5.3 Coordinate Transformations and Pseudodifferential Op- erators on Manifolds
In this section we will show that for every suitable smooth diffeomorphism κ : Rn→ Rn the operator Q : S(Rn) → S(Rn) defined by
Qu(x) = (p(x, Dx)w)(κ(x)), w(x) = u(κ−1(x)), u ∈ S(Rn) (5.17) is again a pseudodifferential operator. This is important to obtain a definition of pseudodifferential operators on a manifold in a way that is essentially independent of the choice of local charts. We note that
Qu= κ∗p(x, Dx)κ∗,−1u, where (κ∗v)(x) = v(κ(x)) and (κ∗,−1u)(x) = u(κ−1(x)).
More precisely, we assume that κ : Rn → Rn is a smooth function such that
∂xjκ∈ Cb∞(Rn) for all j = 1, . . . , n and
0 < c ≤ | det Dκ(x)| ≤ C < ∞ (5.18) for all x ∈ Rnand some constants c, C > 0. In particular, this implies that κ−1: Rn→ Rn is a again smooth and ∂xjκ−1 ∈ Cb∞(Rn) for all j = 1, . . . , n. In the following
∇xκ the total derivative of κ : Rn → Rn. The main result of this section is:
THEOREM 5.11 Let p∈ S1,0m(Rn× Rn), m ∈ R, let κ : Rn → Rn be as above and Q: S(Rn) → S(Rn) be defined by (5.17). Then there is some q ∈ S1,0m(Rn × Rn) such that Qu= q(x, Dx)u for all u ∈ S(Rn). Moreover, q(x, Dx) has the asymptotic expansion
q(x, ξ) ∼ X
α∈Nn0
1
α!∂αξDyαq(x, ξ, y)|˜ y=x, where
˜
q(x, ξ, y) = p(κ(x), A(x, y)−Tξ)| det A(x, y)|−1| det ∇yκ(y)| and A(x, y) =
Z 1 0
∇yκ(x + t(y − x)) dt for all x, y close enough
in the sense that for any N ∈ N0 q(x, ξ) − X
|α|≤N
1
α!∂ξαDyαq(x, ξ, y)|˜ y=x ∈ S1,0m−N −1(Rn× Rn).
In particular, we have
q(x, ξ) = p(κ(x), (∇yκ(x))−Tξ) + r(x, ξ) with r∈ S1,0m−1.
Proof: We only prove the theorem under the additional assumption that sup
x∈Rn
|∇κ(x) − I| ≤ 1 2.
For the proof in the general case we refer to [KG74, Theorem 6.3]. Here |.| is any matrix norm, which is induced by some vector norm. This ensures that
|A(x, y) − I| ≤ 1
2 for all x, y ∈ Rn
and therefore A(x, y)−T = (A(x, y)−1)T exists for all x, y ∈ Rn. Moreover, let χ ∈ S(Rn) with χ(0) = 1 and χε(ξ) = χ(εξ) for ε > 0, ξ ∈ Rn.
Under these assumptions we have Qu = lim
ε→0
Z
Rn×Rn
ei(κ(x)−y)·ξχε(ξ)p(κ(x), ξ)u(κ−1(y))d(y, ξ) (2π)n
= lim
ε→0
Z
Rn×Rn
ei(κ(x)−κ(x′))·ξχε(ξ)p(κ(x), ξ)u(x′)| det ∇x′κ(x′)|d(y, ξ) (2π)n for all u ∈ S(Rn). On the other hand, since
κ(x) − κ(x′) =
Z 1 0
∇x′κ(x′+ t(x − x′)) dt
(x − x′) = A(x, x′)(x − x′), we obtain by the change of variable ξ = A(x, x′)−Tη
Qu = lim
ε→0
Z
Rn×Rn
ei(x−x′)·ηχε(A(x, x′)−Tη)˜q(x, x′, η)u(x′) d(x′, η)
= lim
ε→0Os–
Z Z
e−iy·ηχε(A(x, x + y)−Tη)˜q(x, x + y, η)u(x + y) d(x′, η) due to (κ(x) − κ(x′)) · ξ = (x − x′) · A(x, x)−1ξ, where
˜
q(x, x′, η) = p(κ(x), A(x, x′)−Tη)| det A(x, x′)|−1| det ∇x′κ(x′)|.
Since {χε(A(x, x′)−Tη) : ε ∈ (0, 1)} is bounded in A00(Rn× Rn) with respect to (x′, η) and
χε(A(x, x′)−Tη) →ε→0 1 for all x, x′, η∈ Rn
∂xα′∂ηβχε(A(x, x′)−Tη) →ε→0 0 for all x, x′, η∈ Rn,|α| + |β| > 0, we can apply Corollary 4.10 to conclude that
Qu(x) = lim
ε→0
Z
Rn×Rn
ei(x−x′)·ηq(x, x˜ ′, η)u(x′) d(x′, η)
= lim
ε→0Os–
Z Z
e−iy·ηχε(A(x, x + y)−Tη)˜q(x, x + y, η)u(x + y) d(x′, η).
Finally, an application of Theorem 4.27 finishes the proof.
Next we define pseudodifferential operators on smooth compact manifolds. To this end recall:
Definition 5.12 M is a smooth compact manifold if M is a topological compact space and the following conditions hold:
1. There are finitely many open sets Ω1, . . . ,ΩN ⊂ M such that M =SN j=1Ωj. 2. For every j ∈ {1, . . . , N} there is an open set Uj ⊂ Rn and a continuous
bijective functions κj: Ωj → Uj with continuous inverse, called charts.
3. For every j, k ∈ {1, . . . , N} such that Ωj,k := Ωj ∩ Ωk 6= ∅ the mapping κj,k: κk(Ωj,k) → κj(Ωj,k)
can be extended to some is C∞-diffeomorphism with ˜κj,k: Vx → Vy for some open Vx ⊃ κk(Ωj,k), Vy ⊃ κj(Ωj,k).
A function f : M → C is called smooth if for every j = 1, . . . , N as above uj :=
u◦ κ−1j ∈ C∞(κj(Ωj)). Moreover, if Ω ⊂ M is open
C0∞(Ω) = {f ∈ C∞(M) : supp f ⊂ Ω}.
Before we define pseudodifferential operators on a compact manifold we the defi- nition of certain remainder operators:
Definition 5.13 Let M be a smooth compact manifold as above and let P : C∞(M) → C∞(M) be a linear operator. Then P has a C∞-kernel representation if for every j, k ∈ {1, . . . , N} there is some Kj,k ∈ C∞(Uk× Uj) such that for every u ∈ C0∞(Ωj)
(P u)(κ−1k (x)) = Z
Uj
Kj,k(x, x′)uj(x′) dx′ for all x ∈ Uk, where uj(x) = u(κ−1j (x)) in Uj.
For the following let Φj, j = 1, . . . , N be a smooth partition of unity subordinate to Ωj, j = 1, . . . , N, i.e.,
0 ≤ Φj ∈ C0∞(Ωj),
N
X
j=1
Φj(x) = 1 for all x ∈ M (5.19) Moreover, let Ψj ∈ C0∞(Ωj), j = 1, . . . , N, be such that
Ψj(x) = 1 for all x ∈ supp Φj, j = 1, . . . , N. (5.20) Finally, we define ϕj, ψj ∈ C0∞(Uj) by
ϕj = Φj ◦ κ−1j , ψj = Ψj ◦ κ−1j (5.21)
Definition 5.14 Let M be a smooth compact manifold as above and let P : C∞(M) → C∞(M) be a linear operator. Then P is a pseudodifferential operator of class S1,0m(M) with symbols (pj)Nj=1with respect to the charts κj, j = 1, . . . , N, if the following holds true: Let Φj,Ψj ∈ C∞(M), j = 1, . . . , N satisfy (5.19) and (5.20) and let
(P u)(x) =
N
X
j=1
(ΦjPΨju)(x) +
N
X
j=1
(ΦjP(1 − Ψj)u)(x) for all x ∈ M. (5.22)
Then, using ϕj, ψj defined in (5.21), ΦjPΨju can be written in the form (ΦjPΨju)(κ−1j (x)) = ϕjpj(x, Dx)(ψjuj)(x) for all x ∈ Uj
where uj(x) = u(κ−1j (x)), x ∈ Uj, for all u ∈ C∞(M), j = 1, . . . , N and (ΦjP(1 − Ψj)u) has a C∞-kernel representation.
Remark 5.15 If P is a differential operator, supp P (1 − Ψj)u ⊆ supp(1 − Ψj).
Therefore (ΦjP(1 − Ψj)u)(x) = 0 for all x ∈ M. Hence
(P u)(x) =
N
X
j=1
(ΦjPΨju)(x) for all x ∈ M
in this case.
The following theorem shows that the definition of pseudodifferential operators on M are independent of the choice of charts κ1, . . . , κN.
THEOREM 5.16 LetM be a smooth compact manifolds as above and letΩ′j, Uj′, κ′j, κ′j,k, j, k = 1, . . . , N′ be another sequence satisfying conditions 1.-3 in Definition 5.12 such that
˜
κj,j′ := κj◦ κ−1j′ : κ′j′(Ωj ∩ Ω′j′) → κj(Ωj ∩ Ω′j′), j ∈ {1, . . . , N}, j′ ∈ {1, . . . , N′}, extends to a C∞-diffeomorphim on some open neighborhoods of κ′j′(Ωj ∩ Ω′j′) and κj(Ωj ∩ Ω′j′). If P : C∞(M) → C∞(M) is a pseudodifferential operator in S1,0m(M) with respect to {Ωj, Uj, κj, κj,k : j, k = 1, . . . , N}, then P is in S1,0m(M) with respect to {Ω′j, Uj′, κ′j, κ′j,k : j, k = 1, . . . , N′}.
Proof: For simplicity we assume that ˜κj,j′ can be extended to a C∞-diffeormphism
˜
κj,j′: Rn → Rn such that ∇˜κj,j′,∇˜κ−1j,j′ ∈ Cb∞(Rn). For the proof in the general case we refer to [KG74, Theorem 7.3].
First of all, we prove that for every Ψ, Φ ∈ C∞(M) we have:
supp Φ ∩ supp Ψ = ∅ ⇒ ΦP Ψ has a C∞-kernel representation. (5.23)
We write
ΦP Ψ =
N
X
j=1
Φ(ΦjPΨj)Ψ +
N
X
j=1
Φ(ΦjP(1 − Ψj))Ψ.
Then Φ(ΦjP(1 − Ψj))Ψ has a C∞-kernel representation since ΦjP(1 − Ψj) has one.
Now we have
(Φ(ΦjPΨj)Ψu)(κ−1j (x)) = (ϕϕj)(x)pj(x, Dx)(ψjψuj)(x) for all x ∈ Uj, for all u ∈ C∞(M), j = 1, . . . , N, where uj = u ◦ κ−1j and ϕ = Φ ◦ κ−1j , ψ = Ψ ◦ κ−1j . Since supp(ψjψ) ∩ supp(ϕjϕ) = ∅, Theorem 5.8 implies that there are smooth kj: Rn× (Rn\ {0}) → C such that
(Φ(ΦjP(1 − Ψj))Ψu)(κ−1j (x)) = Z
Rn
(ϕϕj)(x)kj(x, x − y)ψj(y)ψ(y)uj(y) dy.
Hence (ϕϕj)(x)kj(x, x − y)ψj(y)ψ(y) ∈ C∞(Rn× Rn) and Φ(ΦjP(1 − Ψj))Ψ has a C∞-kernel representation. This proves (5.23).
Let Φ′j′,Ψ′j′ ∈ C∞(M), j′ = 1, . . . , N′, satisfy (5.19)-(5.20) with Ωj replaced by Ω′j′. Now let us write P as
P =
N′
X
j′=1
Φ′j′PΨ′j′+
N′
X
j′=1
Φ′j′P(1 − Ψ′j′).
Because of (5.23), the second term has a C∞-kernel representation. For the first term we use that
Φ′j′PΨ′j′ =
N
X
j=1
Φ′j′(ΦjPΨj)Ψ′j′ +
N
X
j=1
Φ′j′(ΦjP(1 − Ψj))Ψ′j′,
where again the second term has a C∞-representation. Moreover, if Ωj ∩ Ω′j′ 6= ∅ by assumption
(Φ′j′ΦjP(ΨjΨ′j′u))(κ−1j (x)) = ϕ′j′(x)ϕj(x)(pj(x, Dx)(ψ′j′ψjuj))(x) for all x ∈ Uj. Hence
(Φ′j′ΦjP(ΨjΨ′j′u))(κ−1j′ (x))
= ˜ϕ′j′(x)
(κ−1j ◦ κ′j′)−1∗
ϕjpj(x, Dx)(ψj′′(κ−1j ◦ κj′)∗( ˜ψju′j′))
(x) for all x ∈ Uj. where u′j′(x) = u((κ′j′)−1(x)), (κ∗v)(x) = v(κ(x)) for κ : Rn→ Rn, and ˜ϕ′j′ = Φ′j′◦κ−1j′ , ψ˜j′′ = Ψ′j′ ◦ κ−1j′ . Now, since ˜κj,j′ := κj ◦ (κ′j′)−1 can be extended to some C∞- diffeomorphism on Rn with ∇xκ˜j,j′,∇xκ˜−1j,j′ ∈ Cb∞(Rn), Theorem 5.11 implies that there is some qj,j′ ∈ S1,0m(Rn× Rn) such that
qj,j′(x, Dx)u = (κ−1j ◦ κ′j′)−1∗
ϕjpj(x, Dx) ψj(κ−1j ◦ κ′j′)∗u .
If we now define
qj′(x, ξ) = X
j′:Ω′
j′∩Ωj6=∅
qj,j′(x, ξ),
we have that
(Φ′j′PΨ′j′u)((κ′j)−1(x)) = ˜ϕ′jqj(x, Dx)(ψj′u′j′)(x) for all x ∈ Uj′, which proves the theorem.