![Page 1: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/1.jpg)
SUBORDINATED PROCESSES AND CAUCHYPROBLEMS
byERKAN NANE
DEPARTMENT OF STATISTICS AND PROBABILITYMICHIGAN STATE UNIVERSITY
![Page 2: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/2.jpg)
OUTLINE
• Introduction and history
• Brownian subordinators
• Cauchy problems on bounded domains
• Other subordinators
• Scaling Limits
• Conclusion and open problems
2
![Page 3: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/3.jpg)
INTRODUCTION AND HISTORY
In recent years, starting with the articles of Burdzy (1993)and (1994), researchers had interest in iterated processes inwhich one changes the time parameter with one-dimensionalBrownian motion.
To define iterated Brownian motion Zt, due to Burdzy (1993),started at z ∈ R, let X+
t , X−t and Yt be three independent
one-dimensional Brownian motions, all started at 0. Two-sidedBrownian motion is defined to be
Xt =
{
X+t , t ≥ 0
X−(−t), t < 0.
Then iterated Brownian motion started at z ∈ R is
Zt = z +X(Yt), t ≥ 0.
3
![Page 4: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/4.jpg)
BM versus IBM: This process has many properties analogousto those of Brownian motion; we list a few
(1) Zt has stationary (but not independent) increments, and isa self-similar process of index 1/4.
(2) Laws of the iterated logarithm (LIL) holds: usual LIL byBurdzy (1993)
lim supt→∞
Z(t)
t1/4(log log(1/t))3/4=
25/4
33/4a.s.
Chung-type LIL by Khoshnevisan and Lewis (1996) and Hu etal. (1995).
(3) Khoshnevisan and Lewis (1999) extended results of Burdzy(1994), to develop a stochastic calculus for iterated Brownianmotion.
4
![Page 5: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/5.jpg)
(4) In 1998, Burdzy and Khoshnevisan showed that IBM can beused to model diffusion in a crack.
(5) Local time of this process was studied by Burdzy andKhoshnevisan (1995), Csaki, Csorgo, Foldes, and Revesz(1996), Shi and Yor (1997), Xiao (1998), and Hu (1999).
(6) Banuelos and DeBlassie (2006) studied the distribution ofexit place for iterated Brownian motion in cones.
(7) DeBlassie (2004) studied the lifetime asymptotics of iteratedBrownian motion in cones and Bounded domains. Nane(2006), in a series of papers, extended some of the resultsof DeBlassie. He also studied the lifetime asymtotics of iteratedBrownian motion in several unbounded domains(parabola-shaped domains, twisted domains...).
5
![Page 6: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/6.jpg)
(8) Khoshnevisan and Lewis (1996) established the modulus ofcontinuity for iterated Brownian motion: with probability one
limδ→0
sup0≤s,t≤1
sup0≤|s−t|≤δ
|Z(s) − Z(t)|δ1/4(log(1/δ))3/4
= 1.
6
![Page 7: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/7.jpg)
–0.1
–0.05
0
0.05
0.1
0.15
0.2
0.05 0.1 0.15 0.2 0.25 0.3
t
Figure 1: Simulations of two Brownian motions7
![Page 8: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/8.jpg)
–0.05
0
0.05
0.1
0.05 0.1 0.15 0.2 0.25 0.3
t
Figure 2: Simulation of IBM Z1t = X(|Yt|)
8
![Page 9: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/9.jpg)
PDE connection
The classical well-known connection of a PDE and a stochasticprocess is the Brownian motion and heat equation connection.LetXt ∈ R
d be Brownian motion started at x. Then the function
u(t, x) = Ex[f (Xt)]
solves the Cauchy problem
∂
∂tu(t, x) = ∆u(t, x), t > 0, x ∈ R
d
u(0, x) = f (x), x ∈ Rd.
In addition to the above properties of IBM there is aninteresting connection between iterated Brownian motion andthe biharmonic operator ∆2; the function
9
![Page 10: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/10.jpg)
u(t, x) = Ex[f (Zt)]
solves the Cauchy problem (Allouba and Zheng (2001) andDeBlassie (2004))
∂
∂tu(t, x) =
∆f (x)√πt
+ ∆2u(t, x); (1)
u(0, x) = f (x).
for t > 0 and x ∈ Rd. The non-Markovian property of IBM is
reflected by the appearance of the initial function f (x) in thePDE.
Let q(t, s) = 2√4πt
exp(
−s2
4t
)
be the transition density of
reflected one-dimensional Brownian motion, |Bt|.The essential argument, using integration by parts twice, is that
10
![Page 11: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/11.jpg)
∂∂tu(t, x)
=
∫ ∞
0
T (s)f (x)∂
∂tq(t, s)ds
=
∫ ∞
0
T (s)f (x)∂2
∂s2q(t, s)ds
= q(t, s)∂
∂s[T (s)f (x)]
∣
∣
∣
∣
s=0
+
∫ ∞
0
∂2
∂s2[T (s)f (x)] q(t, s)ds
= q(t, 0)Lx[T (0)f (x)] +
∫ ∞
0
L2x [T (s)f (x)] q(t, s)ds
=1√πtLxf (x) + L2
x
∫ ∞
0
T (s)f (x) q(t, s)ds
11
![Page 12: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/12.jpg)
Fractional Diffusion
Nigmatullin (1986) gave a Physical derivation of fractionaldiffusion
∂β
∂tβu(t, x) = Lxu(t, x); u(0, x) = f (x) (2)
where 0 < β < 1 and Lx is the generator of some continuousMarkov process X0(t) started at x = 0. Here ∂βg(t)/∂tβ isthe Caputo fractional derivative in time, which can be definedas the inverse Laplace transform of sβg(s) − sβ−1g(0), withg(s) =
∫ ∞0 e−stg(t)dt the usual Laplace transform.
Zaslavsky (1994) used this to model Hamiltonian chaos.
12
![Page 13: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/13.jpg)
Stochastic solution
Baeumer and Meerschaert (2001) and Meerschaert andScheffler (2004) shows that, in the case p(t, x) = T (t)f (x)is a bounded continuous semigroup on a Banach space (withcorresponding process Xt, Et = inf{u : Du > t}, Dt is astable subordinator with index β) , the formula
u(t, x) = Ex(f (XEt)) =
t
β
∫ ∞
0
p(s, x)gβ(t
s1/β)s−1/β−1ds
yields a solution to the fractional Cauchy problem :
∂β
∂tβu(t, x) = Lxu(t, x); u(0, x) = f (x) (3)
Here gβ(t) is the smooth density of the stable subordinator, such
that the Laplace transform gβ(s) =∫ ∞
0 e−stgβ(t) dt = e−sβ
.
13
![Page 14: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/14.jpg)
BROWNIAN SUBORDINATORS
(Baeumer, Meerschaert and Nane (2007)) show that, takingβ = 1/2 in the time-fractional diffusion yields exactly 1-D
distributions X|Bt|d= XEt
.
IfLx is the generator of the semigroupT (t)f (x) = Ex[(f (Xt))]on L1(Rd), then for any f ∈ D(Lx)
∂
∂tu(t, x) =
Lxf (x)√πt
+ Lx2u(t, x); u(0, x) = f (x), (4)
and the fractional Cauchy problem (3) with β = 1/2 have thesame solution
u(t, x) = Ex[f (X(|Bt|))] =2√4πt
∫ ∞
0
T (s)f (x) exp
(
−s2
4t
)
ds.
14
![Page 15: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/15.jpg)
Fourier-Laplace method
The Levy process X0(t) has characteristic function
E[exp(ik ·X0(t))] = exp(tψ(k))
with
ψ(k) = ik·a−1
2k·Qk+
∫
y 6=0
(
eik·y − 1 − ik · y1 + ||y||2
)
ν(dy),
where a ∈ Rd, Q is a nonnegative definite matrix, and ν is a
σ-finite Borel measure on Rd such that
∫
y 6=0
min{1, ||y||2}ν(dy) <∞.
Denote the Fourier transform by
f (k) =
∫
Rd
e−ik·xf (x) dx
15
![Page 16: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/16.jpg)
Meerschaert and Scheffler (2001) shows that Lxf (x) is the
inverse Fourier transform ofψ(k)f (k) for all f ∈ D(Lx), where
D(Lx) = {f ∈ L1(Rd) : ψ(k)f (k) = h(k) ∃ h ∈ L1(Rd)},and
Lxf (x) = a · ∇f (x) +1
2∇ ·Q∇f (x)
+
∫
y 6=0
(
f (x + y) − f (x) − ∇f (x) · y1 + y2
)
ν(dy)
(5)
for all f ∈W 2,1(Rd), the Sobolev space of L1-functions whosefirst and second partial derivatives are all L1-functions.
We can also write Lx = ψ(−i∇) where ∇ =(∂/∂x1, . . . , ∂/∂xd)
′.16
![Page 17: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/17.jpg)
For example, if X0(t) is spherically symmetric stable thenψ(k) = −D‖k‖α and Lx = −D(−∆)α/2, a fractionalderivative in space, using the correspondence kj → −i∂/∂xjfor 1 ≤ j ≤ d.
If X0 has independent stable marginals, then one possibleform is ψ(k) = D
∑
j(ikj)αj and Lx = D
∑
j ∂αj/∂xαj using
Riemann-Liouville fractional derivatives in each variable. Thisform does not coincide with the fractional Laplacian unless allαj = 2.
17
![Page 18: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/18.jpg)
The proof does not use Theorem 0.1 in Allouba and Zheng(2001), rather it relies on a Laplace-Fourier transform argument.
We will use the following notation for the Laplace, Fourier, andFourier-Laplace transforms (respectively):
u(s, x) =
∫ ∞
0
e−stu(t, x)dt;
u(t, k) =
∫
Rd
e−ik·xu(t, x)dx;
u(s, k) =
∫
Rd
e−ik·x∫ ∞
0
e−stu(t, x)dtdx.
18
![Page 19: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/19.jpg)
Let ψ be the characteristic exponent of Xt. Take Fouriertransforms on both sides of (4) to get
∂u(t, k)
∂t=
1√πtψ(k)f (k) + ψ(k)2u(t, k)
using the fact that ψ(k)f (k) is the Fourier transform of Lxf (x).Then take Laplace transforms on both sides to get
su(s, k) − u(t = 0, k) = s−1/2ψ(k)f (k) + ψ(k)2u(s, k),
using the well-known Laplace transform formula∫ ∞
0
t−β
Γ(1 − β)e−stdt = sβ−1, β < 1.
19
![Page 20: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/20.jpg)
Since u(t = 0, k) = f (k), collecting like terms yields
u(s, k) =(1 + s−1/2ψ(k))f (k)
s− ψ(k)2(6)
for s > 0 sufficiently large.
On the other hand, taking Fourier transforms on both sides of(3) with β = 1/2 gives
∂1/2u(t, k)
∂t1/2= ψ(k)u(t, k)
Take Laplace transforms on both sides, using the fact thatsβg(s) − sβ−1g(0) is the Laplace transform of the Caputofractional derivative ∂βg(t)/∂tβ, to get
s1/2u(s, k) − s−1/2f(k) = ψ(k)u(s, k)
20
![Page 21: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/21.jpg)
and collect terms to obtain
u(s, k) =s−1/2f(k)
s1/2 − ψ(k)
=s−1/2f(k)
s1/2 − ψ(k)· s
1/2 + ψ(k)
s1/2 + ψ(k)
=(1 + s−1/2ψ(k))f (k)
s− ψ(k)2
(7)
which agrees with (6). For any fixed k ∈ Rd, the two formulae
are well-defined and equal for all s > 0 sufficiently large.
An easy extension of the argument as above shows that, underthe same conditions, for any k = 2, 3, 4, . . . both the Cauchy
21
![Page 22: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/22.jpg)
problem
∂u(t, x)
∂t=
k−1∑
j=1
t1−j/k
Γ(j/k)Ljxf (x) + Lkxu(t, x);
u(0, x) = f (x)
(8)
and the fractional Cauchy problem:
∂1/k
∂t1/ku(t, x) = Lxu(t, x); u(0, x) = f (x), (9)
have the same unique solution given by
u(t, x) =
∫ ∞
0
p((t/s)β, x)gβ(s) ds
with β = 1/k. Hence the process Zt = X(Et) is also thestochastic solution to this higher order Cauchy problem.
22
![Page 23: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/23.jpg)
Orsingher and Benghin (2004) and (2008) show that for β =1/2n the solution to
∂1/2n
∂t1/2nu(t, x) = ∆xu(t, x); u(0, x) = f (x), (10)
is given by running
In(t) = B1(|B2(|B3(| · · · (Bn+1(t)) · · · |)|)|)Where Bj ’s are independent Brownian motions, i.e., u(t, x) =Ex(f (In(t))) solves (10), and solves (8) for k = 2n.
23
![Page 24: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/24.jpg)
CAUCHY PROBLEMS ON BOUNDED DOMAINS
Since we are working on a bounded domain, the Fouriertransform methods are not useful. Instead we will employ Hilbertspace methods. Hence, given a complete orthonormal basis{ψn(x)} on L2(D), we will call
u(t, n) =
∫
D
ψn(x)u(t, x)dx;
u(s, n) =
∫
D
ψn(x)
∫ ∞
0
e−stu(t, x)dtdx =
∫
D
ψn(x)u(s, x)dx.
the ψn, and ψn-Laplace transforms, respectively.
Let D be bounded and every point of ∂D be regular for DC.The corresponding Markov process is a killed Brownian motion.We denote the eigenvalues and the eigenfunctions of ∆D by
24
![Page 25: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/25.jpg)
{λn, φn}∞n=1, where φn ∈ C∞(D). The corresponding heatkernel is given by
pD(t, x, y) =
∞∑
n=1
e−λntφn(x)φn(y).
The series converges absolutely and uniformly on [t0,∞) ×D ×D for all t0 > 0. In this case, the semigroup given by
TD(t)f (x) = Ex[f (Xt)I(t < τD(X))] =
∫
D
pD(t, x, y)f (y)dy
=
∞∑
n=1
e−λntφn(x)f (n)
solves the Heat equation in D with Dirichlet boundary
25
![Page 26: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/26.jpg)
conditions:
∂u(t, x)
∂t= ∆u(t, x), x ∈ D, t > 0,
u(t, x) = 0, x ∈ ∂D,u(0, x) = f (x), x ∈ D.
26
![Page 27: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/27.jpg)
Fractional Cauchy problems in bounded domains
Let β ∈ (0, 1), D∞ = (0,∞) ×D and define
H∆(D∞) ≡{
u : D∞ → R :∂
∂tu,∂β
∂tβu,∆u ∈ C(D∞),
∣
∣
∣
∣
∂
∂tu(t, x)
∣
∣
∣
∣
≤ g(x)tβ−1, g ∈ L∞(D), t > 0
}
.
Let 0 < γ < 1. Let D be a bounded domain with ∂D ∈ C1,γ,and TD(t) be the killed semigroup of Brownian motion {Xt}in D. Let Et be the process inverse to a stable subordinatorof index β ∈ (0, 1) independent of {Xt}. Let f ∈ D(∆D) ∩C1(D)∩C2(D) for which the eigenfunction expansion (of ∆f )with respect to the complete orthonormal basis {φn : n ∈ N}
27
![Page 28: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/28.jpg)
converges uniformly and absolutely. Then the unique (classical)solution of
u ∈ H∆(D∞) ∩ Cb(D∞) ∩ C1(D)
∂β
∂tβu(t, x) = ∆u(t, x); x ∈ D, t > 0 (11)
u(t, x) = 0, x ∈ ∂D, t > 0,u(0, x) = f (x), x ∈ D.
28
![Page 29: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/29.jpg)
is given by
u(t, x) =
∞∑
n=1
f (n)φn(x)Eβ(−λntβ)
= Ex[f (X(Et))I(τD(X) > Et)]= Ex[f (X(Et))I(τD(X(E)) > t)]
=t
β
∫ ∞
0
TD(l)f (x)gβ(tl−1/β)l−1/β−1dl
=
∫ ∞
0
TD((t/l)β)f (x)gβ(l)dl.
Joint work with Meerschaert and Vellaisamy (2008).
Analytic solution in intervals (0,M ) ⊂ R was obtained byAgrawal (2002).
29
![Page 30: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/30.jpg)
Assume that u(t, x) solves (11). Taking φn- transforms in (11)we obtain
∂β
∂tβu(t, n) = −λnu(t, n). (12)
taking Laplace transforms on both sides of (12), we get
sβu(s, n) − sβ−1u(0, n) = −λnu(s, n) (13)
which leads to
u(s, n) =f (n)sβ−1
sβ + λn. (14)
By inverting the above Laplace transform, we obtain
u(t, n) = f (n)Eβ(−λntβ)
30
![Page 31: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/31.jpg)
in terms of the Mittag-Leffler function defined by
Eβ(z) =
∞∑
k=0
zk
Γ(1 + βk).
Inverting now the φn-transform, we get an L2-convergentsolution of Equation (11) as (for each t ≥ 0 )
u(t, x) =
∞∑
n=1
f (n)φn(x)Eβ(−λntβ) (15)
31
![Page 32: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/32.jpg)
IBM in bounded domains
Let Zt = X(|Yt|) be the iterated Brownian motion, D∞ =(0,∞) ×D and define
H∆2(D∞) ≡{
u : D∞ → R :∂
∂tu,∆2u ∈ C(D∞),∆u ∈ C1(D) ,
∣
∣
∣
∣
∂
∂tu(t, x)
∣
∣
∣
∣
≤ g(x)t−1/2, g ∈ L∞(D), t > 0
}
.
Let D be a domain with ∂D ∈ C1,γ, 0 < γ < 1. Let{Xt} be Brownian motion in R
d, and {Yt} be an independentBrownian motion in R. Let {Et} be the process inverseto a stable subordinator of index β = 1/2 independent of{Xt}. Let f ∈ D(∆D) ∩ C1(D) ∩ C2(D)(⊂ L2(D)) be
32
![Page 33: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/33.jpg)
such that the eigenfunction expansion of ∆f with respect to{φn : n ≥ 1} converges absolutely and uniformly. Then the(classical) solution of
u ∈ H∆2(D∞) ∩ Cb(D∞) ∩ C1(D);∂
∂tu(t, x) =
∆f (x)√πt
+ ∆2u(t, x), x ∈ D, t > 0; (16)
u(t, x) = ∆u(t, x) = 0, t ≥ 0, x ∈ ∂D;
u(0, x) = f (x), x ∈ D
33
![Page 34: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/34.jpg)
is given by
u(t, x) = Ex[f (Zt)I(τD(X) > |Yt|)]= Ex[f (X(Et))I(τD(X) > Et)]= Ex[f (X(Et))I(τD(X(E)) > t)]
= 2
∫ ∞
0
TD(l)f (x)p(t, l)dl, (17)
where TD(l) is the heat semigroup in D, and p(t, l) is thetransition density of one-dimensional Brownian motion {Yt}.
Proof uses again equivalence with fractional Cauchy problemfor β = 1/2.
34
![Page 35: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/35.jpg)
OTHER SUBORDINATORS
A Levy process S = {St, t ≥ 0} with values in R is calledstrictly stable of index α ∈ (0, 2] if its characteristic function isgiven by
E[
exp(iξSt)]
= exp
(
−t|ξ|α1 + iνsgn(ξ) tan(πα2 )
χ
)
, (18)
where −1 ≤ ν ≤ 1 and χ > 0 are constants. When α = 2and χ = 2, S is Brownian motion.
35
![Page 36: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/36.jpg)
For any Borel set I ⊆ R, the occupation measure of S on I isdefined by
µI(A) = λ1{t ∈ I : St ∈ A} (19)
for all Borel sets A ⊆ R, where λ1 is the one-dimensionalLebesgue measure. If µI is absolutely continuous with respectto the Lebesgue measureλ1 on R, we say thatS has a local timeon I and define its local time L(x, I) to be the Radon-Nikodymderivative of µI with respect to λ1, i.e.,
L(x, I) =dµIdλ1
(x), ∀x ∈ R.
In the above, x is the so-called space variable, and I is the timevariable of the local time. If I = [0, t], we will write L(x, I) asL(x, t). Moreover, if x = 0 then we will simply write L(0, t) asLt.
36
![Page 37: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/37.jpg)
Local time as a subordinator
It is well-known (see, e.g. Bertoin (1996)) that a strictly stableLevy process S has a local time if and only if α ∈ (1, 2].
It is well-known (see, e.g. Bertoin (1996)) that the inverse of alocal time Lt of S is a stable subordinator:
Gt = inf{u : Lu > t}, then Gt = ρDt, where Dt is a stablesubordinator of index β = 1 − 1/α.
ρ = π−1Γ(1+1/α)Γ(1−1/α)χ1/αRe{[1+iν tan(πα/2)]−1/α}.Hence Lt = Et/ρ where β = 1 − 1/α for some c > 0.
For β = 1 − 1/α, c > 0, u(t, x) = Ex[f (X(Lt))] solves
∂β
∂tβu(t, x) = cLxu(t, x); u(0, x) = f (x) (20)
37
![Page 38: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/38.jpg)
Symmetric stable subordinators
α-time process is a Markov process subordinated to theabsolute value of an independent one-dimensional symmetricα-stable process:
Zt = X(|St|), where Xt is a Markov process and St is anindependent symmetric α-stable process both started at 0.
This process is self similar with index 1/2α when the outerprocess X is a Brownian motion. In this case Nane (2006)defined the Local time of this process and obtained Laws of theiterated logarithm for the local time for large time.
38
![Page 39: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/39.jpg)
PDE-connection:
Theorem 3 [Nane (2008)]
Let T (s)f (x) = Ex[f (X(s))] be the semigroup of thecontinuous Markov process X(t) and let Lx be its generator.Let α = 1. Let f be a bounded measurable function in thedomain of Lx, with Dijf bounded and Holder continuous for all1 ≤ i, j ≤ n. Then u(t, x) = Ex[f (Zt)] solves
∂2
∂t2u(t, x) = −2Lxf (x)
πt− L2
xu(t, x);
u(0, x) = f (x).
39
![Page 40: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/40.jpg)
For α = l/m 6= 1 rational: the PDE is more complicated sincekernels of symmetric α-stable processes satisfy a higher orderPDE:
[
∂2l
∂s2l+ (−1)l+1 ∂
2m
∂t2m
]
pαt (0, s) = 0.
We also have to assume that we can integrate under the integralas much as we need in the case where the outer process is BM(or in general we can take the operator out of the integral). Thisis valid for α = 1/m, m = 2, 3, · · · by a Lemma in Nane(2008).
40
![Page 41: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/41.jpg)
Theorem 4 [Nane (2008)]
Letα ∈ (0, 2) be rationalα = l/m, where l andm are relativelyprime. Let T (s)f (x) = Ex[f (X(s))] be the semigroup of thecontinuous Markov process X(t) and let Lx be its generator.Let f be a bounded measurable function in the domain of Lx,with Dγf bounded and Holder continuous for all multi index γsuch that |γ| = 2l. Then u(t, x) = Ex[f (Zt)] solves
(−1)l+1 ∂2m
∂t2mu(t, x) = −2
l∑
i=1
(
∂2l−2i
∂s2l−2ipαt (0, s)|s=0
)
L2i−1x f (x)
− L2lxu(t, x);
u(0, x) = f (x).
41
![Page 42: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/42.jpg)
SCALING LIMITS
If Sn = X1 +X2 + · · · +Xn is particle location at time n thenthe scaling limit is r−1/αS[rt] =⇒ Xt.
The limit process Xt is called an α-stable Levy motion.
Another random walk Jn = T1 +T2 + · · ·+Tn records the jumptimes.
If P [Tn > t] ≈ t−β for 0 < β < 1 then r−1/βJ[rt] =⇒ Dt.
Dt is an increasing β-stable Levy motion(a stable subordinator).
Subordinator can give random time change: X(t) → X(Dt).
42
![Page 43: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/43.jpg)
Inverse Stable subordinator
Waiting time random walk has scaling limit r−1/βJ[rt] =⇒ Dt.
The number of jumps by time t isNt = max{n > 0 : Jn ≤ t}.
Renewal process inverse to random walk {Nt ≥ n} = {Jn ≤t}Inverse process has inverse scaling limit r−βN[rt] =⇒ Et.
Inverse stable process {Et ≤ u} = {Du ≥ t} yields thedensity formula:
p(u, t) =d
duP [Et ≤ u] =
d
duP [Du ≥ t]
43
![Page 44: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/44.jpg)
CTRW scaling limits
Particle jump random walk has scaling limit r−1/αS[rt] =⇒ Xt.
Number of jumps has scaling limit r−βN[rt] =⇒ Et.
CTRW is a random walk subordinated to a renewal process
SNt= X1 +X2 + · · · +XNt
CTRW scaling limit is a subordinated process:
r−β/αSNct= (rβ)−1/αSrβ·r−βNct
≈ (rβ)−1/αSrβEt=⇒ XEt
CTRW scaling limit is not Markov, increments are not stationary.
Meerschaert and collaborators (2004).
44
![Page 45: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/45.jpg)
OPEN PROBLEMS
Question 1. For β = 1/2, the inverse stable subordinatorprocessEt and the process |Yt|, where Yt is a one-dimensionalBrownian motion have the same transition density. Is therea similar correspondence between Et for β 6= 1/2 and othersymmetric α-stable process Yt for 1 < α < 2.
Question 2. Looking at the governing PDE for subordinatorsother than Brownian motion, are there any fractional in time PDEwhich has the same solution as the higher order pde?
Question 3. Are there PDE connections of the iteratedprocesses in bounded domain as the PDE connection ofBrownian motion in bounded domains?
45
![Page 46: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/46.jpg)
THANK YOU
46
![Page 47: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/47.jpg)
REFERENCES
O. P. Agrawal, Solution for a fractional diffusion-wave equationdefined in a bounded domain. Fractional order calculus and itsapplications. Nonlinear Dynam. 29 (2002), no. 1-4, 145–155.
H. Allouba and W. Zheng,Brownian-time processes: The pdeconnection and the half-derivative generator, Ann. Prob. 29(2001), no. 2, 1780-1795.
B. Baeumer, M. M. Meerschaert and E. Nane, Browniansubordinators and fractional Cauchy problems, Trans. Amer.Math. Soc. (to appear).
47
![Page 48: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/48.jpg)
B. Baeumer and M.M. Meerschaert, Stochastic solutions forfractional Cauchy problems, Fractional Calculus Appl. Anal. 4(2001), 481-500.
C. Bandle, Isoperimetric Inequalities and Applications, Monog.Stud. Math. 7, Pitnam, Boston, 1980.
P. Becker-Kern, M. M. Meerschaert and H. P. Scheffler, Limittheorems for coupled continuous time random walks, Ann.Probab. 32 (2004), 730–756.
K. Burdzy, Some path properties of iterated Brownian motion,In: Seminar on Stochastic Processes (E. Cinlar, K.L. Chungand M.J. Sharpe, eds.), pp. 67–87, Birkhauser, Boston, 1993.
K. Burdzy, Variation of iterated Brownian motion, In: Workshopsand Conference on Measure-valued Processes, Stochastic
48
![Page 49: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/49.jpg)
Partial Differential Equations and Interacting Particle Systems(D.A. Dawson, ed.), pp. 35–53, Amer. Math. Soc. Providence,RI, 1994.
K. Burdzy and D. Khoshnevisan, The level set of iteratedBrownian motion, Seminarie de probabilites XXIX (Eds.: JAzema, M. Emery, P.-A. Meyer and M. Yor), pp. 231-236,Lecture Notes in Mathematics, 1613, Springer, Berlin, 1995.
K. Burdzy and D. Khoshnevisan, Brownian motion in a Browniancrack, Ann. Appl. Probab. 8 (1998), 708–748.
E. Csaki, A. Foldes and P. Revesz, Strassen theorems for aclass of iterated processes, Trans. Amer. Math. Soc. 349(1997), 1153–1167.
49
![Page 50: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/50.jpg)
E. Csaki, M. Csorgo, A. Foldes, and P. Revesz, The local timeof iterated Brownian motion, J. Theoret. Probab. 9 (1996),717–743.
R. D. DeBlassie, Iterated Brownian motion in an open set, Ann.Appl. Prob. 14 (2004), no. 3, 1529-1558.
Y. Hu, Hausdorff and packing measures of the level sets ofiterated Brownian motion, J. Theoret. Probab. 12 (1999), 313–346.
Y. Hu, D. Pierre-Loti-Viaud, and Z. Shi, Laws of iteratedlogarithm for iterated Wiener proceses, J. Theoret. Probab.8 (1995), 303–319.
50
![Page 51: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/51.jpg)
D. Khoshnevisan, and T.M. Lewis, The uniform modulus ofcontinuity of iterated Brownian motion. J. Theoret. Probab.9 (1996), no. 2, 317–333.
D. Khoshnevisan and T. M. Lewis, Chung’s law of the iteratedlogarithm for iterated Brownian motion, Ann. Inst. H. PoincareProbab. Statist. 32 (1996), 349–359.
D. Khoshnevisan and T. M. Lewis, Stochastic calculus forBrownian motion in a Brownian fracture, Ann. Appl. Probab. 9(1999), 629–667.
M.M. Meerschaert and H.P. Scheffler (2001) Limit Distributionsfor Sums of Independent Random Vectors: Heavy Tails inTheory and Practice. Wiley Interscience, New York.
51
![Page 52: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/52.jpg)
M. M. Meerschaert and H. P. Scheffler, Limit theorems forcontinuous time random walks with infinite mean waiting times,J. Appl. Probab. 41 (2004), 623–638.
M.M. Meerschaert, E. Nane and P. Vellaisamy (2008), FractionalCauchy problems on bounded domains. Submitted
E. Nane, Iterated Brownian motion in parabola-shaped domains,Potential Anal. 24 (2006), 105–123.
E. Nane, Iterated Brownian motion in bounded domains in Rn,
Stochastic Process. Appl. 116 (2006), 905–916.
E. Nane, Higher order PDEŠs and iterated processes, Trans.Amer. Math. Soc. 360 (2008), 2681-2692.
E. Nane, Laws of the iterated logarithm for α-time Brownianmotion, Electron. J. Probab. 11 (2006), 434–459 (electronic).
52
![Page 53: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/53.jpg)
E. Nane, Isoperimetric-type inequalities for iterated Brownianmotion in R
n, Statist. Probab. Lett. 78 (2008), 90–95.
E. Nane, Lifetime asymptotics of iterated Brownian motion inRn, Esaim:PS, 11 (2007), 147–160.
R. R. Nigmatullin, The realization of the generalized transfer ina medium with fractal geometry. Phys. Status Solidi B, 133(1986), 425-430.
E. Orsingher and L. Beghin (2004), Time-fractional telegraphequations and telegraph processes with Brownian time. Prob.Theory Rel. Fields 128, 141–160.
E. Orsingher and L. Beghin (2008), Fractional diffusionequations and processes with randomly-varying time. Annal.Probab. (to appear).
53
![Page 54: SUBORDINATED PROCESSES AND CAUCHY PROBLEMS · Two-sided Brownian motion is defined to be Xt= ... as the inverse Laplace transform of s ... (Rd), the Sobolev space of L1-functions](https://reader035.vdocuments.net/reader035/viewer/2022071402/60ef7331dd8e01459d13a587/html5/thumbnails/54.jpg)
Z. Shi and M. Yor, Integrability and lower limits of the local timeof iterated Brownian motion, Studia Sci. Math. Hungar. 33(1997), 279–298.
Y. Xiao, Local times and related properties of multi-dimensionaliterated Brownian motion, J. Theoret. Probab. 11 (1998), 383–408.
G. Zaslavsky, Fractional kinetic equation for Hamiltonian chaos.Chaotic advection, tracer dynamics and turbulent dispersion.Phys. D 76 (1994), 110-122.
54