APPLICATIONES MATHEMATICAE

29,1 (2002), pp. 51–63

Wojciech M. Zajączkowski (Warszawa)

ON IMBEDDING THEOREMS FOR

WEIGHTED ANISOTROPIC SOBOLEV SPACES

Abstract. Using the Il’in integral representation of functions, imbedding

theorems for weighted anisotropic Sobolev spaces in En are proved. By the
weight we assume a power function of the distance from an (n − 2)-dimen-
sional subspace passing through the domain considered.

1. Introduction. The aim of this paper is to show some imbedding

theorems for weighted Sobolev spaces. We introduce the weighted Sobolev
l,l/2
spaces Wp,µ (Ω T ), l = 2k, k ∈ N ∪ {0}, Ω ⊂ En , Ω T = Ω × (0, T ), µ ∈ R,
p ≥ 1, with the norm
 X 1/p
α α0 p pµ
(1.1) kukWp,µ
l,l/2
(Ω T )
= |D ∂
x x0 u| % (x) dx ,
|α|+2α0 ≤l Ω T

where α = (α1 , . . . , αn ) is a multiindex, |α| = α1 +. . .+αn , Dxα = ∂xα11 . . . ∂xαnn

and % = %(x) is the distance from x to either a subspace of Rn or a point.
In this paper we assume that %(x) = dist(x, M ) where M is an (n − 2)-
dimensional subspace of En . To simplify considerations we assume that M
is defined by x1 = x2 = 0. Finally En ispthe n-dimensional Euclidean space.
Therefore, we can assume that %(x) = x21 + x22 .
l,l/2
More precisely we define Wp,µ (Ω T ) as the closure of the set C0∞ (Ω T \M )
in the norm (1.1).
We use the standard anisotropic notation (see [1]). We consider more
general anisotropic Sobolev spaces Wp,µ l
(En+1 ), where l = (l0 , l1 , . . . , ln ),

2000 Mathematics Subject Classification: Primary 46E35.

Key words and phrases: weighted Sobolev spaces, imbedding theorems, anisotropic
Sobolev spaces.
Research supported by the Polish KBN Grant No. 2 P03A-038-16.

[51]
52 W. M. Zajączkowski

with the norm defined as follows:

n
X
(1.2) kukW l = kukLp,µ (En+1 ) + k∂xlii ukLp,µ (En+1 )
p,µ (E
n+1 )

i=0
≡ kukLp,µ (En+1 ) + kukLl ,
p,µ (E
n+1 )

where kukLp,µ (En+1 ) = ( En+1 |u|p %pµ dx)1/p . Comparing (1.1) and (1.2) we
see that for a norm equivalent to (1.1) we have
l0 = l/2, li = l, i = 1, . . . , n.
We also introduce σ = (σ0 , σ1 , . . . , σn ) and σi = 1/li , i = 0, 1, . . . , n. In the
case of the norm (1.1) we have
σ0 = 2/l, σi = 1/l, i = 1, . . . , n.
We use the following integral representation of a function f with inte-
grable l-derivative (see [1, 3]):
rX
n
(1.3) f (x) = frσ (x) + h−|σ| Φi (x/hσ )Dili f (x + y) dy dh,
0 i=0 En+1

where
Pn x = σ(x0 , x1σ,0. . .σ, 1xn ), y = (y0 , y1 , . . . , yn ), x0 = t, y0 = τ , |σ| = σ0 +
i=1 σi , h = (h , h , . . . , h ), y/h = (y0 /h , y1 /h , . . . , yn /h ), Di
σn σ σ0 σ1 σn

denotes the derivative with respect to the ith argument, and

(1.4) frσ (x) = r−|σ| Φ∗ (y/rσ )f (x + y) dy.
En+1

We assume that Φ∗ , Φi ∈ C0∞ (En+1 ), i = 0, 1, . . . , n, have compact supports

in the first coordinate angle and for any α,
(1.5) Dα Φi (x) dx = 0, i = 0, 1, . . . , n, Dα Φ∗ (x) dx = 0.
Moreover, we assume that the supports of Φ∗ , Φi , i = 0, 1, . . . , n, belong to
the “horn” (see [1, 3])
(1.6) R(l, r, ε) = {y : yi > 0, ai > 0, 0 < ai h < yili < (ai + ε)h,
i = 0, 1, . . . , n, 0 < h < r < ∞},
where ε > 0. If li = l, i = 0, 1, . . . , n, the horn R(l, r, ε) becomes the cone
V (r, ε) = {y : yi > 0, ai > 0, 0 < ai h < yi < (ai + ε)h, i = 0, 1, . . . , n, 0 <
h < r < ∞}. For simplicity we shall omit the ε in R(l, r, ε) and V (r, ε).
We define
r X n
(1.7) fε (x) = frσ (x) + dh h−|σ| Φi (y/hσ )Dili f (x + y) dy.
ε i=0 En+1
 Imbedding theorems for Sobolev spaces 53

From [3] we have the estimate

r X
n τ
(ν(k))
(1.8) hτ −1 Φi (y/h) dh ≤ cχ(y, R(l, r)) |yi |li
ε i=0

where χ(y, R(l, r)) is the characteristic function of the horn R(l, r) and
(ν(k))
Φi is defined below (2.3).
For the reader’s convenience we recall some results. From [3] we recall
the following extension of the Calderón–Zygmund theorem.
Lemma 1.1. Let the support of Φ ∈ C0∞ (En ) be in the first coordinate
cube and En Φ(x) dx = 0. Let 1 < p < ∞ and let
r
vεr (x) = dh h−1−|σ| Φ(y/hσ )u(x + y) dy
ε En
for u ∈ Lp (En ), x, y ∈ En , σ ∈ R+ . Then
kvεr kLp (En ) ≤ cp kukLp (En )
and
vεr → v0r in Lp (En ) as ε → 0.
We also need the Hardy inequality
(1.9) k |x|−γ F (x)kLp (E+ ) ≤ ck |x|−γ+1 f (x)kLp (E+ ) , 1 ≤ p ≤ ∞,
∞ x
where γ 6= 1/p, F (x) = f (ξ) dξ for γ < 1/p, F (x) = f (ξ) dξ for γ > 1/p
x 0
and E+ = {x ∈ E : x > 0}.
In the case of isotropic weighted Sobolev spaces similar results are proved
in [4].
From [1, Ch. 2, Sect. 8] we recall
Definition 1.2. We say that a domain Q satisfies that the R(l, r)-horn
condition if there exist K open subdomains Qk and horns Rk (l, r) such that
K
[ K
[
Q= Qk = (Qk + Rk (l, r)).
k=1 k=1

In [1] this property is called the weak R(l, r)-horn condition.

2. Imbedding theorems for p 6= q. First we prove

Theorem 2.1. Assume that f ∈ Wp,α
l
(Q), Q ⊂ En+1 , l = (l0 , l1 , . . . , ln ),
1 < p < q < ∞, α, β ∈ R+ , 0 < li ∈ Z, 0 ≤ νi ∈ Z, i = 0, 1, . . . , n,
l1 = l2 = l∗ ,
  n n
1 1 X 1 X νi 1
(2.1) κ =1− − − − (α − β) ≥ 0,
p q i=0 li i=0 li l∗
54 W. M. Zajączkowski

where Q satisfies the R(l, r)-horn condition. Assume

(2.10 ) α > β.
Then Dν f ∈ Lq,β (Q) and
(2.2) kD ν f kLq,β (Q) ≤ c1 δ κ kf kLl + c2 δ κ−1 kf kLp,α (Q) ,
p,α (Q)

where the constants c1 , c2 do not depend on f , δ ∈ (0, h0 ), h0 = h0 (Q).

Proof. Let x = (x1 , . . . , xn ), x0 = (x1 , x2 ), x00 = (x3 , . . . , xn ). Then we
introduce the cylindrical coordinates (%x , ϕx , x00 ) connected with x, where
%x = |x0 |, x1 = %x cos ϕx , x2 = %x sin ϕx .
Let k ∈ N0 . Then integrating by parts and using the compactness of the
supports of Φ∗ , Φi , i = 0, 1, . . . , n, we obtain from (1.7) the expression
(2.3) D ν fε (x)
∞ ∞ ∞
(ν(k))
=c Φ∗ (y/rσ ) ... f (x + y(t0 )) dt0 dt1 . . . dtk−1 dy
En+1 %y tk−1 t1
r n
X (ν(k))
+ dh h−|σ|−kσ∗ −(σ,ν) Φi (y/hσ )
ε i=0 En+1
∞ ∞ ∞
× ... Dili f (x + y(t0 )) dt0 . . . dtk−1 dy,
%y tk−1 t1

where σ∗ = σ1 = σ2 so that l1 = l2 = l∗ , ν = (ν0 , ν1 , . . . , νn ), ν(k) =

(ν0 , ν1 + k1 , ν2 + k2 , ν3 , . . . , νn ), k1 + k2 = k,
X
Φ(ν(k)) (x) = ck1 k2 (cos ϕx )k1 (sin ϕx )k2 ∂xν00 ∂xν11 +k1 ∂xν22 +k2 ∂xν33 . . . ∂xνnn Φ(x),
k1 +k2 =k

y = (y0 , y1 , . . . , yn ), y(t0 ) = (y0 , t0 cos ϕy , t0 sin ϕy , y3 , . . . , yn ), and (σ, ν) =

σ0 ν0 + σ1 ν1 + . . . + σn νn . Finally ck1 k2 are determined by the relation
X  k  k
k ∂x1 1 ∂x2 2 k1 k2
∂ %x Φ = ck 1 k 2 ∂x1 ∂x2 Φ.
∂%x ∂%x
k1 +k2 =k

Let us introduce the notation

∞ ∞ ∞
(2.4) F (y) = ... f (y(t0 )) dt0 dt1 . . . dtk−1 ,
%y tk−1 t1
∞ ∞ ∞
(2.5) Fi (y) = ... Dili f (y(t0 )) dt0 dt1 . . . dtk−1 .
%y tk−1 t1
 Imbedding theorems for Sobolev spaces 55

Then we can write (2.3) in the following shorter form:

(2.6) D ν fε (x) = c Φ(ν(k)) (y/rσ )F (x + y) dy
En+1
r n
X (ν(k))
+ c dh h−|σ|−kσ∗ −(σ,ν) Φi (y/hσ )Fi (x + y) dy.
ε i=0 En+1

Now we examine the case κ = 0 (see (2.1)). Consider

(ν(k))
(2.7) kD ν fε (x)kLq,β (Q) ≤ c Φ∗ (y/rσ )F (x + y) dy
Lq,β (Q)
En+1
Xn r
(ν(k))
+c dh h−|σ|−kσ∗ −(σ,ν) Φi (y/hσ )

i=0 ε En+1

|x | 0 β
× 0 Fi (x + y) dy
0
≡ M + N,
|x + y | 0 α−k
Lq (Q)
p
where Fi0 (x) = |x0 |α−k Fi (x) and |x0 | = x21 + x22 .
First we estimate N . Using (1.8) we obtain
Xn X n τ

N ≤c χ(y, R(l, h 0 )) |y | li
i
i=0 En+1 i=0

|x0 |β
× 0 F 0
(x + y) dy ≡ N1 ,
|x + y 0 |α−k i
Lq (En+1 )

where τ = −(1 − 1/p + 1/q)|σ| + (α − β − k)σ∗ .

Since q > p > 1 we can assume that 1/p − 1/q = 1 − 1/s and we
also assume also that α − β − k ≤ 1/s. To estimate N1 we apply the one-
dimensional Young inequality
1 1 1
(2.8) kf ∗ gkLq ≤ kgkLs kf kLp for − = 1 − , 1 ≤ p ≤ q ≤ ∞,
p q s
with respect to the variables x0 , x3 , . . . , xn .
Using (2.8) with respect to x0 we find that the kernel in N1 is estimated
as follows:
∞X n τ s 1/s ∞ 1/s Pn
|yi |li dy0 = (|y0 |l0 + a0 )τ s dy0 (a0 = i=1 |yi |li )
0 i=0 0
∞ 1/s  1/s
1/l0 τ sl0 1 1/l y0 =∞
≤ (y0 + a0 ) dy0 = (y0 + a0 0 )τ sl0 +1 y =0
0
τ sl0 + 1 0

 1/s
1 τ +1/(sl0 )
= − a0 ≡ K1 ,
τ sl0 + 1
56 W. M. Zajączkowski

where we have used the fact that

   
1 1 1 1 1 1
τ sl0 + 1 = − + + + ... + + (α − β − k) sl0 + 1
s l0 l1 l2 ln l∗
 
1 1 1 1 sl0
= −l0 + + + ... + + (α − β − k)
l1 l2 l3 ln l∗
 
2l0 sl0 1 1
≤− + (α − β − k) − l0 + ... +
l∗ l∗ l3 ln
 
l0 1 1
≤ − − l0 + ... + < 0,
l∗ l3 ln
since l1 = l2 = l∗ , α − β − k ≤ 1/s.
Using (2.8) with respect to x3 we see that the kernel K1 is estimated by
∞X
n τ s+1/l0 1/s ∞ 1/s
li 1/l3 (τ s+1/l0 )l3
|yi | dy3 ≤ (y3 + a3 ) dy3
0 i=1 0
 1/s
1 [(τ s+1/l0 )l3 +1]/(sl3 )
= − a3 ,
τ sl3 + l3 /l0 + 1
Pn
where a3 = |y1 |l1 + |y2 |l2 + i=4 |yi |li ,
 
l3 1 1
τ sl3 + + 1 = τ s + + l3
l0 l0 l3
   
1 1 1 1 1 s 1 1
= − + + + + ... + + (α − β − k) + + l3
l0 l1 l2 l3 ln l∗ l0 l3
  
1 1 1 1
≤ − − + + ... + l3 < 0,
l∗ l4 l5 ln
where we have used the fact that l1 = l2 = l∗ , α − β − k ≤ 1/s.
Continuing the calculations we obtain
Xn

N1 ≤ c χ(y 0 − x0 , R(l∗ , h0 ))|x0 − y 0 |−2/s+α−β−k

i=0 E2

|x0 |β 0

0
× 0 α−k Gi (y ) dy ≡ N2 ,
|y | Lq (E2 )

where Gi (y 0 ) = ( Q |Fi0 (y)|p dy0 dy3 . . . dyn )1/p .

To estimate N2 we use Remark 3.1 from [2], which we formulate in the
following form. Assume that
(2.9) p1 ≥ 1, q1 ≥ 1, 1/p1 + 1/q1 ≥ 1, δ < 2/q10 , λ < 2/p01 + 2/q10 ,
 Imbedding theorems for Sobolev spaces 57

where r 0 is dual to r, so 1/r + 1/r 0 = 1. Then

  p0 1/p01
 1 g(y) dy  1
(2.10)  dx ≤ ckgkLq1 (E2 ) ,
 |x|2/p01 +2/q10 −(λ+δ) |x − y|λ |y|δ 
E 2 |y|≤|x|
  q10 1/q10
 1 f (x) dx 
(2.11)   dy ≤ ckf kLp1 (E2 ) .
 |y|δ λ 2/p
|x − y| |x| 1
0 +2/q 0 −(λ+δ) 
1
E2 |y|≤|x|

Inserting p01 := q, q1 := p, δ := δ1 into (2.10) yields

  q 1/q
 1 g(y) dy 
(2.12)  dx ≤ ckgkLp (E2 ) .
 |x|2/s−(λ+δ1 ) |x − y|λ |y|δ1 
E2 |y|≤|x|

Inserting p1 := p, := q, x := y, y := x, f := g, δ := δ2 in (2.11) implies

q10
  q 1/q
 1 g(y) dy 
(2.13)   dx ≤ ckgkLp (E2 ) .
 |x|δ2 |x − y| λ |y| 2/s−(λ+δ 2 ) 
2 E |x|≤|y|

The conditions (2.9) imply for (2.12) the restrictions

(2.14) δ1 < 2(1 − 1/p), λ < 2/s.
For case (2.13) conditions (2.9) yield
(2.15) δ2 < 2/q, λ < 2/s.
Comparing (2.12) with N2 we see that
2 2
(2.16) λ = − (α − β − k), δ1 = α − k, −β = − (λ − δ1 ),
s s
where the last condition follows from the first two, and comparing (2.13)
with N2 we obtain
2 2
(2.17) λ = − (α − β − k), δ2 = −β, α − k = − (λ + δ2 ),
s s
where the last condition also follows from the first two.
However to estimate N2 we need estimate (2.13). Therefore we have to
impose the following restrictions:
(2.18) −β < 2/q, 2/s − (α − β − k) < 2/s,
where the last inequality implies
(2.19) α − k > β.
For k = 0 the condition gives
(2.20) α > β,
so the case α = β cannot be considered. Since the first condition of (2.18) is
trivial and (2.19) with k = 0 is less restrictive we see that (2.20) is exactly
(2.10 ).
58 W. M. Zajączkowski

Now using Remark 3.1 from [2] we obtain the estimate

n
X n
X
(2.21) N1 ≤ c kFi0 kLp (En+1 ) ≤ c kDili f kLp,α (En+1 ) ,
i=0 i=0
where in the second inequality we exploited the Hardy inequality.
Similarly using (1.8) we have
(2.22) M ≤ ckF 0 kLp (En+1 ) ≤ ckf kLp,α (En+1 ) .
From (2.21) and (2.22) we obtain (2.2) with δ = c, κ = 0 and Q = En+1
after letting ε → 0 (see [3, p. 139]).
To show (2.2) with parameter δ and κ > 0 we exploit the considerations
from [1, Ch. 3].
To obtain (2.2) for Q bounded we apply the standard considerations with
a partition of unity. This concludes the proof.
It seems that condition (2.10 ) is artificial. It follows from applying [2] to
estimate the integral N2 . However we do not know how to estimate N2 in a
different way.
From (2.1)0 we see that the case
(2.23) 0≥α≥β
is not included in Theorem 1. Hence we need
Corollary 2. Assume that Q is bounded and satisfies the R(l, h0 )-horn
condition, f ∈ Wp,α
l
(Q), and
(2.24) α ≤ β.
Take α > β such that
0
 X
n n
0 1 1 1 X νi 1
(2.25) κ =1− − − − (α0 − β) ≥ 0.
p q i=0
li i=0 li l∗
Then D f ∈ Lq,β (Q) and
ν

kD ν f kLq,β (Q) ≤ cεκ kf kLl + cεκ

0 0
−1
(2.26) kf kLp,α (Q)
p,α (Q)

for all ε ∈ (0, h0 ), where c does not depend on f and ε.

Proof. Since Q is bounded we have f ∈ Wp,α
l
0 (Q) and

(2.27) kf kW l ≤ ckf kW l .
p,α0
(Q) p,α (Q)

Using Theorem 1 we obtain (2.26). This concludes the proof.

The results of this paper, especially Corollary 2, are necessary for the
proof of the existence of global regular special solutions to Navier–Stokes
equations (see [5]).
 Imbedding theorems for Sobolev spaces 59

3. Imbedding theorems for p = q. First we prove

Theorem 3.1. Assume that f ∈ Wp,α l
(Q), Q ⊂ En+1 and Q satisfies the
R(l, h0 )-horn condition, l = (l0 , l1 , . . . , ln ), 1 < p < ∞, α, β ∈ R+ , α ≥ β,
0 < li ∈ Z, 0 ≤ νi ∈ Z, i = 0, 1, . . . , n, l1 = l2 = l∗ and
X n
νi 1
(3.1) κ ≡1− − (α − β) ≥ 0.
i=0 i
l l∗
Then Dν f ∈ Lp,β (Q) and
(3.2) kD ν f kLp,β (Q) ≤ c1 hκ kf kLl + c2 hκ−1 kf kLp,α (Q) ,
p,α (Q)

where c1 , c2 do not depend on f and h ∈ (0, h0 ), h0 = h0 (Q).

Proof. We consider the case α − β = k + γ, k ∈ N0 , γ ∈ [0, 1).
First we examine the case κ = 0 and γ = 0. Then we write (2.6) in the
form
(ν(k))
(3.3) Dν fε (x) = c φ∗ (y/rσ )F (x + y) dy
En+1
r n
X (ν(k))
+ c dh h−1−|σ| Φi (y/hσ )Fi (x + y) dy.
ε i=0 En+1

Estimating (3.3) implies

(ν(k))
(3.4) kD ν fε kLp,β (Q) ≤ c Φ∗ (y/rσ )F (x + y) dy
Lp,β (Q)
En+1
n r
X
(ν(k))
+c dh h−1−|σ| Φi (y/hσ )Fi (x + y) dy ≡ M0 + M1 .
Lp,β (Q)
i=0 ε En+1
Using estimates for integral operators we have
M0 ≤ ckf kLp,α (Q) .
Next we examine
Xn r
−1−|σ| (ν(k))
M1 ≤ c h Φi (y/hσ )Fi0 (x + y) dy dh
Lp (Q)
i=0 ε
n r
X
−1−|σ| (ν(k)) σ |x | − |x + y |
0 β 0 0 α−k

+c h Φ (y/h ) F 0
(x + y) dy dh
i
|x + y |
0 0 α−k i
Lp (Q)
i=0 ε
≡ M3 + M4 ,
where Fi0 (x) = |x0 |α−k Fi (x).
Lemma 1.1 and the Hardy inequality (1.9) yield
X n n
X
M3 ≤ c 0
kFi kLp (Q) ≤ c kDili f kLp,α (Q) .
i=0 i=0
60 W. M. Zajączkowski

Using (1.8) in M4 implies

Xn X
n −|σ|

M4 ≤ c χ(y, R(l, h 0 )) |y | lj
j
i=0 En+1 j=0
 
 |x0 |β  0

× 0 − 1 F (x + y) dy ≡ M5 .
|x + y 0 |α−k  i
Lp (Q)

Applying the one-dimensional Young inequality

(3.5) kf ∗ gkLp ≤ kgkL1 kf kLp
to M5 with respect to the variables x0 , x3 , . . . , xn , we obtain
n
X
M5 ≤ c χ(y 0 , R(l∗ , h0 ))|y 0 |−2

i=0 E2  
 |x0 |β  0 0

× 0  0 0
− 1kFi (x + y )kLp (En−1 ) dy
|x + y |0 α−k
Lp (Q)
n
X
≤c χ(y 0 − x0 , R(l∗ , h0 ))|x0 − y 0 |−2

i=0 E2
 0β 
 |x |  0 0
  0
×  0 α−k − 1kFi (y )kLp (En−1 ) dy ≡ M6 .
|y | Lp (Q)

Introducing the polar coordinates x0 = (% cos ϕx , % sin ϕx ), y 0 = (η cos ϕy ,

η sin ϕy ) we obtain
n ∞ 2π ∞ 2π
X  1
M6 ≤ c % d% dϕx  η dη dϕy 2 2 − 2%η cos(ϕ − ϕ )
i=0 0 0 % 0
% + η x y
 β  p 1/p
 %  
×  α−k − 1F 0i (η cos ϕy , η sin ϕy ) ≡ M7 ,
η
where F 0i = kFi0 (y 0 )kLp (En−1 ) .
Using the Young inequality (3.5) with respect to ϕx and the expression
π
dϕ 2π
= 2
−π
%2 + η2 − 2%η cos ϕ % − η2

we obtain
n ∞ ∞   p 1/p
X  1  %β  
M7 ≤ c % d%  1 − Fei (η)η dη 
0
≡ M8 ,
|% 2 − η 2 |  η α−k  
i=0 0 %

2π
where Fei0 (η) = ( 0
|F i0 (η cos ϕy , η sin ϕy )|p dϕy )1/p .
 Imbedding theorems for Sobolev spaces 61

Introducing a new variable λ by η = λ% in the inner integral in M8 and

using the generalized Minkowski inequality (see [1, Ch. 1]) we obtain
∞ Xn
λ1−2/p −(α−k)
M8 ≤ c (1 − λ ) dλ kFi0 kLp (En+1 )
1
(λ + 1)(λ − 1) i=0
n
X
≤c kDili f kLp,α (En+1 ) ,
i=0

where in the last inequality the Hardy inequality (1.9) was also used.
Let us consider the case κ = 0 and α − β = k + γ, γ > 0. Then M1 takes
the form
Xn r
−1−|σ|+γ/l∗ (ν(k))
M10 = c dh h Φ i (y/h σ
)F i (x + y) dy .
Lp,β (Q)
i=0 ε En+1

In view of (1.8) we have

Xn X
n −|σ|+γ/l∗

0
M1 ≤ c χ(y, R(l, h 0 )) |y | lj
j
i=0 En+1 j=0

|x0 |β
× 0 Fi (x + y) dy
0
≡ M20 .
|x + y |0 α−k
Lp (Q)

Applying the one-dimensional Young inequality (3.4) with respect to x0 , x3 ,

. . . , xn gives
Xn

0
M2 ≤ c χ(y 0 , R(l∗ , h0 ))|y 0 |−2+γ

i=0 E2

|x0 |β 0 0 0

0
× 0 kF i (x + y )k L (E n−1 ) dy
≡ M30 .
|x + y 0 |α−k p
Lp (E2 )

Changing variables implies

Xn
0 −2+γ |x |
0 β
0
M3 = c χ(y 0
− x 0
, R(l , h 0 ))|x 0
− y | F 0 0
(y ) dy 0
,
∗
|y 0 |α−k i
i=0 2 E Lp (E2 )

0
where = kFi0 (y 0 )kLp (En−1 ) . Introducing the polar coordinates x0 =
F i (y 0 )
(% cos ϕx , % sin ϕx ), y 0 = (η cos ϕ0 , η sin ϕ0 ), ϕ = ϕx − ϕ0 , in M30 yields
n ∞ 2π ∞ 2π
X  dϕx
0 0
M3 ≤ c % d% dϕ  η dη
i=0 0 0 % 0
(% + η − 2%η cos(ϕx − ϕ0 ))1−γ/2
2 2

p 1/p
%β 0 0

0 
× α−k Fi (η cos ϕ , η sin ϕ ) ≡ M40 .
η
62 W. M. Zajączkowski

Using the fact that the integral over ϕx can be made independent of ϕ0 and
then applying the Minkowski inequality (see [1, Ch. 1]) we obtain
Xn ∞ ∞ 2π
 dϕ %β
M40 ≤ c % d%  η dη
i=0 0 % 0
(%2 + η 2 − 2%η cos ϕ)1−γ/2 η α−k
2π 1/p p 1/p
× 0 0 0 p
|Fi (η cos ϕ , η sin ϕ )| dϕ 0  ≡ M50 .

0

Changing variables η = λ% gives

n ∞ ∞ 2π
X  dϕ
0
M5 = c 
% d%  λ dλ λ−(α−k)
i=0 0 1 0
(λ2
+ 1 − 2λ cos ϕ) 1−γ/2

2π 1/p p 1/p

× 0 0 0 p
|Fi (λ% cos ϕ , λ% sin ϕ )| dϕ 0  .

0

Applying the Minkowski inequality (see [1, Ch. 1]) yields

n ∞
X 2π
dϕ
M50 ≤ c λ dλ λ−(α−k)
i=0 1 0
(λ2 + 1 − 2λ cos ϕ)1−γ/2
∞ 2π 1/p
× % d% dϕ0 |Fi0 (λ% cos ϕ0 , λ% sin ϕ0 )|p ≡ M60 .
0 0

Introducing a new variable σ = λ%, dσ = λ d%, implies

n ∞
X 2π
dϕ
M60 = c λ dλ λ−(α−k)−2/p
i=0 1 0
(λ2 + 1 − 2λ cos ϕ)1−γ/2
∞ 2π 1/p
× σ dσ dϕ0 |Fi0 (σ cos ϕ0 , σ sin ϕ0 )|p
0 0
n ∞
X 2π
dϕ
≤c λ dλ λ−(α−k)−2/p kFi kLp (En+1 ) ≡ M70 .
i=0 1 0
(λ2 + 1 − 2λ cos ϕ)1−γ/2

Passing to the Cartesian coordinates z1 = λ cos ϕ, z2 = λ sin ϕ, z 0 = (1, 0),

z = (z1 , z2 ) we write M70 in the form
n
X |z|−(α−k)−2/p
M70 = c kFi kLp (En+1 ) dz .
i=0
|z − z 0 |2−γ
|z|≥1

For |z| large the integral converges for α − k + 2/p > γ.

 Imbedding theorems for Sobolev spaces 63

For |z| in a neighbourhood of 1 we can show convergence by passing to

the coordinates with origin at z 0 . Finally we have
X n
0
M7 ≤ c kFi kLp (En+1 ) .
i=0
Applying now the remarks from [3] we can let ε → 0 to obtain (3.2).
Let κ > 0. Using the considerations from [1, Ch. 3] we obtain (3.2) with
a parameter δ > 0.
For Q bounded we apply the standard considerations with a partition of
unity. This concludes the proof.

References

[1] O. V. Besov, V. P. Il’in and S. M. Nikol’skiı̆, Integral Representations of Functions

and Imbedding Theorems, Nauka, Moscow, 1975 (in Russian).
[2] V. P. Il’in, Some integral inequalities and applications in the theory of differentiable
functions of several variables, Mat. Sb. 54 (1961), 331–380 (in Russian).
[3] A. F. Kocharli, Some weighted imbedding theorems for domains with nonsmooth
boundary, Trudy Mat. Inst. Steklov. 131 (1974), 128–146 (in Russian).
[4] W. M. Zajączkowski, On theorem of embedding for weighted Sobolev spaces, Bull.
Polish Acad. Sci. Math. 33 (1985), 115–121.
[5] —, On global special solutions to Navier–Stokes equations with boundary slip condi-
tions in a cylindrical domain. Existence, preprint 616, Inst. Math., Polish Acad. Sci.,
2001.

Institute of Mathematics Institute of Mathematics and Operations Research

Polish Academy of Sciences Military University of Technology
Śniadeckich 8 S. Kaliskiego 2
00-950 Warszawa, Poland 00-908 Warszawa, Poland
E-mail: wz@impan.gov.pl

Received on 12.10.2000;
revised version on 7.11.2001 (1554)