QUANTUM MECHANICS, FIELD THEORY
AND SPIN SYSTEMS
H. KROG GER
DeH partement de Physique, UniversiteH Laval, QueH bec, QueH ., Canada G1K 7P4
AMSTERDAM } LAUSANNE } NEW YORK } OXFORD } SHANNON } TOKYO
H. KroK ger / Physics Reports 323 (2000) 81}181 81
Email address: hkroger@phy.ulaval.ca (H. KroK ger)
Physics Reports 323 (2000) 81}181
Fractal geometry in quantum mechanics, "eld theory
and spin systems
H. KroK ger
De& partement de Physique, Universite& Laval, Que& bec, Que& ., Canada G1K 7P4
Received May 1999; editor: J. Bagger
Contents
1. Introduction to fractal geometry 84
2. Fractal geometry in quantum mechanics 87
2.1. Brownian motion versus motion in
quantum mechanics 87
2.2. Path integral quantization and fractal
geometry of quantum paths 90
2.3. Numerical simulations for di!erent
potentials 94
2.4. Can we measure experimentally the
geometry of quantum mechanical
propagation? 100
3. Quantum physics on fractal space}time 108
3.1. SchroK dinger and Dirac equation viewed
from statistical mechanics 108
3.2. The principle of scale invariance 109
3.3. Quantum physics on Cantor sets 109
4. Fractal geometry and quantum "eld
theory 110
4.1. Selfsimilarity and renormalization group
equation 110
4.2. Selfsimilarity and scale dependence.
Hadron structure functions in QCD 113
4.3. Geometry of propagation of a relativistic
particle 117
4.4. Nonlocal order parameters for
con"nement in lattice gauge theory 134
4.5. Fractal geometry and critical behavior of
lattice "eld theories 147
5. Fractal geometry and quantum gravity 152
5.1. Random surfaces and quantization of
gravity 152
5.2. Fractal structure 154
5.3. Numerical results from lattice simulations.
Gravity coupled to matter: Ising model and
3state Potts model 160
6. Fractal geometry and spin systems 161
6.1. Critical behavior of spin systems as
a function of noninteger dimension of
space}time: Ising model 162
6.2. Geometry of critical clusters: Ising model 169
6.3. Fractal geometry of topological excitations:
X> spin model 174
7. Concluding discussion and outlook 175
References 177
03701573/00/$ see front matter 2000 Elsevier Science B.V. All rights reserved.
PII: S 0 3 7 0  1 5 7 3 ( 9 9 ) 0 0 0 5 1  4
Abstract
The goal of this article is to review the role of fractal geometry in quantum physics. There are two aspects:
(a) The geometry of underlying space (space}time in relativistic systems) is fractal and one studies the
dynamics of the quantum system. Example: percolation. (b) The underlying space}time is regular, and fractal
geometry which shows up in particular observables is generated by the dynamics of the quantum system.
Example: Brownian motion (imaginary time quantum mechanics), zigzag paths of propagation in quantum
mechanics (Feynman's path integral). Historically, the "rst example of fractal geometry in quantum
mechanics was invoked by Feynman and Hibbs describing the selfsimilarity (fractal behavior) of paths
occurring in the path integral. We discuss the geometry of such paths. We present analytical as well as
numerical results, yielding Hausdor! dimension d
'
"2. Velocitydependent interactions (propagation in
a solid, Brueckner's theory of nuclear matter) allow for d
'
(2. Next, we consider quantum "eld theory. We
discuss the relation of selfsimilarity, the renormalization group equation, scaling laws and critical behavior,
also violation of scale invariance, like logarithmic scaling corrections in hadron structure functions. We
discuss the fractal geometry of paths of the path integral in "eld theory. We present numerical results for the
length of propagation and fractal dimension for the free fermion propagator which is relevant for the
geometry of quark propagation in QCD. Then we look at order parameters for the con"nement phase in
QCD. The fractal dimension of closed monopole current loops is such an order parameter. We discuss
properties of a fractal Wilson loop. We look at critical phenomena, in particular at critical exponents and its
relation to noninteger dimension of space}time by use of an underlying fractal geometry with the purpose to
determine lower or upper critical dimensions. As an example we consider the ;(1) model of lattice gauge
theory. As another topic we discuss fractal geometry and Hausdor! dimension of quantum gravity and also
for gravity coupled to matter, like to the Ising model or to the 3state Potts model. Finally, we study the role
that fractal geometry plays in spin physics, in particular for the purpose to describe critical clusters. 2000
Elsevier Science B.V. All rights reserved.
PACS: 03.65.!w; 03.70.#k; 04.60.!m; 05.45.Df
Keywords: Fractal geometry; Quantum mechanics; Quantum "eld theory; Quantum gravity; Spin systems
H. Kro( ger / Physics Reports 323 (2000) 81}181 83
Fig. 1. Construction of Koch's curve, (a) original generator, (b) 2nd step, (3) 3rd step.
1. Introduction to fractal geometry
The notion of fractal geometry has become quite popular in natural sciences in recent years.
Fractal geometry seems to describe such di!erent phenomena as the shape of clouds and rivers, the
mixture of liquids with di!erent viscosity and solubility (pouring cream into co!ee), the description
of turbulence, in biology to describe plant growth, in medical sciences to describe the shape of brain
tumors or lungs, in models of economy, or in literature to describe the frequency of occurrence of
letters and words.
Because many di!erent fractals occur in nature, it is useful to characterize them. Following
Mandelbrot [104] a fractal is de"ned as an object with two properties: (a) selfsimilarity and (b) its
fractal dimension being di!erent from its topological dimension. In physics, in the theory of phase
transitions (critical phenomena) there are two notions quite similar to the above, namely (a) scale
invariance at the critical point and (b) critical exponents.
Let us explain the notion of selfsimilarity and the fractal dimension with two examples of
fractals: (i) Koch's curve which is an idealized mathematical construction and (ii) the coast line of
England. The construction of Koch's curve is shown in Fig. 1. Koch's curve can be viewed as the
limit of a sequence of curves. Curve (a) is called generator. Curve (b) is obtained by reducing
the generator by a scale factor of three and replacing each piece of straight line of curve (a) by the
reduced generator. Similarly curve (c) is obtained by reducing the generator (a) by a scale factor
84 H. Kro( ger / Physics Reports 323 (2000) 81}181
nine and replacing each piece of straight line of curve (b). This continues ad in"nitum. The curves
are selfsimilar, i.e., parts of each curve are identical (apart from the scale change) to the previous
curve. Now consider the length. Let us denote by a the length of the straight line part of the original
generator which has length l"4a. After scale reduction the straight line part has length
a/3, a(1/3)`,
2
, a(1/3)L and the reduced generator has length l"4a/3, l"4a(1/3)`,
2
, l"4a(1/3)L.
In the limit nPR this goes to zero. The lengths of the curves are: ¸"4a, ¸"4a(4/3),
¸"4a(4/3)`,
2
, ¸"4a(4/3)L in the nth step. In the limit nPRthe length goes to in"nity by
a power law.
As another example let us assume we want to measure the length of the coastal line of England.
One takes a yardstick, representing a straight line of a given length. Let c denote the ratio of the
yardstick length to a "xed unit length. The one walks around the coastline and measures the length
of the coast using the particular yardstick (starting a new step where the previous step leaves o!).
The number of steps multiplied with the yardstick length gives a value ¸(c) for the coastal length.
Then one repeats the same procedure with a smaller yardstick say c'. Doing this for many values of
c yields a function ¸ versus c. Let us assume there is a power law
¸(c)
&
C"
¸
"
c? . (1.1)
This looks very much like the critical behavior of a macroscopic observable at the critical point,
e.g., magnetization of a ferromagnet when temperature approaches the critical temperature. In that
case : would be called a critical exponent. One observes for a wide range of scales c that the length
of the British coast obeys such a power law. The fractal dimension d
''º°
is de"ned by
:"d
''º°
!d
'ºº
, (1.2)
where the topological dimension d
'ºº
"1 for the curve. For the coastline of England one "nds
experimentally d
''º°
+1.25. For Koch's curve, one can choose c as the ratio of the straight line parts
corresponding to nth and 0th step, c"(1/3)L. This satis"es Eq. (1.1), yielding :"(log 4!
log 3)/log 3 and thus d
''º°
"log 4/log 3"1.26
2
.
Why and where does fractal geometry play a role in quantum physics? In the case of non
relativistic quantum mechanics the most obvious example is the geometry of propagation (i.e., of
a typical path) of a massive particle. Under very general assumptions this is a curve of fractal
dimension d
'
"2. The underlying physical reason is Heisenberg's uncertainty principle, which by
itself is closely related to the commutation relation of position and momentum. A heuristic
argument, leading to d
'
"2 goes like this: writing the uncertainty relation
AxAp& , (1.3)
and putting
Ap"m(Ax/At) , (1.4)
we obtain
(Ax)`&At . (1.5)
H. Kro( ger / Physics Reports 323 (2000) 81}181 85
When we consider the propagation between two given points in space and time, where the time
interval ¹ is broken into N subintervals of length At, we "nd for the length of the path
¸"NAx"
¹
At
Ax&
¹
Ax
&Ax¹B
'
. (1.6)
Letting AtP0, and comparing with Eqs. (1.1) and (1.2), implies d
'
"2. This can be made more
rigorous. The result is also supported by numerical simulations.
Another example for the usefulness of fractal geometry in quantum mechanics is percolation, in
particular, the SchroK dinger equation has been studied on a Sierpinski gasket.
If fractal geometry plays a role in nonrelativistic quantum mechanics, one expects it to play
a role in quantum"eld theory, too. For example, one can ask the question: What is the geometry of
propagation for a relativistic quantum particle? Because particle number #uctuates in relativistic
physics, the answer is not as simple as in the nonrelativistic case. In fact, there is no unambiguous
answer. Quantum "eld theories describe manybody systems in nature. Those often display phase
transitions. At the critical point, we know that scaling laws describe the singular behavior of
macroscopic observables. This behavior is related to the mathematical property of selfsimilarity,
which is inherent in the renormalization group equation. On the other hand, selfsimilarity is
a characteristic feature of a fractal. Thus it is not surprising that fractal geometry plays a role for
critical phenomena. E.g., the geometry of critical clusters in certain spin systems (Ising clusters) can
be described by fractal geometry. Fractal geometry of randomwalks and random surfaces has been
studied in twodimensional models of statistical mechanics at the critical point.
In the context of lattice gauge theory, where the Wilson loop plays the role of an order parameter
distinguishing a con"ned from a decon"ned phase, it has been suggested that fractal geometry
plays a role at the decon"nement phase transition. New order parameters of gauge theories in
connection with fractal geometry have been suggested by Polikarpov and coworkers. The fractal
dimension of closed monopole current loops is such an order parameter. For example, the
monopole current loop becomes a fractal object at the transition point. The #ux lines of the Abelian
monopole current in the con"ning phase are found to have a fractal dimension D
'
'1 for ;(1) as
well as for Abelian projected "nite temperature S;(2), but have D
'
"1 in the noncon"ning phase.
Those "ndings by Polikarpov and coworkers strongly suggest that con"nement of quarks and
gluons has something to do with fractal geometry. Another area where fractal geometry has turned
out to be a useful tool, is the determination of upper or lower critical dimensions. Using lattices
which are fractal, it has been possible to simulate physical systems in noninteger space dimensions
and to study nonperturbatively critical behavior as a function of such noninteger dimension. This
has been done for the ;(1) model of lattice gauge theory and for the Ising spin model. For such
purpose fractal geometry is a nice and useful tool, complementary to the perturbative technique of
c expansion.
Finally, in quantum gravity, the dynamics of the systemis determined by curved space}time. It is
tempting to ask: Does curved space}time exhibit any particular features in terms of fractal
geometry? The answer is yes! We discuss, e.g., the relation between the Hausdor! dimension d
'
and
the critical exponent v, related by d
'
"1/v"4 in pure gravity. This establishes that the Hausdor!
dimension can play the role of a critical exponent. In particular, workers have studied the
Hausdor! dimension of topology of 2D quantum gravity as well as gravity coupled to matter, like
to the Ising model or to the 3state Potts model. The "nding that the Hausdor! dimension in
86 H. Kro( ger / Physics Reports 323 (2000) 81}181
Fig. 2. Random motion in a 2D square lattice, from a computer simulation. Figure taken from Ref. [149].
quantum gravity plays the role of a critical exponent (critical exponents determine the universality
class of the theory) suggests that the Hausdor! dimension might play such a role also in other "eld
theories. The problem is to "nd suitable observables.
Many of the questions asked and techniques used in the context of fractal geometry and
quantum "eld theory have been "rstly investigated for spin systems. In particular, the determina
tion of lower or upper critical dimensions, critical behavior as a function of noninteger dimension
of the embedding lattice and the geometry of critical clusters have been investigated "rst for spin
systems. We review those topics, paying attention to the Ising model and the X> model.
In this review, we will present concepts and applications of fractal geometry in quantum systems.
We will try to do this in a pedagogical manner. We hope to give an account of the developments of
recent years.
2. Fractal geometry in quantum mechanics
2.1. Brownian motion versus motion in quantum mechanics
2.1.1. Brownian motion
The subject of this reviewis fractal geometry in quantumphysics. In order to enter this subject let
us consider a system described by classical statistical mechanics: Brownian motion. It describes the
motion of a molecule in a liquid, undergoing collisions with other molecules. Each collision causes
the molecule to change its momentum. As a result one observes a sequence of erratic zigzag
movements (see Figs. 2 and 3). It turns out that physics is well described by making the following
assumptions: The change of momentum of the molecule at a collision depends only upon the last
collision. The change obeys a probabilistic law: It is a Gaussian stochastic process. Following
Itzykson and Drou!e [76] this can be nicely simulated numerically on a regular lattice in D space
dimensions (discretization Ax) and time (discretization At). After each time interval At, the molecule
hops from one lattice site to a neighbor site chosen at random (each neighbor assigned equal
probability). Then one can ask for the probability P(x
¹
, t
¹
; x
"
, t
"
)50 that the molecule arrives at
H. Kro( ger / Physics Reports 323 (2000) 81}181 87
Fig. 3. A trail of a Brownian path in 2D plane, from a computer simulation. Figure taken from Ref. [149].
site x
¹
at time t
¹
after having started from an initial position x
"
at time t
"
. This probability is
determined by three conditions:
(i) If t
¹
equals t
"
it does not move
P(x
¹
, t
"
; x
"
, t
"
)"o
V
¹
V
"
. (2.1)
(ii) The probability is normalized
V
¹
P(x
¹
, t
¹
; x
"
, t
"
)"1 . (2.2)
(iii) At successive times, it can arrive only from neighbor sites
P(x, t#1; x
"
, t
"
)"
1
2d
V¯ "°'¨"'º' '''°' º' V
P(x', t; x
"
, t
"
) . (2.3)
Using the discretized (lattice) Laplace operator
A
'º''
f (x)"
1
2d
B
I¯¹
[ f (x#e
I
)#f (x!e
I
)!2f (x)] , (2.4)
the latter equation can be written as
P(x, t#1; x
"
, t
"
)!P(x, t; x
"
, t
"
)"A
'º''
P(x, t; x
"
, t
"
) . (2.5)
In order to make the transition to the continuum limit, we generalize by allowing time steps a
"
and
lattice steps of length a, by doing the substitution tPt/a
"
and xPx/a. Then p"P/aB de"nes
a probability density. The continuum limit a
"
P0 and aP0 exists and is well de"ned under the
88 H. Kro( ger / Physics Reports 323 (2000) 81}181
condition that the time scale and the spatial scale are related by a
"
Ja`. Choosing in particular
a
"
"(1/2d)a` , (2.6)
the previous Eqs. (2.1), (2.2) and (2.5) yield for the probability density in the continuum limit
lim
R
¹
R
"
p(x
¹
, t
¹
; x
"
, t
"
)"oB(x
¹
!x
"
) , (2.7)
,
dBx p(x, t; x
"
, t
"
)"1 , (2.8)
R
Rt
!
p(x, t; x
"
, t
"
)"0 . (2.9)
Eq. (2.9) is the di!usion equation (with unit di!usion constant). The solution is
p(x, t; x
"
, t
"
)"
1
[4(t
¹
!t
"
)]B`
exp
_
!
(x
¹
!x
"
)`
4(t
¹
!t
"
)
, (2.10)
which is essentially a Gaussian. From the solution Eq. (2.10) one obtains for t
`
't
¹
't
"
the
Kolmogorov equation
,
dBx
¹
p(x
`
, t
`
; x
¹
, t
¹
) p(x
¹
, t
¹
; x
"
, t
"
)"p(x
`
, t
`
; x
"
, t
"
) . (2.11)
This in turn allows to express the probability density p in terms of a multiple integral
p(x
D
, t
D
; x
G
, t
G
)"
,
L¹
_
H¯¹
dBx
H
[4(t
H>¹
!t
H
)]B`
1
[4(t
¹
!t
"
)]B`
exp
_
!
1
4
L¹
H¯"
(x
H>¹
!x
H
)`
t
H>¹
!t
H
, (2.12)
where x
G
"x
"
and x
D
"x
L
. This gives a precise meaning to the path integral expression
p(x
D
, t
D
; x
G
, t
G
)"
,
[dx(t)]exp
_
!
1
4,
R
D
R
G
dt x `
, (2.13)
where x(t
G
)"x
G
and x(t
D
)"x
D
.
From Eq. (2.13) one can obtain a simple plausibility argument which shows that the average
path of the Brownian motion is a fractal curve with fractal dimension d
''º°'
"2. The main
contribution to the path integral comes from con"gurations where jR
D
R
G
dt x `&O(1) or
(Ax)`&At , (2.14)
where Ax means the increment of length of an average curve for a given time increment At. Suppose
that a total time interval ¹"n At is given. Then the length of an average curve, using Eq. (2.14), is
¸"n Ax"
¹
At
Ax"¹(Ax)¹ . (2.15)
Letting AtP0 hence AxP0 and comparing with Eq. (1.2) yields d
''º°'
"2.
H. Kro( ger / Physics Reports 323 (2000) 81}181 89
2.1.2. Relation between Brownian motion and quantum mechanics
As has been shown by Nelson [112] and other authors (see, e.g., Ref. [149]), there is a close
relationship between Brownian motion and quantum mechanics. Brownian motion goes over into
free motion of a massive quantum mechanical particle, when replacing time t Cit and the di!usion
coe$cient d
°'''
C/2m. Then one has the following correspondence between the di!usion equation
Eq. (2.9) and the SchroK dinger equation
R
Rt
p"Ap C!
i
R
Rt
ç"!
`
2m
Aç"Hç . (2.16)
There is as well a correspondence between the path integral of the probability density of Brownian
motion, Eq. (2.13), and the path integral for the transition amplitude of quantum mechanics,
p(x
D
, t
D
; x
G
, t
G
)"
,
[dx(t)]exp
_
!
1
4,
R
D
R
G
dt x `
C¸x
D
, t
D
¦ x
G
, t
G
,"
,
[dx(t)]exp
_
i
,
R
D
R
G
dt
m
2
x `
.
(2.17)
2.2. Path integral quantization and fractal geometry of quantum paths
2.2.1. Analytical results for free motion and the harmonic oscillator
While particles in classical mechanics follow smooth (di!erentiable) trajectories, the situation is
di!erent in quantummechanics. As has been noted by Feynman and Hibbs [56], paths of a massive
particle in quantum mechanics are nondi!erentiable, selfsimilar curves, i.e., zigzag curves (see
Fig. 4). Feynman and Hibbs have noticed in 1965 the property of (stochastic) selfsimilarity, which
plays an eminent role in many areas of modern physics. A decade later, Mandelbrot [104] has
introduced the concept of fractal geometry in nature. So what do we know about the fractal
geometry of quantum mechanical paths?
Due to the close relation between Brownian motion and quantum mechanics, one is tempted to
guess: A typical path of a free massive particle in quantum mechanics is a fractal curve with
d
''º°'
"2. The plausibility argument employed for Brownian motion can be carried over directly to
the path integral of quantum mechanics, thus supporting the guess. Actually, Abbot and Wise [1]
have shown that an average quantum mechanical path of free motion has fractal geometry with
Hausdor! dimension d
'
"2. This result is based on the de"nition of length of monitored paths.
Monitoring a path means to measure the position of a wave packet with some uncertainty in
localization Ax at some discrete times t
¹
, t
`
, t
`
,
2
, t
,
. Experimentally this can be done in the
following way: An electron is emitted from a source and passes through a sequence of screens each
of them carrying several holes (see Fig. 5). In order to determine by which hole the electron has
passed, one uses a source of light emitting photons placed behind each screen. The photon collides
eventually with the electron. Fromthe observation of the de#ected photon one can decide by which
hole the electron has passed. Suppose each hole has a size Ax. In order to determine a fractal
dimension from Eqs. (1.1) and (1.2) one needs to go to the limit Ax&cP0. In order to localize the
electron with uncertainty Ax, the wave length of the photon z"/p must obey z(Ax. Thus
eventually z must go to zero. But that means, according to Heisenberg's uncertainty relation
AxAp5 that the momentum uncertainty Ap of the electron in the plane of the screen would go to
90 H. Kro( ger / Physics Reports 323 (2000) 81}181
Fig. 4. Typical paths of a quantum mechanical particle are highly irregular on a "ne scale. Although a mean velocity can
be de"ned, no meansquare velocity exists at any point. Paths are nondi!erentiable. Figure taken from Ref. [56].
in"nity! This can be interpreted such that the electron is not free but by the measurement
undergoes interaction which in the limit AxP0 becomes totally erratic.
This experimental idea of monitoring the paths underlies Abbot and Wise's de"nition of length
of quantum paths. They construct a wave packet being localized at t
"
"0 at x
"
"0 with
uncertainty in position Ax. The wave packet is denoted by ç(x, t; x
"
"0, t
"
"0, Ax). They compute
the evolution of the wave function for a time interval At and measure the length ¸Al, by
¸Al,"
,
d`x¦x¦¦ç(x, At; x
"
"0, t
"
"0, Ax)¦` . (2.18)
After N measurements of position at time intervals At (¹"NAt), the length of the average path is
¸l,"N¸Al, . (2.19)
H. Kro( ger / Physics Reports 323 (2000) 81}181 91
Fig. 5. Experiment to determine by which hole a particle (electron) has travelled between source and detector. A light
source is placed behind each screen. From the observation of light scattered by the electron one can determine the hole
which the electron has passed. Figure taken from Ref. [56].
Their calculation gives
¸Al,J At/mAx (2.20)
and
¸l,J¹/mAx . (2.21)
According to Eqs. (1.1) and (1.2) this is equivalent to d
'
"2. In summary, Abbot and Wise's result
d
'
"2 for monitored paths can be interpreted such that the erratic paths are generated by
interaction via the position measurement.
Thus it is natural to ask the question: What is the geometry and in particular the Hausdor!
dimension of an unmonitored quantum mechanical path? There is indication that d
'
"2 holds
also. Feynman and Hibbs [56] have shown in 1965 that unmonitored quantum paths are
nondi!erentiable, stochastically selfsimilar curves. They have done a calculation which almost
proves d
'
"2. Their calculation includes the presence of any local potential. Moreover, their
calculation shows the close connection with Heisenberg's uncertainty principle. So let us recall here
the basic steps of Feynman and Hibbs' calculation. They consider the Hamiltonian
H"
!`
2m
#<(x) . (2.22)
92 H. Kro( ger / Physics Reports 323 (2000) 81}181
where < denotes a local potential. The quantum mechanical transition element from a state
¦x
'"
,t"0, to a state ¦x
''
, t"¹, is given by
¸x
''
, t"¹¦x
'"
,t"0,"
x
''
exp
_
!
i
H¹
x
'"
"
,
[dx(t)]exp
_
i
S[x(t)]
. (2.23)
The expression on the r.h.s. is the path integral, i.e., the sum over paths x(t) which start at x
'"
at
t"0 and arrive at x
''
at t"¹. The paths are `weighteda by a phase factor exp[iS[x(t)]/], where
S is the classical action corresponding to the above Hamiltonian for a given path,
S"
,
2
"
dt
m
2
x `!<(x(t)) . (2.24)
Analogously, the transition element of an operator F[x( ] is given by
¸F[x( ],"¸x
''
, t"¹¦F[x( ]¦x
'"
, t"0,"
,
[dx(t)]F[x(t)]exp
_
i
S[x(t)]
. (2.25)
Now suppose time is divided into small slices o, where x
G
,x(t
G
). This gives the action
S"
,¹
G¯¹
o
_
m
2
x
G>¹
!x
G
o
`
!<(x
G
)
. (2.26)
Feynman and Hibbs obtain the general relation
RF
Rx
I
"!
io
F
RS
Rx
I
. (2.27)
Putting F"x
I
, this yields
¸1,"
i
mx
I
x
I>¹
!x
I
o
!
x
I
!x
I¹
o
#ox
I
<'(x
I
)
. (2.28)
In the limit oP0, the potential term becomes negligible and one obtains
m
x
I>¹
!x
I
o
x
I
!
x
I
m
x
I
!x
I¹
o
"
i
¸1, . (2.29)
This means the transition element of a product of position and momentum depends on the order in
time of these two quantities. This leads to the usual operator commutation law between position
and momentum, which implies the Heisenberg uncertainty relation between position and mo
mentum. Now suppose one advances the second term on the l.h.s. of Eq. (2.29) by one time slice o
x
I
m
x
I
!x
I¹
o
"
x
I>¹
m
x
I>¹
!x
I
o
#O(o) . (2.30)
H. Kro( ger / Physics Reports 323 (2000) 81}181 93
Then in the limit At,oP0, Eqs. (2.29) and (2.30) imply
¸(x
I>¹
!x
I
)`,"!
o
im
¸1, ,
¸(Ax)`,J
m
At . (2.31)
This is Feynman and Hibbs' important result on the scaling relation between a time increment At
and the corresponding average length increment of a typical quantum path. Further one can prove
¸(Ax)`,"
2
¸¦Ax¦,` . (2.32)
Feynman and Hibbs scaling relation Eq. (2.31) and Eq. (2.32) imply
¸¦Ax¦,`JAt . (2.33)
Then de"ning the length in analogy to Eq. (2.19), one has
¸¸,"N¸¦Ax¦,"
¹
At
¸¦Ax¦,J
¹
¸¦Ax¦,
, (2.34)
and thus d
'
"2. In summary, Feynman and Hibbs calculation yields d
'
"2 for unmonitored
quantumpaths, in the presence of an arbitrary local potential. Moreover, it shows the close relation
with Heisenberg's uncertainty relation.
2.3. Numerical simulations for diwerent potentials
2.3.1. Free motion
There is further indication that d
'
"2 is the correct result for unmonitored paths. This comes
from numerical simulations. First let us specify the quantum mechanical length de"nition. We
consider a particle to travel from x
'"
at t
'"
to x
''
at t
''
. We divide the time interval into N slices such
that ¹"NAt. The length of the classical trajectory is given by
¸
°'º''
"
,¹
I¯"
¦x
I>¹
!x
I
¦ , (2.35)
where x
I
"x(t
I
), x
"
"x(t
"
)"x
'"
and x
,
"x(t
,
)"x
''
. According to Feynman and Hibbs [56] the
corresponding quantity in quantum mechanics is given by the transition element Eq. (2.25), where
the action is given by Eq. (2.26) and the observable F[x( ] is given by the classical length, Eq. (2.35),
thus
¸¸
DG
,"
,¹
I¯"
¦x
I>¹
!x
I
¦
"
,
d"x
¹
2d"x
,¹
,¹
I¯"
¦x
I>¹
!x
I
¦exp
_
i
S[x]
V
"
¯V
'"
V
,
¯V
''
. (2.36)
94 H. Kro( ger / Physics Reports 323 (2000) 81}181
One should note that this is not an expectation value in the usual sense. This expression is an
average of the observable length involving N time slices where average means summing over all
paths with weight factors given by the exponential action. Feynman and Hibbs refer to transition
elements as `weighted averagesa. The weighting function in quantum mechanics is a complex
function. Thus the transition element is in general complex.
KroK ger et al. [91] have computed this expression numerically, however, for imaginary time. The
latter has been done for the technical reason to be able to compute the path integral by Monte
Carlo simulation. Physically it means to do the Euclidean path integral. For a given initial position
x
'"
and "nal position x
''
and a "xed time interval t
''
!t
'"
"¹"NAt the length ¸¸
DG
, has been
computed from Eq. (2.36) and also the average increment of length ¸¦Ax¦, has been determined
from the path integral. In the continuum limit (AtP0, ¸¦Ax¦,P0) one expects critical behavior,
¸¸
DG
,&(At)&(¸¦Ax¦,)¹B
'
. (2.37)
For consistency it is useful to compute also the weighted average for the observable of length
squared
¸S
DG
,"
,¹
I¯"
(x
I>¹
!x
I
)`
. (2.38)
This can be computed analytically
¸S
DG
,"
D At
2m
2m
D
(x
D
!x
G
)`
¹
!2
#
D¹
m
. (2.39)
From the analytic expression one obtains the critical behavior
¸S
DG
,&(At)&(¸¦Ax¦,)! , (2.40)
with exponents B"0, C"0.
Numerical results for motion in D"3 dimensions are shown in Fig. 6. For the observable ¸S
DG
,
one "nds consistency with the exact result Eq. (2.39). The extracted critical exponents B"0.00037
$0.00025 and C"!0.00073$0.00051 are both compatible with zero. For the length ¸¸
DG
, one
observes a power law behavior with exponents A"0.49971$0.00052 and d
'
"1.9988$0.0021.
This is consistent with d
'
"2 and A"1/2 which implies the scaling lawEq. (2.33). In summary, the
numerical results obtained for quantummechanics in imaginary time are consistent with Hausdor!
dimension d
'
"2 for unmonitored paths of free motion and con"rm also the scaling law.
2.3.2. Local potentials: Coulomb potential and harmonic oscillator
For the case of unmonitored paths, the analytic calculation of Ref. [1] has been generalized by
CampesinoRomeo et al. [26] to include a harmonic oscillator potential, giving also d
'
"2. For
the case of monitored paths, there is Feynman and Hibbs [56] calculation which holds for
arbitrary local potentials. Thus, based on the scaling relation Eq. (2.32), Feynman and Hibbs result
implies d
'
"2 for any local potential. KroK ger et al. [91] have done Monte Carlo simulations for
the Euclidean path integral in the presence of local potentials like the harmonic oscillator and the
Coulomb potential. The numerical results are compatible with d
'
"2 (see Fig. 7). Thus there is
indication for d
'
"2 for both monitored and unmonitored paths in the presence of arbitrary local
potentials.
H. Kro( ger / Physics Reports 323 (2000) 81}181 95
Fig. 6. Average length and length squared for quantum paths versus number of time steps. (a) Length ¸¸, and (b) length
squared ¸S, for free motion, (c) length in presence of harmonic oscillator potential.
One might ask: What is the relationship between monitored and unmonitored paths and why
should d
'
coincide for both? First of all, the employed de"nition of length is di!erent. For
monitored paths, the authors of Refs. [1,26] have de"ned length ¸Al, as the usual quantum
mechanical expectation value of the absolute value of an increment of position. This corresponds to
a state in which the wave function has been evolved for an increment of time At from an original
wave function being characterized by localization uncertainty Ax (Eq. (2.19)). For unmonitored
paths one is interested in the length of the total path. This involves a number of intermediate times.
This leads to Feynman and Hibbs' transition element, which we has been employed in the length
de"nition given in Eq. (2.36). Thus the length de"nition and average are di!erent. However, there is
a common link, which may be considered as physical origin of fractal paths: Heisenberg's
uncertainty relation and behind this the fundamental commutator relation between position and
momentum. For the case of the monitored paths, the process of localization leads via the
uncertainty principle to erratic paths. But also for the case of unmonitored paths, Feynman and
Hibbs' calculation shows that the scaling relation between Ax and At, Eq. (2.31), is directly related
to the commutator relation between position and momentum, which again is directly related to the
uncertainty principle. But why then should the outcome of d
'
agree? One can speculate on this
coincidence. For unmonitored paths the scaling relation Eq. (2.31) is valid for a very large class of
96 H. Kro( ger / Physics Reports 323 (2000) 81}181
Fig. 7. Average length and length squared for quantum paths of velocity dependent action (:"1/2) versus number of
time steps. (a) Length ¸¸, and (b) length squared ¸S,.
interactions, namely all local potentials. Also the numerical simulations in Ref. [91] have given,
within statistical errors, d
'
"2 for all investigated local potentials. In other words there is
indication that d
'
"2 holds for unmonitored paths in the presence of arbitrary local potentials. On
the other hand, monitoring a path means interaction by measurement. If one assumes that such
interaction is represented by a local potential it is plausible that d
'
coincides for monitored and
unmonitored paths.
2.3.3. Velocitydependent potentials in condensed matter and nuclear physics
We have seen above that a local potential which in#uences the dynamics of the quantum system,
like the spectrum or scattering phases, does not in#uence the Hausdor! dimension. If we identify
the Hausdor! dimension with a critical exponent, and use the language of critical phenomena one
can state this as follows: All local potentials fall into the same universality class as the free system.
This leads to the question: Are there interactions which do not fall into this class? Let us consider
velocitydependent interactions. For instance, the action for a massive charged particle interacting
with an electromagnetic "eld
S"
,
dt
m
2
x `!q#
q
c
A) * (2.41)
H. Kro( ger / Physics Reports 323 (2000) 81}181 97
contains a velocitydependent term. An example from nuclear physics is Brueckner's [24] theory of
nuclear matter using a momentumdependent potential of the form <(k)"<
"
#<
`
k`#<
"
k".
A velocity dependence occurs also in condensed matter physics for electrons moving in solids,
where the velocity dependence enters via dispersion relations (see, e.g., Ref. [101]). One should
note, however, that a nonGaussian momentum dependence in the Hamiltonian does not corres
pond to the same functional form in the action (due to the Legendre transformation and expressing
velocity q by momenta p). E.g., a polynomial momentum dependence, like p" in the Hamiltonian
corresponds to a velocity dependence (q )`" in the action. This leads to a scaling behavior Ax vs. At
di!erent from the scaling relation of free motion (Ax)`JAt. In order to keep matters simple, we
choose the following parametrization of the action
S"
,
dt
1
2
mx `#<
"
¦x ¦?#;(x) . (2.42)
Here ;denotes a local potential. We have seen above that the presence of a local potential does not
change the fractal dimension coming from the kinetic term, so we drop the ;term in what follows.
Let us estimate the Hausdor! dimension of the velocity dependent potential S"jdt <
"
¦x ¦? in the
action. The dominant contributions to the path integral come from paths with
At
Ax
At
?
&O(1) . (2.43)
Thus
AxJ(At)'?¹'? . (2.44)
Feynman and Hibbs' calculation of the quantum mechanical scaling law, which for free motion and
motion under local potentials lead to Eq. (2.31), can be generalized to a velocity dependent action
S"
,
dt <
"
x " . (2.45)
It yields
¸(Ax)",J
4<
"
(At)` . (2.46)
It agrees with the estimated scaling law for :"4 given by Eq. (2.44). This implies the following
estimated scaling laws for length and length squared,
¸¸
DG
,J(At)¹?J(Ax)¹'?¹' ,
¸S
DG
,J(At)'?`'?J(Ax)'?`''?¹' .
(2.47)
From this one reads o! the following estimate of the critical exponents:
A"
1
:
, d
'
"
:
1!:
, B"
2!:
:
, C"
2!:
:
. (2.48)
98 H. Kro( ger / Physics Reports 323 (2000) 81}181
Fig. 8. Critical exponents d
'
and C versus potential parameter : for velocity dependent action. The solid line
corresponds to the estimated exponents.
For :"2, the velocitydependent term is equivalent to the kinetic term. In this case the estimated
Hausdor! dimension of the velocitydependent term agrees with the Hausdor! dimension d
'
"2
of the kinetic term.
Numerical simulations (in imaginary time) with the velocity dependent action, Eq. (2.42) have
given the following results [91], presented in Figs. 7 and 8. Fig. 7 shows ¸¸, and ¸S, as function of
the number of time steps (¹"N
R
At). The full lines are "ts using the estimated powerlaw behavior,
Eq. (2.47). For a velocity dependent term corresponding to :"1/2, one observes the critical
exponents A"0.5012$0.0023 and d
'
"2.005$0.009 as well as B"0.0031$0.0079 and
C"!0.0062$0.0159. This is compatible with A"1/2, d
'
"2 and B"C"0. This is not in
agreement with Eq. (2.48) for :"1/2, but it is for :"2. The explanation is that the action has two
velocitydependent terms, the kinetic term (:"2) and the <
"
term (:"1/2). Apparently the :"2
term being the larger one determines the critical exponent. Fig. 8 shows the critical exponents
d
'
and C corresponding to ¸¸, and ¸S,, respectively, as a function of the parameter :. The full
curve shows the estimated critical exponents, Eq. (2.48). For the interval 04:42, the kinetic term
is dominant, while for the domain 24: the <
"
term is dominant. There is a critical value :
°'''
"2.
The numerical data con"rm the estimated critical exponents.
H. Kro( ger / Physics Reports 323 (2000) 81}181 99
2.4. Can we measure experimentally the geometry of quantum mechanical propagation?
2.4.1. Problem with monitoring of paths
In quantum mechanics, position is an observable. The Hausdor! dimension is de"ned via the
length of a path. If one can measure position, one can measure the length of a path by measuring
the position at sequential times. Thus one should be able to obtain experimentally information on
the geometry of paths and in particular measure its Hausdor! dimension. It turns out that there is
a fundamental problem, which will be discussed below. First we ask: What is the physical
motivation to measure d
'
? There are the following reasons:
(a) Abbot and Wise's [1] analytic result giving d
'
"2 for free quantum mechanical motion does
not really correspond to free motion, but rather to an interacting system (see discussion below).
Hence measuring d
'
experimentally for free quantum motion is desirable.
(b) The fractal dimension is closely related to Heisenberg's uncertainty relation and to the
fundamental commutator relation [x( , p( ]"i. The commutator relation can not be measured, but
d
'
being a fundamental quantity in quantum mechanics and being de"ned via an observable
(length), it should be measured experimentally.
(c) For velocitydependent potentials, e.g., for electrons propagating in solids, one expects d
'
O2.
Measuring d
'
would give information on the action.
(d) One expects d
'
O2 in accelerated systems in non#at spacetime, e.g., in the neighborhood of
rotating double stars. A measurement of d
'
in such systems would be very interesting because it
involves quantum mechanics and relativity.
(e) Fractal geometry and selfsimilarity are connected. Selfsimilarity plays a crucial role in
modern physics, e.g., in deep inelastics leptonhadron scattering, Bjorken scaling in the parton
model, quark distribution, structure and splitting functions in the Altarelli}Parisi equation (see
Section 4.2). The quark structure functions F
¹
(x
, Q`), F
`
(x
, Q`) tell us how many quarks with
momentum fraction x
are observed in a hadron (proton) at a scale Q. Here Q` which is the
fourmomentum squared of the exchanged photon can be interpreted as resolution (wavelength).
The structure function depends on this scale Q (logarithmic corrections to Bjorken scaling in QCD)
which has been predicted from perturbative QCD and has been observed experimentally. Thus
measuring the scale dependence and hence d
'
in quantum mechanical propagation would a simple
analogue to measuring the Qdependence of structure functions.
The fundamental problem to measure experimentally d
'
is the following: Abbot and Wise [1]
have suggested to measure the length ¸¸, of quantum paths by sequential measurement of the
position of a particle at regular time intervals At, i.e., via monitoring the path. Here ¸¸, means to
take the average of the observable length according to the rules of quantum mechanics (see
Eq. (2.36)). Let us recall how to measure position in quantum mechanics. One can monitor
a particle being emitted from a source at position x
'"
at time t
'"
to arrive at the detector at position
x
''
at time t
''
by measuring its position at intermediate times t
'
at regular intervals At (see Fig. 5).
This can be done experimentally as described in Ref. [56]. E.g., an electron travelling from source
to detector passes by a number of screens with holes. In order to determine by which hole the
electron has passed the experimentor places a source of light behind each screen emitting photons
parallel to the screen. Eventually, the photon collides with an electron having passed through
100 H. Kro( ger / Physics Reports 323 (2000) 81}181
a hole. From detection of the scattered photon one can determine by which hole the electron has
passed. Thus one determines the length of path by joining the source to the detector by the
experimentally identi"ed holes. This gives a length ¸(Ax) as a function of the resolution of length
Ax, being given by the size of holes and distance between the screens.
According to Eqs. (1.1) and (1.2), the Hausdor! dimension is given by a power law behavior of
the length ¸¸(Ax), in the limit when the spatial resolution Ax goes to zero. In order to extract the
Hausdor! dimension, one needs to send AxP0, which means to decrease the size of holes and the
distance between screens by increasing the number of both. There is no problem with this
procedure when measuring d
'
for a classical system, e.g., the British coast line. However, there is
a fundamental problem with this procedure in quantum mechanics, due to Heisenberg's uncertain
ty relation. Measuring the position proceeds via interaction, e.g., scattering light from the electron
as discussed in Ref. [56]. In order to localize with uncertainty Ax by which hole the electron has
passed, one needs photon wave length z(Ax. Thus by the very measurement of position the
electron interacts with the photon and by collision has an uncertainty of momentum Ap5/Ax.
When going to the limit AxP0, the photon wavelength must go to zero and the uncertainty of the
electron momentum (in plane of screen) Ap goes to in"nity! Thus the path becomes increasingly
erratic. This can be interpreted by saying that monitoring the path creates the fractal (erratic) path.
Such an experiment does not measure the geometry of propagation of a free quantumparticle. This
is an unavoidable problem with any experiment trying to measure d
'
via localization, i.e., by
monitoring the path.
An alternative experiment has been suggested in Refs. [93,94] which avoids this dilemma and
allows to determine the Hausdor! dimension of a free particle without monitoring the path (see
Fig. 10). The idea is to use the concept of topology of paths. In the experiment described below one
measures the interference pattern of the crosssection, and deduces the contributions of homotopy
classes. In each homotopy class the interaction with the vector potential of the magnetic "eld is
analytically known. Thus one can `reconstructa the noninteracting case. Schulman [153] has
noted the topological character of paths in the Aharonov}Bohm e!ect. In the Aharonov}Bohm
experiment an electron is scattered from an (idealized) in"nitely thin and long magnetic #ux tube
(solenoid). The proposed generalized Aharonov}Bohm experiment consists of an array of such #ux
tubes. The spatial resolution Ax is given in this experiment by the distance between neighbor
solenoids. The determination of d
'
requires AxP0 hence many #ux tubes need to be placed
between source and detector. The array of #ux tubes allows to classify paths by its topology. All
paths (going fromx
'"
to x
''
) can be classi"ed by homotopy classes, given by the number and sense of
winding around each of the #ux tubes of the array. An electron propagating through the array of
#ux tubes interacts with the vector potential of the static magnetic "eld. However, for any path
within a given homotopy class the corresponding electromagnetic interaction is a constant, which
is analytically computable. The problem is to "nd the relative weight of each homotopy class
contributing to the propagator.
2.4.2. Aharonov}Bohm propagator
In order to understand the proposed experiment let us review the Aharonov}Bohm experiment
with a single #ux tube and the corresponding calculation of the quantum mechanical propagator
(see Fig. 10, but with only one #ux tube). For the case of the Aharonov}Bohm experiment with
a single #ux tube, the corresponding homotopy classes are simple and the quantum mechanical
H. Kro( ger / Physics Reports 323 (2000) 81}181 101
propagator in 2D (plane perpendicular to #ux) can be computed analytically [171]. We consider
a charged particle (charge q) passing by (scattering from) the solenoid (magnetic #ux ). Classically,
the Lorentz force is zero. The gauge of the vector potential can be chosen such that the vector
potential takes the form A
P
"0, A
F
"/2r. The classical Hamiltonian is given by
H"
1
2j
p!
q
c
A
`
, (2.49)
and the action is given by
S"
,
dt
j
2
x `#
q
c
x ) A(x, t) . (2.50)
When considering quantization by the path integral, there occurs an Aharonov}Bohmphase factor
due to the vector potential present in the action,
exp
_
i
,
2
"
dt
q
c
x ) A
"exp
_
iq
c,
V
''
V
'"
dx ) A
"exp[i:(0'!0#2n
`
)] , (2.51)
when the path winds n
`
"0,$1,$2,
2
times around the solenoid, and :"q/2c. This factor
depends only upon the initial and "nal azimutal angle 0 and the number of windings, but otherwise
it is independent of the path. In other words, paths can be classi"ed by their winding number, they
fall into homotopy classes. The Aharonov}Bohm propagator (2D) has been computed by Wilczek
[171]. It can be decomposed into contributions corresponding to winding number n
`
, given in
spherical coordinates by
K^"
L
`
(r', 0'; r, 0)"
,
>`
`
dz exp[i(z#:)(0'!0#2n
`
)]
j
2i¹
exp
_
ij
2¹
(r'`#r`)
I
'H'
jrr'
i¹
,
(2.52)
where I
J
(z) is the modi"ed Bessel function. The free propagator K''°°
L
`
is given by K^"
L
`
at :"0. Note
that for each winding sector the Aharonov}Bohm propagator factorizes into the Bohm}Aharonov
phase and the free propagator,
K^"
L
`
(r', 0'; r, 0)"exp[i:(0'!0#2n
`
)]K''°°
L
`
(r', 0'; r, 0) . (2.53)
The total propagator (sum over all windings) is
K^"(r', 0'; r, 0)"
>`
K¯`
exp[im(0'!0)]
j
2i¹
exp
_
ij
2¹
(r'`#r`)
I
'K?'
jrr'
i¹
. (2.54)
When letting r', rPRthe di!erential crosssection is obtained which has been "rstly computed in
a di!ererent way by Aharonov and Bohm [2]. Considering r'"r to be large yields velocity
v"(r'#r)/¹, momentum p"jv and the de Broglie wave length z"2/p"¹/jr.
The original Aharonov}Bohm e!ect corresponding to one solenoid can be understood in terms
of the semiclassical propagator which is de"ned as the free propagator times the Aharonov}Bohm
102 H. Kro( ger / Physics Reports 323 (2000) 81}181
Fig. 9. Comparison of (a) semiclassical propagator with (b) Aharonov}Bohm propagator and (c) absolute value of real
part of di!erence between both. Dependence on distance h and on :. Parameters given in dimensionless units: ¸"2
(length of straight line between x
'"
and x
''
), ¹"10, j"1, "1. Corresponding de Broglie wave length z"10¬.
Crossing from quantum mechanical region to classical region at h"5.
phase factor corresponding to the classical path. (see discussion in Ref. [55]). This holds when the
distance h between the solenoid and the classical path of the electron (straight line between slit and
detector) is large compared to the de Broglie wave length z, i.e., h"Ax<z, which is the classical
region. In order to determine the Hausdor! dimension, Ax needs to be sent to zero. In this
limit h"Ax;z, which is the quantum mechanical region. A comparison of the Aharonov}
Bohm propagator Eq. (2.54) with the semiclassical propagator is shown in Fig. 9a}c. This
set of parameters corresponds to the de Broglie wave length z"10 and the crossing of the
quantum mechanical region to the classical region occurs at h"5. One observes that when
the distance h becomes large, the di!erence between the Aharonov}Bohm propagator and the
H. Kro( ger / Physics Reports 323 (2000) 81}181 103
Fig. 10. Setup of generalized Aharonov}Bohm experiment. There are N
1
solenoids positioned in a regular array
separated by distance Ax.
semiclassical propagator tends to zero. However, one observes a marked di!erence for small
distance h (hP0).
2.4.3. Generalized Aharonov}Bohm experiment
The Aharonov}Bohm e!ect in the presence of one solenoid has been measured via an electron
interference experiment. Let us consider now the following generalization: An array of N
`
solenoids
is positioned in a regular array with next neighbor distance Ax (Fig. 10). The array of solenoids is
placed such that the classical trajectory coming from slit A passes through this array, while the
classical trajectory coming from the slit B does not pass through this array. Like in the
Aharonov}Bohm experiment one measures the interference pattern, once when all solenoids
are turned o! and once when all solenoids are turned on. Any change in the interference pattern
is due to a change of wave function which traverses the array of solenoids. The values of
G
(magnetic #ux in solenoid i) are parameters to be chosen by the experimentalist (see below). The
detector measures a squared modulus of the wave function I"¦ç(x, t)¦` and one observes an
interference pattern.
What are the homotopy classes? The quantum mechanical wave function can be expressed in
terms of a path integral,
ç(x, t)"
,
[dy]exp
_
i
S[y]
x
R_
x
"
R
"
!
exp
_
i
S[C]
, (2.55)
where the sum `over historiesa goes over all paths C starting from the source at x
"
, t
"
and going to
the detector at x, t passing via either one of the two slits. Because the action given by Eq. (2.50) has
104 H. Kro( ger / Physics Reports 323 (2000) 81}181
Fig. 11. Examples of topologically equivalent paths.
a free term and a magnetic term the wave function can be factorized
ç(x, t)"
!
exp
_
i
S
''°°
[C]
exp
_
iq
c,
!
dy ) A(y)
"
!
K''°°[C] exp
_
iq
c,
!
dy ) A(y)
. (2.56)
Quantummechanical paths propagate forward in time, but can go forward and backward in space.
In D52 dimensions paths can form loops. We have seen above that the Aharonov}Bohm
propagator in a sector of "xed winding number is given by the free propagator (:"0) in this
winding sector times the Aharonov}Bohm phase factor. This Aharonov}Bohm phase factor is the
same for all those paths which can be mapped onto each other by stretching and deformation
without crossing the solenoid. The winding number n
`
is a topological quantum number which
characterizes the paths. The full Aharonov}Bohm propagator is given by the sum over all winding
sectors. All this carries over to the generalized Aharonov}Bohm experiment with an array of
N
`
solenoids. The full propagator decomposes into homotopy classes. In each homotopy class the
propagator factorizes into the free propagator in this homotopy class and a generalized
Aharonov}Bohm phase factor, given in analogy to Eq. (2.51) by
exp
_
iq
2c
[(0'!0)
'º'
#2[n
¹
¹
#2#n
,
`
,
`
]
,
'º'
"
¹
#2#
,
`
. (2.57)
The topologically di!erent (homotopy) classes are characterized by the winding numbers n
¹
,
2
,n
,
`
,
with n
G
"0,$1,$2,
2
. Because Maxwell's theory is an Abelian gauge theory, homotopy classes
do not depend on the sequential order of winding around individual solenoids. Equivalent paths
with the same winding, but di!erent sequential order are shown in Fig. 11a and b.
The decomposition property of the propagator for a "xed homotopy class has the following
implication being important for the experiment: Changing the magnetic #ux in the solenoid P'
H. Kro( ger / Physics Reports 323 (2000) 81}181 105
and hence :P:', changes the Aharonov}Bohm phase factor in each homotopy class and hence the
total Aharonov}Bohm propagator. But it does not change the free propagator in each homotopy
class. Thus experimentally, one has a handle to measure the free propagator corresponding to
a given homotopy class. We introduce a cuto! in the winding numbers n
G
(n
°"'º''
. This is based on
the assumption that winding numbers beyond the cuto! give contributions to the amplitude which
are in the order of experimental errors and hence cannot be detected. This cuto! makes the number
of homotopy classes "nite. Let us enumerate the homotopy classes by h"1, 2,
2
, N
'
. The
experimentalist chooses a set of #uxes of the solenoids: '¹'
G
, i"1,
2
, N
`
, and measures the
corresponding interference pattern, say I'¹'. Then the experimentalist chooses another set of #uxes
of the solenoids, '`'
G
, i"1,
2
, N
`
, and measures again the interference pattern, I'`'. This is
repeated for N
'
di!erents sets of #uxes. The information obtained is then su$cient to determine the
free propagators in the homotopy classes h"1,
2
, N
'
. Substituting the phase factor, Eq. (2.57),
into the wave function, Eq. (2.56), yields the intensity for N
'
di!erent sets of #uxes,
I'''"
F
K''°°
F
exp
_
iq
2c
[(0'!0)
'º'
#2(n
¹
'''
¹
#2#n
,
`
'''
,
`
)]
`
, f"1,
2
, N
'
. (2.58)
Because a given set of #uxes and a given homotopy class h determines the generalized Aharonov}
Bohm phase factor, and the free propagator in each homotopy class K''°°
F
is independent from the
#uxes, this equation allows to determine the unknown coe$cients K''°°
F
for h"1,
2
, N
'
. Because
K''°°
F
are complex numbers, and vector potentials A and #uxes and are real, we need at least twice
as many sets of #uxes as the number N
'
of homotopy classes considered, N
'
'2N
'
.
How can we determine the length of paths and its Hausdor! dimension? Suppose we have
performed the above experiment and we know the free propagator K''°°
F
for homotopy classes
h"1,
2
, N
'
. From that we can construct the length of an average quantum mechanical path in
the following way. Classically, one de"nes a length of a particle moving along a trajectory (from
x
'"
"x(t
'"
) to x
''
"x(t
''
)) by
¸[x(t
¹
), x(t
`
),
2
, x(t
,¹
)]"
,¹
I¯"
¦x(t
I>¹
)!x(t
I
)¦ (2.59)
and takes the limit AtP0 in the end. In analogy to the classical mechanics the de"nition of length
of trajectories in quantum mechanics also involves the position (observable) at di!erent times. In
quantum mechanics this requires to consider a transition amplitude from some initial state ¦ç
'"
, at
t"t
'"
to some "nal state ¦ç
''
, at t"t
''
. According to Feynman and Hibbs [58] the transition
element for any function F[x(t
¹
), x(t
`
),
2
, x(t
,¹
)] of position x at di!erent time steps t
¹
,
2
, t
,¹
is given by
¸FK,"
¸ç
''
(t
''
) ¦ FK[x(t
¹
),
2
,x(t
,¹
)] ¦ ç
'"
(t
'"
),
¸ç
''
(t
''
) ¦ ç
'"
(t
'"
),
"
j[Dx(t)]dx
''
dx
'"
çH
''
(x
''
)F[x(t
¹
),
2
,x(t
,¹
)]exp[(i/)S]ç
'"
(x
'"
)
j[Dx(t)]dx
''
dx
'"
çH
''
(x
''
)exp[(i/)S]ç
'"
(x
'"
)
. (2.60)
Feynman and Hibbs call this a weighted average. It can be interpreted as a sum over all paths of the
observable F multiplied with the weight of the exponential action. Note that although this has an
interpretation as path integral the matrix element is a quantum mechanical expression which can
106 H. Kro( ger / Physics Reports 323 (2000) 81}181
be de"ned via the SchroK dinger equation. Substituting F by the classical length, Eq. (2.59), and
choosing position eigenstates as initial and "nal states, one obtains (x
I
,x(t
I
))
¸¸K(At),"
,¹
I¯"
¦x
I>¹
!x
I
¦
"
jdx
¹
2dx
,¹
,¹
I¯"
¦x
I>¹
!x
I
¦exp[(i/)S[x
I
]]
jdx
¹
2dx
,¹
exp[(i/)S[x
I
]]
"
!
¸
!
exp[(i/)S[C]]
!
exp[(i/)S[C]]
. (2.61)
The last equation is a short hand notation. Note that each curve C corresponds to pieces of straight
line joining positions at adjacent times, i.e., x(t
I>¹
) with x(t
I
) for k"0,1,
2
,N. Note, however, that
this expression is not well de"ned in the limit AtP0. The average path is a fractal, hence its length
becomes in"nite! This is an example, where an in"nity occurs in the continuum limit of non
relativistic quantum mechanics. We need to introduce a regularization. A natural regularization of
the transition element expressed via the path integral is that given in Eq. (2.61) where At is kept
"nite. However, such regularization is not suitable for the proposed experiment because there is no
measument taken at regular time intervals At. On the other hand the experimentor has at his
disposal the spatial resolution Ax, i.e., the distance between neighbor solenoids. The resolution Ax
comes from an array of #ux tubes. We have seen in the previous sections that the path integral can
be decomposed into corresponding homotopy classes, counting the orientation and winding
number around each solenoid. Thus in analogy to the regularization of the path integral via "nite
temporal resolution At by Eq. (2.61), we de"ne a regularization via "nite spatial resolution Ax by
¸¸K(Ax),"
,
'
F¯¹
¸(h)exp[(i/)S[h]]
,
'
F¯¹
exp[(i/)S[h]]
, (2.62)
where h"1,
2
, N
'
denotes the homotopy classes, N
'
is the cuto! determined from experiment,
exp[(i/)S[h]]"K''°°
F
is the weight factor of the free action determined from the experiment and
¸(h) denotes the classical length of path in the homotopy class h. Note that de"ning the length of
path in a homotopy class is to some extent ambigous. For details see Ref. [93]. In analogy to the
regularization via At by Eq. (2.61), where ¸(C) is given by the classical length of pieces of straight
line, we de"ne here ¸(h) also by the classical length of a curve being an element of homotopy class h.
It starts at x
'"
and arrives at x
''
. It goes by pieces of straight lines always passing in the middle of
a pair of solenoids. Such regularization does not distinguish paths on a scale smaller than Ax. Thus
the length ¸¸K(Ax), is obtained by taking K''°°
F
for homotopy class h from the experiment, construct
¸(h) for homotopy class h from the array of #ux tubes and compute the sum according to Eq. (2.62).
This yields "nally ¸¸K(Ax), in absence of the vector potential, i.e., corresponding to free propaga
tion. Finally, in order to extract the Hausdor! dimension d
'
, one has to measure the length
¸¸K(Ax), for many values of Ax, look for a power law behavior when AxP0 and determine the
critical exponent and thus d
'
. As a consequence of the fact that this experiment is not sensitive to
the zigzagness parallel to the solenoids, we do not measure the length of the path but only its
projection onto the plane perpendicular to the solenoids, i.e., in D"2 dimensions. Nevertheless,
the length as such is physically not so interesting (it depends on Ax anyway). The physically
H. Kro( ger / Physics Reports 323 (2000) 81}181 107
important quantity is the critical exponent (Hausdor! dimension) which corresponds to taking the
limit AxP0. But the latter is expected to be the same in any number of space dimensions.
3. Quantum physics on fractal space}time
3.1. SchroK dinger and Dirac equation viewed from statistical mechanics
The starting point of this section is the observation that quantum mechanics, being today the
generally accepted theory of nonrelativistic microscopic physics, still lacks a complete satisfactory
interpretation. One example is the waveparticle duality. Recall that the wave function ç(x, t) being
a solution of the SchroK dinger equation yields the probability density ç(x, t)çM(x, t) to "nd the
particle at space}time point (x, t). This is a postulate, veri"ed experimentally. But an understanding
of its physical origin is missing. What one would like to have is a microscopic statistical mechanics
model for quantum mechanics. For example, in classical physics the process of di!usion can be
understood in terms of the microscopic statistical mechanics model of Einstein's Brownian motion.
In the quest of such a statistical mechanics model of quantum mechanics, the relation between
di!usion and quantum mechanics has been studied by FeH nyes [51], Nelson [113], Nottale [119],
El Naschie [46] and Nagasawa [110]. Ord [122,125,128] has shown that the SchroK dinger equation
can be obtained from the statistical mechanics of random walks. Ord considers a model, where
a particle moves in one space dimension on a regular space}time lattice (with spacing Ax and At). It
hops in a random manner from one lattice site to another, only forward in time but forward or
backward in space, such that the velocity
v"
¦x
G
!x
H
¦
t
G
!t
H
"const . (3.1)
Moreover, the particle's trajectory is labeled by four colors, say red, green, blue and yellow. With
each change of direction, the color changes, say from red to green, the next time from green to
yellow and so on in a cyclic fashion. For each color state one introduces a corresponding density,
say u
I
, k"1, 2, 3, 4. Conservation of probability in terms of densities yields a coupled set of
equations. Linear combination of the four densities shows that two of them are independent, say
¹
and
`
. Then one goes over to a continuum equation, by taking the limit
AtP0, AxP0,
(Ax)`
At
P2D . (3.2)
The latter implies that the geometry of the random walk is a fractal curve of Hausdor! dimension
d
'
"2 (see Eq. (2.6)). Finally, introducing
ç
>
"
i
2
¹
#
1
2
`
,
ç
"
!i
2
¹
#
1
2
`
,
(3.3)
108 H. Kro( ger / Physics Reports 323 (2000) 81}181
one obtains a twocomponent SchroK dinger equation
(x, t)"
°
'V
`
""R
(
"'"R
0
0 °
'V
`
""R
(
"'"R
1
(2
1
1
. (3.4)
Similar considerations have been put forward also for the relativistic Dirac equation. In order to
better understand the Dirac equation, Feynman has constructed a chessboard model [56,62]. Kac
[80] has developed a stochastic model based on a relativistic random walk which yields the Dirac
equation. Ord [126] has extended his treatment of the SchroK dinger equation also to the Dirac
equation. Another approach, leading in a transparent way to the particle}antiparticle interpreta
tion of the Dirac equation, has been suggested by Ord in Refs. [123,105,124]. This model, called
`spiral modela, considers stochastic trajectories, similar to those described above for the SchroK din
ger equation. There is one di!erence, however. The particle can go forward or backward in time.
When it goes backward it forms a loop or a `spirala. The oriention of such spiral occurs only in one
sense of circulation (anticlockwise). The backward motion corresponds to antiparticles. When the
spiral walks are constructed such that they evolve macroscopically forward in time, the model is
found to satisfy the Dirac equation. The spiral model can be interpreted by saying that it treats time
not as a parameter, but a macroscopic concept with a statistical relation to a microscopic sequence
of events. Concerning fractal geometry, one "nds that if the random walks occur on a typical length
scale Ax (distance between change of direction) which is much larger than the Compton wave
length z, then d
''º°
"2. In the opposite case, when Ax;z, holds d
''º°
"1.
3.2. The principle of scale invariance
Einstein's theory of special relativity (principle of relativity) says that the laws of physics should
keep its form under Lorentz transformations (rotations and boosts). Nottale [119] has proposed
that a more general principle, called `scale relativitya holds. It means a generalization of Einstein's
principle of relativity requiring in addition that the laws of physics should keep its form also under
scale transformations. Einstein's general relativity is formulated in terms of Riemannian geometry
of curved space}time. This is expressed mathematically in terms of continuous and di!erentiable
functions. Nottale's requirement of invariance under scale transformations is directly related to
fractal geometry, with the inherent property of selfsimilarity. Thus Nottale's principle requires that
the laws of physics are formulated on a fractal geometry of space}time. Immediately the following
problem shows up: The fractal geometry of space}time is not di!erentiable. Thus the principle of
relativity, formulated in terms of di!erential calculus, appears to be in con#ict with the requirement
of scale invariance. A way to treat this problem, further details and physical consequences can be
found in Refs. [117}119] and references therein.
3.3. Quantum physics on Cantor sets
While space}time is curved at large length scales (around black holes), it is essentially #at at the
scale of classical physics. One may ask: How does it look like at small scales? One might speculate
that quantum space}time is not #at, but may be similar to a 4dimensional analogue of a Menger
sponge, which is a fractal object [104]. This may occur at the length scale of the de Broglie wave
H. Kro( ger / Physics Reports 323 (2000) 81}181 109
length [45]. One may wonder what happens to Heisenberg's fundamental uncertainty principle,
usually expressed in terms of the multiplication operator X and the di!erential operator
P"
i
d
dx
,
when considering quantum mechanics on a fractal geometry. El Naschie [46] has derived Heisen
berg's uncertainty principle considering a Peano}Hilbert curve as underlying geometry. The
di!erences of views in the interpretation of quantum mechanics (particle versus wave) becomes
obvious, e.g., in Young's double slit experiment. El Naschie showed also that it is possible to
interpret Young's double slit experiment on Peanolike space}time. A generalization of the Peano
curve are Cantor sets. The geometry of Cantor sets is discussed, e.g. in Ref. [50]. El Naschie et al.
have been proposed in a series of papers [47}49,111,127] that the geometry of space}time should
be a manifold made up of Cantor sets. For more details on the construction see the previous
references.
4. Fractal geometry and quantum 5eld theory
4.1. Selfsimilarity and renormalization group equation
At several occasions we have noted the close analogy between the powerlaw behavior of length
with fractal dimension being determined by some exponent, as observed in quantum mechanics,
and critical behavior occuring in condensed matter physics or quantum "eld theory, where power
laws hold with characteristic critical exponents. While mathematically constructed objects, like
Koch's curve or Sierpinski's gasket possess exact selfsimilarity, the trajectories of quantum
mechanical propagation exhibit only stochastic selfsimilarity. On the other hand, physical systems
at a critical point with a secondorder phase transition also exhibit selfsimilarity: Consider a ferro
magnet at a temperature ¹'¹
°'''
(for simplicity let us consider only two spin orientations, up and
down, like in the Ising model). The spins with the same orientation form clusters. The average size
of those clusters is in the order of the correlation length ¸. But there are also smaller and larger
clusters. When the temperature ¹ approaches ¹
°'''
, clusters occur of all sizes. The correlation
length is in"nite. There is no characteristic length scale (¸) anymore. Thus the physics at the critical
point is said to be scaleinvariant. The clusters are stochastically selfsimilar.
Physical properties of a system at the critical point can be obtained from renormalization group
transformations, as suggested by Kadano! [81] and Wilson [173], based on the property of
scaling. Let us consider the renormalization group equation as suggested by StuK ckelberg and
Petermann, GellMann and Low, Callan and Symanzyk (see Ref. [59] and references therein).
Consider an npoint proper Green's function 'L' in QCD,
'L'(p
G
, j, g)"lim
`
[Z
(g
"
, /j)L Z
$
(g
"
, /j)L$ 'L'
S
(p
G
, g
"
, )] , (4.1)
where n
and n
$
are the number of external gluons and quarks, respectively, j is a dimensionful
scale parameter, g
"
is the bare coupling constant, p
G
are the external momenta, Z
and Z
$
are
110 H. Kro( ger / Physics Reports 323 (2000) 81}181
wavefunction renormalization constants, and 'L'
S
is the unrenormalized proper Green's function.
Then holds the renormalization group equation, given by
_
j
R
Rj
#[(g)
R
Rg
#n
¸
(g)#n
$
¸
$
(g)
'L L$'(p
G
, j, g)"0 ,
[(g)"lim
`
_
j
R
Rj
g(g
"
, /j)
,
¸
G
(g)"!lim
`
_
j
R
Rj
lnZ
G
(g
"
, /j)
.
(4.2)
This is a di!erential equation following from the property that the unrenormalized Green's
function is independent from the scale j. A particular solution of the di!erential equation requires
to specify boundary (initial) conditions. Here [(g) is called the [ function and ¸
G
are called the
anomalous dimensions. For example, it determines the scaling exponent of 'LL$' under a scale
transformation of momenta. Suppose one wishes to keep j "xed but change the momenta p
G
,
p
G
Pzp
G
. (4.3)
De"ning t"lnz and denoting by d the naive (classical) dimension one "nds
_
j
R
Rj
#
R
Rt
!d
'L L$'(zp
G
, j, g)"0 . (4.4)
From Eqs. (4.2) and (4.4) one obtains
_
R
Rt
#[(g)
R
Rg
#¸(g)
'L L$'(zp
G
, j, g)"0 ,
¸(g)"d#n
¸
(g)#n
$
¸
$
(g) .
(4.5)
This means that under scale change of momenta, the vertex function behaves as
'LL$'&
H`
zB>LA>L$A$"zA , (4.6)
The scaling exponent, which one would expect naively, d, is modi"ed due to quantum corrections
by the term n
¸
#n
$
¸
$
, hence ¸
G
are called anomalous dimensions. The anomalous dimensions
have been used to describe the fractal dimension in quantum gravity (see Section 5.2).
It has been observed by Shirkov [156] that the renormalization group equation has the property
of functional selfsimilarity. The running coupling g obeys
g (x, y, g)"g (x/¸, y/¸, g (¸, y, g)) , (4.7)
where x and y are momentum and mass, respectively and ¸ is a scale parameter. This equation,
together with the condition g (1, y, g)"g, implies that g is unchanged under the selfsimilar trans
formation
xPx/¸, yPy/¸, gPg (¸, y, g) . (4.8)
H. Kro( ger / Physics Reports 323 (2000) 81}181 111
Selfsimilarity properties of the renormalization group can be used advantageously to construct
an improved approximate solution from a poor approximate solution. This has been elaborated by
Yukalov [175}177]. Gluzman and Yukalov [64] have applied it to resum asymptotic series in the
theory of critical phenomena. For example, they have computed the critical temperature of the
Ising model in D"2 and D"3 dimensions, starting from a few terms of the cumulant expansion.
Another example is the computation of critical exponents of the Landau}Ginzburg model. For
more examples and details see [64]. The renormalization group has mathematically the structure
of a dynamical system. This is most apparent when writing the renormalization group equation in
the following form,
y(t#t', f, s)"y(t, y(t', f, s), s) , (4.9)
which is equivalent to the existence of a function v, such that
R
Rt
y(t, f, s)"v(y(t, f, s)) , (4.10)
and by integration
,
W'RDQ'
W'RDQ'
du
v(u)
"t'!t , (4.11)
In order to see the structure of a dynamical system, let us consider the classical equation of motion
of a massive particle in a constant gravitational "eld. The equation of motion reads in D"1,
z(t)"
_
z(t)"!¹
`
gt`#v
"
t#z
"
z (t)"!gt#v
"
. (4.12)
Writing the initial conditions z(t"0)"z
"
and z (t"0)"v
"
in the form
f"
_
z(t"0)
z (t"0)
(4.13)
allows to express the equation of motion as a function of the initial data f
z(t, f )"
_
!¹
`
gt`#z (t"0)t#z(t"0)
!gt#z (t"0)
, (4.14)
obeying the identity
z(t"0, f )"f . (4.15)
Then computing z(t', f ) from Eq. (4.14) and then replacing f in Eq. (4.14) by z(t', f ), yields
z(t#t', f )"z(t, z(t', f )) , (4.16)
which is selfsimilar. The latter equation has the following meaning: The solution of the equation of
motion at time t#t' corresponding to initial data f at time t"0 equals the solution of motion at
time t corresponding to initial data f ' being given by the evolution of initial data f to time t'.
112 H. Kro( ger / Physics Reports 323 (2000) 81}181
Fig. 12. Kinematics of lepton}hadron scattering. Figure taken from Ref. [77].
4.2. Selfsimilarity and scale dependence. Hadron structure functions in QCD
In quantum physics it happens that physical observables are dependent on the scale of
observation. Examples of scale dependence are: The running coupling `constanta in QCD or the
Q` dependence of structure functions in QCD. A running coupling constant means that the value of
the coupling depends on a dimensionful scale parameter. For instance, the value of the coupling
between the gluon "eld and the quark "eld :
/!"
(corresponding to the "ne structure `constanta : in
QED) at the scale of the Z particle, is given by
:
/!"
(m
8
)+0.125 . (4.17)
The behavior of the coupling constant is determined by the renormalization group equation. As
has been pointed out above this equation exhibits selfsimilarity. The second example, the
dependence of the proton structure function, e.g., on Q`, often denoted by logarithmic violation of
Bjorken scaling, are a fundamental test of the QCD standard model. These scaling violations are
closely related to some anomalous dimensions.
Let us start out by de"ning structure functions. We consider deep inelastic leptonhadron
(electron}proton) scattering. One considers only the leading order in perturbation theory. The
electron emits a photon which is absorbed in the nucleon. The kinematics are shown in Fig. 12. The
electron has initial and "nal four momentum k and k', respectively, and q"k!k' is the four
momentumtransfer due to virtual photon exchange. Q`"!q` is the square of photon momentum
transfer. M is the nucleon mass, v"pq/M is the energy transfer in the laboratory frame,
=`"(p#q)` is the invariant squared mass of the hadrons in the "nal state, x"Q`/(2Mv) is the
Bjorken variable, which runs from 0 to 1, and P"Q/(2x) is the nucleon momentum in the Breit
frame. The Breit frame is a particular reference frame, which allows a convenient interpretation
using the parton model. The Breit frame is de"ned by the requirement, that the electron has the
same energy before and after scattering, i.e., the virtual photon transfers no energy, but only
3momentum. Further the Breit frame can be chosen such that the 3momentum of the proton and
H. Kro( ger / Physics Reports 323 (2000) 81}181 113
the photon are antiparallel. Then
p"
(P`#M`
0
0
P
, q"
0
0
0
!Q
. (4.18)
In the Breit frame P"Q/(2x)PR, thus the nucleon is moving fast. In the parton model, suggested
by Feynman, one makes two assumptions, (a) the partons (quarks) travel approximately in the
same direction as the nucleon, (b) the partons move freely inside the nucleon and the elec
tron}nucleon scattering reaction is decribed (in the impulse approximation) by scattering of an
electron from a parton, and "nally taking an incoherent sum over all parton contributions. The
virtue of the Breit frame is that it allows to interpret the 3momentum transfer Q as a resolution
(wavelength) of a particle probing the structure of the nucleon, i.e. a scale parameter. Bjorken has
suggested to consider the limit
Q`PR, vPR, x"Q`/(2Mv)"const . (4.19)
The inclusive crosssection for electron}proton scattering can be written as
d`o
ddE'
"
:`
q"
E'
E
lIJ=
IJ
, (4.20)
where :"e`/(4) is the "ne structure constant and l
IJ
denotes the leptonic tensor,
l
IJ
"2(k
I
k'
J
#k
I
k
J
#¹
`
q`g
IJ
) . (4.21)
The hadronic tensor =
IJ
can be parametrized in terms of two Lorentz scalars, =
¹
and =
`
, via
=
IJ
(p, q)"
_
!=
¹
g
IJ
!
q
I
q
J
q`
#
=
`
M`
p
I
!
p ) q
q`
q
I
p
J
!
p ) q
q`
q
J
. (4.22)
=
¹
and =
`
are called structure functions. Bjorken predicted in the socalled Bjorken limit the
following behavior of the structure functions,
2M=
¹
(v, Q`)"F
¹
(x)#O(1/Q`) ,
v=
`
(v, Q`)"F
`
(x)#O(1/Q`) . (4.23)
That means, they should be a function of the Bjorken variable x alone and should become
independent of the momentum transfer and scale parameter Q`.
This phenomenon is called Bjorken scaling. Early experimental data are consistent with Bjorken
scaling, as can be seen in Fig. 13. However, later it became obvious, that for "nite Q` there is a small
Q dependence. This can be seen in Fig. 14, which shows Q dependence, particularly strong for small
x. This means that the Bjorken scaling behavior is not rigorously realized in nature. The scaling
symmetry is broken. This Q dependence, however, is a small e!ect. It can be computed in
perturbative QCD, which using the renormalization group, predicts logarithmic corrections. In
Fig. 15 nth moments of the structure function F
`
are shown in a comparison of experiment with
114 H. Kro( ger / Physics Reports 323 (2000) 81}181
Fig. 13. Early experimental data on Q` dependence of structure function v=
`
for x"0.3 from ep scattering at SLAC.
(v=
`
PF
`
in the Bjorken limit). Figure taken from Ref. [148].
Fig. 14. Deuteron structure function F
`
versus Q` at di!erent values of x (data multiplied with constant factors). The
data correspond to muon scattering at CERN and electron scattering at SLAC. The solid line is a QCD"t, taking into
account the theoretically predicted scaling violations and using as "t parameters the strong coupling constant and the
gluon distribution. Figure taken from Ref. [145].
H. Kro( ger / Physics Reports 323 (2000) 81}181 115
Fig. 15. nth moment ¸nF
`
,
L
raised to the power (!1/a
L
) corresponding to the nonsinglet structure function xFJ,
`
.
Perturbative QCD predicts a lnQ` dependence (logarithmic scaling violation). Excellent agreement between experimental
data and theoretical predictions is obtained. Figure taken from Ref. [59].
perturbative QCD predictions. The nth moment of a structure function F is de"ned by
¸F(Q`),
L
"
,
¹
"
dx xL`F(x, Q`) . (4.24)
The excellent agreement is strong support for the QCD standard model.
The scaling violations can be understood as a consequence of the fact that the e!ective strong
coupling constant :
Q
(t) does not vanish fast enough when Q`PR, where t"1/2 ln Q`/Q`
"
. This has
been "rstly suggested by Altarelli and Parisi [4,5]. An excellent presentation is given by Close [31].
Let q(x) denote the probability of a quark carrying the fraction x of nucleon momentum. The
virtual photon mass (Q`)¹` resolves nucleon structure at the length scale (Q`)¹`. Thus increasing
t to t#t increases q(x) to q(x)#q(x). This means that the quark distribution is t dependent. One
can de"ne a function P
OO
(z) which means the variation in the probability per unit t of "nding
a quark inside a quark with z"x/y fraction of its momentum y. Then the Q dependence of q(x) is
given by the Altarelli}Parisi (Gribov}Lipatov}Altarelli}Parisi) equation
dq(x, t)
dt
"g `(t)
,
¹
"
dy
,
¹
"
dz (x!yz)P
OO
(z)q(y, t)
"g `(t)
,
¹
"
dy
y
P
OO
(x/y)q(y, t) . (4.25)
116 H. Kro( ger / Physics Reports 323 (2000) 81}181
This equation becomes even simpler when writing moments of q(x), de"ned by
M'L'(t)"
,
¹
"
dxxLq(x, t) ,
D'L'"
,
¹
"
dz zLP
OO
(z) . (4.26)
Then the Altarelli}Parisi equation yields
dM'L'(t)
dt
"g `
,
¹
"
dxxL
,
¹
"
dy
y
P
OO
(x/y)q(y, t)
"g `
,
¹
"
dy yLq(y, t)
,
¹
"
dz zLP
OO
(z)
"g `M'L'(t)D'L' . (4.27)
The solution of this di!erential equation is
dM'L'(t)
M'L'(t)
"
D'L'
2b
dt
t
,
M'L'(t)
M'L'(t
"
)
"
_
:
Q
(t)
:
Q
(t
"
)
"
'L'
`@
.
(4.28)
This is precisely the QCD result with a logarithmic scaling violation, as displayed in Fig. 15.
4.3. Geometry of propagation of a relativistic particle
4.3.1. Problem of dexnition of fractal dimension in relativistic quantum xeld theory
The Hausdor! dimension of "eld con"gurations has recently become a subject of interest in
quantum gravity (in particular in 2D), where the Hausdor! dimension of the geometry of
space}time has been identi"ed as a critical exponent (see Section 5). Critical exponents of a theory
determine its universality class and thus classify the theory. The above example of quantum gravity
leads us to the question: Can the Hausdor! dimension play a similar role for models like, e.g.,
theory of matter interacting with radiation, in condensed matter, in the theory of medium energy
nuclear physics (Dirac phenomenology), or in high energy physics for the theory of quarks and
gluons (plasma)? A related point of view is the following: In quantum mechanics the Hausdor!
dimension of typical paths may di!er from the standard value d
'
"2 in the presence of velocity
dependent potentials [91], also if one considers quantum mechanics in a background medium
corresponding to curved space}time. Such a situation occurs, e.g., when considering the relativistic
propagation of a particle in nuclear matter (e.g., a neutron star), or when considering a relativistic
particle impinging on the surface and propagating a short distance in ordinary matter.
When trying to generalize the previous considerations from nonrelativistic quantum mechanics
to relativistic quantum mechanics, one is faced with the following problem: (a) In quantum
mechanics, the position of a particle is an observable and can be measured. In quantum"eld theory
H. Kro( ger / Physics Reports 323 (2000) 81}181 117
position is not an observable. (b) The particle number is conserved in quantum mechanics. In
quantum "eld theory particles can be created and annihilated. (c) In quantum mechanics particles
propagate only forward in time. In quantum "eld theory particles propagate forward and back
ward (antiparticles) in time. Thus in a relativistic theory one must look at space and time
dependence, corresponding to a causal propagation of a massive particle. The question raises: Does
it make sense to talk about the length of propagation of a particle, when the particle number can
change (virtual particle creation and annihilation)? From the mathematical point of view as we
have pointed out above, the path integral in imaginary time quantum mechanics is well de"ned in
terms of a stochastic process. This can be generalized to Euclidean path integrals of bosonic
(polynomial) "eld theory which are mathematically well de"ned, allowing an interpretation as
stochastic process [63,149]. The measure gives the dominant contributions of nowhere di!erenti
able curves [63]. However, for fermion "eld theory, there is not yet a strict mathematical form
ulation in terms of a stochastic process, although Osterwalder and Schrader [129] have established
a Feynman}Kac formula for fermion "elds (for some recent progress see Ref. [142]). From those
remarks it is evident that the de"nition of propagator length given by Eq. (2.36) for nonrelativistic
quantum mechanics cannot simply be taken over to relativistic quantum mechanics.
In order to overcome those di$culties and to "nd a physically meaningful de"nition of the
fractal dimension (Hausdor! dimension) two avenues can be pursued: (i) A length of propagation
can be introduced via the hopping parameter expansion, using a space}time lattice as regulator.
From its scaling behavior when aP0 (a being the lattice spacing) one can derive a critical
exponent, i.e., the fractal dimension. (ii) One can consider a dimensionful observable having, e.g.,
volume < as naive dimension. Field theoretic evaluation may give a dependence like <B. Again
from its scaling behavior one deduces a fractal dimension. It is related to the anomalous dimension.
This way of de"ning a fractal dimension has been employed in quantum gravity (see Section 5).
4.3.2. Length of propagation via the hopping parameter expansion
Introducing a length of propagation via the hopping parameter expansion has been suggested in
Ref. [92]. Let us consider in particular the onecomponent scalar "
`>¹
model. In order to de"ne
a fractal dimension we need to express the quantum observable in terms of an elementary length
scale. A suitable elementary length scale is, e.g., the lattice spacing a when the "eld theory is
regularized on a space}time lattice. In the notation of Ref. [109] the scalar model is given by the
Euclidean action
S[]"
,
d"x
1
2
[R
I
"
(x)R
I
"
(x)#m`
"
`
"
(x)]#
g
"
4!
"
"
(x) . (4.29)
After regularization on the lattice it is convenient to express the action as
S"
V
!2«
"
I¯¹
(x)(x#aj( )#`(x)#z[`(x)!1]`!z
, (4.30)
where the "eld and the parameters « and z are related to the bare "eld, bare mass and bare
coupling through
a
"
(x)"(2«)¹`(x), a`m`
"
"
1!2z
«
!8, g
"
"
6z
«`
. (4.31)
118 H. Kro( ger / Physics Reports 323 (2000) 81}181
Introducing the polynomial ;() and the hopping matrix M via
;()"z(`!1)`#`#z ,
M
GH
"
1
2
"
I¯¹
GH>I
#
GHI
,
(4.32)
the action becomes (where i, j denote the lattice sites and, e.g., in D"1 dimension
x
H
"ja, j"0,$1,$2,
2
)
S"!«
GH
G
M
GH
H
#
G
;(
G
) . (4.33)
The (Euclidean) twopoint function is given by
G(x
?
, x
@
)"¸
?
@
,"
1
Z
F(x
?
, x
@
) ,
F"
, _
_
G
d
G
?
@
exp[!S[]] ,
Z"
, _
_
G
d
G
exp[!S[]] . (4.34)
The idea of the hopping parameter expansion is that these integrals can be computed analytically
(for z50), after having made a Taylor expansion in terms of the hopping matrix M(or the hopping
parameter «), and then evaluating the resulting integrals order by order (for more details and
further references see Ref. [109]). We use the abbreviation M"
GH
G
M
GH
H
, and the Taylor
expansion
exp[«M]"1#«M#
1
2!
(«M)`#
1
3!
(«M)`#2 (4.35)
and we de"ne
¸
I
"
,
dI exp[!;()] . (4.36)
Thus one obtains for the partition function Z
Z"Z'"'#«Z'¹'#«`Z'`'#2 (4.37)
where the "rst coe$cients are
Z'"'"¸4
"
, Z'¹'"0, Z'`'"¹r[M`]¸`
`
¸4`
"
, Z'`'"0 . (4.38)
Here < denotes the lattice volume (number of lattice sites), D is the space}time dimension and
¹r[M`]"2D<. Similarly, one expands the numerator F of the twopoint function (for di!erent
lattice sites x
?
, x
@
)
F(x
?
, x
@
)"F'"'(x
?
, x
@
)#«F'¹'(x
?
, x
@
)#«`F'`'(x
?
, x
@
)#2 (4.39)
H. Kro( ger / Physics Reports 323 (2000) 81}181 119
Fig. 16. " model. Graphs from hopping expansion of twopoint function. Graph (d) gives a disconnected vacuum
contribution cancelled by the partition function Z.
with low order coe$cients
F'"'(x
?
, x
@
)"0 ,
F'¹'(x
?
, x
@
)"2M
?@
¸`
`
¸4`
"
,
F'`'(x
?
, x
@
)"4(M`)
?@
¸`
`
¸4`
"
,
F'`'(x
?
, x
@
)"`
`
(M`)
?@
¸"
`
¸4"
"
#"
`
[M
?@
]`¸`
"
¸4`
"
#8M
?@
(M`)
??
¸
"
¸`
`
¸4`
"
#2M
?@
¹r[M`]¸"
`
¸4"
"
, (4.40)
where (M`)
??
"2D. In third order the last term of F(x
?
, x
@
)'`' gives a disconnected vacuum
contributions, see Fig. 16. These vacuum contributions will be cancelled by the partition function
Z (for cancellation of vacuum terms see Ref. [11]). Expanding F(x
?
, x
@
) and 1/Z up to, e.g., third
order in «, one obtains for the twopoint function
G(x
?
, x
@
)"G'"'(x
?
, x
@
)#«G'¹'(x
?
, x
@
)#«`G'`'(x
?
, x
@
)#«`G'`'(x
?
, x
@
)#O(«") (4.41)
where
G'"'(x
?
, x
@
)"0 ,
G'¹'(x
?
, x
@
)"2M
?@
(¸
`
/¸
"
)` ,
G'`'(x
?
, x
@
)"4(M`)
?@
(¸
`
/¸
"
)` ,
G'`'(x
?
, x
@
)"`
`
(M`)
?@
(¸
`
/¸
"
)"#"
`
(M
?@
)`(¸
"
/¸
"
)`#8M
?@
(M`)
??
(¸
"
/¸
"
)(¸
`
/¸
"
)` , (4.42)
which is free of disconnected vacuum contributions. In general one considers the connected
twopoint function
G
°
(x
?
, x
@
)"¸
?
@
,
°
"¸
?
@
,!¸
?
,¸
@
, . (4.43)
which in the symmetric phase coincides with G(x
?
, x
@
).
As can be seen from the example of the twopoint function corresponding to lattice sites x
?
and
x
@
, the hopping expansion given by Eqs. (4.41) and (4.42) has contributions from any number of
hoppings, where order n corresponds to nfold hopping. From the properties of the hopping matrix
M follows, of course, that if lattice sites x
?
and x
@
are more than p lattice spacings apart, then
G
°
(x
?
, x
@
) has no contributions from any order lower than p. In Ref. [92] it has been suggested to
introduce a propagator length via the hopping expansion. To the term of order n, G'L'
°
(x
?
, x
@
), one
assigns a length ¸'L'"na G'L'
°
(x
?
, x
@
). The length of the propagator is de"ned as incoherent sum
120 H. Kro( ger / Physics Reports 323 (2000) 81}181
over all orders of hopping
¸(x
?
, x
@
)"
`
L¯"
na «LG'L'
°
(x
?
,x
@
) . (4.44)
This can be written nonperturbatively as
¸(x
?
, x
@
)"a«
d
d«
G
°
(x
?
, x
@
) . (4.45)
In Section 4.3.4 we will show for the case of the Dirac fermion that the length de"nition introduced
in this section gives a Hausdor! dimension, which is consistent with nonrelativistic quantum
mechanics. As will be shown in the next section, at the critical point of the theory there is a close
connection between the Hausdor! dimension and some critical exponent. The critical behavior is
due to the zeromomentum behavior. The zeromomentum projection of the twopoint function is
given by
GI(k"0)"
L
G(x
L
, x
"
) . (4.46)
Thus it is useful to introduce also a corresponding length de"nition
¸"
L
¸(x
L
, x
"
) , (4.47)
which, apart from a normalization factor 1/<, is the propagator length, Eq. (4.45), averaged over all
lattice sites.
4.3.3. Relation between Hausdorw dimension d
'
and the critical exponent ¸
Euclidean lattice "eld theory is closely related to statistical mechanics. In statistical mechanics,
the correlation function, i.e., the twopoint function is related to the magnetic susceptibility ,, via
,"
L
¸
L
"
,
°
. (4.48)
The sum over lattice sites n corresponds to a projection onto momentum p"0. The scalar model
has two phases, a symmetric and a nonsymmetric phase. There is a critical line in the space of
parameters « and z, where «
°'''
"«
°'''
(z). It separates the two phases and the phase transition is of
second order [99]. When approaching the critical line, physical observables often obey power laws
(see [21]), like
,&
GG°'''
¦«!«
°'''
¦A , (4.49)
where ¸ is a critical exponent. Another example is the physical mass m
0
, which behaves like
am
0
&
GG°'''
¦«!«
°'''
¦J . (4.50)
Critical exponents have been computed from perturbation theory by BreH zin et al. [23]. They "nd in
D"4 space}time dimensions that the critical behavior is modi"ed by logarithmic corrections.
H. Kro( ger / Physics Reports 323 (2000) 81}181 121
E.g., the mass behaves as
am
0
&
GG°'''
¦«!«
°'''
¦¹`¦ln¦«!«
°'''
¦¦¹" , (4.51)
where v"1/2 (this value without logarithmic correction is obtained also in mean "eld theory).
Now suppose that there are no logarithmic corrections, like in D(4. Then one can relate the
magnetic susceptibility to the propagator length. Taking the propagator length as de"ned in
Eq. (4.47), one has
¸"
L
¸(x
L
, x
"
)
"a«
d
d«
L
G
°
(x
L
, x
"
)
"a«
d
d«
,
&
GG°'''
a«
d
d«
¦«!«
°'''
¦A
&
GG°'''
a«(!¸)¦«!«
°'''
¦¹A
&
?"
«¹`A
°'''
a¹`A , (4.52)
where we have used Eq. (4.31) which implies 1/«!1/«
°'''
&a`, and ¦«!«
°'''
¦&«`
°'''
a`. Introducing
an arbitrary "xed (dimensionful) length l
"
and c"a/l
"
, one has
¸&
C"
c¹`A , (4.53)
which by comparison with Eq. (1.1) yields the following relation between d
''º°'
and the critical
exponent ¸,
d
''º°'
"2#2¸ . (4.54)
In statistical mechanics, relations between critical exponents of di!erent observables are called
scaling laws, e.g., scaling laws by Fisher, Widom, etc. If the above de"ned propagator length
¸ would be an observable, then Eq. (4.54) were a scaling law. However, ¸ is not a macroscopic
observable.
Though ¸ not being an observable, it is nevertheless interesting to establish relations between
critical exponents and the fractal dimension of any dynamical quantity, given by the particular
model. Other examples of this kind will be discussed in Section 4.5, where critical exponents will be
related to the fractal dimension of clusters formed at the critical point of a phase transition.
4.3.4. Length of propagation and Hausdorw dimension of the free Wilson}Dirac fermion on the lattice
Let us apply the above de"nition of length of propagation and Hausdor! dimension to a soluble
"eld theory to see if it makes sense. For that purpose one has chosen in Ref. [92] to consider the
noninteracting fermion being described on the lattice by the discretization of the Dirac action.
This lattice action is plagued by the socalled fermion species doubling problem [109]: 16 copies of
fermions with the same mass occur (as poles of the fermion propagator). A way out commonly used
122 H. Kro( ger / Physics Reports 323 (2000) 81}181
is the Wilson action, which lifts the species doubling by adding another term. The (Euclidean)
Wilson}Dirac action on the lattice reads [109]
S[ç, çM]"
KL
çM
K
K
KL
ç
L
. (4.55)
The "elds have been rescaled
a``(am#4r)¹`çPç, «"1/(2ma#8r) for D"4 . (4.56)
Note that the hopping parameter « is now di!erent fromthat of the scalar model, Eq. (4.31), and r is
the Wilson parameter usually chosen to be r"1. There is also a hopping matrix M coming from
the kinetic term of the fermion action and from the Wilson term, allowing the fermion to hop from
one lattice site to the nearest neighbour lattice sites
M
KL
"
"
I¯¹
(r#¸
I
)o
K>I
(
L
#(r!¸
I
)o
KI
(
L
. (4.57)
The matrix K is expressed in terms of the hopping matrix M
K
KL
"o
KL
!«M
KL
. (4.58)
The fermion propagator is given by
¸ç
L
, çM
K
,"(K¹)
LK
. (4.59)
In Ref. [92] we have de"ned the length of the fermion via
¸¸
K L
,"
« (d/d«)¸ç
L
çM
K
,
¸ç
L
çM
K
,
. (4.60)
The indices m, n denote the lattice sites, where the fermion is created and annihilated, respectively.
By expanding the righthand side of Eq. (4.60) in a power series in the hopping parameter «, one
"nds that the pth power of the hopping matrix M, which allows the fermion to hop a distance pa,
gets multiplied with a factor p. Thus ¸¸
KL
, can be interpreted like in the case of the scalar model as
a nonperturbative counter which counts how many times a fermion hops between sites m and n.
The length de"nition given in Eqs. (4.45) and (4.60) is similar but not the same. In the latter, the
propagator has been introduced in the denominator as a normalization. However, as will be shown
below, for the case of the Dirac}Wilson fermion, both length de"nitions will yield the same fractal
dimension.
The classical length is ¸
°'º''
"¦x
K
!x
L
¦. One has to compute ¸¸, as a function of a/¸
°'º''
. The
goal is to extract a critical exponent :,
¸¸,
¸
°'º''
&
?*°'º''"
(a/¸
°'º''
)? . (4.61)
By Eq. (1.1), : is related to d
'
via d
'
"1#:. Because the lattice action, Eq. (4.55), is parametrized
in terms of the hopping parameter «, it is natural to consider another critical exponent [ de"ned by
¸¸,
¸
°'º''
&
GG°'''
«
°'''
!«
«
°'''
@
. (4.62)
H. Kro( ger / Physics Reports 323 (2000) 81}181 123
The critical exponent : is de"ned in the continuum limit aP0. When a goes to zero, the
dimensionless lattice mass ma goes to zero, and « goes to its critical value «
°'''
"1/8r. For a free
Euclidean fermion theory, both exponents turn out to coincide, :"[ (see below).
A power law behavior at the critical point holds at momentum zero. This can be seen from the
Euclidean free fermion propagator for Wilson fermions, given analytically in momentum space by
[109]
I
I
"
1!2r«
"
I¯¹
cos k
I
#2«
"
I¯¹
i¸
I
sink
I
¹
. (4.63)
This is related to the space}time propagator
VW
,¸ç
V
çM
W
, by Fourier transformation
VW
"
1
<
I
e'I 'VW'I
I
, (4.64)
where <"N
¹
N
`
N
`
N
"
is the lattice volume. Note that I
I
has a pole at k"0, «
°'''
"1/2Dr in
D space}time dimensions. One can choose periodic or antiperiodic boundary conditions. They
correspond to the following choice of lattice momenta k
I
(see Ref. [109]), k
I
"2n
I
/N
I
corres
ponds to periodic boundary conditions and k
I
"2(n
I
#1/2)/N
I
corresponds to antiperiodic
boundary conditions. In both cases, n
I
30, 1,
2
, N
I
!1. Thus in the antiperiodic case k°'"
I
"/N
I
is the minimal value of k
I
. In the periodic case, k°'"
I
"0.
One can consider two types of components of the propagator: the unitmatrix component is
given by ¹
"
¹r[] and the ¸
I
component by ¹
"
¹r[¸
I
]. According to Eq. (4.63), the behavior of the
propagator at momentum k"0 is given by
¹
"
¹r[I
I¯"
]"(1!«/«
°'''
)¹ ,
¹
"
¹r[¸
I
I
I¯"
]"0 .
(4.65)
Now let us apply the di!erent de"nitions of propagator length and see what we get, Firstly, we
consider ¸ given by Eq. (4.47) and the unitmatrix component.
¸"
L
¸(x
L
, x
"
)
"a«
d
d«
L
1
4
¹r[
V¯LW¯"
]
"a«
d
d«
1
4
¹r[I
I¯"
]
"a«
d
d«
(1!«/«
°'''
)¹
&
GG°'''
a(1!«/«
°'''
)`
&
GG°'''
(«
°'''
!«)¹
&
?"
a¹ , (4.66)
124 H. Kro( ger / Physics Reports 323 (2000) 81}181
where we have used Eq. (4.56) which implies
(«!«
°'''
)¹&
?"
(!2m«`
°'''
a)¹ . (4.67)
Introducing an arbitrary "xed (dimensionful) length l
"
and c"a/l
"
, one has
¸&
C"
c¹ , (4.68)
which by comparison with Eq. (1.1) yields
d
''º°'
"2 . (4.69)
This shows that a power law like in Eqs. (4.61) and (4.62) holds, with :"["1 for the unitmatrix
component. The analogous calculation for the ¸
I
component yields :"["0 and d
''º°'
"1.
What would have been the result employing another length de"nition? Starting from ¸(x
L
, x
"
),
given by Eq. (4.45), keeping the lattice sites x
L
, x
"
"xed and letting «P«
°'''
with aP0, one has
¸(x
L
, x
"
)"a«
d
d«
1
4
¹r[
V¯LW¯"
]
&
GG°'''
a«
d
d«
1
4
¹r[I
I¯"
]
&
GG°'''
(«
°'''
!«)¹
&
?"
a¹ , (4.70)
because ¹r[
V¯LW¯"
] is dominated by the pole singularity at k"0 and «"«
°'''
. Thus one expects
to "nd the same exponents :, [ and d
''º°'
. Finally, let us consider the length de"nition ¸¸
KL
, given
by Eq. (4.60). By the same reason one has
«
d
d«
1
4
¹r[
V¯KW¯L
]&
GG°'''
(«
°'''
!«)` ,
1
4
¹r[
V¯KW¯L
]&
GG°'''
(«
°'''
!«)¹ , (4.71)
and thus
¸¸
KL
,&
GG°'''
(«
°'''
!«)¹ ,
&
?"
a¹ . (4.72)
Which again yields the same exponenta :, [ and d
''º°'
. This is con"rmed by the numerical results
discussed below.
It is interesting to look also at the in"nite volume limit (thermodynamic limit in statistical
mechanics) <PR. I.e., we "rst go to the in"nite volume limit and then go to the continuum limit
aP0. E.g., in D"1 the computation can be done analytically. One obtains for the unitcomponent
of the propagator
1
4
¹r[
VW¯L
]P
4`
1 if n"0, «(«
°'''
,
¹
`
(«/«
°'''
)L if n51, «(«
°'''
.
(4.73)
H. Kro( ger / Physics Reports 323 (2000) 81}181 125
Employing the length de"nition, Eq. (4.47), which sums over all lattice sites, one obtains
¸"
VL
¸(x
L
, x
"
)
"
VL
a«
d
d«
¹r
_
1
4
VW¯L
"a«
d
d«
`
J¯"
(«/«
°'''
)J
"a«
d
d«
1
1!«/«
°'''
&
GG°'''
(«
°'''
!«)¹
&
?"
a¹ . (4.74)
Thus going to the continuum limit («P«
°'''
and aP0) at the end, we "nd the same critical
behavior as for "nite lattice volume, Eq. (4.66) and hence the same fractal dimension. However,
a di!erent behavior is observed if we employ the length de"nition, Eq. (4.45). Then we "nd
¸(x
L
, x
"
)"a«
d
d«
¹r
_
1
4
VW¯L
"¹
`
an(«/«
°'''
)L
&
?"
0 . (4.75)
Thus there is no critical behavior, the fractal dimension would be d
''º°'
"1. The reason for the
deviation of the result is that the present length de"nition does not project onto momentum zero,
which gives the critical behavior. Unlike to the "nite volume case, now the behavior is not
dominated by its behavior of the propagator at k"0 and «"«
°'''
. The corresponding behavior is
found for the ¸
I
component of the propagator.
Now let us discuss some numerical results. For D"1 dimension, the numerical results for the
¸
¹
component and the unitcomponent are shown in Figs. 17a and b. For the ¸
¹
component, we
have chosen antiperiodic boundary conditions. We have varied N,N
¹
"4, 8, 16,
2
, 1024. In
order to approach «
°'''
we have varied «"1/[2r cos(k)#2 sin(k)], with k"k°'""/N. We have
chosen as classical length ¸
°'º''
"¦x!y¦"1. We have evaluated the space}time propagator,
Eq. (4.59) by doing the Fourier transformation of I
I
, Eq. (4.63), and of (d/d«)I
I
numerically. From
that we have evaluated the length ¸¸,, Eq. (4.60), and hence the exponents : and [, Eqs. (4.61) and
(4.62). The result shows a power law behavior with :"!0.0016, which corresponds to
d
'
"0.9984. The behavior of the unitcomponent, with periodic boundary conditions, is di!erent.
We have varied N"20, 40,
2
, 100. Because for periodic boundary conditions k°'""0, we have
approached «P«
°'''
by choosing «"0.45, 0.475, 0.4875,
2
(decreasing ¦«
°'''
!«¦ by a factor 2 in
each step). Now we have considered the classical length ¸
°'º''
"N/2. The computation of ¸¸, and
d
'
is as above. One observes (Fig. 17b) a power law with the critical exponent varying between
:"1.000 and :"0.9983, and the corresponding fractal dimension varying between d
'
"2.000
and d
'
"1.9983. Generally, one observes that the larger the size of the lattice, the closer one has to
126 H. Kro( ger / Physics Reports 323 (2000) 81}181
Fig. 17. Hausdor! dimension of Dirac}Wilson propagator in D"1. (a) Unitmatrix component, (b) ¸
I
component.
be at «
°'''
before the scaling behavior, Eq. (4.62), is seen. How can it happen that a curve in
topological dimension D"1 shows a fractal dimension larger than 1? The Hausdor! dimension
measures the hopping of the fermion forward and backward on a line (D"1), which can be fractal.
Similar results are obtained in D"2 space}time dimensions, shown in Fig. 18a and b. Now
«
°'''
"1/4r. Firstly, we have considered the ¸
`
component. Because we have a regular, symmetric
lattice, the kdependence is the same for all ¸
I
components. Thus we can interpret the ¸
¹

component as space component and the ¸
`
component as time component. We have chosen
boundary conditions periodic in space and antiperiodic in time. We have varied N
¹
"N
'ºº°°
"
H. Kro( ger / Physics Reports 323 (2000) 81}181 127
Fig. 18. Like Fig. 17, but for D"2.
4, 8, 16 independently from N
`
"N
''°°
"4, 8, 16,
2
, 1024. In order to approach «
°'''
we have
varied «"1/[2r(1#cos(k))#2 sin(k)], where k"k°'"
`
"/N
''°°
. As classical length we have
chosen ¸
°'º''
"N
''°°
/2. We have obtained :"0.0022 and d
'
"1.0022. For the unitcomponent,
we have chosen periodic boundary conditions in space and time. We have varied
N"N
¹
"N
`
"10, 20,
2
, 50. In order to approach «
°'''
we have chosen «"0.225, 0.2375,
2
(decreasing ¦«
°'''
!«¦ by a factor 2 in each step). As classical length we have chosen ¸
°'º''
"N/2. We
"nd d
'
"1.999 to d
'
"1.994 for lattices varying between N"10,
2
, 50.
128 H. Kro( ger / Physics Reports 323 (2000) 81}181
Fig. 19. Like Fig. 17, but for D"4.
Finally, we present the results for D"4 in Fig. 19a and b. Now «
°'''
"1/8r. For the ¸
"

component, we have chosen boundary conditions periodic in space and antiperiodic in time. We
have varied N
¹
"N
`
"N
`
"N
'ºº°°
"4, 8, 16 and N
"
"N
''°°
"4, 8,
2
, 256. In order to ap
proach «
°'''
we have varied «"1/[2r(3#cos(k))#2 sin(k)], where k"k°'"
"
"/N
''°°
. As classical
length we have chosen ¸
°'º''
"N
''°°
/2. As results we obtain :"0.0086 and d
'
"1.008. For the
unit component, we have chosen periodic boundary conditions in space and time. We have
varied N"N
¹
"2"N
"
"4, 8, 12,
2
, 28. In order to approach «
°'''
we have chosen
H. Kro( ger / Physics Reports 323 (2000) 81}181 129
«"0.100, 0.110,
2
(decreasing ¦«
°'''
!«¦ by a factor 2 in each step). As classical length we
have chosen ¸
°'º''
"N/2. We "nd d
'
"2.10 to d
'
"2.05 for lattices varying between
N"4,
2
, 28.
One obtains the following picture: The ¸
I
component of the propagator shows results compat
ible with d
'
"1, i.e., no fractal behavior, in D"1, 2, 4 and di!erent combinations of periodic/anti
periodic boundary conditions. However, the unit component shows results compatible with d
'
"2
for D"1, 2, 4, i.e., the same fractal behavior as in nonrelativistic quantum mechanics. The
numerical results, that is d
'
"1 for the unit component and d
'
"2 for ¸
I
component are
independent of these boundary conditions. For larger lattices, scaling sets in later (closer to «
°'''
).
Numerical errors increase when approaching the singularity «P«
°'''
. Also, numerical errors
increase with the size of the lattice. Nevertheless, one observes for D"4 that when increasing the
lattice size, the Hausdor! dimension moves closer to the value 2.
Let us discuss the results obtained above on the geometry of the Dirac propagator and put it into
perspetive to the nonrelativistic limit.
(a) Firstly, why di!er the results of d
'
for di!erent components? Let us consider periodic
boundary conditions and compare the unit component with the ¸
¹
component of the propagator.
Then the propagator, given in momentum space by Eq. (4.63), projects under Fourier transforma
tion onto the cos(k(x!y)) part for the unit component, but onto the sin(k(x!y)) part for the
¸
¹
component. In other words, the unit and the ¸
¹
component have a di!erent pole structure when
kP0. It should be noted that the unitmatrix component is dominant in the nonrelativistic
domain (k;mc), while the ¸
I
components give the dominant contributions the relativistic domain
(k<mc).
(b) The results for the free fermion propagator on the lattice can be compared with Feynman's
analytical expression for the fermion propagator in the asymptotic regime x`;t` and x`<t`
[54]. Feynman expresses the propagator kernel K
>
(2, 1)"i(i¸
I
RI#m)I
>
(t, x), and gives for the
function I
>
the asymptotic expression
I
>
(t, x)Pexp¦!im[t!(x`/2t)]}, x`;t` . (4.76)
It can be seen that the unit matrix component of the propagation kernel is essentially the
same as for a free particle in nonrelativistic quantum mechanics, where the Hausdor! dimension
is d
'
"2. This is in accord with our result d
'
"2 for the unit component of the fermion
propagator.
(c) The Hausdor! dimension of the Dirac fermion has been also computed by Cannata and
Ferrari [27], by generalizing Abbot and Wise's approach [1] (see Section 2.2.1) to the relativistic
Dirac equation. In Section 2.2.1 we have argued that Abbot and Wise's approach corresponds to
monitoring the path. This applies also to this calculation. Cannata and Ferrari "nd d
'
"2 in the
nonrelativistic quantum mechanical regime and d
'
"1 in the relativistic quantum mechanical
regime, which is consistent with our result. Moreover, they "nd d
'
"1 in the classical regime (for
both, the nonrelativistic as well as the relativistic case). Recently, Polonyi [142] has constructed
a phase space path integral for the Dirac equation. It indicates that d
''º°
"1 when the length scale
is much shorter than the Compton wave length, and d
''º°
"2 when the length scale is much larger
than the Compton wave length. This is compatible also with the prediction of Feynman's
chessboard model [52,56,62] for the Dirac particle.
130 H. Kro( ger / Physics Reports 323 (2000) 81}181
4.3.5. Length and Hausdorw dimension of the quark propagator in lattice QCD
In the previous section we have discussed the length of propagation for the case of the
noninteracting Dirac fermion. In this section we want to extend that to the theory of strong
interactions of quarks and gluons, i.e., QCD. Before doing that let us pause for a moment and
reconsider our discussion of length (via the propagator) in "eld theory versus that of non
relativistic quantum mechanics. We recall the de"nition of fractal dimension, Eq. (1.1), which states
that the total length behaves like a power law as a function of an in"nitesimal length resolution. We
have seen in quantum mechanics that the length behaves like ¸PR, while the length resolution
Ax behaves like Ax&(AtP0. That means we look at the short distance behavior in the
continuum limit. The power law, Eq. (1.1), resembles very much the de"nition of anomalous
dimension in quantum "eld theory, where a dimensionful quantitity under variation of a scale
parameter j scales by a power law, where the exponent is the canonical dimension plus an
`anomalousa contribution. Let us recall the meaning of an anomalous dimension. Considering an
npoint function 'L'(p
¹
,
2
, p
L
), one relates the renormalized to the bare function via
'L'
'°"
"ZL'L'
'º'°
, (4.77)
if there is only one species of "elds. 'L'
'º'°
does not depend on the scale j, thus holds the
renormalization group equation
_
j
R
Rj
#[
R
Rg
#n¸
'L'
'º'°
"0 , (4.78)
which holds if all masses are taken as zero. The anomalous dimension is
¸"!j
R log Z
Rj
. (4.79)
The renormalization group equation says that at ["0, which is the ultraviolet "xed point of QCD,
'L')
'°"
scales as function of momenta under change of scale z like
'L'
'°"
(p
¹
,
2
, p
L
)P'L'
'°"
(p'
¹
,
2
, p'
L
)JzB>LA, pPp'"zp, (4.80)
where d is the naive dimension. This line of thought, namely de"ning fractal dimension as
powerlaw scaling behavior of dimensionful quantities and consequently the relation between
fractal dimension and other critical exponents (scaling laws) has been used in the context of
quantum gravity (see Section 5).
The anomalous dimension ¸ governs the behavior at the "xed point ["0. If this is an ultraviolet
"xed point, like in QCD, then ¸ governs the ultraviolet behavior or shortdistance behavior. (If
there would be some infrared "xed point, it would govern the infrared or longdistance behavior).
Let us formulate a hypothesis: The anomalous dimension, i.e., the di!erence in the dimension of an
observable between classical physics and "eld theory may be due solely to virtual particle creation
and annihilation of relativistic quantum "eld theory and not due to the geometry of propagation.
This is supported by the observation that in nonrelativistic quantummechanics there is no particle
creation and the anomalous dimension is ¸"0. If this hypothesis is valid, one cannot de"ne
a fractal dimension by looking at the scaling behavior of, e.g., a twopoint function (propagator in
the regime pPR.
H. Kro( ger / Physics Reports 323 (2000) 81}181 131
In QCD physics, one is interested in the mechanism of quark con"nement, which is a long
distance phenomenon. Thus in order to understand con"nement, we need to study the infrared
behavior of QCD. Thus we are back to our question we started out: What is a suitable observable?
Fromthe above considerations we conclude: The generalization of the concept of fractal dimension
from classical physics to nonrelativistic quantum mechanics is straight forward. However, the
generalization to quantum "eld theory is problematic. The concept of anomalous dimension is
a very useful tool, but in our opinion it does not play the role of a fractal dimension. Secondly, we
are interested in infrared physics and the anomalous dimension describes the ultraviolet behavior
of QCD. Thus we suggest below to use an observable, which has an interpretation of length
allowing to derive a corresponding fractal dimension.
There is another subtle point, where the treatment in quantum mechanics and in "eld theory
di!er: In quantum mechanics the spatial resolution Ax is an observable, which is at the disposal of
the experimentalist. In "eld theory, the lattice spacing a and a high momentum cuto! regularize
the theory, but are not directly at the experimentalist's disposal. In a highenergy scattering
experiment the experimentalist calibrates the energy of the machine and the detectors correspond
ing to some upper limits of energy and momentum, E
°º`
and P
°º`
, and he measures total energy
and momentum in each scattering reaction, but this does not correspond directly to some
because of virtual particle #uctuations.
4.3.5.1. Dexnition of length via quark hopping on the lattice. The de"nition of length previously
employed for the Dirac}Wilson action can be generalized to the case when matter interacts with
radiation, QED, and also to the theory of strong interaction, QCD. Then the electron}photon
interaction, or quark}gluon interaction, respectively, has the same structure as in Eq. (4.55), but the
hopping matrix M[;] depends now on the gauge "eld via the link variables ;
I
(n) (for details see
Ref. [109]), and the matrix element (K¹)
KL
, Eq. (4.58), occurring in the fermion propagator, must
be replaced by a quantum expectation value ¸(K¹[;])
KL
, which means doing a path integral
over the gauge "eld. Moreover, one has to "x the gauge (see Refs. [15,16]). For the case of QCD
with S; (N"3) gauge symmetry the QCD Wilson action reads (in the notation of Ref. [109])
S
/!"
"S
%
[;]#S
$
[;, ç, çM] ,
S
%
[;]"[
ᮀ
1!
1
N
Re ¹r ;
ᮀ
,
S
$
[;, ç, çM]"
V
çM(x)ç(x)!«
¯"
I¯¯¹
(çM(x#j( )[r#¸
I
];
I
(x)ç(x))
"
VW
çM(y)Q
WV
[;]ç(x) , (4.81)
where «"(2ma#8r)¹ is the hopping parameter, ["2N/g` and the notation ¸
I
"!¸
I
has
been used for negative values of j. The action corresponds to a single quark #avor. The fermion
"eld variable has color and Dirac spinor indices, which have been suppressed. Note that the
132 H. Kro( ger / Physics Reports 323 (2000) 81}181
fermion "eld has been rescaled. The quark matrix Q[;] is related to the hopping matrix M[;] by
Q
WV
[;]"o
WV
!«M
WV
[;] ,
M
WV
[;]"
¯"
I¯¯¹
o
WV>I
(
(r#¸
I
);
I
(x) . (4.82)
Then the quark propagator is given by
¸ç(y)çM(x),"
1
Z,
[d;](Q¹[;])
WV
det(Q[;]) exp[!S
%
[;]] ,
Z"
,
[d;]det(Q[;]) exp[!S
%
[;]] . (4.83)
In QCD lattice calculations it is convenient to use the quenched approximation. Its physical
meaning is the suppression of closed quark loops generated by an e!ective action due to the
determinant of the quark matrix (for more details see Ref. [109]). The quenched approximation
reads
det(K[;])"1 . (4.84)
Let us "rstly consider the quenched approximation. Then
¸ç(y)çM(x),"
1
Z,
[d;](Q¹[;])
WV
exp[!S
%
[;]]
"¸(Q¹[;])
WV
, . (4.85)
This matrix element is not gauge invariant and according to Elitzur's theorem [44] vanishes when
integrating over the gauge con"gurations. Thus one has to "x the gauge.
The de"nition of length of the quark propagator is based again on the hopping expansion.
(Q¹[;])
WV
"(1!«M[;])¹
WV
"
`
J¯"
«J(MJ[;])
WV
. (4.86)
Similarly using
«
d
d«
Q¹"«
d
d«
[1#«M#(«M)`#(«M)`#2]
"0#«M#2(«M)`#3(«M)`#2, (4.87)
the length of the quark propagator can be de"ned, in analogy to Eq. (4.45), by
¸¸
WV
,"a«
R
R«
¸ç(y)çM(x), . (4.88)
The indices x,y denote the lattice sites, where the quark is created and annihilated, respectively.
This observable can be interpreted as a counter of how many times a quark hops between sites
x and y. The continuum limit aP0 corresponds to the limit «P«
°'''
. One expects again a power
H. Kro( ger / Physics Reports 323 (2000) 81}181 133
law behavior
¸¸,&
GG°'''
(«/«
°'''
!1)? , (4.89)
from which a fractal dimension can be extracted.
Let us brie#y look at what happens in the unquenched case. The term det(Q[;]) occuring in
Eq. (4.86) can be interpreted as an e!ective action, det Q"exp[!S
°''
] and expansion in « gives
det(Q[;])"exp[logdetQ[;]]
"exp[¹r log[1!«M[;]]]"exp
_
!
`
J¯¹
«J
l
¹r[MJ[;]]
. (4.90)
The "rst few terms of the e!ective action are
«¹r[M[;]]"0 ,
«`
2
¹r[M`[;]]"48«`<(r`!1) ,
«`
3
¹r[M`[;]]"0 ,
«"
4
¹r[M"[;]]"24«"(1#2r`!r")
ᮀ
1!
1
3
Re ¹r ;
ᮀ
.
(4.91)
Thus the unquenched quark propagator in the hopping parameter expansion is obtained by
expanding the matrix element of the inverse quark matrix (Q¹[;])
WV
as well as the e!ective action
S
°''
. The de"nition of propagator length, Eq. (4.88), remains valid. It can be shown that the e!ective
action by expansion in terms of the gauge coupling gives raise to closed fermion loops [109].
4.3.5.2. Computational aspects. The quark propagator has been computed in quenched approxi
mation by Bernard et al. [15,16]. They have used the Landau and Coulomb gauge and extracted
quark masses. The results show dependence on the choice of gauge. More recent lattice calculations
of quark propagators with determination of quark masses can be found in Refs. [43,65,66,100].
A computation of the length should be feasible with present day computational resources.
Numerically, one has to approach «
°'''
where the problemof critical slowing down occurs. It is hard
to estimate for which region of « one will be able to extract meaningful numbers of d
''º°
. As a guide
one can use the results of the free Wilson fermions.
4.4. Nonlocal order parameters for conxnement in lattice gauge theory
An important problem in nuclear physics is to understand the mecanism of quark con"nement.
In order to explain con"nement, 't Hooft [68] and Mandelstam [102] have proposed the idea of
monopole condensation. The idea is in analogy to the Meissner e!ect, where magnetic "eld lines
are ejected from a superconductor. Placing two (hypothetical) magnetic monopoles of opposite
charge into the superconductor, the magnetic "eld lines are compressed to a #ux tube, such that its
"eld energy is proportional to the length of the tube, giving a linearly increasing, i.e., a con"ning
134 H. Kro( ger / Physics Reports 323 (2000) 81}181
potential. In QCD this mechanism is supposed to be valid after a duality transformation (exchange
of electric and magnetic "elds) [for details see e.g. Ref. [12]]. As alternative mechanism, Nielsen
et al. have proposed the `Copenhagen vacuuma of QCD [114,115], followed by the idea of
stochastic con"nement [6,121]. The latter suggests that the important "eld con"gurations (color
magnetic #ux) are random, i.e., the area of the minimal Wilson loop can be divided into domains
and the "eld con"gurations are uncorrelated from domain to domain, which is a necessary and
su$cient condition for an area law. However, the crucial question remains open, how to derive
from the original QCD action the assumption of stochastically uncorrelated "eld con"gurations.
More discussion on this subject can be found in Ref. [22]. For compact QED in D"3 dimensions,
con"nement can be also understood in terms of an instanton gas [144]. There are criteria for quark
con"nement. E.g., there is a symmetry in gauge theories, which if unbroken leads to charge
con"nement. This is the center symmetry of the gauge group. For the gauge group S;(2) this
symmetry is unbroken for static charges with spin 1/2 [144].
In his classic paper on the con"nement of quarks, Wilson [173] has suggested to consider
a compact formulation of (Abelian) gauge theory on a space}time lattice. Wilson has introduced
a nonlocal order parameter, the socalled Wilson loop, which discriminates between a con"ned
phase and a noncon"ned phase. One should note that local order parameters are ruled out, due to
Elitzur's theorem [44]. Actually, Wegner was the "rst to introduce the concept of a nonlocal order
parameter [169]. Order parameters alternative to the Wilson loop are the Polyakov loop (see
discussion in Refs. [98,109]) and the 't Hooft loop [70]. For nonAbelian gauge theory in
continuum formulation the Wilson loop observable is given by
=
!
"¹r
_
p.o. exp
_
ig
,
!
dx
I
A
I
(x)
, (4.92)
where C is a closed curve and p.o. stands for path ordering of the exponential (see Ref. [38]). This
expression is gauge invariant. In the lattice formulation the Wilson loop observable becomes
=
!
"¹r
_
p.o. _
GH¯!
;
GH
, (4.93)
which is an ordered product of link variables along the closed contour C de"ned on the lattice. It
measures the response of the gauge "elds to an external quarklike source passing around its
perimeter. For a timelike loop it corresponds to the amplitude to produce a quark}antiquark pair,
move them along the world lines of the loop and annihilating it at the latest time. The Wilson loop
allows for an interpretation as a quark}antiquark potential. If one considers a rectangular loop in
one space and time direction with extension R by ¹, the quantum expectation value of the Wilson
loop behaves as
¸=(R, ¹),&
2`
exp[!E
"
(R)¹] , (4.94)
where E
"
is the ground state energy of the gauge "eld associated with the static quark}antiquark
sources separated by distance R. For large loops (R, ¹PR) the behavior of the Wilsonloop can
be of the following kind. Either it behaves like
¸=(C),&exp[!kP(C)] , (4.95)
H. Kro( ger / Physics Reports 323 (2000) 81}181 135
where P(C) is the perimeter of the loop and k is the selfenergy of an isolated quark source. In this
case the energy E
"
does not increase with the distance R, but approaches the selfenergy of the
isolated pair of quark and antiquark; thus there is no con"nement. Or the Wilson loop behaves like
¸=(C),&exp[!KA(C)] , (4.96)
where A(C) is the area of the loop, and K is the string tension. In this case the energy increases
linearly
E
"
(R)P
0`
KR , (4.97)
and there is con"nement. The string tension vanishes in the uncon"ned phase and is nonzero in
the con"ned phase, thus it is also an order parameter. Using strong coupling expansion, the
expectation value of the Wilson loop can be computed analytically, giving an area law for both
compact Abelian ;(1) gauge theory [174] as well as for nonAbelian S;(2) and S;(3) theory [38].
The Wilson loop has been studied nonperturbatively by many workers, using numerical simula
tions on the lattice. Creutz has shown that the compact ;(1) model has a decon"ning phase
transition [37], but the compact S;(2) model has no transition [35]. That is a hint that the
mechanism for con"nement is the same for both, the Abelian as well as the nonAbelian model. On
the other hand signals of con"nement in the continuum(noncompact) formulation of nonAbelian
gauge theories have been searched for in numerical simulations on the lattice. The results are
negative, they are only compatible with a perimeter law [131,155,157]. The reason for this behavior
is related to the fact that the noncompact (continuum) action is not manifestly gauge invariant.
This is supported by the "nding of Palumbo et al. [130], who suggested a noncompact action
using auxiliary "elds, which manifestly conserves gauge invariance and shows an area law behavior
for the Wilson loop.
4.4.1. Fractal Wilson loop in U(1) and SU(2) lattice gauge theory
From the discussion above it is evident that the geometry (area, perimeter) plays a crucial role in
the Wilson loop. The geometry is given by the world lines of propagation of the quark}antiquark
pair. This is where the fractal geometry of propagation enters. The above line of arguments leading
to the potential interpretation, Eq. (4.94), has been a bit too cavalier. A close inspection shows [12]
that some drastic assumption have been made, namely that quarks propagate as classical particles
along classical, i.e., smooth trajectories. What are the physical implications of this assumption? In
previous sections we have discussed the results of the geometry of propagation of massive particles
in nonrelatistic quantum mechanics, for free fermions on the lattice and for quarks on the lattice.
The "nding that the unit component of the free fermion propagator corresponds to a fractal
geometry of propagation (Hausdor! dimension d
'
"2) leads to the suggestion that fractal
geometry should enter into the Wilson loop. E.g., one may consider the Wilson loop, Eq. (4.92),
where the closed loop is given by a fractal curve,
=
!''º°'
"¹r
_
p.o. exp
_
ig
,
!''º°'
dx
I
A
I
(x)
. (4.98)
Such a construction is also possible on the lattice, e.g., by taking a triangular lattice, choosing
C
''º°'
as a closed Koch's curve, and letting the lattice spacing aP0. A fractal Wilson loop with
rectangular geometry has been suggested and discussed in Refs. [89,90]. The construction of
136 H. Kro( ger / Physics Reports 323 (2000) 81}181
Fig. 20. Construction of geometry of fractal Wilson loop. (a) Tree of links, (b) partial conversion of links into plaquettes.
Upper row: Ax/¸"1/2, middle row: Ax/¸"1/4, lower row: Ax/¸"1/8.
the geometry is shown in Fig. 20. It is evident that the particular geometry chosen here is one
from a large number of possibilities. The particular choice is motivated by the desired property
that the fractal `coversa the domain interior of the standard Wilson loop (in the sense as the
H. Kro( ger / Physics Reports 323 (2000) 81}181 137
trajectory of Brownian motion with d
'
"2 `coversa a 2dimensional plane). There is no physical
reason, except for simplicity, why the fractal Wilson loop lives only in a 2dimensional plane. It is
possible to extend it to 3 and 4 dimensions in space}time (note, however, that the trajectory of
Brownian motion has d
'
"2 also in higher dimension and thus does not `covera the space any
more).
4.4.1.1. Properties of fractal Wilson loop. As mentioned above, the standard Wilson loop shows
either an area law or a perimeter law behavior depending on the type of action (compact versus
noncompact), on the symmetry group (;(1), S;(2) or S;(3)), on the space}time dimension and on
the coupling strength (weak versus strong coupling). Thus it is interesting to look at the corre
sponding behavior of the fractal Wilson loop. A fractal curve with Hausdor! dimension d
'
"2
behaves in a certain sense like a 2dimensional surface. One may ask the following questions: Can
a fractal Wilson loop with d
'
"2 give an area law behavior, in particular for the noncompact
action in the strong coupling regime? On the other hand, can such fractal Wilson loop reproduce
a perimeter law for the noncompact Abelian ;(1) theory? In order to address those questions we
have to be more speci"c about the de"nition of the fractal Wilson loop. Recall the quantum
expectation value of the standard Wilson loop in Euclidean noncompact continuum formulation,
where the closed curve C"RP is the boundary of the surface P,
¸=
.
,"
1
Z,
[dA?
I
(x)]¹r
_
p.o. exp
ig
,
..
dx
I
A?
I
(x)t?
exp[!S] ,
F?
IJ
(x)"R
I
A?
J
(x)!R
J
A?
I
(x)!gf ?@AA@
I
(x)AA
J
(x) ,
S"
1
4,
d"xF?
IJ
(x)F?
IJ
(x), Z"
,
[dA?
I
(x)]exp[!S] . (4.99)
Now let us introduce a space}time lattice. We de"ne link variables ;
GH
(A?
I
) and plaquette
variables <
GHIJ
(A?
I
) both being functions of the gauge "eld A?
I
,
;
GH
"exp[ig Ax
I
A?
I
(center ij)t?] ,
<
GHIJ
"exp[igAo
IJ
F?
IJ
(center ijkl)t?] ,
(4.100)
where ij denotes the lattice sites of an elementary link and ijkl denotes those of an elementary
plaquette. Starting from the standard Wilson loop observable =
.
one de"nes the new observable
fractal Wilson loop F
.
, corresponding to an ordered path and ordered surface integral over
a fractal geometry, given by the following construction: One starts from a square loop of size ¸;¸,
with Ax/¸"1/2, 1/4, 1/8,
2
as shown in Fig. 20. Then one grows branches of links from the
boundary RP into the interior of P. Whenever four neighboring and sequential links occur, like
links 3, 4, 5, 6 in top row of Fig. 21a, they are converted to a plaquette,
;
GH
;
HI
;
IJ
;
JG
P<
GHIJ
";
GH
;
HI
;
IJ
;
JG
#O((Ax)") . (4.101)
This introduces an error O(a"). The prescription gives a wellde"ned ordered product of links and
plaquettes. The curve of links, denoted by RP
''º°'
, is indicated by the arrows in Fig. 20b, while the
order of plaquettes, its area being denoted by P
''º°'
, is indicated by the dotted line. However, there
138 H. Kro( ger / Physics Reports 323 (2000) 81}181
is intermittency (links alternate with plaquettes). This de"nes F
.
,
F
.
"¹r
link ord.#plaq. ord. _
..''º°'
_
.''º°'
¦;,<}
. (4.102)
The geometry of RP
''º°'
and P
''º°'
can be characterized as follows: In the limit Ax/¸"2,P0, the
number of plaquettes occuring in P
''º°'
behaves like Cplaquettes &`
`
(¸/Ax)`, thus the area covered
is A(P
''º°'
)"`
`
A(P). The number of links occuring in RP
''º°'
behaves like Clinks &¹
`
(¸/Ax)`, which
indicates that the curve RP
''º°'
has a fractal dimension d
''º°'
"2. Then the following properties hold
[89]:
(a) Because the area of the surface P is of order a`, and the conversion error is of order a", =
.
and F
.
coincide in the classical continuum limit (aP0).
(b) The fractal Wilson loop allows also for an interpretation as a static quark}antiquark
potential.
(c) Let us rescale the "eld variables by A?
I
PB?
I
"(gaA?
I
, F?
IJ
PG?
IJ
"a`F?
IJ
, c"1/(g, and
then make an expansion in powers of c. Although the action is not manifestly gauge invariant under
general local gauge transformations, to lowest order in c, the action is invariant under local gauge
transformations. Moreover, to lowest order in c, the quantum expectation value of the fractal
Wilson loop, ¸F
.
, shows an area law behavior for the gauge group S;(2).
(d) Now let us consider the Abelian case with ;(1) gauge symmetry [90]. In the noncompact
continuum formulation, the expectation value of the standard Wilson loop can be computed
analytically. Using the temporal gauge and considering a square loop in one space and time
direction, R"¹"¸, the result in d space}time dimensions is given by [178],
log=[C(R)]!log=[C(a)]"
1
2B`
(d/2!1)e`
;
,
0
"
du dt[(u!t)`#R`]¹B`!
,
?
"
du dt[(u!t)`#a`]¹B`!
,
0
?
du dt¦u!t¦`B
&const;R if d'3 , (4.103)
which gives a perimeter law for d"4. In the Abelian case, there is no bilinear term A;A in the "eld
strength tensor F, moreover, the ordering plays no role in the Wilson loop and by Stokes theorem
the conversion of four sequential links into a plaquette, Eq. (4.101), becomes exact. This implies that
the fractal Wilson loop F
.
agrees with the standard Wilson loop =
.
. Hence ¸F
.
, obeys a perimeter
law.
4.4.2. Hausdorw dimension of monopole current loops
In Section 2.4 we have used topological properties in order to analyze the fractal geometry
of propagation. In this section we show that fractal geometry may be useful in connection
with topological properties of gauge theories in order to distinguish the con"ned from the
decon"ned phase, thus introducing another order parameter. Let us recall brie#y solutions of
gauge "eld theories, characterized by topological properties, the socalled topological solitons, in
particular, instantons and monopoles. Let us "rst consider the noncompact continuum
H. Kro( ger / Physics Reports 323 (2000) 81}181 139
formulation. For pure Yang}Mills theory holds the following result: There are no static soliton
solutions execpt in 4 spatial dimensions [42] Such a solution, the instanton is characterized by
a topological charge. An explicit solution is given by Belavin et al. [14] and 't Hooft [69]. Although
there are no static solitons solutions in pure Yang}Mills theory in 3 spatial dimensions, such
a solution, a monopole, can exist in the presence of matter "elds. An explicit solution [67,143] can
be given, e.g., for the case of the gauge "eld with S;(2) symmetry coupled to a triplet of Higgs "elds,
i.e. the Georgi}Glashow model for weak and electromagnetic interactions. It starts from S;(2)
theory with three gauge bosons, breaking S;(2) such that two of the bosons become massive (see
[144]). In the unitary gauge, the Higgs "eld and the magnetic "eld B of the massless gauge "eld
then behave like
P
P`
j
"
0
0
1
,
BIP
P`
1
g
xI
r`
, (4.104)
which looks like a magnetic monopole of magnetic charge 1/g located at r"0. The noncompact
Abelian ;(1) gauge theory does not have any monopole solution [28,144]. However, topological
excitations do exist in noncompact gauge theories with periodic boundary conditions. In particu
lar they exist for the Abelian ;(1) theory [152]. Monopoles in compact ;(1) theory are not
physical objects, because they are driven to in"nite mass in the continuum limit [28]. If one
considers the ;(1) group as an unbroken piece of a spontaneously broken nonAbelian symmetry,
the lattice description is compact. Such a theory should possess real magnetic monopoles in the
continuum limit. An example is the above mentioned Georgi}Glashow model for weak and
electromagnetic interactions (see Ref. [144]).
Via techniques used by JoseH et al. [79] to derive the KosterlitzThouless vortex and spinwave
picture, Banks et al. [13] have shown for the compact ;(1) gauge theory on the lattice, by starting
from the periodic Villain action, that the action can be expressed as a gas of monopoles. These
authors have considered the Wilson loop, given on the lattice by Eq. (4.93), which can be
interpreted as as the response of the gauge "elds to an external quarklike source passing around its
perimeter. Such a current of unit electric charge along the closed curve C is given by
J
I
(r)"
#1 if link rPr#j3C ,
!1 if link r#jPr3C ,
0 else .
(4.105)
Denoting the angle of the directed link from r to r#j by 0
I
(r), and the plaquette angle by
0
IJ
(r)"A
I
0
J
(r)!A
J
0
I
(r), then the Wilson loop can be written as a partition function in the
presence of the current coupling to the gauge "eld,
¸=(C),"Z(J)/Z(0) ,
Z( j)"_
PI
,
`
"
d0
I
(r)exp
_
![
PIJ
(1!cos(0
IJ
(r)))
;exp
_
i
PI
0
I
(r)J
I
(r)
.
(4.106)
140 H. Kro( ger / Physics Reports 323 (2000) 81}181
Switching from the Wilson action to the Villain action [166], given by
exp[!S(0)]"
PIJ
>`
L¯`
exp
_
!
1
2
[(0
IJ
(r)!2n)`
, (4.107)
Banks et al. have obtained in D"3 the following expression:
¸=(C),"exp
_
!
1
2[
PPI
J
I
(r)G(r!r')J
I
(r')
;
>`
K'P'¯`
exp
_
!2[
PP
m(r)G(r!r')m(r')
;exp
_
2i
PPIJH
c
IJH
n
I
(n ) )¹
J
J
H
(r)G(r!r')m(r')
>`
K'P'¯`
exp
_
!2[
PP
m(r)G(r!r')m(r')
. (4.108)
Here G(r) is the lattice Coulomb Green's function, m(r) is the magnetic monopole density, B
J
(r) is
the magnetic "eld generated by the current J
I
(r), and B
I
satis"es the 3D Euclidean Maxwell
equations. The partition function factorizes into a part describing photons and a part describing
topological excitations. In D"4 dimensions a similar expression holds for the Wilson loop. But
now, instead of monopoles m(r) there is now a monople current m
I
(r), which is de"ned on the links
of the dual lattice. The monople current forms closed loops (topological charge conservation,
R
I
m
I
(r)"0). The partition function again factorizes. The monopole part reads now
exp
_
!`[
PP
m
I
(r)G(r!r')m
I
(r')
. (4.109)
The physical picture is the following: Electric current loops (J
I
) interact with a gas of monopole
current loops (m
I
). At small coupling, there are few monopole loops with small spatial extent. At
strong coupling the vacuum becomes a gas of monopoles and antimonopoles without strong
correlations.
Shortly after the "nding of the monopole action, several groups have performed nonpertur
bative lattice calculations of monopole density and monopolecurrent density. In their study,
DeGrand and Toussaint [41] have measured those monopole densities. In D"3, the monople
density in principle is obtained from the #ux through a closed surface,
ds ) B"
'"''º°°
ds
?
c
?@A
1
2
(V
@
0
A
!V
A
0
@
)"
'"''º°°
0
IJ
. (4.110)
Taking this de"nition naively, would yield a net #ux zero for each closed surface. However, one
can de"ne the monopole density via the discontinuity from one elementary box to the neighboring
one. One may assume that the plaquette angle 0
IJ
consists of two pieces: physical #uctuations
corresponding to an angle ! to , and strings which carry 2 units of #ux. De"ning
0M
IJ
"0
IJ
!2n
IJ
, where n
IJ
is the number of strings through the plaquette, one measures
the number M of monopoles inside a closed surface by M"V) B, which is expressed on the
H. Kro( ger / Physics Reports 323 (2000) 81}181 141
lattice by
M"
1
2
°'º'°° '"''º°°
0M
IJ
"
1
2
'º` '"''°° °'º'°° '"''º°°
V
I
c
IJH
0M
JH
"
'º` '"''°° °'º'°° '"''º°°
c
IJH
V
I
n
JH
. (4.111)
Analogously, in D"4 dimensions, the monopole currents M
I
can be measured on the lattice by
M
I
(x)"
1
2
c
IJ?@
V
J
0M
?@
(x) . (4.112)
The results of lattice simulations [41] for the monopole density (D"3) and the monopole current
density (D"4) are shown in Fig. 21. In both cases one observes a high density at strong coupling,
and a low density at weak coupling. In D"3 there is a smooth transition, while for D"4 there is
a steep transition at [+1, which occurs in the region where compact ;(1) gauge theory shows
a phase transition between a con"ned ([([
°'''
) and a decon"ned phase (['[
°'''
).
So far, the decomposition of the gauge action into a `photona part and a monopole part has not
been established for nonAbelian gauge theories. On the other hand, t'Hooft [71] has suggested an
Abelian projection of nonAbelian gauge theory. It is based on choosing a particular gauge, the
maximal Abelian gauge, which expresses the theory in terms of Abelian gauge "elds. The maximal
Abelian gauge [95,96] "xes the gauge such that the following expression becomes maximal
(expressed in lattice variables):
R"
VI
¹r[o
`
;
I
(x)o
`
;¹
I
(x)] . (4.113)
Using the Abelian projection in a lattice simulation, Kronfeld et al. [96] have considered the
GeorgiGlashow model, which comprises the S;(2) gauge theory, and studied the monopole
current density. For "nite temperature, there is a phase transition in compact S;(2) as a function of
[. The results obtained are quite similar to the previous ones corresponding to compact ;(1)
theory (at zero temperature). There is a con"ned phase ([([
°'''
) characterized by a high density of
monopole currents and a decon"ned phase (['[
°'''
) with a low density. It has been shown by
Suzuki and Yotsuyanagi [160] for S;(2) lattice gauge theory that the string tension of the original
action agrees with that after Abelian projection (for recent status on Abelian projection see Ref.
[141]).
4.4.2.1. Order parameter. Polikarpov and coworkers [75,135,136] have suggested to consider
the geometry of the closed monopole current loops. They have proposed to measure the fractal
geometry, and in particular the fractal dimension d
''º°'
of those loops, and study its behaviour in
di!erent phases of lattice gauge theories. Results of lattice simulations for ;(1) gauge theory [136]
are shown in Fig. 22. For ['[
°'''
(decon"ned phase) one observes d
''º°'
"1. This means that the
monopole loops are smooth curves. However, for [([
°'''
(con"ned phase) one "nds d
''º°'
'1
142 H. Kro( ger / Physics Reports 323 (2000) 81}181
Fig. 21. (a) Monople density in D"3. The solid circles are the total density of monoples, while the open circles are the
density of isolated monoples, for which no adjacent box contains an antimonopole. (b) Monopole string density versus
[ in Z(10) gauge theory for a 5" lattice. Open circles correspond to stepping down in [, closed circles correspond to
stepping up. Figure taken from Ref. [41].
approaching the value d
''º°'
"2 for ["0 (strong coupling limit). Thus in the con"ned phase the
monopole loops are crinkly, i.e. nonsmooth curves. Thus d
''º°'
plays the role of an order
parameter.
Similar studies for nonAbelian gauge theories have been reported by Polikarpov and co
workers in Ref. [75]. The authors have considered S;(2) lattice gauge theory in D"4 dimensions,
using Abelian projection. They have varied the string tension by using the method of cooling
[163,72]. The string tension is a function of [. In Fig. 23 the fractal dimension of monopole current
loops is plotted versus the Creutz ratio ,(3, 3). The Creutz ratio is de"ned as
,(I, J)"!ln
=(I, J)=(I!1, J!1)
=(I, J!1)=(I!1, J)
, (4.114)
where =(I, J) denotes the expectation value of the rectangular Wilson loop of size I;J. The
general behavior of the Wilson loop is given by
=(R, ¹)&exp[!oA#j(R#¹)#O(1/R, 1/¹)] , (4.115)
where A denotes the area and o is the string tension. The Creutz ratio cancels out the perimeter
term and yields the string tension o for R, ¹PR. In Fig. 23 di!erent sizes of monopoles have been
taken into account. It is generally believed that larger monopole loops are more important for
con"nement. Monopoles are distinguished (type I, II) by di!erent ways of counting the magnetic
H. Kro( ger / Physics Reports 323 (2000) 81}181 143
Fig. 22. Fractal dimension d
''º°'
of monopole current density versus [ for compact ;(1) in D"4 dimensions. Figure
taken from Ref. [136].
charge (for details see Ref. [75]). Fig. 23 shows that when the string tension is nonzero, i.e., when
the systemis in the con"ned phase then the fractal dimension of the monopole loop yields d
''º°'
'1.
4.4.3. Fractal geometry of deconxnement domains and gauge xxing defects
Another way to look at con"nement in lattice gauge theory putting it into perspective of fractal
geometry has been proposed by Polikarpov in Ref. [137]. He has considered "nite temperature
S;(2) lattice gauge theory, which has a phase transition from a con"ning phase at low temperature
to a decon"ning phase at high temperature. For "nite temperature pure gauge theory, the
Polyakov line is an order parameter. It is given by
¸(x)"¹r
,R
_
2¯¹
;
"
(x, ¹) , (4.116)
where N
R
is the size of the lattice in temporal direction, and the temperature is given by
¹"1/(N
R
a)"1/[. Imposing periodic boundary conditions the Polyakov line becomes a (time
ordered) closed loop being gauge invariant. The Polyakov line has the following physical inter
pretation:
¸¸(x),"exp[![F
O
] ,
¸¸(x)¸(x#r),"exp[![<
O
O
(r)] ,
(4.117)
where F
O
is the free energy of an isolated quark relative to the volume and <
O
O
(r) is a static
quarkantiquark potential. In the con"ned phase, the free energy of an isolated quark becomes
144 H. Kro( ger / Physics Reports 323 (2000) 81}181
Fig. 23. Fractal dimension d
''º°'
of monopole current density versus Creutz ratio ,
` `
for S;(2) after Abelian projection
in D"4 dimensions. Monopoles of di!erent sizes are considered. (a) Type I monoploes, (b) type II monopoles (see text).
Figure taken from Ref. [75].
in"nite, and
¸¸,"0 ,
¸¸(x)¸(x#r),&
P`
exp[!¦r¦/¸] ,
(4.118)
and there is a linear con"ning potential <&¦r¦. In the decon"ned phase, one has
¸¸,"MO0 ,
¸¸(x)¸(x#r),&
P`
M` ,
(4.119)
which means that <"const. and free quarks can exist.
Polikarpov [137] has de"ned decon"nement domains via local expectation values of the
Polyakov line. To each spatial site a certain value of ¸ is attached. A spatial site is de"ned to lie in
the decon"nement region, if ¸ attached to this site averaged with six values of ¸ corresponding to
the neighboring sites is nonzero (larger than some threshold parameter c). In this case the
decon"nement phase occupies the elementary threedimensional cube of the dual lattice. The point
lies in the center of this cube. As a result, for each gauge "eld con"guration there exist three
dimensional domains associated to the decon"nement phase. Polikarpov has performed lattice
simulations on a 8`;4 lattice by varying the coupling constant near the phase transition
(4/g`
°'''
+2.32). He has measured the surface S and the volume < of the decon"nement domains
(without saying how S and < are de"ned in detail), for 4/g`
°'''
+2.35 (decon"ned phase). The
relation volume < to the surface area S can be described by a power law <&S?. For a three
dimensional domain without cavities one would obtain :"1.5. The results of simulations are
H. Kro( ger / Physics Reports 323 (2000) 81}181 145
Fig. 24. Finite temperature S;(2) theory. Dependence of volume of decon"nement domain on their surface area.
4/g`"2.35. Figure taken from Ref. [137].
shown in Fig. 24. The data can be "tted by a power law with an exponent :"1.12$0.05, showing
a noninteger dimension of the decon"nement domain.
From the above discussion emerges the picture that the stucture of the QCD vacuum, con"ne
ment of quarks, monoples and its characterization in terms of fractal geometry are interrelated. The
monople picture of QCD however, appears only after partial gauge "xing, choosing the maximal
Abelian gauge. This leads to questions like: What happens in other gauges? What is the role of
gauge "xing defects? The "rst question has not been unambigously answered yet. The role of
gauges di!erent from the maximal Abelian gauge has been investigated by Suzuki et al. [161].
Gauge "xing defects and its geometry have been studied by Polikarpov and Yee [140]. These
authors have looked at Landau gauge "xing and maximal Abelian (MA) gauge "xing. Landau
gauge "xing (R
I
A
I
"0) is achieved on the lattice by maximizing with respect to local gauge
transformations <
V
the following function:
F'"
V
F'(x), F'(x)"
"
I¯¹
Re ¹r(<
V
;
VI
<¹
V>I
) . (4.120)
MA gauge "xing is achieved by maximizing
F`^"
V
F`^(x), F`^(x)"
?
"
I¯¹
¹r(<
V
;
VI
<¹
V>I
z
?
<
V>I
;¹
VI
<¹
V
z
?
) . (4.121)
146 H. Kro( ger / Physics Reports 323 (2000) 81}181
Sites of gauge "xing defects, i.e., resistent sites, can be de"ned as all sites x such that F(x)(F
°'"
where F
°'"
is some parameter. Denoting by N
B
(r) the number of resistant sites inside a hypersphere
of radius r centered at resitant site d, then the fraction of resitant sites averaged over resistant sites is
given by
N(r)"
B
N
B
(r)
B
1 . (4.122)
The fractal dimension is given in the standard way by a power law N(r) versus r, thus
D
''º°'
"
dlog N(r)
dlog r
. (4.123)
The results of lattice simulations [140] are shown in Fig. 25. For Landau gauge "xing, there is
a plateau D
''º°'
+1 over four slices of radius r. Then there is a crossover to a region where
D
''º°'
"4. This is in contrast to MA gauge "xing, which shows no plateau and crossover from
D
''º°'
+0 at small r to D
''º°'
"4 at large r. In conclusion of this section, Landau gauge "xing
defects give additional information (D
''º°'
+1) on the gauge "xed vacuum of QCD, which is
complementary to the monopole currents.
4.5. Fractal geometry and critical behavior of lattice xeld theories
4.5.1. Critical behavior of lattice gauge theories as a function of noninteger dimension of space}time
In condensed matter physics systems described by statistical mechanics as well as in elementary
particle physics systems descibed by relativistic quantum"eld theory show critical behavior, which
is independent to a certain extent from the particular local interaction. One has coined the notion
of universality and universality classes. It means that two systems are in the same universality class,
if they have the same dimension d and order parameters have the same dimensionality D. Also the
critical exponents are the same. In other words for systems of a given symmetry of the order
parameter and interaction range, the critical properties depend solely on the dimensionality d. For
examples see Ref. [21].
In order to investigate critical properties of a given model, it is important to study its dependence
on dimensionality. One way to do this has been suggested by Wilson and Fisher [172,173]. One
makes a perturbative expansion in the parameter c, where d"4!c. In 1980 it has been suggested
by Mandelbrot and coworkers [60,61] to put the model on a lattice of fractal geometry, which has
noninteger dimension d
''º°'
, solve the model and study its critical behavior under variation of d
''º°'
.
In Refs. [60,61] the "rst studies were performed on the Ising model using Sierpinski carpets as
fractal geometry and solving the system by Migdal}Kardano! renormalization group techniques.
One should note that selfsimilar fractal geometries are invariant under scale transformations, but
they are not invariant under translations. The important results obtained was the dependence of
critical properties on the fractal dimension d
''º°'
, but also on topological properties like rami"ca
tion, connectivity and lacunarity (see Ref. [60]). Lacunarity measures the extent of failure of the
fractal to be translationally invariant. In Ref. [61] it is shown that in the limit of low lacunarity, the
physical properties obtained on the fractals are identical to those of the analytically continued
hypercubic lattices. Then Migdal's renormalization recursion relations show for the Ising model on
H. Kro( ger / Physics Reports 323 (2000) 81}181 147
Fig. 25. Fractal dimension D
''º°'
of domain of gauge "xing defects inside hypersphere versus size of hypersphere (log(r/a))
for Landau gauge "xing and maximal Abelian gauge "xing. The homogeneous distribution corresponds to randomly
placed pseudo defects. Figure taken from Ref. [140].
noninteger geometry of dimension d
''º°'
"1#c that the correlation length behaves like
¸&¦¹!¹
°
¦J , (4.124)
and the critical exponent v of the correlation length is related to the dimension via
1/v"c#O(c`) . (4.125)
This result con"rms predictions of realspace renormalization group calculations. Critical proper
ties of spin systems on fractal lattices have been studied by Bhanot et al. [17,18]. This will be further
discussed in Section 6.
The idea to study critical behavior of lattice gauge theories in noninteger space}time dimensions
and in particular look for upper critical dimensions has been pursued by Bhanot and Salvador
[19]. For example, compact Abelian lattice gauge theory with the gauge group Z(2) was found to
have a "rstorder phase transition in d"3#1 dimensions [36]. One may ask the question: What
happens to the phase transition when d is noninteger but close to four? What happens with the
148 H. Kro( ger / Physics Reports 323 (2000) 81}181
order of phase transition? To address such questions, Bhanot and Salvador have interpolated the
dimension of the space}time lattice, by constructing a lattice in analogy to a Sierpinsky carpet. The
trick is, however, to make sure that gauge invariance is not violated, i.e., the fractal lattice is built
from plaquettes and that scale invariance is conserved (under discrete scale transformations). This
is done in the following way: Starting from a D dimensional Euclidean lattice of length ¸, Bhanot
and Salvador construct a fractal lattice, labeled by D, n, c, by blocking and deleting blocks. They
start by picking c¸" sites at random and delete all plaquettes on the hypercube in the forward
direction associated with these sites. They scale the lattice by a factor of 2, i.e., consider blocks of
2" sites. Then a fraction c of those blocks is deleted. The choice of blocks is made randomly. Finally,
the procedure of blocking the lattice and deleting a "xed fraction of blocks is repeated n times. The
procedure is very close to that of constructing a Sierpinsky carpet. D is called the embedding
dimension and n is the number of decimations. The fractal lattice is characterized by its Hausdor!
dimension, which in this case is given by
d
'
"D#ln(1!c)/ln2 . (4.126)
In other words tuning the parameter c allows to choose a lattice of noninteger, i.e., fractal
dimension. Mathematically, a fractal requires that its geometry is selfsimilar on all length scales
down to length scale zero. In the construction of the fractal lattice, this corresponds to let nPR.
Of course on a "nite lattice this cannot be realized, n has an upper limit (n44 in the results shown
below).
Observables, which are sensitive at a phase transition are the expectation value of the average
action per plaquette (corresponding to the free energy E), the histogram P(E) versus E (P(E) is the
probability that the free energy lies between E and E#E), and the latent heat ¸
F
. Another
indicator is the hysteresis curve of the free energy versus [. Indications of a "rst order phase
transition are: Existence of hysteresis, a clear separation of two peaks in the histogram and
a nonvanishing latent heat. The latent heat as function of c"4!d
'
is shown in Fig. 26. It shows
a vanishing latent heat for c'0.017 and n"3, i.e. for d
'
43.983, the transition is of second
order. For smaller values of c, the results are compatible with a "rst order transition. In summary,
the numerical results are compatible with the picture of a second order transition as the dimension
approaches four from below, and switches abruptly to "rst order when the dimension is
four.
4.5.2. Critical behavior of U(1) gauge theory
The previous considerations by Bhanot and Salvador concerning lattice gauge theory with the
Abelian symmetry group Z(2) have been generalized to Z(N) where NPR, which results in the
;(1) model. Ilgenfritz and Schiller [73] and later Ilgenfritz et al. [74] have done a careful analysis of
the ;(1) model on a fractal lattice when approaching the dimension four from below. It is generally
believed (by evidence of numerical simulations, but not really proven) that the compact ;(1) gauge
theory has a "rstorder phase transition in D"4 dimensions. Following Bhanot and Salvador
[19], Ilgenfritz et al. have studied the order of the phase transition as a function of the fractal
dimension. The fractal lattice has been chosen as a selfsimilar Sierpinski carpet. Instead of the
gauge group ;(1) the gauge group Z(N"16) has been used. This means that the link variables take
as values the group elements
;"exp(i0), 0"2k/N, k"0,
2
, N!1 . (4.127)
H. Kro( ger / Physics Reports 323 (2000) 81}181 149
Fig. 26. The latent heat ¸
F
versus 1/n for various values of c"4!d
'
. The points for n"4 were obtained by simulation
on a 32" lattice. The other points correspond to a 16" lattice. Figure taken from Ref. [19].
The authors have looked at the following observables to detect the phase transition and determine
its order: speci"c heat, the histogram P(E) versus E, and the Binder cumulant. The Binder cumulant
is de"ned by
B
*
([)"1!
¸c",
3¸c`,`
, (4.128)
where c denotes the mean plaquette action (corresponding to the free energy), ¸" is the volume of
the embedding lattice and ["1/g` is the coupling. The Binder cumulant has the following
behavior at a phase transition: Its minimum indicates the phase transition point. For a "rstorder
transition, the curve is narrow at the minimum, for a secondorder transition the curve is wide at
the minimum.
The numerical simulations were done on lattices of size 4" up to 10". Some results are presented
in Figs. 27 and 28. In Fig. 27, the spectral density of the mean plaquette action is displayed for
various dimensions (note that d here denotes the coupling dimension de"ned by d"1#2N
º
/N
'
,
where N
º
is the number of active plaquettes and N
'
is the number of retained links. d is not the same
as the Hausdor! dimension d
'
but plays a similar role: dP4 when d
'
P4). It shows a clear double
peak structure ("rst order) as long as d'3.95. The same tendency can be observed in Fig. 28
displaying the cumulant. When the double peak structure of the spectral density disappears
(transition becoming second order), the cumulant starts getting broader. In conclusion, the
"rstorder transition disappears for coupling dimension d(3.95, corresponding to d
'
"3.988,
with embedding volumes 8" and 10".
4.5.3. Critical clusters in
4
4
theory
The correlation length of a manybody system is a characteristic length, which tells something
about the distance of interaction, which is dynamically generated (the original interaction being
150 H. Kro( ger / Physics Reports 323 (2000) 81}181
Fig. 27. Spectral density of the mean plaquette at critical couplings obtained by the multihistrogram method, for various
dimensions. Embedding lattice: 10". Figure taken from Ref. [74].
local). The correlation length when approaching the critical point of a secondorder phase
transition, like in the Ising model when going to the critical temperature ¹P¹
°
, goes to in"nity
following a power law (see Eq. (4.124)). The power law is a way to express mathematically the
property of selfsimilarity. On the other hand, we know that the spins of the Ising model, at the
critical point form clusters of spins of the same orientation, which occur at any size, and the system
is said to be scale invariant (see Fig. 29). Since selfsimilarity at all length scales and scale invariance
are synonymous, it is suggestive to look for a relation between the correlation length and the size of
a critical cluster, and in particular between the corresponding critical exponent v of the correlation
length and the fractal dimension d
'
of the size of the critical cluster. A relation betwen the critical
exponent of the magnetization of the Ising model in 3 dimensions and the fractal dimension of the
Fortuin}Kasteleyn}Coniglio}Klein clusters has been suggested [32] to be of the form
d
'
"d!
[
v
"
d
1#1/o
, (4.129)
where d is the embedding dimension, d
'
is the fractal dimension and [/v is the critical exponent of
the mean magnetization, which scales like m&¸@J on a lattice of size ¸B.
The question, what is the fractal dimension of the FKCK clusters in a "eld theoretic model has
been raised and answered by Jansen and Lang [78]. They have considered the nonlinear sigma
model with O(4) symmetry in four dimensions. While the Ising model is obtained in the limit
zPRof the scalar " model with z" interaction, the O(4) nonlinear sigma model is the zPR
limit of the 4component " model. The action of the model is
S"2«
V
B
I¯¹
?
V
?
V
, ¦¦"1 , (4.130)
where :"1,
2
, 4. The model plays a role in the standard model of the electroweak interaction.
H. Kro( ger / Physics Reports 323 (2000) 81}181 151
Fig. 28. The Binder cumulant B
*
versus [ constructed by means of the multihistogram method for various dimensions
d44 embedded in a 10" lattice. Figure taken from Ref. [74].
The numerical simulations have been done on lattices of size: ¸" with ¸"8, 12, 16,
2
, 28. As
result Jansen and Lang obtain d
'
"3.06(6) and for the critical exponent o"3.25(25). Mean "eld
theory predicts o"3. This result is in agreement with Eq. (4.129).
5. Fractal geometry and quantum gravity
5.1. Random surfaces and quantization of gravity
The Hausdor! dimension of "eld con"gurations has recently become a subject of interest in
quantum gravity (in particular in D"2), where the role of dynamical "elds is played by the
geometry of space}time [8,10,29,40,84,85,116]. The Hausdor! dimension has been de"ned as
a power law relation between two dimensionful observables at a critical point [84]. It is interesting
to establish relations between the Hausdor! dimension and critical exponents. Such a relation is
known to exist between d
'
and the critical exponent v, which is the analogue of a critical exponent
of the spin}spin correlation function in statistical mechanics. In quantum gravity [10] v is de"ned
as the critical exponent of a mass, m(Aj)&(Aj)J, when the cosmological constant tends to its
critical value AjP0, where the mass m(Aj) characterizes the fallo! behavior of a twopoint
function G
I
(r)&exp[!m(Aj)r]. Then holds the following relation (scaling law) d
'
"1/v, relating
the Hausdor! dimension to the critical exponent v [10]. E.g., for quantum gravity in D"2, one
obtains d
'
"4 [10]. Thus also d
'
plays the role of a critical exponent.
Let us start out by de"ning some notation of quantum gravity (see, e.g., Refs. [9,168]). The
Einstein}Hilbert action describing the theory of gravitation is given by
S"
1
16G,
d"x(!gR , (5.1)
152 H. Kro( ger / Physics Reports 323 (2000) 81}181
Fig. 29. Clusters of spinup and spindown in the 2dimensional Ising model. From top left to bottom right
¹/¹
°'''
"0.97, 0.99, 1, 1.01, 1.06, 1.15; ¹"¹
°'''
at middle right. Figure taken from Ref. [21].
where G denotes the coupling constant, g is the determinant of the metric tensor and R is the
curvature scalar. One may also consider a cosmological term
S
°º'°
"
1
16G,
d"x!2(!g . (5.2)
The Einstein}Hilbert action is a successful theory on the classical level. When one tries to
construct a quantum theory of gravitation, several problems arise (see below). The dynamics of the
"eld is given by the curvature of space time. Since the curvature, i.e., the geometry of space}time, is
of central importance in gravity, this suggests the following question: What is the local character of
H. Kro( ger / Physics Reports 323 (2000) 81}181 153
it? In particular, does this geometry have a fractal character? We will try to give some answers to
those questions.
In order to quantize the theory, one constructs a path integral. In order to do so, one has to
regularize the curved space. Regge [146] has suggested a way to do it using simplical manifolds.
Di!erent approaches have been pursued, namely the socalled Regge approach [147] and the
socalled dynamical triangulation approach [165] (see also Ref. [83]). In the Regge approach one
keeps the triangulation "xed and the edge lengths are allowed to vary. In dynamical triangulation
the simplical manifold consists of a set of equilateral simplices which are locally changed. There are
a number of problems when quantizing gravity: For example the measure of the path integral is not
unique, and the action is unbounded from below.
5.2. Fractal structure
Once having established a quantized theory of gravity, one can proceed, either via analytical
methods in D"2, or via numerical simulations by Monte Carlo on the lattice, to compute
quantum expectation values of observables. In particular, one may ask the question: What is the
scaling exponent p in quantum gravity of an observable O,
¸O,
'`
&lengthE , (5.3)
which in the classical theory scales like
O
°'º''
&lengthB , (5.4)
where d

is some integer dimension. In order to study those questions in quantum gravity, Kawai
and Ninomia [84] have employed a renormalization group method and the technique of c
expansion around dimension D"2. They suggest that the fractal dimension is an order parameter
for quantum #uctuations of space}time. This is similar to the proposition by Polikarpov et al. that
the fractal dimension of monopole current loops is an order parameter in lattice QCD (see Section
4.4.2). In particular, the "ndings by Kawai and Ninomia are the following. If the coupling constant
G, introduced for pure gravity by Eq. (5.1), is below some "xed point, G(GH, then the geometry
has a fractal character at short distances and a smooth (nonfractal) character at large distances. If
G'GH, then strong #uctuations occur at large distances. One should note, however, that the
coupling constant G makes sense only when gravity couples to other terms. Kawai and Ninomia
have considered the cosmological term, a fermion mass term, the interaction of the Thirring model
and a general spinless "eld.
As an example, let us consider the Lagrangian of gravity in the presence of the cosmological
term. The cosmological term has a physical interpretation of the vacuum energy of quantum "elds
in the universe. Present astrophysical data allow only for a very small number, which is di$cult to
explain theoretically. This is the socalled cosmological constant problem.
¸"!
1
16G
"
(gR#
"
(g . (5.5)
Making a scale transformation of the "eld,
g
IJ
Pzg
IJ
, (5.6)
154 H. Kro( ger / Physics Reports 323 (2000) 81}181
corresponds to the following change for G and ,
G
"
PG
"
z"`>¹ ,
"
P
"
z"` .
(5.7)
Consequently, G"
"
"`
"
remains unchanged.
The Wilson loop, which is a gauge invariant object in lattice gauge theory and plays the role of
an order parameter in lattice gauge theory, is not a gauge invariant object in quantum gravity.
When considering alternatives for an order parameter, the dynamical role of geometry suggests
that geometry itself may play such a role. In particular, one may consider the fractal dimension of
space}time as a candidate for an order parameter. How can we de"ne the fractal dimension? Let us
start by a simple example: Suppose we have a lattice and we denote by n(¸) the number of lattice
sites at which we can arrive after ¸ steps of hopping (from one site to the nextneighbor site only)
after starting from a given initial point. Here n(¸) corresponds to a volume. If we "nd that n(¸) has
a power law behavior, like n(¸)J¸"', when ¸PRthen we say that D
'
is the fractal dimension of
the lattice. This de"nition, applied to a D dimensional regular lattice, yields n(¸)&¸", and thus
D
'
"D.
In order to construct a corresponding quantity in quantum gravity, one can consider a power
law relation between quantum expectation values of observables, which classically have dimension
of volume and length. For example, in quantum gravity, one may consider the following dimen
sionful quantities,
¸O
*
,"
,
(g
IJ
dxIdxJ
,
¸O
4
,"
,
(g d"x
. (5.8)
On the classical level, the observable O
*
has the dimension of length, while the observable O
4
has
the dimension of volume. Then one may de"ne a scaling exponent d
'
via
¸O
4
,J¸O
*
,B' . (5.9)
Then d
'
may be interpreted as Hausdor! dimension. This way of de"ning d
'
via scaling exponents
of dimensionful quantities is very similar to the scaling exponent, closely related to the anomalous
dimension, of the npoint vertex function in quantum "eld theory. Recall the scaling behavior (see
Section 4.1),
'LL$'&zB>LA>L$A$ , (5.10)
when the momenta p
G
undergo a change of scale p
G
Pzp
G
, and where ¸
, ¸
$
are called the anomalous
dimensions.
A closer look to O
*
as de"ned by Eq. (5.8) reveals that there is a problem, namely, O
*
is not gauge
invariant. That has led Kawai and Ninomia [84] to replace that de"nition by expressions, which
are manifestly gauge invariant and classically have dimension of length and volume, respectively.
In particular they suggest to consider "rstly the mass term for gravity coupled to a fermion "eld
H. Kro( ger / Physics Reports 323 (2000) 81}181 155
and secondly the cosmological term,
O
*
"m
,
eçMçd"x ,
O
4
"
,
(g d"x . (5.11)
Let us now consider the scaling behavior of the fermion mass and the cosmological constant
under change of the scale. The behavior of the fermion mass is given by the renormalization group
equation. The renormalization of the fermionic part of the Lagrangian gives
¸
$
"eçMi¸ID
I
ç!mZ
K
(G)eçMç . (5.12)
The behavior of the wave function renormalization Z
K
under some change of the scale parameter
j used in dimensional regularization is given by the ¸ function (anomalous dimension) of the
renormalization group equation (see Section 4.1)
¸"!(dlnZ
K
/d lnj) . (5.13)
This results in the following behavior of the mass:
mJjA
H
(5.14)
where ¸H"¸¦
%¯%
H
and GH is a "xed point.
Now let us see the behavior of m and under some change of scale. One obtains
Pz" ,
g
IJ
Pz`g
IJ
,
mPz"`
m ,
(5.15)
where 2 is the scale dimension of the fermion mass. Let 2
"
denote the canonical dimension of the
fermion mass, i.e., the dimension in the classical case. From Eqs. (5.14) and (5.15), one "nds
2"D!1#¸H"2
"
#¸H . (5.16)
The theory with the parameters , m is equivalent to the theory with parameters z", z"`
m.
Thus
O
4
&Pz" ,
O
*
&mPz"`
m .
(5.17)
This implies
O
4
&O"'"`
'
*
"O"'
*
, (5.18)
and hence
D
'
"
D
D!2
"
D
1!¸H
. (5.19)
156 H. Kro( ger / Physics Reports 323 (2000) 81}181
The computation of the scale dimension of the fermion mass can be done analytically in D"2
dimensions, using conformal symmetry. This yields
"
(25!c!(13!c
(25!c!(1!c
, (5.20)
where c is the central charge of the conformal (Virasoro) algebra. What is the meaning of conformal
algebra and central charge? In D"2 any holomorphic mapping of the complex plane is angle
preserving. An in"nitesimal conformal transformation of the complex coordinates z, z can be
written as zPz#p(z), p(z)"!zL>¹c for n integer. Note that the transformation does not depend
on z . An analogous transformation law holds for z , being independent of z. These mappings are
generated by di!erential operators (de"ned on functions of z, z ),
l
L
"!zL>¹R
X
, lM
L
"!z L>¹R
X
, (5.21)
which satisfy the commutation relations
[l
K
, l
L
]"(m!n)l
K>L
. (5.22)
An analogous formula holds for lM
L
. Also one has [l
K
, lM
L
]"0. The Lie algebra of the l
L
is called Witt
algebra. So far we have considered the classical case. In quantum theory, there are analogous
operators ¸
L
for integer n, which satisfy
[¸
K
, ¸
L
]"(m!n)¸
K>L
#¹
¹`
(m`!m)o
K>L"
C . (5.23)
The ¸
L
form again a Lie algebra. The new generator C is called central element or central charge.
Its Lie bracket with the whole algebra is zero,
[C, ¸
L
]"0 . (5.24)
This in"nite dimensional algebra is called Virasoro algebra.
This number c depends upon the number N
Q
of scalar "elds and the number N
D
of fermionic
"elds, which couple to quantum gravity. One has
c"N
Q
#¹
`
N
D
. (5.25)
This yields eventually for the fractal dimension,
D
'
"2
(25!c#(13!c
(25!c#(1!c
. (5.26)
This yields the following numbers for D
'
:
(i) cP!R, D
'
P2.
(ii) c"1/2, D
'
"3.
(iii) c"1, D
'
"2#(2+3.41.
Comparison of this result with numerical simulations with random triangularization [83] shows
agreement.
H. Kro( ger / Physics Reports 323 (2000) 81}181 157
Another approach has been taken by Ambjorn and Watabiki [10]. They have determined the
critical behavior of the twopoint correlation function for a geodesic distance. Under the assump
tion of scaling behavior as in statistical mechanics, they "nd a relation between critical exponents,
¸"v(2!p) . (5.27)
This is Fisher's scaling law. Here ¸ denotes the analogue of the spin susceptibility critical exponent,
v is the analogue of the spin}spin correlation length critical exponent and p is the analogue of the
spin}spin correlation function anomalous dimension. They obtain
v"1/d
'
. (5.28)
Pure gravity in D"2 yields
v"¹
"
, d
'
"4 . (5.29)
They consider 2D gravity as a scaling limit of the socalled simplical quantum gravity. In
simplical quantum gravity, the partition function Z and the Einstein}Hilbert action are expressed
as
Z(j, G)"
2¯O
1
c
2
e1'2' ,
S[¹]"jN
2
!
1
4G
(2!2g
2
) ,
(5.30)
where G is the coupling constant, j the cosmological constant, t the class of triangulations of closed
2manifolds, ¹ a triangulation in t, N
2
the number of triangles in ¹, c
2
the symmetry factor and
g
2
the genus of the manifold. One chooses a "xed topology: then the last term, which is
a topological invariant, can be dropped. Consider that the cosmological constant approaches
a critical value, jPj
!
. Let us de"ne Aj"j!j
A
. Then the partition function behaves as
Z(j)&(Aj)`A . (5.31)
It is useful to introduce the geodesic distance between two links. This can be de"ned as the shortest
path of links on the dual lattice, i.e., the shortest `trianglepatha between the two given links on the
original triangulation. Moreover, let t
`
(r) denote the ensemble of triangulations with two marked
links being separated by a geodesic distance r.
Then one can write the twopoint function as
G
I
(r)"
2¯O`'P'
exp[!jN
2
] , (5.32)
which is the sum over triangulations with geodesic distance r, weighted by the exponent of the
action. The large distance behavior is characterized by a `massa m(Aj),
G
I
(r)&
P`
exp[!m(Aj)r] . (5.33)
The twopoint function G
I
(r) can be interpreted as partition function for universes of linear
extension r. Now let us consider the continuum limit. In order that G
I
(r) has a wellde"ned limit,
158 H. Kro( ger / Physics Reports 323 (2000) 81}181
one needs the following conditions to be satis"ed,
m(Aj)rP
P`
MR"const ,
m(Aj)P
I"
0 .
(5.34)
When going to the critical point, the twopoint function is assumed to display some power law
behavior, being given by standard expressions of statistical mechanics,
G
I
(r)&exp[!m(Aj)r] if m(Aj)<1 ,
G
I
(r)&r¹E if m(Aj)r;1 ,
,
I
,
,
dr G
I
(r)&
R`Z(j)
Rj`
&(Aj)A . (5.35)
At the critical point, also the correlation length is assumed to display a power law behavior, thus
m(Aj)P
I"
(Aj)J . (5.36)
In statistical mechanics, there are certain scaling laws, which relate critical exponents. For example,
Fisher's scaling law says
¸"v(2!p) , (5.37)
where ¸ is the susceptibility critical exponent, de"ned by Eq. (5.35), v is the critical exponent of the
correlation length, de"ned by Eq. (5.36), and p is the anomalous scaling exponent of the correlation
function, de"ned by Eq. (5.35).
Now we are ready to introduce a de"nition for the Hausdor! dimension of the geometry of
quantum gravity. Ambjorn et al. [10] suggest to de"ne the Hausdor! dimension by a scaling
relation between a quantum expectation value of naive dimension of volume and the geodesic
distance r. In particular they consider the average number of triangles (which corresponds to
a volume) for a given geodesic distance r,
¸N,
P
"
2¯O`'P'
N
2
exp[!jN
2
]
2¯O`'P'
exp[!jN
2
]
. (5.38)
Then d
'
is given by
¸N,
P
&rB' when rPRand m(Aj)r"const . (5.39)
One computes
¸N,
P
&!
1
G
I
(r)
RG
I
(r)
Rj
&
R
Rj
m(Aj) r¦
KP¯°º"''
"
Rm
RAj
r
KP¯°º"''
&v(Aj)J¹ r¦
KP¯°º"''
"
v
Aj
mr
KP¯°º"''
&
v
Aj
&
1
m¹J
&r¹J , (5.40)
H. Kro( ger / Physics Reports 323 (2000) 81}181 159
where Eqs. (5.35) and (5.36) have been used. Comparison with Eq. (5.39) yields
d
'
"1/v . (5.41)
This has the form of a scaling law.
5.3. Numerical results from lattice simulations. Gravity coupled to matter: Ising model
and 3state Potts model
The previous analytic calculations have been compared with numerical simulations by Catteral
et al. [29]. These authors have considered in 2D (a) pure gravity, (b) gravity coupled to the Ising
spin system, and (c) gravity coupled to the 3state Potts model. Catteral et al. have studied the
loop}loop correlation function
n
JJ
(r, N)"
R¯O`','
1 , (5.42)
where l and l' denote the length of the initial and "nal loops, r denotes the geodesic distance
meaning the minimal number of links that must be traversed from the initial to the "nal loop and
N being the number of triangles. Applying the "nitesize scaling hypothesis yields for n(r, N)
n(r, N)"NNn(r/NO) , (5.43)
where p, q are some critical exponents of the model. The function l
%
"NO is a dynamical length
scale, and as N corresponds to a volume. This can be used to de"ne a Hausdor! dimension
d
'
"1/q . (5.44)
Because the integral of n(r, N) over all geodesic distances must give the total number of points N, p
is not independent, being related by
p"1!1/d
'
. (5.45)
Fig. 30 shows the position r
"
of the peak of the distribution of n(r, N) versus N and also its maximal
value n(r
"
). From the scaling assumption one implies
r
"
&N¹B', n(r
"
)&N¹¹B' . (5.46)
Extracting in this way d
'
from the data yields for pure gravity
r
"
: d
'
"3.133(33) (lattice) d
'
"3.835(59) (dual lattice) , (5.47)
n(r
"
): d
'
"4.040(98) (lattice) d
'
"3.594(77) (dual lattice) . (5.48)
Fig. 31 shows plots of n(r)/N¹¹B' which for a suitable choice of d
'
should become independent of
N. Such a determination yields d
'
"3.79 on the lattice and d
'
"3.150 on the dual lattice. This
should be contrasted to the analytical value of d
'
"4 for pure gravity in 2D.
Finally let us consider gravity coupled to matter, namely to the Ising model and to the 3state
Potts model (for de"nition see Eq. (6.43)). For both models the exact solutions are known [39]. The
Ising model has central charge c"1/2, while the 3state Potts model corresponds to c"4/5. For
the Ising model coupled to gravity the numerical simulations are presented in Figs. 32 and 33. The
160 H. Kro( ger / Physics Reports 323 (2000) 81}181
Fig. 31. Pure gravity. Scaling of the distribution n(r, N). (a) Lattice and (b) dual lattice. Shown are the curves "tted to
distributions after rescaling. The value of d
'
is chosen to minimize the total chisquare of the "ts. Figure taken from
Ref. [29].
Fig. 30. Pure gravity. Scaling of (a) location of the peak r
"
of the distribution n(r, N) and (b) their maximal value n(r
"
).
Figure taken from Ref. [29].
results extracted from n(r, N) are compatible with d
'
"3.9. The analytic result given by Eq. (5.26)
yields for c"1/2 the value d
'
"3. For the 3state Potts model the results are summarized in
Table 2. Again the results are compatible with d
'
"3.9. It is fair to say that it is not fully
understood why there is so little di!erence between the results for pure gravity and gravity coupling
to the Ising or 3state Potts model.
6. Fractal geometry and spin systems
This Section will revisit some topics already discussed in Section 4. We will discuss spin physics
at the critical point and study the geometry of spin clusters. The goal is to establish a relation
between the fractal dimension of the spin clusters and critical exponents. Moreover, we will discuss
the order of the phase transition as a function of noninteger dimension, which means putting the
spin system on a fractal lattice. Finally, we want to discuss topological excitations in the X> spin
H. Kro( ger / Physics Reports 323 (2000) 81}181 161
Fig. 32. Ising model coupled to gravity. Scaling of r
"
and n(r
"
). The scaling behavior is used to extract d
'
from the slope.
Figure taken from Ref. [29].
Fig. 33. Ising model coupled to gravity. Collapsing the data on a single curve using one scaling parameter. Figure taken
from Ref. [29].
model and their relation with fractal geometry. It should be noted that many of the ideas discussed
previously in the context of "eld theory have been originally introduced for spin systems.
6.1. Critical behavior of spin systems as a function of noninteger dimension of space}time:
Ising model
The original idea to study critical behavior for noninteger space}time by use of a fractal lattice is
due to Gefen, Mandelbrot and Aharony [60]. The spin model chosen is the Ising model, where
spins are classical variables taking the values S"$1. The Ising model, where the spins are located
on a regular lattice, is given by the Hamiltonian
H"J
GH
S
G
S
H
, (6.1)
162 H. Kro( ger / Physics Reports 323 (2000) 81}181
Table 1
Ciniglio}Klein clusters of the qstate Potts model: fractal dimensions of the whole cluster (D), of the hull (D
'
), and of the
red bonds (D
"
)
q D D
'
D
"
0 2 2 `
"
1 "¹
"`
`
"
`
"
2 ¹`
`
`
`
¹`
`"
3 ``
¹`
`
`
`
`"
4 ¹`
`
`
`
0
Note: For de"nition of red bonds and hull see Section 6.2 and Fig. 40. Table taken from Ref. [34].
Table 2
Extracted values of d
'
for the Ising and 3state Potts model coupled to gravity
Ising model 3state Potts model
n(r, N) g
SL
(r, N) n(r, N) g
SL
(r, N)
d
'
,` ` d
'
,` ` d
'
,` ` d
'
,` `
(a)
500}1000 3.758(53) 2.6 3.76(12) 0.93 3.752(63) 0.68 4.01(26) 2.5
1000}2000 3.802(55) 0.77 3.75(15) 1.0 3.787(65) 0.29 4.11(18) 1.0
2000}4000 3.833(56) 1.0 3.73(12) 2.5 3.864(63) 1.0 4.04(22) 3.2
4000}8000 3.893(61) 0.88 3.69(09) 3.9 3.870(73) 0.15 4.11(19) 0.41
8000}16000 3.870(87) 0.35 3.80(10) 0.99 3.820(97) 0.58 4.14(15) 0.56
(b)
1000}16000 3.862(74) 1.4 3.851(53) 4.5 3.831(32) 2.4 3.966(64) 12.5
(c)
Position 3.875(53) 3.88(19) 3.879(29) 4.141(58)
Height 4.01(15) 4.36(18) 3.900(41) 4.424(35)
Mass gap 4.51(20) 4.56(43)
Note: Table taken from Ref. [29].
where ¸ij, denotes that the sum runs over nextneighbor sites. The Ising model is a well understood
spin system. In order to study the critical phenomena, Wilson [172,173] had suggested to continue
the theory into noninteger dimension d"4!c and to make an expansion of Feynman diagrams
in terms of c. The general theory of critical phenomena holds that critical properties only depend on
the symmetry of the order parameter and the dimension of the system. For example, Ising models
with shortrange interaction with dimension d51 are believed to exhibit the same critical
properties. The critical temperature ¹
°
'0 in d"2, but goes to ¹
°
"0 in d"1. One wants to
know if d"1 is the lower critical dimension, i.e., if ¹
°
P0 when d"1#c and cP0.
H. Kro( ger / Physics Reports 323 (2000) 81}181 163
Fig. 34. Two Sierpinski carpets, with (a) large and (b) small lacunarity. R"R, b"7, and l"3 (see text). Figure taken
from Ref. [60].
The new idea suggested by Gefen et al. was to construct a space}time domain with noninteger
dimension via a fractal lattice, and then study its critical behavior as a function of the noninteger
dimension. When studying critical phenomena on fractal domains one may raise the question: Do
critical properties only depend on the symmetry of the order parameter and the dimension of the
system? As is well known, there are many sets with coinciding fractal dimension, but otherwise
quite di!erent properties. To put the question in this way: Do critical properties depend on the
geometry of the lattice, given the fractal dimension of the lattice? The answer given by Gefen et al.
is: Yes, they do! In particular, critical properties depend on geometric characteristics, like the order
of ramixcation, the connectivity and the lacunarity. The order of rami"cation R of a point
P measures the smallest number of signi"cant interactions which must be cut in order to isolate an
arbitrarily small bounded part of points surrounding P (Urysohn}Menger order of rami"cation
[106]). For rami"cation R"R, connectivity Q is de"ned as the minimum of the fractal dimen
sionalities of the `cuta required to isolate a bounded in"nite part of the system. Lacunarity has
been introduced by Mandelbrot [103]. It measures the extent of failure of a fractal to be
translational invariant.
The following results are obtained: As fractal domain, Gefen et al. have considered Koch's curve,
the Sierpinski gasket and the Sierpinski carpet. Two Sierpinski carpets with the same fractal
dimension but di!erent lacunarities are shown in Fig. 34. The construction is as follows: A square is
subdivided into b` subsquares, then l` of those subsquares are cut out. This yields a dimension
D"ln(b`!l`)/lnb, and a connectivity Q"ln(b!l)/lnb. Then realspace renormalization
transformations are performed (for a discussion of realspace renormalization see Ref. [21]). When
considering a spin model on a regular lattice this means dividing the sites of the lattice into blocks
and then replacing each block by one single site. The lattice is assumed to have a discrete scaling
symmetry, which means that after blocking the lattice looks like the original one, except that the
lattice spacing has increased aPa'"ba. Finally, one reduces all dimensions in the new lattice by
a factor b.
When considering spins as dynamical variables living on such lattice, blocking means that one
replaces all spins in a particular block on the lattice by a new blockspin,
o'¹'
I
"f (¦o
G
}) , (6.2)
164 H. Kro( ger / Physics Reports 323 (2000) 81}181
where k denotes the block and i runs over the lattice sites within the block k. f is a function to be
speci"ed. One possibility is the majority rule: If most of the spins in a block are up (down), then the
block spin is de"ned to be up (down). Another possibility is the rule of decimation: The block spin
o'¹'
I
is taken to be equal to the value of one particular spin o
G
inside the block. The site i of the
particular spin may be chosen at random from the block, or by taking a `"xeda site, e.g., the top left
corner of a square block. The probability density to "nd the spinsystem in a con"guration ¦o
G
} is
given by the Boltzmann}Gibbs law
P(¦o
G
})"
1
Z
exp[!H(¦o
G
})], Z"
¦
NG
}
exp[!H(¦o
G
})] , (6.3)
where H"[H, H being the Ising Hamiltonian, Eq. (6.1), and ["1/k
"
¹ determines the temper
ature dependence. His called e!ective Hamiltonian. From the distribution of original spins, Eq. (6.3),
and the rule Eq. (6.2) (e.g., majority rule or decimation rule), which associates to each spin
con"guration a corresponding blockspin con"guration, one can construct a distribution of
blockspins P'¹'(¦o'¹'
I
}). Then one de"nes a corresponding blockspin Hamiltonian H'¹' via
Boltzmann}Gibbs' law, because the blockspins should reproduce the same physics (thermodyn
amical observables),
P'¹'(¦o'¹'
I
})"
1
Z'¹'
exp[!H'¹'(¦o'¹'
I
})], Z'¹'"
¦
N
'¹'
I
}
exp[!H'¹'(¦o'¹'
I
})] . (6.4)
Combining Eqs. (6.3) and (6.4) de"nes a mapping H(¦o
G
})PH'¹'(¦o'¹'
I
}). This de"nes a new
e!ective Hamiltonian after one step of renormalization. For each iteration, one has a mapping
F: H'L'PH'L>¹' . (6.5)
For practical reasons, one imposes the constraint that the Hamiltonian after renormalization has
the same functional form as before. In particular, the renormalized Hamiltonian has the same
number and type of parameters ([J for the Ising model). Its value changes under the transforma
tion F. In this way, by iteration of the renormalization transformation, the parameters jump in
a parameter space. This is called a (discrete) renormalization group #ow.
Of particular interest in the parameter space are points which are invariants of the renormaliz
ation group #ow, so called "xed points. One may have attractive "xed points (where all lines of #ow
in a neighborhood of the "xed point tend towards the "xed point), or repulsive "xed points or even
mixed "xed points. For systems with a single phase transition, one has two attractive "xed points,
a low temperature "xed point (¹P0) and a high temperature "xed point (¹PR). The domains of
attraction of these "xed points are separated by a so called critical surface in the parameter space.
Gefen et al. have carried out such a renormalization group analysis, but for the Ising system
living on a fractal geometry (lattice). They found via a majority rule renormalization group analysis
a stable "xed point at ¹"0. Because the point ¹"Ris always a stable "xed point, there must be
an unstable "xed point at a "nite value of ¹, i.e., ¹
°
'0. Combination of exact analytical results
with approximate Migdal}Kadano! recursion relations yields renormalization group #ows, which
are displayed in Fig. 35. Fig. 35a corresponds to b"7, l"5, D+1.65 and Q+0.36. The #ow goes
from an unstable "xed point A to a stable "xed point B. At the "xed point B one "nds y(B)+0.34,
which gives the critical exponent of the correlation length by v"1/y. Fig. 35b corresponds to
H. Kro( ger / Physics Reports 323 (2000) 81}181 165
Fig. 35. Renormalization group #ow diagrams for Ising model on Sierpinski carpets with b"7 and (a) l"5, (b) l"3,
t"tan h(K), t
`
"tanh(K
`
). C is the critical point (K"K
`
). Figure taken from Ref. [60].
b"7, l"3, D+1.90 and Q+0.71. This yields y(B)+0.52. As a rule one observes that ¹
°
and
y"1/v decrease with decreasing D and Q. In summary, one "nds that critical properties like
¹
°
and the critical exponent v depend not only on the dimension D, but also on connectivity and
lacunarity of the system.
As has been pointed out above, when constructing a lattice corresponding to a noninteger
dimension, e.g., via the Sierpinski carpet, then the exact symmetry of translational invariance is
lost. A measure for the violation of this symmetry is the lacunarity. On the other hand, abstract
analytic continuation on hypercubic lattices allows to maintain translational invariance. Thus it is
interesting to study noninteger lattices where translational symmetry is broken only `a little bita,
which means considering lattices of low lacunarity. This has been investigated by Gefen et al. in
Ref. [61]. They claim that in the limit of low lacunarity, the physical properties of systems de"ned
on these fractals become identical to those obtained by abstract analytically continuation of
hypercubic lattices.
An important test of this claim is the computation of critical exponents. Such a quantity is, e.g.,
the critical exponent v,1/y, which characterizes the behavior of the correlation length at the
critical temperature,
¸&¦¹!¹
°
¦J . (6.6)
Migdal's realspace recursion relations [82,108] predict in d"1#c dimensions
y"1/v"c#O(c`) . (6.7)
Gefen et al. have used again a fractal lattice of low lacunarity, constructed by a Sierpinski carpet
and having dimension D"1#c, c;1. Using a dedecoration renormalization group method, they
"nd for the Ising model the critical exponent v, which is consistent with Eq. (6.7).
The study of critical properties of the Ising model on a lattice of fractal geometry, which has been
done by Gefen et al. via analytical renormalization group methods, has been extended by Bhanot
et al. [17] using numerical simulations via the Monte Carlo method. Using a 64;64 lattice
embedded in two space dimensions, they constructed also a fractal lattice of Sierpinski carpet type,
with parameters b"2, l"2 (see above) yielding a fractal dimension D"ln48/ln8+1.86. Two
couplings, [
G
and [
@
, have been introduced to distinguish the coupling of internal links and
166 H. Kro( ger / Physics Reports 323 (2000) 81}181
coupling of boundary links,
Z"
Q
exp
[
G
"°
'"'°'"º' ''"''
s
L
s
K
#[
@
"°
'º""°º'` ''"''
s
L
s
K
. (6.8)
By numerical experiments they have measured the spin}spin correlation, which has been "tted by
the following function:
¸s
x
s
0
,"A([
G
)
exp[!m([
G
)x]
xM'@G'
#
exp[!m([
G
)(64!x)]
(64!x)M'@G'
. (6.9)
From that they have extracted the critical exponents j([
G
) and v, the latter being de"ned by
m([
G
)&m
"
¦[
G
![°
G
¦J . (6.10)
They found
v"1.28$0.05 . (6.11)
They measured the magnetization in the disordered phase,
,
+
"<¹
L
s
L
`
°''º'°°'°°
&A¦[
G
![°
G
¦A , (6.12)
and extracted the critical exponent ¸,
¸"1.90$0.05 . (6.13)
They also measured the spontaneous magnetization in the broken phase,
¸M,/<"<¹
L
s
L
''º'°" º"º'°
&A¦[
G
![°
G
¦@
I
, (6.14)
giving
[I"0.10$0.02 . (6.15)
A plot of the magnetization ,
+
and the spontaneous magnetization in the broken phase ¸M,/< is
given in Fig. 36. Those critical exponents are related among each other by scaling laws, in which
enters the dimension of the system,
d"¸#2[I/v . (6.16)
This yields
d"1.64$0.08 . (6.17)
One observes that the fractal dimension of the Sierpinki carpet and the dimension from the scaling
law do not agree. The reason for this behavior may be the violation of translational invariance and
"niteness of the lattice volume.
A generalized Sierpinski carpet was used also by Bhanot, Duke and Salvador in Ref. [18] in
order to interpolate between integer dimensions for studying critical behavior. The model con
sidered is the Ising model, with the same coupling constant for all bonds connecting active nearest
H. Kro( ger / Physics Reports 323 (2000) 81}181 167
Fig. 36. Critical behavior for Ising model on Sierpinski carpets. Plot of magnetization ,
+
corresponding to disordered
phase and spontaneous magnetization ¸M,/< in broken phase. The solid curve is a "t to the data points. Figure taken
from Ref. [17].
neighbor spins. The "ndings of this study are:
(i) Instead of the Hausdor! dimension, d
'
it is more appropriate to use the dimension
d
LL
"N
@
/N
Q
, (6.18)
where N
Q
is the number of active spins and N
@
is the number of bonds connecting active sites.
(ii) The authors have made a "nite size scaling analysis on fractal lattices. The speci"c heat
C
T
"(¸S`,!¸S,`)/N
Q
, (6.19)
and the magnetic susceptibility
,
+
"(¸M`,!¸M,`)/N
Q
. (6.20)
are expected to behave according to the "nite size scaling hypothesis like
Cº°º'
T
(¸)&¸?J , (6.21)
and
,
º°º'
+
(¸)&¸AJ , (6.22)
as a function of the system size ¸. Here one wants to keep the fractal dimension, respectively
d
""
"xed. This can be done by "xing the parameters b and c,l (see above). Varying the system
size ¸ can be achieved by varying the number n of steps of decimations (i.e., each step correspond
ing to a division of hypercubes and cutting out a central section). The results are shown in Fig. 37.
For the susceptibility one expects the critical exponent ¸ to be positive, thus the peak should
increase when ¸PR. This behavior can be seen in the "gure. One extracts ¸/v+1.23 for
dimension d
""
"1.926.
The critical exponent ¸ has been determined also directly for di!erent dimensions. It is given by
,
+
&¦[![
°
¦A . (6.23)
168 H. Kro( ger / Physics Reports 323 (2000) 81}181
Fig. 37. Finite size behavior of Ising model on Sierpinski carpets. Dimension d
""
"1.926 is "xed. Variation of size
¸ corresponds to variation of decimation steps n. Plot of magnetization ¸M,/N
'
, susceptibility ,
+
and speci"c heat C
T
.
The curve is a "t to the data points. Figure taken from Ref. [18].
The data yield the numbers
d"3, [
°
"0.222(1), ¸"1.26(3) ,
d
""
"2.609, [
°
"0.263(1), ¸"1.37(4) .
(6.24)
For the case d"3 these numbers agree with those given by [
°
"0.221654(6) and ¸+1.25 reported
in Refs. [23,132]. These results show nicely that fractals can be used to interpolate between integer
dimensions for the study of critical phenomena.
6.2. Geometry of critical clusters: Ising model
In the previous section we have dwelled on the point of view that fractal geometry is useful to
de"ne a spin model (Ising model) in noninteger dimensions and to study critical properties as
a function of such a dimension. On the other hand let us recall what we know about a spin system
or a "eld theory at the critical point: The system is described mathematically by a renormalization
group equation, which tells us how the system behaves under a change of scale. In particular, one is
interested in "xed points of the renormalization group equation. At such "xed points the system is
supposed to be scale invariant. This leads to the mathematical concept of selfsimilarity. Physically,
this means that a system at the critical point forms clusters of all sizes. All length scales occur. This
H. Kro( ger / Physics Reports 323 (2000) 81}181 169
can be seen in Fig. 29 for the case of the Ising spin system. Selfsimilarity being a property of
fractals, it is tempting to relate the size and geometry of clusters at the critical point to fractal
geometry. But contrary to the previous section, we now put the model on a regular space}time
lattice and study the dynamically generated fractal geometry of the clusters.
The relation between the renormalization group, selfsimilarity, critical behavior and fractal
dimensions has been discussed in a nice way by Suzuki [159]. In particular he has suggested the
following relation to hold between the embedding dimension d, the fractal dimension d
'
of the
largest cluster at the critical point, the critical exponent of magnetization [ and the critical
exponent of correlation length v,
d
'
"d![/v . (6.25)
This relation has been tested in numerical simulations and has been con"rmed. Let us outline the
argument. Let us consider a magnetic phase transition of a ferromagnet, in particular we consider
the Ising model. At the critical point, there are clusters of spinup and clusters of spindown. Note
that a cluster is being de"ned as a connected domain, where all spins point in the same direction
(see Fig. 29). Suzuki introduced the concept of a resultant dominant cluster at the critical point.
At the critical point there are N> upspins and N downspins. The total number of spins is
N"N>#N. N> and N are close to N/2 with a #uctuation denoted by AM,
N¯"N/2$AM . (6.26)
The total magnetization of the system is given by
M"N>!N . (6.27)
Now let us make the plausible assumption that large clusters are more important to describe the
critical behavior. One de"nes the resultant dominant clusters (RDC) in the following way: One
eliminates smaller clusters of upspins (or downspins), when AM'0 (or AM(0), until the total
number of eliminated spins becomes N/2. The remaining upspin (or downspin) clusters are called
positive (or negative) RDC. For the construction of RDC see Fig. 38. This construction implies that
the number of total spins of a positive RDC is AM. The total magnetization M of the RDC is given
by Eq. (6.27).
Finitesize scaling theory applied at the critical point to the original spins (before introducing
RDC) tells us that the magnetization obeys a power law as a function of the "nite system size ¸,
M(¸)&¸? at ¹"¹
°
. (6.28)
Such power law re#ects selfsimilarity under a transformation of length scale. Likewise it is
plausible that the magnetization, given by the RDC should obey also a power law,
M(¸)&¸B+ at ¹"¹
°
. (6.29)
The scaling exponent d
+
is interpreted as the fractal dimensionality of the resultant dominant
clusters. Analogous to the magnetization, also the susceptibility given by RDC should obey
a power law,
,(¸)&¸BQ at ¹"¹
°
. (6.30)
170 H. Kro( ger / Physics Reports 323 (2000) 81}181
Fig. 38. Resultant dominant clusters which are left by eliminating smaller clusters until the number of eliminated spins
becomes N/2. Figure taken from Ref. [159].
We are interested to "nd scaling relations between d
+
, d
Q
, d and the critical exponents. Let us
denote
c"(¹!¹
°
)/¹
°
, <"¸B . (6.31)
Near the critical temperature, magnetization and susceptibility behave like
M&< ¦¹!¹
°
¦@&<c@,
,&< ¦¹!¹
°
¦A&<cA .
(6.32)
The correlation length behaves as
¸&¦¹!¹
°
¦J&cJ . (6.33)
Going over to a "nite system, the correlation length cannot be larger than the size ¸ of the system,
thus ¸ is replaced by ¸, to give
c&¸¹J . (6.34)
Combining Eqs. (6.32) and (6.34) implies
M(¸)&<¸@J&¸B@J,
,(¸)&<¸AJ&¸B>AJ .
(6.35)
Combining Eqs. (6.29), (6.30) and (6.35) yields
d
+
"d![/v, d
Q
"d#¸/v . (6.36)
So far we did not use explicitly the concept of resultant dominant clusters. We use it now to show
that Eq. (6.36) is consistent with the well known scaling relation
2[#¸"dv . (6.37)
From the de"nition of susceptibility, ,"R¸M(B),/RB¦
¯"
follows that magnetization and suscep
tibility are related via
,&¸M`,/k¹
°
. (6.38)
H. Kro( ger / Physics Reports 323 (2000) 81}181 171
Now for the Ising model, being a classical model, one computes
¸M`,"
G
SX
G
H
SX
H
"
°º"'
(N>!N)(N>!N)"
°º"'
(2AM)`
+(M(¸))` at ¹
°
, (6.39)
where
°º"'
stands for the sum over con"gurations weighted with the Boltzmann factor. The
crucial point is that the approximate equality in the last line is assumed to be satis"ed for the
resultant dominant clusters near the critical point. Further one makes the assumption that this
approximate relation also holds for quantum ferromagnets (e.g., the Heisenberg model). Relation
(6.39) implies for the exponents de"ned in Eqs. (6.29) and (6.30)
d
Q
"2d
+
. (6.40)
Comparing the scaling relations given by Eq. (6.36) and Eq. (6.40), implies the scaling relation
Eq. (6.37) and thus con"rms that the resultant dominant cluster assumption is not inconsistent.
Numerical studies of critical clusters have been carried out to test the scaling relation, Eq. (6.25).
The Ising model near criticality in d"2 and d"3 dimensions has been studied numerically via
Monte Calo methods by Cambier and Nauenberg [25]. They have considered the system closely
below the critical temperature ¹
°
and studied the cluster size distribution, mean cluster surface and
the cluster radius of gyration. The cluster distribution is supposed to have some scaling behavior,
when approaching ¹
°
[20]. They found the predicted scaling behavior in a narrow temperature
range, and the corresponding critical exponents to obey the theoretically predicted scaling rela
tions. Numerical results for the gyration radius of clusters are shown in Fig. 39. The Hausdor!
dimension is de"ned by the scaling law between the size (volume) of the cluster n and the radius of
gyration R
'
,
n&RB'
'
. (6.41)
Taking the average over a range of temperatures, they obtain
d
'
"1.9$0.06 for d"2 ,
d
'
"2.3$0.05 for d"3 .
(6.42)
The value for d"2 is consistent with Eq. (6.25), which gives d
'
"1.875. For d"3 the equation
yields d
'
"2.5.
As hinted by the discrepancy between the Hausdor! dimension obtained by Monte Carlo data
and that from Suzuki's scaling relation, there is a problemwhich has to do with the de"nition of the
critical clusters in the Ising model. The naive de"nition of clusters made of nearest neighbor
parallel spins is not satisfactory [3,20,34]. The clusters come out too large. The clusters are due to
correlation, but there are also purely geometrical e!ects. A new de"nition of a cluster has been
proposed by Coniglio and Klein [32] for the Ising system. It has been generalized by Coniglio and
Peruggi [33] for the qstate Potts model. The starting point was the general belief that Ising
clusters at the critical point should have the following properties: (i) They diverge at the critical
point, (ii) their linear dimension diverges as the correlation length, and (iii) the mean cluster size
given by the second moment of the cluster size distribution diverges as the magnetic susceptibility.
However, it has been observed [162] that the mean cluster size diverges like ¦¹!¹
°
¦AN with
172 H. Kro( ger / Physics Reports 323 (2000) 81}181
Fig. 39. Hausdor! dimension obtained from the gyration radius of clusters (a) for d"2 and (b) for d"3. Similar data
are obtained for other temperatures. Figure taken from Ref. [25].
¸
N
"1.91$0.01, which is di!erent from the susceptibility exponent ¸"1.75. That something was
wrong could be seen also from the extreme case of in"nite temperature and zero magnetic "eld:
Then one expects no correlation, however, the cluster size is di!erent from zero. Then Coniglio and
Klein have suggested the following de"nition of clusters: Let us consider the more general case of
the qstate Potts model. Its Hamiltonian is given by
!H
"º''
/k
"
¹"K
GH
o
NGNH
where o
G
"1, 2,
2
, q . (6.43)
The new clusters are de"ned as nearest neighbor sites in the same state connected by bonds, each
bond being present with probability p"1!e), where K is the nearest neighbor coupling
constant of the Potts Hamiltonian. For the case of the Ising model this is exempli"ed in Fig. 40. In
a typical con"guration, one has to distinguish dangling bonds and a backbone. The backbone is
made of singly connected bonds, called red bonds (such that if one is cut the top is disconnected
from the bottom and the blobs which are made of multiply connected bonds). For each cluster, the
hull is made of all the absent bonds surrounding the cluster. The absent bonds are of two types,
those between sites in the same state and those between sites in di!erent states. Using this modi"ed
de"nition of clusters, Coniglio and Klein [32] found that for the Ising model the critical properties
(i)}(iii) listed above are ful"lled. The same properties are satis"ed also for the qstate Potts model
[33]. Using this new de"nition of critical clusters, Coniglio [34] has been able to compute the
fractal dimension of these clusters, by mapping the Potts model to the Coulomb gas. The results are
given in Table 1.
An alternative way to de"ne critical clusters for the Ising model has been discussed by
Alexandrowicz [3]. Fractal dimensions of critical Ising clusters in 2 dimensions have been studied
using conformal invariance by Stella and Vanderzande [158]. The Coniglio}Klein clusters
of the Ising model in 3 dimensions have been investigated by Wang and Stau!er [167] using
H. Kro( ger / Physics Reports 323 (2000) 81}181 173
Fig. 40. (a) Ising con"guration: `upa and `downa spins are represented, respectively, by dots and circles. (b) Correct
clusters obtained from the con"guration given in (a) by putting bonds between sites in the same state, with probability
p"1!e). Note the spanning clusters of the `upa spins is made of three dangling bonds, four red bonds (if one of these
is cut the cluster does not span anymore), and one blob made of four bonds. Figure taken from Ref. [34].
Monte Carlo simulations. They have measured the cluster radius of gyration R
Q
as a function of the
number s of spins in that cluster. Fitting their data to the expected scaling behavior
R
Q
Js¹"' , (6.44)
yields D
'
"2.51. From the scaling relation D
'
"d![/v one expects D
'
"2.484.
6.3. Fractal geometry of topological excitations: X> spin model
The X> model is a classical spin model, which plays an important role in the physics of
super#uidity. Super#uidity of liquid "He is known to exist since more than half a century ago.
Onsager [120] and Feynman [53] proposed vortex excitations to be responsible for the super#uid
phase transition. In D"2 dimensions, a system with some continuous symmetry cannot produce
spontaneous symmetry breaking. The argument for this goes back to Peierls [133], Landau [97]
and Mermin and Wagner [107]. The importance of vortex excitations (topological excitations) in
2D had been demonstrated theoretically by Kosterlitz and Thouless [88] which has later been
con"rmed by experiments with thin helium"lms [151]. In 3D vortex excitations also play a crucial
role in the transition. This has been shown by Kohring et al. [87] via Monte Carlo simulations in
the 3D X> model.
The X>model is de"ned on a 2D lattice as follows. Let us consider a regular lattice of spacing a.
The lattice sites are denoted by x
G
. Let s
G
denote a classical spin, described a vector of length unity in
the 2D plane being de"ned on each lattice site. Then the Hamiltonian is given by
H"!
J
2
GH
s
G
) s
H
"!
J
2
GH
cos(
G
!
H
) , (6.45)
174 H. Kro( ger / Physics Reports 323 (2000) 81}181
where
G
denotes the angle between spin s
G
and an arbitrary, but globally "xed direction. It is
evident that the Hamiltonian is invariant under global rotations. The model in 3D is given by the
same Hamilton function, except for the lattice sites being now de"ned on a cubic lattice. Vortices
are important in 2D, where the Kosterlitz}Thouless phase transition is explained as a dissociation
of vortex}antivortex pairs.
In D"3 dimensions, the vortex excitations are also called vortex loops. Contrary to the situation
in D"2, in D"3 there is a conservation law giving closed vortex loops. The current is de"ned
to pass through a plaquette formed by four neighboring lattice sites, say i, j, k, l. Let
G
,
2
,
J
denote
the corrsponding spin angles, which take values in the interval (!, ]. The the current is
de"ned by
J"
1
2
[
G
!
H
] , (6.46)
where the sum runs over the four oriented links ¸ij,, ¸jk,, ¸kl, and ¸li, and the values in the
bracket [ ] lie also in the interval (!, ]. Thus J is integer valued. If JO0, the #ux is associated
with a link of the dual lattice perpendicular to the plaquette. Such excited links form closed loops.
The situation with the 3D planar X>model is similar to the Abelian ;(1) lattice gauge model in
4D, where closed monopole current loops exist. Those loops have been shown to display fractal
geometry, depending on the phase of the model. The fractal dimension of the monopole current
loop has been shown to play the role of an order parameter. A similar situation prevails here. The
model has a phase transition at [
°
"0.4542, where ["1/(k
"
¹). Pochinsky et al. [134] have studied
the geometry of such vortex loops from simulations on a lattice (7`}14`) and "nd the behavior for
[([
°
di!erent from that for ['[
°
. In Fig. 41, the fractal dimension is plotted versus the inverse
temperature [. The fractal dimension of a vortex loop is de"ned here as the ratio of the number of
links to the number of sites belonging to the vortex. One observes a behavior compatible with
D
'
"1 for ['[
°
and D
'
'1 for [([
°
.
7. Concluding discussion and outlook
We have given an account of a number of areas where fractal geometry plays a role in quantum
physics. In nonrelativistic quantum mechanics, for the process of percolation, fractal geometry is
suitable to describe the geometry of porous media. In other words, one considers quantum
mechanics with fractal boundary conditions. On the other hand, the dynamics of a free system
without boundary conditions can be characterized by fractal paths. Those paths for most interac
tions (local potentials) correspond to trajectories of Hausdor! dimension d
'
"2. It is interesting to
consider situations, where d
'
O2. An example is a velocity dependent potential, like <&x ", which
may occur, when a particle propagates in a medium, giving rise to modi"ed dispersion relations.
Another example would be to consider quantum mechanics in curved space}time, corresponding
to the situation when a quantum particle visits a neutron star or a black hole. By Einstein's general
theory of relativity, one has a correspondence between a distribution of massive bodies, i.e., the
gravitational forces, and the curvature of space}time. It is tempting to speculate: Assuming that the
Hausdor! dimension of propagation of a quantum particle depends locally on the curvature of
H. Kro( ger / Physics Reports 323 (2000) 81}181 175
Fig. 41. Fractal dimension of vortex loops in 3D X> model versus [. Figure taken from Ref. [134].
space}time, is it possible then to establish a correspondence between the distribution of massive
bodies and the Hausdor! dimension of the probing quantum particle? In other words, can one "nd
a map of the Hausdor! dimension?
This leads us to the next question: Can one measure experimentally the Hausdor! dimension of
propagation of a quantum particle? In principle, one expects the answer to be yes. In practice, it
seems to be quite complicated to devise such an experiment. One severe inherent problemis the fact
that the determination of d
'
requires to use a resolution of length which goes to zero. Then
Heisenberg's uncertainty relation invokes a large uncertainty in momentum, which through the
measuring process interferes with the particle and `createsa an erratic path. We have discussed
a proposal, how to avoid this problem, via the topological Aharonov}Bohm e!ect. With present
day progress in quantum optics and electron interferometry, one might think of doing some
preliminary experiment.
In manybody systems, like spin systems or quantum"eld theory, fractal geometry has proven to
be a useful tool, in particular for the analysis of critical phenomena. Using Sierpinski carpets, it has
been possible to simulate systems in noninteger space dimension and study critical behavior as
a function of space dimension. Doing numerical simulations (Monte Carlo) on those lattices
provides nonperturbative information in noninteger dimensions. Fractal geometry has been used
to analyze the geometry of critical clusters at second order phase transitions. An interesting
question is that of a relation between critical exponents and the Hausdor! dimension. However,
there is no unique way to de"ne such a Hausdor! dimension. In quantum gravity, e.g., the
Hausdor! dimension has been de"ned via comparison of scaling exponents of two observables,
176 H. Kro( ger / Physics Reports 323 (2000) 81}181
which have naive dimension of length and volume, respectively. The Hausdor! dimension de"ned
in such a way is closely related to the anomalous dimension of a vertex function. Another de"nition
of the Hausdor! dimension, based on the hopping expansion of a propagator has been suggested in
"eld theory.
Another interesting subject is the role of fractal geometry at the decon"nement phase transition
at "nite temperature in lattice gauge theory. An observable has been found (the monople current
loop) which is nonfractal in the decon"ned phase, but becomes fractal in the con"ned phase. Thus
it plays the role of an order parameter.
We have emphasized the correspondence between a fractal, being a geometrical object having
the property of selfsimilarity, and the renormalization group in physics, which also has built in the
property of self similarity (generating the running coupling constant).
Finally, let us recall that fractal dimension in classical physics tells us something about an
observable (e.g. the length of a coast line) when the resolution of length goes to zero. A similar
situation prevails when we consider hadron structure functions as function of Q`, where 1/(Q` is
the wave length of resolution. Also the formal instrument of operator product expansion makes
statements about the singularities of a theory at small distances. Thus it is interesting to speculate
about the role that fractal geometry might possibly play in the context of operator product
expansion and hadron structure functions.
References
[1] L.F. Abbot, M.B. Wise, Am. J. Phys. 49 (1981) 37.
[2] Y. Aharonov, D. Bohm, Phys. Rev. 115 (1959) 485.
[3] Z. Alexandrowicz, Phys. Rev. Lett. 60 (1988) 669.
[4] G. Altarelli, G. Parisi, Nucl. Phys. B 126 (1977) 298.
[5] G. Altarelli, Phys. Rep. 81 (1982) 1.
[6] J. Ambjorn, P. Oleson, Nucl. Phys. B 170[FS1] (1980) 60.
[8] J. Ambjorn, G. Thorleifson, Phys. Lett. B 323 (1994) 7.
[9] J. Ambjorn, Quantization of Geometry, in: Fluctuating Geometries in Statistical Mechanics and Field Theory, Les
Houches, 1994.
[10] J. Ambjorn, Y. Watabiki, Niels Bohr Institute, preprint NBIHE9501.
[11] D.J. Amit, Field Theory, the Renormalization Group and Critical Phenomena, World Scienti"c, Singapore,
1984.
[12] M. Bander, Phys. Rep. 75 (1981) 205.
[13] T. Banks, R. Myerson, J. Kogut, Nucl. Phys. B 129 (1977) 493.
[14] A.A. Belavin, A.M. Polyakov, A.S. Schwartz, Y.S. Tyupkin, Phys. Lett. B 59 (1975) 85.
[15] C. Bernard, D. Murphy, A. Sony, K. Lee, Nucl. Phys. B (Proc. Suppl.) 17 (1990) 593.
[16] C. Bernard, A. Sony, K. Lee, Nucl. Phys. B (Proc. Suppl.) 20 (1991) 410.
[17] G. Bhanot, H. Neuberger, J.A. Shapiro, Phys. Rev. Lett. 53 (1984) 2277.
[18] G. Bhanot, D. Duke, R. Salvador, Phys. Lett. B 165 (1985) 355.
[19] G. Bhanot, R. Salvador, Phys. Lett. B 167 (1986) 343.
[20] K. Binder, Ann. Phys. (N.Y.) 98 (1976) 390.
[21] J.J. Binney, N.J. Dowrick, A.J. Fisher, M.E.J. Newman, The Theory of Critical Phenomena, Oxford University
Press, Oxford, 1992.
[22] T.S. BiroH , S.G. Matinyan, B. MuK ller, Chaos and Gauge Field Theory, Lecture Notes in Physics, Vol. 56, World
Scienti"c, Singapore, 1994.
H. Kro( ger / Physics Reports 323 (2000) 81}181 177
[23] E. BreH zin, J.C. Le Guillou, J. ZinnJustin, in: C. Domb, M.S. Green (Eds.), Phase Transitions and Critical
Phenomena, Vol. 6, Academic Press, New York, 1976.
[24] K.A. Brueckner, Phys. Rev. 97 (1955) 1353.
[25] J.L. Cambier, M. Nauenberg, Phys. Rev. B 34 (1986) 8071.
[26] E. CampesinoRomeo, J.C. D'Olivio, M. Sokolovsky, Phys. Lett. A 89 (1982) 321.
[27] F. Cannata, L. Ferrari, Am. J. Phys. 56 (1988) 721.
[28] J.L. Cardy, Nucl. Phys. B 170[FS1] (1980) 369.
[29] S. Catteral, G. Thorleifson, M. Bowick, V. John, Syracuse University, preprint, 1995, SU4240607.
[31] F.E. Close, An Introduction to Quarks and Partons, Academic Press, London, 1979.
[32] A. Coniglio, W. Klein, J. Phys. A: Math. Gen. 13 (1980) 2775.
[33] A. Coniglio, F. Peruggi, J. Phys. A: Math. Gen. 15 (1982) 1873.
[34] A. Coniglio, Phys. Rev. Lett. 62 (1989) 3054.
[35] M. Creutz, Phys. Rev. Lett. 43 (1979) 553.
[36] M. Creutz, L. Jacobs, C. Rebbi, Phys. Rev. Lett. 42 (1979) 1390; Phys. Rev. D 20 (1979) 1915.
[37] M. Creutz, Phys. Rev. D 23 (1981) 1815.
[38] M. Creutz, Quarks, Gluons and Lattices, Cambridge University Press, Cambridge, 1983.
[39] J.M. Daul, Ecole Normale SupeH rieure, preprint, 1994, hepth/9502014.
[40] F. David, Nucl. Phys. B 368 (1992) 671.
[41] T.A. DeGrand, D. Toussaint, Phys. Rev. D 22 (1980) 2478.
[42] S. Deser, Phys. Lett. B 64 (1976) 463.
[43] N. Eicker, U. GlaK ssner, S. GuK sken, H. Hoeber, P. Lacock, Th. Lippert, G. RitzenhoK fer, K. Schilling, G. Siegert,
A. Spitz, P. Ueberholz, J. Vieho!, preprint, WUB 9714, heplat/9704019.
[44] S. Elitzur, Phys. Rev. D 12 (1975) 3978.
[45] M.S. El Naschie, Chaos Solitons Fractals 2 (1992) 437.
[46] M.S. El Naschie, Chaos Solitons Fractals 5 (1995) 881.
[47] M.S. El Naschie, O.E. RoK ssler, I. Prigogine, Quantum Mechanics, Di!usion and Chaotic Fractals, Elsevier,
Amsterdam, 1995.
[48] M.S. El Naschie, L. Nottale, S. Al Athel, G. Ord, Chaos Solitons Fractals 7 (6) (1996).
[49] M.S. El Naschie, O.E. RoK ssler, G. Ord, Chaos Solitons Fractals 7 (5) (1996).
[50] J. Feder, Fractals, Plenum, New York, 1988.
[51] I. FeH nyes, Z. Phys. 132 (1952) 81.
[52] R.P. Feynman, Rev. Mod. Phys. 20 (1948) 367.
[53] R.P. Feynman, in: C.J. Gorter (Ed.), Progress in Low Temperature Physics, Vol. 1, NorthHolland, Amsterdam,
1955, p. 17.
[54] R.P. Feynman, Quantum Electrodynamics, Benjamin, New York, 1961.
[55] R.P. Feynman, R.B. Leighton, M. Sands, The Feynman Lectures on Physics, Vol. II, AddisonWesley, Reading,
MA, 1964.
[56] R.P. Feynman, A.R. Hibbs, Quantum Mechanics and Path Integrals, McGrawHill, New York, 1965.
[58] R.P. Feynman, A.R. Hibbs, Quantum Mechanics and Path Integrals, McGrawHill, New York, 1965
(chapter 7).
[59] P.H. Frampton, Gauge Field Theories, Benjamin/Cummings, Menlo Park, 1987.
[60] Y. Gefen, B.B. Mandelbrot, A. Aharony, Phys. Rev. Lett. 45 (1980) 855.
[61] Y. Gefen, I. Meir, B.B. Mandelbrot, A. Aharony, Phys. Rev. Lett. 50 (1983) 145.
[62] H.A. Gersch, Int. J. Theor. Phys. 20 (1981) 491.
[63] J. Glimm, A. Ja!e, Quantum Physics, Springer, New York, 1981.
[64] S. Gluzman, V.I. Yukalov, Phys. Rev. E 55 (1997) 3983.
[65] B.J. Gough et al., FermilabPub96283, hepph/9610223.
[66] R. Gupta, T. Bhattacharya, heplat/9605039, Phys. Rev. D 55 (1997) 7203.
[67] G. 't Hooft, Nucl. Phys. B 79 (1974) 276.
[68] G. 't Hooft, in: A. Zichichi (Editorice Compositori), High Energy Physics, Bologna (1976); Nucl. Phys. B 190 [FS3]
(1981).
178 H. Kro( ger / Physics Reports 323 (2000) 81}181
[69] G. 't Hooft, Phys. Rev. Lett. 37 (1976) 8.
[70] G. 't Hooft, Nucl. Phys. B 138 (1978) 1.
[71] G. 't Hooft, Nucl. Phys. B 190[FS3] (1981) 455.
[72] E.M. Ilgenfritz, M.L. Laursen, M. MuK llerPreussker, G. Schierholz, U.J. Wiese, Nucl. Phys. B 268 (1986) 693.
[73] E.M. Ilgenfritz, A. Schiller, Phys. Lett. B 242 (1990) 89.
[74] E.M. Ilgenfritz, A. Schiller, H. Markum, Phys. Rev. D 45 (1992) 2949.
[75] T.L. Ivanenko, A.V. Pochinsky, M.I. Polikarpov, Phys. Lett. B 252 (1990) 631.
[76] C. Itzykson, J.M. Drou!e, Statistical Field Theory, Cambridge University Press, Cambridge, 1989.
[77] R.L. Ja!e, Spin, twist and hadron structure in deep inelastic processes, Erice Lecture, 1995, hepph/9602236.
[78] K. Jansen, C.B. Lang, Phys. Rev. Lett. 66 (1991) 3008.
[79] J. JoseH , L. Kadano!, S. Kirkpatrick, D. Nelson, Phys. Rev. B 16 (1977) 1217.
[80] M. Kac, Rocky Mountain, J. Math. 4 (1974) 497.
[81] L.P. Kadano!, Physics 2 (1966) 263.
[82] L.P. Kadano!, Ann. Phys. (N.Y.) 100 (1976) 359.
[83] V. Kazakov, B. Kostov, A.A. Migdal, Phys. Lett. B 157 (1985) 295.
[84] H. Kawai, M. Ninomiya, Nucl. Phys. B 336 (1990) 115.
[85] N. Kawamoto, V.A. Kazakov, Y. Saeki, Y. Watabiki, Phys. Rev. Lett. 68 (1992) 2113.
[87] G. Kohring, R. Schrock, P. Wills, Phys. Rev. Lett. 57 (1986) 1358.
[88] J.M. Kosterlitz, D.J. Thouless, J. Phys. C (Solid State) 6 (1973) 1181.
[89] H. KroK ger, Phys. Lett. B 284 (1992) 357.
[90] H. KroK ger, S. Lantagne, K.J.M. Moriarty, Mod. Phys. Lett. A 8 (1993) 445.
[91] H. KroK ger, S. Lantagne, K.J.M. Moriarty, B. Plache, Phys. Lett. A 199 (1995) 299.
[92] H. KroK ger, Phys. Lett. A 213 (1996) 211.
[93] H. KroK ger, Phys. Rev. A 55 (1997) 951.
[94] H. KroK ger, Phys. Lett. A 226 (1997) 127.
[95] A.S. Kronfeld, G. Schierholz, U.J. Wiese, Nucl. Phys. B 293 (1987) 461.
[96] A.S. Kronfeld, M.L. Lauersen, G. Schierholz, U.J. Wiese, Phys. Lett. B 198 (1987) 516.
[97] L.D. Landau, Zh. Eksp. Teor. Fiz. 7 (1937) 627.
[98] L.D. McLerran, B. Svetisky, Phys. Rev. D 24 (1981) 450.
[99] M. LuK scher, P. Weisz, Nucl. Phys. B 290 (1987) 25.
[100] P.B. Mackenzie, Nucl. Phys. B (Proc. Suppl.) 53 (1997) 23.
[101] O. Madelung, Solid State Theory, Springer, Berlin, 1981.
[102] S. Mandelstam, Phys. Rep. 23 (1976) 245.
[103] B.B. Mandelbrot, C. R. Acad. Sci. Ser. A 288 (1979) 81.
[104] B.B. Mandelbrot, The Fractal Geometry of Nature, Freeman, New York, 1983.
[105] D.G.C. McKeon, G.N. Ord, Phys. Rev. Lett. 69 (1992) 3.
[106] K. Menger, Kurventheorie, Chelsea, New York, 1932, p. 97.
[107] N.D. Mermin, H. Wagner, Phys. Rev. Lett. 17 (1966) 1133.
[108] A.A. Migdal, Sov. Phys. JETP 42 (1975) 743.
[109] I. Montvay, G. MuK nster, Quantum Fields on a Lattice, Cambridge University Press, Cambridge, 1994.
[110] M. Nagasawa, SchroK dinger Equation and Di!usion Theory, BirkhaK user Verlag, Basel, 1993.
[111] M. Nagasawa, M.S. El Naschie, Chaos Solitons Fractals 8 (11) (1997).
[112] E. Nelson, Phys. Rev. 150 (1966) 1079.
[113] E. Nelson, Quantum Fluctuations, Princeton University Press, Princeton, NJ, 1985.
[114] H.B. Nielsen, Phys. Lett. B 80 (1978) 133.
[115] H.B. Nielsen, M. Ninomiya, Nucl. Phys. B 156 (1979) 1.
[116] J. Nishimura, M. Oshikawa, Phys. Lett. B 338 (1994) 187.
[117] L. Nottale, Int. J. Mod. Phys. A 4 (1989) 5047.
[118] L. Nottale, Int. J. Mod. Phys. A 7 (1992) 4899.
[119] L. Nottale, Fractal SpaceTime and Microphysics, World Scienti"c, Singapore, 1993.
[120] L. Onsager, Nuovo Cimento Suppl. 6 (1949) 249.
H. Kro( ger / Physics Reports 323 (2000) 81}181 179
[121] P. Oleson, Nucl. Phys. B 200[FS4] (1982) 381.
[122] G.N. Ord, J. Phys. A 16 (1983) 1869.
[123] G.N. Ord, Int. J. Theor. Phys. 31 (1992) 1177.
[124] G.N. Ord, Phys. Lett. A 173 (1993) 343.
[125] G.N. Ord, Chaos Solitons Fractals 7 (1996) 821.
[126] G.N. Ord, Ann. Phys. 250 (1996) 51.
[127] G.N. Ord, M. Conrad, O.E. RoK ssler, M.S. El Naschie, Chaos, Solitons Fractals 8 (5) (1997).
[128] G.N. Ord, Chaos Solitons Fractals 9 (1998) 1011.
[129] K. Osterwalder, R. Schrader, Phys. Rev. Lett. 29 (1972) 1423; Helv. Phys. Acta 46 (1973) 277; Comm. Math. Phys.
31 (1973) 83.
[130] F. Palumbo, M.I. Polikarpov, A.I. Veselov, Phys. Lett. B 297 (1992) 171.
[131] A. Patrasciu, E. Seiler, I. Stamatescu, Phys. Lett. B 107 (1981) 364.
[132] G.S. Pawley, R.H. Swensen, D.J. Wallace, K.G. Wilson, Phys. Rev. B 29 (1984) 4030.
[133] R. Peierls, Ann. Inst. Henri PoincareH 5 (1935) 177; Helv. Phys. Acta Suppl. 7 (1936) 81.
[134] A.V. Pochinsky, M.I. Polikarpov, B.N. Yurchenko, Phys. Lett. A 154 (1991) 194.
[135] M.I. Polikarpov, U.J. Wiese, Moscow ITEP 3090, Pis'ma Zh. Eksp. Teor. Fiz. 51 (1990) 296.
[136] M.I. Polikarpov, Moscow ITEP 7190.
[137] M.I. Polikarpov, Phys. Lett. B 236 (1990) 61.
[140] M.I. Polikarpov, K. Yee, Phys. Lett. B 333 (1994) 452.
[141] M.I. Polikarpov, Nucl. Phys. B (Proc. Suppl.) 53 (1997) 134.
[142] J. Polonyi, UniversiteH Strasbourg preprint, hepth/9809116.
[143] A.M. Polyakov, JETP Lett. 20 (1974) 194.
[144] A.M. Polyakov, Gauge Fields and Strings, Harwood Publ., Chur, 1978.
[145] B. Povh, K. Rith, C. Scholz, F. Zetsche, Teilchen und Kerne, Springer, Berlin, 1994.
[146] T. Regge, Nuovo. Cimento 19 (1961) 558.
[147] Regge approach: J. Hartle, J. Math. Phys. 26 (1985) 804; 27 (1986) 287; 30 (1989) 452; H. Hamber, in:
K. Osterwalder, R. Stora (Eds.), Critical Phenomena, Random Systems, Gauge Theories, Les Houches, North
Holland, Amsterdam, 1986, p. 375; B. Berg, in: H. Mitter, F. Widder (Eds.), Particle Physics and Astrophysics,
XXVII UniversitaK tswochen, Springer, Berlin, 1989, p. 223.
[148] R.G. Roberts, The Structure of the Proton, Cambridge University Press, Cambridge, 1993.
[149] G. Roepstor!, Path Integral Approach to Quantum Physics, Springer, Berlin, 1994.
[151] I. Rudnick, Phys. Rev. Lett. 40 (1978) 1454.
[152] R. Savit, Phys. Rev. Lett. 39 (1977) 55.
[153] L.S. Schulman, J. Math. Phys. 12 (1971) 304.
[155] E. Seiler, I. Stamatescu, D. Zwanziger, Nucl. Phys. B 239 (1984) 177; 201.
[156] D.V. Shirkov, Sov. Phys. Dokl. 27 (1982) 197.
[157] I. Stamatescu, U. Wol!, D. Zwanziger, Nucl. Phys. B 225[FS9] (1983) 377.
[158] A.L. Stella, C. Vanderzande, Phys. Rev. Lett. 62 (1989) 1067.
[159] M. Suzuki, Prog. Theor. Phys. 69 (1983) 65.
[160] T. Suzuki, I. Yotsuyanagi, Kanazawa Univ., preprint, DPKU8909, 1989.
[161] T. Suzuki, Y. Matsubara, S. Kitahara, S. Ejiri, N. Nakamura, F. Shoji, M. Sei, S. Kato, N. Arasaki, Nucl. Phys.
B (Proc. Suppl.) 53 (1997) 531.
[162] M.F. Sykes, D.S. Gaunt, J. Phys. A: Math. Gen. 9 (1976) 2131.
[163] M. Teper, Phys. Lett. B 162 (1985) 357; B 171 (1986) 81; 86.
[165] Dynamical triangulation: F. David, Nucl. Phys. B 257 (1985) 45; 543; J. Ambjorn, S. Varsted, Phys. Lett.
B 266 (1991) 285; Nucl. Phys. B 373 (1992) 557; J. Ambjorn, B. Durhuus, J. FroK hlich, Nucl. Phys. B 257 (1985)
433.
[166] J. Villain, J. Physique (Paris) 36 (1975) 581.
[167] J.S. Wang, D. Stau!er, Z. Phys. B Cond. Mat. 78 (1990) 145.
[168] S. Weinberg, Gravitation and Cosmology, Wiley, New York, 1972.
[169] F.J. Wegner, J. Math. Phys. 12 (1971) 2259.
180 H. Kro( ger / Physics Reports 323 (2000) 81}181
[171] F. Wilczek, Fractional Statistics and Anyon Superconductivity, World Scienti"c, Singapore, 1990.
[172] K.G. Wilson, M.E. Fisher, Phys. Rev. Lett. 28 (1972) 240.
[173] K.G. Wilson, J. Kogut, Phys. Rep. C 12 (1974) 75.
[174] K.G. Wilson, Phys. Rev. D 10 (1974) 2445.
[175] V.I. Yukalov, Phys. Rev. A 42 (1990) 3324.
[176] V.I. Yukalov, J. Math. Phys. 32 (1991) 1235.
[177] V.I. Yukalov, J. Math. Phys. 33 (1992) 3994.
[178] J. ZinnJustin, Quantum Field Theory and Critical Phenomena, Oxford University Press, Oxford, 1989.
H. Kro( ger / Physics Reports 323 (2000) 81}181 181