Parabolic Hausdorff dimension explained

In fractal geometry, the parabolic Hausdorff dimension is a restricted version of the genuine Hausdorff dimension. Only parabolic cylinders, i. e. rectangles with a distinct non-linear scaling between time and space are permitted as covering sets. It is usefull to determine the Hausdorff dimension of self-similar stochastic processes, such as the geometric Brownian motion or stable Lévy processes plus Borel measurable drift function

f

.

Definitions

We define the

\alpha

-parabolic

\beta

-Hausdorff outer measure for any set

A\subseteq\Rd+1

as

l{P}\alpha-l{H}\beta(A):=\lim\deltainf\left\{

infty
\sum
k=1

\left|Pk\right|\beta:A\subseteq

infty
cup
k=1

Pk,Pk\inl{P}\alpha,\left|Pk\right|\leq\delta\right\}.

where the

\alpha

-parabolic cylinders

\left(Pk\right)k

are contained in

l{P}\alpha:=\left\{[t,t+c] x

d
\prod
i=1

\left[xi,xi+c1/\alpha\right];t,xi\inR,c\in(0,1]\right\}.

We define the

\alpha

-parabolic Hausdorff dimension of

A

as

l{P}\alpha-\dimA:=inf\left\{\beta\geq0:l{P}\alpha-l{H}\beta(A)=0\right\}.

The case

\alpha=1

equals the genuine Hausdorff dimension

\dim

.

Application

Let

\varphi\alpha:=l{P}\alpha-\diml{G}T(f)

. We can calculate the Hausdorff dimension of the fractional Brownian motion

BH

of Hurst index

1/\alpha=H\in(0,1]

plus some measurable drift function

f

. We get

\diml{G}T\left(BH+f\right)=\varphi\alpha\wedge

1
\alpha

\varphi\alpha+\left(1-

1
\alpha

\right)d

and

\diml{R}T\left(BH+f\right)=\varphi\alpha\wedged.

For an isotropic

\alpha

-stable Lévy process

X

for

\alpha\in(0,2]

plus some measurable drift function

f

we get

\diml{G}T(X+f)= \begin{cases} \varphi1,&\alpha\in(0,1],\\ \varphi\alpha\wedge

1
\alpha

\varphi\alpha+\left(1-

1
\alpha

\right)d,&\alpha\in[1,2] \end{cases}

and

\diml{R}T\left(X+f\right)= \begin{cases} \alpha\varphi\alpha\wedged,&\alpha\in(0,1],\\ \varphi\alpha\wedged,&\alpha\in[1,2]. \end{cases}

Inequalities and identities

For

\phi\alpha:=l{P}\alpha-\dimA

one has

\dimA\leq \begin{cases} \phi\alpha\wedge\alpha\phi\alpha+1-\alpha,&\alpha\in(0,1],\\ \phi\alpha\wedge

1
\alpha

\alpha+\left(1-

1
\alpha

\right)d,&\alpha\in[1,infty) \end{cases}

and

\dimA\geq \begin{cases} \alpha\phi\alpha\vee\phi\alpha+\left(1-

1
\alpha

\right)d,&\alpha\in(0,1],\\ \phi\alpha+1-\alpha,&\alpha\in[1,infty). \end{cases}

Further, for the fractional Brownian motion

BH

of Hurst index

1/\alpha=H\in(0,1]

one has

l{P}\alpha-\diml{G}T\left(BH\right)=\alpha\dimT

and for an isotropic

\alpha

-stable Lévy process

X

for

\alpha\in(0,2]

one has

l{P}\alpha-\diml{G}T\left(X\right)=(\alpha\vee1)\dimT

and

\diml{R}T(X)=\alpha\dimT\wedged.

For constant functions

fC

we get

l{P}\alpha-\diml{G}T\left(fC\right)=(\alpha\vee1)\dimT.

If

f\inC\beta(T,Rd)

, i. e.

f

is

\beta

-Hölder continuous, for

\varphi\alpha=l{P}\alpha-\diml{G}T(f)

the estimates

\varphi\alpha\leq \begin{cases} \dimT+\left(

1
\alpha

-\beta\right)d\wedge

\dimT
\alpha\beta

\wedged+1,&\alpha\in(0,1],\\ \alpha\dimT+(1-\alpha\beta)d\wedge

\dimT
\beta

\wedged+1,&\alpha\in\left[1,

1
\beta

\right],\\ \alpha\dimT+

1
\beta

(\dimT-1)+\alpha\wedged+1,&\alpha\in\left[

1
\beta

,infty)\right] \end{cases}

hold.

Finally, for the Brownian motion

B

and

f\inC\beta\left(T,Rd\right)

we get

\diml{G}T(B+f)\leq \begin{cases} d+

1
2

,&\beta\leq

\dimT
d

-

1
2d

,\\ \dimT+(1-\beta)d,&

\dimT
d

-

1
2d

\leq\beta\leq

\dimT
d

\wedge

1,\\
2
\dimT
\beta

,&

\dimT
d

\leq\beta\leq

1
2

,\\ 2\dimT\wedge\dimT+

d
2

,&else \end{cases}

and

\diml{R}T(B+f)\leq \begin{cases}

\dimT
\beta

,&

\dimT
d

\leq\beta\leq

1
2

,\\ 2\dimT\wedged,&

\dimT
d

\leq

1
2

\leq\beta,\\ d,&else. \end{cases}

Sources