CC BY
12Probl. Anal. Issues Anal. Vol. 6(24), No. 2, 2017, pp. 57-80
DOI: 10.15393/j3.art.2017.3850
57
UDC 517.98, 514.745.4
Ya. A. Kopylov
ORLICZ SPACES OF DIFFERENTIAL FORMS ON RIEMANNIAN MANIFOLDS: DUALITY AND COHOMOLOGY
Abstract. We consider Orlicz spaces of differential forms on a Riemannian manifold. A Riesz-type theorem about the func-tionals on Orlicz spaces of forms is proved and other duality theorems are obtained therefrom. We also extend the results on the Holder-Poincare duality for reduced Lq,p-cohomology by Gol'dshtein and Troyanov to -cohomology, where and
are N-functions of class A2 R V2.
Key words: Riemannian manifold,, differential form, exterior differential, Orlicz space, Orlicz cohomology
2010 Mathematical Subject Classification: 58A12, 46E30
Introduction. This article is devoted to the study of the dual spaces of Orlicz spaces of differential forms on an oriented Riemannian manifold X.
Lp-theory of differential forms on Riemannian manifolds has been the subject of many papers and several books since the beginning of the 1980s. In 1976, Atiyah defined L2-cohomology for a Riemannian manifold and initiated various applications of L2-methods to the study of noncom-pact manifolds and quotient spaces of Riemannian manifolds by discrete groups of isometries. The L2-cohomology of such manifolds was studied by Gromov, Cheeger-Gromov and others (see, for example, [2, 3, 12]). In the 1980's, Goldshtein, Kuz'minov, and Shvedov defined the Lp-de Rham complex on a Riemannian manifold M for arbitrary p G [1, to] and began to investigate its cohomology, which they called the Lp-cohomology of M; they obtained many results concerning the density of smooth forms in Lp (see, for example, [5]); the nontriviality and the Hausdorff property of Lp-cohomology on important classes of manifolds (see, for instance, [7, 8, 17]),
©Petrozavodsk State University, 2017
[MglHl
duality for Lp-related spaces of differential forms and the induced duality for Lp-cohomology in [6]; compactly-supported approximation of Lp-forms (see, for example, [16]). In studying the asymptotic invariants of infinite groups and manifolds with pinched negative curvature, Gromov and Pansu also considered Lp-differential forms and lp-simplicial cochains (see [12, 18, 19]). Gol'dstein and Troyanov obtained deep results about the Lqp-cohomology of Riemannian manifolds for q = p in [9, 10, 11].
Like Orlicz function spaces, the Orlicz spaces L^ of differential forms are a natural nonlinear generalization of the spaces Lp. Orlicz spaces of differential forms on domains in Rn were first considered by Iwaniec and Martin in [13] and then by Agarwal, Ding, and Nolder in [1] (see also [4, 14]). In [13], Iwaniec and Martin established a Riesz-type theorem for an Orlicz space of differential forms on a domain in Rn. Orlicz spaces of differential forms on a Riemannian manifold were apparently first examined by Panenko and the author in [15], where de Rham regularization operators were introduced and studied for Orlicz spaces of differential forms.
We prove a Riesz-type theorem for Orlicz spaces of differential forms on a Riemannian manifold and then, using it, describe the dual spaces of Orlicz-Sobolev-type spaces of differential forms, thus generalizing the results of Gol'dshtein, Kuz'minov, and Shvedov obtained in [6] for Lp-related spaces. The so-obtained results are applied for establishing the Holder-Poincare duality for the reduced Orlicz cohomology of X, which extends the Holder-Poincare duality for Lq p-cohomology proved by Gol'dshtein and Troyanov in [11].
The structure of the article is as follows: In Section 1, we recall the main notions and necessary properties of Orlicz function spaces. In Section 2, we give the definition of Orlicz spaces of differential forms on a Rie-mannian manifold. The Riesz-type theorem for Orlicz spaces of differential forms (Theorem 3.1) is the contents of Section 3. Then, in Section 4, we examine the structure of the dual spaces to some -related spaces of differential forms. Finally, in Section 5, we establish a theorem on the Poincare duality for the L$I -cohomology of an oriented Riemannian manifold (Theorem 5.8).
1. N-functions and Orlicz function spaces. Definition 1.1.
A function $ : R ^ R is called an N-function if
(i) $ is even and convex;
(ii) $(x) = 0 ^^ x = 0;
(iii) lim *X£l = 0; lim ^Xe! = to.
x^Q x x^-tt x
An N-function $ has left and right derivatives (which can differ only on an at most countable set, see, for instance, [20, Theorem 1, p. 7]). The left derivative p of $ is left continuous, nondecreasing on (0, to), and such that 0 < y>(t) < to for t > 0, ^(0) = 0, lim y>(t) = to. The function
= inf{t > 0 : p(t) > s}, s > 0,
is called the left inverse of
The functions $, ^ given by
|x| |x|
$w = / ,rn, *(x) = /mdt
are called complementary N-functions.
The N-function ^ complementary to an N-function $ can also be expressed as
^(y) = sup{x|y| - $(x) : x > 0}, y E R.
N-functions are classified in accordance with their growth rates as follows:
Definition 1.2. An N-function $ is said to satisfy the A2-condition for large x (for small x, for all x), which is written as $ E A2(to) ($ E A2 (0), or $ E A2), if there exist constants xQ > 0, K > 2 such that $(2x) < < K$(x) for x > xQ (for 0 < x < xQ, or for all x > 0); and it satisfies the V2-condition for large x (for small x, or for all x), which is denoted symbolically as $ E V2(to) ($ E V2 (0), or $ eV2) if there are constants xQ > 0 and c > 1 such that $(x) < 2c$(cx) for x > xQ (for 0 < x < xQ, or for all x > 0).
Henceforth, let $ be an N-function and let (fi, E,p.) be a measure space.
Definition 1.3. The set L* = L* (fi) = L* (fi, E,^) is defined to be the set of measurable functions f : fi ^ R such that
P*(f) := j $(f № < to. o
Definition 1.4. The linear space
L* = L* (ft) = L* (ft, E,p) =
= {/ : ft ^ R measurable : p*(a/) < to for some a > 0}
is called an Orlicz space on (ft, E,p).
The corresponding Morse-Transue space is the space
M* = M*(ft) = M*(ft, E, p) =
= {/ : ft ^ R measurable : p*(a/) < to for all a > 0}.
For an Orlicz space L* = L* (ft, E,p), the N-function $ is called A2 -regular if $ G A2(to) when p(ft) < to or $ G A2 when p(ft) = to or $ G A2 (0) for ^ the counting measure on countable ft. Let ^ be the complementary N-function to $.
Below we as usual identify two functions equal outside a set of measure zero.
If / G L* then the functional || ■ ||* (called the Orlicz norm) defined
by
$ = llf IIl* (Q) = suP
J
q
: < i
is a seminorm. It becomes a norm if ß satisfies the finite subset property (see [20, p. 59]): if A G E and ß(A) > 0 then there exists B G E, B C A, such that 0 < ß(B) < to.
The equivalent gauge (or Luxemburg) norm of a function / G is defined by the formula
II/Il(*) = II/IIl(*)(q) = inf jk> 0 : k) < •
This is a norm without any constraint on the measure ß (see [20, p. 54, Theorem 3]).
We will need the following familiar assertion (see [20, item (ii), p. 57]): Lemma 1.5. Let
0 < /l < /2 < ■ ■ ■ < /m < • • •
be an increasing sequence of nonnegative measurable functions in the Or-licz space L*(fi) ((fi, E,p) is a measure space) and let fm ^ f a.e. Then ||fm||(*) < llf ||(*) < to. 2. Orlicz spaces of differential forms. Let X be a Riemannian manifold of dimension n. Given x E X, denote by (u(x),6(x)) the scalar product of exterior k-forms u(x) and 6(x) on TxX. This gives a function x ^ (u(x),0(x)) on X.
Let $ : R ^ R and ^ : R ^ R be two complementary N-functions. Denote by L*(X, Ak) the class of all measurable k-forms u such that
p*(u) := J $(|u(x)|)d^x < to. x
Here d^x stands for the volume element of the Riemannian manifold X. We will identify k-forms differing on a set of measure zero.
Given a (not necessarily orientable) Riemannian manifold X, introduce the space L*(X, Ak) as the class of all measurable k-forms u satisfying the condition
p* (au) < to for some a > 0.
The corresponding Morse-Transue space M* (X, Ak) is defined as the class of all measurable k-forms u such that
p*(au) < to for all a > 0.
Obviously, L* (X, Ak) C L* (X, Ak).
As in the case of Orlicz function spaces, the space L*(X, Ak) is endowed with two equivalent norms: the gauge norm
M|(*) =f K> 0 : p^K) ^
and the Orlicz norm
= sup
(u(x), 0(x)) dßx
X
0 g L*(X, Ak), p*(d) < 1
As in the case of function spaces, it can be proved that L*(X, Ak) endowed with one of these norms is a Banach space.
Obviously, the gauge norm of a k-form u is nothing but the gauge norm of its modulus function |u|. The same holds for the Orlicz norm
([15, Lemma 2.1]). Moreover, similarly to the case of Orlicz function spaces ([20, Proposition 10, p. 81]), we have
Lemma 2.1. The Orlicz and gauge norms of a k-form u G L$ (X, Ak) can be calculated by the formulas
u||$ = S^ := sup
eeM *(x,Ak ) ll^lw <1
X
and
|u||($) = XL := sup
eeM *(x,Ak ),
1
X
Proof. For 0 G M*(X, Ak) with ||0||w < 1 we have
(u(x), 0(x))dpX
X
< J |w(x)||0(x)|dpx <
X
< sup
geM *(X ), llsHc®) <1
|u(x)|g(x)dpx
X
= |||u|
.
The last equality here holds by [20, Proposition 10, p. 81]. Thus,
S^ = sup
eeM * (x,Ak ), lWlc®) <i
(u(x), 0(x))
X
<||M
$.
On the other hand, let (gm)meN be a sequence of functions in M*(X) with ||gm||(^) < 1 such that
|w(x)|gm(x)dpx
X
— || |u| ||$ as m —y to.
Since
|u(x)|gm(x)dpx
< J |u(x)||gm(x)|dpx < || |u| ||$,
X
we also have
|w(x)||gm(x)|d^x ^ || ||$ as m ^ to.
x
Consider the sequence (#m)meN of k-forms 0m defined by
|gm(x)| ^Jf if u(x) = 0,
0m (x) =
Then ||0m||(*) = ||gm|| < 1 and
otherwise.
(u(x),0m(x))dßx
x
|^(x)||gm (x)|d^X
x
as m ^ to. Therefore,
||$ < sup
eeM *(x,Afe),
(u(x), 0(x))dßx
K«) ^
<1
x
= |M|$.
Thus, we get the desired equality for the Orlicz norm.
For the gauge norm, the equality ||u||(*) = |||u||(*) is obvious, and one must only prove that
TL = yM^^
which is done in the same manner as for the Orlicz norm with the use of [20, Proposition 10, p. 81]. □
Below, when this does not lead to confusion, we use the abbreviations L* = (L*, ||-||*), L(*! = (L*, |.|(*));
M* = (M*, || ■ ||*), M(*! = (M*, || ■ ||(*)).
3. The Riesz theorem. Let X be an oriented n-dimensional Rie-mannian manifold.
For a k-form u on X, let *u be the Hodge dual of u (an (n — k)-form).
The bilinear function
(w,0) = J w a 0 (1)
defines a pairing
between L*(X, Ak) and L(^(X, Ak) (and between L(*)(X, Ak) and L*(X, Ak)). The integral on the right-hand side of (1) exists because
w A 0 = (-1)kn-k(w, *0)dpx, |(w, *0)x |<|w|x |* 0|x = |w|x |0|x . Hence, we obtain two versions of the Holder inequality:
|(w,0)|<||w||* ||0N(^) (2)
and
|(W,0)I<|M|(*) ||0|^. (3)
Assign to each form 0 G L(*)(X, An-k) the functional
F0 (w) = ^ w A 0. (4)
X
By (2) and (3), we have
F(w)| < ||wN*N0N(^); F(w)| < ||w||(*)||0|k. (5)
Theorem 3.1. If $ is an N-function then the correspondence 0 ^ yields isometric isomorphisms
L«(X, An-k) 4 (M*(X, Ak))'; L*(X, An-k) 4 (M(*) (X, Ak))'.
Proof. Let us prove the first isomorphism.
By (5), ||F^|| < ||0|(^). Show that an arbitrary continuous functional F G (M*(X, Ak)) is representable uniquely in the form (4). Let h : V ^ Rn, V C X be a local chart of X and let U be an open set with compact closure clX U C V; then U is endowed with two metrics: the metric p of the Riemannian manifold X and the metric p induced by h from the standard metric on Rn. It is not hard to see that the L*-spaces (M*-spaces) of k-forms on U L* (U, Ak ,p) and L(*) (U, Ak ,p)
(M*(U, Ak, p) and M(*)(U, Ak, p)) corresponding to these metrics coincide and have equivalent norms. Making use of the Riesz theorem on the general form of a linear functional on the function space M*, we, involving the coordinate representation of differential forms, conclude that every functional / G (M*(U, Ak, p)) is uniquely representable in the form
/(a) = y a A 0f, 0f G LW(U, An-k,p).
X
By the equivalence of the norms in M*(U, Ak, p) and M*(U, Ak, pi), the same holds for functionals in M*(U, Ak,p). Therefore, for FG (M*(X, Ak))' and an open set U with compact closure, there is a unique form 0U G G LW(U, An-k) such that
F(w) ^ y w A 0u for every w G M* (U, Ak).
U
Given two sets Ui and U2 as above, the forms 0Ux and 0U2 coincide on U1 fl U2 by the uniqueness of 0UinU2. Thus, all forms 0U defined for different U agree with each other and thus define an (n — k)-form 0 on X. The form 0 belongs to L(*) (X, An-k) locally, satisfies the condition
F(w) = y w A 0 for all w G M* (X, Ak) with compact support,
X
and is defined by this condition uniquely.
Consider a compact set Y C X. Let g G M* (X) be a function with compact support contained in Y having ||g|* < 1. Let be the k-form on X defined by the formula
^ (x) ,'(—1)k(n-k) |I(X| (*0(x)) if X G Y and 0(x) = 0; g 0 otherwise.
We have
F (^g ) = y ^g A0 = (— 1)k(n-k) y ^ (*0(x))A0(x) = J g(x)|0(x)|d^X .
Y Y Y
Since ||g||$ < 1, this gives
g(x)|9(x)|d^x
Y
= IF (ßg )|<||F |
Hence, using Lemma 2.1, we obtain
H^Iy ||(*) = |||% |||(*) = sup
geM*(Y); ||gH®<1
g(x)I6(x)Idßx
Y
< IIFII.
Let Yi C Y2 C ■ ■ ■ C Ym C ■ ■ ■ C X be an exhaustion of X by compact sets and let 9m be the restriction of 9 to Ym. Put fm = |9m|. Then the sequence {fm}meN satisfies the conditions of Lemma 1.5. Since 11fmH(^) < ||FH, the function lim fm = |9| lies in L(*)(X), and so 9 G
G L(*)(X, An-k) and
m—^^o
||9|(^) = lim ||9m||w < HF|
m—^
(6)
The functionals F and Fq coincide on the set of forms in M* (X, Ak) having compact support, which is, as in the case of Orlicz function spaces, dense in M*(X, Ak). Thus,
F(u) = u A 0
for all u E M*(X, Ak). Combining (2) and (6), we infer that ||FQ|| = = ||0||W.
Let us now establish the second isomorphism
L* (X, An-k) 4 (M(*! (X, Ak))'.
Let F E (M(*! (X, Ak)). Then, as above, we see that there exists a unique (n — k)-form 0 belonging to L* locally that satisfies the condition
F(u) = J u A 0 for all u E M(*!(X, Ak) with compact support. x
Using Lemma 2.1, we verify in the same manner as for || ■ ||* that, given any compact set Y C X,
IY H* <
F.
Because of the inequalities
|| ■ < || ■ ^ < 2| ■ ||(^),
we have
||0|y||(*) < ||F||.
Taking an exhaustion Y1 C Y2 C ■ ■ ■ C Ym C ■ ■ ■ C X of X by compact sets, we as above conclude that 0 G L*.
Now, the functionals F and F# coincide on the dense set of forms with compact support in M(*)(X, Ak) and hence on M(*)(X, Ak). By Lemma 2.1,
|F || = ||Ffl || = sup ll^lm <i
wA0
X
*.
The theorem is completely proved. □
4. The dual spaces to L*-related spaces of differential forms.
Throughout this section, X is an oriented smooth Riemannian manifold of dimension n and ($1, and ($2, ) are pairs of conjugate N-functions. Introduce some spaces of differential forms.
For A G {L, M} and ($*) G {$*, ($*)}, denote by Ak*lM*2)(X) the space A*1 (X, Ak) © A*2 (X, Ak+1) with the norm
||(a,^)||(*1 ),(*2) = ||a||<*i) + ||^|(*2) . Given (a, ft) G Mf*^^)(X) and (w,0) G L^-^(X), where
T^) = if ($i) = $i,
( ') if ($i) = ($i),
we put
((a,ft), (w,0)) = (—1)k (a, 0) + (ft,w). (7)
Theorem 3.1 implies that the pairing (7) defines an isometric isomorphism
(Mf*1 ),<*2 )(X ))' = (X ).
Moreover,
|((a,ft), (w,0))| < ||(a,ft)11<*i),<*2) ■ |(w,0)|.
A differential (k + 1)-form 0 E L11oc (X, Ak+1) on X is called the weak exterior differential (or derivative) of a k-form u E L1oc(X, Ak) (which is written as du = 0) if,
J 0 A u = (— 1)k+iy u A du xx
for any u E Dn-k-1 (X), where D1 (X) is the set of smooth l-forms on X with compact support included in Int X.
Let $1 and $2 be N-functions. For 0 < k < n, put
fik*lM*2)(X) = { u E L<*^(X, Ak) : du E L<*2)(X, Ak+1)} .
This is a Banach space with the norm
||u||<*1 ),<*2 ) = ||u||<*1) + |du|{*2) .
From now on we assume that $1, $2 E A2 n V2, and hence also , ^2 E A2 n V2.
If $ E A2 n V2 then, as is well known, the spaces L* and M* coincide and hence, by Theorem 3.1, the space L* is reflexive. Thus, there is no need in the spaces M*,+. We will often assume that the space fik*) <*2) (X)
is embedded in Lk*) <*2) (X) by identifying a form a E fik*) <*2) (X) with
the pair (a,da) E Lk*1),<*2)(X).
Given a subspace H C Lk* *), denote by H^ the annihilator of H
in (X) with respect to the pairing (7). Since this pairing satisfies
((a,P), (u,0)> = (—1)k(n"k-1)((u,0), (a,P)>,
there is no difference between the pairings between Lk*) <*) (X) and
L^-k"-^(X) and between L^-k-^(X) and Lk** N ,* N(X1).
<*2),<*1 y ' <*2 <*1 )'<*2)V '
The definition of fik*) <* )(X) implies that
fik*1),<*2)(X) = (Dn_k_1(X))±.
put fik*1),<*2),o(X) = (finík;(*0 (X))^. Since Dn-k-1 X C we have fik*1 ) , <*2 ),0 (X) C fik*1),<*2) (X).
Observe that if Ok* \ ,* N 0(X) = 0k \ /* \ (X) then (X) =
<*1),<*2) ,0V > <*1) , <*21 <*2 ) ,<*1 ) ,0V '
= (X)-
Lemma 4.1. The following hold for G A2 f V2:
(1) Smooth forms constitute a dense set in ) <*2 )(X)•
(2) Smooth forms with compact support constitute a dense set in 0k*1) , <*2 ) ,0 (X).
Proof. Item (1) stems from the only theorem of [15] about the properties of the de Rham regularization operators in Orlicz spaces of differential forms. Prove (2). Denote the closure of Dk(X) in L<*1),<*2) (X) by Dk(X). Then, by [21, Theorem 4.7],
•>k\_ f ok f V\\ _ ok
Dk (X) = ((Dk= (x )) = ).
□
Lemma 4.2. If , $2 G A2 nV2 and a form u G ) (^) (X) has compact support then u G ) ($2) o(X).
Proof. Suppose that u G ) ($2)(X) has compact support. Assume first that 9 is a smooth (n — k — 1)-form. By Lemma 4.1, there exists a sequence {uj } of smooth forms with compact support such that Uj ^ u in norm as j ^ to. Then
((u, du), (9, du)) = lim ((u,, duj), (9, d9)) =
= lim I [(—1)kuj A d9 + duj A 9] = lim d(u A 9) = 0. (8)
j^^ J j^^
X
The last equality in (8) is due to the Stokes theorem. Now, let 9 be an arbitrary form in O^^L.(X). By Lemma 4.1, there is a sequence
{9j} of smooth forms converging to 9 in norm as j ^ to. Then
((w,dw), (0, dw)) = lim ((w,dw), (0j,d0j)) =0.
Thus, 0 G 0k*1),<*2),0(X). □
Each pair of forms (w,0) G L^-(X) defines by (7) a continuous linear functional on Lk*) <* )(X) and hence on ) <*2) (X) and
fik*i) <*2) o(X). On the last two spaces, this functional is defined by the formula
F(a) = J[(—1)ka A 0 + da A u]. (9)
x
Theorem 4.3. If $1, $2 E A2 n V2 and are the correspon-
ding complementary functions then any continuous linear functional on fik*i) <*) (X) (on fik*) <*2) 0(X)) can be represented in the form (9). A pair of forms (u,0) defines the zero functional on fik*) <*)(X) (on fik*1),<*2),o(X)) if and only if u E fi^-y--*),0(X) and 0 = du (u E
E fi^—(X) and 0 = du). The norm of the functional (9) on
<*2), <*1) V ' J V 7
fik*1),<*2) (x ) (on fik*1),<*2),o (x )) has the form
||F|| = inf {||0 + dP||^ + ||u + PH^ : P E fin-y^.D^o
( HFH =inf {|0 + ^fe + ||u + Pfe : P E fin-k-*o(X)}) .
Proof. In accordance with [21, Theorem 4.9], if H is a closed subspace in a Banach space Y then Y'/H^ = H', where the isomorphism is induced by the canonical pairing between Y and Y'. Therefore,
(fik*1),<*2) (x ))'=Ln-k-c*!) (X) / (fik*1 ),<*2 )(X ))x =
=(X ,o(x);
vfik*1),<*2),o(X)) = Ln-7,<*7)(X) / (fik*1)'<*2),o(X0 =
=L
--4-1 d (x Vfi™(X)
<*2), <*1) / <*2),<*1) □
Theorem 4.4. If $1, $2 E A2 n V2 and are their complementary
N-functions then the dual of the space fik*) <*2 )(X) is isomorphic to the
completion of Dn-k (X) with respect to the norm
||u|| =inf {||u + d0||<*-y + ||0N(*2y : 0 e Dn-k-1(X)} . (10)
This isomorphism is given by the action
(a,w) = (-l)fcy a Л w. (11)
X
Proof. Consider the embedding j : L<*1 )(X, An-k) ^ L^-4-^.(X) defined by j (w) = (0,w). Let
n: Ln-k-4o(X) ^ LT-y-fc(X V°T-T-fc,o(X)
be the canonical projection. It is not hard to see that n o j is a monomor-phism. Since the set S = {(w,0) : w G Dn-k-1(X),0 G Dn-k(X)} is
dense in Ln-k-1 , (X), n(S) is dense in L^-^-(X) /o^-^- (X). Let w G Dn-k-1(X), 0 G Dn-k(X). Since (w,dw) G O^-^- Q(X), we have n(w, 0) = n(0, 0 — dw) = n o j(0 — dw). Hence, the set n o j(Dn-k) is dense in L^-^.(XWo^zk-^ (X). Moreover,
<*2 ),<*1 P V <*2),<*1 ),0V ! '
||n o j(w)|| . , / . =
II A ^n^-fe-1 (XWnmkri_ (X)
\ v > t\\ /\T/. \ nv ^
<Ф2 <Ф2 )><ф1 )'0
inf {||w + +ne||w : в e ,0(X^.
By Lemma 4.1(2), the set Dn-k-1(X) is dense in O^-^q(X). Hence,
||n o j(w)|| . , / . =
11 JV JULn-kr1_(XWnmkri_ (X)
<«2 >,<«1>V 7 <«2 >,<«1 >,0V 7
= inf {||w + d0||<^ + ||0N<*2y : 0 G Dn"k"1
Thus, the space L^-^-^(X) /(X) is isomorphic to the completion of Dn-k (X) with respect to the norm (10). Now, in view of [21, Theorem 4.9], if H is a closed subspace in a Banach space Y then (Y/H)' = = Hwhere the isomorphism is induced by the canonical pairing between Y and Y'. Thus, (l^-i (XWo^-* (X))' =
' V <*2 ),<*1)V V <*2 ),<*1 ),0V >)
= (O^-k-^. (X))± = Ok* v ,* v(X), and the first claim of the theo-
V <*2 ),<*1),0V ') <*1), <*2) ^
rem is established.
Further, since
((a, da), (0, u)> = (-1)^ a A u,
x
the form a E fik*) <*2)(X) acts at the forms n o j(u), u E Dn-k(X), by the formula
(a, n o j(u)> = (-1)k J a A u.
x
The theorem is proved. □
5. Holder—Poincare duality for L* */7-cohomology. Let X be
an oriented Riemannian manifold of dimension n.
Given N-functions $/ and $//, consider the spaces
Zk*„) (X) = {u E L<*») (X, Ak) : du = 0};
Bk*i),<*„)(X) = {u E L<*») (X, Ak) :
u = dp for some P E L<*) (X, Ak-1)}.
_k 7
Denote by B<*7 ),<*„ )(X) the closure of Bk* ),<*//)(X) in L<*") (X, Ak). The quotient spaces
Hf*i),<*„)(X) := z{W)(X)/bk*i),<*„)(X)
and
Hk*i),<*„)(X) := Z{k*/7)(X)/Bk*i),<*n)(X)
are called the kth L<*7),<*/7) -cohomology and the kth reduced L<*7),<*/7) -cohomology of the Riemannian manifold X, the latter cohomology being a Banach space.
If $/ = $// = $ then we use the notations fik*)(X), H{{*)(X), and
h k*) (x ) instead of fik*),<*)(x), Hf*),<*) (x ), and h {*),<*)(x ) respectively. Thus, the L<*)-cohomology Hk*)(X) (respectively, the reduced _k
L<*)-cohomology H<*)(X)) is the kth cohomology (respectively, the kth reduced cohomology) of the cochain complex {fi**)(X),d}.
The kth interior reduced L<*),<$/7)-cohomology of a Riemannian manifold X is the Banach space
),<*„),o(X) = ),o(X) jdDk-1 (X) ,
where dDk-1 (X) is the closure of dDk (X) in ) (X, Ak) and
),<*„),o(X) = Ke^d : ^^) ^ ftf+1 ^ JnZ*(X>■<*" >.
Thus, a k-form 6 belongs to ) ) 0 (X) if and only if 6 G
G )(X, Ak), d6 = 0, and there is a sequence is a weakly closed forms 6^ G (X) such that
l|6j — 6||<$7) ^ 0 and ||d6j||{*7) ^ 0 as j ^ to.
The quotient (semi)norm on each of the above-introduced cohomolo-gy spaces depends on the choice of the norm on and but the resulting topology does not.
From now on, we assume all N-functions under consideration to belong to A2 n V2.
In [11], Gol'dshtein and Troyanov realized the kth Lq,p-cohomology as the kth cohomology of some Banach complex. Here we apply this approach to L<*),<$/7)-cohomology.
Fix an (n + 1)-tuple of N-functions F = |$0, ..., and put
0F(X) = ,$fe+x(X); ^kF)(X) = ^(W),($fe+i)(X).
Use the unified notation fi(F)(X) for OF (X) and fi|F-)(X). Since the weak exterior differential is a bounded operator d : ^(f) (X) ^ ^(i)1 (X), we obtain a Banach complex
0 ^ ^0f) (X) ^ ^1F)(X) ^ ■ ■ ■ ^ 0(F) (X) ^ ■ ■ ■ ^ ^(!F) (X) ^ 0.
The L<f) -cohomology H(F) (X) (respectively, the reduced L<f) -cohomology _k
H<f) (X)) of X is the kth cohomology (respectively, the kth reduced cohomology) of the Banach complex (fi*F),d).
_k
The above-defined cohomology spaces H{F)(X) and H f (X) in fact depend only on $k-1 and :
H<F)(X) = Hf$fe_i),{$fe)(X) = )(X) /Bi$fe_i),{$fe);
H<F) (X) = H<$fe-i),{$fc) (X) = )(X) /),{$fc) .
Denote by ftkF) 0 (X) the closure of Dk (X) in ftkF)(X). The interior reduced L<f)-cohomology of X is the reduced cohomology of the Banach complex
0 ^ ^0F),O(X) ^ ^1F),0(X) ^ ■ ■ ■ ^ ft<F),0(X) ^ ■ ■ ■ ^ ^F),O(X) ^ 0;
_k _k . /-(X Afc)
H<F),0(X)= H<*fe),{*fe+i),0(X) = Zf*fe),{*fe+i),o(X)/ dD<-1 (X). ^ j
The dual of an (n + 1)-tuple of N-functions F = |$0, ..., is the (n + 1)-tuple F' = |^0, ,..., where and $n-k are complementary N-functions for all k. Henceforth, we assume all N-functions to belong to the class A2 HV2.
Fix an (n + 1)-tuple of N-functions F = |$0, $1,..., and let F' = |^0, ,..., } be its dual (n + 1)-tuple. For -1 < k < n, introduce the vector spaces
P(F) (X) = ),{*„+i) (X) = ) (X, Ak) © L^+^X, Ak+1)
(here ) (X, Ak) = 0 for k = -1,n + 1). If (a, ft) G P{F)(X) with a G ) (X, Ak) and ft G L^+0 (X, Ak+1) then P{F)(X) is endowed with the norm
ll(a,ft)l|p(f> (X) = ^«^{^fe) + llftll{$fe + i). Let dp : P<kF)(X) ^ P<k+)1 (X) be defined as
dp (a, ft) = (ft, 0).
The so-obtained Banach complex ^P*f)(X),dpj has trivial cohomology.
Lemma 5.1. Let F = |$0, ,..., } be an (n+1)-tuple of N-functions and let F' = , ..., } be its dual (n + 1)-tuple. Then the spaces
PkF) (X) and ) (here, as above, the bar changes the type of the
norm) are dual with respect to the pairing
((a,P), (u,0)) = J ((-1)ka A u + P A 0) . (12)
Lemma 5.1 easily follows from Theorem 4.3. Lemma 5.2. The operators
d : PiF)(X) ^ PL(X) and d : P^-1^) ^ P^ (X) are adjoint.
Proof. If (a,P) G Pk-1 (X) and (u,0) G Pn-k-1 (X) then (d(a, P), (u, 0)) = ((P, 0), (u, 0)) = j(-1)kP A
X
((a, P), d(u, 0)) = ((a, P), (0,0)) = / P A 0. □
X
Put
(X) = {(u,dU) G P(F") (X) : U G )} ;
£<f),o(X) = {(u,du) G P(F)(X) : u G fifo^X)} .
Clearly, these spaces form Banach complexes S{f) (X) and £{f),0(X) which are isomorphic to ^{f)(X) and fi{F),0(X) respectively. Introduce the following quotient complex of P^F)(X):
AFy (X ) = P{F) (X ) ,o(x ).
What was said above implies: Proposition 5.3. The graded vector space A^^-(X) possesses the following properties:
(1) A^p-(X) is a Banach space with respect to the norm
y(u, 0)||a = inf {||u + p||<^y + ||0 + } .
(2) AL(X) is dual to E^-*"1^) with respect to the pairing (12).
(3) The differential dp : pL(X) ^ pM1 (X) induces a differential
dA : A^(X) ^ Ak+1 (X) and (Af—"(X),d^) is a Banach com-
{F ) {F ) {F )
plex.
(4) The operators dA : A<-1 (X) ^ A^-(X) and dE : Sn"k-1 (X) ^
{F/) {F) {F)
^ )< (X) are adjoint up to sign with respect to the pairing (12).
Examine the cohomology of the Banach complex (A^^-(X)(X),dA).
{F )
If we put
Zk (A^(X)) = Ker dA : A^(X) ^ A^(X)
and
Bk (a^(X)) = Im dA (A^Fry(X))
and denote by B< (a^-(X)) the closure of Bk (a^(X)) then the cohomology and the reduced cohomology of A^^-(X) are the spaces
{FT)
H< (aF)(X)) = Z< (aFT(X)) /B< (aFT(X)) ; H (af)(X)) = Z< (ATFt)(X)) /B (aFt)(X)) .
We will need the following assertion [11, Lemma 3.1]:
Lemma 5.4. Let I : Y0 x Y1 ^ R be a duality between two reflexive Banach spaces. Let B0, B1, A0, A1 be linear subspaces such that
Bo C Ao = B^ c Yo ; Bi c Ai = BO1 C Yi.
Then the pairing I: (A0/B0) x (A1/B1) ^ R (with the bars standing for closures) is well-defined and induces duality between A0/B0 and A1/B1.
Lemma 5.5. The pairing (12) induces a pairing between the reduced cohomologies of A^^-(X) and S*F)(X).
Proof. We have
Bk"1(AfFTy(X)) C Zk-1(AyF7y(X)) = (Bn-k(EJF)(X)))X C A^X)),
and, similarly,
Im d£-k-1 C Ker d£-k = (Im d^-2)X C E^-) (X),
where the equalities are due to the fact that d^ and d^ are adjoint operators. It remains to apply Lemma 5.4 with X0 = A-^-1 and X1 =
= EnF)k (X). □
Lemma 5.6. The reduced cohomology of the Banach complex
,0
a shift:
(A^^- (X),dA) is isomorphic to the interior cohomology of X up to
H ^r (X) = H k-1 (A^- (X)
{F ')(X )= H
ism is in
j (P) = (0,P).
The isomorphism is induced by the mapping j : Z^— (X) ^ P-k-1 (X),
{F ),0 {F )
Proof. Every element in A^^-(X) is represented by an element (a, P) G
G P-k-^!(X) modulo Ek-1 (X); thus, (a, P) and (a1,p1) represent one
{F ) {F ) ,0
element in A^(X) if and only if a — a1 = u and P — P1 = du, where
- e Ek-r,0»
Further, (a,P) G P-^1(X) represents an element of Zk-1 (a^-(X)) whenever dP(a,P) = (P, 0) G E^ q(X), that is, P G Z-^ q(X). Thus,
Z(-1 (af)(X)) = {(a,P) Gp{-)(X) : P G Z{Fy,0(X)} /E(-),0(X).
Similarly, (a, P) represents an element in Bk-1 ^A^^-(Xif there is (7,^) G P-^X) with (a, P) = dA(7,^) = (5, 0) modulo E^(X), which means that P = du G B^^ Q(X). Thus,
(Af)(X0^(a,P) GP|—-(X) : P G BF^^/E(F1),0(X)
and
Bk-1 (A^(X)) = {(a,P) GP^X) : P G £^,0(X)} /e^X)
Therefore,
k i ( \ {("'в) GP^(X): в G Z*о
H k-1 (A^- (X)) = 4-^-^-}
v <F> ; {(а,в) GP^(X) : в g В^0(Х)}
'в) G Р<-У(X) : в G Z<Fy,o(X)} • R Rk
{(0,$ GP<Fy(X) : в G B
<F '>,0
Thus, the embedding j : Z-^- (X) — P-k-!(X), j(ft) = (0,ft), induces
{F ),0 {F )
an algebraic isomorphism j : H-^^ Q(X) -— Hk-1 ^A^^-(XWe also have the relation
'в) GP-k-l(X) : в G Zk
Hk-1 (a^-(X)) = ^-.
V <F> 7 |(0'в) GP^X) : в G £kF>,o(X)}
The quotient on the right-hand side is endowed with the natural quotient norm and j induces an isometric isomorphism j* : Ну^ту,0(X) -=
4 H"-1 Итт (X )\-D ,
Thus, we have
Theorem 5.7. Let X be a smooth n-dimensional oriented Riemannian manifold and let F = (Ф0, Ф1,... Фп) and F' = (Ф0, Ф1,..., Фп) be dual sequences of N-functions with Ф^ G Д2 П V2. Then the Banach spaces H(X) and Hn-Ty0(X) are dual with respect to the pairing (ш,в) = = / ш Л в for ш G ZjL (X) and в G Z^ (X).
This gives the following duality theorem for £ф/,ф//-cohomology: Theorem 5.8. Let X be an oriented n-dimensional Riemannian manifold. If Ф/, Ф/j are N-functions belonging to Д2 П V2 and Ф/ and Фц
_k
are their respective complementary N-functions then Hф/,ф// (X) is isomorphic to the dual of H^-),(Ф7),0(X) and Hкф/),(ф//) (X) is isomorphic
_у
to the dual of Hф//,ф/,0(X). The dualities are given by the pairing
([ш], [в]) = |ш Л в.
X
Proof. The theorem results from Theorem 5.7 by considering any sequence of N-functions ($0,..., with $(-1 = and = $// and its dual sequence. □
Acknowledgment. The author is indebted to the anonymous referees whose helpful comments and suggestions have improved the exposition of the paper.
References
[1] Agarwal R. P., Ding S., and Nolder C. A. Inequalities for Differential Forms. Springer, 2009. DOI: 10.1007/978-0-387-68417-8
[2] Cheeger J., Gromov M. Bounds on the Von Neumann dimension and the Gauss-Bonnet formula for open manifolds. J. Diff. Geom., 1985, vol. 21, no. 1, pp. 1-34. DOI: 10.1016/0040-9383(86)90039-X
[3] Cheeger J., Gromov M. L2-cohomology and group cohomology. Topology, 1986, vol. 25, no. 1, pp. 189-215. DOI: 10.1016/0040-9383(86)90039-X
[4] Ding S., Xing Y. Imbedding theorems in Orlicz-Sobolev space of differential forms. Nonlinear Anal., 2014, vol. 96, pp. 87-95. DOI: 10.1016/j.na.2013.11.005
[5] Gol'dshtein V. M., Kuz'minov V. I., and Shvedov I. A. A property of the de Rham regularization operator. Siberian Math. J., 1984, vol. 25, no. 2, pp. 251-257. DOI: 10.1007/BF00971462
[6] Gol'dshtein V. M., Kuz'minov V. I., and Shvedov I. A. Dual spaces of spaces of differential forms. Siberian Math. J., 1986, vol. 27, no. 1, pp. 3544. DOI: 10.1007/BF00969340
[7] Gol'dshtein V. M., Kuz'minov V. I., and Shvedov I. A. Reduced Lp-cohomology of warped cylinders. Siberian Math. J., 1990, vol. 31, no. 5, pp. 716-727. DOI: 10.1007/BF00974484
[8] Gol'dshtein V. M., Kuz'minov V. I., and Shvedov I. A. Lp-cohomology of warped cylinders. Siberian Math. J., 1990, vol. 31, no. 6, pp. 919-925. DOI: 10.1007/BF00970057
[9] Gol'dshtein V., Troyanov M. Sobolev inequalities for differential forms and Lq,p-cohomology. J. Geom. Anal., 2006, vol. 16, no. 4, pp. 597-632. DOI: 10.1007/BF02922133
[10] Gol'dshtein V., Troyanov M. Lq,p-cohomology of Riemannian manifolds with negative curvature. In: International Mathematical Series, Sobolev Spaces in Mathematics II. Springer New York, 2009, pp. 199-208. DOI: 10.1007/978-0-387-85650-6_9
[11] Gol'dshtein V., Troyanov M. The Holder-Poincare duality for Lq,p-cohomology, Ann. Global Anal. Geom., 2012, vol. 41, no. 1, pp. 25-45. DOI: 10.1007/s10455-011-9269-x
[12] Gromov M. Asymptotic invariants of infinite groups. Geometric Group Theory, vol. 2. Cambridge Univ. Press, 1993.
[13] Iwaniec T., Martin G. Geometric Function Theory and Nonlinear Analysis. Oxford University Press, 2001.
[14] Johnson C., Ding S. Integral estimates for the potential operator on differential forms. Int. J. Anal., 2013, Article ID 108623, 6 p. DOI: 10.1155/2013/108623
[15] Kopylov Ya. A., Panenko R. A. De Rham regularization operators in Orlicz spaces of differential forms on Riemannian manifolds. Sib. Elektron. Mat. Izv., 2015, vol. 12, pp. 361-371. DOI: 10.17377/semi.2015.12.030
[16] Kuz'minov V. I., Shvedov I. A. On compactly-supported approximation of closed differential forms on Riemannian manifolds of special type. Siberian. Math. J., 1993, vol. 34, no. 3, pp. 486-499. DOI: 10.1007/BF00971224
[17] Kuz'minov V. I., Shvedov I. A. On the operator of exterior derivation on the Riem,a,nnia,n manifolds with cylindrical ends. Siberian. Math. J., 2007, vol. 48, no. 3, pp. 500-507. DOI: 10.1007/s11202-007-0052-y
[18] Pansu P. Cohomologie Lp des variétés a courbure negative, cas du degré 1. Rend. Mat. Univ. Pol. Torino, 1989, Fascicolo Speciale "PDE and Geometry", pp. 95-119.
[19] Pansu P. Cohomologie Lp, espaces homogènes et 'pincement. Preprint, Orsay, 1999.
[20] Rao M. M., Ren Z. D. Theory of Orlicz Spaces. Pure and Applied Mathematics, vol. 146. Marcel Dekker, 1991.
[21] Rudin W. Functional Analysis. McGraw-Hill Book Comp., 1973.
Received May 12, 2017.
In revised form, October 6, 2017.
Accepted August 30, 2017.
Published online October 6, 2017.
Sobolev Institute of Mathematics
4, Akad. Koptyuga st., Novosibirsk 630090, Russia;
Novosibirsk State University
2, Pirogova st., Novosibirsk 630090, Russia
E-mail: yakop@math.nsc.ru
