Directory UMM :Journals:Journal_of_mathematics:DM:
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
Non Tangential Convergence for
the Ornstein-Uhlenbeck Semigroup.
Convergencia no tangencial para el
semigrupo de Ornstein-Uhlenbeck
Ebner Pineda (epineda@uicm.ucla.edu.ve)
Departamento de Matem´atica, Decanato de Ciencia y Tecnolog´ıa, UCLA
Apartado 400 Barquisimeto 3001 Venezuela
Wilfredo Urbina R.(wurbina@euler.ciens.ucv.ve)
Departamento de Matem´aticas, Facultad de Ciencias, UCV.
Apartado 47195, Los Chaguaramos, Caracas 1041-A Venezuela, and
Department of Mathematics and Statistics, University of New Mexico,
Albuquerque, NM, 87131, USA.
Abstract
In this paper we are going to get the non tangential convergence, in
an appropriated parabolic “gaussian cone”, of the Ornstein-Uhlenbeck
semigroup in providing two proofs of this fact. One is a direct proof by
using the truncated non tangential maximal function associated. The
second one is obtained by using a general statement. This second proof
also allows us to get a similar result for the Poisson-Hermite semigroup.
Key words and phrases: Non tangential convergence, OrnsteinUhlenbeck semigroup, Poisson-Hermite semigroup, Hermite expansions.
Resumen
En este art´ıculo vamos a obtener la convergencia no tangencial, en
un “cono gaussiano ”parab´
olico apropiado, del semigrupo de OrnsteinUhlenbeck dando dos pruebas diferentes de ello. La primera es una
prueba directa usando la funci´
on maximal no tangencial truncada asociada. La segunda prueba se obtiene usando principios generales. Esta
u
´ltima prueba nos permite obtener un resultado an´
alogo para el semigrupo de Poisson-Hermite.
Palabras y frases claves: Convergencia no tangencial, semigrupo de
Ornstein-Uhlenbeck, semigrupo de Poisson-Hermite, desarrollos de Hermite.
Received 2006/03/05. Revised 2006/07/02. Accepted 2006/07/25.
MSC (2000): Primary 42C10; Secondary 26A99.
108
1
Ebner Pineda, Wilfredo Urbina
Introduction
2
Let us consider the Gaussian measure γd (x) =
Ornstein-Uhlenbeck differential operator
L=
e−|x|
π d/2
with x ∈ Rd and the
1
△x − hx, ∇x i .
2
(1)
Qd
Let β = (β1 , ..., βd ) ∈ Nd be a multi-index, let β! = i=1 βi !, |β| =
Pd
βd
β1
∂
β
i=1 βi , ∂i = ∂xi , for each 1 ≤ i ≤ d and ∂ = ∂1 ...∂d .
Let us consider the normalized Hermite polynomial of order β, in d variables
d
β
Y
2
2 ∂ i
(2)
hβ (x) = |β| 1 1/2
(−1)βi exi βi (e−xi ),
(2 β!) i=1
∂xi
then, since the one dimensional Hermite polynomials satisfies the Hermite
equation, see [7], then the the normalized Hermite polynomial hβ is an eigenfunction of L, with eigenvalue −|β|,
1
Lhβ (x) = − |β| hβ (x).
(3)
Given a function f ∈ L (γd ) its β-Fourier-Hermite coefficient is defined by
Z
f (x)hβ (x)γd (dx).
fˆ(β) =< f, hβ >γd =
Rd
2
Let Cn be the closed subspace of L (γd ) generated by the linear combinations
of {hβ : |β| = n}. By the orthogonality of the Hermite polynomials with
respect to γd it is easy to see that {Cn } is an orthogonal decomposition of
L2 (γd ),
∞
M
L2 (γd ) =
Cn
n=0
which is called the Wiener chaos.
Let Jn be the orthogonal projection of L2 (γd ) onto Cn . If f is a polynomial,
X
fˆ(β)hβ .
Jn f =
|β|=n
The Ornstein-Uhlenbeck semigroup {Tt }t≥0 is given by
Z
e−2t (|x|2 +|y|2 )−2e−t hx,yi
1
−
1−e−2t
Tt f (x) =
f (y)γd (dy)
e
d/2
Rd
(1 − e−2t )
Z
|y−e−t x|2
1
−
1−e−2t f (y)dy.
e
=
π d/2 (1 − e−2t )d/2 Rd
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
(4)
Non Tangential Convergence for the Ornstein-Uhlenbeck Semigroup.
109
{Tt }t≥0 is a strongly continuous Markov semigroup of contractions on Lp (γd ),
with infinitesimal generator L. Also, by a change of variable we can write,
Z
p
(5)
f ( 1 − e−2t u + e−t x)γd (du).
Tt f (x) =
Rd
Definition 1.1. The maximal function for the Ornstein-Uhlenbeck semigroup
is defined as
T ∗ f (x) = sup |Tt f (x)|
t>0
=
1
|
d/2
(1 − r2 )d/2
00
If f ∈ L1 (γd ), u(x, t) = Pt f (x) is solution of the initial value problem
∂2u
(x, t) = −Lu(x, t)
2
∂t u(x, 0) = f (x)
where u(x, 0) = f (x) means that
lim u(x, t) = f (x), a.e. x
t→0+
We want to prove that this convergence, for the Poisson-Hermite semigroup,
is also non-tangential in the following sense. Let
¾
½
1
∧
1
,
(13)
Γγ (x) = (y, t) ∈ Rd+1
:
|y
−
x|
<
t
∧
+
|x|
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
Non Tangential Convergence for the Ornstein-Uhlenbeck Semigroup.
111
be a “gaussian cone”. Also we want to prove that
lim
(y,t)→x,(y,t)∈Γγ (x)
Pt f (y) = f (x), a.e. x
In order to study the non-tangential convergence for the Ornstein-Uhlenbeck
semigroup we are going to consider the following maximal function, that was
defined by L. Forzani and E. Fabes [3].
Definition 1.3. The non tangential maximal function associated to the Ornstein-Uhlenbeck semigroup is defined as
Tγ∗ f (x) =
sup
(y,t)∈Γp
γ (x)
| Tt f (y) | .
(14)
Using an inequality for a generalized maximal function, obtained by L.
Forzani in [2] (for more details see [8] pag 65–73 and 88–92), it can be proved
that Tγ∗ f is weak (1, 1) and strong (p, p) for 1 < p < ∞, with respect to the
Gaussian measure.
Actually for the non-tangential convergence for the Ornstein-Uhlenbeck
semigroup it is enough to consider a “truncated” maximal function. Let
½
¾
1
1
1
2,0 < t <
Γp (x) = (y, t) ∈ Rd+1
:
|y
−
x|
<
t
,
(15)
∧
+
|x|2 4
be a truncated parabolic “gaussian cone”.
Definition 1.4. The truncated non-tangencial maximal function associated
to the Ornstein-Uhlenbeck semigroup is defined as
T ∗ f (x) =
sup
(y,t)∈Γp (x)
| Tt f (y) | .
(16)
In the next lemma we are going to get a inequality better than (8) for the
truncated non tangential maximal function T ∗ f , which implies, immediately,
that T ∗ f is weak (1, 1) and strong (p, p) for 1 < p < ∞, with respect to the
gaussian measure.
Lemma 1.1.
T ∗ f (x) ≤ Cd Mγd f (x),
for all x ∈ Rd
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
(17)
112
Ebner Pineda, Wilfredo Urbina
Proof. Let us take u(y, t) = Tt f (y) and without loss of generality let us
assume f ≥ 0.
√
Let ao = 0 and aj = j, j ∈ N, then aj < aj+1 ∀j ∈ N, and let us
denote
1
1
Aj (y, t) = {u ∈ Rd : aj−1 (1 − e−2t ) 2 ≤| e−t y − u |< aj (1 − e−2t ) 2 },
the annulus with center e−t y. Now consider for each j ∈ N the ball with
1
center e−t y, and radius aj (1 − e−2t ) 2 and let us denote it by Bj (y, t) =
1
B(e−t y, aj (1 − e−2t ) 2 ), then
Aj (y, t) = Bj (y, t) \ Bj−1 (y, t)
u(y, t)
=
=
1
d
d
d
∞ Z
X
π 2 (1 − e−2t ) 2
1
d
− | e−t y − u |2
1 − e−2t
f (u)du
e
Z
π 2 (1 − e−2t ) 2
Rd
j=1
Aj (y,t)
− | e−t y − u |2
1 − e−2t
f (u)du
e
1
Now if (y, t) ∈ Γp (x) and | e−t y − u |< aj (1 − e−2t ) 2 then,
| e−t x − u |
= | e−t x − e−t y + e−t y − u |
≤ | e−t (x − y) | + | e−t y − u |
1
1
< e−t t 2 + aj (1 − e−2t ) 2
1
1
< t 2 + aj (1 − e−2t ) 2
1
< (1 + aj )(1 − e−2t ) 2 ,
since t < 1 − e−2t if t < 0.8
1
Considering Cj (x, t) = B(e−t x, (1 + aj )(1 − e−2t ) 2 ), we have
u(y, t) ≤
1
d
2
π (1 − e−2t )
d
2
∞
X
j=1
2
e−aj−1
Z
f (u)du
Cj (x,t)
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
Non Tangential Convergence for the Ornstein-Uhlenbeck Semigroup.
113
Now,
Z
f (u)du =
2
2
f (u)e|u| e−|u| du
Cj (x,t)
Cj (x,t)
=
Z
Z
f (u)e|u−e
−t
x|2 +2e−t x.(u−e−t x)+|e−t x|2 −|u|2
e
du
Cj (x,t)
1
(1+aj )2 (1−e−2t )+2(1+aj )(1−e−2t ) 2 |e−t x|+|e−t x|2
≤e
Z
2
f (u)e−|u| du
Cj (x,t)
1
but, | e−t x − u |< (1 + aj )(1 − e−2t ) 2 and therefore,
1
| x − u |=| x − e−t x + e−t x − u |< (1 − e−t ) | x | +(1 + aj )(1 − e−2t ) 2 .
Taking
1
Dj (x, t) = B(x, (1 − e−t ) | x | +(1 + aj )(1 − e−2t ) 2 ),
we get
Z
Cj (x,t)
2
f (u)e−|u| du ≤
≤ Mγd f (x)
Z
Z
Dj (x,t)
−|u|2
e
2
f (u)e−|u| du
du = Mγd f (x)
≤ Mγd f (x)e
Z
2
e−|u−x|
+2x(x−u)−|x|2
du
Dj (x,t)
Dj (x,t)
−|x|2
Z
2
e−|u−x|
+2|x||x−u|
du
Dj (x,t)
1
−|x|2 +2|x|((1−e−t )|x|+(1+aj )(1−e−2t ) 2 )
≤ Mγd f (x)e
Z
e−|u−x| du
Z
e−|w| dw
2
Dj (x,t)
2
= Mγd f (x)e−|x|
1
+2|x|((1−e−t )|x|+(1+aj )(1−e−2t ) 2 )
2
Ej (x,t)
1
where Ej (x, t) = B(0, (1 − e−t ) | x | +(1 + aj )(1 − e−2t ) 2 ).
Since γd is a d−dimensional measure, and using that t <
1
∧ 1 , we get
| x |2 4
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
114
Z
Ebner Pineda, Wilfredo Urbina
2
f (u)e−|u| du
2
Cd Mγd f (x)e−|x|
≤
Cj (x,t)
1
+2|x|((1−e−t )|x|+(1+aj )(1−e−2t ) 2 )
1
×((1 − e−t ) | x | +(1 + aj )(1 − e−2t ) 2 )d
2
Cd Mγd f (x)e−|x|
=
1
+2|x|((1−e−t )|x|+(1+aj )(1−e−2t ) 2
1
d
1
×(1 − e−t ) 2 ((1 − e−t ) 2 | x | +(1 + aj )(1 + e−t ) 2 )d
−|x|2 +2
Cd Mγd f (x)e
≤
−t d
2
×(1 − e
)
Ã
1
(1−e−t )
(1−e−2t ) 2
+2(1+aj )
1
t
t2
1
(1 − e−t ) 2
1
t2
1
−t 2
+ (1 + aj )(1 + e
)
!d
.
Therefore
Z
(1+aj )2 (1−e−2t )+2(1+aj )
f (u) du ≤ e
1
(1−e−2t ) 2
1
2
t
+e−2t |x|2
Cj (x,t)
Z
2
f (u)e−|u| du
Cj (x,t)
1
(1−e−2t ) 2
(1+aj )2 (1−e−2t )+2(1+aj )
1
t2
−|x|2 +2
+|x|2
Cd Mγd f (x)e
≤e
1
(1−e−t )
(1−e−2t ) 2
+2(1+aj )
1
t
t2
1
d
×(1 − e−t ) 2 (
(1+aj )
2
−1
(1−e 2
≤e
×
≤ e(1+aj )
since 0 < t <
1
4
and
2
−1
2
(1−e
Ã
t
)+4(1+aj )
1
2
1
+ (1 + aj )(1 + e−t ) 2 )d
1
(1−e−2t ) 2
1
t2
+
2(1−e−t )
t
1
(1 − e−2t ) 2
)+4(1+aj )
(1 − e−t ) 2
1
t2
√
2+2
1
−t 2
+ (1 + aj )(1 + e
)
!d
d
(1 − e−t ) 2
Cd Mγd f (x)
√
d
.(1 − e−t ) 2 (1 + (1 + aj ) 2)d Cd Mγd f (x),
1 − e−t
< 1, 1 + e−t < 2, if t > 0.
t
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
Non Tangential Convergence for the Ornstein-Uhlenbeck Semigroup.
115
Thus,
u(y, t) ≤
1
d
d
π 2 (1 − e−2t ) 2
≤ Cd Mγd f (x)
×
∞
X
∞
X
2
e−aj−1
j=1
Z
f (u)du
Cj (x,t)
1
d
d
d
π 2 (1 + e−t ) 2 (1 − e−t ) 2
−1
2
2
2
e−aj−1 e(1+aj )
(1−e
√
)+4(1+aj ) 2+2
j=1
≤ Cd Mγd f (x)
∞
1 X
π
1
2
2
e−aj−1 +(1+aj )
−1
2
2
(1−e
√
d
(1 − e−t ) 2 (1 + (1 + aj ) 2)d
√
)+4(1+aj ) 2+2
√
(1 + (1 + aj ) 2)d ,
j=1
since 1 + e−t ≥ 1. Now it is easy to see that
√
1
−a2j−1 + (1 + aj )2 (1 − e− 2 ) + 4(1 + aj ) 2 + 2
√
√
1
1
1p p
= 4 + 4 2 − e− 2 − [−(2(1 − e− 2 ) + 4 2) + e− 2 j] j,
which is negative for j sufficiently big, then
∞
X
2
e−aj−1 +(1+aj )
2
−1
2
(1−e
√
)+4(1+aj ) 2+2
j=1
√
.(1 + (1 + aj ). 2)d < ∞.
Thus u(y, t) ≤ Cd Mγd f (x) and since (y, t) ∈ Γp (x) is arbitrary
T ∗ f (x) =
sup
(y,t)∈Γp (x)
u(y, t) ≤ Cd Mγd f (x).
¤
Now we are ready to establish the convergence result for the Ornstein-Uhlenbeck
semigroup.
Theorem 1.2. The Ornstein-Uhlenbeck semigroup {Tt f } converges in L1 (γd ) a.e
if t → 0+ , for any function f ∈ L1 (γd ),
lim u(x, t) = f (x), a.e. x
t→0+
(18)
Moreover, if u(y, t) = Tt f (y) then u(y, t) tends to f (x) non tangentially ,i.e.
lim
p
(y,t)→x,(y,t)∈Γγ (x)
Tt f (y) = f (x), a.e. x.
Proof. We have,
u(y, t) =
1
d
d
π 2 (1 − e−2t ) 2
Z
Rd
− | e−t y − u |2
1 − e−2t
f (u)du,
e
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
(19)
116
Ebner Pineda, Wilfredo Urbina
considering
Ωf (x) = lim
α→0+
"
sup
p
(y,s)∈Γγ (x),0 δ
depends on x and Mγd g(x) < ǫ.
On the other hand, since
Thus g
u(y, t) − f (x) = u1 (y, t) − f1 (x) + u2 (y, t) − f2 (x)
where
1
i
u (y, t) =
d
d
π 2 (1 − e−2t ) 2
Z
Rd
− | e−t y − u |2
1 − e−2t
e
fi (u)du i = 1, 2,
then we get,
1
1
u (y, t) − f1 (x) =
=
+
1
d
d
π 2 (1 − e−2t ) 2
1
d
d
π 2 (1 − e−2t ) 2
d
d
π 2 (1 − e−2t ) 2
Z
Z
|x−u|≤δ
|x−u|>δ
Z
Rd
− | e−t y − u |2
1 − e−2t
e
(f1 (u) − f1 (x))du
− | e−t y − u |2
1 − e−2t
(f1 (u) − f1 (x))du
e
− | e−t y − u |2
1 − e−2t
e
(f1 (u) − f1 (x))du.
Now we have that if | x |≤ k − 1 and (y, t) ∈ Γpγ (x) with t <
(y, t) ∈ Γp (x). Thus | u − x |≤ δ implies
1
|x|2
| u |=| u − x + x |≤| u − x | + | x |< δ + k − 1 < 1 + k − 1 = k
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
∧
1
,
4
then
Non Tangential Convergence for the Ornstein-Uhlenbeck Semigroup.
117
and then, f1 (u) = f (u) ∧ f1 (x) = f (x). Therefore
¯
¯
¯
¯Z
− | e−t y − u |2
¯
¯
¯
¯
−2t
1−e
(f1 (u) − f1 (x))du¯
e
¯
¯
¯ |x−u|≤δ
¯
¯
¯
¯
−t
2
¯Z
¯
−
|
e
y
−
u
|
¯
¯
¯
¯
1 − e−2t
e
(f (u) − f (x))du¯
¯
¯ |x−u|≤δ
¯
¯
¯
1
d
d
π 2 (1 − e−2t ) 2
=
=
≤
1
d
d
2
π (1 − e−2t ) 2
¯
¯
¯Z
¯
− | e−t y − u |2
¯
¯
¯
¯
−2t
1−e
g(u)du¯
e
¯
¯ Rd
¯
¯
¯
1
d
d
π 2 (1 − e−2t ) 2
T ∗ g(x) ≤ Cd Mγd g(x) ≤ Cd ǫ.
1
Now observe that if (y, t) ∈ Γpγ (x) and t 2 ≤
u − x |≤| u − y | + | y − x | and thus
δ
2
then, | u − x |> δ implies δ δ
¯
¯
¯
d
2
Z
|u−y|> δ
2
|u−x|>δ
− | e−t y − u |2
1 − e−2t
e
du
− | e−t y − u |2
1 − e−2t
| f1 (u) | du
e
1
d
| f1 (x) |
Z
d
π 2 (1 − e−2t ) 2
| f1 (x) |
Z
|u−x|>δ
− | e−t y − u |2
1 − e−2t
du.
e
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
118
Ebner Pineda, Wilfredo Urbina
Now, we have
d
d
|u−y|> δ
2
d
Z
|u−y|> δ
,|u| δ
,|u| δ
,|u| 2δ , | u |< k implies that
Then for 0 < t < log
4k + δ
≤
| e−t y − u |
but 0 < t < log
µ
e−t
=
| e−t y − e−t u + e−t u − u |=| e−t (y − u) − (u − e−t u) |
≥
e−t | y − u | − | u − e−t u |= e−t | y − u | −(1 − e−t ) | u |
≥
e−t
δ
− k(1 − e−t ) = e−t
2
µ
4k + 2δ
4k + δ
¶
and therefore e−t >
µ
¶
−k
δ
+k
2
µ
δ
+k
2
¶
− k,
4k + δ
, then,
4k + 2δ
δ + 2k
2
¶
>
4k + δ
4k + 2δ
=
4k + δ
(2k + δ) − k
4(2k + δ)
=
4k + δ
4k + δ − 4k
δ
−k =
= .
4
4
4
−k
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
Non Tangential Convergence for the Ornstein-Uhlenbeck Semigroup.
Therefore | u − y |>
δ
δ
, | u |< k implies | e−t y − u |> and thus
2
4
1
d
≤
≤
d
π 2 (1 − e−2t ) 2
1
d
2
π (1 − e−2t )
−
e
d
2
Z
e
δ2
+k2
16(1−e−2t )
d
d
π 2 (1 − e−2t ) 2
|u−y|> δ
2
k2
Z
Z
− | e−t y − u |2
1 − e−2t
| f1 (u) | du
e
|u−y|> δ
,|u|δ
− | e−t y − u |2
1 − e−2t
du
e
−|s|2
e 1−e−2t ds
|x−s−e−t y|>δ
−|s|2
e 1−e−2t ds,
|x−s−e−t y|>δ
since, f1 (x) = f (x) as | x |≤µk − 1 < k.
¶
k − 1 − δ/2
Thus taking 0 < t < log
, | x − s − e−t y |> δ implies
k − 1 − 3δ/4
|s|
=
| s − x + e−t y + x − e−t y |=| s − x + e−t y − (e−t y − x) |
≥
| s − x + e−t y | − | e−t y − x | .
But
| e−t y − x |
1
Thus, since t 2 ≤
=
| e−t y − e−t x + e−t x − x |≤ e−t | y − x | +(1 − e−t ) | x |
≤
e−t t 2 + (1 − e−t )(k − 1).
1
δ
,
2
| s − x + e−t y | − | e−t y − x |
1
>
δ − e−t t 2 − (1 − e−t )(k − 1)
≥
δ − e−t
=
δ − (k − 1) + (k − 1 −
δ
− (k − 1)(1 − e−t )
2
δ −t
)e ,
2
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
119
120
Ebner Pineda, Wilfredo Urbina
and as 0 < t < log
µ
|s|
k − 1 − δ/2
k − 1 − 3δ/4
¶
, then e−t >
k − 1 − 3δ/4
. Hence,
k − 1 − δ/2
>
δ − (k − 1) + (k − 1 − δ/2)e−t
>
δ − (k − 1) + (k − 1 − δ/2)
=
δ − (k − 1) + k − 1 − 3δ/4 = δ − 3δ/4 =
k − 1 − 3δ/4
k − 1 − δ/2
δ
if 0 < t < log
Then | x − s − e y |> δ implies | s |>
4
s
Therefore, taking w = √
,
1 − e−2t
−t
d
Z
|u−y|> δ
2
d
Z
|s|> δ
4
1
d
π 2 (1 − e−2t ) 2
≤
=
d
| f (x) |
π 2 (1 − Ze−2t ) 2
| f (x) |
d
π2
|w|>
√
4
δ
.
4
µ
¶
k − 1 − δ/2
.
k − 1 − 3δ/4
− | e−t y − u |2
1 − e−2t
e
| f1 (u) | du
−|s|2
e 1 − e−2t ds
δ
2
e−|w| dw.
1−e−2t
Now since, | x |≤ k − 1 < k, thenf2 (x) = 0. Hence
| u2 (y, t) − f2 (x) |=| u2 (y, t) |≤ T ∗ f2 (x) ≤ Cd Mγd f2 (x)
for (y, t) ∈ Γp (x). Therefore,
| u(y, t) − f (x) |≤| u1 (y, t) − f1 (x) | + | u2 (y, t) − f2 (x) |
=| u1 (y, t) − f1 (x) | + | u2 (y, t) |
−δ
≤ Cd ǫ +
e 16(1−e−2t )
(1 −
+k2
e−2t )
d
2
kf k1,γd +
| f (x) |
π
d
2
Z
2
|w|>
√
4
δ
e−|w| dw
1−e−2t
+Cd Mγd f2 (x),
if (y, t) ∈ Γpγ (x) and
¶
µ
¶
¾
½
µ
k − 1 − δ/2
1
1
4k + 2δ
, log
, 2 ∧
=: a.
0 < t < min log
4k + δ
k − 1 − 3δ/4
|x|
4
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
Non Tangential Convergence for the Ornstein-Uhlenbeck Semigroup.
121
Thus taking supremum on (y, t) ∈ Γpγ (x), 0 < t < α < a and then taking α → 0+
we obtain,
Ωf (x) ≤ Cd (ǫ + Mγd f2 (x))
for all ǫ > 0 and almost every x with | x |≤ k − 1.
Given ǫ > 0, let us take k sufficiently large such that
kf2 k1,γd ≤ Cd ǫ2 ,
then by the estimation of Ω and the weak continuity of Mγd we get
γd ({x ∈ Rd :| x |≤ k − 1, Ωf (x) > ǫ}) ≤ ǫ
and that implies that Ωf (x) = 0 a.e.
¤
A similar proof for the Poisson-Hermite semigroup, using the non-tangential
maximal function defined as
Pγ∗ f (x) =
sup
(y,t)∈Γγ (x)
| Pt f (y) |,
(20)
and its analogous truncated version, should be possible but it has some technical
difficulties that we have been unable to overcome so far.
Let us now prove a general statement for families of linear operators that will
allow us to get a simpler proof of the non-tangential convergence, both for the
Ornstein-Uhlenbeck semigroup and also for the Poisson-Hermite semigroup. It is a
generalization of Theorem 2.2 of J. Duoandikoetxea’s book [1].
Theorem 1.3. Let {Tt }t>0 be a family of linear operators on Lp (Rd , µ) and for any
x ∈ Rd , let Γ(x) be a subset of Rd+1
such that x is in (Γ(x))′ , that is to say x is an
+
accumulation point of Γ(x). Let us define
T ∗ f (x) = sup{|Tt f (y)| : (y, t) ∈ Γ(x)},
for f ∈ Lp (Rd , µ) and x ∈ Rd . If T ∗ is weak (p, q) then the set
½
¾
S = f ∈ Lp (Rd , µ) :
lim
Tt f (y) = f (x) a.e.
(y,t)→x,(y,t)∈Γ(x)
is closed in Lp (Rd , µ).
Proof. Let us consider a sequence (fn ) in S such that fn → f in Lp (Rd , µ), then
|Tt f (y) − f (x)| − |Tt fn (y) − fn (x)| ≤ |Tt (f − fn )(y) − (f (x) − fn (x))|,
this implies that for each n ∈ N, for almost every x,
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
122
Ebner Pineda, Wilfredo Urbina
lim sup
|Tt f (y) − f (x)|
(y,t)→x,(y,t)∈Γ(x)
≤
(y,t)→x,(y,t)∈Γ(x)
lim sup
≤
(y,t)→x,(y,t)∈Γ(x)
lim sup
+
|Tt (f − fn )(y) − (f (x) − fn (x))|
|Tt (f − fn )(y)|
lim sup
(y,t)→x,(y,t)∈Γ(x)
|f (x) − fn (x)|
T ∗ (f − fn )(x) + |f (x) − fn (x)|.
≤
On the other hand, if we know that a ≤ b + c then a > λ implies b > λ2 ∨ c > λ2 .
Then, given λ > 0 and n ∈ N, lim sup(y,t)→x,(y,t)∈Γ(x) |Tt f (y)−f (x)| > λ implies
λ
λ
∨ |f (x) − fn (x)| > a.e.
2
2
T ∗ (f − fn )(x) >
and this implies that, given λ > 0,
Ã(
µ
x:
lim sup
(y,t)→x,(y,t)∈Γ(x)
≤µ
≤
for all n ∈ N. Therefore,
Ã(
µ
x:
µ
|Tt f (y) − f (x)| > λ
¾¶
λ
x : T ∗ (f − fn )(x) >
2
¾¶
µ½
λ
+µ
x : |f (x) − fn (x)| >
2
µ½
¢q ¡ 2
2C
kf − fn kp +
kf − fn kp
λ
λ
lim sup
(y,t)→x,(y,t)∈Γ(x)
x:
lim sup
(y,t)→x,(y,t)∈Γ(x)
|Tt f (y) − f (x)| > 0
¶p
,
)!
=0
)!
= 0,
|Tt f (y) − f (x)| > λ
and since this is true for all λ > 0, we get that
Ã(
µ
)!
as
(
x:
lim sup
(y,t)→x,(y,t)∈Γ(x)
=
|Tt f (y) − f (x)| > 0
∞
[
n=1
(
)
1
x:
lim sup
|Tt f (y) − f (x)| >
n
(y,t)→x,(y,t)∈Γ(x)
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
)
.
Non Tangential Convergence for the Ornstein-Uhlenbeck Semigroup.
123
Thus
lim
(y,t)→x,(y,t)∈Γ(x)
Tt f (y) = f (x) a.e.
and then f ∈ S. Therefore S is a closed set in Lp (Rd , µ).
¤
Finally, as a consequence of this result, we get the non-tangential convergence
for the Ornstein-Uhlenbeck semigroup {Tt }t>0 and the Poisson-Hermite semigroup
{Pt }t>0 .
Corollary 1.4. The Ornstein-Uhlenbeck semigroup {Tt }t>0 and the Poisson-Hermite semigroup {Pt }t>0 verify
lim
p
(y,t)→x,(y,t)∈Γγ (x)
lim
(y,t)→x,(y,t)∈Γγ (x)
Tt f (y) = f (x) a.e. x,
Pt f (y) = f (x) a.e. x.
Proof. Let us discuss the proof for the the Ornstein-Uhlenbeck semigroup {Tt }t>0 .
The proof for the Poisson-Hermite semigroup {Pt }t>0 is totally similar.
P
It is immediate
that
for any given polynomial f (x) = n
k=0 Jk f (x), since Tt f (y) =
¢
¡P
P
n
n
−tk
Jk f (y), we have the non-tangential convergence,
Tt
k=0 Jk f (y) =
k=0 e
lim
p
(y,t)→x,(y,t)∈Γγ (x)
Tt f (y) = f (x),
for all x ∈ Rd . Now considering the set
S=
(
p
f ∈ L (γd ) :
lim
p
(y,t)→x,(y,t)∈Γγ (x)
Tt f (y) = f (x) a.e.
)
,
corresponding to the Ornstein-Uhlenbeck semigroup, then the polynomials are in S.
From the previous result, since non-tangential maximal function for the OrnsteinUhlenbeck semigroup Tγ∗ f is weak (1, 1) with respect to the Gaussian measure, we
get that the set S is closed in Lp (γd ) and since the polynomials are dense in Lp (γd )
then S = Lp (γd ).
¤
Acknowledgement
We want to thank the referees for their suggestions and/or corrections that improved
the presentation of this paper.
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
124
Ebner Pineda, Wilfredo Urbina
References
[1] Duoandikoetxea, J. Fourier Analysis, Graduated Studies in Mathematics, Volume 29, AMS, R. I., 2001.
[2] Forzani, L. Lemas de cubrimiento de tipo Besicovitch y su aplicaci´
on al estudio
del operador maximal de Ornstein-Uhlenbeck. Tesis de Doctorado. Universidad
Nacional de San Luis, Argentina, 1993.
[3] Forzani, L., Fabes, E. Unpublished manuscript, 1994.
[4] Guti´errez, C., Urbina, W., Estimates for the maximal operator of the OrnsteinUhlenbeck semigroup. Proc. Amer. Math. Soc. 113 (1991), 99–104.
[5] Sj¨
ogren P., Operators associated with the Hermite Semigroup-A Survey, J.
Fourier Anal. Appl., (3)(1997), 813–823.
[6] Stein E., Singular Integrals and Differentiability Properties of Functions,
Princeton Univ. Press, Princeton, New Jersey, 1970.
[7] Szeg¨
o, G., Orthogonal polynomials, rev. ed., Amer. Math. Soc. Colloq. Publ.,
vol. 23, Amer. Math. Soc., Providence, R. I., 1959.
[8] Urbina W.m An´
alisis Arm´
onico Gaussiano: una visi´
on panor´
amica, Trabajo
de Ascenso, Facultad de Ciencias, UCV, 1998. Available in
http://euler.ciens.ucv.ve/~wurbina/notes.html
[9] Zygmund, A., Trigonometric Series, 2nd. ed., Cambridge Univ. Press., Cambridge, 1959.
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
aticas Vol. 16 No. 1(2008), pp. 107–124
Non Tangential Convergence for
the Ornstein-Uhlenbeck Semigroup.
Convergencia no tangencial para el
semigrupo de Ornstein-Uhlenbeck
Ebner Pineda (epineda@uicm.ucla.edu.ve)
Departamento de Matem´atica, Decanato de Ciencia y Tecnolog´ıa, UCLA
Apartado 400 Barquisimeto 3001 Venezuela
Wilfredo Urbina R.(wurbina@euler.ciens.ucv.ve)
Departamento de Matem´aticas, Facultad de Ciencias, UCV.
Apartado 47195, Los Chaguaramos, Caracas 1041-A Venezuela, and
Department of Mathematics and Statistics, University of New Mexico,
Albuquerque, NM, 87131, USA.
Abstract
In this paper we are going to get the non tangential convergence, in
an appropriated parabolic “gaussian cone”, of the Ornstein-Uhlenbeck
semigroup in providing two proofs of this fact. One is a direct proof by
using the truncated non tangential maximal function associated. The
second one is obtained by using a general statement. This second proof
also allows us to get a similar result for the Poisson-Hermite semigroup.
Key words and phrases: Non tangential convergence, OrnsteinUhlenbeck semigroup, Poisson-Hermite semigroup, Hermite expansions.
Resumen
En este art´ıculo vamos a obtener la convergencia no tangencial, en
un “cono gaussiano ”parab´
olico apropiado, del semigrupo de OrnsteinUhlenbeck dando dos pruebas diferentes de ello. La primera es una
prueba directa usando la funci´
on maximal no tangencial truncada asociada. La segunda prueba se obtiene usando principios generales. Esta
u
´ltima prueba nos permite obtener un resultado an´
alogo para el semigrupo de Poisson-Hermite.
Palabras y frases claves: Convergencia no tangencial, semigrupo de
Ornstein-Uhlenbeck, semigrupo de Poisson-Hermite, desarrollos de Hermite.
Received 2006/03/05. Revised 2006/07/02. Accepted 2006/07/25.
MSC (2000): Primary 42C10; Secondary 26A99.
108
1
Ebner Pineda, Wilfredo Urbina
Introduction
2
Let us consider the Gaussian measure γd (x) =
Ornstein-Uhlenbeck differential operator
L=
e−|x|
π d/2
with x ∈ Rd and the
1
△x − hx, ∇x i .
2
(1)
Qd
Let β = (β1 , ..., βd ) ∈ Nd be a multi-index, let β! = i=1 βi !, |β| =
Pd
βd
β1
∂
β
i=1 βi , ∂i = ∂xi , for each 1 ≤ i ≤ d and ∂ = ∂1 ...∂d .
Let us consider the normalized Hermite polynomial of order β, in d variables
d
β
Y
2
2 ∂ i
(2)
hβ (x) = |β| 1 1/2
(−1)βi exi βi (e−xi ),
(2 β!) i=1
∂xi
then, since the one dimensional Hermite polynomials satisfies the Hermite
equation, see [7], then the the normalized Hermite polynomial hβ is an eigenfunction of L, with eigenvalue −|β|,
1
Lhβ (x) = − |β| hβ (x).
(3)
Given a function f ∈ L (γd ) its β-Fourier-Hermite coefficient is defined by
Z
f (x)hβ (x)γd (dx).
fˆ(β) =< f, hβ >γd =
Rd
2
Let Cn be the closed subspace of L (γd ) generated by the linear combinations
of {hβ : |β| = n}. By the orthogonality of the Hermite polynomials with
respect to γd it is easy to see that {Cn } is an orthogonal decomposition of
L2 (γd ),
∞
M
L2 (γd ) =
Cn
n=0
which is called the Wiener chaos.
Let Jn be the orthogonal projection of L2 (γd ) onto Cn . If f is a polynomial,
X
fˆ(β)hβ .
Jn f =
|β|=n
The Ornstein-Uhlenbeck semigroup {Tt }t≥0 is given by
Z
e−2t (|x|2 +|y|2 )−2e−t hx,yi
1
−
1−e−2t
Tt f (x) =
f (y)γd (dy)
e
d/2
Rd
(1 − e−2t )
Z
|y−e−t x|2
1
−
1−e−2t f (y)dy.
e
=
π d/2 (1 − e−2t )d/2 Rd
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
(4)
Non Tangential Convergence for the Ornstein-Uhlenbeck Semigroup.
109
{Tt }t≥0 is a strongly continuous Markov semigroup of contractions on Lp (γd ),
with infinitesimal generator L. Also, by a change of variable we can write,
Z
p
(5)
f ( 1 − e−2t u + e−t x)γd (du).
Tt f (x) =
Rd
Definition 1.1. The maximal function for the Ornstein-Uhlenbeck semigroup
is defined as
T ∗ f (x) = sup |Tt f (x)|
t>0
=
1
|
d/2
(1 − r2 )d/2
00
If f ∈ L1 (γd ), u(x, t) = Pt f (x) is solution of the initial value problem
∂2u
(x, t) = −Lu(x, t)
2
∂t u(x, 0) = f (x)
where u(x, 0) = f (x) means that
lim u(x, t) = f (x), a.e. x
t→0+
We want to prove that this convergence, for the Poisson-Hermite semigroup,
is also non-tangential in the following sense. Let
¾
½
1
∧
1
,
(13)
Γγ (x) = (y, t) ∈ Rd+1
:
|y
−
x|
<
t
∧
+
|x|
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
Non Tangential Convergence for the Ornstein-Uhlenbeck Semigroup.
111
be a “gaussian cone”. Also we want to prove that
lim
(y,t)→x,(y,t)∈Γγ (x)
Pt f (y) = f (x), a.e. x
In order to study the non-tangential convergence for the Ornstein-Uhlenbeck
semigroup we are going to consider the following maximal function, that was
defined by L. Forzani and E. Fabes [3].
Definition 1.3. The non tangential maximal function associated to the Ornstein-Uhlenbeck semigroup is defined as
Tγ∗ f (x) =
sup
(y,t)∈Γp
γ (x)
| Tt f (y) | .
(14)
Using an inequality for a generalized maximal function, obtained by L.
Forzani in [2] (for more details see [8] pag 65–73 and 88–92), it can be proved
that Tγ∗ f is weak (1, 1) and strong (p, p) for 1 < p < ∞, with respect to the
Gaussian measure.
Actually for the non-tangential convergence for the Ornstein-Uhlenbeck
semigroup it is enough to consider a “truncated” maximal function. Let
½
¾
1
1
1
2,0 < t <
Γp (x) = (y, t) ∈ Rd+1
:
|y
−
x|
<
t
,
(15)
∧
+
|x|2 4
be a truncated parabolic “gaussian cone”.
Definition 1.4. The truncated non-tangencial maximal function associated
to the Ornstein-Uhlenbeck semigroup is defined as
T ∗ f (x) =
sup
(y,t)∈Γp (x)
| Tt f (y) | .
(16)
In the next lemma we are going to get a inequality better than (8) for the
truncated non tangential maximal function T ∗ f , which implies, immediately,
that T ∗ f is weak (1, 1) and strong (p, p) for 1 < p < ∞, with respect to the
gaussian measure.
Lemma 1.1.
T ∗ f (x) ≤ Cd Mγd f (x),
for all x ∈ Rd
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
(17)
112
Ebner Pineda, Wilfredo Urbina
Proof. Let us take u(y, t) = Tt f (y) and without loss of generality let us
assume f ≥ 0.
√
Let ao = 0 and aj = j, j ∈ N, then aj < aj+1 ∀j ∈ N, and let us
denote
1
1
Aj (y, t) = {u ∈ Rd : aj−1 (1 − e−2t ) 2 ≤| e−t y − u |< aj (1 − e−2t ) 2 },
the annulus with center e−t y. Now consider for each j ∈ N the ball with
1
center e−t y, and radius aj (1 − e−2t ) 2 and let us denote it by Bj (y, t) =
1
B(e−t y, aj (1 − e−2t ) 2 ), then
Aj (y, t) = Bj (y, t) \ Bj−1 (y, t)
u(y, t)
=
=
1
d
d
d
∞ Z
X
π 2 (1 − e−2t ) 2
1
d
− | e−t y − u |2
1 − e−2t
f (u)du
e
Z
π 2 (1 − e−2t ) 2
Rd
j=1
Aj (y,t)
− | e−t y − u |2
1 − e−2t
f (u)du
e
1
Now if (y, t) ∈ Γp (x) and | e−t y − u |< aj (1 − e−2t ) 2 then,
| e−t x − u |
= | e−t x − e−t y + e−t y − u |
≤ | e−t (x − y) | + | e−t y − u |
1
1
< e−t t 2 + aj (1 − e−2t ) 2
1
1
< t 2 + aj (1 − e−2t ) 2
1
< (1 + aj )(1 − e−2t ) 2 ,
since t < 1 − e−2t if t < 0.8
1
Considering Cj (x, t) = B(e−t x, (1 + aj )(1 − e−2t ) 2 ), we have
u(y, t) ≤
1
d
2
π (1 − e−2t )
d
2
∞
X
j=1
2
e−aj−1
Z
f (u)du
Cj (x,t)
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
Non Tangential Convergence for the Ornstein-Uhlenbeck Semigroup.
113
Now,
Z
f (u)du =
2
2
f (u)e|u| e−|u| du
Cj (x,t)
Cj (x,t)
=
Z
Z
f (u)e|u−e
−t
x|2 +2e−t x.(u−e−t x)+|e−t x|2 −|u|2
e
du
Cj (x,t)
1
(1+aj )2 (1−e−2t )+2(1+aj )(1−e−2t ) 2 |e−t x|+|e−t x|2
≤e
Z
2
f (u)e−|u| du
Cj (x,t)
1
but, | e−t x − u |< (1 + aj )(1 − e−2t ) 2 and therefore,
1
| x − u |=| x − e−t x + e−t x − u |< (1 − e−t ) | x | +(1 + aj )(1 − e−2t ) 2 .
Taking
1
Dj (x, t) = B(x, (1 − e−t ) | x | +(1 + aj )(1 − e−2t ) 2 ),
we get
Z
Cj (x,t)
2
f (u)e−|u| du ≤
≤ Mγd f (x)
Z
Z
Dj (x,t)
−|u|2
e
2
f (u)e−|u| du
du = Mγd f (x)
≤ Mγd f (x)e
Z
2
e−|u−x|
+2x(x−u)−|x|2
du
Dj (x,t)
Dj (x,t)
−|x|2
Z
2
e−|u−x|
+2|x||x−u|
du
Dj (x,t)
1
−|x|2 +2|x|((1−e−t )|x|+(1+aj )(1−e−2t ) 2 )
≤ Mγd f (x)e
Z
e−|u−x| du
Z
e−|w| dw
2
Dj (x,t)
2
= Mγd f (x)e−|x|
1
+2|x|((1−e−t )|x|+(1+aj )(1−e−2t ) 2 )
2
Ej (x,t)
1
where Ej (x, t) = B(0, (1 − e−t ) | x | +(1 + aj )(1 − e−2t ) 2 ).
Since γd is a d−dimensional measure, and using that t <
1
∧ 1 , we get
| x |2 4
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
114
Z
Ebner Pineda, Wilfredo Urbina
2
f (u)e−|u| du
2
Cd Mγd f (x)e−|x|
≤
Cj (x,t)
1
+2|x|((1−e−t )|x|+(1+aj )(1−e−2t ) 2 )
1
×((1 − e−t ) | x | +(1 + aj )(1 − e−2t ) 2 )d
2
Cd Mγd f (x)e−|x|
=
1
+2|x|((1−e−t )|x|+(1+aj )(1−e−2t ) 2
1
d
1
×(1 − e−t ) 2 ((1 − e−t ) 2 | x | +(1 + aj )(1 + e−t ) 2 )d
−|x|2 +2
Cd Mγd f (x)e
≤
−t d
2
×(1 − e
)
Ã
1
(1−e−t )
(1−e−2t ) 2
+2(1+aj )
1
t
t2
1
(1 − e−t ) 2
1
t2
1
−t 2
+ (1 + aj )(1 + e
)
!d
.
Therefore
Z
(1+aj )2 (1−e−2t )+2(1+aj )
f (u) du ≤ e
1
(1−e−2t ) 2
1
2
t
+e−2t |x|2
Cj (x,t)
Z
2
f (u)e−|u| du
Cj (x,t)
1
(1−e−2t ) 2
(1+aj )2 (1−e−2t )+2(1+aj )
1
t2
−|x|2 +2
+|x|2
Cd Mγd f (x)e
≤e
1
(1−e−t )
(1−e−2t ) 2
+2(1+aj )
1
t
t2
1
d
×(1 − e−t ) 2 (
(1+aj )
2
−1
(1−e 2
≤e
×
≤ e(1+aj )
since 0 < t <
1
4
and
2
−1
2
(1−e
Ã
t
)+4(1+aj )
1
2
1
+ (1 + aj )(1 + e−t ) 2 )d
1
(1−e−2t ) 2
1
t2
+
2(1−e−t )
t
1
(1 − e−2t ) 2
)+4(1+aj )
(1 − e−t ) 2
1
t2
√
2+2
1
−t 2
+ (1 + aj )(1 + e
)
!d
d
(1 − e−t ) 2
Cd Mγd f (x)
√
d
.(1 − e−t ) 2 (1 + (1 + aj ) 2)d Cd Mγd f (x),
1 − e−t
< 1, 1 + e−t < 2, if t > 0.
t
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
Non Tangential Convergence for the Ornstein-Uhlenbeck Semigroup.
115
Thus,
u(y, t) ≤
1
d
d
π 2 (1 − e−2t ) 2
≤ Cd Mγd f (x)
×
∞
X
∞
X
2
e−aj−1
j=1
Z
f (u)du
Cj (x,t)
1
d
d
d
π 2 (1 + e−t ) 2 (1 − e−t ) 2
−1
2
2
2
e−aj−1 e(1+aj )
(1−e
√
)+4(1+aj ) 2+2
j=1
≤ Cd Mγd f (x)
∞
1 X
π
1
2
2
e−aj−1 +(1+aj )
−1
2
2
(1−e
√
d
(1 − e−t ) 2 (1 + (1 + aj ) 2)d
√
)+4(1+aj ) 2+2
√
(1 + (1 + aj ) 2)d ,
j=1
since 1 + e−t ≥ 1. Now it is easy to see that
√
1
−a2j−1 + (1 + aj )2 (1 − e− 2 ) + 4(1 + aj ) 2 + 2
√
√
1
1
1p p
= 4 + 4 2 − e− 2 − [−(2(1 − e− 2 ) + 4 2) + e− 2 j] j,
which is negative for j sufficiently big, then
∞
X
2
e−aj−1 +(1+aj )
2
−1
2
(1−e
√
)+4(1+aj ) 2+2
j=1
√
.(1 + (1 + aj ). 2)d < ∞.
Thus u(y, t) ≤ Cd Mγd f (x) and since (y, t) ∈ Γp (x) is arbitrary
T ∗ f (x) =
sup
(y,t)∈Γp (x)
u(y, t) ≤ Cd Mγd f (x).
¤
Now we are ready to establish the convergence result for the Ornstein-Uhlenbeck
semigroup.
Theorem 1.2. The Ornstein-Uhlenbeck semigroup {Tt f } converges in L1 (γd ) a.e
if t → 0+ , for any function f ∈ L1 (γd ),
lim u(x, t) = f (x), a.e. x
t→0+
(18)
Moreover, if u(y, t) = Tt f (y) then u(y, t) tends to f (x) non tangentially ,i.e.
lim
p
(y,t)→x,(y,t)∈Γγ (x)
Tt f (y) = f (x), a.e. x.
Proof. We have,
u(y, t) =
1
d
d
π 2 (1 − e−2t ) 2
Z
Rd
− | e−t y − u |2
1 − e−2t
f (u)du,
e
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
(19)
116
Ebner Pineda, Wilfredo Urbina
considering
Ωf (x) = lim
α→0+
"
sup
p
(y,s)∈Γγ (x),0 δ
depends on x and Mγd g(x) < ǫ.
On the other hand, since
Thus g
u(y, t) − f (x) = u1 (y, t) − f1 (x) + u2 (y, t) − f2 (x)
where
1
i
u (y, t) =
d
d
π 2 (1 − e−2t ) 2
Z
Rd
− | e−t y − u |2
1 − e−2t
e
fi (u)du i = 1, 2,
then we get,
1
1
u (y, t) − f1 (x) =
=
+
1
d
d
π 2 (1 − e−2t ) 2
1
d
d
π 2 (1 − e−2t ) 2
d
d
π 2 (1 − e−2t ) 2
Z
Z
|x−u|≤δ
|x−u|>δ
Z
Rd
− | e−t y − u |2
1 − e−2t
e
(f1 (u) − f1 (x))du
− | e−t y − u |2
1 − e−2t
(f1 (u) − f1 (x))du
e
− | e−t y − u |2
1 − e−2t
e
(f1 (u) − f1 (x))du.
Now we have that if | x |≤ k − 1 and (y, t) ∈ Γpγ (x) with t <
(y, t) ∈ Γp (x). Thus | u − x |≤ δ implies
1
|x|2
| u |=| u − x + x |≤| u − x | + | x |< δ + k − 1 < 1 + k − 1 = k
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
∧
1
,
4
then
Non Tangential Convergence for the Ornstein-Uhlenbeck Semigroup.
117
and then, f1 (u) = f (u) ∧ f1 (x) = f (x). Therefore
¯
¯
¯
¯Z
− | e−t y − u |2
¯
¯
¯
¯
−2t
1−e
(f1 (u) − f1 (x))du¯
e
¯
¯
¯ |x−u|≤δ
¯
¯
¯
¯
−t
2
¯Z
¯
−
|
e
y
−
u
|
¯
¯
¯
¯
1 − e−2t
e
(f (u) − f (x))du¯
¯
¯ |x−u|≤δ
¯
¯
¯
1
d
d
π 2 (1 − e−2t ) 2
=
=
≤
1
d
d
2
π (1 − e−2t ) 2
¯
¯
¯Z
¯
− | e−t y − u |2
¯
¯
¯
¯
−2t
1−e
g(u)du¯
e
¯
¯ Rd
¯
¯
¯
1
d
d
π 2 (1 − e−2t ) 2
T ∗ g(x) ≤ Cd Mγd g(x) ≤ Cd ǫ.
1
Now observe that if (y, t) ∈ Γpγ (x) and t 2 ≤
u − x |≤| u − y | + | y − x | and thus
δ
2
then, | u − x |> δ implies δ δ
¯
¯
¯
d
2
Z
|u−y|> δ
2
|u−x|>δ
− | e−t y − u |2
1 − e−2t
e
du
− | e−t y − u |2
1 − e−2t
| f1 (u) | du
e
1
d
| f1 (x) |
Z
d
π 2 (1 − e−2t ) 2
| f1 (x) |
Z
|u−x|>δ
− | e−t y − u |2
1 − e−2t
du.
e
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
118
Ebner Pineda, Wilfredo Urbina
Now, we have
d
d
|u−y|> δ
2
d
Z
|u−y|> δ
,|u| δ
,|u| δ
,|u| 2δ , | u |< k implies that
Then for 0 < t < log
4k + δ
≤
| e−t y − u |
but 0 < t < log
µ
e−t
=
| e−t y − e−t u + e−t u − u |=| e−t (y − u) − (u − e−t u) |
≥
e−t | y − u | − | u − e−t u |= e−t | y − u | −(1 − e−t ) | u |
≥
e−t
δ
− k(1 − e−t ) = e−t
2
µ
4k + 2δ
4k + δ
¶
and therefore e−t >
µ
¶
−k
δ
+k
2
µ
δ
+k
2
¶
− k,
4k + δ
, then,
4k + 2δ
δ + 2k
2
¶
>
4k + δ
4k + 2δ
=
4k + δ
(2k + δ) − k
4(2k + δ)
=
4k + δ
4k + δ − 4k
δ
−k =
= .
4
4
4
−k
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
Non Tangential Convergence for the Ornstein-Uhlenbeck Semigroup.
Therefore | u − y |>
δ
δ
, | u |< k implies | e−t y − u |> and thus
2
4
1
d
≤
≤
d
π 2 (1 − e−2t ) 2
1
d
2
π (1 − e−2t )
−
e
d
2
Z
e
δ2
+k2
16(1−e−2t )
d
d
π 2 (1 − e−2t ) 2
|u−y|> δ
2
k2
Z
Z
− | e−t y − u |2
1 − e−2t
| f1 (u) | du
e
|u−y|> δ
,|u|δ
− | e−t y − u |2
1 − e−2t
du
e
−|s|2
e 1−e−2t ds
|x−s−e−t y|>δ
−|s|2
e 1−e−2t ds,
|x−s−e−t y|>δ
since, f1 (x) = f (x) as | x |≤µk − 1 < k.
¶
k − 1 − δ/2
Thus taking 0 < t < log
, | x − s − e−t y |> δ implies
k − 1 − 3δ/4
|s|
=
| s − x + e−t y + x − e−t y |=| s − x + e−t y − (e−t y − x) |
≥
| s − x + e−t y | − | e−t y − x | .
But
| e−t y − x |
1
Thus, since t 2 ≤
=
| e−t y − e−t x + e−t x − x |≤ e−t | y − x | +(1 − e−t ) | x |
≤
e−t t 2 + (1 − e−t )(k − 1).
1
δ
,
2
| s − x + e−t y | − | e−t y − x |
1
>
δ − e−t t 2 − (1 − e−t )(k − 1)
≥
δ − e−t
=
δ − (k − 1) + (k − 1 −
δ
− (k − 1)(1 − e−t )
2
δ −t
)e ,
2
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
119
120
Ebner Pineda, Wilfredo Urbina
and as 0 < t < log
µ
|s|
k − 1 − δ/2
k − 1 − 3δ/4
¶
, then e−t >
k − 1 − 3δ/4
. Hence,
k − 1 − δ/2
>
δ − (k − 1) + (k − 1 − δ/2)e−t
>
δ − (k − 1) + (k − 1 − δ/2)
=
δ − (k − 1) + k − 1 − 3δ/4 = δ − 3δ/4 =
k − 1 − 3δ/4
k − 1 − δ/2
δ
if 0 < t < log
Then | x − s − e y |> δ implies | s |>
4
s
Therefore, taking w = √
,
1 − e−2t
−t
d
Z
|u−y|> δ
2
d
Z
|s|> δ
4
1
d
π 2 (1 − e−2t ) 2
≤
=
d
| f (x) |
π 2 (1 − Ze−2t ) 2
| f (x) |
d
π2
|w|>
√
4
δ
.
4
µ
¶
k − 1 − δ/2
.
k − 1 − 3δ/4
− | e−t y − u |2
1 − e−2t
e
| f1 (u) | du
−|s|2
e 1 − e−2t ds
δ
2
e−|w| dw.
1−e−2t
Now since, | x |≤ k − 1 < k, thenf2 (x) = 0. Hence
| u2 (y, t) − f2 (x) |=| u2 (y, t) |≤ T ∗ f2 (x) ≤ Cd Mγd f2 (x)
for (y, t) ∈ Γp (x). Therefore,
| u(y, t) − f (x) |≤| u1 (y, t) − f1 (x) | + | u2 (y, t) − f2 (x) |
=| u1 (y, t) − f1 (x) | + | u2 (y, t) |
−δ
≤ Cd ǫ +
e 16(1−e−2t )
(1 −
+k2
e−2t )
d
2
kf k1,γd +
| f (x) |
π
d
2
Z
2
|w|>
√
4
δ
e−|w| dw
1−e−2t
+Cd Mγd f2 (x),
if (y, t) ∈ Γpγ (x) and
¶
µ
¶
¾
½
µ
k − 1 − δ/2
1
1
4k + 2δ
, log
, 2 ∧
=: a.
0 < t < min log
4k + δ
k − 1 − 3δ/4
|x|
4
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
Non Tangential Convergence for the Ornstein-Uhlenbeck Semigroup.
121
Thus taking supremum on (y, t) ∈ Γpγ (x), 0 < t < α < a and then taking α → 0+
we obtain,
Ωf (x) ≤ Cd (ǫ + Mγd f2 (x))
for all ǫ > 0 and almost every x with | x |≤ k − 1.
Given ǫ > 0, let us take k sufficiently large such that
kf2 k1,γd ≤ Cd ǫ2 ,
then by the estimation of Ω and the weak continuity of Mγd we get
γd ({x ∈ Rd :| x |≤ k − 1, Ωf (x) > ǫ}) ≤ ǫ
and that implies that Ωf (x) = 0 a.e.
¤
A similar proof for the Poisson-Hermite semigroup, using the non-tangential
maximal function defined as
Pγ∗ f (x) =
sup
(y,t)∈Γγ (x)
| Pt f (y) |,
(20)
and its analogous truncated version, should be possible but it has some technical
difficulties that we have been unable to overcome so far.
Let us now prove a general statement for families of linear operators that will
allow us to get a simpler proof of the non-tangential convergence, both for the
Ornstein-Uhlenbeck semigroup and also for the Poisson-Hermite semigroup. It is a
generalization of Theorem 2.2 of J. Duoandikoetxea’s book [1].
Theorem 1.3. Let {Tt }t>0 be a family of linear operators on Lp (Rd , µ) and for any
x ∈ Rd , let Γ(x) be a subset of Rd+1
such that x is in (Γ(x))′ , that is to say x is an
+
accumulation point of Γ(x). Let us define
T ∗ f (x) = sup{|Tt f (y)| : (y, t) ∈ Γ(x)},
for f ∈ Lp (Rd , µ) and x ∈ Rd . If T ∗ is weak (p, q) then the set
½
¾
S = f ∈ Lp (Rd , µ) :
lim
Tt f (y) = f (x) a.e.
(y,t)→x,(y,t)∈Γ(x)
is closed in Lp (Rd , µ).
Proof. Let us consider a sequence (fn ) in S such that fn → f in Lp (Rd , µ), then
|Tt f (y) − f (x)| − |Tt fn (y) − fn (x)| ≤ |Tt (f − fn )(y) − (f (x) − fn (x))|,
this implies that for each n ∈ N, for almost every x,
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
122
Ebner Pineda, Wilfredo Urbina
lim sup
|Tt f (y) − f (x)|
(y,t)→x,(y,t)∈Γ(x)
≤
(y,t)→x,(y,t)∈Γ(x)
lim sup
≤
(y,t)→x,(y,t)∈Γ(x)
lim sup
+
|Tt (f − fn )(y) − (f (x) − fn (x))|
|Tt (f − fn )(y)|
lim sup
(y,t)→x,(y,t)∈Γ(x)
|f (x) − fn (x)|
T ∗ (f − fn )(x) + |f (x) − fn (x)|.
≤
On the other hand, if we know that a ≤ b + c then a > λ implies b > λ2 ∨ c > λ2 .
Then, given λ > 0 and n ∈ N, lim sup(y,t)→x,(y,t)∈Γ(x) |Tt f (y)−f (x)| > λ implies
λ
λ
∨ |f (x) − fn (x)| > a.e.
2
2
T ∗ (f − fn )(x) >
and this implies that, given λ > 0,
Ã(
µ
x:
lim sup
(y,t)→x,(y,t)∈Γ(x)
≤µ
≤
for all n ∈ N. Therefore,
Ã(
µ
x:
µ
|Tt f (y) − f (x)| > λ
¾¶
λ
x : T ∗ (f − fn )(x) >
2
¾¶
µ½
λ
+µ
x : |f (x) − fn (x)| >
2
µ½
¢q ¡ 2
2C
kf − fn kp +
kf − fn kp
λ
λ
lim sup
(y,t)→x,(y,t)∈Γ(x)
x:
lim sup
(y,t)→x,(y,t)∈Γ(x)
|Tt f (y) − f (x)| > 0
¶p
,
)!
=0
)!
= 0,
|Tt f (y) − f (x)| > λ
and since this is true for all λ > 0, we get that
Ã(
µ
)!
as
(
x:
lim sup
(y,t)→x,(y,t)∈Γ(x)
=
|Tt f (y) − f (x)| > 0
∞
[
n=1
(
)
1
x:
lim sup
|Tt f (y) − f (x)| >
n
(y,t)→x,(y,t)∈Γ(x)
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
)
.
Non Tangential Convergence for the Ornstein-Uhlenbeck Semigroup.
123
Thus
lim
(y,t)→x,(y,t)∈Γ(x)
Tt f (y) = f (x) a.e.
and then f ∈ S. Therefore S is a closed set in Lp (Rd , µ).
¤
Finally, as a consequence of this result, we get the non-tangential convergence
for the Ornstein-Uhlenbeck semigroup {Tt }t>0 and the Poisson-Hermite semigroup
{Pt }t>0 .
Corollary 1.4. The Ornstein-Uhlenbeck semigroup {Tt }t>0 and the Poisson-Hermite semigroup {Pt }t>0 verify
lim
p
(y,t)→x,(y,t)∈Γγ (x)
lim
(y,t)→x,(y,t)∈Γγ (x)
Tt f (y) = f (x) a.e. x,
Pt f (y) = f (x) a.e. x.
Proof. Let us discuss the proof for the the Ornstein-Uhlenbeck semigroup {Tt }t>0 .
The proof for the Poisson-Hermite semigroup {Pt }t>0 is totally similar.
P
It is immediate
that
for any given polynomial f (x) = n
k=0 Jk f (x), since Tt f (y) =
¢
¡P
P
n
n
−tk
Jk f (y), we have the non-tangential convergence,
Tt
k=0 Jk f (y) =
k=0 e
lim
p
(y,t)→x,(y,t)∈Γγ (x)
Tt f (y) = f (x),
for all x ∈ Rd . Now considering the set
S=
(
p
f ∈ L (γd ) :
lim
p
(y,t)→x,(y,t)∈Γγ (x)
Tt f (y) = f (x) a.e.
)
,
corresponding to the Ornstein-Uhlenbeck semigroup, then the polynomials are in S.
From the previous result, since non-tangential maximal function for the OrnsteinUhlenbeck semigroup Tγ∗ f is weak (1, 1) with respect to the Gaussian measure, we
get that the set S is closed in Lp (γd ) and since the polynomials are dense in Lp (γd )
then S = Lp (γd ).
¤
Acknowledgement
We want to thank the referees for their suggestions and/or corrections that improved
the presentation of this paper.
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124
124
Ebner Pineda, Wilfredo Urbina
References
[1] Duoandikoetxea, J. Fourier Analysis, Graduated Studies in Mathematics, Volume 29, AMS, R. I., 2001.
[2] Forzani, L. Lemas de cubrimiento de tipo Besicovitch y su aplicaci´
on al estudio
del operador maximal de Ornstein-Uhlenbeck. Tesis de Doctorado. Universidad
Nacional de San Luis, Argentina, 1993.
[3] Forzani, L., Fabes, E. Unpublished manuscript, 1994.
[4] Guti´errez, C., Urbina, W., Estimates for the maximal operator of the OrnsteinUhlenbeck semigroup. Proc. Amer. Math. Soc. 113 (1991), 99–104.
[5] Sj¨
ogren P., Operators associated with the Hermite Semigroup-A Survey, J.
Fourier Anal. Appl., (3)(1997), 813–823.
[6] Stein E., Singular Integrals and Differentiability Properties of Functions,
Princeton Univ. Press, Princeton, New Jersey, 1970.
[7] Szeg¨
o, G., Orthogonal polynomials, rev. ed., Amer. Math. Soc. Colloq. Publ.,
vol. 23, Amer. Math. Soc., Providence, R. I., 1959.
[8] Urbina W.m An´
alisis Arm´
onico Gaussiano: una visi´
on panor´
amica, Trabajo
de Ascenso, Facultad de Ciencias, UCV, 1998. Available in
http://euler.ciens.ucv.ve/~wurbina/notes.html
[9] Zygmund, A., Trigonometric Series, 2nd. ed., Cambridge Univ. Press., Cambridge, 1959.
Divulgaciones Matem´
aticas Vol. 16 No. 1(2008), pp. 107–124