THE OPTIMAL EXPONENT IN THE EMBEDDING INTO THE

LEBESGUE SPACES FOR FUNCTIONS WITH GRADIENT IN THE

MORREY SPACE

XAVIER CABRÉ AND FERNANDO CHARRO

arXiv:1907.12982v1 [math.AP] 30 Jul 2019

Abstract. We study the following natural question that, apparently, has not been
well addressed in the literature: Given functions u with support in the unit ball
B1 ⊂ Rn and with gradient in the Morrey space M p,λ (B1 ), where 1 < p < λ < n,
what is the largest range of exponents q for which necessarily u ∈ Lq (B1 )? While
David R. Adams proved in 1975 that this embedding holds for q ≤ λp/(λ − p), an
article from 2011 claimed the embedding in the larger range q < np/(λ − p). Here we
disprove this last statement by constructing a function that provides a counterexample
for q > λp/(λ − p). The function is basically a negative power of the distance to a
set of Haussdorf dimension n − λ. When λ ∈ / Z, this set is a fractal. We also make
a detailed study of the radially symmetric case, a situation in which the exponent q
can go up to np/(λ − p).

1. Introduction
In this article we address the following natural question: Given functions u with
support in the unit ball B1 (0) ⊂ Rn and with gradient in the Morrey space M p,λ (B1 (0)),
where 1 < p < λ < n, what is the optimal range of exponents q such that necessarily
u ∈ Lq (B1 (0))? Apparently, this question has not been well addressed in the literature.
There have been even some important errata on such an exponent.
Our motivation came from the recent work [5] of the first author in collaboration
with A. Figalli, X. Ros-Oton, and J. Serra, on the regularity of stable solutions to
semilinear elliptic equations. The reason is that the results of [5] are based, among
other tools, on Morrey-type estimates for the gradient of a stable solution (see Remark
1.2 below for more details).
The following is the precise statement of the question that we are concerned with.
Given real numbers p and λ such that
1 < p < λ < n,
we wish to know for which exponents q the inequality
kukLq (B1 (0)) ≤ C k∇ukM p,λ(B1 (0)) (1.1)

Key words and phrases. Morrey spaces, Optimal embeddings, Cantor sets.
2010 Mathematics Subject Classification: 42B37, 46E35.
The authors were supported by MINECO grant MTM2014-52402-C3-1-P. X. Cabré is also sup-
ported by MINECO grants MTM2017-84214-C2-1-P and MDM-2014-0445, and is member of the
Catalan research group 2017 SGR 1392. F. Charro was also partially supported by a Juan de la
Cierva fellowship.
1
2 X. CABRÉ AND F. CHARRO

holds true for functions u with support in B1 (0) ⊂ Rn and for a constant C independent
of u, where  Z 
k∇ukpM p,λ(Ω) := sup r λ−n |∇u(x)|p dx
r>0, y∈Ω Ω ∩Br (y)
n
is the Morrey norm of ∇u in a domain Ω ⊂ R . Notice that when λ equals the
dimension n and Ω = B1 (0), (1.1) corresponds to the Sobolev inequality in B1 (0) ⊂ Rn .
In 1975, D. Adams [1, Theorem 3.1] proved the following result. Let us denote in
the sequel
λp np
p1 := and p2 := .
λ−p λ−p
Observe that, clearly, p1 < p2 .
Theorem 1.1 (D. Adams [1]). Let p, λ ∈ R satisfy 1 < p < λ < n and let u : Rn → R
be a Lipschitz function with u ≡ 0 in Rn \ B1 (0). Then, for every q ≤ p1 , inequality
(1.1) holds for a constant C depending only on n, p, and λ.
For the reader’s convenience, in Section 4 we will include his proof in this case p > 1.
The case p = 1, which involves weak spaces, is also treated in [1].
In fact, Adams [1, Proposition 3.1 and Theorems 3.1 and 3.2] proved the following
stronger embedding:
kukM p1 ,λ (B1 (0)) ≤ Ck∇ukM p,λ (B1 (0)) . (1.2)
While inequality (1.2) is dimensionless by scaling, note that the dimensionless exponent
for inequality (1.1) is q = p2 . This suggests that (1.1) could hold with q = p2 , or at
least for q < p2 . In fact, it is easy to prove that among radially symmetric functions,
(1.1) holds for every q < p2 (see [4, Proposition 1.2(i)] and also Theorem 1.5 below).
In 2011, the article [2, Theorem 2.5] claimed that (1.1) held for every q < p2 and for
general functions, not necessarily radial.
Some years later, we realized that the proof they gave in [2] was not correct. See
the corrigenda by the same authors in [3] and [4]. At the same time, we could not find
other works addressing the question of what is the optimal exponent.
In the present paper we show that actually q = p1 is the largest possible exponent
in (1.1). To show this, for every q > p1 we construct a non-radial function, described
in detail below, for which kukLq (B1 (0)) = ∞ while k∇ukM p,λ (B1 (0)) < ∞. An important
feature of the function is that it depends only (up to a cutoff function) on k variables
(x1 , . . . , xk ), where k is the smallest integer such that λ ≤ k. The function is basically
a negative power of the distance to a set of Haussdorf dimension n − λ. When λ
is not an integer this set is a fractal and, therefore, the structure of the function is
not so “simple”. In fact, it will be rather delicate to control the Morrey norm of
its gradient. We could not find a simpler counterexample for λ ∈ / Z, although we
had several candidates that finally did not work. The possibility of finding simpler
examples remains as an open question.
In addition, we also consider a related norm, which we call the “triple norm”, given by
Z
p
|||∇u|||p,λ;Ω := sup |∇u(x)|p |x − y|λ−n dx, (1.3)
y∈Ω Ω
 OPTIMAL EMBEDDINGS FOR LEBESGUE AND MORREY NORMS 3

where Ω ⊂ Rn is a domain. The article [5] on stable solutions to semilinear equations

gives rise naturally to such a norm (see Remark 1.2 below). Note that, clearly, we have
k∇ukM p,λ (Ω) ≤ |||∇u|||p,λ;Ω (1.4)
for every function u. The function that we construct will also satisfy |||∇u|||p,λ;B1(0) < ∞,
and thus q = p1 is the largest possible exponent also for the embedding
kukLq (B1 (0)) ≤ C |||∇u|||p,λ;B1(0) . (1.5)
Instead, among radial functions we show that inequality (1.5) holds for every q ≤ p2 ,
in contrast to inequality (1.1) for radial functions, which holds only for q < p2 .
Remark 1.2. The regularity results from the recent paper [5] on stable solutions to
semilinear equations −∆u = f (u) in a domain Ω ⊂ Rn are based on bounds for a
Morrey norm with p = 2, of ∇u or, given a point y, of the radial derivative ∇u(x) ·
(x − y)/|x − y|. For this, see [5, Lemma 2.1, step 2 in the proof of Theorem 1.2, and
proof of Theorem 7.1], which also lead to the triple norm (1.3). The boundedness
results from [5] up to dimension n ≤ 9 correspond to p = 2 and λ = 2, while in the Lq
results for n ≥ 11 one has p = 2 < λ < n as in our paper. The results of the current
article are used in [5] to determine optimally a range of exponents q for which stable
solutions necessarily belong to Lq in dimensions n ≥ 11.
Summarizing, our main contribution is the following result. It provides a counterex-
ample to the validity of (1.1) and (1.5) for q > p1 , given by a function u which is
basically a negative power of the distance to a set of Haussdorf dimension n − λ. When
λ∈/ Z, this set is a fractal.
Theorem 1.3. Let p, λ ∈ R satisfy 1 < p < λ < n. Then, for every q > p1 :=
λp/(λ − p) there exists a Lipschitz function u : Rn → R with u ≡ 0 in Rn \ B1 (0),
kukLq (B1 (0)) = ∞, and |||∇u|||p,λ;B1 (0) < ∞. (1.6)
In particular, we also have
k∇ukM p,λ (B1 (0)) < ∞.
If λ is an integer, such function u can be taken to be
u(x) = |x′ |−α − 2α + ξ(|x′′ |)


(1.7)

where x = (x′ , x′′ ) ∈ Rλ × Rn−λ , the parameter α satisfies

λ λ−p
≤α< , (1.8)
q p
√
and ξ : R+ → [0, 1] is a cutoff function with ξ ≡ 1 in [0, 1/2) and ξ ≡ 0 in R+ \[0, 3/3).
If k − 1 < λ < k for some integer k ∈ [2, n], the function u can be taken to be
dist(x, Cn,λ )−α − 4α +
(

if k = n
u(x) = (1.9)
dist(x′ , Ck,λ)−α − 4α + ξ(|x′′ |)


if k < n,
4 X. CABRÉ AND F. CHARRO

where x = (x′ , x′′ ) ∈ Rk × Rn−k , α satisfies (1.8), ξ is a cutoff function as above, and
Ck,λ is a set of Hausdorff dimension k − λ in Rk given by
Ck,λ = {0} × Cγ ⊂ Rk−1 × [−1/2, 1/2],
1
where Cγ ⊂ [−1/2, 1/2] is a generalized Cantor set with parameter γ = 1 − 21− k−λ .
We emphasize that the counterexample to the embedding is therefore given by a
function that, up to a cutoff, only depends on k variables, where k is the smallest
integer such that λ ≤ k.
Remark 1.4. The generalized Cantor set Cγ (see [7]) is obtained from the interval
[−1/2, 1/2] by removing at iteration j = 1, 2, . . . the central interval of length γ lj−1
from each remaining segment of length lj−1 = ((1 − γ)/2)j−1; see Figure 2 in Section 6.
The usual Cantor set corresponds to γ = 1/3. The reason for our choice of γ is that
the Hausdorff dimension of Cγ is
− log 2
= k − λ ∈ (0, 1)
log 1−γ
2
(see [7, Theorem 9.3]). In particular, letting λ range from k − 1 to k yields any fractal
dimension between 0 and 1, and (1.9) somehow interpolates between the integer cases
λ = k − 1 and λ = k.
Let us describe briefly how we found that p1 is the optimal exponent. In the case
when λ is an integer, the hint came from the number p1 = λp/(λ − p), which can be
thought of as the Sobolev exponent in dimension λ. It was natural then to choose the
function (1.7), since it gives a counterexample for the Sobolev inequality in Rλ when
q > λp/(λ − p) = p1 and the exponent α is chosen appropriately.
When λ ∈ Z, (1.7) is basically a negative power of the distance to a subspace of
dimension n − λ. Therefore, when λ ∈ / Z, a negative power of the distance to a set
of Hausdorff dimension n − λ became a natural candidate to counterexample. This is
what the function in (1.9) basically is, a power of the distance to Ck,λ × Rn−k .
It may be of interest to recall here the solutions found by R. Schoen and S.-T. Yau
in [8, Section 5] for nonlinear equations with critical exponent. They construct weak
solutions which are singular on a Cantor set with fractional Hausdorff dimension that
can be made arbitrarily small; see [8, Page 70]. Obviously, such equations are closely
related to the Sobolev embedding.
Among radial functions, the optimal ranges of exponents in inequalities (1.1) and
(1.5) are strictly larger than those of Theorems 1.1 and 1.3. This is the content of the
following result, where we show that the exponent q can go up to q2 . Interestingly, the
answer is different for the Morrey and the “triple” norms: We prove that (1.1) is false
for q = p2 while (1.5) holds for this exponent. Here we can include the exponent p = 1.
Theorem 1.5. Let p, λ ∈ R satisfy 1 ≤ p < λ < n, and let p2 := np/(λ − p).
(a) For every 1 ≤ q < p2 and all radially symmetric C 1 functions u vanishing on
∂B1 (0), we have
kukLq (B1 (0)) ≤ C k∇ukM p,λ (B1 (0)) , (1.10)
 OPTIMAL EMBEDDINGS FOR LEBESGUE AND MORREY NORMS 5

where C is a constant depending only on n, p, λ, and q. In addition, this em-

bedding is false for q ≥ p2 .
(b) For all radially symmetric C 1 functions u with compact support in Rn , we have
kukLp2 (Rn ) ≤ C |||∇u|||p,λ;Rn , (1.11)
where C is a constant depending only on n, p, and λ. In addition, p2 is the
optimal exponent in this inequality.
The paper is organized as follows. In Section 2, we prove a monotonicity result,
Lemma 2.1, that we will use several times throughout the paper to optimize the location
of the “singularities” y in the Morrey and “triple” norms, both in the radial and non-
radial cases. In Section 3 we prove the embeddings for radial functions, Theorem 1.5.
In Section 4 we provide for the reader’s convenience D. Adams’ [1] proof of Theorem
1.1 in the general case of non-radial functions. In Sections 5 and 6 we prove Theorem
1.3 on the optimality of the embeddings. We consider separately the case when λ is an
integer in Section 5 (for its simplicity) and the case when λ is a non-integer in Section
6 (which is much more involved).
(m)
Notation. In the sequel BR (x) denotes the open ball in Rm of radius R centered
at x. For simplicity, whenever m or x are omitted, we will consider m = n and x = 0
respectively. By C, we denote constants that may change from line to line. For points
in Rn , we will write x = (x′ , x′′ ) ∈ Rk × Rn−k for k a positive integer specified from the
context. Given a function u, u+ = max{u, 0} is its positive part. As mentioned before,
we will denote p1 = λp/(λ − p) and p2 = np/(λ − p). Finally, dist (t, U) = inf z∈U |t − z|
as usual.

2. On the location of the singularity in the “triple” norm

In this section we prove a monotonicity result that we will use several times in the
sequel to study which locations of the “singularity” y make larger the integral in the
“triple norm”
Z
p
|||∇u|||p,λ;Ω = sup |∇u(x)|p |x − y|λ−n dx.
y∈Ω Ω

Lemma 2.1. Consider a domain Ω ⊂ Rn , convex in the e1 direction, and symmetric

with respect to {z1 = 0}. Let J : Rn → R be given by
Z
J(y) := h(z)|z − y|−θ dz
Ω

with θ > 0 and h a non-negative function in Ω. Then:

(a) J is non-increasing with respect to y1 in {y1 ≥ supz∈Ω z1 }, and non-decreasing
with respect to y1 in {y1 ≤ inf z∈Ω z1 }.
(b1) Suppose that the non-negative function h satisfies that, for some η ∈ [0, supz∈Ω z1 )
and every y1 ∈ [η, supz∈Ω z1 ),
h(z ∗ ) ≥ h(z) for all z ∈ Ω ∩ {z1 ≥ y1 },
6 X. CABRÉ AND F. CHARRO

zs∗ zs
❇❇ ✂✂
❇ ✂
s ❇✂s ✲
0 y1 z1

Figure 1. Monotonicity argument in the proof of Lemma 2.1.

where z ∗ = (2y1 − z1 , z ′ ) ∈ R × Rn−1 is the reflection of z with respect to

the hyperplane {z1 = y1 }. Then, J is non-increasing with respect to y1 in
{η ≤ y1 < supz∈Ω z1 }.
(b2) On the other hand, if the non-negative function h is such that, for some η ∈
(inf z∈Ω z1 , 0] and every y1 ∈ (inf z∈Ω z1 , η],
h(z ∗ ) ≥ h(z) for all z ∈ Ω ∩ {z1 ≤ y1 }
with z ∗ as before, then J is non-decreasing with respect to y1 in {inf z∈Ω z1 <
y1 ≤ η}.
Proof. Observe first that for every z ∈ Ω, the quantity |z − y| is increasing with respect
to y1 in {y1 ≥ supz∈Ω z1 }, and decreasing with respect to y1 in {y1 ≤ inf z∈Ω z1 }.
Since h ≥ 0 and θ > 0, we deduce that J is non-increasing with respect to y1 in
{y1 ≥ supz∈Ω z1 }, and non-decreasing with respect to y1 in {y1 ≤ inf z∈Ω z1 }. This
proves part (a).
Assume now that 0 ≤ η ≤ y1 < supz∈Ω z1 and compute
Z
∂y1 J(y) = θ h(z)(z1 − y1 )|z − y|−θ−2 dz
(ΩZ
=θ h(z)(z1 − y1 )|z − y|−θ−2 dz
Ω∩{z1 ≥y1 }
Z )
+ h(z)(z1 − y1 )|z − y|−θ−2 dz .
Ω∩{z1 ≤y1 }
 OPTIMAL EMBEDDINGS FOR LEBESGUE AND MORREY NORMS 7

For every z ∈ Ω ∩ {z1 ≥ y1 }, let z ∗ = (2y1 − z1 , z ′ ) be its reflection with respect to the
hyperplane {z1 = y1 }; see Figure 1. Then, |z ∗ − y| = |z − y|, while h(z ∗ ) ≥ h(z) by
hypothesis. Therefore, for every z ∈ Ω ∩ {z1 ≥ y1 }, we have

h(z)(z1 − y1 )|z − y|−θ−2 ≤ −h(z ∗ )(z1∗ − y1 )|z ∗ − y|−θ−2

and hence, using that h ≥ 0,

Z
h(z)(z1 − y1 )|z − y|−θ−2 dz
Ω∩{z1 ≥y1 }
Z
≤− h(z)(z1 − y1 )|z − y|−θ−2 dz
∗
(Ω∩{z1 ≥y1 })
Z
≤− h(z)(z1 − y1 )|z − y|−θ−2 dz.
Ω∩{z1 ≤y1 }

Therefore, ∂y1 J(y) ≤ 0 for all η ≤ y1 < supz∈Ω z1 and the conclusion in part (b1)
follows. The statement for inf z∈Ω z1 < y1 ≤ η in part (b2) follows from (b1) by
reflection. 

3. The radial case: Proof of Theorem 1.5

In this section we establish Theorem 1.5 on the embeddings for radial functions. In
the proof we apply Lemma 2.1 (the monotonicity result proved in Section 2), which
will also be used for the non-radial case in Section 6. We point out that the Sobolev
inequalities with monomial weights established in [6] by Ros-Oton and the first author
will be of great use.

Proof of Theorem 1.5. We structure the proof in four parts. In Part 1 we establish
estimate (1.10), while in Part 2 we show that q < p2 is the optimal range of exponents
for this estimate. In Part 3a we prove (1.11); here we will use the results of [6]. Part
3b provides an alternative proof of (1.11). Finally, we show in Part 4 that p2 is the
largest exponent for which (1.11) holds.

Part 1. We proceed now to show estimate (1.10) for 1 ≤ q < p2 . All the constants C
will depend only on n, p, λ, and q. On the one hand we have
Z 1
q
kukLq (B1 (0)) = C |u(r)|q r n−1 dr
0
Z 1 Z 1 q ∞ Z
X 21−j Z 1 q
n−1
≤C |ur (s)| ds r dr = C |ur (s)| ds r n−1 dr
0 r j=1 2−j r
∞ Z 21−j 1 q ∞ 1 q
2n − 1 X −jn
X Z Z
n−1
≤C |ur (s)| ds r dr = C 2 |ur (s)| ds .
j=1 2−j 2−j n j=1 2−j
8 X. CABRÉ AND F. CHARRO

Now, Hölder’s inequality yields

j Z 21−i j
! p1
Z 1 X X p−1
Z 21−i
|ur (s)| ds = |ur (s)| ds ≤ 2−i p |ur (s)|p ds
2−j i=1 2−i i=1 2−i

j
! p1
Z 21−i
−i p−1 +i n−1
X
≤ 2 p p |ur (s)|p sn−1 ds
i=1 2−i

j Z ! p1
−i p−n
X
≤C 2 p |∇u(x)|p dx
i=1 B21−i (0)

j Z ! p1
−i p−n +(1−i) n−λ
X
=C 2 p p 2(1−i)(λ−n) |∇u(x)|p dx
i=1 B21−i (0)∩B1 (0)
j
X λ−p λ−p
≤C 2i p k∇ukM p,λ (B1 (0)) ≤ C 2j p k∇ukM p,λ (B1 (0)) .
i=1

Therefore, we obtain
∞
X q(λ−p)−np
kukqLq (B1 (0)) ≤C 2j p k∇ukqM p,λ (B1 (0)) ,
j=1

and the series is convergent since q < p2 .

Part 2. In order to show that q < p2 is the optimal range of exponents in estimate
(1.10), let q ≥ p2 and consider the function uα (x) = |x|−α − 1, extended by zero
outside B1 (0), with α = (λ − p)/p. Notice that uα vanishes on ∂B1 (0) and that
kuα kLq (B1 (0)) = ∞, since n ≤ αq.
To show that k∇uα kM p,λ (B1 (0)) is finite, let y ∈ B 1 (0) and r > 0. Observe that we
can write −y instead of y in the definition of the Morrey norm. Then,
Z Z
p
|∇uα(x)| dx = C |x|−αp−p dx
B1 (0)∩Br (−y) B (0)∩Br (−y)
Z 1 Z
−αp−p
≤C |x| dx = C |z − y|−αp−p dz
Br (−y) Br (0)

by the change of variables z = x + y. Notice that, upon a rotation, we can assume

y = y1 e1 = (y1 , 0, . . . , 0) with y1 ≥ 0. Denote
Z
J(y1 ) = |z − y1 e1 |−αp−p dz.
Br (0)

We can now apply Lemma 2.1 (with η = 0 and h ≡ 1) and conclude that J is non-
increasing in [0, ∞). Therefore, J(y1 ) ≤ J(0) for all y1 ≥ 0.
As a consequence, we have that
Z Z r
λ−n p λ−n
r |∇uα (x)| dx ≤ Cr sn−αp−p−1 ds ≤ C
B1 (0)∩Br (−y) 0
 OPTIMAL EMBEDDINGS FOR LEBESGUE AND MORREY NORMS 9

independently of r, by our choice of α.

Part 3a. We give here a first proof of estimate (1.11). Recall that u has compact
support. Since u is radial, it suffices to show that
Z ∞  λ−p
np
Z ∞  1p
np
n−1 ′ p λ−1
|u(r)| λ−p r dr ≤C |u (r)| r dr . (3.1)
0 0

In fact, we are going to prove (3.1) with the best constant. For this, we perform the
change of variables r = sa with a = p/(n − λ + p) and v(s) = u(sa ) in the integrals on
both sides of (3.1). We get
Z ∞ Z ∞
np np
n−1
|u(r)| λ−p r dr = a |v(s)| λ−p san−1 ds
0 0

and Z ∞ Z ∞
′ p λ−1 1−p
|u (r)| r dr = a |v ′(s)|p san−1 ds.
0 0
Observe that after the change of variables, both integrals are weighted by the same
power an − 1 ≥ 0. Now, [6, Theorem 1.3], applied with D = A1 + 1 = an, gives that
Z ∞  λ−p
np
Z ∞  p1
np
|v(s)| λ−p san−1 ds ≤ Cp |v ′ (s)|p san−1 ds ,
0 0

with an explicit value of its best constant Cp . Inequality (3.1), and thus (1.11), is now
established.
Moreover, it is also shown in [6] that when 1 < p < λ the R constant Cp is attained in
1,p an−1
W0 (R, |s| ds), the closure of Cc (R) under the norm ( R (|u| + |∇u|p)|s|an−1 ds)1/p ,
1 p

by the functions
p−λ
p
  n+p−λ
v(s) = c1 + c2 |s| p−1 ,
where c1 , c2 > 0 are arbitrary constants. On the other hand, when p = 1, the constant
C1 is not attained by any function in W01,1 (R, |s|an−1 ds). Note however, that knowing
the best constant for (3.1) does not ensure that we know the best constant for (1.11).
Indeed, for
p−λ
n+p−λ
  n+p−λ
uc1 ,c2 (x) = c1 + c2 |x| p−1 , c1 , c2 > 0,
we have seen that
Z
kuc1,c2 k Lp2 (Rn ) =C |∇uc1 ,c2 (x)|p |x|λ−n dx
Rn

with the best constant C, but on the other hand we do not know if the supremum in
the “triple norm” is attained at y = 0. That is, we do not know if
Z
|∇uc1 ,c2 (x)|p |x|λ−n dx = |||∇uc1 ,c2 |||p,λ;Rn . (3.2)
Rn
10 X. CABRÉ AND F. CHARRO

The point here is that |∇uc1,c2 |p is zero at the origin, instead of blowing up as in the
other cases that we consider in the paper. Therefore,
Z
|∇uc1,c2 (x)|p |x − y|λ−n dx
Rn

could perhaps be increasing in |y| in an interval near the origin, and then decrease to
0 as |y| → ∞. In this setting, the reflection argument in the proof of Lemma 2.1 does
not work and we cannot conclude (3.2) as before.
Part 3b. We will provide here an alternative proof of estimate (1.11). First, we
establish the case p = 1 by proving
Z ∞  λ−1
n
Z ∞
n
n−1
|u(r)| λ−1 r dr ≤C |u′ (r)| r λ−1 dr (3.3)
0 0

for every λ ∈ (1, n). One can then deduce the general case applying (3.3) with (n(p −
1) + λ)/p instead of λ to the function ub with b = 1 + n(p − 1)/(λ − p). Note that
1 < (n(p − 1) + λ)/p < n.
To show (3.3), notice that we can assume u ≥ 0. Furthermore, we can also assume
that u is radially decreasing. This follows from [6, Proposition 4.2] (see also [9]) applied
to inequality (3.3) after changing variables as in Part 3 in order to guarantee that both
sides of the inequality are weighted by the same power. Now, on the one hand, the
change of variables t = u(r) yields
Z ∞ Z ∞ Z max(u)
λ−1
′ λ−1 ′ λ−1
|u (r)| r dr = − u (r) r dr = ϕ(t) n dt (3.4)
0 0 0
n
for ϕ(t) = {r : u(r) > t} . On the other hand, by Cavalieri’s principle
Z ∞
1 ∞ n
Z
n n
u(r) λ−1 r n−1
dr = t λ−1 −1 ϕ(t) dt. (3.5)
0 n 0 λ−1
We conclude by proving that
Z ∞ Z ∞ q
1
q−1
q t ϕ(t) dt ≤ ϕ(t) dt
q (3.6)
0 0

for every non-increasing function ϕ = ϕ(t) and every q > 1. Consequently, (3.3) will
follow from (3.4), (3.5), and (3.6). To prove inequality (3.6), denote
Z s Z s q
1
q−1
F1 (s) = q t ϕ(t) dt and F2 (s) = ϕ(t) q dt ,
0 0

and notice that F1 (0) = F2 (0), while

Z s q−1
1 1
′
F2 (s) = qϕ(s) q ϕ(t) dt
q

0
1 q−1
 1

≥ qϕ(s) sϕ(s) q
q = q sq−1 ϕ(s) = F1′ (s).

This establishes (3.6) and hence concludes the proof of (1.11).

 OPTIMAL EMBEDDINGS FOR LEBESGUE AND MORREY NORMS 11

Part 4. Finally, we show that p2 is the largest exponent for which (1.11) holds. Let
q > p2 and consider the function uα (x) = (|x|−α − 1)+ with n/q ≤ α < (λ − p)/p.
Notice that
Z Z 1
q −α

q
kuα kLq (Rn ) = |x| − 1 dx = C (1 − r α )q r n−αq−1 dr = ∞,
B1 (0) 0

by our choice of α.
On the other hand, we are going to see that |||∇uα |||p,λ;Rn < ∞, thus contradicting the
inequality. To prove this, we claim that y = 0 realizes the supremum in the definition
of |||∇uα |||p,λ;Rn , and hence
Z Z 1
p p λ−n
|||∇uα |||p,λ;Rn = |∇uα (x)| |x| dx = C r λ−αp−p−1 dr < ∞,
B1 (0) 0

since λ − αp − p > 0.
We conclude proving the claim by monotonicity. For y ∈ Rn , we have that
Z Z
p λ−n p
|∇uα(x)| |x − y| dx = α |x|−αp−p |x − y|λ−n dx.
B1 (0) B1 (0)

Notice that, upon a rotation, we can assume y = y1 e1 = (y1 , 0, . . . , 0). Denote

Z
J(y1 ) = |x|−αp−p |x − y1 e1 |λ−n dx.
B1 (0)

Since the function J is under the hypotheses of Lemma 2.1 (with η = 0), we conclude
that J is non-increasing in [0, ∞), and therefore that J(y1 ) ≤ J(0) for all y1 ≥ 0. 

4. Embeddings in the general case: Proof of Theorem 1.1

For the reader’s convenience, we provide in this section D. Adams’ proof of Theo-
rem 1.1, see [1, Theorem 3.1]. Recall that we consider a Lipschitz function u : Rn → R
with u ≡ 0 in Rn \ B1 (0), so that integrals in Rn and integrals in B1 (0) coincide.
Proof of Theorem 1.1. The proof is based on the following two claims,
|u(x)| ≤ C(I1 |∇u|)(x), a.e. x. (4.1)
and

λp
1− λp
I1 |∇u| (x) ≤ C Mλ/p |∇u|(x) M0 |∇u|(x) , (4.2)
where Z
I1 f (x) := f (y) |y − x|1−n dy
Rn
is the Riesz potential of f , and
 Z 
β−n
Mβ f (x) := sup r |f (z)| dz , 0≤β≤n
r>0 Br (x)

is the maximal function with parameter β.

12 X. CABRÉ AND F. CHARRO

Once (4.1) and (4.2) are established, we can finish the proof as follows. By Hölder’s
inequality, we have
 Z 
λ
−n
Mλ/p |∇u|(x) = sup r p |∇u(z)| dz
r>0 Br (x)
 Z  p1
λ−n p
≤ C sup r |∇u(z)| dz ≤ Ck∇ukM p,λ(B1 (0)) .
r>0 B1 (0)∩Br (x)

(recall that u ≡ 0 in Rn \ B1 (0)). Then, (4.1) and (4.2) give

p
1− p p
1− λp
|u(x)| ≤ C Mλ/p |∇u|(x) λ M0 |∇u|(x) λ ≤ Ck∇ukM λ
p,λ (B (0))
1
M0 |∇u|(x)

almost everywhere. It follows that

pq
Z

ppq
kukqLq (B1 (0)) q
≤ C k∇ukM p,λ (B1 (0))
λ
M0 |∇u|(x) 1 dx
B1 (0)
pq pq
1− pq
≤ C q k∇ukMλ p,λ (B1 (0)) |B1 (0)|
p
1 kM0 |∇u|kL1p (B1 (0)) ,

where we have applied Hölder’s inequality with exponents p1 /(p1 − q) and p1 /q.
By the well-known Lp estimate for the maximal function M0 when p > 1, there exists
a constant C depending only on n and p such that
kM0 |∇u|kLp (B1 (0)) ≤ Ck∇ukLp (B1 (0)) ,
and therefore
p
1
− p1 + p
λ p1
kuk Lq (B1 (0)) ≤ C |B1 (0)| q 1 k∇uk M p,λ (B1 (0))
≤ C k∇ukM p,λ(B1 (0))

with C depending only on n, p, and λ as desired.

Therefore, it remains to prove claims (4.1) and (4.2).
Consider first estimate (4.1). To prove it, notice that for σ ∈ Rn with |σ| = 1
Z ∞ Z ∞
d
|u(x)| = − u(x + rσ) dr ≤ |∇u(x + rσ)| dr.
0 dr 0

Then, integrating on σ we get

Z ∞Z
|∇u(x + rσ)|r 1−n r n−1 dσdr = C (I1 |∇u|)(x)


|u(x)| ≤ C
0 ∂B1 (0)

and (4.1) is proved.

Next, consider estimate (4.2). We reproduce the argument in [1] to show that for a
given function f with compact support in Rn , we have

p
1− p
I1 f (x) ≤ C Mλ/p f (x) λ M0 f (x) λ , (4.3)
where C depends only on n, p, and λ.
 OPTIMAL EMBEDDINGS FOR LEBESGUE AND MORREY NORMS 13

For f 6≡ 0, let δ > 0 to be determined later and set

Z
I1 f (x) = f (y) |y − x|1−n dy
n
ZR Z
1−n
= f (y) |y − x| dy + f (y) |y − x|1−n dy
{y: |x−y|<δ} {y: |x−y|≥δ}
′
=I +I .
Let
ak (x) = {y : 2k δ ≤ |x − y| < 2k+1 δ} for k ∈ Z.
Then,
∞ Z
X
|I| ≤ |f (y)| |x − y|1−n dy
k=1 a−k (x)
∞
X
1−n
n
≤ 2−k δ 2−k+1 δ M0 f (x) = 2n δM0 f (x).
k=1

Similarly,
∞ Z
X
′
|I | ≤ |f (y)| |x − y|1−n dy
k=0 ak (x)
∞
X
1−n
n− λp λ
≤ 2k δ 2k+1 δ Mλ/p f (x) = Cδ 1− p Mλ/p f (x),
k=0

since p < λ. The choice

 λp
Mλ/p f (x)

δ = δ(x) =
M0 f (x)
finally gives (4.3). 

5. Proof of Theorem 1.3 in the case when λ is an integer

In this section we prove Theorem 1.3 when λ is an integer. The argument is very
simple. The case λ ∈ / Z is the core of our paper and will be considered in Section 6.
As mentioned in the introduction, the choice of the counterexample when λ ∈ Z was
hinted by the number p1 = λp/(λ−p), which can be thought of as the Sobolev exponent
in dimension λ. Then, it is natural to choose a function that provides a counterexample
for the Sobolev inequality in dimension λ and then look at this function embedded in
the n-dimensional space. Namely, we take
u(x) = |x′ |−α − 2α + ξ(|x′′ |),


(5.1)
where x = (x′ , x′′ ) ∈ Rλ × Rn−λ
√ and ξ : R+ → [0, 1] is a cutoff function with ξ ≡ 1 in
[0, 1/2) and ξ ≡ 0 in R+ \ [0, 3/3). Note that clearly u has support in B1 (0) ⊂ Rn .
The rest of the section is devoted to show the following result, which proves Theo-
rem 1.3 when λ is an integer.
14 X. CABRÉ AND F. CHARRO

Proposition 5.1. Let λ be an integer such that 1 < p < λ < n and assume that
q > p1 := λp/(λ − p). Then, for u given by (5.1), we have that kukLq (B1 ) = ∞ and
|||∇u|||p,λ;B1 < ∞ if
λ λ−p
≤α< .
q p
This proves the optimality of the range q ≤ p1 for (1.5) when λ ∈ Z, and in turn also
for (1.1).
(n)
Proof. For every y ∈ B 1 we have that
Z
|∇u(x)|p |x − y|λ−n dx
(n)
B1
Z Z
≤C |x′ |−αp−p |x − y|λ−n dx′′ dx′
(λ) (n−λ)
B1 B1
2 ! λ−n
2
|x′′ − y ′′ |
Z Z 
=C |x′ |−αp−p 1+ |x′ − y ′|λ−n dx′′ dx′ ,
(λ)
B1
(n−λ)
B1 |x′ − y ′ |

x′′ −y ′′
for some constant C independent of y. The change of variables z = |x′ −y ′ |
yields

 ′′ ′′
2 ! λ−n
2
|x − y |
Z
1+ |x′ − y ′ |λ−n dx′′
(n−λ)
B1 |x′ − y ′ |
Z Z 2

λ−n |x′ −y ′ |
λ−n
2
≤ 1 + |z| 2
dz = C 1 + r2 2
r n−λ−1 dr
(n−λ)
B2/|x′ −y ′ | 0
Z 2
|x′ −y ′ |
≤C max{1, r}λ−n r n−λ−1 dr ≤ C (1 + |log |x′ − y ′||) .
0

Therefore, we have
Z Z
p λ−n
|∇u(x)| |x − y| dx ≤ C |x′ |−αp−p (1 + |log |x′ − y ′ ||) dx′ .
(n) (λ)
B1 B1

(λ)
We claim that the last integral is bounded uniformly in y ′ ∈ B 1 . To verify this,
(λ)
since |log |x′ − y ′|| ≤ log 2 for x′ ∈ B1 \ B1 (y ′) and λ − αp − p > 0, it suffices to
control the integral over B1 (y ′ ). But then, calling z := y ′ − x′ , the integral becomes
R ′ −αp−p (λ)
(λ) h(z)|z − y | dz with h(z) = 1 + |log |z|| = 1 − log |z| for z ∈ B1 . Now, since
B 1
(λ)
h is non-negative and radially decreasing in B1 , we can apply Lemma 2.1 with η = 0
and conclude that the largest value of the integral corresponds to y ′ = 0. But since we
have assumed λ − αp − p > 0, the integral with y ′ = 0 is finite.
 OPTIMAL EMBEDDINGS FOR LEBESGUE AND MORREY NORMS 15

On the other hand,

Z Z
kukqLq (B1 ) ≥ uq dx′′ dx′
(λ) (n−λ)
B1/2 B1/2
Z Z 1
2
α q
′ −α ′

q
r −α − 2α r λ−1 dr


=C |x | −2 dx = C
(λ)
B1/2 0

and the last integral is divergent since λ ≤ αq by hypothesis. 

6. Proof of Theorem 1.3 in the general case

In this section we conclude the proof of Theorem 1.3 by considering the case when
λ is not an integer.
Let us motivate first the case n − 1 < λ < n. As we have seen in Section 5, when λ
is either n or n − 1 expression (5.1) provides a counterexample to the embedding (1.5)
for q > p1 , and therefore for (1.1) in view of (1.4) –here, when λ = n the Morrey and
triple norms coincide with the Sobolev norm. In both cases the function that yields
the counterexample is basically a negative power of the distance function, either to the
origin in the case λ = n, or to a line when λ = n − 1. Therefore, when λ is strictly
between n − 1 and n, a negative power of the distance to a fractal set of non-integer
dimension n − λ is a natural candidate to be a counterexample to inequality (1.5).
Let us describe precisely the functions that provide the counterexample. When
n − 1 < λ < n, we consider
uα,n (x) = dist(x, Cn,λ )−α − 4α +


(6.1)
for Cn,λ = {0} × Cγ ⊂ Rn−1 × [−1/2, 1/2], where Cγ is a generalized Cantor set with
parameter
1
γ = 1 − 21− n−λ ∈ (0, 1). (6.2)
The generalized Cantor set Cγ (see [7]) is obtained from the interval [−1/2, 1/2]
by removing at iteration j = 1, 2, . . . the central interval of length γ lj−1 from each

j−1
remaining segment of length lj−1 = (1 − γ)/2 ; see Figure 2. A precise expression
for Cγ is given later in (6.8), (6.9), and (6.10). The usual Cantor set corresponds to
γ = 1/3.
The reason for our choice of γ in (6.2) is that the Hausdorff dimension of Cγ is
− log 2
=n−λ
log 1−γ
2

(see [7, Theorem 9.3]). Thus, letting λ vary between n − 1 and n yields any fractal
dimension between 0 and 1. In particular, (6.1) somehow interpolates the integer cases
λ = n − 1 and λ = n.
(n)
Note that uα,n has support in B3/4 (0). Indeed, if y ∈ Cn,λ then |y| ≤ 1/2, and thus
|x − y| ≥ 1/4 if |x| ≥ 3/4; in particular dist(x, Cn,λ ) ≥ 1/4 and uα,n (x) = 0.
In the case when k − 1 < λ < k for some integer k ∈ {2, . . . , n − 1} we embed into
R the counterexample in Rk by means of an appropriate cutoff function (as we did
n
16 X. CABRÉ AND F. CHARRO

− 21 − γ2 γ
2
1
2

Figure 2. Construction of the generalized Cantor set Cγ

in the previous section when λ ∈ Z). In this way, we reduce the proof to the case
n − 1 < λ < n. More precisely, we consider
uα (x) = uα,k (x′ ) ξ(|x′′|) if k < n, (6.3)
′ ′′ k n−k
where x = (x , x ) ∈ R × R , uα,k is given by (6.1) with n replaced by k, i.e.,
uα,k (x′ ) = dist(x′ , Ck,λ )−α − 4α + ,


(6.4)
√
and ξ : R+ → [0, 1] is a cutoff function with ξ ≡ 1 in [0, 1/2) and √ ξ ≡ 0 in R+ \[0, 3/3).
Note that if uα (x) 6= 0 then necessarily |x′ | ≤ 3/4 and |x′′ | ≤ 3/3; thus |x| < 1.
The rest of the section is devoted to proving the following result, which together
with Proposition 5.1 completes the proof of Theorem 1.3.
Theorem 6.1. Let p, λ, q ∈ R be such that 1 < p < λ < n, k − 1 < λ < k for some
integer k ∈ {2, . . . , n}, and q ≥ 1. Consider
dist(x, Cn,λ )−α − 4α +
(

if k = n
uα (x) =
dist(x′ , Ck,λ)−α − 4α + ξ(|x′′|)


if k ∈ {2, . . . , n − 1},
with x = (x′ , x′′ ) ∈ Rk ×Rn−k , ξ : R+ → [0, 1] a cutoff function as described after (6.4),
and Ck,λ a set of Hausdorff dimension k − λ given by
Ck,λ = {0} × Cγ ⊂ Rk−1 × [−1/2, 1/2],
1
where Cγ ⊂ [−1/2, 1/2] is a generalized Cantor set with parameter γ = 1 − 21− k−λ .
Then,
kuα kLq (B(n) ) = ∞ and |||∇uα |||p,λ;B(n) < ∞
1 1

if
λ λ−p
≤α< . (6.5)
q p
This proves the optimality of the range q ≤ p1 = λp/(λ − p) for inequality (1.5), and
in turn also for (1.1).
 OPTIMAL EMBEDDINGS FOR LEBESGUE AND MORREY NORMS 17

The proof of Theorem 6.1 is divided into two parts. In the first part we reduce the
computations from dimension n to dimension k with a similar argument to the one in
the proof of Proposition 5.1. More precisely we will show, for uα,k given by (6.4), that

|||∇uα |||p,λ;B(n) ≤ C |||∇uα,k |||p,λ;B(k)

1 1

and
kuαkLq (B(n) ) ≥ C −1 kuα,k kLq (B(k) ) ,
1 1

for some constant C, and hence that it is enough to study the case k = n taking
uα = uα,n . In the second part of the proof we will show that (6.5) leads to

kuα,n kLq (B(n) ) = ∞ and |||∇uα,n |||p,λ;B(n) < ∞,

1 1

as desired; this part is the content of the following two propositions.

Proposition 6.2. Let λ, q ∈ R be such that λ > n − 1 and q ≥ 1. Consider uα,n given
by (6.1) with α such that λ ≤ αq. Then, kuα,n kLq (B(n) ) = ∞.
1

Proposition 6.3. Let λ, p ∈ R be such that n − 1 < λ < n, p > 1 and consider uα,n
given by (6.1) with α > 0 satisfying
λ−p
α< . (6.6)
p
Then,
Z
p
|||∇uα,n ||| (n) := sup |∇uα,n (x)|p |x − y|λ−n dx < ∞.
p,λ;B1 (n) (n)
y∈B 1 B1

Let us now prove Theorem 6.1 assuming Propositions 6.2 and 6.3, which will be
established afterwards.

Proof of Theorem 6.1. Since Propositions 6.2 and 6.3 yield the result when k = n, let
(n)
us assume k < n. Let y ∈ B 1 . By (6.3) and (6.4) we have

|∇uα (x)| ≤ C |∇uα,k (x′ )| + uα,k (x′ ) ≤ C|∇uα,k (x′ )|

(n)
for almost every x ∈ B1 , where we have used that the modulus of the gradient of a
distance function is equal to 1 a.e. Therefore,
Z Z Z
p λ−n
|∇uα (x)| |x − y| dx ≤ C |∇uα,k (x′ )|p |x − y|λ−n dx′′ dx′
(n) (k) (n−k)
B1 B1 B1
2 ! λ−n
2
|x′′ − y ′′ |
Z Z 
=C |∇uα,k (x′ )|p |x′ − y ′ |λ−k 1+ |x′ − y ′|k−n dx′′ dx′ .
(k)
B1
(n−k)
B1 |x′ − y ′|
18 X. CABRÉ AND F. CHARRO

x′′ −y ′′
The change of variables z = |x′ −y ′ |
yields
 ′′ ′′
2 ! λ−n
2
|x − y |
Z Z
′ ′ k−n ′′

λ−n
1+ |x − y | dx ≤ 1 + |z|2 2
dz
B1
(n−k) |x′ − y ′| (n−k)
B2/|x′ −y ′ |
Z 2 Z 2
|x′ −y ′ |
λ−n |x′ −y ′ |
2 n−k−1
=C 1+r 2
r dr ≤ C max{1, r}λ−n r n−k−1 dr
0
2
!0
Z 1 Z
|x′ −y ′ |
=C r n−k−1 dr + r λ−k−1 dr ≤C
0 1

independently of y. Therefore,
Z
|∇uα (x)|p |x − y|λ−n dx
(n)
B1
Z
≤ C sup |∇uα,k (x′ )|p |x′ − y ′|λ−k dx′ = C |||∇uα,k |||p,λ;B(k) ,
(k) (k) 1
y ′ ∈B 1 B1

and Proposition 6.3 applied in Rk , i.e., with n replaced by k, yields |||∇uα |||p,λ;B(n) < ∞.
1
On the other hand,
Z Z
q q
kuαk q (n) = ′
uα,k (x ) ξ q (|x′′ |) dx′′ dx′ = C kuα,k kq q (k) ,
L (B1 ) (k) (n−k) L (B3/4 )
B3/4 B√
3/3

which is infinite by Proposition 6.2 applied with n = k, since (as we pointed out) uα,k
(k)
given by (6.4) has support in B 3/4 . 
We devote the rest of the section to the proofs of Propositions 6.2 and 6.3. In the
sequel we assume
n − 1 < λ < n.
Recall that
uα,n (x) = dist(x, Cn,λ )−α − 4α


+
(6.7)
for Cn,λ = {0} × Cγ ⊂ Rn−1 × [−1/2, 1/2], where Cγ is the generalized Cantor set with
1
parameter γ = 1 − 21− n−λ defined in the beginning of this section. We have
  l−1
∞ 2[
1 1 [
− , \ Cγ := Gl,m , (6.8)
2 2 m=1 l=1

where the union is disjoint and Gl,m are the 2l−1 gap-intervals introduced in generation
l, namely1
 l−1 
1−γ γ γ
Gl,m = − , + hl,m (6.9)
2 2 2
1Wewill not need the following precise expression for the gaps, but only to understand their size
and self-similar structure.
 OPTIMAL EMBEDDINGS FOR LEBESGUE AND MORREY NORMS 19

where
 l−1 l−1  j−1
1 1 1−γ 1+γ X 1−γ
hl,m =− + + cj . (6.10)
2 2 2 2 j=1 2
Here cj ∈ {0, 1} for all j = 1, . . . , l − 1, and the index m ∈ {1, 2, . . . , 2l−1 } runs through
all possible choices of the coefficients c1 c2 . . . cl−1 ; see Figure 2 above.
As a consequence of (6.8), we have that Cγ is a compact set of Lebesgue mea-
sure 0. In addition, the set Cγ is self-similar, that is, Cγ = S1 (Cγ ) ∪ S2 (Cγ ) where
S1 (t) = (1 − γ)t/2 − (1 + γ)/4 and S2 (t) = (1 − γ)t/2 + (1 + γ)/4. Finally, the
Hausdorff dimension of Cγ is − log 2/ log((1 − γ)/2)), see [7, Theorem 9.3]. Notice that
1
we have chosen γ = 1 − 21− n−λ such that the Hausdorff dimension of Cγ is n − λ.
6.1. Proof of Proposition 6.2: Computation of the Lq norm. In this subsection
we provide the proof of Proposition 6.2.
Proof of Proposition 6.2. Recall that Cn,λ = {0} × Cγ ⊂ Rn−1 × [−1/2, 1/2]. We will
denote x = (x′ , x′′ ) ∈ Rn−1 × R. Since Cγ has zero Lebesgue measure, (6.8) leads to
Z 1 Z
2
q
q
kuα,n k q (n) ≥ dist(x, Cn,λ )−α − 4α + dx′ dx′′
L (B1 ) (n−1)
− 12 B1/4

∞ X
2 Z l−1 Z
X
q
= dist(x, Cn,λ )−α − 4α +
dx′ dx′′ ,
(n−1)
l=1 m=1 Gl,m B1/4

where Gl,m are given by (6.9) and (6.10).

l−1 ′ ′′ 1−γ l−1 ′′
An affine change of variables x′ = 1−γ


2
t, x = 2
t + hl,m and the self-
similarity of Cn,λ yield
Z Z

q
dist(x, Cn,λ )−α − 4α + dx′ dx′′
(n−1)
Gl,m B1/4
 (n−αq)(l−1) Z γ  α (l−1) !q
1−γ 1−γ
Z
2
= dist(t, Cn,λ )−α − 4α dt′ dt′′
2 l−1 2
n o
− γ2 |t′ |≤ 41 ( 1−γ
2
) +
 (n−αq)(l−1) Z γ
1−γ
Z
2
q
≥ dist(t, Cn,λ )−α − 4α +
dt′ dt′′
2 − γ2
(n−1)
B1/4

for each m ∈ {1, . . . , 2l−1 }. Therefore, adding these 2l−1 integrals of generation l, and
then summing in l, we have
∞  n−αq !l−1 Z γ Z
X 1 − γ 2
q
kuα,n kq q (n) ≥ 2 dist(t, Cn,λ )−α − 4α + dt′ dt′′ .
L (B1 )
l=1
2 (n−1)
− γ2 B1/4

Since the integral on the right-hand side is positive, it is enough to show that the series
diverges. This happens whenever
 n−αq
1−γ
2 ≥ 1,
2
20 X. CABRÉ AND F. CHARRO

which by our choice of γ is equivalent to

− log 2
n − αq ≤ = n − λ.
log 1−γ
2

This inequality holds by hypothesis. 

6.2. Proof of Proposition 6.3: Bound for the “triple norm”. The proof of
Proposition 6.3 has two parts. The first one (Lemma 6.6 below) shows that in order to
bound the triple norm of ∇uα,n , it suffices to only consider points y ∈ {0}×[−1/2, 1/2],
(n)
instead of the full B1 . More precisely, we prove that

Z 1 Z
2
p
|||∇uα,n ||| (n) ≤C sup dist(x, Cn,λ )−αp−p |x − y|λ−n dx′ dx′′ . (6.11)
p,λ;B1 (n−1)
y ′ =0,|y ′′ |≤ 21 − 21 B1/4

Since Cγ has zero Lebesgue measure, by (6.8) we can write the outer integral on the
right-hand side of (6.11) as an infinite sum of integrals over the disjoint gap-intervals
Gl,m of decreasing size.
The second part of the proof of Proposition 6.3, and crucial point in the argument,
is how to estimate these integrals in terms of the size of the gaps in such a way that the
series converges. Here, there are two cases to be considered according to the position
of the singularity y relative to a given gap: the case when y lies on the closure of a
gap (and hence the function x 7→ |x − y|λ−n is singular), and the case when the gap is
uniformly away from y (and hence |x − y|λ−n can be bounded above and factored out
from the integral). We deal with these two cases in Lemmas 6.4 and 6.5 respectively.

Lemma 6.4. Let G = ((1 − γ)/2)l−1 (−γ/2, γ/2) + h for some l ≥ 1 and h ∈ R of the
form (6.10) (i.e. G is a gap-interval introduced in generation l). Assume y ∈ {0} × G
and λ − αp − p > 0. Then, we have that
Z Z
dist(x, Cn,λ )−αp−p |x − y|λ−n dx′ dx′′
(n−1)
G B1/4
 (λ−αp−p)(l−1)  l−1 ! (6.12)
1−γ 1−γ
≤C +l
2 2

for a constant C depending only on n, p, λ, and α.

Proof. Denote a = (γ/2) ((1 − γ)/2)l−1 so that G = (−a, a) + h. To relate dist(x, Cn,λ )
and |x − y| we consider the midpoints between y ′′ and h + a, and between y ′′ and h − a.
 OPTIMAL EMBEDDINGS FOR LEBESGUE AND MORREY NORMS 21

In this way, we have the bound

Z Z
dist(x, Cn,λ )−αp−p |x − y|λ−n dx′ dx′′
(n−1)
G B1/4
Z h+a Z
≤ dist(x, Cn,λ )λ−n−αp−p dx′ dx′′
y ′′ +h+a (n−1)
2
B1/4
Z y ′′ +h+a Z (6.13)
2
λ−n−αp−p ′ ′′
+ |x − y| dx dx
y ′′ +h−a (n−1)
2
B1/4
Z y ′′ +h−a Z
2
+ dist(x, Cn,λ )λ−n−αp−p dx′ dx′′ = I1 + I2 + I3 .
(n−1)
h−a B1/4

Let us estimate I1 first. For this, notice that whenever x′′ ∈ G, we have

dist(x, Cn,λ )2 = |x′ |2 + min (h + a − x′′ )2 , (x′′ − h + a)2 .


(6.14)

Therefore,
Z h+a Z

λ−n−αp−p
I1 = |x′ |2 + (h + a − x′′ )2 2
dx′ dx′′
y ′′ +h+a (n−1)
2
B1/4

and a change to cylindrical coordinates and the change of variables t = h + a − x′′ yield
Z h+a−y ′′ Z 1
2 4
λ−n−αp−p
I1 = C r 2 + t2 2
r n−2 drdt. (6.15)
0 0

Similarly,
1
Z 0 Z
4
λ−n−αp−p
I3 = C r 2 + t2 2
r n−2 drdt. (6.16)
h−a−y ′′
2
0

On the other hand, since y ′ = 0, we have

Z y ′′ +h+a Z
2
λ−n−αp−p
I2 = |x′ |2 + (x′′ − y ′′)2 2
dx′ dx′′
y ′′ +h−a (n−1)
2
B1/4
Z y ′′ +h+a Z 1
2 4
λ−n−αp−p
=C r 2 + (x′′ − y ′′ )2 2
r n−2 drdx′′ (6.17)
y ′′ +h−a
2
0
Z h+a−y ′′ Z 1
2 4
λ−n−αp−p
=C r 2 + t2 2
r n−2 drdt,
h−a−y ′′
2
0

after applying a change to cylindrical coordinates and the change of variables t =

x′′ − y ′′.
22 X. CABRÉ AND F. CHARRO

Then, (6.13) and (6.15)–(6.17), and the change of variables z1 = 2t − h + y ′′ , z2 = 2r

give
Z Z
dist(x, Cn,λ )−αp−p |x − y|λ−n dx′ dx′′
(n−1)
G B1/4
Z h+a−y ′′ Z 1
2 4
λ−n−αp−p
≤ 2C r 2 + t2 2
r n−2 drdt
h−a−y ′′
2
0
1
Z a Z
2
λ−n−αp−p
1−λ+αp+p
=C2 (z1 + h − y ′′)2 + z22 2
z2n−2 dz2 dz1
−a 0
1−λ+αp+p ′′
=C2 J(y − h, 0),
for
1
Z a Z
2
J(y) = |z − y|λ−n−αp−pz2n−2 dz2 dz1 .
−a 0
Notice that we are under the hypotheses of Lemma 2.1 with Ω = [−a, a] × [0, 1/2],
θ = n − λ + αp + p, h(z) = z2n−2 , and η = 0. Therefore, by Lemma 2.1, we know that J
is non-increasing with respect to y1 in [0, ∞), and non-decreasing with respect to y1 in
(−∞, 0]. In particular, J(y ′′ − h, 0) ≤ J(0), with an equality for y ′′ = h. We conclude
Z Z
dist(x, Cn,λ )−αp−p |x − y|λ−n dx′ dx′′
(n−1)
G B1/4
1
(6.18)
Z a Z
2
λ−n−αp−p
1−λ+αp+p
≤C2 z12 + z22 2
z2n−2 dz2 dz1
−a 0

independently of y.
We estimate now the integral on the right-hand side of (6.18) by applying the change
of variables x1 = z1 , z2 = |x1 |x2
Z aZ 1
2
λ−n−αp−p
z12 + z22 2
z2n−2 dz2 dz1
−a 0
1
Z a Z
2|x1 |
λ−n−αp−p
λ−αp−p−1
= |x1 | 1 + x22 2
x2n−2 dx2 dx1 .
−a 0

Notice that |x1 | < a < 1/2, and therefore we can use 1 + x22 ≥ max{1, x22 } to show that
Z 1 Z 1 Z 1
2|x1 |
λ−n−αp−p 2|x1 |
2
1 + x2 2 n−2
x2 dx2 ≤ n−2
x2 dx2 + x2λ−αp−p−2 dx2 .
0 0 1

Assume first that λ − αp − p 6= 1 and recall that λ − αp − p > 0 by hypothesis. Then,

Z Z
dist(x, Cn,λ )−αp−p |x − y|λ−n dx′ dx′′
(n−1)
G B1/4
Z a  
x1λ−αp−p−1 1 + x1−λ+αp+p+1 dx1 ≤ C aλ−αp−p + a .


≤C
0
 OPTIMAL EMBEDDINGS FOR LEBESGUE AND MORREY NORMS 23

On the other hand, if λ − αp − p = 1, then

Z Z
dist(x, Cn,λ )−αp−p |x − y|λ−n dx′ dx′′
(n−1)
G B1/4
Z a
≤C (1 − log 2x1 ) dx1 ≤ C (a + a| log a|) .
0

In both cases,
Z Z
dist(x, Cn,λ )−αp−p |x − y|λ−n dx′ dx′′ ≤ C aλ−αp−p + a(1 + | log a|)


(n−1)
G B1/4

and the lemma is proved. 

The following lemma will allow us to control the triple norm when y belongs to a
gap different from G. Notice that the exponents on the right-hand side of the estimate
are different in Lemmas 6.4 and 6.5.
Lemma 6.5. Let G = ((1 − γ)/2)l−1 (−γ/2, γ/2) + h for some l ≥ 1 and h ∈ R of the
form (6.10) (i.e. G is a gap-interval introduced in generation l). Assume n−αp−p > 0.
Then, we have that
 (n−αp−p)(l−1)  l−1 !
1 − γ 1 − γ
Z Z
dist(x, Cn,λ )−αp−p dx′ dx′′ ≤ C +l
(n−1)
G B1/4 2 2

for a constant C depending only on n, p, and α.

Proof. Let a = (γ/2) ((1 − γ)/2)l−1 so that G = (−a, a) + h. Using cylindrical coordi-
nates and (6.14), we have
Z Z
dist(x, Cn,λ )−αp−p dx′ dx′′
(n−1)
G B1/4
1
Z h+a Z
4
  −αp−p
2 ′′ 2 2
=C r + (h + a − x ) r n−2 drdx′′
h 0
1
Z h Z
4
  −αp−p 
2 ′′ 2 2 n−2 ′′
+ r + (x − h + a) r drdx
h−a 0
1
Z a Z
4
  −αp−p
2 ′′ 2 2
= 2C r + (z ) r n−2 drdz ′′ .
0 0

Notice that this integral is of the same type as the one in the right-hand side of (6.18),
taking λ = n there. It is now easy to check that one can proceed as in the final part
of Lemma 6.4 (taking λ = n there) and complete the proof. 
As mentioned before, the following lemma will also be used in the first part of the
proof of Proposition 6.3 in order to show that to bound the supremum in the definition
of |||∇uα,n |||p,λ;B(n) , it suffices to take y ∈ {0} × [−1/2, 1/2].
1
24 X. CABRÉ AND F. CHARRO

G1,1 G3,1 G2,1

j=0 j=2 y ′′ j=1

Figure 3. Gaps up to generation l = 4, classified according to (6.21).

Lemma 6.6. Let uα,n be given by (6.7). Then,

Z
p
|||∇uα,n ||| (n) = sup |∇uα,n (x)|p |x − y|λ−n dx
p,λ;B1 (n) (n)
y∈B 1 B1
Z 1 Z (6.19)
2
≤C sup dist(x, Cn,λ )−αp−p |x − y|λ−n dx′ dx′′
(n−1)
y ′ =0,|y ′′ |≤ 21 − 12 B1/4

for some constant C depending only on n, p, and α.

We postpone the proof of Lemma 6.6 until the end of the section and proceed instead
with the proof of Proposition 6.3. The idea is to “cluster” the gaps according to their
distance from y, and then use Lemmas 6.4 and 6.5.
Proof of Proposition 6.3. As a result of Lemma 6.6 we can assume that y ′ = 0 and
y ′′ ∈ [−1/2, 1/2]. To simplify the argument below, by Fatou’s lemma we may assume
that y ′′ is not the midpoint of any gap Gl,m given by (6.9) and (6.10).
Each generation l ≥ 1 introduces 2l−1 gaps Gl,m , and we can write
Z 1 Z
2
dist(x, Cn,λ )−αp−p |x − y|λ−n dx′ dx′′
(n−1)
− 21 B1/4
l−1
(6.20)
∞ X
X 2 Z Z
= dist(x, Cn,λ )−αp−p |x − y|λ−n dx′ dx′′ .
(n−1)
l=1 m=1 Gl,m B1/4

Recall that the length of the gap Gl,m is γ ((1 − γ)/2)l−1 .

We classify the 2l−1 gaps of generation l ≥ 2 according to their distance from y ′′ as
follows (see Figures 2 and 3):
(1) We split (−1/2, 1/2) into two halves, and notice that there are exactly 2l−2 gaps
of generation l in each half. We denote the gaps on the half-interval which does
not contain y ′′ by G0l,m , with m = 1, 2, . . . , 2l−2 . Since half of G1,1 = (−γ/2, γ/2)
lies between any of these gaps and y ′′, and the length of G1,1 is γ, we have that
γ
dist(y ′′, G0l,m ) ≥ .
2
l−2
(2) The 2 gaps remaining from step 1 are contained in an interval I of length
(1 − γ)/2. To this interval we apply the procedure in step 1, splitting I into
 OPTIMAL EMBEDDINGS FOR LEBESGUE AND MORREY NORMS 25

two halves. Notice that there are exactly 2l−3 gaps of generation l in each half.
We denote the gaps on the half-interval that is farthest from y ′′ by G1l,m (recall
that y ′′ is not the center of the full interval I), with m = 1, 2, . . . , 2l−3 . These
gaps satisfy
γ 1−γ
dist(y ′′ , G1l,m) ≥ .
2 2

(3) Iterating this procedure, at each step j we find exactly 2l−2−j gaps of genera-
tion l, denoted by Gjl,m with m = 1, 2, . . . , 2l−2−j . They satisfy

 j
γ 1−γ
dist(y ′′
, Gjl,m ) ≥ . (6.21)
2 2

(4) We continue the iteration until j = l − 2, which starts with only two gaps of
generation l left. The farthest from y ′′, denoted by Gl−2
l,1 , satisfies (6.21) with
j = l − 2. On the other hand, we denote the gap closest to y ′′ by Gl,1 .
Summarizing, among the 2l−1 gaps of generation l we have selected one, called Gl,1 ,
in step 4. The gap Gl,1 is the closest to y ′′ among those in generation l. The remaining
2l−1 −1 gaps have been clustered into l −1 families {G0l,m }m , . . . , {Gjl,m }m , . . . , {Gl−2
l,m }m ,
l−2−j
where the j-th family contains 2 gaps of generation l which, in addition, sat-
isfy (6.21).
With this classification, we have

∞ X
2 Zl−1 Z
X
dist(x, Cn,λ )−αp−p |x − y|λ−n dx′ dx′′
(n−1)
l=1 m=1 Gl,m B1/4

X∞ Z Z
= dist(x, Cn,λ )−αp−p |x − y|λ−n dx′ dx′′ (6.22)
(n−1)
l=1 Gl,1 B1/4

X l−2 2l−2−j
∞ X X Z Z
+ dist(x, Cn,λ )−αp−p |x − y|λ−n dx′ dx′′ .
(n−1)
l=2 j=0 m=1 Gjl,m B1/4

There are two cases to study in the sequel since integrals over Gl,1 and Gjl,m are
qualitatively different. The key difference is that (6.21) allows us to control |x − y|
from below and the integrals over Gjl,m become independent of y. Thus, we can apply
Lemma 6.5 to them. This is not possible for integrals over the gaps Gl,1 .
For the integrals over a gap Gl,1 , if y ′′ ∈ Gl,1 then we can apply Lemma 6.4 directly.
/ Gl,1 , we move y ′′ to the closest boundary point of Gl,1 from y ′′ , and
Instead, if y ′′ ∈
with this procedure the integral becomes larger (since all the distances from points in
(n−1)
B1/4 × Gl,1 decrease). With this new point y ′′ the integral can be bounded using
Lemma 6.4.
26 X. CABRÉ AND F. CHARRO

In fact, when we assume y ′′ ∈ Gl,1 Lemma 6.4 yields

X ∞ Z Z
dist(x, Cn,λ )−αp−p |x − y|λ−n dx′ dx′′
(n−1)
l=1 Gl,1 B1/4
∞  (λ−αp−p)(l−1)  l−1 ! (6.23)
X 1−γ 1−γ
≤C +l ≤ C,
l=1
2 2
uniformly in y, since (1 − γ)/2 < 1 and λ − αp − p > 0 by hypothesis (6.6).
On the other hand, given a gap Gjl,m , by (6.21) we have
Z Z
dist(x, Cn,λ )−αp−p |x − y|λ−n dx′ dx′′
(n−1)
Gjl,m B1/4
 (λ−n) j Z
1−γ
Z
≤C dist(x, Cn,λ )−αp−p dx′ dx′′ .
2 Gjl,m
(n−1)
B1/4

Then, Lemma 6.5 leads to

Z Z
dist(x, Cn,λ )−αp−p |x − y|λ−n dx′ dx′′
(n−1)
Gjl,m B1/4
 (λ−n) j+(n−αp−p)(l−1)  (λ−n) j+l−1!
1−γ 1−γ
≤C +l
2 2
1−γ λ−n


uniformly in m. Observe that by our choice of γ, we have 2
= 2 and thus
X l−2 2l−2−j
∞ X X Z Z
dist(x, Cn,λ )−αp−p |x − y|λ−n dx′ dx′′
(n−1)
l=2 j=0 m=1 Gjl,m B1/4

∞ X
l−2 (λ−n) j+(n−αp−p)(l−1)
  (λ−n) j+l−1!
X
l−2−j 1−γ 1−γ
≤C 2 +l
l=2 j=0
2 2
(λ−n+1) (l−1) ! (6.24)
∞ Xl−2  (λ−αp−p)(l−1) 
C X 1−γ 1−γ
= +l
2 l=2 j=0
2 2
∞  (λ−αp−p)(l−1)  (λ−n+1) (l−1) !
C X 1−γ 1−γ
= (l − 1) + l(l − 1) ≤ C,
2 l=2
2 2
uniformly in y, since (1 − γ)/2 < 1, λ − αp − p > 0, and λ > n − 1 by hypothesis.
Then, (6.20), (6.22), (6.23), and (6.24) give
Z 1 Z
2
dist(x, Cn,λ )−αp−p |x − y|λ−n dx′ dx′′ ≤ C
(n−1)
− 12 B1/4

uniformly in y, and the proof is complete. 

We conclude the article with the proof of Lemma 6.6.
 OPTIMAL EMBEDDINGS FOR LEBESGUE AND MORREY NORMS 27

Proof of Lemma 6.6. Note first that the support of uα,n , given by (6.7), is included in
(n−1) (n)
B 1/4 × [−1, 1]. Hence, for any given y ∈ B 1 we have
Z Z 1Z
p λ−n
|∇uα,n (x)| |x − y| dx ≤ C dist(x, Cn,λ )−αp−p |x − y|λ−n dx′ dx′′ ,
(n) (n−1)
B1 −1 B1/4

where we have used that the modulus of the gradient of a distance function is equal to
1 a.e. Then, our goal is to show that
Z 1Z
sup dist(x, Cn,λ )−αp−p |x − y|λ−n dx′ dx′′
(n) (n−1)
y∈B 1 −1 B1/4
Z 1 Z
2
≤C sup dist(x, Cn,λ )−αp−p |x − y|λ−n dx′ dx′′ ,
(n−1)
y ′ =0,|y ′′ |≤ 21 − 12 B1/4

from which (6.19) follows.

(n)
First we will prove that the supremum over y ∈ B 1 is bounded by the supremum
over the axis, i.e., y ∈ {0} × [−1, 1]. Then, that it actually suffices that |y ′′| ≤ 1/2
instead of |y ′′| ≤ 1, and finally that it is enough to integrate x′′ over [−1/2, 1/2]
instead of the whole [−1, 1]. In doing these we will use twice the monotonicity result
in Lemma 2.1.
(n)
Therefore, consider y ∈ B 1 , and let us show that
Z 1Z
J1 (y) := dist(x, Cn,λ )−αp−p |x − y|λ−n dx′ dx′′ ≤ J1 (0, y ′′),
(n−1)
−1 B1/4

(n)
where y = (y ′, y ′′ ) ∈ B 1 , and thus y ′′ ∈ [−1, 1]. In fact, upon a rotation in the x′
variables, we can assume that y ′ = (y1 , 0) ∈ R × Rn−2 with y1 ≥ 0. Hence, we are
(n−1)
under the hypotheses of Lemma 2.1 with Ω = B1/4 ×[−1, 1], h(x) = dist(x, Cn,λ )−αp−p
(which is non-increasing with respect to x1 in {x1 ≥ 0}), θ = n − λ, and η = 0.
Therefore, Lemma 2.1 gives that J1 is non-increasing with respect to y1 in [0, ∞).
Hence we conclude that J1 (y) ≤ J1 (0, y ′′), as desired.
A similar argument shows that we just need to consider the case |y ′′| ≤ 1/2 instead
of |y ′′| ≤ 1. In fact, define
Z 1Z   λ−n
′′ −αp−p ′ 2 ′′ ′′ 2 2
J2 (y ) := dist(x, Cn,λ ) |x | + (x − y ) dx′ dx′′
(n−1)
−1 B1/4

and assume |y ′′| ≥ 1/2. By symmetry, we can assume y ′′ ≥ 1/2. In order to apply
Lemma 2.1, we take x′′ as the direction e1 which is “privileged” in the lemma, while the
(n−1)
rest of the hypotheses are fulfilled for Ω = [−1, 1] × B1/4 , h(x) = dist(x, Cn,λ )−αp−p
(which is non-increasing with respect to x′′ in {x′′ ≥ 1/2}), θ = n − λ, and η = 1/2,
since the set Cn,λ is contained in {0} × [−1/2, 1/2] (see Figure 4). Then, Lemma 2.1
gives that J2 (y) is non-increasing with respect to y ′′ in [1/2, ∞), and therefore it is
enough to study the case y ′′ ≤ 1/2. The case y ′′ ≤ −1/2 follows similarly by taking
η = −1/2 in the lemma.
28 X. CABRÉ AND F. CHARRO

x′′ ✻
1
xs
y ′′ ✏✏
s✏
PP
∗Ps
1 x
2

✲
|x′ |

− 21

−1

Figure 4. Monotonicity argument for J2 in the proof of Lemma 6.6.

(n−1)
Finally, let y = (0, y ′′) with |y ′′| ≤ 1/2. Notice that for every x ∈ B1/4 × [1/2, 1]
and its reflected x∗ with respect to {x′′ = 1/2}, we have |x∗ − y| ≤ |x − y| and
dist(x∗ , Cn,λ) ≤ dist(x, Cn,λ ), which give
dist(x, Cn,λ )−αp−p |x − y|λ−n ≤ dist(x∗ , Cn,λ )−αp−p |x∗ − y|λ−n .
This leads to
Z 1 Z
dist(x, Cn,λ )−αp−p |x − y|λ−n dx′ dx′′
1 (n−1)
2
B1/4
Z 1 Z
2
≤ dist(x, Cn,λ )−αp−p |x − y|λ−n dx′ dx′′ ,
(n−1)
0 B1/4

(n−1)
and similarly for the integral over B1/4 × [−1, − 12 ]. Therefore,
Z 1Z   λ−n
−αp−p ′ 2 ′′ ′′ 2 2
dist(x, Cn,λ ) |x | + (x − y ) dx′ dx′′
(n−1)
−1 B1/4
1
Z
2
Z   λ−n
2
−αp−p ′ 2 ′′ ′′ 2
≤2 dist(x, Cn,λ ) |x | + (x − y ) dx′ dx′′ ,
(n−1)
− 12 B1/4

which completes the proof of the lemma. 

Acknowledgements: The first author would like to thank Joan Mateu and Joan
Verdera for an interesting discussion on the topic of this paper.
 OPTIMAL EMBEDDINGS FOR LEBESGUE AND MORREY NORMS 29

References
[1] Adams D. R.; A note on Riesz potentials, Duke Math. J. 42 (1975), 765–778.
[2] Adams, D. R., Xiao, J.; Morrey potentials and harmonic maps, Comm. Math. Phy. 308 (2011),
439–456.
[3] Adams, D. R., Xiao, J.; Erratum to: Morrey potentials and harmonic maps, Comm. Math. Phys.
339 (2015), 769–771.
[4] Adams, D. R., Xiao, J.; Restrictions of Riesz-Morrey potentials, Ark. Mat. 54 (2016), 201–231.
[5] Cabré, X., Figalli, A., Ros-Oton, X., Serra J.; Stable solutions to semilinear elliptic equations are
smooth up to dimension 9, preprint arXiv:1907.09403.
[6] Cabré, X., Ros-Oton, X.; Sobolev and isoperimetric inequalities with monomial weights, J. Dif-
ferential Equations 255 (2013), 4312–4336.
[7] Falconer, K.; Fractal geometry: mathematical foundations and applications, John Wiley & Sons,
2004.
[8] Schoen, R., Yau, S.-T.; Conformally flat manifolds, Kleinian groups and scalar curvature, Invent.
Math. 92 (1988), 47–71.
[9] Talenti, G.; A weighted version of a rearrangement inequality, Ann. Univ. Ferrara 43 (1997),
121–133.

X.C.1,2,3 — 1 ICREA, Pg. Lluis Companys 23, 08010 Barcelona, Spain & 2 Universitat
Politècnica de Catalunya, Departament de Matemàtiques, Diagonal 647, 08028 Barcelona,
Spain & 3 BGSMath, Campus de Bellaterra, Edifici C, 08193 Bellaterra, Spain.
E-mail address: [email protected]

F.C. — Department of Mathematics, Wayne State University, 656 W. Kirby, De-

troit, MI 48202, USA.
E-mail address: [email protected]