Universität Konstanz
Analytic semigroups of pseudodifferential operators on vector-valued Sobolev spaces
Bienvenido Barraza Martinez Robert Denk
Jairo Hernández Monzón
Konstanzer Schriften in Mathematik Nr. 303, Mai 2012
ISSN 1430-3558
© Fachbereich Mathematik und Statistik Universität Konstanz
Fach D 197, 78457 Konstanz, Germany
Konstanzer Online-Publikations-System (KOPS) URL: http://nbn-resolving.de/urn:nbn:de:bsz:352-193526
OPERATORS ON VECTOR-VALUED SOBOLEV SPACES
B. BARRAZA MARTINEZ, R. DENK, AND J. HERN ´ANDEZ MONZ ´ON
Abstract. In this paper we study continuity and invertibility of pseu- dodifferential operators with non-regular Banach space valued symbols.
The corresponding pseudodifferential operators generate analytic semi- groups on the Sobolev spacesWpk(Rn, E) withk∈N0, 1≤p≤ ∞.Here E is an arbitrary Banach space. We also apply the theory to solve non- autonomous parabolic pseudodifferential equations in Sobolev spaces.
1. Introduction
In the present work, we regard symbols in the space S1,0m,ρ(Rn, E) with m ∈ R, ρ ∈ N0 and E being an arbitrary Banach space. We say that a ∈ Sm,ρ1,0 (Rn, E) ifa∈Cρ(Rn, E) and if for allα ∈Nn0 with|α| ≤ρthere exists a cα >0 such that
∂ξαa(ξ)
E ≤cαhξim−|α| for all ξ∈Rn. (1) Herehξi:=p
1 +|ξ|2,ξ∈Rn. Because of Mikhlin’s theorem (see [BL76] and [NNH02]) the pseudodifferential operator or the Fourier multiplier induced by a,
F−1aF :Lp(Rn, E1)−→Lp(Rn, E2) for 1< p <∞, (2) is continuous if a ∈ S1,00,n+1(Rn,L(E1, E2)) and E1, E2 are Hilbert spaces.
HereF andF−1denote the Fourier transform and the inverse Fourier trans- form, respectively. In practice one would like to have the validity of this result for arbitrary Banach spaces E1, E2. But, this is impossible in light of an observation of G. Pisier: If the Mikhlin theorem is valid onLp(Rn, E) for L(E)-vector-valued symbols, then E is isomorphic to a Hilbert space (see [LLM98] for a proof).
Therefore, in order to obtain versions of Mikhlin’s theorem, one has to, for example, change the space scale Lp(Rn, E), make additional conditions for (1), or impose conditions on the geometry of the Banach space E. On one hand, Amann has shown in [Am97], Theorem 6.2, that for arbitrary
B. Barraza Martinez, Universidad del Norte, Departamento de Matem´aticas, Barranquilla (Colombia)
R. Denk, Universit¨at Konstanz, Fachbereich f¨ur Mathematik und Statistik, Konstanz (Germany)
J. Hern´andez Monz´on, Universidad del Norte, Departamento de Matem´aticas, Barranquilla (Colombia)
E-mail addresses:bbarraza@uninorte.edu.co, robert.denk@uni-konstanz.de, jahernan@uninorte.edu.co.
Date: May 8th, 2012.
2
Banach spaces E1 and E2 and a ∈ S1,0m,n+1(Rn,L(E1, E2)) that the pseu- dodifferential operator F−1aF : Bp,qs+m(Rn, E1) −→ Bp,qs (Rn, E2) (s ∈ R, p, q∈[1,∞]) is continuous, where Bp,qs (Rn, E) denotes the Besov space.
On the other hand, Weis proved in [We01] for UMD spaces E1 and E2
and symbols ainS1,00,1(R,L(E1, E2)) which fulfill a stronger condition than (1), that F−1aF :Lp(R, E1) −→Lp(R, E2) (for 1< p <∞) is continuous.
Similar results were obtained in [PS06]. By a UMD space E one under- stands a Banach space for which the Hilbert transform is continuous from Lp(R, E) toLp(R, E) for 1< p <∞, or equivalently, for which the function m(t) =|t|−1tis a Fourier multiplier onLp(R, E) (1< p <∞). This implies reflexivity of E (see [Am95], Remark 4.4.2).
In order to achieve an optimal value of ρin (1), one has to regard the ge- ometry of the Banach spacesE1, E2. Girardi and Weis considered in [GW03]
Banach spaces E1, E2 of Fourier type p, 1 ≤p ≤2, (i.e. Banach spaces Ei
for which the Fourier transform is continuous ofLp(Rn, Ei) intoLp0(Rn, Ei) with 1p +p10 = 1), and symbols a∈S1,00,ρ(Rn,L(E1, E2)). Under these condi- tions, they proved that for alls∈R,r, q∈[1,∞], the pseudodifferential op- erator F−1aF :Br,qs (Rn, E1)−→Br,qs (Rn, E2) is continuous, ifρ=
hn p
i +1.
For for arbitrary Banach spaces E1, E2 the Fourier type equalsp = 1, and one obtains Theorem 6.2 in [Am97]. If E1,E2 are uniformly convex Banach spaces (thus having Fourier type p >1), see [Bo87], one can chooseρ=n.
Many applications to problems of physics and biology, e.g. models for reaction-diffusion processes (see [Am00]), suggest the necessity to regard Banach space-valued symbols. The space considered in [Am00] is the vector- valued Sobolev-Slobodeckii space W1s(Rn, L1(Y, µ)), s /∈ N, and it is well- known that L1(Y, µ) is not a reflexive space. For this reason, we would like to obtain a version of the Mikhlin theorem (like in [Am97]) for Ba- nach space-valued symbols, in which the correspondent pseudodifferential operator generates an analytic semigroup on Wpk(Rn, E), 1≤p ≤ ∞. As a consequence, one obtains existence and uniqueness of solutions of parabolic pseudodifferential equations.
Therefore, we regard arbitrary Banach spaces Ei, i = 0,1,2, the vector space
V(Rn, Ei) := [
(s,p,q)∈R×[1,∞]×[1,∞]
Bp,qs (Rn, Ei) and symbols a∈S1,0m,ρn(Rn, E1) with regularity
ρn:=
n+ 1, n∈N odd,
n+ 2, n∈N even. (3)
We will show (see Theorem 3.1) that there exists a unique linear map a](D) :V(Rn, E2)−→V(Rn, E0)
with a](D) ∈ L(Bp,qs+m(Rn, E2), Bp,qs (Rn, E0)) such that its restriction on every intersection Cb∞(Rn, E2)∩Bp,qs+m(Rn, E2) (s ∈ R, p, q ∈ [1,∞]) co- incides with the classical definition of the pseudodifferential operator a(D) in [Ku81], which has been defined by means of an oscillatory integral. In
this way, we show that suitable restrictions of the operator −a](D) generate analytic semigroups on the Besov spaces Bp,qs (Rn, E0) andC∞-semigroups on the Sobolev spaces Wpk(Rn, E0). Thus we obtain results similar to those in [Am97]: Using a](D) ∈ L(Bp,qs+m(Rn, E2), Bp,qs (Rn, E0)), the continuous embedding
Bp,1k (Rn, E),→Wpk(Rn, E),→Bkp,∞(Rn, E) (1≤p≤ ∞),
and some properties of Banach space interpolation, we will show that the Wpk−realization of parabolic pseudodifferential operators are the negative generators of analytic semigroups on Wpk(Rn, E0) with 1 ≤ p ≤ ∞ (see Theorem 3.14). Therefore, we obtain the existence and uniqueness of solu- tions for a non-autonomous Cauchy problem in Sobolev spaces Wpk(Rn, E) (see Theorem 4.3).
For vector-valued differential operators, generation of an analytic semi- group on Lp(Rn, E0),C0(Rn, E0) andBU C(Rn, E0) was shown by Amann in [Am01]. Pseudodifferential operators with smooth symbols (with respect to the dual variable ξ) were considered in [Ki03]. In [Ki01], the generation of an analytic semigroup in Lp(Rn, E0) (1 ≤ p < ∞) was shown for suit- able parabolic vector-valued pseudodifferential operators where the symbols are assumed to have a homogeneous principal part and regularity greater than or equal to 2n+ 1. Apart from generalizing these results, the method used in the present paper will allow us to consider x-dependent symbols in a forthcoming publication.
The plan of the paper is as follows: After some preliminary definitions and remarks in Section 2, we prove in Section 3 the main results of the present paper on continuity (Theorem 3.1) and on generation of an analytic semigroup (Theorem 3.14). As an application, we prove in Theorem 4.3 the existence and uniqueness of the solution of a non-autonomous Cauchy problem in the Sobolev spaces Wpk(Rn, E). The constants C1, . . . , C15, M1
are rigourously calculated, because we will explore stability conditions for an adequate family of pseudodifferential operators in Remark 4.4.
2. Preliminary definitions and remarks
In the following, E and Ei always denote arbitrary Banach spaces with norm k·kE and k·kE
i, respectively, and L(E1, E0) the space of linear, con- tinuous maps of E1 into E0. The definitions and properties of functions spaces are taken from [Am97]. In particular,S(Rn, E) denotes the Schwartz space of rapidly decrasing functions, Cbk(Rn, E), k ∈ N0, is the space of all functions u : Rn → E such that ∂αu is bounded and continuous on Rn for all |α| ≤ k, BU Ck(Rn, E) is the space of all u ∈ Ck(Rn, E) such that ∂αu is bounded and uniformly continuous on Rn for all |α| ≤ k and BU C(Rn, E) :=BU C0(Rn, E).
The Besov space Bp,qs , the homogeneous Besov space ˚Bp,qs and the small Besov space spaces bsp,q are defined as follows: For s ∈R and p, q ∈ [1,∞]
one defines the E-valued Besov space of order sby Bp,qs (Rn, E) :=
u∈ S0(Rn, E) :
2jskψj(D)ukL
p(Rn,E)
lq <∞ (4)
with norm
kukBs
p,q(Rn,E):=
2jskψj(D)ukL
p(Rn,E)
lq,
where S0(Rn, E) is the space of theE-valued tempered distributions, ψj(D)u:=F−1(ψjFu) (5) and (ψj)j∈
N0 is a resolution of unity which is constructed in the following way: For ψ∈ S(Rn) :=S(Rn,C) with
suppψ⊂ {x∈Rn:|x| ≤2} and ψ(x) = 1 on |x| ≤1, (6) one definesψ(x) :=e ψ(x)−ψ(2x),ψj(x) :=ψ(2e −jx) forx∈Rnandψ0 :=ψ, ψ−1 := 0. The sequence (ψj)j∈
N0 satisfies:
supp(ψ0)⊂Ω0 :={x∈Rn:|x| ≤2}, supp(ψj)⊂Ωj :=
x∈Rn: 2j−1 ≤ |x| ≤2j+1 for all j∈N, (7) as well as
∞
X
j=0
ψj(ξ) = 1 for all ξ∈Rn. (8) Moreover, for each α∈Nn0 there exists a constant cα>0 such that1
Dαξψj(ξ)
≤cα2−j|α|1Ωj(ξ) for allξ ∈Rn and j∈N0. (9) One defines the Banach spaces ˚Bsp,q(Rn, E) and bsp,q(Rn, E) as the closures of S(Rn, E) and Bp,qs+1(Rn, E), respectively, where the closure is taken with respect to the normk · kBs
p,q.
The following three lemmas will be crucial for the main results in Sec- tion 3.
Lemma 2.1. Let s∈R andp, q∈[1,∞]. Then we have:
a) S(Rn, E),→ Bp,qs (Rn, E) ,→ S0(Rn, E) and S(Rn, E) ,→d Bp,qs (Rn, E) if p, q ∈ [1,∞). Here ,→ denotes continuous embedding and ,→d denotes con- tinuous and dense embedding.
b) For each 1≤q0 ≤q1 ≤ ∞,
Bp,qs 0(Rn, E),→Bsp,q1(Rn, E). (10) c) For each ε >0 and 1≤q0, q1 ≤ ∞,
Bp,qs+ε0(Rn, E),→Bp,qs 1(Rn, E). (11) d) Let s1 ≤s0 and 1≤p0≤p1 ≤ ∞with s1−pn
1 =s0−pn
0. Then
Bps00,q(Rn, E),→Bsp11,q(Rn, E). (12) e)
bsp,q(Rn, E) =
Bp,qs (Rn, E), if1≤p≤ ∞, 1≤q <∞, B˚p,∞s (Rn, E), if 1≤p <∞, q=∞.
f ) For 1≤p <∞ andk∈N,
B˚p,1k (Rn, E),→d Wpk(Rn, E),→d B˚p,∞k (Rn, E) =bkp,∞(Rn, E), (13)
11Ωdenotes the characteristic function of a set Ω.
B∞,1k (Rn, E),→W∞k (Rn, E),→Bk∞,∞(Rn, E), (14) bk∞,1(Rn, E) =Bk∞,1(Rn, E),→d BU Ck(Rn, E),→d bk∞,∞(Rn, E), (15) B˚∞,1k (Rn, E),→d C0k(Rn, E),→d B˚∞,∞k (Rn, E), (16) B∞,1k (Rn, E),→Cbk(Rn, E),→B∞,∞k (Rn, E). (17) g) If p, q, q1, q2 ∈[1,∞], −∞< s1< s2 <∞ andθ∈(0,1), then2
Bp,qs11(Rn, E), Bp,qs22(Rn, E)
θ,q
∼=Bp,q(1−θ)s1+θs2(Rn, E). (18)
Proof. See [Am95], Chap. 5 or [Sc86].
By Definition and Lemma 2.1 a), it follows immediately that S(Rn, E),→d B˚p,qs (Rn, E),→Bp,qs (Rn, E) (s∈R, p, q∈[1,∞]), Bp,qs+1(Rn, E),→d bsp,q(Rn, E),→Bp,qs (Rn, E) (s∈R, p, q∈[1,∞]), B˚sp,q(Rn, E) =Bp,qs (Rn, E) (s∈R, p, q∈[1,∞)).
(19)
Also one can easily see, using again the previous Lemma, that if B ∈ {b,B, B}, 1˚ ≤p, q1, q2≤ ∞,t > s and E1,→E0, then
Bp,qt
1(Rn, E1),→ Bp,qs
2(Rn, E0) . (20)
If E1
,→d E0, the continuous embedding (20) is always dense, except when B=B andq1 orq2 is ∞. Moreover
Bp,qs 1(Rn, E1),→ Bp,qs 2(Rn, E0) (1≤q1≤q2 ≤ ∞). (21) Lemma 2.2. Let s∈R, p∈[1,∞] andq ∈[1,∞). Then
Cb∞(Rn, E)∩Bp,qs (Rn, E) ;k·kBs p,q
d
,→Bp,qs (Rn, E). (22) Proof. For alls∈R,p∈[1,∞] and q∈[1,∞) it is clear that
Cb∞(Rn, E)∩Bp,qs (Rn, E) ;k·kBs p,q
,→Bp,qs (Rn, E).
It only remains to prove the density. In the case p <∞, the density in (22) follows from Lemma 2.1 a) andS(Rn, E)⊂Cb∞(Rn, E)∩Bp,qs (Rn, E). Now, let u∈B∞,qs (Rn, E) withq ∈[1,∞), (ψj)j∈
N0 a partition of unity as in (8) and uN :=
N
P
k=0
ψk(D)u,N ∈N. By definition (4), ψj(D)u∈L∞(Rn, E) for all j∈N0, and thereforeuN ∈L∞(Rn, E). On the other hand, let (ϕj)j∈
N0
be another partition of the unity as in (8). Then one obtains from (5) and the fact that convolution corresponds to multiplication of the Fourier transform3 that
ϕj(D) (ψk(D)u) = (2π)−n2 ϕˇj ∗(ψk(D)u)∈ OM(Rn, E) (23)
2Here (E0, E1)θ,q denotes the real interpolation space with exponentθ and parameter q between the Banach spacesE1 andE0.
3OM(Rn, E) denotes the space ofE-valued slowly increasing smooth functions onRn. For the result on convolution see [Am97], Th. 3.6.
for all j, k∈N0 where ˇϕj :=F−1ϕj. Therefore ϕj(D)(ψk(D)u) is a regular tempered distribution. From (23), ˇϕj ∈ L1(Rn), ψk(D)u ∈ L∞(Rn, E0) and Theorem 1.9.9 in [Am03] it follows that for all j, k∈N0
kϕj(D) (ψk(D)u)kL
∞ ≤ kϕˇjkL
1kψk(D)ukL
∞ =cϕkψk(D)ukL
∞ (24) where the constant cϕ := max{kϕkˇ L1,k(ϕ)e ∨kL1} is independent ofj and k.
Thus
ϕj(D) (u−uN) =ϕj(D)u−
N
X
k=0
ϕj(D) (ψk(D)u)∈L∞(Rn, E).
From supp (ψj), supp (ϕj)⊂Ωj, (24) and
∞
P
k=0
ψk(D)u=uin S0(Rn, E), we see that
ku−uNkqBs
∞,q(Rn,E)
=
∞
X
j=0
2jsq
ϕj(D) u−
N
X
k=0
F−1(ψkFu)
q
L∞(Rn,E)
=
∞
X
j=0
2jsq ϕj(D)
X∞
k=0
F−1(ψkFu)−
N
X
k=0
F−1(ψkFu)
q
L∞(Rn,E)
(5)=
∞
X
j=0
2jsq F−1
ϕj
∞
X
k=N+1
ψkFu
q
L∞(Rn,E)
= 2N sq
F−1(ϕNψN+1Fu)
q
L∞(Rn,E)
+ 2(N+1)sq
F−1(ϕN+1(ψN+1+ψN+2)Fu)
q
L∞(Rn,E)
+
∞
X
j=N+2
2jsq
F−1(ϕjFu)]
q
L∞(Rn,E)
since
1
X
r=−1
ϕjψj+r =ϕj
(5)= 2N sqkψN+1(D) (ϕN(D)u)kqL
∞
+ 2(N+1)sqk(ψN+1+ψN+2) (D) (ϕN+1(D)u)kq∞ +
∞
X
j=N+2
2jsqkϕj(D)ukq∞}
(24)
≤ cn,q,s,ψn
2N sqkϕN(D)ukq∞+ 2(N+1)sqk(ϕN+1(D)u)kq∞ +
∞
X
j=N+2
2jsqkϕj(D)ukq∞o
=cn,q,s,ψ
∞
X
j=N
2jsqkϕj(D)ukq∞ −→
N→∞0,
becauseuis inBs∞,q(Rn, E). This implies thatuN ∈B∞,qs (Rn, E) anduN → u inB∞,qs (Rn, E). FurthermoreuN is a regular distribution in Cb∞(Rn, E),
because we have, setting χk :=P1
r=−1ψk+r, uN =
N
X
k=0
F−1(ψkFu) =
N
X
k=0
F−1((χkψk)Fu)
=
N
X
k=0
F−1(χkF(ψk(D)u)) =
N
X
k=0
cnχˇk∗(ψk(D)u) . Thus, it follows from Theorem 1.9.9 in [Am03], for each α∈Nn0, that
∂αuN =cn
N
X
k=0
(∂αχˇk)
| {z }
∈L1
∗(ψk(D)u)
| {z }
∈L∞
∈BU C(Rn, E) .
This finishes the proof.
Lemma 2.3. Let (ψj)j∈
N0 as in (8), B :=
x∈Rn: 12 ≤ |x| ≤2 , m∈ R and a∈S1,0m,n+1(Rn, E). Then (ψja)∨ ∈L1(Rn, E) and
(ψja)∨
L1(Rn,E)≤C12jmkakSm,n+1
1,0 (Rn,E) (j∈N0), where the constantC1is given byC1 := 23m2 ωnC1∗max
vol(B), P
|α|=n+1
nn+12 with C1∗ := kψke Cn+1
b
P
|α|=n+1 β≤α
α β
2|β| and ωn denoting the volume of the unit ball in Rn.
Proof. For allj∈N0, we have
k(ψja)∨kL1(Rn,E)=k(ψa(2e j·))∨kL1(Rn,E) (25) where ψe= ψ in the case j = 0. Now, let 1B be the characteristic function of B, and let α ∈Nn0 with|α| ≤n+ 1 and j ∈N. Then for all ξ ∈Rn we obtain4
Dα(ψa(2e j·))(ξ) E
≤ X
β≤α
α β
2j|β|
|{z}
=2|β|2(j−1)|β|
|Dα−βψ(ξ)|e 1B(ξ)
(Dβa)(2jξ) E
≤ X
β≤α
α β
2|β|kψke Cn+1
b 1B(ξ) max
|β|≤n+1sup
Ωj
hξi|β|
Dβa(ξ) E
≤C1∗kakSm,n+1
1,0 1B(ξ) sup
Ωj
hξim
≤C1∗kakSm,n+1
1,0 1B(ξ)·
( 1 + 22(j+1)m/2
ifm≥0, 1 + 22(j−1)m/2
ifm <0,
≤23m2 C1∗2jmkakSm,n+1
1,0 1B(ξ) .
An analogous result is obtained for j= 0 with ψe=ψ. Therefore, Dα ψa(2e j·)
∈L1(Rn, E),
4Dα:= (−i)|α|∂α.
and for all j∈N0 and |α| ≤n+ 1 we have
Dα(ψa(2e j·))
L1(Rn,E)≤23m2 C1∗2jmkakSm,n+1 1,0 . From the last inequality, the equality
xαF−1(ψa(2e j·)) = (−1)|α|F−1 Dα(ψa(2e j·))
∈C0(Rn, E), (25) and the inequality |x|n+1 ≤nn+12 P
|α|=n+1|xα|one concludes that
(ψja)∨
L1(Rn,E)
= Z
B1(0)
F−1(ψa(2e j·))(x) Edx+
Z
Rn\B1(0)
F−1(ψa(2e j·))(x) Edx
≤ Z
B1(0)
Z
B
|ψ(ξ)| ka(2e jξ)kEdξ dx
+ Z
Rn\B1(0)
nn+12 P
|α|=n+1|xα|
|x|n+1
F−1
ψae 2j· (x)
Edx
≤23m2 C1∗ωnvol(B)2jmkakSm,n+1 1,0
+ X
|α|=n+1
nn+12 Z
Rn\B1(0)
1
|x|n+1 F−1
Dα
ψae 2j· (x)
Edx
≤23m2 C1∗ωnvol(B)2jmkakSm,n+1 1,0
+ X
|α|=n+1
nn+12 Z
Rn\B1(0)
1
|x|n+1 Z
B
Dα
ψae 2j· (ξ)
Edξ
dx
≤23m2 C1∗ωnvol(B)2jmkakSm,n+1 1,0
+ 23m2 C1∗
X
|α|=n+1
nn+12
2jmkakSm,n+1 1,0
Z
Rn\B1(0)
1
|x|n+1dx
≤C12jmkakSm,n+1 1,0 , with C1 := 23m2 ωnC1∗max
(
vol(B), P
|α|=n+1
nn+32 )
.
Remark 2.4. Leta∈S1,0m,ρ(Rn, E1) andu∈Cb∞(Rn, E2). For the definition of
[a(D)u](x) = Os− Z Z
R2n
eiξ·ηa(ξ)•u(x−η)d(ξ, η)
(2n)n, x∈Rn, (26) (cf. [Ku81]), the condition ρ≤n+ 1is not sufficient, because
bx(ξ, η) :=a(ξ)•u(x−η)∈ Amδ,τ
withδ = 0and τ = 0, and the oscillatory integral in (26)exists ifn+τ <2l and n+m
1−δ < 2l0, where 2l ( l ∈ N0) denotes the necessary number of derivatives in the variable ξ (see [Ku81]). We obtain the condition n < 2l
and therefore we consider in this paperρ≥ρn, whereρn is the smallest even number greater than n (see (3)).
3. Main results
Theorem 3.1 (and definition of the operator a](D)). Let m ∈ R, a ∈ Sm,ρ1,0 n(Rn;E1) and •:E1×E2 →E0 a multiplication5. Then it holds:
a) There exists a map T :V(Rn, E2)−→V(Rn, E0), such that i) T is linear.
ii) T
Bp,qs+m:Bp,qs+m(Rn, E2)−→Bsp,q(Rn, E0) is continuous for all s∈R and p, q∈[1,∞].
iii) T
(Bs+mp,q (Rn,E2)∩Cb∞(Rn,E2)) =a(D)for all(s, p, q)∈R×[1,∞]×
[1,∞], where [a(D)u] (x) =os−
Z Z
eiξ·ηa(ξ)•u(x−η)d(ξ, η)
(2π)n , u∈Cb∞(Rn, E2). b) LetS :V(Rn, E2)−→V(Rn, E0) be another map satisfaying i)-iii).
Then T =S. We denote this operator T by a](D).
Proof. The proof of a) is done in four steps:
I) We will show that for all (s,(p, q))∈R×[1,∞]2 a(D) :
Bp,qs+m(Rn, E2)∩Cb∞(Rn, E2) ;k·kBs+m p,q (Rn,E2)
→Bp,qs (Rn, E0) is continuous. For this purpose one calculatesψj(D) (a(D)u) with u∈Bp,qs+m(Rn, E2)∩Cb∞(Rn, E2):
[ψj(D) (a(D)u)] (x) = [(ψja) (D)]u(x)
= (2π)−n2
(ψja)∨∗•χj(D)u
| {z }
∈Lp(Rn,E0)
(x) , where χj := P1
i=−1ψj+i and ∗• is the convolution relative to the multiplication • (see [Am97]). From this and from Lemma 2.3 it follows for allu∈Bp,qs+m(Rn, E2)∩Cb∞(Rn, E2) that
ka(D)ukBs
p,q(Rn,E0)≤C2kakSm,n+1
1,0 (Rn,E1)kukBs+m
p,q (Rn,E2), withC2 := 3(2π)−n2C1.
II) Because we have the embedding
Bp,qs (Rn, E)∩Cb∞(Rn, E) ;k·kBs p,q
d
,→Bp,qs (Rn, E) (27) for all 1 ≤ p ≤ ∞ and 1≤q < ∞ (see Lemma 2.2), there exists a unique linear continuous extension
as,p,q(D) :Bp,qs+m(Rn, E2)→Bp,qs (Rn, E0) of a(D) giving I).
5that is, a continuous bilinear map with an operator norm smaller than or equal to 1.
III) One defines T :V(Rn, E2) −→V(Rn, E0) by T
Bp,qs+m :=as,p,q(D) for all (s, p, q)∈R×[1,∞]×[1,∞) and by as−1,p,q(D) in the other cases. T is well-defined on V(Rn, E2) and fulfillsa) for all q <∞.
IV) The case q =∞ can be treated using real interpolation theory.6 In fact, from Bs+mp,∞ (Rn, E2) = (Bp,1s+m−1(Rn, E2), Bp,1s+m+1(Rn, E2))1
2,∞, Bp,∞s (Rn, E0) = (Bp,1s−1(Rn, E0), Bs+1p,1 (Rn, E0))1
2,∞, and from Bp,1s+m−1(Rn, E2) −−−−−−−→as−1,p,1(D)
continuous Bp,1s−1(Rn, E0)
∪ ∪
Bs+mp,∞ (Rn, E2) −−−−−−−→as−1,p,1(D) Bp,∞s (Rn, E0)
∪ ∪
Bp,1s+m+1(Rn, E2) −−−−−−−→as−1,p,1(D)
continuous Bp,1s+1(Rn, E0)
it follows thatas−1,p,1(D) :Bp,∞s+m(Rn, E2)→Bp,∞s (Rn, E0) is conti- nouos. To proveb), letu∈V(Rn, E2). Then there exists a (s, p, q)∈ R×[1,∞]×[1,∞) such thatu∈Bp,qs+m(Rn, E2). From this, (27) and a) assertion b) follows.
Corollary 3.2. Let m, s∈R,1≤p, q≤ ∞ and B ∈{B,B, b}. Then˚
a7→a(D))] ∈ L
S1,0m,ρn(Rn, E1),L Bm+sp,q (Rn, E2),Bp,qs (Rn, E0) with
a](D)u Bs
p,q(Rn;E0)≤C3kakSm,n+1
1,0 (Rn;E1)kukBs+m p,q (Rn;E2)
for all u∈ Bp,qs+m(Rn;E2), where C3 = 3(2π)−n2C1.
3.1. Analytic semigroups. In this subsection we will assume thatE1 ,→ E0. Fors∈R,p, q∈[1,∞],m∈R+,B ∈{B,B, b}˚ anda∈S1,0m,ρn(Rn,L(E1, E0)), we denote A := a(D)]
Bs+m
p,q (Rn,E1) with domain D(A) =Bp,qs+m(Rn, E1) and E :=Bsp,q(Rn, E0). Since E1,→E0 andm >0, it holds that
D(A),→ Bp,qs+m(Rn, E0),→ Bp,qs (Rn, E0) =E.
From this and Corollary 3.2 it follows that
A:D(A)⊂E →E is a continuous linear operator. (28) These notations will be used in Proposition 3.5 and Corollary 3.6 to prove that the operator −A in (28) is sectorial and hence it generates an analytic semigroup on Bp,qs (Rn, E0), when the symbol a is parabolic in the sense of Definition 3.4. In the following we introduce the notation ρ(A) for the resolvent set of A,R(λ, A) := (λI−A)−1, with λ∈ρ(A), for the resolvent operator of A and
Lis(E1, E0) :={T ∈ L(E1, E0) :T is bijective}.
6A good summary of vector-valued Besov spaces and interpolation properties of these spaces can be found in [Am97], Chapter 5 and [Lu95], Chapter 1.
Proposition 3.3. Let M >0, r≥0, θ∈[0,2π]be constants, Stθ,r:=
n
λ=µeiθ:µ≥r o
a ray in C, and set
Aθ,M := {A∈ L(E1, E0) : (λI−A)∈ Lis(E1, E0) and P1
j=0(1 +|λ|)1−jkR(λ, A)kL(E
0,Ej)≤M for allλ∈Stθ,r}.
Then there exist constants εθ := ε(θ, M)>0, rθ >0 and C4 := C4(M) :=
M
M+1max{2M + 3,4M2 + 7M + 1} such that Σεθ,rθ
θ ⊂ ρ(A) holds for all A∈ Aθ,M where
Σεθ,rθ
θ :={λ∈C:|λ| ≥rθ and arg(λ)∈[θ−εθ, θ+εθ]}. Moreover,
1
X
j=0
(1 +|λ|)1−jkR(λ, A)kL(E
0,Ej)≤C4 for all λ∈Σεθ,rθ
θ and A∈ Aθ,M. Proof. First we will show two assertions.
Assertion I: Let θ = 0 in the hypotheses. Then there exist constants ε0 > 0, r0 > r and Mf:= M(2MM+1+3) so that Σε0,r00 ⊂ ρ(A) for all A ∈ A0,M and
kR(λ, A)kL(E
0)≤ Mf
1 +|λ| ∀λ∈Σε0,r00 and A∈ A0,M.
Proof of Assertion I: We select a r0 > r,ε0 := arcsin(2(M1+1)) and define Σ∗ε0,r0 :={λ∈C: Reλ≥r0 and |arg(λ)| ≤ε0}
and
s:= |λ|2
Reλ forλ∈Σ∗ε0,r0. (29) Then it follows from |λ| = |iReλ−Imλ| and λ−s = Imλ[iReReλ−Imλ λ] that for all λ∈Σ∗ε0,r0:
1
s|λ−s|= Reλ
|λ|2
|Imλ| |iReλ−Imλ|
Reλ =
Imλ
|λ|
=|sin(arg (λ))| ≤ 1 2 (M+ 1) (Note that ε0 ∈ 0,π6
and f(t) = sin (t) is increasing in −π6,π6
). Thus
|λ−s| ≤ 1 +s
2 (M+ 1) for all λ∈Σ∗ε0,r0. (30) It is clear due to the hypotheses that {s∈R:s≥r0} ⊂St0,r ⊂ρ(A) for all A∈ A0,M. For allλ∈Σ∗ε0,r0 and sas in (29) we get that
(λ−s) (sI−A)−1 L(E
0)
≤ |λ−s| M 1 +s ≤ 1
2 for all A∈ A0,M. From this and a Neumann series argument it follows that B := I + (λ− s)(sI −A)−1−1
∈ L(E0) and kBkL(E
0) ≤ 2. On the other hand we have that
λI−A= (sI−A)h
I+ (λ−s) (sI−A)−1i
= (sI−A)B−1.
Therefore (λI−A)−1∈ L(E0) and kR(λ, A)kL(E
0)≤ kBkL(E
0)kR(s, A)kL(E
0)
≤ 2M 1 +s
= 2M
1 +|λ|
|λ−s+s|+ 1 1 +s
≤ 2M 1 +|λ|
|λ−s|
1 +s + 1
≤ Mf
1 +|λ| for all λ∈Σ∗ε0,r0 and A∈ A0,M
with Mf := M(2MM+1+3). Choosing er0 ≥ r0 appropriately, we obtain a sector Σε0,0
er0 ⊂Σ∗ε0,r0 which finishes the proof of Assertion I.
Assertion II: An analogous result to Assertion I also is valid if the hypotheses are fulfilled for aθ∈(0,2π).
Proof of Assertion II: Letθ∈(0,2π) andO:C→Cdefined byO(z) :=
e−iθz,z∈C, a rotation function. It is clear thatOis bijective andO−1(z) = eiθz, z ∈C. By hypotheses one knows that Stθ,r ⊂ρ(A) for all A ∈ Aθ,M and furthermore
kR(λ, A)kL(E
0) ≤ M
1 +|λ| for allλ∈Stθ and A∈ Aθ,M. Thereforee−iθA∈ A0,M for allA∈ Aθ,M, becauseR µ, e−iθA
=eiθR(λ, A) for all λ=µeiθ ∈Stθ ⊂ρ(A), and
1
X
j=0
(1 +|µ|)1−j R
µ, e−iθA L(E
0,Ej) =
1
X
j=0
(1 +|λ|)1−jkR(λ, A)kL(E
0,Ej)≤M for all λ∈Stθ,r and A∈ Aθ,M. Thus, we get from the Assertion I that
R
w, e−iθA L(E
0)≤ Mf
1 +|w| for allw∈Σε0,r0
0 and A∈ Aθ,M. Now, let Σεθ,rθ
θ :=O−1 Σε0,r00
. Then for eachλ∈Σεθ,rθ
θ (and hence e−iθλ∈ Σε0,r00) and each A ∈ Aθ,M it holds that R(λ, A) = e−iθR e−iθλ, e−iθA
∈ L(E0). Then Σεθ,rθ
θ ⊂ρ(A) and kR(λ, A)kL(E
0) ≤ Mf
1 +|λ| for all λ∈Σεθ,rθ
θ, A∈ Aθ,M.
Consequently Assertion II is proven. In order to finish the proof of this proposition it is sufficient to show that
kR(λ, A)kL(E
0,E1)≤C4 for all λ∈Σεθ,rθ
θ and A∈ Aθ,M. (31) There C4 := maxn
M , Mf
1 + 2Mfo
. For it, let A ∈ Aθ,M,λ ∈Σεθ,rθ
θ and λθ :=rθeiθ ∈Stθ,r. From|λθ|=rθ ≤ |λ|and
x= (λθ−A)−1[(λ−A)x+ (λθ−λ)x] for all x∈E1
it follows that kxkE
1 ≤ kR(λθ, A)kL(E
0,E1)
k(λ−A)xkE
0+|λθ−λ| kxkE
0
(32)
≤M
k(λ−A)xkE
0 + 2|λ| kxkE
0
(due to definition ofAθ,M).
Moreover, it follows from Assertion II that kR(λ, A)ykE
0 ≤ Mf
1 +|λ|kykE
0 for all y∈E0, A∈ Aθ,M andλ∈Σεθ,rθ
θ, and therefore for all x∈E1 (and hence (λ−A)x∈E0) we obtain
kxkE
0 =kR(λ, A) [(λ−A)x]kE
0 ≤ Mf
1 +|λ|k(λ−A)xkE
0
for all A∈ Aθ,M, λ∈Σεθ,rθ
θ. From this and (32) we get kxkE
1 ≤M
k(λ−A)xkE
0 + 2Mf |λ|
1 +|λ|k(λ−A)xkE
0
≤M
1 + 2Mf
k(λ−A)xkE
0
for all x ∈ E1, A ∈ Aθ,M and λ ∈ Σεθ,rθ
θ. As (λ−A) ∈ Lis(E1, E0), this
yields the assertion (31).
Definition 3.4. Let m∈ R+, ρ ∈N0 and a∈S1,0m,ρ(Rn,L(E1, E0)). Then the symbol a is called parabolic in S1,0m,ρ(Rn,L(E1, E0)) with constants ω andκ, if there are constantsω≥0andκ >0, so that for all(ξ, µ)∈Rn×R+0 with |ξ, µ| ≥ω andθ∈
−π2,π2
, a(ξ) +µmeiθI :E1 −→E0 is bijective, h
a(ξ) +µmeiθIi−1
∈ L(E0, E1)
and
h
a(ξ) +µmeiθIi−1
L(E0,E1)
≤κhξ, µi−m. (33) Here, hξ, µi:=
q
1 +|ξ|2+µ2 and |ξ, µ|:=
q
|ξ|2+µ2.
In the following proposition, we denote for R≥0 and 0< θ≤π Σθ,R :={λ∈C:|λ| ≥R and |arg(λ)| ≤θ}, and in his proof we will use the inequality
1 +tm ≤2 (1 +t)m for all t≥0, (34) where m ≥ 0 (this inequality is obtained from 1 ≤ (1 +t)m and tm ≤ (1 +t)m for allt≥0).
Proposition 3.5. Let s ∈ R, p, q ∈ [1,∞], m ∈ R+, ρn as in (3) and A ⊂ S1,0m,ρn(Rn,L(E1, E0)) be bounded. Moreover, assume that all a ∈ A are parabolic in S1,0m,ρn(Rn,L(E1, E0)) with the same constants ω ≥ 0 and κ >0, andA:= a(D)]
Bp,qs+m(Rn,E1),a∈ A. ThenΣπ
2,R⊂ρ(−A)forR=ωm and
(1 +|λ|)1−j(λ+a)−1 ∈S1,0−jm,ρn(Rn,L(E0, Ej)), j= 0,1, (35)