Trace Results on Domains with Self-Similar Fractal Boundaries

Yves Achdou ∗, Nicoletta Tchou †.

January 3, 2008

Abstract
This work deals with trace theorems for a family of ramified bidimensional domains
Ω with a self-similar fractal boundary Γ∞ . The fractal boundary Γ∞ is supplied with a
probability measure µ called the self-similar measure. Emphasis is put on the case when the
domain is not a ǫ − δ domain and the fractal is not post-critically finite, for which classical
results cannot be used. It is proven that the trace of a square integrable function belongs
to Lpµ for all real numbers p ≥ 1. A counterexample shows that the trace of a function in
H 1 (Ω) may not belong to BM O(µ) (and therefore may not belong to L∞ µ ). Finally, it is
proven that the traces of the functions in H 1 (Ω) belong to H s (Γ∞ ) for all real numbers s
such that 0 ≤ s < dH /4, where dH is the Hausdorff dimension of Γ∞ . Examples of functions
whose traces do not belong to H s (Γ∞ ) for all s > dH /4 are supplied.
There is an important contrast with the case when Γ∞ is post-critically finite, for which the
square integrable functions have their traces in H s (Γ∞ ) for all s such that 0 ≤ s < dH /2.

1 Introduction
This work deals with some properties of H 1 (Ω) for a family of ramified domains Ω of R2 with
a self-similar fractal boundary called Γ∞ below, see Figure 1. The domain Ω depends on a
parameter a, 1/2 ≤ a ≤ a∗ . The restriction a ≤ a∗ allows for the construction of Ω as a union
of non-overlapping sub-domains, see (2) below. The inessential condition 1/2 ≤ a ensures that
the Hausdorff dimension of Γ∞ is not smaller than one.
Such a geometry can be seen as a bidimensional idealization of the bronchial tree, for example.
Indeed, this work is part of a wider project aimed at simulating the diffusion of medical sprays
in lungs. Since the exchanges between the lungs and the circulatory system take place only
in the last generations of the bronchial tree (the smallest structures), reasonable models for
the diffusion of, e.g., oxygen may involve a non-homogeneous Neumann or Robin condition on
the boundary Γ∞ . Similarly, the lungs are mechanically coupled to the diaphragm, which also
implies non-homogeneous boundary conditions on Γ∞ , if one is interested in a coupled fluid-
structure model. It is therefore necessary to study traces of functions on Γ∞ . In that respect,
the present work is a continuation of [2].
Sobolev spaces of functions defined in irregular domains have been widely studied in the litera-
ture:

• Jones [6] (and Vodopjanov et al [18] in the case n = 2, see also [11, 7]) have studied the
open bounded subsets Ω of Rn such that there exists a continuous extension operator from
∗
UFR Mathématiques, Université Paris Diderot, Case 7012, 75251 Paris Cedex 05, France and Laboratoire
Jacques-Louis Lions, Université Paris 6, 75252 Paris Cedex 05. [email protected]
†
IRMAR, Université de Rennes 1, Rennes, France, [email protected]

1
 W ℓ,p (Ω) in W ℓ,p (Rn ), for all nonnegative integers ℓ and real numbers p, 1 ≤ p ≤ ∞. Jones
has proved that if Ω is a ǫ − δ domain for some parameters ǫ, δ > 0, see [6, 7] for the
definition, then the above extension property is true. Moreover, in dimension two, if the
extension property stated above is true, then Ω is a ǫ − δ domain for some parameters
ǫ, δ > 0. In dimension two, the definition of such domains is equivalent to that of quasi-
disks, see [11].

• Jonsson and Wallin [7] have considered closed subsets F of Rn supplied with a Borel
measure µ such that there exists a positive real number d and two positive constants c1
and c2 with
c1 r d ≤ µ(B(x, r)) ≤ c2 r d ,
for all x ∈ F and r < 1 (here B(x, r) is the ball in F with center x and radius r, with
respect to the Euclidean distance in Rn ); in [7], these sets are called d-sets. Sobolev and
Besov spaces can be defined on the d-sets. Using as a main ingredient Whitney extension
theory, Jonsson and Wallin have proved extension and trace results for Besov and Sobolev
spaces. In particular, see [7] page 103, there exists a continuous trace operator from
α− n−d ,p n−d
W α,p (Rn ) onto W p (F ), if 1 ≤ p ≤ ∞ and 0 < α − p < 1.

• There is also a a growing interest on analysis on self-similar fractal sets, see for instance
Kigami [8], Strichartz [16, 17], Mosco[14, 13] and references therein. These works aim at
intrinsically defining function spaces using Dirichlet forms and a different metric from the
Euclidean one. Most of the results concerning analysis on fractal sets are subject to the
important assumption that the set is post-critically finite (or p.c.f.), see [8], page 23 for
the definition.

• The question of extensions or traces naturally arises in boundary value or transmission

problems in domains with fractal boundaries. Results in this direction have been given in
[15, 10, 9] for the Koch flake. Here also, the assumption that the fractal set is p.c.f. is
generally made.

Our goal here is to study the traces of functions of H 1 (Ω) on the fractal boundary Γ∞ . Note
that this is different from considering the traces of functions of H 1 (R2 ) on Γ∞ .

Following Hutchinson[5] and Kigami[8], Γ∞ is supplied with a probability measure µ called

the self-similar measure and defined in (17) below. It will also be seen that with µ, Γ∞ is a
log 2
dH -set, where dH = − log a ≥ 1, so the results of Jonsson and Wallin can be applied.
On the other hand, it is possible to define a trace operator which maps H 1 (Ω) on L2µ (Γ∞ ). This
has been done in [2].
We will see that if 1/2 ≤ a < a∗ , then Ω is a ǫ − δ domain and Γ∞ is post-critically finite.
In this case, the results of [6] and those of [7] can be combined to obtain trace results. This
is done in § 4.1 below: we prove that the trace of a function in H 1 (Ω) belongs to H s (Γ∞ ) for
0 ≤ s ≤ dH /2.
The main part of the present paper focuses on the case a = a∗ for which Ω is not a ǫ − δ domain
and Γ∞ is not post-critically finite. Up to our knowledge, this case is not studied in the available
literature.
The main results are

• Theorem 4 , which states that if 1/2 ≤ a ≤ a∗ then the trace of a function in H 1 (Ω)
belongs to Lpµ (Γ∞ ), for all real numbers p such that 1 ≤ p < ∞.

2
 • Theorem 5 in the case a = a∗ which gives an example of a function in H 1 (Ω) whose trace
on Γ∞ has not a bounded mean oscillation with respect to µ, (and, as a consequence, does
not belong to L∞ ∞
µ (Γ )).

• Theorem 6 in the case a = a∗ , which states that the trace of a function in H 1 (Ω) belongs
to H s (Γ∞ ) defined in (29) below, for all real numbers s such that 0 ≤ s < dH /4, where
dH is the Hausdorff dimension of Γ∞ , and Proposition 5, which states the existence of a
function in H 1 (Ω) whose trace does not belong to H s (Γ∞ ) for all s > dH /4. Note the
important contrast with the case a < a∗ for which the trace of a function in H 1 (Ω) belongs
to H s (Γ∞ ) for 0 ≤ s ≤ dH /2. Similar results concerning the spaces W 1,p (Ω), 2 ≤ p < ∞,
are given in Remark 11.

The article is organized as follows: the geometry is presented in section 2; in particular, the
critical case a = a∗ when Ω is not a quasi-disk and Γ∞ is not a post-critical set is carefully
discussed. In section 3, we recall some of the results of [2] on the space H 1 (Ω), concerning
Poincaré inequality and the construction of the trace operator. The main part of the paper
is Section 4 where the accurate trace results mentioned above are given. Finally, for the ease
of the reader, the most technical proofs (involving geometrical lemmas) are postponed to two
appendices at the end of the paper.

2 The geometry
Let a be a positive parameter. Consider two similitudes Fi , i = 1, 2 respectively defined by the
following:      
(−1)i 1 − √a2 + √a2 x1 + (−1)i x2
Fi (x) =    .
1 + √a2 + √a2 x2 + (−1)i+1 x1

The similitude Fi has the dilation ratio a and the rotation angle (−1)i+1 π/4.
Consider also two points in R2 , P1 = (−1, 0),
√ P2 =√(1, 0), and define P3 = F1 (P√ 1 ) = (−1,
√1),
P4 = F2 (P2 ) = (1, 1), P5 = F1 (P2 ) = (−1+a 2, 1+a 2) and P6 = F2 (P1 ) = (1−a 2, 1+a 2).
Let Y 0 be the open hexagonal subset of R2 defined as the convex hull of the last six points.
 
Y 0 = Interior Conv(P1 , P2 , P3 , P4 , P5 , P6 ) .
√
It is easily seen that one must choose a ≤ 2/2 to prevent F1 (Y 0 ) and F2 (Y 0 ) from overlapping.
For n ≥ 1, we call An the set containing all the 2n mappings from {1, . . . , n} to {1, 2}. Similarly
A∞ = {1, 2}N . We define

Mσ = Fσ(1) ◦ · · · ◦ Fσ(n) for σ ∈ An , (1)

and the ramified open domain, see Figure 1,

  
∞
Ω = Interior Y 0 ∪ ∪ ∪ Mσ (Y 0 ) . (2)
n=1 σ∈An

Stronger constraints must be imposed on a to prevent the sets Mσ (Y 0 ), σ ∈ An , n > 0, from

overlapping. It can be shown by elementary geometrical arguments that the condition is
√ √
2 2a5 + 2a4 + 2a2 + 2a − 2 ≤ 0, (3)

3
i.e. ,
a ≤ a∗ ≃ 0.593465.
In all what follows, we will take 1/2 ≤ a ≤ a∗ .
We call Γ∞ the self similar set associated to the similitudes F1 and F2 , i.e. the unique compact
subset of R2 such that
Γ∞ = F1 (Γ∞ ) ∪ F2 (Γ∞ ).
The Moran condition, see [12, 8], is that there exists a nonempty bounded open subset O of
R2 such that F1 (O) ∩ F2 (O) = ∅ and F1 (O) ∪ F2 (O) ⊂ O. In our case, one can take O = Ω.
As shown in [12, 8] this condition allows for computing the Hausdorff dimension of Γ∞ ; we find
that
dimH (Γ∞ ) = − log 2/ log a. (4)
For instance, if a = a∗ , then dimH (Γ∞ ) ≃ 1.3284371.
We split the boundary of Ω into Γ∞ , Γ0 = [−1, 1] × {0} and Σ = ∂Ω\(Γ0 ∪ Γ∞ ). For what
follows, it is important to define the polygonal open domain Y N obtained by stopping the above
construction at step N + 1,
  
N N
0
Y = Interior Y ∪ ∪ ∪ Mσ (Y ) 0 . (5)
n=1 σ∈An

We introduce the open domains Ωσ = Mσ (Ω) and ΩN = ∪σ∈AN Ωσ = Ω\Y N −1 . We also define
the sets Γσ = Mσ (Γ0 ) and ΓN = ∪σ∈AN Γσ . The one-dimensional Lebesgue measure of Γσ for
σ ∈ AN and of ΓN are given by

|Γσ | = aN |Γ0 | and |ΓN | = (2a)N |Γ0 |.

It is clear that if a > 1/2 then limN →∞ |ΓN | = +∞.

45◦ 3.5

3.0

2.5

2.0
P5 P6

1.5

2a
1.0
P3 P4

0.5

0.0
−3 −2 −1 0 1 2 3
P1 P2

Figure 1: Left, the construction (more exactly Y 3 ). Right, the ramified domain Ω for the critical
value a = a∗ .

4
2.1 The sub-critical case a < a∗
In the case when 1/2 ≤ a < a∗ , F1 (Γ∞ ) ∩ F2 (Γ∞ ) is empty. As a consequence
Lemma 1 If 1/2 ≤ a < a∗ , there exists ǫ > 0 and δ > 0 such that Ω is a ǫ − δ domain as
defined by Jones [6], see also [7] or in an equivalent manner a quasi-disk, see [11].
Proof. Some of the details will be skipped for brevity. The proof is similar to that given in
[11] page 71 for a different domain Ω. We have to show that ∂Ω is a quasi-circle, i.e. the image
of a circle by a quasiconformal mapping of the plane into itself. By Ahlfors theorem, see [3],
this is equivalent to the following: there exists c > 0 such that for any x and y on ∂Ω, if γ is
a continuous sub-arc of ∂Ω which joins x and y with minimal diameter, i.e. γ is a continuous
function from [0, 1] to ∂Ω which minimizes diam(γ[0, 1]) subject to γ(0) = x and γ(1) = y, then
for each t ∈ [0, 1],
|x − γ(t)| ≤ c|x − y|. (6)
This condition mainly stems from the following two facts:
1. there exists a constant δ > 0 such that if x ∈ F1 (Ω) and y ∈ F2 (Ω) then |x − y| ≥ δ: on
the other hand, if γ is a sub-arc as above, then diam(γ[0, 1]) ≤ diam(Ω), so (6) is satisfied
with c = diam(Ω)/δ. By an elementary scaling, the same is true if x ∈ Mσ (F1 (Ω)) and
y ∈ Mσ (F2 (Ω)).

2. We call Σ2 the polygonal line Σ2 = [P2 , P4 ] ∪ [P4 , P6 ] (see theSbeginning of § 2 for the
definition of the points P2 , P4 and P6 ). Take x ∈ Σ2 and y ∈ n≥1 F22n Σ2 . Then there
exists a constant C > 0 such that |x − y| ≥SC, so (6) is satisfied with c = diam(Ω)/C. The
same is true if x ∈ Mσ (Σ2 ) and y ∈ Mσ ( n≥1 F22n Σ2 ). By symmetry the same is true if
S
x ∈ Mσ (Σ1 ) and y ∈ Mσ ( n≥1 F12n Σ1 ), where Σ1 = [P1 , P3 ] ∪ [P3 , P5 ].
P
For all the points x, y on ∂Ω, the minimal sub-arc between x and y can be split into pi=1 γi
where p ≤ 3, γi is a minimal sub-arc joining its endpoints, and γi is either made of a finite
number of straight line segments or falls into the two cases above. This leads to (6).

Remark 1 If a < a∗ , the set Γ∞ is a post-critically finite set as defined in [8] page 23, since
F1 (Γ∞ ) ∩ F2 (Γ∞ ) = ∅.

2.2 The critical case a = a∗

In the critical case when a = a∗ , it will be proved below that F1 (Γ∞ ) ∩ F2 (Γ∞ ) ⊂ Γ∞ is a non
countable set and that Ω is not a ǫ − δ domain.
We introduce the similitudes

G1 = F1 ◦ F2 and G2 = F2 ◦ F1 . (7)

It will sometimes be convenient to use the redundant notations

Gc1 = G2 and Gc2 = G1 , (analogously, F1c = F2 and F2c = F1 ). (8)

We need some preliminary lemmas before characterizing

Ξ∞ = F1 (Γ∞ ) ∩ F2 (Γ∞ ) (9)

in Proposition 1 below. These lemmas will also be useful in Appendix B.

To simplify the notations, we set a = a∗ .

5
Lemma 2 Let H be the number
√
H = sup x2 = (1 + 3a/ 2)/(1 − a2 ).
x∈Ω

For all k ∈ N, for all n ≥ 2k + 2 and all σ ∈ An with

σ(2i − 1) 6= σ(2i), ∀1 ≤ i ≤ k, and σ(2k + 1) = σ(2k + 2), (10)

we have
Ha2k ≥ infσ (H − x2 ) ≥ δa2k (11)
x∈Ω

for δ > 0 independent of k, n and σ.

Proof. Introduce √
h1 = sup x2 = 1 + 3a/ 2
x∈Y 1

and
h2 = sup x2 = sup x2 .
x∈F1 ◦F1 (Ω) x∈F2 ◦F2 (Ω)

We have h1 < h2 < H = h1 /(1 − a2 ). The bound h2 < H can be obtained by realizing that

h2 = sup x2 + a4 H,
x∈F1 ◦F1 ◦F2 (Y 0)

and explicitly computing supx∈F1 ◦F1 ◦F2 (Y 0 ) x2 .

Therefore, there exists δ > 0 such that h2 < H − δ.
Let σ ∈ An satisfy (10). By a simple scaling argument, we see that
k−1
X 1 − a2k
sup x2 = h1 a2i + h2 a2k = h1 + h2 a2k = H(1 − a2k ) + h2 a2k ≤ H − δa2k ,
x∈Ωσ 1 − a2
i=0

which yields the lower bound in (11), and

k−1
X
sup x2 = h1 a2i + h2 a2k ≥ H(1 − a2k ),
x∈Ωσ i=0

which yields the upper bound in (11).

Lemma 3 There exists two positive constants c and C such that for all n > 0,

1. for all σ ∈ An , the distance of Gσ(1) ◦ · · · ◦ Gσ(n) (Γ0 ) to the vertical axis {x1 = 0} is greater
than c,

2. for all k ∈ {1, . . . , n − 1} and σ, σ ′ ∈ An such that σ(i) = σ ′ (i) for all i < k and σ(k) 6=
σ ′ (k),

ca2(k−1) ≤ d(Gσ(1) ◦ · · · ◦ Gσ(n) (Γ0 ), Gσ′ (1) ◦ · · · ◦ Gσ′ (n) (Γ0 )) ≤ Ca2(k−1) .

where d(x, y) is the Euclidean distance between the two points x, y ∈ R2 .

6
Proof. The quantity minx∈Gσ(1) ◦···◦Gσ(n) (Γ0 ) |x1 | can be bounded from below, because its min-
imum with respect to σ is achieved for σ = (2, 1, 1, 1, 1, . . . , 1): this quantity is larger than
the abscissa c of G2 (M ), where M is the fixed point of G1 : elementary calculus yields that
2
c = (1 − √a2 + a2 ) 1−2a
1−a2 > 0.
For proving the second point,

d(Gσ(1) ◦ · · · ◦ Gσ(n) (Γ0 ), Gσ′ (1) ◦ · · · ◦ Gσ′ (n) (Γ0 ))

=a2(k−1) d(Gσ(k) ◦ · · · ◦ Gσ(n) (Γ0 ), Gσ′ (k) ◦ · · · ◦ Gσ′ (n) (Γ0 )) ≥ ca2(k−1) ,

where we have used the first point and the fact that Gσ(k) ◦ · · · ◦ Gσ(n) (Γ0 ) and Gσ′ (k) ◦ · · · ◦
Gσ′ (n) (Γ0 ) are separated by the axis {x1 = 0}. We have obtained the lower bound. The upper
bound comes from the fact that d(Gσ(k) ◦ · · · ◦ Gσ(n) (Γ0 ), Gσ′ (k) ◦ · · · ◦ Gσ′ (n) (Γ0 )) is bounded
from above by the diameter of Ω.

Definition 1 For σ ∈ AN , N = 2(n + 2), we define the integer prox(σ) by



 if σ(1, 2, 3, 4) 6= (1, 2, 2, 2) and σ(1, 2, 3, 4) 6= (2, 1, 1, 1), then prox(σ) = 0,

else if σ(5) = σ(6), then prox(σ) = 1,

 σ(2i − 1) 6= σ(2i), ∀i, 3 ≤ i ≤ k + 1,
 else prox(σ) is the largest number k s.t.
k ≤ n + 1.
(12)

We will see in Lemma 4 below that the larger prox(σ) is, the closer Γσ to the axis x1 = 0.

Remark 2 Note that, for k ∈ {1, . . . , n}, there are 22n−k+1 maps σ in AN such that prox(σ) =
k, there are 2n+1 maps σ in AN such that prox(σ) = n + 1, and there are 7 · 22n+1 maps σ in
AN such that prox(σ) = 0.

Lemma 4 There exists two positive numbers H ′ and δ′ such that for all n ≥ 0, for all σ ∈ A2n+4 ,
the distance of Γσ to the axis {x1 = 0} is greater than δ′ a2prox(σ) and smaller than H ′ a2prox(σ) .

Proof. The proof uses Lemma 2 and the fact that the image of the line {x2 = H} by F1 ◦ F2 ◦
F2 ◦ F2 or F2 ◦ F1 ◦ F1 ◦ F1 is exactly the axis {x1 = 0}.

Proposition 1 The set Ξ∞ given by (9) is contained in the vertical axis {x1 = 0}, and is
characterized by
n o
Ξ∞ = lim G1 ◦ F2 ◦ F2 ◦ Gσ(1) ◦ · · · ◦ Gσ(n) (O); σ ∈ A∞ (13)
n→∞

where O = (0, 0) is the origin. Moreover, for P ∈ Ξ∞ , there exists a unique σ ∈ A∞ such that

P = lim G1 ◦ F2 ◦ F2 ◦ Gσ(1) ◦ · · · ◦ Gσ(n) (O) = lim G2 ◦ F1 ◦ F1 ◦ Gcσ(1) ◦ · · · ◦ Gcσ(n) (O). (14)

n→∞ n→∞

The Hausdorff dimension of Ξ∞ is dimH (Γ∞ )/2.

Proof. Since F1 (Γ∞ ) = Γ∞ ∩ {x1 ≤ 0} and F2 (Γ∞ ) = Γ∞ ∩ {x1 ≥ 0}, we immediately

see that Ξ∞ = Γ∞ ∩ {x1 = 0}. Then (13) is a consequence of Lemma 4. Moreover, if P =
limn→∞ G1 ◦ F2 ◦ F2 ◦ Gσ(1) ◦ · · · ◦ Gσ(n) (O), then by symmetry, we also have P = limn→∞ G2 ◦
F1 ◦ F1 ◦ Gcσ(1) ◦ · · · ◦ Gcσ(n) (O). Finally, for σ, σ ′ ∈ A∞ such that σ 6= σ ′ , there exists k > 0 such
that σ(i) = σ ′ (i) for all i < k and σ(k) 6= σ ′ (k): from the second statement of Lemma 3, we see
that d(G1 ◦ F2 ◦ F2 ◦ Gσ(1) ◦ · · · ◦ Gσ(n) (O), G1 ◦ F2 ◦ F2 ◦ Gσ′ (1) ◦ · · · ◦ Gσ′ (n) (O) ≥ ca2k , for all

7
n > k. This yields the uniqueness in (14).
The set Ξ∞ is the image by G1 ◦ F2 ◦ F2 of the self-similar fractal associated with the similitudes
G1 and G2 , whose dilation ratii 2
√ are a2 < 1/2. The latter is a Cantor set contained in the
horizontal line x2 = (1 + 3a/ 2)(1 − a ): since the Moran condition is satisfied, its Hausdorff
dimension is − log(2)/ log(a2 ) = − log(2)/(2 log(a)) = 1/2 dimH (Γ∞ ). Therefore, the Hausdorff
dimension of Ξ∞ is dimH (Γ∞ )/2.
To illustrate Proposition 1, we display in Figure 2 the sets Y 0 , F1 (Y 0 ), G1 (Y 0 ), G1 ◦ F2 (Y 0 ),
G1 ◦ F2 ◦ F2 (Y 0 ), G1 ◦ F2 ◦ F2 ◦ F1 (Y 0 ), G1 ◦ F2 ◦ F2 ◦ G1 (Y 0 ), G1 ◦ F2 ◦ F2 ◦ G1 ◦ F1 (Y 0 ),
G1 ◦ F2 ◦ F2 ◦ G1 ◦ G1 (Y 0 ),. . . .

Figure 2: In dark, Y 0 and the images of Y 0 by F1 , G1 , G1 ◦ F2 , G1 ◦ F2 ◦ F2 , G1 ◦ F2 ◦ F2 ◦ F1 ,

G1 ◦ F2 ◦ F2 ◦ G1 , G1 ◦ F2 ◦ F2 ◦ G1 ◦ F1 , G1 ◦ F2 ◦ F2 ◦ G1 ◦ G1 ,. . . .

Remark 3 A consequence of (14) is that Ω is not a ǫ − δ domain if a = a∗ .

Remark 4 A consequence of (13) and (14) is that the set Γ∞ is not post-critically finite if
a = a∗ . Note that most of the available results on function spaces on self-similar fractals are
valid for post-critically finite sets only.

Remark 5 Other examples of non post-critically finite fractal boundaries may be constructed
by taking similitudes with rotation angles ±π/(2ℓ+1 ), where ℓ is a positive integer, and a suitable
dilation factor.

3 The space H 1 (Ω)

Let H 1 (Ω) be the space
 of functions in L 2
(Ω) withnfirst order partial derivatives 2
in L (Ω). We
1 1 n
also define V(Ω) = v ∈ H (Ω); v|Γ0 = 0 and V(Y ) = v ∈ H (Y ); v|Γ0 = 0 .
The results stated below are important for the study of elliptic boundary value problems in Ω.
We refer to [2] for their proofs.
We will sometimes use the notation . to indicate that there may arise constants in the estimates,
which are independent of the index n in Ωn (recall that ΩnSis the union
S of all Mσ (Ω), σ ∈ An )
or Γn ( Γn is the union of all Mσ (Γ0 ), σ ∈ An ), Y n = Y 0 ∪ 1≤p≤n σ∈Ap Mσ (Y 0 ), or the index
σ in Ωσ = Mσ (Ω) or Γσ = Mσ (Γ0 ).

8
3.1 Poincaré inequality and consequences
Theorem 1 There exists a constant C > 0, such that

∀v ∈ V(Ω), kvk2L2 (Ω) ≤ Ck∇vk2L2 (Ω) .

Corollary 1 There exists a positive constant C such that for all v ∈ H 1 (Ω),
 
kvk2L2 (Ω) ≤ C k∇vk2L2 (Ω) + kv|Γ0 k2L2 (Γ0 ) .

Corollary 2 There exists a positive constant C such that for all integer n ≥ 0 and for all
σ ∈ An , for all v ∈ H 1 (Ωσ ),
 
kvk2L2 (Ωσ ) ≤ C a2n k∇vk2L2 (Ωσ ) + an kv|Γσ k2L2 (Γσ ) ,

and for all v ∈ H 1 (Ωn )

 
kvk2L2 (Ωn ) ≤ C a2n k∇vk2L2 (Ωn ) + an kv|Γn k2L2 (Γn ) .

We need to estimate kvk2L2 (Ωn ) when v ∈ H 1 (Ω):

Lemma 5 There exists a positive constant C such that for all v ∈ H 1 (Ω), for all n ≥ 0,
 
kvk2L2 (Ωn ) ≤ C (2a2 )n k∇vk2L2 (Ω) + kv|Γ0 k2L2 (Γ0 ) . (15)

Condition (3) implies 2a2 < 1, because 2(a∗ )2 ∼ 0.7044022575. Thus, (15) implies the Rellich
type theorem:

Theorem 2 (Compactness) The imbedding of H 1 (Ω) in L2 (Ω) is compact.

The following lemma will be useful for defining a trace operator on Γ∞ :

Lemma 6 There exists a positive constant C such that for all v ∈ H 1 (Ω), for all integers p ≥ 0,
X Z  
(v|Γσ )2 ≤ C(2a)p k∇vk2L2 (Ω) + kvk2L2 (Ω) . (16)
σ∈Ap Γσ

|Γp |
Remark 6 Note that |Γ0 |
= (2a)p , so (16) is equivalent to
Z
1 X
(v|Γσ )2 . k∇vk2L2 (Ω) + kvk2L2 (Ω) .
|Γp | Γσ
σ∈Ap

Corollary 3 There exists a positive constant C such that for all v ∈ H 1 (Ω), for all integers
p ≥ 0,
X Z
(v|Γσ − hv|Γ0 i)2 ≤ C(2a)p k∇vk2L2 (Ω) ,
σ∈Ap Γσ

where hv|Γ0 i is the mean value of v|Γ0 on Γ0 .

9
3.2 A trace operator on Γ∞
For defining traces on Γ∞ , we recall the classical result on self-similar measures, see [4]:
Theorem 3 There exists a unique Borel regular probability measure µ on Γ∞ such that for any
Borel set A ⊂ Γ∞ ,
1
1

µ(A) = µ F1−1 (A) + µ F2−1 (A) . (17)
2 2
The measure µ is called the self-similar measure defined in the self-similar triplet (Γ∞ , F1 , F2 ).
Proposition 2 For 1/2 ≤ a ≤ a∗ , the measure µ is d-measure on Γ∞ , with d = − log 2/ log a,
according to the definition in [7], page 28: there exists two positive constants c1 and c2 such that
c1 r d ≤ µ(B(x, r)) ≤ c2 r d ,
for any r 0 < r < 1 and x ∈ Γ∞ , where B(x, r) is the ball of Γ∞ centered at x and with radius
r. In other words the closed set is a d-set, see [7], page 28.
Proof. The proof stems from the Moran condition in § 2. It is due to Moran [12] and has been
extended by Kigami, see [8], §1.5, especially Proposition 1.5.8 and Theorem 1.5.7.
Let L2µ be the Hilbert space of the functions defined µ a.e. on Γ∞ that are µ-measurable and
qR
square integrable with respect to µ, with the norm kvkL2µ = 2
Γ∞ v dµ.
A Hilbertian basis of L2µ can be constructed with e.g. Haar wavelets.
Similarly we define Lpµ , p ∈ [1, +∞) as the space of the measurable functions v on Γ∞ such that
R p
R p

1/p
Γ∞ v dµ < ∞, endowed with the norm kvkLµ = Γ∞ v dµ .
p

Remark 7 It can be proved that, even if a = a∗ , µ(F1 (Γ∞ ) ∩ F2 (Γ∞ )) = 0, see [12, 5, 8].
Similarly, introducing the set
N ∞ = ∪n∈N ∪σ∈An Mσ (F1 (Γ∞ ) ∩ F2 (Γ∞ )) , (18)
we have
µ (N ∞ ) = 0.
For all x ∈ Γ∞ \N ∞ , there exists a unique sequence σx ∈ A∞ such that x = limn→∞ Fσx (1) ◦
. . . Fσx (n) (O), where O is the origin. The mapping ℵ : Γ∞ \N ∞ → Γ0 defined by
X
ℵ(x) = (−1 + 2 (σx (i) − 1)2−i , 0) (19)
i≥1

is injective and such that the Lebesgue measure of ℵ(Γ∞ ) is 2, and we have the formula
Z Z
1
f dx = f ◦ ℵ dµ, ∀f ∈ L1 (Γ0 ). (20)
2 Γ0 ∞
Γ \N ∞

Formula (20) is first proven for the characteristic functions of the intervals (−1 + p2−n+1 , −1 +
(p + 1)2−n+1 ), n ∈ N, p ∈ {0, . . . , 2n − 1}, then for f ∈ L1 (Γ0 ) by density.
Consider the sequence of linear operators ℓn : H 1 (Ω) → L2µ ,
X  1 Z 
n
ℓ (v) = v dx 1Mσ (Γ∞ ) , (21)
|Γσ | Γσ
σ∈An

where |Γσ | is the one-dimensional Lebesgue measure of Γσ , see [2].

Proposition 3 The sequence (ℓn )n converges in L(H 1 (Ω), L2µ ) to an operator that we call ℓ∞ .
Proof. See [2].

10
4 Finer results on traces
4.1 The sub-critical case when a < a∗ .
In the case when a < a∗ , the theory of Jones (see [6, 7]), or of Vodopjanov et al [18] see also
[11] page 70, can be applied since Ω is a ǫ − δ domain (Lemma 1): it is possible to define a
continuous extension operator form H 1 (Ω) to H 1 (R2 ). Then, since Γ∞ is dH -set where dH =
log 2
− log a (Proposition 2), one can use the trace theorem by Jonsson and Wallin, see [7] Theorem 1
page 103, which states that
 ZZ 
1 2 d /2 ∞ 2 ∞ (v(x) − v(y))2
H (R )|Γ∞ = H H
(Γ ) = v ∈ Lµ (Γ ), dµx dµy < ∞ .
Γ∞ ×Γ∞ |x − y|2dH

Combining the two points above, a function in H 1 (Ω) has a trace in H dH /2 (Γ∞ ).

4.2 The critical case when a = a∗ .

We aim at finding more information on the operator ℓ∞ . We will successively answer the
questions: does the trace of u ∈ H 1 (Ω) belong to Lpµ (Γ∞ ), 1 ≤ p < ∞, to BM O(Γ∞ , µ), and
finally to the Sobolev spaces H s (Γ∞ )?

4.2.1 The trace functions belong to Lpµ , 1 ≤ p < ∞.

The main result is as follows:

Theorem 4 For any p ∈ [1, ∞) and u ∈ H 1 (Ω),

ℓ∞ (u) ∈ Lpµ ,

and there exists a constant C > 0 such that

kℓ∞ (u)kLpµ ≤ CkukH 1 (Ω) , ∀u ∈ H 1 (Ω).

Moreover, for all u ∈ H 1 (Ω), ℓn (u) converges to ℓ∞ (u) in Lpµ as n tends to ∞.

Proof. The proof relies on trace theorems obtained for a different and simpler geometry in [1].
For completeness, we recall the proof of these results in Appendix A. Let Fe1 and Fe2 be the
affine maps in R2
3 x1 x2 3 x1 x2
Fe1 (x) = (− + , 3 + ), Fe2 (x) = ( + , 3 + ). (22)
2 2 2 2 2 2

We introduce the points Pe1 = (−1, 0), Pe2 = (1, 0), Pe3 = Fe1 (Pe1 ) = (−2, 3), Pe4 = Fe2 (Pe2 ) = (2, 3),
Pe5 = Fe1 (Pe2 ) = (−1, 3) and Pe6 = Fe2 (Pe1 ) = (1, 3). Note that the last four points are aligned. We
introduce the trapezoidal domain
 
Ye 0 = Interior Conv(Pe1 , Pe2 , Pe3 , Pe4 )

e 0 = [Pe1 , Pe2 ] = Γ0 . As above, for σ ∈ An , we define

and Γ

fσ = Feσ(1) ◦ · · · ◦ Feσ(n) ,
M (23)

11
 3 f
P 3
f5
P f
P 6
f4
P

0
−3 −2 −1 0 1 2 3
f1
P f2
P

e
Figure 3: The domain Ω.

and the open domain   

∞
e e 0
Ω = Interior Y ∪ ∪ f e 0
∪ Mσ (Y ) , (24)
n=1 σ∈An

see Figure 3.
We also define the sets Γ eσ = M fσ (Γ e 0 ) and Γ
e N = ∪σ∈A Γe σ . Finally, we denote by Γ e∞ the self-
N
similar set associated to the similitudes Fe1 and Fe2 , which is the straight line segment between
the points (−3, 6) and (3, 6). It is important to realize that the self-similar measure defined in
e ∞ , Fe1 , Fe2 ) is 1/6 times the Lebesgue measure on Γ
the self-similar triplet (Γ e∞ , i.e. µ
e = 61 dx.
As above, we introduce the sequence of linear operators ℓen : H 1 (Ω) e → L2 (Γ e ∞ , dx) by
Z !
X 1
en
ℓ (v) = v(x) dx 1M e∞ ).
fσ (Γ (25)
e
|Γσ | eσ
Γ
σ∈An

A result similar to Proposition 3 holds: the sequence (ℓen )n converges in L(H 1 (Ω),
e L2 ) to an
e
µ
operator that we call ℓe∞ .
We partition the domains Y 0 and Ye 0 into six non-overlapping triangles which are numbered
as shown in Figure 4, and we call Q (respectively Q)e the interior node to Y 0 (respectively Ye 0 ).
We call T and Te the two sets of triangles. There exists a continuous, one to one and piecewise
linear function ψ from Y 0 onto Ye 0 , such that

• its restriction to the triangles in T is linear,

• ψ maps each triangle in T to the triangle in Te with the same index,

• the restriction of ψ to Γ0 is the identity,

• for i = 1, 2,
Fi (ψ −1 (x)) = ψ −1 (Fei (x)), e0.
∀x ∈ Γ

12
 4

5 3

5 4 3
6 Q 2 ψ
6 e 2
Q
1
1

Figure 4: The domains Y 0 and Ye 0 are partitioned into six triangles

e
This construction allows for the definition of the continuous linear operator Ψ : H 1 (Ω) → H 1 (Ω):

Ψ(u) = u
e
u fσ )−1
e = u|Mσ (Y 0 ) ◦ Mσ ◦ ψ −1 ◦ (M in fσ (Ye 0 ).
M

With ℓn defined in (21), we use the mapping Ψ for bounding

Z
p
n
kℓ (u)kLp = |ℓn (u)|p dµ
µ
Γ∞  p
Z X  Z  
 1 
=  σ|
u(x) dx 1Mσ (Γ )  dµ
∞
Γ ∞  |Γ Γσ 
σ∈An Z 
X 1  1 p
=  u
n  n 0
2 a |Γ | Γσ
σ∈An
X 1  1 Z p

=  f −1 
e ◦ Mσ ◦ ψ ◦ (Mσ )  .
u
n  n 0
2 a |Γ | Γσ
σ∈An

Therefore,
X 1  1 Z p

kℓn (u)kpLp = 
 u
e ◦ f
Mσ


µ 2n |Γ0 | Γ0
σ∈An  
X 1  1 Z p

=  u u)kp p e ∞ .
e = 1/6kℓen (e
2n  |Γ
fσ | Γfσ  L (Γ ,dx)
σ∈An

Similarly
kℓn (u) − ℓn+k (u)kpLp = 1/6kℓen (e u)kp p
u) − ℓen+k (e e ∞ ,dx) , ∀n, k ∈ N. (26)
µ L (Γ

Finally, in Appendix A, we prove the following result:

e ℓe∞ (e
e ∈ H 1 (Ω),
Proposition 4 For any p ∈ [1, ∞) and u e ∞ , dx), and there exists a
u) ∈ Lp (Γ
constant C > 0 such that

kℓe∞ (e
u)kLp (Γe ∞ ,dx) ≤ Cke
ukH 1 (Ω)
e , ∀e e
u ∈ H 1 (Ω).

e ℓen (e
e ∈ H 1 (Ω),
Moreover, for all u u) converges to ℓe∞ (e e ∞ , dx) as n tends to ∞.
u) in Lp (Γ

13
Combining (26) and Proposition 4, it is easy to prove that the sequence ℓn (u) is a Cauchy
sequence in Lpµ and to identify its limit as ℓ∞ (u) ∈ Lpµ .
Remark 8 Note that Theorem 4 holds for bounded domains of the type
  
∞
Ω = Interior Y 0 ∪ ∪ ∪ Mσ (Y 0 ) ,
n=1 σ∈An

where F1 and F2 are two similitudes, Mσ is defined by (1) and Y 0 is a polygonal domain, as
soon as the sets Mσ (Y 0 ), σ ∈ An , n > 0, do not overlap. The dilation ratii of F1 and F2 may
differ from each other.

4.2.2 The trace functions may not belong to BM O(Γ∞ , µ). A counterexample
We recall the definition of space of the functions with Bounded Mean Oscillation on Γ∞ , µ. This
space is referred to as BM O(Γ∞ , µ) or BM O for brevity.
Definition 2
( )
BM O(Γ∞ , µ) = φ ∈ L1 (Γ∞ , µ) s.t. sup h|φ − hφiB(P,r),µ |iB(P,r),µ < ∞ , (27)
r>0,P ∈Γ∞

where B(P, r) is the ball in Γ∞ centered in P and with radius r, and where, for a µ-measurable
subset X of Γ∞ , Z
1
hviX,µ = vdµ.
µ(X) X
It is easily seen that L∞ (Γ∞ , µ) is a subset of BM O.
Our goal is to give an example of a function u∗ ∈ H 1 (Ω) whose trace ℓ∞ (u∗ ) does not belong to
BM O. As a consequence, ℓ∞ (u∗ ) will not belong to L∞ (Γ∞ , µ).
• We choose u∗ |Y 0 = 0.
We also introduce a smooth nonnegative function φ1 taking the constant values 0 on Γ0
and F2 (Γ0 ), and 1 on F1 (Γ0 ). We also define the function φ2 by φ2 (x, y) = φ1 (−x, y). We
introduce Z
I= |∇φ1 |2 .
Y0
Remember that for any σ ∈ An , if φσi is the function defined on Mσ (Y 0 ) by φσi = φi ◦
(Mσ )−1 , we have Z
|∇φσi |2 = I.
Mσ (Y 0 )

• The function u∗ will be antisymmetric with respect to the axis x1 = 0, i.e. ∀(x, y) ∈ Ω,
(−x, y) ∈ Ω and u∗ (x, y) := −u∗ (−x, y). Therefore, we will focus on its restriction to
F1 (Ω).
• Let us choose some point P ∗ ∈ Ξ∞ , for instance
P ∗ = lim G1 ◦ F2 ◦ F2 ◦ (G2 )n (O)
n→∞
= lim F1 ◦ F2 ◦ F2 ◦ F2 ◦ F2 ◦ F1 ◦ F2 ◦ F1 ◦ · · · ◦ F2 ◦ F1 (O),
n→∞ | {z }
n times

where G1 and G2 are defined in (7). We denote by σ ∗ the sequence σ ∗ = (1, 2, 2, 2, 2, 1, 2, 1, . . . 2, 1 . . . ) ∈

A∞ and by σn∗ ∈ An the truncated sequence such that σn∗ (j) = σ ∗ (j) for any j ≤ n.

14
 • The
P last2 ingredient isP
an arbitrary sequence of positive real numbers α = (αi )i∈N such that
1
i∈N αi < +∞ and i∈N αi = +∞, (for instance take αi = i ).

• Assume that u∗ |Y n−1 has already been constructed for some n ≥ 1, and that u∗ |Γσ is
constant for all σ ∈ An . For σ ∈ An such that σ(1) = 1, we define u∗ |Y σ by the following:

u∗ |Y σ is constant and take the same value as u∗ |Γσ , if σ 6= σn∗ ,

u∗ |Y σ = u∗ |Γσ + αn φσ∗ (n) ◦ (Mσ )−1 , if σ = σn∗ ,

where in the last line we have identified u∗ |Γσ with its constant value. We have defined u∗
in Y n ∩ F1 (Ω), and we fix u∗ (x, y) for (x, y) ∈ Y n ∩ F2 (Ω) by u∗ (x, y) = −u∗ (−x, y). It is
easily seen that u∗ |Γσ is constant for all σ ∈ An+1 , so the recursion can be continued.

Theorem 5 The function u∗ constructed above satisfies u∗ ∈ H 1 (Ω) and ℓ∞ (u∗ ) ∈ / BM O(Γ∞ , µ).
P P
Proof. By using the bound ( ni=1 αi )2 ≤ n ni=1 α2i , it is possible to verify that
N −1 n
!
X X
∗ 2 2n 2
ku kL2 (Y N ) ≤ C a n αi ,
n=1 i=1

and that the quantity in the right hand side is bounded independently of N .
Moreover,
N
X −1
∗ 2
k∇u kL2 (Y N ) = 2I α2n ,
n=1

which is bounded independently of N . Therefore ∈ H 1 (Ω). u∗

We want to prove that ℓ∞ (u∗ ) ∈
/ BM O(Γ∞ , µ). For any positive number r, we know that

hℓn (u∗ )iB(P ∗ ,r),µ = 0, ∀n ∈ N,

because u∗ is antisymmetric with respect to the axis x1 = 0 (note that the ball B(P ∗ , r) is
symmetric w.r.t. the axis x1 = 0). Therefore, we also have

hℓ∞ (u∗ )iB(P ∗ ,r),µ = 0.

Moreover, from the construction of u∗ and since the numbers αn are positive, we have that

|ℓn (u∗ )| ≤ |ℓ∞ (u∗ )|, ∀n ∈ N.

Therefore

h|ℓ∞ (u∗ ) − hℓ∞ (u∗ )iB(P ∗ ,r),µ |iB(P ∗ ,r),µ ≥ h|ℓn (u∗ )|iB(P ∗ ,r),µ , ∀n ∈ N.

Symmetry considerations lead to

Z
∗ 1
n
h|ℓ (u )|iB(P ∗ ,r),µ = ℓn (u∗ )dµ.
µ(B(P , r) ∩ F1 (Γ∞ ))
∗
B(P ∗ ,r)∩F 1 (Γ∞ )

Let us now make the important observation that for any n ∈ N, there exists a positive number
rn such that B(P ∗ , rn ) ∩ F1 (Γ∞ ) ⊂ Mσn∗ (Γ∞ ). On the other hand, for any x ∈ Mσn∗ (Γ∞ ),

X  1 Z 
1
Z n
X
n ∗ ∗ ∗
ℓ (u )(x) = u (x) dx 1 ∞
Mσ (Γ ) (x) = ∗ u (x) dx = αi .
|Γσ | Γσ |Γσn | Γσn∗
σ∈An i=1

15
This implies that
n
X
n ∗
h|ℓ (u )|iB(P ∗ ,rn ),µ = αi ,
i=1

thus
n
X
∞ ∗ ∞ ∗
h|ℓ (u ) − hℓ (u )iB(P ∗ ,rn ),µ |iB(P ∗ ,rn ),µ ≥ αi . (28)
i=1

The right hand side of (28) tends to +∞ as n → ∞. We have proved that ℓ∞ (u∗ ) ∈
/ BM O(Γ∞ , µ).

Remark 9 Note that a similar counterexample can be found for bounded domains of the type
  
∞
0
Ω = Interior Y ∪ ∪ ∪ Mσ (Y ) 0 ,
n=1 σ∈An

where F1 and F2 are two similitudes, Mσ is defined by (1) and Y 0 is a polygonal domain, as
soon as

• The sets Mσ (Y 0 ), σ ∈ An , n > 0, do not overlap.

• The self-invariant set Γ∞ is such that F1 (Γ∞ ) ∩ F2 (Γ∞ ) 6= ∅, (in our setting this occurs
for a = a∗ ).

• The domain Ω has a symmetry axis which contains F1 (Γ∞ ) ∩ F2 (Γ∞ ).

For instance, the same counterexample can be constructed for the geometry in Figure 3.
Moreover it seems that similar counterexamples can be found under relaxed geometrical assump-
tions, (in particular, the symmetry assumption does not seem crucial).

4.2.3 Trace theorems in H s (Γ∞ ).

For brevity, let us denote dH the Hausdorff dimension of Γ∞ :

dH = − log 2/ log a∗ .

We define for 0 < s < 1, the space H s (Γ∞ ) as

 Z Z 
|v(x) − v(y)|2
H s (Γ∞ ) = v ∈ L2µ (Γ∞ ) s.t. dµ(x)dµ(y) < +∞ , (29)
Γ∞ Γ∞ d(x, y)dH +2s

where d(x, y) is the Euclidean distance between the two points x and y. Note that the space
H s (Γ∞ ) correspond to the space Bs2,2 in [7], see page 103.
The space H s (Γ∞ ) is a Hilbert space with the norm
q
k · kH s (Γ∞ ) = k · k2L2 + | · |2H s (Γ∞ ) ,
µ

where Z Z 1/2
|v(x) − v(y)|2
|v|H s (Γ∞ ) = dµ(x)dµ(y) . (30)
Γ∞ Γ∞ d(x, y)dH +2s
Our goal is to prove the following

16
Theorem 6 For any s < dH /4,

ℓ∞ (u) ∈ H s (Γ∞ ), ∀u ∈ H 1 (Ω),

and there exists a positive constant C such that

kℓ∞ (u)kH s (Γ∞ ) ≤ CkukH 1 (Ω) , ∀u ∈ H 1 (Ω). (31)

We first state two technical lemmas.

The first lemma is an easy bound (invariant by scaling) on the difference of the mean values of
ℓ∞ (u) on F1 (Γ∞ ) and F2 (Γ∞ ) by the H 1 (Ω) semi-norm of u.
Lemma 7 There exists a positive constant C such that

|hℓ∞ (u)iF1 (Γ∞ ),µ − hℓ∞ (u)iF2 (Γ∞ ),µ | ≤ Ck∇ukL2 (Ω) , ∀u ∈ H 1 (Ω). (32)

Remark 10 The bound (32) implies that for all positive integer p and for all σ ∈ Ap ,

|hℓ∞ (u)iF1 (Mσ (Γ∞ )),µ − hℓ∞ (u)iF2 (Mσ (Γ∞ )),µ | ≤ Ck∇ukL2 (Mσ (Ω)) , ∀u ∈ H 1 (Ω),

where C is exactly the same constant as in (32) (i.e. C does not depend of σ and p).

The second lemma is an explicit bound on the H s -semi-norm of the Haar mother wavelet g0 on
Γ∞ ,
g0 = 1F1 (Γ∞ ) − 1F2 (Γ∞ ) . (33)

Lemma 8 For all real number s such that 0 < s < d4H ,
Z Z
0 2 1
|g |H s (Γ∞ ) = 2 dH +2s
dµ(x)dµ(y) < ∞. (34)
F1 (Γ∞ ) F2 (Γ∞ ) d(x, y)

dH
Moreover, if s > 4 , then
Z Z
1
|g0 |2H s (Γ∞ ) =2 dµ(x)dµ(y) = ∞.
F1 (Γ∞ ) F2 (Γ∞ ) d(x, y)dH +2s

Proof. Since it is rather technical, the proof is postponed to the Appendix B.

It is easy to construct a function u in H 1 (Ω) whose trace is g0 . Therefore, as corollary of Lemma
8, we have the following
Proposition 5 There exists a function u ∈ H 1 (Ω) such that

ℓ∞ (u) ∈
/ H s (Γ∞ ), ∀s > dH /4.

Proof of Theorem 6 For s, 0 < s < dH /4, we set τ = 2s and we have to prove that there
exists a constant C such that for all u ∈ H 1 (Ω),
Z Z
|ℓ∞ (u)(x) − ℓ∞ (u)(y)|2
dH +τ
dµ(x)dµ(y) ≤ Ckuk2H 1 (Ω) . (35)
Γ∞ Γ∞ d(x, y)

Let us decompose the integral in (35) into three terms:

Z Z
|ℓ∞ (u)(x) − ℓ∞ (u)(y)|2
dµ(x)dµ(y) = I + II + 2III, (36)
Γ∞ Γ∞ d(x, y)dH +τ

17
where
Z Z
|ℓ∞ (u)(x) − ℓ∞ (u)(y)|2
I = dµ(x)dµ(y),
F1 (Γ∞ ) F1 (Γ∞ ) d(x, y)dH +τ
Z Z
|ℓ∞ (u)(x) − ℓ∞ (u)(y)|2
II = dµ(x)dµ(y),
F2 (Γ∞ ) F2 (Γ∞ ) d(x, y)dH +τ
Z Z
|ℓ∞ (u)(x) − ℓ∞ (u)(y)|2
III = dµ(x)dµ(y).
F1 (Γ∞ ) F2 (Γ∞ ) d(x, y)dH +τ
We first estimate the term III: from the identity
 ∞ 
ℓ (u)(x) − hℓ∞ (u)iF1 (Γ∞ ),µ
ℓ∞ (u)(x) − ℓ∞ (u)(y) =  +hℓ∞ (u)iF1 (Γ∞ ),µ − hℓ∞ (u)iF2 (Γ∞ ),µ  ,
+hℓ∞ (u)iF2 (Γ∞ ),µ − ℓ∞ (u)(y)
we deduce that
III ≤ 3(III1 + III2 + III3 ),
Z Z
|ℓ∞ (u)(x) − hℓ∞ (u)iF1 (Γ∞ ),µ |2
III1 = dµ(x)dµ(y),
F1 (Γ∞ ) F2 (Γ∞ ) d(x, y)dH +τ
Z Z
|hℓ∞ (u)iF1 (Γ∞ ),µ − hℓ∞ (u)iF2 (Γ∞ ),µ |2
III2 = dµ(x)dµ(y),
F1 (Γ∞ ) F2 (Γ∞ ) d(x, y)dH +τ
Z Z
|hℓ∞ (u)iF2 (Γ∞ ),µ − ℓ∞ (u)(y)|2
III3 = dµ(x)dµ(y).
F1 (Γ∞ ) F2 (Γ∞ ) d(x, y)dH +τ
We can bound III2 thanks to Lemmas 7 and 8, because
Z Z
∞ ∞ 2 1
III2 = |hℓ (u)iF1 (Γ∞ ),µ − hℓ (u)iF2 (Γ∞ ),µ | dµ(x)dµ(y)
F1 (Γ∞ ) F2 (Γ∞ ) d(x, y)dH +τ
≤ Ckuk2H 1 (Ω) .
Bounds on III1 and III3 are obtained in the same way, so let us only consider III1 : we fix
η > 0 such that (dH + τ )(1 + η) < 3dH /2, which also reads
(dH + τ )(1 + η) − dH < dH /2. (37)
Since from Theorem 4, ℓ∞ (u) ∈ Lpµ ,
for all p ∈ [2, +∞), we have the following Hölder inequality
 ! η 
Z Z 2(1+η)
1+η
 ∞ ∞ 
 |ℓ (u)(x) − hℓ (u)iF1 (Γ∞ ),µ | η dµ(x)dµ(y) 
 F1 (Γ∞ ) F2 (Γ∞ ) 
III1 ≤  Z Z ! 1 ,
 1 1+η 
 
(d +τ )(1+η)
dµ(x)dµ(y)
F1 (Γ∞ ) F2 (Γ∞ ) d(x, y)) H
! 1
1+η
1 0 2
because the second factor is exactly 2 |g | (dH +τ )(1+η)−dH ∞ < ∞ from (37) and Lemma
H 2 (Γ )
8. Therefore, for all u ∈ H 1 (Ω),
Z Z ! η
1+η
2(1+η)
∞ ∞
III1 . |ℓ (u)(x) − hℓ (u)iF1 (Γ∞ ),µ | η dµ(x)dµ(y)
F1 (Γ∞ ) F2 (Γ∞ )
Z ! η
1+η
2(1+η)
. |ℓ∞ (u)(x) − hℓ∞ (u)iF1 (Γ∞ ),µ | η dµ(x)
F1 (Γ∞ )

18
 2(1+η)
and from Theorem 4 (in the particular case p = η ),
Z
III1 . |∇u(x)|2 dx.
F1 (Ω)

Finally, we have obtained that there exists a positive constant C such that for all u ∈ H 1 (Ω),
Z
III ≤ C |∇u(x)|2 dx.
Ω

We are left with finding an estimate for I and II. Since the argument is the same for the two
terms, we focus on I. Using the change of variables x = F1 (x′ ) and y = F1 (y ′ ) and the definition
of dH , we obtain that
Z Z
1 |ℓ∞ (u ◦ F1 )(x′ ) − ℓ∞ (u ◦ F1 )(y ′ )|2
I= dµ(x′ )dµ(y ′ )
4 Γ∞ Γ∞ (a · d(x′ , y ′ ))dH +τ
Z Z
τ
−1 |ℓ∞ (u ◦ F1 )(x′ ) − ℓ∞ (u ◦ F1 )(y ′ )|2
=2 dH
dµ(x′ )dµ(y ′ ).
Γ∞ Γ∞ (d(x′ , y ′ ))dH +τ
This allows for a recursive argument, because we can decompose the last integral into the sum
of three terms as in (36). We obtain that
 Z 
2
C |∇u(x)| dx
 F1 (Ω)

τ
−1  
I≤2 dH
 X2 Z Z ∞ (u ◦ F ◦ F )(x′′ ) − ℓ∞ (u ◦ F ◦ F )(y ′′ )|2 ,
 τ
−1 |ℓ 1 i 1 i ′′ ′′ 
+2 dH ′′ , y ′′ ))dH +τ
dµ(x )dµ(y )
Γ∞ Γ∞ (d(x
i=1

and this implies that

Z Z
|ℓ∞ (u)(x) − ℓ∞ (u)(y)|2
dµ(x)dµ(y)
Γ∞ Γ∞ d(x, y)dH +τ
 Z Z Z ! 
τ τ
2 −1 2 −1 2
 C |∇u(x)| dx + 2 dH |∇u(x)| dx + 2 dH |∇u(x)| dx  (38)
 Ω F (Ω) F2 (Ω) 
≤  Z Z 1 .
τ X |ℓ∞ (u ◦ Mσ )(x) − ℓ∞ (u ◦ Mσ )(y)|2 
 +22( dH −1) dµ(x)dµ(y) 
Γ∞ Γ∞ d(x, y)dH +τ
σ∈A2

Note that the last term in the right hand side of (38) is exactly
X Z Z
|ℓ∞ (u)(x) − ℓ∞ (u)(y)|2
dµ(x)dµ(y).
Mσ (Γ∞ ) Mσ (Γ∞ ) d(x, y)dH +τ
σ∈A2

Carrying the recursion, we find that

Z X Z
|ℓ∞ (u)(x) − ℓ∞ (u)(y)|2 |ℓ∞ (u)(x) − ℓ∞ (u)(y)|2
dµ(x)dµ(y) − dµ(x)dµ(y)
Γ∞ ×Γ∞ d(x, y)dH +τ ∞ ∞ d(x, y)dH +τ
σ∈An Mσ (Γ )×Mσ (Γ )
Z
|ℓ∞ (u)(x) − ℓ∞ (u)(y)|2
= 1Γ∞ ×Γ∞ \∪σ∈An Mσ (Γ∞ )×Mσ (Γ∞ ) dµ(x)dµ(y)
Γ∞ ×Γ∞ d(x, y)dH +τ
X m( τ −1) X Z
≤C 2 dH |∇u(x)|2 dx.
m<n σ∈Am Mσ (Ω)

19
From this and Fatou lemma, we obtain that
Z Z X m( τ −1) X Z
|ℓ∞ (u)(x) − ℓ∞ (u)(y)|2
dµ(x)dµ(y) ≤ C 2 dH |∇u(x)|2 dx
Γ∞ Γ∞ d(x, y)dH +τ Mσ (Ω)
m∈N σ∈Am
X X k( τ −1) Z
=C 2 dH |∇u(x)|2 dx
m∈N 0≤k≤m Ωm \Ωm−1
!Z
1
≤C ( dτ −1)
|∇u(x)|2 dx,
1−2 H Ω

dH
because τ < 2 . This concludes the proof.

Remark 11 It is easy to extend Theorem 6 and Proposition 5 to the Sobolev space W 1,p (Ω),
2 ≤ p < ∞:
For any s < 1 − 2p + d2p
H
,
u|Γ∞ ∈ W s,p(Γ∞ ), ∀u ∈ W 1,p (Ω),
and there exists a positive constant C such that

ku|Γ∞ kW s,p (Γ∞ ) ≤ CkukW 1,p (Ω) , ∀u ∈ W 1,p (Ω).

Moreover, there exists a function u ∈ W 1,p (Ω) such that

2 dH
/ W s,p (Γ∞ ),
u|Γ∞ ∈ ∀s > 1 − + .
p 2p

A Proof of Proposition 4
The proof of Proposition 4 uses arguments that have already been written in [1] for a slightly
different construction with the same affine maps Fei (see (22)) but with a different reference
domain Ye 0 .
The first step is an extension result:

Lemma 9 Let Ω b be the trapezoidal domain whose vertices are (−1, 0), (1, 0), (−3, 6) and (3, 6).
e
With Ω defined in (24), there exists an extension operator J bounded from W 1,q (Ω)e to W 1,q (Ω),
b
for all q, 1 ≤ q < 2.

Proof. This is Lemma 11 in [1].

As a consequence we have a density result (Theorem 4 in [1]):

Proposition 6 The space C ∞ (Ω) e is dense in W 1,q (Ω),

e for all q, 1 ≤ q < 2. More precisely, for
e there exists a sequence (φej )j∈N , φej ∈ C ∞ (Ω)
e ∈ W 1,q (Ω),
q, 1 ≤ q < 2, for any u b such that

u − φej kW 1,q (Ω)

lim ke e = 0.
n→0

b
Remark 12 Observe that for any function φ ∈ C 1 (Ω),

ℓe∞ (φ1Ωe ) = φ1Γe ∞ . (39)

20
 fσ defined in (23),
Indeed, with M
 2
X Z  1
Z

kℓen (φ1Ωe ) − φ1Γe ∞ k2L2 (Γe ∞ ,dx) = φ(x) − φ dx ≤ C2−2n k∇φk2L∞ (Ω)
 e σ Γeσ b ,
σ∈An
fσ (Γ
M e∞ ) Γ
where the last estimate comes from a Taylor expansion of φ. This yields that
lim kℓen (φ1Ωe ) − φ1Γe ∞ k2L2 (Γe ∞ ,dx) = 0,
n→∞

and (39).
Similarly for any p ∈ [1, ∞[,
kℓen (φ1Ωe ) − φ1Γe ∞ kp p e ∞ ,dx) ≤ C2−pn k∇φkp ∞ b . (40)
L (Γ L (Ω)

b
Therefore for all p ∈ [1, ∞) and φ ∈ C 1 (Ω),
lim kℓen (φ1Ωe ) − φ1Γe ∞ kp p e ∞ ,dx) = 0.
n→∞ L (Γ

Lemma 10 For any p ∈ [1, ∞) there exists q ∈ [1, 2) and a constant C independent of n such
that
kℓen (φ)kLp (Γb ∞ ,dx) ≤ CkφkW 1,q (Ω)
e ,
e
∀φ ∈ W 1,q (Ω). (41)

Proof. Call Γ b n the smallest straight line segment containing Γ

e n . For all p ∈ [1, ∞), there exists
q ∈ (1, 2) and a positive constant c independent of n such that
kψkLp (Γe n ,dx) ≤ ckψkW 1,q (Ω)
b ,
b
∀ψ ∈ W 1,q (Ω). (42)
Then, using the extension operator J mentioned in Lemma 9, we see that for any p ∈ [1, ∞)
there exists q ∈ [1, 2) and a constant C independent of n such that
Z Z
en p p
kℓ (φ)k p e ∞ ≤ 3 |φ(x)| dx = 3 |J (φ)(x)|p dx
L (Γ ,dx)
ZΓe n e
Γ n

≤ 3 |J (φ)(x)|p dx
bn
Γ
≤ 3ckJ (φ)kp b ≤ Ckφkp e .
W 1,q (Ω) W 1,q (Ω)

where c is the constant appearing in (42).

We are now ready to prove Proposition 4.

Proof of Proposition 4 Consider u e fix p ∈ [1, ∞) and take q = q(p) ∈ [1, 2) as in

e ∈ H 1 (Ω),
b such that
Lemma 10. Thanks to Proposition 6, for any ǫ > 0, there exists a function φ ∈ C ∞ (Ω)
ke
u − φkW 1,q (Ω)
e ≤ ǫ. (43)
Then for all nonnegative integers n and k,
kℓen (e
u) − ℓen+k (e
u)kLp (Γe ∞ ,dx)
≤kℓen (e
u − φ)kLp (Γe ∞ ,dx) + kℓen (φ) − ℓen+k (φ)kLp (Γe ∞ ,dx) + kℓen+k (φ − u
e)kLp (Γe ∞ ,dx) .
Using (43) and (41), we obtain that
kℓen (e
u) − ℓen+k (e
u)kLp (Γe ∞ ,dx) ≤ kℓen (φ) − ℓen+k (φ)kLp (Γe ∞ ,dx) + 2Cε. (44)

On the other hand, (40) implies that (ℓen (φ))n∈N is a Cauchy sequence in Lp (Γ e ∞ , dx). From
en
this and (44), we see that (ℓ (e p e ∞
u))n∈N is a Cauchy sequence in L (Γ , dx). Finally, it is easy to
e ∞ , dx)-limit and we obtain the desired result using (41).
identify the Lp (Γ

21
B Proof of Lemma 8.
B.1 Preliminary definitions and lemmas
Hereafter, in order to make the notation simpler, we write a instead of a∗ .
Take N = 2(n + 2), and consider

ΞN,1 = ∪σ∈An Ξσ,1 ,

ΞN,2 = ∪σ∈An Ξσ,2 ,

where

Ξσ,1 = G1 ◦ F2 ◦ F2 ◦ Gσ(1) ◦ · · · ◦ Gσ(n) (Γ0 ), for σ ∈ An ,

Ξσ,2 = Gc1 ◦ F2c ◦ F2c ◦ Gcσ(1) ◦ ··· ◦ Gcσ(n) (Γ0 ), for σ ∈ An ,

where G1 and G2 are defined in (7) and Gc1 and Gc2 are defined in (8). From these definitions,
we see that ΞN,i , i = 1, 2, is made of #(An ) = 2n non-overlapping straight lines and that
ΞN,i ⊂ F i (ΓN −1 ) ⊂ ΓN . The one-dimensional Lebesgue measure of Ξσ,i , for σ ∈ An is
 σ,i 
Ξ  = a2n+4 |Γ0 |.

Recalling the definition of prox(σ) given in (12) for σ ∈ AN , we introduce a mapping ι from
AN to N which allows for sorting the σ ∈ AN such that prox(σ) = k, for any fixed integer k,
k = 1, . . . , n + 1.

Definition 3 Take N = 2(n + 2). For σ ∈ AN , such that σ(1, 2, 3, 4) = (1, 2, 2, 2), let us define
the integer ι(σ) by

 prox(σ)

 X

 ι(σ) = 2prox(σ)−q (σ(2q + 3) − 1), if 1 ≤ prox(σ) ≤ n,

q=1 (45)
 Xn


 ι(σ) =
 2n−q (σ(2q + 3) − 1), if prox(σ) = n + 1,

q=1

see Figure 5.
For σ ∈ AN , such that σ(1, 2, 3, 4) = (2, 1, 1, 1), we choose ι(σ) = ι(σ c ), where σ c (i) = 3 − σ(i),
for all i = 1, . . . , N .

Properties of prox and ι

1. If 1 ≤ prox(σ) ≤ n, then ι(σ) ∈ {0, . . . , 2prox(σ) − 1}, and if prox(σ) = n + 1 then

ι(σ) ∈ {0, . . . , 2n − 1}.

2. For σ ∈ AN , there exists 22(n−min(prox(σ),n)) maps η ∈ AN such that

η(1, 2, 3, 4) = σ(1, 2, 3, 4),

prox(η) = prox(σ),
ι(η) = ι(σ).

These maps satisfy η(j) = σ(j) for all j ∈ {1, . . . , 4 + 2 min(prox(σ), n)}.

22
 prox = 3
prox = 1, ι = 0
prox = 4

prox = 2, ι = 0

ι=0
8
2a ι=1
2a6 ι=2
ι=3

prox = 2, ι = 1
2a4
prox = 2, ι = 2
ι=4
ι=5
ι=6

ι=7

prox = 2, ι = 3

prox = 1, ι = 1

Figure 5: Zoom on F1 ◦ F2 ◦ F2 ◦ F2 (Y 5 ). The values of prox(σ) and of ι(σ) for n = 3, i.e.

N = 10.

23
 3. For σ ∈ AN = A2n+4 , and m < n, the mapping σ ′ ∈ A2m+4 defined by σ ′ (i) = σ(i),
i = 1, . . . , 2m + 4 is such that

prox(σ ′ ) = min(prox(σ), m + 1),

 
′ ι(σ)
ι(σ ) = ,
2min(prox(σ),n)−min(prox(σ′ ),m)

where hereafter, [u] stands for the integer part of u.

A consequence of Lemma 3 is the following:

Lemma 11 For all N = 2n + 4, for all σ 6= η ∈ AN such that σ(1, 2, 3, 4) = η(1, 2, 3, 4) =

6 ι(η),
(1, 2, 2, 2), prox(η) = prox(σ) = k, and ι(σ) =
1
inf |x2 − y2 | ≥ (|ι(σ) − ι(η)| − 1)a2k+2 . (46)
x∈Γ σ,y∈Γη 2

Proof. For a given k, 0 ≤ k ≤ n + 1, call Sk the set of all the maps µ ∈ AN such that
µ(1, 2, 3, 4) = (1, 2, 2, 2) and prox(µ) = k. If k < n + 1, the lines Γµ , µ ∈ Sk are either horizontal
or vertical. If ι(µ) < ι(ν), any x ∈ Γµ and x′ ∈ Γν satisfy x′2 < x2 . We introduce the notations

x2,min (i) = min x2 , and x2,max (i) = max x2 .

x∈Γµ , µ∈Sk , ι(µ)=i x∈Γµ , µ∈Sk , ι(µ)=i

Therefore, assuming ι(η) > ι(σ),

ι(η)−1
X
inf
σ η
|x2 − y2 | ≥ (x2,min (i) − x2,max (i + 1)) . (47)
x∈Γ ,y∈Γ
i=ι(σ)

By an elementary geometrical argument, (cf Figure 5), we see that x2,min (2j) − x2,max (2j + 1) >
2a2k+2 . This and (47) yield (46).
Similarly, if k = n + 1, then the lines Γµ , µ ∈ Sk are vertical and aligned, and if µ, ν ∈ Sk satisfy
ι(µ) < ι(ν), minx∈Γµ x2 − maxx∈Γν x2 > 0. Therefore, (47) is still valid. By an elementary
geometrical argument, we see that x2,min (2j) − x2,max (2j + 1) > 2a2(n+1)+2 . This and (47) yield
(46).

Remark 13 The estimate (46) is very far from optimal, but for what follows, we do not need
a better one.

Lemma 12 For all N = 2n + 4, for all σ, η ∈ AN such that σ(1, 2, 3, 4) = η(1, 2, 3, 4) =

(1, 2, 2, 2) and prox(η) < prox(σ),
1
inf |x2 − y2 | ≥ |ι(σ ′ ) − ι(η)|a2prox(η)+2 ,
x∈Γ σ,y∈Γη 2

where σ ′ ∈ A2m+4 , m = prox(σ) is defined by σ ′ (i) = σ(i), i = 1, . . . , 2m + 4 and

 
ι(σ)
prox(σ ′ ) = prox(η) + 1 and ι(σ ′ ) = min(prox(σ),n)−prox(η) .
2

24
Proof. The first elementary observation is that if ι(η) = ι(σ ′ ) ± 1, then

inf |x2 − y2 | > a2prox(η)+2 , (48)

x∈Γσ ,y∈Γη

see Figure 5.
The second observation is that if for example ι(η) > ι(σ ′ ) + 1, then calling µ a map in AN such
that prox(µ) = prox(η) and ι(µ) = ι(σ ′ ) + 1, we have

inf |x2 − y2 | ≥ inf |x2 − y2 | + inf |x2 − y2 |

x∈Γσ ,y∈Γη x∈Γσ ,y∈Γµ x∈Γµ ,y∈Γη
1

≥ a2prox(η)+2 + ι(η) − ι(σ ′ ) − 2 a2prox(η)+2
2
1

= ι(η) − ι(σ ′ ) a2prox(η)+2 ,
2
where we have used (48) and (46).
The argument is the same if ι(η) < ι(σ ′ ) − 1.

Corollary 4 There exists a constant c such that, for all N = 2n + 4, for all σ, η ∈ AN such
that σ(1, 2, 3, 4) = (1, 2, 2, 2), η(1, 2, 3, 4) = (2, 1, 1, 1) and prox(σ) > prox(η), we have
s   2
σ η 2prox(σ) 2prox(η) 2
ι(σ)
d(Γ , Γ ) ≥ c (a +a ) + ι(η) − min(prox(σ),n)−prox(η) a4prox(η) , (49)
2

Proof. Take x ∈ Γσ and y ∈ Γη . We know from Lemma 4 that |x1 | ≥ δ′ a2prox(σ) and |y1 | ≥
δ′ a2prox(η) . Since x1 < 0 < y1 , we have y1 − x1 ≥ δ′ (a2prox(σ) + a2prox(η) ).
The lower estimate for |y2 − x2 | comes from Lemma 12.

B.2 Proof of Lemma 8

The proof of Lemma 8 will be decomposed into several steps:

Assertion 1
Z Z
dH −(4n+8) 1
If τ > 2 , then lim (2a) dx dy = ∞. (50)
n→∞ ΞN,1 ΞN,2 d(x, y)dH +τ

Assertion 2
Z Z
dH 1
If τ > 2 , then dµ(x)dµ(y) = ∞. (51)
F1 (Γ∞ ) F2 (Γ∞ ) d(x, y)dH +τ

Assertion 3
Z Z
dH −(4n+8) 1
If τ < 2 , then lim (2a) dx dy = 0. (52)
n→∞ ΞN,1 ΞN,2 d(x, y)dH +τ

Assertion 4
X X Z Z
dH −4n dx dy
If τ < 2 , then (2a) ≤ C, (53)
Γσ Γ σ′ d(x, y)dH +τ
σ∈AN ,σ(1)=1 σ′ ∈AN ,σ′ (1)=2

where C > 0 is independent of n.

25
Assertion 5
Z Z
dH 1
If τ < 2 , then dµ(x)dµ(y) < ∞. (54)
F1 (Γ∞ ) F2 (Γ∞ ) d(x, y)dH +τ

In order to support assertions 1 and 4, we plot in the top of Figure 6

X X Z Z
−2N dx dy
(2a) dH +t
′ ′ Γσ Γσ ′ d(xσ , xσ′ )
σ∈AN ,σ(1)=1 σ ∈AN ,σ (1)=2

versus N , for a = 0.593465 ≈ a∗ where xσ is the midpoint of Γσ , for t = 0.4dH , t = 0.5dH and
t = 0.6dH . We see that for large values of N the function is convex for t = 0.6dH and concave
if t = 0.4dH . On the bottom part of Figure 6, we plot the same quantity for a = 0.58673 < a∗
so dH = 1.3 instead of 1.3284371, and t = dH , and we see that the function tends to a constant
value as N tends to infinity. This supports the theoretical results obtained in § 4.1 in the sub-
critical case a < a∗ .

0.35
"t=0.6d"
"t=0.5d"
0.3 "t=0.4d"

0.25

0.2

0.15

0.1

0.05

0
0 2 4 6 8 10 12 14 16

0.18
"t=dprime"
0.16

0.14

0.12

0.1

0.08

0.06

0.04

0.02
0 2 4 6 8 10 12 14 16

P P R R dx dy
Figure 6: The value of (2a)−2N σ∈AN ,σ(1)=1 σ′ ∈AN ,σ′ (1)=2 Γσ Γσ ′ d(xσ ,xσ ′ )dH +t versus N for
dH = 1.3284371, t = 0.4dH , t = 0.5dH and t = 0.6dH (top) and dH = 1.3, t = dH (bottom).

Proof of Assertions 1 and 3. As an easy consequence of Lemma 3, there exist two positive

26
numbers c and C, such that for all n > 0 and for all σ ∈ An , ∀x ∈ Ξσ,1 ,

ca2n ≤ d(x, y) ≤ Ca2n , ∀y ∈ Ξσ,2 , (55)


′ σ(i) = σ ′ (i),
∀i < n − k,
ca2(n−k) ≤ d(x, y) ≤ Ca2(n−k) , ∀y ∈ Ξσ ,2 , σ ′ ∈ An s.t. (56)
σ(n − k) 6= σ ′ (n − k).

Observe that if σ ∈ An , for all k ∈ {0, . . . , n − 1}, there exists 2k maps σ ′ ∈ An such that
σ(i) = σ ′ (i) for i < n − k, and σ(n − k) 6= σ ′ (n − k). From (56), we deduce that
Z Z X n−1 n−1
a−4n−8 X X dx dy X
k 1 X 2n+k
dH +τ
≥ 2 =
4
σ∈A σ′ ∈A Ξσ,1 Ξσ ′ ,2 d(x, y) σ∈A k=0
(Ca2(n−k) )dH +τ k=0
(Ca2(n−k) )dH +τ
n n n
n−1
X
−dH −τ 2(n−k)( dτ +1)
=C 2n+k 2 H

k=0
n−1
X
n(3+ d2τ ) −k(1+ d2τ )
= C −dH −τ 2 H 2 H ,
k=0

dH
where we have used the fact that adH = 1/2. From the bound above, we see that if τ > 2 ,
then there exists a positive number ν such that 3 + d2τH ≥ 4 + ν and
X X Z Z
dx dy
−4n−8
a & 2(4+ν)n ,
σ,1 σ ′ ,2 d(x, y)dH +τ
′ Ξ
σ∈An σ ∈AnΞ

which yields (50).

On the other hand, from the lower bounds in (55) and (56), we see in a similar way that
Z Z n−1
a−4n−8 X X dx dy 2τ
−dH −τ n(3+ dH )
2τ
−dH −τ n(3+ dH )
X −k(1+ 2τ )
≤ c 2 + c 2 2 dH
.
4 σ,1 σ ′ ,2 d(x, y)dH +τ
′ Ξ
σ∈An σ ∈AnΞ k=0

dH
From this, we see that if τ < 2 ,
there exists some positive number ν such that
X X Z Z
dx dy
a−4n−8 . 2(4−ν)n ,
σ,1 σ ′ ,2 d(x, y)dH +τ
′ Ξ
σ∈An σ ∈An Ξ

which yields (52).

Proof of Assertion 2. We need the notations

Γ∞,σ,1 = G1 ◦ F2 ◦ F2 ◦ Gσ(1) ◦ · · · ◦ Gσ(n) (Γ∞ \N ∞ ), for σ ∈ An , (57)

Γ∞,σ,2 = Gc1 ◦ F2c ◦ F2c ◦ Gcσ(1) ◦ · · · ◦ Gcσ(n) (Γ∞ \N ∞ ), for σ ∈ An , (58)
∞,σ,1
Γ∞,N,1 = ∪σ∈An Γ , and Γ∞,N,2 = ∪σ∈An Γ∞,σ,2 . (59)

We define the mapping ℵN,1 : Γ∞,N,1 → ΞN,1 by

ℵN,1 (x) = G1 ◦F2 ◦F2 ◦Gσ(1) ◦· · ·◦Gσ(n) (ℵ(y)), if x = G1 ◦F2 ◦F2 ◦Gσ(1) ◦· · ·◦Gσ(n) (y), y ∈ Γ∞ ,

where the function ℵ has been defined in (19). We define in a similar way the mapping ℵN,2 :
Γ∞,N,2 → ΞN,2 . A consequence of (20) is that
Z Z
1
(2a)−2n−4 f (x)dx = f ◦ ℵN,i dµ, ∀f ∈ L1 (ΞN,i ), i = 1, 2. (60)
2 ΞN,i Γ∞,N,i

27
Simple geometric considerations yield that there exists a constant C such that

d(x, ℵN,i (x)) ≤ Ca2n+4 ∀x ∈ Γ∞,N,i , i = 1, 2,

and that

d(x, y) ≤ d ℵN,1 (x), ℵN,2 (y) + Ca2n+4 ∀x ∈ Γ∞,N,1 , y ∈ Γ∞,N,2 .
Therefore
Z Z Z Z
1 1
dµ(x)dµ(y) ≥ dµ(x)dµ(y)
Γ∞,N,1 Γ∞,N,2 d(x, y)dH +τ Γ∞,N,1 Γ∞,N,2 (d (ℵN,1 (x), ℵN,2 (y)) + Ca2n+4 )dH +τ
Z Z
1 1
= (2a)−4n−8 dxdy,
4 ΞN,1 ΞN,2 (d (x, y) + Ca2n+4 )dH +τ

and a slight modification to the proof of Assertion 1 yields that if τ > d2H , then
Z Z
1
lim dµ(x)dµ(y) = +∞,
n→∞ Γ∞,N,1 Γ∞,N,2 d(x, y)dH +τ

which implies (51).

Proof of Assertion 4. Note that
X X Z Z
−4n−8 dx dy
(2a)
Γσ Γσ ′ d(x, y)dH +τ
σ∈AN ,σ(1)=1 σ′ ∈AN ,σ′ (1)=2
X X (61)
1
≤2−4n−8 .
d(Γ , Γσ′ )dH +τ
σ
σ∈AN ,σ(1)=1 σ′ ∈AN ,σ′ (1)=2

′
First of all observe that if prox(σ) = 0 or prox(σ ′ ) = 0, d(Γσ , Γσ ) > c for some positive constant
c independent of n and the contribution of such (σ, σ ′ ) to the right hand side of (61) is bounded
by a constant independent of n.
We are left with finding a bound for
X X 1
2−4n−8 = 2−4n−8 (2I + II),
d(Γσ , Γσ′ )dH +τ
σ∈AN ,σ(1)=1,prox(σ)>0 σ′ ∈AN ,σ′ (1)=2,prox(σ′ )>0

where
n+1
XX j−1 X X 1
I= ,
d(Γσ , Γσ′ )dH +τ
j=1 k=1 σ∈AN ,σ(1)=1,prox(σ)=j σ′ ∈AN ,σ′ (1)=1,prox(σ′ )=k
n+1
X X X 1
II = .
d(Γσ , Γσ′ )dH +τ
j=1 σ∈AN ,σ(1)=1,prox(σ)=j σ′ ∈AN ,σ′ (1)=1,prox(σ′ )=j

28
But, from (49),
n+1
XX j−1 2min(n,j)
X −1 2X
k −1
22(n−min(j,n)) 22(n−k)
I.   dH +τ
j=1 k=1 i=0 ℓ=0
 h i2 2
i
(a2j + a2k )2 + ℓ − 2min(j,n)−k a4k

n+1
XX j−1 2min(n,j)
X −1 X 2k
24n−2k−2 min(j,n)
. p−(dH +τ )
j=1 k=1
a2k(dH +τ ) i=0 p=1
n+1
XX j−1 X j−1
n X X 23n−2k n
24n−2k−min(j,n) 24n−2k−j
. = +
j=1 k=1
a2k(dH +τ )
j=1 k=1
a2k(dH +τ ) a2k(dH +τ )
k=1
 
X j−1
n X n
X
2τ 2τ
−j+ d k −n+ d k
= 24n  2 H + 2 H .

j=1 k=1 k=1

From this, it is easy to prove that if τ < d2H , then 2−4n I is bounded by a constant independent
of n.
A similar argument can be used for proving that if τ < d2H , then 2−4n II is bounded by a constant
independent of n.
Proof of Assertion 5. We use the mapping ℵ(N ) : Γ∞ \N ∞ → ΓN (where N ∞ is defined by
(18)) by
ℵ(N ) (x) = Mσ (ℵ(y)) if x = Mσ (y), with y ∈ Γ∞ \N ∞ and σ ∈ AN ,
with ℵ has been defined in (19). As a consequence of (20),
Z Z
1 −2n−4
(2a) f (x)dx = f ◦ ℵN dµ, ∀f ∈ L1 (ΓN ).
2 ΓN ∞
Γ \N ∞

It is also easy to see that for all x ∈ Γ∞ \N ∞ , limn→∞ d(ℵN (x), x) = 0 and that
1 1
lim = ∀x ∈ F1 (Γ∞ )\N ∞ , y ∈ F2 (Γ∞ )\N ∞ . (62)
n→∞ d(ℵN (x), ℵN (y))dH +τ d(x, y)dH +τ
Therefore,
X X Z Z
1 dx dy
(2a)−4n−8
4 Γσ Γ σ′ d(x, y)dH +τ
σ∈AN ,σ(1)=1 σ′ ∈AN ,σ′ (1)=2
Z Z
1 dx dy
= (2a)−4n−8 dH +τ
4 F (ΓN−1 ) F2 (ΓN−1 ) d(x, y)
Z Z1
1
= N (x), ℵN (y))dH +τ
dµ(x)dµ(y).
F1 (Γ∞ )\N ∞ ∞
F2 (Γ )\N ∞ d(ℵ
From this, (53), (62) and Fatou lemma, we deduce that
Z Z
1
dH +τ
dµ(x)dµ(y) < ∞,
F1 (Γ∞ )\N ∞ F2 (Γ∞ )\N ∞ d(x, y)

and we obtain (54) because

Z Z Z Z
1 1
d +τ
dµ(x)dµ(y) = dH +τ
dµ(x)dµ(y).
F1 (Γ∞ ) F2 (Γ∞ ) d(x, y) F1 (Γ∞ )\N ∞ F2 (Γ∞ )\N ∞ d(x, y)
H

29
References
[1] Y. Achdou, C. Sabot, and N. Tchou. Diffusion and propagation problems in some ramified
domains with a fractal boundary. M2AN Math. Model. Numer. Anal., 40(4):623–652, 2006.

[2] Y. Achdou and N. Tchou. Neumann conditions on fractal boundaries. Asymptotic Analysis,
53(1-2):61–82, 2007.

[3] L. V. Ahlfors. Lectures on quasiconformal mappings. Manuscript prepared with the assis-
tance of Clifford J. Earle, Jr. Van Nostrand Mathematical Studies, No. 10. D. Van Nostrand
Co., Inc., Toronto, Ont.-New York-London, 1966.

[4] K. Falconer. Techniques in fractal geometry. John Wiley & Sons Ltd., Chichester, 1997.

[5] J.E. Hutchinson. Fractals and self-similarity. Indiana Univ. Math. J., 30(5):713–747, 1981.

[6] P.W. Jones. Quasiconformal mappings and extendability of functions in Sobolev spaces.
Acta Math., 147(1-2):71–88, 1981.

[7] A. Jonsson and H. Wallin. Function spaces on subsets of Rn . Math. Rep., 2(1):xiv+221,
1984.

[8] J. Kigami. Analysis on fractals, volume 143 of Cambridge Tracts in Mathematics. Cam-
bridge University Press, Cambridge, 2001.

[9] M.R. Lancia. A transmission problem with a fractal interface. Z. Anal. Anwendungen,
21(1):113–133, 2002.

[10] M.R. Lancia. Second order transmission problems across a fractal surface. Rend. Accad.
Naz. Sci. XL Mem. Mat. Appl. (5), 27:191–213, 2003.

[11] V.G. Maz’ja. Sobolev spaces. Springer Series in Soviet Mathematics. Springer-Verlag, Berlin,
1985. Translated from the Russian by T. O. Shaposhnikova.

[12] P. A. P. Moran. Additive functions of intervals and Hausdorff measure. Proc. Cambridge
Philos. Soc., 42:15–23, 1946.

[13] U. Mosco. Dirichlet forms and self-similarity. In New directions in Dirichlet forms, volume 8
of AMS/IP Stud. Adv. Math., pages 117–155. Amer. Math. Soc., Providence, RI, 1998.

[14] U. Mosco. Energy functionals on certain fractal structures. J. Convex Anal., 9(2):581–600,
2002. Special issue on optimization (Montpellier, 2000).

[15] U. Mosco and M. A. Vivaldi. Variational problems with fractal layers. Rend. Accad. Naz.
Sci. XL Mem. Mat. Appl. (5), 27:237–251, 2003.

[16] R. S. Strichartz. Function spaces on fractals. J. Funct. Anal., 198(1):43–83, 2003.

[17] R.S. Strichartz. Analysis on fractals. Notices Amer. Math. Soc., 46(10):1199–1208, 1999.

[18] S. K. Vodop′ janov, V. M. Gol′ dšteı̆n, and T. G. Latfullin. A criterion for the extension of
functions of the class L12 from unbounded plane domains. Sibirsk. Mat. Zh., 20(2):416–419,
464, 1979.

30