Download as pdf or txt
Download as pdf or txt
You are on page 1of 16

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/267094387

Composition Operators on Sobolev Spaces

Chapter · January 2005


DOI: 10.1090/conm/382/07079

CITATIONS READS

16 626

1 author:

Sergey Konstantinovich Vodopyanov


Sobolev Institute of mathematics
233 PUBLICATIONS 2,937 CITATIONS

SEE PROFILE

All content following this page was uploaded by Sergey Konstantinovich Vodopyanov on 20 October 2014.

The user has requested enhancement of the downloaded file.


Contemporary Mathematics
Volume 00, 1997

Description of Composition Operators of Sobolev Spaces

S. K. Vodopyanov

Dedicated to Lawrence Zalcman on his 60th birthday

Abstract. We obtain necessary and sufficient conditions for the boundedness


(or isomorphic correspondence) of the composition operator of a mapping with
the functions of Sobolev spaces having first generalized derivatives.

Consider an arbitrary open set D in the Euclidean space Rn , n ≥ 1. The


Lebesgue space Lp (D), 1 ≤ p ≤ ∞, consists of functions measurable on D having
³R ´1/p
finite norm kf | Lp (D)k = |f (x)|p dx , where dx is Lebesgue measure. The
D
Sobolev space Wp1 (D) (L1p (D)), 1 ≤ p ≤ ∞, consists of functions locally summable
on D having generalized first derivatives and finite norm (semi-norm)

kf | Wp1 (D)k = kf | Lp (D)k + k∇f | Lp (D)k


(kf | L1p (D)k = k∇f | Lp (D)k),
¡ ∂f ∂f
¢
where ∇f = ∂x 1
, . . . , ∂x n
is the generalized gradient of f .
Let D and D0 be open sets in Rn , n ≥ 1, and ϕ : D → D0 be a measurable
mapping.
Definition. A mapping ϕ induces a bounded operator ϕ∗ : Wp1 (D0 ) → Wp1 (D)
by composition if the following properties hold.
(a) If two quasicontinuous functions f1 , f2 ∈ Wp1 (D0 ) differ on a set of p-capacity
zero, then the functions f1 ◦ ϕ, f2 ◦ ϕ differ on a set of measure zero.
(b) If f ∈ Wp1 (D0 ) is a quasicontinuous representative, then f ◦ ϕ ∈ Wp1 (D) but
f ◦ ϕ is not required to be quasicontinuous.
(c) The mapping ϕ∗ : f 7→ f˜ ◦ ϕ, where f˜ is a quasicontinuous representative
of f , is a bounded operator Wp1 (D0 ) → Wp1 (D).

2000 Mathematics Subject Classification. Primary 46E35, 3065.


Key words and phrases. Sobolev space, embedding theorem, quasiconformal mapping.
This research was carried out with the partial support of the Russian Foundation for Basic
Research (Grant no. 03–01–00899) and the Program of Support of Leading Scientific Schools of
the Russian Federation (Grant no. 311.2003.1).

c
°1997 American Mathematical Society
1
2 S. K. VODOPYANOV

If the operator ϕ∗ is surjective, then we say that ϕ induces an isomorphic


operator ϕ∗ : Wp1 (D0 ) → Wp1 (D) by composition.

For the definition and properties p-capacity, see [CV, GR, M3, R1].
The main goal of the paper is to obtain necessary and sufficient conditions for
the mapping ϕ to induce a bounded operator (or an isomorphism) ϕ∗ : Wp1 (D0 ) →
Wp1 (D) by composition. In particular, we give a positive solution of the problem in
the following cases:
1) ϕ∗ : Wp1 (D0 ) → Wp1 (D) is a bounded operator, 1 ≤ p < n, in Theorem 1;
2) ϕ∗ : Wp1 (Rn ) → Wp1 (Rn ) is a bounded operator, n < p < ∞, in Theorem 2;
3) ϕ∗ : Wn1 (Rn ) → Wn1 (Rn ) is a bounded operator in Theorem 3;
4) ϕ∗ : Wp1 (D0 ) → Wp1 (D) is an isomorphism, p 6= n, in Theorem 4
Note that an analytic description of mappings ϕ : D → D0 , inducing a bounded
operator ϕ∗ : L1p (D0 ) → L1p (D) of homogeneous Sobolev spaces, is obtained in
[V1, VU1, VU2], where one can also find a detailed bibliography on this question.
The main difference between papers [V1, VU1-VU3] and those on this subject
written earlier is that we do not assume that ϕ is a diffeomorphism or a Lipschitz
homeomorphism, as was done in [M1, M2], nor even a homeomorphism [V2–V4].
1. To formulate the main results we need the following concepts. A function
f : D → R is said to be absolutely continuous on a straight line l having non-empty
intersection with D if it is absolutely continuous on an arbitrary closed segment of
this line which is contained in D. The function f : D → R belongs to the class
ACL(D) (absolutely continuous on almost all lines) if it is absolutely continuous
on almost all straight lines parallel to the jth coordinate axis, j = 1, . . . , n. Recall
that f belongs to the Sobolev space L11 (D) if and only if f is locally summable and
can be modified on a set of measure zero to belong to the class ACL(D) and its
∂f
partial derivatives ∂x i
(x), i = 1, . . . , n, existing almost everywhere, are just the
first generalized derivatives of f in D.
A mapping ϕ : D → D0 belongs to the class ACL(D) if its coordinate functions
ϕj belong to ACL(D), j = 1, . . . , n. Then the formal Jacobian matrix Dϕ(x) =
¡ ∂ϕi ¢
∂xj (x) , i, j = 1, . . . , n, and its Jacobian determinant are defined at almost all
points of D. The norm |Dϕ(x)| of the matrix Dϕ(x) is the norm of the linear
operator determined by the matrix in the Euclidean space Rn .
A mapping ϕ : D → D0 has Luzin’s condition N if the image of any set of
measure zero is a set of measure zero.
We formulate below a change-of-variable formula in the Lebesgue integral in
the form which is used in proofs of the basic results of the paper.

Proposition 1. ( [V5], Corollary 5.1) Let A be a measurable set. Assume


that a mapping ϕ : A → Rn has approximate partial derivatives in A. Then there
exists a set Σϕ ⊂ A of measure zero such that for any non-negative measurable
function u : A → R, one has the change-of-variable formula

 
Z Z X
(1) u(x)|J(x, ϕ)| dx =  u(x) dy.
A Rn x∈ϕ−1 (y)∩(A\Σϕ )

If ϕ has Luzin’s condition N , then Σϕ = ∅.


COMPOSITION OPERATORS OF SOBOLEV SPACES 3

1
Note that any mapping whose coordinate functions belong to Wq,loc (D), q > n,
1
(are monotone and belong to Wn,loc (D)) has a continuous representative possessing
Luzin’s condition N [MM, V5–V7].
We say that a mapping ϕ : D → D0 of class ACL(D) has the finite distortion
if Dϕ(x) = 0 almost everywhere in Z = {x ∈ D : J(x, ϕ) = 0}.
For a mapping ϕ : D → D0 of the class ACL(D), we define the distortion
function
³ ´ p1
 P |Dϕ|p (x)
 ,
|J(x,ϕ)|
(2) D0 3 y 7→ Hp (y) = x∈ϕ−1 (y)\Σϕ , J(x,ϕ)6=0


0, if {x ∈ ϕ−1 (y) \ Σϕ : J(x, ϕ) 6= 0} = ∅

(henceforth Σϕ ⊂ D is the set from Proposition 1). The distortion function was
introduced in [V1].
Two functions f : D0 → R and g : D0 → R are said to be equivalent (f ∼ g) if
αf (x) ≤ g(x) ≤ βf (x) for almost all x ∈ D0 , where 0 < α ≤ β < ∞ are constants
independent of the choice of x ∈ D0 .
To prove the main statements of the paper we need the following result, separate
fragments of which are contained in [VU1, VU2]. As usual, B(a, r) = {y ∈ Rn :
|y − x| < r} denotes a Euclidean ball in Rn .

Lemma 1. Let D and D0 be open sets in Rn . If ϕ : D → D0 induces a bounded


embedding operator ϕ∗ : Wp1 (D0 ) → Wp1 (D) (or ϕ∗ : L1p (D0 ) → L1p (D)), 1 ≤ p < ∞,
then ϕ can be modified on a set of measure zero to belong to ACL(D), has finite
distortion and kHp (·) | L∞ (D0 )k ≤ T kϕ∗ k, where T is some constant.

Proof. For the operator ϕ∗ : L1p (D0 ) → L1p (D), the lemma is proved in [VU2].
For the reader’s convenience, we provide the details of the proof in the case of non-
homogeneous Sobolev spaces. For proving ϕ ∈ ACL(D), we consider finite functions
ξN ∈ C0∞ (Rn ), N ∈ N, equal to 1 on the compact set B(0, N ). For any coordinate
function yj : D0 → R, j = 1, . . . , n, ξN (y) · yj ∈ Wp1 (D0 ). Then the functions
ϕ∗ (ξN yj )(x) = ξN (ϕ(x))ϕj (x) can be modified on a set of measure zero to belong
to the class Wq1 (D) ∩ ACL(D).
Fix arbitrary coordinate axes and consider the family LN consisting of all
straight lines l which are parallel to the chosen coordinate axes and such that on
every line l ∈ LN , the modified function ϕ∗ (ξN yj ), N ∈ N, is absolutely continuous.
T∞
As the set {ϕ∗ (ξN yj )}N ∈N is countable, the projection of the set L = LN along
N =1
to the chosen axes to the subspace Rn−1 has full Lebesgue (n − 1)-measure.
The modification of the function ϕ∗ (ξN yj )(x) = ξN (ϕ(x))ϕj (x), N ∈ N,
implies also a modification of the mapping ϕ. Now we show that the modified
mapping ϕ e is absolutely continuous on any line l ∈ L. First and foremost, we
notice that the set ϕe−1 (B(0, N )) ∩ l is open on a line l since ξN (y) · yj = yj for
y ∈ B(0, N ), and the modified functions ϕ∗ (ξN yj )(x), N ∈ N, are continuous on
the line l. We choose two arbitrary points x1 , x2 ⊂ D on the line l in such a
way that the segment [x1 , x2 ] ⊂ D. Because the set ϕ e−1 (B(0, N )) ∩ l is open
S

on the line l and [x1 , x2 ] ⊂ ϕe−1 (B(0, N )), there exists a number N1 such that
N =1
4 S. K. VODOPYANOV

e−1 (B(0, N1 )). Consequently, ϕ


[x1 , x2 ] ⊂ ϕ e is absolutely continuous on [x1 , x2 ] since

e [x1 ,x2 ] = (ϕ∗ (ξN1 y1 )|[x1 ,x2 ] , . . . , ϕ∗ (ξN1 yn )|[x1 ,x2 ] )


ϕ| a. e.

and the modified functions ϕ∗ (ξN1 yj ) ∈ ACL([x1 , x2 ]), j = 1, . . . , n. Thus we have


proved ϕe ∈ ACL(D).
Fix a cut-off function η ∈ C0∞ (Rn ) equal to 1 on B(0, 1) and 0 outside the ball
B(0, 2). By substituting the functions hj (z) = (zj − yj )η( z−y
r ), j = 1, . . . , n, where
n
zj − yj is the j-th coordinate of the
° vector z −
° y ∈ R , B(y, 2r) ⊂ D0 , into the
∗ ∗ ° 1 0 °
inequality k∇ϕ f | Lp (D)k ≤ kϕ k f | Wp (D ) , we arrive at the inequality
µ Z ¶1/p
(3) |Dϕ|p (x) dx ≤ Ckϕ∗ k|B(y, 2r)|1/p , r ∈ (0, 1),
ϕ−1 (B(y,r))

since khj | Lp (D0 )k ≤ C|B(y, 2r)|1/p for all the balls B(y, 2r) ⊂ D0 , r ∈ (0, 1) (here
C is a constant depending only on the dimension n and the index p).
Let Z = {x ∈ D : J(x, ϕ) = 0}. We want to show that
Z
(4) |Dϕ|p (x) dx = 0.
Z

By formula (1), |ϕ(Z \ Σϕ )| = 0. Fix ε > 0 and an open set U ⊂ D0 such that
U ⊃ ϕ(Z\Σϕ ) and |U | < ε. We can choose a covering {B(yi , ri )}, ri < 12 dist(yi , ∂U ),
of U of finite multiplicity such that B(yi , 2ri ) ⊂ U , i ∈ N, and this collection of balls
also constitutes a covering of U of finite multiplicity. Moreover,
P the multiplicity M
of the covering {B(yi , 2ri )} is independent of U , and |B(yi , 2ri )| < M ε. Then,
i
applying the inequality (3), we derive
Z Z ∞
X Z
|Dh ϕ|p (x) d = |Dh ϕ|p (x) dx ≤ |Dh ϕ|q (x) dx
Z i=1
Z\Σϕ ϕ−1 (B(yi ,ri ))

X
≤ C p kϕ∗ kp |B(yi , 2ri )| ≤ C1p kϕ∗ kp |U |.
i=1

As ε > 0 is arbitrary, (4) is proved. Hence |Dh ϕ| = 0 almost everywhere on Z.


Changing variables on the left hand side of (3) by formula (1), we obtain
Z X |Dϕ|p (x) p
(5) dy ≤ C 0 kϕ∗ kp |B(y, r)|.
|J(x, ϕ)|
B(y,r) x∈ϕ−1 (y)\Σϕ ,
|J(x,ϕ)|6=0

Applying the Lebesgue differentiability theorem, we infer ess sup Hp (y) ≤ C 0 kϕ∗ k.
y∈D 0
The lemma is proved. ¤

Lemma 1, Theorems 1 and 3 and Proposition 4 of [VU2] imply the following


result.
COMPOSITION OPERATORS OF SOBOLEV SPACES 5

Theorem 1. Let D and D0 be open sets in Rn . A mapping ϕ : D → D0


induces a bounded embedding operator ϕ∗ : Wp1 (D0 ) → Wp1 (D), 1 ≤ p < n, if and
only if one of the following conditions holds:
1) ϕ can be modified on a set of measure zero to belong to the class ACL(D),
has finite distortion and Hp (·) ∈ L∞ (D0 );
2) ϕ induces a bounded embedding operator ϕ∗ : L1p (D0 ) → L1p (D).
The norm of the operator ϕ∗ : Wp1 (D0 ) → Wp1 (D), 1 ≤ p < n, is equivalent to
the value
µ X ¶ p1
1
ess sup + kHp (·) | L∞ (D0 )k.
y∈D 0 −1
|J(x, ϕ)|
x∈ϕ (y)\Σϕ

Proof. By Lemma 1, the boundedness of the operator ϕ∗ : Wp1 (D0 ) → Wp1 (D)
(or ϕ∗ : L1p (D0 ) → L1p (D)), 1 ≤ p < n, implies that ϕ can be modified on a set of
measure zero to belong to ACL(D), ϕ has finite distortion and Hp (·) ∈ L∞ (D0 ).
Moreover, if the operator ϕ∗ : Wp1 (D0 ) → Wp1 (D) is bounded then the modified
mapping ϕ e induces a bounded operator ϕ∗ : L1p (D0 ) → L1p (D), 1 ≤ p < n, by
Theorem 1 of [VU2].
Now suppose that the operator ϕ∗ : L1p (D0 ) → L1p (D), 1 ≤ p < n, is bounded.
The essential part of the converse assertion is a verification that the bounded-
ness of the quantity kHp (·) | L∞ (D0 )k implies the boundedness of the operator
ϕ∗ : Wp1 (D0 ) → Wp1 (D). Indeed, by the pointwise relation |J(x, ϕ)| ≤ |Dϕ|n (x)
and Jensen inequality, we have
µ X ¶1
|Dϕ|p (x) p
Hp (y) =
−1
|J(x, ϕ)|
x∈ϕ (y)\Σϕ
µ X µ ¶ np ¶ p1
|Dϕ|n (x) 1
= · p

x∈ϕ−1 (y)\Σϕ
|J(x, ϕ)| |J(x, ϕ)|1− n
µ X ¶ p1 µ X ¶ p1 − n1
1 1
≥ p ≥
x∈ϕ−1 (y)\Σϕ
|J(x, ϕ)|1− n x∈ϕ−1 (y)\Σϕ
|J(x, ϕ)|

for almost all y ∈ D0 . It remains to observe


P ([VU2], Proposition 4), that the
1
essential boundedness of Jϕ−1 (y) = |J(x,ϕ)| is a necessary and sufficient
x∈ϕ−1 (y)\Σϕ
condition for the operator ϕ∗ : Lp (D0 ) → Lp (D) to be bounded with norm equiv-
1
alent to kJϕ−1 (·) p | L∞ (D0 )k. If ϕ has the property that capp (A) = 0, A ⊂ D0 ,
implies |ϕ−1 (A)| = 0, then ϕ∗ : Lp (D0 ) ∩ L1p (D0 ) → Lp (D) ∩ L1p (D) is a bounded
operator ϕ∗ : Wp1 (D0 ) → Wp1 (D) with the above-mentioned estimate of its norm.
¤

Remark. In Theorem 1 and other statements, the sufficient part of an as-


sertion is proved by the following scheme. We consider a mapping ϕ : D → D0
having the ACL-property. For any smooth function f : D0 → R, the composition

f ◦ϕ : D → R possesses the ACL-property and has first partial derivatives ∂x i
(f ◦ϕ)
calculated by the classical chain rule. Applying formula (1) we see that the oper-
ator ϕ∗ : Wp1 (D0 ) ∩ C 1 (D0 ) → Wp1 (D) is bounded. It follows that the preimage
of a set of capacity zero is a set of measure zero. It follows that the extension of
6 S. K. VODOPYANOV

the operator ϕ∗ : Wp1 (D0 ) ∩ C 1 (D0 ) → Wp1 (D) by continuity coincides with the
composition operator.
Let E ⊂ Rn be a measurable set. A mapping ϕ : E → Rn is called Lipschitz if
|ϕ(y) − ϕ(x)| ≤ L|y − x| for some constant L and all points x, y ∈ E.
Proposition 2. Suppose ϕ : D → D0 induces a bounded operator ϕ∗ : L1p (D0 ) →
L1p (D),n < p < ∞. Then
1 p
1) |J(x, ϕ)| n ≤ |Dϕ|(x) ≤ kHp (·) | L∞ (D0 )k p−n for almost all x ∈ D;
1
2) ϕ ∈ W∞,loc (D);
3) the mapping ϕ can be modified on a set of measure zero to be a continuous
mapping Φ : D → Rn which is Lipschitz on every ball B ⊂ D;
4) the mapping Φ has Luzin’s property N .
Proof. Lemma 1 implies that ϕ can be modified on a set of measure zero to
belong to ACL(D), to have the finite distortion and Hp (·) ∈ L∞ (D0 ). In the case
J(x, ϕ) = 0 for almost all x ∈ D, the statement of the proposition is evident because
the ACL-property and finite distortion imply that ϕ differs from a mapping which
is constant on every connected component of the open set D on a set of measure
zero. Therefore, we assume that J(x, ϕ) 6= 0 on a set of positive measure.
By the pointwise relation |J(x, ϕ)| ≤ |Dϕ|n (x) and Jensen’s inequality, we have
µ X ¶1
|Dϕ|p (z) p
(6) Hp (y) =
|J(z, ϕ)|
z∈(ϕ−1 (y)∩V )\Σϕ ,
J(z,ϕ)6=0
µ X µ ¶ np ¶ p1
|Dϕ|n (z) 1 1 1
= · p ≥ |J(x, ϕ)| n − p
|J(z, ϕ)| |J(z, ϕ)|1− n
z∈(ϕ−1 (y)∩V )\Σϕ ,
J(z,ϕ)6=0

for almost all y = ϕ(x) ∈ D0 . It follows that


1 1
|J(x, ϕ)| n − p ≤ kHp (·) | L∞ (D0 )k
for almost all x ∈ D. Indeed, if
1 1
E = {y ∈ D0 : Hp (y) < |J(x, ϕ)| n − p , ϕ(x) = y}
then |E| = 0. Hece, by the change-of-variable formula, J(x, ϕ) = 0 almost every-
where on the set ϕ−1 (E).
Further,
µ X ¶1
|Dϕ|p (z) p
Hp (y) =
|J(z, ϕ)|
z∈(ϕ−1 (y)∩V )\Σϕ ,
J(z,ϕ)6=0
µ ¶ p−n
n µ
X ¶ p1
1
≥ |Dϕ|p (z)
kHp (·) | L∞ (D0 )k
z∈(ϕ−1 (y)∩V )\Σϕ ,
J(z,ϕ)6=0
µ ¶ p−n
n
1
≥ |Dϕ|(x).
kHp (·) | L∞ (D0 )k
COMPOSITION OPERATORS OF SOBOLEV SPACES 7

Evaluating the quantity Hp (y) by means of kHp (·) | L∞ (D0 )k, we obtain assertion 1.
The boundedness of the norm of the gradient and the ACL-property of the
mapping ϕ imply assertion 2.
Modifying the mapping on a set of measure zero, we obtain assertion 3. Asser-
tion 4 is a consequence of the fact that Φ is locally Lipschitz. ¤

Theorem 2. The mapping ϕ : Rn → Rn induces a bounded embedding operator


ϕ : Wp1 (Rn ) → Wp1 (Rn ), n < p < ∞, if and only if ϕ has the following properties:

1) ϕ can be modified on a set of measure zero to be Lipschitz on every ball


B ⊂ D, ϕ hasR the finitePdistortion and Hp (·) ∈ L∞ (Rn );
1 ∗ p
2) V = sup |J(x,ϕ)| dy ≤ ckϕ k , for some constant c.
z∈Rn B(z,1) x∈ϕ−1 (y), J(x,ϕ)6=0
The norm of the operator ϕ∗ : Wp1 (Rn ) → Wp1 (Rn ), n < p < ∞, is equivalent
1
to V p + kHp (·) | L∞ (Rn )k.
Proof. Let the embedding operator ϕ∗ : Wp1 (Rn ) → Wp1 (Rn ), n < p < ∞, be
bounded. We proved in Lemma 1 that the mapping ϕ can be modified on a set of
measure zero to belong to the class ACL(D), that it has finite distortion and that
kHp (·) | L∞ (D0 )k ≤ ckϕ∗ k.
Proposition 2 implies that ϕ can be modified on a set of measure zero to be
Lipschitz on every ball B ⊂ D. We obtain condition 2 and the estimate V ≤ ckϕ∗ kp ,
where c is some°constant, if we °substitute the ° function h(·)
°p = η(·−z) instead of f in
p
the inequality °ϕ∗ f | Wp1 (Rn )° ≤ kϕ∗ kp °f | Wp1 (Rn )° , where η ∈ C0∞ (B(0, 2))
is a test function such that η ≡ 1 on B(0, 1) and z ∈ Rn is arbitrary.
To prove the sufficiency of conditions it remains only to estimate
kϕ∗ (f ) | Lp (Rn )k since by Theorem 1 of [VU2], the operator ϕ∗ : L1p (Rn ) → L1p (Rn )
is bounded with norm kϕ∗ k equivalent to kHp (·) | L∞ (Rn )k. To do this, we fix a
covering {Bj = B(zj , 1/2)} of the space Rn of a finite multiplicity such that the
covering {2Bj = B(zj , 1)} also has finite multiplicity. Let hj (·) = ξ(· − zj ), where
ξ ∈ C0∞ (B(0, 1)) is a test function such that ξ ≡ 1 on B(0, 1/2) and 0 ≤ ξ ≤ 1.
Assuming ϕ and f ∈ Wp1 (Rn ) to be continuous, we have
X
kϕ∗ (f ) | Lp (Rn )kp ≤ kϕ∗ (f hj ) | Lp (Rn )kp
j
X X
−1
≤ |ϕ (2Bj )|kf hj | C(ϕ−1 (2Bj ))kp ≤ |ϕ−1 (2Bj )|kf | C(2Bj )kp
j j
X
−1
≤ sup |ϕ (2Bj )| kf | Wp1 (2Bj )kp ≤ V Ckf | Wp1 (Rn )kp ,
j
j

since by (1),
Z X 1
|ϕ−1 (2Bj )| = dy ≤ V.
−1
|J(x, ϕ)|
B(zj ,1) x∈ϕ (y),J(x,ϕ)6=0

Here C depends only on the multiplicity of the covering and the norm of the
embedding operator i : Wp1 (Rn ) → C(Rn ). ¤

For a mapping ϕ : D → D0 of the class ACL(D), we introduce the characteristic


Kp (x) = inf{k : |Dϕ|(x) ≤ k|J(x, ϕ)|1/p }, x ∈ Ω.
8 S. K. VODOPYANOV

¡ ¢
Recall that the capacity cap e; Wp1 (Ω) of the compact e ⊂ Ω in the space
Wn1 (Ω) is the quantity
¡ ¢
cap e; Wp1 (Ω) = inf kg | Wp1 (Ω)kp ,

where the infimum is taken over all continuous functions g ∈ Wp1 (Ω) such that g ≥ 1
on e. For properties and applications of capacity see, for instance, [CV, M3, MS, R1].
Theorem 3. A mapping ϕ : Rn → Rn induces a bounded embedding operator
ϕ∗ : Wn1 (Rn ) → Wn1 (Rn ) if and only if
1) ϕ can be modified on a set of measure zero to belongs to ACL(Rn ) and has
the finite distortion;
2) Kn (·) ∈ L∞ (Rn );
3) the Banach indicatrix M (y, ϕ) = #{x ∈ ϕ−1 (y) \ Σϕ } belongs to L∞ (Rn );
4) the least constant in the inequality
Z X 1 ¡ ¢
dy ≤ C cap e; Wn1 (Rn )
−1
|J(x, ϕ)|
e x∈ϕ (y)\Σϕ , J(x,ϕ)6=0

is bounded where e is an arbitrary compact, the diameter of which does not


exceed 1.
The norm of the operator ϕ∗ : Wn1 (Rn ) → Wn1 (Rn ) does not exceed
1 1
C n + kM (·, ϕ) | L∞ (Rn )k n · kKn (·) | L∞ (Rn )k.
If the Jacobian J(x, ϕ) has a single sign then ϕ can be modified on a set of
measure zero to be a mapping with bounded distortion of finite multiplicity [R2].
If, additionally, kM (·, ϕ) | L∞ (Rn )k = 1, then ϕ is a quasiconformal mapping.
Proof. By Lemma 1, the boundedness of the operator ϕ∗ : Wn1 (D0 ) → Wn1 (D)
implies that ϕ can be modified on a set of measure zero to belong to ACL(D),
ϕ has the finite distortion and Hn (·) ∈ L∞ (D0 ). By Theorem 1 of [VU2], the
operator ϕ∗ : L1n (D0 ) → L1n (D)) is bounded and its norm is equivalent to the value
kHn (·) | L∞ (D0 )k. From Theorem 5 of [VU2], it follows that conditions 1–3 are
necessary and sufficient for the operator ϕ∗ : L1n (Rn ) → L1n (Rn ) to be bounded.
Moreover, its norm does not exceed
1
kM (·, ϕ) | L∞ (Rn )k n · kKn (·) | L∞ (Rn )k.
Now, let f ∈ Wn1 (Rn ). As is known [MS],
Z X 1
kf ◦ ϕ | Ln (R )k = |f (y)|n
n n
dy
|J(x, ϕ)|
Rn x∈ϕ−1 (y)\Σϕ , J(x,ϕ)6=0

≤ kik · kf | Wn1 (Rn )kn


if and only if condition 4 of the theorem holds. The least constant kik in this
inequality is equivalent to the least constant C of condition 4. ¤

To complete our exposition, we formulate the following result.


Corollary 1. ([V8, V9]) A homeomorphism ϕ : Rn → Rn induces a bounded
embedding operator ϕ∗ : Wn1 (Rn ) → Wn1 (Rn ) if and only if
1
1) ϕ ∈ Wn,loc (Rn );
2) the mapping ϕ is quasiconformal and Kn ≤ c1 kϕ∗ k;
COMPOSITION OPERATORS OF SOBOLEV SPACES 9

3) R = sup sup |ϕ−1 (x) − ϕ−1 (y)| ≤ c2 kϕ∗ k where c2 depends on the
x∈Rn y∈B(x,1)
coefficient of distortion Kn .
The norm of the operator ϕ∗ : Wn1 (Rn ) → Wn1 (Rn ) does not exceed the quantity
(c3 R + 1)Kn . (Here c1 , c3 are constants.)
Proof. If a homeomorphism ϕ : Rn → Rn induces a bounded embedding
operator ϕ∗ : Wn1 (Rn ) → Wn1 (Rn ) then it is evident that ϕ ∈ Wn,loc
1
(Rn ).
Suppose that the embedding operator ϕ∗ : Wn1 (Rn ) → Wn1 (Rn ) is bounded.
Then Lemma 1 and [VU2], Theorem 1 imply the boundedness of the operator
ϕ∗ : L1n (Rn ) → L1n (Rn ) and the estimate on its norm. This condition is equivalent
to the quasiconformality of the mapping ϕ; see, for instance, [Mo, Vä]. Moreover,
the mapping ψ = ϕ−1 is also quasiconformal; therefore,
max |ψ(z) − ψ(y)|
y∈S(z,1)
sup = D < ∞,
z∈Rn min |ψ(z) − ψ(y)|
y∈S(z,1)

where D depends on the quasiconformality coefficient Kn [Mo, Vä]. (We may


express this by saying that the image of a ball is a quasiball.) We now obtain
condition 3 and the estimate

R = sup sup |ϕ−1 (x) − ϕ−1 (y)| ≤ c4 kϕ∗ k,


x∈Rn y∈B(x,1)

where c4 is a constant, by substituting the function h(·) = η(· − z) in place of f in


the inequality ° ∗ ° ° °
°ϕ f | Wn1 (Rn )° ≤ kϕ∗ k°f | Wn1 (Rn )°,
where η ∈ C0∞ (B(0, 2)) is a test function such that η ≡ 1 on B(0, 1) and z ∈ Rn is
arbitrary.
To verify the sufficiency of the conditions, it remains only to estimate
kϕ∗ (f ) | Ln (Rn )k since in view of [V8, V9] (see also [VU2], Theorem 1), the
operator ϕ∗ : L1n (Rn ) → L1n (Rn ) is bounded and its norm equals Kn . To do this,
we fix a covering {Bj = B(zj , 1/2)} of the space Rn of finite multiplicity such that
the covering {2Bj = B(zj , 1)} is also of finite multiplicity. Let hj (·) = ξ(· − zj )
where ξ ∈ C0∞ (B(0, 1)) is a test function such that ξ ≡ 1 on B(0, 1/2) and 0 ≤ ξ ≤ 1.
Applying Poincaré’s inequality, we obtain
X
kϕ∗ (f ) | Ln (Rn )kn ≤ C1 kϕ∗ (f hj ) | Ln (Rn )kn
j
X
≤ C1 C2 R n k∇(ϕ∗ (f hj )) | Ln (ϕ−1 (2Bj ))kn
j
X
≤ C1 C2 R Knnn
k∇(f hj ) | Ln (2Bj )kn
j

≤ C2 C3 Rn Knn kf | Wn1 (Rn )kn ,

where C1 , C2 , C3 are constants depending on the constant in Poincaré’s inequality


and the multiplicity of the covering. The boundedness of the operator
ϕ∗ : Wn1 (Rn ) → Wn1 (Rn ) and the estimate on its norm are proved. ¤
10 S. K. VODOPYANOV

Proposition 3. Let f be a function of Sobolev class, as in Lemma 1,


Theorems 1–3, Proposition 2 and Corollary 1. The partial derivatives of the com-
position f ◦ ϕ are calculated by the formula
 P n
 ∂f ∂ϕj
∂(f ◦ ϕ) ∂yj (ϕ(x)) ∂xi (x), if J(x, ϕ) 6= 0,
(x) = j=1
∂xi 
0 otherwise
for almost all x ∈ D.
Proof. We can assume that ϕ has the ACL-property. In the case of a smooth
function f , the chain rule is derived by taking into consideration the ACL-property
of the mapping ϕ and its bounded distortion. The general case is obtained as a
result of approximation of an arbitrary function of Sobolev class by a sequence of
smooth functions. Indeed, if, for example, fn → f in L1p (D0 ) (or Wp1 (D0 )), then we
can suppose that ∂f ∂f 0 0
∂yj (y) → ∂yj (y) everywhere in D \ Σ where Σ ⊂ D is a set of
n

−1
measure zero. Then by (1), the Jacobian J(x, ϕ) vanishes on ϕ (Σ) a.e. Thus we
have fn ◦ ϕ → f ◦ ϕ in L1p (D) (or Wp1 (D)) and
n
X ∂fn X ∂fn
∂(fn ◦ ϕ) ∂ϕj ∂ϕj
(x) = (ϕ(x)) (x) → (ϕ(x)) (x) a.e.
∂xi j=1
∂yj ∂xi j=1
∂yj ∂xi

on D \ (ϕ−1 (Σ) ∪ {x ∈ D : J(x, ϕ) = 0}) and 0 = ∂(f∂xn ◦ϕ)


i
(x) = ∂(f ◦ϕ)
∂xi (x) a.e. on
−1
ϕ (Σ) ∪ {x ∈ D : J(x, ϕ) = 0}, since ϕ has finite distortion. ¤

2. Let E ⊂ Rn be a measurable set. A mapping ϕ : E → Rn is called bi-


Lipschitz if it is Lipschitz and has an inverse mapping ϕ−1 : ϕ(E) → E which is
also Lipschitz.
A homeomorphism ϕ : D → D0 of two open sets is called quasi-isometric if for
some constant M the relations
|ϕ(y) − ϕ(x)| |ϕ−1 (y) − ϕ−1 (z)|
lim ≤ M, lim ≤M
y→x |y − x| y→z |y − z|
hold at all points x ∈ D and z ∈ D0 .
It is well-known (see, for instance, [VGR]) that any quasi-isometric mapping
ϕ : D → D0 induces an isomorphism ϕ∗ : Wp1 (D0 ) → Wp1 (D), 1 ≤ p < ∞, of
Sobolev spaces. Conversely, results of [G] may be interpreted to say that if a
homeomorphism ϕ : D → D0 induces an isomorphism ϕ∗ : L1p (D0 ) → L1p (D),
1 ≤ p < ∞, p 6= n, of homogeneous Sobolev spaces, then ϕ : D → Rn is quasi-
isometric.
Recall that two open sets Ω1 and Ω2 are said to be (1, p)-equivalent [VG1]
if, for any function f ∈ Wp1 (Ω1 ) (f ∈ Wp1 (Ω2 )), the restriction f |Ω1 ∩Ω2 can be
extended in a unique way to be a function f˜ ∈ Wp1 (Ω2 ) (f˜ ∈ Wp1 (Ω1 )). It follows
that the operator Wp1 (Ω1 ) 3 f 7→ f˜ ∈ Wp1 (Ω2 ) is an isomorphism of vector spaces.
This definition implies that not only open sets Ω1 and Ω2 are (1, p)-equivalent but
Ω2 and Ω1 are also (1, p)-equivalent. Moreover, if open sets Ω1 and Ω2 are (1, p)-
equivalent and open sets Ω2 and Ω3 are (1, p)-equivalent, then the open sets Ω1 and
Ω3 are (1, p)-equivalent.
In [VG1], a stronger result is obtained comparing with [G]: if a measurable
mapping ϕ : D → D0 induces an isomorphism ϕ∗ : Wp1 (D0 ) → Wp1 (D), p > n, of
COMPOSITION OPERATORS OF SOBOLEV SPACES 11

Sobolev spaces, then ϕ : D → Rn can be modified on a set of measure zero in such


a way that the modified mapping Φ : D → Rn becomes quasi-isometric. Moreover,
the open sets D0 and Φ(D) are (1, p)-equivalent. In this paper, we extend this result
to the whole scale of Sobolev spaces.
Theorem 4. A mapping ϕ : D → D0 induces an isomorphism ϕ∗ : Wp1 (D0 ) →
Wp1 (D),1 ≤ p < ∞, p 6= n, if and only if ϕ : D → D0 coincides almost everywhere
with a quasi-isometric homeomorphism Φ : D → Rn . Moreover, the open sets D0
and Φ(D) are (1, p)-equivalent.
Proof. The sufficiency of the condition is verified in the standard way; for
details, see [V9, VG1].
The necessity of the theorem in the case 1 ≤ p < n and D 6= Rn or D0 6= Rn is
a new result. The proof given below is also new. Other proofs of the necessity for
p > n (1 < p < n and D = D0 = Rn ), known earlier, can be found in [VG1, GR]
([R]).
In the proof given below, we reduce the cases p > n and p < n to the same
situation.
Common part of arguments. By Lemma 1, we can suppose that the mapping ϕ
belongs to the class ACL(D), has finite distortion and Hp (·) ∈ L∞ (D0 ), 1 ≤ p < ∞,
p 6= n. Let
Z = {x ∈ D : J(x, ϕ) = 0}.
There exists a family {Ej }, j = 0, 1, . . . , of pairwise disjoint bounded sets
Ej ⊂ D \ Z such that |E0 | = 0, |Ej | > 0 for all j ∈ N, ϕ|Ej is a bi-Lipschitz
S

mapping for all j ≥ 1, and D = Z ∪ Ej (see, for instance, [V5, VU4]). We
j=0
assume that the set Σϕ from Proposition 1 is included in E0 and Z ∩ E0 = ∅.
Observe that the assumption |ϕ(Ei ) ∩ ϕ(Ej )| > 0 for some i, j ≥ 1, i 6= j, leads
immediately to a contradiction with the isomorphism of the operator ϕ∗ . Indeed,
by this assumption, for any compact set A ⊂ ϕ(Ei )∩ϕ(Ej ) of positive measure, the
pre-images ϕ−1 (A) ∩ Ei and ϕ−1 (A) ∩ Ej are disjoint, and each of them is compact
and has positive measure. Then a finite smooth function ψ, which equals 1 (0) on
ϕ−1 (A) ∩ Ei (ϕ−1 (A) ∩ Ej ) belongs to the Sobolev space Wp1 (D) and cannot be a
composition of any function f ∈ Wp1 (D0 ) with the mapping ϕ.
It follows that outside of some set F ⊂ D\(Z ∪E0 ) of measure zero the mapping
ϕ is bijective on D \ (Z ∪ E0 ∪ F ). Hence, on ϕ(D) \ ϕ(E0 ∪ Z ∪ F ), there exists
an inverse mapping ψ. From this, it follows that ϕ∗ −1 (g)(y) = g ◦ ψ(y) for all
y ∈ ϕ(D) \ ϕ(E0 ∪ Z ∪ F ).
Since the mapping ψ is Lipschitz on every set ϕ(Ei \ F ), it has an approximate
differential (see, for instance, [F]) almost everywhere on its range of definition.
Fix a cut-off function η ∈ C0∞ (Rn ) which equals 1 on B(0, 1) and 0 outside
of B(0, 2). Let x be a density point of D \ (Z ∪ E0 ∪ F ). Substituting the
function hj (z) =° (z − x)j η(°z−x
° r ), j =° 1, . . . , n, in place of f in the inequality
°ϕ∗ −1 f | L1p (D0 )° ≤ kϕ∗ −1 k°f | L1p (D)°, where (z − x)j denotes the j-th coordi-
nate of the vector z − x and B(x, 2r) ⊂ D, we have
Z
(7) |Dψ|p (y) dy ≤ Ckϕ∗ −1 kp |B(x, 2r)|,
ϕ(B(x,r)\(E0 ∪Z∪F ))
12 S. K. VODOPYANOV

in which C is a constant depending only on the dimension n and the index p.


Applying formula (1) to the left-hand side of (7), we obtain
Z Z
p |Dψ|p (ϕ(z))
|Dψ| (y) dy = dz
|J(ϕ(z), ψ)|
ϕ(B(x,r)\(E0 ∪Z∪F )) B(x,r)\(E0 ∪Z∪F )

≤ Ckϕ∗ −1 kp |B(x, 2r)|.


By the Lebesgue differentiability theorem, it follows that
|Dψ|p (y)
≤ Ckϕ∗ −1 kp
|J(y, ψ)|
for almost all y ∈ ϕ(D \(E0 ∪Z ∪F )), since the mapping ϕ is injective and possesses
Luzin’s property N on D \ (E0 ∪ Z ∪ F ). Hence we have the pointwise estimate
(8) |Dψ|p (y) ≤ Ckϕ∗ −1 kp |J(y, ψ)|
between the norm of the formal differential Dψ(y) and its Jacobian for almost all
y ∈ ϕ(D \ (E0 ∪ Z ∪ F )).
Case p > n. In this case, the mapping ϕ has the following additional properties.
1) By Proposition 2, ϕ can be modified on a set of measure zero to a mapping
Φ : D → Rn which is Lipschitz on every ball B ⊂ D, has finite distortion and whose
Jacobian satisfies |J(y, Φ)| ≤ γ < ∞ almost everywhere in D for some constant γ.
Moreover, Φ possesses Luzin’s property N and is differentiable almost everywhere
in D. Therefore, |Φ(E0 ∪ F )| = 0. Since Φ(Z) has measure zero, by formula (1),
|Φ(E0 ∪ F ∪ Z)| = 0. Thus, the mapping ψ is defined almost everywhere on ϕ(D).
2) Applying arguments used in proving of (6), we obtain from (8)
Dψ|p (y)
Ckϕ∗ −1 kp ≥
|J(y, ψ)|
µ ¶p
|Dψ|n (y) n 1 p
= · n −1
p ≥ |J(y, ψ)|
|J(y, ψ)| |J(y, ψ)|1− n
for almost all y = ϕ(x) ∈ ϕ(D). Therefore, the Jacobian |J(y, ψ)| is bounded from
above: |J(y, ψ)| ≤ α < ∞ almost everywhere in ϕ(D).
3) |J(x, Φ)| = |J −1 (y, ψ)|, y = Φ(x), for almost all points x ∈ D \ Z. It follows
that α−1 ≤ |J(x, Φ)| ≤ γ for almost all points x ∈ D \ Z.
4) From the previous property, it follows that the locally Lipschitz mapping
Φ : D → Rn is invertible on D \ Z outside a set of measure zero and satisfies not
only the pointwise estimate |DΦ|p (x) ≤ Kp |J(x, Φ)| for almost all x ∈ D, where
Kp is some constant, but also the pointwise estimate |DΦ|n (x) ≤ Kn |J(x, Φ)| for
almost all x ∈ D, where Kn is a constant depending on n and kϕ∗ k. By [VU2],
Theorem 5, the mapping ϕ induces a bounded operator ϕ∗ : L1n (D0 ) → L1n (D). By
[VU2], Corollary 4, we have |Z| = 0, where Z = {x ∈ D : J(x, ϕ) = 0}.
Case 1 ≤ p < n. In this case, both the set Z = {x ∈ D : J(x, ϕ) = 0} of zeros
of the Jacobian and its image ϕ(Z) have measure zero by [VU2], Theorem 4, and
Proposition 1. In addition, the mapping ϕ has the following additional properties.
5) Since ϕ is almost bijective, it follows from Proposition 1 (see also [VU2],
Proposition 4) that |J −1 (x, ϕ)| = |J(y, ψ)|, y = ϕ(x), for almost all points x ∈ D.
6) By Theorem 3 of [VU2], the mapping ψ : ϕ(D \ (E0 ∪ F )) → D \ (E0 ∪ F )
has Luzin’s property N and its Jacobian is bounded from above almost everywhere:
COMPOSITION OPERATORS OF SOBOLEV SPACES 13

|J(y, ψ)| ≤ α < ∞. It follows that |J(x, ϕ)| = |J(y, ψ)|−1 ≥ α−1 > 0, y = ϕ(x), for
almost all points x ∈ D \ (E0 ∪ F ).
7) |Dψ|p (y) ≤ Ckϕ∗ −1 kp |J(y, ψ)| ≤ αCkϕ∗ −1 kp and |J(y, ψ)| ≥ γ −1 > 0 for
almost all y ∈ ϕ(D \ ϕ(E0 ∪ F )). Here the constant C is independent of the choice
of domains and the mapping ϕ.
We use estimate (8) to prove the last property. Indeed,
µ ¶p
∗ −1 p |Dψ|p (y) |Dψ|n (y) n 1 1
Ckϕ k ≥ = 1− p ≥ p ;
|J(y, ψ)| |J(y, ψ)| |J(y, ψ)| n |J(y, ψ)|1− n
therefore, |J(y, ψ)| ≥ γ −1 > 0 for almost all y ∈ ϕ(D \ (E0 ∪ F )).
8) |Dϕ|(x) ≤ L for almost all x ∈ D.
This property follows from the inequalities |Dϕ|p (x) ≤ C1 |J(x, ϕ)| and
|J(x, ϕ)| ≤ γ, which are valid for almost all x ∈ D (the first property follows
from Lemma 1 and the second one is a consequence of properties 5 and 7).
9) On every ball B ⊂ D, the mapping ϕ coincides almost everywhere with a
Lipschitz mapping Φ.
10) The image Φ(E0 ∪ F ) has measure zero.
11) The mapping Φ : D → Rn is Lipschitz on every ball B ⊂ D and invertible
outside of a set of measure zero; and |DΦ|n (x) ≤ Kn |J(x, Φ)| for almost all x ∈ D,
where Kn is some constant depending on n and kΦ∗ k. By Theorem 5 of [VU2],
the operator Φ∗ : L1n (D0 ) → L1n (D) is bounded.
Note that the findings of 4) and 11) coincide. Further, we have a common part
of arguments.
12) The mapping Φ is topologically non-degenerate, i.e., if a point Φ(z) does
not belong to the image Φ(S(z, r)) of the sphere S(z, r), where B(z, r) b D, then
the topological degree µ(Φ(z), Φ, B(z, r)) does not vanish.
To prove this, we show that the assumption µ(Φ(z), Φ, B(x, r)) = 0 contradicts
the condition of injectivity almost everywhere. Indeed, in this case, we substitute
the characteristic function χV of those connected component V of the complement
Rn \ Φ(B(z, r)) to which the point Φ(z) belongs, in place of f in the change-of-
variable formula with topological degree (see, for instance, [VG2, V5])
Z Z
(9) f (Φ(x))J(x, Φ) dx = f (y)µ(y, Φ, B(z, r)) dy
B(z,r) Rn

to obtain Z
J(x, Φ) dx = 0.
B(z,r)∩Φ−1 (V )

Since, by the continuity of Φ, we have |B(z, r) ∩ Φ−1 (V )| > 0, formula (1)


implies |Φ(B(z, r)) ∩ V | > 0 (as J(x, Φ) 6= 0 almost everywhere). On the
other hand, the sets Z+ = {x ∈ B(z, r) ∩ Φ−1 (V ) : J(x, Φ) > 0} and
Z− = {x ∈ B(z, r) ∩ Φ−1 (V ) : J(x, Φ) < 0} do not intersect, and each of them
has positive measure (if one of them had measure zero, then from (9) we would ob-
tain µ(Φ(z), Φ, B(z, r)) 6= 0). It follows that each of the images Φ(Z+ ) and Φ(Z− )
has positive measure and theyPintersect on a set of positive measure (as we have
0 = µ(y, Φ, B(z, r)) = sign J(x, Φ) for almost all y ∈ V ; see, for
x∈Φ−1 (y)∩B(z,r)
instance, [V5]).
14 S. K. VODOPYANOV

13) For any point y ∈ D0 , the one-dimensional Hausdorff measure of the pre-
image Φ−1 (y) equals zero.
Indeed, it was mentioned above that the operator Φ∗ : L1n (D0 ) → Wn1 (D) is
bounded and therefore the pre-image of a set of n-capacity zero in D0 is a set of
n-capacity zero in D (see, for instance, [V10]). In particular, the pre-image of a
point has n-capacity zero. Hence, it has one-dimensional Hausdorff measure zero
[CV, R1, R2].
Observe that the last property means that the mapping Φ is light: the pre-
image Φ−1 (y) is totally disconnected for any point y ∈ D0 .
14) The mapping Φ is a homeomorphism.
We first show that Φ is an open mapping. Let x ∈ D. Then, by the previous
property, there exists a ball B(x, r) such that Φ(x) ∈/ Φ(S(x, r)). By property 12,
we have µ(Φ(x), Φ, B(x, r)) 6= 0. Hence, by properties of the topological degree,
Φ(x) is an interior point of the image Φ(B(x, r)).
It is evident that a mapping which is continuous, open and injective almost
everywhere is injective. Hence, Φ is a homeomorphism.
15) The mapping Φ is a quasi-isometric homeomorphism.
Indeed, Φ is locally Lipschitz and the inequality |DΦ|n (x) ≤ Kn J(x, Φ) holds
a.e. if Φ is a sense-preserving homeomorphism (in other words, if its topological
degree is positive); otherwise, |DΦ|n (x) ≤ −Kn J(x, Φ) a.e. By properties of quasi-
conformal mappings [Mo], the inverse mapping Φ−1 = Ψ is also quasiconformal. As
α−1 ≤ |J(x, Φ)| ≤ γ a.e. in D, the mapping Φ is a quasi-isometric homeomorphism.
16) The domains D0 and Φ(D) are (1, p)-equivalent.
Indeed, on the one hand, by assumption, the operator ϕ∗ : Wp1 (D0 ) → Wp1 (D)
is an isomorphism. On the other hand, the operator Φ∗ : Wp1 (Φ(D)) → Wp1 (D),
Φ∗ (f ) = f ◦Φ, is also an isomorphism. Therefore, Φ∗ −1 ◦ϕ∗ : Wp1 (D0 ) → Wp1 (Φ(D))
is an isomorphism such that g|D0 ∩Φ(D) = (Φ∗ −1 ◦ ϕ∗ )(g)|D0 ∩Φ(D) . ¤

Properties of (1, p)-equivalent domains can be found in [VG3].


Generalizations of some results of the paper for mappings defined on Carnot
groups can be found in [VU5].

References
[CV] V. M. Chernikov and S. K. Vodop0 yanov, Sobolev Spaces and Hypoelliptic Equations. II,
Siberian Adv. Math. 6 (1996), no. 4, 64–96.
[F] H. Federer, Geometric Measure Theory, Springer-Verlag, Berlin, 1969.
[G] F. W. Gehring, Lipschitz mappings and the p-capacity of rings in n-space, Advances in
the Theory of Riemann Surfaces, Ann. Math. Studies, vol. 66, 1971, pp. 175–193.
[GR] V. M. Gol0 dshteı̆n and Yu. G. Reshetnyak, Quasiconformal Mappings and Sobolev Spaces,
Kluwer Academic Publishers, Dordrecht, 1990.
[MM] O. Martio and J. Malý, Luzin’s condition (N) and mappings of the class W 1,n , J. Reine
Angew. Math. 485 (1995), 19–36.
[M1] V. G. Maz0 ya, Classes of Sets and Embedding Theorems of Function Spaces. Some Prob-
lems of the Theory of Elliptic Operators, Ph. D. Thesis, Leningrad Univ., Leningrad, 1961.
[M2] V. G. Maz0 ya, On weak solutions of Dirichlet and Neumann problems, Trudy Moskovskogo
Matematicheskogo Obschestva. T. 20, Moscow Univ., Moscow, 1969, pp. 137–172.
[M3] V. G. Maz0 ya, Sobolev Spaces, Leningrad Univ., Leningrad, 1985.
[MS] V. G. Maz0 ya and T. O. Shaposhnikova, Multipliers in Spaces of Differentiable Functions,
Leningrad Univ., Leningrad, 1986.
[Mo] G. D. Mostow, Quasi-conformal mappings in n-space and the rigidity of hyperbolic space
forms, Inst. Hautes Etudes Sci. Publ. Math. 34 (1968), 53–104.
View publication stats

COMPOSITION OPERATORS OF SOBOLEV SPACES 15

[R1] Yu. G. Reshetnyak, The concept of capacity in the theory of functions with generalized
derivatives, Sibirsk. Mat. Ž. 10 (1969), 1109–1138.
[R2] Yu. G. Reshetnyak, Space Mappings with Bounded Distortion, American Mathematical
Society, Providence, RI, 1989.
[R] A. S. Romanov, On the change of variable in spaces of Bessel and Riesz potentials, Func-
tional Analysis and Mathematical Physics, Academy of Science of USSR, Siberian branch,
Institute of Mathematics, Novosibirsk, 1985, pp. 117–133.
[Vä] J. Väisälä, Lectures on n-dimensional Quasiconformal Mappings, Springer-Verlag, Berlin-
New York, 1971.
[V1] S. K. Vodopyanov, Composition operators of Sobolev spaces, Modern Problems of Function
Theory and its Applications, Saratov, Jan. 2002, Saratov, 2002, pp. 42–43.
[V2] S. K. Vodopyanov, Taylor Formula and Function Spaces, Novosibirsk State University,
Novosibirsk, 1988.
[V3] S. K. Vodopyanov, Weighted Sobolev spaces and the theory of mappings, Tezisy dokladov
Vsesouznoi matematicheskoi shkoly “Potential Theory”. Katciveli, June 26 — July 3, 1991,
Kiev: Institute of Mathematics AN Ukrainy, 1991, p. 7.
[V4] S. K. Vodopyanov, Geometric Properties of Function Spaces with a Generalized Smooth-
ness. Dr. Sc. Thesis, Sobolev Institute of Mathematics, Novosibirsk, 1992.
[V5] S. K. Vodopyanov, P-differentiability on Carnot groups in different topologies and related
topics, Proceedings on Analysis and Geometry (S. K. Vodopyanov, ed.), Sobolev Institute
Press, Novosibirsk, 2000, pp. 603–670.
[V6] S. K. Vodopyanov, Monotone functions and quasiconformal mappings on Carnot groups,
Siberian Math. J. 37 (1996), 1113–1136.
[V7] S. K. Vodopyanov, Differentiability of maps of Carnot group of Sobolev spaces, Sb. Math.
194 (2003), 857–877.
[V8] S. K. Vodopyanov, Lp -theory of potential and quasiconformal mappings on homogeneous
groups, Sovrem. Probl. Geom. Analiz, (Trudy Inst, Mat.), vol. 14, Nauka, Novosibirsk,
1989, pp. 45–89.
[V9] S. K. Vodopyanov, Mappings of homogeneous groups and embeddings of function spaces,
Siberian Math. J. 30 (1989), 685-698.
[V10] S. K. Vodopyanov, Topological and geometric properties of mappings with an integrable
Jacobian in Sobolev classes. I, Siberian Math. J. 41 (2000), 19–39.
[VG1] S. K. Vodopyanov and V. M. Gol0 dšteı̆n, Functional characterizations of quasi-isometric
mappings, Sibirsk. Mat. Ž. 17 (1976), 768–773.
[VG2] S. K. Vodopyanov and V. M. Gol0 dšteı̆n, Quasiconformal mappings and spaces of functions
with the first generalized derivatives, Sibirsk. Mat. Ž. 17 (1976), 515–531.
[VG3] S. K. Vodopyanov and V. M. Gol0 dšteı̆n, A test of the removability of sets for L1p spaces
of quasiconformal and quasi-isomorphic mappings, Sibirsk. Mat. Ž. 18 (1977), 48–68.
[VGR] S. K. Vodopyanov, V. M. Gol’dshtein, and Yu. G. Reshetnyak, The geometric properties
of functions with generalized first derivatives, Uspekhi Mat. Nauk 34 (1979), 17–65.
[VU1] S. K. Vodopyanov and A. D. Ukhlov, Substitution operators in Sobolev spaces, Dokl. Akad.
Nauk 386 (2002), 730–734.
[VU2] S. K. Vodopyanov and A. D. Ukhlov, Superposition operators in Sobolev spaces, Russian
Math. (Iz. VUZ) 46 (2002), no. 10, 9–31.
[VU3] S. K. Vodopyanov and A. D. Ukhlov, Sobolev spaces and (P, Q)-quasiconformal mappings
of Carnot groups, Siberian Math. J. 39 (1998), 665–682.
[VU4] S. K. Vodopyanov and A. D. Ukhlov, Approximately differentiable transformations and
the change of variables on nilpotent groups, Siberian Math. J. 37 (1996), 62–78.
[VU5] S. K. Vodopy0 anov and A. D. Ukhlov, Set functions and their applications in the theory
of Lebesgue and Sobolev spaces. I, Siberian Adv. Math. 14 (2004), no. 4, 1–48.

Sobolev Institute of Mathematics, Academician Koptug pr., 4, 630090 Novosi-


birsk
E-mail address: vodopis@math.nsc.ru

You might also like