Selcuk Journal of
Applied Mathematics
Sel¸cuk J. Appl. Math.Vol. 4, No. 1, pp. 3–24, 2003
On solvability of one elliptic system in
weighted Sobolev spaces
Lina N. Buldygerova
Department of Mathematics, Novosibirsk State University, Pirogova 2, 630090 Novosibirsk, Russia; e-mail: b lina@gorodok.net
Received: March 11, 2003
Summary. In this article we consider an elliptic system in R3 and
prove a theorem on unique solvability in a special scale of weighted Sobolev spaces.
Key words: elliptic systems, weighted Sobolev spaces, uncondi-tional solvability
2000 Mathematics Subject Classification: 35J45, 35C15
1. Introduction
In the present article we consider the following elliptic system of partial differential equations in R3
(1.1)
u++∇u−= f+(x), div u+ = f−(x).
This system arises in various problems of mathematical physics. There are results concerning solvability of the system in various func-tional spaces. In particular, from results by V. N. Maslennikova (see, for example, [1]) it follows that, in the case of f−(x) ≡ 0, the sys-tem (1.1) is uniquely solvable in the Sobolev space
(1.2)
3
1
Wp1(R3)× Wp2(R3)
The research was supported in part by the Russian Foundation for Basic
for p > 3/2. However, the system (1.1) is not unconditionally solvable in (1.2) for p ≤ 3/2. As is shown in the book [2], in the case of f−(x) ≡ 0 for the system (1.1) to be solvable in (1.2) it is necessary and sufficient that
R3
f+(x) dx = 0.
To research solvability of the system (1.1) it is more convenient to use the weighted Sobolev spaces Wp,σl introduced by G. V. Demidenko in [3]. In the present article we study solvability of the system (1.1) in the spaces and show that, for suitable values of σ, the system is unconditionally solvable for any p > 1.
By definition [3], the norm in the space
Wp,σl (Rn), l = (l1, . . . , ln), li ∈ N, 1 < p < ∞, σ ∈ R1, is defined by u(x), Wl p,σ(Rn) = 0≤β/l≤1 (1 + x )−σ(1−β/l)Dβ xu(x), Lp(Rn), where x 2= n i=1 x2li i, β = (β1, . . . , βn), β/l = n i=1 βi/li.
In the isotropic case l1 = . . . = ln = l the norm is equivalent to the
norm
0≤|β|≤l
(1 + |x|)−σ(l−|β|)Dβxu(x), Lp(Rn).
It was proved in [3] that the set of compactly-supported infinitely differentiable functions is everywhere dense in Wp,σl (Rn) for σ ≤ 1.
Hereinafter, we consider the weighted Sobolev spaces Wp,σl (Rn) for 0 < σ < 1.
We give a result on solvability of the system (1.1) in the weighted Sobolev spaces Wp,σl .
Theorem 1. Let 1 < p < ∞, 1 −2p3 < σ < 2p3, 1/p + 1/p = 1. Then
for every right-hand side f+(x) ∈ 3 1 Wp,σ1 (R3), (1 +|x|)2(1−σ)∇ f+(x) ∈ 3 1 Lp(R3), (1 +|x|)2(1−σ)f−(x) ∈ Lp(R3),
the system (1.1) has a unique solution u+(x) ∈ 3 1 Wp,σ1 (R3), u−(x) ∈ Wp,σ2 (R3); moreover, the following inequality holds:
(1.3) 3 j=1 u+j(x), Wp,σ1 (R3) + u−(x), Wp,σ2 (R3) ≤ c(3 j=1 fj+(x), Wp,σ1 (R3) + (1 + |x|)2(1−σ)∇ fj+(x), Lp(R3) +(1 + |x|)2(1−σ)f−(x), Lp(R3))
with some constant c > 0 independent of f+(x) and f−(x). Remark 1. It was proved in [3] that the operator
⎛ ⎜ ⎜ ⎝ 1 0 0 Dx1 0 1 0 Dx2 0 0 1 Dx3 Dx1 Dx2 Dx3 0 ⎞ ⎟ ⎟ ⎠: 3 1 Wp,11 (R3)× Wp,12 (R3) → 3 1 Wp,11 (R3)× Lp(R3) is an isomorphism for 1 < p < 3/2.
2. The scheme of the proof of the theorem
We outline the basic points of the proof of solvability of the system (1.1).
First, by analogy with [2, Chapter 3], we construct a sequence of approximate solutions to the system (1.1) by using S. V. Uspenskii’s integral representation [6] for summable functions
(2.1) ϕ(x) = lim h→0(2π) −n h−1 h v−|α|−1 Rn Rn
e(ix−yvα ξ)G(ξ)ϕ(y) dξdydv, where (2.2) G(ξ) = 2mξ 2mexp(−ξ 2m), ξ 2= n i=1 ξi2/αi, m ∈ N.
We use (2.1) and (2.2) for αi = 1/2, n = 3.
To this end we consider the following system with a parameter
ξ ∈ R3 ⎧ ⎪ ⎨ ⎪ ⎩ v++ L(iξ)v−=f+(ξ), M (iξ)v+=f−(ξ), where L(iξ) = ⎛ ⎝iξiξ12 iξ3 ⎞
⎠, M (iξ) = (iξ1, iξ2, iξ3).
The system is obtained by formal application of the Fourier operator to the system (1.1) under the condition that the vector-functions f+(x) and f−(x) are infinitely differentiable and compactly-support-ed. Since det(M (iξ)L(iξ)) = −|ξ|2 = 0 for ξ ∈ R3\{0}, we obviously
obtain (2.3) v+(ξ) = (I +L(iξ) |ξ|2 M (iξ))f+(ξ) − L(iξ) |ξ|2 f−(ξ), (2.4) v−(ξ) = −M (iξ)|ξ|2 f+(ξ) + |ξ|12f−(ξ). We construct the vector-functions
(2.5) u+k(x) = (2π)−3/2 k 1/k v−1 R3 eixξG(ξvα)v+(ξ) dξ dv, (2.6) u−k(x) = (2π)−3/2 k 1/k v−1 R3 eixξG(ξvα)v−(ξ) dξ dv.
It follows from the definition of the vector-functions v+(ξ) and v−(ξ)
that ⎧ ⎨ ⎩ u+k(x) + L(Dx)u−k(x) = fk+(x), M (Dx)u+k(x) = fk−(x), where fk+(x) = (2π)−3 k 1/k v−|α|−1 R3 R3 e(ix−yvα ξ)G(ξ)f+(y) dξ dy dv,
fk−(x) = (2π)−3 k 1/k v−|α|−1 R3 R3 e(ix−yvα ξ)G(ξ)f−(y) dξ dy dv. By the integral representation (2.1) we can consider the vector-functions (2.5) and (2.6) as an approximate solution to (1.1).
The vector-functions (2.3) and (2.4) can have singularities only at ξ = 0. Therefore, by (2.2) the vector-functions u+k(x) and u−k(x) are infinitely differentiable. It is obvious that we can indicate a natural number m1 such that if m ≥ m1 in (2.2) then these vector-functions
are summable with an arbitrary power p ≥ 1. To this end, it suffices to take m1such that the functions k(ξ) ∈ C∞(R3\{0}), homogeneous
in α of degree q ∈ [−1, 0], satisfy the inequality |
R3
eixξξ 2m1e−ξ2m1k(ξ) dξ| ≤ c(1+x )−(|α|+1), x ∈ R
3, |α| = 3/2.
Henceforth we suppose that m ≥ m1 in (2.2).
In the following section we prove the estimates
(2.7) 3 j=1 u+k,j(x), Wp,σ1 (R3) + u−k(x), Wp,σ2 (R3) ≤ c[3 j=1 ( |γ|=1 (1 + |x|)2(1−σ)Dγ xfj+(x), Lp(R3) + fj+(x), Wp,σ1 (R3)) +f−(x), Lp(R3) + |x|2(1−σ)f−(x), Lp(R3)]
with some constant c > 0 independent of f+(x), f−(x) and k, and also establish the convergence
(2.8) 3 j=1 (u+k1,j(x) − u+k2,j(x)), Wp,σ1 (R3) +(u−k 1(x) − u−k2(x)), Wp,σ2 (R3) → 0, k1, k2 → ∞
under the conditions
1 < p < ∞, 1− 3
2p < σ < 3
2p, 1/p + 1/p
= 1.
By completeness of the space
3
1
from (2.7) and (2.8) it follows that there exists a vector-function u+(x) ∈ 3 1 Wp,σ1 (R3), u−(x) ∈ Wp,σ2 (R3)
that is a solution to the system (1.1) and satisfies the estimate (1.3). Prove the uniqueness of solutions, i. e. show that if u(x) is a solu-tion to the system
(2.9) ⎧ ⎨ ⎩ u++ L(Dx)u−= 0, x ∈ R3, M (Dx)u+= 0, then u(x) = 0 nearly everywhere in R3.
First, we assume that u+(x) and u−(x) are compactly-supported. Applying the Fourier operator to the system (2.9), we obtain the following system of linear equations:
⎧ ⎪ ⎨ ⎪ ⎩ u++ L(iξ)u−= 0, x ∈ R3, M (iξ)u+= 0.
We can rewrite the system in the matrix form L(iξ)u = 0, where L(iξ) = I L(iξ) M (iξ) 0 .
Since det(M (iξ)L(iξ)) = 0 for ξ ∈ R3\{0} thenu(ξ) = 0 nearly every-
where in R3; moreover, by continuityu(ξ) ≡ 0, ξ ∈ R 3. Hence, u(x) is the zero solution, and the system (1.1) has a unique compactly-supported solution. Then, taking into account the estimates from Lemma 2 of [4], for β, |β| = 2, the solution to (1.1) satisfies the inequality
(2.10) Dβxu−(x), Lp(R3) ≤ c(3
j=1
fj+(x), Wp,σ1 (R3) + f−(x), Lp(R3)) with some constant c > 0 independent of f+(x) and f−(x).
Now we consider a more general case. Let u(x) ∈ 3
1 W
1
p,σ(R3)× Wp,σ2 (R3) be a solution to the system (2.9). Prove that u(x) = 0 nearly
everywhere in R3. Since u−(x) ∈ Wp,σ2 (R3) it follows that for every
ε > 0 there is a compactly-supported function u−ε(x) ∈ Wp,σ2 (R3) such that
u−(x) − u−ε(x), Wp,σ2 (R3) ≤ ε.
Consider the function u+ε(x) = −L(Dx)u−ε(x). Obviously, u+ε(x) is compactly-supported too. We have the following system
⎧ ⎨ ⎩
u+ε(x) + L(Dx)u−ε(x) ≡ 0, x ∈ R3,
M (Dx)u+ε(x) ≡ − u−ε(x).
According to (2.10), for β, |β| = 2, the estimate holds: Dβ
xu−ε(x), Lp(R3) ≤ c − u−ε(x) Lp(R3)
with some constant c > 0 independent of ε. Since u+(x) and u−(x) satisfy the system (2.9) then u−(x) ≡ 0. Consequently,
− u− ε(x), Lp(R3) = (u−(x) − u−ε(x)), Lp(R3) ≤ c |γ|=2 Dγx(u−(x) − u−ε(x)), Lp(R3) ≤ cε. Hence, for β, |β| = 2, we have
Dβ
xu−ε(x), Lp(R3) ≤ cε,
Dβ
xu−(x), Lp(R3) ≤ Dβx(u−(x) − u−ε(x)), Lp(R3)
+Dxβu−ε(x), Lp(R3) ≤ ε + cε.
Since the inequality holds for every ε > 0 then Dβxu−(x), Lp(R3) =
0 for |β| = 2. From the estimate [7]
|x|−2σu−(x), Lp(R3) ≤ c|x|2(1−σ) u−(x), Lp(R3)
it follows that u−(x) = 0 nearly everywhere in R3. Then u+(x) = 0
and u(x) = 0 nearly everywhere in R3. Thus, we proved the
3. Estimates for approximate solutions
To obtain estimates for approximate solutions to the system (1.1), we rewrite the vector-functions (2.5) and (2.6) as follows:
u+k(x) = u+,+k (x) + u+,−k (x) = (2π)−3/2 k 1/k v−1 R3 eixξG(ξvα)(v+,+(ξ) + v+,−(ξ)) dξ dv, u−k(x) = u−,+k (x) + u−,−k (x) = (2π)−3/2 k 1/k v−1 R3 eixξG(ξvα)(v−,+(ξ) + v−,−(ξ)) dξ dv, where
v+,+(ξ) = (I +L(iξ)|ξ|2 M (iξ))f+(ξ), v+,−(ξ) = −L(iξ)|ξ|2 f−(ξ),
v−,+(ξ) = −M (iξ)|ξ|2 f+(ξ), v−,−(ξ) = |ξ|12f−(ξ).
Estimates for the higher-order derivatives of the vector-functions u+k(x) and u−k(x) follow from the lemmas [4].
Lemma 1. Let β = (β1, β2, β3),|β| = 1. Then the following estimates
hold: (3.1) 3 j=1 Dβ xu+,+k,j (x), Lp(R3) ≤ c 3 j=1 |γ|=1 Dγ xfj+(x), Lp(R3), (3.2) 3 j=1 Dβ xu+,−k,j (x), Lp(R3) ≤ cf−(x), Lp(R3)
with some constant c > 0 independent of f+(x), f−(x) and k; more-over, (3.3) 3 j=1 Dβxu+k1,j(x) − Dβxuk+2,j(x), Lp(R3) → 0, k1, k2→ ∞.
Lemma 2. Let β = (β1, β2, β3),|β| = 2. Then the following estimates hold: (3.4) Dβxu−,+k (x), Lp(R3) ≤ c 3 j=1 |γ|=1 Dγ xfj+(x), Lp(R3), (3.5) Dβxu−,−k (x), Lp(R3) ≤ cf−(x), Lp(R3)
with some constant c > 0 independent of f+(x), f−(x) and k; more-over,
(3.6) Dβxu−k1(x) − Dxβu−k2(x), Lp(R3) → 0, k1, k2 → ∞.
To obtain estimates for u+k(x) and Dxγu−k(x), |γ| < 2, we consider the functions (3.7) Kh(x) = h−1 h v−1 R3 eixξG(ξvα)k(ξ) dξ dv, 0 < h < 1, where the function k(ξ) ∈ C∞(R3\{0}) is homogeneous in α of degree
q < 0. The derivatives of k(ξ) satisfy the estimates |Dξβk(ξ)| ≤ cβξ q−βα, ξ = 0,
where cβ > 0 are constants.
We will use the following lemma [4].
Lemma 3. Let |α| + q > 0. There is m0 such that if m ≥ m0 in
the definition (2.2) of the function G(ξ) then the following uniform estimate holds:
x |α|+q|K
h(x)| ≤ c, x ∈ Rn,
with some constant c > 0 independent of h.
Henceforth we deal with integrals of the form (3.7) for q ∈ [−1, 0]. Now we establish estimates for the vector-functions u+k(x) and Dγxu−k(x), |γ| < 2. Lemma 4. Let 1 < p < ∞, 1− 3 2p < σ < 3 2p, 1/p + 1/p = 1.
Then the following inequalities hold: (3.8) 3 j=1 |x|−σu+,+ k,j (x), Lp(R3)
≤ c3 j=1 |γ|=1 |x|(1−σ)Dxγfj+(x), Lp(R3), (3.9) 3 j=1 |x|−σu+,−k,j (x), Lp(R3) ≤ c|x|(1−σ)f−(x), Lp(R3)
with some constant c > 0 independent of f+(x), f−(x) and k; more-over, (3.10) 3 j=1 (1 + |x|)−σ(u+ k1,j(x) − u+k2,j(x)), Lp(R3) → 0, k1, k2 → ∞.
Proof. From the definition of the vector-function u+,+k (x) we have (3.11) u+,+k (x) = 3 l=1 (2π)−3/2 k 1/k v−1 R3 eixξG(ξvα) ×(iξl)−1(iξl)v+,+(ξ) dξ dv = 3 l=1 (2π)−3/2 k 1/k v−1 R3 eixξG(ξvα)(iξl)−1
×(I + L(iξ)|ξ|2 M (iξ))Dylf+(ξ) dξ dv = 3 l=1 (2π)−3 k 1/k v−1 R3 R3 ei(x−y)ξG(ξvα)(iξl)−1 ×(I +L(iξ) |ξ|2 M (iξ))dξ Dylf+(y) dy dv. Introduce the notation
Kk,l(x) = k 1/k v−1 R3 eixξG(ξvα)Ql(ξ) dξ dv,
Then u+,+k (x) = 3 l=1 (2π)−3 R3
Kk,l(x − y)Dylf+(y) dy.
The entries of the matrix L(iξ)|ξ|2 M (iξ) are homogeneous in α = (1/2, 1/2, 1/2) of degree 0. Therefore, all entries Ql(ξ) are homogeneous
in α of degree q = −1/2 < 0. Since |α| + q = 3/2 − 1/2 = 1 > 0 then the conditions of Lemma 3 are satisfied. Hence, applying this lemma, we obtain the inequality
3 j=1 |x|−σu+,+ k,j (x), Lp(R3) ≤ c3 j=1 3 l=1 x qσ R3 x − y −|α|−q|Dylfj+(y)| dy, Lp(R3) = c 3 j=1 3 l=1 |x|qσ R3 x − y −|α|−qy q(1−σ) ×y q(σ−1)|Dylfj+(y)| dy, Lp(R3). Taking into account the inequality
3 i=1 |yi|(1−σ)/|α|≤ cy 1−σ, 0≤ σ ≤ 1, we have 3 j=1 |x|−σu+,+k,j (x), Lp(R3) ≤ c3 j=1 3 l=1 R3 3 i=1 |xi|σq/|α||xi− yi|−1−q/|α||yi|q/|α|(1−σ) ×y q(σ−1)|D ylfj+(y)| dy, Lp(R3) = c 3 j=1 3 l=1 ( R1 R1 R1 3 i=1 1/|xi|−σq/|α|1/|xi− yi|1+(q/|α|)
×1/|yi|q/|α|(σ−1)Fl,j(y) dy, Lp(R1)pdx2dx3)1/p = c 3 j=1 3 l=1 ( R1 R1 R1 R1 [ R1 1/|x1|−σq/|α|1/|x1− y1|1+(q/|α|)
×1/|y1|q/|α|(σ−1)Fl,j(y) dy1] 2 i=1 1/|xi|−σq/|α|1/|xi− yi|1+(q/|α|) ×1/|yi|q/|α|(σ−1)dy2dy3, Lp(R1)pdx2dx3)1/p = c 3 j=1 3 l=1 ( R1 R1 R1 R1 KFl,j(x1, y2, y3) ×2 i=1 1/|xi|−σq/|α|1/|xi− yi|1+(q/|α|) ×1/|yi|q/|α|(σ−1)dy2dy3, Lp(R1)pdx2dx3)1/p.
Applying Minkowski’s inequality, we obtain
3 j=1 |x|−σu+,+k,j (x), Lp(R3) ≤ c3 j=1 3 l=1 ( R1 R1 [ R1 R1 KFl,j(x1, y2, y3), Lp(R1) ×2 i=1 1/|xi|−σq/|α|1/|xi− yi|1+(q/|α|) ×1/|yi|q/|α|(σ−1)dy2dy3]pdx2dx3)1/p. As σ < 2p3 then −qσ/|α| < −q/p < 1/p. From the conditions of the lemma we find that
σ − 1 > − 3
2p, q(σ − 1)/|α| < −q/p
< 1/p.
Since
−qσ/|α| < 1/p, q(σ − 1)/|α| < 1/p, −q/|α| > 0, using the Hardy–Littlewood inequality [5], we have
3 j=1 |x|−σu+,+ k,j (x), Lp(R3) ≤ c3 j=1 3 l=1 ( R1 R1 [ R1 R1 Fl,j(x1, y2, y3), Lp(R1)
×2 i=1 1/|xi|−σq/|α|1/|xi− yi|1+(q/|α|) ×1/|yi|q/|α|(σ−1)dy2dy3]pdx2dx3)1/p = c 3 j=1 3 l=1 ( R1 R1 R1 Fl,j(x1, y2, y3), Lp(R1) ×2 i=1 1/|xi|−σq/|α|1/|xi− yi|1+(q/|α|) ×1/|yi|q/|α|(σ−1)dy2dy3, Lp(R1)pdx3)1/p.
Applying Minkowski’s inequality, we obtain
3 j=1 |x|−σu+,+ k,j (x), Lp(R3) ≤ c3 j=1 3 l=1 ( R1 [ R1 R1 Fl,j(x1, y2, y3), Lp(R1) ×1/|x2|−σq/|α|1/|x2− y2|1+(q/|α|)1/|y2|q/|α|(σ−1)dy2, Lp(R1) ×1/|x3|−σq/|α|1/|x3− y3|1+(q/|α|)1/|y3|q/|α|(σ−1)dy3]pdx3)1/p.
Using the Hardy–Littlewood inequality two times, we have
3 j=1 |x|−σu+,+ k,j (x), Lp(R3) ≤ c3 j=1 3 l=1 ( R1 [ R1 Fl,j(x1, x2, y3), Lp(R1), Lp(R1) ×1/|x3|−σq/|α|1/|x3− y3|1+(q/|α|)1/|y3|q/|α|(σ−1)dy3]pdx3)1/p = c 3 j=1 3 l=1 R1 Fl,j(x1, x2, y3), Lp(R1), Lp(R1) ×1/|x3|−σq/|α|1/|x3− y3|1+(q/|α|)1/|y3|q/|α|(σ−1)dy3, Lp(R1) ≤ c3 j=1 3 l=1 Fl,j(x1, x2, x3), Lp(R1), Lp(R1), Lp(R1)
= c 3 j=1 3 l=1 ( R1 Fl,j(x), Lp(R1), Lp(R1)pdx3)1/p = c 3 j=1 3 l=1 ( R1 R1 Fl,j(x), Lp(R1)pdx2dx3)1/p = c 3 j=1 3 l=1 Fl,j(x), Lp(R3) = c 3 j=1 3 l=1 x q(σ−1)Dxlfj+(x) Lp(R3).
Since q = −1/2 and x 1/2 ∼ |x| it follows that
3 j=1 |x|−σu+,+k,j (x), Lp(R3) ≤ c 3 j=1 3 l=1 |x|(1−σ)Dxlfj+(x) Lp(R3). Prove (3.9). From the definition of the vector-function u+,−k (x) we obtain u+,−k (x) = (2π)−3/2 k 1/k v−1 R3 eixξG(ξvα)(−L(iξ) |ξ|2 )f−(ξ) dξ dv = (2π)−3 k 1/k v−1 R3 R3 ei(x−y)ξG(ξvα)(−L(iξ) |ξ|2 ) dξ f−(y) dy dv. All entries of the matrix L(iξ)|ξ|2 are homogeneous in α = (1/2, 1/2, 1/2) of degree q = −1/2 < 0. Therefore, as above we have the estimate
3 j=1 |x|−σu+,− k,j (x), Lp(R3) ≤ c R3 3 i=1 |xi|qσ/|α||xi− yi|−1−q/|α||yi|q(1−σ)/|α|
×y q(σ−1)|f−(y)| dy, Lp(R3).
Applying the Hardy–Littlewood inequality three times, we obtain the estimate (3.9): 3 j=1 |x|−σu+,− k,j (x), Lp(R3) ≤ cx q(σ−1)f−(x), Lp(R3) ≤ c|x|(1−σ)f−(x), Lp(R3).
To prove the convergence (3.10), we have to prove that (3.13) (1 + |x|)−σ(u+,+k1,j(x) − u+,+k2,j(x)), Lp(R3) → 0, (3.14) (1 + |x|)−σ(u+,−k1,j(x) − u+,−k2,j(x)), Lp(R3) → 0
for every j = 1, 2, 3 as k1, k2 → ∞. Obviously, it suffices to establish
(3.13) and (3.14) for compactly-supported f+(x).
Prove (3.13). By the representation (3.11) and the notation (3.12) for the matrix Ql(ξ) = (Qj,il (ξ)), it suffices to consider the function
wk(x) = (2π)−3 k 1/k v−1 R3 R3 ei(x−y)ξG(ξvα)Qj,il (ξ) dξ Dylfi+(y) dy dv and prove the convergence
(3.15) (1 + |x|)−σ(wk1(x) − wk2(x)), Lp(R3) → 0, k1, k2 → ∞. We have wk1(x) − wk2(x) = (2π)−3 k1 1/k1 v−1 × R3 R3 ei(x−y)ξG(ξvα)Qj,il (ξ) dξ Dylfi+(y) dy dv −(2π)−3 k2 1/k2 v−1 R3 R3 ei(x−y)ξG(ξvα)Qj,il (ξ) dξ Dylfi+(y) dy dv = (2π)−3 1/k 2 1/k1 v−1 R3 R3 ei(x−y)ξG(ξvα)Qj,il (ξ) dξ Dylfi+(y) dy dv +(2π)−3 k1 1/k2 v−1 R3 R3 ei(x−y)ξG(ξvα)Qj,il (ξ) dξ Dylfi+(y) dy dv −(2π)−3 k1 1/k2 v−1 R3 R3 ei(x−y)ξG(ξvα)Qj,il (ξ) dξ Dylfi+(y) dy dv −(2π)−3 k2 k1 v−1 R3 R3 ei(x−y)ξG(ξvα)Qj,il (ξ) dξ Dylfi+(y) dy dv
= (2π)−3 1/k 2 1/k1 v−1 R3 R3 ei(x−y)ξG(ξvα)Qj,il (ξ) dξ Dylfi+(y) dy dv −(2π)−3 k2 k1 v−1 R3 R3 ei(x−y)ξG(ξvα)Qj,il (ξ) dξ Dylfi+(y) dy dv. Using Minkowski’s inequality, we obtain
(1 + |x|)−σ(wk1(x) − wk2(x)), Lp(R3) ≤ | 1/k 2 1/k1 v−1(1 + |x|)−σ R3 R3 ei(x−y)ξG(ξvα)Qj,il (ξ) dξ ×Dylfi+(y) dy, Lp(R3) dv| +| k2 k1 v−1(1 + |x|)−σ R3 R3 ei(x−y)ξG(ξvα)Qj,il (ξ) dξ ×Dylfi+(y) dy, Lp(R3) dv| = Ik11,k2 + Ik21,k2.
Consider the first summand. As the functions Qj,il (ξ) are homoge-neous in α of degree q = −1/2 < 0 and (1 + |x|)−σ≤ 1, from Young’s inequality we obtain Ik11,k2 ≤ | 1/k 2 1/k1 v−1 R3 R3 ei(x−y)ξG(ξvα)Qj,il (ξ) dξ ×Dylfi+(y) dy, Lp(R3) dv| ≤ | 1/k 2 1/k1 v−1 R3
eixξG(ξvα)Qj,il (ξ) dξ, L1(R3) dv|Dylfi+(y), Lp(R3)
(change of variables: s = ξvα, z = x/vα) =| 1/k 2 1/k1 v−1−qdv| R3
eiszG(s)Qj,il (s) dξ, L1(R3)Dylfi+(y), Lp(R3). Recalling that q < 0, we find that
Consider the second summand Ik21,k2. Using the inequalities x − y (1 + x )−1≤ a(1 + y ), |x|−σ ≤ ax qσ, q = −1/2 < 0, we have Ik21,k2 ≤ | k2 k1 v−1(1 + x )qσ R3 R3 ei(x−y)ξG(ξvα)Qj,il (ξ) dξ ×Dylfi+(y) dy, Lp(R3) dv| ≤ c| k2 k1 v−1 R3 x − y qσ| R3 ei(x−y)ξG(ξvα)Qj,il (ξ) dξ| ×(1 + y )−qσ|Dylfi+(y)| dy, Lp(R3) dv|.
As the functions Qj,il (ξ) are homogeneous in α of degree q = −1/2 < 0, from Young’s inequality we obtain
Ik21,k2 ≤ c| k2 k1 v−1x qσ R3 eixξG(ξvα)Qj,il (ξ) dξ, Lp(R3) dv| ×(1 + y )−qσ|D ylfi+(y)|, L1(R3) (change of variables: s = ξvα, z = x/vα) = c| k2 k1 v−1−|α|/p−q+qσdv|z qσ R3 eiszG(s)Qj,il (s) dξ, Lp(R3) ×(1 + y )−qσ|D ylfi+(y)|, L1(R3). Since 1 +|α|/p+ q(1 − σ) > 1 + q(σ − 1) + q(1 − σ) = 1 and the functions fi+(y) are compactly-supported, we find that
Ik21,k2 → 0, k1, k2→ ∞.
The above arguments yield (3.15) and consequently (3.13). Simi-larly, we can prove (3.14).
Lemma 5. Let β = (β1, β2, β3), α = (1/2, 1/2, 1/2), βα < 1, 1 < p < ∞, 1− 3 2p < σ < 3 2p, 1/p + 1/p = 1.
Then the following inequalities hold:
(3.16) x −σ(1−βα)Dβxu−,+k (x), Lp(R3) ≤ c1 3 j=1 |x|1−σDxjfj+(x), Lp(R3) +c2 3 j=1 |x|2(1−σ)D xjfj+(x), Lp(R3), (3.17) x −σ(1−βα)Dβxu−,−k (x), Lp(R3) ≤ c(|x|1−σf−(x), Lp(R3) + |x|2(1−σ)f−(x), Lp(R3))
with constants c1, c2 > 0 independent of f+(x), f−(x) and k; more-over,
(3.18) (1 + x )−σ(1−βα)(Dβxu−k1(x) − Dxβu−k2(x)), Lp(R3) → 0, k1, k2 → ∞.
Proof. From the definition of u−,+k (x) we have Dβxu−,+k (x) = (2π)−3/2 k 1/k v−1 R3
eixξG(ξvα)(iξ)β(−M (iξ)
|ξ|2 )f+(ξ) dξ dv = (2π)−3/2 3 j=1 k 1/k v−1 R3 eixξG(ξvα)−(iξ)β |ξ|2 Dyjfj+(ξ) dξ dv = (2π)−3 3 j=1 k 1/k v−1 R3 R3 ei(x−y)ξG(ξvα)−(iξ) β |ξ|2 dξ)Dyjfj+(y) dy dv = (2π)−3 3 j=1 ( R3 k 1/k v−1 R3 ei(x−y)ξG(ξvα)−(iξ) β |ξ|2 dξ dv Dyjfj+(y) dy).
Introduce the notation KQ,k(x) = k 1/k v−1 R3 eixξG(ξvα)Q(ξ) dξ dv,
where Q(ξ) = −(iξ)|ξ|2β. Then
Dβxu−,+k (x) = (2π)−3 3 j=1 R3
KQ,k(x − y)Dyjfj+(y) dy.
Since the function Q(ξ) is homogeneous in α of degree q = βα−1 < 0, we obtain |α| + q = 3/2 + βα − 1 ≥ 1/2 > 0. Hence, using Lemma 3, we have x −σ(1−βα)Dxβu−,+k (x), Lp(R3) ≤ c3 j=1 x qσ R3
x−y −|α|−qy q(1−σ)y q(σ−1)|D
yjfj+(y)| dy, Lp(R3) ≤ c3 j=1 R3 3 i=1 |xi|qσ/|α||xi− yi|−1−q/|α|
×|yi|q(1−σ)/|α|y q(σ−1)|Dyjfj+(y)| dy, Lp(R3).
As σ < 2p3 then
−qσ/|α| < −q/p < 1/p. From the conditions of the lemma we find that
σ − 1 > − 3
2p, q(σ − 1)/|α| < −q/p
< 1/p.
Since
−qσ/|α| < 1/p, q(σ − 1)/|α| < 1/p, −q/|α| > 0,
by the Hardy–Littlewood inequality, we have x −σ(1−βα)Dβ xu−,+k (x), Lp(R3) ≤ c3 j=1 x (1−σ)(1−βα)D xjfj+(x), Lp(R3) and the estimate (3.16)
Prove (3.17). From the definition of the vector-functions u−,−k (x) we obtain Dxβu−,−k (x) = (2π)−3/2 k 1/k v−1 R3 eixξG(ξvα)(iξ)β |ξ|2 f−(ξ) dξ dv = (2π)−3 k 1/k v−1 R3 R3 ei(x−y)ξG(ξvα)(iξ)β |ξ|2 dξ f−(y) dy dv = (2π)−3 R3
KQ,k(x − y)f−(y) dy,
where KQ,k(x) = k 1/k v−1 R3 eixξG(ξvα)Q(ξ) dξ dv, Q(ξ) = (iξ)β |ξ|2 .
Since the function Q(ξ) is homogeneous in α of degree q = βα−1 < 0, |α| + q = 3/2 + βα − 1 ≥ 1/2 > 0, then, as above, by Lemma 3 we have the inequality
x −σ(1−βα)Dxβu−,−k (x), Lp(R3) = (2π)−3x qσ
R3
KQ,k(x − y)f−(y) dy, Lp(R3)
≤ cx qσ R3
x − y −|α|−qy q(1−σ)y q(σ−1)|f−(y)| dy, L p(R3) ≤ c R3 3 i=1 |xi|qσ/|α||xi− yi|−1−q/|α|
|yi|q(1−σ)/|α|y q(σ−1)|f−(y)| dy, Lp(R3).
Therefore, by the Hardy–Littlewood inequality, we arrive at (3.17): x −σ(1−βα)Dβxu−,−k (x), Lp(R3) ≤ x q(σ−1)|f−(x)|, Lp(R3)
≤ c(|x|1−σf−(x), Lp(R3) + |x|2(1−σ)f−(x), Lp(R3)). The convergence (3.18) can be established in the same way as (3.10).
Taking into account Lemmas 1, 2, 4 and 5, for the approximate solutions we obtain the following estimates
u−k(x), Wp,σ2 (R3) ≤ c[ 3 j=1 ( |γ|=1 Dxγfj+(x), Lp(R3) +(|x|1−σDxjfj+(x), Lp(R3) + |x|2(1−σ)Dxjfj+(x), Lp(R3)) +f−(x), Lp(R3) + |x|1−σf−(x), Lp(R3) +|x|2(1−σ)f−(x), Lp(R3)], 3 j=1 u+ k,j(x), Wp,σ1 (R3) ≤ c[3 j=1 ( |γ|=1 |x|(1−σ)Dγ xfj+(x), Lp(R3) + Dγxfj+(x), Lp(R3)) +f−(x), Lp(R3) + |x|1−σf−(x), Lp(R3)] for p ∈ (1, ∞) and σ ∈ (1 −2p3,2p3 ). Since |x|1−σ ≤ c(1 + |x|)(1−σ)≤ c(1 + |x|)2(1−σ)
and for the Laplace operator the following inequality holds [7]: |x|−2σu, L p(R3) + |x|−σ∇u, Lp(R3) ≤ c|x|2(1−σ) u, Lp(R3), we have 3 j=1 u+k,j(x), Wp,σ1 (R3) ≤ c[3 j=1 ( |γ|=1 (1 + |x|)2(1−σ)Dxγfj+(x), Lp(R3) + fj+(x), Wp,σ1 (R3)) +f−(x), Lp(R3) + |x|2(1−σ)f−(x), Lp(R3)], u−k(x), Wp,σ2 (R3) ≤ c[3 j=1 ( |γ|=1 (1 + |x|)2(1−σ)Dxγfj+(x), Lp(R3) + fj+(x), Wp,σ1 (R3)) +f−(x), Lp(R3) + |x|2(1−σ)f−(x), Lp(R3)].
Hence, we obtain the estimate (2.7). From Lemmas 1, 2, 4, 5 we have also the convergence (2.8) for p ∈ (1, ∞) and σ ∈ (1 −2p3,2p3).
The theorem is proved.
Acknowledgment. The author is very grateful to Prof. G. V. Demi-denko for statement of the problem and his assistance.
References
1. Maslennikova, V. N. (1968): Lp-estimates and the asymptotic behavior as t
→ ∞ of a solution of the Cauchy problem for a Sobolev system [in Russian],
Trudy Mat. Inst. Steklov. Akad Nauk SSSR,103, 117–141.
2. Demidenko, G. V. and Uspenskii, S. V. (1998): Partial Differential Equations
and Systems Not Solvable with Respect to the Highest-Order Derivative [in
Russian], Nauchnaya Kniga, Novosibirsk; English transl.: Marcel Dekker, New York, 2003.
3. Demidenko, G. V. (1994): On weighted Sobolev spaces and integral operators determined by quasi-elliptic equations, Russian Acad. Sci. Dokl. Math. 49, No. 1, 113–118.
4. Demidenko, G. V. (2001): Isomorphic properties of one class differential op-erators and their applications, Siberian Math. J.,42, No. 5, 865–883. 5. Hardy, G. H., Littlewood, J. E., and Polya, G. (1934): Inequalities, Univeristy
Press, Cambridge.
6. Uspenskii, S. V. (1972): On a representation of functions defined by one class of hypoelliptic operators [in Russian], Trudy Mat. Inst. Steklov. Akad Nauk SSSR,117, 292–299.
7. Demidenko, G. V. (1998): On quasielliptic operators in Rn, Siberian Math. J.,39, No. 5, 884–893.