variable order subordination in the sense of bochner and pseudo-differential operators
TRANSCRIPT
Math. Nachr. 284, No. 8–9, 987 – 1002 (2011) / DOI 10.1002/mana.200810246
Variable order subordination in the sense of Bochnerand pseudo-differential operators
Kristian Evans∗ and Niels Jacob∗∗
Department of Mathematics, Swansea University, Swansea SA2 8PP, UK
Received 3 November 2008, revised 30 March 2009, accepted 1 May 2009Published online 9 May 2011
Key words Subordination in the sense of Bochner, pseudo-differential operators of variable order, Fellersemigroups
MSC (2010) 47006, 47007, 47G30, 35S05, 60J35
For a large class of pseudo-differential operators with a negative definite symbol q(x, ξ) in the sense of Hohand for a large family of x-dependent Bernstein functions f (x, ·) we prove that the pseudo-differential operatorwith symbol −f (x, q(x, ξ)) has an extension generating a Feller semigroup.
c© 2011 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
1 Introduction
The main purpose of this note is to investigate subordination (in the sense of Bochner) of variable order.Bochner introduced a method called subordination which was used to obtain a new process from a given oneby a random time change. His original papers are [4] and [5]. We will however, study subordination using an an-alytic approach. There is a long history of constructing a functional calculus for generators of subordinate semi-groups, with the first general results obtained by Phillips [32]. Many calculi for this topic have been proposed,however, we only mention the paper of Berg, Boyadzhiev and de Laubenfels [3], Faraut [9], the monograph ofde Laubenfels [6] and the papers of Schilling [33] and [34]. We should note that Hirsch [13]–[14] had obtainedrelated results prior to this. The representation of fractional powers of generators is the best known result, this isdue to Balakrishnan [1], see also Yosida [38], Krasnosel’skii et al. [24] and Nollau [30]–[31]. At the root of ourwork is the result that for a continuous negative definite function ψ and a Bernstein function f , f ◦ ψ is also acontinuous negative definite function.
Subordination has also been studied on the level of pseudo-differential operators. In particular, in the casewhere −q(x,D) generates a Feller or sub-Markovian semigroup, in many cases −f(q(x, ξ)), f being a Bernsteinfunction, is also the symbol of a pseudo-differential operator generating a Feller or sub-Markovian semigrouprespectively.
This now leads us to the next step, i.e., subordination of variable order. By subordination of variable orderwe mean the case when we replace a fixed Bernstein function f by a family of Bernstein functions, f(x, ·)depending on x. Pseudo-differential operators of variable order of differentiation have already been studied byUnterberger and Bokobza [37], and in particular by Leopold [25], [26]. Feller semigroups obtained from the
symbol(1 + |ξ|2
)r(x)have been studied by Jacob and Leopold [21], where further work is due to Negoro [29],
in particular to Kikuchi and Negoro [22], [23]. It should also be noted that a Weyl-Hormander calculus can beused to consider operators of variable order of differentiation, see Baldus [2]. Moreover, Hoh [17] has shown thatwhen
f(x, q(x, ξ)) = (q(x, ξ))m (x)
∗ e-mail: [email protected], Phone: +44 1792 292905, Fax: +44 1792 295843∗∗ Corresponding author: e-mail: [email protected], Phone: +44 1792 295461, Fax: +44 1792 295843
c© 2011 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
988 K. Evans and N. Jacob: Variable order subordination
where 0 ≤ m(x) ≤ 1, then under certain conditions the pseudo-differential operator
−p(x,D)u = −(2π)−n2
∫Rn
eix·ξ (q(x, ξ))m (x) u(ξ) dξ
extends to the generator of a Feller semigroup. An approach using Dirichlet forms was recently proposed byUemura [35], [36].
The aim of this paper is to extend these ideas; we want to enlarge the class of examples and obtain a generalproof showing that the pseudo-differential −p(x,D) with symbol −f(x, q(x, ξ)) extends to the generator of aFeller semigroup, see our main result, Theorem 3.10. All our notations are standard and if not explained in thetext they can be found in [18]–[20].
2 Hoh’s symbolic calculus
Before starting with our main considerations we need to recollect some basic results from Hoh’s symbolic calcu-lus, see Hoh [15] or [16], compare also [19].
Definition 2.1 A continuous negative definite function ψ : Rn → R belongs to the class Λ if for all α ∈ N
n0
it satisfies ∣∣∂αξ (1 + ψ(ξ))
∣∣ ≤ c|α |(1 + ψ(ξ))2−ρ ( |α |)
2 , (2.1)
where ρ(k) = k ∧ 2 for k ∈ Nn0 .
Definition 2.2 A. Let m ∈ R and let ψ ∈ Λ. We then call a C∞-function q : Rn × R
n → C a symbol in theclass Sm,ψ
ρ (Rn ) if for all α, β ∈ Nn0 there are constants cα,β ≥ 0 such that
∣∣∂βx ∂α
ξ q(x, ξ)∣∣ ≤ cα,β (1 + ψ(ξ))
m −ρ ( |α |)2 (2.2)
holds for all x ∈ Rn and ξ ∈ R
n . We call m ∈ R the order of the symbol q(x, ξ).B. Let ψ ∈ Λ and suppose that for an arbitrarily often differentiable function q : R
n × Rn → C the estimate∣∣∂α
ξ ∂βx q(x, ξ)
∣∣ ≤ cα,β (1 + ψ(ξ))m2 (2.3)
holds for all α, β ∈ Nn0 and x, ξ ∈ R
n . In this case we call q a symbol of the class Sm,ψ0 (Rn ).
Note that Sm,ψρ (Rn ) ⊂ Sm,ψ
0 (Rn ). For q ∈ Sm,ψ0 (Rn ), hence also for q ∈ Sm,ψ
ρ (Rn ), we can define onS(Rn ) the pseudo-differential operator q(x,D) by
q(x,D)u(x) := (2π)−n2
∫Rn
eix·ξ q(x, ξ)u(ξ) dξ (2.4)
and we denote the classes of these operators by Ψm,ψρ (Rn ) and Ψm,ψ
0 (Rn ), respectively.
Theorem 2.3 If q ∈ Sm,ψ0 (Rn ) then q(x,D) maps S(Rn ) continuously into itself.
Let ψ : Rn → R be a fixed continuous negative definite function. For s ∈ R and u ∈ S(Rn ) (or u ∈ S′(Rn ))
we define the norm
‖u‖2ψ ,s =
∥∥(1 + ψ(D))s2 u∥∥2
0 =∫
Rn
(1 + ψ(s))s |u(ξ)|2dξ, (2.5)
where || · ||0 denotes the norm in L2(Rn ).The space Hψ,s(Rn ) is defined as
Hψ,s(Rn ) := {u ∈ S′(Rn ); ‖u‖ψ ,s < ∞}. (2.6)
The scale Hψ,s(Rn ), s ∈ Rn , and more general spaces have been systematically investigated in [10] and [11],
see also [19]. In particular we know that if for some ρ1 > 0 and c1 > 0 the estimate ψ(ξ) ≥ c1 |ξ|ρ1 holds for allξ ∈ R
n , |ξ| ≥ R, R ≥ 0, then the space Hψ,s(Rn ) is continuously embedded into C∞(Rn ) provided s > nρ1
.
c© 2011 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.mn-journal.com
Math. Nachr. 284, No. 8–9 (2011) / www.mn-journal.com 989
The following result is due to Hoh [15] and is of much importance, compare also [19], Theorem 2.4.23.A.
Theorem 2.4 Let ψ ∈ Λ. For q1 ∈ Sm 1 ,ψρ (Rn ) and q2 ∈ Sm 2 ,ψ
ρ (Rn ) the symbol q of the operatorq(x,D) := q1(x,D) ◦ q2(x,D) is given by
q(x, ξ) = q1(x, ξ) · q2(x, ξ) +n∑
j=1
∂ξjq1(x, ξ)Dxj
q2(x, ξ) + qr1 (x, ξ) (2.7)
with qr1 ∈ Sm 1 +m 2 −2,ψ0 (Rn ).
Remark 2.5 An easy calculation yields q1 · q2 ∈ Sm 1 +m 2 ,ψρ (Rn ), ∂ξj
q1 ∈ Sm 1 −1,ψρ (Rn ) and
Dxjq2 ∈ Sm 2 ,ψ
ρ (Rn ). Hence the second term on the right-hand side in (2.7) belongs to Sm 1 +m 2 −1,ψρ (Rn ).
On S(Rn ) we may define the bilinear form
B(u, v) := (q(x,D)u, v)0 , q ∈ Sm,ψρ (Rn ). (2.8)
Now using Hoh’s symbolic calculus we obtain estimates for the bilinear form, compare again Hoh [15] or see[19].
Theorem 2.6 Let q ∈ Sm,ψρ (Rn ) be real valued and m > 0. It follows that
|B(u, v)| ≤ c ‖u‖ψ , m2‖v‖ψ , m
2(2.9)
holds for all u, v ∈ S(Rn ). Hence the bilinear form B has a continuous extension onto Hψ, m2 (Rn ). If in addition
for all x ∈ Rn
q(x, ξ) ≥ δ0(1 + ψ(ξ))m2 for |ξ| ≥ R (2.10)
with some δ0 > 0 and R ≥ 0, and
lim|ξ |→∞
ψ(ξ) = ∞ (2.11)
holds, then we have for all u ∈ Hψ, m2 (Rn ) the Garding inequality
ReB(u, u) ≥ δ0
2||u||2ψ , m
2− λ0 ||u||20 . (2.12)
3 Subordination of variable order
The method discussed in this section improves the ideas of [8]. Let f : Rn × (0,∞) → R be an arbitrarily often
differentiable function such that for y ∈ Rn fixed the function s → f(y, s) is a Bernstein function. We assume
that with some 0 < r1 ≤ 1 we have
supy∈Rn
f(y, s) ≤ c1sr1 for s ≥ γ0 (3.1)
as well as for some 0 < r0 such that r0 < r1 it holds
infy∈Rn
f(y, s) ≥ c2sr0 for s ≥ γ0 . (3.2)
In our applications we will consider symbols f(x, q(x, ξ)) where q(x, ξ) ≥ λ0(1 + ψ(ξ)) for some real-valuedcontinuous negative definite function ψ. Thus we can always confine ourselves to the case where γ0 ≥ 1. Considerthe negative definite symbol
p(x, ξ) = f(x, q(x, ξ)) (3.3)
where the symbol q(x, ξ) is comparable with a fixed continuous negative definite function ψ satisfyinglimξ→∞ ψ(ξ) = ∞ , i.e.,
0 < c3 ≤ q(x, ξ)ψ(ξ)
≤ c4 , (3.4)
www.mn-journal.com c© 2011 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
990 K. Evans and N. Jacob: Variable order subordination
for all x ∈ Rn and ξ ∈ R
n . Note that the lower bounds imply ψ(ξ) > 0. Since ξ → ψ(ξ) − ψ(0) is also acontinuous negative definite function, the lower bound in (3.4) corresponds to an estimate c(1 + ψ(ξ)) ≤ q(x, ξ)for ψ being a continuous negative function which might have a zero. We find using Lemma 3.9.34.B in [18]
p(x, ξ) = f(x, q(x, ξ)) ≤ c1(1 + ψ(ξ))r1 (3.5)
and
p(x, ξ) ≥ c2(1 + ψ(ξ))r0 (3.6)
i.e., p(x, ξ) is bounded above and below by continuous negative definite functions.Under suitable assumptions on f , see below, the pseudo-differential operator
p(x,D)u(x) = (2π)−n2
∫Rn
eix·ξ p(x, ξ)u(ξ) dξ
= (2π)−n2
∫Rn
eix·ξ f(x, q(x, ξ))u(ξ) dξ (3.7)
has a symbol p ∈ S2r1 +2ε,ψρ (Rn ). The following section gives a detailed proof of this result.
3.1 Estimates for p(x, ξ)
Let f : Rn × (0,∞) → R be an arbitrarily often differentiable function such that for each x ∈ R
n the functionf(x, ·) : (0,∞) → R is a Bernstein function. For every Bernstein function h : (0,∞) → R the estimates
∣∣h(k)(s)∣∣ ≤ k!
skh(s), s > 0 and k ∈ N0 (3.8)
hold, compare [18], Lemma 3.9.34.D. Hence for f as above we find
∣∣f (k)(x, s)∣∣ ≤ k!
skf(x, s), s > 0, x ∈ R
n and k ∈ N0 , (3.9)
where
f (k)(x, s) =∂kf(x, s)
∂sk.
We assume now in addition:There exists η > 0 and δ0 > 0 such that for ε ∈ (0, η) and for all s ≥ δ0 it follows that
∣∣∂αx ∂k
s f(x, s)∣∣ ≤ cα,k,ε
1sk
f(x, s)sε (3.10)
holds for all x ∈ Rn and s ≥ δ0 with cα,k,ε independent of x and s.
Example 3.1 (Compare Hoh [17]) Consider f(x, s) = sm (x) for 0 < m ≤ m(x) ≤ M < 1. It follows that
∂ks sm (x) = Pk (m(x))
1sk
sm (x) (3.11)
where Pk (t) is a polynomial of degree less or equal to k. If we assume in addition that m(·) ∈ C∞(Rn ) and|∂αm(x)| ≤ mα for all α ∈ N
n0 we find using (3.11) that
∂αx ∂k
s sm (x) = ∂αx
(Pk (m(x))
1sk
sm (x))
=1sk
∑β≤α
(αβ
)∂α−β Pk (m(x))
∑β1 +···+βl ′=β
l′=1,...,|β |
c{βj }
l∏j=1
(∂βj (m(x) ln s)
)sm (x) .
c© 2011 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.mn-journal.com
Math. Nachr. 284, No. 8–9 (2011) / www.mn-journal.com 991
Thus we arrive at
∣∣∂αx ∂k
s sm (x)∣∣ ≤ cα,k,m
Q(ln s)sk
sm (x)
provided s ≥ 1 (otherwise, if s ≥ δ0 > 0, we need to treat the terms involving ln s a bit differently, for examplewe may switch to | ln s|) where Q is a suitable polynomial. Since for ε > 0 we find a constant cε such thatQ(ln s) ≤ cεs
ε holds, we arrive at (3.10).
Example 3.2 Consider f(x, s) = sm (x )
2
(1− e−4s
m (x )2)
where m : Rn → R is arbitrarily often differentiable
such that 0 < m ≤ m(x) ≤ M < 2 and∣∣∂α
x m(x)∣∣ ≤ mα . A straight forward calculation, compare [7] shows
that for all α ∈ Nn0 and k ∈ N0 there exists ε > 0, 0 < ε < η, such that∣∣∣∣∂α
x ∂ks s
m (x )2
(1 − e−4s
m (x )2
)∣∣∣∣ ≤ cα,k,ε1sk
sm (x )+ ε
2
(1 − e−4s
m (x )2
). (3.12)
We now return to the general case. It should be noted here that due to their length calculations may have to besplit over many lines. To avoid any confusion if a calculation is written as∑
×∑
it means ∑∑and not (∑)(∑)
.
Eventually we need various controls on the symbol
(x, ξ) −→ f(x, q(x, ξ))
where q(x, ξ) comes from a certain symbol class which we will fix later. For this we use a formula to calculatehigher order derivatives of composed functions which is due to Fraenkel [12], compare also [18], p. 15.
Let u : Rm → C and vj : R
n → R, j = 1, . . . ,m, be smooth functions. Then for α ∈ Nn0 it holds with
v = (v1 , . . . vm )
∂αu(v(x)) = ∂αu(v1(x), . . . vm (x)) (3.13)
=∑
1≤|σ |≤|α |σ∈N
m0
(∂σu)(v(x))σ!
∑γ 1 +···+γ m =α
γ j ∈Nn0
Pγ 1 (σ1 , v1 ;x) · . . . · Pγ m (σm , vm ;x)
where for γ ∈ Nn0
Pγ (N, v;x) :=∑
ρ∈R(γ ,N )
N !ρ!
(∂β (1)v(x)
β(1)!
)ρ1
· . . . ·(
∂β (r)v(x)β(r)!
)ρr
(3.14)
with
R(γ,N) :=
⎧⎨⎩ρ ∈ N
r0
∣∣∣∣∣∣r∑
j=1
ρjβ(j) = γ and |ρ| = N
⎫⎬⎭ , (3.15)
Nn0,γ :=
{β ∈ N
n0 | 0 < β ≤ γ
}(3.16)
www.mn-journal.com c© 2011 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
992 K. Evans and N. Jacob: Variable order subordination
and with |Nn0,γ | = r an enumeration of N
n0,γ is given by β(1), . . . , β(r). In our concrete problem many reductions
happen.We consider first f : R
n × (0,∞) → R artificial as f : Rn × R
n × (0,∞) → R by setting f(x, s) =f(x1 , . . . , xn , 1, . . . , 1, s). Next we introduce the 2n + 1 functions
vj (x, ξ) =
⎧⎨⎩
xj , 1 ≤ j ≤ n,1, n + 1 ≤ j ≤ 2n,q(x, ξ), j = 2n + 1.
(3.17)
In the following multi-indices in N2n0 will be split as α =
(α(1) , α(2)
)where α(1) acts on the x-variables and
α(2) acts on the ξ variables. Our problem is to estimate
∂βx ∂α
ξ f(v1(x, ξ), . . . , v2n+1(x, ξ))
= ∂βx ∂α
ξ f(x1 , . . . , xn , 1, . . . , 1, q(x1 , . . . , xn , ξ1 , . . . ξn ))
=∑
1≤|σ |≤|α |+|β |σ∈N
2 n + 10
(∂σf)(v(x, ξ))σ!
×∑
γ 1 +···+γ 2 n + 1 =ω
γ j ∈N2 n0
Pγ 1 (σ1 , v1 ;x, ξ) · . . . · Pγ 2 n + 1 (σ2n+1 , v2n+1;x, ξ) (3.18)
where ω = (β, α). If γj =(δj1 , δ
j2
)then
N2n0,γ j =
{(ζ, τ) ∈ N
2n0 | |ζ| + |τ | > 0 and 0 < ζ ≤ δj
1 , 0 < τ ≤ δj2
}.
Let an enumeration of N2n0,γ j : η(1) = (ζ(1), τ(1)), . . . , η(rj ) = (ζ(rj ), τ(rj )) where rj = rj (γj ) be given.
Then we have with σ = (σ1 , . . . , σ2n+1)
R(γj , σj
)=
{ρ ∈ N
rj
0
∣∣∣∣∣rj∑
l=1
ρlη(l) = γj and |ρ| = σj
}
=
{ρ ∈ N
rj
0
∣∣∣∣∣rj∑
l=1
ρl(ζ(l), τ(l)) = γj and |ρ| = σj
}
=
{ρ ∈ N
rj
0
∣∣∣∣∣rj∑
l=1
ρlζ(l) = δj1 ,
rj∑l=1
ρlτ(l) = δj2 and |ρ| = σj
}
and
Pγ j (σj , vj , x, ξ) =∑
ρ∈R(γ j ,σj )
σj !ρ!
(∂
ζ (1)x ∂
τ (1)ξ vj (x, ξ)
ζ(1)!τ(1)!
)ρ1
· . . . ·(
∂ζ (rj )x ∂
τ (rj )ξ vj (x, ξ)
ζ(rj )!τ(rj )!
)ρr j
where σ ∈ N2n+10 is such that σ = (σx, σξ , σ2n+1), σx, σξ ∈ N
n0 and σ2n+1 ∈ N0 .
When σξ �= 0 then ∂σf = 0 therefore∑1≤|σ |≤|α |+|β |
σ∈N2 n + 10
reduces to ∑1≤|σx |+σ2 n + 1 ≤|α |+|β |
σ∈N2 n + 10 , σξ =0
Consider Pγ j (σj , vj , x, ξ)
c© 2011 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.mn-journal.com
Math. Nachr. 284, No. 8–9 (2011) / www.mn-journal.com 993
(i) 1 ≤ j ≤ n implies
∂ζ (k)x ∂
τ (k)ξ vj (x, ξ) = ∂ζ (k)
x ∂τ (k)ξ xj =
{1, ζ(k) = εj , τ(k) = 00, otherwise
therefore for 1 ≤ j ≤ n
Pγj(σj , vj , x, ξ) =
∑ρ∈R(γ j ,σj )
σj !ρ!
(∂
ζ (1)x ∂
τ (1)ξ vj (x, ξ)
ζ(1)!τ(1)!
)ρ1
· . . . ·(
∂ζ (rj )x ∂
τ (rj )ξ vj (x, ξ)
ζ(rj )!τ(rj )!
)ρr j
=∑
ρ∈{
ρ′∈Nr j0
∣∣∑ r jl = 1 ρ′
l ζ (l)=σj and ζ (l)=εj} σj !
ρ!
(∂
ζ (1)x ∂
τ (1)ξ vj (x, ξ)
ζ(1)!τ(1)!
)ρ1
· . . . ·(
∂ζ (rj )x ∂
τ (rj )ξ vj (x, ξ)
ζ(rj )!τ(rj )!
)ρr j
,
i.e., in this case Pγj(σj , vj , x, ξ) = cj,σ ;
(ii) n + 1 ≤ j ≤ 2n implies
∂ζ (k)x ∂
τ (k)ξ vj (x, ξ) = 0 whenever ζ(k) �= 0 or τ(k) �= 0, i.e., (ζ(k), τ(k)) �= 0,
i.e., for n + 1 ≤ j ≤ 2n
Pγj(σj , vj , x, ξ) =
∑ρ∈R(γ j ,σj )
σj !ρ!
(∂
ζ (1)x ∂
τ (1)ξ vj (x, ξ)
ζ(1)!τ(1)!
)ρ1
· . . . ·(
∂ζ (rj )x ∂
τ (rj )ξ vj (x, ξ)
ζ(rj )!τ(rj )!
)ρr j
= 0;
(iii) finally let j = 2n + 1 and set r = rj , i.e.,
R(γ2n+1 , σ2n+1
)=
{ρ ∈ N
r0
∣∣∣∣∣r∑
l=1
ρl(ζ(l), τ(l)) = γ2n+1 , |ρ| = σ2n+1
}
then
Pγ 2 n + 1 (σ2n+1 , v2n+1 , x, ξ) = Pγ 2 n + 1 (σ2n+1 , q(x, ξ))
=∑
ρ∈R(γ 2 n + 1 ,σ2 n + 1 )
σ2n+1!ρ!
(∂
ζ (1)x ∂
τ (1)ξ q(x, ξ)
ζ(1)!τ(1)!
)ρ1
· . . . ·(
∂ζ (rj )x ∂
τ (rj )ξ q(x, ξ)
ζ(rj )!τ(rj )!
)ρr j
.
We observe that
(∂σf)(x1 , . . . xn , s) = 0 if σξ �= 0 ∈ Nn0 .
www.mn-journal.com c© 2011 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
994 K. Evans and N. Jacob: Variable order subordination
Thus we find using the previous calculations and (3.18)
∂βx ∂α
ξ f(x, q(x, ξ)) =∑
1≤|σx |+σ2 n + 1 ≤|α |+|β |σ=(σx ,0,σ2 n + 1 )∈N
2 n + 10
(∂σf
σ!
)(x, q(x, ξ))
×∑
γ 1 +···+γ 2 n + 1 =(β ,α)
γ j ∈N2 n0
Pγ 1 (σ1 , x1) · . . . · Pγ 2 n + 1 (σ2n+1 , q(x, ξ))
=∑
1≤|σx |+σ2 n + 1 ≤|α |+|β |σ=(σx ,0,σ2 n + 1 )∈N
2 n + 10
(∂σf
σ!
)(x, q(x, ξ))
×∑
δ 11 +···+δ 2 n + 1
1 =β
δ 12 +···+δ 2 n + 1
2 =α
δjk ∈N
n0
P(δ 11 ,δ 1
2 )(σ1 , x1) · . . . · P(δ 2 n + 11 ,δ 2 n + 1
2 )(σ2n+1 , q(x, ξ))
=∑
1≤|σx |+σ2 n + 1 ≤|α |+|β |σ=(σx ,0,σ2 n + 1 )∈N
2 n + 10
(∂σf
σ!
)(x, q(x, ξ))
×∑
δ 11 +···+δn
1 +δ 2 n + 11 =β
δ 12 +···+δn
2 +δ 2 n + 12 =α
δjk ∈N
n0 ,δ j
k =0, n+1≤j≤2n
P(δ 11 ,δ 1
2 )(σ1 , x1) · . . . · P(δ 2 n + 11 ,δ 2 n + 1
2 )(σ2n+1 , q(x, ξ))
=∑
1≤|σx |+σ2 n + 1 ≤|α |+|β |σ=(σx ,0,σ2 n + 1 )∈N
2 n + 10
(∂σf
σ!
)(x, q(x, ξ))
×∑
δ 11 +···+δn
1 +δ 2 n + 11 =β
δ 2 n + 12 =α
δjk ∈N
n0 ,δ j
k =0, n+1≤j≤2n
δj1 ∈{0,εj }, 1≤j≤n, δ j
2 =0, 1≤j≤n
P(δ 11 ,δ 1
2 )(σ1 , x1) · . . . · P(δ 2 n + 11 ,δ 2 n + 1
2 )(σ2n+1 , q(x, ξ))
=∑
1≤|σx |+σ2 n + 1 ≤|α |+|β |σ=(σx ,0,σ2 n + 1 )∈N
2 n + 10
(∂σf
σ!
)(x, q(x, ξ))
×∑
δ 11 +···+δn
1 +δ 2 n + 11 =β
δ 2 n + 12 =α
δjk ∈N
n0 ,δ j
k =0, n+1≤j≤2n
δj1 ∈{0,εj }, 1≤j≤n, δ j
2 =0, 1≤j≤n
C(δ11 , . . . , δn
1 , δ12 , . . . , δn
2)P(δ 2 n + 1
1 ,α)(σ2n+1 , q(x, ξ))
=∑
1≤|σx |+σ2 n + 1 ≤|α |+|β |σ=(σx ,0,σ2 n + 1 )∈N
2 n + 10
(∂σf
σ!
)(x, q(x, ξ))
c© 2011 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.mn-journal.com
Math. Nachr. 284, No. 8–9 (2011) / www.mn-journal.com 995
×∑
δ 11 +···+δn
1 +δ 2 n + 11 =β
δjk ∈N
n0 ,δ j
k =0, n+1≤j≤2n
δj1 ∈{0,εj }, 1≤j≤n, δ j
2 =0, 1≤j≤n
C(δ11 , . . . , δn
1)
×∑
ρ∈R((δ 2 n + 11 ,α),σ2 n + 1 )
σ2n+1!ρ!
(∂
ζ (1)x ∂
τ (1)ξ q(x, ξ)
ζ(1)!τ(1)!
)ρ1
· . . . ·(
∂ζ (rj )x ∂
τ (rj )ξ q(x, ξ)
ζ(rj )!τ(rj )!
)ρr 2 n + 1
.
In order to estimate ∂βx ∂α
ξ f(x, q(x, ξ)) we assume further∣∣∂ζx∂τ
ξ q(x, ξ)∣∣ ≤ cζ τ (1 + ψ(ξ))
2−ρ ( |τ |)2 (3.19)
and the ellipticity condition
q(x, ξ) ≥ γ0(1 + ψ(ξ)). (3.20)
Taking (3.10) into account we find for every ε > 0 but sufficiently small that∣∣∂βx ∂α
ξ f(x, q(x, ξ))∣∣ ≤ C ′
α,β ,ε
∑1≤|σx |+σ2 n + 1 ≤|α |+|β |σ=(σx ,0,σ2 n + 1 )∈N
2 n + 10
1q(x, ξ)σ2 n + 1
f(x, q(x, ξ))q(x, ξ)ε
×∑
δ 11 +···+δn
1 +δ 2 n + 11 =β
δjk ∈N
n0 ,δ j
k =0, n+1≤j≤2n
δj1 ∈{0,εj }, 1≤j≤n, δ j
2 =0, 1≤j≤n
×∑
ρ∈R((δ 2 n + 11 ,α),σ2 n + 1 )
(1 + ψ(ξ))(2−( 2∧τ ( 1 ) ) )
2 ρ1 · . . . · (1 + ψ(ξ))(2−( 2∧τ ( r ) ) )
2 ρr 2 n + 1
≤ C ′α,β ,ε
∑1≤|σx |+σ2 n + 1 ≤|α |+|β |σ=(σx ,0,σ2 n + 1 )∈N
2 n + 10
1(1 + ψ(ξ))σ2 n + 1
f(x, q(x, ξ))q(x, ξ)ε
×∑
δ 11 +···+δn
1 +δ 2 n + 11 =β
δjk ∈N
n0 ,δ j
k =0, n+1≤j≤2n
δj1 ∈{0,εj }, 1≤j≤n, δ j
2 =0, 1≤j≤n
×∑
ρ∈R((δ 2 n + 11 ,α),σ2 n + 1 )
(1 + ψ(ξ))ρ1 +···+ρr 2 n + 1 (1 + ψ(ξ))−( (2∧τ ( 1 ) )2 ρ1 +···+ (2∧τ ( r ) )
2 ρr 2 n + 1 ).
Since for ρ ∈ R((
δ2n+11 , δ2n+1
2
), σ2n+1
), it follows that ρ1 + · · · + ρr2 n + 1 = σ2n+1 and we arrive at∣∣∂β
x ∂αξ f(x, q(x, ξ))
∣∣ ≤ Cα,β ,εf(x, q(x, ξ))q(x, ξ)ε∑
1≤|σx |+σ2 n + 1 ≤|α |+|β |σ=(σx ,0,σ2 n + 1 )∈N
2 n + 10
×∑
δ 11 +···+δn
1 +δ 2 n + 11 =β
δjk ∈N
n0 ,δ j
k =0, n+1≤j≤2n
δj1 ∈{0,εj }, 1≤j≤n, δ j
2 =0, 1≤j≤n
×∑
ρ∈R((δ 2 n + 11 ,α),σ2 n + 1 )
(1 + ψ(ξ))−12 ((2∧τ (1))ρ1 +···+(2∧τ (r))ρr ) .
www.mn-journal.com c© 2011 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
996 K. Evans and N. Jacob: Variable order subordination
Denote the last sum by UR((δ 2 n + 11 ,δ 2 n + 1
2 ),σ2 n + 1 ), i.e.,
U(ψ )R((δ 2 n + 1
1 ,δ 2 n + 12 ),σ2 n + 1 ) =
∑ρ∈R((γ 2 n + 1
1 ,γ 2 n + 12 )σ2 n + 1 )
(1 + ψ(ξ))−12 ((2∧τ (1))ρ1 +···+(2∧τ (r))ρr ) .
If |α| = 1 we get a contribution from R((
δ2n+11 , δ2n+1
2
), 1)
of the type |τ(l)| = 1, ρl = 1, and we have theestimate ∣∣∣∣U (ψ )
R((δ 2 n + 11 ,δ 2 n + 1
2 ),1)
∣∣∣∣ ≤ (1 + ψ(ξ))−12 = (1 + ψ(ξ))−
12 (2∧|α |) .
If |α| = 2 we get at least one contribution from R((
δ2n+11 , δ2n+1
2
), 2)
of the type |τ(l)| = 2, ρl = 1 or |τ(l)| =|τ(k)| = 1 and ρl = ρk = 1, l �= k, hence we get that the estimate∣∣∣∣U (ψ )
R((δ 2 n + 11 ,δ 2 n + 1
2 ),2)
∣∣∣∣ ≤ (1 + ψ(ξ))−1 = (1 + ψ(ξ))−12 (2∧|α |)
holds. Finally, for |α| ≥ 3 we find the estimate by analysing the possible terms in R((
γ2n+11 , γ2n+1
2
)σ2n+1
)for
σ2n+1 ≥ 3 ∣∣∣∣U (ψ )R((δ 2 n + 1
1 ,δ 2 n + 12 ),k)
∣∣∣∣ ≤ (1 + ψ(ξ))−12 (2∧|α |)
with k ≥ 3. Thus we have proved
Theorem 3.3 Suppose that f : Rn×(0,∞) → R is arbitrarily often differentiable and that f(x, ·) : (0,∞) →
R is a Bernstein function. Suppose further that (3.10) holds. Let q : Rn × R
n → R be an arbitrarily oftendifferentiable function satisfying with a fixed continuous negative definite function ψ : R
n → R conditions (3.19)and (3.20). Then for every ε > 0 sufficiently small and all α, β ∈ N
n0 it holds∣∣∂β
x ∂αξ f(x, q(x, ξ))
∣∣ ≤ Cα,β ,εf(x, q(x, ξ))q(x, ξ)ε(1 + ψ(ξ))−12 ρ(|α |) . (3.21)
Now, Theorem 3.3 implies together with Theorem 2.5.4 in [19] and (3.1) that p ∈ S2r1 +2ε,ψρ (Rn ) and that
p(x,D) maps the space Hψ,2r1 +2ε+s(Rn ) to the space Hψ,s(Rn ) i.e.,
||p(x,D)u||ψ ,s ≤ c ||u||ψ ,2r1 +2ε+s .
In particular p(x,D) is continuous from Hψ,s(Rn ) to Hψ,s−2r1 −2ε(Rn ) i.e.,
||p(x,D)u||ψ ,s−2r1 −2ε ≤ c′||u||ψ ,s .
We now consider the bilinear form
B(u, v) := (p(x,D)u, v)0 , u, v ∈ S(Rn ).
Since p ∈ S2r1 +2ε,ψρ (Rn ) we may apply Theorem 2.6 to get
|B(u, v)| ≤ κ||u||ψ ,r1 +ε ||v||ψ ,r1 +ε , (3.22)
for some κ > 0 and all u, v ∈ S(Rn ) i.e., B has a continuous extension onto Hψ,r1 +ε(Rn ) again denoted by B.Furthermore we have
Proposition 3.4 For u ∈ Hψ,2r1 +2ε(Rn ), r1 + ε − 12 < r0 , we have the Garding inequality
B(u, u) ≥ δ1 ||u||2ψ ,r0− λ0 ||u||20 , (3.23)
for some λ0 ≥ 0.
c© 2011 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.mn-journal.com
Math. Nachr. 284, No. 8–9 (2011) / www.mn-journal.com 997
P r o o f. We have the lower bound
p(x, ξ) ≥ c(1 + ψ(ξ))r0 .
This implies
r(x, ξ) = p(x, ξ) − c(1 + ψ(ξ))r0 ≥ 0.
Now Theorem 2.5.5 in [19], which is due to Hoh, see [15], gives
(r(x,D)u, u)0 ≥ −k||u||22 r 1 + 2 ε−12
(3.24)
since (r(x,D)u, u)0 is real-valued, where
(r(x,D)u, u)0 = ReB(u, u) − c||u||2ψ ,r0. (3.25)
Putting (3.24) and (3.25) together we get
ReB(u, u) ≥ c ||u||2ψ ,r0− k||u||22 r 1 + 2 ε−1
2.
Under the assumption that
ψ(ξ) ≥ c0 |ξ|ρ0 (3.26)
and r1 + ε − 12 < r0 we get for every ε0 > 0
(1 + ψ(ξ))2 r 1 + 2 ε−1
2 ≤ ε20(1 + ψ(ξ))r0 + c2(ε0)
which leads to
||u||22 r 1 + 2 ε−12
≤ ε0 ||u||2ψ ,r0+ c(ε0)||u||20
implying the result.
We are now dealing with the space Hψ,r0 (Rn ) and the space Hψ,r1 +ε(Rn ) which is the smaller of thetwo. Since our estimates for B are in different spaces we seek to introduce an intermediate space, compareLouhivaara and Simader [27] and [28].
Firstly, we consider the symmetric part B of B i.e.,
B(u, v) =12(B(u, v) + B(v, u)),
on Hψ,r1 +ε(Rn ). Then
Bλ0 (u, v) := B(u, v) + λ0(u, v)0 , (3.27)
We have
|Bλ0 (u, v)| ≤ κ||u||ψ ,r1 +ε ||v||ψ ,r1 +ε
and
Bλ0 (u, u) ≥ γ||u||2ψ ,r0.
Since Bλ0 (u, v) is a scalar product on Hψ,r1 +ε(Rn ) we may consider the completion of Hψ,r1 +ε(Rn ) withrespect to Bλ0 (u, v). We denote this new intermediate space by Hpλ 0 (Rn ) . Clearly we have
Hψ,r1 +ε(Rn ) ↪→ Hpλ 0 (Rn ) ↪→ Hψ,r0 (Rn ) (3.28)
in the sense of continuous embeddings.
www.mn-journal.com c© 2011 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
998 K. Evans and N. Jacob: Variable order subordination
Lemma 3.5 The bilinear form Bλ0 is continuous on Hpλ 0 (Rn ), provided r1 + ε − 12 < r0 .
P r o o f. We find by using Corollary 2.4.23 in [19] that
12(pλ0 (x,D) + p∗λ0
(x,D)) =12(pλ0 (x,D) + pλ0 (x,D)) + r1(x,D)
= pλ0 (x,D) + r1(x,D)
where r1 ∈ S2r1 +2ε−1,ψρ (Rn ) and we used that p(x, ξ) is real-valued. Consider
|Bλ0 (u, v)| = |(pλ0 (x,D))u, v)0 |
≤ 12|((pλ0 (x,D) + p∗λ0
(x,D))u, v)0 | + |(r1(x,D)u, v)0 |
= |Bλ0 (u, v)| + |(r1(x,D)u, v)0 |.
We know that Bλ0 (u, v) is continuous on Hpλ 0 (Rn ) therefore our calculations are reduced to estimating|(r1(x,D)u, v)0 |.We know that r1 ∈ S2r1 +2ε−1,ψ
ρ (Rn ) which implies by Theorem 2.6 that
|(r1(x,D)u, v)0 | ≤ c ||u||ψ ,r1 +ε− 12||v||ψ ,r1 +ε− 1
2.
If r1 + ε − 12 < r0 we get
||u||ψ ,r1 +ε− 12≤ ||u||ψ ,r0 ≤ c||u||pλ 0
implying the result by (3.28).
By the Lax-Milgram theorem, for every g ∈ (Hpλ 0 )∗ ⊂ S′(Rn ) there exists a unique element u ∈ Hpλ 0
satisfying
Bλ0 (u, v) =⟨g, v⟩
(3.29)
for all v ∈ Hpλ 0 . We call u the variational solution to p(x,D)u + λ0u = g. From (3.28) we get
Hψ,−r0 (Rn ) = (Hψ,r0 (Rn ))∗ ↪→ (Hpλ 0 (Rn ))∗,
hence for g ∈ Hψ,−r0 (Rn ) there exists a unique u ∈ Hpλ 0 (Rn ) satisfying (3.29).
Proposition 3.6 For every g ∈ Hψ,−r0 (Rn ) there exists a unique u ∈ Hψ,r0 (Rn ) such that
pλ0 (x,D)u = p(x,D)u + λ0u = g (3.30)
holds as an equality in S′(Rn ).
P r o o f. Denote by u ∈ Hpλ 0 (Rn ) the unique solution to (3.30) for g ∈ Hψ,r0 (Rn ) given and take a sequence(uk )k∈N, uk ∈ S(Rn ) converging in Hpλ 0 (Rn ) to u.
It follows from
(pλ0 (x,D)uk , v)0 = Bλ0 (uk , v), v ∈ S(Rn )
and the continuity of pλ0 (x,D) from Hψ,s(Rn ) to Hψ,s−2r1 −2ε(Rn ) that for k → ∞⟨pλ0 (x,D)u, v
⟩= Bλ0 (u, v) =
⟨g, v⟩
for all v ∈ S(Rn ). Thus pλ0 (x,D)u = g in S′(Rn ) and the uniqueness follows from (3.23).
In order to get more regularity for variational solutions we have
c© 2011 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.mn-journal.com
Math. Nachr. 284, No. 8–9 (2011) / www.mn-journal.com 999
Theorem 3.7 Assume (3.10), (3.19), (3.20) and (3.2), i.e., for some γ > 0
γ(1 + ψ(ξ))r0 ≤ f0(γ0(1 + ψ(ξ))) (3.31)
holds where
f0(s) := infx∈Rn
f(x, s). (3.32)
Then for every η > 0 sufficiently small the function p−1λ (x, ξ) belongs to the class S−2r0 +2η ,ψ
ρ (Rn ).
P r o o f. Let us assume for simplicity that δ0 = γ0 , compare (3.10). For λ > 0 let pλ (x, ξ) = λ+f(x, q(x, ξ)).From (2.27) in [18] we find with l = |α| + |β|∣∣∣∣∂α
ξ ∂βx
1pλ (x, ξ)
∣∣∣∣ ≤ 1pλ (x, ξ)
∑α1 +···+αl =α
β 1 +···+β l =β
c{αj ,β j }
l∏j=1
∣∣∣∣∣∂αj
ξ ∂β j
x pλ(x, ξ)pλ(x, ξ)
∣∣∣∣∣
=1
λ + f(x, q(x, ξ))
∑α1 +···+αl =α
β 1 +···+β l =β
c{αj ,β j }
l∏j=1
∣∣∣∣∣∂αj
ξ ∂β j
x pλ (x, ξ)pλ(x, ξ)
∣∣∣∣∣ .
From the definition of f1(s), (3.32), together with (3.20) we get
∣∣∣∣∂αξ ∂β
x
1pλ (x, ξ)
∣∣∣∣ ≤ 1f1(γ0(1 + ψ(ξ)))
∑α1 +···+αl =α
β 1 +···+β l =β
c{αj ,β j }
l∏j=1
∣∣∣∣∣∂αj
ξ ∂β j
x pλ(x, ξ)pλ (x, ξ)
∣∣∣∣∣
≤ cα,β1
f1(γ0(1 + ψ(ξ)))
∑α1 +···+αl =α
β 1 +···+β l =β
l∏j=1
∣∣∣∣∣∂αj
ξ ∂β j
x (λ + f(x, q(x, ξ)))λ + f(x, q(x, ξ))
∣∣∣∣∣
= cα,β1
f1(γ0(1 + ψ(ξ)))
∑α1 +···+αl =α
β 1 +···+β l =β
l∏j=1
∣∣∣∣∣∂αj
ξ ∂β j
x λ + ∂αj
ξ ∂β j
x f(x, q(x, ξ))λ + f(x, q(x, ξ))
∣∣∣∣∣ .
When α = β = 0 we find∣∣∣∣∂αξ ∂β
x
1pλ (x, ξ)
∣∣∣∣ ≤ c0,01
f1(γ0(1 + ψ(ξ))).
Moreover whenever αj = βj = 0, then∣∣∣∣∣∂αj
ξ ∂β j
x λ + ∂αj
ξ ∂β j
x f(x, q(x, ξ))λ + f(x, q(x, ξ))
∣∣∣∣∣ = 1.
Therefore we now only have to consider
∣∣∣∣∂αξ ∂β
x
1pλ (x, ξ)
∣∣∣∣ ≤ cα,β1
f1(γ0(1 + ψ(ξ)))
∑α1 +···+αl =α
β 1 +···+β l =β
αj +β j >0
l∏j=1
∣∣∣∣∣∂αj
ξ ∂β j
x f(x, q(x, ξ))λ + f(x, q(x, ξ))
∣∣∣∣∣
www.mn-journal.com c© 2011 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
1000 K. Evans and N. Jacob: Variable order subordination
Further, using (3.21) we get∣∣∣∣∂αξ ∂β
x
1pλ(x, ξ)
∣∣∣∣ ≤ 1f1(γ0(1 + ψ(ξ)))
∑α1 +···+αl =α
β 1 +···+β l =β
αj +β j >0
×l∏
j=1
cαj β j εf(x, q(x, ξ))q(x, ξ)ε(1 + ψ(ξ))−ρ (|α j |)
2
λ + f(x, q(x, ξ))
=1
f1(γ0(1 + ψ(ξ)))
∑α1 +···+αl =α
β 1 +···+β l =β
αj +β j >0
[f(x, q(x, ξ))q(x, ξ)ε
λ + f(x, q(x, ξ)))
]l⎛⎝ l∏
j=1
cαj β j ε
⎞⎠
× (1 + ψ(ξ))−12 (ρ(|α1 |)+···+ρ(|αl |))
≤ γα,β ,ε1
f1(γ0(1 + ψ(ξ)))
∑α1 +···+αl =α
β 1 +···+β l =β
αj +β j >0
[f(x, q(x, ξ))q(x, ξ)ε
λ + f(x, q(x, ξ))
]l
(1 + ψ(ξ))−12 ρ(|α |)
using the sub-additivity of ρ. Since∣∣∣ f (x,q(x,ξ))λ+f (x,q(x,ξ))
∣∣∣ ≤ 1 we find
∣∣∣∣∂αξ ∂β
x
1pλ(x, ξ)
∣∣∣∣ ≤ cα,β ,ε
f1(γ0(1 + ψ(ξ)))
∑α1 +···+αl =α
β 1 +···+β l =β
αj +β j >0
q(x, ξ)lε(1 + ψ(ξ))−12 ρ(|α |) . (3.33)
Since q(x, ξ) ≤ c0,0(1 + ψ(ξ)) it follows further that given η > 0 we can find cα,β ,η > 0 such that∣∣∣∣∂αξ ∂β
x
1pλ(x, ξ)
∣∣∣∣ ≤ cα,β ,η
f0(γ0(1 + ψ(ξ)))(1 + ψ(ξ))η− 1
2 ρ(|α |) . (3.34)
Taking into account (3.31) we eventually arrive at∣∣∣∣∂αξ ∂β
x
1pλ(x, ξ)
∣∣∣∣ ≤ cα,β ,η (1 + ψ(ξ))−2 r 0 + 2 η −ρ ( |α |)
2 , (3.35)
i.e., p−1λ ∈ S−2r0 +2η ,ψ
ρ (Rn ) proving the theorem.
We can now prove
Theorem 3.8 Let p(x, ξ) be given by (3.3) where we assume for q condition (3.4). For f it is supposed that(3.1) and (3.2) hold. Then we have for any ε > 0 and ˜η > 0 that p ∈ S2r1 +2ε,ψ
ρ (Rn ) and p−1λ ∈ S−2r0 +2η ,ψ
ρ (Rn )where we assume that r1 −r0 < 1
2 . Let u ∈ Hpλ 0 (Rn ) ⊂ Hψ,r0 (Rn ) be the solution to (3.30) for g ∈ Hψ,k (Rn ),k ≥ 0. Then it follows that u ∈ Hψ,k+2r0 −2η (Rn ).
P r o o f. The statements for p and p−1λ have already been proved, i.e., Theorems 3.3 and 3.7, respectively.
From Theorem 2.4 it follows that
p−1λ0
(x,D) ◦ pλ0 (x,D) = id + r(x,D) (3.36)
with r ∈ S2r1 +2ε−2r0 +2η−1,ψ0 (Rn ). Since pλ0 (x,D)u = g we deduce from (3.36) that
u = p−1λ0
(x,D) ◦ pλ0 (x,D)u − r(x,D)u
= p−1λ0
(x,D)g − r(x,D)u.
c© 2011 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.mn-journal.com
Math. Nachr. 284, No. 8–9 (2011) / www.mn-journal.com 1001
Now, p−1λ0
(x,D)g ∈ Hψ,2r0 −2η+k (Rn ) and r(x,D)u ∈ Hψ,3r0 −2η−2r1 −2ε+1(Rn ) implying that u ∈ Hψ,t(Rn )for t = (k + 2r0 − 2η) ∧ (3r0 − 2η − 2r1 − 2ε + 1) > r0 . With a finite number of iterations we arrive atu ∈ Hψ,k+2r0 −2η (Rn ).
Corollary 3.9 If k + 2r0 − 2η > nρ1
then u ∈ C∞(Rn ).
We finally arrive at
Theorem 3.10 Let f : Rn × (0,∞) → R be an arbitrarily often differentiable function such that for
y ∈ Rn fixed, the function s → f(y, s) is a Bernstein function. Moreover assume (3.1) and (3.2). In addi-
tion let ψ : Rn → R be a continuous negative definite function in the class Λ which satisfies in addition (3.26).
For an elliptic symbol q ∈ S2,ψρ (Rn ) satisfying (3.4) we define p(x, ξ) by (3.3). We know that for any ε > 0 and
η > 0 it holds p ∈ S2r1 +2ε,ψρ (Rn ) and 1
p+λ ∈ S−2r0 +2η ,ψρ (Rn ). If r1 − r0 < 1
2 then −p(x,D) extends to agenerator of a Feller semigroup on C∞(Rn ).
P r o o f. The statements for p and p−1λ have already been proved, i.e., Theorems 3.3 and 3.7, respectively. We
want to apply the Hille-Yosida-Ray theorem, Theorem 4.5.3 in [18]. We know that p(x,D) maps Hψ,2r1 +2ε+k (Rn )into Hψ,k (Rn ). Hence if k > n
2ρ1the operator
(− p(x,D),Hψ,2r1 +2ε+k (Rn )
)is densely defined on C∞(Rn )
with range in C∞(Rn ). That −p(x,D) satisfies the positive maximum principle on Hψ,2r1 +2ε+k (Rn ) followsfrom Theorem 2.6.1 in [19]. Now, for λ ≥ λ0 we know that for g ∈ Hψ,k+2r1 −2r0 +2η+2ε(Rn ) we have a uniquesolution to pλ (x,D)u = g belonging to Hψ,k+2r1 +2ε(Rn ) implying the theorem.
References
[1] A. V. Balakrishnan, Fractional powers of closed operators and the semigroups generated by them, Pac. J. Math. 10,419–437 (1960).
[2] F. Baldus, Application of the Weyl-Hormander calculus to generators of Feller semigroups, Math. Nachr. 252, 3–23(2003).
[3] Chr. Berg, K. Boyadzhiev, and R. de Laubenfels, Generation of generators of holomorphic semigroups, J. Austral. Math.Soc. (Ser.A) 55, 246–269 (1993).
[4] S. Bochner, Diffusion equation and stochastic processes, Proc. Natl. Acad. Sci. U.S.A. 35, 368–370 (1949).[5] S. Bochner, Harmonic analysis and the theory of Probability, in: California Monographs in Mathematical Science (Uni-
versity of California Press, Berkeley, CA, 1955).[6] R. de Laubenfels, Existence Families, Functional Calculi and Evolution Equations, Lecture Notes in Mathematics
Vol. 1570 (Springer Verlag, Berlin, 1994).[7] K. P. Evans, Subordination in the Sense of Bochner of Variable Order, PhD-thesis (Swansea University, Swansea, 2008).[8] K. P. Evans and N. Jacob, Feller semigroups obtained by variable order subordination, Rev. Mat. Complut. 20, 293–307
(2007).[9] J. Faraut, Semi-groupe de mesures complexes et cacul symbolique sur les generateurs infinitesmaux de semi-groupes
d’opeerateurs, Ann. Inst. Fourier 20, 235–301 (1970).[10] W. Farkas, N. Jacob, and R. Schilling, Function spaces related to continuous negative definite functions: ψ-Bessel
potential spaces, Diss. Math. CCCXCIII, 1–62 (2001).[11] W. Farkas, N. Jacob, and R. Schilling, Feller semigroups, Lp -sub-Markovian semigroups, and applications to pseudo-
differential operators with negative definite symbols, Forum Math. 13, 51–90 (2001).[12] L. E. Fraenkel, Formulae for higher derivatives of composite functions, Math. Proc. Cambridge Phil. Soc. 83, 159–165
(1978).[13] F. Hirsch, Extension des proprietes des puissances fractionaire, in: Seminorie de Theorie du Potentiel, Lecture Notes in
Mathematics Vol. 563 (Springer Verlag, Berlin, 1976), pp. 100–120.[14] F. Hirsch, Domaines d’operateurs representes comme integrales des resolvantes, J. Funct. Anal. 23, 199–217 (1976).[15] W. Hoh, A symbolic calculus for pseudo-differential operators generating Feller semigroups, Osaka J. Math. 35, 798–
820 (1998).[16] W. Hoh, Pseudo Differential Operators Generating Markov Processes (Habilitationsschrift, Universitat Bielefeld,
Bielefeld, 1998).[17] W. Hoh, Pseudo differential operators with negative symbols of variable order, Rev. Mat. Iberoam. 16, 219–241 (2000).[18] N. Jacob, Pseudo-differential operators and Markov processes, in: Fourier Analysis and Semigroups Vol. 1 (Imperial
College Press, London, 2001).[19] N. Jacob, Pseudo-differential operators and Markov processes, in: Generators and their Potential Theory Vol. 2 (Imperial
College Press, London, 2002).
www.mn-journal.com c© 2011 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
1002 K. Evans and N. Jacob: Variable order subordination
[20] N. Jacob, Pseudo-differential Operators and Markov Processes, in: Markov Proceses and Applications Vol. 3 (ImperialCollege Press, London, 2005).
[21] N. Jacob and H.-G. Leopold, Pseudo-differential operators with variable order of differentiation generating Fellersemigroups, Integr. Equat. Oper. Th. 17, 544–553 (1993).
[22] K. Kikuchi and A. Negoro, Pseudo differential operators and spaces of variable order of differentiation, Rep. Fac. LiberalArts, Shizuoka University 31, 19–27 (1995).
[23] K. Kikuchi and A. Negoro, On Markov processes generated by pseudo differential operators of variable order, OsakaJ. Math. 34, 319–335 (1997).
[24] M. A. Krasnosel’skii, P. P. Zabreiko, E. J. Pustylnik, and P. E. Sbolevskii, Integral Operators in Spaces of SummableFunctions, in: Monographs and Textbooks on Mechanics of Solids and Fluids, Ser. Mechanics Analysis (NoordhoffInternational Publishing, Leyden, 1976).
[25] H.-G. Leopold, Pseudodifferentialoperatoren und Funktionenraume variabler Glattheit, Dissertation B (Friedrich-Schiller-Universitat Jena, Jena 1987).
[26] H.-G. Leopold, On function spaces of variable order of differentiation, Forum Math. 3, 69–82 (1991).[27] I. Louhivaara and C. Simader, Fredholmsche verallgemeinerte Dirichlet-probleme fur koerzitive lineare partielle Differ-
entialgleichungen, in: Proceedings of the Rolf Nevanlinna Symposium on Complex Analysis Vol. 7 (Silivri. Publ. Math.Research Inst. Istanbul, Istanbul, 1978), pp. 45–57.
[28] I. Louhivaara and C. Simader, Uber koerzitive lineare partielle Differentialoperatoren: Fredholmsche verallgemein-erte Dirichletprobleme und deren Klasseneinteilung, in: Complex Analysis and its Applications, edited by Boboljubov,N. N., et al. A collection of papers dedicated to I. N. Vekua on his 70th birthday (Izdat, Nauka Moscow, 1978),pp. 342–345.
[29] A. Negoro, Stable-like processes: Construction of the transition density and behaviour of sample path near t = 0, OsakaJ. Math. 31, 189–214 (1994).
[30] V. Nollau, Uber Potenzen von linearen Operatoren in Banachschen Raumen, Acta Sci. Math. 28, 107–121 (1967).[31] V. Nollau, Uber den Logorithmus abgeschlossener Operatoren in Banachschen Raumen, Acta Sci. Math. 30, 161–174
(1969).[32] R. S. Phillips, On the generation of semi-groups of linear operators, Pac. J. Math. 2, 343–369 (1952).[33] R. L. Schilling, On the domain of the generator of a subordinate semigroup, in: Proceedings of the Conference on
Potential Theory, edited by Kral, J., et al. (Walter de Gruyter Verlag, Berlin, 1996), pp. 449–462.[34] R. L. Schilling, Subordination in the sense of Bochner and a related functional calculus, J. Austral. Math. Soc. (Ser.A),
64, 368–396 (1998).[35] T. Uemura, On path properties of symmetric stable-like processes for one dimension, Potential Anal. 16, 76–91 (2002).[36] T. Uemura, On symmetric stable-like processes: some path properties and generators, J. Theor. Probab. 17(3), 541–555
(2004).[37] A. Unterberger and J. Bokobza, Les operateurs pseud differentiels d’ordre variable, C. R. Acad. Sci. Paris 261, 2271–
2273 (1965).[38] K. Yosida, Functional Analysis, fourth ed., Grundlehren der mathematischen Wissenschaften Band 123 (Springer Verlag,
Berlin, 1974).
c© 2011 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.mn-journal.com