-1 is continuous. It remains
to show that range T = E*. Assume not, then there is 0 E E*, 0 0 0 such that coo E E we then (OITv)E- = 0 V v E F. Representing 0 as 0(u) = (ulcpo), find 0 = (OIT,,)E` = Tv('po) = a('po,v)
(iii) then implies coo = 0, i.e.,
b v E F,
= 0: a contradiction.
Notice that in particular Theorem 1 yields
Proposition 1. Let w be a closed form in X. If fX wnrl = 0 for all rl with d77 = 0, then w is exact.
2.8 Poincare and Poincare-Lefschetz Dualities in Cohomology
567
Compact manifolds with boundary. Given a smooth compact Riemannian manifold X with boundary, we may define, as we have seen, three sets of cohomology groups depending on k = 0, ... , n: The de Rham cohomology groups k HaR(X) _ closed forms of degree k in X
exact forms of degree k in X '
no boundary condition is imposed; for w E Zk(X), we denote by [w]x the cohomology class of w,
The de Rham cohomology groups of the boundary aX k HaR(aX) _ closed forms of degree k in aX
exact forms of degree k in aX
for w E Zk(8X), we denote by [w]ax the cohomology class of w in HHR(aX),
The relative cohomology group of relative closed forms modulo relative exact forms i.e. HkdR (X
ax)
closed forms of degree k in X which are null on OX exact forms of degree k in X which are null on OX
that we have already defined in Sec. 5.2.6. The relative cohomology class of w E Zk (X, aX) in HHR (X, aX) is denoted by [w] rel.
The inclusion map i : OX -f X clearly induces a map i# : HHR(X) HHR(OX ). As forms in Zk (X, OX) and in Bk (X, DX) are just special closed and exact forms, it is easily seen that an inclusion map j* : HHR (X, OX) -# HR(X) is well defined by 7#([w]rel) = [w]x. Finally, one can define a coboundary operator b : HHR(aX) -, Hk+1(aX) by b([w]eX) _ [dw]rel
where ID is an extension of w in X. In fact, if w extends w to X, we have i#(di.-5) = d(i#w) = dw = 0
on aX,
i.e., dW E Zk+1(X, aX ); moreover, if w is another extension of w, then
and i#(w-w)=0, i.e., [dD]rel is independent of the choice of the extension w of w. Also, if w E Bk (aX ), w = da, a E Dk+l (aX ), and a E Dk+1(X) is an extension of a, we have dw = d(w - da)
and i#(w - da) = w - w = 0.
5. Cartesian Currents in Riemannian Manifolds
568
Recalling now that a sequence of vector spaces
+ Vi --f Vi+l - .. .
Vi-1
is said to be exact if for all i the kernel of fi is equal to the image of its predecessor
fi-1
ker fi = Im fi- 1,
we can state
Theorem 2. The long sequence
... ?#
HaR(X)
2#' HHR(aX) a ' HHR1 (X, aX)
'- Hk+1(X)
is exact.
Proof. For the reader's convenience we give the simple proof. (i) First we show that the composition of two successive maps gives zero. Let [w]re1 E HaR(X, aX). We have 2#(7#([w]re1))
= i#(1w)x) = [i#W]X = 0
as i#w = 0. Let [w]x E Hk(X). We have b(i#[w]x) = 6([i#w]ax) _ [di#w]re1 = [di#w]rel = [2#dw]re1 = 0.
Let [w]ax E Hk(aX). We have j#(8([w]ax)) = j#([dw]rel) = [dw]X = 0 as dw is exact in X (though not relatively exact). (ii) Suppose that i# ([w] x) = 0, we need to construct c:i so that j# ([W]re1) = [LO]X.
From the assumption [i#w]ax = 0, i.e., i#w is exact in OX, i#w = dcp in OX. We extend cp to cp in X, define W = w - cp, and verify ill (Lo
- dip) = 0,
j#([w]re1] = [W -
x = [w].
(iii) Suppose that [w]ax E Hk(DX) and 8([w]ax) = 0. Extend w tow in X, i#iv = w; since [dZ]re] = a([w]ax) = 0, dw is relatively exact, i.e., div = dcp with i#cp = 0. If we define = w - cp we have
dw=di3-dcp=0 i#[W]x = [i#w]ax =
[i#cp]ax = [w]ax.
(iv) Finally, let [w]rel E HHR(X,9X) and j#[W]rel = [w]x = 0, W being relatively
closed, du) = 0 and i#w = 0. Then w is exact in X, w = dep, and [i#coiax E Hk-1(aX), as di#cc = i#w = 0, 8(i#cP]ax) = [dcP]rei = [w]rel
2.8 Poincare and Poincare-Lefschetz Dualities in Cohomology
569
Returning to harmonic forms, we recall Remark 1 in Sec. 5.2.6. i.e., that Hodge * operator yields an isomorphism *:xN-'HTk
we easily infer that HaR(X) and HaR k(X, aX) are isomorphic. However the duality given by
* : HdR(X) - HHRk(X,aX) depends on the metric (and on the orientation) of X. On an oriented manifold, an algebraic isomorphism between HdkR(X) and the dual of HHRk(X,aX) is instead given by the map (w, 71) - JwAm
x It is in fact easily seen that fx (w + d17)A(C + do,) = f x wA if w E Zk (X), ii E Dk-1(X), e E Zn'-k (X, aX), and o, E Bn-k-1(X, aX). Therefore such a map induces a bilinear pairing Poinc< [w], [c] > : HaR(X) X HaR k (X, aX) -+ R
between the k-cohomology and the relative (n - k)-cohomology. We have
Theorem 3 (Poincare-Lefschetz duality). The pairing
Poinc<[],[]>:HHR(X) xHaRk(X,aX) _+R given by Poinc< [w], [l;] > :=
J x
wA
- HaR(X,oX)
'-
is non degenerate. Moreover the diagram i#
HHR(X)
'+
x Poinc b
HHRk(X,aX)
HHR'(OX)
x PoincT
'#
x Poinc i#
HHRk(X)
...
HHRk(aX)
b ...
is commutative. PoincT denotes the transposed duality, PoincT< 1771 r'!1' [w] > Poinc< [W], 1771 ]rel >
The commutativity of Al
xF1 A2
B1
xF2 B2
means F2(f(a,),b2) = F1(a,,g(b2)), Val E A1, Vb2 E B2.
570
5. Cartesian Currents in Riemannian Manifolds
Proof. If [w] is a non trivial cohomology class we can represent it by a non-zero harmonic form h with n h = 0. The form *h is harmonic and t (*h) _ *n h = 0,
consequently *h represents a relative cohomology class [*h] E H ' (X,BX), and we have
r
f
Poinc<[w],[*h]>=J hA*h=J Ih12#0. x
x
Similarly one proves that, if [S]rel ; 0, we can find [w] in HHR(X) such that Poinc< [w], 0. The commutativity of the diagram is trivial.
3 Currents and Real Homology of Compact Manifolds Let X be an oriented compact smooth submanifold of dimension n in Rn+N without boundary. We introduce in Sec. 5.3.2 for k := 0, 1.... the k-homology group Hk (X, R) of a X setting Zk(X) Bk(X)
{TENk(U) I sptTCX, 8T=0} {8S I S E Nk+l(U), spt S C X}
and
Hk(X,R) := Zk(X)/Bk(X) where U is a tubular neighbourhood of X, i.e., compare Sec. 5.3.1, as normal cycles in X modulo boundaries of normal currents on X. Two sets of pairings are naturally associated to homology and cohomology. One is the set of de Rham pairings between currents and forms which factors to homology and cohomology
<,>:Hk(X,lI)xHaR(X) defined by < [T], [w] > := T (w). The others are the pairings (w,71) E Dk(X) x
Dn-k(X) _
fwA7]
x which, by Stokes theorem, induce Poincare pairings in cohomology
Poinc<, >: HHR(X) X HHRk(X) -> R given by Poinc < [w], [ri] > := f wnr7. Classical theorems, respectively de Rham's
and Poincare's duality theorems, state in fact that de Rham and Poincare pairings are non degenerate; consequently there are isomorphisms, Poincare duality isomorphisms P : HHRk(X) -' Hk(X,
3 Currents and Real Homology of Compact Manifolds
571
between HHR k (X) and Hk (X, R). In Sec. 5.3.2 we show that the Poincare' duality
isomorphism is actually induced by the map again denoted by P, P : Z,-k(X) -a Zk (X) given by w -+ fX Aw. Moreover, by regularizing a k-cycle T E Zk (X) Dn+N-k(U), and integrating such a we construct an (n + N - k)-form w E form on the fibers of the projection map -7r, we construct a so-called Poincare
dual of T, PT E Dn-k(X). The cohomology class of PT turns out to depend only on the homology class of T, therefore T --# PT yields an explicit way to describe surjectivity of P. Non degeneracy of Poincare duality, which we already proved in Sec. 5.2.8 as consequence of Hodge decomposition theorem, then yields injectivity of the Poincare' duality isomorphism and also yields non-degeneracy of de Rham pairing, as < [T], [w] > = Poinc < [w], [PT] > =
fAPT,
x compare Sec. 5.3.2. We then deduce the weak closure of the real homology classes, hence in particular the existence of mass minimizing real cycles in each homology class.
A completely parallel theory holds in the case of compact manifolds with boundary. One has to introduce the relative homology groups, compare (1) in Sec. 5.3.3. The Poincare-Lefschetz duality isomorphism this time consists of three sets of maps
P# :HdRk(X,BX) -> Hk(X,lR) P' : HHR -k (X) -> Hk (X, 8X, IR)
P
:Hank-1(8X)
- Hk (BX, IR)
which actually give a morphism between the long exact sequences in cohomology and homology, compare (9) in Sec. 5.3.3. The duality forms-currents_ T(w) induces three sets of de Rham pairings at the homological level
<,>a:Hk(X,IR)xH,R(X) ---+
IR
< , >' : Hk (X, 8X, IR) x HdR(X, 8X) -> 1R < , > : Kk (BX, R) x HdR (8X) --+ 1R
which are related each other by the commutative diagram in de Rham's theorem, Theorem 5 in Sec. 5.3.3. As consequence we deduce the closure of absolute and relative homology classes and consequently the existence of mass minimizing cycles in each homology class. Still another set of pairings is relevant in homology: these are the so-called intersection indices of cycles of complementary dimension.
In Sec. 5.3.4 we shall define Kronecker's index for two'arbitrary currents S E Nk(X) and T E N,,-k(X) for which spt 8S f1 spt T = spt 8T f1 spt S = 0 and see that it induces a pairing Hk(X,1R) X Hn_k(X,8X,1R) -> R.
572
5. Cartesian Currents in Riemannian Manifolds
It turns out that
ix(S,T) = J Ps APT,
x therefore non degeneracy of Poincare duality is equivalent to non-degeneracy of the intersection index.
3.1 Currents on Manifolds Let X be a smooth manifold of dimension m. As already mentioned, by Whitney embedding theorem we may and do assume that X is a smooth (CO°) subman-
ifold of some l1P', X C R'. Moreover we can assume that X has a tubular neighbourhood U D X, U C Rn, and there is a C°° retraction map 7r : U -> X, rr(U)=X,ir(y)=yVyEX.
Definition 1. A k-dimensional current T E Dk(U) is said to be a current in X if T(w) = 0 for any null form w E Dk(U) to X. The class of k-dimensional currents in X is denoted by Dk(X). From the characterization of null forms to X, compare the end of Sec. 5.2.1, we easily infer
Proposition 1. Let T E Vk(U). Then T E Dk(X) if and only if for some E C°°(U) with ( = 1 on X we have T = 7r# (T we have
T=7r#(TLc)
Actually, if T E Dk(X),
b'(EC°°(U), (=1 on X.
Proof. In fact for any w E Dk(U) the form w - (7r#w E Dk(U) is a null form
to X, hence T(w) - T((ir#w) = 0. Conversely if q is a null form to X, then 7r#w = 0, hence T(77) = a# (T L () (y) = T ((7r#77)=0.
Remark 1. Proposition 1 extends to the case of Ck manifolds, k > 1, provided T is a vector valued distribution of order -k. In this case one in fact can assume that the retraction map is Ck hence 7r#(TLC) is well defined. In particular Proposition 1 holds true for flat chains on C' manifolds. Trivially sptT C X if T E Dk (X) . In fact, if w is supported outside X and
w is a null form to X, we have T(w) = 0. But in general it is not true that spt T C X implies T E Dk(X). Consider for instance the 1-dimensional current in R2 integration of 1-forms along the real axis el against the normal vector e2 L(w) :=
J
<w, e2 > dxl.
Clearly spt L C l[8xi, while for example on the null form W(x1) dx2 we have T (,p(xl)dx2) =
f(x1)dx1.
Notice that M(T) < oo, while M(8T) = +oo, compare 10 in Sec. 2.2.4. However we have
3.1 Currents on Manifolds
573
Theorem 1 (Flatness theorem). Let X be a closed submanifold of Pn. If T is a flat chain in U, T E Fk(U), and sptT C X, then T E Dk(X). Proof. Theorem 1 is a consequence of Proposition 7 in Sec. 5.1.3. In fact the X and the identity map id : U -+ U trivially agree on retraction map 7r : U
X, hence on spt T. Proposition 7 in Sec. 5.1.3 then yields T = 'r# (T L () V ( E C°°(U), C = 1 on X, and the conclusion follows from Proposition 1. By Theorem 1 normal currents on a submanifold X on Rn may be identified to currents in the ambient space IRn with support in X. We shall now see that a constancy theorem holds also for currents on submanifolds. Let X be an m-dimensional C°° submanifold of RTh with locally finite area. Suppose that X is oriented by a smooth rn-vector field X. Then the current
Q X := T(X,1,..) is a locally m-rectifiable current in R'.
Theorem 2 (Constancy theorem). Let .fl be an open set in ][MTh such that .flflX is connected and letT E D,(X), m = dim X, be such that sptaTfl,fl = 0. Then there exists a constant c E IR such that T= c Q X
in D'(Q f1 U)
U being a tubular neighbourhood of X. Moreover c is an integer if T is i.m. rectifiable.
X be the retraction map, and let ( E C°° (U), = 1 Proof. As usual let -7r : U ,(l2 C D on X. We cover .fl fl X by a system of coordinates Vi : V -> Di f1 X
so that cj#U = X; the maps cpi have inverses, and more precisely there are smooth maps z/ii : U fl it-' (.Qi fl X) -> V which are onto and such that W o Oi = ir,
Oi o wi = id,.
Therefore the currents Wi# (T L () belong to D,,,, (V) and, as spt aT fl .fl = 0, we
have fti# (T L () = 0 on Vi. It follows from the constancy theorem in the flat case, Theorem 1 in Sec. 4.3.1 that
'bi#(TL() =G;QV on V, hence
T = it#(T LO) =Wi#4Ji#(T L() = ciEX
on ,fli fl U,
since cpi#U = X. As the .(li's overlap on X, we infer that the constant ci are all equal to a c independent of the chart T = c QX ]J
on Dm(.fli fl u),
and the conclusion follows by means of a decomposition of unity associated to the covering {.Qi}.
5. Cartesian Currents in Riemannian Manifolds
574
Remark 2. Theorem 1 holds also for flat chains on Cl manifolds, compare Remark 1. More generally we have
Theorem 3 (Constancy theorem). Let E be an r-dimensional oriented, connected and embedded submanifold of U of class C', where U is an open set in ]Rn, and let T E F;°` (U). If (spt T) \Z is closed relative to U and spt 8T C U\E, then there exists a real number p such that
spt(T-pjZ]) CU\E. Moreover p is an integer if T is locally integer rectifiable.
If f : X --+ Y is a Lipschitz map between C' manifolds, every T E Nk(X) has a Lipschitz image f#T E Nk(Y). Also if fo, f, : X - Y are two maps and h : [0,1] x X - Y is an homotopy map between fo and fl, h(0, x) = fo(x), h(1, x) = f, (x), then the homotopy formula holds (1)
fl#T- fo#T=Bh#(Q0,11 xT)+h#(Q0,1J x8T)
for T E Nk (X) . Also a corresponding homotopy formula for forms holds (2)
fl#w - fo#w = d[h#(w)](o,i) + [h#(dw)] (o,l),
[i7] (o,l) being the integration along the fibers of the components of rl which are not
tangential to X. Also the standard estimates for homotopies hold true, compare Sec. 2.2.3.
3.2 Manifolds Without Boundary: Poincare and de Rham Dualities By Poincare duality isomorphism one usually refers to the duality between the k-homology group and the (n - k)-de Rham cohomology group of an oriented compact smooth manifold without boundary. In this subsection we shall describe such an isomorphism in terms of currents and forms.
Poincare duality isomorphism on open sets. Let U be an open set in R". For 0 < k < n consider the vector spaces of currents
Zk(U, R) = {T E Dk(U) 18T = 0, M(T) < +oo} Bk(U,R) = {8S I S E Dk+l(U), M(S) <+oc} and set Hk(U, R) = Zk(U, R)lBk(U, R) Analogously we define de Rham's cohomology groups in U by HdR(U, R) := Zk(U, R)/Bk(U,
3.2 Poincare and de Rham Dualities
575
where
Zk(U,R) _ {w E L1(U,AkRn) n£k(U) I dw = 0} Bk(U,1R) _ {dr) 177 E L' (U, AkRn) n £k-1(U)}
Then we have
Theorem 1.
HaRk(U,R) and Hk(U,R) are isomorphic.
To prove this we consider the map P : Dk(U) I M(T) < +oo} given by
L1(U,An-kRn)
p(W) :_ (-1)k(n-k)Rn LW = As 0(1Rn L W) _
(-1)k-1Wn
J
n £n-k(U) , {T E
Au.
L dw the map P factorizes to a map still denoted by
P P : H a R k (U, R) --+ H k (U,1 [ 8) .
We shall prove that P is an isomorphism, indeed the isomorphism of Theorem 1. Of course this amount to prove
Proposition 1. We have (2)
Let T be a cycle of finite mass, aT = 0. Then there is a (n - k) form PT E An-kR2) n £n-k(U) such that for some S E Dk+l(U), M(S) < +oo L' (U, we have
T= 1 APT+aS.
(1)
UU
(ii) Let w E L'(U, An-kRn) n £n-k (U) be such that 1R L W = aS for some S E Dk+1(U), M(S) < +oo. Then w = dr7, 77 E L1 (U, An-k-lRn) n £n-k-1(U).
Proof. We prove both claims by regularizing T. (i) is in fact already proved in (-1)k(n-k)f. Proposition 7 in Sec. 5.1.2, by taking PT :_ To prove (ii), we choose a convolution kernel p and a regularized distance
d. For0<E
Wi L wE = (Rn L W)E = (aS)E = as'.
On the other hand by Proposition 6 in Sec. 5.1.2 (3)
SE = IEBn L 17e
i 6, E L'(U,
An-k-'Rn) n £n-k-1(U)
From (2) and (3) we conclude that wE = (_l)n-kdr)E. As w6 and w are also cohomologous to a form L1(U,
An-k-lRn) n £n-k-1(U) the claim follows.
576
5. Cartesian Currents in Riemannian Manifolds
Remark 1. We observe that the same reasoning of Proposition 1 implies the following:
(i) if the current T in (i) of Proposition 1 has compact support in U, then the decomposition (1) holds with spt PT and spt S arbitrarily close to spt T; (ii) if the form w and the current S in (ii) of Proposition 1 have compact support in U, then one finds a potential for w with support arbitrarily to spt w U spt S.
(iii) In particular we infer that de Rham's (n - k)-cohomology group of smooth forms with compact support is isomorphic to the k-homology group of currents with compact supports {T E Dk(U) I OT = 0, sptT CC U}/{ 8S I S E Dk+1(U)}.
From now on we shall refer to the map P : Hd k (U, R) - Hk (U, ]R) as to the Poincare duality isomorphism and to (n - k)-forms PT satisfying (1) as to Poincare's dual forms of T. Poincare's dual forms PT, T E Zk(U) are constructed mollifying T, compare Sec. 5.1.2, in particular there are not unique. However they preserve some of the properties of T, e.g., PT E L' (U, do-kin) if
M(T) < oo, sptPT CC U if sptT CC U. Also (ii) of Proposition 1 states that the cohomology class of any Poincare dual form PT E Zn-k (U) of a given cycle T E Zk(U) is uniquely defined by the homology class of T. Let X be a compact oriented and boundaryless submanifold in U of dimension k. A form w E Zn-k(U) with compact support in U such that Q X (7l) =
f7lAw U
for all closed forms ?] E Zk (U) is classically referred as to a Poincare's dual form of X. In principle therefore our notion of Poincare dual form is slightly stronger,
as we require equality (1). Actually this is not the case, in fact one can show that every classical Poincare form is actually a Poincare form in our sense. This amounts to show that for every (n - k)-form w for which (4)
R' Lw(a) = fwAa = 0
V a E Dk(U), da = 0
U
(in particular w is closed) we can find a current S E Dk+1 (U) such that 1R L w = 8S. This is equivalent to say that the linear functional S' : Bk+1(U) --+ Ifs (5)
S'(,(3) :=
J
was
if 3 =
doe
extends as a linear continuous functional on Dk+1(U). We have in fact
Proposition 2. Let U be a smooth bounded open set in Rn and let w be a form LP(U,An-kJn) n £n-k(U), p > 1, such that in
3.2 Poincare and de Rham Dualities
fwAa=0
577
b a E Dk(U) with da = 0,
U
in particular dw = 0. Then there exists S E Dk+1 (U) such that 11 n L W = 3S. Proof. First assume p > 1, and let w = >2 wa (x)dxa. We solve the problem
dQa = wa O6
.Qa E Wo'p(U).
By the elliptic regularity we find a unique .fla which actually belongs to Coo (U) n W2,P(U). Consequently the form (1 := >2 Qa dxa solves
d8ffl + Sdfl = w,
compare (7) in Sec. 5.2.4, 8 being the adjoint of d with respect to zero values on W. Since dw = 0, we infer SdD E ker d n ker 5 and by the assumption, taking into account Proposition 2 in Sec. 5.2.4, Sdfl = w - dSS? E ker J', hence 8d.fl = 0. Therefore we find w = dij, rl E 6,_k-1(U) n W1,P(U, An-k-lRn), and consequently ][8n LW = IRn L drl = (-1)n-ka(Rn L 77).
If p = 1, we can still solve (6), compare Stampacchia [609], and find a weak solution which belongs to C°°(U) n W1,9 (U) for all q < n/ (n - 1). The proof is then concluded as previously.
Integration along the fibers and the lift of currents. Let X be a C2 compact, oriented n-dimensional manifold without boundary in IEPn+N and denote by X : X -* 11nRn+N and by N : X -> ANIRn+N respectively the unit tangent n-vector field to X and a unit normal vector field to X so that
X/vV has the same orientation of Rn+N. Recall now that X is a tubular neighbourhood retract, and more precisely we can find an Eo > 0 such that each point in
U = {x E Rn+N
I
dist (x, X) < Eo }
has a unique point of least distance ir(x) in X, and that -7r : U -+ X is a retraction. Moreover, we have a system of charts cpa : fla -> Va C X of X, and on each chart a system of N independent vectors via), ... , VNa) in such that via)n...nvN) =N on K,; consequently the maps
Ua x B(0, Eo) C Ua x RN -+ 7r- 1(V,) given by N
(7)
Y a (x, t) _ 0a (x) + E tiv(a) (x) i=1
578
5. Cartesian Currents in Riemannian Manifolds
are diffeomorphisms. Of course, taking into account the special form of 0a and denoting by El, ... , EN the standard basis of RN, we see that ANDca (x, t) (Eln ... AEN) = v1A ... AvN (co(x)) = N(W(x))
for x E .(la, t E B(0, Eo). Also note that 1 ° ga (x), A(x)t),
1(x, b) =
where A(x) E MN x N and det A > 0, or in other words 7r : U --> X is an oriented fiber bundle over X. To any (k+N)-form w E Dk+N(U) we now associate a k-form 7r*w in Dk(X) by "integrating w along the fibers of the fibration it : U --> X". We define 7rw by (8)
<7r*w, v > :=
r
J 7r-'(X)
<w(z), vAN(x) > dH'(z) V v E AkTXX.
Clearly 7r* defines a map from Dk+N (U) into Dk (X). In the special case in which
X = .(l is an open set of R'", U = .(2 x RN introducing coordinates el, . .(2 and El, ... , EN in R", we have
X = e1A... hen
N = E1A... AEN,
and any form w E Dk+N l/ f7 x RN) has the form
E
w«Q(x, y) dxandyQ.
aI+j8I=k+N
Then 7r (x, y) = x and (9)
7c*(w)(x) := >
(fWa(X,S)dS)dXa.
Notice that the formula for a product of currents gives (10)
for T E Dk(fl), W E Dk+N(Q X RN)
T x RN(w) = T(7r*(w))
Proposition 3. We have
11
(i) 7<* d = d7r* . (ii) 7r*(7r#77Aw) = 77A7r*w, d w E Dk+N(U), V 77 E DP(X), p + k < n. (iii) For 77 E D,-k (X) and w E Dk+N (U)
f x
7)^7r*w =
f 7#77nw, U
i.e., 7r# (IRn+N L w) = (-1) N(n-k) Q X L -7r*w.
. .
, e,,, in
3.2 Poincare and de Rham Dualities
579
If Px is a Poincare dual form of X in U then 7r.Px = 1.
(iv)
Proof. The first two claims are local, consequently it suffices to prove them for forms 77 and w respectively with compact support in a chart V and in 7r-1(V), we omit here the index a, compare (7). Also we first prove the claims in the case that V =: S2 is an open set of Rn so that -7r-1(.f2) = S2 x RN. In this case we can write 7r(x, y) = x and
(fWa(X)d) dxa.
7r.w := F, Ial=k
Now each component of w takes the form = a(x, y) dxandyO. Therefore
d = ax, (x, y)dxindxandya + v(i, 0) (-1)ka, (x, y)dxandy,3+a hence
0
(i,
7r.dt; =
Q)(_1)k
( l
\
faye (x, y) dy) dxa = 0
(fax(x)d) dxiAdxa
1/31
2.=1
and we conclude n
7r.dw = E T, (
f
axi (x, y) dy) dxindxa = d7r.w.
i=1 dal=k
Writing 7r#77 = E ,7.y (x) dxy the second claim follows easily as 17I=n
7r* (7r# ('y) Aw) (x)
E U 77-y (x)aao (x, y) dy I dx'yAdxa
171=h
Ial=k 77(x)A7r.w(x).
The general case follows from the flat case because (11)
7r.(O#w) = V#(-7r*w),
and cp being the diffeomorphisms in (7) (index omitted). In fact co#[d7r.w] = dco#7r*w = (by (11)) d7r.O#w
= (by (i) in the flat case) 7r.do#w
= 7r*0#(dw) = (again by (11)) cp#(7r.d(.1)
580
5. Cartesian Currents in R.iemannian Manifolds
which proves that in any coordinate chart d7r* - 7rd = 0. Similarly one proves (9).
Let us prove (11). We have ca#w(x, t) = Ak+NDC,(x, t)w(q5(x, t)), therefore for v E AkRn, < 0#W, vAe1A ... ^EN > = <W (O(x, t) ), Ak+NDO(x, t) (vAe1 A ... /BEN) >.
But on the account of the particular form of 0 AkDc(x, t) (v) = AkDcp(x)(v)
ANDO(x, t)Eln... AEN = I ANDO(x, t) I N(((x, t))
... Err) are orthogonal, hence
AkDO(v), ANDO(ElA
< O#W, vAe1 A ... AEN > = < w (q5(x, t) ), AkD p(x) (v) NV (cp(x, t)) >,
consequently <7r* (0#w), v >
r
_ / <w(o(x,t)),AkDV(x)v/
J
<w(z), AkDW(x)vAN(V(x)) > d7-lN (z)
7-1(rv(x))
_ <7r*w(cp(x)),AkDco(x)v> =. Proof of (iii). We use the general coarea formula
f gIVNfI dH =f dnn f g(z)dxN(z) x
x
7r-1(x)
where f : U --+ X and VN f (x) = projection of V f onto the level line of through x. In this case we have for r) E D,-k(X)
f
71A7r*(W) =
f
f (#Aw) = x
x
x
dH f d7-lr'<7r#7]Aw, ex/vV> 7r-1(x)
= f <7r#7)nw, e1A... /den+N>IVN7rI dz = f 7r#7?Aw U
U
since ON7rI = 1. This proves the claim together with
QXj L7r*(w) = (_I)("-k)k
f
r)n7r*(w)
x
_
(-1)(n-k)k f#()Aw
_
(-1)(n-k)k7r#(IRn+N L W).
=
U
(-1)(n-k)[k+N-k]
¶
Ull(WA7r#(7/))
f
3.2 Poincare and de Rham Dualities
581
Finally (iv) follows at once as for 77 E Dn(X) by (iii) and (1) we have
f
fi7ir*(Px) = f 7r#(r1)APx = x U X
771
hence 7r* (Px) = 1.
Fig. 5.3. The lift of a current.
Remark 2. Let cp :.Q -+ L be a diffeomorphism of two open sets of 1R
and let
0: (1xRN -,A x R N be of the type h : 1? x RN __ RN smooth and ah (x, t) > 0.
(12) 0(x, t) = (cp(x), h(x, t)),
Then it is easily seen that cp#7r* simply by changing variables. In particular we obtain that if 7r : E --* X is an oriented fiber bundle, then the integration along the fibers is well defined and gives a map 7r*
: