Volume1, Number 1(2010), 54 72
ON INFINITE DIFFERENTIABILITY OF SOLUTIONS
OF NONHOMOGENEOUS ALMOST
HYPOELLIPTIC EQUATIONS
H.G. Ghazaryan, V.N. Margaryan
Communiated by T.V. Tararykova
Keywords and phrases: hypoellipti operator (polynomial),almost hypoellipti
operator (polynomial),weighted Sobolevspaes.
Mathematis Subjet Classiation: 12E10.
Abstrat. A linear dierential operator
P (D)
with onstant oeients is alledalmost hypoellipti if all derivatives
P (ν) (ξ)
of the harateristi polynomialP (ξ)
an be estimated abovevia
P (ξ)
. In this paperitis proved that allsolutionsof theequation
P (D)u = f
wheref
and all its derivatives are square integrable with aertain exponential weight, whih are square integrable with the same weight, are
alsosuhthatalltheirderivativesare squareintegrablewith thisweight,ifandonly
if the operator
P (D)
is almost hypoellipti.1 Introdution
After in 1950's L. Hormander introdued the onept of a hypoellipti dierential
equation
P (D)u = f
all distributional solutionsu
of whih with an innitelydierentiableright-handside
f
areinnitelydierentiable(see[15℄,[16℄),aproblem arose of nding additional assumptions on solutionsu
of more general, non-hypoellipti equationsensuring that these solutions are innitely dierentiable.
In [14℄, [8℄ L. Garding, B. Malgrange and L. Ehrenprie studied the lass of
partiallyhypoelliptiequationsallsolutionsofwhihwithaninnitelydierentiable
right-handside are innitely dierentiable under the a priori assumption that they
are innitely dierentiable with respet to aertain group ofthe variables.
In [3℄ Ya.S. Bugrov onstruted an example of a non-hypoellipti equation, all
solutions of whih in the half-spae are innitely dierentiable provided they are
square integrable inthe half-spae together with some of their derivatives.
In [4℄, [6℄ V.I. Burenkov onsidered the equation
P (D)u = f
in the ylinderΩ = Ω m × E n − m
with0 ≤ m < n
whereΩ m
is an open set inE m
(ifm = 0
then
Ω = E n
) andf
and all its derivatives arem
-loally square integrable onΩ
,i.e. square integrable on
Ω m × E n − m
for all ompatsQ m ⊂ Ω m
(ifm = 0
squareintegrable on
E n
). Neessary and suient onditions onP
were found ensuringthat all solutions
u
of this equation with any suhf
, whih arem
-loally squareintegrableon
Ω
together withsome of theirderivativeare of the samelass asf
(inpartiular are innitely dierentiable).
The lass of suh operators is essentially wider than the lass of hypoellipti
operators.
In [10℄, [17℄ the notionof an almost hypoellipti polynomial wasintroduedand
some suient onditions for almost hypoelliptiity were found in terms of the
homogeneity orders and multipliityof the roots of ertainsubpolynomials.
In [9℄ O.R. Gabrielyan obtained neessary and suient onditions for almost
hypoelliptiptiity of two-dimensional polynomials in terms of multipliity of the
rootsof the ertain homogeneoussubpolynomials.
We use the following standard notations:
N
the set of all natural numbers,N 0 = N ∪ { 0 }
,N 0 n = N 0 × . . . × N 0
the set of alln
-dimensionalmulti-indies,E n
and
R n
then
-dimensional eulidean spaes of points (vetors)x = (x 1 , . . . , x n )
and
ξ = (ξ 1 , . . . , ξ n )
respetively. Forξ ∈ R n
,x ∈ E n
andα ∈ N 0 n
we put| ξ | = p ξ 1 2 + . . . + ξ n 2 , | α | = α 1 + . . . + α n
,ξ α = ξ 1 α 1 ...ξ n α n , D α = D α 1 1 ...D α n n
,whereD j = ∂ξ ∂
j
or
D j = 1 i ∂x ∂ j (j = 1, ...n)
. Finallywe putC n = R n × iR n
.For alinear dierentialoperator with onstant oeients
P (D) = P
α
γ α D α
, letP (ξ) = P
α
γ α ξ α
be its harateristi polynomial (omplete symbol),where the sum extends over a nite olletion of multi-indies(P ) = { α ∈ N 0 n , γ α 6 = 0 }
and letm = ord P = max {| α | , α ∈ (P ) }
.An operator
P (D)
(a polynomialP (ξ))
is alled hypoellipti (see [15℄), if all solutionsu ∈ D ′
(D ′ = D ′ (E n )
isthe set ofdistributions)of theequationP (D)u = f
are innitely dierentiable(belong toC ∞ = C ∞ (E n ))
for allf ∈ C ∞ .
This holds if and only if (see [16℄, Theorem 11.1.1) all solutions
u ∈ D ′
of theequation
P (D)u = 0
are innitely dierentiable.L.Hormanderhasproved(see[16℄,Theorem11.1.1andTheorem11.1.3),thatan
operator
P (D)
ishypoelliptiifandonlyifanyofthefollowingequivalentonditions issatised.1)
Sing Supp u = Sing Supp P (D)u
for any open setΩ ⊂ E n
andu ∈ D ′ ,
2)if
0 6 = ν ∈ N 0 n ,
thenP (ν ) (ξ)/P (ξ) ≡ D ν P (ξ)/P (ξ) → 0
as| ξ | → ∞
,3)
d P (ξ) → ∞
as| ξ | → ∞ ,
whered P (ξ)
is the distane from the pointξ ∈ R n
tothe surfae
{ ζ ; ζ ∈ C n , P (ζ) = 0 }
.The following question naturally arises. Let the symbol
P (ξ)
of the operatorP (D)
satisfy aweaker ondition thanondition 2):| P ν (ξ) | / [1 + | P (ξ) | ] ≤ C < ∞ ∀ ξ ∈ R n , ∀ ν ∈ N 0 n , (1.1)
orlet 3)be replaed by the requirementthat
d P (ξ) ≥ ε > 0 (1.2)
forallsuiently large
ξ ∈ R n
.Whih onditionsshould beimposed ona funtionf ∈ C ∞
and on solutionsu ∈ D ′ (E n )
of the equationP (D)u = f
toensure innitedierentiability of
u
inE n
?Weshall use the followinglasses of operatorsand funtions.
Denition 1.1. An operator
P (D)
(and a polynomialP (ξ))
is alled almosthypoelliptiifthepolynomial
P
satisesoneof(equivalent)onditions(1.1),(1.2).By Lemma 11.1.4 of [16℄ for any polynomial
P (ξ)
there exists a onstantσ 1 = σ 1 (P ) > 0
suh that the following inequality holds:X
| ν | >0
d | P ν | (ξ) | P (ν) (ξ) | ≤ σ 1 | P (ξ) |∀ ξ ∈ R n . (1.3)
On the other hand if a polynomial
P ∈ I n ,
i.e| P (ξ) | → ∞
as| ξ | → ∞
(inthis onnetion we note that in [11℄ - [12℄ neessary and suient onditions in
terms of the oeients of
P
ensuring thatP ∈ I 2
were found), then it is almosthypoelliptiif and only if
ρ P = lim
t →∞ inf
| ξ | = t d P (ξ) > 0 · (1.4)
For any
δ > 0
andm ∈ N,
letL 2,δ ≡ L 2,δ (E n )
be the set of all funtionsu
loallyintegrableon
E n
with nitenorms|| u || L 2,δ =
Z
E n
| u(x) | 2 e − 2δ | x | dx
1/2
(1.5)
and
H δ m = H δ m (E n )
bethe set of allfuntionsu ∈ L 2,δ
withnite norms|| u || H δ m = X
| α |≤ m
|| D α u || L 2,δ . (1.6)
Finally weput
H δ ∞ =
\ ∞ m=1
H δ m ·
It is obvious that
L 2, δ
andH δ m
are Banah spaes andH δ ∞
is aFrehet spae suhthat
H δ ∞ ⊂ C ∞
for anyδ > 0
.Let
N (P, δ) = { u ∈ L 2,δ : P (D)u ∈ H δ ∞ } ,
where
P (D)u
is understoodin the distributionalsense, i.e.u ∈ N (P, δ)
means thatu ∈ L 2,δ
and there existsf ∈ H δ ∞
suh that (here and the sequelR
means
R
E n
) Z
uP ( − D)ϕdx = Z
f ϕdx ¯ ∀ ϕ ∈ C 0 ∞ ,
where
C 0 ∞
is the set of allfuntions inC ∞
with ompat supportThe followingstatement is the main result of the artile.
Theorem. An operator
P ∈ I n
is almost hypoellipti if and only if there exists a numberδ > 0
suhthatN (P, δ) ⊂ H δ ∞
.Remark.The ase
δ = 0
is quitedierent from the aseδ > 0.
In [4,6℄ it isprovedthat
N (P, 0) ⊂ H 0 ∞
if and only if there existc, M > 0
suh that| P (ξ) | ≥ c
for allξ ∈ R n
satisfying| ξ | ≥ M ·
Let
ε > 0
andL 2,δ,ε
be the set of all funtionu
loally integrable onE n
withnite norms
|| u || L 2,δ,ε =
Z
E n
| u(x) | 2 e − 2δ | x | ε dx
1/2
·
Considerthespae
H δ,ε ∞
andthesetsN(P, δ, ε)
obtainedbyreplaingL 2,δ
byL 2,δ,ε
intheabovedenitions.In[7℄V.I.Burenkovproved thatif
ε > 1
thenN (P, δ, ε) ⊂ H δ,ε
if and only if the operator
P
is hypoellipti. An interesting question arises: what happens in the ase0 < ε < 1?
2 Estimates for funtions in
H δ ∞
Lemma 2.1. Let
P ∈ I n
be an almost hypoellipti polynomial, the numberρ P
bedened by formula (1.4) and
σ 1
be the minimal number for whih inequality (1.3)is satised. Then for any
ρ ∈ (0, ρ P )
there exists a numberσ 2 = σ 2 (ρ, P ) > 0
suhthat
X
| α | >0
ρ | α | | P (α) (ξ) | ≤ σ 1 | P (ξ) | + σ 2 ∀ ξ ∈ R n . (2.1)
Proof. Let
ρ ∈ (0, ρ P )
. Then by the denition of the numberρ P
there exists anumber
M = M (ρ) > 0
suhthatd P (ξ) ≥ ρ ∀ ξ ∈ R n ; | ξ | ≥ M. (2.2)
Let usput
σ 2 = σ 2 (ρ, P ) = max
| ξ |≤ M
X
| ν | >0
ρ | ν | | P (ν) (ξ) | ,
then estimate(2.1) follows by estimates (1.3) and (2.2).
For onveniene instead of the weight funtion
e − δ | x |
we onsider the equivalentsmooth weightfuntion
g δ ∈ C ∞
.Weassume that
g ∈ C ∞
is axed positive funtion suh that for someκ > 0 κ − 1 e −| x | ≤ g(x) ≤ κe −| x | ∀ x ∈ E n ·
Moreover, we suppose that for any
α ∈ N 0 n
there exists a numberκ α > 0 (κ 0 = κ)
for whih
| D α g(x) | ≤ κ α 2ptg(x) ∀ x ∈ E n ·
Note, that for example
g
may be onstruted as the regularization (i.e. the averaging) of the funtionH
, dened byH(x) = e −| x |
for| x | > 1
andH(x) = e − 1
for
| x | ≤ 1
by means of a xed nonnegative weight funtionϕ ∈ C 0 ∞
suh thatR ϕ(x)dx = 1
.For
δ > 0
,we setg δ (x) = g(δx)
. Then by the denition ofg
κ − 1 e − δ | x | ≤ g δ (x) ≤ κe − δ | x | ∀ x ∈ E n , (2.3)
| D α g δ (x) | ≤ κ α δ | α | g δ (x) ∀ α ∈ N 0 n , ∀ x ∈ E n . (2.4)
We start with the statements proved in [13℄ (see Lemma 1.1and Lemma 1.2in
[13℄)
Lemma 2.2. Let
m ∈ N, a i , b i ≥ 0 (i = 0, 1, ..., m)
,d > 0
andt > 1
. Thena) if
a 0 = b 0
anda k ≤ b k + d
k − 1
X
j=0
t j a j (k = 1, 2, ..., m), (2.5)
then for
σ 3 = 2[2dt m − 1 + 1] m
X m k=0
a k ≤ σ 3
X m k=0
b k , (2.6)
b) if
a m = b m
andt k a k ≤ t k b k + d X m j=k+1
t j a j (k = 0, 1, ..., m − 1), (2.7)
then
X m k=0
t k a k ≤ X m k=0
(1 + d) k t k b k . (2.8)
Lemma 2.3. If the set
G ⊂ E n
is ontained in the ballS T = { x ∈ E n , | x | ≤ T } ,
then for any
δ > 0
sup
y ∈ G
g δ (x + y) ≤ σ 4 g δ (x) ∀ x ∈ E n , (2.9) sup
y ∈ G | g δ (x + y) − g δ (x) | ≤ σ 5 g δ (x) ∀ x ∈ E n , (2.10)
where the funtion
g δ
is as dened above,σ 4 = κ 2 e δT
andσ 5 = κ 2 (δT )e δT (max
| α | =1 κ α ) √ n
.In the sequel for a xed almost hypoellipti polynomial
P
of orderm
, xedfuntion
g
, and for anyρ ∈ (0, ρ P )
, let positive numbersκ α
,| α | ≤ m
, satisfyinequalities(2.3), (2.4), numbers
σ 1 (ρ), σ 2 (ρ)
, satisfy inequality (2.1) andh = h(g, m) = X
0< | α | <m
κ α
α! , (2.11)
∆ 0 = ∆ 0 (g, m, ρ) = ρ min
0< | α |≤ m
( α!
2κ α
1/ | α |
[(1 + h) m σ 1 ] 1/ | α | )
· (2.12)
Lemma 2.4. Let
P (D)
be an almost hypoellipti operator of degreem
and thenumber
ρ P
be dened by formula (1.4), then it follows that for anyρ ∈ (0, ρ P )
,δ ∈ (0, ∆ 0 (ρ)]
andu ∈ H δ ∞
X
| α | >0
ρ | α | k [P (α) (D)u]g δ k L 2 ≤ 2(1 + h) m [σ 1 (ρ) k [P (D)u]g δ k L 2 +
+σ 2 (ρ) k ug δ k L 2 ]. (2.13)
Proof. First we prove that for any
t > 0
,δ ∈ (0, t]
andu ∈ H δ ∞ X
| α | >0
t | α | k [P (α) (D)u]g δ k L 2 ≤ (1 + h) m X
| α | >0
t | α | k{ P (α) (D)[ug δ ] k L 2 . (2.14)
Put for
j = 1, 2, ..., m a j = X
| α | =j
t | α | k [P (α) (D)u]g δ k L 2 , b j = X
| α | =j
t | α | k P (α) (D)[ug δ ] k L 2 .
It isobvious that
a m = b m
.Then by the Leibnitz formulaitfollows thatfor anyt > 0
andj = 1, 2, ..., m − 1 a j = X
| α | =j
t | α | k [P (α) (D)u k L 2 = X
| α | =j
t | α | {k P (α) (D)[ug δ ] −
− X
| β | >0
1
β! [P (α+β) (D)u]D β g δ k L 2 } ≤ b j +
+ X
| α | =j
t | α | X
| β | >0
1
β! [P (α+β) (D)u]D β g δ k L 2 .
Applying property (2.4) of the funtion
g δ
, we get that for anyδ ∈ (0, t]
a j ≤ b j + X
| α | =j
t | α | X
| β | >0
κ β
β! δ | β | k [P (α+β) (D)u]g δ k L 2 ≤
≤ b j + X m l=j+1
t l X
| γ | =l
k [P (γ) (D)u]g δ k L 2
X
0<ν<γ
κ ν
ν! =
= b j + h X m l=j+1
a j j = 1, ..., m − 1 ·
This means that the numbers
α j , b j j = 1, ..., m
satisfy ondition b) of Lemma2.2, whih proves estimate (2.14). Hene, by Lemma 2.1, inequality (2.14) and the
Parsevalequality we onlude that
X
| α | >0
ρ | α | k [P (α) (D)u]g δ k L 2 ≤ (1 + h) m X
| α | >0
ρ | α | k P (α) (D)[ug δ ] k L 2 =
= (1 + h) m X
| α | >0
ρ | α | k P (α) (ξ)F (ug δ )(ξ) k L 2 ≤
≤ (1 + h) m [σ 1 k P (ξ)F (ug δ ) k L 2 + σ 2 (ρ) k F (ug δ ) k L 2 ] =
= (1 + h) m [σ 1 k P (D)(ug δ ) k L 2 + σ 2 (ρ) k ug δ k L 2 ] , (2.15)
where
F
is the Fourier transform.Using the Leibnitz formula and estimates (2.4), for the rst summand in the
right-handside weobtain
k P (D)(ug δ ) k L 2 ≤ k [P (D)u]g δ k L 2 + X
| α | >0
1
α! k [P (α) (D)u]D α g δ k L 2 ≤
≤ k [P (D)u] g δ k L 2 + X
| α | >0
κ α
α! δ | α | k [P (α) (D)u]g δ k L 2 . (2.16)
It follows from(2.15) - (2.16) that
X
| α | >0
ρ | α | k [P (α) (D)u]g δ k L 2 ≤ (1 + h) m σ 1 [ k [P (D)u]g δ k L 2 +
+ X
| α | >0
κ α
α! δ | α | k [P (α) (D)u]g δ k L 2 ] + (1 + h) m σ 2 (ρ) k ug δ k L 2 .
Then it islear that
X
| α | >0
ρ | α | k [P (α) (D)u]g δ k L 2 [1 − (1 + h) m σ 1
κ α
α!
δ ρ
| α |
] ≤
≤ (1 + h) m σ 1 k [P (D)u]g δ k L 2 + (1 + h) m σ 2 k ug δ k L 2 .
Sine
1 − (1 + h) m σ 1
κ α
α!
δ ρ
| α |
≥ 1
2 ∀ α ∈ N 0 n , 0 < | α | ≤ m
for any
δ ∈ (0, ∆ 0 (ρ))
, this leads toinequality (2.13).For any
ρ ∈ (0, ρ P )
we denote by∆ 1 = ∆ 1 (ρ, g)
the greatest of numbersδ ∈ (0, ∆ 0 )
for whih1 − σ 1
"
κ α
α!
δ ρ
| α |
+ X
0<γ<α
κ γ
γ!
δ ρ
| γ | #
≥ 1
2 (2.17)
for any
α ∈ N 0 n
;0 6 = | α | < m
.Lemma 2.5. Let
P ∈ I n
be an almost hypoellipti operator andρ ∈ (0, ρ P )
, thenfor any
k ∈ N 0
there exist positive numbersA k,j
,j = 0, 1, ..., k
, andB k
suh thatfor all
δ ∈ (0, ∆ 1 ]
andu ∈ H δ ∞ X
| β |≤ k
X
| α | >0
ρ | α | k [D β P (α) (D)u]g δ k L 2 ≤
≤ X k
j=0
A k,j
X
| β | =j
k [D β P (D)u]g δ k L 2 + B k k ug δ k L 2 . (2.18)
Proof. We prove the result by indution in
k
. Fork = 0
inequality (2.18) followsby (2.13) with the onstants
A 0,0 = 2(1 + h) m σ 1
andB 0 = 2(1 + h) m σ 2
.Assuming thatinequalities (2.18)hold for
k ≤ r
,let usprovethat they hold fork = r + 1
.Bythe indutiveassumption itfollows that forany
δ ∈ (0, ∆ 1 )
andu ∈ H δ ∞
thefollowing inequality holds:
X
| β |≤ r+1
X
| α | >0
ρ | α | [D β P (α) (D)u]g δ
L 2 ≤
≤ X
| β | =r+1
X
| α | >0
ρ | α | [D β P (α) (D)u]g δ
L 2 +
+ X r
j=0
A r,j X
| β | =j
[D β P (D)u]g δ
L 2 + B r k ug δ k L 2 . (2.19)
If
β, ν ∈ N 0 n
and0 6 = ν ≤ β
, then| β − ν | ≤ r
. Hene, by the Leibnitz formulaand estimates (2.4) we onlude that for some positive onstants
C 1 = C 1 (g, ρ)
,C 2 = C 2 (g, ρ)
and for anyδ ∈ (0, ρ]
the followinginequalitiesholdX
| β | =r+1
X
| α | >0
ρ | α | [D β P (α) (D)u]g δ
L 2 = X
| β | =r+1
X
| α | >0
ρ | α |
D β P (α) (D)[ug δ ] −
− X
γ+ν 6 =0,ν ≤ β
C β ν
γ! [D β − ν P (α+γ) (D)u]D γ+ν g δ L 2
≤
≤ X
| β | =r+1
X
| α | >0
ρ | α | D β P (α) (D)[ug δ ]
L 2 +
+ X
| β | =r+1
X
| α | >0
ρ | α | X
γ 6 =0
1 γ!
[D β P (α+γ) (D)u]D γ g δ ]
L 2 +
+ X
| β | =r+1
X
| α | >0
ρ | α | X
0 6 =ν ≤ β
X
γ
C β ν γ!
[D β − ν P (α+γ) (D)u]D γ+ν g δ ]
L 2 ≤
≤ X
| β | =r+1
X
| α | >0
ρ | α | D β P (α) (D)[ug δ ]
L 2 +
+ X
| β | =r+1
X
| α |≥ 2
ρ | α | [D β P (α) (D)u]g δ ]
L 2
"
X
0 6 =ν<α
δ ρ
| γ | κ γ
γ!
# +
+C 1
X
| β | =r+1
X
| α | >0
ρ | α | X
0 6 =ν ≤ β
X
γ 6 =0
C β ν
γ! κ ν+γ δ | ν+γ | [D β − ν P (α+γ) (D)u]D γ+ν g δ ]
L 2 ≤
≤ X
| β | =r+1
X
| α | >0
ρ | α | D β P (α) (D)[ug δ ]
L 2 +
+ X
| β | =r+1
X
| α |≥ 2
ρ | α | [D β P (α) (D)u]g δ ]
L 2
"
X
0 6 =ν<α
δ ρ
| γ | κ γ
γ!
# +
+C 2
X
| β |≤ r
X
| α |≥ 1
ρ | α | [D β P (α) (D)u]g δ ]
L 2 ·
From this and by the indutive assumption we get that for any
δ ∈ (0, ρ]
andu ∈ H δ ∞
X
| β | =r+1
X
| α | >0
ρ | α | [D β P (α) (D)u]g δ
L
2 ≤
≤ X
| β | =r+1
X
| α | >0
ρ | α | D β P (α) (D)[ug δ ]
L 2 +
+ X
| β | =r+1
X
| α |≥ 2
ρ | α | [D β P (α) (D)u]g δ ]
L 2
"
X
0 6 =ν<α
δ ρ
| γ | κ γ
γ!
# +
+ X r
j=0
C 1 A r,j
X
| β | =j
[D β P (α) (D)u]g δ ]
L 2 + C 1 B r k ug δ k L 2 . (2.20)
Let
ρ ∈ (0, ρ P )
and letM = M(ρ)
be the minimal of numbers satisfying theinequality(2.2) and
σ 6 = max
| ξ |≤ M
X
| β | =r+1
X
| α | >0
ρ | α | | ξ β P (α) (ξ) | .
Toevaluatetherstsummandintheright-handsideof(2.20)weusetheParseval
equality,estimates (2.4) and Lemma 2.1. We onlude that for any
ρ ∈ (0, ρ P ) X
| β | =r+1
X
| α | >0
ρ | α | D β P (α) (D)[ug δ ]
L 2 =
= X
| β | =r+1
X
| α | >0
ρ | α | ξ β P (α) (ξ)F (ug δ )]
L 2 ≤
≤ σ 1
X
| β | =r+1
ξ β P (ξ)F (ug δ )
L 2 + σ 6 k F (ug δ ) k L 2 =
= σ 1
X
| β | =r+1
D β P (D)[ug δ ]
L 2 + σ 6 k u g δ k L 2 . (2.21)
Toevaluatetherstsummandintheright-handsideof(2.21)weusetheLeibnitz
formula and estimates (2.3). Then by indutive assumption we onlude that for
somepositiveonstant
C 3 = C 3 (g, ρ)
,andforanyδ ∈ (0, ρ]
the folowinginequalities holdX
| β | =r+1
|| D β P (D)[ u g δ ] || L 2 ≤ X
| β | =r+1
|| [ D β P (D) u ] g δ || L 2 +
+ X
| β | =r+1
X
γ+ν 6 =0; ν ≤ β
C β ν
γ! || [D β − ν P (γ) (D)u ] D γ+ν g δ || L 2 ≤
≤ X
| β | =r+1
|| [ D β P (D) u ] g δ || L 2 +
+ X
| β | =r+1
X
| γ | >0
κ γ
γ ! δ | γ | || [D β P (γ) (D)u ] g δ || L 2 +
+ X
| β | =r+1
X
0<ν<β
(δ | ν | κ ν ) · C β ν || [D β − ν P (D)u] g δ || L 2 +
+ X
| β | =r+1
X
| γ | >0
X
| ν | >0
C β ν γ!
κ γ+ν
γ! δ | γ+ν | || [D β − ν P (γ) (D)u] g δ || L 2 ≤
≤ X
| β | =r+1
|| [ D β P (D) u ] g δ || L 2 +
+ X
| β | =r+1
X
| γ | >0
κ γ γ!
δ ρ
| γ |
ρ | γ | || [D β P (γ) (D)u ] g δ || L 2 +
+C 3
X r
j=0
A j,r
X
| β | =j
|| [D β P (D)u ] g δ || L 2 + B r || ug δ || L 2
. (2.22)
Applyingestimates(2.20)(2.22)weobtainthatforany
δ ∈ (0, ρ]
andu ∈ H δ ∞ X
| β | =r+1
X
| α | >0
ρ | α | || D β P (α) (D)[ u g δ ] || L 2 ≤
≤ σ 1
X
| β | =r+1
|| [ D β P (D) u ] g δ || L 2 +
+σ 1
X
| β | =r+1
X
| γ | >0
κ γ
γ!
δ ρ
| γ |
ρ | γ | || [D β P (γ) (D)u ] g δ || L 2 +
+ X
| β | =r+1
X
| α |≥ 2
ρ | α | || [ D β P (α) (D) u ] g δ || L 2
X
0<γ<α
κ γ
γ!
δ ρ
| γ |
+
+[σ 1 (C 2 + C 3 ) + C 1 ] X r
j=0
A j,r
X
| β | =j
|| [D β P (D)u ] g δ || L 2 +
+ { [σ 1 (C 2 + C 3 ) + C 1 ] B r + σ 6 } || ug δ || L 2 . (2.23)
Applying estimates (2.19) and (2.23) we obtain that for any
δ ∈ (0, ρ]
andu ∈ H δ ∞
X
| β | =r+1
X
| α | >0
ρ | α | || D β P (α) (D)[ u g δ ] || L 2 ≤ σ 1
X
| β | =r+1
|| [ D β P (D) u ] g δ || L 2 +
+[σ 1 (C 2 + C 3 ) + C 1 + 1]
X r j=0
A j,r
X
| β | =j
|| [D β P (D)u ] g δ || L 2 +
+[σ 1 (C 2 + C 3 ) + C 1 + 1] B r + σ 6 } || u g δ || L 2 ]+
+σ 1
X
| β | =r+1
X
| γ | >0
κ γ
γ!
δ ρ
| γ |
ρ | γ | || [D β P (γ) (D)u ] g δ || L 2 +
+σ 1
X
| β | =r+1
X
| α |≥ 2
ρ | α | || [ D β P (α) (D) u ] g δ || L 2
X
0<γ<α
κ γ
γ!
δ ρ
| γ |
·
Estimate (2.18) immediately follows by this inequality and by inequality (2.17)
inthe denition of
∆ 1 (ρ, g)
with the onstantsA j,r+1 = 2[σ 1 (C 2 + C 3 ) + C 1 + 1] A j,r j = 0, 1, ..., r; A r+1,r+1 = 2σ 1 ; B r+1 = 2 { [σ 1 (C 2 + C 3 ) C 1 + 1] B r + σ 6 } ,
whihin turn ompletes the proof of the lemma.
3 Density of smooth funtions in weighted Sobolev spaes
In this setion we onsider almost hypoellipti operators in weighted Sobolev
funtion spaes
H δ m = H δ m (E n )
andH δ ∞ = H δ ∞ (E n ) .
We begin with a generalresulton linear dierentialoperators with onstant oeients.
Lemma 3.1. For any linear dierential operator
P (D)
with onstant oeientsand for any
δ > 0
the setH δ ∞
is dense inN (P, δ) = { u ∈ L 2,δ ; P (D)u ∈ H δ ∞ }
withrespet to the topology, indued by the seminorms
k u k P,k,δ = k ug δ k L 2 + X
| α |≤ k
k [D α (P (D)u)]g δ k L 2 , k = 0, 1, . . . .
Proof. Assuming that
S 1 = { x ∈ E n : | x | < 1 } , ϕ ∈ C 0 ∞ (S 1 ), R
ϕ(x)dx = 1, u ∈ L 2, δ
andε > 0,
wedenoteϕ ε (x) = ε − n ϕ(x/ε),
and setu ε (x) = u ∗ ϕ ε = Z
u(x − y)ϕ ε (y)dy = ε − n Z
u(x − y)ϕ(y/ε)dy ·
The funtion
u ε
is alled a regularization (or molliation) ofu
(for theproperties of
u ε
see for example [5℄, Chapter 1, or[1℄, Chapter 2,Setion 17).Firstweprovethat
u ε ∈ H δ ∞ ·
Weobserve thatthe followingtakesplae foranyk ∈ N 0 ,
using property (2.10) of the funtiong δ
and Young's inequalityX
| α |≤ k
|| (D α u ε ) g δ || L 2 = X
| α |≤ k
Z
u(x − y) D α ϕ ε (y) g δ (x)dy
L 2
≤
≤ X
| α |≤ k
"
|| (ug δ ) ∗ D α ϕ ε || L 2 +
Z
u(x − y) [ g δ (x − y) − g δ (x) ]D α ϕ ε (y)dy
L 2
#
≤
≤ X
| α |≤ k
[ || (u g δ ) ∗ D α ϕ ε || L 2 + σ 5 (ε) || | u g δ | ∗ | D α ϕ ε | || L 2 ] ≤
≤ (1 + σ 5 (ε)) X
| α |≤ k
|| ug δ || L 2 || D α ϕ ε || L 1 =
= (1 + σ 5 (ε)) || ug δ || L 2
X
| α |≤ k
ε −| α | || D α ϕ || L 1 < ∞·
Sine
k ∈ N 0
is arbitrary, itfollows thatu ε ∈ H δ ∞ ·
To omplete the proof it remainsto show that as
ε → 0
|| u ε − u || P, k, δ → 0 · (3.1)
Let a funtion
v ∈ L 2, loc (E n )
and a linear dierential operatorQ(D)
satisfythe followingondition :
Q(D)v ∈ L 2, loc (E n ) ·
Then (see [2℄, 6.3(2))Q(D)v ε (x) = [Q(D)v] ε (x)
forallx ∈ E n
and by the ontinuityinthe mean offuntionsu ∈ L 2
|| v ε − v || L 2 → 0
as
ε → 0 .
Therefore,usingproperty(2.10)oftheweightfuntion
g
andYoung'sinequalitythe followingholds for any
k ∈ N 0
|| u ε − u || P, k, δ = || [ u ε − u ]g δ || L 2 + X
| α |≤ k
|| [ D α (P (D)u ε ) − D α (P (D)u) ]g δ || L 2 =
= || [ u ε − u ]g δ || L 2 + X
| α |≤ k
|| { [D α (P (D)u) ] ε − D α (P (D)u) } g δ || L 2 ≤
≤ || (u g δ ) ε − (u g δ ) || L 2 + || (u g δ ) ε − (u ε g δ ) || L 2 +
+ X
| α |≤ k
|| [ (D α P (D)u) g δ ] ε − (D α P (D)u) g δ || L 2 +
+ X
| α |≤ k
|| [ (D α P (D)u) g δ ] ε − (D α P (D)u) ε g δ || L 2 ≤
≤ || (u g δ ) ε − (u g δ ) || L 2 + X
| α |≤ k
|| [ (D α (P (D)u) g δ ] ε − [ D α (P (D)u) ] g δ || L 2 +
+
Z
u(x − y) [g δ (x − y) − g δ (x)]ϕ ε (y)dy
L 2
+
+ X
| α |≤ k
Z
[D α P (D)u](x − y) [ g δ (x − y) − g δ (x) ]ϕ ε (y)dy
L 2
≤
≤ || (u g δ ) ε − (u g δ ) || L 2 + X
| α |≤ k
|| [(D α (P (D)u) g δ ] ε − [ D α (P (D)u) ] g δ || L 2 +
+σ 5 (ε) || | ug δ | ∗ | ϕ ε | || L 2 + σ 5 (ε) X
| α |≤ k
|| [ | D α (P (D)u | ∗ | ϕ ε | || L 2 ≤
≤ || (u g δ ) ε − (u g δ ) || L 2 + X
| α |≤ k
|| [(D α (P (D)u) g δ ] ε − [ D α (P (D)u) ] g δ || L 2 +
+σ 5 (ε) {|| ug δ || L 2 + X
| α |≤ k
|| [ D α (P (D)u) ] g δ || L 2 }·
Sine
σ 5 (ε) → 0
asε → 0,
the right-hand side tends to zero asε → 0.
Thus,(3.1) is true and the proof isomplete.
4 Proof of the main result
Theorem 4.1. Let
P ∈ I n
be an almost hypoellipti operator,ρ ∈ (0, ρ P )
andδ ∈ (0, ∆ 1 (ρ))
, thenN (P, δ) ⊂ H δ ∞
.Proof.Let
u ∈ N (P, δ)
,ϕ ∈ C 0 ∞ (S 1 )
,ϕ ≥ 0
,R
ϕ(x)dx = 1
,ε > 0
,ϕ ε (x) = ε − n ϕ( x ε )
and let
u ε = u ∗ ϕ ε
be a regularization ofu
. Then (see the proof of Lemma 3.1)u ε ∈ H δ ∞
and by the Lemma 2.5for anyδ ∈ (0, ∆ 1 (ρ))
andk ∈ N 0
X
| β |≤ k
X
| α | >0
ρ | α | [D α P (α) (D)u ε ]g δ
L 2 ≤
≤ X k
j=0
A k,j
X
| β | =j
[D β P (D)u ε ]g δ
L 2 + B k k u ε g δ k L 2 .
Notethat by Lemma 3.1
X k j=0
A k,j
X
| β | =j
|| [ D β P (D)u ε ] g δ || L 2 + B k || u ε g δ || L 2 →
→ X k
j=0
A k,j
X
| β | =j
|| [ D β P (D)u ] g δ || L 2 + B k || u g δ || L 2
as
ε → 0
. Hene, there existnumbersε 0 > 0
andC = C(ε 0 ) > 0
suh thatX
| β |≤ k
X
| α | >0
ρ | α | || [ D α P (α) (D)u ε ] g δ || L 2 ≤ C ∀ ε ∈ (0, ε 0 ).
On the other hand, sine
P (α 0 ) (ξ) = const 6 = 0
for some multi-indexα 0 6 = 0,
by this inequality itfollows that for some onstant
C 1 > 0 X
| β |≤ k
|| (D β u ε ) g δ || L 2 ≤ C 1 ∀ ε ∈ (0, ε 0 ). (4.1)
This means that the set
{ u ε ; ε ∈ (0, ε 0 ) }
is uniformly bounded inH δ ∞
.Therefore, using Lemmas 2.5 and 2.6 we get that for some onstant
C 2 > 0
andany
ε 1 , ε 2 ∈ (0, ε 0 )
C 2
X
| β |≤ k
|| D β (u ε 1 − u ε 2 ) g δ || L 2 ≤
≤ X
| β |≤ k
X
| α | >0
ρ | α | || D β P (α) (D)(u ε 1 − u ε 2 ) g δ || L 2 ≤
≤ X k
j=0
A k,j
X
| β | =j
|| [ D β P (D)u ε 1 − D β P (D)u ε 2 ].g δ || L 2 +
+B k || (u ε 1 − u ε 2 ) g δ || L 2 ≤
≤ X
| β |≤ k
X
| α | >0
ρ | α | || [ D β P (α) (D)u ε 1 − D β P (α) (D)u) ] g δ || L 2 +
+ X
| β |≤ k
X
| α | >0
ρ | α | || [ D β P (α) (D)u ε 2 − D β P (α) (D)u) ] g δ || L 2 +
+B k || (u ε 1 − u) g δ || L 2 + B k || (u ε 2 − u) g δ || L 2 ·
Hene
X
| β |≤ k
|| D β (u ε 1 − u ε 2 ) g δ || L 2 → 0 as ε 1 , ε 2 → 0 + ·
From thisand (3.2) weget thatforany bounded set
G ⊂ E n
andanyk ∈ N 0
there exists a number
C 3 = C 3 (G, k) > 0
suh thatX
| β |≤ k
|| (D β u ε ) || L 2 (G) ≤ C 3 ∀ ε ∈ (0, ε 0 ),
and
X
| β |≤ k
|| D β (u ε 1 − u ε 2 ) || L 2 (G) → 0 as ε 1 + ε 2 → 0 ·
Sine the Sobolev spae
H k ≡ W 2 k
is omplete and|| u ε − u || L 2 (G) → 0
asε → 0
(see the proof of Lemma 3.1) it follows thatu ∈ H k ·
Moreover, sineG ⊂ E n
andk ∈ N 0
are arbitrary,we haveu ∈ H loc ∞ (E n ) ·
Passing in (4.1) to the limit as
ε → 0
we onlude thatu ∈ H δ k
for anyk ∈ N 0 .
Heneu ∈ H δ ∞ (E n ) ·
Remark 4.1Itfollowsby estimate(4.1)thatfor any
δ ∈ (0, ∆ 1 (ρ))
thesetN (P, δ)
is ontinuously embeddedin
H δ ∞
in the topologyofH δ ∞
.Sine
[D β (P (D)u)]
.g δ ∈ L 2
for anyβ ∈ N 0 n
andu ∈ N (P, δ)
, the followingorollary isan immediate onsequene of the aboveTheorem 4.1:
Corollary 4.1. Let
P (D)
be an almost hypoellipti operator,ρ ∈ (0, ρ P )
,δ ∈ (0, ∆ 1 (ρ))
,f ∈ H δ ∞
and letu ∈ L 2,δ
be a solution of the equationP (D)u = f
.Then
u ∈ H δ ∞
.Using Theorem 3.1and noting that
H δ ∞ (E n ) ⊂ H loc ∞ (E n ) ⊂ C ∞ (E n ),
we an prove
Corollary 4.2. Let the assumptions of Corollary 4.1 hold. Then
u ∈ C ∞ (E n )
.Theorem4.2. Let
P (D)
be a linear dierential operator with onstantoeients,suh that
N (P, δ) ⊂ H δ ∞
for aδ > 0
. Thenρ P ≥ δ
and operatorP (D)
is almosthypoellipti.
Proof.Firstnotethatbythelosedgraphtheoremandtheondition
N(P, δ) ⊂ H δ ∞
there existnumbers
k ∈ N
andC 1 > 0
suh thatX n
j=1
k (D j u)g δ k L 2 ≤ C 1 k u k P,k,δ ∀ u ∈ N (P, δ). (4.2)
To prove the theorem itis sues toshow that
ρ P ≥ δ
.Suppose to the ontrary that
ρ P < δ
. Then by the denition of the numberρ P
it follows that there exists a sequene
{ ξ s }
of points inR n
suh that| ξ s | → ∞
ass → ∞
andd P (ξ s ) ≤ ρ P + δ
2 s = 1, 2, . . . . (4.3)
Let
ζ s ∈ D(P )
besuh thatd P (ξ s ) = | ξ s − ζ s | s = 1, 2, . . . .
Then by (4.3)
| Im ζ s | ≤ d P (ξ s ) ≤ ρ P + δ
2 < δ s = 1, 2, . . . . (4.4)
Weset
u s (x) = e i(x, ζ s ) (s ∈ N )
. Thenu s ∈ N (P, δ)
and by (4.2) we obtain thatX n
j=1
k (D j u s )g δ k L 2 ≤ C 1 k u s k P,k,δ s = 1, 2, . . . . (4.5)
By estimates (2.3), (4.5) and by the denition of the points
{ ζ s }
we obtaink u s k P,k,δ = k u s · g δ k L 2 + X
| β |≤ k
[D β P (D)u s ]g δ
L 2 =
= k u s · g δ k L 2 ≤ κ e ρP 2 +δ | x | e − δ | x |
L 2 ≡ C 2 , s = 1, 2, . . . . (4.6)
On the otherhand, by estimates (2.3) and (4.4)
X n j=1
k (D j u s )g δ k L 2 = X n
j=1
| ζ j s | e i(x,ζ s ) g δ
L 2 ≥
≥ X n
j=1
| ζ j s | e −| Imζ s |·| x | g δ
L 2 ≥ κ − 1 X n
j=1
| ζ j s | e − ( | Imζ s | +δ) ·| x |
L 2 ≥
≥ κ − 1 X n j=1
| ζ j s | e − ρP 2 +δ | x | e − δ | x |
L 2 ≡ C 3 X n
j=1
| ζ j s | . (4.7)
From (4.5) (4.7) itfollows that
X n j=1
| ζ j s | ≤ C 4 , s = 1, 2, . . . ,
where
C 4 = C 1 C 2 /C 3
.From this and (4.3) weget
| ξ s | ≤ | ζ s − ξ s | + | ζ s | = d P (ξ s ) + | ζ s | ≤ ρ P + δ 2 + C 4 .
Therefore,thesequene
{ ξ s }
isbounded, whihontraditstheassumption.Thisontradition ompletes the proof.
UsingthestatementsofTheorems4.1and4.2,wearriveatthemainresultstated
inSetion 1.
Referenes
[1℄ R.A.Adams,Sobolevspaes.Aademipress,NewYorkSanFranisoLondon,1975.
[2℄ O.V. Besov, V.P. Il'in, S.M. Nikolskii, Integral representations of funtions and embedding
theorems.Nauka,Mosow,1975(inRussian).Englishtransl.JohnWileyandsons,NewYork,
v.1,1978,v.2,1979.
[3℄ Ya.S. Bugrov, Embedding theorems for some funtion spaes. Pro. SteklovInst.Math., 77
(1965),4564(inRussian).
[4℄ V.I. Burenkov, An analogue of Hormander's theorem on hypoelliptiity for funtions
onverging to 0 at innity. Pro. 7th Soviet Czehoslovak Seminar. Yerevan, 1982, 63 -
67(inRussian).
[5℄ V.I.Burenkov, Sobolev spaes on domains. B.G.Teubner, TeubnerTextezur Mathemati,
137,StuttgartLeipzig,1998.
[6℄ V.I.Burenkov,Investigationofspaesofdierentiablefuntionsdenedonirregulardomains.
Dotor'sdegreethesis.SteklovInst.Math.,Mosow,1982(in Russian).
[7℄ V.I.Burenkov,Conditionalhypoelliptiity andFouriermultipliersinweighted
L p
-spaeswithanexponential weight. Pro.oftheSummerShool"Funtionspaes,dierentialoperators,
nonlinear analysisheld in Fridrihrodain 1993. B.G. Teubner, Stuttgart-Leipzig. Teubner-
TextezurMathematik,133(1993),256265.
[8℄ L.Ehrenpreis, Solutionsof someproblems of division. 4. Amer. J. Math.,82 (1960),522
588.
[9℄ O.R.Gabrielyan,Comparisonofpowerandstrengthofpolynomialsin
R 2
.ComplexAnalysis,Dierential Equationsand Related Topis. Pro. ISAAC Conferene on Analysis,Yerevan,
2002,4151.
[10℄ H.G.Ghazaryan,Someestimatesofderivativesofpolynomialswithonstantoeients.Izv.
AN Armenii,Matematika,34, no.3(1999),4463(in Russian).English transl.Journalof
ContemporaryMathematialAnalysis(ArmenianAademyofSienes),34,no.3(1999).
[11℄ H.G. Ghazaryan, V.N. Margaryan, On the behaviour of nonellipti polynomials at innity.
Izv.ANArmenii,Matematika,39, no.3(2004),118(inRussian).English transl.Journal
ofContemporaryMathematialAnalysis(ArmenianAademyofSienes),39,no.3(2004).
[12℄ H.G. Ghazaryan, V.N. Margaryan, Behaviour at innity of polynomials in two variables.
Topisin Analysisandits Appliations,NATOSi.Series.KluwerAad.Publ.Dortreht
BostonLondon,147(2004),163190.
[13℄ H.G. Ghazaryan, V.N. Margaryan, On a lass of almost hypoellipti operators. Izv. AN
Armenii, Matematika, 41, no. 6 (2006), 39 56 (in Russian). English transl. Journal of
ContemporaryMathematialAnalysis(ArmenianAademyofSienes),41,no.6(2006),30
46.
[14℄ L. Garding, B. Malgrange, Operateurs dierentiels partiellement hypoelliptiques. Math.
Sand.,9(1961),521.
[15℄ L.Hormander,Onthetheoryofgeneralpartialdierentialoperators.AtaMath.,94(1955),
161248.
[16℄ L.Hormander,Theanalysis of linearpartialdierential operators2. SpringerVerlag,1983.
[17℄ G.G.Kazaryan,Onalmosthypoelliptipolynomials.DokladyRoss.Aad.Nauk.Matematika,
398,no.6(2004),701703(inRussian).
[18℄ A.N. Kolmogorov, S.V. Fomin, Elements of theory of funtions and funtional analysis.
Nauka,Mosow,1972(inRussian).
HaikGhazaryanandVahaganMargaryan
Departmentofmathematisandmathematialmodelling
RussianArmenian(Slavoni)StateUniversity
123OvsepEminSt
0051Yerevan,Armenia
E-mail:haikghazaryanmail.ru,mar tikoyahoo.om
Reeived:25.12.2009