Journal of Functional Analysis 289 (2025) 111099

Contents lists available at ScienceDirect

Journal of Functional Analysis

journal homepage: www.elsevier.com/locate/jfa

Regular Article

Calculus for parametric boundary problems with

global projection conditions
Jörg Seiler
Dipartimento di Matematica, Università di Torino, Italy

a r t i c l e i n f o a b s t r a c t

Article history: A pseudodifferential calculus for parameter-dependent oper­

Received 16 January 2025 ators on smooth manifolds with boundary in the spirit of
Accepted 5 June 2025 Boutet de Monvel’s algebra is constructed. The calculus con­
Available online 18 June 2025
tains, in particular, the resolvents of realizations of differential
Communicated by Guido De
Philippis
operators subject to global projection boundary conditions
(spectral boundary conditions are a particular example); re­
MSC: solvent trace asymptotics are easily derived. The calculus is
primary 35S15, 47L80, 47A10 related to but different from the calculi developed by Grubb
secondary 35J58, 58J40 and Grubb-Seeley. We use ideas from the theory of pseudodif­
ferential operators on manifolds with edges due to Schulze, in
Keywords: particular the concept of operator-valued symbols twisted by
Boundary value problems with a group-action. Parameter-ellipticity in the calculus is char­
parameter acterized by the invertibility of three principal symbols: the
Global projection boundary
homogeneous principal symbol, the principal boundary sym­
conditions
Pseudodifferential operators bol, and the so-called principal limit symbol. The principal
Resolvent trace asymptotics boundary symbol has, in general, a singularity in the co­
variable/parameter space, the principal limit symbol is a new
ingredient of the calculus.
© 2025 The Author. Published by Elsevier Inc. This is an
open access article under the CC BY license (http://
creativecommons.org/licenses/by/4.0/).

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

E-mail address: joerg.seiler@unito.it.

https://doi.org/10.1016/j.jfa.2025.111099
0022-1236/© 2025 The Author. Published by Elsevier Inc. This is an open access article under the CC BY
license (http://creativecommons.org/licenses/by/4.0/).
2 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

2. Boutet’s algebra with strong parameter-dependence . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

3. Symbols with expansion at infinity -- the abstract setting . . . . . . . . . . . . . . . . . . . . . . . . 19
4. Singular Green symbols with expansion at infinity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
5. The full calculus with expansion at infinity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
6. Operators of Toeplitz type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
7. Applications to boundary value problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
8. The calculus for manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
Data availability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

1. Introduction

Parameter-dependent pseudodifferential operators have proven to be a very powerful

tool in partial differential equations and in geometric analysis. One of the main goals
in this approach is to obtain a precise description of the resolvent of certain (differen­
tial) operators and to use this structure in the study of the specific problem. Typical
applications concern maximal regularity of evolution equations, the analysis of spectral
and heat trace asymptotics, and complex powers of operators (just to name a few). This
idea has been introduced in the 1960’s, see for example the pioneering works of Seeley
[24--26] on boundary value problems, and has been refined over the following decades, in
particular under the aspect of ``algebras'' or ``calculi'' of pseudodifferential operators.
In [5], Boutet de Monvel introduced a calculus of pseudodifferential operators which
contains all classical differential boundary value problems as well as the parametrices
(i.e., inverses modulo smoothing remainders) of Shapiro-Lopatinskij elliptic problems.
Nowadays this calculus is referred to as Boutet de Monvel’s algebra. The elements in
this calculus are block-matrix operators of the form
(︃ )︃ C ∞ (M, E0 ) C ∞ (M, E1 )
A+ + G K
: ⊕ −→ ⊕ ,
T Q
C ∞ (∂M, F0 ) C ∞ (∂M, F1 )

where M is a smooth compact manifold with boundary, Ej and Fj are vector-bundles over
M and ∂M , respectively. The bundles F0 or F1 are also allowed to be zero-dimensional.
A+ is the restriction of a pseudodifferential operator A from an ambient manifold of M
to M itself, G is a so-called singular Green operator, T is a trace operator, K is a Poisson
or potential operator, Q is a pseudodifferential operator on the boundary. Elements with
A+ = 0 we shall refer to as generalized singular Green operators. Initially defined on
smooth sections, the operators extend by continuity to scales of Sobolev spaces (which
are L2 -Bessel potential spaces in this paper; however, as shown in Grubb-Kokholm [10],
one can also consider the Lp -scale). For nice introductions to this calculus we refer the
reader to Schrohe [16] and Schulze [21]. The study of resolvents of differential operators
subject to classical differential boundary conditions leads to a parameter-dependent ver­
sion of Boutet de Monvel’s algebra where the parameter enters in a ``strong'' way, i.e., as
additional co-variable; the trick to consider a parameter as additional co-variable goes
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 3

back to Agmon [1]. Details are given in Section 2.3 or can be found in Schrohe, Schulze
[17], for example.
Differential operators equipped with boundary conditions like spectral boundary con­
ditions cannot be handled in Boutet’s original calculus and not every elliptic differential
operator can be supplemented with boundary conditions to become a Shapiro-Lopatinskii
elliptic problem, since it must satisfy the so-called Atiyah-Bott condition, cf. [2]. Based
on a systematic study of such effects by Nazaikinskij, Schulze, Shatalov, and Sternin in
[14], Schulze in [20,13] introduced an extended calculus that resolves this shortcoming.
The elements of the extended calculus have the form
(︃ )︃ (︃ )︃ (︃ )︃
1 0 A+ + G K 1 0
(1.1)
0 P1 T Q 0 P0

with pseudodifferential projections (i.e., zero order idempotents) Pj on the boundary. The
operator action refers to the scale of spaces where the full Sobolev spaces are substituted
by closed subspaces determined by the projections. In this paper we extend this approach
to parameter-dependent boundary problems; in this way, resolvents of elliptic differential
operators subject to global projection boundary conditions can be treated. As it turns
out, the calculus we develop also allows to obtain resolvent trace asymptotics and thus
provides, in particular, a method to derive asymptotics of first order operators subject
to APS boundary conditions due Grubb and Seeley [11,7] and of second order operators
due to Grubb [9], see Sections 7.2 and 7.3.
Since the structure of the operators in (1.1) reminds of classical Toeplitz operators on
the unit circle, we shall also use the terminology of operators of Toeplitz type. In [28] the
author has considered such operators from a general point of view: given a calculus, does
there exist an associated calculus of Toeplitz operators? As it turns out, if the calculus
is closed under taking adjoints and if ellipticity in the calculus is characterized by the
invertibility of certain principal symbols, then there is a canonical way to characterize
ellipticity and existence of a parametrix for operators of Toeplitz type whenever the
projections belong to the original calculus. We summarize these results in Section 6. In
this set-up, the calculus may also be a parameter-dependent one. For this reason, the
main issue in this paper is actually to construct a pseudodifferential calculus which

i) contains Boutet de Monvel’s algebra of strongly parameter-dependent operators and

pseudodifferential operators on the boundary which do not depend on the parameter
and where
ii) parameter-ellipticity is characterized by the invertibility of certain principal symbols.

Point i) has already been addressed by Grubb in [8]. It is remarkable that the resulting
calculus is in appearance quite different from ours. The reason might be that [8] is focused
on resolvent trace expansions while for us point ii) is central. Let us emphasize once more
that, once such a calculus has been established, we can substitute the projections in (1.1)
by arbitrary projections from the calculus.
4 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

The present paper uses ideas of the author’s work [29] where the analogues of i) and
ii) have been addressed for operators on manifolds without boundary, i.e., a calculus
containing both strongly parameter-dependent and parameter-independent pseudodif­
ferential operators on Rn or on closed manifolds has been constructed. The paper [29]
takes much inspiration from the concept of pseudodifferential operators with finite reg­
ularity due to Grubb [6]. The regularity number indicates how far an operator is from
being strongly parameter-dependent; parameter-independent zero-order operators (in
particular projections) have regularity zero, strongly parameter-dependent operators
have infinite regularity. The monograph [6] introduces another Boutet de Monvel type
calculus for boundary problems which allows the construction of resolvents for a wide
class of operators (which includes certain nonlocal perturbations of differential operators)
but, at the same time has some limitations, for example it only yields finite resolvent
trace expansions. An important requirement for the notion of ellipticity in [6] is that of
positive regularity. The main finding of [29] is a new ``geometric'' interpretation of the
regularity as a specific kind of singular structure (i.e., non-smoothness) of symbols in
both co-variable and parameter. This permits not only to eliminate the requirement of
positive regularity in the boundaryless case but also leads to a hierarchy of principal
symbols in the calculus characterizing the ellipticity.
In the present work we shall find symbolic structures in the spirit of [29] for boundary
value problems. The homogeneous principal symbol and the principal boundary symbol
are already known from the classic Boutet de Monvel’s calculus, though in our calculus
the principal boundary symbol generally has a singularity which is absent in the classical
setting. We shall introduce two further new principal symbols, called principal limit
symbol and principal angular symbol. Let us mention that the ellipticity condition (III)
in [6, Definition 1.5.5] can be interpreted in terms of the principal angular symbol.
Let us touch upon another key aspect of this paper. For simplicity of presentation
let us focus here on Poisson operators, similar observations hold for trace and singular
Green operators. A Poisson operator on the half-space has the form
ˆ
′ ′
[op(k)(μ)u](x) = e−ix ξ k(x′ , ξ ′ , μ; xn )ˆ︁
u(ξ ′ ) d¯ξ ′ ,

where x = (x′ , xn ), u
ˆ︁ is the Fourier transform, and the symbol-kernel k has the form

k(x′ , ξ ′ , μ; xn ) = [ξ ′ , μ]1/2 k′ (x′ , ξ ′ , μ; [ξ ′ , μ]xn ),

where [·] is a smoothed norm function and k′ (x′ , ξ ′ , μ; t) is smooth and rapidly decreasing
in t ∈ R+ ; note the ``twisting'' between xn and the co-variable and parameter (ξ ′ , μ).
Denoting with κλ , λ > 0, the dilation group-action acting on functions by (κλ u)(t) =
λ1/2 u(λt), k can be written as

k(x′ , ξ ′ , μ; xn ) = (κ[ξ′ ,μ] k′ )(x′ , ξ ′ , μ; xn ).

 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 5

This leads to the fact that Poisson operators can be viewed as pseudodifferential op­
erators with operator-valued symbols twisted by a group-action in the sense of Schulze
(see the textbooks [18,19], for example); we will recall this concept in Section 2.1.1.
The mapping k ↦→ k′ = κ−1 [ξ ′ ,μ] k provides a canonical passage from twisted to un-twisted
symbols. It is important for our analysis that the operation of un-twisting persists on
the operator-level, i.e., we establish a one-to-one correspondence between twisted and
un-twisted Poisson operators, see Section 4.4.
Twisted operator-valued symbols have been systematically employed by Schulze for
the construction of calculi of pseudodifferential operators on manifolds with singularities,
in particular on manifolds with edges, see for instance [15,18]. The local model of a
manifold with edge (also called manifold with fibered boundary) is a wedge Rq × X ∧ ,
where X ∧ is an infinite cone with base X, a closed smooth manifold. The half-space
is a special case, where X is simply a point. The so-called edge algebra of Schulze is a
calculus that, on a formal level, has many similarities with Boutet de Monvel’s algebra;
there also exists a version with global projection edge-conditions (without parameter),
see [22,23,13]. The parameter-dependent edge algebra contains operators with symbol
kernels, which have a twisted structure similar to the one described above. It seems
plausible that the operation of un-twisting extends to this more general setting and thus
one can hope to apply our approach to manifolds with edges, too. This will be subject
of future work.
The structure of the paper is as follows. In Section 2 we recall the concept of operator­
valued symbols twisted by a group-action and review the strongly parameter-dependent
Boutet de Monvel calculus. In Section 3 we intoduce various symbol classes; of particu­
lar importance are the so-called symbols with expansion at infinity. In Section 4 we shall
describe generalized singular Green symbols by associated classes of symbol-kernels and
shall prove the (un-)twisting on operator-level discussed above. Enlarging Boutet’s cal­
culus with strong parameter-dependence by the class of generalized Green symbols with
expansion at infinity yields our new calculus as is discussed in Section 5. In Section 6
we summarize results on Toeplitz type pseudodifferential operators which we apply in
Section 7. In particular, we consider boundary value problems for differential operators
subject to global projection boundary conditions and derive the structure of the resolvent
of the associated realization and resolvent trace asymptotics. We compare this with pre­
vious work of Grubb. In Section 8 we indicate how the calculus can be lifted to compact
manifolds with boundary.

2. Boutet’s algebra with strong parameter-dependence

In this section we describe Boutet de Monvel’s algebra with parameter on the half­
space Rn+ . The parameter enters in a ``strong way'', which is suited for describing
resolvents of classical differential boundary value problems.
6 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

2.1. The algebra of singular Green operators

The description of the class of generalized singular Green operators is based on the
use of pseudodifferential operators with twisted operator-valued symbols in the sense of
Schulze.

2.1.1. Twisted operator-valued symbols - the abstract setting

A group-action {κλ } on a Hilbert space E is a strongly continuous map

λ ↦→ κλ : R+ −→ L (E)

such that κλ κσ = κλσ for all λ, σ > 0 and κ1 = 1. In particular, κ−1λ = κ1/λ . By a
well-known result from the theory of operator semi-groups there exist constants C ≥ 0
and M ∈ R such that

∥κλ ∥L (E) ≤ C max(λM , λ−M ).

In the following definition and throughout the whole paper, y ↦→ [y] denotes a smooth
positive function defined on Rm (we shall use the same notation independent of the
specific value of m) which coincides with the usual modulus |y| outside the unit-ball.

Definition 2.1. Let s ∈ R and E be a Hilbert space with group-action {κλ }. Define
𝒲 s (Rn−1 , E) as the closure of S (Rn−1 , E) with respect to the norm
(︂ ˆ )︂1/2
∥u∥𝒲 s (Rn−1 ,E) = [ξ ′ ]2s ∥κ−1 ˆ︁ ′ )∥2E dξ ′
[ξ ′ ] u(ξ .
Rn−1

ˆ︁ denotes the standard Fourier transform,

Here, u
ˆ
′ ′
′
ˆ︁ ) =
u(ξ e−ix ξ u(x′ ) dx′ .
Rn−1

The spaces 𝒲 s (Rn−1 , E) are called (abstract) edge Sobolev spaces associated with E. By
construction it is obvious that the Fourier multiplier with symbol κ[ξ′ ] induces an iso­
metric isomorphism H s (Rn−1 , E) −→ 𝒲 s (Rn−1 , E) for every s ∈ R where H s (Rn−1 , E)
is the standard Bessel potential space with norm ∥[·]s u∥
ˆ︁ L2 (Rn−1 ,E) .

Definition 2.2. For j = 0, 1 let Ej be Hilbert spaces with group-action {κj,λ }. The space
d
S1,0 (Rn−1 ; E0 , E1 ) consists of all smooth functions p : Rn−1 × Rn−1 → L (E0 , E1 ) with

′ ′ ′
∥κ−1 α β ′ ′ ′ d−|α |
1,[ξ ′ ] {Dξ ′ Dx′ p(x , ξ )}κ0,[ξ ] ∥L (E0 ,E1 ) ≲ [ξ ]
′

for every choice of multi-indices α′ , β ′ ∈ N0n−1 .

 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 7

With a symbol p ∈ S1,0

d
(Rn−1 ; E0 , E1 ) we associate the pseudodifferential operator

op(p) = p(x′ , D′ ) : S (Rn−1 , E0 ) −→ S (Rn−1 , E1 )

in the usual way, i.e.,

ˆ
′ ′
′
[op(p)u](x ) = eix ξ p(x′ , ξ ′ )ˆ︁
u(ξ ′ ) d¯ξ ′ ,

where integration is over Rn−1 and d¯ξ ′ = (2π)−(n−1) dξ ′ . The following theorem is a
consequence of [27, Theorem 3.14] and the use of order reducing operators.

Theorem 2.3. With the notation from Definition 2.2,

op(p) : 𝒲 s (Rn−1 , E0 ) −→ 𝒲 s−d (Rn−1 , E1 ), s ∈ R.

There is a full calculus for such pseudodifferential operators, including composition,

formal adjoint, ellipticity and parametrix construction; we will present it in the version
with parameter. In the following we shall use the notation

κ(ξ ′ , μ) := κ[ξ′ ,μ] , (ξ ′ , μ) ∈ Rn−1 × R+ . (2.1)

Definition 2.4. For j = 0, 1, let Ej be Hilbert spaces with group-action {κj,λ }. The space
d
S1,0 (Rn−1 × R+ ; E0 , E1 ) consists of all smooth functions p : Rn−1 × Rn−1 × R+ →
L (E0 , E1 ) with
′ ′ ′
∥κ−1 ′ α β ′ ′ ′ ′
1 (ξ , μ){Dξ ′ Dx′ Dμ p(x , ξ , μ)}κ0 (ξ , μ)∥L (E0 ,E1 ) ≲ [ξ , μ]
j d−|α |−j
(2.2)

for every choice of multi-indices α′ , β ′ ∈ N0n−1 and j ∈ N0 .

Given p ∈ S1,0
d
(Rn−1 × R+ ; E0 , E1 ) define pμ (x′ , ξ ′ ) := p(x′ , ξ ′ , μ). Then pμ is a sym­
d n−1
bol in S1,0 (R ; E0 , E1 ) for every μ. Thus we can consider the associated family of
pseudodifferential operators; we shall use the notation

op(p)(μ) = p(x′ , D′ , μ) := op(pμ ).

The class of regularizing or smoothing symbols is

S −∞ (Rn−1 × R+ ; E0 , E1 ) := ∩ S1,0
d
(Rn−1 × R+ ; E0 , E1 );
d∈R

it is easy to see that

S −∞ (Rn−1 × R+ ; E0 , E1 ) = S (Rn−1 × R+ , Cb∞ (Rn−1

x′ ; L (E0 , E1 ))),
8 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

i.e., smoothing symbols are rapidly decreasing in (ξ ′ , μ).

If V is a conic subset of Rm for some m, a function p : Rn−1 × V → L (E0 , E1 ) shall
be called twisted homogeneous of degree d (with respect to the variable v) provided

p(x′ , λv) = λd κ1,λ p(x′ , v) κ−1

0,λ ∀ (x′ , v) ∈ Rn−1 × V ∀ λ > 0.

Definition 2.5. For j = 0, 1 let Ej be Hilbert spaces with group-action {κj,λ }. Then
d
Shom (Rn−1 × R+ ; E0 , E1 ) denotes the space of all functions p which are twisted homo­
geneous in the above sense with V = (Rn−1 × R+ ) \ {0} and which satisfy the estimates
′ ′ ′
∥κ−1 β ′ ′ ′
1,|ξ ′ ,μ| {Dξ ′ Dx′ Dμ p(x , ξ , μ)}κ0,|ξ ,μ| ∥L (E0 ,E1 ) ≲ |ξ , μ|
α j
′
d−|α |−j
(2.3)

for every choice of multi-indices α′ , β ′ ∈ N0n−1 and j ∈ N0 .

If χ(ξ ′ , μ) is a zero-excision function (i.e., χ ∈ C ∞ (Rn−1 × R+ ) such that χ ≡ 0 in

some neighborhood of the origin and 1 − χ has compact support) and p ∈ Shom d
(Rn−1 ×
R+ ; E0 , E1 ) then χp ∈ S1,0 (R
d n−1
× R+ ; E0 , E1 ). Hence the following definition makes
sense:

Definition 2.6. For j = 0, 1 let Ej be Hilbert spaces with group-action {κj,λ }. A symbol
p ∈ S1,0
d
(Rn−1 × R+ ; E0 , E1 ) is called (twisted) poly-homogeneous provided there exists
d−j
a sequence of symbols p(d−j) ∈ Shom (Rn−1 × R+ ; E0 , E1 ) such that, for every N ∈ N0 ,

∑︂
N −1
p− χ(ξ ′ , μ)p(d−j) ∈ S1,0
d−N
(Rn−1 × R+ ; E0 , E1 ).
j=0

The space of such symbols shall be denoted by S d (Rn−1 × R+ ; E0 , E1 ), the leading

component p(d) is the so-called homogeneous principal symbol of p.

The homogeneous principal symbol can be obtained from the full symbol, i.e.,

p(d) (x′ , ξ ′ , μ) = lim λ−d κ−1 ′ ′

1,λ p(x , λξ , λμ)κ0,λ , (ξ ′ , μ) ̸= 0. (2.4)
λ→+∞

d
Theorem 2.7. Let pj ∈ S1,0j
(Rn−1 × R+ ; Ej , Ej+1 ), j = 0, 1. Then there exists a unique
symbol p1 #p0 ∈ S1,0 (R
d0 +d1 n−1
× R+ ; E0 , E2 ), the so-called Leibniz product of p0 and p1 ,
such that, for all μ,

op(p1 )(μ)op(p0 )(μ) = op(p1 #p0 )(μ).

In case both p0 and p1 are poly-homogeneous, so is p1 #p0 ; moreover

(d1 ) (d0 )
(p1 #p0 )(d0 +d1 ) = p1 p0 .
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 9

The Leibniz product of two symbols (not only here but also for all other classes of
symbols in this paper) is given by the formula
¨
′ ′
′ ′
(p#q)(x , ξ ) = Os − e−iy η p(x′ , ξ ′ + η ′ )q(x′ + y ′ , ξ ′ ) dy ′ d¯η ′

where the integral is an oscillatory integral (for the precise definition of the oscillatory
integral see Theorem 4.10, below). In this formula p and q may also depend on the
parameter μ.

Definition 2.8. A symbol p ∈ S d (Rn−1 × R+ ; E0 , E1 ) is called elliptic if its homogeneous

principal symbol is pointwise invertible whenever (ξ ′ , μ) ̸= 0 and

∥p(d) (x′ , ξ ′ , μ)−1 ∥L (E1 ,E0 ) ≲ 1 ∀ x′ ∈ Rn−1 ∀ |ξ ′ , μ| = 1. (2.5)

This estimate (2.5) is necessary (only) for ensuring the boundedness of the inverse in
−d
the x′ -variable and implies (p(d) )−1 ∈ Shom (Rn−1 × R+ ; E1 , E0 ).

Theorem 2.9. Let p ∈ S d (Rn−1 × R+ ; E0 , E1 ) be elliptic. Then there exists a symbol q ∈

S −d (Rn−1 ×R+ ; E1 , E0 ) such that q#p−1 and p#q−1 are symbols of order −∞ that van­
ish for sufficiently large μ. In particular, op(p)(μ) : 𝒲 s (Rn−1 , E0 ) → 𝒲 s−d (Rn−1 , E1 )
is invertible for large μ with op(p)(μ)−1 = op(q)(μ).

We also shall need the notion of the formal adjoint of an operator.

Definition 2.10. Let E and E ˜︁ be Hilbert spaces with group-action {κλ } and {˜︁
κλ }, respec­
˜︁
tively. Let H be a Hilbert space. We call (E, H, E) a Hilbert-triple if the inner product
of H induces a continuous non-degenerate sesquilinear pairing E × E ˜︁ → C that permits
˜︁
to identify the dual space of E with E and vice versa. Moreover, we require

(κλ e, e) ˜︁−1
˜︁ H = (e, κ ˜︁ H ∀e∈E ˜︁
∀ e˜︁ ∈ E ∀ λ > 0.
λ e)

Let (Ej , Hj , E˜︁j ), j = 0, 1, be Hilbert-triples. If T ∈ L (E0 , E1 ) then there exists a

unique operator T ∗ ∈ L (E ˜︁1 , E
˜︁0 ) such that (T e0 , e˜︁1 ) ˜︂ = (e0 , T ∗ e˜︁1 )H for every e0 ∈ E0
H
and e˜︁1 ∈ E˜︁1 . By slight abuse of language we call T ∗ the adjoint of T . Moreover, if
A : S (Rn , E0 ) → S (Rn , E1 ), then the operator A∗ : S (Rn , E ˜︁1 ) → S (Rn , E ˜︁0 ) defined
by

(Au, v)L2 (Rn ,H1 ) = (u, A∗ v)L2 (Rn ,H0 ) , ˜︁1 ),

u ∈ S (Rn , E0 ), v ∈ S (Rn , E

is called the formal adjoint of A.

Theorem 2.11. Let (Ej , Hj , E ˜︁j ), j = 0, 1, be Hilbert triples and p ∈ S d (Rn−1 ×

1,0
R+ ; E0 , E1 ). Then there exists a unique symbol p(∗) ∈ S1,0d ˜︁1 , E
(Rn−1 × R+ ; E ˜︁0 ) such
that, for all μ,
10 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

op(p)(μ)∗ = op(p(∗) )(μ).

If p is poly-homogeneous then so is p(∗) and (p(∗) )(d) = (p(d) )∗ , where ∗ on the right is
the pointwise adjoint principal symbol.

The formal adjoint symbol (not only here but also for all other classes of symbols in
this paper) is given by the formula
¨
′ ′
p(∗) (x′ , ξ ′ ) = Os − e−iy η p(x′ + y ′ , ξ ′ + η ′ )∗ dy ′ d¯η ′ .

Of course, all the above considerations include the case of trivial group-actions, i.e.,
κj ≡ 1. To emphasize this particular case we shall use a different notation:

Definition 2.12. If E0 and E1 carry the trivial group action we set

d
S1,0 (Rn−1 × R+ ; L (E0 , E1 )) := S1,0
d
(Rn−1 × R+ ; E0 , E1 ),
d
Shom (Rn−1 × R+ ; L (E0 , E1 )) := Shom
d
(Rn−1 × R+ ; E0 , E1 ),
S d (Rn−1 × R+ ; L (E0 , E1 )) := S d (Rn−1 × R+ ; E0 , E1 ).

Let us finally comment on the topology of symbol spaces. Using the estimates (2.2) and
d
(2.3) it is obvious how to define a Fréchet topology on the spaces S1,0 (Rn−1 ×R+ ; E0 , E1 )
d
and Shom (Rn−1 ×R+ ; E0 , E1 ), respectively. The class of poly-homogeneous symbols then
carries the projective topology under the maps

p ↦→ p(d−j) : S d (Rn−1 × R+ ; E0 , E1 ) −→ Shom

d
(Rn−1 × R+ ; E0 , E1 ),
∑︂
p ↦→ p − χp(d−j) : S d (Rn−1 × R+ ; E0 , E1 ) −→ S1,0
d−N
(Rn−1 × R+ ; E0 , E1 ),
j<N

with j, N ∈ N0 . This makes S d (Rn−1 × R+ ; E0 , E1 ) a Frèchet space, too. Both maps

(p1 , p0 ) ↦→ p1 #p0 and p ↦→ p(∗) are then continuous in the respective symbol spaces, cf.
Theorems 2.7 and 2.11.

2.1.2. Sobolev spaces and the standard group-action

Let us now apply the above abstract concept to operators on the half-space.

Definition 2.13. Let I = R or I = R+ . For φ ∈ D (I) define κλ φ ∈ D (I) by (κλ φ)(t) =

λ1/2 φ(λt). For a distribution T ∈ D ′ (I) define κλ T ∈ D ′ (I) by

(κλ T )(φ) = T (κ−1

λ φ), φ ∈ D (I).
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 11

It is easily seen that κλ of the previous definition induces a group-action on the spaces
H (R), H s,δ (R+ ), and H0s,δ (R+ ) for any choice of s, δ ∈ R. We shall refer to it as the
s,δ

dilation group-action. Here,

H0s (R+ ) = {u ∈ H s (R) | supp u ⊆ R+ },

H s (R+ ) = {u ∈ D ′ (R+ ) | u = v|R+ for some v ∈ H s (R)};

moreover,

H s,δ (R) = {[·]−δ u | u ∈ H s (R)}

and analogously for the other spaces. Note that every κλ is unitary as an opera­
tor on L2 (R) and L2 (R+ ), respectively. Moreover, (H s,δ (R), L2 (R), H −s,−δ (R)) and
(H0s,δ (R+ ), L2 (R+ ), H −s,−δ (R+ )) are Hilbert triples in the sense of Definition 2.10. The
restriction of the Schwartz space to the half-axis is denoted by

S (R+ ) = {u ∈ D ′ (R+ ) | u = v|R+ for some v ∈ S (R)}.

The following result is the main motivation for introducing edge Sobolev spaces in­
volving group-actions.

Theorem 2.14. Let u ∈ S (Rn ) ∼

= S (Rn−1 , S (R)). Then, for every s ∈ R,

∥u∥H s (Rn ) = ∥u∥𝒲 s (Rn−1 ,H s (R)) .

This yields an isometric isomorphism of H s (Rn ) and 𝒲 s (Rn−1 , H s (R)). By this result
one then obtains the identifications

H0s (Rn+ ) = 𝒲 s (Rn−1 , H0s (R+ )), H s (Rn+ ) = 𝒲 s (Rn−1 , H s (R+ ))

Remark 2.15. Recall notation (2.1). For each μ ∈ R+ and s ∈ R the map

op(κ)(μ) : H s (Rn−1 , H s (R+ )) −→ 𝒲 s (Rn−1 , H s (R+ )) = H s (Rn+ )

is an isomorphism with inverse op(κ−1 )(μ); we shall refer to it as the group-action oper­
ator. If u ∈ S (Rn+ ) ∼
= S (Rn−1 , S (R+ )) then we can write explicitly
(︁ ′ 1/2 ′ ′ )︁
[op(κ)(μ)u](x) = Fξ−1
′ →x′ [ξ , μ] ˆ︁ , [ξ , μ]xn ) (x′ ),
u(ξ

ˆ︁ indicates the partial Fourier transform of u in the variable x′ .

where u

In the sequel we shall frequently apply the following simple observation which is easily
proved by induction.
12 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

Lemma 2.16. Let u ∈ S (R+ ) and {κλ } be the dilation group-action. Then

dℓ 1
ℓ
κλ u = ℓ κλ Θℓ u
dλ λ
for every ℓ ∈ N, where

∏︂
ℓ−1 (︂ 1 )︂
Θℓ = t∂t + − k : S (R+ ) −→ S (R+ ).
2
k=0

Note that κλ and Θℓ commute.

If a : U ⊂ Rm → R+ is a smooth function, the previous lemma in combination with

the chain-rule yields that Dyα κa(y) u, |α| ≥ 1, is a finite linear-combination of terms of
the form

a(y)−m Dyα1 a(y) · . . . · Dyαm a(y)κa(y) Θm u,

(2.6)
with 1 ≤ m ≤ |α|, α1 + . . . + αm = α.

2.1.3. Generalized singular Green operators

In the following we shall extend the dilation group-action from Definition 2.13 to
D (I, C L ) (by component-wise application) and to D ′ (I, C L ) ⊕ C M as κλ ⊕ 1. By slight
′

abuse of notation we write 𝜿λ := κλ ⊕ 1 for any choice of L and M . Correspondingly,

we write
(︃ )︃ D ′ (I, C L ) D ′ (I, C L )
′ ′ κ(ξ ′ , μ) 0
𝜿(ξ , μ) := κ(ξ , μ) ⊕ 1 = : ⊕ −→ ⊕ . (2.7)
0 1
CM CM

In order to have a managable formalism at hand, let us introduce so-called weight-data:

Definition 2.17. A weight-datum is a tuple 𝔤 = ((L0 , M0 ), (L1 , M1 )) with Lj , Mj ∈ N0 ,

j = 0, 1. Its inverse weight–datum is 𝔤−1 := ((L1 , M1 ), (L0 , M0 )).
A datum 𝔤1 is said to be composable with 𝔤0 provided 𝔤j = ((Lj , Mj ), (Lj+1 , Mj+1 )),
j = 0, 1. In this case we define the product weight-datum 𝔤1 𝔤0 := ((L0 , M0 ), (L2 , M2 )).

d;0
Definition 2.18. BG (Rn−1 × R+ ; 𝔤), 𝔤 = ((L0 , M0 ), (L1 , M1 )), denotes the space
′ ′
∩ S d (Rn−1 × R+ ; H0s,δ (R+ , C L0 ) ⊕ C M0 , H s ,δ (R+ , C L1 ) ⊕ C M1 ).
s,s′ ,δ,δ ′ ∈R

d;0
Elements of BG (Rn−1 × R+ ; 𝔤) are so-called generalized singular Green symbols.

In the previous definition d represents the order of the symbol while the superscript
0 indicates that these symbols have type or class 0. Below we shall define symbols of
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 13

positive type. Note also that we allow the values M0 = 0 or M1 = 0, respectively, where
X ⊕ C 0 := X × {0} := X. In any case, using the topology of spaces of operator-valued
d;0
symbols, we naturally obtain a Fréchet topology on BG (Rn−1 × R+ ; 𝔤).
d;0
Any generalized singular Green symbol g ∈ BG (R n−1
× R+ ; 𝔤) has a representation
in block-matrix form, i.e.,
(︃ )︃
′ ′ g(x′ , ξ ′ , μ) k(x′ , ξ ′ , μ)
g(x , ξ , μ) = ; (2.8)
t(x′ , ξ ′ , μ) q(x′ , ξ ′ , μ)

then g is called a singular Green symbol, k is a Poisson symbol, t is(︃a trace

)︃ symbol and q is
g
a usual pseudodifferential symbol. Note that if M0 = 0 then g = , while g = (g k)
t
in case M1 = 0. The abstract theory of operator-valued symbols yields, in particular,
that

H0s (Rn+ , C L0 ) H s−d (Rn+ , C L1 )

op(g)(μ) : ⊕ −→ ⊕ , s ∈ R,
H s (Rn−1 , C M0 ) H s−d (Rn−1 , C M1 )

continuously for every μ ∈ R+ . Moreover, the class of generalized singular Green opera­
tors is closed under taking the formal adjoint. Extension by zero from R+ to R defines
a meaningful embedding

−1/2 1
e+ : H s (R+ ) −→ H0 (R+ ), s>− . (2.9)
2

In this sense, generalized singular Green symbols g of order d and type 0 can be consid­
ered as operator-valued symbols belonging to

′ ′
∩ S d (Rn−1 × R+ ; H s,δ (R+ , C L0 ) ⊕ C M0 , H s ,δ (R+ , C L1 ) ⊕ C M1 ),
s>−1/2,
s′ ,δ,δ ′ ∈R

hence induce operators

H s (Rn+ , C L0 ) H s−d (Rn+ , C L1 ) 1

op(g)(μ) : ⊕ −→ ⊕ , s>− .
H s (Rn−1 , C M0 ) H s−d (Rn−1 , C M1 ) 2

The operator of differentiation on D ′ (R+ , C L ), denoted by ∂+ , induces (constant)

operator-valued symbols
(︃ )︃
∂+ 0
𝝏 + := ∈ S 1 (Rn−1 × R+ ; H s,δ (R+ , C L ) ⊕ C M , H s−1,δ (R+ , C L ) ⊕ C M )
0 0
(2.10)
14 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

for every s, δ ∈ R (by slight abuse of notation, we shall use the same notation for every
choice of L and M ). Note that 𝝏 + is twisted homogeneous of degree 1 with respect to
{𝜿λ }.

d;r
Definition 2.19. Let r ∈ N0 . Then BG (Rn−1 × R+ ; 𝔤), 𝔤 = ((L0 , M0 ), (L1 , M1 )), denotes
the space of all symbols of the form

∑︂
r
g = g0 + gj 𝝏 j+ , d−j;0
gj ∈ BG (Rn−1 × R+ ; 𝔤).
j=1

d is called the order of g while r is its type (also called class in the literature).

By construction, if g is as in the previous definition, then

′ ′
g∈ ∩ S d (Rn−1 × R+ ; H s,δ (R+ , C L0 ) ⊕ C M0 , H s ,δ (R+ , C L1 ) ⊕ C M1 ),
s>r−1/2,
s′ ,δ,δ ′ ∈R

hence induces maps

H s (Rn+ , C L0 ) H s−d (Rn+ , C L1 ) 1

op(g)(μ) : ⊕ −→ ⊕ , s>r− . (2.11)
H s (Rn−1 , C M0 ) H s−d (Rn−1 , C M1 ) 2

Moreover, g has the homogeneous principal symbol

∑︂
r
g(d) (x′ , ξ ′ , μ) = g0 + (x′ , ξ ′ , μ)𝝏 j+
(d) (d−j)
gj
j=1

∑︁
r
which is defined whenever (ξ ′ , μ) ̸= 0. Using the map (g0 , . . . , gr ) ↦→ g0 + gj 𝝏 j+ and
j=1
d;r
factoring out its kernel, we obtain a natural Fréchet topology on BG (Rn−1 × R+ ; 𝔤).

Remark 2.20. Using symbols without parameter, cf. Definition 2.2, we can define in
the same way the classes of parameter-independent generalized singular Green symbols
BG d;r
(Rn−1 ; 𝔤). In the special case n = 1 there is also no dependence on (x′ , ξ ′ ), in
particular the order d is obsolete. To avoid confusion we use the notation ΓrG for the
class of generalized singular Green operators of type r on the half-axis R+ ; to be precise,
if 𝔤 = ((L0 , M0 ), (L1 , M1 )),

′ ′
Γ0G (R+ ; 𝔤) = ∩ L (H0s,δ (R+ , C L0 ) ⊕ C M0 , H s ,δ (R+ , C L1 ) ⊕ C M1 ))
s,s′ ,δ,δ ′ ∈R

and operators of type r > 0 are then defined analogously to Definition 2.19.
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 15

2.2. Pseudodifferential operators and the transmission property

d
We shall write S1,0 (Rn × R+ ; L (C L0 , C L1 )) for the standard Hörmander class of
pseudodifferential symbols a = a(x, ξ, μ) and S d (Rn ×R+ ; L (C L0 , C L1 )) for the subclass
of poly-homogeneous symbols. As before, we shall denote the homogeneous components
of a poly-homogeneous symbol a of order d by a(d−ℓ) , ℓ ∈ N0 . We werite x = (x′ , xn )
and ξ = (ξ ′ , ξn ).

Definition 2.21. a ∈ S d (Rn ×R+ ; L (C L0 , C L1 )), d ∈ Z, is said to satisfy the transmission

condition (with respect to xn = 0) provided
⃓
(∂xkn ∂ξα′ ∂μj a(d−ℓ) )(x′ , 0, ξ ′ , 1, μ)⃓(ξ′ ,μ)=0
′

⃓
= (−1)d−ℓ−|α |−j (∂xkn ∂ξα′ ∂μj a(d−ℓ) )(x′ , 0, ξ ′ , −1, μ)⃓(ξ′ ,μ)=0
′ ′

for every choice of α′ ∈ N0n−1 and j, k, ℓ ∈ N0 . Write Str

d
(Rn × R+ ; L (C L0 , C L1 )) for the
space of all such symbols.

d
Str (Rn ×R+ ; L (C L0 , C L1 )) is a closed subspace of S d (Rn ×R+ ; L (C L0 , C L1 )). More­
over, symbols of order −∞ and symbols that vanish to infinite order in xn = 0 always
satisfy the transmission condition. The transmission condition is preserved under com­
position, formal adjoint (in case d ≤ 0), and parametrix construction. With e+ from
(2.9), for a symbol a(x, ξ, μ) we then define

op+ (a)(x′ , ξ ′ , μ) = r+ opxn (a)(x′ , ξ ′ , μ) e+ ,

where opxn (a)(x′ , ξ ′ , μ) = a(x′ , xn , ξ ′ , Dxn , μ) means the pseudodifferential operator on

R with respect to the xn -variable with co-variable ξn and r+ : S ′ (R) → D ′ (R+ ) is the
operator of restriction.

d
Theorem 2.22. Let Str (Rn × R+ ; L (C L0 , C L1 )). Then, for all s > − 12 and δ ∈ R,

op+ (a) ∈ S1,0

d
(Rn−1 × R+ ; H s,δ (R+ , C L0 ), H s−d,δ (R+ , C L1 )).

In case a(x, ξ, μ) = a(x′ , ξ, μ) does not depend on the xn -variable,

op+ (a) ∈ S d (Rn−1 × R+ ; H s,δ (R+ , C L0 ), H s−d,δ (R+ , C L1 ))

is poly-homogeneous with homogeneous principal symbol op+ (a(d) )(x′ , ξ ′ , μ).

2.3. Boundary symbols with strong parameter-dependence

Definition 2.23. Let d ∈ Z and r ∈ N0 . Then B d;r (Rn−1 × R+ ; 𝔤) with weight-datum

𝔤 = ((L0 , M0 ), (L1 , M1 )), denotes the space of all symbols of the form
16 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

(︃ )︃
op+ (a)(x′ , ξ ′ , μ) 0
p(x′ , ξ ′ , μ) = + g(x′ , ξ ′ , μ),
0 0

where a ∈ Strd
(Rn × R+ ; L (C L0 , C L1 )) satisfies the transmission condition and g is
d;r
a generalized singular Green symbol belonging to BG (Rn−1 × R+ ; 𝔤). Any such p is
called a boundary symbol. p is called a regularizing boundary symbol if both a and g
have order d = −∞. The class of regularizing boundary symbols of type r is denoted by
B −∞;r (Rn−1 × R+ ; 𝔤).

Note that the representation of p in the previous definition is not unique. It can be
shown that if p has another representation with pseudodifferential part op+ (b)(x′ , ξ ′ , μ)
then there exists a smoothing symbol r ∈ S −∞ (Rn × R+ ; L (C L0 , C L1 )) such that
a(x, ξ; μ) − b(x, ξ; μ) = r(x, ξ; μ) whenever xn ≥ 0.
If p is as in Definition 2.23 then op(p)(μ) has property (2.11). With p we associate
its homogeneous principal symbol

(d)
σψ (p)(x, ξ, μ) = a(d) (x, ξ, μ), xn ≥ 0, (ξ, μ) ̸= 0,

as well as its principal boundary symbol

(︃ )︃
op+ (a0 )(x′ , ξ ′ , μ)
(d)
σ∂ (p)(x′ , ξ ′ , μ) =
(d) 0 + g(d) (x′ , ξ ′ , μ), (ξ ′ , μ) ̸= 0;
0 0

here a0 denotes the symbol with coefficients ``frozen'' at the boundary, i.e.,

a0 (x, ξ, μ) = a(x′ , 0, ξ, μ).

Note that, for every s > r − 1/2,

(d)
σ∂ (p) ∈ Shom
d
(Rn−1 × R+ ; H s (R+ , C L0 ) ⊕ C M0 , H s−d (R+ , C L1 ) ⊕ C M1 ).

Theorem 2.24. Let d ≤ 0. Then p ↦→ p(∗) induces continuous maps

B d;0 (Rn−1 × R+ ; 𝔤) −→ B d;0 (Rn−1 × R+ ; 𝔤−1 ),

d;0
BG d;0
(Rn−1 × R+ ; 𝔤) −→ BG (Rn−1 × R+ ; 𝔤−1 ).

Both homogeneous principal symbol and principal boundary symbol behave naturally un­
der taking the formal adjoint, i.e.

σψ (p(∗) ) = σψ (p)∗ , σ∂ (p(∗) ) = σ∂ (p)∗ ,

(d) (d) (d) (d)

where ∗ on the right-hand sides means the adjoint of (L1 × L0 )-matrices and the adjoint
with respect to the corresponding L2 (R+ )-pairings, respectively
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 17

Theorem 2.25. Let 𝔤1 be composable with 𝔤0 , cf. Definition 2.17. Then (p1 , p0 ) ↦→ p1 #p0
induces continuous maps

B d1 ;r1 (Rn−1 × R+ ; 𝔤1 ) × B d0 ;r0 (Rn−1 × R+ ; 𝔤0 )

−→ B d;r (Rn−1 × R+ ; 𝔤1 𝔤0 ),

where

d = d0 + d1 , r = max(r1 + d0 , r0 ).

Both homogeneous principal symbol and principal boundary symbol are multiplicative,
i.e.,

(d +d1 ) (d ) (d )
σψ 0 (p1 #p0 ) = σψ 1 (p1 )σψ 0 (p0 ),
(d0 +d1 ) (d1 ) (d0 )
σ∂ (p1 #p0 ) = σ∂ (p1 )σ∂ (p0 ).

The class of generalized singular Green operators is a two-sided ideal with respect to
composition. Moreover, (p1 , p0 ) ↦→ p1 #p0 restricts to

d1 ;r1 d0 ;r0
BG (Rn−1 × R+ ; 𝔤1 ) × BG (Rn−1 × R+ ; 𝔤0 )
d0 +d1 ;r0
−→ BG (Rn−1 × R+ ; 𝔤1 𝔤0 ).

If in the previous theorem op+ (aj ) is the pseudodifferential part of pj then p1 #p0
can be written with the pseudodifferential part op+ (a1 #a0 ).

2.3.1. Reduction of orders

ˆ︁ ) = 0 whenever τ ≤ 0. Let c > 2|ψ ′ (t)| for every
Let ψ ∈ S (R) with ψ(0) = 1 and ψ(τ
t and define
(︂ (︂ ξn )︂ )︂d
λd− (ξ, μ) = [ξ ′ , μ]ψ ′
− iξn , d ∈ Z.
c[ξ , μ]

Theorem 2.26. The following affirmations are true:

a) op+ (λd− )(ξ ′ , μ) is an elliptic element of B d;0 (Rn−1 ×R+ ; (1, 0), (1, 0)) for every d ∈ Z.
b) op+ (λd−1 )#op+ (λd−0 ) = op+ (λd−0 +d1 ) for all d0 , d1 ∈ Z.
c) op(op+ (λd− ))(μ) : H s (Rn+ ) → H s−d (Rn+ ) is an isomorphism for every s ∈ R and
every μ ≥ 0.
d) If a ∈ Str
m
(Rn × R+ ) has the transmission property then, for all d ∈ Z,

op+ (λd− )#op+ (a) = op+ (λd− #a).

18 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

For every L, M ∈ N0 let us define

(︃ + d )︃
op (λ− )(ξ ′ , μ) ⊗ 1L 0
𝝀d− (ξ ′ , μ) = , (2.12)
0 [ξ ′ , μ]d ⊗ 1M

i.e., the diagonal matrix with L entries op+ (λd− )(ξ ′ , μ) and M entries [ξ ′ , μ]d (we shall
use the same notation for any choice of L and M ). Then the analogue of Theorem 2.26
holds true for 𝝀d− , d ∈ Z.

2.3.2. Ellipticity and parametrix construction

Let d ∈ Z and d+ := max(d, 0). A symbol p ∈ B d;d+ (Rn−1 × R+ ; (L, M0 ), (L, M1 ))
(note L = L0 = L1 ) is called elliptic provided both homogeneous principal symbol and
principal boundary symbol are pointwise invertible on their domain and

|σψ (p)−1 (x, ξ, μ)| ≲ 1

(d) n
∀ x∈R
xn ≥0 ∀ |ξ, μ| = 1

as well as, for some (and then for all) s > d+ − 1/2,

∥σ∂ (p)−1 (x′ , ξ ′ , μ)∥L (H s−d (R+ ,C L )⊕C M1 ,H s (R+ ,C L )⊕C M0 ) ≲ 1

(d)

uniformly for x′ ∈ Rn−1 and |ξ ′ , μ| = 1. These conditions assure that σψ (p)−1 can be
(d)

extended to a symbol

−d
σψ (p)−1 ∈ Shom
(d)
(Rn × R+ ; L (C L )),

and that

−d
σ∂ (p)−1 ∈ Shom
(d)
(Rn−1 × R+ ; H s−d (R+ , C L ) ⊕ C M0 , H s (R+ , C L ) ⊕ C M0 );

it can be shown that the invertibility of the principal boundary symbol is independent
of the choice of s > d+ − 12 .

Definition 2.27. Let p ∈ B d;d+ (Rn−1 × R+ ; 𝔤), 𝔤 = ((L, M0 ), (L, M1 )). A parametrix for
p is any symbol q ∈ B −d;(−d)+ (Rn−1 × R+ ; 𝔤−1 ) such that

1 − q#p ∈ B −∞;d+ (Rn−1 × R+ ; 𝔤−1 𝔤),

1 − p#q ∈ B −∞;(−d)+ (Rn−1 × R+ ; 𝔤𝔤−1 ).

The main theorem of parameter-elliptic theory is then the following:

Theorem 2.28. Let p ∈ B d;d+ (Rn−1 × R+ ; 𝔤). Then p is elliptic if and only if p has a
parametrix. In this case there exists a μ0 such that op(p)(μ) is invertible for μ ≥ μ0 and
the parametrix q can be chosen in such a way that

op(q)(μ) = op(p)(μ)−1 ∀ μ ≥ μ0 .
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 19

Let us also remark that the order reductions of (2.12) allow to reduce the study of
ellipticity and parametrix to the particular case of symbols of order and type zero:

Remark 2.29. Let p ∈ B d;d+ (Rn−1 × R+ ; 𝔤), 𝔤 = ((L, M0 ), (L, M1 )). If d ≥ 0 then
p#𝝀−d
− ∈ B
0;0
(Rn−1 × R+ ; 𝔤) and p is elliptic if and only if p#𝝀−d − is. If d ≤ 0 then
𝝀−d
− #p ∈ B 0;0
(Rn−1
× R+ ; 𝔤) and p is elliptic if and only if 𝝀 −d
− #p is.

3. Symbols with expansion at infinity -- the abstract setting

3.1. Twisted operator-valued symbols revisited

Let us return to the abstract setting of Section 2.1.1 of Hilbert spaces equipped with
a group-action. By slight modifications we will introduce symbol classes

S˜︁1,0
d,ν
(Rn−1 × R+ ; E0 , E1 ), S˜︁hom
d,ν
(Rn−1 × R+ ; E0 , E1 ), S˜︁d,ν (Rn−1 × R+ ; E0 , E1 )

with d, ν ∈ R. We limit ourselves to list here the necessary modifications to be made and
refer the reader to [30] where these classes appear in a related but different context.

• In Definition 2.2 substitute the estimates (2.2) by

′ ′
∥κ−1 ′ α β j ′ ′ ′
1 (ξ , μ){Dξ ′ Dx′ Dμ p(x , ξ , μ)}κ0 (ξ , μ)∥L (E0 ,E1 )
′
≲ ⟨ξ ′ ⟩ν−|α | ⟨ξ ′ , μ⟩d−ν−j ;

again d is the order of the symbol while ν is called the regularity number.
• The corresponding classes of regularizing symbols are

S˜︁d−∞,ν−∞ (Rn−1 × R+ ; E0 , E1 )
:= ∩ S1,0
d−N,ν−N
(Rn−1 × R+ ; E0 , E1 ).
N >0

Note that these classes only depend on the difference d − ν.

• In order to define S˜︁hom
d,ν
(Rn−1 × R+ ; E0 , E1 ), in Definition 2.5 let V = (Rn−1 \ {0}) ×
R+ and rather than (2.3) consider the estimates

′ ′
∥κ−1 β ′ ′
1,|ξ ′ ,μ| {Dξ ′ Dx′ Dμ p(x , ξ , μ)}κ0,|ξ ,μ| ∥L (E0 ,E1 )
α j
′

′
≲ |ξ ′ |ν−|α | |ξ ′ , μ|d−ν−j .

• For the definition of poly-homogeneous symbols, in Definition 2.6 ask for the exis­
tence of homogeneous components p(d−j,ν−j) ∈ S˜︁hom
d−j,ν−j
(Rn−1 × R+ ; E0 , E1 ) such
that, for every N ∈ N0 ,
20 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

∑︂
N −1
p− χ(ξ ′ )p(d−j,ν−j) ∈ S˜︁1,0
d−N,ν−N
(Rn−1 × R+ ; E0 , E1 )
j=0

(note that χ = χ(ξ ′ ) is a zero-excision function of ξ ′ only). The leading component

p(d,ν) is the homogeneous principal symbol of p. It can be recovered from the full
symbol as in formula (2.4), for ξ ′ ̸= 0.
• The behavior under Leibniz product and formal adjoint is analogous to that of Theo­
rems 2.7 and 2.11, respectively; both order and regularity number are additive under
composition, the homogeneous principal symbol behaves multiplicatively.

As described in the end of Section 2.1.1, all the above spaces can be equipped with
natural Fréchet topologies.

Remark 3.1. S1,0 d

(Rn−1 × R+ ; E0 , E1 ) is continuously embedded in S˜︁1,0
d,0
(Rn−1 × R+ ;
E0 , E1 ); analogously for the subclasses of homogeneous and poly-homogeneous symbols.

Again the case of trivial group actions κj ≡ 1 is included; in this case we shall write

S˜︁1,0
d,ν
(Rn−1 × R+ ; L (E0 , E1 )) := S˜︁1,0
d,ν
(Rn−1 × R+ ; E0 , E1 ),
S˜︁hom
d,ν
(Rn−1 × R+ ; L (E0 , E1 )) := S˜︁hom
d,ν
(Rn−1 × R+ ; E0 , E1 ), (3.1)
S˜︁d,ν (Rn−1 × R+ ; L (E0 , E1 )) := S d,ν (Rn−1 × R+ ; E0 , E1 ).

3.1.1. Symbols with values in Fréchet spaces

We shall also need a variant of the symbol spaces of Definition 2.12 and (3.1) where
the space L (E0 , E1 ) is substituted by a general Fréchet space F (i.e., the symbols take
(ℓ)
values in F ). Let F be equipped with a system of semi-norms {∥| · ∥|F | ℓ ∈ N}
When before we asked symbol estimates with trivial group-action and using the
(ℓ)
operator-norm of L (E0 , E1 ), we now require such estimates for every semi-norm ∥| · ∥|F ,
ℓ ∈ N. This leads to spaces

d
S1,0 (Rn−1 × R+ ; F ), d
Shom (Rn−1 × R+ ; F ), S d (Rn−1 × R+ ; F ),

as well as

S˜︁1,0
d,ν
(Rn−1 × R+ ; F ), S˜︁hom
d,ν
(Rn−1 × R+ ; F ), S˜︁d,ν (Rn−1 × R+ ; F ).

These are all Fréchet spaces again.

d
Remark 3.2. S1,0 (Rn−1 × R+ ; F ) is continuously embedded in S˜︁1,0
d,0
(Rn−1 × R+ ; F ); anal­
ogously for the subclasses of homogeneous and poly-homogeneous symbols.
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 21

We shall also need classes of symbols not depending on the parameter μ, namely

d
S1,0 (Rn−1 ; F ), d
Shom (Rn−1 ; F ), S d (Rn−1 ; F ).

These are constructed similarly as above, by simply cancelling the parameter μ in the
d
definitions of S1,0 (Rn−1 × R+ ; F ), Shom
d
(Rn−1 × R+ ; F ), and S d (Rn−1 × R+ ; F ), respec­
tively.

Remark 3.3. If p(x′ , ξ ′ ) ∈ S1,0

d
(Rn−1 ; F ) is considered as a μ-dependent symbol then
p ∈ S˜︁1,0 (Rn−1 × R+ ; F ). Analogously in the poly-homogeneous case.
d,d

3.2. F -valued symbols with expansion at infinity

In this section we describe two subclasses of S˜︁1,0

d,ν
(Rn−1 × R+ ; F ) which will be crucial
for this paper. In case F = L (C , C ) these classes have been introduced in [29] where
L M

operators on manifolds without boundary were considered. The extension to F -valued

symbols is straight-forward.

Definition 3.4. Let S ˜︁ d,ν (Rn−1 × R+ ; F ) denote the space of all symbols p(x′ , ξ ′ , μ) ∈
1,0
S˜︁1,0
d,ν
(Rn−1 × R+ ; F ) for which exist symbols

p∞ ′ ′ ν+j
[d,ν+j] (x , ξ ) ∈ S1,0 (R
n−1
; F ), j ∈ N0 ,

such that, for every N ∈ N0 ,

∑︂
N −1
p(x′ , ξ ′ , μ) − p∞ ′ ′ ′
[d,ν+j] (x , ξ )[ξ , μ]
d−ν−j
∈ S˜︁1,0
d,ν+N
(Rn−1 × R+ ; F ).
j=0

The symbol p∞
[d,ν] shall be called the principal limit-symbol of p.

The principal limit-symbol can be recovered from the full symbol:

p∞ ′ ′ ′
[d,ν] (x , ξ ) = lim [ξ , μ]
ν−d
p(x′ , ξ ′ , μ) (convergence in S ν+1 (Rn−1 ; F )).
μ→+∞

˜︁ d,ν (Rn−1 × R+ ; F ) is given by the projective topology under the

A Fréchet topology on S 1,0
maps
∑︂
p ↦→ p∞
[d,ν+j] , p ↦→ p − p∞ ′
[d,ν+j] [ξ , μ]
d−ν−j
(j, N ∈ N0 ).
j<N

Remark 3.5. In Definition 3.4 we require an expansion in powers of [ξ ′ , μ]. As shown in

[29, Theorem 7.7] there is a great flexibility in the choice of substitutes for [ξ ′ , μ] without
˜︁ d,ν (Rn−1 × R+ ; F ) itself. For example, we can take the standard
altering the class S
22 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

symbol ⟨ξ ′ , μ⟩ = (1 + |ξ ′ |2 + |μ|2 )1/2 , i.e., p belongs to S˜︁ d,ν (Rn−1 × R+ ; F ) if and only if
there exist symbols p˜︁∞ ν+j
[d,ν+j] ∈ S1,0 (R
n−1
; F ) such that

∑︂
N −1
p(x′ , ξ ′ , μ) − p˜︁∞ ′ ′ ′
[d,ν+j] (x , ξ )⟨ξ , μ⟩
d−ν−j
∈ S˜︁1,0
d,ν+N
(Rn−1 × R+ ; F )
j=0

for every N ∈ N0 . In this case even p˜︁∞ ∞

[d,ν] = p[d,ν] .

3.2.1. A characterization of smoothing symbols

The class of smoothing symbols with expansion at infinity is

˜︁ d−∞,ν−∞ (Rn−1 × R+ ; F ) =
S ∩ S˜︁ d−N,ν−N (Rn−1 × R+ ; F ).
1,0
N >0

˜︁ d−N,ν−N (Rn−1 × R+ ; F ) one has

Since for p ∈ S 1,0

p∞ ∞
[d,ν+j] = p[d−N,ν−N +j] , j ≥ 0,

for a smoothing symbol p all symbols p∞[d,ν+j] are smoothing, too.

In the following proposition, let S1,0 (R+ ; F ) be the space of all p ∈ C ∞ (R+ , F )
d
(ℓ)
with ∥|Dμj p(μ)∥|F ≲ ⟨μ⟩d−j for every j, ℓ. Similarly S d (R+ ; F ) is the subspace of poly­
homogeneous symbols in μ.

Proposition 3.6. For a symbol p(x′ , ξ ′ , μ) it holds

˜︁ d−∞,ν−∞ (Rn−1 × R+ ; F ) ⇐⇒ p ∈ S d−ν (R+ ; S −∞ (Rn−1 ; F )).

p∈S

In this case

p(d−ν) (μ) = p∞
[d,ν] μ
d−ν
.

Proof. It is easy to verify that

S˜︁d−∞,ν−∞ (Rn−1 × R+ ; F ) = S1,0

d−ν
(R+ ; S −∞ (Rn−1 ; F )).

Therefore (cf. Remark 3.5) p ∈ S ˜︁ d−∞,ν−∞ (Rn−1 × R+ ; F ) if and only if there exist
symbols p˜︁j (x′ , ξ ′ ) ∈ S −∞ (Rn−1 ; F ), with p˜︁0 = p∞
[d,ν] , such that, for every N ∈ N0 ,

∑︂
N −1
p(x′ , ξ ′ , μ) − p˜︁j (x′ , ξ ′ )⟨ξ ′ , μ⟩d−ν−j ∈ S1,0
d−ν−N
(R+ ; S −∞ (Rn−1 ; F )). (3.2)
j=0

On the other hand, p ∈ S d−ν (R+ ; S −∞ (Rn−1 ; F )) if and only if there exist symbols
qj (x′ , ξ ′ ) ∈ S −∞ (Rn−1 ; F ) such that, for every N ∈ N0 ,
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 23

∑︂
N −1
p(x′ , ξ ′ , μ) − qj (x′ , ξ ′ )χ(μ)μd−ν−j ∈ S1,0
d−ν−N
(R+ ; S −∞ (Rn−1 ; F )), (3.3)
j=0

where χ(μ) is an arbitrarily fixed zero-excision function. Let us now denote by

a(ρ−j) (μ) = aρ,j μρ−j the homogeneous components of aρ (μ) = ⟨μ⟩ρ ∈ S ρ (R+ ); note
that aρ,0 = 1. Since ⟨ξ ′ , ⟨ξ ′ ⟩μ⟩ρ = ⟨ξ ′ ⟩ρ ⟨μ⟩ρ ,

∑︂
L−1
⟨ξ ′ , μ⟩ρ = ˜︁ ′ ⟩−1 μ)aρ,k ⟨ξ ′ ⟩k μρ−k + ⟨ξ ′ ⟩ρ r˜︁L (⟨ξ ′ ⟩−1 μ),
χ(⟨ξ ρ−L
r˜︁L ∈ S1,0 (R+ )
k=0

for every L ∈ N0 , where the zero-excision function χ ˜︁ is chosen in such a way that
′ −1 ′ −1 ′
χ(⟨ξ ⟩ μ) = χ(⟨ξ
χ(μ)˜︁ ˜︁ ⟩ μ) for all ξ and μ. By direct calculation one checks that

⟨ξ ′ ⟩ρ r˜︁L (⟨ξ ′ ⟩−1 μ) ∈ S1,0

ρ−L
(R+ ; S max(L,ρ) (Rn−1 )),
′ −1 −m
(1 − χ)(⟨ξ
˜︁ ⟩ μ)χ(μ) ∈ S1,0 (R+ ; S m (Rn−1 )) ∀ m.

This yields

∑︂
L−1
⟨ξ ′ , μ⟩ρ ≡ χ(μ)aρ,k ⟨ξ ′ ⟩k μρ−k ρ−L
mod S1,0 (R+ ; S max(L,ρ) (Rn−1 )).
k=0

Taking ρ = d − ν − j and L = N − j, we find

∑︂
N −1
′
⟨ξ , μ⟩d−ν−j
= ad−ν−j,k−j ⟨ξ ′ ⟩k−j χ(μ)μd−ν−k + rjN , 0 ≤ j ≤ N − 1, (3.4)
k=j

with rjN ∈ S1,0

d−ν−N
(R+ ; S max(N,d−ν)−j (Rn−1 )). Equivalently,
⎛ ⎞ ⎛ ⎞⎛ ⎞ ⎛ ⎞
⟨ξ ′ , μ⟩d−ν α00 α01 ··· α0,N −1 χ(μ)μd−ν r0N
⎜ ⟨ξ ′ , μ⟩d−ν−1 ⎟ ⎜ 0 α11 ··· α1,N −1 ⎟ ⎜ χ(μ)μd−ν−1 ⎟ ⎜ r1N ⎟
⎜ ⎟ ⎜ ⎟⎜ ⎟ ⎜ ⎟
⎜ .. ⎟ = ⎝ .. .. .. .. ⎠⎜ .. ⎟ +⎜ .. ⎟
⎝ . ⎠ . . . . ⎝ . ⎠ ⎝ . ⎠
⟨ξ ′ , μ⟩d−ν−(N −1) 0 ··· 0 αN −1,N −1 χ(μ)μd−ν−(N −1) N
rN −1

with αjk (ξ ′ ) = ad−ν−j,k−j ⟨ξ ′ ⟩k−j for j ≤ k, where αjj = 1 for all j. Since the inverse
(βjk ) of the upper-right triangular matrix (αjk ) has the same structure, i.e., βjk (ξ ′ ) =
bjk ⟨ξ ′ ⟩k−j for j ≤ k for certain numbers bjk where bjj = 1 for all j, the latter system,
hence (3.4), is equivalent to

∑︂
N −1
χ(μ)μd−ν−j = bjk ⟨ξ ′ ⟩k−j ⟨ξ ′ , μ⟩d−ν−k + r˜︁jN , 0 ≤ j ≤ N − 1, (3.5)
k=j

with r˜︁jN ∈ S1,0

d−ν−N
(R+ ; S max(N,d−ν)−j (Rn−1 )).
24 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

Clearly (3.4) and (3.5) imply that we can transform an expansion (3.2) into an ex­
pansion (3.3) and vice versa, without changing the coefficient of the leading term. This
yields the claim. □

3.2.2. The subclass of poly-homogeneous symbols

To define the subclass of poly-homogeneous symbols we need some further preparation.
First of all let
n−1
S+ = {(ξ ′ , μ) ∈ Rn−1 × R+ | |ξ ′ |2 + μ2 = 1},
ˆ︁ n−1 = Sn−1 \ {(0, 1)}
S + +

be the unit semi-sphere and the punctured unit semi-sphere, respectively. The latter
shall be identified with (0, 1] × Sn−2 via
√︁
(r, ϕ) ↦→ (rϕ, ˆ︁ n−1 .
1 − r2 ) : (0, 1] × Sn−2 −→ S +

For a Fréchet space F we let CB∞ ((0, ε]; F ), ε > 0, be the space of all smooth functions
u : (0, ε] → F such that (r∂r )j u is bounded on (0, ε] for every j ∈ N0 .

Definition 3.7. rν CB∞ (S ˆ︁ n−1 ; F ) consists of all functions pˆ︁ ∈ C ∞ (S

ˆ︁ n−1 ; F ) such that
+ +
r−ν p(r,
ˆ︁ ϕ) belongs to CB∞ ((0, ε], C ∞ (Sn−2 ; F )) for some (and then for all) ε ∈ (0, 1).
The subspace of all pˆ︁ such that r−ν p(r,ˆ︁ ϕ) extends to a function in C ∞ ([0, ε] × Sn−2 ; F )
is denoted by rν CT∞ (Sˆ︁ n−1
+ ; F ).

These spaces again are Fréchet in a natural way, hence it makes also sense to consider
Cb∞ (Rn−1 ν ∞ ˆ︁ n−1 ′ ∞
x′ , r CB/T (S+ ; F )) including an additional variable x with Cb -dependence
(i.e., all partial derivatives with x′ remain bounded functions in x′ ). The subscript T in
the definition indicates the existence of a Taylor expansion in r centered in r = 0.

ˆ︁ n−1 of functions p(x′ , ξ ′ , μ) defined on Rn−1 ×

Theorem 3.8. The restriction to Rn−1 × S +
(Rn−1 \ {0}) × R+ yields isomorphisms
(︁ )︁
S˜︁hom
d,ν ∼ C ∞ Rn−1 , rν C ∞ (S
(Rn−1 × R+ ; F ) = ˆ︁ n−1 ; F ) ;
b B +

the inverse map is given by homogeneous extension of degree d, i.e.,

(︂ (ξ ′ , μ) )︂
p(x′ , ξ ′ , μ) = |ξ ′ , μ|d pˆ︁ x′ , ′ , ξ ′ ̸= 0.
|ξ , μ|

˜︁ d,ν (Rn−1 × R+ ; F ) consists of all smooth F -valued functions p(x′ , ξ ′ , μ)

Definition 3.9. S hom
which are homogeneous in (ξ ′ , μ) of degree d on Rn−1 × (Rn−1 \ {0}) × R+ and have the
property
⃓ (︁ ν ∞ ˆ︁ n−1
)︁
p⃓Rn−1 ×Sˆ︁ n−1 ∈ Cb∞ Rn−1
x′ , r CT (S+ ; F ) .
+
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 25

˜︁ d,ν (Rn−1 × R+ ; F ) is
The so-called principal angular symbol of p ∈ S hom

(︂ ξ ′ √︁ )︂
p⟨ν⟩ (x′ , ξ ′ ) := |ξ ′ |ν lim r−ν p x′ , r ′ , 1 − r2 ∈ Shom
ν
(Rn−1 ; F ).
r→0+ |ξ |

The principal angular symbol is determined by the leading term in the Taylor expan­
˜︁ d,ν (Rn−1 × R+ ; F ) ⊂ S˜︁d,ν (Rn−1 × R+ ; F ). Moreover
sion. By construction, S hom hom

d
Shom ˜︁ d,0 (Rn−1 × R+ ; F ),
(Rn−1 × R+ ; F ) ⊂ S hom

since C ∞ (S+
n−1 ˆ︁ n−1 ; F ).
; F ) ⊂ CT∞ (S +

Proposition 3.10. Let p(x′ , ξ ′ , μ) ∈ S ˜︁ d,ν (Rn−1 × R+ ; F ) and χ(ξ ′ ) be a zero-excision

hom
function. Then χ(ξ ′ )p(x′ , ξ ′ , μ) belongs to S ˜︁ d,ν (Rn−1 × R+ ; F ) and has principal limit­
1,0
symbol

(χp)∞ ′ ′ ′ ′ ′
[d,ν] (x , ξ ) = χ(ξ )p⟨ν⟩ (x , ξ ).

Due to the latter proposition the following definition makes sense:

Definition 3.11. A symbol p belongs to the space S ˜︁ d,ν (Rn−1 × R+ ; F ) provided there
˜︁
exist homogeneous components p(d−j,ν−j) ∈ S d−j,ν−j
(Rn−1 ×R+ ; F ) such that, for every
hom
N ∈ N0 ,

∑︂
N −1
p(x′ , ξ ′ , μ) − ˜︁ d−N,ν−N (Rn−1 × R+ ; F ).
χ(ξ ′ )p(d−j,ν−j) (x′ , ξ ′ , μ) ∈ S 1,0
j=0

By definition, the principal angular symbol of p is that of its homogeneous principal

symbol, i.e., p⟨d,ν⟩ (x′ , ξ ′ ) := p⟨ν⟩ (x′ , ξ ′ ).
(d,ν)

˜︁ d,ν (Rn−1 × R+ ; F ) carries a natural projective topology which makes it a

Again, S
Fréchet space. By construction, the principal angular symbol of p ∈ S ˜︁ d,ν (Rn−1 × R+ ; F )
satisfies
(︂ ξ ′ √︁ )︂
p⟨d,ν⟩ (x′ , ξ ′ ) := |ξ ′ |ν lim r−ν p(d.ν) x′ , r ′ , 1 − r2 ∈ Shom
ν
(Rn−1 ; F ).
r→0+ |ξ |

The principal angular symbol is a ``subordinate'' or ``secondary'' principal symbol, since

it is uniquely determined by the homogeneous principal symbol. However, it plays an
important role in the calculus, cf. Section 3.3.

˜︁ d,ν (Rn−1 × R+ ; F ) as in Definition 3.11. Due to Proposition 3.10,

Remark 3.12. Let p ∈ S

∑︂
N −1
p∞ ′ ′
χ(ξ ′ )p⟨ν−j⟩ (x′ , ξ ′ ) ∈ S ν−N (Rn−1 ; F ).
(d−j,ν−j)
[d,ν] (x , ξ ) −
j=0
26 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

Thus the principal limit symbol p∞

[d,ν] is poly-homogeneous and its homogeneous principal
symbol coincides with the principal angular symbol of p,

p⟨d,ν⟩ (x′ , ξ ′ ) = (p∞

[d,ν] )
(ν) ′ ′
(x , ξ ).

The following result was already used in [29] (in case F = C) but without providing
an explicit proof. For convenience of the reader we give a proof here.

˜︁ d,0 (Rn−1 × R+ ; F ) and

Proposition 3.13. Let p(x′ , ξ ′ , μ) ∈ S d (Rn−1 × R+ ; F ). Then p ∈ S
p has limit symbol

p∞ ′ ′
[d,0] (x , ξ ) = p
(d) ′
(x , 0, 1),

i.e., the homogeneous principal symbol evaluated in (ξ ′ , μ) = (0, 1).

Proof. By [29, Proposition 5.3], p ∈ S˜︁ d,0 (Rn−1 × R+ ; F ) and the principal limit symbol
1,0
is as stated. We have to show that p belongs to the poly-homogeneous class.
Given a zero-excision function χ(ξ ′ ) let us choose a zero-excision function χ(ξ
˜︁ ′ , μ) such
that χ˜︁
χ = χ. Then

∑︂
N −1 (︂ ∑︂
N −1 )︂
p− χp(d−j) = χ p − ˜︁ (d−j) + (1 − χ)p.
χp
j=0 j=0

Since (1−χ) ∈ S˜︁ −N,−N (Rn−1 ×R+ ) and p ∈ S ˜︁ d,0 (Rn−1 ×R+ ; F ), the second term on the
1,0 1,0
˜︁ d−N,−N (Rn−1 × R+ ; F ). The first term is of the form χpN ,
right-hand side belongs to S 1,0
where pN ∈ S d−N (Rn−1 × R+ ; F ) ⊂ S ˜︁ d−N,0 (Rn−1 × R+ ; F ) and χ ∈ S ˜︁ 0,0 (Rn−1 × R+ ).
1,0 1,0
Thus, to complete the proof, it remains to show that

˜︁ d−N,0 (Rn−1 × R+ ; F ) ⊂ S
S ˜︁ d−N,−N (Rn−1 × R+ ; F ).
1,0 1,0

For the latter rename d − N by d and let a ∈ S ˜︁ d,0 (Rn−1 × R+ ; F ). We will show that
1,0
a belongs to S ˜︁ d,−L
(Rn−1
× R+ ; F ) for arbitrary L ∈ N0 . This is equivalent to showing
1,0
that ˜︁ ′ L
a := ⟨ξ ⟩ a ∈ S1,0 (R˜︁ d+L,0 n−1
× R+ ; F ). We define ˜︁ a[0] = . . . = ˜︁
a[L−1] = 0 and
a[L+j] = ⟨ξ ′ ⟩L a[j] for j ≥ 0. Then, for N ≤ L,
˜︁

∑︂
N −1
˜︁
a− a[j] [ξ ′ , μ]d+L−j = ⟨ξ ′ ⟩L a ∈ S˜︁1,0
˜︁ d+L,L
(Rn−1 × R+ ; F ) ⊂ S˜︁1,0
d+L,N
(Rn−1 × R+ ; F ),
j=0

since S˜︁1,0
d,ν1
(Rn−1 × R+ ; F ) ⊂ S˜︁1,0
d,ν0
(Rn−1 × R+ ; F ) whenever ν0 ≤ ν1 . If N > L,

∑︂
N −1 (︂ N −L−1
∑︂ )︂
˜︁
a− a[j] [ξ ′ , μ]d+L−j = ⟨ξ ′ ⟩L a −
˜︁ a[j] [ξ ′ , μ]d−j ∈ S˜︁1,0
d+L,N
(Rn−1 × R+ ; F ),
j=0 j=0
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 27

since ⟨ξ ′ ⟩L S˜︁1,0
d,N −L
(Rn−1 × R+ ; F ) = S˜︁1,0
d+L,N
(Rn−1 × R+ ; F ). □

Remark 3.14. If p(x′ , ξ ′ ) ∈ S1,0

d
(Rn−1 ; F ) is considered as a μ-dependent symbol then
p∈S˜︁ (R
d,d n−1
× R+ ; F ). Analogously in the poly-homogeneous case.
1,0

3.3. A result on the invertibility of homogeneous components

For later purpose we shall discuss in this section a specific result concerning the
invertibility of homogeneous symbols. It explains the importance of the principal angular
symbol.

˜︁ d,ν (Rn−1 × R+ ; L (E0 , E1 )). Assume

Lemma 3.15. Let p ∈ S hom

(1) p(x′ , ξ ′ , μ) is invertible whenever ξ ′ ̸= 0 and

∥p(x′ , ξ ′ , μ)−1 |ξ ′ |ν ∥L (E1 ,E0 ) ≲ 1 ∀ x′ ∈ Rn−1 ˆ︁ n−1 ,

∀ (ξ ′ , μ) ∈ S +

(2) The principal angular symbol p⟨ν⟩ (x′ , ξ ′ ) is invertible whenever ξ ′ ̸= 0 and

∥p⟨ν⟩ (x′ , ξ ′ )−1 ∥L (E1 ,E0 ) ≲ 1 ∀ x′ ∈ Rn−1 ∀ |ξ ′ | = 1.

˜︁ −d,−ν (Rn−1 × R+ ; L (E1 , E0 )).

Then p−1 ∈ S hom

Proof. By multiplication with |ξ ′ |−ν |ξ ′ , μ|ν we may assume without loss of generality
that ν = 0. Let pˆ︁ denote the restriction of p to Rn−1 ×Sˆ︁ n−1 . If ω ∈ C ∞ ([0, 1)) has
x′ + 0
compact support and ω ≡ 1 near r = 0, it is clear from (1) that (1 − ω)(|ξ|)ˆ︁ p−1 is a
(︁ )︁
ˆ︁
function in Cb∞ Rn−1 , CT∞ (S + ; L (E1 , E0 )) . Using the notation of Definition 3.7 and
n−1

′
ˆ︁ , r, ϕ) extends to a smooth function up to and including r = 0 and
Definition 3.9, p(x

ˆ︁ ′ , r, ϕ) = p⟨ν⟩ (x′ , ϕ).

lim p(x
r→0

Thus (1) and (2) imply the pointwise invertibility of pˆ︁ together with a uniform estimate of
pˆ︁−1 (x′ , r, ϕ) on Rn−1 × [0, ε] × Sn−2 for every ε ∈ (0, 1). Therefore ω(|ξ|)ˆ︁
p(x′ , ξ ′ , μ)−1 lies
(︁ n−1 ∞ n−1 )︁ (︁ n−1 ∞ n−1 )︁
in Cb R∞ ˆ︁
, CT (S −1 ∞
ˆ︁ ∈ Cb Rx′ , CT (S ˆ︁
+ ; L (E1 , E0 )) . Thus p + ; L (E1 , E0 ))
and homogeneous extension of degree −d yields the claim. □

Remark 3.16. The reasoning in the proof of Lemma 3.15 shows also the following: If
p(x′ , ξ ′ , μ) is constant in x′ for |x′ | ≥ C for some C ≥ 0, conditions (1) and (2) are
equivalent to

ˆ︁ n−1 ,
(1′ ) p(x′ , ξ ′ , μ) is invertible for all x′ ∈ Rn−1 and (ξ ′ , μ) ∈ S +
(2 ) The principal angular symbol p⟨ν⟩ (x , ξ ) is invertible whenever |ξ ′ | = 1.
′ ′ ′
28 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

We shall need a version of the previous result, concerning the related space

Sd,ν n−1
× R+ ; F ) := Shom
d ˜︁ d,ν (Rn−1 × R+ ; F )
(Rn−1 × R+ ; F ) + S
hom (R hom (3.6)

where ν ∈ N0 . Note that

Sd,0 n−1 ˜︁ d,0 (Rn−1 × R+ ; F )

× R+ ; F ) = S
hom (R hom

and

Sd,ν n−1 ˜︁ d,0 (Rn−1 × R+ ; F ),

× R+ ; F ) ⊂ S ν ≥ 1.
hom (R hom

In any case we can associate with p ∈ Sd,ν hom (R

n−1
× R+ ; F ) its principal angular symbol
p⟨0⟩ ∈ Shom (R
0 n−1
; F ). Indeed, if p = p0 + p˜︁ with p0 ∈ Shom
d
(Rn−1 × R+ ; F ) and p˜︁ ∈
˜︁
S d,ν
(Rn−1
× R+ ; F ), then
hom

{︄
′ ′ p0 (x′ , 0, 1) + p˜︁⟨0⟩ (x′ , ξ ′ ) :ν=0
p⟨0⟩ (x , ξ ) = .
′
p0 (x , 0, 1) :ν≥1

Lemma 3.17. Let p ∈ Sd,ν

hom (R
n−1
× R+ ; L (E0 , E1 )), ν ∈ N0 . Assume

i) p(x′ , ξ ′ , μ) is invertible whenever ξ ′ ̸= 0 and

∥p(x′ , ξ ′ , μ)−1 ∥L (E1 ,E0 ) ≲ 1 ∀ x′ ∈ Rn−1 ˆ︁ n−1 ,

∀ (ξ ′ , μ) ∈ S +

ii) The principal angular symbol p⟨0⟩ (x′ , ξ ′ ) is invertible whenever ξ ′ ̸= 0 and

∥p⟨0⟩ (x′ , ξ ′ )−1 ∥L (E1 ,E0 ) ≲ 1 ∀ x′ ∈ Rn−1 ∀ |ξ ′ | = 1.

Then p−1 ∈ S−d,ν

hom (R
n−1
× R+ ; L (E1 , E0 )) (note the identical regularity number).

Proof. In case ν = 0 this is a special case of Lemma 3.15. We thus assume ν ≥ 1. By

multiplication with |ξ ′ , μ|−d we may assume without loss of generality that d = 0. On
the one hand it follows from [30, Proposition 3.8] that p−1 belongs to Shom 0
(Rn−1 ×
R+ ; L (E1 , E0 )) + S˜︁hom (R
0,ν n−1 −1
× R+ ; L (E1 , E0 )), hence can be written as p = p0 + p. ˜︁
−1
On the other hand, by Lemma 3.15, p belongs to Shom (R ˜︁ 0,0 n−1
× R+ ; L (E1 , E0 )), say
p−1 = q.
˜︁ Then

˜︁ 0,0 (Rn−1 × R+ ; L (E1 , E0 )) ∩ S˜︁0,ν (Rn−1 × R+ ; L (E1 , E0 )).

p˜︁ = q˜︁ − p0 ∈ S hom hom

˜︁ 0,ν (Rn−1 × R+ ; L (E1 , E0 )).

The latter space coincides with S □
hom

The analogue of Remark 3.16 holds for Lemma 3.17.

 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 29

4. Singular Green symbols with expansion at infinity

4.1. Singular Green symbols with finite regularity number

d;0
In Section 2.1.3 we have introduced the classes BG (Rn−1 × R+ ; 𝔤) of strongly
parameter-dependent generalized singular Green symbols of type zero as a realization
of the abstract classes S d (Rn−1 × R+ ; E0 , E1 ), cf. Definition 2.18. From this resulted
symbols of positive type r ∈ N in Definition 2.19.
We proceed in the analogous way for Green symbols with finite regularity number,
based on the symbol classes S˜︁d,ν (Rn−1 × R+ ; E0 , E1 ) from Section 3.1. Again we shall
use the dilation group-action as explained in the beginning of Section 2.1.2 and (2.7).

˜︁ d,ν;0 (Rn−1 × R+ ; 𝔤), 𝔤 = ((L0 , M0 ), (L1 , M1 )), denotes the space

Definition 4.1. B G

S˜︁d,ν (Rn−1 × R+ ; H0s,δ (R+ , C L0 ) ⊕ C M0 , H s ,δ (R+ , C L1 ) ⊕ C M1 ).

′ ′
∩
s,s′ ,δ,δ ′ ∈R

˜︁ d,ν;r (Rn−1 × R+ ; 𝔤) denotes the space of all symbols

Moreover, B G

∑︂
r
g = g0 + gj 𝝏 j+ , ˜︁ d−j,ν;0 ((Rn−1 × R+ ; 𝔤).
gj ∈ B G
j=1

˜︁ d,ν;r (Rn−1 × R+ ; 𝔤) has its homogeneous principal symbol

g∈B G

∑︂
r
g(d,ν) (x′ , ξ ′ , μ) = g0 (x′ , ξ ′ , μ)𝝏 j+
(d,ν) (d−j,ν)
+ gj (4.1)
j=1

which is defined whenever ξ ′ ̸= 0. Moreover, op(g)(μ) induces maps as in (2.11).

Le us remark that in Definition 4.1 we employ a notation already used in [30] in a
different context; actually the respective spaces are different (that of [30] is larger).

4.2. From operator-valued symbols to symbol-kernels

Generalized singular Green symbols g of type zero from each of the above introduced
classes are functions of (x′ , ξ ′ , μ) taking values in Γ0G (R+ ; 𝔤), 𝔤 = ((L0 , M0 ), (L1 , M1 )),
the space of type zero generalized singular Green operators on R+, cf. Remark 2.20. We
can characterize these operators on R+ as follows:

′ ′
a) Poisson operators: ∩s′ ,δ′ L (C M0 , H s ,δ (R+ , C L1 )) is the space of all operators of the
form

c ↦→ φc, φ ∈ S (R+ , C L1 ×M0 ),

30 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

b) Trace operators of type zero: ∩s,δ L (H0s,δ (R+ , C L0 ), C M1 ) is the space of all operators
of the form

ˆ
+∞

u ↦→ φ(yn )u(yn ) dyn , φ ∈ S (R+ , C M1 ×L0 ),

0

′ ′
c) Green operators of type zero: ∩s,δ,s′ ,δ′ L (H0s,δ (R+ , C L0 ), H s ,δ (R+ , C L1 )) is the
space of all operators of the form

ˆ
+∞

u ↦→ ψ(·, yn )u(yn ) dyn , ψ ∈ S (R+ × R+ , C L1 ×L0 ),

0

where C k×ℓ := L (C ℓ , C k ) is (isomorphic to) the space of complex (k ×ℓ)-matrices. Thus

if we represent a generalized singular Green symbol g of type zero in block-matrix form
(2.8) and use S (Rn+ ) ∼
= S (Rn−1 , S (R+ )) then the Poisson operator can be written as
ˆ
eix ξ ˜︁
′ ′
[op(k)(μ)u](x) = k(x′ , ξ ′ , μ; xn )ˆ︁
u(ξ ′ ) d¯ξ ′ , u ∈ S (Rn−1 , C M0 ),

the trace operator can be written as

ˆ ˆ
+∞
′ ′
′
[op(t)(μ)u](x ) = eix ξ ˜︁
t(x′ , ξ ′ , μ; yn )ˆ︁
u(ξ ′ , yn ) dyn d¯ξ ′ , u ∈ S (Rn+ , C L0 ),
0

the Green operator can be written as

ˆ ˆ
+∞
′ ′
[op(g)(μ)u](x) = ˜︁ ′ , ξ ′ , μ; xn , yn )ˆ︁
eix ξ g(x u(ξ ′ , yn ) dyn d¯ξ ′ , u ∈ S (Rn+ , C L0 ),
0

where x = (x′ , xn ) ∈ Rn+ and u ˆ︁ indicates the (partial) Fourier transform in the x′ ­
variable. The involved so-called symbol-kernels are functions of (x′ , ξ ′ , μ) and take values
in the spaces of rapidly decreasing functions as indicated above in a), b), and c). In
order to simplify notation we shall identify symbols and symbol-kernels, i.e., also denote
symbol kernels by k, t, and g rather than ˜︁k, ˜︁ ˜︁ respectively.
t, and g,
For the assertions of the following Theorems 4.2 and 4.4 in the case of strongly
parameter-dependent symbols see Remark 2.1.18 and Theorem 2.1.19 of [17] as well
as Theorems 3.7 and 3.9 of [16].

Theorem 4.2. Let 𝔤 = ((L0 , M0 ), (L1 , M1 )) and g(x′ , ξ ′ , μ) be smooth with values in
L (L2 (R+ , C L0 ) ⊕ C M0 , L2 (R+ , C L1 ) ⊕ C M1 )). The following are equivalent:
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 31

d;0
a) g ∈ BG (Rn−1 × R+ ; 𝔤).
b) If g∗ = g(x′ , ξ ′ , μ)∗ is the pointwise adjoint of g, then
′ ′
g∈ ∩ S d (Rn−1 × R+ ; L2 (R+ , C L0 ) ⊕ C M0 , H s ,δ (R+ , C L1 ) ⊕ C M1 ),
s′ ,δ ′ ∈R
′ ′
g∗ ∈ ∩ S d (Rn−1 × R+ ; L2 (R+ , C L1 ) ⊕ C M1 , H s ,δ (R+ , C L0 ) ⊕ C M0 ).
s′ ,δ ′ ∈R

c) g′ := 𝜿−1 g𝜿 ∈ S d (Rn−1 × R+ ; Γ0G (R+ ; 𝔤)).

For c) recall the notation from (2.7). The analogous result is true for symbols with finite
d;0
regularity number, i.e., replacing BG ˜︁ d,ν;0 and S d by S˜︁d,ν , respectively.
by B G

Part c) of(︃the previous

)︃ theorem can be reformulated on the level of symbol-kernels.
′ g′ k′
Write g := and let g ′ , k′ , and t′ denote (also) the associated symbol-kernels.
t′ q ′
If g, k, and t are the symbol-kernels associated with g, then by direct calculation one
finds that g′ = 𝜿−1 g𝜿 is equivalent to q = q ′ and

g(x′ , ξ ′ , μ; xn , yn ) = [ξ ′ , μ]g ′ (x′ , ξ ′ , μ; [ξ ′ , μ]xn , [ξ ′ , μ]yn ),

k(x′ , ξ ′ , μ; xn ) = [ξ ′ , μ]1/2 k′ (x′ , ξ ′ , μ; [ξ ′ , μ]xn ), (4.2)
′ ′ ′ 1/2 ′ ′ ′ ′
t(x , ξ , μ; yn ) = [ξ , μ] t (x , ξ , μ; [ξ , μ]yn ).

Theorem 4.3. For g(x′ , ξ ′ , μ; xn , yn ) the following are equivalent:

a) g has the form (4.2) with g ′ ∈ S1,0

d
(Rn−1 × R+ ; S (R+ × R+ , C L1 ×L0 )).
b) For every choice of (multi-)indices, g satisfies the estimates
′ ′ ′ ′
∥xℓn Dxℓ n ynm Dymn Dμj Dξα′ Dxβ′ g(x′ , ξ ′ , μ; xn , yn )∥L2 (R+,xn ×R+,yn ,C L1 ×L0 )
′ ′ ′
≲ [ξ ′ , μ]d−|α |−j+ℓ −ℓ+m −m .

An analogous result is true for symbol-kernels with finite regularity number ν. Similar
results hold for Poisson and trace symbol-kernels.

For the characterization of generalized singular Green symbols of positive type r > 0
recall the notations from (2.7) and (2.10) and note that
(︃ )︃
[ξ ′ , μ]j 0
𝜿−1 (ξ ′ , μ) 𝝏 j+ 𝜿(ξ ′ , μ) = 𝝏 j+ . (4.3)
0 1

This yields immediately the following result:

Theorem 4.4. The following are equivalent:

32 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

d;r
a) g ∈ BG (Rn−1 × R+ ; 𝔤).
b) g′ := 𝜿−1 g𝜿 ∈ S d (Rn−1 × R+ ; ΓrG (R+ ; 𝔤)).

d;r
The analogous result is true for symbols with finite regularity number, i.e., replacing BG
˜︁
by B d,ν;r
and S d by S˜︁d,ν , respectively.
G

4.3. Singular Green symbols with expansion at infinity

The following definition is a natural consequence of Theorem 4.4.

˜︁ d,ν;r (Rn−1 × R+ ; 𝔤) consists of all symbols g such that

Definition 4.5. B G

˜︁ d,ν (Rn−1 × R+ ; Γr (R+ ; 𝔤)).

g′ := 𝜿−1 g𝜿 ∈ S G

If g is as in the previous definition then, due to (4.3), it has the form

∑︂
r
g = g0 + gj 𝝏 j+ , ˜︁ d−j,ν;0 (Rn−1 × R+ ; 𝔤).
gj ∈ B G
j=1

Remark 4.6. In view of the above Theorem 4.2,

˜︁ d,ν;r (Rn−1 × R+ ; 𝔤) ⊂ B
B ˜︁ d,ν;r (Rn−1 × R+ ; 𝔤),
G G
d;r
BG ˜︁ d,0;r (Rn−1 × R+ ; 𝔤).
(Rn−1 × R+ ; 𝔤) ⊂ B G

Due to the first inclusion in the previous remark, with each generalized singular Green
symbol g ∈ B ˜︁ d,ν;r (Rn−1 × R+ ; 𝔤) we can associate its homogeneous principal symbol
G
g(d,ν) (x′ , ξ ′ , μ), ξ ′ ̸= 0, as in (4.1). If g′ = 𝜿−1 g𝜿 as in Definition 4.5, then

g(d,ν) (x′ , ξ ′ , μ) = 𝜿|ξ′ ,μ| g′ (x′ , ξ ′ , μ)𝜿−1

(d,ν)
|ξ ′ ,μ| .

For the following definition recall Definition 3.4.

˜︁ d,ν;r (Rn−1 × R+ ; 𝔤) is
Definition 4.7. The principal limit-symbol of g ∈ B G

∞
g[d,ν] := (𝜿−1 g𝜿)∞
[d,ν] ∈ S (R
ν n−1
; ΓrG (R+ ; 𝔤)).

Again, as a subordinate principal symbol, we define the principal angular symbol

g⟨d,ν⟩ := (𝜿−1 g𝜿)⟨d,ν⟩ ∈ Shom

ν
(Rn−1 ; ΓrG (R+ ; 𝔤)),

d;r
cf. Definition 3.11. If g ∈ BG (Rn−1 × R+ ; 𝔤) then

∞
g[d,0] (x′ , ξ ′ ) = (𝜿−1 g𝜿)(d) (x′ , 0, 1) = g(d) (x′ , 0, 1) = g⟨0⟩ (x′ , ξ ′ ).
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 33

4.4. (Un-)twisting on operator-level

)︃ (︃
κ 0
In Sections 4.2 and 4.3 we have seen that conjugation by 𝜿 = establishes,
0 1
on the level of symbols, a one-to-one correspondence between generalized singular Green
symbols and symbols with values in generalized singular Green operators on the half­
axis. In this Section we shall show that this correspondence persists on operator-level,
i.e., conjugation with the group-action operator establishes a one-to-one correspondence
between the various classes.

Theorem 4.8. Let 𝔅d;r d;r ˜︁ d,ν;r ˜︁ d,ν;r and let 𝔖d represent
G represent one choice of BG , BG , or B G
˜︁ d,ν . Let 𝔤 = ((L0 , M0 ), (L1 , M1 )).
the corresponding choice S d , S˜︁d,ν , or S

a) Let g ∈ 𝔅d;r
G (R
n−1
× R+ ; 𝔤). Then there exists a unique symbol g′ ∈ 𝔖d (Rn−1 ×
R+ ; ΓG (R+ ; 𝔤)) such that
r

op(g′ )(μ) = op(𝜿−1 )(μ)op(g)(μ)op(𝜿)(μ).

We shall write 𝜿−1 #g#𝜿 := g′ . One has

𝜿−1 #g#𝜿 − 𝜿−1 g 𝜿 ∈ 𝔖d−1 (Rn−1 × R+ ; ΓrG (R+ ; 𝔤)) (4.4)

b) Let g′ ∈ 𝔖d (Rn−1 × R+ ; ΓrG (R+ ; 𝔤)). Then there exists a unique symbol g ∈
𝔅d;r
G (R
n−1
× R+ ; 𝔤) such that

op(g)(μ) = op(𝜿)(μ)op(g′ )(μ)op(𝜿−1 )(μ).

We shall write 𝜿#g′ #𝜿−1 := g. One has

𝜿#g′ #𝜿−1 − 𝜿g′ 𝜿−1 ∈ 𝔅d−1;r

G (Rn−1 × R+ ; 𝔤). (4.5)

The proof of this theorem will occupy the remaining part of this section. Let us remark
that in the proof we shall obtain a complete asymptotic expansions of 𝜿−1 #g#𝜿 and
𝜿#g′ #𝜿−1 which include (4.4) and (4.5) as special cases (see Proposition 4.9 and the
subsequent paragraph). The main case to verify is that of type r = 0 which we will
assume from now on. We will first prove a) and (4.4). The proof makes essential use of
the concept of oscillatory-integrals and related techniques in the spirit of [12], here in a
variant for amplitude functions with values in(︃Fréchet
)︃ spaces (see Theorem 4.10).
g k
We start out from the representation g = = 𝜿g0′ 𝜿−1 with a symbol
t q

(︃ )︃
g0′ k0′
g0′ = ∈ 𝔖d (Rn−1 × R+ ; Γ0G (R+ ; 𝔤)), 𝔤 = ((L0 , M0 ), (L1 , M1 ));
t′0 q0′
34 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

this is possible by Theorem 4.2 and Definition 4.5, respectively. Since 𝜿 is independent
of x′ , we find that

op(𝜿−1 )(μ)op(g)(μ)op(𝜿)(μ) = op(𝜿−1 )(μ)op(𝜿g0′ )

(︃ )︃ (4.6)
op(κ−1 )(μ)op(κg0′ )(μ) op(κ−1 )(μ)op(κk0′ )(μ)
= .
op(t′0 )(μ) op(q0′ )(μ)

We thus have to show the existence of

(︃ ′ )︃
g k′
g′ = ′ ′ ∈ 𝔖d (Rn−1 × R+ ; Γ0G (R+ ; 𝔤))
t q

such that op(g′ )(μ) coincides with the right-hand side of (4.6) and such that (4.4) is
valid. To this end we obviously must define t′ := t′0 and q ′ := q0′ and have to find g ′ and
k′ such that

op(g ′ )(μ) = op(κ−1 )(μ)op(κg0′ )(μ), op(k′ )(μ) = op(κ−1 )(μ)op(κk0′ )(μ)

as well as

g ′ − g0′ ∈ 𝔖d−1 (Rn−1 × R+ ; Γ0G ((L0 , 0), (L1 , 0)))

and

k′ − k0′ ∈ 𝔖d−1 (Rn−1 × R+ ; Γ0G ((0, M0 ), (L1 , 0))). (4.7)

We shall now verify this for the Poisson part, i.e., show the existence of k′ together with
(4.7). The proof for the Green part g ′ works in the same way and only comes along with
lengthier notation, since the symbol-kernel of Green symbol depends on (xn , yn ) while
that of a Poisson symbol only on xn (note that the group-action and the group-action
operator act from the left which only has an effect on the xn -variable).
If u ∈ S (Rn−1 , C M0 ) then, by direct computation,

[op(κ−1 )(μ)op(κk0′ )(μ)u](x′ )

˘
′ ′ ′ ′ ′′ ′′
= ei(x −y )η +i(y −y )η p(η ′ , y ′ , η ′′ , μ)u(y ′′ ) dy ′′ d¯η ′′ dy ′ d¯η ′

(understood as iterated integrals) where

p(η ′ , y ′ , η ′′ , μ) = κ(η ′ , μ)−1 κ(η ′′ , μ)k0′ (y ′ , η ′′ , μ) = κω(η′ ,η′′ ,μ) k0′ (y ′ , η ′′ , μ)

where

ω(η ′ , η ′′ , μ) := [η ′ , μ]−1 [η ′′ , μ]. (4.8)

 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 35

Now let us pass from the setting of operator-valued symbols to the formulation with
symbol-kernels. Recall that the symbol-kernel of k0′ satisfies

k0′ (x′ , ξ ′ , μ; xn ) ∈ 𝔖d (Rn−1 × R+ ; S (R+ , C L1 ×M0 )).

Then p has the symbol-kernel

p(η ′ , y ′ , η ′′ , μ; xn ) = ω(η ′ , η ′′ , μ)1/2 k0′ (x′ , ξ ′ , μ; ω(η ′ , η ′′ , μ)xn )

(4.9)
= (κω(η′ ,η′′ ,μ) k0′ )(y ′ , η ′′ , μ; xn )

(i.e., the group-action acts on the xn -variable) and

[op(κ−1 )(μ)op(κk0′ )(μ)u](x)

˘
′ ′ ′ ′ ′′ ′′
= ei(x −y )η +i(y −y )η p(η ′ , y ′ , η ′′ , μ; xn )u(y ′′ ) dy ′′ d¯η ′′ dy ′ d¯η ′ .

From (2.6) it easily follows that, for an arbitrary choice of multi-indices,

′ ′′ ′ ′ ′′
|Dηα′ Dηα′′ Dyβ′ p(η ′ , y ′ , η ′′ , μ; xn )| ≲ C(μ)⟨η ′ , μ⟩− 2 −|α | ⟨η ′′ , μ⟩d+ 2 −|α | ,
1 1

where C(μ) = 1 if 𝔖d = S d and C(μ) = ⟨μ⟩(−ν)+ otherwise. Therefore, for fixed μ

− 1 ,d+ 1
and xn , p(η ′ , y ′ , η ′′ , μ; xn ) is a double symbol from the class S1,02 2
in the sense of
[12, Definition 2.1, Chapter 2] (where the variables (η , y , η ) correspond to (ξ, x′ , ξ ′ ) in
′ ′ ′′

[12]). We thus can apply the results of [12], in particular Theorem 2.5 and Theorem 3.1
of Chapter 2, to obtain the following:

Proposition 4.9. With the above notation,

[op(κ−1 )(μ)op(κk0′ )(μ)u](x) = [op(pL )(μ; xn )u](x′ ),

where
¨
′ ′
′ ′
pL (x , ξ , μ; xn ) = Os − e−iy η p(ξ ′ + η ′ , x′ + y ′ , ξ ′ , μ; xn ) dy ′ d¯η ′

d
is, for each μ and xn , a symbol in S1,0 (Rn−1 ; C L1 ×M0 ). Moreover,

∑︂
N −1 ⃓
′ ′ 1 α′ α′ ′ ′ ′ ⃓
pL (x , ξ , μ; xn ) = ∂ ′ Dy ′ p(η , y , ξ , μ; xn )⃓ ′
α′ ! η y =x′ +
|α′ |=0 η ′ =ξ ′
(4.10)
∑︂ ˆ (1 − θ)N −1
1

+N ′!
rγ ′ ,θ (x′ , ξ ′ , μ; xn ) dθ
′
γ
|γ |=N 0
36 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

where
¨
′ ′ ′ ′
rγ ′ ,θ (x′ , ξ ′ , μ; xn ) = Os − e−iy η (∂ηγ′ Dyγ′ p)(ξ ′ + θη ′ , x′ + y ′ , ξ ′ , μ; xn ) dy ′ d¯η ′

d−N
is, for each μ and xn , a symbol in S1,0 (Rn−1 ; C L1 ×M0 ).

By virtue of (2.6) the terms in the first sum on the right-hand side of (4.10) satisfy
⃓
′ ′ ⃓ ′
∂ηα′ Dyα′ p(η ′ , y ′ , ξ ′ , μ; xn )⃓y′ =x′ ∈ 𝔖−|α | (Rn−1 × R+ ; S (R+ , C L1 ×M0 ));
η ′ =ξ ′

note here that ω(ξ ′ , ξ ′ , μ) ≡ 1. In particular, the first term (with α′ = 0) is

⃓
⃓
p(η ′ , y ′ , μ; xn )⃓y′ =x′ = k0′ (x′ , ξ ′ , μ; xn ).
η ′ =ξ ′

Hence, to complete the proof of Theorem 4.4 it remains to show that

ˆ1
(1 − θ)N −1 rγ ′ ,θ (x′ , ξ ′ , μ; xn ) dθ ∈ 𝔖d−N (Rn−1 × R+ ; S (R+ , C L1 ×M0 )) (4.11)
0

for every γ ′ with |γ ′ | = N . To this end we make use of the following result:

Theorem 4.10 (and Definition). Let F be a Frèchet space whose topology is given by a
system of semi-norms {∥| · ∥|(ℓ) | ℓ ∈ N}. A smooth function a(y ′ , η ′ ) on Rn−1 × Rn−1
with values in F is called an amplitude function if there exists a sequence τ = (τℓ ) of
real numbers such that
′ ′
∥|Dηα′ Dyβ′ a(y ′ , η ′ )∥|(ℓ) ≲ ⟨η ′ ⟩τℓ

uniformly in (y ′ , η ′ ) for every choice of indices ℓ, α′ , β ′ . Then

¨ ¨
′ ′ ′ ′
Os[a] := Os − eiy η a(y ′ , η ′ ) dy ′ d¯η ′ := lim eiy η χ(εy ′ , εη ′ )a(y ′ , η ′ ) dy ′ d¯η ′
ε→0

∞
exists in F , where χ ∈ Ccomp (Rn−1 × Rn−1 ) with χ(0, 0) = 1. The space 𝒜τ (Rn−1 ; F )
of all such amplitudes is a Fréchet space in the obvious way and the map a →
↦ Os[a] is
continuous.

If τ 0 , τ 1 are two sequences and B : F0 × F1 → F is a bilinear continuous mapping

between Fréchet spaces, then B induces a continuous map
0 1
B : 𝒜τ (Rn−1 ; F0 ) × 𝒜τ (Rn−1 ; F1 ) −→ 𝒜τ (Rn−1 ; F ),

with a resulting sequence τ = τ (τ 0 , τ 1 ).

 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 37

Proposition 4.11. With the above notation and |γ ′ | = N , the function

′ ′
a(y ′ , η ′ ; x′ , ξ ′ , μ, θ; xn ) := (∂ηγ′ Dyγ′ p)(ξ ′ + θη ′ , x′ + y ′ , ξ ′ , μ; xn )

is an amplitude function of (y ′ , η ′ ) with values in

(︁ )︁
C [0, 1], 𝔖d−N (Rn−1 × R+ ; S (R+ , C L1 ×M0 )) .

This result yields that the oscillatory integral defining rγ ′ ,θ in Proposition 4.9 con­
(︁ )︁
verges in C [0, 1], 𝔖d−N (Rn−1 × R+ ; S (R+ , C L1 ×M0 )) , hence integration with respect
to θ yields (4.11) and thus finishes the proof of Theorem 4.4.
The proof of Proposition 4.11 is again split into different steps. To begin with, let us
recall that S (R+ ) is a nuclear Fréchet space and that

ˆ︁ π F
S (R+ , F ) = S (R+ )⊗

with the completed projective tensor-product. By a classical result thus every function
u ∈ S (R+ , F ) can be written as a series

∑︂
+∞
u(t) = cℓ uℓ (t)fℓ
ℓ=0

where (cℓ ) ⊂ C is an absolutely summable numerical sequence and (uℓ ), (fℓ ) are in­
finitesimal sequences in S (R+ ) and F , respectively. Applying this to the symbol-kernel
k0′ it is therefore no restriction of generality to assume that k0′ has the form

k0′ (x′ , ξ ′ , μ; xn ) = r(x′ , ξ ′ , μ)ϕ(xn )

with r(x′ , ξ ′ , μ) ∈ 𝔖d (Rn−1 × R+ ; C L1 ×M0 ) and ϕ(xn ) ∈ S (R+ ). Correspondingly,

p(η ′ , y ′ , η ′′ , μ; xn ) = r(y ′ , η ′′ , μ) (κω(η′ ,η′′ ,μ) ϕ)(xn ) . (4.12)

⏞ ⏟⏟ ⏞
=:pϕ (η ′ ,η ′′ ,μ;xn )

It is obvious that
′
a1 (y ′ , x′ , ξ ′ , μ) := Dyγ′ r(x′ + y ′ , ξ ′ , μ)

is an amplitude function in y ′ with values in 𝔖d (Rn−1 × R+ ; C L1 ×M0 ). In fact, whenever

f ∈ Cb∞ (Rn−1 , F ) with some Fréchet space F , then f (x′ + y ′ ) is an amplitude function
of y ′ with values in Cb∞ (Rn−1 , F ).

Lemma 4.12. With the above notation and |γ ′ | = N , the function

′
aϕ (η ′ , ξ ′ , μ, θ; xn ) := (∂ηγ′ pϕ )(ξ ′ + θη ′ , ξ ′ , μ; xn )
38 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

(︁ )︁
is an amplitude in η ′ with values in C [0, 1], S −N (Rn−1 × R+ ; S (R+ , C L1 ×M0 )) .

Proof. For simplicity of presentation we shall show the lemma only in case N = 0; the
general case is verified in exactly the same way and only comes along with more complex
notation due to the involved chain rule.
Next note that it is enough to show that aϕ (η ′ , ξ ′ , μ; xn ) := aϕ (η ′ , ξ ′ , μ, 1; xn ) is an
amplitude function with values in S 0 (Rn−1 × R+ ; S (R+ , C L1 ×M0 )); in fact, if a(y ′ , η ′ )
is an amplitude function with values in a Fréchet space F then ˜︁ a(y ′ , η ′ ; θ) := a(y ′ , θη ′ )
is an amplitude with values in C ([0, 1], F ).
We need to show that the homogeneous components aϕ (η ′ , ξ ′ , μ; xn ) are amplitude
(−j)

−j
functions with values in Shom (Rn−1 × R+ ; S (R+ , C L1 ×M0 )) and that

∑︂
L−1
aϕ (η ′ , ξ ′ , μ; xn ) − χ(ξ ′ , μ)aϕ (η ′ , ξ ′ , μ; xn )
(−j)

j=0

−L
is an amplitude with values in S1,0 (Rn−1 × R+ ; S (R+ )) for arbitrary L.
′
Due to (2.6), ∂ηα′ aϕ (η ′ , ξ ′ , μ; xn ) is a finite linear combination of terms

(︂ ∏︂
m )︂
α′
[ξ ′ + η ′ , μ]m ∂η′ℓ [ξ ′ + η ′ , μ]−1 κω(ξ′ +η′ ,ξ′ ,μ) Θm ϕ(xn ) (4.13)
ℓ=1

with 0 ≤ m ≤ |α′ | and α1′ + . . . + αm

′
= α′ . Therefore, since ω(ξ ′ , ξ ′ , μ) ≡ 1,
′
|α |
⃓ ∑︂
∂ηα′ aϕ (η ′ , ξ ′ , μ; xn )⃓η′ =0
′
′
=α! qα′ ,m (ξ ′ , μ)Θm ϕ(xn )
m=0

′
with certain qα′ ,m (ξ ′ , μ) ∈ S −|α | (Rn−1 ×R+ ) which are positively homogeneous of degree
−|α′ | for |ξ ′ , μ| ≥ 1. Thus, by Taylor expansion,
′
|α |
∑︂
L−1 ∑︂ ′
′ ′
aϕ (η , ξ , μ; xn ) = qα′ ,m (ξ ′ , μ)Θm ϕ(xn )η ′ α +
|α′ |=0 m=0

∑︂ ˆ1
′ β′ (1 − t)L−1 β ′
+N η (∂η′ aϕ )(tη ′ , ξ ′ , μ; xn ) dt
β′!
|β ′ |=L 0

′
It remains to show that (∂ηβ′ aϕ )(η ′ , ξ ′ , μ; xn ), |β ′ | = N , is an amplitude with values in
−L
S1,0 (Rn−1 × R+ ; S (R+ )) because then we have verified our claim, since then aϕ has the
homogeneous components

∑︂ ∑︂
j
′
(η ′ , ξ ′ , μ; xn ) = qα′ ,m (ξ ′ , μ)Θm ϕ(xn )η ′ α ,
(−j)
aϕ
|α′ |=j m=0
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 39

′
which are clearly amplitude functions of η ′ . As seen from (4.13), (∂ηβ′ aϕ )(η ′ , ξ ′ , μ; xn ) is
a linear combination of terms

q(ξ ′ + η ′ , μ)κω(ξ′ +η′ ,ξ′ ,μ) ψ(xn ), q ∈ S −L (Rn−1 × R+ ), ψ ∈ S (R+ ).

From
′ ′ ′ ′ ′ ′
|Dηγ′ Dξα′ Dμj q(ξ ′ + η ′ , μ)| ≲ ⟨ξ ′ + η ′ , μ⟩−L−|α |−|γ |−j
≤ ⟨η ′ ⟩L+|α |+j ⟨ξ ′ , μ⟩−L−|α |−j

it follows that q(ξ ′ + η ′ , μ) is an amplitude function with values in S −L (Rn−1 × R+ ).

Finally, let

b(η ′ ; ξ ′ , μ; xn ) = κω(ξ′ +η′ ,ξ′ ,μ) ψ(xn ).

′ ′ ′
Then xℓn Dxℓ n Dηγ′ Dξα′ Dμj b is a finite linear combination of terms

(︂ ∏︂
m )︂
′ γ ′ α′ ′
ω ℓ −ℓ ω −m Dη′i Dξ′i Dμji ω κω xℓn Dxℓ n Θm ψ
i=1

with m ≤ |γ ′ | + |α′ | + j, γ1′ + . . . + γm

′
= γ ′ , α1′ + . . . + αm
′
= α′ , and j1 + . . . + jm = j;
here, ω is a short notation for ω(ξ ′ + η ′ , ξ ′ , μ). We can estimate
′ ′ ′ ′
|ω(ξ ′ + η ′ , ξ ′ , μ)ℓ −ℓ | = [ξ ′ , μ]ℓ −ℓ [ξ ′ + η ′ , μ]ℓ−ℓ ≲ ⟨η ′ ⟩ℓ+ℓ

as well as

γ′ α′ [ξ ′ + η ′ , μ] ∑︂ ′′ j′ γ′ ′′′ j ′′
ω −1 |Dη′i Dξ′i Dμji ω| ≲ |Dαi Dμi [ξ ′ , μ]||Dη′i Dαi Dμi [ξ ′ + η ′ , μ]−1 |
[ξ ′ , μ]
α′′ ′′′
i +αi =αi
′

ji′ +ji′′ =ji

∑︂ ′′ ′ ′′′
|−ji′′ −|γi′ |
≲ [ξ ′ , μ]−|αi |−ji [ξ ′ + η ′ , μ]−|αi
α′′ ′′′
i +αi =αi
′

ji′ +ji′′ =ji

′ ′
≲ ⟨ξ ′ , μ⟩−|αi |−ji ⟨η ′ ⟩|αi |+ji

This yields
′ ′ ′ ′ ′ ′
|xℓn Dxℓ n Dηγ′ Dξα′ Dμj b(η ′ ; ξ ′ , μ; xn )| ≲ ⟨ξ ′ , μ⟩−|α |−j ⟨η ′ ⟩ℓ+ℓ +|α |+j+ 2 .
1

0
Hence b is an amplitude function with values in S1,0 (Rn−1 × R+ ; S (R+ )). Thus
′
(∂ηβ′ aϕ )(η ′ , ξ ′ , μ; xn ) is as claimed and the proof is complete. □

This finishes the proof of Theorem 4.8.a). The proof of b) is very similar and therefore
we shall only indicate the necessary adjustments in the above proof but leave the details
40 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

to the reader. Using block-matrix representations as above, we need to find g ∈ 𝔅d;r

G
such that
(︃ )︃
op(κ)(μ)op(g ′ )(μ)op(κ−1 )(μ) op(κ)(μ)op(k′ )(μ)
op(g)(μ) = .
op(t′ )(μ)op(κ−1 )(μ) op(q ′ )(μ)

Therefore q = q ′ and t = t′ κ−1 . Then determine symbols κ#g ′ and κ#k′ such that

op(κ)(μ)op(g ′ )(μ) = op(κ#g ′ )(μ), op(κ)(μ)op(k′ )(μ) = op(κ#k′ )(μ)

and define g := (κ#g ′ )κ−1 and k := κ#k′ . One then has to show that

κ−1 gκ = κ−1 (κ#g ′ ) ∈ 𝔖(Rn−1 × R+ ; ΓrG (R+ ; (L0 , 0), (L1 , 0)),
κ−1 k = κ−1 (κ#k′ ) ∈ 𝔖(Rn−1 × R+ ; ΓrG (R+ ; (0, M0 ), (0, M1 ))

and, in order to verify (4.5),

κ−1 gκ − g ′ ∈ 𝔖d−1 (Rn−1 × R+ ; ΓrG (R+ ; (L0 , 0), (L1 , 0)),

κ−1 k − k′ ∈ 𝔖d−1 (Rn−1 × R+ ; ΓrG (R+ ; (0, M0 ), (0, M1 )).

Treating as above only on the Poisson part, to calculate κ#k′ we have

[op(κ)(μ)op(k′ )(μ)u](x′ )
˘
′ ′ ′ ′ ′′ ′′
= ei(x −y )η +i(y −y )η p′ (η ′ , y ′ , η ′′ , μ)u(y ′′ ) dy ′′ d¯η ′′ dy ′ d¯η ′

with p′ (η ′ , y ′ , η ′′ , μ) = κ(η ′ , μ)k′ (y ′ , η ′′ , μ). Thus passing to symbol-kernels, in (4.9), p

needs to be replaced by

p(η ′ , y ′ , η ′′ , μ; xn ) = (κ(η ′ , μ)k′ )(y ′ , η ′′ , μ; xn ).

Then the analogue of Proposition 4.9 and (4.10) for op(κ)(μ)op(k′ )(μ) is valid; in partic­
ular, the leading term in (4.10) is (κ(ξ ′ , μ)k′ )(x′ , ξ ′ , μ; xn ). In (4.11) we have to substitute
rγ ′ ,θ by κ−1 (ξ ′ , μ)rγ ′ ,θ and, correspondingly, in Proposition 4.11, a by κ−1 (ξ ′ , μ)a. With
k′ (x′ , ξ ′ , μ; xn ) = r(x′ , ξ ′ , μ)ϕ(xn ) in (4.12) we then need to consider instead

p(η ′ , y ′ , η ′′ , μ; xn ) = r(y ′ , η ′′ , μ)(κ−1 (η ′′ , μ)κ(η ′ , μ)ϕ)(xn ) = (κ1/ω(η′ ,η′′ ,μ) ϕ(xn ),

⏞ ⏟⏟ ⏞
=:pϕ (η ′ ,η ′′ ,μ;xn )

with ω as defined in (4.8). Therefore, to prove b), it remains to repeat the proof of
Lemma 4.12 with the function ω replaced by 1/ω.

˜︁ d,ν;r (Rn−1 × R+ ; 𝔤). Then g∞ = (𝜿−1 #g#𝜿)∞ .

Corollary 4.13. Let g ∈ B G [d,ν] [d,ν]
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 41

4.5. The calculus of generalized singular Green symbols

The composition of generalized singular Green operators on R+ induces continuous

mappings

ΓrG1 (R+ ; 𝔤1 ) × ΓrG0 (R+ ; 𝔤0 ) −→ ΓrG0 (R+ ; 𝔤1 𝔤0 ),

taking the formal adjoint with respect to the L2 -inner product yields continuous map­
pings

Γ0G (R+ ; 𝔤) −→ Γ0G (R+ ; 𝔤−1 ).

˜︁ dj ,νj (Rn−1 × R+ ; Γrj (R+ ; 𝔤j )) then, following the proof in Section 8 of [29],
Given gj′ ∈ S G
one can show that
¨
′ ′
′ ′ ′ ′
(g1 #g0 )(x , ξ , μ) := Os − e−iy η g1′ (x′ + y ′ , ξ ′ , μ)g0′ (x′ , ξ ′ + η ′ , μ) dy ′ d¯η ′

˜︁ d0 +d1 ,ν0 +ν1 (Rn−1 × R+ ; Γr0 (R+ ; 𝔤1 𝔤0 )) and

is an oscillatory integral converging in S G

op(g1′ #g0′ )(μ) = op(g1′ )(μ)op(g0′ )(μ),

(g1′ #g0′ )∞ ′ ∞ ′ ∞
[d0 +d1 ,ν0 +ν1 ] = (g1 )[d1 ,ν1 ] (g0 )[d0 0,ν0 ] .

˜︁ d,ν (Rn−1 × R+ ; Γ0 (R+ ; 𝔤)) with d ≤ 0 then

Likewise, if g′ ∈ S G
ˆ
′ ′
g′ (x′ , ξ ′ , μ) := Os − e−iy η g′ (x′ + y ′ , ξ ′ + η ′ , μ)∗ dy ′ d¯η ′
(∗)

˜︁ d,ν (Rn−1 × R+ ; Γ0 (R+ ; 𝔤−1 )) and

converges in S G

∞
op(g′ ) = op(g′ )∗ , (g′
(∗) ∞
= (g′ [d,ν] )(∗) .
(∗)
)[d,ν]

Combining this with Theorem 4.8 and Corollary 4.13 we immediately get the analogue
of Theorem 2.25 for generalized singular Green symbols with limit symbol at infinity.
At the same time, this way of reasoning also yields an alternative proof for the algebra
property of generalized Green symbols with strong parameter dependence.

Theorem 4.14. Let d ≤ 0. Then g ↦→ g(∗) induces continuous maps

˜︁ d,ν;0 (Rn−1 × R+ ; 𝔤) −→ B
B ˜︁ d,ν;0 (Rn−1 × R+ ; 𝔤−1 ).
G G

Both homogeneous principal symbol and principal limit symbol are well-behaved,

(g(∗) )(d,ν) = (g(d,ν) )∗ , (g(∗) )∞ ∞

[d,ν] = (g[d,ν] )
(∗)
.
42 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

Proof. Due to Theorem 4.8 we can write g = 𝜿#g′ #𝜿−1 . Since the dilation semi-group
is unitary on L2 , the group-action operator satisfies op(𝜿)(μ)∗ = op(𝜿−1 )(μ). Therefore

g(∗) = (𝜿#g′ #𝜿−1 )(∗) = 𝜿#(g′ )(∗) #𝜿−1 .

Then the claim follows immediately by Corollary 4.13. □

Theorem 4.15. Let 𝔤1 be composable with 𝔤0 . Then (g1 , g0 ) ↦→ g1 #g0 induces continuous
maps

˜︁ d1 ,ν1 ;r1 (Rn−1 × R+ ; 𝔤1 ) × B

B ˜︁ d0 ,ν0 ;r0 (Rn−1 × R+ ; 𝔤0 )
G G
˜︁ d0 +d1 ,ν0 +ν1 ;r0 (Rn−1 × R+ ; 𝔤1 𝔤0 ).
−→ B G

Both homogeneous principal symbol and principal limit symbol are multiplicative,

(d1 ,ν1 ) (d0 ,ν0 )

(g1 #g0 )(d0 +d1 ,ν0 +ν1 ) = g1 g0 ,
(g1 #g0 )∞ ∞ ∞
[d0 +d1 ,ν0 +ν1 ] = (g1 )[d1 ,ν1 ] (g0 )[d0 ,ν0 ] .

Proof. According to Theorem 4.8 we can write gj = 𝜿#gj′ #𝜿−1 . Then

g1 #g0 = (𝜿#g1′ #𝜿−1 )#(𝜿#g0′ #𝜿−1 ) = 𝜿#(g1′ #g0′ )#𝜿−1 .

From this together with Corollary 4.13 the claim follows immediately. □

˜︁ d,ν;r . We shall
The analogue of the previous theorem also holds true in the classes B G
also need the following result on asymptotic summation.

Proposition 4.16. Let gj ∈ B ˜︁ d−j,ν−j;r (Rn−1 × R+ ; 𝔤), j ∈ N0 . Then there exists a symbol
G
g∈B ˜︁ d,ν;r
(Rn−1 × R+ ; 𝔤) such that, for every N ∈ N0 ,
G

∑︂
N −1
g− ˜︁ d−N,ν−N ;r (Rn−1 × R+ ; 𝔤).
gj ∈ B G
j=0

Proof. By conjugation with 𝜿, the claim follows from the corresponding claim of asymp­
totic summation for a sequence of symbols gj′ ∈ S ˜︁ d−j,ν−j (Rn−1 × R+ ; Γr (R+ ; 𝔤)) in the
G
˜︁ (R
class S d,ν n−1
× R+ ; ΓG (R+ ; 𝔤)). In turn this is a special case of asymptotic summation
r

of a sequence of symbols pj ∈ S ˜︁ d−j,ν−j (Rn−1 × R+ ; F ) in the class S

˜︁ d,ν (Rn−1 × R+ ; F ),
where F is some Fréchet space. In case F = C this has been proved in [29], the general
case is verified in the same way. □

Proposition 4.17. Let gj ∈ B ˜︁ d−j,ν;r (Rn−1 × R+ ; 𝔤), j ∈ N0 . Then there exists a symbol
G
g∈B ˜︁ d,ν;r (Rn−1 × R+ ; 𝔤) such that, for every N ∈ N0 ,
G
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 43

∑︂
N −1
g− ˜︁ d−N,ν;r (Rn−1 × R+ ; 𝔤).
gj ∈ B G
j=0

The proof of this result is analogous to that of Proposition 4.16. Let us conclude this
section with a remark concerning regularizing symbols:

Remark 4.18. For arbitrary weight-datum 𝔤,

−∞;r
BG ˜︁ d−∞,ν−∞;r (Rn−1 × R+ ; 𝔤) ⊂ B
(Rn−1 × R+ ; 𝔤) ⊂ B ˜︁ d−∞,ν−∞;r (Rn−1 × R+ ; 𝔤),
G G

and
−∞;r
BG (Rn−1 × R+ ; 𝔤) = ˜︁ d−N,ν;r (Rn−1 × R+ ; 𝔤)
∩ B G
N ∈N0

= ∩ B˜︁Gd−N,ν;r (Rn−1 × R+ ; 𝔤)
N ∈N0

= S −∞ (Rn−1 × R+ ; ΓrG (R+ , 𝔤)).

5. The full calculus with expansion at infinity

5.1. Definition and symbolic structure

The full calculus is obtained by enlarging the strongly parameter-dependent calculus

by the new class of generalized singular Green symbols.

Definition 5.1. Let d, ν ∈ Z, r ∈ N0 , and 𝔤 = ((L0 , M0 ), (L1 , M1 )). Then

Bd,ν;r d;r
(Rn−1 × R+ ; 𝔤) := BG ˜︁ d,ν;r (Rn−1 × R+ ; 𝔤).
(Rn−1 × R+ ; 𝔤) + B
G G

It is obvious by construction that

Bd,ν;r ˜︁ d,ν;r (Rn−1 × R+ ; 𝔤),

(Rn−1 × R+ ; 𝔤) = B ν ≤ 0,
G G

Bd,ν;r ˜︁ d,0;r (Rn−1 × R+ ; 𝔤),

(Rn−1 × R+ ; 𝔤) ⊂ B ν ≥ 1.
G G

Therefore we can associate with g ∈ Bd,ν;rG (Rn−1 × R+ ; 𝔤) its homogeneous principal

∞
symbol g(d,ν− ) and the principal limit-symbol g[d,ν −]
where ν− := min(0, ν).

Definition 5.2. Let d ∈ Z, ν, r ∈ N0 , and 𝔤 = ((L0 , M0 ), (L1 , M1 )). Then

˜︁ d,ν;r (Rn−1 × R+ ; 𝔤).

Bd,ν;r (Rn−1 × R+ ; 𝔤) := B d;r (Rn−1 × R+ ; 𝔤) + B G

The elements of Bd,ν;r (Rn−1 × R+ ; 𝔤) are called boundary symbols. The class of regu­
larizing boundary symbols is

B−∞,ν;r (Rn−1 × R+ ; 𝔤) = B −∞;r (Rn−1 × R+ ; 𝔤).

44 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

Equivalently, any p ∈ Bd,ν;r (Rn−1 × R+ ; 𝔤) has the form

(︃ )︃
′ ′ op+ (a)(x′ , ξ ′ , μ) 0
p(x , ξ , μ) = + g(x′ , ξ ′ , μ),
0 0

where a ∈ S d (Rn × R+ ; L (C L0 , C L1 )) satisfies the transmission condition and g is a

generalized singular Green symbol from Bd,ν;r G (Rn−1 × R+ ; 𝔤). Then the homogeneous
principal symbol of p is

(d)
σψ (p)(x, ξ, μ) = a(d) (x, ξ, μ), xn ≥ 0, (ξ, μ) ̸= 0. (5.1)

Definition 5.3. Representing p ∈ Bd,ν;r (Rn−1 × R+ ; 𝔤) as p = p0 + g ˜︁ with p0 ∈

d;r
B (R n−1 ˜︁ d,ν;r
˜︁ ∈ BG (R
× R+ ; 𝔤) and g n−1
× R+ ; 𝔤), the principal boundary symbol of p
is

(p)(x′ , ξ ′ , μ) := σ∂ (p0 )(x′ , ξ ′ , μ) + g

˜︁(d,ν) (x′ , ξ ′ , μ), ξ ′ ̸= 0,
(d,ν) (d)
σ∂

the principal limit symbol of p is

(p)(x′ , ξ ′ ) := σ∂ (p0 )(x′ , 0, 1) + g ∞

(x′ , ξ ′ ).
(d)
[d]
σ∞ ˜︁[d,0]

∞
˜︁[d,0]
Note that, in the previous definition, g = 0 in case ν ≥ 1. The subordinate
principal angular symbol is
(︂ ξ ′ √︁ )︂
ˆ︁⟨d⟩ (p)(x′ , ξ ′ ) : = lim σ∂ (p) x′ , r ′ , 1 − r2
(d,ν)
σ
r→0+ |ξ |
= σ∂ (p0 )(x′ , 0, 1) + g
˜︁⟨d,0⟩ (x′ , ξ ′ ), ξ ′ ̸= 0.
(d)

ˆ︁⟨d⟩ (p) is the homogeneous principal

˜︁⟨d,0⟩ = 0 in case ν ≥ 1. Moreover, σ
Also here g
[d]
symbol of the limit symbol σ∞ (p), i.e.,

ˆ︁⟨d⟩ (p) = (σ∞

σ [d]
(p))(0) .

Remark 5.4. For p ∈ B d;r (Rn−1 ×R+ ; 𝔤) considered as an element of Bd,0;r (Rn−1 ×R+ ; 𝔤)
we have

(p)(x′ , ξ ′ ) = σ
ˆ︁⟨d⟩ (p)(x′ , ξ ′ ) = σ∂ (p)(x′ , 0, 1),
[d] (d)
σ∞

i.e., both principal limit-symbol and principal angular symbol are determined by the
principal boundary symbol. This is why ellipticity for symbols with regularity number
ν ≥ 1 (thus, in particular, for strongly parameter-dependent symbols) is covered by
the homogeneous principal symbol and the boundary symbol alone. The principal limit­
symbol will be relevant for symbols with regularity number ν = 0.
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 45

For a further description of the principal symbols we need to understand better the
behavior of pseudodifferential operators on the half-axis under conjugation with the
dilation group-action. By direct calculation, if a(xn , ξn ) ∈ Str
d
(R), then

κ−1 + +
λ op (a)κλ = op (aλ ), aλ (xn , ξn ) = a(xn /λ, λξn )

for every λ > 0. This already yields formula (5.2) in the following lemma.

Lemma 5.5. Let a(x′ , ξ, μ) ∈ Str

d
(Rn × R+ ; L (C L1 , C L2 )) be independent of the xn ­
variable. Define

a′ (x′ , ξ ′ , μ; ξn ) = a(x′ , ξ ′ , [ξ ′ , μ]ξn , μ).

Then a′ ∈ S d (Rn−1 × R+ ; Str

d
(R, L (C L1 , C L2 ))) and

κ−1 (ξ ′ , μ)op+ (a)(ξ ′ , μ)κ(ξ ′ , μ) = op+ (a′ )(ξ ′ , μ). (5.2)

The homogeneous principal symbol of a′ is

a′ (d) (x′ , ξ ′ , μ; ξn ) = a(d) (x′ , ξ ′ , |ξ ′ , μ|ξn , μ).

Proof. For convenience of notation let L1 = L2 = 1. It is a well-known fact that a′ ∈

d
S1,0 (Rn−1 × R+ ; Str d
(R)); see, for example, [16, p. 10] for a short explanation (where ξ ′
has to be substituted by (ξ ′ , μ)). It remains to show the poly-homogeneity.
Let χ(ξ ′ , μ) and χ(ξ,
ˆ︁ μ) be two zero-excision functions such that χˆ︁ χ = χ and χ(ξ ′ , μ) =
0 whenever |ξ ′ , μ| ≤ 1. Then, for arbitrary N ∈ N0 ,

∑︂
N −1
a(x′ , ξ, μ) = ˆ︁ μ)a(d−j) (x′ , ξ, μ) + rN (x′ , ξ, μ)
χ(ξ,
j=0

with rN ∈ Str
d−N
(Rn × R+ ). Therefore

∑︂
N −1
′ ′ ′
a (x , ξ , μ; ξn ) = ˆ︁ ′ , [ξ ′ , μ]ξn , μ)a(d−j) (x′ , ξ ′ , [ξ ′ , μ]ξn , μ) + rN
χ(ξ ′
(x′ , ξ, μ; ξn )
j=0

′
with rN ∈ S1,0
d−N
(Rn−1 × R+ ; Str
d−N
(R)) ⊂ S1,0
d−N
(Rn−1 × R+ ; Str
d
(R)). Moreover,

ˆ︁ ′ , [ξ ′ , μ]ξn , μ)a(d−j) (x′ , ξ ′ , [ξ ′ , μ]ξn , μ) = χ(ξ ′ , μ)a(d−j) (x′ , ξ ′ , |ξ ′ , μ|ξn , μ)+

χ(ξ
+ (1 − χ)(ξ ′ , μ)(ˆ︁
χa(d−j) )′ (x′ , ξ ′ , μ; ξn ).

The second summand belongs to S −∞ (Rn−1 × R+ ; Str

d−j
(R)). Thus
46 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

∑︂
N −1
a′ (x′ , ξ ′ , μ; ξn ) ≡ χ(ξ ′ , μ)a(d−j) (x′ , ξ ′ , |ξ ′ , μ|ξn , μ)
j=0

d−N
modulo S1,0 (Rn−1 × R+ ; Str
d
(R)). This completes the proof. □

As a consequence of Lemma 5.5, Theorem 4.4 and Definition 4.5,

𝜿−1 (p)(x′ , ξ ′ , μ)𝜿|ξ′ ,μ| ∈ Sd,ν

(d,ν)
|ξ ′ ,μ| σ∂ hom (R
n−1
× R+ ; B d,r (R+ ; 𝔤))

where the latter space is that of (3.6) with F = B d,r (R+ ; 𝔤). Moreover,

[d]
σ∞ (p) ∈ S 0 (Rn−1 ; B d;r (R+ , 𝔤)).

5.1.1. Composition with powers of xn

If a = a(xn ) ∈ C ∞ (R+ , L (C L )) is a smooth function on R+ we may consider the
associated operator op+ (a), i.e. the operator of multiplication by a, as a symbol constant
in the variables (x′ , ξ ′ , μ). For simplicity of notation we shall use the same notation a
for the function a, the associated operator on R+ , and the associated pseudodifferential
symbol.
A case of particular importance is a(xn ) = xjn idC L , j ∈ N0 . In this case we shall write
simply xjn for every choice of the value L. Note that
⃓
[(κ−1 j n j ⃓ −j j
λ xn κλ )u](x ) = (t u(λt)) t=λ−1 x = λ xn u(xn ), n

i.e., the symbol xjn is twisted homogeneous of degree −j; in particular, xjn ∈ S −j (Rn−1 ×
R+ ; E0 , E1 ) for every choice of spaces E0 , E1 carrying the dilation group-action such
that xjn ∈ L (E0 , E1 ). We will use the notation
(︃ )︃
xjn 0
xnj := ∈ S −j (Rn−1 × R+ ; E0 ⊕ C M0 , E1 ⊕ C M1 )
0 0

without explicitly indicating the involved spaces Eℓ and numbers Mj .

d;r ˜︁ d,ν;r
Lemma 5.6. Let 𝔅d;r
G represent one choice of BG , BG , or Bd,ν;r
G .
d;r d−j;r
If g ∈ 𝔅G (R n−1
× R+ ; 𝔤), then xn g, gxn ∈ 𝔅G (R
j j n−1
× R+ ; 𝔤).

Though rather obvious (since xjn is strongly parameter-dependent of order −j), this
result is technically very important for the calculus and will be used various times in the
sequel.

5.2. Composition and formal adjoint

It is immediate from Theorems 2.24 and 4.14 that the calculus is closed under taking
the formal adjoint:
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 47

Theorem 5.7. Let d ≤ 0. Then the map p ↦→ p(∗) (formal adjoint with respect to the
L2 -inner product) induces a continuous map

Bd,ν;0 (Rn−1 × R+ ; 𝔤) −→ Bd,ν;0 (Rn−1 × R+ ; 𝔤−1 ).

All principal symbols behave natural under taking the formal adjoint.

It will require more work to verify the closedness under composition.

Theorem 5.8. Let 𝔤1 be composable with 𝔤0 . Then (p1 , p0 ) ↦→ p1 #p0 induces continuous
maps

Bd1 ,ν1 ;r1 (Rn−1 × R+ ; 𝔤1 ) × Bd0 ,ν0 ;r0 (Rn−1 × R+ ; 𝔤0 )

−→ Bd,ν;r (Rn−1 × R+ ; 𝔤1 𝔤0 ),

where (recall that νj ≥ 0)

d = d0 + d1 , ν = min(ν0 , ν1 ), r = max(r1 + d0 , r0 ).

All principal symbols are multiplicative, i.e.,

(d +d1 ) (d ) (d )
σψ 0 (p1 #p0 ) = σψ 1 (p1 )σψ 0 (p0 ),
(d0 +d1 ,ν) (d1 ,ν1 ) (d0 ,ν0 )
σ∂ (p1 #p0 ) = σ∂ (p1 )σ∂ (p0 ),
[d0 +d1 ] [d1 ] [d0 ]
σ∞ (p1 #p0 ) = σ∞ (p1 )#σ∞ (p0 ).

Consequently, the principal angular symbol behaves multiplicatively. The class of gener­
alized singular Green operators is a two-sided ideal with respect to composition.

Before proving this result let us first make some informal discussion. Let us write
˜︁j with pj,0 ∈ B dj ;rj (Rn−1 × R+ ; 𝔤j ) and g
pj = pj,0 + g ˜︁ dj ,νj ;rj (Rn−1 × R+ ; 𝔤1 ).
˜︁j ∈ B G
Then

p1 #p0 = p1,0 #p0,0 + p1,0 #˜︁ ˜︁1 #p0,0 + g

g0 + g ˜︁1 #˜︁
g0 .

The first and the last summand on the right-hand side are covered by Theorem 2.25 and
Theorem 4.15, respectively. Hence it remains to analyze the mixed terms. Writing pj,0
as the sum of a pseudodifferential part and a generalized singular Green operator, the
composition of the generalized singular Green operators is again covered by Theorem 4.15
d ;r
together with the fact that BGj j (Rn−1 × R+ ; 𝔤j ) ⊂ B ˜︁ dj ,0;rj (Rn−1 × R+ ; 𝔤j ). Hence
G
it remains to analyze compositions of the pseudodifferential part with the generalized
singular Green operators. For the first mixed term this means to analyze compositions
of the form op+ (a)#g and op+ (a)#k with a generalized singular Green symbol g and
48 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

a Poisson symbol k, for the second mixed term that of g#op+ (a) and t#op+ (a) with a
˜︁ G -classes. We shall focus on
trace symbol t, where g, k, t belong to the corresponding B
the terms involving g, since the other terms are treated analogously. Using the explicit
form of generalized singular Green symbols of type r it is not difficult to see that the
case of general type follows from that of type r = 0.

Lemma 5.9. Let a(x′ , ξ, μ) ∈ Str

d
(Rn × R+ ; L (C L1 , C L2 )) be independent of the xn ­
˜︁ j j (Rn−1 × R+ ; 𝔤j ), j = 0, 2, with 𝔤j = ((Lj , 0), (Lj+1 , 0)) for
variable. Let gj ∈ B
d ,ν;r
G
j = 0, 1, 2. Then

˜︁ d0 +d,ν;r0 (Rn−1 × R+ ; 𝔤1 𝔤0 ),
op+ (a)g0 ∈ B G
˜︁ d2 +d,ν;r (Rn−1 × R+ ; 𝔤2 𝔤1 ),
g2 op+ (a) ∈ B r = max(r2 + d, 0).
G

Moreover,

(op+ (a)g0 )(d0 +d,ν) (x′ , ξ ′ , μ) = op+ (a(d) )(x′ , ξ ′ , μ)g0 (x′ , ξ ′ , μ),
(d0 ,ν)

(op+ (a)g0 )∞ ′ ′ + (d)

[d0 +d,ν] (x , ξ ) = op (a )(x′ , 0, 1)g0,[d,ν]
∞
(x′ , ξ ′ , μ),

and analogously for g2 op+ (a) with factors interchanged.

Proof. First we recall the continuity of the maps

(a′ , g0′ ) ↦→ op+ (a′ )g0′ : Str

d
(R; L (C L1 , C L2 )) × ΓrG0 (R+ ; 𝔤0 ) −→ ΓrG0 (R+ ; 𝔤1 𝔤0 ),
(g2′ , a′ ) ↦→ g2′ op+ (a′ ) : ΓrG2 (R+ ; 𝔤2 ) × Str
d
(R; L (C L1 , C L2 ))× −→ ΓrG (R+ ; 𝔤2 𝔤1 ).

Then the result follows immediately, since

κop+ (a)g0 κ−1 = κop+ (a)κ−1 κg0 κ−1 = op+ (a′ )g0′ ,
κg2 op+ (a)κ−1 = κg2 κ−1 κop+ (a)κ−1 = g2′ op+ (a′ ),

where gj′ ∈ S˜︁ dj ,ν (Rn−1 × R+ ; Γrj (R+ ; 𝔤j )) and, due to Lemma 5.5, op+ (a′ ) ∈ S d (Rn−1 ×
G
R+ ; B d,0 (R+ ; 𝔤1 )). □

Theorem 5.10. Let 𝔤j = ((Lj , 0), (Lj+1 , 0)). Let g0 ∈ B ˜︁ d0 ,ν;0 (Rn−1 × R+ ; 𝔤0 ) and a ∈
G
Str (R × R+ ; L (C , C )). Then op (a)#g0 ∈ B
d n L1 L2 + ˜︁ d0 +d,ν;0
(Rn−1 × R+ ; 𝔤1 𝔤0 ) and all
G
principal symbols behave multiplicatively.

Proof. For convenience of notation let us assume L0 = L1 = L2 = 1. First we consider

op+ (a) and g0 as operator-valued symbols in the sense of Theorem 2.22 and Defini­
tion 4.1, respectively. Then
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 49

∑︂
N −1
1 α′ + ′
op+ (a)#g0 = ∂ ′ op (a)Dxα′ g0 + rN ,
α′ ! ξ
|α′ |=0

for arbitrary N , where

∑︂ ˆ (1 − θ)N −1
1
′ ′
rN (x , ξ , μ) = N rγ ′ ,θ (x′ , ξ ′ , μ) dθ
′
γ′!
|γ |=N 0

where
¨
′ ′ ′ ′
rγ ′ ,θ (x′ , ξ ′ , μ) = Os − e−iy η ∂ηγ′ op+ (a)(x′ , ξ ′ + θη ′ , μ)Dyγ′ g0 (x′ + y ′ , ξ ′ , μ) dy ′ d¯η ′ .

′ d−|α′ |
Since ∂ξα′ a ∈ Str (Rn × R+ ),

′ d−|α′ |
∂ξα′ op+ (a) ∈ S1,0 (Rn−1 × R+ ; H s,δ (R+ ), H s−d,δ (R+ ))
′
⊂ S˜︁1,0
d−|α |,0
(Rn−1 × R+ ; H s,δ (R+ ), H s−d,δ (R+ ))

for every s and δ, the oscillatory integral defining rN converges in

S˜︁1,0
′ ′
∩
′ ′
d0 +d−N,ν;0
(Rn−1 × R+ ; H0s,δ (R+ ), H s ,δ (R+ )). (5.3)
s,s ,δ,δ

For |α′ | ≤ N − 1 write

N −1−|α′ |
′ ∑︂ 1 k α′ ′
∂ξα′ a(x, ξ, μ) = ∂ ∂ ′ a(x′ , 0, ξ, μ)xkn + xnN −|α | aα′ ,N (x, ξ, μ)
k! xn ξ
k=0

Since xℓn is twisted homogeneous of degree −ℓ, we have

xnN −|α | op+ (aα′ ,N ) ∈ S˜︁1,0

′ ′
d−N,0
(Rn−1 × R+ ; H s,δ (R+ ), H s−d,δ−N +|α | (R+ ))

for every s and δ. Altogether,

∑︂
N −1 ∑︂ 1 ′ ′
op+ (a)#g0 = ′
op+ (∂ξα′ ak )xkn Dxα′ g0 + r˜︁N ,
α!
ℓ=0 |α′ |+k=ℓ

where ak (x′ , ξ, μ) = (∂xkn a)(x′ , 0, ξ, μ)/k! and r˜︁N belongs to the space in (5.3). By the
previous Lemma 5.9,
∑︂ 1
g˜︁ℓ :=
′ ′
˜︁ d0 +d−ℓ,ν;0 (Rn−1 × R+ ; 𝔤1 𝔤0 ).
op+ (∂ξα′ ak )xkn Dxα′ g0 ∈ B
′
α! G
|α′ |+k=ℓ
50 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

By asymptotic summation there exists a g˜︁ ∈ B ˜︁ d0 +d,ν;0 (Rn−1 × R+ ; 𝔤1 𝔤0 ) such that

∑︁ ˜︁ 0
d +d−N,ν;0
G
g˜︁ − ℓ<N g˜︁ℓ ∈ B G (Rn−1 × R+ ; 𝔤) for every N ∈ N0 . In view of Remark 4.18
we can conclude that

−∞;0
g˜︁ − op+ (a)#g0 ∈ BG (Rn−1 × R+ ; 𝔤1 𝔤0 ).

This shows the first claim of the theorem and that

op+ (a)#g0 = op+ (a0 )g0 ˜︁ d0 +d−1,ν;0 (Rn−1 × R+ ; 𝔤1 𝔤0 ).

mod B G

The multiplicativity of the principal symbols then holds by Lemma 5.9. □

Theorem 5.11. Let 𝔤j = ((Lj , 0), (Lj+1 , 0)). Let g2 ∈ B ˜︁ d2 ,ν;0 (Rn−1 × R+ ; 𝔤2 ) and a ∈
G
Str (R × R+ ; L (C , C )). Then g2 #op (a) ∈ B
d n L1 L2 + ˜︁ d2 +d,ν;d +
(Rn−1 × R+ ; 𝔤2 𝔤1 ) and all
G
principal symbols behave multiplicatively.

Proof. Suppose the claim is true in case d2 ≤ 0. For general d2 we make use the reduction
of orders λd− (ξ, μ) from Section 2.3.1 by writing

g2 #op+ (a) = op+ (λd−2 )#(˜︁

g2 #op+ (a)),

where g˜︁2 := op+ (λ−d ˜︁ 0,ν;0 Rn−1 × R+ ; 𝔤2 ) due to Theorem 5.10. By the claim
− )#g2 ∈ BG
2

with d2 = 0, g˜︁2 #op+ (a) belongs to B ˜︁ d,ν;d+ (Rn−1 × R+ ; 𝔤2 𝔤1 ). Applying Theorem 5.10
G
yields g2 #op+ (a) ∈ B ˜︁ d2 +d,ν;d+ (Rn−1 × R+ ; 𝔤2 𝔤1 ). The multiplicativity of the principal
G
symbols follows analogously, first applying the case d2 = 0 to g˜︁2 #op+ (a) and then using
twice Theorem 5.10.
Next let us assume d2 ≤ 0. As a first case assume that a has order 0 and is independent
of ξ, i.e., a ∈ Cb∞ (Rn ) and op+ (a)(x′ ) is the operator of multiplication by a(x′ , ·). Writing
g2 #op+ (a) = (op+ (a∗ )#g2 )(∗) , the result follows from Theorems 5.10 and 5.7.
(∗)

Finally consider the general case a ∈ Str d

((Rn × R+ ; L (C L1 , C L2 )). Let us introduce
for this proof the space Sconstd
(Rn × R+ ; F ) of poly-homogeneous symbols of (ξ, μ) with
values in a Fréchet space F and Str,const d
(Rn × R+ ; F ) the subspace of symbols with
the transmission property (i.e., symbols whose homogeneous components satisfy the
symmetry-relations of Definition 2.21). Being a closed subspace of a nuclear Fréchet
d
space, Str,const (Rn × R+ ) is nuclear itself. It follows that

d
Str,const (Rn × R+ ; F ) = Str,const
d ˆ︁ π F
(Rn × R+ )⊗

with the completed projective tensor product. Taking F = Cb∞ (Rnx , L (C L1 , C L2 )) we

thus can represent a in the form

∑︂
+∞
a(x, ξ, μ) = λℓ aℓ (x)pℓ (ξ, μ)
ℓ=0
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 51

where (λℓ ) ⊂ C is an absolutely summable sequence, (aℓ ) ⊂ Cb∞ (Rn ) and (pℓ ) ⊂
d
Str,const (Rn × R+ ) are zero-sequences. Then

∑︂
+∞
(︁ )︁
g2 #op+ (a) = λℓ g2 #op+ (aℓ ) op+ (pℓ )
ℓ=0

and the result follows from the previous consideration and Lemma 5.9. □

5.3. Ellipticity and parametrix construction

We shall need the following result:

Lemma 5.12. Let g ∈ Bd,ν;rG (Rn−1 × R+ ; 𝔤), ν ∈ N0 with g(d,0) = 0. Then g ∈

d−1,ν−1;r
BG (Rn−1
× R+ ; 𝔤).

d;r
Proof. Only the case ν ≥ 1 needs to be proved. Write g = g0 + g ˜︁ with g0 ∈ BG (Rn−1 ×
˜︁
˜︁ ∈ B
R+ ; 𝔤) and g d,ν;r (d)
˜︁ ′ )
˜︁(d,ν) . Let χ(ξ, μ) and χ(ξ
(Rn−1 × R+ ; 𝔤). Then g(d,0) = g0 + g
G
be zero-excision functions such that χ˜︁ ˜︁ Then
χ = χ.

(d) (d)
g = (g0 − χg0 ) + (˜︁
g − χ˜︁
˜︁g(d,ν) ) + r, ˜︁g(d,ν) .
r = χg0 + χ˜︁

Now define

′ (d)
= 𝜿−1
(d)
|ξ ′ ,μ| g0 𝜿|ξ ,μ| ∈ Shom (R × R+ ; F ),
d n−1
g0 ′

˜︁′ (d,ν) = 𝜿−1

g ˜︁ d,ν (Rn−1 × R+ ; F ),
˜︁(d,ν) 𝜿|ξ′ ,μ| ∈ S
|ξ ′ ,μ| g hom

where F := ΓrG (R+ ; 𝔤). Also let r′ = 𝜿−1

|ξ ′ ,μ| r𝜿|ξ ,μ| .
′

′ (d) ′ (d)
From g0 = −˜︁ g′ (d,ν) it follows that g0 ∈ C ∞ (S+
n−1
, Cb∞ (Rn−1 , F )) vanishes to
order ν in (ξ ′ , μ) = (0, 1). As shown in Step 2 of the proof of [29, Proposition 5.3] this
′ (d)
implies χg0 ∈ S ˜︁ d,ν (Rn−1 × R+ ; F ). But then

′ (d) ′ (d)
r′ = χ(g0 ˜︁g′ (d,ν) ) = (1 − χ)χg
+ χ˜︁ ˜︁ 0

together with (1 − χ)
˜︁ ∈ S˜︁ 0−∞,0−∞ (Rn−1 × R+ ) yields r′ ∈ S
˜︁ d−∞,ν−∞ (Rn−1 × R+ ; F ),
˜︁
hence r ∈ B d−∞,ν−∞;r
(R n−1
× R+ ; 𝔤). The claim follows. □
G

Definition 5.13. Let d ∈ Z, ν ∈ N0 , and 𝔤 = ((L, M0 ), (L, M1 )). A symbol p ∈

Bd,ν;d+ (Rn−1 × R+ ; 𝔤) is called elliptic provided

(E1) the homogeneous principal symbol is pointwise invertible and

|σψ (p)−1 (x, ξ, μ)| ≲ 1

(d) n
∀ x∈R
xn ≥0 ∀ |ξ, μ| = 1,
52 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

(E2) for some (and then for all) s > d+ − 12 , the principal boundary symbol is pointwise
invertible and

(p)−1 (x′ , ξ ′ , μ)∥L (H s−d (R+ ,C L )⊕C M1 ,H s (R+ ,C L )⊕C M0 ) ≲ 1

(d,ν)
∥σ∂

uniformly for x′ ∈ Rn−1 and |ξ ′ , μ| = 1, ξ ′ ̸= 0,

(E3) for some (and then for all) s > d+ − 12 the following operator is an isomorphism:

L2 (Rn−1 , H s (R+ , C L )) L2 (Rn−1 , H s−d (R+ , C L )

[d]
op(σ∞ (p)) : ⊕ −→ ⊕ .
L2 (Rn−1 , C M0 ) L2 (Rn−1 , C M1 )

For the independence of condition (E2) of the specific value of s see [21, Theorem
3.30] for instance. Due to the spectral-invariance of pseudodifferential operators (see the
proof of Proposition 5.18 for more details), an equivalent formulation of (E3) is then as
follows:

(E3) For some (and then for all) s > d+ − 12 the principal limit symbol is invertible with
respect to the Leibniz product in

S 0 (Rn−1 ; L (H s (R+ , C L ) ⊕ C M0 , H s−d (R+ , C L ) ⊕ C M1 )),

i.e., there exists a symbol

q∞ ∈ S 0 (Rn−1 ; L (H s−d (R+ , C L ) ⊕ C M1 , H s (R+ , C L ) ⊕ C M0 ))

[d] [d]
such that σ∞ (p)#q∞ = 1 and q∞ #σ∞ (p) = 1.

Let us now show the independence of (E3) of s:

Lemma 5.14. If (E3) holds for some s0 > d+ − 1

2 then it holds for all s > d+ − 12 .

Proof. We consider the second version of (E3) and shall use the notation from Defini­
tion 5.3 and (5.1). Assume (E3) holds for s = s0 . (︃ + )︃
′ (d) ′ −1 ′ op (b)(x′ ) 0
Define b(x , ξn ) = a (x , 0, 0, ξn , 1) and q(x ) := . Then q ∈
0 0
S 0 (Rn−1 ; B −d;0 (R+ ; g−1 )) and

[d]
q#σ∞ (p) = 1 − gL , [d]
σ∞ (p)#q = 1 − gR (5.4)

with

gL ∈ S 0 (Rn−1 ; ΓG+ (R+ ; g−1 𝔤))

d

⊂ S 0 (Rn−1 ; L (H s0 (R+ , C L ) ⊕ C M0 , H s0 (R+ , C L ) ⊕ C M0 )),

 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 53

(R+ ; 𝔤𝔤−1 ))
(−d)+
gR ∈ S 0 (Rn−1 ; ΓG
⊂ S 0 (Rn−1 ; L (H s0 −d (R+ , C L ) ⊕ C M1 , H s0 −d (R+ , C L ) ⊕ C M1 )).

Applying the Leibniz product with q∞ from the right to the first identity in (5.4) and
from the left to the second identity gives q = q∞ − gL #q∞ and q = q∞ − q∞ #gR .
Resolving for q∞ yields q∞ = q + gL #q + gL #q∞ #qR . The second term on the
right belongs to S 0 (Rn−1 ; ΓG + (R+ ; g−1 )). The same is true for the third term. In
(−d)
∑︁(−d)+ j j
fact, write gR = j=0 gR 𝝏 + with gR j
∈ S 0 (Rn−1 ; Γ0G (R+ ; 𝔤𝔤−1 )). Then, using the
mapping property of generalized singular Green operators on R+ ,

j
q∞ #gR ∈ S 0 (Rn−1 ; L (H0s,δ (R+ , C L ) ⊕ C M1 , H s0 (R+ , C L ) ⊕ C M0 ))

for all s, δ ∈ R. Since

′ ′
gL ∈ S 0 (Rn−1 ; L (H s0 (R+ , C L ) ⊕ C M0 , H s ,δ (R+ , C L ) ⊕ C M0 ))

for all s′ , δ ′ ∈ R, it follows that gL #q∞ #gR

j
∈ S 0 (Rn−1 ; Γ0G (R+ ; 𝔤−1 )).
Altogether we have verified that q∞ ∈ S 0 (Rn−1 ; B −d;(−d)+ (R+ ; 𝔤−1 )). Using the
density of H ∞ (R+ , C L ) in H s (R+ , C L ) and the mapping properties of op(q∞ ) and
[d] [d]
op(σ∞ (p)), respectively, it follows that op(q∞ ) is the inverse of op(σ∞ (p)) in the first
[d]
version of (E3) for any choice of s, hence q∞ is the inverse of σ∞ (p) with respect to the
Leibniz product for any choice of s in the second version of (E3). □

Definition 5.15. Let d ∈ Z, ν ∈ N0 . A parametrix for p ∈ Bd,ν;d+ (Rn−1 × R+ ; 𝔤),

𝔤 = ((L, M0 ), (L, M1 )) is any symbol q ∈ B−d,ν;(−d)+ (Rn−1 × R+ ; 𝔤−1 ) such that

1 − q#p ∈ B −∞;d+ (Rn−1 × R+ ; 𝔤−1 𝔤),

1 − p#q ∈ B −∞;(−d)+ (Rn−1 × R+ ; 𝔤𝔤−1 ).

Theorem 5.16. Let p ∈ Bd,ν;d+ (Rn−1 × R+ ; 𝔤). Then p is elliptic if and only if p has a
parametrix. In this case there exists a μ0 such that op(p)(μ) is invertible for μ ≥ μ0 and
the parametrix q can be chosen in such a way that

op(q)(μ) = op(p)(μ)−1 ∀ μ ≥ μ0 .

The remaining part of this section is dedicated to the proof of Theorem 5.16. Using
the order reductions from Section 2.3.1 and an analogue of Remark 2.29 we may assume
without loss of generality that p ∈ B0,ν;0 (Rn−1 × R+ ; 𝔤), i.e., p is of order and type zero.
Correspondingly, the parametrix will also be of order and type zero.
(0)
Let p have the pseudodifferential part op+ (a). By the invertibility of σψ (p) we find
a symbol b ∈ Str0
(Rn × R+ ; L (C L )) such that (1 −(︃a#b)(x, ξ, μ) )︃ and (1 − b#a)(x, ξ, μ)
op+ (b) 0
are of order −∞ on R+ × R × R+ . Thus if q0 :=
n n
then
0 0
54 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

p#q0 ≡ 1 − gR , q0 #p ≡ 1 − gL , (5.5)

modulo regularizing symbols of type zero, where gL ∈ B0,ν;0

G (Rn−1 × R+ ; 𝔤−1 𝔤) and
(Rn−1 × R+ ; 𝔤𝔤−1 ). Applying σ∂
(0,ν)
gR ∈ B0,ν;0
G to these identities and resolving for
(0,ν) −1
σ∂ (p) we find

(p)−1 = σ∂ σ∂ (p)−1 gR .
(0,ν) (0,ν) (0,ν) (0,ν) (0,ν) (0,ν) (0,ν)
σ∂ (q0 ) + σ∂ (q0 )gR + gL (5.6)

Lemma 5.17. The third term on the right-hand side of (5.6) is the homogeneous principal
symbol of a generalized singular Green symbol from the class B0,ν;0
G (Rn−1 × R+ ; 𝔤−1 ).

Proof. Let p′ (x′ , ξ ′ , μ) = 𝜿−1 (p)(x′ , ξ ′ , μ)𝜿|ξ′ ,μ| . Then

(0,ν)
|ξ ′ ,μ| σ∂

p′ ∈ S0,ν
hom (R
n−1
× R+ ; L (H s (R+ , C L ) ⊕ C M0 ), H s (R+ , C L ) ⊕ C M1 )).

Due to ellipticity condition (E2), p′ satisfies assumption i) of Lemma 3.17. By definition,

the principal angular symbol of p′ coincides with the principal angular symbol of p. The
latter coincides with the homogeneous principal symbol of the limit symbol of p; by
ellipticity condition (E3), it is invertible and the inverse is the homogeneous principal
symbol of q∞ . Hence p′ also satisfies assumption ii) of Lemma 3.17. In conclusion

p′ −1 ∈ S0,ν
hom (R
n−1
× R+ ; L (H s (R+ , C L ) ⊕ C M1 ), H s (R+ , C L ) ⊕ C M0 ))

and p′ (x′ , ξ ′ , μ)−1 = 𝜿−1 (p)(x′ , ξ ′ , μ)−1 𝜿|ξ′ ,μ| . Conjugating gL

(0,ν) (0,ν) (0,ν)
|ξ ′ ,μ| σ∂ and gR
with 𝜿|ξ′ ,μ| yields symbols gL
′ (0,ν)
and gR
′ (0,ν) ˜︁
from Shom (R0,ν n−1
× R+ ; ΓG (R+ ; 𝔤−1 𝔤))
0

and S ˜︁ 0,ν n−1 −1

× R+ ; ΓG (R+ ; 𝔤𝔤 )), respectively. The third term on the right-hand
0
hom (R
′ (0,ν) ′ ′ (0,ν)
side of (5.6) conjugated with 𝜿|ξ′ ,μ| coincides with gL p gR , hence belongs to
˜︁ 0,ν (Rn−1 × R+ ; Γ0 (R+ ; 𝔤−1 )). This shows the claim. □
S hom G

By the previous lemma thus there exists a g ∈ B0,ν;0

G (Rn−1 × R+ ; 𝔤−1 ) with

σ∂ (p)−1 gR .
(0,ν) (0,ν) (0,ν) (0,ν) (0,ν)
g(0,ν) = σ∂ (q0 )gR + gL

Renaming q0 + g as q0 again, we thus find the identities (5.5) where, in addition,

(0,ν) (0,ν)
the homogeneous principal symbols gL and gR vanish, i.e., gL and gR belong to
−1,ν−1;0 −1 −1,ν−1;0
BG (Rn−1
× R+ ; 𝔤 𝔤) and BG (Rn−1
× R+ ; 𝔤𝔤−1 ), respectively
Sofar the reasoning is the same for any value ν ∈ N0 . However, from here on we must
distinguish the cases ν ≥ 1 and ν = 0.
−ℓ,ν−1;0
In case ν−1 ≥ 0 we have gR #ℓ
∈ BG (Rn−1 ×R+ ; 𝔤−1 𝔤) for every ℓ. By asymptotic
−1,ν−1;0
summation there exists g0 ∈ BG (Rn−1
× R+ ; 𝔤−1 𝔤) such that, for every N ∈ N,

∑︂
N −1
g0 − #ℓ
gR ∈ B−N,ν−1;0
G (Rn−1 × R+ ; 𝔤−1 𝔤).
ℓ=1
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 55

Then q := q0 + q0 #g0 ∈ B0,ν;0 (Rn−1 × R+ ; 𝔤−1 ) satisfies p#q ≡ 1 − gR and q#p ≡

1 − gL modulo regularizing symbols of type zero, where

gL ∈ ∩ B−1−N,ν−1;0
G (Rn−1 × R+ ; 𝔤𝔤−1 ) ⊂ B −∞;0 (Rn−1 × R+ ; 𝔤𝔤−1 ),
N ∈N

and analogously for gR . Thus we have found

rR := 1 − p#q ∈ B −∞;0 (Rn−1 × R+ ; 𝔤−1 𝔤),

(5.7)
rL := 1 − q#p ∈ B −∞;0 (Rn−1 × R+ ; 𝔤𝔤−1 ).

Also in case ν = 0, which we will consider now, we shall construct a parametrix

q satisfying (5.7); however, the reasoning is more complex and makes use of the
principal limit symbol. Let us return to the above point where we have constructed
q0 ∈ B0,ν;0 (Rn−1 × R+ ; 𝔤−1 ) such that

p#q0 ≡ 1 − gR , q0 #p ≡ 1 − gL
−1,−1;0
modulo regularizing symbols of type zero, where gL ∈ BG (Rn−1 × R+ ; 𝔤𝔤−1 ) =
˜︁
BG −1,−1;0
(Rn−1 −1
× R+ ; 𝔤𝔤 ) and analogously gR ∈ BG ˜︁ −1,−1;0
(Rn−1
× R+ ; 𝔤−1 𝔤). Noting
#ℓ
that gR ∈ B ˜︁ −ℓ,−ℓ;0
(Rn−1 × R+ ; 𝔤−1 𝔤), by asymptotic summation we can construct
G
˜︁
g0 ∈ BG −1,−1;0
(R n−1
× R+ ; 𝔤−1 𝔤) such that q1 := q0 + q0 #g0 satisfies

p#q1 ≡ 1 − gR , q1 #p ≡ 1 − gL ,

modulo regularizing symbols of type zero where

˜︁ 0−∞,0−∞;0 (Rn−1 × R+ ; 𝔤𝔤−1 ),

gL ∈ B ˜︁ 0−∞,0−∞;0 (Rn−1 × R+ ; 𝔤−1 𝔤).
gR ∈ B
G G

[0]
Now we apply σ∞ to these identities and resolve for q∞ (see ellipticity condition (E3));
this yields

[0] [0] ∞ ∞ ∞
q∞ = σ∞ (q1 ) + σ∞ (q1 )#gR,[0,0] + gL,[0,0] #q∞ #gR,[0,0] .

∞ ∞
Note that gL/R,[0,0] = gL/R,[−N,−N ] for every N ∈ N and therefore

∞
gL,[0,0]) ∈ S −∞ (Rn−1 ; Γ0G (R+ ; 𝔤𝔤−1 )), ∞
gR,[0,0]) ∈ S −∞ (Rn−1 ; Γ0G (R+ ; 𝔤−1 𝔤)).

It follows that

g1′ := σ∞
[0] ∞
(q1 )#gR,[0,0] ∞
+ gL,[0,0] ∞
#q∞ #gR,[0,0] ∈ S −∞ (Rn−1 ; Γ0G (R+ ; 𝔤−1 )).

Since

˜︁ 0−∞,0−∞ (Rn−1 × R+ ; Γ0 (R+ ; 𝔤−1 )),

S −∞ (Rn−1 ; Γ0G (R+ ; 𝔤−1 )) ⊂ S G
56 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

g1 := 𝜿g1′ 𝜿−1 lies in B ˜︁ 0−∞,0−∞;0 (Rn−1 × R+ ; 𝔤−1 ) and has principal limit symbol
G
∞
g1,[0,0] = g1′ . Hence q2 := q1 + q1 #g1 satisfies

p#q2 ≡ 1 − gR , q2 #p ≡ 1 − gL ,

modulo regularizing symbols of type zero where

˜︁ 0−∞,0−∞;0 (Rn−1 × R+ ; 𝔤𝔤−1 ),

gL ∈ B ˜︁ 0−∞,0−∞;0 (Rn−1 × R+ ; 𝔤−1 𝔤)
gR ∈ B
G G

∞
have both vanishing principal limit symbol, i.e., gL/R,[0,0] = 0. By the following Propo­
˜︁
sition 5.18 there exists g2 ∈ B 0−∞,0−∞;0
(Rn−1 −1
× R+ ; 𝔤 𝔤) such that q := q2 + q2 #g2
G
satisfies (5.7).

Proposition 5.18. Let g ∈ B ˜︁ 0−∞,0−∞;0 (Rn−1 × R+ ; 𝔤), 𝔤 = ((L, M ), (L, M )), have van­
G
ishing principal limit symbol. Then 1 + op(g)(μ) is invertible for large μ ∈ R+ and there
˜︁ 0−∞,0−∞;0 (Rn−1 × R+ ; 𝔤−1 ) such that, for these μ,
exists an h ∈ B G

(1 − op(g))(μ)−1 = (1 − op(h))(μ).

Proof. By Theorem 4.8 and Corollary 4.13,

˜︁ 0−∞,0−∞ (Rn−1 × R+ ; Γ0 (R+ , 𝔤))

g′ := 𝜿−1 #g#𝜿 ∈ S G

has vanishing principal limit symbol. By Proposition 3.6 this is equivalent to

g′ ∈ S −1 (R+ ; S −∞ (Rn−1 ; Γ0G (R+ , 𝔤))).

Then g′ #ℓ ∈ S −ℓ (R+ ; S −∞ (Rn−1 × R+ ; Γ0G (R+ , 𝔤))) and, by asymptotic summation,

there exists h0′ ∈ S˜︁−1 (R+ ; S −∞ (Rn−1 ; Γ0G (R+ , 𝔤))) such that

∑︂
N −1
h0′ − g′ #ℓ ∈ S −1−N (R+ ; S −∞ (Rn−1 ; Γ0G (R+ , 𝔤)))
ℓ=1

for every N ∈ N0 . Therefore,

(1 − h0′ )#(1 − g′ ) = 1 − r′L , (1 − g′ )#(1 − h0′ ) = 1 − r′R

with r′L , r′R ∈ S˜︁−∞ (R+ ; S −∞ (Rn−1 ; Γ0G (R+ , 𝔤))).

Next we note that by (a slight extension of) a classical result of Beals [3,4] on the
characterization of pseudodifferential operators via mapping properties of certain com­
mutators (for a proof see, for instance, [30, Corollary A.3]),

𝒜 := {op(p) | p(x′ , ξ ′ ) ∈ Cb∞ (Rn−1 × Rn−1 ; L (E))}, E := L2 (R+ , C L ) ⊕ C M ,

 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 57

is spectral-invariant in L (L2 (Rn−1 , E)), i.e., an element of 𝒜 has an inverse in 𝒜 if

and only if it is invertible as an operator in L (L2 (Rn−1 , E)). Moreover, inversion is a
continuous operation in 𝒜. From this it follows that there exists a p′ (μ) ∈ Cb∞ (R+ ; 𝒜)
such that op(p′ )(μ) = χ(μ)(1 − op(r′L ))(μ)−1 for a suitable zero-excision function χ.
Then

(1 − op(r′L )(μ)−1 = 1 + op(r′L )(μ) + op(r′L )(μ)(1 − op(r′L )(μ)−1 op(r′L )(μ)
= 1 + op(r′L + r′L #p′ #r′L )(μ)

shows the existence of s′L ∈ S −∞ (R+ ; S −∞ (Rn−1 ; Γ0G (R+ , 𝔤))) such that
′ −1 ′
(1 − op(rL ))(μ) = 1 + op(sL )(μ) for large μ. For the latter note that

(r0 , p, r1 ) ↦→ r0 pr1 : Γ0G (R+ ; 𝔤) × L (E) × Γ0G (R+ ; 𝔤) −→ Γ0G (R+ ; 𝔤)

is continuous. Analogously we find s′R for 1 − r′R .

Thus if h′ := h0′ − s′L + s′L #h0′ , then (1 − op(g′ ))(μ)−1 = 1 − op(h′ )(μ) for large
enough μ. Then the claim follows with h := 𝜿#h′ #𝜿−1 . □

Summing up, we have constructed q with (5.7) in both cases ν ≥ 1 and ν = 0. Then
the following lemma completes the proof of Theorem 5.16.

Lemma 5.19. Let r ∈ B −∞;0 (Rn−1 × R+ ; 𝔤), 𝔤 = ((L, M ), (L, M )). Then 1 − op(r)(μ) is
invertible for large μ ∈ R+ and there exists an s ∈ B −∞;0 (Rn−1 × R+ ; 𝔤) such that, for
these μ,

(1 − op(r))(μ)−1 = (1 − op(s))(μ).

−∞;0
Proof. If r = g ∈ BG (Rn−1 × R+ ; 𝔤) the proof is a simpler version of that of Propo­
sition 5.18; we leave the details to the reader.
Let op+ (a)(μ) be the pseudodifferential part of r, where a ∈ S −∞ (Rn × R+ ; L (C L )).
Then there exists a b ∈ S −∞ (Rn × R+ ; L (C L )) such that

(1 − op(a)(μ))−1 = (1 − op(b)(μ)

for large enough μ as pseudodifferential operators on Rn . But then

1 − op+ (1 − a)#op+ (1 − b) = op+ ((1 − a)#(1 − b)) − op+ (1 − a)#op+ (1 − b)

= op+ (a#b) − op+ (a)#op+ (b) =: g

−∞;0
with g ∈ BG (Rn−1 × R+ ; (L, 0), (L, 0)). This yields
(︃ )︃
op+ (1 − b) 0
(1 − r)# =1+g
0 1
58 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

−∞;0
with g ∈ BG (Rn−1 × R+ ; 𝔤). Now we are in the situation from the beginning of the
proof. □

6. Operators of Toeplitz type

6.1. An abstract framework for parameter-dependent operators

In this section we present some results of [28] that we shall apply below to boundary
value problems. For convenience of the reader we adapt the formalism to better fit with
the current paper and include a self-contained proof of Theorem 6.3.
We begin with a number of definitions. Let

𝔊 = {𝔤 = (g0 , g1 ) | g0 , g1 ∈ G}

be a set of ``weight-data''. The inverse weight-datum of 𝔤 = (g0 , g1 ) is 𝔤−1 = (g1 , g0 ), for

two weight-data 𝔤j = (gj , gj+1 ), j = 0, 1, let 𝔤1 𝔤0 = (g0 , g2 ). Assume that with every g ∈
G is associated a Hilbert space H(g) and with every 𝔤 = (g0 , g1 ) ∈ 𝔊 there are associated
two vector-spaces of parameter-dependent operators L0 (R+ ; 𝔤) and L−∞ (R+ ; 𝔤), the first
containing the second, such that:

(1) If A ∈ L0 (R+ ; 𝔤) then A(μ) ∈ L (H(g0 ), H(g1 )) for every μ ∈ R+ ,

(2) Taking (μ-wise) the adjoint of operators induces maps

Ld (R+ ; 𝔤) −→ Ld (R+ ; 𝔤−1 ), d ∈ {0, −∞},

(3) (μ-wise) composition of operators induces maps

Ld1 (R+ ; 𝔤1 ) × Ld0 (R+ ; 𝔤0 ) −→ Ld0 +d1 (R+ ; 𝔤1 𝔤0 )

whenever 𝔤1 𝔤0 is defined and d0 , d1 ∈ {0, −∞}.

Writing simply 1 for the identity map in each H(g), we require that 1 ∈ L0 (R+ ; (g, g))
for every g ∈ G.

Definition 6.1. A parametrix of A ∈ L0 (R+ ; 𝔤) is any B ∈ L0 (R+ ; 𝔤−1 ) satisfying

1 − AB ∈ L−∞ (R+ ; 𝔤𝔤−1 ), 1 − BA ∈ L−∞ (R+ ; 𝔤−1 𝔤).

We assume the existence of a principal symbol map, i.e., with every A ∈ L0 (R+ ; 𝔤)
with arbitrary weight 𝔤 = (g0 , g1 ) is associated a tuple

σ(A) = (σ1 (A), . . . , σN (A)) (N some fixed integer)

 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 59

consisting of bundle homomorphisms σk (A) : Ek (g0 ) −→ Ek (g1 ), where Ek (g) represents

a Hilbert space bundle (finite or infinite-dimensional) associated with g ∈ G. We require
that the principal symbols vanish for operators of order −∞, that σk (1) = 1 for every
k, and

(4) each principal symbol σk is compatible with the addition, composition, and adjoint,
i.e. σk (A + B) = σk (A) + σk (B) whenever A + B is defined, σk (AB) = σk (A)σk (B)
whenever AB is defined, and σk (A∗ ) = σk (A)∗ for every A.
(5) A ∈ L0 (R+ ; 𝔤) possesses a parametrix if and only if σk (A) is a bundle isomorphism
for every k = 1, . . . , N .

6.1.1. Ellipticity and parametrices

We now consider Toeplitz type operators associated with two projections. Given 𝔤 =
(g0 , g1 ) and two projections Πj ∈ L0 (R+ ; 𝔤j ), 𝔤j = (gj , gj ), j = 0, 1, let us set, for
d ∈ {0, −∞},
{︁ }︁
Ld (R+ ; 𝔤, Π0 , Π1 ) = A ∈ Ld (R+ ; 𝔤) | A(1 − Π0 ) = (1 − Π1 )A = 0

Definition 6.2. Given A ∈ L0 (R+ ; 𝔤, Π0 , Π1 ), any B ∈ L0 (R+ ; 𝔤−1 , Π1 , Π0 ) satisfying

Π1 − AB ∈ L−∞ (R+ ; 𝔤𝔤−1 , Π1 , Π1 ),

Π0 − BA ∈ L−∞ (R+ ; 𝔤−1 𝔤, Π0 , Π0 )

is called a parametrix of A.

The following theorem shows how ellipticity ``descents'' to the class of Toeplitz type
operators:

Theorem 6.3. For A ∈ L0 (R+ ; 𝔤, Π0 , Π1 ), 𝔤 = (g0 , g1 ), the following are equivalent:

a) A possesses a parametrix in the sense of Definition 6.2.

b) A is elliptic, i.e., for every k = 1, . . . , N the following map is bijective:
(︁ )︁ (︁ )︁
σk (A) : σk (Π0 ) Ek (g0 ) −→ σk (Π1 ) Ek (g1 ) .

Since, by assumption, A = Π1 AΠ0 and every principal symbol is multiplicative un­

der composition, σk (A) = σk (Π1 )σk (A)σk (Π0 ), i.e., the requirement in b) is (only) the
bijectivity.

Proof. It is clear that a) implies b). So let us start out from b).
Step 1: Let a ∈ L (H0 , H1 ) and let πj ∈ L (Hj ) be projections such that a(1 − π0 ) =
(1 − π1 )a = 0 and a : im π0 → im π1 bijectively. Define a0 = a∗ a + (1 − π0 )∗ (1 − π0 ). Then
60 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

a0 ∈ L (H0 ) is invertible. In fact, define t ∈ L (H0 , H1 ⊕ H0 ) by tx = (ax, (1 − π0 )x).

Then t is injective and, since t(π0 x + (1 − π0 )y) = (aπ0 x, (1 − π0 )y) for all x, y ∈ H0 ,
im t = im π1 ⊕ im (1 − π0 ) is closed in H1 ⊕ H0 . Hence t is upper semi-fredholm and
therefore a0 = t∗ t is a self-adjoint, injective Fredholm operator, i.e., an isomorphism.
Step 2: Let 𝔤0 = 𝔤−1 𝔤; then

A0 (μ) := A(μ)∗ A(μ) + (1 − Π0 )(μ)∗ (1 − Π0 )(μ) ∈ L0 (R+ ; 𝔤0 )

and

σk (A0 ) = σk (A)∗ σk (A) + σk (1 − Π0 )∗ σk (1 − Π0 ).

Applying the first step in each fibre of Ek (g0 ) shows that σk (A0 ) : Ek (g0 ) → Ek (g0 ) is an
isomorphism. Hence A0 admits a parametrix B0 ∈ L0 (R+ ; 𝔤0 ). Then BL := Π0 B0 A∗ Π1
belongs to L0 (R+ ; 𝔤−1 , Π1 , Π0 ) and

BL (μ)A(μ) = Π0 (μ)B0 (μ)A(μ)∗ A(μ)Π0 (μ) = Π0 (μ)B0 (μ)A0 (μ)Π0 (μ) ≡ Π0 (μ)

modulo L−∞ (R+ ; 𝔤0 , Π0 , Π0 ). Hence BL is a parametrix from the left.

Step 3: Let the notation be as in Step 1. Then a∗ : im π1∗ → im π0∗ is bijective. In fact,
consider im πj as Hilbert spaces (with the inner product induced from Hj ) and denote
by ˆ︁
a the map a : im π0 → im π1 , by π ˆ︁1 the map π1 : H1 → im π1 , and by ι0 the inclusion
−1
map im π0 → H0 . Let b = ι0 ˆ︁ a π ˆ︁1 : H1 → H0 . Then ba vanishes on ker π0 and on im π0
it coincides with ι0 ˆ︁a−1 ˆ︁
a = ι0 . Thus ba = π0 . Similarly, ab = π1 . Hence a∗ b∗ = π0∗ and
b∗ a∗ = π1∗ . It follows that b∗ : im π0∗ → im π1∗ is the inverse of a∗ : im π1∗ → im π0∗ .
Step 4: Applying Step 3 in each fibre to σk (A)∗ = σk (A∗ ) we see that A∗ is an elliptic
element of L0 (R+ ; 𝔤−1 , Π∗1 , Π∗0 ). By Step 2, A∗ has a parametrix from the left. Taking
the adjoint yields BR ∈ L0 (R+ ; 𝔤, Π0 , Π1 ) which is a parametrix for A from the right, i.e.
ABR ≡ Π1 modulo a remainder of order −∞. But then BL ≡ BL (ABR ) = (BL A)BR ≡
BR shows that both BL and BR are (two-sided) parametrices for A. □

6.1.2. Invertibility for large values of the parameter

In order to obtain parametrices that are inverses (in the sense explained below) for
large values of μ we need to require:

(6) Whenever R ∈ L−∞ (R+ ; 𝔤) with 𝔤 = (g, g), there exists a μ0 ≥ 0 and an S ∈
L−∞ (R+ ; 𝔤) such that, for all μ ≥ μ0 ,

(1 − R(μ))(1 − S(μ)) = (1 − S(μ))(1 − R(μ)) = 1.

Lemma 6.4. Let R ∈ L−∞ (R+ ; 𝔤, Π, Π), 𝔤 = (g, g). Then there exists an S ∈
L−∞ (R+ ; 𝔤, Π, Π) such that
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 61

(Π(μ) − R(μ))(Π(μ) − S(μ)) = (Π(μ) − S(μ))(Π(μ) − R(μ)) = Π(μ)

for all sufficiently large μ.

Proof. First, by (6), there exists an S˜︁ ∈ L−∞ (R+ ; 𝔤) such that (1 − R(μ))(1 − S(μ))
˜︁ =
˜︁
(1 − S(μ))(1 − R(μ)) = 1 for large μ. Multiplying with Π(μ) from the left and the right,
we see that S := ΠSΠ˜︁ is as desired. □

As an immediate consequence of this lemma and Theorem 6.3 we get:

Theorem 6.5. Assume (6). Then for every elliptic A ∈ L0 (R+ ; 𝔤, Π0 , Π1 ) exists a
parametrix B ∈ L0 (R+ ; 𝔤−1 , Π1 , Π0 ) such that, for all sufficiently large μ,

A(μ)B(μ) = Π1 (μ), B(μ)A(μ) = Π0 (μ)

Obviously, using the notation of Theorem 6.5, the map

(︁ )︁ (︁ )︁
A(μ) : Π0 (μ) H(g0 ) −→ Π1 (μ) H(g1 )

is an isomorphism whenever μ is sufficiently large. The inverse is induced by B(μ).

7. Applications to boundary value problems

We will apply the above results on Toeplitz type operators to boundary value problems

7.1. Reformulation of the ellipticity conditions

In the abstract setting of Section 6, ellipticity was required to be characterized by

the (sole) invertibility of certain principal symbols which were assumed to be bundle
homomorphisms in certain Hilbert space bundles. Note that condition (E3) is of such
a kind while, at a first glance, conditions (E1) and (E2) are not, since they contain
estimates on the inverse (which are necessary in view of the non-compactness of Rn+).
However, there is a simple way to reformulate these conditions appropriately, based on
˜︁ the following are equivalent:
the following observation. For a ∈ Cb∞ (Ω, L (E, E))

(1) a(x) is invertible for every x ∈ Ω and ∥a(x)−1 ∥L (E,E)

˜︁ ≲ 1 uniformly in x ∈ Ω.
−1 ∞
(2) a(x) is invertible for every x ∈ Ω and x ↦→ a(x) ∈ Cb (Ω, L (E, ˜︁ E)).
˜︁
(3) The operator of multiplication f ↦→ af is invertible in L (L (Ω, E), L2 (Ω, E)).
2

Let us take this point of view to reformulate ellipticity condition (E1). We consider
(d)
σψ (p) as a bundle homomorphism

(d) n
σψ (p) : S+ × L2 (Rn+ , C L ) −→ S+
n
× L2 (Rn+ , C L ) (7.1)
62 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

(d)
which acts in the fibre over (ξ, μ) as operator of multiplication by σψ (p)(·, ξ, μ). Then
condition (E1) is equivalent to

(E1′ ) The bundle homomorphism (7.1) is an isomorphism.

Since the involved bundle is trivial, this is of course the same as requiring

(E1′ ) Whenever |ξ ′ , μ| = 1, the following map is an isomorphism:

σψ (p)(·, ξ ′ , μ) : L2 (Rn+ , C L ) −→ L2 (Rn+ , C L ).

(d)

(d,ν)
Analogously, the boundary symbol σ∂ (p) will be identified with a bundle homomor­
phism

σ∂
(d,ν) ˆ︁ n−1 ×L2 (Rn−1 , H s (R+ , C L ) ⊕ C M0 )
(p) : S +
(7.2)
ˆ︁ n−1 × L2 (Rn−1 , H s−d (R+ , C L ) ⊕ C M1 )
−→ S +

which acts in the fibre over (ξ ′ , μ) as operator of multiplication by σ∂ (p)(·, ξ ′ , μ).

(d,ν)

Then conditions (E2) and (E3) together are equivalent to

(E2′ ) The bundle hommorphism (7.2) is an isomorphism for some (and then for all)
s > d+ − 12 .
′
(E3 ) For some (and then for all) s > d+ − 12 , the following operator is invertible:

L2 (Rn−1 , H s (R+ , C L )) L2 (Rn−1 , H s−d (R+ , C L ))

[d]
op(σ∞ (p)) : ⊕ −→ ⊕ .
L2 (Rn−1 , C M0 ) L2 (Rn−1 , C M1 )

(For (E3′ ) recall the comment given after Definition 5.13). Again due to the triviality
of the involved bundles we can write equivalently

(E2′ ) Whenever |ξ ′ , μ| = 1, the following map is an isomorphism for some (and then for
all) s > d+ − 12 :

L2 (Rn−1 , H s (R+ , C L )) L2 (Rn−1 , H s−d (R+ , C L ))

σ∂ (p)(·, ξ ′ , μ)
(d,ν)
: ⊕ −→ ⊕ .
L2 (Rn−1 , C M0 ) L2 (Rn−1 , C M1 )

Let be given weight-data 𝔤 = ((L, M0 ), (L, M1 )) and 𝔤j = ((L, Mj ), (L, Mj )) for

j = 0, 1. Suppose

πj (x′ , ξ ′ , μ) ∈ B0,ν;0 (Rn−1 × R+ ; 𝔤j )

 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 63

(0)
are two projections, i.e. πj #πj = πj , where ν ∈ N0 . Since πj is a projection, σψ (πj ),
(0,ν) [0] (0)
σ∂ (πj ), and σ∞ (πj ) are projections, too. Recall that we consider σψ (πj ) as a bun­
(0,ν) [0]
dle morphism as in (7.1), σ∂ (πj ) as a bundle morphism as in (7.2), and σ∞ (πj ) ∈
S 0 (Rn−1 ; L (H s (R+ , C Lj ) ⊕ C Mj )) for s > −1/2. In particular,

(0)
im σψ (πj ) ⊂ S+
n
× L2 (Rn+ , C L ),

ims σ∂
(0,ν) ˆ︁ n−1 × L2 (Rn−1 , H s (R+ , C L ) ⊕ C Mj ),
(πj ) ⊂ S +

ims op(σ∞
[0]
(πj )) ⊂ L2 (Rn−1 , L (H s (R+ , C L ) ⊕ C M1 );

here we added the superscript s in ims to indicate the involved Sobolev-regularity.

Definition 7.1. With the above notation and d ∈ Z, ν ∈ N0 , let us define

Bd,ν;d+ (Rn−1 × R+ ; 𝔤, π0 , π1 )
= {p ∈ Bd,ν;d+ (Rn−1 × R+ ; 𝔤) | (1 − π1 )#p = p#(1 − π0 ) = 0}.

Theorem 7.2. Let p ∈ Bd,ν;d+ (Rn−1 × R+ ; 𝔤, π0 , π1 ) with d ∈ Z, ν ∈ N0 . Then there

exists a symbol q ∈ B−d,ν;(−d)+ (Rn−1 × R+ ; 𝔤−1 , π1 , π0 ) such that

op(q)(μ)op(p)(μ) = op(π0 )(μ), op(p)(μ)op(q)(μ) = op(π1 )(μ),

for sufficiently large μ if and only if the following conditions hold true:

(d) (0) (0)

(Π1) σψ (p) of (7.1) restricts to an isomorphism im σψ (π0 ) → im σψ (π1 ),
(d,ν) (0,ν) (0,ν)
(Π2) σ∂ (p) of (7.2) restricts to an isomorphism ims σ∂ (π0 ) → ims−d σ∂ (π1 ),
[d] [0] ([0]
(Π3) op(σ∞ (p)) restricts to an isomorphism ims σ∞ (π0 ) → ims−d σ∞ (π1 )

for some s > d+ − 12 .

Proof. The case d = 0 is an immediate consequence of the results from Section 6.1.1.
The general case can be reduced to it by the use of the order reductions from (2.12). Let
us show this in case d ≥ 1 (the case d ≤ −1 is verified analogously).
˜︁ = p#𝝀−d
Consider p − . Then

˜︁ = π1 #p#𝝀−d
p d −d
− #𝝀− #π0 #𝝀− = π1 #˜︁
p#˜︁
π0

with the projection

˜︁0 = 𝝀d− #π0 #𝝀−d

π − ∈B
0,ν;0
(Rn−1 × R+ ; 𝔤0 ).
64 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

This shows that p ˜︁ ∈ B0,ν;0 (Rn−1 × R+ ; 𝔤, π

˜︁0 , π1 ). Since op(𝝀d− ) is invertible with inverse
−d
op(𝝀− ) and by the multiplicativity under composition of all principal symbols, all prin­
cipal symbols associated with 𝝀±d − are isomorphisms and it is immediate by construction
that p satisfies (Π1) − (Π3) if and only if p ˜︁ does (with d = 0 and with respect to the
˜︁0 and π1 ). Therefore we can apply the theorem with d = 0 to p,
projetcions π ˜︁ yielding a
corresponding symbol q. ˜︁ Then the claim follows with q := 𝝀−d − #˜︁
q. □

7.2. BVP with global projection conditions

Let A be a differential operator of order d and let

∑︂
A(μ) = eiθ μd − A = eiθ μd − aα (x)Dxα , μ ≥ 0,
|α|≤d

for some fixed 0 ≤ θ < 2π and where aα ∈ Cb∞ (Rn , C L×L ). We will apply Theorem 7.2
to analyze the invertibility of the problem

(︃ )︃ H s−d (Rn+ , C L )
A(μ) ⊕ 1
: H s (Rn+ , C L ) −→ (︂ d−1 )︂, s>d− , (7.3)
T 1 2
Π ⊕ H s−j− 2 (Rn−1 , C mj )
j=0

where T = ΠS is a so-called global projection boundary condition as described in the

following Section 7.2.1.

7.2.1. Global projection boundary conditions

For j ∈ N0 let us define

∂j u
γj : S (R+ , C L ) −→ C L , u ↦→ (0)
∂xjn

(we shall use the same notation for different values of L). Then γj is a strongly parameter­
dependent trace symbol (constant in (x′ , ξ ′ , μ)) of order j + 12 and of type j + 1. In fact,
the symbol-kernels

g0 (ξ ′ , μ; xn ) = [ξ ′ , μ] exp(−[ξ ′ , μ]yn ), g1 (ξ ′ , μ; yn ) = exp(−[ξ ′ , μ]yn )

1/2−k;0
define singular Green symbols gk ∈ BG (Rn−1 × R+ ; (L, 0), (0, L)) and, by integra­
tion by parts,

γ0 = g0 (ξ ′ , μ) − g1 (ξ ′ , μ)∂+ .

j
For j ≥ 1 note that γj u = γ0 (∂+ u). Note that the principal symbol and the principal
limit symbol are
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 65

σ (j+1/2) (γj ) = σ∞
[j+1/2]
(γj ) = γj .

Obviously,

∂j u
[op(γj )u](x′ ) = (x′ , 0), u ∈ S (Rn+ , C L );
∂xjn

By abuse of notation (and to obtain standard notation from the literature) we shall write
γj instead of op(γj ). If sjk (x′ , ξ ′ ) ∈ S j−k (Rn−1 ; L (C L , C mj )), j, k ∈ N0 , with mj ∈ N0
are pseudodifferential symbols then

∑︂
j
′ ′
sj (x , ξ ) = sjk (x′ , ξ ′ )γk : S (R+ , C L ) −→ C mj (7.4)
k=0

defines a trace symbol of order j + 1/2, type j + 1, and regularity number 0, i.e.,

j+1/2,0;j+1
sj ∈ BG (Rn−1 × R+ ; (L, 0), (0, mj )), j ∈ N0

(note that if in (7.4) the summation index k would exceed j then sj would have negative
regularity number). Writing Sjk := op(sjk ) and Sj = op(sj ) we thus have

⎛ ⎞ ⎛ ⎞⎛ ⎞
S0 S00 0 ··· 0 γ0
⎜ .. ⎟ ⎜ .. .. .. .. ⎟⎜ .. ⎟
⎜ . ⎟ ⎜ . . . . ⎟⎜ . ⎟
⎜ . ⎟=⎜ . ⎟⎜ ⎟.
⎝ . ⎠ ⎝ . ..
. ⎠⎝ .. ⎠
. . 0 .
Sd−1 Sd−1,0 ··· ··· Sd−1,d−1 γd−1

Definition 7.3. Using the previous notation and writing S = (S0 , . . . , Sd−1 )t , a global
projection boundary condition has the form T = ΠS with an idempotent Π =
(Πjk )0≤j,k≤d−1 , where Πjk = 0 whenever k > j and, otherwise, Πjk = op(πjk ) with
πjk ∈ S j−k (Rn−1 ; L (C ℓk , C mj )).

The assumption that Π is a left lower triangular matrix ensures that the com­
ponents of Π have non-negative regularity number. More explicitly, T has the form
T = (T0 , . . . , Td−1 )t with

∑︂
j ∑︂
j ∑︂
k
Tj = Πjk Sk = op(πjk )op(skℓ )γℓ ;
k=0 k=0 ℓ=0

j+1/2,0;j+1
note that Tj ∈ BG (Rn−1 × R+ ; (L, 0), (0, mj )). T has the mapping property
from (7.3). Let us remark that we admit also the case that some of the mj are equal to
zero; in this case the condition Tj is considered ``void'' respectively ``not present''.
66 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

Remark 7.4. Let us compare the structure of the operator in (7.3) with that considered
by Grubb in [6]. On the one hand the condition T in (7.3) is more general, since the
projection Π and all terms Sjk can be pseudodifferential operators, while in [6] one has
Π = 1 and all Sjj are operators of multiplication with a function sjj (x′ ), cf. [6, (1.5.28)].
On the other hand, each condition Tk in [6, (1.5.28)] contains a trace operator Tk′ which
we have to require to be either equal to zero or to be parameter-dependent. Similarly,
A(μ) in [6, (1.5.19)] is permitted to contain a parameter-independent singular Green
operator G which we have to require either to be equal to zero or to be parameter­
dependent.

7.2.2. Invertibility of realizations

In order to apply Theorem 7.2 we need to transform the problem in an equivalent
form. To this end let

Λj (μ) = [D′ , μ]d−j− 2

1
Λ(μ) = diag(Λ0 (μ), . . . , Λd−1 (μ)),

and then consider

(︃ )︃ (︃ )︃ (︃ )︃ (︃ )︃
A(μ) A(μ) 1 0 A(μ)
= = ,
Λ(μ)T Λ(μ)ΠS 0 ΠΛ (μ) Λ(μ)ΠS

with

ΠΛ (μ) = Λ(μ)ΠΛ(μ)−1 = op(πΛ )(μ), πΛ ∈ S 0,0 (Rn−1 × R+ ; L (C M )),

where we use the identification

C M = C m0 ⊕ . . . ⊕ C md−1 , M = m0 + . . . + md−1 .

Therefore
(︃ )︃
A(μ)
= op(p)(μ), p ∈ Bd,0;d (Rn−1 × R+ ; (L, 0), (L, M ), 1, πΛ ). (7.5)
Λ(μ)T

By construction of p, its homogeneous principal symbol is

(d)
∑︂
σψ (p)(x, ξ, μ) = eiθ μd − aα (x)ξ α ,
|α|=d

its principal boundary symbol is

(︄ )︄
σ∂ (A)(x′ , ξ ′ , μ)
(d,0)
σ∂ (p)(x′ , ξ ′ , μ)
(d,0)
=
σ∂ (ΛT )(x′ , ξ ′ , μ)
(d,0)

where
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 67

∑︂ ′
(A)(x′ , ξ ′ , μ) = eiθ μd − aα (x′ , 0)ξ ′ α Dxαnn
(d,0)
σ∂
|α|=d

and
⎛ ⎞
|ξ ′ , μ|d− 2 π00 (x′ , ξ ′ )s0 (x′ , ξ ′ )γ0
1 (0) (0)
⎜ .. ⎟
⎜ ⎟
⎜ . ⎟
⎜ ⎟
⎜ ′ d−j− 12 ∑︁ j ∑︁ k
(j−k) ′ ′ (k−ℓ) ′ ′ ⎟
⎜ |ξ , μ| π (x , ξ )s (x , ξ )γ ℓ⎟
σ∂ (ΛT )(x′ , ξ ′ , μ) = ⎜
(d,0) jk kℓ ⎟.
⎜ k=0 ℓ=0 ⎟
⎜ .. ⎟
⎜ . ⎟
⎜ ⎟
⎝ ′ 1− 1 d−1 ∑︁ ∑︁k ⎠
πjk (x′ , ξ ′ )skℓ (x′ , ξ ′ )γℓ
(j−k) (k−ℓ)
|ξ , μ| 2
k=0 ℓ=0

If we introduce the ``mixed'' or ``Douglis-Nirenberg-type'' homogeneous principal symbols

⎛ ⎞
(0)
π
⎜ 00 . ⎟ ′ ′
′ ′ ⎜
σhom (Π)(x , ξ ) = ⎝ .. .. ⎟ (x , ξ ),
. ⎠
(d−1) (0)
πd−1,0 · · · πd−1,d−1
⎛ ⎞
(0)
s00
⎜ .. ⎟ ′ ′
σhom (S)(x′ , ξ ′ ) = ⎜
⎝ .
..
. ⎟ (x , ξ )
⎠
(d−1) (0)
sd−1,0 · · · sd−1,d−1

and note that the factors |ξ ′ , μ|d−j− 2 are equal to 1 whenever |ξ ′ , μ| = 1, we have
1

⎞ ⎛
γ0
σ∂ (ΛT )(x′ , ξ ′ , μ) = σhom (Π)(x′ , ξ ′ )σhom (S)(x′ , ξ ′ ) ⎝ ... ⎠ , |ξ ′ , μ| = 1.
(d,0)

γd−1

Finally, the principal limit symbol of p is

⎛ ⎞
eiθ − a0,d (x′ , 0)Dxdn
′ ′
⎜ (π00 #s00 )(x , ξ )γ0 ⎟
[d]
σ∞ (p)(x′ , ξ ′ ) = ⎜
⎝ .. ⎟.
⎠
.
′ ′
(πd−1,d−1 #sd−1,d−1 )(x , ξ )γd−1

For the latter identity note that Λj (μ)op(πjk )op(skℓ )γℓ is a trace operator with symbol
belonging to B ˜︁ d,j−ℓ;ℓ+1 (Rn−1 × R+ ; (L, 0), (0, mj )), hence its principal limit symbol σ∞
[d]
G
is zero unless ℓ = j; hence the principal limit-symbol of Λj (μ)Tj coincides with that
of Λj (μ)op(πjj )op(sjj )γj = Λj (μ)op(πjj #sjj )γj which is the one appearing above, due
the multiplicativity of the principal limit symbol under composition. Moreover recall
that the principal limit symbol of A(μ) coincides with its boundary symbol evaluated in
(ξ ′ , μ) = (0, 1).
68 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

Theorem 7.5. Let p from (7.5) satisfy (Π1) − (Π3) of Theorem 7.2 (with π0 = 1 and
π1 = (πjk )). Then there exist a pseudodifferential operator

−d
B+ (μ) = op+ (b)(μ), b ∈ Str (Rn × R+ ; L (C L )),

a singular Green operator

G(μ) = op(g)(μ), ˜︁ −d,0;0 (Rn−1 × R+ ; (L, 0), (L, 0)),

g∈B G

and trace operators

˜︁ −j− 12 ,0;0
Kj (μ) = op(kj )(μ), kj ∈ B G (Rn−1 × R+ ; (0, mj ), (L, 0)),

such that, with K(μ) = (K0 (μ), . . . , Kd−1 (μ)),

(︃ )︃ (︃ )︃
eiθ μd − A 1 0
( B+ (μ) + G(μ) K(μ)) = ,
T 0 Π
(︃ )︃
eiθ μd − A
(B+ (μ) + G(μ) K(μ)) =1
T

for sufficiently large μ. In particular, the map (7.3) is bijective for large μ.

Proof. By Theorem 7.2 there exists a q ∈ B−d,0;0 (Rn−1 × R+ ; (L, M ), (L, 0), πΛ , 1) such
that, for large enough μ,
(︃ )︃ (︃ )︃
1 0 A(μ)
1 = op(q)(μ)op(p)(μ) = op(q)(μ)
0 Λ(μ) T

and
(︃ )︃ (︃ )︃ (︃
)︃ (︃ )︃
1 0 1 0 1 0 1 0
= = op(p)(μ)op(q)(μ)
0 ΠΛ (μ) 0 Λ(μ) 0 Π 0 Λ(μ)−1
(︃ )︃ (︃ )︃
1 0 A(μ)
= op(q)(μ).
0 Λ(μ) T

The latter is equivalent to

(︃ )︃ (︃ )︃ (︃ )︃
1 0 A(μ) 1 0
= op(q)(μ) .
0 Π T 0 Λ(μ)
(︃ )︃
1 0
It remains to observe that op(q)(μ) = ( B+ (μ) + G(μ) K(μ)) as stated. □
0 Λ(μ)

Let us reformulate the assumption of the previous theorem, i.e., that p satisfies (Π1)−
(Π3), in a more explicit way:
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 69

(Π1) The homogeneous principal symbol of A(μ) is invertible whenever |ξ ′ , μ| = 1 and

x ∈ Rn + with uniform estimate
⃦(︂ ∑︂ )︂−1 ⃦
⃦ iθ d ⃦
⃦ e μ − aα (x)ξ α ⃦ ≲ 1.
L (C L )
|α|=d

(Π2) For some s > d − 1

2 the maps

L2 (Rn−1 , H s−d (R+ , C L )

σ∂ (p)(·, ξ ′ , μ)
(d,0)
, H (R+ , C )) −→ (︁⊕
2 n−1 s L
: L (R )︁ .
σhom (Π)(·, ξ ′ ) L2 (Rn−1 , C M )

are isomorphisms for all (ξ ′ , μ) with |ξ ′ , μ| = 1, ξ ′ ̸= 0 (a(·) means here the operator
of multiplication with the function a).
(Π3) For some s > d − 12 the following map is an isomorphism:
⎛ ⎞
eiθ − a0,d (x′ , 0)Dxdn L2 (Rn−1 , H s−d (R+ , C L ))
⎜ Π00 S00 γ0 ⎟ ⊕
⎜ .. ⎟ : L2 (Rn−1 , H s (R+ , C L )) −→
⎝ ⎠ d−1
. ⊕ Πjj (L2 (Rn−1 , C mj ))
Πd−1,d−1 Sd−1,d−1 γd−1 j=0

In case we require that all involved principal symbols are constant in x or x′ for large
|x| and |x′ |, respectively, the conditions become:
∑︁
(Π1) eiθ μd − aα (x)ξ α is invertible for all (x′ , ξ ′ , μ) with |ξ ′ , μ| = 1.
|α|=d
(Π2) For some s > d − 12 ,

H s−d (R+ , C L )
σ∂ (p)(x′ , ξ ′ , μ)
(d,0)
: H (R+ , C )) −→
s L
⊕ ,
σhom (Π)(x′ , ξ ′ )(C M )

is an isomorphism for all (x′ , ξ ′ , μ) with |ξ ′ , μ| = 1, ξ ′ ̸= 0.

(Π3) As above.

7.2.3. Resolvent and trace expansions

With the above notation let AT denote the unbounded operator in L2 (Rn+ , C L ) acting
as A on the domain

𝒟(AT ) = {u ∈ H d (Rn+ , C L ) | T u = 0}.

By Theorem 7.5, eiθ μd − AT is invertible for sufficiently large μ with inverse

(eiθ μd − AT )−1 = B+ (μ) + G(μ), μ >> 1.

70 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

Corollary 7.6. With the above notation and assumptions, Γθ := {eiθ μ | μ ≥ 0} is a ray
of minimal growth for AT , i.e., λ − AT is invertible for sufficiently large λ ∈ Γθ with
uniform estimate

1
∥(λ − AT )−1 ∥L (L2 (Rn+ ,C L )) ≲ .
|λ|

Proof. Using notation from Theorem 7.5 observe that

op+ (b)(x′ , ξ ′ , μ) + g(x′ , ξ ′ , μ) ∈ S˜︁1,0

−d,0
(Rn−1 × R+ ; L (L2 (R+ , C L )),

since the dilation group-action is a group of unitary operators on L2(R+ , C L ). More­

over, p(x′ , ξ, μ) ∈ S˜︁1,0
−d,0
(Rn−1 × R+ ; L (E)) implies pμ (x′ , ξ ′ ) := μd p(x′ , ξ, μ) ∈
0
S1,0 (Rn−1
; L (E)) uniformly μ, hence μd op(p)(μ) = op(pμ ) ∈ L (L2 (Rn−1 , E)) is uni­
formly bounded in μ. □

Given a trace class operator K : L2 (Rn+ , C L ) → L2 (Rn+ , C L ) which is also an integral

operator with continuous kernel k ∈ C (Rn+ × Rn+ , C L×L ), its trace coincides with

ˆ ˆ ˆ∞
Tr K = tr k(x, x) dx = tr k(x′ , xn , x′ , xn ) dxn dx′ .
Rn
+ Rn−1 0

If p(μ) = op+ (a)(μ) + g(μ) ∈ Bd,ν;0 (Rn−1 × R+ ; (L, 0), (L, 0)) has order d < −n, then
op(p)(μ) ∈ L (L2 (Rn+ , C L )) for every μ with continuous integral kernel
ˆ ˆ
′ ′
k(x, y, μ) = ei(x−y)ξ a(x, ξ, μ) dξ + ei(x −y )ξ g(x′ , ξ ′ , μ; xn , yn ) d¯ξ ′ .

Thus, supposing in addition the trace class property, we get

ˆ ˆ
Tr op(p)(μ) = tr a(x, ξ, μ) dxd¯ξ+
Rn Rn
+

ˆ ˆ ˆ
+∞

+ tr g(x′ , ξ ′ , μ; xn , xn ) dxn dx′ d¯ξ ′ .

Rn−1 Rn−1 0

To have the trace class property it is not sufficient to assume negative order d < −n
but one needs to require decay of the coefficients in x, essentially like ⟨x⟩−n−ε for some
ε > 0. Note however, that the symbol-kernel of the singular Green symbol g is rapidly
decreasing in xn and yn . Thus it makes sense to introduce the symbol Tr+ g as

ˆ
+∞
′ ′
(Tr+ g)(x , ξ , μ) := tr g(x′ , ξ ′ , μ; xn , xn ) dxn ;
0
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 71

in fact, for each fixed (x′ , ξ ′ , μ) this is the trace of the trace class operator in L2 (R+ , C L )
with integral kernel (xn , yn ) ↦→ g(x′ , ξ ′ , μ; xn , yn ).

˜︁ d,ν;0 (Rn−1 × R+ ; (L, 0), (L, 0)) with arbitrary d, ν ∈ R, then Tr+ g ∈
Lemma 7.7. If g ∈ B
˜︁ (R
S d,ν n−1
× R+ ).

Proof. Recall that, by definition, g has a symbol-kernel

g(x′ , ξ ′ , μ; xn , xn ) = [ξ ′ , μ]g ′ (x′ , ξ ′ , μ; [ξ ′ , μ]xn , [ξ ′ , μ]xn )

˜︁ d,ν (Rn−1 × R+ ; S (R+ × R+ , C L×L )). Inserting this in the defining expres­
where g ′ ∈ S
sion of Tr+ g and making the change of variables t := [ξ ′ , μ]xn immediately yields the
claim. □

As an immediate corollary of [29, Theorem 6.3] (which is based on techniques for trace
expansions developed by Grubb and Seeley, in particular in [11, Theorem 2.1]) we get:

˜︁ d,ν;0 (Rn−1 × R+ ; (L, 0), (L, 0)) with d < −n + 1 and d − ν ≤ 0.

Theorem 7.8. Let g ∈ B G
Then
ˆ ∑︂ ∑︂
+∞ +∞
(︁ )︁
Tr+ g(x′ , ξ ′ , μ) d¯ξ ′ ∼μ→+∞ cj (x′ )μd−j+n−1 + c′j (x′ ) log μ + c′′j (x′ ) μd−ν−j
j=0 j=0

with certain continuous and bounded functions cj , c′j , c′′j .

Concerning the pseudodifferential part, using the asymptotic expansion of a in homo­

d−j
geneous components a(d−j) (x, ξ, μ) ∈ Shom (Rn × R+ ; L (C L )), it is easy to see that

ˆ ∑︂
+∞
tr a(x, ξ, μ) d¯ξ ∼μ→+∞ aj (x)μd−j+n
j=0
Rn
ˆ
provided d < −n with coefficients aj (x) = a(d−j) (x, ξ, 1) d¯ξ.
Let us now consider P (μ) = D(μd eiθ − AT )−ℓ , where D is a differential operator
of order m and ℓ is so large that m − dℓ < −n. By Theorem 7.5, for large enough μ,
P (μ) = op(p)(μ) with a symbol

p ∈ Bm−dℓ,0;0 (Rn−1 × R+ ; (L, 0), (L, 0)).

The above results imply at once the following:

Theorem 7.9. With the above notation and ellipticity assumptions, and supposing that
the coefficients of D are decaying in x sufficiently fast,
72 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

−1
∑︂ ∑︂
+∞
(︁ ′ )︁ m−j
Tr D(λeiθ − AT )−ℓ ∼λ→+∞
m−j
−ℓ
αj λ d + αj log λ + αj′′ λ d −ℓ
j=−n j=0

for certain coefficients αj , αj′ , and αj′′ (depending on A, T , and D).

If in the previous theorem the differential operator D is tangential near the boundary,
then α0′ = . . . = αm−1
′
= 0 because in this case p ∈ Bm−dℓ,m;0 (Rn−1 × R+ ; (L, 0), (L, 0))
has regularity number m.

7.3. Example: Laplace type operators

The paper [9] investigates trace expansions for D(λ − AT )−ℓ , ℓ > n+m 2 , where A is
a Laplace type operator on a compact manifold acting on sections of an L-dimensional
vector-bundle, D is a differential operator of order m, and T is a boundary condition
related to spectral boundary conditions. We shall consider the analogous problem in
the half-space; for simplicity we shall assume that A and T have constant coefficients
outside
(︃ some)︃ compact set. We will explain that the assumptions made in [9] imply that
μ2 eiθ − A
is an elliptic problem in the sense of Theorem 7.5 for every θ ∈ (0, 2π) and
T
therefore the trace expansion follows from Theorem 7.9 with d = 2.
So let A be a second order differential operator on Rn+ with C L×L -valued coefficients
such that, near the boundary,

A = Dx2n + A′ + xn A2 + A1 ,

where A′ is a second order self-adjoint, positive differential operator on the boundary and
Aj is a differential operator of order j, cf. [9, Assumption 1.1]. Assume that e−θ μ2 − A
for some θ ∈ (0, 2π) satisfies ellipticity assumption (Π1) (for example, this is the case
when A is, as assumed in [9], principally self-adjoint, i.e., A∗ − A has order one). The
boundary condition has the form
(︃ )︃
Πγ0 u
Tu = ,
(1 − Π)(γ1 u + Bγ0 u)

where B is a first order pseudo-differential operator on the boundary and Π is a zero

order projection, cf. [9, (1.4)]. Thus, in our notation, d = 2, S00 = S11 = 1, S10 = B,
Π00 = Π, Π11 = 1 − Π, and S01 = Π01 = Π10 = 0. Let us first consider the principal
limit symbol, i.e.,

⎛ ⎞ L2 (Rn−1 , L2 (R+ , C L ))
e iθ
− Dx2n ⊕
⎝ Πγ0 ⎠ : L2 (Rn−1 , H 2 (R+ , C L )) −→ Π(L2 (Rn−1 , C L )) .
(1 − Π)γ1 ⊕
(1 − Π)(L2 (Rn−1 , C L ))
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 73

This operator is bijective whenever θ ̸= 0. In fact, in this case, eiθ − Dx2n is surjective
with kernel

{︁ }︁
N = N (θ) = u(x) = v(x′ )e−ρxn | v ∈ L2 (Rn−1 , C L ) , ρ = ei(θ−π)/2 .

Hence it suffices to verify that

(︃ )︃ Π(L2 (Rn−1 , C L ))
Πγ0
: N −→ ⊕
(1 − Π)γ1
(1 − Π)(L2 (Rn−1 , C L ))

is bijective. u ∈ N belongs to the kernel of this map if and only if Πv = 0 and −ρ(1 −
Π)v = 0, i.e., v = 0. This shows the injectivity. Given f, g ∈ L2 (Rn−1 , C L ) take u ∈ N
with v := Πf − ρ1 (1 − Π)g. Then Πγ0 u = Πf and (1 − Π)γ1 u = (1 − Π)g. This shows the
surjectivity. In other words, ellipticity condition (Π3) is satisfied.
Let us now look at the bijectivity of the principal boundary symbol

⎛ ⎞ L2 (R+ , C L )
eiθ μ2 − a′ (2) (x′ , ξ ′ ) − Dx2n ⊕
⎝ π (0) (x′ , ξ ′ )γ0 ⎠ : H 2 (R+ , C L ) −→ π (0) (x′ , ξ ′ )(C L ) ,
(0) ′ ′ (1) ′ ′
(1 − π (x , ξ ))(γ1 + b (x , ξ )γ0 ) ⊕
(1 − π (0) (x′ , ξ ′ ))(C L )

where ξ ′ ̸= 0 and a′ (2) , π (0) , and b(1) are the homogeneous principal symbols of A′ , Π,
and B, respectively.
ˆ Let σ = σ(x′ , ξ ′ , μ) denote the square root of a′ (2) (x′ , ξ ′ ) − eiθ μ2 ,
1
i.e., σ = λ1/2 (λ − a′ (2) + eiθ μ2 )−1 dλ, where λ1/2 is the principal branch of the
2πi
Γ
square root on C \ (−∞, 0] and Γ is a Jordan curve around the poles of the integrand.
The operator in the first row is surjective with kernel

{︁ }︁
N = N (x′ , ξ ′ , μ) = ϕ(xn ) = e−σxn z | z ∈ C L .

Thus we have to verify the bijectivity of

(︃ )︃ π (0) (x′ , ξ ′ )(C L )

π (0) (x′ , ξ ′ )γ0
: N −→ ⊕ ,
(1 − π (x , ξ ))(γ1 + b(1) (x′ , ξ ′ )γ0 )
(0) ′ ′
(1 − π (0) (x′ , ξ ′ ))(C L )

i.e., the bijectivity of

(︃ )︃ π (0) (x′ , ξ ′ )(C L )

π (0) (x′ , ξ ′ )
: C L −→ ⊕ . (7.6)
(1 − π (x , ξ ))(−σ + b(1) (x′ , ξ ′ ))
(0) ′ ′
(1 − π (0) (x′ , ξ ′ ))(C L )
74 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

Since both domain and co-domain have dimension L it suffices to verify the injectivity.
To this end we assume that π (0) commutes with a′ (hence with σ), cf. [9, Assumption
(2)

2.4], and that

−σ + (1 − π (0) )b(1) (1 − π (0) ) : C L −→ C L

is an isomorphism whenever ξ ′ ̸= 0, cf. [9, (2.3)] and [9, Assumption 2.7]. If z is in the
kernel of (7.6), then z = (1 − π (0) (x′ , ξ ′ ))z, hence

0 = (1 − π (0) (x′ , ξ ′ ))(−σ + b(1) (x′ , ξ ′ ))z = [−σ + (1 − π (0) )b(1) (1 − π (0) )](x′ , ξ ′ )z.

It follows z = 0. We thus have verified ellipticity assumption (Π2).

8. The calculus for manifolds

In this section we indicate how the calculus extends to compact manifolds. We dis­
cuss coordinate-invariance and shall prove that the principal limit symbol has a global
analogue. Moreover, we state how the other principal symbols look like globally. The
technique to define operators on a manifold is to use local coordinates and partitions
of unity. We do not enter in details concerning global composition and parametrix con­
struction.

8.1. Twisting with group-action revisited

Both characterization and definition of generalized singular Green symbols, cf. The­
orem 4.4 and Definition 4.5, made use of the symbol κ(ξ ′ , μ) = κ[ξ′ ,μ] associated with
the dilation group-action. We shall show now that we may substitute the smoothed
norm-function [ξ ′ , μ] by more general symbols of first order. In fact, let a(x′ , ξ ′ , μ) ∈
S 1 (Rn−1 × R+ ) be homogeneous of degree 1 in the large. Let us write (cf. (2.7))
(︃ )︃
˜︁
κ 0
˜︁ ′ , ξ ′ , μ) := κa(x′ ,ξ′ ,μ) ,
κ(x ˜︁ =
𝜿 .
0 1

Proposition 8.1. Let 𝔅d;r d;r ˜︁ d,ν;r and let 𝔖d represent the
G represent one choice of BG or BG
˜︁ d,ν . Assume that
corresponding choice S d or S

[ξ ′ , μ] ≲ a(x′ , ξ ′ , μ) ≲ [ξ ′ , μ]

uniformly in (x′ , ξ ′ , μ). Then

g ∈ 𝔅d;r
G (R
n−1
˜︁ −1 g˜︁
× R+ ; 𝔤) ⇐⇒ 𝜿 𝜿 ∈ 𝔖d (Rn−1 × R+ ; 𝔤).
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 75

Proof. It is sufficient to consider the case r = 0. For simplicity of presentation we shall

show the result for Poisson symbols. The case of trace symbols then follows by duality,
the case of singular Green symbols works in the same way and only involves more lengthy
notation due to the (xn , yn )-dependence of the symbol-kernels.
Let k ∈ 𝔅d;r G (R
n−1
× R+ ; (0, M0 ), (L, 0)) be a Poisson symbol with symbol-kernel
k(x , ξ , μ; xn ) = [ξ , μ]1/2 k′ (x′ , ξ ′ , μ; [ξ ′ , μ]xn ) where
′ ′ ′

k′ (x′ , ξ ′ , μ; xn ) ∈ 𝔖d (Rn−1 × R+ ; S (R+ , C L×M0 ).

Since x′ enters here simply as a Cb∞ -parameter it is no loss of generality to assume that
k′ is constant in x′ . Moreover, we may assume L = M0 = 1. Obviously,

k(x′ , ξ ′ , μ; xn ) = a(x′ , ξ ′ , μ)1/2 k′′ (x′ , ξ ′ , μ; a(x′ , ξ ′ , μ)xn ),

where

k′′ (x′ , ξ ′ , μ; xn ) = p(x′ , ξ ′ , μ)1/2 k′ (ξ ′ , μ; p(x′ , ξ ′ , μ)xn ),

[ξ ′ , μ]
p(x′ , ξ ′ , μ) := .
a(x′ , ξ ′ , μ)

Now the claim is equivalent to the statement

k′ ∈ 𝔖d (Rn−1 × R+ ; S (R+ )) ⇐⇒ k′′ ∈ 𝔖d (Rn−1 × R+ ; S (R+ )).

Since both p and 1/p belong to S 0 (Rn−1 × R+ ) We only show direction ``⇒'', since
the other direction works in the same way replacing p with 1/p. As explained below
Proposition 4.11, by a tensor-product argument we may assume that k′ has the form

k′ (ξ ′ , μ; xn ) = ˜︁
k′ (ξ ′ , μ)ϕ(xn )

with ˜︁
k′ ∈ 𝔖d (Rn−1 × R+ ) and ϕ ∈ S (R+ ). Then

k′′ (x′ , ξ ′ , μ; xn ) = ˜︁
k′ (ξ ′ , μ)ϕp (x′ , ξ ′ , μ), ϕp (x′ , ξ ′ , μ) = p(x′ , ξ ′ , μ)ϕ(p(x′ , ξ ′ , μ)xn ).

Since p is homogeneous of degree 0 in the large and 1 ≲ p ≲ 1,

˜︁ 0,0 (Rn−1 × R+ ; S (R+ )).

ϕp ∈ S 0 (Rn−1 × R+ ; S (R+ )) ⊂ S

This implies at once the desired property for k′′ . □

76 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

8.2. Invariance under coordinate changes

We will study the invariance of the calculus on the half-space. In the present work we
shall limit outselves to diffeomorphisms of the half-space that leave the ``normal-direction
untouched'', i.e., have the form

˜︁ ′ , xn ) = (χ(x′ ), xn ),
χ(x

where χ is a diffeomorphism of Rn−1 . Since our calculus concerns operators with globally
bounded coefficients we need to impose some control at infinity of the diffeomorphism
χ, namely that ∂j χk ∈ Cb∞ (Rn−1 ), 1 ≤ j, k ≤ n − 1, and |det χ′ | are uniformly bounded
from above and below by some positive constants.
Given a symbol p ∈ Bd,ν;r (Rn−1 × R+ ; 𝔤), 𝔤 = ((L0 , M0 ), (L1 , M1 )), the pull-back of
op(p)(μ) is defined by

χ∗ op(p)(μ)u = [op(p)(μ)(u ◦ χ−1 )] ◦ χ., u ∈ S (Rn−1 , H s (R+ , C L0 ) ⊕ C M0 ).

Note that we consider op(p)(μ) as a pseudodifferential operator on Rn−1 acting on

vector-valued Sobolev spaces (having values in Sobolev spaces of the half-axis). For this
reason we shall use the notation χ∗ rather than χ
˜︁∗ .

Lemma 8.2. Let a ∈ S ˜︁ d,ν (Rn−1 × R+ ). Then there exists a unique symbol χ∗ a in the
same class such that χ∗ op(a)(μ) = op(χ∗ a)(μ) for all μ. Moreover,

(χ∗ a)(d,ν) (x′ , ξ ′ , μ) = a(d,ν) (χ(x′ ), χ′ (x′ )−t ξ ′ , μ), (χ∗ a)∞ ∗ ∞
[d,ν] = χ (a[d,ν] ).

Proof. Except the formula for the limit symbol this is [29, Theorem 7.1]. There it is also
shown that the class S˜︁1,0
d,ν
(Rn−1 × R+ ) is invariant. We have

∑︂
N −1
a(x′ , ξ ′ , μ) ≡ aj (x′ , ξ ′ )pj (ξ ′ , μ) mod S˜︁1,0
d,ν+N
(Rn−1 × R+ )
j=0

where for brevity we write here aj := a∞ ′ ′

[d,ν+j] and pj (ξ , μ) := [ξ , μ]
d−ν−j
. Therefore

∑︂
N −1
χ∗ a ≡ (χ∗ aj )#(χ∗ pj ) mod S˜︁1,0
d,ν+N
(Rn−1 × R+ ).
j=0

˜︁ d−ν−j,0 (Rn−1 × R+ ). Inserting the corresponding

Now χ∗ pj ∈ S d−ν−j (Rn−1 × R+ ) ⊂ S
expansions of χ∗ pj yields

∑︂
N −1 (︂ ∑︂ )︂
χ∗ a ≡ (χ∗ ak )#(χ∗ pk )[d−ν−k,ℓ] pj mod S˜︁1,0
d,ν+N
(Rn−1 × R+ ).
j=0 k+ℓ=j
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 77

Since (χ∗ p0 )[d−ν,0] (x′ , ξ ′ ) = (χ∗ p0 )(d−ν) (x′ , 0, 1) = 1, the leading coefficient is χ∗ a0 =
χ∗ (a∞
[d,ν+j] ). □

Theorem 8.3. Let p ∈ Bd,ν;r (Rn−1 × R+ ; 𝔤). Then there exists a unique symbol χ∗ p ∈
Bd,ν;r (Rn−1 × R+ ; 𝔤) such that χ∗ op(p)(μ) = op(χ∗ p)(μ) for all μ. For the associated
principal symbols, the following relations are valid:

σψ (χ∗ p)(x, ξ, μ) = σψ (p)(χ(x′ ), xn , χ′ (x′ )−t ξ ′ , ξn , μ),

(d) (d)

(χ∗ p)(x′ , ξ ′ , μ) = σψ (p)(χ(x′ ), χ′ (x′ )−t ξ ′ , μ),

(d,ν) (d,ν)
σ∂
[d]
σ∞ (χ∗ p) = χ∗ σ∞
[d]
(p),

[d]
where the latter means the symbol of the pull-back of op(σ∞ (p)) under χ.

Proof. The result is known to be true if p is strongly parameter-dependent, i.e., with

p also χ∗ p belongs to B d;r (Rn−1 × R+ ; 𝔤) and, in this case, the homogeneous principal
symbol and principal boundary symbol behave as indicated. Since the principal limit
symbol, in this case, coincides with the principal boundary symbol evaluated in (ξ ′ , μ) =
(0, 1), the claim for the limit symbol is trivially true. Hence it remains to show the claim
for p ∈ B ˜︁ d,ν;r (Rn−1 × R+ ; 𝔤).
G
Since the diffeomorphism does not effect xn it leaves the operator ∂+ invariant. For
this reason it suffices to consider generalized singular Green symbols of type r = 0. For
simplicity of presentation we shall prove the theorem for Poisson symbols only; trace
and singular Green symbols are treated in an analogous way. So let k ∈ B ˜︁ d,ν;0 (Rn−1 ×
G
R+ ; (0, M0 ), (L, 0)) and k′ = κ−1 #k ∈ S ˜︁ d,ν;0 (Rn−1 × R+ ; (0, M0 ), (L, 0)). As already
G
observed in the proof of Proposition 8.1, it is no restriction to assume 𝔤 = ((0, 1), (1, 0))
and that k′ is of product form, i.e.,

k′ (x′ , ξ ′ , μ; xn ) = a(x′ , ξ ′ , μ)ϕ(xn )

˜︁ d,ν (Rn−1 × R+ ) and ϕ ∈ S (R+ ) (we identify symbols with symbol-kernels).

with a ∈ S
Then

∞
k′ [d,ν] (x′ , ξ ′ ; xn ) = a∞ ′ ′
[d,ν] (x , ξ )ϕ(xn ).

Moreover, op(k′ )(μ) = op(φ)op(a)(μ), where op(a)(μ) is a pseudodifferential operator

on Rn−1 while [op(φ)u](x) = u(x′ )ϕ(xn ) is a Poisson operator. Therefore

k = κ#k′ = kϕ #a, kϕ (x′ , ξ ′ , μ; xn ) = [ξ ′ , μ]1/2 ϕ([ξ ′ , μ]xn ),

hence χ∗ k = (χ∗ kϕ )#(χ∗ a). Since kϕ ∈ BG 0;0

(Rn−1 × R+ ; 𝔤), also χ∗ kϕ belongs to the
˜︁ d,ν (Rn−1 × R+ ) by Lemma 8.2. This yields χ∗ k ∈ B
same class; χ∗ a ∈ S ˜︁ d,ν;0 (Rn−1 ×
G
R+ ; 𝔤). Moreover,
78 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

⃓
(χ∗ kϕ )∞ ′ ′ ∗
[0,0] (x , ξ ; xn ) = (χ kϕ ) (x , ξ , μ; xn )⃓(ξ′ ,μ)=(0,1)
(0) ′ ′

⃓
= kϕ (χ(x′ ), 0, 1; xn ) = |ξ ′ , μ|1/2 ϕ(|ξ ′ , μ|xn )⃓(ξ′ ,μ)=(0,1)
(0)

= ϕ(xn )

and therefore, again using Lemma 8.2,

(χ∗ k)∞ ′ ′ ∗ ∞ ∗ ∞ ′ ′
[d,ν] (x , ξ ; xn ) = ((χ kϕ )[0,0] #(χ a)[d,ν] )(x , ξ ; xn )
∞
= (χ∗ a∞ ′ ′ ∗ ′ ′ ′
[d,ν] )(x , ξ )ϕ(xn ) = (χ k [d,ν] )(x , ξ ; xn )

= (χ∗ k[d,ν]
∞
)(x′ , ξ ′ ; xn ).

This gives the transformation rule for the principal limit symbol. The principal boundary
symbol is treated similarly, again starting out from χ∗ k = (χ∗ kϕ )#(χ∗ a). □

8.3. Operators on compact manifolds

Let M be a (smooth) compact Riemannian manifold with boundary ∂M , embedded in

a closed manifold denoted by 2M . Moreover, for j = 0, 1, let Fj be a Hermitean vector
bundle over ∂M and Ej the restriction to M of a Hermitian vector bundle 2Ej over
2M . Let 𝒰 ∼ = ∂M × (−1, 1) be a tubular neighborhood with a splitting of coordinates
(x′ , xn ), where xn ≥ 0 on M . We assume that the Riemannian and Hermitean structure
is compatible with the product structure of 𝒰. We want to define the classes

Bd,ν;r (M × R+ ; 𝔤), 𝔤 = ((E0 , F0 ), (E1 , F1 )),

with d ∈ Z and r, ν ∈ N0 . Elements will be operator-families

(︃ )︃ H s (int M, E0 ) H s−d (int M, E1 )

A+ (μ) + G(μ) K(μ)
P (μ) = : ⊕ −→ ⊕
T (μ) Q(μ)
H s (∂M, F0 ) H s−d (∂M, F1 )

whenever s > r − 12 . The pseudodifferential part is of the form

A+ (μ) = r+ A(μ)e+ , A(μ) ∈ Ld (2M ; 2E0 , 2E1 ),

where A(μ) is a strongly parameter-dependent pseudodifferential operator defined on

2M .
It remains to introduce the classes of generalized singular Green operators.

Definition 8.4. We define

⎛ ⎛ ′ ⎞⎞
H0s (M, E0 ) H s (int M, E1 )
B −∞;0 (M × R+ ; 𝔤) = ∩ S ⎝R+ ; L ⎝ ⊕ , ⊕ ⎠⎠
s,s′ ∈R
H s (∂M, F0 ) H s′ (∂M, F1 )
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 79

∑︁
r−1
and B −∞;r (M × R+ ; 𝔤) as the space of all operator-families G(μ) = Gj (μ)𝝏 j+ , with
j=0
Gj (μ) ∈ B −∞;0 (M × R+ ; 𝔤) and 𝝏 + as in (2.10), where ∂+ represents a fixed choice of
a first order differential operator which coincides with ∂xn near the boundary.

Modulo this class of globally defined regularizing operators, all other contributions to
the class of generalized singular Green operators are defined locally near the boundary.

Lemma 8.5. Let g ∈ B ˜︁ d,ν;r (Rn−1 × R+ ; 𝔤) and ω ∈ C ∞ (R+ ) be a compactly supported

G
function with ω = 1 near the origin. Then
(︃ )︃ (︃ )︃
ω 0 ω 0 −∞;r
g− g, g − g ∈ BG (Rn−1 × R+ ; 𝔤).
0 0 0 0

Proof. The claim follows from Lemma 5.6 by writing (1−ω)(xn ) = aj (xn )xjn for arbitrary
j ∈ N where aj (xn ) = (1 − ω)(xn )x−j ∞
n ∈ Cb (R+ ). □

In the following definition ω is a cut-off function as in Lemma 8.5, considered as a

function defined on M with support in the collar-neighborhood of ∂M . The definition
does not depend on the choice of ω.

Definition 8.6. Let d, ν ∈ R and let 𝔅d;r d;r ˜︁ d,ν;r . The space
G be one choice of BG or BG
𝔅d;r
G (M × R+ ; 𝔤), 𝔤 = ((E0 , F0 ), (E1 , F1 )), consist of all operator-families

C ∞ (M, E0 ) C ∞ (M, E1 )
P (μ) : ⊕ → ⊕
C ∞ (∂M, F0 ) C ∞ (∂M, F1 )

satisfying:
(︃ )︃ (︃ )︃
1−ω 0 1−ω 0
(1) P (μ) and P (μ) belong to B −∞;r (M × R+ ; 𝔤).
0 0 0 0
(2) For every chart U of ∂M such that Ej |U ×[0,1) (︃ = U ×[0,
)︃ 1)×C (︃
Lj
and)︃Fj |U = U ×C Mj
ϕω 0 ψω 0
and for every ϕ, ψ ∈ C0∞ (U ) the operator P (μ) transported to
0 ϕ 0 ψ
Rn+ via the local trivialization is a generalized singular Green operator op(p)(μ) with
symbol p ∈ 𝔅d;r G (R
n−1
× R+ ; ((L0 , M0 ), (L1 , M1 ))).

Definition 8.7. Let d ∈ Z and ν ∈ N0 . Then we set

{︃(︃ )︃ }︃
A+ (μ) 0 d;r
B (M × R+ ; 𝔤) =
d;r
| A(μ) ∈ L (2M ; 2E0 , 2E1 ) + BG
d
(M × R+ ; 𝔤)
0 0

and

˜︁ d,ν;r (M × R+ ; 𝔤).
Bd,ν;r (M × R+ ; 𝔤) = B d;r (M × R+ ; 𝔤) + B G
80 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

8.3.1. Principal symbols for the global calculus

The relevant principal symbols for the calculus on the half-space Rn+ lead to glob­
ally defined analogues in the manifold case. Given P (μ) ∈ Bd,ν;r (M × R+ ; 𝔤), 𝔤 =
((E0 , F0 ), (E1 , F1 )), the global homogeneous principal symbol is a bundle-morphism

σψd (P ) : π ∗ E0 −→ π ∗ E1 , π : (T ∗ M × R+ ) \ 0 −→ M,

where π is the canonical projection, i.e., π(vx , μ) = x for vx ∈ Tx∗ M ; equivalently, we

may consider π : {(vx , μ) | ∥vx ∥2 + μ2 = 1} → M , with norm induced by some fixed
Riemannian metric. The global principal boundary symbol is a bundle-morphism

(︄ )︄ (︄ )︄
H s (R+ ) ⊗ E0′ H s−d (R+ ) ⊗ E1′ 1
: π∂∗ −→ π∂∗
(d,ν)
σ∂ (P ) ⊕ ⊕ , s>r− ,
F0 F1 2

where Ej′ = Ej |∂M and π∂ : (T ∗ ∂M \ 0) × R+ → ∂M is the canonical projection or,

equivalently, π∂ considered with domain {(vx′ ′ , μ) | ∥vx′ ′ ∥2 + μ2 = 1, vx′ ′ ̸= 0}. The
principal angular symbol is a bundle-morphism

(︄ )︄ (︄ )︄
H s (R+ ) ⊗ E0′ H s−d (R+ ) ⊗ E1′ 1
⟨d⟩
ˆ︁
σ (P ) : ˆ︁∂∗
π ⊕ −→ ˆ︁∂∗
π ⊕ , s>r− ,
F0 F1 2

where πˆ︁∂ : T ∗ ∂M \ 0 → ∂M is the canonical projection which, equivalently, can also be

restricted to the unit co-sphere bundle of the boundary.
In order to define the global analogue of the principal limit symbol, let ϕj ∈
∞
C0 (Uj ), j = 1, . . . , N , be a partition of unity such that every union Uj ∪ Uk is
contained in a chart domain Ujk over which all involved bundles are trivial. Let
(︃jk ∈ B )︃ (R (︃ × R+ ; ((L
d,ν;r n−1
p )︃ 0 , M0 ), (L1 , M1 ))) be the local symbol associated with
ϕj ω 0 ϕk ω 0
P (μ) . The principal limit operator of P (μ) is defined as
0 ϕj 0 ϕk

∑︂
N
[d] [d]
σ∞ (P ) = σ∞ (Pjk )
j,k=1

where σ∞ (Pjk ) transported in Rn+ coincides with op(p∞

[d]
jk,[d,0] ). By construction,

′ ′
H s (∂M, H s (R+ ) ⊗ E0′ ) H s (∂M, H s −d (R+ ) ⊗ E1′ )
[d]
σ∞ (P ) : ⊕ −→ ⊕
H s (∂M, F0 ) H s (∂M, F1 )

for every s′ > r − 1

2 and s ∈ R.
 J. Seiler / Journal of Functional Analysis 289 (2025) 111099 81

Data availability

No data was used for the research described in the article.

References

[1] S. Agmon, On the eigenfunctions and on the eigenvalues of general elliptic boundary value problems,
Commun. Pure Appl. Math. 15 (1962) 627--663.
[2] M.F. Atiyah, R. Bott, The index problem for manifolds with boundary, in: Coll. Differential Analysis,
Tata Institute Bombay. Oxford University Press, Oxford, 1964, pp. 175--186.
[3] R. Beals, Characterization of pseudodifferential operators and applications, Duke Math. J. 44 (1977)
45--57.
[4] R. Beals, Correction to: ``Characterization of pseudodifferential operators and applications'', Duke
Math. J. 46 (1979) 215.
[5] L. Boutet de Monvel, Boundary problems for pseudo-differential operators, Acta Math. 126 (1--2)
(1971) 11--51.
[6] G. Grubb, Functional Calculus of Pseudo-Differential Boundary Problems, 2nd edition, Birkhäuser,
Basel, 1996.
[7] G. Grubb, Parametrized pseudodifferential operators and geometric invariants, in: Microlocal Anal­
ysis and Spectral Theory (Lucca, 1996), in: NATO Adv. Sci. Inst. Ser. C: Math. Phys. Sci., vol. 490,
Kluwer Acad. Publ., Dordrecht, 1997, pp. 115--164.
[8] G. Grubb, A weakly polyhomogeneous calculus for pseudodifferential boundary problems, J. Funct.
Anal. 184 (1) (2001) 19--76.
[9] G. Grubb, Spectral boundary conditions for generalizations of Laplace and Dirac operators, Com­
mun. Math. Phys. 240 (2003) 243--280.
[10] G. Grubb, N.J. Kokholm, A global calculus of parameter-dependent pseudodifferential boundary
problems in Lp -Sobolev spaces, Acta Math. 171 (2) (1993) 165--229.
[11] G. Grubb, R.T. Seeley, Weakly parametric pseudodifferential operators and Atiyah-Patodi-Singer
boundary problems, Invent. Math. 121 (3) (1995) 481--529.
[12] H. Kumano-go, Pseudo-Differential Operators, MIT Press, Cambridge (MA), 1981.
[13] X. Liu, B.-W. Schulze, Boundary Value Problems with Global Projection Conditions. Opera­
tor Theory: Advances and Applications 265. Advances in Partial Differential Equations (Basel),
Birkhäuser/Springer, Cham, 2018.
[14] V. Nazaikinskij, B.-W. Schulze, V. Shatalov, B.Ju. Sternin, On general boundary value problems
for elliptic equations, Sb. Math. 189 (9--10) (1998) 1573--1586.
[15] S. Rempel, B.-W. Schulze, Index Theory of Elliptic Boundary Problems, Akademie-Verlag, Berlin,
1982.
[16] E. Schrohe, An introduction to Boutet de Monvel’s calculus, in: J.B. Gil, et al. (Eds.), Approaches
to Singular Analysis, Operator Theory, in: Advances in PDE, vol. 125, Birkhäuser, 2001.
[17] E. Schrohe, B.-W. Schulze, Boundary value problems in Boutet de Monvel’s calculus for manifolds
with conical singularities I, in: M. Demuth, E. Schrohe, B.-W. Schulze (Eds.), Pseudo-Differential
Operators and Mathematical Physics, Math. Topics, in: Advances in Part. Diff. Equ., vol. 5,
Akademie Verlag, Berlin, 1994.
[18] B.-W. Schulze, Pseudo-Differential Operators on Manifolds with Singularities, Studies in Mathe­
matics and Its Applications, vol. 24, North-Holland Publishing Co., Amsterdam, 1991.
[19] B.-W. Schulze, Pseudo-Differential Boundary Value Problems, Conical Singularities, and Asymp­
totics, Mathematical Topics, vol. 4, Akademie Verlag, Berlin, 1994.
[20] B.-W. Schulze, An algebra of boundary value problems not requiring Shapiro-Lopatinskij conditions,
J. Funct. Anal. 179 (2) (2001) 374--408.
[21] B.-W. Schulze, Boundary value problems with the transmission property, in: Pseudo-Differential
Operators: Complex Analysis and Partial Differential Equations, in: Oper. Theory Adv. Appl.,
vol. 205, Birkhäuser Verlag, Basel, 2010, pp. 1--50.
[22] B.-W. Schulze, J. Seiler, Boundary value problems with global projection conditions, J. Funct. Anal.
206 (2) (2004) 449--498.
[23] B.-W. Schulze, J. Seiler, Edge operators with conditions of Toeplitz type, J. Inst. Math. Jussieu
5 (1) (2006) 101--123.
82 J. Seiler / Journal of Functional Analysis 289 (2025) 111099

[24] R.T. Seeley, Complex powers of an elliptic operator, in: Singular Integrals, Proc. Sympos. Pure
Math., Chicago, Ill., 1966, Amer. Math. Soc., 1967, pp. 288--307.
[25] R.T. Seeley, The resolvent of an elliptic boundary problem, Am. J. Math. 91 (1969) 889--920.
[26] R.T. Seeley, Topics in pseudo-differential operators, in: C.I.M.E. Estivo (Ed.), Pseudo-Diff. Opera­
tors (C.I.M.E., Stresa, 1968), Centro Internazionale Matematico Estivo (C.I.M.E.), Ed. Cremonese,
Rome, 1969, pp. 167--305.
[27] J. Seiler, Continuity of edge and corner pseudodifferential operators, Math. Nachr. 205 (1999)
163--182.
[28] J. Seiler, Ellipticity in pseudodifferential algebras of Toeplitz type, J. Funct. Anal. 263 (5) (2012)
1408--1434.
[29] J. Seiler, Parametric pseudodifferential operators with point-singularity in the covariable, Ann.
Glob. Anal. Geom. 61 (3) (2022) 553--592.
[30] J. Seiler, Singular Green operators in the edge algebra formalism, J. Math. Anal. Appl. 511 (1)
(2022) 126041.