ALGEBRAS OF PSEUDO-DIFFERENTIAL OPERATORS

ACTING ON HOLOMORPHIC SOBOLEV SPACES

DAVID SCOTT WINTERROSE ∗

Abstract. We search for pseudo-differential operators acting on holomorphic Sobolev spaces.

The operators should mirror the standard Sobolev mapping property in the holomorphic analogues.
arXiv:2110.09389v1 [math.AP] 18 Oct 2021

The setting is a closed real-analytic Riemannian manifold, or Lie group with a bi-invariant metric,
and the holomorphic Sobolev spaces are defined on a fixed Grauert tube about the core manifold.
We find that every pseudo-differential operator in the commutant of the Laplacian is of this kind.
Moreover, so are all the operators in the commutant of certain analytic pseudo-differential operators,
but for more general tubes, provided that an old statement of Boutet de Monvel holds true generally.
In the Lie group setting, we find even larger algebras, and characterize all their elliptic elements.
These latter algebras are determined by global matrix-valued symbols.

Key words. Holomorphic Sobolev Spaces; Grauert Tubes; Pseudodifferential Operators.

AMS subject classifications. 58J40, 32W25, 22E30, 32C05, 35S05

1. Introduction. Let (M, g) be a compact real-analytic Riemannian manifold.

This manifold always admits a (non-unique) Bruhat-Whitney complexification MC ,
and, inside MC , the family of Grauert tubes {Mǫ }ǫ<ǫ0 . See e.g. [19] for further details.
Let us denote by Ψ(M ) the algebra of standard pseudo-differential operators on M ,
and by O(Mǫ ) the holomorphic functions on Mǫ (O(N ) on a complex manifold N ).
Then, if P ∈ Ψ(M ), we seek a commutative diagram:

e
P
H1 H2
Rǫ Rǫ

∞ P ∞
C (M ) C (M )

where H1 , H2 ⊂ O(Mǫ ) are function spaces, Rǫ is the restriction map

Rǫ : O(Mǫ ) → C ω (M ) : f 7→ f |M .

That is, we seek necessary and/or sufficient conditions for the existence of such a Pe ,
and we would also like the two function spaces H1 and H2 above to be Hilbert spaces.
An operator P with this property will then be called ǫ-holomorphically extendible.
Our question will lead us to commutative diagrams of the form

e
P
HH s (Mǫ ) HH s−d (Mǫ )

Rǫ Rǫ

P
H s (M ) H s−d (M )

where HH s (Mǫ ) are the holomorphic Sobolev spaces indexed by s ∈ R, defined later,
and P ∈ Ψd (M ) is a (standard) pseudo-differential operator of degree d ∈ R.

∗ Departmentof Applied Mathematics and Computer Science, Technical University of Denmark,

Kgs. Lyngby, Denmark (dawin@dtu.dk).
1
2 D.S. WINTERROSE

Why do we ask this? We seek a parallel to the Boutet de Monvel calculus [6, 1],
but adapted to solve real-analytic elliptic boundary value problems in Hilbert spaces
of holomorphic functions. This research path was inspired by Karamehmedović [11],
and further additions in Winterrose [21]. Unfortunately, no such calculus is known.
It is not clear if it even exists. As a beginning, we look for P with the above property,
and, especially, parametrices with a similar extension property if P is also elliptic.
The ǫ-holomorphic extension property can be viewed as an abstract global analogue
of the property in Winterrose [20] for (local) analytic pseudo-differential operators.
This work, and the reason for undertaking it, is further detailed in [21].

Consider a compact Lie group G with Lie algebra g of left-invariant vector fields.
It admits a complexification GC itself a complex Lie group [8, pp. 110, Lemma 3],
and the natural Cartan decomposition diffeomorphism
G × g → GC : (x, Y ) 7→ exp(iY )x,
where i comes from the complex structure on GC . See Hall [8] and references therein.
Give G the bi-invariant metric induced by an Ad(G)-invariant inner product (·, ·)g .
Relative to this metric, the tube Gǫ , of any radius ǫ ∈ (0, ∞], is just
 
Gǫ = exp(iY )  |Y |g < ǫ G.
The universal property of GC ensures that every irreducible unitary representation ξ
of G extends holomorphically to GC . Write [ξ] ∈ G b for its unitary equivalence class,
and dξ for the dimension of its representation space. Write ∆ = ∆G for the Laplacian.
The Fourier transform on G is F = FG . See also Folland [5, Chapter 5] for details.
Recall that components of ξ are eigenfunctions for −∆G for a single eigenvalue λξ ,
and Gb forms a basis for L2 (G) via Peter-Weyl’s theorem [5, Theorem 5.12].

Combining this, we get our prototypical example, the algebra generated by ∆G .

The Laplacian extends to an operator (∆G )C on holomorphic functions on GC or Gǫ .
If u ∈ O(Gǫ ), and z = exp(iY )x with x ∈ G, we can write
X
(∆G )C u(z) = dξ Tr(−λξ ξ(z)F u(ξ))
b
[ξ]∈G
X  Z 
= dξ Tr − λξ ξ(x) u(exp(iY )y)ξ(y)∗ dy ,
b G
[ξ]∈G

which converges absolutely uniformly for Y in compact subsets of {Y ∈ g | |Y |g < ǫ}.

Therefore (∆G )C u is well-defined, holomorphic on Gǫ , and (∆G )C is bi-G-invariant.
If f ∈ C 0 (C) has no more than polynomial growth, we can replace −λξ with f (−λξ ).
For example, if s ∈ R, we form
s
X
(I − (∆G )C ) 2 u(z) = dξ Tr(hξis ξ(z)F u(ξ)),
b
[ξ]∈G

1
where hξi = (1 + λ2ξ ) 2 , and this leads to natural ”holomorphic Sobolev spaces” on Gǫ :
s
HH s (Gǫ ) = {u ∈ O(Gǫ ) | (I − (∆G )C ) 2 u ∈ HL2 (Gǫ )}.
Unfortunately, these ideas do not carry over directly to the Grauert tubes Mǫ of M .
s s
How should we define (I − ∆C ) 2 , analogous to (I − (∆G )C ) 2 , on Mǫ when s 6∈ 2N0 ?
 MAPPING HOLOMORPHIC SOBOLEV SPACES 3

We will overcome this problem by using the Poisson transform due to Stenzel [18].
This transform relates the holomorphic functions on Mǫ to their restrictions on M .
It is used by Stenzel to get an analogue of the Segal-Bargmann transform [8] for Mǫ .
Arising from a Fourier integral operator, it has well-understood mapping properties,
and an entire tube is not required to get a Hilbert space structure on its image [18].
The Poisson transform relies on an instance of a theorem of Boutet de Monvel [2].
This theorem has also been used in several contexts by Zelditch, see e.g. [23, 22, 24],
but it has not been proved in the generality originally stated by Boutet de Monvel.
So we refer to the general statement as a conjecture, although in [2] it is a theorem.
It should be said that a very rough sketch is provided in [2].

On M the conjecture implies a supply of operators with the ǫ-extension property.

Elements in the Ψ(M )-commutant of P ∈ Diff(M ) for which it holds has the property,
and their extensions mirror the Sobolev mapping property [16, pp. 69, Theorem 8.2].
As a special case, it implies such a parametrix for any formally normal A ∈ Ψ(M ).
This is contained in Theorems 5.11 and 5.15, and the two Corollaries 5.14 and 5.12.
Of course, without a proof in the general case, it is only certain for P = ∆g .

On the compact Lie group G, we set up an even larger algebra in Definition 5.17.
This is accomplished by exploiting the fact that Ψ(G) admits a bijection p 7→ Op(p),
b = S d are symbols due to Ruzhansky, Turunen and Wirth [15].
where p ∈ S d (G × G)
These operators are the quantizations of holomorphic symbols in a subalgebra of S d ,
and Theorem 5.28 characterizes ellipticity in it, resembling [15, pp. 10, Theorem 4.1].
Also here, the Sobolev mapping property is mirrored, in Theorem 5.20.

2. Notation. We write B(H) for the bounded operators on a normed space H,

and B(H1 , H2 ) for operators bounded from one normed space H1 into another H2 .
1S denotes the indicator of a (sub)set S. In multiple contexts, I is the identity map.
Fix a smooth positive 1-density ω0 on M , and form the L2 -space on M relative to it.
Then H s (M ) is the Sobolev space of order s ∈ R on M , normed by
s
||u||H s (M) = ||(I − ∆g ) 2 u||L2 (M) for any u ∈ H s (M ),
s
where ∆g is the Laplacian of (M, g), and (I − ∆g ) 2 formed from the calculus of [16].
On a Kähler manifold-with-boundary N , boundary ∂N is automatically orientable,
and it is equipped with the volume form induced by the Kähler volume form on N .
The Sobolev spaces H s (N ◦ ) and H s (∂N ) are then defined relative to these choices,
and the space Os (N ) is the closure in H s (∂N ) of

{u|∂N | u ∈ C ∞ (N )} ∩ O(N ◦ )

The distribution space is denoted D′ (M ), or D′ (N ◦ , Ωp,q ) for distribution (p, q)-forms.

Finally, ∂ and ∂ are the Dolbeault operators on N ◦ .

Beyond notation already introduced for G, dx is the normalized Haar measure,

and gl(k, K) is the Lie algebra of k × k matrices with entries contained in a field K.
The associated general linear group is GL(k, K), with entries in K. Here K is C or R.
Also, U(m) are the m × m unitary matrices, and O(m) the (real) orthogonal matrices.
On matrices, || · || is the norm induced by the Euclidean 2-norm, and Tr is the trace.
4 D.S. WINTERROSE

3. Preliminaries. Before proceeding, we need some definitions and theorems.

Mainly, theory by Ruzhansky, Turunen and Wirth [15], Wirth and Ruzhansky [14],
and some minor lemmas in Winterrose [21] derived from [14]. Let d ∈ R.

b
Analogous to the Schwartz functions on Rn , there exists a sequence space S(G).
d b b→∪
Define spaces S (G) of a : G b gl(dξ , C) with a([ξ]) ∈ gl(dξ , C) and
[ξ]∈G
X
dξ hξi2d Tr(a(ξ)∗ a(ξ)) < ∞,
b
[ξ]∈G

which are equipped with the inner products

X
(a, b)S d (G)
b = dξ hξi2d Tr(b(ξ)∗ a(ξ)) b
for any a, b ∈ S d (G),
b
[ξ]∈G

and give S(G) = ∩k∈Z S k (G) b the Frechet topology induced by the collection of norms.
Then S(G) is sequentially dense in every S k (G) by truncating the sequences in S k (G).
It is the right space; F : C ∞ (G) → S(G) b is a well-defined linear homeomorphism.
Give each space the pairing
X
S −k (G) × S k (G) → C : (a, b) 7→ ha, bi = dξ Tr(a(ξ)b(ξ)),
b
[ξ]∈G

and then Riesz’ theorem and denseness of S(G) implies that

S ′ (G) ∼ b
= ∪k∈Z S k (G).

b × S(G).
This is understood in the above pairing, which extends to S ′ (G) b
Definition 3.1. Overloading notation, F and F −1 extend by duality:
b : f 7→ F f by
1. Extend F to F : D′ (G) → S ′ (G)

hF f (ξ), a(ξ)i = hf (x), F −1 a(x−1 )i for all b

a ∈ S(G).

b → D′ (G) : a 7→ F −1 a by
2. Extend F −1 to F −1 : S ′ (G)

hF −1 a(x), f (x−1 )i = ha(ξ), F f (ξ)i for all f ∈ C ∞ (G).

Continuous operators P : C ∞ (G) → C ∞ (G) decompose by Peter-Weyl’s theorem.

The Peter-Weyl expansion of u ∈ C ∞ (G) converges in C ∞ (G), so we have
X
P u(x) = dξ Tr(ξ(x)p(x, ξ)F u(ξ)) for all x ∈ G,
b
[ξ]∈G

where p(x, ξ) = ξ(x)∗ (P ξ)(x) is the ”matrix-symbol” of P . This defines P uniquely.

It has the general property that
b
p ∈ C ∞ (G; S ′ (G)),

and we will write P = Op(p). The map p 7→ Op(p) is the operator quantization map.
It is customary to suppress the brackets [ξ], since p is constant on each class.
 MAPPING HOLOMORPHIC SOBOLEV SPACES 5

In order to describe the symbol space S d , we need to set up some more structure.
Take a faithful representation, ρ : G → U(m) for some m ∈ N, onto a Lie subgroup.
Using real matrices to represent C, we have

U(m) ∼
= O(2m) ∩ GL(m, C) ⊂ GL(2m, R),

and so we may assume ρ : G → O(2m). Choose a complement to dρ(g) in gl(2m, R).

By the inverse function theorem, there is an open V ⊂ GL(2m, R) with

ρ(G) ⊂ V,

and a neighbourhood U of 0 in the complement, giving a diffeomorphism

G × U → V : (g, Y ) 7→ ρ(g) exp(Y ),

which in turn gives a smooth ”projection” map

̺ : V → G : ρ(g) exp(Y ) 7→ g.

It absorbs G from the left, that is,

x̺(y) = ̺(ρ(x)y) for any (x, y) ∈ G × V.

Fix a cutoff χ ∈ C0∞ (V ) equal to 1 on a ρ(G)-invariant neighbourhood of ρ(G) in V .

Note that χ(ρ(x)y) = χ(ρ(x)) = 1 for x ∈ G and y in this neighbourhood.

This choice of embedding of G into a Euclidean space is fixed once and for all.
This is the structure implicit in Ruzhansky, Turunen and Wirth [15].
Definition 3.2. Associated to these choices, we define two operators δxα and δξα .
Given any α ∈ Nm×m
0 with m as in the embedding ρ : G → O(2m), put

dα (x) = (ρ(x−1 ) − I)α for all x ∈ G,

where α is viewed as a multi-index, the power is a product over terms indexed by α.

1. Define δxα : C ∞ (G) → C ∞ (G) : u 7→ δxα u by
 
δxα u(x) = ∂yα [χ(u ◦ ̺)(ρ(x)y)] y=I for all x ∈ G.

b → S ′ (G)
2. Define δξα : S ′ (G) b : a 7→ δ α a by
ξ

δξα a(ξ) = F dα F −1 a for all b

[ξ] ∈ G.

b sequences.
The ”difference operators” δξα satisfy a Leibniz-type rule on the S ′ (G)
Write eij for the m × m matrix with 1 at the i, j entry and zero else.
Lemma 3.3 (Ruzhansky, Turunen and Wirth [15]).
m
X
e e e e b
δξ ij (ab) = δξ ij (a)b + aδξ ij (b) + δξeik (a)δξ kj (b) for any a, b ∈ S ′ (G).
k=1
6 D.S. WINTERROSE

Let d ∈ R. Let us write X β = X1β1 ◦ · · · ◦ Xnβn for any β ∈ Nn0 if {Xj }nj=1 ⊂ g.
b = S d for d ∈ R, if the following holds:
Definition 3.4. Write p ∈ S d (G × G)
b
1. p ∈ C ∞ (G; S ′ (G)).
2. Given any ordered basis X = (X1 , · · · , Xn ) of g, we have
sup hξi|α|−d ||δξα Xxβ p(x, ξ)|| < ∞ for any α ∈ Nm×m
0 and β ∈ Nn0 .
b
(x,[ξ])∈G×G

The space S d has the Frechet topology induced by the above collection of semi-norms.
Given a sequence {pj }∞
j=0 with pj ∈ S
dj
and dj → −∞ as j → ∞, we write
∞
X k−1
X
p∼ pj if p− pj ∈ S maxj≥k dj for each k ∈ N.
j=0 j=0

These matrix-symbols are called the Hörmander symbols.

Theorem 3.5 (Ruzhansky, Turunen and Wirth [15]).
Ψd (G) = Op S d .

Definition 3.6. Let d1 , d2 ∈ R. Given any p ∈ S d1 and q ∈ S d2 , define:

b a symbol p ⊙ q by
1. Pointwise in (x, [η]) ∈ G × G
X Z  
(p ⊙ q)(x, η) = dξ Tr ξ(y −1 x)p(x, ξ) η(x−1 y)q(y, η) dy.
b G
[ξ]∈G

b a symbol p† by
2. Pointwise in (x, [η]) ∈ G × G
X Z  
p† (x, η) = dξ Tr ξ(y −1 x)p(y, ξ)∗ η(x−1 y) dy.
b G
[ξ]∈G

Both of the two sums in the above definition are uniformly absolutely convergent.
To see this, integrate by parts entry-wise with I − ∆y to summon powers of hξi−1 .
Then the sums are dominated by convergent sums scaling with hηi, independent of x,
because p and its derivatives in x grow polynomially in hξi at a fixed rate.
Theorem 3.7 (See e.g. Winterrose [21]). Let d1 , d2 ∈ R. The following holds:
1. The map S d1 × S d2 → S d1 +d2 : (p, q) 7→ p ⊙ q is well-defined, and
∞
X X 1
Op(p)Op(q) = Op(p ⊙ q) with p⊙q ∼ δ α (p)δxα (q).
α! ξ
N =0 |α|=N

2. The map S d1 → S d1 : p 7→ p† is well-defined, and

∞
X X 1
Op(p)∗ = Op(p† ) with p† ∼ δ α δ α (p∗ ).
α! ξ x
N =0 |α|=N

Both maps are continuous with respect to the Frechet topologies on the symbol spaces.
 MAPPING HOLOMORPHIC SOBOLEV SPACES 7

In our applications to compact Lie groups, we need the following two lemmas.
Both are proved in Winterrose [21], but are just easy adaptations of proofs in [14],
where a parameter Y ∈ K running in a topological space K has just been inserted.
It represents Y ∈ g in the Cartan decomposition exp(iY )x = z ∈ GC later.
Lemma 3.8 (Winterrose [21]). Let K be compact, and {pY }Y ∈K ⊂ S d bounded.
If all x-derivatives of pY (x, ξ) are continuous in (x, Y ), then for s ∈ R we have

sup ||Op(pY )||B(H s (G),H s−d (G)) < ∞,

Y ∈K

which is to say, Op(pY ) is bounded uniformly in Y from H s (G) to H s−d (G).

Lemma 3.9 (Winterrose [21]). Let K be compact, and {pY }Y ∈K ⊂ S d bounded.
Assume that each pY (x, ξ) is point-wise invertible as a matrix belonging to gl(dξ , C).
If all x-derivatives of pY (x, ξ) are continuous in (x, Y ), then

sup sup hξid ||pY (x, ξ)−1 || < ∞,

Y ∈K (x,[ξ])∈G×G
b

and {p−1
Y }Y ∈K ⊂ S
−d
is bounded, with the same continuity property.

Recall the calculus of poly-homogeneous (PHG) operators Ψdphg (M ) in Ψd (M ).

These have a unique homogeneous principal symbol, the classical principal symbol,
and elliptic A ∈ Ψdphg (M ) with d > 0 have either discrete spectrum σ(A), or σ(A) = C.
Choose such an A ∈ Ψdphg (M ), but with a discrete spectrum in the right half-plane.
It is parameter elliptic w.r.t Λ if its classical principal symbol does not valuate in Λ.
Suppose A is parameter-elliptic w.r.t. a [π − θ0 , π + θ0 ]-sector for some θ0 ∈ (0, π).
Given k ∈ N0 and z ∈ C with Re (z) − k < 0, we can unambiguously define
Z
1
Az = λz−k (λI − A)−1 Ak dλ,
2πi ΓR

where ΓR is the counter-clockwise relative to σ(A) \ {0} boundary of

ΛR = ∪θ∈[−π+θ0 ,π−θ0 ] eiθ [R, ∞) ⊃ σ(A) \ {0},

and the integral converges in B(H N (M )) for all N ∈ N by [16, pp. 86, Theorem 9.3].
By the Sobolev embedding theorem, it converges weakly in the topology of C ∞ (M ).
Suppose A is parameter elliptic w.r.t. to a [θ0′ , 2π − θ0′ ]-sector for some θ0′ ∈ (0, 12 π).
In the same way, for Re (z) > 0, we can form
Z Z
z 1 z 1
e−tA = e−tλ (λI − A)−1 dλ + (λI − A)−1 dλ,
2πi Γ′R 2πi RS1

where RS1 is oriented counter-clockwise, Γ′R = ∂Λ′R with R > 0 such that
′
Λ′R = ∪θ′ ∈[−θ0′ ,θ0′ ] eiθ [R, ∞) ⊃ σ(A) \ {0},

and Γ′R is oriented counter-clockwise relative to σ(A) \ {0} inside the right half-plane.
In particular, if A is formally self-adjoint and elliptic, its spectrum must be discrete,
and is semi-bounded from below if its classical principal symbol is positive on T ∗ M \0.
This is [16, pp. 71, Theorem 8.3-8.4] and [16, pp. 86, Corollary 9.3].
8 D.S. WINTERROSE

4. The Boutet de Monvel Conjecture. (A theorem in a prototypical case.)

A theorem was announced by Boutet de Monvel in an old 1978 conference paper [2],
but only recently was the case P = ∆g fully proved in [17] and (incompletely) in [23].
Several possible approaches to proving it generally are discussed by Zelditch in [23].
To our knowledge, the general statement is not yet fully proven.

Let d > 0, and let P ∈ Ψdphg (M ) be analytic with classical principal symbol p.
Suppose that P is formally self-adjoint, elliptic, p|T ∗ M\0 > 0, and that

{ξ ∈ Tx∗ M | p(x, ξ) ≤ 1} is strictly convex for all x ∈ M.

In that case, P is semi-lower bounded, and we assume it has no negative eigenvalues.

1
A real-analytic complete Hamiltonian flow ϕt : T ∗ M \ 0 → T ∗ M \ 0 arises from p d .
Given any t ∈ R and (x, ξ) ∈ T ∗ M \ 0, it satisfies

ϕt (x, sξ) = s · ϕt (x, ξ) for all s > 0,

and ϕt (x, ξ) extends holomorphically in t to {t ∈ C | |t| < ǫ} for ǫ > 0 small enough.
The extension is then a map
∗
{t ∈ C | |t| < ǫ} → T1,0 MC \ 0 : t 7→ ϕt (x, ξ),
∗
where T1,0 MC is the holomorphic cotangent bundle of MC .

Conjecture 4.1 (Boutet de Monvel [2]. The statement here is a bit different).
1
Let Pǫ denote the Schwartz kernel of e−ǫP d , and put

Φx (ξ) = (πx ϕ 1 )(x, ξ) for any (x, ξ) ∈ Bǫ∗ M \ 0,

ip(x,ξ) d

where πx is the bundle projection, and

1
Bǫ∗ M = {(x, ξ) ∈ T ∗ M | p(x, ξ) d < ǫ}.

There is a maximal ǫ0 > 0 such that for any ǫ ∈ (0, ǫ0 ) the following holds:
1. The extended flows combine into a real-analytic diffeomorphism

Φ : Bǫ∗ M \ 0 → MǫΦ \ M ⊂ MC : (x, ξ) 7→ Φx (ξ).

2. The set MǫΦ is open in MC with (orientable) C ω -boundary ∂MǫΦ .

3. The set MǫΦ is strictly pseudo-convex.
The last point says that MǫΦ admits a Kähler structure, and a global potential for it.
There is a maximal ǫ′0 ∈ (0, ǫ0 ] such that for any ǫ ∈ (0, ǫ′0 ) the following holds:
1. The map x 7→ Pǫ (x, y) extends holomorphically to MǫΦ for each fixed y ∈ M .
2. The kernel Pǫ |∂MǫΦ ×M induces a complex-phase PHG FIO Sǫ of order − n−1
4 .
n−1
3. Sǫ defines a homeomorphism Sǫ : H s (M ) → Os+ 4 (∂MǫΦ ) for any s ∈ R.

The first three points simply generalize the basic properties of the Grauert tubes.
In the special case that P = ∆g , MǫΦ is exactly the Grauert tube Mǫ .
 MAPPING HOLOMORPHIC SOBOLEV SPACES 9

The abbreviation ”PHG FIO” is for poly-homogeneous Fourier integral operator.

In this article, we only remind the reader of the classical article by Hörmander [10],
and, in the situation of globally defined complex phase FIO, Melin and Sjöstrand [12].
The theory in the complex case runs parallel to the real case, but is less widely known.
In the proven instance of 4.1, there are precise details on the structure of Sǫ :

Theorem 4.2 (Stenzel [17, 18] and Zelditch [23]. See Boutet de Monvel [4]).
Conjecture 4.1 is true for the Laplacian P = ∆g of (M, g) for certain ǫ0 ≥ ǫ′0 > 0.
The positive homogeneous canonical relation of Sǫ arises from the graph of

|ξ|x
T ∗ M \ 0 → T ∗ (∂Mǫ ) \ 0 : (x, ξ) 7→ α(x,ξ) ,
ǫ
where the cotangent vector is

α(x,ξ) = (ι∗∂Mǫ Im ∂ρ)exp −1 ♯ ,

x (iǫ|ξ|x ξ )

and ρ is the unique tube function of Mǫ , a Kähler potential for the structure on Mǫ .
This function is the unique real-analytic solution ρ : Mǫ → [0, ∞) of
( √
(i∂∂ ρ)∧n = 0 in Mǫ \ M,
Lev(ρ)|M = 12 g,

where Lev(ρ) is the Levi form of ρ. The second condition ensures isometric inclusion.
It has, and for uniqueness must have, the following properties:
1. ρ is invariant under the unique anti-holomorphic involution on Mǫ .
2. ρ is strictly plurisubharmonic on Mǫ .
3. ρ|M = 0 and dρ|M = 0.
− n−1
Finally, Sǫ∗ Sǫ ∈ Ψphg2 (M ) is elliptic, with classical principal symbol

− n−1
T ∗ M → [0, ∞) : (x, ξ) 7→ |ξ ♯ |x 2
.

Example 4.3 (Torus Tn ). Equip the n-torus Tn ∼

= (R/2πZ)n with the flat metric.
n
The Bruhat-Whitney complexification of T is

TnC ∼
= (C/2πZ)n ,

and the complexified geodesic at ([z], ξ) ∈ TnC × Cn ∼

= T TnC is

t 7→ [z + tξ].

The Grauert tube function of Tnǫ for any ǫ > 0 is

√
ρ : Tnǫ → [0, ∞) : [z] 7→ |Im (z)|,

where Tnǫ ∼
= (Rǫ /2πZ)n for Rǫ = {z ∈ C | |Im (z) < ǫ}, 2πZ acting on the real part.
10 D.S. WINTERROSE

5. Operators with the ǫ-Extension Property. Let n = dim M throughout.

The Poisson transform in [18] is built from the solution operator to
(
∆h u = 0 in Mǫ ,
u|∂Mǫ = f on ∂Mǫ ,

where ∆h is the Kähler Laplacian associated to the Kähler metric h from Theorem 4.2,
and f is, at first, a smooth function on ∂Mǫ , and we look for u ∈ C 2 (Mǫ ) ∩ C 0 (Mǫ ).
The unique solvability of this Dirichlet problem for the Laplacian ∆h is well-known,
but we need details on extensions of the solution operator.

Assume therefore that (N, h) is a smooth Riemannian manifold-with-boundary.

e,
In our case, N is the closure of an open subset N ◦ in a compact smooth manifold N
◦
where N = N ⊔ ∂N , the topological boundary is a co-dimension one hypersurface.
This ”enveloping manifold” N e carries a metric coinciding with h on N .
Definition 5.1. Given any s ∈ R, we define the H s -extendible distributions
e )},
H̄ s (N ◦ ) = {u|N ◦ ∈ D′ (N ◦ ) | u ∈ H s (N

and equip this space with the norm

||v||H̄ s (N ◦ ) = inf ||u||H s (Ne ) if v ∈ H̄ s (N ◦ ).

u|N ◦ =v

These spaces of s ∈ R extendible distributions are ingredients in the next theorem.

It states that the Dirichlet problem for ∆h on N is uniquely solvable in these spaces.
A clear exposition (of the invertible case) is found in Grubb [7, pp. 320-326].
Theorem 5.2 (Outlined in Grubb [7]. Alternatively, see Boutet de Monvel [3]).
Let s ∈ R. There are bounded, linear, and mutually inverse bijections
1
Kγ : H s− 2 (∂N ) → H̄ s (N ◦ ) ∩ ker (∆h |H̄ s (N ◦ ) ),
1
γ0 : H̄ s (N ◦ ) ∩ ker (∆h |H̄ s (N ◦ ) ) → H s− 2 (∂N ).
1
If f ∈ H s− 2 (∂N ), they together uniquely solve
(
∆h (Kγ f ) = 0 in N,
γ0 (Kγ f ) = f on ∂N.

The operator γ0 acts as the restriction operator onto ∂N for smooth functions on N ,
and Kγ is a Poisson operator (see [7]), restricting to a continuous linear map

Kγ : C ∞ (∂N ) → C ∞ (N ).

Relative to fixed smooth positive 1-densities on N and ∂N , it has the two properties:
1. Kγ admits a unique formal L2 -adjoint Kγ∗ .
2. Kγ∗ Kγ ∈ Ψ−1 ∗
phg (∂N ) has classical principal symbol positive on T N \ 0.
 MAPPING HOLOMORPHIC SOBOLEV SPACES 11

In the case that (N ◦ , h) is Kähler, we can use the theorem to characterize Os (∂N ).
Naturally, the boundary ∂N carries the Riemannian structure induced from N .
Lemma 5.3. In the above setting, if (N ◦ , h) is Kähler, we have
n  o
1 
Os (∂N ) = γ0 u ∈ H̄ s+ 2 (N ◦ ) ∩ C ∞ (N ◦ )  ∂(u|N ◦ ) = 0 .

Proof. Write u ∈ Os (∂N ) as u = γ0 (Kγ u). We must show that ∂(Kγ u|N ◦ ) = 0.
Take a sequence {uj }∞ ∞ ◦
j=1 ⊂ C (N ) with uj ∈ O(N ) such that

uj |∂N → u in H s (∂N ) as j → ∞.

Since holomorphic functions are harmonic for ∆h , we have

uj |N ◦ ∈ ker (∆h |H̄ s (N ◦ ) ).

Consequently, we can write

1
uj = Kγ (uj |∂N ) → Kγ u in H̄ s+ 2 (N ◦ ) as j → ∞,

and then also

0 = ∂(uj |N ◦ ) → ∂(Kγ u|N ◦ ) in D′ (N ◦ , Ω0,1 ) as j → ∞.

Thus ∂(Kγ u|N ◦ ) = 0, and Kγ u ∈ C ∞ (N ◦ ) by ellipticity of ∆h .

Choose ǫ′0 ≤ ǫ0 so that the Boutet de Monvel Theorem 4.2 holds, and ǫ ∈ (0, ǫ′0 ).
Using Theorem 5.2, we construct the Poisson transform as defined by Stenzel in [18].
The Poisson transform is well-behaved, and can be unitarized by polar decomposition.
It is very closely related to the restriction map [18].

Let h be the Kähler metric on the closure of a tube Mǫ , and let s ∈ R be arbitrary.
n+1
√
Definition 5.4. If f ∈ H s− 4 (M ), Pǫ f is the extension of e−ǫ −∆g f to Mǫ .
Suppose Kǫ is the Poisson operator solving the Dirichlet problem for ∆h on Mǫ .
Then, by Theorem 5.2 it is a linear homeomorphism
1
Kǫ : H s− 2 (∂Mǫ ) → H̄ s (Mǫ ) ∩ ker(∆h |H̄ s (Mǫ ) ),

and so, by Theorem 4.2, Pǫ = Kǫ Sǫ realizes a linear homeomorphism

n+1 1
Pǫ : H s− 4 (M ) → Kǫ Os− 2 (∂Mǫ ).

In the special case that s = 0, we have

1
Kǫ O− 2 (∂Mǫ ) = HL2 (Mǫ ),

which follows from H̄ 0 (Mǫ ) = L2 (Mǫ ), as any L2 -function can be extended by zero,
and Lemma 5.3 ensures that the functions are holomorphic on the interior.
n+1 1
Theorem 5.5. The composite Pǫ : H s− 4 (M ) → Kǫ Os− 2 (∂Mǫ ) is well-defined.
Furthermore, it is a linear homeomorphism.
12 D.S. WINTERROSE

Our approach is to indirectly define spaces as images under Pǫ of any order s ∈ R.

These will coincide with the natural definition on Gǫ when M = G.
Definition 5.6.
n+1
HH s (Mǫ ) = Pǫ H s− 4 (M ).

Using the inverse Pǫ−1 , the image space is made to inherit the Hilbert space structure.
In the case M = G and P = ∆G , it has the inner product induced by
s
||u||HH s (Gǫ ) = ||Pǫ (I − ∆G ) 2 Pǫ−1 u||L2 (Gǫ ) if u ∈ HH s (Gǫ ),
n+1
and Pǫ : H s− 4 (G) → HH s (Gǫ ) is automatically a bounded isomorphism this way.
This is clear from
s
||Pǫ u||HH s (Gǫ ) = ||Pǫ (I − ∆G ) 2 u||L2 (Gǫ )
≤ ||Pǫ || −
n+1 ||u|| n+1 .
B(H 4 2
(G),HL (Gǫ )) H s− 4 (G)

Proposition 5.7.
s
HH s (Gǫ ) = {u ∈ O(Gǫ ) | (I − (∆G )C ) 2 u ∈ HL2 (Gǫ )}.

Proof. Take any u ∈ C ∞ (G), and for any z ∈ Gǫ write

s
X s
Pǫ (I − ∆G ) 2 u(z) = dξ Tr(Pǫ ξ(z)F (I − ∆G ) 2 u)
b
[ξ]∈G
X  Z √ 
= s
dξ Tr hξi ξ(z) u(x)e−ǫ λξ ξ(x)∗ dx
b G
[ξ]∈G
X  Z √ 
= dξ Tr hξis ξ(z) (e−ǫ −∆ u)(x)ξ(x)∗ dx ,
b G
[ξ]∈G

where we use that the Peter-Weyl expansion of u converges in the topology of C ∞ (G),
and so the first sum converges in HL2 (Gǫ ), thus uniformly on compact subsets of Gǫ .
This shows that
s s
Pǫ (I − ∆G ) 2 u = (I − (∆G )C ) 2 Pǫ u,
n+1
which, on the left hand side, must extend to u ∈ H s− 4 (G) just by the continuity.
n+1
To see equality for u ∈ H s− 4 (G), pick {uk }∞ ∞
k=1 ⊂ C (G) such that

n+1
uk → u in H − 4 (G) as k → ∞,

and note that

s s
(I − (∆G )C ) 2 Pǫ uk → (I − (∆G )C ) 2 Pǫ u uniformly on K ⊂⊂ Gǫ as k → ∞,

while the same type of convergence is implied by HL2 -convergence on the left side.
n+1
Therefore, the two limits agree as holomorphic functions on Gǫ when u ∈ H s− 4 (G),
and the proposition follows.
 MAPPING HOLOMORPHIC SOBOLEV SPACES 13

5.1. On a Compact Riemannian Manifold M . Let M carry the metric g.

The key is the following simple lemma:
Lemma 5.8. Let d ∈ N and let P ∈ Ψdphg (M ) be elliptic and formally self-adjoint.
Assume that P has classical principal symbol p positive on T ∗ M \ 0, and

σ(P ) ⊂ [0, ∞).

Then, if A : C ∞ (M ) → C ∞ (M ) is continuous with [A, P ] = 0, the following holds:

1
1. [A, P d ] = 0.
1
2. [A, e−ǫP d ] = 0 for any ǫ > 0.

Proof. By hypothesis, P is parameter-elliptic w.r.t. any closed sector in C\(0, ∞).

Thus the two operators are defined, given appropriate contours in the complex plane.
That is, we can choose R > 0 so that if u ∈ C ∞ (M ), we have
Z
1 1 1−d
P u=
d λ d (λI − P )−1 P u dλ,
2πi ΓR

and
Z Z
−tP
1
1 1
−tλ d −1 1
e d
u= e (λI − P ) u dλ + (λI − P )−1 u dλ,
2πi Γ′R 2πi RS1

1−d
where λ 7→ λ d is defined by the principal logarithm with branch cut along (−∞, 0],
and ΓR and Γ′R are keyhole contours, RS1 encircles no eigenvalues except possibly 0.
The commutators are easily calculated to be
1
 1 Z 1−d
 1
[A, P d ]u = A λ d (λI − P )−1 P u dλ − P d Au
2πi ΓR
Z h i
1 1−d
= λ d A(λI − P )−1 P − (λI − P )−1 P A u dλ = 0,
2πi ΓR

and also
1  1 Z 1  1
[A, e−ǫP d ]u = A e−ǫλ d (λI − P )−1 u dλ − e−ǫP d Au + 0
2πi Γ′R
Z 1 h i
1
= e−ǫλ d A(λI − P )−1 − (λI − P )−1 A u dλ = 0.
2πi Γ′R

In the above lemma, I + P is always invertible, (I + P )s is defined for any s ∈ R.

Of course, in that case, A also commutes with any such power:

[A, (I + P )s ] = 0,

where the argument is similar to the above calculations.

Assume that P and ǫ′0 ≤ ǫ0 are chosen so Conjecture 4.1 holds, and ǫ ∈ (0, ǫ′0 ).
The operator P satisfies the conditions of Lemma 5.8, so the calculus is applicable.
This is possible in at least one non-trivial case, P = ∆g , by Theorem 4.2.
14 D.S. WINTERROSE

Let h be the Kähler metric on the closure of MǫΦ , and let s ∈ R be arbitrary.
n+1 1
Definition 5.9. If f ∈ H s− 4 (M ), Pǫ f is the extension of e−ǫP d f to MǫΦ .
Suppose Kǫ is the Poisson operator solving the Dirichlet problem for ∆h on MǫΦ .
As before, by Theorem 5.2 it is a linear homeomorphism
1
Kǫ : H s− 2 (∂MǫΦ ) → H̄ s (MǫΦ ) ∩ ker(∆h |H̄ s (MǫΦ ) ),

and Pǫ = Kǫ Sǫ realizes a well-defined linear homeomorphism

n+1 1
Pǫ : H s− 4 (M ) → Kǫ Os− 2 (∂MǫΦ ).

Definition 5.10.
1
HH s (MǫΦ ) = Kǫ Os− 2 (∂MǫΦ ).

Note that I + P ∈ Ψdphg (M ) is invertible, the inverse always belongs to Ψ−d

phg (M ).
So each space is a Hilbert space with the inner product induced by the norm
s
||u||HH s (MǫΦ ) = ||Pǫ (I + P ) d Pǫ−1 u||L2 (MǫΦ ) if u ∈ HH s (MǫΦ ),

and, automatically, Pǫ becomes a bounded isomorphism onto this space.

′
Theorem 5.11. Let d′ ∈ R and let A ∈ Ψd (M ) commute with P , [A, P ] = 0.
Then, if s ∈ R, the diagram commutes, and consists of bounded operators:

Pǫ APǫ−1 ′
HH s (MǫΦ ) HH s−d (MǫΦ )

Rǫ Rǫ

A ′
H s (M ) H s−d (M )

Proof. By construction, we have

1
Rǫ Pǫ | s−
n+1 = e−ǫP d | n+1 for any s ∈ R,
H 4 (M) H s− 4 (M)

and so by Lemma 5.8, if u ∈ C ∞ (M ) this implies

1 1
Rǫ Pǫ Au = e−ǫP d Au = Ae−ǫP d u = ARǫ Pǫ u,

which then extends by continuity to all u ∈ H s (M ). Therefore the diagram commutes.

In particular, if u ∈ HH s (Mǫ ), we have
s s
(I + P ) d Rǫ u = Rǫ Pǫ (I + P ) d Pǫ−1 u,

and so we get
s′ s′
||Rǫ u||H s′ (M) ≤ ||(I + P )− d ||B(H s′ (M),L2 (M)) ||(I + P ) d Rǫ u||L2 (M)
s
≤ ||(I + P )− d ||B(H s′ (M),L2 (M)) ||Rǫ ||B(L2 (MǫΦ ),L2 (M)) ||u||HH s′ (MǫΦ ) ,

which shows that the Rǫ are bounded in the diagram

 MAPPING HOLOMORPHIC SOBOLEV SPACES 15

The fact that the conjecture holds for P = ∆g leads to some interesting corollaries.
Those operators commuting with ∆g always preserve ǫ-extendible functions:
′
Corollary 5.12. Let d′ ∈ R and let A ∈ Ψd (M ) commute with ∆g , [A, ∆g ] = 0.
Then, if s ∈ R, the diagram commutes, and consists of bounded operators:

Pǫ APǫ−1 ′
HH s (Mǫ ) HH s−d (Mǫ )

Rǫ Rǫ

A ′
H s (M ) H s−d (M )

Proposition 5.13. Let d′ ∈ R. Let {ηk }∞ k=0 be a sequence of complex numbers.

d′
If there is a C > 0 with |ηk | ≤ Chλk i for all k ∈ N0 , we can meaningfully define
∞
X
A : C ∞ (M ) → C ∞ (M ) : u 7→ ηk (u, φk )L2 (M) φk .
k=0

Then A is continuous, [A, ∆g ] = 0, and the conclusion of Corollary 5.12 holds.

Proof. Taking any s ∈ R and u ∈ H s (M ), we have
∞
X ′
||Au||2H s−d′ (M) = hλk i2(s−d ) |ηk |2 |(u, φk )L2 (M) |2
k=0
X∞   2
 s 
≤C  (I − ∆g )− 2 u, φk  = C||u||2H s (M) ,
L2 (M)
k=0

and so by the Sobolev embedding theorem, A is continuous from C ∞ (M ) into itself.

It is not necessary that A ∈ Ψ(M ), we can use Lemma 5.8 just as in Theorem 5.11.
Repeating the arguments lead to the conclusions of Corollary 5.12
Corollary 5.14. ∆g admits a parametrix for which the diagram commutes.
1
Proof. Put ηk = λk when λk 6= 0 and zero otherwise.
Thus, the commutant of ∆g consists of operators with the ǫ-extension property.
In particular, any operator in the functional calculus of ∆g is also in the commutant,
and we will soon find an even larger algebra on G.

The most interesting implication of the Conjecture 4.1 is a generalization of this.

′
Let d′ > 0 and assume that A ∈ Ψdphg (M ) is analytic, elliptic and formally normal.
In that case, let a be the classical principal symbol of A, and put

P = A∗ A ∈ Ψdphg (M ) with d = 2d′ ,

which then has the unique classical principal symbol

p = |a|2 and p|T ∗ M\0 > 0.

This P is elliptic, formally self-adjoint, and can not have any negative eigenvalues.
But, we still need a convexity assumption to satisfy the conditions of Conjecture 4.1,
and so we require of a that {ξ ∈ Tx∗ M | |a(x, ξ)| ≤ 1} is strictly convex for any x ∈ M .
16 D.S. WINTERROSE

′
Theorem 5.15. A admits a parametrix B ∈ Ψ−d ∗
phg (M ) such that [A A, B] = 0.
∗
Consequently, if Conjecture 4.1 holds for P = A A, the diagram commutes:

Pǫ BPǫ−1 ′
HH s (MǫΦ ) HH s+d (MǫΦ )

Rǫ Rǫ

B ′
H s (M ) H s+d (M )

Proof. If A is invertible, A−1 commutes with P on C ∞ (M ), and so we are done.

Assume that A is not invertible. Then 0 is an eigenvalue of some multiplicity N ∈ N.
The spectrum of P consistsR only of eigenvalues, {ηk }∞ k=1 , counted with multiplicity,
and, by [16, Theorem 9.3], Γ λ1 (λI − P )−1 dλ converges in B(H N (M )) for all N ∈ N,
where Γ is a circle inside the resolvent set of P , enclosing 0 but no other eigenvalues.
By the Sobolev embedding theorem, we get a continuous operator
Z
∞ ∞ 1 1
B : C (M ) → C (M ) : u 7→ − (λI − P )−1 A∗ u dλ.
2πi Γ λ

Take u ∈ C ∞ (M ), and use the formal normality of A to write

Z
1  1 
[P, B]u = − P (λI − P )−1 A∗ u dλ − BP u
2πi λ
Z Γh i
1 1
=− P (λI − P )−1 A∗ − (λI − P )−1 A∗ P u dλ = 0,
2πi Γ λ

and, if {ψk }∞ ∞
k=1 is an ON eigenbasis associated to {ηk }k=1 , we have
Z
1 1
ABu = BAu = − (λI − P )−1 P u dλ
2πi Γ λ
h 1 Z 1 i 1
Z hX∞ i
= dλ u − (λI − P )−1 (u, ψk )L2 (M) ψk dλ
2πi Γ λ 2πi Γ
k=1
∞
X h 1 Z 1 i
= Iu − (u, ψk )L2 (M) dλ ψk
2πi Γ λ − ηk
k=1
N
X
= Iu − (u, ψk )L2 (M) ψk ,
k=1

where the remainder is simply the projection onto the finite-dimensional kernel of A.
′
So B can only differ from a parametrix B ′ ∈ Ψ−d
phg (M ) of A by a smoothing operator:

(B − B ′ )u = B(I − AB ′ )u − (I − BA)B ′ u
Z  XN 
= B[R(·, y)](x) − ψk (x)((B ′ )∗ ψk )(y) u(y)ω0 (y),
M k=1

where ω0 is the 1-density relative to which L2 (M ) is defined, R the kernel of I − AB ′ .

′
It follows that B ∈ Ψ−d
phg (M ), and that it is a parametrix for A.
 MAPPING HOLOMORPHIC SOBOLEV SPACES 17

An analogue of [18, pp. 6, Theorem 1] holds for the holomorphic Sobolev spaces.
The proof is very similar, but we generalize it, and fill in some details.
Theorem 5.16. Suppose that Conjecture 4.1 holds for an operator P as in 4.1,
− n−1
and that the classical principal symbol of Sǫ∗ Sǫ ∈ Ψphg2 (M ) is positive on T ∗ M \ 0.
Then, given any s ∈ R, the following holds:
1. There is a well-defined, positive operator
s s 1 s n+1
Qǫ = ((I + P ) d Sǫ∗ (Kǫ∗ Kǫ )Sǫ (I + P ) d )− 2 (I − ∆g ) 2 ∈ Ψphg
4
(M ),

2. It unitarizes Pǫ , in that sense that

||Pǫ Qǫ u||HH s (MǫΦ ) = ||u||H s (M) for any u ∈ H s (M ).

Here Kǫ∗ is understood to be the formal L2 -adjoint of Kǫ obtained from Theorem 5.2,
s
and H s (M ) is normed so that (I − ∆g ) 2 : H s (M ) → L2 (M ) is an isometry.
Proof. Using Theorems 5.2 and 4.2 and the complex Egorov theorem, we get
− n+1
Sǫ∗ (Kǫ∗ Kǫ )Sǫ ∈ Ψphg2 (M ),

which is then elliptic, and its classical principal symbol is strictly positive on T ∗ M \ 0.
It has no negative eigenvalues. We show that it is also invertible.

Regarding Pǫ as a bounded map into L2 (MǫΦ ), we form its Hilbert adjoint Pǫ∗ .
n+1 n+1
Then Pǫ∗ Pǫ : H − 4 (M ) → H − 4 (M ) is injective with dense range, and
n+1
Pǫ∗ Pǫ |C ∞ (M) = (I − ∆g ) 4 Sǫ∗ (Kǫ∗ Kǫ )Sǫ ∈ Ψ0phg (M ).
n+1
But this shows that Pǫ∗ Pǫ is a Fredholm operator on H − 4 (M ), so has closed range.
Therefore, we have a bijective and continuous, hence invertible, operator

Sǫ∗ (Kǫ∗ Kǫ )Sǫ : C ∞ (M ) → C ∞ (M ),

and the inverse is again a pseudo-differential operator by [16, pp. 70, Theorem 8.2].
Applying [16, pp. 90-91, Theorem 10.1] and [16, pp. 93, Proposition 10.3], Qǫ exists.
It follows that Qǫ is well-defined, elliptic, invertible, and also formally self-adjoint.
Taking any u ∈ C ∞ (M ), we see that

1 1

Qǫ u, u L2 (MǫΦ )
= Qǫ2 u, Qǫ2 u L2 (MǫΦ )
>0 if u 6= 0,

and
s s

||(Kǫ Sǫ )u||2HH s (MǫΦ ) = (Kǫ Sǫ )(I + P ) d u, (Kǫ Sǫ )(I + P ) d u L2 (MǫΦ )
s
∗

s
= (I + P ) (Kǫ Sǫ ) (Kǫ Sǫ )(I + P ) u, u L2 (M)
d d

s s

= (I − ∆g ) 2 Q−1 −1
ǫ u, (I − ∆g ) Qǫ u L2 (M)
2

= ||Q−1 2
ǫ u||H s (M) .
18 D.S. WINTERROSE

5.2. On a Compact Lie Group G. Let {Gǫ }ǫ>0 be the Grauert tubes of G.
It is possible to form a subalgebra of Ψ(G) with the desired properties for any ǫ > 0.
That is, operators acting as bounded maps between the holomorphic Sobolev spaces,
and with an analogous notion of, and conditions for, ellipticity within the subalgebra.
We will show that the subalgebra is non-trivial. Let d ∈ R.
Definition 5.17. Define Sǫd to be those p ∈ S d satisfying:
b Each
1. The map G → gl(dξ , C) : x 7→ p(x, ξ) is real-analytic for every [ξ] ∈ G.
of these extend holomorphically to Gǫ .
b when |Y |g < ǫ.
2. Write pY (x, ξ) = p(exp(iY )x, ξ) for every (x, [ξ]) ∈ G × G
Then {pY }|Y |g <ǫ is a bounded subset of S d .

Take p ∈ Sǫd . Let us adopt the notation pY as it is appears in the second point.
If u ∈ C ω (G) extends to Gǫ , then since G is totally real in GC , we have
X
Op(p)u(exp(iY )x) = dξ Tr(ξ(exp(iY )x)pY (x, ξ)F u(ξ))
b
[ξ]∈G
X  Z 
= dξ Tr ξ(x)pY (x, ξ) u(exp(iY )y)ξ(y)∗ dy ,
b G
[ξ]∈G

where the sum is then uniformly absolutely convergent on x ∈ G and |Y | ≤ r < ǫ.

This gives us the following simple result.
Proposition 5.18. If u ∈ C ω (G) extends holomorphically to Gǫ , so does Op(p)u.
Theorem 5.19. Let d1 , d2 ∈ R. If p ∈ Sǫd1 and q ∈ Sǫd2 , then p ⊙ q ∈ Sǫd1 +d2 .
b is fixed, then p(x, η) extends holomorphically in x ∈ G to z ∈ Gǫ .
Proof. If [η] ∈ G
This is clear from the formula
X Z  
(p ⊙ q)(z, η) = dξ Tr ξ(y)p(z, ξ) η(y)∗ q(zy −1 , η) dy,
b G
[ξ]∈G

and, if |Y |g < ǫ, it also shows that

(p ⊙ q)Y (x, η) = (p ⊙ q)(exp(iY )x, η)

= (pY ⊙ qY )(x, η).

Then, because {pY }|Y |g <ǫ ⊂ S d1 and {pY }|Y |g <ǫ ⊂ S d2 are bounded sets by definition,
and the symbolic product is continuous, {(p ⊙ q)Y }|Y |g <ǫ is a bounded set in S d1 +d2 .
This shows that p ⊙ q ∈ Sǫd1 +d2 .
It therefore follows that Ψǫ (G) = ∪d∈R Op Sǫd is actually a subalgebra of Ψ(G).
The holomorphic extension of Op(p)u can also be expressed in terms of {Op(pY )}|Y |g <ǫ .
In fact, with x and Y as above, we have

Op(p)u(z) = Op(pY )(Lexp(iY ) u)(x) if z = exp(iY )x ∈ Gǫ ,

which by itself is not a pseudo-differential operator.

 MAPPING HOLOMORPHIC SOBOLEV SPACES 19

Next, we get the holomorphic mapping property using the Cartan decomposition.
It mirrors precisely the standard Sobolev mapping property of Ψ(G).
Theorem 5.20. If p ∈ Sǫd and s ∈ R, then Op(p) admits a unique ǫ-extension.
That is, the diagram commutes, and consists of bounded operators:

^
Op(p)
HH s (Gǫ ) HH s−d (Gǫ )

Rǫ Rǫ

Op(p)
H s (G) H s−d (G)
s
Proof. Because (I − ∆G ) 2 ∈ OpSǫs for all s ∈ R it is enough to prove s = d = 0.
Using the boundedness of {pY }|Y |g <ǫ , Lemma 3.8, and [9, Lemma 5], we get
Z Z hZ i dY
2
|Op(p)u(z)| dz = |[Op(p)u](exp(iY )x)|2 dx
Gǫ |Y |g <ǫ G Θ(Y )2
Z h Z i dY
= |[Op(pY )Lexp(iY ) u](x)|2 dx
|Y |g <ǫ G Θ(Y )2
Z hZ i dY
≤C |u(exp(iY )x)|2 dx
|Y |g <ǫ G Θ(Y )2
Z
=C |u(z)|2 dz,
Gǫ

where C > 0 is the constant in the uniform bound

||Op(pY )Lexp(iY ) u||2L2 (G) ≤ C||Lexp(iY ) u||2L2 (G) ,

and Y 7→ Θ(Y )−2 is the Ad(G)-invariant Jacobian from [9, Lemma 5].
Remark 5.21. Take any two symbols p ∈ Sǫd1 and q ∈ Sǫd2 for some d1 , d2 ∈ R.
Then (p ⊙ q)Y = pY ⊙ qY so {pY ⊙ qY − pY qY }|Y |g <ǫ ⊂ S d1 +d2 −1 is a bounded set,
and we must have

p ⊙ q − pq ∈ Sǫd1 +d2 −1 .

It does not appear possible to include more terms, as these involve the δx operator,
which contains a cutoff, so the higher order terms may not extend holomorphically.
Nevertheless, we only need this to hold for the principal term.
Proposition 5.22. The space Sǫd contains non-trivial symbols for each d ∈ R.
Proof. The weighted Bargmann space HL2 (GC , νt ) of [8] is infinite-dimensional.
Hence so is the space of the restrictions of these functions to Gǫ , and so also O(Gǫ ).
d
Let Sinv be the d ∈ R symbols of bi-invariant operators on G (depending only on [ξ]).
Then, we have that
d
O(Gǫ′ )|Gǫ ⊗ Sinv ⊂ Sǫd when ǫ′ > ǫ,

and this space is infinite-dimensional, the symbols depend on both z and [ξ].
20 D.S. WINTERROSE

The spaces Op Sǫk contain the real-analytic differential operators of degree k ∈ N.

This is easy to prove by writing out their matrix-symbols. Put

gǫ = {Y ∈ g | |Y | < ǫ}.

In the following, (X1 , · · · , Xn ) is an ordered basis of left-invariant vector fields for g,

and we write X β = X1β1 ◦ · · · ◦ Xnβn for any β = (β1 , · · · , βn ) ∈ Nn0 .
Proposition 5.23. If P ∈ Diffk (G) is real-analytic, P ∈ Op Sǫk for some ǫ > 0.
Proof. In any sufficiently small chart U of G, we have
X
P |U = gβ X β ,
|β|≤k

where gβ ∈ C ω (U ) extend holomorphically into exp(igǫ )U for ǫ > 0 depending on U .

It follows that if z = exp(iY )x ∈ exp(igǫ )U , we have
X
ξ(z)−1 P ξ(z) = ξ(x)−1 ξ(exp(iY ))−1 gα (z)X β ξ(exp(iY )x)
|β|≤k
X h i
= gβ (exp(iY )x) ξ(x)−1 X β ξ(x)
|β|≤k
X
= gα (exp(iY )x)pβ (ξ),
|β|≤k

and (x, Y ) 7→ gα (exp(iY )x) has all its x-derivatives uniformly bounded in Y ∈ gǫ .
The symbols pβ ∈ S |β| are independent of all (x, Y ) ∈ G × gǫ , and given by

pβ (ξ) = ξ(x)−1 X β ξ(x) = (X β ξ)(e).

Taking a finite cover of G by such U , we get ǫ > 0 so {pY }Y ∈gǫ is bounded in S k .

Thus we have P ∈ Op Sǫd for some ǫ > 0.
Corollary 5.24. If P above is left-invariant, then P ∈ Op Sǫk for all ǫ > 0.
Proof. In this case, the coefficients gα are globally defined and constant.
To characterize ellipticity for these symbols, we need to sum them asymptotically.
It will be necessary to use the following, similar to [14, pp. 20, Lemma 7.2].
Lemma 5.25. Let d ∈ R and p ∈ Sǫd , and put

R(z) = Fξ−1 [p(z, ξ)] ∈ D′ (G) for any z ∈ Gǫ ,

where (x, Y ) 7→ p(exp(iY )x, ξ) has been extended by continuity to x ∈ G and |Y | ≤ ǫ.

It is alternatively viewed as a map

G × gǫ → D′ (G) : (x, Y ) 7→ R(exp(iY )x) = R(x, Y ).

Then, we have
n
R ∈ C ∞ (G × gǫ , H −d−⌈ 2 ⌉ (G)),

and the x-derivatives of R are uniformly bounded in Y up to gǫ .

 MAPPING HOLOMORPHIC SOBOLEV SPACES 21

Proof. Moving derivatives under Fξ−1 , we can differentiate R in the weak sense.
The Xxβ derivatives are just

Xxβ R(x, Y ) = Fξ−1 [Xxβ pY (x, ξ)] in C ∞ (G × gǫ , D′ (G)),

Using the Parseval identity, if x ∈ G, we have

X n
 
||R(x, Y )||H −d−⌈ n2 ⌉ (G) = dξ hξi−2d−2⌈ 2 ⌉ Tr pY (x, ξ)∗ pY (x, ξ)
b
[ξ]∈G
X n
h i2
≤ d2ξ hξi−2⌈ 2 ⌉ hξi−d ||pY (x, ξ)|| < ∞,
b
[ξ]∈G

n
and, similarly, for any weak x-derivative of x 7→ R(x, Y ), we find it in H −d−⌈ 2 ⌉ (G).
n
So (x, Y ) 7→ R(x, Y ) ∈ H −d−⌈ 2 ⌉ (G) has weak x-derivatives uniformly bounded in Y .
It remains to show smoothness in norm, the derivatives will agree with the weak ones.
In fact, even more is true. The function z 7→ R(z) is strongly holomorphic.

It is sufficient to prove weak holomorphy due to [13, pp. 82-83, Theorem 3.31].
By localizing, we may assume that z runs inside a poly-disc D(0, δ)n about 0 in Cn .
n
Take v ∈ H d+⌈ 2 ⌉ (G) and any closed piecewise C 1 curve γ in D(0, δ), and write
Z Z
hv, R(z)i dzj = hv, Fξ−1 [p(z, ξ)]i dzj
γ γ
Z hZ d+⌈ n ⌉ d+⌈ n ⌉ i
2 2
= (I − ∆G ) 2 v(x)Fξ−1 [hξi− 2 p(z, ξ)] dx dzj
γ G
Z h X  i
= dξ Tr F v(ξ)∗ p(z, ξ) dzj
γ b
[ξ]∈G
X  Z 
= dξ Tr F v(ξ)∗ p(z, ξ) dzj = 0.
b γ
[ξ]∈G

It follows from Morera’s theorem and Osgood’s lemma that

n
R ∈ O(Gǫ , H −d−⌈ 2 ⌉ (G)),
n
and G × gǫ → H −d−⌈ 2 ⌉ (G) : (x, Y ) 7→ R(x, Y ) is then norm-differentiable a fortiori.
The x-derivatives are uniformly bounded in Y up to gǫ , since the weak ones are.
d
Lemma 5.26. Let {pj }∞j=0 be a sequence with pj ∈ Sǫ j and dj ց −∞ as j → ∞.
P ∞
Then we can construct p ∈ Sǫd0 with p ∼ j=0 pj in the sense that
k−1
X
p− pj ∈ Sǫdk for each k ∈ N.
j=0

Proof. In this proof, we let β ∈ Nn0 and α ∈ Nm×m

0 be arbitrary multi-indices.
Take an approximate identity {ψt }t>0 ⊂ C ∞ (G) for the convolution product ∗ on G.
In this way, if f ∈ H s (G) for any s ∈ R, we have

||ψt ∗ f − f ||L2 (G) → 0 as t → 0.

22 D.S. WINTERROSE

d
Construct functions Rj : G × gǫ → D′ (G) from pj ∈ Sǫ j just as in Lemma 5.25.
n
Then Rj ∈ C ∞ (G × gǫ , H −dj −⌈ 2 ⌉ (G)), and we can pick {tj }∞
j=0 such that

1
sup ||Xxβ Rj (x, Y ) − ψtj ∗ (Xxβ Rj (x, Y ))||H −dj −⌈ n2 ⌉ (G) < when |β| ≤ j.
(x,Y )∈G×gǫ 2j+1

Take N ∈ N0 with −dj − ⌈ n2 ⌉ ≥ 0 for j ≥ N , and observe that

||pj (z, ξ)(1 − F ψtj (ξ))||2 ≤ ||pj (z, ·)(1 − F ψtj (·))||2ℓ2 (G)
b

= ||Rj (z) − ψtj ∗ (Rj (z))||2L2 (G)

1
≤ ||Rj (z) − ψtj ∗ (Rj (z))||2H −dj −⌈ n2 ⌉ (G) < ,
4j+1
b
where the estimates are uniform in z = exp(iY )x for (x, Y ) ∈ G × gǫ and [ξ] ∈ G.
Consequently, we have
∞
X
||pj (z, ξ)(1 − F ψtj (ξ))|| < 1.
j=N

It follows that p ∈ Sǫd0 if we put

∞
X
p(z, ξ) = pj (z, ξ)(1 − F ψtj (ξ)) b
for any (z, [ξ]) ∈ Gǫ × G,
j=0

which converges absolutely uniformly (in matrix norm) on z ∈ Gǫ for each [ξ] ∈ G, b
β
and x 7→ pY (x, ξ) is differentiable, Xx falls onto x 7→ pj (exp(iY )x, ξ) under the sum.
Let rN denote the sum starting at j = N . Take k ∈ N, and write
k−1
X N
X −1 N
X −1
p(z, ξ) − pj (z, ξ) = pj (z, ξ) − pj (z, ξ)F ψtj (ξ) + rN (z, ξ),
j=0 j=k j=k

where the first term is in Sǫdk , and the second belongs to Sǫ−∞ by Lemma 3.3 for δξα .
It suffices that rN ∈ Sǫdk for large N ∈ N0 . But by [14, Lemma 7.1], we get
h i
hξi|α|−dk ||δξα Xxβ (rN )Y (x, ξ)|| ≤ Cα,β sup hξi|α|−dk ||Xxβ (rN )Y (x, ξ)||
b
[ξ]∈G
hX
∞ i
≤ Cα,β ||Xxβ (pj )Y (x, ·)(1 − F ψtj (·))||H |α|−dk (G)
b
j=N
hX
∞ i
= Cα,β ||Xxβ Rj (z) − ψtj ∗ Xxβ Rj (z)||H |α|−dk (G) ,
j=N
| {z }
≤1

where N is chosen so that

n
dN + ⌈ ⌉ < dk − |α|,
2
and Cα,β > 0 is a constant depending only on α and β.
 MAPPING HOLOMORPHIC SOBOLEV SPACES 23

The above means that we can sum symbols in the {Sǫd }d∈R classes asymptotically.
This allows us to get inverses in Sǫ−d mod Sǫ−∞ for elliptic symbols in Sǫd .
Theorem 5.27. Let q0 ∈ Sǫ−d . The following holds:
1. If q0 p − 1 ∈ Sǫ−1 , then there is a qL ∈ Sǫ−d such that qL ⊙ p − 1 ∈ Sǫ−∞ .
2. If pq0 − 1 ∈ Sǫ−1 , then there is a qR ∈ Sǫ−d such that p ⊙ qR − 1 ∈ Sǫ−∞ .
Finally, if both qL and qR as above exist, then qL − qR ∈ Sǫ−∞ .
Proof. If both exist, then qL − qR = qL ⊙ (1 − p ⊙ qR ) − (1 − qL ⊙ p) ⊙ qR ∈ Sǫ−∞ .
Let N ∈ N. In either case, Lemma 5.26 allows us to build the inverse q as follows.

r = 1 − q0 ⊙ p. Define the symbol sequence qj = r⊙j ⊙ q0 for j ∈ N0 .

Left: PutP
∞
Then put q ∼ j=0 qj with q ∈ Sǫ−d , and write

 N
X −1 
Sǫ−N ∋ q − qj ⊙ p − r⊙N = q ⊙ p − 1.
j=0

Right: Put
P∞r = 1 − p ⊙ q0 . Define the symbol sequence qj = q0 ⊙ r⊙j for j ∈ N0 .
Then put q ∼ j=0 qj with q ∈ Sǫ−d , and write

 N
X −1 
Sǫ−N ∋ p ⊙ q − qj − r⊙N = p ⊙ q − 1.
j=0

b so that:
Theorem 5.28. p is elliptic in Sǫd if and only if there is a finite F ⊂ G
b \ F ).
1. p(z, ξ) is invertible for all (z, [ξ]) ∈ Gǫ × (G
2. The family of inverses satisfy

sup hξid ||p(z, ξ)−1 || < ∞.

b )
(z,[ξ])∈Gǫ ×(G\F

Proof. Assume p has a parametrix in Sǫd . That is, q0 ∈ Sǫ−d with pq0 − 1 ∈ Sǫ−1 .
b we have
In that case, there is a C > 0 so that, uniformly in (z, [ξ]) ∈ Gǫ × G,

||p(z, ξ)q0 (z, ξ) − I|| ≤ Chξi−1 ,

and so (pq0 )(z, ξ) is invertible for hξi ≥ R > 0. So is p, and

 
||p(z, ξ)−1 || ≤ ||q0 (z, ξ)|| (p(z, ξ)q0 (z, ξ))−1  ≤ C ′ hξi−d ,

where C ′ > 0, and the inequality holds for [ξ] ∈ G b except the finite set with hξi < R.
Conversely, if the two points above hold, consider the bounded subset {pY }Y ∈gǫ ⊂ S d .
It fulfils the hypotheses of Lemma 3.9 with K = gǫ and the above estimate over z ∈ Gǫ .
b then χF ∈ S −∞ , it gives
Thus, if we put χF (ξ) = 1F ([ξ])Idξ for all [ξ] ∈ G, ǫ

(χF + (1 − χF )p)−1 ∈ Sǫ−d ,

and so p ∈ Sǫd has a parametrix in Sǫd .

24 D.S. WINTERROSE

As a consequence, elliptic left-invariant operators admit a parametrix in Op Sǫ .

More generally, only the leading term in p has to belong to such an operator:
Proposition 5.29. Suppose that p of P = Op(p) ∈ Ψk (G) is of the form

p = pk + pk−1 ,

where the following holds:

1. pk ∈ S k is elliptic and independent of x ∈ G.
2. pk−1 ∈ Sǫk−1 for a fixed ǫ > 0.
Then P has a parametrix in Op Sǫ−k for this particular ǫ > 0.
Proof. Observe that

sup hξi−k ||pY (x, ξ) − pk (ξ)|| = Oλξ →∞ (hξi−1 ),

(x,Y )∈G×gǫ

and the symbols pY (x, ξ) are invertible for all (x, Y ) ∈ G × gǫ if hξi is large enough.
This is because pk (ξ) is invertible for all sufficiently large hξi by [15, Theorem 4.1].
Furthermore, it gives a finite F ⊂ G b and a C > 0 such that

||hξik pk (ξ)−1 || ≤ C b \ F,
for all [ξ] ∈ G

and for hξi suitably large, we can then write

 −1 
 
hξik ||pY (x, ξ)−1 || =  I + pk (ξ)−1 pk−1 (exp(iY )x, ξ) hξik pk (ξ)−1 
 −1
≤ 1 − ||pk (ξ)−1 pk−1 (exp(iY )x, ξ)|| ||hξik pk (ξ)−1 ||
 −1
≤ 1 − Chξi−k ||pk−1 (exp(iY )x, ξ)|| C,

where the last sum is made smaller than 12 for all (x, Y ) ∈ G×gǫ if hξi is large enough.
Therefore the conditions of Theorem 5.28 are fulfilled.
In other words, the ”leading term” determines if P has the property that we seek.
These observations indicate that Op Sǫd contain many non-trivial operators.
Example 5.30. Operators in Ψ(Tn ), acting on u ∈ C ∞ (Tn ), are of the form
X
Op(p)u(x) = eik·x p(x, k)FTn u(k) for all x ∈ Tn ,
k∈Zn

and we may take p to be the symbol

n
X
p(x, k) = |k|2 + gj (x)kj + f (x),
j=1

where {gj }nj=1 , f are bounded holomorphic on Tnǫ (2π-periodic on the polystrip Rnǫ ).
Quantization of this particular symbol gives the Laplacian ∆Tn plus lower order terms.
The result above says that p is elliptic in Sǫ2 (Tn × Zn ), e.g. via p(x, k)−1 for large k,
which is not a sum of products of functions in x and k separately.
 MAPPING HOLOMORPHIC SOBOLEV SPACES 25

6. Outlook. The Boutet de Monvel theorem/conjecture is yet not fully proven,

but it has important implications for differential operators on real-analytic manifolds.
We expect that the results on Lie groups extend to compact homogeneous spaces,
where the chosen metric is related to a bi-invariant metric on the group acting on it.
Our hope is that an analogous conjecture can be made for manifolds with boundary,
and, if true, that it reveals a path to studying real-analytic boundary value problems
in Sobolev spaces of holomorphic functions.

REFERENCES

[1] L. Boutet de Monvel, Boundary problems for pseudo-differential operators, Acta Math., 126
(1971), pp. 11–51.
[2] L. Boutet de Monvel, Convergence dans le domaine complexe des series de fonctions propres,
Journées Équations aux dérivées partielles, (1978).
[3] L. Boutet de Monvel, On the Index of Toeplitz Operators of Several Complex Variables,
Inventiones Mathematicae, 50 (1979), pp. 249–272.
[4] L. Boutet de Monvel and V. Guuillemin, The Spectral Theory of Toeplitz Operators, vol. 99,
Princeton University Press, 1981.
[5] G. B. Folland, A Course in Abstract Harmonic Analysis, Chapman and Hall/CRC, 2015.
[6] G. Grubb, Functional Calculus of Pseudodifferential Boundary Problems, vol. 65 of Progress
in Mathematics, Birkhäuser Basel, 1996.
[7] G. Grubb, Distributions and Operators, vol. 252 of Graduate Texts in Mathematics, Springer,
2008.
[8] B. Hall, The Segal-Bargmann ”Coherent State” Transform for Compact Lie Groups, Journal
of Functional Analysis, 122 (1994), pp. 103–151.
[9] B. Hall, Phase Space Bounds for Quantum Mechanics on a Compact Lie Group, Communi-
cations in Mathematical Physics, 184 (1997), p. 233–250.
[10] L. Hörmander, Fourier integral operators. I., Acta Mathematica, 127 (1971), pp. 79–183.
[11] M. Karamehmedović, On analytic continuability of the missing cauchy datum for helmholtz
boundary problems, Proceedings of the American Mathematical Society, 143 (2015),
pp. 1515–1530.
[12] A. Melin and J. Sjöstrand, Fourier integral operators with complex-valued phase functions,
in Chazarain J. (eds) Fourier Integral Operators and Partial Differential Equations, vol. 459
of Lecture Notes in Mathematics, Springer, Berlin, Heidelberg, 1975.
[13] W. Rudin, Functional Analysis, McGraw-Hill, 2 ed., 1991.
[14] M. Ruzhansky and J. Wirth, Global functional calculus for operators on compact Lie groups,
Journal of Functional Analysis, 267 (2006), pp. 144–172.
[15] M. V. Ruzhansky, V. Turunen, and J. Wirth, Hörmander Class of Pseudo-Differential
Operators on Compact Lie Groups and Global Hypoellipticity, Journal of Fourier Analysis
and Applications, 20 (2014), p. 476–499.
[16] M. A. Shubin, Pseudodifferential Operators and Spectral Theory, Springer, 2001.
[17] M. Stenzel, On the analytic continuation of the Poisson kernel, Manuscripta Mathematica,
144 (2014), p. 253–276.
[18] M. Stenzel, The Poisson Transform on a Compact Real-Analytic Riemannian Manifold,
Monatshefte für Mathematik, 178 (2015), pp. 299–309.
[19] M. Stenzel, F. Golse, and E. Leichtnam, Intrinsic Microlocal Analysis and Inversion For-
mulae for the Heat Equation on Compact Real-Analytic Riemannian Manifolds, Annales
Scientifiques de l’École Normale Supérieure, 29 (1996), pp. 669–736.
[20] D. S. Winterrose, A holomorphic mapping property of analytic pseudo-differential operators,
to appear, (2021).
[21] D. S. Winterrose, Complexifications, Pseudodifferential Operators, and the Poisson Trans-
form, PhD thesis, Technical University of Denmark, 2021.
[22] S. M. Zelditch, Complex zeros of real ergodic eigenfunctions, Inventiones mathematicae, 167
(2007), p. 419–443.
[23] S. M. Zelditch, Pluri-Potential Theory on Grauert Tubes of Real Analytic Riemannian Man-
ifolds, I, Spectral Geometry, 84 (2012), pp. 299–339.
[24] S. M. Zelditch, Ergodicity and intersections of nodal sets and geodesics on real analytic
surfaces, Journal of Differential Geometry, 96 (2014), pp. 305–351.