q
d
For these values, the statement of the theorem followsfrom Theorem 1.18.6/2 (see also Theorem 1.18.7/2) and (12). For q = 1, one has to replace Theorem 1.18.7/2 by Theorem 1.18.7/3. S t e p 4. X ’ denotes the dual operator to X . Similarly, K’(x)is the dual operator t o K ( x ) .If A is a reflexive Banach space and if 1 < G < 00, then 1 1 [L,(A)]’= L,+4’) where - + - = 1 , (24) (T (TI in the sense of the usual interpretation, that means that the general linear continuous functional over L,(A) can be represented in the form
J <W), W )dz,
h ( 4 € L.Y(A),44 € L,*(A’).
(25)
Rn
( * , I ) is the linear continuous functional over A . See for instance R. E. EDWARDS [l], Theorem 8.20.5. Using the third step, it is not hard to see that X ’ belongs to
L(L,,(A;),LJA;)) where
_1- - _ 1 0’
s’
- 1
- -1 , q
r’
s
(TI
< q’
(p’
5 s’ < 00).
(26)
It holds l l K ’ ( z ) l l ~ ; +=~ ;l l K ( ~ ) 1 1 A , . - . A , . If we use the usual properties of vectorvalued integrals, it follows for f E L,,(A,) and g E L,(A;)
I
H”
( f ( 4 7
.m(4)dx = J
(Wf)(4,B(4) dx
RII
= Rn
=
J
J
Rn
( K ( z - Y) f(Y), 9(+ dY dx Rn
(f(y),
1 K’(s -- Y) 9(4 dx) dY -
Rn
Now, by the preliminary remarks, one obtains that
2.2.3. Singular Integral Operators
157
Setting ( i h ) ( 2 ) = h ( - x ) , we find that &?'I? is a n operator of type (10). It holds (9) with K' instead of K and L( A ; ,Ah) instead of L(A,, A,). Further, one has to replace (12) by IlX'f(x)II&4A&
5
CllfIIt,4Ah
.
But then one obtains from the third step that X ' belongs to L(L,,(A;),L,,(A,$), if _1 - _1 -- 1 --,1 (q < 5' 5 p ' ) . u' 5' q Together with (26), this proves that X ' is an element of L(L,,(A;),L,,(AA)),if 1 < 0' < q'
(q
< 5' < 0 O ) .
Step 5 . The theorem is now a consequence of X" = X . R e m a r k 2. * For values s with 1 < s 5 p , the theorem coincides essentially with [l]. The corresponding first three Theorem 2 of the paper written by J. T. SCHWARTZ steps of the proof are more or less a repetition of the proof given by J. T. SCHWARTZ. See also N. DUNFORD, J. T. SCHWARTZ [l, 111, chapter 11.11. The assumption that the Banach spaces A , and A , are reflexive is not necessary for this part of the theorem. The theorem in the formulation given above and the last two steps of the proof go back to H. TRIEBEL[19], Theorem 3.2. The theorem generalizes a corresponding result in L. HORMANDER [2], Theorem 2.1, for the scalar case.
2.2.3.
Singular Integral Operators
Later on, we shall need mapping properties of singular integral operators. The methods developed in the last subsection are sufficient to prove the needed results. [2]. We shall follow the treatment given by L. HORMANDER The Banach spaces A , and A , of Theorem 2.2.2 are identified with the complex plane C . Let L, = Lp(C).The unit sphere in R, (and also a point on it) are denoted by w, , let dw, be the corresponding surface element. T h e o r e m 1. Let K ( x ) be a complex-valued locally integrable function with
K ( x ) = 0 if 1x1 < 1 and K ( t x ) = t-"K(x) if t 2 1 and 1 . 1 2 1. (1) Further it is assumed that (2.2.2/9)holds with q = 1 and A , = A , = C . Then the operator Z from Theorem 2.2.2 can be extended to an operator belonging to L(L,, L,) for all p with 1 < p < 00 if and only if
J' K(o,) d o ,
= 0.
(2)
% I
If this condition fails t h n %.? cannot be extendd to an operator belonging to L(L,,,L,)
for any p with 1 < p < 00. Proof. Step 1 . Using (l),it is easy t o see, that K ( x ) is a regular distribution belonging t o S'(R,). We wish t o show that the Fourier transform FK belongs to L ,
158
2.2. Integral Operatore and Fourier Multipliers
if and only if (2) holds. Since K ( z ) is locally integrable, we have, by (2.2.2/9), for
lYl IB-l,
K ( z - y)
- K ( z )EL1.
Consequently,
F ( K ( z - y)
- K ( z ) )= (e-i
is a continuous and uniformely bounded function with respect to y. Hence it fol-P 00. The lows that ( F K )(5) is continuous in R,,- (01, and bounded if function t"K(ts) - K ( z ) with t > 1 is zero if lzl < t-l and if 121 > 1. $inm it holdsF(t"K(tz)) (t)= F ( K ( x ) ) ( + ) , one obtains that
( B K ) ( E ~- ~( )F K ) ( ~=)
j"
e-i+>WK(tz)dz
j"
e-'(tr'*l/>K(y)dy
t-a<
=
1 2 1< 1
1 < 191 < I 1
j j
=
1
It follows that
e-i(el-'~y)lyl-nK(on)lyln-1 do,,dlyl.
(3)
'Dn
If FK belongs to La,then the left-hand side of (4) is uniformely bounded with respect to t. Considering t 00, one obtains (2). Conversely, we w u m e that (2) is valid. It follows that # ( ~ ~ ) ( 5 t - 1 )( F K ) ( ~=) (e-i(~~t-1~) - 1) ~ ( w , d)o,- 4 Y l
j' J
-
Hence for
1
= 1 it follows that
I(FW(Et-')
lYl
*
'D*
- (FK)(t)Is c j"
IK(on)Id o n *
mn
+
Now, one obtains that (FK) (5) is bounded for 5 0. Hence, either (FK) ([) is a. bounded function or the s u m of a bounded function and a singular distribution whose support is the origin. But such a singular distribution is the finite linear combination of the &distribution and its derivatives. To exclude the second possibility, it is sufficient to show that (FK)(q6)-P 0 if = D(R,,).It holds that
(FK)(v6)= K ( F q 6 ) = Since
one obtains that
1 5
j
E
-P
0, where q6(s)= q
K ( z )(Fq)(Ex) E"
21
an
and q(z) E C$'(R,,)
169
2.2.3. Singular Integral Operatore
But this is the desired assertion. Hence FK E L, Step 2. By (2.2.114)for q E Cg(R,,)we have n
(Xq)(x) = P'(2n)T FK
*
.
Pq.
(5)
If (2) is satisfied, then FK EL,. Since F is a unitary mapping from L, onto L,, one obtaim by the last formula that X (after extension) belongs to L(Ls,Ls). But now, it follows fromTheorem 2.2.2 that X (after extension) is an element of L(Lp,Lp), 1 < p < 00. If (2) is not satisfied, then FK is not an element of L, . By (5)and the continuity of FK in R,, - (0}, it follows that X cannot belong to L(L,, L,). Together with Theorem 2.2.2, this yields X # L(Lp,Lp), 1 < p < 00, Theorem 2. Let k(x) be cz mp&x-ualued function, differentiable in R,, {0},
-
where
2 E Rn
k(tz) = t-"k(s) if
Waf) (4=
j
lz-yl b l
and
0 < t < 00,
k(x - Y ) f ( Y )dY
+
(6)
I w - Y ) (f(9) - f ( 4 ) dY7
I.z-ar1.z 1
f E G'(Rn)
(10)
7
then X s ,after extension by continuity, belongs to L(L,, Lp),1 < p < 00 ,0 Further, for f E Lp we huve
X8f +Zof if L?
~10.
(11)
Proof. Step 1.We want to apply Theorem 1 to ifl. (2.2.219)is satisfied with q = 1, B = 2 and
If tls - yI 2 1 and tlsl 2 1, then on a suitable path, it follows that
5 E < CO.
For this purpose, we ahow that
lzl 2 2 and lyl 5 3. By (8) and the integration
If for instance tlzl < 1 and tJs- yI 2 1 with
151
2 2 and Iyi
3, then
160
2.2. Integral Operators and Fourier Multipliers
Together with (6))whence it follows (2.2.219) with q cable and it holds t h a t
Il.X,fllLp 5
=
1. Hence, Theorem 1 is appli(12)
CpllfllLp.
Step 2. Replacing x in (9) by E X and y by EY, one obtains by (6) that ( S e j ) (EX)
= J" k ( x Iz-yI 21
- Y) ~ ( E Y )dy-
Now, (12) yields
Il.XefllLp S c p l l f l l ~ ~ independently of (%of
E.
(13)
Using (7), one obtains for 0 < E < 1 that
-x
e f )
(XI
=
J
1.i-gI g e
k ( x - Y) (f(Y) - f ( 4 )dY,
f
E Com(R,l).
The modulus of the integral can be estimated by C E . Since the integral vanishes for large values of 1x1, one obtains (11) for f E C$(Bn).But now, (13) holds for E = 0 , too, and for f E C?(R,l).Hence, one obtains .XoE L(Lp,L,) (after extension) and the validity of (11) for all f E L,. R e m a r k 1. With the aid of the developed methods, one can also consider other integral operators. If we set
k(x) =
IxI-O,
0 < (x
c n,
n it is easy t o see that (2.2.219) holds with B = 2 and q = - . On this way, one can OL
obtain a new proof of Theorem 1.18.913. We do not go into detail and refer t o L. HORMANDER [2]. Further references are given in Remark 1.18.9/4. R e m a r k 2. * The investigation of singular integrals in one dimension goes back to D. &BERT and H. PO IN CAR^. I n the twenties and in the thirties, F. G. TRICOMI, G. GIRAUD,and S. G. IVIICHLIN have carried over the results t o the case of several dimensions. References corresponding t o these results and also ahort descriptions can be found in the first section of S. G. MICHLIN [3]. This book contains also a comprehemive treatment of the work of S. G. MICHLINon the theory of singular integrals and singular integral equations. A new period in the theory of singular integrals in L,-spaces began in 1952 by the paper written by A. P. C A L D E R ~ N , A. ZYGMUND [l]. Theorem 2 agrees with a corresponding result found by A. P. CALDER6N, A. ZYGMUND[I], where (8)may be replaced by essentially weaker conditions. For instance, Theorem 2 remains valid, if k(x) is locally integrable, and it holds (6)) (7), and I k ( ~ n) k(-wn)I max (0, log I k ( W n ) k(-on)~) don < 03
1
+
-
%I
We mentioned above that we essentially followed the treatments given by J. T. SCHWARTZ [l] and L. HORMANDER [2]. (See also N. DUNFORD, J. T. SCHWARTZ [l,111, 11.7 and 11.11.) A. P. C A L D E R h , A. ZYGMIJND [2] extended their considerations t o singular integrals whose kernels have the form
161
2.2.4. Multiplier Theorem
Here, for every point x E R, , a formula analogous to (7) holds. Further in this context we refer to the paper by M. COTLAR[l]. Comprehensive treatments of this theory can be found in E. M. STEIN,G. WEISS[ 5 ] ,E. M. STEIN[5], U. NERI[l],and A. ZYGMUND [4]. Further references to the papers written by A. P. CALDER6N, A. ZYGMUND are given by S. G. MICHLIN[3]. We quote also the surveys by A. P. CALDER~N [6], J. PEETRE[13], and R. S. STRICHARTZ [3]. Singularintegrals in Lp-spaceswith weights E 21, B. MUCKENHOUPT, are considered by E. M. STEIN[l] and later by P. K R ~ [l, R. L. WHEEDEN[l], and T. WALSH[l,21. Anisotropic singular integrals (partly also in spaces with weights) are treated by E. B. FABES,N. M. R ~ R [l], E 0. V. BEsov, V. P. IL'IN, P. I. LIZORKIN[l], C. SADOSKY [l], Ju. S. NIKOL'SKIJ [l], N. M. R I V ~ R[l]. E
2.2.4.
Multiplier Theorem
Let =
{
E IE
=
(tjjjm3-m
9
t j
complex,
IIEII~, =
(. c m
I=-m
1 1tjlr)T
< a) 9
where 1 5 r < a.For 1 p 5 00, the spaces Lp(Zr) have the same meaning as in Subsection 1.18.4, where now X = R,, and ,u denotes the Lebesgue measure. If
K ( s ) = (Kk,j(x))-m c k , j c m K k , j ( x ) E L?(Rn) is a matrix with complex coefficients, then the operator X , 9
( S f(4 ) = is considered, where
s'(&),
I K ( x - Y )f ( Y )dY,
(1)
(2)
4
f = (fj(x)}$-m f j ( x ) 0 if ljl h N, f j ( x )E COm(Rn). (3) Theorem. Let K ( x ) be the matrix defined above. Let ( F K k , j )(5) be a regular distri9
bution for all k and j , having classical derivatives in R,
- (0) u p to order
[t] +
l.*)
Further it i s assumed that there exists a positive number B such that for all R > 0 and for
all multi-indices LY with
(a) Then X (after extension) belongs for all p with 1 < p < 00 to L(Lp(lZ),Lp(12)), and it holds IlXll 5 BB, where B depends only on p and n. ( b )If additionally Kk,j(X) E 0 if k j , then .f(after extension) belongs for all p and all r with 1 < p < 03 and 1 < r < 03 to L(Lp(Zr), Lp(lr)),and it holds IlXll BB, where B depends only o n p , r , and n.
+
*) As usual,
[t]
n
is the largest integer smaller than or equal to - .
11 Triebel, Interpolation
2
162
2.2. Integral Operators and Fourier Multipliers
Proof. Step 1. We begin with preliminary conaiderations. Let y(x) 2 0 be a func1
tion belonging to C$(R,) with y ( z )> 0 if -I 1x1 5 in the set {z 13 < 1x1 < 2}. If we set 1/2 ~m
[:I
\
fiwhose support,is contained
-1
tion. For x = 1 + - Parseval's formula yields
1 (1 +
2S(X12)"
R,
cOD
k,j=
IS'k,j(S)l2dx m
5 cB22".
n
Since x > - one obtains by the triangle inequality for integrals 2
It
From (8) we find that
= c'B.
Now, one obtains by (lo),(ll),and the construction of (FK,c,j)'(x)
c
a,
k,j= -a
I(FKk,j)' (x)I2
5 CB2.
*) Numbers depending only on n,p and r are denoted by c and c'.
(10)
163
2.2.4. Multiplier Theorem
Step 2. After these preliminary conaideratiom, we introduce the matrix N
=
Q[j(z)
C gZr,j(z), an(,)= ( Q c j ( z ) ) - m i=-n
< k , j < co
,
(13)
and the operator Y N ,
( g N f($1 ) =j
- Y) f(Y)dY
R"
(14)
2
defined for elements of type (3). With (lo), it follows atj@) E L ~ ( R ,Further, ). by (lo), C N ( z )belongs to L(l,, I,) for almost all z and llGN(z)llt2+~, E Ll(Rn).In this step, we want to show that the aammptions of Theorem 2.2.2 with A, = A, = I , , q = 1, and p = r = 2 are satisfied for the operator g N .With the aid of the remarks in 2.2.1 and Parseval's formula, it follows that m
c
IIgNflli,o) =k = - - a ,
1c m
j=-
j
II~Nfllzs(l,,S
- Y) fj(Y) dyll
h
H"
Now, uaing the definitions of giij(z)and tains from (12) ,
8
acj(z
a:&)
in (8) and (13), respectively, one ob(16)
C~211flle(ls)~
Here, c is independent of N. Hence it follows that (2.2.2/12) holds where p and q = 1. To obtain (2.2.2/9), we prove that
=
T
=
2
In the same manner as in (10) where the number 1 in the last integral is omitted, it follows that
5 t that
Hence, one obtains for IyI lZl
j
zzt
[
m
kJ=
C
-m
1
1gj&
s CB(2'l)+.
- y) - gi,j(z)12]'dz
We need a second estimate of such a type. We assume 2't glk,j(z) - g:,j(z
If we use 29 5 1, lyl 05 5 x that
I 5-
R, k,j=
11'
m
- y) = P-l[(l - e-i(Yfc))
s t, and the construction of
lO"(1
5
1. It holds that
(F&,j)' (pKk,j)',
- e-'(vee)) ( F K k , , j ) ' (t)lad [
(19)
(03. it follows from (7) for
- C~22z(n-21a1)221f2.
164
2.2. Integral Operatom and Fourier Multipliers
- cB 5
a,
C
l=-w
n
min ((2zt)B-Y, 2't)
5 c'B.
But this coincides with (17). Hence, the assumptions of Theorem 2.2.2 with A, = A, = I , , q = 1, and p = r = 2 are satisfied. It follows that
[ 1 IlsNf(s)Ilb"21"
1
5 c"L
Ilf(4ll;
dz]&
1
(22)
Rm
where c depends only on n and p (but not on N). f is of type (3). Step 3. For the operator g N ,the limit process N + 00 is considered. Since [Lp(a2)]' = L P ~ ( lin 2 )the sense of the formulas (2.2.2/24)and (2.2.2/25),formula (22) is eqmvalent to
and equivalent to
1
C 1( P K k , j ) ( f ) k,j= -w R 00
'
F f j ( E )F - l h k ( 8
Whence it follows that This proves the part (a) of the theorem.
5 cBllfI14(Za)IlhllLp/(u
2.2.4. Multiplier Theorem
165
Step 4. To prove the part (b), we consider again the operator g Ndefined in (13) and (14). Now, it holds U t j ( s ) = 0 if k j . Since
llQN(s
- y ) - QN(s)~~L(~r,= Z,)
k
la&(z
- 3) - Qsc”lk(,)l
(21) remains true after replacing L(l,, I,) by L(lr, Zr). In the same manner, it follows l l ~ ( s ) l l ~ ( t , ,E t ,L1(R,J. ) Considering (22) with p = r for f ( j ) ( s ) = (. . ., 0 , f j ( s )0, , .. .), and summing the so obtained estimates, we find that m
But now, the assumptions of Theorem 2.2.2 with A, = A, = I,, q = 1, and p = r are satisfied. Hence, it holds (22) with 1, instead of I , . Using the considerations of the third step with 1, instead of I,, one obtains now the part (b) of the theorem. Remark 1. Both the theorem and the above proof go back to H. TFLIEBEL [19], Theorem 3.5. But both the theorem and the proof are generalizations of a result [2], Theorem 2.5, for the scalar-valued case to the vectorfound in L. HORMANDER valued case. The proof shows that one can weaken the assumption (4) for the proof of part (b). Further it is easy t o see that the theorem and the proof remain valid, if D”(FKk,j)(t)are regular distributions satisfying (4). Remark 2. Giving up the representation (2) as an integral operator, one can generalize the theorem. The remarks in Subsection 2.2.1 show that (2) may be represented as m
Wf) (4 = {(en)+F-l ( j = c - FKk,jFij)} -m
Now, T = and if for all f
=
is said to be a multiplier-matrix of type ( p , p),if Tk,j ES’(R,,), with f j E X(R,) and f j = 0 for ljl > N we have
where Cis a number independent of f . By the proof of the theorem, it follows immediately that T is a multiplier-matrix of type ( p , p ) , where 1 < p < CQ, if (4) is satisfied with Tic,jinstead of FKk,,i.A similar assertioia holds for the generalization of the part (b) of the theorem. This version explains also the notation “multiplier”. R e m a r k 3. * The assumption (4) is satisfied, if there exists a positive number B such that for all multi-indices 01 with 0
5
For the scalar case, this coincides essentially with the condition originally formulated by S. G. MICHLTN[l, 2,3]. The condition (4) or (24), respectively, may be replaced for the scalar case by the following assumption: For x = ( x , , . . . , x,,), where xj + 0 ,
166
2.2. Integral Operatom and Fourier Multipliers
j = 1, . . . , n, K ( x ) has continuous partial derivatives up to the order n inclusively. Further there exists a positive number B such that for all multi-indices a = (a1,. . ., u,,) where mj is either 0 or 1, and for all x = ( x l , . . .,xn) where xj 0, j = 1, . . ., n,
+
141. . . *WPK(x)I 5 B . Then X E L(L, , L p )and IlXll I; BB) where B depends only on n and p . This theorem [2] and it is a generalization of the original formulation by is due to P. I. LIZORKIN S. G. MICHLIN.A proof of this theorem may be found also in S. M. NIKOL'SKIJ [7], p. 69. These statements remain valid also for the generalization described in Remark2 (with respect to the scalar case). R e m a r k 4. * The first multiplier theorem for trigonometrical series was proved by J. MARCINKIEWICZ [2] in 1939. Using these results, S. G. MICFXLIN [l, 2 , 3 ] obtained similar results for Fourier integrals (scalar case). In the framework of the theory of translation-invariant operators, L. HORMANDER [2] described further generalizations and simplifications. Remarks 2 shows that the statement of the theorem for the scalar case is equivalent to IIF-YFK * Ff)ll~p 6 B B l l f l l ~ ~f, € S ( R n ) *
(25)
FK is called (Fourier) multiplier of type ( p ,p ) . Generalizing this notion, one denotes by M; the set of all distributions T E S' such that for all f E S where Cisindependent of f . Here 1 < p, q c co. The elements of M! are said to be (Fourier) multipliers of type ( p , q). It holds that and
Mi=L,,
M$cL,
if
1 c p < co
1 1 1 1 1 0 Mq P
= M$
if
such that for all R > 0 and all a with
B IPT(x)l 5 12116( for all x
1011
5
E R,,
1
+
with x
El
- we have
+ 0.
This is the usual formulation of the multiplier theorems of M I C ~ I K and HORYANDER. Generalizations are given by P. I. LIZORKIN [ 2 , 5 , 7 , 8 ] , where the last two papers [l,21, are surveys. Further results can be found .in L. CATTABRIGA[l], P. KRI~E S. IGARI, S. KURATSUBO [l])and G. 0. OKIKIOLU[l].In the paper by J. T. SOHWARTZ [l], these considerations axe carried over to the vector-valued case. Generalizations of these investigations are given by W. LITTMAN, C. MCCARTHY,N. M. R ~ R [l] E
2.2.4. Multiplier Theorem
167
and N. M. RIVI~RE[l], where these papers contain rather general results. Further, we refer also to papers written by P. I. LIZORKIN [lo, 111. Multipliers in L,-spaces [l].Tceatments of with weights are considered by P. KREE[2] and A. I. KAMZOLOV aspects of the theory of multipliers can be found in E. M. STEIN[a, 51 and R. LARSEN [l]. The book written by R. LARSEN contains a comprehensive bibliography. Remark 5. * The connection between interpolation theory and the theory of multipliers was considered by J. PEETRE[14], W. LITTMAN [l], and J. LOFSTROM [l]. The aim is to weaken the assumption (4) of the theorem in dependence on p . For this purpose, one “interpolates” between ‘ ‘ M ; = L,” and “ ( 4 ) ” and contains conditions depending on p . For details, we refer to the cited papers. See also C. FEFFERMAN, E. M. STEIN[l] and R. JOHNSON [2]. Remark 6. Clearly, one can extend the definition of multipliers in the sense of (26) to p = q = 1 and p = q = 00. The set of all these multipliers is denoted by M i and M g ,respectively. Since these cases are not important for the later considerations, we do not go into detail here. But we formulate an interesting result: M i = M z is the set of all finite Borel-measures i n R,. A proof can be found in E. M. STEIN[5], p. 94/95. Now we consider a special multiplier which is not contained in the last theorem. Lemma. If Q = {x I x = ( x l , . . ., x,), Ixjl < a,}, aj > 0 , then the characteristic function xQ(x)of Q belongs to M! for all p with 1 < p < 00. Proof. Step 1. First, we consider the one-dimensional case. It holds that
If f
=
E S ( R l ) ,it
follows that
-a
1
m
sina(x - y ) f(Y) dY 2 - Y
--m
sin ax 1 cos ax -cos a y f ( y ) dy - 1irnIT 7L 2 - Y L I-?’-ylZe
e l 0
-
-1 I
Theorem 2.2.312 yields that xr-.,.l belongs to M$. Step 2. We consider the case of several dimensions. Temporarily we set (?if)
(21 3
..
* 3
Xj-1,
6 ,xj+1,. . ., x,)
1 e-izjEjf(x)d x j , a,
= (2n)-+
f
E S(R,),
--03
and extend the definition of Fj to S‘(R,) in the usual way. Similarly, we define F;l. If x,(x) denotes the characteristic function of the set {x I -aj < xj < uj), then n
XQ(X) =
Il Xj(x)J=1
Let f
E S(Rn).Uaing
=
the first step, one obtains that
II IIFy'XjFjf(x1,
* * - 9
s cllf
IILp(Rn) *
xj-1,xjs z j + l ,
-
* *>
zn)IILp(Ri)II~(R,,-i)
..
Since F-lXQFf = F-1x1FF-1x2FF-1. Ff, the desired result follows by iteration. R e m a r k 7. The question arises for which domains 52 c R, the characteristic function x&) belongs to M ; . C. FEFFERMAN [2] has shown that the chracteristic function of a ball is an element of M; ,if and only if p = 2. This result was extended [4] t o a large class of closed sets in R,. One can generalize the by B. S. &JAGIN question and ask for which values of p the function
belongs to M$ . A summary of results in this direction can be found in C. FEFFERW [3]. Sufficient conditions for multipliers in dependence on p are also given by W. TREBELS
[4].
I n this section, the spaces Bi,q(Rn)and F;,JRn) containing the spaces HJR,) and W;(Rn)as special cases are introduced. At the same time, we shall prove a number of important properties. As it will be shown later, Bi,q(Rn)are the Besov spaces, H;(R,) are the Lebesgue (Liouville, Bessel-potential) spaces, and W;(Rn)are the SobolevSlobodeckij spaces.
2.3.1.
Definitions
s
The spaces ,?;(A),where 1 5 p co, - 00 < s < 00, have the same meaning as in (1.18.2/1).If A is the complex plane, we will write Zi instead of Zi(A).As usual, the support of the distribution f is denoted by supp f . The convergence of a series in S' is marked by . F and F-l are the Fourier transform and the inverse Fourier transform in S', respectively. Further we set Mj
= {E
15 E R,,, 2j-l 5 151 I - Zit'},
Mo = {5 I 5 E Rn 151 S 2 ) * 9
j = 1,2,.
..,
(1)
2.3.1. Definitions
169
andfor -co < s < c o , l < p < co,andq= coonesets
(c) For -co < s < co and 1 < p <
H;(Rn) = Fi,dRn)(d) For 1 < p < 00 one sets
00
one sets
if s = O , 1 , 2 ,..., Wi(R,) = (Hi(Rn) (7) B;,p(R,,) if 0 < s =l=integer. R e m a r k 1. The considerations in Subsection 2.2.1 show that any distribution aj E S ( R n )with supp Faj c Mj is an analytic function. Sometimes we shall make use of a, = aj(s)E Cm(R,,). R e m a r k 2. * The spaces Wi((Rn)with s = 1 , 2 , 3 , . . . coincide with the well[l,2 , 3 , 4 ] in 1935-1938. The known Sobolev spaces introduced by s. L. SOBOLEV extension of the definition of the spaces W;(R,) to values s with 0 < s =I= integer (Slobodeckij spaces) is closely related to the investigation of boundary values of [l], L. N. SLOBOfunctions belonging to Sobolev spaces. We refer to N. ARONSZAJN DECKIJ [l], and E. GAGLIARDO [l]. As it will be shown later, the spaces with s > 0 coincide with the well-known Lebesgue (or Liouville, or Bessel-potential) K. T. SMITH [l] and A. P. CALDER6N [2]. spaces introduced by N. ARONSZAJN, The spaces B;,q(RIL) with s > 0 are the Besov spaces, see 0. V. BESOV[l, 21. The definition of the spaces Bi,q(R,)described above was given by H. TRIEBEL[19] and is closely related to corresponding definitions formulated by S. M. NIXOL'SKIJ [7]
and J. PEETRE [15]. The definition of the spaces Fi,q(R,)goes back t o H. TRIEBEL [19]. I n this connection, we refer t o the papers by P. I. LIZORKIN [12,14], and the spaces L;,q(R,) introduced there which are similar t o the spaces Fi,q(R,), and by J. PEETRE [39]. Finally, we mention the spaces introduced by S. M. NIKOL’SKIJ [2, 3,4] (and often called Nikol‘skij spaces) which coincide with the spaces Bi,m(Rn). In this connection, we remark that (in contrast to the notat,ions used here) the Nikol’skij spaces are often denoted by Hi(R,) and the Lebesgue spaces by Li(R,). Many references, particularly t o papers of Soviet mathematicians (0.V. BESOV, V. I. BURENKOV, K. K. GOLOVKIN, V. P. IL’IN, L. D. KUDRJAVCEV,P. I. LIZORKIN, S. M. NIKOL’SKIJ, S . V. USPENSKIJ, . . .), can be found in S. M. NIKOL’SKIJ[7]. Further we refer t o the survey papers by N. ARONSZAJN, P. MULLA,P. SZEPTYCKI [l], R. A. ADAMS,N. ARONSZAJN, K. T. S ~ [l], H and E. MAGENES[l]. Purther references, particularly t o the spaces B;,,,(R,)with s 5 0, are given in Remark 2.3.412. See also Remark 2.3.2/1. R e m a r k 3. By the definition of the spaces Bi,q(R,,),ff;(R,J,and W;,(R,J,we restrict ourselves to the case 1 < p < 03. As mentioned above, for s > 0 these spaces coincide with the Besov spaces, Lebesgue spaces, and Sobolev-Slobodeckij spaces, respectively. But for these types of spaces, the limit cases p = 1 and p = co are also of interest. One could try to carry over Definition 1 (or one of the later norms equivalent t o the above oneif 1 < p < co) t o these cases. We do not go into detail here. but we add some remarks : 1. The later considerations are based on Theorem 2.2.4 which is valid only for 1 < p < 03. By an extension of the considerations to p = 1 and p = co,one has t o use Remark 2.2.416 (and a generalization t o the vector-valued case). 2. By the treatment of the spaces with p = 1 and p = 00, there arise new difficulties. The results obtained for 1 < p < 03 are not always true for these limit cases. Particularly, there are several non-equivalent possibilities of definition. (See Remark 2.3.3/5 and Remark 1.13.4/2.) 3. One of the main aims of this book is the application of the theory of function spaces t o elliptic differential operators. But here again a restriction to 1 < p < 00 is meaningful. 4. The definition of the spaces BL,m(R,,)in analogy t o the norms in Theorem 2.5.1 is meaningful and gives interesting spaces. See for instance E. M. STEW[5], p. 141. The spacea BL,m(R,) are closely related to the Holder spaces Ct(R,,) and V t ( R )defined in Section 2.7. The latter spaces are also necessary for the formulation of the embedding theorems in Section 2.8. R e m a r k 4. * The main aim of this chapter is the systematic consideration of the spaces Bi,q(Rn),F&(Rn), Hi(R,,), and W;(R,) (embedding theorems, traces on the boundary, interpolation theory, equivalent norms, duality theory, structure theory, . . .). At the same time, we shall be concerned with the restriction of these spaces t o R,+= {x I x E R n ;x , ~> O } . Further, we consider some properties of Holder spaces. I n the later chapters, we deal with the restriction of the mentioned spaces t o arbitrary domains L?c R,. Furthermore, Sobolev-Besov spaces with weights are considered. The extensive theory of anisotropic spaces is not treated here, but there are some remarks in the last section of this chapter. To this part of the theory, we do not give references here, although numerous new publications have appeared since the publication of S. M. NIKOL‘SKIJ[7], where a bibliography of this subject up t o 1967-69 can be found. But we want t o mention here shortly some classes of isotropic spaces which are important but not conaidered later on. Lebesgue-Besov spaces
2.3.1. Definitions
171
defined on general structures are introduced by M. H. TAIBLESON [2] and vectorvalued Sobolev-Slobodeckij-Besov spaces (with values in Hilbert spaces and Banach [l], P. GRISVARD [4,6], N. N. FROLOV [l] and other spaces) are treated by J. WLOKA mathemat,icians. Interpolation theorems and embedding theorems can be found in P. GRISVARD [4,6]. Sobolev-Besov spaces WL and B;,q with 0 < p < 1and 0 < p < 00 are considered by J. PEETRE [32,35,36] (this includes interpolation theorems, duality theorems, and embedding theorems) and T. M. FLEW.[2] (duality theory). [36] is an extension of Definition The definition of the spaces Bi,q by J. PEETRE l(a) and Theorem 2.3.2(a) to the values 0 < p < oc) and 0 < q < a.An inter[3]) can be found in polation theory for the Hormander spaces (see L. HORMANDER M. SCHECHTER [5,6]. We mention also the Morrey spaces, John-Nirenberg spaces, and the Campanato spaces. A short summary and a description of interpolation [l] and J. PEETRE [20]. The interpolation properties are given by G. STAMPACCHIA theory of these spaces was developed by S. CAMPANATO, M. K. V. MIJRTHY,S. SPANNE, References can be found in G. STAMPACCHIA [11 and J. PEETRE and G. STAMPACCHIA. [20]. As it will be shown later, the Lebesgue spaces are closely related to (elliptic) Bessel-potentials.Similar considerationsfor Lebesgue spaces based on parabolic potentials are given by R. J. BAQBY [l]. In this paper, there is also given an interpolation theory for spaces of such a type (complex method). C. S. HERZ[l] and R. JOHNSON [l I introduced spaces defined with the aid of Riesz potentials. Finally, we mention E. M. STEIN[l]. There are the theory of the Hp-spaces treated by C. FEFFERMAN, also interpolation results for the complex method. Similar results for the real interpolation method are obtained by N. M. RIVIERE, Y. SAGHER [l] and C. FEFFERMAN, N. M. RIVIERE, Y. SAQHER [l]. See also M. ZAFRAN [l]. For the later considerations special systems of functions are of interest. Definition 2. If N i s a natural number, then @N denotes the set of all SY8bn-S of functions ( q ~ , ~ ( x with ) } ~ -the ~ following properties :
There exists a positive number c1 such that
c 00
c1
5 k=O
(FQ)k)
(8.
(9)
For aity multi-index a,there exists a poaitive number c 2 ( a ) such that
F u r t h r , one sets @ =
u GN.
N=1
Example. It is not difficult to describe examples of systems of functions of the above type. Let p(z) E X(R,) be a function with (Fv) (5) 2 0 and
SUPPFF c (5 I 5 E Rn, 2-N 5 151 5 2N),
( ~ 9(5) ) > 0 if
1
-5
12
-
-
IEI 5 1 2 .
(11)
and choosing eo in a suitable way, then ( @ k } g a E @N and it holds that
z 00
k=O
( F e d (5) = c.
(13)
We shall make use of these examples later on.
2.3.2.
The Spaces Bi,q(Rn)and Fi,q(Rn)
As usual, CF (R,) denotes the set of all complex-valued infinitely differentiable functions with compact support defined in R, . “c” means alwaysa continuousembedding. Theorem. (a) Let -00 < s < co, 1 < p < 00, and 1 q a.Then BRT;6q(Rn) is a Banach space. Further, for a n y system of functions {Q)k}& E @ we have
Bi,q(Rn)= {f I f
E s’(Rn)
; Ilf
=
Il{f * ~ k } I I $ ( L P )< a)-
(1)
Here, llfll$;,g is an equivalent mmn in the space Bi,q(Rn).If q < co, then C$’(R,) and S(R,) are dense in B&(R,). On the other hand, C$(R,) and X(R,) are not dense in
q, m(Rn).
(b) Let -CQ < s < co, 1 c p < co, and 1 < q < C Q . Then F;,q(Rn) is a Banach space. Further, for a n y system of functions {Q)k)&, E @, we have Fi,q(Rn) = { f I f Here, 11f Fi,q(Rn)*
I I f II*F.,,*= II{/ * Y ~IIL~(I:I ) < 00) -
E R(Rn),
(2)
is a n equivalent norm in the space Fiiq(Rn).C;(R,) and X(R,) are dense in
2.3.2. The Spaces Bi,q(Rn)and Fi,((R,,)
Proof. Step 1. We prove (2). Let f
E Fi,q(Rn)and
173
let
m
be a representation in the sense of Definition 2.3.1/1(b).Let { Q ) k } & ) E @ N . Then aj * p)k = F-lF(aj * Q)k) = F-'(2n)"12 FQ)$Uj = 0 , if either j < k - N - 1 or j > k + N + 1. Using the remarks in 2.2.1, whence it follows that
f * p)k F I c -0
a,
aj
* qk
c
k+N+1
=
j=k-N-1
(aj * vk) (2).
The functions ak(z)belong to Lp(Rn).Now, we approximate these functions,
CF(R,,) 3 akBE 2 a,+ if E 1 0 . (6) If we set K j , j ( z )= ~ ) j ( for z ) j = 0, 1,2, . . . and Kj,j(z) = 0 for j < 0, then, for fixed E, r, and M, Theorem 2.2.4(b) is applicable ((2.2.4124) is an immediate consequence of (2.3.1/8) and (2.3.1/10)).One obtains that instead of akfr,€.) (Here, the numbers 2sj do not give any trouble, one considers 28jak+r,E The number c is independent of E , r , and M. A limit process shows that the last ineWith quality is true for E = 0, too, and hence it holds also for akcr instead of ak+r,E. M = 00 on the right-hand side, it follows that
Il{f * Q)k}llLp(2i) 5 cll{ak}IILp(li)* Construction of the infimum yields
Il{f * v k } k ) l h ( l i ) s
(7)
CllfIIF;,,.
Step 2. We prove the conversion to (7). We suppose that the left-hand side of (7) is finite. Let e(z) E CZ(R,) be a real function such that e(z) = 1 if 2-N 5 1x1 5 2N, e(z) E CF((5 1 2 - N - 1 < 151 < 2 N + ' } ) . Let &(Z) = @(2-'Z) if k = 1,2,. . . We choose a real function po(x)such that
eo(4 Let
{Q)k}&,E @N
s
= 1 if 1 4 2N, and let K j , j ( z )= 0 for
( F K j , j () z ) = (2n)-n(
03
C
1=0
e o ( 4 E CI?({E I 151 < 2N+'}). < 0, and
Fqz)-'ej(s), j = 0 , 1 , 2 , .
Then the assumptions of Theorem 2.2.4(b) are satisfied. Let vk = Q)k
* Kk,k
n
(Fvk = (2n)TpvkFKk,k).
..
174
2.3. The Spaces Bp,q(Rn),li'i,q(Rn),H;(R,,L and W;(Rn)
Then by Theorem 2.2.4(b) (after a continuity argument analogously to the first step with respect to the functions Q)k * f E Lp)
we find that
* qk,the desired inequality is a consequence of 5 Il{ak)\ILp(l;) c\l{f * Q ) k ) l l $ ( 1 : ) *
Let N = 1. Choosing ak = f IlfIIF:,,
Now let N > 1. We consider a system n
{ ~ ~E } gfor~which (2.3.1/13) with
(9) xk
instead of ,ok and c = (2n)-T holds. If Kk,k= x k for k = 0, 1 , 2 , . . . and Kk,k= 0 for k < 0, then the assumptions of Theorem 2.2.4(b)are satisfied. Hence, it holds that and
l({f*Wk*Xk+r}IIL,(Zi) s C ( I { ' / ' k * f } l l ~ ~ ( l : ) ,
c
k+lV+l
f*yk=
l=k-N-1
r = - N - l , . - - ? N +1,
f * Y k *XI.
(Here x1 = 0 for 1 < 0.) Setting
and
SUPpUk m
t
M/<, k
c %(z) =
k=O
m
k=O
(
=
0, 1, 2 , . . .
lV+ 1
r=-N-1
(see (2.3.1/1))
f * yk * X k t r )
c (f * W
=
k=O
Yk)
7f
(8) and (10) prove (9). Now, together with the first step, one obtains (2).
-
Step 3. The proof of (1) follows the same line. One has to replace I( llLp(p)by (1 * llgLp). Further, it is sufficient to use the scalar case of Theorem 2.2.4(b)where the number c in IlYk
* ak+rllLp
Cllak+rllLp
is independent of k. The limiting cases q = 1 and q = 00 do not give any trouble in these considerations.
2.3.2.The Spaces Bi,(R,) and FiJR,)
176
Step 4. We prove (3). The third, fourth, and fifth inclusion follow immediately by the monotony of the 1,-spaces. The second (and hence also the last but one) inclusion is a consequence of II{2(s+e)jII f * vjIIj}IIlm.
I I ~ I I BI ~ ,C~I I { ~ ~ ~*I IvjIILp}IIh / 5
Here f E Bi:&(R,,).If we choose a suitable system {qj]T4 E @ (for instance such one as in Example 2.3.1), then the first inclusion for f E ,S(Bn)(and sufficiently large a) follows from 2J(s+qF--1(Ffpj * Ff)llLp
5 - C12j'8+8)(( (1
x
C
[lul82n
+ 1212)" P - l ( F f
sup (1
XER,
+
*
F y j ) I(L ,
IZ12)-o+s+e
(DdPcpj(S)l].
Let {qj}so E @ be a system described in Example 2.3.1 where (2.3.1/13) holds, with n
of e k and with c = (zn)-T. Let { y j } z 0E @ with (Fyj) (5) = 1 for 5 E supp Fqj. Then the last inclusion is a consequence of
'pk instead
and of the just proved first inclusion. Step 5. We prove (d). By the monotony of the 1,-spaces and the trivial assertion B&(Rn) = F;,p(Rl,),one obtains the right-hand aide of (4a) and the left-hand side of (4b). If {yj}j?o E @, then the left-hand side of (4a) follows from
176
2.3. The Spaces BiJR,,), Pi,p(R,,),Hi(Rm),and Wi(Rm)
6 c'lIfllF;,q. Step 6 . The last two steps show that (3) is true, too, after replacing B by F . Step 7. Using (1)and (2), we find out that Bi,q(R,,)and Fi,q(R,,)are normed spaces. To prove the completeness of BP,q(R,,),we consider two systems of functions {pj);~ E @ and { y i s o E CP. Here, {p,>sosatisfies (2.3.1/13) with pj inatead of and with c = (2n)-T, and it holds Fyj(6) = 1 for and
S(f) = {f
* pj>jmpo
9
S
It follows that
E L(Bi,q(Rn), l@p))
m
R ( { f j l )7 S
E supp Fp,.
C vj(z)* IjCz), j=o
R
(12)
E L(li(Lp), Bi,q(Rn)).
(13)
The convergence of the series (13) in S'(R,,) can be obtained analogously to the estimate (ll),where one has to take into consideration 1; c li-". Further, one has to useTheorem 2.2.4(b) in the same manner as in the first steps. Now, it is not hard to see that S is a coretraction and R is a retraction in the sense of Definition 1.2.4. By Theorem 1.2.4, it follows that S is an isomorphic mapping from B;,q(R,) onto a complemented (and hence closed) subspace of li(Lp).Whence one obtains the completeness of Bi,q(R,)and (in a similar way) of F;,q(Rn). Step 8. If 1 < q < 00 (resp. 1 6 q < a), and if the functions uj(z) have the same that meaning as in Definition 2.3.1/1, then it follows by the theorem of LEBESQUE the functions
M
C
j=O
u,(x) are dense in F;,*(Rn)(resp. in Bi,q(Rn)).Approximation of
q(z)in LJR,,) by functions belonging to C$(R,) or to S(R,) shows that both C$(R,,) and S(R,) are dense subsets in F;,,(R,,) (resp. in Bi,q(Rn)).Finally, it remains to prove that S(Rn)(and hence also C$(R,,))is not dense in B;,,,(R,,). The mapping properties of F and the first inclusion in (3) show that it is sufficient to prove that the functions p E S(Rn)with Fp, E Cg(Rn)are not dense in B;,,,(R,,). Now, let {ui},To be a sequence of functions with
supp uj c N j (see (2.3.1/1)) and 2sj~~u,~~Lp = 1. Then the consideratiom after formula (13) show that
a,
C
j=O
u,(z) belongs to
B;,m(Rn).On the other hand, it is not hard to see that it is impossible to approximate f by functions of the above type. Remark 1. Norms of the type (1)for the spaces BiJR,,) are first used by J. PEETRE [El, while (2.3.1/2a) essentially goes back to S. M. NIEOL'SKIJ [7]; see also Remark 2.3.1/2. In J. PEETRE [12], there are shortly mentioned norms of type (2). The
2.3.3. The Spaces H;(RJ
177
described proof is a modification of corresponding considerations by H. TRIEBEL r191.*) Definition. T h completion of CZ(R,) in Bi,m(Rn),-03 < s < 00, 1 < p < 03, is denoted by S;,a(Rn). R e m a r k 2. The part (a) of the theorem shows that h;,w(Rn)does not coincide with Further, it follows by (3) that S(R,,)is dense in g i , w ( R n )too. , Bi,m(Rn).
The Spaces Hi(&)
2.3.3.
The spaces Hi(R,,)have been defined in (2.3.1/6). The following theorem shows that these spaces are of special interest. Theorem. (a) Let
-00
< s < 00 and 1 < p < 00. Then 6
I f E S'(Rn),IIjIIH; = IIP-Vl + IzIa)-ji-WtILp< a}Here, IlfllHI is an equivalent norm to llfll .
Hi(%)
P
(b) I f 8 = 1,2,3,
Here,
= {f
. . .and 1 < p
<
5.2
03,
then
~ l f ~ ~ wis, an equivalent norm to IlfllH;. ?
Proof. Step 1. Let f E S'(Rn)be an element such that 9 = F-l( 1
Let
(1)
{cpj};o
E @.
+ lzI2)FFf 8
I
HP
< 03. We set
E Lp(Rn).
We shall use Theorem 2.2.4(a) with
~ j , ~ (=z~-1(28j(1 ) + Kj,k
llfll
=0
~zla)-$pqj) if j
= 0,1,2,
. . .,
otherwise,
and = {gj],Cm with go = g and gj = 0 otherwise. It follows (after a continuity argument as in the first step of the proof of Theorem 2.3.2)
IIf IIF;,, S cII{F-YFf Fqj
IIL,,(z,) = cII{P-YPKj,o * Pg)}I I ~ ( u * g}I1~p(tr)S ~ " l l ~=l~l "~I l f l l ~ ~ * . *
= c'II{Kj,o
*
2''))
(3)
Step 2. We prove the conversion to (3). Let f €F;,g(Rn)= Hi(R,,).We choose a system { ~ j } , ? E ~ @ such that
*) Recently, J. PEETRE [38] considered applications of the spaces Pin(B,,).
12 Triebel, Interpolation
(see Example 2.3.1). Further, we choose a second system (~,}i”,~ E @ such that
( P y j ) (6)= 1 if 5 E SUPP Pvj
Then for and
fi
=
2j.f
-
if j = 0, 1 , 2 , . . .,
* vj
= 0 otherwise,
fj
K , , j ( x ) = y j ( z ) if j = 0, 1 , 2 , . . ., Kk,,(z) = 0 otherwise,
the assumptions of Theorem 2.2.4(a) are satisfied. It follows (after a continuity
and
c 2”jf * 03
vj
j-0
* yj
n
=
(2n)P
c 28jF-l(Pf m
j=O
*
Ffpj)
Whence it follows the conversion t o ( 3 ) and hence the proof of (a). Step 3. If s is a natural number, then it is not hard t o see that
< p < m, in the sense of Remark 2.2.414. For
is a multiplier belonging t o M ! , 1
f
E S(Rn),it
follows that
5 c’IlF-l(1 + I z l ” Q y I L p . 8
IlO“fllr,
=
cIlF-l*..
.z.FfllLp
Now, one obtains in the usual way by completion llfll ,I cllfllHi for f WP Step 4 . To prove the converse inequality, we consider
E H;(Rn).
Further, let e ( t ) be an infinitely differentiable function on R , such that e(t) = 0 for It[ 5 3, e ( t ) = 1 for t 2 1 , e(t) 2 0 for t 2 0, and e(t) = -e(-t). Then 8
179
2.3.3. The Spaces Hi(R,)
is a multiplier belonging to M; in the sense of Remark 2.2.414. Hence, one obtains
s c’llf
+ c‘ j =c1 11 llFil@”(ti) ~ ~ F ~ f [ I L p ( R 1 ) I I L p (- R n _ , ) n
llLp(Rm)
Here, F , (resp. Pi’) denotes the one-dimensional (inverse) Fourier transform. Since $ ( t ) is a one-dimensional multiplier and aince it holds Fil@i‘lf obtains
=
c- auf
a$ ’
one (5)
Together with the third step, this yields the part (b) of the theorem. R e m a r k 1. * For s = 0 one obtains F:,2(R,L)= H:(Rn) = L,,(R,) and
IIfII L, II{f * vj}IILp,12) {pj}jao_oE @ * (6) Theorems of such a type are very important in functional analysis. They are called theorems of Paley-Littlewood type. Treatments on several aspects of theorems of Paley-Littlewood type can be found in E.M.STEIN [a] and W.LITTUN, C. MCCARTHY, N. M. RIVIERE [l]. R e m a r k 2. I n the last two steps, we proved a little more than stated in the theorem, namely N
where l l f l l , , , and l l f l l ~ , are equivalent norms. Here s = 1, 2, . . . and 1 < p < 00. P P Assertions of such a type on equivalent norms are very important. Later on, we shall prove numerous theorems of such a type. R e m a r k 3. * The last theorem describes the usual definition of the Sobolev spaces W”,R,,)and the Lebesgue spaces H i (R,,).References may be found in Remark 2.3.1/2. Later on, we shall obtain for the spaces B;JR,J equivalent norms, too. It is also possible t o find equivalent norms for the spaces Hi(Rn) if s integer with the aid of singular integrals, hypersingular integrals and differences. We refer t o [l,21, P. I. LIZORKIN[9], C. FEFFERMAN [l], E. M. STEIN [ 3 ] , R. S. STRICHARTZ and R. L. WHEEDEN[I, 2,3]. In the framework of approximation theory, one obtains also numerous equivalent norms in the spaces HE(Rn).We referto P. L. BmZER,E. GORLICH[l], E. GORLICH[l], and W. TREBELS[1, 2,3]. (See also P. L. B m ZER,W. TREBELS [l] and P. L. BUTZER, H. BERENS[l].)Similar considerations for anisotropic spaces can be found in R. J. BAGBY[2]. R e m a r k 4. * For -a < s < 00, one obtains by Theorem 2.3.2(d)
+
Bi,2(Rn)c fl;(Rr,) c Bi,p(Rn), 2 S P < a ) q.p(R,,= ) ffp,,) = q),2(&), 1 < P 5 2, 12*
(8)
and Hi(R,) = B;,p(Rn).One can replace Hi by W: in (8). Further, together with (2.3.2/3), one obtains that
HFe(R,),W F ( R n )c B;,q(Rn)c HO,-"(R,),W",-"(R,), (9) -co < s < 00, 1 < p < co, E > 0 , l 5 q 5 00. Inclusion properties of type (8) and (9) for Sobolev-Slobodeckij-Lebesgue-Besov spaces are well-known. We refer [ l ] (one-dimensional case for periodical functions); 0. V. BESOV to I. I. HIRSCHMAN [2]; A. P. CALDER~N [2]; S. V. USPENSKIJ[l]; J. L. LIONS,E. MAGENES [l,1111; M. H. TAIBLESON[ l , I], Theorem 15; P. I. LIZORKIN[ 4 ] ; and N. ARONSZAJN, F. MULLA, P. SZEPTYCKI [ l ] . An improvement of (8) is impossible in the following 00. sense. The embedding H;(R,) c Bi,q(Rn)is true if and only if max [2, p ] q Similarly, Hi(R,) 3 Bi,q(Rn)holds if and only if 1 5 q 5 min [2, p ] . We refer t o M. H. TAIBLESON [ l , I], Theorem 20; and K. K. GOLOVEIN[3].
s
R e m a r k 5. I n Remark 2.3.113, we mentioned the possibility to extend the considerations t o the limit cases p = 1 and p = co. If Hi(R,) and H",R,) are defined by (1)) and if Wi(R,) and W$(R,) are defined by (2), where s = 1 , 2 , 3 , . ., then Hi(Rn)9 Wi(R,) and HS,(R,) WL(R,) for s = 1 , 2 , 3 , . . . and n 2 2. (For 1 < p < co these spaces coincide, respectively.) For n = 1 and s = 2 , 4 , 6, . . ., we have Hi(R1) = W$(Rl)and Hd,(R1)= W$(Rl),while for n = 1 and s = 1 , 3 , 5 , . . . there is Hi(Rl) $: Wi(R1) and H",(Rl) W$(Rl). We refer t o E. M. STEIN [5], pp. 135/160. Further, it follows by Remark 1.13.412 that formula (7) for p = 1 and p = 00 is untrue, too.
.
+
+
2.3.4.
Lift Property
It is easy t o see that
+ lx12)"B'f, 8
-a < s < 00, (1) is a continuous one-to-one mapping from S(R,) onto S(R,) and from S'(R,) onto S'(R,).*) It holds Iil = I-,. T h e o r e m . Let -co < s, 0 < 00 and 1 < p < co. Then I , is a continuousone-to-one mapping from F,",,(R,,)onto F;Z(R,,), 1 < q < 00, and fTom B;,q(Rn)onto B;;:(R,), 15qsco. P r o o f . Let {qk)& E @. If
I,f = F-I(l
yk
=
* F-l
~ph
2ks
(2)
+
((1 I x , 2 d then {yk)&, belongs also to @. It follows t h a t
I,?/* yk
n
=
F-'((2n)TPyA FIJ)
= F-'((2n)" 2"Fqk
n
. Ff) = (2n)T2ksf * ~ ) k .
*) As usual, the notation "onto" means that the range of the operator coincides with the whole space.
2.4.1. Interpolation of the Spaces Bi,,(R,)
181
Now, one obtains by (2.3.2/1) and (2.3.2/2) that
-
l l M J sP d- * IlfIl$,p =
and
- lIfllBU .
IILfllga-.
PIP
PIP
R e m a r k 1. It follows immediately by (2.3.3/1) and (1) that the spaces Hi(R,) Fi,2(R,)have the lift property described in the theorem.
R e m a r k 2. * We just mentioned that the spaces B;,q(Rr,)for s > 0 coincide with the well-known Besov spaces. But on the basis of the usual definition of these spaces it is more difficult to prove the l i f t property. Conversely, the l i f t property for the spaces B;,q(R,) with s > 0 was used t o define the spaces Bi,q(R,)for s j0. Several authors discussed tJhequestion what the spaces B;,q(R,) are. In this context, we refer to [2]; N. ARONSZAJN, F. MULLA, P. SZEPTYCKI El]; M. H. TAIBLESON A. P. CALDER~N [l, I]; S. M. NIKOL'SKIJ,J. L. LIONS, P. I. LIZORKIN[l]; and S. M. NIKOL'SKIJ ["I, Theorem 8.9.1.
2.4.
Interpolation Theory for the Spaces Bi,q(Rn)and J",q(Rn)
One of the main aims of this chapter is t o give a treatment of the spaces B;,q(R,,)and F;,q(R,) (containing the spaces Hi(R,,) and Wi(R,,)as special cases) on the basis of
interpolation theory. I n this section, we shall prove several interpolation theorems which are of fundamental importance for the later considerations.
2.4.1.
Interpolation of the Spaces B&(Rn)
For a comprehensive description of the interpolation properties of the spaces B;,q(Rr,), we introduce the spaces Bi,q,(r)(R,J and B;$)(R,) generalizing the spaces given in Definition 2.3.1/l(a) and Theorem 2.3.2(a). These new spaces are not important for the later considerations. Later on, we shall be interested only in interpolation spaces of B;JRrZ)and Fi,q(Rn)which are also spaces of the type B;,q(Rn)or F;,q(Rn). The scalar (complex) Lorentz-sequence spaces introduced in Subsection 1.18.3 are denoted by lq,r,and the scalar (complex) Lorentz spaces defined in R, with respect to the Lebesgue-measure, introduced in Subsecttion 1.18.6, are denoted by Lq,r.rP has the same meaning as in Definition 2.3.1/2. D e f i n i t i o n . Let (Q?,~)?=~ E @. Further let l j r j c o . (a) For 1 5 q 5 co one sets
B;,q,(r)(Rtr)= {f I f E
fif(&t),
IIfIIgig(,) 8
.
03
< s < co, 1 < p < co, and
= II(2"'IIf
* p j IIL~,,}
111,
< a}.
(1)
(b) For 1 < q < co one sets
Bg;)(Rn) = {f I f
E S'(Rn), llfIlB.*(*) = P.P
ll{28j.llf
* ~;llLp}llzg,~ < .I.
(2)
182
2.4. Interpolation Theory for the Spaces B:,q(Bn)and F:,,(R,,)
R e m a r k 1. Clearly, it holds
B;.q.(P)(R,l) = q,q(R,J and q,Ip’(R,,) = q , q ( R , l ) . Using the previous methods (proof of Theorem 2.3.2, particularly Step,7),we realize that Bi,q,(r)(R,,) and B:$’(R,,) are Banach spaces, independent of the choice of (qk}gO E@. To prove the last assertion, we remark that Theorem 2.2.4 is also true for the Lorentz spaces Lp,ras follows immediately by interpolation on the basis of Theorem 2.2.4 and Theorem l.l8.6/2. Now, the method of the proof of Theorem 2.3.2 and the scalar case of Theorem 2.2.4 with Lp,rinstead of Lp give the desired statement’. We remind of the Definition 2.3.2 of the spaces &,,a(Rll). T h e o r e m . (a) Let -co < so, s1 < co, so and 0 < 8 < 1. Then
$1 sl,
1
(Bzq,(Rn),B2ql(RI1))@,q = Bi,q(Rn) where
(b) Let -co < s < 0 < 8 < 1. Then
s
=
(1 -
00,
1 < p < co, 1 5 qo,ql
e)so + esl ,
1
= (1
1
5 qo,ql,q 5
- 8 ) $0
+ 881.
00,
(3)
S co, qo $1 ql, 1 5 1’5 00, and
e
1-0
5
00,
-4 Po +-,q1
and
1
-P= -
1-0 PO
+ -.8
P1
(6)
If additionally p = q, then (7)
(Bz,q,(Rn),B23,q,(%))e,q= Bi,p(&).
(d) Let -co < so, s1 < co, 1 0 < 8 < 1. Then
5 qo < co, 1 S q1 5
00,
1 < p o , p l < co, and
[B;,qe(Rii))B2,q1(Rn)le= Bi,q(Rn) with (6)* (e) Let
-00
< so,sl <
co,so $: sl, 1
(8)
< p < co,and 0 < 8 < 1. Then
[ B ~ m ( R n ) , B ~ m ( ~ n=) &,,(R,J ~e
where
=
(1 - @ s o
+ 88,.
(9)
Proof. s t e p 1. We use the two systems {P)k)& E @ and {v k },& E @ introduced in Step 7 of the proof of Theorem 2.3.2, and we use the operators S and R defined in (2.3.2/12) and (2.3.2/13). We remarked there that S is a coretraction from B;,q(Rn) into l;(Lp)and R is a corresponding retraction from li(Lp)onto B;,q(Rn).The last B2,,ql(R,L)) is an interpolation couple (in the part of (2.3.2/3) shows that {B;,qo(R,l), If P is an sense of the notations of Subsect’ion 1.2.1 one can choose a2 = S’(RII)). interpolation functor, then one obtains by Theorem 1.2.4 that
2.4.1.Interpolation of t h e Spaces B;,JR,,)
183
Step 2. All the statements of the theorem are consequences of (10)by specialization. (a) Part (a) of the theorem follows from Theorem 1.18.2. (b) One obtains part (b) of the theorem from Theorem 1.18.3/2 if one takes into consideration that one can in (1.18.3/8) by (and similarly for the other spaces). (c) To prove (c), replace we use (1.18.1/3) where A,i = 2jSoLp,(Rn) and Bj = 2J"lLp,(R,,). One obtains as an easy consequence of Definition 1.3.2 and Theorem 1.18.6/2 that
-
Here, means that the numbers of estimates are independent of i. Now, it follows (c) if one takes into consideration Bi,p,(p,(Rn) = B;,p(R,,).(d) Assuming q1 < 03, then (d) follows from (1.18.1/4) similarly t o (c). Here, one has to use a formula corresponding to (ll),see (1.18.1/15) and (1.18.6/15). For q1 = co, one obtains the desired result from (1.18.1/12). (e) To prove (e), we use (1.18.1/16). We must show that CF(Rn)is dense in the interpolation space. Let { y k } E oE @ be a system such that f
5 f * y j , see Example m
=
S' i=O
2.3.1. Then
2 f * y j approximates f in the M
=
i-0
norm of the interpolation space. But such functions can be approximated by functions belonging to Cg(Rn)in the desired way. R e m a r k 2. We just mentioned that we are above all interested in interpolation results which lead to spaces of type BiJR"): (3), (7), (8), and (9). Formula (9) is of special interest, since it shows that &&(Rn) may be obtained by interpolation of R,m(Rn)R e m a r k 3. By the extension of (9) t o different values of p , there arise new difficulties. But a t least, one obtains for - co < so, s1 < 00, 1 < p o , p1 < co,and 0 < 8 < 1 with the aid of the above method by (1.18.1/13) and (1.18.1/14) that
where s
=
(1
- 0 ) so + 88, and-1 P
=
1-8 e f-. PO Pl
R e m a r k 4. I n contrast t o (3), the parameter q in (5) is determined by q o , q l , and 8. The question arises whether one is able to determine the interpolation spaces on the left-hand side of (5) if the second index of the real interpolation method does not coincide with q. We quote a result by J. PEETRE [15], Theorem 1 : If the parameters have the same meaning as in part (c) of the theorem, inclusively formula (6), and if 1 5 r 5 co,then
Bi,min(q,r),(v(k) c (Bz,q,(Rn),B;,q,(Rn))e,rc Bi,maxcq,n,(r,(Rn) * Here, the parameters min (q, r ) and max (q, r ) cannot be improved, a t least in the 12
n
PO
P1
case po < p1 and so - - 5s1--.
R e m a r k 5. * From the very beginning, the interpolation theory for function spaces and for spaces of distributions was closely related t o the development of the abstract interpolation theory. We refer t o E. GAGLIARDO [4], J. L. LIONS [3, I], and J. L. ~ O N S ,E. MAGENES [l, particularly 1111. After preliminary results in the just mentioned papers, formula (3) was proved essentially by J. L. LIONS,J. PEETRE [Z] and E. MAGENES[l]. As a powerful tool, they used strongly continuous semi-
184
2.4. Interpolation Theory for the Spaces B:,(R,,) and F;,q(R,,)
groups of operators in the sense of Section 1.13. Here, in this book, we have the converse situation. On the basis of the just proved formula (3) and the theory developed in Section 1.13, we shall obtain equivalent norms for these spaces later on. Particularly, we shall derive formulas usually used for the definition of the Besov spaces B;,q(Rn).The further development of the interpolation theory of function [3,4] spaces without weights in R,, is characterized by papers by A. P. CALDER~N (complexmethod), J. L. LIONS[lo], J. PEETRE [ l l , 151, P. GRISVARD [a], M. H. TAIBLESON [l,113, and many other papers on interpolation of function spaces (in R, or in domains, with and without weights) which we shall mention later on (see Remark 2.4.2/3).The paper by J. PEETRE [15] and the unpublished lecture notes by J. PEETRE [12] are of special importance for the methods considered here. There is the first description of the methods developed here. Further, there can be found (5)and the generalization described in Remark 4. The interesting special case (7) was proved [4] and M. H. TAIBLESON before by P. GRISVARD[4]. (8) goes back to P. GRISVARD [l,111. Before this time, J. L. LIONS,E. MAOENES[l,V] and J. L. LIONS[lo] had proved special caseg.
2.4.2.
Interpolation of the Spaces FiJZtJ
Analogously to the previous Subsection 2.4.1, we generalize the spaces F;,*(Rn)of Definition 2.3.1/l(b) and introduce the spaces F;,O(R,) and F&(n(Rn).These new spaces are not important for the later considerations. Later on, we are interested or F;JR,,). only in interpolation spaces which coincide with Bdp,q(Rn) Definition. Let ( p k } k ~Eo@. Further let -co < s < co and 1 < p , q < co. (a) One sets
F:,P(Rn) = (fI f ES‘(Rn),IIfII,+(p)P.I =
II
5 r 5 co. One sets F;,q,(r)(Rn)= (f I f E fi‘(Rn),IIfIIpP.%-dr)
=
(b) Let additionally 1
Remark 1. Clearly,
II(2jIf
* pjI}IIlq,pIIz,p< “1.
II II(2”If * pjIHIzqIILp,,< “1.
FP>P ’*@) (Rn) = F;,p(Rn) = Bi,p(Rn) and F;,q,(p)(Rn)= p;,q(Rn)* To prove the independence of the spaces of the choice of the system {pk}goE @ in the same manner as in Theorem 2.3.2, one needs Theorem 2.2.4(b) with LP(lq,J [resp. LP,JZq)]instead of LP(lq).But this follows by interpolation from Theorem 2.2.4(b), Theorem 1.18.4 with po = p1 = p , and Theorem 1.18.3/2 [resp. Theorem 2.2.4(b) and Theorem l.lS.S/Z]. Afterwards, one can show as in the seventh step of the proof of Theorem 2.3.2 that F $ ’ ) ( R , , and ) Fi,q,(r)(Rn) are Banach spaces. (Since both types of spaces are interpolation spaces, one obtains the last assertion also as a consequence of the following theorem.) Theorem 1. Let -a<so,sl < co, 1 < p o , p l , q o , q l c 00, 0 c 8 < 1, and s = (1
- ep,,
+ esl,
1
1-8
P
Po
+-P81
-= -
and
1
1-8
q
!lo
-=-
+ -.e
Q1
(1)
2.4.2. Interpolation of the Spaces F;,JR,,)
CF&.(Rn), Fz,ql(Rn)1e= Fi,q(Rn)
185
(7)
-
Proof. The proof is analogous t o the proof of Theorem 2.4.1. The formula correspondinn t o (2.4.1/10) is llfll ~ ( { ( ~ ~ , ~ ~ f f n ) , ~ ~ , ~ , Hf ( ~ n* )~) }~) } l l ~ ( { ~ ~ ~ ( ~ (a) is a consequence of Theorem 1.18.4 and Theorem 1.18.2. One obtains (b) from Theorem 1.18.4 and Theorem 1.18.3/2, (c) follows from Theorem l.l8.6/2. Finally, (d) is a consequence of Theorem 1.18.4 and Theorem 1.18.1 with Aj = 2 j S 4 and Bj = 2jW, where C denotes the complex plane. R e m a r k 2. The theorem is fairly general and contains a number of interesting special cases. We remind of Bi,p(Rn)= Fi,p(Rn)and H;(Rn) = Fi,2(Rn). ( a ) I f - m < s O , s l < c o , 1 < p 0 , p 1 < ~ , a n d O < 8 < 1 , t h e n i t f o l l o w s f r o m(2) and (4) that
(B2a,po(Rn), B2x,p,(Rrz))e,p = Bi,p(XJ where s and p are determined by (1). This is also a special case of (2.4.1/7). 9
(b) If -co < so,sl < a,so from (2) that
+- sl, 1 < p o , p l <
03,
and 0 < 8 < 1, then it follows
( B ~ o , p o (H21(Rn))e,p ~n)~ = (H:o(Rn), H21(RJ)e,p= B;,p(Rn) where s and p are again determined by (1).
7
(c) If -co < s < co, 1 < p , , p1 <
03,
(8)
(9)
and 0 < 8 < 1, then it follows from (6) that
(10) (Hio(Rn)> Hil(Rrt))e,p= Hi(Rr,)> where s and p are again determined by (1). This formula is also animmediate conseL,,l(Rn))e,p = Lp(R,,)and Theorem 2.3.4. quence of (LpO(Rn),
(d) If -a < so, s1 < co, 1 < p,,, pl < that
03,
[H;,(Rn), H21(fL)le= Hi(&),
where s and p are determined by (1).
and 0 < 8 < 1, then it follows f~.om(7) (11)
~),~~
186
2.4. Interpolation Theory for the Spaces Bi,,(R,) and B';,&,)
(e) If -a < so,s1 < co, 1 < p o , p1 < co, and 0 < 8 < 1, then it follows from (7) that (12) [Hz(Rn),B2,p1(Rn)10= Fi,q(Rn) 9
1 1-8 where s and p are determined by (1) and - = ____ q
e +--. P1
R e m a r k 3. * Essentially, the theorem goes back to H. TRIEBEL[19]. The corresponding results are proved there directly, without use of Theorem 1.2.4. But numerous special cases are known before. As mentioned in Remark 2.4.115, formula (8) (as a special case of (2.4.1/7))is due t o P. GRISVARD [4]. The second part of (9) was formulated by J. PEETRE [15] without proof. (11) can be found in A. P. CALDER~N [3], and also in M. SCHECHTER [5, 61. T h e o r e m 2. Let -co < so,sl < co, so sl, 1 < p < 00, 1 5 q o , q l , q 6 co, and 0 < 8 < 1. If s = (1 - 0 ) so + 8sl, then
+
(BzqI(Rn) B2,qt(Rn))o ,q = (B2qo(Rn)F2,ql(Rn))e ,q
(13)
9
= (F2qa(Rn), F;q,(Rn))o,q = Bi,q(Rn)*
Proof. The first part of (13) coincides with (2.4.1/3). Then the other parts follow from the inclusion properties (2.3.2/4). R e m a r k 4. Interesting special cases of (13) are (Bzqo(&),H;(Rn))e,q = (H:(Rn), f$(Rn))e,q = Bi,q(Rn) 9
(14)
where the parameters have the same meaning as in the theorem. If additionally 0 5 s1 < co,then (Bzqo(Rn),W;(Rn))o,q = (H:(Rn), W2(Rn))o,q= Bi,q(Rn)*
For 0
s so, s1 <
03,
so
(15)
+ sl, one obtains that
(W',.(Rn),W ; ( m ) o , q = q , q ( R n ) f
(16)
R e m a r k 5. Theorem 2.3.3(b) shows that for s = 1 , 2 , 3 , . . . the spaces Wi(Rn) coincide with the usual Sobolev spaces. Further, it holds WE(Rn)= Lp(Rn). But then formula (16) is equivalent with one of the usual definitions of the Besov spaces, namely the definition by interpolation of Sobolev spaces. R e m a r k 6. By (13) and (1.3.3/5),it follows that
where the parameters have the same meaning as in Theorem 2. On the right-hand side of (17), one may replace Bgqoby H;, F2qoor W;, similarly for B2,q,.Estimates of type (17) are often needed rn the theory of elliptic differential operators. Sometimes they are called multiplicative inequalities.
2.5.1. Equivalent Norms and Tranalation Groups
2.5.
187
Equivalent Norms in the Spaces Bi,a(RJ
In Theorem 2.3.3 and in Remark 2.3.312, we have obtained equivalent norms for the Lebesgue spaces Hi(R,J, -co < s < co, and particularly for the Sobolev spaces WT(R,), m = 1 , 2 , 3 , . . ., which coincide with the usual definitions of these spaces. This section is concerned with equivalent norms for the Besov spaces B&(R,,), 0 < s < 00. Formula (2.4.2/16) shows that the spaces BiJR,,) for s > 0 are mterpolation spaces of Sobolev spaces. On this basis, one can describe the main motive of the considerations of this section as follows. I n the space Lp(Rn),we choose a strongly continuous semi-group (resp. n strongly continuous semi-groups commutative t o each other) of operators with the infinitesimal operator A (resp. A,, . .,A,) such that Wr(R,) = D(Ak)(or
.
n D ( A f ) )for suitable natural numbers m and k. 11
=
j-1
Then we can apply the results of Sections 1.13, 1.14, and 1.15, and we obtain a large number of equivalent norms in the interpolation space BE,')"(Rn) = (WF(R,), Lp(Rn))e,q. There are three types of semi-groups under consideration : Translation groups, Gauss-Weierstrass semi-groups, and Cauchy-Poisson semi-groups. Finally, we determine equivalent norms by approximation of functions belonging to Lp(Rn) by smooth functions.
2.6.1.
Equivalent Norms and Translation (froups
I n the space LJR,), there are considered the strongly continuous commutative groups ( G j ( t ) ) j' = 1, . ., n, of isometric operators,
.
-
[aj(t)f l (2)= f ( ~ 1* ,
9
xj-1, xj
+ t,
2 j + 1 9
* * *
9
zn),
f
E Lp(Rn)*
(1)
For our purpose, only the semi-group{aj(t)}O~t<,is of interest. The corresponding infinitesimal operator will be denoted by Aj, its domain of definition by D(Aj). Similarly, AT and D(AY) are defined for m = 1 , 2 , . Further, K m has the same meaning as in Definition 1.13.3. Temporarily we set
..
m
1 , 2 , . . ., j = 1,. . . , m , 1 < p < 00. L e m m a . (a) S(R,) is dense in WgI(R,)and =
where w,,,and (b) It holdsP'hd
llfll *q, are equivalent m w .
188
2.5. Equivalent Norms in the Spaces B;,JR,,)
and n
=
~m
n D ( A ~=) w;(R,). j=1
(5)
Proof. Xtep 1. One obtains formula (3) in the same manner as in the fourth step of the proof of Theorem 2.3.3; see also Remark 2.3.312. If v E CF(RI,)with ~ ( x=) 1 if 1x1 5 1, then i t is not hard t o see that f E WEj(Rn)can be approximated in
(3
Wgj(Rn)by the functions p - f(z) if N -+ 11m11~, = 1. For 0
00.
Let 0
<= w ( z ) E C$'(R,J
with
< h < co and g e L p ( R , , )we , see
This is SOBOLEV'S well-known mollification method, see S. L. SOBOLEV[4]. (For n = 1 this method is also known as STEKLOV'S mollification method.) It holds that 1101~11L, = 1. Whence, it follows that (g)h ELrdRn) C " ( R ~ ~and ) II(g)hllLp l l g I / L p * Since for functions of c$'(RrI)it holds II(g)h - gllLp(Rn)+ 0 if h 10, one obtains by the last estimate that the last relation is true for arbitrary functions g belonging t o
Lp(Rn).If g has a compact support then (g)h E C$'(RIJ.Setting g ( z ) = ak
it follows for D" = -, 0 p ( ( g ) h (2))=
s
ax;
Dzmh(z
R.
(3
- f(x),
5 k 5 m, by the definition of derivatives of distributions
- Y ) g ( Y ) dy
= (-1)Ia'
s Dimh(z -
Rri
g ( Y ) dy = oh * D"g. (7)
v (i1 f(x) ( - f ( x ) in WEj(RI,)if h 10. Hence, CF(RIJ,and so also S(R,), is dense
Using the preliminary considerations, whence it follows that C$'(Rn)3 tends to v
v
(3
-
in W;j(Rn)*
Step 2. It is not hard to see that for f E S(Rn)c D(AY)we have
where D ( A y ) is normed by ll*ll~pj or by Il*l&;,. By Theorem 1.13.1/2, (0, co) belongs to the resolvent set of Aj . Particularly, (Aj - E)"' is an isomorphic mapping from D ( A y ) onto Lp(Rn).Since X(Rn)is dense in W;j(RI,)and D(AY) = D((Aj- E ) m ) is closed, one obtains D(Ay) 3 W g j ( R I , )Here, . ( 8 ) holds for f E WEj(&,), too. On the other hand, using Fourier t,ransformations, i t follows that (Aj - E)IJ1is an isomorphic mapping from S(Rn) onto S(Rn). Completion shows that (Aj - E)" is also an isomorphic mapping from W;fj(Rn) onto LJR,). But then we have D(A7) = W g j ( R n )Formula . ( 5 )is a consequence of Remark 2.3.312. R e m a r k 1. The first step contains a complete description of SOBOLEV'S molli[4]) which is very important. We used a similar fication method (see S. L. SOBOLEV method in the first chapter several times.
2.5.1. Equivalent Norma and Translation Groups
R e m a r k 2 . I n the case considered here we have K m = 1.13.4/2. We sha.11
189
n
n D(A7). See Remark
and The norms II.()L;(Q6,A) and and 1.13.2.
11.11 L:,j0,6),~) have the same meaning as in Subsections 1.13.4
T h e o r e m . Let 0 < s < 00, 1 < p < co, and 1 5 q w i t h O 5 k < s a n d l > ~ - k , a n d i f 0 < 6 6 co,then
00.
If k and I are integers
where
A11 these norms are equivalent to 11 f 11 B;,f. Proof. The proof follows immediately from (2.4.2/16) with so = 0 and s1 = m > s
= O m , the last lemma, a,nd Theorem 1.13.6/1.
R e m a r k 3. Clearly, on the basis of the considerations to Theorem 1.13.6/1 (see particularly (1.13.6/9)),one obtains numerous further equivalent norms. For instance, ajf
C by C and - by Pf. Further, one can add in j=1 IalSk ax! (lo), (ll),or (12) norms of the form Ilf IIq or 11 f 11 with 0 5 o < s. This is a conse-
in (10) one can replace
I1
quence of (2.3.3/8) and (2.3.213). Those norms in which 1 is as small as possible are of special interest. For 0 < s < co,we set and
+ {s}, [s] integer, O 6 {s} < 1 , = [s]- + {s}-, [s]- integer, O < {s}+
= [s]
1.
The smallest number for I is I = 1 + [{a)']. Then we have k = [s]-. This shows that the spaces Bi,q(R,,),where s is an integer, hold a special position. R e m a r k 4. The spaces W;(RJ,0 < s < 00, 1 < p < 00, are defined in (2.3.1/7). For the Sobolev space W;(Rn)with s = 1, 2 , 3 , . . . , we obtained in Theorem 2.3.3(b) and Remark 2.3.312 a satisfactory description. Now, for the Slobodeckij spaces
190
2.5. Equivalent Norms in the Spaces Bi,JRJ
W;(R,,)= B;,p(Rn),0 < s
+ integer, it follows from the Theorem and Remark 3
R e m a r k 5. * By the last theorem, we obtained the usual definitions for the Slobodeckij spaces W”,R,), 0 < s integer, 1 < p < 00 (N. ARONSZAJK[l], L. N. SLOBODECKIJ[l], E. GAGLIARDO [l]) and for the Besov spaces Bi,q(Rn), 0 < s < 00, 1 < p < co, 1 5 q 5 co (0.V. BESOV [1, 21). Numerous further equivalent norms, partly considered in the following subsections, can be found in S. M. NIKOL’SKIJ [7] and M. H. TAIBLESON[1, I]. I n this connection we refer also t o P. L. BUTZER,H. BERENS[l]; P. L. BUTZER,R. J. NESSEL [l]; and P. I. hZORKIN [13]. Further references are given in Remark 2.3.3/3 and Remark 2.3.1/2.
+
2.6.2.
Equivalent Norms and Gauss-Weierstrass Semi-Groups
On the basis of the considerations on analytic semi-groups in Subsection 1.14.5, we obtain in an easy way further equivalent norms in the space B;,q(Rn). Lemma. Let 1 < p < co and n
[ W ( t )f] (x)= (4nt)-T
J
R”
lx-yp
e - 7 f ( y )d y , 0 < t < co,f
E LJR,).
If additionally W ( 0 )= E , then { W(t)}Ost< is an analytic semi-group i n the s Lp(Rn).I f A is the corresponding infinitesimal operator, then
Amf = A m / , D(Am)= W26n(R,,), m = 1 , 2 , . . . Proof. Step 1. By the well-known formula
(1)
p
(2)
if follows that
II W)II 6
1-
(4 )
For f E S(Rn),(1) is the uniquely determined bounded solution of the classical Cauchy problem of the heat conduction equation
The uniqueness of the solution shows that
W(t, + t z ) f = W ( t , ) [
W 2 )
fl,
0
s t,,
t2
<
03 *
Together with (4), the semi-group property follows from the last equation. With the aid of (3), one shows for f E S(Rn)in the usual way that
I l w ( t ) f - f l l L p + o if t l 0 .
(6)
2.5.2. Equivalent Norms and Gauss-Weierstrass Semi-Groups
191
By (4), it follows that (6) is true for f E LJR,), too. Hence, { W(t))ostt< a, is a strongly continuous semi-group with B = 0 in (1.13.1/1). Step 2. One obtains by ( 5 )and the definition of A that S(R,) c D(A) and Af = Af if f E S ( R , ) . Here we used A W ( t )f = W ( t )Af. By Theorem 2.3.3, it holds for f E S(R,) that
II(d - E ) fllr.,
=
l1F-V
+ 1512)Y I I L ,
- llfllJ";.
Since A is a closed operator, and since S(R,) is dense in v , ( R , , ) , it follows that D ( A ) 3 W;(R,) and Af = Af if f E W&,(R,).Conversely, let f e L p ( R I z ) Then . Fl(1 I[l2)-l F f belongs t o W;(R,,)and
+
-(A - E ) (F-'(l
+
1512)-'
Ff)
=
f.
Whence it follows D ( A ) = W8,(R,,).Further, one obtains that ( A - E ) m is an isomorphic mapping from D ( A m )onto L p ( R f f As ) . before, i t follows D(Anl)= Wim(R,). Step 3. By derivation it follows that the assumptions of Definition 1.14.5 are satisfied. Hence { W(t))Ost