arXiv:1505.01566v1 [math.

AP] 6 May 2015

FOURIER INTEGRAL OPERATORS ALGEBRA AND

FUNDAMENTAL SOLUTIONS TO HYPERBOLIC SYSTEMS
WITH POLYNOMIALLY BOUNDED COEFFICIENTS ON Rn
ALESSIA ASCANELLI AND SANDRO CORIASCO
Abstract. We study the composition of an arbitrary number of Fourier integral
operators A j , j 1, . . . , M, M 2, defined through symbols belonging to the socalled SG classes. We give conditions ensuring that the composition A1 AM
of such operators still belongs to the same class. Through this, we are then able to
show well-posedness in weighted Sobolev spaces for first order hyperbolic systems
of partial differential equations with coefficients in SG classes, by constructing the
associated fundamental solutions. These results expand the existing theory for the
study of the properties at infinity of the solutions to hyperbolic Cauchy problems
on Rn with polynomially bounded coefficients.

Contents
1. Introduction
Acknowledgements
2. SG symbols and Fourier integral operators
3. Multiproducts of SG phase functions
4. Composition of SG Fourier integral operators
5. Fundamental solution to hyperbolic systems in SG classes
References

1
3
3
6
20
29
33

1. Introduction
We deal with a class of Fourier integral operators globally defined on Rn, namely,
the SG Fourier integral operators (SG FIOs, for short, in the sequel), that is, the
class of FIOs defined through symbols belonging to the so-called SG classes.
The class Sm, pR2n q of SG symbols of order pm, q P R2 is given by all the
functions apx, q P C8 pRn Rn q with the property that, for any multiindices , P
Zn` , there exist constants C 0 such that the conditions
(1.1)

|D Dx apx, q| C xxym||xy|| ,

px, q P Rn Rn ,

hold. Here xxy p1 ` |x|2 q1{2 when x P Rn , and Z` is the set of non-negative
integers. These classes, together with corresponding classes of pseudo-differential
operators OppSm, q, were first introduced in the 70s by H.O. Cordes [10] and
C. Parenti [27], see also R. Melrose [26]. They form a graded algebra with respect
to composition, i.e.,
OppSm1 ,1 q OppSm2 ,2 q OppSm1 `m2 ,1 `2 q,
2010 Mathematics Subject Classification. Primary: 58J40; Secondary: 35S05, 35S30, 47G30, 58J45.
Key words and phrases. Fourier integral operator, multi-product, phase function, hyperbolic first
order systems.
1

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

whose residual elements are operators with symbols in

S8,8 pR2n q
Sm, pR2n q SpR2n q,
pm,qPR2

that is, those having kernel in SpR2n q, continuously mapping S1 pRn q to SpRn q.
Operators in OppSm, q are continuous on SpRn q, and extend uniquely to continuous operators on S1 pRn q and from Hs, pRn q to Hsm, pRn q, where Hr, pRn q,
r, P R, denotes the weighted Sobolev (or Sobolev-Kato) space
Hr, pRn q tu P S1 pRn q : }u}r, }x.yr xDy u}L2 8u.
An operator A Oppaq, is called elliptic (or Sm, -elliptic) if a P Sm, pR2n q and
there exists R 0 such that
Cxxym xy |apx, q|,

|x| ` || R,

for some constant C 0. An elliptic SG operator A P OppSm, q admits a parametrix

P P OppSm, q such that
PA I ` K1 ,
8,8

AP I ` K2 ,

2n

for suitable K1 , K2 P OppS

pR qq, where I denotes the identity operator. In
such a case, A turns out to be a Fredholm operator on the scale of functional spaces
Hr, pRn q, r, P R.
In 1987, E. Schrohe [29] introduced a class of non-compact manifolds, the socalled SG manifolds, on which a version of SG calculus can be defined. Such
manifolds admit a finite atlas, whose changes of coordinates behave like symbols
of order p0, 1q (see [29] for details and additional technical hypotheses). A relevant
example of SG manifolds are the manifolds with cylindrical ends, where also the
concept of classical SG operator makes sense, see, e. g. [7, 15, 20, 23, 25, 26]. With
up denoting the Fourier transform of u P SpRn q, given by

(1.2)
uppq eix upxq dx,

for any a P Sm, pR2n q, P P the set of SG phase functions, see Section 2 below ,
the SG FIOs are defined, for u P SpRn q, as

(1.3)
u pOp paquqpxq p2qn eipx,q apx, qp
upq d,
and

(1.4)

pOp paquqpxq

p2q

"

eipxpy,qq apy, qupyq dyd.

Here the operators Op paq and Op paq are sometimes called SG FIOs of type I and
type II, respectively, with symbol a and SG phase function . Note that a type II
operator satisfies Op paq Op paq , that is, it is the formal L2 -adjoint of the type
I operator Op paq.
The analysis of SG FIOs started in [11], where composition results with the
corresponding classes of pseudodifferential operators, and of SG FIOs of type I
and type II with regular phase functions, have been proved, as well as the basic
continuity proprties in SpRn q and S1 pRn q of operators in the class. A version of the
Asada-Fujiwara L2 pRn q-continuity theorem was also proved there, for operators
Op paq with symbol a P S0,0 pR2n q and regular SG phase function P Pr , see
Definition 2.4. Applications to SG hyperbolic Cauchy problems were initially
given in [12, 17].
Many authors have, since then, expanded the SG FIOs theory in various directions. To mention a few, see, e.g., G.D. Andrews [1], M. Ruzhansky, M. Sugimoto

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

[28], E. Cordero, F. Nicola, L Rodino [9], and the recent works by S. Coriasco and
M. Ruzhansky [18], S. Coriasco and R. Schulz [19, 20]. Concerning applications
to SG hyperbolic problems and propagation of singularities, see, e.g., A. Ascanelli
and M. Cappiello [2, 3, 4], M. Cappiello [8], S. Coriasco, K. Johansson, J, Toft [13],
S. Coriasco, L. Maniccia [14]. Concerning applications to anisotropic evolution
equations of Schrodinger

type see, e.g., A. Ascanelli, M. Cappiello [5].

Here our aim is to expand the results in [11, 12], through the study of the
composition of M 2 SG FIOs A j : Op j pa j q with regular SG phase functions
j P Pr p j q see Definition 2.4 below and symbols a j P Sm j , j pR2n q, j 1, . . . , M.
To our best knowledge, the composition of SG FIOs with different phase functions
of the type that we consider in this paper has not been studied by other authors.
First, we shall prove, under suitable assumptions, the existence of a SG phase
function P Pr pq, called the multi-product of the SG phase functions 1 , . . . , M ,
and of a symbol a P Sm, pR2n q, with m : m1 ` ` mM , : 1 ` ` M , such
that
(1.5)

A Op paq : A1 AM ,

see Theorem 4.3 below for the precise statement.

Subsequently, we apply such result to study a class of hyperbolic Cauchy problems. We focus on first order systems of partial differential equations of hyperbolic
type with pt, xqdepending coefficients in SG classes. By means of Theorem 4.3, we
construct the fundamental solution tEpt, squ0stT to the system. The existence of
the fundamental solution provides, via Duhamels formula, existence and uniqueness of the solution to the system, for any given Cauchy data in the weighted
Sobolev spaces Hr, pRn q. A remarkable feature, typical for these classes of hyperbolic problems, is the well-posedness with loss/gain of decay at infinity, observed for
the first time in [2], see also Section 5 below. We need these results in the study of
certain stochastic equations, which will be treated in the forthcoming paper [6].
This paper is organized as follows. Section 2 is devoted to fixing notation and
recalling some basic definitions and known results on SG symbols and Fourier
integral operators, which will be used throughout the paper. In Section 3 we
perform the first step of the proof of our main result, Theorem 4.3, defining and
studying the multi-product of M 2 regular SG phase functions. In Section 4 we
prove Theorem 4.3, showing the existence, under suitable hypotheses, of P Pr
and a P Sm, such that (1.5) holds. Finally, in Section 5 we obtain the fundamental
solution to SG hyperbolic first order systems.
Acknowledgements
The authors were supported by the INdAM-GNAMPA grant Equazioni Differenziali a Derivate Parziali di Evoluzione e Stocastiche (Coordinator: S. Coriasco,
Dep. of Mathematics G. Peano, University of Turin).
2. SG symbols and Fourier integral operators
In this section we fix some notation and recall some of the results proved in [11],
which will be used below. SG pseudodifferential operators apx, Dq Oppaq can be
introduced by means of the usual left-quantization

pOppaquqpxq p2qn eix apx, qp

upqd, u P SpRn q,

p the Fourier transform of u defined in (1.2), starting from symbols apx, q P

with u
C8 pRn Rn q satisfying (1.1). Symbols of this type belong to the class denoted by
Sm, pR2n q, and the corresponding operators constitute the class OppSm, pR2n qq. In

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

the sequel we will often simply write Sm, , fixing the dimension of the base space
to n. For m, P R, l P Z` , a P Sm, , the quantities
m,

~a~l

max sup xxym`|| xy`|| |Bx B apx, q|

|`|l x,PRn

are a family of seminorms, defining the Frechet topology of Sm, . The continuity
properties of the elements of OppSm, q on the scale of spaces Hr, , m, , r, P R,
is expressed more precisely in the next Theorem 2.1 (see [10] and the references
quoted therein for the result on more general classes of SG type symbols).
Theorem 2.1. Let a P Sm, pRn q, m, P R. Then, for any r, P R, Oppaq P
LpHr, pRn q, Hrm,pRn qq, and there exists a constant C 0, depending only on
n, m, , r, , such that
(2.1)

}Oppaq}LpHr, pRn q,Hrm, pRn qq C~a~

m,

r n2 s`1

where rss denotes the integer part of s P R.

We now introduce the class of SG phase functions. Here and in what follows,
A B means that A . B and B . A, where A . B means that A c B, for a suitable
constant c 0.
Definition 2.2 (SG phase function). A real valued function P C8 pR2n q belongs to
the class P of SG phase functions if it satisfies the following conditions:
(1) P S1,1 pR2n q;
(2) x1x px, qy xy as |px, q| 8;
(3) x1 px, qy xxy as |px, q| 8.
Functions of class P are those used in the construction of the SG FIOs calculus.
The SG FIOs of type I and type II, Op paq and Op pbq, are defined as in (1.3)
and (1.4), respectively, with P P and a, b P Sm, . The next Theorem 2.3 about
composition between SG pseudodifferential operators and SG FIOs was originally
proved in [11], see also [13, 16, 22].
Theorem 2.3. Let P P and assume p P St, pR2n q, a, b P Sm, pR2n q. Then,
Opppq Op paq Op pc1 ` r1 q Op pc1 q mod OppS8,8 pR2d qq,
Opppq Op pbq Op pc2 ` r2 q Op pc2 q

mod OppS8,8 pR2d qq,

Op paq Opppq Op pc3 ` r3 q Op pc3 q mod OppS8,8 pR2d qq,

Op pbq Opppq Op pc4 ` r4 q Op pc4 q

mod OppS8,8 pR2d qq,

for some c j P Sm`t,` pR2n q, r j P S8,8 pR2d q, j 1, . . . , 4.

To obtain the composition of SG FIOs of type I and type II, some more hypotheses
are needed, leading to the definition of the classes Pr and Pr pq of regular SG phase
functions.
Definition 2.4 (Regular SG phase function). Let P r0, 1q and r 0. A function
P P belongs to the class Pr pq if it satisfies the following conditions:
(1) | detp2x qpx, q| r, @px, q;
(2) the function Jpx, q : px, q x is such that

(2.2)

sup
x,PRn
|`|2

|D Dx Jpx, q|
xxy1||xy1||

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

If only condition (1) holds, we write P Pr .

Remark 2.5. Notice that condition (2.2) means that Jpx, q{ is bounded with constant
1 in S1,1 . Notice also that condition (1) in Definition 2.4 is authomatically fulfilled when
condition (2) holds true for a sufficiently small P r0, 1q.
For P N, we also introduce the seminorms
}J}2, :

sup

2|`|2` px,qPR

and

|D Dx Jpx, q|
2n

xxy1||xy1||

}J} : sup

x,PRn
|`|1

|D Dx Jpx, q|
xxy1||xy1||

` }J}2, .

We notice that P Pr pq means that (1) of Definition 2.4 and }J}0 hold, and
then we define the following subclass of the class of regular SG phase functions:
Definition 2.6. Let P r0, 1q, r 0, 0. A function belongs to the class Pr p, q if
P Pr pq and }J} for the corresponding J.
Theorem 2.7 below shows that the composition of SG FIOs of type I and type II
with the same regular SG phase functions is a SG pseudodifferential operator.
Theorem 2.7. Let P Pr and assume a P Sm, pR2n q, b P St, pR2n q. Then,
Op paq Op pbq Oppc5 ` r5 q Oppc5 q mod OppS8,8 q,
Op pbq Op paq Oppc6 ` r6 q Oppc6 q mod OppS8,8 q,
for some c j P Sm`t,` pR2n q, r j P S8,8 pR2d q, j 5, 6.
Furthermore, asymptotic formulae can be given for c j , j 1, . . . , 6, in terms of ,
p, a and b, see [11]. A generalization of Theorems 2.3 and 2.7 to operators defined
by means of broader, generalized SG classes was proved in [13, 22], together with
similar asymptotic expansions, studied by means of the criteria obtained in [21].
Remark 2.8. In particular, in Section 5 we will make use of the following (first order)
expansion of the symbol of c1 , coming from [11]:
c1 px, q ppx, 1x px, qqapx, q ` spx, q,

s P Sm`t1,`1 pR2n q.

Finally, when a P Sm, is elliptic and P Pr , the corresponding SG FIOs admit a

parametrix, that is, there exist b1 , b2 P Sm, such that
(2.3)

Op paq Op pb1 q Op pb1 q Op paq I

mod OppS8,8 q,

(2.4)

Op paq Op pb2 q Op pb2 q Op paq I

mod OppS8,8 q,

where I is the identity operator, see again [11, 13, 22].

In this paper we extend the existing theory of SG FIOs, dealing with the composition of SG FIOs of type I with different phase functions. We then apply it to
compute the fundamental solution to SG hyperbolic systems with coefficients of
polynomial growth.
The following result is going to be used in Sections 3 and 5. Given a symbol
a P Cpr0, Ts; S,1q with P r0, 1s, let us consider the eikonal equation
#
Bt pt, s, x, q apt, x, 1xpt, s, x, qq, t P r0, T0s
(2.5)
ps, s, x, q x ,
s P r0, T0 s,
with 0 T0 T. By an extension of the theory developed in [12], it is possible to
prove that the following Proposition 2.9 holds true.

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

Proposition 2.9. For any small enough T0 P r0, Ts, equation (2.5) admits a unique
solution P C1 pr0, T0 s2t,s , S1,1 pRnx, qq, satisfying J P C1 pr0, T0 s2t,s , S,1 pRnx, qq and
Bs pt, s, x, q aps, 1 pt, s, x, q, q,

(2.6)

for any t, s P r0, T0s. Moreover, for every h 0 there exists ch 1 and Th P r0, T0 s such
that pt, s, x, q P Pr pch |t s|q, with }J}2,h ch |t s| for all 0 s t Th .
In the sequel we will sometimes write ts px, q : pt, s, x, q, for a solution of
(2.5).
3. Multiproducts of SG phase functions
The first step in our construction is to define the multi-product of regular SG
phase functions and to analyze its properties, which we perform in the present
section, following mainly [24].
Let us consider a sequence t j u j1 of regular SG phase functions j px, q P Pr p j q
with
8

(3.1)
j : 0 1{4.
j1

By Definition 2.4 and assumption (3.1) we have that the sequence tJk px, q{k uk1
is bounded in S1,1 and for every P N that there exists a constant c 0 such that
}Jk }2, c k

(3.2)

and

}Jk }2, c 0 .

k1

Notice that from (2.2) we have c0 1. This will be useful in the proof of Theorem
3.10 at the end of the present section.
Example 3.1. A simple realization of a sequence t j u j1 satisfying (3.1) and (2.2) can be
obtained using the phase function pt, s, x, q solving the eikonal equation (2.5). Indeed, it
is sufficient to take a partition
s t`1 t t1 t0 t,
of the interval rs, ts and define
j px, q

pt j1 , t j , x, q 1 j ` 1
x
j ` 2.

In fact, from Proposition 2.9 we know that j P Pr p j q with j c0 pt j1 t j q for

1 j ` 1 and with j 0 for j ` 2. Condition (3.1) is fulfilled if we choose T0
small enough, since
8

j1

`1

c0 pt j1 t j q c0 pt sq c0 T0

j1

1
4

if T0 p4c0 q . Moreover, again from Proposition 2.9, we know that }J j }2,0 c0 |t j

t j1 | j for all 1 j ` 1 and J j 0 for j ` 2, so each one of the J j satisfies (2.2).
With a fixed integer M 1, we denote
pX, q px0 , x1 , . . . , xM , 1 , . . . , M , M`1 q : px, T, , q,
pT, q px1 , . . . , xM , 1 , . . . , M q,
and define the function of 2pM ` 1qn real variables
(3.3)

pX, q :

`
j1

j px j1 , j q x j j ` M`1 pxM , M`1 q.

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

For every fixed px, q P R2n , the critical points pY, Nq pY, Nqpx, q of the function
r q px, T, , q are the solutions to the system
of 2Mn variables pT,
#
1 j pX, q 1j, px j1 , j q x j 0
j 1, . . . , M,
1x j pX, q 1j`1,x px j , j`1 q j 0

j 1, . . . , M,

in the unknowns pT, q. That is pY, Nq pY1 , . . . , YM , N1 , . . . , NM qpx, q satisfies, if

M 1,
#
Y1 11, px, N1 q
(3.4)
N1 12,x pY1 , q,
or, if M 2,

(3.5)

Y1 11, px, N1 q

&Y j 1 pY j1 , N j q,
j,
1

pY j , N j`1 q,
j

j`1,x

%N 1
pY , q.
M

M`1,x

j 2, . . . , M
j 1, . . . , M 1

In the sequel we will only refer to the system (3.5), tacitly meaning (3.4) when
M 1. Definition 3.2 below of the multi product of SG phase functions is analogous

type.
to the one given in [24] for (local) symbols of Hormander
Definition 3.2 (Multi-product of SG phase functions). If, for every fixed px, q P R2n ,
the system (3.5) admits a unique solution pY, Nq pY, Nqpx, q, we define
px, q p1 7 7 M`1 qpx, q : px, Ypx, q, Npx, q, q.

(3.6)

The function is called multi-product of the SG phase functions 1 , . . . , M`1 .

Example 3.3. The simplest case of a well-defined multi-product of SG phase functions is
given by the sharp product 7 0 , where P Pr and 0 px, q x . Indeed, the critical
points pY, Nq of the function
r 1 , 1 q px, x1 , 1 , q px, 1 q x1 1 ` x1
px

are given by pY, Nqpx, q p1 px, q, q. The multi-product 7 0 is so defined by

px, q px, 1 px, q, , q px, q 1 px, qp q px, q.
Similarly, the multi-product 0 7 is well defined. Indeed, the function
r 1 , 1 q px, x1 , 1 , q x 1 x1 1 ` px1 , q
px

has critical points pY, Nqpx, q px, 1x px, qq, and

px, q px, x, 1xpx, q, q px xq 1x px, q ` px, q px, q.

Notice that we have proved here above that for every P Pr the identity
7 0 0 7
holds true. That is, the multi-product of SG phase functions defined in (3.6) admits the
trivial phase function 0 px, q x as identity element.
Example 3.4. A situation where (3.6) is well defined, which is interesting for applications,
see Section 5, is given by the multi-product of solutions to the eikonal equation (2.5) on
different, neighboring time intervals. Indeed, the critical points pY, Nqpx, q of the function
rtsr px1 , 1 q : tsr px, x1 , 1 , q pt, s, x, 1 q x1 1 ` ps, r, x1 , q

are given by
(3.7)

#
1rst,x1 px, x1 , 1 , q 1 ` 1x ps, r, x1 , q 0
1rst,1 px, x1 , 1 , q 1 pt, s, x, 1 q x1 0.

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

The Jacobian matrix with respect to px1 , 1 q of the system (3.7) is

2

xx ps, r, x1 , q
I
Jpt, s, r, x, x1 , 1 , q
,
I
2 pt, s, x, 1 q
where I is the pn nq-dimensional unit matrix. By (2.5), det Jpt, r, r, x, x1 , 1 , q 1.
Thus, taking a small interval r0, T0 s such that det Jpt, s, r, x, x1 , 1 , q 0 for all r, s, t
such that 0 r s t T0 and all pX, q P R4n , by the implicit function theorem
it follows that the system (3.7) admits a unique solution pY, Nqtsr pYtsr , Ntsr qpx, q
pYpt, s, r, x, q, Npt, s, r, x, qq. The multi-product
tsr px, q pt, s, r, x, q pts 7 sr qpx, q tsr px, Ytsr px, q, Ntsr px, q, q
pt, s, x, Ntsrpx, qq Ytsr px, q Ntsr px, q ` ps, r, Ytsr px, q, q
is then well defined. Moreover, it is quite simple to show, in view of to Proposition 2.9, that
the multi-product ts 7 sr satisfies the associative law
ts 7 sr tr ,

(3.8)

0 r s t T0 .

Indeed, pt, s, r, x, q does not depend on s:

d
rpt, s, r, x, qs pBs qpt, s, x, Ntsrpx, qq`1 pt, s, x, Ntsrpx, qqpBs Nqpt, s, r, x, q
ds
pBs Yqpt, s, r, x, q Npt, s, r, x, q Ypt, s, r, x, q pBs Nqpt, s, r, x, q
` pBt qps, r, Ytsr px, q, q ` 1x ps, r, Ytsr px, q, q pBs Yqpt, s, r, x, q 0,
since, by (2.5), (2.6) and the definition (3.7) of the critical point pY, Nqtsr , we have
1x ps, r, Ytsr px, q, q Npt, s, r, x, q,
1 pt, s, x, Ntsrpx, qq Ypt, s, r, x, q,
pBt qps, r, Ytsr px, q, q

aps, Ytsr px, q, 1x ps, r, Ytsr px, q, qq

aps, Ytsr px, q, Ntsr px, qq,

pBs qpt, s, x, Ntsrpx, qq aps, 1 pt, s, x, Ntsrpx, qq, Ntsr px, qq

aps, Ytsr px, q, Ntsr px, xiqq.
This gives, with 0 px, q x ,
pts 7 sr qpx, q pt, s, r, x, q pt, r, r, x, q ptr 7 rr qpx, q ptr 7 0 qpx, q
tr px, q,
by Example 3.3, as claimed.
Now we want to show that under assumption (3.1) the multi-product px, q of
Definition 3.2 is well defined on R2n , and it is a regular SG phase function itself. To
this aim, we switch from the system (3.5) in the unknown pY, Nq to the equivalent
Nq
py1 , . . . , yM , 1 , . . . , M q P R2Mn as follows.
system (3.10) in the unknown pY,
Define
$ 0
z : 0

j
j
&
z : k1 yk , j 1, . . . , M

(3.9)

j : M
j 1, . . . , M

k j k ,

% M`1

: 0,

and then consider the system

#
1
yk Jk,
px ` zk1 , ` k q,
(3.10)
1
k Jk`1,x px ` zk , ` j`1 q,
We have that:

k 1, . . . , M
k 1, . . . , M.

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

Lemma 3.5. For every fixed px, q P R2n , pY, Nqpx, q is a solution of (3.5) if and only if
Nqpx,

pY,
q py1 , . . . , yM , 1 , . . . , m qpx, q, defined by
$
y1 Y1 x

&y Y Y
j 2, . . . , M
j
j
j1
(3.11)

N
j 1, . . . , M 1
j
j
j`1

%
M NM ,
is a solution of (3.10).

Proof. Substituting (3.11) in (3.9), we immediately get the relation

#
Yj x ` zj
(3.12)
N j ` j.
By this, it follows that pY, Nq is a solution of (3.5) if and only if
#
x ` z j 1j, px ` z j1 , ` j q
j 1, . . . , M
` j 1j`1,x px ` z j , ` j`1 q

j 1, . . . , M;

by substituting j px, q J j px, q ` x we obtain

#
z j z j1 J1j, px ` z j1 , ` j q

j 1, . . . , M

j`1

J1j`1,x px

`z ,`

j`1

j 1, . . . , M,

which is exactly (3.10), in view of (3.9).

We are then reduced to prove the following Theorem 3.6.

Theorem 3.6. Under the assumption (3.1), for every fixed px, q P R2n there exists a
Nqpx,
Nq

satisfies
unique solution pY,
q of (3.10). Moreover, the solution pY,
4
4
(3.13)
|k | k`1 xy, k 1, . . . , M,
|yk | k xxy,
3
3
and the functions z j and j in (3.9) satisfy
1
1
xxy,
| j | xy, j 1, . . . , M.
3
3
Remark 3.7. We aim at obtaining a solution pY, Nq such that p., Y, N, ..q P Pr pq.
By Definition 3.2, recalling that a smooth function, it is enough to show that pY, Nq
is of class C8 pR2n q, that Y j P S1,0 , N j P S0,1 , and that xY j px, qy xxy as |x| 8,
xN j px, qy xy as || 8. To get these last equivalences, it is sufficient to prove the
existence of a constant k P p0, 1q such that |Y j px, q x| kxxy and |N j px, q | kxy.
Indeed, the following implication holds:
(3.14)

(3.15)

|z j |

|b| kxay, k P p0, 1q, a, b P Rn p1 kqxay xa ` by p1 ` kqxay.

Formula (3.14) gives precisely the desired estimates, with k 1{3, owing to (3.12).
Theorem 3.6 then ensures that the multi-product is well-defined. We show that pY, Nq P
C8 pR2n q in the subsequent Theorem 3.8.
Proof of Theorem 3.6. We divide the proof into two steps. In step one we suppose
Nq
of (3.10) and prove that such solution satisfies
the existence of a solution pY,
(3.13) and that (3.14) holds. In step two we show, by a fixed point argument, the
Nq.

existence and uniqueness of the solution pY,

Nq
is a solution of (3.10), then by (3.10) and (2.2) we get, for any
Step 1. If pY,
px, q P R2n ,
#
|yk | k xx ` zk1 y
|k | k`1 x ` k`1 y

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

10

for k 1, . . . , M. Now, using the inequality

xx ` yy xxy ` |y| @x, y P Rn

(3.16)

and definition (3.9), we get, for k 1, . . . , M and any px, q P R2n ,

$

|yk | k xxy ` |zk1 | k xxy `

|y j |,

&
j1

(3.17)
M

|k | k`1 xy ` |k`1 | k`1 xy `

| j |,

%
j1

so that

(3.18)

M
M
M
M

k xxy `
|yk | : M xxy `
|yk | ,

& |yk |
k1
k1
k1

k1

M
M
M
M

|k | : M`1 xy `
k`1 xy `
|k | .
% |k |
k1

k1

k1

k1

The two inequalities here above are of the form pxxy ` q with 0 1{4
by assumption (3.1), so they give

1
xxy xxy,
1
3

and, coming back to (3.18), we have, for any px, q P R2n ,

|zk |

j1

|y j |

1
xxy,
3

|k |

| j |

jk

1
xy,
3

that is (3.14). Substituting in (3.17) we obtain

1
1
4
4
|yk | k xxy ` xxy k xxy,
|k | k`1 xy ` xy k`1 xy,
3
3
3
3
that is (3.14).
Nq
of (3.10) satisfies (3.14) for
Step 2. Since we have shown that every solution pY,
2n
any px, q P R , to show existence and uniqueness of a solution to (3.10) in R2Mn it
Nq
in the space
is sufficient to show existence and uniqueness of pY,
+
#
M
M

1
1
2Mn
|k | xy ,
x, : py1 , . . . , yM , 1 , . . . , M q P R
:
|yk | xxy,
3
3
k1
k1
px, q P R2n , which is a metric space with norm
}py1 , . . . , yM , 1 , . . . , M q} :

k1

xxy1 |yk | ` xy1 |k | .

We define the map

T Tx, :
by Tpy1 , . . . , yM , 1 , . . . , M q : pw1 , . . . , wM , 1 , . . . , M q, where, for k 1, . . . , M,
px, q P R2n ,
#
1
wk Jk,
px ` zk1 , ` k q
1
k Jk`1,x
px ` zk , ` k`1 q.

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

11

The map T is well defined. Indeed, by (2.2), (3.16) and (3.14) we have, for any
px, q P R2n ,
#

(3.19)

|wk | k xx ` zk1 y k pxxy ` 31 xxyq 43 k xxy

|k | k`1 x ` k`1 y k`1 pxy ` 31 xyq 34 k`1 xy,

so that
M

k1

|wk |

4
1
xxy
k xxy,
3
3
k1

and

|k |

k1

4
1
xy
k`1 xy.
3
3
k1

Nq
Nqpx,
pY,

By (3.10), to show existence and uniqueness of pY,

q is equivalent
Nq
of the map T. We show
to show existence and uniqueness of a fixed point pY,
here below that, under assumption (3.1), T is a contraction on , so it admits a
Nq.

unique fixed point pY,

Let us consider two arbitrary points
r Nq
r pr
pY, Nq py1 , . . . , yM , 1 , . . . , M q, pY,
y1 , . . . , r
yM , r1 , . . . , rM q P ,

and let

r Nq
r pw
r1 , . . . , w
rM ,
r1 , . . . ,
rM q.
TpY, Nq pw1 , . . . , wM , 1 , . . . , M q, TpY,

For every fixed k 1, . . . , M, px, q P R2n , we have

1
1
rk wk Jk,
px ` zk1 , ` k q
px ` r
zk1 , ` rk q Jk,
w
1
2
pr
zk1 zk1 q
Jk,x
px ` zk1 ` pr
zk1 zk1 q, ` k qd

` prk k q

1
0

and from (2.2) we get

r k wk | k
|w

k1

|r
z

k1

2
Jk,
px ` zk1 , ` k ` prk k qqd

1
k
k
k 1
k
k
k1
r
r
| ` | |xx ` z y x ` ` p qy d .
0

By inequality (3.15) with b zk and k 1{3 we get 32 xxy xx ` zk y 34 xxy; the same
inequality with b k ` prk k q and k 1{3 gives 23 xy x ` k ` prk k qy
4
r
3 xy; substituting these inequalities into the estimate of |wk wk | we come to
r k wk | k
|w

1
1
|r
zk1 zk1 | ` |rk k |2xxy xy d

j1

|r
y j y j | ` |r
j j |2xxyxy1 .

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

12

Similarly:

2
rk k | |r
|
zk zk | Jk`1,xx
px ` zk ` pr
zk zk q, ` k`1 qd
0

k`1
k`1
k
k`1
k`1
k`1
2
r
r
` |
| Jk`1,x, px ` z , `
` p
qqd
0

zk zk |2xxy1 xy ` |rk`1 k`1 |

k`1 |r
k`1

|r
y j y j |2xxy1xy ` |r
k`1 k`1 | .
j1

Thus

r Nq}
r
}TpY, Nq TpY,

r k wk | ` xy1 |
rk k |
xxy1 |w

k1
M

k1

`k`1

xxy1 |r
y j y j | ` 2xy1 |r
j j|
j1

j1

k1

|r
y j y j |2xxy1 ` |r
j j |xy1

maxtk , k`1 u3

|r
y j y j |xxy1 ` |r
j j |xy1
j1

r Nq}
r .
30 }pY, Nq pY,

This shows that the map T is Lipschitz continuous, with Lispchitz constant 30 1.
It follows that T is a strict contraction on , which then admits a unique fixed point
Nq
P , for any px, q P R2n . Such fixed point obviously gives the unique
pY,
solution of (3.10). The proof is complete.

Nq
Nqpx,
pY,

Theorem 3.8. The unique solution pY,

q of (3.10) is of class C8 pR2n q.
Proof. For pY, Nq P R2Mn and px, q P R2n , we define the function
FpY, N; x, q : pF1 , . . . , FM qpY, N; x, q,
with values in R2M , where for all k 1, . . . , M,

1
1
px ` zk , ` k`1 q .
px ` zk1 , ` k q, k Jk`1,x
Fk pY, N; x, q : yk Jk,

We apply the implicit function Theorem to the function F, which is clearly of class
C8 with respect to all variables, being Jk a C8 function for all k 1, . . . , M. For
every fixed px, q we have that
Nqpx,

FppY,
q; x, q 0,
Nq
is the solution of (3.10). Moreover, we are going to prove here below
since pY,
that

BF

(3.20)
ppY, Nqpx, q; x, q , 0.
det
BpY, Nq
Nqpx,

This means that the implicitly defined function pY,

q has the same regularity
8
2n
as F, so it is of class C pR q. To complete the proof, it remains only to show that
(3.20) holds true.

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

13

BF
Let us compute the entries of the 2M 2M matrix BpY,Nq
pY, N; x, q. For every fixed
k 1, . . . , M, px, q P R2n , we have

$
2
k1
k
2
k
k`1

J
px
`
z
,

q,
J
px
`
z
,

q
,1 j k 1

k`1,xx
& k,x

1
2
k
k`1
Fk,y jpY, N; x, q 1, J
px ` z , ` q ,
jk
k`1,xx

%
p0, 0q,
k ` 1 j M,

and

$
p0, 0q,
1 jk1

&
2
k1
k
1
jk
Fk, jpY, N; x, q Jk, px ` z , ` q, 1 ,

% J2 px ` zk1 , ` k q, J2
k
k`1
px ` z , ` q , k ` 1 j M,
k,
k`1,x

so we can write

BF
pY, N; x, q
BpY, Nq

I H11 pY, N; x, q H12 pY, N; x, q

H21 pY, N; x, q I H22 pY, N; x, q

where I stands for the identity M M matrix, and

2
0
0
0
J1,
2

..

0
J2,x
. 0
0
, H1,2
H1,1

..
..
...

.
.
2
2
0
JM,x
JM,x
0

H2,1

2
J2,xx
2
J3,xx
..
.
2
JM`1,xx

0
2
J3,xx

..
.

0
0
..
.
2
JM`1,xx

H2,2

2
J2,
..
.

..
.
0

2
J2,x

0
..
.

0
..
.

..
.

Let us estimate the matrix norm of each one of the Hi j :

}H11 pY, N; x, q} max

j1,...,M

}H12 pY, N; x, q} max

j1,...,M

}H21 pY, N; x, q} max

j1,...,M

}H22 pY, N; x, q} max

j1,...,M

|ph11 qi j | max

|ph12 qi j | max

j1,...,M

i1

j1,...,M

i1
M

|ph21 qi j | max

|ph22 qi j | max

j1,...,M

i1

j1,...,M

i1

i j`1
j

2
J1,
2
J2,
..
.
2
JM,

2
J2,x
..
.
.

2
JM,x
0

j1

i xx ` zi1 yx ` i y1

i1
M

i`1 xx ` zi y1 x ` i`1 y

i j

j1

i1

i`1

j.

j1

Nqpx,

With the choice pY, Nq pY,

q these estimates become, via formula (3.12)
and Remark 3.7,
Nqpx,

}H11 ppY,
q; x, q}

j,

Nqpx,

}H12 ppY,
q; x, q} 2xxyxy1

i ,

i1

j1

Nqpx,

}H21 ppY,
q; x, q} 2xxy1xy

i1

i ,

Nqpx,

}H22 ppY,
q; x, q}

j1

j.

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

14

Now, since detpI H11 q 1, being H11 triangular with null diagonal, we have
BF
Nqpx,

ppY,
q; x, q
BpY, Nq

Nqpx,
Nqpx,

I H11 ppY,
q; x, q
xyxxy1 H12 ppY,
q; x, q
det
Nqpx,
Nqpx,

xxyxy1 H21 ppY,

q; x, q
I H22 ppY,
q; x, q

H11 ppY, Nqpx, q; x, q

xyxxy H12 ppY, Nqpx,
q; x, q
det I
Nqpx,
Nqpx,

xxyxy1 H21 ppY,

q; x, q
H22 ppY,
q; x, q

det

detpI Apx, qq,

with
Nqpx,
Nqpx,

}Apx, q} maxt}H11 ppY,

q; x, q} ` }xxyxy1 H21 ppY,
q; x, q},
1
Nqpx,
Nqpx,

}H22 ppY,
q; x, q} ` }xyxxy H12 ppY,
q; x, q}u
3

j 30

j1

3
,
4

and applying Proposition 3.9 below, cfr. [24], we get detpI Apx, qq 42M 0.
That is, (3.20) holds true, and the proof is complete.

Proposition 3.9 (Proposition 5.3, page 336 in [24]). Let A pai j q1i, j be a real
matrix and suppose that there exists a constant c0 P r0, 1q such that
}A} : max

j1,...,

Then,

|ai j | c0 .

i1

p1 c0 q detpI Aq p1 ` c0 q .
The following theorem gives crucial estimates of the unique C8 solution pY, Nq
of (3.5).
Theorem 3.10. Under the assumptions (3.1) and (2.2), the unique C8 solution pY, Nqpx, q
of (3.5) satisfies:

(3.21)

|B Bx pY j Y j1 qpx, q| c, j xy|| xxy1||,

(3.22)

|B Bx pN j N j`1 qpx, q| c, j`1 xy1|| xxy||,

for all , P Zn` , j 1, . . . , M, x, P Rn , with constants c, not depending on j and M.

Moreover,
(3.23)

tpY j Y j1 qpx, q{ j u j1 is bounded in S0,1 ,

(3.24)

tpN j N j`1 qpx, q{ j`1 u j1 is bounded in S1,0 .

Proof. Estimates (3.21), (3.22) in the case 0 have already been proved, see
(3.13) and (3.11). To prove the same estimates for | ` | 1, it is sufficient, by
Nqpx,

(3.13), (3.11) and (3.2), to show that the solution pY,

q of (3.10) is such that

(3.25)

|B Bx yk px, q| c, }Jk }2,|`|1xy|| xxy1||,

(3.26)

|B Bx k px, q| c, }Jk`1 }2,|`|1xy1|| xxy||,

for | ` | 1, k 1, . . . , M, x, P Rn . Estimates (3.25), (3.26) are going to be

proved by induction on N | ` |.
Step N 1. We need to check (3.25), (3.26) for the derivatives of order 1. Let us start

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

15

with the derivatives with respect to x. By definition (3.10) of yk , k , k 1, . . . , M,

x, P Rn , we have
#
2
2
y1 Jk,x
p. ` zk1 , .. ` k qp1 ` pzk1 q1x q ` Jk,
p. ` zk1 , .. ` k qpk q1x
(3.27) 1k,x
2
2
k,x Jk`1,xx
p. ` zk , .. ` k`1 qp1 ` pzk q1x q ` Jk`1,x
p. ` zk , .. ` k`1 qpk`1 q1x .
By (2.2), setting hpx, q xxyxy1 , we obtain
!
)
}y1k,x } ` h }1k,x } k 1 ` }pzk1 q1x } ` x. ` zk1 yx.. ` k y1 }pk q1x }
!
)
` k`1 h x. ` zk y1 x.. ` k`1 yp1 ` }pzk q1x }q ` }pk`1 q1x } ;
from (3.14) we have 23 xxy xx ` zk1 y 34 xxy and 23 xy x ` k y 43 xy, so we
come to
}y1k,x } ` h }1k,x } k 1 ` }pzk1 q1x } ` 2 h }pk q1x }

(
` k`1 2 ` 2}pzk q1x } ` h }pk`1 q1x }
#
+
M
M

1
1
k 1 `
}yk,x } ` 2 h
}k,x }
k1

` k`1 2 ` 2

k1

}y1k,x }

`h

}1k,x }

k1

k1

where we have used also definition (3.9). Summing for k 1, . . . , M, we get, for
any x, P Rn ,
M

}y1k,x } ` h }1k,x } M

k1

1`
#

}y1k,x } ` 2 h

k1

` M`1 2 ` 2
#

3 M`1 1 `

}1k,x }

k1
M

}y1k,x }

`h

k1
M

k1

}1k,x }

k1

}y1k,x }

`h

}1k,x }

This last inequality immediately gives

(3.28)

k1

}y1k,x } ` h }1k,x }

3M`1
30

1 3M`1
1 30

with 1 30 1 3{4 1{4 0, so that the amount (3.28) is finite (bounded by

3). Coming back to (3.27) and substituting there the estimate here above we get
)
!
}y1k,x } }Jk }2,0 1 ` }pzk1 q1x } ` 2 h }1k,x }
#
+
M

2}Jk }2,0 1 `
}y1k,x } ` h }1k,x }

2}Jk }2,0 1 `

k1

30
1 30

: c0,1 }Jk }2,0 ,

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

16

that is (3.25) with 0 and || 1. With similar computations we obtain

`

}1k,x px, q} }Jk`1 }2,0 x. ` zk y1 x. ` k`1 yp1 ` }pzk q1x }q ` }pk`1 q1x } px, q
`

}Jk`1 }2,0 2 h1 p1 ` }pzk q1x }q ` }pk`1 q1x } px, q

2}Jk`1 }2,0 h1 1 ` }pzk q1x } ` h }pk`1 q1x } px, q

2}Jk`1 }2,0 h1 1 `
}y1k,x } ` h }1k,x }
px, q
k1

2xxy xy}Jk`1 }2,0 1 `

1

C0,1 }Jk`1 }2,0 xxy1 xy,

30
1 30

x, P Rn ,

and also
}y1k, px, q} C1,0 }Jk }2,0 xxyxy1 ,

}1k, px, q} C1,0 }Jk`1 }2,0 ,

x, P Rn .

The step N 1 is complete.

Step N { N ` 1. Let us now suppose that (3.25), (3.26) hold for 1 | ` | N,
N 1, x, P Rn , and prove the same estimates for | ` | N ` 1. If we substitute
(3.2) into (3.25), (3.26) we immediately get

(3.29)

|B Bx yk px, q| c1, xy|| xxy1||,

(3.30)

|B Bx k px, q| c1, xy1|| xxy||,

for 1 | ` | N and k 1, . . . , M. These estimates are going to be used to

bound the derivatives Bx B with | ` | N of the functions y1k,x , y1k, , 1k,x , 1k, (i.e.

the derivatives Bx B with | ` | N ` 1 of the functions yk , k ). Let us start by

computing, from (3.27), the derivative

2
Bx B y1k,x Bx B Jk,x
(3.31)
p. ` zk1 , .. ` k q 1 ` pzk1 q1x

2
` Bx B Jk,
p. ` zk1 , .. ` k q pk q1x .

To obtain an estimate of (3.31), we use Faa di Bruno formula, write the derivatives
of zk and k as derivatives with respect to yk and k by (3.9), and finally we apply
(3.29), (3.30), obtaining

2
|Bx B Jk,x
px ` zk1 px, q, ` k px, qq |

Cq,r,,}Jk }2,q`r xyq xxyr

1 ``r 1 ``q
i ,0
i ,0

xyp1|1 |q``p1|q |q xxyp1|1|q``p1|r |q

C, }Jk }2,|`| xy|| xxy||
and

2
|Bx B Jk,
px ` zk1 px, q, ` k px, qq | C, }Jk }2,|`|xy1|| xxy1||.

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

17

Thus, coming back to (3.31), substituting these last two estimates and using (3.9)
we come to
M

|Bx B y1j,x px, q| ` 2xxyxy1 |Bx B 1j,x px, q|

|Bx B y1k,x px, q| }Jk }2,0
j1

C1, }Jk }2,|`|

}Jk }2,0

j1

xy|| xxy|| ` C2, xy1|| xxy1||xyxxy1

|Bx B y1j,x px, q| ` 2xxyxy1 |Bx B 1j,x px, q|

r, }Jk }2,|`| xy|| xxy||.

`C

(3.32)

Working similarly on the terms Bx B 1k,x coming from the derivatives in (3.27), we
get the corresponding estimate:

|Bx B 1k,x px, q|

}Jk`1 }2,0 xxy

xy

j1

|Bx B y1j,x px, q| ` 2xxyxy1 |Bx B 1j,x px, q|

r1 }Jk`1 }2,|`|xy1|| xxy1||.

`C
,

(3.33)

Now summing up for k 1, . . . , M inequalities (3.32) and (3.33) we have

M

|Bx B y1k,x px, q| ` 2xxyxy1 |Bx B 1k,x px, q|

k1

k1

`C ,
3c0 0

}Jk }2,0 ` 2

}Jk`1 }2,0

k1
M

}Jk }2,|`| ` 2

k1

k1

k1

k1

|Bx B y1k,x px, q| ` 2xxyxy1 |Bx B 1k,x px, q|

}Jk`1 }2,|`| xy|| xxy||

|Bx B y1k,x px, q| ` 2xxyxy1 |Bx B 1k,x px, q| ` 3c|`|0 C , xy|| xxy||,

where c0 , c|`| are the constants defined in (3.2). In particular, notice that, by (2.2),
we have c0 1. From this, we finally obtain
M

k1

|Bx B y1k,x px, q| ` 2xxyxy1 |Bx B 1k,x px, q| C 1,

0
xy|| xxy||
1 30

C 1, xy|| xxy||

(3.34)

by the choice of 0 in (3.1). Substituting (3.34) in (3.32) and (3.33) we get

(3.35)

|Bx B y1k,x px, q| C, }Jk }2,|`|xy|| xxy||

(3.36)

|Bx B 1k,x px, q| C, }Jk`1 }2,|`|xy1|| xxy1||.

All the computations from (3.31) to (3.36) on the functions y1k,x and 1k,x can be
repeated on the functions y1k, and 1k, with minor changes. In this way we finally
obtain the estimates corresponding to (3.35) and (3.36), namely

(3.37)

|Bx B y1k, px, q| C, }Jk }2,|`| xy1|| xxy1||

(3.38)

|Bx B 1k, px, q| C, }Jk`1 }2,|`| xy|| xxy||.

The proof is complete, since (3.35)-(3.38) are the desired estimates (3.25) and (3.26)
for all the derivatives of order N ` 1 of the functions yk and k .


FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

18

We conclude with a Theorem that summarizes what we have proved throughout

the present section, and gives the main properties of the multi-products of regular
SG phase functions.
Theorem 3.11. Under assumptions (3.1) and (2.2), the multi-product px, q of Definition
3.2 is well defined for every M 1 and has the following properties.
(1) There exists k 1 such that px, q p1 7 7 M`1 qpx, q P Pr pkM`1 q and,
setting
JM`1 px, q : p1 7 7 M`1 qpx, q x ,
the sequence tJM`1 { M`1 uM1 is bounded in S1,1 pR2n q.
(2) The following relations hold:
#
1x px, q 11,x px, N1 px, qq
1 px, q 1M`1, pYM px, q, q,
where pY, Nq is the critical point (3.5).
(3) The associative law holds: 1 7 p2 7 7 M`1 q p1 7 7 M q 7 M`1 .
(4) For any 0 there exist 0 1{4 and c 1 such that, if j P Pr p j , q
for all j and 0 , then P Pr pc M`1 , q.
Proof. By theorems 3.6 and 3.8 we know that, for any M 1, is a well-defined
smooth function on R2n . We start by showing (1). We write, with Y0 px, q
YM`1 px, q : x, NM`1 px, q : ,
JM`1 px, q

j pY j1 px, q, N j px, qq Y j px, q N j px, q

j1

`M`1 pYM px, q, q x

M`1

j1

M`1

j1

j pY j1 , N j q Y j N j px, q

J j pY j1 , N j q pY j Y j1 q N j px, q.

This gives that

M`1
j J j pY j1 , N j q Y j Y j1
JM`1

N j is bounded in S1,1
M`1

M`1
j
j
j1
since tJ j { j u j1 is bounded in S1,1 , (3.23) holds, and xN j px, qy xy. Now, the
boundedness proved here above implies the existence of a positive constant k such
that
}JM`1 }2 kM`1 k0 ,

(3.39)

and taking 0 small enough, so that k0 1, we obtain that P Pr pkM`1 q.

Statement (1) is proved. Statement (4) immediately follows. Indeed, if j P
Pr p j , q, then we have }JM`1 } pk ` 1qM`1 , with k coming from (3.39), and we
obtain }JM`1 } c M`1 and P Pr pc M`1 , q if we choose c such that c 0 1.
Let us now come to (2), which is quite simple. Indeed, from (3.2) and (3.5), we
have
px, q :

j pY j1 px, q, N j px, qq Y j px, q N j px, q ` M`1 pYM px, q, q.

j1

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

19

A derivation of the expression above with respect to x and the use of (3.5) give
1x px, q

1j,x pY j1 px, q, N j px, qq Y1j1,x px, q

j1

`1j,pY j1 px, q, N j px, qq N1j,x px, q

Y1j,x px, q N j px, q Y j px, q N1j,x px, q

1
` 1M`1,x pYM px, q, qYM,x
px, q

1
11,x px, N1 px, qq Y1,x
px, q N1 px, q
M

1
`
pY1j1,x N j1 Y1j,x N j q ` NM YM,x
px, q
j2

11,x px, N1 px, qq,

which is exactly the first equality in (2). The second equality can be obtained
similarly, by derivation with respect to of px, q.
Finally, we deal with (3). We want to show that
p1 72 7 7 M q 7 M`1 1 7 7 M`1 .

(3.40)

To this aim, let us denote

r : 1 7 7 M ,

and compute by (3.3), with M 1, the product

r 7 M`1 qpx, q px,

r Npx,
r q Npx,
r q, q,
r qq Ypx,
r q ` M`1 pYpx,
(3.41) p

r Nq
r Nqpx,
r pY,
r
where pY,
q is the 2ndimensional critical point given by
#
r
r1 px, Nq,
r
Y

(3.42)
r q.
r 1
N
pY,
M`1,x

r 7 M`1 is well-defined by (1) (eventually, with a smaller 0 ). Now,

Notice that
r Npx,
r qq p1 7 7 M qpx, Npx,
r qq in (3.41), using
we compute the value of px,
r
(3.3) with M 1 in place of M and N in place of , obtaining
r Npx,
r qq
px,
M1

r qq, N j px, Npx,
r qqq Y j px, Npx,
r qq N j px, Npx,
r qq
j pY j1 px, Npx,
(3.43)
j1

r qq, Npx,
r qq,
` M pY M1 px, Npx,

Nq
given by
with the 2pM 1qndimensional critical point pY,
$

Y 0 x

&Y j 1 pY j1 , N j q
j 1, . . . , M 1
j,
(3.44)
1
j`1 q j 1, . . . , M 1

N j j`1,x pY j , N

%N N,
M

r in place of . Moreover, we
obtained from (3.5), with M 1 in place of M and N
have from (3.42) and (2), with M 1 in place of M, that
(3.45)

r q
r1 px, Npx,
r qq p1 7 7 M q1 px, Npx,
r qq
Ypx,

r qq, Npx,
r qq.
1M, pY M1 px, Npx,

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

20

Summing up, from (3.45), the second equation in (3.42), and (3.44), we have that
r N
r solves system (3.5), and thus it is the 2Mndi 1 , , N M1 , Nq
pY 1 , , Y M1 , Y,
mensional critical point needed to define the multi-product 1 7 7 M`1 , which
turns out to be given, in view of (3.3), by
p1 7 7 M`1 qpx, q
M1

r qq, N j px, Npx,
r qqq Y j px, Npx,
r qq N j px, Npx,
r qq

j pY j1 px, Npx,
j1

r q, q.
r qq, Npx,
r qq Ypx,
q Npx,
q ` M`1 pYpx,
` M pY M1 px, Npx,

We observe that this last expression coincides with (3.41) after substituting (3.43)
r 7 M`1 , that is (3.40). Similarly, we can
in it. This gives that 1 7 7 M`1
prove the corresponding law 1 7 p2 7 7 M`1 q 1 7 7 M`1 , completing
the proof of (3).

4. Composition of SG Fourier integral operators
We can now prove our main theorem on compositions of regular SG FIOs. We
start with an invertibility result for I Op p1q and I Op p1q when is a
regular phase function. Theorem 4.1 below gives more precise versions of (2.3),
(2.4), with a slight additional restriction on , for FIOs with constant, nonvanishing
symbol.
Theorem 4.1. Assume that P Pr pq with 0
exists q P S0,0 pR2n q such that

1
4

sufficiently small. Then, there

(4.1)

I Op pqq Op pqq I I,

(4.2)

I Op pqq Op pqq I I.

Moreover, if the family of SG phase functions ts px, qu is such that the family tJs px, qu
ts px, q x u is bounded in S1,1 , then the corresponding family tqs u is also bounded in
S0,0 .
Proof. For u P SpRn q we have, by definition of type I and type II SG FIOs,
"
(4.3)
ppI I quqpxq p2qn
eippx,qpy,qq upyq dyd.
The map
x,y : x,y pq px, y, q

1
0

1x px ` tpy xq, q dt

is globally invertible on Rn . In fact, its Jacobian is given by the matrix

1
1
2
2x px ` tpy xq, q dt I `
Jx
px ` tpy xq, q dt
0

which has nonvanishing determinant, in view of the hypothesis P Pr pq, 0

1
4 . Moreover, condition (2) in Definition 2.2 implies that is coercive, and these
two properties give its global invertibility on Rn , see [11, Theorems 11 and 12] and
the references quoted therein. Finally, x,y is also a SG diffeomorphism with 0-order
parameter-dependence, that is both px, y, q and 1 px, y, q belong to S0,0,1 pR3n q, the
space of SG amplitudes of order p0, 0, 1q, see [10, 11], and satisfy xpx, y, qy xy,
x1 px, y, qy xy, uniformly with respect to x, y P Rn . In (4.3) we can then
change variable, setting
px, y, q 1 px, y, q,

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

and obtain
ppI

I quqpxq

upxq ` p2q

"

21

eipxyq a0 px, y, qupyq dyd ppI ` A0 quqpxq,

with
2
a0 px, y, q detpI ` Jx
px, y, qq1 |1 px,y,q 1.

By the results on composition of SG functions in [11, 22], we find that a0 P S0,0,0 pR3d q,
the space of SG-amplitudes of order p0, 0, 0q. Since the seminorms of a0 can be
controlled by means of the parameter , and the map associating a0 with the symbol
a P S0,0 such that A0 Oppaq is continuous, the same holds for the seminorms of
a. By general arguments, see [10, 24, 30, 31], it turns out that pI ` Oppaqq1 exists
in OppS0,0 q. Then, setting Q I pI ` Oppaqq1 , using Theorem 2.3 we find
Q Op pqq for some q P S0,0 and I Op pqq I, which is the first part of (4.1).
The remaining statements follow by arguments analogous to those used in the

proof of [24, Theorem 6.1].
The next Theorem 4.2 is one of our main results.
Theorem 4.2. Let j P Pr p j q, j 1, 2, be such that 0 1 ` 2
sufficiently small 0. Then, there exists p P S0,0 pR2n q such that
(4.4)

I1 I2 Op1 72 ppq,

(4.5)

I2 I1 Op1 72 ppq.

1
4

for some

Moreover, if the families of SG phase functions t js px, qu, j 1, 2, are such that the
families tJ js px, qu t js px, qxu are bounded in S1,1 , j 1, 2, then, the corresponding
family tps px, qu is also bounded in S0,0 .
We will achieve the proof of Theorem 4.2 through various intermediate results,
adapting the analogous scheme in [24]. Before getting to that, let us first state and
prove our main Theorem 4.3, which is obtained as a consequence of Theorems 4.1
and 4.2.
Theorem 4.3. Let j P Pr p j q, j 1, 2, . . . , M, M 2, be such that 1 ` `M
for some sufficiently small 0, and set

1
4

0 px, q x ,
1 1 ,
j 1 7 7 j , j 2, . . . , M
M, j j 7 j`1 7 7M , j 1, . . . , M 1,
M,M M ,
M,M`1 px, q x .
Assume also a j P Sm j , j pR2n q, and set A j Op j pa j q, j 1, . . . , M. Then, the following
holds true.
(1) Given q j , qM, j P S0,0 pR2n q, j 1, . . . , M, such that
Op j pq j q I j I,

IM,j OpM,j pqM, j q I,

set Qj Op j pq j q, QM, j OpM,j pqM, j q, and

R j I j1 A j Qj ,

RM, j QM, j A j IM,j`1 ,

j 1, . . . , M.

Then, R j , RM, j P OppS0,0 pR2n qq, j 1, . . . , M, and

(4.6)

A A1 AM R1 RM IM IM,1 RM,1 RM,M .

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

22

(2) There exists a P Sm, pR2n q, m m1 ` ` mM , 1 ` ` M such that,

setting 1 7 7M ,
A A1 AM Op paq.
(3) For any l P Z` there exist l1 P Z` , Cl 0 such that
m,

~a~l

(4.7)

Cl

m ,

~a j ~l1 j j .

j1

Proof. The existence of q j , qM, j P S0,0 , j 1, . . . , M, with the desired properties

follows from Theorem 4.1. We also notice that, trivially, I0 IM,M`1 I, so that,
inserting either I Q1 I1 QM IM or I IM,1 QM,1 IM,M QM,M ,
we indeed find
A1 AM I0 A1 Q1 I1 A2 IM AM QM IM
R1 RM IM
IM,1 QM,1 A1 IM,2 QM,2 A2 IM,M QM,M AM IM,M`1
IM,1 RM,1 RM,M ,
as claimed. Now, we observe that, again in view of Theorem 4.1, there exists
p j P S0,0 such that I j Op j pp j q I, j 1, . . . , M. Setting Pj Op j pp j q, and
inserting it into the definition of R j , by Theorem 4.2 we then find, for j 1, . . . , M,
R j pI j1 I j q pPj A j q Qj I j1 7 j pPj A j q Qj I j pPj A j q Qj .
Theorem 2.7 implies that Pj A j P OppSm j , j q, and Theorem 2.3 then implies that
pPj A j q Qj Op j pd j q, for some d j P Sm j , j , j 1, . . . , M. Another application
of Theorem 2.3 gives that
R j I j Op j pd j q P OppSm j , j q, j 1, . . . , M,
so that the standard composition rules for SG pseudodifferential operators and
a further application of Theorem 2.3 imply, for 1 7 7M and a suitable
a P Sm, ,
A A1 AM Op paq,
as claimed. Similar considerations hold for RM, j , j 1, . . . , M and the representation
formula
A1 AM IM,1 RM,1 RM,M .
The estimate (4.7) follows from the composition results in [11], applied repeatedly
to (4.6), observing that the amplitudes of the resulting operators depend continuously on those of the involved factors. The proof is complete.

To start proving Theorem 4.2, with two SG phase functions 1 , 2 as in the
corresponding hypotheses and u P SpRn q, let us write, as it is possible,
$
1
1 1
1
rpI1 I2 quspxq
eip1 px, qx `2 px ,qq uppq d1 dx1 d.
Now, with 1 72 , set

(4.8)

0 px, x1 , 1 , q 1 px, 1 q x1 1 ` 2 px1 , q px, q,

and consider, in the sense of oscillatory integrals,

"
1 1
(4.9)
ppx, q
ei0 px,x , ,q d1 dx1 .

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

Then, we can write

rpI1 I2 quspxq

ppq d,
eipx,q ppx, q u

23

u P SpRn q,

which gives the desired claim, if we show that (4.9) indeed defines a symbol
p P S0,0 pR2n q. Let us now define the adapted cut-off functions which will be
needed for the proof of this fact.
Definition 4.4. We set
px, x1, 1 , q a px, x1 q a p, 1 q,
where, with a 0 to be fixed later and w, w1 P Rn , we assume
a pw, w1 q papw w1 qxwy1 q,
for a fixed cut-off function P C8
pRn q. In particular, we also assume that, for all w P Rn ,
0
0 pwq 1, supp B 2 p0q, |B 1 p0q 1, w < B 1 p0q 0 pwq 1, where
3

Br pw0 q is the closed ball in Rn centred at w0 with radius r 0.

For the proof of the next lemma see, e.g., [11].
Lemma 4.5. i) For any multiindeces 1 , 2 P Zn` , the function a pw, w1 q introduced in
Definition 4.4 satisfies, for all w, w1 P Rn ,
`2

|Bw11

(4.10)

a pw, w1 q| . xwy|1 | xw1 y|2 | .

ii) For any multiindeces 1 2 , 1 , 2 P Zn` , the function px, x1 , 1 , q introduced in

Definition 4.4 satisfies, for all x, x1 , , 1 , the estimates
(4.11)

`2

|Bx11 `2 B11

px, x1 , 1 , q| . xxy|1 | xx1 y|2 | xy|1 | x1 y|2 | .

Remark 4.6. In view of Definition 4.4,

1 px, x1 , 1 , q 1 a px, x1 q ` a px, x1 q a px, x1 q a p, 1 q
1 a px, x1 q ` a px, x1 q p1 a p, 1 qq,
which implies that on suppp1 px, x1, 1 , qq either |x x1 |

1
1
xxy or | 1 | xy.
2a
2a

Now write p in (4.9) as p p0 ` p8 with

"
1 1
(4.12)
p0 px, q
ei0 px,x , ,q px, x1 , 1 , q d1 dx1 ,
"
1 1
(4.13)
p8 px, q
ei0 px,x , ,q p1 px, x1 , 1 , qq d1 dx1 .
We analyze separately p0 and p8 .
Proposition 4.7. Under the hypotheses of Theorem 4.2, for p8 defined in (4.13) we have
p8 P S8,8 pR2n q.
Proof. Define
8 px, x1 , 1 , q 1 px, 1 q x1 1 ` 2 px1 , q x ,
so we have from (4.8)
0 px, x1 , 1 , q 8 px, x1 , 1 , q ` x px, q
and

p8 px, q eiJpx,q p18 px, q,

where we have set Jpx, q px, q x and
"
1 1
r
p8 px, q
ei8 px,x , ,q p1 px, x1 , 1 , qq d1 dx1 .

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

24

It is straightforward, since J P S1,1 for small 0, that it is enough to prove that

pr8 P S8,8 to achieve the desired result. Also, in view of the definition of 8 ,
1
18,x px, x1 , 1 , q 1 ` J1,x
px, 1 q,

1
18,1 px, x1 , 1 , q x x1 ` J1,
px, 1 q,

1
18,x1 px, x1 , 1 , q 1 ` J2,x
px1 , q,
1
18, px, x1 , 1 , q x1 x ` J2,
px1 , q.

Then, on suppp1 px, x1 , 1 , qq, for a known c 0 and a sufficiently small 0,

depending on 1 , 2 , and , there exist suitable k1 , k2 0, such that either
|18,x1 px, 1 , x1 , q| | 1 | cxy | 1 | c| 1 | p1 cq| 1 |
k1 pxy ` x1 yq 0,
or
|18,1 px, 1 , x1 , q| |x x1 | cxxy |x x1 | c|x x1 | p1 cq|x x1 |
k2 pxxy ` xx1 yq 0.
Let us set, for b 2a 0,
(4.14)
"
1 1
pr18 px, q
ei8 px, ,x ,q p1 px, x1 , 1 , qq b px, x1 q d1 dx1 ,
(4.15)

pr28 px, q

(4.16)

pr38 px, q

"

ei8 px, ,x ,q p1 px, x1 , 1 , qq p1 b px, x1 qq b p, 1 q d1 dx1 ,

"

ei8 px, ,x ,q p1 px, x1 , 1 , qq p1 b px, x1 qq p1 b p, 1 qq d1 dx1 ,

so that

Then, the operator

pr8 px, q pr18 px, q ` pr28 px, q ` pr38 px, q.

TV i|18,x1 px, x1 , 1 , q|2 18,x1 px, x1 , 1 , q x1 Vpx, x1 , 1 , q x1

such that

TV ei8 px, ,x ,q ei8 px, ,x ,q

is well defined on the support if the integrand of (4.14), and, respectively, the
operator
TC i|18,1 px, x1 , 1 , q|2 18,1 px, x1 , 1 , q 1 Cpx, x1 , 1 , q 1
such that

TC ei8 px, ,x ,q ei8 px, ,x ,q

is well defined on the support of the integrand of (4.15). Both TV and TC are well
defined on the support of the integrand of (4.16). Notice also that the coefficients
of TV satisfy, on the support of the integrand of (4.14), estimates of the type
(4.17)

|Bx1 B1 Vpx, x1 , 1 , q| . xx1 y|| x1 y|| pxy ` x1 yq1 .

Since there xxy xx1 y, the same holds with x in place of x1 . Similarly, the coefficients
of TC satisfy, on the support of the integrand of (4.15), estimates of the type
(4.18)

|Bx1 B1 Cpx, x1 , 1 , q| . xx1 y|| x1 y|| pxxy ` xx1 yq1 ,

as well as the analogous ones with in place of 1 , since xy x1 y there. Moreover,

both (4.17) and (4.18) hold on the support of the intengrand in (4.16). The claim

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

25

then follows by repeated integration by parts, using TC and/or TV in the expressions


of p38 , p28 , and p18 , and recalling Lemma 4.5.
Proposition 4.8. Under the hypotheses of Theorem 4.2, for p0 defined in (4.12) we have
p0 P S0,0 pR2n q.
To prove Proposition 4.8, we will use the change of variables
#
x1 Ypx, q ` y px, q1
(4.19)
1 Npx, q ` px, q,
1

where px, q xxy 2 xy 2 P S 2 , 2 and pY, Nq pYpx, q, Npx, qq is the unique

solution of
#
Ypx, q 11 px, Npx, qq
Npx, q 12x pYpx, q, q,
1 1

see (3.4) of Section 3 above. With as in Definition 4.4, let

py, ; x, q px, Ypx, q ` y px, q1 , Npx, q ` px, q, q,
py, ; x, q 0 px, Ypx, q ` y px, q1 , Npx, q ` px, q, q,
so that
p0 px, q

"

epy,;x,qpy, ; x, q dyd.

By construction, on supp ,
|Ypx, q ` y px, q1 x|

2
xxy,
3a

|Npx, q ` px, q |

2
xy,
3a

which implies that, for a sufficiently large a 0 and a suitable k P p0, 1q, on supp
we also have by (3.12) and (3.14)
1

|y| px, q1 kxxy

and || px, q kxy
|y|, || k pxxyxyq 2 .

Furthermore, recalling that x11 px, qy xxy and x12x px, qy xy, we find that, on
supp , for any P r0, 1s,
(4.20)

xYpx, q ` y px, q1 y xxy,

xNpx, q ` px, qy xy.

The next Lemma 4.9 can be proved analysing the Taylor expansions of py, ; x, q.
Lemma 4.9. Let
2

A1 p; x, q px, q

1
0

A2 py; x, q px, q2
B1 p; x, q px, q2

1
0

B2 py; x, q px, q

2
p1 qJ1
px, Npx, q ` px, qq d,

2
p1 qJ2xx
pYpx, q ` y px, q1 , q d,

2
J1
px, Npx, q ` px, qq d,

1
0

2
J1xx
pYpx, q ` y px, q1 , q d.

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

26

Then
py, ; x, q y ` p1 px, Npx, q ` px, qq 1 px, Npx, qqq
11 px, Npx, qq px, q
` p2 pYpx, q ` y px, q1 , q 2 pYpx, q, qq
12x pYpx, q, qq y px, q1

(4.21)

y ` rA1 p; x, qs ` rA2 py; x, qys y,

1y py, ; x, q ` r12x pYpx, q ` y px, q1 , q 12x pYpx, q, qs px, q1
` B2 py; x, qy,

(4.22)

1 py, ; x, q y ` r11 px, Npx, q ` px, qq 11 px, Npx, qqs px, q

y ` B1 p; x, q.

(4.23)

Proof. By the definition (4.8) of 0 and of the multi-product of phase functions (3.3)
and (3.6), recalling (4.19), we can write
0 px, x1 , 1 , q 1 px, 1 q x1 1 ` 2 px1 , q
1 px, Npx, qq ` Ypx, q Npx, q 2 pYpx, q, q,
which implies
0 px,Ypx, q ` y px, q1 , Npx, q ` px, q, q
1 px, Npx, q ` px, qq pYpx, q ` y px, q1 q pNpx, q ` px, qq
`2 pYpx, q ` y px, q1 , q 1 px, Npx, qq 2 pYpx, q, q ` Ypx, q Npx, q
y ` p1 px, Npx, q ` px, qq 1 px, Npx, qqq Ypx, q px, q
` p2 pYpx, q ` y px, q1 , q 2 pYpx, q, qq y Npx, q px, q1 .
Then, recalling that Ypx, q 11 px, Npx, qq and Npx, q 12x pYpx, q, q, we get
py, ; x, q y ` p1 px, Npx, q ` px, qq 1 px, Npx, qqq
Ypx, q px, q ` p2 pYpx, q ` y px, q1 , q 2 pYpx, q, qq
y Npx, q px, q1
y ` p1 px, Npx, q ` px, qq 1 px, Npx, qqq
11 px, Npx, qq px, q
` p2 pYpx, q ` y px, q1 , q 2 pYpx, q, qq
12x pYpx, q, qq y px, q1
y ` rA1 p; x, qs ` rA2 py; x, qys y,
that is (4.21) and its subsequent expression in terms of A1 , A2 . Then (4.22) and
(4.23) immediately follow taking derivatives with respect to y, in (4.21), and then
looking at the definitions of B1 , B2 .

Lemma 4.10. For A1 , A2 , B1 , B2 defined in Lemma 4.9 we have, for all x, y, , P Rn in
supp ,

|1 |
2

|1 |
2

}Bx B B pA1 , B1 qp; x, q} . xy||

}Bx B B y pA2 , B2 qpy; x, q} . xy||
a
where xy, y : 1 ` |y|2 ` ||2 , y, P Rn .

xxy||
xxy||

|1 |
2
|1 |
2

xy, y|`| ,
xy, y|`| ,

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

27

Proof. The result follows from the Faa di Bruno formula for the derivatives of the
composed functions, the properties of X P S1,0 , N P S0,1 stated above, the fact that,
on supp , (4.20) holds for any P r0, 1s, as well as
Ypx, q ` y px, q1 P S1,0 xy, y,

Npx, q ` px, q P S0,1 xy, y,

recalling that the seminorms of J1 and J2 involving their derivatives up to order 2

are proportional to P p0, 1q.
The proof works by induction on the order of the derivatives. Let us give an
idea of the step | ` ` 1 | 1. Let e j be the multiindex such that |e j | 1, with
components 0 everywhere apart from the j-th. Then, for instance, on supp ,
e
Bxj B1 p; x, q

e
pBxj 2 q

` 2

0
1,0

PS

1
0

2
J1
p. . . q d

1
0

3
J1x
p. . . q d

3
J1
p. . . q d Bxj pNpx, q ` px, qq

` S1,0 xy, y S1,0 xy, y,

1 2
1 3
1 3
since 2 P S1,1 , 0 J1
p. . . q d P S1,1 , 0 J1x
p. . . q d P S0,1 , 0 J1
p. . . q d P S1,2 ,
0,1
and Npx, q ` px, q P S ||. Similarly,
e

Bj B1 p; x, q pBj 2 q
` 2

1
0

1
0

2
J1
p. . . q d
e

3
J1
p. . . q d Bj pNpx, q ` px, qq

P S0,1 ` S0,1 xy, y S0,1 xy, y,

1
e
3
J1
Bj B1 p; x, q 2
p. . . q d p px, qq P S1{2,1{2 .
0

The estimates for general multiindeces follow by induction.

Lemma 4.11. On supp ,

|1y py, ; x, q| ` |1 py, ; x, q| |y| ` ||.
Proof. From Lemmas 4.9 and 4.10, on supp , for P p0, 1q,
}B1 p; x, q} . }B1 p; x, q} . ||,
}B2 py; x, q} . }B2 py; x, qy} . |y|,
which imply
|1y py, ; x, q| . || ` |y|,

|1y py, ; x, q| & || |y|,

|1 py, ; x, q| . |y| ` ||,

|1 py, ; x, q| & |y| ||.

These give
|1y py, ; x, q| ` |1 py, ; x, q| . p1 ` qp|y| ` ||q,
|1y py, ; x, q| ` |1 py, ; x, q| & p1 qp|y| ` ||q,
as claimed.

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

28

Lemma 4.12. On supp , for any multiindeces , , 1 , 1 , and all x, y, , ,

$

0
if |1 | 2,

&
1
if |1 | 1,
1
1
|Bx B y B B 1y py, ; x, q| .
|1 |
|1 |
xxy|| 2 xy|| 2 xy, y1`|`| if |1 | 0,

%
| ` ` 1 | 0;
$

0
if |1 | 2,

&
1
if |1 | 1,
1
1
|Bx B y B B 1 py, ; x, q| .
|1 |
|1 |
|| 2

xy|| 2 xy, y1`|`| if |1 | 0,

xxy

%
| ` 1 ` | 0.
Proof. The results follow from Lemma 4.10 and the estimates (4.20).
1

Lemma 4.13. On supp , for any multiindeces , , , , and all x, y, , ,

$
|1 |
|1 |

xy|| 2 xxy1|| 2 xy, y3`|`| if |1 | 0,

&
1
1
|1 |
|1 |
|Bx B y B B 1x py, ; x, q| . xy|| 2 xxy1|| 2 xy, y3`|`| if |1 | 0,

%
xy|| xxy1|| xy, y3`|`|
if 1 1 0;
$
|1 |
|1 |

xy1|| 2 xxy|| 2 xy, y3`|`| if |1 | 0,

&
1
1
|1 |
|1 |
|Bx B y B B 1 py, ; x, q| . xy1|| 2 xxy|| 2 xy, y3`|`| if |1 | 0,

%
xy1|| xxy|| xy, y3`|`|
if 1 1 0.
Proof. The results follow from Lemma 4.10, observing that

1x py, ; x, q dx rpA1 p; x, qq s ` dx rpA2 py; x, qyq ys,

1 py, ; x, q d rpA1 p; x, qq s ` d rpA2 py; x, qyq ys.

Lemma 4.14. For any multiindeces , , 1 , 1 , and all x, y, , ,
1

|B B Bx B y py, ; x, q| . xy||

|1 |
2

xxy||

|1 |
2

Proof. Immediate, by the definition of , the hypotheses on , the properties

Ypx, q P S1,0 , Npx, q P S0,1 , and the estimates (4.20).

Lemma 4.15. Let
py, ; x, q 1 ` |1y py, ; x, q|2 ` |1 py, ; x, q|2 .
Then, on supp , for any multiindeces , , 1 , 1 , and all x, y, , ,

1 1
1
. xy|| xxy|| xy, y2`|`| .
B B Bx B y

py, ; x,

Proof. Immediate, by Lemmas 4.9, 4.10, 4.11, and 4.12.

The next Lemma 4.16 is a straightforward consequence of Lemma 4.15 and the
definition of transpose operator.
Lemma 4.16. Let us define the operator
1
p1 i1y py, ; x, q y i1 py, ; x, q q

eipy,;x,q . Then,

M
such that Meipy,;x,q

M M0 ` M1 y ` M2 ,

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

29

where, on supp , for any multiindeces , , 1 , 1 , and all x, y, , ,

1

}B B Bx B y rpM0 , M1 , M2 qpy, ; x, qs} . xy|| xxy|| xy, y1`|`| .

Proof of Proposition 4.8. Using the operator M defined in Lemma 4.16, we have, for
arbitrary k P Z` ,
"
p0 px, q
eipy,;x,qppt Mqk qpy, ; x, q dyd.
Notice that, from the analysis above, for any k P Z` , any multiindeces 1 , 1 , and
all x, y, , ,
1

|pB Bx ppt Mqk qqpy, ; x, q| . xxy| | xy| | xy, yk`| ` | .

Then, for any fixed , P Zn` , and arbitrary k P Z` , we find

B Bx p0 px, q

"

B1 Bx 1 eipy,;x,q pB2 Bx 2 ppt Mqk qpy, ; x, qq dyd.

1
1
1 `2 `
1

Choosing k such that k ` 6| ` | p2n ` 1q, from the results in Lemmas 4.13,
4.14, and 4.16 above, we get
"

|Bx B p0 px, q| . xxy|| xy||

xy, yp2n`1q dyd . xxy|| xy|| ,
as claimed.

Remark 4.17. Let us notice that we have proved here above that the seminorms of p0 are
controlled by those of 1 and 2 . This implies that, if J1 and J2 are bounded in S1,1 , so
is p0 in S0,0 . The boundedness conditions of Theorem 4.2 are so fulfilled, and the proof of
Theorem 4.2 is complete.
5. Fundamental solution to hyperbolic systems in SG classes
In the present section we apply the results of Sections 3 and 4 to construct
the fundamental solution Ept, sq to the Cauchy problem for a first order system
of partial differential equations of hyperbolic type, with coefficients in SG classes
and roots of (possibly) variable multiplicity. A standard argument, which we omit
here, gives then the solution, via Ept, sq and Duhamels formula, see Theorem 5.1
below. We follow the approach in [24, Section 10.7].
Let us consider the Cauchy problem
#
LWpt, xq Fpt, xq pt, xq P p0, Ts Rn ,
(5.1)
Wp0, xq W0 pxq x P Rn ,
where
(5.2)

Lpt, x, Dt , Dx q Dt ` pt, x, Dxq ` Rpt, x, Dx q,

is an m m diagonal operator matrix whose entries j pt, x, Dxq, j 1, . . . , m, are

pseudo-differential operators with symbols j pt, x, q P Cpr0, Ts; S,1q, P r0, 1s, and
R is an m m-operator matrix with elements in Cpr0, Ts, S1,0q. The case 0
corresponds to symbols uniformly bounded in the space variable, while the case
1 is the standard situation of SG symbols with equal order components.
Assume also that the system (5.2) is of hyperbolic type, that is, j pt, x, q P R,
j 1, . . . , m. Notice that, differently from [12, 17], here we do not impose any
separation condition at infinity on the j , j 1, . . . , m. Indeed, the results

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

30

presented below apply both to the constant as well as the variable multiplicities
cases.
For 0 T0 T, we define T0 : tpt, sq| 0 s t T0 u. The fundamental
solution of (5.1) is a family tEpt, sq|pt, sq P T0 u of SG FIOs, satisfying
#
LEpt, sq 0 pt, sq P T0 ,
(5.3)
Eps, sq I
s P r0, T0s.
In this section we aim to show that, if T0 is small enough, it is possible to construct
the family tEpt, squ satisfying (5.3).
As a consequence of (5.3), it is quite easy to get the following:
Theorem 5.1. For every F P Cpr0, Ts; Hr,pRn qq and G P Hr, pRn q, the solution Wpt, xq of
the Cauchy problem (5.1) exists uniquely, it belongs to the class Cpr0, T0 s, Hrp1q,pRn qq,
and it is given by
t
Wptq Ept, 0qG ` i Ept, sqFpsqds, t P r0, T0s.
0

Remark 5.2. Theorem 5.1 gives well-posedness of the Cauchy problem (5.1) in SpRn q and
S1 pRn q; moreover it gives well posedness with loss/gain of decay (depending on the sign
of r) of (5.1) in weighted Sobolev spaces Hr, pRn q. This phenomenon is quite common in
the theory of hyperbolic partial differential equations with SG type coefficients, see [2, 4, 5].
We remark that in the symmetric case 1 the Cauchy problem (5.1) turns out to be
well-posed also in Hr, pRn q.
To begin, consider SG phase functions j j pt, s, x, q, 1 j m, defined on
T0 R2n , and define the operator matrix

I1 pt, sq
0

..
I pt, sq
,
.
0
Im pt, sq
where I j : Op j p1q, 1 j m. From Theorem 2.3 (see Remark 2.8) we see that

B j
Dt I j ` j pt, x, DxqI j ei j pt,s,x,q
pt, s, x, qd
Bt

` ei j pt,s,x,q j pt, x, 1j,xpt, s, x, qqd

` ei j pt,s,x,qb0, j pt, s, x, qd,

where b0, j pt, sq P S1,0 S0,0 . The first two integrals in the right-hand side of
the equation here above cancel if we choose j , j 1, . . . , m, to be the solution
to the eikonal equation (2.5) associated with the symbol a j , j 1, . . . , m.
By Proposition 2.9, this is possible, provided that T0 is small enough. Writing
B0, j : Op j pb0, j q, we define the family tW1 pt, sq; pt, sq P T0 u of SG FIOs by

B0,1 pt, s, x, Dxq

0

..
W1 pt, s, x, Dxq : i
` Rpt, x, Dx qI pt, s, x, Dx q,
.
0

B0,m pt, s, x, Dxq

and we denote by w1 pt, s, x, q the symbol of W1 pt, s, x, q. Notice that

(5.4)

Lpt, x, DxqI pt, s, x, Dxq iW1 pt, s, x, Dxq,

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

31

that is, iW1 is the residual of system (5.1) for I . We define then by induction the
sequence of m m-matrices of SG FIOs, denoted by tW pt, sq; pt, sq P T0 uPN , as
W`1 pt, s, x, Dx q

(5.5)

W1 pt, , x, DxqW p, s, x, Dxqd,

and we denote by w`1 pt, s, x, q the symbol of W`1 pt, s, x, Dxq. We are now going
to prove that the operator norms of W , seen as operators from the Sobolev space
Hr, into Hrp1qp1q, for any fixed pr, q P R2 can be estimated from above by
}W pt, sq}LpHr, ,Hrp1qp1q, q

(5.6)

1
C1
r, |t s|

p 1q!

1
C1
r, T0

p 1q!

for all pt, sq P T0 and P N, where Cr, is a constant which only depends on r, .
To deal with the operator norms in (5.6), we need to explicitly write the matrices
W . An induction in (5.5) easily shows that
W pt, sq

(5.7)

t 1
s

...

W1 pt, 1 q . . . W1 p2 , 1 qd1 . . . d1 .

The integrand is a product of 1 m m-matrices of SG FIOs, therefore it is an

operator matrix whose entries consist of m2 summands of compositions of 1
SG FIOs. Denoting by Q1 . . . Q1 one of these compositions, where each of
the Q j is one of the m2 entries of the m m-matrix of SG FIOs W1 , we have from
Example 3.3 and (2) of Theorem 4.3 that Q1 . . . Q1 is again a SG FIO with
symbol q1,...,1 P Sp1qp1q,0 S0,0 . Moreover, from (3) of Theorem 4.3, for all
P N there exists C 0 and 1 P N0 such that
p1qp1q,0

|||q1,...,1 pt, 1 , . . . , 1 q|||

C2
|||q1 pt, 1 q|||1,0
. . . |||q1 p2 , 1 q|||1,0
,

1
1
where for j 1, . . . , 1, q j pt, sq denotes the symbol of the SG FIO Q j pt, sq, pt, sq P
T0 . Now we set
:

sup

8,
sup |||q j pt, sq|||1,0
1

j1,...,1 pt,sqPT0

so that
p1qp1q,0

|||q1,...,1 pt, 1 , . . . , 1 q|||

C2
1 .

The continuity of the SG FIOs Q1 . . .Q1 pt, 1 , , 1 q : Hr, Hrpn1qp1q,

(see Theorem 2.1) and the previous inequality give that for every r, there exist
constants Cr, 0 (depending only on the indeces of the Sobolev space) and
r, P N0 such that for all u P Hr,
(5.8)

}Q1 pt, 1 q . . . Q1 p2 1 qu}rpn1qp1q,

p1qp1q,0

Cr, |||q1,...,1 pt, 1 , . . . , 1 q|||r,

}u}r,

Cr, C2
1 }u}r, .
r,
Therefore, in the operator matrix W1 pt, 1 q . . . W1 p2 , 1 q, the operator norm of
each entry can be bounded from above by m2 Cr, C2
1 . Now by (5.7) and
r,

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

32

(5.8) we deduce that

}W pt, sq}LpHr, ,Hrp1qp1q, q
2
t 1
}W1 pt, 1 q . . . W1 p2 , 1 q}LpHr, ,Hrp1qp1q, q d1 . . . d1
...

(5.9)

Cr, C2
1
r,

t 1
s

...

m2 Cr, C2
1 |t s|1
r,
p 1q!

d1 . . . d1

1
C 1
r, |t s|

p 1q!

for a new constant C r, depending only on r, , which yields the claim (5.6).
Now, using the estimate (5.6), we can show that the sequence of SG FIOs, defined
for all pt, sq P T0 and all N P N by
EN pt, sq I pt, sq `

(5.10)

I pt, q

W p, sqd,

is a well-defined SG FIO in LpHr,, Hr`1, q for every r, , and converges, as N 8,

to the well-defined SG FIO, belonging to LpHr, , Hr`1,q, given by
t
8

(5.11)
Ept, sq I pt, sq ` I pt, q
W p, sqd.
s

Ept, sq in (5.11) is the fundamental solution to the system (5.1) in the sense that
it satisfies (5.3). Indeed, at symbols level, with the notations EN Op peN q,
E Op peq and W1 W1 Op p1 q, for every l P N and | ` | , we have

|B Bx eN pt, s, x, q|
t
N

|B Bx w p, s, x, q|d

s 1

N t 1 2

B Bx 1 pt, 1 , . . . , 1 , x, qdn1 . . . d1 d
...

1 s

1 s

...

2
s

p1qp1q,0

|||1 pt, 1 , . . . , 1 q|||

xxy1||xy||

N m2 C2

1 |t s|1

p 1q!

so

|||eN pt, sq|||1,0

N1

xxyp1qp1q||xy|| d1 . . . d

pC1 |t s|q
!

for a new constant C1 0. Then, for N 8 we get

|||ept, sq|||1,0
exppC1 pt sqq 8.

Thus, the SG FIO (5.11) has a well-defined symbol. On the other hand, at operators
level, by definitions (5.10) and (5.2) we have
(5.12)

LEN LI i

W pt, sq `

t
s

LI pt, q

W p, sqd.

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

33

An induction shows that

t
N1
N

W p, sqd.
(5.13)
W pt, sq ipLI qpt, sq i pLI qpt, q
s

Indeed, for N 2 we have by (5.4) and (5.5)

W1 pt, sq ` W2 pt, sq ipLI qpt, sq i

pLI qpt, qW1 p, sqd;

the induction step N N ` 1 works as follows:

N`1

W pt, sq WN`1 pt, sq `

W pt, sq

pLI qpt, qWN p, sqd ipLI qpt, sq i

ipLI qpt, sq i

t
s

pLI qpt, q

pLI qpt, q

N1

W p, sqd

W p, sqd.

Substituting (5.13) into (5.12) we get

t
pLEN qpt, sq pLI qpt, qWN p, sqd.
s

Now, for N 8, }WN pt, sq}LpHr, ,HrpN1qp1q, q 0 because of (5.9); thus LEN
LE 0. Moreover, its easy to verify that Eps, sq I. So, (5.3) is fulfilled, and we
have constructed the fundamental solution to L. As it concerns the dependence
of the fundamental solution on the parameters pt, sq, we finally notice that the SG
FIO-valued map pt, sq Ept, sq belongs to CpT0 q, since E is obtained by continuous
operations of operators which are continuous in t, s, see (5.11).
References
[1] G.D. Andrews, SG Fourier integral operators with closure under composition, Dissertation. Imperial College, London, 2009.
[2] Ascanelli, A., Cappiello, M., Log-Lipschitz regularity for SG hyperbolic systems. J. Differential
Equations, 230 (2006), 556578.
[3] Ascanelli, A., Cappiello, M., Holder

continuity in time for SG hyperbolic systems. J. Differential

Equations, 244 (2008), 20912121.
[4] Ascanelli, A., Cappiello, M., The Cauchy problem for finitely degenerate hyperbolic equations
with polynomial coefficients. Osaka J.Math., 47, n.2 (2010), 423438.
[5] Ascanelli, A., Cappiello, M., Weighted energy estimates for p-evolution equations in SG classes.
Journal of Evolution Equations, (2015), DOI: 10.1007/s00028-015-0274-6
[6] Ascanelli, A., Coriasco, S., Su,
A., Solution theory of hyperbolic stochastic partial differential
equations with polynomially bounded coefficients (2015), in preparation.
[7] Battisti, U., Coriasco, S., Wodzicki residue for operators on manifolds with cylindrical ends. Ann.
Global Anal. Geom., Vol. 40, No. 2 (2011), 223249.
[8] Cappiello, M., Fourier integral operators of infinite order and applications to SG-hyperbolic equations. Tsukuba J. Math., 28 (2004), 311361.
[9] Cordero, E., Nicola, F., Rodino, L., On the global boundedness of Fourier integral operators. Ann.
Global Anal. Geom., 38 (2010), 373398.
[10] H. O. Cordes The technique of pseudodifferential operators. Cambridge Univ. Press, 1995.
[11] S. Coriasco. Fourier integral operators in SG classes I. Composition theorems and action on SG
Sobolev spaces. Rend. Sem. Mat. Univ. Pol. Torino, 57, 4:249302, 1999.
[12] S. Coriasco. Fourier integral operators in SG classes II. Application to SG hyperbolic Cauchy
problems. Ann. Univ. Ferrara, 47:81122, 1998.
[13] Coriasco, S., Johansson, K., Toft, J., Calculus, continuity and global wave-front properties for
Fourier integral operators on Rd . Preprint arXiv:1307.6249 (2014).
[14] Coriasco, S., Maniccia, L., Wave front set at infinity and hyperbolic linear operators with multiple
characteristics. Ann. Global Anal. Geom., 24 (2003), 375400.

FIO ALGEBRA AND FUNDAMENTAL SOLUTION TO SG HYPERBOLIC SYSTEMS

34

[15] Coriasco, S., Maniccia, L., On the spectral asymptotics of operators on manifolds with ends. Abstr.
Appl. Anal., Vol. 2013, Article ID 909782 (2013).
[16] Coriasco, S., Panarese, P., Fourier integral operators defined by classical symbols with exit behaviour. Math. Nachr., 242 (2002), 61-78.
[17] S. Coriasco and L. Rodino. Cauchy problem for SG-hyperbolic equations with constant multiplicities. Ric. di Matematica, 48, (Suppl.):2543, 1999.
[18] Coriasco, S., Ruzhansky, M., Global Lp -continuity of Fourier integral operators. Trans. Amer. Math.
Soc., Vol. 366, No. 5 (2014), 2575-2596.
[19] Coriasco, S., Schulz, R., Global wave front set of tempered oscillatory integrals with inhomogeneous phase functions. J. Fourier Anal. Appl., Vol. 19, No. 5 (2013), 10931121.
[20] Coriasco, S., Schulz, R., SG-Lagrangian submanifolds and their parametrization. Preprint
arXiv:1406.1888 (2014).
[21] Coriasco, S., Toft, J., Asymptotic expansions for Hormander

symbol classes in the calculus of

pseudo-differential operators. J. Pseudo-Differ. Oper. Appl., Vol. 5, No. 1 (2014), 2741.
[22] Coriasco, S., Toft, J., Calculus for Fourier integral operators in generalized SG classes. Preprint,
arXiv:1412.8050 (2014).
[23] Y. V. Egorov and B.-W. Schulze. Pseudo-differential operators, singularities, applications, volume 93 of
Operator Theory: Advances and Applications. Birkhauser Verlag, Basel, 1997.
[24] H. Kumano-go. Pseudo-differential operators. MIT Press, 1981.
[25] L. Maniccia and P. Panarese. Eigenvalue asymptotics for a class of md-elliptic dos on manifolds
with cylindrical exits. Ann. Mat. Pura Appl. (4), 181, 3:283308, 2002.
[26] R. Melrose. Geometric scattering theory. Stanford Lectures. Cambridge University Press, Cambridge,
1995.
[27] C. Parenti. Operatori pseudodifferenziali in Rn e applicazioni. Ann. Mat. Pura Appl., 93:359389,
1972.
[28] Ruzhansky, M., Sugimoto, M., Global L2 boundedness theorems for a class of Fourier integral
operators. Comm. Partial Differential Equations, 31 (2006), 547569.
[29] Schrohe, E., Spaces of weighted symbols and weighted Sobolev spaces on manifolds In: Proceedings, Oberwolfach. (H.O. Cordes, B. Gramsch, H. Widom, eds.), pp. 360-377. Springer LMN, Vol.
1256. New York, 1986.
[30] B.-W. Schulze. Boundary value problems and singular pseudo-differential operators. Pure and Applied
Mathematics (New York). John Wiley & Sons Ltd., Chichester, 1998.
[31] M. A. Shubin. Pseudodifferential operators and spectral theory. Springer Series in Soviet Mathematics.
Springer-Verlag, Berlin, 1987.
Dipartimento di Matematica, Universita` degli Studi di Ferrara, via Machiavelli 30, 44121
Ferrara, Italy
E-mail address: [email protected]
Dipartimento di Matematica G. Peano, Universita` degli Studi di Torino, via Carlo Alberto
10. 10123 Torino, Italy
E-mail address: [email protected]