Multifractality and multiscaling in two dimensional fragmentation
Extracted source text ยท arXiv cond-mat/9604089
\documentstyle[12pt]{article}
\voffset=-30mm
\hoffset=-10mm
\textwidth=165mm
\textheight=235mm
\begin{document}
\begin{titlepage}
\vspace{35mm}
\begin{center}
{\Large \bf Multifractality and Multiscaling in two dimensional
fragmentation}
\end{center}
\vskip 10mm
\begin{center}
{\bf M K Hassan$^{\rm a b}$ and G J Rodgers$^{\rm a}$}
\end{center}
\vskip2mm
\begin{center}
\footnotesize
\begin{tabular}{ll}
a\ &Department of Physics, Brunel University, Uxbridge, Middlesex UB8
3PH, UK \\
b\ &Department of Physics,
Shahjalal Science and Technology University, Sylhet, Bangladesh
\end{tabular}
\end{center}
\vspace{20mm}
\begin{center}
{\bf Abstract}
\end{center}
\noindent
\vspace{20mm}
\noindent
We consider two models (A and B) which can describe both two dimensional
fragmentation and stochastic fractals. Model A exhibits multifractality
on a unique support when describing a fragmentation process and
on one of infinitely many possible supports when describing
stochastic fractals. Model B obeys simple scaling.
\vspace{8mm}
\noindent
PACS numbers: 05.20-y,02.50-r
\vspace{3mm}
\noindent
Keywords: \hspace{4mm}Fragmentation, stochastic, fractal, multifractal
\vspace{15mm}
\vspace{3mm}
\noindent
\footnotesize {E-mail: Kamrul.Hassan@brunel.ac.uk and
G.J.Rodgers@brunel.ac.uk}
\end{titlepage}
\newpage
%%-Main Body of the paper-%%
\vspace{5mm}
Fragmentation is an irreversible kinetic process
in which a collection of fragments are sequentially broken. There have
been a
number of different analytical approaches to the problem of the
kinetics of
fragmentation of one dimensional particles. These have included using the
maximum entropy method [1], using statistical and combinatorial
arguments [2,3] and using a
kinetic equation. It is the kinetic equation approach, developed by
Filippov [4] after it's original proposal by Kolmogorov [5],
that has provided most of our theoretical understanding. This has
resulted in numerous exact and explicit solutions for the
particle size distribution function [6,7], in addition to scaling
solutions [8,9].
In one dimension the size or mass of the particles is the
only dynamical quantity of interest. However, there has been a recent
theoretical
interest [10-12,20] in the kinetics of fragmentation of multidimensional
objects. This is motivated by a desire to move away from
characterising particles solely by their volume or equivalently, mass
and towards
an understanding of the physical role played by shape in the
fragmentation of particles. In reality, particles are identified by their
shape.
The studies of fragmentation phenomenon in two dimensions [10-12, 20] have
revealed unexpectedly rich patterns with interesting and novel
statistics.
We attempt to clarify the origin of this behaviour
and invoke the idea of multifractality to characterise such pattern.
We find that some fragmentation rules do not have a
single measure support instead they have an infinite number. Each
measure yields different spectrum of
exponents to characterize the system.
We also consider a second model which exhibits simple scaling and we seek
to explain the difference between these models.
\noindent
The obvious extension of the rate equation to two dimensions
[10-13]] is \begin{eqnarray}
{{\partial f(x,y;t)}\over{\partial t}} =
-f(x,y;t)\int_0^xdx_1\int_0^ydy_1F(x_{1},y_1,x,y)+ \nonumber \\
s\int_x^{\infty}dx_1\int_y^{\infty}dy_1f(x_1,y_1;t)F(x,y,x_1,y_1)
\end{eqnarray}
where $f(x,y;t)$ is the concentration of particles of sides $x$
and $y$ at time $t$ and $s=1,2,3$ or $4$.
$F(x_1,y_1,x,y)$ describes the rate with which objects having sides
$x$ and $y$ break to produce fragments of sides $x_1$, $x-x_1$ and
$y_1$, $y-y_1$.
The two integrals over the two variables implies that the two orthogonal
cracks are placed on an objects such that the cracks are equal and
parallel to the sides. Evidently they will produce four new fragments at
each time step to describe fragmentation process ($s=4$). However, we can
choose $s=1,2,3$ which simply implies that $s$ fragments are kept and $(4-s)$
fragments are removed at each time step. This process creates a
stochastic fractals [14,15]] at long times. In this letter, we consider
our two variables $x$ and $y$ to be lengths. This is in contrast to [12],
where one variable was associated with energy and the other with mass.
We choose to study a homogeneous rate kernel [12],
\begin{equation}
F(x_1,y_1,x,y)=x^{\beta_1}y^{\beta_2}
\end{equation}
This model (called A in the abstract) describes a system in which particles
are selected for fragmentation with a rate determined by their
area
and shape. The relative importance of area and shape is
determined by $\beta_1$ and $\beta_2$.
Once a particles has been chosen for fragmentation, products
of any area and shape are equally likely so that the daughter distribution is
Poisson. Substituting this choice of kernel into the rate equation yields
\begin{equation}
{{\partial f(x,y,;t)}\over{\partial t}} =
-x^{\beta_1+1}y^{\beta_2+1}f(x,y;t)+
s\int_x^{\infty}\int_y^{\infty}dx_1dy_1x_1^{\beta_1}y_1^{\beta_2}f(x_1,y_1;t)
\end{equation}
\noindent
We now define the moments of the probability distribution
function $f(x,y;t)$ as
\begin{equation}
M_{m,n}(t) = \int_0^{\infty}\int_0^{\infty}dxdyx^{m-1}y^{n-1}f(x,y;t)
\end{equation}
We can then obtain a rate
equation for the moments,
\begin{equation}
{{\partial M_{m,n}(t)}\over{\partial t}}=({s\over{mn}}-1)M_{m+\beta_1+1,n+
\beta_2+1}(t)
\end{equation}
An interesting feature of the above equation is that there are
infinitely many conserved (time independent) moments. Moments
$M_{m,n}(t)$ that are conserved satisfy $mn=s$. This simply reflects the
fact that fragments with a given area
can have an infinite number of different shapes.
Using Charlesby's method, the moment equation can be iterated to
get all
the derivatives of the moments [11,12]. These can then be substituted into a
Taylor series expansion of
$M_{m,n}(t)$ about $t=0$ to give
\begin{equation}
M_{m,n}(t) = \ _{2}F_2(a_1,a_2;b_1,b_2;-t)
\end{equation}
\noindent
$_2F_2$ is a generalised hypergeometric function [15], where
\begin{equation}
a_{1/2} = {m \over{2(\beta_1 +1)}}+{n\over{2(\beta_2+1)}}
\mp\sqrt{{(m(\beta_2+1)-n(\beta_1+1))^2+4s
(\beta_1+1)(\beta_2+1)}\over{4(\beta_1+1)^2(\beta_2+1)^2}}
\end{equation}
and
\begin{equation}
b_1= {m\over{(\beta_1+1)}}
\end{equation}
\begin{equation}
b_2 = {n\over{(\beta_2+1)}}
\end{equation}
We are only interested in the long time behaviour of the moments.
The asymptotic
expansion of the generalised hypergeometric function for large time $t$ gives
\begin{equation}
M_{m,n}(t)\approx {{\Gamma (b_1) \Gamma (b_2) \Gamma(a_1-a_2)}
\over{\Gamma (a_2) \Gamma(b_1-a_1)\Gamma(b_2-a_1)}}t^{-a_1}
\end{equation}
For $s=4$ the conserved moments can be written as
$M_{m^{\ast},{{4}\over{m^{\ast}}}}(t)$ where,
$m^{\ast}$ is any number. We will only consider $m^\ast$ positive.
Obviously, $M_{2,2}(t)$ can be identifyed as the
area of the system. For convenience, we
choose $\beta_1=\beta_2=\beta$ since it does not change the physics of
the mechanism under investigation.
We may choose to
associate each hidden conserved quantity of the $s=4$
process with the set of points in ${\cal R}^2$. This space can
then be subdivided into boxes of size
\begin{equation}
\delta_{m\ast} =
\sqrt{{{M_{m^\ast,{{4}\over{m^\ast}}}(t)}\over{M_{1,1}(t)}})}.
\end{equation}
such that $\mu_i(\delta_{m^\ast})$ denotes the measure within the
$i^{th}$ box that depends on the choice of $s$. This choice of
$\delta_{m^\ast}$ ensures that for a pure fragmentation process ($s=4$),
we recover the complete set of points in ${\cal R}^2$. We
now express the moment $M_{m,1}(t)$ in terms of $\delta_{m^\ast}$
\begin{equation}
M_{m,1}(\delta_{m^\ast}) \sim
\delta_{m^\ast}^{-\gamma(m^\ast)(\sqrt{(m-1)^2+4s}-m-1)} \end{equation}
where
\begin{equation}
\gamma(m^\ast)={{2}\over{(m^\ast+{{4}\over{m^\ast}})-\sqrt{(m^\ast
-{{4}\over{m^\ast}})^2+4s}-2+\sqrt{4s}}}
\end{equation}
When $m=1$, the exponent of equation (13) gives the
Hausdorff-Besicovitch dimension $D_f$.
Note that $\gamma(m^\ast)= \gamma(4/m^\ast)$ and that in the limit
$m^\ast \rightarrow \infty$, when $s>1$, $\gamma(m^\ast) \rightarrow
1/(\sqrt{s}-1)$. We
also see that for $s=4$, $\gamma(m^\ast)=1$ and $D_f=2$, independent of
$m^\ast$, as we ensured with our choice of $\delta_{m^\ast}$.
However, for $1<s<4$ there exist infinitely many
supports for different values of $m^\ast$ and each is subdevided
by a corresponding $\delta_{m^\ast}$. We also can write the $d-$measure
of the weighted box number $M_{m,1}(t)$ as
\begin{equation}
M_{m,1}(d, \delta_{m^\ast})=\sum_i
\mu_i^{k(m)}\delta_{m^\ast}^d =
N(k(m),\delta_{m^\ast)})\delta_{m\ast}^d.
\end{equation}
Where $ N(k(m),\delta_{m^\ast}^d)$ is the $k(m)^{th}$ moment of the
measure such that $N(0,\delta_{m^\ast})$ is the number of boxes
require to cover the support of dimension $d=D_f$. Hence we can write the
weighted box number as
\begin{equation}
N(k(m),\delta_{m^\ast}^d)=
\sum_i\mu_i^{k(m)} \sim
\delta_{m^\ast}^{-\tau(k(m))}
\end{equation}
We thus see that $M_{m,1}(\delta_{m^\ast})$ can be partioned into boxes
of sides $\delta_{m^\ast}$ such that the probabilities $\{\mu_i\}$
are normalised if we let
\begin{equation}
m=1+(s-1)k
\end{equation}
when $\delta_{m^\ast}\rightarrow 0$ we require that the measure
$M_{m,1}(d,\delta_{m^\ast})$ tends to a finite value. This occurs when
$d=\tau(k)$. Combining (13),(15) and (17) gives the mass exponent
$\tau(k)$ as
\begin{equation}
\tau(k)=\gamma(m^\ast)\{\sqrt{(s-1)^2k^2+4s}-(s-1)k-2\}
\end{equation}
\noindent
This expression meets some essential requirement; namely $\tau(0)$ is the
dimension of
the support and $\tau(1)=0$.
We thus see that there exist a spectrum of mass exponents $\tau(k)$ that
characterise the distribution of the particle size distribution.
The mass exponent is nonlinear which indicates the existence of a
fractal subset for each support whether or not the support itself is
fractal.
To find this
fractal subset we use the usual Legendre transform of the independent variables
$\tau$ and $k$ to the independent variable $\alpha$ and $f(\alpha(k))$;
\begin{equation}
\alpha(k)=-{{d\tau(k)}\over{dk}}
\end{equation}
and
\begin{equation}
f(\alpha(k))=k\alpha(k)+\tau(k)
\end{equation}
These relations yield
\begin{equation}
\alpha(k)=\gamma(m^\ast){{(s-1)\sqrt{(s-1)^2k^2+4s}-(s-1)^2k}\over{\sqrt{(s-1)^2k^2+4s}}}
\end{equation}
and
\begin{equation}
f(\alpha(k))=\gamma(m^\ast){{4s-2\sqrt{(s-1)^2k^2+4s}}\over{\sqrt{(s-1^2k^2+4s}}}
\end{equation}
It is interesting to note that all the quantities of interest,
$\gamma(m^\ast)$, $\tau(k)$ and $f(\alpha(k))$ are independent of $\beta$
when $\beta_1=\beta_2=\beta$.
When $\beta_1 \neq \beta_2$
all the quantities depend on both $\beta_1$ and $\beta_2$ so
the analogous expressions are much more complex. However, the basic
picture is unchanged. For all $\beta_1$ and $\beta_2$ the $f-\alpha$
spectrum obeys a simple scaling relationship with respect to $\gamma$,
namely $f(\alpha)=\gamma h(\alpha/\gamma)$. In figure $1$ the $f-\alpha$
spectrum is plotted for three different values of $m^\ast$.
Physically, the $f(\alpha(k))$ versus $\alpha$ curve simply suggest
the existence of intertwined fractal subsets describing the measure.
We find that when fragments are removed from the system at each time step
there
exist a range of fractal dimension $1.4641 \leq D_f(m^\ast) \leq 2$ and
there are an infinitely manny $f-\alpha$ spectra for each $D_f$. All the
$D_f(m^\ast)$ compete on equal footings to be the support on which the
measure can distributed in a given realisation. This reflects that in
addition to the entropy from the location of the fractal subset there is
another source of entropy from $D_f(m^\ast)$ which is absent in the pure
fragmmentation ($s=4$) process.
As an aside, let us mention the connection between these models
for $1<s<4$ and
those of random sequential adsorption (rsa) (see [17] for a recent review).
One can imagine that at each time step, the fragmentation event is a
deposition
in which $4-s$ fragments are deposited on the substrate and play no further
part in the
kinetics and $s$ regions of the substrate survive for future deposition.
The difference between true rsa and our system is that in our system
deposition can only take place in rather restricted set of position
also as time proceeds the deposited particles get smaller in size.
However, in the long time the pattern created looks rather like that
created by the deposition of a mixture of sizes of ractangles. This
is the $2-d$ variant of the $1-d$ system of mixture deposition
studied by [18]. The $s=2$ version of this system studied in [19] the
context of deposition of needles. In rsa, one of the interesting
observables is that system reaches a jamming limit at long time
which is less than random close packing. However, in this case we find
instead of having a jamming limit the number density shows power law
behaviour with non-trivial exponent.
In general, the jamming limit can be found using the following relation
\begin{equation}
\theta(t \rightarrow \infty)=1-\int_0^\infty dx\int_0^\infty dy f(x,y;t)
\end{equation}
This gives
\begin{equation}
1-\theta(t \rightarrow \infty) \sim t^{\sqrt{s}-1}
\end{equation}
To understand the role
played by the dimension and shape we consider now a different model ie
\begin{equation}
F(x_1,y_1;x,y)=x^{\beta_1}y^{\beta_2}\delta(2x_1-x)\delta(2y_1-y)
\end{equation}
Substitute this into the rate equation to get,
\begin{equation}
{{\partial f(x,y;t)}\over{\partial
t}}=-{1\over{2^2}}x^{\beta_1}y^{\beta_2}f(x,y;t)+2^{\beta_1+\beta_2}sf(2x,2y;t)
\end{equation}
And the rate equation for the moments
\begin{equation}
{{\partial M_{m,n}(t)}\over{\partial
t}}=-({1\over{2^2}}-{s\over{2^{m+n}}})M_{m+\beta_1,n+\beta_2}(t)
\end{equation}
This again gives infinitely many hidden conserved quantity. As before the
condition for the conserved dynamical quantity can be
obtained from the requirement of time independent moments
\begin{equation}
m+n=2+{{{\ln}s}\over{{\ln}2}}
\end{equation}
This linear relation between $m^,$s and $n^,$s implies asymptotic power
law decay of the moments with linear exponent in $m^,$s and $n^,$s in time:
\begin{equation}
M_{m,n}(t) \sim A(m,n)t^{-\alpha(m+n)}
\end{equation}
Substitute this into the rate equation for the moments to give
a difference equation
\begin{equation}
\alpha(m+n+\beta_1+\beta_2)=\alpha(m+n)+1
\end{equation}
Iterating the above difference equation and using the appropriate
boundary condition gives
\begin{equation}
\alpha(m+n)= {{{{m+n-(2+{{{\ln}s}\over{{\ln}2}})}}}\over{(\beta_1+\beta_2)}}
\end{equation}
This gives us the power law decay of all the moments and allows us to
show that the average number of fragments $<N(t)>$ and the average area
are related by
\begin{equation}
<N(t)> \sim <xy>^{-{{{\ln}s}\over{2{\ln}2}}}
\end{equation}
Consequently, the fractal dimension for $s<4$ is given by
\begin{equation}
D_f={{{\ln}s}\over{{\ln}2}}
\end{equation}
and we have $f(\alpha)=D_f$ and $ \tau(k)=D_f(1-k)$ (self-similar). In
short, model B exhibits simple scaling.
\vspace{3mm}
These results give us the opportunity to ask why one needs an
infinite number of independent exponents to characterise the scaling
relations in model A while model B only exhibits scaling.
To find the answer we need to go back to the nature of the models
themselves and search for the things we lost in moving from the model A
to the model B. In the model A we had stochastic
homogenieity which implies the fragmentation of an object
possesses ergodic probability distribution. In this model, two
orthogonal cracks are placed independently parallel to the sides ie they
can pass through
any points in Euclidean space. While the
model B describe two orthogonal cracks are allowed to
place only at the middle of the objects to produce successfully four equal
sized fragments. It implies that the size is no longer intrigued with
shape ie shape is determined by the initial condition. Note that it is
one of the infinitely many possibility of the former model. Thus if there
is a mixture of particles of different size and shape, and if any fragments
are
equally likely to be picked, in the second model, once a fragments with
definite shape
is picked for fragmentation that will only produce of that shape. Thus it is
the broken ergodicity in shape that causes the absence of multiscaling.
In conclusion, we have studied the two different models to understand the
fragmentation phenomenon when there are more than one dynamical
variable. We found significantly different behaviour between the two
models, although both models have an infinite number of conserved moments.
Interestingly, these models also helps us to explain the occurence of
multifractality in fragmenting systems, which
is not yet fully understood.
\newpage
\begin{flushleft}
{\large \bf Acknowledgement}
\vskip5mm
MKH would like to thank the CVCP for an ORS award.
\end{flushleft}
\vspace{8mm}
\begin{flushleft}
{\large \bf References}
1. Englemen R 1991 J. Phys: condensed matter {\bf 3} 1019
2. Mark H and Simha R 1940 Trans. Faraday Soc. {\bf 35} 611
3. Montroll E W and Simha R 1940 J. Chem. Phys. {\bf 8} 721
4. Filippov A F 1961 Theory Probab. Appl. {\bf 4} 275
5. Kolmogorov A N 1941 Dan USSR (Doklady) {\bf 31} 99
6. Ziff R M and McGrady E D 1985 J. Phys. A {\bf 18} 3027
7. McGrady E D and Ziff R M 1987 Phys. Rev Lett {\bf 58} 892
8. Williams M M R 1990 Aero. Sci. and Technol. {\bf 12} 538
9. Cheng Z and Redner S 1990 J Phys A {\bf 23} 1233
10. Rodgers G J and Hassan M K 1994 Physical Rev. E {\bf 50} 3458
11. Krapivsky P L and Ben-Naim E 1994 Phys Rev E {\bf 50} 3508
12. Boyer D Tarjus G and Viot P 1995 Phys Rev E {\bf 50} 1043
13. Singh P and Hassan M K 1996 Phys. Rev. E in press
14. Krapivsky P L Ben-Naim E 1994 Physics Letters A {\bf 196} 168
15. Hassan M K and Rodgers G J Physics Letters A {\bf 208} 95
16. Luke Y L 1969 The special functions and their approximation {\bf 1}
( New York, Academic Press)
17. Feder J, 1988 Fractals (New York, Plenum).
18. Evans J W 1991 Rev. Mod. Phys. {\bf 65} 1281
19. Krapivsky P L 1992 J. Stat. Phys. {\bf 69} 135
20. Tarjus G and Viot P 1991 Phys. Rev. Lett. {\bf 67} 1875
\end{flushleft}
\newpage
{\bf Figure}
\vspace{5mm}
1. Three of the $f-\alpha$ spectra for model A when $s=3$ and
$\beta_1=\beta_2$. The three curves are for $m^\ast =2,4$ and $\infty$.
\end{document}