Simba23032018 SergiBaena

Four definitions for the fractional Laplacian
N. Accomazzo (UPV/EHU), S. Baena (UB), A. Becerra Tomé (US), J.

Martı́nez (BCAM), A. Rodrı́guez (UCM), I. Soler (UM)
VIII Escuela-Taller de Análisis Funcional

Basque Center for Applied Mathematics (BCAM)
Nap time, March 8, 20181
1 International Women’s Day

Group 3 Fractional Laplacian VIII Escuela-Taller 1 / 40
A pointwise definition of the
fractional Laplacian

Laplace fractional operator: several points of view
Functional analysis: M. Riesz, S. Bochner, W. Feller, E. Hille, R. S. Phillips,

A. V. Balakrishnan, T. Kato, Martı́nez–Carracedo y Sanz–Alix, K. Yosida
potencial theory for fractional laplacian: N. S. Landkof
Lévy’s processes: K. Bogdan e.a.
Partial Derivative Ecuations: L. Caffarelli y L. Silvestre
Scattering theory: C. R. Graham y M. Zworski, S-Y. A. Chang y M.d.M.

González
Kato’s square root (solved by P. Auscher e.a.)

Laplace fractional operator: several points of view
Functional analysis: M. Riesz, S. Bochner, W. Feller, E. Hille, R. S. Phillips,

A. V. Balakrishnan, T. Kato, Martı́nez–Carracedo y Sanz–Alix, K. Yosida
potencial theory for fractional laplacian: N. S. Landkof
Lévy’s processes: K. Bogdan e.a.
Partial Derivative Ecuations: L. Caffarelli y L. Silvestre
Scattering theory: C. R. Graham y M. Zworski, S-Y. A. Chang y M.d.M.

González
Kato’s square root (solved by P. Auscher e.a.)
Basic example of fractional operator : fractional laplacean

The working space
We are going to work with the space S (Rn ) of L. Schwartz’ rapidly decreasing
functions.

The working space
functions.
S (Rn ) is the space C ∞ (Rn ) of functions that
kf kp = sup sup (1 + |x |2 )p/2 |∂α f (x )| < ∞ p ∈ N ∪ {0}

|α|≤p x ∈Rn

The working space
functions.
S (Rn ) is the space C ∞ (Rn ) of functions that
kf kp = sup sup (1 + |x |2 )p/2 |∂α f (x )| < ∞ p ∈ N ∪ {0}

|α|≤p x ∈Rn
This space endowed with the metric topology

∞ kf − g kp
d (f , g ) = ∑ 2−p 1 + k f − g kp
p =0

First definition motivation
Let f ∈ C 2 (a, b ), then for every x ∈ (a, b ) one has
2f (x ) − f (x + y ) − f (x − y )
−f 00 (x ) = lı́m
y →0 y2

2f (x ) − f (x + y ) − f (x − y )
−f 00 (x ) = lı́m
y →0 y2
If we introduce the spherical and solid averaging operators

ˆ x +y
f (x + y ) + f (x − y ) 1
My f ( x ) = Ay f ( x ) = f (t ) dt
2 2y x −y

2f (x ) − f (x + y ) − f (x − y )
−f 00 (x ) = lı́m
y →0 y2
If we introduce the spherical and solid averaging operators

ˆ x +y
f (x + y ) + f (x − y ) 1
My f ( x ) = Ay f ( x ) = f (t ) dt
2 2y x −y
then we can reformulate −f 00 (x ) like this
f (x ) − My f (x ) f ( x ) − Ay f ( x )
−f 00 (x ) = 2 lı́m 2
= 6 lı́m
y →0 y y →0 y2

n
∂2 f
Thus, if we bear in mind that −∆f = − ∑ 2
and making an extension of the
k =1 ∂xk
spherical and solid averaging operators
ˆ ˆ
1 1
My f ( x ) = f (y ) d σ (y ) Ay f ( x ) = f (y ) dy
σ(n−1) r (n−1) S (x ,r ) ωn r n B ( x ,r )

n
∂2 f
Thus, if we bear in mind that −∆f = − ∑ 2
and making an extension of the
k =1 ∂xk
spherical and solid averaging operators
ˆ ˆ
1 1
My f ( x ) = f (y ) d σ (y ) Ay f ( x ) = f (y ) dy
σ(n−1) r (n−1) S (x ,r ) ωn r n B ( x ,r )
then we have
f ( x ) − My f ( x ) f ( x ) − Ay f ( x )
−∆f (x ) = 2n lı́m 2
= 2(n + 2) lı́m
y →0 y y →0 y2

Finally, as a generalization of the operator (−∆)f one can define (−∆)s f as an
Rn non local operator. If we have u ∈ S (Rn ), we define:

ˆ
γ(n, s ) 2u (x ) − u (x + y ) − u (x − y )
(−∆) u (x ) =
s
dy , s ∈ (0, 1)
2 Rn |y |n+2s

ˆ
γ(n, s ) 2u (x ) − u (x + y ) − u (x − y )
(−∆) u (x ) =
s
dy , s ∈ (0, 1)
2 Rn |y |n+2s
Observation 1: γ(n, s ) is, at this moment, an unknown constant

ˆ
γ(n, s ) 2u (x ) − u (x + y ) − u (x − y )
(−∆) u (x ) =
s
dy , s ∈ (0, 1)
2 Rn |y |n+2s

Observation 2: this is a linear operator.

ˆ
γ(n, s ) 2u (x ) − u (x + y ) − u (x − y )
(−∆) u (x ) =
s
dy , s ∈ (0, 1)
2 Rn |y |n+2s

Observation 2: this is a linear operator.
Observation 3: since as s → 1− the fractional Laplacean tends (at least, formally
right now) to (−∆), one might surmise that in the regime 1/2 < s < 1 the
operator (−∆)s should display properties closer to those of the classical
Laplacian, whereas since (−∆)s → I as s → 0+ , the stronger discrepancies
might present themselves in the range 0 < s < 1/2

(−∆u (x )) is well-defined
It is important to observe that the integral in the right-hand side is convergent. In

order to see this, it suffices to write:
ˆ ˆ
2u (x ) − u (x + y ) − u (x − y ) 2u (x ) − u (x + y ) − u (x − y )
dy = dy +
Rn |y |n+2s |y |≤1 | y | n + 2s
ˆ
2u (x ) − u (x + y ) − u (x − y )
+ dy = (a) + (b )
|y |≥1 | y | n + 2s


ˆ ˆ
dy = dy +
Rn |y |n+2s |y |≤1 | y | n + 2s
ˆ
2u (x ) − u (x + y ) − u (x − y )
+ dy = (a) + (b )
|y |≥1 | y | n + 2s
Starting with (a):
ˆ
|2u (x ) − u (x + y ) − u (x − y )|
|(a)| ≤ dy
|y |≤1 |y |n+2s


ˆ ˆ
dy = dy +
Rn |y |n+2s |y |≤1 | y | n + 2s
ˆ
2u (x ) − u (x + y ) − u (x − y )
+ dy = (a) + (b )
|y |≥1 | y | n + 2s
Starting with (a):
ˆ
|2u (x ) − u (x + y ) − u (x − y )|
|(a)| ≤ dy
|y |≤1 |y |n+2s
Using Taylor 2u (x ) − u (x + y ) − u (x − y ) = −h∇2 u (x )y , y i + o (|x |3 )

Therefore
ˆ
|h∇2 u (x )y , y i + o (|x |3 )|
= dy
|y |≤1 |y |n+2s

Therefore
ˆ
|h∇2 u (x )y , y i + o (|x |3 )|
= dy
|y |≤1 |y |n+2s
Using Cauchy-Schwarz inequality
ˆ ˆ
|∇2 u (x )y ||y | + |o (|x |3 )| |∇2 u (x )||y |2 + |o (|x |3 )|
≤ 2
dy ≤ dy
|y |≤1 |y | n + s
|y |≤1 |y |n+2s
ˆ
1
=C dy =
|y |≤1 | y | n −2(1−s )

Therefore
ˆ
|h∇2 u (x )y , y i + o (|x |3 )|
= dy
|y |≤1 |y |n+2s
ˆ ˆ
|∇2 u (x )y ||y | + |o (|x |3 )| |∇2 u (x )||y |2 + |o (|x |3 )|
≤ 2
dy ≤ dy
|y |≤1 |y | n + s
|y |≤1 |y |n+2s
ˆ
1
=C dy =
|y |≤1 | y | n −2(1−s )

Therefore
ˆ
|h∇2 u (x )y , y i + o (|x |3 )|
= dy
|y |≤1 |y |n+2s
ˆ ˆ
|∇2 u (x )y ||y | + |o (|x |3 )| |∇2 u (x )||y |2 + |o (|x |3 )|
≤ 2
dy ≤ dy
|y |≤1 |y | n + s
|y |≤1 |y |n+2s
ˆ ˆ 1ˆ
1 1
=C dy = d σ(ω ) dr < ∞
|y |≤1 | y | n −2(1−s ) 0 Sn − 1 r 1−2s

Therefore
ˆ
|h∇2 u (x )y , y i + o (|x |3 )|
= dy
|y |≤1 |y |n+2s
ˆ ˆ
|∇2 u (x )y ||y | + |o (|x |3 )| |∇2 u (x )||y |2 + |o (|x |3 )|
≤ 2
dy ≤ dy
|y |≤1 |y | n + s
|y |≤1 |y |n+2s
ˆ ˆ 1ˆ
1 1
=C dy = d σ(ω ) dr < ∞
|y |≤1 | y | n −2(1−s ) 0 Sn − 1 r 1−2s
Now, for (b):

ˆ ˆ
2u (x ) − u (x + y ) − u (x − y ) 1
dy ≤ 4 k u k dy < ∞

n + 2 s L∞ (R )
n
n +2s
|y | |y |≥1 |y |

|y |≥1

Translations and dilations
Let h ∈ Rn and λ > 0, the translation and dilation operators are defined,
respectively, by
τh f (x ) = f (x + h ); δλ f (x ) = f (λx )
for every f : Rn → R and every x ∈ Rn

respectively, by
τh f (x ) = f (x + h ); δλ f (x ) = f (λx )
Proposition 1
Let u ∈ S(Rn ), then for every h ∈ Rn and λ > 0 we have
(−∆)s (τh u ) = τh ((−∆)s u )
and
(−∆)s (δλ u ) = λ2s δλ ((−∆)s u )

respectively, by
τh f (x ) = f (x + h ); δλ f (x ) = f (λx )
Proposition 1
Let u ∈ S(Rn ), then for every h ∈ Rn and λ > 0 we have
(−∆)s (τh u ) = τh ((−∆)s u )
and
(−∆)s (δλ u ) = λ2s δλ ((−∆)s u )
In particular, (−∆)s is a homogeneous operator of order 2s.

Orthogonal group
Recall that the orthogonal group is
O(n ) = {T ∈ Mn (R) : T t T = TT t = I }

Orthogonal group
O(n ) = {T ∈ Mn (R) : T t T = TT t = I }
The usual Laplacian satisfies ∆(u ◦ T ) = ∆u ◦ T for every T ∈ O(n ).
What about the fractional Laplacian?

Orthogonal group
O(n ) = {T ∈ Mn (R) : T t T = TT t = I }
We say that a function f : Rn → R has spherical symmetry if f (x ) = f ∗ (|x |) for

some f ∗ : Rn → R or, equivalently, if f (Tx ) = f (x ) for every T ∈ O(n ) and
every x ∈ R

Orthogonal group
O(n ) = {T ∈ Mn (R) : T t T = TT t = I }
We say that a function f : Rn → R has spherical symmetry if f (x ) = f ∗ (|x |) for

some f ∗ : Rn → R or, equivalently, if f (Tx ) = f (x ) for every T ∈ O(n ) and
every x ∈ R
Proposition 2
Let u ∈ S(Rn ) (actually, it is enough that u ∈ C 2 (Rn ) ∩ L∞ (Rn )) be a function
with spherical symmetry. Then, (−∆)s has spherical symmetry.

Orthogonal group (Cont.)
PROOF.-

PROOF.-
Let’s see that (−∆)s u (Tx ) = (−∆)s u (x ) for each T ∈ O(n ) and each x ∈ Rn

PROOF.-
ˆ
γ(n, s ) 2u ∗ (|Tx |) − u ∗ (|Tx + y |) − u ∗ (|Tx − y |)
(−∆)s u (Tx ) = dy
2 Rn |y |2n+s
ˆ
γ(n, s ) 2u ∗ (|x |) − u ∗ (|x + T t y |) − u ∗ (|x − T t y |)
= dy
2 Rn |y |2n+s

PROOF.-
ˆ
γ(n, s ) 2u ∗ (|Tx |) − u ∗ (|Tx + y |) − u ∗ (|Tx − y |)
(−∆)s u (Tx ) = dy
2 Rn |y |2n+s
ˆ
= dy
2 Rn |y |2n+s
Change of variable: z = T t y

PROOF.-
ˆ
γ(n, s ) 2u ∗ (|Tx |) − u ∗ (|Tx + y |) − u ∗ (|Tx − y |)
(−∆)s u (Tx ) = dy
2 Rn |y |2n+s
ˆ
= dy
2 Rn |y |2n+s
ˆ
γ(n, s ) 2u ∗ (|x |) − u ∗ (|x + z |) − u ∗ (|x − z |)
= dz
2 Rn |Tz |2n+s

PROOF.-
ˆ
γ(n, s ) 2u ∗ (|Tx |) − u ∗ (|Tx + y |) − u ∗ (|Tx − y |)
(−∆)s u (Tx ) = dy
2 Rn |y |2n+s
ˆ
= dy
2 Rn |y |2n+s
ˆ
γ(n, s ) 2u ∗ (|x |) − u ∗ (|x + z |) − u ∗ (|x − z |)
= dz
2 Rn |Tz |2n+s
ˆ
γ(n, s ) 2u ∗ (|x |) − u ∗ (|x + z |) − u ∗ (|x − z |)
= dz = (−∆)s u (x )
2 Rn |z |2n+s

Alternative expression for the fractional Laplacian
Now we find a new pointwise expression for the fractional Laplacian which will be
useful when we prove the equivalence of the different definitions.

Alternative expression for the fractional Laplacian
Now we find a new pointwise expression for the fractional Laplacian which will be
useful when we prove the equivalence of the different definitions.
Theorem
Let u ∈ S(Rn ), then
ˆ
u (x ) − u (y )
(−∆)s u (x ) = γ(n, s )P.V . dy
Rn | x − y | n + 2s
where P.V. means the Cauchy’s principal value, i.e.
ˆ ˆ
u (x ) − u (y ) u (x ) − u (y )
P.V . n + 2s
dy = lim n+2s
dy
Rn | x − y | ε →0 +
|x −y |>ε |x − y |

Alternative expression for the fractional Laplacian (Cont.)
PROOF.-

PROOF.-
ˆ
1 2u (x ) − u (x + y ) − u (x − y )
(−∆) u (x ) = lim
s
dy
2 ε →0+ |y |>ε |y |n+2s

PROOF.-
ˆ
1 2u (x ) − u (x + y ) − u (x − y )
(−∆) u (x ) = lim
s
dy
2 ε →0+ |y |>ε |y |n+2s
ˆ ˆ
1 u (x ) − u (x + y ) 1 u (x ) − u (x − y )
= lim 2
dy + lim dy
2 ε →0+ |y |>ε |y | n + s 2 ε →0+ |y |>ε |y |n+2s

PROOF.-
ˆ
1 2u (x ) − u (x + y ) − u (x − y )
(−∆) u (x ) = lim
s
dy
2 ε →0+ |y |>ε |y |n+2s
ˆ ˆ
1 u (x ) − u (x + y ) 1 u (x ) − u (x − y )
= lim 2
dy + lim dy
2 ε →0+ |y |>ε |y | n + s 2 ε →0+ |y |>ε |y |n+2s
Changes of variables:
x + y = z in the first integral
x − y = z in the second integral

PROOF.-
ˆ
1 2u (x ) − u (x + y ) − u (x − y )
(−∆) u (x ) = lim
s
dy
2 ε →0+ |y |>ε |y |n+2s
ˆ ˆ
1 u (x ) − u (x + y ) 1 u (x ) − u (x − y )
= lim 2
dy + lim dy
2 ε →0+ |y |>ε |y | n + s 2 ε →0+ |y |>ε |y |n+2s
ˆ ˆ
1 u (x ) − u (z ) 1 u (x ) − u (z )
= lim dz + lim dz
2 ε →0+ |z −x |>ε |z − x |n+2s 2 ε →0+ |x −z |>ε | x − z | n + 2s

PROOF.-
ˆ
1 2u (x ) − u (x + y ) − u (x − y )
(−∆) u (x ) = lim
s
dy
2 ε →0+ |y |>ε |y |n+2s
ˆ ˆ
1 u (x ) − u (x + y ) 1 u (x ) − u (x − y )
= lim 2
dy + lim dy
2 ε →0+ |y |>ε |y | n + s 2 ε →0+ |y |>ε |y |n+2s
ˆ ˆ
1 u (x ) − u (z ) 1 u (x ) − u (z )
= lim dz + lim dz
2 ε →0+ |z −x |>ε |z − x |n+2s 2 ε →0+ |x −z |>ε | x − z | n + 2s
ˆ
u (x ) − u (z )
= lim dz
ε →0+ |x −z |>ε |x − z |n+2s
Another two definitions of the
fractional Laplacian

Fourier transform
We recall the definition of the Fourier transform, F , of a function f ∈ S(Rn ):

ˆ
F (f )(ξ ) = fˆ(ξ ) = (2π )−n/2 f (x )e −ix ·ξ dx, ξ ∈ Rn ,
Rn

Fourier transform

ˆ
−n/2
ˆ
F (f )(ξ ) = f (ξ ) = (2π ) f (x )e −ix ·ξ dx, ξ ∈ Rn ,
Rn
whose inverse function is given by
ˆ
F −1 (f )(x ) = (2π )−n/2 f (ξ )e ix ·ξ d ξ, x ∈ Rn ,
Rn

Fourier transform

ˆ
F (f )(ξ ) = fˆ(ξ ) = f (x )e −2πix ·ξ dx, ξ ∈ Rn ,
Rn

ˆ
F −1 (f )(x ) = f (ξ )e 2πix ·ξ d ξ, x ∈ Rn ,
Rn

Fourier transform

ˆ
F (f )(ξ ) = fˆ(ξ ) = f (x )e −2πix ·ξ dx, ξ ∈ Rn ,
Rn

ˆ
F −1 (f )(x ) = f (ξ )e 2πix ·ξ d ξ, x ∈ Rn ,
Rn
so that
ˆ
−1 −n/2
f (x ) = F ◦ F (f )(x ) = (2π ) fˆ(ξ )e ix ·ξ d ξ, x ∈ Rn .
Rn

Definition of (−∆)s via the heat semigroup e t∆
We will define (−∆)s f in terms of the heat semigroup e t∆ , which is nothing but
an operator such that maps every function f ∈ S(Rn ) to the solution of the heat
equation with initial data given by f :
(
vt = ∆v , (x, t ) ∈ Rn × (0, ∞)
v (x, 0) = f (x ), x ∈ Rn .
Using Fourier transform and its inverse and with a bit of magic, we can write
ˆ ˆ
t∆ −n/2 −t | ξ |2 ˆ ix ·ξ
e f (x ) := v (x, t ) = (2π ) e f ( ξ )e d ξ = Wt (x − z )f (z )dz,
Rn Rn
where
|x |2
Wt (x ) = (4πt )−n/2 e − 4t , x ∈ Rn ,
is the Gauss-Weierstrass kernel.

Definition of (−∆)s via the heat semigroup e t∆
Inspired by the following numerical identity: for λ > 0,

ˆ ∞
1 dt
s
λ = (e −tλ − 1) 1+s , 0 < s < 1,
Γ(−s ) 0 t
where ˆ ∞
dr
Γ(−s ) = (e −r − 1 ) < 0;
0 r 1+s
we can think of (−∆)s as the following operator
ˆ ∞
1 dt
(−∆)s f (x ) ∼ (e t∆ f (x ) − f (x )) 1+s , 0 < s < 1.
Γ(−s ) 0 t

Definition of (−∆)s via the Fourier Transform
By the well-known properties of F with respect to derivatives, we have that, for

f ∈ S(Rn ),
F [−∆f ](ξ ) = |ξ |2 F (f )(ξ ), ξ ∈ Rn ,
so it is reasonable to write something like
(−∆)s f (x ) ∼ F −1 [| · |2s F (f )](x ), x ∈ Rn , 0 < s < 1.


f ∈ S(Rn ),
F [−∆f ](ξ ) = |ξ |2 F (f )(ξ ), ξ ∈ Rn ,
(−∆)s f (x ) ∼ F −1 [| · |2s F (f )](x ), x ∈ Rn , 0 < s < 1.


f ∈ S(Rn ),
F [−∆f ](ξ ) = |2πξ |2 F (f )(ξ ), ξ ∈ Rn ,
(−∆)s f (x ) ∼ F −1 [|2π · |2s F (f )](x ), x ∈ Rn , 0 < s < 1.

∼ is =
Theorem (Lemma 2.1. P. Stinga’s PhD thesis)

Given f ∈ S(Rn ) and 0 < s < 1,
ˆ ∞
1 dt
F −1 [| · |2s F (f )](x ) = (e t∆ f (x ) − f (x )) 1+s , x ∈ Rn
Γ(−s ) 0 t
and this two functions coincide in a pointwise way with (−∆)s f (x ) when the
constant γ(n, s ) in its definition is given by
4s Γ(n/2 + s )
γ(n, s ) = > 0.
−π n/2 Γ(−s )

Everybody wants to be the fractional Laplacian
Let x ∈ Rn . By Fubini’s theorem and inverse Fourier formula,

ˆ ∞ ˆ ˆ ∞
1 dt 1 2 dt
(e t∆ f (x ) − f (x )) = (e −t |ξ | − 1) 1+s fˆ(ξ )e ix ·ξ d ξ
Γ(−s ) 0 t 1+s Γ(−s ) Rn 0 t
ˆ ˆ ∞
1 dr
= (e −r − 1) 1+s |ξ |2s fˆ(y )e ix ·ξ dy
Γ(−s ) Rn 0 r
ˆ
= |ξ |2s fˆ(ξ )e ix ·ξ d ξ = F −1 [| · |2s F (f )](x ).
Rn
Since f ∈ S(Rn ), we have that

ˆ ∞
dt
|e t∆ f (x ) − f (x )| 1+s < ∞,
0 t
and so Tonelli authorises us to apply Fubini’s theorem.


ˆ ∞ ˆ ˆ ∞
1 dt 1 2 dt
Γ(−s ) 0 t 1+s Γ(−s ) Rn 0 t
ˆ ˆ ∞
1 dr
= (e −r − 1) 1+s |ξ |2s fˆ(y )e ix ·ξ dy
Γ(−s ) Rn 0 r
ˆ
Rn

ˆ ∞
dt
|e t∆ f (x ) − f (x )| 1+s < ∞,
0 t


ˆ ∞ ˆ ˆ ∞
1 dt 1 2 dt
Γ(−s ) 0 t 1+s Γ(−s ) Rn 0 t
ˆ ˆ ∞
1 dr
= (e −r − 1) 1+s |ξ |2s fˆ(y )e ix ·ξ dy
Γ(−s ) Rn 0 r
ˆ
Rn

ˆ ∞
dt
|e t∆ f (x ) − f (x )| 1+s < ∞,
0 t


ˆ ∞ ˆ ˆ ∞
1 dt 1 2 dt
Γ(−s ) 0 t 1+s Γ(−s ) Rn 0 t
ˆ ˆ ∞
1 dr
= (e −r − 1) 1+s |ξ |2s fˆ(y )e ix ·ξ dy
Γ(−s ) Rn 0 r
ˆ
Rn

ˆ ∞
dt
|e t∆ f (x ) − f (x )| 1+s < ∞,
0 t


ˆ ∞ ˆ ˆ ∞
1 dt 1 2 dt
Γ(−s ) 0 t 1+s Γ(−s ) Rn 0 t
ˆ ˆ ∞
1 dr
= (e −r − 1) 1+s |ξ |2s fˆ(y )e ix ·ξ dy
Γ(−s ) Rn 0 r
ˆ
Rn

ˆ ∞
dt
|e t∆ f (x ) − f (x )| 1+s < ∞,
0 t

Next, we will see that

ˆ ∞ ˆ
1 dt 4s Γ(n/2 + s ) f (x ) − f (z )
(e t∆ f (x ) − f (x )) = P.V. dz, x ∈ Rn .
Γ(−s ) 0 t 1+s −π n/2 Γ(−s ) Rn |x − z |n+2s
Let ε > 0. Using that kWt (x − ·)kL1 (Rn ) = 1 for any x ∈ Rn ,

ˆ ∞ ˆ ∞ˆ
dt dt
(e t∆ f (x ) − f (x )) = Wt (x − z )(f (z ) − f (x ))dz
0 t 1+s 0 Rn t 1+s
= Iε + IIε .


ˆ ∞ ˆ
1 dt 4s Γ(n/2 + s ) f (x ) − f (z )
(e t∆ f (x ) − f (x )) = P.V. dz, x ∈ Rn .
Γ(−s ) 0 t 1+s −π n/2 Γ(−s ) Rn |x − z |n+2s

ˆ ∞ ˆ ∞ˆ
dt dt
0 t 1+s 0 Rn t 1+s
= Iε + IIε .


ˆ ∞ ˆ
1 dt 4s Γ(n/2 + s ) f (x ) − f (z )
(e t∆ f (x ) − f (x )) = P.V. dz, x ∈ Rn .
Γ(−s ) 0 t 1+s −π n/2 Γ(−s ) Rn |x − z |n+2s

ˆ ∞ ˆ ∞ˆ
dt dt
0 t 1+s 0 Rn t 1+s
= Iε + IIε .


ˆ ∞ ˆ
1 dt 4s Γ(n/2 + s ) f (x ) − f (z )
(e t∆ f (x ) − f (x )) = P.V. dz, x ∈ Rn .
Γ(−s ) 0 t 1+s −π n/2 Γ(−s ) Rn |x − z |n+2s

ˆ ∞ ˆ ∞ˆ
dt dt
0 t 1+s 0 Rn t 1+s
= Iε + IIε .


ˆ ∞ ˆ
1 dt 4s Γ(n/2 + s ) f (x ) − f (z )
(e t∆ f (x ) − f (x )) = P.V. dz, x ∈ Rn .
Γ(−s ) 0 t 1+s −π n/2 Γ(−s ) Rn |x − z |n+2s

ˆ ∞ ˆ ∞ˆ
dt dt
0 t 1+s 0 Rn t 1+s
= Iε + IIε .

Using Fubini’s theorem,

ˆ ˆ ∞ |x −z |2 dt
Iε = (4πt )−n/2 e − 4t (f (z ) − f (x )) dz
|x −z |>ε 0 t 1+s
ˆ ˆ ∞ |x −z |2 dt
= (f (z ) − f (x )) (4πt )−n/2 e − 4t dz
|x −z |>ε 0 t 1+s
ˆ
4s Γ(n/2 + s ) 1
= (f (x ) − f (z )) dz
|x −z |>ε −π n/2 |x − z |n+2s
|x −z |2
where we used the change of variables r = 4t .
Observe that Iε converges absolutely for any ε > 0 since f is bounded, so the use
of Fubini’s theorem is licit.


ˆ ˆ ∞ |x −z |2 dt
Iε = (4πt )−n/2 e − 4t (f (z ) − f (x )) dz
|x −z |>ε 0 t 1+s
ˆ ˆ ∞ |x −z |2 dt
= (f (z ) − f (x )) (4πt )−n/2 e − 4t dz
|x −z |>ε 0 t 1+s
ˆ
4s Γ(n/2 + s ) 1
= (f (x ) − f (z )) dz
|x −z |>ε −π n/2 |x − z |n+2s
|x −z |2


ˆ ˆ ∞ |x −z |2 dt
Iε = (4πt )−n/2 e − 4t (f (z ) − f (x )) dz
|x −z |>ε 0 t 1+s
ˆ ˆ ∞ |x −z |2 dt
= (f (z ) − f (x )) (4πt )−n/2 e − 4t dz
|x −z |>ε 0 t 1+s
ˆ
4s Γ(n/2 + s ) 1
= (f (x ) − f (z )) dz
|x −z |>ε −π n/2 |x − z |n+2s
|x −z |2


ˆ ˆ ∞ |x −z |2 dt
Iε = (4πt )−n/2 e − 4t (f (z ) − f (x )) dz
|x −z |>ε 0 t 1+s
ˆ ˆ ∞ |x −z |2 dt
= (f (z ) − f (x )) (4πt )−n/2 e − 4t dz
|x −z |>ε 0 t 1+s
ˆ
4s Γ(n/2 + s ) 1
= (f (x ) − f (z )) dz
|x −z |>ε −π n/2 |x − z |n+2s
|x −z |2

Using polar coordinates,
ˆ ∞ˆ dt
IIε = Wt (x − z )(f (z ) − f (x ))dz
0 |x −z |<ε t 1+s
ˆ ∞ ˆ ε ˆ
r2 dt
= (4πt )−n/2 e − 4t r n−1 (f (x + rz 0 ) − f (z ))dS (z 0 )dr .
0 0 |z 0 |=1 t 1+s
By Taylor’s theorem, using the symmetry of the sphere,

ˆ
(f (x + rz 0 ) − f (z ))dS (z 0 ) = Cn r 2 ∆f (x ) + O (r 3 ),
|z 0 |=1
thus
ˆ ˆ ∞ r2
ε e − 4t dt
|IIε | ≤ Cn,∆f (x ) r n +1
0 0 t n/2+s t
ˆ ε
= Cn,∆f (x ) r n+1 Cn,s r −n−2s dr = Cn,∆f (x ),s ε2(1−s ) .
0

ˆ ∞ˆ dt
IIε = Wt (x − z )(f (z ) − f (x ))dz
0 |x −z |<ε t 1+s
ˆ ∞ ˆ ε ˆ
r2 dt
0 0 |z 0 |=1 t 1+s

ˆ
|z 0 |=1
thus
ˆ ˆ ∞ r2
ε e − 4t dt
|IIε | ≤ Cn,∆f (x ) r n +1
0 0 t n/2+s t
ˆ ε
0

ˆ ∞ˆ dt
IIε = Wt (x − z )(f (z ) − f (x ))dz
0 |x −z |<ε t 1+s
ˆ ∞ ˆ ε ˆ
r2 dt
0 0 |z 0 |=1 t 1+s

ˆ
|z 0 |=1
thus
ˆ ˆ ∞ r2
ε e − 4t dt
|IIε | ≤ Cn,∆f (x ) r n +1
0 0 t n/2+s t
ˆ ε
0

ˆ ∞ˆ dt
IIε = Wt (x − z )(f (z ) − f (x ))dz
0 |x −z |<ε t 1+s
ˆ ∞ ˆ ε ˆ
r2 dt
0 0 |z 0 |=1 t 1+s

ˆ
|z 0 |=1
thus
ˆ ˆ ∞ r2
ε e − 4t dt
|IIε | ≤ Cn,∆f (x ) r n +1
0 0 t n/2+s t
ˆ ε
0

ˆ ∞ˆ dt
IIε = Wt (x − z )(f (z ) − f (x ))dz
0 |x −z |<ε t 1+s
ˆ ∞ ˆ ε ˆ
r2 dt
0 0 |z 0 |=1 t 1+s

ˆ
|z 0 |=1
thus
ˆ ˆ ∞ r2
ε e − 4t dt
|IIε | ≤ Cn,∆f (x ) r n +1
0 0 t n/2+s t
ˆ ε
0

Everybody IS the fractional Laplacian
This proves that IIε → 0 as ε → 0, so

ˆ ∞ˆ dt
Wt (x − z )(f (z ) − f (x ))dz = lı́m Iε + IIε
0 Rn t 1+s ε →0
ˆ
4s Γ(n/2 + s ) f (x ) − f (z )
= P.V. n+2s
dz
−π n/2 Rn | x − z |
. .
^
This kind of computations (bearing in mind the exact expression of the constant
γ(n, s )) also prove the following pointwise convergence
(−∆)s f (x ) → −∆f (x ), x ∈ Rn as s → 0+ ,
when f ∈ C 2 (Rn ) ∩ L∞ (Rn ) (observe that, in S(Rn ) this is obvious by the

definition via Fourier transform).


ˆ ∞ˆ dt
0 Rn t 1+s ε →0
ˆ
4s Γ(n/2 + s ) f (x ) − f (z )
= P.V. n+2s
dz
−π n/2 Rn | x − z |
. .
^
(−∆)s f (x ) → −∆f (x ), x ∈ Rn as s → 0+ ,



ˆ ∞ˆ dt
0 Rn t 1+s ε →0
ˆ
4s Γ(n/2 + s ) f (x ) − f (z )
= P.V. n+2s
dz
−π n/2 Rn | x − z |
. .
^
(−∆)s f (x ) → −∆f (x ), x ∈ Rn as s → 0+ ,



ˆ ∞ˆ dt
0 Rn t 1+s ε →0
ˆ
4s Γ(n/2 + s ) f (x ) − f (z )
= P.V. n+2s
dz
−π n/2 Rn | x − z |
. .
^
(−∆)s f (x ) → −∆f (x ), x ∈ Rn as s → 0+ ,


And last but not least

Extension Problem
Let s ∈ (0, 1) and consider a = 1 − 2s. We want to solve the extension problem
 La U (x, y ) = divx ,y (y a ∇x ,y U ) = 0, x ∈ Rn+ , y > 0,


U (x, 0) = u (x ),
U (x, y ) → 0 as y → ∞.

The previous system can be written as


 −∆x U (x, y ) = ∂yy + ya ∂y U (x, y ),
 x ∈ Rn+ , y > 0,
U (x, 0) = u (x ), (1)
U (x, y ) → 0 as y → ∞.



Extension Problem


U (x, 0) = u (x ),
U (x, y ) → 0 as y → ∞.



 −∆x U (x, y ) = ∂yy + ya ∂y U (x, y ),
 x ∈ Rn+ , y > 0,
U (x, 0) = u (x ), (1)
U (x, y ) → 0 as y → ∞.



Extension Problem


U (x, 0) = u (x ),
U (x, y ) → 0 as y → ∞.



 −∆x U (x, y ) = ∂yy + ya ∂y U (x, y ),
 x ∈ Rn+ , y > 0,
U (x, 0) = u (x ), (1)
U (x, y ) → 0 as y → ∞.



Extension Problem
Theorem 1 (Extension Theorem)

Let u ∈ S(Rn ). Then, the solution U to the extension problem (1) is given by
ˆ
U (x, y ) = (Ps (·, y ) ? u )(x ) = Ps (x − z, y )u (z ) dz, (2)
Rn
where
Γ(n/2 + s ) y 2s
Ps (x, y ) = (3)
π n/2 Γ(s ) (y 2 + |x |2 )(n+2s )/2
is the Poisson Kernel for the extension problem in the half-space Rn++1 . For U as
in (2) one has
22s −1 Γ(s )
(−∆)s u (x ) = − lı́m y 1−2s ∂y U (x, y ). (4)
Γ ( 1 − s ) y →0+

Extension Problem
Theorem 1 (Extension Theorem)

Let u ∈ S(Rn ). Then, the solution U to the extension problem (1) is given by
ˆ
U (x, y ) = (Ps (·, y ) ? u )(x ) = Ps (x − z, y )u (z ) dz, (2)
Rn
where
Γ(n/2 + s ) y 2s
Ps (x, y ) = (3)
π n/2 Γ(s ) (y 2 + |x |2 )(n+2s )/2
is the Poisson Kernel for the extension problem in the half-space Rn++1 . For U as
in (2) one has
22s −1 Γ(s )
(−∆)s u (x ) = − lı́m y 1−2s ∂y U (x, y ). (4)
Γ ( 1 − s ) y →0+

Extension Problem
PROOF
If we take a partial Fourier transform of (1)
(
∂yy Û (ξ, y ) + 1−y2s ∂y Û (ξ, y ) − 4π 2 |ξ |2 Û (ξ, y ) = 0 in Rn++1 ,
Û (ξ, 0) = û (ξ ), Û (ξ, y ) → 0 as y → ∞, x ∈ Rn .
If we fix ξ ∈ Rn \ {0} and Y (y ) = Yξ (y ) = Û (ξ, y ),

00 0
y 2 Y (y ) + (1 − 2s )yY (y ) − 4π 2 |ξ |2 y 2 Y (y ) = 0 y in R+ ,

Y (0) = û (ξ ), y (y ) → 0 as y → ∞,
then it can be compared with the generalized modified Bessel equation:

00 0
y 2 Y + (1 − 2α)yY (y ) + [ β2 γ2 y 2γ + (α − ν2 γ2 )]Y (y ) = 0 (5)
α = s, γ = 1, ν = s, β = 2π |ξ |.

Extension Problem
PROOF
(
Û (ξ, 0) = û (ξ ), Û (ξ, y ) → 0 as y → ∞, x ∈ Rn .

00 0

Y (0) = û (ξ ), y (y ) → 0 as y → ∞,

00 0
α = s, γ = 1, ν = s, β = 2π |ξ |.

Extension Problem
PROOF
(
Û (ξ, 0) = û (ξ ), Û (ξ, y ) → 0 as y → ∞, x ∈ Rn .

00 0

Y (0) = û (ξ ), y (y ) → 0 as y → ∞,

00 0
α = s, γ = 1, ν = s, β = 2π |ξ |.

The general solutions of (5) are given by
Û (ξ, y ) = Ay s Is (2π |ξ |y ) + By s Ks (2π |ξ |y )
where Is and Ks are the Bessel functions of second and third kind,both
independent solutions of the modified Bessel equation of order s
00 0
z 2 φ + zφ − (z 2 + s 2 )φ = 0 (6)
where
φ solution of (6) =⇒ Y (y ) = y α φ( βy γ ) solution of (5).

The general solutions of (5) are given by
Û (ξ, y ) = Ay s Is (2π |ξ |y ) + By s Ks (2π |ξ |y )
where Is and Ks are the Bessel functions of second and third kind,both
independent solutions of the modified Bessel equation of order s
00 0
z 2 φ + zφ − (z 2 + s 2 )φ = 0 (6)
where
φ solution of (6) =⇒ Y (y ) = y α φ( βy γ ) solution of (5).

(z /2)s +2k
Js ( z ) = ∑k∞=0 (−1)k , |z | < ∞, |arg (z )| < π,
Γ (k + 1) Γ (k + s + 1)
(z /2)s +2k
Is (z ) = ∑k∞=0 , |z | < ∞, |arg (z )| < π,
Γ (k + 1) Γ (k + s + 1)
π I − s ( z ) − Is ( z )
Ks ( z ) = , |arg (z )| < π.
2 sin πs

Extension Problem
The condition Û (ξ, y ) → 0 as y → ∞ forces A = 0. Using Is asymptotic behavior,
π y s I−s (2π |ξ |y ) − y s Is (2π |ξ |y )

By s Ks (2π |ξ |y ) = B
2 sin πs
Bπ2 s − 1 h π i
→ (2π |ξ |)−s = Γ(s )Γ(s − 1) =
Γ(1 − s ) sin πs sin πs
s −
= B2 Γ(s )(2π |ξ |) .
1 − s
In order to fulfill the condition Û (ξ, 0) = û (ξ ), we impose
(2π |ξ |)s û (ξ ) s
Û (ξ, y ) = y Ks (2π |ξ |y ). (7)
2s −1 Γ(s )

Extension Problem

By s Ks (2π |ξ |y ) = B
2 sin πs
Bπ2 s − 1 h π i
→ (2π |ξ |)−s = Γ(s )Γ(s − 1) =
s −
= B2 Γ(s )(2π |ξ |) .
1 − s
(2π |ξ |)s û (ξ ) s
Û (ξ, y ) = y Ks (2π |ξ |y ). (7)
2s −1 Γ(s )

Extension Problem

By s Ks (2π |ξ |y ) = B
2 sin πs
Bπ2 s − 1 h π i
→ (2π |ξ |)−s = Γ(s )Γ(s − 1) =
s −
= B2 Γ(s )(2π |ξ |) .
1 − s
(2π |ξ |)s û (ξ ) s
Û (ξ, y ) = y Ks (2π |ξ |y ). (7)
2s −1 Γ(s )

We want to prove
U (x, y ) = (Ps (·, y ) ? u )(x ).
Taking inverse Fourier transform and using (7), we have to show that
(2π |ξ |)s s Γ(n/2 + s ) y 2s

−1
F ξ →x y K s ( 2π | ξ | y ) = .
2s −1 Γ(s ) π n/2 Γ(s ) (y 2 + |x |2 )(n+2s )/2
Since the function in the left hand-side of (33) is spherically symmetric, proving
(33) is equivalent to establishing the follow identity
Γ(n/2 + s ) y 2s
Fξ →x (2π s |ξ |s y s Ks (2π |ξ |y )) = n/ 2
.
π (y 2 + |x |2 )(n+2s )/2
(Hankel transform : H ≡ H −1 for radial functions.)

We want to prove
U (x, y ) = (Ps (·, y ) ? u )(x ).
(2π |ξ |)s s Γ(n/2 + s ) y 2s

−1
F ξ →x y K s ( 2π | ξ | y ) = .
2s −1 Γ(s ) π n/2 Γ(s ) (y 2 + |x |2 )(n+2s )/2
Γ(n/2 + s ) y 2s
.
π (y 2 + |x |2 )(n+2s )/2

We want to prove
U (x, y ) = (Ps (·, y ) ? u )(x ).
(2π |ξ |)s s Γ(n/2 + s ) y 2s

−1
F ξ →x y K s ( 2π | ξ | y ) = .
2s −1 Γ(s ) π n/2 Γ(s ) (y 2 + |x |2 )(n+2s )/2
Γ(n/2 + s ) y 2s
.
π (y 2 + |x |2 )(n+2s )/2

We want to prove
U (x, y ) = (Ps (·, y ) ? u )(x ).
(2π |ξ |)s s Γ(n/2 + s ) y 2s

−1
F ξ →x y K s ( 2π | ξ | y ) = .
2s −1 Γ(s ) π n/2 Γ(s ) (y 2 + |x |2 )(n+2s )/2
Γ(n/2 + s ) y 2s
.
π (y 2 + |x |2 )(n+2s )/2

Theorem 2 (Fourier-Bessel Representation)
Let u (x ) = f (|x |), and suppose that
t 7→ t n/2 f (t )Jn/2−1 (2π |ξ |t ) ∈ L1 (Rn ).
Then, ˆ ∞
−n/2+1
û (ξ ) = 2π |ξ | t n/2 f (t )Jn/2−1 (2π |ξ |t ) dt.
0
Then, the latter identity (33) is equivalent to

ˆ
22 π s +1 y s ∞ n/2+s
t Ks (2πyt )Jn/2−1 (2π |ξ |t ) dt
|x |n/2−1 0
Γ(n/2 + s ) y 2s
= n/2 .
π Γ(s ) (y 2 + |x |2 )(n+2s )/2

Theorem 2 (Fourier-Bessel Representation)
Let u (x ) = f (|x |), and suppose that
t 7→ t n/2 f (t )Jn/2−1 (2π |ξ |t ) ∈ L1 (Rn ).
Then, ˆ ∞
−n/2+1
û (ξ ) = 2π |ξ | t n/2 f (t )Jn/2−1 (2π |ξ |t ) dt.
0
Then, the latter identity (33) is equivalent to

ˆ
22 π s +1 y s ∞ n/2+s
t Ks (2πyt )Jn/2−1 (2π |ξ |t ) dt
|x |n/2−1 0
Γ(n/2 + s ) y 2s
= n/2 .
π Γ(s ) (y 2 + |x |2 )(n+2s )/2

Let’s establish
22s −1 Γ(s )
(−∆)s u (x ) = − lı́m y 1−2s ∂y U (x, y ).
Γ ( 1 − s ) y →0+
\
Recall that (− ∆)s u (ξ ) = (2π |ξ |)2s û (ξ ). Using the equalities
0 s
Ks ( z ) = Ks (z ) − Ks +1 (z )
z
and
2s
Ks (z ) − Ks +1 (z ) = −Ks −1 (z ) = −K1−s (z ),
z
we obtain
(2π |ξ |)s +1 û (ξ ) 1−s
y 1−2s ∂y Û (ξ, y ) = y K1−s (2π |ξ |y ).
22s −1 Γ(s )

Let’s establish
22s −1 Γ(s )
(−∆)s u (x ) = − lı́m y 1−2s ∂y U (x, y ).
Γ ( 1 − s ) y →0+
\
0 s
Ks ( z ) = Ks (z ) − Ks +1 (z )
z
and
2s
Ks (z ) − Ks +1 (z ) = −Ks −1 (z ) = −K1−s (z ),
z
we obtain
(2π |ξ |)s +1 û (ξ ) 1−s
y 1−2s ∂y Û (ξ, y ) = y K1−s (2π |ξ |y ).
22s −1 Γ(s )

Let’s establish
22s −1 Γ(s )
(−∆)s u (x ) = − lı́m y 1−2s ∂y U (x, y ).
Γ ( 1 − s ) y →0+
\
0 s
Ks ( z ) = Ks (z ) − Ks +1 (z )
z
and
2s
Ks (z ) − Ks +1 (z ) = −Ks −1 (z ) = −K1−s (z ),
z
we obtain
(2π |ξ |)s +1 û (ξ ) 1−s
y 1−2s ∂y Û (ξ, y ) = y K1−s (2π |ξ |y ).
22s −1 Γ(s )

Let’s establish
22s −1 Γ(s )
(−∆)s u (x ) = − lı́m y 1−2s ∂y U (x, y ).
Γ ( 1 − s ) y →0+
\
0 s
Ks ( z ) = Ks (z ) − Ks +1 (z )
z
and
2s
Ks (z ) − Ks +1 (z ) = −Ks −1 (z ) = −K1−s (z ),
z
we obtain
(2π |ξ |)s +1 û (ξ ) 1−s
y 1−2s ∂y Û (ξ, y ) = y K1−s (2π |ξ |y ).
22s −1 Γ(s )

As before, we have
lı́m y 1−s K1−s (2π |ξ |y ) = 2−s Γ(1 − s )(2π |ξ |)s −1 .

y →0+
We finally reach the conclusion
Γ (1 − s )
lı́m y 1−2s ∂y Û (ξ, y ) = − 2s −1 (2π |ξ |)2s û (ξ ).
y →0+ 2 Γ (s )
Remark 1 (Alternative proof of (4))

Using that ˆ
Ps (x, y ) dx = 1, y > 0,
Rn
we will show another proof.

As before, we have

y →0+
Γ (1 − s )
y →0+ 2 Γ (s )

Using that ˆ
Ps (x, y ) dx = 1, y > 0,
Rn

As before, we have

y →0+
Γ (1 − s )
y →0+ 2 Γ (s )

Using that ˆ
Ps (x, y ) dx = 1, y > 0,
Rn

Let u ∈ S(Rn ) and consider the solution U (x, y ) = (Ps (·, y ) ? u )(x ) to the
extension problem (1). We can write
ˆ
Γ(n/2 + s ) u (z ) − u (x )
U (x, y ) = dz + u (x ).
π Γ(s ) R (y + |z − x |2 )(n+2s )/2
n/ 2 n 2
Differentiating both sides respect to y we obtain

ˆ
1−2s Γ(n/2 + s ) u (z ) − u (x )
y ∂y U (x, y ) = 2s n/2 dz + O (y 2 ).
π Γ(s ) R (y + |z − x |2 )(n+2s )/2
n 2

ˆ
Γ(n/2 + s ) u (z ) − u (x )
U (x, y ) = dz + u (x ).
π Γ(s ) R (y + |z − x |2 )(n+2s )/2
n/ 2 n 2

ˆ
1−2s Γ(n/2 + s ) u (z ) − u (x )
y ∂y U (x, y ) = 2s n/2 dz + O (y 2 ).
π Γ(s ) R (y + |z − x |2 )(n+2s )/2
n 2

ˆ
Γ(n/2 + s ) u (z ) − u (x )
U (x, y ) = dz + u (x ).
π Γ(s ) R (y + |z − x |2 )(n+2s )/2
n/ 2 n 2

ˆ
1−2s Γ(n/2 + s ) u (z ) − u (x )
y ∂y U (x, y ) = 2s n/2 dz + O (y 2 ).
π Γ(s ) R (y + |z − x |2 )(n+2s )/2
n 2

Now, letting y → 0+ and using the Lebesgue dominated convergence theorem, we
thus find
ˆ
Γ(n/2 + s ) u (z ) − u (x )
lı́m y 1−2s ∂y U (x, y ) = 2s P.V. dz
y →0+ π n/2 Γ(s ) Rn (|z − x |2 )(n+2s )/2
Γ(n/2 + s )
= −2s γ(n, s )−1 (−∆)s u (x ).
π n/2 Γ(s )
Finally, recall that

s22s Γ(n/2 + s )
γ(n, s ) = .
π n/2 Γ(1 − s )

Now, letting y → 0+ and using the Lebesgue dominated convergence theorem, we
thus find
ˆ
Γ(n/2 + s ) u (z ) − u (x )
lı́m y 1−2s ∂y U (x, y ) = 2s P.V. dz
y →0+ π n/2 Γ(s ) Rn (|z − x |2 )(n+2s )/2
Γ(n/2 + s )
= −2s γ(n, s )−1 (−∆)s u (x ).
π n/2 Γ(s )
Finally, recall that

s22s Γ(n/2 + s )
γ(n, s ) = .
π n/2 Γ(1 − s )

La exposición está basada en:
1. L. Caffarelli y L. Silvestre, An extension problem related to the fractional

Laplacean, Comm. Partial Differential Equations 32 (2007), no. 7-9,
1245-1260;
2. N. Garofalo, Fractional thoughts, arXiv:1712.03347v3;
3. M. Kwaśnicki, Ten equivalent definitions of the fractional Laplace operator,
Fract. Calc. Appl. Anal. 20 (2017), no. 1, 7-51;
4. L. Silvestre, Regularity of the obstacle problem for a fractional power of the
Laplace operator, Comm. Pure Appl. Math. 60 (2007), no. 1, 67-112 y Tesis
Doctoral, 2005 (con el mismo nombre).
5. P. R. Stinga, Fractional powers of second order partial differential operators:
extension problem and regularity theory, Tesis Doctoral, 2010;
6. P. R. Stinga y J. L. Torrea, Extension problem and Harnack’s inequality for
some fractional operators, Comm. Partial Differential Equations 35 (2010),
no. 11, 2092-2122.

Thanks for your attention
Eskerrik asko zuen arretarengatik

Simba23032018 SergiBaena

Uploaded by

Copyright:

Available Formats

You might also like

Simba23032018 SergiBaena

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Simba23032018 SergiBaena

Uploaded by

Copyright:

Available Formats

Four definitions for the fractional Laplacian

N. Accomazzo (UPV/EHU), S. Baena (UB), A. Becerra Tomé (US), J.

VIII Escuela-Taller de Análisis Funcional

1 International Women’s Day

Group 3 Fractional Laplacian VIII Escuela-Taller 2 / 40

Functional analysis: M. Riesz, S. Bochner, W. Feller, E. Hille, R. S. Phillips,

potencial theory for fractional laplacian: N. S. Landkof

Lévy’s processes: K. Bogdan e.a.

Partial Derivative Ecuations: L. Caffarelli y L. Silvestre

Scattering theory: C. R. Graham y M. Zworski, S-Y. A. Chang y M.d.M.

Kato’s square root (solved by P. Auscher e.a.)

Group 3 Fractional Laplacian VIII Escuela-Taller 3 / 40

Functional analysis: M. Riesz, S. Bochner, W. Feller, E. Hille, R. S. Phillips,

potencial theory for fractional laplacian: N. S. Landkof

Lévy’s processes: K. Bogdan e.a.

Partial Derivative Ecuations: L. Caffarelli y L. Silvestre

Scattering theory: C. R. Graham y M. Zworski, S-Y. A. Chang y M.d.M.

Kato’s square root (solved by P. Auscher e.a.)

Basic example of fractional operator : fractional laplacean

Group 3 Fractional Laplacian VIII Escuela-Taller 3 / 40

Group 3 Fractional Laplacian VIII Escuela-Taller 4 / 40

S (Rn ) is the space C ∞ (Rn ) of functions that

kf kp = sup sup (1 + |x |2 )p/2 |∂α f (x )| < ∞ p ∈ N ∪ {0}

Group 3 Fractional Laplacian VIII Escuela-Taller 4 / 40

S (Rn ) is the space C ∞ (Rn ) of functions that

kf kp = sup sup (1 + |x |2 )p/2 |∂α f (x )| < ∞ p ∈ N ∪ {0}

This space endowed with the metric topology

Group 3 Fractional Laplacian VIII Escuela-Taller 4 / 40

Let f ∈ C 2 (a, b ), then for every x ∈ (a, b ) one has

Group 3 Fractional Laplacian VIII Escuela-Taller 5 / 40

Let f ∈ C 2 (a, b ), then for every x ∈ (a, b ) one has

If we introduce the spherical and solid averaging operators

Group 3 Fractional Laplacian VIII Escuela-Taller 5 / 40

Let f ∈ C 2 (a, b ), then for every x ∈ (a, b ) one has

If we introduce the spherical and solid averaging operators

then we can reformulate −f 00 (x ) like this

Group 3 Fractional Laplacian VIII Escuela-Taller 5 / 40

Group 3 Fractional Laplacian VIII Escuela-Taller 6 / 40

Group 3 Fractional Laplacian VIII Escuela-Taller 6 / 40

Group 3 Fractional Laplacian VIII Escuela-Taller 7 / 40

Group 3 Fractional Laplacian VIII Escuela-Taller 7 / 40

Observation 1: γ(n, s ) is, at this moment, an unknown constant

Group 3 Fractional Laplacian VIII Escuela-Taller 7 / 40

Observation 1: γ(n, s ) is, at this moment, an unknown constant

Group 3 Fractional Laplacian VIII Escuela-Taller 7 / 40

Observation 1: γ(n, s ) is, at this moment, an unknown constant

Group 3 Fractional Laplacian VIII Escuela-Taller 7 / 40

It is important to observe that the integral in the right-hand side is convergent. In

Group 3 Fractional Laplacian VIII Escuela-Taller 8 / 40

It is important to observe that the integral in the right-hand side is convergent. In

Group 3 Fractional Laplacian VIII Escuela-Taller 8 / 40

It is important to observe that the integral in the right-hand side is convergent. In

Group 3 Fractional Laplacian VIII Escuela-Taller 8 / 40

Group 3 Fractional Laplacian VIII Escuela-Taller 9 / 40

Group 3 Fractional Laplacian VIII Escuela-Taller 9 / 40

Group 3 Fractional Laplacian VIII Escuela-Taller 9 / 40

Group 3 Fractional Laplacian VIII Escuela-Taller 9 / 40