a

r
X
i
v
:
0
8
0
9
.
0
5
7
0
v
1


[
m
a
t
h
.
F
A
]


3

S
e
p

2
0
0
8
On sharp embeddings of Besov and Triebel-Lizorkin spaces
in the subcritical case
Jan Vybral
Mathematisches Institut, Universitat Jena
Ernst-Abbe-Platz 2, 07740 Jena, Germany
email: vybiral@mathematik.uni-jena.de
September 3, 2008
Abstract
We discuss the growth envelopes of Fourier-analytically dened Besov and Triebel-Lizorkin
spaces B
s
p,q
(R
n
) and F
s
p,q
(R
n
) for s =
p
= nmax(
1
p
1, 0). These results may be also reformu-
lated as optimal embeddings into the scale of Lorentz spaces L
p,q
(R
n
). We close several open
problems outlined already by H. Triebel in [11] and explicitly formulated by D. D. Haroske in
[4].
AMS Classication: 46E35, 46E30
Keywords and phrases: Besov spaces, Triebel-Lizorkin spaces, rearrangement invariant spaces,
Lorentz spaces, growth envelopes
J. Vybral On sharp embeddings . . .
1 Introduction and main results
We denote by B
s
p,q
(R
n
) and F
s
p,q
(R
n
) the Fourier-analytic Besov and Triebel-Lizorkin spaces (see
Denition 2.4 for details). The embeddings of these function spaces (and other spaces of smooth
functions) play an important role in functional analysis. If s >
n
p
, then these spaces are continuously
embedded into C(R
n
), the space of all complex-valued bounded and uniformly continuous functions
on R
n
normed in the usual way. If s <
n
p
then these function spaces contain also unbounded
functions. This statement holds true also for s =
n
p
under some additional restrictions on the
parameters p and q. We refer to [8, Theorem 3.3.1] for a complete overview.
To describe the singularities of these unbounded elements, we use the technique of the non-
increasing rearrangement.
Denition 1.1. Let be the Lebesgue measure in R
n
. If h is a measurable function on R
n
, we
dene the non-increasing rearrangement of h through
h

(t) = sup{ > 0 : {x R

n
: |h(x)| > } > t}, t (0, ). (1.1)
To be able to apply this procedure to elements of A
s
p,q
(R
n
) (with A standing for B or F), we have to
know whether all the distributions of A
s
p,q
(R
n
) may be interpreted as measurable functions. This
is the case if, and only if, A
s
p,q
(R
n
) L
loc
1
(R
n
), the space of all measurable, locally-integrable
functions on R
n
. A complete treatment of this question may be found in [8, Theorem 3.3.2]:
B
s
p,q
(R
n
) L
loc
1
(R
n
)
_

_
either s >
p
:= nmax(
1
p
1, 0),
or s =
p
, 1 < p , 0 < q min(p, 2),
or s =
p
, 0 < p 1, 0 < q 1
(1.2)
and
F
s
p,q
(R
n
) L
loc
1
(R
n
)
_

_
either s >
p
,
or s =
p
, 1 p < , 0 < q 2,
or s =
p
, 0 < p < 1, 0 < q .
(1.3)
Let us assume, that a function space X is embedded into L
loc
1
(R
n
). The growth envelope function
of X was dened by D. D. Haroske and H. Triebel (see [3], [4], [11] and references given there) by
E
X
G
(t) := sup
||f|X||1
f

(t), 0 < t < 1.

If E
X
G
(t) t

for 0 < t < 1 and some > 0, then we dene the growth envelope index u
X
as the OK?
inmum of all numbers v, 0 < v , such that
__

0
_
f

(t)
E
X
G
(t)
_
v
dt
t
_
1/v
c ||f|X|| (1.4)
(with the usual modication for v = ) holds for some > 0, c > 0 and all f X.
The pair E
G
(X) = (E
X
G
, u
X
) is called growth envelope for the function space X.
In the case
p
< s <
n
p
, the growth envelopes of A
s
p,q
(R
n
) are known, cf. [11, Theorem 15.2] and
[4, Theorem 8.1]. If s =
p
and (1.2) or (1.3) is fullled in the B or F case, respectively, then the
known information is not complete, cf. [11, Rem. 12.5, 15.1] and [4, Prop. 8.12, 8.14 and Rem.
8.15]:
Theorem 1.2. (i) Let 1 < p < and 0 < q min(p, 2). Then
E
G
(B
0
p,q
) = (t

1
p
, u) with q u p.
2
J. Vybral On sharp embeddings . . .
(ii) Let 1 p < and 0 < q 2. Then
E
G
(F
0
p,q
) = (t

1
p
, p).
(iii) Let 0 < p 1 and 0 < q 1. Then
E
G
(B
p
p,q
) = (t
1
, u) with q u 1.
(iv) Let 0 < p < 1 and 0 < q . Then
E
G
(F
p
p,q
) = (t
1
, u) with p u 1.
We ll all the above mentioned gaps.
Theorem 1.3. (i) Let 1 p < and 0 < q min(p, 2). Then
E
G
(B
0
p,q
) = (t

1
p
, p).
(ii) Let 0 < p < 1 and 0 < q 1. Then
E
G
(B
p
p,q
) = (t
1
, q).
(iii) Let 0 < p < 1 and 0 < q . Then
E
G
(F
p
p,q
) = (t
1
, p).
We also reformulate these results as optimal embeddings into the scale of Lorentz spaces (cf.
Denition 2.1):
Theorem 1.4. (i) Let 1 p < and 0 < q min(p, 2). Then
B
0
p,q
(R
n
) L
p
(R
n
).
(ii) Let 0 < p < 1 and 0 < q 1. Then
B
p
p,q
(R
n
) L
1,q
(R
n
). (1.5)
(iii) Let 0 < p < 1 and 0 < q . Then
F
p
p,q
(R
n
) L
1,p
(R
n
)
and all these embeddings are optimal with respect to the second ne parameter of the scale of the
Lorentz spaces.
Remark 1.5. (i) Let us also observe, that (1.5) improves [8, Theorem 3.2.1] and [7, Theorem 2.2.3],
where the embedding B
n(
1
p
1)
p,q
(R
n
) L
1
(R
n
) is proved for all 0 < p < 1 and 0 < q 1.
(ii) Let us also mention, that growth envelopes for function spaces with minimal smoothness were
recently studied in [2]. These authors worked with spaces dened by dierences and therefore their
results are of a dierent nature.
3
J. Vybral On sharp embeddings . . .
2 Preliminaries, notation and denitions
We use standard notation: N denotes the collection of all natural numbers, R
n
is the Euclidean
n-dimensional space, where n N, and C stands for the complex plane.
Denition 2.1. (i) Let 0 < p . We denote by L
p
(R
n
) the Lebesgue spaces endowed with the
quasi-norm
||f|L
p
(R
n
)|| =
_

_
__
R
n
|f(x)|
p
dx
_
1/p
, 0 < p < ,
ess sup
xR
n
|f(x)|, p = .
(ii) Let 0 < p, q . Then the Lorentz space L
p,q
(R
n
) consists of all f L
loc
1
(R
n
) such that the
quantity
||f|L
p,q
(R
n
)|| =
_

_
__

0
[t
1
p
f

(t)]
q
dt
t
_
1/q
, 0 < q < ,
sup
0<t<
t
1
p
f

(t), q =
is nite
Remark 2.2. These denitions are well-known, we refer to [1, Chapter 4.4] for details and further
references. We shall need only very few properties of these spaces. Obviously, L
p,p
(R
n
) = L
p
(R
n
).
If 0 < q
1
q
2
, then L
p,q
1
(R
n
) L
p,q
2
(R
n
) - so the Lorentz spaces are monotonically ordered
in q. We shall make use of the following lemma:
Lemma 2.3. Let 0 < q < 1. Then the || |L
1,q
(R
n
)|| is the q-norm, it means
||f
1
+f
2
|L
1,q
(R
n
)||
q
||f
1
|L
1,q
(R
n
)||
q
+||f
2
|L
1,q
(R
n
)||
q
holds for all f
1
, f
2
L
1,q
(R
n
).
Proof. First note, that the function s s
q
is increasing for all 0 < q < on (0, ). This leads to
the identity
(|f|
q
)

(t) = (f

)
q
(t), (2.1)
which holds for all t > 0, 0 < q < and all measurable functions f. The reader may also consult
[1, Proposition 2.1.7]. Using (2.1) and 0 < q < 1, we obtain
||f
1
+f
2
|L
1,q
(R
n
)||
q
=
_

0
t
q1
((f
1
+f
2
)

(t))
q
dt
_

0
t
q1
((|f
1
| +|f
2
|)

(t))
q
dt
=
_

0
t
q1
((|f
1
| +|f
2
|)
q
)

(t)dt
_

0
t
q1
(|f
1
|
q
+|f
2
|
q
)

(t)dt.
We observe, that t :t
q1
is a decreasing function on (0, ) and that
_

0
(|f
1
|
q
+|f
2
|
q
)

(t)dt
_

0
(|f
1
|
q
)

(t)dt +
_

0
(|f
2
|
q
)

(t)dt
holds for all (0, ). Hence, by Hardys lemma (cf. [1, Proposition 2.3.6]),
||f
1
+f
2
|L
1,q
(R
n
)||
q

_

0
t
q1
(|f
1
|
q
)

(t)dt +
_

0
t
q1
(|f
2
|
q
)

(t)dt
= ||f
1
|L
1,q
(R
n
)||
q
+||f
2
|L
1,q
(R
n
)||
q
.
4
J. Vybral On sharp embeddings . . .
Let S(R
n
) be the Schwartz space of all complex-valued rapidly decreasing, innitely dierentiable
functions on R
n
and let S

(R
n
) be its dual - the space of all tempered distributions.
For f S

(R
n
) we denote by

f = Ff its Fourier transform and by f

or F
1
f its inverse Fourier
transform.
We give a Fourier-analytic denition of Besov and Triebel-Lizorkin spaces, which relies on the
so-called dyadic resolution of unity. Let S(R
n
) with
(x) = 1 if |x| 1 and (x) = 0 if |x|
3
2
. (2.2)
We put
0
= and
j
(x) = (2
j
x) (2
j+1
x) for j N and x R
n
. This leads to the identity

j=0

j
(x) = 1, x R
n
.
Denition 2.4. (i) Let s R, 0 < p, q . Then B
s
pq
(R
n
) is the collection of all f S

(R
n
) such
that
||f|B
s
pq
(R
n
)|| =
_

j=0
2
jsq
||(
j

f)

|L
p
(R
n
)||
q
_
1/q
< (2.3)
(with the usual modication for q = ).
(ii) Let s R, 0 < p < , 0 < q . Then F
s
pq
(R
n
) is the collection of all f S

(R
n
) such that
||f|F
s
pq
(R
n
)|| =

j=0
2
jsq
|(
j

f)

()|
q
_
1/q
|L
p
(R
n
)

< (2.4)
(with the usual modication for q = ).
Remark 2.5. These spaces have a long history. In this context we recommend [6], [9], [10] and
[12] as standard references. We point out that the spaces B
s
pq
(R
n
) and F
s
pq
(R
n
) are independent
of the choice of in the sense of equivalent (quasi-)norms. Special cases of these two scales in-
clude Lebesgue spaces, Sobolev spaces, Holder-Zygmund spaces and many other important function
spaces.
We introduce the sequence spaces associated with the Besov and Triebel-Lizorkin spaces. Let
m Z
n
and j N
0
. Then Q
j m
denotes the closed cube in R
n
with sides parallel to the coordinate
axes, centred at 2
j
m, and with side length 2
j
. By
j m
=
Q
j m
we denote the characteristic
function of Q
j m
. If
= {
j m
C : j N
0
, m Z
n
},
< s < and 0 < p, q , we set
|||b
s
pq
|| =
_

j=0
2
j(s
n
p
)q
_

mZ
n
|
j m
|
p
_q
p
_1
q
(2.5)
appropriately modied if p = and/or q = . If p < , we dene also
|||f
s
pq
|| =

j=0

mZ
n
|2
js

j m

j m
()|
q
_
1/q
|L
p
(R
n
)

. (2.6)
The connection between the function spaces B
s
pq
(R
n
), F
s
pq
(R
n
) and the sequence spaces b
s
pq
, f
s
pq
may be given by various decomposition techniques, we refer to [12, Chapters 2 and 3] for details
and further references.
All the unimportant constants are denoted by the letter c, whose meaning may dier from one
occurrence to another. If {a
n
}

n=1
and {b
n
}

n=1
are two sequences of positive real numbers, we
write a
n
b
n
if, and only if, there is a positive real number c > 0 such that a
n
c b
n
, n N.
Furthermore, a
n
b
n
means that a
n
b
n
and simultaneously b
n
a
n
.
5
J. Vybral On sharp embeddings . . .
3 Proofs of the main results
3.1 Proof of Theorem 1.3 (i)
In view of Theorem 1.2, it is enough to prove, that for 1 p < and 0 < q min(p, 2) the index
u associated to B
0
p,q
(R
n
) is greater or equal to p.
We assume in contrary that (1.4) is fullled for some 0 < v < p, > 0, c > 0 and all f B
0
p,q
(R
n
).
Let be a non-vanishing C

function in R
n
supported in [0, 1]
n
with
_
R
n
(x)dx = 0.
Let J N be such that 2
Jn
< and consider the function
f
j
=
2
(jJ)n

m=1

j m
(2
j
(x (m, 0, . . . , 0))), j > J, (3.1)
where

j m
=
1
m
1
p
log
1
v
(m+ 1)
, m = 1, . . . , 2
(jJ)n
Then (3.1) represents an atomic decomposition of f in the space B
0
p,q
(R
n
) according to [12, Chapter
1.5] and we obtain (recall that v < p)
||f
j
|B
0
p,q
(R
n
)|| 2
j
n
p
_
_
2
(jJ)n

m=1

p
j m
_
_
1/p
2
j
n
p
_

m=1
m
1
(log(m + 1))

p
v
_
1/p
2
j
n
p
. (3.2)
On the other hand,
__

0
_
f

j
(t)t
1
p
_
v
dt
t
_
1/v

_
_
2
Jn
0
f

j
(t)
v
t
v/p1
dt
_
1/v

_
_
2
(jJ)n

m=1

v
j m
_
c 2
jn
m
c 2
jn
(m1)
t
v/p1
dt
_
_
1/v

_
_
2
(jJ)n

m=1

v
j m
2
jnv/p
m
v/p1
_
_
1/v
= 2
j
n
p
_
_
2
(jJ)n

m=1
1
mlog(m+ 1)
_
_
1/v
.
As the last series is divergent for j , this is in a contradiction with (3.2) and (1.4) cannot hold
for all f
j
, j > J.
Remark 3.1. Observe, that Theorem 1.4 (i) is a direct consequence of Theorem 1.3 (i). The
embeddings B
0
1,q
(R
n
) B
0
1,1
(R
n
) L
1
(R
n
) if p = 1 and B
0
p,q
(R
n
) F
0
p,2
(R
n
) = L
p
(R
n
) if 1 <
p < show, that B
0
p,q
(R
n
) L
p
(R
n
). And Theorem 1.3 (i) implies that if B
0
p,q
(R
n
) L
p,v
(R
n
)
for some 0 < v < , then p v. This proves the optimality of Theorem 1.4 (i) in the frame of the
scala of Lorentz spaces.
3.2 Proof of Theorem 1.3 (ii) and Theorem 1.4 (ii)
Let 0 < p < 1, 0 < q 1 and s =
p
= n
_
1
p
1
_
. We prove rst Theorem 1.4 (ii), i.e. we show
that
B
n
p
n
p,q
(R
n
) L
1,q
(R
n
),
or, equivalently,
__

0
[tf

(t)]
q
dt
t
_
1/q
c ||f|B
n
p
n
p,q
(R
n
)||, f B
n
p
n
pq
(R
n
).
6
J. Vybral On sharp embeddings . . .
Let
f =

j=0
f
j
=

j=0

mZ
n

j m
a
j m
be the optimal atomic decomposition of an f B
n
p
n
p,q
(R
n
), again in the sense of [12]. Then
||f|B
n
p
n
p,q
(R
n
)||
_
_

j=0
2
jqn
_

mZ
n
|
j m
|
p
_
q/p
_
_
1/q
(3.3)
and by Lemma 2.3
||f|L
1,q
(R
n
)|| = ||

j=0
f
j
|L
1,q
(R
n
)||
_
_

j=0
||f
j
|L
1,q
(R
n
)||
q
_
_
1/q
. (3.4)
We shall need only one property of the atoms a
j m
, namely that their support is contained in the
cube

Q
j m
- a cube centred at the point 2
j
m with sides parallel to the coordinate axes and side
length 2
j
, where > 1 is xed and independent of f. We denote by
j m
(x) the characteristic
functions of

Q
j m
and by
j l
the characteristic function of the interval (l2
jn
, (l + 1)2
jn
). Hence
f
j
(x) c

mZ
n
|
j m
|
j m
(x), x R
n
and
||f
j
|L
1,q
(R
n
)||
_
_

0

l=0
[(
j
)

j l
(t)]
q
t
q1
dt
_
1/q

l=0
[(
j
)

l
]
q
_
2
jn
(l+1)
2
jn
l
t
q1
dt
_
1/q
2
jn
_

l=0
[(
j
)

l
]
q
(l + 1)
q1
_
1/q
2
jn
||
j
|
p
||. (3.5)
The last inequality follows by (l + 1)
q1
1 and
p

q
if p q. If p > q, the same follows by
Holders inequality with respect to indices =
p
q
and

=
p
pq
:
_

l=0
[(
j
)

l
]
q
(l + 1)
q1
_
1/q

l=0
[(
j
)

l
]
q
p
q
_1
q

q
p

l=0
(l + 1)
(q1)
p
pq
_1
q

pq
p
c ||
j
|
p
||.
Here, we used that for 0 < q < p < 1 the exponent
(q1)p
pq
= 1+
(p1)q
pq
is strictly smaller than 1.
The proof now follows by (3.3), (3.4) and (3.5).
||f|L
1,q
(R
n
)||
_
_

j=0
||f
j
|L
1,q
(R
n
)||
q
_
_
1/q
c
_
_

j=0
2
jnq
||
j
|
p
||
q
_
_
1/q
c ||f|B
p
p,q
(R
n
)||.
Remark 3.2. We actually proved, that (1.4) holds for X = B
n
p
n
pq
(R
n
), v = q and = . This,
together with Theorem 1.2 (iii) implies immediately Theorem 1.3 (ii).
3.3 Proof of Theorem 1.3 (iii) and Theorem 1.4 (iii)
Let 0 < p < 1 and 0 < q . By the Jawerth embedding (cf. [5] or [13]) and Theorem 1.3 (ii) we
get for any 0 < p < p < 1
F
p
p,q
(R
n
) B

p
p,p
(R
n
) L
1,p
(R
n
).
7
J. Vybral On sharp embeddings . . .
References
[1] C. Bennett and R. Sharpley, Interpolation of operators, Academic Press, San Diego, 1988.
[2] A. M. Caetano, A. Gogatishvili and B. Opic, Sharp embeddings of Besov spaces involving only
logarithmic smoothness, J. Appr. Theory 152 (2008), 188-214.
[3] D. D. Haroske, Limiting embeddings, entropy numbers and envelopes in function spaces, Ha-
bilitationsschrift, Friedrich-Schiller-Universit at Jena, Germany, 2002.
[4] D. D. Haroske, Envelopes and sharp embeddings of function spaces, Chapman & Hall / CRC,
Boca Raton, 2007.
[5] B. Jawerth, Some observations on Besov and Lizorkin-Triebel spaces, Math. Scand. 40 (1977),
94-104.
[6] J. Peetre, New thoughts on Besov spaces, Duke Univ. Math. Series, Durham, 1976.
[7] W. Sickel and T. Runst, Sobolev spaces of fractional order, Nemytskij operators, and nonlinear
partial dierential equations. de Gruyter Series in Nonlinear Analysis and Applications, 3.
Walter de Gruyter & Co., Berlin, 1996.
[8] W. Sickel and H. Triebel, H older inequalities and sharp embeddings in function spaces of B
s
pq
and F
s
pq
type, Z. Anal. Anwendungen, 14 (1995), 105-140.
[9] H. Triebel, Theory of function spaces, Birkhauser, Basel, 1983.
[10] H. Triebel, Theory of function spaces II, Birkhauser, Basel, 1992.
[11] H. Triebel, The structure of functions, Birkhauser, Basel, 2001.
[12] H. Triebel, Theory of function spaces III, Birkhauser, Basel, 2006.
[13] J. Vybral, A new proof of the Jawerth-Franke embedding, Rev. Mat. Complut. 21 (2008),
75-82.
8