Composition operators on the Schwartz space

arXiv:1511.03072v1 [[Link]] 10 Nov 2015

Antonio Galbis, Enrique Jordá

March 21, 2022

Abstract
We study composition operators on the Schwartz space of rapidly
decreasing functions. We prove that such a composition operator is
never a compact operator and we obtain necessary or sufficient con-
ditions for the range of the composition operator to be closed. These
conditions are expressed in terms of multipliers for the Schwartz class
and the closed range property of the corresponding operator considered
in the space of smooth functions.
Keywords: Composition operator; Composite function problem; Space
of rapidly decreasing functions. MSC2010: 47B33, 46F05, 47A05

1 Introduction.
Composition operators, as well as multiplication operators, are widely stud-
ied in the last years mainly in spaces of holomorphic functions. One can
check [2, 13] and the references given therein. Much less seems to be known
about the functional analytic properties of the composition operator when
it is defined in real spaces of smooth functions. To be more precise let
us denote by C ∞ (Rd ) the Fréchet space of all smooth functions endowed
with its natural topology of uniform convergence of the derivatives on the
compact sets of Rd . In the case that ϕ : Rd → Rd is an analytic function,
properties of the composition operator Cϕ : C ∞ (Rd ) → C ∞ (Rd ), f 7→ f ◦ ϕ,
are well known since long ([4, 7, 14]). However, when the symbol ϕ is only
assumed to be smooth, the operator is not well understood. In particu-
lar, to characterize when the operator has closed range becomes a tough
problem. Yet, there are some recent important advances in this direction
concerning the one variable case. Kenessey and Wengenroth characterized
in [6] the injective symbols ϕ : R → Rd such that Cϕ : C ∞ (Rd ) → C ∞ (R)
has closed range. A sufficient condition on the symbol ϕ : R → R for
the operator Cϕ : C ∞ (R) → C ∞ (R) to have closed range was obtained by

1
Przestacki. This condition is also necessary under some mild assumptions
on ϕ ([9, 10, 11]). A characterization of those classes of ultradifferentiable
functions which are closed under composition was obtained in [3].
Besides the space of smooth functions and the space of real analytic
functions, one of the most important spaces in mathematical analysis is the
Schwartz space S(Rd ) of rapidly decreasing functions. The multipliers in
S(Rd ) are the functions F : Rd → Rd such that the multiplication opera-
tor MF : S(Rd ) → S(Rd ), f 7→ F f, is well defined and continuous. The
set of all multipliers is denoted by OM (Rd ) and consists of those smooth
functions whose derivatives of arbitrary order have polynomial growth at
infinity. Bonet, Frerick and the second author have characterized in [1] the
multipliers ϕ ∈ OM (:= OM (R)) such that Mϕ : S(R) → S(R), f 7→ f ϕ, has
closed range.
In this paper we study composition operators defined in the Schwartz
space S(R) of one variable rapidly decreasing functions. We characterize
the smooth functions ϕ : R → R for which the composition operator Cϕ :
S(R) → S(R), f 7→ f ◦ ϕ, is well defined and continuous and, if so, we
study some of its properties. In particular, we prove that Cϕ can never
be a compact operator and obtain necessary or sufficient conditions for the
range of the composition operator to be closed in S(R). These conditions
are expressed in terms of the multiplication operator studied in [1] and
the closed range of the corresponding operator considered in the space of
smooth functions, involving then the conditions considered by Przestacki in
[9, 10, 11]. We remark that the characterization of the symbols is valid for
the several variables case.
Recall that S(R) consists of those smooth functions f : R → R with the
property that

πn (f ) := sup sup (1 + |x|2 )n |f (j) (x)| < ∞

x∈R 1≤j≤n

for each n ∈ N. S(R) is a Fréchet space when endowed with the topology gen-
erated by the sequence of seminorms (πn )n∈N . Usually the Schwartz space
refers to the space of complex valued functions satisfying this conditions,
which is of fundamental importance in harmonic analysis. If we denote this
space by SC (R), then we have SC (R) = S(R) ⊕ iS(R). Since our objec-
tive is to study the behaviour of the composition operator, we can restrict
to the real space and the conclusions remain valid for the corresponding
composition operator defined in SC (R).

2
2 Characterization of symbols.
Definition 2.1 A function ϕ ∈ C ∞ (R) is called a symbol for S(R) if f ◦ϕ ∈
S(R) whenever f ∈ S(R).
If ϕ is a symbol, then the composition operator Cϕ : S(R) → S(R),
f 7→ f ◦ ϕ, is continuous by the closed graph theorem. Our first aim is
to characterize the symbols for S(R) and we begin with a simple necessary
condition which, in particular, implies that every symbol ϕ for S(R) is a
semiproper map, that is, for every compact set K ⊂ R there exists a compact
set L ⊂ R such that ϕ(R) ∩ K ⊂ ϕ(L).

Lemma 2.2 If ϕ is a symbol for S(R) then

lim |ϕ(x)| = ∞.
|x|→∞

Consequently, either ϕ(R) = R or ϕ(R) = [a, +∞) for some a or ϕ(R) =

(−∞, b] for some b.
Proof. We proceed by contradiction. Assume that there exists a sequence
(xj ) with lim |xj | = ∞ and ℓ ∈ R such that lim ϕ(xj ) = ℓ. Let f be a
j→∞ j→∞
compactly supported function such that f (ℓ) = 1. Then f ∈ S(R), however
f ◦ϕ ∈
/ S(R) since

lim |xj (f ◦ ϕ) (xj )| = ∞.

j→∞
✷

For the proof of the next result we recall Faà di Bruno formula (see e.g.
[5, 1.3.1]):
k 1 !k n
ϕ′ (x) ϕ(n) (x)

(n)
X n!
(f ◦ ϕ) (x) = f (k) (ϕ(x)) ...
k1 ! . . . kn ! 1! n!

where the sum is extended over all (k1 , . . . , kn ) ∈ Nn0 such that k1 + 2k2 +
. . . + nkn = n and k := k1 + . . . + kn .

Theorem 2.3 A function ϕ ∈ C ∞ (R) is a symbol for S(R) if and only if

the following conditions are satisfied:

3
 (i) For all j ∈ N0 there exist C, p > 0 such that

ϕ(j) (x) ≤ C(1 + |ϕ(x)|2 )p

for every x ∈ R.

(ii) There exists k > 0 such that |ϕ(x)| ≥ |x|1/k for all |x| ≥ k.

Proof. Let us first assume that ϕ is a symbol for S(R) and prove that
condition (i) is satisfied. If this is not the case, there exist n ∈ N and a
sequence (xj )j ⊂ R such that

|xj | + 1 < |xj+1 |

and

ϕ(n) (xj ) ≥ j(1 + |ϕ(xj )|2 )j

for every j ∈ N0 .
By Lemma 2.2 we can assume, taking a subsequence if necessary,

|ϕ(xj )| + 1 < |ϕ(xj+1 )|, j ∈ N0 .

Let us consider ρ ∈ D[− 12 , 21 ] with ρ′ (0) = 1 and ρ(j) (0) = 0 for 2 ≤ j ≤ n,
and define
∞
X ρ(x − yj )
f (x) := , yj := ϕ(xj ).
(1 + |yj |2 )j
j=1

The terms of the sum defining f are disjointly supported. Moreover

∞
[ 1 1
supp f ⊆ [yj − , yj + ].
2 2
j=1

If x ∈ Ik := [yk − 12 , yk + 21 ] then

ρ(j) (x − yk )
f (j)(x) = .
(1 + |yk |2 )k
Hence, for each j, m ∈ N there is C > 0 not depending on k such that

(1 + |x|2 )m
(1 + |x|2 )m f (j) (x) ≤ kρ(j) k∞ ≤ C(1 + |yk |2 )m−k , (2.1)
(1 + |yk |2 )k

4
from where it follows

lim (1 + |x|2 )m f (j) (x) = 0,

|x|→∞

i.e. f ∈ S(R). Now, an easy consequence of the Faà di Bruno formula gives

ρ′ (0)ϕ(n) (xk ) ϕ(n) (xk )

(f ◦ ϕ)(n) (xk ) = = ≥ k,
(1 + |yk |2 )k (1 + |ϕ(xk )|2 )k
and then f ◦ ϕ 6∈ S(R), contradicting that ϕ is a symbol for S(R).
Assume that ϕ is a symbol for S(R) but (ii) does not hold. Let (xj )j be
a sequence in R such that |xj | ≥ j and |ϕ(xj )|j ≤ |xj |. We take ρ as above.
By Lemma 2.2, we can assume without loss of generality,

|ϕ(xj )| + 1 < |ϕ(xj+1 )|, j ∈ N0 .

Now we define
∞
X ρ(x − yj )
f (x) := , yj := ϕ(xj ).
|yj |j
j=1

We have f ∈ S(R) since there exists C independent on k such that (1+x2 ) ≤

C(1 + yk2 ) for each x ∈ Ik = [yk − 12 , yk + 21 ]. However

lim inf |xj | · |(f ◦ ϕ) (xj )| = lim inf |xj |/|ϕ(xj )|j ≥ 1,
j→∞ j→∞

and hence f ◦ ϕ is not in S(R), a contradiction since ϕ is a symbol for S(R).

Let us now assume that ϕ satisfies conditions (i) and (ii) and prove
that ϕ is a symbol for S(R). Condition (ii) in the statement implies the
boundedness of the function h(x) := (1 + |x|2 )/(1 + |ϕ(x)|2 )k . Thus, for a
fixed n ∈ N, we can get C > 0 and p > 0 satisfying (i) for 1 ≤ j ≤ n and
also |h(x)|n ≤ C for all x ∈ R. This implies that,

(1 + x2 )n ≤ C(1 + |ϕ(x)|2 )kn (2.2)

and

ϕ(j) (x) ≤ C(1 + |ϕ(x)|2 )p (2.3)

for every 1 ≤ j ≤ n and x ∈ R.

From condition (2.3) and Faà di Bruno formula we get constants M > 0 and
t ∈ N, such that, for each 1 ≤ j ≤ n and f ∈ S(R)

5
 X
(f ◦ ϕ)(j) (x) ≤ M (1 + |ϕ(x)|2 )t f (ℓ) (ϕ(x)) . (2.4)
1≤ℓ≤j

We assume without loss of generality t > n. Hence

n
sup sup 1 + x2 (f ◦ ϕ)(j) (x)
x∈R 1≤j≤n

is less than or equal to

X
M C sup (1 + |ϕ(x)|2 )kn+t (f (ℓ) (ϕ(x)) ≤ nM Cπkn+t(f ) < ∞.
x∈R 1≤ℓ≤n

and we conclude f ◦ ϕ ∈ S(R). ✷

Remark 2.4 Lemma 2.2 and Theorem 2.3 remain valid for symbols in
S(Rn ). This is a consequence of the fact that there is ρ ∈ D(Rn ) with
∂ α ρ(0) = 1 for |α| = 1, ρ(α) (0) = 0 for |α| > 1, as follows from the Borel
theorem together with the existence of compactly supported C ∞ functions.
Thus, test functions and constant symbols are examples of symbols for S(Rd )
which do not belong to OM (Rd ).
From the conditions in Theorem 2.3 the following examples are followed.

Example 2.5 (a) Non constant polynomials are symbols for S(R). Also,
if P (x) is a polynomial with lim P (x) = +∞ then
|x|→+∞

ϕ(x) = exp (P (x))

is a symbol for S(R).

(b) The symbols for S(R) are stable under products. Moreover, if ϕ, ψ are
symbols and |ψ(x)| ≤ c|ϕ(x)| for some 0 < c < 1 then also ϕ + ψ is a
symbol.

3 Compactness of composition operators

We recall that a continuous operator T : E → F between two locally convex
spaces is said to be compact if there is some 0-neighbourhood U in E with
the property that T (U ) is a relatively compact subset of F.

6
 We now prove that a composition operator acting on S(R) is never com-
pact. This contrasts with the behavior of composition operators in Banach
spaces of analytic functions. In fact, compactness of composition operators
has been extensively studied on various spaces of analytic functions.
The key for the lack of compactness in our setting is the following well-
known lemma, the proof of which is included for the convenience of the
reader. For a bounded function f : R → R we denote

kf k∞ = sup{|f (x)| : x ∈ R}.

Lemma 3.1 For f ∈ D[a, b] and n ∈ N let define kf kn := nk=0 kf (k) k∞ .

P
The norms k · kn and k · kn+1 are not equivalent. Moreover the system of
norms (k · kn )n∈N defines the topology on D[a, b].
Proof. We argue by contradiction and assume that k · kn and k · kn+1
are equivalent norms. Then, from kf (k) k∞ = k(f ′ )(k−1) k∞ and proceeding
inductively we conclude that k · kk is equivalent to k · kk+1 for k ≥ n. This
is a contradiction since D[a, b] is not normable. ✷

Proposition 3.2 Let ϕ ∈ C ∞ (R) be given and assume ϕ([a, b]) = [c, d] and
|ϕ′ (x)| ≥ δ > 0 for all x ∈ [a, b]. Then Cϕ : D[c, d] → C ∞ (R), f 7→ f ◦ ϕ, is
not compact.
Proof. Let n ∈ N be fixed. The Faá Di Bruno formula implies that there
exist polynomials Q1 , Q2 , . . . , Qn−1 of n variables such that

(n)
n
(x) = f (n) ϕ(x) ϕ′ (x)


f ◦ϕ
Pn−1 (m) ϕ(x) Qm ϕ′ (x), ϕ′′ (x), . . . , ϕ(n) (x) .



+ m=1 f

This implies the existence of λn > 0 such that, for all x ∈ [a, b],

(n) Pn−1 (k)
≥ f (n) ϕ(x) δn − λn k=0


f ◦ϕ (x) kf k∞
(3.1)
f (n) ϕ(x) δn − λn kf kn−1 .


=

For each 0-neighbourhood U in D[c, d] there exist ε > 0 and p ∈ N such

that

{f ∈ D[c, d] : kf kp−1 ≤ ε} ⊆ U.

7
According to Lemma 3.1 there exists a sequence (fj )j ⊂ D[c, d] such that
kfj kp−1 = ε and
(p)
δp kfj k∞ > λp ε + j.
(p) (p)

Now we take xj ∈ [a, b] such that kfj k∞ = fj ϕ(xj ) . We get from
(3.1)

(p)
(p)
sup{ fj ◦ ϕ (x) : x ∈ [a, b]} ≥ fj ◦ ϕ (xj ) ≥ j

for every j ∈ N, from where it follows that Cϕ (U ) is unbounded. ✷

Theorem 3.3 Let X be a locally convex space such that D(R) ֒→ X ֒→

C ∞ (R) with continuous inclusions and let ϕ ∈ C ∞ (R) be a non constant
function with the property that f ◦ ϕ ∈ X for each f ∈ X. Then Cϕ : X → X
is not compact. In particular, Cϕ : S(R) → S(R) is not compact for any
symbol ϕ for S(R).
Proof. Since ϕ is non constant there exists [a, b] ⊆ R such that ϕ restricted
to [a, b] is monotonic, and then there exists δ > 0 such that |ϕ′ (x)| ≥ δ
for each x ∈ [a, b]. The hypothesis imply the continuity of the inclusion
i1 : D[c, d] ֒→ X for [c, d] = ϕ([a, b]). Proposition 3.2 implies that the
composition operator

Ĉϕ : D[c, d] → C ∞ (R), f 7→ f ◦ ϕ,

is not compact. The conclusion follows since Ĉϕ decomposes as i2 ◦ Cϕ ◦ i1 ,

with i2 being the continuous inclusion from X into C ∞ (R).
✷

4 Closed range composition operators

Our aim is to obtain necessary or sufficient conditions for the range of the
composition operator to be closed in S(R). We will relate the closed range
property of a composition operator Cϕ : S(R) → S(R) with the closed
range property of Cϕ : C ∞ (R) → C ∞ (R) (characterized by [10]) and the
closed range property of multiplication operators on S(R), which has been
characterized in [1]. Concrete examples of composition operators on S(R)
lacking the closed range property are provided.

8
Lemma 4.1 Let ϕ be a symbol for S(R). If Cϕ : S(R) → S(R) has closed
range then Cϕ : C ∞ (R) → C ∞ (R) has also closed range.
Proof. Let (fn )n be a sequence in C ∞ (R) such that (fn ◦ ϕ)n is convergent
to g in C ∞ (R). Our aim is to find f ∈ C ∞ (R) such that g = f ◦ ϕ. For each
k ∈ N we consider χk ∈ D[−2k, 2k] with the property that χk |[−k,k] = 1.
Since lim |ϕ(x)| = ∞ there exist Mk ∈ N such that ϕ−1 ([−2k, 2k]) ⊆
|x|→∞
[−Mk , Mk ]. Then (h ◦ ϕ) · (χk ◦ ϕ) ∈ D[−Mk , Mk ] for each h ∈ C ∞ (R) and
k ∈ N. Hence we have

lim (fn · χk ) ◦ ϕ = lim (fn ◦ ϕ) · (χk ◦ ϕ) = g · (χk ◦ ϕ)

n→∞ n→∞

in D[−Mk , Mk ], which is a topological subspace of both C ∞ (R) and S(R).

Hence there exists hk ∈ S(R) such that

g · (χk ◦ ϕ) = hk ◦ ϕ.
We observe that,
for every natural

numbers k ≤ q, the condition |ϕ(x)| ≤ k
implies hk ϕ(x) = hq ϕ(x) , since

g(x)(χk ◦ ϕ)(x) = g(x)(χq ◦ ϕ)(x) = g(x).

Consequently we can define f ϕ(x) := hk ϕ(x) whenever |ϕ(x)| ≤ k. It
easily follows that f is C ∞ in the interior of ϕ(R) and f ϕ(x) = g(x) for

all x ∈ R. In the case ϕ(R) = R we are done. In the case ϕ(R) = [a, +∞),
f admits right derivatives of every order at a, and, by Whitney extension
theorem [15] we can extend f to a function in C ∞ (R). A similar argument
gives the conclusion in the case that ϕ(R) = (−∞, b]. ✷

In order to obtain examples of composition operators on S(R) lacking

the closed range property we first consider symbols satisfying the additional
condition that there exist (qj )j ⊂ N and positive constants (Cj )j such that

q j
ϕ(j) (x) ≤ Cj 1 + x2 (1 + |ϕ(x)|) (∗)

Examples of symbols with property (∗) :

(a) Each symbol ϕ ∈ OM .

(b) Every smooth function ϕ such that ϕ(x) = sign(x)e|x| or ϕ(x) =

2
sign(x)ex for large values of |x|.

9
Proposition 4.2 Let ϕ be a surjective symbol with property (∗) such that
Cϕ : S(R) → S(R) has closed range. Then f ∈ C ∞ (R) and f ◦ ϕ ∈ S(R)
imply f ∈ S(R).
Proof. Let χ ∈ D[−1, 1] be a test function such that 0 ≤ χ ≤ 1 and χ(x) = 1
for x ∈ [− 21 , 21 ] and consider χk (x) := χ( xk ) and fk := f · χk ∈ S(R). We
claim that B := (fk ◦ ϕ)k is a bounded set in S(R). Once the claim is proved
we can proceed as follows. Since S(R) is a Montel space we can assume,
passing to a subsequence if necessary, that there exist

g := S(R) − lim fk ◦ ϕ.
k→∞

Since Cϕ has closed range then g = h◦ϕ for some h ∈ S(R). The surjectivity
of ϕ and the identity f ◦ ϕ = h ◦ ϕ imply f = h ∈ S(R).
Finally, we check that B is a bounded set in S(R), that is,

p
sup sup 1 + x2 · (fk ◦ ϕ)(ℓ) (x) (4.1)
k∈N x∈R

is finite for every p and ℓ. We observe that

(fk ◦ ϕ)(x) = (f ◦ ϕ)(x) · χk (ϕ(x)),

and, by Leibniz’s formula, (4.1) is less than or equal to a constant times

p
sup sup sup 1 + x2 · (f ◦ ϕ)(ℓ−n) (x) · (χk ◦ ϕ)(n) (x) . (4.2)
k∈N 0≤n≤ℓ x∈R

According to Faà di Bruno formula,

n  mj
(m1 +...+mn )
X
Y
(χk ◦ ϕ)(n) (x) = Cm,n χk ϕ(x) ϕ(j) (x)
m j=1

where m = (m1 , . . . , mn ) satisfies m1 + 2m2 + . . . + nmn = n. It turns out

that there are constants An , Bn > 0 such that
n
X 1 Y mj
(χk ◦ ϕ)(n) (x) ≤ An ϕ(j) (x)
m
km1 +...+mn
j=1

n 
1
mj qj 
(1 + |ϕ(x)|)mj .
X Y
≤ Bn 1 + x2
m
km1 +...+mn
j=1

10
Since |ϕ(x)| ≤ k whenever (χk ◦ϕ)(n) (x) 6= 0 we conclude that (χk ◦ ϕ)(n) (x)
is dominated by some polynomial that depends on n but is independent on
k. Since f ◦ ϕ ∈ S(R) then (4.2) is finite and the claim is proved. ✷

Corollary 4.3 Let ϕ(x) = sign(x)e|x| for large values of |x|. Then the range
of
Cϕ : S(R) → S(R)
is not closed.
1
Proof. Take f ∈ C ∞ (R) such that f (x) = 0 for x < 0 while f (x) = x for
x ≥ 1. Then f ◦ ϕ ∈ S(R) while f ∈
/ S(R). ✷

A multiplier of the space S(R) is a smooth function F satisfying

F · S(R) ⊂ S(R).

It is known that F ∈ C ∞ (R) is a multiplier of S(R) if, and only if, for each
k ∈ N there exist C > 0 and j ∈ N such that

j
F (k) (x) ≤ C 1 + x2 .

The space of multipliers of S(R) is denoted by OM . It is obvious that F ∈ OM

is equivalent to F ′ ∈ OM . For F ∈ OM we denote by MF the multiplication
operator
MF : S(R) → S(R), f 7→ F · f.
We now present some sufficient conditions, in terms of multipliers, for the
closed range property of composition operators.

Proposition 4.4 Let ϕ ∈ OM be a symbol for S(R) such that Mϕ′ : S(R) →
S(R) has closed range. Then Cϕ : S(R) → S(R) has also closed range.
Proof. Let (fn )n ⊂ S(R) be a sequence such that (fn ◦ ϕ)n converges to
g in S(R). According to [1, Lemma 1.1], ϕ′ does not have flat points in its
zero set and then we can apply [9] to conclude that Cϕ : C ∞ (R) → C ∞ (R)
has closed range. Consequently there is f ∈ C ∞ (R) such that g = f ◦ ϕ. We
aim to prove that f ∈ S(R). Since ϕ is a symbol with polynomial growth
we have that for each T > 0 there exists N > 0 such that

sup(1 + |ϕ(x)|2 )T |f (ϕ(x))| ≤ sup(1 + x2 )N |g(x)| < ∞. (4.3)

x∈R x∈R

11
In the case ϕ(R) = R, (4.3) is equivalent to

sup(1 + |x|2 )T |f (x)| < ∞. (4.4)

x∈R

In the case ϕ(R) = [a, ∞) we can assume without loss of generality that the
restriction of f to (−∞, a − 1) is identically null and we also

obtain (4.4).
Since (fn ◦ ϕ)n converges to g in S(R) then Mϕ′ (fn′ ◦ ϕ) n converges to
g′ = f ′ ◦ ϕ · ϕ′ in S(R). Consequently

g′ ∈ Mϕ′ S(R) = Mϕ′ S(R) .

Since the set of zeros of ϕ′ is discrete, we can conclude that

f ′ ◦ ϕ ∈ S(R).

Moreover, Mϕ′ : S(R) → S(R) is an isomorphism into its range, from where
it follows that
lim fn′ ◦ ϕ = f ′ ◦ ϕ in S(R).
n→∞

Now we can proceed inductively to prove that f (k) ◦ ϕ ∈ S(R) for every
k ∈ N. Finally, as in (4.3) and (4.4) we get

sup(1 + x2 )T |f (k) (x)| < ∞

x∈R

for every T ∈ N and k ∈ N0 . The proof is done. ✷

In [1], the multipliers of S(R) which have closed range are characterized
as the functions F ∈ OM such that there exist N, T, c > 0 such that, if we
set Ix,T := [x − 1/(1 + x2 )T , x + 1/(1 + x2 )T ], then we have for each x ∈ R
(a) The cardinality of the set Z(F ) ∩ Ix,T , the zeros counted with their
multiplicity, is smaller than N.
(b) (1 + x2 )T |F (x)| > c ki=1 |x − xi |, (xi )ki=1 being the zeros of F in Ix,T
Q
counting multiplicities. If there are no zeros in Ix,T then in the right
side one writes only c.
If I ⊂ R is a closed unbounded interval, the space S(I) can be defined
in a natural way. For example, if I = [a, +∞),

S(I) := {f ∈ C ∞ (I) : sup sup (1 + x2 )n f (j) (x) < ∞ for each n ∈ N}.
x∈I 1≤j≤n

We are considering of course the existence of right derivatives in a. Then

S(I) is a (F N )-space.

12
Remark 4.5 The proof of the main theorem of [1] (with the obvious mod-
ifications) permits to characterize the functions F : I → R such that
MF : S(I) → S(I) is well defined and has closed range as those functions

T
satisfying F (j) (x) ≤ C 1 + x2 , (T = T (j)) for every j ∈ N0 and x ∈ I
(multiplier condition) and moreover, (a) and (b) above are satisfied for each
x ∈ I, for Ix,T = [x − 1/(1 + x2 )T , x + 1/(1 + x2 )T ] ∩ I
We see below that, in the case that ϕ ∈ OM is not surjective, we only
need to require that Cϕ : C ∞ (R) → C ∞ (R) have closed range and ϕ′ satisfies
conditions (a) and (b) in some unbounded interval I.

Theorem 4.6 Let ϕ be a non surjective symbol for S(R) and assume that
Cϕ : C ∞ (R) → C ∞ (R) has closed range and there exists a closed unbounded
interval I such that the multiplication operator Mϕ′ : S(I) → S(I) is well
defined and it has closed range. Then Cϕ : S(R) → S(R) has also closed
range.
Proof. Let us assume that ϕ(R) = [a, ∞[ and ϕ′ is a closed range multiplier
for S((−∞, b]). Let (fn )n ⊂ S(R) be a sequence such that (fn ◦ϕ)n converges
in S(R). Since Cϕ : C ∞ (R) → C ∞ (R) has closed range, there exists f ∈
C ∞ (R) such that
lim (fn ◦ ϕ)n = f ◦ ϕ ∈ S(R). (4.5)
n→∞

Our aim is to show that f ∈ S(R). By cutting off after multiplying by a

convenient test function, we can assume that the support of f is contained
in [a − 1, ∞). Calculating the derivative in (4.5) we get

lim (fn′ ◦ ϕ) · ϕ′ = f ′ ◦ ϕ · ϕ′ ∈ S(R).

(4.6)
n→∞

Thus the convergence of (4.6) is also in S((−∞, b]). From the hypothesis on
ϕ′ we get that ϕ′ is a multiplier on S((−∞, b]) and then it has polynomial
increase as x → −∞. Hence also ϕ has polynomial increase as x → −∞.
Since moreover Mϕ′ : S ((−∞, b]) → S ((−∞, b]) has closed range we have
that ϕ′ has no flat points in its zero set in (−∞, b] and Mϕ′ is an isomorphism
into its image, hence

lim f ′ ◦ ϕ = f ′ ◦ ϕ ∈ S((−∞, b]).

n→∞ n

Inductively we get

lim f (j) ◦ ϕ = f (j) ◦ ϕ ∈ S((−∞, b]). (4.7)

n→∞ n

13
We use now the polynomial increase of ϕ in (−∞, b] to get that for each
k ∈ N there exist N ∈ N and C > 0 such that

lim (1 + ϕ(x)2 )k f (j) (ϕ(x)) ≤ C lim (1 + x2 )N (f (j) ◦ ϕ)(x) = 0

x→−∞ x→−∞

for each 1 ≤ j ≤ k. Finally, as in 4.4, from ϕ(R) = [a, ∞[, lim ϕ(x) = +∞
x→−∞
and f identically null in (−∞, a − 1] we get

k
lim 1 + x2 |f (j) (x)| = 0
|x|→+∞

for every 1 ≤ j ≤ k. In the case that ϕ′ is a closed range multiplier for

S([b, ∞)) the argument is similar. ✷

From the proof, it follows the following version for surjective symbols.

Theorem 4.7 Let ϕ be a surjective symbol for S(R) such that Cϕ : C ∞ (R) →
C ∞ (R) has closed range. Assume there exist closed unbounded intervals
I1 , I2 such that R \ (I1 ∪ I2 ) is bounded and the multiplication operator
Mϕ′ : S(Ij ) → S(Ij ) is well defined and it has closed range, j = 1, 2. Then
Cϕ : S(R) → S(R) has also closed range.
The examples below illustrate that for non surjective symbols we need
only one good branch to have closed range. However, for surjective symbols
we need control on both branches.

Example 4.8 (a) Consider the function

−e−x + 2, x ≤ 0

ϕ1 (x) :=
x, x≥1
and let ϕˆ1 a C ∞ extension of ϕ1 to R, which always exists by Whitney’s
theorem [15]. Then ϕˆ1 is a surjective symbol and Cϕˆ1 : S(R) →
S(R) does not have closed range by Proposition 4.2 (see the proof
of Corollary 4.3). Since ϕˆ1 satisfies that for each x ∈ R there exists
y ∈ ϕˆ1 −1 (x) ∩ (R \ (0, 1)), and then ϕˆ1 ′ (y) = ϕ′ (y) 6= 0, we conclude
from the main theorem in [9] that Cϕˆ1 : C ∞ (R) → C ∞ (R) has closed
range.
1
(b) Let x0 > 1 such that x0 e−x0 = 2e and consider the function
 −x
xe x ≤ x0
ϕ2 (x) := 1
(2x0 − x) + 2e x ≥ 2x0

14
 Let us observe that ϕ2 (R \ (x0 , 2x0 )) ⊆ (−∞, e−1 ] and
1
ϕ2 (x0 ) = ϕ2 (2x0 ) = .
2e
If ϕˆ2 is an smooth extension of ϕ2 which is real analytic on (x0 , 2x0 ),
extension that always exists due to a Whitney extension’s theorem
[15], then ϕˆ2 does not have any flat critical point. A similar approach
to that of the previous example gives that Cϕˆ2 : C ∞ (R) → C ∞ (R)
has closed range. Now we apply Theorem 4.6 to conclude that also
Cϕˆ2 : S(R) → S(R) has closed range.
Finally we obtain some kind of converse of Proposition 4.4 and Theorem
4.6.

Theorem 4.9 Let ϕ be a surjective symbol for S(R) such that Cϕ : S(R) →
S(R) has closed range. We assume that there exists k such that ϕ′ (x) 6= 0
if |x| > k. Then there exists a closed unbounded interval I such that the
continuous linear mapping Cϕ : S(ϕ(I)) → S(I) is surjective. Moreover

Mϕ′ : S(I) → S(I)

is well defined and it has closed range of codimension less or equal 1.

Proof. Since f (x) 7→ f (−x) is an isometry we assume without loss of
generality that

lim ϕ(x) = +∞ and lim ϕ(x) = −∞.

x→+∞ x→−∞

From the hypothesis we can take a > 0 such that ϕ′ (x) > 0 for all x ∈
[a, ∞) and ϕ−1 (ϕ[a, ∞)) = [a, ∞). We consider I = [a, ∞) and denote
ϕ(I) = [b, ∞). Let ε > 0 such that ϕ′ > 0 in [a − ε, a + ε] and let δ > 0 and
η > 0 such that ϕ([a − ε, a]) = [b − δ, b] and ϕ([a, a + ε]) = [b, b + η].
We check first that Cϕ : S(ϕ(I)) → S(I) has closed range. Notice that
the mapping is well defined since ϕ is a symbol for S(R). Assume there is
(fn )n in S(ϕ(I)) such that (fn ◦ϕ)n converges to g in S(I). We aim to obtain
f ∈ S(ϕ(I)) such that g = f ◦ ϕ. By Seeley-Mityagin’s theorem [8, 12] there
exists a continuous linear extension operator E : C ∞ (I) → C ∞ (R). Let ρ
be a test function such that 0 ≤ ρ ≤ 1 and with the properties that there
is 0 < t < ε/2 with ρ = 1 in (a − t, a + t) and its support is contained in
(a− ε/2, a+ ε/2). For f ∈ C ∞ (I), we define T (f )(x) = E(f )(x) if x ≥ a and
T (f )(x) = E(f )(x)ρ(x) if x < a. Now T : C ∞ (I) → C ∞ (R) is a continuous
linear extension operator and

15
 supp (T (C ∞ (I))) ⊆ (a − ε/2, ∞). (4.8)
In the above formula we mean that the supports of the functions in
T (C ∞ (I))) are closed intervals contained in (a − ε/2, ∞). Since ϕ−1 is
smooth in [b−δ, b+η] we can use Whitney’s extension theorem to consider an
extension γ of (ϕ|[a−ε,a+ε] )−1 which is smooth in (−∞, b+η). Since γ(b−δ) =
a − ε, there is s > 0 such that γ(x) < a − ε/2 if x ∈ (b − δ − 2s, b − δ + 2s).
If we consider α ∈ D(b − δ − 2s, b − δ + 2s) such that 0 ≤ α ≤ 1 and α = 1
in (b − δ − s, b − δ + s), we can define

(ϕ|[a−ε,a+ε] )−1 (x), x ∈ [b − δ, b + η)


h(x) := .
(αγ)(x), x ∈ (−∞, b − δ)
We have that h : (−∞, b+η) → R is a smooth extension of (ϕ|[a−ε,a+ε] )−1
which satisfies

h(−∞, b − δ) ⊆ (−∞, a − ε/2) (4.9)

and also

h(x) = 0 for x ≤ b − δ − 2s. (4.10)

Now we define the following function

ˆ T (fn ◦ ϕ) ◦ h(x), x ∈ (−∞, b)
fn (x) :=
fn (x), x ∈ [b, ∞)

Notice that differentiability of fn implies that fˆn is smooth on a neighbour-

hood on b. We mean, near b we have fˆn (x) = T (fn ◦ ϕ) ◦ h(x) if x < b
and also fˆn (x) = fn (x) = fn ◦ ϕ ◦ (ϕ|[a−ε,a+ε] )−1 (x) = T (fn ◦ ϕ) ◦ h(x) if
x ∈ [b, b + η), i.e fˆn near b is the composition between two smooth functions.
By (4.8) and (4.9) we can write

 0, x ∈ (−∞, b − δ)
fˆn (x) = T (fn ◦ ϕ) ◦ (ϕ|[a−ε,a] )−1 (x), x ∈ [b − δ, b]
fn (x), x ∈ [b, ∞)


Since fn ∈ S(ϕ(I)) we get fˆn ∈ S(R). Moreover we have

x ∈ ϕ−1 (−∞, b − δ)

   0,
ˆ
fn ◦ ϕ (x) = T (fn ◦ ϕ)(x), x ∈ ϕ−1 [b − δ, b]
(fn ◦ ϕ)(x), x ∈ ϕ−1 (b, ∞) = (a, ∞)


16
 Since ϕ is semiproper then K = ϕ−1 ([b − δ, b]) is a compact set. From
the fact that T is an extension operator it follows that T (fn ◦ ϕ) converges
to T (g) in C ∞ (R). In particular, T (fn ◦ ϕ) and all its derivatives converge
uniformly to T (g) on the compact set K. Together with the convergence
of (fn ◦ ϕ)n to g in S(I) we get that fˆn ◦ ϕ converges to T (g) in S(R).
Hence there exists fˆ ∈ S(R) such that T (g) = fˆ ◦ ϕ. If we denote by f the
restriction of fˆ to ϕ(I), we have that f ∈ S(ϕ(I)) and g = f ◦ ϕ.
Each f ∈ S(I) which is compactly supported satisfies that f ◦ ϕ−1 is
compactly supported in ϕ(I) and C ∞ . Since Cϕ : S(ϕ(I)) → S(I) has
closed range and the range contains all compactly supported functions in
S(I) we conclude that it is surjective. We proceed to show that under these
conditions ϕ′ is a multiplier for S(I) and Mϕ′ : S(I) → S(I) has closed
range.
Claim. (f ◦ ϕ)ϕ′ ∈ S(I) for each f ∈ S(ϕ(I))
Proof of the claim. Since ϕ is a symbol for S(R), for each n ∈ N there exists
N ∈ N such that |ϕ(j) (x)| ≤ (1 + |ϕ(x))|N for 1 ≤ j ≤ n and there exists
k ∈ N such that |x| ≤ |ϕ(x)|k if |x| > k. Applying Faá di Bruno formula and
Leibnitz rule we can then obtain that for each n, k ∈ N there exist C, M > 0
such that

n
X
sup(1 + |x|2 )k ((f ◦ ϕ)ϕ′ )(n) (x) ≤ C sup (1 + |t|2 )M |f (j) (t)|,
x∈I t∈ϕ(I) j=1

the right term being bounded since f ∈ S(ϕ(I)). The claim is proved.
Surjectivity of Cϕ : S(ϕ(I)) → S(I) permits to conclude that

Mϕ′ : S(I) → S(I)

is well defined, and then continuous by the closed graph theorem.

Let T : C ∞ (I) → C ∞ (R) be the continuos linear extension operator used
in the first part of the proof. The construction implies T (S(I)) ⊆ S(R) and
then T : S(I) → S(R) is continuous by the closed graph theorem. Define
v ∈ S(I)′ as
Z ∞
v(f ) = T (f )(t)dt.
−∞
√ 2 R∞
Let f0 = (1/ π)e−x ∈ S(R) with −∞ f0 (t)dt = 1. For f ∈ S(I) define
Z x
g(x) = (T (f )(t) − λf0 (t))dt λ = v(f ).
−∞

17
We have by L’Hopital rule that g ∈ S(R) and

T (f ) = g ′ + λf0 , λ = v(f ).
From this we get that for each f ∈ S(ϕ(I)) there exists g ∈ S(R) such that

T (f ◦ ϕ) = g ′ + λf0 , λ = v(f ◦ ϕ).

If we restrict to I we get

(f ◦ ϕ) = g′ + λf0 , λ = v(f ◦ ϕ). (4.11)

Surjectivity of Cϕ : S(ϕ(I)) → S(I) permits to get h ∈ S(ϕ(I)) such that
g(x) = (h ◦ ϕ) (x) for each x ∈ I. This means that if f ∈ S(ϕ(I)) and
f ◦ ϕ ∈ Ker(v) then

f ◦ ϕ = (h′ ◦ ϕ)ϕ′ ∈ Mϕ′ (S(I)).

Since S(I) = {f ◦ ϕ : f ∈ S(ϕ(I)} we conclude that Mϕ′ (S(I)) contains the
closed hyperplane Ker(v), and then it is closed.
✷

Theorem 4.10 Let ϕ be a surjective symbol for S(R) such that Cϕ : S(R) →
S(R) has closed range. Assume that there exists k such that ϕ′ (x) 6= 0 if
|x| > k. Then ϕ ∈ OM and there exist c > 0, T > 0 such that

−T
ϕ′ (x) ≥ c 1 + x2

for |x| large enough.

Proof. The proof of Theorem 4.9 shows that under the hypothesis there are
in fact two intervals, I1 ⊂ [0, ∞) and I2 ⊂ [−∞, 0) such that Mϕ′ : S(Ij ) →
S(Ij ) is a well defined closed range operator. Hence we get the conclusion
from Remark 4.5 and the equivalence ϕ ∈ OM if and only if ϕ′ ∈ OM . ✷

Acknowledgements
The research was partially supported by the projects MTM2013-43540-P,
GVA Prometeo II/2013/013 and ACOMP/2015/186.

18
References
[1] J. Bonet, L. Frerick, E. Jordá; The division problem for tempered dis-
tributions of one variable. J. Funct. Anal.262 (2012), 2349–2358.

[2] C.C. Cowen, B.D. MacCluer; Composition operators on spaces of an-

alytic functions. Studies in Advanced Mathematics. CRC Press, Boca
Raton, FL, 1995.

[3] C. Fernández, A. Galbis; Superposition in classes of ultradifferentiable

functions. Publ. Res. Inst. Math. Sci. 42 (2006), 399–419.

[4] G. Glaeser; Fonctions composes diffrentiables. (French) Ann. of Math.

(2) 77 (1963), 193–209.

[5] S. Krantz, H. Parks; A primer on real analytic functions. Birkhäuser,

Basel (1992).

[6] N. Kenessey, J. Wengenroth; Composition operators with closed range

for smooth injective symbols R → Rd . J. Funct. Anal. 260 (2011), 2997–
3006.

[7] B. Malgrange; Ideals of differentiable functions. Tata Institute of Fun-

damental Research Studies in Mathematics, No. 3 Tata Institute of
Fundamental Research, Bombay; Oxford University Press, London 1967
vii+106 pp.

[8] B. S. Mitjagin; Approximate dimension and bases in nuclear spaces.

(Russian) Uspehi Mat. Nauk 16 100 (1961) no. 4, 63–132.

[9] A. Przestacki; Composition operators with closed range for one-

dimensional smooth symbols. J. Math. Anal. Appl. 399 (2013), 225–
228.

[10] A. Przestacki; Characterization of composition operators with closed

range for one-dimensional smooth symbols. J. Funct. Anal. 266 (2014),
5847–5857.

[11] A. Przestacki; Corrigendum to ”Characterization of composition oper-

ators with closed range for one-dimensional smooth symbols” [J. Funct.
Anal. 266 (9) (2014) 5847–5857]. J. Funct. Anal. 269 (2015), no. 8,
2665–667.

19
[12] R. T. Seeley; Extension of C ∞ functions defined in a half space. Proc.
Amer. Math. Soc. 15 (1964), 625–626.

[13] J. H. Shapiro; Composition operators and classical function theory. Uni-

versitext: Tracts in Mathematics. Springer-Verlag, New York, 1993.

[14] J. C. Tougeron; Idéaux de fonctions différentiables. Ergebnisse der

Mathematik und ihrer Grenzgebiete, Band 71. Springer-Verlag, Berlin-
New York, 1972. vii+219 pp.

[15] H. Whitney; Analytic extensions of differentiable functions defined in

closed sets. Trans. Amer. Math. Soc. 36 (1934), no. 1, 63–89.

Antonio Galbis: Departament d’Anàlisi Matemàtica, Universitat de València,

46100 Burjassot (València), Spain. E-mail: [Link]@[Link]
Enrique Jordá: Instituto Universitario de MAtemática Pura y Aplicada
IUMPA, Universitat Politècnica de València, Plaza Ferrándiz y Carbonell,
s/n E-03801 Alcoy (Alicante), Spain. E-mail: ejorda@[Link]

20