INSTITUTE OF PHYSICS PUBLISHING

JOURNAL OF PHYSICS G: NUCLEAR AND PARTICLE PHYSICS

J. Phys. G: Nucl. Part. Phys. 32 (2006) 12791293

doi:10.1088/0954-3899/32/9/006

A comparative study of multifractal moments in

relativistic heavy-ion collisions
Shafiq Ahmad and M Ayaz Ahmad
Department of Physics, Aligarh Muslim University, Aligarh-202002, India
E-mail: sahmad2004amu@yahoo.co.in

Received 4 March 2006

Published 14 August 2006
Online at stacks.iop.org/JPhysG/32/1279
Abstract
A study of the fractal structure of relativistic shower particles produced in the
interactions of 28Si and 12C nuclei at 4.5 A GeV/c with nuclear emulsion using
the methods of Takagi moments, Tq , factorial moments, Fq , and modified
multifractal moments, Gq , has been performed. The dependences of these
moments on the number of bins M are found to follow power law behaviour
for the experimental data in pseudorapidity phase space. The calculated values
of the generalized fractal dimensions, Dq , are found to decrease with the
increasing order of moments, q. This observation indicates the existence
of multifractality and a self-similar cascade mechanism in multiparticle
production. The multifractal Bernoulli representation Dq = D + c ln q/(q 1)
is found to support the linear dependence of Dq on ln q/(q 1). Finally, from
the analysis of the data it can be concluded that no universality in the values
of the multifractal specific heat, c, calculated from Tq , Fq and Gq moments
is found. Some consistency seems to be observed in the values of the specific
heat obtained from the Takagi method.

1. Introduction
There has been much interest in studying large density fluctuations observed in high-energy
hadronic and heavy-ion collisions. They have attracted special attention to understand the
underlying mechanism responsible for particle production. The existence of large fluctuations
may also indicate a phase transition. Various methods [1, 2] have been suggested to identify
the existence of non-statistical fluctuations. Bialas and Peschanski [1] were the first to
introduce the most suitable method known as scaled factorial moments (SFMs) to study the
non-statistical fluctuations in the distributions of relativistic shower particles produced in
high-energy collisions. The proposal of these factorial moments was made in analogy with
the phenomenon known as intermittency in the hydrodynamics of turbulent fluid flow. Also,
in high-energy physics, the power law behaviour of the scaled factorial moment is known as
0954-3899/06/091279+15$30.00 2006 IOP Publishing Ltd Printed in the UK

1279

1280

S Ahmad and M A Ahmad

intermittency. One of the possible characteristics of the scaled factorial moment analysis is that
it can detect and characterize the dynamical fluctuation and it is also capable of filtering out the
statistical noise. In this method, the scaled factorial moments, Fq , are computed as a function
of decreasing phase space size. The values of Fq for purely statistical fluctuations saturate
with decreasing phase space size, whereas in dynamical fluctuation Fq moments are supposed
to increase with decreasing phase space size and exhibit power law behaviour. Evidence for
intermittency received tremendous support when this behaviour was observed in electron
positron annihilation [35], hadronhadron [6], hadronnucleus [7, 8] and nucleusnucleus
collisions [911]. A comprehensive review on the subject can be found in [12].
The self-similarity observed in the power law dependence of scaled factorial moments
reveals a connection between intermittency and fractality. Hwa [2] was the first to provide
the idea of using multifractal moments, Gq , to study the multifractality and self-similarity in
multiparticle production. If the particle production process exhibits self-similar behaviour,
also the Gq moments show a remarkable power law dependence on phase space bin size.
However, if the multiplicity is low, the Gq moments are found to be dominated by statistical
fluctuations. In order to suppress the statistical contribution, a modified form of Gq moments
in terms of the step function was suggested by Hwa and Pan [2], which can act as a filter
for the low multiplicity events [2, 13]. Also, this method has been used to understand the
multifractality in lepton, hadron and nuclear collisions [1416]. However, the shortcoming
in all these analyses is that the experimental data on the scaled factorial moments, Fq , and
on the multifractal moments, Gq , only approximately show the expected linear behaviour
on a loglog plot. To overcome this difficulty, Takagi [17] has proposed a new method for
studying the multifractal structure of multiparticle production and successfully applied this
method to study fractality in UA5 data on protonantiproton interactions [18] and TASSO and
DELPHI data on electronpositron annihilations [19, 20]. It has been pointed out by Takagi
that the deviations from the linear behaviour in a loglog plot may be partly due to the fact that
the above methods are unable to give the required mathematical limit, the number of points
tending to infinity.
In the present study, we have investigated the existence of the multifractal
structure of multiplicity distributions of relativistic shower particles produced from 28Si
and 12CAgBr interactions at 4.5 A GeV using the methods of Takagi moments, Tq , scaled
factorial moments, Fq , and modified multifractal moments, Gq . The variation of the
generalized dimensions, Dq , with the order of the moments q corresponding to the three
types of moments is investigated. We have also looked at the behaviour of the multifractal
specific heat as obtained from the generalized dimensions, Dq , obtained from the Tq moments,
Fq moments and Gq moments, respectively.

2. Experimental details
In the present work, two stacks of BR-2 emulsion exposed to 4.5 A GeV/c silicon and
carbon beams at the Synchrophasotron of the Joint Institute of Nuclear Research (JINR),
Dubna, Russia, have been utilized. The method of line scanning has been adopted to scan
the stacks, which was carried out with Japan-made NIKON (LABOPHOT and Tc-BIOPHOT)
microscopes with an 8 cm movable stage using 40 objectives and 10 eyepieces. The
interactions due to beam tracks at an angle of <2 to the mean direction and lying in the
emulsion at depths >35 m from either surface of the pellicles were included in the final
statistics. Some other relevant details about the present experiment may be found in our
earlier publications [21].

A comparative study of multifractal moments in relativistic heavy-ion collisions

1281

2.1. Classification of tracks

All charged secondaries in these events were classified, in accordance with the emulsion
terminology, into the following groups [22].
2.1.1. Black track producing particles (Nb). Tracks with specific ionization g > 10 (g =
g/g0, where g0 is the Plateau ionization of a relativistic singly charged particle and g is the
ionization of the charged secondary) have been taken as black tracks. These correspond to
protons of relative velocity < 0.3 and range in emulsion L < 3.0 mm.
2.1.2. Grey track producing particles (Ng). Tracks with specific ionization 1.4
g
10
corresponding to protons with velocity in the interval 0.3

0.7 and range L  3.0 mm in
nuclear emulsion are called grey tracks.
2.1.3. Shower tracks producing particles (Ns). Tracks with specific ionization g < 1.4
corresponding to protons with relative velocity > 0.7 are classified as shower tracks. These
tracks are mostly due to relativistic pions with small admixture of charged K-mesons and fast
protons.
In order to eliminate all the possible backgrounds due to overlap (where a from
a 0 decay converts into e+e pair) close to shower tracks near vertex, special care was
taken to exclude such e+e pairs from the primary shower tracks while performing angular
measurements. Usually all shower tracks in the forward direction were followed more than
100200 m from the interaction vertex for angular measurement. The tracks due to e+e pair
can be easily recognized from the grain density measurement, which is initially much larger
than the grain density of a single charged pions or proton track. It may also be mentioned that
the tracks of an electron and positron when followed downstream in nuclear emulsion showed
considerable amount of Coulomb scattering as compared to the energetic charged pions.
Such e+e pairs were eliminated from the data.
2.2. Grouping of targets
The black and grey tracks together in an event are known as heavily ionizing tracks (Nh = Nb
+ Ng). In an emulsion experiment, the exact identification of the target is not possible since
the medium is composed of H, C, N, O, Ag and Br nuclei. The events produced due to the
collisions with the different targets are usually classified into three main categories on the
basis of the multiplicity of heavily ionizing tracks. Thus, the events with Nh
1, 2
Nh

7 and Nh  8 are classified as collisions with hydrogen (H, AT = 1), the group of light nuclei
(CNO, AT = 14) and the group of heavy nuclei (AgBr, <AT> = 94), respectively. However,
the grouping of events on the basis of Nh values only does not lead to the right percentage of
interactions due to the light and heavy groups of nuclei. In fact, a considerable fraction of
events with Nh
7 are produced in the interactions in the heavy group of nuclei. Therefore
we have used the following criteria:
AgBr events: (i) Nh > 7, or
(ii) Nh
7 and at least one track with range, R
10 m and no track with 10 < R < 50 m;
CNO events: (i) 2
Nh
7 and no tracks with R
10 m;
H events: (i) Nh = 0, or
(ii) Nh =1 and no track with R
50 m.
In the present investigation, only interactions with Ns  8 are considered, since the events
with low multiplicity (Ns < 8) would have large statistical fluctuations. The numbers of

1282

S Ahmad and M A Ahmad

interactions with Ns  8 for silicon and carbon beams with emulsion nuclei are 445 and 307,
respectively. Out of these events only 291 and 258 events, respectively, were finally selected as
genuine AgBr interactions. Only AgBr events were considered to maximize the contribution
of dynamical fluctuations [7].
Having measured the space angle ( s) of each track, the pseudorapidity () of the
relativistic shower particles was found using the relation = ln tan(s /2). Its accuracy
is 0.1 unit in pseudorapidity. At very high energies, is found to be a good approximation
of rapidity, Y, since experimentally it is not always possible to measure energy and momentum
of a particle, distributions in rapidity space are generally studied in terms of pseudorapidity,
, instead of the rapidity variable, Y.
Nuclear emulsion is a global 4 detector and possesses a very high spatial resolution
(0.1 mrad) as compared to all other detectors currently used in high-energy physics [23, 24].
Because of this advantage, nuclear emulsion is used as a very effective detector for studying
the multifractal behaviour of multiparticle production.
3. Mathematical formalism and experimental results
3.1. Takagi moments
Takagi [17] proposed a new method to understand the multifractal structure of multiplicity
distributions of charged hadrons in pseudorapidity phase space. In this method, a single event
contains n relativistic shower particles distributed in the interval min <  < max in -phase
space. The multiplicity of n charged particles changes from event to event according to the
distribution Pn(), where  = max min. The pseudorapidity interval is divided into M
bins of equal size = /M. The multiplicity distribution for a single bin is represented
as Pn() for n = 0, 1, 2, 3, . . . , where it is assumed that the inclusive rapidity distribution
dn/d is constant and Pn() is independent of the location of the bin. If  denotes the
number of independent events, then the shower particles produced in  events are distributed
in M bins of size . Let N be the total number of hadrons produced in these events and nij
the multiplicity of particles in the jth bin of the ith event. In accordance with the theory of
multifractals [25], the normalized density, Pij , is defined by
Pij = nij /N.
This is also true for N . The Takagi moment of order q is defined as
Tq () = ln



M

Pij

for

q > 0.

(1)

i=1 j =1

It is expected to behave like a linear function of the logarithm of the resolution R(),
Tq () = Aq + Bq ln R(),

(2)

where Aq and Bq are constants independent of . If such a linear relation is observed for a
considerable range of R(), a generalized dimension may be defined as
Dq = Bq /(q 1).

(3)

For a sufficiently large number of events , the summation in equation (1) is evaluated as


M

Pij = nq /(N q1 n).

(4)

i=1 j =1

Using equations (1)(4), the following linear relation is easily obtained for the simplest choice

A comparative study of multifractal moments in relativistic heavy-ion collisions

27

27

24

q=6
q=5

24

q=4

21

q=3

18

15

ln <n >

18

ln <n >

(a) Si-AgBr

q=2

12
9

q=6
12

(b) C-AgBr

28

21

1283

q=5

q=4
q=3

15
12

q=2

9
<n ln n>/<n>

0
0.5

1.0

1.5

2.0

2.5

3.0

3.5

0
0.5

ln <n>

<n ln n>/<n>

1.0

1.5

2.0

2.5

3.0

3.5

ln <n>

Figure 1. Variation of lnnq  as a function of lnn in the interactions of (a)

(b) 12CAgBr at 4.5 A GeV/c.

28SiAgBr

and

R() = :
lnnq  = Aq + {(q 1)Dq + 1} ln .

(5)

In order to verify the above relation, it was observed [18] that the variation of lnn  as a

function of ln c for UA5 data on pp collisions at s = 200 GeV shows a considerable

deviation from the linear behaviour for the larger c region (c is the pseudorapidity in the
centre-of-mass system). This deviation may be due to non-flat behaviour of dn/d in the
projectile and target fragmentation regions where || is large. It was suggested [17] that n
would be a better choice for R(). From equation (5), one gets the following relation:
q

lnnq  = Aq + {(q 1)Dq + 1} lnn

= Aq + Kq lnn,

(6)

where the slope Kq is related to the generalized dimension, Dq , by the following relation:
Dq = (Kq 1)/(q 1).

(7)

The existence of linear behaviour over a considerable range of n indicates the presence of
the fractal structure in the multiparticle production. The values of Dq for q  2 in expression
(3) can be determined easily from the slopes of the plot between lnnq  and lnn. The value
of information dimension, D1, for q = 1 can be determined from the simple relation
n ln n/n = C1 + D1 lnn.

(8)

Here, the idea of a multifractal structure of the relativistic shower particles has been investigated
for the AgBr target by looking at the behaviour of lnnq  versus lnn for q = 26 in a single
bin of varying size in pseudorapidity space. For this purpose, the analysis of the multiplicity
distribution in the central pseudorapidity region (peak 1.5 < < peak + 1.5) has been
considered, which covers most of the produced shower particles. The initial rapidity interval
 = 3.0 was subsequently reduced in steps of 0.25. The values of n ln n/n and nq  are
calculated for each interval. The values of lnnq  as a function of lnn for different q are
shown in figure 1 for 28SiAgBr and 12CAgBr collisions. All the observed points clearly
follow an excellent linear relation for the whole range of n. The values of the slopes, Kq, for
28
SiAgBr and 12CAgBr collisions were obtained by fitting relation (6) to the experimental

1284

S Ahmad and M A Ahmad

points. The linear behaviour of lnnq  with lnn in the above figure gives an indication of the
fractal structure in multiparticle production in -space. Similar observations have also been
reported by other workers [26].
3.2. Scaled factorial moments
It is believed that the multiplicity fluctuations in small phase space bins can reveal important
aspects of the multiparticle production mechanism such as an intermittent pattern of
fluctuations [1, 26, 27]. For the study of fluctuations in pseudorapidity distributions, a given
pseudorapidity interval of total length,  = max min, is divided into M bins of equal width,
= /M. The fluctuations in phase space are estimated by calculating the horizontally
scaled factorials (SFMs), Fq , of order q for a sample of events of varying multiplicities using
the relation [1]
Fq H =

Nev

M
nm,i (nm,i 1) . . . (nm,i q + 1)
M q1

,
Nev i=1 m=1
N q

(9)

where nm,i is the number of relativistic charged particles in the mth bin (m = 1, 2, 3, . . . , M)
of ith event, N represents the mean multiplicity of the hadrons in the pseudorapidity interval
 = M and Nev is the number of events. The scaled factorial moments are sensitive to
the shape of the pseudorapidity distribution. Thus for non-flat pseudorapidity, the corrected
horizontal scaled factorial moments are given as
Fq corr
H = Fq H /RF ,

(10)

where the correction factor, RF, suggested by Fialkowski et al [28] is given as

RF =

M
1
q
M nm q /N q .
M m=1

(11)

In the case of flat pseudorapidity distribution, however, the factor RF becomes insignificant.
The corrected horizontal moments are compared with the corresponding vertical moments
discussed below.
Another method known as vertical scaled factorial moment analysis is suggested to correct
for the non-uniform shape of -distribution and Fqv is expressed as
Fq V =

Nev
M
1
nm,i (nm,i 1) . . . (nm,i q + 1)
1

,
Nev i=1 M m=1
n m q

(12)

 ev
where n m  = N1ev N
i=1 nm,i is the average number of particles in the mth bins for the entire set
having number of events, Nev. The two definitions (9) and (12) become identical if the single
particle -distribution is flat. However, if the distribution is not flat, one should either consider
vertically averaged moments or apply a correction factor [28] to the horizontal moments.
It has been shown [1] that if the fluctuations are purely statistical, a saturation of Fq 
with decreasing resolution (or increasing number of bins M) is expected. However, if the
fluctuations are dynamical in nature, then in the limit of small bin size the scaled factorial
moments would obey [1] the following power law:


 q
Fq H (M)q
,
as 0.
(13)

The power law dependence of Fq H with q is known as intermittency, which predicts a linear
rise of lnFq H as a function of ln M.

A comparative study of multifractal moments in relativistic heavy-ion collisions

5.0

5.5
5.0
4.5

(a) 28Si -AgBr

4.5

q=6

FqHorizontal (corr)
FqVertical

4.0

q=5

4.0
q=4

ln <Fq>

ln <Fq>

(b) 12C-AgBr

q=6

FqHorizontal (corr)
FqVertical

q=5

3.5

3.5
3.0
2.5

q=3

2.0

q=4

3.0
q=3
2.5
2.0
1.5

1.5

1.0

1.0
q=2

0.5
0.0
0.5

1285

1.0

1.5

2.0

2.5

ln M

3.0

q=2

0.5
3.5

0.0
0.5

1.0

1.5

2.0

2.5

3.0

3.5

ln M

Figure 2. Variation of lnFq  as a function of ln M in the interactions of (a) 28SiAgBr and

(b) 12CAgBr at 4.5 A GeV/c. The cross points (x) are associated with the horizontally averaged
scaled factorial moments and the open circles () are for the vertically averaged scaled factorial
moments. The solid lines represent the best linear fits. The error bars are shown only on horizontal
moments.

The strength of the intermittency is characterized by the value of q. The value of q is

obtained by performing best fits according to the following relation:
q =

 lnFq H
.
 ln M

(14)

The power law behaviour of the scaled factorial moments can also be interpreted in terms of
fractal properties of the underlying physical process.
In order to study the intermittency in multiparticle production, the whole pseudorapidity
phase space is divided into number of bins, M = 230, and the average scaled factorial
moments, Fq corr
and Fq V , for q = 26 are calculated using equations (10) and (12),
H
respectively. In figure 2, lnFq corr
H and lnFq V are displayed as a function of ln M for our
data. The cross points and open circles in the diagram represent horizontally normalized
and vertically normalized averaged moments, respectively. One can readily see that there
is virtually no difference between two moments. The corrected horizontal moments are
generally in good agreement with vertical moments. It is evident from the figure that the
values of lnFq corr
H increase with decreasing bin size (i.e. increasing resolution). The linear
rise in the values of lnFq corr
H with ln M in figure 2 indicates the power law behaviour, which
clearly suggests that an intermittent behaviour is being observed. The error bars in figures 2
and 3 are estimated by considering them as independent statistical errors only and the solid
lines drawn indicate the least-squares fit to the respective experimental data points. Though the
effect of point-to-point correlations of the statistical errors for different bin sizes has not been
taken into consideration in the present study, it is expected that the exclusion of the correlation
of the statistical errors will not change the main result appreciably [6, 29]. The values of the
intermittency indices, q, for various orders of moments, q, along with correlation coefficient
(R) values for the best fitted lines of figures 2 and 3 are listed in table 1. It is observed that the
parameter, q, increases with the increasing order of the moments. The increasing trends of
q with q are in accordance with the predictions of the -cascade model [1, 30].

1286

S Ahmad and M A Ahmad

0

-1

-1

-3
q=3

-4
-5

q=4
28

-9
0.5

q=3

-4
-5

q=4

-6

-6

-8

ln <Gq>

ln <Gq>

-3

-7

q=2

-2

q=2

-2

q=5

(a) Si-AgBr
Experimental data
Monte Carlo data
1.0

1.5

2.0

-7
q=6
2.5

3.0

3.5

-8

12

q=5

(b) C-AgBr
Experimental data
Monte Carlo data

-9
0.5

1.0

1.5

2.0

q=6
2.5

3.0

3.5

ln M

ln M

Figure 3. Variation of lnGq  as a function of ln M in the interactions of (a) 28SiAgBr and

(b) 12CAgBr at 4.5 A GeV/c. Open circles () indicate the uncorrelated Monte Carlo data in
(a) and (b).

Table 1. Values of q, R, q, q1 qdyn and  q. The errors are statistical. Values in the
parentheses represent the correlation coefficient (R).
Energy
4.5 A GeV/c q q (R)
28SiAgBr

12CAgBr

2
3
4
5
6
2
3
4
5
6

0.12 0.03 (0.95)

0.24 0.05 (0.99)
0.58 0.05 (0.99)
0.95 0.06 (1.01)
1.17 0.23 (0.99)
0.15 0.03 (0.98)
0.37 0.05 (0.99)
0.72 0.04 (0.99)
1.09 0.02 (0.99)
1.51 0.03 (0.99)

qexp (R)

qstat

qdyn

q1 qdyn

 q
= qstat q

0.70 0.01 (0.97)

1.35 0.01 (0.98)
1.92 0.02 (0.99)
2.41 0.04 (0.97)
2.84 0.06 (0.99)
0.62 0.01 (0.98)
1.21 0.02 (0.96)
1.74 0.03 (0.98)
2.23 0.04 (0.96)
2.68 0.06 (0.99)

0.79 0.04
1.67 0.01
2.63 0.02
3.59 0.03
4.37 0.05
0.79 0.01
1.68 0.03
2.59 0.03
3.50 0.04
4.39 0.05

0.92 0.11
1.68 0.16
2.29 0.03
2.83 0.05
3.48 0.06
0.83 0.02
1.53 0.05
2.15 0.03
2.73 0.05
3.30 0.06

0.09 0.01
0.32 0.01
0.75 0.03
1.17 0.05
1.50 0.07
0.17 0.02
0.47 0.03
0.85 0.04
1.27 0.05
1.71 0.06

0.09 0.01
0.32 0.02
0.72 0.03
1.17 0.05
1.53 0.07
0.17 0.02
0.47 0.04
0.85 0.04
1.27 0.06
1.70 0.07

3.3. Modified multifractal moments

The modified multifractal moments proposed by Hwa and Pan [2] are more closely related to
the fractal behaviour of particles produced in the multiparticle system. They have been used
to minimize the contribution of the statistical fluctuations. In this method N single charged
shower particles in a given interval  = max min are distributed into M nonempty bins of
width = /M. A modified, Gq moment is defined by the following relation,
Gq =

M

(nj /N)q (nj q),

(15)

j =1

where q is a positive integer, nj denotes the number of charged particles in the jth bin and

N= M
j =1 nj is the total number of particles detected in an event. The step function, (nj q)

q = M (nj /N )q in order to minimize the
has been added to the original definition of G
j =1

A comparative study of multifractal moments in relativistic heavy-ion collisions

statistical noise:
(nj q) =


1,
0,

if
if

nj  q
nj < q.

1287

(16)

For very large multiplicity, the value of the step function is unity, so the modified Gq moments
give the same result as the old G q moments. However, for low multiplicity events, the ordinary
G q moment is dominated by the statistical fluctuations.
The multifractal moment Gq  for M nonempty bins is defined as
Gq  =

Nev
1

Gq ,
Nev 1

(17)

where Nev stands for the total number of events in a given ensemble. According to the theory
of multifractals, if the multiplicity distribution of charged particles has a fractal structure, then
the power law behaviour of Gq moments can be expressed by the following relation:
Gq  M q ,

(18)

where q is the fractal index and can be found from a linear dependence of lnGq  on ln M
over all windows as
 lnGq 
.
(19)
q =
 ln(M)
In order to examine the dependence of lnGq  on ln M, the values of the multifractal
moments Gq  for q = 26 have been calculated using equations (15) and (17) for our data
at 4.5 A GeV/c. The variations of lnGq  with ln M of the emitted relativistic shower particles
in -phase space are shown in figure 3 for 28SiAgBr and 12CAgBr collisions with Ns  8.
It is clear from the figure that the lnGq  now exhibits a linear dependence on ln M for q > 1
without saturation. This linear behaviour is found to satisfy the power law dependence as
described in expression (18). The values of q obtained using the least-squares fitting of the
experimental points in the above figure are listed in table 1. The observed linear dependence
of the multifractal moments gives an evidence of self-similarity for the mechanism of particle
production and an initial indication of the fractal structure in multiparticle production. A
similar trend of power law dependence of lnGq  on ln M for q >1 has also been verified on
p, p p and e+e data [31, 32].
To study the effect of statistical contribution to Gq moments, simple uncorrelated Monte
Carlo events, (MC-RAND) were generated randomly in -space according to the following
criteria:
(i) N such particles in each event are distributed randomly in the given -interval,
(ii) the multiplicity distribution and pseudorapidity distribution of generated events should be
similar to those of the experimental data.


with ln M for the uncorrelated MC events are also included in
The variations of ln Gstat
q
figure 3 by open circles with the corresponding experimental data. The slopes of the plots, qstat ,
thus obtained are also listed in table 1 along with the experimental data. A clear difference
of the pseudorapidity fluctuations is observed for statistically generated events. The pattern
of variation of lnGq  on ln M for uncorrelated MC events is more or less similar to that of
the experimental data, but the magnitudes of Gq moments are always significantly less in
comparison to the experimental values. This feature
 can be attributed to the existence of the
on M is shown by the relation
dynamical fluctuations. The power dependence of Gstat
q
Gq stat M q .
stat

(20)

1288

S Ahmad and M A Ahmad

2.0

2.0

1.8
12

1.6

(a) 28Si-AgBr

1.6

1.4

q
q

1.4

(b)

C-AgBr
q
q

1.2

1.2

q and q

q and q

1.8

1.0
0.8
0.6

1.0
0.8
0.6

0.4

0.4

0.2

0.2
0.0

0.0
2

Figure 4. Comparison of q and  q with the order of moment q in the interactions of

(a) 28SiAgBr and (b) 12CAgBr at 4.5 A GeV/c.

The dynamical contribution to the Gq moment suggested by Chiu et al [14] is expressed as


1q


Gq dyn = Gq  Gstat
M .
(21)
q
If Gq dyn also exhibits a power law dependence on M as
dyn

Gq dyn M q ,

(22)

then the dynamical contribution to q can be extracted from the experimental q and the qstat
using the following formula [2, 12, 14]:
qdyn = q qstat + q 1.

(23)

dyn
Gq

= M 1q provided Gq  is purely statistical, which implies from

Relation (21) reduces to
dyn
dyn
equation (23) that q = q 1 for trivial dynamics. The dynamical values of q obtained
stat
by inserting the values of q and q in equation (23) for q = 26 are listed in table 1. It
dyn

is evident from this table that the values of q deviate significantly from (q 1), which
indicates the presence of the dynamical fluctuations.
Hwa and Pan [2] developed a relation between the Gq moments and Fq moments, where
the intermittency index ( q) and fractal index ( q) are approximately related as
q (q 1) qdyn = qstat q .

(24)

The above equation is not an exact relation because Fq and Gq moments are different and
approach each other only in the limiting case of infinite N. The deviation of q from zero
gives a measure of non-statistical fluctuations of dynamical origin, which is equivalent
to the

deviation of q from q 1. Figures 4(a) and (b) give a comparison of q and q = qstat q
as a function of q obtained from table 1 for our data in -phase space. It is evident from
the figure that the values of q are not exactly equal to  q for 28Si and 12C projectiles,
respectively. They show a similar trend for all q except at q = 6, while these values are close
to one another only up to q
3, and beyond q > 3 they are different. This difference can be
attributed to the difference in the definitions of Fq and Gq moments, since the former is closely
related to the correlation function while the latter gives a measure of the fractal structures.
Further, it can be concluded that the deviation of q from the statistical qstat is a measure of the

A comparative study of multifractal moments in relativistic heavy-ion collisions

1.00

0.90

(a)
0.95
0.90

28

Si-AgBr
Tq
Fq
Gq

0.87

(b)

0.84
0.81

0.85

Dq

Dq

1289

12

C-AgBr
Tq
Fq
Gq

0.78
0.75

0.80

0.72

0.75

0.69
0.70
0.65

0.66
2

0.63

Figure 5. Dependence of the generalized dimension Dq on q for (a) 28SiAgBr and (b) 12CAgBr
collisions using Tq , Fq and Gq moments methods at 4.5 A GeV.

real dynamics involved. This is in agreement with the observation reported by other workers
[2, 26, 33].
4. Multifractality
The generalized dimensions, Dq , which characterize the fractal behaviour, are determined
from Tq , Fq and Gq moments analyses using the following relations:
Dq = (Kq 1)/(q 1)

(25)

Dq = 1 {q /(q 1)}

(26)

Dq = q /(q 1).

(27)

and
The values of the generalized dimensions, Dq , for different values of q obtained for our data
using above relations are shown in figures 5(a) and (b). From the figure it is observed that
the values of Dq decrease with the increasing order of the moments, q, and is always less
than 1 for all q. The decreasing trend of Dq with increasing q clearly gives an indication of
multifractal characteristics, which supports an interpretation in terms of a cascade mechanism
in multiparticle production. The values of the generalized dimension determined by the
intermittency analysis are consistently larger and do not match with the other moments. This
difference may be due to different approaches in the definitions of two moments, whose
differences become more prominent when N is low. In Takagis method, the total number
of particles could be made arbitrarily large by making the total number of events very large.
Another reason for the difference in Dq values may be attributed to the different ranges of
and the definitions of the modified Gq moments and Tq moments, where it is assumed
that the statistical fluctuations are suppressed, but actually they are not eliminated completely.
A similar behaviour of the multifractal structure is obtained in pp collisions [18] and e+e
annihilation [19, 20] using Takagis method.
In order to exclude the empty bins, we have extended our analysis of Takagis approach
to negative values of q. A linear relation between Dq and q for the values of q close to zero

1290

S Ahmad and M A Ahmad

Table 2. Values of (D0 Dq ) for different targets using the Tq moment method for our data. Upper
row and lower row represent the corresponding values for 28Si and 12C projectiles, respectively.

Type of
interactions
Nucleon
CNO
AgBr

(D0 Dq )
q=2

q=3

q=4

q=5

q=6

0.2840.002
0.2520.001
0.2230.001
0.2140.002
0.0940.001
0.1420.002

0.3640.003
0.3020.002
0.2820.002
0.2610.001
0.2040.002
0.2240.002

0.4320.001
0.3920.002
0.2990.003
0.3260.001
0.2570.004
0.2940.002

0.4740.002
0.4310.001
0.3420.002
0.3860.002
0.2990.002
0.3250.003

0.4990.002
0.4310.003
0.3520.003
0.3860.003
0.3120.003
0.3290.002

(0.5 < q < 0.5) is represented by

Dq = mq + c.

(28)
28

12

The values of the intercept (c) in the case of Si and C projectiles are found to be
approximately 1.002 0.003, 1.006 0.001 and 1.004 0.002, respectively, for nucleon,
CNO and AgBr targets in -space. The errors are statistical. This result corresponds to
D0 1 for q = 0 from equation (28). This result is explained by the absence of empty bins in
our analysis. The inhomogeneity of the multiparticle density distribution can be measured by
the quantity D0 Dq . The difference (D0 Dq ) or (1 Dq ) for different targets is illustrated
in table 2 for 28Si and 12C projectiles. We observe from the table that the values of (1 Dq )
decrease with increasing target mass. The variations of lnnq  as a function of lnn for
different q and Dq versus q plots are not shown in the figure for nucleon and CNO targets to
avoid more figures.
5. Multifractal specific heat
Recently, a multifractal Bernoulli distribution [34] has been introduced to describe the
transition from monofractality to multifractality, which is also believed to play a crucial role
in the multiparticle production at high energies. The following relation for the multifractal
Bernoulli function is given by
Dq = D + c ln q/(q 1),

(29)

where Dq is the generalized dimension of order q. The constant c in equation (29) can
be interpreted as the multifractal specific heat of the system provided thermodynamical
interpretation of the multifractality is used [35]. It is widely believed in thermodynamics
that the specific heat of gases and solids is constant and independent of temperature [36]
in a wide range of q. This analogy of constant specific heat approximation should also be
applicable to the multifractal specific heat.
Bershadskii [34] also compared the experimental data for 238U-Em interactions at
0.96 A GeV [33] and 197Au-Em interactions at 10.6 A GeV [33] with equation (29) and
found the values of the multifractal specific heats of 1/4 and 1/3 for heavy and light ions,
respectively.
In order to find the values of the multifractal specific heat, we have plotted the values of
the generalized fractal dimension, Dq , obtained from Tq , Fq and Gq moments analyses
against ln(q)/(q 1) in figures 6(a)(c) for 28SiAgBr and 28CAgBr interactions at
4.5 A GeV/c in -phase space. The straight lines shown in the above figures are best linear
fits to the experimental points. One can see from the figures that the values of Dq reveal a

A comparative study of multifractal moments in relativistic heavy-ion collisions

0.82

1291

(a) Tq - moment
28

Si-AgBr
C-AgBr

0.90

(b) Fq - moment

12

28

0.80

Si-AgBr
C-AgBr

12

Dq

0.84

0.78
Dq

0.78
0.76

0.72
0.74

0.66
0.72
0.36 0.42 0.48 0.54 0.60 0.66 0.72

0.36 0.42 0.48 0.54 0.60 0.66 0.72

lnq /(q-1)

lnq/(q-1)

0.95
0.90

Dq

0.85

(c) Gq - moment
28
Si-AgBr
12
C-AgBr

0.80
0.75
0.70
0.65

0.36 0.42 0.48 0.54 0.60 0.66 0.72

lnq/(q-1)

Figure 6. Dependence of the generalized dimension Dq on ln q/(q 1) for 28SiAgBr and

12CAgBr collisions using (a) T moment, (b) F moment and (c) G moment methods. Straight
q
q
q
lines represent the best fit which is in agreement with the multifractal Bernoulli representation
(equation (29)).
Table 3. Values of multifractal specific heat using Tq , Fq and Gq methods.
Interactions
28SiAgBr
12CAgBr
28SiAgBr
16OAgBr
32SAgBr

Energy
(A GeV)

Tq moments

Fq moments

Gq moments

References

4.5
4.5
14.5
60
200

0.30 0.03
0.29 0.03
0.31 0.03
0.26 0.01
0.37 0.04

1.23 0.23
0.72 0.16
3.17 0.31
1.40 0.30
0.66 0.13

0.96 0.16
0.66 0.06
1.97 0.15
1.95 0.33
0.32 0.20

Present work
Present work
[37]
[37]
[37]

linear increase as a function of ln(q)/(q 1). The linear behaviour in the figures indicates
good agreement between the experimental data and the multifractal Bernoulli representation.
The calculated multifractal specific heats extracted from these figures are shown in table 3
corresponding to Tq , Fq and Gq moments along with other results [37]. It is observed from
the table that there is no systematic variation in the values of c as determined by the different

1292

S Ahmad and M A Ahmad

methods. Moreover, no dependence is seen in the values of the multifractal specific heat
with the projectile mass and their energies. The obvious reason for this difference is due
to different values of the generalized dimension, Dq , obtained in the analyses of different
methods. However, it is very interesting to observe that in Takagis method some consistency
is observed in the values of the specific heat obtained from Tq moments.
6. Conclusions
The following conclusions may be drawn from the result of multifractal analysis of the data.
The study of Tq , Fq and Gq moments clearly exhibits a linear dependence for q = 26 in
28
SiAgBr collisions and 12CAgBr collisions at 4.5 A GeV/c. The decreasing trend of the
generalized dimension, Dq , with the increasing order of the moments indicates multifractality
and gives an evidence for the self-similar cascade mechanism in multiparticle production. The
value of D0 = 1 for different targets implies that there is no contribution of empty bins to our
analysis and the quantity (1 D0) decreases with increasing target mass.
The linear behaviour of Dq as a function of ln(q)/(q 1) favours the multifractal Bernoulli
representation for our data. No systematic behaviour in the values of the multifractal specific
heat determined by different methods is found. Some consistency seems to be observed in the
values of the specific heat obtained from Takagis method in -space.
References
[1] Bialas A and Peschanski R 1986 Nucl. Phys. B 273 703
Bialas A and Peschanski R 1988 Nucl. Phys. B 308 837
[2] Hwa R C and Pan J 1992 Phys. Rev. D 45 1476
Hwa R C 1990 Phys. Rev. D 41 1456
[3] Buschbeck B, Lipa R and Peschanski R 1988 Phys. Lett. B 215 788
[4] Abreu P et al (DELPHI Collaboration) 1990 Phys. Lett. B 247 137
[5] Braunschweig W et al (TASSO Collaboration) 1989 Phys. Lett. B 231 548
[6] Ajinenko I V et al (NA22 Collaboration) 1989 Phys. Lett. B 222 306
[7] Holynski R et al 1989 Phys. Rev. Lett. 62 733
[8] Holynski R et al (KLM Collaboration) 1989 Phys. Rev. C 40 R2449
[9] Adamovich M I et al (EMU01 Collaboration) 1990 Phys. Rev. Lett. 65 412
[10] Akensson T et al (HELIOS Collaboration) 1990 Phys. Lett. B 252 303
[11] Sengupta et al 1990 Phys. Lett. B 236 219
[12] Bozek P et al 1995 Phys. Rep. 252 101
De Wolf E A et al 1996 Phys. Rep. 270 1
Kittel W and De Wolf E A 2005 Soft Multihadron Dynamics (Singapore: World Scientific)
[13] Dremin I M 1988 Mod. Phys. Lett. A 3 1333
Carruthers P 1989 Int. J. Mod. Phys. A 4 5587
Brax Ph and Peschanski R 1990 Nucl. Phys. B 346 650
[14] Chiu C B and Hwa R C 1992 Phys. Rev. D 45 2276
Chiu C B et al 1990 Mod. Phys. Lett. A 5 2651
[15] Ghosh D et al 1991 Phys. Lett. B 272 5
[16] Jain P L et al 1992 Phys. Rev. C 46 721
[17] Takagi F 1994 Phys. Rev. Lett. 72 32
Takagi F 1984 Phys. Rev. Lett. 53 427
[18] Ansorge R E et al (UA5 Collaboration) 1989 Z. Phys. C 43 357
[19] Braunschweig W et al (TASSO Collaboration) 1989 Z. Phys. C 41 159
[20] Abreu P et al (DELPHI Collaboration) 1991 Z. Phys. C 52 271
[21] Tariq M et al 1992 Int. J. Mod. Phys. E 1 859
Tariq M et al 1993 Nuovo Cimento A 106 617
[22] Powell C F et al 1959 The Study of Elementary Particles by Photographic Methods (Oxford: Pergamon) p 450
[23] Bradnova V et al 1992 JINR, Dubna, LHE-0983-2 1

A comparative study of multifractal moments in relativistic heavy-ion collisions

[24] Adamovich et al 1991 Phys. Lett. B 262 369
Aggarwal M 1998 Physics and Astrophysics of Quark-Gluon Plasma (New Delhi)
[25] Hentschel H G E and Procaccia I 1983 Physica D 8 435
[26] Parashar N et al 1995 J. Phys. G: Nucl. Part. Phys. 21 173
Ghosh D et al 1998 Phys. Rev. C 58 3553
Rashid H et al 2000 Int. J. Mod. Phys. E 9 917
[27] Frisch U et al 1978 J. Fluid Mech. 87 719
[28] Fialkowski K et al 1989 Acta Phys. Pol. B 20 639
[29] Agababyan N M et al 1996 Phys. Lett. B 382 305
Ghosh D et al 2004 Int. J. Phys. A 78 353
[30] Kolmogorov A N 1941 Akad. Sci. USSR 30 301
Novikov E A and Stewart R W 1964 Inzh. Acad. Sci. USSR Ser. Fiz. 3 408
[31] Derado I et al 1992 Phys. Lett. B 283 151
[32] Albajar C et al (UA 1 Collaboration) 1992 Z. Phys. C 56 37
[33] Jain P L et al 1993 Nucl. Phys. A 561 651
Jain P L et al 1990 Phys. Lett. B 241 273
Jain P L and Singh G 1996 Nucl. Phys. A 596 700
Singh G and Jain P L 1994 Phys. Rev. C 50 2508
[34] Bershadskii A 1999 Phys. Rev. C 59 364
[35] Stanely E and Meakin P 1988 Nature 355 405
[36] Landau L D and Lifshitz E M 1934 M. Phys. Sz. Sowjet 6 244
Landau L D and Lifshitz E M 1980 Statistical Phys. Part I (London: Pergamon)
[37] Ghosh D et al 2003 Fractals 11 331

1293